aboutsummaryrefslogtreecommitdiffstats
path: root/kernel
diff options
context:
space:
mode:
authorIngo Molnar <mingo@elte.hu>2009-04-06 03:02:57 -0400
committerIngo Molnar <mingo@elte.hu>2009-04-06 03:02:57 -0400
commitf541ae326fa120fa5c57433e4d9a133df212ce41 (patch)
treebdbd94ec72cfc601118051cb35e8617d55510177 /kernel
parente255357764f92afcafafbd4879b222b8c752065a (diff)
parent0221c81b1b8eb0cbb6b30a0ced52ead32d2b4e4c (diff)
Merge branch 'linus' into perfcounters/core-v2
Merge reason: we have gathered quite a few conflicts, need to merge upstream Conflicts: arch/powerpc/kernel/Makefile arch/x86/ia32/ia32entry.S arch/x86/include/asm/hardirq.h arch/x86/include/asm/unistd_32.h arch/x86/include/asm/unistd_64.h arch/x86/kernel/cpu/common.c arch/x86/kernel/irq.c arch/x86/kernel/syscall_table_32.S arch/x86/mm/iomap_32.c include/linux/sched.h kernel/Makefile Signed-off-by: Ingo Molnar <mingo@elte.hu>
Diffstat (limited to 'kernel')
-rw-r--r--kernel/Makefile1
-rw-r--r--kernel/async.c18
-rw-r--r--kernel/audit.c9
-rw-r--r--kernel/audit_tree.c2
-rw-r--r--kernel/auditfilter.c16
-rw-r--r--kernel/auditsc.c34
-rw-r--r--kernel/cgroup.c435
-rw-r--r--kernel/cgroup_debug.c2
-rw-r--r--kernel/cpu.c6
-rw-r--r--kernel/cpuset.c254
-rw-r--r--kernel/exec_domain.c23
-rw-r--r--kernel/exit.c245
-rw-r--r--kernel/extable.c48
-rw-r--r--kernel/fork.c83
-rw-r--r--kernel/futex.c201
-rw-r--r--kernel/irq/Makefile1
-rw-r--r--kernel/irq/chip.c7
-rw-r--r--kernel/irq/handle.c42
-rw-r--r--kernel/irq/internals.h3
-rw-r--r--kernel/irq/manage.c223
-rw-r--r--kernel/irq/numa_migrate.c11
-rw-r--r--kernel/irq/pm.c79
-rw-r--r--kernel/irq/spurious.c14
-rw-r--r--kernel/kallsyms.c19
-rw-r--r--kernel/kexec.c22
-rw-r--r--kernel/kmod.c12
-rw-r--r--kernel/kprobes.c19
-rw-r--r--kernel/kthread.c4
-rw-r--r--kernel/latencytop.c83
-rw-r--r--kernel/lockdep.c555
-rw-r--r--kernel/lockdep_internals.h45
-rw-r--r--kernel/lockdep_proc.c22
-rw-r--r--kernel/lockdep_states.h9
-rw-r--r--kernel/module.c387
-rw-r--r--kernel/mutex-debug.c9
-rw-r--r--kernel/mutex-debug.h18
-rw-r--r--kernel/mutex.c121
-rw-r--r--kernel/mutex.h22
-rw-r--r--kernel/ns_cgroup.c14
-rw-r--r--kernel/panic.c115
-rw-r--r--kernel/params.c26
-rw-r--r--kernel/pid.c33
-rw-r--r--kernel/pid_namespace.c15
-rw-r--r--kernel/posix-cpu-timers.c3
-rw-r--r--kernel/power/disk.c143
-rw-r--r--kernel/power/main.c55
-rw-r--r--kernel/power/snapshot.c9
-rw-r--r--kernel/power/swsusp.c18
-rw-r--r--kernel/printk.c26
-rw-r--r--kernel/ptrace.c103
-rw-r--r--kernel/rcupdate.c44
-rw-r--r--kernel/rcutorture.c25
-rw-r--r--kernel/relay.c14
-rw-r--r--kernel/sched.c1134
-rw-r--r--kernel/sched_clock.c38
-rw-r--r--kernel/sched_cpupri.h2
-rw-r--r--kernel/sched_debug.c8
-rw-r--r--kernel/sched_fair.c59
-rw-r--r--kernel/sched_features.h4
-rw-r--r--kernel/sched_rt.c537
-rw-r--r--kernel/sched_stats.h7
-rw-r--r--kernel/signal.c71
-rw-r--r--kernel/slow-work.c640
-rw-r--r--kernel/smp.c432
-rw-r--r--kernel/softirq.c35
-rw-r--r--kernel/spinlock.c18
-rw-r--r--kernel/stop_machine.c4
-rw-r--r--kernel/sys.c5
-rw-r--r--kernel/sysctl.c26
-rw-r--r--kernel/sysctl_check.c1
-rw-r--r--kernel/time/Makefile2
-rw-r--r--kernel/time/clockevents.c20
-rw-r--r--kernel/time/clocksource.c76
-rw-r--r--kernel/time/ntp.c444
-rw-r--r--kernel/time/timecompare.c191
-rw-r--r--kernel/timer.c178
-rw-r--r--kernel/trace/Kconfig132
-rw-r--r--kernel/trace/Makefile13
-rw-r--r--kernel/trace/blktrace.c1549
-rw-r--r--kernel/trace/events.c14
-rw-r--r--kernel/trace/ftrace.c1133
-rw-r--r--kernel/trace/kmemtrace.c339
-rw-r--r--kernel/trace/ring_buffer.c693
-rw-r--r--kernel/trace/trace.c3034
-rw-r--r--kernel/trace/trace.h315
-rw-r--r--kernel/trace/trace_boot.c36
-rw-r--r--kernel/trace/trace_branch.c278
-rw-r--r--kernel/trace/trace_clock.c109
-rw-r--r--kernel/trace/trace_event_profile.c31
-rw-r--r--kernel/trace/trace_event_types.h173
-rw-r--r--kernel/trace/trace_events.c824
-rw-r--r--kernel/trace/trace_events_filter.c427
-rw-r--r--kernel/trace/trace_events_stage_1.h39
-rw-r--r--kernel/trace/trace_events_stage_2.h176
-rw-r--r--kernel/trace/trace_events_stage_3.h281
-rw-r--r--kernel/trace/trace_export.c102
-rw-r--r--kernel/trace/trace_functions.c369
-rw-r--r--kernel/trace/trace_functions_graph.c635
-rw-r--r--kernel/trace/trace_hw_branches.c185
-rw-r--r--kernel/trace/trace_irqsoff.c54
-rw-r--r--kernel/trace/trace_mmiotrace.c45
-rw-r--r--kernel/trace/trace_nop.c6
-rw-r--r--kernel/trace/trace_output.c1017
-rw-r--r--kernel/trace/trace_output.h71
-rw-r--r--kernel/trace/trace_power.c194
-rw-r--r--kernel/trace/trace_printk.c270
-rw-r--r--kernel/trace/trace_sched_switch.c24
-rw-r--r--kernel/trace/trace_sched_wakeup.c96
-rw-r--r--kernel/trace/trace_selftest.c169
-rw-r--r--kernel/trace/trace_stack.c19
-rw-r--r--kernel/trace/trace_stat.c326
-rw-r--r--kernel/trace/trace_stat.h31
-rw-r--r--kernel/trace/trace_syscalls.c250
-rw-r--r--kernel/trace/trace_sysprof.c23
-rw-r--r--kernel/trace/trace_workqueue.c288
-rw-r--r--kernel/tracepoint.c7
-rw-r--r--kernel/tsacct.c6
-rw-r--r--kernel/user.c16
-rw-r--r--kernel/utsname_sysctl.c2
-rw-r--r--kernel/workqueue.c63
120 files changed, 16606 insertions, 4937 deletions
diff --git a/kernel/Makefile b/kernel/Makefile
index 9ef39e5b0211..63c697529ca1 100644
--- a/kernel/Makefile
+++ b/kernel/Makefile
@@ -93,6 +93,7 @@ obj-$(CONFIG_HAVE_GENERIC_DMA_COHERENT) += dma-coherent.o
93obj-$(CONFIG_FUNCTION_TRACER) += trace/ 93obj-$(CONFIG_FUNCTION_TRACER) += trace/
94obj-$(CONFIG_TRACING) += trace/ 94obj-$(CONFIG_TRACING) += trace/
95obj-$(CONFIG_SMP) += sched_cpupri.o 95obj-$(CONFIG_SMP) += sched_cpupri.o
96obj-$(CONFIG_SLOW_WORK) += slow-work.o
96obj-$(CONFIG_PERF_COUNTERS) += perf_counter.o 97obj-$(CONFIG_PERF_COUNTERS) += perf_counter.o
97 98
98ifneq ($(CONFIG_SCHED_OMIT_FRAME_POINTER),y) 99ifneq ($(CONFIG_SCHED_OMIT_FRAME_POINTER),y)
diff --git a/kernel/async.c b/kernel/async.c
index f565891f2c9b..968ef9457d4e 100644
--- a/kernel/async.c
+++ b/kernel/async.c
@@ -49,6 +49,7 @@ asynchronous and synchronous parts of the kernel.
49*/ 49*/
50 50
51#include <linux/async.h> 51#include <linux/async.h>
52#include <linux/bug.h>
52#include <linux/module.h> 53#include <linux/module.h>
53#include <linux/wait.h> 54#include <linux/wait.h>
54#include <linux/sched.h> 55#include <linux/sched.h>
@@ -387,20 +388,11 @@ static int async_manager_thread(void *unused)
387 388
388static int __init async_init(void) 389static int __init async_init(void)
389{ 390{
390 if (async_enabled) 391 async_enabled =
391 if (IS_ERR(kthread_run(async_manager_thread, NULL, 392 !IS_ERR(kthread_run(async_manager_thread, NULL, "async/mgr"));
392 "async/mgr")))
393 async_enabled = 0;
394 return 0;
395}
396 393
397static int __init setup_async(char *str) 394 WARN_ON(!async_enabled);
398{ 395 return 0;
399 async_enabled = 1;
400 return 1;
401} 396}
402 397
403__setup("fastboot", setup_async);
404
405
406core_initcall(async_init); 398core_initcall(async_init);
diff --git a/kernel/audit.c b/kernel/audit.c
index ce6d8ea3131e..9442c3533ba9 100644
--- a/kernel/audit.c
+++ b/kernel/audit.c
@@ -766,6 +766,9 @@ static int audit_receive_msg(struct sk_buff *skb, struct nlmsghdr *nlh)
766 766
767 audit_log_format(ab, " msg="); 767 audit_log_format(ab, " msg=");
768 size = nlmsg_len(nlh); 768 size = nlmsg_len(nlh);
769 if (size > 0 &&
770 ((unsigned char *)data)[size - 1] == '\0')
771 size--;
769 audit_log_n_untrustedstring(ab, data, size); 772 audit_log_n_untrustedstring(ab, data, size);
770 } 773 }
771 audit_set_pid(ab, pid); 774 audit_set_pid(ab, pid);
@@ -1382,7 +1385,7 @@ void audit_log_n_string(struct audit_buffer *ab, const char *string,
1382int audit_string_contains_control(const char *string, size_t len) 1385int audit_string_contains_control(const char *string, size_t len)
1383{ 1386{
1384 const unsigned char *p; 1387 const unsigned char *p;
1385 for (p = string; p < (const unsigned char *)string + len && *p; p++) { 1388 for (p = string; p < (const unsigned char *)string + len; p++) {
1386 if (*p == '"' || *p < 0x21 || *p > 0x7e) 1389 if (*p == '"' || *p < 0x21 || *p > 0x7e)
1387 return 1; 1390 return 1;
1388 } 1391 }
@@ -1437,13 +1440,13 @@ void audit_log_d_path(struct audit_buffer *ab, const char *prefix,
1437 /* We will allow 11 spaces for ' (deleted)' to be appended */ 1440 /* We will allow 11 spaces for ' (deleted)' to be appended */
1438 pathname = kmalloc(PATH_MAX+11, ab->gfp_mask); 1441 pathname = kmalloc(PATH_MAX+11, ab->gfp_mask);
1439 if (!pathname) { 1442 if (!pathname) {
1440 audit_log_format(ab, "<no memory>"); 1443 audit_log_string(ab, "<no_memory>");
1441 return; 1444 return;
1442 } 1445 }
1443 p = d_path(path, pathname, PATH_MAX+11); 1446 p = d_path(path, pathname, PATH_MAX+11);
1444 if (IS_ERR(p)) { /* Should never happen since we send PATH_MAX */ 1447 if (IS_ERR(p)) { /* Should never happen since we send PATH_MAX */
1445 /* FIXME: can we save some information here? */ 1448 /* FIXME: can we save some information here? */
1446 audit_log_format(ab, "<too long>"); 1449 audit_log_string(ab, "<too_long>");
1447 } else 1450 } else
1448 audit_log_untrustedstring(ab, p); 1451 audit_log_untrustedstring(ab, p);
1449 kfree(pathname); 1452 kfree(pathname);
diff --git a/kernel/audit_tree.c b/kernel/audit_tree.c
index 8ad9545b8db9..917ab9525568 100644
--- a/kernel/audit_tree.c
+++ b/kernel/audit_tree.c
@@ -385,6 +385,7 @@ static int tag_chunk(struct inode *inode, struct audit_tree *tree)
385 mutex_lock(&inode->inotify_mutex); 385 mutex_lock(&inode->inotify_mutex);
386 if (inotify_clone_watch(&old->watch, &chunk->watch) < 0) { 386 if (inotify_clone_watch(&old->watch, &chunk->watch) < 0) {
387 mutex_unlock(&inode->inotify_mutex); 387 mutex_unlock(&inode->inotify_mutex);
388 put_inotify_watch(&old->watch);
388 free_chunk(chunk); 389 free_chunk(chunk);
389 return -ENOSPC; 390 return -ENOSPC;
390 } 391 }
@@ -394,6 +395,7 @@ static int tag_chunk(struct inode *inode, struct audit_tree *tree)
394 chunk->dead = 1; 395 chunk->dead = 1;
395 inotify_evict_watch(&chunk->watch); 396 inotify_evict_watch(&chunk->watch);
396 mutex_unlock(&inode->inotify_mutex); 397 mutex_unlock(&inode->inotify_mutex);
398 put_inotify_watch(&old->watch);
397 put_inotify_watch(&chunk->watch); 399 put_inotify_watch(&chunk->watch);
398 return 0; 400 return 0;
399 } 401 }
diff --git a/kernel/auditfilter.c b/kernel/auditfilter.c
index fbf24d121d97..a6fe71fd5d1b 100644
--- a/kernel/auditfilter.c
+++ b/kernel/auditfilter.c
@@ -135,18 +135,18 @@ static void audit_remove_watch(struct audit_watch *watch)
135static inline void audit_free_rule(struct audit_entry *e) 135static inline void audit_free_rule(struct audit_entry *e)
136{ 136{
137 int i; 137 int i;
138 138 struct audit_krule *erule = &e->rule;
139 /* some rules don't have associated watches */ 139 /* some rules don't have associated watches */
140 if (e->rule.watch) 140 if (erule->watch)
141 audit_put_watch(e->rule.watch); 141 audit_put_watch(erule->watch);
142 if (e->rule.fields) 142 if (erule->fields)
143 for (i = 0; i < e->rule.field_count; i++) { 143 for (i = 0; i < erule->field_count; i++) {
144 struct audit_field *f = &e->rule.fields[i]; 144 struct audit_field *f = &erule->fields[i];
145 kfree(f->lsm_str); 145 kfree(f->lsm_str);
146 security_audit_rule_free(f->lsm_rule); 146 security_audit_rule_free(f->lsm_rule);
147 } 147 }
148 kfree(e->rule.fields); 148 kfree(erule->fields);
149 kfree(e->rule.filterkey); 149 kfree(erule->filterkey);
150 kfree(e); 150 kfree(e);
151} 151}
152 152
diff --git a/kernel/auditsc.c b/kernel/auditsc.c
index 8cbddff6c283..7d6ac7c1f414 100644
--- a/kernel/auditsc.c
+++ b/kernel/auditsc.c
@@ -66,6 +66,7 @@
66#include <linux/syscalls.h> 66#include <linux/syscalls.h>
67#include <linux/inotify.h> 67#include <linux/inotify.h>
68#include <linux/capability.h> 68#include <linux/capability.h>
69#include <linux/fs_struct.h>
69 70
70#include "audit.h" 71#include "audit.h"
71 72
@@ -328,6 +329,14 @@ static int audit_match_filetype(struct audit_context *ctx, int which)
328 */ 329 */
329 330
330#ifdef CONFIG_AUDIT_TREE 331#ifdef CONFIG_AUDIT_TREE
332static void audit_set_auditable(struct audit_context *ctx)
333{
334 if (!ctx->prio) {
335 ctx->prio = 1;
336 ctx->current_state = AUDIT_RECORD_CONTEXT;
337 }
338}
339
331static int put_tree_ref(struct audit_context *ctx, struct audit_chunk *chunk) 340static int put_tree_ref(struct audit_context *ctx, struct audit_chunk *chunk)
332{ 341{
333 struct audit_tree_refs *p = ctx->trees; 342 struct audit_tree_refs *p = ctx->trees;
@@ -741,17 +750,9 @@ void audit_filter_inodes(struct task_struct *tsk, struct audit_context *ctx)
741 rcu_read_unlock(); 750 rcu_read_unlock();
742} 751}
743 752
744static void audit_set_auditable(struct audit_context *ctx)
745{
746 if (!ctx->prio) {
747 ctx->prio = 1;
748 ctx->current_state = AUDIT_RECORD_CONTEXT;
749 }
750}
751
752static inline struct audit_context *audit_get_context(struct task_struct *tsk, 753static inline struct audit_context *audit_get_context(struct task_struct *tsk,
753 int return_valid, 754 int return_valid,
754 int return_code) 755 long return_code)
755{ 756{
756 struct audit_context *context = tsk->audit_context; 757 struct audit_context *context = tsk->audit_context;
757 758
@@ -1023,7 +1024,7 @@ static int audit_log_single_execve_arg(struct audit_context *context,
1023{ 1024{
1024 char arg_num_len_buf[12]; 1025 char arg_num_len_buf[12];
1025 const char __user *tmp_p = p; 1026 const char __user *tmp_p = p;
1026 /* how many digits are in arg_num? 3 is the length of a=\n */ 1027 /* how many digits are in arg_num? 3 is the length of " a=" */
1027 size_t arg_num_len = snprintf(arg_num_len_buf, 12, "%d", arg_num) + 3; 1028 size_t arg_num_len = snprintf(arg_num_len_buf, 12, "%d", arg_num) + 3;
1028 size_t len, len_left, to_send; 1029 size_t len, len_left, to_send;
1029 size_t max_execve_audit_len = MAX_EXECVE_AUDIT_LEN; 1030 size_t max_execve_audit_len = MAX_EXECVE_AUDIT_LEN;
@@ -1109,7 +1110,7 @@ static int audit_log_single_execve_arg(struct audit_context *context,
1109 * so we can be sure nothing was lost. 1110 * so we can be sure nothing was lost.
1110 */ 1111 */
1111 if ((i == 0) && (too_long)) 1112 if ((i == 0) && (too_long))
1112 audit_log_format(*ab, "a%d_len=%zu ", arg_num, 1113 audit_log_format(*ab, " a%d_len=%zu", arg_num,
1113 has_cntl ? 2*len : len); 1114 has_cntl ? 2*len : len);
1114 1115
1115 /* 1116 /*
@@ -1129,7 +1130,7 @@ static int audit_log_single_execve_arg(struct audit_context *context,
1129 buf[to_send] = '\0'; 1130 buf[to_send] = '\0';
1130 1131
1131 /* actually log it */ 1132 /* actually log it */
1132 audit_log_format(*ab, "a%d", arg_num); 1133 audit_log_format(*ab, " a%d", arg_num);
1133 if (too_long) 1134 if (too_long)
1134 audit_log_format(*ab, "[%d]", i); 1135 audit_log_format(*ab, "[%d]", i);
1135 audit_log_format(*ab, "="); 1136 audit_log_format(*ab, "=");
@@ -1137,7 +1138,6 @@ static int audit_log_single_execve_arg(struct audit_context *context,
1137 audit_log_n_hex(*ab, buf, to_send); 1138 audit_log_n_hex(*ab, buf, to_send);
1138 else 1139 else
1139 audit_log_format(*ab, "\"%s\"", buf); 1140 audit_log_format(*ab, "\"%s\"", buf);
1140 audit_log_format(*ab, "\n");
1141 1141
1142 p += to_send; 1142 p += to_send;
1143 len_left -= to_send; 1143 len_left -= to_send;
@@ -1165,7 +1165,7 @@ static void audit_log_execve_info(struct audit_context *context,
1165 1165
1166 p = (const char __user *)axi->mm->arg_start; 1166 p = (const char __user *)axi->mm->arg_start;
1167 1167
1168 audit_log_format(*ab, "argc=%d ", axi->argc); 1168 audit_log_format(*ab, "argc=%d", axi->argc);
1169 1169
1170 /* 1170 /*
1171 * we need some kernel buffer to hold the userspace args. Just 1171 * we need some kernel buffer to hold the userspace args. Just
@@ -1478,7 +1478,7 @@ static void audit_log_exit(struct audit_context *context, struct task_struct *ts
1478 case 0: 1478 case 0:
1479 /* name was specified as a relative path and the 1479 /* name was specified as a relative path and the
1480 * directory component is the cwd */ 1480 * directory component is the cwd */
1481 audit_log_d_path(ab, " name=", &context->pwd); 1481 audit_log_d_path(ab, "name=", &context->pwd);
1482 break; 1482 break;
1483 default: 1483 default:
1484 /* log the name's directory component */ 1484 /* log the name's directory component */
@@ -2149,7 +2149,7 @@ int audit_set_loginuid(struct task_struct *task, uid_t loginuid)
2149 * __audit_mq_open - record audit data for a POSIX MQ open 2149 * __audit_mq_open - record audit data for a POSIX MQ open
2150 * @oflag: open flag 2150 * @oflag: open flag
2151 * @mode: mode bits 2151 * @mode: mode bits
2152 * @u_attr: queue attributes 2152 * @attr: queue attributes
2153 * 2153 *
2154 */ 2154 */
2155void __audit_mq_open(int oflag, mode_t mode, struct mq_attr *attr) 2155void __audit_mq_open(int oflag, mode_t mode, struct mq_attr *attr)
@@ -2196,7 +2196,7 @@ void __audit_mq_sendrecv(mqd_t mqdes, size_t msg_len, unsigned int msg_prio,
2196/** 2196/**
2197 * __audit_mq_notify - record audit data for a POSIX MQ notify 2197 * __audit_mq_notify - record audit data for a POSIX MQ notify
2198 * @mqdes: MQ descriptor 2198 * @mqdes: MQ descriptor
2199 * @u_notification: Notification event 2199 * @notification: Notification event
2200 * 2200 *
2201 */ 2201 */
2202 2202
diff --git a/kernel/cgroup.c b/kernel/cgroup.c
index 9edb5c4b79b4..382109b5baeb 100644
--- a/kernel/cgroup.c
+++ b/kernel/cgroup.c
@@ -94,7 +94,6 @@ struct cgroupfs_root {
94 char release_agent_path[PATH_MAX]; 94 char release_agent_path[PATH_MAX];
95}; 95};
96 96
97
98/* 97/*
99 * The "rootnode" hierarchy is the "dummy hierarchy", reserved for the 98 * The "rootnode" hierarchy is the "dummy hierarchy", reserved for the
100 * subsystems that are otherwise unattached - it never has more than a 99 * subsystems that are otherwise unattached - it never has more than a
@@ -102,6 +101,39 @@ struct cgroupfs_root {
102 */ 101 */
103static struct cgroupfs_root rootnode; 102static struct cgroupfs_root rootnode;
104 103
104/*
105 * CSS ID -- ID per subsys's Cgroup Subsys State(CSS). used only when
106 * cgroup_subsys->use_id != 0.
107 */
108#define CSS_ID_MAX (65535)
109struct css_id {
110 /*
111 * The css to which this ID points. This pointer is set to valid value
112 * after cgroup is populated. If cgroup is removed, this will be NULL.
113 * This pointer is expected to be RCU-safe because destroy()
114 * is called after synchronize_rcu(). But for safe use, css_is_removed()
115 * css_tryget() should be used for avoiding race.
116 */
117 struct cgroup_subsys_state *css;
118 /*
119 * ID of this css.
120 */
121 unsigned short id;
122 /*
123 * Depth in hierarchy which this ID belongs to.
124 */
125 unsigned short depth;
126 /*
127 * ID is freed by RCU. (and lookup routine is RCU safe.)
128 */
129 struct rcu_head rcu_head;
130 /*
131 * Hierarchy of CSS ID belongs to.
132 */
133 unsigned short stack[0]; /* Array of Length (depth+1) */
134};
135
136
105/* The list of hierarchy roots */ 137/* The list of hierarchy roots */
106 138
107static LIST_HEAD(roots); 139static LIST_HEAD(roots);
@@ -185,6 +217,8 @@ struct cg_cgroup_link {
185static struct css_set init_css_set; 217static struct css_set init_css_set;
186static struct cg_cgroup_link init_css_set_link; 218static struct cg_cgroup_link init_css_set_link;
187 219
220static int cgroup_subsys_init_idr(struct cgroup_subsys *ss);
221
188/* css_set_lock protects the list of css_set objects, and the 222/* css_set_lock protects the list of css_set objects, and the
189 * chain of tasks off each css_set. Nests outside task->alloc_lock 223 * chain of tasks off each css_set. Nests outside task->alloc_lock
190 * due to cgroup_iter_start() */ 224 * due to cgroup_iter_start() */
@@ -567,6 +601,9 @@ static struct backing_dev_info cgroup_backing_dev_info = {
567 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK, 601 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK,
568}; 602};
569 603
604static int alloc_css_id(struct cgroup_subsys *ss,
605 struct cgroup *parent, struct cgroup *child);
606
570static struct inode *cgroup_new_inode(mode_t mode, struct super_block *sb) 607static struct inode *cgroup_new_inode(mode_t mode, struct super_block *sb)
571{ 608{
572 struct inode *inode = new_inode(sb); 609 struct inode *inode = new_inode(sb);
@@ -585,13 +622,18 @@ static struct inode *cgroup_new_inode(mode_t mode, struct super_block *sb)
585 * Call subsys's pre_destroy handler. 622 * Call subsys's pre_destroy handler.
586 * This is called before css refcnt check. 623 * This is called before css refcnt check.
587 */ 624 */
588static void cgroup_call_pre_destroy(struct cgroup *cgrp) 625static int cgroup_call_pre_destroy(struct cgroup *cgrp)
589{ 626{
590 struct cgroup_subsys *ss; 627 struct cgroup_subsys *ss;
628 int ret = 0;
629
591 for_each_subsys(cgrp->root, ss) 630 for_each_subsys(cgrp->root, ss)
592 if (ss->pre_destroy) 631 if (ss->pre_destroy) {
593 ss->pre_destroy(ss, cgrp); 632 ret = ss->pre_destroy(ss, cgrp);
594 return; 633 if (ret)
634 break;
635 }
636 return ret;
595} 637}
596 638
597static void free_cgroup_rcu(struct rcu_head *obj) 639static void free_cgroup_rcu(struct rcu_head *obj)
@@ -685,6 +727,22 @@ static void cgroup_d_remove_dir(struct dentry *dentry)
685 remove_dir(dentry); 727 remove_dir(dentry);
686} 728}
687 729
730/*
731 * A queue for waiters to do rmdir() cgroup. A tasks will sleep when
732 * cgroup->count == 0 && list_empty(&cgroup->children) && subsys has some
733 * reference to css->refcnt. In general, this refcnt is expected to goes down
734 * to zero, soon.
735 *
736 * CGRP_WAIT_ON_RMDIR flag is modified under cgroup's inode->i_mutex;
737 */
738DECLARE_WAIT_QUEUE_HEAD(cgroup_rmdir_waitq);
739
740static void cgroup_wakeup_rmdir_waiters(const struct cgroup *cgrp)
741{
742 if (unlikely(test_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags)))
743 wake_up_all(&cgroup_rmdir_waitq);
744}
745
688static int rebind_subsystems(struct cgroupfs_root *root, 746static int rebind_subsystems(struct cgroupfs_root *root,
689 unsigned long final_bits) 747 unsigned long final_bits)
690{ 748{
@@ -857,16 +915,16 @@ static int cgroup_remount(struct super_block *sb, int *flags, char *data)
857 } 915 }
858 916
859 ret = rebind_subsystems(root, opts.subsys_bits); 917 ret = rebind_subsystems(root, opts.subsys_bits);
918 if (ret)
919 goto out_unlock;
860 920
861 /* (re)populate subsystem files */ 921 /* (re)populate subsystem files */
862 if (!ret) 922 cgroup_populate_dir(cgrp);
863 cgroup_populate_dir(cgrp);
864 923
865 if (opts.release_agent) 924 if (opts.release_agent)
866 strcpy(root->release_agent_path, opts.release_agent); 925 strcpy(root->release_agent_path, opts.release_agent);
867 out_unlock: 926 out_unlock:
868 if (opts.release_agent) 927 kfree(opts.release_agent);
869 kfree(opts.release_agent);
870 mutex_unlock(&cgroup_mutex); 928 mutex_unlock(&cgroup_mutex);
871 mutex_unlock(&cgrp->dentry->d_inode->i_mutex); 929 mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
872 return ret; 930 return ret;
@@ -969,15 +1027,13 @@ static int cgroup_get_sb(struct file_system_type *fs_type,
969 /* First find the desired set of subsystems */ 1027 /* First find the desired set of subsystems */
970 ret = parse_cgroupfs_options(data, &opts); 1028 ret = parse_cgroupfs_options(data, &opts);
971 if (ret) { 1029 if (ret) {
972 if (opts.release_agent) 1030 kfree(opts.release_agent);
973 kfree(opts.release_agent);
974 return ret; 1031 return ret;
975 } 1032 }
976 1033
977 root = kzalloc(sizeof(*root), GFP_KERNEL); 1034 root = kzalloc(sizeof(*root), GFP_KERNEL);
978 if (!root) { 1035 if (!root) {
979 if (opts.release_agent) 1036 kfree(opts.release_agent);
980 kfree(opts.release_agent);
981 return -ENOMEM; 1037 return -ENOMEM;
982 } 1038 }
983 1039
@@ -1071,7 +1127,8 @@ static int cgroup_get_sb(struct file_system_type *fs_type,
1071 mutex_unlock(&cgroup_mutex); 1127 mutex_unlock(&cgroup_mutex);
1072 } 1128 }
1073 1129
1074 return simple_set_mnt(mnt, sb); 1130 simple_set_mnt(mnt, sb);
1131 return 0;
1075 1132
1076 free_cg_links: 1133 free_cg_links:
1077 free_cg_links(&tmp_cg_links); 1134 free_cg_links(&tmp_cg_links);
@@ -1279,6 +1336,12 @@ int cgroup_attach_task(struct cgroup *cgrp, struct task_struct *tsk)
1279 set_bit(CGRP_RELEASABLE, &oldcgrp->flags); 1336 set_bit(CGRP_RELEASABLE, &oldcgrp->flags);
1280 synchronize_rcu(); 1337 synchronize_rcu();
1281 put_css_set(cg); 1338 put_css_set(cg);
1339
1340 /*
1341 * wake up rmdir() waiter. the rmdir should fail since the cgroup
1342 * is no longer empty.
1343 */
1344 cgroup_wakeup_rmdir_waiters(cgrp);
1282 return 0; 1345 return 0;
1283} 1346}
1284 1347
@@ -1624,10 +1687,10 @@ static struct inode_operations cgroup_dir_inode_operations = {
1624 .rename = cgroup_rename, 1687 .rename = cgroup_rename,
1625}; 1688};
1626 1689
1627static int cgroup_create_file(struct dentry *dentry, int mode, 1690static int cgroup_create_file(struct dentry *dentry, mode_t mode,
1628 struct super_block *sb) 1691 struct super_block *sb)
1629{ 1692{
1630 static struct dentry_operations cgroup_dops = { 1693 static const struct dentry_operations cgroup_dops = {
1631 .d_iput = cgroup_diput, 1694 .d_iput = cgroup_diput,
1632 }; 1695 };
1633 1696
@@ -1670,7 +1733,7 @@ static int cgroup_create_file(struct dentry *dentry, int mode,
1670 * @mode: mode to set on new directory. 1733 * @mode: mode to set on new directory.
1671 */ 1734 */
1672static int cgroup_create_dir(struct cgroup *cgrp, struct dentry *dentry, 1735static int cgroup_create_dir(struct cgroup *cgrp, struct dentry *dentry,
1673 int mode) 1736 mode_t mode)
1674{ 1737{
1675 struct dentry *parent; 1738 struct dentry *parent;
1676 int error = 0; 1739 int error = 0;
@@ -1688,6 +1751,33 @@ static int cgroup_create_dir(struct cgroup *cgrp, struct dentry *dentry,
1688 return error; 1751 return error;
1689} 1752}
1690 1753
1754/**
1755 * cgroup_file_mode - deduce file mode of a control file
1756 * @cft: the control file in question
1757 *
1758 * returns cft->mode if ->mode is not 0
1759 * returns S_IRUGO|S_IWUSR if it has both a read and a write handler
1760 * returns S_IRUGO if it has only a read handler
1761 * returns S_IWUSR if it has only a write hander
1762 */
1763static mode_t cgroup_file_mode(const struct cftype *cft)
1764{
1765 mode_t mode = 0;
1766
1767 if (cft->mode)
1768 return cft->mode;
1769
1770 if (cft->read || cft->read_u64 || cft->read_s64 ||
1771 cft->read_map || cft->read_seq_string)
1772 mode |= S_IRUGO;
1773
1774 if (cft->write || cft->write_u64 || cft->write_s64 ||
1775 cft->write_string || cft->trigger)
1776 mode |= S_IWUSR;
1777
1778 return mode;
1779}
1780
1691int cgroup_add_file(struct cgroup *cgrp, 1781int cgroup_add_file(struct cgroup *cgrp,
1692 struct cgroup_subsys *subsys, 1782 struct cgroup_subsys *subsys,
1693 const struct cftype *cft) 1783 const struct cftype *cft)
@@ -1695,6 +1785,7 @@ int cgroup_add_file(struct cgroup *cgrp,
1695 struct dentry *dir = cgrp->dentry; 1785 struct dentry *dir = cgrp->dentry;
1696 struct dentry *dentry; 1786 struct dentry *dentry;
1697 int error; 1787 int error;
1788 mode_t mode;
1698 1789
1699 char name[MAX_CGROUP_TYPE_NAMELEN + MAX_CFTYPE_NAME + 2] = { 0 }; 1790 char name[MAX_CGROUP_TYPE_NAMELEN + MAX_CFTYPE_NAME + 2] = { 0 };
1700 if (subsys && !test_bit(ROOT_NOPREFIX, &cgrp->root->flags)) { 1791 if (subsys && !test_bit(ROOT_NOPREFIX, &cgrp->root->flags)) {
@@ -1705,7 +1796,8 @@ int cgroup_add_file(struct cgroup *cgrp,
1705 BUG_ON(!mutex_is_locked(&dir->d_inode->i_mutex)); 1796 BUG_ON(!mutex_is_locked(&dir->d_inode->i_mutex));
1706 dentry = lookup_one_len(name, dir, strlen(name)); 1797 dentry = lookup_one_len(name, dir, strlen(name));
1707 if (!IS_ERR(dentry)) { 1798 if (!IS_ERR(dentry)) {
1708 error = cgroup_create_file(dentry, 0644 | S_IFREG, 1799 mode = cgroup_file_mode(cft);
1800 error = cgroup_create_file(dentry, mode | S_IFREG,
1709 cgrp->root->sb); 1801 cgrp->root->sb);
1710 if (!error) 1802 if (!error)
1711 dentry->d_fsdata = (void *)cft; 1803 dentry->d_fsdata = (void *)cft;
@@ -2287,6 +2379,7 @@ static struct cftype files[] = {
2287 .write_u64 = cgroup_tasks_write, 2379 .write_u64 = cgroup_tasks_write,
2288 .release = cgroup_tasks_release, 2380 .release = cgroup_tasks_release,
2289 .private = FILE_TASKLIST, 2381 .private = FILE_TASKLIST,
2382 .mode = S_IRUGO | S_IWUSR,
2290 }, 2383 },
2291 2384
2292 { 2385 {
@@ -2326,6 +2419,17 @@ static int cgroup_populate_dir(struct cgroup *cgrp)
2326 if (ss->populate && (err = ss->populate(ss, cgrp)) < 0) 2419 if (ss->populate && (err = ss->populate(ss, cgrp)) < 0)
2327 return err; 2420 return err;
2328 } 2421 }
2422 /* This cgroup is ready now */
2423 for_each_subsys(cgrp->root, ss) {
2424 struct cgroup_subsys_state *css = cgrp->subsys[ss->subsys_id];
2425 /*
2426 * Update id->css pointer and make this css visible from
2427 * CSS ID functions. This pointer will be dereferened
2428 * from RCU-read-side without locks.
2429 */
2430 if (css->id)
2431 rcu_assign_pointer(css->id->css, css);
2432 }
2329 2433
2330 return 0; 2434 return 0;
2331} 2435}
@@ -2337,6 +2441,7 @@ static void init_cgroup_css(struct cgroup_subsys_state *css,
2337 css->cgroup = cgrp; 2441 css->cgroup = cgrp;
2338 atomic_set(&css->refcnt, 1); 2442 atomic_set(&css->refcnt, 1);
2339 css->flags = 0; 2443 css->flags = 0;
2444 css->id = NULL;
2340 if (cgrp == dummytop) 2445 if (cgrp == dummytop)
2341 set_bit(CSS_ROOT, &css->flags); 2446 set_bit(CSS_ROOT, &css->flags);
2342 BUG_ON(cgrp->subsys[ss->subsys_id]); 2447 BUG_ON(cgrp->subsys[ss->subsys_id]);
@@ -2375,7 +2480,7 @@ static void cgroup_unlock_hierarchy(struct cgroupfs_root *root)
2375 * Must be called with the mutex on the parent inode held 2480 * Must be called with the mutex on the parent inode held
2376 */ 2481 */
2377static long cgroup_create(struct cgroup *parent, struct dentry *dentry, 2482static long cgroup_create(struct cgroup *parent, struct dentry *dentry,
2378 int mode) 2483 mode_t mode)
2379{ 2484{
2380 struct cgroup *cgrp; 2485 struct cgroup *cgrp;
2381 struct cgroupfs_root *root = parent->root; 2486 struct cgroupfs_root *root = parent->root;
@@ -2412,6 +2517,10 @@ static long cgroup_create(struct cgroup *parent, struct dentry *dentry,
2412 goto err_destroy; 2517 goto err_destroy;
2413 } 2518 }
2414 init_cgroup_css(css, ss, cgrp); 2519 init_cgroup_css(css, ss, cgrp);
2520 if (ss->use_id)
2521 if (alloc_css_id(ss, parent, cgrp))
2522 goto err_destroy;
2523 /* At error, ->destroy() callback has to free assigned ID. */
2415 } 2524 }
2416 2525
2417 cgroup_lock_hierarchy(root); 2526 cgroup_lock_hierarchy(root);
@@ -2554,9 +2663,11 @@ static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry)
2554 struct cgroup *cgrp = dentry->d_fsdata; 2663 struct cgroup *cgrp = dentry->d_fsdata;
2555 struct dentry *d; 2664 struct dentry *d;
2556 struct cgroup *parent; 2665 struct cgroup *parent;
2666 DEFINE_WAIT(wait);
2667 int ret;
2557 2668
2558 /* the vfs holds both inode->i_mutex already */ 2669 /* the vfs holds both inode->i_mutex already */
2559 2670again:
2560 mutex_lock(&cgroup_mutex); 2671 mutex_lock(&cgroup_mutex);
2561 if (atomic_read(&cgrp->count) != 0) { 2672 if (atomic_read(&cgrp->count) != 0) {
2562 mutex_unlock(&cgroup_mutex); 2673 mutex_unlock(&cgroup_mutex);
@@ -2572,17 +2683,39 @@ static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry)
2572 * Call pre_destroy handlers of subsys. Notify subsystems 2683 * Call pre_destroy handlers of subsys. Notify subsystems
2573 * that rmdir() request comes. 2684 * that rmdir() request comes.
2574 */ 2685 */
2575 cgroup_call_pre_destroy(cgrp); 2686 ret = cgroup_call_pre_destroy(cgrp);
2687 if (ret)
2688 return ret;
2576 2689
2577 mutex_lock(&cgroup_mutex); 2690 mutex_lock(&cgroup_mutex);
2578 parent = cgrp->parent; 2691 parent = cgrp->parent;
2579 2692 if (atomic_read(&cgrp->count) || !list_empty(&cgrp->children)) {
2580 if (atomic_read(&cgrp->count)
2581 || !list_empty(&cgrp->children)
2582 || !cgroup_clear_css_refs(cgrp)) {
2583 mutex_unlock(&cgroup_mutex); 2693 mutex_unlock(&cgroup_mutex);
2584 return -EBUSY; 2694 return -EBUSY;
2585 } 2695 }
2696 /*
2697 * css_put/get is provided for subsys to grab refcnt to css. In typical
2698 * case, subsystem has no reference after pre_destroy(). But, under
2699 * hierarchy management, some *temporal* refcnt can be hold.
2700 * To avoid returning -EBUSY to a user, waitqueue is used. If subsys
2701 * is really busy, it should return -EBUSY at pre_destroy(). wake_up
2702 * is called when css_put() is called and refcnt goes down to 0.
2703 */
2704 set_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
2705 prepare_to_wait(&cgroup_rmdir_waitq, &wait, TASK_INTERRUPTIBLE);
2706
2707 if (!cgroup_clear_css_refs(cgrp)) {
2708 mutex_unlock(&cgroup_mutex);
2709 schedule();
2710 finish_wait(&cgroup_rmdir_waitq, &wait);
2711 clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
2712 if (signal_pending(current))
2713 return -EINTR;
2714 goto again;
2715 }
2716 /* NO css_tryget() can success after here. */
2717 finish_wait(&cgroup_rmdir_waitq, &wait);
2718 clear_bit(CGRP_WAIT_ON_RMDIR, &cgrp->flags);
2586 2719
2587 spin_lock(&release_list_lock); 2720 spin_lock(&release_list_lock);
2588 set_bit(CGRP_REMOVED, &cgrp->flags); 2721 set_bit(CGRP_REMOVED, &cgrp->flags);
@@ -2707,6 +2840,8 @@ int __init cgroup_init(void)
2707 struct cgroup_subsys *ss = subsys[i]; 2840 struct cgroup_subsys *ss = subsys[i];
2708 if (!ss->early_init) 2841 if (!ss->early_init)
2709 cgroup_init_subsys(ss); 2842 cgroup_init_subsys(ss);
2843 if (ss->use_id)
2844 cgroup_subsys_init_idr(ss);
2710 } 2845 }
2711 2846
2712 /* Add init_css_set to the hash table */ 2847 /* Add init_css_set to the hash table */
@@ -3083,18 +3218,19 @@ int cgroup_clone(struct task_struct *tsk, struct cgroup_subsys *subsys,
3083} 3218}
3084 3219
3085/** 3220/**
3086 * cgroup_is_descendant - see if @cgrp is a descendant of current task's cgrp 3221 * cgroup_is_descendant - see if @cgrp is a descendant of @task's cgrp
3087 * @cgrp: the cgroup in question 3222 * @cgrp: the cgroup in question
3223 * @task: the task in question
3088 * 3224 *
3089 * See if @cgrp is a descendant of the current task's cgroup in 3225 * See if @cgrp is a descendant of @task's cgroup in the appropriate
3090 * the appropriate hierarchy. 3226 * hierarchy.
3091 * 3227 *
3092 * If we are sending in dummytop, then presumably we are creating 3228 * If we are sending in dummytop, then presumably we are creating
3093 * the top cgroup in the subsystem. 3229 * the top cgroup in the subsystem.
3094 * 3230 *
3095 * Called only by the ns (nsproxy) cgroup. 3231 * Called only by the ns (nsproxy) cgroup.
3096 */ 3232 */
3097int cgroup_is_descendant(const struct cgroup *cgrp) 3233int cgroup_is_descendant(const struct cgroup *cgrp, struct task_struct *task)
3098{ 3234{
3099 int ret; 3235 int ret;
3100 struct cgroup *target; 3236 struct cgroup *target;
@@ -3104,7 +3240,7 @@ int cgroup_is_descendant(const struct cgroup *cgrp)
3104 return 1; 3240 return 1;
3105 3241
3106 get_first_subsys(cgrp, NULL, &subsys_id); 3242 get_first_subsys(cgrp, NULL, &subsys_id);
3107 target = task_cgroup(current, subsys_id); 3243 target = task_cgroup(task, subsys_id);
3108 while (cgrp != target && cgrp!= cgrp->top_cgroup) 3244 while (cgrp != target && cgrp!= cgrp->top_cgroup)
3109 cgrp = cgrp->parent; 3245 cgrp = cgrp->parent;
3110 ret = (cgrp == target); 3246 ret = (cgrp == target);
@@ -3137,10 +3273,12 @@ void __css_put(struct cgroup_subsys_state *css)
3137{ 3273{
3138 struct cgroup *cgrp = css->cgroup; 3274 struct cgroup *cgrp = css->cgroup;
3139 rcu_read_lock(); 3275 rcu_read_lock();
3140 if ((atomic_dec_return(&css->refcnt) == 1) && 3276 if (atomic_dec_return(&css->refcnt) == 1) {
3141 notify_on_release(cgrp)) { 3277 if (notify_on_release(cgrp)) {
3142 set_bit(CGRP_RELEASABLE, &cgrp->flags); 3278 set_bit(CGRP_RELEASABLE, &cgrp->flags);
3143 check_for_release(cgrp); 3279 check_for_release(cgrp);
3280 }
3281 cgroup_wakeup_rmdir_waiters(cgrp);
3144 } 3282 }
3145 rcu_read_unlock(); 3283 rcu_read_unlock();
3146} 3284}
@@ -3240,3 +3378,232 @@ static int __init cgroup_disable(char *str)
3240 return 1; 3378 return 1;
3241} 3379}
3242__setup("cgroup_disable=", cgroup_disable); 3380__setup("cgroup_disable=", cgroup_disable);
3381
3382/*
3383 * Functons for CSS ID.
3384 */
3385
3386/*
3387 *To get ID other than 0, this should be called when !cgroup_is_removed().
3388 */
3389unsigned short css_id(struct cgroup_subsys_state *css)
3390{
3391 struct css_id *cssid = rcu_dereference(css->id);
3392
3393 if (cssid)
3394 return cssid->id;
3395 return 0;
3396}
3397
3398unsigned short css_depth(struct cgroup_subsys_state *css)
3399{
3400 struct css_id *cssid = rcu_dereference(css->id);
3401
3402 if (cssid)
3403 return cssid->depth;
3404 return 0;
3405}
3406
3407bool css_is_ancestor(struct cgroup_subsys_state *child,
3408 const struct cgroup_subsys_state *root)
3409{
3410 struct css_id *child_id = rcu_dereference(child->id);
3411 struct css_id *root_id = rcu_dereference(root->id);
3412
3413 if (!child_id || !root_id || (child_id->depth < root_id->depth))
3414 return false;
3415 return child_id->stack[root_id->depth] == root_id->id;
3416}
3417
3418static void __free_css_id_cb(struct rcu_head *head)
3419{
3420 struct css_id *id;
3421
3422 id = container_of(head, struct css_id, rcu_head);
3423 kfree(id);
3424}
3425
3426void free_css_id(struct cgroup_subsys *ss, struct cgroup_subsys_state *css)
3427{
3428 struct css_id *id = css->id;
3429 /* When this is called before css_id initialization, id can be NULL */
3430 if (!id)
3431 return;
3432
3433 BUG_ON(!ss->use_id);
3434
3435 rcu_assign_pointer(id->css, NULL);
3436 rcu_assign_pointer(css->id, NULL);
3437 spin_lock(&ss->id_lock);
3438 idr_remove(&ss->idr, id->id);
3439 spin_unlock(&ss->id_lock);
3440 call_rcu(&id->rcu_head, __free_css_id_cb);
3441}
3442
3443/*
3444 * This is called by init or create(). Then, calls to this function are
3445 * always serialized (By cgroup_mutex() at create()).
3446 */
3447
3448static struct css_id *get_new_cssid(struct cgroup_subsys *ss, int depth)
3449{
3450 struct css_id *newid;
3451 int myid, error, size;
3452
3453 BUG_ON(!ss->use_id);
3454
3455 size = sizeof(*newid) + sizeof(unsigned short) * (depth + 1);
3456 newid = kzalloc(size, GFP_KERNEL);
3457 if (!newid)
3458 return ERR_PTR(-ENOMEM);
3459 /* get id */
3460 if (unlikely(!idr_pre_get(&ss->idr, GFP_KERNEL))) {
3461 error = -ENOMEM;
3462 goto err_out;
3463 }
3464 spin_lock(&ss->id_lock);
3465 /* Don't use 0. allocates an ID of 1-65535 */
3466 error = idr_get_new_above(&ss->idr, newid, 1, &myid);
3467 spin_unlock(&ss->id_lock);
3468
3469 /* Returns error when there are no free spaces for new ID.*/
3470 if (error) {
3471 error = -ENOSPC;
3472 goto err_out;
3473 }
3474 if (myid > CSS_ID_MAX)
3475 goto remove_idr;
3476
3477 newid->id = myid;
3478 newid->depth = depth;
3479 return newid;
3480remove_idr:
3481 error = -ENOSPC;
3482 spin_lock(&ss->id_lock);
3483 idr_remove(&ss->idr, myid);
3484 spin_unlock(&ss->id_lock);
3485err_out:
3486 kfree(newid);
3487 return ERR_PTR(error);
3488
3489}
3490
3491static int __init cgroup_subsys_init_idr(struct cgroup_subsys *ss)
3492{
3493 struct css_id *newid;
3494 struct cgroup_subsys_state *rootcss;
3495
3496 spin_lock_init(&ss->id_lock);
3497 idr_init(&ss->idr);
3498
3499 rootcss = init_css_set.subsys[ss->subsys_id];
3500 newid = get_new_cssid(ss, 0);
3501 if (IS_ERR(newid))
3502 return PTR_ERR(newid);
3503
3504 newid->stack[0] = newid->id;
3505 newid->css = rootcss;
3506 rootcss->id = newid;
3507 return 0;
3508}
3509
3510static int alloc_css_id(struct cgroup_subsys *ss, struct cgroup *parent,
3511 struct cgroup *child)
3512{
3513 int subsys_id, i, depth = 0;
3514 struct cgroup_subsys_state *parent_css, *child_css;
3515 struct css_id *child_id, *parent_id = NULL;
3516
3517 subsys_id = ss->subsys_id;
3518 parent_css = parent->subsys[subsys_id];
3519 child_css = child->subsys[subsys_id];
3520 depth = css_depth(parent_css) + 1;
3521 parent_id = parent_css->id;
3522
3523 child_id = get_new_cssid(ss, depth);
3524 if (IS_ERR(child_id))
3525 return PTR_ERR(child_id);
3526
3527 for (i = 0; i < depth; i++)
3528 child_id->stack[i] = parent_id->stack[i];
3529 child_id->stack[depth] = child_id->id;
3530 /*
3531 * child_id->css pointer will be set after this cgroup is available
3532 * see cgroup_populate_dir()
3533 */
3534 rcu_assign_pointer(child_css->id, child_id);
3535
3536 return 0;
3537}
3538
3539/**
3540 * css_lookup - lookup css by id
3541 * @ss: cgroup subsys to be looked into.
3542 * @id: the id
3543 *
3544 * Returns pointer to cgroup_subsys_state if there is valid one with id.
3545 * NULL if not. Should be called under rcu_read_lock()
3546 */
3547struct cgroup_subsys_state *css_lookup(struct cgroup_subsys *ss, int id)
3548{
3549 struct css_id *cssid = NULL;
3550
3551 BUG_ON(!ss->use_id);
3552 cssid = idr_find(&ss->idr, id);
3553
3554 if (unlikely(!cssid))
3555 return NULL;
3556
3557 return rcu_dereference(cssid->css);
3558}
3559
3560/**
3561 * css_get_next - lookup next cgroup under specified hierarchy.
3562 * @ss: pointer to subsystem
3563 * @id: current position of iteration.
3564 * @root: pointer to css. search tree under this.
3565 * @foundid: position of found object.
3566 *
3567 * Search next css under the specified hierarchy of rootid. Calling under
3568 * rcu_read_lock() is necessary. Returns NULL if it reaches the end.
3569 */
3570struct cgroup_subsys_state *
3571css_get_next(struct cgroup_subsys *ss, int id,
3572 struct cgroup_subsys_state *root, int *foundid)
3573{
3574 struct cgroup_subsys_state *ret = NULL;
3575 struct css_id *tmp;
3576 int tmpid;
3577 int rootid = css_id(root);
3578 int depth = css_depth(root);
3579
3580 if (!rootid)
3581 return NULL;
3582
3583 BUG_ON(!ss->use_id);
3584 /* fill start point for scan */
3585 tmpid = id;
3586 while (1) {
3587 /*
3588 * scan next entry from bitmap(tree), tmpid is updated after
3589 * idr_get_next().
3590 */
3591 spin_lock(&ss->id_lock);
3592 tmp = idr_get_next(&ss->idr, &tmpid);
3593 spin_unlock(&ss->id_lock);
3594
3595 if (!tmp)
3596 break;
3597 if (tmp->depth >= depth && tmp->stack[depth] == rootid) {
3598 ret = rcu_dereference(tmp->css);
3599 if (ret) {
3600 *foundid = tmpid;
3601 break;
3602 }
3603 }
3604 /* continue to scan from next id */
3605 tmpid = tmpid + 1;
3606 }
3607 return ret;
3608}
3609
diff --git a/kernel/cgroup_debug.c b/kernel/cgroup_debug.c
index daca6209202d..0c92d797baa6 100644
--- a/kernel/cgroup_debug.c
+++ b/kernel/cgroup_debug.c
@@ -40,9 +40,7 @@ static u64 taskcount_read(struct cgroup *cont, struct cftype *cft)
40{ 40{
41 u64 count; 41 u64 count;
42 42
43 cgroup_lock();
44 count = cgroup_task_count(cont); 43 count = cgroup_task_count(cont);
45 cgroup_unlock();
46 return count; 44 return count;
47} 45}
48 46
diff --git a/kernel/cpu.c b/kernel/cpu.c
index 79e40f00dcb8..395b6974dc8d 100644
--- a/kernel/cpu.c
+++ b/kernel/cpu.c
@@ -281,7 +281,7 @@ int __ref cpu_down(unsigned int cpu)
281 goto out; 281 goto out;
282 } 282 }
283 283
284 cpu_clear(cpu, cpu_active_map); 284 set_cpu_active(cpu, false);
285 285
286 /* 286 /*
287 * Make sure the all cpus did the reschedule and are not 287 * Make sure the all cpus did the reschedule and are not
@@ -296,7 +296,7 @@ int __ref cpu_down(unsigned int cpu)
296 err = _cpu_down(cpu, 0); 296 err = _cpu_down(cpu, 0);
297 297
298 if (cpu_online(cpu)) 298 if (cpu_online(cpu))
299 cpu_set(cpu, cpu_active_map); 299 set_cpu_active(cpu, true);
300 300
301out: 301out:
302 cpu_maps_update_done(); 302 cpu_maps_update_done();
@@ -333,7 +333,7 @@ static int __cpuinit _cpu_up(unsigned int cpu, int tasks_frozen)
333 goto out_notify; 333 goto out_notify;
334 BUG_ON(!cpu_online(cpu)); 334 BUG_ON(!cpu_online(cpu));
335 335
336 cpu_set(cpu, cpu_active_map); 336 set_cpu_active(cpu, true);
337 337
338 /* Now call notifier in preparation. */ 338 /* Now call notifier in preparation. */
339 raw_notifier_call_chain(&cpu_chain, CPU_ONLINE | mod, hcpu); 339 raw_notifier_call_chain(&cpu_chain, CPU_ONLINE | mod, hcpu);
diff --git a/kernel/cpuset.c b/kernel/cpuset.c
index f76db9dcaa05..026faccca869 100644
--- a/kernel/cpuset.c
+++ b/kernel/cpuset.c
@@ -128,10 +128,6 @@ static inline struct cpuset *task_cs(struct task_struct *task)
128 return container_of(task_subsys_state(task, cpuset_subsys_id), 128 return container_of(task_subsys_state(task, cpuset_subsys_id),
129 struct cpuset, css); 129 struct cpuset, css);
130} 130}
131struct cpuset_hotplug_scanner {
132 struct cgroup_scanner scan;
133 struct cgroup *to;
134};
135 131
136/* bits in struct cpuset flags field */ 132/* bits in struct cpuset flags field */
137typedef enum { 133typedef enum {
@@ -521,6 +517,7 @@ static int validate_change(const struct cpuset *cur, const struct cpuset *trial)
521 return 0; 517 return 0;
522} 518}
523 519
520#ifdef CONFIG_SMP
524/* 521/*
525 * Helper routine for generate_sched_domains(). 522 * Helper routine for generate_sched_domains().
526 * Do cpusets a, b have overlapping cpus_allowed masks? 523 * Do cpusets a, b have overlapping cpus_allowed masks?
@@ -815,6 +812,18 @@ static void do_rebuild_sched_domains(struct work_struct *unused)
815 812
816 put_online_cpus(); 813 put_online_cpus();
817} 814}
815#else /* !CONFIG_SMP */
816static void do_rebuild_sched_domains(struct work_struct *unused)
817{
818}
819
820static int generate_sched_domains(struct cpumask **domains,
821 struct sched_domain_attr **attributes)
822{
823 *domains = NULL;
824 return 1;
825}
826#endif /* CONFIG_SMP */
818 827
819static DECLARE_WORK(rebuild_sched_domains_work, do_rebuild_sched_domains); 828static DECLARE_WORK(rebuild_sched_domains_work, do_rebuild_sched_domains);
820 829
@@ -1026,101 +1035,70 @@ static void cpuset_migrate_mm(struct mm_struct *mm, const nodemask_t *from,
1026 mutex_unlock(&callback_mutex); 1035 mutex_unlock(&callback_mutex);
1027} 1036}
1028 1037
1038/*
1039 * Rebind task's vmas to cpuset's new mems_allowed, and migrate pages to new
1040 * nodes if memory_migrate flag is set. Called with cgroup_mutex held.
1041 */
1042static void cpuset_change_nodemask(struct task_struct *p,
1043 struct cgroup_scanner *scan)
1044{
1045 struct mm_struct *mm;
1046 struct cpuset *cs;
1047 int migrate;
1048 const nodemask_t *oldmem = scan->data;
1049
1050 mm = get_task_mm(p);
1051 if (!mm)
1052 return;
1053
1054 cs = cgroup_cs(scan->cg);
1055 migrate = is_memory_migrate(cs);
1056
1057 mpol_rebind_mm(mm, &cs->mems_allowed);
1058 if (migrate)
1059 cpuset_migrate_mm(mm, oldmem, &cs->mems_allowed);
1060 mmput(mm);
1061}
1062
1029static void *cpuset_being_rebound; 1063static void *cpuset_being_rebound;
1030 1064
1031/** 1065/**
1032 * update_tasks_nodemask - Update the nodemasks of tasks in the cpuset. 1066 * update_tasks_nodemask - Update the nodemasks of tasks in the cpuset.
1033 * @cs: the cpuset in which each task's mems_allowed mask needs to be changed 1067 * @cs: the cpuset in which each task's mems_allowed mask needs to be changed
1034 * @oldmem: old mems_allowed of cpuset cs 1068 * @oldmem: old mems_allowed of cpuset cs
1069 * @heap: if NULL, defer allocating heap memory to cgroup_scan_tasks()
1035 * 1070 *
1036 * Called with cgroup_mutex held 1071 * Called with cgroup_mutex held
1037 * Return 0 if successful, -errno if not. 1072 * No return value. It's guaranteed that cgroup_scan_tasks() always returns 0
1073 * if @heap != NULL.
1038 */ 1074 */
1039static int update_tasks_nodemask(struct cpuset *cs, const nodemask_t *oldmem) 1075static void update_tasks_nodemask(struct cpuset *cs, const nodemask_t *oldmem,
1076 struct ptr_heap *heap)
1040{ 1077{
1041 struct task_struct *p; 1078 struct cgroup_scanner scan;
1042 struct mm_struct **mmarray;
1043 int i, n, ntasks;
1044 int migrate;
1045 int fudge;
1046 struct cgroup_iter it;
1047 int retval;
1048 1079
1049 cpuset_being_rebound = cs; /* causes mpol_dup() rebind */ 1080 cpuset_being_rebound = cs; /* causes mpol_dup() rebind */
1050 1081
1051 fudge = 10; /* spare mmarray[] slots */ 1082 scan.cg = cs->css.cgroup;
1052 fudge += cpumask_weight(cs->cpus_allowed);/* imagine 1 fork-bomb/cpu */ 1083 scan.test_task = NULL;
1053 retval = -ENOMEM; 1084 scan.process_task = cpuset_change_nodemask;
1054 1085 scan.heap = heap;
1055 /* 1086 scan.data = (nodemask_t *)oldmem;
1056 * Allocate mmarray[] to hold mm reference for each task
1057 * in cpuset cs. Can't kmalloc GFP_KERNEL while holding
1058 * tasklist_lock. We could use GFP_ATOMIC, but with a
1059 * few more lines of code, we can retry until we get a big
1060 * enough mmarray[] w/o using GFP_ATOMIC.
1061 */
1062 while (1) {
1063 ntasks = cgroup_task_count(cs->css.cgroup); /* guess */
1064 ntasks += fudge;
1065 mmarray = kmalloc(ntasks * sizeof(*mmarray), GFP_KERNEL);
1066 if (!mmarray)
1067 goto done;
1068 read_lock(&tasklist_lock); /* block fork */
1069 if (cgroup_task_count(cs->css.cgroup) <= ntasks)
1070 break; /* got enough */
1071 read_unlock(&tasklist_lock); /* try again */
1072 kfree(mmarray);
1073 }
1074
1075 n = 0;
1076
1077 /* Load up mmarray[] with mm reference for each task in cpuset. */
1078 cgroup_iter_start(cs->css.cgroup, &it);
1079 while ((p = cgroup_iter_next(cs->css.cgroup, &it))) {
1080 struct mm_struct *mm;
1081
1082 if (n >= ntasks) {
1083 printk(KERN_WARNING
1084 "Cpuset mempolicy rebind incomplete.\n");
1085 break;
1086 }
1087 mm = get_task_mm(p);
1088 if (!mm)
1089 continue;
1090 mmarray[n++] = mm;
1091 }
1092 cgroup_iter_end(cs->css.cgroup, &it);
1093 read_unlock(&tasklist_lock);
1094 1087
1095 /* 1088 /*
1096 * Now that we've dropped the tasklist spinlock, we can 1089 * The mpol_rebind_mm() call takes mmap_sem, which we couldn't
1097 * rebind the vma mempolicies of each mm in mmarray[] to their 1090 * take while holding tasklist_lock. Forks can happen - the
1098 * new cpuset, and release that mm. The mpol_rebind_mm() 1091 * mpol_dup() cpuset_being_rebound check will catch such forks,
1099 * call takes mmap_sem, which we couldn't take while holding 1092 * and rebind their vma mempolicies too. Because we still hold
1100 * tasklist_lock. Forks can happen again now - the mpol_dup() 1093 * the global cgroup_mutex, we know that no other rebind effort
1101 * cpuset_being_rebound check will catch such forks, and rebind 1094 * will be contending for the global variable cpuset_being_rebound.
1102 * their vma mempolicies too. Because we still hold the global
1103 * cgroup_mutex, we know that no other rebind effort will
1104 * be contending for the global variable cpuset_being_rebound.
1105 * It's ok if we rebind the same mm twice; mpol_rebind_mm() 1095 * It's ok if we rebind the same mm twice; mpol_rebind_mm()
1106 * is idempotent. Also migrate pages in each mm to new nodes. 1096 * is idempotent. Also migrate pages in each mm to new nodes.
1107 */ 1097 */
1108 migrate = is_memory_migrate(cs); 1098 cgroup_scan_tasks(&scan);
1109 for (i = 0; i < n; i++) {
1110 struct mm_struct *mm = mmarray[i];
1111
1112 mpol_rebind_mm(mm, &cs->mems_allowed);
1113 if (migrate)
1114 cpuset_migrate_mm(mm, oldmem, &cs->mems_allowed);
1115 mmput(mm);
1116 }
1117 1099
1118 /* We're done rebinding vmas to this cpuset's new mems_allowed. */ 1100 /* We're done rebinding vmas to this cpuset's new mems_allowed. */
1119 kfree(mmarray);
1120 cpuset_being_rebound = NULL; 1101 cpuset_being_rebound = NULL;
1121 retval = 0;
1122done:
1123 return retval;
1124} 1102}
1125 1103
1126/* 1104/*
@@ -1141,6 +1119,7 @@ static int update_nodemask(struct cpuset *cs, struct cpuset *trialcs,
1141{ 1119{
1142 nodemask_t oldmem; 1120 nodemask_t oldmem;
1143 int retval; 1121 int retval;
1122 struct ptr_heap heap;
1144 1123
1145 /* 1124 /*
1146 * top_cpuset.mems_allowed tracks node_stats[N_HIGH_MEMORY]; 1125 * top_cpuset.mems_allowed tracks node_stats[N_HIGH_MEMORY];
@@ -1175,12 +1154,18 @@ static int update_nodemask(struct cpuset *cs, struct cpuset *trialcs,
1175 if (retval < 0) 1154 if (retval < 0)
1176 goto done; 1155 goto done;
1177 1156
1157 retval = heap_init(&heap, PAGE_SIZE, GFP_KERNEL, NULL);
1158 if (retval < 0)
1159 goto done;
1160
1178 mutex_lock(&callback_mutex); 1161 mutex_lock(&callback_mutex);
1179 cs->mems_allowed = trialcs->mems_allowed; 1162 cs->mems_allowed = trialcs->mems_allowed;
1180 cs->mems_generation = cpuset_mems_generation++; 1163 cs->mems_generation = cpuset_mems_generation++;
1181 mutex_unlock(&callback_mutex); 1164 mutex_unlock(&callback_mutex);
1182 1165
1183 retval = update_tasks_nodemask(cs, &oldmem); 1166 update_tasks_nodemask(cs, &oldmem, &heap);
1167
1168 heap_free(&heap);
1184done: 1169done:
1185 return retval; 1170 return retval;
1186} 1171}
@@ -1192,8 +1177,10 @@ int current_cpuset_is_being_rebound(void)
1192 1177
1193static int update_relax_domain_level(struct cpuset *cs, s64 val) 1178static int update_relax_domain_level(struct cpuset *cs, s64 val)
1194{ 1179{
1180#ifdef CONFIG_SMP
1195 if (val < -1 || val >= SD_LV_MAX) 1181 if (val < -1 || val >= SD_LV_MAX)
1196 return -EINVAL; 1182 return -EINVAL;
1183#endif
1197 1184
1198 if (val != cs->relax_domain_level) { 1185 if (val != cs->relax_domain_level) {
1199 cs->relax_domain_level = val; 1186 cs->relax_domain_level = val;
@@ -1355,19 +1342,22 @@ static int cpuset_can_attach(struct cgroup_subsys *ss,
1355 struct cgroup *cont, struct task_struct *tsk) 1342 struct cgroup *cont, struct task_struct *tsk)
1356{ 1343{
1357 struct cpuset *cs = cgroup_cs(cont); 1344 struct cpuset *cs = cgroup_cs(cont);
1358 int ret = 0;
1359 1345
1360 if (cpumask_empty(cs->cpus_allowed) || nodes_empty(cs->mems_allowed)) 1346 if (cpumask_empty(cs->cpus_allowed) || nodes_empty(cs->mems_allowed))
1361 return -ENOSPC; 1347 return -ENOSPC;
1362 1348
1363 if (tsk->flags & PF_THREAD_BOUND) { 1349 /*
1364 mutex_lock(&callback_mutex); 1350 * Kthreads bound to specific cpus cannot be moved to a new cpuset; we
1365 if (!cpumask_equal(&tsk->cpus_allowed, cs->cpus_allowed)) 1351 * cannot change their cpu affinity and isolating such threads by their
1366 ret = -EINVAL; 1352 * set of allowed nodes is unnecessary. Thus, cpusets are not
1367 mutex_unlock(&callback_mutex); 1353 * applicable for such threads. This prevents checking for success of
1368 } 1354 * set_cpus_allowed_ptr() on all attached tasks before cpus_allowed may
1355 * be changed.
1356 */
1357 if (tsk->flags & PF_THREAD_BOUND)
1358 return -EINVAL;
1369 1359
1370 return ret < 0 ? ret : security_task_setscheduler(tsk, 0, NULL); 1360 return security_task_setscheduler(tsk, 0, NULL);
1371} 1361}
1372 1362
1373static void cpuset_attach(struct cgroup_subsys *ss, 1363static void cpuset_attach(struct cgroup_subsys *ss,
@@ -1706,6 +1696,7 @@ static struct cftype files[] = {
1706 .read_u64 = cpuset_read_u64, 1696 .read_u64 = cpuset_read_u64,
1707 .write_u64 = cpuset_write_u64, 1697 .write_u64 = cpuset_write_u64,
1708 .private = FILE_MEMORY_PRESSURE, 1698 .private = FILE_MEMORY_PRESSURE,
1699 .mode = S_IRUGO,
1709 }, 1700 },
1710 1701
1711 { 1702 {
@@ -1913,10 +1904,9 @@ int __init cpuset_init(void)
1913static void cpuset_do_move_task(struct task_struct *tsk, 1904static void cpuset_do_move_task(struct task_struct *tsk,
1914 struct cgroup_scanner *scan) 1905 struct cgroup_scanner *scan)
1915{ 1906{
1916 struct cpuset_hotplug_scanner *chsp; 1907 struct cgroup *new_cgroup = scan->data;
1917 1908
1918 chsp = container_of(scan, struct cpuset_hotplug_scanner, scan); 1909 cgroup_attach_task(new_cgroup, tsk);
1919 cgroup_attach_task(chsp->to, tsk);
1920} 1910}
1921 1911
1922/** 1912/**
@@ -1932,15 +1922,15 @@ static void cpuset_do_move_task(struct task_struct *tsk,
1932 */ 1922 */
1933static void move_member_tasks_to_cpuset(struct cpuset *from, struct cpuset *to) 1923static void move_member_tasks_to_cpuset(struct cpuset *from, struct cpuset *to)
1934{ 1924{
1935 struct cpuset_hotplug_scanner scan; 1925 struct cgroup_scanner scan;
1936 1926
1937 scan.scan.cg = from->css.cgroup; 1927 scan.cg = from->css.cgroup;
1938 scan.scan.test_task = NULL; /* select all tasks in cgroup */ 1928 scan.test_task = NULL; /* select all tasks in cgroup */
1939 scan.scan.process_task = cpuset_do_move_task; 1929 scan.process_task = cpuset_do_move_task;
1940 scan.scan.heap = NULL; 1930 scan.heap = NULL;
1941 scan.to = to->css.cgroup; 1931 scan.data = to->css.cgroup;
1942 1932
1943 if (cgroup_scan_tasks(&scan.scan)) 1933 if (cgroup_scan_tasks(&scan))
1944 printk(KERN_ERR "move_member_tasks_to_cpuset: " 1934 printk(KERN_ERR "move_member_tasks_to_cpuset: "
1945 "cgroup_scan_tasks failed\n"); 1935 "cgroup_scan_tasks failed\n");
1946} 1936}
@@ -2033,7 +2023,7 @@ static void scan_for_empty_cpusets(struct cpuset *root)
2033 remove_tasks_in_empty_cpuset(cp); 2023 remove_tasks_in_empty_cpuset(cp);
2034 else { 2024 else {
2035 update_tasks_cpumask(cp, NULL); 2025 update_tasks_cpumask(cp, NULL);
2036 update_tasks_nodemask(cp, &oldmems); 2026 update_tasks_nodemask(cp, &oldmems, NULL);
2037 } 2027 }
2038 } 2028 }
2039} 2029}
@@ -2069,7 +2059,9 @@ static int cpuset_track_online_cpus(struct notifier_block *unused_nb,
2069 } 2059 }
2070 2060
2071 cgroup_lock(); 2061 cgroup_lock();
2062 mutex_lock(&callback_mutex);
2072 cpumask_copy(top_cpuset.cpus_allowed, cpu_online_mask); 2063 cpumask_copy(top_cpuset.cpus_allowed, cpu_online_mask);
2064 mutex_unlock(&callback_mutex);
2073 scan_for_empty_cpusets(&top_cpuset); 2065 scan_for_empty_cpusets(&top_cpuset);
2074 ndoms = generate_sched_domains(&doms, &attr); 2066 ndoms = generate_sched_domains(&doms, &attr);
2075 cgroup_unlock(); 2067 cgroup_unlock();
@@ -2092,11 +2084,12 @@ static int cpuset_track_online_nodes(struct notifier_block *self,
2092 cgroup_lock(); 2084 cgroup_lock();
2093 switch (action) { 2085 switch (action) {
2094 case MEM_ONLINE: 2086 case MEM_ONLINE:
2095 top_cpuset.mems_allowed = node_states[N_HIGH_MEMORY];
2096 break;
2097 case MEM_OFFLINE: 2087 case MEM_OFFLINE:
2088 mutex_lock(&callback_mutex);
2098 top_cpuset.mems_allowed = node_states[N_HIGH_MEMORY]; 2089 top_cpuset.mems_allowed = node_states[N_HIGH_MEMORY];
2099 scan_for_empty_cpusets(&top_cpuset); 2090 mutex_unlock(&callback_mutex);
2091 if (action == MEM_OFFLINE)
2092 scan_for_empty_cpusets(&top_cpuset);
2100 break; 2093 break;
2101 default: 2094 default:
2102 break; 2095 break;
@@ -2206,26 +2199,24 @@ static const struct cpuset *nearest_hardwall_ancestor(const struct cpuset *cs)
2206} 2199}
2207 2200
2208/** 2201/**
2209 * cpuset_zone_allowed_softwall - Can we allocate on zone z's memory node? 2202 * cpuset_node_allowed_softwall - Can we allocate on a memory node?
2210 * @z: is this zone on an allowed node? 2203 * @node: is this an allowed node?
2211 * @gfp_mask: memory allocation flags 2204 * @gfp_mask: memory allocation flags
2212 * 2205 *
2213 * If we're in interrupt, yes, we can always allocate. If 2206 * If we're in interrupt, yes, we can always allocate. If __GFP_THISNODE is
2214 * __GFP_THISNODE is set, yes, we can always allocate. If zone 2207 * set, yes, we can always allocate. If node is in our task's mems_allowed,
2215 * z's node is in our tasks mems_allowed, yes. If it's not a 2208 * yes. If it's not a __GFP_HARDWALL request and this node is in the nearest
2216 * __GFP_HARDWALL request and this zone's nodes is in the nearest 2209 * hardwalled cpuset ancestor to this task's cpuset, yes. If the task has been
2217 * hardwalled cpuset ancestor to this tasks cpuset, yes. 2210 * OOM killed and has access to memory reserves as specified by the TIF_MEMDIE
2218 * If the task has been OOM killed and has access to memory reserves 2211 * flag, yes.
2219 * as specified by the TIF_MEMDIE flag, yes.
2220 * Otherwise, no. 2212 * Otherwise, no.
2221 * 2213 *
2222 * If __GFP_HARDWALL is set, cpuset_zone_allowed_softwall() 2214 * If __GFP_HARDWALL is set, cpuset_node_allowed_softwall() reduces to
2223 * reduces to cpuset_zone_allowed_hardwall(). Otherwise, 2215 * cpuset_node_allowed_hardwall(). Otherwise, cpuset_node_allowed_softwall()
2224 * cpuset_zone_allowed_softwall() might sleep, and might allow a zone 2216 * might sleep, and might allow a node from an enclosing cpuset.
2225 * from an enclosing cpuset.
2226 * 2217 *
2227 * cpuset_zone_allowed_hardwall() only handles the simpler case of 2218 * cpuset_node_allowed_hardwall() only handles the simpler case of hardwall
2228 * hardwall cpusets, and never sleeps. 2219 * cpusets, and never sleeps.
2229 * 2220 *
2230 * The __GFP_THISNODE placement logic is really handled elsewhere, 2221 * The __GFP_THISNODE placement logic is really handled elsewhere,
2231 * by forcibly using a zonelist starting at a specified node, and by 2222 * by forcibly using a zonelist starting at a specified node, and by
@@ -2264,20 +2255,17 @@ static const struct cpuset *nearest_hardwall_ancestor(const struct cpuset *cs)
2264 * GFP_USER - only nodes in current tasks mems allowed ok. 2255 * GFP_USER - only nodes in current tasks mems allowed ok.
2265 * 2256 *
2266 * Rule: 2257 * Rule:
2267 * Don't call cpuset_zone_allowed_softwall if you can't sleep, unless you 2258 * Don't call cpuset_node_allowed_softwall if you can't sleep, unless you
2268 * pass in the __GFP_HARDWALL flag set in gfp_flag, which disables 2259 * pass in the __GFP_HARDWALL flag set in gfp_flag, which disables
2269 * the code that might scan up ancestor cpusets and sleep. 2260 * the code that might scan up ancestor cpusets and sleep.
2270 */ 2261 */
2271 2262int __cpuset_node_allowed_softwall(int node, gfp_t gfp_mask)
2272int __cpuset_zone_allowed_softwall(struct zone *z, gfp_t gfp_mask)
2273{ 2263{
2274 int node; /* node that zone z is on */
2275 const struct cpuset *cs; /* current cpuset ancestors */ 2264 const struct cpuset *cs; /* current cpuset ancestors */
2276 int allowed; /* is allocation in zone z allowed? */ 2265 int allowed; /* is allocation in zone z allowed? */
2277 2266
2278 if (in_interrupt() || (gfp_mask & __GFP_THISNODE)) 2267 if (in_interrupt() || (gfp_mask & __GFP_THISNODE))
2279 return 1; 2268 return 1;
2280 node = zone_to_nid(z);
2281 might_sleep_if(!(gfp_mask & __GFP_HARDWALL)); 2269 might_sleep_if(!(gfp_mask & __GFP_HARDWALL));
2282 if (node_isset(node, current->mems_allowed)) 2270 if (node_isset(node, current->mems_allowed))
2283 return 1; 2271 return 1;
@@ -2306,15 +2294,15 @@ int __cpuset_zone_allowed_softwall(struct zone *z, gfp_t gfp_mask)
2306} 2294}
2307 2295
2308/* 2296/*
2309 * cpuset_zone_allowed_hardwall - Can we allocate on zone z's memory node? 2297 * cpuset_node_allowed_hardwall - Can we allocate on a memory node?
2310 * @z: is this zone on an allowed node? 2298 * @node: is this an allowed node?
2311 * @gfp_mask: memory allocation flags 2299 * @gfp_mask: memory allocation flags
2312 * 2300 *
2313 * If we're in interrupt, yes, we can always allocate. 2301 * If we're in interrupt, yes, we can always allocate. If __GFP_THISNODE is
2314 * If __GFP_THISNODE is set, yes, we can always allocate. If zone 2302 * set, yes, we can always allocate. If node is in our task's mems_allowed,
2315 * z's node is in our tasks mems_allowed, yes. If the task has been 2303 * yes. If the task has been OOM killed and has access to memory reserves as
2316 * OOM killed and has access to memory reserves as specified by the 2304 * specified by the TIF_MEMDIE flag, yes.
2317 * TIF_MEMDIE flag, yes. Otherwise, no. 2305 * Otherwise, no.
2318 * 2306 *
2319 * The __GFP_THISNODE placement logic is really handled elsewhere, 2307 * The __GFP_THISNODE placement logic is really handled elsewhere,
2320 * by forcibly using a zonelist starting at a specified node, and by 2308 * by forcibly using a zonelist starting at a specified node, and by
@@ -2322,20 +2310,16 @@ int __cpuset_zone_allowed_softwall(struct zone *z, gfp_t gfp_mask)
2322 * any node on the zonelist except the first. By the time any such 2310 * any node on the zonelist except the first. By the time any such
2323 * calls get to this routine, we should just shut up and say 'yes'. 2311 * calls get to this routine, we should just shut up and say 'yes'.
2324 * 2312 *
2325 * Unlike the cpuset_zone_allowed_softwall() variant, above, 2313 * Unlike the cpuset_node_allowed_softwall() variant, above,
2326 * this variant requires that the zone be in the current tasks 2314 * this variant requires that the node be in the current task's
2327 * mems_allowed or that we're in interrupt. It does not scan up the 2315 * mems_allowed or that we're in interrupt. It does not scan up the
2328 * cpuset hierarchy for the nearest enclosing mem_exclusive cpuset. 2316 * cpuset hierarchy for the nearest enclosing mem_exclusive cpuset.
2329 * It never sleeps. 2317 * It never sleeps.
2330 */ 2318 */
2331 2319int __cpuset_node_allowed_hardwall(int node, gfp_t gfp_mask)
2332int __cpuset_zone_allowed_hardwall(struct zone *z, gfp_t gfp_mask)
2333{ 2320{
2334 int node; /* node that zone z is on */
2335
2336 if (in_interrupt() || (gfp_mask & __GFP_THISNODE)) 2321 if (in_interrupt() || (gfp_mask & __GFP_THISNODE))
2337 return 1; 2322 return 1;
2338 node = zone_to_nid(z);
2339 if (node_isset(node, current->mems_allowed)) 2323 if (node_isset(node, current->mems_allowed))
2340 return 1; 2324 return 1;
2341 /* 2325 /*
diff --git a/kernel/exec_domain.c b/kernel/exec_domain.c
index 667c841c2952..c35452cadded 100644
--- a/kernel/exec_domain.c
+++ b/kernel/exec_domain.c
@@ -18,6 +18,7 @@
18#include <linux/syscalls.h> 18#include <linux/syscalls.h>
19#include <linux/sysctl.h> 19#include <linux/sysctl.h>
20#include <linux/types.h> 20#include <linux/types.h>
21#include <linux/fs_struct.h>
21 22
22 23
23static void default_handler(int, struct pt_regs *); 24static void default_handler(int, struct pt_regs *);
@@ -145,28 +146,6 @@ __set_personality(u_long personality)
145 return 0; 146 return 0;
146 } 147 }
147 148
148 if (atomic_read(&current->fs->count) != 1) {
149 struct fs_struct *fsp, *ofsp;
150
151 fsp = copy_fs_struct(current->fs);
152 if (fsp == NULL) {
153 module_put(ep->module);
154 return -ENOMEM;
155 }
156
157 task_lock(current);
158 ofsp = current->fs;
159 current->fs = fsp;
160 task_unlock(current);
161
162 put_fs_struct(ofsp);
163 }
164
165 /*
166 * At that point we are guaranteed to be the sole owner of
167 * current->fs.
168 */
169
170 current->personality = personality; 149 current->personality = personality;
171 oep = current_thread_info()->exec_domain; 150 oep = current_thread_info()->exec_domain;
172 current_thread_info()->exec_domain = ep; 151 current_thread_info()->exec_domain = ep;
diff --git a/kernel/exit.c b/kernel/exit.c
index f52c24eb8a8f..7a14a2b504f5 100644
--- a/kernel/exit.c
+++ b/kernel/exit.c
@@ -46,6 +46,7 @@
46#include <linux/blkdev.h> 46#include <linux/blkdev.h>
47#include <linux/task_io_accounting_ops.h> 47#include <linux/task_io_accounting_ops.h>
48#include <linux/tracehook.h> 48#include <linux/tracehook.h>
49#include <linux/fs_struct.h>
49#include <linux/init_task.h> 50#include <linux/init_task.h>
50#include <trace/sched.h> 51#include <trace/sched.h>
51 52
@@ -61,11 +62,6 @@ DEFINE_TRACE(sched_process_wait);
61 62
62static void exit_mm(struct task_struct * tsk); 63static void exit_mm(struct task_struct * tsk);
63 64
64static inline int task_detached(struct task_struct *p)
65{
66 return p->exit_signal == -1;
67}
68
69static void __unhash_process(struct task_struct *p) 65static void __unhash_process(struct task_struct *p)
70{ 66{
71 nr_threads--; 67 nr_threads--;
@@ -365,16 +361,12 @@ static void reparent_to_kthreadd(void)
365void __set_special_pids(struct pid *pid) 361void __set_special_pids(struct pid *pid)
366{ 362{
367 struct task_struct *curr = current->group_leader; 363 struct task_struct *curr = current->group_leader;
368 pid_t nr = pid_nr(pid);
369 364
370 if (task_session(curr) != pid) { 365 if (task_session(curr) != pid)
371 change_pid(curr, PIDTYPE_SID, pid); 366 change_pid(curr, PIDTYPE_SID, pid);
372 set_task_session(curr, nr); 367
373 } 368 if (task_pgrp(curr) != pid)
374 if (task_pgrp(curr) != pid) {
375 change_pid(curr, PIDTYPE_PGID, pid); 369 change_pid(curr, PIDTYPE_PGID, pid);
376 set_task_pgrp(curr, nr);
377 }
378} 370}
379 371
380static void set_special_pids(struct pid *pid) 372static void set_special_pids(struct pid *pid)
@@ -432,7 +424,6 @@ EXPORT_SYMBOL(disallow_signal);
432void daemonize(const char *name, ...) 424void daemonize(const char *name, ...)
433{ 425{
434 va_list args; 426 va_list args;
435 struct fs_struct *fs;
436 sigset_t blocked; 427 sigset_t blocked;
437 428
438 va_start(args, name); 429 va_start(args, name);
@@ -465,11 +456,7 @@ void daemonize(const char *name, ...)
465 456
466 /* Become as one with the init task */ 457 /* Become as one with the init task */
467 458
468 exit_fs(current); /* current->fs->count--; */ 459 daemonize_fs_struct();
469 fs = init_task.fs;
470 current->fs = fs;
471 atomic_inc(&fs->count);
472
473 exit_files(current); 460 exit_files(current);
474 current->files = init_task.files; 461 current->files = init_task.files;
475 atomic_inc(&current->files->count); 462 atomic_inc(&current->files->count);
@@ -568,30 +555,6 @@ void exit_files(struct task_struct *tsk)
568 } 555 }
569} 556}
570 557
571void put_fs_struct(struct fs_struct *fs)
572{
573 /* No need to hold fs->lock if we are killing it */
574 if (atomic_dec_and_test(&fs->count)) {
575 path_put(&fs->root);
576 path_put(&fs->pwd);
577 kmem_cache_free(fs_cachep, fs);
578 }
579}
580
581void exit_fs(struct task_struct *tsk)
582{
583 struct fs_struct * fs = tsk->fs;
584
585 if (fs) {
586 task_lock(tsk);
587 tsk->fs = NULL;
588 task_unlock(tsk);
589 put_fs_struct(fs);
590 }
591}
592
593EXPORT_SYMBOL_GPL(exit_fs);
594
595#ifdef CONFIG_MM_OWNER 558#ifdef CONFIG_MM_OWNER
596/* 559/*
597 * Task p is exiting and it owned mm, lets find a new owner for it 560 * Task p is exiting and it owned mm, lets find a new owner for it
@@ -735,119 +698,6 @@ static void exit_mm(struct task_struct * tsk)
735} 698}
736 699
737/* 700/*
738 * Return nonzero if @parent's children should reap themselves.
739 *
740 * Called with write_lock_irq(&tasklist_lock) held.
741 */
742static int ignoring_children(struct task_struct *parent)
743{
744 int ret;
745 struct sighand_struct *psig = parent->sighand;
746 unsigned long flags;
747 spin_lock_irqsave(&psig->siglock, flags);
748 ret = (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN ||
749 (psig->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT));
750 spin_unlock_irqrestore(&psig->siglock, flags);
751 return ret;
752}
753
754/*
755 * Detach all tasks we were using ptrace on.
756 * Any that need to be release_task'd are put on the @dead list.
757 *
758 * Called with write_lock(&tasklist_lock) held.
759 */
760static void ptrace_exit(struct task_struct *parent, struct list_head *dead)
761{
762 struct task_struct *p, *n;
763 int ign = -1;
764
765 list_for_each_entry_safe(p, n, &parent->ptraced, ptrace_entry) {
766 __ptrace_unlink(p);
767
768 if (p->exit_state != EXIT_ZOMBIE)
769 continue;
770
771 /*
772 * If it's a zombie, our attachedness prevented normal
773 * parent notification or self-reaping. Do notification
774 * now if it would have happened earlier. If it should
775 * reap itself, add it to the @dead list. We can't call
776 * release_task() here because we already hold tasklist_lock.
777 *
778 * If it's our own child, there is no notification to do.
779 * But if our normal children self-reap, then this child
780 * was prevented by ptrace and we must reap it now.
781 */
782 if (!task_detached(p) && thread_group_empty(p)) {
783 if (!same_thread_group(p->real_parent, parent))
784 do_notify_parent(p, p->exit_signal);
785 else {
786 if (ign < 0)
787 ign = ignoring_children(parent);
788 if (ign)
789 p->exit_signal = -1;
790 }
791 }
792
793 if (task_detached(p)) {
794 /*
795 * Mark it as in the process of being reaped.
796 */
797 p->exit_state = EXIT_DEAD;
798 list_add(&p->ptrace_entry, dead);
799 }
800 }
801}
802
803/*
804 * Finish up exit-time ptrace cleanup.
805 *
806 * Called without locks.
807 */
808static void ptrace_exit_finish(struct task_struct *parent,
809 struct list_head *dead)
810{
811 struct task_struct *p, *n;
812
813 BUG_ON(!list_empty(&parent->ptraced));
814
815 list_for_each_entry_safe(p, n, dead, ptrace_entry) {
816 list_del_init(&p->ptrace_entry);
817 release_task(p);
818 }
819}
820
821static void reparent_thread(struct task_struct *p, struct task_struct *father)
822{
823 if (p->pdeath_signal)
824 /* We already hold the tasklist_lock here. */
825 group_send_sig_info(p->pdeath_signal, SEND_SIG_NOINFO, p);
826
827 list_move_tail(&p->sibling, &p->real_parent->children);
828
829 /* If this is a threaded reparent there is no need to
830 * notify anyone anything has happened.
831 */
832 if (same_thread_group(p->real_parent, father))
833 return;
834
835 /* We don't want people slaying init. */
836 if (!task_detached(p))
837 p->exit_signal = SIGCHLD;
838
839 /* If we'd notified the old parent about this child's death,
840 * also notify the new parent.
841 */
842 if (!ptrace_reparented(p) &&
843 p->exit_state == EXIT_ZOMBIE &&
844 !task_detached(p) && thread_group_empty(p))
845 do_notify_parent(p, p->exit_signal);
846
847 kill_orphaned_pgrp(p, father);
848}
849
850/*
851 * When we die, we re-parent all our children. 701 * When we die, we re-parent all our children.
852 * Try to give them to another thread in our thread 702 * Try to give them to another thread in our thread
853 * group, and if no such member exists, give it to 703 * group, and if no such member exists, give it to
@@ -886,17 +736,51 @@ static struct task_struct *find_new_reaper(struct task_struct *father)
886 return pid_ns->child_reaper; 736 return pid_ns->child_reaper;
887} 737}
888 738
739/*
740* Any that need to be release_task'd are put on the @dead list.
741 */
742static void reparent_thread(struct task_struct *father, struct task_struct *p,
743 struct list_head *dead)
744{
745 if (p->pdeath_signal)
746 group_send_sig_info(p->pdeath_signal, SEND_SIG_NOINFO, p);
747
748 list_move_tail(&p->sibling, &p->real_parent->children);
749
750 if (task_detached(p))
751 return;
752 /*
753 * If this is a threaded reparent there is no need to
754 * notify anyone anything has happened.
755 */
756 if (same_thread_group(p->real_parent, father))
757 return;
758
759 /* We don't want people slaying init. */
760 p->exit_signal = SIGCHLD;
761
762 /* If it has exited notify the new parent about this child's death. */
763 if (!p->ptrace &&
764 p->exit_state == EXIT_ZOMBIE && thread_group_empty(p)) {
765 do_notify_parent(p, p->exit_signal);
766 if (task_detached(p)) {
767 p->exit_state = EXIT_DEAD;
768 list_move_tail(&p->sibling, dead);
769 }
770 }
771
772 kill_orphaned_pgrp(p, father);
773}
774
889static void forget_original_parent(struct task_struct *father) 775static void forget_original_parent(struct task_struct *father)
890{ 776{
891 struct task_struct *p, *n, *reaper; 777 struct task_struct *p, *n, *reaper;
892 LIST_HEAD(ptrace_dead); 778 LIST_HEAD(dead_children);
779
780 exit_ptrace(father);
893 781
894 write_lock_irq(&tasklist_lock); 782 write_lock_irq(&tasklist_lock);
895 reaper = find_new_reaper(father); 783 reaper = find_new_reaper(father);
896 /*
897 * First clean up ptrace if we were using it.
898 */
899 ptrace_exit(father, &ptrace_dead);
900 784
901 list_for_each_entry_safe(p, n, &father->children, sibling) { 785 list_for_each_entry_safe(p, n, &father->children, sibling) {
902 p->real_parent = reaper; 786 p->real_parent = reaper;
@@ -904,13 +788,16 @@ static void forget_original_parent(struct task_struct *father)
904 BUG_ON(p->ptrace); 788 BUG_ON(p->ptrace);
905 p->parent = p->real_parent; 789 p->parent = p->real_parent;
906 } 790 }
907 reparent_thread(p, father); 791 reparent_thread(father, p, &dead_children);
908 } 792 }
909
910 write_unlock_irq(&tasklist_lock); 793 write_unlock_irq(&tasklist_lock);
794
911 BUG_ON(!list_empty(&father->children)); 795 BUG_ON(!list_empty(&father->children));
912 796
913 ptrace_exit_finish(father, &ptrace_dead); 797 list_for_each_entry_safe(p, n, &dead_children, sibling) {
798 list_del_init(&p->sibling);
799 release_task(p);
800 }
914} 801}
915 802
916/* 803/*
@@ -1422,6 +1309,18 @@ static int wait_task_zombie(struct task_struct *p, int options,
1422 return retval; 1309 return retval;
1423} 1310}
1424 1311
1312static int *task_stopped_code(struct task_struct *p, bool ptrace)
1313{
1314 if (ptrace) {
1315 if (task_is_stopped_or_traced(p))
1316 return &p->exit_code;
1317 } else {
1318 if (p->signal->flags & SIGNAL_STOP_STOPPED)
1319 return &p->signal->group_exit_code;
1320 }
1321 return NULL;
1322}
1323
1425/* 1324/*
1426 * Handle sys_wait4 work for one task in state TASK_STOPPED. We hold 1325 * Handle sys_wait4 work for one task in state TASK_STOPPED. We hold
1427 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold 1326 * read_lock(&tasklist_lock) on entry. If we return zero, we still hold
@@ -1432,7 +1331,7 @@ static int wait_task_stopped(int ptrace, struct task_struct *p,
1432 int options, struct siginfo __user *infop, 1331 int options, struct siginfo __user *infop,
1433 int __user *stat_addr, struct rusage __user *ru) 1332 int __user *stat_addr, struct rusage __user *ru)
1434{ 1333{
1435 int retval, exit_code, why; 1334 int retval, exit_code, *p_code, why;
1436 uid_t uid = 0; /* unneeded, required by compiler */ 1335 uid_t uid = 0; /* unneeded, required by compiler */
1437 pid_t pid; 1336 pid_t pid;
1438 1337
@@ -1442,22 +1341,16 @@ static int wait_task_stopped(int ptrace, struct task_struct *p,
1442 exit_code = 0; 1341 exit_code = 0;
1443 spin_lock_irq(&p->sighand->siglock); 1342 spin_lock_irq(&p->sighand->siglock);
1444 1343
1445 if (unlikely(!task_is_stopped_or_traced(p))) 1344 p_code = task_stopped_code(p, ptrace);
1446 goto unlock_sig; 1345 if (unlikely(!p_code))
1447
1448 if (!ptrace && p->signal->group_stop_count > 0)
1449 /*
1450 * A group stop is in progress and this is the group leader.
1451 * We won't report until all threads have stopped.
1452 */
1453 goto unlock_sig; 1346 goto unlock_sig;
1454 1347
1455 exit_code = p->exit_code; 1348 exit_code = *p_code;
1456 if (!exit_code) 1349 if (!exit_code)
1457 goto unlock_sig; 1350 goto unlock_sig;
1458 1351
1459 if (!unlikely(options & WNOWAIT)) 1352 if (!unlikely(options & WNOWAIT))
1460 p->exit_code = 0; 1353 *p_code = 0;
1461 1354
1462 /* don't need the RCU readlock here as we're holding a spinlock */ 1355 /* don't need the RCU readlock here as we're holding a spinlock */
1463 uid = __task_cred(p)->uid; 1356 uid = __task_cred(p)->uid;
@@ -1613,7 +1506,7 @@ static int wait_consider_task(struct task_struct *parent, int ptrace,
1613 */ 1506 */
1614 *notask_error = 0; 1507 *notask_error = 0;
1615 1508
1616 if (task_is_stopped_or_traced(p)) 1509 if (task_stopped_code(p, ptrace))
1617 return wait_task_stopped(ptrace, p, options, 1510 return wait_task_stopped(ptrace, p, options,
1618 infop, stat_addr, ru); 1511 infop, stat_addr, ru);
1619 1512
@@ -1817,7 +1710,7 @@ SYSCALL_DEFINE4(wait4, pid_t, upid, int __user *, stat_addr,
1817 pid = find_get_pid(-upid); 1710 pid = find_get_pid(-upid);
1818 } else if (upid == 0) { 1711 } else if (upid == 0) {
1819 type = PIDTYPE_PGID; 1712 type = PIDTYPE_PGID;
1820 pid = get_pid(task_pgrp(current)); 1713 pid = get_task_pid(current, PIDTYPE_PGID);
1821 } else /* upid > 0 */ { 1714 } else /* upid > 0 */ {
1822 type = PIDTYPE_PID; 1715 type = PIDTYPE_PID;
1823 pid = find_get_pid(upid); 1716 pid = find_get_pid(upid);
diff --git a/kernel/extable.c b/kernel/extable.c
index e136ed8d82ba..7f8f263f8524 100644
--- a/kernel/extable.c
+++ b/kernel/extable.c
@@ -15,11 +15,22 @@
15 along with this program; if not, write to the Free Software 15 along with this program; if not, write to the Free Software
16 Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA 16 Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
17*/ 17*/
18#include <linux/ftrace.h>
19#include <linux/memory.h>
18#include <linux/module.h> 20#include <linux/module.h>
21#include <linux/mutex.h>
19#include <linux/init.h> 22#include <linux/init.h>
20#include <linux/ftrace.h> 23
21#include <asm/uaccess.h>
22#include <asm/sections.h> 24#include <asm/sections.h>
25#include <asm/uaccess.h>
26
27/*
28 * mutex protecting text section modification (dynamic code patching).
29 * some users need to sleep (allocating memory...) while they hold this lock.
30 *
31 * NOT exported to modules - patching kernel text is a really delicate matter.
32 */
33DEFINE_MUTEX(text_mutex);
23 34
24extern struct exception_table_entry __start___ex_table[]; 35extern struct exception_table_entry __start___ex_table[];
25extern struct exception_table_entry __stop___ex_table[]; 36extern struct exception_table_entry __stop___ex_table[];
@@ -41,31 +52,50 @@ const struct exception_table_entry *search_exception_tables(unsigned long addr)
41 return e; 52 return e;
42} 53}
43 54
44__notrace_funcgraph int core_kernel_text(unsigned long addr) 55static inline int init_kernel_text(unsigned long addr)
56{
57 if (addr >= (unsigned long)_sinittext &&
58 addr <= (unsigned long)_einittext)
59 return 1;
60 return 0;
61}
62
63int core_kernel_text(unsigned long addr)
45{ 64{
46 if (addr >= (unsigned long)_stext && 65 if (addr >= (unsigned long)_stext &&
47 addr <= (unsigned long)_etext) 66 addr <= (unsigned long)_etext)
48 return 1; 67 return 1;
49 68
50 if (system_state == SYSTEM_BOOTING && 69 if (system_state == SYSTEM_BOOTING &&
51 addr >= (unsigned long)_sinittext && 70 init_kernel_text(addr))
52 addr <= (unsigned long)_einittext)
53 return 1; 71 return 1;
54 return 0; 72 return 0;
55} 73}
56 74
57__notrace_funcgraph int __kernel_text_address(unsigned long addr) 75int __kernel_text_address(unsigned long addr)
58{ 76{
59 if (core_kernel_text(addr)) 77 if (core_kernel_text(addr))
60 return 1; 78 return 1;
61 return __module_text_address(addr) != NULL; 79 if (is_module_text_address(addr))
80 return 1;
81 /*
82 * There might be init symbols in saved stacktraces.
83 * Give those symbols a chance to be printed in
84 * backtraces (such as lockdep traces).
85 *
86 * Since we are after the module-symbols check, there's
87 * no danger of address overlap:
88 */
89 if (init_kernel_text(addr))
90 return 1;
91 return 0;
62} 92}
63 93
64int kernel_text_address(unsigned long addr) 94int kernel_text_address(unsigned long addr)
65{ 95{
66 if (core_kernel_text(addr)) 96 if (core_kernel_text(addr))
67 return 1; 97 return 1;
68 return module_text_address(addr) != NULL; 98 return is_module_text_address(addr);
69} 99}
70 100
71/* 101/*
@@ -81,5 +111,5 @@ int func_ptr_is_kernel_text(void *ptr)
81 addr = (unsigned long) dereference_function_descriptor(ptr); 111 addr = (unsigned long) dereference_function_descriptor(ptr);
82 if (core_kernel_text(addr)) 112 if (core_kernel_text(addr))
83 return 1; 113 return 1;
84 return module_text_address(addr) != NULL; 114 return is_module_text_address(addr);
85} 115}
diff --git a/kernel/fork.c b/kernel/fork.c
index 4640a3e0085e..381d7f9b70fb 100644
--- a/kernel/fork.c
+++ b/kernel/fork.c
@@ -60,6 +60,7 @@
60#include <linux/tty.h> 60#include <linux/tty.h>
61#include <linux/proc_fs.h> 61#include <linux/proc_fs.h>
62#include <linux/blkdev.h> 62#include <linux/blkdev.h>
63#include <linux/fs_struct.h>
63#include <trace/sched.h> 64#include <trace/sched.h>
64#include <linux/magic.h> 65#include <linux/magic.h>
65 66
@@ -284,7 +285,7 @@ static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
284 mm->free_area_cache = oldmm->mmap_base; 285 mm->free_area_cache = oldmm->mmap_base;
285 mm->cached_hole_size = ~0UL; 286 mm->cached_hole_size = ~0UL;
286 mm->map_count = 0; 287 mm->map_count = 0;
287 cpus_clear(mm->cpu_vm_mask); 288 cpumask_clear(mm_cpumask(mm));
288 mm->mm_rb = RB_ROOT; 289 mm->mm_rb = RB_ROOT;
289 rb_link = &mm->mm_rb.rb_node; 290 rb_link = &mm->mm_rb.rb_node;
290 rb_parent = NULL; 291 rb_parent = NULL;
@@ -681,38 +682,21 @@ fail_nomem:
681 return retval; 682 return retval;
682} 683}
683 684
684static struct fs_struct *__copy_fs_struct(struct fs_struct *old)
685{
686 struct fs_struct *fs = kmem_cache_alloc(fs_cachep, GFP_KERNEL);
687 /* We don't need to lock fs - think why ;-) */
688 if (fs) {
689 atomic_set(&fs->count, 1);
690 rwlock_init(&fs->lock);
691 fs->umask = old->umask;
692 read_lock(&old->lock);
693 fs->root = old->root;
694 path_get(&old->root);
695 fs->pwd = old->pwd;
696 path_get(&old->pwd);
697 read_unlock(&old->lock);
698 }
699 return fs;
700}
701
702struct fs_struct *copy_fs_struct(struct fs_struct *old)
703{
704 return __copy_fs_struct(old);
705}
706
707EXPORT_SYMBOL_GPL(copy_fs_struct);
708
709static int copy_fs(unsigned long clone_flags, struct task_struct *tsk) 685static int copy_fs(unsigned long clone_flags, struct task_struct *tsk)
710{ 686{
687 struct fs_struct *fs = current->fs;
711 if (clone_flags & CLONE_FS) { 688 if (clone_flags & CLONE_FS) {
712 atomic_inc(&current->fs->count); 689 /* tsk->fs is already what we want */
690 write_lock(&fs->lock);
691 if (fs->in_exec) {
692 write_unlock(&fs->lock);
693 return -EAGAIN;
694 }
695 fs->users++;
696 write_unlock(&fs->lock);
713 return 0; 697 return 0;
714 } 698 }
715 tsk->fs = __copy_fs_struct(current->fs); 699 tsk->fs = copy_fs_struct(fs);
716 if (!tsk->fs) 700 if (!tsk->fs)
717 return -ENOMEM; 701 return -ENOMEM;
718 return 0; 702 return 0;
@@ -841,6 +825,8 @@ static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
841 atomic_set(&sig->live, 1); 825 atomic_set(&sig->live, 1);
842 init_waitqueue_head(&sig->wait_chldexit); 826 init_waitqueue_head(&sig->wait_chldexit);
843 sig->flags = 0; 827 sig->flags = 0;
828 if (clone_flags & CLONE_NEWPID)
829 sig->flags |= SIGNAL_UNKILLABLE;
844 sig->group_exit_code = 0; 830 sig->group_exit_code = 0;
845 sig->group_exit_task = NULL; 831 sig->group_exit_task = NULL;
846 sig->group_stop_count = 0; 832 sig->group_stop_count = 0;
@@ -1126,7 +1112,7 @@ static struct task_struct *copy_process(unsigned long clone_flags,
1126 goto bad_fork_cleanup_mm; 1112 goto bad_fork_cleanup_mm;
1127 if ((retval = copy_io(clone_flags, p))) 1113 if ((retval = copy_io(clone_flags, p)))
1128 goto bad_fork_cleanup_namespaces; 1114 goto bad_fork_cleanup_namespaces;
1129 retval = copy_thread(0, clone_flags, stack_start, stack_size, p, regs); 1115 retval = copy_thread(clone_flags, stack_start, stack_size, p, regs);
1130 if (retval) 1116 if (retval)
1131 goto bad_fork_cleanup_io; 1117 goto bad_fork_cleanup_io;
1132 1118
@@ -1185,10 +1171,6 @@ static struct task_struct *copy_process(unsigned long clone_flags,
1185#endif 1171#endif
1186 clear_all_latency_tracing(p); 1172 clear_all_latency_tracing(p);
1187 1173
1188 /* Our parent execution domain becomes current domain
1189 These must match for thread signalling to apply */
1190 p->parent_exec_id = p->self_exec_id;
1191
1192 /* ok, now we should be set up.. */ 1174 /* ok, now we should be set up.. */
1193 p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL); 1175 p->exit_signal = (clone_flags & CLONE_THREAD) ? -1 : (clone_flags & CSIGNAL);
1194 p->pdeath_signal = 0; 1176 p->pdeath_signal = 0;
@@ -1226,10 +1208,13 @@ static struct task_struct *copy_process(unsigned long clone_flags,
1226 set_task_cpu(p, smp_processor_id()); 1208 set_task_cpu(p, smp_processor_id());
1227 1209
1228 /* CLONE_PARENT re-uses the old parent */ 1210 /* CLONE_PARENT re-uses the old parent */
1229 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) 1211 if (clone_flags & (CLONE_PARENT|CLONE_THREAD)) {
1230 p->real_parent = current->real_parent; 1212 p->real_parent = current->real_parent;
1231 else 1213 p->parent_exec_id = current->parent_exec_id;
1214 } else {
1232 p->real_parent = current; 1215 p->real_parent = current;
1216 p->parent_exec_id = current->self_exec_id;
1217 }
1233 1218
1234 spin_lock(&current->sighand->siglock); 1219 spin_lock(&current->sighand->siglock);
1235 1220
@@ -1265,8 +1250,6 @@ static struct task_struct *copy_process(unsigned long clone_flags,
1265 p->signal->leader_pid = pid; 1250 p->signal->leader_pid = pid;
1266 tty_kref_put(p->signal->tty); 1251 tty_kref_put(p->signal->tty);
1267 p->signal->tty = tty_kref_get(current->signal->tty); 1252 p->signal->tty = tty_kref_get(current->signal->tty);
1268 set_task_pgrp(p, task_pgrp_nr(current));
1269 set_task_session(p, task_session_nr(current));
1270 attach_pid(p, PIDTYPE_PGID, task_pgrp(current)); 1253 attach_pid(p, PIDTYPE_PGID, task_pgrp(current));
1271 attach_pid(p, PIDTYPE_SID, task_session(current)); 1254 attach_pid(p, PIDTYPE_SID, task_session(current));
1272 list_add_tail_rcu(&p->tasks, &init_task.tasks); 1255 list_add_tail_rcu(&p->tasks, &init_task.tasks);
@@ -1490,6 +1473,7 @@ void __init proc_caches_init(void)
1490 mm_cachep = kmem_cache_create("mm_struct", 1473 mm_cachep = kmem_cache_create("mm_struct",
1491 sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN, 1474 sizeof(struct mm_struct), ARCH_MIN_MMSTRUCT_ALIGN,
1492 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL); 1475 SLAB_HWCACHE_ALIGN|SLAB_PANIC, NULL);
1476 vm_area_cachep = KMEM_CACHE(vm_area_struct, SLAB_PANIC);
1493 mmap_init(); 1477 mmap_init();
1494} 1478}
1495 1479
@@ -1545,12 +1529,16 @@ static int unshare_fs(unsigned long unshare_flags, struct fs_struct **new_fsp)
1545{ 1529{
1546 struct fs_struct *fs = current->fs; 1530 struct fs_struct *fs = current->fs;
1547 1531
1548 if ((unshare_flags & CLONE_FS) && 1532 if (!(unshare_flags & CLONE_FS) || !fs)
1549 (fs && atomic_read(&fs->count) > 1)) { 1533 return 0;
1550 *new_fsp = __copy_fs_struct(current->fs); 1534
1551 if (!*new_fsp) 1535 /* don't need lock here; in the worst case we'll do useless copy */
1552 return -ENOMEM; 1536 if (fs->users == 1)
1553 } 1537 return 0;
1538
1539 *new_fsp = copy_fs_struct(fs);
1540 if (!*new_fsp)
1541 return -ENOMEM;
1554 1542
1555 return 0; 1543 return 0;
1556} 1544}
@@ -1666,8 +1654,13 @@ SYSCALL_DEFINE1(unshare, unsigned long, unshare_flags)
1666 1654
1667 if (new_fs) { 1655 if (new_fs) {
1668 fs = current->fs; 1656 fs = current->fs;
1657 write_lock(&fs->lock);
1669 current->fs = new_fs; 1658 current->fs = new_fs;
1670 new_fs = fs; 1659 if (--fs->users)
1660 new_fs = NULL;
1661 else
1662 new_fs = fs;
1663 write_unlock(&fs->lock);
1671 } 1664 }
1672 1665
1673 if (new_mm) { 1666 if (new_mm) {
@@ -1706,7 +1699,7 @@ bad_unshare_cleanup_sigh:
1706 1699
1707bad_unshare_cleanup_fs: 1700bad_unshare_cleanup_fs:
1708 if (new_fs) 1701 if (new_fs)
1709 put_fs_struct(new_fs); 1702 free_fs_struct(new_fs);
1710 1703
1711bad_unshare_cleanup_thread: 1704bad_unshare_cleanup_thread:
1712bad_unshare_out: 1705bad_unshare_out:
diff --git a/kernel/futex.c b/kernel/futex.c
index 438701adce23..6b50a024bca2 100644
--- a/kernel/futex.c
+++ b/kernel/futex.c
@@ -114,7 +114,9 @@ struct futex_q {
114}; 114};
115 115
116/* 116/*
117 * Split the global futex_lock into every hash list lock. 117 * Hash buckets are shared by all the futex_keys that hash to the same
118 * location. Each key may have multiple futex_q structures, one for each task
119 * waiting on a futex.
118 */ 120 */
119struct futex_hash_bucket { 121struct futex_hash_bucket {
120 spinlock_t lock; 122 spinlock_t lock;
@@ -189,8 +191,7 @@ static void drop_futex_key_refs(union futex_key *key)
189/** 191/**
190 * get_futex_key - Get parameters which are the keys for a futex. 192 * get_futex_key - Get parameters which are the keys for a futex.
191 * @uaddr: virtual address of the futex 193 * @uaddr: virtual address of the futex
192 * @shared: NULL for a PROCESS_PRIVATE futex, 194 * @fshared: 0 for a PROCESS_PRIVATE futex, 1 for PROCESS_SHARED
193 * &current->mm->mmap_sem for a PROCESS_SHARED futex
194 * @key: address where result is stored. 195 * @key: address where result is stored.
195 * 196 *
196 * Returns a negative error code or 0 197 * Returns a negative error code or 0
@@ -200,9 +201,7 @@ static void drop_futex_key_refs(union futex_key *key)
200 * offset_within_page). For private mappings, it's (uaddr, current->mm). 201 * offset_within_page). For private mappings, it's (uaddr, current->mm).
201 * We can usually work out the index without swapping in the page. 202 * We can usually work out the index without swapping in the page.
202 * 203 *
203 * fshared is NULL for PROCESS_PRIVATE futexes 204 * lock_page() might sleep, the caller should not hold a spinlock.
204 * For other futexes, it points to &current->mm->mmap_sem and
205 * caller must have taken the reader lock. but NOT any spinlocks.
206 */ 205 */
207static int get_futex_key(u32 __user *uaddr, int fshared, union futex_key *key) 206static int get_futex_key(u32 __user *uaddr, int fshared, union futex_key *key)
208{ 207{
@@ -299,41 +298,6 @@ static int get_futex_value_locked(u32 *dest, u32 __user *from)
299 return ret ? -EFAULT : 0; 298 return ret ? -EFAULT : 0;
300} 299}
301 300
302/*
303 * Fault handling.
304 */
305static int futex_handle_fault(unsigned long address, int attempt)
306{
307 struct vm_area_struct * vma;
308 struct mm_struct *mm = current->mm;
309 int ret = -EFAULT;
310
311 if (attempt > 2)
312 return ret;
313
314 down_read(&mm->mmap_sem);
315 vma = find_vma(mm, address);
316 if (vma && address >= vma->vm_start &&
317 (vma->vm_flags & VM_WRITE)) {
318 int fault;
319 fault = handle_mm_fault(mm, vma, address, 1);
320 if (unlikely((fault & VM_FAULT_ERROR))) {
321#if 0
322 /* XXX: let's do this when we verify it is OK */
323 if (ret & VM_FAULT_OOM)
324 ret = -ENOMEM;
325#endif
326 } else {
327 ret = 0;
328 if (fault & VM_FAULT_MAJOR)
329 current->maj_flt++;
330 else
331 current->min_flt++;
332 }
333 }
334 up_read(&mm->mmap_sem);
335 return ret;
336}
337 301
338/* 302/*
339 * PI code: 303 * PI code:
@@ -589,10 +553,9 @@ static void wake_futex(struct futex_q *q)
589 * The waiting task can free the futex_q as soon as this is written, 553 * The waiting task can free the futex_q as soon as this is written,
590 * without taking any locks. This must come last. 554 * without taking any locks. This must come last.
591 * 555 *
592 * A memory barrier is required here to prevent the following store 556 * A memory barrier is required here to prevent the following store to
593 * to lock_ptr from getting ahead of the wakeup. Clearing the lock 557 * lock_ptr from getting ahead of the wakeup. Clearing the lock at the
594 * at the end of wake_up_all() does not prevent this store from 558 * end of wake_up() does not prevent this store from moving.
595 * moving.
596 */ 559 */
597 smp_wmb(); 560 smp_wmb();
598 q->lock_ptr = NULL; 561 q->lock_ptr = NULL;
@@ -692,9 +655,16 @@ double_lock_hb(struct futex_hash_bucket *hb1, struct futex_hash_bucket *hb2)
692 } 655 }
693} 656}
694 657
658static inline void
659double_unlock_hb(struct futex_hash_bucket *hb1, struct futex_hash_bucket *hb2)
660{
661 spin_unlock(&hb1->lock);
662 if (hb1 != hb2)
663 spin_unlock(&hb2->lock);
664}
665
695/* 666/*
696 * Wake up all waiters hashed on the physical page that is mapped 667 * Wake up waiters matching bitset queued on this futex (uaddr).
697 * to this virtual address:
698 */ 668 */
699static int futex_wake(u32 __user *uaddr, int fshared, int nr_wake, u32 bitset) 669static int futex_wake(u32 __user *uaddr, int fshared, int nr_wake, u32 bitset)
700{ 670{
@@ -750,9 +720,9 @@ futex_wake_op(u32 __user *uaddr1, int fshared, u32 __user *uaddr2,
750 struct futex_hash_bucket *hb1, *hb2; 720 struct futex_hash_bucket *hb1, *hb2;
751 struct plist_head *head; 721 struct plist_head *head;
752 struct futex_q *this, *next; 722 struct futex_q *this, *next;
753 int ret, op_ret, attempt = 0; 723 int ret, op_ret;
754 724
755retryfull: 725retry:
756 ret = get_futex_key(uaddr1, fshared, &key1); 726 ret = get_futex_key(uaddr1, fshared, &key1);
757 if (unlikely(ret != 0)) 727 if (unlikely(ret != 0))
758 goto out; 728 goto out;
@@ -763,16 +733,13 @@ retryfull:
763 hb1 = hash_futex(&key1); 733 hb1 = hash_futex(&key1);
764 hb2 = hash_futex(&key2); 734 hb2 = hash_futex(&key2);
765 735
766retry:
767 double_lock_hb(hb1, hb2); 736 double_lock_hb(hb1, hb2);
768 737retry_private:
769 op_ret = futex_atomic_op_inuser(op, uaddr2); 738 op_ret = futex_atomic_op_inuser(op, uaddr2);
770 if (unlikely(op_ret < 0)) { 739 if (unlikely(op_ret < 0)) {
771 u32 dummy; 740 u32 dummy;
772 741
773 spin_unlock(&hb1->lock); 742 double_unlock_hb(hb1, hb2);
774 if (hb1 != hb2)
775 spin_unlock(&hb2->lock);
776 743
777#ifndef CONFIG_MMU 744#ifndef CONFIG_MMU
778 /* 745 /*
@@ -788,26 +755,16 @@ retry:
788 goto out_put_keys; 755 goto out_put_keys;
789 } 756 }
790 757
791 /*
792 * futex_atomic_op_inuser needs to both read and write
793 * *(int __user *)uaddr2, but we can't modify it
794 * non-atomically. Therefore, if get_user below is not
795 * enough, we need to handle the fault ourselves, while
796 * still holding the mmap_sem.
797 */
798 if (attempt++) {
799 ret = futex_handle_fault((unsigned long)uaddr2,
800 attempt);
801 if (ret)
802 goto out_put_keys;
803 goto retry;
804 }
805
806 ret = get_user(dummy, uaddr2); 758 ret = get_user(dummy, uaddr2);
807 if (ret) 759 if (ret)
808 return ret; 760 goto out_put_keys;
761
762 if (!fshared)
763 goto retry_private;
809 764
810 goto retryfull; 765 put_futex_key(fshared, &key2);
766 put_futex_key(fshared, &key1);
767 goto retry;
811 } 768 }
812 769
813 head = &hb1->chain; 770 head = &hb1->chain;
@@ -834,9 +791,7 @@ retry:
834 ret += op_ret; 791 ret += op_ret;
835 } 792 }
836 793
837 spin_unlock(&hb1->lock); 794 double_unlock_hb(hb1, hb2);
838 if (hb1 != hb2)
839 spin_unlock(&hb2->lock);
840out_put_keys: 795out_put_keys:
841 put_futex_key(fshared, &key2); 796 put_futex_key(fshared, &key2);
842out_put_key1: 797out_put_key1:
@@ -869,6 +824,7 @@ retry:
869 hb1 = hash_futex(&key1); 824 hb1 = hash_futex(&key1);
870 hb2 = hash_futex(&key2); 825 hb2 = hash_futex(&key2);
871 826
827retry_private:
872 double_lock_hb(hb1, hb2); 828 double_lock_hb(hb1, hb2);
873 829
874 if (likely(cmpval != NULL)) { 830 if (likely(cmpval != NULL)) {
@@ -877,16 +833,18 @@ retry:
877 ret = get_futex_value_locked(&curval, uaddr1); 833 ret = get_futex_value_locked(&curval, uaddr1);
878 834
879 if (unlikely(ret)) { 835 if (unlikely(ret)) {
880 spin_unlock(&hb1->lock); 836 double_unlock_hb(hb1, hb2);
881 if (hb1 != hb2)
882 spin_unlock(&hb2->lock);
883 837
884 ret = get_user(curval, uaddr1); 838 ret = get_user(curval, uaddr1);
839 if (ret)
840 goto out_put_keys;
885 841
886 if (!ret) 842 if (!fshared)
887 goto retry; 843 goto retry_private;
888 844
889 goto out_put_keys; 845 put_futex_key(fshared, &key2);
846 put_futex_key(fshared, &key1);
847 goto retry;
890 } 848 }
891 if (curval != *cmpval) { 849 if (curval != *cmpval) {
892 ret = -EAGAIN; 850 ret = -EAGAIN;
@@ -923,9 +881,7 @@ retry:
923 } 881 }
924 882
925out_unlock: 883out_unlock:
926 spin_unlock(&hb1->lock); 884 double_unlock_hb(hb1, hb2);
927 if (hb1 != hb2)
928 spin_unlock(&hb2->lock);
929 885
930 /* drop_futex_key_refs() must be called outside the spinlocks. */ 886 /* drop_futex_key_refs() must be called outside the spinlocks. */
931 while (--drop_count >= 0) 887 while (--drop_count >= 0)
@@ -1063,7 +1019,7 @@ static int fixup_pi_state_owner(u32 __user *uaddr, struct futex_q *q,
1063 struct futex_pi_state *pi_state = q->pi_state; 1019 struct futex_pi_state *pi_state = q->pi_state;
1064 struct task_struct *oldowner = pi_state->owner; 1020 struct task_struct *oldowner = pi_state->owner;
1065 u32 uval, curval, newval; 1021 u32 uval, curval, newval;
1066 int ret, attempt = 0; 1022 int ret;
1067 1023
1068 /* Owner died? */ 1024 /* Owner died? */
1069 if (!pi_state->owner) 1025 if (!pi_state->owner)
@@ -1076,11 +1032,9 @@ static int fixup_pi_state_owner(u32 __user *uaddr, struct futex_q *q,
1076 * in the user space variable. This must be atomic as we have 1032 * in the user space variable. This must be atomic as we have
1077 * to preserve the owner died bit here. 1033 * to preserve the owner died bit here.
1078 * 1034 *
1079 * Note: We write the user space value _before_ changing the 1035 * Note: We write the user space value _before_ changing the pi_state
1080 * pi_state because we can fault here. Imagine swapped out 1036 * because we can fault here. Imagine swapped out pages or a fork
1081 * pages or a fork, which was running right before we acquired 1037 * that marked all the anonymous memory readonly for cow.
1082 * mmap_sem, that marked all the anonymous memory readonly for
1083 * cow.
1084 * 1038 *
1085 * Modifying pi_state _before_ the user space value would 1039 * Modifying pi_state _before_ the user space value would
1086 * leave the pi_state in an inconsistent state when we fault 1040 * leave the pi_state in an inconsistent state when we fault
@@ -1136,7 +1090,7 @@ retry:
1136handle_fault: 1090handle_fault:
1137 spin_unlock(q->lock_ptr); 1091 spin_unlock(q->lock_ptr);
1138 1092
1139 ret = futex_handle_fault((unsigned long)uaddr, attempt++); 1093 ret = get_user(uval, uaddr);
1140 1094
1141 spin_lock(q->lock_ptr); 1095 spin_lock(q->lock_ptr);
1142 1096
@@ -1185,10 +1139,11 @@ retry:
1185 if (unlikely(ret != 0)) 1139 if (unlikely(ret != 0))
1186 goto out; 1140 goto out;
1187 1141
1142retry_private:
1188 hb = queue_lock(&q); 1143 hb = queue_lock(&q);
1189 1144
1190 /* 1145 /*
1191 * Access the page AFTER the futex is queued. 1146 * Access the page AFTER the hash-bucket is locked.
1192 * Order is important: 1147 * Order is important:
1193 * 1148 *
1194 * Userspace waiter: val = var; if (cond(val)) futex_wait(&var, val); 1149 * Userspace waiter: val = var; if (cond(val)) futex_wait(&var, val);
@@ -1204,20 +1159,23 @@ retry:
1204 * a wakeup when *uaddr != val on entry to the syscall. This is 1159 * a wakeup when *uaddr != val on entry to the syscall. This is
1205 * rare, but normal. 1160 * rare, but normal.
1206 * 1161 *
1207 * for shared futexes, we hold the mmap semaphore, so the mapping 1162 * For shared futexes, we hold the mmap semaphore, so the mapping
1208 * cannot have changed since we looked it up in get_futex_key. 1163 * cannot have changed since we looked it up in get_futex_key.
1209 */ 1164 */
1210 ret = get_futex_value_locked(&uval, uaddr); 1165 ret = get_futex_value_locked(&uval, uaddr);
1211 1166
1212 if (unlikely(ret)) { 1167 if (unlikely(ret)) {
1213 queue_unlock(&q, hb); 1168 queue_unlock(&q, hb);
1214 put_futex_key(fshared, &q.key);
1215 1169
1216 ret = get_user(uval, uaddr); 1170 ret = get_user(uval, uaddr);
1171 if (ret)
1172 goto out_put_key;
1217 1173
1218 if (!ret) 1174 if (!fshared)
1219 goto retry; 1175 goto retry_private;
1220 goto out; 1176
1177 put_futex_key(fshared, &q.key);
1178 goto retry;
1221 } 1179 }
1222 ret = -EWOULDBLOCK; 1180 ret = -EWOULDBLOCK;
1223 if (unlikely(uval != val)) { 1181 if (unlikely(uval != val)) {
@@ -1248,16 +1206,13 @@ retry:
1248 if (!abs_time) 1206 if (!abs_time)
1249 schedule(); 1207 schedule();
1250 else { 1208 else {
1251 unsigned long slack;
1252 slack = current->timer_slack_ns;
1253 if (rt_task(current))
1254 slack = 0;
1255 hrtimer_init_on_stack(&t.timer, 1209 hrtimer_init_on_stack(&t.timer,
1256 clockrt ? CLOCK_REALTIME : 1210 clockrt ? CLOCK_REALTIME :
1257 CLOCK_MONOTONIC, 1211 CLOCK_MONOTONIC,
1258 HRTIMER_MODE_ABS); 1212 HRTIMER_MODE_ABS);
1259 hrtimer_init_sleeper(&t, current); 1213 hrtimer_init_sleeper(&t, current);
1260 hrtimer_set_expires_range_ns(&t.timer, *abs_time, slack); 1214 hrtimer_set_expires_range_ns(&t.timer, *abs_time,
1215 current->timer_slack_ns);
1261 1216
1262 hrtimer_start_expires(&t.timer, HRTIMER_MODE_ABS); 1217 hrtimer_start_expires(&t.timer, HRTIMER_MODE_ABS);
1263 if (!hrtimer_active(&t.timer)) 1218 if (!hrtimer_active(&t.timer))
@@ -1354,7 +1309,7 @@ static int futex_lock_pi(u32 __user *uaddr, int fshared,
1354 struct futex_hash_bucket *hb; 1309 struct futex_hash_bucket *hb;
1355 u32 uval, newval, curval; 1310 u32 uval, newval, curval;
1356 struct futex_q q; 1311 struct futex_q q;
1357 int ret, lock_taken, ownerdied = 0, attempt = 0; 1312 int ret, lock_taken, ownerdied = 0;
1358 1313
1359 if (refill_pi_state_cache()) 1314 if (refill_pi_state_cache())
1360 return -ENOMEM; 1315 return -ENOMEM;
@@ -1374,7 +1329,7 @@ retry:
1374 if (unlikely(ret != 0)) 1329 if (unlikely(ret != 0))
1375 goto out; 1330 goto out;
1376 1331
1377retry_unlocked: 1332retry_private:
1378 hb = queue_lock(&q); 1333 hb = queue_lock(&q);
1379 1334
1380retry_locked: 1335retry_locked:
@@ -1458,6 +1413,7 @@ retry_locked:
1458 * exit to complete. 1413 * exit to complete.
1459 */ 1414 */
1460 queue_unlock(&q, hb); 1415 queue_unlock(&q, hb);
1416 put_futex_key(fshared, &q.key);
1461 cond_resched(); 1417 cond_resched();
1462 goto retry; 1418 goto retry;
1463 1419
@@ -1564,6 +1520,13 @@ retry_locked:
1564 } 1520 }
1565 } 1521 }
1566 1522
1523 /*
1524 * If fixup_pi_state_owner() faulted and was unable to handle the
1525 * fault, unlock it and return the fault to userspace.
1526 */
1527 if (ret && (rt_mutex_owner(&q.pi_state->pi_mutex) == current))
1528 rt_mutex_unlock(&q.pi_state->pi_mutex);
1529
1567 /* Unqueue and drop the lock */ 1530 /* Unqueue and drop the lock */
1568 unqueue_me_pi(&q); 1531 unqueue_me_pi(&q);
1569 1532
@@ -1591,22 +1554,18 @@ uaddr_faulted:
1591 */ 1554 */
1592 queue_unlock(&q, hb); 1555 queue_unlock(&q, hb);
1593 1556
1594 if (attempt++) {
1595 ret = futex_handle_fault((unsigned long)uaddr, attempt);
1596 if (ret)
1597 goto out_put_key;
1598 goto retry_unlocked;
1599 }
1600
1601 ret = get_user(uval, uaddr); 1557 ret = get_user(uval, uaddr);
1602 if (!ret) 1558 if (ret)
1603 goto retry; 1559 goto out_put_key;
1604 1560
1605 if (to) 1561 if (!fshared)
1606 destroy_hrtimer_on_stack(&to->timer); 1562 goto retry_private;
1607 return ret; 1563
1564 put_futex_key(fshared, &q.key);
1565 goto retry;
1608} 1566}
1609 1567
1568
1610/* 1569/*
1611 * Userspace attempted a TID -> 0 atomic transition, and failed. 1570 * Userspace attempted a TID -> 0 atomic transition, and failed.
1612 * This is the in-kernel slowpath: we look up the PI state (if any), 1571 * This is the in-kernel slowpath: we look up the PI state (if any),
@@ -1619,7 +1578,7 @@ static int futex_unlock_pi(u32 __user *uaddr, int fshared)
1619 u32 uval; 1578 u32 uval;
1620 struct plist_head *head; 1579 struct plist_head *head;
1621 union futex_key key = FUTEX_KEY_INIT; 1580 union futex_key key = FUTEX_KEY_INIT;
1622 int ret, attempt = 0; 1581 int ret;
1623 1582
1624retry: 1583retry:
1625 if (get_user(uval, uaddr)) 1584 if (get_user(uval, uaddr))
@@ -1635,7 +1594,6 @@ retry:
1635 goto out; 1594 goto out;
1636 1595
1637 hb = hash_futex(&key); 1596 hb = hash_futex(&key);
1638retry_unlocked:
1639 spin_lock(&hb->lock); 1597 spin_lock(&hb->lock);
1640 1598
1641 /* 1599 /*
@@ -1700,14 +1658,7 @@ pi_faulted:
1700 * we have to drop the mmap_sem in order to call get_user(). 1658 * we have to drop the mmap_sem in order to call get_user().
1701 */ 1659 */
1702 spin_unlock(&hb->lock); 1660 spin_unlock(&hb->lock);
1703 1661 put_futex_key(fshared, &key);
1704 if (attempt++) {
1705 ret = futex_handle_fault((unsigned long)uaddr, attempt);
1706 if (ret)
1707 goto out;
1708 uval = 0;
1709 goto retry_unlocked;
1710 }
1711 1662
1712 ret = get_user(uval, uaddr); 1663 ret = get_user(uval, uaddr);
1713 if (!ret) 1664 if (!ret)
diff --git a/kernel/irq/Makefile b/kernel/irq/Makefile
index 4dd5b1edac98..3394f8f52964 100644
--- a/kernel/irq/Makefile
+++ b/kernel/irq/Makefile
@@ -4,3 +4,4 @@ obj-$(CONFIG_GENERIC_IRQ_PROBE) += autoprobe.o
4obj-$(CONFIG_PROC_FS) += proc.o 4obj-$(CONFIG_PROC_FS) += proc.o
5obj-$(CONFIG_GENERIC_PENDING_IRQ) += migration.o 5obj-$(CONFIG_GENERIC_PENDING_IRQ) += migration.o
6obj-$(CONFIG_NUMA_MIGRATE_IRQ_DESC) += numa_migrate.o 6obj-$(CONFIG_NUMA_MIGRATE_IRQ_DESC) += numa_migrate.o
7obj-$(CONFIG_PM_SLEEP) += pm.o
diff --git a/kernel/irq/chip.c b/kernel/irq/chip.c
index 122fef4b0bd3..c687ba4363f2 100644
--- a/kernel/irq/chip.c
+++ b/kernel/irq/chip.c
@@ -81,6 +81,7 @@ void dynamic_irq_cleanup(unsigned int irq)
81 desc->handle_irq = handle_bad_irq; 81 desc->handle_irq = handle_bad_irq;
82 desc->chip = &no_irq_chip; 82 desc->chip = &no_irq_chip;
83 desc->name = NULL; 83 desc->name = NULL;
84 clear_kstat_irqs(desc);
84 spin_unlock_irqrestore(&desc->lock, flags); 85 spin_unlock_irqrestore(&desc->lock, flags);
85} 86}
86 87
@@ -293,7 +294,8 @@ static inline void mask_ack_irq(struct irq_desc *desc, int irq)
293 desc->chip->mask_ack(irq); 294 desc->chip->mask_ack(irq);
294 else { 295 else {
295 desc->chip->mask(irq); 296 desc->chip->mask(irq);
296 desc->chip->ack(irq); 297 if (desc->chip->ack)
298 desc->chip->ack(irq);
297 } 299 }
298} 300}
299 301
@@ -479,7 +481,8 @@ handle_edge_irq(unsigned int irq, struct irq_desc *desc)
479 kstat_incr_irqs_this_cpu(irq, desc); 481 kstat_incr_irqs_this_cpu(irq, desc);
480 482
481 /* Start handling the irq */ 483 /* Start handling the irq */
482 desc->chip->ack(irq); 484 if (desc->chip->ack)
485 desc->chip->ack(irq);
483 desc = irq_remap_to_desc(irq, desc); 486 desc = irq_remap_to_desc(irq, desc);
484 487
485 /* Mark the IRQ currently in progress.*/ 488 /* Mark the IRQ currently in progress.*/
diff --git a/kernel/irq/handle.c b/kernel/irq/handle.c
index f51eaee921b6..343acecae629 100644
--- a/kernel/irq/handle.c
+++ b/kernel/irq/handle.c
@@ -17,6 +17,7 @@
17#include <linux/kernel_stat.h> 17#include <linux/kernel_stat.h>
18#include <linux/rculist.h> 18#include <linux/rculist.h>
19#include <linux/hash.h> 19#include <linux/hash.h>
20#include <trace/irq.h>
20#include <linux/bootmem.h> 21#include <linux/bootmem.h>
21 22
22#include "internals.h" 23#include "internals.h"
@@ -82,19 +83,21 @@ static struct irq_desc irq_desc_init = {
82 83
83void init_kstat_irqs(struct irq_desc *desc, int cpu, int nr) 84void init_kstat_irqs(struct irq_desc *desc, int cpu, int nr)
84{ 85{
85 unsigned long bytes;
86 char *ptr;
87 int node; 86 int node;
88 87 void *ptr;
89 /* Compute how many bytes we need per irq and allocate them */
90 bytes = nr * sizeof(unsigned int);
91 88
92 node = cpu_to_node(cpu); 89 node = cpu_to_node(cpu);
93 ptr = kzalloc_node(bytes, GFP_ATOMIC, node); 90 ptr = kzalloc_node(nr * sizeof(*desc->kstat_irqs), GFP_ATOMIC, node);
94 printk(KERN_DEBUG " alloc kstat_irqs on cpu %d node %d\n", cpu, node);
95 91
96 if (ptr) 92 /*
97 desc->kstat_irqs = (unsigned int *)ptr; 93 * don't overwite if can not get new one
94 * init_copy_kstat_irqs() could still use old one
95 */
96 if (ptr) {
97 printk(KERN_DEBUG " alloc kstat_irqs on cpu %d node %d\n",
98 cpu, node);
99 desc->kstat_irqs = ptr;
100 }
98} 101}
99 102
100static void init_one_irq_desc(int irq, struct irq_desc *desc, int cpu) 103static void init_one_irq_desc(int irq, struct irq_desc *desc, int cpu)
@@ -237,6 +240,7 @@ struct irq_desc irq_desc[NR_IRQS] __cacheline_aligned_in_smp = {
237 } 240 }
238}; 241};
239 242
243static unsigned int kstat_irqs_all[NR_IRQS][NR_CPUS];
240int __init early_irq_init(void) 244int __init early_irq_init(void)
241{ 245{
242 struct irq_desc *desc; 246 struct irq_desc *desc;
@@ -253,6 +257,7 @@ int __init early_irq_init(void)
253 for (i = 0; i < count; i++) { 257 for (i = 0; i < count; i++) {
254 desc[i].irq = i; 258 desc[i].irq = i;
255 init_alloc_desc_masks(&desc[i], 0, true); 259 init_alloc_desc_masks(&desc[i], 0, true);
260 desc[i].kstat_irqs = kstat_irqs_all[i];
256 } 261 }
257 return arch_early_irq_init(); 262 return arch_early_irq_init();
258} 263}
@@ -268,6 +273,11 @@ struct irq_desc *irq_to_desc_alloc_cpu(unsigned int irq, int cpu)
268} 273}
269#endif /* !CONFIG_SPARSE_IRQ */ 274#endif /* !CONFIG_SPARSE_IRQ */
270 275
276void clear_kstat_irqs(struct irq_desc *desc)
277{
278 memset(desc->kstat_irqs, 0, nr_cpu_ids * sizeof(*(desc->kstat_irqs)));
279}
280
271/* 281/*
272 * What should we do if we get a hw irq event on an illegal vector? 282 * What should we do if we get a hw irq event on an illegal vector?
273 * Each architecture has to answer this themself. 283 * Each architecture has to answer this themself.
@@ -329,6 +339,9 @@ irqreturn_t no_action(int cpl, void *dev_id)
329 return IRQ_NONE; 339 return IRQ_NONE;
330} 340}
331 341
342DEFINE_TRACE(irq_handler_entry);
343DEFINE_TRACE(irq_handler_exit);
344
332/** 345/**
333 * handle_IRQ_event - irq action chain handler 346 * handle_IRQ_event - irq action chain handler
334 * @irq: the interrupt number 347 * @irq: the interrupt number
@@ -341,11 +354,15 @@ irqreturn_t handle_IRQ_event(unsigned int irq, struct irqaction *action)
341 irqreturn_t ret, retval = IRQ_NONE; 354 irqreturn_t ret, retval = IRQ_NONE;
342 unsigned int status = 0; 355 unsigned int status = 0;
343 356
357 WARN_ONCE(!in_irq(), "BUG: IRQ handler called from non-hardirq context!");
358
344 if (!(action->flags & IRQF_DISABLED)) 359 if (!(action->flags & IRQF_DISABLED))
345 local_irq_enable_in_hardirq(); 360 local_irq_enable_in_hardirq();
346 361
347 do { 362 do {
363 trace_irq_handler_entry(irq, action);
348 ret = action->handler(irq, action->dev_id); 364 ret = action->handler(irq, action->dev_id);
365 trace_irq_handler_exit(irq, action, ret);
349 if (ret == IRQ_HANDLED) 366 if (ret == IRQ_HANDLED)
350 status |= action->flags; 367 status |= action->flags;
351 retval |= ret; 368 retval |= ret;
@@ -360,6 +377,11 @@ irqreturn_t handle_IRQ_event(unsigned int irq, struct irqaction *action)
360} 377}
361 378
362#ifndef CONFIG_GENERIC_HARDIRQS_NO__DO_IRQ 379#ifndef CONFIG_GENERIC_HARDIRQS_NO__DO_IRQ
380
381#ifdef CONFIG_ENABLE_WARN_DEPRECATED
382# warning __do_IRQ is deprecated. Please convert to proper flow handlers
383#endif
384
363/** 385/**
364 * __do_IRQ - original all in one highlevel IRQ handler 386 * __do_IRQ - original all in one highlevel IRQ handler
365 * @irq: the interrupt number 387 * @irq: the interrupt number
@@ -480,12 +502,10 @@ void early_init_irq_lock_class(void)
480 } 502 }
481} 503}
482 504
483#ifdef CONFIG_SPARSE_IRQ
484unsigned int kstat_irqs_cpu(unsigned int irq, int cpu) 505unsigned int kstat_irqs_cpu(unsigned int irq, int cpu)
485{ 506{
486 struct irq_desc *desc = irq_to_desc(irq); 507 struct irq_desc *desc = irq_to_desc(irq);
487 return desc ? desc->kstat_irqs[cpu] : 0; 508 return desc ? desc->kstat_irqs[cpu] : 0;
488} 509}
489#endif
490EXPORT_SYMBOL(kstat_irqs_cpu); 510EXPORT_SYMBOL(kstat_irqs_cpu);
491 511
diff --git a/kernel/irq/internals.h b/kernel/irq/internals.h
index 40416a81a0f5..01ce20eab38f 100644
--- a/kernel/irq/internals.h
+++ b/kernel/irq/internals.h
@@ -12,9 +12,12 @@ extern void compat_irq_chip_set_default_handler(struct irq_desc *desc);
12 12
13extern int __irq_set_trigger(struct irq_desc *desc, unsigned int irq, 13extern int __irq_set_trigger(struct irq_desc *desc, unsigned int irq,
14 unsigned long flags); 14 unsigned long flags);
15extern void __disable_irq(struct irq_desc *desc, unsigned int irq, bool susp);
16extern void __enable_irq(struct irq_desc *desc, unsigned int irq, bool resume);
15 17
16extern struct lock_class_key irq_desc_lock_class; 18extern struct lock_class_key irq_desc_lock_class;
17extern void init_kstat_irqs(struct irq_desc *desc, int cpu, int nr); 19extern void init_kstat_irqs(struct irq_desc *desc, int cpu, int nr);
20extern void clear_kstat_irqs(struct irq_desc *desc);
18extern spinlock_t sparse_irq_lock; 21extern spinlock_t sparse_irq_lock;
19 22
20#ifdef CONFIG_SPARSE_IRQ 23#ifdef CONFIG_SPARSE_IRQ
diff --git a/kernel/irq/manage.c b/kernel/irq/manage.c
index a3a5dc9ef346..1516ab77355c 100644
--- a/kernel/irq/manage.c
+++ b/kernel/irq/manage.c
@@ -109,7 +109,7 @@ int irq_set_affinity(unsigned int irq, const struct cpumask *cpumask)
109/* 109/*
110 * Generic version of the affinity autoselector. 110 * Generic version of the affinity autoselector.
111 */ 111 */
112int do_irq_select_affinity(unsigned int irq, struct irq_desc *desc) 112static int setup_affinity(unsigned int irq, struct irq_desc *desc)
113{ 113{
114 if (!irq_can_set_affinity(irq)) 114 if (!irq_can_set_affinity(irq))
115 return 0; 115 return 0;
@@ -133,7 +133,7 @@ set_affinity:
133 return 0; 133 return 0;
134} 134}
135#else 135#else
136static inline int do_irq_select_affinity(unsigned int irq, struct irq_desc *d) 136static inline int setup_affinity(unsigned int irq, struct irq_desc *d)
137{ 137{
138 return irq_select_affinity(irq); 138 return irq_select_affinity(irq);
139} 139}
@@ -149,19 +149,33 @@ int irq_select_affinity_usr(unsigned int irq)
149 int ret; 149 int ret;
150 150
151 spin_lock_irqsave(&desc->lock, flags); 151 spin_lock_irqsave(&desc->lock, flags);
152 ret = do_irq_select_affinity(irq, desc); 152 ret = setup_affinity(irq, desc);
153 spin_unlock_irqrestore(&desc->lock, flags); 153 spin_unlock_irqrestore(&desc->lock, flags);
154 154
155 return ret; 155 return ret;
156} 156}
157 157
158#else 158#else
159static inline int do_irq_select_affinity(int irq, struct irq_desc *desc) 159static inline int setup_affinity(unsigned int irq, struct irq_desc *desc)
160{ 160{
161 return 0; 161 return 0;
162} 162}
163#endif 163#endif
164 164
165void __disable_irq(struct irq_desc *desc, unsigned int irq, bool suspend)
166{
167 if (suspend) {
168 if (!desc->action || (desc->action->flags & IRQF_TIMER))
169 return;
170 desc->status |= IRQ_SUSPENDED;
171 }
172
173 if (!desc->depth++) {
174 desc->status |= IRQ_DISABLED;
175 desc->chip->disable(irq);
176 }
177}
178
165/** 179/**
166 * disable_irq_nosync - disable an irq without waiting 180 * disable_irq_nosync - disable an irq without waiting
167 * @irq: Interrupt to disable 181 * @irq: Interrupt to disable
@@ -182,10 +196,7 @@ void disable_irq_nosync(unsigned int irq)
182 return; 196 return;
183 197
184 spin_lock_irqsave(&desc->lock, flags); 198 spin_lock_irqsave(&desc->lock, flags);
185 if (!desc->depth++) { 199 __disable_irq(desc, irq, false);
186 desc->status |= IRQ_DISABLED;
187 desc->chip->disable(irq);
188 }
189 spin_unlock_irqrestore(&desc->lock, flags); 200 spin_unlock_irqrestore(&desc->lock, flags);
190} 201}
191EXPORT_SYMBOL(disable_irq_nosync); 202EXPORT_SYMBOL(disable_irq_nosync);
@@ -215,15 +226,21 @@ void disable_irq(unsigned int irq)
215} 226}
216EXPORT_SYMBOL(disable_irq); 227EXPORT_SYMBOL(disable_irq);
217 228
218static void __enable_irq(struct irq_desc *desc, unsigned int irq) 229void __enable_irq(struct irq_desc *desc, unsigned int irq, bool resume)
219{ 230{
231 if (resume)
232 desc->status &= ~IRQ_SUSPENDED;
233
220 switch (desc->depth) { 234 switch (desc->depth) {
221 case 0: 235 case 0:
236 err_out:
222 WARN(1, KERN_WARNING "Unbalanced enable for IRQ %d\n", irq); 237 WARN(1, KERN_WARNING "Unbalanced enable for IRQ %d\n", irq);
223 break; 238 break;
224 case 1: { 239 case 1: {
225 unsigned int status = desc->status & ~IRQ_DISABLED; 240 unsigned int status = desc->status & ~IRQ_DISABLED;
226 241
242 if (desc->status & IRQ_SUSPENDED)
243 goto err_out;
227 /* Prevent probing on this irq: */ 244 /* Prevent probing on this irq: */
228 desc->status = status | IRQ_NOPROBE; 245 desc->status = status | IRQ_NOPROBE;
229 check_irq_resend(desc, irq); 246 check_irq_resend(desc, irq);
@@ -253,7 +270,7 @@ void enable_irq(unsigned int irq)
253 return; 270 return;
254 271
255 spin_lock_irqsave(&desc->lock, flags); 272 spin_lock_irqsave(&desc->lock, flags);
256 __enable_irq(desc, irq); 273 __enable_irq(desc, irq, false);
257 spin_unlock_irqrestore(&desc->lock, flags); 274 spin_unlock_irqrestore(&desc->lock, flags);
258} 275}
259EXPORT_SYMBOL(enable_irq); 276EXPORT_SYMBOL(enable_irq);
@@ -389,9 +406,9 @@ int __irq_set_trigger(struct irq_desc *desc, unsigned int irq,
389 * allocate special interrupts that are part of the architecture. 406 * allocate special interrupts that are part of the architecture.
390 */ 407 */
391static int 408static int
392__setup_irq(unsigned int irq, struct irq_desc * desc, struct irqaction *new) 409__setup_irq(unsigned int irq, struct irq_desc *desc, struct irqaction *new)
393{ 410{
394 struct irqaction *old, **p; 411 struct irqaction *old, **old_ptr;
395 const char *old_name = NULL; 412 const char *old_name = NULL;
396 unsigned long flags; 413 unsigned long flags;
397 int shared = 0; 414 int shared = 0;
@@ -423,8 +440,8 @@ __setup_irq(unsigned int irq, struct irq_desc * desc, struct irqaction *new)
423 * The following block of code has to be executed atomically 440 * The following block of code has to be executed atomically
424 */ 441 */
425 spin_lock_irqsave(&desc->lock, flags); 442 spin_lock_irqsave(&desc->lock, flags);
426 p = &desc->action; 443 old_ptr = &desc->action;
427 old = *p; 444 old = *old_ptr;
428 if (old) { 445 if (old) {
429 /* 446 /*
430 * Can't share interrupts unless both agree to and are 447 * Can't share interrupts unless both agree to and are
@@ -447,8 +464,8 @@ __setup_irq(unsigned int irq, struct irq_desc * desc, struct irqaction *new)
447 464
448 /* add new interrupt at end of irq queue */ 465 /* add new interrupt at end of irq queue */
449 do { 466 do {
450 p = &old->next; 467 old_ptr = &old->next;
451 old = *p; 468 old = *old_ptr;
452 } while (old); 469 } while (old);
453 shared = 1; 470 shared = 1;
454 } 471 }
@@ -488,7 +505,7 @@ __setup_irq(unsigned int irq, struct irq_desc * desc, struct irqaction *new)
488 desc->status |= IRQ_NO_BALANCING; 505 desc->status |= IRQ_NO_BALANCING;
489 506
490 /* Set default affinity mask once everything is setup */ 507 /* Set default affinity mask once everything is setup */
491 do_irq_select_affinity(irq, desc); 508 setup_affinity(irq, desc);
492 509
493 } else if ((new->flags & IRQF_TRIGGER_MASK) 510 } else if ((new->flags & IRQF_TRIGGER_MASK)
494 && (new->flags & IRQF_TRIGGER_MASK) 511 && (new->flags & IRQF_TRIGGER_MASK)
@@ -499,7 +516,7 @@ __setup_irq(unsigned int irq, struct irq_desc * desc, struct irqaction *new)
499 (int)(new->flags & IRQF_TRIGGER_MASK)); 516 (int)(new->flags & IRQF_TRIGGER_MASK));
500 } 517 }
501 518
502 *p = new; 519 *old_ptr = new;
503 520
504 /* Reset broken irq detection when installing new handler */ 521 /* Reset broken irq detection when installing new handler */
505 desc->irq_count = 0; 522 desc->irq_count = 0;
@@ -511,7 +528,7 @@ __setup_irq(unsigned int irq, struct irq_desc * desc, struct irqaction *new)
511 */ 528 */
512 if (shared && (desc->status & IRQ_SPURIOUS_DISABLED)) { 529 if (shared && (desc->status & IRQ_SPURIOUS_DISABLED)) {
513 desc->status &= ~IRQ_SPURIOUS_DISABLED; 530 desc->status &= ~IRQ_SPURIOUS_DISABLED;
514 __enable_irq(desc, irq); 531 __enable_irq(desc, irq, false);
515 } 532 }
516 533
517 spin_unlock_irqrestore(&desc->lock, flags); 534 spin_unlock_irqrestore(&desc->lock, flags);
@@ -549,90 +566,117 @@ int setup_irq(unsigned int irq, struct irqaction *act)
549 566
550 return __setup_irq(irq, desc, act); 567 return __setup_irq(irq, desc, act);
551} 568}
569EXPORT_SYMBOL_GPL(setup_irq);
552 570
553/** 571 /*
554 * free_irq - free an interrupt 572 * Internal function to unregister an irqaction - used to free
555 * @irq: Interrupt line to free 573 * regular and special interrupts that are part of the architecture.
556 * @dev_id: Device identity to free
557 *
558 * Remove an interrupt handler. The handler is removed and if the
559 * interrupt line is no longer in use by any driver it is disabled.
560 * On a shared IRQ the caller must ensure the interrupt is disabled
561 * on the card it drives before calling this function. The function
562 * does not return until any executing interrupts for this IRQ
563 * have completed.
564 *
565 * This function must not be called from interrupt context.
566 */ 574 */
567void free_irq(unsigned int irq, void *dev_id) 575static struct irqaction *__free_irq(unsigned int irq, void *dev_id)
568{ 576{
569 struct irq_desc *desc = irq_to_desc(irq); 577 struct irq_desc *desc = irq_to_desc(irq);
570 struct irqaction **p; 578 struct irqaction *action, **action_ptr;
571 unsigned long flags; 579 unsigned long flags;
572 580
573 WARN_ON(in_interrupt()); 581 WARN(in_interrupt(), "Trying to free IRQ %d from IRQ context!\n", irq);
574 582
575 if (!desc) 583 if (!desc)
576 return; 584 return NULL;
577 585
578 spin_lock_irqsave(&desc->lock, flags); 586 spin_lock_irqsave(&desc->lock, flags);
579 p = &desc->action; 587
588 /*
589 * There can be multiple actions per IRQ descriptor, find the right
590 * one based on the dev_id:
591 */
592 action_ptr = &desc->action;
580 for (;;) { 593 for (;;) {
581 struct irqaction *action = *p; 594 action = *action_ptr;
582 595
583 if (action) { 596 if (!action) {
584 struct irqaction **pp = p; 597 WARN(1, "Trying to free already-free IRQ %d\n", irq);
598 spin_unlock_irqrestore(&desc->lock, flags);
585 599
586 p = &action->next; 600 return NULL;
587 if (action->dev_id != dev_id) 601 }
588 continue;
589 602
590 /* Found it - now remove it from the list of entries */ 603 if (action->dev_id == dev_id)
591 *pp = action->next; 604 break;
605 action_ptr = &action->next;
606 }
592 607
593 /* Currently used only by UML, might disappear one day.*/ 608 /* Found it - now remove it from the list of entries: */
609 *action_ptr = action->next;
610
611 /* Currently used only by UML, might disappear one day: */
594#ifdef CONFIG_IRQ_RELEASE_METHOD 612#ifdef CONFIG_IRQ_RELEASE_METHOD
595 if (desc->chip->release) 613 if (desc->chip->release)
596 desc->chip->release(irq, dev_id); 614 desc->chip->release(irq, dev_id);
597#endif 615#endif
598 616
599 if (!desc->action) { 617 /* If this was the last handler, shut down the IRQ line: */
600 desc->status |= IRQ_DISABLED; 618 if (!desc->action) {
601 if (desc->chip->shutdown) 619 desc->status |= IRQ_DISABLED;
602 desc->chip->shutdown(irq); 620 if (desc->chip->shutdown)
603 else 621 desc->chip->shutdown(irq);
604 desc->chip->disable(irq); 622 else
605 } 623 desc->chip->disable(irq);
606 spin_unlock_irqrestore(&desc->lock, flags); 624 }
607 unregister_handler_proc(irq, action); 625 spin_unlock_irqrestore(&desc->lock, flags);
626
627 unregister_handler_proc(irq, action);
628
629 /* Make sure it's not being used on another CPU: */
630 synchronize_irq(irq);
608 631
609 /* Make sure it's not being used on another CPU */
610 synchronize_irq(irq);
611#ifdef CONFIG_DEBUG_SHIRQ
612 /*
613 * It's a shared IRQ -- the driver ought to be
614 * prepared for it to happen even now it's
615 * being freed, so let's make sure.... We do
616 * this after actually deregistering it, to
617 * make sure that a 'real' IRQ doesn't run in
618 * parallel with our fake
619 */
620 if (action->flags & IRQF_SHARED) {
621 local_irq_save(flags);
622 action->handler(irq, dev_id);
623 local_irq_restore(flags);
624 }
625#endif
626 kfree(action);
627 return;
628 }
629 printk(KERN_ERR "Trying to free already-free IRQ %d\n", irq);
630#ifdef CONFIG_DEBUG_SHIRQ 632#ifdef CONFIG_DEBUG_SHIRQ
631 dump_stack(); 633 /*
632#endif 634 * It's a shared IRQ -- the driver ought to be prepared for an IRQ
633 spin_unlock_irqrestore(&desc->lock, flags); 635 * event to happen even now it's being freed, so let's make sure that
634 return; 636 * is so by doing an extra call to the handler ....
637 *
638 * ( We do this after actually deregistering it, to make sure that a
639 * 'real' IRQ doesn't run in * parallel with our fake. )
640 */
641 if (action->flags & IRQF_SHARED) {
642 local_irq_save(flags);
643 action->handler(irq, dev_id);
644 local_irq_restore(flags);
635 } 645 }
646#endif
647 return action;
648}
649
650/**
651 * remove_irq - free an interrupt
652 * @irq: Interrupt line to free
653 * @act: irqaction for the interrupt
654 *
655 * Used to remove interrupts statically setup by the early boot process.
656 */
657void remove_irq(unsigned int irq, struct irqaction *act)
658{
659 __free_irq(irq, act->dev_id);
660}
661EXPORT_SYMBOL_GPL(remove_irq);
662
663/**
664 * free_irq - free an interrupt allocated with request_irq
665 * @irq: Interrupt line to free
666 * @dev_id: Device identity to free
667 *
668 * Remove an interrupt handler. The handler is removed and if the
669 * interrupt line is no longer in use by any driver it is disabled.
670 * On a shared IRQ the caller must ensure the interrupt is disabled
671 * on the card it drives before calling this function. The function
672 * does not return until any executing interrupts for this IRQ
673 * have completed.
674 *
675 * This function must not be called from interrupt context.
676 */
677void free_irq(unsigned int irq, void *dev_id)
678{
679 kfree(__free_irq(irq, dev_id));
636} 680}
637EXPORT_SYMBOL(free_irq); 681EXPORT_SYMBOL(free_irq);
638 682
@@ -679,11 +723,12 @@ int request_irq(unsigned int irq, irq_handler_t handler,
679 * the behavior is classified as "will not fix" so we need to 723 * the behavior is classified as "will not fix" so we need to
680 * start nudging drivers away from using that idiom. 724 * start nudging drivers away from using that idiom.
681 */ 725 */
682 if ((irqflags & (IRQF_SHARED|IRQF_DISABLED)) 726 if ((irqflags & (IRQF_SHARED|IRQF_DISABLED)) ==
683 == (IRQF_SHARED|IRQF_DISABLED)) 727 (IRQF_SHARED|IRQF_DISABLED)) {
684 pr_warning("IRQ %d/%s: IRQF_DISABLED is not " 728 pr_warning(
685 "guaranteed on shared IRQs\n", 729 "IRQ %d/%s: IRQF_DISABLED is not guaranteed on shared IRQs\n",
686 irq, devname); 730 irq, devname);
731 }
687 732
688#ifdef CONFIG_LOCKDEP 733#ifdef CONFIG_LOCKDEP
689 /* 734 /*
@@ -709,15 +754,13 @@ int request_irq(unsigned int irq, irq_handler_t handler,
709 if (!handler) 754 if (!handler)
710 return -EINVAL; 755 return -EINVAL;
711 756
712 action = kmalloc(sizeof(struct irqaction), GFP_ATOMIC); 757 action = kzalloc(sizeof(struct irqaction), GFP_KERNEL);
713 if (!action) 758 if (!action)
714 return -ENOMEM; 759 return -ENOMEM;
715 760
716 action->handler = handler; 761 action->handler = handler;
717 action->flags = irqflags; 762 action->flags = irqflags;
718 cpus_clear(action->mask);
719 action->name = devname; 763 action->name = devname;
720 action->next = NULL;
721 action->dev_id = dev_id; 764 action->dev_id = dev_id;
722 765
723 retval = __setup_irq(irq, desc, action); 766 retval = __setup_irq(irq, desc, action);
diff --git a/kernel/irq/numa_migrate.c b/kernel/irq/numa_migrate.c
index 7f9b80434e32..243d6121e50e 100644
--- a/kernel/irq/numa_migrate.c
+++ b/kernel/irq/numa_migrate.c
@@ -17,16 +17,11 @@ static void init_copy_kstat_irqs(struct irq_desc *old_desc,
17 struct irq_desc *desc, 17 struct irq_desc *desc,
18 int cpu, int nr) 18 int cpu, int nr)
19{ 19{
20 unsigned long bytes;
21
22 init_kstat_irqs(desc, cpu, nr); 20 init_kstat_irqs(desc, cpu, nr);
23 21
24 if (desc->kstat_irqs != old_desc->kstat_irqs) { 22 if (desc->kstat_irqs != old_desc->kstat_irqs)
25 /* Compute how many bytes we need per irq and allocate them */ 23 memcpy(desc->kstat_irqs, old_desc->kstat_irqs,
26 bytes = nr * sizeof(unsigned int); 24 nr * sizeof(*desc->kstat_irqs));
27
28 memcpy(desc->kstat_irqs, old_desc->kstat_irqs, bytes);
29 }
30} 25}
31 26
32static void free_kstat_irqs(struct irq_desc *old_desc, struct irq_desc *desc) 27static void free_kstat_irqs(struct irq_desc *old_desc, struct irq_desc *desc)
diff --git a/kernel/irq/pm.c b/kernel/irq/pm.c
new file mode 100644
index 000000000000..638d8bedec14
--- /dev/null
+++ b/kernel/irq/pm.c
@@ -0,0 +1,79 @@
1/*
2 * linux/kernel/irq/pm.c
3 *
4 * Copyright (C) 2009 Rafael J. Wysocki <rjw@sisk.pl>, Novell Inc.
5 *
6 * This file contains power management functions related to interrupts.
7 */
8
9#include <linux/irq.h>
10#include <linux/module.h>
11#include <linux/interrupt.h>
12
13#include "internals.h"
14
15/**
16 * suspend_device_irqs - disable all currently enabled interrupt lines
17 *
18 * During system-wide suspend or hibernation device interrupts need to be
19 * disabled at the chip level and this function is provided for this purpose.
20 * It disables all interrupt lines that are enabled at the moment and sets the
21 * IRQ_SUSPENDED flag for them.
22 */
23void suspend_device_irqs(void)
24{
25 struct irq_desc *desc;
26 int irq;
27
28 for_each_irq_desc(irq, desc) {
29 unsigned long flags;
30
31 spin_lock_irqsave(&desc->lock, flags);
32 __disable_irq(desc, irq, true);
33 spin_unlock_irqrestore(&desc->lock, flags);
34 }
35
36 for_each_irq_desc(irq, desc)
37 if (desc->status & IRQ_SUSPENDED)
38 synchronize_irq(irq);
39}
40EXPORT_SYMBOL_GPL(suspend_device_irqs);
41
42/**
43 * resume_device_irqs - enable interrupt lines disabled by suspend_device_irqs()
44 *
45 * Enable all interrupt lines previously disabled by suspend_device_irqs() that
46 * have the IRQ_SUSPENDED flag set.
47 */
48void resume_device_irqs(void)
49{
50 struct irq_desc *desc;
51 int irq;
52
53 for_each_irq_desc(irq, desc) {
54 unsigned long flags;
55
56 if (!(desc->status & IRQ_SUSPENDED))
57 continue;
58
59 spin_lock_irqsave(&desc->lock, flags);
60 __enable_irq(desc, irq, true);
61 spin_unlock_irqrestore(&desc->lock, flags);
62 }
63}
64EXPORT_SYMBOL_GPL(resume_device_irqs);
65
66/**
67 * check_wakeup_irqs - check if any wake-up interrupts are pending
68 */
69int check_wakeup_irqs(void)
70{
71 struct irq_desc *desc;
72 int irq;
73
74 for_each_irq_desc(irq, desc)
75 if ((desc->status & IRQ_WAKEUP) && (desc->status & IRQ_PENDING))
76 return -EBUSY;
77
78 return 0;
79}
diff --git a/kernel/irq/spurious.c b/kernel/irq/spurious.c
index dd364c11e56e..4d568294de3e 100644
--- a/kernel/irq/spurious.c
+++ b/kernel/irq/spurious.c
@@ -104,7 +104,7 @@ static int misrouted_irq(int irq)
104 return ok; 104 return ok;
105} 105}
106 106
107static void poll_spurious_irqs(unsigned long dummy) 107static void poll_all_shared_irqs(void)
108{ 108{
109 struct irq_desc *desc; 109 struct irq_desc *desc;
110 int i; 110 int i;
@@ -123,11 +123,23 @@ static void poll_spurious_irqs(unsigned long dummy)
123 123
124 try_one_irq(i, desc); 124 try_one_irq(i, desc);
125 } 125 }
126}
127
128static void poll_spurious_irqs(unsigned long dummy)
129{
130 poll_all_shared_irqs();
126 131
127 mod_timer(&poll_spurious_irq_timer, 132 mod_timer(&poll_spurious_irq_timer,
128 jiffies + POLL_SPURIOUS_IRQ_INTERVAL); 133 jiffies + POLL_SPURIOUS_IRQ_INTERVAL);
129} 134}
130 135
136#ifdef CONFIG_DEBUG_SHIRQ
137void debug_poll_all_shared_irqs(void)
138{
139 poll_all_shared_irqs();
140}
141#endif
142
131/* 143/*
132 * If 99,900 of the previous 100,000 interrupts have not been handled 144 * If 99,900 of the previous 100,000 interrupts have not been handled
133 * then assume that the IRQ is stuck in some manner. Drop a diagnostic 145 * then assume that the IRQ is stuck in some manner. Drop a diagnostic
diff --git a/kernel/kallsyms.c b/kernel/kallsyms.c
index 7b8b0f21a5b1..374faf9bfdc7 100644
--- a/kernel/kallsyms.c
+++ b/kernel/kallsyms.c
@@ -161,6 +161,25 @@ unsigned long kallsyms_lookup_name(const char *name)
161 return module_kallsyms_lookup_name(name); 161 return module_kallsyms_lookup_name(name);
162} 162}
163 163
164int kallsyms_on_each_symbol(int (*fn)(void *, const char *, struct module *,
165 unsigned long),
166 void *data)
167{
168 char namebuf[KSYM_NAME_LEN];
169 unsigned long i;
170 unsigned int off;
171 int ret;
172
173 for (i = 0, off = 0; i < kallsyms_num_syms; i++) {
174 off = kallsyms_expand_symbol(off, namebuf);
175 ret = fn(data, namebuf, NULL, kallsyms_addresses[i]);
176 if (ret != 0)
177 return ret;
178 }
179 return module_kallsyms_on_each_symbol(fn, data);
180}
181EXPORT_SYMBOL_GPL(kallsyms_on_each_symbol);
182
164static unsigned long get_symbol_pos(unsigned long addr, 183static unsigned long get_symbol_pos(unsigned long addr,
165 unsigned long *symbolsize, 184 unsigned long *symbolsize,
166 unsigned long *offset) 185 unsigned long *offset)
diff --git a/kernel/kexec.c b/kernel/kexec.c
index c7fd6692939d..5a758c6e4950 100644
--- a/kernel/kexec.c
+++ b/kernel/kexec.c
@@ -42,7 +42,7 @@
42note_buf_t* crash_notes; 42note_buf_t* crash_notes;
43 43
44/* vmcoreinfo stuff */ 44/* vmcoreinfo stuff */
45unsigned char vmcoreinfo_data[VMCOREINFO_BYTES]; 45static unsigned char vmcoreinfo_data[VMCOREINFO_BYTES];
46u32 vmcoreinfo_note[VMCOREINFO_NOTE_SIZE/4]; 46u32 vmcoreinfo_note[VMCOREINFO_NOTE_SIZE/4];
47size_t vmcoreinfo_size; 47size_t vmcoreinfo_size;
48size_t vmcoreinfo_max_size = sizeof(vmcoreinfo_data); 48size_t vmcoreinfo_max_size = sizeof(vmcoreinfo_data);
@@ -1409,6 +1409,7 @@ static int __init crash_save_vmcoreinfo_init(void)
1409 VMCOREINFO_OFFSET(list_head, prev); 1409 VMCOREINFO_OFFSET(list_head, prev);
1410 VMCOREINFO_OFFSET(vm_struct, addr); 1410 VMCOREINFO_OFFSET(vm_struct, addr);
1411 VMCOREINFO_LENGTH(zone.free_area, MAX_ORDER); 1411 VMCOREINFO_LENGTH(zone.free_area, MAX_ORDER);
1412 log_buf_kexec_setup();
1412 VMCOREINFO_LENGTH(free_area.free_list, MIGRATE_TYPES); 1413 VMCOREINFO_LENGTH(free_area.free_list, MIGRATE_TYPES);
1413 VMCOREINFO_NUMBER(NR_FREE_PAGES); 1414 VMCOREINFO_NUMBER(NR_FREE_PAGES);
1414 VMCOREINFO_NUMBER(PG_lru); 1415 VMCOREINFO_NUMBER(PG_lru);
@@ -1450,11 +1451,7 @@ int kernel_kexec(void)
1450 error = device_suspend(PMSG_FREEZE); 1451 error = device_suspend(PMSG_FREEZE);
1451 if (error) 1452 if (error)
1452 goto Resume_console; 1453 goto Resume_console;
1453 error = disable_nonboot_cpus();
1454 if (error)
1455 goto Resume_devices;
1456 device_pm_lock(); 1454 device_pm_lock();
1457 local_irq_disable();
1458 /* At this point, device_suspend() has been called, 1455 /* At this point, device_suspend() has been called,
1459 * but *not* device_power_down(). We *must* 1456 * but *not* device_power_down(). We *must*
1460 * device_power_down() now. Otherwise, drivers for 1457 * device_power_down() now. Otherwise, drivers for
@@ -1464,12 +1461,15 @@ int kernel_kexec(void)
1464 */ 1461 */
1465 error = device_power_down(PMSG_FREEZE); 1462 error = device_power_down(PMSG_FREEZE);
1466 if (error) 1463 if (error)
1467 goto Enable_irqs; 1464 goto Resume_devices;
1468 1465 error = disable_nonboot_cpus();
1466 if (error)
1467 goto Enable_cpus;
1468 local_irq_disable();
1469 /* Suspend system devices */ 1469 /* Suspend system devices */
1470 error = sysdev_suspend(PMSG_FREEZE); 1470 error = sysdev_suspend(PMSG_FREEZE);
1471 if (error) 1471 if (error)
1472 goto Power_up_devices; 1472 goto Enable_irqs;
1473 } else 1473 } else
1474#endif 1474#endif
1475 { 1475 {
@@ -1483,13 +1483,13 @@ int kernel_kexec(void)
1483#ifdef CONFIG_KEXEC_JUMP 1483#ifdef CONFIG_KEXEC_JUMP
1484 if (kexec_image->preserve_context) { 1484 if (kexec_image->preserve_context) {
1485 sysdev_resume(); 1485 sysdev_resume();
1486 Power_up_devices:
1487 device_power_up(PMSG_RESTORE);
1488 Enable_irqs: 1486 Enable_irqs:
1489 local_irq_enable(); 1487 local_irq_enable();
1490 device_pm_unlock(); 1488 Enable_cpus:
1491 enable_nonboot_cpus(); 1489 enable_nonboot_cpus();
1490 device_power_up(PMSG_RESTORE);
1492 Resume_devices: 1491 Resume_devices:
1492 device_pm_unlock();
1493 device_resume(PMSG_RESTORE); 1493 device_resume(PMSG_RESTORE);
1494 Resume_console: 1494 Resume_console:
1495 resume_console(); 1495 resume_console();
diff --git a/kernel/kmod.c b/kernel/kmod.c
index a27a5f64443d..b750675251e5 100644
--- a/kernel/kmod.c
+++ b/kernel/kmod.c
@@ -50,7 +50,8 @@ static struct workqueue_struct *khelper_wq;
50char modprobe_path[KMOD_PATH_LEN] = "/sbin/modprobe"; 50char modprobe_path[KMOD_PATH_LEN] = "/sbin/modprobe";
51 51
52/** 52/**
53 * request_module - try to load a kernel module 53 * __request_module - try to load a kernel module
54 * @wait: wait (or not) for the operation to complete
54 * @fmt: printf style format string for the name of the module 55 * @fmt: printf style format string for the name of the module
55 * @...: arguments as specified in the format string 56 * @...: arguments as specified in the format string
56 * 57 *
@@ -63,7 +64,7 @@ char modprobe_path[KMOD_PATH_LEN] = "/sbin/modprobe";
63 * If module auto-loading support is disabled then this function 64 * If module auto-loading support is disabled then this function
64 * becomes a no-operation. 65 * becomes a no-operation.
