aboutsummaryrefslogtreecommitdiffstats
path: root/kernel
diff options
context:
space:
mode:
Diffstat (limited to 'kernel')
-rw-r--r--kernel/Kconfig.locks103
-rw-r--r--kernel/Makefile5
-rw-r--r--kernel/acct.c10
-rw-r--r--kernel/audit.c144
-rw-r--r--kernel/audit.h11
-rw-r--r--kernel/audit_watch.c5
-rw-r--r--kernel/auditfilter.c202
-rw-r--r--kernel/auditsc.c510
-rw-r--r--kernel/cgroup.c332
-rw-r--r--kernel/cgroup_freezer.c8
-rw-r--r--kernel/cpu.c25
-rw-r--r--kernel/cred.c10
-rw-r--r--kernel/debug/debug_core.c32
-rw-r--r--kernel/debug/kdb/kdb_bt.c2
-rw-r--r--kernel/debug/kdb/kdb_io.c33
-rw-r--r--kernel/debug/kdb/kdb_main.c33
-rw-r--r--kernel/events/callchain.c38
-rw-r--r--kernel/events/core.c316
-rw-r--r--kernel/events/internal.h82
-rw-r--r--kernel/events/ring_buffer.c10
-rw-r--r--kernel/events/uprobes.c256
-rw-r--r--kernel/exit.c100
-rw-r--r--kernel/fork.c56
-rw-r--r--kernel/irq/chip.c1
-rw-r--r--kernel/irq/dummychip.c2
-rw-r--r--kernel/irq/irqdomain.c33
-rw-r--r--kernel/jump_label.c1
-rw-r--r--kernel/kexec.c1
-rw-r--r--kernel/kmod.c7
-rw-r--r--kernel/kprobes.c247
-rw-r--r--kernel/kthread.c186
-rw-r--r--kernel/lockdep.c39
-rw-r--r--kernel/pid.c1
-rw-r--r--kernel/pid_namespace.c2
-rw-r--r--kernel/power/Kconfig4
-rw-r--r--kernel/power/poweroff.c2
-rw-r--r--kernel/power/process.c2
-rw-r--r--kernel/power/qos.c1
-rw-r--r--kernel/ptrace.c3
-rw-r--r--kernel/rcupdate.c4
-rw-r--r--kernel/rcutiny.c33
-rw-r--r--kernel/rcutiny_plugin.h10
-rw-r--r--kernel/rcutorture.c159
-rw-r--r--kernel/rcutree.c933
-rw-r--r--kernel/rcutree.h56
-rw-r--r--kernel/rcutree_plugin.h597
-rw-r--r--kernel/rcutree_trace.c22
-rw-r--r--kernel/resource.c50
-rw-r--r--kernel/sched/Makefile2
-rw-r--r--kernel/sched/core.c765
-rw-r--r--kernel/sched/cputime.c530
-rw-r--r--kernel/sched/fair.c81
-rw-r--r--kernel/sched/features.h10
-rw-r--r--kernel/sched/rt.c5
-rw-r--r--kernel/sched/sched.h69
-rw-r--r--kernel/signal.c21
-rw-r--r--kernel/smpboot.c233
-rw-r--r--kernel/smpboot.h4
-rw-r--r--kernel/softirq.c117
-rw-r--r--kernel/srcu.c4
-rw-r--r--kernel/sys.c17
-rw-r--r--kernel/sysctl.c21
-rw-r--r--kernel/task_work.c111
-rw-r--r--kernel/taskstats.c39
-rw-r--r--kernel/time.c2
-rw-r--r--kernel/time/Kconfig4
-rw-r--r--kernel/time/alarmtimer.c118
-rw-r--r--kernel/time/clockevents.c24
-rw-r--r--kernel/time/jiffies.c32
-rw-r--r--kernel/time/tick-sched.c8
-rw-r--r--kernel/time/timekeeping.c138
-rw-r--r--kernel/timer.c118
-rw-r--r--kernel/trace/Kconfig10
-rw-r--r--kernel/trace/Makefile8
-rw-r--r--kernel/trace/ftrace.c322
-rw-r--r--kernel/trace/ring_buffer.c4