65 */ 66 */
66int request_module(const char *fmt, ...) 67int __request_module(bool wait, const char *fmt, ...)
67{ 68{
68 va_list args; 69 va_list args;
69 char module_name[MODULE_NAME_LEN]; 70 char module_name[MODULE_NAME_LEN];
@@ -108,11 +109,12 @@ int request_module(const char *fmt, ...)
108 return -ENOMEM; 109 return -ENOMEM;
109 } 110 }
110 111
111 ret = call_usermodehelper(modprobe_path, argv, envp, 1); 112 ret = call_usermodehelper(modprobe_path, argv, envp,
113 wait ? UMH_WAIT_PROC : UMH_WAIT_EXEC);
112 atomic_dec(&kmod_concurrent); 114 atomic_dec(&kmod_concurrent);
113 return ret; 115 return ret;
114} 116}
115EXPORT_SYMBOL(request_module); 117EXPORT_SYMBOL(__request_module);
116#endif /* CONFIG_MODULES */ 118#endif /* CONFIG_MODULES */
117 119
118struct subprocess_info { 120struct subprocess_info {
@@ -167,7 +169,7 @@ static int ____call_usermodehelper(void *data)
167 } 169 }
168 170
169 /* We can run anywhere, unlike our parent keventd(). */ 171 /* We can run anywhere, unlike our parent keventd(). */
170 set_cpus_allowed_ptr(current, CPU_MASK_ALL_PTR); 172 set_cpus_allowed_ptr(current, cpu_all_mask);
171 173
172 /* 174 /*
173 * Our parent is keventd, which runs with elevated scheduling priority. 175 * Our parent is keventd, which runs with elevated scheduling priority.
diff --git a/kernel/kprobes.c b/kernel/kprobes.c
index 7ba8cd9845cb..5016bfb682b9 100644
--- a/kernel/kprobes.c
+++ b/kernel/kprobes.c
@@ -43,6 +43,7 @@
43#include <linux/seq_file.h> 43#include <linux/seq_file.h>
44#include <linux/debugfs.h> 44#include <linux/debugfs.h>
45#include <linux/kdebug.h> 45#include <linux/kdebug.h>
46#include <linux/memory.h>
46 47
47#include <asm-generic/sections.h> 48#include <asm-generic/sections.h>
48#include <asm/cacheflush.h> 49#include <asm/cacheflush.h>
@@ -699,9 +700,10 @@ int __kprobes register_kprobe(struct kprobe *p)
699 goto out; 700 goto out;
700 } 701 }
701 702
703 mutex_lock(&text_mutex);
702 ret = arch_prepare_kprobe(p); 704 ret = arch_prepare_kprobe(p);
703 if (ret) 705 if (ret)
704 goto out; 706 goto out_unlock_text;
705 707
706 INIT_HLIST_NODE(&p->hlist); 708 INIT_HLIST_NODE(&p->hlist);
707 hlist_add_head_rcu(&p->hlist, 709 hlist_add_head_rcu(&p->hlist,
@@ -710,6 +712,8 @@ int __kprobes register_kprobe(struct kprobe *p)
710 if (kprobe_enabled) 712 if (kprobe_enabled)
711 arch_arm_kprobe(p); 713 arch_arm_kprobe(p);
712 714
715out_unlock_text:
716 mutex_unlock(&text_mutex);
713out: 717out:
714 mutex_unlock(&kprobe_mutex); 718 mutex_unlock(&kprobe_mutex);
715 719
@@ -746,8 +750,11 @@ valid_p:
746 * enabled and not gone - otherwise, the breakpoint would 750 * enabled and not gone - otherwise, the breakpoint would
747 * already have been removed. We save on flushing icache. 751 * already have been removed. We save on flushing icache.
748 */ 752 */
749 if (kprobe_enabled && !kprobe_gone(old_p)) 753 if (kprobe_enabled && !kprobe_gone(old_p)) {
754 mutex_lock(&text_mutex);
750 arch_disarm_kprobe(p); 755 arch_disarm_kprobe(p);
756 mutex_unlock(&text_mutex);
757 }
751 hlist_del_rcu(&old_p->hlist); 758 hlist_del_rcu(&old_p->hlist);
752 } else { 759 } else {
753 if (p->break_handler && !kprobe_gone(p)) 760 if (p->break_handler && !kprobe_gone(p))
@@ -912,10 +919,8 @@ static int __kprobes pre_handler_kretprobe(struct kprobe *p,
912 ri->rp = rp; 919 ri->rp = rp;
913 ri->task = current; 920 ri->task = current;
914 921
915 if (rp->entry_handler && rp->entry_handler(ri, regs)) { 922 if (rp->entry_handler && rp->entry_handler(ri, regs))
916 spin_unlock_irqrestore(&rp->lock, flags);
917 return 0; 923 return 0;
918 }
919 924
920 arch_prepare_kretprobe(ri, regs); 925 arch_prepare_kretprobe(ri, regs);
921 926
@@ -1280,12 +1285,14 @@ static void __kprobes enable_all_kprobes(void)
1280 if (kprobe_enabled) 1285 if (kprobe_enabled)
1281 goto already_enabled; 1286 goto already_enabled;
1282 1287
1288 mutex_lock(&text_mutex);
1283 for (i = 0; i < KPROBE_TABLE_SIZE; i++) { 1289 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
1284 head = &kprobe_table[i]; 1290 head = &kprobe_table[i];
1285 hlist_for_each_entry_rcu(p, node, head, hlist) 1291 hlist_for_each_entry_rcu(p, node, head, hlist)
1286 if (!kprobe_gone(p)) 1292 if (!kprobe_gone(p))
1287 arch_arm_kprobe(p); 1293 arch_arm_kprobe(p);
1288 } 1294 }
1295 mutex_unlock(&text_mutex);
1289 1296
1290 kprobe_enabled = true; 1297 kprobe_enabled = true;
1291 printk(KERN_INFO "Kprobes globally enabled\n"); 1298 printk(KERN_INFO "Kprobes globally enabled\n");
@@ -1310,6 +1317,7 @@ static void __kprobes disable_all_kprobes(void)
1310 1317
1311 kprobe_enabled = false; 1318 kprobe_enabled = false;
1312 printk(KERN_INFO "Kprobes globally disabled\n"); 1319 printk(KERN_INFO "Kprobes globally disabled\n");
1320 mutex_lock(&text_mutex);
1313 for (i = 0; i < KPROBE_TABLE_SIZE; i++) { 1321 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
1314 head = &kprobe_table[i]; 1322 head = &kprobe_table[i];
1315 hlist_for_each_entry_rcu(p, node, head, hlist) { 1323 hlist_for_each_entry_rcu(p, node, head, hlist) {
@@ -1318,6 +1326,7 @@ static void __kprobes disable_all_kprobes(void)
1318 } 1326 }
1319 } 1327 }
1320 1328
1329 mutex_unlock(&text_mutex);
1321 mutex_unlock(&kprobe_mutex); 1330 mutex_unlock(&kprobe_mutex);
1322 /* Allow all currently running kprobes to complete */ 1331 /* Allow all currently running kprobes to complete */
1323 synchronize_sched(); 1332 synchronize_sched();
diff --git a/kernel/kthread.c b/kernel/kthread.c
index 4fbc456f393d..84bbadd4d021 100644
--- a/kernel/kthread.c
+++ b/kernel/kthread.c
@@ -110,7 +110,7 @@ static void create_kthread(struct kthread_create_info *create)
110 */ 110 */
111 sched_setscheduler(create->result, SCHED_NORMAL, &param); 111 sched_setscheduler(create->result, SCHED_NORMAL, &param);
112 set_user_nice(create->result, KTHREAD_NICE_LEVEL); 112 set_user_nice(create->result, KTHREAD_NICE_LEVEL);
113 set_cpus_allowed_ptr(create->result, CPU_MASK_ALL_PTR); 113 set_cpus_allowed_ptr(create->result, cpu_all_mask);
114 } 114 }
115 complete(&create->done); 115 complete(&create->done);
116} 116}
@@ -240,7 +240,7 @@ int kthreadd(void *unused)
240 set_task_comm(tsk, "kthreadd"); 240 set_task_comm(tsk, "kthreadd");
241 ignore_signals(tsk); 241 ignore_signals(tsk);
242 set_user_nice(tsk, KTHREAD_NICE_LEVEL); 242 set_user_nice(tsk, KTHREAD_NICE_LEVEL);
243 set_cpus_allowed_ptr(tsk, CPU_MASK_ALL_PTR); 243 set_cpus_allowed_ptr(tsk, cpu_all_mask);
244 244
245 current->flags |= PF_NOFREEZE | PF_FREEZER_NOSIG; 245 current->flags |= PF_NOFREEZE | PF_FREEZER_NOSIG;
246 246
diff --git a/kernel/latencytop.c b/kernel/latencytop.c
index 449db466bdbc..ca07c5c0c914 100644
--- a/kernel/latencytop.c
+++ b/kernel/latencytop.c
@@ -9,6 +9,44 @@
9 * as published by the Free Software Foundation; version 2 9 * as published by the Free Software Foundation; version 2
10 * of the License. 10 * of the License.
11 */ 11 */
12
13/*
14 * CONFIG_LATENCYTOP enables a kernel latency tracking infrastructure that is
15 * used by the "latencytop" userspace tool. The latency that is tracked is not
16 * the 'traditional' interrupt latency (which is primarily caused by something
17 * else consuming CPU), but instead, it is the latency an application encounters
18 * because the kernel sleeps on its behalf for various reasons.
19 *
20 * This code tracks 2 levels of statistics:
21 * 1) System level latency
22 * 2) Per process latency
23 *
24 * The latency is stored in fixed sized data structures in an accumulated form;
25 * if the "same" latency cause is hit twice, this will be tracked as one entry
26 * in the data structure. Both the count, total accumulated latency and maximum
27 * latency are tracked in this data structure. When the fixed size structure is
28 * full, no new causes are tracked until the buffer is flushed by writing to
29 * the /proc file; the userspace tool does this on a regular basis.
30 *
31 * A latency cause is identified by a stringified backtrace at the point that
32 * the scheduler gets invoked. The userland tool will use this string to
33 * identify the cause of the latency in human readable form.
34 *
35 * The information is exported via /proc/latency_stats and /proc/<pid>/latency.
36 * These files look like this:
37 *
38 * Latency Top version : v0.1
39 * 70 59433 4897 i915_irq_wait drm_ioctl vfs_ioctl do_vfs_ioctl sys_ioctl
40 * | | | |
41 * | | | +----> the stringified backtrace
42 * | | +---------> The maximum latency for this entry in microseconds
43 * | +--------------> The accumulated latency for this entry (microseconds)
44 * +-------------------> The number of times this entry is hit
45 *
46 * (note: the average latency is the accumulated latency divided by the number
47 * of times)
48 */
49
12#include <linux/latencytop.h> 50#include <linux/latencytop.h>
13#include <linux/kallsyms.h> 51#include <linux/kallsyms.h>
14#include <linux/seq_file.h> 52#include <linux/seq_file.h>
@@ -72,7 +110,7 @@ account_global_scheduler_latency(struct task_struct *tsk, struct latency_record
72 firstnonnull = i; 110 firstnonnull = i;
73 continue; 111 continue;
74 } 112 }
75 for (q = 0 ; q < LT_BACKTRACEDEPTH ; q++) { 113 for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
76 unsigned long record = lat->backtrace[q]; 114 unsigned long record = lat->backtrace[q];
77 115
78 if (latency_record[i].backtrace[q] != record) { 116 if (latency_record[i].backtrace[q] != record) {
@@ -101,31 +139,52 @@ account_global_scheduler_latency(struct task_struct *tsk, struct latency_record
101 memcpy(&latency_record[i], lat, sizeof(struct latency_record)); 139 memcpy(&latency_record[i], lat, sizeof(struct latency_record));
102} 140}
103 141
104static inline void store_stacktrace(struct task_struct *tsk, struct latency_record *lat) 142/*
143 * Iterator to store a backtrace into a latency record entry
144 */
145static inline void store_stacktrace(struct task_struct *tsk,
146 struct latency_record *lat)
105{ 147{
106 struct stack_trace trace; 148 struct stack_trace trace;
107 149
108 memset(&trace, 0, sizeof(trace)); 150 memset(&trace, 0, sizeof(trace));
109 trace.max_entries = LT_BACKTRACEDEPTH; 151 trace.max_entries = LT_BACKTRACEDEPTH;
110 trace.entries = &lat->backtrace[0]; 152 trace.entries = &lat->backtrace[0];
111 trace.skip = 0;
112 save_stack_trace_tsk(tsk, &trace); 153 save_stack_trace_tsk(tsk, &trace);
113} 154}
114 155
156/**
157 * __account_scheduler_latency - record an occured latency
158 * @tsk - the task struct of the task hitting the latency
159 * @usecs - the duration of the latency in microseconds
160 * @inter - 1 if the sleep was interruptible, 0 if uninterruptible
161 *
162 * This function is the main entry point for recording latency entries
163 * as called by the scheduler.
164 *
165 * This function has a few special cases to deal with normal 'non-latency'
166 * sleeps: specifically, interruptible sleep longer than 5 msec is skipped
167 * since this usually is caused by waiting for events via select() and co.
168 *
169 * Negative latencies (caused by time going backwards) are also explicitly
170 * skipped.
171 */
115void __sched 172void __sched
116account_scheduler_latency(struct task_struct *tsk, int usecs, int inter) 173__account_scheduler_latency(struct task_struct *tsk, int usecs, int inter)
117{ 174{
118 unsigned long flags; 175 unsigned long flags;
119 int i, q; 176 int i, q;
120 struct latency_record lat; 177 struct latency_record lat;
121 178
122 if (!latencytop_enabled)
123 return;
124
125 /* Long interruptible waits are generally user requested... */ 179 /* Long interruptible waits are generally user requested... */
126 if (inter && usecs > 5000) 180 if (inter && usecs > 5000)
127 return; 181 return;
128 182
183 /* Negative sleeps are time going backwards */
184 /* Zero-time sleeps are non-interesting */
185 if (usecs <= 0)
186 return;
187
129 memset(&lat, 0, sizeof(lat)); 188 memset(&lat, 0, sizeof(lat));
130 lat.count = 1; 189 lat.count = 1;
131 lat.time = usecs; 190 lat.time = usecs;
@@ -143,12 +202,12 @@ account_scheduler_latency(struct task_struct *tsk, int usecs, int inter)
143 if (tsk->latency_record_count >= LT_SAVECOUNT) 202 if (tsk->latency_record_count >= LT_SAVECOUNT)
144 goto out_unlock; 203 goto out_unlock;
145 204
146 for (i = 0; i < LT_SAVECOUNT ; i++) { 205 for (i = 0; i < LT_SAVECOUNT; i++) {
147 struct latency_record *mylat; 206 struct latency_record *mylat;
148 int same = 1; 207 int same = 1;
149 208
150 mylat = &tsk->latency_record[i]; 209 mylat = &tsk->latency_record[i];
151 for (q = 0 ; q < LT_BACKTRACEDEPTH ; q++) { 210 for (q = 0; q < LT_BACKTRACEDEPTH; q++) {
152 unsigned long record = lat.backtrace[q]; 211 unsigned long record = lat.backtrace[q];
153 212
154 if (mylat->backtrace[q] != record) { 213 if (mylat->backtrace[q] != record) {
@@ -186,7 +245,7 @@ static int lstats_show(struct seq_file *m, void *v)
186 for (i = 0; i < MAXLR; i++) { 245 for (i = 0; i < MAXLR; i++) {
187 if (latency_record[i].backtrace[0]) { 246 if (latency_record[i].backtrace[0]) {
188 int q; 247 int q;
189 seq_printf(m, "%i %li %li ", 248 seq_printf(m, "%i %lu %lu ",
190 latency_record[i].count, 249 latency_record[i].count,
191 latency_record[i].time, 250 latency_record[i].time,
192 latency_record[i].max); 251 latency_record[i].max);
@@ -223,7 +282,7 @@ static int lstats_open(struct inode *inode, struct file *filp)
223 return single_open(filp, lstats_show, NULL); 282 return single_open(filp, lstats_show, NULL);
224} 283}
225 284
226static struct file_operations lstats_fops = { 285static const struct file_operations lstats_fops = {
227 .open = lstats_open, 286 .open = lstats_open,
228 .read = seq_read, 287 .read = seq_read,
229 .write = lstats_write, 288 .write = lstats_write,
@@ -236,4 +295,4 @@ static int __init init_lstats_procfs(void)
236 proc_create("latency_stats", 0644, NULL, &lstats_fops); 295 proc_create("latency_stats", 0644, NULL, &lstats_fops);
237 return 0; 296 return 0;
238} 297}
239__initcall(init_lstats_procfs); 298device_initcall(init_lstats_procfs);
diff --git a/kernel/lockdep.c b/kernel/lockdep.c
index 06b0c3568f0b..81b5f33970b8 100644
--- a/kernel/lockdep.c
+++ b/kernel/lockdep.c
@@ -41,6 +41,8 @@
41#include <linux/utsname.h> 41#include <linux/utsname.h>
42#include <linux/hash.h> 42#include <linux/hash.h>
43#include <linux/ftrace.h> 43#include <linux/ftrace.h>
44#include <linux/stringify.h>
45#include <trace/lockdep.h>
44 46
45#include <asm/sections.h> 47#include <asm/sections.h>
46 48
@@ -310,12 +312,14 @@ EXPORT_SYMBOL(lockdep_on);
310#if VERBOSE 312#if VERBOSE
311# define HARDIRQ_VERBOSE 1 313# define HARDIRQ_VERBOSE 1
312# define SOFTIRQ_VERBOSE 1 314# define SOFTIRQ_VERBOSE 1
315# define RECLAIM_VERBOSE 1
313#else 316#else
314# define HARDIRQ_VERBOSE 0 317# define HARDIRQ_VERBOSE 0
315# define SOFTIRQ_VERBOSE 0 318# define SOFTIRQ_VERBOSE 0
319# define RECLAIM_VERBOSE 0
316#endif 320#endif
317 321
318#if VERBOSE || HARDIRQ_VERBOSE || SOFTIRQ_VERBOSE 322#if VERBOSE || HARDIRQ_VERBOSE || SOFTIRQ_VERBOSE || RECLAIM_VERBOSE
319/* 323/*
320 * Quick filtering for interesting events: 324 * Quick filtering for interesting events:
321 */ 325 */
@@ -430,30 +434,24 @@ atomic_t nr_find_usage_forwards_checks;
430atomic_t nr_find_usage_forwards_recursions; 434atomic_t nr_find_usage_forwards_recursions;
431atomic_t nr_find_usage_backwards_checks; 435atomic_t nr_find_usage_backwards_checks;
432atomic_t nr_find_usage_backwards_recursions; 436atomic_t nr_find_usage_backwards_recursions;
433# define debug_atomic_inc(ptr) atomic_inc(ptr)
434# define debug_atomic_dec(ptr) atomic_dec(ptr)
435# define debug_atomic_read(ptr) atomic_read(ptr)
436#else
437# define debug_atomic_inc(ptr) do { } while (0)
438# define debug_atomic_dec(ptr) do { } while (0)
439# define debug_atomic_read(ptr) 0
440#endif 437#endif
441 438
442/* 439/*
443 * Locking printouts: 440 * Locking printouts:
444 */ 441 */
445 442
443#define __USAGE(__STATE) \
444 [LOCK_USED_IN_##__STATE] = "IN-"__stringify(__STATE)"-W", \
445 [LOCK_ENABLED_##__STATE] = __stringify(__STATE)"-ON-W", \
446 [LOCK_USED_IN_##__STATE##_READ] = "IN-"__stringify(__STATE)"-R",\
447 [LOCK_ENABLED_##__STATE##_READ] = __stringify(__STATE)"-ON-R",
448
446static const char *usage_str[] = 449static const char *usage_str[] =
447{ 450{
448 [LOCK_USED] = "initial-use ", 451#define LOCKDEP_STATE(__STATE) __USAGE(__STATE)
449 [LOCK_USED_IN_HARDIRQ] = "in-hardirq-W", 452#include "lockdep_states.h"
450 [LOCK_USED_IN_SOFTIRQ] = "in-softirq-W", 453#undef LOCKDEP_STATE
451 [LOCK_ENABLED_SOFTIRQS] = "softirq-on-W", 454 [LOCK_USED] = "INITIAL USE",
452 [LOCK_ENABLED_HARDIRQS] = "hardirq-on-W",
453 [LOCK_USED_IN_HARDIRQ_READ] = "in-hardirq-R",
454 [LOCK_USED_IN_SOFTIRQ_READ] = "in-softirq-R",
455 [LOCK_ENABLED_SOFTIRQS_READ] = "softirq-on-R",
456 [LOCK_ENABLED_HARDIRQS_READ] = "hardirq-on-R",
457}; 455};
458 456
459const char * __get_key_name(struct lockdep_subclass_key *key, char *str) 457const char * __get_key_name(struct lockdep_subclass_key *key, char *str)
@@ -461,46 +459,45 @@ const char * __get_key_name(struct lockdep_subclass_key *key, char *str)
461 return kallsyms_lookup((unsigned long)key, NULL, NULL, NULL, str); 459 return kallsyms_lookup((unsigned long)key, NULL, NULL, NULL, str);
462} 460}
463 461
464void 462static inline unsigned long lock_flag(enum lock_usage_bit bit)
465get_usage_chars(struct lock_class *class, char *c1, char *c2, char *c3, char *c4)
466{ 463{
467 *c1 = '.', *c2 = '.', *c3 = '.', *c4 = '.'; 464 return 1UL << bit;
468 465}
469 if (class->usage_mask & LOCKF_USED_IN_HARDIRQ)
470 *c1 = '+';
471 else
472 if (class->usage_mask & LOCKF_ENABLED_HARDIRQS)
473 *c1 = '-';
474 466
475 if (class->usage_mask & LOCKF_USED_IN_SOFTIRQ) 467static char get_usage_char(struct lock_class *class, enum lock_usage_bit bit)
476 *c2 = '+'; 468{
477 else 469 char c = '.';
478 if (class->usage_mask & LOCKF_ENABLED_SOFTIRQS)
479 *c2 = '-';
480 470
481 if (class->usage_mask & LOCKF_ENABLED_HARDIRQS_READ) 471 if (class->usage_mask & lock_flag(bit + 2))
482 *c3 = '-'; 472 c = '+';
483 if (class->usage_mask & LOCKF_USED_IN_HARDIRQ_READ) { 473 if (class->usage_mask & lock_flag(bit)) {
484 *c3 = '+'; 474 c = '-';
485 if (class->usage_mask & LOCKF_ENABLED_HARDIRQS_READ) 475 if (class->usage_mask & lock_flag(bit + 2))
486 *c3 = '?'; 476 c = '?';
487 } 477 }
488 478
489 if (class->usage_mask & LOCKF_ENABLED_SOFTIRQS_READ) 479 return c;
490 *c4 = '-'; 480}
491 if (class->usage_mask & LOCKF_USED_IN_SOFTIRQ_READ) { 481
492 *c4 = '+'; 482void get_usage_chars(struct lock_class *class, char usage[LOCK_USAGE_CHARS])
493 if (class->usage_mask & LOCKF_ENABLED_SOFTIRQS_READ) 483{
494 *c4 = '?'; 484 int i = 0;
495 } 485
486#define LOCKDEP_STATE(__STATE) \
487 usage[i++] = get_usage_char(class, LOCK_USED_IN_##__STATE); \
488 usage[i++] = get_usage_char(class, LOCK_USED_IN_##__STATE##_READ);
489#include "lockdep_states.h"
490#undef LOCKDEP_STATE
491
492 usage[i] = '\0';
496} 493}
497 494
498static void print_lock_name(struct lock_class *class) 495static void print_lock_name(struct lock_class *class)
499{ 496{
500 char str[KSYM_NAME_LEN], c1, c2, c3, c4; 497 char str[KSYM_NAME_LEN], usage[LOCK_USAGE_CHARS];
501 const char *name; 498 const char *name;
502 499
503 get_usage_chars(class, &c1, &c2, &c3, &c4); 500 get_usage_chars(class, usage);
504 501
505 name = class->name; 502 name = class->name;
506 if (!name) { 503 if (!name) {
@@ -513,7 +510,7 @@ static void print_lock_name(struct lock_class *class)
513 if (class->subclass) 510 if (class->subclass)
514 printk("/%d", class->subclass); 511 printk("/%d", class->subclass);
515 } 512 }
516 printk("){%c%c%c%c}", c1, c2, c3, c4); 513 printk("){%s}", usage);
517} 514}
518 515
519static void print_lockdep_cache(struct lockdep_map *lock) 516static void print_lockdep_cache(struct lockdep_map *lock)
@@ -1263,9 +1260,49 @@ check_usage(struct task_struct *curr, struct held_lock *prev,
1263 bit_backwards, bit_forwards, irqclass); 1260 bit_backwards, bit_forwards, irqclass);
1264} 1261}
1265 1262
1266static int 1263static const char *state_names[] = {
1267check_prev_add_irq(struct task_struct *curr, struct held_lock *prev, 1264#define LOCKDEP_STATE(__STATE) \
1268 struct held_lock *next) 1265 __stringify(__STATE),
1266#include "lockdep_states.h"
1267#undef LOCKDEP_STATE
1268};
1269
1270static const char *state_rnames[] = {
1271#define LOCKDEP_STATE(__STATE) \
1272 __stringify(__STATE)"-READ",
1273#include "lockdep_states.h"
1274#undef LOCKDEP_STATE
1275};
1276
1277static inline const char *state_name(enum lock_usage_bit bit)
1278{
1279 return (bit & 1) ? state_rnames[bit >> 2] : state_names[bit >> 2];
1280}
1281
1282static int exclusive_bit(int new_bit)
1283{
1284 /*
1285 * USED_IN
1286 * USED_IN_READ
1287 * ENABLED
1288 * ENABLED_READ
1289 *
1290 * bit 0 - write/read
1291 * bit 1 - used_in/enabled
1292 * bit 2+ state
1293 */
1294
1295 int state = new_bit & ~3;
1296 int dir = new_bit & 2;
1297
1298 /*
1299 * keep state, bit flip the direction and strip read.
1300 */
1301 return state | (dir ^ 2);
1302}
1303
1304static int check_irq_usage(struct task_struct *curr, struct held_lock *prev,
1305 struct held_lock *next, enum lock_usage_bit bit)
1269{ 1306{
1270 /* 1307 /*
1271 * Prove that the new dependency does not connect a hardirq-safe 1308 * Prove that the new dependency does not connect a hardirq-safe
@@ -1273,38 +1310,34 @@ check_prev_add_irq(struct task_struct *curr, struct held_lock *prev,
1273 * the backwards-subgraph starting at <prev>, and the 1310 * the backwards-subgraph starting at <prev>, and the
1274 * forwards-subgraph starting at <next>: 1311 * forwards-subgraph starting at <next>:
1275 */ 1312 */
1276 if (!check_usage(curr, prev, next, LOCK_USED_IN_HARDIRQ, 1313 if (!check_usage(curr, prev, next, bit,
1277 LOCK_ENABLED_HARDIRQS, "hard")) 1314 exclusive_bit(bit), state_name(bit)))
1278 return 0; 1315 return 0;
1279 1316
1317 bit++; /* _READ */
1318
1280 /* 1319 /*
1281 * Prove that the new dependency does not connect a hardirq-safe-read 1320 * Prove that the new dependency does not connect a hardirq-safe-read
1282 * lock with a hardirq-unsafe lock - to achieve this we search 1321 * lock with a hardirq-unsafe lock - to achieve this we search
1283 * the backwards-subgraph starting at <prev>, and the 1322 * the backwards-subgraph starting at <prev>, and the
1284 * forwards-subgraph starting at <next>: 1323 * forwards-subgraph starting at <next>:
1285 */ 1324 */
1286 if (!check_usage(curr, prev, next, LOCK_USED_IN_HARDIRQ_READ, 1325 if (!check_usage(curr, prev, next, bit,
1287 LOCK_ENABLED_HARDIRQS, "hard-read")) 1326 exclusive_bit(bit), state_name(bit)))
1288 return 0; 1327 return 0;
1289 1328
1290 /* 1329 return 1;
1291 * Prove that the new dependency does not connect a softirq-safe 1330}
1292 * lock with a softirq-unsafe lock - to achieve this we search 1331
1293 * the backwards-subgraph starting at <prev>, and the 1332static int
1294 * forwards-subgraph starting at <next>: 1333check_prev_add_irq(struct task_struct *curr, struct held_lock *prev,
1295 */ 1334 struct held_lock *next)
1296 if (!check_usage(curr, prev, next, LOCK_USED_IN_SOFTIRQ, 1335{
1297 LOCK_ENABLED_SOFTIRQS, "soft")) 1336#define LOCKDEP_STATE(__STATE) \
1298 return 0; 1337 if (!check_irq_usage(curr, prev, next, LOCK_USED_IN_##__STATE)) \
1299 /*
1300 * Prove that the new dependency does not connect a softirq-safe-read
1301 * lock with a softirq-unsafe lock - to achieve this we search
1302 * the backwards-subgraph starting at <prev>, and the
1303 * forwards-subgraph starting at <next>:
1304 */
1305 if (!check_usage(curr, prev, next, LOCK_USED_IN_SOFTIRQ_READ,
1306 LOCK_ENABLED_SOFTIRQS, "soft"))
1307 return 0; 1338 return 0;
1339#include "lockdep_states.h"
1340#undef LOCKDEP_STATE
1308 1341
1309 return 1; 1342 return 1;
1310} 1343}
@@ -1861,9 +1894,9 @@ print_irq_inversion_bug(struct task_struct *curr, struct lock_class *other,
1861 curr->comm, task_pid_nr(curr)); 1894 curr->comm, task_pid_nr(curr));
1862 print_lock(this); 1895 print_lock(this);
1863 if (forwards) 1896 if (forwards)
1864 printk("but this lock took another, %s-irq-unsafe lock in the past:\n", irqclass); 1897 printk("but this lock took another, %s-unsafe lock in the past:\n", irqclass);
1865 else 1898 else
1866 printk("but this lock was taken by another, %s-irq-safe lock in the past:\n", irqclass); 1899 printk("but this lock was taken by another, %s-safe lock in the past:\n", irqclass);
1867 print_lock_name(other); 1900 print_lock_name(other);
1868 printk("\n\nand interrupts could create inverse lock ordering between them.\n\n"); 1901 printk("\n\nand interrupts could create inverse lock ordering between them.\n\n");
1869 1902
@@ -1933,7 +1966,7 @@ void print_irqtrace_events(struct task_struct *curr)
1933 print_ip_sym(curr->softirq_disable_ip); 1966 print_ip_sym(curr->softirq_disable_ip);
1934} 1967}
1935 1968
1936static int hardirq_verbose(struct lock_class *class) 1969static int HARDIRQ_verbose(struct lock_class *class)
1937{ 1970{
1938#if HARDIRQ_VERBOSE 1971#if HARDIRQ_VERBOSE
1939 return class_filter(class); 1972 return class_filter(class);
@@ -1941,7 +1974,7 @@ static int hardirq_verbose(struct lock_class *class)
1941 return 0; 1974 return 0;
1942} 1975}
1943 1976
1944static int softirq_verbose(struct lock_class *class) 1977static int SOFTIRQ_verbose(struct lock_class *class)
1945{ 1978{
1946#if SOFTIRQ_VERBOSE 1979#if SOFTIRQ_VERBOSE
1947 return class_filter(class); 1980 return class_filter(class);
@@ -1949,185 +1982,95 @@ static int softirq_verbose(struct lock_class *class)
1949 return 0; 1982 return 0;
1950} 1983}
1951 1984
1985static int RECLAIM_FS_verbose(struct lock_class *class)
1986{
1987#if RECLAIM_VERBOSE
1988 return class_filter(class);
1989#endif
1990 return 0;
1991}
1992
1952#define STRICT_READ_CHECKS 1 1993#define STRICT_READ_CHECKS 1
1953 1994
1954static int mark_lock_irq(struct task_struct *curr, struct held_lock *this, 1995static int (*state_verbose_f[])(struct lock_class *class) = {
1996#define LOCKDEP_STATE(__STATE) \
1997 __STATE##_verbose,
1998#include "lockdep_states.h"
1999#undef LOCKDEP_STATE
2000};
2001
2002static inline int state_verbose(enum lock_usage_bit bit,
2003 struct lock_class *class)
2004{
2005 return state_verbose_f[bit >> 2](class);
2006}
2007
2008typedef int (*check_usage_f)(struct task_struct *, struct held_lock *,
2009 enum lock_usage_bit bit, const char *name);
2010
2011static int
2012mark_lock_irq(struct task_struct *curr, struct held_lock *this,
1955 enum lock_usage_bit new_bit) 2013 enum lock_usage_bit new_bit)
1956{ 2014{
1957 int ret = 1; 2015 int excl_bit = exclusive_bit(new_bit);
2016 int read = new_bit & 1;
2017 int dir = new_bit & 2;
1958 2018
1959 switch(new_bit) { 2019 /*
1960 case LOCK_USED_IN_HARDIRQ: 2020 * mark USED_IN has to look forwards -- to ensure no dependency
1961 if (!valid_state(curr, this, new_bit, LOCK_ENABLED_HARDIRQS)) 2021 * has ENABLED state, which would allow recursion deadlocks.
1962 return 0; 2022 *
1963 if (!valid_state(curr, this, new_bit, 2023 * mark ENABLED has to look backwards -- to ensure no dependee
1964 LOCK_ENABLED_HARDIRQS_READ)) 2024 * has USED_IN state, which, again, would allow recursion deadlocks.
1965 return 0; 2025 */
1966 /* 2026 check_usage_f usage = dir ?
1967 * just marked it hardirq-safe, check that this lock 2027 check_usage_backwards : check_usage_forwards;
1968 * took no hardirq-unsafe lock in the past: 2028
1969 */ 2029 /*
1970 if (!check_usage_forwards(curr, this, 2030 * Validate that this particular lock does not have conflicting
1971 LOCK_ENABLED_HARDIRQS, "hard")) 2031 * usage states.
1972 return 0; 2032 */
1973#if STRICT_READ_CHECKS 2033 if (!valid_state(curr, this, new_bit, excl_bit))
1974 /* 2034 return 0;
1975 * just marked it hardirq-safe, check that this lock 2035
1976 * took no hardirq-unsafe-read lock in the past: 2036 /*
1977 */ 2037 * Validate that the lock dependencies don't have conflicting usage
1978 if (!check_usage_forwards(curr, this, 2038 * states.
1979 LOCK_ENABLED_HARDIRQS_READ, "hard-read")) 2039 */
1980 return 0; 2040 if ((!read || !dir || STRICT_READ_CHECKS) &&
1981#endif 2041 !usage(curr, this, excl_bit, state_name(new_bit & ~1)))
1982 if (hardirq_verbose(hlock_class(this))) 2042 return 0;
1983 ret = 2; 2043
1984 break; 2044 /*
1985 case LOCK_USED_IN_SOFTIRQ: 2045 * Check for read in write conflicts
1986 if (!valid_state(curr, this, new_bit, LOCK_ENABLED_SOFTIRQS)) 2046 */
1987 return 0; 2047 if (!read) {
1988 if (!valid_state(curr, this, new_bit, 2048 if (!valid_state(curr, this, new_bit, excl_bit + 1))
1989 LOCK_ENABLED_SOFTIRQS_READ))
1990 return 0;
1991 /*
1992 * just marked it softirq-safe, check that this lock
1993 * took no softirq-unsafe lock in the past:
1994 */
1995 if (!check_usage_forwards(curr, this,
1996 LOCK_ENABLED_SOFTIRQS, "soft"))
1997 return 0;
1998#if STRICT_READ_CHECKS
1999 /*
2000 * just marked it softirq-safe, check that this lock
2001 * took no softirq-unsafe-read lock in the past:
2002 */
2003 if (!check_usage_forwards(curr, this,
2004 LOCK_ENABLED_SOFTIRQS_READ, "soft-read"))
2005 return 0;
2006#endif
2007 if (softirq_verbose(hlock_class(this)))
2008 ret = 2;
2009 break;
2010 case LOCK_USED_IN_HARDIRQ_READ:
2011 if (!valid_state(curr, this, new_bit, LOCK_ENABLED_HARDIRQS))
2012 return 0;
2013 /*
2014 * just marked it hardirq-read-safe, check that this lock
2015 * took no hardirq-unsafe lock in the past:
2016 */
2017 if (!check_usage_forwards(curr, this,
2018 LOCK_ENABLED_HARDIRQS, "hard"))
2019 return 0;
2020 if (hardirq_verbose(hlock_class(this)))
2021 ret = 2;
2022 break;
2023 case LOCK_USED_IN_SOFTIRQ_READ:
2024 if (!valid_state(curr, this, new_bit, LOCK_ENABLED_SOFTIRQS))
2025 return 0;
2026 /*
2027 * just marked it softirq-read-safe, check that this lock
2028 * took no softirq-unsafe lock in the past:
2029 */
2030 if (!check_usage_forwards(curr, this,
2031 LOCK_ENABLED_SOFTIRQS, "soft"))
2032 return 0;
2033 if (softirq_verbose(hlock_class(this)))
2034 ret = 2;
2035 break;
2036 case LOCK_ENABLED_HARDIRQS:
2037 if (!valid_state(curr, this, new_bit, LOCK_USED_IN_HARDIRQ))
2038 return 0;
2039 if (!valid_state(curr, this, new_bit,
2040 LOCK_USED_IN_HARDIRQ_READ))
2041 return 0;
2042 /*
2043 * just marked it hardirq-unsafe, check that no hardirq-safe
2044 * lock in the system ever took it in the past:
2045 */
2046 if (!check_usage_backwards(curr, this,
2047 LOCK_USED_IN_HARDIRQ, "hard"))
2048 return 0;
2049#if STRICT_READ_CHECKS
2050 /*
2051 * just marked it hardirq-unsafe, check that no
2052 * hardirq-safe-read lock in the system ever took
2053 * it in the past:
2054 */
2055 if (!check_usage_backwards(curr, this,
2056 LOCK_USED_IN_HARDIRQ_READ, "hard-read"))
2057 return 0;
2058#endif
2059 if (hardirq_verbose(hlock_class(this)))
2060 ret = 2;
2061 break;
2062 case LOCK_ENABLED_SOFTIRQS:
2063 if (!valid_state(curr, this, new_bit, LOCK_USED_IN_SOFTIRQ))
2064 return 0;
2065 if (!valid_state(curr, this, new_bit,
2066 LOCK_USED_IN_SOFTIRQ_READ))
2067 return 0;
2068 /*
2069 * just marked it softirq-unsafe, check that no softirq-safe
2070 * lock in the system ever took it in the past:
2071 */
2072 if (!check_usage_backwards(curr, this,
2073 LOCK_USED_IN_SOFTIRQ, "soft"))
2074 return 0;
2075#if STRICT_READ_CHECKS
2076 /*
2077 * just marked it softirq-unsafe, check that no
2078 * softirq-safe-read lock in the system ever took
2079 * it in the past:
2080 */
2081 if (!check_usage_backwards(curr, this,
2082 LOCK_USED_IN_SOFTIRQ_READ, "soft-read"))
2083 return 0;
2084#endif
2085 if (softirq_verbose(hlock_class(this)))
2086 ret = 2;
2087 break;
2088 case LOCK_ENABLED_HARDIRQS_READ:
2089 if (!valid_state(curr, this, new_bit, LOCK_USED_IN_HARDIRQ))
2090 return 0;
2091#if STRICT_READ_CHECKS
2092 /*
2093 * just marked it hardirq-read-unsafe, check that no
2094 * hardirq-safe lock in the system ever took it in the past:
2095 */
2096 if (!check_usage_backwards(curr, this,
2097 LOCK_USED_IN_HARDIRQ, "hard"))
2098 return 0;
2099#endif
2100 if (hardirq_verbose(hlock_class(this)))
2101 ret = 2;
2102 break;
2103 case LOCK_ENABLED_SOFTIRQS_READ:
2104 if (!valid_state(curr, this, new_bit, LOCK_USED_IN_SOFTIRQ))
2105 return 0; 2049 return 0;
2106#if STRICT_READ_CHECKS 2050
2107 /* 2051 if (STRICT_READ_CHECKS &&
2108 * just marked it softirq-read-unsafe, check that no 2052 !usage(curr, this, excl_bit + 1,
2109 * softirq-safe lock in the system ever took it in the past: 2053 state_name(new_bit + 1)))
2110 */
2111 if (!check_usage_backwards(curr, this,
2112 LOCK_USED_IN_SOFTIRQ, "soft"))
2113 return 0; 2054 return 0;
2114#endif
2115 if (softirq_verbose(hlock_class(this)))
2116 ret = 2;
2117 break;
2118 default:
2119 WARN_ON(1);
2120 break;
2121 } 2055 }
2122 2056
2123 return ret; 2057 if (state_verbose(new_bit, hlock_class(this)))
2058 return 2;
2059
2060 return 1;
2124} 2061}
2125 2062
2063enum mark_type {
2064#define LOCKDEP_STATE(__STATE) __STATE,
2065#include "lockdep_states.h"
2066#undef LOCKDEP_STATE
2067};
2068
2126/* 2069/*
2127 * Mark all held locks with a usage bit: 2070 * Mark all held locks with a usage bit:
2128 */ 2071 */
2129static int 2072static int
2130mark_held_locks(struct task_struct *curr, int hardirq) 2073mark_held_locks(struct task_struct *curr, enum mark_type mark)
2131{ 2074{
2132 enum lock_usage_bit usage_bit; 2075 enum lock_usage_bit usage_bit;
2133 struct held_lock *hlock; 2076 struct held_lock *hlock;
@@ -2136,17 +2079,12 @@ mark_held_locks(struct task_struct *curr, int hardirq)
2136 for (i = 0; i < curr->lockdep_depth; i++) { 2079 for (i = 0; i < curr->lockdep_depth; i++) {
2137 hlock = curr->held_locks + i; 2080 hlock = curr->held_locks + i;
2138 2081
2139 if (hardirq) { 2082 usage_bit = 2 + (mark << 2); /* ENABLED */
2140 if (hlock->read) 2083 if (hlock->read)
2141 usage_bit = LOCK_ENABLED_HARDIRQS_READ; 2084 usage_bit += 1; /* READ */
2142 else 2085
2143 usage_bit = LOCK_ENABLED_HARDIRQS; 2086 BUG_ON(usage_bit >= LOCK_USAGE_STATES);
2144 } else { 2087
2145 if (hlock->read)
2146 usage_bit = LOCK_ENABLED_SOFTIRQS_READ;
2147 else
2148 usage_bit = LOCK_ENABLED_SOFTIRQS;
2149 }
2150 if (!mark_lock(curr, hlock, usage_bit)) 2088 if (!mark_lock(curr, hlock, usage_bit))
2151 return 0; 2089 return 0;
2152 } 2090 }
@@ -2200,7 +2138,7 @@ void trace_hardirqs_on_caller(unsigned long ip)
2200 * We are going to turn hardirqs on, so set the 2138 * We are going to turn hardirqs on, so set the
2201 * usage bit for all held locks: 2139 * usage bit for all held locks:
2202 */ 2140 */
2203 if (!mark_held_locks(curr, 1)) 2141 if (!mark_held_locks(curr, HARDIRQ))
2204 return; 2142 return;
2205 /* 2143 /*
2206 * If we have softirqs enabled, then set the usage 2144 * If we have softirqs enabled, then set the usage
@@ -2208,7 +2146,7 @@ void trace_hardirqs_on_caller(unsigned long ip)
2208 * this bit from being set before) 2146 * this bit from being set before)
2209 */ 2147 */
2210 if (curr->softirqs_enabled) 2148 if (curr->softirqs_enabled)
2211 if (!mark_held_locks(curr, 0)) 2149 if (!mark_held_locks(curr, SOFTIRQ))
2212 return; 2150 return;
2213 2151
2214 curr->hardirq_enable_ip = ip; 2152 curr->hardirq_enable_ip = ip;
@@ -2288,7 +2226,7 @@ void trace_softirqs_on(unsigned long ip)
2288 * enabled too: 2226 * enabled too:
2289 */ 2227 */
2290 if (curr->hardirqs_enabled) 2228 if (curr->hardirqs_enabled)
2291 mark_held_locks(curr, 0); 2229 mark_held_locks(curr, SOFTIRQ);
2292} 2230}
2293 2231
2294/* 2232/*
@@ -2317,6 +2255,48 @@ void trace_softirqs_off(unsigned long ip)
2317 debug_atomic_inc(&redundant_softirqs_off); 2255 debug_atomic_inc(&redundant_softirqs_off);
2318} 2256}
2319 2257
2258static void __lockdep_trace_alloc(gfp_t gfp_mask, unsigned long flags)
2259{
2260 struct task_struct *curr = current;
2261
2262 if (unlikely(!debug_locks))
2263 return;
2264
2265 /* no reclaim without waiting on it */
2266 if (!(gfp_mask & __GFP_WAIT))
2267 return;
2268
2269 /* this guy won't enter reclaim */
2270 if ((curr->flags & PF_MEMALLOC) && !(gfp_mask & __GFP_NOMEMALLOC))
2271 return;
2272
2273 /* We're only interested __GFP_FS allocations for now */
2274 if (!(gfp_mask & __GFP_FS))
2275 return;
2276
2277 if (DEBUG_LOCKS_WARN_ON(irqs_disabled_flags(flags)))
2278 return;
2279
2280 mark_held_locks(curr, RECLAIM_FS);
2281}
2282
2283static void check_flags(unsigned long flags);
2284
2285void lockdep_trace_alloc(gfp_t gfp_mask)
2286{
2287 unsigned long flags;
2288
2289 if (unlikely(current->lockdep_recursion))
2290 return;
2291
2292 raw_local_irq_save(flags);
2293 check_flags(flags);
2294 current->lockdep_recursion = 1;
2295 __lockdep_trace_alloc(gfp_mask, flags);
2296 current->lockdep_recursion = 0;
2297 raw_local_irq_restore(flags);
2298}
2299
2320static int mark_irqflags(struct task_struct *curr, struct held_lock *hlock) 2300static int mark_irqflags(struct task_struct *curr, struct held_lock *hlock)
2321{ 2301{
2322 /* 2302 /*
@@ -2345,19 +2325,35 @@ static int mark_irqflags(struct task_struct *curr, struct held_lock *hlock)
2345 if (!hlock->hardirqs_off) { 2325 if (!hlock->hardirqs_off) {
2346 if (hlock->read) { 2326 if (hlock->read) {
2347 if (!mark_lock(curr, hlock, 2327 if (!mark_lock(curr, hlock,
2348 LOCK_ENABLED_HARDIRQS_READ)) 2328 LOCK_ENABLED_HARDIRQ_READ))
2349 return 0; 2329 return 0;
2350 if (curr->softirqs_enabled) 2330 if (curr->softirqs_enabled)
2351 if (!mark_lock(curr, hlock, 2331 if (!mark_lock(curr, hlock,
2352 LOCK_ENABLED_SOFTIRQS_READ)) 2332 LOCK_ENABLED_SOFTIRQ_READ))
2353 return 0; 2333 return 0;
2354 } else { 2334 } else {
2355 if (!mark_lock(curr, hlock, 2335 if (!mark_lock(curr, hlock,
2356 LOCK_ENABLED_HARDIRQS)) 2336 LOCK_ENABLED_HARDIRQ))
2357 return 0; 2337 return 0;
2358 if (curr->softirqs_enabled) 2338 if (curr->softirqs_enabled)
2359 if (!mark_lock(curr, hlock, 2339 if (!mark_lock(curr, hlock,
2360 LOCK_ENABLED_SOFTIRQS)) 2340 LOCK_ENABLED_SOFTIRQ))
2341 return 0;
2342 }
2343 }
2344
2345 /*
2346 * We reuse the irq context infrastructure more broadly as a general
2347 * context checking code. This tests GFP_FS recursion (a lock taken
2348 * during reclaim for a GFP_FS allocation is held over a GFP_FS
2349 * allocation).
2350 */
2351 if (!hlock->trylock && (curr->lockdep_reclaim_gfp & __GFP_FS)) {
2352 if (hlock->read) {
2353 if (!mark_lock(curr, hlock, LOCK_USED_IN_RECLAIM_FS_READ))
2354 return 0;
2355 } else {
2356 if (!mark_lock(curr, hlock, LOCK_USED_IN_RECLAIM_FS))
2361 return 0; 2357 return 0;
2362 } 2358 }
2363 } 2359 }
@@ -2412,6 +2408,10 @@ static inline int separate_irq_context(struct task_struct *curr,
2412 return 0; 2408 return 0;
2413} 2409}
2414 2410
2411void lockdep_trace_alloc(gfp_t gfp_mask)
2412{
2413}
2414
2415#endif 2415#endif
2416 2416
2417/* 2417/*
@@ -2445,14 +2445,13 @@ static int mark_lock(struct task_struct *curr, struct held_lock *this,
2445 return 0; 2445 return 0;
2446 2446
2447 switch (new_bit) { 2447 switch (new_bit) {
2448 case LOCK_USED_IN_HARDIRQ: 2448#define LOCKDEP_STATE(__STATE) \
2449 case LOCK_USED_IN_SOFTIRQ: 2449 case LOCK_USED_IN_##__STATE: \
2450 case LOCK_USED_IN_HARDIRQ_READ: 2450 case LOCK_USED_IN_##__STATE##_READ: \
2451 case LOCK_USED_IN_SOFTIRQ_READ: 2451 case LOCK_ENABLED_##__STATE: \
2452 case LOCK_ENABLED_HARDIRQS: 2452 case LOCK_ENABLED_##__STATE##_READ:
2453 case LOCK_ENABLED_SOFTIRQS: 2453#include "lockdep_states.h"
2454 case LOCK_ENABLED_HARDIRQS_READ: 2454#undef LOCKDEP_STATE
2455 case LOCK_ENABLED_SOFTIRQS_READ:
2456 ret = mark_lock_irq(curr, this, new_bit); 2455 ret = mark_lock_irq(curr, this, new_bit);
2457 if (!ret) 2456 if (!ret)
2458 return 0; 2457 return 0;
@@ -2925,6 +2924,8 @@ void lock_set_class(struct lockdep_map *lock, const char *name,
2925} 2924}
2926EXPORT_SYMBOL_GPL(lock_set_class); 2925EXPORT_SYMBOL_GPL(lock_set_class);
2927 2926
2927DEFINE_TRACE(lock_acquire);
2928
2928/* 2929/*
2929 * We are not always called with irqs disabled - do that here, 2930 * We are not always called with irqs disabled - do that here,
2930 * and also avoid lockdep recursion: 2931 * and also avoid lockdep recursion:
@@ -2935,6 +2936,8 @@ void lock_acquire(struct lockdep_map *lock, unsigned int subclass,
2935{ 2936{
2936 unsigned long flags; 2937 unsigned long flags;
2937 2938
2939 trace_lock_acquire(lock, subclass, trylock, read, check, nest_lock, ip);
2940
2938 if (unlikely(current->lockdep_recursion)) 2941 if (unlikely(current->lockdep_recursion))
2939 return; 2942 return;
2940 2943
@@ -2949,11 +2952,15 @@ void lock_acquire(struct lockdep_map *lock, unsigned int subclass,
2949} 2952}
2950EXPORT_SYMBOL_GPL(lock_acquire); 2953EXPORT_SYMBOL_GPL(lock_acquire);
2951 2954
2955DEFINE_TRACE(lock_release);
2956
2952void lock_release(struct lockdep_map *lock, int nested, 2957void lock_release(struct lockdep_map *lock, int nested,
2953 unsigned long ip) 2958 unsigned long ip)
2954{ 2959{
2955 unsigned long flags; 2960 unsigned long flags;
2956 2961
2962 trace_lock_release(lock, nested, ip);
2963
2957 if (unlikely(current->lockdep_recursion)) 2964 if (unlikely(current->lockdep_recursion))
2958 return; 2965 return;
2959 2966
@@ -2966,6 +2973,16 @@ void lock_release(struct lockdep_map *lock, int nested,
2966} 2973}
2967EXPORT_SYMBOL_GPL(lock_release); 2974EXPORT_SYMBOL_GPL(lock_release);
2968 2975
2976void lockdep_set_current_reclaim_state(gfp_t gfp_mask)
2977{
2978 current->lockdep_reclaim_gfp = gfp_mask;
2979}
2980
2981void lockdep_clear_current_reclaim_state(void)
2982{
2983 current->lockdep_reclaim_gfp = 0;
2984}
2985
2969#ifdef CONFIG_LOCK_STAT 2986#ifdef CONFIG_LOCK_STAT
2970static int 2987static int
2971print_lock_contention_bug(struct task_struct *curr, struct lockdep_map *lock, 2988print_lock_contention_bug(struct task_struct *curr, struct lockdep_map *lock,
@@ -3092,10 +3109,14 @@ found_it:
3092 lock->ip = ip; 3109 lock->ip = ip;
3093} 3110}
3094 3111
3112DEFINE_TRACE(lock_contended);
3113
3095void lock_contended(struct lockdep_map *lock, unsigned long ip) 3114void lock_contended(struct lockdep_map *lock, unsigned long ip)
3096{ 3115{
3097 unsigned long flags; 3116 unsigned long flags;
3098 3117
3118 trace_lock_contended(lock, ip);
3119
3099 if (unlikely(!lock_stat)) 3120 if (unlikely(!lock_stat))
3100 return; 3121 return;
3101 3122
@@ -3111,10 +3132,14 @@ void lock_contended(struct lockdep_map *lock, unsigned long ip)
3111} 3132}
3112EXPORT_SYMBOL_GPL(lock_contended); 3133EXPORT_SYMBOL_GPL(lock_contended);
3113 3134
3135DEFINE_TRACE(lock_acquired);
3136
3114void lock_acquired(struct lockdep_map *lock, unsigned long ip) 3137void lock_acquired(struct lockdep_map *lock, unsigned long ip)
3115{ 3138{
3116 unsigned long flags; 3139 unsigned long flags;
3117 3140
3141 trace_lock_acquired(lock, ip);
3142
3118 if (unlikely(!lock_stat)) 3143 if (unlikely(!lock_stat))
3119 return; 3144 return;
3120 3145
diff --git a/kernel/lockdep_internals.h b/kernel/lockdep_internals.h
index 56b196932c08..a2cc7e9a6e84 100644
--- a/kernel/lockdep_internals.h
+++ b/kernel/lockdep_internals.h
@@ -7,6 +7,45 @@
7 */ 7 */
8 8
9/* 9/*
10 * Lock-class usage-state bits:
11 */
12enum lock_usage_bit {
13#define LOCKDEP_STATE(__STATE) \
14 LOCK_USED_IN_##__STATE, \
15 LOCK_USED_IN_##__STATE##_READ, \
16 LOCK_ENABLED_##__STATE, \
17 LOCK_ENABLED_##__STATE##_READ,
18#include "lockdep_states.h"
19#undef LOCKDEP_STATE
20 LOCK_USED,
21 LOCK_USAGE_STATES
22};
23
24/*
25 * Usage-state bitmasks:
26 */
27#define __LOCKF(__STATE) LOCKF_##__STATE = (1 << LOCK_##__STATE),
28
29enum {
30#define LOCKDEP_STATE(__STATE) \
31 __LOCKF(USED_IN_##__STATE) \
32 __LOCKF(USED_IN_##__STATE##_READ) \
33 __LOCKF(ENABLED_##__STATE) \
34 __LOCKF(ENABLED_##__STATE##_READ)
35#include "lockdep_states.h"
36#undef LOCKDEP_STATE
37 __LOCKF(USED)
38};
39
40#define LOCKF_ENABLED_IRQ (LOCKF_ENABLED_HARDIRQ | LOCKF_ENABLED_SOFTIRQ)
41#define LOCKF_USED_IN_IRQ (LOCKF_USED_IN_HARDIRQ | LOCKF_USED_IN_SOFTIRQ)
42
43#define LOCKF_ENABLED_IRQ_READ \
44 (LOCKF_ENABLED_HARDIRQ_READ | LOCKF_ENABLED_SOFTIRQ_READ)
45#define LOCKF_USED_IN_IRQ_READ \
46 (LOCKF_USED_IN_HARDIRQ_READ | LOCKF_USED_IN_SOFTIRQ_READ)
47
48/*
10 * MAX_LOCKDEP_ENTRIES is the maximum number of lock dependencies 49 * MAX_LOCKDEP_ENTRIES is the maximum number of lock dependencies
11 * we track. 50 * we track.
12 * 51 *
@@ -31,8 +70,10 @@
31extern struct list_head all_lock_classes; 70extern struct list_head all_lock_classes;
32extern struct lock_chain lock_chains[]; 71extern struct lock_chain lock_chains[];
33 72
34extern void 73#define LOCK_USAGE_CHARS (1+LOCK_USAGE_STATES/2)
35get_usage_chars(struct lock_class *class, char *c1, char *c2, char *c3, char *c4); 74
75extern void get_usage_chars(struct lock_class *class,
76 char usage[LOCK_USAGE_CHARS]);
36 77
37extern const char * __get_key_name(struct lockdep_subclass_key *key, char *str); 78extern const char * __get_key_name(struct lockdep_subclass_key *key, char *str);
38 79
diff --git a/kernel/lockdep_proc.c b/kernel/lockdep_proc.c
index 13716b813896..d7135aa2d2c4 100644
--- a/kernel/lockdep_proc.c
+++ b/kernel/lockdep_proc.c
@@ -84,7 +84,7 @@ static int l_show(struct seq_file *m, void *v)
84{ 84{
85 struct lock_class *class = v; 85 struct lock_class *class = v;
86 struct lock_list *entry; 86 struct lock_list *entry;
87 char c1, c2, c3, c4; 87 char usage[LOCK_USAGE_CHARS];
88 88
89 if (v == SEQ_START_TOKEN) { 89 if (v == SEQ_START_TOKEN) {
90 seq_printf(m, "all lock classes:\n"); 90 seq_printf(m, "all lock classes:\n");
@@ -100,8 +100,8 @@ static int l_show(struct seq_file *m, void *v)
100 seq_printf(m, " BD:%5ld", lockdep_count_backward_deps(class)); 100 seq_printf(m, " BD:%5ld", lockdep_count_backward_deps(class));
101#endif 101#endif
102 102
103 get_usage_chars(class, &c1, &c2, &c3, &c4); 103 get_usage_chars(class, usage);
104 seq_printf(m, " %c%c%c%c", c1, c2, c3, c4); 104 seq_printf(m, " %s", usage);
105 105
106 seq_printf(m, ": "); 106 seq_printf(m, ": ");
107 print_name(m, class); 107 print_name(m, class);
@@ -300,27 +300,27 @@ static int lockdep_stats_show(struct seq_file *m, void *v)
300 nr_uncategorized++; 300 nr_uncategorized++;
301 if (class->usage_mask & LOCKF_USED_IN_IRQ) 301 if (class->usage_mask & LOCKF_USED_IN_IRQ)
302 nr_irq_safe++; 302 nr_irq_safe++;
303 if (class->usage_mask & LOCKF_ENABLED_IRQS) 303 if (class->usage_mask & LOCKF_ENABLED_IRQ)
304 nr_irq_unsafe++; 304 nr_irq_unsafe++;
305 if (class->usage_mask & LOCKF_USED_IN_SOFTIRQ) 305 if (class->usage_mask & LOCKF_USED_IN_SOFTIRQ)
306 nr_softirq_safe++; 306 nr_softirq_safe++;
307 if (class->usage_mask & LOCKF_ENABLED_SOFTIRQS) 307 if (class->usage_mask & LOCKF_ENABLED_SOFTIRQ)
308 nr_softirq_unsafe++; 308 nr_softirq_unsafe++;
309 if (class->usage_mask & LOCKF_USED_IN_HARDIRQ) 309 if (class->usage_mask & LOCKF_USED_IN_HARDIRQ)
310 nr_hardirq_safe++; 310 nr_hardirq_safe++;
311 if (class->usage_mask & LOCKF_ENABLED_HARDIRQS) 311 if (class->usage_mask & LOCKF_ENABLED_HARDIRQ)
312 nr_hardirq_unsafe++; 312 nr_hardirq_unsafe++;
313 if (class->usage_mask & LOCKF_USED_IN_IRQ_READ) 313 if (class->usage_mask & LOCKF_USED_IN_IRQ_READ)
314 nr_irq_read_safe++; 314 nr_irq_read_safe++;
315 if (class->usage_mask & LOCKF_ENABLED_IRQS_READ) 315 if (class->usage_mask & LOCKF_ENABLED_IRQ_READ)
316 nr_irq_read_unsafe++; 316 nr_irq_read_unsafe++;
317 if (class->usage_mask & LOCKF_USED_IN_SOFTIRQ_READ) 317 if (class->usage_mask & LOCKF_USED_IN_SOFTIRQ_READ)
318 nr_softirq_read_safe++; 318 nr_softirq_read_safe++;
319 if (class->usage_mask & LOCKF_ENABLED_SOFTIRQS_READ) 319 if (class->usage_mask & LOCKF_ENABLED_SOFTIRQ_READ)
320 nr_softirq_read_unsafe++; 320 nr_softirq_read_unsafe++;
321 if (class->usage_mask & LOCKF_USED_IN_HARDIRQ_READ) 321 if (class->usage_mask & LOCKF_USED_IN_HARDIRQ_READ)
322 nr_hardirq_read_safe++; 322 nr_hardirq_read_safe++;
323 if (class->usage_mask & LOCKF_ENABLED_HARDIRQS_READ) 323 if (class->usage_mask & LOCKF_ENABLED_HARDIRQ_READ)
324 nr_hardirq_read_unsafe++; 324 nr_hardirq_read_unsafe++;
325 325
326#ifdef CONFIG_PROVE_LOCKING 326#ifdef CONFIG_PROVE_LOCKING
@@ -601,6 +601,10 @@ static void seq_stats(struct seq_file *m, struct lock_stat_data *data)
601static void seq_header(struct seq_file *m) 601static void seq_header(struct seq_file *m)
602{ 602{
603 seq_printf(m, "lock_stat version 0.3\n"); 603 seq_printf(m, "lock_stat version 0.3\n");
604
605 if (unlikely(!debug_locks))
606 seq_printf(m, "*WARNING* lock debugging disabled!! - possibly due to a lockdep warning\n");
607
604 seq_line(m, '-', 0, 40 + 1 + 10 * (14 + 1)); 608 seq_line(m, '-', 0, 40 + 1 + 10 * (14 + 1));
605 seq_printf(m, "%40s %14s %14s %14s %14s %14s %14s %14s %14s " 609 seq_printf(m, "%40s %14s %14s %14s %14s %14s %14s %14s %14s "
606 "%14s %14s\n", 610 "%14s %14s\n",
diff --git a/kernel/lockdep_states.h b/kernel/lockdep_states.h
new file mode 100644
index 000000000000..995b0cc2b84c
--- /dev/null
+++ b/kernel/lockdep_states.h
@@ -0,0 +1,9 @@
1/*
2 * Lockdep states,
3 *
4 * please update XXX_LOCK_USAGE_STATES in include/linux/lockdep.h whenever
5 * you add one, or come up with a nice dynamic solution.
6 */
7LOCKDEP_STATE(HARDIRQ)
8LOCKDEP_STATE(SOFTIRQ)
9LOCKDEP_STATE(RECLAIM_FS)
diff --git a/kernel/module.c b/kernel/module.c
index ba22484a987e..c268a771595c 100644
--- a/kernel/module.c
+++ b/kernel/module.c
@@ -51,6 +51,7 @@
51#include <linux/tracepoint.h> 51#include <linux/tracepoint.h>
52#include <linux/ftrace.h> 52#include <linux/ftrace.h>
53#include <linux/async.h> 53#include <linux/async.h>
54#include <linux/percpu.h>
54 55
55#if 0 56#if 0
56#define DEBUGP printk 57#define DEBUGP printk
@@ -67,7 +68,8 @@
67 68
68/* List of modules, protected by module_mutex or preempt_disable 69/* List of modules, protected by module_mutex or preempt_disable
69 * (delete uses stop_machine/add uses RCU list operations). */ 70 * (delete uses stop_machine/add uses RCU list operations). */
70static DEFINE_MUTEX(module_mutex); 71DEFINE_MUTEX(module_mutex);
72EXPORT_SYMBOL_GPL(module_mutex);
71static LIST_HEAD(modules); 73static LIST_HEAD(modules);
72 74
73/* Waiting for a module to finish initializing? */ 75/* Waiting for a module to finish initializing? */
@@ -75,7 +77,7 @@ static DECLARE_WAIT_QUEUE_HEAD(module_wq);
75 77
76static BLOCKING_NOTIFIER_HEAD(module_notify_list); 78static BLOCKING_NOTIFIER_HEAD(module_notify_list);
77 79
78/* Bounds of module allocation, for speeding __module_text_address */ 80/* Bounds of module allocation, for speeding __module_address */
79static unsigned long module_addr_min = -1UL, module_addr_max = 0; 81static unsigned long module_addr_min = -1UL, module_addr_max = 0;
80 82
81int register_module_notifier(struct notifier_block * nb) 83int register_module_notifier(struct notifier_block * nb)
@@ -185,17 +187,6 @@ extern const unsigned long __start___kcrctab_unused_gpl[];
185#define symversion(base, idx) ((base != NULL) ? ((base) + (idx)) : NULL) 187#define symversion(base, idx) ((base != NULL) ? ((base) + (idx)) : NULL)
186#endif 188#endif
187 189
188struct symsearch {
189 const struct kernel_symbol *start, *stop;
190 const unsigned long *crcs;
191 enum {
192 NOT_GPL_ONLY,
193 GPL_ONLY,
194 WILL_BE_GPL_ONLY,
195 } licence;
196 bool unused;
197};
198
199static bool each_symbol_in_section(const struct symsearch *arr, 190static bool each_symbol_in_section(const struct symsearch *arr,
200 unsigned int arrsize, 191 unsigned int arrsize,
201 struct module *owner, 192 struct module *owner,
@@ -216,10 +207,8 @@ static bool each_symbol_in_section(const struct symsearch *arr,
216} 207}
217 208
218/* Returns true as soon as fn returns true, otherwise false. */ 209/* Returns true as soon as fn returns true, otherwise false. */
219static bool each_symbol(bool (*fn)(const struct symsearch *arr, 210bool each_symbol(bool (*fn)(const struct symsearch *arr, struct module *owner,
220 struct module *owner, 211 unsigned int symnum, void *data), void *data)
221 unsigned int symnum, void *data),
222 void *data)
223{ 212{
224 struct module *mod; 213 struct module *mod;
225 const struct symsearch arr[] = { 214 const struct symsearch arr[] = {
@@ -272,6 +261,7 @@ static bool each_symbol(bool (*fn)(const struct symsearch *arr,
272 } 261 }
273 return false; 262 return false;
274} 263}
264EXPORT_SYMBOL_GPL(each_symbol);
275 265
276struct find_symbol_arg { 266struct find_symbol_arg {
277 /* Input */ 267 /* Input */
@@ -282,7 +272,7 @@ struct find_symbol_arg {
282 /* Output */ 272 /* Output */
283 struct module *owner; 273 struct module *owner;
284 const unsigned long *crc; 274 const unsigned long *crc;
285 unsigned long value; 275 const struct kernel_symbol *sym;
286}; 276};
287 277
288static bool find_symbol_in_section(const struct symsearch *syms, 278static bool find_symbol_in_section(const struct symsearch *syms,
@@ -323,17 +313,17 @@ static bool find_symbol_in_section(const struct symsearch *syms,
323 313
324 fsa->owner = owner; 314 fsa->owner = owner;
325 fsa->crc = symversion(syms->crcs, symnum); 315 fsa->crc = symversion(syms->crcs, symnum);
326 fsa->value = syms->start[symnum].value; 316 fsa->sym = &syms->start[symnum];
327 return true; 317 return true;
328} 318}
329 319
330/* Find a symbol, return value, (optional) crc and (optional) module 320/* Find a symbol and return it, along with, (optional) crc and
331 * which owns it */ 321 * (optional) module which owns it */
332static unsigned long find_symbol(const char *name, 322const struct kernel_symbol *find_symbol(const char *name,
333 struct module **owner, 323 struct module **owner,
334 const unsigned long **crc, 324 const unsigned long **crc,
335 bool gplok, 325 bool gplok,
336 bool warn) 326 bool warn)
337{ 327{
338 struct find_symbol_arg fsa; 328 struct find_symbol_arg fsa;
339 329
@@ -346,15 +336,16 @@ static unsigned long find_symbol(const char *name,
346 *owner = fsa.owner; 336 *owner = fsa.owner;
347 if (crc) 337 if (crc)
348 *crc = fsa.crc; 338 *crc = fsa.crc;
349 return fsa.value; 339 return fsa.sym;
350 } 340 }
351 341
352 DEBUGP("Failed to find symbol %s\n", name); 342 DEBUGP("Failed to find symbol %s\n", name);
353 return -ENOENT; 343 return NULL;
354} 344}
345EXPORT_SYMBOL_GPL(find_symbol);
355 346
356/* Search for module by name: must hold module_mutex. */ 347/* Search for module by name: must hold module_mutex. */
357static struct module *find_module(const char *name) 348struct module *find_module(const char *name)
358{ 349{
359 struct module *mod; 350 struct module *mod;
360 351
@@ -364,8 +355,37 @@ static struct module *find_module(const char *name)
364 } 355 }
365 return NULL; 356 return NULL;
366} 357}
358EXPORT_SYMBOL_GPL(find_module);
367 359
368#ifdef CONFIG_SMP 360#ifdef CONFIG_SMP
361
362#ifdef CONFIG_HAVE_DYNAMIC_PER_CPU_AREA
363
364static void *percpu_modalloc(unsigned long size, unsigned long align,
365 const char *name)
366{
367 void *ptr;
368
369 if (align > PAGE_SIZE) {
370 printk(KERN_WARNING "%s: per-cpu alignment %li > %li\n",
371 name, align, PAGE_SIZE);
372 align = PAGE_SIZE;
373 }
374
375 ptr = __alloc_reserved_percpu(size, align);
376 if (!ptr)
377 printk(KERN_WARNING
378 "Could not allocate %lu bytes percpu data\n", size);
379 return ptr;
380}
381
382static void percpu_modfree(void *freeme)
383{
384 free_percpu(freeme);
385}
386
387#else /* ... !CONFIG_HAVE_DYNAMIC_PER_CPU_AREA */
388
369/* Number of blocks used and allocated. */ 389/* Number of blocks used and allocated. */
370static unsigned int pcpu_num_used, pcpu_num_allocated; 390static unsigned int pcpu_num_used, pcpu_num_allocated;
371/* Size of each block. -ve means used. */ 391/* Size of each block. -ve means used. */
@@ -480,21 +500,6 @@ static void percpu_modfree(void *freeme)
480 } 500 }
481} 501}
482 502
483static unsigned int find_pcpusec(Elf_Ehdr *hdr,
484 Elf_Shdr *sechdrs,
485 const char *secstrings)
486{
487 return find_sec(hdr, sechdrs, secstrings, ".data.percpu");
488}
489
490static void percpu_modcopy(void *pcpudest, const void *from, unsigned long size)
491{
492 int cpu;
493
494 for_each_possible_cpu(cpu)
495 memcpy(pcpudest + per_cpu_offset(cpu), from, size);
496}
497
498static int percpu_modinit(void) 503static int percpu_modinit(void)
499{ 504{
500 pcpu_num_used = 2; 505 pcpu_num_used = 2;
@@ -513,7 +518,26 @@ static int percpu_modinit(void)
513 return 0; 518 return 0;
514} 519}
515__initcall(percpu_modinit); 520__initcall(percpu_modinit);
521
522#endif /* CONFIG_HAVE_DYNAMIC_PER_CPU_AREA */
523
524static unsigned int find_pcpusec(Elf_Ehdr *hdr,
525 Elf_Shdr *sechdrs,
526 const char *secstrings)
527{
528 return find_sec(hdr, sechdrs, secstrings, ".data.percpu");
529}
530
531static void percpu_modcopy(void *pcpudest, const void *from, unsigned long size)
532{
533 int cpu;
534
535 for_each_possible_cpu(cpu)
536 memcpy(pcpudest + per_cpu_offset(cpu), from, size);
537}
538
516#else /* ... !CONFIG_SMP */ 539#else /* ... !CONFIG_SMP */
540
517static inline void *percpu_modalloc(unsigned long size, unsigned long align, 541static inline void *percpu_modalloc(unsigned long size, unsigned long align,
518 const char *name) 542 const char *name)
519{ 543{
@@ -535,6 +559,7 @@ static inline void percpu_modcopy(void *pcpudst, const void *src,
535 /* pcpusec should be 0, and size of that section should be 0. */ 559 /* pcpusec should be 0, and size of that section should be 0. */
536 BUG_ON(size != 0); 560 BUG_ON(size != 0);
537} 561}
562
538#endif /* CONFIG_SMP */ 563#endif /* CONFIG_SMP */
539 564
540#define MODINFO_ATTR(field) \ 565#define MODINFO_ATTR(field) \
@@ -607,7 +632,7 @@ static int already_uses(struct module *a, struct module *b)
607} 632}
608 633
609/* Module a uses b */ 634/* Module a uses b */
610static int use_module(struct module *a, struct module *b) 635int use_module(struct module *a, struct module *b)
611{ 636{
612 struct module_use *use; 637 struct module_use *use;
613 int no_warn, err; 638 int no_warn, err;
@@ -640,6 +665,7 @@ static int use_module(struct module *a, struct module *b)
640 no_warn = sysfs_create_link(b->holders_dir, &a->mkobj.kobj, a->name); 665 no_warn = sysfs_create_link(b->holders_dir, &a->mkobj.kobj, a->name);
641 return 1; 666 return 1;
642} 667}
668EXPORT_SYMBOL_GPL(use_module);
643 669
644/* Clear the unload stuff of the module. */ 670/* Clear the unload stuff of the module. */
645static void module_unload_free(struct module *mod) 671static void module_unload_free(struct module *mod)
@@ -822,7 +848,7 @@ SYSCALL_DEFINE2(delete_module, const char __user *, name_user,
822 mutex_lock(&module_mutex); 848 mutex_lock(&module_mutex);
823 /* Store the name of the last unloaded module for diagnostic purposes */ 849 /* Store the name of the last unloaded module for diagnostic purposes */
824 strlcpy(last_unloaded_module, mod->name, sizeof(last_unloaded_module)); 850 strlcpy(last_unloaded_module, mod->name, sizeof(last_unloaded_module));
825 unregister_dynamic_debug_module(mod->name); 851 ddebug_remove_module(mod->name);
826 free_module(mod); 852 free_module(mod);
827 853
828 out: 854 out:
@@ -860,7 +886,7 @@ void __symbol_put(const char *symbol)
860 struct module *owner; 886 struct module *owner;
861 887
862 preempt_disable(); 888 preempt_disable();
863 if (IS_ERR_VALUE(find_symbol(symbol, &owner, NULL, true, false))) 889 if (!find_symbol(symbol, &owner, NULL, true, false))
864 BUG(); 890 BUG();
865 module_put(owner); 891 module_put(owner);
866 preempt_enable(); 892 preempt_enable();
@@ -874,8 +900,10 @@ void symbol_put_addr(void *addr)
874 if (core_kernel_text((unsigned long)addr)) 900 if (core_kernel_text((unsigned long)addr))
875 return; 901 return;
876 902
877 if (!(modaddr = module_text_address((unsigned long)addr))) 903 /* module_text_address is safe here: we're supposed to have reference
878 BUG(); 904 * to module from symbol_get, so it can't go away. */
905 modaddr = __module_text_address((unsigned long)addr);
906 BUG_ON(!modaddr);
879 module_put(modaddr); 907 module_put(modaddr);
880} 908}
881EXPORT_SYMBOL_GPL(symbol_put_addr); 909EXPORT_SYMBOL_GPL(symbol_put_addr);
@@ -915,10 +943,11 @@ static inline void module_unload_free(struct module *mod)
915{ 943{
916} 944}
917 945
918static inline int use_module(struct module *a, struct module *b) 946int use_module(struct module *a, struct module *b)
919{ 947{
920 return strong_try_module_get(b) == 0; 948 return strong_try_module_get(b) == 0;
921} 949}
950EXPORT_SYMBOL_GPL(use_module);
922 951
923static inline void module_unload_init(struct module *mod) 952static inline void module_unload_init(struct module *mod)
924{ 953{
@@ -961,12 +990,12 @@ static struct module_attribute *modinfo_attrs[] = {
961 990
962static const char vermagic[] = VERMAGIC_STRING; 991static const char vermagic[] = VERMAGIC_STRING;
963 992
964static int try_to_force_load(struct module *mod, const char *symname) 993static int try_to_force_load(struct module *mod, const char *reason)
965{ 994{
966#ifdef CONFIG_MODULE_FORCE_LOAD 995#ifdef CONFIG_MODULE_FORCE_LOAD
967 if (!test_taint(TAINT_FORCED_MODULE)) 996 if (!test_taint(TAINT_FORCED_MODULE))
968 printk("%s: no version for \"%s\" found: kernel tainted.\n", 997 printk(KERN_WARNING "%s: %s: kernel tainted.\n",
969 mod->name, symname); 998 mod->name, reason);
970 add_taint_module(mod, TAINT_FORCED_MODULE); 999 add_taint_module(mod, TAINT_FORCED_MODULE);
971 return 0; 1000 return 0;
972#else 1001#else
@@ -1023,9 +1052,9 @@ static inline int check_modstruct_version(Elf_Shdr *sechdrs,
1023{ 1052{
1024 const unsigned long *crc; 1053 const unsigned long *crc;
1025 1054
1026 if (IS_ERR_VALUE(find_symbol("struct_module", NULL, &crc, true, false))) 1055 if (!find_symbol("module_layout", NULL, &crc, true, false))
1027 BUG(); 1056 BUG();
1028 return check_version(sechdrs, versindex, "struct_module", mod, crc); 1057 return check_version(sechdrs, versindex, "module_layout", mod, crc);
1029} 1058}
1030 1059
1031/* First part is kernel version, which we ignore if module has crcs. */ 1060/* First part is kernel version, which we ignore if module has crcs. */
@@ -1064,25 +1093,25 @@ static inline int same_magic(const char *amagic, const char *bmagic,
1064 1093
1065/* Resolve a symbol for this module. I.e. if we find one, record usage. 1094/* Resolve a symbol for this module. I.e. if we find one, record usage.
1066 Must be holding module_mutex. */ 1095 Must be holding module_mutex. */
1067static unsigned long resolve_symbol(Elf_Shdr *sechdrs, 1096static const struct kernel_symbol *resolve_symbol(Elf_Shdr *sechdrs,
1068 unsigned int versindex, 1097 unsigned int versindex,
1069 const char *name, 1098 const char *name,
1070 struct module *mod) 1099 struct module *mod)
1071{ 1100{
1072 struct module *owner; 1101 struct module *owner;
1073 unsigned long ret; 1102 const struct kernel_symbol *sym;
1074 const unsigned long *crc; 1103 const unsigned long *crc;
1075 1104
1076 ret = find_symbol(name, &owner, &crc, 1105 sym = find_symbol(name, &owner, &crc,
1077 !(mod->taints & (1 << TAINT_PROPRIETARY_MODULE)), true); 1106 !(mod->taints & (1 << TAINT_PROPRIETARY_MODULE)), true);
1078 if (!IS_ERR_VALUE(ret)) { 1107 /* use_module can fail due to OOM,
1079 /* use_module can fail due to OOM, 1108 or module initialization or unloading */
1080 or module initialization or unloading */ 1109 if (sym) {
1081 if (!check_version(sechdrs, versindex, name, mod, crc) || 1110 if (!check_version(sechdrs, versindex, name, mod, crc) ||
1082 !use_module(mod, owner)) 1111 !use_module(mod, owner))
1083 ret = -EINVAL; 1112 sym = NULL;
1084 } 1113 }
1085 return ret; 1114 return sym;
1086} 1115}
1087 1116
1088/* 1117/*
@@ -1457,6 +1486,9 @@ static void free_module(struct module *mod)
1457 /* Module unload stuff */ 1486 /* Module unload stuff */
1458 module_unload_free(mod); 1487 module_unload_free(mod);
1459 1488
1489 /* Free any allocated parameters. */
1490 destroy_params(mod->kp, mod->num_kp);
1491
1460 /* release any pointers to mcount in this module */ 1492 /* release any pointers to mcount in this module */
1461 ftrace_release(mod->module_core, mod->core_size); 1493 ftrace_release(mod->module_core, mod->core_size);
1462 1494
@@ -1479,17 +1511,15 @@ static void free_module(struct module *mod)
1479void *__symbol_get(const char *symbol) 1511void *__symbol_get(const char *symbol)
1480{ 1512{
1481 struct module *owner; 1513 struct module *owner;
1482 unsigned long value; 1514 const struct kernel_symbol *sym;
1483 1515
1484 preempt_disable(); 1516 preempt_disable();
1485 value = find_symbol(symbol, &owner, NULL, true, true); 1517 sym = find_symbol(symbol, &owner, NULL, true, true);
1486 if (IS_ERR_VALUE(value)) 1518 if (sym && strong_try_module_get(owner))
1487 value = 0; 1519 sym = NULL;
1488 else if (strong_try_module_get(owner))
1489 value = 0;
1490 preempt_enable(); 1520 preempt_enable();
1491 1521
1492 return (void *)value; 1522 return sym ? (void *)sym->value : NULL;
1493} 1523}
1494EXPORT_SYMBOL_GPL(__symbol_get); 1524EXPORT_SYMBOL_GPL(__symbol_get);
1495 1525
@@ -1517,8 +1547,7 @@ static int verify_export_symbols(struct module *mod)
1517 1547
1518 for (i = 0; i < ARRAY_SIZE(arr); i++) { 1548 for (i = 0; i < ARRAY_SIZE(arr); i++) {
1519 for (s = arr[i].sym; s < arr[i].sym + arr[i].num; s++) { 1549 for (s = arr[i].sym; s < arr[i].sym + arr[i].num; s++) {
1520 if (!IS_ERR_VALUE(find_symbol(s->name, &owner, 1550 if (find_symbol(s->name, &owner, NULL, true, false)) {
1521 NULL, true, false))) {
1522 printk(KERN_ERR 1551 printk(KERN_ERR
1523 "%s: exports duplicate symbol %s" 1552 "%s: exports duplicate symbol %s"
1524 " (owned by %s)\n", 1553 " (owned by %s)\n",
@@ -1542,6 +1571,7 @@ static int simplify_symbols(Elf_Shdr *sechdrs,
1542 unsigned long secbase; 1571 unsigned long secbase;
1543 unsigned int i, n = sechdrs[symindex].sh_size / sizeof(Elf_Sym); 1572 unsigned int i, n = sechdrs[symindex].sh_size / sizeof(Elf_Sym);
1544 int ret = 0; 1573 int ret = 0;
1574 const struct kernel_symbol *ksym;
1545 1575
1546 for (i = 1; i < n; i++) { 1576 for (i = 1; i < n; i++) {
1547 switch (sym[i].st_shndx) { 1577 switch (sym[i].st_shndx) {
@@ -1561,13 +1591,14 @@ static int simplify_symbols(Elf_Shdr *sechdrs,
1561 break; 1591 break;
1562 1592
1563 case SHN_UNDEF: 1593 case SHN_UNDEF:
1564 sym[i].st_value 1594 ksym = resolve_symbol(sechdrs, versindex,
1565 = resolve_symbol(sechdrs, versindex, 1595 strtab + sym[i].st_name, mod);
1566 strtab + sym[i].st_name, mod);
1567
1568 /* Ok if resolved. */ 1596 /* Ok if resolved. */
1569 if (!IS_ERR_VALUE(sym[i].st_value)) 1597 if (ksym) {
1598 sym[i].st_value = ksym->value;
1570 break; 1599 break;
1600 }
1601
1571 /* Ok if weak. */ 1602 /* Ok if weak. */
1572 if (ELF_ST_BIND(sym[i].st_info) == STB_WEAK) 1603 if (ELF_ST_BIND(sym[i].st_info) == STB_WEAK)
1573 break; 1604 break;
@@ -1642,8 +1673,7 @@ static void layout_sections(struct module *mod,
1642 if ((s->sh_flags & masks[m][0]) != masks[m][0] 1673 if ((s->sh_flags & masks[m][0]) != masks[m][0]
1643 || (s->sh_flags & masks[m][1]) 1674 || (s->sh_flags & masks[m][1])
1644 || s->sh_entsize != ~0UL 1675 || s->sh_entsize != ~0UL
1645 || strncmp(secstrings + s->sh_name, 1676 || strstarts(secstrings + s->sh_name, ".init"))
1646 ".init", 5) == 0)
1647 continue; 1677 continue;
1648 s->sh_entsize = get_offset(mod, &mod->core_size, s, i); 1678 s->sh_entsize = get_offset(mod, &mod->core_size, s, i);
1649 DEBUGP("\t%s\n", secstrings + s->sh_name); 1679 DEBUGP("\t%s\n", secstrings + s->sh_name);
@@ -1660,8 +1690,7 @@ static void layout_sections(struct module *mod,
1660 if ((s->sh_flags & masks[m][0]) != masks[m][0] 1690 if ((s->sh_flags & masks[m][0]) != masks[m][0]
1661 || (s->sh_flags & masks[m][1]) 1691 || (s->sh_flags & masks[m][1])
1662 || s->sh_entsize != ~0UL 1692 || s->sh_entsize != ~0UL
1663 || strncmp(secstrings + s->sh_name, 1693 || !strstarts(secstrings + s->sh_name, ".init"))
1664 ".init", 5) != 0)
1665 continue; 1694 continue;
1666 s->sh_entsize = (get_offset(mod, &mod->init_size, s, i) 1695 s->sh_entsize = (get_offset(mod, &mod->init_size, s, i)
1667 | INIT_OFFSET_MASK); 1696 | INIT_OFFSET_MASK);
@@ -1794,8 +1823,7 @@ static char elf_type(const Elf_Sym *sym,
1794 else 1823 else
1795 return 'b'; 1824 return 'b';
1796 } 1825 }
1797 if (strncmp(secstrings + sechdrs[sym->st_shndx].sh_name, 1826 if (strstarts(secstrings + sechdrs[sym->st_shndx].sh_name, ".debug"))
1798 ".debug", strlen(".debug")) == 0)
1799 return 'n'; 1827 return 'n';
1800 return '?'; 1828 return '?';
1801} 1829}
@@ -1827,19 +1855,13 @@ static inline void add_kallsyms(struct module *mod,
1827} 1855}
1828#endif /* CONFIG_KALLSYMS */ 1856#endif /* CONFIG_KALLSYMS */
1829 1857
1830static void dynamic_printk_setup(struct mod_debug *debug, unsigned int num) 1858static void dynamic_debug_setup(struct _ddebug *debug, unsigned int num)
1831{ 1859{
1832#ifdef CONFIG_DYNAMIC_PRINTK_DEBUG 1860#ifdef CONFIG_DYNAMIC_DEBUG
1833 unsigned int i; 1861 if (ddebug_add_module(debug, num, debug->modname))
1834 1862 printk(KERN_ERR "dynamic debug error adding module: %s\n",
1835 for (i = 0; i < num; i++) { 1863 debug->modname);
1836 register_dynamic_debug_module(debug[i].modname, 1864#endif
1837 debug[i].type,
1838 debug[i].logical_modname,
1839 debug[i].flag_names,
1840 debug[i].hash, debug[i].hash2);
1841 }
1842#endif /* CONFIG_DYNAMIC_PRINTK_DEBUG */
1843} 1865}
1844 1866
1845static void *module_alloc_update_bounds(unsigned long size) 1867static void *module_alloc_update_bounds(unsigned long size)
@@ -1870,8 +1892,7 @@ static noinline struct module *load_module(void __user *umod,
1870 unsigned int symindex = 0; 1892 unsigned int symindex = 0;
1871 unsigned int strindex = 0; 1893 unsigned int strindex = 0;
1872 unsigned int modindex, versindex, infoindex, pcpuindex; 1894 unsigned int modindex, versindex, infoindex, pcpuindex;
1873 unsigned int num_kp, num_mcount; 1895 unsigned int num_mcount;
1874 struct kernel_param *kp;
1875 struct module *mod; 1896 struct module *mod;
1876 long err = 0; 1897 long err = 0;
1877 void *percpu = NULL, *ptr = NULL; /* Stops spurious gcc warning */ 1898 void *percpu = NULL, *ptr = NULL; /* Stops spurious gcc warning */
@@ -1888,12 +1909,6 @@ static noinline struct module *load_module(void __user *umod,
1888 if (len > 64 * 1024 * 1024 || (hdr = vmalloc(len)) == NULL) 1909 if (len > 64 * 1024 * 1024 || (hdr = vmalloc(len)) == NULL)
1889 return ERR_PTR(-ENOMEM); 1910 return ERR_PTR(-ENOMEM);
1890 1911
1891 /* Create stop_machine threads since the error path relies on
1892 * a non-failing stop_machine call. */
1893 err = stop_machine_create();
1894 if (err)
1895 goto free_hdr;
1896
1897 if (copy_from_user(hdr, umod, len) != 0) { 1912 if (copy_from_user(hdr, umod, len) != 0) {
1898 err = -EFAULT; 1913 err = -EFAULT;
1899 goto free_hdr; 1914 goto free_hdr;
@@ -1934,9 +1949,12 @@ static noinline struct module *load_module(void __user *umod,
1934 } 1949 }
1935#ifndef CONFIG_MODULE_UNLOAD 1950#ifndef CONFIG_MODULE_UNLOAD
1936 /* Don't load .exit sections */ 1951 /* Don't load .exit sections */
1937 if (strncmp(secstrings+sechdrs[i].sh_name, ".exit", 5) == 0) 1952 if (strstarts(secstrings+sechdrs[i].sh_name, ".exit"))
1938 sechdrs[i].sh_flags &= ~(unsigned long)SHF_ALLOC; 1953 sechdrs[i].sh_flags &= ~(unsigned long)SHF_ALLOC;
1939#endif 1954#endif
1955 /* Don't keep __versions around; it's just for loading. */
1956 if (strcmp(secstrings + sechdrs[i].sh_name, "__versions") == 0)
1957 sechdrs[i].sh_flags &= ~(unsigned long)SHF_ALLOC;
1940 } 1958 }
1941 1959
1942 modindex = find_sec(hdr, sechdrs, secstrings, 1960 modindex = find_sec(hdr, sechdrs, secstrings,
@@ -1978,7 +1996,7 @@ static noinline struct module *load_module(void __user *umod,
1978 modmagic = get_modinfo(sechdrs, infoindex, "vermagic"); 1996 modmagic = get_modinfo(sechdrs, infoindex, "vermagic");
1979 /* This is allowed: modprobe --force will invalidate it. */ 1997 /* This is allowed: modprobe --force will invalidate it. */
1980 if (!modmagic) { 1998 if (!modmagic) {
1981 err = try_to_force_load(mod, "magic"); 1999 err = try_to_force_load(mod, "bad vermagic");
1982 if (err) 2000 if (err)
1983 goto free_hdr; 2001 goto free_hdr;
1984 } else if (!same_magic(modmagic, vermagic, versindex)) { 2002 } else if (!same_magic(modmagic, vermagic, versindex)) {
@@ -2015,14 +2033,6 @@ static noinline struct module *load_module(void __user *umod,
2015 if (err < 0) 2033 if (err < 0)
2016 goto free_mod; 2034 goto free_mod;
2017 2035
2018#if defined(CONFIG_MODULE_UNLOAD) && defined(CONFIG_SMP)
2019 mod->refptr = percpu_modalloc(sizeof(local_t), __alignof__(local_t),
2020 mod->name);
2021 if (!mod->refptr) {
2022 err = -ENOMEM;
2023 goto free_mod;
2024 }
2025#endif
2026 if (pcpuindex) { 2036 if (pcpuindex) {
2027 /* We have a special allocation for this section. */ 2037 /* We have a special allocation for this section. */
2028 percpu = percpu_modalloc(sechdrs[pcpuindex].sh_size, 2038 percpu = percpu_modalloc(sechdrs[pcpuindex].sh_size,
@@ -2030,7 +2040,7 @@ static noinline struct module *load_module(void __user *umod,
2030 mod->name); 2040 mod->name);
2031 if (!percpu) { 2041 if (!percpu) {
2032 err = -ENOMEM; 2042 err = -ENOMEM;
2033 goto free_percpu; 2043 goto free_mod;
2034 } 2044 }
2035 sechdrs[pcpuindex].sh_flags &= ~(unsigned long)SHF_ALLOC; 2045 sechdrs[pcpuindex].sh_flags &= ~(unsigned long)SHF_ALLOC;
2036 mod->percpu = percpu; 2046 mod->percpu = percpu;
@@ -2082,6 +2092,14 @@ static noinline struct module *load_module(void __user *umod,
2082 /* Module has been moved. */ 2092 /* Module has been moved. */
2083 mod = (void *)sechdrs[modindex].sh_addr; 2093 mod = (void *)sechdrs[modindex].sh_addr;
2084 2094
2095#if defined(CONFIG_MODULE_UNLOAD) && defined(CONFIG_SMP)
2096 mod->refptr = percpu_modalloc(sizeof(local_t), __alignof__(local_t),
2097 mod->name);
2098 if (!mod->refptr) {
2099 err = -ENOMEM;
2100 goto free_init;
2101 }
2102#endif
2085 /* Now we've moved module, initialize linked lists, etc. */ 2103 /* Now we've moved module, initialize linked lists, etc. */
2086 module_unload_init(mod); 2104 module_unload_init(mod);
2087 2105
@@ -2116,8 +2134,8 @@ static noinline struct module *load_module(void __user *umod,
2116 2134
2117 /* Now we've got everything in the final locations, we can 2135 /* Now we've got everything in the final locations, we can
2118 * find optional sections. */ 2136 * find optional sections. */
2119 kp = section_objs(hdr, sechdrs, secstrings, "__param", sizeof(*kp), 2137 mod->kp = section_objs(hdr, sechdrs, secstrings, "__param",
2120 &num_kp); 2138 sizeof(*mod->kp), &mod->num_kp);
2121 mod->syms = section_objs(hdr, sechdrs, secstrings, "__ksymtab", 2139 mod->syms = section_objs(hdr, sechdrs, secstrings, "__ksymtab",
2122 sizeof(*mod->syms), &mod->num_syms); 2140 sizeof(*mod->syms), &mod->num_syms);
2123 mod->crcs = section_addr(hdr, sechdrs, secstrings, "__kcrctab"); 2141 mod->crcs = section_addr(hdr, sechdrs, secstrings, "__kcrctab");
@@ -2167,8 +2185,8 @@ static noinline struct module *load_module(void __user *umod,
2167 || (mod->num_unused_gpl_syms && !mod->unused_gpl_crcs) 2185 || (mod->num_unused_gpl_syms && !mod->unused_gpl_crcs)
2168#endif 2186#endif
2169 ) { 2187 ) {
2170 printk(KERN_WARNING "%s: No versions for exported symbols.\n", mod->name); 2188 err = try_to_force_load(mod,
2171 err = try_to_force_load(mod, "nocrc"); 2189 "no versions for exported symbols");
2172 if (err) 2190 if (err)
2173 goto cleanup; 2191 goto cleanup;
2174 } 2192 }
@@ -2213,12 +2231,13 @@ static noinline struct module *load_module(void __user *umod,
2213 add_kallsyms(mod, sechdrs, symindex, strindex, secstrings); 2231 add_kallsyms(mod, sechdrs, symindex, strindex, secstrings);
2214 2232
2215 if (!mod->taints) { 2233 if (!mod->taints) {
2216 struct mod_debug *debug; 2234 struct _ddebug *debug;
2217 unsigned int num_debug; 2235 unsigned int num_debug;
2218 2236
2219 debug = section_objs(hdr, sechdrs, secstrings, "__verbose", 2237 debug = section_objs(hdr, sechdrs, secstrings, "__verbose",
2220 sizeof(*debug), &num_debug); 2238 sizeof(*debug), &num_debug);
2221 dynamic_printk_setup(debug, num_debug); 2239 if (debug)
2240 dynamic_debug_setup(debug, num_debug);
2222 } 2241 }
2223 2242
2224 /* sechdrs[0].sh_size is always zero */ 2243 /* sechdrs[0].sh_size is always zero */
@@ -2262,11 +2281,11 @@ static noinline struct module *load_module(void __user *umod,
2262 */ 2281 */
2263 list_add_rcu(&mod->list, &modules); 2282 list_add_rcu(&mod->list, &modules);
2264 2283
2265 err = parse_args(mod->name, mod->args, kp, num_kp, NULL); 2284 err = parse_args(mod->name, mod->args, mod->kp, mod->num_kp, NULL);
2266 if (err < 0) 2285 if (err < 0)
2267 goto unlink; 2286 goto unlink;
2268 2287
2269 err = mod_sysfs_setup(mod, kp, num_kp); 2288 err = mod_sysfs_setup(mod, mod->kp, mod->num_kp);
2270 if (err < 0) 2289 if (err < 0)
2271 goto unlink; 2290 goto unlink;
2272 add_sect_attrs(mod, hdr->e_shnum, secstrings, sechdrs); 2291 add_sect_attrs(mod, hdr->e_shnum, secstrings, sechdrs);
@@ -2275,12 +2294,13 @@ static noinline struct module *load_module(void __user *umod,
2275 /* Get rid of temporary copy */ 2294 /* Get rid of temporary copy */
2276 vfree(hdr); 2295 vfree(hdr);
2277 2296
2278 stop_machine_destroy();
2279 /* Done! */ 2297 /* Done! */
2280 return mod; 2298 return mod;
2281 2299
2282 unlink: 2300 unlink:
2283 stop_machine(__unlink_module, mod, NULL); 2301 /* Unlink carefully: kallsyms could be walking list. */
2302 list_del_rcu(&mod->list);
2303 synchronize_sched();
2284 module_arch_cleanup(mod); 2304 module_arch_cleanup(mod);
2285 cleanup: 2305 cleanup:
2286 kobject_del(&mod->mkobj.kobj); 2306 kobject_del(&mod->mkobj.kobj);
@@ -2288,20 +2308,21 @@ static noinline struct module *load_module(void __user *umod,
2288 ftrace_release(mod->module_core, mod->core_size); 2308 ftrace_release(mod->module_core, mod->core_size);
2289 free_unload: 2309 free_unload:
2290 module_unload_free(mod); 2310 module_unload_free(mod);
2311#if defined(CONFIG_MODULE_UNLOAD) && defined(CONFIG_SMP)
2312 free_init:
2313 percpu_modfree(mod->refptr);
2314#endif
2291 module_free(mod, mod->module_init); 2315 module_free(mod, mod->module_init);
2292 free_core: 2316 free_core:
2293 module_free(mod, mod->module_core); 2317 module_free(mod, mod->module_core);
2318 /* mod will be freed with core. Don't access it beyond this line! */
2294 free_percpu: 2319 free_percpu:
2295 if (percpu) 2320 if (percpu)
2296 percpu_modfree(percpu); 2321 percpu_modfree(percpu);
2297#if defined(CONFIG_MODULE_UNLOAD) && defined(CONFIG_SMP)
2298 percpu_modfree(mod->refptr);
2299#endif
2300 free_mod: 2322 free_mod:
2301 kfree(args); 2323 kfree(args);
2302 free_hdr: 2324 free_hdr:
2303 vfree(hdr); 2325 vfree(hdr);
2304 stop_machine_destroy();
2305 return ERR_PTR(err); 2326 return ERR_PTR(err);
2306 2327
2307 truncated: 2328 truncated:
@@ -2578,6 +2599,25 @@ unsigned long module_kallsyms_lookup_name(const char *name)
2578 preempt_enable(); 2599 preempt_enable();
2579 return ret; 2600 return ret;
2580} 2601}
2602
2603int module_kallsyms_on_each_symbol(int (*fn)(void *, const char *,
2604 struct module *, unsigned long),
2605 void *data)
2606{
2607 struct module *mod;
2608 unsigned int i;
2609 int ret;
2610
2611 list_for_each_entry(mod, &modules, list) {
2612 for (i = 0; i < mod->num_symtab; i++) {
2613 ret = fn(data, mod->strtab + mod->symtab[i].st_name,
2614 mod, mod->symtab[i].st_value);
2615 if (ret != 0)
2616 return ret;
2617 }
2618 }
2619 return 0;
2620}
2581#endif /* CONFIG_KALLSYMS */ 2621#endif /* CONFIG_KALLSYMS */
2582 2622
2583static char *module_flags(struct module *mod, char *buf) 2623static char *module_flags(struct module *mod, char *buf)
@@ -2713,29 +2753,31 @@ const struct exception_table_entry *search_module_extables(unsigned long addr)
2713} 2753}
2714 2754
2715/* 2755/*
2716 * Is this a valid module address? 2756 * is_module_address - is this address inside a module?
2757 * @addr: the address to check.
2758 *
2759 * See is_module_text_address() if you simply want to see if the address
2760 * is code (not data).
2717 */ 2761 */
2718int is_module_address(unsigned long addr) 2762bool is_module_address(unsigned long addr)
2719{ 2763{
2720 struct module *mod; 2764 bool ret;
2721 2765
2722 preempt_disable(); 2766 preempt_disable();
2723 2767 ret = __module_address(addr) != NULL;
2724 list_for_each_entry_rcu(mod, &modules, list) {
2725 if (within_module_core(addr, mod)) {
2726 preempt_enable();
2727 return 1;
2728 }
2729 }
2730
2731 preempt_enable(); 2768 preempt_enable();
2732 2769
2733 return 0; 2770 return ret;
2734} 2771}
2735 2772
2736 2773/*
2737/* Is this a valid kernel address? */ 2774 * __module_address - get the module which contains an address.
2738__notrace_funcgraph struct module *__module_text_address(unsigned long addr) 2775 * @addr: the address.
2776 *
2777 * Must be called with preempt disabled or module mutex held so that
2778 * module doesn't get freed during this.
2779 */
2780struct module *__module_address(unsigned long addr)
2739{ 2781{
2740 struct module *mod; 2782 struct module *mod;
2741 2783
@@ -2743,22 +2785,51 @@ __notrace_funcgraph struct module *__module_text_address(unsigned long addr)
2743 return NULL; 2785 return NULL;
2744 2786
2745 list_for_each_entry_rcu(mod, &modules, list) 2787 list_for_each_entry_rcu(mod, &modules, list)
2746 if (within(addr, mod->module_init, mod->init_text_size) 2788 if (within_module_core(addr, mod)
2747 || within(addr, mod->module_core, mod->core_text_size)) 2789 || within_module_init(addr, mod))
2748 return mod; 2790 return mod;
2749 return NULL; 2791 return NULL;
2750} 2792}
2793EXPORT_SYMBOL_GPL(__module_address);
2751 2794
2752struct module *module_text_address(unsigned long addr) 2795/*
2796 * is_module_text_address - is this address inside module code?
2797 * @addr: the address to check.
2798 *
2799 * See is_module_address() if you simply want to see if the address is
2800 * anywhere in a module. See kernel_text_address() for testing if an
2801 * address corresponds to kernel or module code.
2802 */
2803bool is_module_text_address(unsigned long addr)
2753{ 2804{
2754 struct module *mod; 2805 bool ret;
2755 2806
2756 preempt_disable(); 2807 preempt_disable();
2757 mod = __module_text_address(addr); 2808 ret = __module_text_address(addr) != NULL;
2758 preempt_enable(); 2809 preempt_enable();
2759 2810
2811 return ret;
2812}
2813
2814/*
2815 * __module_text_address - get the module whose code contains an address.
2816 * @addr: the address.
2817 *
2818 * Must be called with preempt disabled or module mutex held so that
2819 * module doesn't get freed during this.
2820 */
2821struct module *__module_text_address(unsigned long addr)
2822{
2823 struct module *mod = __module_address(addr);
2824 if (mod) {
2825 /* Make sure it's within the text section. */
2826 if (!within(addr, mod->module_init, mod->init_text_size)
2827 && !within(addr, mod->module_core, mod->core_text_size))
2828 mod = NULL;
2829 }
2760 return mod; 2830 return mod;
2761} 2831}
2832EXPORT_SYMBOL_GPL(__module_text_address);
2762 2833
2763/* Don't grab lock, we're oopsing. */ 2834/* Don't grab lock, we're oopsing. */
2764void print_modules(void) 2835void print_modules(void)
@@ -2778,9 +2849,17 @@ void print_modules(void)
2778} 2849}
2779 2850
2780#ifdef CONFIG_MODVERSIONS 2851#ifdef CONFIG_MODVERSIONS
2781/* Generate the signature for struct module here, too, for modversions. */ 2852/* Generate the signature for all relevant module structures here.
2782void struct_module(struct module *mod) { return; } 2853 * If these change, we don't want to try to parse the module. */
2783EXPORT_SYMBOL(struct_module); 2854void module_layout(struct module *mod,
2855 struct modversion_info *ver,
2856 struct kernel_param *kp,
2857 struct kernel_symbol *ks,
2858 struct marker *marker,
2859 struct tracepoint *tp)
2860{
2861}
2862EXPORT_SYMBOL(module_layout);
2784#endif 2863#endif
2785 2864
2786#ifdef CONFIG_MARKERS 2865#ifdef CONFIG_MARKERS
diff --git a/kernel/mutex-debug.c b/kernel/mutex-debug.c
index 1d94160eb532..50d022e5a560 100644
--- a/kernel/mutex-debug.c
+++ b/kernel/mutex-debug.c
@@ -26,11 +26,6 @@
26/* 26/*
27 * Must be called with lock->wait_lock held. 27 * Must be called with lock->wait_lock held.
28 */ 28 */
29void debug_mutex_set_owner(struct mutex *lock, struct thread_info *new_owner)
30{
31 lock->owner = new_owner;
32}
33
34void debug_mutex_lock_common(struct mutex *lock, struct mutex_waiter *waiter) 29void debug_mutex_lock_common(struct mutex *lock, struct mutex_waiter *waiter)
35{ 30{
36 memset(waiter, MUTEX_DEBUG_INIT, sizeof(*waiter)); 31 memset(waiter, MUTEX_DEBUG_INIT, sizeof(*waiter));
@@ -59,7 +54,6 @@ void debug_mutex_add_waiter(struct mutex *lock, struct mutex_waiter *waiter,
59 54
60 /* Mark the current thread as blocked on the lock: */ 55 /* Mark the current thread as blocked on the lock: */
61 ti->task->blocked_on = waiter; 56 ti->task->blocked_on = waiter;
62 waiter->lock = lock;
63} 57}
64 58
65void mutex_remove_waiter(struct mutex *lock, struct mutex_waiter *waiter, 59void mutex_remove_waiter(struct mutex *lock, struct mutex_waiter *waiter,
@@ -82,7 +76,7 @@ void debug_mutex_unlock(struct mutex *lock)
82 DEBUG_LOCKS_WARN_ON(lock->magic != lock); 76 DEBUG_LOCKS_WARN_ON(lock->magic != lock);
83 DEBUG_LOCKS_WARN_ON(lock->owner != current_thread_info()); 77 DEBUG_LOCKS_WARN_ON(lock->owner != current_thread_info());
84 DEBUG_LOCKS_WARN_ON(!lock->wait_list.prev && !lock->wait_list.next); 78 DEBUG_LOCKS_WARN_ON(!lock->wait_list.prev && !lock->wait_list.next);
85 DEBUG_LOCKS_WARN_ON(lock->owner != current_thread_info()); 79 mutex_clear_owner(lock);
86} 80}
87 81
88void debug_mutex_init(struct mutex *lock, const char *name, 82void debug_mutex_init(struct mutex *lock, const char *name,
@@ -95,7 +89,6 @@ void debug_mutex_init(struct mutex *lock, const char *name,
95 debug_check_no_locks_freed((void *)lock, sizeof(*lock)); 89 debug_check_no_locks_freed((void *)lock, sizeof(*lock));
96 lockdep_init_map(&lock->dep_map, name, key, 0); 90 lockdep_init_map(&lock->dep_map, name, key, 0);
97#endif 91#endif
98 lock->owner = NULL;
99 lock->magic = lock; 92 lock->magic = lock;
100} 93}
101 94
diff --git a/kernel/mutex-debug.h b/kernel/mutex-debug.h
index babfbdfc534b..6b2d735846a5 100644
--- a/kernel/mutex-debug.h
+++ b/kernel/mutex-debug.h
@@ -13,14 +13,6 @@
13/* 13/*
14 * This must be called with lock->wait_lock held. 14 * This must be called with lock->wait_lock held.
15 */ 15 */
16extern void
17debug_mutex_set_owner(struct mutex *lock, struct thread_info *new_owner);
18
19static inline void debug_mutex_clear_owner(struct mutex *lock)
20{
21 lock->owner = NULL;
22}
23
24extern void debug_mutex_lock_common(struct mutex *lock, 16extern void debug_mutex_lock_common(struct mutex *lock,
25 struct mutex_waiter *waiter); 17 struct mutex_waiter *waiter);
26extern void debug_mutex_wake_waiter(struct mutex *lock, 18extern void debug_mutex_wake_waiter(struct mutex *lock,
@@ -35,6 +27,16 @@ extern void debug_mutex_unlock(struct mutex *lock);
35extern void debug_mutex_init(struct mutex *lock, const char *name, 27extern void debug_mutex_init(struct mutex *lock, const char *name,
36 struct lock_class_key *key); 28 struct lock_class_key *key);
37 29
30static inline void mutex_set_owner(struct mutex *lock)
31{
32 lock->owner = current_thread_info();
33}
34
35static inline void mutex_clear_owner(struct mutex *lock)
36{
37 lock->owner = NULL;
38}
39
38#define spin_lock_mutex(lock, flags) \ 40#define spin_lock_mutex(lock, flags) \
39 do { \ 41 do { \
40 struct mutex *l = container_of(lock, struct mutex, wait_lock); \ 42 struct mutex *l = container_of(lock, struct mutex, wait_lock); \
diff --git a/kernel/mutex.c b/kernel/mutex.c
index 4f45d4b658ef..5d79781394a3 100644
--- a/kernel/mutex.c
+++ b/kernel/mutex.c
@@ -10,6 +10,11 @@
10 * Many thanks to Arjan van de Ven, Thomas Gleixner, Steven Rostedt and 10 * Many thanks to Arjan van de Ven, Thomas Gleixner, Steven Rostedt and
11 * David Howells for suggestions and improvements. 11 * David Howells for suggestions and improvements.
12 * 12 *
13 * - Adaptive spinning for mutexes by Peter Zijlstra. (Ported to mainline
14 * from the -rt tree, where it was originally implemented for rtmutexes
15 * by Steven Rostedt, based on work by Gregory Haskins, Peter Morreale
16 * and Sven Dietrich.
17 *
13 * Also see Documentation/mutex-design.txt. 18 * Also see Documentation/mutex-design.txt.
14 */ 19 */
15#include <linux/mutex.h> 20#include <linux/mutex.h>
@@ -46,6 +51,7 @@ __mutex_init(struct mutex *lock, const char *name, struct lock_class_key *key)
46 atomic_set(&lock->count, 1); 51 atomic_set(&lock->count, 1);
47 spin_lock_init(&lock->wait_lock); 52 spin_lock_init(&lock->wait_lock);
48 INIT_LIST_HEAD(&lock->wait_list); 53 INIT_LIST_HEAD(&lock->wait_list);
54 mutex_clear_owner(lock);
49 55
50 debug_mutex_init(lock, name, key); 56 debug_mutex_init(lock, name, key);
51} 57}
@@ -91,6 +97,7 @@ void inline __sched mutex_lock(struct mutex *lock)
91 * 'unlocked' into 'locked' state. 97 * 'unlocked' into 'locked' state.
92 */ 98 */
93 __mutex_fastpath_lock(&lock->count, __mutex_lock_slowpath); 99 __mutex_fastpath_lock(&lock->count, __mutex_lock_slowpath);
100 mutex_set_owner(lock);
94} 101}
95 102
96EXPORT_SYMBOL(mutex_lock); 103EXPORT_SYMBOL(mutex_lock);
@@ -115,6 +122,14 @@ void __sched mutex_unlock(struct mutex *lock)
115 * The unlocking fastpath is the 0->1 transition from 'locked' 122 * The unlocking fastpath is the 0->1 transition from 'locked'
116 * into 'unlocked' state: 123 * into 'unlocked' state:
117 */ 124 */
125#ifndef CONFIG_DEBUG_MUTEXES
126 /*
127 * When debugging is enabled we must not clear the owner before time,
128 * the slow path will always be taken, and that clears the owner field
129 * after verifying that it was indeed current.
130 */
131 mutex_clear_owner(lock);
132#endif
118 __mutex_fastpath_unlock(&lock->count, __mutex_unlock_slowpath); 133 __mutex_fastpath_unlock(&lock->count, __mutex_unlock_slowpath);
119} 134}
120 135
@@ -129,21 +144,75 @@ __mutex_lock_common(struct mutex *lock, long state, unsigned int subclass,
129{ 144{
130 struct task_struct *task = current; 145 struct task_struct *task = current;
131 struct mutex_waiter waiter; 146 struct mutex_waiter waiter;
132 unsigned int old_val;
133 unsigned long flags; 147 unsigned long flags;
134 148
149 preempt_disable();
150 mutex_acquire(&lock->dep_map, subclass, 0, ip);
151#if defined(CONFIG_SMP) && !defined(CONFIG_DEBUG_MUTEXES)
152 /*
153 * Optimistic spinning.
154 *
155 * We try to spin for acquisition when we find that there are no
156 * pending waiters and the lock owner is currently running on a
157 * (different) CPU.
158 *
159 * The rationale is that if the lock owner is running, it is likely to
160 * release the lock soon.
161 *
162 * Since this needs the lock owner, and this mutex implementation
163 * doesn't track the owner atomically in the lock field, we need to
164 * track it non-atomically.
165 *
166 * We can't do this for DEBUG_MUTEXES because that relies on wait_lock
167 * to serialize everything.
168 */
169
170 for (;;) {
171 struct thread_info *owner;
172
173 /*
174 * If there's an owner, wait for it to either
175 * release the lock or go to sleep.
176 */
177 owner = ACCESS_ONCE(lock->owner);
178 if (owner && !mutex_spin_on_owner(lock, owner))
179 break;
180
181 if (atomic_cmpxchg(&lock->count, 1, 0) == 1) {
182 lock_acquired(&lock->dep_map, ip);
183 mutex_set_owner(lock);
184 preempt_enable();
185 return 0;
186 }
187
188 /*
189 * When there's no owner, we might have preempted between the
190 * owner acquiring the lock and setting the owner field. If
191 * we're an RT task that will live-lock because we won't let
192 * the owner complete.
193 */
194 if (!owner && (need_resched() || rt_task(task)))
195 break;
196
197 /*
198 * The cpu_relax() call is a compiler barrier which forces
199 * everything in this loop to be re-loaded. We don't need
200 * memory barriers as we'll eventually observe the right
201 * values at the cost of a few extra spins.
202 */
203 cpu_relax();
204 }
205#endif
135 spin_lock_mutex(&lock->wait_lock, flags); 206 spin_lock_mutex(&lock->wait_lock, flags);
136 207
137 debug_mutex_lock_common(lock, &waiter); 208 debug_mutex_lock_common(lock, &waiter);
138 mutex_acquire(&lock->dep_map, subclass, 0, ip);
139 debug_mutex_add_waiter(lock, &waiter, task_thread_info(task)); 209 debug_mutex_add_waiter(lock, &waiter, task_thread_info(task));
140 210
141 /* add waiting tasks to the end of the waitqueue (FIFO): */ 211 /* add waiting tasks to the end of the waitqueue (FIFO): */
142 list_add_tail(&waiter.list, &lock->wait_list); 212 list_add_tail(&waiter.list, &lock->wait_list);
143 waiter.task = task; 213 waiter.task = task;
144 214
145 old_val = atomic_xchg(&lock->count, -1); 215 if (atomic_xchg(&lock->count, -1) == 1)
146 if (old_val == 1)
147 goto done; 216 goto done;
148 217
149 lock_contended(&lock->dep_map, ip); 218 lock_contended(&lock->dep_map, ip);
@@ -158,8 +227,7 @@ __mutex_lock_common(struct mutex *lock, long state, unsigned int subclass,
158 * that when we release the lock, we properly wake up the 227 * that when we release the lock, we properly wake up the
159 * other waiters: 228 * other waiters:
160 */ 229 */
161 old_val = atomic_xchg(&lock->count, -1); 230 if (atomic_xchg(&lock->count, -1) == 1)
162 if (old_val == 1)
163 break; 231 break;
164 232
165 /* 233 /*
@@ -173,21 +241,22 @@ __mutex_lock_common(struct mutex *lock, long state, unsigned int subclass,
173 spin_unlock_mutex(&lock->wait_lock, flags); 241 spin_unlock_mutex(&lock->wait_lock, flags);
174 242
175 debug_mutex_free_waiter(&waiter); 243 debug_mutex_free_waiter(&waiter);
244 preempt_enable();
176 return -EINTR; 245 return -EINTR;
177 } 246 }
178 __set_task_state(task, state); 247 __set_task_state(task, state);
179 248
180 /* didnt get the lock, go to sleep: */ 249 /* didnt get the lock, go to sleep: */
181 spin_unlock_mutex(&lock->wait_lock, flags); 250 spin_unlock_mutex(&lock->wait_lock, flags);
182 schedule(); 251 __schedule();
183 spin_lock_mutex(&lock->wait_lock, flags); 252 spin_lock_mutex(&lock->wait_lock, flags);
184 } 253 }
185 254
186done: 255done:
187 lock_acquired(&lock->dep_map, ip); 256 lock_acquired(&lock->dep_map, ip);
188 /* got the lock - rejoice! */ 257 /* got the lock - rejoice! */
189 mutex_remove_waiter(lock, &waiter, task_thread_info(task)); 258 mutex_remove_waiter(lock, &waiter, current_thread_info());
190 debug_mutex_set_owner(lock, task_thread_info(task)); 259 mutex_set_owner(lock);
191 260
192 /* set it to 0 if there are no waiters left: */ 261 /* set it to 0 if there are no waiters left: */
193 if (likely(list_empty(&lock->wait_list))) 262 if (likely(list_empty(&lock->wait_list)))
@@ -196,6 +265,7 @@ done:
196 spin_unlock_mutex(&lock->wait_lock, flags); 265 spin_unlock_mutex(&lock->wait_lock, flags);
197 266
198 debug_mutex_free_waiter(&waiter); 267 debug_mutex_free_waiter(&waiter);
268 preempt_enable();
199 269
200 return 0; 270 return 0;
201} 271}
@@ -222,7 +292,8 @@ int __sched
222mutex_lock_interruptible_nested(struct mutex *lock, unsigned int subclass) 292mutex_lock_interruptible_nested(struct mutex *lock, unsigned int subclass)
223{ 293{
224 might_sleep(); 294 might_sleep();
225 return __mutex_lock_common(lock, TASK_INTERRUPTIBLE, subclass, _RET_IP_); 295 return __mutex_lock_common(lock, TASK_INTERRUPTIBLE,
296 subclass, _RET_IP_);
226} 297}
227 298
228EXPORT_SYMBOL_GPL(mutex_lock_interruptible_nested); 299EXPORT_SYMBOL_GPL(mutex_lock_interruptible_nested);
@@ -260,8 +331,6 @@ __mutex_unlock_common_slowpath(atomic_t *lock_count, int nested)
260 wake_up_process(waiter->task); 331 wake_up_process(waiter->task);
261 } 332 }
262 333
263 debug_mutex_clear_owner(lock);
264
265 spin_unlock_mutex(&lock->wait_lock, flags); 334 spin_unlock_mutex(&lock->wait_lock, flags);
266} 335}
267 336
@@ -298,18 +367,30 @@ __mutex_lock_interruptible_slowpath(atomic_t *lock_count);
298 */ 367 */
299int __sched mutex_lock_interruptible(struct mutex *lock) 368int __sched mutex_lock_interruptible(struct mutex *lock)
300{ 369{
370 int ret;
371
301 might_sleep(); 372 might_sleep();
302 return __mutex_fastpath_lock_retval 373 ret = __mutex_fastpath_lock_retval
303 (&lock->count, __mutex_lock_interruptible_slowpath); 374 (&lock->count, __mutex_lock_interruptible_slowpath);
375 if (!ret)
376 mutex_set_owner(lock);
377
378 return ret;
304} 379}
305 380
306EXPORT_SYMBOL(mutex_lock_interruptible); 381EXPORT_SYMBOL(mutex_lock_interruptible);
307 382
308int __sched mutex_lock_killable(struct mutex *lock) 383int __sched mutex_lock_killable(struct mutex *lock)
309{ 384{
385 int ret;
386
310 might_sleep(); 387 might_sleep();
311 return __mutex_fastpath_lock_retval 388 ret = __mutex_fastpath_lock_retval
312 (&lock->count, __mutex_lock_killable_slowpath); 389 (&lock->count, __mutex_lock_killable_slowpath);
390 if (!ret)
391 mutex_set_owner(lock);
392
393 return ret;
313} 394}
314EXPORT_SYMBOL(mutex_lock_killable); 395EXPORT_SYMBOL(mutex_lock_killable);
315 396
@@ -352,9 +433,10 @@ static inline int __mutex_trylock_slowpath(atomic_t *lock_count)
352 433
353 prev = atomic_xchg(&lock->count, -1); 434 prev = atomic_xchg(&lock->count, -1);
354 if (likely(prev == 1)) { 435 if (likely(prev == 1)) {
355 debug_mutex_set_owner(lock, current_thread_info()); 436 mutex_set_owner(lock);
356 mutex_acquire(&lock->dep_map, 0, 1, _RET_IP_); 437 mutex_acquire(&lock->dep_map, 0, 1, _RET_IP_);
357 } 438 }
439
358 /* Set it back to 0 if there are no waiters: */ 440 /* Set it back to 0 if there are no waiters: */
359 if (likely(list_empty(&lock->wait_list))) 441 if (likely(list_empty(&lock->wait_list)))
360 atomic_set(&lock->count, 0); 442 atomic_set(&lock->count, 0);
@@ -380,8 +462,13 @@ static inline int __mutex_trylock_slowpath(atomic_t *lock_count)
380 */ 462 */
381int __sched mutex_trylock(struct mutex *lock) 463int __sched mutex_trylock(struct mutex *lock)
382{ 464{
383 return __mutex_fastpath_trylock(&lock->count, 465 int ret;
384 __mutex_trylock_slowpath); 466
467 ret = __mutex_fastpath_trylock(&lock->count, __mutex_trylock_slowpath);
468 if (ret)
469 mutex_set_owner(lock);
470
471 return ret;
385} 472}
386 473
387EXPORT_SYMBOL(mutex_trylock); 474EXPORT_SYMBOL(mutex_trylock);
diff --git a/kernel/mutex.h b/kernel/mutex.h
index a075dafbb290..67578ca48f94 100644
--- a/kernel/mutex.h
+++ b/kernel/mutex.h
@@ -16,8 +16,26 @@
16#define mutex_remove_waiter(lock, waiter, ti) \ 16#define mutex_remove_waiter(lock, waiter, ti) \
17 __list_del((waiter)->list.prev, (waiter)->list.next) 17 __list_del((waiter)->list.prev, (waiter)->list.next)
18 18
19#define debug_mutex_set_owner(lock, new_owner) do { } while (0) 19#ifdef CONFIG_SMP
20#define debug_mutex_clear_owner(lock) do { } while (0) 20static inline void mutex_set_owner(struct mutex *lock)
21{
22 lock->owner = current_thread_info();
23}
24
25static inline void mutex_clear_owner(struct mutex *lock)
26{
27 lock->owner = NULL;
28}
29#else
30static inline void mutex_set_owner(struct mutex *lock)
31{
32}
33
34static inline void mutex_clear_owner(struct mutex *lock)
35{
36}
37#endif
38
21#define debug_mutex_wake_waiter(lock, waiter) do { } while (0) 39#define debug_mutex_wake_waiter(lock, waiter) do { } while (0)
22#define debug_mutex_free_waiter(waiter) do { } while (0) 40#define debug_mutex_free_waiter(waiter) do { } while (0)
23#define debug_mutex_add_waiter(lock, waiter, ti) do { } while (0) 41#define debug_mutex_add_waiter(lock, waiter, ti) do { } while (0)
diff --git a/kernel/ns_cgroup.c b/kernel/ns_cgroup.c
index 78bc3fdac0d2..5aa854f9e5ae 100644
--- a/kernel/ns_cgroup.c
+++ b/kernel/ns_cgroup.c
@@ -34,7 +34,7 @@ int ns_cgroup_clone(struct task_struct *task, struct pid *pid)
34 34
35/* 35/*
36 * Rules: 36 * Rules:
37 * 1. you can only enter a cgroup which is a child of your current 37 * 1. you can only enter a cgroup which is a descendant of your current
38 * cgroup 38 * cgroup
39 * 2. you can only place another process into a cgroup if 39 * 2. you can only place another process into a cgroup if
40 * a. you have CAP_SYS_ADMIN 40 * a. you have CAP_SYS_ADMIN
@@ -45,21 +45,15 @@ int ns_cgroup_clone(struct task_struct *task, struct pid *pid)
45static int ns_can_attach(struct cgroup_subsys *ss, 45static int ns_can_attach(struct cgroup_subsys *ss,
46 struct cgroup *new_cgroup, struct task_struct *task) 46 struct cgroup *new_cgroup, struct task_struct *task)
47{ 47{
48 struct cgroup *orig;
49
50 if (current != task) { 48 if (current != task) {
51 if (!capable(CAP_SYS_ADMIN)) 49 if (!capable(CAP_SYS_ADMIN))
52 return -EPERM; 50 return -EPERM;
53 51
54 if (!cgroup_is_descendant(new_cgroup)) 52 if (!cgroup_is_descendant(new_cgroup, current))
55 return -EPERM; 53 return -EPERM;
56 } 54 }
57 55
58 if (atomic_read(&new_cgroup->count) != 0) 56 if (!cgroup_is_descendant(new_cgroup, task))
59 return -EPERM;
60
61 orig = task_cgroup(task, ns_subsys_id);
62 if (orig && orig != new_cgroup->parent)
63 return -EPERM; 57 return -EPERM;
64 58
65 return 0; 59 return 0;
@@ -77,7 +71,7 @@ static struct cgroup_subsys_state *ns_create(struct cgroup_subsys *ss,
77 71
78 if (!capable(CAP_SYS_ADMIN)) 72 if (!capable(CAP_SYS_ADMIN))
79 return ERR_PTR(-EPERM); 73 return ERR_PTR(-EPERM);
80 if (!cgroup_is_descendant(cgroup)) 74 if (!cgroup_is_descendant(cgroup, current))
81 return ERR_PTR(-EPERM); 75 return ERR_PTR(-EPERM);
82 76
83 ns_cgroup = kzalloc(sizeof(*ns_cgroup), GFP_KERNEL); 77 ns_cgroup = kzalloc(sizeof(*ns_cgroup), GFP_KERNEL);
diff --git a/kernel/panic.c b/kernel/panic.c
index 32fe4eff1b89..3fd8c5bf8b39 100644
--- a/kernel/panic.c
+++ b/kernel/panic.c
@@ -8,19 +8,19 @@
8 * This function is used through-out the kernel (including mm and fs) 8 * This function is used through-out the kernel (including mm and fs)
9 * to indicate a major problem. 9 * to indicate a major problem.
10 */ 10 */
11#include <linux/debug_locks.h>
12#include <linux/interrupt.h>
13#include <linux/kallsyms.h>
14#include <linux/notifier.h>
11#include <linux/module.h> 15#include <linux/module.h>
12#include <linux/sched.h> 16#include <linux/random.h>
13#include <linux/delay.h>
14#include <linux/reboot.h> 17#include <linux/reboot.h>
15#include <linux/notifier.h> 18#include <linux/delay.h>
16#include <linux/init.h> 19#include <linux/kexec.h>
20#include <linux/sched.h>
17#include <linux/sysrq.h> 21#include <linux/sysrq.h>
18#include <linux/interrupt.h> 22#include <linux/init.h>
19#include <linux/nmi.h> 23#include <linux/nmi.h>
20#include <linux/kexec.h>
21#include <linux/debug_locks.h>
22#include <linux/random.h>
23#include <linux/kallsyms.h>
24#include <linux/dmi.h> 24#include <linux/dmi.h>
25 25
26int panic_on_oops; 26int panic_on_oops;
@@ -52,19 +52,15 @@ EXPORT_SYMBOL(panic_blink);
52 * 52 *
53 * This function never returns. 53 * This function never returns.
54 */ 54 */
55
56NORET_TYPE void panic(const char * fmt, ...) 55NORET_TYPE void panic(const char * fmt, ...)
57{ 56{
58 long i;
59 static char buf[1024]; 57 static char buf[1024];
60 va_list args; 58 va_list args;
61#if defined(CONFIG_S390) 59 long i;
62 unsigned long caller = (unsigned long) __builtin_return_address(0);
63#endif
64 60
65 /* 61 /*
66 * It's possible to come here directly from a panic-assertion and not 62 * It's possible to come here directly from a panic-assertion and
67 * have preempt disabled. Some functions called from here want 63 * not have preempt disabled. Some functions called from here want
68 * preempt to be disabled. No point enabling it later though... 64 * preempt to be disabled. No point enabling it later though...
69 */ 65 */
70 preempt_disable(); 66 preempt_disable();
@@ -77,7 +73,6 @@ NORET_TYPE void panic(const char * fmt, ...)
77#ifdef CONFIG_DEBUG_BUGVERBOSE 73#ifdef CONFIG_DEBUG_BUGVERBOSE
78 dump_stack(); 74 dump_stack();
79#endif 75#endif
80 bust_spinlocks(0);
81 76
82 /* 77 /*
83 * If we have crashed and we have a crash kernel loaded let it handle 78 * If we have crashed and we have a crash kernel loaded let it handle
@@ -86,14 +81,12 @@ NORET_TYPE void panic(const char * fmt, ...)
86 */ 81 */
87 crash_kexec(NULL); 82 crash_kexec(NULL);
88 83
89#ifdef CONFIG_SMP
90 /* 84 /*
91 * Note smp_send_stop is the usual smp shutdown function, which 85 * Note smp_send_stop is the usual smp shutdown function, which
92 * unfortunately means it may not be hardened to work in a panic 86 * unfortunately means it may not be hardened to work in a panic
93 * situation. 87 * situation.
94 */ 88 */
95 smp_send_stop(); 89 smp_send_stop();
96#endif
97 90
98 atomic_notifier_call_chain(&panic_notifier_list, 0, buf); 91 atomic_notifier_call_chain(&panic_notifier_list, 0, buf);
99 92
@@ -102,19 +95,21 @@ NORET_TYPE void panic(const char * fmt, ...)
102 95
103 if (panic_timeout > 0) { 96 if (panic_timeout > 0) {
104 /* 97 /*
105 * Delay timeout seconds before rebooting the machine. 98 * Delay timeout seconds before rebooting the machine.
106 * We can't use the "normal" timers since we just panicked.. 99 * We can't use the "normal" timers since we just panicked.
107 */ 100 */
108 printk(KERN_EMERG "Rebooting in %d seconds..",panic_timeout); 101 printk(KERN_EMERG "Rebooting in %d seconds..", panic_timeout);
102
109 for (i = 0; i < panic_timeout*1000; ) { 103 for (i = 0; i < panic_timeout*1000; ) {
110 touch_nmi_watchdog(); 104 touch_nmi_watchdog();
111 i += panic_blink(i); 105 i += panic_blink(i);
112 mdelay(1); 106 mdelay(1);
113 i++; 107 i++;
114 } 108 }
115 /* This will not be a clean reboot, with everything 109 /*
116 * shutting down. But if there is a chance of 110 * This will not be a clean reboot, with everything
117 * rebooting the system it will be rebooted. 111 * shutting down. But if there is a chance of
112 * rebooting the system it will be rebooted.
118 */ 113 */
119 emergency_restart(); 114 emergency_restart();
120 } 115 }
@@ -127,38 +122,44 @@ NORET_TYPE void panic(const char * fmt, ...)
127 } 122 }
128#endif 123#endif
129#if defined(CONFIG_S390) 124#if defined(CONFIG_S390)
130 disabled_wait(caller); 125 {
126 unsigned long caller;
127
128 caller = (unsigned long)__builtin_return_address(0);
129 disabled_wait(caller);
130 }
131#endif 131#endif
132 local_irq_enable(); 132 local_irq_enable();
133 for (i = 0;;) { 133 for (i = 0; ; ) {
134 touch_softlockup_watchdog(); 134 touch_softlockup_watchdog();
135 i += panic_blink(i); 135 i += panic_blink(i);
136 mdelay(1); 136 mdelay(1);
137 i++; 137 i++;
138 } 138 }
139 bust_spinlocks(0);
139} 140}
140 141
141EXPORT_SYMBOL(panic); 142EXPORT_SYMBOL(panic);
142 143
143 144
144struct tnt { 145struct tnt {
145 u8 bit; 146 u8 bit;
146 char true; 147 char true;
147 char false; 148 char false;
148}; 149};
149 150
150static const struct tnt tnts[] = { 151static const struct tnt tnts[] = {
151 { TAINT_PROPRIETARY_MODULE, 'P', 'G' }, 152 { TAINT_PROPRIETARY_MODULE, 'P', 'G' },
152 { TAINT_FORCED_MODULE, 'F', ' ' }, 153 { TAINT_FORCED_MODULE, 'F', ' ' },
153 { TAINT_UNSAFE_SMP, 'S', ' ' }, 154 { TAINT_UNSAFE_SMP, 'S', ' ' },
154 { TAINT_FORCED_RMMOD, 'R', ' ' }, 155 { TAINT_FORCED_RMMOD, 'R', ' ' },
155 { TAINT_MACHINE_CHECK, 'M', ' ' }, 156 { TAINT_MACHINE_CHECK, 'M', ' ' },
156 { TAINT_BAD_PAGE, 'B', ' ' }, 157 { TAINT_BAD_PAGE, 'B', ' ' },
157 { TAINT_USER, 'U', ' ' }, 158 { TAINT_USER, 'U', ' ' },
158 { TAINT_DIE, 'D', ' ' }, 159 { TAINT_DIE, 'D', ' ' },
159 { TAINT_OVERRIDDEN_ACPI_TABLE, 'A', ' ' }, 160 { TAINT_OVERRIDDEN_ACPI_TABLE, 'A', ' ' },
160 { TAINT_WARN, 'W', ' ' }, 161 { TAINT_WARN, 'W', ' ' },
161 { TAINT_CRAP, 'C', ' ' }, 162 { TAINT_CRAP, 'C', ' ' },
162}; 163};
163 164
164/** 165/**
@@ -195,7 +196,8 @@ const char *print_tainted(void)
195 *s = 0; 196 *s = 0;
196 } else 197 } else
197 snprintf(buf, sizeof(buf), "Not tainted"); 198 snprintf(buf, sizeof(buf), "Not tainted");
198 return(buf); 199
200 return buf;
199} 201}
200 202
201int test_taint(unsigned flag) 203int test_taint(unsigned flag)
@@ -211,7 +213,8 @@ unsigned long get_taint(void)
211 213
212void add_taint(unsigned flag) 214void add_taint(unsigned flag)
213{ 215{
214 debug_locks = 0; /* can't trust the integrity of the kernel anymore */ 216 /* can't trust the integrity of the kernel anymore: */
217 debug_locks = 0;
215 set_bit(flag, &tainted_mask); 218 set_bit(flag, &tainted_mask);
216} 219}
217EXPORT_SYMBOL(add_taint); 220EXPORT_SYMBOL(add_taint);
@@ -266,8 +269,8 @@ static void do_oops_enter_exit(void)
266} 269}
267 270
268/* 271/*
269 * Return true if the calling CPU is allowed to print oops-related info. This 272 * Return true if the calling CPU is allowed to print oops-related info.
270 * is a bit racy.. 273 * This is a bit racy..
271 */ 274 */
272int oops_may_print(void) 275int oops_may_print(void)
273{ 276{
@@ -276,20 +279,22 @@ int oops_may_print(void)
276 279
277/* 280/*
278 * Called when the architecture enters its oops handler, before it prints 281 * Called when the architecture enters its oops handler, before it prints
279 * anything. If this is the first CPU to oops, and it's oopsing the first time 282 * anything. If this is the first CPU to oops, and it's oopsing the first
280 * then let it proceed. 283 * time then let it proceed.
281 * 284 *
282 * This is all enabled by the pause_on_oops kernel boot option. We do all this 285 * This is all enabled by the pause_on_oops kernel boot option. We do all
283 * to ensure that oopses don't scroll off the screen. It has the side-effect 286 * this to ensure that oopses don't scroll off the screen. It has the
284 * of preventing later-oopsing CPUs from mucking up the display, too. 287 * side-effect of preventing later-oopsing CPUs from mucking up the display,
288 * too.
285 * 289 *
286 * It turns out that the CPU which is allowed to print ends up pausing for the 290 * It turns out that the CPU which is allowed to print ends up pausing for
287 * right duration, whereas all the other CPUs pause for twice as long: once in 291 * the right duration, whereas all the other CPUs pause for twice as long:
288 * oops_enter(), once in oops_exit(). 292 * once in oops_enter(), once in oops_exit().
289 */ 293 */
290void oops_enter(void) 294void oops_enter(void)
291{ 295{
292 debug_locks_off(); /* can't trust the integrity of the kernel anymore */ 296 /* can't trust the integrity of the kernel anymore: */
297 debug_locks_off();
293 do_oops_enter_exit(); 298 do_oops_enter_exit();
294} 299}
295 300
diff --git a/kernel/params.c b/kernel/params.c
index a1e3025b19a9..de273ec85bd2 100644
--- a/kernel/params.c
+++ b/kernel/params.c
@@ -24,6 +24,9 @@
24#include <linux/err.h> 24#include <linux/err.h>
25#include <linux/slab.h> 25#include <linux/slab.h>
26 26
27/* We abuse the high bits of "perm" to record whether we kmalloc'ed. */
28#define KPARAM_KMALLOCED 0x80000000
29
27#if 0 30#if 0
28#define DEBUGP printk 31#define DEBUGP printk
29#else 32#else
@@ -217,7 +220,19 @@ int param_set_charp(const char *val, struct kernel_param *kp)
217 return -ENOSPC; 220 return -ENOSPC;
218 } 221 }
219 222
220 *(char **)kp->arg = (char *)val; 223 if (kp->perm & KPARAM_KMALLOCED)
224 kfree(*(char **)kp->arg);
225
226 /* This is a hack. We can't need to strdup in early boot, and we
227 * don't need to; this mangled commandline is preserved. */
228 if (slab_is_available()) {
229 kp->perm |= KPARAM_KMALLOCED;
230 *(char **)kp->arg = kstrdup(val, GFP_KERNEL);
231 if (!kp->arg)
232 return -ENOMEM;
233 } else
234 *(const char **)kp->arg = val;
235
221 return 0; 236 return 0;
222} 237}
223 238
@@ -571,6 +586,15 @@ void module_param_sysfs_remove(struct module *mod)
571} 586}
572#endif 587#endif
573 588
589void destroy_params(const struct kernel_param *params, unsigned num)
590{
591 unsigned int i;
592
593 for (i = 0; i < num; i++)
594 if (params[i].perm & KPARAM_KMALLOCED)
595 kfree(*(char **)params[i].arg);
596}
597
574static void __init kernel_add_sysfs_param(const char *name, 598static void __init kernel_add_sysfs_param(const char *name,
575 struct kernel_param *kparam, 599 struct kernel_param *kparam,
576 unsigned int name_skip) 600 unsigned int name_skip)
diff --git a/kernel/pid.c b/kernel/pid.c
index 1b3586fe753a..b2e5f78fd281 100644
--- a/kernel/pid.c
+++ b/kernel/pid.c
@@ -403,6 +403,8 @@ struct pid *get_task_pid(struct task_struct *task, enum pid_type type)
403{ 403{
404 struct pid *pid; 404 struct pid *pid;
405 rcu_read_lock(); 405 rcu_read_lock();
406 if (type != PIDTYPE_PID)
407 task = task->group_leader;
406 pid = get_pid(task->pids[type].pid); 408 pid = get_pid(task->pids[type].pid);
407 rcu_read_unlock(); 409 rcu_read_unlock();
408 return pid; 410 return pid;
@@ -450,11 +452,24 @@ pid_t pid_vnr(struct pid *pid)
450} 452}
451EXPORT_SYMBOL_GPL(pid_vnr); 453EXPORT_SYMBOL_GPL(pid_vnr);
452 454
453pid_t task_pid_nr_ns(struct task_struct *tsk, struct pid_namespace *ns) 455pid_t __task_pid_nr_ns(struct task_struct *task, enum pid_type type,
456 struct pid_namespace *ns)
454{ 457{
455 return pid_nr_ns(task_pid(tsk), ns); 458 pid_t nr = 0;
459
460 rcu_read_lock();
461 if (!ns)
462 ns = current->nsproxy->pid_ns;
463 if (likely(pid_alive(task))) {
464 if (type != PIDTYPE_PID)
465 task = task->group_leader;
466 nr = pid_nr_ns(task->pids[type].pid, ns);
467 }
468 rcu_read_unlock();
469
470 return nr;
456} 471}
457EXPORT_SYMBOL(task_pid_nr_ns); 472EXPORT_SYMBOL(__task_pid_nr_ns);
458 473
459pid_t task_tgid_nr_ns(struct task_struct *tsk, struct pid_namespace *ns) 474pid_t task_tgid_nr_ns(struct task_struct *tsk, struct pid_namespace *ns)
460{ 475{
@@ -462,18 +477,6 @@ pid_t task_tgid_nr_ns(struct task_struct *tsk, struct pid_namespace *ns)
462} 477}
463EXPORT_SYMBOL(task_tgid_nr_ns); 478EXPORT_SYMBOL(task_tgid_nr_ns);
464 479
465pid_t task_pgrp_nr_ns(struct task_struct *tsk, struct pid_namespace *ns)
466{
467 return pid_nr_ns(task_pgrp(tsk), ns);
468}
469EXPORT_SYMBOL(task_pgrp_nr_ns);
470
471pid_t task_session_nr_ns(struct task_struct *tsk, struct pid_namespace *ns)
472{
473 return pid_nr_ns(task_session(tsk), ns);
474}
475EXPORT_SYMBOL(task_session_nr_ns);
476
477struct pid_namespace *task_active_pid_ns(struct task_struct *tsk) 480struct pid_namespace *task_active_pid_ns(struct task_struct *tsk)
478{ 481{
479 return ns_of_pid(task_pid(tsk)); 482 return ns_of_pid(task_pid(tsk));
diff --git a/kernel/pid_namespace.c b/kernel/pid_namespace.c
index fab8ea86fac3..2d1001b4858d 100644
--- a/kernel/pid_namespace.c
+++ b/kernel/pid_namespace.c
@@ -152,6 +152,7 @@ void zap_pid_ns_processes(struct pid_namespace *pid_ns)
152{ 152{
153 int nr; 153 int nr;
154 int rc; 154 int rc;
155 struct task_struct *task;
155 156
156 /* 157 /*
157 * The last thread in the cgroup-init thread group is terminating. 158 * The last thread in the cgroup-init thread group is terminating.
@@ -169,7 +170,19 @@ void zap_pid_ns_processes(struct pid_namespace *pid_ns)
169 read_lock(&tasklist_lock); 170 read_lock(&tasklist_lock);
170 nr = next_pidmap(pid_ns, 1); 171 nr = next_pidmap(pid_ns, 1);
171 while (nr > 0) { 172 while (nr > 0) {
172 kill_proc_info(SIGKILL, SEND_SIG_PRIV, nr); 173 rcu_read_lock();
174
175 /*
176 * Use force_sig() since it clears SIGNAL_UNKILLABLE ensuring
177 * any nested-container's init processes don't ignore the
178 * signal
179 */
180 task = pid_task(find_vpid(nr), PIDTYPE_PID);
181 if (task)
182 force_sig(SIGKILL, task);
183
184 rcu_read_unlock();
185
173 nr = next_pidmap(pid_ns, nr); 186 nr = next_pidmap(pid_ns, nr);
174 } 187 }
175 read_unlock(&tasklist_lock); 188 read_unlock(&tasklist_lock);
diff --git a/kernel/posix-cpu-timers.c b/kernel/posix-cpu-timers.c
index e976e505648d..8e5d9a68b022 100644
--- a/kernel/posix-cpu-timers.c
+++ b/kernel/posix-cpu-timers.c
@@ -1370,7 +1370,8 @@ static inline int fastpath_timer_check(struct task_struct *tsk)
1370 if (task_cputime_expired(&group_sample, &sig->cputime_expires)) 1370 if (task_cputime_expired(&group_sample, &sig->cputime_expires))
1371 return 1; 1371 return 1;
1372 } 1372 }
1373 return 0; 1373
1374 return sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY;
1374} 1375}
1375 1376
1376/* 1377/*
diff --git a/kernel/power/disk.c b/kernel/power/disk.c
index 4a4a206b1979..5f21ab2bbcdf 100644
--- a/kernel/power/disk.c
+++ b/kernel/power/disk.c
@@ -22,6 +22,7 @@
22#include <linux/console.h> 22#include <linux/console.h>
23#include <linux/cpu.h> 23#include <linux/cpu.h>
24#include <linux/freezer.h> 24#include <linux/freezer.h>
25#include <asm/suspend.h>
25 26
26#include "power.h" 27#include "power.h"
27 28
@@ -214,7 +215,7 @@ static int create_image(int platform_mode)
214 return error; 215 return error;
215 216
216 device_pm_lock(); 217 device_pm_lock();
217 local_irq_disable(); 218
218 /* At this point, device_suspend() has been called, but *not* 219 /* At this point, device_suspend() has been called, but *not*
219 * device_power_down(). We *must* call device_power_down() now. 220 * device_power_down(). We *must* call device_power_down() now.
220 * Otherwise, drivers for some devices (e.g. interrupt controllers) 221 * Otherwise, drivers for some devices (e.g. interrupt controllers)
@@ -225,13 +226,25 @@ static int create_image(int platform_mode)
225 if (error) { 226 if (error) {
226 printk(KERN_ERR "PM: Some devices failed to power down, " 227 printk(KERN_ERR "PM: Some devices failed to power down, "
227 "aborting hibernation\n"); 228 "aborting hibernation\n");
228 goto Enable_irqs; 229 goto Unlock;
229 } 230 }
231
232 error = platform_pre_snapshot(platform_mode);
233 if (error || hibernation_test(TEST_PLATFORM))
234 goto Platform_finish;
235
236 error = disable_nonboot_cpus();
237 if (error || hibernation_test(TEST_CPUS)
238 || hibernation_testmode(HIBERNATION_TEST))
239 goto Enable_cpus;
240
241 local_irq_disable();
242
230 sysdev_suspend(PMSG_FREEZE); 243 sysdev_suspend(PMSG_FREEZE);
231 if (error) { 244 if (error) {
232 printk(KERN_ERR "PM: Some devices failed to power down, " 245 printk(KERN_ERR "PM: Some devices failed to power down, "
233 "aborting hibernation\n"); 246 "aborting hibernation\n");
234 goto Power_up_devices; 247 goto Enable_irqs;
235 } 248 }
236 249
237 if (hibernation_test(TEST_CORE)) 250 if (hibernation_test(TEST_CORE))
@@ -247,17 +260,28 @@ static int create_image(int platform_mode)
247 restore_processor_state(); 260 restore_processor_state();
248 if (!in_suspend) 261 if (!in_suspend)
249 platform_leave(platform_mode); 262 platform_leave(platform_mode);
263
250 Power_up: 264 Power_up:
251 sysdev_resume(); 265 sysdev_resume();
252 /* NOTE: device_power_up() is just a resume() for devices 266 /* NOTE: device_power_up() is just a resume() for devices
253 * that suspended with irqs off ... no overall powerup. 267 * that suspended with irqs off ... no overall powerup.
254 */ 268 */
255 Power_up_devices: 269
256 device_power_up(in_suspend ?
257 (error ? PMSG_RECOVER : PMSG_THAW) : PMSG_RESTORE);
258 Enable_irqs: 270 Enable_irqs:
259 local_irq_enable(); 271 local_irq_enable();
272
273 Enable_cpus:
274 enable_nonboot_cpus();
275
276 Platform_finish:
277 platform_finish(platform_mode);
278
279 device_power_up(in_suspend ?
280 (error ? PMSG_RECOVER : PMSG_THAW) : PMSG_RESTORE);
281
282 Unlock:
260 device_pm_unlock(); 283 device_pm_unlock();
284
261 return error; 285 return error;
262} 286}
263 287
@@ -265,7 +289,7 @@ static int create_image(int platform_mode)
265 * hibernation_snapshot - quiesce devices and create the hibernation 289 * hibernation_snapshot - quiesce devices and create the hibernation
266 * snapshot image. 290 * snapshot image.
267 * @platform_mode - if set, use the platform driver, if available, to 291 * @platform_mode - if set, use the platform driver, if available, to
268 * prepare the platform frimware for the power transition. 292 * prepare the platform firmware for the power transition.
269 * 293 *
270 * Must be called with pm_mutex held 294 * Must be called with pm_mutex held
271 */ 295 */
@@ -291,25 +315,9 @@ int hibernation_snapshot(int platform_mode)
291 if (hibernation_test(TEST_DEVICES)) 315 if (hibernation_test(TEST_DEVICES))
292 goto Recover_platform; 316 goto Recover_platform;
293 317
294 error = platform_pre_snapshot(platform_mode); 318 error = create_image(platform_mode);
295 if (error || hibernation_test(TEST_PLATFORM)) 319 /* Control returns here after successful restore */
296 goto Finish;
297
298 error = disable_nonboot_cpus();
299 if (!error) {
300 if (hibernation_test(TEST_CPUS))
301 goto Enable_cpus;
302
303 if (hibernation_testmode(HIBERNATION_TEST))
304 goto Enable_cpus;
305 320
306 error = create_image(platform_mode);
307 /* Control returns here after successful restore */
308 }
309 Enable_cpus:
310 enable_nonboot_cpus();
311 Finish:
312 platform_finish(platform_mode);
313 Resume_devices: 321 Resume_devices:
314 device_resume(in_suspend ? 322 device_resume(in_suspend ?
315 (error ? PMSG_RECOVER : PMSG_THAW) : PMSG_RESTORE); 323 (error ? PMSG_RECOVER : PMSG_THAW) : PMSG_RESTORE);
@@ -331,19 +339,33 @@ int hibernation_snapshot(int platform_mode)
331 * kernel. 339 * kernel.
332 */ 340 */
333 341
334static int resume_target_kernel(void) 342static int resume_target_kernel(bool platform_mode)
335{ 343{
336 int error; 344 int error;
337 345
338 device_pm_lock(); 346 device_pm_lock();
339 local_irq_disable(); 347
340 error = device_power_down(PMSG_QUIESCE); 348 error = device_power_down(PMSG_QUIESCE);
341 if (error) { 349 if (error) {
342 printk(KERN_ERR "PM: Some devices failed to power down, " 350 printk(KERN_ERR "PM: Some devices failed to power down, "
343 "aborting resume\n"); 351 "aborting resume\n");
344 goto Enable_irqs; 352 goto Unlock;
345 } 353 }
346 sysdev_suspend(PMSG_QUIESCE); 354
355 error = platform_pre_restore(platform_mode);
356 if (error)
357 goto Cleanup;
358
359 error = disable_nonboot_cpus();
360 if (error)
361 goto Enable_cpus;
362
363 local_irq_disable();
364
365 error = sysdev_suspend(PMSG_QUIESCE);
366 if (error)
367 goto Enable_irqs;
368
347 /* We'll ignore saved state, but this gets preempt count (etc) right */ 369 /* We'll ignore saved state, but this gets preempt count (etc) right */
348 save_processor_state(); 370 save_processor_state();
349 error = restore_highmem(); 371 error = restore_highmem();
@@ -366,11 +388,23 @@ static int resume_target_kernel(void)
366 swsusp_free(); 388 swsusp_free();
367 restore_processor_state(); 389 restore_processor_state();
368 touch_softlockup_watchdog(); 390 touch_softlockup_watchdog();
391
369 sysdev_resume(); 392 sysdev_resume();
370 device_power_up(PMSG_RECOVER); 393
371 Enable_irqs: 394 Enable_irqs:
372 local_irq_enable(); 395 local_irq_enable();
396
397 Enable_cpus:
398 enable_nonboot_cpus();
399
400 Cleanup:
401 platform_restore_cleanup(platform_mode);
402
403 device_power_up(PMSG_RECOVER);
404
405 Unlock:
373 device_pm_unlock(); 406 device_pm_unlock();
407
374 return error; 408 return error;
375} 409}
376 410
@@ -378,7 +412,7 @@ static int resume_target_kernel(void)
378 * hibernation_restore - quiesce devices and restore the hibernation 412 * hibernation_restore - quiesce devices and restore the hibernation
379 * snapshot image. If successful, control returns in hibernation_snaphot() 413 * snapshot image. If successful, control returns in hibernation_snaphot()
380 * @platform_mode - if set, use the platform driver, if available, to 414 * @platform_mode - if set, use the platform driver, if available, to
381 * prepare the platform frimware for the transition. 415 * prepare the platform firmware for the transition.
382 * 416 *
383 * Must be called with pm_mutex held 417 * Must be called with pm_mutex held
384 */ 418 */
@@ -390,19 +424,10 @@ int hibernation_restore(int platform_mode)
390 pm_prepare_console(); 424 pm_prepare_console();
391 suspend_console(); 425 suspend_console();
392 error = device_suspend(PMSG_QUIESCE); 426 error = device_suspend(PMSG_QUIESCE);
393 if (error)
394 goto Finish;
395
396 error = platform_pre_restore(platform_mode);
397 if (!error) { 427 if (!error) {
398 error = disable_nonboot_cpus(); 428 error = resume_target_kernel(platform_mode);
399 if (!error) 429 device_resume(PMSG_RECOVER);
400 error = resume_target_kernel();
401 enable_nonboot_cpus();
402 } 430 }
403 platform_restore_cleanup(platform_mode);
404 device_resume(PMSG_RECOVER);
405 Finish:
406 resume_console(); 431 resume_console();
407 pm_restore_console(); 432 pm_restore_console();
408 return error; 433 return error;
@@ -438,38 +463,46 @@ int hibernation_platform_enter(void)
438 goto Resume_devices; 463 goto Resume_devices;
439 } 464 }
440 465
466 device_pm_lock();
467
468 error = device_power_down(PMSG_HIBERNATE);
469 if (error)
470 goto Unlock;
471
441 error = hibernation_ops->prepare(); 472 error = hibernation_ops->prepare();
442 if (error) 473 if (error)
443 goto Resume_devices; 474 goto Platofrm_finish;
444 475
445 error = disable_nonboot_cpus(); 476 error = disable_nonboot_cpus();
446 if (error) 477 if (error)
447 goto Finish; 478 goto Platofrm_finish;
448 479
449 device_pm_lock();
450 local_irq_disable(); 480 local_irq_disable();
451 error = device_power_down(PMSG_HIBERNATE); 481 sysdev_suspend(PMSG_HIBERNATE);
452 if (!error) { 482 hibernation_ops->enter();
453 sysdev_suspend(PMSG_HIBERNATE); 483 /* We should never get here */
454 hibernation_ops->enter(); 484 while (1);
455 /* We should never get here */
456 while (1);
457 }
458 local_irq_enable();
459 device_pm_unlock();
460 485
461 /* 486 /*
462 * We don't need to reenable the nonboot CPUs or resume consoles, since 487 * We don't need to reenable the nonboot CPUs or resume consoles, since
463 * the system is going to be halted anyway. 488 * the system is going to be halted anyway.
464 */ 489 */
465 Finish: 490 Platofrm_finish:
466 hibernation_ops->finish(); 491 hibernation_ops->finish();
492
493 device_power_up(PMSG_RESTORE);
494
495 Unlock:
496 device_pm_unlock();
497
467 Resume_devices: 498 Resume_devices:
468 entering_platform_hibernation = false; 499 entering_platform_hibernation = false;
469 device_resume(PMSG_RESTORE); 500 device_resume(PMSG_RESTORE);
470 resume_console(); 501 resume_console();
502
471 Close: 503 Close:
472 hibernation_ops->end(); 504 hibernation_ops->end();
505
473 return error; 506 return error;
474} 507}
475 508
diff --git a/kernel/power/main.c b/kernel/power/main.c
index c9632f841f64..f172f41858bb 100644
--- a/kernel/power/main.c
+++ b/kernel/power/main.c
@@ -287,17 +287,32 @@ void __attribute__ ((weak)) arch_suspend_enable_irqs(void)
287 */ 287 */
288static int suspend_enter(suspend_state_t state) 288static int suspend_enter(suspend_state_t state)
289{ 289{
290 int error = 0; 290 int error;
291 291
292 device_pm_lock(); 292 device_pm_lock();
293 arch_suspend_disable_irqs();
294 BUG_ON(!irqs_disabled());
295 293
296 if ((error = device_power_down(PMSG_SUSPEND))) { 294 error = device_power_down(PMSG_SUSPEND);
295 if (error) {
297 printk(KERN_ERR "PM: Some devices failed to power down\n"); 296 printk(KERN_ERR "PM: Some devices failed to power down\n");
298 goto Done; 297 goto Done;
299 } 298 }
300 299
300 if (suspend_ops->prepare) {
301 error = suspend_ops->prepare();
302 if (error)
303 goto Power_up_devices;
304 }
305
306 if (suspend_test(TEST_PLATFORM))
307 goto Platfrom_finish;
308
309 error = disable_nonboot_cpus();
310 if (error || suspend_test(TEST_CPUS))
311 goto Enable_cpus;
312
313 arch_suspend_disable_irqs();
314 BUG_ON(!irqs_disabled());
315
301 error = sysdev_suspend(PMSG_SUSPEND); 316 error = sysdev_suspend(PMSG_SUSPEND);
302 if (!error) { 317 if (!error) {
303 if (!suspend_test(TEST_CORE)) 318 if (!suspend_test(TEST_CORE))
@@ -305,11 +320,22 @@ static int suspend_enter(suspend_state_t state)
305 sysdev_resume(); 320 sysdev_resume();
306 } 321 }
307 322
308 device_power_up(PMSG_RESUME);
309 Done:
310 arch_suspend_enable_irqs(); 323 arch_suspend_enable_irqs();
311 BUG_ON(irqs_disabled()); 324 BUG_ON(irqs_disabled());
325
326 Enable_cpus:
327 enable_nonboot_cpus();
328
329 Platfrom_finish:
330 if (suspend_ops->finish)
331 suspend_ops->finish();
332
333 Power_up_devices:
334 device_power_up(PMSG_RESUME);
335
336 Done:
312 device_pm_unlock(); 337 device_pm_unlock();
338
313 return error; 339 return error;
314} 340}
315 341
@@ -341,23 +367,8 @@ int suspend_devices_and_enter(suspend_state_t state)
341 if (suspend_test(TEST_DEVICES)) 367 if (suspend_test(TEST_DEVICES))
342 goto Recover_platform; 368 goto Recover_platform;
343 369
344 if (suspend_ops->prepare) { 370 suspend_enter(state);
345 error = suspend_ops->prepare();
346 if (error)
347 goto Resume_devices;
348 }
349
350 if (suspend_test(TEST_PLATFORM))
351 goto Finish;
352
353 error = disable_nonboot_cpus();
354 if (!error && !suspend_test(TEST_CPUS))
355 suspend_enter(state);
356 371
357 enable_nonboot_cpus();
358 Finish:
359 if (suspend_ops->finish)
360 suspend_ops->finish();
361 Resume_devices: 372 Resume_devices:
362 suspend_test_start(); 373 suspend_test_start();
363 device_resume(PMSG_RESUME); 374 device_resume(PMSG_RESUME);
diff --git a/kernel/power/snapshot.c b/kernel/power/snapshot.c
index f5fc2d7680f2..33e2e4a819f9 100644
--- a/kernel/power/snapshot.c
+++ b/kernel/power/snapshot.c
@@ -321,13 +321,10 @@ static int create_mem_extents(struct list_head *list, gfp_t gfp_mask)
321 321
322 INIT_LIST_HEAD(list); 322 INIT_LIST_HEAD(list);
323 323
324 for_each_zone(zone) { 324 for_each_populated_zone(zone) {
325 unsigned long zone_start, zone_end; 325 unsigned long zone_start, zone_end;
326 struct mem_extent *ext, *cur, *aux; 326 struct mem_extent *ext, *cur, *aux;
327 327
328 if (!populated_zone(zone))
329 continue;
330
331 zone_start = zone->zone_start_pfn; 328 zone_start = zone->zone_start_pfn;
332 zone_end = zone->zone_start_pfn + zone->spanned_pages; 329 zone_end = zone->zone_start_pfn + zone->spanned_pages;
333 330
@@ -804,8 +801,8 @@ static unsigned int count_free_highmem_pages(void)
804 struct zone *zone; 801 struct zone *zone;
805 unsigned int cnt = 0; 802 unsigned int cnt = 0;
806 803
807 for_each_zone(zone) 804 for_each_populated_zone(zone)
808 if (populated_zone(zone) && is_highmem(zone)) 805 if (is_highmem(zone))
809 cnt += zone_page_state(zone, NR_FREE_PAGES); 806 cnt += zone_page_state(zone, NR_FREE_PAGES);
810 807
811 return cnt; 808 return cnt;
diff --git a/kernel/power/swsusp.c b/kernel/power/swsusp.c
index a92c91451559..78c35047586d 100644
--- a/kernel/power/swsusp.c
+++ b/kernel/power/swsusp.c
@@ -51,6 +51,7 @@
51#include <linux/highmem.h> 51#include <linux/highmem.h>
52#include <linux/time.h> 52#include <linux/time.h>
53#include <linux/rbtree.h> 53#include <linux/rbtree.h>
54#include <linux/io.h>
54 55
55#include "power.h" 56#include "power.h"
56 57
@@ -229,17 +230,16 @@ int swsusp_shrink_memory(void)
229 size = count_data_pages() + PAGES_FOR_IO + SPARE_PAGES; 230 size = count_data_pages() + PAGES_FOR_IO + SPARE_PAGES;
230 tmp = size; 231 tmp = size;
231 size += highmem_size; 232 size += highmem_size;
232 for_each_zone (zone) 233 for_each_populated_zone(zone) {
233 if (populated_zone(zone)) { 234 tmp += snapshot_additional_pages(zone);
234 tmp += snapshot_additional_pages(zone); 235 if (is_highmem(zone)) {
235 if (is_highmem(zone)) { 236 highmem_size -=
236 highmem_size -=
237 zone_page_state(zone, NR_FREE_PAGES); 237 zone_page_state(zone, NR_FREE_PAGES);
238 } else { 238 } else {
239 tmp -= zone_page_state(zone, NR_FREE_PAGES); 239 tmp -= zone_page_state(zone, NR_FREE_PAGES);
240 tmp += zone->lowmem_reserve[ZONE_NORMAL]; 240 tmp += zone->lowmem_reserve[ZONE_NORMAL];
241 }
242 } 241 }
242 }
243 243
244 if (highmem_size < 0) 244 if (highmem_size < 0)
245 highmem_size = 0; 245 highmem_size = 0;
diff --git a/kernel/printk.c b/kernel/printk.c
index e3602d0755b0..5052b5497c67 100644
--- a/kernel/printk.c
+++ b/kernel/printk.c
@@ -32,6 +32,7 @@
32#include <linux/security.h> 32#include <linux/security.h>
33#include <linux/bootmem.h> 33#include <linux/bootmem.h>
34#include <linux/syscalls.h> 34#include <linux/syscalls.h>
35#include <linux/kexec.h>
35 36
36#include <asm/uaccess.h> 37#include <asm/uaccess.h>
37 38
@@ -135,6 +136,24 @@ static char *log_buf = __log_buf;
135static int log_buf_len = __LOG_BUF_LEN; 136static int log_buf_len = __LOG_BUF_LEN;
136static unsigned logged_chars; /* Number of chars produced since last read+clear operation */ 137static unsigned logged_chars; /* Number of chars produced since last read+clear operation */
137 138
139#ifdef CONFIG_KEXEC
140/*
141 * This appends the listed symbols to /proc/vmcoreinfo
142 *
143 * /proc/vmcoreinfo is used by various utiilties, like crash and makedumpfile to
144 * obtain access to symbols that are otherwise very difficult to locate. These
145 * symbols are specifically used so that utilities can access and extract the
146 * dmesg log from a vmcore file after a crash.
147 */
148void log_buf_kexec_setup(void)
149{
150 VMCOREINFO_SYMBOL(log_buf);
151 VMCOREINFO_SYMBOL(log_end);
152 VMCOREINFO_SYMBOL(log_buf_len);
153 VMCOREINFO_SYMBOL(logged_chars);
154}
155#endif
156
138static int __init log_buf_len_setup(char *str) 157static int __init log_buf_len_setup(char *str)
139{ 158{
140 unsigned size = memparse(str, &str); 159 unsigned size = memparse(str, &str);
@@ -1292,8 +1311,11 @@ EXPORT_SYMBOL(printk_ratelimit);
1292bool printk_timed_ratelimit(unsigned long *caller_jiffies, 1311bool printk_timed_ratelimit(unsigned long *caller_jiffies,
1293 unsigned int interval_msecs) 1312 unsigned int interval_msecs)
1294{ 1313{
1295 if (*caller_jiffies == 0 || time_after(jiffies, *caller_jiffies)) { 1314 if (*caller_jiffies == 0
1296 *caller_jiffies = jiffies + msecs_to_jiffies(interval_msecs); 1315 || !time_in_range(jiffies, *caller_jiffies,
1316 *caller_jiffies
1317 + msecs_to_jiffies(interval_msecs))) {
1318 *caller_jiffies = jiffies;
1297 return true; 1319 return true;
1298 } 1320 }
1299 return false; 1321 return false;
diff --git a/kernel/ptrace.c b/kernel/ptrace.c
index c9cf48b21f05..aaad0ec34194 100644
--- a/kernel/ptrace.c
+++ b/kernel/ptrace.c
@@ -60,11 +60,15 @@ static void ptrace_untrace(struct task_struct *child)
60{ 60{
61 spin_lock(&child->sighand->siglock); 61 spin_lock(&child->sighand->siglock);
62 if (task_is_traced(child)) { 62 if (task_is_traced(child)) {
63 if (child->signal->flags & SIGNAL_STOP_STOPPED) { 63 /*
64 * If the group stop is completed or in progress,
65 * this thread was already counted as stopped.
66 */
67 if (child->signal->flags & SIGNAL_STOP_STOPPED ||
68 child->signal->group_stop_count)
64 __set_task_state(child, TASK_STOPPED); 69 __set_task_state(child, TASK_STOPPED);
65 } else { 70 else
66 signal_wake_up(child, 1); 71 signal_wake_up(child, 1);
67 }
68 } 72 }
69 spin_unlock(&child->sighand->siglock); 73 spin_unlock(&child->sighand->siglock);
70} 74}
@@ -235,18 +239,58 @@ out:
235 return retval; 239 return retval;
236} 240}
237 241
238static inline void __ptrace_detach(struct task_struct *child, unsigned int data) 242/*
243 * Called with irqs disabled, returns true if childs should reap themselves.
244 */
245static int ignoring_children(struct sighand_struct *sigh)
239{ 246{
240 child->exit_code = data; 247 int ret;
241 /* .. re-parent .. */ 248 spin_lock(&sigh->siglock);
242 __ptrace_unlink(child); 249 ret = (sigh->action[SIGCHLD-1].sa.sa_handler == SIG_IGN) ||
243 /* .. and wake it up. */ 250 (sigh->action[SIGCHLD-1].sa.sa_flags & SA_NOCLDWAIT);
244 if (child->exit_state != EXIT_ZOMBIE) 251 spin_unlock(&sigh->siglock);
245 wake_up_process(child); 252 return ret;
253}
254
255/*
256 * Called with tasklist_lock held for writing.
257 * Unlink a traced task, and clean it up if it was a traced zombie.
258 * Return true if it needs to be reaped with release_task().
259 * (We can't call release_task() here because we already hold tasklist_lock.)
260 *
261 * If it's a zombie, our attachedness prevented normal parent notification
262 * or self-reaping. Do notification now if it would have happened earlier.
263 * If it should reap itself, return true.
264 *
265 * If it's our own child, there is no notification to do.
266 * But if our normal children self-reap, then this child
267 * was prevented by ptrace and we must reap it now.
268 */
269static bool __ptrace_detach(struct task_struct *tracer, struct task_struct *p)
270{
271 __ptrace_unlink(p);
272
273 if (p->exit_state == EXIT_ZOMBIE) {
274 if (!task_detached(p) && thread_group_empty(p)) {
275 if (!same_thread_group(p->real_parent, tracer))
276 do_notify_parent(p, p->exit_signal);
277 else if (ignoring_children(tracer->sighand))
278 p->exit_signal = -1;
279 }
280 if (task_detached(p)) {
281 /* Mark it as in the process of being reaped. */
282 p->exit_state = EXIT_DEAD;
283 return true;
284 }
285 }
286
287 return false;
246} 288}
247 289
248int ptrace_detach(struct task_struct *child, unsigned int data) 290int ptrace_detach(struct task_struct *child, unsigned int data)
249{ 291{
292 bool dead = false;
293
250 if (!valid_signal(data)) 294 if (!valid_signal(data))
251 return -EIO; 295 return -EIO;
252 296
@@ -255,14 +299,45 @@ int ptrace_detach(struct task_struct *child, unsigned int data)
255 clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE); 299 clear_tsk_thread_flag(child, TIF_SYSCALL_TRACE);
256 300
257 write_lock_irq(&tasklist_lock); 301 write_lock_irq(&tasklist_lock);
258 /* protect against de_thread()->release_task() */ 302 /*
259 if (child->ptrace) 303 * This child can be already killed. Make sure de_thread() or
260 __ptrace_detach(child, data); 304 * our sub-thread doing do_wait() didn't do release_task() yet.
305 */
306 if (child->ptrace) {
307 child->exit_code = data;
308 dead = __ptrace_detach(current, child);
309 }
261 write_unlock_irq(&tasklist_lock); 310 write_unlock_irq(&tasklist_lock);
262 311
312 if (unlikely(dead))
313 release_task(child);
314
263 return 0; 315 return 0;
264} 316}
265 317
318/*
319 * Detach all tasks we were using ptrace on.
320 */
321void exit_ptrace(struct task_struct *tracer)
322{
323 struct task_struct *p, *n;
324 LIST_HEAD(ptrace_dead);
325
326 write_lock_irq(&tasklist_lock);
327 list_for_each_entry_safe(p, n, &tracer->ptraced, ptrace_entry) {
328 if (__ptrace_detach(tracer, p))
329 list_add(&p->ptrace_entry, &ptrace_dead);
330 }
331 write_unlock_irq(&tasklist_lock);
332
333 BUG_ON(!list_empty(&tracer->ptraced));
334
335 list_for_each_entry_safe(p, n, &ptrace_dead, ptrace_entry) {
336 list_del_init(&p->ptrace_entry);
337 release_task(p);
338 }
339}
340
266int ptrace_readdata(struct task_struct *tsk, unsigned long src, char __user *dst, int len) 341int ptrace_readdata(struct task_struct *tsk, unsigned long src, char __user *dst, int len)
267{ 342{
268 int copied = 0; 343 int copied = 0;
@@ -612,8 +687,6 @@ SYSCALL_DEFINE4(ptrace, long, request, long, pid, long, addr, long, data)
612 goto out_put_task_struct; 687 goto out_put_task_struct;
613 688
614 ret = arch_ptrace(child, request, addr, data); 689 ret = arch_ptrace(child, request, addr, data);
615 if (ret < 0)
616 goto out_put_task_struct;
617 690
618 out_put_task_struct: 691 out_put_task_struct:
619 put_task_struct(child); 692 put_task_struct(child);
diff --git a/kernel/rcupdate.c b/kernel/rcupdate.c
index cae8a059cf47..2c7b8457d0d2 100644
--- a/kernel/rcupdate.c
+++ b/kernel/rcupdate.c
@@ -122,6 +122,8 @@ static void rcu_barrier_func(void *type)
122 } 122 }
123} 123}
124 124
125static inline void wait_migrated_callbacks(void);
126
125/* 127/*
126 * Orchestrate the specified type of RCU barrier, waiting for all 128 * Orchestrate the specified type of RCU barrier, waiting for all
127 * RCU callbacks of the specified type to complete. 129 * RCU callbacks of the specified type to complete.
@@ -147,6 +149,7 @@ static void _rcu_barrier(enum rcu_barrier type)
147 complete(&rcu_barrier_completion); 149 complete(&rcu_barrier_completion);
148 wait_for_completion(&rcu_barrier_completion); 150 wait_for_completion(&rcu_barrier_completion);
149 mutex_unlock(&rcu_barrier_mutex); 151 mutex_unlock(&rcu_barrier_mutex);
152 wait_migrated_callbacks();
150} 153}
151 154
152/** 155/**
@@ -176,9 +179,50 @@ void rcu_barrier_sched(void)
176} 179}
177EXPORT_SYMBOL_GPL(rcu_barrier_sched); 180EXPORT_SYMBOL_GPL(rcu_barrier_sched);
178 181
182static atomic_t rcu_migrate_type_count = ATOMIC_INIT(0);
183static struct rcu_head rcu_migrate_head[3];
184static DECLARE_WAIT_QUEUE_HEAD(rcu_migrate_wq);
185
186static void rcu_migrate_callback(struct rcu_head *notused)
187{
188 if (atomic_dec_and_test(&rcu_migrate_type_count))
189 wake_up(&rcu_migrate_wq);
190}
191
192static inline void wait_migrated_callbacks(void)
193{
194 wait_event(rcu_migrate_wq, !atomic_read(&rcu_migrate_type_count));
195}
196
197static int __cpuinit rcu_barrier_cpu_hotplug(struct notifier_block *self,
198 unsigned long action, void *hcpu)
199{
200 if (action == CPU_DYING) {
201 /*
202 * preempt_disable() in on_each_cpu() prevents stop_machine(),
203 * so when "on_each_cpu(rcu_barrier_func, (void *)type, 1);"
204 * returns, all online cpus have queued rcu_barrier_func(),
205 * and the dead cpu(if it exist) queues rcu_migrate_callback()s.
206 *
207 * These callbacks ensure _rcu_barrier() waits for all
208 * RCU callbacks of the specified type to complete.
209 */
210 atomic_set(&rcu_migrate_type_count, 3);
211 call_rcu_bh(rcu_migrate_head, rcu_migrate_callback);
212 call_rcu_sched(rcu_migrate_head + 1, rcu_migrate_callback);
213 call_rcu(rcu_migrate_head + 2, rcu_migrate_callback);
214 } else if (action == CPU_POST_DEAD) {
215 /* rcu_migrate_head is protected by cpu_add_remove_lock */
216 wait_migrated_callbacks();
217 }
218
219 return NOTIFY_OK;
220}
221
179void __init rcu_init(void) 222void __init rcu_init(void)
180{ 223{
181 __rcu_init(); 224 __rcu_init();
225 hotcpu_notifier(rcu_barrier_cpu_hotplug, 0);
182} 226}
183 227
184void rcu_scheduler_starting(void) 228void rcu_scheduler_starting(void)
diff --git a/kernel/rcutorture.c b/kernel/rcutorture.c
index 7c4142a79f0a..9b4a975a4b4a 100644
--- a/kernel/rcutorture.c
+++ b/kernel/rcutorture.c
@@ -126,6 +126,7 @@ static atomic_t n_rcu_torture_mberror;
126static atomic_t n_rcu_torture_error; 126static atomic_t n_rcu_torture_error;
127static long n_rcu_torture_timers = 0; 127static long n_rcu_torture_timers = 0;
128static struct list_head rcu_torture_removed; 128static struct list_head rcu_torture_removed;
129static cpumask_var_t shuffle_tmp_mask;
129 130
130static int stutter_pause_test = 0; 131static int stutter_pause_test = 0;
131 132
@@ -889,10 +890,9 @@ static int rcu_idle_cpu; /* Force all torture tasks off this CPU */
889 */ 890 */
890static void rcu_torture_shuffle_tasks(void) 891static void rcu_torture_shuffle_tasks(void)
891{ 892{
892 cpumask_t tmp_mask;
893 int i; 893 int i;
894 894
895 cpus_setall(tmp_mask); 895 cpumask_setall(shuffle_tmp_mask);
896 get_online_cpus(); 896 get_online_cpus();
897 897
898 /* No point in shuffling if there is only one online CPU (ex: UP) */ 898 /* No point in shuffling if there is only one online CPU (ex: UP) */
@@ -902,29 +902,29 @@ static void rcu_torture_shuffle_tasks(void)
902 } 902 }
903 903
904 if (rcu_idle_cpu != -1) 904 if (rcu_idle_cpu != -1)
905 cpu_clear(rcu_idle_cpu, tmp_mask); 905 cpumask_clear_cpu(rcu_idle_cpu, shuffle_tmp_mask);
906 906
907 set_cpus_allowed_ptr(current, &tmp_mask); 907 set_cpus_allowed_ptr(current, shuffle_tmp_mask);
908 908
909 if (reader_tasks) { 909 if (reader_tasks) {
910 for (i = 0; i < nrealreaders; i++) 910 for (i = 0; i < nrealreaders; i++)
911 if (reader_tasks[i]) 911 if (reader_tasks[i])
912 set_cpus_allowed_ptr(reader_tasks[i], 912 set_cpus_allowed_ptr(reader_tasks[i],
913 &tmp_mask); 913 shuffle_tmp_mask);
914 } 914 }
915 915
916 if (fakewriter_tasks) { 916 if (fakewriter_tasks) {
917 for (i = 0; i < nfakewriters; i++) 917 for (i = 0; i < nfakewriters; i++)
918 if (fakewriter_tasks[i]) 918 if (fakewriter_tasks[i])
919 set_cpus_allowed_ptr(fakewriter_tasks[i], 919 set_cpus_allowed_ptr(fakewriter_tasks[i],
920 &tmp_mask); 920 shuffle_tmp_mask);
921 } 921 }
922 922
923 if (writer_task) 923 if (writer_task)
924 set_cpus_allowed_ptr(writer_task, &tmp_mask); 924 set_cpus_allowed_ptr(writer_task, shuffle_tmp_mask);
925 925
926 if (stats_task) 926 if (stats_task)
927 set_cpus_allowed_ptr(stats_task, &tmp_mask); 927 set_cpus_allowed_ptr(stats_task, shuffle_tmp_mask);
928 928
929 if (rcu_idle_cpu == -1) 929 if (rcu_idle_cpu == -1)
930 rcu_idle_cpu = num_online_cpus() - 1; 930 rcu_idle_cpu = num_online_cpus() - 1;
@@ -1012,6 +1012,7 @@ rcu_torture_cleanup(void)
1012 if (shuffler_task) { 1012 if (shuffler_task) {
1013 VERBOSE_PRINTK_STRING("Stopping rcu_torture_shuffle task"); 1013 VERBOSE_PRINTK_STRING("Stopping rcu_torture_shuffle task");
1014 kthread_stop(shuffler_task); 1014 kthread_stop(shuffler_task);
1015 free_cpumask_var(shuffle_tmp_mask);
1015 } 1016 }
1016 shuffler_task = NULL; 1017 shuffler_task = NULL;
1017 1018
@@ -1190,10 +1191,18 @@ rcu_torture_init(void)
1190 } 1191 }
1191 if (test_no_idle_hz) { 1192 if (test_no_idle_hz) {
1192 rcu_idle_cpu = num_online_cpus() - 1; 1193 rcu_idle_cpu = num_online_cpus() - 1;
1194
1195 if (!alloc_cpumask_var(&shuffle_tmp_mask, GFP_KERNEL)) {
1196 firsterr = -ENOMEM;
1197 VERBOSE_PRINTK_ERRSTRING("Failed to alloc mask");
1198 goto unwind;
1199 }
1200
1193 /* Create the shuffler thread */ 1201 /* Create the shuffler thread */
1194 shuffler_task = kthread_run(rcu_torture_shuffle, NULL, 1202 shuffler_task = kthread_run(rcu_torture_shuffle, NULL,
1195 "rcu_torture_shuffle"); 1203 "rcu_torture_shuffle");
1196 if (IS_ERR(shuffler_task)) { 1204 if (IS_ERR(shuffler_task)) {
1205 free_cpumask_var(shuffle_tmp_mask);
1197 firsterr = PTR_ERR(shuffler_task); 1206 firsterr = PTR_ERR(shuffler_task);
1198 VERBOSE_PRINTK_ERRSTRING("Failed to create shuffler"); 1207 VERBOSE_PRINTK_ERRSTRING("Failed to create shuffler");
1199 shuffler_task = NULL; 1208 shuffler_task = NULL;
diff --git a/kernel/relay.c b/kernel/relay.c
index 9d79b7854fa6..bc188549788f 100644
--- a/kernel/relay.c
+++ b/kernel/relay.c
@@ -677,9 +677,7 @@ int relay_late_setup_files(struct rchan *chan,
677 */ 677 */
678 for_each_online_cpu(i) { 678 for_each_online_cpu(i) {
679 if (unlikely(!chan->buf[i])) { 679 if (unlikely(!chan->buf[i])) {
680 printk(KERN_ERR "relay_late_setup_files: CPU %u " 680 WARN_ONCE(1, KERN_ERR "CPU has no buffer!\n");
681 "has no buffer, it must have!\n", i);
682 BUG();
683 err = -EINVAL; 681 err = -EINVAL;
684 break; 682 break;
685 } 683 }
@@ -750,7 +748,7 @@ size_t relay_switch_subbuf(struct rchan_buf *buf, size_t length)
750 * from the scheduler (trying to re-grab 748 * from the scheduler (trying to re-grab
751 * rq->lock), so defer it. 749 * rq->lock), so defer it.
752 */ 750 */
753 __mod_timer(&buf->timer, jiffies + 1); 751 mod_timer(&buf->timer, jiffies + 1);
754 } 752 }
755 753
756 old = buf->data; 754 old = buf->data;
@@ -797,13 +795,15 @@ void relay_subbufs_consumed(struct rchan *chan,
797 if (!chan) 795 if (!chan)
798 return; 796 return;
799 797
800 if (cpu >= NR_CPUS || !chan->buf[cpu]) 798 if (cpu >= NR_CPUS || !chan->buf[cpu] ||
799 subbufs_consumed > chan->n_subbufs)
801 return; 800 return;
802 801
803 buf = chan->buf[cpu]; 802 buf = chan->buf[cpu];
804 buf->subbufs_consumed += subbufs_consumed; 803 if (subbufs_consumed > buf->subbufs_produced - buf->subbufs_consumed)
805 if (buf->subbufs_consumed > buf->subbufs_produced)
806 buf->subbufs_consumed = buf->subbufs_produced; 804 buf->subbufs_consumed = buf->subbufs_produced;
805 else
806 buf->subbufs_consumed += subbufs_consumed;
807} 807}
808EXPORT_SYMBOL_GPL(relay_subbufs_consumed); 808EXPORT_SYMBOL_GPL(relay_subbufs_consumed);
809 809
diff --git a/kernel/sched.c b/kernel/sched.c
index 78f4424b7c43..39e708602169 100644
--- a/kernel/sched.c
+++ b/kernel/sched.c
@@ -331,6 +331,13 @@ static DEFINE_PER_CPU(struct rt_rq, init_rt_rq) ____cacheline_aligned_in_smp;
331 */ 331 */
332static DEFINE_SPINLOCK(task_group_lock); 332static DEFINE_SPINLOCK(task_group_lock);
333 333
334#ifdef CONFIG_SMP
335static int root_task_group_empty(void)
336{
337 return list_empty(&root_task_group.children);
338}
339#endif
340
334#ifdef CONFIG_FAIR_GROUP_SCHED 341#ifdef CONFIG_FAIR_GROUP_SCHED
335#ifdef CONFIG_USER_SCHED 342#ifdef CONFIG_USER_SCHED
336# define INIT_TASK_GROUP_LOAD (2*NICE_0_LOAD) 343# define INIT_TASK_GROUP_LOAD (2*NICE_0_LOAD)
@@ -391,6 +398,13 @@ static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
391 398
392#else 399#else
393 400
401#ifdef CONFIG_SMP
402static int root_task_group_empty(void)
403{
404 return 1;
405}
406#endif
407
394static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { } 408static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
395static inline struct task_group *task_group(struct task_struct *p) 409static inline struct task_group *task_group(struct task_struct *p)
396{ 410{
@@ -467,11 +481,17 @@ struct rt_rq {
467 struct rt_prio_array active; 481 struct rt_prio_array active;
468 unsigned long rt_nr_running; 482 unsigned long rt_nr_running;
469#if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED 483#if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
470 int highest_prio; /* highest queued rt task prio */ 484 struct {
485 int curr; /* highest queued rt task prio */
486#ifdef CONFIG_SMP
487 int next; /* next highest */
488#endif
489 } highest_prio;
471#endif 490#endif
472#ifdef CONFIG_SMP 491#ifdef CONFIG_SMP
473 unsigned long rt_nr_migratory; 492 unsigned long rt_nr_migratory;
474 int overloaded; 493 int overloaded;
494 struct plist_head pushable_tasks;
475#endif 495#endif
476 int rt_throttled; 496 int rt_throttled;
477 u64 rt_time; 497 u64 rt_time;
@@ -549,7 +569,6 @@ struct rq {
549 unsigned long nr_running; 569 unsigned long nr_running;
550 #define CPU_LOAD_IDX_MAX 5 570 #define CPU_LOAD_IDX_MAX 5
551 unsigned long cpu_load[CPU_LOAD_IDX_MAX]; 571 unsigned long cpu_load[CPU_LOAD_IDX_MAX];
552 unsigned char idle_at_tick;
553#ifdef CONFIG_NO_HZ 572#ifdef CONFIG_NO_HZ
554 unsigned long last_tick_seen; 573 unsigned long last_tick_seen;
555 unsigned char in_nohz_recently; 574 unsigned char in_nohz_recently;
@@ -591,6 +610,7 @@ struct rq {
591 struct root_domain *rd; 610 struct root_domain *rd;
592 struct sched_domain *sd; 611 struct sched_domain *sd;
593 612
613 unsigned char idle_at_tick;
594 /* For active balancing */ 614 /* For active balancing */
595 int active_balance; 615 int active_balance;
596 int push_cpu; 616 int push_cpu;
@@ -619,9 +639,6 @@ struct rq {
619 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */ 639 /* could above be rq->cfs_rq.exec_clock + rq->rt_rq.rt_runtime ? */
620 640
621 /* sys_sched_yield() stats */ 641 /* sys_sched_yield() stats */
622 unsigned int yld_exp_empty;
623 unsigned int yld_act_empty;
624 unsigned int yld_both_empty;
625 unsigned int yld_count; 642 unsigned int yld_count;
626 643
627 /* schedule() stats */ 644 /* schedule() stats */
@@ -1114,7 +1131,7 @@ static void hrtick_start(struct rq *rq, u64 delay)
1114 if (rq == this_rq()) { 1131 if (rq == this_rq()) {
1115 hrtimer_restart(timer); 1132 hrtimer_restart(timer);
1116 } else if (!rq->hrtick_csd_pending) { 1133 } else if (!rq->hrtick_csd_pending) {
1117 __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd); 1134 __smp_call_function_single(cpu_of(rq), &rq->hrtick_csd, 0);
1118 rq->hrtick_csd_pending = 1; 1135 rq->hrtick_csd_pending = 1;
1119 } 1136 }
1120} 1137}
@@ -1204,10 +1221,10 @@ static void resched_task(struct task_struct *p)
1204 1221
1205 assert_spin_locked(&task_rq(p)->lock); 1222 assert_spin_locked(&task_rq(p)->lock);
1206 1223
1207 if (unlikely(test_tsk_thread_flag(p, TIF_NEED_RESCHED))) 1224 if (test_tsk_need_resched(p))
1208 return; 1225 return;
1209 1226
1210 set_tsk_thread_flag(p, TIF_NEED_RESCHED); 1227 set_tsk_need_resched(p);
1211 1228
1212 cpu = task_cpu(p); 1229 cpu = task_cpu(p);
1213 if (cpu == smp_processor_id()) 1230 if (cpu == smp_processor_id())
@@ -1263,7 +1280,7 @@ void wake_up_idle_cpu(int cpu)
1263 * lockless. The worst case is that the other CPU runs the 1280 * lockless. The worst case is that the other CPU runs the
1264 * idle task through an additional NOOP schedule() 1281 * idle task through an additional NOOP schedule()
1265 */ 1282 */
1266 set_tsk_thread_flag(rq->idle, TIF_NEED_RESCHED); 1283 set_tsk_need_resched(rq->idle);
1267 1284
1268 /* NEED_RESCHED must be visible before we test polling */ 1285 /* NEED_RESCHED must be visible before we test polling */
1269 smp_mb(); 1286 smp_mb();
@@ -1631,21 +1648,42 @@ static inline void update_shares_locked(struct rq *rq, struct sched_domain *sd)
1631 1648
1632#endif 1649#endif
1633 1650
1651#ifdef CONFIG_PREEMPT
1652
1634/* 1653/*
1635 * double_lock_balance - lock the busiest runqueue, this_rq is locked already. 1654 * fair double_lock_balance: Safely acquires both rq->locks in a fair
1655 * way at the expense of forcing extra atomic operations in all
1656 * invocations. This assures that the double_lock is acquired using the
1657 * same underlying policy as the spinlock_t on this architecture, which
1658 * reduces latency compared to the unfair variant below. However, it
1659 * also adds more overhead and therefore may reduce throughput.
1636 */ 1660 */
1637static int double_lock_balance(struct rq *this_rq, struct rq *busiest) 1661static inline int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1662 __releases(this_rq->lock)
1663 __acquires(busiest->lock)
1664 __acquires(this_rq->lock)
1665{
1666 spin_unlock(&this_rq->lock);
1667 double_rq_lock(this_rq, busiest);
1668
1669 return 1;
1670}
1671
1672#else
1673/*
1674 * Unfair double_lock_balance: Optimizes throughput at the expense of
1675 * latency by eliminating extra atomic operations when the locks are
1676 * already in proper order on entry. This favors lower cpu-ids and will
1677 * grant the double lock to lower cpus over higher ids under contention,
1678 * regardless of entry order into the function.
1679 */
1680static int _double_lock_balance(struct rq *this_rq, struct rq *busiest)
1638 __releases(this_rq->lock) 1681 __releases(this_rq->lock)
1639 __acquires(busiest->lock) 1682 __acquires(busiest->lock)
1640 __acquires(this_rq->lock) 1683 __acquires(this_rq->lock)
1641{ 1684{
1642 int ret = 0; 1685 int ret = 0;
1643 1686
1644 if (unlikely(!irqs_disabled())) {
1645 /* printk() doesn't work good under rq->lock */
1646 spin_unlock(&this_rq->lock);
1647 BUG_ON(1);
1648 }
1649 if (unlikely(!spin_trylock(&busiest->lock))) { 1687 if (unlikely(!spin_trylock(&busiest->lock))) {
1650 if (busiest < this_rq) { 1688 if (busiest < this_rq) {
1651 spin_unlock(&this_rq->lock); 1689 spin_unlock(&this_rq->lock);
@@ -1658,6 +1696,22 @@ static int double_lock_balance(struct rq *this_rq, struct rq *busiest)
1658 return ret; 1696 return ret;
1659} 1697}
1660 1698
1699#endif /* CONFIG_PREEMPT */
1700
1701/*
1702 * double_lock_balance - lock the busiest runqueue, this_rq is locked already.
1703 */
1704static int double_lock_balance(struct rq *this_rq, struct rq *busiest)
1705{
1706 if (unlikely(!irqs_disabled())) {
1707 /* printk() doesn't work good under rq->lock */
1708 spin_unlock(&this_rq->lock);
1709 BUG_ON(1);
1710 }
1711
1712 return _double_lock_balance(this_rq, busiest);
1713}
1714
1661static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest) 1715static inline void double_unlock_balance(struct rq *this_rq, struct rq *busiest)
1662 __releases(busiest->lock) 1716 __releases(busiest->lock)
1663{ 1717{
@@ -1726,6 +1780,9 @@ static void update_avg(u64 *avg, u64 sample)
1726 1780
1727static void enqueue_task(struct rq *rq, struct task_struct *p, int wakeup) 1781static void enqueue_task(struct rq *rq, struct task_struct *p, int wakeup)
1728{ 1782{
1783 if (wakeup)
1784 p->se.start_runtime = p->se.sum_exec_runtime;
1785
1729 sched_info_queued(p); 1786 sched_info_queued(p);
1730 p->sched_class->enqueue_task(rq, p, wakeup); 1787 p->sched_class->enqueue_task(rq, p, wakeup);
1731 p->se.on_rq = 1; 1788 p->se.on_rq = 1;
@@ -1733,10 +1790,15 @@ static void enqueue_task(struct rq *rq, struct task_struct *p, int wakeup)
1733 1790
1734static void dequeue_task(struct rq *rq, struct task_struct *p, int sleep) 1791static void dequeue_task(struct rq *rq, struct task_struct *p, int sleep)
1735{ 1792{
1736 if (sleep && p->se.last_wakeup) { 1793 if (sleep) {
1737 update_avg(&p->se.avg_overlap, 1794 if (p->se.last_wakeup) {
1738 p->se.sum_exec_runtime - p->se.last_wakeup); 1795 update_avg(&p->se.avg_overlap,
1739 p->se.last_wakeup = 0; 1796 p->se.sum_exec_runtime - p->se.last_wakeup);
1797 p->se.last_wakeup = 0;
1798 } else {
1799 update_avg(&p->se.avg_wakeup,
1800 sysctl_sched_wakeup_granularity);
1801 }
1740 } 1802 }
1741 1803
1742 sched_info_dequeued(p); 1804 sched_info_dequeued(p);
@@ -2041,7 +2103,7 @@ unsigned long wait_task_inactive(struct task_struct *p, long match_state)
2041 * it must be off the runqueue _entirely_, and not 2103 * it must be off the runqueue _entirely_, and not
2042 * preempted! 2104 * preempted!
2043 * 2105 *
2044 * So if it wa still runnable (but just not actively 2106 * So if it was still runnable (but just not actively
2045 * running right now), it's preempted, and we should 2107 * running right now), it's preempted, and we should
2046 * yield - it could be a while. 2108 * yield - it could be a while.
2047 */ 2109 */
@@ -2312,7 +2374,7 @@ static int try_to_wake_up(struct task_struct *p, unsigned int state, int sync)
2312 sync = 0; 2374 sync = 0;
2313 2375
2314#ifdef CONFIG_SMP 2376#ifdef CONFIG_SMP
2315 if (sched_feat(LB_WAKEUP_UPDATE)) { 2377 if (sched_feat(LB_WAKEUP_UPDATE) && !root_task_group_empty()) {
2316 struct sched_domain *sd; 2378 struct sched_domain *sd;
2317 2379
2318 this_cpu = raw_smp_processor_id(); 2380 this_cpu = raw_smp_processor_id();
@@ -2390,6 +2452,22 @@ out_activate:
2390 activate_task(rq, p, 1); 2452 activate_task(rq, p, 1);
2391 success = 1; 2453 success = 1;
2392 2454
2455 /*
2456 * Only attribute actual wakeups done by this task.
2457 */
2458 if (!in_interrupt()) {
2459 struct sched_entity *se = &current->se;
2460 u64 sample = se->sum_exec_runtime;
2461
2462 if (se->last_wakeup)
2463 sample -= se->last_wakeup;
2464 else
2465 sample -= se->start_runtime;
2466 update_avg(&se->avg_wakeup, sample);
2467
2468 se->last_wakeup = se->sum_exec_runtime;
2469 }
2470
2393out_running: 2471out_running:
2394 trace_sched_wakeup(rq, p, success); 2472 trace_sched_wakeup(rq, p, success);
2395 check_preempt_curr(rq, p, sync); 2473 check_preempt_curr(rq, p, sync);
@@ -2400,8 +2478,6 @@ out_running:
2400 p->sched_class->task_wake_up(rq, p); 2478 p->sched_class->task_wake_up(rq, p);
2401#endif 2479#endif
2402out: 2480out:
2403 current->se.last_wakeup = current->se.sum_exec_runtime;
2404
2405 task_rq_unlock(rq, &flags); 2481 task_rq_unlock(rq, &flags);
2406 2482
2407 return success; 2483 return success;
@@ -2432,6 +2508,8 @@ static void __sched_fork(struct task_struct *p)
2432 p->se.nr_migrations = 0; 2508 p->se.nr_migrations = 0;
2433 p->se.last_wakeup = 0; 2509 p->se.last_wakeup = 0;
2434 p->se.avg_overlap = 0; 2510 p->se.avg_overlap = 0;
2511 p->se.start_runtime = 0;
2512 p->se.avg_wakeup = sysctl_sched_wakeup_granularity;
2435 2513
2436#ifdef CONFIG_SCHEDSTATS 2514#ifdef CONFIG_SCHEDSTATS
2437 p->se.wait_start = 0; 2515 p->se.wait_start = 0;
@@ -2494,6 +2572,8 @@ void sched_fork(struct task_struct *p, int clone_flags)
2494 /* Want to start with kernel preemption disabled. */ 2572 /* Want to start with kernel preemption disabled. */
2495 task_thread_info(p)->preempt_count = 1; 2573 task_thread_info(p)->preempt_count = 1;
2496#endif 2574#endif
2575 plist_node_init(&p->pushable_tasks, MAX_PRIO);
2576
2497 put_cpu(); 2577 put_cpu();
2498} 2578}
2499 2579
@@ -2537,7 +2617,7 @@ void wake_up_new_task(struct task_struct *p, unsigned long clone_flags)
2537#ifdef CONFIG_PREEMPT_NOTIFIERS 2617#ifdef CONFIG_PREEMPT_NOTIFIERS
2538 2618
2539/** 2619/**
2540 * preempt_notifier_register - tell me when current is being being preempted & rescheduled 2620 * preempt_notifier_register - tell me when current is being preempted & rescheduled
2541 * @notifier: notifier struct to register 2621 * @notifier: notifier struct to register
2542 */ 2622 */
2543void preempt_notifier_register(struct preempt_notifier *notifier) 2623void preempt_notifier_register(struct preempt_notifier *notifier)
@@ -2634,6 +2714,12 @@ static void finish_task_switch(struct rq *rq, struct task_struct *prev)
2634{ 2714{
2635 struct mm_struct *mm = rq->prev_mm; 2715 struct mm_struct *mm = rq->prev_mm;
2636 long prev_state; 2716 long prev_state;
2717#ifdef CONFIG_SMP
2718 int post_schedule = 0;
2719
2720 if (current->sched_class->needs_post_schedule)
2721 post_schedule = current->sched_class->needs_post_schedule(rq);
2722#endif
2637 2723
2638 rq->prev_mm = NULL; 2724 rq->prev_mm = NULL;
2639 2725
@@ -2653,7 +2739,7 @@ static void finish_task_switch(struct rq *rq, struct task_struct *prev)
2653 perf_counter_task_sched_in(current, cpu_of(rq)); 2739 perf_counter_task_sched_in(current, cpu_of(rq));
2654 finish_lock_switch(rq, prev); 2740 finish_lock_switch(rq, prev);
2655#ifdef CONFIG_SMP 2741#ifdef CONFIG_SMP
2656 if (current->sched_class->post_schedule) 2742 if (post_schedule)
2657 current->sched_class->post_schedule(rq); 2743 current->sched_class->post_schedule(rq);
2658#endif 2744#endif
2659 2745
@@ -2975,6 +3061,7 @@ int can_migrate_task(struct task_struct *p, struct rq *rq, int this_cpu,
2975 struct sched_domain *sd, enum cpu_idle_type idle, 3061 struct sched_domain *sd, enum cpu_idle_type idle,
2976 int *all_pinned) 3062 int *all_pinned)
2977{ 3063{
3064 int tsk_cache_hot = 0;
2978 /* 3065 /*
2979 * We do not migrate tasks that are: 3066 * We do not migrate tasks that are:
2980 * 1) running (obviously), or 3067 * 1) running (obviously), or
@@ -2998,10 +3085,11 @@ int can_migrate_task(struct task_struct *p, struct rq *rq, int this_cpu,
2998 * 2) too many balance attempts have failed. 3085 * 2) too many balance attempts have failed.
2999 */ 3086 */
3000 3087
3001 if (!task_hot(p, rq->clock, sd) || 3088 tsk_cache_hot = task_hot(p, rq->clock, sd);
3002 sd->nr_balance_failed > sd->cache_nice_tries) { 3089 if (!tsk_cache_hot ||
3090 sd->nr_balance_failed > sd->cache_nice_tries) {
3003#ifdef CONFIG_SCHEDSTATS 3091#ifdef CONFIG_SCHEDSTATS
3004 if (task_hot(p, rq->clock, sd)) { 3092 if (tsk_cache_hot) {
3005 schedstat_inc(sd, lb_hot_gained[idle]); 3093 schedstat_inc(sd, lb_hot_gained[idle]);
3006 schedstat_inc(p, se.nr_forced_migrations); 3094 schedstat_inc(p, se.nr_forced_migrations);
3007 } 3095 }
@@ -3009,7 +3097,7 @@ int can_migrate_task(struct task_struct *p, struct rq *rq, int this_cpu,
3009 return 1; 3097 return 1;
3010 } 3098 }
3011 3099
3012 if (task_hot(p, rq->clock, sd)) { 3100 if (tsk_cache_hot) {
3013 schedstat_inc(p, se.nr_failed_migrations_hot); 3101 schedstat_inc(p, se.nr_failed_migrations_hot);
3014 return 0; 3102 return 0;
3015 } 3103 }
@@ -3049,6 +3137,16 @@ next:
3049 pulled++; 3137 pulled++;
3050 rem_load_move -= p->se.load.weight; 3138 rem_load_move -= p->se.load.weight;
3051 3139
3140#ifdef CONFIG_PREEMPT
3141 /*
3142 * NEWIDLE balancing is a source of latency, so preemptible kernels
3143 * will stop after the first task is pulled to minimize the critical
3144 * section.
3145 */
3146 if (idle == CPU_NEWLY_IDLE)
3147 goto out;
3148#endif
3149
3052 /* 3150 /*
3053 * We only want to steal up to the prescribed amount of weighted load. 3151 * We only want to steal up to the prescribed amount of weighted load.
3054 */ 3152 */
@@ -3095,9 +3193,15 @@ static int move_tasks(struct rq *this_rq, int this_cpu, struct rq *busiest,
3095 sd, idle, all_pinned, &this_best_prio); 3193 sd, idle, all_pinned, &this_best_prio);
3096 class = class->next; 3194 class = class->next;
3097 3195
3196#ifdef CONFIG_PREEMPT
3197 /*
3198 * NEWIDLE balancing is a source of latency, so preemptible
3199 * kernels will stop after the first task is pulled to minimize
3200 * the critical section.
3201 */
3098 if (idle == CPU_NEWLY_IDLE && this_rq->nr_running) 3202 if (idle == CPU_NEWLY_IDLE && this_rq->nr_running)
3099 break; 3203 break;
3100 3204#endif
3101 } while (class && max_load_move > total_load_moved); 3205 } while (class && max_load_move > total_load_moved);
3102 3206
3103 return total_load_moved > 0; 3207 return total_load_moved > 0;
@@ -3147,246 +3251,480 @@ static int move_one_task(struct rq *this_rq, int this_cpu, struct rq *busiest,
3147 3251
3148 return 0; 3252 return 0;
3149} 3253}
3150 3254/********** Helpers for find_busiest_group ************************/
3151/* 3255/*
3152 * find_busiest_group finds and returns the busiest CPU group within the 3256 * sd_lb_stats - Structure to store the statistics of a sched_domain
3153 * domain. It calculates and returns the amount of weighted load which 3257 * during load balancing.
3154 * should be moved to restore balance via the imbalance parameter.
3155 */ 3258 */
3156static struct sched_group * 3259struct sd_lb_stats {
3157find_busiest_group(struct sched_domain *sd, int this_cpu, 3260 struct sched_group *busiest; /* Busiest group in this sd */
3158 unsigned long *imbalance, enum cpu_idle_type idle, 3261 struct sched_group *this; /* Local group in this sd */
3159 int *sd_idle, const struct cpumask *cpus, int *balance) 3262 unsigned long total_load; /* Total load of all groups in sd */
3160{ 3263 unsigned long total_pwr; /* Total power of all groups in sd */
3161 struct sched_group *busiest = NULL, *this = NULL, *group = sd->groups; 3264 unsigned long avg_load; /* Average load across all groups in sd */
3162 unsigned long max_load, avg_load, total_load, this_load, total_pwr; 3265
3163 unsigned long max_pull; 3266 /** Statistics of this group */
3164 unsigned long busiest_load_per_task, busiest_nr_running; 3267 unsigned long this_load;
3165 unsigned long this_load_per_task, this_nr_running; 3268 unsigned long this_load_per_task;
3166 int load_idx, group_imb = 0; 3269 unsigned long this_nr_running;
3270
3271 /* Statistics of the busiest group */
3272 unsigned long max_load;
3273 unsigned long busiest_load_per_task;
3274 unsigned long busiest_nr_running;
3275
3276 int group_imb; /* Is there imbalance in this sd */
3167#if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) 3277#if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT)
3168 int power_savings_balance = 1; 3278 int power_savings_balance; /* Is powersave balance needed for this sd */
3169 unsigned long leader_nr_running = 0, min_load_per_task = 0; 3279 struct sched_group *group_min; /* Least loaded group in sd */
3170 unsigned long min_nr_running = ULONG_MAX; 3280 struct sched_group *group_leader; /* Group which relieves group_min */
3171 struct sched_group *group_min = NULL, *group_leader = NULL; 3281 unsigned long min_load_per_task; /* load_per_task in group_min */
3282 unsigned long leader_nr_running; /* Nr running of group_leader */
3283 unsigned long min_nr_running; /* Nr running of group_min */
3172#endif 3284#endif
3285};
3173 3286
3174 max_load = this_load = total_load = total_pwr = 0; 3287/*
3175 busiest_load_per_task = busiest_nr_running = 0; 3288 * sg_lb_stats - stats of a sched_group required for load_balancing
3176 this_load_per_task = this_nr_running = 0; 3289 */
3290struct sg_lb_stats {
3291 unsigned long avg_load; /*Avg load across the CPUs of the group */
3292 unsigned long group_load; /* Total load over the CPUs of the group */
3293 unsigned long sum_nr_running; /* Nr tasks running in the group */
3294 unsigned long sum_weighted_load; /* Weighted load of group's tasks */
3295 unsigned long group_capacity;
3296 int group_imb; /* Is there an imbalance in the group ? */
3297};
3298
3299/**
3300 * group_first_cpu - Returns the first cpu in the cpumask of a sched_group.
3301 * @group: The group whose first cpu is to be returned.
3302 */
3303static inline unsigned int group_first_cpu(struct sched_group *group)
3304{
3305 return cpumask_first(sched_group_cpus(group));
3306}
3307
3308/**
3309 * get_sd_load_idx - Obtain the load index for a given sched domain.
3310 * @sd: The sched_domain whose load_idx is to be obtained.
3311 * @idle: The Idle status of the CPU for whose sd load_icx is obtained.
3312 */
3313static inline int get_sd_load_idx(struct sched_domain *sd,
3314 enum cpu_idle_type idle)
3315{
3316 int load_idx;
3177 3317
3178 if (idle == CPU_NOT_IDLE) 3318 switch (idle) {
3319 case CPU_NOT_IDLE:
3179 load_idx = sd->busy_idx; 3320 load_idx = sd->busy_idx;
3180 else if (idle == CPU_NEWLY_IDLE) 3321 break;
3322
3323 case CPU_NEWLY_IDLE:
3181 load_idx = sd->newidle_idx; 3324 load_idx = sd->newidle_idx;
3182 else 3325 break;
3326 default:
3183 load_idx = sd->idle_idx; 3327 load_idx = sd->idle_idx;
3328 break;
3329 }
3184 3330
3185 do { 3331 return load_idx;
3186 unsigned long load, group_capacity, max_cpu_load, min_cpu_load; 3332}
3187 int local_group;
3188 int i;
3189 int __group_imb = 0;
3190 unsigned int balance_cpu = -1, first_idle_cpu = 0;
3191 unsigned long sum_nr_running, sum_weighted_load;
3192 unsigned long sum_avg_load_per_task;
3193 unsigned long avg_load_per_task;
3194 3333
3195 local_group = cpumask_test_cpu(this_cpu,
3196 sched_group_cpus(group));
3197 3334
3198 if (local_group) 3335#if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT)
3199 balance_cpu = cpumask_first(sched_group_cpus(group)); 3336/**
3337 * init_sd_power_savings_stats - Initialize power savings statistics for
3338 * the given sched_domain, during load balancing.
3339 *
3340 * @sd: Sched domain whose power-savings statistics are to be initialized.
3341 * @sds: Variable containing the statistics for sd.
3342 * @idle: Idle status of the CPU at which we're performing load-balancing.
3343 */
3344static inline void init_sd_power_savings_stats(struct sched_domain *sd,
3345 struct sd_lb_stats *sds, enum cpu_idle_type idle)
3346{
3347 /*
3348 * Busy processors will not participate in power savings
3349 * balance.
3350 */
3351 if (idle == CPU_NOT_IDLE || !(sd->flags & SD_POWERSAVINGS_BALANCE))
3352 sds->power_savings_balance = 0;
3353 else {
3354 sds->power_savings_balance = 1;
3355 sds->min_nr_running = ULONG_MAX;
3356 sds->leader_nr_running = 0;
3357 }
3358}
3200 3359
3201 /* Tally up the load of all CPUs in the group */ 3360/**
3202 sum_weighted_load = sum_nr_running = avg_load = 0; 3361 * update_sd_power_savings_stats - Update the power saving stats for a
3203 sum_avg_load_per_task = avg_load_per_task = 0; 3362 * sched_domain while performing load balancing.
3363 *
3364 * @group: sched_group belonging to the sched_domain under consideration.
3365 * @sds: Variable containing the statistics of the sched_domain
3366 * @local_group: Does group contain the CPU for which we're performing
3367 * load balancing ?
3368 * @sgs: Variable containing the statistics of the group.
3369 */
3370static inline void update_sd_power_savings_stats(struct sched_group *group,
3371 struct sd_lb_stats *sds, int local_group, struct sg_lb_stats *sgs)
3372{
3204 3373
3205 max_cpu_load = 0; 3374 if (!sds->power_savings_balance)
3206 min_cpu_load = ~0UL; 3375 return;
3207 3376
3208 for_each_cpu_and(i, sched_group_cpus(group), cpus) { 3377 /*
3209 struct rq *rq = cpu_rq(i); 3378 * If the local group is idle or completely loaded
3379 * no need to do power savings balance at this domain
3380 */
3381 if (local_group && (sds->this_nr_running >= sgs->group_capacity ||
3382 !sds->this_nr_running))
3383 sds->power_savings_balance = 0;
3210 3384
3211 if (*sd_idle && rq->nr_running) 3385 /*
3212 *sd_idle = 0; 3386 * If a group is already running at full capacity or idle,
3387 * don't include that group in power savings calculations
3388 */
3389 if (!sds->power_savings_balance ||
3390 sgs->sum_nr_running >= sgs->group_capacity ||
3391 !sgs->sum_nr_running)
3392 return;
3213 3393
3214 /* Bias balancing toward cpus of our domain */ 3394 /*
3215 if (local_group) { 3395 * Calculate the group which has the least non-idle load.
3216 if (idle_cpu(i) && !first_idle_cpu) { 3396 * This is the group from where we need to pick up the load
3217 first_idle_cpu = 1; 3397 * for saving power
3218 balance_cpu = i; 3398 */
3219 } 3399 if ((sgs->sum_nr_running < sds->min_nr_running) ||
3400 (sgs->sum_nr_running == sds->min_nr_running &&
3401 group_first_cpu(group) > group_first_cpu(sds->group_min))) {
3402 sds->group_min = group;
3403 sds->min_nr_running = sgs->sum_nr_running;
3404 sds->min_load_per_task = sgs->sum_weighted_load /
3405 sgs->sum_nr_running;
3406 }
3220 3407
3221 load = target_load(i, load_idx); 3408 /*
3222 } else { 3409 * Calculate the group which is almost near its
3223 load = source_load(i, load_idx); 3410 * capacity but still has some space to pick up some load
3224 if (load > max_cpu_load) 3411 * from other group and save more power
3225 max_cpu_load = load; 3412 */
3226 if (min_cpu_load > load) 3413 if (sgs->sum_nr_running > sgs->group_capacity - 1)
3227 min_cpu_load = load; 3414 return;
3228 }
3229 3415
3230 avg_load += load; 3416 if (sgs->sum_nr_running > sds->leader_nr_running ||
3231 sum_nr_running += rq->nr_running; 3417 (sgs->sum_nr_running == sds->leader_nr_running &&
3232 sum_weighted_load += weighted_cpuload(i); 3418 group_first_cpu(group) < group_first_cpu(sds->group_leader))) {
3419 sds->group_leader = group;
3420 sds->leader_nr_running = sgs->sum_nr_running;
3421 }
3422}
3233 3423
3234 sum_avg_load_per_task += cpu_avg_load_per_task(i); 3424/**
3235 } 3425 * check_power_save_busiest_group - see if there is potential for some power-savings balance
3426 * @sds: Variable containing the statistics of the sched_domain
3427 * under consideration.
3428 * @this_cpu: Cpu at which we're currently performing load-balancing.
3429 * @imbalance: Variable to store the imbalance.
3430 *
3431 * Description:
3432 * Check if we have potential to perform some power-savings balance.
3433 * If yes, set the busiest group to be the least loaded group in the
3434 * sched_domain, so that it's CPUs can be put to idle.
3435 *
3436 * Returns 1 if there is potential to perform power-savings balance.
3437 * Else returns 0.
3438 */
3439static inline int check_power_save_busiest_group(struct sd_lb_stats *sds,
3440 int this_cpu, unsigned long *imbalance)
3441{
3442 if (!sds->power_savings_balance)
3443 return 0;
3236 3444
3237 /* 3445 if (sds->this != sds->group_leader ||
3238 * First idle cpu or the first cpu(busiest) in this sched group 3446 sds->group_leader == sds->group_min)
3239 * is eligible for doing load balancing at this and above 3447 return 0;
3240 * domains. In the newly idle case, we will allow all the cpu's
3241 * to do the newly idle load balance.
3242 */
3243 if (idle != CPU_NEWLY_IDLE && local_group &&
3244 balance_cpu != this_cpu && balance) {
3245 *balance = 0;
3246 goto ret;
3247 }
3248 3448
3249 total_load += avg_load; 3449 *imbalance = sds->min_load_per_task;
3250 total_pwr += group->__cpu_power; 3450 sds->busiest = sds->group_min;
3251 3451
3252 /* Adjust by relative CPU power of the group */ 3452 if (sched_mc_power_savings >= POWERSAVINGS_BALANCE_WAKEUP) {
3253 avg_load = sg_div_cpu_power(group, 3453 cpu_rq(this_cpu)->rd->sched_mc_preferred_wakeup_cpu =
3254 avg_load * SCHED_LOAD_SCALE); 3454 group_first_cpu(sds->group_leader);
3455 }
3255 3456
3457 return 1;
3256 3458
3257 /* 3459}
3258 * Consider the group unbalanced when the imbalance is larger 3460#else /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */
3259 * than the average weight of two tasks. 3461static inline void init_sd_power_savings_stats(struct sched_domain *sd,
3260 * 3462 struct sd_lb_stats *sds, enum cpu_idle_type idle)
3261 * APZ: with cgroup the avg task weight can vary wildly and 3463{
3262 * might not be a suitable number - should we keep a 3464 return;
3263 * normalized nr_running number somewhere that negates 3465}
3264 * the hierarchy? 3466
3265 */ 3467static inline void update_sd_power_savings_stats(struct sched_group *group,
3266 avg_load_per_task = sg_div_cpu_power(group, 3468 struct sd_lb_stats *sds, int local_group, struct sg_lb_stats *sgs)
3267 sum_avg_load_per_task * SCHED_LOAD_SCALE); 3469{
3470 return;
3471}
3472
3473static inline int check_power_save_busiest_group(struct sd_lb_stats *sds,
3474 int this_cpu, unsigned long *imbalance)
3475{
3476 return 0;
3477}
3478#endif /* CONFIG_SCHED_MC || CONFIG_SCHED_SMT */
3479
3480
3481/**
3482 * update_sg_lb_stats - Update sched_group's statistics for load balancing.
3483 * @group: sched_group whose statistics are to be updated.
3484 * @this_cpu: Cpu for which load balance is currently performed.
3485 * @idle: Idle status of this_cpu
3486 * @load_idx: Load index of sched_domain of this_cpu for load calc.
3487 * @sd_idle: Idle status of the sched_domain containing group.
3488 * @local_group: Does group contain this_cpu.
3489 * @cpus: Set of cpus considered for load balancing.
3490 * @balance: Should we balance.
3491 * @sgs: variable to hold the statistics for this group.
3492 */
3493static inline void update_sg_lb_stats(struct sched_group *group, int this_cpu,
3494 enum cpu_idle_type idle, int load_idx, int *sd_idle,
3495 int local_group, const struct cpumask *cpus,
3496 int *balance, struct sg_lb_stats *sgs)
3497{
3498 unsigned long load, max_cpu_load, min_cpu_load;
3499 int i;
3500 unsigned int balance_cpu = -1, first_idle_cpu = 0;
3501 unsigned long sum_avg_load_per_task;
3502 unsigned long avg_load_per_task;
3503
3504 if (local_group)
3505 balance_cpu = group_first_cpu(group);
3268 3506
3269 if ((max_cpu_load - min_cpu_load) > 2*avg_load_per_task) 3507 /* Tally up the load of all CPUs in the group */
3270 __group_imb = 1; 3508 sum_avg_load_per_task = avg_load_per_task = 0;
3509 max_cpu_load = 0;
3510 min_cpu_load = ~0UL;
3271 3511
3272 group_capacity = group->__cpu_power / SCHED_LOAD_SCALE; 3512 for_each_cpu_and(i, sched_group_cpus(group), cpus) {
3513 struct rq *rq = cpu_rq(i);
3273 3514
3515 if (*sd_idle && rq->nr_running)
3516 *sd_idle = 0;
3517
3518 /* Bias balancing toward cpus of our domain */
3274 if (local_group) { 3519 if (local_group) {
3275 this_load = avg_load; 3520 if (idle_cpu(i) && !first_idle_cpu) {
3276 this = group; 3521 first_idle_cpu = 1;
3277 this_nr_running = sum_nr_running; 3522 balance_cpu = i;
3278 this_load_per_task = sum_weighted_load; 3523 }
3279 } else if (avg_load > max_load && 3524
3280 (sum_nr_running > group_capacity || __group_imb)) { 3525 load = target_load(i, load_idx);
3281 max_load = avg_load; 3526 } else {
3282 busiest = group; 3527 load = source_load(i, load_idx);
3283 busiest_nr_running = sum_nr_running; 3528 if (load > max_cpu_load)
3284 busiest_load_per_task = sum_weighted_load; 3529 max_cpu_load = load;
3285 group_imb = __group_imb; 3530 if (min_cpu_load > load)
3531 min_cpu_load = load;
3286 } 3532 }
3287 3533
3288#if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) 3534 sgs->group_load += load;
3289 /* 3535 sgs->sum_nr_running += rq->nr_running;
3290 * Busy processors will not participate in power savings 3536 sgs->sum_weighted_load += weighted_cpuload(i);
3291 * balance.
3292 */
3293 if (idle == CPU_NOT_IDLE ||
3294 !(sd->flags & SD_POWERSAVINGS_BALANCE))
3295 goto group_next;
3296 3537
3297 /* 3538 sum_avg_load_per_task += cpu_avg_load_per_task(i);
3298 * If the local group is idle or completely loaded 3539 }
3299 * no need to do power savings balance at this domain
3300 */
3301 if (local_group && (this_nr_running >= group_capacity ||
3302 !this_nr_running))
3303 power_savings_balance = 0;
3304 3540
3305 /* 3541 /*
3306 * If a group is already running at full capacity or idle, 3542 * First idle cpu or the first cpu(busiest) in this sched group
3307 * don't include that group in power savings calculations 3543 * is eligible for doing load balancing at this and above
3308 */ 3544 * domains. In the newly idle case, we will allow all the cpu's
3309 if (!power_savings_balance || sum_nr_running >= group_capacity 3545 * to do the newly idle load balance.
3310 || !sum_nr_running) 3546 */
3311 goto group_next; 3547 if (idle != CPU_NEWLY_IDLE && local_group &&
3548 balance_cpu != this_cpu && balance) {
3549 *balance = 0;
3550 return;
3551 }
3312 3552
3313 /* 3553 /* Adjust by relative CPU power of the group */
3314 * Calculate the group which has the least non-idle load. 3554 sgs->avg_load = sg_div_cpu_power(group,
3315 * This is the group from where we need to pick up the load 3555 sgs->group_load * SCHED_LOAD_SCALE);
3316 * for saving power
3317 */
3318 if ((sum_nr_running < min_nr_running) ||
3319 (sum_nr_running == min_nr_running &&
3320 cpumask_first(sched_group_cpus(group)) >
3321 cpumask_first(sched_group_cpus(group_min)))) {
3322 group_min = group;
3323 min_nr_running = sum_nr_running;
3324 min_load_per_task = sum_weighted_load /
3325 sum_nr_running;
3326 }
3327 3556
3328 /* 3557
3329 * Calculate the group which is almost near its 3558 /*
3330 * capacity but still has some space to pick up some load 3559 * Consider the group unbalanced when the imbalance is larger
3331 * from other group and save more power 3560 * than the average weight of two tasks.
3332 */ 3561 *
3333 if (sum_nr_running <= group_capacity - 1) { 3562 * APZ: with cgroup the avg task weight can vary wildly and
3334 if (sum_nr_running > leader_nr_running || 3563 * might not be a suitable number - should we keep a
3335 (sum_nr_running == leader_nr_running && 3564 * normalized nr_running number somewhere that negates
3336 cpumask_first(sched_group_cpus(group)) < 3565 * the hierarchy?
3337 cpumask_first(sched_group_cpus(group_leader)))) { 3566 */
3338 group_leader = group; 3567 avg_load_per_task = sg_div_cpu_power(group,
3339 leader_nr_running = sum_nr_running; 3568 sum_avg_load_per_task * SCHED_LOAD_SCALE);
3340 } 3569
3570 if ((max_cpu_load - min_cpu_load) > 2*avg_load_per_task)
3571 sgs->group_imb = 1;
3572
3573 sgs->group_capacity = group->__cpu_power / SCHED_LOAD_SCALE;
3574
3575}
3576
3577/**
3578 * update_sd_lb_stats - Update sched_group's statistics for load balancing.
3579 * @sd: sched_domain whose statistics are to be updated.
3580 * @this_cpu: Cpu for which load balance is currently performed.
3581 * @idle: Idle status of this_cpu
3582 * @sd_idle: Idle status of the sched_domain containing group.
3583 * @cpus: Set of cpus considered for load balancing.
3584 * @balance: Should we balance.
3585 * @sds: variable to hold the statistics for this sched_domain.
3586 */
3587static inline void update_sd_lb_stats(struct sched_domain *sd, int this_cpu,
3588 enum cpu_idle_type idle, int *sd_idle,
3589 const struct cpumask *cpus, int *balance,
3590 struct sd_lb_stats *sds)
3591{
3592 struct sched_group *group = sd->groups;
3593 struct sg_lb_stats sgs;
3594 int load_idx;
3595
3596 init_sd_power_savings_stats(sd, sds, idle);
3597 load_idx = get_sd_load_idx(sd, idle);
3598
3599 do {
3600 int local_group;
3601
3602 local_group = cpumask_test_cpu(this_cpu,
3603 sched_group_cpus(group));
3604 memset(&sgs, 0, sizeof(sgs));
3605 update_sg_lb_stats(group, this_cpu, idle, load_idx, sd_idle,
3606 local_group, cpus, balance, &sgs);
3607
3608 if (local_group && balance && !(*balance))
3609 return;
3610
3611 sds->total_load += sgs.group_load;
3612 sds->total_pwr += group->__cpu_power;
3613
3614 if (local_group) {
3615 sds->this_load = sgs.avg_load;
3616 sds->this = group;
3617 sds->this_nr_running = sgs.sum_nr_running;
3618 sds->this_load_per_task = sgs.sum_weighted_load;
3619 } else if (sgs.avg_load > sds->max_load &&
3620 (sgs.sum_nr_running > sgs.group_capacity ||
3621 sgs.group_imb)) {
3622 sds->max_load = sgs.avg_load;
3623 sds->busiest = group;
3624 sds->busiest_nr_running = sgs.sum_nr_running;
3625 sds->busiest_load_per_task = sgs.sum_weighted_load;
3626 sds->group_imb = sgs.group_imb;
3341 } 3627 }
3342group_next: 3628
3343#endif 3629 update_sd_power_savings_stats(group, sds, local_group, &sgs);
3344 group = group->next; 3630 group = group->next;
3345 } while (group != sd->groups); 3631 } while (group != sd->groups);
3346 3632
3347 if (!busiest || this_load >= max_load || busiest_nr_running == 0) 3633}
3348 goto out_balanced;
3349
3350 avg_load = (SCHED_LOAD_SCALE * total_load) / total_pwr;
3351 3634
3352 if (this_load >= avg_load || 3635/**
3353 100*max_load <= sd->imbalance_pct*this_load) 3636 * fix_small_imbalance - Calculate the minor imbalance that exists
3354 goto out_balanced; 3637 * amongst the groups of a sched_domain, during
3638 * load balancing.
3639 * @sds: Statistics of the sched_domain whose imbalance is to be calculated.
3640 * @this_cpu: The cpu at whose sched_domain we're performing load-balance.
3641 * @imbalance: Variable to store the imbalance.
3642 */
3643static inline void fix_small_imbalance(struct sd_lb_stats *sds,
3644 int this_cpu, unsigned long *imbalance)
3645{
3646 unsigned long tmp, pwr_now = 0, pwr_move = 0;
3647 unsigned int imbn = 2;
3648
3649 if (sds->this_nr_running) {
3650 sds->this_load_per_task /= sds->this_nr_running;
3651 if (sds->busiest_load_per_task >
3652 sds->this_load_per_task)
3653 imbn = 1;
3654 } else
3655 sds->this_load_per_task =
3656 cpu_avg_load_per_task(this_cpu);
3355 3657
3356 busiest_load_per_task /= busiest_nr_running; 3658 if (sds->max_load - sds->this_load + sds->busiest_load_per_task >=
3357 if (group_imb) 3659 sds->busiest_load_per_task * imbn) {
3358 busiest_load_per_task = min(busiest_load_per_task, avg_load); 3660 *imbalance = sds->busiest_load_per_task;
3661 return;
3662 }
3359 3663
3360 /* 3664 /*
3361 * We're trying to get all the cpus to the average_load, so we don't 3665 * OK, we don't have enough imbalance to justify moving tasks,
3362 * want to push ourselves above the average load, nor do we wish to 3666 * however we may be able to increase total CPU power used by
3363 * reduce the max loaded cpu below the average load, as either of these 3667 * moving them.
3364 * actions would just result in more rebalancing later, and ping-pong
3365 * tasks around. Thus we look for the minimum possible imbalance.
3366 * Negative imbalances (*we* are more loaded than anyone else) will
3367 * be counted as no imbalance for these purposes -- we can't fix that
3368 * by pulling tasks to us. Be careful of negative numbers as they'll
3369 * appear as very large values with unsigned longs.
3370 */ 3668 */
3371 if (max_load <= busiest_load_per_task)
3372 goto out_balanced;
3373 3669
3670 pwr_now += sds->busiest->__cpu_power *
3671 min(sds->busiest_load_per_task, sds->max_load);
3672 pwr_now += sds->this->__cpu_power *
3673 min(sds->this_load_per_task, sds->this_load);
3674 pwr_now /= SCHED_LOAD_SCALE;
3675
3676 /* Amount of load we'd subtract */
3677 tmp = sg_div_cpu_power(sds->busiest,
3678 sds->busiest_load_per_task * SCHED_LOAD_SCALE);
3679 if (sds->max_load > tmp)
3680 pwr_move += sds->busiest->__cpu_power *
3681 min(sds->busiest_load_per_task, sds->max_load - tmp);
3682
3683 /* Amount of load we'd add */
3684 if (sds->max_load * sds->busiest->__cpu_power <
3685 sds->busiest_load_per_task * SCHED_LOAD_SCALE)
3686 tmp = sg_div_cpu_power(sds->this,
3687 sds->max_load * sds->busiest->__cpu_power);
3688 else
3689 tmp = sg_div_cpu_power(sds->this,
3690 sds->busiest_load_per_task * SCHED_LOAD_SCALE);
3691 pwr_move += sds->this->__cpu_power *
3692 min(sds->this_load_per_task, sds->this_load + tmp);
3693 pwr_move /= SCHED_LOAD_SCALE;
3694
3695 /* Move if we gain throughput */
3696 if (pwr_move > pwr_now)
3697 *imbalance = sds->busiest_load_per_task;
3698}
3699
3700/**
3701 * calculate_imbalance - Calculate the amount of imbalance present within the
3702 * groups of a given sched_domain during load balance.
3703 * @sds: statistics of the sched_domain whose imbalance is to be calculated.
3704 * @this_cpu: Cpu for which currently load balance is being performed.
3705 * @imbalance: The variable to store the imbalance.
3706 */
3707static inline void calculate_imbalance(struct sd_lb_stats *sds, int this_cpu,
3708 unsigned long *imbalance)
3709{
3710 unsigned long max_pull;
3374 /* 3711 /*
3375 * In the presence of smp nice balancing, certain scenarios can have 3712 * In the presence of smp nice balancing, certain scenarios can have
3376 * max load less than avg load(as we skip the groups at or below 3713 * max load less than avg load(as we skip the groups at or below
3377 * its cpu_power, while calculating max_load..) 3714 * its cpu_power, while calculating max_load..)
3378 */ 3715 */
3379 if (max_load < avg_load) { 3716 if (sds->max_load < sds->avg_load) {
3380 *imbalance = 0; 3717 *imbalance = 0;
3381 goto small_imbalance; 3718 return fix_small_imbalance(sds, this_cpu, imbalance);
3382 } 3719 }
3383 3720
3384 /* Don't want to pull so many tasks that a group would go idle */ 3721 /* Don't want to pull so many tasks that a group would go idle */
3385 max_pull = min(max_load - avg_load, max_load - busiest_load_per_task); 3722 max_pull = min(sds->max_load - sds->avg_load,
3723 sds->max_load - sds->busiest_load_per_task);
3386 3724
3387 /* How much load to actually move to equalise the imbalance */ 3725 /* How much load to actually move to equalise the imbalance */
3388 *imbalance = min(max_pull * busiest->__cpu_power, 3726 *imbalance = min(max_pull * sds->busiest->__cpu_power,
3389 (avg_load - this_load) * this->__cpu_power) 3727 (sds->avg_load - sds->this_load) * sds->this->__cpu_power)
3390 / SCHED_LOAD_SCALE; 3728 / SCHED_LOAD_SCALE;
3391 3729
3392 /* 3730 /*
@@ -3395,78 +3733,110 @@ group_next:
3395 * a think about bumping its value to force at least one task to be 3733 * a think about bumping its value to force at least one task to be
3396 * moved 3734 * moved
3397 */ 3735 */
3398 if (*imbalance < busiest_load_per_task) { 3736 if (*imbalance < sds->busiest_load_per_task)
3399 unsigned long tmp, pwr_now, pwr_move; 3737 return fix_small_imbalance(sds, this_cpu, imbalance);
3400 unsigned int imbn;
3401
3402small_imbalance:
3403 pwr_move = pwr_now = 0;
3404 imbn = 2;
3405 if (this_nr_running) {
3406 this_load_per_task /= this_nr_running;
3407 if (busiest_load_per_task > this_load_per_task)
3408 imbn = 1;
3409 } else
3410 this_load_per_task = cpu_avg_load_per_task(this_cpu);
3411 3738
3412 if (max_load - this_load + busiest_load_per_task >= 3739}
3413 busiest_load_per_task * imbn) { 3740/******* find_busiest_group() helpers end here *********************/
3414 *imbalance = busiest_load_per_task;
3415 return busiest;
3416 }
3417 3741
3418 /* 3742/**
3419 * OK, we don't have enough imbalance to justify moving tasks, 3743 * find_busiest_group - Returns the busiest group within the sched_domain
3420 * however we may be able to increase total CPU power used by 3744 * if there is an imbalance. If there isn't an imbalance, and
3421 * moving them. 3745 * the user has opted for power-savings, it returns a group whose
3422 */ 3746 * CPUs can be put to idle by rebalancing those tasks elsewhere, if
3747 * such a group exists.
3748 *
3749 * Also calculates the amount of weighted load which should be moved
3750 * to restore balance.
3751 *
3752 * @sd: The sched_domain whose busiest group is to be returned.
3753 * @this_cpu: The cpu for which load balancing is currently being performed.
3754 * @imbalance: Variable which stores amount of weighted load which should
3755 * be moved to restore balance/put a group to idle.
3756 * @idle: The idle status of this_cpu.
3757 * @sd_idle: The idleness of sd
3758 * @cpus: The set of CPUs under consideration for load-balancing.
3759 * @balance: Pointer to a variable indicating if this_cpu
3760 * is the appropriate cpu to perform load balancing at this_level.
3761 *
3762 * Returns: - the busiest group if imbalance exists.
3763 * - If no imbalance and user has opted for power-savings balance,
3764 * return the least loaded group whose CPUs can be
3765 * put to idle by rebalancing its tasks onto our group.
3766 */
3767static struct sched_group *
3768find_busiest_group(struct sched_domain *sd, int this_cpu,
3769 unsigned long *imbalance, enum cpu_idle_type idle,
3770 int *sd_idle, const struct cpumask *cpus, int *balance)
3771{
3772 struct sd_lb_stats sds;
3423 3773
3424 pwr_now += busiest->__cpu_power * 3774 memset(&sds, 0, sizeof(sds));
3425 min(busiest_load_per_task, max_load);
3426 pwr_now += this->__cpu_power *
3427 min(this_load_per_task, this_load);
3428 pwr_now /= SCHED_LOAD_SCALE;
3429
3430 /* Amount of load we'd subtract */
3431 tmp = sg_div_cpu_power(busiest,
3432 busiest_load_per_task * SCHED_LOAD_SCALE);
3433 if (max_load > tmp)
3434 pwr_move += busiest->__cpu_power *
3435 min(busiest_load_per_task, max_load - tmp);
3436
3437 /* Amount of load we'd add */
3438 if (max_load * busiest->__cpu_power <
3439 busiest_load_per_task * SCHED_LOAD_SCALE)
3440 tmp = sg_div_cpu_power(this,
3441 max_load * busiest->__cpu_power);
3442 else
3443 tmp = sg_div_cpu_power(this,
3444 busiest_load_per_task * SCHED_LOAD_SCALE);
3445 pwr_move += this->__cpu_power *
3446 min(this_load_per_task, this_load + tmp);
3447 pwr_move /= SCHED_LOAD_SCALE;
3448 3775
3449 /* Move if we gain throughput */ 3776 /*
3450 if (pwr_move > pwr_now) 3777 * Compute the various statistics relavent for load balancing at
3451 *imbalance = busiest_load_per_task; 3778 * this level.
3452 } 3779 */
3780 update_sd_lb_stats(sd, this_cpu, idle, sd_idle, cpus,
3781 balance, &sds);
3782
3783 /* Cases where imbalance does not exist from POV of this_cpu */
3784 /* 1) this_cpu is not the appropriate cpu to perform load balancing
3785 * at this level.
3786 * 2) There is no busy sibling group to pull from.
3787 * 3) This group is the busiest group.
3788 * 4) This group is more busy than the avg busieness at this
3789 * sched_domain.
3790 * 5) The imbalance is within the specified limit.
3791 * 6) Any rebalance would lead to ping-pong
3792 */
3793 if (balance && !(*balance))
3794 goto ret;
3453 3795
3454 return busiest; 3796 if (!sds.busiest || sds.busiest_nr_running == 0)
3797 goto out_balanced;
3455 3798
3456out_balanced: 3799 if (sds.this_load >= sds.max_load)
3457#if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) 3800 goto out_balanced;
3458 if (idle == CPU_NOT_IDLE || !(sd->flags & SD_POWERSAVINGS_BALANCE))
3459 goto ret;
3460 3801
3461 if (this == group_leader && group_leader != group_min) { 3802 sds.avg_load = (SCHED_LOAD_SCALE * sds.total_load) / sds.total_pwr;
3462 *imbalance = min_load_per_task; 3803
3463 if (sched_mc_power_savings >= POWERSAVINGS_BALANCE_WAKEUP) { 3804 if (sds.this_load >= sds.avg_load)
3464 cpu_rq(this_cpu)->rd->sched_mc_preferred_wakeup_cpu = 3805 goto out_balanced;
3465 cpumask_first(sched_group_cpus(group_leader)); 3806
3466 } 3807 if (100 * sds.max_load <= sd->imbalance_pct * sds.this_load)
3467 return group_min; 3808 goto out_balanced;
3468 } 3809
3469#endif 3810 sds.busiest_load_per_task /= sds.busiest_nr_running;
3811 if (sds.group_imb)
3812 sds.busiest_load_per_task =
3813 min(sds.busiest_load_per_task, sds.avg_load);
3814
3815 /*
3816 * We're trying to get all the cpus to the average_load, so we don't
3817 * want to push ourselves above the average load, nor do we wish to
3818 * reduce the max loaded cpu below the average load, as either of these
3819 * actions would just result in more rebalancing later, and ping-pong
3820 * tasks around. Thus we look for the minimum possible imbalance.
3821 * Negative imbalances (*we* are more loaded than anyone else) will
3822 * be counted as no imbalance for these purposes -- we can't fix that
3823 * by pulling tasks to us. Be careful of negative numbers as they'll
3824 * appear as very large values with unsigned longs.
3825 */
3826 if (sds.max_load <= sds.busiest_load_per_task)
3827 goto out_balanced;
3828
3829 /* Looks like there is an imbalance. Compute it */
3830 calculate_imbalance(&sds, this_cpu, imbalance);
3831 return sds.busiest;
3832
3833out_balanced:
3834 /*
3835 * There is no obvious imbalance. But check if we can do some balancing
3836 * to save power.
3837 */
3838 if (check_power_save_busiest_group(&sds, this_cpu, imbalance))
3839 return sds.busiest;
3470ret: 3840ret:
3471 *imbalance = 0; 3841 *imbalance = 0;
3472 return NULL; 3842 return NULL;
@@ -3510,19 +3880,23 @@ find_busiest_queue(struct sched_group *group, enum cpu_idle_type idle,
3510 */ 3880 */
3511#define MAX_PINNED_INTERVAL 512 3881#define MAX_PINNED_INTERVAL 512
3512 3882
3883/* Working cpumask for load_balance and load_balance_newidle. */
3884static DEFINE_PER_CPU(cpumask_var_t, load_balance_tmpmask);
3885
3513/* 3886/*
3514 * Check this_cpu to ensure it is balanced within domain. Attempt to move 3887 * Check this_cpu to ensure it is balanced within domain. Attempt to move
3515 * tasks if there is an imbalance. 3888 * tasks if there is an imbalance.
3516 */ 3889 */
3517static int load_balance(int this_cpu, struct rq *this_rq, 3890static int load_balance(int this_cpu, struct rq *this_rq,
3518 struct sched_domain *sd, enum cpu_idle_type idle, 3891 struct sched_domain *sd, enum cpu_idle_type idle,
3519 int *balance, struct cpumask *cpus) 3892 int *balance)
3520{ 3893{
3521 int ld_moved, all_pinned = 0, active_balance = 0, sd_idle = 0; 3894 int ld_moved, all_pinned = 0, active_balance = 0, sd_idle = 0;
3522 struct sched_group *group; 3895 struct sched_group *group;
3523 unsigned long imbalance; 3896 unsigned long imbalance;
3524 struct rq *busiest; 3897 struct rq *busiest;
3525 unsigned long flags; 3898 unsigned long flags;
3899 struct cpumask *cpus = __get_cpu_var(load_balance_tmpmask);
3526 3900
3527 cpumask_setall(cpus); 3901 cpumask_setall(cpus);
3528 3902
@@ -3677,8 +4051,7 @@ out:
3677 * this_rq is locked. 4051 * this_rq is locked.
3678 */ 4052 */
3679static int 4053static int
3680load_balance_newidle(int this_cpu, struct rq *this_rq, struct sched_domain *sd, 4054load_balance_newidle(int this_cpu, struct rq *this_rq, struct sched_domain *sd)
3681 struct cpumask *cpus)
3682{ 4055{
3683 struct sched_group *group; 4056 struct sched_group *group;
3684 struct rq *busiest = NULL; 4057 struct rq *busiest = NULL;
@@ -3686,6 +4059,7 @@ load_balance_newidle(int this_cpu, struct rq *this_rq, struct sched_domain *sd,
3686 int ld_moved = 0; 4059 int ld_moved = 0;
3687 int sd_idle = 0; 4060 int sd_idle = 0;
3688 int all_pinned = 0; 4061 int all_pinned = 0;
4062 struct cpumask *cpus = __get_cpu_var(load_balance_tmpmask);
3689 4063
3690 cpumask_setall(cpus); 4064 cpumask_setall(cpus);
3691 4065
@@ -3826,10 +4200,6 @@ static void idle_balance(int this_cpu, struct rq *this_rq)
3826 struct sched_domain *sd; 4200 struct sched_domain *sd;
3827 int pulled_task = 0; 4201 int pulled_task = 0;
3828 unsigned long next_balance = jiffies + HZ; 4202 unsigned long next_balance = jiffies + HZ;
3829 cpumask_var_t tmpmask;
3830
3831 if (!alloc_cpumask_var(&tmpmask, GFP_ATOMIC))
3832 return;
3833 4203
3834 for_each_domain(this_cpu, sd) { 4204 for_each_domain(this_cpu, sd) {
3835 unsigned long interval; 4205 unsigned long interval;
@@ -3840,7 +4210,7 @@ static void idle_balance(int this_cpu, struct rq *this_rq)
3840 if (sd->flags & SD_BALANCE_NEWIDLE) 4210 if (sd->flags & SD_BALANCE_NEWIDLE)
3841 /* If we've pulled tasks over stop searching: */ 4211 /* If we've pulled tasks over stop searching: */
3842 pulled_task = load_balance_newidle(this_cpu, this_rq, 4212 pulled_task = load_balance_newidle(this_cpu, this_rq,
3843 sd, tmpmask); 4213 sd);
3844 4214
3845 interval = msecs_to_jiffies(sd->balance_interval); 4215 interval = msecs_to_jiffies(sd->balance_interval);
3846 if (time_after(next_balance, sd->last_balance + interval)) 4216 if (time_after(next_balance, sd->last_balance + interval))
@@ -3855,7 +4225,6 @@ static void idle_balance(int this_cpu, struct rq *this_rq)
3855 */ 4225 */
3856 this_rq->next_balance = next_balance; 4226 this_rq->next_balance = next_balance;
3857 } 4227 }
3858 free_cpumask_var(tmpmask);
3859} 4228}
3860 4229
3861/* 4230/*
@@ -4005,11 +4374,6 @@ static void rebalance_domains(int cpu, enum cpu_idle_type idle)
4005 unsigned long next_balance = jiffies + 60*HZ; 4374 unsigned long next_balance = jiffies + 60*HZ;
4006 int update_next_balance = 0; 4375 int update_next_balance = 0;
4007 int need_serialize; 4376 int need_serialize;
4008 cpumask_var_t tmp;
4009
4010 /* Fails alloc? Rebalancing probably not a priority right now. */
4011 if (!alloc_cpumask_var(&tmp, GFP_ATOMIC))
4012 return;
4013 4377
4014 for_each_domain(cpu, sd) { 4378 for_each_domain(cpu, sd) {
4015 if (!(sd->flags & SD_LOAD_BALANCE)) 4379 if (!(sd->flags & SD_LOAD_BALANCE))
@@ -4034,7 +4398,7 @@ static void rebalance_domains(int cpu, enum cpu_idle_type idle)
4034 } 4398 }
4035 4399
4036 if (time_after_eq(jiffies, sd->last_balance + interval)) { 4400 if (time_after_eq(jiffies, sd->last_balance + interval)) {
4037 if (load_balance(cpu, rq, sd, idle, &balance, tmp)) { 4401 if (load_balance(cpu, rq, sd, idle, &balance)) {
4038 /* 4402 /*
4039 * We've pulled tasks over so either we're no 4403 * We've pulled tasks over so either we're no
4040 * longer idle, or one of our SMT siblings is 4404 * longer idle, or one of our SMT siblings is
@@ -4068,8 +4432,6 @@ out:
4068 */ 4432 */
4069 if (likely(update_next_balance)) 4433 if (likely(update_next_balance))
4070 rq->next_balance = next_balance; 4434 rq->next_balance = next_balance;
4071
4072 free_cpumask_var(tmp);
4073} 4435}
4074 4436
4075/* 4437/*
@@ -4119,6 +4481,11 @@ static void run_rebalance_domains(struct softirq_action *h)
4119#endif 4481#endif
4120} 4482}
4121 4483
4484static inline int on_null_domain(int cpu)
4485{
4486 return !rcu_dereference(cpu_rq(cpu)->sd);
4487}
4488
4122/* 4489/*
4123 * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing. 4490 * Trigger the SCHED_SOFTIRQ if it is time to do periodic load balancing.
4124 * 4491 *
@@ -4176,7 +4543,9 @@ static inline void trigger_load_balance(struct rq *rq, int cpu)
4176 cpumask_test_cpu(cpu, nohz.cpu_mask)) 4543 cpumask_test_cpu(cpu, nohz.cpu_mask))
4177 return; 4544 return;
4178#endif 4545#endif
4179 if (time_after_eq(jiffies, rq->next_balance)) 4546 /* Don't need to rebalance while attached to NULL domain */
4547 if (time_after_eq(jiffies, rq->next_balance) &&
4548 likely(!on_null_domain(cpu)))
4180 raise_softirq(SCHED_SOFTIRQ); 4549 raise_softirq(SCHED_SOFTIRQ);
4181} 4550}
4182 4551
@@ -4490,10 +4859,7 @@ void scheduler_tick(void)
4490#endif 4859#endif
4491} 4860}
4492 4861
4493#if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \ 4862unsigned long get_parent_ip(unsigned long addr)
4494 defined(CONFIG_PREEMPT_TRACER))
4495
4496static inline unsigned long get_parent_ip(unsigned long addr)
4497{ 4863{
4498 if (in_lock_functions(addr)) { 4864 if (in_lock_functions(addr)) {
4499 addr = CALLER_ADDR2; 4865 addr = CALLER_ADDR2;
@@ -4503,6 +4869,9 @@ static inline unsigned long get_parent_ip(unsigned long addr)
4503 return addr; 4869 return addr;
4504} 4870}
4505 4871
4872#if defined(CONFIG_PREEMPT) && (defined(CONFIG_DEBUG_PREEMPT) || \
4873 defined(CONFIG_PREEMPT_TRACER))
4874
4506void __kprobes add_preempt_count(int val) 4875void __kprobes add_preempt_count(int val)
4507{ 4876{
4508#ifdef CONFIG_DEBUG_PREEMPT 4877#ifdef CONFIG_DEBUG_PREEMPT
@@ -4594,11 +4963,33 @@ static inline void schedule_debug(struct task_struct *prev)
4594#endif 4963#endif
4595} 4964}
4596 4965
4966static void put_prev_task(struct rq *rq, struct task_struct *prev)
4967{
4968 if (prev->state == TASK_RUNNING) {
4969 u64 runtime = prev->se.sum_exec_runtime;
4970
4971 runtime -= prev->se.prev_sum_exec_runtime;
4972 runtime = min_t(u64, runtime, 2*sysctl_sched_migration_cost);
4973
4974 /*
4975 * In order to avoid avg_overlap growing stale when we are
4976 * indeed overlapping and hence not getting put to sleep, grow
4977 * the avg_overlap on preemption.
4978 *
4979 * We use the average preemption runtime because that
4980 * correlates to the amount of cache footprint a task can
4981 * build up.
4982 */
4983 update_avg(&prev->se.avg_overlap, runtime);
4984 }
4985 prev->sched_class->put_prev_task(rq, prev);
4986}
4987
4597/* 4988/*
4598 * Pick up the highest-prio task: 4989 * Pick up the highest-prio task:
4599 */ 4990 */
4600static inline struct task_struct * 4991static inline struct task_struct *
4601pick_next_task(struct rq *rq, struct task_struct *prev) 4992pick_next_task(struct rq *rq)
4602{ 4993{
4603 const struct sched_class *class; 4994 const struct sched_class *class;
4604 struct task_struct *p; 4995 struct task_struct *p;
@@ -4629,15 +5020,13 @@ pick_next_task(struct rq *rq, struct task_struct *prev)
4629/* 5020/*
4630 * schedule() is the main scheduler function. 5021 * schedule() is the main scheduler function.
4631 */ 5022 */
4632asmlinkage void __sched schedule(void) 5023asmlinkage void __sched __schedule(void)
4633{ 5024{
4634 struct task_struct *prev, *next; 5025 struct task_struct *prev, *next;
4635 unsigned long *switch_count; 5026 unsigned long *switch_count;
4636 struct rq *rq; 5027 struct rq *rq;
4637 int cpu; 5028 int cpu;
4638 5029
4639need_resched:
4640 preempt_disable();
4641 cpu = smp_processor_id(); 5030 cpu = smp_processor_id();
4642 rq = cpu_rq(cpu); 5031 rq = cpu_rq(cpu);
4643 rcu_qsctr_inc(cpu); 5032 rcu_qsctr_inc(cpu);
@@ -4672,8 +5061,8 @@ need_resched_nonpreemptible:
4672 if (unlikely(!rq->nr_running)) 5061 if (unlikely(!rq->nr_running))
4673 idle_balance(cpu, rq); 5062 idle_balance(cpu, rq);
4674 5063
4675 prev->sched_class->put_prev_task(rq, prev); 5064 put_prev_task(rq, prev);
4676 next = pick_next_task(rq, prev); 5065 next = pick_next_task(rq);
4677 5066
4678 if (likely(prev != next)) { 5067 if (likely(prev != next)) {
4679 sched_info_switch(prev, next); 5068 sched_info_switch(prev, next);
@@ -4695,13 +5084,80 @@ need_resched_nonpreemptible:
4695 5084
4696 if (unlikely(reacquire_kernel_lock(current) < 0)) 5085 if (unlikely(reacquire_kernel_lock(current) < 0))
4697 goto need_resched_nonpreemptible; 5086 goto need_resched_nonpreemptible;
5087}
4698 5088
5089asmlinkage void __sched schedule(void)
5090{
5091need_resched:
5092 preempt_disable();
5093 __schedule();
4699 preempt_enable_no_resched(); 5094 preempt_enable_no_resched();
4700 if (unlikely(test_thread_flag(TIF_NEED_RESCHED))) 5095 if (unlikely(test_thread_flag(TIF_NEED_RESCHED)))
4701 goto need_resched; 5096 goto need_resched;
4702} 5097}
4703EXPORT_SYMBOL(schedule); 5098EXPORT_SYMBOL(schedule);
4704 5099
5100#ifdef CONFIG_SMP
5101/*
5102 * Look out! "owner" is an entirely speculative pointer
5103 * access and not reliable.
5104 */
5105int mutex_spin_on_owner(struct mutex *lock, struct thread_info *owner)
5106{
5107 unsigned int cpu;
5108 struct rq *rq;
5109
5110 if (!sched_feat(OWNER_SPIN))
5111 return 0;
5112
5113#ifdef CONFIG_DEBUG_PAGEALLOC
5114 /*
5115 * Need to access the cpu field knowing that
5116 * DEBUG_PAGEALLOC could have unmapped it if
5117 * the mutex owner just released it and exited.
5118 */
5119 if (probe_kernel_address(&owner->cpu, cpu))
5120 goto out;
5121#else
5122 cpu = owner->cpu;
5123#endif
5124
5125 /*
5126 * Even if the access succeeded (likely case),
5127 * the cpu field may no longer be valid.
5128 */
5129 if (cpu >= nr_cpumask_bits)
5130 goto out;
5131
5132 /*
5133 * We need to validate that we can do a
5134 * get_cpu() and that we have the percpu area.
5135 */
5136 if (!cpu_online(cpu))
5137 goto out;
5138
5139 rq = cpu_rq(cpu);
5140
5141 for (;;) {
5142 /*
5143 * Owner changed, break to re-assess state.
5144 */
5145 if (lock->owner != owner)
5146 break;
5147
5148 /*
5149 * Is that owner really running on that cpu?
5150 */
5151 if (task_thread_info(rq->curr) != owner || need_resched())
5152 return 0;
5153
5154 cpu_relax();
5155 }
5156out:
5157 return 1;
5158}
5159#endif
5160
4705#ifdef CONFIG_PREEMPT 5161#ifdef CONFIG_PREEMPT
4706/* 5162/*
4707 * this is the entry point to schedule() from in-kernel preemption 5163 * this is the entry point to schedule() from in-kernel preemption
@@ -4729,7 +5185,7 @@ asmlinkage void __sched preempt_schedule(void)
4729 * between schedule and now. 5185 * between schedule and now.
4730 */ 5186 */
4731 barrier(); 5187 barrier();
4732 } while (unlikely(test_thread_flag(TIF_NEED_RESCHED))); 5188 } while (need_resched());
4733} 5189}
4734EXPORT_SYMBOL(preempt_schedule); 5190EXPORT_SYMBOL(preempt_schedule);
4735 5191
@@ -4758,7 +5214,7 @@ asmlinkage void __sched preempt_schedule_irq(void)
4758 * between schedule and now. 5214 * between schedule and now.
4759 */ 5215 */
4760 barrier(); 5216 barrier();
4761 } while (unlikely(test_thread_flag(TIF_NEED_RESCHED))); 5217 } while (need_resched());
4762} 5218}
4763 5219
4764#endif /* CONFIG_PREEMPT */ 5220#endif /* CONFIG_PREEMPT */
@@ -4819,11 +5275,17 @@ void __wake_up_locked(wait_queue_head_t *q, unsigned int mode)
4819 __wake_up_common(q, mode, 1, 0, NULL); 5275 __wake_up_common(q, mode, 1, 0, NULL);
4820} 5276}
4821 5277
5278void __wake_up_locked_key(wait_queue_head_t *q, unsigned int mode, void *key)
5279{
5280 __wake_up_common(q, mode, 1, 0, key);
5281}
5282
4822/** 5283/**
4823 * __wake_up_sync - wake up threads blocked on a waitqueue. 5284 * __wake_up_sync_key - wake up threads blocked on a waitqueue.
4824 * @q: the waitqueue 5285 * @q: the waitqueue
4825 * @mode: which threads 5286 * @mode: which threads
4826 * @nr_exclusive: how many wake-one or wake-many threads to wake up 5287 * @nr_exclusive: how many wake-one or wake-many threads to wake up
5288 * @key: opaque value to be passed to wakeup targets
4827 * 5289 *
4828 * The sync wakeup differs that the waker knows that it will schedule 5290 * The sync wakeup differs that the waker knows that it will schedule
4829 * away soon, so while the target thread will be woken up, it will not 5291 * away soon, so while the target thread will be woken up, it will not
@@ -4832,8 +5294,8 @@ void __wake_up_locked(wait_queue_head_t *q, unsigned int mode)
4832 * 5294 *
4833 * On UP it can prevent extra preemption. 5295 * On UP it can prevent extra preemption.
4834 */ 5296 */
4835void 5297void __wake_up_sync_key(wait_queue_head_t *q, unsigned int mode,
4836__wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive) 5298 int nr_exclusive, void *key)
4837{ 5299{
4838 unsigned long flags; 5300 unsigned long flags;
4839 int sync = 1; 5301 int sync = 1;
@@ -4845,9 +5307,18 @@ __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive)
4845 sync = 0; 5307 sync = 0;
4846 5308
4847 spin_lock_irqsave(&q->lock, flags); 5309 spin_lock_irqsave(&q->lock, flags);
4848 __wake_up_common(q, mode, nr_exclusive, sync, NULL); 5310 __wake_up_common(q, mode, nr_exclusive, sync, key);
4849 spin_unlock_irqrestore(&q->lock, flags); 5311 spin_unlock_irqrestore(&q->lock, flags);
4850} 5312}
5313EXPORT_SYMBOL_GPL(__wake_up_sync_key);
5314
5315/*
5316 * __wake_up_sync - see __wake_up_sync_key()
5317 */
5318void __wake_up_sync(wait_queue_head_t *q, unsigned int mode, int nr_exclusive)
5319{
5320 __wake_up_sync_key(q, mode, nr_exclusive, NULL);
5321}
4851EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */ 5322EXPORT_SYMBOL_GPL(__wake_up_sync); /* For internal use only */
4852 5323
4853/** 5324/**
@@ -5232,7 +5703,7 @@ SYSCALL_DEFINE1(nice, int, increment)
5232 if (increment > 40) 5703 if (increment > 40)
5233 increment = 40; 5704 increment = 40;
5234 5705
5235 nice = PRIO_TO_NICE(current->static_prio) + increment; 5706 nice = TASK_NICE(current) + increment;
5236 if (nice < -20) 5707 if (nice < -20)
5237 nice = -20; 5708 nice = -20;
5238 if (nice > 19) 5709 if (nice > 19)
@@ -6505,7 +6976,7 @@ static void migrate_dead_tasks(unsigned int dead_cpu)
6505 if (!rq->nr_running) 6976 if (!rq->nr_running)
6506 break; 6977 break;
6507 update_rq_clock(rq); 6978 update_rq_clock(rq);
6508 next = pick_next_task(rq, rq->curr); 6979 next = pick_next_task(rq);
6509 if (!next) 6980 if (!next)
6510 break; 6981 break;
6511 next->sched_class->put_prev_task(rq, next); 6982 next->sched_class->put_prev_task(rq, next);
@@ -7336,7 +7807,7 @@ cpu_to_core_group(int cpu, const struct cpumask *cpu_map,
7336{ 7807{
7337 int group; 7808 int group;
7338 7809
7339 cpumask_and(mask, &per_cpu(cpu_sibling_map, cpu), cpu_map); 7810 cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map);
7340 group = cpumask_first(mask); 7811 group = cpumask_first(mask);
7341 if (sg) 7812 if (sg)
7342 *sg = &per_cpu(sched_group_core, group).sg; 7813 *sg = &per_cpu(sched_group_core, group).sg;
@@ -7365,7 +7836,7 @@ cpu_to_phys_group(int cpu, const struct cpumask *cpu_map,
7365 cpumask_and(mask, cpu_coregroup_mask(cpu), cpu_map); 7836 cpumask_and(mask, cpu_coregroup_mask(cpu), cpu_map);
7366 group = cpumask_first(mask); 7837 group = cpumask_first(mask);
7367#elif defined(CONFIG_SCHED_SMT) 7838#elif defined(CONFIG_SCHED_SMT)
7368 cpumask_and(mask, &per_cpu(cpu_sibling_map, cpu), cpu_map); 7839 cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map);
7369 group = cpumask_first(mask); 7840 group = cpumask_first(mask);
7370#else 7841#else
7371 group = cpu; 7842 group = cpu;
@@ -7708,7 +8179,7 @@ static int __build_sched_domains(const struct cpumask *cpu_map,
7708 SD_INIT(sd, SIBLING); 8179 SD_INIT(sd, SIBLING);
7709 set_domain_attribute(sd, attr); 8180 set_domain_attribute(sd, attr);
7710 cpumask_and(sched_domain_span(sd), 8181 cpumask_and(sched_domain_span(sd),
7711 &per_cpu(cpu_sibling_map, i), cpu_map); 8182 topology_thread_cpumask(i), cpu_map);
7712 sd->parent = p; 8183 sd->parent = p;
7713 p->child = sd; 8184 p->child = sd;
7714 cpu_to_cpu_group(i, cpu_map, &sd->groups, tmpmask); 8185 cpu_to_cpu_group(i, cpu_map, &sd->groups, tmpmask);
@@ -7719,7 +8190,7 @@ static int __build_sched_domains(const struct cpumask *cpu_map,
7719 /* Set up CPU (sibling) groups */ 8190 /* Set up CPU (sibling) groups */
7720 for_each_cpu(i, cpu_map) { 8191 for_each_cpu(i, cpu_map) {
7721 cpumask_and(this_sibling_map, 8192 cpumask_and(this_sibling_map,
7722 &per_cpu(cpu_sibling_map, i), cpu_map); 8193 topology_thread_cpumask(i), cpu_map);
7723 if (i != cpumask_first(this_sibling_map)) 8194 if (i != cpumask_first(this_sibling_map))
7724 continue; 8195 continue;
7725 8196
@@ -8300,11 +8771,15 @@ static void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq)
8300 __set_bit(MAX_RT_PRIO, array->bitmap); 8771 __set_bit(MAX_RT_PRIO, array->bitmap);
8301 8772
8302#if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED 8773#if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
8303 rt_rq->highest_prio = MAX_RT_PRIO; 8774 rt_rq->highest_prio.curr = MAX_RT_PRIO;
8775#ifdef CONFIG_SMP
8776 rt_rq->highest_prio.next = MAX_RT_PRIO;
8777#endif
8304#endif 8778#endif
8305#ifdef CONFIG_SMP 8779#ifdef CONFIG_SMP
8306 rt_rq->rt_nr_migratory = 0; 8780 rt_rq->rt_nr_migratory = 0;
8307 rt_rq->overloaded = 0; 8781 rt_rq->overloaded = 0;
8782 plist_head_init(&rq->rt.pushable_tasks, &rq->lock);
8308#endif 8783#endif
8309 8784
8310 rt_rq->rt_time = 0; 8785 rt_rq->rt_time = 0;
@@ -8391,6 +8866,9 @@ void __init sched_init(void)
8391#ifdef CONFIG_USER_SCHED 8866#ifdef CONFIG_USER_SCHED
8392 alloc_size *= 2; 8867 alloc_size *= 2;
8393#endif 8868#endif
8869#ifdef CONFIG_CPUMASK_OFFSTACK
8870 alloc_size += num_possible_cpus() * cpumask_size();
8871#endif
8394 /* 8872 /*
8395 * As sched_init() is called before page_alloc is setup, 8873 * As sched_init() is called before page_alloc is setup,
8396 * we use alloc_bootmem(). 8874 * we use alloc_bootmem().
@@ -8428,6 +8906,12 @@ void __init sched_init(void)
8428 ptr += nr_cpu_ids * sizeof(void **); 8906 ptr += nr_cpu_ids * sizeof(void **);
8429#endif /* CONFIG_USER_SCHED */ 8907#endif /* CONFIG_USER_SCHED */
8430#endif /* CONFIG_RT_GROUP_SCHED */ 8908#endif /* CONFIG_RT_GROUP_SCHED */
8909#ifdef CONFIG_CPUMASK_OFFSTACK
8910 for_each_possible_cpu(i) {
8911 per_cpu(load_balance_tmpmask, i) = (void *)ptr;
8912 ptr += cpumask_size();
8913 }
8914#endif /* CONFIG_CPUMASK_OFFSTACK */
8431 } 8915 }
8432 8916
8433#ifdef CONFIG_SMP 8917#ifdef CONFIG_SMP
@@ -9572,7 +10056,7 @@ cpuacct_destroy(struct cgroup_subsys *ss, struct cgroup *cgrp)
9572 10056
9573static u64 cpuacct_cpuusage_read(struct cpuacct *ca, int cpu) 10057static u64 cpuacct_cpuusage_read(struct cpuacct *ca, int cpu)
9574{ 10058{
9575 u64 *cpuusage = percpu_ptr(ca->cpuusage, cpu); 10059 u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu);
9576 u64 data; 10060 u64 data;
9577 10061
9578#ifndef CONFIG_64BIT 10062#ifndef CONFIG_64BIT
@@ -9591,7 +10075,7 @@ static u64 cpuacct_cpuusage_read(struct cpuacct *ca, int cpu)
9591 10075
9592static void cpuacct_cpuusage_write(struct cpuacct *ca, int cpu, u64 val) 10076static void cpuacct_cpuusage_write(struct cpuacct *ca, int cpu, u64 val)
9593{ 10077{
9594 u64 *cpuusage = percpu_ptr(ca->cpuusage, cpu); 10078 u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu);
9595 10079
9596#ifndef CONFIG_64BIT 10080#ifndef CONFIG_64BIT
9597 /* 10081 /*
@@ -9680,14 +10164,14 @@ static void cpuacct_charge(struct task_struct *tsk, u64 cputime)
9680 struct cpuacct *ca; 10164 struct cpuacct *ca;
9681 int cpu; 10165 int cpu;
9682 10166
9683 if (!cpuacct_subsys.active) 10167 if (unlikely(!cpuacct_subsys.active))
9684 return; 10168 return;
9685 10169
9686 cpu = task_cpu(tsk); 10170 cpu = task_cpu(tsk);
9687 ca = task_ca(tsk); 10171 ca = task_ca(tsk);
9688 10172
9689 for (; ca; ca = ca->parent) { 10173 for (; ca; ca = ca->parent) {
9690 u64 *cpuusage = percpu_ptr(ca->cpuusage, cpu); 10174 u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu);
9691 *cpuusage += cputime; 10175 *cpuusage += cputime;
9692 } 10176 }
9693} 10177}
diff --git a/kernel/sched_clock.c b/kernel/sched_clock.c
index a0b0852414cc..819f17ac796e 100644
--- a/kernel/sched_clock.c
+++ b/kernel/sched_clock.c
@@ -24,11 +24,12 @@
24 * The clock: sched_clock_cpu() is monotonic per cpu, and should be somewhat 24 * The clock: sched_clock_cpu() is monotonic per cpu, and should be somewhat
25 * consistent between cpus (never more than 2 jiffies difference). 25 * consistent between cpus (never more than 2 jiffies difference).
26 */ 26 */
27#include <linux/sched.h>
28#include <linux/percpu.h>
29#include <linux/spinlock.h> 27#include <linux/spinlock.h>
30#include <linux/ktime.h> 28#include <linux/hardirq.h>
31#include <linux/module.h> 29#include <linux/module.h>
30#include <linux/percpu.h>
31#include <linux/ktime.h>
32#include <linux/sched.h>
32 33
33/* 34/*
34 * Scheduler clock - returns current time in nanosec units. 35 * Scheduler clock - returns current time in nanosec units.
@@ -43,6 +44,7 @@ unsigned long long __attribute__((weak)) sched_clock(void)
43static __read_mostly int sched_clock_running; 44static __read_mostly int sched_clock_running;
44 45
45#ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK 46#ifdef CONFIG_HAVE_UNSTABLE_SCHED_CLOCK
47__read_mostly int sched_clock_stable;
46 48
47struct sched_clock_data { 49struct sched_clock_data {
48 /* 50 /*
@@ -87,7 +89,7 @@ void sched_clock_init(void)
87} 89}
88 90
89/* 91/*
90 * min,max except they take wrapping into account 92 * min, max except they take wrapping into account
91 */ 93 */
92 94
93static inline u64 wrap_min(u64 x, u64 y) 95static inline u64 wrap_min(u64 x, u64 y)
@@ -111,15 +113,13 @@ static u64 __update_sched_clock(struct sched_clock_data *scd, u64 now)
111 s64 delta = now - scd->tick_raw; 113 s64 delta = now - scd->tick_raw;
112 u64 clock, min_clock, max_clock; 114 u64 clock, min_clock, max_clock;
113 115
114 WARN_ON_ONCE(!irqs_disabled());
115
116 if (unlikely(delta < 0)) 116 if (unlikely(delta < 0))
117 delta = 0; 117 delta = 0;
118 118
119 /* 119 /*
120 * scd->clock = clamp(scd->tick_gtod + delta, 120 * scd->clock = clamp(scd->tick_gtod + delta,
121 * max(scd->tick_gtod, scd->clock), 121 * max(scd->tick_gtod, scd->clock),
122 * scd->tick_gtod + TICK_NSEC); 122 * scd->tick_gtod + TICK_NSEC);
123 */ 123 */
124 124
125 clock = scd->tick_gtod + delta; 125 clock = scd->tick_gtod + delta;
@@ -148,8 +148,20 @@ static void lock_double_clock(struct sched_clock_data *data1,
148 148
149u64 sched_clock_cpu(int cpu) 149u64 sched_clock_cpu(int cpu)
150{ 150{
151 struct sched_clock_data *scd = cpu_sdc(cpu);
152 u64 now, clock, this_clock, remote_clock; 151 u64 now, clock, this_clock, remote_clock;
152 struct sched_clock_data *scd;
153
154 if (sched_clock_stable)
155 return sched_clock();
156
157 scd = cpu_sdc(cpu);
158
159 /*
160 * Normally this is not called in NMI context - but if it is,
161 * trying to do any locking here is totally lethal.
162 */
163 if (unlikely(in_nmi()))
164 return scd->clock;
153 165
154 if (unlikely(!sched_clock_running)) 166 if (unlikely(!sched_clock_running))
155 return 0ull; 167 return 0ull;
@@ -195,14 +207,18 @@ u64 sched_clock_cpu(int cpu)
195 207
196void sched_clock_tick(void) 208void sched_clock_tick(void)
197{ 209{
198 struct sched_clock_data *scd = this_scd(); 210 struct sched_clock_data *scd;
199 u64 now, now_gtod; 211 u64 now, now_gtod;
200 212
213 if (sched_clock_stable)
214 return;
215
201 if (unlikely(!sched_clock_running)) 216 if (unlikely(!sched_clock_running))
202 return; 217 return;
203 218
204 WARN_ON_ONCE(!irqs_disabled()); 219 WARN_ON_ONCE(!irqs_disabled());
205 220
221 scd = this_scd();
206 now_gtod = ktime_to_ns(ktime_get()); 222 now_gtod = ktime_to_ns(ktime_get());
207 now = sched_clock(); 223 now = sched_clock();
208 224
@@ -250,7 +266,7 @@ u64 sched_clock_cpu(int cpu)
250 return sched_clock(); 266 return sched_clock();
251} 267}
252 268
253#endif 269#endif /* CONFIG_HAVE_UNSTABLE_SCHED_CLOCK */
254 270
255unsigned long long cpu_clock(int cpu) 271unsigned long long cpu_clock(int cpu)
256{ 272{
diff --git a/kernel/sched_cpupri.h b/kernel/sched_cpupri.h
index 642a94ef8a0a..9a7e859b8fbf 100644
--- a/kernel/sched_cpupri.h
+++ b/kernel/sched_cpupri.h
@@ -25,7 +25,7 @@ struct cpupri {
25 25
26#ifdef CONFIG_SMP 26#ifdef CONFIG_SMP
27int cpupri_find(struct cpupri *cp, 27int cpupri_find(struct cpupri *cp,
28 struct task_struct *p, cpumask_t *lowest_mask); 28 struct task_struct *p, struct cpumask *lowest_mask);
29void cpupri_set(struct cpupri *cp, int cpu, int pri); 29void cpupri_set(struct cpupri *cp, int cpu, int pri);
30int cpupri_init(struct cpupri *cp, bool bootmem); 30int cpupri_init(struct cpupri *cp, bool bootmem);
31void cpupri_cleanup(struct cpupri *cp); 31void cpupri_cleanup(struct cpupri *cp);
diff --git a/kernel/sched_debug.c b/kernel/sched_debug.c
index 16eeba4e4169..467ca72f1657 100644
--- a/kernel/sched_debug.c
+++ b/kernel/sched_debug.c
@@ -272,7 +272,6 @@ static void print_cpu(struct seq_file *m, int cpu)
272 P(nr_switches); 272 P(nr_switches);
273 P(nr_load_updates); 273 P(nr_load_updates);
274 P(nr_uninterruptible); 274 P(nr_uninterruptible);
275 SEQ_printf(m, " .%-30s: %lu\n", "jiffies", jiffies);
276 PN(next_balance); 275 PN(next_balance);
277 P(curr->pid); 276 P(curr->pid);
278 PN(clock); 277 PN(clock);
@@ -287,9 +286,6 @@ static void print_cpu(struct seq_file *m, int cpu)
287#ifdef CONFIG_SCHEDSTATS 286#ifdef CONFIG_SCHEDSTATS
288#define P(n) SEQ_printf(m, " .%-30s: %d\n", #n, rq->n); 287#define P(n) SEQ_printf(m, " .%-30s: %d\n", #n, rq->n);
289 288
290 P(yld_exp_empty);
291 P(yld_act_empty);
292 P(yld_both_empty);
293 P(yld_count); 289 P(yld_count);
294 290
295 P(sched_switch); 291 P(sched_switch);
@@ -314,7 +310,7 @@ static int sched_debug_show(struct seq_file *m, void *v)
314 u64 now = ktime_to_ns(ktime_get()); 310 u64 now = ktime_to_ns(ktime_get());
315 int cpu; 311 int cpu;
316 312
317 SEQ_printf(m, "Sched Debug Version: v0.08, %s %.*s\n", 313 SEQ_printf(m, "Sched Debug Version: v0.09, %s %.*s\n",
318 init_utsname()->release, 314 init_utsname()->release,
319 (int)strcspn(init_utsname()->version, " "), 315 (int)strcspn(init_utsname()->version, " "),
320 init_utsname()->version); 316 init_utsname()->version);
@@ -325,6 +321,7 @@ static int sched_debug_show(struct seq_file *m, void *v)
325 SEQ_printf(m, " .%-40s: %Ld\n", #x, (long long)(x)) 321 SEQ_printf(m, " .%-40s: %Ld\n", #x, (long long)(x))
326#define PN(x) \ 322#define PN(x) \
327 SEQ_printf(m, " .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x)) 323 SEQ_printf(m, " .%-40s: %Ld.%06ld\n", #x, SPLIT_NS(x))
324 P(jiffies);
328 PN(sysctl_sched_latency); 325 PN(sysctl_sched_latency);
329 PN(sysctl_sched_min_granularity); 326 PN(sysctl_sched_min_granularity);
330 PN(sysctl_sched_wakeup_granularity); 327 PN(sysctl_sched_wakeup_granularity);
@@ -397,6 +394,7 @@ void proc_sched_show_task(struct task_struct *p, struct seq_file *m)
397 PN(se.vruntime); 394 PN(se.vruntime);
398 PN(se.sum_exec_runtime); 395 PN(se.sum_exec_runtime);
399 PN(se.avg_overlap); 396 PN(se.avg_overlap);
397 PN(se.avg_wakeup);
400 398
401 nr_switches = p->nvcsw + p->nivcsw; 399 nr_switches = p->nvcsw + p->nivcsw;
402 400
diff --git a/kernel/sched_fair.c b/kernel/sched_fair.c
index 0566f2a03c42..3816f217f119 100644
--- a/kernel/sched_fair.c
+++ b/kernel/sched_fair.c
@@ -1314,16 +1314,63 @@ out:
1314} 1314}
1315#endif /* CONFIG_SMP */ 1315#endif /* CONFIG_SMP */
1316 1316
1317static unsigned long wakeup_gran(struct sched_entity *se) 1317/*
1318 * Adaptive granularity
1319 *
1320 * se->avg_wakeup gives the average time a task runs until it does a wakeup,
1321 * with the limit of wakeup_gran -- when it never does a wakeup.
1322 *
1323 * So the smaller avg_wakeup is the faster we want this task to preempt,
1324 * but we don't want to treat the preemptee unfairly and therefore allow it
1325 * to run for at least the amount of time we'd like to run.
1326 *
1327 * NOTE: we use 2*avg_wakeup to increase the probability of actually doing one
1328 *
1329 * NOTE: we use *nr_running to scale with load, this nicely matches the
1330 * degrading latency on load.
1331 */
1332static unsigned long
1333adaptive_gran(struct sched_entity *curr, struct sched_entity *se)
1334{
1335 u64 this_run = curr->sum_exec_runtime - curr->prev_sum_exec_runtime;
1336 u64 expected_wakeup = 2*se->avg_wakeup * cfs_rq_of(se)->nr_running;
1337 u64 gran = 0;
1338
1339 if (this_run < expected_wakeup)
1340 gran = expected_wakeup - this_run;
1341
1342 return min_t(s64, gran, sysctl_sched_wakeup_granularity);
1343}
1344
1345static unsigned long
1346wakeup_gran(struct sched_entity *curr, struct sched_entity *se)
1318{ 1347{
1319 unsigned long gran = sysctl_sched_wakeup_granularity; 1348 unsigned long gran = sysctl_sched_wakeup_granularity;
1320 1349
1350 if (cfs_rq_of(curr)->curr && sched_feat(ADAPTIVE_GRAN))
1351 gran = adaptive_gran(curr, se);
1352
1321 /* 1353 /*
1322 * More easily preempt - nice tasks, while not making it harder for 1354 * Since its curr running now, convert the gran from real-time
1323 * + nice tasks. 1355 * to virtual-time in his units.
1324 */ 1356 */
1325 if (!sched_feat(ASYM_GRAN) || se->load.weight > NICE_0_LOAD) 1357 if (sched_feat(ASYM_GRAN)) {
1326 gran = calc_delta_fair(sysctl_sched_wakeup_granularity, se); 1358 /*
1359 * By using 'se' instead of 'curr' we penalize light tasks, so
1360 * they get preempted easier. That is, if 'se' < 'curr' then
1361 * the resulting gran will be larger, therefore penalizing the
1362 * lighter, if otoh 'se' > 'curr' then the resulting gran will
1363 * be smaller, again penalizing the lighter task.
1364 *
1365 * This is especially important for buddies when the leftmost
1366 * task is higher priority than the buddy.
1367 */
1368 if (unlikely(se->load.weight != NICE_0_LOAD))
1369 gran = calc_delta_fair(gran, se);
1370 } else {
1371 if (unlikely(curr->load.weight != NICE_0_LOAD))
1372 gran = calc_delta_fair(gran, curr);
1373 }
1327 1374
1328 return gran; 1375 return gran;
1329} 1376}
@@ -1350,7 +1397,7 @@ wakeup_preempt_entity(struct sched_entity *curr, struct sched_entity *se)
1350 if (vdiff <= 0) 1397 if (vdiff <= 0)
1351 return -1; 1398 return -1;
1352 1399
1353 gran = wakeup_gran(curr); 1400 gran = wakeup_gran(curr, se);
1354 if (vdiff > gran) 1401 if (vdiff > gran)
1355 return 1; 1402 return 1;
1356 1403
diff --git a/kernel/sched_features.h b/kernel/sched_features.h
index da5d93b5d2c6..4569bfa7df9b 100644
--- a/kernel/sched_features.h
+++ b/kernel/sched_features.h
@@ -1,5 +1,6 @@
1SCHED_FEAT(NEW_FAIR_SLEEPERS, 1) 1SCHED_FEAT(NEW_FAIR_SLEEPERS, 1)
2SCHED_FEAT(NORMALIZED_SLEEPER, 1) 2SCHED_FEAT(NORMALIZED_SLEEPER, 0)
3SCHED_FEAT(ADAPTIVE_GRAN, 1)
3SCHED_FEAT(WAKEUP_PREEMPT, 1) 4SCHED_FEAT(WAKEUP_PREEMPT, 1)
4SCHED_FEAT(START_DEBIT, 1) 5SCHED_FEAT(START_DEBIT, 1)
5SCHED_FEAT(AFFINE_WAKEUPS, 1) 6SCHED_FEAT(AFFINE_WAKEUPS, 1)
@@ -13,3 +14,4 @@ SCHED_FEAT(LB_WAKEUP_UPDATE, 1)
13SCHED_FEAT(ASYM_EFF_LOAD, 1) 14SCHED_FEAT(ASYM_EFF_LOAD, 1)
14SCHED_FEAT(WAKEUP_OVERLAP, 0) 15SCHED_FEAT(WAKEUP_OVERLAP, 0)
15SCHED_FEAT(LAST_BUDDY, 1) 16SCHED_FEAT(LAST_BUDDY, 1)
17SCHED_FEAT(OWNER_SPIN, 1)
diff --git a/kernel/sched_rt.c b/kernel/sched_rt.c
index da932f4c8524..299d012b4394 100644
--- a/kernel/sched_rt.c
+++ b/kernel/sched_rt.c
@@ -3,6 +3,40 @@
3 * policies) 3 * policies)
4 */ 4 */
5 5
6static inline struct task_struct *rt_task_of(struct sched_rt_entity *rt_se)
7{
8 return container_of(rt_se, struct task_struct, rt);
9}
10
11#ifdef CONFIG_RT_GROUP_SCHED
12
13static inline struct rq *rq_of_rt_rq(struct rt_rq *rt_rq)
14{
15 return rt_rq->rq;
16}
17
18static inline struct rt_rq *rt_rq_of_se(struct sched_rt_entity *rt_se)
19{
20 return rt_se->rt_rq;
21}
22
23#else /* CONFIG_RT_GROUP_SCHED */
24
25static inline struct rq *rq_of_rt_rq(struct rt_rq *rt_rq)
26{
27 return container_of(rt_rq, struct rq, rt);
28}
29
30static inline struct rt_rq *rt_rq_of_se(struct sched_rt_entity *rt_se)
31{
32 struct task_struct *p = rt_task_of(rt_se);
33 struct rq *rq = task_rq(p);
34
35 return &rq->rt;
36}
37
38#endif /* CONFIG_RT_GROUP_SCHED */
39
6#ifdef CONFIG_SMP 40#ifdef CONFIG_SMP
7 41
8static inline int rt_overloaded(struct rq *rq) 42static inline int rt_overloaded(struct rq *rq)
@@ -37,25 +71,69 @@ static inline void rt_clear_overload(struct rq *rq)
37 cpumask_clear_cpu(rq->cpu, rq->rd->rto_mask); 71 cpumask_clear_cpu(rq->cpu, rq->rd->rto_mask);
38} 72}
39 73
40static void update_rt_migration(struct rq *rq) 74static void update_rt_migration(struct rt_rq *rt_rq)
41{ 75{
42 if (rq->rt.rt_nr_migratory && (rq->rt.rt_nr_running > 1)) { 76 if (rt_rq->rt_nr_migratory && (rt_rq->rt_nr_running > 1)) {
43 if (!rq->rt.overloaded) { 77 if (!rt_rq->overloaded) {
44 rt_set_overload(rq); 78 rt_set_overload(rq_of_rt_rq(rt_rq));
45 rq->rt.overloaded = 1; 79 rt_rq->overloaded = 1;
46 } 80 }
47 } else if (rq->rt.overloaded) { 81 } else if (rt_rq->overloaded) {
48 rt_clear_overload(rq); 82 rt_clear_overload(rq_of_rt_rq(rt_rq));
49 rq->rt.overloaded = 0; 83 rt_rq->overloaded = 0;
50 } 84 }
51} 85}
52#endif /* CONFIG_SMP */
53 86
54static inline struct task_struct *rt_task_of(struct sched_rt_entity *rt_se) 87static void inc_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
88{
89 if (rt_se->nr_cpus_allowed > 1)
90 rt_rq->rt_nr_migratory++;
91
92 update_rt_migration(rt_rq);
93}
94
95static void dec_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
96{
97 if (rt_se->nr_cpus_allowed > 1)
98 rt_rq->rt_nr_migratory--;
99
100 update_rt_migration(rt_rq);
101}
102
103static void enqueue_pushable_task(struct rq *rq, struct task_struct *p)
104{
105 plist_del(&p->pushable_tasks, &rq->rt.pushable_tasks);
106 plist_node_init(&p->pushable_tasks, p->prio);
107 plist_add(&p->pushable_tasks, &rq->rt.pushable_tasks);
108}
109
110static void dequeue_pushable_task(struct rq *rq, struct task_struct *p)
111{
112 plist_del(&p->pushable_tasks, &rq->rt.pushable_tasks);
113}
114
115#else
116
117static inline void enqueue_pushable_task(struct rq *rq, struct task_struct *p)
55{ 118{
56 return container_of(rt_se, struct task_struct, rt);
57} 119}
58 120
121static inline void dequeue_pushable_task(struct rq *rq, struct task_struct *p)
122{
123}
124
125static inline
126void inc_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
127{
128}
129
130static inline
131void dec_rt_migration(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
132{
133}
134
135#endif /* CONFIG_SMP */
136
59static inline int on_rt_rq(struct sched_rt_entity *rt_se) 137static inline int on_rt_rq(struct sched_rt_entity *rt_se)
60{ 138{
61 return !list_empty(&rt_se->run_list); 139 return !list_empty(&rt_se->run_list);
@@ -79,16 +157,6 @@ static inline u64 sched_rt_period(struct rt_rq *rt_rq)
79#define for_each_leaf_rt_rq(rt_rq, rq) \ 157#define for_each_leaf_rt_rq(rt_rq, rq) \
80 list_for_each_entry_rcu(rt_rq, &rq->leaf_rt_rq_list, leaf_rt_rq_list) 158 list_for_each_entry_rcu(rt_rq, &rq->leaf_rt_rq_list, leaf_rt_rq_list)
81 159
82static inline struct rq *rq_of_rt_rq(struct rt_rq *rt_rq)
83{
84 return rt_rq->rq;
85}
86
87static inline struct rt_rq *rt_rq_of_se(struct sched_rt_entity *rt_se)
88{
89 return rt_se->rt_rq;
90}
91
92#define for_each_sched_rt_entity(rt_se) \ 160#define for_each_sched_rt_entity(rt_se) \
93 for (; rt_se; rt_se = rt_se->parent) 161 for (; rt_se; rt_se = rt_se->parent)
94 162
@@ -108,7 +176,7 @@ static void sched_rt_rq_enqueue(struct rt_rq *rt_rq)
108 if (rt_rq->rt_nr_running) { 176 if (rt_rq->rt_nr_running) {
109 if (rt_se && !on_rt_rq(rt_se)) 177 if (rt_se && !on_rt_rq(rt_se))
110 enqueue_rt_entity(rt_se); 178 enqueue_rt_entity(rt_se);
111 if (rt_rq->highest_prio < curr->prio) 179 if (rt_rq->highest_prio.curr < curr->prio)
112 resched_task(curr); 180 resched_task(curr);
113 } 181 }
114} 182}
@@ -176,19 +244,6 @@ static inline u64 sched_rt_period(struct rt_rq *rt_rq)
176#define for_each_leaf_rt_rq(rt_rq, rq) \ 244#define for_each_leaf_rt_rq(rt_rq, rq) \
177 for (rt_rq = &rq->rt; rt_rq; rt_rq = NULL) 245 for (rt_rq = &rq->rt; rt_rq; rt_rq = NULL)
178 246
179static inline struct rq *rq_of_rt_rq(struct rt_rq *rt_rq)
180{
181 return container_of(rt_rq, struct rq, rt);
182}
183
184static inline struct rt_rq *rt_rq_of_se(struct sched_rt_entity *rt_se)
185{
186 struct task_struct *p = rt_task_of(rt_se);
187 struct rq *rq = task_rq(p);
188
189 return &rq->rt;
190}
191
192#define for_each_sched_rt_entity(rt_se) \ 247#define for_each_sched_rt_entity(rt_se) \
193 for (; rt_se; rt_se = NULL) 248 for (; rt_se; rt_se = NULL)
194 249
@@ -473,7 +528,7 @@ static inline int rt_se_prio(struct sched_rt_entity *rt_se)
473 struct rt_rq *rt_rq = group_rt_rq(rt_se); 528 struct rt_rq *rt_rq = group_rt_rq(rt_se);
474 529
475 if (rt_rq) 530 if (rt_rq)
476 return rt_rq->highest_prio; 531 return rt_rq->highest_prio.curr;
477#endif 532#endif
478 533
479 return rt_task_of(rt_se)->prio; 534 return rt_task_of(rt_se)->prio;
@@ -547,91 +602,174 @@ static void update_curr_rt(struct rq *rq)
547 } 602 }
548} 603}
549 604
550static inline 605#if defined CONFIG_SMP
551void inc_rt_tasks(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) 606
607static struct task_struct *pick_next_highest_task_rt(struct rq *rq, int cpu);
608
609static inline int next_prio(struct rq *rq)
552{ 610{
553 WARN_ON(!rt_prio(rt_se_prio(rt_se))); 611 struct task_struct *next = pick_next_highest_task_rt(rq, rq->cpu);
554 rt_rq->rt_nr_running++; 612
555#if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED 613 if (next && rt_prio(next->prio))
556 if (rt_se_prio(rt_se) < rt_rq->highest_prio) { 614 return next->prio;
557#ifdef CONFIG_SMP 615 else
558 struct rq *rq = rq_of_rt_rq(rt_rq); 616 return MAX_RT_PRIO;
559#endif 617}
618
619static void
620inc_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio)
621{
622 struct rq *rq = rq_of_rt_rq(rt_rq);
623
624 if (prio < prev_prio) {
625
626 /*
627 * If the new task is higher in priority than anything on the
628 * run-queue, we know that the previous high becomes our
629 * next-highest.
630 */
631 rt_rq->highest_prio.next = prev_prio;
560 632
561 rt_rq->highest_prio = rt_se_prio(rt_se);
562#ifdef CONFIG_SMP
563 if (rq->online) 633 if (rq->online)
564 cpupri_set(&rq->rd->cpupri, rq->cpu, 634 cpupri_set(&rq->rd->cpupri, rq->cpu, prio);
565 rt_se_prio(rt_se));
566#endif
567 }
568#endif
569#ifdef CONFIG_SMP
570 if (rt_se->nr_cpus_allowed > 1) {
571 struct rq *rq = rq_of_rt_rq(rt_rq);
572 635
573 rq->rt.rt_nr_migratory++; 636 } else if (prio == rt_rq->highest_prio.curr)
574 } 637 /*
638 * If the next task is equal in priority to the highest on
639 * the run-queue, then we implicitly know that the next highest
640 * task cannot be any lower than current
641 */
642 rt_rq->highest_prio.next = prio;
643 else if (prio < rt_rq->highest_prio.next)
644 /*
645 * Otherwise, we need to recompute next-highest
646 */
647 rt_rq->highest_prio.next = next_prio(rq);
648}
575 649
576 update_rt_migration(rq_of_rt_rq(rt_rq)); 650static void
577#endif 651dec_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio)
578#ifdef CONFIG_RT_GROUP_SCHED 652{
579 if (rt_se_boosted(rt_se)) 653 struct rq *rq = rq_of_rt_rq(rt_rq);
580 rt_rq->rt_nr_boosted++;
581 654
582 if (rt_rq->tg) 655 if (rt_rq->rt_nr_running && (prio <= rt_rq->highest_prio.next))
583 start_rt_bandwidth(&rt_rq->tg->rt_bandwidth); 656 rt_rq->highest_prio.next = next_prio(rq);
584#else 657
585 start_rt_bandwidth(&def_rt_bandwidth); 658 if (rq->online && rt_rq->highest_prio.curr != prev_prio)
586#endif 659 cpupri_set(&rq->rd->cpupri, rq->cpu, rt_rq->highest_prio.curr);
587} 660}
588 661
662#else /* CONFIG_SMP */
663
589static inline 664static inline
590void dec_rt_tasks(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) 665void inc_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio) {}
591{ 666static inline
592#ifdef CONFIG_SMP 667void dec_rt_prio_smp(struct rt_rq *rt_rq, int prio, int prev_prio) {}
593 int highest_prio = rt_rq->highest_prio; 668
594#endif 669#endif /* CONFIG_SMP */
595 670
596 WARN_ON(!rt_prio(rt_se_prio(rt_se)));
597 WARN_ON(!rt_rq->rt_nr_running);
598 rt_rq->rt_nr_running--;
599#if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED 671#if defined CONFIG_SMP || defined CONFIG_RT_GROUP_SCHED
672static void
673inc_rt_prio(struct rt_rq *rt_rq, int prio)
674{
675 int prev_prio = rt_rq->highest_prio.curr;
676
677 if (prio < prev_prio)
678 rt_rq->highest_prio.curr = prio;
679
680 inc_rt_prio_smp(rt_rq, prio, prev_prio);
681}
682
683static void
684dec_rt_prio(struct rt_rq *rt_rq, int prio)
685{
686 int prev_prio = rt_rq->highest_prio.curr;
687
600 if (rt_rq->rt_nr_running) { 688 if (rt_rq->rt_nr_running) {
601 struct rt_prio_array *array;
602 689
603 WARN_ON(rt_se_prio(rt_se) < rt_rq->highest_prio); 690 WARN_ON(prio < prev_prio);
604 if (rt_se_prio(rt_se) == rt_rq->highest_prio) { 691
605 /* recalculate */ 692 /*
606 array = &rt_rq->active; 693 * This may have been our highest task, and therefore
607 rt_rq->highest_prio = 694 * we may have some recomputation to do
695 */
696 if (prio == prev_prio) {
697 struct rt_prio_array *array = &rt_rq->active;
698
699 rt_rq->highest_prio.curr =
608 sched_find_first_bit(array->bitmap); 700 sched_find_first_bit(array->bitmap);
609 } /* otherwise leave rq->highest prio alone */ 701 }
702
610 } else 703 } else
611 rt_rq->highest_prio = MAX_RT_PRIO; 704 rt_rq->highest_prio.curr = MAX_RT_PRIO;
612#endif
613#ifdef CONFIG_SMP
614 if (rt_se->nr_cpus_allowed > 1) {
615 struct rq *rq = rq_of_rt_rq(rt_rq);
616 rq->rt.rt_nr_migratory--;
617 }
618 705
619 if (rt_rq->highest_prio != highest_prio) { 706 dec_rt_prio_smp(rt_rq, prio, prev_prio);
620 struct rq *rq = rq_of_rt_rq(rt_rq); 707}
621 708
622 if (rq->online) 709#else
623 cpupri_set(&rq->rd->cpupri, rq->cpu, 710
624 rt_rq->highest_prio); 711static inline void inc_rt_prio(struct rt_rq *rt_rq, int prio) {}
625 } 712static inline void dec_rt_prio(struct rt_rq *rt_rq, int prio) {}
713
714#endif /* CONFIG_SMP || CONFIG_RT_GROUP_SCHED */
626 715
627 update_rt_migration(rq_of_rt_rq(rt_rq));
628#endif /* CONFIG_SMP */
629#ifdef CONFIG_RT_GROUP_SCHED 716#ifdef CONFIG_RT_GROUP_SCHED
717
718static void
719inc_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
720{
721 if (rt_se_boosted(rt_se))
722 rt_rq->rt_nr_boosted++;
723
724 if (rt_rq->tg)
725 start_rt_bandwidth(&rt_rq->tg->rt_bandwidth);
726}
727
728static void
729dec_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
730{
630 if (rt_se_boosted(rt_se)) 731 if (rt_se_boosted(rt_se))
631 rt_rq->rt_nr_boosted--; 732 rt_rq->rt_nr_boosted--;
632 733
633 WARN_ON(!rt_rq->rt_nr_running && rt_rq->rt_nr_boosted); 734 WARN_ON(!rt_rq->rt_nr_running && rt_rq->rt_nr_boosted);
634#endif 735}
736
737#else /* CONFIG_RT_GROUP_SCHED */
738
739static void
740inc_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
741{
742 start_rt_bandwidth(&def_rt_bandwidth);
743}
744
745static inline
746void dec_rt_group(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq) {}
747
748#endif /* CONFIG_RT_GROUP_SCHED */
749
750static inline
751void inc_rt_tasks(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
752{
753 int prio = rt_se_prio(rt_se);
754
755 WARN_ON(!rt_prio(prio));
756 rt_rq->rt_nr_running++;
757
758 inc_rt_prio(rt_rq, prio);
759 inc_rt_migration(rt_se, rt_rq);
760 inc_rt_group(rt_se, rt_rq);
761}
762
763static inline
764void dec_rt_tasks(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
765{
766 WARN_ON(!rt_prio(rt_se_prio(rt_se)));
767 WARN_ON(!rt_rq->rt_nr_running);
768 rt_rq->rt_nr_running--;
769
770 dec_rt_prio(rt_rq, rt_se_prio(rt_se));
771 dec_rt_migration(rt_se, rt_rq);
772 dec_rt_group(rt_se, rt_rq);
635} 773}
636 774
637static void __enqueue_rt_entity(struct sched_rt_entity *rt_se) 775static void __enqueue_rt_entity(struct sched_rt_entity *rt_se)
@@ -718,6 +856,9 @@ static void enqueue_task_rt(struct rq *rq, struct task_struct *p, int wakeup)
718 856
719 enqueue_rt_entity(rt_se); 857 enqueue_rt_entity(rt_se);
720 858
859 if (!task_current(rq, p) && p->rt.nr_cpus_allowed > 1)
860 enqueue_pushable_task(rq, p);
861
721 inc_cpu_load(rq, p->se.load.weight); 862 inc_cpu_load(rq, p->se.load.weight);
722} 863}
723 864
@@ -728,6 +869,8 @@ static void dequeue_task_rt(struct rq *rq, struct task_struct *p, int sleep)
728 update_curr_rt(rq); 869 update_curr_rt(rq);
729 dequeue_rt_entity(rt_se); 870 dequeue_rt_entity(rt_se);
730 871
872 dequeue_pushable_task(rq, p);
873
731 dec_cpu_load(rq, p->se.load.weight); 874 dec_cpu_load(rq, p->se.load.weight);
732} 875}
733 876
@@ -878,7 +1021,7 @@ static struct sched_rt_entity *pick_next_rt_entity(struct rq *rq,
878 return next; 1021 return next;
879} 1022}
880 1023
881static struct task_struct *pick_next_task_rt(struct rq *rq) 1024static struct task_struct *_pick_next_task_rt(struct rq *rq)
882{ 1025{
883 struct sched_rt_entity *rt_se; 1026 struct sched_rt_entity *rt_se;
884 struct task_struct *p; 1027 struct task_struct *p;
@@ -900,6 +1043,18 @@ static struct task_struct *pick_next_task_rt(struct rq *rq)
900 1043
901 p = rt_task_of(rt_se); 1044 p = rt_task_of(rt_se);
902 p->se.exec_start = rq->clock; 1045 p->se.exec_start = rq->clock;
1046
1047 return p;
1048}
1049
1050static struct task_struct *pick_next_task_rt(struct rq *rq)
1051{
1052 struct task_struct *p = _pick_next_task_rt(rq);
1053
1054 /* The running task is never eligible for pushing */
1055 if (p)
1056 dequeue_pushable_task(rq, p);
1057
903 return p; 1058 return p;
904} 1059}
905 1060
@@ -907,6 +1062,13 @@ static void put_prev_task_rt(struct rq *rq, struct task_struct *p)
907{ 1062{
908 update_curr_rt(rq); 1063 update_curr_rt(rq);
909 p->se.exec_start = 0; 1064 p->se.exec_start = 0;
1065
1066 /*
1067 * The previous task needs to be made eligible for pushing
1068 * if it is still active
1069 */
1070 if (p->se.on_rq && p->rt.nr_cpus_allowed > 1)
1071 enqueue_pushable_task(rq, p);
910} 1072}
911 1073
912#ifdef CONFIG_SMP 1074#ifdef CONFIG_SMP
@@ -1080,7 +1242,7 @@ static struct rq *find_lock_lowest_rq(struct task_struct *task, struct rq *rq)
1080 } 1242 }
1081 1243
1082 /* If this rq is still suitable use it. */ 1244 /* If this rq is still suitable use it. */
1083 if (lowest_rq->rt.highest_prio > task->prio) 1245 if (lowest_rq->rt.highest_prio.curr > task->prio)
1084 break; 1246 break;
1085 1247
1086 /* try again */ 1248 /* try again */
@@ -1091,6 +1253,31 @@ static struct rq *find_lock_lowest_rq(struct task_struct *task, struct rq *rq)
1091 return lowest_rq; 1253 return lowest_rq;
1092} 1254}
1093 1255
1256static inline int has_pushable_tasks(struct rq *rq)
1257{
1258 return !plist_head_empty(&rq->rt.pushable_tasks);
1259}
1260
1261static struct task_struct *pick_next_pushable_task(struct rq *rq)
1262{
1263 struct task_struct *p;
1264
1265 if (!has_pushable_tasks(rq))
1266 return NULL;
1267
1268 p = plist_first_entry(&rq->rt.pushable_tasks,
1269 struct task_struct, pushable_tasks);
1270
1271 BUG_ON(rq->cpu != task_cpu(p));
1272 BUG_ON(task_current(rq, p));
1273 BUG_ON(p->rt.nr_cpus_allowed <= 1);
1274
1275 BUG_ON(!p->se.on_rq);
1276 BUG_ON(!rt_task(p));
1277
1278 return p;
1279}
1280
1094/* 1281/*
1095 * If the current CPU has more than one RT task, see if the non 1282 * If the current CPU has more than one RT task, see if the non
1096 * running task can migrate over to a CPU that is running a task 1283 * running task can migrate over to a CPU that is running a task
@@ -1100,13 +1287,11 @@ static int push_rt_task(struct rq *rq)
1100{ 1287{
1101 struct task_struct *next_task; 1288 struct task_struct *next_task;
1102 struct rq *lowest_rq; 1289 struct rq *lowest_rq;
1103 int ret = 0;
1104 int paranoid = RT_MAX_TRIES;
1105 1290
1106 if (!rq->rt.overloaded) 1291 if (!rq->rt.overloaded)
1107 return 0; 1292 return 0;
1108 1293
1109 next_task = pick_next_highest_task_rt(rq, -1); 1294 next_task = pick_next_pushable_task(rq);
1110 if (!next_task) 1295 if (!next_task)
1111 return 0; 1296 return 0;
1112 1297
@@ -1135,16 +1320,34 @@ static int push_rt_task(struct rq *rq)
1135 struct task_struct *task; 1320 struct task_struct *task;
1136 /* 1321 /*
1137 * find lock_lowest_rq releases rq->lock 1322 * find lock_lowest_rq releases rq->lock
1138 * so it is possible that next_task has changed. 1323 * so it is possible that next_task has migrated.
1139 * If it has, then try again. 1324 *
1325 * We need to make sure that the task is still on the same
1326 * run-queue and is also still the next task eligible for
1327 * pushing.
1140 */ 1328 */
1141 task = pick_next_highest_task_rt(rq, -1); 1329 task = pick_next_pushable_task(rq);
1142 if (unlikely(task != next_task) && task && paranoid--) { 1330 if (task_cpu(next_task) == rq->cpu && task == next_task) {
1143 put_task_struct(next_task); 1331 /*
1144 next_task = task; 1332 * If we get here, the task hasnt moved at all, but
1145 goto retry; 1333 * it has failed to push. We will not try again,
1334 * since the other cpus will pull from us when they
1335 * are ready.
1336 */
1337 dequeue_pushable_task(rq, next_task);
1338 goto out;
1146 } 1339 }
1147 goto out; 1340
1341 if (!task)
1342 /* No more tasks, just exit */
1343 goto out;
1344
1345 /*
1346 * Something has shifted, try again.
1347 */
1348 put_task_struct(next_task);
1349 next_task = task;
1350 goto retry;
1148 } 1351 }
1149 1352
1150 deactivate_task(rq, next_task, 0); 1353 deactivate_task(rq, next_task, 0);
@@ -1155,23 +1358,12 @@ static int push_rt_task(struct rq *rq)
1155 1358
1156 double_unlock_balance(rq, lowest_rq); 1359 double_unlock_balance(rq, lowest_rq);
1157 1360
1158 ret = 1;
1159out: 1361out:
1160 put_task_struct(next_task); 1362 put_task_struct(next_task);
1161 1363
1162 return ret; 1364 return 1;
1163} 1365}
1164 1366
1165/*
1166 * TODO: Currently we just use the second highest prio task on
1167 * the queue, and stop when it can't migrate (or there's
1168 * no more RT tasks). There may be a case where a lower
1169 * priority RT task has a different affinity than the
1170 * higher RT task. In this case the lower RT task could
1171 * possibly be able to migrate where as the higher priority
1172 * RT task could not. We currently ignore this issue.
1173 * Enhancements are welcome!
1174 */
1175static void push_rt_tasks(struct rq *rq) 1367static void push_rt_tasks(struct rq *rq)
1176{ 1368{
1177 /* push_rt_task will return true if it moved an RT */ 1369 /* push_rt_task will return true if it moved an RT */
@@ -1182,33 +1374,35 @@ static void push_rt_tasks(struct rq *rq)
1182static int pull_rt_task(struct rq *this_rq) 1374static int pull_rt_task(struct rq *this_rq)
1183{ 1375{
1184 int this_cpu = this_rq->cpu, ret = 0, cpu; 1376 int this_cpu = this_rq->cpu, ret = 0, cpu;
1185 struct task_struct *p, *next; 1377 struct task_struct *p;
1186 struct rq *src_rq; 1378 struct rq *src_rq;
1187 1379
1188 if (likely(!rt_overloaded(this_rq))) 1380 if (likely(!rt_overloaded(this_rq)))
1189 return 0; 1381 return 0;
1190 1382
1191 next = pick_next_task_rt(this_rq);
1192
1193 for_each_cpu(cpu, this_rq->rd->rto_mask) { 1383 for_each_cpu(cpu, this_rq->rd->rto_mask) {
1194 if (this_cpu == cpu) 1384 if (this_cpu == cpu)
1195 continue; 1385 continue;
1196 1386
1197 src_rq = cpu_rq(cpu); 1387 src_rq = cpu_rq(cpu);
1388
1389 /*
1390 * Don't bother taking the src_rq->lock if the next highest
1391 * task is known to be lower-priority than our current task.
1392 * This may look racy, but if this value is about to go
1393 * logically higher, the src_rq will push this task away.
1394 * And if its going logically lower, we do not care
1395 */
1396 if (src_rq->rt.highest_prio.next >=
1397 this_rq->rt.highest_prio.curr)
1398 continue;
1399
1198 /* 1400 /*
1199 * We can potentially drop this_rq's lock in 1401 * We can potentially drop this_rq's lock in
1200 * double_lock_balance, and another CPU could 1402 * double_lock_balance, and another CPU could
1201 * steal our next task - hence we must cause 1403 * alter this_rq
1202 * the caller to recalculate the next task
1203 * in that case:
1204 */ 1404 */
1205 if (double_lock_balance(this_rq, src_rq)) { 1405 double_lock_balance(this_rq, src_rq);
1206 struct task_struct *old_next = next;
1207
1208 next = pick_next_task_rt(this_rq);
1209 if (next != old_next)
1210 ret = 1;
1211 }
1212 1406
1213 /* 1407 /*
1214 * Are there still pullable RT tasks? 1408 * Are there still pullable RT tasks?
@@ -1222,7 +1416,7 @@ static int pull_rt_task(struct rq *this_rq)
1222 * Do we have an RT task that preempts 1416 * Do we have an RT task that preempts
1223 * the to-be-scheduled task? 1417 * the to-be-scheduled task?
1224 */ 1418 */
1225 if (p && (!next || (p->prio < next->prio))) { 1419 if (p && (p->prio < this_rq->rt.highest_prio.curr)) {
1226 WARN_ON(p == src_rq->curr); 1420 WARN_ON(p == src_rq->curr);
1227 WARN_ON(!p->se.on_rq); 1421 WARN_ON(!p->se.on_rq);
1228 1422
@@ -1232,12 +1426,9 @@ static int pull_rt_task(struct rq *this_rq)
1232 * This is just that p is wakeing up and hasn't 1426 * This is just that p is wakeing up and hasn't
1233 * had a chance to schedule. We only pull 1427 * had a chance to schedule. We only pull
1234 * p if it is lower in priority than the 1428 * p if it is lower in priority than the
1235 * current task on the run queue or 1429 * current task on the run queue
1236 * this_rq next task is lower in prio than
1237 * the current task on that rq.
1238 */ 1430 */
1239 if (p->prio < src_rq->curr->prio || 1431 if (p->prio < src_rq->curr->prio)
1240 (next && next->prio < src_rq->curr->prio))
1241 goto skip; 1432 goto skip;
1242 1433
1243 ret = 1; 1434 ret = 1;
@@ -1250,13 +1441,7 @@ static int pull_rt_task(struct rq *this_rq)
1250 * case there's an even higher prio task 1441 * case there's an even higher prio task
1251 * in another runqueue. (low likelyhood 1442 * in another runqueue. (low likelyhood
1252 * but possible) 1443 * but possible)
1253 *
1254 * Update next so that we won't pick a task
1255 * on another cpu with a priority lower (or equal)
1256 * than the one we just picked.
1257 */ 1444 */
1258 next = p;
1259
1260 } 1445 }
1261 skip: 1446 skip:
1262 double_unlock_balance(this_rq, src_rq); 1447 double_unlock_balance(this_rq, src_rq);
@@ -1268,24 +1453,27 @@ static int pull_rt_task(struct rq *this_rq)
1268static void pre_schedule_rt(struct rq *rq, struct task_struct *prev) 1453static void pre_schedule_rt(struct rq *rq, struct task_struct *prev)
1269{ 1454{
1270 /* Try to pull RT tasks here if we lower this rq's prio */ 1455 /* Try to pull RT tasks here if we lower this rq's prio */
1271 if (unlikely(rt_task(prev)) && rq->rt.highest_prio > prev->prio) 1456 if (unlikely(rt_task(prev)) && rq->rt.highest_prio.curr > prev->prio)
1272 pull_rt_task(rq); 1457 pull_rt_task(rq);
1273} 1458}
1274 1459
1460/*
1461 * assumes rq->lock is held
1462 */
1463static int needs_post_schedule_rt(struct rq *rq)
1464{
1465 return has_pushable_tasks(rq);
1466}
1467
1275static void post_schedule_rt(struct rq *rq) 1468static void post_schedule_rt(struct rq *rq)
1276{ 1469{
1277 /* 1470 /*
1278 * If we have more than one rt_task queued, then 1471 * This is only called if needs_post_schedule_rt() indicates that
1279 * see if we can push the other rt_tasks off to other CPUS. 1472 * we need to push tasks away
1280 * Note we may release the rq lock, and since
1281 * the lock was owned by prev, we need to release it
1282 * first via finish_lock_switch and then reaquire it here.
1283 */ 1473 */
1284 if (unlikely(rq->rt.overloaded)) { 1474 spin_lock_irq(&rq->lock);
1285 spin_lock_irq(&rq->lock); 1475 push_rt_tasks(rq);
1286 push_rt_tasks(rq); 1476 spin_unlock_irq(&rq->lock);
1287 spin_unlock_irq(&rq->lock);
1288 }
1289} 1477}
1290 1478
1291/* 1479/*
@@ -1296,7 +1484,8 @@ static void task_wake_up_rt(struct rq *rq, struct task_struct *p)
1296{ 1484{
1297 if (!task_running(rq, p) && 1485 if (!task_running(rq, p) &&
1298 !test_tsk_need_resched(rq->curr) && 1486 !test_tsk_need_resched(rq->curr) &&
1299 rq->rt.overloaded) 1487 has_pushable_tasks(rq) &&
1488 p->rt.nr_cpus_allowed > 1)
1300 push_rt_tasks(rq); 1489 push_rt_tasks(rq);
1301} 1490}
1302 1491
@@ -1332,6 +1521,24 @@ static void set_cpus_allowed_rt(struct task_struct *p,
1332 if (p->se.on_rq && (weight != p->rt.nr_cpus_allowed)) { 1521 if (p->se.on_rq && (weight != p->rt.nr_cpus_allowed)) {
1333 struct rq *rq = task_rq(p); 1522 struct rq *rq = task_rq(p);
1334 1523
1524 if (!task_current(rq, p)) {
1525 /*
1526 * Make sure we dequeue this task from the pushable list
1527 * before going further. It will either remain off of
1528 * the list because we are no longer pushable, or it
1529 * will be requeued.
1530 */
1531 if (p->rt.nr_cpus_allowed > 1)
1532 dequeue_pushable_task(rq, p);
1533
1534 /*
1535 * Requeue if our weight is changing and still > 1
1536 */
1537 if (weight > 1)
1538 enqueue_pushable_task(rq, p);
1539
1540 }
1541
1335 if ((p->rt.nr_cpus_allowed <= 1) && (weight > 1)) { 1542 if ((p->rt.nr_cpus_allowed <= 1) && (weight > 1)) {
1336 rq->rt.rt_nr_migratory++; 1543 rq->rt.rt_nr_migratory++;
1337 } else if ((p->rt.nr_cpus_allowed > 1) && (weight <= 1)) { 1544 } else if ((p->rt.nr_cpus_allowed > 1) && (weight <= 1)) {
@@ -1339,7 +1546,7 @@ static void set_cpus_allowed_rt(struct task_struct *p,
1339 rq->rt.rt_nr_migratory--; 1546 rq->rt.rt_nr_migratory--;
1340 } 1547 }
1341 1548
1342 update_rt_migration(rq); 1549 update_rt_migration(&rq->rt);
1343 } 1550 }
1344 1551
1345 cpumask_copy(&p->cpus_allowed, new_mask); 1552 cpumask_copy(&p->cpus_allowed, new_mask);
@@ -1354,7 +1561,7 @@ static void rq_online_rt(struct rq *rq)
1354 1561
1355 __enable_runtime(rq); 1562 __enable_runtime(rq);
1356 1563
1357 cpupri_set(&rq->rd->cpupri, rq->cpu, rq->rt.highest_prio); 1564 cpupri_set(&rq->rd->cpupri, rq->cpu, rq->rt.highest_prio.curr);
1358} 1565}
1359 1566
1360/* Assumes rq->lock is held */ 1567/* Assumes rq->lock is held */
@@ -1446,7 +1653,7 @@ static void prio_changed_rt(struct rq *rq, struct task_struct *p,
1446 * can release the rq lock and p could migrate. 1653 * can release the rq lock and p could migrate.
1447 * Only reschedule if p is still on the same runqueue. 1654 * Only reschedule if p is still on the same runqueue.
1448 */ 1655 */
1449 if (p->prio > rq->rt.highest_prio && rq->curr == p) 1656 if (p->prio > rq->rt.highest_prio.curr && rq->curr == p)
1450 resched_task(p); 1657 resched_task(p);
1451#else 1658#else
1452 /* For UP simply resched on drop of prio */ 1659 /* For UP simply resched on drop of prio */
@@ -1517,6 +1724,9 @@ static void set_curr_task_rt(struct rq *rq)
1517 struct task_struct *p = rq->curr; 1724 struct task_struct *p = rq->curr;
1518 1725
1519 p->se.exec_start = rq->clock; 1726 p->se.exec_start = rq->clock;
1727
1728 /* The running task is never eligible for pushing */
1729 dequeue_pushable_task(rq, p);
1520} 1730}
1521 1731
1522static const struct sched_class rt_sched_class = { 1732static const struct sched_class rt_sched_class = {
@@ -1539,6 +1749,7 @@ static const struct sched_class rt_sched_class = {
1539 .rq_online = rq_online_rt, 1749 .rq_online = rq_online_rt,
1540 .rq_offline = rq_offline_rt, 1750 .rq_offline = rq_offline_rt,
1541 .pre_schedule = pre_schedule_rt, 1751 .pre_schedule = pre_schedule_rt,
1752 .needs_post_schedule = needs_post_schedule_rt,
1542 .post_schedule = post_schedule_rt, 1753 .post_schedule = post_schedule_rt,
1543 .task_wake_up = task_wake_up_rt, 1754 .task_wake_up = task_wake_up_rt,
1544 .switched_from = switched_from_rt, 1755 .switched_from = switched_from_rt,
diff --git a/kernel/sched_stats.h b/kernel/sched_stats.h
index a8f93dd374e1..32d2bd4061b0 100644
--- a/kernel/sched_stats.h
+++ b/kernel/sched_stats.h
@@ -4,7 +4,7 @@
4 * bump this up when changing the output format or the meaning of an existing 4 * bump this up when changing the output format or the meaning of an existing
5 * format, so that tools can adapt (or abort) 5 * format, so that tools can adapt (or abort)
6 */ 6 */
7#define SCHEDSTAT_VERSION 14 7#define SCHEDSTAT_VERSION 15
8 8
9static int show_schedstat(struct seq_file *seq, void *v) 9static int show_schedstat(struct seq_file *seq, void *v)
10{ 10{
@@ -26,9 +26,8 @@ static int show_schedstat(struct seq_file *seq, void *v)
26 26
27 /* runqueue-specific stats */ 27 /* runqueue-specific stats */
28 seq_printf(seq, 28 seq_printf(seq,
29 "cpu%d %u %u %u %u %u %u %u %u %u %llu %llu %lu", 29 "cpu%d %u %u %u %u %u %u %llu %llu %lu",
30 cpu, rq->yld_both_empty, 30 cpu, rq->yld_count,
31 rq->yld_act_empty, rq->yld_exp_empty, rq->yld_count,
32 rq->sched_switch, rq->sched_count, rq->sched_goidle, 31 rq->sched_switch, rq->sched_count, rq->sched_goidle,
33 rq->ttwu_count, rq->ttwu_local, 32 rq->ttwu_count, rq->ttwu_local,
34 rq->rq_cpu_time, 33 rq->rq_cpu_time,
diff --git a/kernel/signal.c b/kernel/signal.c
index 2a74fe87c0dd..d8034737db4c 100644
--- a/kernel/signal.c
+++ b/kernel/signal.c
@@ -55,10 +55,22 @@ static int sig_handler_ignored(void __user *handler, int sig)
55 (handler == SIG_DFL && sig_kernel_ignore(sig)); 55 (handler == SIG_DFL && sig_kernel_ignore(sig));
56} 56}
57 57
58static int sig_ignored(struct task_struct *t, int sig) 58static int sig_task_ignored(struct task_struct *t, int sig,
59 int from_ancestor_ns)
59{ 60{
60 void __user *handler; 61 void __user *handler;
61 62
63 handler = sig_handler(t, sig);
64
65 if (unlikely(t->signal->flags & SIGNAL_UNKILLABLE) &&
66 handler == SIG_DFL && !from_ancestor_ns)
67 return 1;
68
69 return sig_handler_ignored(handler, sig);
70}
71
72static int sig_ignored(struct task_struct *t, int sig, int from_ancestor_ns)
73{
62 /* 74 /*
63 * Blocked signals are never ignored, since the 75 * Blocked signals are never ignored, since the
64 * signal handler may change by the time it is 76 * signal handler may change by the time it is
@@ -67,14 +79,13 @@ static int sig_ignored(struct task_struct *t, int sig)
67 if (sigismember(&t->blocked, sig) || sigismember(&t->real_blocked, sig)) 79 if (sigismember(&t->blocked, sig) || sigismember(&t->real_blocked, sig))
68 return 0; 80 return 0;
69 81
70 handler = sig_handler(t, sig); 82 if (!sig_task_ignored(t, sig, from_ancestor_ns))
71 if (!sig_handler_ignored(handler, sig))
72 return 0; 83 return 0;
73 84
74 /* 85 /*
75 * Tracers may want to know about even ignored signals. 86 * Tracers may want to know about even ignored signals.
76 */ 87 */
77 return !tracehook_consider_ignored_signal(t, sig, handler); 88 return !tracehook_consider_ignored_signal(t, sig);
78} 89}
79 90
80/* 91/*
@@ -318,7 +329,7 @@ int unhandled_signal(struct task_struct *tsk, int sig)
318 return 1; 329 return 1;
319 if (handler != SIG_IGN && handler != SIG_DFL) 330 if (handler != SIG_IGN && handler != SIG_DFL)
320 return 0; 331 return 0;
321 return !tracehook_consider_fatal_signal(tsk, sig, handler); 332 return !tracehook_consider_fatal_signal(tsk, sig);
322} 333}
323 334
324 335
@@ -624,7 +635,7 @@ static int check_kill_permission(int sig, struct siginfo *info,
624 * Returns true if the signal should be actually delivered, otherwise 635 * Returns true if the signal should be actually delivered, otherwise
625 * it should be dropped. 636 * it should be dropped.
626 */ 637 */
627static int prepare_signal(int sig, struct task_struct *p) 638static int prepare_signal(int sig, struct task_struct *p, int from_ancestor_ns)
628{ 639{
629 struct signal_struct *signal = p->signal; 640 struct signal_struct *signal = p->signal;
630 struct task_struct *t; 641 struct task_struct *t;
@@ -708,7 +719,7 @@ static int prepare_signal(int sig, struct task_struct *p)
708 } 719 }
709 } 720 }
710 721
711 return !sig_ignored(p, sig); 722 return !sig_ignored(p, sig, from_ancestor_ns);
712} 723}
713 724
714/* 725/*
@@ -777,7 +788,7 @@ static void complete_signal(int sig, struct task_struct *p, int group)
777 !(signal->flags & (SIGNAL_UNKILLABLE | SIGNAL_GROUP_EXIT)) && 788 !(signal->flags & (SIGNAL_UNKILLABLE | SIGNAL_GROUP_EXIT)) &&
778 !sigismember(&t->real_blocked, sig) && 789 !sigismember(&t->real_blocked, sig) &&
779 (sig == SIGKILL || 790 (sig == SIGKILL ||
780 !tracehook_consider_fatal_signal(t, sig, SIG_DFL))) { 791 !tracehook_consider_fatal_signal(t, sig))) {
781 /* 792 /*
782 * This signal will be fatal to the whole group. 793 * This signal will be fatal to the whole group.
783 */ 794 */
@@ -813,8 +824,8 @@ static inline int legacy_queue(struct sigpending *signals, int sig)
813 return (sig < SIGRTMIN) && sigismember(&signals->signal, sig); 824 return (sig < SIGRTMIN) && sigismember(&signals->signal, sig);
814} 825}
815 826
816static int send_signal(int sig, struct siginfo *info, struct task_struct *t, 827static int __send_signal(int sig, struct siginfo *info, struct task_struct *t,
817 int group) 828 int group, int from_ancestor_ns)
818{ 829{
819 struct sigpending *pending; 830 struct sigpending *pending;
820 struct sigqueue *q; 831 struct sigqueue *q;
@@ -822,7 +833,8 @@ static int send_signal(int sig, struct siginfo *info, struct task_struct *t,
822 trace_sched_signal_send(sig, t); 833 trace_sched_signal_send(sig, t);
823 834
824 assert_spin_locked(&t->sighand->siglock); 835 assert_spin_locked(&t->sighand->siglock);
825 if (!prepare_signal(sig, t)) 836
837 if (!prepare_signal(sig, t, from_ancestor_ns))
826 return 0; 838 return 0;
827 839
828 pending = group ? &t->signal->shared_pending : &t->pending; 840 pending = group ? &t->signal->shared_pending : &t->pending;
@@ -871,6 +883,8 @@ static int send_signal(int sig, struct siginfo *info, struct task_struct *t,
871 break; 883 break;
872 default: 884 default:
873 copy_siginfo(&q->info, info); 885 copy_siginfo(&q->info, info);
886 if (from_ancestor_ns)
887 q->info.si_pid = 0;
874 break; 888 break;
875 } 889 }
876 } else if (!is_si_special(info)) { 890 } else if (!is_si_special(info)) {
@@ -889,6 +903,20 @@ out_set:
889 return 0; 903 return 0;
890} 904}
891 905
906static int send_signal(int sig, struct siginfo *info, struct task_struct *t,
907 int group)
908{
909 int from_ancestor_ns = 0;
910
911#ifdef CONFIG_PID_NS
912 if (!is_si_special(info) && SI_FROMUSER(info) &&
913 task_pid_nr_ns(current, task_active_pid_ns(t)) <= 0)
914 from_ancestor_ns = 1;
915#endif
916
917 return __send_signal(sig, info, t, group, from_ancestor_ns);
918}
919
892int print_fatal_signals; 920int print_fatal_signals;
893 921
894static void print_fatal_signal(struct pt_regs *regs, int signr) 922static void print_fatal_signal(struct pt_regs *regs, int signr)
@@ -1133,7 +1161,7 @@ int kill_pid_info_as_uid(int sig, struct siginfo *info, struct pid *pid,
1133 if (sig && p->sighand) { 1161 if (sig && p->sighand) {
1134 unsigned long flags; 1162 unsigned long flags;
1135 spin_lock_irqsave(&p->sighand->siglock, flags); 1163 spin_lock_irqsave(&p->sighand->siglock, flags);
1136 ret = __group_send_sig_info(sig, info, p); 1164 ret = __send_signal(sig, info, p, 1, 0);
1137 spin_unlock_irqrestore(&p->sighand->siglock, flags); 1165 spin_unlock_irqrestore(&p->sighand->siglock, flags);
1138 } 1166 }
1139out_unlock: 1167out_unlock:
@@ -1320,7 +1348,7 @@ int send_sigqueue(struct sigqueue *q, struct task_struct *t, int group)
1320 goto ret; 1348 goto ret;
1321 1349
1322 ret = 1; /* the signal is ignored */ 1350 ret = 1; /* the signal is ignored */
1323 if (!prepare_signal(sig, t)) 1351 if (!prepare_signal(sig, t, 0))
1324 goto out; 1352 goto out;
1325 1353
1326 ret = 0; 1354 ret = 0;
@@ -1575,7 +1603,15 @@ static void ptrace_stop(int exit_code, int clear_code, siginfo_t *info)
1575 read_lock(&tasklist_lock); 1603 read_lock(&tasklist_lock);
1576 if (may_ptrace_stop()) { 1604 if (may_ptrace_stop()) {
1577 do_notify_parent_cldstop(current, CLD_TRAPPED); 1605 do_notify_parent_cldstop(current, CLD_TRAPPED);
1606 /*
1607 * Don't want to allow preemption here, because
1608 * sys_ptrace() needs this task to be inactive.
1609 *
1610 * XXX: implement read_unlock_no_resched().
1611 */
1612 preempt_disable();
1578 read_unlock(&tasklist_lock); 1613 read_unlock(&tasklist_lock);
1614 preempt_enable_no_resched();
1579 schedule(); 1615 schedule();
1580 } else { 1616 } else {
1581 /* 1617 /*
@@ -1836,9 +1872,16 @@ relock:
1836 1872
1837 /* 1873 /*
1838 * Global init gets no signals it doesn't want. 1874 * Global init gets no signals it doesn't want.
1875 * Container-init gets no signals it doesn't want from same
1876 * container.
1877 *
1878 * Note that if global/container-init sees a sig_kernel_only()
1879 * signal here, the signal must have been generated internally
1880 * or must have come from an ancestor namespace. In either
1881 * case, the signal cannot be dropped.
1839 */ 1882 */
1840 if (unlikely(signal->flags & SIGNAL_UNKILLABLE) && 1883 if (unlikely(signal->flags & SIGNAL_UNKILLABLE) &&
1841 !signal_group_exit(signal)) 1884 !sig_kernel_only(signr))
1842 continue; 1885 continue;
1843 1886
1844 if (sig_kernel_stop(signr)) { 1887 if (sig_kernel_stop(signr)) {
diff --git a/kernel/slow-work.c b/kernel/slow-work.c
new file mode 100644
index 000000000000..cf2bc01186ef
--- /dev/null
+++ b/kernel/slow-work.c
@@ -0,0 +1,640 @@
1/* Worker thread pool for slow items, such as filesystem lookups or mkdirs
2 *
3 * Copyright (C) 2008 Red Hat, Inc. All Rights Reserved.
4 * Written by David Howells (dhowells@redhat.com)
5 *
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public Licence
8 * as published by the Free Software Foundation; either version
9 * 2 of the Licence, or (at your option) any later version.
10 *
11 * See Documentation/slow-work.txt
12 */
13
14#include <linux/module.h>
15#include <linux/slow-work.h>
16#include <linux/kthread.h>
17#include <linux/freezer.h>
18#include <linux/wait.h>
19
20#define SLOW_WORK_CULL_TIMEOUT (5 * HZ) /* cull threads 5s after running out of
21 * things to do */
22#define SLOW_WORK_OOM_TIMEOUT (5 * HZ) /* can't start new threads for 5s after
23 * OOM */
24
25static void slow_work_cull_timeout(unsigned long);
26static void slow_work_oom_timeout(unsigned long);
27
28#ifdef CONFIG_SYSCTL
29static int slow_work_min_threads_sysctl(struct ctl_table *, int, struct file *,
30 void __user *, size_t *, loff_t *);
31
32static int slow_work_max_threads_sysctl(struct ctl_table *, int , struct file *,
33 void __user *, size_t *, loff_t *);
34#endif
35
36/*
37 * The pool of threads has at least min threads in it as long as someone is
38 * using the facility, and may have as many as max.
39 *
40 * A portion of the pool may be processing very slow operations.
41 */
42static unsigned slow_work_min_threads = 2;
43static unsigned slow_work_max_threads = 4;
44static unsigned vslow_work_proportion = 50; /* % of threads that may process
45 * very slow work */
46
47#ifdef CONFIG_SYSCTL
48static const int slow_work_min_min_threads = 2;
49static int slow_work_max_max_threads = 255;
50static const int slow_work_min_vslow = 1;
51static const int slow_work_max_vslow = 99;
52
53ctl_table slow_work_sysctls[] = {
54 {
55 .ctl_name = CTL_UNNUMBERED,
56 .procname = "min-threads",
57 .data = &slow_work_min_threads,
58 .maxlen = sizeof(unsigned),
59 .mode = 0644,
60 .proc_handler = slow_work_min_threads_sysctl,
61 .extra1 = (void *) &slow_work_min_min_threads,
62 .extra2 = &slow_work_max_threads,
63 },
64 {
65 .ctl_name = CTL_UNNUMBERED,
66 .procname = "max-threads",
67 .data = &slow_work_max_threads,
68 .maxlen = sizeof(unsigned),
69 .mode = 0644,
70 .proc_handler = slow_work_max_threads_sysctl,
71 .extra1 = &slow_work_min_threads,
72 .extra2 = (void *) &slow_work_max_max_threads,
73 },
74 {
75 .ctl_name = CTL_UNNUMBERED,
76 .procname = "vslow-percentage",
77 .data = &vslow_work_proportion,
78 .maxlen = sizeof(unsigned),
79 .mode = 0644,
80 .proc_handler = &proc_dointvec_minmax,
81 .extra1 = (void *) &slow_work_min_vslow,
82 .extra2 = (void *) &slow_work_max_vslow,
83 },
84 { .ctl_name = 0 }
85};
86#endif
87
88/*
89 * The active state of the thread pool
90 */
91static atomic_t slow_work_thread_count;
92static atomic_t vslow_work_executing_count;
93
94static bool slow_work_may_not_start_new_thread;
95static bool slow_work_cull; /* cull a thread due to lack of activity */
96static DEFINE_TIMER(slow_work_cull_timer, slow_work_cull_timeout, 0, 0);
97static DEFINE_TIMER(slow_work_oom_timer, slow_work_oom_timeout, 0, 0);
98static struct slow_work slow_work_new_thread; /* new thread starter */
99
100/*
101 * The queues of work items and the lock governing access to them. These are
102 * shared between all the CPUs. It doesn't make sense to have per-CPU queues
103 * as the number of threads bears no relation to the number of CPUs.
104 *
105 * There are two queues of work items: one for slow work items, and one for
106 * very slow work items.
107 */
108static LIST_HEAD(slow_work_queue);
109static LIST_HEAD(vslow_work_queue);
110static DEFINE_SPINLOCK(slow_work_queue_lock);
111
112/*
113 * The thread controls. A variable used to signal to the threads that they
114 * should exit when the queue is empty, a waitqueue used by the threads to wait
115 * for signals, and a completion set by the last thread to exit.
116 */
117static bool slow_work_threads_should_exit;
118static DECLARE_WAIT_QUEUE_HEAD(slow_work_thread_wq);
119static DECLARE_COMPLETION(slow_work_last_thread_exited);
120
121/*
122 * The number of users of the thread pool and its lock. Whilst this is zero we
123 * have no threads hanging around, and when this reaches zero, we wait for all
124 * active or queued work items to complete and kill all the threads we do have.
125 */
126static int slow_work_user_count;
127static DEFINE_MUTEX(slow_work_user_lock);
128
129/*
130 * Calculate the maximum number of active threads in the pool that are
131 * permitted to process very slow work items.
132 *
133 * The answer is rounded up to at least 1, but may not equal or exceed the
134 * maximum number of the threads in the pool. This means we always have at
135 * least one thread that can process slow work items, and we always have at
136 * least one thread that won't get tied up doing so.
137 */
138static unsigned slow_work_calc_vsmax(void)
139{
140 unsigned vsmax;
141
142 vsmax = atomic_read(&slow_work_thread_count) * vslow_work_proportion;
143 vsmax /= 100;
144 vsmax = max(vsmax, 1U);
145 return min(vsmax, slow_work_max_threads - 1);
146}
147
148/*
149 * Attempt to execute stuff queued on a slow thread. Return true if we managed
150 * it, false if there was nothing to do.
151 */
152static bool slow_work_execute(void)
153{
154 struct slow_work *work = NULL;
155 unsigned vsmax;
156 bool very_slow;
157
158 vsmax = slow_work_calc_vsmax();
159
160 /* see if we can schedule a new thread to be started if we're not
161 * keeping up with the work */
162 if (!waitqueue_active(&slow_work_thread_wq) &&
163 (!list_empty(&slow_work_queue) || !list_empty(&vslow_work_queue)) &&
164 atomic_read(&slow_work_thread_count) < slow_work_max_threads &&
165 !slow_work_may_not_start_new_thread)
166 slow_work_enqueue(&slow_work_new_thread);
167
168 /* find something to execute */
169 spin_lock_irq(&slow_work_queue_lock);
170 if (!list_empty(&vslow_work_queue) &&
171 atomic_read(&vslow_work_executing_count) < vsmax) {
172 work = list_entry(vslow_work_queue.next,
173 struct slow_work, link);
174 if (test_and_set_bit_lock(SLOW_WORK_EXECUTING, &work->flags))
175 BUG();
176 list_del_init(&work->link);
177 atomic_inc(&vslow_work_executing_count);
178 very_slow = true;
179 } else if (!list_empty(&slow_work_queue)) {
180 work = list_entry(slow_work_queue.next,
181 struct slow_work, link);
182 if (test_and_set_bit_lock(SLOW_WORK_EXECUTING, &work->flags))
183 BUG();
184 list_del_init(&work->link);
185 very_slow = false;
186 } else {
187 very_slow = false; /* avoid the compiler warning */
188 }
189 spin_unlock_irq(&slow_work_queue_lock);
190
191 if (!work)
192 return false;
193
194 if (!test_and_clear_bit(SLOW_WORK_PENDING, &work->flags))
195 BUG();
196
197 work->ops->execute(work);
198
199 if (very_slow)
200 atomic_dec(&vslow_work_executing_count);
201 clear_bit_unlock(SLOW_WORK_EXECUTING, &work->flags);
202
203 /* if someone tried to enqueue the item whilst we were executing it,
204 * then it'll be left unenqueued to avoid multiple threads trying to
205 * execute it simultaneously
206 *
207 * there is, however, a race between us testing the pending flag and
208 * getting the spinlock, and between the enqueuer setting the pending
209 * flag and getting the spinlock, so we use a deferral bit to tell us
210 * if the enqueuer got there first
211 */
212 if (test_bit(SLOW_WORK_PENDING, &work->flags)) {
213 spin_lock_irq(&slow_work_queue_lock);
214
215 if (!test_bit(SLOW_WORK_EXECUTING, &work->flags) &&
216 test_and_clear_bit(SLOW_WORK_ENQ_DEFERRED, &work->flags))
217 goto auto_requeue;
218
219 spin_unlock_irq(&slow_work_queue_lock);
220 }
221
222 work->ops->put_ref(work);
223 return true;
224
225auto_requeue:
226 /* we must complete the enqueue operation
227 * - we transfer our ref on the item back to the appropriate queue
228 * - don't wake another thread up as we're awake already
229 */
230 if (test_bit(SLOW_WORK_VERY_SLOW, &work->flags))
231 list_add_tail(&work->link, &vslow_work_queue);
232 else
233 list_add_tail(&work->link, &slow_work_queue);
234 spin_unlock_irq(&slow_work_queue_lock);
235 return true;
236}
237
238/**
239 * slow_work_enqueue - Schedule a slow work item for processing
240 * @work: The work item to queue
241 *
242 * Schedule a slow work item for processing. If the item is already undergoing
243 * execution, this guarantees not to re-enter the execution routine until the
244 * first execution finishes.
245 *
246 * The item is pinned by this function as it retains a reference to it, managed
247 * through the item operations. The item is unpinned once it has been
248 * executed.
249 *
250 * An item may hog the thread that is running it for a relatively large amount
251 * of time, sufficient, for example, to perform several lookup, mkdir, create
252 * and setxattr operations. It may sleep on I/O and may sleep to obtain locks.
253 *
254 * Conversely, if a number of items are awaiting processing, it may take some
255 * time before any given item is given attention. The number of threads in the
256 * pool may be increased to deal with demand, but only up to a limit.
257 *
258 * If SLOW_WORK_VERY_SLOW is set on the work item, then it will be placed in
259 * the very slow queue, from which only a portion of the threads will be
260 * allowed to pick items to execute. This ensures that very slow items won't
261 * overly block ones that are just ordinarily slow.
262 *
263 * Returns 0 if successful, -EAGAIN if not.
264 */
265int slow_work_enqueue(struct slow_work *work)
266{
267 unsigned long flags;
268
269 BUG_ON(slow_work_user_count <= 0);
270 BUG_ON(!work);
271 BUG_ON(!work->ops);
272 BUG_ON(!work->ops->get_ref);
273
274 /* when honouring an enqueue request, we only promise that we will run
275 * the work function in the future; we do not promise to run it once
276 * per enqueue request
277 *
278 * we use the PENDING bit to merge together repeat requests without
279 * having to disable IRQs and take the spinlock, whilst still
280 * maintaining our promise
281 */
282 if (!test_and_set_bit_lock(SLOW_WORK_PENDING, &work->flags)) {
283 spin_lock_irqsave(&slow_work_queue_lock, flags);
284
285 /* we promise that we will not attempt to execute the work
286 * function in more than one thread simultaneously
287 *
288 * this, however, leaves us with a problem if we're asked to
289 * enqueue the work whilst someone is executing the work
290 * function as simply queueing the work immediately means that
291 * another thread may try executing it whilst it is already
292 * under execution
293 *
294 * to deal with this, we set the ENQ_DEFERRED bit instead of
295 * enqueueing, and the thread currently executing the work
296 * function will enqueue the work item when the work function
297 * returns and it has cleared the EXECUTING bit
298 */
299 if (test_bit(SLOW_WORK_EXECUTING, &work->flags)) {
300 set_bit(SLOW_WORK_ENQ_DEFERRED, &work->flags);
301 } else {
302 if (work->ops->get_ref(work) < 0)
303 goto cant_get_ref;
304 if (test_bit(SLOW_WORK_VERY_SLOW, &work->flags))
305 list_add_tail(&work->link, &vslow_work_queue);
306 else
307 list_add_tail(&work->link, &slow_work_queue);
308 wake_up(&slow_work_thread_wq);
309 }
310
311 spin_unlock_irqrestore(&slow_work_queue_lock, flags);
312 }
313 return 0;
314
315cant_get_ref:
316 spin_unlock_irqrestore(&slow_work_queue_lock, flags);
317 return -EAGAIN;
318}
319EXPORT_SYMBOL(slow_work_enqueue);
320
321/*
322 * Worker thread culling algorithm
323 */
324static bool slow_work_cull_thread(void)
325{
326 unsigned long flags;
327 bool do_cull = false;
328
329 spin_lock_irqsave(&slow_work_queue_lock, flags);
330
331 if (slow_work_cull) {
332 slow_work_cull = false;
333
334 if (list_empty(&slow_work_queue) &&
335 list_empty(&vslow_work_queue) &&
336 atomic_read(&slow_work_thread_count) >
337 slow_work_min_threads) {
338 mod_timer(&slow_work_cull_timer,
339 jiffies + SLOW_WORK_CULL_TIMEOUT);
340 do_cull = true;
341 }
342 }
343
344 spin_unlock_irqrestore(&slow_work_queue_lock, flags);
345 return do_cull;
346}
347
348/*
349 * Determine if there is slow work available for dispatch
350 */
351static inline bool slow_work_available(int vsmax)
352{
353 return !list_empty(&slow_work_queue) ||
354 (!list_empty(&vslow_work_queue) &&
355 atomic_read(&vslow_work_executing_count) < vsmax);
356}
357
358/*
359 * Worker thread dispatcher
360 */
361static int slow_work_thread(void *_data)
362{
363 int vsmax;
364
365 DEFINE_WAIT(wait);
366
367 set_freezable();
368 set_user_nice(current, -5);
369
370 for (;;) {
371 vsmax = vslow_work_proportion;
372 vsmax *= atomic_read(&slow_work_thread_count);
373 vsmax /= 100;
374
375 prepare_to_wait(&slow_work_thread_wq, &wait,
376 TASK_INTERRUPTIBLE);
377 if (!freezing(current) &&
378 !slow_work_threads_should_exit &&
379 !slow_work_available(vsmax) &&
380 !slow_work_cull)
381 schedule();
382 finish_wait(&slow_work_thread_wq, &wait);
383
384 try_to_freeze();
385
386 vsmax = vslow_work_proportion;
387 vsmax *= atomic_read(&slow_work_thread_count);
388 vsmax /= 100;
389
390 if (slow_work_available(vsmax) && slow_work_execute()) {
391 cond_resched();
392 if (list_empty(&slow_work_queue) &&
393 list_empty(&vslow_work_queue) &&
394 atomic_read(&slow_work_thread_count) >
395 slow_work_min_threads)
396 mod_timer(&slow_work_cull_timer,
397 jiffies + SLOW_WORK_CULL_TIMEOUT);
398 continue;
399 }
400
401 if (slow_work_threads_should_exit)
402 break;
403
404 if (slow_work_cull && slow_work_cull_thread())
405 break;
406 }
407
408 if (atomic_dec_and_test(&slow_work_thread_count))
409 complete_and_exit(&slow_work_last_thread_exited, 0);
410 return 0;
411}
412
413/*
414 * Handle thread cull timer expiration
415 */
416static void slow_work_cull_timeout(unsigned long data)
417{
418 slow_work_cull = true;
419 wake_up(&slow_work_thread_wq);
420}
421
422/*
423 * Get a reference on slow work thread starter
424 */
425static int slow_work_new_thread_get_ref(struct slow_work *work)
426{
427 return 0;
428}
429
430/*
431 * Drop a reference on slow work thread starter
432 */
433static void slow_work_new_thread_put_ref(struct slow_work *work)
434{
435}
436
437/*
438 * Start a new slow work thread
439 */
440static void slow_work_new_thread_execute(struct slow_work *work)
441{
442 struct task_struct *p;
443
444 if (slow_work_threads_should_exit)
445 return;
446
447 if (atomic_read(&slow_work_thread_count) >= slow_work_max_threads)
448 return;
449
450 if (!mutex_trylock(&slow_work_user_lock))
451 return;
452
453 slow_work_may_not_start_new_thread = true;
454 atomic_inc(&slow_work_thread_count);
455 p = kthread_run(slow_work_thread, NULL, "kslowd");
456 if (IS_ERR(p)) {
457 printk(KERN_DEBUG "Slow work thread pool: OOM\n");
458 if (atomic_dec_and_test(&slow_work_thread_count))
459 BUG(); /* we're running on a slow work thread... */
460 mod_timer(&slow_work_oom_timer,
461 jiffies + SLOW_WORK_OOM_TIMEOUT);
462 } else {
463 /* ratelimit the starting of new threads */
464 mod_timer(&slow_work_oom_timer, jiffies + 1);
465 }
466
467 mutex_unlock(&slow_work_user_lock);
468}
469
470static const struct slow_work_ops slow_work_new_thread_ops = {
471 .get_ref = slow_work_new_thread_get_ref,
472 .put_ref = slow_work_new_thread_put_ref,
473 .execute = slow_work_new_thread_execute,
474};
475
476/*
477 * post-OOM new thread start suppression expiration
478 */
479static void slow_work_oom_timeout(unsigned long data)
480{
481 slow_work_may_not_start_new_thread = false;
482}
483
484#ifdef CONFIG_SYSCTL
485/*
486 * Handle adjustment of the minimum number of threads
487 */
488static int slow_work_min_threads_sysctl(struct ctl_table *table, int write,
489 struct file *filp, void __user *buffer,
490 size_t *lenp, loff_t *ppos)
491{
492 int ret = proc_dointvec_minmax(table, write, filp, buffer, lenp, ppos);
493 int n;
494
495 if (ret == 0) {
496 mutex_lock(&slow_work_user_lock);
497 if (slow_work_user_count > 0) {
498 /* see if we need to start or stop threads */
499 n = atomic_read(&slow_work_thread_count) -
500 slow_work_min_threads;
501
502 if (n < 0 && !slow_work_may_not_start_new_thread)
503 slow_work_enqueue(&slow_work_new_thread);
504 else if (n > 0)
505 mod_timer(&slow_work_cull_timer,
506 jiffies + SLOW_WORK_CULL_TIMEOUT);
507 }
508 mutex_unlock(&slow_work_user_lock);
509 }
510
511 return ret;
512}
513
514/*
515 * Handle adjustment of the maximum number of threads
516 */
517static int slow_work_max_threads_sysctl(struct ctl_table *table, int write,
518 struct file *filp, void __user *buffer,
519 size_t *lenp, loff_t *ppos)
520{
521 int ret = proc_dointvec_minmax(table, write, filp, buffer, lenp, ppos);
522 int n;
523
524 if (ret == 0) {
525 mutex_lock(&slow_work_user_lock);
526 if (slow_work_user_count > 0) {
527 /* see if we need to stop threads */
528 n = slow_work_max_threads -
529 atomic_read(&slow_work_thread_count);
530
531 if (n < 0)
532 mod_timer(&slow_work_cull_timer,
533 jiffies + SLOW_WORK_CULL_TIMEOUT);
534 }
535 mutex_unlock(&slow_work_user_lock);
536 }
537
538 return ret;
539}
540#endif /* CONFIG_SYSCTL */
541
542/**
543 * slow_work_register_user - Register a user of the facility
544 *
545 * Register a user of the facility, starting up the initial threads if there
546 * aren't any other users at this point. This will return 0 if successful, or
547 * an error if not.
548 */
549int slow_work_register_user(void)
550{
551 struct task_struct *p;
552 int loop;
553
554 mutex_lock(&slow_work_user_lock);
555
556 if (slow_work_user_count == 0) {
557 printk(KERN_NOTICE "Slow work thread pool: Starting up\n");
558 init_completion(&slow_work_last_thread_exited);
559
560 slow_work_threads_should_exit = false;
561 slow_work_init(&slow_work_new_thread,
562 &slow_work_new_thread_ops);
563 slow_work_may_not_start_new_thread = false;
564 slow_work_cull = false;
565
566 /* start the minimum number of threads */
567 for (loop = 0; loop < slow_work_min_threads; loop++) {
568 atomic_inc(&slow_work_thread_count);
569 p = kthread_run(slow_work_thread, NULL, "kslowd");
570 if (IS_ERR(p))
571 goto error;
572 }
573 printk(KERN_NOTICE "Slow work thread pool: Ready\n");
574 }
575
576 slow_work_user_count++;
577 mutex_unlock(&slow_work_user_lock);
578 return 0;
579
580error:
581 if (atomic_dec_and_test(&slow_work_thread_count))
582 complete(&slow_work_last_thread_exited);
583 if (loop > 0) {
584 printk(KERN_ERR "Slow work thread pool:"
585 " Aborting startup on ENOMEM\n");
586 slow_work_threads_should_exit = true;
587 wake_up_all(&slow_work_thread_wq);
588 wait_for_completion(&slow_work_last_thread_exited);
589 printk(KERN_ERR "Slow work thread pool: Aborted\n");
590 }
591 mutex_unlock(&slow_work_user_lock);
592 return PTR_ERR(p);
593}
594EXPORT_SYMBOL(slow_work_register_user);
595
596/**
597 * slow_work_unregister_user - Unregister a user of the facility
598 *
599 * Unregister a user of the facility, killing all the threads if this was the
600 * last one.
601 */
602void slow_work_unregister_user(void)
603{
604 mutex_lock(&slow_work_user_lock);
605
606 BUG_ON(slow_work_user_count <= 0);
607
608 slow_work_user_count--;
609 if (slow_work_user_count == 0) {
610 printk(KERN_NOTICE "Slow work thread pool: Shutting down\n");
611 slow_work_threads_should_exit = true;
612 wake_up_all(&slow_work_thread_wq);
613 wait_for_completion(&slow_work_last_thread_exited);
614 printk(KERN_NOTICE "Slow work thread pool:"
615 " Shut down complete\n");
616 }
617
618 del_timer_sync(&slow_work_cull_timer);
619
620 mutex_unlock(&slow_work_user_lock);
621}
622EXPORT_SYMBOL(slow_work_unregister_user);
623
624/*
625 * Initialise the slow work facility
626 */
627static int __init init_slow_work(void)
628{
629 unsigned nr_cpus = num_possible_cpus();
630
631 if (slow_work_max_threads < nr_cpus)
632 slow_work_max_threads = nr_cpus;
633#ifdef CONFIG_SYSCTL
634 if (slow_work_max_max_threads < nr_cpus * 2)
635 slow_work_max_max_threads = nr_cpus * 2;
636#endif
637 return 0;
638}
639
640subsys_initcall(init_slow_work);
diff --git a/kernel/smp.c b/kernel/smp.c
index bbedbb7efe32..858baac568ee 100644
--- a/kernel/smp.c
+++ b/kernel/smp.c
@@ -2,40 +2,82 @@
2 * Generic helpers for smp ipi calls 2 * Generic helpers for smp ipi calls
3 * 3 *
4 * (C) Jens Axboe <jens.axboe@oracle.com> 2008 4 * (C) Jens Axboe <jens.axboe@oracle.com> 2008
5 *
6 */ 5 */
7#include <linux/init.h>
8#include <linux/module.h>
9#include <linux/percpu.h>
10#include <linux/rcupdate.h> 6#include <linux/rcupdate.h>
11#include <linux/rculist.h> 7#include <linux/rculist.h>
8#include <linux/kernel.h>
9#include <linux/module.h>
10#include <linux/percpu.h>
11#include <linux/init.h>
12#include <linux/smp.h> 12#include <linux/smp.h>
13#include <linux/cpu.h>
13 14
14static DEFINE_PER_CPU(struct call_single_queue, call_single_queue); 15static DEFINE_PER_CPU(struct call_single_queue, call_single_queue);
15static LIST_HEAD(call_function_queue); 16
16__cacheline_aligned_in_smp DEFINE_SPINLOCK(call_function_lock); 17static struct {
18 struct list_head queue;
19 spinlock_t lock;
20} call_function __cacheline_aligned_in_smp =
21 {
22 .queue = LIST_HEAD_INIT(call_function.queue),
23 .lock = __SPIN_LOCK_UNLOCKED(call_function.lock),
24 };
17 25
18enum { 26enum {
19 CSD_FLAG_WAIT = 0x01, 27 CSD_FLAG_LOCK = 0x01,
20 CSD_FLAG_ALLOC = 0x02,
21 CSD_FLAG_LOCK = 0x04,
22}; 28};
23 29
24struct call_function_data { 30struct call_function_data {
25 struct call_single_data csd; 31 struct call_single_data csd;
26 spinlock_t lock; 32 spinlock_t lock;
27 unsigned int refs; 33 unsigned int refs;
28 struct rcu_head rcu_head; 34 cpumask_var_t cpumask;
29 unsigned long cpumask_bits[];
30}; 35};
31 36
32struct call_single_queue { 37struct call_single_queue {
33 struct list_head list; 38 struct list_head list;
34 spinlock_t lock; 39 spinlock_t lock;
40};
41
42static DEFINE_PER_CPU(struct call_function_data, cfd_data) = {
43 .lock = __SPIN_LOCK_UNLOCKED(cfd_data.lock),
44};
45
46static int
47hotplug_cfd(struct notifier_block *nfb, unsigned long action, void *hcpu)
48{
49 long cpu = (long)hcpu;
50 struct call_function_data *cfd = &per_cpu(cfd_data, cpu);
51
52 switch (action) {
53 case CPU_UP_PREPARE:
54 case CPU_UP_PREPARE_FROZEN:
55 if (!alloc_cpumask_var_node(&cfd->cpumask, GFP_KERNEL,
56 cpu_to_node(cpu)))
57 return NOTIFY_BAD;
58 break;
59
60#ifdef CONFIG_CPU_HOTPLUG
61 case CPU_UP_CANCELED:
62 case CPU_UP_CANCELED_FROZEN:
63
64 case CPU_DEAD:
65 case CPU_DEAD_FROZEN:
66 free_cpumask_var(cfd->cpumask);
67 break;
68#endif
69 };
70
71 return NOTIFY_OK;
72}
73
74static struct notifier_block __cpuinitdata hotplug_cfd_notifier = {
75 .notifier_call = hotplug_cfd,
35}; 76};
36 77
37static int __cpuinit init_call_single_data(void) 78static int __cpuinit init_call_single_data(void)
38{ 79{
80 void *cpu = (void *)(long)smp_processor_id();
39 int i; 81 int i;
40 82
41 for_each_possible_cpu(i) { 83 for_each_possible_cpu(i) {
@@ -44,29 +86,63 @@ static int __cpuinit init_call_single_data(void)
44 spin_lock_init(&q->lock); 86 spin_lock_init(&q->lock);
45 INIT_LIST_HEAD(&q->list); 87 INIT_LIST_HEAD(&q->list);
46 } 88 }
89
90 hotplug_cfd(&hotplug_cfd_notifier, CPU_UP_PREPARE, cpu);
91 register_cpu_notifier(&hotplug_cfd_notifier);
92
47 return 0; 93 return 0;
48} 94}
49early_initcall(init_call_single_data); 95early_initcall(init_call_single_data);
50 96
51static void csd_flag_wait(struct call_single_data *data) 97/*
98 * csd_lock/csd_unlock used to serialize access to per-cpu csd resources
99 *
100 * For non-synchronous ipi calls the csd can still be in use by the
101 * previous function call. For multi-cpu calls its even more interesting
102 * as we'll have to ensure no other cpu is observing our csd.
103 */
104static void csd_lock_wait(struct call_single_data *data)
52{ 105{
53 /* Wait for response */ 106 while (data->flags & CSD_FLAG_LOCK)
54 do {
55 if (!(data->flags & CSD_FLAG_WAIT))
56 break;
57 cpu_relax(); 107 cpu_relax();
58 } while (1); 108}
109
110static void csd_lock(struct call_single_data *data)
111{
112 csd_lock_wait(data);
113 data->flags = CSD_FLAG_LOCK;
114
115 /*
116 * prevent CPU from reordering the above assignment
117 * to ->flags with any subsequent assignments to other
118 * fields of the specified call_single_data structure:
119 */
120 smp_mb();
121}
122
123static void csd_unlock(struct call_single_data *data)
124{
125 WARN_ON(!(data->flags & CSD_FLAG_LOCK));
126
127 /*
128 * ensure we're all done before releasing data:
129 */
130 smp_mb();
131
132 data->flags &= ~CSD_FLAG_LOCK;
59} 133}
60 134
61/* 135/*
62 * Insert a previously allocated call_single_data element for execution 136 * Insert a previously allocated call_single_data element
63 * on the given CPU. data must already have ->func, ->info, and ->flags set. 137 * for execution on the given CPU. data must already have
138 * ->func, ->info, and ->flags set.
64 */ 139 */
65static void generic_exec_single(int cpu, struct call_single_data *data) 140static
141void generic_exec_single(int cpu, struct call_single_data *data, int wait)
66{ 142{
67 struct call_single_queue *dst = &per_cpu(call_single_queue, cpu); 143 struct call_single_queue *dst = &per_cpu(call_single_queue, cpu);
68 int wait = data->flags & CSD_FLAG_WAIT, ipi;
69 unsigned long flags; 144 unsigned long flags;
145 int ipi;
70 146
71 spin_lock_irqsave(&dst->lock, flags); 147 spin_lock_irqsave(&dst->lock, flags);
72 ipi = list_empty(&dst->list); 148 ipi = list_empty(&dst->list);
@@ -74,24 +150,21 @@ static void generic_exec_single(int cpu, struct call_single_data *data)
74 spin_unlock_irqrestore(&dst->lock, flags); 150 spin_unlock_irqrestore(&dst->lock, flags);
75 151
76 /* 152 /*
77 * Make the list addition visible before sending the ipi. 153 * The list addition should be visible before sending the IPI
154 * handler locks the list to pull the entry off it because of
155 * normal cache coherency rules implied by spinlocks.
156 *
157 * If IPIs can go out of order to the cache coherency protocol
158 * in an architecture, sufficient synchronisation should be added
159 * to arch code to make it appear to obey cache coherency WRT
160 * locking and barrier primitives. Generic code isn't really
161 * equipped to do the right thing...
78 */ 162 */
79 smp_mb();
80
81 if (ipi) 163 if (ipi)
82 arch_send_call_function_single_ipi(cpu); 164 arch_send_call_function_single_ipi(cpu);
83 165
84 if (wait) 166 if (wait)
85 csd_flag_wait(data); 167 csd_lock_wait(data);
86}
87
88static void rcu_free_call_data(struct rcu_head *head)
89{
90 struct call_function_data *data;
91
92 data = container_of(head, struct call_function_data, rcu_head);
93
94 kfree(data);
95} 168}
96 169
97/* 170/*
@@ -104,99 +177,83 @@ void generic_smp_call_function_interrupt(void)
104 int cpu = get_cpu(); 177 int cpu = get_cpu();
105 178
106 /* 179 /*
107 * It's ok to use list_for_each_rcu() here even though we may delete 180 * Ensure entry is visible on call_function_queue after we have
108 * 'pos', since list_del_rcu() doesn't clear ->next 181 * entered the IPI. See comment in smp_call_function_many.
182 * If we don't have this, then we may miss an entry on the list
183 * and never get another IPI to process it.
184 */
185 smp_mb();
186
187 /*
188 * It's ok to use list_for_each_rcu() here even though we may
189 * delete 'pos', since list_del_rcu() doesn't clear ->next
109 */ 190 */
110 rcu_read_lock(); 191 list_for_each_entry_rcu(data, &call_function.queue, csd.list) {
111 list_for_each_entry_rcu(data, &call_function_queue, csd.list) {
112 int refs; 192 int refs;
113 193
114 if (!cpumask_test_cpu(cpu, to_cpumask(data->cpumask_bits))) 194 spin_lock(&data->lock);
195 if (!cpumask_test_cpu(cpu, data->cpumask)) {
196 spin_unlock(&data->lock);
115 continue; 197 continue;
198 }
199 cpumask_clear_cpu(cpu, data->cpumask);
200 spin_unlock(&data->lock);
116 201
117 data->csd.func(data->csd.info); 202 data->csd.func(data->csd.info);
118 203
119 spin_lock(&data->lock); 204 spin_lock(&data->lock);
120 cpumask_clear_cpu(cpu, to_cpumask(data->cpumask_bits));
121 WARN_ON(data->refs == 0); 205 WARN_ON(data->refs == 0);
122 data->refs--; 206 refs = --data->refs;
123 refs = data->refs; 207 if (!refs) {
208 spin_lock(&call_function.lock);
209 list_del_rcu(&data->csd.list);
210 spin_unlock(&call_function.lock);
211 }
124 spin_unlock(&data->lock); 212 spin_unlock(&data->lock);
125 213
126 if (refs) 214 if (refs)
127 continue; 215 continue;
128 216
129 spin_lock(&call_function_lock); 217 csd_unlock(&data->csd);
130 list_del_rcu(&data->csd.list);
131 spin_unlock(&call_function_lock);
132
133 if (data->csd.flags & CSD_FLAG_WAIT) {
134 /*
135 * serialize stores to data with the flag clear
136 * and wakeup
137 */
138 smp_wmb();
139 data->csd.flags &= ~CSD_FLAG_WAIT;
140 }
141 if (data->csd.flags & CSD_FLAG_ALLOC)
142 call_rcu(&data->rcu_head, rcu_free_call_data);
143 } 218 }
144 rcu_read_unlock();
145 219
146 put_cpu(); 220 put_cpu();
147} 221}
148 222
149/* 223/*
150 * Invoked by arch to handle an IPI for call function single. Must be called 224 * Invoked by arch to handle an IPI for call function single. Must be
151 * from the arch with interrupts disabled. 225 * called from the arch with interrupts disabled.
152 */ 226 */
153void generic_smp_call_function_single_interrupt(void) 227void generic_smp_call_function_single_interrupt(void)
154{ 228{
155 struct call_single_queue *q = &__get_cpu_var(call_single_queue); 229 struct call_single_queue *q = &__get_cpu_var(call_single_queue);
230 unsigned int data_flags;
156 LIST_HEAD(list); 231 LIST_HEAD(list);
157 232
158 /* 233 spin_lock(&q->lock);
159 * Need to see other stores to list head for checking whether 234 list_replace_init(&q->list, &list);
160 * list is empty without holding q->lock 235 spin_unlock(&q->lock);
161 */ 236
162 smp_read_barrier_depends(); 237 while (!list_empty(&list)) {
163 while (!list_empty(&q->list)) { 238 struct call_single_data *data;
164 unsigned int data_flags; 239
165 240 data = list_entry(list.next, struct call_single_data, list);
166 spin_lock(&q->lock); 241 list_del(&data->list);
167 list_replace_init(&q->list, &list); 242
168 spin_unlock(&q->lock); 243 /*
169 244 * 'data' can be invalid after this call if flags == 0
170 while (!list_empty(&list)) { 245 * (when called through generic_exec_single()),
171 struct call_single_data *data; 246 * so save them away before making the call:
172 247 */
173 data = list_entry(list.next, struct call_single_data, 248 data_flags = data->flags;
174 list); 249
175 list_del(&data->list); 250 data->func(data->info);
176 251
177 /*
178 * 'data' can be invalid after this call if
179 * flags == 0 (when called through
180 * generic_exec_single(), so save them away before
181 * making the call.
182 */
183 data_flags = data->flags;
184
185 data->func(data->info);
186
187 if (data_flags & CSD_FLAG_WAIT) {
188 smp_wmb();
189 data->flags &= ~CSD_FLAG_WAIT;
190 } else if (data_flags & CSD_FLAG_LOCK) {
191 smp_wmb();
192 data->flags &= ~CSD_FLAG_LOCK;
193 } else if (data_flags & CSD_FLAG_ALLOC)
194 kfree(data);
195 }
196 /* 252 /*
197 * See comment on outer loop 253 * Unlocked CSDs are valid through generic_exec_single():
198 */ 254 */
199 smp_read_barrier_depends(); 255 if (data_flags & CSD_FLAG_LOCK)
256 csd_unlock(data);
200 } 257 }
201} 258}
202 259
@@ -215,65 +272,45 @@ static DEFINE_PER_CPU(struct call_single_data, csd_data);
215int smp_call_function_single(int cpu, void (*func) (void *info), void *info, 272int smp_call_function_single(int cpu, void (*func) (void *info), void *info,
216 int wait) 273 int wait)
217{ 274{
218 struct call_single_data d; 275 struct call_single_data d = {
276 .flags = 0,
277 };
219 unsigned long flags; 278 unsigned long flags;
220 /* prevent preemption and reschedule on another processor, 279 int this_cpu;
221 as well as CPU removal */
222 int me = get_cpu();
223 int err = 0; 280 int err = 0;
224 281
282 /*
283 * prevent preemption and reschedule on another processor,
284 * as well as CPU removal
285 */
286 this_cpu = get_cpu();
287
225 /* Can deadlock when called with interrupts disabled */ 288 /* Can deadlock when called with interrupts disabled */
226 WARN_ON(irqs_disabled()); 289 WARN_ON_ONCE(irqs_disabled() && !oops_in_progress);
227 290
228 if (cpu == me) { 291 if (cpu == this_cpu) {
229 local_irq_save(flags); 292 local_irq_save(flags);
230 func(info); 293 func(info);
231 local_irq_restore(flags); 294 local_irq_restore(flags);
232 } else if ((unsigned)cpu < nr_cpu_ids && cpu_online(cpu)) { 295 } else {
233 struct call_single_data *data; 296 if ((unsigned)cpu < nr_cpu_ids && cpu_online(cpu)) {
297 struct call_single_data *data = &d;
298
299 if (!wait)
300 data = &__get_cpu_var(csd_data);
234 301
235 if (!wait) { 302 csd_lock(data);
236 /* 303
237 * We are calling a function on a single CPU 304 data->func = func;
238 * and we are not going to wait for it to finish. 305 data->info = info;
239 * We first try to allocate the data, but if we 306 generic_exec_single(cpu, data, wait);
240 * fail, we fall back to use a per cpu data to pass
241 * the information to that CPU. Since all callers
242 * of this code will use the same data, we must
243 * synchronize the callers to prevent a new caller
244 * from corrupting the data before the callee
245 * can access it.
246 *
247 * The CSD_FLAG_LOCK is used to let us know when
248 * the IPI handler is done with the data.
249 * The first caller will set it, and the callee
250 * will clear it. The next caller must wait for
251 * it to clear before we set it again. This
252 * will make sure the callee is done with the
253 * data before a new caller will use it.
254 */
255 data = kmalloc(sizeof(*data), GFP_ATOMIC);
256 if (data)
257 data->flags = CSD_FLAG_ALLOC;
258 else {
259 data = &per_cpu(csd_data, me);
260 while (data->flags & CSD_FLAG_LOCK)
261 cpu_relax();
262 data->flags = CSD_FLAG_LOCK;
263 }
264 } else { 307 } else {
265 data = &d; 308 err = -ENXIO; /* CPU not online */
266 data->flags = CSD_FLAG_WAIT;
267 } 309 }
268
269 data->func = func;
270 data->info = info;
271 generic_exec_single(cpu, data);
272 } else {
273 err = -ENXIO; /* CPU not online */
274 } 310 }
275 311
276 put_cpu(); 312 put_cpu();
313
277 return err; 314 return err;
278} 315}
279EXPORT_SYMBOL(smp_call_function_single); 316EXPORT_SYMBOL(smp_call_function_single);
@@ -283,23 +320,26 @@ EXPORT_SYMBOL(smp_call_function_single);
283 * @cpu: The CPU to run on. 320 * @cpu: The CPU to run on.
284 * @data: Pre-allocated and setup data structure 321 * @data: Pre-allocated and setup data structure
285 * 322 *
286 * Like smp_call_function_single(), but allow caller to pass in a pre-allocated 323 * Like smp_call_function_single(), but allow caller to pass in a
287 * data structure. Useful for embedding @data inside other structures, for 324 * pre-allocated data structure. Useful for embedding @data inside
288 * instance. 325 * other structures, for instance.
289 *
290 */ 326 */
291void __smp_call_function_single(int cpu, struct call_single_data *data) 327void __smp_call_function_single(int cpu, struct call_single_data *data,
328 int wait)
292{ 329{
330 csd_lock(data);
331
293 /* Can deadlock when called with interrupts disabled */ 332 /* Can deadlock when called with interrupts disabled */
294 WARN_ON((data->flags & CSD_FLAG_WAIT) && irqs_disabled()); 333 WARN_ON_ONCE(wait && irqs_disabled() && !oops_in_progress);
295 334
296 generic_exec_single(cpu, data); 335 generic_exec_single(cpu, data, wait);
297} 336}
298 337
299/* FIXME: Shim for archs using old arch_send_call_function_ipi API. */ 338/* Deprecated: shim for archs using old arch_send_call_function_ipi API. */
339
300#ifndef arch_send_call_function_ipi_mask 340#ifndef arch_send_call_function_ipi_mask
301#define arch_send_call_function_ipi_mask(maskp) \ 341# define arch_send_call_function_ipi_mask(maskp) \
302 arch_send_call_function_ipi(*(maskp)) 342 arch_send_call_function_ipi(*(maskp))
303#endif 343#endif
304 344
305/** 345/**
@@ -307,7 +347,8 @@ void __smp_call_function_single(int cpu, struct call_single_data *data)
307 * @mask: The set of cpus to run on (only runs on online subset). 347 * @mask: The set of cpus to run on (only runs on online subset).
308 * @func: The function to run. This must be fast and non-blocking. 348 * @func: The function to run. This must be fast and non-blocking.
309 * @info: An arbitrary pointer to pass to the function. 349 * @info: An arbitrary pointer to pass to the function.
310 * @wait: If true, wait (atomically) until function has completed on other CPUs. 350 * @wait: If true, wait (atomically) until function has completed
351 * on other CPUs.
311 * 352 *
312 * If @wait is true, then returns once @func has returned. Note that @wait 353 * If @wait is true, then returns once @func has returned. Note that @wait
313 * will be implicitly turned on in case of allocation failures, since 354 * will be implicitly turned on in case of allocation failures, since
@@ -318,27 +359,27 @@ void __smp_call_function_single(int cpu, struct call_single_data *data)
318 * must be disabled when calling this function. 359 * must be disabled when calling this function.
319 */ 360 */
320void smp_call_function_many(const struct cpumask *mask, 361void smp_call_function_many(const struct cpumask *mask,
321 void (*func)(void *), void *info, 362 void (*func)(void *), void *info, bool wait)
322 bool wait)
323{ 363{
324 struct call_function_data *data; 364 struct call_function_data *data;
325 unsigned long flags; 365 unsigned long flags;
326 int cpu, next_cpu; 366 int cpu, next_cpu, this_cpu = smp_processor_id();
327 367
328 /* Can deadlock when called with interrupts disabled */ 368 /* Can deadlock when called with interrupts disabled */
329 WARN_ON(irqs_disabled()); 369 WARN_ON_ONCE(irqs_disabled() && !oops_in_progress);
330 370
331 /* So, what's a CPU they want? Ignoring this one. */ 371 /* So, what's a CPU they want? Ignoring this one. */
332 cpu = cpumask_first_and(mask, cpu_online_mask); 372 cpu = cpumask_first_and(mask, cpu_online_mask);
333 if (cpu == smp_processor_id()) 373 if (cpu == this_cpu)
334 cpu = cpumask_next_and(cpu, mask, cpu_online_mask); 374 cpu = cpumask_next_and(cpu, mask, cpu_online_mask);
375
335 /* No online cpus? We're done. */ 376 /* No online cpus? We're done. */
336 if (cpu >= nr_cpu_ids) 377 if (cpu >= nr_cpu_ids)
337 return; 378 return;
338 379
339 /* Do we have another CPU which isn't us? */ 380 /* Do we have another CPU which isn't us? */
340 next_cpu = cpumask_next_and(cpu, mask, cpu_online_mask); 381 next_cpu = cpumask_next_and(cpu, mask, cpu_online_mask);
341 if (next_cpu == smp_processor_id()) 382 if (next_cpu == this_cpu)
342 next_cpu = cpumask_next_and(next_cpu, mask, cpu_online_mask); 383 next_cpu = cpumask_next_and(next_cpu, mask, cpu_online_mask);
343 384
344 /* Fastpath: do that cpu by itself. */ 385 /* Fastpath: do that cpu by itself. */
@@ -347,43 +388,40 @@ void smp_call_function_many(const struct cpumask *mask,
347 return; 388 return;
348 } 389 }
349 390
350 data = kmalloc(sizeof(*data) + cpumask_size(), GFP_ATOMIC); 391 data = &__get_cpu_var(cfd_data);
351 if (unlikely(!data)) { 392 csd_lock(&data->csd);
352 /* Slow path. */
353 for_each_online_cpu(cpu) {
354 if (cpu == smp_processor_id())
355 continue;
356 if (cpumask_test_cpu(cpu, mask))
357 smp_call_function_single(cpu, func, info, wait);
358 }
359 return;
360 }
361 393
362 spin_lock_init(&data->lock); 394 spin_lock_irqsave(&data->lock, flags);
363 data->csd.flags = CSD_FLAG_ALLOC;
364 if (wait)
365 data->csd.flags |= CSD_FLAG_WAIT;
366 data->csd.func = func; 395 data->csd.func = func;
367 data->csd.info = info; 396 data->csd.info = info;
368 cpumask_and(to_cpumask(data->cpumask_bits), mask, cpu_online_mask); 397 cpumask_and(data->cpumask, mask, cpu_online_mask);
369 cpumask_clear_cpu(smp_processor_id(), to_cpumask(data->cpumask_bits)); 398 cpumask_clear_cpu(this_cpu, data->cpumask);
370 data->refs = cpumask_weight(to_cpumask(data->cpumask_bits)); 399 data->refs = cpumask_weight(data->cpumask);
371 400
372 spin_lock_irqsave(&call_function_lock, flags); 401 spin_lock(&call_function.lock);
373 list_add_tail_rcu(&data->csd.list, &call_function_queue); 402 /*
374 spin_unlock_irqrestore(&call_function_lock, flags); 403 * Place entry at the _HEAD_ of the list, so that any cpu still
404 * observing the entry in generic_smp_call_function_interrupt()
405 * will not miss any other list entries:
406 */
407 list_add_rcu(&data->csd.list, &call_function.queue);
408 spin_unlock(&call_function.lock);
409
410 spin_unlock_irqrestore(&data->lock, flags);
375 411
376 /* 412 /*
377 * Make the list addition visible before sending the ipi. 413 * Make the list addition visible before sending the ipi.
414 * (IPIs must obey or appear to obey normal Linux cache
415 * coherency rules -- see comment in generic_exec_single).
378 */ 416 */
379 smp_mb(); 417 smp_mb();
380 418
381 /* Send a message to all CPUs in the map */ 419 /* Send a message to all CPUs in the map */
382 arch_send_call_function_ipi_mask(to_cpumask(data->cpumask_bits)); 420 arch_send_call_function_ipi_mask(data->cpumask);
383 421
384 /* optionally wait for the CPUs to complete */ 422 /* Optionally wait for the CPUs to complete */
385 if (wait) 423 if (wait)
386 csd_flag_wait(&data->csd); 424 csd_lock_wait(&data->csd);
387} 425}
388EXPORT_SYMBOL(smp_call_function_many); 426EXPORT_SYMBOL(smp_call_function_many);
389 427
@@ -391,7 +429,8 @@ EXPORT_SYMBOL(smp_call_function_many);
391 * smp_call_function(): Run a function on all other CPUs. 429 * smp_call_function(): Run a function on all other CPUs.
392 * @func: The function to run. This must be fast and non-blocking. 430 * @func: The function to run. This must be fast and non-blocking.
393 * @info: An arbitrary pointer to pass to the function. 431 * @info: An arbitrary pointer to pass to the function.
394 * @wait: If true, wait (atomically) until function has completed on other CPUs. 432 * @wait: If true, wait (atomically) until function has completed
433 * on other CPUs.
395 * 434 *
396 * Returns 0. 435 * Returns 0.
397 * 436 *
@@ -407,26 +446,27 @@ int smp_call_function(void (*func)(void *), void *info, int wait)
407 preempt_disable(); 446 preempt_disable();
408 smp_call_function_many(cpu_online_mask, func, info, wait); 447 smp_call_function_many(cpu_online_mask, func, info, wait);
409 preempt_enable(); 448 preempt_enable();
449
410 return 0; 450 return 0;
411} 451}
412EXPORT_SYMBOL(smp_call_function); 452EXPORT_SYMBOL(smp_call_function);
413 453
414void ipi_call_lock(void) 454void ipi_call_lock(void)
415{ 455{
416 spin_lock(&call_function_lock); 456 spin_lock(&call_function.lock);
417} 457}
418 458
419void ipi_call_unlock(void) 459void ipi_call_unlock(void)
420{ 460{
421 spin_unlock(&call_function_lock); 461 spin_unlock(&call_function.lock);
422} 462}
423 463
424void ipi_call_lock_irq(void) 464void ipi_call_lock_irq(void)
425{ 465{
426 spin_lock_irq(&call_function_lock); 466 spin_lock_irq(&call_function.lock);
427} 467}
428 468
429void ipi_call_unlock_irq(void) 469void ipi_call_unlock_irq(void)
430{ 470{
431 spin_unlock_irq(&call_function_lock); 471 spin_unlock_irq(&call_function.lock);
432} 472}
diff --git a/kernel/softirq.c b/kernel/softirq.c
index 0365b4899a3d..d105a82543d0 100644
--- a/kernel/softirq.c
+++ b/kernel/softirq.c
@@ -21,8 +21,10 @@
21#include <linux/freezer.h> 21#include <linux/freezer.h>
22#include <linux/kthread.h> 22#include <linux/kthread.h>
23#include <linux/rcupdate.h> 23#include <linux/rcupdate.h>
24#include <linux/ftrace.h>
24#include <linux/smp.h> 25#include <linux/smp.h>
25#include <linux/tick.h> 26#include <linux/tick.h>
27#include <trace/irq.h>
26 28
27#include <asm/irq.h> 29#include <asm/irq.h>
28/* 30/*
@@ -52,6 +54,11 @@ static struct softirq_action softirq_vec[NR_SOFTIRQS] __cacheline_aligned_in_smp
52 54
53static DEFINE_PER_CPU(struct task_struct *, ksoftirqd); 55static DEFINE_PER_CPU(struct task_struct *, ksoftirqd);
54 56
57char *softirq_to_name[NR_SOFTIRQS] = {
58 "HI", "TIMER", "NET_TX", "NET_RX", "BLOCK",
59 "TASKLET", "SCHED", "HRTIMER", "RCU"
60};
61
55/* 62/*
56 * we cannot loop indefinitely here to avoid userspace starvation, 63 * we cannot loop indefinitely here to avoid userspace starvation,
57 * but we also don't want to introduce a worst case 1/HZ latency 64 * but we also don't want to introduce a worst case 1/HZ latency
@@ -79,13 +86,23 @@ static void __local_bh_disable(unsigned long ip)
79 WARN_ON_ONCE(in_irq()); 86 WARN_ON_ONCE(in_irq());
80 87
81 raw_local_irq_save(flags); 88 raw_local_irq_save(flags);
82 add_preempt_count(SOFTIRQ_OFFSET); 89 /*
90 * The preempt tracer hooks into add_preempt_count and will break
91 * lockdep because it calls back into lockdep after SOFTIRQ_OFFSET
92 * is set and before current->softirq_enabled is cleared.
93 * We must manually increment preempt_count here and manually
94 * call the trace_preempt_off later.
95 */
96 preempt_count() += SOFTIRQ_OFFSET;
83 /* 97 /*
84 * Were softirqs turned off above: 98 * Were softirqs turned off above:
85 */ 99 */
86 if (softirq_count() == SOFTIRQ_OFFSET) 100 if (softirq_count() == SOFTIRQ_OFFSET)
87 trace_softirqs_off(ip); 101 trace_softirqs_off(ip);
88 raw_local_irq_restore(flags); 102 raw_local_irq_restore(flags);
103
104 if (preempt_count() == SOFTIRQ_OFFSET)
105 trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1));
89} 106}
90#else /* !CONFIG_TRACE_IRQFLAGS */ 107#else /* !CONFIG_TRACE_IRQFLAGS */
91static inline void __local_bh_disable(unsigned long ip) 108static inline void __local_bh_disable(unsigned long ip)
@@ -169,6 +186,9 @@ EXPORT_SYMBOL(local_bh_enable_ip);
169 */ 186 */
170#define MAX_SOFTIRQ_RESTART 10 187#define MAX_SOFTIRQ_RESTART 10
171 188
189DEFINE_TRACE(softirq_entry);
190DEFINE_TRACE(softirq_exit);
191
172asmlinkage void __do_softirq(void) 192asmlinkage void __do_softirq(void)
173{ 193{
174 struct softirq_action *h; 194 struct softirq_action *h;
@@ -180,7 +200,7 @@ asmlinkage void __do_softirq(void)
180 account_system_vtime(current); 200 account_system_vtime(current);
181 201
182 __local_bh_disable((unsigned long)__builtin_return_address(0)); 202 __local_bh_disable((unsigned long)__builtin_return_address(0));
183 trace_softirq_enter(); 203 lockdep_softirq_enter();
184 204
185 cpu = smp_processor_id(); 205 cpu = smp_processor_id();
186restart: 206restart:
@@ -195,12 +215,14 @@ restart:
195 if (pending & 1) { 215 if (pending & 1) {
196 int prev_count = preempt_count(); 216 int prev_count = preempt_count();
197 217
218 trace_softirq_entry(h, softirq_vec);
198 h->action(h); 219 h->action(h);
199 220 trace_softirq_exit(h, softirq_vec);
200 if (unlikely(prev_count != preempt_count())) { 221 if (unlikely(prev_count != preempt_count())) {
201 printk(KERN_ERR "huh, entered softirq %td %p" 222 printk(KERN_ERR "huh, entered softirq %td %s %p"
202 "with preempt_count %08x," 223 "with preempt_count %08x,"
203 " exited with %08x?\n", h - softirq_vec, 224 " exited with %08x?\n", h - softirq_vec,
225 softirq_to_name[h - softirq_vec],
204 h->action, prev_count, preempt_count()); 226 h->action, prev_count, preempt_count());
205 preempt_count() = prev_count; 227 preempt_count() = prev_count;
206 } 228 }
@@ -220,7 +242,7 @@ restart:
220 if (pending) 242 if (pending)
221 wakeup_softirqd(); 243 wakeup_softirqd();
222 244
223 trace_softirq_exit(); 245 lockdep_softirq_exit();
224 246
225 account_system_vtime(current); 247 account_system_vtime(current);
226 _local_bh_enable(); 248 _local_bh_enable();
@@ -496,7 +518,7 @@ static int __try_remote_softirq(struct call_single_data *cp, int cpu, int softir
496 cp->flags = 0; 518 cp->flags = 0;
497 cp->priv = softirq; 519 cp->priv = softirq;
498 520
499 __smp_call_function_single(cpu, cp); 521 __smp_call_function_single(cpu, cp, 0);
500 return 0; 522 return 0;
501 } 523 }
502 return 1; 524 return 1;
@@ -626,6 +648,7 @@ static int ksoftirqd(void * __bind_cpu)
626 preempt_enable_no_resched(); 648 preempt_enable_no_resched();
627 cond_resched(); 649 cond_resched();
628 preempt_disable(); 650 preempt_disable();
651 rcu_qsctr_inc((long)__bind_cpu);
629 } 652 }
630 preempt_enable(); 653 preempt_enable();
631 set_current_state(TASK_INTERRUPTIBLE); 654 set_current_state(TASK_INTERRUPTIBLE);
diff --git a/kernel/spinlock.c b/kernel/spinlock.c
index 29ab20749dd3..7932653c4ebd 100644
--- a/kernel/spinlock.c
+++ b/kernel/spinlock.c
@@ -121,7 +121,8 @@ unsigned long __lockfunc _read_lock_irqsave(rwlock_t *lock)
121 local_irq_save(flags); 121 local_irq_save(flags);
122 preempt_disable(); 122 preempt_disable();
123 rwlock_acquire_read(&lock->dep_map, 0, 0, _RET_IP_); 123 rwlock_acquire_read(&lock->dep_map, 0, 0, _RET_IP_);
124 LOCK_CONTENDED(lock, _raw_read_trylock, _raw_read_lock); 124 LOCK_CONTENDED_FLAGS(lock, _raw_read_trylock, _raw_read_lock,
125 _raw_read_lock_flags, &flags);
125 return flags; 126 return flags;
126} 127}
127EXPORT_SYMBOL(_read_lock_irqsave); 128EXPORT_SYMBOL(_read_lock_irqsave);
@@ -151,7 +152,8 @@ unsigned long __lockfunc _write_lock_irqsave(rwlock_t *lock)
151 local_irq_save(flags); 152 local_irq_save(flags);
152 preempt_disable(); 153 preempt_disable();
153 rwlock_acquire(&lock->dep_map, 0, 0, _RET_IP_); 154 rwlock_acquire(&lock->dep_map, 0, 0, _RET_IP_);
154 LOCK_CONTENDED(lock, _raw_write_trylock, _raw_write_lock); 155 LOCK_CONTENDED_FLAGS(lock, _raw_write_trylock, _raw_write_lock,
156 _raw_write_lock_flags, &flags);
155 return flags; 157 return flags;
156} 158}
157EXPORT_SYMBOL(_write_lock_irqsave); 159EXPORT_SYMBOL(_write_lock_irqsave);
@@ -299,16 +301,8 @@ unsigned long __lockfunc _spin_lock_irqsave_nested(spinlock_t *lock, int subclas
299 local_irq_save(flags); 301 local_irq_save(flags);
300 preempt_disable(); 302 preempt_disable();
301 spin_acquire(&lock->dep_map, subclass, 0, _RET_IP_); 303 spin_acquire(&lock->dep_map, subclass, 0, _RET_IP_);
302 /* 304 LOCK_CONTENDED_FLAGS(lock, _raw_spin_trylock, _raw_spin_lock,
303 * On lockdep we dont want the hand-coded irq-enable of 305 _raw_spin_lock_flags, &flags);
304 * _raw_spin_lock_flags() code, because lockdep assumes
305 * that interrupts are not re-enabled during lock-acquire:
306 */
307#ifdef CONFIG_LOCKDEP
308 LOCK_CONTENDED(lock, _raw_spin_trylock, _raw_spin_lock);
309#else
310 _raw_spin_lock_flags(lock, &flags);
311#endif
312 return flags; 306 return flags;
313} 307}
314EXPORT_SYMBOL(_spin_lock_irqsave_nested); 308EXPORT_SYMBOL(_spin_lock_irqsave_nested);
diff --git a/kernel/stop_machine.c b/kernel/stop_machine.c
index 0cd415ee62a2..912823e2a11b 100644
--- a/kernel/stop_machine.c
+++ b/kernel/stop_machine.c
@@ -44,7 +44,7 @@ static DEFINE_MUTEX(setup_lock);
44static int refcount; 44static int refcount;
45static struct workqueue_struct *stop_machine_wq; 45static struct workqueue_struct *stop_machine_wq;
46static struct stop_machine_data active, idle; 46static struct stop_machine_data active, idle;
47static const cpumask_t *active_cpus; 47static const struct cpumask *active_cpus;
48static void *stop_machine_work; 48static void *stop_machine_work;
49 49
50static void set_state(enum stopmachine_state newstate) 50static void set_state(enum stopmachine_state newstate)
@@ -170,7 +170,7 @@ int __stop_machine(int (*fn)(void *), void *data, const struct cpumask *cpus)
170 * doesn't hit this CPU until we're ready. */ 170 * doesn't hit this CPU until we're ready. */
171 get_cpu(); 171 get_cpu();
172 for_each_online_cpu(i) { 172 for_each_online_cpu(i) {
173 sm_work = percpu_ptr(stop_machine_work, i); 173 sm_work = per_cpu_ptr(stop_machine_work, i);
174 INIT_WORK(sm_work, stop_cpu); 174 INIT_WORK(sm_work, stop_cpu);
175 queue_work_on(i, stop_machine_wq, sm_work); 175 queue_work_on(i, stop_machine_wq, sm_work);
176 } 176 }
diff --git a/kernel/sys.c b/kernel/sys.c
index 7306f9421aac..14c4c5613118 100644
--- a/kernel/sys.c
+++ b/kernel/sys.c
@@ -35,6 +35,7 @@
35#include <linux/seccomp.h> 35#include <linux/seccomp.h>
36#include <linux/cpu.h> 36#include <linux/cpu.h>
37#include <linux/ptrace.h> 37#include <linux/ptrace.h>
38#include <linux/fs_struct.h>
38 39
39#include <linux/compat.h> 40#include <linux/compat.h>
40#include <linux/syscalls.h> 41#include <linux/syscalls.h>
@@ -1014,10 +1015,8 @@ SYSCALL_DEFINE2(setpgid, pid_t, pid, pid_t, pgid)
1014 if (err) 1015 if (err)
1015 goto out; 1016 goto out;
1016 1017
1017 if (task_pgrp(p) != pgrp) { 1018 if (task_pgrp(p) != pgrp)
1018 change_pid(p, PIDTYPE_PGID, pgrp); 1019 change_pid(p, PIDTYPE_PGID, pgrp);
1019 set_task_pgrp(p, pid_nr(pgrp));
1020 }
1021 1020
1022 err = 0; 1021 err = 0;
1023out: 1022out:
diff --git a/kernel/sysctl.c b/kernel/sysctl.c
index c5ef44ff850f..82350f8f04f6 100644
--- a/kernel/sysctl.c
+++ b/kernel/sysctl.c
@@ -48,6 +48,7 @@
48#include <linux/acpi.h> 48#include <linux/acpi.h>
49#include <linux/reboot.h> 49#include <linux/reboot.h>
50#include <linux/ftrace.h> 50#include <linux/ftrace.h>
51#include <linux/slow-work.h>
51 52
52#include <asm/uaccess.h> 53#include <asm/uaccess.h>
53#include <asm/processor.h> 54#include <asm/processor.h>
@@ -95,12 +96,9 @@ static int sixty = 60;
95static int neg_one = -1; 96static int neg_one = -1;
96#endif 97#endif
97 98
98#if defined(CONFIG_MMU) && defined(CONFIG_FILE_LOCKING)
99static int two = 2;
100#endif
101
102static int zero; 99static int zero;
103static int one = 1; 100static int one = 1;
101static int two = 2;
104static unsigned long one_ul = 1; 102static unsigned long one_ul = 1;
105static int one_hundred = 100; 103static int one_hundred = 100;
106 104
@@ -900,6 +898,14 @@ static struct ctl_table kern_table[] = {
900 .proc_handler = &scan_unevictable_handler, 898 .proc_handler = &scan_unevictable_handler,
901 }, 899 },
902#endif 900#endif
901#ifdef CONFIG_SLOW_WORK
902 {
903 .ctl_name = CTL_UNNUMBERED,
904 .procname = "slow-work",
905 .mode = 0555,
906 .child = slow_work_sysctls,
907 },
908#endif
903/* 909/*
904 * NOTE: do not add new entries to this table unless you have read 910 * NOTE: do not add new entries to this table unless you have read
905 * Documentation/sysctl/ctl_unnumbered.txt 911 * Documentation/sysctl/ctl_unnumbered.txt
@@ -1010,7 +1016,7 @@ static struct ctl_table vm_table[] = {
1010 .data = &dirty_expire_interval, 1016 .data = &dirty_expire_interval,
1011 .maxlen = sizeof(dirty_expire_interval), 1017 .maxlen = sizeof(dirty_expire_interval),
1012 .mode = 0644, 1018 .mode = 0644,
1013 .proc_handler = &proc_dointvec_userhz_jiffies, 1019 .proc_handler = &proc_dointvec,
1014 }, 1020 },
1015 { 1021 {
1016 .ctl_name = VM_NR_PDFLUSH_THREADS, 1022 .ctl_name = VM_NR_PDFLUSH_THREADS,
@@ -1373,10 +1379,7 @@ static struct ctl_table fs_table[] = {
1373 .data = &lease_break_time, 1379 .data = &lease_break_time,
1374 .maxlen = sizeof(int), 1380 .maxlen = sizeof(int),
1375 .mode = 0644, 1381 .mode = 0644,
1376 .proc_handler = &proc_dointvec_minmax, 1382 .proc_handler = &proc_dointvec,
1377 .strategy = &sysctl_intvec,
1378 .extra1 = &zero,
1379 .extra2 = &two,
1380 }, 1383 },
1381#endif 1384#endif
1382#ifdef CONFIG_AIO 1385#ifdef CONFIG_AIO
@@ -1417,7 +1420,10 @@ static struct ctl_table fs_table[] = {
1417 .data = &suid_dumpable, 1420 .data = &suid_dumpable,
1418 .maxlen = sizeof(int), 1421 .maxlen = sizeof(int),
1419 .mode = 0644, 1422 .mode = 0644,
1420 .proc_handler = &proc_dointvec, 1423 .proc_handler = &proc_dointvec_minmax,
1424 .strategy = &sysctl_intvec,
1425 .extra1 = &zero,
1426 .extra2 = &two,
1421 }, 1427 },
1422#if defined(CONFIG_BINFMT_MISC) || defined(CONFIG_BINFMT_MISC_MODULE) 1428#if defined(CONFIG_BINFMT_MISC) || defined(CONFIG_BINFMT_MISC_MODULE)
1423 { 1429 {
diff --git a/kernel/sysctl_check.c b/kernel/sysctl_check.c
index fafeb48f27c0..b38423ca711a 100644
--- a/kernel/sysctl_check.c
+++ b/kernel/sysctl_check.c
@@ -219,6 +219,7 @@ static const struct trans_ctl_table trans_net_ipv4_conf_vars_table[] = {
219 { NET_IPV4_CONF_ARP_IGNORE, "arp_ignore" }, 219 { NET_IPV4_CONF_ARP_IGNORE, "arp_ignore" },
220 { NET_IPV4_CONF_PROMOTE_SECONDARIES, "promote_secondaries" }, 220 { NET_IPV4_CONF_PROMOTE_SECONDARIES, "promote_secondaries" },
221 { NET_IPV4_CONF_ARP_ACCEPT, "arp_accept" }, 221 { NET_IPV4_CONF_ARP_ACCEPT, "arp_accept" },
222 { NET_IPV4_CONF_ARP_NOTIFY, "arp_notify" },
222 {} 223 {}
223}; 224};
224 225
diff --git a/kernel/time/Makefile b/kernel/time/Makefile
index 905b0b50792d..0b0a6366c9d4 100644
--- a/kernel/time/Makefile
+++ b/kernel/time/Makefile
@@ -1,4 +1,4 @@
1obj-y += timekeeping.o ntp.o clocksource.o jiffies.o timer_list.o 1obj-y += timekeeping.o ntp.o clocksource.o jiffies.o timer_list.o timecompare.o
2 2
3obj-$(CONFIG_GENERIC_CLOCKEVENTS_BUILD) += clockevents.o 3obj-$(CONFIG_GENERIC_CLOCKEVENTS_BUILD) += clockevents.o
4obj-$(CONFIG_GENERIC_CLOCKEVENTS) += tick-common.o 4obj-$(CONFIG_GENERIC_CLOCKEVENTS) += tick-common.o
diff --git a/kernel/time/clockevents.c b/kernel/time/clockevents.c
index ea2f48af83cf..d13be216a790 100644
--- a/kernel/time/clockevents.c
+++ b/kernel/time/clockevents.c
@@ -68,6 +68,17 @@ void clockevents_set_mode(struct clock_event_device *dev,
68 if (dev->mode != mode) { 68 if (dev->mode != mode) {
69 dev->set_mode(mode, dev); 69 dev->set_mode(mode, dev);
70 dev->mode = mode; 70 dev->mode = mode;
71
72 /*
73 * A nsec2cyc multiplicator of 0 is invalid and we'd crash
74 * on it, so fix it up and emit a warning:
75 */
76 if (mode == CLOCK_EVT_MODE_ONESHOT) {
77 if (unlikely(!dev->mult)) {
78 dev->mult = 1;
79 WARN_ON(1);
80 }
81 }
71 } 82 }
72} 83}
73 84
@@ -168,15 +179,6 @@ void clockevents_register_device(struct clock_event_device *dev)
168 BUG_ON(dev->mode != CLOCK_EVT_MODE_UNUSED); 179 BUG_ON(dev->mode != CLOCK_EVT_MODE_UNUSED);
169 BUG_ON(!dev->cpumask); 180 BUG_ON(!dev->cpumask);
170 181
171 /*
172 * A nsec2cyc multiplicator of 0 is invalid and we'd crash
173 * on it, so fix it up and emit a warning:
174 */
175 if (unlikely(!dev->mult)) {
176 dev->mult = 1;
177 WARN_ON(1);
178 }
179
180 spin_lock(&clockevents_lock); 182 spin_lock(&clockevents_lock);
181 183
182 list_add(&dev->list, &clockevent_devices); 184 list_add(&dev->list, &clockevent_devices);
diff --git a/kernel/time/clocksource.c b/kernel/time/clocksource.c
index ca89e1593f08..c46c931a7fe7 100644
--- a/kernel/time/clocksource.c
+++ b/kernel/time/clocksource.c
@@ -31,6 +31,82 @@
31#include <linux/sched.h> /* for spin_unlock_irq() using preempt_count() m68k */ 31#include <linux/sched.h> /* for spin_unlock_irq() using preempt_count() m68k */
32#include <linux/tick.h> 32#include <linux/tick.h>
33 33
34void timecounter_init(struct timecounter *tc,
35 const struct cyclecounter *cc,
36 u64 start_tstamp)
37{
38 tc->cc = cc;
39 tc->cycle_last = cc->read(cc);
40 tc->nsec = start_tstamp;
41}
42EXPORT_SYMBOL(timecounter_init);
43
44/**
45 * timecounter_read_delta - get nanoseconds since last call of this function
46 * @tc: Pointer to time counter
47 *
48 * When the underlying cycle counter runs over, this will be handled
49 * correctly as long as it does not run over more than once between
50 * calls.
51 *
52 * The first call to this function for a new time counter initializes
53 * the time tracking and returns an undefined result.
54 */
55static u64 timecounter_read_delta(struct timecounter *tc)
56{
57 cycle_t cycle_now, cycle_delta;
58 u64 ns_offset;
59
60 /* read cycle counter: */
61 cycle_now = tc->cc->read(tc->cc);
62
63 /* calculate the delta since the last timecounter_read_delta(): */
64 cycle_delta = (cycle_now - tc->cycle_last) & tc->cc->mask;
65
66 /* convert to nanoseconds: */
67 ns_offset = cyclecounter_cyc2ns(tc->cc, cycle_delta);
68
69 /* update time stamp of timecounter_read_delta() call: */
70 tc->cycle_last = cycle_now;
71
72 return ns_offset;
73}
74
75u64 timecounter_read(struct timecounter *tc)
76{
77 u64 nsec;
78
79 /* increment time by nanoseconds since last call */
80 nsec = timecounter_read_delta(tc);
81 nsec += tc->nsec;
82 tc->nsec = nsec;
83
84 return nsec;
85}
86EXPORT_SYMBOL(timecounter_read);
87
88u64 timecounter_cyc2time(struct timecounter *tc,
89 cycle_t cycle_tstamp)
90{
91 u64 cycle_delta = (cycle_tstamp - tc->cycle_last) & tc->cc->mask;
92 u64 nsec;
93
94 /*
95 * Instead of always treating cycle_tstamp as more recent
96 * than tc->cycle_last, detect when it is too far in the
97 * future and treat it as old time stamp instead.
98 */
99 if (cycle_delta > tc->cc->mask / 2) {
100 cycle_delta = (tc->cycle_last - cycle_tstamp) & tc->cc->mask;
101 nsec = tc->nsec - cyclecounter_cyc2ns(tc->cc, cycle_delta);
102 } else {
103 nsec = cyclecounter_cyc2ns(tc->cc, cycle_delta) + tc->nsec;
104 }
105
106 return nsec;
107}
108EXPORT_SYMBOL(timecounter_cyc2time);
109
34/* XXX - Would like a better way for initializing curr_clocksource */ 110/* XXX - Would like a better way for initializing curr_clocksource */
35extern struct clocksource clocksource_jiffies; 111extern struct clocksource clocksource_jiffies;
36 112
diff --git a/kernel/time/ntp.c b/kernel/time/ntp.c
index f5f793d92415..7fc64375ff43 100644
--- a/kernel/time/ntp.c
+++ b/kernel/time/ntp.c
@@ -1,71 +1,129 @@
1/* 1/*
2 * linux/kernel/time/ntp.c
3 *
4 * NTP state machine interfaces and logic. 2 * NTP state machine interfaces and logic.
5 * 3 *
6 * This code was mainly moved from kernel/timer.c and kernel/time.c 4 * This code was mainly moved from kernel/timer.c and kernel/time.c
7 * Please see those files for relevant copyright info and historical 5 * Please see those files for relevant copyright info and historical
8 * changelogs. 6 * changelogs.
9 */ 7 */
10
11#include <linux/mm.h>
12#include <linux/time.h>
13#include <linux/timex.h>
14#include <linux/jiffies.h>
15#include <linux/hrtimer.h>
16#include <linux/capability.h> 8#include <linux/capability.h>
17#include <linux/math64.h>
18#include <linux/clocksource.h> 9#include <linux/clocksource.h>
19#include <linux/workqueue.h> 10#include <linux/workqueue.h>
20#include <asm/timex.h> 11#include <linux/hrtimer.h>
12#include <linux/jiffies.h>
13#include <linux/math64.h>
14#include <linux/timex.h>
15#include <linux/time.h>
16#include <linux/mm.h>
21 17
22/* 18/*
23 * Timekeeping variables 19 * NTP timekeeping variables:
24 */ 20 */
25unsigned long tick_usec = TICK_USEC; /* USER_HZ period (usec) */
26unsigned long tick_nsec; /* ACTHZ period (nsec) */
27u64 tick_length;
28static u64 tick_length_base;
29 21
30static struct hrtimer leap_timer; 22/* USER_HZ period (usecs): */
23unsigned long tick_usec = TICK_USEC;
31 24
32#define MAX_TICKADJ 500 /* microsecs */ 25/* ACTHZ period (nsecs): */
33#define MAX_TICKADJ_SCALED (((u64)(MAX_TICKADJ * NSEC_PER_USEC) << \ 26unsigned long tick_nsec;
34 NTP_SCALE_SHIFT) / NTP_INTERVAL_FREQ) 27
28u64 tick_length;
29static u64 tick_length_base;
30
31static struct hrtimer leap_timer;
32
33#define MAX_TICKADJ 500LL /* usecs */
34#define MAX_TICKADJ_SCALED \
35 (((MAX_TICKADJ * NSEC_PER_USEC) << NTP_SCALE_SHIFT) / NTP_INTERVAL_FREQ)
35 36
36/* 37/*
37 * phase-lock loop variables 38 * phase-lock loop variables
38 */ 39 */
39/* TIME_ERROR prevents overwriting the CMOS clock */
40static int time_state = TIME_OK; /* clock synchronization status */
41int time_status = STA_UNSYNC; /* clock status bits */
42static long time_tai; /* TAI offset (s) */
43static s64 time_offset; /* time adjustment (ns) */
44static long time_constant = 2; /* pll time constant */
45long time_maxerror = NTP_PHASE_LIMIT; /* maximum error (us) */
46long time_esterror = NTP_PHASE_LIMIT; /* estimated error (us) */
47static s64 time_freq; /* frequency offset (scaled ns/s)*/
48static long time_reftime; /* time at last adjustment (s) */
49long time_adjust;
50static long ntp_tick_adj;
51 40
41/*
42 * clock synchronization status
43 *
44 * (TIME_ERROR prevents overwriting the CMOS clock)
45 */
46static int time_state = TIME_OK;
47
48/* clock status bits: */
49int time_status = STA_UNSYNC;
50
51/* TAI offset (secs): */
52static long time_tai;
53
54/* time adjustment (nsecs): */
55static s64 time_offset;
56
57/* pll time constant: */
58static long time_constant = 2;
59
60/* maximum error (usecs): */
61long time_maxerror = NTP_PHASE_LIMIT;
62
63/* estimated error (usecs): */
64long time_esterror = NTP_PHASE_LIMIT;
65
66/* frequency offset (scaled nsecs/secs): */
67static s64 time_freq;
68
69/* time at last adjustment (secs): */
70static long time_reftime;
71
72long time_adjust;
73
74/* constant (boot-param configurable) NTP tick adjustment (upscaled) */
75static s64 ntp_tick_adj;
76
77/*
78 * NTP methods:
79 */
80
81/*
82 * Update (tick_length, tick_length_base, tick_nsec), based
83 * on (tick_usec, ntp_tick_adj, time_freq):
84 */
52static void ntp_update_frequency(void) 85static void ntp_update_frequency(void)
53{ 86{
54 u64 second_length = (u64)(tick_usec * NSEC_PER_USEC * USER_HZ) 87 u64 second_length;
55 << NTP_SCALE_SHIFT; 88 u64 new_base;
56 second_length += (s64)ntp_tick_adj << NTP_SCALE_SHIFT; 89
57 second_length += time_freq; 90 second_length = (u64)(tick_usec * NSEC_PER_USEC * USER_HZ)
91 << NTP_SCALE_SHIFT;
92
93 second_length += ntp_tick_adj;
94 second_length += time_freq;
58 95
59 tick_length_base = second_length; 96 tick_nsec = div_u64(second_length, HZ) >> NTP_SCALE_SHIFT;
97 new_base = div_u64(second_length, NTP_INTERVAL_FREQ);
60 98
61 tick_nsec = div_u64(second_length, HZ) >> NTP_SCALE_SHIFT; 99 /*
62 tick_length_base = div_u64(tick_length_base, NTP_INTERVAL_FREQ); 100 * Don't wait for the next second_overflow, apply
101 * the change to the tick length immediately:
102 */
103 tick_length += new_base - tick_length_base;
104 tick_length_base = new_base;
105}
106
107static inline s64 ntp_update_offset_fll(s64 offset64, long secs)
108{
109 time_status &= ~STA_MODE;
110
111 if (secs < MINSEC)
112 return 0;
113
114 if (!(time_status & STA_FLL) && (secs <= MAXSEC))
115 return 0;
116
117 time_status |= STA_MODE;
118
119 return div_s64(offset64 << (NTP_SCALE_SHIFT - SHIFT_FLL), secs);
63} 120}
64 121
65static void ntp_update_offset(long offset) 122static void ntp_update_offset(long offset)
66{ 123{
67 long mtemp;
68 s64 freq_adj; 124 s64 freq_adj;
125 s64 offset64;
126 long secs;
69 127
70 if (!(time_status & STA_PLL)) 128 if (!(time_status & STA_PLL))
71 return; 129 return;
@@ -84,24 +142,23 @@ static void ntp_update_offset(long offset)
84 * Select how the frequency is to be controlled 142 * Select how the frequency is to be controlled
85 * and in which mode (PLL or FLL). 143 * and in which mode (PLL or FLL).
86 */ 144 */
87 if (time_status & STA_FREQHOLD || time_reftime == 0) 145 secs = xtime.tv_sec - time_reftime;
88 time_reftime = xtime.tv_sec; 146 if (unlikely(time_status & STA_FREQHOLD))
89 mtemp = xtime.tv_sec - time_reftime; 147 secs = 0;
148
90 time_reftime = xtime.tv_sec; 149 time_reftime = xtime.tv_sec;
91 150
92 freq_adj = (s64)offset * mtemp; 151 offset64 = offset;
93 freq_adj <<= NTP_SCALE_SHIFT - 2 * (SHIFT_PLL + 2 + time_constant); 152 freq_adj = (offset64 * secs) <<
94 time_status &= ~STA_MODE; 153 (NTP_SCALE_SHIFT - 2 * (SHIFT_PLL + 2 + time_constant));
95 if (mtemp >= MINSEC && (time_status & STA_FLL || mtemp > MAXSEC)) {
96 freq_adj += div_s64((s64)offset << (NTP_SCALE_SHIFT - SHIFT_FLL),
97 mtemp);
98 time_status |= STA_MODE;
99 }
100 freq_adj += time_freq;
101 freq_adj = min(freq_adj, MAXFREQ_SCALED);
102 time_freq = max(freq_adj, -MAXFREQ_SCALED);
103 154
104 time_offset = div_s64((s64)offset << NTP_SCALE_SHIFT, NTP_INTERVAL_FREQ); 155 freq_adj += ntp_update_offset_fll(offset64, secs);
156
157 freq_adj = min(freq_adj + time_freq, MAXFREQ_SCALED);
158
159 time_freq = max(freq_adj, -MAXFREQ_SCALED);
160
161 time_offset = div_s64(offset64 << NTP_SCALE_SHIFT, NTP_INTERVAL_FREQ);
105} 162}
106 163
107/** 164/**
@@ -111,15 +168,15 @@ static void ntp_update_offset(long offset)
111 */ 168 */
112void ntp_clear(void) 169void ntp_clear(void)
113{ 170{
114 time_adjust = 0; /* stop active adjtime() */ 171 time_adjust = 0; /* stop active adjtime() */
115 time_status |= STA_UNSYNC; 172 time_status |= STA_UNSYNC;
116 time_maxerror = NTP_PHASE_LIMIT; 173 time_maxerror = NTP_PHASE_LIMIT;
117 time_esterror = NTP_PHASE_LIMIT; 174 time_esterror = NTP_PHASE_LIMIT;
118 175
119 ntp_update_frequency(); 176 ntp_update_frequency();
120 177
121 tick_length = tick_length_base; 178 tick_length = tick_length_base;
122 time_offset = 0; 179 time_offset = 0;
123} 180}
124 181
125/* 182/*
@@ -140,8 +197,8 @@ static enum hrtimer_restart ntp_leap_second(struct hrtimer *timer)
140 xtime.tv_sec--; 197 xtime.tv_sec--;
141 wall_to_monotonic.tv_sec++; 198 wall_to_monotonic.tv_sec++;
142 time_state = TIME_OOP; 199 time_state = TIME_OOP;
143 printk(KERN_NOTICE "Clock: " 200 printk(KERN_NOTICE
144 "inserting leap second 23:59:60 UTC\n"); 201 "Clock: inserting leap second 23:59:60 UTC\n");
145 hrtimer_add_expires_ns(&leap_timer, NSEC_PER_SEC); 202 hrtimer_add_expires_ns(&leap_timer, NSEC_PER_SEC);
146 res = HRTIMER_RESTART; 203 res = HRTIMER_RESTART;
147 break; 204 break;
@@ -150,8 +207,8 @@ static enum hrtimer_restart ntp_leap_second(struct hrtimer *timer)
150 time_tai--; 207 time_tai--;
151 wall_to_monotonic.tv_sec--; 208 wall_to_monotonic.tv_sec--;
152 time_state = TIME_WAIT; 209 time_state = TIME_WAIT;
153 printk(KERN_NOTICE "Clock: " 210 printk(KERN_NOTICE
154 "deleting leap second 23:59:59 UTC\n"); 211 "Clock: deleting leap second 23:59:59 UTC\n");
155 break; 212 break;
156 case TIME_OOP: 213 case TIME_OOP:
157 time_tai++; 214 time_tai++;
@@ -179,7 +236,7 @@ static enum hrtimer_restart ntp_leap_second(struct hrtimer *timer)
179 */ 236 */
180void second_overflow(void) 237void second_overflow(void)
181{ 238{
182 s64 time_adj; 239 s64 delta;
183 240
184 /* Bump the maxerror field */ 241 /* Bump the maxerror field */
185 time_maxerror += MAXFREQ / NSEC_PER_USEC; 242 time_maxerror += MAXFREQ / NSEC_PER_USEC;
@@ -192,24 +249,30 @@ void second_overflow(void)
192 * Compute the phase adjustment for the next second. The offset is 249 * Compute the phase adjustment for the next second. The offset is
193 * reduced by a fixed factor times the time constant. 250 * reduced by a fixed factor times the time constant.
194 */ 251 */
195 tick_length = tick_length_base; 252 tick_length = tick_length_base;
196 time_adj = shift_right(time_offset, SHIFT_PLL + time_constant); 253
197 time_offset -= time_adj; 254 delta = shift_right(time_offset, SHIFT_PLL + time_constant);
198 tick_length += time_adj; 255 time_offset -= delta;
199 256 tick_length += delta;
200 if (unlikely(time_adjust)) { 257
201 if (time_adjust > MAX_TICKADJ) { 258 if (!time_adjust)
202 time_adjust -= MAX_TICKADJ; 259 return;
203 tick_length += MAX_TICKADJ_SCALED; 260
204 } else if (time_adjust < -MAX_TICKADJ) { 261 if (time_adjust > MAX_TICKADJ) {
205 time_adjust += MAX_TICKADJ; 262 time_adjust -= MAX_TICKADJ;
206 tick_length -= MAX_TICKADJ_SCALED; 263 tick_length += MAX_TICKADJ_SCALED;
207 } else { 264 return;
208 tick_length += (s64)(time_adjust * NSEC_PER_USEC /
209 NTP_INTERVAL_FREQ) << NTP_SCALE_SHIFT;
210 time_adjust = 0;
211 }
212 } 265 }
266
267 if (time_adjust < -MAX_TICKADJ) {
268 time_adjust += MAX_TICKADJ;
269 tick_length -= MAX_TICKADJ_SCALED;
270 return;
271 }
272
273 tick_length += (s64)(time_adjust * NSEC_PER_USEC / NTP_INTERVAL_FREQ)
274 << NTP_SCALE_SHIFT;
275 time_adjust = 0;
213} 276}
214 277
215#ifdef CONFIG_GENERIC_CMOS_UPDATE 278#ifdef CONFIG_GENERIC_CMOS_UPDATE
@@ -233,12 +296,13 @@ static void sync_cmos_clock(struct work_struct *work)
233 * This code is run on a timer. If the clock is set, that timer 296 * This code is run on a timer. If the clock is set, that timer
234 * may not expire at the correct time. Thus, we adjust... 297 * may not expire at the correct time. Thus, we adjust...
235 */ 298 */
236 if (!ntp_synced()) 299 if (!ntp_synced()) {
237 /* 300 /*
238 * Not synced, exit, do not restart a timer (if one is 301 * Not synced, exit, do not restart a timer (if one is
239 * running, let it run out). 302 * running, let it run out).
240 */ 303 */
241 return; 304 return;
305 }
242 306
243 getnstimeofday(&now); 307 getnstimeofday(&now);
244 if (abs(now.tv_nsec - (NSEC_PER_SEC / 2)) <= tick_nsec / 2) 308 if (abs(now.tv_nsec - (NSEC_PER_SEC / 2)) <= tick_nsec / 2)
@@ -270,7 +334,116 @@ static void notify_cmos_timer(void)
270static inline void notify_cmos_timer(void) { } 334static inline void notify_cmos_timer(void) { }
271#endif 335#endif
272 336
273/* adjtimex mainly allows reading (and writing, if superuser) of 337/*
338 * Start the leap seconds timer:
339 */
340static inline void ntp_start_leap_timer(struct timespec *ts)
341{
342 long now = ts->tv_sec;
343
344 if (time_status & STA_INS) {
345 time_state = TIME_INS;
346 now += 86400 - now % 86400;
347 hrtimer_start(&leap_timer, ktime_set(now, 0), HRTIMER_MODE_ABS);
348
349 return;
350 }
351
352 if (time_status & STA_DEL) {
353 time_state = TIME_DEL;
354 now += 86400 - (now + 1) % 86400;
355 hrtimer_start(&leap_timer, ktime_set(now, 0), HRTIMER_MODE_ABS);
356 }
357}
358
359/*
360 * Propagate a new txc->status value into the NTP state:
361 */
362static inline void process_adj_status(struct timex *txc, struct timespec *ts)
363{
364 if ((time_status & STA_PLL) && !(txc->status & STA_PLL)) {
365 time_state = TIME_OK;
366 time_status = STA_UNSYNC;
367 }
368
369 /*
370 * If we turn on PLL adjustments then reset the
371 * reference time to current time.
372 */
373 if (!(time_status & STA_PLL) && (txc->status & STA_PLL))
374 time_reftime = xtime.tv_sec;
375
376 /* only set allowed bits */
377 time_status &= STA_RONLY;
378 time_status |= txc->status & ~STA_RONLY;
379
380 switch (time_state) {
381 case TIME_OK:
382 ntp_start_leap_timer(ts);
383 break;
384 case TIME_INS:
385 case TIME_DEL:
386 time_state = TIME_OK;
387 ntp_start_leap_timer(ts);
388 case TIME_WAIT:
389 if (!(time_status & (STA_INS | STA_DEL)))
390 time_state = TIME_OK;
391 break;
392 case TIME_OOP:
393 hrtimer_restart(&leap_timer);
394 break;
395 }
396}
397/*
398 * Called with the xtime lock held, so we can access and modify
399 * all the global NTP state:
400 */
401static inline void process_adjtimex_modes(struct timex *txc, struct timespec *ts)
402{
403 if (txc->modes & ADJ_STATUS)
404 process_adj_status(txc, ts);
405
406 if (txc->modes & ADJ_NANO)
407 time_status |= STA_NANO;
408
409 if (txc->modes & ADJ_MICRO)
410 time_status &= ~STA_NANO;
411
412 if (txc->modes & ADJ_FREQUENCY) {
413 time_freq = txc->freq * PPM_SCALE;
414 time_freq = min(time_freq, MAXFREQ_SCALED);
415 time_freq = max(time_freq, -MAXFREQ_SCALED);
416 }
417
418 if (txc->modes & ADJ_MAXERROR)
419 time_maxerror = txc->maxerror;
420
421 if (txc->modes & ADJ_ESTERROR)
422 time_esterror = txc->esterror;
423
424 if (txc->modes & ADJ_TIMECONST) {
425 time_constant = txc->constant;
426 if (!(time_status & STA_NANO))
427 time_constant += 4;
428 time_constant = min(time_constant, (long)MAXTC);
429 time_constant = max(time_constant, 0l);
430 }
431
432 if (txc->modes & ADJ_TAI && txc->constant > 0)
433 time_tai = txc->constant;
434
435 if (txc->modes & ADJ_OFFSET)
436 ntp_update_offset(txc->offset);
437
438 if (txc->modes & ADJ_TICK)
439 tick_usec = txc->tick;
440
441 if (txc->modes & (ADJ_TICK|ADJ_FREQUENCY|ADJ_OFFSET))
442 ntp_update_frequency();
443}
444
445/*
446 * adjtimex mainly allows reading (and writing, if superuser) of
274 * kernel time-keeping variables. used by xntpd. 447 * kernel time-keeping variables. used by xntpd.
275 */ 448 */
276int do_adjtimex(struct timex *txc) 449int do_adjtimex(struct timex *txc)
@@ -291,11 +464,14 @@ int do_adjtimex(struct timex *txc)
291 if (txc->modes && !capable(CAP_SYS_TIME)) 464 if (txc->modes && !capable(CAP_SYS_TIME))
292 return -EPERM; 465 return -EPERM;
293 466
294 /* if the quartz is off by more than 10% something is VERY wrong! */ 467 /*
468 * if the quartz is off by more than 10% then
469 * something is VERY wrong!
470 */
295 if (txc->modes & ADJ_TICK && 471 if (txc->modes & ADJ_TICK &&
296 (txc->tick < 900000/USER_HZ || 472 (txc->tick < 900000/USER_HZ ||
297 txc->tick > 1100000/USER_HZ)) 473 txc->tick > 1100000/USER_HZ))
298 return -EINVAL; 474 return -EINVAL;
299 475
300 if (txc->modes & ADJ_STATUS && time_state != TIME_OK) 476 if (txc->modes & ADJ_STATUS && time_state != TIME_OK)
301 hrtimer_cancel(&leap_timer); 477 hrtimer_cancel(&leap_timer);
@@ -305,7 +481,6 @@ int do_adjtimex(struct timex *txc)
305 481
306 write_seqlock_irq(&xtime_lock); 482 write_seqlock_irq(&xtime_lock);
307 483
308 /* If there are input parameters, then process them */
309 if (txc->modes & ADJ_ADJTIME) { 484 if (txc->modes & ADJ_ADJTIME) {
310 long save_adjust = time_adjust; 485 long save_adjust = time_adjust;
311 486
@@ -315,98 +490,24 @@ int do_adjtimex(struct timex *txc)
315 ntp_update_frequency(); 490 ntp_update_frequency();
316 } 491 }
317 txc->offset = save_adjust; 492 txc->offset = save_adjust;
318 goto adj_done; 493 } else {
319 }
320 if (txc->modes) {
321 long sec;
322
323 if (txc->modes & ADJ_STATUS) {
324 if ((time_status & STA_PLL) &&
325 !(txc->status & STA_PLL)) {
326 time_state = TIME_OK;
327 time_status = STA_UNSYNC;
328 }
329 /* only set allowed bits */
330 time_status &= STA_RONLY;
331 time_status |= txc->status & ~STA_RONLY;
332
333 switch (time_state) {
334 case TIME_OK:
335 start_timer:
336 sec = ts.tv_sec;
337 if (time_status & STA_INS) {
338 time_state = TIME_INS;
339 sec += 86400 - sec % 86400;
340 hrtimer_start(&leap_timer, ktime_set(sec, 0), HRTIMER_MODE_ABS);
341 } else if (time_status & STA_DEL) {
342 time_state = TIME_DEL;
343 sec += 86400 - (sec + 1) % 86400;
344 hrtimer_start(&leap_timer, ktime_set(sec, 0), HRTIMER_MODE_ABS);
345 }
346 break;
347 case TIME_INS:
348 case TIME_DEL:
349 time_state = TIME_OK;
350 goto start_timer;
351 break;
352 case TIME_WAIT:
353 if (!(time_status & (STA_INS | STA_DEL)))
354 time_state = TIME_OK;
355 break;
356 case TIME_OOP:
357 hrtimer_restart(&leap_timer);
358 break;
359 }
360 }
361
362 if (txc->modes & ADJ_NANO)
363 time_status |= STA_NANO;
364 if (txc->modes & ADJ_MICRO)
365 time_status &= ~STA_NANO;
366
367 if (txc->modes & ADJ_FREQUENCY) {
368 time_freq = (s64)txc->freq * PPM_SCALE;
369 time_freq = min(time_freq, MAXFREQ_SCALED);
370 time_freq = max(time_freq, -MAXFREQ_SCALED);
371 }
372
373 if (txc->modes & ADJ_MAXERROR)
374 time_maxerror = txc->maxerror;
375 if (txc->modes & ADJ_ESTERROR)
376 time_esterror = txc->esterror;
377
378 if (txc->modes & ADJ_TIMECONST) {
379 time_constant = txc->constant;
380 if (!(time_status & STA_NANO))
381 time_constant += 4;
382 time_constant = min(time_constant, (long)MAXTC);
383 time_constant = max(time_constant, 0l);
384 }
385
386 if (txc->modes & ADJ_TAI && txc->constant > 0)
387 time_tai = txc->constant;
388
389 if (txc->modes & ADJ_OFFSET)
390 ntp_update_offset(txc->offset);
391 if (txc->modes & ADJ_TICK)
392 tick_usec = txc->tick;
393 494
394 if (txc->modes & (ADJ_TICK|ADJ_FREQUENCY|ADJ_OFFSET)) 495 /* If there are input parameters, then process them: */
395 ntp_update_frequency(); 496 if (txc->modes)
396 } 497 process_adjtimex_modes(txc, &ts);
397 498
398 txc->offset = shift_right(time_offset * NTP_INTERVAL_FREQ, 499 txc->offset = shift_right(time_offset * NTP_INTERVAL_FREQ,
399 NTP_SCALE_SHIFT); 500 NTP_SCALE_SHIFT);
400 if (!(time_status & STA_NANO)) 501 if (!(time_status & STA_NANO))
401 txc->offset /= NSEC_PER_USEC; 502 txc->offset /= NSEC_PER_USEC;
503 }
402 504
403adj_done:
404 result = time_state; /* mostly `TIME_OK' */ 505 result = time_state; /* mostly `TIME_OK' */
405 if (time_status & (STA_UNSYNC|STA_CLOCKERR)) 506 if (time_status & (STA_UNSYNC|STA_CLOCKERR))
406 result = TIME_ERROR; 507 result = TIME_ERROR;
407 508
408 txc->freq = shift_right((time_freq >> PPM_SCALE_INV_SHIFT) * 509 txc->freq = shift_right((time_freq >> PPM_SCALE_INV_SHIFT) *
409 (s64)PPM_SCALE_INV, NTP_SCALE_SHIFT); 510 PPM_SCALE_INV, NTP_SCALE_SHIFT);
410 txc->maxerror = time_maxerror; 511 txc->maxerror = time_maxerror;
411 txc->esterror = time_esterror; 512 txc->esterror = time_esterror;
412 txc->status = time_status; 513 txc->status = time_status;
@@ -425,6 +526,7 @@ adj_done:
425 txc->calcnt = 0; 526 txc->calcnt = 0;
426 txc->errcnt = 0; 527 txc->errcnt = 0;
427 txc->stbcnt = 0; 528 txc->stbcnt = 0;
529
428 write_sequnlock_irq(&xtime_lock); 530 write_sequnlock_irq(&xtime_lock);
429 531
430 txc->time.tv_sec = ts.tv_sec; 532 txc->time.tv_sec = ts.tv_sec;
@@ -440,6 +542,8 @@ adj_done:
440static int __init ntp_tick_adj_setup(char *str) 542static int __init ntp_tick_adj_setup(char *str)
441{ 543{
442 ntp_tick_adj = simple_strtol(str, NULL, 0); 544 ntp_tick_adj = simple_strtol(str, NULL, 0);
545 ntp_tick_adj <<= NTP_SCALE_SHIFT;
546
443 return 1; 547 return 1;
444} 548}
445 549
diff --git a/kernel/time/timecompare.c b/kernel/time/timecompare.c
new file mode 100644
index 000000000000..71e7f1a19156
--- /dev/null
+++ b/kernel/time/timecompare.c
@@ -0,0 +1,191 @@
1/*
2 * Copyright (C) 2009 Intel Corporation.
3 * Author: Patrick Ohly <patrick.ohly@intel.com>
4 *
5 * This program is free software; you can redistribute it and/or modify
6 * it under the terms of the GNU General Public License as published by
7 * the Free Software Foundation; either version 2 of the License, or
8 * (at your option) any later version.
9 *
10 * This program is distributed in the hope that it will be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
14 *
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write to the Free Software
17 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
18 */
19
20#include <linux/timecompare.h>
21#include <linux/module.h>
22#include <linux/math64.h>
23
24/*
25 * fixed point arithmetic scale factor for skew
26 *
27 * Usually one would measure skew in ppb (parts per billion, 1e9), but
28 * using a factor of 2 simplifies the math.
29 */
30#define TIMECOMPARE_SKEW_RESOLUTION (((s64)1)<<30)
31
32ktime_t timecompare_transform(struct timecompare *sync,
33 u64 source_tstamp)
34{
35 u64 nsec;
36
37 nsec = source_tstamp + sync->offset;
38 nsec += (s64)(source_tstamp - sync->last_update) * sync->skew /
39 TIMECOMPARE_SKEW_RESOLUTION;
40
41 return ns_to_ktime(nsec);
42}
43EXPORT_SYMBOL(timecompare_transform);
44
45int timecompare_offset(struct timecompare *sync,
46 s64 *offset,
47 u64 *source_tstamp)
48{
49 u64 start_source = 0, end_source = 0;
50 struct {
51 s64 offset;
52 s64 duration_target;
53 } buffer[10], sample, *samples;
54 int counter = 0, i;
55 int used;
56 int index;
57 int num_samples = sync->num_samples;
58
59 if (num_samples > sizeof(buffer)/sizeof(buffer[0])) {
60 samples = kmalloc(sizeof(*samples) * num_samples, GFP_ATOMIC);
61 if (!samples) {
62 samples = buffer;
63 num_samples = sizeof(buffer)/sizeof(buffer[0]);
64 }
65 } else {
66 samples = buffer;
67 }
68
69 /* run until we have enough valid samples, but do not try forever */
70 i = 0;
71 counter = 0;
72 while (1) {
73 u64 ts;
74 ktime_t start, end;
75
76 start = sync->target();
77 ts = timecounter_read(sync->source);
78 end = sync->target();
79
80 if (!i)
81 start_source = ts;
82
83 /* ignore negative durations */
84 sample.duration_target = ktime_to_ns(ktime_sub(end, start));
85 if (sample.duration_target >= 0) {
86 /*
87 * assume symetric delay to and from source:
88 * average target time corresponds to measured
89 * source time
90 */
91 sample.offset =
92 ktime_to_ns(ktime_add(end, start)) / 2 -
93 ts;
94
95 /* simple insertion sort based on duration */
96 index = counter - 1;
97 while (index >= 0) {
98 if (samples[index].duration_target <
99 sample.duration_target)
100 break;
101 samples[index + 1] = samples[index];
102 index--;
103 }
104 samples[index + 1] = sample;
105 counter++;
106 }
107
108 i++;
109 if (counter >= num_samples || i >= 100000) {
110 end_source = ts;
111 break;
112 }
113 }
114
115 *source_tstamp = (end_source + start_source) / 2;
116
117 /* remove outliers by only using 75% of the samples */
118 used = counter * 3 / 4;
119 if (!used)
120 used = counter;
121 if (used) {
122 /* calculate average */
123 s64 off = 0;
124 for (index = 0; index < used; index++)
125 off += samples[index].offset;
126 *offset = div_s64(off, used);
127 }
128
129 if (samples && samples != buffer)
130 kfree(samples);
131
132 return used;
133}
134EXPORT_SYMBOL(timecompare_offset);
135
136void __timecompare_update(struct timecompare *sync,
137 u64 source_tstamp)
138{
139 s64 offset;
140 u64 average_time;
141
142 if (!timecompare_offset(sync, &offset, &average_time))
143 return;
144
145 if (!sync->last_update) {
146 sync->last_update = average_time;
147 sync->offset = offset;
148 sync->skew = 0;
149 } else {
150 s64 delta_nsec = average_time - sync->last_update;
151
152 /* avoid division by negative or small deltas */
153 if (delta_nsec >= 10000) {
154 s64 delta_offset_nsec = offset - sync->offset;
155 s64 skew; /* delta_offset_nsec *
156 TIMECOMPARE_SKEW_RESOLUTION /
157 delta_nsec */
158 u64 divisor;
159
160 /* div_s64() is limited to 32 bit divisor */
161 skew = delta_offset_nsec * TIMECOMPARE_SKEW_RESOLUTION;
162 divisor = delta_nsec;
163 while (unlikely(divisor >= ((s64)1) << 32)) {
164 /* divide both by 2; beware, right shift
165 of negative value has undefined
166 behavior and can only be used for
167 the positive divisor */
168 skew = div_s64(skew, 2);
169 divisor >>= 1;
170 }
171 skew = div_s64(skew, divisor);
172
173 /*
174 * Calculate new overall skew as 4/16 the
175 * old value and 12/16 the new one. This is
176 * a rather arbitrary tradeoff between
177 * only using the latest measurement (0/16 and
178 * 16/16) and even more weight on past measurements.
179 */
180#define TIMECOMPARE_NEW_SKEW_PER_16 12
181 sync->skew =
182 div_s64((16 - TIMECOMPARE_NEW_SKEW_PER_16) *
183 sync->skew +
184 TIMECOMPARE_NEW_SKEW_PER_16 * skew,
185 16);
186 sync->last_update = average_time;
187 sync->offset = offset;
188 }
189 }
190}
191EXPORT_SYMBOL(__timecompare_update);
diff --git a/kernel/timer.c b/kernel/timer.c
index 13dd64fe143d..b4555568b4e4 100644
--- a/kernel/timer.c
+++ b/kernel/timer.c
@@ -491,14 +491,18 @@ static inline void debug_timer_free(struct timer_list *timer)
491 debug_object_free(timer, &timer_debug_descr); 491 debug_object_free(timer, &timer_debug_descr);
492} 492}
493 493
494static void __init_timer(struct timer_list *timer); 494static void __init_timer(struct timer_list *timer,
495 const char *name,
496 struct lock_class_key *key);
495 497
496void init_timer_on_stack(struct timer_list *timer) 498void init_timer_on_stack_key(struct timer_list *timer,
499 const char *name,
500 struct lock_class_key *key)
497{ 501{
498 debug_object_init_on_stack(timer, &timer_debug_descr); 502 debug_object_init_on_stack(timer, &timer_debug_descr);
499 __init_timer(timer); 503 __init_timer(timer, name, key);
500} 504}
501EXPORT_SYMBOL_GPL(init_timer_on_stack); 505EXPORT_SYMBOL_GPL(init_timer_on_stack_key);
502 506
503void destroy_timer_on_stack(struct timer_list *timer) 507void destroy_timer_on_stack(struct timer_list *timer)
504{ 508{
@@ -512,7 +516,9 @@ static inline void debug_timer_activate(struct timer_list *timer) { }
512static inline void debug_timer_deactivate(struct timer_list *timer) { } 516static inline void debug_timer_deactivate(struct timer_list *timer) { }
513#endif 517#endif
514 518
515static void __init_timer(struct timer_list *timer) 519static void __init_timer(struct timer_list *timer,
520 const char *name,
521 struct lock_class_key *key)
516{ 522{
517 timer->entry.next = NULL; 523 timer->entry.next = NULL;
518 timer->base = __raw_get_cpu_var(tvec_bases); 524 timer->base = __raw_get_cpu_var(tvec_bases);
@@ -521,6 +527,7 @@ static void __init_timer(struct timer_list *timer)
521 timer->start_pid = -1; 527 timer->start_pid = -1;
522 memset(timer->start_comm, 0, TASK_COMM_LEN); 528 memset(timer->start_comm, 0, TASK_COMM_LEN);
523#endif 529#endif
530 lockdep_init_map(&timer->lockdep_map, name, key, 0);
524} 531}
525 532
526/** 533/**
@@ -530,19 +537,23 @@ static void __init_timer(struct timer_list *timer)
530 * init_timer() must be done to a timer prior calling *any* of the 537 * init_timer() must be done to a timer prior calling *any* of the
531 * other timer functions. 538 * other timer functions.
532 */ 539 */
533void init_timer(struct timer_list *timer) 540void init_timer_key(struct timer_list *timer,
541 const char *name,
542 struct lock_class_key *key)
534{ 543{
535 debug_timer_init(timer); 544 debug_timer_init(timer);
536 __init_timer(timer); 545 __init_timer(timer, name, key);
537} 546}
538EXPORT_SYMBOL(init_timer); 547EXPORT_SYMBOL(init_timer_key);
539 548
540void init_timer_deferrable(struct timer_list *timer) 549void init_timer_deferrable_key(struct timer_list *timer,
550 const char *name,
551 struct lock_class_key *key)
541{ 552{
542 init_timer(timer); 553 init_timer_key(timer, name, key);
543 timer_set_deferrable(timer); 554 timer_set_deferrable(timer);
544} 555}
545EXPORT_SYMBOL(init_timer_deferrable); 556EXPORT_SYMBOL(init_timer_deferrable_key);
546 557
547static inline void detach_timer(struct timer_list *timer, 558static inline void detach_timer(struct timer_list *timer,
548 int clear_pending) 559 int clear_pending)
@@ -589,11 +600,14 @@ static struct tvec_base *lock_timer_base(struct timer_list *timer,
589 } 600 }
590} 601}
591 602
592int __mod_timer(struct timer_list *timer, unsigned long expires) 603static inline int
604__mod_timer(struct timer_list *timer, unsigned long expires, bool pending_only)
593{ 605{
594 struct tvec_base *base, *new_base; 606 struct tvec_base *base, *new_base;
595 unsigned long flags; 607 unsigned long flags;
596 int ret = 0; 608 int ret;
609
610 ret = 0;
597 611
598 timer_stats_timer_set_start_info(timer); 612 timer_stats_timer_set_start_info(timer);
599 BUG_ON(!timer->function); 613 BUG_ON(!timer->function);
@@ -603,6 +617,9 @@ int __mod_timer(struct timer_list *timer, unsigned long expires)
603 if (timer_pending(timer)) { 617 if (timer_pending(timer)) {
604 detach_timer(timer, 0); 618 detach_timer(timer, 0);
605 ret = 1; 619 ret = 1;
620 } else {
621 if (pending_only)
622 goto out_unlock;
606 } 623 }
607 624
608 debug_timer_activate(timer); 625 debug_timer_activate(timer);
@@ -629,42 +646,28 @@ int __mod_timer(struct timer_list *timer, unsigned long expires)
629 646
630 timer->expires = expires; 647 timer->expires = expires;
631 internal_add_timer(base, timer); 648 internal_add_timer(base, timer);
649
650out_unlock:
632 spin_unlock_irqrestore(&base->lock, flags); 651 spin_unlock_irqrestore(&base->lock, flags);
633 652
634 return ret; 653 return ret;
635} 654}
636 655
637EXPORT_SYMBOL(__mod_timer);
638
639/** 656/**
640 * add_timer_on - start a timer on a particular CPU 657 * mod_timer_pending - modify a pending timer's timeout
641 * @timer: the timer to be added 658 * @timer: the pending timer to be modified
642 * @cpu: the CPU to start it on 659 * @expires: new timeout in jiffies
643 * 660 *
644 * This is not very scalable on SMP. Double adds are not possible. 661 * mod_timer_pending() is the same for pending timers as mod_timer(),
662 * but will not re-activate and modify already deleted timers.
663 *
664 * It is useful for unserialized use of timers.
645 */ 665 */
646void add_timer_on(struct timer_list *timer, int cpu) 666int mod_timer_pending(struct timer_list *timer, unsigned long expires)
647{ 667{
648 struct tvec_base *base = per_cpu(tvec_bases, cpu); 668 return __mod_timer(timer, expires, true);
649 unsigned long flags;
650
651 timer_stats_timer_set_start_info(timer);
652 BUG_ON(timer_pending(timer) || !timer->function);
653 spin_lock_irqsave(&base->lock, flags);
654 timer_set_base(timer, base);
655 debug_timer_activate(timer);
656 internal_add_timer(base, timer);
657 /*
658 * Check whether the other CPU is idle and needs to be
659 * triggered to reevaluate the timer wheel when nohz is
660 * active. We are protected against the other CPU fiddling
661 * with the timer by holding the timer base lock. This also
662 * makes sure that a CPU on the way to idle can not evaluate
663 * the timer wheel.
664 */
665 wake_up_idle_cpu(cpu);
666 spin_unlock_irqrestore(&base->lock, flags);
667} 669}
670EXPORT_SYMBOL(mod_timer_pending);
668 671
669/** 672/**
670 * mod_timer - modify a timer's timeout 673 * mod_timer - modify a timer's timeout
@@ -688,9 +691,6 @@ void add_timer_on(struct timer_list *timer, int cpu)
688 */ 691 */
689int mod_timer(struct timer_list *timer, unsigned long expires) 692int mod_timer(struct timer_list *timer, unsigned long expires)
690{ 693{
691 BUG_ON(!timer->function);
692
693 timer_stats_timer_set_start_info(timer);
694 /* 694 /*
695 * This is a common optimization triggered by the 695 * This is a common optimization triggered by the
696 * networking code - if the timer is re-modified 696 * networking code - if the timer is re-modified
@@ -699,12 +699,62 @@ int mod_timer(struct timer_list *timer, unsigned long expires)
699 if (timer->expires == expires && timer_pending(timer)) 699 if (timer->expires == expires && timer_pending(timer))
700 return 1; 700 return 1;
701 701
702 return __mod_timer(timer, expires); 702 return __mod_timer(timer, expires, false);
703} 703}
704
705EXPORT_SYMBOL(mod_timer); 704EXPORT_SYMBOL(mod_timer);
706 705
707/** 706/**
707 * add_timer - start a timer
708 * @timer: the timer to be added
709 *
710 * The kernel will do a ->function(->data) callback from the
711 * timer interrupt at the ->expires point in the future. The
712 * current time is 'jiffies'.
713 *
714 * The timer's ->expires, ->function (and if the handler uses it, ->data)
715 * fields must be set prior calling this function.
716 *
717 * Timers with an ->expires field in the past will be executed in the next
718 * timer tick.
719 */
720void add_timer(struct timer_list *timer)
721{
722 BUG_ON(timer_pending(timer));
723 mod_timer(timer, timer->expires);
724}
725EXPORT_SYMBOL(add_timer);
726
727/**
728 * add_timer_on - start a timer on a particular CPU
729 * @timer: the timer to be added
730 * @cpu: the CPU to start it on
731 *
732 * This is not very scalable on SMP. Double adds are not possible.
733 */
734void add_timer_on(struct timer_list *timer, int cpu)
735{
736 struct tvec_base *base = per_cpu(tvec_bases, cpu);
737 unsigned long flags;
738
739 timer_stats_timer_set_start_info(timer);
740 BUG_ON(timer_pending(timer) || !timer->function);
741 spin_lock_irqsave(&base->lock, flags);
742 timer_set_base(timer, base);
743 debug_timer_activate(timer);
744 internal_add_timer(base, timer);
745 /*
746 * Check whether the other CPU is idle and needs to be
747 * triggered to reevaluate the timer wheel when nohz is
748 * active. We are protected against the other CPU fiddling
749 * with the timer by holding the timer base lock. This also
750 * makes sure that a CPU on the way to idle can not evaluate
751 * the timer wheel.
752 */
753 wake_up_idle_cpu(cpu);
754 spin_unlock_irqrestore(&base->lock, flags);
755}
756
757/**
708 * del_timer - deactive a timer. 758 * del_timer - deactive a timer.
709 * @timer: the timer to be deactivated 759 * @timer: the timer to be deactivated
710 * 760 *
@@ -733,7 +783,6 @@ int del_timer(struct timer_list *timer)
733 783
734 return ret; 784 return ret;
735} 785}
736
737EXPORT_SYMBOL(del_timer); 786EXPORT_SYMBOL(del_timer);
738 787
739#ifdef CONFIG_SMP 788#ifdef CONFIG_SMP
@@ -767,7 +816,6 @@ out:
767 816
768 return ret; 817 return ret;
769} 818}
770
771EXPORT_SYMBOL(try_to_del_timer_sync); 819EXPORT_SYMBOL(try_to_del_timer_sync);
772 820
773/** 821/**
@@ -789,6 +837,15 @@ EXPORT_SYMBOL(try_to_del_timer_sync);
789 */ 837 */
790int del_timer_sync(struct timer_list *timer) 838int del_timer_sync(struct timer_list *timer)
791{ 839{
840#ifdef CONFIG_LOCKDEP
841 unsigned long flags;
842
843 local_irq_save(flags);
844 lock_map_acquire(&timer->lockdep_map);
845 lock_map_release(&timer->lockdep_map);
846 local_irq_restore(flags);
847#endif
848
792 for (;;) { 849 for (;;) {
793 int ret = try_to_del_timer_sync(timer); 850 int ret = try_to_del_timer_sync(timer);
794 if (ret >= 0) 851 if (ret >= 0)
@@ -796,7 +853,6 @@ int del_timer_sync(struct timer_list *timer)
796 cpu_relax(); 853 cpu_relax();
797 } 854 }
798} 855}
799
800EXPORT_SYMBOL(del_timer_sync); 856EXPORT_SYMBOL(del_timer_sync);
801#endif 857#endif
802 858
@@ -861,10 +917,36 @@ static inline void __run_timers(struct tvec_base *base)
861 917
862 set_running_timer(base, timer); 918 set_running_timer(base, timer);
863 detach_timer(timer, 1); 919 detach_timer(timer, 1);
920
864 spin_unlock_irq(&base->lock); 921 spin_unlock_irq(&base->lock);
865 { 922 {
866 int preempt_count = preempt_count(); 923 int preempt_count = preempt_count();
924
925#ifdef CONFIG_LOCKDEP
926 /*
927 * It is permissible to free the timer from
928 * inside the function that is called from
929 * it, this we need to take into account for
930 * lockdep too. To avoid bogus "held lock
931 * freed" warnings as well as problems when
932 * looking into timer->lockdep_map, make a
933 * copy and use that here.
934 */
935 struct lockdep_map lockdep_map =
936 timer->lockdep_map;
937#endif
938 /*
939 * Couple the lock chain with the lock chain at
940 * del_timer_sync() by acquiring the lock_map
941 * around the fn() call here and in
942 * del_timer_sync().
943 */
944 lock_map_acquire(&lockdep_map);
945
867 fn(data); 946 fn(data);
947
948 lock_map_release(&lockdep_map);
949
868 if (preempt_count != preempt_count()) { 950 if (preempt_count != preempt_count()) {
869 printk(KERN_ERR "huh, entered %p " 951 printk(KERN_ERR "huh, entered %p "
870 "with preempt_count %08x, exited" 952 "with preempt_count %08x, exited"
@@ -1268,7 +1350,7 @@ signed long __sched schedule_timeout(signed long timeout)
1268 expire = timeout + jiffies; 1350 expire = timeout + jiffies;
1269 1351
1270 setup_timer_on_stack(&timer, process_timeout, (unsigned long)current); 1352 setup_timer_on_stack(&timer, process_timeout, (unsigned long)current);
1271 __mod_timer(&timer, expire); 1353 __mod_timer(&timer, expire, false);
1272 schedule(); 1354 schedule();
1273 del_singleshot_timer_sync(&timer); 1355 del_singleshot_timer_sync(&timer);
1274 1356
diff --git a/kernel/trace/Kconfig b/kernel/trace/Kconfig
index 34e707e5ab87..2246141bda4d 100644
--- a/kernel/trace/Kconfig
+++ b/kernel/trace/Kconfig
@@ -9,6 +9,9 @@ config USER_STACKTRACE_SUPPORT
9config NOP_TRACER 9config NOP_TRACER
10 bool 10 bool
11 11
12config HAVE_FTRACE_NMI_ENTER
13 bool
14
12config HAVE_FUNCTION_TRACER 15config HAVE_FUNCTION_TRACER
13 bool 16 bool
14 17
@@ -31,12 +34,20 @@ config HAVE_FTRACE_MCOUNT_RECORD
31config HAVE_HW_BRANCH_TRACER 34config HAVE_HW_BRANCH_TRACER
32 bool 35 bool
33 36
37config HAVE_FTRACE_SYSCALLS
38 bool
39
34config TRACER_MAX_TRACE 40config TRACER_MAX_TRACE
35 bool 41 bool
36 42
37config RING_BUFFER 43config RING_BUFFER
38 bool 44 bool
39 45
46config FTRACE_NMI_ENTER
47 bool
48 depends on HAVE_FTRACE_NMI_ENTER
49 default y
50
40config TRACING 51config TRACING
41 bool 52 bool
42 select DEBUG_FS 53 select DEBUG_FS
@@ -44,13 +55,29 @@ config TRACING
44 select STACKTRACE if STACKTRACE_SUPPORT 55 select STACKTRACE if STACKTRACE_SUPPORT
45 select TRACEPOINTS 56 select TRACEPOINTS
46 select NOP_TRACER 57 select NOP_TRACER
58 select BINARY_PRINTF
59
60#
61# Minimum requirements an architecture has to meet for us to
62# be able to offer generic tracing facilities:
63#
64config TRACING_SUPPORT
65 bool
66 # PPC32 has no irqflags tracing support, but it can use most of the
67 # tracers anyway, they were tested to build and work. Note that new
68 # exceptions to this list aren't welcomed, better implement the
69 # irqflags tracing for your architecture.
70 depends on TRACE_IRQFLAGS_SUPPORT || PPC32
71 depends on STACKTRACE_SUPPORT
72 default y
73
74if TRACING_SUPPORT
47 75
48menu "Tracers" 76menu "Tracers"
49 77
50config FUNCTION_TRACER 78config FUNCTION_TRACER
51 bool "Kernel Function Tracer" 79 bool "Kernel Function Tracer"
52 depends on HAVE_FUNCTION_TRACER 80 depends on HAVE_FUNCTION_TRACER
53 depends on DEBUG_KERNEL
54 select FRAME_POINTER 81 select FRAME_POINTER
55 select KALLSYMS 82 select KALLSYMS
56 select TRACING 83 select TRACING
@@ -72,18 +99,16 @@ config FUNCTION_GRAPH_TRACER
72 help 99 help
73 Enable the kernel to trace a function at both its return 100 Enable the kernel to trace a function at both its return
74 and its entry. 101 and its entry.
75 It's first purpose is to trace the duration of functions and 102 Its first purpose is to trace the duration of functions and
76 draw a call graph for each thread with some informations like 103 draw a call graph for each thread with some information like
77 the return value. 104 the return value. This is done by setting the current return
78 This is done by setting the current return address on the current 105 address on the current task structure into a stack of calls.
79 task structure into a stack of calls.
80 106
81config IRQSOFF_TRACER 107config IRQSOFF_TRACER
82 bool "Interrupts-off Latency Tracer" 108 bool "Interrupts-off Latency Tracer"
83 default n 109 default n
84 depends on TRACE_IRQFLAGS_SUPPORT 110 depends on TRACE_IRQFLAGS_SUPPORT
85 depends on GENERIC_TIME 111 depends on GENERIC_TIME
86 depends on DEBUG_KERNEL
87 select TRACE_IRQFLAGS 112 select TRACE_IRQFLAGS
88 select TRACING 113 select TRACING
89 select TRACER_MAX_TRACE 114 select TRACER_MAX_TRACE
@@ -106,7 +131,6 @@ config PREEMPT_TRACER
106 default n 131 default n
107 depends on GENERIC_TIME 132 depends on GENERIC_TIME
108 depends on PREEMPT 133 depends on PREEMPT
109 depends on DEBUG_KERNEL
110 select TRACING 134 select TRACING
111 select TRACER_MAX_TRACE 135 select TRACER_MAX_TRACE
112 help 136 help
@@ -127,13 +151,13 @@ config SYSPROF_TRACER
127 bool "Sysprof Tracer" 151 bool "Sysprof Tracer"
128 depends on X86 152 depends on X86
129 select TRACING 153 select TRACING
154 select CONTEXT_SWITCH_TRACER
130 help 155 help
131 This tracer provides the trace needed by the 'Sysprof' userspace 156 This tracer provides the trace needed by the 'Sysprof' userspace
132 tool. 157 tool.
133 158
134config SCHED_TRACER 159config SCHED_TRACER
135 bool "Scheduling Latency Tracer" 160 bool "Scheduling Latency Tracer"
136 depends on DEBUG_KERNEL
137 select TRACING 161 select TRACING
138 select CONTEXT_SWITCH_TRACER 162 select CONTEXT_SWITCH_TRACER
139 select TRACER_MAX_TRACE 163 select TRACER_MAX_TRACE
@@ -143,16 +167,30 @@ config SCHED_TRACER
143 167
144config CONTEXT_SWITCH_TRACER 168config CONTEXT_SWITCH_TRACER
145 bool "Trace process context switches" 169 bool "Trace process context switches"
146 depends on DEBUG_KERNEL
147 select TRACING 170 select TRACING
148 select MARKERS 171 select MARKERS
149 help 172 help
150 This tracer gets called from the context switch and records 173 This tracer gets called from the context switch and records
151 all switching of tasks. 174 all switching of tasks.
152 175
176config EVENT_TRACER
177 bool "Trace various events in the kernel"
178 select TRACING
179 help
180 This tracer hooks to various trace points in the kernel
181 allowing the user to pick and choose which trace point they
182 want to trace.
183
184config FTRACE_SYSCALLS
185 bool "Trace syscalls"
186 depends on HAVE_FTRACE_SYSCALLS
187 select TRACING
188 select KALLSYMS
189 help
190 Basic tracer to catch the syscall entry and exit events.
191
153config BOOT_TRACER 192config BOOT_TRACER
154 bool "Trace boot initcalls" 193 bool "Trace boot initcalls"
155 depends on DEBUG_KERNEL
156 select TRACING 194 select TRACING
157 select CONTEXT_SWITCH_TRACER 195 select CONTEXT_SWITCH_TRACER
158 help 196 help
@@ -165,13 +203,11 @@ config BOOT_TRACER
165 representation of the delays during initcalls - but the raw 203 representation of the delays during initcalls - but the raw
166 /debug/tracing/trace text output is readable too. 204 /debug/tracing/trace text output is readable too.
167 205
168 ( Note that tracing self tests can't be enabled if this tracer is 206 You must pass in ftrace=initcall to the kernel command line
169 selected, because the self-tests are an initcall as well and that 207 to enable this on bootup.
170 would invalidate the boot trace. )
171 208
172config TRACE_BRANCH_PROFILING 209config TRACE_BRANCH_PROFILING
173 bool "Trace likely/unlikely profiler" 210 bool "Trace likely/unlikely profiler"
174 depends on DEBUG_KERNEL
175 select TRACING 211 select TRACING
176 help 212 help
177 This tracer profiles all the the likely and unlikely macros 213 This tracer profiles all the the likely and unlikely macros
@@ -224,7 +260,6 @@ config BRANCH_TRACER
224 260
225config POWER_TRACER 261config POWER_TRACER
226 bool "Trace power consumption behavior" 262 bool "Trace power consumption behavior"
227 depends on DEBUG_KERNEL
228 depends on X86 263 depends on X86
229 select TRACING 264 select TRACING
230 help 265 help
@@ -236,7 +271,6 @@ config POWER_TRACER
236config STACK_TRACER 271config STACK_TRACER
237 bool "Trace max stack" 272 bool "Trace max stack"
238 depends on HAVE_FUNCTION_TRACER 273 depends on HAVE_FUNCTION_TRACER
239 depends on DEBUG_KERNEL
240 select FUNCTION_TRACER 274 select FUNCTION_TRACER
241 select STACKTRACE 275 select STACKTRACE
242 select KALLSYMS 276 select KALLSYMS
@@ -266,11 +300,66 @@ config HW_BRANCH_TRACER
266 This tracer records all branches on the system in a circular 300 This tracer records all branches on the system in a circular
267 buffer giving access to the last N branches for each cpu. 301 buffer giving access to the last N branches for each cpu.
268 302
303config KMEMTRACE
304 bool "Trace SLAB allocations"
305 select TRACING
306 help
307 kmemtrace provides tracing for slab allocator functions, such as
308 kmalloc, kfree, kmem_cache_alloc, kmem_cache_free etc.. Collected
309 data is then fed to the userspace application in order to analyse
310 allocation hotspots, internal fragmentation and so on, making it
311 possible to see how well an allocator performs, as well as debug
312 and profile kernel code.
313
314 This requires an userspace application to use. See
315 Documentation/vm/kmemtrace.txt for more information.
316
317 Saying Y will make the kernel somewhat larger and slower. However,
318 if you disable kmemtrace at run-time or boot-time, the performance
319 impact is minimal (depending on the arch the kernel is built for).
320
321 If unsure, say N.
322
323config WORKQUEUE_TRACER
324 bool "Trace workqueues"
325 select TRACING
326 help
327 The workqueue tracer provides some statistical informations
328 about each cpu workqueue thread such as the number of the
329 works inserted and executed since their creation. It can help
330 to evaluate the amount of work each of them have to perform.
331 For example it can help a developer to decide whether he should
332 choose a per cpu workqueue instead of a singlethreaded one.
333
334config BLK_DEV_IO_TRACE
335 bool "Support for tracing block io actions"
336 depends on SYSFS
337 depends on BLOCK
338 select RELAY
339 select DEBUG_FS
340 select TRACEPOINTS
341 select TRACING
342 select STACKTRACE
343 help
344 Say Y here if you want to be able to trace the block layer actions
345 on a given queue. Tracing allows you to see any traffic happening
346 on a block device queue. For more information (and the userspace
347 support tools needed), fetch the blktrace tools from:
348
349 git://git.kernel.dk/blktrace.git
350
351 Tracing also is possible using the ftrace interface, e.g.:
352
353 echo 1 > /sys/block/sda/sda1/trace/enable
354 echo blk > /sys/kernel/debug/tracing/current_tracer
355 cat /sys/kernel/debug/tracing/trace_pipe
356
357 If unsure, say N.
358
269config DYNAMIC_FTRACE 359config DYNAMIC_FTRACE
270 bool "enable/disable ftrace tracepoints dynamically" 360 bool "enable/disable ftrace tracepoints dynamically"
271 depends on FUNCTION_TRACER 361 depends on FUNCTION_TRACER
272 depends on HAVE_DYNAMIC_FTRACE 362 depends on HAVE_DYNAMIC_FTRACE
273 depends on DEBUG_KERNEL
274 default y 363 default y
275 help 364 help
276 This option will modify all the calls to ftrace dynamically 365 This option will modify all the calls to ftrace dynamically
@@ -296,7 +385,7 @@ config FTRACE_SELFTEST
296 385
297config FTRACE_STARTUP_TEST 386config FTRACE_STARTUP_TEST
298 bool "Perform a startup test on ftrace" 387 bool "Perform a startup test on ftrace"
299 depends on TRACING && DEBUG_KERNEL && !BOOT_TRACER 388 depends on TRACING
300 select FTRACE_SELFTEST 389 select FTRACE_SELFTEST
301 help 390 help
302 This option performs a series of startup tests on ftrace. On bootup 391 This option performs a series of startup tests on ftrace. On bootup
@@ -306,7 +395,7 @@ config FTRACE_STARTUP_TEST
306 395
307config MMIOTRACE 396config MMIOTRACE
308 bool "Memory mapped IO tracing" 397 bool "Memory mapped IO tracing"
309 depends on HAVE_MMIOTRACE_SUPPORT && DEBUG_KERNEL && PCI 398 depends on HAVE_MMIOTRACE_SUPPORT && PCI
310 select TRACING 399 select TRACING
311 help 400 help
312 Mmiotrace traces Memory Mapped I/O access and is meant for 401 Mmiotrace traces Memory Mapped I/O access and is meant for
@@ -328,3 +417,6 @@ config MMIOTRACE_TEST
328 Say N, unless you absolutely know what you are doing. 417 Say N, unless you absolutely know what you are doing.
329 418
330endmenu 419endmenu
420
421endif # TRACING_SUPPORT
422
diff --git a/kernel/trace/Makefile b/kernel/trace/Makefile
index 349d5a93653f..2630f5121ec1 100644
--- a/kernel/trace/Makefile
+++ b/kernel/trace/Makefile
@@ -19,6 +19,10 @@ obj-$(CONFIG_FUNCTION_TRACER) += libftrace.o
19obj-$(CONFIG_RING_BUFFER) += ring_buffer.o 19obj-$(CONFIG_RING_BUFFER) += ring_buffer.o
20 20
21obj-$(CONFIG_TRACING) += trace.o 21obj-$(CONFIG_TRACING) += trace.o
22obj-$(CONFIG_TRACING) += trace_clock.o
23obj-$(CONFIG_TRACING) += trace_output.o
24obj-$(CONFIG_TRACING) += trace_stat.o
25obj-$(CONFIG_TRACING) += trace_printk.o
22obj-$(CONFIG_CONTEXT_SWITCH_TRACER) += trace_sched_switch.o 26obj-$(CONFIG_CONTEXT_SWITCH_TRACER) += trace_sched_switch.o
23obj-$(CONFIG_SYSPROF_TRACER) += trace_sysprof.o 27obj-$(CONFIG_SYSPROF_TRACER) += trace_sysprof.o
24obj-$(CONFIG_FUNCTION_TRACER) += trace_functions.o 28obj-$(CONFIG_FUNCTION_TRACER) += trace_functions.o
@@ -33,5 +37,14 @@ obj-$(CONFIG_FUNCTION_GRAPH_TRACER) += trace_functions_graph.o
33obj-$(CONFIG_TRACE_BRANCH_PROFILING) += trace_branch.o 37obj-$(CONFIG_TRACE_BRANCH_PROFILING) += trace_branch.o
34obj-$(CONFIG_HW_BRANCH_TRACER) += trace_hw_branches.o 38obj-$(CONFIG_HW_BRANCH_TRACER) += trace_hw_branches.o
35obj-$(CONFIG_POWER_TRACER) += trace_power.o 39obj-$(CONFIG_POWER_TRACER) += trace_power.o
40obj-$(CONFIG_KMEMTRACE) += kmemtrace.o
41obj-$(CONFIG_WORKQUEUE_TRACER) += trace_workqueue.o
42obj-$(CONFIG_BLK_DEV_IO_TRACE) += blktrace.o
43obj-$(CONFIG_EVENT_TRACER) += trace_events.o
44obj-$(CONFIG_EVENT_TRACER) += events.o
45obj-$(CONFIG_EVENT_TRACER) += trace_export.o
46obj-$(CONFIG_FTRACE_SYSCALLS) += trace_syscalls.o
47obj-$(CONFIG_EVENT_PROFILE) += trace_event_profile.o
48obj-$(CONFIG_EVENT_TRACER) += trace_events_filter.o
36 49
37libftrace-y := ftrace.o 50libftrace-y := ftrace.o
diff --git a/kernel/trace/blktrace.c b/kernel/trace/blktrace.c
new file mode 100644
index 000000000000..947c5b3f90c4
--- /dev/null
+++ b/kernel/trace/blktrace.c
@@ -0,0 +1,1549 @@
1/*
2 * Copyright (C) 2006 Jens Axboe <axboe@kernel.dk>
3 *
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License version 2 as
6 * published by the Free Software Foundation.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
11 * GNU General Public License for more details.
12 *
13 * You should have received a copy of the GNU General Public License
14 * along with this program; if not, write to the Free Software
15 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
16 *
17 */
18#include <linux/kernel.h>
19#include <linux/blkdev.h>
20#include <linux/blktrace_api.h>
21#include <linux/percpu.h>
22#include <linux/init.h>
23#include <linux/mutex.h>
24#include <linux/debugfs.h>
25#include <linux/time.h>
26#include <trace/block.h>
27#include <linux/uaccess.h>
28#include "trace_output.h"
29
30static unsigned int blktrace_seq __read_mostly = 1;
31
32static struct trace_array *blk_tr;
33static bool blk_tracer_enabled __read_mostly;
34
35/* Select an alternative, minimalistic output than the original one */
36#define TRACE_BLK_OPT_CLASSIC 0x1
37
38static struct tracer_opt blk_tracer_opts[] = {
39 /* Default disable the minimalistic output */
40 { TRACER_OPT(blk_classic, TRACE_BLK_OPT_CLASSIC) },
41 { }
42};
43
44static struct tracer_flags blk_tracer_flags = {
45 .val = 0,
46 .opts = blk_tracer_opts,
47};
48
49/* Global reference count of probes */
50static atomic_t blk_probes_ref = ATOMIC_INIT(0);
51
52static void blk_register_tracepoints(void);
53static void blk_unregister_tracepoints(void);
54
55/*
56 * Send out a notify message.
57 */
58static void trace_note(struct blk_trace *bt, pid_t pid, int action,
59 const void *data, size_t len)
60{
61 struct blk_io_trace *t;
62 struct ring_buffer_event *event = NULL;
63 int pc = 0;
64 int cpu = smp_processor_id();
65 bool blk_tracer = blk_tracer_enabled;
66
67 if (blk_tracer) {
68 pc = preempt_count();
69 event = trace_buffer_lock_reserve(blk_tr, TRACE_BLK,
70 sizeof(*t) + len,
71 0, pc);
72 if (!event)
73 return;
74 t = ring_buffer_event_data(event);
75 goto record_it;
76 }
77
78 if (!bt->rchan)
79 return;
80
81 t = relay_reserve(bt->rchan, sizeof(*t) + len);
82 if (t) {
83 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
84 t->time = ktime_to_ns(ktime_get());
85record_it:
86 t->device = bt->dev;
87 t->action = action;
88 t->pid = pid;
89 t->cpu = cpu;
90 t->pdu_len = len;
91 memcpy((void *) t + sizeof(*t), data, len);
92
93 if (blk_tracer)
94 trace_buffer_unlock_commit(blk_tr, event, 0, pc);
95 }
96}
97
98/*
99 * Send out a notify for this process, if we haven't done so since a trace
100 * started
101 */
102static void trace_note_tsk(struct blk_trace *bt, struct task_struct *tsk)
103{
104 tsk->btrace_seq = blktrace_seq;
105 trace_note(bt, tsk->pid, BLK_TN_PROCESS, tsk->comm, sizeof(tsk->comm));
106}
107
108static void trace_note_time(struct blk_trace *bt)
109{
110 struct timespec now;
111 unsigned long flags;
112 u32 words[2];
113
114 getnstimeofday(&now);
115 words[0] = now.tv_sec;
116 words[1] = now.tv_nsec;
117
118 local_irq_save(flags);
119 trace_note(bt, 0, BLK_TN_TIMESTAMP, words, sizeof(words));
120 local_irq_restore(flags);
121}
122
123void __trace_note_message(struct blk_trace *bt, const char *fmt, ...)
124{
125 int n;
126 va_list args;
127 unsigned long flags;
128 char *buf;
129
130 if (unlikely(bt->trace_state != Blktrace_running &&
131 !blk_tracer_enabled))
132 return;
133
134 local_irq_save(flags);
135 buf = per_cpu_ptr(bt->msg_data, smp_processor_id());
136 va_start(args, fmt);
137 n = vscnprintf(buf, BLK_TN_MAX_MSG, fmt, args);
138 va_end(args);
139
140 trace_note(bt, 0, BLK_TN_MESSAGE, buf, n);
141 local_irq_restore(flags);
142}
143EXPORT_SYMBOL_GPL(__trace_note_message);
144
145static int act_log_check(struct blk_trace *bt, u32 what, sector_t sector,
146 pid_t pid)
147{
148 if (((bt->act_mask << BLK_TC_SHIFT) & what) == 0)
149 return 1;
150 if (sector < bt->start_lba || sector > bt->end_lba)
151 return 1;
152 if (bt->pid && pid != bt->pid)
153 return 1;
154
155 return 0;
156}
157
158/*
159 * Data direction bit lookup
160 */
161static const u32 ddir_act[2] = { BLK_TC_ACT(BLK_TC_READ),
162 BLK_TC_ACT(BLK_TC_WRITE) };
163
164/* The ilog2() calls fall out because they're constant */
165#define MASK_TC_BIT(rw, __name) ((rw & (1 << BIO_RW_ ## __name)) << \
166 (ilog2(BLK_TC_ ## __name) + BLK_TC_SHIFT - BIO_RW_ ## __name))
167
168/*
169 * The worker for the various blk_add_trace*() types. Fills out a
170 * blk_io_trace structure and places it in a per-cpu subbuffer.
171 */
172static void __blk_add_trace(struct blk_trace *bt, sector_t sector, int bytes,
173 int rw, u32 what, int error, int pdu_len, void *pdu_data)
174{
175 struct task_struct *tsk = current;
176 struct ring_buffer_event *event = NULL;
177 struct blk_io_trace *t;
178 unsigned long flags = 0;
179 unsigned long *sequence;
180 pid_t pid;
181 int cpu, pc = 0;
182 bool blk_tracer = blk_tracer_enabled;
183
184 if (unlikely(bt->trace_state != Blktrace_running && !blk_tracer))
185 return;
186
187 what |= ddir_act[rw & WRITE];
188 what |= MASK_TC_BIT(rw, BARRIER);
189 what |= MASK_TC_BIT(rw, SYNCIO);
190 what |= MASK_TC_BIT(rw, AHEAD);
191 what |= MASK_TC_BIT(rw, META);
192 what |= MASK_TC_BIT(rw, DISCARD);
193
194 pid = tsk->pid;
195 if (unlikely(act_log_check(bt, what, sector, pid)))
196 return;
197 cpu = raw_smp_processor_id();
198
199 if (blk_tracer) {
200 tracing_record_cmdline(current);
201
202 pc = preempt_count();
203 event = trace_buffer_lock_reserve(blk_tr, TRACE_BLK,
204 sizeof(*t) + pdu_len,
205 0, pc);
206 if (!event)
207 return;
208 t = ring_buffer_event_data(event);
209 goto record_it;
210 }
211
212 /*
213 * A word about the locking here - we disable interrupts to reserve
214 * some space in the relay per-cpu buffer, to prevent an irq
215 * from coming in and stepping on our toes.
216 */
217 local_irq_save(flags);
218
219 if (unlikely(tsk->btrace_seq != blktrace_seq))
220 trace_note_tsk(bt, tsk);
221
222 t = relay_reserve(bt->rchan, sizeof(*t) + pdu_len);
223 if (t) {
224 sequence = per_cpu_ptr(bt->sequence, cpu);
225
226 t->magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
227 t->sequence = ++(*sequence);
228 t->time = ktime_to_ns(ktime_get());
229record_it:
230 /*
231 * These two are not needed in ftrace as they are in the
232 * generic trace_entry, filled by tracing_generic_entry_update,
233 * but for the trace_event->bin() synthesizer benefit we do it
234 * here too.
235 */
236 t->cpu = cpu;
237 t->pid = pid;
238
239 t->sector = sector;
240 t->bytes = bytes;
241 t->action = what;
242 t->device = bt->dev;
243 t->error = error;
244 t->pdu_len = pdu_len;
245
246 if (pdu_len)
247 memcpy((void *) t + sizeof(*t), pdu_data, pdu_len);
248
249 if (blk_tracer) {
250 trace_buffer_unlock_commit(blk_tr, event, 0, pc);
251 return;
252 }
253 }
254
255 local_irq_restore(flags);
256}
257
258static struct dentry *blk_tree_root;
259static DEFINE_MUTEX(blk_tree_mutex);
260
261static void blk_trace_free(struct blk_trace *bt)
262{
263 debugfs_remove(bt->msg_file);
264 debugfs_remove(bt->dropped_file);
265 relay_close(bt->rchan);
266 free_percpu(bt->sequence);
267 free_percpu(bt->msg_data);
268 kfree(bt);
269}
270
271static void blk_trace_cleanup(struct blk_trace *bt)
272{
273 blk_trace_free(bt);
274 if (atomic_dec_and_test(&blk_probes_ref))
275 blk_unregister_tracepoints();
276}
277
278int blk_trace_remove(struct request_queue *q)
279{
280 struct blk_trace *bt;
281
282 bt = xchg(&q->blk_trace, NULL);
283 if (!bt)
284 return -EINVAL;
285
286 if (bt->trace_state != Blktrace_running)
287 blk_trace_cleanup(bt);
288
289 return 0;
290}
291EXPORT_SYMBOL_GPL(blk_trace_remove);
292
293static int blk_dropped_open(struct inode *inode, struct file *filp)
294{
295 filp->private_data = inode->i_private;
296
297 return 0;
298}
299
300static ssize_t blk_dropped_read(struct file *filp, char __user *buffer,
301 size_t count, loff_t *ppos)
302{
303 struct blk_trace *bt = filp->private_data;
304 char buf[16];
305
306 snprintf(buf, sizeof(buf), "%u\n", atomic_read(&bt->dropped));
307
308 return simple_read_from_buffer(buffer, count, ppos, buf, strlen(buf));
309}
310
311static const struct file_operations blk_dropped_fops = {
312 .owner = THIS_MODULE,
313 .open = blk_dropped_open,
314 .read = blk_dropped_read,
315};
316
317static int blk_msg_open(struct inode *inode, struct file *filp)
318{
319 filp->private_data = inode->i_private;
320
321 return 0;
322}
323
324static ssize_t blk_msg_write(struct file *filp, const char __user *buffer,
325 size_t count, loff_t *ppos)
326{
327 char *msg;
328 struct blk_trace *bt;
329
330 if (count > BLK_TN_MAX_MSG)
331 return -EINVAL;
332
333 msg = kmalloc(count, GFP_KERNEL);
334 if (msg == NULL)
335 return -ENOMEM;
336
337 if (copy_from_user(msg, buffer, count)) {
338 kfree(msg);
339 return -EFAULT;
340 }
341
342 bt = filp->private_data;
343 __trace_note_message(bt, "%s", msg);
344 kfree(msg);
345
346 return count;
347}
348
349static const struct file_operations blk_msg_fops = {
350 .owner = THIS_MODULE,
351 .open = blk_msg_open,
352 .write = blk_msg_write,
353};
354
355/*
356 * Keep track of how many times we encountered a full subbuffer, to aid
357 * the user space app in telling how many lost events there were.
358 */
359static int blk_subbuf_start_callback(struct rchan_buf *buf, void *subbuf,
360 void *prev_subbuf, size_t prev_padding)
361{
362 struct blk_trace *bt;
363
364 if (!relay_buf_full(buf))
365 return 1;
366
367 bt = buf->chan->private_data;
368 atomic_inc(&bt->dropped);
369 return 0;
370}
371
372static int blk_remove_buf_file_callback(struct dentry *dentry)
373{
374 struct dentry *parent = dentry->d_parent;
375 debugfs_remove(dentry);
376
377 /*
378 * this will fail for all but the last file, but that is ok. what we
379 * care about is the top level buts->name directory going away, when
380 * the last trace file is gone. Then we don't have to rmdir() that
381 * manually on trace stop, so it nicely solves the issue with
382 * force killing of running traces.
383 */
384
385 debugfs_remove(parent);
386 return 0;
387}
388
389static struct dentry *blk_create_buf_file_callback(const char *filename,
390 struct dentry *parent,
391 int mode,
392 struct rchan_buf *buf,
393 int *is_global)
394{
395 return debugfs_create_file(filename, mode, parent, buf,
396 &relay_file_operations);
397}
398
399static struct rchan_callbacks blk_relay_callbacks = {
400 .subbuf_start = blk_subbuf_start_callback,
401 .create_buf_file = blk_create_buf_file_callback,
402 .remove_buf_file = blk_remove_buf_file_callback,
403};
404
405/*
406 * Setup everything required to start tracing
407 */
408int do_blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
409 struct blk_user_trace_setup *buts)
410{
411 struct blk_trace *old_bt, *bt = NULL;
412 struct dentry *dir = NULL;
413 int ret, i;
414
415 if (!buts->buf_size || !buts->buf_nr)
416 return -EINVAL;
417
418 strncpy(buts->name, name, BLKTRACE_BDEV_SIZE);
419 buts->name[BLKTRACE_BDEV_SIZE - 1] = '\0';
420
421 /*
422 * some device names have larger paths - convert the slashes
423 * to underscores for this to work as expected
424 */
425 for (i = 0; i < strlen(buts->name); i++)
426 if (buts->name[i] == '/')
427 buts->name[i] = '_';
428
429 bt = kzalloc(sizeof(*bt), GFP_KERNEL);
430 if (!bt)
431 return -ENOMEM;
432
433 ret = -ENOMEM;
434 bt->sequence = alloc_percpu(unsigned long);
435 if (!bt->sequence)
436 goto err;
437
438 bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
439 if (!bt->msg_data)
440 goto err;
441
442 ret = -ENOENT;
443
444 mutex_lock(&blk_tree_mutex);
445 if (!blk_tree_root) {
446 blk_tree_root = debugfs_create_dir("block", NULL);
447 if (!blk_tree_root) {
448 mutex_unlock(&blk_tree_mutex);
449 goto err;
450 }
451 }
452 mutex_unlock(&blk_tree_mutex);
453
454 dir = debugfs_create_dir(buts->name, blk_tree_root);
455
456 if (!dir)
457 goto err;
458
459 bt->dir = dir;
460 bt->dev = dev;
461 atomic_set(&bt->dropped, 0);
462
463 ret = -EIO;
464 bt->dropped_file = debugfs_create_file("dropped", 0444, dir, bt,
465 &blk_dropped_fops);
466 if (!bt->dropped_file)
467 goto err;
468
469 bt->msg_file = debugfs_create_file("msg", 0222, dir, bt, &blk_msg_fops);
470 if (!bt->msg_file)
471 goto err;
472
473 bt->rchan = relay_open("trace", dir, buts->buf_size,
474 buts->buf_nr, &blk_relay_callbacks, bt);
475 if (!bt->rchan)
476 goto err;
477
478 bt->act_mask = buts->act_mask;
479 if (!bt->act_mask)
480 bt->act_mask = (u16) -1;
481
482 bt->start_lba = buts->start_lba;
483 bt->end_lba = buts->end_lba;
484 if (!bt->end_lba)
485 bt->end_lba = -1ULL;
486
487 bt->pid = buts->pid;
488 bt->trace_state = Blktrace_setup;
489
490 ret = -EBUSY;
491 old_bt = xchg(&q->blk_trace, bt);
492 if (old_bt) {
493 (void) xchg(&q->blk_trace, old_bt);
494 goto err;
495 }
496
497 if (atomic_inc_return(&blk_probes_ref) == 1)
498 blk_register_tracepoints();
499
500 return 0;
501err:
502 blk_trace_free(bt);
503 return ret;
504}
505
506int blk_trace_setup(struct request_queue *q, char *name, dev_t dev,
507 char __user *arg)
508{
509 struct blk_user_trace_setup buts;
510 int ret;
511
512 ret = copy_from_user(&buts, arg, sizeof(buts));
513 if (ret)
514 return -EFAULT;
515
516 ret = do_blk_trace_setup(q, name, dev, &buts);
517 if (ret)
518 return ret;
519
520 if (copy_to_user(arg, &buts, sizeof(buts)))
521 return -EFAULT;
522
523 return 0;
524}
525EXPORT_SYMBOL_GPL(blk_trace_setup);
526
527int blk_trace_startstop(struct request_queue *q, int start)
528{
529 int ret;
530 struct blk_trace *bt = q->blk_trace;
531
532 if (bt == NULL)
533 return -EINVAL;
534
535 /*
536 * For starting a trace, we can transition from a setup or stopped
537 * trace. For stopping a trace, the state must be running
538 */
539 ret = -EINVAL;
540 if (start) {
541 if (bt->trace_state == Blktrace_setup ||
542 bt->trace_state == Blktrace_stopped) {
543 blktrace_seq++;
544 smp_mb();
545 bt->trace_state = Blktrace_running;
546
547 trace_note_time(bt);
548 ret = 0;
549 }
550 } else {
551 if (bt->trace_state == Blktrace_running) {
552 bt->trace_state = Blktrace_stopped;
553 relay_flush(bt->rchan);
554 ret = 0;
555 }
556 }
557
558 return ret;
559}
560EXPORT_SYMBOL_GPL(blk_trace_startstop);
561
562/**
563 * blk_trace_ioctl: - handle the ioctls associated with tracing
564 * @bdev: the block device
565 * @cmd: the ioctl cmd
566 * @arg: the argument data, if any
567 *
568 **/
569int blk_trace_ioctl(struct block_device *bdev, unsigned cmd, char __user *arg)
570{
571 struct request_queue *q;
572 int ret, start = 0;
573 char b[BDEVNAME_SIZE];
574
575 q = bdev_get_queue(bdev);
576 if (!q)
577 return -ENXIO;
578
579 mutex_lock(&bdev->bd_mutex);
580
581 switch (cmd) {
582 case BLKTRACESETUP:
583 bdevname(bdev, b);
584 ret = blk_trace_setup(q, b, bdev->bd_dev, arg);
585 break;
586 case BLKTRACESTART:
587 start = 1;
588 case BLKTRACESTOP:
589 ret = blk_trace_startstop(q, start);
590 break;
591 case BLKTRACETEARDOWN:
592 ret = blk_trace_remove(q);
593 break;
594 default:
595 ret = -ENOTTY;
596 break;
597 }
598
599 mutex_unlock(&bdev->bd_mutex);
600 return ret;
601}
602
603/**
604 * blk_trace_shutdown: - stop and cleanup trace structures
605 * @q: the request queue associated with the device
606 *
607 **/
608void blk_trace_shutdown(struct request_queue *q)
609{
610 if (q->blk_trace) {
611 blk_trace_startstop(q, 0);
612 blk_trace_remove(q);
613 }
614}
615
616/*
617 * blktrace probes
618 */
619
620/**
621 * blk_add_trace_rq - Add a trace for a request oriented action
622 * @q: queue the io is for
623 * @rq: the source request
624 * @what: the action
625 *
626 * Description:
627 * Records an action against a request. Will log the bio offset + size.
628 *
629 **/
630static void blk_add_trace_rq(struct request_queue *q, struct request *rq,
631 u32 what)
632{
633 struct blk_trace *bt = q->blk_trace;
634 int rw = rq->cmd_flags & 0x03;
635
636 if (likely(!bt))
637 return;
638
639 if (blk_discard_rq(rq))
640 rw |= (1 << BIO_RW_DISCARD);
641
642 if (blk_pc_request(rq)) {
643 what |= BLK_TC_ACT(BLK_TC_PC);
644 __blk_add_trace(bt, 0, rq->data_len, rw, what, rq->errors,
645 sizeof(rq->cmd), rq->cmd);
646 } else {
647 what |= BLK_TC_ACT(BLK_TC_FS);
648 __blk_add_trace(bt, rq->hard_sector, rq->hard_nr_sectors << 9,
649 rw, what, rq->errors, 0, NULL);
650 }
651}
652
653static void blk_add_trace_rq_abort(struct request_queue *q, struct request *rq)
654{
655 blk_add_trace_rq(q, rq, BLK_TA_ABORT);
656}
657
658static void blk_add_trace_rq_insert(struct request_queue *q, struct request *rq)
659{
660 blk_add_trace_rq(q, rq, BLK_TA_INSERT);
661}
662
663static void blk_add_trace_rq_issue(struct request_queue *q, struct request *rq)
664{
665 blk_add_trace_rq(q, rq, BLK_TA_ISSUE);
666}
667
668static void blk_add_trace_rq_requeue(struct request_queue *q,
669 struct request *rq)
670{
671 blk_add_trace_rq(q, rq, BLK_TA_REQUEUE);
672}
673
674static void blk_add_trace_rq_complete(struct request_queue *q,
675 struct request *rq)
676{
677 blk_add_trace_rq(q, rq, BLK_TA_COMPLETE);
678}
679
680/**
681 * blk_add_trace_bio - Add a trace for a bio oriented action
682 * @q: queue the io is for
683 * @bio: the source bio
684 * @what: the action
685 *
686 * Description:
687 * Records an action against a bio. Will log the bio offset + size.
688 *
689 **/
690static void blk_add_trace_bio(struct request_queue *q, struct bio *bio,
691 u32 what)
692{
693 struct blk_trace *bt = q->blk_trace;
694
695 if (likely(!bt))
696 return;
697
698 __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw, what,
699 !bio_flagged(bio, BIO_UPTODATE), 0, NULL);
700}
701
702static void blk_add_trace_bio_bounce(struct request_queue *q, struct bio *bio)
703{
704 blk_add_trace_bio(q, bio, BLK_TA_BOUNCE);
705}
706
707static void blk_add_trace_bio_complete(struct request_queue *q, struct bio *bio)
708{
709 blk_add_trace_bio(q, bio, BLK_TA_COMPLETE);
710}
711
712static void blk_add_trace_bio_backmerge(struct request_queue *q,
713 struct bio *bio)
714{
715 blk_add_trace_bio(q, bio, BLK_TA_BACKMERGE);
716}
717
718static void blk_add_trace_bio_frontmerge(struct request_queue *q,
719 struct bio *bio)
720{
721 blk_add_trace_bio(q, bio, BLK_TA_FRONTMERGE);
722}
723
724static void blk_add_trace_bio_queue(struct request_queue *q, struct bio *bio)
725{
726 blk_add_trace_bio(q, bio, BLK_TA_QUEUE);
727}
728
729static void blk_add_trace_getrq(struct request_queue *q,
730 struct bio *bio, int rw)
731{
732 if (bio)
733 blk_add_trace_bio(q, bio, BLK_TA_GETRQ);
734 else {
735 struct blk_trace *bt = q->blk_trace;
736
737 if (bt)
738 __blk_add_trace(bt, 0, 0, rw, BLK_TA_GETRQ, 0, 0, NULL);
739 }
740}
741
742
743static void blk_add_trace_sleeprq(struct request_queue *q,
744 struct bio *bio, int rw)
745{
746 if (bio)
747 blk_add_trace_bio(q, bio, BLK_TA_SLEEPRQ);
748 else {
749 struct blk_trace *bt = q->blk_trace;
750
751 if (bt)
752 __blk_add_trace(bt, 0, 0, rw, BLK_TA_SLEEPRQ,
753 0, 0, NULL);
754 }
755}
756
757static void blk_add_trace_plug(struct request_queue *q)
758{
759 struct blk_trace *bt = q->blk_trace;
760
761 if (bt)
762 __blk_add_trace(bt, 0, 0, 0, BLK_TA_PLUG, 0, 0, NULL);
763}
764
765static void blk_add_trace_unplug_io(struct request_queue *q)
766{
767 struct blk_trace *bt = q->blk_trace;
768
769 if (bt) {
770 unsigned int pdu = q->rq.count[READ] + q->rq.count[WRITE];
771 __be64 rpdu = cpu_to_be64(pdu);
772
773 __blk_add_trace(bt, 0, 0, 0, BLK_TA_UNPLUG_IO, 0,
774 sizeof(rpdu), &rpdu);
775 }
776}
777
778static void blk_add_trace_unplug_timer(struct request_queue *q)
779{
780 struct blk_trace *bt = q->blk_trace;
781
782 if (bt) {
783 unsigned int pdu = q->rq.count[READ] + q->rq.count[WRITE];
784 __be64 rpdu = cpu_to_be64(pdu);
785
786 __blk_add_trace(bt, 0, 0, 0, BLK_TA_UNPLUG_TIMER, 0,
787 sizeof(rpdu), &rpdu);
788 }
789}
790
791static void blk_add_trace_split(struct request_queue *q, struct bio *bio,
792 unsigned int pdu)
793{
794 struct blk_trace *bt = q->blk_trace;
795
796 if (bt) {
797 __be64 rpdu = cpu_to_be64(pdu);
798
799 __blk_add_trace(bt, bio->bi_sector, bio->bi_size, bio->bi_rw,
800 BLK_TA_SPLIT, !bio_flagged(bio, BIO_UPTODATE),
801 sizeof(rpdu), &rpdu);
802 }
803}
804
805/**
806 * blk_add_trace_remap - Add a trace for a remap operation
807 * @q: queue the io is for
808 * @bio: the source bio
809 * @dev: target device
810 * @from: source sector
811 * @to: target sector
812 *
813 * Description:
814 * Device mapper or raid target sometimes need to split a bio because
815 * it spans a stripe (or similar). Add a trace for that action.
816 *
817 **/
818static void blk_add_trace_remap(struct request_queue *q, struct bio *bio,
819 dev_t dev, sector_t from, sector_t to)
820{
821 struct blk_trace *bt = q->blk_trace;
822 struct blk_io_trace_remap r;
823
824 if (likely(!bt))
825 return;
826
827 r.device = cpu_to_be32(dev);
828 r.device_from = cpu_to_be32(bio->bi_bdev->bd_dev);
829 r.sector = cpu_to_be64(to);
830
831 __blk_add_trace(bt, from, bio->bi_size, bio->bi_rw, BLK_TA_REMAP,
832 !bio_flagged(bio, BIO_UPTODATE), sizeof(r), &r);
833}
834
835/**
836 * blk_add_driver_data - Add binary message with driver-specific data
837 * @q: queue the io is for
838 * @rq: io request
839 * @data: driver-specific data
840 * @len: length of driver-specific data
841 *
842 * Description:
843 * Some drivers might want to write driver-specific data per request.
844 *
845 **/
846void blk_add_driver_data(struct request_queue *q,
847 struct request *rq,
848 void *data, size_t len)
849{
850 struct blk_trace *bt = q->blk_trace;
851
852 if (likely(!bt))
853 return;
854
855 if (blk_pc_request(rq))
856 __blk_add_trace(bt, 0, rq->data_len, 0, BLK_TA_DRV_DATA,
857 rq->errors, len, data);
858 else
859 __blk_add_trace(bt, rq->hard_sector, rq->hard_nr_sectors << 9,
860 0, BLK_TA_DRV_DATA, rq->errors, len, data);
861}
862EXPORT_SYMBOL_GPL(blk_add_driver_data);
863
864static void blk_register_tracepoints(void)
865{
866 int ret;
867
868 ret = register_trace_block_rq_abort(blk_add_trace_rq_abort);
869 WARN_ON(ret);
870 ret = register_trace_block_rq_insert(blk_add_trace_rq_insert);
871 WARN_ON(ret);
872 ret = register_trace_block_rq_issue(blk_add_trace_rq_issue);
873 WARN_ON(ret);
874 ret = register_trace_block_rq_requeue(blk_add_trace_rq_requeue);
875 WARN_ON(ret);
876 ret = register_trace_block_rq_complete(blk_add_trace_rq_complete);
877 WARN_ON(ret);
878 ret = register_trace_block_bio_bounce(blk_add_trace_bio_bounce);
879 WARN_ON(ret);
880 ret = register_trace_block_bio_complete(blk_add_trace_bio_complete);
881 WARN_ON(ret);
882 ret = register_trace_block_bio_backmerge(blk_add_trace_bio_backmerge);
883 WARN_ON(ret);
884 ret = register_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge);
885 WARN_ON(ret);
886 ret = register_trace_block_bio_queue(blk_add_trace_bio_queue);
887 WARN_ON(ret);
888 ret = register_trace_block_getrq(blk_add_trace_getrq);
889 WARN_ON(ret);
890 ret = register_trace_block_sleeprq(blk_add_trace_sleeprq);
891 WARN_ON(ret);
892 ret = register_trace_block_plug(blk_add_trace_plug);
893 WARN_ON(ret);
894 ret = register_trace_block_unplug_timer(blk_add_trace_unplug_timer);
895 WARN_ON(ret);
896 ret = register_trace_block_unplug_io(blk_add_trace_unplug_io);
897 WARN_ON(ret);
898 ret = register_trace_block_split(blk_add_trace_split);
899 WARN_ON(ret);
900 ret = register_trace_block_remap(blk_add_trace_remap);
901 WARN_ON(ret);
902}
903
904static void blk_unregister_tracepoints(void)
905{
906 unregister_trace_block_remap(blk_add_trace_remap);
907 unregister_trace_block_split(blk_add_trace_split);
908 unregister_trace_block_unplug_io(blk_add_trace_unplug_io);
909 unregister_trace_block_unplug_timer(blk_add_trace_unplug_timer);
910 unregister_trace_block_plug(blk_add_trace_plug);
911 unregister_trace_block_sleeprq(blk_add_trace_sleeprq);
912 unregister_trace_block_getrq(blk_add_trace_getrq);
913 unregister_trace_block_bio_queue(blk_add_trace_bio_queue);
914 unregister_trace_block_bio_frontmerge(blk_add_trace_bio_frontmerge);
915 unregister_trace_block_bio_backmerge(blk_add_trace_bio_backmerge);
916 unregister_trace_block_bio_complete(blk_add_trace_bio_complete);
917 unregister_trace_block_bio_bounce(blk_add_trace_bio_bounce);
918 unregister_trace_block_rq_complete(blk_add_trace_rq_complete);
919 unregister_trace_block_rq_requeue(blk_add_trace_rq_requeue);
920 unregister_trace_block_rq_issue(blk_add_trace_rq_issue);
921 unregister_trace_block_rq_insert(blk_add_trace_rq_insert);
922 unregister_trace_block_rq_abort(blk_add_trace_rq_abort);
923
924 tracepoint_synchronize_unregister();
925}
926
927/*
928 * struct blk_io_tracer formatting routines
929 */
930
931static void fill_rwbs(char *rwbs, const struct blk_io_trace *t)
932{
933 int i = 0;
934 int tc = t->action >> BLK_TC_SHIFT;
935
936 if (t->action == BLK_TN_MESSAGE) {
937 rwbs[i++] = 'N';
938 goto out;
939 }
940
941 if (tc & BLK_TC_DISCARD)
942 rwbs[i++] = 'D';
943 else if (tc & BLK_TC_WRITE)
944 rwbs[i++] = 'W';
945 else if (t->bytes)
946 rwbs[i++] = 'R';
947 else
948 rwbs[i++] = 'N';
949
950 if (tc & BLK_TC_AHEAD)
951 rwbs[i++] = 'A';
952 if (tc & BLK_TC_BARRIER)
953 rwbs[i++] = 'B';
954 if (tc & BLK_TC_SYNC)
955 rwbs[i++] = 'S';
956 if (tc & BLK_TC_META)
957 rwbs[i++] = 'M';
958out:
959 rwbs[i] = '\0';
960}
961
962static inline
963const struct blk_io_trace *te_blk_io_trace(const struct trace_entry *ent)
964{
965 return (const struct blk_io_trace *)ent;
966}
967
968static inline const void *pdu_start(const struct trace_entry *ent)
969{
970 return te_blk_io_trace(ent) + 1;
971}
972
973static inline u32 t_sec(const struct trace_entry *ent)
974{
975 return te_blk_io_trace(ent)->bytes >> 9;
976}
977
978static inline unsigned long long t_sector(const struct trace_entry *ent)
979{
980 return te_blk_io_trace(ent)->sector;
981}
982
983static inline __u16 t_error(const struct trace_entry *ent)
984{
985 return te_blk_io_trace(ent)->error;
986}
987
988static __u64 get_pdu_int(const struct trace_entry *ent)
989{
990 const __u64 *val = pdu_start(ent);
991 return be64_to_cpu(*val);
992}
993
994static void get_pdu_remap(const struct trace_entry *ent,
995 struct blk_io_trace_remap *r)
996{
997 const struct blk_io_trace_remap *__r = pdu_start(ent);
998 __u64 sector = __r->sector;
999
1000 r->device = be32_to_cpu(__r->device);
1001 r->device_from = be32_to_cpu(__r->device_from);
1002 r->sector = be64_to_cpu(sector);
1003}
1004
1005typedef int (blk_log_action_t) (struct trace_iterator *iter, const char *act);
1006
1007static int blk_log_action_classic(struct trace_iterator *iter, const char *act)
1008{
1009 char rwbs[6];
1010 unsigned long long ts = iter->ts;
1011 unsigned long nsec_rem = do_div(ts, NSEC_PER_SEC);
1012 unsigned secs = (unsigned long)ts;
1013 const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1014
1015 fill_rwbs(rwbs, t);
1016
1017 return trace_seq_printf(&iter->seq,
1018 "%3d,%-3d %2d %5d.%09lu %5u %2s %3s ",
1019 MAJOR(t->device), MINOR(t->device), iter->cpu,
1020 secs, nsec_rem, iter->ent->pid, act, rwbs);
1021}
1022
1023static int blk_log_action(struct trace_iterator *iter, const char *act)
1024{
1025 char rwbs[6];
1026 const struct blk_io_trace *t = te_blk_io_trace(iter->ent);
1027
1028 fill_rwbs(rwbs, t);
1029 return trace_seq_printf(&iter->seq, "%3d,%-3d %2s %3s ",
1030 MAJOR(t->device), MINOR(t->device), act, rwbs);
1031}
1032
1033static int blk_log_generic(struct trace_seq *s, const struct trace_entry *ent)
1034{
1035 char cmd[TASK_COMM_LEN];
1036
1037 trace_find_cmdline(ent->pid, cmd);
1038
1039 if (t_sec(ent))
1040 return trace_seq_printf(s, "%llu + %u [%s]\n",
1041 t_sector(ent), t_sec(ent), cmd);
1042 return trace_seq_printf(s, "[%s]\n", cmd);
1043}
1044
1045static int blk_log_with_error(struct trace_seq *s,
1046 const struct trace_entry *ent)
1047{
1048 if (t_sec(ent))
1049 return trace_seq_printf(s, "%llu + %u [%d]\n", t_sector(ent),
1050 t_sec(ent), t_error(ent));
1051 return trace_seq_printf(s, "%llu [%d]\n", t_sector(ent), t_error(ent));
1052}
1053
1054static int blk_log_remap(struct trace_seq *s, const struct trace_entry *ent)
1055{
1056 struct blk_io_trace_remap r = { .device = 0, };
1057
1058 get_pdu_remap(ent, &r);
1059 return trace_seq_printf(s, "%llu + %u <- (%d,%d) %llu\n",
1060 t_sector(ent),
1061 t_sec(ent), MAJOR(r.device), MINOR(r.device),
1062 (unsigned long long)r.sector);
1063}
1064
1065static int blk_log_plug(struct trace_seq *s, const struct trace_entry *ent)
1066{
1067 char cmd[TASK_COMM_LEN];
1068
1069 trace_find_cmdline(ent->pid, cmd);
1070
1071 return trace_seq_printf(s, "[%s]\n", cmd);
1072}
1073
1074static int blk_log_unplug(struct trace_seq *s, const struct trace_entry *ent)
1075{
1076 char cmd[TASK_COMM_LEN];
1077
1078 trace_find_cmdline(ent->pid, cmd);
1079
1080 return trace_seq_printf(s, "[%s] %llu\n", cmd, get_pdu_int(ent));
1081}
1082
1083static int blk_log_split(struct trace_seq *s, const struct trace_entry *ent)
1084{
1085 char cmd[TASK_COMM_LEN];
1086
1087 trace_find_cmdline(ent->pid, cmd);
1088
1089 return trace_seq_printf(s, "%llu / %llu [%s]\n", t_sector(ent),
1090 get_pdu_int(ent), cmd);
1091}
1092
1093static int blk_log_msg(struct trace_seq *s, const struct trace_entry *ent)
1094{
1095 int ret;
1096 const struct blk_io_trace *t = te_blk_io_trace(ent);
1097
1098 ret = trace_seq_putmem(s, t + 1, t->pdu_len);
1099 if (ret)
1100 return trace_seq_putc(s, '\n');
1101 return ret;
1102}
1103
1104/*
1105 * struct tracer operations
1106 */
1107
1108static void blk_tracer_print_header(struct seq_file *m)
1109{
1110 if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1111 return;
1112 seq_puts(m, "# DEV CPU TIMESTAMP PID ACT FLG\n"
1113 "# | | | | | |\n");
1114}
1115
1116static void blk_tracer_start(struct trace_array *tr)
1117{
1118 blk_tracer_enabled = true;
1119 trace_flags &= ~TRACE_ITER_CONTEXT_INFO;
1120}
1121
1122static int blk_tracer_init(struct trace_array *tr)
1123{
1124 blk_tr = tr;
1125 blk_tracer_start(tr);
1126 return 0;
1127}
1128
1129static void blk_tracer_stop(struct trace_array *tr)
1130{
1131 blk_tracer_enabled = false;
1132 trace_flags |= TRACE_ITER_CONTEXT_INFO;
1133}
1134
1135static void blk_tracer_reset(struct trace_array *tr)
1136{
1137 blk_tracer_stop(tr);
1138}
1139
1140static const struct {
1141 const char *act[2];
1142 int (*print)(struct trace_seq *s, const struct trace_entry *ent);
1143} what2act[] = {
1144 [__BLK_TA_QUEUE] = {{ "Q", "queue" }, blk_log_generic },
1145 [__BLK_TA_BACKMERGE] = {{ "M", "backmerge" }, blk_log_generic },
1146 [__BLK_TA_FRONTMERGE] = {{ "F", "frontmerge" }, blk_log_generic },
1147 [__BLK_TA_GETRQ] = {{ "G", "getrq" }, blk_log_generic },
1148 [__BLK_TA_SLEEPRQ] = {{ "S", "sleeprq" }, blk_log_generic },
1149 [__BLK_TA_REQUEUE] = {{ "R", "requeue" }, blk_log_with_error },
1150 [__BLK_TA_ISSUE] = {{ "D", "issue" }, blk_log_generic },
1151 [__BLK_TA_COMPLETE] = {{ "C", "complete" }, blk_log_with_error },
1152 [__BLK_TA_PLUG] = {{ "P", "plug" }, blk_log_plug },
1153 [__BLK_TA_UNPLUG_IO] = {{ "U", "unplug_io" }, blk_log_unplug },
1154 [__BLK_TA_UNPLUG_TIMER] = {{ "UT", "unplug_timer" }, blk_log_unplug },
1155 [__BLK_TA_INSERT] = {{ "I", "insert" }, blk_log_generic },
1156 [__BLK_TA_SPLIT] = {{ "X", "split" }, blk_log_split },
1157 [__BLK_TA_BOUNCE] = {{ "B", "bounce" }, blk_log_generic },
1158 [__BLK_TA_REMAP] = {{ "A", "remap" }, blk_log_remap },
1159};
1160
1161static enum print_line_t print_one_line(struct trace_iterator *iter,
1162 bool classic)
1163{
1164 struct trace_seq *s = &iter->seq;
1165 const struct blk_io_trace *t;
1166 u16 what;
1167 int ret;
1168 bool long_act;
1169 blk_log_action_t *log_action;
1170
1171 t = te_blk_io_trace(iter->ent);
1172 what = t->action & ((1 << BLK_TC_SHIFT) - 1);
1173 long_act = !!(trace_flags & TRACE_ITER_VERBOSE);
1174 log_action = classic ? &blk_log_action_classic : &blk_log_action;
1175
1176 if (t->action == BLK_TN_MESSAGE) {
1177 ret = log_action(iter, long_act ? "message" : "m");
1178 if (ret)
1179 ret = blk_log_msg(s, iter->ent);
1180 goto out;
1181 }
1182
1183 if (unlikely(what == 0 || what >= ARRAY_SIZE(what2act)))
1184 ret = trace_seq_printf(s, "Bad pc action %x\n", what);
1185 else {
1186 ret = log_action(iter, what2act[what].act[long_act]);
1187 if (ret)
1188 ret = what2act[what].print(s, iter->ent);
1189 }
1190out:
1191 return ret ? TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
1192}
1193
1194static enum print_line_t blk_trace_event_print(struct trace_iterator *iter,
1195 int flags)
1196{
1197 if (!trace_print_context(iter))
1198 return TRACE_TYPE_PARTIAL_LINE;
1199
1200 return print_one_line(iter, false);
1201}
1202
1203static int blk_trace_synthesize_old_trace(struct trace_iterator *iter)
1204{
1205 struct trace_seq *s = &iter->seq;
1206 struct blk_io_trace *t = (struct blk_io_trace *)iter->ent;
1207 const int offset = offsetof(struct blk_io_trace, sector);
1208 struct blk_io_trace old = {
1209 .magic = BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION,
1210 .time = iter->ts,
1211 };
1212
1213 if (!trace_seq_putmem(s, &old, offset))
1214 return 0;
1215 return trace_seq_putmem(s, &t->sector,
1216 sizeof(old) - offset + t->pdu_len);
1217}
1218
1219static enum print_line_t
1220blk_trace_event_print_binary(struct trace_iterator *iter, int flags)
1221{
1222 return blk_trace_synthesize_old_trace(iter) ?
1223 TRACE_TYPE_HANDLED : TRACE_TYPE_PARTIAL_LINE;
1224}
1225
1226static enum print_line_t blk_tracer_print_line(struct trace_iterator *iter)
1227{
1228 if (!(blk_tracer_flags.val & TRACE_BLK_OPT_CLASSIC))
1229 return TRACE_TYPE_UNHANDLED;
1230
1231 return print_one_line(iter, true);
1232}
1233
1234static struct tracer blk_tracer __read_mostly = {
1235 .name = "blk",
1236 .init = blk_tracer_init,
1237 .reset = blk_tracer_reset,
1238 .start = blk_tracer_start,
1239 .stop = blk_tracer_stop,
1240 .print_header = blk_tracer_print_header,
1241 .print_line = blk_tracer_print_line,
1242 .flags = &blk_tracer_flags,
1243};
1244
1245static struct trace_event trace_blk_event = {
1246 .type = TRACE_BLK,
1247 .trace = blk_trace_event_print,
1248 .binary = blk_trace_event_print_binary,
1249};
1250
1251static int __init init_blk_tracer(void)
1252{
1253 if (!register_ftrace_event(&trace_blk_event)) {
1254 pr_warning("Warning: could not register block events\n");
1255 return 1;
1256 }
1257
1258 if (register_tracer(&blk_tracer) != 0) {
1259 pr_warning("Warning: could not register the block tracer\n");
1260 unregister_ftrace_event(&trace_blk_event);
1261 return 1;
1262 }
1263
1264 return 0;
1265}
1266
1267device_initcall(init_blk_tracer);
1268
1269static int blk_trace_remove_queue(struct request_queue *q)
1270{
1271 struct blk_trace *bt;
1272
1273 bt = xchg(&q->blk_trace, NULL);
1274 if (bt == NULL)
1275 return -EINVAL;
1276
1277 if (atomic_dec_and_test(&blk_probes_ref))
1278 blk_unregister_tracepoints();
1279
1280 blk_trace_free(bt);
1281 return 0;
1282}
1283
1284/*
1285 * Setup everything required to start tracing
1286 */
1287static int blk_trace_setup_queue(struct request_queue *q, dev_t dev)
1288{
1289 struct blk_trace *old_bt, *bt = NULL;
1290 int ret = -ENOMEM;
1291
1292 bt = kzalloc(sizeof(*bt), GFP_KERNEL);
1293 if (!bt)
1294 return -ENOMEM;
1295
1296 bt->msg_data = __alloc_percpu(BLK_TN_MAX_MSG, __alignof__(char));
1297 if (!bt->msg_data)
1298 goto free_bt;
1299
1300 bt->dev = dev;
1301 bt->act_mask = (u16)-1;
1302 bt->end_lba = -1ULL;
1303
1304 old_bt = xchg(&q->blk_trace, bt);
1305 if (old_bt != NULL) {
1306 (void)xchg(&q->blk_trace, old_bt);
1307 ret = -EBUSY;
1308 goto free_bt;
1309 }
1310
1311 if (atomic_inc_return(&blk_probes_ref) == 1)
1312 blk_register_tracepoints();
1313 return 0;
1314
1315free_bt:
1316 blk_trace_free(bt);
1317 return ret;
1318}
1319
1320/*
1321 * sysfs interface to enable and configure tracing
1322 */
1323
1324static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1325 struct device_attribute *attr,
1326 char *buf);
1327static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1328 struct device_attribute *attr,
1329 const char *buf, size_t count);
1330#define BLK_TRACE_DEVICE_ATTR(_name) \
1331 DEVICE_ATTR(_name, S_IRUGO | S_IWUSR, \
1332 sysfs_blk_trace_attr_show, \
1333 sysfs_blk_trace_attr_store)
1334
1335static BLK_TRACE_DEVICE_ATTR(enable);
1336static BLK_TRACE_DEVICE_ATTR(act_mask);
1337static BLK_TRACE_DEVICE_ATTR(pid);
1338static BLK_TRACE_DEVICE_ATTR(start_lba);
1339static BLK_TRACE_DEVICE_ATTR(end_lba);
1340
1341static struct attribute *blk_trace_attrs[] = {
1342 &dev_attr_enable.attr,
1343 &dev_attr_act_mask.attr,
1344 &dev_attr_pid.attr,
1345 &dev_attr_start_lba.attr,
1346 &dev_attr_end_lba.attr,
1347 NULL
1348};
1349
1350struct attribute_group blk_trace_attr_group = {
1351 .name = "trace",
1352 .attrs = blk_trace_attrs,
1353};
1354
1355static const struct {
1356 int mask;
1357 const char *str;
1358} mask_maps[] = {
1359 { BLK_TC_READ, "read" },
1360 { BLK_TC_WRITE, "write" },
1361 { BLK_TC_BARRIER, "barrier" },
1362 { BLK_TC_SYNC, "sync" },
1363 { BLK_TC_QUEUE, "queue" },
1364 { BLK_TC_REQUEUE, "requeue" },
1365 { BLK_TC_ISSUE, "issue" },
1366 { BLK_TC_COMPLETE, "complete" },
1367 { BLK_TC_FS, "fs" },
1368 { BLK_TC_PC, "pc" },
1369 { BLK_TC_AHEAD, "ahead" },
1370 { BLK_TC_META, "meta" },
1371 { BLK_TC_DISCARD, "discard" },
1372 { BLK_TC_DRV_DATA, "drv_data" },
1373};
1374
1375static int blk_trace_str2mask(const char *str)
1376{
1377 int i;
1378 int mask = 0;
1379 char *s, *token;
1380
1381 s = kstrdup(str, GFP_KERNEL);
1382 if (s == NULL)
1383 return -ENOMEM;
1384 s = strstrip(s);
1385
1386 while (1) {
1387 token = strsep(&s, ",");
1388 if (token == NULL)
1389 break;
1390
1391 if (*token == '\0')
1392 continue;
1393
1394 for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1395 if (strcasecmp(token, mask_maps[i].str) == 0) {
1396 mask |= mask_maps[i].mask;
1397 break;
1398 }
1399 }
1400 if (i == ARRAY_SIZE(mask_maps)) {
1401 mask = -EINVAL;
1402 break;
1403 }
1404 }
1405 kfree(s);
1406
1407 return mask;
1408}
1409
1410static ssize_t blk_trace_mask2str(char *buf, int mask)
1411{
1412 int i;
1413 char *p = buf;
1414
1415 for (i = 0; i < ARRAY_SIZE(mask_maps); i++) {
1416 if (mask & mask_maps[i].mask) {
1417 p += sprintf(p, "%s%s",
1418 (p == buf) ? "" : ",", mask_maps[i].str);
1419 }
1420 }
1421 *p++ = '\n';
1422
1423 return p - buf;
1424}
1425
1426static struct request_queue *blk_trace_get_queue(struct block_device *bdev)
1427{
1428 if (bdev->bd_disk == NULL)
1429 return NULL;
1430
1431 return bdev_get_queue(bdev);
1432}
1433
1434static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1435 struct device_attribute *attr,
1436 char *buf)
1437{
1438 struct hd_struct *p = dev_to_part(dev);
1439 struct request_queue *q;
1440 struct block_device *bdev;
1441 ssize_t ret = -ENXIO;
1442
1443 lock_kernel();
1444 bdev = bdget(part_devt(p));
1445 if (bdev == NULL)
1446 goto out_unlock_kernel;
1447
1448 q = blk_trace_get_queue(bdev);
1449 if (q == NULL)
1450 goto out_bdput;
1451
1452 mutex_lock(&bdev->bd_mutex);
1453
1454 if (attr == &dev_attr_enable) {
1455 ret = sprintf(buf, "%u\n", !!q->blk_trace);
1456 goto out_unlock_bdev;
1457 }
1458
1459 if (q->blk_trace == NULL)
1460 ret = sprintf(buf, "disabled\n");
1461 else if (attr == &dev_attr_act_mask)
1462 ret = blk_trace_mask2str(buf, q->blk_trace->act_mask);
1463 else if (attr == &dev_attr_pid)
1464 ret = sprintf(buf, "%u\n", q->blk_trace->pid);
1465 else if (attr == &dev_attr_start_lba)
1466 ret = sprintf(buf, "%llu\n", q->blk_trace->start_lba);
1467 else if (attr == &dev_attr_end_lba)
1468 ret = sprintf(buf, "%llu\n", q->blk_trace->end_lba);
1469
1470out_unlock_bdev:
1471 mutex_unlock(&bdev->bd_mutex);
1472out_bdput:
1473 bdput(bdev);
1474out_unlock_kernel:
1475 unlock_kernel();
1476 return ret;
1477}
1478
1479static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1480 struct device_attribute *attr,
1481 const char *buf, size_t count)
1482{
1483 struct block_device *bdev;
1484 struct request_queue *q;
1485 struct hd_struct *p;
1486 u64 value;
1487 ssize_t ret = -EINVAL;
1488
1489 if (count == 0)
1490 goto out;
1491
1492 if (attr == &dev_attr_act_mask) {
1493 if (sscanf(buf, "%llx", &value) != 1) {
1494 /* Assume it is a list of trace category names */
1495 ret = blk_trace_str2mask(buf);
1496 if (ret < 0)
1497 goto out;
1498 value = ret;
1499 }
1500 } else if (sscanf(buf, "%llu", &value) != 1)
1501 goto out;
1502
1503 ret = -ENXIO;
1504
1505 lock_kernel();
1506 p = dev_to_part(dev);
1507 bdev = bdget(part_devt(p));
1508 if (bdev == NULL)
1509 goto out_unlock_kernel;
1510
1511 q = blk_trace_get_queue(bdev);
1512 if (q == NULL)
1513 goto out_bdput;
1514
1515 mutex_lock(&bdev->bd_mutex);
1516
1517 if (attr == &dev_attr_enable) {
1518 if (value)
1519 ret = blk_trace_setup_queue(q, bdev->bd_dev);
1520 else
1521 ret = blk_trace_remove_queue(q);
1522 goto out_unlock_bdev;
1523 }
1524
1525 ret = 0;
1526 if (q->blk_trace == NULL)
1527 ret = blk_trace_setup_queue(q, bdev->bd_dev);
1528
1529 if (ret == 0) {
1530 if (attr == &dev_attr_act_mask)
1531 q->blk_trace->act_mask = value;
1532 else if (attr == &dev_attr_pid)
1533 q->blk_trace->pid = value;
1534 else if (attr == &dev_attr_start_lba)
1535 q->blk_trace->start_lba = value;
1536 else if (attr == &dev_attr_end_lba)
1537 q->blk_trace->end_lba = value;
1538 }
1539
1540out_unlock_bdev:
1541 mutex_unlock(&bdev->bd_mutex);
1542out_bdput:
1543 bdput(bdev);
1544out_unlock_kernel:
1545 unlock_kernel();
1546out:
1547 return ret ? ret : count;
1548}
1549
diff --git a/kernel/trace/events.c b/kernel/trace/events.c
new file mode 100644
index 000000000000..246f2aa6dc46
--- /dev/null
+++ b/kernel/trace/events.c
@@ -0,0 +1,14 @@
1/*
2 * This is the place to register all trace points as events.
3 */
4
5#include <linux/stringify.h>
6
7#include <trace/trace_events.h>
8
9#include "trace_output.h"
10
11#include "trace_events_stage_1.h"
12#include "trace_events_stage_2.h"
13#include "trace_events_stage_3.h"
14
diff --git a/kernel/trace/ftrace.c b/kernel/trace/ftrace.c
index fdf913dfc7e8..f1ed080406c3 100644
--- a/kernel/trace/ftrace.c
+++ b/kernel/trace/ftrace.c
@@ -27,6 +27,9 @@
27#include <linux/sysctl.h> 27#include <linux/sysctl.h>
28#include <linux/ctype.h> 28#include <linux/ctype.h>
29#include <linux/list.h> 29#include <linux/list.h>
30#include <linux/hash.h>
31
32#include <trace/sched.h>
30 33
31#include <asm/ftrace.h> 34#include <asm/ftrace.h>
32 35
@@ -44,14 +47,14 @@
44 ftrace_kill(); \ 47 ftrace_kill(); \
45 } while (0) 48 } while (0)
46 49
50/* hash bits for specific function selection */
51#define FTRACE_HASH_BITS 7
52#define FTRACE_FUNC_HASHSIZE (1 << FTRACE_HASH_BITS)
53
47/* ftrace_enabled is a method to turn ftrace on or off */ 54/* ftrace_enabled is a method to turn ftrace on or off */
48int ftrace_enabled __read_mostly; 55int ftrace_enabled __read_mostly;
49static int last_ftrace_enabled; 56static int last_ftrace_enabled;
50 57
51/* set when tracing only a pid */
52struct pid *ftrace_pid_trace;
53static struct pid * const ftrace_swapper_pid = &init_struct_pid;
54
55/* Quick disabling of function tracer. */ 58/* Quick disabling of function tracer. */
56int function_trace_stop; 59int function_trace_stop;
57 60
@@ -61,9 +64,7 @@ int function_trace_stop;
61 */ 64 */
62static int ftrace_disabled __read_mostly; 65static int ftrace_disabled __read_mostly;
63 66
64static DEFINE_SPINLOCK(ftrace_lock); 67static DEFINE_MUTEX(ftrace_lock);
65static DEFINE_MUTEX(ftrace_sysctl_lock);
66static DEFINE_MUTEX(ftrace_start_lock);
67 68
68static struct ftrace_ops ftrace_list_end __read_mostly = 69static struct ftrace_ops ftrace_list_end __read_mostly =
69{ 70{
@@ -134,9 +135,6 @@ static void ftrace_test_stop_func(unsigned long ip, unsigned long parent_ip)
134 135
135static int __register_ftrace_function(struct ftrace_ops *ops) 136static int __register_ftrace_function(struct ftrace_ops *ops)
136{ 137{
137 /* should not be called from interrupt context */
138 spin_lock(&ftrace_lock);
139
140 ops->next = ftrace_list; 138 ops->next = ftrace_list;
141 /* 139 /*
142 * We are entering ops into the ftrace_list but another 140 * We are entering ops into the ftrace_list but another
@@ -172,18 +170,12 @@ static int __register_ftrace_function(struct ftrace_ops *ops)
172#endif 170#endif
173 } 171 }
174 172
175 spin_unlock(&ftrace_lock);
176
177 return 0; 173 return 0;
178} 174}
179 175
180static int __unregister_ftrace_function(struct ftrace_ops *ops) 176static int __unregister_ftrace_function(struct ftrace_ops *ops)
181{ 177{
182 struct ftrace_ops **p; 178 struct ftrace_ops **p;
183 int ret = 0;
184
185 /* should not be called from interrupt context */
186 spin_lock(&ftrace_lock);
187 179
188 /* 180 /*
189 * If we are removing the last function, then simply point 181 * If we are removing the last function, then simply point
@@ -192,17 +184,15 @@ static int __unregister_ftrace_function(struct ftrace_ops *ops)
192 if (ftrace_list == ops && ops->next == &ftrace_list_end) { 184 if (ftrace_list == ops && ops->next == &ftrace_list_end) {
193 ftrace_trace_function = ftrace_stub; 185 ftrace_trace_function = ftrace_stub;
194 ftrace_list = &ftrace_list_end; 186 ftrace_list = &ftrace_list_end;
195 goto out; 187 return 0;
196 } 188 }
197 189
198 for (p = &ftrace_list; *p != &ftrace_list_end; p = &(*p)->next) 190 for (p = &ftrace_list; *p != &ftrace_list_end; p = &(*p)->next)
199 if (*p == ops) 191 if (*p == ops)
200 break; 192 break;
201 193
202 if (*p != ops) { 194 if (*p != ops)
203 ret = -1; 195 return -1;
204 goto out;
205 }
206 196
207 *p = (*p)->next; 197 *p = (*p)->next;
208 198
@@ -223,21 +213,15 @@ static int __unregister_ftrace_function(struct ftrace_ops *ops)
223 } 213 }
224 } 214 }
225 215
226 out: 216 return 0;
227 spin_unlock(&ftrace_lock);
228
229 return ret;
230} 217}
231 218
232static void ftrace_update_pid_func(void) 219static void ftrace_update_pid_func(void)
233{ 220{
234 ftrace_func_t func; 221 ftrace_func_t func;
235 222
236 /* should not be called from interrupt context */
237 spin_lock(&ftrace_lock);
238
239 if (ftrace_trace_function == ftrace_stub) 223 if (ftrace_trace_function == ftrace_stub)
240 goto out; 224 return;
241 225
242 func = ftrace_trace_function; 226 func = ftrace_trace_function;
243 227
@@ -254,23 +238,29 @@ static void ftrace_update_pid_func(void)
254#else 238#else
255 __ftrace_trace_function = func; 239 __ftrace_trace_function = func;
256#endif 240#endif
257
258 out:
259 spin_unlock(&ftrace_lock);
260} 241}
261 242
243/* set when tracing only a pid */
244struct pid *ftrace_pid_trace;
245static struct pid * const ftrace_swapper_pid = &init_struct_pid;
246
262#ifdef CONFIG_DYNAMIC_FTRACE 247#ifdef CONFIG_DYNAMIC_FTRACE
248
263#ifndef CONFIG_FTRACE_MCOUNT_RECORD 249#ifndef CONFIG_FTRACE_MCOUNT_RECORD
264# error Dynamic ftrace depends on MCOUNT_RECORD 250# error Dynamic ftrace depends on MCOUNT_RECORD
265#endif 251#endif
266 252
267/* 253static struct hlist_head ftrace_func_hash[FTRACE_FUNC_HASHSIZE] __read_mostly;
268 * Since MCOUNT_ADDR may point to mcount itself, we do not want 254
269 * to get it confused by reading a reference in the code as we 255struct ftrace_func_probe {
270 * are parsing on objcopy output of text. Use a variable for 256 struct hlist_node node;
271 * it instead. 257 struct ftrace_probe_ops *ops;
272 */ 258 unsigned long flags;
273static unsigned long mcount_addr = MCOUNT_ADDR; 259 unsigned long ip;
260 void *data;
261 struct rcu_head rcu;
262};
263
274 264
275enum { 265enum {
276 FTRACE_ENABLE_CALLS = (1 << 0), 266 FTRACE_ENABLE_CALLS = (1 << 0),
@@ -284,13 +274,13 @@ enum {
284 274
285static int ftrace_filtered; 275static int ftrace_filtered;
286 276
287static LIST_HEAD(ftrace_new_addrs); 277static struct dyn_ftrace *ftrace_new_addrs;
288 278
289static DEFINE_MUTEX(ftrace_regex_lock); 279static DEFINE_MUTEX(ftrace_regex_lock);
290 280
291struct ftrace_page { 281struct ftrace_page {
292 struct ftrace_page *next; 282 struct ftrace_page *next;
293 unsigned long index; 283 int index;
294 struct dyn_ftrace records[]; 284 struct dyn_ftrace records[];
295}; 285};
296 286
@@ -305,6 +295,19 @@ static struct ftrace_page *ftrace_pages;
305 295
306static struct dyn_ftrace *ftrace_free_records; 296static struct dyn_ftrace *ftrace_free_records;
307 297
298/*
299 * This is a double for. Do not use 'break' to break out of the loop,
300 * you must use a goto.
301 */
302#define do_for_each_ftrace_rec(pg, rec) \
303 for (pg = ftrace_pages_start; pg; pg = pg->next) { \
304 int _____i; \
305 for (_____i = 0; _____i < pg->index; _____i++) { \
306 rec = &pg->records[_____i];
307
308#define while_for_each_ftrace_rec() \
309 } \
310 }
308 311
309#ifdef CONFIG_KPROBES 312#ifdef CONFIG_KPROBES
310 313
@@ -338,7 +341,7 @@ static inline int record_frozen(struct dyn_ftrace *rec)
338 341
339static void ftrace_free_rec(struct dyn_ftrace *rec) 342static void ftrace_free_rec(struct dyn_ftrace *rec)
340{ 343{
341 rec->ip = (unsigned long)ftrace_free_records; 344 rec->freelist = ftrace_free_records;
342 ftrace_free_records = rec; 345 ftrace_free_records = rec;
343 rec->flags |= FTRACE_FL_FREE; 346 rec->flags |= FTRACE_FL_FREE;
344} 347}
@@ -349,23 +352,22 @@ void ftrace_release(void *start, unsigned long size)
349 struct ftrace_page *pg; 352 struct ftrace_page *pg;
350 unsigned long s = (unsigned long)start; 353 unsigned long s = (unsigned long)start;
351 unsigned long e = s + size; 354 unsigned long e = s + size;
352 int i;
353 355
354 if (ftrace_disabled || !start) 356 if (ftrace_disabled || !start)
355 return; 357 return;
356 358
357 /* should not be called from interrupt context */ 359 mutex_lock(&ftrace_lock);
358 spin_lock(&ftrace_lock); 360 do_for_each_ftrace_rec(pg, rec) {
359 361 if ((rec->ip >= s) && (rec->ip < e)) {
360 for (pg = ftrace_pages_start; pg; pg = pg->next) { 362 /*
361 for (i = 0; i < pg->index; i++) { 363 * rec->ip is changed in ftrace_free_rec()
362 rec = &pg->records[i]; 364 * It should not between s and e if record was freed.
363 365 */
364 if ((rec->ip >= s) && (rec->ip < e)) 366 FTRACE_WARN_ON(rec->flags & FTRACE_FL_FREE);
365 ftrace_free_rec(rec); 367 ftrace_free_rec(rec);
366 } 368 }
367 } 369 } while_for_each_ftrace_rec();
368 spin_unlock(&ftrace_lock); 370 mutex_unlock(&ftrace_lock);
369} 371}
370 372
371static struct dyn_ftrace *ftrace_alloc_dyn_node(unsigned long ip) 373static struct dyn_ftrace *ftrace_alloc_dyn_node(unsigned long ip)
@@ -382,7 +384,7 @@ static struct dyn_ftrace *ftrace_alloc_dyn_node(unsigned long ip)
382 return NULL; 384 return NULL;
383 } 385 }
384 386
385 ftrace_free_records = (void *)rec->ip; 387 ftrace_free_records = rec->freelist;
386 memset(rec, 0, sizeof(*rec)); 388 memset(rec, 0, sizeof(*rec));
387 return rec; 389 return rec;
388 } 390 }
@@ -414,8 +416,8 @@ ftrace_record_ip(unsigned long ip)
414 return NULL; 416 return NULL;
415 417
416 rec->ip = ip; 418 rec->ip = ip;
417 419 rec->newlist = ftrace_new_addrs;
418 list_add(&rec->list, &ftrace_new_addrs); 420 ftrace_new_addrs = rec;
419 421
420 return rec; 422 return rec;
421} 423}
@@ -461,10 +463,10 @@ static void ftrace_bug(int failed, unsigned long ip)
461static int 463static int
462__ftrace_replace_code(struct dyn_ftrace *rec, int enable) 464__ftrace_replace_code(struct dyn_ftrace *rec, int enable)
463{ 465{
464 unsigned long ip, fl;
465 unsigned long ftrace_addr; 466 unsigned long ftrace_addr;
467 unsigned long ip, fl;
466 468
467 ftrace_addr = (unsigned long)ftrace_caller; 469 ftrace_addr = (unsigned long)FTRACE_ADDR;
468 470
469 ip = rec->ip; 471 ip = rec->ip;
470 472
@@ -473,7 +475,7 @@ __ftrace_replace_code(struct dyn_ftrace *rec, int enable)
473 * it is not enabled then do nothing. 475 * it is not enabled then do nothing.
474 * 476 *
475 * If this record is not to be traced and 477 * If this record is not to be traced and
476 * it is enabled then disabled it. 478 * it is enabled then disable it.
477 * 479 *
478 */ 480 */
479 if (rec->flags & FTRACE_FL_NOTRACE) { 481 if (rec->flags & FTRACE_FL_NOTRACE) {
@@ -493,7 +495,7 @@ __ftrace_replace_code(struct dyn_ftrace *rec, int enable)
493 if (fl == (FTRACE_FL_FILTER | FTRACE_FL_ENABLED)) 495 if (fl == (FTRACE_FL_FILTER | FTRACE_FL_ENABLED))
494 return 0; 496 return 0;
495 497
496 /* Record is not filtered and is not enabled do nothing */ 498 /* Record is not filtered or enabled, do nothing */
497 if (!fl) 499 if (!fl)
498 return 0; 500 return 0;
499 501
@@ -515,7 +517,7 @@ __ftrace_replace_code(struct dyn_ftrace *rec, int enable)
515 517
516 } else { 518 } else {
517 519
518 /* if record is not enabled do nothing */ 520 /* if record is not enabled, do nothing */
519 if (!(rec->flags & FTRACE_FL_ENABLED)) 521 if (!(rec->flags & FTRACE_FL_ENABLED))
520 return 0; 522 return 0;
521 523
@@ -531,41 +533,41 @@ __ftrace_replace_code(struct dyn_ftrace *rec, int enable)
531 533
532static void ftrace_replace_code(int enable) 534static void ftrace_replace_code(int enable)
533{ 535{
534 int i, failed;
535 struct dyn_ftrace *rec; 536 struct dyn_ftrace *rec;
536 struct ftrace_page *pg; 537 struct ftrace_page *pg;
538 int failed;
537 539
538 for (pg = ftrace_pages_start; pg; pg = pg->next) { 540 do_for_each_ftrace_rec(pg, rec) {
539 for (i = 0; i < pg->index; i++) { 541 /*
540 rec = &pg->records[i]; 542 * Skip over free records, records that have
541 543 * failed and not converted.
542 /* 544 */
543 * Skip over free records and records that have 545 if (rec->flags & FTRACE_FL_FREE ||
544 * failed. 546 rec->flags & FTRACE_FL_FAILED ||
545 */ 547 !(rec->flags & FTRACE_FL_CONVERTED))
546 if (rec->flags & FTRACE_FL_FREE || 548 continue;
547 rec->flags & FTRACE_FL_FAILED)
548 continue;
549 549
550 /* ignore updates to this record's mcount site */ 550 /* ignore updates to this record's mcount site */
551 if (get_kprobe((void *)rec->ip)) { 551 if (get_kprobe((void *)rec->ip)) {
552 freeze_record(rec); 552 freeze_record(rec);
553 continue; 553 continue;
554 } else { 554 } else {
555 unfreeze_record(rec); 555 unfreeze_record(rec);
556 } 556 }
557 557
558 failed = __ftrace_replace_code(rec, enable); 558 failed = __ftrace_replace_code(rec, enable);
559 if (failed && (rec->flags & FTRACE_FL_CONVERTED)) { 559 if (failed) {
560 rec->flags |= FTRACE_FL_FAILED; 560 rec->flags |= FTRACE_FL_FAILED;
561 if ((system_state == SYSTEM_BOOTING) || 561 if ((system_state == SYSTEM_BOOTING) ||
562 !core_kernel_text(rec->ip)) { 562 !core_kernel_text(rec->ip)) {
563 ftrace_free_rec(rec); 563 ftrace_free_rec(rec);
564 } else 564 } else {
565 ftrace_bug(failed, rec->ip); 565 ftrace_bug(failed, rec->ip);
566 } 566 /* Stop processing */
567 return;
568 }
567 } 569 }
568 } 570 } while_for_each_ftrace_rec();
569} 571}
570 572
571static int 573static int
@@ -576,7 +578,7 @@ ftrace_code_disable(struct module *mod, struct dyn_ftrace *rec)
576 578
577 ip = rec->ip; 579 ip = rec->ip;
578 580
579 ret = ftrace_make_nop(mod, rec, mcount_addr); 581 ret = ftrace_make_nop(mod, rec, MCOUNT_ADDR);
580 if (ret) { 582 if (ret) {
581 ftrace_bug(ret, ip); 583 ftrace_bug(ret, ip);
582 rec->flags |= FTRACE_FL_FAILED; 584 rec->flags |= FTRACE_FL_FAILED;
@@ -585,6 +587,24 @@ ftrace_code_disable(struct module *mod, struct dyn_ftrace *rec)
585 return 1; 587 return 1;
586} 588}
587 589
590/*
591 * archs can override this function if they must do something
592 * before the modifying code is performed.
593 */
594int __weak ftrace_arch_code_modify_prepare(void)
595{
596 return 0;
597}
598
599/*
600 * archs can override this function if they must do something
601 * after the modifying code is performed.
602 */
603int __weak ftrace_arch_code_modify_post_process(void)
604{
605 return 0;
606}
607
588static int __ftrace_modify_code(void *data) 608static int __ftrace_modify_code(void *data)
589{ 609{
590 int *command = data; 610 int *command = data;
@@ -607,7 +627,17 @@ static int __ftrace_modify_code(void *data)
607 627
608static void ftrace_run_update_code(int command) 628static void ftrace_run_update_code(int command)
609{ 629{
630 int ret;
631
632 ret = ftrace_arch_code_modify_prepare();
633 FTRACE_WARN_ON(ret);
634 if (ret)
635 return;
636
610 stop_machine(__ftrace_modify_code, &command, NULL); 637 stop_machine(__ftrace_modify_code, &command, NULL);
638
639 ret = ftrace_arch_code_modify_post_process();
640 FTRACE_WARN_ON(ret);
611} 641}
612 642
613static ftrace_func_t saved_ftrace_func; 643static ftrace_func_t saved_ftrace_func;
@@ -631,13 +661,10 @@ static void ftrace_startup(int command)
631 if (unlikely(ftrace_disabled)) 661 if (unlikely(ftrace_disabled))
632 return; 662 return;
633 663
634 mutex_lock(&ftrace_start_lock);
635 ftrace_start_up++; 664 ftrace_start_up++;
636 command |= FTRACE_ENABLE_CALLS; 665 command |= FTRACE_ENABLE_CALLS;
637 666
638 ftrace_startup_enable(command); 667 ftrace_startup_enable(command);
639
640 mutex_unlock(&ftrace_start_lock);
641} 668}
642 669
643static void ftrace_shutdown(int command) 670static void ftrace_shutdown(int command)
@@ -645,7 +672,6 @@ static void ftrace_shutdown(int command)
645 if (unlikely(ftrace_disabled)) 672 if (unlikely(ftrace_disabled))
646 return; 673 return;
647 674
648 mutex_lock(&ftrace_start_lock);
649 ftrace_start_up--; 675 ftrace_start_up--;
650 if (!ftrace_start_up) 676 if (!ftrace_start_up)
651 command |= FTRACE_DISABLE_CALLS; 677 command |= FTRACE_DISABLE_CALLS;
@@ -656,11 +682,9 @@ static void ftrace_shutdown(int command)
656 } 682 }
657 683
658 if (!command || !ftrace_enabled) 684 if (!command || !ftrace_enabled)
659 goto out; 685 return;
660 686
661 ftrace_run_update_code(command); 687 ftrace_run_update_code(command);
662 out:
663 mutex_unlock(&ftrace_start_lock);
664} 688}
665 689
666static void ftrace_startup_sysctl(void) 690static void ftrace_startup_sysctl(void)
@@ -670,7 +694,6 @@ static void ftrace_startup_sysctl(void)
670 if (unlikely(ftrace_disabled)) 694 if (unlikely(ftrace_disabled))
671 return; 695 return;
672 696
673 mutex_lock(&ftrace_start_lock);
674 /* Force update next time */ 697 /* Force update next time */
675 saved_ftrace_func = NULL; 698 saved_ftrace_func = NULL;
676 /* ftrace_start_up is true if we want ftrace running */ 699 /* ftrace_start_up is true if we want ftrace running */
@@ -678,7 +701,6 @@ static void ftrace_startup_sysctl(void)
678 command |= FTRACE_ENABLE_CALLS; 701 command |= FTRACE_ENABLE_CALLS;
679 702
680 ftrace_run_update_code(command); 703 ftrace_run_update_code(command);
681 mutex_unlock(&ftrace_start_lock);
682} 704}
683 705
684static void ftrace_shutdown_sysctl(void) 706static void ftrace_shutdown_sysctl(void)
@@ -688,13 +710,11 @@ static void ftrace_shutdown_sysctl(void)
688 if (unlikely(ftrace_disabled)) 710 if (unlikely(ftrace_disabled))
689 return; 711 return;
690 712
691 mutex_lock(&ftrace_start_lock);
692 /* ftrace_start_up is true if ftrace is running */ 713 /* ftrace_start_up is true if ftrace is running */
693 if (ftrace_start_up) 714 if (ftrace_start_up)
694 command |= FTRACE_DISABLE_CALLS; 715 command |= FTRACE_DISABLE_CALLS;
695 716
696 ftrace_run_update_code(command); 717 ftrace_run_update_code(command);
697 mutex_unlock(&ftrace_start_lock);
698} 718}
699 719
700static cycle_t ftrace_update_time; 720static cycle_t ftrace_update_time;
@@ -703,19 +723,21 @@ unsigned long ftrace_update_tot_cnt;
703 723
704static int ftrace_update_code(struct module *mod) 724static int ftrace_update_code(struct module *mod)
705{ 725{
706 struct dyn_ftrace *p, *t; 726 struct dyn_ftrace *p;
707 cycle_t start, stop; 727 cycle_t start, stop;
708 728
709 start = ftrace_now(raw_smp_processor_id()); 729 start = ftrace_now(raw_smp_processor_id());
710 ftrace_update_cnt = 0; 730 ftrace_update_cnt = 0;
711 731
712 list_for_each_entry_safe(p, t, &ftrace_new_addrs, list) { 732 while (ftrace_new_addrs) {
713 733
714 /* If something went wrong, bail without enabling anything */ 734 /* If something went wrong, bail without enabling anything */
715 if (unlikely(ftrace_disabled)) 735 if (unlikely(ftrace_disabled))
716 return -1; 736 return -1;
717 737
718 list_del_init(&p->list); 738 p = ftrace_new_addrs;
739 ftrace_new_addrs = p->newlist;
740 p->flags = 0L;
719 741
720 /* convert record (i.e, patch mcount-call with NOP) */ 742 /* convert record (i.e, patch mcount-call with NOP) */
721 if (ftrace_code_disable(mod, p)) { 743 if (ftrace_code_disable(mod, p)) {
@@ -781,13 +803,16 @@ enum {
781 FTRACE_ITER_CONT = (1 << 1), 803 FTRACE_ITER_CONT = (1 << 1),
782 FTRACE_ITER_NOTRACE = (1 << 2), 804 FTRACE_ITER_NOTRACE = (1 << 2),
783 FTRACE_ITER_FAILURES = (1 << 3), 805 FTRACE_ITER_FAILURES = (1 << 3),
806 FTRACE_ITER_PRINTALL = (1 << 4),
807 FTRACE_ITER_HASH = (1 << 5),
784}; 808};
785 809
786#define FTRACE_BUFF_MAX (KSYM_SYMBOL_LEN+4) /* room for wildcards */ 810#define FTRACE_BUFF_MAX (KSYM_SYMBOL_LEN+4) /* room for wildcards */
787 811
788struct ftrace_iterator { 812struct ftrace_iterator {
789 struct ftrace_page *pg; 813 struct ftrace_page *pg;
790 unsigned idx; 814 int hidx;
815 int idx;
791 unsigned flags; 816 unsigned flags;
792 unsigned char buffer[FTRACE_BUFF_MAX+1]; 817 unsigned char buffer[FTRACE_BUFF_MAX+1];
793 unsigned buffer_idx; 818 unsigned buffer_idx;
@@ -795,15 +820,89 @@ struct ftrace_iterator {
795}; 820};
796 821
797static void * 822static void *
823t_hash_next(struct seq_file *m, void *v, loff_t *pos)
824{
825 struct ftrace_iterator *iter = m->private;
826 struct hlist_node *hnd = v;
827 struct hlist_head *hhd;
828
829 WARN_ON(!(iter->flags & FTRACE_ITER_HASH));
830
831 (*pos)++;
832
833 retry:
834 if (iter->hidx >= FTRACE_FUNC_HASHSIZE)
835 return NULL;
836
837 hhd = &ftrace_func_hash[iter->hidx];
838
839 if (hlist_empty(hhd)) {
840 iter->hidx++;
841 hnd = NULL;
842 goto retry;
843 }
844
845 if (!hnd)
846 hnd = hhd->first;
847 else {
848 hnd = hnd->next;
849 if (!hnd) {
850 iter->hidx++;
851 goto retry;
852 }
853 }
854
855 return hnd;
856}
857
858static void *t_hash_start(struct seq_file *m, loff_t *pos)
859{
860 struct ftrace_iterator *iter = m->private;
861 void *p = NULL;
862
863 iter->flags |= FTRACE_ITER_HASH;
864
865 return t_hash_next(m, p, pos);
866}
867
868static int t_hash_show(struct seq_file *m, void *v)
869{
870 struct ftrace_func_probe *rec;
871 struct hlist_node *hnd = v;
872 char str[KSYM_SYMBOL_LEN];
873
874 rec = hlist_entry(hnd, struct ftrace_func_probe, node);
875
876 if (rec->ops->print)
877 return rec->ops->print(m, rec->ip, rec->ops, rec->data);
878
879 kallsyms_lookup(rec->ip, NULL, NULL, NULL, str);
880 seq_printf(m, "%s:", str);
881
882 kallsyms_lookup((unsigned long)rec->ops->func, NULL, NULL, NULL, str);
883 seq_printf(m, "%s", str);
884
885 if (rec->data)
886 seq_printf(m, ":%p", rec->data);
887 seq_putc(m, '\n');
888
889 return 0;
890}
891
892static void *
798t_next(struct seq_file *m, void *v, loff_t *pos) 893t_next(struct seq_file *m, void *v, loff_t *pos)
799{ 894{
800 struct ftrace_iterator *iter = m->private; 895 struct ftrace_iterator *iter = m->private;
801 struct dyn_ftrace *rec = NULL; 896 struct dyn_ftrace *rec = NULL;
802 897
898 if (iter->flags & FTRACE_ITER_HASH)
899 return t_hash_next(m, v, pos);
900
803 (*pos)++; 901 (*pos)++;
804 902
805 /* should not be called from interrupt context */ 903 if (iter->flags & FTRACE_ITER_PRINTALL)
806 spin_lock(&ftrace_lock); 904 return NULL;
905
807 retry: 906 retry:
808 if (iter->idx >= iter->pg->index) { 907 if (iter->idx >= iter->pg->index) {
809 if (iter->pg->next) { 908 if (iter->pg->next) {
@@ -832,7 +931,6 @@ t_next(struct seq_file *m, void *v, loff_t *pos)
832 goto retry; 931 goto retry;
833 } 932 }
834 } 933 }
835 spin_unlock(&ftrace_lock);
836 934
837 return rec; 935 return rec;
838} 936}
@@ -842,6 +940,23 @@ static void *t_start(struct seq_file *m, loff_t *pos)
842 struct ftrace_iterator *iter = m->private; 940 struct ftrace_iterator *iter = m->private;
843 void *p = NULL; 941 void *p = NULL;
844 942
943 mutex_lock(&ftrace_lock);
944 /*
945 * For set_ftrace_filter reading, if we have the filter
946 * off, we can short cut and just print out that all
947 * functions are enabled.
948 */
949 if (iter->flags & FTRACE_ITER_FILTER && !ftrace_filtered) {
950 if (*pos > 0)
951 return t_hash_start(m, pos);
952 iter->flags |= FTRACE_ITER_PRINTALL;
953 (*pos)++;
954 return iter;
955 }
956
957 if (iter->flags & FTRACE_ITER_HASH)
958 return t_hash_start(m, pos);
959
845 if (*pos > 0) { 960 if (*pos > 0) {
846 if (iter->idx < 0) 961 if (iter->idx < 0)
847 return p; 962 return p;
@@ -851,18 +966,31 @@ static void *t_start(struct seq_file *m, loff_t *pos)
851 966
852 p = t_next(m, p, pos); 967 p = t_next(m, p, pos);
853 968
969 if (!p)
970 return t_hash_start(m, pos);
971
854 return p; 972 return p;
855} 973}
856 974
857static void t_stop(struct seq_file *m, void *p) 975static void t_stop(struct seq_file *m, void *p)
858{ 976{
977 mutex_unlock(&ftrace_lock);
859} 978}
860 979
861static int t_show(struct seq_file *m, void *v) 980static int t_show(struct seq_file *m, void *v)
862{ 981{
982 struct ftrace_iterator *iter = m->private;
863 struct dyn_ftrace *rec = v; 983 struct dyn_ftrace *rec = v;
864 char str[KSYM_SYMBOL_LEN]; 984 char str[KSYM_SYMBOL_LEN];
865 985
986 if (iter->flags & FTRACE_ITER_HASH)
987 return t_hash_show(m, v);
988
989 if (iter->flags & FTRACE_ITER_PRINTALL) {
990 seq_printf(m, "#### all functions enabled ####\n");
991 return 0;
992 }
993
866 if (!rec) 994 if (!rec)
867 return 0; 995 return 0;
868 996
@@ -941,23 +1069,16 @@ static void ftrace_filter_reset(int enable)
941 struct ftrace_page *pg; 1069 struct ftrace_page *pg;
942 struct dyn_ftrace *rec; 1070 struct dyn_ftrace *rec;
943 unsigned long type = enable ? FTRACE_FL_FILTER : FTRACE_FL_NOTRACE; 1071 unsigned long type = enable ? FTRACE_FL_FILTER : FTRACE_FL_NOTRACE;
944 unsigned i;
945 1072
946 /* should not be called from interrupt context */ 1073 mutex_lock(&ftrace_lock);
947 spin_lock(&ftrace_lock);
948 if (enable) 1074 if (enable)
949 ftrace_filtered = 0; 1075 ftrace_filtered = 0;
950 pg = ftrace_pages_start; 1076 do_for_each_ftrace_rec(pg, rec) {
951 while (pg) { 1077 if (rec->flags & FTRACE_FL_FAILED)
952 for (i = 0; i < pg->index; i++) { 1078 continue;
953 rec = &pg->records[i]; 1079 rec->flags &= ~type;
954 if (rec->flags & FTRACE_FL_FAILED) 1080 } while_for_each_ftrace_rec();
955 continue; 1081 mutex_unlock(&ftrace_lock);
956 rec->flags &= ~type;
957 }
958 pg = pg->next;
959 }
960 spin_unlock(&ftrace_lock);
961} 1082}
962 1083
963static int 1084static int
@@ -1008,16 +1129,6 @@ ftrace_notrace_open(struct inode *inode, struct file *file)
1008 return ftrace_regex_open(inode, file, 0); 1129 return ftrace_regex_open(inode, file, 0);
1009} 1130}
1010 1131
1011static ssize_t
1012ftrace_regex_read(struct file *file, char __user *ubuf,
1013 size_t cnt, loff_t *ppos)
1014{
1015 if (file->f_mode & FMODE_READ)
1016 return seq_read(file, ubuf, cnt, ppos);
1017 else
1018 return -EPERM;
1019}
1020
1021static loff_t 1132static loff_t
1022ftrace_regex_lseek(struct file *file, loff_t offset, int origin) 1133ftrace_regex_lseek(struct file *file, loff_t offset, int origin)
1023{ 1134{
@@ -1038,86 +1149,536 @@ enum {
1038 MATCH_END_ONLY, 1149 MATCH_END_ONLY,
1039}; 1150};
1040 1151
1041static void 1152/*
1042ftrace_match(unsigned char *buff, int len, int enable) 1153 * (static function - no need for kernel doc)
1154 *
1155 * Pass in a buffer containing a glob and this function will
1156 * set search to point to the search part of the buffer and
1157 * return the type of search it is (see enum above).
1158 * This does modify buff.
1159 *
1160 * Returns enum type.
1161 * search returns the pointer to use for comparison.
1162 * not returns 1 if buff started with a '!'
1163 * 0 otherwise.
1164 */
1165static int
1166ftrace_setup_glob(char *buff, int len, char **search, int *not)
1043{ 1167{
1044 char str[KSYM_SYMBOL_LEN];
1045 char *search = NULL;
1046 struct ftrace_page *pg;
1047 struct dyn_ftrace *rec;
1048 int type = MATCH_FULL; 1168 int type = MATCH_FULL;
1049 unsigned long flag = enable ? FTRACE_FL_FILTER : FTRACE_FL_NOTRACE; 1169 int i;
1050 unsigned i, match = 0, search_len = 0;
1051 int not = 0;
1052 1170
1053 if (buff[0] == '!') { 1171 if (buff[0] == '!') {
1054 not = 1; 1172 *not = 1;
1055 buff++; 1173 buff++;
1056 len--; 1174 len--;
1057 } 1175 } else
1176 *not = 0;
1177
1178 *search = buff;
1058 1179
1059 for (i = 0; i < len; i++) { 1180 for (i = 0; i < len; i++) {
1060 if (buff[i] == '*') { 1181 if (buff[i] == '*') {
1061 if (!i) { 1182 if (!i) {
1062 search = buff + i + 1; 1183 *search = buff + 1;
1063 type = MATCH_END_ONLY; 1184 type = MATCH_END_ONLY;
1064 search_len = len - (i + 1);
1065 } else { 1185 } else {
1066 if (type == MATCH_END_ONLY) { 1186 if (type == MATCH_END_ONLY)
1067 type = MATCH_MIDDLE_ONLY; 1187 type = MATCH_MIDDLE_ONLY;
1068 } else { 1188 else
1069 match = i;
1070 type = MATCH_FRONT_ONLY; 1189 type = MATCH_FRONT_ONLY;
1071 }
1072 buff[i] = 0; 1190 buff[i] = 0;
1073 break; 1191 break;
1074 } 1192 }
1075 } 1193 }
1076 } 1194 }
1077 1195
1078 /* should not be called from interrupt context */ 1196 return type;
1079 spin_lock(&ftrace_lock); 1197}
1080 if (enable) 1198
1081 ftrace_filtered = 1; 1199static int ftrace_match(char *str, char *regex, int len, int type)
1082 pg = ftrace_pages_start; 1200{
1083 while (pg) { 1201 int matched = 0;
1084 for (i = 0; i < pg->index; i++) { 1202 char *ptr;
1085 int matched = 0; 1203
1086 char *ptr; 1204 switch (type) {
1087 1205 case MATCH_FULL:
1088 rec = &pg->records[i]; 1206 if (strcmp(str, regex) == 0)
1089 if (rec->flags & FTRACE_FL_FAILED) 1207 matched = 1;
1208 break;
1209 case MATCH_FRONT_ONLY:
1210 if (strncmp(str, regex, len) == 0)
1211 matched = 1;
1212 break;
1213 case MATCH_MIDDLE_ONLY:
1214 if (strstr(str, regex))
1215 matched = 1;
1216 break;
1217 case MATCH_END_ONLY:
1218 ptr = strstr(str, regex);
1219 if (ptr && (ptr[len] == 0))
1220 matched = 1;
1221 break;
1222 }
1223
1224 return matched;
1225}
1226
1227static int
1228ftrace_match_record(struct dyn_ftrace *rec, char *regex, int len, int type)
1229{
1230 char str[KSYM_SYMBOL_LEN];
1231
1232 kallsyms_lookup(rec->ip, NULL, NULL, NULL, str);
1233 return ftrace_match(str, regex, len, type);
1234}
1235
1236static void ftrace_match_records(char *buff, int len, int enable)
1237{
1238 unsigned int search_len;
1239 struct ftrace_page *pg;
1240 struct dyn_ftrace *rec;
1241 unsigned long flag;
1242 char *search;
1243 int type;
1244 int not;
1245
1246 flag = enable ? FTRACE_FL_FILTER : FTRACE_FL_NOTRACE;
1247 type = ftrace_setup_glob(buff, len, &search, &not);
1248
1249 search_len = strlen(search);
1250
1251 mutex_lock(&ftrace_lock);
1252 do_for_each_ftrace_rec(pg, rec) {
1253
1254 if (rec->flags & FTRACE_FL_FAILED)
1255 continue;
1256
1257 if (ftrace_match_record(rec, search, search_len, type)) {
1258 if (not)
1259 rec->flags &= ~flag;
1260 else
1261 rec->flags |= flag;
1262 }
1263 /*
1264 * Only enable filtering if we have a function that
1265 * is filtered on.
1266 */
1267 if (enable && (rec->flags & FTRACE_FL_FILTER))
1268 ftrace_filtered = 1;
1269 } while_for_each_ftrace_rec();
1270 mutex_unlock(&ftrace_lock);
1271}
1272
1273static int
1274ftrace_match_module_record(struct dyn_ftrace *rec, char *mod,
1275 char *regex, int len, int type)
1276{
1277 char str[KSYM_SYMBOL_LEN];
1278 char *modname;
1279
1280 kallsyms_lookup(rec->ip, NULL, NULL, &modname, str);
1281
1282 if (!modname || strcmp(modname, mod))
1283 return 0;
1284
1285 /* blank search means to match all funcs in the mod */
1286 if (len)
1287 return ftrace_match(str, regex, len, type);
1288 else
1289 return 1;
1290}
1291
1292static void ftrace_match_module_records(char *buff, char *mod, int enable)
1293{
1294 unsigned search_len = 0;
1295 struct ftrace_page *pg;
1296 struct dyn_ftrace *rec;
1297 int type = MATCH_FULL;
1298 char *search = buff;
1299 unsigned long flag;
1300 int not = 0;
1301
1302 flag = enable ? FTRACE_FL_FILTER : FTRACE_FL_NOTRACE;
1303
1304 /* blank or '*' mean the same */
1305 if (strcmp(buff, "*") == 0)
1306 buff[0] = 0;
1307
1308 /* handle the case of 'dont filter this module' */
1309 if (strcmp(buff, "!") == 0 || strcmp(buff, "!*") == 0) {
1310 buff[0] = 0;
1311 not = 1;
1312 }
1313
1314 if (strlen(buff)) {
1315 type = ftrace_setup_glob(buff, strlen(buff), &search, &not);
1316 search_len = strlen(search);
1317 }
1318
1319 mutex_lock(&ftrace_lock);
1320 do_for_each_ftrace_rec(pg, rec) {
1321
1322 if (rec->flags & FTRACE_FL_FAILED)
1323 continue;
1324
1325 if (ftrace_match_module_record(rec, mod,
1326 search, search_len, type)) {
1327 if (not)
1328 rec->flags &= ~flag;
1329 else
1330 rec->flags |= flag;
1331 }
1332 if (enable && (rec->flags & FTRACE_FL_FILTER))
1333 ftrace_filtered = 1;
1334
1335 } while_for_each_ftrace_rec();
1336 mutex_unlock(&ftrace_lock);
1337}
1338
1339/*
1340 * We register the module command as a template to show others how
1341 * to register the a command as well.
1342 */
1343
1344static int
1345ftrace_mod_callback(char *func, char *cmd, char *param, int enable)
1346{
1347 char *mod;
1348
1349 /*
1350 * cmd == 'mod' because we only registered this func
1351 * for the 'mod' ftrace_func_command.
1352 * But if you register one func with multiple commands,
1353 * you can tell which command was used by the cmd
1354 * parameter.
1355 */
1356
1357 /* we must have a module name */
1358 if (!param)
1359 return -EINVAL;
1360
1361 mod = strsep(&param, ":");
1362 if (!strlen(mod))
1363 return -EINVAL;
1364
1365 ftrace_match_module_records(func, mod, enable);
1366 return 0;
1367}
1368
1369static struct ftrace_func_command ftrace_mod_cmd = {
1370 .name = "mod",
1371 .func = ftrace_mod_callback,
1372};
1373
1374static int __init ftrace_mod_cmd_init(void)
1375{
1376 return register_ftrace_command(&ftrace_mod_cmd);
1377}
1378device_initcall(ftrace_mod_cmd_init);
1379
1380static void
1381function_trace_probe_call(unsigned long ip, unsigned long parent_ip)
1382{
1383 struct ftrace_func_probe *entry;
1384 struct hlist_head *hhd;
1385 struct hlist_node *n;
1386 unsigned long key;
1387 int resched;
1388
1389 key = hash_long(ip, FTRACE_HASH_BITS);
1390
1391 hhd = &ftrace_func_hash[key];
1392
1393 if (hlist_empty(hhd))
1394 return;
1395
1396 /*
1397 * Disable preemption for these calls to prevent a RCU grace
1398 * period. This syncs the hash iteration and freeing of items
1399 * on the hash. rcu_read_lock is too dangerous here.
1400 */
1401 resched = ftrace_preempt_disable();
1402 hlist_for_each_entry_rcu(entry, n, hhd, node) {
1403 if (entry->ip == ip)
1404 entry->ops->func(ip, parent_ip, &entry->data);
1405 }
1406 ftrace_preempt_enable(resched);
1407}
1408
1409static struct ftrace_ops trace_probe_ops __read_mostly =
1410{
1411 .func = function_trace_probe_call,
1412};
1413
1414static int ftrace_probe_registered;
1415
1416static void __enable_ftrace_function_probe(void)
1417{
1418 int i;
1419
1420 if (ftrace_probe_registered)
1421 return;
1422
1423 for (i = 0; i < FTRACE_FUNC_HASHSIZE; i++) {
1424 struct hlist_head *hhd = &ftrace_func_hash[i];
1425 if (hhd->first)
1426 break;
1427 }
1428 /* Nothing registered? */
1429 if (i == FTRACE_FUNC_HASHSIZE)
1430 return;
1431
1432 __register_ftrace_function(&trace_probe_ops);
1433 ftrace_startup(0);
1434 ftrace_probe_registered = 1;
1435}
1436
1437static void __disable_ftrace_function_probe(void)
1438{
1439 int i;
1440
1441 if (!ftrace_probe_registered)
1442 return;
1443
1444 for (i = 0; i < FTRACE_FUNC_HASHSIZE; i++) {
1445 struct hlist_head *hhd = &ftrace_func_hash[i];
1446 if (hhd->first)
1447 return;
1448 }
1449
1450 /* no more funcs left */
1451 __unregister_ftrace_function(&trace_probe_ops);
1452 ftrace_shutdown(0);
1453 ftrace_probe_registered = 0;
1454}
1455
1456
1457static void ftrace_free_entry_rcu(struct rcu_head *rhp)
1458{
1459 struct ftrace_func_probe *entry =
1460 container_of(rhp, struct ftrace_func_probe, rcu);
1461
1462 if (entry->ops->free)
1463 entry->ops->free(&entry->data);
1464 kfree(entry);
1465}
1466
1467
1468int
1469register_ftrace_function_probe(char *glob, struct ftrace_probe_ops *ops,
1470 void *data)
1471{
1472 struct ftrace_func_probe *entry;
1473 struct ftrace_page *pg;
1474 struct dyn_ftrace *rec;
1475 int type, len, not;
1476 unsigned long key;
1477 int count = 0;
1478 char *search;
1479
1480 type = ftrace_setup_glob(glob, strlen(glob), &search, &not);
1481 len = strlen(search);
1482
1483 /* we do not support '!' for function probes */
1484 if (WARN_ON(not))
1485 return -EINVAL;
1486
1487 mutex_lock(&ftrace_lock);
1488 do_for_each_ftrace_rec(pg, rec) {
1489
1490 if (rec->flags & FTRACE_FL_FAILED)
1491 continue;
1492
1493 if (!ftrace_match_record(rec, search, len, type))
1494 continue;
1495
1496 entry = kmalloc(sizeof(*entry), GFP_KERNEL);
1497 if (!entry) {
1498 /* If we did not process any, then return error */
1499 if (!count)
1500 count = -ENOMEM;
1501 goto out_unlock;
1502 }
1503
1504 count++;
1505
1506 entry->data = data;
1507
1508 /*
1509 * The caller might want to do something special
1510 * for each function we find. We call the callback
1511 * to give the caller an opportunity to do so.
1512 */
1513 if (ops->callback) {
1514 if (ops->callback(rec->ip, &entry->data) < 0) {
1515 /* caller does not like this func */
1516 kfree(entry);
1090 continue; 1517 continue;
1091 kallsyms_lookup(rec->ip, NULL, NULL, NULL, str);
1092 switch (type) {
1093 case MATCH_FULL:
1094 if (strcmp(str, buff) == 0)
1095 matched = 1;
1096 break;
1097 case MATCH_FRONT_ONLY:
1098 if (memcmp(str, buff, match) == 0)
1099 matched = 1;
1100 break;
1101 case MATCH_MIDDLE_ONLY:
1102 if (strstr(str, search))
1103 matched = 1;
1104 break;
1105 case MATCH_END_ONLY:
1106 ptr = strstr(str, search);
1107 if (ptr && (ptr[search_len] == 0))
1108 matched = 1;
1109 break;
1110 } 1518 }
1111 if (matched) { 1519 }
1112 if (not) 1520
1113 rec->flags &= ~flag; 1521 entry->ops = ops;
1114 else 1522 entry->ip = rec->ip;
1115 rec->flags |= flag; 1523
1524 key = hash_long(entry->ip, FTRACE_HASH_BITS);
1525 hlist_add_head_rcu(&entry->node, &ftrace_func_hash[key]);
1526
1527 } while_for_each_ftrace_rec();
1528 __enable_ftrace_function_probe();
1529
1530 out_unlock:
1531 mutex_unlock(&ftrace_lock);
1532
1533 return count;
1534}
1535
1536enum {
1537 PROBE_TEST_FUNC = 1,
1538 PROBE_TEST_DATA = 2
1539};
1540
1541static void
1542__unregister_ftrace_function_probe(char *glob, struct ftrace_probe_ops *ops,
1543 void *data, int flags)
1544{
1545 struct ftrace_func_probe *entry;
1546 struct hlist_node *n, *tmp;
1547 char str[KSYM_SYMBOL_LEN];
1548 int type = MATCH_FULL;
1549 int i, len = 0;
1550 char *search;
1551
1552 if (glob && (strcmp(glob, "*") || !strlen(glob)))
1553 glob = NULL;
1554 else {
1555 int not;
1556
1557 type = ftrace_setup_glob(glob, strlen(glob), &search, &not);
1558 len = strlen(search);
1559
1560 /* we do not support '!' for function probes */
1561 if (WARN_ON(not))
1562 return;
1563 }
1564
1565 mutex_lock(&ftrace_lock);
1566 for (i = 0; i < FTRACE_FUNC_HASHSIZE; i++) {
1567 struct hlist_head *hhd = &ftrace_func_hash[i];
1568
1569 hlist_for_each_entry_safe(entry, n, tmp, hhd, node) {
1570
1571 /* break up if statements for readability */
1572 if ((flags & PROBE_TEST_FUNC) && entry->ops != ops)
1573 continue;
1574
1575 if ((flags & PROBE_TEST_DATA) && entry->data != data)
1576 continue;
1577
1578 /* do this last, since it is the most expensive */
1579 if (glob) {
1580 kallsyms_lookup(entry->ip, NULL, NULL,
1581 NULL, str);
1582 if (!ftrace_match(str, glob, len, type))
1583 continue;
1116 } 1584 }
1585
1586 hlist_del(&entry->node);
1587 call_rcu(&entry->rcu, ftrace_free_entry_rcu);
1117 } 1588 }
1118 pg = pg->next;
1119 } 1589 }
1120 spin_unlock(&ftrace_lock); 1590 __disable_ftrace_function_probe();
1591 mutex_unlock(&ftrace_lock);
1592}
1593
1594void
1595unregister_ftrace_function_probe(char *glob, struct ftrace_probe_ops *ops,
1596 void *data)
1597{
1598 __unregister_ftrace_function_probe(glob, ops, data,
1599 PROBE_TEST_FUNC | PROBE_TEST_DATA);
1600}
1601
1602void
1603unregister_ftrace_function_probe_func(char *glob, struct ftrace_probe_ops *ops)
1604{
1605 __unregister_ftrace_function_probe(glob, ops, NULL, PROBE_TEST_FUNC);
1606}
1607
1608void unregister_ftrace_function_probe_all(char *glob)
1609{
1610 __unregister_ftrace_function_probe(glob, NULL, NULL, 0);
1611}
1612
1613static LIST_HEAD(ftrace_commands);
1614static DEFINE_MUTEX(ftrace_cmd_mutex);
1615
1616int register_ftrace_command(struct ftrace_func_command *cmd)
1617{
1618 struct ftrace_func_command *p;
1619 int ret = 0;
1620
1621 mutex_lock(&ftrace_cmd_mutex);
1622 list_for_each_entry(p, &ftrace_commands, list) {
1623 if (strcmp(cmd->name, p->name) == 0) {
1624 ret = -EBUSY;
1625 goto out_unlock;
1626 }
1627 }
1628 list_add(&cmd->list, &ftrace_commands);
1629 out_unlock:
1630 mutex_unlock(&ftrace_cmd_mutex);
1631
1632 return ret;
1633}
1634
1635int unregister_ftrace_command(struct ftrace_func_command *cmd)
1636{
1637 struct ftrace_func_command *p, *n;
1638 int ret = -ENODEV;
1639
1640 mutex_lock(&ftrace_cmd_mutex);
1641 list_for_each_entry_safe(p, n, &ftrace_commands, list) {
1642 if (strcmp(cmd->name, p->name) == 0) {
1643 ret = 0;
1644 list_del_init(&p->list);
1645 goto out_unlock;
1646 }
1647 }
1648 out_unlock:
1649 mutex_unlock(&ftrace_cmd_mutex);
1650
1651 return ret;
1652}
1653
1654static int ftrace_process_regex(char *buff, int len, int enable)
1655{
1656 char *func, *command, *next = buff;
1657 struct ftrace_func_command *p;
1658 int ret = -EINVAL;
1659
1660 func = strsep(&next, ":");
1661
1662 if (!next) {
1663 ftrace_match_records(func, len, enable);
1664 return 0;
1665 }
1666
1667 /* command found */
1668
1669 command = strsep(&next, ":");
1670
1671 mutex_lock(&ftrace_cmd_mutex);
1672 list_for_each_entry(p, &ftrace_commands, list) {
1673 if (strcmp(p->name, command) == 0) {
1674 ret = p->func(func, command, next, enable);
1675 goto out_unlock;
1676 }
1677 }
1678 out_unlock:
1679 mutex_unlock(&ftrace_cmd_mutex);
1680
1681 return ret;
1121} 1682}
1122 1683
1123static ssize_t 1684static ssize_t
@@ -1187,7 +1748,10 @@ ftrace_regex_write(struct file *file, const char __user *ubuf,
1187 if (isspace(ch)) { 1748 if (isspace(ch)) {
1188 iter->filtered++; 1749 iter->filtered++;
1189 iter->buffer[iter->buffer_idx] = 0; 1750 iter->buffer[iter->buffer_idx] = 0;
1190 ftrace_match(iter->buffer, iter->buffer_idx, enable); 1751 ret = ftrace_process_regex(iter->buffer,
1752 iter->buffer_idx, enable);
1753 if (ret)
1754 goto out;
1191 iter->buffer_idx = 0; 1755 iter->buffer_idx = 0;
1192 } else 1756 } else
1193 iter->flags |= FTRACE_ITER_CONT; 1757 iter->flags |= FTRACE_ITER_CONT;
@@ -1226,7 +1790,7 @@ ftrace_set_regex(unsigned char *buf, int len, int reset, int enable)
1226 if (reset) 1790 if (reset)
1227 ftrace_filter_reset(enable); 1791 ftrace_filter_reset(enable);
1228 if (buf) 1792 if (buf)
1229 ftrace_match(buf, len, enable); 1793 ftrace_match_records(buf, len, enable);
1230 mutex_unlock(&ftrace_regex_lock); 1794 mutex_unlock(&ftrace_regex_lock);
1231} 1795}
1232 1796
@@ -1276,15 +1840,13 @@ ftrace_regex_release(struct inode *inode, struct file *file, int enable)
1276 if (iter->buffer_idx) { 1840 if (iter->buffer_idx) {
1277 iter->filtered++; 1841 iter->filtered++;
1278 iter->buffer[iter->buffer_idx] = 0; 1842 iter->buffer[iter->buffer_idx] = 0;
1279 ftrace_match(iter->buffer, iter->buffer_idx, enable); 1843 ftrace_match_records(iter->buffer, iter->buffer_idx, enable);
1280 } 1844 }
1281 1845
1282 mutex_lock(&ftrace_sysctl_lock); 1846 mutex_lock(&ftrace_lock);
1283 mutex_lock(&ftrace_start_lock);
1284 if (ftrace_start_up && ftrace_enabled) 1847 if (ftrace_start_up && ftrace_enabled)
1285 ftrace_run_update_code(FTRACE_ENABLE_CALLS); 1848 ftrace_run_update_code(FTRACE_ENABLE_CALLS);
1286 mutex_unlock(&ftrace_start_lock); 1849 mutex_unlock(&ftrace_lock);
1287 mutex_unlock(&ftrace_sysctl_lock);
1288 1850
1289 kfree(iter); 1851 kfree(iter);
1290 mutex_unlock(&ftrace_regex_lock); 1852 mutex_unlock(&ftrace_regex_lock);
@@ -1303,31 +1865,31 @@ ftrace_notrace_release(struct inode *inode, struct file *file)
1303 return ftrace_regex_release(inode, file, 0); 1865 return ftrace_regex_release(inode, file, 0);
1304} 1866}
1305 1867
1306static struct file_operations ftrace_avail_fops = { 1868static const struct file_operations ftrace_avail_fops = {
1307 .open = ftrace_avail_open, 1869 .open = ftrace_avail_open,
1308 .read = seq_read, 1870 .read = seq_read,
1309 .llseek = seq_lseek, 1871 .llseek = seq_lseek,
1310 .release = ftrace_avail_release, 1872 .release = ftrace_avail_release,
1311}; 1873};
1312 1874
1313static struct file_operations ftrace_failures_fops = { 1875static const struct file_operations ftrace_failures_fops = {
1314 .open = ftrace_failures_open, 1876 .open = ftrace_failures_open,
1315 .read = seq_read, 1877 .read = seq_read,
1316 .llseek = seq_lseek, 1878 .llseek = seq_lseek,
1317 .release = ftrace_avail_release, 1879 .release = ftrace_avail_release,
1318}; 1880};
1319 1881
1320static struct file_operations ftrace_filter_fops = { 1882static const struct file_operations ftrace_filter_fops = {
1321 .open = ftrace_filter_open, 1883 .open = ftrace_filter_open,
1322 .read = ftrace_regex_read, 1884 .read = seq_read,
1323 .write = ftrace_filter_write, 1885 .write = ftrace_filter_write,
1324 .llseek = ftrace_regex_lseek, 1886 .llseek = ftrace_regex_lseek,
1325 .release = ftrace_filter_release, 1887 .release = ftrace_filter_release,
1326}; 1888};
1327 1889
1328static struct file_operations ftrace_notrace_fops = { 1890static const struct file_operations ftrace_notrace_fops = {
1329 .open = ftrace_notrace_open, 1891 .open = ftrace_notrace_open,
1330 .read = ftrace_regex_read, 1892 .read = seq_read,
1331 .write = ftrace_notrace_write, 1893 .write = ftrace_notrace_write,
1332 .llseek = ftrace_regex_lseek, 1894 .llseek = ftrace_regex_lseek,
1333 .release = ftrace_notrace_release, 1895 .release = ftrace_notrace_release,
@@ -1360,6 +1922,10 @@ static void *g_start(struct seq_file *m, loff_t *pos)
1360 1922
1361 mutex_lock(&graph_lock); 1923 mutex_lock(&graph_lock);
1362 1924
1925 /* Nothing, tell g_show to print all functions are enabled */
1926 if (!ftrace_graph_count && !*pos)
1927 return (void *)1;
1928
1363 p = g_next(m, p, pos); 1929 p = g_next(m, p, pos);
1364 1930
1365 return p; 1931 return p;
@@ -1378,6 +1944,11 @@ static int g_show(struct seq_file *m, void *v)
1378 if (!ptr) 1944 if (!ptr)
1379 return 0; 1945 return 0;
1380 1946
1947 if (ptr == (unsigned long *)1) {
1948 seq_printf(m, "#### all functions enabled ####\n");
1949 return 0;
1950 }
1951
1381 kallsyms_lookup(*ptr, NULL, NULL, NULL, str); 1952 kallsyms_lookup(*ptr, NULL, NULL, NULL, str);
1382 1953
1383 seq_printf(m, "%s\n", str); 1954 seq_printf(m, "%s\n", str);
@@ -1420,53 +1991,53 @@ ftrace_graph_open(struct inode *inode, struct file *file)
1420 return ret; 1991 return ret;
1421} 1992}
1422 1993
1423static ssize_t
1424ftrace_graph_read(struct file *file, char __user *ubuf,
1425 size_t cnt, loff_t *ppos)
1426{
1427 if (file->f_mode & FMODE_READ)
1428 return seq_read(file, ubuf, cnt, ppos);
1429 else
1430 return -EPERM;
1431}
1432
1433static int 1994static int
1434ftrace_set_func(unsigned long *array, int idx, char *buffer) 1995ftrace_set_func(unsigned long *array, int *idx, char *buffer)
1435{ 1996{
1436 char str[KSYM_SYMBOL_LEN];
1437 struct dyn_ftrace *rec; 1997 struct dyn_ftrace *rec;
1438 struct ftrace_page *pg; 1998 struct ftrace_page *pg;
1999 int search_len;
1439 int found = 0; 2000 int found = 0;
1440 int i, j; 2001 int type, not;
2002 char *search;
2003 bool exists;
2004 int i;
1441 2005
1442 if (ftrace_disabled) 2006 if (ftrace_disabled)
1443 return -ENODEV; 2007 return -ENODEV;
1444 2008
1445 /* should not be called from interrupt context */ 2009 /* decode regex */
1446 spin_lock(&ftrace_lock); 2010 type = ftrace_setup_glob(buffer, strlen(buffer), &search, &not);
2011 if (not)
2012 return -EINVAL;
2013
2014 search_len = strlen(search);
1447 2015
1448 for (pg = ftrace_pages_start; pg; pg = pg->next) { 2016 mutex_lock(&ftrace_lock);
1449 for (i = 0; i < pg->index; i++) { 2017 do_for_each_ftrace_rec(pg, rec) {
1450 rec = &pg->records[i];
1451 2018
1452 if (rec->flags & (FTRACE_FL_FAILED | FTRACE_FL_FREE)) 2019 if (*idx >= FTRACE_GRAPH_MAX_FUNCS)
1453 continue; 2020 break;
2021
2022 if (rec->flags & (FTRACE_FL_FAILED | FTRACE_FL_FREE))
2023 continue;
1454 2024
1455 kallsyms_lookup(rec->ip, NULL, NULL, NULL, str); 2025 if (ftrace_match_record(rec, search, search_len, type)) {
1456 if (strcmp(str, buffer) == 0) { 2026 /* ensure it is not already in the array */
2027 exists = false;
2028 for (i = 0; i < *idx; i++)
2029 if (array[i] == rec->ip) {
2030 exists = true;
2031 break;
2032 }
2033 if (!exists) {
2034 array[(*idx)++] = rec->ip;
1457 found = 1; 2035 found = 1;
1458 for (j = 0; j < idx; j++)
1459 if (array[j] == rec->ip) {
1460 found = 0;
1461 break;
1462 }
1463 if (found)
1464 array[idx] = rec->ip;
1465 break;
1466 } 2036 }
1467 } 2037 }
1468 } 2038 } while_for_each_ftrace_rec();
1469 spin_unlock(&ftrace_lock); 2039
2040 mutex_unlock(&ftrace_lock);
1470 2041
1471 return found ? 0 : -EINVAL; 2042 return found ? 0 : -EINVAL;
1472} 2043}
@@ -1534,13 +2105,11 @@ ftrace_graph_write(struct file *file, const char __user *ubuf,
1534 } 2105 }
1535 buffer[index] = 0; 2106 buffer[index] = 0;
1536 2107
1537 /* we allow only one at a time */ 2108 /* we allow only one expression at a time */
1538 ret = ftrace_set_func(array, ftrace_graph_count, buffer); 2109 ret = ftrace_set_func(array, &ftrace_graph_count, buffer);
1539 if (ret) 2110 if (ret)
1540 goto out; 2111 goto out;
1541 2112
1542 ftrace_graph_count++;
1543
1544 file->f_pos += read; 2113 file->f_pos += read;
1545 2114
1546 ret = read; 2115 ret = read;
@@ -1552,7 +2121,7 @@ ftrace_graph_write(struct file *file, const char __user *ubuf,
1552 2121
1553static const struct file_operations ftrace_graph_fops = { 2122static const struct file_operations ftrace_graph_fops = {
1554 .open = ftrace_graph_open, 2123 .open = ftrace_graph_open,
1555 .read = ftrace_graph_read, 2124 .read = seq_read,
1556 .write = ftrace_graph_write, 2125 .write = ftrace_graph_write,
1557}; 2126};
1558#endif /* CONFIG_FUNCTION_GRAPH_TRACER */ 2127#endif /* CONFIG_FUNCTION_GRAPH_TRACER */
@@ -1604,7 +2173,7 @@ static int ftrace_convert_nops(struct module *mod,
1604 unsigned long addr; 2173 unsigned long addr;
1605 unsigned long flags; 2174 unsigned long flags;
1606 2175
1607 mutex_lock(&ftrace_start_lock); 2176 mutex_lock(&ftrace_lock);
1608 p = start; 2177 p = start;
1609 while (p < end) { 2178 while (p < end) {
1610 addr = ftrace_call_adjust(*p++); 2179 addr = ftrace_call_adjust(*p++);
@@ -1623,7 +2192,7 @@ static int ftrace_convert_nops(struct module *mod,
1623 local_irq_save(flags); 2192 local_irq_save(flags);
1624 ftrace_update_code(mod); 2193 ftrace_update_code(mod);
1625 local_irq_restore(flags); 2194 local_irq_restore(flags);
1626 mutex_unlock(&ftrace_start_lock); 2195 mutex_unlock(&ftrace_lock);
1627 2196
1628 return 0; 2197 return 0;
1629} 2198}
@@ -1700,7 +2269,7 @@ ftrace_pid_read(struct file *file, char __user *ubuf,
1700 if (ftrace_pid_trace == ftrace_swapper_pid) 2269 if (ftrace_pid_trace == ftrace_swapper_pid)
1701 r = sprintf(buf, "swapper tasks\n"); 2270 r = sprintf(buf, "swapper tasks\n");
1702 else if (ftrace_pid_trace) 2271 else if (ftrace_pid_trace)
1703 r = sprintf(buf, "%u\n", pid_nr(ftrace_pid_trace)); 2272 r = sprintf(buf, "%u\n", pid_vnr(ftrace_pid_trace));
1704 else 2273 else
1705 r = sprintf(buf, "no pid\n"); 2274 r = sprintf(buf, "no pid\n");
1706 2275
@@ -1796,7 +2365,7 @@ ftrace_pid_write(struct file *filp, const char __user *ubuf,
1796 if (ret < 0) 2365 if (ret < 0)
1797 return ret; 2366 return ret;
1798 2367
1799 mutex_lock(&ftrace_start_lock); 2368 mutex_lock(&ftrace_lock);
1800 if (val < 0) { 2369 if (val < 0) {
1801 /* disable pid tracing */ 2370 /* disable pid tracing */
1802 if (!ftrace_pid_trace) 2371 if (!ftrace_pid_trace)
@@ -1835,12 +2404,12 @@ ftrace_pid_write(struct file *filp, const char __user *ubuf,
1835 ftrace_startup_enable(0); 2404 ftrace_startup_enable(0);
1836 2405
1837 out: 2406 out:
1838 mutex_unlock(&ftrace_start_lock); 2407 mutex_unlock(&ftrace_lock);
1839 2408
1840 return cnt; 2409 return cnt;
1841} 2410}
1842 2411
1843static struct file_operations ftrace_pid_fops = { 2412static const struct file_operations ftrace_pid_fops = {
1844 .read = ftrace_pid_read, 2413 .read = ftrace_pid_read,
1845 .write = ftrace_pid_write, 2414 .write = ftrace_pid_write,
1846}; 2415};
@@ -1863,7 +2432,6 @@ static __init int ftrace_init_debugfs(void)
1863 "'set_ftrace_pid' entry\n"); 2432 "'set_ftrace_pid' entry\n");
1864 return 0; 2433 return 0;
1865} 2434}
1866
1867fs_initcall(ftrace_init_debugfs); 2435fs_initcall(ftrace_init_debugfs);
1868 2436
1869/** 2437/**
@@ -1898,17 +2466,17 @@ int register_ftrace_function(struct ftrace_ops *ops)
1898 if (unlikely(ftrace_disabled)) 2466 if (unlikely(ftrace_disabled))
1899 return -1; 2467 return -1;
1900 2468
1901 mutex_lock(&ftrace_sysctl_lock); 2469 mutex_lock(&ftrace_lock);
1902 2470
1903 ret = __register_ftrace_function(ops); 2471 ret = __register_ftrace_function(ops);
1904 ftrace_startup(0); 2472 ftrace_startup(0);
1905 2473
1906 mutex_unlock(&ftrace_sysctl_lock); 2474 mutex_unlock(&ftrace_lock);
1907 return ret; 2475 return ret;
1908} 2476}
1909 2477
1910/** 2478/**
1911 * unregister_ftrace_function - unresgister a function for profiling. 2479 * unregister_ftrace_function - unregister a function for profiling.
1912 * @ops - ops structure that holds the function to unregister 2480 * @ops - ops structure that holds the function to unregister
1913 * 2481 *
1914 * Unregister a function that was added to be called by ftrace profiling. 2482 * Unregister a function that was added to be called by ftrace profiling.
@@ -1917,10 +2485,10 @@ int unregister_ftrace_function(struct ftrace_ops *ops)
1917{ 2485{
1918 int ret; 2486 int ret;
1919 2487
1920 mutex_lock(&ftrace_sysctl_lock); 2488 mutex_lock(&ftrace_lock);
1921 ret = __unregister_ftrace_function(ops); 2489 ret = __unregister_ftrace_function(ops);
1922 ftrace_shutdown(0); 2490 ftrace_shutdown(0);
1923 mutex_unlock(&ftrace_sysctl_lock); 2491 mutex_unlock(&ftrace_lock);
1924 2492
1925 return ret; 2493 return ret;
1926} 2494}
@@ -1935,7 +2503,7 @@ ftrace_enable_sysctl(struct ctl_table *table, int write,
1935 if (unlikely(ftrace_disabled)) 2503 if (unlikely(ftrace_disabled))
1936 return -ENODEV; 2504 return -ENODEV;
1937 2505
1938 mutex_lock(&ftrace_sysctl_lock); 2506 mutex_lock(&ftrace_lock);
1939 2507
1940 ret = proc_dointvec(table, write, file, buffer, lenp, ppos); 2508 ret = proc_dointvec(table, write, file, buffer, lenp, ppos);
1941 2509
@@ -1964,7 +2532,7 @@ ftrace_enable_sysctl(struct ctl_table *table, int write,
1964 } 2532 }
1965 2533
1966 out: 2534 out:
1967 mutex_unlock(&ftrace_sysctl_lock); 2535 mutex_unlock(&ftrace_lock);
1968 return ret; 2536 return ret;
1969} 2537}
1970 2538
@@ -2029,6 +2597,38 @@ free:
2029 return ret; 2597 return ret;
2030} 2598}
2031 2599
2600static void
2601ftrace_graph_probe_sched_switch(struct rq *__rq, struct task_struct *prev,
2602 struct task_struct *next)
2603{
2604 unsigned long long timestamp;
2605 int index;
2606
2607 /*
2608 * Does the user want to count the time a function was asleep.
2609 * If so, do not update the time stamps.
2610 */
2611 if (trace_flags & TRACE_ITER_SLEEP_TIME)
2612 return;
2613
2614 timestamp = trace_clock_local();
2615
2616 prev->ftrace_timestamp = timestamp;
2617
2618 /* only process tasks that we timestamped */
2619 if (!next->ftrace_timestamp)
2620 return;
2621
2622 /*
2623 * Update all the counters in next to make up for the
2624 * time next was sleeping.
2625 */
2626 timestamp -= next->ftrace_timestamp;
2627
2628 for (index = next->curr_ret_stack; index >= 0; index--)
2629 next->ret_stack[index].calltime += timestamp;
2630}
2631
2032/* Allocate a return stack for each task */ 2632/* Allocate a return stack for each task */
2033static int start_graph_tracing(void) 2633static int start_graph_tracing(void)
2034{ 2634{
@@ -2050,6 +2650,13 @@ static int start_graph_tracing(void)
2050 ret = alloc_retstack_tasklist(ret_stack_list); 2650 ret = alloc_retstack_tasklist(ret_stack_list);
2051 } while (ret == -EAGAIN); 2651 } while (ret == -EAGAIN);
2052 2652
2653 if (!ret) {
2654 ret = register_trace_sched_switch(ftrace_graph_probe_sched_switch);
2655 if (ret)
2656 pr_info("ftrace_graph: Couldn't activate tracepoint"
2657 " probe to kernel_sched_switch\n");
2658 }
2659
2053 kfree(ret_stack_list); 2660 kfree(ret_stack_list);
2054 return ret; 2661 return ret;
2055} 2662}
@@ -2080,7 +2687,13 @@ int register_ftrace_graph(trace_func_graph_ret_t retfunc,
2080{ 2687{
2081 int ret = 0; 2688 int ret = 0;
2082 2689
2083 mutex_lock(&ftrace_sysctl_lock); 2690 mutex_lock(&ftrace_lock);
2691
2692 /* we currently allow only one tracer registered at a time */
2693 if (atomic_read(&ftrace_graph_active)) {
2694 ret = -EBUSY;
2695 goto out;
2696 }
2084 2697
2085 ftrace_suspend_notifier.notifier_call = ftrace_suspend_notifier_call; 2698 ftrace_suspend_notifier.notifier_call = ftrace_suspend_notifier_call;
2086 register_pm_notifier(&ftrace_suspend_notifier); 2699 register_pm_notifier(&ftrace_suspend_notifier);
@@ -2098,21 +2711,26 @@ int register_ftrace_graph(trace_func_graph_ret_t retfunc,
2098 ftrace_startup(FTRACE_START_FUNC_RET); 2711 ftrace_startup(FTRACE_START_FUNC_RET);
2099 2712
2100out: 2713out:
2101 mutex_unlock(&ftrace_sysctl_lock); 2714 mutex_unlock(&ftrace_lock);
2102 return ret; 2715 return ret;
2103} 2716}
2104 2717
2105void unregister_ftrace_graph(void) 2718void unregister_ftrace_graph(void)
2106{ 2719{
2107 mutex_lock(&ftrace_sysctl_lock); 2720 mutex_lock(&ftrace_lock);
2721
2722 if (!unlikely(atomic_read(&ftrace_graph_active)))
2723 goto out;
2108 2724
2109 atomic_dec(&ftrace_graph_active); 2725 atomic_dec(&ftrace_graph_active);
2726 unregister_trace_sched_switch(ftrace_graph_probe_sched_switch);
2110 ftrace_graph_return = (trace_func_graph_ret_t)ftrace_stub; 2727 ftrace_graph_return = (trace_func_graph_ret_t)ftrace_stub;
2111 ftrace_graph_entry = ftrace_graph_entry_stub; 2728 ftrace_graph_entry = ftrace_graph_entry_stub;
2112 ftrace_shutdown(FTRACE_STOP_FUNC_RET); 2729 ftrace_shutdown(FTRACE_STOP_FUNC_RET);
2113 unregister_pm_notifier(&ftrace_suspend_notifier); 2730 unregister_pm_notifier(&ftrace_suspend_notifier);
2114 2731
2115 mutex_unlock(&ftrace_sysctl_lock); 2732 out:
2733 mutex_unlock(&ftrace_lock);
2116} 2734}
2117 2735
2118/* Allocate a return stack for newly created task */ 2736/* Allocate a return stack for newly created task */
@@ -2127,6 +2745,7 @@ void ftrace_graph_init_task(struct task_struct *t)
2127 t->curr_ret_stack = -1; 2745 t->curr_ret_stack = -1;
2128 atomic_set(&t->tracing_graph_pause, 0); 2746 atomic_set(&t->tracing_graph_pause, 0);
2129 atomic_set(&t->trace_overrun, 0); 2747 atomic_set(&t->trace_overrun, 0);
2748 t->ftrace_timestamp = 0;
2130 } else 2749 } else
2131 t->ret_stack = NULL; 2750 t->ret_stack = NULL;
2132} 2751}
diff --git a/kernel/trace/kmemtrace.c b/kernel/trace/kmemtrace.c
new file mode 100644
index 000000000000..ae201b3eda89
--- /dev/null
+++ b/kernel/trace/kmemtrace.c
@@ -0,0 +1,339 @@
1/*
2 * Memory allocator tracing
3 *
4 * Copyright (C) 2008 Eduard - Gabriel Munteanu
5 * Copyright (C) 2008 Pekka Enberg <penberg@cs.helsinki.fi>
6 * Copyright (C) 2008 Frederic Weisbecker <fweisbec@gmail.com>
7 */
8
9#include <linux/dcache.h>
10#include <linux/debugfs.h>
11#include <linux/fs.h>
12#include <linux/seq_file.h>
13#include <trace/kmemtrace.h>
14
15#include "trace.h"
16#include "trace_output.h"
17
18/* Select an alternative, minimalistic output than the original one */
19#define TRACE_KMEM_OPT_MINIMAL 0x1
20
21static struct tracer_opt kmem_opts[] = {
22 /* Default disable the minimalistic output */
23 { TRACER_OPT(kmem_minimalistic, TRACE_KMEM_OPT_MINIMAL) },
24 { }
25};
26
27static struct tracer_flags kmem_tracer_flags = {
28 .val = 0,
29 .opts = kmem_opts
30};
31
32
33static bool kmem_tracing_enabled __read_mostly;
34static struct trace_array *kmemtrace_array;
35
36static int kmem_trace_init(struct trace_array *tr)
37{
38 int cpu;
39 kmemtrace_array = tr;
40
41 for_each_cpu_mask(cpu, cpu_possible_map)
42 tracing_reset(tr, cpu);
43
44 kmem_tracing_enabled = true;
45
46 return 0;
47}
48
49static void kmem_trace_reset(struct trace_array *tr)
50{
51 kmem_tracing_enabled = false;
52}
53
54static void kmemtrace_headers(struct seq_file *s)
55{
56 /* Don't need headers for the original kmemtrace output */
57 if (!(kmem_tracer_flags.val & TRACE_KMEM_OPT_MINIMAL))
58 return;
59
60 seq_printf(s, "#\n");
61 seq_printf(s, "# ALLOC TYPE REQ GIVEN FLAGS "
62 " POINTER NODE CALLER\n");
63 seq_printf(s, "# FREE | | | | "
64 " | | | |\n");
65 seq_printf(s, "# |\n\n");
66}
67
68/*
69 * The two following functions give the original output from kmemtrace,
70 * or something close to....perhaps they need some missing things
71 */
72static enum print_line_t
73kmemtrace_print_alloc_original(struct trace_iterator *iter,
74 struct kmemtrace_alloc_entry *entry)
75{
76 struct trace_seq *s = &iter->seq;
77 int ret;
78
79 /* Taken from the old linux/kmemtrace.h */
80 ret = trace_seq_printf(s, "type_id %d call_site %lu ptr %lu "
81 "bytes_req %lu bytes_alloc %lu gfp_flags %lu node %d\n",
82 entry->type_id, entry->call_site, (unsigned long) entry->ptr,
83 (unsigned long) entry->bytes_req, (unsigned long) entry->bytes_alloc,
84 (unsigned long) entry->gfp_flags, entry->node);
85
86 if (!ret)
87 return TRACE_TYPE_PARTIAL_LINE;
88
89 return TRACE_TYPE_HANDLED;
90}
91
92static enum print_line_t
93kmemtrace_print_free_original(struct trace_iterator *iter,
94 struct kmemtrace_free_entry *entry)
95{
96 struct trace_seq *s = &iter->seq;
97 int ret;
98
99 /* Taken from the old linux/kmemtrace.h */
100 ret = trace_seq_printf(s, "type_id %d call_site %lu ptr %lu\n",
101 entry->type_id, entry->call_site, (unsigned long) entry->ptr);
102
103 if (!ret)
104 return TRACE_TYPE_PARTIAL_LINE;
105
106 return TRACE_TYPE_HANDLED;
107}
108
109
110/* The two other following provide a more minimalistic output */
111static enum print_line_t
112kmemtrace_print_alloc_compress(struct trace_iterator *iter,
113 struct kmemtrace_alloc_entry *entry)
114{
115 struct trace_seq *s = &iter->seq;
116 int ret;
117
118 /* Alloc entry */
119 ret = trace_seq_printf(s, " + ");
120 if (!ret)
121 return TRACE_TYPE_PARTIAL_LINE;
122
123 /* Type */
124 switch (entry->type_id) {
125 case KMEMTRACE_TYPE_KMALLOC:
126 ret = trace_seq_printf(s, "K ");
127 break;
128 case KMEMTRACE_TYPE_CACHE:
129 ret = trace_seq_printf(s, "C ");
130 break;
131 case KMEMTRACE_TYPE_PAGES:
132 ret = trace_seq_printf(s, "P ");
133 break;
134 default:
135 ret = trace_seq_printf(s, "? ");
136 }
137
138 if (!ret)
139 return TRACE_TYPE_PARTIAL_LINE;
140
141 /* Requested */
142 ret = trace_seq_printf(s, "%4zu ", entry->bytes_req);
143 if (!ret)
144 return TRACE_TYPE_PARTIAL_LINE;
145
146 /* Allocated */
147 ret = trace_seq_printf(s, "%4zu ", entry->bytes_alloc);
148 if (!ret)
149 return TRACE_TYPE_PARTIAL_LINE;
150
151 /* Flags
152 * TODO: would be better to see the name of the GFP flag names
153 */
154 ret = trace_seq_printf(s, "%08x ", entry->gfp_flags);
155 if (!ret)
156 return TRACE_TYPE_PARTIAL_LINE;
157
158 /* Pointer to allocated */
159 ret = trace_seq_printf(s, "0x%tx ", (ptrdiff_t)entry->ptr);
160 if (!ret)
161 return TRACE_TYPE_PARTIAL_LINE;
162
163 /* Node */
164 ret = trace_seq_printf(s, "%4d ", entry->node);
165 if (!ret)
166 return TRACE_TYPE_PARTIAL_LINE;
167
168 /* Call site */
169 ret = seq_print_ip_sym(s, entry->call_site, 0);
170 if (!ret)
171 return TRACE_TYPE_PARTIAL_LINE;
172
173 if (!trace_seq_printf(s, "\n"))
174 return TRACE_TYPE_PARTIAL_LINE;
175
176 return TRACE_TYPE_HANDLED;
177}
178
179static enum print_line_t
180kmemtrace_print_free_compress(struct trace_iterator *iter,
181 struct kmemtrace_free_entry *entry)
182{
183 struct trace_seq *s = &iter->seq;
184 int ret;
185
186 /* Free entry */
187 ret = trace_seq_printf(s, " - ");
188 if (!ret)
189 return TRACE_TYPE_PARTIAL_LINE;
190
191 /* Type */
192 switch (entry->type_id) {
193 case KMEMTRACE_TYPE_KMALLOC:
194 ret = trace_seq_printf(s, "K ");
195 break;
196 case KMEMTRACE_TYPE_CACHE:
197 ret = trace_seq_printf(s, "C ");
198 break;
199 case KMEMTRACE_TYPE_PAGES:
200 ret = trace_seq_printf(s, "P ");
201 break;
202 default:
203 ret = trace_seq_printf(s, "? ");
204 }
205
206 if (!ret)
207 return TRACE_TYPE_PARTIAL_LINE;
208
209 /* Skip requested/allocated/flags */
210 ret = trace_seq_printf(s, " ");
211 if (!ret)
212 return TRACE_TYPE_PARTIAL_LINE;
213
214 /* Pointer to allocated */
215 ret = trace_seq_printf(s, "0x%tx ", (ptrdiff_t)entry->ptr);
216 if (!ret)
217 return TRACE_TYPE_PARTIAL_LINE;
218
219 /* Skip node */
220 ret = trace_seq_printf(s, " ");
221 if (!ret)
222 return TRACE_TYPE_PARTIAL_LINE;
223
224 /* Call site */
225 ret = seq_print_ip_sym(s, entry->call_site, 0);
226 if (!ret)
227 return TRACE_TYPE_PARTIAL_LINE;
228
229 if (!trace_seq_printf(s, "\n"))
230 return TRACE_TYPE_PARTIAL_LINE;
231
232 return TRACE_TYPE_HANDLED;
233}
234
235static enum print_line_t kmemtrace_print_line(struct trace_iterator *iter)
236{
237 struct trace_entry *entry = iter->ent;
238
239 switch (entry->type) {
240 case TRACE_KMEM_ALLOC: {
241 struct kmemtrace_alloc_entry *field;
242 trace_assign_type(field, entry);
243 if (kmem_tracer_flags.val & TRACE_KMEM_OPT_MINIMAL)
244 return kmemtrace_print_alloc_compress(iter, field);
245 else
246 return kmemtrace_print_alloc_original(iter, field);
247 }
248
249 case TRACE_KMEM_FREE: {
250 struct kmemtrace_free_entry *field;
251 trace_assign_type(field, entry);
252 if (kmem_tracer_flags.val & TRACE_KMEM_OPT_MINIMAL)
253 return kmemtrace_print_free_compress(iter, field);
254 else
255 return kmemtrace_print_free_original(iter, field);
256 }
257
258 default:
259 return TRACE_TYPE_UNHANDLED;
260 }
261}
262
263/* Trace allocations */
264void kmemtrace_mark_alloc_node(enum kmemtrace_type_id type_id,
265 unsigned long call_site,
266 const void *ptr,
267 size_t bytes_req,
268 size_t bytes_alloc,
269 gfp_t gfp_flags,
270 int node)
271{
272 struct ring_buffer_event *event;
273 struct kmemtrace_alloc_entry *entry;
274 struct trace_array *tr = kmemtrace_array;
275
276 if (!kmem_tracing_enabled)
277 return;
278
279 event = trace_buffer_lock_reserve(tr, TRACE_KMEM_ALLOC,
280 sizeof(*entry), 0, 0);
281 if (!event)
282 return;
283 entry = ring_buffer_event_data(event);
284
285 entry->call_site = call_site;
286 entry->ptr = ptr;
287 entry->bytes_req = bytes_req;
288 entry->bytes_alloc = bytes_alloc;
289 entry->gfp_flags = gfp_flags;
290 entry->node = node;
291
292 trace_buffer_unlock_commit(tr, event, 0, 0);
293}
294EXPORT_SYMBOL(kmemtrace_mark_alloc_node);
295
296void kmemtrace_mark_free(enum kmemtrace_type_id type_id,
297 unsigned long call_site,
298 const void *ptr)
299{
300 struct ring_buffer_event *event;
301 struct kmemtrace_free_entry *entry;
302 struct trace_array *tr = kmemtrace_array;
303
304 if (!kmem_tracing_enabled)
305 return;
306
307 event = trace_buffer_lock_reserve(tr, TRACE_KMEM_FREE,
308 sizeof(*entry), 0, 0);
309 if (!event)
310 return;
311 entry = ring_buffer_event_data(event);
312 entry->type_id = type_id;
313 entry->call_site = call_site;
314 entry->ptr = ptr;
315
316 trace_buffer_unlock_commit(tr, event, 0, 0);
317}
318EXPORT_SYMBOL(kmemtrace_mark_free);
319
320static struct tracer kmem_tracer __read_mostly = {
321 .name = "kmemtrace",
322 .init = kmem_trace_init,
323 .reset = kmem_trace_reset,
324 .print_line = kmemtrace_print_line,
325 .print_header = kmemtrace_headers,
326 .flags = &kmem_tracer_flags
327};
328
329void kmemtrace_init(void)
330{
331 /* earliest opportunity to start kmem tracing */
332}
333
334static int __init init_kmem_tracer(void)
335{
336 return register_tracer(&kmem_tracer);
337}
338
339device_initcall(init_kmem_tracer);
diff --git a/kernel/trace/ring_buffer.c b/kernel/trace/ring_buffer.c
index bd38c5cfd8ad..960cbf44c844 100644
--- a/kernel/trace/ring_buffer.c
+++ b/kernel/trace/ring_buffer.c
@@ -4,21 +4,92 @@
4 * Copyright (C) 2008 Steven Rostedt <srostedt@redhat.com> 4 * Copyright (C) 2008 Steven Rostedt <srostedt@redhat.com>
5 */ 5 */
6#include <linux/ring_buffer.h> 6#include <linux/ring_buffer.h>
7#include <linux/trace_clock.h>
8#include <linux/ftrace_irq.h>
7#include <linux/spinlock.h> 9#include <linux/spinlock.h>
8#include <linux/debugfs.h> 10#include <linux/debugfs.h>
9#include <linux/uaccess.h> 11#include <linux/uaccess.h>
12#include <linux/hardirq.h>
10#include <linux/module.h> 13#include <linux/module.h>
11#include <linux/percpu.h> 14#include <linux/percpu.h>
12#include <linux/mutex.h> 15#include <linux/mutex.h>
13#include <linux/sched.h> /* used for sched_clock() (for now) */
14#include <linux/init.h> 16#include <linux/init.h>
15#include <linux/hash.h> 17#include <linux/hash.h>
16#include <linux/list.h> 18#include <linux/list.h>
19#include <linux/cpu.h>
17#include <linux/fs.h> 20#include <linux/fs.h>
18 21
19#include "trace.h" 22#include "trace.h"
20 23
21/* 24/*
25 * The ring buffer is made up of a list of pages. A separate list of pages is
26 * allocated for each CPU. A writer may only write to a buffer that is
27 * associated with the CPU it is currently executing on. A reader may read
28 * from any per cpu buffer.
29 *
30 * The reader is special. For each per cpu buffer, the reader has its own
31 * reader page. When a reader has read the entire reader page, this reader
32 * page is swapped with another page in the ring buffer.
33 *
34 * Now, as long as the writer is off the reader page, the reader can do what
35 * ever it wants with that page. The writer will never write to that page
36 * again (as long as it is out of the ring buffer).
37 *
38 * Here's some silly ASCII art.
39 *
40 * +------+
41 * |reader| RING BUFFER
42 * |page |
43 * +------+ +---+ +---+ +---+
44 * | |-->| |-->| |
45 * +---+ +---+ +---+
46 * ^ |
47 * | |
48 * +---------------+
49 *
50 *
51 * +------+
52 * |reader| RING BUFFER
53 * |page |------------------v
54 * +------+ +---+ +---+ +---+
55 * | |-->| |-->| |
56 * +---+ +---+ +---+
57 * ^ |
58 * | |
59 * +---------------+
60 *
61 *
62 * +------+
63 * |reader| RING BUFFER
64 * |page |------------------v
65 * +------+ +---+ +---+ +---+
66 * ^ | |-->| |-->| |
67 * | +---+ +---+ +---+
68 * | |
69 * | |
70 * +------------------------------+
71 *
72 *
73 * +------+
74 * |buffer| RING BUFFER
75 * |page |------------------v
76 * +------+ +---+ +---+ +---+
77 * ^ | | | |-->| |
78 * | New +---+ +---+ +---+
79 * | Reader------^ |
80 * | page |
81 * +------------------------------+
82 *
83 *
84 * After we make this swap, the reader can hand this page off to the splice
85 * code and be done with it. It can even allocate a new page if it needs to
86 * and swap that into the ring buffer.
87 *
88 * We will be using cmpxchg soon to make all this lockless.
89 *
90 */
91
92/*
22 * A fast way to enable or disable all ring buffers is to 93 * A fast way to enable or disable all ring buffers is to
23 * call tracing_on or tracing_off. Turning off the ring buffers 94 * call tracing_on or tracing_off. Turning off the ring buffers
24 * prevents all ring buffers from being recorded to. 95 * prevents all ring buffers from being recorded to.
@@ -57,7 +128,9 @@ enum {
57 RB_BUFFERS_DISABLED = 1 << RB_BUFFERS_DISABLED_BIT, 128 RB_BUFFERS_DISABLED = 1 << RB_BUFFERS_DISABLED_BIT,
58}; 129};
59 130
60static long ring_buffer_flags __read_mostly = RB_BUFFERS_ON; 131static unsigned long ring_buffer_flags __read_mostly = RB_BUFFERS_ON;
132
133#define BUF_PAGE_HDR_SIZE offsetof(struct buffer_data_page, data)
61 134
62/** 135/**
63 * tracing_on - enable all tracing buffers 136 * tracing_on - enable all tracing buffers
@@ -89,59 +162,92 @@ EXPORT_SYMBOL_GPL(tracing_off);
89 * tracing_off_permanent - permanently disable ring buffers 162 * tracing_off_permanent - permanently disable ring buffers
90 * 163 *
91 * This function, once called, will disable all ring buffers 164 * This function, once called, will disable all ring buffers
92 * permanenty. 165 * permanently.
93 */ 166 */
94void tracing_off_permanent(void) 167void tracing_off_permanent(void)
95{ 168{
96 set_bit(RB_BUFFERS_DISABLED_BIT, &ring_buffer_flags); 169 set_bit(RB_BUFFERS_DISABLED_BIT, &ring_buffer_flags);
97} 170}
98 171
172/**
173 * tracing_is_on - show state of ring buffers enabled
174 */
175int tracing_is_on(void)
176{
177 return ring_buffer_flags == RB_BUFFERS_ON;
178}
179EXPORT_SYMBOL_GPL(tracing_is_on);
180
99#include "trace.h" 181#include "trace.h"
100 182
101/* Up this if you want to test the TIME_EXTENTS and normalization */ 183#define RB_EVNT_HDR_SIZE (offsetof(struct ring_buffer_event, array))
102#define DEBUG_SHIFT 0 184#define RB_ALIGNMENT 4U
185#define RB_MAX_SMALL_DATA 28
186
187enum {
188 RB_LEN_TIME_EXTEND = 8,
189 RB_LEN_TIME_STAMP = 16,
190};
103 191
104/* FIXME!!! */ 192static inline int rb_null_event(struct ring_buffer_event *event)
105u64 ring_buffer_time_stamp(int cpu)
106{ 193{
107 u64 time; 194 return event->type == RINGBUF_TYPE_PADDING && event->time_delta == 0;
195}
108 196
109 preempt_disable_notrace(); 197static inline int rb_discarded_event(struct ring_buffer_event *event)
110 /* shift to debug/test normalization and TIME_EXTENTS */ 198{
111 time = sched_clock() << DEBUG_SHIFT; 199 return event->type == RINGBUF_TYPE_PADDING && event->time_delta;
112 preempt_enable_no_resched_notrace(); 200}
113 201
114 return time; 202static void rb_event_set_padding(struct ring_buffer_event *event)
203{
204 event->type = RINGBUF_TYPE_PADDING;
205 event->time_delta = 0;
115} 206}
116EXPORT_SYMBOL_GPL(ring_buffer_time_stamp);
117 207
118void ring_buffer_normalize_time_stamp(int cpu, u64 *ts) 208/**
209 * ring_buffer_event_discard - discard an event in the ring buffer
210 * @buffer: the ring buffer
211 * @event: the event to discard
212 *
213 * Sometimes a event that is in the ring buffer needs to be ignored.
214 * This function lets the user discard an event in the ring buffer
215 * and then that event will not be read later.
216 *
217 * Note, it is up to the user to be careful with this, and protect
218 * against races. If the user discards an event that has been consumed
219 * it is possible that it could corrupt the ring buffer.
220 */
221void ring_buffer_event_discard(struct ring_buffer_event *event)
119{ 222{
120 /* Just stupid testing the normalize function and deltas */ 223 event->type = RINGBUF_TYPE_PADDING;
121 *ts >>= DEBUG_SHIFT; 224 /* time delta must be non zero */
225 if (!event->time_delta)
226 event->time_delta = 1;
122} 227}
123EXPORT_SYMBOL_GPL(ring_buffer_normalize_time_stamp);
124 228
125#define RB_EVNT_HDR_SIZE (sizeof(struct ring_buffer_event)) 229static unsigned
126#define RB_ALIGNMENT_SHIFT 2 230rb_event_data_length(struct ring_buffer_event *event)
127#define RB_ALIGNMENT (1 << RB_ALIGNMENT_SHIFT) 231{
128#define RB_MAX_SMALL_DATA 28 232 unsigned length;
129 233
130enum { 234 if (event->len)
131 RB_LEN_TIME_EXTEND = 8, 235 length = event->len * RB_ALIGNMENT;
132 RB_LEN_TIME_STAMP = 16, 236 else
133}; 237 length = event->array[0];
238 return length + RB_EVNT_HDR_SIZE;
239}
134 240
135/* inline for ring buffer fast paths */ 241/* inline for ring buffer fast paths */
136static inline unsigned 242static unsigned
137rb_event_length(struct ring_buffer_event *event) 243rb_event_length(struct ring_buffer_event *event)
138{ 244{
139 unsigned length;
140
141 switch (event->type) { 245 switch (event->type) {
142 case RINGBUF_TYPE_PADDING: 246 case RINGBUF_TYPE_PADDING:
143 /* undefined */ 247 if (rb_null_event(event))
144 return -1; 248 /* undefined */
249 return -1;
250 return rb_event_data_length(event);
145 251
146 case RINGBUF_TYPE_TIME_EXTEND: 252 case RINGBUF_TYPE_TIME_EXTEND:
147 return RB_LEN_TIME_EXTEND; 253 return RB_LEN_TIME_EXTEND;
@@ -150,11 +256,7 @@ rb_event_length(struct ring_buffer_event *event)
150 return RB_LEN_TIME_STAMP; 256 return RB_LEN_TIME_STAMP;
151 257
152 case RINGBUF_TYPE_DATA: 258 case RINGBUF_TYPE_DATA:
153 if (event->len) 259 return rb_event_data_length(event);
154 length = event->len << RB_ALIGNMENT_SHIFT;
155 else
156 length = event->array[0];
157 return length + RB_EVNT_HDR_SIZE;
158 default: 260 default:
159 BUG(); 261 BUG();
160 } 262 }
@@ -179,7 +281,7 @@ unsigned ring_buffer_event_length(struct ring_buffer_event *event)
179EXPORT_SYMBOL_GPL(ring_buffer_event_length); 281EXPORT_SYMBOL_GPL(ring_buffer_event_length);
180 282
181/* inline for ring buffer fast paths */ 283/* inline for ring buffer fast paths */
182static inline void * 284static void *
183rb_event_data(struct ring_buffer_event *event) 285rb_event_data(struct ring_buffer_event *event)
184{ 286{
185 BUG_ON(event->type != RINGBUF_TYPE_DATA); 287 BUG_ON(event->type != RINGBUF_TYPE_DATA);
@@ -209,7 +311,7 @@ EXPORT_SYMBOL_GPL(ring_buffer_event_data);
209 311
210struct buffer_data_page { 312struct buffer_data_page {
211 u64 time_stamp; /* page time stamp */ 313 u64 time_stamp; /* page time stamp */
212 local_t commit; /* write commited index */ 314 local_t commit; /* write committed index */
213 unsigned char data[]; /* data of buffer page */ 315 unsigned char data[]; /* data of buffer page */
214}; 316};
215 317
@@ -225,14 +327,25 @@ static void rb_init_page(struct buffer_data_page *bpage)
225 local_set(&bpage->commit, 0); 327 local_set(&bpage->commit, 0);
226} 328}
227 329
330/**
331 * ring_buffer_page_len - the size of data on the page.
332 * @page: The page to read
333 *
334 * Returns the amount of data on the page, including buffer page header.
335 */
336size_t ring_buffer_page_len(void *page)
337{
338 return local_read(&((struct buffer_data_page *)page)->commit)
339 + BUF_PAGE_HDR_SIZE;
340}
341
228/* 342/*
229 * Also stolen from mm/slob.c. Thanks to Mathieu Desnoyers for pointing 343 * Also stolen from mm/slob.c. Thanks to Mathieu Desnoyers for pointing
230 * this issue out. 344 * this issue out.
231 */ 345 */
232static inline void free_buffer_page(struct buffer_page *bpage) 346static void free_buffer_page(struct buffer_page *bpage)
233{ 347{
234 if (bpage->page) 348 free_page((unsigned long)bpage->page);
235 free_page((unsigned long)bpage->page);
236 kfree(bpage); 349 kfree(bpage);
237} 350}
238 351
@@ -246,7 +359,7 @@ static inline int test_time_stamp(u64 delta)
246 return 0; 359 return 0;
247} 360}
248 361
249#define BUF_PAGE_SIZE (PAGE_SIZE - offsetof(struct buffer_data_page, data)) 362#define BUF_PAGE_SIZE (PAGE_SIZE - BUF_PAGE_HDR_SIZE)
250 363
251/* 364/*
252 * head_page == tail_page && head == tail then buffer is empty. 365 * head_page == tail_page && head == tail then buffer is empty.
@@ -260,7 +373,7 @@ struct ring_buffer_per_cpu {
260 struct list_head pages; 373 struct list_head pages;
261 struct buffer_page *head_page; /* read from head */ 374 struct buffer_page *head_page; /* read from head */
262 struct buffer_page *tail_page; /* write to tail */ 375 struct buffer_page *tail_page; /* write to tail */
263 struct buffer_page *commit_page; /* commited pages */ 376 struct buffer_page *commit_page; /* committed pages */
264 struct buffer_page *reader_page; 377 struct buffer_page *reader_page;
265 unsigned long overrun; 378 unsigned long overrun;
266 unsigned long entries; 379 unsigned long entries;
@@ -273,12 +386,17 @@ struct ring_buffer {
273 unsigned pages; 386 unsigned pages;
274 unsigned flags; 387 unsigned flags;
275 int cpus; 388 int cpus;
276 cpumask_var_t cpumask;
277 atomic_t record_disabled; 389 atomic_t record_disabled;
390 cpumask_var_t cpumask;
278 391
279 struct mutex mutex; 392 struct mutex mutex;
280 393
281 struct ring_buffer_per_cpu **buffers; 394 struct ring_buffer_per_cpu **buffers;
395
396#ifdef CONFIG_HOTPLUG_CPU
397 struct notifier_block cpu_notify;
398#endif
399 u64 (*clock)(void);
282}; 400};
283 401
284struct ring_buffer_iter { 402struct ring_buffer_iter {
@@ -299,11 +417,35 @@ struct ring_buffer_iter {
299 _____ret; \ 417 _____ret; \
300 }) 418 })
301 419
420/* Up this if you want to test the TIME_EXTENTS and normalization */
421#define DEBUG_SHIFT 0
422
423u64 ring_buffer_time_stamp(struct ring_buffer *buffer, int cpu)
424{
425 u64 time;
426
427 preempt_disable_notrace();
428 /* shift to debug/test normalization and TIME_EXTENTS */
429 time = buffer->clock() << DEBUG_SHIFT;
430 preempt_enable_no_resched_notrace();
431
432 return time;
433}
434EXPORT_SYMBOL_GPL(ring_buffer_time_stamp);
435
436void ring_buffer_normalize_time_stamp(struct ring_buffer *buffer,
437 int cpu, u64 *ts)
438{
439 /* Just stupid testing the normalize function and deltas */
440 *ts >>= DEBUG_SHIFT;
441}
442EXPORT_SYMBOL_GPL(ring_buffer_normalize_time_stamp);
443
302/** 444/**
303 * check_pages - integrity check of buffer pages 445 * check_pages - integrity check of buffer pages
304 * @cpu_buffer: CPU buffer with pages to test 446 * @cpu_buffer: CPU buffer with pages to test
305 * 447 *
306 * As a safty measure we check to make sure the data pages have not 448 * As a safety measure we check to make sure the data pages have not
307 * been corrupted. 449 * been corrupted.
308 */ 450 */
309static int rb_check_pages(struct ring_buffer_per_cpu *cpu_buffer) 451static int rb_check_pages(struct ring_buffer_per_cpu *cpu_buffer)
@@ -421,7 +563,6 @@ static void rb_free_cpu_buffer(struct ring_buffer_per_cpu *cpu_buffer)
421 struct list_head *head = &cpu_buffer->pages; 563 struct list_head *head = &cpu_buffer->pages;
422 struct buffer_page *bpage, *tmp; 564 struct buffer_page *bpage, *tmp;
423 565
424 list_del_init(&cpu_buffer->reader_page->list);
425 free_buffer_page(cpu_buffer->reader_page); 566 free_buffer_page(cpu_buffer->reader_page);
426 567
427 list_for_each_entry_safe(bpage, tmp, head, list) { 568 list_for_each_entry_safe(bpage, tmp, head, list) {
@@ -437,6 +578,11 @@ static void rb_free_cpu_buffer(struct ring_buffer_per_cpu *cpu_buffer)
437 */ 578 */
438extern int ring_buffer_page_too_big(void); 579extern int ring_buffer_page_too_big(void);
439 580
581#ifdef CONFIG_HOTPLUG_CPU
582static int rb_cpu_notify(struct notifier_block *self,
583 unsigned long action, void *hcpu);
584#endif
585
440/** 586/**
441 * ring_buffer_alloc - allocate a new ring_buffer 587 * ring_buffer_alloc - allocate a new ring_buffer
442 * @size: the size in bytes per cpu that is needed. 588 * @size: the size in bytes per cpu that is needed.
@@ -469,12 +615,23 @@ struct ring_buffer *ring_buffer_alloc(unsigned long size, unsigned flags)
469 615
470 buffer->pages = DIV_ROUND_UP(size, BUF_PAGE_SIZE); 616 buffer->pages = DIV_ROUND_UP(size, BUF_PAGE_SIZE);
471 buffer->flags = flags; 617 buffer->flags = flags;
618 buffer->clock = trace_clock_local;
472 619
473 /* need at least two pages */ 620 /* need at least two pages */
474 if (buffer->pages == 1) 621 if (buffer->pages == 1)
475 buffer->pages++; 622 buffer->pages++;
476 623
624 /*
625 * In case of non-hotplug cpu, if the ring-buffer is allocated
626 * in early initcall, it will not be notified of secondary cpus.
627 * In that off case, we need to allocate for all possible cpus.
628 */
629#ifdef CONFIG_HOTPLUG_CPU
630 get_online_cpus();
631 cpumask_copy(buffer->cpumask, cpu_online_mask);
632#else
477 cpumask_copy(buffer->cpumask, cpu_possible_mask); 633 cpumask_copy(buffer->cpumask, cpu_possible_mask);
634#endif
478 buffer->cpus = nr_cpu_ids; 635 buffer->cpus = nr_cpu_ids;
479 636
480 bsize = sizeof(void *) * nr_cpu_ids; 637 bsize = sizeof(void *) * nr_cpu_ids;
@@ -490,6 +647,13 @@ struct ring_buffer *ring_buffer_alloc(unsigned long size, unsigned flags)
490 goto fail_free_buffers; 647 goto fail_free_buffers;
491 } 648 }
492 649
650#ifdef CONFIG_HOTPLUG_CPU
651 buffer->cpu_notify.notifier_call = rb_cpu_notify;
652 buffer->cpu_notify.priority = 0;
653 register_cpu_notifier(&buffer->cpu_notify);
654#endif
655
656 put_online_cpus();
493 mutex_init(&buffer->mutex); 657 mutex_init(&buffer->mutex);
494 658
495 return buffer; 659 return buffer;
@@ -503,6 +667,7 @@ struct ring_buffer *ring_buffer_alloc(unsigned long size, unsigned flags)
503 667
504 fail_free_cpumask: 668 fail_free_cpumask:
505 free_cpumask_var(buffer->cpumask); 669 free_cpumask_var(buffer->cpumask);
670 put_online_cpus();
506 671
507 fail_free_buffer: 672 fail_free_buffer:
508 kfree(buffer); 673 kfree(buffer);
@@ -519,15 +684,29 @@ ring_buffer_free(struct ring_buffer *buffer)
519{ 684{
520 int cpu; 685 int cpu;
521 686
687 get_online_cpus();
688
689#ifdef CONFIG_HOTPLUG_CPU
690 unregister_cpu_notifier(&buffer->cpu_notify);
691#endif
692
522 for_each_buffer_cpu(buffer, cpu) 693 for_each_buffer_cpu(buffer, cpu)
523 rb_free_cpu_buffer(buffer->buffers[cpu]); 694 rb_free_cpu_buffer(buffer->buffers[cpu]);
524 695
696 put_online_cpus();
697
525 free_cpumask_var(buffer->cpumask); 698 free_cpumask_var(buffer->cpumask);
526 699
527 kfree(buffer); 700 kfree(buffer);
528} 701}
529EXPORT_SYMBOL_GPL(ring_buffer_free); 702EXPORT_SYMBOL_GPL(ring_buffer_free);
530 703
704void ring_buffer_set_clock(struct ring_buffer *buffer,
705 u64 (*clock)(void))
706{
707 buffer->clock = clock;
708}
709
531static void rb_reset_cpu(struct ring_buffer_per_cpu *cpu_buffer); 710static void rb_reset_cpu(struct ring_buffer_per_cpu *cpu_buffer);
532 711
533static void 712static void
@@ -627,16 +806,15 @@ int ring_buffer_resize(struct ring_buffer *buffer, unsigned long size)
627 return size; 806 return size;
628 807
629 mutex_lock(&buffer->mutex); 808 mutex_lock(&buffer->mutex);
809 get_online_cpus();
630 810
631 nr_pages = DIV_ROUND_UP(size, BUF_PAGE_SIZE); 811 nr_pages = DIV_ROUND_UP(size, BUF_PAGE_SIZE);
632 812
633 if (size < buffer_size) { 813 if (size < buffer_size) {
634 814
635 /* easy case, just free pages */ 815 /* easy case, just free pages */
636 if (RB_WARN_ON(buffer, nr_pages >= buffer->pages)) { 816 if (RB_WARN_ON(buffer, nr_pages >= buffer->pages))
637 mutex_unlock(&buffer->mutex); 817 goto out_fail;
638 return -1;
639 }
640 818
641 rm_pages = buffer->pages - nr_pages; 819 rm_pages = buffer->pages - nr_pages;
642 820
@@ -655,10 +833,8 @@ int ring_buffer_resize(struct ring_buffer *buffer, unsigned long size)
655 * add these pages to the cpu_buffers. Otherwise we just free 833 * add these pages to the cpu_buffers. Otherwise we just free
656 * them all and return -ENOMEM; 834 * them all and return -ENOMEM;
657 */ 835 */
658 if (RB_WARN_ON(buffer, nr_pages <= buffer->pages)) { 836 if (RB_WARN_ON(buffer, nr_pages <= buffer->pages))
659 mutex_unlock(&buffer->mutex); 837 goto out_fail;
660 return -1;
661 }
662 838
663 new_pages = nr_pages - buffer->pages; 839 new_pages = nr_pages - buffer->pages;
664 840
@@ -683,13 +859,12 @@ int ring_buffer_resize(struct ring_buffer *buffer, unsigned long size)
683 rb_insert_pages(cpu_buffer, &pages, new_pages); 859 rb_insert_pages(cpu_buffer, &pages, new_pages);
684 } 860 }
685 861
686 if (RB_WARN_ON(buffer, !list_empty(&pages))) { 862 if (RB_WARN_ON(buffer, !list_empty(&pages)))
687 mutex_unlock(&buffer->mutex); 863 goto out_fail;
688 return -1;
689 }
690 864
691 out: 865 out:
692 buffer->pages = nr_pages; 866 buffer->pages = nr_pages;
867 put_online_cpus();
693 mutex_unlock(&buffer->mutex); 868 mutex_unlock(&buffer->mutex);
694 869
695 return size; 870 return size;
@@ -699,15 +874,20 @@ int ring_buffer_resize(struct ring_buffer *buffer, unsigned long size)
699 list_del_init(&bpage->list); 874 list_del_init(&bpage->list);
700 free_buffer_page(bpage); 875 free_buffer_page(bpage);
701 } 876 }
877 put_online_cpus();
702 mutex_unlock(&buffer->mutex); 878 mutex_unlock(&buffer->mutex);
703 return -ENOMEM; 879 return -ENOMEM;
704}
705EXPORT_SYMBOL_GPL(ring_buffer_resize);
706 880
707static inline int rb_null_event(struct ring_buffer_event *event) 881 /*
708{ 882 * Something went totally wrong, and we are too paranoid
709 return event->type == RINGBUF_TYPE_PADDING; 883 * to even clean up the mess.
884 */
885 out_fail:
886 put_online_cpus();
887 mutex_unlock(&buffer->mutex);
888 return -1;
710} 889}
890EXPORT_SYMBOL_GPL(ring_buffer_resize);
711 891
712static inline void * 892static inline void *
713__rb_data_page_index(struct buffer_data_page *bpage, unsigned index) 893__rb_data_page_index(struct buffer_data_page *bpage, unsigned index)
@@ -811,7 +991,7 @@ rb_event_index(struct ring_buffer_event *event)
811 return (addr & ~PAGE_MASK) - (PAGE_SIZE - BUF_PAGE_SIZE); 991 return (addr & ~PAGE_MASK) - (PAGE_SIZE - BUF_PAGE_SIZE);
812} 992}
813 993
814static inline int 994static int
815rb_is_commit(struct ring_buffer_per_cpu *cpu_buffer, 995rb_is_commit(struct ring_buffer_per_cpu *cpu_buffer,
816 struct ring_buffer_event *event) 996 struct ring_buffer_event *event)
817{ 997{
@@ -825,7 +1005,7 @@ rb_is_commit(struct ring_buffer_per_cpu *cpu_buffer,
825 rb_commit_index(cpu_buffer) == index; 1005 rb_commit_index(cpu_buffer) == index;
826} 1006}
827 1007
828static inline void 1008static void
829rb_set_commit_event(struct ring_buffer_per_cpu *cpu_buffer, 1009rb_set_commit_event(struct ring_buffer_per_cpu *cpu_buffer,
830 struct ring_buffer_event *event) 1010 struct ring_buffer_event *event)
831{ 1011{
@@ -850,7 +1030,7 @@ rb_set_commit_event(struct ring_buffer_per_cpu *cpu_buffer,
850 local_set(&cpu_buffer->commit_page->page->commit, index); 1030 local_set(&cpu_buffer->commit_page->page->commit, index);
851} 1031}
852 1032
853static inline void 1033static void
854rb_set_commit_to_write(struct ring_buffer_per_cpu *cpu_buffer) 1034rb_set_commit_to_write(struct ring_buffer_per_cpu *cpu_buffer)
855{ 1035{
856 /* 1036 /*
@@ -896,7 +1076,7 @@ static void rb_reset_reader_page(struct ring_buffer_per_cpu *cpu_buffer)
896 cpu_buffer->reader_page->read = 0; 1076 cpu_buffer->reader_page->read = 0;
897} 1077}
898 1078
899static inline void rb_inc_iter(struct ring_buffer_iter *iter) 1079static void rb_inc_iter(struct ring_buffer_iter *iter)
900{ 1080{
901 struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; 1081 struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer;
902 1082
@@ -926,7 +1106,7 @@ static inline void rb_inc_iter(struct ring_buffer_iter *iter)
926 * and with this, we can determine what to place into the 1106 * and with this, we can determine what to place into the
927 * data field. 1107 * data field.
928 */ 1108 */
929static inline void 1109static void
930rb_update_event(struct ring_buffer_event *event, 1110rb_update_event(struct ring_buffer_event *event,
931 unsigned type, unsigned length) 1111 unsigned type, unsigned length)
932{ 1112{
@@ -938,15 +1118,11 @@ rb_update_event(struct ring_buffer_event *event,
938 break; 1118 break;
939 1119
940 case RINGBUF_TYPE_TIME_EXTEND: 1120 case RINGBUF_TYPE_TIME_EXTEND:
941 event->len = 1121 event->len = DIV_ROUND_UP(RB_LEN_TIME_EXTEND, RB_ALIGNMENT);
942 (RB_LEN_TIME_EXTEND + (RB_ALIGNMENT-1))
943 >> RB_ALIGNMENT_SHIFT;
944 break; 1122 break;
945 1123
946 case RINGBUF_TYPE_TIME_STAMP: 1124 case RINGBUF_TYPE_TIME_STAMP:
947 event->len = 1125 event->len = DIV_ROUND_UP(RB_LEN_TIME_STAMP, RB_ALIGNMENT);
948 (RB_LEN_TIME_STAMP + (RB_ALIGNMENT-1))
949 >> RB_ALIGNMENT_SHIFT;
950 break; 1126 break;
951 1127
952 case RINGBUF_TYPE_DATA: 1128 case RINGBUF_TYPE_DATA:
@@ -955,16 +1131,14 @@ rb_update_event(struct ring_buffer_event *event,
955 event->len = 0; 1131 event->len = 0;
956 event->array[0] = length; 1132 event->array[0] = length;
957 } else 1133 } else
958 event->len = 1134 event->len = DIV_ROUND_UP(length, RB_ALIGNMENT);
959 (length + (RB_ALIGNMENT-1))
960 >> RB_ALIGNMENT_SHIFT;
961 break; 1135 break;
962 default: 1136 default:
963 BUG(); 1137 BUG();
964 } 1138 }
965} 1139}
966 1140
967static inline unsigned rb_calculate_event_length(unsigned length) 1141static unsigned rb_calculate_event_length(unsigned length)
968{ 1142{
969 struct ring_buffer_event event; /* Used only for sizeof array */ 1143 struct ring_buffer_event event; /* Used only for sizeof array */
970 1144
@@ -990,6 +1164,7 @@ __rb_reserve_next(struct ring_buffer_per_cpu *cpu_buffer,
990 struct ring_buffer *buffer = cpu_buffer->buffer; 1164 struct ring_buffer *buffer = cpu_buffer->buffer;
991 struct ring_buffer_event *event; 1165 struct ring_buffer_event *event;
992 unsigned long flags; 1166 unsigned long flags;
1167 bool lock_taken = false;
993 1168
994 commit_page = cpu_buffer->commit_page; 1169 commit_page = cpu_buffer->commit_page;
995 /* we just need to protect against interrupts */ 1170 /* we just need to protect against interrupts */
@@ -1003,7 +1178,30 @@ __rb_reserve_next(struct ring_buffer_per_cpu *cpu_buffer,
1003 struct buffer_page *next_page = tail_page; 1178 struct buffer_page *next_page = tail_page;
1004 1179
1005 local_irq_save(flags); 1180 local_irq_save(flags);
1006 __raw_spin_lock(&cpu_buffer->lock); 1181 /*
1182 * Since the write to the buffer is still not
1183 * fully lockless, we must be careful with NMIs.
1184 * The locks in the writers are taken when a write
1185 * crosses to a new page. The locks protect against
1186 * races with the readers (this will soon be fixed
1187 * with a lockless solution).
1188 *
1189 * Because we can not protect against NMIs, and we
1190 * want to keep traces reentrant, we need to manage
1191 * what happens when we are in an NMI.
1192 *
1193 * NMIs can happen after we take the lock.
1194 * If we are in an NMI, only take the lock
1195 * if it is not already taken. Otherwise
1196 * simply fail.
1197 */
1198 if (unlikely(in_nmi())) {
1199 if (!__raw_spin_trylock(&cpu_buffer->lock))
1200 goto out_reset;
1201 } else
1202 __raw_spin_lock(&cpu_buffer->lock);
1203
1204 lock_taken = true;
1007 1205
1008 rb_inc_page(cpu_buffer, &next_page); 1206 rb_inc_page(cpu_buffer, &next_page);
1009 1207
@@ -1012,7 +1210,7 @@ __rb_reserve_next(struct ring_buffer_per_cpu *cpu_buffer,
1012 1210
1013 /* we grabbed the lock before incrementing */ 1211 /* we grabbed the lock before incrementing */
1014 if (RB_WARN_ON(cpu_buffer, next_page == reader_page)) 1212 if (RB_WARN_ON(cpu_buffer, next_page == reader_page))
1015 goto out_unlock; 1213 goto out_reset;
1016 1214
1017 /* 1215 /*
1018 * If for some reason, we had an interrupt storm that made 1216 * If for some reason, we had an interrupt storm that made
@@ -1021,12 +1219,12 @@ __rb_reserve_next(struct ring_buffer_per_cpu *cpu_buffer,
1021 */ 1219 */
1022 if (unlikely(next_page == commit_page)) { 1220 if (unlikely(next_page == commit_page)) {
1023 WARN_ON_ONCE(1); 1221 WARN_ON_ONCE(1);
1024 goto out_unlock; 1222 goto out_reset;
1025 } 1223 }
1026 1224
1027 if (next_page == head_page) { 1225 if (next_page == head_page) {
1028 if (!(buffer->flags & RB_FL_OVERWRITE)) 1226 if (!(buffer->flags & RB_FL_OVERWRITE))
1029 goto out_unlock; 1227 goto out_reset;
1030 1228
1031 /* tail_page has not moved yet? */ 1229 /* tail_page has not moved yet? */
1032 if (tail_page == cpu_buffer->tail_page) { 1230 if (tail_page == cpu_buffer->tail_page) {
@@ -1050,7 +1248,7 @@ __rb_reserve_next(struct ring_buffer_per_cpu *cpu_buffer,
1050 cpu_buffer->tail_page = next_page; 1248 cpu_buffer->tail_page = next_page;
1051 1249
1052 /* reread the time stamp */ 1250 /* reread the time stamp */
1053 *ts = ring_buffer_time_stamp(cpu_buffer->cpu); 1251 *ts = ring_buffer_time_stamp(buffer, cpu_buffer->cpu);
1054 cpu_buffer->tail_page->page->time_stamp = *ts; 1252 cpu_buffer->tail_page->page->time_stamp = *ts;
1055 } 1253 }
1056 1254
@@ -1060,7 +1258,7 @@ __rb_reserve_next(struct ring_buffer_per_cpu *cpu_buffer,
1060 if (tail < BUF_PAGE_SIZE) { 1258 if (tail < BUF_PAGE_SIZE) {
1061 /* Mark the rest of the page with padding */ 1259 /* Mark the rest of the page with padding */
1062 event = __rb_page_index(tail_page, tail); 1260 event = __rb_page_index(tail_page, tail);
1063 event->type = RINGBUF_TYPE_PADDING; 1261 rb_event_set_padding(event);
1064 } 1262 }
1065 1263
1066 if (tail <= BUF_PAGE_SIZE) 1264 if (tail <= BUF_PAGE_SIZE)
@@ -1100,12 +1298,13 @@ __rb_reserve_next(struct ring_buffer_per_cpu *cpu_buffer,
1100 1298
1101 return event; 1299 return event;
1102 1300
1103 out_unlock: 1301 out_reset:
1104 /* reset write */ 1302 /* reset write */
1105 if (tail <= BUF_PAGE_SIZE) 1303 if (tail <= BUF_PAGE_SIZE)
1106 local_set(&tail_page->write, tail); 1304 local_set(&tail_page->write, tail);
1107 1305
1108 __raw_spin_unlock(&cpu_buffer->lock); 1306 if (likely(lock_taken))
1307 __raw_spin_unlock(&cpu_buffer->lock);
1109 local_irq_restore(flags); 1308 local_irq_restore(flags);
1110 return NULL; 1309 return NULL;
1111} 1310}
@@ -1192,7 +1391,7 @@ rb_reserve_next_event(struct ring_buffer_per_cpu *cpu_buffer,
1192 if (RB_WARN_ON(cpu_buffer, ++nr_loops > 1000)) 1391 if (RB_WARN_ON(cpu_buffer, ++nr_loops > 1000))
1193 return NULL; 1392 return NULL;
1194 1393
1195 ts = ring_buffer_time_stamp(cpu_buffer->cpu); 1394 ts = ring_buffer_time_stamp(cpu_buffer->buffer, cpu_buffer->cpu);
1196 1395
1197 /* 1396 /*
1198 * Only the first commit can update the timestamp. 1397 * Only the first commit can update the timestamp.
@@ -1265,7 +1464,6 @@ static DEFINE_PER_CPU(int, rb_need_resched);
1265 * ring_buffer_lock_reserve - reserve a part of the buffer 1464 * ring_buffer_lock_reserve - reserve a part of the buffer
1266 * @buffer: the ring buffer to reserve from 1465 * @buffer: the ring buffer to reserve from
1267 * @length: the length of the data to reserve (excluding event header) 1466 * @length: the length of the data to reserve (excluding event header)
1268 * @flags: a pointer to save the interrupt flags
1269 * 1467 *
1270 * Returns a reseverd event on the ring buffer to copy directly to. 1468 * Returns a reseverd event on the ring buffer to copy directly to.
1271 * The user of this interface will need to get the body to write into 1469 * The user of this interface will need to get the body to write into
@@ -1278,9 +1476,7 @@ static DEFINE_PER_CPU(int, rb_need_resched);
1278 * If NULL is returned, then nothing has been allocated or locked. 1476 * If NULL is returned, then nothing has been allocated or locked.
1279 */ 1477 */
1280struct ring_buffer_event * 1478struct ring_buffer_event *
1281ring_buffer_lock_reserve(struct ring_buffer *buffer, 1479ring_buffer_lock_reserve(struct ring_buffer *buffer, unsigned long length)
1282 unsigned long length,
1283 unsigned long *flags)
1284{ 1480{
1285 struct ring_buffer_per_cpu *cpu_buffer; 1481 struct ring_buffer_per_cpu *cpu_buffer;
1286 struct ring_buffer_event *event; 1482 struct ring_buffer_event *event;
@@ -1347,15 +1543,13 @@ static void rb_commit(struct ring_buffer_per_cpu *cpu_buffer,
1347 * ring_buffer_unlock_commit - commit a reserved 1543 * ring_buffer_unlock_commit - commit a reserved
1348 * @buffer: The buffer to commit to 1544 * @buffer: The buffer to commit to
1349 * @event: The event pointer to commit. 1545 * @event: The event pointer to commit.
1350 * @flags: the interrupt flags received from ring_buffer_lock_reserve.
1351 * 1546 *
1352 * This commits the data to the ring buffer, and releases any locks held. 1547 * This commits the data to the ring buffer, and releases any locks held.
1353 * 1548 *
1354 * Must be paired with ring_buffer_lock_reserve. 1549 * Must be paired with ring_buffer_lock_reserve.
1355 */ 1550 */
1356int ring_buffer_unlock_commit(struct ring_buffer *buffer, 1551int ring_buffer_unlock_commit(struct ring_buffer *buffer,
1357 struct ring_buffer_event *event, 1552 struct ring_buffer_event *event)
1358 unsigned long flags)
1359{ 1553{
1360 struct ring_buffer_per_cpu *cpu_buffer; 1554 struct ring_buffer_per_cpu *cpu_buffer;
1361 int cpu = raw_smp_processor_id(); 1555 int cpu = raw_smp_processor_id();
@@ -1438,7 +1632,7 @@ int ring_buffer_write(struct ring_buffer *buffer,
1438} 1632}
1439EXPORT_SYMBOL_GPL(ring_buffer_write); 1633EXPORT_SYMBOL_GPL(ring_buffer_write);
1440 1634
1441static inline int rb_per_cpu_empty(struct ring_buffer_per_cpu *cpu_buffer) 1635static int rb_per_cpu_empty(struct ring_buffer_per_cpu *cpu_buffer)
1442{ 1636{
1443 struct buffer_page *reader = cpu_buffer->reader_page; 1637 struct buffer_page *reader = cpu_buffer->reader_page;
1444 struct buffer_page *head = cpu_buffer->head_page; 1638 struct buffer_page *head = cpu_buffer->head_page;
@@ -1528,12 +1722,15 @@ EXPORT_SYMBOL_GPL(ring_buffer_record_enable_cpu);
1528unsigned long ring_buffer_entries_cpu(struct ring_buffer *buffer, int cpu) 1722unsigned long ring_buffer_entries_cpu(struct ring_buffer *buffer, int cpu)
1529{ 1723{
1530 struct ring_buffer_per_cpu *cpu_buffer; 1724 struct ring_buffer_per_cpu *cpu_buffer;
1725 unsigned long ret;
1531 1726
1532 if (!cpumask_test_cpu(cpu, buffer->cpumask)) 1727 if (!cpumask_test_cpu(cpu, buffer->cpumask))
1533 return 0; 1728 return 0;
1534 1729
1535 cpu_buffer = buffer->buffers[cpu]; 1730 cpu_buffer = buffer->buffers[cpu];
1536 return cpu_buffer->entries; 1731 ret = cpu_buffer->entries;
1732
1733 return ret;
1537} 1734}
1538EXPORT_SYMBOL_GPL(ring_buffer_entries_cpu); 1735EXPORT_SYMBOL_GPL(ring_buffer_entries_cpu);
1539 1736
@@ -1545,12 +1742,15 @@ EXPORT_SYMBOL_GPL(ring_buffer_entries_cpu);
1545unsigned long ring_buffer_overrun_cpu(struct ring_buffer *buffer, int cpu) 1742unsigned long ring_buffer_overrun_cpu(struct ring_buffer *buffer, int cpu)
1546{ 1743{
1547 struct ring_buffer_per_cpu *cpu_buffer; 1744 struct ring_buffer_per_cpu *cpu_buffer;
1745 unsigned long ret;
1548 1746
1549 if (!cpumask_test_cpu(cpu, buffer->cpumask)) 1747 if (!cpumask_test_cpu(cpu, buffer->cpumask))
1550 return 0; 1748 return 0;
1551 1749
1552 cpu_buffer = buffer->buffers[cpu]; 1750 cpu_buffer = buffer->buffers[cpu];
1553 return cpu_buffer->overrun; 1751 ret = cpu_buffer->overrun;
1752
1753 return ret;
1554} 1754}
1555EXPORT_SYMBOL_GPL(ring_buffer_overrun_cpu); 1755EXPORT_SYMBOL_GPL(ring_buffer_overrun_cpu);
1556 1756
@@ -1627,9 +1827,14 @@ static void rb_iter_reset(struct ring_buffer_iter *iter)
1627 */ 1827 */
1628void ring_buffer_iter_reset(struct ring_buffer_iter *iter) 1828void ring_buffer_iter_reset(struct ring_buffer_iter *iter)
1629{ 1829{
1630 struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; 1830 struct ring_buffer_per_cpu *cpu_buffer;
1631 unsigned long flags; 1831 unsigned long flags;
1632 1832
1833 if (!iter)
1834 return;
1835
1836 cpu_buffer = iter->cpu_buffer;
1837
1633 spin_lock_irqsave(&cpu_buffer->reader_lock, flags); 1838 spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
1634 rb_iter_reset(iter); 1839 rb_iter_reset(iter);
1635 spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); 1840 spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
@@ -1803,7 +2008,7 @@ static void rb_advance_reader(struct ring_buffer_per_cpu *cpu_buffer)
1803 2008
1804 event = rb_reader_event(cpu_buffer); 2009 event = rb_reader_event(cpu_buffer);
1805 2010
1806 if (event->type == RINGBUF_TYPE_DATA) 2011 if (event->type == RINGBUF_TYPE_DATA || rb_discarded_event(event))
1807 cpu_buffer->entries--; 2012 cpu_buffer->entries--;
1808 2013
1809 rb_update_read_stamp(cpu_buffer, event); 2014 rb_update_read_stamp(cpu_buffer, event);
@@ -1864,9 +2069,6 @@ rb_buffer_peek(struct ring_buffer *buffer, int cpu, u64 *ts)
1864 struct buffer_page *reader; 2069 struct buffer_page *reader;
1865 int nr_loops = 0; 2070 int nr_loops = 0;
1866 2071
1867 if (!cpumask_test_cpu(cpu, buffer->cpumask))
1868 return NULL;
1869
1870 cpu_buffer = buffer->buffers[cpu]; 2072 cpu_buffer = buffer->buffers[cpu];
1871 2073
1872 again: 2074 again:
@@ -1889,9 +2091,18 @@ rb_buffer_peek(struct ring_buffer *buffer, int cpu, u64 *ts)
1889 2091
1890 switch (event->type) { 2092 switch (event->type) {
1891 case RINGBUF_TYPE_PADDING: 2093 case RINGBUF_TYPE_PADDING:
1892 RB_WARN_ON(cpu_buffer, 1); 2094 if (rb_null_event(event))
2095 RB_WARN_ON(cpu_buffer, 1);
2096 /*
2097 * Because the writer could be discarding every
2098 * event it creates (which would probably be bad)
2099 * if we were to go back to "again" then we may never
2100 * catch up, and will trigger the warn on, or lock
2101 * the box. Return the padding, and we will release
2102 * the current locks, and try again.
2103 */
1893 rb_advance_reader(cpu_buffer); 2104 rb_advance_reader(cpu_buffer);
1894 return NULL; 2105 return event;
1895 2106
1896 case RINGBUF_TYPE_TIME_EXTEND: 2107 case RINGBUF_TYPE_TIME_EXTEND:
1897 /* Internal data, OK to advance */ 2108 /* Internal data, OK to advance */
@@ -1906,7 +2117,8 @@ rb_buffer_peek(struct ring_buffer *buffer, int cpu, u64 *ts)
1906 case RINGBUF_TYPE_DATA: 2117 case RINGBUF_TYPE_DATA:
1907 if (ts) { 2118 if (ts) {
1908 *ts = cpu_buffer->read_stamp + event->time_delta; 2119 *ts = cpu_buffer->read_stamp + event->time_delta;
1909 ring_buffer_normalize_time_stamp(cpu_buffer->cpu, ts); 2120 ring_buffer_normalize_time_stamp(buffer,
2121 cpu_buffer->cpu, ts);
1910 } 2122 }
1911 return event; 2123 return event;
1912 2124
@@ -1951,8 +2163,12 @@ rb_iter_peek(struct ring_buffer_iter *iter, u64 *ts)
1951 2163
1952 switch (event->type) { 2164 switch (event->type) {
1953 case RINGBUF_TYPE_PADDING: 2165 case RINGBUF_TYPE_PADDING:
1954 rb_inc_iter(iter); 2166 if (rb_null_event(event)) {
1955 goto again; 2167 rb_inc_iter(iter);
2168 goto again;
2169 }
2170 rb_advance_iter(iter);
2171 return event;
1956 2172
1957 case RINGBUF_TYPE_TIME_EXTEND: 2173 case RINGBUF_TYPE_TIME_EXTEND:
1958 /* Internal data, OK to advance */ 2174 /* Internal data, OK to advance */
@@ -1967,7 +2183,8 @@ rb_iter_peek(struct ring_buffer_iter *iter, u64 *ts)
1967 case RINGBUF_TYPE_DATA: 2183 case RINGBUF_TYPE_DATA:
1968 if (ts) { 2184 if (ts) {
1969 *ts = iter->read_stamp + event->time_delta; 2185 *ts = iter->read_stamp + event->time_delta;
1970 ring_buffer_normalize_time_stamp(cpu_buffer->cpu, ts); 2186 ring_buffer_normalize_time_stamp(buffer,
2187 cpu_buffer->cpu, ts);
1971 } 2188 }
1972 return event; 2189 return event;
1973 2190
@@ -1995,10 +2212,19 @@ ring_buffer_peek(struct ring_buffer *buffer, int cpu, u64 *ts)
1995 struct ring_buffer_event *event; 2212 struct ring_buffer_event *event;
1996 unsigned long flags; 2213 unsigned long flags;
1997 2214
2215 if (!cpumask_test_cpu(cpu, buffer->cpumask))
2216 return NULL;
2217
2218 again:
1998 spin_lock_irqsave(&cpu_buffer->reader_lock, flags); 2219 spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
1999 event = rb_buffer_peek(buffer, cpu, ts); 2220 event = rb_buffer_peek(buffer, cpu, ts);
2000 spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); 2221 spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
2001 2222
2223 if (event && event->type == RINGBUF_TYPE_PADDING) {
2224 cpu_relax();
2225 goto again;
2226 }
2227
2002 return event; 2228 return event;
2003} 2229}
2004 2230
@@ -2017,10 +2243,16 @@ ring_buffer_iter_peek(struct ring_buffer_iter *iter, u64 *ts)
2017 struct ring_buffer_event *event; 2243 struct ring_buffer_event *event;
2018 unsigned long flags; 2244 unsigned long flags;
2019 2245
2246 again:
2020 spin_lock_irqsave(&cpu_buffer->reader_lock, flags); 2247 spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
2021 event = rb_iter_peek(iter, ts); 2248 event = rb_iter_peek(iter, ts);
2022 spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); 2249 spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
2023 2250
2251 if (event && event->type == RINGBUF_TYPE_PADDING) {
2252 cpu_relax();
2253 goto again;
2254 }
2255
2024 return event; 2256 return event;
2025} 2257}
2026 2258
@@ -2035,24 +2267,37 @@ ring_buffer_iter_peek(struct ring_buffer_iter *iter, u64 *ts)
2035struct ring_buffer_event * 2267struct ring_buffer_event *
2036ring_buffer_consume(struct ring_buffer *buffer, int cpu, u64 *ts) 2268ring_buffer_consume(struct ring_buffer *buffer, int cpu, u64 *ts)
2037{ 2269{
2038 struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu]; 2270 struct ring_buffer_per_cpu *cpu_buffer;
2039 struct ring_buffer_event *event; 2271 struct ring_buffer_event *event = NULL;
2040 unsigned long flags; 2272 unsigned long flags;
2041 2273
2274 again:
2275 /* might be called in atomic */
2276 preempt_disable();
2277
2042 if (!cpumask_test_cpu(cpu, buffer->cpumask)) 2278 if (!cpumask_test_cpu(cpu, buffer->cpumask))
2043 return NULL; 2279 goto out;
2044 2280
2281 cpu_buffer = buffer->buffers[cpu];
2045 spin_lock_irqsave(&cpu_buffer->reader_lock, flags); 2282 spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
2046 2283
2047 event = rb_buffer_peek(buffer, cpu, ts); 2284 event = rb_buffer_peek(buffer, cpu, ts);
2048 if (!event) 2285 if (!event)
2049 goto out; 2286 goto out_unlock;
2050 2287
2051 rb_advance_reader(cpu_buffer); 2288 rb_advance_reader(cpu_buffer);
2052 2289
2053 out: 2290 out_unlock:
2054 spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); 2291 spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
2055 2292
2293 out:
2294 preempt_enable();
2295
2296 if (event && event->type == RINGBUF_TYPE_PADDING) {
2297 cpu_relax();
2298 goto again;
2299 }
2300
2056 return event; 2301 return event;
2057} 2302}
2058EXPORT_SYMBOL_GPL(ring_buffer_consume); 2303EXPORT_SYMBOL_GPL(ring_buffer_consume);
@@ -2131,6 +2376,7 @@ ring_buffer_read(struct ring_buffer_iter *iter, u64 *ts)
2131 struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer; 2376 struct ring_buffer_per_cpu *cpu_buffer = iter->cpu_buffer;
2132 unsigned long flags; 2377 unsigned long flags;
2133 2378
2379 again:
2134 spin_lock_irqsave(&cpu_buffer->reader_lock, flags); 2380 spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
2135 event = rb_iter_peek(iter, ts); 2381 event = rb_iter_peek(iter, ts);
2136 if (!event) 2382 if (!event)
@@ -2140,6 +2386,11 @@ ring_buffer_read(struct ring_buffer_iter *iter, u64 *ts)
2140 out: 2386 out:
2141 spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); 2387 spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
2142 2388
2389 if (event && event->type == RINGBUF_TYPE_PADDING) {
2390 cpu_relax();
2391 goto again;
2392 }
2393
2143 return event; 2394 return event;
2144} 2395}
2145EXPORT_SYMBOL_GPL(ring_buffer_read); 2396EXPORT_SYMBOL_GPL(ring_buffer_read);
@@ -2232,6 +2483,7 @@ int ring_buffer_empty(struct ring_buffer *buffer)
2232 if (!rb_per_cpu_empty(cpu_buffer)) 2483 if (!rb_per_cpu_empty(cpu_buffer))
2233 return 0; 2484 return 0;
2234 } 2485 }
2486
2235 return 1; 2487 return 1;
2236} 2488}
2237EXPORT_SYMBOL_GPL(ring_buffer_empty); 2489EXPORT_SYMBOL_GPL(ring_buffer_empty);
@@ -2244,12 +2496,16 @@ EXPORT_SYMBOL_GPL(ring_buffer_empty);
2244int ring_buffer_empty_cpu(struct ring_buffer *buffer, int cpu) 2496int ring_buffer_empty_cpu(struct ring_buffer *buffer, int cpu)
2245{ 2497{
2246 struct ring_buffer_per_cpu *cpu_buffer; 2498 struct ring_buffer_per_cpu *cpu_buffer;
2499 int ret;
2247 2500
2248 if (!cpumask_test_cpu(cpu, buffer->cpumask)) 2501 if (!cpumask_test_cpu(cpu, buffer->cpumask))
2249 return 1; 2502 return 1;
2250 2503
2251 cpu_buffer = buffer->buffers[cpu]; 2504 cpu_buffer = buffer->buffers[cpu];
2252 return rb_per_cpu_empty(cpu_buffer); 2505 ret = rb_per_cpu_empty(cpu_buffer);
2506
2507
2508 return ret;
2253} 2509}
2254EXPORT_SYMBOL_GPL(ring_buffer_empty_cpu); 2510EXPORT_SYMBOL_GPL(ring_buffer_empty_cpu);
2255 2511
@@ -2268,18 +2524,36 @@ int ring_buffer_swap_cpu(struct ring_buffer *buffer_a,
2268{ 2524{
2269 struct ring_buffer_per_cpu *cpu_buffer_a; 2525 struct ring_buffer_per_cpu *cpu_buffer_a;
2270 struct ring_buffer_per_cpu *cpu_buffer_b; 2526 struct ring_buffer_per_cpu *cpu_buffer_b;
2527 int ret = -EINVAL;
2271 2528
2272 if (!cpumask_test_cpu(cpu, buffer_a->cpumask) || 2529 if (!cpumask_test_cpu(cpu, buffer_a->cpumask) ||
2273 !cpumask_test_cpu(cpu, buffer_b->cpumask)) 2530 !cpumask_test_cpu(cpu, buffer_b->cpumask))
2274 return -EINVAL; 2531 goto out;
2275 2532
2276 /* At least make sure the two buffers are somewhat the same */ 2533 /* At least make sure the two buffers are somewhat the same */
2277 if (buffer_a->pages != buffer_b->pages) 2534 if (buffer_a->pages != buffer_b->pages)
2278 return -EINVAL; 2535 goto out;
2536
2537 ret = -EAGAIN;
2538
2539 if (ring_buffer_flags != RB_BUFFERS_ON)
2540 goto out;
2541
2542 if (atomic_read(&buffer_a->record_disabled))
2543 goto out;
2544
2545 if (atomic_read(&buffer_b->record_disabled))
2546 goto out;
2279 2547
2280 cpu_buffer_a = buffer_a->buffers[cpu]; 2548 cpu_buffer_a = buffer_a->buffers[cpu];
2281 cpu_buffer_b = buffer_b->buffers[cpu]; 2549 cpu_buffer_b = buffer_b->buffers[cpu];
2282 2550
2551 if (atomic_read(&cpu_buffer_a->record_disabled))
2552 goto out;
2553
2554 if (atomic_read(&cpu_buffer_b->record_disabled))
2555 goto out;
2556
2283 /* 2557 /*
2284 * We can't do a synchronize_sched here because this 2558 * We can't do a synchronize_sched here because this
2285 * function can be called in atomic context. 2559 * function can be called in atomic context.
@@ -2298,18 +2572,21 @@ int ring_buffer_swap_cpu(struct ring_buffer *buffer_a,
2298 atomic_dec(&cpu_buffer_a->record_disabled); 2572 atomic_dec(&cpu_buffer_a->record_disabled);
2299 atomic_dec(&cpu_buffer_b->record_disabled); 2573 atomic_dec(&cpu_buffer_b->record_disabled);
2300 2574
2301 return 0; 2575 ret = 0;
2576out:
2577 return ret;
2302} 2578}
2303EXPORT_SYMBOL_GPL(ring_buffer_swap_cpu); 2579EXPORT_SYMBOL_GPL(ring_buffer_swap_cpu);
2304 2580
2305static void rb_remove_entries(struct ring_buffer_per_cpu *cpu_buffer, 2581static void rb_remove_entries(struct ring_buffer_per_cpu *cpu_buffer,
2306 struct buffer_data_page *bpage) 2582 struct buffer_data_page *bpage,
2583 unsigned int offset)
2307{ 2584{
2308 struct ring_buffer_event *event; 2585 struct ring_buffer_event *event;
2309 unsigned long head; 2586 unsigned long head;
2310 2587
2311 __raw_spin_lock(&cpu_buffer->lock); 2588 __raw_spin_lock(&cpu_buffer->lock);
2312 for (head = 0; head < local_read(&bpage->commit); 2589 for (head = offset; head < local_read(&bpage->commit);
2313 head += rb_event_length(event)) { 2590 head += rb_event_length(event)) {
2314 2591
2315 event = __rb_data_page_index(bpage, head); 2592 event = __rb_data_page_index(bpage, head);
@@ -2340,8 +2617,8 @@ static void rb_remove_entries(struct ring_buffer_per_cpu *cpu_buffer,
2340 */ 2617 */
2341void *ring_buffer_alloc_read_page(struct ring_buffer *buffer) 2618void *ring_buffer_alloc_read_page(struct ring_buffer *buffer)
2342{ 2619{
2343 unsigned long addr;
2344 struct buffer_data_page *bpage; 2620 struct buffer_data_page *bpage;
2621 unsigned long addr;
2345 2622
2346 addr = __get_free_page(GFP_KERNEL); 2623 addr = __get_free_page(GFP_KERNEL);
2347 if (!addr) 2624 if (!addr)
@@ -2349,6 +2626,8 @@ void *ring_buffer_alloc_read_page(struct ring_buffer *buffer)
2349 2626
2350 bpage = (void *)addr; 2627 bpage = (void *)addr;
2351 2628
2629 rb_init_page(bpage);
2630
2352 return bpage; 2631 return bpage;
2353} 2632}
2354 2633
@@ -2368,6 +2647,7 @@ void ring_buffer_free_read_page(struct ring_buffer *buffer, void *data)
2368 * ring_buffer_read_page - extract a page from the ring buffer 2647 * ring_buffer_read_page - extract a page from the ring buffer
2369 * @buffer: buffer to extract from 2648 * @buffer: buffer to extract from
2370 * @data_page: the page to use allocated from ring_buffer_alloc_read_page 2649 * @data_page: the page to use allocated from ring_buffer_alloc_read_page
2650 * @len: amount to extract
2371 * @cpu: the cpu of the buffer to extract 2651 * @cpu: the cpu of the buffer to extract
2372 * @full: should the extraction only happen when the page is full. 2652 * @full: should the extraction only happen when the page is full.
2373 * 2653 *
@@ -2377,12 +2657,12 @@ void ring_buffer_free_read_page(struct ring_buffer *buffer, void *data)
2377 * to swap with a page in the ring buffer. 2657 * to swap with a page in the ring buffer.
2378 * 2658 *
2379 * for example: 2659 * for example:
2380 * rpage = ring_buffer_alloc_page(buffer); 2660 * rpage = ring_buffer_alloc_read_page(buffer);
2381 * if (!rpage) 2661 * if (!rpage)
2382 * return error; 2662 * return error;
2383 * ret = ring_buffer_read_page(buffer, &rpage, cpu, 0); 2663 * ret = ring_buffer_read_page(buffer, &rpage, len, cpu, 0);
2384 * if (ret) 2664 * if (ret >= 0)
2385 * process_page(rpage); 2665 * process_page(rpage, ret);
2386 * 2666 *
2387 * When @full is set, the function will not return true unless 2667 * When @full is set, the function will not return true unless
2388 * the writer is off the reader page. 2668 * the writer is off the reader page.
@@ -2393,72 +2673,118 @@ void ring_buffer_free_read_page(struct ring_buffer *buffer, void *data)
2393 * responsible for that. 2673 * responsible for that.
2394 * 2674 *
2395 * Returns: 2675 * Returns:
2396 * 1 if data has been transferred 2676 * >=0 if data has been transferred, returns the offset of consumed data.
2397 * 0 if no data has been transferred. 2677 * <0 if no data has been transferred.
2398 */ 2678 */
2399int ring_buffer_read_page(struct ring_buffer *buffer, 2679int ring_buffer_read_page(struct ring_buffer *buffer,
2400 void **data_page, int cpu, int full) 2680 void **data_page, size_t len, int cpu, int full)
2401{ 2681{
2402 struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu]; 2682 struct ring_buffer_per_cpu *cpu_buffer = buffer->buffers[cpu];
2403 struct ring_buffer_event *event; 2683 struct ring_buffer_event *event;
2404 struct buffer_data_page *bpage; 2684 struct buffer_data_page *bpage;
2685 struct buffer_page *reader;
2405 unsigned long flags; 2686 unsigned long flags;
2406 int ret = 0; 2687 unsigned int commit;
2688 unsigned int read;
2689 u64 save_timestamp;
2690 int ret = -1;
2691
2692 if (!cpumask_test_cpu(cpu, buffer->cpumask))
2693 goto out;
2694
2695 /*
2696 * If len is not big enough to hold the page header, then
2697 * we can not copy anything.
2698 */
2699 if (len <= BUF_PAGE_HDR_SIZE)
2700 goto out;
2701
2702 len -= BUF_PAGE_HDR_SIZE;
2407 2703
2408 if (!data_page) 2704 if (!data_page)
2409 return 0; 2705 goto out;
2410 2706
2411 bpage = *data_page; 2707 bpage = *data_page;
2412 if (!bpage) 2708 if (!bpage)
2413 return 0; 2709 goto out;
2414 2710
2415 spin_lock_irqsave(&cpu_buffer->reader_lock, flags); 2711 spin_lock_irqsave(&cpu_buffer->reader_lock, flags);
2416 2712
2417 /* 2713 reader = rb_get_reader_page(cpu_buffer);
2418 * rb_buffer_peek will get the next ring buffer if 2714 if (!reader)
2419 * the current reader page is empty. 2715 goto out_unlock;
2420 */ 2716
2421 event = rb_buffer_peek(buffer, cpu, NULL); 2717 event = rb_reader_event(cpu_buffer);
2422 if (!event) 2718
2423 goto out; 2719 read = reader->read;
2720 commit = rb_page_commit(reader);
2424 2721
2425 /* check for data */
2426 if (!local_read(&cpu_buffer->reader_page->page->commit))
2427 goto out;
2428 /* 2722 /*
2429 * If the writer is already off of the read page, then simply 2723 * If this page has been partially read or
2430 * switch the read page with the given page. Otherwise 2724 * if len is not big enough to read the rest of the page or
2431 * we need to copy the data from the reader to the writer. 2725 * a writer is still on the page, then
2726 * we must copy the data from the page to the buffer.
2727 * Otherwise, we can simply swap the page with the one passed in.
2432 */ 2728 */
2433 if (cpu_buffer->reader_page == cpu_buffer->commit_page) { 2729 if (read || (len < (commit - read)) ||
2434 unsigned int read = cpu_buffer->reader_page->read; 2730 cpu_buffer->reader_page == cpu_buffer->commit_page) {
2731 struct buffer_data_page *rpage = cpu_buffer->reader_page->page;
2732 unsigned int rpos = read;
2733 unsigned int pos = 0;
2734 unsigned int size;
2435 2735
2436 if (full) 2736 if (full)
2437 goto out; 2737 goto out_unlock;
2438 /* The writer is still on the reader page, we must copy */ 2738
2439 bpage = cpu_buffer->reader_page->page; 2739 if (len > (commit - read))
2440 memcpy(bpage->data, 2740 len = (commit - read);
2441 cpu_buffer->reader_page->page->data + read, 2741
2442 local_read(&bpage->commit) - read); 2742 size = rb_event_length(event);
2743
2744 if (len < size)
2745 goto out_unlock;
2443 2746
2444 /* consume what was read */ 2747 /* save the current timestamp, since the user will need it */
2445 cpu_buffer->reader_page += read; 2748 save_timestamp = cpu_buffer->read_stamp;
2446 2749
2750 /* Need to copy one event at a time */
2751 do {
2752 memcpy(bpage->data + pos, rpage->data + rpos, size);
2753
2754 len -= size;
2755
2756 rb_advance_reader(cpu_buffer);
2757 rpos = reader->read;
2758 pos += size;
2759
2760 event = rb_reader_event(cpu_buffer);
2761 size = rb_event_length(event);
2762 } while (len > size);
2763
2764 /* update bpage */
2765 local_set(&bpage->commit, pos);
2766 bpage->time_stamp = save_timestamp;
2767
2768 /* we copied everything to the beginning */
2769 read = 0;
2447 } else { 2770 } else {
2448 /* swap the pages */ 2771 /* swap the pages */
2449 rb_init_page(bpage); 2772 rb_init_page(bpage);
2450 bpage = cpu_buffer->reader_page->page; 2773 bpage = reader->page;
2451 cpu_buffer->reader_page->page = *data_page; 2774 reader->page = *data_page;
2452 cpu_buffer->reader_page->read = 0; 2775 local_set(&reader->write, 0);
2776 reader->read = 0;
2453 *data_page = bpage; 2777 *data_page = bpage;
2778
2779 /* update the entry counter */
2780 rb_remove_entries(cpu_buffer, bpage, read);
2454 } 2781 }
2455 ret = 1; 2782 ret = read;
2456 2783
2457 /* update the entry counter */ 2784 out_unlock:
2458 rb_remove_entries(cpu_buffer, bpage);
2459 out:
2460 spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags); 2785 spin_unlock_irqrestore(&cpu_buffer->reader_lock, flags);
2461 2786
2787 out:
2462 return ret; 2788 return ret;
2463} 2789}
2464 2790
@@ -2466,7 +2792,7 @@ static ssize_t
2466rb_simple_read(struct file *filp, char __user *ubuf, 2792rb_simple_read(struct file *filp, char __user *ubuf,
2467 size_t cnt, loff_t *ppos) 2793 size_t cnt, loff_t *ppos)
2468{ 2794{
2469 long *p = filp->private_data; 2795 unsigned long *p = filp->private_data;
2470 char buf[64]; 2796 char buf[64];
2471 int r; 2797 int r;
2472 2798
@@ -2482,9 +2808,9 @@ static ssize_t
2482rb_simple_write(struct file *filp, const char __user *ubuf, 2808rb_simple_write(struct file *filp, const char __user *ubuf,
2483 size_t cnt, loff_t *ppos) 2809 size_t cnt, loff_t *ppos)
2484{ 2810{
2485 long *p = filp->private_data; 2811 unsigned long *p = filp->private_data;
2486 char buf[64]; 2812 char buf[64];
2487 long val; 2813 unsigned long val;
2488 int ret; 2814 int ret;
2489 2815
2490 if (cnt >= sizeof(buf)) 2816 if (cnt >= sizeof(buf))
@@ -2509,7 +2835,7 @@ rb_simple_write(struct file *filp, const char __user *ubuf,
2509 return cnt; 2835 return cnt;
2510} 2836}
2511 2837
2512static struct file_operations rb_simple_fops = { 2838static const struct file_operations rb_simple_fops = {
2513 .open = tracing_open_generic, 2839 .open = tracing_open_generic,
2514 .read = rb_simple_read, 2840 .read = rb_simple_read,
2515 .write = rb_simple_write, 2841 .write = rb_simple_write,
@@ -2532,3 +2858,42 @@ static __init int rb_init_debugfs(void)
2532} 2858}
2533 2859
2534fs_initcall(rb_init_debugfs); 2860fs_initcall(rb_init_debugfs);
2861
2862#ifdef CONFIG_HOTPLUG_CPU
2863static int rb_cpu_notify(struct notifier_block *self,
2864 unsigned long action, void *hcpu)
2865{
2866 struct ring_buffer *buffer =
2867 container_of(self, struct ring_buffer, cpu_notify);
2868 long cpu = (long)hcpu;
2869
2870 switch (action) {
2871 case CPU_UP_PREPARE:
2872 case CPU_UP_PREPARE_FROZEN:
2873 if (cpu_isset(cpu, *buffer->cpumask))
2874 return NOTIFY_OK;
2875
2876 buffer->buffers[cpu] =
2877 rb_allocate_cpu_buffer(buffer, cpu);
2878 if (!buffer->buffers[cpu]) {
2879 WARN(1, "failed to allocate ring buffer on CPU %ld\n",
2880 cpu);
2881 return NOTIFY_OK;
2882 }
2883 smp_wmb();
2884 cpu_set(cpu, *buffer->cpumask);
2885 break;
2886 case CPU_DOWN_PREPARE:
2887 case CPU_DOWN_PREPARE_FROZEN:
2888 /*
2889 * Do nothing.
2890 * If we were to free the buffer, then the user would
2891 * lose any trace that was in the buffer.
2892 */
2893 break;
2894 default:
2895 break;
2896 }
2897 return NOTIFY_OK;
2898}
2899#endif
diff --git a/kernel/trace/trace.c b/kernel/trace/trace.c
index 17bb88d86ac2..a0174a40c563 100644
--- a/kernel/trace/trace.c
+++ b/kernel/trace/trace.c
@@ -11,32 +11,33 @@
11 * Copyright (C) 2004-2006 Ingo Molnar 11 * Copyright (C) 2004-2006 Ingo Molnar
12 * Copyright (C) 2004 William Lee Irwin III 12 * Copyright (C) 2004 William Lee Irwin III
13 */ 13 */
14#include <linux/ring_buffer.h>
14#include <linux/utsrelease.h> 15#include <linux/utsrelease.h>
16#include <linux/stacktrace.h>
17#include <linux/writeback.h>
15#include <linux/kallsyms.h> 18#include <linux/kallsyms.h>
16#include <linux/seq_file.h> 19#include <linux/seq_file.h>
17#include <linux/notifier.h> 20#include <linux/notifier.h>
21#include <linux/irqflags.h>
18#include <linux/debugfs.h> 22#include <linux/debugfs.h>
19#include <linux/pagemap.h> 23#include <linux/pagemap.h>
20#include <linux/hardirq.h> 24#include <linux/hardirq.h>
21#include <linux/linkage.h> 25#include <linux/linkage.h>
22#include <linux/uaccess.h> 26#include <linux/uaccess.h>
27#include <linux/kprobes.h>
23#include <linux/ftrace.h> 28#include <linux/ftrace.h>
24#include <linux/module.h> 29#include <linux/module.h>
25#include <linux/percpu.h> 30#include <linux/percpu.h>
31#include <linux/splice.h>
26#include <linux/kdebug.h> 32#include <linux/kdebug.h>
27#include <linux/ctype.h> 33#include <linux/ctype.h>
28#include <linux/init.h> 34#include <linux/init.h>
29#include <linux/poll.h> 35#include <linux/poll.h>
30#include <linux/gfp.h> 36#include <linux/gfp.h>
31#include <linux/fs.h> 37#include <linux/fs.h>
32#include <linux/kprobes.h>
33#include <linux/writeback.h>
34
35#include <linux/stacktrace.h>
36#include <linux/ring_buffer.h>
37#include <linux/irqflags.h>
38 38
39#include "trace.h" 39#include "trace.h"
40#include "trace_output.h"
40 41
41#define TRACE_BUFFER_FLAGS (RB_FL_OVERWRITE) 42#define TRACE_BUFFER_FLAGS (RB_FL_OVERWRITE)
42 43
@@ -44,14 +45,25 @@ unsigned long __read_mostly tracing_max_latency;
44unsigned long __read_mostly tracing_thresh; 45unsigned long __read_mostly tracing_thresh;
45 46
46/* 47/*
48 * On boot up, the ring buffer is set to the minimum size, so that
49 * we do not waste memory on systems that are not using tracing.
50 */
51static int ring_buffer_expanded;
52
53/*
47 * We need to change this state when a selftest is running. 54 * We need to change this state when a selftest is running.
48 * A selftest will lurk into the ring-buffer to count the 55 * A selftest will lurk into the ring-buffer to count the
49 * entries inserted during the selftest although some concurrent 56 * entries inserted during the selftest although some concurrent
50 * insertions into the ring-buffer such as ftrace_printk could occurred 57 * insertions into the ring-buffer such as trace_printk could occurred
51 * at the same time, giving false positive or negative results. 58 * at the same time, giving false positive or negative results.
52 */ 59 */
53static bool __read_mostly tracing_selftest_running; 60static bool __read_mostly tracing_selftest_running;
54 61
62/*
63 * If a tracer is running, we do not want to run SELFTEST.
64 */
65static bool __read_mostly tracing_selftest_disabled;
66
55/* For tracers that don't implement custom flags */ 67/* For tracers that don't implement custom flags */
56static struct tracer_opt dummy_tracer_opt[] = { 68static struct tracer_opt dummy_tracer_opt[] = {
57 { } 69 { }
@@ -73,7 +85,7 @@ static int dummy_set_flag(u32 old_flags, u32 bit, int set)
73 * of the tracer is successful. But that is the only place that sets 85 * of the tracer is successful. But that is the only place that sets
74 * this back to zero. 86 * this back to zero.
75 */ 87 */
76int tracing_disabled = 1; 88static int tracing_disabled = 1;
77 89
78static DEFINE_PER_CPU(local_t, ftrace_cpu_disabled); 90static DEFINE_PER_CPU(local_t, ftrace_cpu_disabled);
79 91
@@ -91,6 +103,9 @@ static inline void ftrace_enable_cpu(void)
91 103
92static cpumask_var_t __read_mostly tracing_buffer_mask; 104static cpumask_var_t __read_mostly tracing_buffer_mask;
93 105
106/* Define which cpu buffers are currently read in trace_pipe */
107static cpumask_var_t tracing_reader_cpumask;
108
94#define for_each_tracing_cpu(cpu) \ 109#define for_each_tracing_cpu(cpu) \
95 for_each_cpu(cpu, tracing_buffer_mask) 110 for_each_cpu(cpu, tracing_buffer_mask)
96 111
@@ -109,14 +124,21 @@ static cpumask_var_t __read_mostly tracing_buffer_mask;
109 */ 124 */
110int ftrace_dump_on_oops; 125int ftrace_dump_on_oops;
111 126
112static int tracing_set_tracer(char *buf); 127static int tracing_set_tracer(const char *buf);
128
129#define BOOTUP_TRACER_SIZE 100
130static char bootup_tracer_buf[BOOTUP_TRACER_SIZE] __initdata;
131static char *default_bootup_tracer;
113 132
114static int __init set_ftrace(char *str) 133static int __init set_ftrace(char *str)
115{ 134{
116 tracing_set_tracer(str); 135 strncpy(bootup_tracer_buf, str, BOOTUP_TRACER_SIZE);
136 default_bootup_tracer = bootup_tracer_buf;
137 /* We are using ftrace early, expand it */
138 ring_buffer_expanded = 1;
117 return 1; 139 return 1;
118} 140}
119__setup("ftrace", set_ftrace); 141__setup("ftrace=", set_ftrace);
120 142
121static int __init set_ftrace_dump_on_oops(char *str) 143static int __init set_ftrace_dump_on_oops(char *str)
122{ 144{
@@ -133,13 +155,6 @@ ns2usecs(cycle_t nsec)
133 return nsec; 155 return nsec;
134} 156}
135 157
136cycle_t ftrace_now(int cpu)
137{
138 u64 ts = ring_buffer_time_stamp(cpu);
139 ring_buffer_normalize_time_stamp(cpu, &ts);
140 return ts;
141}
142
143/* 158/*
144 * The global_trace is the descriptor that holds the tracing 159 * The global_trace is the descriptor that holds the tracing
145 * buffers for the live tracing. For each CPU, it contains 160 * buffers for the live tracing. For each CPU, it contains
@@ -156,6 +171,20 @@ static struct trace_array global_trace;
156 171
157static DEFINE_PER_CPU(struct trace_array_cpu, global_trace_cpu); 172static DEFINE_PER_CPU(struct trace_array_cpu, global_trace_cpu);
158 173
174cycle_t ftrace_now(int cpu)
175{
176 u64 ts;
177
178 /* Early boot up does not have a buffer yet */
179 if (!global_trace.buffer)
180 return trace_clock_local();
181
182 ts = ring_buffer_time_stamp(global_trace.buffer, cpu);
183 ring_buffer_normalize_time_stamp(global_trace.buffer, cpu, &ts);
184
185 return ts;
186}
187
159/* 188/*
160 * The max_tr is used to snapshot the global_trace when a maximum 189 * The max_tr is used to snapshot the global_trace when a maximum
161 * latency is reached. Some tracers will use this to store a maximum 190 * latency is reached. Some tracers will use this to store a maximum
@@ -186,9 +215,6 @@ int tracing_is_enabled(void)
186 return tracer_enabled; 215 return tracer_enabled;
187} 216}
188 217
189/* function tracing enabled */
190int ftrace_function_enabled;
191
192/* 218/*
193 * trace_buf_size is the size in bytes that is allocated 219 * trace_buf_size is the size in bytes that is allocated
194 * for a buffer. Note, the number of bytes is always rounded 220 * for a buffer. Note, the number of bytes is always rounded
@@ -229,7 +255,7 @@ static DECLARE_WAIT_QUEUE_HEAD(trace_wait);
229 255
230/* trace_flags holds trace_options default values */ 256/* trace_flags holds trace_options default values */
231unsigned long trace_flags = TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK | 257unsigned long trace_flags = TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK |
232 TRACE_ITER_ANNOTATE; 258 TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO | TRACE_ITER_SLEEP_TIME;
233 259
234/** 260/**
235 * trace_wake_up - wake up tasks waiting for trace input 261 * trace_wake_up - wake up tasks waiting for trace input
@@ -280,13 +306,17 @@ static const char *trace_options[] = {
280 "block", 306 "block",
281 "stacktrace", 307 "stacktrace",
282 "sched-tree", 308 "sched-tree",
283 "ftrace_printk", 309 "trace_printk",
284 "ftrace_preempt", 310 "ftrace_preempt",
285 "branch", 311 "branch",
286 "annotate", 312 "annotate",
287 "userstacktrace", 313 "userstacktrace",
288 "sym-userobj", 314 "sym-userobj",
289 "printk-msg-only", 315 "printk-msg-only",
316 "context-info",
317 "latency-format",
318 "global-clock",
319 "sleep-time",
290 NULL 320 NULL
291}; 321};
292 322
@@ -326,146 +356,37 @@ __update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu)
326 data->rt_priority = tsk->rt_priority; 356 data->rt_priority = tsk->rt_priority;
327 357
328 /* record this tasks comm */ 358 /* record this tasks comm */
329 tracing_record_cmdline(current); 359 tracing_record_cmdline(tsk);
330} 360}
331 361
332/** 362ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf, size_t cnt)
333 * trace_seq_printf - sequence printing of trace information
334 * @s: trace sequence descriptor
335 * @fmt: printf format string
336 *
337 * The tracer may use either sequence operations or its own
338 * copy to user routines. To simplify formating of a trace
339 * trace_seq_printf is used to store strings into a special
340 * buffer (@s). Then the output may be either used by
341 * the sequencer or pulled into another buffer.
342 */
343int
344trace_seq_printf(struct trace_seq *s, const char *fmt, ...)
345{ 363{
346 int len = (PAGE_SIZE - 1) - s->len; 364 int len;
347 va_list ap;
348 int ret; 365 int ret;
349 366
350 if (!len) 367 if (!cnt)
351 return 0;
352
353 va_start(ap, fmt);
354 ret = vsnprintf(s->buffer + s->len, len, fmt, ap);
355 va_end(ap);
356
357 /* If we can't write it all, don't bother writing anything */
358 if (ret >= len)
359 return 0;
360
361 s->len += ret;
362
363 return len;
364}
365
366/**
367 * trace_seq_puts - trace sequence printing of simple string
368 * @s: trace sequence descriptor
369 * @str: simple string to record
370 *
371 * The tracer may use either the sequence operations or its own
372 * copy to user routines. This function records a simple string
373 * into a special buffer (@s) for later retrieval by a sequencer
374 * or other mechanism.
375 */
376static int
377trace_seq_puts(struct trace_seq *s, const char *str)
378{
379 int len = strlen(str);
380
381 if (len > ((PAGE_SIZE - 1) - s->len))
382 return 0;
383
384 memcpy(s->buffer + s->len, str, len);
385 s->len += len;
386
387 return len;
388}
389
390static int
391trace_seq_putc(struct trace_seq *s, unsigned char c)
392{
393 if (s->len >= (PAGE_SIZE - 1))
394 return 0;
395
396 s->buffer[s->len++] = c;
397
398 return 1;
399}
400
401static int
402trace_seq_putmem(struct trace_seq *s, void *mem, size_t len)
403{
404 if (len > ((PAGE_SIZE - 1) - s->len))
405 return 0; 368 return 0;
406 369
407 memcpy(s->buffer + s->len, mem, len); 370 if (s->len <= s->readpos)
408 s->len += len; 371 return -EBUSY;
409
410 return len;
411}
412
413#define MAX_MEMHEX_BYTES 8
414#define HEX_CHARS (MAX_MEMHEX_BYTES*2 + 1)
415
416static int
417trace_seq_putmem_hex(struct trace_seq *s, void *mem, size_t len)
418{
419 unsigned char hex[HEX_CHARS];
420 unsigned char *data = mem;
421 int i, j;
422
423#ifdef __BIG_ENDIAN
424 for (i = 0, j = 0; i < len; i++) {
425#else
426 for (i = len-1, j = 0; i >= 0; i--) {
427#endif
428 hex[j++] = hex_asc_hi(data[i]);
429 hex[j++] = hex_asc_lo(data[i]);
430 }
431 hex[j++] = ' ';
432
433 return trace_seq_putmem(s, hex, j);
434}
435
436static int
437trace_seq_path(struct trace_seq *s, struct path *path)
438{
439 unsigned char *p;
440 372
441 if (s->len >= (PAGE_SIZE - 1)) 373 len = s->len - s->readpos;
442 return 0; 374 if (cnt > len)
443 p = d_path(path, s->buffer + s->len, PAGE_SIZE - s->len); 375 cnt = len;
444 if (!IS_ERR(p)) { 376 ret = copy_to_user(ubuf, s->buffer + s->readpos, cnt);
445 p = mangle_path(s->buffer + s->len, p, "\n"); 377 if (ret == cnt)
446 if (p) { 378 return -EFAULT;
447 s->len = p - s->buffer;
448 return 1;
449 }
450 } else {
451 s->buffer[s->len++] = '?';
452 return 1;
453 }
454 379
455 return 0; 380 cnt -= ret;
456}
457 381
458static void 382 s->readpos += cnt;
459trace_seq_reset(struct trace_seq *s) 383 return cnt;
460{
461 s->len = 0;
462 s->readpos = 0;
463} 384}
464 385
465ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf, size_t cnt) 386static ssize_t trace_seq_to_buffer(struct trace_seq *s, void *buf, size_t cnt)
466{ 387{
467 int len; 388 int len;
468 int ret; 389 void *ret;
469 390
470 if (s->len <= s->readpos) 391 if (s->len <= s->readpos)
471 return -EBUSY; 392 return -EBUSY;
@@ -473,11 +394,11 @@ ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf, size_t cnt)
473 len = s->len - s->readpos; 394 len = s->len - s->readpos;
474 if (cnt > len) 395 if (cnt > len)
475 cnt = len; 396 cnt = len;
476 ret = copy_to_user(ubuf, s->buffer + s->readpos, cnt); 397 ret = memcpy(buf, s->buffer + s->readpos, cnt);
477 if (ret) 398 if (!ret)
478 return -EFAULT; 399 return -EFAULT;
479 400
480 s->readpos += len; 401 s->readpos += cnt;
481 return cnt; 402 return cnt;
482} 403}
483 404
@@ -489,7 +410,7 @@ trace_print_seq(struct seq_file *m, struct trace_seq *s)
489 s->buffer[len] = 0; 410 s->buffer[len] = 0;
490 seq_puts(m, s->buffer); 411 seq_puts(m, s->buffer);
491 412
492 trace_seq_reset(s); 413 trace_seq_init(s);
493} 414}
494 415
495/** 416/**
@@ -543,7 +464,7 @@ update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
543 464
544 ftrace_enable_cpu(); 465 ftrace_enable_cpu();
545 466
546 WARN_ON_ONCE(ret); 467 WARN_ON_ONCE(ret && ret != -EAGAIN);
547 468
548 __update_max_tr(tr, tsk, cpu); 469 __update_max_tr(tr, tsk, cpu);
549 __raw_spin_unlock(&ftrace_max_lock); 470 __raw_spin_unlock(&ftrace_max_lock);
@@ -556,6 +477,8 @@ update_max_tr_single(struct trace_array *tr, struct task_struct *tsk, int cpu)
556 * Register a new plugin tracer. 477 * Register a new plugin tracer.
557 */ 478 */
558int register_tracer(struct tracer *type) 479int register_tracer(struct tracer *type)
480__releases(kernel_lock)
481__acquires(kernel_lock)
559{ 482{
560 struct tracer *t; 483 struct tracer *t;
561 int len; 484 int len;
@@ -594,9 +517,12 @@ int register_tracer(struct tracer *type)
594 else 517 else
595 if (!type->flags->opts) 518 if (!type->flags->opts)
596 type->flags->opts = dummy_tracer_opt; 519 type->flags->opts = dummy_tracer_opt;
520 if (!type->wait_pipe)
521 type->wait_pipe = default_wait_pipe;
522
597 523
598#ifdef CONFIG_FTRACE_STARTUP_TEST 524#ifdef CONFIG_FTRACE_STARTUP_TEST
599 if (type->selftest) { 525 if (type->selftest && !tracing_selftest_disabled) {
600 struct tracer *saved_tracer = current_trace; 526 struct tracer *saved_tracer = current_trace;
601 struct trace_array *tr = &global_trace; 527 struct trace_array *tr = &global_trace;
602 int i; 528 int i;
@@ -638,8 +564,26 @@ int register_tracer(struct tracer *type)
638 out: 564 out:
639 tracing_selftest_running = false; 565 tracing_selftest_running = false;
640 mutex_unlock(&trace_types_lock); 566 mutex_unlock(&trace_types_lock);
641 lock_kernel();
642 567
568 if (ret || !default_bootup_tracer)
569 goto out_unlock;
570
571 if (strncmp(default_bootup_tracer, type->name, BOOTUP_TRACER_SIZE))
572 goto out_unlock;
573
574 printk(KERN_INFO "Starting tracer '%s'\n", type->name);
575 /* Do we want this tracer to start on bootup? */
576 tracing_set_tracer(type->name);
577 default_bootup_tracer = NULL;
578 /* disable other selftests, since this will break it. */
579 tracing_selftest_disabled = 1;
580#ifdef CONFIG_FTRACE_STARTUP_TEST
581 printk(KERN_INFO "Disabling FTRACE selftests due to running tracer '%s'\n",
582 type->name);
583#endif
584
585 out_unlock:
586 lock_kernel();
643 return ret; 587 return ret;
644} 588}
645 589
@@ -658,6 +602,15 @@ void unregister_tracer(struct tracer *type)
658 602
659 found: 603 found:
660 *t = (*t)->next; 604 *t = (*t)->next;
605
606 if (type == current_trace && tracer_enabled) {
607 tracer_enabled = 0;
608 tracing_stop();
609 if (current_trace->stop)
610 current_trace->stop(&global_trace);
611 current_trace = &nop_trace;
612 }
613
661 if (strlen(type->name) != max_tracer_type_len) 614 if (strlen(type->name) != max_tracer_type_len)
662 goto out; 615 goto out;
663 616
@@ -689,19 +642,20 @@ void tracing_reset_online_cpus(struct trace_array *tr)
689} 642}
690 643
691#define SAVED_CMDLINES 128 644#define SAVED_CMDLINES 128
645#define NO_CMDLINE_MAP UINT_MAX
692static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1]; 646static unsigned map_pid_to_cmdline[PID_MAX_DEFAULT+1];
693static unsigned map_cmdline_to_pid[SAVED_CMDLINES]; 647static unsigned map_cmdline_to_pid[SAVED_CMDLINES];
694static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN]; 648static char saved_cmdlines[SAVED_CMDLINES][TASK_COMM_LEN];
695static int cmdline_idx; 649static int cmdline_idx;
696static DEFINE_SPINLOCK(trace_cmdline_lock); 650static raw_spinlock_t trace_cmdline_lock = __RAW_SPIN_LOCK_UNLOCKED;
697 651
698/* temporary disable recording */ 652/* temporary disable recording */
699atomic_t trace_record_cmdline_disabled __read_mostly; 653static atomic_t trace_record_cmdline_disabled __read_mostly;
700 654
701static void trace_init_cmdlines(void) 655static void trace_init_cmdlines(void)
702{ 656{
703 memset(&map_pid_to_cmdline, -1, sizeof(map_pid_to_cmdline)); 657 memset(&map_pid_to_cmdline, NO_CMDLINE_MAP, sizeof(map_pid_to_cmdline));
704 memset(&map_cmdline_to_pid, -1, sizeof(map_cmdline_to_pid)); 658 memset(&map_cmdline_to_pid, NO_CMDLINE_MAP, sizeof(map_cmdline_to_pid));
705 cmdline_idx = 0; 659 cmdline_idx = 0;
706} 660}
707 661
@@ -738,13 +692,12 @@ void tracing_start(void)
738 return; 692 return;
739 693
740 spin_lock_irqsave(&tracing_start_lock, flags); 694 spin_lock_irqsave(&tracing_start_lock, flags);
741 if (--trace_stop_count) 695 if (--trace_stop_count) {
742 goto out; 696 if (trace_stop_count < 0) {
743 697 /* Someone screwed up their debugging */
744 if (trace_stop_count < 0) { 698 WARN_ON_ONCE(1);
745 /* Someone screwed up their debugging */ 699 trace_stop_count = 0;
746 WARN_ON_ONCE(1); 700 }
747 trace_stop_count = 0;
748 goto out; 701 goto out;
749 } 702 }
750 703
@@ -794,8 +747,7 @@ void trace_stop_cmdline_recording(void);
794 747
795static void trace_save_cmdline(struct task_struct *tsk) 748static void trace_save_cmdline(struct task_struct *tsk)
796{ 749{
797 unsigned map; 750 unsigned pid, idx;
798 unsigned idx;
799 751
800 if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT)) 752 if (!tsk->pid || unlikely(tsk->pid > PID_MAX_DEFAULT))
801 return; 753 return;
@@ -806,17 +758,24 @@ static void trace_save_cmdline(struct task_struct *tsk)
806 * nor do we want to disable interrupts, 758 * nor do we want to disable interrupts,
807 * so if we miss here, then better luck next time. 759 * so if we miss here, then better luck next time.
808 */ 760 */
809 if (!spin_trylock(&trace_cmdline_lock)) 761 if (!__raw_spin_trylock(&trace_cmdline_lock))
810 return; 762 return;
811 763
812 idx = map_pid_to_cmdline[tsk->pid]; 764 idx = map_pid_to_cmdline[tsk->pid];
813 if (idx >= SAVED_CMDLINES) { 765 if (idx == NO_CMDLINE_MAP) {
814 idx = (cmdline_idx + 1) % SAVED_CMDLINES; 766 idx = (cmdline_idx + 1) % SAVED_CMDLINES;
815 767
816 map = map_cmdline_to_pid[idx]; 768 /*
817 if (map <= PID_MAX_DEFAULT) 769 * Check whether the cmdline buffer at idx has a pid
818 map_pid_to_cmdline[map] = (unsigned)-1; 770 * mapped. We are going to overwrite that entry so we
771 * need to clear the map_pid_to_cmdline. Otherwise we
772 * would read the new comm for the old pid.
773 */
774 pid = map_cmdline_to_pid[idx];
775 if (pid != NO_CMDLINE_MAP)
776 map_pid_to_cmdline[pid] = NO_CMDLINE_MAP;
819 777
778 map_cmdline_to_pid[idx] = tsk->pid;
820 map_pid_to_cmdline[tsk->pid] = idx; 779 map_pid_to_cmdline[tsk->pid] = idx;
821 780
822 cmdline_idx = idx; 781 cmdline_idx = idx;
@@ -824,33 +783,37 @@ static void trace_save_cmdline(struct task_struct *tsk)
824 783
825 memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN); 784 memcpy(&saved_cmdlines[idx], tsk->comm, TASK_COMM_LEN);
826 785
827 spin_unlock(&trace_cmdline_lock); 786 __raw_spin_unlock(&trace_cmdline_lock);
828} 787}
829 788
830char *trace_find_cmdline(int pid) 789void trace_find_cmdline(int pid, char comm[])
831{ 790{
832 char *cmdline = "<...>";
833 unsigned map; 791 unsigned map;
834 792
835 if (!pid) 793 if (!pid) {
836 return "<idle>"; 794 strcpy(comm, "<idle>");
795 return;
796 }
837 797
838 if (pid > PID_MAX_DEFAULT) 798 if (pid > PID_MAX_DEFAULT) {
839 goto out; 799 strcpy(comm, "<...>");
800 return;
801 }
840 802
803 __raw_spin_lock(&trace_cmdline_lock);
841 map = map_pid_to_cmdline[pid]; 804 map = map_pid_to_cmdline[pid];
842 if (map >= SAVED_CMDLINES) 805 if (map != NO_CMDLINE_MAP)
843 goto out; 806 strcpy(comm, saved_cmdlines[map]);
844 807 else
845 cmdline = saved_cmdlines[map]; 808 strcpy(comm, "<...>");
846 809
847 out: 810 __raw_spin_unlock(&trace_cmdline_lock);
848 return cmdline;
849} 811}
850 812
851void tracing_record_cmdline(struct task_struct *tsk) 813void tracing_record_cmdline(struct task_struct *tsk)
852{ 814{
853 if (atomic_read(&trace_record_cmdline_disabled)) 815 if (atomic_read(&trace_record_cmdline_disabled) || !tracer_enabled ||
816 !tracing_is_on())
854 return; 817 return;
855 818
856 trace_save_cmdline(tsk); 819 trace_save_cmdline(tsk);
@@ -864,7 +827,7 @@ tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags,
864 827
865 entry->preempt_count = pc & 0xff; 828 entry->preempt_count = pc & 0xff;
866 entry->pid = (tsk) ? tsk->pid : 0; 829 entry->pid = (tsk) ? tsk->pid : 0;
867 entry->tgid = (tsk) ? tsk->tgid : 0; 830 entry->tgid = (tsk) ? tsk->tgid : 0;
868 entry->flags = 831 entry->flags =
869#ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT 832#ifdef CONFIG_TRACE_IRQFLAGS_SUPPORT
870 (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) | 833 (irqs_disabled_flags(flags) ? TRACE_FLAG_IRQS_OFF : 0) |
@@ -876,78 +839,132 @@ tracing_generic_entry_update(struct trace_entry *entry, unsigned long flags,
876 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0); 839 (need_resched() ? TRACE_FLAG_NEED_RESCHED : 0);
877} 840}
878 841
842struct ring_buffer_event *trace_buffer_lock_reserve(struct trace_array *tr,
843 unsigned char type,
844 unsigned long len,
845 unsigned long flags, int pc)
846{
847 struct ring_buffer_event *event;
848
849 event = ring_buffer_lock_reserve(tr->buffer, len);
850 if (event != NULL) {
851 struct trace_entry *ent = ring_buffer_event_data(event);
852
853 tracing_generic_entry_update(ent, flags, pc);
854 ent->type = type;
855 }
856
857 return event;
858}
859static void ftrace_trace_stack(struct trace_array *tr,
860 unsigned long flags, int skip, int pc);
861static void ftrace_trace_userstack(struct trace_array *tr,
862 unsigned long flags, int pc);
863
864static inline void __trace_buffer_unlock_commit(struct trace_array *tr,
865 struct ring_buffer_event *event,
866 unsigned long flags, int pc,
867 int wake)
868{
869 ring_buffer_unlock_commit(tr->buffer, event);
870
871 ftrace_trace_stack(tr, flags, 6, pc);
872 ftrace_trace_userstack(tr, flags, pc);
873
874 if (wake)
875 trace_wake_up();
876}
877
878void trace_buffer_unlock_commit(struct trace_array *tr,
879 struct ring_buffer_event *event,
880 unsigned long flags, int pc)
881{
882 __trace_buffer_unlock_commit(tr, event, flags, pc, 1);
883}
884
885struct ring_buffer_event *
886trace_current_buffer_lock_reserve(unsigned char type, unsigned long len,
887 unsigned long flags, int pc)
888{
889 return trace_buffer_lock_reserve(&global_trace,
890 type, len, flags, pc);
891}
892
893void trace_current_buffer_unlock_commit(struct ring_buffer_event *event,
894 unsigned long flags, int pc)
895{
896 return __trace_buffer_unlock_commit(&global_trace, event, flags, pc, 1);
897}
898
899void trace_nowake_buffer_unlock_commit(struct ring_buffer_event *event,
900 unsigned long flags, int pc)
901{
902 return __trace_buffer_unlock_commit(&global_trace, event, flags, pc, 0);
903}
904
879void 905void
880trace_function(struct trace_array *tr, struct trace_array_cpu *data, 906trace_function(struct trace_array *tr,
881 unsigned long ip, unsigned long parent_ip, unsigned long flags, 907 unsigned long ip, unsigned long parent_ip, unsigned long flags,
882 int pc) 908 int pc)
883{ 909{
884 struct ring_buffer_event *event; 910 struct ring_buffer_event *event;
885 struct ftrace_entry *entry; 911 struct ftrace_entry *entry;
886 unsigned long irq_flags;
887 912
888 /* If we are reading the ring buffer, don't trace */ 913 /* If we are reading the ring buffer, don't trace */
889 if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled)))) 914 if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled))))
890 return; 915 return;
891 916
892 event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry), 917 event = trace_buffer_lock_reserve(tr, TRACE_FN, sizeof(*entry),
893 &irq_flags); 918 flags, pc);
894 if (!event) 919 if (!event)
895 return; 920 return;
896 entry = ring_buffer_event_data(event); 921 entry = ring_buffer_event_data(event);
897 tracing_generic_entry_update(&entry->ent, flags, pc);
898 entry->ent.type = TRACE_FN;
899 entry->ip = ip; 922 entry->ip = ip;
900 entry->parent_ip = parent_ip; 923 entry->parent_ip = parent_ip;
901 ring_buffer_unlock_commit(tr->buffer, event, irq_flags); 924 ring_buffer_unlock_commit(tr->buffer, event);
902} 925}
903 926
904#ifdef CONFIG_FUNCTION_GRAPH_TRACER 927#ifdef CONFIG_FUNCTION_GRAPH_TRACER
905static void __trace_graph_entry(struct trace_array *tr, 928static int __trace_graph_entry(struct trace_array *tr,
906 struct trace_array_cpu *data,
907 struct ftrace_graph_ent *trace, 929 struct ftrace_graph_ent *trace,
908 unsigned long flags, 930 unsigned long flags,
909 int pc) 931 int pc)
910{ 932{
911 struct ring_buffer_event *event; 933 struct ring_buffer_event *event;
912 struct ftrace_graph_ent_entry *entry; 934 struct ftrace_graph_ent_entry *entry;
913 unsigned long irq_flags;
914 935
915 if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled)))) 936 if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled))))
916 return; 937 return 0;
917 938
918 event = ring_buffer_lock_reserve(global_trace.buffer, sizeof(*entry), 939 event = trace_buffer_lock_reserve(&global_trace, TRACE_GRAPH_ENT,
919 &irq_flags); 940 sizeof(*entry), flags, pc);
920 if (!event) 941 if (!event)
921 return; 942 return 0;
922 entry = ring_buffer_event_data(event); 943 entry = ring_buffer_event_data(event);
923 tracing_generic_entry_update(&entry->ent, flags, pc);
924 entry->ent.type = TRACE_GRAPH_ENT;
925 entry->graph_ent = *trace; 944 entry->graph_ent = *trace;
926 ring_buffer_unlock_commit(global_trace.buffer, event, irq_flags); 945 ring_buffer_unlock_commit(global_trace.buffer, event);
946
947 return 1;
927} 948}
928 949
929static void __trace_graph_return(struct trace_array *tr, 950static void __trace_graph_return(struct trace_array *tr,
930 struct trace_array_cpu *data,
931 struct ftrace_graph_ret *trace, 951 struct ftrace_graph_ret *trace,
932 unsigned long flags, 952 unsigned long flags,
933 int pc) 953 int pc)
934{ 954{
935 struct ring_buffer_event *event; 955 struct ring_buffer_event *event;
936 struct ftrace_graph_ret_entry *entry; 956 struct ftrace_graph_ret_entry *entry;
937 unsigned long irq_flags;
938 957
939 if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled)))) 958 if (unlikely(local_read(&__get_cpu_var(ftrace_cpu_disabled))))
940 return; 959 return;
941 960
942 event = ring_buffer_lock_reserve(global_trace.buffer, sizeof(*entry), 961 event = trace_buffer_lock_reserve(&global_trace, TRACE_GRAPH_RET,
943 &irq_flags); 962 sizeof(*entry), flags, pc);
944 if (!event) 963 if (!event)
945 return; 964 return;
946 entry = ring_buffer_event_data(event); 965 entry = ring_buffer_event_data(event);
947 tracing_generic_entry_update(&entry->ent, flags, pc);
948 entry->ent.type = TRACE_GRAPH_RET;
949 entry->ret = *trace; 966 entry->ret = *trace;
950 ring_buffer_unlock_commit(global_trace.buffer, event, irq_flags); 967 ring_buffer_unlock_commit(global_trace.buffer, event);
951} 968}
952#endif 969#endif
953 970
@@ -957,31 +974,23 @@ ftrace(struct trace_array *tr, struct trace_array_cpu *data,
957 int pc) 974 int pc)
958{ 975{
959 if (likely(!atomic_read(&data->disabled))) 976 if (likely(!atomic_read(&data->disabled)))
960 trace_function(tr, data, ip, parent_ip, flags, pc); 977 trace_function(tr, ip, parent_ip, flags, pc);
961} 978}
962 979
963static void ftrace_trace_stack(struct trace_array *tr, 980static void __ftrace_trace_stack(struct trace_array *tr,
964 struct trace_array_cpu *data, 981 unsigned long flags,
965 unsigned long flags, 982 int skip, int pc)
966 int skip, int pc)
967{ 983{
968#ifdef CONFIG_STACKTRACE 984#ifdef CONFIG_STACKTRACE
969 struct ring_buffer_event *event; 985 struct ring_buffer_event *event;
970 struct stack_entry *entry; 986 struct stack_entry *entry;
971 struct stack_trace trace; 987 struct stack_trace trace;
972 unsigned long irq_flags;
973 988
974 if (!(trace_flags & TRACE_ITER_STACKTRACE)) 989 event = trace_buffer_lock_reserve(tr, TRACE_STACK,
975 return; 990 sizeof(*entry), flags, pc);
976
977 event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
978 &irq_flags);
979 if (!event) 991 if (!event)
980 return; 992 return;
981 entry = ring_buffer_event_data(event); 993 entry = ring_buffer_event_data(event);
982 tracing_generic_entry_update(&entry->ent, flags, pc);
983 entry->ent.type = TRACE_STACK;
984
985 memset(&entry->caller, 0, sizeof(entry->caller)); 994 memset(&entry->caller, 0, sizeof(entry->caller));
986 995
987 trace.nr_entries = 0; 996 trace.nr_entries = 0;
@@ -990,38 +999,43 @@ static void ftrace_trace_stack(struct trace_array *tr,
990 trace.entries = entry->caller; 999 trace.entries = entry->caller;
991 1000
992 save_stack_trace(&trace); 1001 save_stack_trace(&trace);
993 ring_buffer_unlock_commit(tr->buffer, event, irq_flags); 1002 ring_buffer_unlock_commit(tr->buffer, event);
994#endif 1003#endif
995} 1004}
996 1005
1006static void ftrace_trace_stack(struct trace_array *tr,
1007 unsigned long flags,
1008 int skip, int pc)
1009{
1010 if (!(trace_flags & TRACE_ITER_STACKTRACE))
1011 return;
1012
1013 __ftrace_trace_stack(tr, flags, skip, pc);
1014}
1015
997void __trace_stack(struct trace_array *tr, 1016void __trace_stack(struct trace_array *tr,
998 struct trace_array_cpu *data,
999 unsigned long flags, 1017 unsigned long flags,
1000 int skip) 1018 int skip, int pc)
1001{ 1019{
1002 ftrace_trace_stack(tr, data, flags, skip, preempt_count()); 1020 __ftrace_trace_stack(tr, flags, skip, pc);
1003} 1021}
1004 1022
1005static void ftrace_trace_userstack(struct trace_array *tr, 1023static void ftrace_trace_userstack(struct trace_array *tr,
1006 struct trace_array_cpu *data, 1024 unsigned long flags, int pc)
1007 unsigned long flags, int pc)
1008{ 1025{
1009#ifdef CONFIG_STACKTRACE 1026#ifdef CONFIG_STACKTRACE
1010 struct ring_buffer_event *event; 1027 struct ring_buffer_event *event;
1011 struct userstack_entry *entry; 1028 struct userstack_entry *entry;
1012 struct stack_trace trace; 1029 struct stack_trace trace;
1013 unsigned long irq_flags;
1014 1030
1015 if (!(trace_flags & TRACE_ITER_USERSTACKTRACE)) 1031 if (!(trace_flags & TRACE_ITER_USERSTACKTRACE))
1016 return; 1032 return;
1017 1033
1018 event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry), 1034 event = trace_buffer_lock_reserve(tr, TRACE_USER_STACK,
1019 &irq_flags); 1035 sizeof(*entry), flags, pc);
1020 if (!event) 1036 if (!event)
1021 return; 1037 return;
1022 entry = ring_buffer_event_data(event); 1038 entry = ring_buffer_event_data(event);
1023 tracing_generic_entry_update(&entry->ent, flags, pc);
1024 entry->ent.type = TRACE_USER_STACK;
1025 1039
1026 memset(&entry->caller, 0, sizeof(entry->caller)); 1040 memset(&entry->caller, 0, sizeof(entry->caller));
1027 1041
@@ -1031,70 +1045,58 @@ static void ftrace_trace_userstack(struct trace_array *tr,
1031 trace.entries = entry->caller; 1045 trace.entries = entry->caller;
1032 1046
1033 save_stack_trace_user(&trace); 1047 save_stack_trace_user(&trace);
1034 ring_buffer_unlock_commit(tr->buffer, event, irq_flags); 1048 ring_buffer_unlock_commit(tr->buffer, event);
1035#endif 1049#endif
1036} 1050}
1037 1051
1038void __trace_userstack(struct trace_array *tr, 1052#ifdef UNUSED
1039 struct trace_array_cpu *data, 1053static void __trace_userstack(struct trace_array *tr, unsigned long flags)
1040 unsigned long flags)
1041{ 1054{
1042 ftrace_trace_userstack(tr, data, flags, preempt_count()); 1055 ftrace_trace_userstack(tr, flags, preempt_count());
1043} 1056}
1057#endif /* UNUSED */
1044 1058
1045static void 1059static void
1046ftrace_trace_special(void *__tr, void *__data, 1060ftrace_trace_special(void *__tr,
1047 unsigned long arg1, unsigned long arg2, unsigned long arg3, 1061 unsigned long arg1, unsigned long arg2, unsigned long arg3,
1048 int pc) 1062 int pc)
1049{ 1063{
1050 struct ring_buffer_event *event; 1064 struct ring_buffer_event *event;
1051 struct trace_array_cpu *data = __data;
1052 struct trace_array *tr = __tr; 1065 struct trace_array *tr = __tr;
1053 struct special_entry *entry; 1066 struct special_entry *entry;
1054 unsigned long irq_flags;
1055 1067
1056 event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry), 1068 event = trace_buffer_lock_reserve(tr, TRACE_SPECIAL,
1057 &irq_flags); 1069 sizeof(*entry), 0, pc);
1058 if (!event) 1070 if (!event)
1059 return; 1071 return;
1060 entry = ring_buffer_event_data(event); 1072 entry = ring_buffer_event_data(event);
1061 tracing_generic_entry_update(&entry->ent, 0, pc);
1062 entry->ent.type = TRACE_SPECIAL;
1063 entry->arg1 = arg1; 1073 entry->arg1 = arg1;
1064 entry->arg2 = arg2; 1074 entry->arg2 = arg2;
1065 entry->arg3 = arg3; 1075 entry->arg3 = arg3;
1066 ring_buffer_unlock_commit(tr->buffer, event, irq_flags); 1076 trace_buffer_unlock_commit(tr, event, 0, pc);
1067 ftrace_trace_stack(tr, data, irq_flags, 4, pc);
1068 ftrace_trace_userstack(tr, data, irq_flags, pc);
1069
1070 trace_wake_up();
1071} 1077}
1072 1078
1073void 1079void
1074__trace_special(void *__tr, void *__data, 1080__trace_special(void *__tr, void *__data,
1075 unsigned long arg1, unsigned long arg2, unsigned long arg3) 1081 unsigned long arg1, unsigned long arg2, unsigned long arg3)
1076{ 1082{
1077 ftrace_trace_special(__tr, __data, arg1, arg2, arg3, preempt_count()); 1083 ftrace_trace_special(__tr, arg1, arg2, arg3, preempt_count());
1078} 1084}
1079 1085
1080void 1086void
1081tracing_sched_switch_trace(struct trace_array *tr, 1087tracing_sched_switch_trace(struct trace_array *tr,
1082 struct trace_array_cpu *data,
1083 struct task_struct *prev, 1088 struct task_struct *prev,
1084 struct task_struct *next, 1089 struct task_struct *next,
1085 unsigned long flags, int pc) 1090 unsigned long flags, int pc)
1086{ 1091{
1087 struct ring_buffer_event *event; 1092 struct ring_buffer_event *event;
1088 struct ctx_switch_entry *entry; 1093 struct ctx_switch_entry *entry;
1089 unsigned long irq_flags;
1090 1094
1091 event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry), 1095 event = trace_buffer_lock_reserve(tr, TRACE_CTX,
1092 &irq_flags); 1096 sizeof(*entry), flags, pc);
1093 if (!event) 1097 if (!event)
1094 return; 1098 return;
1095 entry = ring_buffer_event_data(event); 1099 entry = ring_buffer_event_data(event);
1096 tracing_generic_entry_update(&entry->ent, flags, pc);
1097 entry->ent.type = TRACE_CTX;
1098 entry->prev_pid = prev->pid; 1100 entry->prev_pid = prev->pid;
1099 entry->prev_prio = prev->prio; 1101 entry->prev_prio = prev->prio;
1100 entry->prev_state = prev->state; 1102 entry->prev_state = prev->state;
@@ -1102,29 +1104,23 @@ tracing_sched_switch_trace(struct trace_array *tr,
1102 entry->next_prio = next->prio; 1104 entry->next_prio = next->prio;
1103 entry->next_state = next->state; 1105 entry->next_state = next->state;
1104 entry->next_cpu = task_cpu(next); 1106 entry->next_cpu = task_cpu(next);
1105 ring_buffer_unlock_commit(tr->buffer, event, irq_flags); 1107 trace_buffer_unlock_commit(tr, event, flags, pc);
1106 ftrace_trace_stack(tr, data, flags, 5, pc);
1107 ftrace_trace_userstack(tr, data, flags, pc);
1108} 1108}
1109 1109
1110void 1110void
1111tracing_sched_wakeup_trace(struct trace_array *tr, 1111tracing_sched_wakeup_trace(struct trace_array *tr,
1112 struct trace_array_cpu *data,
1113 struct task_struct *wakee, 1112 struct task_struct *wakee,
1114 struct task_struct *curr, 1113 struct task_struct *curr,
1115 unsigned long flags, int pc) 1114 unsigned long flags, int pc)
1116{ 1115{
1117 struct ring_buffer_event *event; 1116 struct ring_buffer_event *event;
1118 struct ctx_switch_entry *entry; 1117 struct ctx_switch_entry *entry;
1119 unsigned long irq_flags;
1120 1118
1121 event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry), 1119 event = trace_buffer_lock_reserve(tr, TRACE_WAKE,
1122 &irq_flags); 1120 sizeof(*entry), flags, pc);
1123 if (!event) 1121 if (!event)
1124 return; 1122 return;
1125 entry = ring_buffer_event_data(event); 1123 entry = ring_buffer_event_data(event);
1126 tracing_generic_entry_update(&entry->ent, flags, pc);
1127 entry->ent.type = TRACE_WAKE;
1128 entry->prev_pid = curr->pid; 1124 entry->prev_pid = curr->pid;
1129 entry->prev_prio = curr->prio; 1125 entry->prev_prio = curr->prio;
1130 entry->prev_state = curr->state; 1126 entry->prev_state = curr->state;
@@ -1132,11 +1128,10 @@ tracing_sched_wakeup_trace(struct trace_array *tr,
1132 entry->next_prio = wakee->prio; 1128 entry->next_prio = wakee->prio;
1133 entry->next_state = wakee->state; 1129 entry->next_state = wakee->state;
1134 entry->next_cpu = task_cpu(wakee); 1130 entry->next_cpu = task_cpu(wakee);
1135 ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
1136 ftrace_trace_stack(tr, data, flags, 6, pc);
1137 ftrace_trace_userstack(tr, data, flags, pc);
1138 1131
1139 trace_wake_up(); 1132 ring_buffer_unlock_commit(tr->buffer, event);
1133 ftrace_trace_stack(tr, flags, 6, pc);
1134 ftrace_trace_userstack(tr, flags, pc);
1140} 1135}
1141 1136
1142void 1137void
@@ -1157,66 +1152,7 @@ ftrace_special(unsigned long arg1, unsigned long arg2, unsigned long arg3)
1157 data = tr->data[cpu]; 1152 data = tr->data[cpu];
1158 1153
1159 if (likely(atomic_inc_return(&data->disabled) == 1)) 1154 if (likely(atomic_inc_return(&data->disabled) == 1))
1160 ftrace_trace_special(tr, data, arg1, arg2, arg3, pc); 1155 ftrace_trace_special(tr, arg1, arg2, arg3, pc);
1161
1162 atomic_dec(&data->disabled);
1163 local_irq_restore(flags);
1164}
1165
1166#ifdef CONFIG_FUNCTION_TRACER
1167static void
1168function_trace_call_preempt_only(unsigned long ip, unsigned long parent_ip)
1169{
1170 struct trace_array *tr = &global_trace;
1171 struct trace_array_cpu *data;
1172 unsigned long flags;
1173 long disabled;
1174 int cpu, resched;
1175 int pc;
1176
1177 if (unlikely(!ftrace_function_enabled))
1178 return;
1179
1180 pc = preempt_count();
1181 resched = ftrace_preempt_disable();
1182 local_save_flags(flags);
1183 cpu = raw_smp_processor_id();
1184 data = tr->data[cpu];
1185 disabled = atomic_inc_return(&data->disabled);
1186
1187 if (likely(disabled == 1))
1188 trace_function(tr, data, ip, parent_ip, flags, pc);
1189
1190 atomic_dec(&data->disabled);
1191 ftrace_preempt_enable(resched);
1192}
1193
1194static void
1195function_trace_call(unsigned long ip, unsigned long parent_ip)
1196{
1197 struct trace_array *tr = &global_trace;
1198 struct trace_array_cpu *data;
1199 unsigned long flags;
1200 long disabled;
1201 int cpu;
1202 int pc;
1203
1204 if (unlikely(!ftrace_function_enabled))
1205 return;
1206
1207 /*
1208 * Need to use raw, since this must be called before the
1209 * recursive protection is performed.
1210 */
1211 local_irq_save(flags);
1212 cpu = raw_smp_processor_id();
1213 data = tr->data[cpu];
1214 disabled = atomic_inc_return(&data->disabled);
1215
1216 if (likely(disabled == 1)) {
1217 pc = preempt_count();
1218 trace_function(tr, data, ip, parent_ip, flags, pc);
1219 }
1220 1156
1221 atomic_dec(&data->disabled); 1157 atomic_dec(&data->disabled);
1222 local_irq_restore(flags); 1158 local_irq_restore(flags);
@@ -1229,6 +1165,7 @@ int trace_graph_entry(struct ftrace_graph_ent *trace)
1229 struct trace_array_cpu *data; 1165 struct trace_array_cpu *data;
1230 unsigned long flags; 1166 unsigned long flags;
1231 long disabled; 1167 long disabled;
1168 int ret;
1232 int cpu; 1169 int cpu;
1233 int pc; 1170 int pc;
1234 1171
@@ -1244,15 +1181,18 @@ int trace_graph_entry(struct ftrace_graph_ent *trace)
1244 disabled = atomic_inc_return(&data->disabled); 1181 disabled = atomic_inc_return(&data->disabled);
1245 if (likely(disabled == 1)) { 1182 if (likely(disabled == 1)) {
1246 pc = preempt_count(); 1183 pc = preempt_count();
1247 __trace_graph_entry(tr, data, trace, flags, pc); 1184 ret = __trace_graph_entry(tr, trace, flags, pc);
1185 } else {
1186 ret = 0;
1248 } 1187 }
1249 /* Only do the atomic if it is not already set */ 1188 /* Only do the atomic if it is not already set */
1250 if (!test_tsk_trace_graph(current)) 1189 if (!test_tsk_trace_graph(current))
1251 set_tsk_trace_graph(current); 1190 set_tsk_trace_graph(current);
1191
1252 atomic_dec(&data->disabled); 1192 atomic_dec(&data->disabled);
1253 local_irq_restore(flags); 1193 local_irq_restore(flags);
1254 1194
1255 return 1; 1195 return ret;
1256} 1196}
1257 1197
1258void trace_graph_return(struct ftrace_graph_ret *trace) 1198void trace_graph_return(struct ftrace_graph_ret *trace)
@@ -1270,7 +1210,7 @@ void trace_graph_return(struct ftrace_graph_ret *trace)
1270 disabled = atomic_inc_return(&data->disabled); 1210 disabled = atomic_inc_return(&data->disabled);
1271 if (likely(disabled == 1)) { 1211 if (likely(disabled == 1)) {
1272 pc = preempt_count(); 1212 pc = preempt_count();
1273 __trace_graph_return(tr, data, trace, flags, pc); 1213 __trace_graph_return(tr, trace, flags, pc);
1274 } 1214 }
1275 if (!trace->depth) 1215 if (!trace->depth)
1276 clear_tsk_trace_graph(current); 1216 clear_tsk_trace_graph(current);
@@ -1279,30 +1219,122 @@ void trace_graph_return(struct ftrace_graph_ret *trace)
1279} 1219}
1280#endif /* CONFIG_FUNCTION_GRAPH_TRACER */ 1220#endif /* CONFIG_FUNCTION_GRAPH_TRACER */
1281 1221
1282static struct ftrace_ops trace_ops __read_mostly =
1283{
1284 .func = function_trace_call,
1285};
1286 1222
1287void tracing_start_function_trace(void) 1223/**
1224 * trace_vbprintk - write binary msg to tracing buffer
1225 *
1226 */
1227int trace_vbprintk(unsigned long ip, const char *fmt, va_list args)
1288{ 1228{
1289 ftrace_function_enabled = 0; 1229 static raw_spinlock_t trace_buf_lock =
1230 (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
1231 static u32 trace_buf[TRACE_BUF_SIZE];
1290 1232
1291 if (trace_flags & TRACE_ITER_PREEMPTONLY) 1233 struct ring_buffer_event *event;
1292 trace_ops.func = function_trace_call_preempt_only; 1234 struct trace_array *tr = &global_trace;
1293 else 1235 struct trace_array_cpu *data;
1294 trace_ops.func = function_trace_call; 1236 struct bprint_entry *entry;
1237 unsigned long flags;
1238 int resched;
1239 int cpu, len = 0, size, pc;
1240
1241 if (unlikely(tracing_selftest_running || tracing_disabled))
1242 return 0;
1243
1244 /* Don't pollute graph traces with trace_vprintk internals */
1245 pause_graph_tracing();
1246
1247 pc = preempt_count();
1248 resched = ftrace_preempt_disable();
1249 cpu = raw_smp_processor_id();
1250 data = tr->data[cpu];
1251
1252 if (unlikely(atomic_read(&data->disabled)))
1253 goto out;
1254
1255 /* Lockdep uses trace_printk for lock tracing */
1256 local_irq_save(flags);
1257 __raw_spin_lock(&trace_buf_lock);
1258 len = vbin_printf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1259
1260 if (len > TRACE_BUF_SIZE || len < 0)
1261 goto out_unlock;
1262
1263 size = sizeof(*entry) + sizeof(u32) * len;
1264 event = trace_buffer_lock_reserve(tr, TRACE_BPRINT, size, flags, pc);
1265 if (!event)
1266 goto out_unlock;
1267 entry = ring_buffer_event_data(event);
1268 entry->ip = ip;
1269 entry->fmt = fmt;
1270
1271 memcpy(entry->buf, trace_buf, sizeof(u32) * len);
1272 ring_buffer_unlock_commit(tr->buffer, event);
1273
1274out_unlock:
1275 __raw_spin_unlock(&trace_buf_lock);
1276 local_irq_restore(flags);
1277
1278out:
1279 ftrace_preempt_enable(resched);
1280 unpause_graph_tracing();
1295 1281
1296 register_ftrace_function(&trace_ops); 1282 return len;
1297 ftrace_function_enabled = 1;
1298} 1283}
1284EXPORT_SYMBOL_GPL(trace_vbprintk);
1299 1285
1300void tracing_stop_function_trace(void) 1286int trace_vprintk(unsigned long ip, const char *fmt, va_list args)
1301{ 1287{
1302 ftrace_function_enabled = 0; 1288 static raw_spinlock_t trace_buf_lock = __RAW_SPIN_LOCK_UNLOCKED;
1303 unregister_ftrace_function(&trace_ops); 1289 static char trace_buf[TRACE_BUF_SIZE];
1290
1291 struct ring_buffer_event *event;
1292 struct trace_array *tr = &global_trace;
1293 struct trace_array_cpu *data;
1294 int cpu, len = 0, size, pc;
1295 struct print_entry *entry;
1296 unsigned long irq_flags;
1297
1298 if (tracing_disabled || tracing_selftest_running)
1299 return 0;
1300
1301 pc = preempt_count();
1302 preempt_disable_notrace();
1303 cpu = raw_smp_processor_id();
1304 data = tr->data[cpu];
1305
1306 if (unlikely(atomic_read(&data->disabled)))
1307 goto out;
1308
1309 pause_graph_tracing();
1310 raw_local_irq_save(irq_flags);
1311 __raw_spin_lock(&trace_buf_lock);
1312 len = vsnprintf(trace_buf, TRACE_BUF_SIZE, fmt, args);
1313
1314 len = min(len, TRACE_BUF_SIZE-1);
1315 trace_buf[len] = 0;
1316
1317 size = sizeof(*entry) + len + 1;
1318 event = trace_buffer_lock_reserve(tr, TRACE_PRINT, size, irq_flags, pc);
1319 if (!event)
1320 goto out_unlock;
1321 entry = ring_buffer_event_data(event);
1322 entry->ip = ip;
1323
1324 memcpy(&entry->buf, trace_buf, len);
1325 entry->buf[len] = 0;
1326 ring_buffer_unlock_commit(tr->buffer, event);
1327
1328 out_unlock:
1329 __raw_spin_unlock(&trace_buf_lock);
1330 raw_local_irq_restore(irq_flags);
1331 unpause_graph_tracing();
1332 out:
1333 preempt_enable_notrace();
1334
1335 return len;
1304} 1336}
1305#endif 1337EXPORT_SYMBOL_GPL(trace_vprintk);
1306 1338
1307enum trace_file_type { 1339enum trace_file_type {
1308 TRACE_FILE_LAT_FMT = 1, 1340 TRACE_FILE_LAT_FMT = 1,
@@ -1345,10 +1377,25 @@ __find_next_entry(struct trace_iterator *iter, int *ent_cpu, u64 *ent_ts)
1345{ 1377{
1346 struct ring_buffer *buffer = iter->tr->buffer; 1378 struct ring_buffer *buffer = iter->tr->buffer;
1347 struct trace_entry *ent, *next = NULL; 1379 struct trace_entry *ent, *next = NULL;
1380 int cpu_file = iter->cpu_file;
1348 u64 next_ts = 0, ts; 1381 u64 next_ts = 0, ts;
1349 int next_cpu = -1; 1382 int next_cpu = -1;
1350 int cpu; 1383 int cpu;
1351 1384
1385 /*
1386 * If we are in a per_cpu trace file, don't bother by iterating over
1387 * all cpu and peek directly.
1388 */
1389 if (cpu_file > TRACE_PIPE_ALL_CPU) {
1390 if (ring_buffer_empty_cpu(buffer, cpu_file))
1391 return NULL;
1392 ent = peek_next_entry(iter, cpu_file, ent_ts);
1393 if (ent_cpu)
1394 *ent_cpu = cpu_file;
1395
1396 return ent;
1397 }
1398
1352 for_each_tracing_cpu(cpu) { 1399 for_each_tracing_cpu(cpu) {
1353 1400
1354 if (ring_buffer_empty_cpu(buffer, cpu)) 1401 if (ring_buffer_empty_cpu(buffer, cpu))
@@ -1376,8 +1423,8 @@ __find_next_entry(struct trace_iterator *iter, int *ent_cpu, u64 *ent_ts)
1376} 1423}
1377 1424
1378/* Find the next real entry, without updating the iterator itself */ 1425/* Find the next real entry, without updating the iterator itself */
1379static struct trace_entry * 1426struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
1380find_next_entry(struct trace_iterator *iter, int *ent_cpu, u64 *ent_ts) 1427 int *ent_cpu, u64 *ent_ts)
1381{ 1428{
1382 return __find_next_entry(iter, ent_cpu, ent_ts); 1429 return __find_next_entry(iter, ent_cpu, ent_ts);
1383} 1430}
@@ -1426,19 +1473,32 @@ static void *s_next(struct seq_file *m, void *v, loff_t *pos)
1426 return ent; 1473 return ent;
1427} 1474}
1428 1475
1476/*
1477 * No necessary locking here. The worst thing which can
1478 * happen is loosing events consumed at the same time
1479 * by a trace_pipe reader.
1480 * Other than that, we don't risk to crash the ring buffer
1481 * because it serializes the readers.
1482 *
1483 * The current tracer is copied to avoid a global locking
1484 * all around.
1485 */
1429static void *s_start(struct seq_file *m, loff_t *pos) 1486static void *s_start(struct seq_file *m, loff_t *pos)
1430{ 1487{
1431 struct trace_iterator *iter = m->private; 1488 struct trace_iterator *iter = m->private;
1489 static struct tracer *old_tracer;
1490 int cpu_file = iter->cpu_file;
1432 void *p = NULL; 1491 void *p = NULL;
1433 loff_t l = 0; 1492 loff_t l = 0;
1434 int cpu; 1493 int cpu;
1435 1494
1495 /* copy the tracer to avoid using a global lock all around */
1436 mutex_lock(&trace_types_lock); 1496 mutex_lock(&trace_types_lock);
1437 1497 if (unlikely(old_tracer != current_trace && current_trace)) {
1438 if (!current_trace || current_trace != iter->trace) { 1498 old_tracer = current_trace;
1439 mutex_unlock(&trace_types_lock); 1499 *iter->trace = *current_trace;
1440 return NULL;
1441 } 1500 }
1501 mutex_unlock(&trace_types_lock);
1442 1502
1443 atomic_inc(&trace_record_cmdline_disabled); 1503 atomic_inc(&trace_record_cmdline_disabled);
1444 1504
@@ -1449,9 +1509,12 @@ static void *s_start(struct seq_file *m, loff_t *pos)
1449 1509
1450 ftrace_disable_cpu(); 1510 ftrace_disable_cpu();
1451 1511
1452 for_each_tracing_cpu(cpu) { 1512 if (cpu_file == TRACE_PIPE_ALL_CPU) {
1453 ring_buffer_iter_reset(iter->buffer_iter[cpu]); 1513 for_each_tracing_cpu(cpu)
1454 } 1514 ring_buffer_iter_reset(iter->buffer_iter[cpu]);
1515 } else
1516 ring_buffer_iter_reset(iter->buffer_iter[cpu_file]);
1517
1455 1518
1456 ftrace_enable_cpu(); 1519 ftrace_enable_cpu();
1457 1520
@@ -1469,155 +1532,6 @@ static void *s_start(struct seq_file *m, loff_t *pos)
1469static void s_stop(struct seq_file *m, void *p) 1532static void s_stop(struct seq_file *m, void *p)
1470{ 1533{
1471 atomic_dec(&trace_record_cmdline_disabled); 1534 atomic_dec(&trace_record_cmdline_disabled);
1472 mutex_unlock(&trace_types_lock);
1473}
1474
1475#ifdef CONFIG_KRETPROBES
1476static inline const char *kretprobed(const char *name)
1477{
1478 static const char tramp_name[] = "kretprobe_trampoline";
1479 int size = sizeof(tramp_name);
1480
1481 if (strncmp(tramp_name, name, size) == 0)
1482 return "[unknown/kretprobe'd]";
1483 return name;
1484}
1485#else
1486static inline const char *kretprobed(const char *name)
1487{
1488 return name;
1489}
1490#endif /* CONFIG_KRETPROBES */
1491
1492static int
1493seq_print_sym_short(struct trace_seq *s, const char *fmt, unsigned long address)
1494{
1495#ifdef CONFIG_KALLSYMS
1496 char str[KSYM_SYMBOL_LEN];
1497 const char *name;
1498
1499 kallsyms_lookup(address, NULL, NULL, NULL, str);
1500
1501 name = kretprobed(str);
1502
1503 return trace_seq_printf(s, fmt, name);
1504#endif
1505 return 1;
1506}
1507
1508static int
1509seq_print_sym_offset(struct trace_seq *s, const char *fmt,
1510 unsigned long address)
1511{
1512#ifdef CONFIG_KALLSYMS
1513 char str[KSYM_SYMBOL_LEN];
1514 const char *name;
1515
1516 sprint_symbol(str, address);
1517 name = kretprobed(str);
1518
1519 return trace_seq_printf(s, fmt, name);
1520#endif
1521 return 1;
1522}
1523
1524#ifndef CONFIG_64BIT
1525# define IP_FMT "%08lx"
1526#else
1527# define IP_FMT "%016lx"
1528#endif
1529
1530int
1531seq_print_ip_sym(struct trace_seq *s, unsigned long ip, unsigned long sym_flags)
1532{
1533 int ret;
1534
1535 if (!ip)
1536 return trace_seq_printf(s, "0");
1537
1538 if (sym_flags & TRACE_ITER_SYM_OFFSET)
1539 ret = seq_print_sym_offset(s, "%s", ip);
1540 else
1541 ret = seq_print_sym_short(s, "%s", ip);
1542
1543 if (!ret)
1544 return 0;
1545
1546 if (sym_flags & TRACE_ITER_SYM_ADDR)
1547 ret = trace_seq_printf(s, " <" IP_FMT ">", ip);
1548 return ret;
1549}
1550
1551static inline int seq_print_user_ip(struct trace_seq *s, struct mm_struct *mm,
1552 unsigned long ip, unsigned long sym_flags)
1553{
1554 struct file *file = NULL;
1555 unsigned long vmstart = 0;
1556 int ret = 1;
1557
1558 if (mm) {
1559 const struct vm_area_struct *vma;
1560
1561 down_read(&mm->mmap_sem);
1562 vma = find_vma(mm, ip);
1563 if (vma) {
1564 file = vma->vm_file;
1565 vmstart = vma->vm_start;
1566 }
1567 if (file) {
1568 ret = trace_seq_path(s, &file->f_path);
1569 if (ret)
1570 ret = trace_seq_printf(s, "[+0x%lx]", ip - vmstart);
1571 }
1572 up_read(&mm->mmap_sem);
1573 }
1574 if (ret && ((sym_flags & TRACE_ITER_SYM_ADDR) || !file))
1575 ret = trace_seq_printf(s, " <" IP_FMT ">", ip);
1576 return ret;
1577}
1578
1579static int
1580seq_print_userip_objs(const struct userstack_entry *entry, struct trace_seq *s,
1581 unsigned long sym_flags)
1582{
1583 struct mm_struct *mm = NULL;
1584 int ret = 1;
1585 unsigned int i;
1586
1587 if (trace_flags & TRACE_ITER_SYM_USEROBJ) {
1588 struct task_struct *task;
1589 /*
1590 * we do the lookup on the thread group leader,
1591 * since individual threads might have already quit!
1592 */
1593 rcu_read_lock();
1594 task = find_task_by_vpid(entry->ent.tgid);
1595 if (task)
1596 mm = get_task_mm(task);
1597 rcu_read_unlock();
1598 }
1599
1600 for (i = 0; i < FTRACE_STACK_ENTRIES; i++) {
1601 unsigned long ip = entry->caller[i];
1602
1603 if (ip == ULONG_MAX || !ret)
1604 break;
1605 if (i && ret)
1606 ret = trace_seq_puts(s, " <- ");
1607 if (!ip) {
1608 if (ret)
1609 ret = trace_seq_puts(s, "??");
1610 continue;
1611 }
1612 if (!ret)
1613 break;
1614 if (ret)
1615 ret = seq_print_user_ip(s, mm, ip, sym_flags);
1616 }
1617
1618 if (mm)
1619 mmput(mm);
1620 return ret;
1621} 1535}
1622 1536
1623static void print_lat_help_header(struct seq_file *m) 1537static void print_lat_help_header(struct seq_file *m)
@@ -1658,11 +1572,11 @@ print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1658 total = entries + 1572 total = entries +
1659 ring_buffer_overruns(iter->tr->buffer); 1573 ring_buffer_overruns(iter->tr->buffer);
1660 1574
1661 seq_printf(m, "%s latency trace v1.1.5 on %s\n", 1575 seq_printf(m, "# %s latency trace v1.1.5 on %s\n",
1662 name, UTS_RELEASE); 1576 name, UTS_RELEASE);
1663 seq_puts(m, "-----------------------------------" 1577 seq_puts(m, "# -----------------------------------"
1664 "---------------------------------\n"); 1578 "---------------------------------\n");
1665 seq_printf(m, " latency: %lu us, #%lu/%lu, CPU#%d |" 1579 seq_printf(m, "# latency: %lu us, #%lu/%lu, CPU#%d |"
1666 " (M:%s VP:%d, KP:%d, SP:%d HP:%d", 1580 " (M:%s VP:%d, KP:%d, SP:%d HP:%d",
1667 nsecs_to_usecs(data->saved_latency), 1581 nsecs_to_usecs(data->saved_latency),
1668 entries, 1582 entries,
@@ -1684,121 +1598,24 @@ print_trace_header(struct seq_file *m, struct trace_iterator *iter)
1684#else 1598#else
1685 seq_puts(m, ")\n"); 1599 seq_puts(m, ")\n");
1686#endif 1600#endif
1687 seq_puts(m, " -----------------\n"); 1601 seq_puts(m, "# -----------------\n");
1688 seq_printf(m, " | task: %.16s-%d " 1602 seq_printf(m, "# | task: %.16s-%d "
1689 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n", 1603 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
1690 data->comm, data->pid, data->uid, data->nice, 1604 data->comm, data->pid, data->uid, data->nice,
1691 data->policy, data->rt_priority); 1605 data->policy, data->rt_priority);
1692 seq_puts(m, " -----------------\n"); 1606 seq_puts(m, "# -----------------\n");
1693 1607
1694 if (data->critical_start) { 1608 if (data->critical_start) {
1695 seq_puts(m, " => started at: "); 1609 seq_puts(m, "# => started at: ");
1696 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags); 1610 seq_print_ip_sym(&iter->seq, data->critical_start, sym_flags);
1697 trace_print_seq(m, &iter->seq); 1611 trace_print_seq(m, &iter->seq);
1698 seq_puts(m, "\n => ended at: "); 1612 seq_puts(m, "\n# => ended at: ");
1699 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags); 1613 seq_print_ip_sym(&iter->seq, data->critical_end, sym_flags);
1700 trace_print_seq(m, &iter->seq); 1614 trace_print_seq(m, &iter->seq);
1701 seq_puts(m, "\n"); 1615 seq_puts(m, "#\n");
1702 }
1703
1704 seq_puts(m, "\n");
1705}
1706
1707static void
1708lat_print_generic(struct trace_seq *s, struct trace_entry *entry, int cpu)
1709{
1710 int hardirq, softirq;
1711 char *comm;
1712
1713 comm = trace_find_cmdline(entry->pid);
1714
1715 trace_seq_printf(s, "%8.8s-%-5d ", comm, entry->pid);
1716 trace_seq_printf(s, "%3d", cpu);
1717 trace_seq_printf(s, "%c%c",
1718 (entry->flags & TRACE_FLAG_IRQS_OFF) ? 'd' :
1719 (entry->flags & TRACE_FLAG_IRQS_NOSUPPORT) ? 'X' : '.',
1720 ((entry->flags & TRACE_FLAG_NEED_RESCHED) ? 'N' : '.'));
1721
1722 hardirq = entry->flags & TRACE_FLAG_HARDIRQ;
1723 softirq = entry->flags & TRACE_FLAG_SOFTIRQ;
1724 if (hardirq && softirq) {
1725 trace_seq_putc(s, 'H');
1726 } else {
1727 if (hardirq) {
1728 trace_seq_putc(s, 'h');
1729 } else {
1730 if (softirq)
1731 trace_seq_putc(s, 's');
1732 else
1733 trace_seq_putc(s, '.');
1734 }
1735 }
1736
1737 if (entry->preempt_count)
1738 trace_seq_printf(s, "%x", entry->preempt_count);
1739 else
1740 trace_seq_puts(s, ".");
1741}
1742
1743unsigned long preempt_mark_thresh = 100;
1744
1745static void
1746lat_print_timestamp(struct trace_seq *s, u64 abs_usecs,
1747 unsigned long rel_usecs)
1748{
1749 trace_seq_printf(s, " %4lldus", abs_usecs);
1750 if (rel_usecs > preempt_mark_thresh)
1751 trace_seq_puts(s, "!: ");
1752 else if (rel_usecs > 1)
1753 trace_seq_puts(s, "+: ");
1754 else
1755 trace_seq_puts(s, " : ");
1756}
1757
1758static const char state_to_char[] = TASK_STATE_TO_CHAR_STR;
1759
1760static int task_state_char(unsigned long state)
1761{
1762 int bit = state ? __ffs(state) + 1 : 0;
1763
1764 return bit < sizeof(state_to_char) - 1 ? state_to_char[bit] : '?';
1765}
1766
1767/*
1768 * The message is supposed to contain an ending newline.
1769 * If the printing stops prematurely, try to add a newline of our own.
1770 */
1771void trace_seq_print_cont(struct trace_seq *s, struct trace_iterator *iter)
1772{
1773 struct trace_entry *ent;
1774 struct trace_field_cont *cont;
1775 bool ok = true;
1776
1777 ent = peek_next_entry(iter, iter->cpu, NULL);
1778 if (!ent || ent->type != TRACE_CONT) {
1779 trace_seq_putc(s, '\n');
1780 return;
1781 } 1616 }
1782 1617
1783 do { 1618 seq_puts(m, "#\n");
1784 cont = (struct trace_field_cont *)ent;
1785 if (ok)
1786 ok = (trace_seq_printf(s, "%s", cont->buf) > 0);
1787
1788 ftrace_disable_cpu();
1789
1790 if (iter->buffer_iter[iter->cpu])
1791 ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
1792 else
1793 ring_buffer_consume(iter->tr->buffer, iter->cpu, NULL);
1794
1795 ftrace_enable_cpu();
1796
1797 ent = peek_next_entry(iter, iter->cpu, NULL);
1798 } while (ent && ent->type == TRACE_CONT);
1799
1800 if (!ok)
1801 trace_seq_putc(s, '\n');
1802} 1619}
1803 1620
1804static void test_cpu_buff_start(struct trace_iterator *iter) 1621static void test_cpu_buff_start(struct trace_iterator *iter)
@@ -1818,472 +1635,89 @@ static void test_cpu_buff_start(struct trace_iterator *iter)
1818 trace_seq_printf(s, "##### CPU %u buffer started ####\n", iter->cpu); 1635 trace_seq_printf(s, "##### CPU %u buffer started ####\n", iter->cpu);
1819} 1636}
1820 1637
1821static enum print_line_t
1822print_lat_fmt(struct trace_iterator *iter, unsigned int trace_idx, int cpu)
1823{
1824 struct trace_seq *s = &iter->seq;
1825 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1826 struct trace_entry *next_entry;
1827 unsigned long verbose = (trace_flags & TRACE_ITER_VERBOSE);
1828 struct trace_entry *entry = iter->ent;
1829 unsigned long abs_usecs;
1830 unsigned long rel_usecs;
1831 u64 next_ts;
1832 char *comm;
1833 int S, T;
1834 int i;
1835
1836 if (entry->type == TRACE_CONT)
1837 return TRACE_TYPE_HANDLED;
1838
1839 test_cpu_buff_start(iter);
1840
1841 next_entry = find_next_entry(iter, NULL, &next_ts);
1842 if (!next_entry)
1843 next_ts = iter->ts;
1844 rel_usecs = ns2usecs(next_ts - iter->ts);
1845 abs_usecs = ns2usecs(iter->ts - iter->tr->time_start);
1846
1847 if (verbose) {
1848 comm = trace_find_cmdline(entry->pid);
1849 trace_seq_printf(s, "%16s %5d %3d %d %08x %08x [%08lx]"
1850 " %ld.%03ldms (+%ld.%03ldms): ",
1851 comm,
1852 entry->pid, cpu, entry->flags,
1853 entry->preempt_count, trace_idx,
1854 ns2usecs(iter->ts),
1855 abs_usecs/1000,
1856 abs_usecs % 1000, rel_usecs/1000,
1857 rel_usecs % 1000);
1858 } else {
1859 lat_print_generic(s, entry, cpu);
1860 lat_print_timestamp(s, abs_usecs, rel_usecs);
1861 }
1862 switch (entry->type) {
1863 case TRACE_FN: {
1864 struct ftrace_entry *field;
1865
1866 trace_assign_type(field, entry);
1867
1868 seq_print_ip_sym(s, field->ip, sym_flags);
1869 trace_seq_puts(s, " (");
1870 seq_print_ip_sym(s, field->parent_ip, sym_flags);
1871 trace_seq_puts(s, ")\n");
1872 break;
1873 }
1874 case TRACE_CTX:
1875 case TRACE_WAKE: {
1876 struct ctx_switch_entry *field;
1877
1878 trace_assign_type(field, entry);
1879
1880 T = task_state_char(field->next_state);
1881 S = task_state_char(field->prev_state);
1882 comm = trace_find_cmdline(field->next_pid);
1883 trace_seq_printf(s, " %5d:%3d:%c %s [%03d] %5d:%3d:%c %s\n",
1884 field->prev_pid,
1885 field->prev_prio,
1886 S, entry->type == TRACE_CTX ? "==>" : " +",
1887 field->next_cpu,
1888 field->next_pid,
1889 field->next_prio,
1890 T, comm);
1891 break;
1892 }
1893 case TRACE_SPECIAL: {
1894 struct special_entry *field;
1895
1896 trace_assign_type(field, entry);
1897
1898 trace_seq_printf(s, "# %ld %ld %ld\n",
1899 field->arg1,
1900 field->arg2,
1901 field->arg3);
1902 break;
1903 }
1904 case TRACE_STACK: {
1905 struct stack_entry *field;
1906
1907 trace_assign_type(field, entry);
1908
1909 for (i = 0; i < FTRACE_STACK_ENTRIES; i++) {
1910 if (i)
1911 trace_seq_puts(s, " <= ");
1912 seq_print_ip_sym(s, field->caller[i], sym_flags);
1913 }
1914 trace_seq_puts(s, "\n");
1915 break;
1916 }
1917 case TRACE_PRINT: {
1918 struct print_entry *field;
1919
1920 trace_assign_type(field, entry);
1921
1922 seq_print_ip_sym(s, field->ip, sym_flags);
1923 trace_seq_printf(s, ": %s", field->buf);
1924 if (entry->flags & TRACE_FLAG_CONT)
1925 trace_seq_print_cont(s, iter);
1926 break;
1927 }
1928 case TRACE_BRANCH: {
1929 struct trace_branch *field;
1930
1931 trace_assign_type(field, entry);
1932
1933 trace_seq_printf(s, "[%s] %s:%s:%d\n",
1934 field->correct ? " ok " : " MISS ",
1935 field->func,
1936 field->file,
1937 field->line);
1938 break;
1939 }
1940 case TRACE_USER_STACK: {
1941 struct userstack_entry *field;
1942
1943 trace_assign_type(field, entry);
1944
1945 seq_print_userip_objs(field, s, sym_flags);
1946 trace_seq_putc(s, '\n');
1947 break;
1948 }
1949 default:
1950 trace_seq_printf(s, "Unknown type %d\n", entry->type);
1951 }
1952 return TRACE_TYPE_HANDLED;
1953}
1954
1955static enum print_line_t print_trace_fmt(struct trace_iterator *iter) 1638static enum print_line_t print_trace_fmt(struct trace_iterator *iter)
1956{ 1639{
1957 struct trace_seq *s = &iter->seq; 1640 struct trace_seq *s = &iter->seq;
1958 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK); 1641 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
1959 struct trace_entry *entry; 1642 struct trace_entry *entry;
1960 unsigned long usec_rem; 1643 struct trace_event *event;
1961 unsigned long long t;
1962 unsigned long secs;
1963 char *comm;
1964 int ret;
1965 int S, T;
1966 int i;
1967 1644
1968 entry = iter->ent; 1645 entry = iter->ent;
1969 1646
1970 if (entry->type == TRACE_CONT)
1971 return TRACE_TYPE_HANDLED;
1972
1973 test_cpu_buff_start(iter); 1647 test_cpu_buff_start(iter);
1974 1648
1975 comm = trace_find_cmdline(iter->ent->pid); 1649 event = ftrace_find_event(entry->type);
1976
1977 t = ns2usecs(iter->ts);
1978 usec_rem = do_div(t, 1000000ULL);
1979 secs = (unsigned long)t;
1980
1981 ret = trace_seq_printf(s, "%16s-%-5d ", comm, entry->pid);
1982 if (!ret)
1983 return TRACE_TYPE_PARTIAL_LINE;
1984 ret = trace_seq_printf(s, "[%03d] ", iter->cpu);
1985 if (!ret)
1986 return TRACE_TYPE_PARTIAL_LINE;
1987 ret = trace_seq_printf(s, "%5lu.%06lu: ", secs, usec_rem);
1988 if (!ret)
1989 return TRACE_TYPE_PARTIAL_LINE;
1990
1991 switch (entry->type) {
1992 case TRACE_FN: {
1993 struct ftrace_entry *field;
1994
1995 trace_assign_type(field, entry);
1996
1997 ret = seq_print_ip_sym(s, field->ip, sym_flags);
1998 if (!ret)
1999 return TRACE_TYPE_PARTIAL_LINE;
2000 if ((sym_flags & TRACE_ITER_PRINT_PARENT) &&
2001 field->parent_ip) {
2002 ret = trace_seq_printf(s, " <-");
2003 if (!ret)
2004 return TRACE_TYPE_PARTIAL_LINE;
2005 ret = seq_print_ip_sym(s,
2006 field->parent_ip,
2007 sym_flags);
2008 if (!ret)
2009 return TRACE_TYPE_PARTIAL_LINE;
2010 }
2011 ret = trace_seq_printf(s, "\n");
2012 if (!ret)
2013 return TRACE_TYPE_PARTIAL_LINE;
2014 break;
2015 }
2016 case TRACE_CTX:
2017 case TRACE_WAKE: {
2018 struct ctx_switch_entry *field;
2019
2020 trace_assign_type(field, entry);
2021
2022 T = task_state_char(field->next_state);
2023 S = task_state_char(field->prev_state);
2024 ret = trace_seq_printf(s, " %5d:%3d:%c %s [%03d] %5d:%3d:%c\n",
2025 field->prev_pid,
2026 field->prev_prio,
2027 S,
2028 entry->type == TRACE_CTX ? "==>" : " +",
2029 field->next_cpu,
2030 field->next_pid,
2031 field->next_prio,
2032 T);
2033 if (!ret)
2034 return TRACE_TYPE_PARTIAL_LINE;
2035 break;
2036 }
2037 case TRACE_SPECIAL: {
2038 struct special_entry *field;
2039
2040 trace_assign_type(field, entry);
2041 1650
2042 ret = trace_seq_printf(s, "# %ld %ld %ld\n", 1651 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2043 field->arg1, 1652 if (iter->iter_flags & TRACE_FILE_LAT_FMT) {
2044 field->arg2, 1653 if (!trace_print_lat_context(iter))
2045 field->arg3); 1654 goto partial;
2046 if (!ret) 1655 } else {
2047 return TRACE_TYPE_PARTIAL_LINE; 1656 if (!trace_print_context(iter))
2048 break; 1657 goto partial;
2049 }
2050 case TRACE_STACK: {
2051 struct stack_entry *field;
2052
2053 trace_assign_type(field, entry);
2054
2055 for (i = 0; i < FTRACE_STACK_ENTRIES; i++) {
2056 if (i) {
2057 ret = trace_seq_puts(s, " <= ");
2058 if (!ret)
2059 return TRACE_TYPE_PARTIAL_LINE;
2060 }
2061 ret = seq_print_ip_sym(s, field->caller[i],
2062 sym_flags);
2063 if (!ret)
2064 return TRACE_TYPE_PARTIAL_LINE;
2065 } 1658 }
2066 ret = trace_seq_puts(s, "\n");
2067 if (!ret)
2068 return TRACE_TYPE_PARTIAL_LINE;
2069 break;
2070 }
2071 case TRACE_PRINT: {
2072 struct print_entry *field;
2073
2074 trace_assign_type(field, entry);
2075
2076 seq_print_ip_sym(s, field->ip, sym_flags);
2077 trace_seq_printf(s, ": %s", field->buf);
2078 if (entry->flags & TRACE_FLAG_CONT)
2079 trace_seq_print_cont(s, iter);
2080 break;
2081 }
2082 case TRACE_GRAPH_RET: {
2083 return print_graph_function(iter);
2084 }
2085 case TRACE_GRAPH_ENT: {
2086 return print_graph_function(iter);
2087 } 1659 }
2088 case TRACE_BRANCH: {
2089 struct trace_branch *field;
2090 1660
2091 trace_assign_type(field, entry); 1661 if (event)
1662 return event->trace(iter, sym_flags);
2092 1663
2093 trace_seq_printf(s, "[%s] %s:%s:%d\n", 1664 if (!trace_seq_printf(s, "Unknown type %d\n", entry->type))
2094 field->correct ? " ok " : " MISS ", 1665 goto partial;
2095 field->func,
2096 field->file,
2097 field->line);
2098 break;
2099 }
2100 case TRACE_USER_STACK: {
2101 struct userstack_entry *field;
2102
2103 trace_assign_type(field, entry);
2104 1666
2105 ret = seq_print_userip_objs(field, s, sym_flags);
2106 if (!ret)
2107 return TRACE_TYPE_PARTIAL_LINE;
2108 ret = trace_seq_putc(s, '\n');
2109 if (!ret)
2110 return TRACE_TYPE_PARTIAL_LINE;
2111 break;
2112 }
2113 }
2114 return TRACE_TYPE_HANDLED; 1667 return TRACE_TYPE_HANDLED;
1668partial:
1669 return TRACE_TYPE_PARTIAL_LINE;
2115} 1670}
2116 1671
2117static enum print_line_t print_raw_fmt(struct trace_iterator *iter) 1672static enum print_line_t print_raw_fmt(struct trace_iterator *iter)
2118{ 1673{
2119 struct trace_seq *s = &iter->seq; 1674 struct trace_seq *s = &iter->seq;
2120 struct trace_entry *entry; 1675 struct trace_entry *entry;
2121 int ret; 1676 struct trace_event *event;
2122 int S, T;
2123 1677
2124 entry = iter->ent; 1678 entry = iter->ent;
2125 1679
2126 if (entry->type == TRACE_CONT) 1680 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2127 return TRACE_TYPE_HANDLED; 1681 if (!trace_seq_printf(s, "%d %d %llu ",
2128 1682 entry->pid, iter->cpu, iter->ts))
2129 ret = trace_seq_printf(s, "%d %d %llu ", 1683 goto partial;
2130 entry->pid, iter->cpu, iter->ts);
2131 if (!ret)
2132 return TRACE_TYPE_PARTIAL_LINE;
2133
2134 switch (entry->type) {
2135 case TRACE_FN: {
2136 struct ftrace_entry *field;
2137
2138 trace_assign_type(field, entry);
2139
2140 ret = trace_seq_printf(s, "%x %x\n",
2141 field->ip,
2142 field->parent_ip);
2143 if (!ret)
2144 return TRACE_TYPE_PARTIAL_LINE;
2145 break;
2146 }
2147 case TRACE_CTX:
2148 case TRACE_WAKE: {
2149 struct ctx_switch_entry *field;
2150
2151 trace_assign_type(field, entry);
2152
2153 T = task_state_char(field->next_state);
2154 S = entry->type == TRACE_WAKE ? '+' :
2155 task_state_char(field->prev_state);
2156 ret = trace_seq_printf(s, "%d %d %c %d %d %d %c\n",
2157 field->prev_pid,
2158 field->prev_prio,
2159 S,
2160 field->next_cpu,
2161 field->next_pid,
2162 field->next_prio,
2163 T);
2164 if (!ret)
2165 return TRACE_TYPE_PARTIAL_LINE;
2166 break;
2167 } 1684 }
2168 case TRACE_SPECIAL:
2169 case TRACE_USER_STACK:
2170 case TRACE_STACK: {
2171 struct special_entry *field;
2172 1685
2173 trace_assign_type(field, entry); 1686 event = ftrace_find_event(entry->type);
1687 if (event)
1688 return event->raw(iter, 0);
2174 1689
2175 ret = trace_seq_printf(s, "# %ld %ld %ld\n", 1690 if (!trace_seq_printf(s, "%d ?\n", entry->type))
2176 field->arg1, 1691 goto partial;
2177 field->arg2,
2178 field->arg3);
2179 if (!ret)
2180 return TRACE_TYPE_PARTIAL_LINE;
2181 break;
2182 }
2183 case TRACE_PRINT: {
2184 struct print_entry *field;
2185
2186 trace_assign_type(field, entry);
2187 1692
2188 trace_seq_printf(s, "# %lx %s", field->ip, field->buf);
2189 if (entry->flags & TRACE_FLAG_CONT)
2190 trace_seq_print_cont(s, iter);
2191 break;
2192 }
2193 }
2194 return TRACE_TYPE_HANDLED; 1693 return TRACE_TYPE_HANDLED;
1694partial:
1695 return TRACE_TYPE_PARTIAL_LINE;
2195} 1696}
2196 1697
2197#define SEQ_PUT_FIELD_RET(s, x) \
2198do { \
2199 if (!trace_seq_putmem(s, &(x), sizeof(x))) \
2200 return 0; \
2201} while (0)
2202
2203#define SEQ_PUT_HEX_FIELD_RET(s, x) \
2204do { \
2205 BUILD_BUG_ON(sizeof(x) > MAX_MEMHEX_BYTES); \
2206 if (!trace_seq_putmem_hex(s, &(x), sizeof(x))) \
2207 return 0; \
2208} while (0)
2209
2210static enum print_line_t print_hex_fmt(struct trace_iterator *iter) 1698static enum print_line_t print_hex_fmt(struct trace_iterator *iter)
2211{ 1699{
2212 struct trace_seq *s = &iter->seq; 1700 struct trace_seq *s = &iter->seq;
2213 unsigned char newline = '\n'; 1701 unsigned char newline = '\n';
2214 struct trace_entry *entry; 1702 struct trace_entry *entry;
2215 int S, T; 1703 struct trace_event *event;
2216 1704
2217 entry = iter->ent; 1705 entry = iter->ent;
2218 1706
2219 if (entry->type == TRACE_CONT) 1707 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2220 return TRACE_TYPE_HANDLED; 1708 SEQ_PUT_HEX_FIELD_RET(s, entry->pid);
2221 1709 SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
2222 SEQ_PUT_HEX_FIELD_RET(s, entry->pid); 1710 SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
2223 SEQ_PUT_HEX_FIELD_RET(s, iter->cpu);
2224 SEQ_PUT_HEX_FIELD_RET(s, iter->ts);
2225
2226 switch (entry->type) {
2227 case TRACE_FN: {
2228 struct ftrace_entry *field;
2229
2230 trace_assign_type(field, entry);
2231
2232 SEQ_PUT_HEX_FIELD_RET(s, field->ip);
2233 SEQ_PUT_HEX_FIELD_RET(s, field->parent_ip);
2234 break;
2235 }
2236 case TRACE_CTX:
2237 case TRACE_WAKE: {
2238 struct ctx_switch_entry *field;
2239
2240 trace_assign_type(field, entry);
2241
2242 T = task_state_char(field->next_state);
2243 S = entry->type == TRACE_WAKE ? '+' :
2244 task_state_char(field->prev_state);
2245 SEQ_PUT_HEX_FIELD_RET(s, field->prev_pid);
2246 SEQ_PUT_HEX_FIELD_RET(s, field->prev_prio);
2247 SEQ_PUT_HEX_FIELD_RET(s, S);
2248 SEQ_PUT_HEX_FIELD_RET(s, field->next_cpu);
2249 SEQ_PUT_HEX_FIELD_RET(s, field->next_pid);
2250 SEQ_PUT_HEX_FIELD_RET(s, field->next_prio);
2251 SEQ_PUT_HEX_FIELD_RET(s, T);
2252 break;
2253 } 1711 }
2254 case TRACE_SPECIAL:
2255 case TRACE_USER_STACK:
2256 case TRACE_STACK: {
2257 struct special_entry *field;
2258 1712
2259 trace_assign_type(field, entry); 1713 event = ftrace_find_event(entry->type);
2260 1714 if (event) {
2261 SEQ_PUT_HEX_FIELD_RET(s, field->arg1); 1715 enum print_line_t ret = event->hex(iter, 0);
2262 SEQ_PUT_HEX_FIELD_RET(s, field->arg2); 1716 if (ret != TRACE_TYPE_HANDLED)
2263 SEQ_PUT_HEX_FIELD_RET(s, field->arg3); 1717 return ret;
2264 break;
2265 }
2266 } 1718 }
2267 SEQ_PUT_FIELD_RET(s, newline);
2268
2269 return TRACE_TYPE_HANDLED;
2270}
2271
2272static enum print_line_t print_printk_msg_only(struct trace_iterator *iter)
2273{
2274 struct trace_seq *s = &iter->seq;
2275 struct trace_entry *entry = iter->ent;
2276 struct print_entry *field;
2277 int ret;
2278
2279 trace_assign_type(field, entry);
2280
2281 ret = trace_seq_printf(s, field->buf);
2282 if (!ret)
2283 return TRACE_TYPE_PARTIAL_LINE;
2284 1719
2285 if (entry->flags & TRACE_FLAG_CONT) 1720 SEQ_PUT_FIELD_RET(s, newline);
2286 trace_seq_print_cont(s, iter);
2287 1721
2288 return TRACE_TYPE_HANDLED; 1722 return TRACE_TYPE_HANDLED;
2289} 1723}
@@ -2292,59 +1726,37 @@ static enum print_line_t print_bin_fmt(struct trace_iterator *iter)
2292{ 1726{
2293 struct trace_seq *s = &iter->seq; 1727 struct trace_seq *s = &iter->seq;
2294 struct trace_entry *entry; 1728 struct trace_entry *entry;
1729 struct trace_event *event;
2295 1730
2296 entry = iter->ent; 1731 entry = iter->ent;
2297 1732
2298 if (entry->type == TRACE_CONT) 1733 if (trace_flags & TRACE_ITER_CONTEXT_INFO) {
2299 return TRACE_TYPE_HANDLED; 1734 SEQ_PUT_FIELD_RET(s, entry->pid);
2300 1735 SEQ_PUT_FIELD_RET(s, iter->cpu);
2301 SEQ_PUT_FIELD_RET(s, entry->pid); 1736 SEQ_PUT_FIELD_RET(s, iter->ts);
2302 SEQ_PUT_FIELD_RET(s, entry->cpu);
2303 SEQ_PUT_FIELD_RET(s, iter->ts);
2304
2305 switch (entry->type) {
2306 case TRACE_FN: {
2307 struct ftrace_entry *field;
2308
2309 trace_assign_type(field, entry);
2310
2311 SEQ_PUT_FIELD_RET(s, field->ip);
2312 SEQ_PUT_FIELD_RET(s, field->parent_ip);
2313 break;
2314 }
2315 case TRACE_CTX: {
2316 struct ctx_switch_entry *field;
2317
2318 trace_assign_type(field, entry);
2319
2320 SEQ_PUT_FIELD_RET(s, field->prev_pid);
2321 SEQ_PUT_FIELD_RET(s, field->prev_prio);
2322 SEQ_PUT_FIELD_RET(s, field->prev_state);
2323 SEQ_PUT_FIELD_RET(s, field->next_pid);
2324 SEQ_PUT_FIELD_RET(s, field->next_prio);
2325 SEQ_PUT_FIELD_RET(s, field->next_state);
2326 break;
2327 } 1737 }
2328 case TRACE_SPECIAL:
2329 case TRACE_USER_STACK:
2330 case TRACE_STACK: {
2331 struct special_entry *field;
2332 1738
2333 trace_assign_type(field, entry); 1739 event = ftrace_find_event(entry->type);
2334 1740 return event ? event->binary(iter, 0) : TRACE_TYPE_HANDLED;
2335 SEQ_PUT_FIELD_RET(s, field->arg1);
2336 SEQ_PUT_FIELD_RET(s, field->arg2);
2337 SEQ_PUT_FIELD_RET(s, field->arg3);
2338 break;
2339 }
2340 }
2341 return 1;
2342} 1741}
2343 1742
2344static int trace_empty(struct trace_iterator *iter) 1743static int trace_empty(struct trace_iterator *iter)
2345{ 1744{
2346 int cpu; 1745 int cpu;
2347 1746
1747 /* If we are looking at one CPU buffer, only check that one */
1748 if (iter->cpu_file != TRACE_PIPE_ALL_CPU) {
1749 cpu = iter->cpu_file;
1750 if (iter->buffer_iter[cpu]) {
1751 if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
1752 return 0;
1753 } else {
1754 if (!ring_buffer_empty_cpu(iter->tr->buffer, cpu))
1755 return 0;
1756 }
1757 return 1;
1758 }
1759
2348 for_each_tracing_cpu(cpu) { 1760 for_each_tracing_cpu(cpu) {
2349 if (iter->buffer_iter[cpu]) { 1761 if (iter->buffer_iter[cpu]) {
2350 if (!ring_buffer_iter_empty(iter->buffer_iter[cpu])) 1762 if (!ring_buffer_iter_empty(iter->buffer_iter[cpu]))
@@ -2368,10 +1780,15 @@ static enum print_line_t print_trace_line(struct trace_iterator *iter)
2368 return ret; 1780 return ret;
2369 } 1781 }
2370 1782
1783 if (iter->ent->type == TRACE_BPRINT &&
1784 trace_flags & TRACE_ITER_PRINTK &&
1785 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
1786 return trace_print_bprintk_msg_only(iter);
1787
2371 if (iter->ent->type == TRACE_PRINT && 1788 if (iter->ent->type == TRACE_PRINT &&
2372 trace_flags & TRACE_ITER_PRINTK && 1789 trace_flags & TRACE_ITER_PRINTK &&
2373 trace_flags & TRACE_ITER_PRINTK_MSGONLY) 1790 trace_flags & TRACE_ITER_PRINTK_MSGONLY)
2374 return print_printk_msg_only(iter); 1791 return trace_print_printk_msg_only(iter);
2375 1792
2376 if (trace_flags & TRACE_ITER_BIN) 1793 if (trace_flags & TRACE_ITER_BIN)
2377 return print_bin_fmt(iter); 1794 return print_bin_fmt(iter);
@@ -2382,9 +1799,6 @@ static enum print_line_t print_trace_line(struct trace_iterator *iter)
2382 if (trace_flags & TRACE_ITER_RAW) 1799 if (trace_flags & TRACE_ITER_RAW)
2383 return print_raw_fmt(iter); 1800 return print_raw_fmt(iter);
2384 1801
2385 if (iter->iter_flags & TRACE_FILE_LAT_FMT)
2386 return print_lat_fmt(iter, iter->idx, iter->cpu);
2387
2388 return print_trace_fmt(iter); 1802 return print_trace_fmt(iter);
2389} 1803}
2390 1804
@@ -2426,30 +1840,40 @@ static struct seq_operations tracer_seq_ops = {
2426}; 1840};
2427 1841
2428static struct trace_iterator * 1842static struct trace_iterator *
2429__tracing_open(struct inode *inode, struct file *file, int *ret) 1843__tracing_open(struct inode *inode, struct file *file)
2430{ 1844{
1845 long cpu_file = (long) inode->i_private;
1846 void *fail_ret = ERR_PTR(-ENOMEM);
2431 struct trace_iterator *iter; 1847 struct trace_iterator *iter;
2432 struct seq_file *m; 1848 struct seq_file *m;
2433 int cpu; 1849 int cpu, ret;
2434 1850
2435 if (tracing_disabled) { 1851 if (tracing_disabled)
2436 *ret = -ENODEV; 1852 return ERR_PTR(-ENODEV);
2437 return NULL;
2438 }
2439 1853
2440 iter = kzalloc(sizeof(*iter), GFP_KERNEL); 1854 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
2441 if (!iter) { 1855 if (!iter)
2442 *ret = -ENOMEM; 1856 return ERR_PTR(-ENOMEM);
2443 goto out;
2444 }
2445 1857
1858 /*
1859 * We make a copy of the current tracer to avoid concurrent
1860 * changes on it while we are reading.
1861 */
2446 mutex_lock(&trace_types_lock); 1862 mutex_lock(&trace_types_lock);
1863 iter->trace = kzalloc(sizeof(*iter->trace), GFP_KERNEL);
1864 if (!iter->trace)
1865 goto fail;
1866
1867 if (current_trace)
1868 *iter->trace = *current_trace;
1869
2447 if (current_trace && current_trace->print_max) 1870 if (current_trace && current_trace->print_max)
2448 iter->tr = &max_tr; 1871 iter->tr = &max_tr;
2449 else 1872 else
2450 iter->tr = inode->i_private; 1873 iter->tr = &global_trace;
2451 iter->trace = current_trace;
2452 iter->pos = -1; 1874 iter->pos = -1;
1875 mutex_init(&iter->mutex);
1876 iter->cpu_file = cpu_file;
2453 1877
2454 /* Notify the tracer early; before we stop tracing. */ 1878 /* Notify the tracer early; before we stop tracing. */
2455 if (iter->trace && iter->trace->open) 1879 if (iter->trace && iter->trace->open)
@@ -2459,20 +1883,24 @@ __tracing_open(struct inode *inode, struct file *file, int *ret)
2459 if (ring_buffer_overruns(iter->tr->buffer)) 1883 if (ring_buffer_overruns(iter->tr->buffer))
2460 iter->iter_flags |= TRACE_FILE_ANNOTATE; 1884 iter->iter_flags |= TRACE_FILE_ANNOTATE;
2461 1885
1886 if (iter->cpu_file == TRACE_PIPE_ALL_CPU) {
1887 for_each_tracing_cpu(cpu) {
2462 1888
2463 for_each_tracing_cpu(cpu) { 1889 iter->buffer_iter[cpu] =
2464 1890 ring_buffer_read_start(iter->tr->buffer, cpu);
1891 }
1892 } else {
1893 cpu = iter->cpu_file;
2465 iter->buffer_iter[cpu] = 1894 iter->buffer_iter[cpu] =
2466 ring_buffer_read_start(iter->tr->buffer, cpu); 1895 ring_buffer_read_start(iter->tr->buffer, cpu);
2467
2468 if (!iter->buffer_iter[cpu])
2469 goto fail_buffer;
2470 } 1896 }
2471 1897
2472 /* TODO stop tracer */ 1898 /* TODO stop tracer */
2473 *ret = seq_open(file, &tracer_seq_ops); 1899 ret = seq_open(file, &tracer_seq_ops);
2474 if (*ret) 1900 if (ret < 0) {
1901 fail_ret = ERR_PTR(ret);
2475 goto fail_buffer; 1902 goto fail_buffer;
1903 }
2476 1904
2477 m = file->private_data; 1905 m = file->private_data;
2478 m->private = iter; 1906 m->private = iter;
@@ -2482,7 +1910,6 @@ __tracing_open(struct inode *inode, struct file *file, int *ret)
2482 1910
2483 mutex_unlock(&trace_types_lock); 1911 mutex_unlock(&trace_types_lock);
2484 1912
2485 out:
2486 return iter; 1913 return iter;
2487 1914
2488 fail_buffer: 1915 fail_buffer:
@@ -2490,10 +1917,12 @@ __tracing_open(struct inode *inode, struct file *file, int *ret)
2490 if (iter->buffer_iter[cpu]) 1917 if (iter->buffer_iter[cpu])
2491 ring_buffer_read_finish(iter->buffer_iter[cpu]); 1918 ring_buffer_read_finish(iter->buffer_iter[cpu]);
2492 } 1919 }
1920 fail:
2493 mutex_unlock(&trace_types_lock); 1921 mutex_unlock(&trace_types_lock);
1922 kfree(iter->trace);
2494 kfree(iter); 1923 kfree(iter);
2495 1924
2496 return ERR_PTR(-ENOMEM); 1925 return fail_ret;
2497} 1926}
2498 1927
2499int tracing_open_generic(struct inode *inode, struct file *filp) 1928int tracing_open_generic(struct inode *inode, struct file *filp)
@@ -2505,12 +1934,17 @@ int tracing_open_generic(struct inode *inode, struct file *filp)
2505 return 0; 1934 return 0;
2506} 1935}
2507 1936
2508int tracing_release(struct inode *inode, struct file *file) 1937static int tracing_release(struct inode *inode, struct file *file)
2509{ 1938{
2510 struct seq_file *m = (struct seq_file *)file->private_data; 1939 struct seq_file *m = (struct seq_file *)file->private_data;
2511 struct trace_iterator *iter = m->private; 1940 struct trace_iterator *iter;
2512 int cpu; 1941 int cpu;
2513 1942
1943 if (!(file->f_mode & FMODE_READ))
1944 return 0;
1945
1946 iter = m->private;
1947
2514 mutex_lock(&trace_types_lock); 1948 mutex_lock(&trace_types_lock);
2515 for_each_tracing_cpu(cpu) { 1949 for_each_tracing_cpu(cpu) {
2516 if (iter->buffer_iter[cpu]) 1950 if (iter->buffer_iter[cpu])
@@ -2525,33 +1959,38 @@ int tracing_release(struct inode *inode, struct file *file)
2525 mutex_unlock(&trace_types_lock); 1959 mutex_unlock(&trace_types_lock);
2526 1960
2527 seq_release(inode, file); 1961 seq_release(inode, file);
1962 mutex_destroy(&iter->mutex);
1963 kfree(iter->trace);
2528 kfree(iter); 1964 kfree(iter);
2529 return 0; 1965 return 0;
2530} 1966}
2531 1967
2532static int tracing_open(struct inode *inode, struct file *file) 1968static int tracing_open(struct inode *inode, struct file *file)
2533{ 1969{
2534 int ret;
2535
2536 __tracing_open(inode, file, &ret);
2537
2538 return ret;
2539}
2540
2541static int tracing_lt_open(struct inode *inode, struct file *file)
2542{
2543 struct trace_iterator *iter; 1970 struct trace_iterator *iter;
2544 int ret; 1971 int ret = 0;
2545 1972
2546 iter = __tracing_open(inode, file, &ret); 1973 /* If this file was open for write, then erase contents */
1974 if ((file->f_mode & FMODE_WRITE) &&
1975 !(file->f_flags & O_APPEND)) {
1976 long cpu = (long) inode->i_private;
2547 1977
2548 if (!ret) 1978 if (cpu == TRACE_PIPE_ALL_CPU)
2549 iter->iter_flags |= TRACE_FILE_LAT_FMT; 1979 tracing_reset_online_cpus(&global_trace);
1980 else
1981 tracing_reset(&global_trace, cpu);
1982 }
2550 1983
1984 if (file->f_mode & FMODE_READ) {
1985 iter = __tracing_open(inode, file);
1986 if (IS_ERR(iter))
1987 ret = PTR_ERR(iter);
1988 else if (trace_flags & TRACE_ITER_LATENCY_FMT)
1989 iter->iter_flags |= TRACE_FILE_LAT_FMT;
1990 }
2551 return ret; 1991 return ret;
2552} 1992}
2553 1993
2554
2555static void * 1994static void *
2556t_next(struct seq_file *m, void *v, loff_t *pos) 1995t_next(struct seq_file *m, void *v, loff_t *pos)
2557{ 1996{
@@ -2623,21 +2062,22 @@ static int show_traces_open(struct inode *inode, struct file *file)
2623 return ret; 2062 return ret;
2624} 2063}
2625 2064
2626static struct file_operations tracing_fops = { 2065static ssize_t
2627 .open = tracing_open, 2066tracing_write_stub(struct file *filp, const char __user *ubuf,
2628 .read = seq_read, 2067 size_t count, loff_t *ppos)
2629 .llseek = seq_lseek, 2068{
2630 .release = tracing_release, 2069 return count;
2631}; 2070}
2632 2071
2633static struct file_operations tracing_lt_fops = { 2072static const struct file_operations tracing_fops = {
2634 .open = tracing_lt_open, 2073 .open = tracing_open,
2635 .read = seq_read, 2074 .read = seq_read,
2075 .write = tracing_write_stub,
2636 .llseek = seq_lseek, 2076 .llseek = seq_lseek,
2637 .release = tracing_release, 2077 .release = tracing_release,
2638}; 2078};
2639 2079
2640static struct file_operations show_traces_fops = { 2080static const struct file_operations show_traces_fops = {
2641 .open = show_traces_open, 2081 .open = show_traces_open,
2642 .read = seq_read, 2082 .read = seq_read,
2643 .release = seq_release, 2083 .release = seq_release,
@@ -2730,7 +2170,7 @@ err_unlock:
2730 return err; 2170 return err;
2731} 2171}
2732 2172
2733static struct file_operations tracing_cpumask_fops = { 2173static const struct file_operations tracing_cpumask_fops = {
2734 .open = tracing_open_generic, 2174 .open = tracing_open_generic,
2735 .read = tracing_cpumask_read, 2175 .read = tracing_cpumask_read,
2736 .write = tracing_cpumask_write, 2176 .write = tracing_cpumask_write,
@@ -2740,57 +2180,62 @@ static ssize_t
2740tracing_trace_options_read(struct file *filp, char __user *ubuf, 2180tracing_trace_options_read(struct file *filp, char __user *ubuf,
2741 size_t cnt, loff_t *ppos) 2181 size_t cnt, loff_t *ppos)
2742{ 2182{
2743 int i; 2183 struct tracer_opt *trace_opts;
2184 u32 tracer_flags;
2185 int len = 0;
2744 char *buf; 2186 char *buf;
2745 int r = 0; 2187 int r = 0;
2746 int len = 0; 2188 int i;
2747 u32 tracer_flags = current_trace->flags->val;
2748 struct tracer_opt *trace_opts = current_trace->flags->opts;
2749 2189
2750 2190
2751 /* calulate max size */ 2191 /* calculate max size */
2752 for (i = 0; trace_options[i]; i++) { 2192 for (i = 0; trace_options[i]; i++) {
2753 len += strlen(trace_options[i]); 2193 len += strlen(trace_options[i]);
2754 len += 3; /* "no" and space */ 2194 len += 3; /* "no" and newline */
2755 } 2195 }
2756 2196
2197 mutex_lock(&trace_types_lock);
2198 tracer_flags = current_trace->flags->val;
2199 trace_opts = current_trace->flags->opts;
2200
2757 /* 2201 /*
2758 * Increase the size with names of options specific 2202 * Increase the size with names of options specific
2759 * of the current tracer. 2203 * of the current tracer.
2760 */ 2204 */
2761 for (i = 0; trace_opts[i].name; i++) { 2205 for (i = 0; trace_opts[i].name; i++) {
2762 len += strlen(trace_opts[i].name); 2206 len += strlen(trace_opts[i].name);
2763 len += 3; /* "no" and space */ 2207 len += 3; /* "no" and newline */
2764 } 2208 }
2765 2209
2766 /* +2 for \n and \0 */ 2210 /* +2 for \n and \0 */
2767 buf = kmalloc(len + 2, GFP_KERNEL); 2211 buf = kmalloc(len + 2, GFP_KERNEL);
2768 if (!buf) 2212 if (!buf) {
2213 mutex_unlock(&trace_types_lock);
2769 return -ENOMEM; 2214 return -ENOMEM;
2215 }
2770 2216
2771 for (i = 0; trace_options[i]; i++) { 2217 for (i = 0; trace_options[i]; i++) {
2772 if (trace_flags & (1 << i)) 2218 if (trace_flags & (1 << i))
2773 r += sprintf(buf + r, "%s ", trace_options[i]); 2219 r += sprintf(buf + r, "%s\n", trace_options[i]);
2774 else 2220 else
2775 r += sprintf(buf + r, "no%s ", trace_options[i]); 2221 r += sprintf(buf + r, "no%s\n", trace_options[i]);
2776 } 2222 }
2777 2223
2778 for (i = 0; trace_opts[i].name; i++) { 2224 for (i = 0; trace_opts[i].name; i++) {
2779 if (tracer_flags & trace_opts[i].bit) 2225 if (tracer_flags & trace_opts[i].bit)
2780 r += sprintf(buf + r, "%s ", 2226 r += sprintf(buf + r, "%s\n",
2781 trace_opts[i].name); 2227 trace_opts[i].name);
2782 else 2228 else
2783 r += sprintf(buf + r, "no%s ", 2229 r += sprintf(buf + r, "no%s\n",
2784 trace_opts[i].name); 2230 trace_opts[i].name);
2785 } 2231 }
2232 mutex_unlock(&trace_types_lock);
2786 2233
2787 r += sprintf(buf + r, "\n");
2788 WARN_ON(r >= len + 2); 2234 WARN_ON(r >= len + 2);
2789 2235
2790 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r); 2236 r = simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2791 2237
2792 kfree(buf); 2238 kfree(buf);
2793
2794 return r; 2239 return r;
2795} 2240}
2796 2241
@@ -2828,6 +2273,34 @@ static int set_tracer_option(struct tracer *trace, char *cmp, int neg)
2828 return 0; 2273 return 0;
2829} 2274}
2830 2275
2276static void set_tracer_flags(unsigned int mask, int enabled)
2277{
2278 /* do nothing if flag is already set */
2279 if (!!(trace_flags & mask) == !!enabled)
2280 return;
2281
2282 if (enabled)
2283 trace_flags |= mask;
2284 else
2285 trace_flags &= ~mask;
2286
2287 if (mask == TRACE_ITER_GLOBAL_CLK) {
2288 u64 (*func)(void);
2289
2290 if (enabled)
2291 func = trace_clock_global;
2292 else
2293 func = trace_clock_local;
2294
2295 mutex_lock(&trace_types_lock);
2296 ring_buffer_set_clock(global_trace.buffer, func);
2297
2298 if (max_tr.buffer)
2299 ring_buffer_set_clock(max_tr.buffer, func);
2300 mutex_unlock(&trace_types_lock);
2301 }
2302}
2303
2831static ssize_t 2304static ssize_t
2832tracing_trace_options_write(struct file *filp, const char __user *ubuf, 2305tracing_trace_options_write(struct file *filp, const char __user *ubuf,
2833 size_t cnt, loff_t *ppos) 2306 size_t cnt, loff_t *ppos)
@@ -2855,17 +2328,16 @@ tracing_trace_options_write(struct file *filp, const char __user *ubuf,
2855 int len = strlen(trace_options[i]); 2328 int len = strlen(trace_options[i]);
2856 2329
2857 if (strncmp(cmp, trace_options[i], len) == 0) { 2330 if (strncmp(cmp, trace_options[i], len) == 0) {
2858 if (neg) 2331 set_tracer_flags(1 << i, !neg);
2859 trace_flags &= ~(1 << i);
2860 else
2861 trace_flags |= (1 << i);
2862 break; 2332 break;
2863 } 2333 }
2864 } 2334 }
2865 2335
2866 /* If no option could be set, test the specific tracer options */ 2336 /* If no option could be set, test the specific tracer options */
2867 if (!trace_options[i]) { 2337 if (!trace_options[i]) {
2338 mutex_lock(&trace_types_lock);
2868 ret = set_tracer_option(current_trace, cmp, neg); 2339 ret = set_tracer_option(current_trace, cmp, neg);
2340 mutex_unlock(&trace_types_lock);
2869 if (ret) 2341 if (ret)
2870 return ret; 2342 return ret;
2871 } 2343 }
@@ -2875,7 +2347,7 @@ tracing_trace_options_write(struct file *filp, const char __user *ubuf,
2875 return cnt; 2347 return cnt;
2876} 2348}
2877 2349
2878static struct file_operations tracing_iter_fops = { 2350static const struct file_operations tracing_iter_fops = {
2879 .open = tracing_open_generic, 2351 .open = tracing_open_generic,
2880 .read = tracing_trace_options_read, 2352 .read = tracing_trace_options_read,
2881 .write = tracing_trace_options_write, 2353 .write = tracing_trace_options_write,
@@ -2908,7 +2380,7 @@ tracing_readme_read(struct file *filp, char __user *ubuf,
2908 readme_msg, strlen(readme_msg)); 2380 readme_msg, strlen(readme_msg));
2909} 2381}
2910 2382
2911static struct file_operations tracing_readme_fops = { 2383static const struct file_operations tracing_readme_fops = {
2912 .open = tracing_open_generic, 2384 .open = tracing_open_generic,
2913 .read = tracing_readme_read, 2385 .read = tracing_readme_read,
2914}; 2386};
@@ -2930,7 +2402,7 @@ tracing_ctrl_write(struct file *filp, const char __user *ubuf,
2930{ 2402{
2931 struct trace_array *tr = filp->private_data; 2403 struct trace_array *tr = filp->private_data;
2932 char buf[64]; 2404 char buf[64];
2933 long val; 2405 unsigned long val;
2934 int ret; 2406 int ret;
2935 2407
2936 if (cnt >= sizeof(buf)) 2408 if (cnt >= sizeof(buf))
@@ -2985,13 +2457,105 @@ tracing_set_trace_read(struct file *filp, char __user *ubuf,
2985 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r); 2457 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
2986} 2458}
2987 2459
2988static int tracing_set_tracer(char *buf) 2460int tracer_init(struct tracer *t, struct trace_array *tr)
2461{
2462 tracing_reset_online_cpus(tr);
2463 return t->init(tr);
2464}
2465
2466static int tracing_resize_ring_buffer(unsigned long size)
2467{
2468 int ret;
2469
2470 /*
2471 * If kernel or user changes the size of the ring buffer
2472 * we use the size that was given, and we can forget about
2473 * expanding it later.
2474 */
2475 ring_buffer_expanded = 1;
2476
2477 ret = ring_buffer_resize(global_trace.buffer, size);
2478 if (ret < 0)
2479 return ret;
2480
2481 ret = ring_buffer_resize(max_tr.buffer, size);
2482 if (ret < 0) {
2483 int r;
2484
2485 r = ring_buffer_resize(global_trace.buffer,
2486 global_trace.entries);
2487 if (r < 0) {
2488 /*
2489 * AARGH! We are left with different
2490 * size max buffer!!!!
2491 * The max buffer is our "snapshot" buffer.
2492 * When a tracer needs a snapshot (one of the
2493 * latency tracers), it swaps the max buffer
2494 * with the saved snap shot. We succeeded to
2495 * update the size of the main buffer, but failed to
2496 * update the size of the max buffer. But when we tried
2497 * to reset the main buffer to the original size, we
2498 * failed there too. This is very unlikely to
2499 * happen, but if it does, warn and kill all
2500 * tracing.
2501 */
2502 WARN_ON(1);
2503 tracing_disabled = 1;
2504 }
2505 return ret;
2506 }
2507
2508 global_trace.entries = size;
2509
2510 return ret;
2511}
2512
2513/**
2514 * tracing_update_buffers - used by tracing facility to expand ring buffers
2515 *
2516 * To save on memory when the tracing is never used on a system with it
2517 * configured in. The ring buffers are set to a minimum size. But once
2518 * a user starts to use the tracing facility, then they need to grow
2519 * to their default size.
2520 *
2521 * This function is to be called when a tracer is about to be used.
2522 */
2523int tracing_update_buffers(void)
2524{
2525 int ret = 0;
2526
2527 mutex_lock(&trace_types_lock);
2528 if (!ring_buffer_expanded)
2529 ret = tracing_resize_ring_buffer(trace_buf_size);
2530 mutex_unlock(&trace_types_lock);
2531
2532 return ret;
2533}
2534
2535struct trace_option_dentry;
2536
2537static struct trace_option_dentry *
2538create_trace_option_files(struct tracer *tracer);
2539
2540static void
2541destroy_trace_option_files(struct trace_option_dentry *topts);
2542
2543static int tracing_set_tracer(const char *buf)
2989{ 2544{
2545 static struct trace_option_dentry *topts;
2990 struct trace_array *tr = &global_trace; 2546 struct trace_array *tr = &global_trace;
2991 struct tracer *t; 2547 struct tracer *t;
2992 int ret = 0; 2548 int ret = 0;
2993 2549
2994 mutex_lock(&trace_types_lock); 2550 mutex_lock(&trace_types_lock);
2551
2552 if (!ring_buffer_expanded) {
2553 ret = tracing_resize_ring_buffer(trace_buf_size);
2554 if (ret < 0)
2555 goto out;
2556 ret = 0;
2557 }
2558
2995 for (t = trace_types; t; t = t->next) { 2559 for (t = trace_types; t; t = t->next) {
2996 if (strcmp(t->name, buf) == 0) 2560 if (strcmp(t->name, buf) == 0)
2997 break; 2561 break;
@@ -3007,9 +2571,14 @@ static int tracing_set_tracer(char *buf)
3007 if (current_trace && current_trace->reset) 2571 if (current_trace && current_trace->reset)
3008 current_trace->reset(tr); 2572 current_trace->reset(tr);
3009 2573
2574 destroy_trace_option_files(topts);
2575
3010 current_trace = t; 2576 current_trace = t;
2577
2578 topts = create_trace_option_files(current_trace);
2579
3011 if (t->init) { 2580 if (t->init) {
3012 ret = t->init(tr); 2581 ret = tracer_init(t, tr);
3013 if (ret) 2582 if (ret)
3014 goto out; 2583 goto out;
3015 } 2584 }
@@ -3072,9 +2641,9 @@ static ssize_t
3072tracing_max_lat_write(struct file *filp, const char __user *ubuf, 2641tracing_max_lat_write(struct file *filp, const char __user *ubuf,
3073 size_t cnt, loff_t *ppos) 2642 size_t cnt, loff_t *ppos)
3074{ 2643{
3075 long *ptr = filp->private_data; 2644 unsigned long *ptr = filp->private_data;
3076 char buf[64]; 2645 char buf[64];
3077 long val; 2646 unsigned long val;
3078 int ret; 2647 int ret;
3079 2648
3080 if (cnt >= sizeof(buf)) 2649 if (cnt >= sizeof(buf))
@@ -3094,54 +2663,96 @@ tracing_max_lat_write(struct file *filp, const char __user *ubuf,
3094 return cnt; 2663 return cnt;
3095} 2664}
3096 2665
3097static atomic_t tracing_reader;
3098
3099static int tracing_open_pipe(struct inode *inode, struct file *filp) 2666static int tracing_open_pipe(struct inode *inode, struct file *filp)
3100{ 2667{
2668 long cpu_file = (long) inode->i_private;
3101 struct trace_iterator *iter; 2669 struct trace_iterator *iter;
2670 int ret = 0;
3102 2671
3103 if (tracing_disabled) 2672 if (tracing_disabled)
3104 return -ENODEV; 2673 return -ENODEV;
3105 2674
3106 /* We only allow for reader of the pipe */ 2675 mutex_lock(&trace_types_lock);
3107 if (atomic_inc_return(&tracing_reader) != 1) { 2676
3108 atomic_dec(&tracing_reader); 2677 /* We only allow one reader per cpu */
3109 return -EBUSY; 2678 if (cpu_file == TRACE_PIPE_ALL_CPU) {
2679 if (!cpumask_empty(tracing_reader_cpumask)) {
2680 ret = -EBUSY;
2681 goto out;
2682 }
2683 cpumask_setall(tracing_reader_cpumask);
2684 } else {
2685 if (!cpumask_test_cpu(cpu_file, tracing_reader_cpumask))
2686 cpumask_set_cpu(cpu_file, tracing_reader_cpumask);
2687 else {
2688 ret = -EBUSY;
2689 goto out;
2690 }
3110 } 2691 }
3111 2692
3112 /* create a buffer to store the information to pass to userspace */ 2693 /* create a buffer to store the information to pass to userspace */
3113 iter = kzalloc(sizeof(*iter), GFP_KERNEL); 2694 iter = kzalloc(sizeof(*iter), GFP_KERNEL);
3114 if (!iter) 2695 if (!iter) {
3115 return -ENOMEM; 2696 ret = -ENOMEM;
2697 goto out;
2698 }
3116 2699
3117 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) { 2700 /*
3118 kfree(iter); 2701 * We make a copy of the current tracer to avoid concurrent
3119 return -ENOMEM; 2702 * changes on it while we are reading.
2703 */
2704 iter->trace = kmalloc(sizeof(*iter->trace), GFP_KERNEL);
2705 if (!iter->trace) {
2706 ret = -ENOMEM;
2707 goto fail;
3120 } 2708 }
2709 if (current_trace)
2710 *iter->trace = *current_trace;
3121 2711
3122 mutex_lock(&trace_types_lock); 2712 if (!alloc_cpumask_var(&iter->started, GFP_KERNEL)) {
2713 ret = -ENOMEM;
2714 goto fail;
2715 }
3123 2716
3124 /* trace pipe does not show start of buffer */ 2717 /* trace pipe does not show start of buffer */
3125 cpumask_setall(iter->started); 2718 cpumask_setall(iter->started);
3126 2719
2720 iter->cpu_file = cpu_file;
3127 iter->tr = &global_trace; 2721 iter->tr = &global_trace;
3128 iter->trace = current_trace; 2722 mutex_init(&iter->mutex);
3129 filp->private_data = iter; 2723 filp->private_data = iter;
3130 2724
3131 if (iter->trace->pipe_open) 2725 if (iter->trace->pipe_open)
3132 iter->trace->pipe_open(iter); 2726 iter->trace->pipe_open(iter);
2727
2728out:
3133 mutex_unlock(&trace_types_lock); 2729 mutex_unlock(&trace_types_lock);
2730 return ret;
3134 2731
3135 return 0; 2732fail:
2733 kfree(iter->trace);
2734 kfree(iter);
2735 mutex_unlock(&trace_types_lock);
2736 return ret;
3136} 2737}
3137 2738
3138static int tracing_release_pipe(struct inode *inode, struct file *file) 2739static int tracing_release_pipe(struct inode *inode, struct file *file)
3139{ 2740{
3140 struct trace_iterator *iter = file->private_data; 2741 struct trace_iterator *iter = file->private_data;
3141 2742
2743 mutex_lock(&trace_types_lock);
2744
2745 if (iter->cpu_file == TRACE_PIPE_ALL_CPU)
2746 cpumask_clear(tracing_reader_cpumask);
2747 else
2748 cpumask_clear_cpu(iter->cpu_file, tracing_reader_cpumask);
2749
2750 mutex_unlock(&trace_types_lock);
2751
3142 free_cpumask_var(iter->started); 2752 free_cpumask_var(iter->started);
2753 mutex_destroy(&iter->mutex);
2754 kfree(iter->trace);
3143 kfree(iter); 2755 kfree(iter);
3144 atomic_dec(&tracing_reader);
3145 2756
3146 return 0; 2757 return 0;
3147} 2758}
@@ -3167,67 +2778,57 @@ tracing_poll_pipe(struct file *filp, poll_table *poll_table)
3167 } 2778 }
3168} 2779}
3169 2780
3170/* 2781
3171 * Consumer reader. 2782void default_wait_pipe(struct trace_iterator *iter)
3172 */
3173static ssize_t
3174tracing_read_pipe(struct file *filp, char __user *ubuf,
3175 size_t cnt, loff_t *ppos)
3176{ 2783{
3177 struct trace_iterator *iter = filp->private_data; 2784 DEFINE_WAIT(wait);
3178 ssize_t sret;
3179 2785
3180 /* return any leftover data */ 2786 prepare_to_wait(&trace_wait, &wait, TASK_INTERRUPTIBLE);
3181 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3182 if (sret != -EBUSY)
3183 return sret;
3184 2787
3185 trace_seq_reset(&iter->seq); 2788 if (trace_empty(iter))
2789 schedule();
3186 2790
3187 mutex_lock(&trace_types_lock); 2791 finish_wait(&trace_wait, &wait);
3188 if (iter->trace->read) { 2792}
3189 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos); 2793
3190 if (sret) 2794/*
3191 goto out; 2795 * This is a make-shift waitqueue.
3192 } 2796 * A tracer might use this callback on some rare cases:
2797 *
2798 * 1) the current tracer might hold the runqueue lock when it wakes up
2799 * a reader, hence a deadlock (sched, function, and function graph tracers)
2800 * 2) the function tracers, trace all functions, we don't want
2801 * the overhead of calling wake_up and friends
2802 * (and tracing them too)
2803 *
2804 * Anyway, this is really very primitive wakeup.
2805 */
2806void poll_wait_pipe(struct trace_iterator *iter)
2807{
2808 set_current_state(TASK_INTERRUPTIBLE);
2809 /* sleep for 100 msecs, and try again. */
2810 schedule_timeout(HZ / 10);
2811}
2812
2813/* Must be called with trace_types_lock mutex held. */
2814static int tracing_wait_pipe(struct file *filp)
2815{
2816 struct trace_iterator *iter = filp->private_data;
3193 2817
3194waitagain:
3195 sret = 0;
3196 while (trace_empty(iter)) { 2818 while (trace_empty(iter)) {
3197 2819
3198 if ((filp->f_flags & O_NONBLOCK)) { 2820 if ((filp->f_flags & O_NONBLOCK)) {
3199 sret = -EAGAIN; 2821 return -EAGAIN;
3200 goto out;
3201 } 2822 }
3202 2823
3203 /* 2824 mutex_unlock(&iter->mutex);
3204 * This is a make-shift waitqueue. The reason we don't use
3205 * an actual wait queue is because:
3206 * 1) we only ever have one waiter
3207 * 2) the tracing, traces all functions, we don't want
3208 * the overhead of calling wake_up and friends
3209 * (and tracing them too)
3210 * Anyway, this is really very primitive wakeup.
3211 */
3212 set_current_state(TASK_INTERRUPTIBLE);
3213 iter->tr->waiter = current;
3214
3215 mutex_unlock(&trace_types_lock);
3216
3217 /* sleep for 100 msecs, and try again. */
3218 schedule_timeout(HZ/10);
3219
3220 mutex_lock(&trace_types_lock);
3221 2825
3222 iter->tr->waiter = NULL; 2826 iter->trace->wait_pipe(iter);
3223 2827
3224 if (signal_pending(current)) { 2828 mutex_lock(&iter->mutex);
3225 sret = -EINTR;
3226 goto out;
3227 }
3228 2829
3229 if (iter->trace != current_trace) 2830 if (signal_pending(current))
3230 goto out; 2831 return -EINTR;
3231 2832
3232 /* 2833 /*
3233 * We block until we read something and tracing is disabled. 2834 * We block until we read something and tracing is disabled.
@@ -3240,13 +2841,59 @@ waitagain:
3240 */ 2841 */
3241 if (!tracer_enabled && iter->pos) 2842 if (!tracer_enabled && iter->pos)
3242 break; 2843 break;
2844 }
2845
2846 return 1;
2847}
2848
2849/*
2850 * Consumer reader.
2851 */
2852static ssize_t
2853tracing_read_pipe(struct file *filp, char __user *ubuf,
2854 size_t cnt, loff_t *ppos)
2855{
2856 struct trace_iterator *iter = filp->private_data;
2857 static struct tracer *old_tracer;
2858 ssize_t sret;
3243 2859
3244 continue; 2860 /* return any leftover data */
2861 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
2862 if (sret != -EBUSY)
2863 return sret;
2864
2865 trace_seq_init(&iter->seq);
2866
2867 /* copy the tracer to avoid using a global lock all around */
2868 mutex_lock(&trace_types_lock);
2869 if (unlikely(old_tracer != current_trace && current_trace)) {
2870 old_tracer = current_trace;
2871 *iter->trace = *current_trace;
2872 }
2873 mutex_unlock(&trace_types_lock);
2874
2875 /*
2876 * Avoid more than one consumer on a single file descriptor
2877 * This is just a matter of traces coherency, the ring buffer itself
2878 * is protected.
2879 */
2880 mutex_lock(&iter->mutex);
2881 if (iter->trace->read) {
2882 sret = iter->trace->read(iter, filp, ubuf, cnt, ppos);
2883 if (sret)
2884 goto out;
3245 } 2885 }
3246 2886
2887waitagain:
2888 sret = tracing_wait_pipe(filp);
2889 if (sret <= 0)
2890 goto out;
2891
3247 /* stop when tracing is finished */ 2892 /* stop when tracing is finished */
3248 if (trace_empty(iter)) 2893 if (trace_empty(iter)) {
2894 sret = 0;
3249 goto out; 2895 goto out;
2896 }
3250 2897
3251 if (cnt >= PAGE_SIZE) 2898 if (cnt >= PAGE_SIZE)
3252 cnt = PAGE_SIZE - 1; 2899 cnt = PAGE_SIZE - 1;
@@ -3267,8 +2914,8 @@ waitagain:
3267 iter->seq.len = len; 2914 iter->seq.len = len;
3268 break; 2915 break;
3269 } 2916 }
3270 2917 if (ret != TRACE_TYPE_NO_CONSUME)
3271 trace_consume(iter); 2918 trace_consume(iter);
3272 2919
3273 if (iter->seq.len >= cnt) 2920 if (iter->seq.len >= cnt)
3274 break; 2921 break;
@@ -3277,7 +2924,7 @@ waitagain:
3277 /* Now copy what we have to the user */ 2924 /* Now copy what we have to the user */
3278 sret = trace_seq_to_user(&iter->seq, ubuf, cnt); 2925 sret = trace_seq_to_user(&iter->seq, ubuf, cnt);
3279 if (iter->seq.readpos >= iter->seq.len) 2926 if (iter->seq.readpos >= iter->seq.len)
3280 trace_seq_reset(&iter->seq); 2927 trace_seq_init(&iter->seq);
3281 2928
3282 /* 2929 /*
3283 * If there was nothing to send to user, inspite of consuming trace 2930 * If there was nothing to send to user, inspite of consuming trace
@@ -3287,20 +2934,165 @@ waitagain:
3287 goto waitagain; 2934 goto waitagain;
3288 2935
3289out: 2936out:
3290 mutex_unlock(&trace_types_lock); 2937 mutex_unlock(&iter->mutex);
3291 2938
3292 return sret; 2939 return sret;
3293} 2940}
3294 2941
2942static void tracing_pipe_buf_release(struct pipe_inode_info *pipe,
2943 struct pipe_buffer *buf)
2944{
2945 __free_page(buf->page);
2946}
2947
2948static void tracing_spd_release_pipe(struct splice_pipe_desc *spd,
2949 unsigned int idx)
2950{
2951 __free_page(spd->pages[idx]);
2952}
2953
2954static struct pipe_buf_operations tracing_pipe_buf_ops = {
2955 .can_merge = 0,
2956 .map = generic_pipe_buf_map,
2957 .unmap = generic_pipe_buf_unmap,
2958 .confirm = generic_pipe_buf_confirm,
2959 .release = tracing_pipe_buf_release,
2960 .steal = generic_pipe_buf_steal,
2961 .get = generic_pipe_buf_get,
2962};
2963
2964static size_t
2965tracing_fill_pipe_page(size_t rem, struct trace_iterator *iter)
2966{
2967 size_t count;
2968 int ret;
2969
2970 /* Seq buffer is page-sized, exactly what we need. */
2971 for (;;) {
2972 count = iter->seq.len;
2973 ret = print_trace_line(iter);
2974 count = iter->seq.len - count;
2975 if (rem < count) {
2976 rem = 0;
2977 iter->seq.len -= count;
2978 break;
2979 }
2980 if (ret == TRACE_TYPE_PARTIAL_LINE) {
2981 iter->seq.len -= count;
2982 break;
2983 }
2984
2985 trace_consume(iter);
2986 rem -= count;
2987 if (!find_next_entry_inc(iter)) {
2988 rem = 0;
2989 iter->ent = NULL;
2990 break;
2991 }
2992 }
2993
2994 return rem;
2995}
2996
2997static ssize_t tracing_splice_read_pipe(struct file *filp,
2998 loff_t *ppos,
2999 struct pipe_inode_info *pipe,
3000 size_t len,
3001 unsigned int flags)
3002{
3003 struct page *pages[PIPE_BUFFERS];
3004 struct partial_page partial[PIPE_BUFFERS];
3005 struct trace_iterator *iter = filp->private_data;
3006 struct splice_pipe_desc spd = {
3007 .pages = pages,
3008 .partial = partial,
3009 .nr_pages = 0, /* This gets updated below. */
3010 .flags = flags,
3011 .ops = &tracing_pipe_buf_ops,
3012 .spd_release = tracing_spd_release_pipe,
3013 };
3014 static struct tracer *old_tracer;
3015 ssize_t ret;
3016 size_t rem;
3017 unsigned int i;
3018
3019 /* copy the tracer to avoid using a global lock all around */
3020 mutex_lock(&trace_types_lock);
3021 if (unlikely(old_tracer != current_trace && current_trace)) {
3022 old_tracer = current_trace;
3023 *iter->trace = *current_trace;
3024 }
3025 mutex_unlock(&trace_types_lock);
3026
3027 mutex_lock(&iter->mutex);
3028
3029 if (iter->trace->splice_read) {
3030 ret = iter->trace->splice_read(iter, filp,
3031 ppos, pipe, len, flags);
3032 if (ret)
3033 goto out_err;
3034 }
3035
3036 ret = tracing_wait_pipe(filp);
3037 if (ret <= 0)
3038 goto out_err;
3039
3040 if (!iter->ent && !find_next_entry_inc(iter)) {
3041 ret = -EFAULT;
3042 goto out_err;
3043 }
3044
3045 /* Fill as many pages as possible. */
3046 for (i = 0, rem = len; i < PIPE_BUFFERS && rem; i++) {
3047 pages[i] = alloc_page(GFP_KERNEL);
3048 if (!pages[i])
3049 break;
3050
3051 rem = tracing_fill_pipe_page(rem, iter);
3052
3053 /* Copy the data into the page, so we can start over. */
3054 ret = trace_seq_to_buffer(&iter->seq,
3055 page_address(pages[i]),
3056 iter->seq.len);
3057 if (ret < 0) {
3058 __free_page(pages[i]);
3059 break;
3060 }
3061 partial[i].offset = 0;
3062 partial[i].len = iter->seq.len;
3063
3064 trace_seq_init(&iter->seq);
3065 }
3066
3067 mutex_unlock(&iter->mutex);
3068
3069 spd.nr_pages = i;
3070
3071 return splice_to_pipe(pipe, &spd);
3072
3073out_err:
3074 mutex_unlock(&iter->mutex);
3075
3076 return ret;
3077}
3078
3295static ssize_t 3079static ssize_t
3296tracing_entries_read(struct file *filp, char __user *ubuf, 3080tracing_entries_read(struct file *filp, char __user *ubuf,
3297 size_t cnt, loff_t *ppos) 3081 size_t cnt, loff_t *ppos)
3298{ 3082{
3299 struct trace_array *tr = filp->private_data; 3083 struct trace_array *tr = filp->private_data;
3300 char buf[64]; 3084 char buf[96];
3301 int r; 3085 int r;
3302 3086
3303 r = sprintf(buf, "%lu\n", tr->entries >> 10); 3087 mutex_lock(&trace_types_lock);
3088 if (!ring_buffer_expanded)
3089 r = sprintf(buf, "%lu (expanded: %lu)\n",
3090 tr->entries >> 10,
3091 trace_buf_size >> 10);
3092 else
3093 r = sprintf(buf, "%lu\n", tr->entries >> 10);
3094 mutex_unlock(&trace_types_lock);
3095
3304 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r); 3096 return simple_read_from_buffer(ubuf, cnt, ppos, buf, r);
3305} 3097}
3306 3098
@@ -3344,28 +3136,11 @@ tracing_entries_write(struct file *filp, const char __user *ubuf,
3344 val <<= 10; 3136 val <<= 10;
3345 3137
3346 if (val != global_trace.entries) { 3138 if (val != global_trace.entries) {
3347 ret = ring_buffer_resize(global_trace.buffer, val); 3139 ret = tracing_resize_ring_buffer(val);
3348 if (ret < 0) {
3349 cnt = ret;
3350 goto out;
3351 }
3352
3353 ret = ring_buffer_resize(max_tr.buffer, val);
3354 if (ret < 0) { 3140 if (ret < 0) {
3355 int r;
3356 cnt = ret; 3141 cnt = ret;
3357 r = ring_buffer_resize(global_trace.buffer,
3358 global_trace.entries);
3359 if (r < 0) {
3360 /* AARGH! We are left with different
3361 * size max buffer!!!! */
3362 WARN_ON(1);
3363 tracing_disabled = 1;
3364 }
3365 goto out; 3142 goto out;
3366 } 3143 }
3367
3368 global_trace.entries = val;
3369 } 3144 }
3370 3145
3371 filp->f_pos += cnt; 3146 filp->f_pos += cnt;
@@ -3393,7 +3168,7 @@ static int mark_printk(const char *fmt, ...)
3393 int ret; 3168 int ret;
3394 va_list args; 3169 va_list args;
3395 va_start(args, fmt); 3170 va_start(args, fmt);
3396 ret = trace_vprintk(0, -1, fmt, args); 3171 ret = trace_vprintk(0, fmt, args);
3397 va_end(args); 3172 va_end(args);
3398 return ret; 3173 return ret;
3399} 3174}
@@ -3433,42 +3208,288 @@ tracing_mark_write(struct file *filp, const char __user *ubuf,
3433 return cnt; 3208 return cnt;
3434} 3209}
3435 3210
3436static struct file_operations tracing_max_lat_fops = { 3211static const struct file_operations tracing_max_lat_fops = {
3437 .open = tracing_open_generic, 3212 .open = tracing_open_generic,
3438 .read = tracing_max_lat_read, 3213 .read = tracing_max_lat_read,
3439 .write = tracing_max_lat_write, 3214 .write = tracing_max_lat_write,
3440}; 3215};
3441 3216
3442static struct file_operations tracing_ctrl_fops = { 3217static const struct file_operations tracing_ctrl_fops = {
3443 .open = tracing_open_generic, 3218 .open = tracing_open_generic,
3444 .read = tracing_ctrl_read, 3219 .read = tracing_ctrl_read,
3445 .write = tracing_ctrl_write, 3220 .write = tracing_ctrl_write,
3446}; 3221};
3447 3222
3448static struct file_operations set_tracer_fops = { 3223static const struct file_operations set_tracer_fops = {
3449 .open = tracing_open_generic, 3224 .open = tracing_open_generic,
3450 .read = tracing_set_trace_read, 3225 .read = tracing_set_trace_read,
3451 .write = tracing_set_trace_write, 3226 .write = tracing_set_trace_write,
3452}; 3227};
3453 3228
3454static struct file_operations tracing_pipe_fops = { 3229static const struct file_operations tracing_pipe_fops = {
3455 .open = tracing_open_pipe, 3230 .open = tracing_open_pipe,
3456 .poll = tracing_poll_pipe, 3231 .poll = tracing_poll_pipe,
3457 .read = tracing_read_pipe, 3232 .read = tracing_read_pipe,
3233 .splice_read = tracing_splice_read_pipe,
3458 .release = tracing_release_pipe, 3234 .release = tracing_release_pipe,
3459}; 3235};
3460 3236
3461static struct file_operations tracing_entries_fops = { 3237static const struct file_operations tracing_entries_fops = {
3462 .open = tracing_open_generic, 3238 .open = tracing_open_generic,
3463 .read = tracing_entries_read, 3239 .read = tracing_entries_read,
3464 .write = tracing_entries_write, 3240 .write = tracing_entries_write,
3465}; 3241};
3466 3242
3467static struct file_operations tracing_mark_fops = { 3243static const struct file_operations tracing_mark_fops = {
3468 .open = tracing_open_generic, 3244 .open = tracing_open_generic,
3469 .write = tracing_mark_write, 3245 .write = tracing_mark_write,
3470}; 3246};
3471 3247
3248struct ftrace_buffer_info {
3249 struct trace_array *tr;
3250 void *spare;
3251 int cpu;
3252 unsigned int read;
3253};
3254
3255static int tracing_buffers_open(struct inode *inode, struct file *filp)
3256{
3257 int cpu = (int)(long)inode->i_private;
3258 struct ftrace_buffer_info *info;
3259
3260 if (tracing_disabled)
3261 return -ENODEV;
3262
3263 info = kzalloc(sizeof(*info), GFP_KERNEL);
3264 if (!info)
3265 return -ENOMEM;
3266
3267 info->tr = &global_trace;
3268 info->cpu = cpu;
3269 info->spare = ring_buffer_alloc_read_page(info->tr->buffer);
3270 /* Force reading ring buffer for first read */
3271 info->read = (unsigned int)-1;
3272 if (!info->spare)
3273 goto out;
3274
3275 filp->private_data = info;
3276
3277 return 0;
3278
3279 out:
3280 kfree(info);
3281 return -ENOMEM;
3282}
3283
3284static ssize_t
3285tracing_buffers_read(struct file *filp, char __user *ubuf,
3286 size_t count, loff_t *ppos)
3287{
3288 struct ftrace_buffer_info *info = filp->private_data;
3289 unsigned int pos;
3290 ssize_t ret;
3291 size_t size;
3292
3293 if (!count)
3294 return 0;
3295
3296 /* Do we have previous read data to read? */
3297 if (info->read < PAGE_SIZE)
3298 goto read;
3299
3300 info->read = 0;
3301
3302 ret = ring_buffer_read_page(info->tr->buffer,
3303 &info->spare,
3304 count,
3305 info->cpu, 0);
3306 if (ret < 0)
3307 return 0;
3308
3309 pos = ring_buffer_page_len(info->spare);
3310
3311 if (pos < PAGE_SIZE)
3312 memset(info->spare + pos, 0, PAGE_SIZE - pos);
3313
3314read:
3315 size = PAGE_SIZE - info->read;
3316 if (size > count)
3317 size = count;
3318
3319 ret = copy_to_user(ubuf, info->spare + info->read, size);
3320 if (ret == size)
3321 return -EFAULT;
3322 size -= ret;
3323
3324 *ppos += size;
3325 info->read += size;
3326
3327 return size;
3328}
3329
3330static int tracing_buffers_release(struct inode *inode, struct file *file)
3331{
3332 struct ftrace_buffer_info *info = file->private_data;
3333
3334 ring_buffer_free_read_page(info->tr->buffer, info->spare);
3335 kfree(info);
3336
3337 return 0;
3338}
3339
3340struct buffer_ref {
3341 struct ring_buffer *buffer;
3342 void *page;
3343 int ref;
3344};
3345
3346static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
3347 struct pipe_buffer *buf)
3348{
3349 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
3350
3351 if (--ref->ref)
3352 return;
3353
3354 ring_buffer_free_read_page(ref->buffer, ref->page);
3355 kfree(ref);
3356 buf->private = 0;
3357}
3358
3359static int buffer_pipe_buf_steal(struct pipe_inode_info *pipe,
3360 struct pipe_buffer *buf)
3361{
3362 return 1;
3363}
3364
3365static void buffer_pipe_buf_get(struct pipe_inode_info *pipe,
3366 struct pipe_buffer *buf)
3367{
3368 struct buffer_ref *ref = (struct buffer_ref *)buf->private;
3369
3370 ref->ref++;
3371}
3372
3373/* Pipe buffer operations for a buffer. */
3374static struct pipe_buf_operations buffer_pipe_buf_ops = {
3375 .can_merge = 0,
3376 .map = generic_pipe_buf_map,
3377 .unmap = generic_pipe_buf_unmap,
3378 .confirm = generic_pipe_buf_confirm,
3379 .release = buffer_pipe_buf_release,
3380 .steal = buffer_pipe_buf_steal,
3381 .get = buffer_pipe_buf_get,
3382};
3383
3384/*
3385 * Callback from splice_to_pipe(), if we need to release some pages
3386 * at the end of the spd in case we error'ed out in filling the pipe.
3387 */
3388static void buffer_spd_release(struct splice_pipe_desc *spd, unsigned int i)
3389{
3390 struct buffer_ref *ref =
3391 (struct buffer_ref *)spd->partial[i].private;
3392
3393 if (--ref->ref)
3394 return;
3395
3396 ring_buffer_free_read_page(ref->buffer, ref->page);
3397 kfree(ref);
3398 spd->partial[i].private = 0;
3399}
3400
3401static ssize_t
3402tracing_buffers_splice_read(struct file *file, loff_t *ppos,
3403 struct pipe_inode_info *pipe, size_t len,
3404 unsigned int flags)
3405{
3406 struct ftrace_buffer_info *info = file->private_data;
3407 struct partial_page partial[PIPE_BUFFERS];
3408 struct page *pages[PIPE_BUFFERS];
3409 struct splice_pipe_desc spd = {
3410 .pages = pages,
3411 .partial = partial,
3412 .flags = flags,
3413 .ops = &buffer_pipe_buf_ops,
3414 .spd_release = buffer_spd_release,
3415 };
3416 struct buffer_ref *ref;
3417 int size, i;
3418 size_t ret;
3419
3420 /*
3421 * We can't seek on a buffer input
3422 */
3423 if (unlikely(*ppos))
3424 return -ESPIPE;
3425
3426
3427 for (i = 0; i < PIPE_BUFFERS && len; i++, len -= size) {
3428 struct page *page;
3429 int r;
3430
3431 ref = kzalloc(sizeof(*ref), GFP_KERNEL);
3432 if (!ref)
3433 break;
3434
3435 ref->buffer = info->tr->buffer;
3436 ref->page = ring_buffer_alloc_read_page(ref->buffer);
3437 if (!ref->page) {
3438 kfree(ref);
3439 break;
3440 }
3441
3442 r = ring_buffer_read_page(ref->buffer, &ref->page,
3443 len, info->cpu, 0);
3444 if (r < 0) {
3445 ring_buffer_free_read_page(ref->buffer,
3446 ref->page);
3447 kfree(ref);
3448 break;
3449 }
3450
3451 /*
3452 * zero out any left over data, this is going to
3453 * user land.
3454 */
3455 size = ring_buffer_page_len(ref->page);
3456 if (size < PAGE_SIZE)
3457 memset(ref->page + size, 0, PAGE_SIZE - size);
3458
3459 page = virt_to_page(ref->page);
3460
3461 spd.pages[i] = page;
3462 spd.partial[i].len = PAGE_SIZE;
3463 spd.partial[i].offset = 0;
3464 spd.partial[i].private = (unsigned long)ref;
3465 spd.nr_pages++;
3466 }
3467
3468 spd.nr_pages = i;
3469
3470 /* did we read anything? */
3471 if (!spd.nr_pages) {
3472 if (flags & SPLICE_F_NONBLOCK)
3473 ret = -EAGAIN;
3474 else
3475 ret = 0;
3476 /* TODO: block */
3477 return ret;
3478 }
3479
3480 ret = splice_to_pipe(pipe, &spd);
3481
3482 return ret;
3483}
3484
3485static const struct file_operations tracing_buffers_fops = {
3486 .open = tracing_buffers_open,
3487 .read = tracing_buffers_read,
3488 .release = tracing_buffers_release,
3489 .splice_read = tracing_buffers_splice_read,
3490 .llseek = no_llseek,
3491};
3492
3472#ifdef CONFIG_DYNAMIC_FTRACE 3493#ifdef CONFIG_DYNAMIC_FTRACE
3473 3494
3474int __weak ftrace_arch_read_dyn_info(char *buf, int size) 3495int __weak ftrace_arch_read_dyn_info(char *buf, int size)
@@ -3500,7 +3521,7 @@ tracing_read_dyn_info(struct file *filp, char __user *ubuf,
3500 return r; 3521 return r;
3501} 3522}
3502 3523
3503static struct file_operations tracing_dyn_info_fops = { 3524static const struct file_operations tracing_dyn_info_fops = {
3504 .open = tracing_open_generic, 3525 .open = tracing_open_generic,
3505 .read = tracing_read_dyn_info, 3526 .read = tracing_read_dyn_info,
3506}; 3527};
@@ -3515,6 +3536,9 @@ struct dentry *tracing_init_dentry(void)
3515 if (d_tracer) 3536 if (d_tracer)
3516 return d_tracer; 3537 return d_tracer;
3517 3538
3539 if (!debugfs_initialized())
3540 return NULL;
3541
3518 d_tracer = debugfs_create_dir("tracing", NULL); 3542 d_tracer = debugfs_create_dir("tracing", NULL);
3519 3543
3520 if (!d_tracer && !once) { 3544 if (!d_tracer && !once) {
@@ -3526,15 +3550,350 @@ struct dentry *tracing_init_dentry(void)
3526 return d_tracer; 3550 return d_tracer;
3527} 3551}
3528 3552
3553static struct dentry *d_percpu;
3554
3555struct dentry *tracing_dentry_percpu(void)
3556{
3557 static int once;
3558 struct dentry *d_tracer;
3559
3560 if (d_percpu)
3561 return d_percpu;
3562
3563 d_tracer = tracing_init_dentry();
3564
3565 if (!d_tracer)
3566 return NULL;
3567
3568 d_percpu = debugfs_create_dir("per_cpu", d_tracer);
3569
3570 if (!d_percpu && !once) {
3571 once = 1;
3572 pr_warning("Could not create debugfs directory 'per_cpu'\n");
3573 return NULL;
3574 }
3575
3576 return d_percpu;
3577}
3578
3579static void tracing_init_debugfs_percpu(long cpu)
3580{
3581 struct dentry *d_percpu = tracing_dentry_percpu();
3582 struct dentry *entry, *d_cpu;
3583 /* strlen(cpu) + MAX(log10(cpu)) + '\0' */
3584 char cpu_dir[7];
3585
3586 if (cpu > 999 || cpu < 0)
3587 return;
3588
3589 sprintf(cpu_dir, "cpu%ld", cpu);
3590 d_cpu = debugfs_create_dir(cpu_dir, d_percpu);
3591 if (!d_cpu) {
3592 pr_warning("Could not create debugfs '%s' entry\n", cpu_dir);
3593 return;
3594 }
3595
3596 /* per cpu trace_pipe */
3597 entry = debugfs_create_file("trace_pipe", 0444, d_cpu,
3598 (void *) cpu, &tracing_pipe_fops);
3599 if (!entry)
3600 pr_warning("Could not create debugfs 'trace_pipe' entry\n");
3601
3602 /* per cpu trace */
3603 entry = debugfs_create_file("trace", 0644, d_cpu,
3604 (void *) cpu, &tracing_fops);
3605 if (!entry)
3606 pr_warning("Could not create debugfs 'trace' entry\n");
3607
3608 entry = debugfs_create_file("trace_pipe_raw", 0444, d_cpu,
3609 (void *) cpu, &tracing_buffers_fops);
3610 if (!entry)
3611 pr_warning("Could not create debugfs 'trace_pipe_raw' entry\n");
3612}
3613
3529#ifdef CONFIG_FTRACE_SELFTEST 3614#ifdef CONFIG_FTRACE_SELFTEST
3530/* Let selftest have access to static functions in this file */ 3615/* Let selftest have access to static functions in this file */
3531#include "trace_selftest.c" 3616#include "trace_selftest.c"
3532#endif 3617#endif
3533 3618
3619struct trace_option_dentry {
3620 struct tracer_opt *opt;
3621 struct tracer_flags *flags;
3622 struct dentry *entry;
3623};
3624
3625static ssize_t
3626trace_options_read(struct file *filp, char __user *ubuf, size_t cnt,
3627 loff_t *ppos)
3628{
3629 struct trace_option_dentry *topt = filp->private_data;
3630 char *buf;
3631
3632 if (topt->flags->val & topt->opt->bit)
3633 buf = "1\n";
3634 else
3635 buf = "0\n";
3636
3637 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
3638}
3639
3640static ssize_t
3641trace_options_write(struct file *filp, const char __user *ubuf, size_t cnt,
3642 loff_t *ppos)
3643{
3644 struct trace_option_dentry *topt = filp->private_data;
3645 unsigned long val;
3646 char buf[64];
3647 int ret;
3648
3649 if (cnt >= sizeof(buf))
3650 return -EINVAL;
3651
3652 if (copy_from_user(&buf, ubuf, cnt))
3653 return -EFAULT;
3654
3655 buf[cnt] = 0;
3656
3657 ret = strict_strtoul(buf, 10, &val);
3658 if (ret < 0)
3659 return ret;
3660
3661 ret = 0;
3662 switch (val) {
3663 case 0:
3664 /* do nothing if already cleared */
3665 if (!(topt->flags->val & topt->opt->bit))
3666 break;
3667
3668 mutex_lock(&trace_types_lock);
3669 if (current_trace->set_flag)
3670 ret = current_trace->set_flag(topt->flags->val,
3671 topt->opt->bit, 0);
3672 mutex_unlock(&trace_types_lock);
3673 if (ret)
3674 return ret;
3675 topt->flags->val &= ~topt->opt->bit;
3676 break;
3677 case 1:
3678 /* do nothing if already set */
3679 if (topt->flags->val & topt->opt->bit)
3680 break;
3681
3682 mutex_lock(&trace_types_lock);
3683 if (current_trace->set_flag)
3684 ret = current_trace->set_flag(topt->flags->val,
3685 topt->opt->bit, 1);
3686 mutex_unlock(&trace_types_lock);
3687 if (ret)
3688 return ret;
3689 topt->flags->val |= topt->opt->bit;
3690 break;
3691
3692 default:
3693 return -EINVAL;
3694 }
3695
3696 *ppos += cnt;
3697
3698 return cnt;
3699}
3700
3701
3702static const struct file_operations trace_options_fops = {
3703 .open = tracing_open_generic,
3704 .read = trace_options_read,
3705 .write = trace_options_write,
3706};
3707
3708static ssize_t
3709trace_options_core_read(struct file *filp, char __user *ubuf, size_t cnt,
3710 loff_t *ppos)
3711{
3712 long index = (long)filp->private_data;
3713 char *buf;
3714
3715 if (trace_flags & (1 << index))
3716 buf = "1\n";
3717 else
3718 buf = "0\n";
3719
3720 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
3721}
3722
3723static ssize_t
3724trace_options_core_write(struct file *filp, const char __user *ubuf, size_t cnt,
3725 loff_t *ppos)
3726{
3727 long index = (long)filp->private_data;
3728 char buf[64];
3729 unsigned long val;
3730 int ret;
3731
3732 if (cnt >= sizeof(buf))
3733 return -EINVAL;
3734
3735 if (copy_from_user(&buf, ubuf, cnt))
3736 return -EFAULT;
3737
3738 buf[cnt] = 0;
3739
3740 ret = strict_strtoul(buf, 10, &val);
3741 if (ret < 0)
3742 return ret;
3743
3744 switch (val) {
3745 case 0:
3746 trace_flags &= ~(1 << index);
3747 break;
3748 case 1:
3749 trace_flags |= 1 << index;
3750 break;
3751
3752 default:
3753 return -EINVAL;
3754 }
3755
3756 *ppos += cnt;
3757
3758 return cnt;
3759}
3760
3761static const struct file_operations trace_options_core_fops = {
3762 .open = tracing_open_generic,
3763 .read = trace_options_core_read,
3764 .write = trace_options_core_write,
3765};
3766
3767static struct dentry *trace_options_init_dentry(void)
3768{
3769 struct dentry *d_tracer;
3770 static struct dentry *t_options;
3771
3772 if (t_options)
3773 return t_options;
3774
3775 d_tracer = tracing_init_dentry();
3776 if (!d_tracer)
3777 return NULL;
3778
3779 t_options = debugfs_create_dir("options", d_tracer);
3780 if (!t_options) {
3781 pr_warning("Could not create debugfs directory 'options'\n");
3782 return NULL;
3783 }
3784
3785 return t_options;
3786}
3787
3788static void
3789create_trace_option_file(struct trace_option_dentry *topt,
3790 struct tracer_flags *flags,
3791 struct tracer_opt *opt)
3792{
3793 struct dentry *t_options;
3794 struct dentry *entry;
3795
3796 t_options = trace_options_init_dentry();
3797 if (!t_options)
3798 return;
3799
3800 topt->flags = flags;
3801 topt->opt = opt;
3802
3803 entry = debugfs_create_file(opt->name, 0644, t_options, topt,
3804 &trace_options_fops);
3805
3806 topt->entry = entry;
3807
3808}
3809
3810static struct trace_option_dentry *
3811create_trace_option_files(struct tracer *tracer)
3812{
3813 struct trace_option_dentry *topts;
3814 struct tracer_flags *flags;
3815 struct tracer_opt *opts;
3816 int cnt;
3817
3818 if (!tracer)
3819 return NULL;
3820
3821 flags = tracer->flags;
3822
3823 if (!flags || !flags->opts)
3824 return NULL;
3825
3826 opts = flags->opts;
3827
3828 for (cnt = 0; opts[cnt].name; cnt++)
3829 ;
3830
3831 topts = kcalloc(cnt + 1, sizeof(*topts), GFP_KERNEL);
3832 if (!topts)
3833 return NULL;
3834
3835 for (cnt = 0; opts[cnt].name; cnt++)
3836 create_trace_option_file(&topts[cnt], flags,
3837 &opts[cnt]);
3838
3839 return topts;
3840}
3841
3842static void
3843destroy_trace_option_files(struct trace_option_dentry *topts)
3844{
3845 int cnt;
3846
3847 if (!topts)
3848 return;
3849
3850 for (cnt = 0; topts[cnt].opt; cnt++) {
3851 if (topts[cnt].entry)
3852 debugfs_remove(topts[cnt].entry);
3853 }
3854
3855 kfree(topts);
3856}
3857
3858static struct dentry *
3859create_trace_option_core_file(const char *option, long index)
3860{
3861 struct dentry *t_options;
3862 struct dentry *entry;
3863
3864 t_options = trace_options_init_dentry();
3865 if (!t_options)
3866 return NULL;
3867
3868 entry = debugfs_create_file(option, 0644, t_options, (void *)index,
3869 &trace_options_core_fops);
3870
3871 return entry;
3872}
3873
3874static __init void create_trace_options_dir(void)
3875{
3876 struct dentry *t_options;
3877 struct dentry *entry;
3878 int i;
3879
3880 t_options = trace_options_init_dentry();
3881 if (!t_options)
3882 return;
3883
3884 for (i = 0; trace_options[i]; i++) {
3885 entry = create_trace_option_core_file(trace_options[i], i);
3886 if (!entry)
3887 pr_warning("Could not create debugfs %s entry\n",
3888 trace_options[i]);
3889 }
3890}
3891
3534static __init int tracer_init_debugfs(void) 3892static __init int tracer_init_debugfs(void)
3535{ 3893{
3536 struct dentry *d_tracer; 3894 struct dentry *d_tracer;
3537 struct dentry *entry; 3895 struct dentry *entry;
3896 int cpu;
3538 3897
3539 d_tracer = tracing_init_dentry(); 3898 d_tracer = tracing_init_dentry();
3540 3899
@@ -3548,18 +3907,15 @@ static __init int tracer_init_debugfs(void)
3548 if (!entry) 3907 if (!entry)
3549 pr_warning("Could not create debugfs 'trace_options' entry\n"); 3908 pr_warning("Could not create debugfs 'trace_options' entry\n");
3550 3909
3910 create_trace_options_dir();
3911
3551 entry = debugfs_create_file("tracing_cpumask", 0644, d_tracer, 3912 entry = debugfs_create_file("tracing_cpumask", 0644, d_tracer,
3552 NULL, &tracing_cpumask_fops); 3913 NULL, &tracing_cpumask_fops);
3553 if (!entry) 3914 if (!entry)
3554 pr_warning("Could not create debugfs 'tracing_cpumask' entry\n"); 3915 pr_warning("Could not create debugfs 'tracing_cpumask' entry\n");
3555 3916
3556 entry = debugfs_create_file("latency_trace", 0444, d_tracer, 3917 entry = debugfs_create_file("trace", 0644, d_tracer,
3557 &global_trace, &tracing_lt_fops); 3918 (void *) TRACE_PIPE_ALL_CPU, &tracing_fops);
3558 if (!entry)
3559 pr_warning("Could not create debugfs 'latency_trace' entry\n");
3560
3561 entry = debugfs_create_file("trace", 0444, d_tracer,
3562 &global_trace, &tracing_fops);
3563 if (!entry) 3919 if (!entry)
3564 pr_warning("Could not create debugfs 'trace' entry\n"); 3920 pr_warning("Could not create debugfs 'trace' entry\n");
3565 3921
@@ -3590,8 +3946,8 @@ static __init int tracer_init_debugfs(void)
3590 if (!entry) 3946 if (!entry)
3591 pr_warning("Could not create debugfs 'README' entry\n"); 3947 pr_warning("Could not create debugfs 'README' entry\n");
3592 3948
3593 entry = debugfs_create_file("trace_pipe", 0644, d_tracer, 3949 entry = debugfs_create_file("trace_pipe", 0444, d_tracer,
3594 NULL, &tracing_pipe_fops); 3950 (void *) TRACE_PIPE_ALL_CPU, &tracing_pipe_fops);
3595 if (!entry) 3951 if (!entry)
3596 pr_warning("Could not create debugfs " 3952 pr_warning("Could not create debugfs "
3597 "'trace_pipe' entry\n"); 3953 "'trace_pipe' entry\n");
@@ -3619,77 +3975,12 @@ static __init int tracer_init_debugfs(void)
3619#ifdef CONFIG_SYSPROF_TRACER 3975#ifdef CONFIG_SYSPROF_TRACER
3620 init_tracer_sysprof_debugfs(d_tracer); 3976 init_tracer_sysprof_debugfs(d_tracer);
3621#endif 3977#endif
3622 return 0;
3623}
3624
3625int trace_vprintk(unsigned long ip, int depth, const char *fmt, va_list args)
3626{
3627 static DEFINE_SPINLOCK(trace_buf_lock);
3628 static char trace_buf[TRACE_BUF_SIZE];
3629
3630 struct ring_buffer_event *event;
3631 struct trace_array *tr = &global_trace;
3632 struct trace_array_cpu *data;
3633 int cpu, len = 0, size, pc;
3634 struct print_entry *entry;
3635 unsigned long irq_flags;
3636
3637 if (tracing_disabled || tracing_selftest_running)
3638 return 0;
3639
3640 pc = preempt_count();
3641 preempt_disable_notrace();
3642 cpu = raw_smp_processor_id();
3643 data = tr->data[cpu];
3644
3645 if (unlikely(atomic_read(&data->disabled)))
3646 goto out;
3647
3648 pause_graph_tracing();
3649 spin_lock_irqsave(&trace_buf_lock, irq_flags);
3650 len = vsnprintf(trace_buf, TRACE_BUF_SIZE, fmt, args);
3651
3652 len = min(len, TRACE_BUF_SIZE-1);
3653 trace_buf[len] = 0;
3654
3655 size = sizeof(*entry) + len + 1;
3656 event = ring_buffer_lock_reserve(tr->buffer, size, &irq_flags);
3657 if (!event)
3658 goto out_unlock;
3659 entry = ring_buffer_event_data(event);
3660 tracing_generic_entry_update(&entry->ent, irq_flags, pc);
3661 entry->ent.type = TRACE_PRINT;
3662 entry->ip = ip;
3663 entry->depth = depth;
3664
3665 memcpy(&entry->buf, trace_buf, len);
3666 entry->buf[len] = 0;
3667 ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
3668
3669 out_unlock:
3670 spin_unlock_irqrestore(&trace_buf_lock, irq_flags);
3671 unpause_graph_tracing();
3672 out:
3673 preempt_enable_notrace();
3674
3675 return len;
3676}
3677EXPORT_SYMBOL_GPL(trace_vprintk);
3678 3978
3679int __ftrace_printk(unsigned long ip, const char *fmt, ...) 3979 for_each_tracing_cpu(cpu)
3680{ 3980 tracing_init_debugfs_percpu(cpu);
3681 int ret;
3682 va_list ap;
3683
3684 if (!(trace_flags & TRACE_ITER_PRINTK))
3685 return 0;
3686 3981
3687 va_start(ap, fmt); 3982 return 0;
3688 ret = trace_vprintk(ip, task_curr_ret_stack(current), fmt, ap);
3689 va_end(ap);
3690 return ret;
3691} 3983}
3692EXPORT_SYMBOL_GPL(__ftrace_printk);
3693 3984
3694static int trace_panic_handler(struct notifier_block *this, 3985static int trace_panic_handler(struct notifier_block *this,
3695 unsigned long event, void *unused) 3986 unsigned long event, void *unused)
@@ -3750,14 +4041,15 @@ trace_printk_seq(struct trace_seq *s)
3750 4041
3751 printk(KERN_TRACE "%s", s->buffer); 4042 printk(KERN_TRACE "%s", s->buffer);
3752 4043
3753 trace_seq_reset(s); 4044 trace_seq_init(s);
3754} 4045}
3755 4046
3756void ftrace_dump(void) 4047static void __ftrace_dump(bool disable_tracing)
3757{ 4048{
3758 static DEFINE_SPINLOCK(ftrace_dump_lock); 4049 static DEFINE_SPINLOCK(ftrace_dump_lock);
3759 /* use static because iter can be a bit big for the stack */ 4050 /* use static because iter can be a bit big for the stack */
3760 static struct trace_iterator iter; 4051 static struct trace_iterator iter;
4052 unsigned int old_userobj;
3761 static int dump_ran; 4053 static int dump_ran;
3762 unsigned long flags; 4054 unsigned long flags;
3763 int cnt = 0, cpu; 4055 int cnt = 0, cpu;
@@ -3769,21 +4061,26 @@ void ftrace_dump(void)
3769 4061
3770 dump_ran = 1; 4062 dump_ran = 1;
3771 4063
3772 /* No turning back! */
3773 tracing_off(); 4064 tracing_off();
3774 ftrace_kill(); 4065
4066 if (disable_tracing)
4067 ftrace_kill();
3775 4068
3776 for_each_tracing_cpu(cpu) { 4069 for_each_tracing_cpu(cpu) {
3777 atomic_inc(&global_trace.data[cpu]->disabled); 4070 atomic_inc(&global_trace.data[cpu]->disabled);
3778 } 4071 }
3779 4072
4073 old_userobj = trace_flags & TRACE_ITER_SYM_USEROBJ;
4074
3780 /* don't look at user memory in panic mode */ 4075 /* don't look at user memory in panic mode */
3781 trace_flags &= ~TRACE_ITER_SYM_USEROBJ; 4076 trace_flags &= ~TRACE_ITER_SYM_USEROBJ;
3782 4077
3783 printk(KERN_TRACE "Dumping ftrace buffer:\n"); 4078 printk(KERN_TRACE "Dumping ftrace buffer:\n");
3784 4079
4080 /* Simulate the iterator */
3785 iter.tr = &global_trace; 4081 iter.tr = &global_trace;
3786 iter.trace = current_trace; 4082 iter.trace = current_trace;
4083 iter.cpu_file = TRACE_PIPE_ALL_CPU;
3787 4084
3788 /* 4085 /*
3789 * We need to stop all tracing on all CPUS to read the 4086 * We need to stop all tracing on all CPUS to read the
@@ -3819,13 +4116,30 @@ void ftrace_dump(void)
3819 else 4116 else
3820 printk(KERN_TRACE "---------------------------------\n"); 4117 printk(KERN_TRACE "---------------------------------\n");
3821 4118
4119 /* Re-enable tracing if requested */
4120 if (!disable_tracing) {
4121 trace_flags |= old_userobj;
4122
4123 for_each_tracing_cpu(cpu) {
4124 atomic_dec(&global_trace.data[cpu]->disabled);
4125 }
4126 tracing_on();
4127 }
4128
3822 out: 4129 out:
3823 spin_unlock_irqrestore(&ftrace_dump_lock, flags); 4130 spin_unlock_irqrestore(&ftrace_dump_lock, flags);
3824} 4131}
3825 4132
4133/* By default: disable tracing after the dump */
4134void ftrace_dump(void)
4135{
4136 __ftrace_dump(true);
4137}
4138
3826__init static int tracer_alloc_buffers(void) 4139__init static int tracer_alloc_buffers(void)
3827{ 4140{
3828 struct trace_array_cpu *data; 4141 struct trace_array_cpu *data;
4142 int ring_buf_size;
3829 int i; 4143 int i;
3830 int ret = -ENOMEM; 4144 int ret = -ENOMEM;
3831 4145
@@ -3835,11 +4149,21 @@ __init static int tracer_alloc_buffers(void)
3835 if (!alloc_cpumask_var(&tracing_cpumask, GFP_KERNEL)) 4149 if (!alloc_cpumask_var(&tracing_cpumask, GFP_KERNEL))
3836 goto out_free_buffer_mask; 4150 goto out_free_buffer_mask;
3837 4151
4152 if (!alloc_cpumask_var(&tracing_reader_cpumask, GFP_KERNEL))
4153 goto out_free_tracing_cpumask;
4154
4155 /* To save memory, keep the ring buffer size to its minimum */
4156 if (ring_buffer_expanded)
4157 ring_buf_size = trace_buf_size;
4158 else
4159 ring_buf_size = 1;
4160
3838 cpumask_copy(tracing_buffer_mask, cpu_possible_mask); 4161 cpumask_copy(tracing_buffer_mask, cpu_possible_mask);
3839 cpumask_copy(tracing_cpumask, cpu_all_mask); 4162 cpumask_copy(tracing_cpumask, cpu_all_mask);
4163 cpumask_clear(tracing_reader_cpumask);
3840 4164
3841 /* TODO: make the number of buffers hot pluggable with CPUS */ 4165 /* TODO: make the number of buffers hot pluggable with CPUS */
3842 global_trace.buffer = ring_buffer_alloc(trace_buf_size, 4166 global_trace.buffer = ring_buffer_alloc(ring_buf_size,
3843 TRACE_BUFFER_FLAGS); 4167 TRACE_BUFFER_FLAGS);
3844 if (!global_trace.buffer) { 4168 if (!global_trace.buffer) {
3845 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n"); 4169 printk(KERN_ERR "tracer: failed to allocate ring buffer!\n");
@@ -3850,7 +4174,7 @@ __init static int tracer_alloc_buffers(void)
3850 4174
3851 4175
3852#ifdef CONFIG_TRACER_MAX_TRACE 4176#ifdef CONFIG_TRACER_MAX_TRACE
3853 max_tr.buffer = ring_buffer_alloc(trace_buf_size, 4177 max_tr.buffer = ring_buffer_alloc(ring_buf_size,
3854 TRACE_BUFFER_FLAGS); 4178 TRACE_BUFFER_FLAGS);
3855 if (!max_tr.buffer) { 4179 if (!max_tr.buffer) {
3856 printk(KERN_ERR "tracer: failed to allocate max ring buffer!\n"); 4180 printk(KERN_ERR "tracer: failed to allocate max ring buffer!\n");
@@ -3871,14 +4195,10 @@ __init static int tracer_alloc_buffers(void)
3871 trace_init_cmdlines(); 4195 trace_init_cmdlines();
3872 4196
3873 register_tracer(&nop_trace); 4197 register_tracer(&nop_trace);
4198 current_trace = &nop_trace;
3874#ifdef CONFIG_BOOT_TRACER 4199#ifdef CONFIG_BOOT_TRACER
3875 register_tracer(&boot_tracer); 4200 register_tracer(&boot_tracer);
3876 current_trace = &boot_tracer;
3877 current_trace->init(&global_trace);
3878#else
3879 current_trace = &nop_trace;
3880#endif 4201#endif
3881
3882 /* All seems OK, enable tracing */ 4202 /* All seems OK, enable tracing */
3883 tracing_disabled = 0; 4203 tracing_disabled = 0;
3884 4204
@@ -3886,14 +4206,38 @@ __init static int tracer_alloc_buffers(void)
3886 &trace_panic_notifier); 4206 &trace_panic_notifier);
3887 4207
3888 register_die_notifier(&trace_die_notifier); 4208 register_die_notifier(&trace_die_notifier);
3889 ret = 0; 4209
4210 return 0;
3890 4211
3891out_free_cpumask: 4212out_free_cpumask:
4213 free_cpumask_var(tracing_reader_cpumask);
4214out_free_tracing_cpumask:
3892 free_cpumask_var(tracing_cpumask); 4215 free_cpumask_var(tracing_cpumask);
3893out_free_buffer_mask: 4216out_free_buffer_mask:
3894 free_cpumask_var(tracing_buffer_mask); 4217 free_cpumask_var(tracing_buffer_mask);
3895out: 4218out:
3896 return ret; 4219 return ret;
3897} 4220}
4221
4222__init static int clear_boot_tracer(void)
4223{
4224 /*
4225 * The default tracer at boot buffer is an init section.
4226 * This function is called in lateinit. If we did not
4227 * find the boot tracer, then clear it out, to prevent
4228 * later registration from accessing the buffer that is
4229 * about to be freed.
4230 */
4231 if (!default_bootup_tracer)
4232 return 0;
4233
4234 printk(KERN_INFO "ftrace bootup tracer '%s' not registered.\n",
4235 default_bootup_tracer);
4236 default_bootup_tracer = NULL;
4237
4238 return 0;
4239}
4240
3898early_initcall(tracer_alloc_buffers); 4241early_initcall(tracer_alloc_buffers);
3899fs_initcall(tracer_init_debugfs); 4242fs_initcall(tracer_init_debugfs);
4243late_initcall(clear_boot_tracer);
diff --git a/kernel/trace/trace.h b/kernel/trace/trace.h
index 4d3d381bfd95..cb0ce3fc36d3 100644
--- a/kernel/trace/trace.h
+++ b/kernel/trace/trace.h
@@ -9,6 +9,8 @@
9#include <linux/mmiotrace.h> 9#include <linux/mmiotrace.h>
10#include <linux/ftrace.h> 10#include <linux/ftrace.h>
11#include <trace/boot.h> 11#include <trace/boot.h>
12#include <trace/kmemtrace.h>
13#include <trace/power.h>
12 14
13enum trace_type { 15enum trace_type {
14 __TRACE_FIRST_TYPE = 0, 16 __TRACE_FIRST_TYPE = 0,
@@ -16,9 +18,9 @@ enum trace_type {
16 TRACE_FN, 18 TRACE_FN,
17 TRACE_CTX, 19 TRACE_CTX,
18 TRACE_WAKE, 20 TRACE_WAKE,
19 TRACE_CONT,
20 TRACE_STACK, 21 TRACE_STACK,
21 TRACE_PRINT, 22 TRACE_PRINT,
23 TRACE_BPRINT,
22 TRACE_SPECIAL, 24 TRACE_SPECIAL,
23 TRACE_MMIO_RW, 25 TRACE_MMIO_RW,
24 TRACE_MMIO_MAP, 26 TRACE_MMIO_MAP,
@@ -29,9 +31,14 @@ enum trace_type {
29 TRACE_GRAPH_ENT, 31 TRACE_GRAPH_ENT,
30 TRACE_USER_STACK, 32 TRACE_USER_STACK,
31 TRACE_HW_BRANCHES, 33 TRACE_HW_BRANCHES,
34 TRACE_SYSCALL_ENTER,
35 TRACE_SYSCALL_EXIT,
36 TRACE_KMEM_ALLOC,
37 TRACE_KMEM_FREE,
32 TRACE_POWER, 38 TRACE_POWER,
39 TRACE_BLK,
33 40
34 __TRACE_LAST_TYPE 41 __TRACE_LAST_TYPE,
35}; 42};
36 43
37/* 44/*
@@ -42,7 +49,6 @@ enum trace_type {
42 */ 49 */
43struct trace_entry { 50struct trace_entry {
44 unsigned char type; 51 unsigned char type;
45 unsigned char cpu;
46 unsigned char flags; 52 unsigned char flags;
47 unsigned char preempt_count; 53 unsigned char preempt_count;
48 int pid; 54 int pid;
@@ -60,13 +66,13 @@ struct ftrace_entry {
60 66
61/* Function call entry */ 67/* Function call entry */
62struct ftrace_graph_ent_entry { 68struct ftrace_graph_ent_entry {
63 struct trace_entry ent; 69 struct trace_entry ent;
64 struct ftrace_graph_ent graph_ent; 70 struct ftrace_graph_ent graph_ent;
65}; 71};
66 72
67/* Function return entry */ 73/* Function return entry */
68struct ftrace_graph_ret_entry { 74struct ftrace_graph_ret_entry {
69 struct trace_entry ent; 75 struct trace_entry ent;
70 struct ftrace_graph_ret ret; 76 struct ftrace_graph_ret ret;
71}; 77};
72extern struct tracer boot_tracer; 78extern struct tracer boot_tracer;
@@ -112,12 +118,18 @@ struct userstack_entry {
112}; 118};
113 119
114/* 120/*
115 * ftrace_printk entry: 121 * trace_printk entry:
116 */ 122 */
123struct bprint_entry {
124 struct trace_entry ent;
125 unsigned long ip;
126 const char *fmt;
127 u32 buf[];
128};
129
117struct print_entry { 130struct print_entry {
118 struct trace_entry ent; 131 struct trace_entry ent;
119 unsigned long ip; 132 unsigned long ip;
120 int depth;
121 char buf[]; 133 char buf[];
122}; 134};
123 135
@@ -170,15 +182,45 @@ struct trace_power {
170 struct power_trace state_data; 182 struct power_trace state_data;
171}; 183};
172 184
185struct kmemtrace_alloc_entry {
186 struct trace_entry ent;
187 enum kmemtrace_type_id type_id;
188 unsigned long call_site;
189 const void *ptr;
190 size_t bytes_req;
191 size_t bytes_alloc;
192 gfp_t gfp_flags;
193 int node;
194};
195
196struct kmemtrace_free_entry {
197 struct trace_entry ent;
198 enum kmemtrace_type_id type_id;
199 unsigned long call_site;
200 const void *ptr;
201};
202
203struct syscall_trace_enter {
204 struct trace_entry ent;
205 int nr;
206 unsigned long args[];
207};
208
209struct syscall_trace_exit {
210 struct trace_entry ent;
211 int nr;
212 unsigned long ret;
213};
214
215
173/* 216/*
174 * trace_flag_type is an enumeration that holds different 217 * trace_flag_type is an enumeration that holds different
175 * states when a trace occurs. These are: 218 * states when a trace occurs. These are:
176 * IRQS_OFF - interrupts were disabled 219 * IRQS_OFF - interrupts were disabled
177 * IRQS_NOSUPPORT - arch does not support irqs_disabled_flags 220 * IRQS_NOSUPPORT - arch does not support irqs_disabled_flags
178 * NEED_RESCED - reschedule is requested 221 * NEED_RESCED - reschedule is requested
179 * HARDIRQ - inside an interrupt handler 222 * HARDIRQ - inside an interrupt handler
180 * SOFTIRQ - inside a softirq handler 223 * SOFTIRQ - inside a softirq handler
181 * CONT - multiple entries hold the trace item
182 */ 224 */
183enum trace_flag_type { 225enum trace_flag_type {
184 TRACE_FLAG_IRQS_OFF = 0x01, 226 TRACE_FLAG_IRQS_OFF = 0x01,
@@ -186,7 +228,6 @@ enum trace_flag_type {
186 TRACE_FLAG_NEED_RESCHED = 0x04, 228 TRACE_FLAG_NEED_RESCHED = 0x04,
187 TRACE_FLAG_HARDIRQ = 0x08, 229 TRACE_FLAG_HARDIRQ = 0x08,
188 TRACE_FLAG_SOFTIRQ = 0x10, 230 TRACE_FLAG_SOFTIRQ = 0x10,
189 TRACE_FLAG_CONT = 0x20,
190}; 231};
191 232
192#define TRACE_BUF_SIZE 1024 233#define TRACE_BUF_SIZE 1024
@@ -198,6 +239,7 @@ enum trace_flag_type {
198 */ 239 */
199struct trace_array_cpu { 240struct trace_array_cpu {
200 atomic_t disabled; 241 atomic_t disabled;
242 void *buffer_page; /* ring buffer spare */
201 243
202 /* these fields get copied into max-trace: */ 244 /* these fields get copied into max-trace: */
203 unsigned long trace_idx; 245 unsigned long trace_idx;
@@ -262,10 +304,10 @@ extern void __ftrace_bad_type(void);
262 do { \ 304 do { \
263 IF_ASSIGN(var, ent, struct ftrace_entry, TRACE_FN); \ 305 IF_ASSIGN(var, ent, struct ftrace_entry, TRACE_FN); \
264 IF_ASSIGN(var, ent, struct ctx_switch_entry, 0); \ 306 IF_ASSIGN(var, ent, struct ctx_switch_entry, 0); \
265 IF_ASSIGN(var, ent, struct trace_field_cont, TRACE_CONT); \
266 IF_ASSIGN(var, ent, struct stack_entry, TRACE_STACK); \ 307 IF_ASSIGN(var, ent, struct stack_entry, TRACE_STACK); \
267 IF_ASSIGN(var, ent, struct userstack_entry, TRACE_USER_STACK);\ 308 IF_ASSIGN(var, ent, struct userstack_entry, TRACE_USER_STACK);\
268 IF_ASSIGN(var, ent, struct print_entry, TRACE_PRINT); \ 309 IF_ASSIGN(var, ent, struct print_entry, TRACE_PRINT); \
310 IF_ASSIGN(var, ent, struct bprint_entry, TRACE_BPRINT); \
269 IF_ASSIGN(var, ent, struct special_entry, 0); \ 311 IF_ASSIGN(var, ent, struct special_entry, 0); \
270 IF_ASSIGN(var, ent, struct trace_mmiotrace_rw, \ 312 IF_ASSIGN(var, ent, struct trace_mmiotrace_rw, \
271 TRACE_MMIO_RW); \ 313 TRACE_MMIO_RW); \
@@ -279,7 +321,15 @@ extern void __ftrace_bad_type(void);
279 IF_ASSIGN(var, ent, struct ftrace_graph_ret_entry, \ 321 IF_ASSIGN(var, ent, struct ftrace_graph_ret_entry, \
280 TRACE_GRAPH_RET); \ 322 TRACE_GRAPH_RET); \
281 IF_ASSIGN(var, ent, struct hw_branch_entry, TRACE_HW_BRANCHES);\ 323 IF_ASSIGN(var, ent, struct hw_branch_entry, TRACE_HW_BRANCHES);\
282 IF_ASSIGN(var, ent, struct trace_power, TRACE_POWER); \ 324 IF_ASSIGN(var, ent, struct trace_power, TRACE_POWER); \
325 IF_ASSIGN(var, ent, struct kmemtrace_alloc_entry, \
326 TRACE_KMEM_ALLOC); \
327 IF_ASSIGN(var, ent, struct kmemtrace_free_entry, \
328 TRACE_KMEM_FREE); \
329 IF_ASSIGN(var, ent, struct syscall_trace_enter, \
330 TRACE_SYSCALL_ENTER); \
331 IF_ASSIGN(var, ent, struct syscall_trace_exit, \
332 TRACE_SYSCALL_EXIT); \
283 __ftrace_bad_type(); \ 333 __ftrace_bad_type(); \
284 } while (0) 334 } while (0)
285 335
@@ -287,7 +337,8 @@ extern void __ftrace_bad_type(void);
287enum print_line_t { 337enum print_line_t {
288 TRACE_TYPE_PARTIAL_LINE = 0, /* Retry after flushing the seq */ 338 TRACE_TYPE_PARTIAL_LINE = 0, /* Retry after flushing the seq */
289 TRACE_TYPE_HANDLED = 1, 339 TRACE_TYPE_HANDLED = 1,
290 TRACE_TYPE_UNHANDLED = 2 /* Relay to other output functions */ 340 TRACE_TYPE_UNHANDLED = 2, /* Relay to other output functions */
341 TRACE_TYPE_NO_CONSUME = 3 /* Handled but ask to not consume */
291}; 342};
292 343
293 344
@@ -297,8 +348,8 @@ enum print_line_t {
297 * flags value in struct tracer_flags. 348 * flags value in struct tracer_flags.
298 */ 349 */
299struct tracer_opt { 350struct tracer_opt {
300 const char *name; /* Will appear on the trace_options file */ 351 const char *name; /* Will appear on the trace_options file */
301 u32 bit; /* Mask assigned in val field in tracer_flags */ 352 u32 bit; /* Mask assigned in val field in tracer_flags */
302}; 353};
303 354
304/* 355/*
@@ -307,28 +358,51 @@ struct tracer_opt {
307 */ 358 */
308struct tracer_flags { 359struct tracer_flags {
309 u32 val; 360 u32 val;
310 struct tracer_opt *opts; 361 struct tracer_opt *opts;
311}; 362};
312 363
313/* Makes more easy to define a tracer opt */ 364/* Makes more easy to define a tracer opt */
314#define TRACER_OPT(s, b) .name = #s, .bit = b 365#define TRACER_OPT(s, b) .name = #s, .bit = b
315 366
316/* 367
317 * A specific tracer, represented by methods that operate on a trace array: 368/**
369 * struct tracer - a specific tracer and its callbacks to interact with debugfs
370 * @name: the name chosen to select it on the available_tracers file
371 * @init: called when one switches to this tracer (echo name > current_tracer)
372 * @reset: called when one switches to another tracer
373 * @start: called when tracing is unpaused (echo 1 > tracing_enabled)
374 * @stop: called when tracing is paused (echo 0 > tracing_enabled)
375 * @open: called when the trace file is opened
376 * @pipe_open: called when the trace_pipe file is opened
377 * @wait_pipe: override how the user waits for traces on trace_pipe
378 * @close: called when the trace file is released
379 * @read: override the default read callback on trace_pipe
380 * @splice_read: override the default splice_read callback on trace_pipe
381 * @selftest: selftest to run on boot (see trace_selftest.c)
382 * @print_headers: override the first lines that describe your columns
383 * @print_line: callback that prints a trace
384 * @set_flag: signals one of your private flags changed (trace_options file)
385 * @flags: your private flags
318 */ 386 */
319struct tracer { 387struct tracer {
320 const char *name; 388 const char *name;
321 /* Your tracer should raise a warning if init fails */
322 int (*init)(struct trace_array *tr); 389 int (*init)(struct trace_array *tr);
323 void (*reset)(struct trace_array *tr); 390 void (*reset)(struct trace_array *tr);
324 void (*start)(struct trace_array *tr); 391 void (*start)(struct trace_array *tr);
325 void (*stop)(struct trace_array *tr); 392 void (*stop)(struct trace_array *tr);
326 void (*open)(struct trace_iterator *iter); 393 void (*open)(struct trace_iterator *iter);
327 void (*pipe_open)(struct trace_iterator *iter); 394 void (*pipe_open)(struct trace_iterator *iter);
395 void (*wait_pipe)(struct trace_iterator *iter);
328 void (*close)(struct trace_iterator *iter); 396 void (*close)(struct trace_iterator *iter);
329 ssize_t (*read)(struct trace_iterator *iter, 397 ssize_t (*read)(struct trace_iterator *iter,
330 struct file *filp, char __user *ubuf, 398 struct file *filp, char __user *ubuf,
331 size_t cnt, loff_t *ppos); 399 size_t cnt, loff_t *ppos);
400 ssize_t (*splice_read)(struct trace_iterator *iter,
401 struct file *filp,
402 loff_t *ppos,
403 struct pipe_inode_info *pipe,
404 size_t len,
405 unsigned int flags);
332#ifdef CONFIG_FTRACE_STARTUP_TEST 406#ifdef CONFIG_FTRACE_STARTUP_TEST
333 int (*selftest)(struct tracer *trace, 407 int (*selftest)(struct tracer *trace,
334 struct trace_array *tr); 408 struct trace_array *tr);
@@ -339,7 +413,8 @@ struct tracer {
339 int (*set_flag)(u32 old_flags, u32 bit, int set); 413 int (*set_flag)(u32 old_flags, u32 bit, int set);
340 struct tracer *next; 414 struct tracer *next;
341 int print_max; 415 int print_max;
342 struct tracer_flags *flags; 416 struct tracer_flags *flags;
417 struct tracer_stat *stats;
343}; 418};
344 419
345struct trace_seq { 420struct trace_seq {
@@ -348,6 +423,16 @@ struct trace_seq {
348 unsigned int readpos; 423 unsigned int readpos;
349}; 424};
350 425
426static inline void
427trace_seq_init(struct trace_seq *s)
428{
429 s->len = 0;
430 s->readpos = 0;
431}
432
433
434#define TRACE_PIPE_ALL_CPU -1
435
351/* 436/*
352 * Trace iterator - used by printout routines who present trace 437 * Trace iterator - used by printout routines who present trace
353 * results to users and which routines might sleep, etc: 438 * results to users and which routines might sleep, etc:
@@ -356,6 +441,8 @@ struct trace_iterator {
356 struct trace_array *tr; 441 struct trace_array *tr;
357 struct tracer *trace; 442 struct tracer *trace;
358 void *private; 443 void *private;
444 int cpu_file;
445 struct mutex mutex;
359 struct ring_buffer_iter *buffer_iter[NR_CPUS]; 446 struct ring_buffer_iter *buffer_iter[NR_CPUS];
360 447
361 /* The below is zeroed out in pipe_read */ 448 /* The below is zeroed out in pipe_read */
@@ -371,6 +458,7 @@ struct trace_iterator {
371 cpumask_var_t started; 458 cpumask_var_t started;
372}; 459};
373 460
461int tracer_init(struct tracer *t, struct trace_array *tr);
374int tracing_is_enabled(void); 462int tracing_is_enabled(void);
375void trace_wake_up(void); 463void trace_wake_up(void);
376void tracing_reset(struct trace_array *tr, int cpu); 464void tracing_reset(struct trace_array *tr, int cpu);
@@ -379,26 +467,50 @@ int tracing_open_generic(struct inode *inode, struct file *filp);
379struct dentry *tracing_init_dentry(void); 467struct dentry *tracing_init_dentry(void);
380void init_tracer_sysprof_debugfs(struct dentry *d_tracer); 468void init_tracer_sysprof_debugfs(struct dentry *d_tracer);
381 469
470struct ring_buffer_event;
471
472struct ring_buffer_event *trace_buffer_lock_reserve(struct trace_array *tr,
473 unsigned char type,
474 unsigned long len,
475 unsigned long flags,
476 int pc);
477void trace_buffer_unlock_commit(struct trace_array *tr,
478 struct ring_buffer_event *event,
479 unsigned long flags, int pc);
480
481struct ring_buffer_event *
482trace_current_buffer_lock_reserve(unsigned char type, unsigned long len,
483 unsigned long flags, int pc);
484void trace_current_buffer_unlock_commit(struct ring_buffer_event *event,
485 unsigned long flags, int pc);
486void trace_nowake_buffer_unlock_commit(struct ring_buffer_event *event,
487 unsigned long flags, int pc);
488
382struct trace_entry *tracing_get_trace_entry(struct trace_array *tr, 489struct trace_entry *tracing_get_trace_entry(struct trace_array *tr,
383 struct trace_array_cpu *data); 490 struct trace_array_cpu *data);
491
492struct trace_entry *trace_find_next_entry(struct trace_iterator *iter,
493 int *ent_cpu, u64 *ent_ts);
494
384void tracing_generic_entry_update(struct trace_entry *entry, 495void tracing_generic_entry_update(struct trace_entry *entry,
385 unsigned long flags, 496 unsigned long flags,
386 int pc); 497 int pc);
387 498
499void default_wait_pipe(struct trace_iterator *iter);
500void poll_wait_pipe(struct trace_iterator *iter);
501
388void ftrace(struct trace_array *tr, 502void ftrace(struct trace_array *tr,
389 struct trace_array_cpu *data, 503 struct trace_array_cpu *data,
390 unsigned long ip, 504 unsigned long ip,
391 unsigned long parent_ip, 505 unsigned long parent_ip,
392 unsigned long flags, int pc); 506 unsigned long flags, int pc);
393void tracing_sched_switch_trace(struct trace_array *tr, 507void tracing_sched_switch_trace(struct trace_array *tr,
394 struct trace_array_cpu *data,
395 struct task_struct *prev, 508 struct task_struct *prev,
396 struct task_struct *next, 509 struct task_struct *next,
397 unsigned long flags, int pc); 510 unsigned long flags, int pc);
398void tracing_record_cmdline(struct task_struct *tsk); 511void tracing_record_cmdline(struct task_struct *tsk);
399 512
400void tracing_sched_wakeup_trace(struct trace_array *tr, 513void tracing_sched_wakeup_trace(struct trace_array *tr,
401 struct trace_array_cpu *data,
402 struct task_struct *wakee, 514 struct task_struct *wakee,
403 struct task_struct *cur, 515 struct task_struct *cur,
404 unsigned long flags, int pc); 516 unsigned long flags, int pc);
@@ -408,14 +520,12 @@ void trace_special(struct trace_array *tr,
408 unsigned long arg2, 520 unsigned long arg2,
409 unsigned long arg3, int pc); 521 unsigned long arg3, int pc);
410void trace_function(struct trace_array *tr, 522void trace_function(struct trace_array *tr,
411 struct trace_array_cpu *data,
412 unsigned long ip, 523 unsigned long ip,
413 unsigned long parent_ip, 524 unsigned long parent_ip,
414 unsigned long flags, int pc); 525 unsigned long flags, int pc);
415 526
416void trace_graph_return(struct ftrace_graph_ret *trace); 527void trace_graph_return(struct ftrace_graph_ret *trace);
417int trace_graph_entry(struct ftrace_graph_ent *trace); 528int trace_graph_entry(struct ftrace_graph_ent *trace);
418void trace_hw_branch(struct trace_array *tr, u64 from, u64 to);
419 529
420void tracing_start_cmdline_record(void); 530void tracing_start_cmdline_record(void);
421void tracing_stop_cmdline_record(void); 531void tracing_stop_cmdline_record(void);
@@ -434,15 +544,11 @@ void update_max_tr(struct trace_array *tr, struct task_struct *tsk, int cpu);
434void update_max_tr_single(struct trace_array *tr, 544void update_max_tr_single(struct trace_array *tr,
435 struct task_struct *tsk, int cpu); 545 struct task_struct *tsk, int cpu);
436 546
437extern cycle_t ftrace_now(int cpu); 547void __trace_stack(struct trace_array *tr,
548 unsigned long flags,
549 int skip, int pc);
438 550
439#ifdef CONFIG_FUNCTION_TRACER 551extern cycle_t ftrace_now(int cpu);
440void tracing_start_function_trace(void);
441void tracing_stop_function_trace(void);
442#else
443# define tracing_start_function_trace() do { } while (0)
444# define tracing_stop_function_trace() do { } while (0)
445#endif
446 552
447#ifdef CONFIG_CONTEXT_SWITCH_TRACER 553#ifdef CONFIG_CONTEXT_SWITCH_TRACER
448typedef void 554typedef void
@@ -456,10 +562,10 @@ struct tracer_switch_ops {
456 void *private; 562 void *private;
457 struct tracer_switch_ops *next; 563 struct tracer_switch_ops *next;
458}; 564};
459
460char *trace_find_cmdline(int pid);
461#endif /* CONFIG_CONTEXT_SWITCH_TRACER */ 565#endif /* CONFIG_CONTEXT_SWITCH_TRACER */
462 566
567extern void trace_find_cmdline(int pid, char comm[]);
568
463#ifdef CONFIG_DYNAMIC_FTRACE 569#ifdef CONFIG_DYNAMIC_FTRACE
464extern unsigned long ftrace_update_tot_cnt; 570extern unsigned long ftrace_update_tot_cnt;
465#define DYN_FTRACE_TEST_NAME trace_selftest_dynamic_test_func 571#define DYN_FTRACE_TEST_NAME trace_selftest_dynamic_test_func
@@ -469,6 +575,8 @@ extern int DYN_FTRACE_TEST_NAME(void);
469#ifdef CONFIG_FTRACE_STARTUP_TEST 575#ifdef CONFIG_FTRACE_STARTUP_TEST
470extern int trace_selftest_startup_function(struct tracer *trace, 576extern int trace_selftest_startup_function(struct tracer *trace,
471 struct trace_array *tr); 577 struct trace_array *tr);
578extern int trace_selftest_startup_function_graph(struct tracer *trace,
579 struct trace_array *tr);
472extern int trace_selftest_startup_irqsoff(struct tracer *trace, 580extern int trace_selftest_startup_irqsoff(struct tracer *trace,
473 struct trace_array *tr); 581 struct trace_array *tr);
474extern int trace_selftest_startup_preemptoff(struct tracer *trace, 582extern int trace_selftest_startup_preemptoff(struct tracer *trace,
@@ -488,18 +596,11 @@ extern int trace_selftest_startup_branch(struct tracer *trace,
488#endif /* CONFIG_FTRACE_STARTUP_TEST */ 596#endif /* CONFIG_FTRACE_STARTUP_TEST */
489 597
490extern void *head_page(struct trace_array_cpu *data); 598extern void *head_page(struct trace_array_cpu *data);
491extern int trace_seq_printf(struct trace_seq *s, const char *fmt, ...);
492extern void trace_seq_print_cont(struct trace_seq *s,
493 struct trace_iterator *iter);
494
495extern int
496seq_print_ip_sym(struct trace_seq *s, unsigned long ip,
497 unsigned long sym_flags);
498extern ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf,
499 size_t cnt);
500extern long ns2usecs(cycle_t nsec); 599extern long ns2usecs(cycle_t nsec);
501extern int 600extern int
502trace_vprintk(unsigned long ip, int depth, const char *fmt, va_list args); 601trace_vbprintk(unsigned long ip, const char *fmt, va_list args);
602extern int
603trace_vprintk(unsigned long ip, const char *fmt, va_list args);
503 604
504extern unsigned long trace_flags; 605extern unsigned long trace_flags;
505 606
@@ -580,7 +681,11 @@ enum trace_iterator_flags {
580 TRACE_ITER_ANNOTATE = 0x2000, 681 TRACE_ITER_ANNOTATE = 0x2000,
581 TRACE_ITER_USERSTACKTRACE = 0x4000, 682 TRACE_ITER_USERSTACKTRACE = 0x4000,
582 TRACE_ITER_SYM_USEROBJ = 0x8000, 683 TRACE_ITER_SYM_USEROBJ = 0x8000,
583 TRACE_ITER_PRINTK_MSGONLY = 0x10000 684 TRACE_ITER_PRINTK_MSGONLY = 0x10000,
685 TRACE_ITER_CONTEXT_INFO = 0x20000, /* Print pid/cpu/time */
686 TRACE_ITER_LATENCY_FMT = 0x40000,
687 TRACE_ITER_GLOBAL_CLK = 0x80000,
688 TRACE_ITER_SLEEP_TIME = 0x100000,
584}; 689};
585 690
586/* 691/*
@@ -601,12 +706,12 @@ extern struct tracer nop_trace;
601 * preempt_enable (after a disable), a schedule might take place 706 * preempt_enable (after a disable), a schedule might take place
602 * causing an infinite recursion. 707 * causing an infinite recursion.
603 * 708 *
604 * To prevent this, we read the need_recshed flag before 709 * To prevent this, we read the need_resched flag before
605 * disabling preemption. When we want to enable preemption we 710 * disabling preemption. When we want to enable preemption we
606 * check the flag, if it is set, then we call preempt_enable_no_resched. 711 * check the flag, if it is set, then we call preempt_enable_no_resched.
607 * Otherwise, we call preempt_enable. 712 * Otherwise, we call preempt_enable.
608 * 713 *
609 * The rational for doing the above is that if need resched is set 714 * The rational for doing the above is that if need_resched is set
610 * and we have yet to reschedule, we are either in an atomic location 715 * and we have yet to reschedule, we are either in an atomic location
611 * (where we do not need to check for scheduling) or we are inside 716 * (where we do not need to check for scheduling) or we are inside
612 * the scheduler and do not want to resched. 717 * the scheduler and do not want to resched.
@@ -627,7 +732,7 @@ static inline int ftrace_preempt_disable(void)
627 * 732 *
628 * This is a scheduler safe way to enable preemption and not miss 733 * This is a scheduler safe way to enable preemption and not miss
629 * any preemption checks. The disabled saved the state of preemption. 734 * any preemption checks. The disabled saved the state of preemption.
630 * If resched is set, then we were either inside an atomic or 735 * If resched is set, then we are either inside an atomic or
631 * are inside the scheduler (we would have already scheduled 736 * are inside the scheduler (we would have already scheduled
632 * otherwise). In this case, we do not want to call normal 737 * otherwise). In this case, we do not want to call normal
633 * preempt_enable, but preempt_enable_no_resched instead. 738 * preempt_enable, but preempt_enable_no_resched instead.
@@ -664,4 +769,118 @@ static inline void trace_branch_disable(void)
664} 769}
665#endif /* CONFIG_BRANCH_TRACER */ 770#endif /* CONFIG_BRANCH_TRACER */
666 771
772/* set ring buffers to default size if not already done so */
773int tracing_update_buffers(void);
774
775/* trace event type bit fields, not numeric */
776enum {
777 TRACE_EVENT_TYPE_PRINTF = 1,
778 TRACE_EVENT_TYPE_RAW = 2,
779};
780
781struct ftrace_event_field {
782 struct list_head link;
783 char *name;
784 char *type;
785 int offset;
786 int size;
787};
788
789struct ftrace_event_call {
790 char *name;
791 char *system;
792 struct dentry *dir;
793 int enabled;
794 int (*regfunc)(void);
795 void (*unregfunc)(void);
796 int id;
797 int (*raw_init)(void);
798 int (*show_format)(struct trace_seq *s);
799 int (*define_fields)(void);
800 struct list_head fields;
801 struct filter_pred **preds;
802
803#ifdef CONFIG_EVENT_PROFILE
804 atomic_t profile_count;
805 int (*profile_enable)(struct ftrace_event_call *);
806 void (*profile_disable)(struct ftrace_event_call *);
807#endif
808};
809
810struct event_subsystem {
811 struct list_head list;
812 const char *name;
813 struct dentry *entry;
814 struct filter_pred **preds;
815};
816
817#define events_for_each(event) \
818 for (event = __start_ftrace_events; \
819 (unsigned long)event < (unsigned long)__stop_ftrace_events; \
820 event++)
821
822#define MAX_FILTER_PRED 8
823
824struct filter_pred;
825
826typedef int (*filter_pred_fn_t) (struct filter_pred *pred, void *event);
827
828struct filter_pred {
829 filter_pred_fn_t fn;
830 u64 val;
831 char *str_val;
832 int str_len;
833 char *field_name;
834 int offset;
835 int not;
836 int or;
837 int compound;
838 int clear;
839};
840
841int trace_define_field(struct ftrace_event_call *call, char *type,
842 char *name, int offset, int size);
843extern void filter_free_pred(struct filter_pred *pred);
844extern void filter_print_preds(struct filter_pred **preds,
845 struct trace_seq *s);
846extern int filter_parse(char **pbuf, struct filter_pred *pred);
847extern int filter_add_pred(struct ftrace_event_call *call,
848 struct filter_pred *pred);
849extern void filter_free_preds(struct ftrace_event_call *call);
850extern int filter_match_preds(struct ftrace_event_call *call, void *rec);
851extern void filter_free_subsystem_preds(struct event_subsystem *system);
852extern int filter_add_subsystem_pred(struct event_subsystem *system,
853 struct filter_pred *pred);
854
855void event_trace_printk(unsigned long ip, const char *fmt, ...);
856extern struct ftrace_event_call __start_ftrace_events[];
857extern struct ftrace_event_call __stop_ftrace_events[];
858
859#define for_each_event(event) \
860 for (event = __start_ftrace_events; \
861 (unsigned long)event < (unsigned long)__stop_ftrace_events; \
862 event++)
863
864extern const char *__start___trace_bprintk_fmt[];
865extern const char *__stop___trace_bprintk_fmt[];
866
867/*
868 * The double __builtin_constant_p is because gcc will give us an error
869 * if we try to allocate the static variable to fmt if it is not a
870 * constant. Even with the outer if statement optimizing out.
871 */
872#define event_trace_printk(ip, fmt, args...) \
873do { \
874 __trace_printk_check_format(fmt, ##args); \
875 tracing_record_cmdline(current); \
876 if (__builtin_constant_p(fmt)) { \
877 static const char *trace_printk_fmt \
878 __attribute__((section("__trace_printk_fmt"))) = \
879 __builtin_constant_p(fmt) ? fmt : NULL; \
880 \
881 __trace_bprintk(ip, trace_printk_fmt, ##args); \
882 } else \
883 __trace_printk(ip, fmt, ##args); \
884} while (0)
885
667#endif /* _LINUX_KERNEL_TRACE_H */ 886#endif /* _LINUX_KERNEL_TRACE_H */
diff --git a/kernel/trace/trace_boot.c b/kernel/trace/trace_boot.c
index 366c8c333e13..7a30fc4c3642 100644
--- a/kernel/trace/trace_boot.c
+++ b/kernel/trace/trace_boot.c
@@ -11,6 +11,7 @@
11#include <linux/kallsyms.h> 11#include <linux/kallsyms.h>
12 12
13#include "trace.h" 13#include "trace.h"
14#include "trace_output.h"
14 15
15static struct trace_array *boot_trace; 16static struct trace_array *boot_trace;
16static bool pre_initcalls_finished; 17static bool pre_initcalls_finished;
@@ -27,13 +28,13 @@ void start_boot_trace(void)
27 28
28void enable_boot_trace(void) 29void enable_boot_trace(void)
29{ 30{
30 if (pre_initcalls_finished) 31 if (boot_trace && pre_initcalls_finished)
31 tracing_start_sched_switch_record(); 32 tracing_start_sched_switch_record();
32} 33}
33 34
34void disable_boot_trace(void) 35void disable_boot_trace(void)
35{ 36{
36 if (pre_initcalls_finished) 37 if (boot_trace && pre_initcalls_finished)
37 tracing_stop_sched_switch_record(); 38 tracing_stop_sched_switch_record();
38} 39}
39 40
@@ -42,6 +43,9 @@ static int boot_trace_init(struct trace_array *tr)
42 int cpu; 43 int cpu;
43 boot_trace = tr; 44 boot_trace = tr;
44 45
46 if (!tr)
47 return 0;
48
45 for_each_cpu(cpu, cpu_possible_mask) 49 for_each_cpu(cpu, cpu_possible_mask)
46 tracing_reset(tr, cpu); 50 tracing_reset(tr, cpu);
47 51
@@ -128,10 +132,9 @@ void trace_boot_call(struct boot_trace_call *bt, initcall_t fn)
128{ 132{
129 struct ring_buffer_event *event; 133 struct ring_buffer_event *event;
130 struct trace_boot_call *entry; 134 struct trace_boot_call *entry;
131 unsigned long irq_flags;
132 struct trace_array *tr = boot_trace; 135 struct trace_array *tr = boot_trace;
133 136
134 if (!pre_initcalls_finished) 137 if (!tr || !pre_initcalls_finished)
135 return; 138 return;
136 139
137 /* Get its name now since this function could 140 /* Get its name now since this function could
@@ -140,18 +143,13 @@ void trace_boot_call(struct boot_trace_call *bt, initcall_t fn)
140 sprint_symbol(bt->func, (unsigned long)fn); 143 sprint_symbol(bt->func, (unsigned long)fn);
141 preempt_disable(); 144 preempt_disable();
142 145
143 event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry), 146 event = trace_buffer_lock_reserve(tr, TRACE_BOOT_CALL,
144 &irq_flags); 147 sizeof(*entry), 0, 0);
145 if (!event) 148 if (!event)
146 goto out; 149 goto out;
147 entry = ring_buffer_event_data(event); 150 entry = ring_buffer_event_data(event);
148 tracing_generic_entry_update(&entry->ent, 0, 0);
149 entry->ent.type = TRACE_BOOT_CALL;
150 entry->boot_call = *bt; 151 entry->boot_call = *bt;
151 ring_buffer_unlock_commit(tr->buffer, event, irq_flags); 152 trace_buffer_unlock_commit(tr, event, 0, 0);
152
153 trace_wake_up();
154
155 out: 153 out:
156 preempt_enable(); 154 preempt_enable();
157} 155}
@@ -160,27 +158,21 @@ void trace_boot_ret(struct boot_trace_ret *bt, initcall_t fn)
160{ 158{
161 struct ring_buffer_event *event; 159 struct ring_buffer_event *event;
162 struct trace_boot_ret *entry; 160 struct trace_boot_ret *entry;
163 unsigned long irq_flags;
164 struct trace_array *tr = boot_trace; 161 struct trace_array *tr = boot_trace;
165 162
166 if (!pre_initcalls_finished) 163 if (!tr || !pre_initcalls_finished)
167 return; 164 return;
168 165
169 sprint_symbol(bt->func, (unsigned long)fn); 166 sprint_symbol(bt->func, (unsigned long)fn);
170 preempt_disable(); 167 preempt_disable();
171 168
172 event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry), 169 event = trace_buffer_lock_reserve(tr, TRACE_BOOT_RET,
173 &irq_flags); 170 sizeof(*entry), 0, 0);
174 if (!event) 171 if (!event)
175 goto out; 172 goto out;
176 entry = ring_buffer_event_data(event); 173 entry = ring_buffer_event_data(event);
177 tracing_generic_entry_update(&entry->ent, 0, 0);
178 entry->ent.type = TRACE_BOOT_RET;
179 entry->boot_ret = *bt; 174 entry->boot_ret = *bt;
180 ring_buffer_unlock_commit(tr->buffer, event, irq_flags); 175 trace_buffer_unlock_commit(tr, event, 0, 0);
181
182 trace_wake_up();
183
184 out: 176 out:
185 preempt_enable(); 177 preempt_enable();
186} 178}
diff --git a/kernel/trace/trace_branch.c b/kernel/trace/trace_branch.c
index 6c00feb3bac7..ad8c22efff41 100644
--- a/kernel/trace/trace_branch.c
+++ b/kernel/trace/trace_branch.c
@@ -14,12 +14,17 @@
14#include <linux/hash.h> 14#include <linux/hash.h>
15#include <linux/fs.h> 15#include <linux/fs.h>
16#include <asm/local.h> 16#include <asm/local.h>
17
17#include "trace.h" 18#include "trace.h"
19#include "trace_stat.h"
20#include "trace_output.h"
18 21
19#ifdef CONFIG_BRANCH_TRACER 22#ifdef CONFIG_BRANCH_TRACER
20 23
24static struct tracer branch_trace;
21static int branch_tracing_enabled __read_mostly; 25static int branch_tracing_enabled __read_mostly;
22static DEFINE_MUTEX(branch_tracing_mutex); 26static DEFINE_MUTEX(branch_tracing_mutex);
27
23static struct trace_array *branch_tracer; 28static struct trace_array *branch_tracer;
24 29
25static void 30static void
@@ -28,7 +33,7 @@ probe_likely_condition(struct ftrace_branch_data *f, int val, int expect)
28 struct trace_array *tr = branch_tracer; 33 struct trace_array *tr = branch_tracer;
29 struct ring_buffer_event *event; 34 struct ring_buffer_event *event;
30 struct trace_branch *entry; 35 struct trace_branch *entry;
31 unsigned long flags, irq_flags; 36 unsigned long flags;
32 int cpu, pc; 37 int cpu, pc;
33 const char *p; 38 const char *p;
34 39
@@ -47,15 +52,13 @@ probe_likely_condition(struct ftrace_branch_data *f, int val, int expect)
47 if (atomic_inc_return(&tr->data[cpu]->disabled) != 1) 52 if (atomic_inc_return(&tr->data[cpu]->disabled) != 1)
48 goto out; 53 goto out;
49 54
50 event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry), 55 pc = preempt_count();
51 &irq_flags); 56 event = trace_buffer_lock_reserve(tr, TRACE_BRANCH,
57 sizeof(*entry), flags, pc);
52 if (!event) 58 if (!event)
53 goto out; 59 goto out;
54 60
55 pc = preempt_count();
56 entry = ring_buffer_event_data(event); 61 entry = ring_buffer_event_data(event);
57 tracing_generic_entry_update(&entry->ent, flags, pc);
58 entry->ent.type = TRACE_BRANCH;
59 62
60 /* Strip off the path, only save the file */ 63 /* Strip off the path, only save the file */
61 p = f->file + strlen(f->file); 64 p = f->file + strlen(f->file);
@@ -70,7 +73,7 @@ probe_likely_condition(struct ftrace_branch_data *f, int val, int expect)
70 entry->line = f->line; 73 entry->line = f->line;
71 entry->correct = val == expect; 74 entry->correct = val == expect;
72 75
73 ring_buffer_unlock_commit(tr->buffer, event, irq_flags); 76 ring_buffer_unlock_commit(tr->buffer, event);
74 77
75 out: 78 out:
76 atomic_dec(&tr->data[cpu]->disabled); 79 atomic_dec(&tr->data[cpu]->disabled);
@@ -88,8 +91,6 @@ void trace_likely_condition(struct ftrace_branch_data *f, int val, int expect)
88 91
89int enable_branch_tracing(struct trace_array *tr) 92int enable_branch_tracing(struct trace_array *tr)
90{ 93{
91 int ret = 0;
92
93 mutex_lock(&branch_tracing_mutex); 94 mutex_lock(&branch_tracing_mutex);
94 branch_tracer = tr; 95 branch_tracer = tr;
95 /* 96 /*
@@ -100,7 +101,7 @@ int enable_branch_tracing(struct trace_array *tr)
100 branch_tracing_enabled++; 101 branch_tracing_enabled++;
101 mutex_unlock(&branch_tracing_mutex); 102 mutex_unlock(&branch_tracing_mutex);
102 103
103 return ret; 104 return 0;
104} 105}
105 106
106void disable_branch_tracing(void) 107void disable_branch_tracing(void)
@@ -128,11 +129,6 @@ static void stop_branch_trace(struct trace_array *tr)
128 129
129static int branch_trace_init(struct trace_array *tr) 130static int branch_trace_init(struct trace_array *tr)
130{ 131{
131 int cpu;
132
133 for_each_online_cpu(cpu)
134 tracing_reset(tr, cpu);
135
136 start_branch_trace(tr); 132 start_branch_trace(tr);
137 return 0; 133 return 0;
138} 134}
@@ -142,22 +138,53 @@ static void branch_trace_reset(struct trace_array *tr)
142 stop_branch_trace(tr); 138 stop_branch_trace(tr);
143} 139}
144 140
145struct tracer branch_trace __read_mostly = 141static enum print_line_t trace_branch_print(struct trace_iterator *iter,
142 int flags)
143{
144 struct trace_branch *field;
145
146 trace_assign_type(field, iter->ent);
147
148 if (trace_seq_printf(&iter->seq, "[%s] %s:%s:%d\n",
149 field->correct ? " ok " : " MISS ",
150 field->func,
151 field->file,
152 field->line))
153 return TRACE_TYPE_PARTIAL_LINE;
154
155 return TRACE_TYPE_HANDLED;
156}
157
158
159static struct trace_event trace_branch_event = {
160 .type = TRACE_BRANCH,
161 .trace = trace_branch_print,
162};
163
164static struct tracer branch_trace __read_mostly =
146{ 165{
147 .name = "branch", 166 .name = "branch",
148 .init = branch_trace_init, 167 .init = branch_trace_init,
149 .reset = branch_trace_reset, 168 .reset = branch_trace_reset,
150#ifdef CONFIG_FTRACE_SELFTEST 169#ifdef CONFIG_FTRACE_SELFTEST
151 .selftest = trace_selftest_startup_branch, 170 .selftest = trace_selftest_startup_branch,
152#endif 171#endif /* CONFIG_FTRACE_SELFTEST */
153}; 172};
154 173
155__init static int init_branch_trace(void) 174__init static int init_branch_tracer(void)
156{ 175{
176 int ret;
177
178 ret = register_ftrace_event(&trace_branch_event);
179 if (!ret) {
180 printk(KERN_WARNING "Warning: could not register "
181 "branch events\n");
182 return 1;
183 }
157 return register_tracer(&branch_trace); 184 return register_tracer(&branch_trace);
158} 185}
186device_initcall(init_branch_tracer);
159 187
160device_initcall(init_branch_trace);
161#else 188#else
162static inline 189static inline
163void trace_likely_condition(struct ftrace_branch_data *f, int val, int expect) 190void trace_likely_condition(struct ftrace_branch_data *f, int val, int expect)
@@ -183,66 +210,39 @@ void ftrace_likely_update(struct ftrace_branch_data *f, int val, int expect)
183} 210}
184EXPORT_SYMBOL(ftrace_likely_update); 211EXPORT_SYMBOL(ftrace_likely_update);
185 212
186struct ftrace_pointer { 213extern unsigned long __start_annotated_branch_profile[];
187 void *start; 214extern unsigned long __stop_annotated_branch_profile[];
188 void *stop;
189 int hit;
190};
191 215
192static void * 216static int annotated_branch_stat_headers(struct seq_file *m)
193t_next(struct seq_file *m, void *v, loff_t *pos)
194{ 217{
195 const struct ftrace_pointer *f = m->private; 218 seq_printf(m, " correct incorrect %% ");
196 struct ftrace_branch_data *p = v; 219 seq_printf(m, " Function "
197 220 " File Line\n"
198 (*pos)++; 221 " ------- --------- - "
199 222 " -------- "
200 if (v == (void *)1) 223 " ---- ----\n");
201 return f->start; 224 return 0;
202
203 ++p;
204
205 if ((void *)p >= (void *)f->stop)
206 return NULL;
207
208 return p;
209} 225}
210 226
211static void *t_start(struct seq_file *m, loff_t *pos) 227static inline long get_incorrect_percent(struct ftrace_branch_data *p)
212{ 228{
213 void *t = (void *)1; 229 long percent;
214 loff_t l = 0;
215
216 for (; t && l < *pos; t = t_next(m, t, &l))
217 ;
218 230
219 return t; 231 if (p->correct) {
220} 232 percent = p->incorrect * 100;
233 percent /= p->correct + p->incorrect;
234 } else
235 percent = p->incorrect ? 100 : -1;
221 236
222static void t_stop(struct seq_file *m, void *p) 237 return percent;
223{
224} 238}
225 239
226static int t_show(struct seq_file *m, void *v) 240static int branch_stat_show(struct seq_file *m, void *v)
227{ 241{
228 const struct ftrace_pointer *fp = m->private;
229 struct ftrace_branch_data *p = v; 242 struct ftrace_branch_data *p = v;
230 const char *f; 243 const char *f;
231 long percent; 244 long percent;
232 245
233 if (v == (void *)1) {
234 if (fp->hit)
235 seq_printf(m, " miss hit %% ");
236 else
237 seq_printf(m, " correct incorrect %% ");
238 seq_printf(m, " Function "
239 " File Line\n"
240 " ------- --------- - "
241 " -------- "
242 " ---- ----\n");
243 return 0;
244 }
245
246 /* Only print the file, not the path */ 246 /* Only print the file, not the path */
247 f = p->file + strlen(p->file); 247 f = p->file + strlen(p->file);
248 while (f >= p->file && *f != '/') 248 while (f >= p->file && *f != '/')
@@ -252,11 +252,7 @@ static int t_show(struct seq_file *m, void *v)
252 /* 252 /*
253 * The miss is overlayed on correct, and hit on incorrect. 253 * The miss is overlayed on correct, and hit on incorrect.
254 */ 254 */
255 if (p->correct) { 255 percent = get_incorrect_percent(p);
256 percent = p->incorrect * 100;
257 percent /= p->correct + p->incorrect;
258 } else
259 percent = p->incorrect ? 100 : -1;
260 256
261 seq_printf(m, "%8lu %8lu ", p->correct, p->incorrect); 257 seq_printf(m, "%8lu %8lu ", p->correct, p->incorrect);
262 if (percent < 0) 258 if (percent < 0)
@@ -267,76 +263,118 @@ static int t_show(struct seq_file *m, void *v)
267 return 0; 263 return 0;
268} 264}
269 265
270static struct seq_operations tracing_likely_seq_ops = { 266static void *annotated_branch_stat_start(void)
271 .start = t_start, 267{
272 .next = t_next, 268 return __start_annotated_branch_profile;
273 .stop = t_stop, 269}
274 .show = t_show, 270
271static void *
272annotated_branch_stat_next(void *v, int idx)
273{
274 struct ftrace_branch_data *p = v;
275
276 ++p;
277
278 if ((void *)p >= (void *)__stop_annotated_branch_profile)
279 return NULL;
280
281 return p;
282}
283
284static int annotated_branch_stat_cmp(void *p1, void *p2)
285{
286 struct ftrace_branch_data *a = p1;
287 struct ftrace_branch_data *b = p2;
288
289 long percent_a, percent_b;
290
291 percent_a = get_incorrect_percent(a);
292 percent_b = get_incorrect_percent(b);
293
294 if (percent_a < percent_b)
295 return -1;
296 if (percent_a > percent_b)
297 return 1;
298 else
299 return 0;
300}
301
302static struct tracer_stat annotated_branch_stats = {
303 .name = "branch_annotated",
304 .stat_start = annotated_branch_stat_start,
305 .stat_next = annotated_branch_stat_next,
306 .stat_cmp = annotated_branch_stat_cmp,
307 .stat_headers = annotated_branch_stat_headers,
308 .stat_show = branch_stat_show
275}; 309};
276 310
277static int tracing_branch_open(struct inode *inode, struct file *file) 311__init static int init_annotated_branch_stats(void)
278{ 312{
279 int ret; 313 int ret;
280 314
281 ret = seq_open(file, &tracing_likely_seq_ops); 315 ret = register_stat_tracer(&annotated_branch_stats);
282 if (!ret) { 316 if (!ret) {
283 struct seq_file *m = file->private_data; 317 printk(KERN_WARNING "Warning: could not register "
284 m->private = (void *)inode->i_private; 318 "annotated branches stats\n");
319 return 1;
285 } 320 }
286 321 return 0;
287 return ret;
288} 322}
289 323fs_initcall(init_annotated_branch_stats);
290static const struct file_operations tracing_branch_fops = {
291 .open = tracing_branch_open,
292 .read = seq_read,
293 .llseek = seq_lseek,
294};
295 324
296#ifdef CONFIG_PROFILE_ALL_BRANCHES 325#ifdef CONFIG_PROFILE_ALL_BRANCHES
326
297extern unsigned long __start_branch_profile[]; 327extern unsigned long __start_branch_profile[];
298extern unsigned long __stop_branch_profile[]; 328extern unsigned long __stop_branch_profile[];
299 329
300static const struct ftrace_pointer ftrace_branch_pos = { 330static int all_branch_stat_headers(struct seq_file *m)
301 .start = __start_branch_profile, 331{
302 .stop = __stop_branch_profile, 332 seq_printf(m, " miss hit %% ");
303 .hit = 1, 333 seq_printf(m, " Function "
304}; 334 " File Line\n"
335 " ------- --------- - "
336 " -------- "
337 " ---- ----\n");
338 return 0;
339}
305 340
306#endif /* CONFIG_PROFILE_ALL_BRANCHES */ 341static void *all_branch_stat_start(void)
342{
343 return __start_branch_profile;
344}
307 345
308extern unsigned long __start_annotated_branch_profile[]; 346static void *
309extern unsigned long __stop_annotated_branch_profile[]; 347all_branch_stat_next(void *v, int idx)
348{
349 struct ftrace_branch_data *p = v;
310 350
311static const struct ftrace_pointer ftrace_annotated_branch_pos = { 351 ++p;
312 .start = __start_annotated_branch_profile,
313 .stop = __stop_annotated_branch_profile,
314};
315 352
316static __init int ftrace_branch_init(void) 353 if ((void *)p >= (void *)__stop_branch_profile)
317{ 354 return NULL;
318 struct dentry *d_tracer;
319 struct dentry *entry;
320 355
321 d_tracer = tracing_init_dentry(); 356 return p;
357}
322 358
323 entry = debugfs_create_file("profile_annotated_branch", 0444, d_tracer, 359static struct tracer_stat all_branch_stats = {
324 (void *)&ftrace_annotated_branch_pos, 360 .name = "branch_all",
325 &tracing_branch_fops); 361 .stat_start = all_branch_stat_start,
326 if (!entry) 362 .stat_next = all_branch_stat_next,
327 pr_warning("Could not create debugfs " 363 .stat_headers = all_branch_stat_headers,
328 "'profile_annotatet_branch' entry\n"); 364 .stat_show = branch_stat_show
365};
329 366
330#ifdef CONFIG_PROFILE_ALL_BRANCHES 367__init static int all_annotated_branch_stats(void)
331 entry = debugfs_create_file("profile_branch", 0444, d_tracer, 368{
332 (void *)&ftrace_branch_pos, 369 int ret;
333 &tracing_branch_fops);
334 if (!entry)
335 pr_warning("Could not create debugfs"
336 " 'profile_branch' entry\n");
337#endif
338 370
371 ret = register_stat_tracer(&all_branch_stats);
372 if (!ret) {
373 printk(KERN_WARNING "Warning: could not register "
374 "all branches stats\n");
375 return 1;
376 }
339 return 0; 377 return 0;
340} 378}
341 379fs_initcall(all_annotated_branch_stats);
342device_initcall(ftrace_branch_init); 380#endif /* CONFIG_PROFILE_ALL_BRANCHES */
diff --git a/kernel/trace/trace_clock.c b/kernel/trace/trace_clock.c
new file mode 100644
index 000000000000..b588fd81f7f9
--- /dev/null
+++ b/kernel/trace/trace_clock.c
@@ -0,0 +1,109 @@
1/*
2 * tracing clocks
3 *
4 * Copyright (C) 2009 Red Hat, Inc., Ingo Molnar <mingo@redhat.com>
5 *
6 * Implements 3 trace clock variants, with differing scalability/precision
7 * tradeoffs:
8 *
9 * - local: CPU-local trace clock
10 * - medium: scalable global clock with some jitter
11 * - global: globally monotonic, serialized clock
12 *
13 * Tracer plugins will chose a default from these clocks.
14 */
15#include <linux/spinlock.h>
16#include <linux/hardirq.h>
17#include <linux/module.h>
18#include <linux/percpu.h>
19#include <linux/sched.h>
20#include <linux/ktime.h>
21#include <linux/trace_clock.h>
22
23/*
24 * trace_clock_local(): the simplest and least coherent tracing clock.
25 *
26 * Useful for tracing that does not cross to other CPUs nor
27 * does it go through idle events.
28 */
29u64 notrace trace_clock_local(void)
30{
31 unsigned long flags;
32 u64 clock;
33
34 /*
35 * sched_clock() is an architecture implemented, fast, scalable,
36 * lockless clock. It is not guaranteed to be coherent across
37 * CPUs, nor across CPU idle events.
38 */
39 raw_local_irq_save(flags);
40 clock = sched_clock();
41 raw_local_irq_restore(flags);
42
43 return clock;
44}
45
46/*
47 * trace_clock(): 'inbetween' trace clock. Not completely serialized,
48 * but not completely incorrect when crossing CPUs either.
49 *
50 * This is based on cpu_clock(), which will allow at most ~1 jiffy of
51 * jitter between CPUs. So it's a pretty scalable clock, but there
52 * can be offsets in the trace data.
53 */
54u64 notrace trace_clock(void)
55{
56 return cpu_clock(raw_smp_processor_id());
57}
58
59
60/*
61 * trace_clock_global(): special globally coherent trace clock
62 *
63 * It has higher overhead than the other trace clocks but is still
64 * an order of magnitude faster than GTOD derived hardware clocks.
65 *
66 * Used by plugins that need globally coherent timestamps.
67 */
68
69static u64 prev_trace_clock_time;
70
71static raw_spinlock_t trace_clock_lock ____cacheline_aligned_in_smp =
72 (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
73
74u64 notrace trace_clock_global(void)
75{
76 unsigned long flags;
77 int this_cpu;
78 u64 now;
79
80 raw_local_irq_save(flags);
81
82 this_cpu = raw_smp_processor_id();
83 now = cpu_clock(this_cpu);
84 /*
85 * If in an NMI context then dont risk lockups and return the
86 * cpu_clock() time:
87 */
88 if (unlikely(in_nmi()))
89 goto out;
90
91 __raw_spin_lock(&trace_clock_lock);
92
93 /*
94 * TODO: if this happens often then maybe we should reset
95 * my_scd->clock to prev_trace_clock_time+1, to make sure
96 * we start ticking with the local clock from now on?
97 */
98 if ((s64)(now - prev_trace_clock_time) < 0)
99 now = prev_trace_clock_time + 1;
100
101 prev_trace_clock_time = now;
102
103 __raw_spin_unlock(&trace_clock_lock);
104
105 out:
106 raw_local_irq_restore(flags);
107
108 return now;
109}
diff --git a/kernel/trace/trace_event_profile.c b/kernel/trace/trace_event_profile.c
new file mode 100644
index 000000000000..22cba9970776
--- /dev/null
+++ b/kernel/trace/trace_event_profile.c
@@ -0,0 +1,31 @@
1/*
2 * trace event based perf counter profiling
3 *
4 * Copyright (C) 2009 Red Hat Inc, Peter Zijlstra <pzijlstr@redhat.com>
5 *
6 */
7
8#include "trace.h"
9
10int ftrace_profile_enable(int event_id)
11{
12 struct ftrace_event_call *event;
13
14 for_each_event(event) {
15 if (event->id == event_id)
16 return event->profile_enable(event);
17 }
18
19 return -EINVAL;
20}
21
22void ftrace_profile_disable(int event_id)
23{
24 struct ftrace_event_call *event;
25
26 for_each_event(event) {
27 if (event->id == event_id)
28 return event->profile_disable(event);
29 }
30}
31
diff --git a/kernel/trace/trace_event_types.h b/kernel/trace/trace_event_types.h
new file mode 100644
index 000000000000..fd78bee71dd7
--- /dev/null
+++ b/kernel/trace/trace_event_types.h
@@ -0,0 +1,173 @@
1#undef TRACE_SYSTEM
2#define TRACE_SYSTEM ftrace
3
4/*
5 * We cheat and use the proto type field as the ID
6 * and args as the entry type (minus 'struct')
7 */
8TRACE_EVENT_FORMAT(function, TRACE_FN, ftrace_entry, ignore,
9 TRACE_STRUCT(
10 TRACE_FIELD(unsigned long, ip, ip)
11 TRACE_FIELD(unsigned long, parent_ip, parent_ip)
12 ),
13 TP_RAW_FMT(" %lx <-- %lx")
14);
15
16TRACE_EVENT_FORMAT(funcgraph_entry, TRACE_GRAPH_ENT,
17 ftrace_graph_ent_entry, ignore,
18 TRACE_STRUCT(
19 TRACE_FIELD(unsigned long, graph_ent.func, func)
20 TRACE_FIELD(int, graph_ent.depth, depth)
21 ),
22 TP_RAW_FMT("--> %lx (%d)")
23);
24
25TRACE_EVENT_FORMAT(funcgraph_exit, TRACE_GRAPH_RET,
26 ftrace_graph_ret_entry, ignore,
27 TRACE_STRUCT(
28 TRACE_FIELD(unsigned long, ret.func, func)
29 TRACE_FIELD(int, ret.depth, depth)
30 ),
31 TP_RAW_FMT("<-- %lx (%d)")
32);
33
34TRACE_EVENT_FORMAT(wakeup, TRACE_WAKE, ctx_switch_entry, ignore,
35 TRACE_STRUCT(
36 TRACE_FIELD(unsigned int, prev_pid, prev_pid)
37 TRACE_FIELD(unsigned char, prev_prio, prev_prio)
38 TRACE_FIELD(unsigned char, prev_state, prev_state)
39 TRACE_FIELD(unsigned int, next_pid, next_pid)
40 TRACE_FIELD(unsigned char, next_prio, next_prio)
41 TRACE_FIELD(unsigned char, next_state, next_state)
42 TRACE_FIELD(unsigned int, next_cpu, next_cpu)
43 ),
44 TP_RAW_FMT("%u:%u:%u ==+ %u:%u:%u [%03u]")
45);
46
47TRACE_EVENT_FORMAT(context_switch, TRACE_CTX, ctx_switch_entry, ignore,
48 TRACE_STRUCT(
49 TRACE_FIELD(unsigned int, prev_pid, prev_pid)
50 TRACE_FIELD(unsigned char, prev_prio, prev_prio)
51 TRACE_FIELD(unsigned char, prev_state, prev_state)
52 TRACE_FIELD(unsigned int, next_pid, next_pid)
53 TRACE_FIELD(unsigned char, next_prio, next_prio)
54 TRACE_FIELD(unsigned char, next_state, next_state)
55 TRACE_FIELD(unsigned int, next_cpu, next_cpu)
56 ),
57 TP_RAW_FMT("%u:%u:%u ==+ %u:%u:%u [%03u]")
58);
59
60TRACE_EVENT_FORMAT(special, TRACE_SPECIAL, special_entry, ignore,
61 TRACE_STRUCT(
62 TRACE_FIELD(unsigned long, arg1, arg1)
63 TRACE_FIELD(unsigned long, arg2, arg2)
64 TRACE_FIELD(unsigned long, arg3, arg3)
65 ),
66 TP_RAW_FMT("(%08lx) (%08lx) (%08lx)")
67);
68
69/*
70 * Stack-trace entry:
71 */
72
73/* #define FTRACE_STACK_ENTRIES 8 */
74
75TRACE_EVENT_FORMAT(kernel_stack, TRACE_STACK, stack_entry, ignore,
76 TRACE_STRUCT(
77 TRACE_FIELD(unsigned long, caller[0], stack0)
78 TRACE_FIELD(unsigned long, caller[1], stack1)
79 TRACE_FIELD(unsigned long, caller[2], stack2)
80 TRACE_FIELD(unsigned long, caller[3], stack3)
81 TRACE_FIELD(unsigned long, caller[4], stack4)
82 TRACE_FIELD(unsigned long, caller[5], stack5)
83 TRACE_FIELD(unsigned long, caller[6], stack6)
84 TRACE_FIELD(unsigned long, caller[7], stack7)
85 ),
86 TP_RAW_FMT("\t=> (%08lx)\n\t=> (%08lx)\n\t=> (%08lx)\n\t=> (%08lx)\n"
87 "\t=> (%08lx)\n\t=> (%08lx)\n\t=> (%08lx)\n\t=> (%08lx)\n")
88);
89
90TRACE_EVENT_FORMAT(user_stack, TRACE_USER_STACK, userstack_entry, ignore,
91 TRACE_STRUCT(
92 TRACE_FIELD(unsigned long, caller[0], stack0)
93 TRACE_FIELD(unsigned long, caller[1], stack1)
94 TRACE_FIELD(unsigned long, caller[2], stack2)
95 TRACE_FIELD(unsigned long, caller[3], stack3)
96 TRACE_FIELD(unsigned long, caller[4], stack4)
97 TRACE_FIELD(unsigned long, caller[5], stack5)
98 TRACE_FIELD(unsigned long, caller[6], stack6)
99 TRACE_FIELD(unsigned long, caller[7], stack7)
100 ),
101 TP_RAW_FMT("\t=> (%08lx)\n\t=> (%08lx)\n\t=> (%08lx)\n\t=> (%08lx)\n"
102 "\t=> (%08lx)\n\t=> (%08lx)\n\t=> (%08lx)\n\t=> (%08lx)\n")
103);
104
105TRACE_EVENT_FORMAT(bprint, TRACE_BPRINT, bprint_entry, ignore,
106 TRACE_STRUCT(
107 TRACE_FIELD(unsigned long, ip, ip)
108 TRACE_FIELD(char *, fmt, fmt)
109 TRACE_FIELD_ZERO_CHAR(buf)
110 ),
111 TP_RAW_FMT("%08lx (%d) fmt:%p %s")
112);
113
114TRACE_EVENT_FORMAT(print, TRACE_PRINT, print_entry, ignore,
115 TRACE_STRUCT(
116 TRACE_FIELD(unsigned long, ip, ip)
117 TRACE_FIELD_ZERO_CHAR(buf)
118 ),
119 TP_RAW_FMT("%08lx (%d) fmt:%p %s")
120);
121
122TRACE_EVENT_FORMAT(branch, TRACE_BRANCH, trace_branch, ignore,
123 TRACE_STRUCT(
124 TRACE_FIELD(unsigned int, line, line)
125 TRACE_FIELD_SPECIAL(char func[TRACE_FUNC_SIZE+1], func, func)
126 TRACE_FIELD_SPECIAL(char file[TRACE_FUNC_SIZE+1], file, file)
127 TRACE_FIELD(char, correct, correct)
128 ),
129 TP_RAW_FMT("%u:%s:%s (%u)")
130);
131
132TRACE_EVENT_FORMAT(hw_branch, TRACE_HW_BRANCHES, hw_branch_entry, ignore,
133 TRACE_STRUCT(
134 TRACE_FIELD(u64, from, from)
135 TRACE_FIELD(u64, to, to)
136 ),
137 TP_RAW_FMT("from: %llx to: %llx")
138);
139
140TRACE_EVENT_FORMAT(power, TRACE_POWER, trace_power, ignore,
141 TRACE_STRUCT(
142 TRACE_FIELD(ktime_t, state_data.stamp, stamp)
143 TRACE_FIELD(ktime_t, state_data.end, end)
144 TRACE_FIELD(int, state_data.type, type)
145 TRACE_FIELD(int, state_data.state, state)
146 ),
147 TP_RAW_FMT("%llx->%llx type:%u state:%u")
148);
149
150TRACE_EVENT_FORMAT(kmem_alloc, TRACE_KMEM_ALLOC, kmemtrace_alloc_entry, ignore,
151 TRACE_STRUCT(
152 TRACE_FIELD(enum kmemtrace_type_id, type_id, type_id)
153 TRACE_FIELD(unsigned long, call_site, call_site)
154 TRACE_FIELD(const void *, ptr, ptr)
155 TRACE_FIELD(size_t, bytes_req, bytes_req)
156 TRACE_FIELD(size_t, bytes_alloc, bytes_alloc)
157 TRACE_FIELD(gfp_t, gfp_flags, gfp_flags)
158 TRACE_FIELD(int, node, node)
159 ),
160 TP_RAW_FMT("type:%u call_site:%lx ptr:%p req:%lu alloc:%lu"
161 " flags:%x node:%d")
162);
163
164TRACE_EVENT_FORMAT(kmem_free, TRACE_KMEM_FREE, kmemtrace_free_entry, ignore,
165 TRACE_STRUCT(
166 TRACE_FIELD(enum kmemtrace_type_id, type_id, type_id)
167 TRACE_FIELD(unsigned long, call_site, call_site)
168 TRACE_FIELD(const void *, ptr, ptr)
169 ),
170 TP_RAW_FMT("type:%u call_site:%lx ptr:%p")
171);
172
173#undef TRACE_SYSTEM
diff --git a/kernel/trace/trace_events.c b/kernel/trace/trace_events.c
new file mode 100644
index 000000000000..64ec4d278ffb
--- /dev/null
+++ b/kernel/trace/trace_events.c
@@ -0,0 +1,824 @@
1/*
2 * event tracer
3 *
4 * Copyright (C) 2008 Red Hat Inc, Steven Rostedt <srostedt@redhat.com>
5 *
6 * - Added format output of fields of the trace point.
7 * This was based off of work by Tom Zanussi <tzanussi@gmail.com>.
8 *
9 */
10
11#include <linux/debugfs.h>
12#include <linux/uaccess.h>
13#include <linux/module.h>
14#include <linux/ctype.h>
15
16#include "trace_output.h"
17
18#define TRACE_SYSTEM "TRACE_SYSTEM"
19
20static DEFINE_MUTEX(event_mutex);
21
22int trace_define_field(struct ftrace_event_call *call, char *type,
23 char *name, int offset, int size)
24{
25 struct ftrace_event_field *field;
26
27 field = kzalloc(sizeof(*field), GFP_KERNEL);
28 if (!field)
29 goto err;
30
31 field->name = kstrdup(name, GFP_KERNEL);
32 if (!field->name)
33 goto err;
34
35 field->type = kstrdup(type, GFP_KERNEL);
36 if (!field->type)
37 goto err;
38
39 field->offset = offset;
40 field->size = size;
41 list_add(&field->link, &call->fields);
42
43 return 0;
44
45err:
46 if (field) {
47 kfree(field->name);
48 kfree(field->type);
49 }
50 kfree(field);
51
52 return -ENOMEM;
53}
54
55static void ftrace_clear_events(void)
56{
57 struct ftrace_event_call *call = (void *)__start_ftrace_events;
58
59
60 while ((unsigned long)call < (unsigned long)__stop_ftrace_events) {
61
62 if (call->enabled) {
63 call->enabled = 0;
64 call->unregfunc();
65 }
66 call++;
67 }
68}
69
70static void ftrace_event_enable_disable(struct ftrace_event_call *call,
71 int enable)
72{
73
74 switch (enable) {
75 case 0:
76 if (call->enabled) {
77 call->enabled = 0;
78 call->unregfunc();
79 }
80 break;
81 case 1:
82 if (!call->enabled) {
83 call->enabled = 1;
84 call->regfunc();
85 }
86 break;
87 }
88}
89
90static int ftrace_set_clr_event(char *buf, int set)
91{
92 struct ftrace_event_call *call = __start_ftrace_events;
93 char *event = NULL, *sub = NULL, *match;
94 int ret = -EINVAL;
95
96 /*
97 * The buf format can be <subsystem>:<event-name>
98 * *:<event-name> means any event by that name.
99 * :<event-name> is the same.
100 *
101 * <subsystem>:* means all events in that subsystem
102 * <subsystem>: means the same.
103 *
104 * <name> (no ':') means all events in a subsystem with
105 * the name <name> or any event that matches <name>
106 */
107
108 match = strsep(&buf, ":");
109 if (buf) {
110 sub = match;
111 event = buf;
112 match = NULL;
113
114 if (!strlen(sub) || strcmp(sub, "*") == 0)
115 sub = NULL;
116 if (!strlen(event) || strcmp(event, "*") == 0)
117 event = NULL;
118 }
119
120 mutex_lock(&event_mutex);
121 for_each_event(call) {
122
123 if (!call->name || !call->regfunc)
124 continue;
125
126 if (match &&
127 strcmp(match, call->name) != 0 &&
128 strcmp(match, call->system) != 0)
129 continue;
130
131 if (sub && strcmp(sub, call->system) != 0)
132 continue;
133
134 if (event && strcmp(event, call->name) != 0)
135 continue;
136
137 ftrace_event_enable_disable(call, set);
138
139 ret = 0;
140 }
141 mutex_unlock(&event_mutex);
142
143 return ret;
144}
145
146/* 128 should be much more than enough */
147#define EVENT_BUF_SIZE 127
148
149static ssize_t
150ftrace_event_write(struct file *file, const char __user *ubuf,
151 size_t cnt, loff_t *ppos)
152{
153 size_t read = 0;
154 int i, set = 1;
155 ssize_t ret;
156 char *buf;
157 char ch;
158
159 if (!cnt || cnt < 0)
160 return 0;
161
162 ret = tracing_update_buffers();
163 if (ret < 0)
164 return ret;
165
166 ret = get_user(ch, ubuf++);
167 if (ret)
168 return ret;
169 read++;
170 cnt--;
171
172 /* skip white space */
173 while (cnt && isspace(ch)) {
174 ret = get_user(ch, ubuf++);
175 if (ret)
176 return ret;
177 read++;
178 cnt--;
179 }
180
181 /* Only white space found? */
182 if (isspace(ch)) {
183 file->f_pos += read;
184 ret = read;
185 return ret;
186 }
187
188 buf = kmalloc(EVENT_BUF_SIZE+1, GFP_KERNEL);
189 if (!buf)
190 return -ENOMEM;
191
192 if (cnt > EVENT_BUF_SIZE)
193 cnt = EVENT_BUF_SIZE;
194
195 i = 0;
196 while (cnt && !isspace(ch)) {
197 if (!i && ch == '!')
198 set = 0;
199 else
200 buf[i++] = ch;
201
202 ret = get_user(ch, ubuf++);
203 if (ret)
204 goto out_free;
205 read++;
206 cnt--;
207 }
208 buf[i] = 0;
209
210 file->f_pos += read;
211
212 ret = ftrace_set_clr_event(buf, set);
213 if (ret)
214 goto out_free;
215
216 ret = read;
217
218 out_free:
219 kfree(buf);
220
221 return ret;
222}
223
224static void *
225t_next(struct seq_file *m, void *v, loff_t *pos)
226{
227 struct ftrace_event_call *call = m->private;
228 struct ftrace_event_call *next = call;
229
230 (*pos)++;
231
232 for (;;) {
233 if ((unsigned long)call >= (unsigned long)__stop_ftrace_events)
234 return NULL;
235
236 /*
237 * The ftrace subsystem is for showing formats only.
238 * They can not be enabled or disabled via the event files.
239 */
240 if (call->regfunc)
241 break;
242
243 call++;
244 next = call;
245 }
246
247 m->private = ++next;
248
249 return call;
250}
251
252static void *t_start(struct seq_file *m, loff_t *pos)
253{
254 return t_next(m, NULL, pos);
255}
256
257static void *
258s_next(struct seq_file *m, void *v, loff_t *pos)
259{
260 struct ftrace_event_call *call = m->private;
261 struct ftrace_event_call *next;
262
263 (*pos)++;
264
265 retry:
266 if ((unsigned long)call >= (unsigned long)__stop_ftrace_events)
267 return NULL;
268
269 if (!call->enabled) {
270 call++;
271 goto retry;
272 }
273
274 next = call;
275 m->private = ++next;
276
277 return call;
278}
279
280static void *s_start(struct seq_file *m, loff_t *pos)
281{
282 return s_next(m, NULL, pos);
283}
284
285static int t_show(struct seq_file *m, void *v)
286{
287 struct ftrace_event_call *call = v;
288
289 if (strcmp(call->system, TRACE_SYSTEM) != 0)
290 seq_printf(m, "%s:", call->system);
291 seq_printf(m, "%s\n", call->name);
292
293 return 0;
294}
295
296static void t_stop(struct seq_file *m, void *p)
297{
298}
299
300static int
301ftrace_event_seq_open(struct inode *inode, struct file *file)
302{
303 int ret;
304 const struct seq_operations *seq_ops;
305
306 if ((file->f_mode & FMODE_WRITE) &&
307 !(file->f_flags & O_APPEND))
308 ftrace_clear_events();
309
310 seq_ops = inode->i_private;
311 ret = seq_open(file, seq_ops);
312 if (!ret) {
313 struct seq_file *m = file->private_data;
314
315 m->private = __start_ftrace_events;
316 }
317 return ret;
318}
319
320static ssize_t
321event_enable_read(struct file *filp, char __user *ubuf, size_t cnt,
322 loff_t *ppos)
323{
324 struct ftrace_event_call *call = filp->private_data;
325 char *buf;
326
327 if (call->enabled)
328 buf = "1\n";
329 else
330 buf = "0\n";
331
332 return simple_read_from_buffer(ubuf, cnt, ppos, buf, 2);
333}
334
335static ssize_t
336event_enable_write(struct file *filp, const char __user *ubuf, size_t cnt,
337 loff_t *ppos)
338{
339 struct ftrace_event_call *call = filp->private_data;
340 char buf[64];
341 unsigned long val;
342 int ret;
343
344 if (cnt >= sizeof(buf))
345 return -EINVAL;
346
347 if (copy_from_user(&buf, ubuf, cnt))
348 return -EFAULT;
349
350 buf[cnt] = 0;
351
352 ret = strict_strtoul(buf, 10, &val);
353 if (ret < 0)
354 return ret;
355
356 ret = tracing_update_buffers();
357 if (ret < 0)
358 return ret;
359
360 switch (val) {
361 case 0:
362 case 1:
363 mutex_lock(&event_mutex);
364 ftrace_event_enable_disable(call, val);
365 mutex_unlock(&event_mutex);
366 break;
367
368 default:
369 return -EINVAL;
370 }
371
372 *ppos += cnt;
373
374 return cnt;
375}
376
377#undef FIELD
378#define FIELD(type, name) \
379 #type, "common_" #name, offsetof(typeof(field), name), \
380 sizeof(field.name)
381
382static int trace_write_header(struct trace_seq *s)
383{
384 struct trace_entry field;
385
386 /* struct trace_entry */
387 return trace_seq_printf(s,
388 "\tfield:%s %s;\toffset:%zu;\tsize:%zu;\n"
389 "\tfield:%s %s;\toffset:%zu;\tsize:%zu;\n"
390 "\tfield:%s %s;\toffset:%zu;\tsize:%zu;\n"
391 "\tfield:%s %s;\toffset:%zu;\tsize:%zu;\n"
392 "\tfield:%s %s;\toffset:%zu;\tsize:%zu;\n"
393 "\n",
394 FIELD(unsigned char, type),
395 FIELD(unsigned char, flags),
396 FIELD(unsigned char, preempt_count),
397 FIELD(int, pid),
398 FIELD(int, tgid));
399}
400
401static ssize_t
402event_format_read(struct file *filp, char __user *ubuf, size_t cnt,
403 loff_t *ppos)
404{
405 struct ftrace_event_call *call = filp->private_data;
406 struct trace_seq *s;
407 char *buf;
408 int r;
409
410 if (*ppos)
411 return 0;
412
413 s = kmalloc(sizeof(*s), GFP_KERNEL);
414 if (!s)
415 return -ENOMEM;
416
417 trace_seq_init(s);
418
419 /* If any of the first writes fail, so will the show_format. */
420
421 trace_seq_printf(s, "name: %s\n", call->name);
422 trace_seq_printf(s, "ID: %d\n", call->id);
423 trace_seq_printf(s, "format:\n");
424 trace_write_header(s);
425
426 r = call->show_format(s);
427 if (!r) {
428 /*
429 * ug! The format output is bigger than a PAGE!!
430 */
431 buf = "FORMAT TOO BIG\n";
432 r = simple_read_from_buffer(ubuf, cnt, ppos,
433 buf, strlen(buf));
434 goto out;
435 }
436
437 r = simple_read_from_buffer(ubuf, cnt, ppos,
438 s->buffer, s->len);
439 out:
440 kfree(s);
441 return r;
442}
443
444static ssize_t
445event_id_read(struct file *filp, char __user *ubuf, size_t cnt, loff_t *ppos)
446{
447 struct ftrace_event_call *call = filp->private_data;
448 struct trace_seq *s;
449 int r;
450
451 if (*ppos)
452 return 0;
453
454 s = kmalloc(sizeof(*s), GFP_KERNEL);
455 if (!s)
456 return -ENOMEM;
457
458 trace_seq_init(s);
459 trace_seq_printf(s, "%d\n", call->id);
460
461 r = simple_read_from_buffer(ubuf, cnt, ppos,
462 s->buffer, s->len);
463 kfree(s);
464 return r;
465}
466
467static ssize_t
468event_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
469 loff_t *ppos)
470{
471 struct ftrace_event_call *call = filp->private_data;
472 struct trace_seq *s;
473 int r;
474
475 if (*ppos)
476 return 0;
477
478 s = kmalloc(sizeof(*s), GFP_KERNEL);
479 if (!s)
480 return -ENOMEM;
481
482 trace_seq_init(s);
483
484 filter_print_preds(call->preds, s);
485 r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
486
487 kfree(s);
488
489 return r;
490}
491
492static ssize_t
493event_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
494 loff_t *ppos)
495{
496 struct ftrace_event_call *call = filp->private_data;
497 char buf[64], *pbuf = buf;
498 struct filter_pred *pred;
499 int err;
500
501 if (cnt >= sizeof(buf))
502 return -EINVAL;
503
504 if (copy_from_user(&buf, ubuf, cnt))
505 return -EFAULT;
506
507 pred = kzalloc(sizeof(*pred), GFP_KERNEL);
508 if (!pred)
509 return -ENOMEM;
510
511 err = filter_parse(&pbuf, pred);
512 if (err < 0) {
513 filter_free_pred(pred);
514 return err;
515 }
516
517 if (pred->clear) {
518 filter_free_preds(call);
519 filter_free_pred(pred);
520 return cnt;
521 }
522
523 if (filter_add_pred(call, pred)) {
524 filter_free_pred(pred);
525 return -EINVAL;
526 }
527
528 *ppos += cnt;
529
530 return cnt;
531}
532
533static ssize_t
534subsystem_filter_read(struct file *filp, char __user *ubuf, size_t cnt,
535 loff_t *ppos)
536{
537 struct event_subsystem *system = filp->private_data;
538 struct trace_seq *s;
539 int r;
540
541 if (*ppos)
542 return 0;
543
544 s = kmalloc(sizeof(*s), GFP_KERNEL);
545 if (!s)
546 return -ENOMEM;
547
548 trace_seq_init(s);
549
550 filter_print_preds(system->preds, s);
551 r = simple_read_from_buffer(ubuf, cnt, ppos, s->buffer, s->len);
552
553 kfree(s);
554
555 return r;
556}
557
558static ssize_t
559subsystem_filter_write(struct file *filp, const char __user *ubuf, size_t cnt,
560 loff_t *ppos)
561{
562 struct event_subsystem *system = filp->private_data;
563 char buf[64], *pbuf = buf;
564 struct filter_pred *pred;
565 int err;
566
567 if (cnt >= sizeof(buf))
568 return -EINVAL;
569
570 if (copy_from_user(&buf, ubuf, cnt))
571 return -EFAULT;
572
573 pred = kzalloc(sizeof(*pred), GFP_KERNEL);
574 if (!pred)
575 return -ENOMEM;
576
577 err = filter_parse(&pbuf, pred);
578 if (err < 0) {
579 filter_free_pred(pred);
580 return err;
581 }
582
583 if (pred->clear) {
584 filter_free_subsystem_preds(system);
585 filter_free_pred(pred);
586 return cnt;
587 }
588
589 if (filter_add_subsystem_pred(system, pred)) {
590 filter_free_subsystem_preds(system);
591 filter_free_pred(pred);
592 return -EINVAL;
593 }
594
595 *ppos += cnt;
596
597 return cnt;
598}
599
600static const struct seq_operations show_event_seq_ops = {
601 .start = t_start,
602 .next = t_next,
603 .show = t_show,
604 .stop = t_stop,
605};
606
607static const struct seq_operations show_set_event_seq_ops = {
608 .start = s_start,
609 .next = s_next,
610 .show = t_show,
611 .stop = t_stop,
612};
613
614static const struct file_operations ftrace_avail_fops = {
615 .open = ftrace_event_seq_open,
616 .read = seq_read,
617 .llseek = seq_lseek,
618 .release = seq_release,
619};
620
621static const struct file_operations ftrace_set_event_fops = {
622 .open = ftrace_event_seq_open,
623 .read = seq_read,
624 .write = ftrace_event_write,
625 .llseek = seq_lseek,
626 .release = seq_release,
627};
628
629static const struct file_operations ftrace_enable_fops = {
630 .open = tracing_open_generic,
631 .read = event_enable_read,
632 .write = event_enable_write,
633};
634
635static const struct file_operations ftrace_event_format_fops = {
636 .open = tracing_open_generic,
637 .read = event_format_read,
638};
639
640static const struct file_operations ftrace_event_id_fops = {
641 .open = tracing_open_generic,
642 .read = event_id_read,
643};
644
645static const struct file_operations ftrace_event_filter_fops = {
646 .open = tracing_open_generic,
647 .read = event_filter_read,
648 .write = event_filter_write,
649};
650
651static const struct file_operations ftrace_subsystem_filter_fops = {
652 .open = tracing_open_generic,
653 .read = subsystem_filter_read,
654 .write = subsystem_filter_write,
655};
656
657static struct dentry *event_trace_events_dir(void)
658{
659 static struct dentry *d_tracer;
660 static struct dentry *d_events;
661
662 if (d_events)
663 return d_events;
664
665 d_tracer = tracing_init_dentry();
666 if (!d_tracer)
667 return NULL;
668
669 d_events = debugfs_create_dir("events", d_tracer);
670 if (!d_events)
671 pr_warning("Could not create debugfs "
672 "'events' directory\n");
673
674 return d_events;
675}
676
677static LIST_HEAD(event_subsystems);
678
679static struct dentry *
680event_subsystem_dir(const char *name, struct dentry *d_events)
681{
682 struct event_subsystem *system;
683
684 /* First see if we did not already create this dir */
685 list_for_each_entry(system, &event_subsystems, list) {
686 if (strcmp(system->name, name) == 0)
687 return system->entry;
688 }
689
690 /* need to create new entry */
691 system = kmalloc(sizeof(*system), GFP_KERNEL);
692 if (!system) {
693 pr_warning("No memory to create event subsystem %s\n",
694 name);
695 return d_events;
696 }
697
698 system->entry = debugfs_create_dir(name, d_events);
699 if (!system->entry) {
700 pr_warning("Could not create event subsystem %s\n",
701 name);
702 kfree(system);
703 return d_events;
704 }
705
706 system->name = name;
707 list_add(&system->list, &event_subsystems);
708
709 system->preds = NULL;
710
711 return system->entry;
712}
713
714static int
715event_create_dir(struct ftrace_event_call *call, struct dentry *d_events)
716{
717 struct dentry *entry;
718 int ret;
719
720 /*
721 * If the trace point header did not define TRACE_SYSTEM
722 * then the system would be called "TRACE_SYSTEM".
723 */
724 if (strcmp(call->system, "TRACE_SYSTEM") != 0)
725 d_events = event_subsystem_dir(call->system, d_events);
726
727 if (call->raw_init) {
728 ret = call->raw_init();
729 if (ret < 0) {
730 pr_warning("Could not initialize trace point"
731 " events/%s\n", call->name);
732 return ret;
733 }
734 }
735
736 call->dir = debugfs_create_dir(call->name, d_events);
737 if (!call->dir) {
738 pr_warning("Could not create debugfs "
739 "'%s' directory\n", call->name);
740 return -1;
741 }
742
743 if (call->regfunc) {
744 entry = debugfs_create_file("enable", 0644, call->dir, call,
745 &ftrace_enable_fops);
746 if (!entry)
747 pr_warning("Could not create debugfs "
748 "'%s/enable' entry\n", call->name);
749 }
750
751 if (call->id) {
752 entry = debugfs_create_file("id", 0444, call->dir, call,
753 &ftrace_event_id_fops);
754 if (!entry)
755 pr_warning("Could not create debugfs '%s/id' entry\n",
756 call->name);
757 }
758
759 if (call->define_fields) {
760 ret = call->define_fields();
761 if (ret < 0) {
762 pr_warning("Could not initialize trace point"
763 " events/%s\n", call->name);
764 return ret;
765 }
766 entry = debugfs_create_file("filter", 0644, call->dir, call,
767 &ftrace_event_filter_fops);
768 if (!entry)
769 pr_warning("Could not create debugfs "
770 "'%s/filter' entry\n", call->name);
771 }
772
773 /* A trace may not want to export its format */
774 if (!call->show_format)
775 return 0;
776
777 entry = debugfs_create_file("format", 0444, call->dir, call,
778 &ftrace_event_format_fops);
779 if (!entry)
780 pr_warning("Could not create debugfs "
781 "'%s/format' entry\n", call->name);
782
783 return 0;
784}
785
786static __init int event_trace_init(void)
787{
788 struct ftrace_event_call *call = __start_ftrace_events;
789 struct dentry *d_tracer;
790 struct dentry *entry;
791 struct dentry *d_events;
792
793 d_tracer = tracing_init_dentry();
794 if (!d_tracer)
795 return 0;
796
797 entry = debugfs_create_file("available_events", 0444, d_tracer,
798 (void *)&show_event_seq_ops,
799 &ftrace_avail_fops);
800 if (!entry)
801 pr_warning("Could not create debugfs "
802 "'available_events' entry\n");
803
804 entry = debugfs_create_file("set_event", 0644, d_tracer,
805 (void *)&show_set_event_seq_ops,
806 &ftrace_set_event_fops);
807 if (!entry)
808 pr_warning("Could not create debugfs "
809 "'set_event' entry\n");
810
811 d_events = event_trace_events_dir();
812 if (!d_events)
813 return 0;
814
815 for_each_event(call) {
816 /* The linker may leave blanks */
817 if (!call->name)
818 continue;
819 event_create_dir(call, d_events);
820 }
821
822 return 0;
823}
824fs_initcall(event_trace_init);
diff --git a/kernel/trace/trace_events_filter.c b/kernel/trace/trace_events_filter.c
new file mode 100644
index 000000000000..026be412f356
--- /dev/null
+++ b/kernel/trace/trace_events_filter.c
@@ -0,0 +1,427 @@
1/*
2 * trace_events_filter - generic event filtering
3 *
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of the GNU General Public License as published by
6 * the Free Software Foundation; either version 2 of the License, or
7 * (at your option) any later version.
8 *
9 * This program is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
12 * GNU General Public License for more details.
13 *
14 * You should have received a copy of the GNU General Public License
15 * along with this program; if not, write to the Free Software
16 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
17 *
18 * Copyright (C) 2009 Tom Zanussi <tzanussi@gmail.com>
19 */
20
21#include <linux/debugfs.h>
22#include <linux/uaccess.h>
23#include <linux/module.h>
24#include <linux/ctype.h>
25
26#include "trace.h"
27#include "trace_output.h"
28
29static int filter_pred_64(struct filter_pred *pred, void *event)
30{
31 u64 *addr = (u64 *)(event + pred->offset);
32 u64 val = (u64)pred->val;
33 int match;
34
35 match = (val == *addr) ^ pred->not;
36
37 return match;
38}
39
40static int filter_pred_32(struct filter_pred *pred, void *event)
41{
42 u32 *addr = (u32 *)(event + pred->offset);
43 u32 val = (u32)pred->val;
44 int match;
45
46 match = (val == *addr) ^ pred->not;
47
48 return match;
49}
50
51static int filter_pred_16(struct filter_pred *pred, void *event)
52{
53 u16 *addr = (u16 *)(event + pred->offset);
54 u16 val = (u16)pred->val;
55 int match;
56
57 match = (val == *addr) ^ pred->not;
58
59 return match;
60}
61
62static int filter_pred_8(struct filter_pred *pred, void *event)
63{
64 u8 *addr = (u8 *)(event + pred->offset);
65 u8 val = (u8)pred->val;
66 int match;
67
68 match = (val == *addr) ^ pred->not;
69
70 return match;
71}
72
73static int filter_pred_string(struct filter_pred *pred, void *event)
74{
75 char *addr = (char *)(event + pred->offset);
76 int cmp, match;
77
78 cmp = strncmp(addr, pred->str_val, pred->str_len);
79
80 match = (!cmp) ^ pred->not;
81
82 return match;
83}
84
85/* return 1 if event matches, 0 otherwise (discard) */
86int filter_match_preds(struct ftrace_event_call *call, void *rec)
87{
88 int i, matched, and_failed = 0;
89 struct filter_pred *pred;
90
91 for (i = 0; i < MAX_FILTER_PRED; i++) {
92 if (call->preds[i]) {
93 pred = call->preds[i];
94 if (and_failed && !pred->or)
95 continue;
96 matched = pred->fn(pred, rec);
97 if (!matched && !pred->or) {
98 and_failed = 1;
99 continue;
100 } else if (matched && pred->or)
101 return 1;
102 } else
103 break;
104 }
105
106 if (and_failed)
107 return 0;
108
109 return 1;
110}
111
112void filter_print_preds(struct filter_pred **preds, struct trace_seq *s)
113{
114 char *field_name;
115 struct filter_pred *pred;
116 int i;
117
118 if (!preds) {
119 trace_seq_printf(s, "none\n");
120 return;
121 }
122
123 for (i = 0; i < MAX_FILTER_PRED; i++) {
124 if (preds[i]) {
125 pred = preds[i];
126 field_name = pred->field_name;
127 if (i)
128 trace_seq_printf(s, pred->or ? "|| " : "&& ");
129 trace_seq_printf(s, "%s ", field_name);
130 trace_seq_printf(s, pred->not ? "!= " : "== ");
131 if (pred->str_val)
132 trace_seq_printf(s, "%s\n", pred->str_val);
133 else
134 trace_seq_printf(s, "%llu\n", pred->val);
135 } else
136 break;
137 }
138}
139
140static struct ftrace_event_field *
141find_event_field(struct ftrace_event_call *call, char *name)
142{
143 struct ftrace_event_field *field;
144
145 list_for_each_entry(field, &call->fields, link) {
146 if (!strcmp(field->name, name))
147 return field;
148 }
149
150 return NULL;
151}
152
153void filter_free_pred(struct filter_pred *pred)
154{
155 if (!pred)
156 return;
157
158 kfree(pred->field_name);
159 kfree(pred->str_val);
160 kfree(pred);
161}
162
163void filter_free_preds(struct ftrace_event_call *call)
164{
165 int i;
166
167 if (call->preds) {
168 for (i = 0; i < MAX_FILTER_PRED; i++)
169 filter_free_pred(call->preds[i]);
170 kfree(call->preds);
171 call->preds = NULL;
172 }
173}
174
175void filter_free_subsystem_preds(struct event_subsystem *system)
176{
177 struct ftrace_event_call *call = __start_ftrace_events;
178 int i;
179
180 if (system->preds) {
181 for (i = 0; i < MAX_FILTER_PRED; i++)
182 filter_free_pred(system->preds[i]);
183 kfree(system->preds);
184 system->preds = NULL;
185 }
186
187 events_for_each(call) {
188 if (!call->name || !call->regfunc)
189 continue;
190
191 if (!strcmp(call->system, system->name))
192 filter_free_preds(call);
193 }
194}
195
196static int __filter_add_pred(struct ftrace_event_call *call,
197 struct filter_pred *pred)
198{
199 int i;
200
201 if (call->preds && !pred->compound)
202 filter_free_preds(call);
203
204 if (!call->preds) {
205 call->preds = kzalloc(MAX_FILTER_PRED * sizeof(pred),
206 GFP_KERNEL);
207 if (!call->preds)
208 return -ENOMEM;
209 }
210
211 for (i = 0; i < MAX_FILTER_PRED; i++) {
212 if (!call->preds[i]) {
213 call->preds[i] = pred;
214 return 0;
215 }
216 }
217
218 return -ENOMEM;
219}
220
221static int is_string_field(const char *type)
222{
223 if (strchr(type, '[') && strstr(type, "char"))
224 return 1;
225
226 return 0;
227}
228
229int filter_add_pred(struct ftrace_event_call *call, struct filter_pred *pred)
230{
231 struct ftrace_event_field *field;
232
233 field = find_event_field(call, pred->field_name);
234 if (!field)
235 return -EINVAL;
236
237 pred->offset = field->offset;
238
239 if (is_string_field(field->type)) {
240 if (!pred->str_val)
241 return -EINVAL;
242 pred->fn = filter_pred_string;
243 pred->str_len = field->size;
244 return __filter_add_pred(call, pred);
245 } else {
246 if (pred->str_val)
247 return -EINVAL;
248 }
249
250 switch (field->size) {
251 case 8:
252 pred->fn = filter_pred_64;
253 break;
254 case 4:
255 pred->fn = filter_pred_32;
256 break;
257 case 2:
258 pred->fn = filter_pred_16;
259 break;
260 case 1:
261 pred->fn = filter_pred_8;
262 break;
263 default:
264 return -EINVAL;
265 }
266
267 return __filter_add_pred(call, pred);
268}
269
270static struct filter_pred *copy_pred(struct filter_pred *pred)
271{
272 struct filter_pred *new_pred = kmalloc(sizeof(*pred), GFP_KERNEL);
273 if (!new_pred)
274 return NULL;
275
276 memcpy(new_pred, pred, sizeof(*pred));
277
278 if (pred->field_name) {
279 new_pred->field_name = kstrdup(pred->field_name, GFP_KERNEL);
280 if (!new_pred->field_name) {
281 kfree(new_pred);
282 return NULL;
283 }
284 }
285
286 if (pred->str_val) {
287 new_pred->str_val = kstrdup(pred->str_val, GFP_KERNEL);
288 if (!new_pred->str_val) {
289 filter_free_pred(new_pred);
290 return NULL;
291 }
292 }
293
294 return new_pred;
295}
296
297int filter_add_subsystem_pred(struct event_subsystem *system,
298 struct filter_pred *pred)
299{
300 struct ftrace_event_call *call = __start_ftrace_events;
301 struct filter_pred *event_pred;
302 int i;
303
304 if (system->preds && !pred->compound)
305 filter_free_subsystem_preds(system);
306
307 if (!system->preds) {
308 system->preds = kzalloc(MAX_FILTER_PRED * sizeof(pred),
309 GFP_KERNEL);
310 if (!system->preds)
311 return -ENOMEM;
312 }
313
314 for (i = 0; i < MAX_FILTER_PRED; i++) {
315 if (!system->preds[i]) {
316 system->preds[i] = pred;
317 break;
318 }
319 }
320
321 if (i == MAX_FILTER_PRED)
322 return -EINVAL;
323
324 events_for_each(call) {
325 int err;
326
327 if (!call->name || !call->regfunc)
328 continue;
329
330 if (strcmp(call->system, system->name))
331 continue;
332
333 if (!find_event_field(call, pred->field_name))
334 continue;
335
336 event_pred = copy_pred(pred);
337 if (!event_pred)
338 goto oom;
339
340 err = filter_add_pred(call, event_pred);
341 if (err)
342 filter_free_pred(event_pred);
343 if (err == -ENOMEM)
344 goto oom;
345 }
346
347 return 0;
348
349oom:
350 system->preds[i] = NULL;
351 return -ENOMEM;
352}
353
354int filter_parse(char **pbuf, struct filter_pred *pred)
355{
356 char *tmp, *tok, *val_str = NULL;
357 int tok_n = 0;
358
359 /* field ==/!= number, or/and field ==/!= number, number */
360 while ((tok = strsep(pbuf, " \n"))) {
361 if (tok_n == 0) {
362 if (!strcmp(tok, "0")) {
363 pred->clear = 1;
364 return 0;
365 } else if (!strcmp(tok, "&&")) {
366 pred->or = 0;
367 pred->compound = 1;
368 } else if (!strcmp(tok, "||")) {
369 pred->or = 1;
370 pred->compound = 1;
371 } else
372 pred->field_name = tok;
373 tok_n = 1;
374 continue;
375 }
376 if (tok_n == 1) {
377 if (!pred->field_name)
378 pred->field_name = tok;
379 else if (!strcmp(tok, "!="))
380 pred->not = 1;
381 else if (!strcmp(tok, "=="))
382 pred->not = 0;
383 else {
384 pred->field_name = NULL;
385 return -EINVAL;
386 }
387 tok_n = 2;
388 continue;
389 }
390 if (tok_n == 2) {
391 if (pred->compound) {
392 if (!strcmp(tok, "!="))
393 pred->not = 1;
394 else if (!strcmp(tok, "=="))
395 pred->not = 0;
396 else {
397 pred->field_name = NULL;
398 return -EINVAL;
399 }
400 } else {
401 val_str = tok;
402 break; /* done */
403 }
404 tok_n = 3;
405 continue;
406 }
407 if (tok_n == 3) {
408 val_str = tok;
409 break; /* done */
410 }
411 }
412
413 pred->field_name = kstrdup(pred->field_name, GFP_KERNEL);
414 if (!pred->field_name)
415 return -ENOMEM;
416
417 pred->val = simple_strtoull(val_str, &tmp, 10);
418 if (tmp == val_str) {
419 pred->str_val = kstrdup(val_str, GFP_KERNEL);
420 if (!pred->str_val)
421 return -ENOMEM;
422 }
423
424 return 0;
425}
426
427
diff --git a/kernel/trace/trace_events_stage_1.h b/kernel/trace/trace_events_stage_1.h
new file mode 100644
index 000000000000..38985f9b379c
--- /dev/null
+++ b/kernel/trace/trace_events_stage_1.h
@@ -0,0 +1,39 @@
1/*
2 * Stage 1 of the trace events.
3 *
4 * Override the macros in <trace/trace_event_types.h> to include the following:
5 *
6 * struct ftrace_raw_<call> {
7 * struct trace_entry ent;
8 * <type> <item>;
9 * <type2> <item2>[<len>];
10 * [...]
11 * };
12 *
13 * The <type> <item> is created by the __field(type, item) macro or
14 * the __array(type2, item2, len) macro.
15 * We simply do "type item;", and that will create the fields
16 * in the structure.
17 */
18
19#undef TRACE_FORMAT
20#define TRACE_FORMAT(call, proto, args, fmt)
21
22#undef __array
23#define __array(type, item, len) type item[len];
24
25#undef __field
26#define __field(type, item) type item;
27
28#undef TP_STRUCT__entry
29#define TP_STRUCT__entry(args...) args
30
31#undef TRACE_EVENT
32#define TRACE_EVENT(name, proto, args, tstruct, assign, print) \
33 struct ftrace_raw_##name { \
34 struct trace_entry ent; \
35 tstruct \
36 }; \
37 static struct ftrace_event_call event_##name
38
39#include <trace/trace_event_types.h>
diff --git a/kernel/trace/trace_events_stage_2.h b/kernel/trace/trace_events_stage_2.h
new file mode 100644
index 000000000000..30743f7d4110
--- /dev/null
+++ b/kernel/trace/trace_events_stage_2.h
@@ -0,0 +1,176 @@
1/*
2 * Stage 2 of the trace events.
3 *
4 * Override the macros in <trace/trace_event_types.h> to include the following:
5 *
6 * enum print_line_t
7 * ftrace_raw_output_<call>(struct trace_iterator *iter, int flags)
8 * {
9 * struct trace_seq *s = &iter->seq;
10 * struct ftrace_raw_<call> *field; <-- defined in stage 1
11 * struct trace_entry *entry;
12 * int ret;
13 *
14 * entry = iter->ent;
15 *
16 * if (entry->type != event_<call>.id) {
17 * WARN_ON_ONCE(1);
18 * return TRACE_TYPE_UNHANDLED;
19 * }
20 *
21 * field = (typeof(field))entry;
22 *
23 * ret = trace_seq_printf(s, <TP_printk> "\n");
24 * if (!ret)
25 * return TRACE_TYPE_PARTIAL_LINE;
26 *
27 * return TRACE_TYPE_HANDLED;
28 * }
29 *
30 * This is the method used to print the raw event to the trace
31 * output format. Note, this is not needed if the data is read
32 * in binary.
33 */
34
35#undef __entry
36#define __entry field
37
38#undef TP_printk
39#define TP_printk(fmt, args...) fmt "\n", args
40
41#undef TRACE_EVENT
42#define TRACE_EVENT(call, proto, args, tstruct, assign, print) \
43enum print_line_t \
44ftrace_raw_output_##call(struct trace_iterator *iter, int flags) \
45{ \
46 struct trace_seq *s = &iter->seq; \
47 struct ftrace_raw_##call *field; \
48 struct trace_entry *entry; \
49 int ret; \
50 \
51 entry = iter->ent; \
52 \
53 if (entry->type != event_##call.id) { \
54 WARN_ON_ONCE(1); \
55 return TRACE_TYPE_UNHANDLED; \
56 } \
57 \
58 field = (typeof(field))entry; \
59 \
60 ret = trace_seq_printf(s, #call ": " print); \
61 if (!ret) \
62 return TRACE_TYPE_PARTIAL_LINE; \
63 \
64 return TRACE_TYPE_HANDLED; \
65}
66
67#include <trace/trace_event_types.h>
68
69/*
70 * Setup the showing format of trace point.
71 *
72 * int
73 * ftrace_format_##call(struct trace_seq *s)
74 * {
75 * struct ftrace_raw_##call field;
76 * int ret;
77 *
78 * ret = trace_seq_printf(s, #type " " #item ";"
79 * " offset:%u; size:%u;\n",
80 * offsetof(struct ftrace_raw_##call, item),
81 * sizeof(field.type));
82 *
83 * }
84 */
85
86#undef TP_STRUCT__entry
87#define TP_STRUCT__entry(args...) args
88
89#undef __field
90#define __field(type, item) \
91 ret = trace_seq_printf(s, "\tfield:" #type " " #item ";\t" \
92 "offset:%u;\tsize:%u;\n", \
93 (unsigned int)offsetof(typeof(field), item), \
94 (unsigned int)sizeof(field.item)); \
95 if (!ret) \
96 return 0;
97
98#undef __array
99#define __array(type, item, len) \
100 ret = trace_seq_printf(s, "\tfield:" #type " " #item "[" #len "];\t" \
101 "offset:%u;\tsize:%u;\n", \
102 (unsigned int)offsetof(typeof(field), item), \
103 (unsigned int)sizeof(field.item)); \
104 if (!ret) \
105 return 0;
106
107#undef __entry
108#define __entry "REC"
109
110#undef TP_printk
111#define TP_printk(fmt, args...) "%s, %s\n", #fmt, #args
112
113#undef TP_fast_assign
114#define TP_fast_assign(args...) args
115
116#undef TRACE_EVENT
117#define TRACE_EVENT(call, proto, args, tstruct, func, print) \
118static int \
119ftrace_format_##call(struct trace_seq *s) \
120{ \
121 struct ftrace_raw_##call field; \
122 int ret; \
123 \
124 tstruct; \
125 \
126 trace_seq_printf(s, "\nprint fmt: " print); \
127 \
128 return ret; \
129}
130
131#include <trace/trace_event_types.h>
132
133#undef __field
134#define __field(type, item) \
135 ret = trace_define_field(event_call, #type, #item, \
136 offsetof(typeof(field), item), \
137 sizeof(field.item)); \
138 if (ret) \
139 return ret;
140
141#undef __array
142#define __array(type, item, len) \
143 ret = trace_define_field(event_call, #type "[" #len "]", #item, \
144 offsetof(typeof(field), item), \
145 sizeof(field.item)); \
146 if (ret) \
147 return ret;
148
149#define __common_field(type, item) \
150 ret = trace_define_field(event_call, #type, "common_" #item, \
151 offsetof(typeof(field.ent), item), \
152 sizeof(field.ent.item)); \
153 if (ret) \
154 return ret;
155
156#undef TRACE_EVENT
157#define TRACE_EVENT(call, proto, args, tstruct, func, print) \
158int \
159ftrace_define_fields_##call(void) \
160{ \
161 struct ftrace_raw_##call field; \
162 struct ftrace_event_call *event_call = &event_##call; \
163 int ret; \
164 \
165 __common_field(unsigned char, type); \
166 __common_field(unsigned char, flags); \
167 __common_field(unsigned char, preempt_count); \
168 __common_field(int, pid); \
169 __common_field(int, tgid); \
170 \
171 tstruct; \
172 \
173 return ret; \
174}
175
176#include <trace/trace_event_types.h>
diff --git a/kernel/trace/trace_events_stage_3.h b/kernel/trace/trace_events_stage_3.h
new file mode 100644
index 000000000000..9d2fa78cecca
--- /dev/null
+++ b/kernel/trace/trace_events_stage_3.h
@@ -0,0 +1,281 @@
1/*
2 * Stage 3 of the trace events.
3 *
4 * Override the macros in <trace/trace_event_types.h> to include the following:
5 *
6 * static void ftrace_event_<call>(proto)
7 * {
8 * event_trace_printk(_RET_IP_, "<call>: " <fmt>);
9 * }
10 *
11 * static int ftrace_reg_event_<call>(void)
12 * {
13 * int ret;
14 *
15 * ret = register_trace_<call>(ftrace_event_<call>);
16 * if (!ret)
17 * pr_info("event trace: Could not activate trace point "
18 * "probe to <call>");
19 * return ret;
20 * }
21 *
22 * static void ftrace_unreg_event_<call>(void)
23 * {
24 * unregister_trace_<call>(ftrace_event_<call>);
25 * }
26 *
27 * For those macros defined with TRACE_FORMAT:
28 *
29 * static struct ftrace_event_call __used
30 * __attribute__((__aligned__(4)))
31 * __attribute__((section("_ftrace_events"))) event_<call> = {
32 * .name = "<call>",
33 * .regfunc = ftrace_reg_event_<call>,
34 * .unregfunc = ftrace_unreg_event_<call>,
35 * }
36 *
37 *
38 * For those macros defined with TRACE_EVENT:
39 *
40 * static struct ftrace_event_call event_<call>;
41 *
42 * static void ftrace_raw_event_<call>(proto)
43 * {
44 * struct ring_buffer_event *event;
45 * struct ftrace_raw_<call> *entry; <-- defined in stage 1
46 * unsigned long irq_flags;
47 * int pc;
48 *
49 * local_save_flags(irq_flags);
50 * pc = preempt_count();
51 *
52 * event = trace_current_buffer_lock_reserve(event_<call>.id,
53 * sizeof(struct ftrace_raw_<call>),
54 * irq_flags, pc);
55 * if (!event)
56 * return;
57 * entry = ring_buffer_event_data(event);
58 *
59 * <assign>; <-- Here we assign the entries by the __field and
60 * __array macros.
61 *
62 * trace_current_buffer_unlock_commit(event, irq_flags, pc);
63 * }
64 *
65 * static int ftrace_raw_reg_event_<call>(void)
66 * {
67 * int ret;
68 *
69 * ret = register_trace_<call>(ftrace_raw_event_<call>);
70 * if (!ret)
71 * pr_info("event trace: Could not activate trace point "
72 * "probe to <call>");
73 * return ret;
74 * }
75 *
76 * static void ftrace_unreg_event_<call>(void)
77 * {
78 * unregister_trace_<call>(ftrace_raw_event_<call>);
79 * }
80 *
81 * static struct trace_event ftrace_event_type_<call> = {
82 * .trace = ftrace_raw_output_<call>, <-- stage 2
83 * };
84 *
85 * static int ftrace_raw_init_event_<call>(void)
86 * {
87 * int id;
88 *
89 * id = register_ftrace_event(&ftrace_event_type_<call>);
90 * if (!id)
91 * return -ENODEV;
92 * event_<call>.id = id;
93 * return 0;
94 * }
95 *
96 * static struct ftrace_event_call __used
97 * __attribute__((__aligned__(4)))
98 * __attribute__((section("_ftrace_events"))) event_<call> = {
99 * .name = "<call>",
100 * .system = "<system>",
101 * .raw_init = ftrace_raw_init_event_<call>,
102 * .regfunc = ftrace_reg_event_<call>,
103 * .unregfunc = ftrace_unreg_event_<call>,
104 * .show_format = ftrace_format_<call>,
105 * }
106 *
107 */
108
109#undef TP_FMT
110#define TP_FMT(fmt, args...) fmt "\n", ##args
111
112#ifdef CONFIG_EVENT_PROFILE
113#define _TRACE_PROFILE(call, proto, args) \
114static void ftrace_profile_##call(proto) \
115{ \
116 extern void perf_tpcounter_event(int); \
117 perf_tpcounter_event(event_##call.id); \
118} \
119 \
120static int ftrace_profile_enable_##call(struct ftrace_event_call *call) \
121{ \
122 int ret = 0; \
123 \
124 if (!atomic_inc_return(&call->profile_count)) \
125 ret = register_trace_##call(ftrace_profile_##call); \
126 \
127 return ret; \
128} \
129 \
130static void ftrace_profile_disable_##call(struct ftrace_event_call *call) \
131{ \
132 if (atomic_add_negative(-1, &call->profile_count)) \
133 unregister_trace_##call(ftrace_profile_##call); \
134}
135
136#define _TRACE_PROFILE_INIT(call) \
137 .profile_count = ATOMIC_INIT(-1), \
138 .profile_enable = ftrace_profile_enable_##call, \
139 .profile_disable = ftrace_profile_disable_##call,
140
141#else
142#define _TRACE_PROFILE(call, proto, args)
143#define _TRACE_PROFILE_INIT(call)
144#endif
145
146#define _TRACE_FORMAT(call, proto, args, fmt) \
147static void ftrace_event_##call(proto) \
148{ \
149 event_trace_printk(_RET_IP_, #call ": " fmt); \
150} \
151 \
152static int ftrace_reg_event_##call(void) \
153{ \
154 int ret; \
155 \
156 ret = register_trace_##call(ftrace_event_##call); \
157 if (ret) \
158 pr_info("event trace: Could not activate trace point " \
159 "probe to " #call "\n"); \
160 return ret; \
161} \
162 \
163static void ftrace_unreg_event_##call(void) \
164{ \
165 unregister_trace_##call(ftrace_event_##call); \
166} \
167 \
168static struct ftrace_event_call event_##call; \
169 \
170static int ftrace_init_event_##call(void) \
171{ \
172 int id; \
173 \
174 id = register_ftrace_event(NULL); \
175 if (!id) \
176 return -ENODEV; \
177 event_##call.id = id; \
178 return 0; \
179}
180
181#undef TRACE_FORMAT
182#define TRACE_FORMAT(call, proto, args, fmt) \
183_TRACE_FORMAT(call, PARAMS(proto), PARAMS(args), PARAMS(fmt)) \
184_TRACE_PROFILE(call, PARAMS(proto), PARAMS(args)) \
185static struct ftrace_event_call __used \
186__attribute__((__aligned__(4))) \
187__attribute__((section("_ftrace_events"))) event_##call = { \
188 .name = #call, \
189 .system = __stringify(TRACE_SYSTEM), \
190 .raw_init = ftrace_init_event_##call, \
191 .regfunc = ftrace_reg_event_##call, \
192 .unregfunc = ftrace_unreg_event_##call, \
193 _TRACE_PROFILE_INIT(call) \
194}
195
196#undef __entry
197#define __entry entry
198
199#undef TRACE_EVENT
200#define TRACE_EVENT(call, proto, args, tstruct, assign, print) \
201_TRACE_PROFILE(call, PARAMS(proto), PARAMS(args)) \
202 \
203static struct ftrace_event_call event_##call; \
204 \
205static void ftrace_raw_event_##call(proto) \
206{ \
207 struct ftrace_event_call *call = &event_##call; \
208 struct ring_buffer_event *event; \
209 struct ftrace_raw_##call *entry; \
210 unsigned long irq_flags; \
211 int pc; \
212 \
213 local_save_flags(irq_flags); \
214 pc = preempt_count(); \
215 \
216 event = trace_current_buffer_lock_reserve(event_##call.id, \
217 sizeof(struct ftrace_raw_##call), \
218 irq_flags, pc); \
219 if (!event) \
220 return; \
221 entry = ring_buffer_event_data(event); \
222 \
223 assign; \
224 \
225 if (call->preds && !filter_match_preds(call, entry)) \
226 ring_buffer_event_discard(event); \
227 \
228 trace_nowake_buffer_unlock_commit(event, irq_flags, pc); \
229 \
230} \
231 \
232static int ftrace_raw_reg_event_##call(void) \
233{ \
234 int ret; \
235 \
236 ret = register_trace_##call(ftrace_raw_event_##call); \
237 if (ret) \
238 pr_info("event trace: Could not activate trace point " \
239 "probe to " #call "\n"); \
240 return ret; \
241} \
242 \
243static void ftrace_raw_unreg_event_##call(void) \
244{ \
245 unregister_trace_##call(ftrace_raw_event_##call); \
246} \
247 \
248static struct trace_event ftrace_event_type_##call = { \
249 .trace = ftrace_raw_output_##call, \
250}; \
251 \
252static int ftrace_raw_init_event_##call(void) \
253{ \
254 int id; \
255 \
256 id = register_ftrace_event(&ftrace_event_type_##call); \
257 if (!id) \
258 return -ENODEV; \
259 event_##call.id = id; \
260 INIT_LIST_HEAD(&event_##call.fields); \
261 return 0; \
262} \
263 \
264static struct ftrace_event_call __used \
265__attribute__((__aligned__(4))) \
266__attribute__((section("_ftrace_events"))) event_##call = { \
267 .name = #call, \
268 .system = __stringify(TRACE_SYSTEM), \
269 .raw_init = ftrace_raw_init_event_##call, \
270 .regfunc = ftrace_raw_reg_event_##call, \
271 .unregfunc = ftrace_raw_unreg_event_##call, \
272 .show_format = ftrace_format_##call, \
273 .define_fields = ftrace_define_fields_##call, \
274 _TRACE_PROFILE_INIT(call) \
275}
276
277#include <trace/trace_event_types.h>
278
279#undef _TRACE_PROFILE
280#undef _TRACE_PROFILE_INIT
281
diff --git a/kernel/trace/trace_export.c b/kernel/trace/trace_export.c
new file mode 100644
index 000000000000..4d9952d3df50
--- /dev/null
+++ b/kernel/trace/trace_export.c
@@ -0,0 +1,102 @@
1/*
2 * trace_export.c - export basic ftrace utilities to user space
3 *
4 * Copyright (C) 2009 Steven Rostedt <srostedt@redhat.com>
5 */
6#include <linux/stringify.h>
7#include <linux/kallsyms.h>
8#include <linux/seq_file.h>
9#include <linux/debugfs.h>
10#include <linux/uaccess.h>
11#include <linux/ftrace.h>
12#include <linux/module.h>
13#include <linux/init.h>
14#include <linux/fs.h>
15
16#include "trace_output.h"
17
18
19#undef TRACE_STRUCT
20#define TRACE_STRUCT(args...) args
21
22#undef TRACE_FIELD
23#define TRACE_FIELD(type, item, assign) \
24 ret = trace_seq_printf(s, "\tfield:" #type " " #item ";\t" \
25 "offset:%u;\tsize:%u;\n", \
26 (unsigned int)offsetof(typeof(field), item), \
27 (unsigned int)sizeof(field.item)); \
28 if (!ret) \
29 return 0;
30
31
32#undef TRACE_FIELD_SPECIAL
33#define TRACE_FIELD_SPECIAL(type_item, item, cmd) \
34 ret = trace_seq_printf(s, "\tfield special:" #type_item ";\t" \
35 "offset:%u;\tsize:%u;\n", \
36 (unsigned int)offsetof(typeof(field), item), \
37 (unsigned int)sizeof(field.item)); \
38 if (!ret) \
39 return 0;
40
41#undef TRACE_FIELD_ZERO_CHAR
42#define TRACE_FIELD_ZERO_CHAR(item) \
43 ret = trace_seq_printf(s, "\tfield: char " #item ";\t" \
44 "offset:%u;\tsize:0;\n", \
45 (unsigned int)offsetof(typeof(field), item)); \
46 if (!ret) \
47 return 0;
48
49
50#undef TP_RAW_FMT
51#define TP_RAW_FMT(args...) args
52
53#undef TRACE_EVENT_FORMAT
54#define TRACE_EVENT_FORMAT(call, proto, args, fmt, tstruct, tpfmt) \
55static int \
56ftrace_format_##call(struct trace_seq *s) \
57{ \
58 struct args field; \
59 int ret; \
60 \
61 tstruct; \
62 \
63 trace_seq_printf(s, "\nprint fmt: \"%s\"\n", tpfmt); \
64 \
65 return ret; \
66}
67
68#include "trace_event_types.h"
69
70#undef TRACE_ZERO_CHAR
71#define TRACE_ZERO_CHAR(arg)
72
73#undef TRACE_FIELD
74#define TRACE_FIELD(type, item, assign)\
75 entry->item = assign;
76
77#undef TRACE_FIELD
78#define TRACE_FIELD(type, item, assign)\
79 entry->item = assign;
80
81#undef TP_CMD
82#define TP_CMD(cmd...) cmd
83
84#undef TRACE_ENTRY
85#define TRACE_ENTRY entry
86
87#undef TRACE_FIELD_SPECIAL
88#define TRACE_FIELD_SPECIAL(type_item, item, cmd) \
89 cmd;
90
91#undef TRACE_EVENT_FORMAT
92#define TRACE_EVENT_FORMAT(call, proto, args, fmt, tstruct, tpfmt) \
93 \
94static struct ftrace_event_call __used \
95__attribute__((__aligned__(4))) \
96__attribute__((section("_ftrace_events"))) event_##call = { \
97 .name = #call, \
98 .id = proto, \
99 .system = __stringify(TRACE_SYSTEM), \
100 .show_format = ftrace_format_##call, \
101}
102#include "trace_event_types.h"
diff --git a/kernel/trace/trace_functions.c b/kernel/trace/trace_functions.c
index 9236d7e25a16..c9a0b7df44ff 100644
--- a/kernel/trace/trace_functions.c
+++ b/kernel/trace/trace_functions.c
@@ -9,6 +9,7 @@
9 * Copyright (C) 2004-2006 Ingo Molnar 9 * Copyright (C) 2004-2006 Ingo Molnar
10 * Copyright (C) 2004 William Lee Irwin III 10 * Copyright (C) 2004 William Lee Irwin III
11 */ 11 */
12#include <linux/ring_buffer.h>
12#include <linux/debugfs.h> 13#include <linux/debugfs.h>
13#include <linux/uaccess.h> 14#include <linux/uaccess.h>
14#include <linux/ftrace.h> 15#include <linux/ftrace.h>
@@ -16,52 +17,388 @@
16 17
17#include "trace.h" 18#include "trace.h"
18 19
19static void start_function_trace(struct trace_array *tr) 20/* function tracing enabled */
21static int ftrace_function_enabled;
22
23static struct trace_array *func_trace;
24
25static void tracing_start_function_trace(void);
26static void tracing_stop_function_trace(void);
27
28static int function_trace_init(struct trace_array *tr)
20{ 29{
30 func_trace = tr;
21 tr->cpu = get_cpu(); 31 tr->cpu = get_cpu();
22 tracing_reset_online_cpus(tr);
23 put_cpu(); 32 put_cpu();
24 33
25 tracing_start_cmdline_record(); 34 tracing_start_cmdline_record();
26 tracing_start_function_trace(); 35 tracing_start_function_trace();
36 return 0;
27} 37}
28 38
29static void stop_function_trace(struct trace_array *tr) 39static void function_trace_reset(struct trace_array *tr)
30{ 40{
31 tracing_stop_function_trace(); 41 tracing_stop_function_trace();
32 tracing_stop_cmdline_record(); 42 tracing_stop_cmdline_record();
33} 43}
34 44
35static int function_trace_init(struct trace_array *tr) 45static void function_trace_start(struct trace_array *tr)
36{ 46{
37 start_function_trace(tr); 47 tracing_reset_online_cpus(tr);
38 return 0;
39} 48}
40 49
41static void function_trace_reset(struct trace_array *tr) 50static void
51function_trace_call_preempt_only(unsigned long ip, unsigned long parent_ip)
52{
53 struct trace_array *tr = func_trace;
54 struct trace_array_cpu *data;
55 unsigned long flags;
56 long disabled;
57 int cpu, resched;
58 int pc;
59
60 if (unlikely(!ftrace_function_enabled))
61 return;
62
63 pc = preempt_count();
64 resched = ftrace_preempt_disable();
65 local_save_flags(flags);
66 cpu = raw_smp_processor_id();
67 data = tr->data[cpu];
68 disabled = atomic_inc_return(&data->disabled);
69
70 if (likely(disabled == 1))
71 trace_function(tr, ip, parent_ip, flags, pc);
72
73 atomic_dec(&data->disabled);
74 ftrace_preempt_enable(resched);
75}
76
77static void
78function_trace_call(unsigned long ip, unsigned long parent_ip)
42{ 79{
43 stop_function_trace(tr); 80 struct trace_array *tr = func_trace;
81 struct trace_array_cpu *data;
82 unsigned long flags;
83 long disabled;
84 int cpu;
85 int pc;
86
87 if (unlikely(!ftrace_function_enabled))
88 return;
89
90 /*
91 * Need to use raw, since this must be called before the
92 * recursive protection is performed.
93 */
94 local_irq_save(flags);
95 cpu = raw_smp_processor_id();
96 data = tr->data[cpu];
97 disabled = atomic_inc_return(&data->disabled);
98
99 if (likely(disabled == 1)) {
100 pc = preempt_count();
101 trace_function(tr, ip, parent_ip, flags, pc);
102 }
103
104 atomic_dec(&data->disabled);
105 local_irq_restore(flags);
44} 106}
45 107
46static void function_trace_start(struct trace_array *tr) 108static void
109function_stack_trace_call(unsigned long ip, unsigned long parent_ip)
47{ 110{
48 tracing_reset_online_cpus(tr); 111 struct trace_array *tr = func_trace;
112 struct trace_array_cpu *data;
113 unsigned long flags;
114 long disabled;
115 int cpu;
116 int pc;
117
118 if (unlikely(!ftrace_function_enabled))
119 return;
120
121 /*
122 * Need to use raw, since this must be called before the
123 * recursive protection is performed.
124 */
125 local_irq_save(flags);
126 cpu = raw_smp_processor_id();
127 data = tr->data[cpu];
128 disabled = atomic_inc_return(&data->disabled);
129
130 if (likely(disabled == 1)) {
131 pc = preempt_count();
132 trace_function(tr, ip, parent_ip, flags, pc);
133 /*
134 * skip over 5 funcs:
135 * __ftrace_trace_stack,
136 * __trace_stack,
137 * function_stack_trace_call
138 * ftrace_list_func
139 * ftrace_call
140 */
141 __trace_stack(tr, flags, 5, pc);
142 }
143
144 atomic_dec(&data->disabled);
145 local_irq_restore(flags);
146}
147
148
149static struct ftrace_ops trace_ops __read_mostly =
150{
151 .func = function_trace_call,
152};
153
154static struct ftrace_ops trace_stack_ops __read_mostly =
155{
156 .func = function_stack_trace_call,
157};
158
159/* Our two options */
160enum {
161 TRACE_FUNC_OPT_STACK = 0x1,
162};
163
164static struct tracer_opt func_opts[] = {
165#ifdef CONFIG_STACKTRACE
166 { TRACER_OPT(func_stack_trace, TRACE_FUNC_OPT_STACK) },
167#endif
168 { } /* Always set a last empty entry */
169};
170
171static struct tracer_flags func_flags = {
172 .val = 0, /* By default: all flags disabled */
173 .opts = func_opts
174};
175
176static void tracing_start_function_trace(void)
177{
178 ftrace_function_enabled = 0;
179
180 if (trace_flags & TRACE_ITER_PREEMPTONLY)
181 trace_ops.func = function_trace_call_preempt_only;
182 else
183 trace_ops.func = function_trace_call;
184
185 if (func_flags.val & TRACE_FUNC_OPT_STACK)
186 register_ftrace_function(&trace_stack_ops);
187 else
188 register_ftrace_function(&trace_ops);
189
190 ftrace_function_enabled = 1;
191}
192
193static void tracing_stop_function_trace(void)
194{
195 ftrace_function_enabled = 0;
196 /* OK if they are not registered */
197 unregister_ftrace_function(&trace_stack_ops);
198 unregister_ftrace_function(&trace_ops);
199}
200
201static int func_set_flag(u32 old_flags, u32 bit, int set)
202{
203 if (bit == TRACE_FUNC_OPT_STACK) {
204 /* do nothing if already set */
205 if (!!set == !!(func_flags.val & TRACE_FUNC_OPT_STACK))
206 return 0;
207
208 if (set) {
209 unregister_ftrace_function(&trace_ops);
210 register_ftrace_function(&trace_stack_ops);
211 } else {
212 unregister_ftrace_function(&trace_stack_ops);
213 register_ftrace_function(&trace_ops);
214 }
215
216 return 0;
217 }
218
219 return -EINVAL;
49} 220}
50 221
51static struct tracer function_trace __read_mostly = 222static struct tracer function_trace __read_mostly =
52{ 223{
53 .name = "function", 224 .name = "function",
54 .init = function_trace_init, 225 .init = function_trace_init,
55 .reset = function_trace_reset, 226 .reset = function_trace_reset,
56 .start = function_trace_start, 227 .start = function_trace_start,
228 .wait_pipe = poll_wait_pipe,
229 .flags = &func_flags,
230 .set_flag = func_set_flag,
57#ifdef CONFIG_FTRACE_SELFTEST 231#ifdef CONFIG_FTRACE_SELFTEST
58 .selftest = trace_selftest_startup_function, 232 .selftest = trace_selftest_startup_function,
59#endif 233#endif
60}; 234};
61 235
236#ifdef CONFIG_DYNAMIC_FTRACE
237static void
238ftrace_traceon(unsigned long ip, unsigned long parent_ip, void **data)
239{
240 long *count = (long *)data;
241
242 if (tracing_is_on())
243 return;
244
245 if (!*count)
246 return;
247
248 if (*count != -1)
249 (*count)--;
250
251 tracing_on();
252}
253
254static void
255ftrace_traceoff(unsigned long ip, unsigned long parent_ip, void **data)
256{
257 long *count = (long *)data;
258
259 if (!tracing_is_on())
260 return;
261
262 if (!*count)
263 return;
264
265 if (*count != -1)
266 (*count)--;
267
268 tracing_off();
269}
270
271static int
272ftrace_trace_onoff_print(struct seq_file *m, unsigned long ip,
273 struct ftrace_probe_ops *ops, void *data);
274
275static struct ftrace_probe_ops traceon_probe_ops = {
276 .func = ftrace_traceon,
277 .print = ftrace_trace_onoff_print,
278};
279
280static struct ftrace_probe_ops traceoff_probe_ops = {
281 .func = ftrace_traceoff,
282 .print = ftrace_trace_onoff_print,
283};
284
285static int
286ftrace_trace_onoff_print(struct seq_file *m, unsigned long ip,
287 struct ftrace_probe_ops *ops, void *data)
288{
289 char str[KSYM_SYMBOL_LEN];
290 long count = (long)data;
291
292 kallsyms_lookup(ip, NULL, NULL, NULL, str);
293 seq_printf(m, "%s:", str);
294
295 if (ops == &traceon_probe_ops)
296 seq_printf(m, "traceon");
297 else
298 seq_printf(m, "traceoff");
299
300 if (count == -1)
301 seq_printf(m, ":unlimited\n");
302 else
303 seq_printf(m, ":count=%ld", count);
304 seq_putc(m, '\n');
305
306 return 0;
307}
308
309static int
310ftrace_trace_onoff_unreg(char *glob, char *cmd, char *param)
311{
312 struct ftrace_probe_ops *ops;
313
314 /* we register both traceon and traceoff to this callback */
315 if (strcmp(cmd, "traceon") == 0)
316 ops = &traceon_probe_ops;
317 else
318 ops = &traceoff_probe_ops;
319
320 unregister_ftrace_function_probe_func(glob, ops);
321
322 return 0;
323}
324
325static int
326ftrace_trace_onoff_callback(char *glob, char *cmd, char *param, int enable)
327{
328 struct ftrace_probe_ops *ops;
329 void *count = (void *)-1;
330 char *number;
331 int ret;
332
333 /* hash funcs only work with set_ftrace_filter */
334 if (!enable)
335 return -EINVAL;
336
337 if (glob[0] == '!')
338 return ftrace_trace_onoff_unreg(glob+1, cmd, param);
339
340 /* we register both traceon and traceoff to this callback */
341 if (strcmp(cmd, "traceon") == 0)
342 ops = &traceon_probe_ops;
343 else
344 ops = &traceoff_probe_ops;
345
346 if (!param)
347 goto out_reg;
348
349 number = strsep(&param, ":");
350
351 if (!strlen(number))
352 goto out_reg;
353
354 /*
355 * We use the callback data field (which is a pointer)
356 * as our counter.
357 */
358 ret = strict_strtoul(number, 0, (unsigned long *)&count);
359 if (ret)
360 return ret;
361
362 out_reg:
363 ret = register_ftrace_function_probe(glob, ops, count);
364
365 return ret;
366}
367
368static struct ftrace_func_command ftrace_traceon_cmd = {
369 .name = "traceon",
370 .func = ftrace_trace_onoff_callback,
371};
372
373static struct ftrace_func_command ftrace_traceoff_cmd = {
374 .name = "traceoff",
375 .func = ftrace_trace_onoff_callback,
376};
377
378static int __init init_func_cmd_traceon(void)
379{
380 int ret;
381
382 ret = register_ftrace_command(&ftrace_traceoff_cmd);
383 if (ret)
384 return ret;
385
386 ret = register_ftrace_command(&ftrace_traceon_cmd);
387 if (ret)
388 unregister_ftrace_command(&ftrace_traceoff_cmd);
389 return ret;
390}
391#else
392static inline int init_func_cmd_traceon(void)
393{
394 return 0;
395}
396#endif /* CONFIG_DYNAMIC_FTRACE */
397
62static __init int init_function_trace(void) 398static __init int init_function_trace(void)
63{ 399{
400 init_func_cmd_traceon();
64 return register_tracer(&function_trace); 401 return register_tracer(&function_trace);
65} 402}
66
67device_initcall(init_function_trace); 403device_initcall(init_function_trace);
404
diff --git a/kernel/trace/trace_functions_graph.c b/kernel/trace/trace_functions_graph.c
index 930c08e5b38e..d28687e7b3a7 100644
--- a/kernel/trace/trace_functions_graph.c
+++ b/kernel/trace/trace_functions_graph.c
@@ -1,7 +1,7 @@
1/* 1/*
2 * 2 *
3 * Function graph tracer. 3 * Function graph tracer.
4 * Copyright (c) 2008 Frederic Weisbecker <fweisbec@gmail.com> 4 * Copyright (c) 2008-2009 Frederic Weisbecker <fweisbec@gmail.com>
5 * Mostly borrowed from function tracer which 5 * Mostly borrowed from function tracer which
6 * is Copyright (c) Steven Rostedt <srostedt@redhat.com> 6 * is Copyright (c) Steven Rostedt <srostedt@redhat.com>
7 * 7 *
@@ -12,6 +12,12 @@
12#include <linux/fs.h> 12#include <linux/fs.h>
13 13
14#include "trace.h" 14#include "trace.h"
15#include "trace_output.h"
16
17struct fgraph_data {
18 pid_t last_pid;
19 int depth;
20};
15 21
16#define TRACE_GRAPH_INDENT 2 22#define TRACE_GRAPH_INDENT 2
17 23
@@ -20,9 +26,11 @@
20#define TRACE_GRAPH_PRINT_CPU 0x2 26#define TRACE_GRAPH_PRINT_CPU 0x2
21#define TRACE_GRAPH_PRINT_OVERHEAD 0x4 27#define TRACE_GRAPH_PRINT_OVERHEAD 0x4
22#define TRACE_GRAPH_PRINT_PROC 0x8 28#define TRACE_GRAPH_PRINT_PROC 0x8
29#define TRACE_GRAPH_PRINT_DURATION 0x10
30#define TRACE_GRAPH_PRINT_ABS_TIME 0X20
23 31
24static struct tracer_opt trace_opts[] = { 32static struct tracer_opt trace_opts[] = {
25 /* Display overruns ? */ 33 /* Display overruns? (for self-debug purpose) */
26 { TRACER_OPT(funcgraph-overrun, TRACE_GRAPH_PRINT_OVERRUN) }, 34 { TRACER_OPT(funcgraph-overrun, TRACE_GRAPH_PRINT_OVERRUN) },
27 /* Display CPU ? */ 35 /* Display CPU ? */
28 { TRACER_OPT(funcgraph-cpu, TRACE_GRAPH_PRINT_CPU) }, 36 { TRACER_OPT(funcgraph-cpu, TRACE_GRAPH_PRINT_CPU) },
@@ -30,26 +38,103 @@ static struct tracer_opt trace_opts[] = {
30 { TRACER_OPT(funcgraph-overhead, TRACE_GRAPH_PRINT_OVERHEAD) }, 38 { TRACER_OPT(funcgraph-overhead, TRACE_GRAPH_PRINT_OVERHEAD) },
31 /* Display proc name/pid */ 39 /* Display proc name/pid */
32 { TRACER_OPT(funcgraph-proc, TRACE_GRAPH_PRINT_PROC) }, 40 { TRACER_OPT(funcgraph-proc, TRACE_GRAPH_PRINT_PROC) },
41 /* Display duration of execution */
42 { TRACER_OPT(funcgraph-duration, TRACE_GRAPH_PRINT_DURATION) },
43 /* Display absolute time of an entry */
44 { TRACER_OPT(funcgraph-abstime, TRACE_GRAPH_PRINT_ABS_TIME) },
33 { } /* Empty entry */ 45 { } /* Empty entry */
34}; 46};
35 47
36static struct tracer_flags tracer_flags = { 48static struct tracer_flags tracer_flags = {
37 /* Don't display overruns and proc by default */ 49 /* Don't display overruns and proc by default */
38 .val = TRACE_GRAPH_PRINT_CPU | TRACE_GRAPH_PRINT_OVERHEAD, 50 .val = TRACE_GRAPH_PRINT_CPU | TRACE_GRAPH_PRINT_OVERHEAD |
51 TRACE_GRAPH_PRINT_DURATION,
39 .opts = trace_opts 52 .opts = trace_opts
40}; 53};
41 54
42/* pid on the last trace processed */ 55/* pid on the last trace processed */
43static pid_t last_pid[NR_CPUS] = { [0 ... NR_CPUS-1] = -1 };
44 56
45static int graph_trace_init(struct trace_array *tr) 57
58/* Add a function return address to the trace stack on thread info.*/
59int
60ftrace_push_return_trace(unsigned long ret, unsigned long func, int *depth)
46{ 61{
47 int cpu, ret; 62 unsigned long long calltime;
63 int index;
64
65 if (!current->ret_stack)
66 return -EBUSY;
67
68 /* The return trace stack is full */
69 if (current->curr_ret_stack == FTRACE_RETFUNC_DEPTH - 1) {
70 atomic_inc(&current->trace_overrun);
71 return -EBUSY;
72 }
73
74 calltime = trace_clock_local();
75
76 index = ++current->curr_ret_stack;
77 barrier();
78 current->ret_stack[index].ret = ret;
79 current->ret_stack[index].func = func;
80 current->ret_stack[index].calltime = calltime;
81 *depth = index;
82
83 return 0;
84}
85
86/* Retrieve a function return address to the trace stack on thread info.*/
87void
88ftrace_pop_return_trace(struct ftrace_graph_ret *trace, unsigned long *ret)
89{
90 int index;
91
92 index = current->curr_ret_stack;
93
94 if (unlikely(index < 0)) {
95 ftrace_graph_stop();
96 WARN_ON(1);
97 /* Might as well panic, otherwise we have no where to go */
98 *ret = (unsigned long)panic;
99 return;
100 }
48 101
49 for_each_online_cpu(cpu) 102 *ret = current->ret_stack[index].ret;
50 tracing_reset(tr, cpu); 103 trace->func = current->ret_stack[index].func;
104 trace->calltime = current->ret_stack[index].calltime;
105 trace->overrun = atomic_read(&current->trace_overrun);
106 trace->depth = index;
107 barrier();
108 current->curr_ret_stack--;
51 109
52 ret = register_ftrace_graph(&trace_graph_return, 110}
111
112/*
113 * Send the trace to the ring-buffer.
114 * @return the original return address.
115 */
116unsigned long ftrace_return_to_handler(void)
117{
118 struct ftrace_graph_ret trace;
119 unsigned long ret;
120
121 ftrace_pop_return_trace(&trace, &ret);
122 trace.rettime = trace_clock_local();
123 ftrace_graph_return(&trace);
124
125 if (unlikely(!ret)) {
126 ftrace_graph_stop();
127 WARN_ON(1);
128 /* Might as well panic. What else to do? */
129 ret = (unsigned long)panic;
130 }
131
132 return ret;
133}
134
135static int graph_trace_init(struct trace_array *tr)
136{
137 int ret = register_ftrace_graph(&trace_graph_return,
53 &trace_graph_entry); 138 &trace_graph_entry);
54 if (ret) 139 if (ret)
55 return ret; 140 return ret;
@@ -112,15 +197,15 @@ print_graph_cpu(struct trace_seq *s, int cpu)
112static enum print_line_t 197static enum print_line_t
113print_graph_proc(struct trace_seq *s, pid_t pid) 198print_graph_proc(struct trace_seq *s, pid_t pid)
114{ 199{
115 int i; 200 char comm[TASK_COMM_LEN];
116 int ret;
117 int len;
118 char comm[8];
119 int spaces = 0;
120 /* sign + log10(MAX_INT) + '\0' */ 201 /* sign + log10(MAX_INT) + '\0' */
121 char pid_str[11]; 202 char pid_str[11];
203 int spaces = 0;
204 int ret;
205 int len;
206 int i;
122 207
123 strncpy(comm, trace_find_cmdline(pid), 7); 208 trace_find_cmdline(pid, comm);
124 comm[7] = '\0'; 209 comm[7] = '\0';
125 sprintf(pid_str, "%d", pid); 210 sprintf(pid_str, "%d", pid);
126 211
@@ -153,17 +238,25 @@ print_graph_proc(struct trace_seq *s, pid_t pid)
153 238
154/* If the pid changed since the last trace, output this event */ 239/* If the pid changed since the last trace, output this event */
155static enum print_line_t 240static enum print_line_t
156verif_pid(struct trace_seq *s, pid_t pid, int cpu) 241verif_pid(struct trace_seq *s, pid_t pid, int cpu, struct fgraph_data *data)
157{ 242{
158 pid_t prev_pid; 243 pid_t prev_pid;
244 pid_t *last_pid;
159 int ret; 245 int ret;
160 246
161 if (last_pid[cpu] != -1 && last_pid[cpu] == pid) 247 if (!data)
162 return TRACE_TYPE_HANDLED; 248 return TRACE_TYPE_HANDLED;
163 249
164 prev_pid = last_pid[cpu]; 250 last_pid = &(per_cpu_ptr(data, cpu)->last_pid);
165 last_pid[cpu] = pid;
166 251
252 if (*last_pid == pid)
253 return TRACE_TYPE_HANDLED;
254
255 prev_pid = *last_pid;
256 *last_pid = pid;
257
258 if (prev_pid == -1)
259 return TRACE_TYPE_HANDLED;
167/* 260/*
168 * Context-switch trace line: 261 * Context-switch trace line:
169 262
@@ -175,34 +268,34 @@ verif_pid(struct trace_seq *s, pid_t pid, int cpu)
175 ret = trace_seq_printf(s, 268 ret = trace_seq_printf(s,
176 " ------------------------------------------\n"); 269 " ------------------------------------------\n");
177 if (!ret) 270 if (!ret)
178 TRACE_TYPE_PARTIAL_LINE; 271 return TRACE_TYPE_PARTIAL_LINE;
179 272
180 ret = print_graph_cpu(s, cpu); 273 ret = print_graph_cpu(s, cpu);
181 if (ret == TRACE_TYPE_PARTIAL_LINE) 274 if (ret == TRACE_TYPE_PARTIAL_LINE)
182 TRACE_TYPE_PARTIAL_LINE; 275 return TRACE_TYPE_PARTIAL_LINE;
183 276
184 ret = print_graph_proc(s, prev_pid); 277 ret = print_graph_proc(s, prev_pid);
185 if (ret == TRACE_TYPE_PARTIAL_LINE) 278 if (ret == TRACE_TYPE_PARTIAL_LINE)
186 TRACE_TYPE_PARTIAL_LINE; 279 return TRACE_TYPE_PARTIAL_LINE;
187 280
188 ret = trace_seq_printf(s, " => "); 281 ret = trace_seq_printf(s, " => ");
189 if (!ret) 282 if (!ret)
190 TRACE_TYPE_PARTIAL_LINE; 283 return TRACE_TYPE_PARTIAL_LINE;
191 284
192 ret = print_graph_proc(s, pid); 285 ret = print_graph_proc(s, pid);
193 if (ret == TRACE_TYPE_PARTIAL_LINE) 286 if (ret == TRACE_TYPE_PARTIAL_LINE)
194 TRACE_TYPE_PARTIAL_LINE; 287 return TRACE_TYPE_PARTIAL_LINE;
195 288
196 ret = trace_seq_printf(s, 289 ret = trace_seq_printf(s,
197 "\n ------------------------------------------\n\n"); 290 "\n ------------------------------------------\n\n");
198 if (!ret) 291 if (!ret)
199 TRACE_TYPE_PARTIAL_LINE; 292 return TRACE_TYPE_PARTIAL_LINE;
200 293
201 return ret; 294 return TRACE_TYPE_HANDLED;
202} 295}
203 296
204static bool 297static struct ftrace_graph_ret_entry *
205trace_branch_is_leaf(struct trace_iterator *iter, 298get_return_for_leaf(struct trace_iterator *iter,
206 struct ftrace_graph_ent_entry *curr) 299 struct ftrace_graph_ent_entry *curr)
207{ 300{
208 struct ring_buffer_iter *ring_iter; 301 struct ring_buffer_iter *ring_iter;
@@ -211,65 +304,123 @@ trace_branch_is_leaf(struct trace_iterator *iter,
211 304
212 ring_iter = iter->buffer_iter[iter->cpu]; 305 ring_iter = iter->buffer_iter[iter->cpu];
213 306
214 if (!ring_iter) 307 /* First peek to compare current entry and the next one */
215 return false; 308 if (ring_iter)
216 309 event = ring_buffer_iter_peek(ring_iter, NULL);
217 event = ring_buffer_iter_peek(ring_iter, NULL); 310 else {
311 /* We need to consume the current entry to see the next one */
312 ring_buffer_consume(iter->tr->buffer, iter->cpu, NULL);
313 event = ring_buffer_peek(iter->tr->buffer, iter->cpu,
314 NULL);
315 }
218 316
219 if (!event) 317 if (!event)
220 return false; 318 return NULL;
221 319
222 next = ring_buffer_event_data(event); 320 next = ring_buffer_event_data(event);
223 321
224 if (next->ent.type != TRACE_GRAPH_RET) 322 if (next->ent.type != TRACE_GRAPH_RET)
225 return false; 323 return NULL;
226 324
227 if (curr->ent.pid != next->ent.pid || 325 if (curr->ent.pid != next->ent.pid ||
228 curr->graph_ent.func != next->ret.func) 326 curr->graph_ent.func != next->ret.func)
229 return false; 327 return NULL;
328
329 /* this is a leaf, now advance the iterator */
330 if (ring_iter)
331 ring_buffer_read(ring_iter, NULL);
332
333 return next;
334}
335
336/* Signal a overhead of time execution to the output */
337static int
338print_graph_overhead(unsigned long long duration, struct trace_seq *s)
339{
340 /* If duration disappear, we don't need anything */
341 if (!(tracer_flags.val & TRACE_GRAPH_PRINT_DURATION))
342 return 1;
343
344 /* Non nested entry or return */
345 if (duration == -1)
346 return trace_seq_printf(s, " ");
347
348 if (tracer_flags.val & TRACE_GRAPH_PRINT_OVERHEAD) {
349 /* Duration exceeded 100 msecs */
350 if (duration > 100000ULL)
351 return trace_seq_printf(s, "! ");
230 352
231 return true; 353 /* Duration exceeded 10 msecs */
354 if (duration > 10000ULL)
355 return trace_seq_printf(s, "+ ");
356 }
357
358 return trace_seq_printf(s, " ");
359}
360
361static int print_graph_abs_time(u64 t, struct trace_seq *s)
362{
363 unsigned long usecs_rem;
364
365 usecs_rem = do_div(t, NSEC_PER_SEC);
366 usecs_rem /= 1000;
367
368 return trace_seq_printf(s, "%5lu.%06lu | ",
369 (unsigned long)t, usecs_rem);
232} 370}
233 371
234static enum print_line_t 372static enum print_line_t
235print_graph_irq(struct trace_seq *s, unsigned long addr, 373print_graph_irq(struct trace_iterator *iter, unsigned long addr,
236 enum trace_type type, int cpu, pid_t pid) 374 enum trace_type type, int cpu, pid_t pid)
237{ 375{
238 int ret; 376 int ret;
377 struct trace_seq *s = &iter->seq;
239 378
240 if (addr < (unsigned long)__irqentry_text_start || 379 if (addr < (unsigned long)__irqentry_text_start ||
241 addr >= (unsigned long)__irqentry_text_end) 380 addr >= (unsigned long)__irqentry_text_end)
242 return TRACE_TYPE_UNHANDLED; 381 return TRACE_TYPE_UNHANDLED;
243 382
244 if (type == TRACE_GRAPH_ENT) { 383 /* Absolute time */
245 ret = trace_seq_printf(s, "==========> | "); 384 if (tracer_flags.val & TRACE_GRAPH_PRINT_ABS_TIME) {
246 } else { 385 ret = print_graph_abs_time(iter->ts, s);
247 /* Cpu */ 386 if (!ret)
248 if (tracer_flags.val & TRACE_GRAPH_PRINT_CPU) { 387 return TRACE_TYPE_PARTIAL_LINE;
249 ret = print_graph_cpu(s, cpu); 388 }
250 if (ret == TRACE_TYPE_PARTIAL_LINE)
251 return TRACE_TYPE_PARTIAL_LINE;
252 }
253 /* Proc */
254 if (tracer_flags.val & TRACE_GRAPH_PRINT_PROC) {
255 ret = print_graph_proc(s, pid);
256 if (ret == TRACE_TYPE_PARTIAL_LINE)
257 return TRACE_TYPE_PARTIAL_LINE;
258 389
259 ret = trace_seq_printf(s, " | "); 390 /* Cpu */
260 if (!ret) 391 if (tracer_flags.val & TRACE_GRAPH_PRINT_CPU) {
261 return TRACE_TYPE_PARTIAL_LINE; 392 ret = print_graph_cpu(s, cpu);
262 } 393 if (ret == TRACE_TYPE_PARTIAL_LINE)
394 return TRACE_TYPE_PARTIAL_LINE;
395 }
396 /* Proc */
397 if (tracer_flags.val & TRACE_GRAPH_PRINT_PROC) {
398 ret = print_graph_proc(s, pid);
399 if (ret == TRACE_TYPE_PARTIAL_LINE)
400 return TRACE_TYPE_PARTIAL_LINE;
401 ret = trace_seq_printf(s, " | ");
402 if (!ret)
403 return TRACE_TYPE_PARTIAL_LINE;
404 }
263 405
264 /* No overhead */ 406 /* No overhead */
265 if (tracer_flags.val & TRACE_GRAPH_PRINT_OVERHEAD) { 407 ret = print_graph_overhead(-1, s);
266 ret = trace_seq_printf(s, " "); 408 if (!ret)
267 if (!ret) 409 return TRACE_TYPE_PARTIAL_LINE;
268 return TRACE_TYPE_PARTIAL_LINE; 410
269 } 411 if (type == TRACE_GRAPH_ENT)
412 ret = trace_seq_printf(s, "==========>");
413 else
414 ret = trace_seq_printf(s, "<==========");
415
416 if (!ret)
417 return TRACE_TYPE_PARTIAL_LINE;
418
419 /* Don't close the duration column if haven't one */
420 if (tracer_flags.val & TRACE_GRAPH_PRINT_DURATION)
421 trace_seq_printf(s, " |");
422 ret = trace_seq_printf(s, "\n");
270 423
271 ret = trace_seq_printf(s, "<========== |\n");
272 }
273 if (!ret) 424 if (!ret)
274 return TRACE_TYPE_PARTIAL_LINE; 425 return TRACE_TYPE_PARTIAL_LINE;
275 return TRACE_TYPE_HANDLED; 426 return TRACE_TYPE_HANDLED;
@@ -288,7 +439,7 @@ print_graph_duration(unsigned long long duration, struct trace_seq *s)
288 sprintf(msecs_str, "%lu", (unsigned long) duration); 439 sprintf(msecs_str, "%lu", (unsigned long) duration);
289 440
290 /* Print msecs */ 441 /* Print msecs */
291 ret = trace_seq_printf(s, msecs_str); 442 ret = trace_seq_printf(s, "%s", msecs_str);
292 if (!ret) 443 if (!ret)
293 return TRACE_TYPE_PARTIAL_LINE; 444 return TRACE_TYPE_PARTIAL_LINE;
294 445
@@ -321,52 +472,47 @@ print_graph_duration(unsigned long long duration, struct trace_seq *s)
321 472
322} 473}
323 474
324/* Signal a overhead of time execution to the output */
325static int
326print_graph_overhead(unsigned long long duration, struct trace_seq *s)
327{
328 /* Duration exceeded 100 msecs */
329 if (duration > 100000ULL)
330 return trace_seq_printf(s, "! ");
331
332 /* Duration exceeded 10 msecs */
333 if (duration > 10000ULL)
334 return trace_seq_printf(s, "+ ");
335
336 return trace_seq_printf(s, " ");
337}
338
339/* Case of a leaf function on its call entry */ 475/* Case of a leaf function on its call entry */
340static enum print_line_t 476static enum print_line_t
341print_graph_entry_leaf(struct trace_iterator *iter, 477print_graph_entry_leaf(struct trace_iterator *iter,
342 struct ftrace_graph_ent_entry *entry, struct trace_seq *s) 478 struct ftrace_graph_ent_entry *entry,
479 struct ftrace_graph_ret_entry *ret_entry, struct trace_seq *s)
343{ 480{
344 struct ftrace_graph_ret_entry *ret_entry; 481 struct fgraph_data *data = iter->private;
345 struct ftrace_graph_ret *graph_ret; 482 struct ftrace_graph_ret *graph_ret;
346 struct ring_buffer_event *event;
347 struct ftrace_graph_ent *call; 483 struct ftrace_graph_ent *call;
348 unsigned long long duration; 484 unsigned long long duration;
349 int ret; 485 int ret;
350 int i; 486 int i;
351 487
352 event = ring_buffer_read(iter->buffer_iter[iter->cpu], NULL);
353 ret_entry = ring_buffer_event_data(event);
354 graph_ret = &ret_entry->ret; 488 graph_ret = &ret_entry->ret;
355 call = &entry->graph_ent; 489 call = &entry->graph_ent;
356 duration = graph_ret->rettime - graph_ret->calltime; 490 duration = graph_ret->rettime - graph_ret->calltime;
357 491
358 /* Overhead */ 492 if (data) {
359 if (tracer_flags.val & TRACE_GRAPH_PRINT_OVERHEAD) { 493 int cpu = iter->cpu;
360 ret = print_graph_overhead(duration, s); 494 int *depth = &(per_cpu_ptr(data, cpu)->depth);
361 if (!ret) 495
362 return TRACE_TYPE_PARTIAL_LINE; 496 /*
497 * Comments display at + 1 to depth. Since
498 * this is a leaf function, keep the comments
499 * equal to this depth.
500 */
501 *depth = call->depth - 1;
363 } 502 }
364 503
365 /* Duration */ 504 /* Overhead */
366 ret = print_graph_duration(duration, s); 505 ret = print_graph_overhead(duration, s);
367 if (ret == TRACE_TYPE_PARTIAL_LINE) 506 if (!ret)
368 return TRACE_TYPE_PARTIAL_LINE; 507 return TRACE_TYPE_PARTIAL_LINE;
369 508
509 /* Duration */
510 if (tracer_flags.val & TRACE_GRAPH_PRINT_DURATION) {
511 ret = print_graph_duration(duration, s);
512 if (ret == TRACE_TYPE_PARTIAL_LINE)
513 return TRACE_TYPE_PARTIAL_LINE;
514 }
515
370 /* Function */ 516 /* Function */
371 for (i = 0; i < call->depth * TRACE_GRAPH_INDENT; i++) { 517 for (i = 0; i < call->depth * TRACE_GRAPH_INDENT; i++) {
372 ret = trace_seq_printf(s, " "); 518 ret = trace_seq_printf(s, " ");
@@ -386,33 +532,34 @@ print_graph_entry_leaf(struct trace_iterator *iter,
386} 532}
387 533
388static enum print_line_t 534static enum print_line_t
389print_graph_entry_nested(struct ftrace_graph_ent_entry *entry, 535print_graph_entry_nested(struct trace_iterator *iter,
390 struct trace_seq *s, pid_t pid, int cpu) 536 struct ftrace_graph_ent_entry *entry,
537 struct trace_seq *s, int cpu)
391{ 538{
392 int i;
393 int ret;
394 struct ftrace_graph_ent *call = &entry->graph_ent; 539 struct ftrace_graph_ent *call = &entry->graph_ent;
540 struct fgraph_data *data = iter->private;
541 int ret;
542 int i;
395 543
396 /* No overhead */ 544 if (data) {
397 if (tracer_flags.val & TRACE_GRAPH_PRINT_OVERHEAD) { 545 int cpu = iter->cpu;
398 ret = trace_seq_printf(s, " "); 546 int *depth = &(per_cpu_ptr(data, cpu)->depth);
399 if (!ret) 547
400 return TRACE_TYPE_PARTIAL_LINE; 548 *depth = call->depth;
401 } 549 }
402 550
403 /* Interrupt */ 551 /* No overhead */
404 ret = print_graph_irq(s, call->func, TRACE_GRAPH_ENT, cpu, pid); 552 ret = print_graph_overhead(-1, s);
405 if (ret == TRACE_TYPE_UNHANDLED) { 553 if (!ret)
406 /* No time */ 554 return TRACE_TYPE_PARTIAL_LINE;
555
556 /* No time */
557 if (tracer_flags.val & TRACE_GRAPH_PRINT_DURATION) {
407 ret = trace_seq_printf(s, " | "); 558 ret = trace_seq_printf(s, " | ");
408 if (!ret) 559 if (!ret)
409 return TRACE_TYPE_PARTIAL_LINE; 560 return TRACE_TYPE_PARTIAL_LINE;
410 } else {
411 if (ret == TRACE_TYPE_PARTIAL_LINE)
412 return TRACE_TYPE_PARTIAL_LINE;
413 } 561 }
414 562
415
416 /* Function */ 563 /* Function */
417 for (i = 0; i < call->depth * TRACE_GRAPH_INDENT; i++) { 564 for (i = 0; i < call->depth * TRACE_GRAPH_INDENT; i++) {
418 ret = trace_seq_printf(s, " "); 565 ret = trace_seq_printf(s, " ");
@@ -428,20 +575,40 @@ print_graph_entry_nested(struct ftrace_graph_ent_entry *entry,
428 if (!ret) 575 if (!ret)
429 return TRACE_TYPE_PARTIAL_LINE; 576 return TRACE_TYPE_PARTIAL_LINE;
430 577
431 return TRACE_TYPE_HANDLED; 578 /*
579 * we already consumed the current entry to check the next one
580 * and see if this is a leaf.
581 */
582 return TRACE_TYPE_NO_CONSUME;
432} 583}
433 584
434static enum print_line_t 585static enum print_line_t
435print_graph_entry(struct ftrace_graph_ent_entry *field, struct trace_seq *s, 586print_graph_prologue(struct trace_iterator *iter, struct trace_seq *s,
436 struct trace_iterator *iter, int cpu) 587 int type, unsigned long addr)
437{ 588{
438 int ret; 589 struct fgraph_data *data = iter->private;
439 struct trace_entry *ent = iter->ent; 590 struct trace_entry *ent = iter->ent;
591 int cpu = iter->cpu;
592 int ret;
440 593
441 /* Pid */ 594 /* Pid */
442 if (verif_pid(s, ent->pid, cpu) == TRACE_TYPE_PARTIAL_LINE) 595 if (verif_pid(s, ent->pid, cpu, data) == TRACE_TYPE_PARTIAL_LINE)
443 return TRACE_TYPE_PARTIAL_LINE; 596 return TRACE_TYPE_PARTIAL_LINE;
444 597
598 if (type) {
599 /* Interrupt */
600 ret = print_graph_irq(iter, addr, type, cpu, ent->pid);
601 if (ret == TRACE_TYPE_PARTIAL_LINE)
602 return TRACE_TYPE_PARTIAL_LINE;
603 }
604
605 /* Absolute time */
606 if (tracer_flags.val & TRACE_GRAPH_PRINT_ABS_TIME) {
607 ret = print_graph_abs_time(iter->ts, s);
608 if (!ret)
609 return TRACE_TYPE_PARTIAL_LINE;
610 }
611
445 /* Cpu */ 612 /* Cpu */
446 if (tracer_flags.val & TRACE_GRAPH_PRINT_CPU) { 613 if (tracer_flags.val & TRACE_GRAPH_PRINT_CPU) {
447 ret = print_graph_cpu(s, cpu); 614 ret = print_graph_cpu(s, cpu);
@@ -460,54 +627,65 @@ print_graph_entry(struct ftrace_graph_ent_entry *field, struct trace_seq *s,
460 return TRACE_TYPE_PARTIAL_LINE; 627 return TRACE_TYPE_PARTIAL_LINE;
461 } 628 }
462 629
463 if (trace_branch_is_leaf(iter, field)) 630 return 0;
464 return print_graph_entry_leaf(iter, field, s); 631}
632
633static enum print_line_t
634print_graph_entry(struct ftrace_graph_ent_entry *field, struct trace_seq *s,
635 struct trace_iterator *iter)
636{
637 int cpu = iter->cpu;
638 struct ftrace_graph_ent *call = &field->graph_ent;
639 struct ftrace_graph_ret_entry *leaf_ret;
640
641 if (print_graph_prologue(iter, s, TRACE_GRAPH_ENT, call->func))
642 return TRACE_TYPE_PARTIAL_LINE;
643
644 leaf_ret = get_return_for_leaf(iter, field);
645 if (leaf_ret)
646 return print_graph_entry_leaf(iter, field, leaf_ret, s);
465 else 647 else
466 return print_graph_entry_nested(field, s, iter->ent->pid, cpu); 648 return print_graph_entry_nested(iter, field, s, cpu);
467 649
468} 650}
469 651
470static enum print_line_t 652static enum print_line_t
471print_graph_return(struct ftrace_graph_ret *trace, struct trace_seq *s, 653print_graph_return(struct ftrace_graph_ret *trace, struct trace_seq *s,
472 struct trace_entry *ent, int cpu) 654 struct trace_entry *ent, struct trace_iterator *iter)
473{ 655{
474 int i;
475 int ret;
476 unsigned long long duration = trace->rettime - trace->calltime; 656 unsigned long long duration = trace->rettime - trace->calltime;
657 struct fgraph_data *data = iter->private;
658 pid_t pid = ent->pid;
659 int cpu = iter->cpu;
660 int ret;
661 int i;
477 662
478 /* Pid */ 663 if (data) {
479 if (verif_pid(s, ent->pid, cpu) == TRACE_TYPE_PARTIAL_LINE) 664 int cpu = iter->cpu;
480 return TRACE_TYPE_PARTIAL_LINE; 665 int *depth = &(per_cpu_ptr(data, cpu)->depth);
481 666
482 /* Cpu */ 667 /*
483 if (tracer_flags.val & TRACE_GRAPH_PRINT_CPU) { 668 * Comments display at + 1 to depth. This is the
484 ret = print_graph_cpu(s, cpu); 669 * return from a function, we now want the comments
485 if (ret == TRACE_TYPE_PARTIAL_LINE) 670 * to display at the same level of the bracket.
486 return TRACE_TYPE_PARTIAL_LINE; 671 */
672 *depth = trace->depth - 1;
487 } 673 }
488 674
489 /* Proc */ 675 if (print_graph_prologue(iter, s, 0, 0))
490 if (tracer_flags.val & TRACE_GRAPH_PRINT_PROC) { 676 return TRACE_TYPE_PARTIAL_LINE;
491 ret = print_graph_proc(s, ent->pid);
492 if (ret == TRACE_TYPE_PARTIAL_LINE)
493 return TRACE_TYPE_PARTIAL_LINE;
494
495 ret = trace_seq_printf(s, " | ");
496 if (!ret)
497 return TRACE_TYPE_PARTIAL_LINE;
498 }
499 677
500 /* Overhead */ 678 /* Overhead */
501 if (tracer_flags.val & TRACE_GRAPH_PRINT_OVERHEAD) { 679 ret = print_graph_overhead(duration, s);
502 ret = print_graph_overhead(duration, s); 680 if (!ret)
503 if (!ret) 681 return TRACE_TYPE_PARTIAL_LINE;
504 return TRACE_TYPE_PARTIAL_LINE;
505 }
506 682
507 /* Duration */ 683 /* Duration */
508 ret = print_graph_duration(duration, s); 684 if (tracer_flags.val & TRACE_GRAPH_PRINT_DURATION) {
509 if (ret == TRACE_TYPE_PARTIAL_LINE) 685 ret = print_graph_duration(duration, s);
510 return TRACE_TYPE_PARTIAL_LINE; 686 if (ret == TRACE_TYPE_PARTIAL_LINE)
687 return TRACE_TYPE_PARTIAL_LINE;
688 }
511 689
512 /* Closing brace */ 690 /* Closing brace */
513 for (i = 0; i < trace->depth * TRACE_GRAPH_INDENT; i++) { 691 for (i = 0; i < trace->depth * TRACE_GRAPH_INDENT; i++) {
@@ -528,7 +706,7 @@ print_graph_return(struct ftrace_graph_ret *trace, struct trace_seq *s,
528 return TRACE_TYPE_PARTIAL_LINE; 706 return TRACE_TYPE_PARTIAL_LINE;
529 } 707 }
530 708
531 ret = print_graph_irq(s, trace->func, TRACE_GRAPH_RET, cpu, ent->pid); 709 ret = print_graph_irq(iter, trace->func, TRACE_GRAPH_RET, cpu, pid);
532 if (ret == TRACE_TYPE_PARTIAL_LINE) 710 if (ret == TRACE_TYPE_PARTIAL_LINE)
533 return TRACE_TYPE_PARTIAL_LINE; 711 return TRACE_TYPE_PARTIAL_LINE;
534 712
@@ -536,61 +714,73 @@ print_graph_return(struct ftrace_graph_ret *trace, struct trace_seq *s,
536} 714}
537 715
538static enum print_line_t 716static enum print_line_t
539print_graph_comment(struct print_entry *trace, struct trace_seq *s, 717print_graph_comment(struct trace_seq *s, struct trace_entry *ent,
540 struct trace_entry *ent, struct trace_iterator *iter) 718 struct trace_iterator *iter)
541{ 719{
542 int i; 720 unsigned long sym_flags = (trace_flags & TRACE_ITER_SYM_MASK);
721 struct fgraph_data *data = iter->private;
722 struct trace_event *event;
723 int depth = 0;
543 int ret; 724 int ret;
725 int i;
544 726
545 /* Pid */ 727 if (data)
546 if (verif_pid(s, ent->pid, iter->cpu) == TRACE_TYPE_PARTIAL_LINE) 728 depth = per_cpu_ptr(data, iter->cpu)->depth;
547 return TRACE_TYPE_PARTIAL_LINE;
548
549 /* Cpu */
550 if (tracer_flags.val & TRACE_GRAPH_PRINT_CPU) {
551 ret = print_graph_cpu(s, iter->cpu);
552 if (ret == TRACE_TYPE_PARTIAL_LINE)
553 return TRACE_TYPE_PARTIAL_LINE;
554 }
555
556 /* Proc */
557 if (tracer_flags.val & TRACE_GRAPH_PRINT_PROC) {
558 ret = print_graph_proc(s, ent->pid);
559 if (ret == TRACE_TYPE_PARTIAL_LINE)
560 return TRACE_TYPE_PARTIAL_LINE;
561 729
562 ret = trace_seq_printf(s, " | "); 730 if (print_graph_prologue(iter, s, 0, 0))
563 if (!ret) 731 return TRACE_TYPE_PARTIAL_LINE;
564 return TRACE_TYPE_PARTIAL_LINE;
565 }
566 732
567 /* No overhead */ 733 /* No overhead */
568 if (tracer_flags.val & TRACE_GRAPH_PRINT_OVERHEAD) { 734 ret = print_graph_overhead(-1, s);
569 ret = trace_seq_printf(s, " "); 735 if (!ret)
736 return TRACE_TYPE_PARTIAL_LINE;
737
738 /* No time */
739 if (tracer_flags.val & TRACE_GRAPH_PRINT_DURATION) {
740 ret = trace_seq_printf(s, " | ");
570 if (!ret) 741 if (!ret)
571 return TRACE_TYPE_PARTIAL_LINE; 742 return TRACE_TYPE_PARTIAL_LINE;
572 } 743 }
573 744
574 /* No time */
575 ret = trace_seq_printf(s, " | ");
576 if (!ret)
577 return TRACE_TYPE_PARTIAL_LINE;
578
579 /* Indentation */ 745 /* Indentation */
580 if (trace->depth > 0) 746 if (depth > 0)
581 for (i = 0; i < (trace->depth + 1) * TRACE_GRAPH_INDENT; i++) { 747 for (i = 0; i < (depth + 1) * TRACE_GRAPH_INDENT; i++) {
582 ret = trace_seq_printf(s, " "); 748 ret = trace_seq_printf(s, " ");
583 if (!ret) 749 if (!ret)
584 return TRACE_TYPE_PARTIAL_LINE; 750 return TRACE_TYPE_PARTIAL_LINE;
585 } 751 }
586 752
587 /* The comment */ 753 /* The comment */
588 ret = trace_seq_printf(s, "/* %s", trace->buf); 754 ret = trace_seq_printf(s, "/* ");
589 if (!ret) 755 if (!ret)
590 return TRACE_TYPE_PARTIAL_LINE; 756 return TRACE_TYPE_PARTIAL_LINE;
591 757
592 if (ent->flags & TRACE_FLAG_CONT) 758 switch (iter->ent->type) {
593 trace_seq_print_cont(s, iter); 759 case TRACE_BPRINT:
760 ret = trace_print_bprintk_msg_only(iter);
761 if (ret != TRACE_TYPE_HANDLED)
762 return ret;
763 break;
764 case TRACE_PRINT:
765 ret = trace_print_printk_msg_only(iter);
766 if (ret != TRACE_TYPE_HANDLED)
767 return ret;
768 break;
769 default:
770 event = ftrace_find_event(ent->type);
771 if (!event)
772 return TRACE_TYPE_UNHANDLED;
773
774 ret = event->trace(iter, sym_flags);
775 if (ret != TRACE_TYPE_HANDLED)
776 return ret;
777 }
778
779 /* Strip ending newline */
780 if (s->buffer[s->len - 1] == '\n') {
781 s->buffer[s->len - 1] = '\0';
782 s->len--;
783 }
594 784
595 ret = trace_seq_printf(s, " */\n"); 785 ret = trace_seq_printf(s, " */\n");
596 if (!ret) 786 if (!ret)
@@ -603,62 +793,91 @@ print_graph_comment(struct print_entry *trace, struct trace_seq *s,
603enum print_line_t 793enum print_line_t
604print_graph_function(struct trace_iterator *iter) 794print_graph_function(struct trace_iterator *iter)
605{ 795{
606 struct trace_seq *s = &iter->seq;
607 struct trace_entry *entry = iter->ent; 796 struct trace_entry *entry = iter->ent;
797 struct trace_seq *s = &iter->seq;
608 798
609 switch (entry->type) { 799 switch (entry->type) {
610 case TRACE_GRAPH_ENT: { 800 case TRACE_GRAPH_ENT: {
611 struct ftrace_graph_ent_entry *field; 801 struct ftrace_graph_ent_entry *field;
612 trace_assign_type(field, entry); 802 trace_assign_type(field, entry);
613 return print_graph_entry(field, s, iter, 803 return print_graph_entry(field, s, iter);
614 iter->cpu);
615 } 804 }
616 case TRACE_GRAPH_RET: { 805 case TRACE_GRAPH_RET: {
617 struct ftrace_graph_ret_entry *field; 806 struct ftrace_graph_ret_entry *field;
618 trace_assign_type(field, entry); 807 trace_assign_type(field, entry);
619 return print_graph_return(&field->ret, s, entry, iter->cpu); 808 return print_graph_return(&field->ret, s, entry, iter);
620 }
621 case TRACE_PRINT: {
622 struct print_entry *field;
623 trace_assign_type(field, entry);
624 return print_graph_comment(field, s, entry, iter);
625 } 809 }
626 default: 810 default:
627 return TRACE_TYPE_UNHANDLED; 811 return print_graph_comment(s, entry, iter);
628 } 812 }
813
814 return TRACE_TYPE_HANDLED;
629} 815}
630 816
631static void print_graph_headers(struct seq_file *s) 817static void print_graph_headers(struct seq_file *s)
632{ 818{
633 /* 1st line */ 819 /* 1st line */
634 seq_printf(s, "# "); 820 seq_printf(s, "# ");
821 if (tracer_flags.val & TRACE_GRAPH_PRINT_ABS_TIME)
822 seq_printf(s, " TIME ");
635 if (tracer_flags.val & TRACE_GRAPH_PRINT_CPU) 823 if (tracer_flags.val & TRACE_GRAPH_PRINT_CPU)
636 seq_printf(s, "CPU "); 824 seq_printf(s, "CPU");
637 if (tracer_flags.val & TRACE_GRAPH_PRINT_PROC) 825 if (tracer_flags.val & TRACE_GRAPH_PRINT_PROC)
638 seq_printf(s, "TASK/PID "); 826 seq_printf(s, " TASK/PID ");
639 if (tracer_flags.val & TRACE_GRAPH_PRINT_OVERHEAD) 827 if (tracer_flags.val & TRACE_GRAPH_PRINT_DURATION)
640 seq_printf(s, "OVERHEAD/"); 828 seq_printf(s, " DURATION ");
641 seq_printf(s, "DURATION FUNCTION CALLS\n"); 829 seq_printf(s, " FUNCTION CALLS\n");
642 830
643 /* 2nd line */ 831 /* 2nd line */
644 seq_printf(s, "# "); 832 seq_printf(s, "# ");
833 if (tracer_flags.val & TRACE_GRAPH_PRINT_ABS_TIME)
834 seq_printf(s, " | ");
645 if (tracer_flags.val & TRACE_GRAPH_PRINT_CPU) 835 if (tracer_flags.val & TRACE_GRAPH_PRINT_CPU)
646 seq_printf(s, "| "); 836 seq_printf(s, "| ");
647 if (tracer_flags.val & TRACE_GRAPH_PRINT_PROC) 837 if (tracer_flags.val & TRACE_GRAPH_PRINT_PROC)
648 seq_printf(s, "| | "); 838 seq_printf(s, " | | ");
649 if (tracer_flags.val & TRACE_GRAPH_PRINT_OVERHEAD) { 839 if (tracer_flags.val & TRACE_GRAPH_PRINT_DURATION)
650 seq_printf(s, "| "); 840 seq_printf(s, " | | ");
651 seq_printf(s, "| | | | |\n"); 841 seq_printf(s, " | | | |\n");
652 } else 842}
653 seq_printf(s, " | | | | |\n"); 843
844static void graph_trace_open(struct trace_iterator *iter)
845{
846 /* pid and depth on the last trace processed */
847 struct fgraph_data *data = alloc_percpu(struct fgraph_data);
848 int cpu;
849
850 if (!data)
851 pr_warning("function graph tracer: not enough memory\n");
852 else
853 for_each_possible_cpu(cpu) {
854 pid_t *pid = &(per_cpu_ptr(data, cpu)->last_pid);
855 int *depth = &(per_cpu_ptr(data, cpu)->depth);
856 *pid = -1;
857 *depth = 0;
858 }
859
860 iter->private = data;
654} 861}
862
863static void graph_trace_close(struct trace_iterator *iter)
864{
865 free_percpu(iter->private);
866}
867
655static struct tracer graph_trace __read_mostly = { 868static struct tracer graph_trace __read_mostly = {
656 .name = "function_graph", 869 .name = "function_graph",
657 .init = graph_trace_init, 870 .open = graph_trace_open,
658 .reset = graph_trace_reset, 871 .close = graph_trace_close,
872 .wait_pipe = poll_wait_pipe,
873 .init = graph_trace_init,
874 .reset = graph_trace_reset,
659 .print_line = print_graph_function, 875 .print_line = print_graph_function,
660 .print_header = print_graph_headers, 876 .print_header = print_graph_headers,
661 .flags = &tracer_flags, 877 .flags = &tracer_flags,
878#ifdef CONFIG_FTRACE_SELFTEST
879 .selftest = trace_selftest_startup_function_graph,
880#endif
662}; 881};
663 882
664static __init int init_graph_trace(void) 883static __init int init_graph_trace(void)
diff --git a/kernel/trace/trace_hw_branches.c b/kernel/trace/trace_hw_branches.c
index 649df22d435f..7bfdf4c2347f 100644
--- a/kernel/trace/trace_hw_branches.c
+++ b/kernel/trace/trace_hw_branches.c
@@ -1,30 +1,53 @@
1/* 1/*
2 * h/w branch tracer for x86 based on bts 2 * h/w branch tracer for x86 based on bts
3 * 3 *
4 * Copyright (C) 2008 Markus Metzger <markus.t.metzger@gmail.com> 4 * Copyright (C) 2008-2009 Intel Corporation.
5 * 5 * Markus Metzger <markus.t.metzger@gmail.com>, 2008-2009
6 */ 6 */
7 7#include <linux/spinlock.h>
8#include <linux/module.h> 8#include <linux/kallsyms.h>
9#include <linux/fs.h>
10#include <linux/debugfs.h> 9#include <linux/debugfs.h>
11#include <linux/ftrace.h> 10#include <linux/ftrace.h>
12#include <linux/kallsyms.h> 11#include <linux/module.h>
12#include <linux/cpu.h>
13#include <linux/smp.h>
14#include <linux/fs.h>
13 15
14#include <asm/ds.h> 16#include <asm/ds.h>
15 17
16#include "trace.h" 18#include "trace.h"
19#include "trace_output.h"
17 20
18 21
19#define SIZEOF_BTS (1 << 13) 22#define SIZEOF_BTS (1 << 13)
20 23
24/*
25 * The tracer lock protects the below per-cpu tracer array.
26 * It needs to be held to:
27 * - start tracing on all cpus
28 * - stop tracing on all cpus
29 * - start tracing on a single hotplug cpu
30 * - stop tracing on a single hotplug cpu
31 * - read the trace from all cpus
32 * - read the trace from a single cpu
33 */
34static DEFINE_SPINLOCK(bts_tracer_lock);
21static DEFINE_PER_CPU(struct bts_tracer *, tracer); 35static DEFINE_PER_CPU(struct bts_tracer *, tracer);
22static DEFINE_PER_CPU(unsigned char[SIZEOF_BTS], buffer); 36static DEFINE_PER_CPU(unsigned char[SIZEOF_BTS], buffer);
23 37
24#define this_tracer per_cpu(tracer, smp_processor_id()) 38#define this_tracer per_cpu(tracer, smp_processor_id())
25#define this_buffer per_cpu(buffer, smp_processor_id()) 39#define this_buffer per_cpu(buffer, smp_processor_id())
26 40
41static int __read_mostly trace_hw_branches_enabled;
42static struct trace_array *hw_branch_trace __read_mostly;
43
27 44
45/*
46 * Start tracing on the current cpu.
47 * The argument is ignored.
48 *
49 * pre: bts_tracer_lock must be locked.
50 */
28static void bts_trace_start_cpu(void *arg) 51static void bts_trace_start_cpu(void *arg)
29{ 52{
30 if (this_tracer) 53 if (this_tracer)
@@ -42,14 +65,20 @@ static void bts_trace_start_cpu(void *arg)
42 65
43static void bts_trace_start(struct trace_array *tr) 66static void bts_trace_start(struct trace_array *tr)
44{ 67{
45 int cpu; 68 spin_lock(&bts_tracer_lock);
46 69
47 tracing_reset_online_cpus(tr); 70 on_each_cpu(bts_trace_start_cpu, NULL, 1);
71 trace_hw_branches_enabled = 1;
48 72
49 for_each_cpu(cpu, cpu_possible_mask) 73 spin_unlock(&bts_tracer_lock);
50 smp_call_function_single(cpu, bts_trace_start_cpu, NULL, 1);
51} 74}
52 75
76/*
77 * Stop tracing on the current cpu.
78 * The argument is ignored.
79 *
80 * pre: bts_tracer_lock must be locked.
81 */
53static void bts_trace_stop_cpu(void *arg) 82static void bts_trace_stop_cpu(void *arg)
54{ 83{
55 if (this_tracer) { 84 if (this_tracer) {
@@ -60,26 +89,60 @@ static void bts_trace_stop_cpu(void *arg)
60 89
61static void bts_trace_stop(struct trace_array *tr) 90static void bts_trace_stop(struct trace_array *tr)
62{ 91{
63 int cpu; 92 spin_lock(&bts_tracer_lock);
93
94 trace_hw_branches_enabled = 0;
95 on_each_cpu(bts_trace_stop_cpu, NULL, 1);
96
97 spin_unlock(&bts_tracer_lock);
98}
99
100static int __cpuinit bts_hotcpu_handler(struct notifier_block *nfb,
101 unsigned long action, void *hcpu)
102{
103 unsigned int cpu = (unsigned long)hcpu;
64 104
65 for_each_cpu(cpu, cpu_possible_mask) 105 spin_lock(&bts_tracer_lock);
106
107 if (!trace_hw_branches_enabled)
108 goto out;
109
110 switch (action) {
111 case CPU_ONLINE:
112 case CPU_DOWN_FAILED:
113 smp_call_function_single(cpu, bts_trace_start_cpu, NULL, 1);
114 break;
115 case CPU_DOWN_PREPARE:
66 smp_call_function_single(cpu, bts_trace_stop_cpu, NULL, 1); 116 smp_call_function_single(cpu, bts_trace_stop_cpu, NULL, 1);
117 break;
118 }
119
120 out:
121 spin_unlock(&bts_tracer_lock);
122 return NOTIFY_DONE;
67} 123}
68 124
125static struct notifier_block bts_hotcpu_notifier __cpuinitdata = {
126 .notifier_call = bts_hotcpu_handler
127};
128
69static int bts_trace_init(struct trace_array *tr) 129static int bts_trace_init(struct trace_array *tr)
70{ 130{
71 tracing_reset_online_cpus(tr); 131 hw_branch_trace = tr;
132
72 bts_trace_start(tr); 133 bts_trace_start(tr);
73 134
74 return 0; 135 return 0;
75} 136}
76 137
138static void bts_trace_reset(struct trace_array *tr)
139{
140 bts_trace_stop(tr);
141}
142
77static void bts_trace_print_header(struct seq_file *m) 143static void bts_trace_print_header(struct seq_file *m)
78{ 144{
79 seq_puts(m, 145 seq_puts(m, "# CPU# TO <- FROM\n");
80 "# CPU# FROM TO FUNCTION\n");
81 seq_puts(m,
82 "# | | | |\n");
83} 146}
84 147
85static enum print_line_t bts_trace_print_line(struct trace_iterator *iter) 148static enum print_line_t bts_trace_print_line(struct trace_iterator *iter)
@@ -87,15 +150,15 @@ static enum print_line_t bts_trace_print_line(struct trace_iterator *iter)
87 struct trace_entry *entry = iter->ent; 150 struct trace_entry *entry = iter->ent;
88 struct trace_seq *seq = &iter->seq; 151 struct trace_seq *seq = &iter->seq;
89 struct hw_branch_entry *it; 152 struct hw_branch_entry *it;
153 unsigned long symflags = TRACE_ITER_SYM_OFFSET;
90 154
91 trace_assign_type(it, entry); 155 trace_assign_type(it, entry);
92 156
93 if (entry->type == TRACE_HW_BRANCHES) { 157 if (entry->type == TRACE_HW_BRANCHES) {
94 if (trace_seq_printf(seq, "%4d ", entry->cpu) && 158 if (trace_seq_printf(seq, "%4d ", iter->cpu) &&
95 trace_seq_printf(seq, "0x%016llx -> 0x%016llx ", 159 seq_print_ip_sym(seq, it->to, symflags) &&
96 it->from, it->to) && 160 trace_seq_printf(seq, "\t <- ") &&
97 (!it->from || 161 seq_print_ip_sym(seq, it->from, symflags) &&
98 seq_print_ip_sym(seq, it->from, /* sym_flags = */ 0)) &&
99 trace_seq_printf(seq, "\n")) 162 trace_seq_printf(seq, "\n"))
100 return TRACE_TYPE_HANDLED; 163 return TRACE_TYPE_HANDLED;
101 return TRACE_TYPE_PARTIAL_LINE;; 164 return TRACE_TYPE_PARTIAL_LINE;;
@@ -103,26 +166,42 @@ static enum print_line_t bts_trace_print_line(struct trace_iterator *iter)
103 return TRACE_TYPE_UNHANDLED; 166 return TRACE_TYPE_UNHANDLED;
104} 167}
105 168
106void trace_hw_branch(struct trace_array *tr, u64 from, u64 to) 169void trace_hw_branch(u64 from, u64 to)
107{ 170{
171 struct trace_array *tr = hw_branch_trace;
108 struct ring_buffer_event *event; 172 struct ring_buffer_event *event;
109 struct hw_branch_entry *entry; 173 struct hw_branch_entry *entry;
110 unsigned long irq; 174 unsigned long irq1;
175 int cpu;
111 176
112 event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry), &irq); 177 if (unlikely(!tr))
113 if (!event)
114 return; 178 return;
179
180 if (unlikely(!trace_hw_branches_enabled))
181 return;
182
183 local_irq_save(irq1);
184 cpu = raw_smp_processor_id();
185 if (atomic_inc_return(&tr->data[cpu]->disabled) != 1)
186 goto out;
187
188 event = trace_buffer_lock_reserve(tr, TRACE_HW_BRANCHES,
189 sizeof(*entry), 0, 0);
190 if (!event)
191 goto out;
115 entry = ring_buffer_event_data(event); 192 entry = ring_buffer_event_data(event);
116 tracing_generic_entry_update(&entry->ent, 0, from); 193 tracing_generic_entry_update(&entry->ent, 0, from);
117 entry->ent.type = TRACE_HW_BRANCHES; 194 entry->ent.type = TRACE_HW_BRANCHES;
118 entry->ent.cpu = smp_processor_id();
119 entry->from = from; 195 entry->from = from;
120 entry->to = to; 196 entry->to = to;
121 ring_buffer_unlock_commit(tr->buffer, event, irq); 197 trace_buffer_unlock_commit(tr, event, 0, 0);
198
199 out:
200 atomic_dec(&tr->data[cpu]->disabled);
201 local_irq_restore(irq1);
122} 202}
123 203
124static void trace_bts_at(struct trace_array *tr, 204static void trace_bts_at(const struct bts_trace *trace, void *at)
125 const struct bts_trace *trace, void *at)
126{ 205{
127 struct bts_struct bts; 206 struct bts_struct bts;
128 int err = 0; 207 int err = 0;
@@ -137,18 +216,29 @@ static void trace_bts_at(struct trace_array *tr,
137 216
138 switch (bts.qualifier) { 217 switch (bts.qualifier) {
139 case BTS_BRANCH: 218 case BTS_BRANCH:
140 trace_hw_branch(tr, bts.variant.lbr.from, bts.variant.lbr.to); 219 trace_hw_branch(bts.variant.lbr.from, bts.variant.lbr.to);
141 break; 220 break;
142 } 221 }
143} 222}
144 223
224/*
225 * Collect the trace on the current cpu and write it into the ftrace buffer.
226 *
227 * pre: bts_tracer_lock must be locked
228 */
145static void trace_bts_cpu(void *arg) 229static void trace_bts_cpu(void *arg)
146{ 230{
147 struct trace_array *tr = (struct trace_array *) arg; 231 struct trace_array *tr = (struct trace_array *) arg;
148 const struct bts_trace *trace; 232 const struct bts_trace *trace;
149 unsigned char *at; 233 unsigned char *at;
150 234
151 if (!this_tracer) 235 if (unlikely(!tr))
236 return;
237
238 if (unlikely(atomic_read(&tr->data[raw_smp_processor_id()]->disabled)))
239 return;
240
241 if (unlikely(!this_tracer))
152 return; 242 return;
153 243
154 ds_suspend_bts(this_tracer); 244 ds_suspend_bts(this_tracer);
@@ -158,11 +248,11 @@ static void trace_bts_cpu(void *arg)
158 248
159 for (at = trace->ds.top; (void *)at < trace->ds.end; 249 for (at = trace->ds.top; (void *)at < trace->ds.end;
160 at += trace->ds.size) 250 at += trace->ds.size)
161 trace_bts_at(tr, trace, at); 251 trace_bts_at(trace, at);
162 252
163 for (at = trace->ds.begin; (void *)at < trace->ds.top; 253 for (at = trace->ds.begin; (void *)at < trace->ds.top;
164 at += trace->ds.size) 254 at += trace->ds.size)
165 trace_bts_at(tr, trace, at); 255 trace_bts_at(trace, at);
166 256
167out: 257out:
168 ds_resume_bts(this_tracer); 258 ds_resume_bts(this_tracer);
@@ -170,26 +260,43 @@ out:
170 260
171static void trace_bts_prepare(struct trace_iterator *iter) 261static void trace_bts_prepare(struct trace_iterator *iter)
172{ 262{
173 int cpu; 263 spin_lock(&bts_tracer_lock);
264
265 on_each_cpu(trace_bts_cpu, iter->tr, 1);
266
267 spin_unlock(&bts_tracer_lock);
268}
269
270static void trace_bts_close(struct trace_iterator *iter)
271{
272 tracing_reset_online_cpus(iter->tr);
273}
274
275void trace_hw_branch_oops(void)
276{
277 spin_lock(&bts_tracer_lock);
278
279 trace_bts_cpu(hw_branch_trace);
174 280
175 for_each_cpu(cpu, cpu_possible_mask) 281 spin_unlock(&bts_tracer_lock);
176 smp_call_function_single(cpu, trace_bts_cpu, iter->tr, 1);
177} 282}
178 283
179struct tracer bts_tracer __read_mostly = 284struct tracer bts_tracer __read_mostly =
180{ 285{
181 .name = "hw-branch-tracer", 286 .name = "hw-branch-tracer",
182 .init = bts_trace_init, 287 .init = bts_trace_init,
183 .reset = bts_trace_stop, 288 .reset = bts_trace_reset,
184 .print_header = bts_trace_print_header, 289 .print_header = bts_trace_print_header,
185 .print_line = bts_trace_print_line, 290 .print_line = bts_trace_print_line,
186 .start = bts_trace_start, 291 .start = bts_trace_start,
187 .stop = bts_trace_stop, 292 .stop = bts_trace_stop,
188 .open = trace_bts_prepare 293 .open = trace_bts_prepare,
294 .close = trace_bts_close
189}; 295};
190 296
191__init static int init_bts_trace(void) 297__init static int init_bts_trace(void)
192{ 298{
299 register_hotcpu_notifier(&bts_hotcpu_notifier);
193 return register_tracer(&bts_tracer); 300 return register_tracer(&bts_tracer);
194} 301}
195device_initcall(init_bts_trace); 302device_initcall(init_bts_trace);
diff --git a/kernel/trace/trace_irqsoff.c b/kernel/trace/trace_irqsoff.c
index 62a78d943534..b923d13e2fad 100644
--- a/kernel/trace/trace_irqsoff.c
+++ b/kernel/trace/trace_irqsoff.c
@@ -1,5 +1,5 @@
1/* 1/*
2 * trace irqs off criticall timings 2 * trace irqs off critical timings
3 * 3 *
4 * Copyright (C) 2007-2008 Steven Rostedt <srostedt@redhat.com> 4 * Copyright (C) 2007-2008 Steven Rostedt <srostedt@redhat.com>
5 * Copyright (C) 2008 Ingo Molnar <mingo@redhat.com> 5 * Copyright (C) 2008 Ingo Molnar <mingo@redhat.com>
@@ -32,6 +32,8 @@ enum {
32 32
33static int trace_type __read_mostly; 33static int trace_type __read_mostly;
34 34
35static int save_lat_flag;
36
35#ifdef CONFIG_PREEMPT_TRACER 37#ifdef CONFIG_PREEMPT_TRACER
36static inline int 38static inline int
37preempt_trace(void) 39preempt_trace(void)
@@ -95,7 +97,7 @@ irqsoff_tracer_call(unsigned long ip, unsigned long parent_ip)
95 disabled = atomic_inc_return(&data->disabled); 97 disabled = atomic_inc_return(&data->disabled);
96 98
97 if (likely(disabled == 1)) 99 if (likely(disabled == 1))
98 trace_function(tr, data, ip, parent_ip, flags, preempt_count()); 100 trace_function(tr, ip, parent_ip, flags, preempt_count());
99 101
100 atomic_dec(&data->disabled); 102 atomic_dec(&data->disabled);
101} 103}
@@ -153,7 +155,7 @@ check_critical_timing(struct trace_array *tr,
153 if (!report_latency(delta)) 155 if (!report_latency(delta))
154 goto out_unlock; 156 goto out_unlock;
155 157
156 trace_function(tr, data, CALLER_ADDR0, parent_ip, flags, pc); 158 trace_function(tr, CALLER_ADDR0, parent_ip, flags, pc);
157 159
158 latency = nsecs_to_usecs(delta); 160 latency = nsecs_to_usecs(delta);
159 161
@@ -177,7 +179,7 @@ out:
177 data->critical_sequence = max_sequence; 179 data->critical_sequence = max_sequence;
178 data->preempt_timestamp = ftrace_now(cpu); 180 data->preempt_timestamp = ftrace_now(cpu);
179 tracing_reset(tr, cpu); 181 tracing_reset(tr, cpu);
180 trace_function(tr, data, CALLER_ADDR0, parent_ip, flags, pc); 182 trace_function(tr, CALLER_ADDR0, parent_ip, flags, pc);
181} 183}
182 184
183static inline void 185static inline void
@@ -210,7 +212,7 @@ start_critical_timing(unsigned long ip, unsigned long parent_ip)
210 212
211 local_save_flags(flags); 213 local_save_flags(flags);
212 214
213 trace_function(tr, data, ip, parent_ip, flags, preempt_count()); 215 trace_function(tr, ip, parent_ip, flags, preempt_count());
214 216
215 per_cpu(tracing_cpu, cpu) = 1; 217 per_cpu(tracing_cpu, cpu) = 1;
216 218
@@ -244,7 +246,7 @@ stop_critical_timing(unsigned long ip, unsigned long parent_ip)
244 atomic_inc(&data->disabled); 246 atomic_inc(&data->disabled);
245 247
246 local_save_flags(flags); 248 local_save_flags(flags);
247 trace_function(tr, data, ip, parent_ip, flags, preempt_count()); 249 trace_function(tr, ip, parent_ip, flags, preempt_count());
248 check_critical_timing(tr, data, parent_ip ? : ip, cpu); 250 check_critical_timing(tr, data, parent_ip ? : ip, cpu);
249 data->critical_start = 0; 251 data->critical_start = 0;
250 atomic_dec(&data->disabled); 252 atomic_dec(&data->disabled);
@@ -353,33 +355,26 @@ void trace_preempt_off(unsigned long a0, unsigned long a1)
353} 355}
354#endif /* CONFIG_PREEMPT_TRACER */ 356#endif /* CONFIG_PREEMPT_TRACER */
355 357
356/*
357 * save_tracer_enabled is used to save the state of the tracer_enabled
358 * variable when we disable it when we open a trace output file.
359 */
360static int save_tracer_enabled;
361
362static void start_irqsoff_tracer(struct trace_array *tr) 358static void start_irqsoff_tracer(struct trace_array *tr)
363{ 359{
364 register_ftrace_function(&trace_ops); 360 register_ftrace_function(&trace_ops);
365 if (tracing_is_enabled()) { 361 if (tracing_is_enabled())
366 tracer_enabled = 1; 362 tracer_enabled = 1;
367 save_tracer_enabled = 1; 363 else
368 } else {
369 tracer_enabled = 0; 364 tracer_enabled = 0;
370 save_tracer_enabled = 0;
371 }
372} 365}
373 366
374static void stop_irqsoff_tracer(struct trace_array *tr) 367static void stop_irqsoff_tracer(struct trace_array *tr)
375{ 368{
376 tracer_enabled = 0; 369 tracer_enabled = 0;
377 save_tracer_enabled = 0;
378 unregister_ftrace_function(&trace_ops); 370 unregister_ftrace_function(&trace_ops);
379} 371}
380 372
381static void __irqsoff_tracer_init(struct trace_array *tr) 373static void __irqsoff_tracer_init(struct trace_array *tr)
382{ 374{
375 save_lat_flag = trace_flags & TRACE_ITER_LATENCY_FMT;
376 trace_flags |= TRACE_ITER_LATENCY_FMT;
377
383 tracing_max_latency = 0; 378 tracing_max_latency = 0;
384 irqsoff_trace = tr; 379 irqsoff_trace = tr;
385 /* make sure that the tracer is visible */ 380 /* make sure that the tracer is visible */
@@ -390,30 +385,19 @@ static void __irqsoff_tracer_init(struct trace_array *tr)
390static void irqsoff_tracer_reset(struct trace_array *tr) 385static void irqsoff_tracer_reset(struct trace_array *tr)
391{ 386{
392 stop_irqsoff_tracer(tr); 387 stop_irqsoff_tracer(tr);
388
389 if (!save_lat_flag)
390 trace_flags &= ~TRACE_ITER_LATENCY_FMT;
393} 391}
394 392
395static void irqsoff_tracer_start(struct trace_array *tr) 393static void irqsoff_tracer_start(struct trace_array *tr)
396{ 394{
397 tracer_enabled = 1; 395 tracer_enabled = 1;
398 save_tracer_enabled = 1;
399} 396}
400 397
401static void irqsoff_tracer_stop(struct trace_array *tr) 398static void irqsoff_tracer_stop(struct trace_array *tr)
402{ 399{
403 tracer_enabled = 0; 400 tracer_enabled = 0;
404 save_tracer_enabled = 0;
405}
406
407static void irqsoff_tracer_open(struct trace_iterator *iter)
408{
409 /* stop the trace while dumping */
410 tracer_enabled = 0;
411}
412
413static void irqsoff_tracer_close(struct trace_iterator *iter)
414{
415 /* restart tracing */
416 tracer_enabled = save_tracer_enabled;
417} 401}
418 402
419#ifdef CONFIG_IRQSOFF_TRACER 403#ifdef CONFIG_IRQSOFF_TRACER
@@ -431,8 +415,6 @@ static struct tracer irqsoff_tracer __read_mostly =
431 .reset = irqsoff_tracer_reset, 415 .reset = irqsoff_tracer_reset,
432 .start = irqsoff_tracer_start, 416 .start = irqsoff_tracer_start,
433 .stop = irqsoff_tracer_stop, 417 .stop = irqsoff_tracer_stop,
434 .open = irqsoff_tracer_open,
435 .close = irqsoff_tracer_close,
436 .print_max = 1, 418 .print_max = 1,
437#ifdef CONFIG_FTRACE_SELFTEST 419#ifdef CONFIG_FTRACE_SELFTEST
438 .selftest = trace_selftest_startup_irqsoff, 420 .selftest = trace_selftest_startup_irqsoff,
@@ -459,8 +441,6 @@ static struct tracer preemptoff_tracer __read_mostly =
459 .reset = irqsoff_tracer_reset, 441 .reset = irqsoff_tracer_reset,
460 .start = irqsoff_tracer_start, 442 .start = irqsoff_tracer_start,
461 .stop = irqsoff_tracer_stop, 443 .stop = irqsoff_tracer_stop,
462 .open = irqsoff_tracer_open,
463 .close = irqsoff_tracer_close,
464 .print_max = 1, 444 .print_max = 1,
465#ifdef CONFIG_FTRACE_SELFTEST 445#ifdef CONFIG_FTRACE_SELFTEST
466 .selftest = trace_selftest_startup_preemptoff, 446 .selftest = trace_selftest_startup_preemptoff,
@@ -489,8 +469,6 @@ static struct tracer preemptirqsoff_tracer __read_mostly =
489 .reset = irqsoff_tracer_reset, 469 .reset = irqsoff_tracer_reset,
490 .start = irqsoff_tracer_start, 470 .start = irqsoff_tracer_start,
491 .stop = irqsoff_tracer_stop, 471 .stop = irqsoff_tracer_stop,
492 .open = irqsoff_tracer_open,
493 .close = irqsoff_tracer_close,
494 .print_max = 1, 472 .print_max = 1,
495#ifdef CONFIG_FTRACE_SELFTEST 473#ifdef CONFIG_FTRACE_SELFTEST
496 .selftest = trace_selftest_startup_preemptirqsoff, 474 .selftest = trace_selftest_startup_preemptirqsoff,
diff --git a/kernel/trace/trace_mmiotrace.c b/kernel/trace/trace_mmiotrace.c
index 80e503ef6136..8e37fcddd8b4 100644
--- a/kernel/trace/trace_mmiotrace.c
+++ b/kernel/trace/trace_mmiotrace.c
@@ -12,6 +12,7 @@
12#include <asm/atomic.h> 12#include <asm/atomic.h>
13 13
14#include "trace.h" 14#include "trace.h"
15#include "trace_output.h"
15 16
16struct header_iter { 17struct header_iter {
17 struct pci_dev *dev; 18 struct pci_dev *dev;
@@ -183,21 +184,22 @@ static enum print_line_t mmio_print_rw(struct trace_iterator *iter)
183 switch (rw->opcode) { 184 switch (rw->opcode) {
184 case MMIO_READ: 185 case MMIO_READ:
185 ret = trace_seq_printf(s, 186 ret = trace_seq_printf(s,
186 "R %d %lu.%06lu %d 0x%llx 0x%lx 0x%lx %d\n", 187 "R %d %u.%06lu %d 0x%llx 0x%lx 0x%lx %d\n",
187 rw->width, secs, usec_rem, rw->map_id, 188 rw->width, secs, usec_rem, rw->map_id,
188 (unsigned long long)rw->phys, 189 (unsigned long long)rw->phys,
189 rw->value, rw->pc, 0); 190 rw->value, rw->pc, 0);
190 break; 191 break;
191 case MMIO_WRITE: 192 case MMIO_WRITE:
192 ret = trace_seq_printf(s, 193 ret = trace_seq_printf(s,
193 "W %d %lu.%06lu %d 0x%llx 0x%lx 0x%lx %d\n", 194 "W %d %u.%06lu %d 0x%llx 0x%lx 0x%lx %d\n",
194 rw->width, secs, usec_rem, rw->map_id, 195 rw->width, secs, usec_rem, rw->map_id,
195 (unsigned long long)rw->phys, 196 (unsigned long long)rw->phys,
196 rw->value, rw->pc, 0); 197 rw->value, rw->pc, 0);
197 break; 198 break;
198 case MMIO_UNKNOWN_OP: 199 case MMIO_UNKNOWN_OP:
199 ret = trace_seq_printf(s, 200 ret = trace_seq_printf(s,
200 "UNKNOWN %lu.%06lu %d 0x%llx %02x,%02x,%02x 0x%lx %d\n", 201 "UNKNOWN %u.%06lu %d 0x%llx %02lx,%02lx,"
202 "%02lx 0x%lx %d\n",
201 secs, usec_rem, rw->map_id, 203 secs, usec_rem, rw->map_id,
202 (unsigned long long)rw->phys, 204 (unsigned long long)rw->phys,
203 (rw->value >> 16) & 0xff, (rw->value >> 8) & 0xff, 205 (rw->value >> 16) & 0xff, (rw->value >> 8) & 0xff,
@@ -229,14 +231,14 @@ static enum print_line_t mmio_print_map(struct trace_iterator *iter)
229 switch (m->opcode) { 231 switch (m->opcode) {
230 case MMIO_PROBE: 232 case MMIO_PROBE:
231 ret = trace_seq_printf(s, 233 ret = trace_seq_printf(s,
232 "MAP %lu.%06lu %d 0x%llx 0x%lx 0x%lx 0x%lx %d\n", 234 "MAP %u.%06lu %d 0x%llx 0x%lx 0x%lx 0x%lx %d\n",
233 secs, usec_rem, m->map_id, 235 secs, usec_rem, m->map_id,
234 (unsigned long long)m->phys, m->virt, m->len, 236 (unsigned long long)m->phys, m->virt, m->len,
235 0UL, 0); 237 0UL, 0);
236 break; 238 break;
237 case MMIO_UNPROBE: 239 case MMIO_UNPROBE:
238 ret = trace_seq_printf(s, 240 ret = trace_seq_printf(s,
239 "UNMAP %lu.%06lu %d 0x%lx %d\n", 241 "UNMAP %u.%06lu %d 0x%lx %d\n",
240 secs, usec_rem, m->map_id, 0UL, 0); 242 secs, usec_rem, m->map_id, 0UL, 0);
241 break; 243 break;
242 default: 244 default:
@@ -255,18 +257,15 @@ static enum print_line_t mmio_print_mark(struct trace_iterator *iter)
255 const char *msg = print->buf; 257 const char *msg = print->buf;
256 struct trace_seq *s = &iter->seq; 258 struct trace_seq *s = &iter->seq;
257 unsigned long long t = ns2usecs(iter->ts); 259 unsigned long long t = ns2usecs(iter->ts);
258 unsigned long usec_rem = do_div(t, 1000000ULL); 260 unsigned long usec_rem = do_div(t, USEC_PER_SEC);
259 unsigned secs = (unsigned long)t; 261 unsigned secs = (unsigned long)t;
260 int ret; 262 int ret;
261 263
262 /* The trailing newline must be in the message. */ 264 /* The trailing newline must be in the message. */
263 ret = trace_seq_printf(s, "MARK %lu.%06lu %s", secs, usec_rem, msg); 265 ret = trace_seq_printf(s, "MARK %u.%06lu %s", secs, usec_rem, msg);
264 if (!ret) 266 if (!ret)
265 return TRACE_TYPE_PARTIAL_LINE; 267 return TRACE_TYPE_PARTIAL_LINE;
266 268
267 if (entry->flags & TRACE_FLAG_CONT)
268 trace_seq_print_cont(s, iter);
269
270 return TRACE_TYPE_HANDLED; 269 return TRACE_TYPE_HANDLED;
271} 270}
272 271
@@ -308,21 +307,17 @@ static void __trace_mmiotrace_rw(struct trace_array *tr,
308{ 307{
309 struct ring_buffer_event *event; 308 struct ring_buffer_event *event;
310 struct trace_mmiotrace_rw *entry; 309 struct trace_mmiotrace_rw *entry;
311 unsigned long irq_flags; 310 int pc = preempt_count();
312 311
313 event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry), 312 event = trace_buffer_lock_reserve(tr, TRACE_MMIO_RW,
314 &irq_flags); 313 sizeof(*entry), 0, pc);
315 if (!event) { 314 if (!event) {
316 atomic_inc(&dropped_count); 315 atomic_inc(&dropped_count);
317 return; 316 return;
318 } 317 }
319 entry = ring_buffer_event_data(event); 318 entry = ring_buffer_event_data(event);
320 tracing_generic_entry_update(&entry->ent, 0, preempt_count());
321 entry->ent.type = TRACE_MMIO_RW;
322 entry->rw = *rw; 319 entry->rw = *rw;
323 ring_buffer_unlock_commit(tr->buffer, event, irq_flags); 320 trace_buffer_unlock_commit(tr, event, 0, pc);
324
325 trace_wake_up();
326} 321}
327 322
328void mmio_trace_rw(struct mmiotrace_rw *rw) 323void mmio_trace_rw(struct mmiotrace_rw *rw)
@@ -338,21 +333,17 @@ static void __trace_mmiotrace_map(struct trace_array *tr,
338{ 333{
339 struct ring_buffer_event *event; 334 struct ring_buffer_event *event;
340 struct trace_mmiotrace_map *entry; 335 struct trace_mmiotrace_map *entry;
341 unsigned long irq_flags; 336 int pc = preempt_count();
342 337
343 event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry), 338 event = trace_buffer_lock_reserve(tr, TRACE_MMIO_MAP,
344 &irq_flags); 339 sizeof(*entry), 0, pc);
345 if (!event) { 340 if (!event) {
346 atomic_inc(&dropped_count); 341 atomic_inc(&dropped_count);
347 return; 342 return;
348 } 343 }
349 entry = ring_buffer_event_data(event); 344 entry = ring_buffer_event_data(event);
350 tracing_generic_entry_update(&entry->ent, 0, preempt_count());
351 entry->ent.type = TRACE_MMIO_MAP;
352 entry->map = *map; 345 entry->map = *map;
353 ring_buffer_unlock_commit(tr->buffer, event, irq_flags); 346 trace_buffer_unlock_commit(tr, event, 0, pc);
354
355 trace_wake_up();
356} 347}
357 348
358void mmio_trace_mapping(struct mmiotrace_map *map) 349void mmio_trace_mapping(struct mmiotrace_map *map)
@@ -368,5 +359,5 @@ void mmio_trace_mapping(struct mmiotrace_map *map)
368 359
369int mmio_trace_printk(const char *fmt, va_list args) 360int mmio_trace_printk(const char *fmt, va_list args)
370{ 361{
371 return trace_vprintk(0, -1, fmt, args); 362 return trace_vprintk(0, fmt, args);
372} 363}
diff --git a/kernel/trace/trace_nop.c b/kernel/trace/trace_nop.c
index b9767acd30ac..394f94417e2f 100644
--- a/kernel/trace/trace_nop.c
+++ b/kernel/trace/trace_nop.c
@@ -47,12 +47,7 @@ static void stop_nop_trace(struct trace_array *tr)
47 47
48static int nop_trace_init(struct trace_array *tr) 48static int nop_trace_init(struct trace_array *tr)
49{ 49{
50 int cpu;
51 ctx_trace = tr; 50 ctx_trace = tr;
52
53 for_each_online_cpu(cpu)
54 tracing_reset(tr, cpu);
55
56 start_nop_trace(tr); 51 start_nop_trace(tr);
57 return 0; 52 return 0;
58} 53}
@@ -96,6 +91,7 @@ struct tracer nop_trace __read_mostly =
96 .name = "nop", 91 .name = "nop",
97 .init = nop_trace_init, 92 .init = nop_trace_init,
98 .reset = nop_trace_reset, 93 .reset = nop_trace_reset,
94 .wait_pipe = poll_wait_pipe,
99#ifdef CONFIG_FTRACE_SELFTEST 95#ifdef CONFIG_FTRACE_SELFTEST
100 .selftest = trace_selftest_startup_nop, 96 .selftest = trace_selftest_startup_nop,
101#endif 97#endif
diff --git a/kernel/trace/trace_output.c b/kernel/trace/trace_output.c
new file mode 100644
index 000000000000..d72b9a63b247
--- /dev/null
+++ b/kernel/trace/trace_output.c
@@ -0,0 +1,1017 @@
1/*
2 * trace_output.c
3 *
4 * Copyright (C) 2008 Red Hat Inc, Steven Rostedt <srostedt@redhat.com>
5 *
6 */
7
8#include <linux/module.h>
9#include <linux/mutex.h>
10#include <linux/ftrace.h>
11
12#include "trace_output.h"
13
14/* must be a power of 2 */
15#define EVENT_HASHSIZE 128
16
17static DEFINE_MUTEX(trace_event_mutex);
18static struct hlist_head event_hash[EVENT_HASHSIZE] __read_mostly;
19
20static int next_event_type = __TRACE_LAST_TYPE + 1;
21
22enum print_line_t trace_print_bprintk_msg_only(struct trace_iterator *iter)
23{
24 struct trace_seq *s = &iter->seq;
25 struct trace_entry *entry = iter->ent;
26 struct bprint_entry *field;
27 int ret;
28
29 trace_assign_type(field, entry);
30
31 ret = trace_seq_bprintf(s, field->fmt, field->buf);
32 if (!ret)
33 return TRACE_TYPE_PARTIAL_LINE;
34
35 return TRACE_TYPE_HANDLED;
36}
37
38enum print_line_t trace_print_printk_msg_only(struct trace_iterator *iter)
39{
40 struct trace_seq *s = &iter->seq;
41 struct trace_entry *entry = iter->ent;
42 struct print_entry *field;
43 int ret;
44
45 trace_assign_type(field, entry);
46
47 ret = trace_seq_printf(s, "%s", field->buf);
48 if (!ret)
49 return TRACE_TYPE_PARTIAL_LINE;
50
51 return TRACE_TYPE_HANDLED;
52}
53
54/**
55 * trace_seq_printf - sequence printing of trace information
56 * @s: trace sequence descriptor
57 * @fmt: printf format string
58 *
59 * The tracer may use either sequence operations or its own
60 * copy to user routines. To simplify formating of a trace
61 * trace_seq_printf is used to store strings into a special
62 * buffer (@s). Then the output may be either used by
63 * the sequencer or pulled into another buffer.
64 */
65int
66trace_seq_printf(struct trace_seq *s, const char *fmt, ...)
67{
68 int len = (PAGE_SIZE - 1) - s->len;
69 va_list ap;
70 int ret;
71
72 if (!len)
73 return 0;
74
75 va_start(ap, fmt);
76 ret = vsnprintf(s->buffer + s->len, len, fmt, ap);
77 va_end(ap);
78
79 /* If we can't write it all, don't bother writing anything */
80 if (ret >= len)
81 return 0;
82
83 s->len += ret;
84
85 return len;
86}
87
88int trace_seq_bprintf(struct trace_seq *s, const char *fmt, const u32 *binary)
89{
90 int len = (PAGE_SIZE - 1) - s->len;
91 int ret;
92
93 if (!len)
94 return 0;
95
96 ret = bstr_printf(s->buffer + s->len, len, fmt, binary);
97
98 /* If we can't write it all, don't bother writing anything */
99 if (ret >= len)
100 return 0;
101
102 s->len += ret;
103
104 return len;
105}
106
107/**
108 * trace_seq_puts - trace sequence printing of simple string
109 * @s: trace sequence descriptor
110 * @str: simple string to record
111 *
112 * The tracer may use either the sequence operations or its own
113 * copy to user routines. This function records a simple string
114 * into a special buffer (@s) for later retrieval by a sequencer
115 * or other mechanism.
116 */
117int trace_seq_puts(struct trace_seq *s, const char *str)
118{
119 int len = strlen(str);
120
121 if (len > ((PAGE_SIZE - 1) - s->len))
122 return 0;
123
124 memcpy(s->buffer + s->len, str, len);
125 s->len += len;
126
127 return len;
128}
129
130int trace_seq_putc(struct trace_seq *s, unsigned char c)
131{
132 if (s->len >= (PAGE_SIZE - 1))
133 return 0;
134
135 s->buffer[s->len++] = c;
136
137 return 1;
138}
139
140int trace_seq_putmem(struct trace_seq *s, const void *mem, size_t len)
141{
142 if (len > ((PAGE_SIZE - 1) - s->len))
143 return 0;
144
145 memcpy(s->buffer + s->len, mem, len);
146 s->len += len;
147
148 return len;
149}
150
151int trace_seq_putmem_hex(struct trace_seq *s, const void *mem, size_t len)
152{
153 unsigned char hex[HEX_CHARS];
154 const unsigned char *data = mem;
155 int i, j;
156
157#ifdef __BIG_ENDIAN
158 for (i = 0, j = 0; i < len; i++) {
159#else
160 for (i = len-1, j = 0; i >= 0; i--) {
161#endif
162 hex[j++] = hex_asc_hi(data[i]);
163 hex[j++] = hex_asc_lo(data[i]);
164 }
165 hex[j++] = ' ';
166
167 return trace_seq_putmem(s, hex, j);
168}
169
170void *trace_seq_reserve(struct trace_seq *s, size_t len)
171{
172 void *ret;
173
174 if (len > ((PAGE_SIZE - 1) - s->len))
175 return NULL;
176
177 ret = s->buffer + s->len;
178 s->len += len;
179
180 return ret;
181}
182
183int trace_seq_path(struct trace_seq *s, struct path *path)
184{
185 unsigned char *p;
186
187 if (s->len >= (PAGE_SIZE - 1))
188 return 0;
189 p = d_path(path, s->buffer + s->len, PAGE_SIZE - s->len);
190 if (!IS_ERR(p)) {
191 p = mangle_path(s->buffer + s->len, p, "\n");
192 if (p) {
193 s->len = p - s->buffer;
194 return 1;
195 }
196 } else {
197 s->buffer[s->len++] = '?';
198 return 1;
199 }
200
201 return 0;
202}
203
204#ifdef CONFIG_KRETPROBES
205static inline const char *kretprobed(const char *name)
206{
207 static const char tramp_name[] = "kretprobe_trampoline";
208 int size = sizeof(tramp_name);
209
210 if (strncmp(tramp_name, name, size) == 0)
211 return "[unknown/kretprobe'd]";
212 return name;
213}
214#else
215static inline const char *kretprobed(const char *name)
216{
217 return name;
218}
219#endif /* CONFIG_KRETPROBES */
220
221static int
222seq_print_sym_short(struct trace_seq *s, const char *fmt, unsigned long address)
223{
224#ifdef CONFIG_KALLSYMS
225 char str[KSYM_SYMBOL_LEN];
226 const char *name;
227
228 kallsyms_lookup(address, NULL, NULL, NULL, str);
229
230 name = kretprobed(str);
231
232 return trace_seq_printf(s, fmt, name);
233#endif
234 return 1;
235}
236
237static int
238seq_print_sym_offset(struct trace_seq *s, const char *fmt,
239 unsigned long address)
240{
241#ifdef CONFIG_KALLSYMS
242 char str[KSYM_SYMBOL_LEN];
243 const char *name;
244
245 sprint_symbol(str, address);
246 name = kretprobed(str);
247
248 return trace_seq_printf(s, fmt, name);
249#endif
250 return 1;
251}
252
253#ifndef CONFIG_64BIT
254# define IP_FMT "%08lx"
255#else
256# define IP_FMT "%016lx"
257#endif
258
259int seq_print_user_ip(struct trace_seq *s, struct mm_struct *mm,
260 unsigned long ip, unsigned long sym_flags)
261{
262 struct file *file = NULL;
263 unsigned long vmstart = 0;
264 int ret = 1;
265
266 if (mm) {
267 const struct vm_area_struct *vma;
268
269 down_read(&mm->mmap_sem);
270 vma = find_vma(mm, ip);
271 if (vma) {
272 file = vma->vm_file;
273 vmstart = vma->vm_start;
274 }
275 if (file) {
276 ret = trace_seq_path(s, &file->f_path);
277 if (ret)
278 ret = trace_seq_printf(s, "[+0x%lx]",
279 ip - vmstart);
280 }
281 up_read(&mm->mmap_sem);
282 }
283 if (ret && ((sym_flags & TRACE_ITER_SYM_ADDR) || !file))
284 ret = trace_seq_printf(s, " <" IP_FMT ">", ip);
285 return ret;
286}
287
288int
289seq_print_userip_objs(const struct userstack_entry *entry, struct trace_seq *s,
290 unsigned long sym_flags)
291{
292 struct mm_struct *mm = NULL;
293 int ret = 1;
294 unsigned int i;
295
296 if (trace_flags & TRACE_ITER_SYM_USEROBJ) {
297 struct task_struct *task;
298 /*
299 * we do the lookup on the thread group leader,
300 * since individual threads might have already quit!
301 */
302 rcu_read_lock();
303 task = find_task_by_vpid(entry->ent.tgid);
304 if (task)
305 mm = get_task_mm(task);
306 rcu_read_unlock();
307 }
308
309 for (i = 0; i < FTRACE_STACK_ENTRIES; i++) {
310 unsigned long ip = entry->caller[i];
311
312 if (ip == ULONG_MAX || !ret)
313 break;
314 if (i && ret)
315 ret = trace_seq_puts(s, " <- ");
316 if (!ip) {
317 if (ret)
318 ret = trace_seq_puts(s, "??");
319 continue;
320 }
321 if (!ret)
322 break;
323 if (ret)
324 ret = seq_print_user_ip(s, mm, ip, sym_flags);
325 }
326
327 if (mm)
328 mmput(mm);
329 return ret;
330}
331
332int
333seq_print_ip_sym(struct trace_seq *s, unsigned long ip, unsigned long sym_flags)
334{
335 int ret;
336
337 if (!ip)
338 return trace_seq_printf(s, "0");
339
340 if (sym_flags & TRACE_ITER_SYM_OFFSET)
341 ret = seq_print_sym_offset(s, "%s", ip);
342 else
343 ret = seq_print_sym_short(s, "%s", ip);
344
345 if (!ret)
346 return 0;
347
348 if (sym_flags & TRACE_ITER_SYM_ADDR)
349 ret = trace_seq_printf(s, " <" IP_FMT ">", ip);
350 return ret;
351}
352
353static int
354lat_print_generic(struct trace_seq *s, struct trace_entry *entry, int cpu)
355{
356 int hardirq, softirq;
357 char comm[TASK_COMM_LEN];
358
359 trace_find_cmdline(entry->pid, comm);
360 hardirq = entry->flags & TRACE_FLAG_HARDIRQ;
361 softirq = entry->flags & TRACE_FLAG_SOFTIRQ;
362
363 if (!trace_seq_printf(s, "%8.8s-%-5d %3d%c%c%c",
364 comm, entry->pid, cpu,
365 (entry->flags & TRACE_FLAG_IRQS_OFF) ? 'd' :
366 (entry->flags & TRACE_FLAG_IRQS_NOSUPPORT) ?
367 'X' : '.',
368 (entry->flags & TRACE_FLAG_NEED_RESCHED) ?
369 'N' : '.',
370 (hardirq && softirq) ? 'H' :
371 hardirq ? 'h' : softirq ? 's' : '.'))
372 return 0;
373
374 if (entry->preempt_count)
375 return trace_seq_printf(s, "%x", entry->preempt_count);
376 return trace_seq_puts(s, ".");
377}
378
379static unsigned long preempt_mark_thresh = 100;
380
381static int
382lat_print_timestamp(struct trace_seq *s, u64 abs_usecs,
383 unsigned long rel_usecs)
384{
385 return trace_seq_printf(s, " %4lldus%c: ", abs_usecs,
386 rel_usecs > preempt_mark_thresh ? '!' :
387 rel_usecs > 1 ? '+' : ' ');
388}
389
390int trace_print_context(struct trace_iterator *iter)
391{
392 struct trace_seq *s = &iter->seq;
393 struct trace_entry *entry = iter->ent;
394 unsigned long long t = ns2usecs(iter->ts);
395 unsigned long usec_rem = do_div(t, USEC_PER_SEC);
396 unsigned long secs = (unsigned long)t;
397 char comm[TASK_COMM_LEN];
398
399 trace_find_cmdline(entry->pid, comm);
400
401 return trace_seq_printf(s, "%16s-%-5d [%03d] %5lu.%06lu: ",
402 comm, entry->pid, iter->cpu, secs, usec_rem);
403}
404
405int trace_print_lat_context(struct trace_iterator *iter)
406{
407 u64 next_ts;
408 int ret;
409 struct trace_seq *s = &iter->seq;
410 struct trace_entry *entry = iter->ent,
411 *next_entry = trace_find_next_entry(iter, NULL,
412 &next_ts);
413 unsigned long verbose = (trace_flags & TRACE_ITER_VERBOSE);
414 unsigned long abs_usecs = ns2usecs(iter->ts - iter->tr->time_start);
415 unsigned long rel_usecs;
416
417 if (!next_entry)
418 next_ts = iter->ts;
419 rel_usecs = ns2usecs(next_ts - iter->ts);
420
421 if (verbose) {
422 char comm[TASK_COMM_LEN];
423
424 trace_find_cmdline(entry->pid, comm);
425
426 ret = trace_seq_printf(s, "%16s %5d %3d %d %08x %08lx [%08lx]"
427 " %ld.%03ldms (+%ld.%03ldms): ", comm,
428 entry->pid, iter->cpu, entry->flags,
429 entry->preempt_count, iter->idx,
430 ns2usecs(iter->ts),
431 abs_usecs / USEC_PER_MSEC,
432 abs_usecs % USEC_PER_MSEC,
433 rel_usecs / USEC_PER_MSEC,
434 rel_usecs % USEC_PER_MSEC);
435 } else {
436 ret = lat_print_generic(s, entry, iter->cpu);
437 if (ret)
438 ret = lat_print_timestamp(s, abs_usecs, rel_usecs);
439 }
440
441 return ret;
442}
443
444static const char state_to_char[] = TASK_STATE_TO_CHAR_STR;
445
446static int task_state_char(unsigned long state)
447{
448 int bit = state ? __ffs(state) + 1 : 0;
449
450 return bit < sizeof(state_to_char) - 1 ? state_to_char[bit] : '?';
451}
452
453/**
454 * ftrace_find_event - find a registered event
455 * @type: the type of event to look for
456 *
457 * Returns an event of type @type otherwise NULL
458 */
459struct trace_event *ftrace_find_event(int type)
460{
461 struct trace_event *event;
462 struct hlist_node *n;
463 unsigned key;
464
465 key = type & (EVENT_HASHSIZE - 1);
466
467 hlist_for_each_entry_rcu(event, n, &event_hash[key], node) {
468 if (event->type == type)
469 return event;
470 }
471
472 return NULL;
473}
474
475/**
476 * register_ftrace_event - register output for an event type
477 * @event: the event type to register
478 *
479 * Event types are stored in a hash and this hash is used to
480 * find a way to print an event. If the @event->type is set
481 * then it will use that type, otherwise it will assign a
482 * type to use.
483 *
484 * If you assign your own type, please make sure it is added
485 * to the trace_type enum in trace.h, to avoid collisions
486 * with the dynamic types.
487 *
488 * Returns the event type number or zero on error.
489 */
490int register_ftrace_event(struct trace_event *event)
491{
492 unsigned key;
493 int ret = 0;
494
495 mutex_lock(&trace_event_mutex);
496
497 if (!event) {
498 ret = next_event_type++;
499 goto out;
500 }
501
502 if (!event->type)
503 event->type = next_event_type++;
504 else if (event->type > __TRACE_LAST_TYPE) {
505 printk(KERN_WARNING "Need to add type to trace.h\n");
506 WARN_ON(1);
507 }
508
509 if (ftrace_find_event(event->type))
510 goto out;
511
512 if (event->trace == NULL)
513 event->trace = trace_nop_print;
514 if (event->raw == NULL)
515 event->raw = trace_nop_print;
516 if (event->hex == NULL)
517 event->hex = trace_nop_print;
518 if (event->binary == NULL)
519 event->binary = trace_nop_print;
520
521 key = event->type & (EVENT_HASHSIZE - 1);
522
523 hlist_add_head_rcu(&event->node, &event_hash[key]);
524
525 ret = event->type;
526 out:
527 mutex_unlock(&trace_event_mutex);
528
529 return ret;
530}
531
532/**
533 * unregister_ftrace_event - remove a no longer used event
534 * @event: the event to remove
535 */
536int unregister_ftrace_event(struct trace_event *event)
537{
538 mutex_lock(&trace_event_mutex);
539 hlist_del(&event->node);
540 mutex_unlock(&trace_event_mutex);
541
542 return 0;
543}
544
545/*
546 * Standard events
547 */
548
549enum print_line_t trace_nop_print(struct trace_iterator *iter, int flags)
550{
551 return TRACE_TYPE_HANDLED;
552}
553
554/* TRACE_FN */
555static enum print_line_t trace_fn_trace(struct trace_iterator *iter, int flags)
556{
557 struct ftrace_entry *field;
558 struct trace_seq *s = &iter->seq;
559
560 trace_assign_type(field, iter->ent);
561
562 if (!seq_print_ip_sym(s, field->ip, flags))
563 goto partial;
564
565 if ((flags & TRACE_ITER_PRINT_PARENT) && field->parent_ip) {
566 if (!trace_seq_printf(s, " <-"))
567 goto partial;
568 if (!seq_print_ip_sym(s,
569 field->parent_ip,
570 flags))
571 goto partial;
572 }
573 if (!trace_seq_printf(s, "\n"))
574 goto partial;
575
576 return TRACE_TYPE_HANDLED;
577
578 partial:
579 return TRACE_TYPE_PARTIAL_LINE;
580}
581
582static enum print_line_t trace_fn_raw(struct trace_iterator *iter, int flags)
583{
584 struct ftrace_entry *field;
585
586 trace_assign_type(field, iter->ent);
587
588 if (!trace_seq_printf(&iter->seq, "%lx %lx\n",
589 field->ip,
590 field->parent_ip))
591 return TRACE_TYPE_PARTIAL_LINE;
592
593 return TRACE_TYPE_HANDLED;
594}
595
596static enum print_line_t trace_fn_hex(struct trace_iterator *iter, int flags)
597{
598 struct ftrace_entry *field;
599 struct trace_seq *s = &iter->seq;
600
601 trace_assign_type(field, iter->ent);
602
603 SEQ_PUT_HEX_FIELD_RET(s, field->ip);
604 SEQ_PUT_HEX_FIELD_RET(s, field->parent_ip);
605
606 return TRACE_TYPE_HANDLED;
607}
608
609static enum print_line_t trace_fn_bin(struct trace_iterator *iter, int flags)
610{
611 struct ftrace_entry *field;
612 struct trace_seq *s = &iter->seq;
613
614 trace_assign_type(field, iter->ent);
615
616 SEQ_PUT_FIELD_RET(s, field->ip);
617 SEQ_PUT_FIELD_RET(s, field->parent_ip);
618
619 return TRACE_TYPE_HANDLED;
620}
621
622static struct trace_event trace_fn_event = {
623 .type = TRACE_FN,
624 .trace = trace_fn_trace,
625 .raw = trace_fn_raw,
626 .hex = trace_fn_hex,
627 .binary = trace_fn_bin,
628};
629
630/* TRACE_CTX an TRACE_WAKE */
631static enum print_line_t trace_ctxwake_print(struct trace_iterator *iter,
632 char *delim)
633{
634 struct ctx_switch_entry *field;
635 char comm[TASK_COMM_LEN];
636 int S, T;
637
638
639 trace_assign_type(field, iter->ent);
640
641 T = task_state_char(field->next_state);
642 S = task_state_char(field->prev_state);
643 trace_find_cmdline(field->next_pid, comm);
644 if (!trace_seq_printf(&iter->seq,
645 " %5d:%3d:%c %s [%03d] %5d:%3d:%c %s\n",
646 field->prev_pid,
647 field->prev_prio,
648 S, delim,
649 field->next_cpu,
650 field->next_pid,
651 field->next_prio,
652 T, comm))
653 return TRACE_TYPE_PARTIAL_LINE;
654
655 return TRACE_TYPE_HANDLED;
656}
657
658static enum print_line_t trace_ctx_print(struct trace_iterator *iter, int flags)
659{
660 return trace_ctxwake_print(iter, "==>");
661}
662
663static enum print_line_t trace_wake_print(struct trace_iterator *iter,
664 int flags)
665{
666 return trace_ctxwake_print(iter, " +");
667}
668
669static int trace_ctxwake_raw(struct trace_iterator *iter, char S)
670{
671 struct ctx_switch_entry *field;
672 int T;
673
674 trace_assign_type(field, iter->ent);
675
676 if (!S)
677 task_state_char(field->prev_state);
678 T = task_state_char(field->next_state);
679 if (!trace_seq_printf(&iter->seq, "%d %d %c %d %d %d %c\n",
680 field->prev_pid,
681 field->prev_prio,
682 S,
683 field->next_cpu,
684 field->next_pid,
685 field->next_prio,
686 T))
687 return TRACE_TYPE_PARTIAL_LINE;
688
689 return TRACE_TYPE_HANDLED;
690}
691
692static enum print_line_t trace_ctx_raw(struct trace_iterator *iter, int flags)
693{
694 return trace_ctxwake_raw(iter, 0);
695}
696
697static enum print_line_t trace_wake_raw(struct trace_iterator *iter, int flags)
698{
699 return trace_ctxwake_raw(iter, '+');
700}
701
702
703static int trace_ctxwake_hex(struct trace_iterator *iter, char S)
704{
705 struct ctx_switch_entry *field;
706 struct trace_seq *s = &iter->seq;
707 int T;
708
709 trace_assign_type(field, iter->ent);
710
711 if (!S)
712 task_state_char(field->prev_state);
713 T = task_state_char(field->next_state);
714
715 SEQ_PUT_HEX_FIELD_RET(s, field->prev_pid);
716 SEQ_PUT_HEX_FIELD_RET(s, field->prev_prio);
717 SEQ_PUT_HEX_FIELD_RET(s, S);
718 SEQ_PUT_HEX_FIELD_RET(s, field->next_cpu);
719 SEQ_PUT_HEX_FIELD_RET(s, field->next_pid);
720 SEQ_PUT_HEX_FIELD_RET(s, field->next_prio);
721 SEQ_PUT_HEX_FIELD_RET(s, T);
722
723 return TRACE_TYPE_HANDLED;
724}
725
726static enum print_line_t trace_ctx_hex(struct trace_iterator *iter, int flags)
727{
728 return trace_ctxwake_hex(iter, 0);
729}
730
731static enum print_line_t trace_wake_hex(struct trace_iterator *iter, int flags)
732{
733 return trace_ctxwake_hex(iter, '+');
734}
735
736static enum print_line_t trace_ctxwake_bin(struct trace_iterator *iter,
737 int flags)
738{
739 struct ctx_switch_entry *field;
740 struct trace_seq *s = &iter->seq;
741
742 trace_assign_type(field, iter->ent);
743
744 SEQ_PUT_FIELD_RET(s, field->prev_pid);
745 SEQ_PUT_FIELD_RET(s, field->prev_prio);
746 SEQ_PUT_FIELD_RET(s, field->prev_state);
747 SEQ_PUT_FIELD_RET(s, field->next_pid);
748 SEQ_PUT_FIELD_RET(s, field->next_prio);
749 SEQ_PUT_FIELD_RET(s, field->next_state);
750
751 return TRACE_TYPE_HANDLED;
752}
753
754static struct trace_event trace_ctx_event = {
755 .type = TRACE_CTX,
756 .trace = trace_ctx_print,
757 .raw = trace_ctx_raw,
758 .hex = trace_ctx_hex,
759 .binary = trace_ctxwake_bin,
760};
761
762static struct trace_event trace_wake_event = {
763 .type = TRACE_WAKE,
764 .trace = trace_wake_print,
765 .raw = trace_wake_raw,
766 .hex = trace_wake_hex,
767 .binary = trace_ctxwake_bin,
768};
769
770/* TRACE_SPECIAL */
771static enum print_line_t trace_special_print(struct trace_iterator *iter,
772 int flags)
773{
774 struct special_entry *field;
775
776 trace_assign_type(field, iter->ent);
777
778 if (!trace_seq_printf(&iter->seq, "# %ld %ld %ld\n",
779 field->arg1,
780 field->arg2,
781 field->arg3))
782 return TRACE_TYPE_PARTIAL_LINE;
783
784 return TRACE_TYPE_HANDLED;
785}
786
787static enum print_line_t trace_special_hex(struct trace_iterator *iter,
788 int flags)
789{
790 struct special_entry *field;
791 struct trace_seq *s = &iter->seq;
792
793 trace_assign_type(field, iter->ent);
794
795 SEQ_PUT_HEX_FIELD_RET(s, field->arg1);
796 SEQ_PUT_HEX_FIELD_RET(s, field->arg2);
797 SEQ_PUT_HEX_FIELD_RET(s, field->arg3);
798
799 return TRACE_TYPE_HANDLED;
800}
801
802static enum print_line_t trace_special_bin(struct trace_iterator *iter,
803 int flags)
804{
805 struct special_entry *field;
806 struct trace_seq *s = &iter->seq;
807
808 trace_assign_type(field, iter->ent);
809
810 SEQ_PUT_FIELD_RET(s, field->arg1);
811 SEQ_PUT_FIELD_RET(s, field->arg2);
812 SEQ_PUT_FIELD_RET(s, field->arg3);
813
814 return TRACE_TYPE_HANDLED;
815}
816
817static struct trace_event trace_special_event = {
818 .type = TRACE_SPECIAL,
819 .trace = trace_special_print,
820 .raw = trace_special_print,
821 .hex = trace_special_hex,
822 .binary = trace_special_bin,
823};
824
825/* TRACE_STACK */
826
827static enum print_line_t trace_stack_print(struct trace_iterator *iter,
828 int flags)
829{
830 struct stack_entry *field;
831 struct trace_seq *s = &iter->seq;
832 int i;
833
834 trace_assign_type(field, iter->ent);
835
836 for (i = 0; i < FTRACE_STACK_ENTRIES; i++) {
837 if (i) {
838 if (!trace_seq_puts(s, " <= "))
839 goto partial;
840
841 if (!seq_print_ip_sym(s, field->caller[i], flags))
842 goto partial;
843 }
844 if (!trace_seq_puts(s, "\n"))
845 goto partial;
846 }
847
848 return TRACE_TYPE_HANDLED;
849
850 partial:
851 return TRACE_TYPE_PARTIAL_LINE;
852}
853
854static struct trace_event trace_stack_event = {
855 .type = TRACE_STACK,
856 .trace = trace_stack_print,
857 .raw = trace_special_print,
858 .hex = trace_special_hex,
859 .binary = trace_special_bin,
860};
861
862/* TRACE_USER_STACK */
863static enum print_line_t trace_user_stack_print(struct trace_iterator *iter,
864 int flags)
865{
866 struct userstack_entry *field;
867 struct trace_seq *s = &iter->seq;
868
869 trace_assign_type(field, iter->ent);
870
871 if (!seq_print_userip_objs(field, s, flags))
872 goto partial;
873
874 if (!trace_seq_putc(s, '\n'))
875 goto partial;
876
877 return TRACE_TYPE_HANDLED;
878
879 partial:
880 return TRACE_TYPE_PARTIAL_LINE;
881}
882
883static struct trace_event trace_user_stack_event = {
884 .type = TRACE_USER_STACK,
885 .trace = trace_user_stack_print,
886 .raw = trace_special_print,
887 .hex = trace_special_hex,
888 .binary = trace_special_bin,
889};
890
891/* TRACE_BPRINT */
892static enum print_line_t
893trace_bprint_print(struct trace_iterator *iter, int flags)
894{
895 struct trace_entry *entry = iter->ent;
896 struct trace_seq *s = &iter->seq;
897 struct bprint_entry *field;
898
899 trace_assign_type(field, entry);
900
901 if (!seq_print_ip_sym(s, field->ip, flags))
902 goto partial;
903
904 if (!trace_seq_puts(s, ": "))
905 goto partial;
906
907 if (!trace_seq_bprintf(s, field->fmt, field->buf))
908 goto partial;
909
910 return TRACE_TYPE_HANDLED;
911
912 partial:
913 return TRACE_TYPE_PARTIAL_LINE;
914}
915
916
917static enum print_line_t
918trace_bprint_raw(struct trace_iterator *iter, int flags)
919{
920 struct bprint_entry *field;
921 struct trace_seq *s = &iter->seq;
922
923 trace_assign_type(field, iter->ent);
924
925 if (!trace_seq_printf(s, ": %lx : ", field->ip))
926 goto partial;
927
928 if (!trace_seq_bprintf(s, field->fmt, field->buf))
929 goto partial;
930
931 return TRACE_TYPE_HANDLED;
932
933 partial:
934 return TRACE_TYPE_PARTIAL_LINE;
935}
936
937
938static struct trace_event trace_bprint_event = {
939 .type = TRACE_BPRINT,
940 .trace = trace_bprint_print,
941 .raw = trace_bprint_raw,
942};
943
944/* TRACE_PRINT */
945static enum print_line_t trace_print_print(struct trace_iterator *iter,
946 int flags)
947{
948 struct print_entry *field;
949 struct trace_seq *s = &iter->seq;
950
951 trace_assign_type(field, iter->ent);
952
953 if (!seq_print_ip_sym(s, field->ip, flags))
954 goto partial;
955
956 if (!trace_seq_printf(s, ": %s", field->buf))
957 goto partial;
958
959 return TRACE_TYPE_HANDLED;
960
961 partial:
962 return TRACE_TYPE_PARTIAL_LINE;
963}
964
965static enum print_line_t trace_print_raw(struct trace_iterator *iter, int flags)
966{
967 struct print_entry *field;
968
969 trace_assign_type(field, iter->ent);
970
971 if (!trace_seq_printf(&iter->seq, "# %lx %s", field->ip, field->buf))
972 goto partial;
973
974 return TRACE_TYPE_HANDLED;
975
976 partial:
977 return TRACE_TYPE_PARTIAL_LINE;
978}
979
980static struct trace_event trace_print_event = {
981 .type = TRACE_PRINT,
982 .trace = trace_print_print,
983 .raw = trace_print_raw,
984};
985
986
987static struct trace_event *events[] __initdata = {
988 &trace_fn_event,
989 &trace_ctx_event,
990 &trace_wake_event,
991 &trace_special_event,
992 &trace_stack_event,
993 &trace_user_stack_event,
994 &trace_bprint_event,
995 &trace_print_event,
996 NULL
997};
998
999__init static int init_events(void)
1000{
1001 struct trace_event *event;
1002 int i, ret;
1003
1004 for (i = 0; events[i]; i++) {
1005 event = events[i];
1006
1007 ret = register_ftrace_event(event);
1008 if (!ret) {
1009 printk(KERN_WARNING "event %d failed to register\n",
1010 event->type);
1011 WARN_ON_ONCE(1);
1012 }
1013 }
1014
1015 return 0;
1016}
1017device_initcall(init_events);
diff --git a/kernel/trace/trace_output.h b/kernel/trace/trace_output.h
new file mode 100644
index 000000000000..e0bde39c2dd9
--- /dev/null
+++ b/kernel/trace/trace_output.h
@@ -0,0 +1,71 @@
1#ifndef __TRACE_EVENTS_H
2#define __TRACE_EVENTS_H
3
4#include "trace.h"
5
6typedef enum print_line_t (*trace_print_func)(struct trace_iterator *iter,
7 int flags);
8
9struct trace_event {
10 struct hlist_node node;
11 int type;
12 trace_print_func trace;
13 trace_print_func raw;
14 trace_print_func hex;
15 trace_print_func binary;
16};
17
18extern enum print_line_t
19trace_print_bprintk_msg_only(struct trace_iterator *iter);
20extern enum print_line_t
21trace_print_printk_msg_only(struct trace_iterator *iter);
22
23extern int trace_seq_printf(struct trace_seq *s, const char *fmt, ...)
24 __attribute__ ((format (printf, 2, 3)));
25extern int
26trace_seq_bprintf(struct trace_seq *s, const char *fmt, const u32 *binary);
27extern int
28seq_print_ip_sym(struct trace_seq *s, unsigned long ip,
29 unsigned long sym_flags);
30extern ssize_t trace_seq_to_user(struct trace_seq *s, char __user *ubuf,
31 size_t cnt);
32extern int trace_seq_puts(struct trace_seq *s, const char *str);
33extern int trace_seq_putc(struct trace_seq *s, unsigned char c);
34extern int trace_seq_putmem(struct trace_seq *s, const void *mem, size_t len);
35extern int trace_seq_putmem_hex(struct trace_seq *s, const void *mem,
36 size_t len);
37extern void *trace_seq_reserve(struct trace_seq *s, size_t len);
38extern int trace_seq_path(struct trace_seq *s, struct path *path);
39extern int seq_print_userip_objs(const struct userstack_entry *entry,
40 struct trace_seq *s, unsigned long sym_flags);
41extern int seq_print_user_ip(struct trace_seq *s, struct mm_struct *mm,
42 unsigned long ip, unsigned long sym_flags);
43
44extern int trace_print_context(struct trace_iterator *iter);
45extern int trace_print_lat_context(struct trace_iterator *iter);
46
47extern struct trace_event *ftrace_find_event(int type);
48extern int register_ftrace_event(struct trace_event *event);
49extern int unregister_ftrace_event(struct trace_event *event);
50
51extern enum print_line_t trace_nop_print(struct trace_iterator *iter,
52 int flags);
53
54#define MAX_MEMHEX_BYTES 8
55#define HEX_CHARS (MAX_MEMHEX_BYTES*2 + 1)
56
57#define SEQ_PUT_FIELD_RET(s, x) \
58do { \
59 if (!trace_seq_putmem(s, &(x), sizeof(x))) \
60 return TRACE_TYPE_PARTIAL_LINE; \
61} while (0)
62
63#define SEQ_PUT_HEX_FIELD_RET(s, x) \
64do { \
65 BUILD_BUG_ON(sizeof(x) > MAX_MEMHEX_BYTES); \
66 if (!trace_seq_putmem_hex(s, &(x), sizeof(x))) \
67 return TRACE_TYPE_PARTIAL_LINE; \
68} while (0)
69
70#endif
71
diff --git a/kernel/trace/trace_power.c b/kernel/trace/trace_power.c
index 7bda248daf55..bae791ebcc51 100644
--- a/kernel/trace/trace_power.c
+++ b/kernel/trace/trace_power.c
@@ -11,15 +11,113 @@
11 11
12#include <linux/init.h> 12#include <linux/init.h>
13#include <linux/debugfs.h> 13#include <linux/debugfs.h>
14#include <linux/ftrace.h> 14#include <trace/power.h>
15#include <linux/kallsyms.h> 15#include <linux/kallsyms.h>
16#include <linux/module.h> 16#include <linux/module.h>
17 17
18#include "trace.h" 18#include "trace.h"
19#include "trace_output.h"
19 20
20static struct trace_array *power_trace; 21static struct trace_array *power_trace;
21static int __read_mostly trace_power_enabled; 22static int __read_mostly trace_power_enabled;
22 23
24static void probe_power_start(struct power_trace *it, unsigned int type,
25 unsigned int level)
26{
27 if (!trace_power_enabled)
28 return;
29
30 memset(it, 0, sizeof(struct power_trace));
31 it->state = level;
32 it->type = type;
33 it->stamp = ktime_get();
34}
35
36
37static void probe_power_end(struct power_trace *it)
38{
39 struct ring_buffer_event *event;
40 struct trace_power *entry;
41 struct trace_array_cpu *data;
42 struct trace_array *tr = power_trace;
43
44 if (!trace_power_enabled)
45 return;
46
47 preempt_disable();
48 it->end = ktime_get();
49 data = tr->data[smp_processor_id()];
50
51 event = trace_buffer_lock_reserve(tr, TRACE_POWER,
52 sizeof(*entry), 0, 0);
53 if (!event)
54 goto out;
55 entry = ring_buffer_event_data(event);
56 entry->state_data = *it;
57 trace_buffer_unlock_commit(tr, event, 0, 0);
58 out:
59 preempt_enable();
60}
61
62static void probe_power_mark(struct power_trace *it, unsigned int type,
63 unsigned int level)
64{
65 struct ring_buffer_event *event;
66 struct trace_power *entry;
67 struct trace_array_cpu *data;
68 struct trace_array *tr = power_trace;
69
70 if (!trace_power_enabled)
71 return;
72
73 memset(it, 0, sizeof(struct power_trace));
74 it->state = level;
75 it->type = type;
76 it->stamp = ktime_get();
77 preempt_disable();
78 it->end = it->stamp;
79 data = tr->data[smp_processor_id()];
80
81 event = trace_buffer_lock_reserve(tr, TRACE_POWER,
82 sizeof(*entry), 0, 0);
83 if (!event)
84 goto out;
85 entry = ring_buffer_event_data(event);
86 entry->state_data = *it;
87 trace_buffer_unlock_commit(tr, event, 0, 0);
88 out:
89 preempt_enable();
90}
91
92static int tracing_power_register(void)
93{
94 int ret;
95
96 ret = register_trace_power_start(probe_power_start);
97 if (ret) {
98 pr_info("power trace: Couldn't activate tracepoint"
99 " probe to trace_power_start\n");
100 return ret;
101 }
102 ret = register_trace_power_end(probe_power_end);
103 if (ret) {
104 pr_info("power trace: Couldn't activate tracepoint"
105 " probe to trace_power_end\n");
106 goto fail_start;
107 }
108 ret = register_trace_power_mark(probe_power_mark);
109 if (ret) {
110 pr_info("power trace: Couldn't activate tracepoint"
111 " probe to trace_power_mark\n");
112 goto fail_end;
113 }
114 return ret;
115fail_end:
116 unregister_trace_power_end(probe_power_end);
117fail_start:
118 unregister_trace_power_start(probe_power_start);
119 return ret;
120}
23 121
24static void start_power_trace(struct trace_array *tr) 122static void start_power_trace(struct trace_array *tr)
25{ 123{
@@ -31,6 +129,14 @@ static void stop_power_trace(struct trace_array *tr)
31 trace_power_enabled = 0; 129 trace_power_enabled = 0;
32} 130}
33 131
132static void power_trace_reset(struct trace_array *tr)
133{
134 trace_power_enabled = 0;
135 unregister_trace_power_start(probe_power_start);
136 unregister_trace_power_end(probe_power_end);
137 unregister_trace_power_mark(probe_power_mark);
138}
139
34 140
35static int power_trace_init(struct trace_array *tr) 141static int power_trace_init(struct trace_array *tr)
36{ 142{
@@ -38,6 +144,7 @@ static int power_trace_init(struct trace_array *tr)
38 power_trace = tr; 144 power_trace = tr;
39 145
40 trace_power_enabled = 1; 146 trace_power_enabled = 1;
147 tracing_power_register();
41 148
42 for_each_cpu(cpu, cpu_possible_mask) 149 for_each_cpu(cpu, cpu_possible_mask)
43 tracing_reset(tr, cpu); 150 tracing_reset(tr, cpu);
@@ -85,7 +192,7 @@ static struct tracer power_tracer __read_mostly =
85 .init = power_trace_init, 192 .init = power_trace_init,
86 .start = start_power_trace, 193 .start = start_power_trace,
87 .stop = stop_power_trace, 194 .stop = stop_power_trace,
88 .reset = stop_power_trace, 195 .reset = power_trace_reset,
89 .print_line = power_print_line, 196 .print_line = power_print_line,
90}; 197};
91 198
@@ -94,86 +201,3 @@ static int init_power_trace(void)
94 return register_tracer(&power_tracer); 201 return register_tracer(&power_tracer);
95} 202}
96device_initcall(init_power_trace); 203device_initcall(init_power_trace);
97
98void trace_power_start(struct power_trace *it, unsigned int type,
99 unsigned int level)
100{
101 if (!trace_power_enabled)
102 return;
103
104 memset(it, 0, sizeof(struct power_trace));
105 it->state = level;
106 it->type = type;
107 it->stamp = ktime_get();
108}
109EXPORT_SYMBOL_GPL(trace_power_start);
110
111
112void trace_power_end(struct power_trace *it)
113{
114 struct ring_buffer_event *event;
115 struct trace_power *entry;
116 struct trace_array_cpu *data;
117 unsigned long irq_flags;
118 struct trace_array *tr = power_trace;
119
120 if (!trace_power_enabled)
121 return;
122
123 preempt_disable();
124 it->end = ktime_get();
125 data = tr->data[smp_processor_id()];
126
127 event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
128 &irq_flags);
129 if (!event)
130 goto out;
131 entry = ring_buffer_event_data(event);
132 tracing_generic_entry_update(&entry->ent, 0, 0);
133 entry->ent.type = TRACE_POWER;
134 entry->state_data = *it;
135 ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
136
137 trace_wake_up();
138
139 out:
140 preempt_enable();
141}
142EXPORT_SYMBOL_GPL(trace_power_end);
143
144void trace_power_mark(struct power_trace *it, unsigned int type,
145 unsigned int level)
146{
147 struct ring_buffer_event *event;
148 struct trace_power *entry;
149 struct trace_array_cpu *data;
150 unsigned long irq_flags;
151 struct trace_array *tr = power_trace;
152
153 if (!trace_power_enabled)
154 return;
155
156 memset(it, 0, sizeof(struct power_trace));
157 it->state = level;
158 it->type = type;
159 it->stamp = ktime_get();
160 preempt_disable();
161 it->end = it->stamp;
162 data = tr->data[smp_processor_id()];
163
164 event = ring_buffer_lock_reserve(tr->buffer, sizeof(*entry),
165 &irq_flags);
166 if (!event)
167 goto out;
168 entry = ring_buffer_event_data(event);
169 tracing_generic_entry_update(&entry->ent, 0, 0);
170 entry->ent.type = TRACE_POWER;
171 entry->state_data = *it;
172 ring_buffer_unlock_commit(tr->buffer, event, irq_flags);
173
174 trace_wake_up();
175
176 out:
177 preempt_enable();
178}
179EXPORT_SYMBOL_GPL(trace_power_mark);
diff --git a/kernel/trace/trace_printk.c b/kernel/trace/trace_printk.c
new file mode 100644
index 000000000000..eb81556107fe
--- /dev/null
+++ b/kernel/trace/trace_printk.c
@@ -0,0 +1,270 @@
1/*
2 * trace binary printk
3 *
4 * Copyright (C) 2008 Lai Jiangshan <laijs@cn.fujitsu.com>
5 *
6 */
7#include <linux/seq_file.h>
8#include <linux/debugfs.h>
9#include <linux/uaccess.h>
10#include <linux/kernel.h>
11#include <linux/ftrace.h>
12#include <linux/string.h>
13#include <linux/module.h>
14#include <linux/marker.h>
15#include <linux/mutex.h>
16#include <linux/ctype.h>
17#include <linux/list.h>
18#include <linux/slab.h>
19#include <linux/fs.h>
20
21#include "trace.h"
22
23#ifdef CONFIG_MODULES
24
25/*
26 * modules trace_printk()'s formats are autosaved in struct trace_bprintk_fmt
27 * which are queued on trace_bprintk_fmt_list.
28 */
29static LIST_HEAD(trace_bprintk_fmt_list);
30
31/* serialize accesses to trace_bprintk_fmt_list */
32static DEFINE_MUTEX(btrace_mutex);
33
34struct trace_bprintk_fmt {
35 struct list_head list;
36 char fmt[0];
37};
38
39static inline struct trace_bprintk_fmt *lookup_format(const char *fmt)
40{
41 struct trace_bprintk_fmt *pos;
42 list_for_each_entry(pos, &trace_bprintk_fmt_list, list) {
43 if (!strcmp(pos->fmt, fmt))
44 return pos;
45 }
46 return NULL;
47}
48
49static
50void hold_module_trace_bprintk_format(const char **start, const char **end)
51{
52 const char **iter;
53
54 mutex_lock(&btrace_mutex);
55 for (iter = start; iter < end; iter++) {
56 struct trace_bprintk_fmt *tb_fmt = lookup_format(*iter);
57 if (tb_fmt) {
58 *iter = tb_fmt->fmt;
59 continue;
60 }
61
62 tb_fmt = kmalloc(offsetof(struct trace_bprintk_fmt, fmt)
63 + strlen(*iter) + 1, GFP_KERNEL);
64 if (tb_fmt) {
65 list_add_tail(&tb_fmt->list, &trace_bprintk_fmt_list);
66 strcpy(tb_fmt->fmt, *iter);
67 *iter = tb_fmt->fmt;
68 } else
69 *iter = NULL;
70 }
71 mutex_unlock(&btrace_mutex);
72}
73
74static int module_trace_bprintk_format_notify(struct notifier_block *self,
75 unsigned long val, void *data)
76{
77 struct module *mod = data;
78 if (mod->num_trace_bprintk_fmt) {
79 const char **start = mod->trace_bprintk_fmt_start;
80 const char **end = start + mod->num_trace_bprintk_fmt;
81
82 if (val == MODULE_STATE_COMING)
83 hold_module_trace_bprintk_format(start, end);
84 }
85 return 0;
86}
87
88#else /* !CONFIG_MODULES */
89__init static int
90module_trace_bprintk_format_notify(struct notifier_block *self,
91 unsigned long val, void *data)
92{
93 return 0;
94}
95#endif /* CONFIG_MODULES */
96
97
98__initdata_or_module static
99struct notifier_block module_trace_bprintk_format_nb = {
100 .notifier_call = module_trace_bprintk_format_notify,
101};
102
103int __trace_bprintk(unsigned long ip, const char *fmt, ...)
104 {
105 int ret;
106 va_list ap;
107
108 if (unlikely(!fmt))
109 return 0;
110
111 if (!(trace_flags & TRACE_ITER_PRINTK))
112 return 0;
113
114 va_start(ap, fmt);
115 ret = trace_vbprintk(ip, fmt, ap);
116 va_end(ap);
117 return ret;
118}
119EXPORT_SYMBOL_GPL(__trace_bprintk);
120
121int __ftrace_vbprintk(unsigned long ip, const char *fmt, va_list ap)
122 {
123 if (unlikely(!fmt))
124 return 0;
125
126 if (!(trace_flags & TRACE_ITER_PRINTK))
127 return 0;
128
129 return trace_vbprintk(ip, fmt, ap);
130}
131EXPORT_SYMBOL_GPL(__ftrace_vbprintk);
132
133int __trace_printk(unsigned long ip, const char *fmt, ...)
134{
135 int ret;
136 va_list ap;
137
138 if (!(trace_flags & TRACE_ITER_PRINTK))
139 return 0;
140
141 va_start(ap, fmt);
142 ret = trace_vprintk(ip, fmt, ap);
143 va_end(ap);
144 return ret;
145}
146EXPORT_SYMBOL_GPL(__trace_printk);
147
148int __ftrace_vprintk(unsigned long ip, const char *fmt, va_list ap)
149{
150 if (!(trace_flags & TRACE_ITER_PRINTK))
151 return 0;
152
153 return trace_vprintk(ip, fmt, ap);
154}
155EXPORT_SYMBOL_GPL(__ftrace_vprintk);
156
157static void *
158t_next(struct seq_file *m, void *v, loff_t *pos)
159{
160 const char **fmt = m->private;
161 const char **next = fmt;
162
163 (*pos)++;
164
165 if ((unsigned long)fmt >= (unsigned long)__stop___trace_bprintk_fmt)
166 return NULL;
167
168 next = fmt;
169 m->private = ++next;
170
171 return fmt;
172}
173
174static void *t_start(struct seq_file *m, loff_t *pos)
175{
176 return t_next(m, NULL, pos);
177}
178
179static int t_show(struct seq_file *m, void *v)
180{
181 const char **fmt = v;
182 const char *str = *fmt;
183 int i;
184
185 seq_printf(m, "0x%lx : \"", (unsigned long)fmt);
186
187 /*
188 * Tabs and new lines need to be converted.
189 */
190 for (i = 0; str[i]; i++) {
191 switch (str[i]) {
192 case '\n':
193 seq_puts(m, "\\n");
194 break;
195 case '\t':
196 seq_puts(m, "\\t");
197 break;
198 case '\\':
199 seq_puts(m, "\\");
200 break;
201 case '"':
202 seq_puts(m, "\\\"");
203 break;
204 default:
205 seq_putc(m, str[i]);
206 }
207 }
208 seq_puts(m, "\"\n");
209
210 return 0;
211}
212
213static void t_stop(struct seq_file *m, void *p)
214{
215}
216
217static const struct seq_operations show_format_seq_ops = {
218 .start = t_start,
219 .next = t_next,
220 .show = t_show,
221 .stop = t_stop,
222};
223
224static int
225ftrace_formats_open(struct inode *inode, struct file *file)
226{
227 int ret;
228
229 ret = seq_open(file, &show_format_seq_ops);
230 if (!ret) {
231 struct seq_file *m = file->private_data;
232
233 m->private = __start___trace_bprintk_fmt;
234 }
235 return ret;
236}
237
238static const struct file_operations ftrace_formats_fops = {
239 .open = ftrace_formats_open,
240 .read = seq_read,
241 .llseek = seq_lseek,
242 .release = seq_release,
243};
244
245static __init int init_trace_printk_function_export(void)
246{
247 struct dentry *d_tracer;
248 struct dentry *entry;
249
250 d_tracer = tracing_init_dentry();
251 if (!d_tracer)
252 return 0;
253
254 entry = debugfs_create_file("printk_formats", 0444, d_tracer,
255 NULL, &ftrace_formats_fops);
256 if (!entry)
257 pr_warning("Could not create debugfs "
258 "'printk_formats' entry\n");
259
260 return 0;
261}
262
263fs_initcall(init_trace_printk_function_export);
264
265static __init int init_trace_printk(void)
266{
267 return register_module_notifier(&module_trace_bprintk_format_nb);
268}
269
270early_initcall(init_trace_printk);
diff --git a/kernel/trace/trace_sched_switch.c b/kernel/trace/trace_sched_switch.c
index df175cb4564f..de35f200abd3 100644
--- a/kernel/trace/trace_sched_switch.c
+++ b/kernel/trace/trace_sched_switch.c
@@ -18,6 +18,7 @@ static struct trace_array *ctx_trace;
18static int __read_mostly tracer_enabled; 18static int __read_mostly tracer_enabled;
19static int sched_ref; 19static int sched_ref;
20static DEFINE_MUTEX(sched_register_mutex); 20static DEFINE_MUTEX(sched_register_mutex);
21static int sched_stopped;
21 22
22static void 23static void
23probe_sched_switch(struct rq *__rq, struct task_struct *prev, 24probe_sched_switch(struct rq *__rq, struct task_struct *prev,
@@ -28,7 +29,7 @@ probe_sched_switch(struct rq *__rq, struct task_struct *prev,
28 int cpu; 29 int cpu;
29 int pc; 30 int pc;
30 31
31 if (!sched_ref) 32 if (!sched_ref || sched_stopped)
32 return; 33 return;
33 34
34 tracing_record_cmdline(prev); 35 tracing_record_cmdline(prev);
@@ -43,7 +44,7 @@ probe_sched_switch(struct rq *__rq, struct task_struct *prev,
43 data = ctx_trace->data[cpu]; 44 data = ctx_trace->data[cpu];
44 45
45 if (likely(!atomic_read(&data->disabled))) 46 if (likely(!atomic_read(&data->disabled)))
46 tracing_sched_switch_trace(ctx_trace, data, prev, next, flags, pc); 47 tracing_sched_switch_trace(ctx_trace, prev, next, flags, pc);
47 48
48 local_irq_restore(flags); 49 local_irq_restore(flags);
49} 50}
@@ -66,7 +67,7 @@ probe_sched_wakeup(struct rq *__rq, struct task_struct *wakee, int success)
66 data = ctx_trace->data[cpu]; 67 data = ctx_trace->data[cpu];
67 68
68 if (likely(!atomic_read(&data->disabled))) 69 if (likely(!atomic_read(&data->disabled)))
69 tracing_sched_wakeup_trace(ctx_trace, data, wakee, current, 70 tracing_sched_wakeup_trace(ctx_trace, wakee, current,
70 flags, pc); 71 flags, pc);
71 72
72 local_irq_restore(flags); 73 local_irq_restore(flags);
@@ -93,7 +94,7 @@ static int tracing_sched_register(void)
93 ret = register_trace_sched_switch(probe_sched_switch); 94 ret = register_trace_sched_switch(probe_sched_switch);
94 if (ret) { 95 if (ret) {
95 pr_info("sched trace: Couldn't activate tracepoint" 96 pr_info("sched trace: Couldn't activate tracepoint"
96 " probe to kernel_sched_schedule\n"); 97 " probe to kernel_sched_switch\n");
97 goto fail_deprobe_wake_new; 98 goto fail_deprobe_wake_new;
98 } 99 }
99 100
@@ -185,12 +186,6 @@ void tracing_sched_switch_assign_trace(struct trace_array *tr)
185 ctx_trace = tr; 186 ctx_trace = tr;
186} 187}
187 188
188static void start_sched_trace(struct trace_array *tr)
189{
190 tracing_reset_online_cpus(tr);
191 tracing_start_sched_switch_record();
192}
193
194static void stop_sched_trace(struct trace_array *tr) 189static void stop_sched_trace(struct trace_array *tr)
195{ 190{
196 tracing_stop_sched_switch_record(); 191 tracing_stop_sched_switch_record();
@@ -199,7 +194,8 @@ static void stop_sched_trace(struct trace_array *tr)
199static int sched_switch_trace_init(struct trace_array *tr) 194static int sched_switch_trace_init(struct trace_array *tr)
200{ 195{
201 ctx_trace = tr; 196 ctx_trace = tr;
202 start_sched_trace(tr); 197 tracing_reset_online_cpus(tr);
198 tracing_start_sched_switch_record();
203 return 0; 199 return 0;
204} 200}
205 201
@@ -211,13 +207,12 @@ static void sched_switch_trace_reset(struct trace_array *tr)
211 207
212static void sched_switch_trace_start(struct trace_array *tr) 208static void sched_switch_trace_start(struct trace_array *tr)
213{ 209{
214 tracing_reset_online_cpus(tr); 210 sched_stopped = 0;
215 tracing_start_sched_switch();
216} 211}
217 212
218static void sched_switch_trace_stop(struct trace_array *tr) 213static void sched_switch_trace_stop(struct trace_array *tr)
219{ 214{
220 tracing_stop_sched_switch(); 215 sched_stopped = 1;
221} 216}
222 217
223static struct tracer sched_switch_trace __read_mostly = 218static struct tracer sched_switch_trace __read_mostly =
@@ -227,6 +222,7 @@ static struct tracer sched_switch_trace __read_mostly =
227 .reset = sched_switch_trace_reset, 222 .reset = sched_switch_trace_reset,
228 .start = sched_switch_trace_start, 223 .start = sched_switch_trace_start,
229 .stop = sched_switch_trace_stop, 224 .stop = sched_switch_trace_stop,
225 .wait_pipe = poll_wait_pipe,
230#ifdef CONFIG_FTRACE_SELFTEST 226#ifdef CONFIG_FTRACE_SELFTEST
231 .selftest = trace_selftest_startup_sched_switch, 227 .selftest = trace_selftest_startup_sched_switch,
232#endif 228#endif
diff --git a/kernel/trace/trace_sched_wakeup.c b/kernel/trace/trace_sched_wakeup.c
index 42ae1e77b6b3..3c5ad6b2ec84 100644
--- a/kernel/trace/trace_sched_wakeup.c
+++ b/kernel/trace/trace_sched_wakeup.c
@@ -25,12 +25,15 @@ static int __read_mostly tracer_enabled;
25static struct task_struct *wakeup_task; 25static struct task_struct *wakeup_task;
26static int wakeup_cpu; 26static int wakeup_cpu;
27static unsigned wakeup_prio = -1; 27static unsigned wakeup_prio = -1;
28static int wakeup_rt;
28 29
29static raw_spinlock_t wakeup_lock = 30static raw_spinlock_t wakeup_lock =
30 (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED; 31 (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
31 32
32static void __wakeup_reset(struct trace_array *tr); 33static void __wakeup_reset(struct trace_array *tr);
33 34
35static int save_lat_flag;
36
34#ifdef CONFIG_FUNCTION_TRACER 37#ifdef CONFIG_FUNCTION_TRACER
35/* 38/*
36 * irqsoff uses its own tracer function to keep the overhead down: 39 * irqsoff uses its own tracer function to keep the overhead down:
@@ -71,7 +74,7 @@ wakeup_tracer_call(unsigned long ip, unsigned long parent_ip)
71 if (task_cpu(wakeup_task) != cpu) 74 if (task_cpu(wakeup_task) != cpu)
72 goto unlock; 75 goto unlock;
73 76
74 trace_function(tr, data, ip, parent_ip, flags, pc); 77 trace_function(tr, ip, parent_ip, flags, pc);
75 78
76 unlock: 79 unlock:
77 __raw_spin_unlock(&wakeup_lock); 80 __raw_spin_unlock(&wakeup_lock);
@@ -151,7 +154,8 @@ probe_wakeup_sched_switch(struct rq *rq, struct task_struct *prev,
151 if (unlikely(!tracer_enabled || next != wakeup_task)) 154 if (unlikely(!tracer_enabled || next != wakeup_task))
152 goto out_unlock; 155 goto out_unlock;
153 156
154 trace_function(wakeup_trace, data, CALLER_ADDR1, CALLER_ADDR2, flags, pc); 157 trace_function(wakeup_trace, CALLER_ADDR1, CALLER_ADDR2, flags, pc);
158 tracing_sched_switch_trace(wakeup_trace, prev, next, flags, pc);
155 159
156 /* 160 /*
157 * usecs conversion is slow so we try to delay the conversion 161 * usecs conversion is slow so we try to delay the conversion
@@ -182,13 +186,10 @@ out:
182 186
183static void __wakeup_reset(struct trace_array *tr) 187static void __wakeup_reset(struct trace_array *tr)
184{ 188{
185 struct trace_array_cpu *data;
186 int cpu; 189 int cpu;
187 190
188 for_each_possible_cpu(cpu) { 191 for_each_possible_cpu(cpu)
189 data = tr->data[cpu];
190 tracing_reset(tr, cpu); 192 tracing_reset(tr, cpu);
191 }
192 193
193 wakeup_cpu = -1; 194 wakeup_cpu = -1;
194 wakeup_prio = -1; 195 wakeup_prio = -1;
@@ -213,6 +214,7 @@ static void wakeup_reset(struct trace_array *tr)
213static void 214static void
214probe_wakeup(struct rq *rq, struct task_struct *p, int success) 215probe_wakeup(struct rq *rq, struct task_struct *p, int success)
215{ 216{
217 struct trace_array_cpu *data;
216 int cpu = smp_processor_id(); 218 int cpu = smp_processor_id();
217 unsigned long flags; 219 unsigned long flags;
218 long disabled; 220 long disabled;
@@ -224,7 +226,7 @@ probe_wakeup(struct rq *rq, struct task_struct *p, int success)
224 tracing_record_cmdline(p); 226 tracing_record_cmdline(p);
225 tracing_record_cmdline(current); 227 tracing_record_cmdline(current);
226 228
227 if (likely(!rt_task(p)) || 229 if ((wakeup_rt && !rt_task(p)) ||
228 p->prio >= wakeup_prio || 230 p->prio >= wakeup_prio ||
229 p->prio >= current->prio) 231 p->prio >= current->prio)
230 return; 232 return;
@@ -252,9 +254,10 @@ probe_wakeup(struct rq *rq, struct task_struct *p, int success)
252 254
253 local_save_flags(flags); 255 local_save_flags(flags);
254 256
255 wakeup_trace->data[wakeup_cpu]->preempt_timestamp = ftrace_now(cpu); 257 data = wakeup_trace->data[wakeup_cpu];
256 trace_function(wakeup_trace, wakeup_trace->data[wakeup_cpu], 258 data->preempt_timestamp = ftrace_now(cpu);
257 CALLER_ADDR1, CALLER_ADDR2, flags, pc); 259 tracing_sched_wakeup_trace(wakeup_trace, p, current, flags, pc);
260 trace_function(wakeup_trace, CALLER_ADDR1, CALLER_ADDR2, flags, pc);
258 261
259out_locked: 262out_locked:
260 __raw_spin_unlock(&wakeup_lock); 263 __raw_spin_unlock(&wakeup_lock);
@@ -262,12 +265,6 @@ out:
262 atomic_dec(&wakeup_trace->data[cpu]->disabled); 265 atomic_dec(&wakeup_trace->data[cpu]->disabled);
263} 266}
264 267
265/*
266 * save_tracer_enabled is used to save the state of the tracer_enabled
267 * variable when we disable it when we open a trace output file.
268 */
269static int save_tracer_enabled;
270
271static void start_wakeup_tracer(struct trace_array *tr) 268static void start_wakeup_tracer(struct trace_array *tr)
272{ 269{
273 int ret; 270 int ret;
@@ -289,7 +286,7 @@ static void start_wakeup_tracer(struct trace_array *tr)
289 ret = register_trace_sched_switch(probe_wakeup_sched_switch); 286 ret = register_trace_sched_switch(probe_wakeup_sched_switch);
290 if (ret) { 287 if (ret) {
291 pr_info("sched trace: Couldn't activate tracepoint" 288 pr_info("sched trace: Couldn't activate tracepoint"
292 " probe to kernel_sched_schedule\n"); 289 " probe to kernel_sched_switch\n");
293 goto fail_deprobe_wake_new; 290 goto fail_deprobe_wake_new;
294 } 291 }
295 292
@@ -306,13 +303,10 @@ static void start_wakeup_tracer(struct trace_array *tr)
306 303
307 register_ftrace_function(&trace_ops); 304 register_ftrace_function(&trace_ops);
308 305
309 if (tracing_is_enabled()) { 306 if (tracing_is_enabled())
310 tracer_enabled = 1; 307 tracer_enabled = 1;
311 save_tracer_enabled = 1; 308 else
312 } else {
313 tracer_enabled = 0; 309 tracer_enabled = 0;
314 save_tracer_enabled = 0;
315 }
316 310
317 return; 311 return;
318fail_deprobe_wake_new: 312fail_deprobe_wake_new:
@@ -324,54 +318,54 @@ fail_deprobe:
324static void stop_wakeup_tracer(struct trace_array *tr) 318static void stop_wakeup_tracer(struct trace_array *tr)
325{ 319{
326 tracer_enabled = 0; 320 tracer_enabled = 0;
327 save_tracer_enabled = 0;
328 unregister_ftrace_function(&trace_ops); 321 unregister_ftrace_function(&trace_ops);
329 unregister_trace_sched_switch(probe_wakeup_sched_switch); 322 unregister_trace_sched_switch(probe_wakeup_sched_switch);
330 unregister_trace_sched_wakeup_new(probe_wakeup); 323 unregister_trace_sched_wakeup_new(probe_wakeup);
331 unregister_trace_sched_wakeup(probe_wakeup); 324 unregister_trace_sched_wakeup(probe_wakeup);
332} 325}
333 326
334static int wakeup_tracer_init(struct trace_array *tr) 327static int __wakeup_tracer_init(struct trace_array *tr)
335{ 328{
329 save_lat_flag = trace_flags & TRACE_ITER_LATENCY_FMT;
330 trace_flags |= TRACE_ITER_LATENCY_FMT;
331
336 tracing_max_latency = 0; 332 tracing_max_latency = 0;
337 wakeup_trace = tr; 333 wakeup_trace = tr;
338 start_wakeup_tracer(tr); 334 start_wakeup_tracer(tr);
339 return 0; 335 return 0;
340} 336}
341 337
338static int wakeup_tracer_init(struct trace_array *tr)
339{
340 wakeup_rt = 0;
341 return __wakeup_tracer_init(tr);
342}
343
344static int wakeup_rt_tracer_init(struct trace_array *tr)
345{
346 wakeup_rt = 1;
347 return __wakeup_tracer_init(tr);
348}
349
342static void wakeup_tracer_reset(struct trace_array *tr) 350static void wakeup_tracer_reset(struct trace_array *tr)
343{ 351{
344 stop_wakeup_tracer(tr); 352 stop_wakeup_tracer(tr);
345 /* make sure we put back any tasks we are tracing */ 353 /* make sure we put back any tasks we are tracing */
346 wakeup_reset(tr); 354 wakeup_reset(tr);
355
356 if (!save_lat_flag)
357 trace_flags &= ~TRACE_ITER_LATENCY_FMT;
347} 358}
348 359
349static void wakeup_tracer_start(struct trace_array *tr) 360static void wakeup_tracer_start(struct trace_array *tr)
350{ 361{
351 wakeup_reset(tr); 362 wakeup_reset(tr);
352 tracer_enabled = 1; 363 tracer_enabled = 1;
353 save_tracer_enabled = 1;
354} 364}
355 365
356static void wakeup_tracer_stop(struct trace_array *tr) 366static void wakeup_tracer_stop(struct trace_array *tr)
357{ 367{
358 tracer_enabled = 0; 368 tracer_enabled = 0;
359 save_tracer_enabled = 0;
360}
361
362static void wakeup_tracer_open(struct trace_iterator *iter)
363{
364 /* stop the trace while dumping */
365 tracer_enabled = 0;
366}
367
368static void wakeup_tracer_close(struct trace_iterator *iter)
369{
370 /* forget about any processes we were recording */
371 if (save_tracer_enabled) {
372 wakeup_reset(iter->tr);
373 tracer_enabled = 1;
374 }
375} 369}
376 370
377static struct tracer wakeup_tracer __read_mostly = 371static struct tracer wakeup_tracer __read_mostly =
@@ -381,8 +375,20 @@ static struct tracer wakeup_tracer __read_mostly =
381 .reset = wakeup_tracer_reset, 375 .reset = wakeup_tracer_reset,
382 .start = wakeup_tracer_start, 376 .start = wakeup_tracer_start,
383 .stop = wakeup_tracer_stop, 377 .stop = wakeup_tracer_stop,
384 .open = wakeup_tracer_open, 378 .print_max = 1,
385 .close = wakeup_tracer_close, 379#ifdef CONFIG_FTRACE_SELFTEST
380 .selftest = trace_selftest_startup_wakeup,
381#endif
382};
383
384static struct tracer wakeup_rt_tracer __read_mostly =
385{
386 .name = "wakeup_rt",
387 .init = wakeup_rt_tracer_init,
388 .reset = wakeup_tracer_reset,
389 .start = wakeup_tracer_start,
390 .stop = wakeup_tracer_stop,
391 .wait_pipe = poll_wait_pipe,
386 .print_max = 1, 392 .print_max = 1,
387#ifdef CONFIG_FTRACE_SELFTEST 393#ifdef CONFIG_FTRACE_SELFTEST
388 .selftest = trace_selftest_startup_wakeup, 394 .selftest = trace_selftest_startup_wakeup,
@@ -397,6 +403,10 @@ __init static int init_wakeup_tracer(void)
397 if (ret) 403 if (ret)
398 return ret; 404 return ret;
399 405
406 ret = register_tracer(&wakeup_rt_tracer);
407 if (ret)
408 return ret;
409
400 return 0; 410 return 0;
401} 411}
402device_initcall(init_wakeup_tracer); 412device_initcall(init_wakeup_tracer);
diff --git a/kernel/trace/trace_selftest.c b/kernel/trace/trace_selftest.c
index bc8e80a86bca..08f4eb2763d1 100644
--- a/kernel/trace/trace_selftest.c
+++ b/kernel/trace/trace_selftest.c
@@ -1,5 +1,6 @@
1/* Include in trace.c */ 1/* Include in trace.c */
2 2
3#include <linux/stringify.h>
3#include <linux/kthread.h> 4#include <linux/kthread.h>
4#include <linux/delay.h> 5#include <linux/delay.h>
5 6
@@ -9,11 +10,12 @@ static inline int trace_valid_entry(struct trace_entry *entry)
9 case TRACE_FN: 10 case TRACE_FN:
10 case TRACE_CTX: 11 case TRACE_CTX:
11 case TRACE_WAKE: 12 case TRACE_WAKE:
12 case TRACE_CONT:
13 case TRACE_STACK: 13 case TRACE_STACK:
14 case TRACE_PRINT: 14 case TRACE_PRINT:
15 case TRACE_SPECIAL: 15 case TRACE_SPECIAL:
16 case TRACE_BRANCH: 16 case TRACE_BRANCH:
17 case TRACE_GRAPH_ENT:
18 case TRACE_GRAPH_RET:
17 return 1; 19 return 1;
18 } 20 }
19 return 0; 21 return 0;
@@ -99,9 +101,6 @@ static inline void warn_failed_init_tracer(struct tracer *trace, int init_ret)
99 101
100#ifdef CONFIG_DYNAMIC_FTRACE 102#ifdef CONFIG_DYNAMIC_FTRACE
101 103
102#define __STR(x) #x
103#define STR(x) __STR(x)
104
105/* Test dynamic code modification and ftrace filters */ 104/* Test dynamic code modification and ftrace filters */
106int trace_selftest_startup_dynamic_tracing(struct tracer *trace, 105int trace_selftest_startup_dynamic_tracing(struct tracer *trace,
107 struct trace_array *tr, 106 struct trace_array *tr,
@@ -125,17 +124,17 @@ int trace_selftest_startup_dynamic_tracing(struct tracer *trace,
125 func(); 124 func();
126 125
127 /* 126 /*
128 * Some archs *cough*PowerPC*cough* add charachters to the 127 * Some archs *cough*PowerPC*cough* add characters to the
129 * start of the function names. We simply put a '*' to 128 * start of the function names. We simply put a '*' to
130 * accomodate them. 129 * accommodate them.
131 */ 130 */
132 func_name = "*" STR(DYN_FTRACE_TEST_NAME); 131 func_name = "*" __stringify(DYN_FTRACE_TEST_NAME);
133 132
134 /* filter only on our function */ 133 /* filter only on our function */
135 ftrace_set_filter(func_name, strlen(func_name), 1); 134 ftrace_set_filter(func_name, strlen(func_name), 1);
136 135
137 /* enable tracing */ 136 /* enable tracing */
138 ret = trace->init(tr); 137 ret = tracer_init(trace, tr);
139 if (ret) { 138 if (ret) {
140 warn_failed_init_tracer(trace, ret); 139 warn_failed_init_tracer(trace, ret);
141 goto out; 140 goto out;
@@ -209,7 +208,7 @@ trace_selftest_startup_function(struct tracer *trace, struct trace_array *tr)
209 ftrace_enabled = 1; 208 ftrace_enabled = 1;
210 tracer_enabled = 1; 209 tracer_enabled = 1;
211 210
212 ret = trace->init(tr); 211 ret = tracer_init(trace, tr);
213 if (ret) { 212 if (ret) {
214 warn_failed_init_tracer(trace, ret); 213 warn_failed_init_tracer(trace, ret);
215 goto out; 214 goto out;
@@ -247,6 +246,90 @@ trace_selftest_startup_function(struct tracer *trace, struct trace_array *tr)
247} 246}
248#endif /* CONFIG_FUNCTION_TRACER */ 247#endif /* CONFIG_FUNCTION_TRACER */
249 248
249
250#ifdef CONFIG_FUNCTION_GRAPH_TRACER
251
252/* Maximum number of functions to trace before diagnosing a hang */
253#define GRAPH_MAX_FUNC_TEST 100000000
254
255static void __ftrace_dump(bool disable_tracing);
256static unsigned int graph_hang_thresh;
257
258/* Wrap the real function entry probe to avoid possible hanging */
259static int trace_graph_entry_watchdog(struct ftrace_graph_ent *trace)
260{
261 /* This is harmlessly racy, we want to approximately detect a hang */
262 if (unlikely(++graph_hang_thresh > GRAPH_MAX_FUNC_TEST)) {
263 ftrace_graph_stop();
264 printk(KERN_WARNING "BUG: Function graph tracer hang!\n");
265 if (ftrace_dump_on_oops)
266 __ftrace_dump(false);
267 return 0;
268 }
269
270 return trace_graph_entry(trace);
271}
272
273/*
274 * Pretty much the same than for the function tracer from which the selftest
275 * has been borrowed.
276 */
277int
278trace_selftest_startup_function_graph(struct tracer *trace,
279 struct trace_array *tr)
280{
281 int ret;
282 unsigned long count;
283
284 /*
285 * Simulate the init() callback but we attach a watchdog callback
286 * to detect and recover from possible hangs
287 */
288 tracing_reset_online_cpus(tr);
289 ret = register_ftrace_graph(&trace_graph_return,
290 &trace_graph_entry_watchdog);
291 if (ret) {
292 warn_failed_init_tracer(trace, ret);
293 goto out;
294 }
295 tracing_start_cmdline_record();
296
297 /* Sleep for a 1/10 of a second */
298 msleep(100);
299
300 /* Have we just recovered from a hang? */
301 if (graph_hang_thresh > GRAPH_MAX_FUNC_TEST) {
302 tracing_selftest_disabled = true;
303 ret = -1;
304 goto out;
305 }
306
307 tracing_stop();
308
309 /* check the trace buffer */
310 ret = trace_test_buffer(tr, &count);
311
312 trace->reset(tr);
313 tracing_start();
314
315 if (!ret && !count) {
316 printk(KERN_CONT ".. no entries found ..");
317 ret = -1;
318 goto out;
319 }
320
321 /* Don't test dynamic tracing, the function tracer already did */
322
323out:
324 /* Stop it if we failed */
325 if (ret)
326 ftrace_graph_stop();
327
328 return ret;
329}
330#endif /* CONFIG_FUNCTION_GRAPH_TRACER */
331
332
250#ifdef CONFIG_IRQSOFF_TRACER 333#ifdef CONFIG_IRQSOFF_TRACER
251int 334int
252trace_selftest_startup_irqsoff(struct tracer *trace, struct trace_array *tr) 335trace_selftest_startup_irqsoff(struct tracer *trace, struct trace_array *tr)
@@ -256,7 +339,7 @@ trace_selftest_startup_irqsoff(struct tracer *trace, struct trace_array *tr)
256 int ret; 339 int ret;
257 340
258 /* start the tracing */ 341 /* start the tracing */
259 ret = trace->init(tr); 342 ret = tracer_init(trace, tr);
260 if (ret) { 343 if (ret) {
261 warn_failed_init_tracer(trace, ret); 344 warn_failed_init_tracer(trace, ret);
262 return ret; 345 return ret;
@@ -268,6 +351,14 @@ trace_selftest_startup_irqsoff(struct tracer *trace, struct trace_array *tr)
268 local_irq_disable(); 351 local_irq_disable();
269 udelay(100); 352 udelay(100);
270 local_irq_enable(); 353 local_irq_enable();
354
355 /*
356 * Stop the tracer to avoid a warning subsequent
357 * to buffer flipping failure because tracing_stop()
358 * disables the tr and max buffers, making flipping impossible
359 * in case of parallels max irqs off latencies.
360 */
361 trace->stop(tr);
271 /* stop the tracing. */ 362 /* stop the tracing. */
272 tracing_stop(); 363 tracing_stop();
273 /* check both trace buffers */ 364 /* check both trace buffers */
@@ -310,7 +401,7 @@ trace_selftest_startup_preemptoff(struct tracer *trace, struct trace_array *tr)
310 } 401 }
311 402
312 /* start the tracing */ 403 /* start the tracing */
313 ret = trace->init(tr); 404 ret = tracer_init(trace, tr);
314 if (ret) { 405 if (ret) {
315 warn_failed_init_tracer(trace, ret); 406 warn_failed_init_tracer(trace, ret);
316 return ret; 407 return ret;
@@ -322,6 +413,14 @@ trace_selftest_startup_preemptoff(struct tracer *trace, struct trace_array *tr)
322 preempt_disable(); 413 preempt_disable();
323 udelay(100); 414 udelay(100);
324 preempt_enable(); 415 preempt_enable();
416
417 /*
418 * Stop the tracer to avoid a warning subsequent
419 * to buffer flipping failure because tracing_stop()
420 * disables the tr and max buffers, making flipping impossible
421 * in case of parallels max preempt off latencies.
422 */
423 trace->stop(tr);
325 /* stop the tracing. */ 424 /* stop the tracing. */
326 tracing_stop(); 425 tracing_stop();
327 /* check both trace buffers */ 426 /* check both trace buffers */
@@ -364,10 +463,10 @@ trace_selftest_startup_preemptirqsoff(struct tracer *trace, struct trace_array *
364 } 463 }
365 464
366 /* start the tracing */ 465 /* start the tracing */
367 ret = trace->init(tr); 466 ret = tracer_init(trace, tr);
368 if (ret) { 467 if (ret) {
369 warn_failed_init_tracer(trace, ret); 468 warn_failed_init_tracer(trace, ret);
370 goto out; 469 goto out_no_start;
371 } 470 }
372 471
373 /* reset the max latency */ 472 /* reset the max latency */
@@ -381,31 +480,35 @@ trace_selftest_startup_preemptirqsoff(struct tracer *trace, struct trace_array *
381 /* reverse the order of preempt vs irqs */ 480 /* reverse the order of preempt vs irqs */
382 local_irq_enable(); 481 local_irq_enable();
383 482
483 /*
484 * Stop the tracer to avoid a warning subsequent
485 * to buffer flipping failure because tracing_stop()
486 * disables the tr and max buffers, making flipping impossible
487 * in case of parallels max irqs/preempt off latencies.
488 */
489 trace->stop(tr);
384 /* stop the tracing. */ 490 /* stop the tracing. */
385 tracing_stop(); 491 tracing_stop();
386 /* check both trace buffers */ 492 /* check both trace buffers */
387 ret = trace_test_buffer(tr, NULL); 493 ret = trace_test_buffer(tr, NULL);
388 if (ret) { 494 if (ret)
389 tracing_start();
390 goto out; 495 goto out;
391 }
392 496
393 ret = trace_test_buffer(&max_tr, &count); 497 ret = trace_test_buffer(&max_tr, &count);
394 if (ret) { 498 if (ret)
395 tracing_start();
396 goto out; 499 goto out;
397 }
398 500
399 if (!ret && !count) { 501 if (!ret && !count) {
400 printk(KERN_CONT ".. no entries found .."); 502 printk(KERN_CONT ".. no entries found ..");
401 ret = -1; 503 ret = -1;
402 tracing_start();
403 goto out; 504 goto out;
404 } 505 }
405 506
406 /* do the test by disabling interrupts first this time */ 507 /* do the test by disabling interrupts first this time */
407 tracing_max_latency = 0; 508 tracing_max_latency = 0;
408 tracing_start(); 509 tracing_start();
510 trace->start(tr);
511
409 preempt_disable(); 512 preempt_disable();
410 local_irq_disable(); 513 local_irq_disable();
411 udelay(100); 514 udelay(100);
@@ -413,6 +516,7 @@ trace_selftest_startup_preemptirqsoff(struct tracer *trace, struct trace_array *
413 /* reverse the order of preempt vs irqs */ 516 /* reverse the order of preempt vs irqs */
414 local_irq_enable(); 517 local_irq_enable();
415 518
519 trace->stop(tr);
416 /* stop the tracing. */ 520 /* stop the tracing. */
417 tracing_stop(); 521 tracing_stop();
418 /* check both trace buffers */ 522 /* check both trace buffers */
@@ -428,9 +532,10 @@ trace_selftest_startup_preemptirqsoff(struct tracer *trace, struct trace_array *
428 goto out; 532 goto out;
429 } 533 }
430 534
431 out: 535out:
432 trace->reset(tr);
433 tracing_start(); 536 tracing_start();
537out_no_start:
538 trace->reset(tr);
434 tracing_max_latency = save_max; 539 tracing_max_latency = save_max;
435 540
436 return ret; 541 return ret;
@@ -496,7 +601,7 @@ trace_selftest_startup_wakeup(struct tracer *trace, struct trace_array *tr)
496 wait_for_completion(&isrt); 601 wait_for_completion(&isrt);
497 602
498 /* start the tracing */ 603 /* start the tracing */
499 ret = trace->init(tr); 604 ret = tracer_init(trace, tr);
500 if (ret) { 605 if (ret) {
501 warn_failed_init_tracer(trace, ret); 606 warn_failed_init_tracer(trace, ret);
502 return ret; 607 return ret;
@@ -557,7 +662,7 @@ trace_selftest_startup_sched_switch(struct tracer *trace, struct trace_array *tr
557 int ret; 662 int ret;
558 663
559 /* start the tracing */ 664 /* start the tracing */
560 ret = trace->init(tr); 665 ret = tracer_init(trace, tr);
561 if (ret) { 666 if (ret) {
562 warn_failed_init_tracer(trace, ret); 667 warn_failed_init_tracer(trace, ret);
563 return ret; 668 return ret;
@@ -589,10 +694,10 @@ trace_selftest_startup_sysprof(struct tracer *trace, struct trace_array *tr)
589 int ret; 694 int ret;
590 695
591 /* start the tracing */ 696 /* start the tracing */
592 ret = trace->init(tr); 697 ret = tracer_init(trace, tr);
593 if (ret) { 698 if (ret) {
594 warn_failed_init_tracer(trace, ret); 699 warn_failed_init_tracer(trace, ret);
595 return 0; 700 return ret;
596 } 701 }
597 702
598 /* Sleep for a 1/10 of a second */ 703 /* Sleep for a 1/10 of a second */
@@ -604,6 +709,11 @@ trace_selftest_startup_sysprof(struct tracer *trace, struct trace_array *tr)
604 trace->reset(tr); 709 trace->reset(tr);
605 tracing_start(); 710 tracing_start();
606 711
712 if (!ret && !count) {
713 printk(KERN_CONT ".. no entries found ..");
714 ret = -1;
715 }
716
607 return ret; 717 return ret;
608} 718}
609#endif /* CONFIG_SYSPROF_TRACER */ 719#endif /* CONFIG_SYSPROF_TRACER */
@@ -616,7 +726,7 @@ trace_selftest_startup_branch(struct tracer *trace, struct trace_array *tr)
616 int ret; 726 int ret;
617 727
618 /* start the tracing */ 728 /* start the tracing */
619 ret = trace->init(tr); 729 ret = tracer_init(trace, tr);
620 if (ret) { 730 if (ret) {
621 warn_failed_init_tracer(trace, ret); 731 warn_failed_init_tracer(trace, ret);
622 return ret; 732 return ret;
@@ -631,6 +741,11 @@ trace_selftest_startup_branch(struct tracer *trace, struct trace_array *tr)
631 trace->reset(tr); 741 trace->reset(tr);
632 tracing_start(); 742 tracing_start();
633 743
744 if (!ret && !count) {
745 printk(KERN_CONT ".. no entries found ..");
746 ret = -1;
747 }
748
634 return ret; 749 return ret;
635} 750}
636#endif /* CONFIG_BRANCH_TRACER */ 751#endif /* CONFIG_BRANCH_TRACER */
diff --git a/kernel/trace/trace_stack.c b/kernel/trace/trace_stack.c
index d0871bc0aca5..c750f65f9661 100644
--- a/kernel/trace/trace_stack.c
+++ b/kernel/trace/trace_stack.c
@@ -245,16 +245,31 @@ static int trace_lookup_stack(struct seq_file *m, long i)
245#endif 245#endif
246} 246}
247 247
248static void print_disabled(struct seq_file *m)
249{
250 seq_puts(m, "#\n"
251 "# Stack tracer disabled\n"
252 "#\n"
253 "# To enable the stack tracer, either add 'stacktrace' to the\n"
254 "# kernel command line\n"
255 "# or 'echo 1 > /proc/sys/kernel/stack_tracer_enabled'\n"
256 "#\n");
257}
258
248static int t_show(struct seq_file *m, void *v) 259static int t_show(struct seq_file *m, void *v)
249{ 260{
250 long i; 261 long i;
251 int size; 262 int size;
252 263
253 if (v == SEQ_START_TOKEN) { 264 if (v == SEQ_START_TOKEN) {
254 seq_printf(m, " Depth Size Location" 265 seq_printf(m, " Depth Size Location"
255 " (%d entries)\n" 266 " (%d entries)\n"
256 " ----- ---- --------\n", 267 " ----- ---- --------\n",
257 max_stack_trace.nr_entries); 268 max_stack_trace.nr_entries);
269
270 if (!stack_tracer_enabled && !max_stack_size)
271 print_disabled(m);
272
258 return 0; 273 return 0;
259 } 274 }
260 275
diff --git a/kernel/trace/trace_stat.c b/kernel/trace/trace_stat.c
new file mode 100644
index 000000000000..acdebd771a93
--- /dev/null
+++ b/kernel/trace/trace_stat.c
@@ -0,0 +1,326 @@
1/*
2 * Infrastructure for statistic tracing (histogram output).
3 *
4 * Copyright (C) 2008 Frederic Weisbecker <fweisbec@gmail.com>
5 *
6 * Based on the code from trace_branch.c which is
7 * Copyright (C) 2008 Steven Rostedt <srostedt@redhat.com>
8 *
9 */
10
11
12#include <linux/list.h>
13#include <linux/debugfs.h>
14#include "trace_stat.h"
15#include "trace.h"
16
17
18/* List of stat entries from a tracer */
19struct trace_stat_list {
20 struct list_head list;
21 void *stat;
22};
23
24/* A stat session is the stats output in one file */
25struct tracer_stat_session {
26 struct list_head session_list;
27 struct tracer_stat *ts;
28 struct list_head stat_list;
29 struct mutex stat_mutex;
30 struct dentry *file;
31};
32
33/* All of the sessions currently in use. Each stat file embed one session */
34static LIST_HEAD(all_stat_sessions);
35static DEFINE_MUTEX(all_stat_sessions_mutex);
36
37/* The root directory for all stat files */
38static struct dentry *stat_dir;
39
40
41static void reset_stat_session(struct tracer_stat_session *session)
42{
43 struct trace_stat_list *node, *next;
44
45 list_for_each_entry_safe(node, next, &session->stat_list, list)
46 kfree(node);
47
48 INIT_LIST_HEAD(&session->stat_list);
49}
50
51static void destroy_session(struct tracer_stat_session *session)
52{
53 debugfs_remove(session->file);
54 reset_stat_session(session);
55 mutex_destroy(&session->stat_mutex);
56 kfree(session);
57}
58
59/*
60 * For tracers that don't provide a stat_cmp callback.
61 * This one will force an immediate insertion on tail of
62 * the list.
63 */
64static int dummy_cmp(void *p1, void *p2)
65{
66 return 1;
67}
68
69/*
70 * Initialize the stat list at each trace_stat file opening.
71 * All of these copies and sorting are required on all opening
72 * since the stats could have changed between two file sessions.
73 */
74static int stat_seq_init(struct tracer_stat_session *session)
75{
76 struct trace_stat_list *iter_entry, *new_entry;
77 struct tracer_stat *ts = session->ts;
78 void *stat;
79 int ret = 0;
80 int i;
81
82 mutex_lock(&session->stat_mutex);
83 reset_stat_session(session);
84
85 if (!ts->stat_cmp)
86 ts->stat_cmp = dummy_cmp;
87
88 stat = ts->stat_start();
89 if (!stat)
90 goto exit;
91
92 /*
93 * The first entry. Actually this is the second, but the first
94 * one (the stat_list head) is pointless.
95 */
96 new_entry = kmalloc(sizeof(struct trace_stat_list), GFP_KERNEL);
97 if (!new_entry) {
98 ret = -ENOMEM;
99 goto exit;
100 }
101
102 INIT_LIST_HEAD(&new_entry->list);
103
104 list_add(&new_entry->list, &session->stat_list);
105
106 new_entry->stat = stat;
107
108 /*
109 * Iterate over the tracer stat entries and store them in a sorted
110 * list.
111 */
112 for (i = 1; ; i++) {
113 stat = ts->stat_next(stat, i);
114
115 /* End of insertion */
116 if (!stat)
117 break;
118
119 new_entry = kmalloc(sizeof(struct trace_stat_list), GFP_KERNEL);
120 if (!new_entry) {
121 ret = -ENOMEM;
122 goto exit_free_list;
123 }
124
125 INIT_LIST_HEAD(&new_entry->list);
126 new_entry->stat = stat;
127
128 list_for_each_entry_reverse(iter_entry, &session->stat_list,
129 list) {
130
131 /* Insertion with a descendent sorting */
132 if (ts->stat_cmp(iter_entry->stat,
133 new_entry->stat) >= 0) {
134
135 list_add(&new_entry->list, &iter_entry->list);
136 break;
137 }
138 }
139
140 /* The current larger value */
141 if (list_empty(&new_entry->list))
142 list_add(&new_entry->list, &session->stat_list);
143 }
144exit:
145 mutex_unlock(&session->stat_mutex);
146 return ret;
147
148exit_free_list:
149 reset_stat_session(session);
150 mutex_unlock(&session->stat_mutex);
151 return ret;
152}
153
154
155static void *stat_seq_start(struct seq_file *s, loff_t *pos)
156{
157 struct tracer_stat_session *session = s->private;
158
159 /* Prevent from tracer switch or stat_list modification */
160 mutex_lock(&session->stat_mutex);
161
162 /* If we are in the beginning of the file, print the headers */
163 if (!*pos && session->ts->stat_headers)
164 return SEQ_START_TOKEN;
165
166 return seq_list_start(&session->stat_list, *pos);
167}
168
169static void *stat_seq_next(struct seq_file *s, void *p, loff_t *pos)
170{
171 struct tracer_stat_session *session = s->private;
172
173 if (p == SEQ_START_TOKEN)
174 return seq_list_start(&session->stat_list, *pos);
175
176 return seq_list_next(p, &session->stat_list, pos);
177}
178
179static void stat_seq_stop(struct seq_file *s, void *p)
180{
181 struct tracer_stat_session *session = s->private;
182 mutex_unlock(&session->stat_mutex);
183}
184
185static int stat_seq_show(struct seq_file *s, void *v)
186{
187 struct tracer_stat_session *session = s->private;
188 struct trace_stat_list *l = list_entry(v, struct trace_stat_list, list);
189
190 if (v == SEQ_START_TOKEN)
191 return session->ts->stat_headers(s);
192
193 return session->ts->stat_show(s, l->stat);
194}
195
196static const struct seq_operations trace_stat_seq_ops = {
197 .start = stat_seq_start,
198 .next = stat_seq_next,
199 .stop = stat_seq_stop,
200 .show = stat_seq_show
201};
202
203/* The session stat is refilled and resorted at each stat file opening */
204static int tracing_stat_open(struct inode *inode, struct file *file)
205{
206 int ret;
207
208 struct tracer_stat_session *session = inode->i_private;
209
210 ret = seq_open(file, &trace_stat_seq_ops);
211 if (!ret) {
212 struct seq_file *m = file->private_data;
213 m->private = session;
214 ret = stat_seq_init(session);
215 }
216
217 return ret;
218}
219
220/*
221 * Avoid consuming memory with our now useless list.
222 */
223static int tracing_stat_release(struct inode *i, struct file *f)
224{
225 struct tracer_stat_session *session = i->i_private;
226
227 mutex_lock(&session->stat_mutex);
228 reset_stat_session(session);
229 mutex_unlock(&session->stat_mutex);
230
231 return 0;
232}
233
234static const struct file_operations tracing_stat_fops = {
235 .open = tracing_stat_open,
236 .read = seq_read,
237 .llseek = seq_lseek,
238 .release = tracing_stat_release
239};
240
241static int tracing_stat_init(void)
242{
243 struct dentry *d_tracing;
244
245 d_tracing = tracing_init_dentry();
246
247 stat_dir = debugfs_create_dir("trace_stat", d_tracing);
248 if (!stat_dir)
249 pr_warning("Could not create debugfs "
250 "'trace_stat' entry\n");
251 return 0;
252}
253
254static int init_stat_file(struct tracer_stat_session *session)
255{
256 if (!stat_dir && tracing_stat_init())
257 return -ENODEV;
258
259 session->file = debugfs_create_file(session->ts->name, 0644,
260 stat_dir,
261 session, &tracing_stat_fops);
262 if (!session->file)
263 return -ENOMEM;
264 return 0;
265}
266
267int register_stat_tracer(struct tracer_stat *trace)
268{
269 struct tracer_stat_session *session, *node, *tmp;
270 int ret;
271
272 if (!trace)
273 return -EINVAL;
274
275 if (!trace->stat_start || !trace->stat_next || !trace->stat_show)
276 return -EINVAL;
277
278 /* Already registered? */
279 mutex_lock(&all_stat_sessions_mutex);
280 list_for_each_entry_safe(node, tmp, &all_stat_sessions, session_list) {
281 if (node->ts == trace) {
282 mutex_unlock(&all_stat_sessions_mutex);
283 return -EINVAL;
284 }
285 }
286 mutex_unlock(&all_stat_sessions_mutex);
287
288 /* Init the session */
289 session = kmalloc(sizeof(struct tracer_stat_session), GFP_KERNEL);
290 if (!session)
291 return -ENOMEM;
292
293 session->ts = trace;
294 INIT_LIST_HEAD(&session->session_list);
295 INIT_LIST_HEAD(&session->stat_list);
296 mutex_init(&session->stat_mutex);
297 session->file = NULL;
298
299 ret = init_stat_file(session);
300 if (ret) {
301 destroy_session(session);
302 return ret;
303 }
304
305 /* Register */
306 mutex_lock(&all_stat_sessions_mutex);
307 list_add_tail(&session->session_list, &all_stat_sessions);
308 mutex_unlock(&all_stat_sessions_mutex);
309
310 return 0;
311}
312
313void unregister_stat_tracer(struct tracer_stat *trace)
314{
315 struct tracer_stat_session *node, *tmp;
316
317 mutex_lock(&all_stat_sessions_mutex);
318 list_for_each_entry_safe(node, tmp, &all_stat_sessions, session_list) {
319 if (node->ts == trace) {
320 list_del(&node->session_list);
321 destroy_session(node);
322 break;
323 }
324 }
325 mutex_unlock(&all_stat_sessions_mutex);
326}
diff --git a/kernel/trace/trace_stat.h b/kernel/trace/trace_stat.h
new file mode 100644
index 000000000000..202274cf7f3d
--- /dev/null
+++ b/kernel/trace/trace_stat.h
@@ -0,0 +1,31 @@
1#ifndef __TRACE_STAT_H
2#define __TRACE_STAT_H
3
4#include <linux/seq_file.h>
5
6/*
7 * If you want to provide a stat file (one-shot statistics), fill
8 * an iterator with stat_start/stat_next and a stat_show callbacks.
9 * The others callbacks are optional.
10 */
11struct tracer_stat {
12 /* The name of your stat file */
13 const char *name;
14 /* Iteration over statistic entries */
15 void *(*stat_start)(void);
16 void *(*stat_next)(void *prev, int idx);
17 /* Compare two entries for stats sorting */
18 int (*stat_cmp)(void *p1, void *p2);
19 /* Print a stat entry */
20 int (*stat_show)(struct seq_file *s, void *p);
21 /* Print the headers of your stat entries */
22 int (*stat_headers)(struct seq_file *s);
23};
24
25/*
26 * Destroy or create a stat file
27 */
28extern int register_stat_tracer(struct tracer_stat *trace);
29extern void unregister_stat_tracer(struct tracer_stat *trace);
30
31#endif /* __TRACE_STAT_H */
diff --git a/kernel/trace/trace_syscalls.c b/kernel/trace/trace_syscalls.c
new file mode 100644
index 000000000000..a2a3af29c943
--- /dev/null
+++ b/kernel/trace/trace_syscalls.c
@@ -0,0 +1,250 @@
1#include <linux/kernel.h>
2#include <linux/ftrace.h>
3#include <asm/syscall.h>
4
5#include "trace_output.h"
6#include "trace.h"
7
8/* Keep a counter of the syscall tracing users */
9static int refcount;
10
11/* Prevent from races on thread flags toggling */
12static DEFINE_MUTEX(syscall_trace_lock);
13
14/* Option to display the parameters types */
15enum {
16 TRACE_SYSCALLS_OPT_TYPES = 0x1,
17};
18
19static struct tracer_opt syscalls_opts[] = {
20 { TRACER_OPT(syscall_arg_type, TRACE_SYSCALLS_OPT_TYPES) },
21 { }
22};
23
24static struct tracer_flags syscalls_flags = {
25 .val = 0, /* By default: no parameters types */
26 .opts = syscalls_opts
27};
28
29enum print_line_t
30print_syscall_enter(struct trace_iterator *iter, int flags)
31{
32 struct trace_seq *s = &iter->seq;
33 struct trace_entry *ent = iter->ent;
34 struct syscall_trace_enter *trace;
35 struct syscall_metadata *entry;
36 int i, ret, syscall;
37
38 trace_assign_type(trace, ent);
39
40 syscall = trace->nr;
41
42 entry = syscall_nr_to_meta(syscall);
43 if (!entry)
44 goto end;
45
46 ret = trace_seq_printf(s, "%s(", entry->name);
47 if (!ret)
48 return TRACE_TYPE_PARTIAL_LINE;
49
50 for (i = 0; i < entry->nb_args; i++) {
51 /* parameter types */
52 if (syscalls_flags.val & TRACE_SYSCALLS_OPT_TYPES) {
53 ret = trace_seq_printf(s, "%s ", entry->types[i]);
54 if (!ret)
55 return TRACE_TYPE_PARTIAL_LINE;
56 }
57 /* parameter values */
58 ret = trace_seq_printf(s, "%s: %lx%s ", entry->args[i],
59 trace->args[i],
60 i == entry->nb_args - 1 ? ")" : ",");
61 if (!ret)
62 return TRACE_TYPE_PARTIAL_LINE;
63 }
64
65end:
66 trace_seq_printf(s, "\n");
67 return TRACE_TYPE_HANDLED;
68}
69
70enum print_line_t
71print_syscall_exit(struct trace_iterator *iter, int flags)
72{
73 struct trace_seq *s = &iter->seq;
74 struct trace_entry *ent = iter->ent;
75 struct syscall_trace_exit *trace;
76 int syscall;
77 struct syscall_metadata *entry;
78 int ret;
79
80 trace_assign_type(trace, ent);
81
82 syscall = trace->nr;
83
84 entry = syscall_nr_to_meta(syscall);
85 if (!entry) {
86 trace_seq_printf(s, "\n");
87 return TRACE_TYPE_HANDLED;
88 }
89
90 ret = trace_seq_printf(s, "%s -> 0x%lx\n", entry->name,
91 trace->ret);
92 if (!ret)
93 return TRACE_TYPE_PARTIAL_LINE;
94
95 return TRACE_TYPE_HANDLED;
96}
97
98void start_ftrace_syscalls(void)
99{
100 unsigned long flags;
101 struct task_struct *g, *t;
102
103 mutex_lock(&syscall_trace_lock);
104
105 /* Don't enable the flag on the tasks twice */
106 if (++refcount != 1)
107 goto unlock;
108
109 arch_init_ftrace_syscalls();
110 read_lock_irqsave(&tasklist_lock, flags);
111
112 do_each_thread(g, t) {
113 set_tsk_thread_flag(t, TIF_SYSCALL_FTRACE);
114 } while_each_thread(g, t);
115
116 read_unlock_irqrestore(&tasklist_lock, flags);
117
118unlock:
119 mutex_unlock(&syscall_trace_lock);
120}
121
122void stop_ftrace_syscalls(void)
123{
124 unsigned long flags;
125 struct task_struct *g, *t;
126
127 mutex_lock(&syscall_trace_lock);
128
129 /* There are perhaps still some users */
130 if (--refcount)
131 goto unlock;
132
133 read_lock_irqsave(&tasklist_lock, flags);
134
135 do_each_thread(g, t) {
136 clear_tsk_thread_flag(t, TIF_SYSCALL_FTRACE);
137 } while_each_thread(g, t);
138
139 read_unlock_irqrestore(&tasklist_lock, flags);
140
141unlock:
142 mutex_unlock(&syscall_trace_lock);
143}
144
145void ftrace_syscall_enter(struct pt_regs *regs)
146{
147 struct syscall_trace_enter *entry;
148 struct syscall_metadata *sys_data;
149 struct ring_buffer_event *event;
150 int size;
151 int syscall_nr;
152
153 syscall_nr = syscall_get_nr(current, regs);
154
155 sys_data = syscall_nr_to_meta(syscall_nr);
156 if (!sys_data)
157 return;
158
159 size = sizeof(*entry) + sizeof(unsigned long) * sys_data->nb_args;
160
161 event = trace_current_buffer_lock_reserve(TRACE_SYSCALL_ENTER, size,
162 0, 0);
163 if (!event)
164 return;
165
166 entry = ring_buffer_event_data(event);
167 entry->nr = syscall_nr;
168 syscall_get_arguments(current, regs, 0, sys_data->nb_args, entry->args);
169
170 trace_current_buffer_unlock_commit(event, 0, 0);
171 trace_wake_up();
172}
173
174void ftrace_syscall_exit(struct pt_regs *regs)
175{
176 struct syscall_trace_exit *entry;
177 struct syscall_metadata *sys_data;
178 struct ring_buffer_event *event;
179 int syscall_nr;
180
181 syscall_nr = syscall_get_nr(current, regs);
182
183 sys_data = syscall_nr_to_meta(syscall_nr);
184 if (!sys_data)
185 return;
186
187 event = trace_current_buffer_lock_reserve(TRACE_SYSCALL_EXIT,
188 sizeof(*entry), 0, 0);
189 if (!event)
190 return;
191
192 entry = ring_buffer_event_data(event);
193 entry->nr = syscall_nr;
194 entry->ret = syscall_get_return_value(current, regs);
195
196 trace_current_buffer_unlock_commit(event, 0, 0);
197 trace_wake_up();
198}
199
200static int init_syscall_tracer(struct trace_array *tr)
201{
202 start_ftrace_syscalls();
203
204 return 0;
205}
206
207static void reset_syscall_tracer(struct trace_array *tr)
208{
209 stop_ftrace_syscalls();
210 tracing_reset_online_cpus(tr);
211}
212
213static struct trace_event syscall_enter_event = {
214 .type = TRACE_SYSCALL_ENTER,
215 .trace = print_syscall_enter,
216};
217
218static struct trace_event syscall_exit_event = {
219 .type = TRACE_SYSCALL_EXIT,
220 .trace = print_syscall_exit,
221};
222
223static struct tracer syscall_tracer __read_mostly = {
224 .name = "syscall",
225 .init = init_syscall_tracer,
226 .reset = reset_syscall_tracer,
227 .flags = &syscalls_flags,
228};
229
230__init int register_ftrace_syscalls(void)
231{
232 int ret;
233
234 ret = register_ftrace_event(&syscall_enter_event);
235 if (!ret) {
236 printk(KERN_WARNING "event %d failed to register\n",
237 syscall_enter_event.type);
238 WARN_ON_ONCE(1);
239 }
240
241 ret = register_ftrace_event(&syscall_exit_event);
242 if (!ret) {
243 printk(KERN_WARNING "event %d failed to register\n",
244 syscall_exit_event.type);
245 WARN_ON_ONCE(1);
246 }
247
248 return register_tracer(&syscall_tracer);
249}
250device_initcall(register_ftrace_syscalls);
diff --git a/kernel/trace/trace_sysprof.c b/kernel/trace/trace_sysprof.c
index eaca5ad803ff..91fd19c2149f 100644
--- a/kernel/trace/trace_sysprof.c
+++ b/kernel/trace/trace_sysprof.c
@@ -88,7 +88,7 @@ static void backtrace_address(void *data, unsigned long addr, int reliable)
88 } 88 }
89} 89}
90 90
91const static struct stacktrace_ops backtrace_ops = { 91static const struct stacktrace_ops backtrace_ops = {
92 .warning = backtrace_warning, 92 .warning = backtrace_warning,
93 .warning_symbol = backtrace_warning_symbol, 93 .warning_symbol = backtrace_warning_symbol,
94 .stack = backtrace_stack, 94 .stack = backtrace_stack,
@@ -226,15 +226,6 @@ static void stop_stack_timers(void)
226 stop_stack_timer(cpu); 226 stop_stack_timer(cpu);
227} 227}
228 228
229static void start_stack_trace(struct trace_array *tr)
230{
231 mutex_lock(&sample_timer_lock);
232 tracing_reset_online_cpus(tr);
233 start_stack_timers();
234 tracer_enabled = 1;
235 mutex_unlock(&sample_timer_lock);
236}
237
238static void stop_stack_trace(struct trace_array *tr) 229static void stop_stack_trace(struct trace_array *tr)
239{ 230{
240 mutex_lock(&sample_timer_lock); 231 mutex_lock(&sample_timer_lock);
@@ -247,12 +238,18 @@ static int stack_trace_init(struct trace_array *tr)
247{ 238{
248 sysprof_trace = tr; 239 sysprof_trace = tr;
249 240
250 start_stack_trace(tr); 241 tracing_start_cmdline_record();
242
243 mutex_lock(&sample_timer_lock);
244 start_stack_timers();
245 tracer_enabled = 1;
246 mutex_unlock(&sample_timer_lock);
251 return 0; 247 return 0;
252} 248}
253 249
254static void stack_trace_reset(struct trace_array *tr) 250static void stack_trace_reset(struct trace_array *tr)
255{ 251{
252 tracing_stop_cmdline_record();
256 stop_stack_trace(tr); 253 stop_stack_trace(tr);
257} 254}
258 255
@@ -317,7 +314,7 @@ sysprof_sample_write(struct file *filp, const char __user *ubuf,
317 return cnt; 314 return cnt;
318} 315}
319 316
320static struct file_operations sysprof_sample_fops = { 317static const struct file_operations sysprof_sample_fops = {
321 .read = sysprof_sample_read, 318 .read = sysprof_sample_read,
322 .write = sysprof_sample_write, 319 .write = sysprof_sample_write,
323}; 320};
@@ -330,5 +327,5 @@ void init_tracer_sysprof_debugfs(struct dentry *d_tracer)
330 d_tracer, NULL, &sysprof_sample_fops); 327 d_tracer, NULL, &sysprof_sample_fops);
331 if (entry) 328 if (entry)
332 return; 329 return;
333 pr_warning("Could not create debugfs 'dyn_ftrace_total_info' entry\n"); 330 pr_warning("Could not create debugfs 'sysprof_sample_period' entry\n");
334} 331}
diff --git a/kernel/trace/trace_workqueue.c b/kernel/trace/trace_workqueue.c
new file mode 100644
index 000000000000..797201e4a137
--- /dev/null
+++ b/kernel/trace/trace_workqueue.c
@@ -0,0 +1,288 @@
1/*
2 * Workqueue statistical tracer.
3 *
4 * Copyright (C) 2008 Frederic Weisbecker <fweisbec@gmail.com>
5 *
6 */
7
8
9#include <trace/workqueue.h>
10#include <linux/list.h>
11#include <linux/percpu.h>
12#include "trace_stat.h"
13#include "trace.h"
14
15
16/* A cpu workqueue thread */
17struct cpu_workqueue_stats {
18 struct list_head list;
19/* Useful to know if we print the cpu headers */
20 bool first_entry;
21 int cpu;
22 pid_t pid;
23/* Can be inserted from interrupt or user context, need to be atomic */
24 atomic_t inserted;
25/*
26 * Don't need to be atomic, works are serialized in a single workqueue thread
27 * on a single CPU.
28 */
29 unsigned int executed;
30};
31
32/* List of workqueue threads on one cpu */
33struct workqueue_global_stats {
34 struct list_head list;
35 spinlock_t lock;
36};
37
38/* Don't need a global lock because allocated before the workqueues, and
39 * never freed.
40 */
41static DEFINE_PER_CPU(struct workqueue_global_stats, all_workqueue_stat);
42#define workqueue_cpu_stat(cpu) (&per_cpu(all_workqueue_stat, cpu))
43
44/* Insertion of a work */
45static void
46probe_workqueue_insertion(struct task_struct *wq_thread,
47 struct work_struct *work)
48{
49 int cpu = cpumask_first(&wq_thread->cpus_allowed);
50 struct cpu_workqueue_stats *node, *next;
51 unsigned long flags;
52
53 spin_lock_irqsave(&workqueue_cpu_stat(cpu)->lock, flags);
54 list_for_each_entry_safe(node, next, &workqueue_cpu_stat(cpu)->list,
55 list) {
56 if (node->pid == wq_thread->pid) {
57 atomic_inc(&node->inserted);
58 goto found;
59 }
60 }
61 pr_debug("trace_workqueue: entry not found\n");
62found:
63 spin_unlock_irqrestore(&workqueue_cpu_stat(cpu)->lock, flags);
64}
65
66/* Execution of a work */
67static void
68probe_workqueue_execution(struct task_struct *wq_thread,
69 struct work_struct *work)
70{
71 int cpu = cpumask_first(&wq_thread->cpus_allowed);
72 struct cpu_workqueue_stats *node, *next;
73 unsigned long flags;
74
75 spin_lock_irqsave(&workqueue_cpu_stat(cpu)->lock, flags);
76 list_for_each_entry_safe(node, next, &workqueue_cpu_stat(cpu)->list,
77 list) {
78 if (node->pid == wq_thread->pid) {
79 node->executed++;
80 goto found;
81 }
82 }
83 pr_debug("trace_workqueue: entry not found\n");
84found:
85 spin_unlock_irqrestore(&workqueue_cpu_stat(cpu)->lock, flags);
86}
87
88/* Creation of a cpu workqueue thread */
89static void probe_workqueue_creation(struct task_struct *wq_thread, int cpu)
90{
91 struct cpu_workqueue_stats *cws;
92 unsigned long flags;
93
94 WARN_ON(cpu < 0);
95
96 /* Workqueues are sometimes created in atomic context */
97 cws = kzalloc(sizeof(struct cpu_workqueue_stats), GFP_ATOMIC);
98 if (!cws) {
99 pr_warning("trace_workqueue: not enough memory\n");
100 return;
101 }
102 INIT_LIST_HEAD(&cws->list);
103 cws->cpu = cpu;
104
105 cws->pid = wq_thread->pid;
106
107 spin_lock_irqsave(&workqueue_cpu_stat(cpu)->lock, flags);
108 if (list_empty(&workqueue_cpu_stat(cpu)->list))
109 cws->first_entry = true;
110 list_add_tail(&cws->list, &workqueue_cpu_stat(cpu)->list);
111 spin_unlock_irqrestore(&workqueue_cpu_stat(cpu)->lock, flags);
112}
113
114/* Destruction of a cpu workqueue thread */
115static void probe_workqueue_destruction(struct task_struct *wq_thread)
116{
117 /* Workqueue only execute on one cpu */
118 int cpu = cpumask_first(&wq_thread->cpus_allowed);
119 struct cpu_workqueue_stats *node, *next;
120 unsigned long flags;
121
122 spin_lock_irqsave(&workqueue_cpu_stat(cpu)->lock, flags);
123 list_for_each_entry_safe(node, next, &workqueue_cpu_stat(cpu)->list,
124 list) {
125 if (node->pid == wq_thread->pid) {
126 list_del(&node->list);
127 kfree(node);
128 goto found;
129 }
130 }
131
132 pr_debug("trace_workqueue: don't find workqueue to destroy\n");
133found:
134 spin_unlock_irqrestore(&workqueue_cpu_stat(cpu)->lock, flags);
135
136}
137
138static struct cpu_workqueue_stats *workqueue_stat_start_cpu(int cpu)
139{
140 unsigned long flags;
141 struct cpu_workqueue_stats *ret = NULL;
142
143
144 spin_lock_irqsave(&workqueue_cpu_stat(cpu)->lock, flags);
145
146 if (!list_empty(&workqueue_cpu_stat(cpu)->list))
147 ret = list_entry(workqueue_cpu_stat(cpu)->list.next,
148 struct cpu_workqueue_stats, list);
149
150 spin_unlock_irqrestore(&workqueue_cpu_stat(cpu)->lock, flags);
151
152 return ret;
153}
154
155static void *workqueue_stat_start(void)
156{
157 int cpu;
158 void *ret = NULL;
159
160 for_each_possible_cpu(cpu) {
161 ret = workqueue_stat_start_cpu(cpu);
162 if (ret)
163 return ret;
164 }
165 return NULL;
166}
167
168static void *workqueue_stat_next(void *prev, int idx)
169{
170 struct cpu_workqueue_stats *prev_cws = prev;
171 int cpu = prev_cws->cpu;
172 unsigned long flags;
173 void *ret = NULL;
174
175 spin_lock_irqsave(&workqueue_cpu_stat(cpu)->lock, flags);
176 if (list_is_last(&prev_cws->list, &workqueue_cpu_stat(cpu)->list)) {
177 spin_unlock_irqrestore(&workqueue_cpu_stat(cpu)->lock, flags);
178 do {
179 cpu = cpumask_next(cpu, cpu_possible_mask);
180 if (cpu >= nr_cpu_ids)
181 return NULL;
182 } while (!(ret = workqueue_stat_start_cpu(cpu)));
183 return ret;
184 }
185 spin_unlock_irqrestore(&workqueue_cpu_stat(cpu)->lock, flags);
186
187 return list_entry(prev_cws->list.next, struct cpu_workqueue_stats,
188 list);
189}
190
191static int workqueue_stat_show(struct seq_file *s, void *p)
192{
193 struct cpu_workqueue_stats *cws = p;
194 unsigned long flags;
195 int cpu = cws->cpu;
196 struct pid *pid;
197 struct task_struct *tsk;
198
199 spin_lock_irqsave(&workqueue_cpu_stat(cpu)->lock, flags);
200 if (&cws->list == workqueue_cpu_stat(cpu)->list.next)
201 seq_printf(s, "\n");
202 spin_unlock_irqrestore(&workqueue_cpu_stat(cpu)->lock, flags);
203
204 pid = find_get_pid(cws->pid);
205 if (pid) {
206 tsk = get_pid_task(pid, PIDTYPE_PID);
207 if (tsk) {
208 seq_printf(s, "%3d %6d %6u %s\n", cws->cpu,
209 atomic_read(&cws->inserted), cws->executed,
210 tsk->comm);
211 put_task_struct(tsk);
212 }
213 put_pid(pid);
214 }
215
216 return 0;
217}
218
219static int workqueue_stat_headers(struct seq_file *s)
220{
221 seq_printf(s, "# CPU INSERTED EXECUTED NAME\n");
222 seq_printf(s, "# | | | |\n");
223 return 0;
224}
225
226struct tracer_stat workqueue_stats __read_mostly = {
227 .name = "workqueues",
228 .stat_start = workqueue_stat_start,
229 .stat_next = workqueue_stat_next,
230 .stat_show = workqueue_stat_show,
231 .stat_headers = workqueue_stat_headers
232};
233
234
235int __init stat_workqueue_init(void)
236{
237 if (register_stat_tracer(&workqueue_stats)) {
238 pr_warning("Unable to register workqueue stat tracer\n");
239 return 1;
240 }
241
242 return 0;
243}
244fs_initcall(stat_workqueue_init);
245
246/*
247 * Workqueues are created very early, just after pre-smp initcalls.
248 * So we must register our tracepoints at this stage.
249 */
250int __init trace_workqueue_early_init(void)
251{
252 int ret, cpu;
253
254 ret = register_trace_workqueue_insertion(probe_workqueue_insertion);
255 if (ret)
256 goto out;
257
258 ret = register_trace_workqueue_execution(probe_workqueue_execution);
259 if (ret)
260 goto no_insertion;
261
262 ret = register_trace_workqueue_creation(probe_workqueue_creation);
263 if (ret)
264 goto no_execution;
265
266 ret = register_trace_workqueue_destruction(probe_workqueue_destruction);
267 if (ret)
268 goto no_creation;
269
270 for_each_possible_cpu(cpu) {
271 spin_lock_init(&workqueue_cpu_stat(cpu)->lock);
272 INIT_LIST_HEAD(&workqueue_cpu_stat(cpu)->list);
273 }
274
275 return 0;
276
277no_creation:
278 unregister_trace_workqueue_creation(probe_workqueue_creation);
279no_execution:
280 unregister_trace_workqueue_execution(probe_workqueue_execution);
281no_insertion:
282 unregister_trace_workqueue_insertion(probe_workqueue_insertion);
283out:
284 pr_warning("trace_workqueue: unable to trace workqueues\n");
285
286 return 1;
287}
288early_initcall(trace_workqueue_early_init);
diff --git a/kernel/tracepoint.c b/kernel/tracepoint.c
index 79602740bbb5..1ef5d3a601c7 100644
--- a/kernel/tracepoint.c
+++ b/kernel/tracepoint.c
@@ -272,12 +272,15 @@ static void disable_tracepoint(struct tracepoint *elem)
272 * 272 *
273 * Updates the probe callback corresponding to a range of tracepoints. 273 * Updates the probe callback corresponding to a range of tracepoints.
274 */ 274 */
275void tracepoint_update_probe_range(struct tracepoint *begin, 275void
276 struct tracepoint *end) 276tracepoint_update_probe_range(struct tracepoint *begin, struct tracepoint *end)
277{ 277{
278 struct tracepoint *iter; 278 struct tracepoint *iter;
279 struct tracepoint_entry *mark_entry; 279 struct tracepoint_entry *mark_entry;
280 280
281 if (!begin)
282 return;
283
281 mutex_lock(&tracepoints_mutex); 284 mutex_lock(&tracepoints_mutex);
282 for (iter = begin; iter < end; iter++) { 285 for (iter = begin; iter < end; iter++) {
283 mark_entry = get_tracepoint(iter->name); 286 mark_entry = get_tracepoint(iter->name);
diff --git a/kernel/tsacct.c b/kernel/tsacct.c
index 43f891b05a4b..00d59d048edf 100644
--- a/kernel/tsacct.c
+++ b/kernel/tsacct.c
@@ -122,8 +122,10 @@ void acct_update_integrals(struct task_struct *tsk)
122 if (likely(tsk->mm)) { 122 if (likely(tsk->mm)) {
123 cputime_t time, dtime; 123 cputime_t time, dtime;
124 struct timeval value; 124 struct timeval value;
125 unsigned long flags;
125 u64 delta; 126 u64 delta;
126 127
128 local_irq_save(flags);
127 time = tsk->stime + tsk->utime; 129 time = tsk->stime + tsk->utime;
128 dtime = cputime_sub(time, tsk->acct_timexpd); 130 dtime = cputime_sub(time, tsk->acct_timexpd);
129 jiffies_to_timeval(cputime_to_jiffies(dtime), &value); 131 jiffies_to_timeval(cputime_to_jiffies(dtime), &value);
@@ -131,10 +133,12 @@ void acct_update_integrals(struct task_struct *tsk)
131 delta = delta * USEC_PER_SEC + value.tv_usec; 133 delta = delta * USEC_PER_SEC + value.tv_usec;
132 134
133 if (delta == 0) 135 if (delta == 0)
134 return; 136 goto out;
135 tsk->acct_timexpd = time; 137 tsk->acct_timexpd = time;
136 tsk->acct_rss_mem1 += delta * get_mm_rss(tsk->mm); 138 tsk->acct_rss_mem1 += delta * get_mm_rss(tsk->mm);
137 tsk->acct_vm_mem1 += delta * tsk->mm->total_vm; 139 tsk->acct_vm_mem1 += delta * tsk->mm->total_vm;
140 out:
141 local_irq_restore(flags);
138 } 142 }
139} 143}
140 144
diff --git a/kernel/user.c b/kernel/user.c
index 6a9b696128c8..850e0ba41c1e 100644
--- a/kernel/user.c
+++ b/kernel/user.c
@@ -20,7 +20,7 @@
20 20
21struct user_namespace init_user_ns = { 21struct user_namespace init_user_ns = {
22 .kref = { 22 .kref = {
23 .refcount = ATOMIC_INIT(1), 23 .refcount = ATOMIC_INIT(2),
24 }, 24 },
25 .creator = &root_user, 25 .creator = &root_user,
26}; 26};
@@ -286,14 +286,12 @@ int __init uids_sysfs_init(void)
286/* work function to remove sysfs directory for a user and free up 286/* work function to remove sysfs directory for a user and free up
287 * corresponding structures. 287 * corresponding structures.
288 */ 288 */
289static void remove_user_sysfs_dir(struct work_struct *w) 289static void cleanup_user_struct(struct work_struct *w)
290{ 290{
291 struct user_struct *up = container_of(w, struct user_struct, work); 291 struct user_struct *up = container_of(w, struct user_struct, work);
292 unsigned long flags; 292 unsigned long flags;
293 int remove_user = 0; 293 int remove_user = 0;
294 294
295 if (up->user_ns != &init_user_ns)
296 return;
297 /* Make uid_hash_remove() + sysfs_remove_file() + kobject_del() 295 /* Make uid_hash_remove() + sysfs_remove_file() + kobject_del()
298 * atomic. 296 * atomic.
299 */ 297 */
@@ -312,9 +310,11 @@ static void remove_user_sysfs_dir(struct work_struct *w)
312 if (!remove_user) 310 if (!remove_user)
313 goto done; 311 goto done;
314 312
315 kobject_uevent(&up->kobj, KOBJ_REMOVE); 313 if (up->user_ns == &init_user_ns) {
316 kobject_del(&up->kobj); 314 kobject_uevent(&up->kobj, KOBJ_REMOVE);
317 kobject_put(&up->kobj); 315 kobject_del(&up->kobj);
316 kobject_put(&up->kobj);
317 }
318 318
319 sched_destroy_user(up); 319 sched_destroy_user(up);
320 key_put(up->uid_keyring); 320 key_put(up->uid_keyring);
@@ -335,7 +335,7 @@ static void free_user(struct user_struct *up, unsigned long flags)
335 atomic_inc(&up->__count); 335 atomic_inc(&up->__count);
336 spin_unlock_irqrestore(&uidhash_lock, flags); 336 spin_unlock_irqrestore(&uidhash_lock, flags);
337 337
338 INIT_WORK(&up->work, remove_user_sysfs_dir); 338 INIT_WORK(&up->work, cleanup_user_struct);
339 schedule_work(&up->work); 339 schedule_work(&up->work);
340} 340}
341 341
diff --git a/kernel/utsname_sysctl.c b/kernel/utsname_sysctl.c
index 3b34b3545936..92359cc747a7 100644
--- a/kernel/utsname_sysctl.c
+++ b/kernel/utsname_sysctl.c
@@ -37,7 +37,7 @@ static void put_uts(ctl_table *table, int write, void *which)
37 up_write(&uts_sem); 37 up_write(&uts_sem);
38} 38}
39 39
40#ifdef CONFIG_PROC_FS 40#ifdef CONFIG_PROC_SYSCTL
41/* 41/*
42 * Special case of dostring for the UTS structure. This has locks 42 * Special case of dostring for the UTS structure. This has locks
43 * to observe. Should this be in kernel/sys.c ???? 43 * to observe. Should this be in kernel/sys.c ????
diff --git a/kernel/workqueue.c b/kernel/workqueue.c
index 1f0c509b40d3..b6b966ce1451 100644
--- a/kernel/workqueue.c
+++ b/kernel/workqueue.c
@@ -33,6 +33,7 @@
33#include <linux/kallsyms.h> 33#include <linux/kallsyms.h>
34#include <linux/debug_locks.h> 34#include <linux/debug_locks.h>
35#include <linux/lockdep.h> 35#include <linux/lockdep.h>
36#include <trace/workqueue.h>
36 37
37/* 38/*
38 * The per-CPU workqueue (if single thread, we always use the first 39 * The per-CPU workqueue (if single thread, we always use the first
@@ -48,8 +49,6 @@ struct cpu_workqueue_struct {
48 49
49 struct workqueue_struct *wq; 50 struct workqueue_struct *wq;
50 struct task_struct *thread; 51 struct task_struct *thread;
51
52 int run_depth; /* Detect run_workqueue() recursion depth */
53} ____cacheline_aligned; 52} ____cacheline_aligned;
54 53
55/* 54/*
@@ -125,9 +124,13 @@ struct cpu_workqueue_struct *get_wq_data(struct work_struct *work)
125 return (void *) (atomic_long_read(&work->data) & WORK_STRUCT_WQ_DATA_MASK); 124 return (void *) (atomic_long_read(&work->data) & WORK_STRUCT_WQ_DATA_MASK);
126} 125}
127 126
127DEFINE_TRACE(workqueue_insertion);
128
128static void insert_work(struct cpu_workqueue_struct *cwq, 129static void insert_work(struct cpu_workqueue_struct *cwq,
129 struct work_struct *work, struct list_head *head) 130 struct work_struct *work, struct list_head *head)
130{ 131{
132 trace_workqueue_insertion(cwq->thread, work);
133
131 set_wq_data(work, cwq); 134 set_wq_data(work, cwq);
132 /* 135 /*
133 * Ensure that we get the right work->data if we see the 136 * Ensure that we get the right work->data if we see the
@@ -259,16 +262,11 @@ int queue_delayed_work_on(int cpu, struct workqueue_struct *wq,
259} 262}
260EXPORT_SYMBOL_GPL(queue_delayed_work_on); 263EXPORT_SYMBOL_GPL(queue_delayed_work_on);
261 264
265DEFINE_TRACE(workqueue_execution);
266
262static void run_workqueue(struct cpu_workqueue_struct *cwq) 267static void run_workqueue(struct cpu_workqueue_struct *cwq)
263{ 268{
264 spin_lock_irq(&cwq->lock); 269 spin_lock_irq(&cwq->lock);
265 cwq->run_depth++;
266 if (cwq->run_depth > 3) {
267 /* morton gets to eat his hat */
268 printk("%s: recursion depth exceeded: %d\n",
269 __func__, cwq->run_depth);
270 dump_stack();
271 }
272 while (!list_empty(&cwq->worklist)) { 270 while (!list_empty(&cwq->worklist)) {
273 struct work_struct *work = list_entry(cwq->worklist.next, 271 struct work_struct *work = list_entry(cwq->worklist.next,
274 struct work_struct, entry); 272 struct work_struct, entry);
@@ -284,7 +282,7 @@ static void run_workqueue(struct cpu_workqueue_struct *cwq)
284 */ 282 */
285 struct lockdep_map lockdep_map = work->lockdep_map; 283 struct lockdep_map lockdep_map = work->lockdep_map;
286#endif 284#endif
287 285 trace_workqueue_execution(cwq->thread, work);
288 cwq->current_work = work; 286 cwq->current_work = work;
289 list_del_init(cwq->worklist.next); 287 list_del_init(cwq->worklist.next);
290 spin_unlock_irq(&cwq->lock); 288 spin_unlock_irq(&cwq->lock);
@@ -311,7 +309,6 @@ static void run_workqueue(struct cpu_workqueue_struct *cwq)
311 spin_lock_irq(&cwq->lock); 309 spin_lock_irq(&cwq->lock);
312 cwq->current_work = NULL; 310 cwq->current_work = NULL;
313 } 311 }
314 cwq->run_depth--;
315 spin_unlock_irq(&cwq->lock); 312 spin_unlock_irq(&cwq->lock);
316} 313}
317 314
@@ -368,29 +365,20 @@ static void insert_wq_barrier(struct cpu_workqueue_struct *cwq,
368 365
369static int flush_cpu_workqueue(struct cpu_workqueue_struct *cwq) 366static int flush_cpu_workqueue(struct cpu_workqueue_struct *cwq)
370{ 367{
371 int active; 368 int active = 0;
372 369 struct wq_barrier barr;
373 if (cwq->thread == current) {
374 /*
375 * Probably keventd trying to flush its own queue. So simply run
376 * it by hand rather than deadlocking.
377 */
378 run_workqueue(cwq);
379 active = 1;
380 } else {
381 struct wq_barrier barr;
382 370
383 active = 0; 371 WARN_ON(cwq->thread == current);
384 spin_lock_irq(&cwq->lock);
385 if (!list_empty(&cwq->worklist) || cwq->current_work != NULL) {
386 insert_wq_barrier(cwq, &barr, &cwq->worklist);
387 active = 1;
388 }
389 spin_unlock_irq(&cwq->lock);
390 372
391 if (active) 373 spin_lock_irq(&cwq->lock);
392 wait_for_completion(&barr.done); 374 if (!list_empty(&cwq->worklist) || cwq->current_work != NULL) {
375 insert_wq_barrier(cwq, &barr, &cwq->worklist);
376 active = 1;
393 } 377 }
378 spin_unlock_irq(&cwq->lock);
379
380 if (active)
381 wait_for_completion(&barr.done);
394 382
395 return active; 383 return active;
396} 384}
@@ -416,7 +404,7 @@ void flush_workqueue(struct workqueue_struct *wq)
416 might_sleep(); 404 might_sleep();
417 lock_map_acquire(&wq->lockdep_map); 405 lock_map_acquire(&wq->lockdep_map);
418 lock_map_release(&wq->lockdep_map); 406 lock_map_release(&wq->lockdep_map);
419 for_each_cpu_mask_nr(cpu, *cpu_map) 407 for_each_cpu(cpu, cpu_map)
420 flush_cpu_workqueue(per_cpu_ptr(wq->cpu_wq, cpu)); 408 flush_cpu_workqueue(per_cpu_ptr(wq->cpu_wq, cpu));
421} 409}
422EXPORT_SYMBOL_GPL(flush_workqueue); 410EXPORT_SYMBOL_GPL(flush_workqueue);
@@ -547,7 +535,7 @@ static void wait_on_work(struct work_struct *work)
547 wq = cwq->wq; 535 wq = cwq->wq;
548 cpu_map = wq_cpu_map(wq); 536 cpu_map = wq_cpu_map(wq);
549 537
550 for_each_cpu_mask_nr(cpu, *cpu_map) 538 for_each_cpu(cpu, cpu_map)
551 wait_on_cpu_work(per_cpu_ptr(wq->cpu_wq, cpu), work); 539 wait_on_cpu_work(per_cpu_ptr(wq->cpu_wq, cpu), work);
552} 540}
553 541
@@ -765,6 +753,8 @@ init_cpu_workqueue(struct workqueue_struct *wq, int cpu)
765 return cwq; 753 return cwq;
766} 754}
767 755
756DEFINE_TRACE(workqueue_creation);
757
768static int create_workqueue_thread(struct cpu_workqueue_struct *cwq, int cpu) 758static int create_workqueue_thread(struct cpu_workqueue_struct *cwq, int cpu)
769{ 759{
770 struct sched_param param = { .sched_priority = MAX_RT_PRIO-1 }; 760 struct sched_param param = { .sched_priority = MAX_RT_PRIO-1 };
@@ -787,6 +777,8 @@ static int create_workqueue_thread(struct cpu_workqueue_struct *cwq, int cpu)
787 sched_setscheduler_nocheck(p, SCHED_FIFO, &param); 777 sched_setscheduler_nocheck(p, SCHED_FIFO, &param);
788 cwq->thread = p; 778 cwq->thread = p;
789 779
780 trace_workqueue_creation(cwq->thread, cpu);
781
790 return 0; 782 return 0;
791} 783}
792 784
@@ -868,6 +860,8 @@ struct workqueue_struct *__create_workqueue_key(const char *name,
868} 860}
869EXPORT_SYMBOL_GPL(__create_workqueue_key); 861EXPORT_SYMBOL_GPL(__create_workqueue_key);
870 862
863DEFINE_TRACE(workqueue_destruction);
864
871static void cleanup_workqueue_thread(struct cpu_workqueue_struct *cwq) 865static void cleanup_workqueue_thread(struct cpu_workqueue_struct *cwq)
872{ 866{
873 /* 867 /*
@@ -891,6 +885,7 @@ static void cleanup_workqueue_thread(struct cpu_workqueue_struct *cwq)
891 * checks list_empty(), and a "normal" queue_work() can't use 885 * checks list_empty(), and a "normal" queue_work() can't use
892 * a dead CPU. 886 * a dead CPU.
893 */ 887 */
888 trace_workqueue_destruction(cwq->thread);
894 kthread_stop(cwq->thread); 889 kthread_stop(cwq->thread);
895 cwq->thread = NULL; 890 cwq->thread = NULL;
896} 891}
@@ -911,7 +906,7 @@ void destroy_workqueue(struct workqueue_struct *wq)
911 list_del(&wq->list); 906 list_del(&wq->list);
912 spin_unlock(&workqueue_lock); 907 spin_unlock(&workqueue_lock);
913 908
914 for_each_cpu_mask_nr(cpu, *cpu_map) 909 for_each_cpu(cpu, cpu_map)
915 cleanup_workqueue_thread(per_cpu_ptr(wq->cpu_wq, cpu)); 910 cleanup_workqueue_thread(per_cpu_ptr(wq->cpu_wq, cpu));
916 cpu_maps_update_done(); 911 cpu_maps_update_done();
917 912
generated by cgit v1.2.2 (git 2.25.0) at 2024-11-28 07:35:00 -0500