-rw-r--r--kernel/trace/trace.c23
-rw-r--r--kernel/trace/trace.h5
-rw-r--r--kernel/trace/trace_event_perf.c3
-rw-r--r--kernel/trace/trace_events.c116
-rw-r--r--kernel/trace/trace_events_filter.c2
-rw-r--r--kernel/trace/trace_functions.c29
-rw-r--r--kernel/trace/trace_functions_graph.c5
-rw-r--r--kernel/trace/trace_irqsoff.c5
-rw-r--r--kernel/trace/trace_sched_wakeup.c5
-rw-r--r--kernel/trace/trace_selftest.c304
-rw-r--r--kernel/trace/trace_stack.c4
-rw-r--r--kernel/trace/trace_syscalls.c2
-rw-r--r--kernel/tsacct.c12
-rw-r--r--kernel/user.c8
-rw-r--r--kernel/user_namespace.c128
-rw-r--r--kernel/watchdog.c263
-rw-r--r--kernel/workqueue.c1242
93 files changed, 5755 insertions, 3937 deletions
diff --git a/kernel/Kconfig.locks b/kernel/Kconfig.locks
index 2251882daf53..44511d100eaa 100644
--- a/kernel/Kconfig.locks
+++ b/kernel/Kconfig.locks
@@ -87,6 +87,9 @@ config ARCH_INLINE_WRITE_UNLOCK_IRQ
87config ARCH_INLINE_WRITE_UNLOCK_IRQRESTORE 87config ARCH_INLINE_WRITE_UNLOCK_IRQRESTORE
88 bool 88 bool
89 89
90config UNINLINE_SPIN_UNLOCK
91 bool
92
90# 93#
91# lock_* functions are inlined when: 94# lock_* functions are inlined when:
92# - DEBUG_SPINLOCK=n and GENERIC_LOCKBREAK=n and ARCH_INLINE_*LOCK=y 95# - DEBUG_SPINLOCK=n and GENERIC_LOCKBREAK=n and ARCH_INLINE_*LOCK=y
@@ -103,100 +106,120 @@ config ARCH_INLINE_WRITE_UNLOCK_IRQRESTORE
103# - DEBUG_SPINLOCK=n and ARCH_INLINE_*LOCK=y 106# - DEBUG_SPINLOCK=n and ARCH_INLINE_*LOCK=y
104# 107#
105 108
109if !DEBUG_SPINLOCK
110
106config INLINE_SPIN_TRYLOCK 111config INLINE_SPIN_TRYLOCK
107 def_bool !DEBUG_SPINLOCK && ARCH_INLINE_SPIN_TRYLOCK 112 def_bool y
113 depends on ARCH_INLINE_SPIN_TRYLOCK
108 114
109config INLINE_SPIN_TRYLOCK_BH 115config INLINE_SPIN_TRYLOCK_BH
110 def_bool !DEBUG_SPINLOCK && ARCH_INLINE_SPIN_TRYLOCK_BH 116 def_bool y
117 depends on ARCH_INLINE_SPIN_TRYLOCK_BH
111 118
112config INLINE_SPIN_LOCK 119config INLINE_SPIN_LOCK
113 def_bool !DEBUG_SPINLOCK && !GENERIC_LOCKBREAK && ARCH_INLINE_SPIN_LOCK 120 def_bool y
121 depends on !GENERIC_LOCKBREAK && ARCH_INLINE_SPIN_LOCK
114 122
115config INLINE_SPIN_LOCK_BH 123config INLINE_SPIN_LOCK_BH
116 def_bool !DEBUG_SPINLOCK && !GENERIC_LOCKBREAK && \ 124 def_bool y
117 ARCH_INLINE_SPIN_LOCK_BH 125 depends on !GENERIC_LOCKBREAK && ARCH_INLINE_SPIN_LOCK_BH
118 126
119config INLINE_SPIN_LOCK_IRQ 127config INLINE_SPIN_LOCK_IRQ
120 def_bool !DEBUG_SPINLOCK && !GENERIC_LOCKBREAK && \ 128 def_bool y
121 ARCH_INLINE_SPIN_LOCK_IRQ 129 depends on !GENERIC_LOCKBREAK && ARCH_INLINE_SPIN_LOCK_IRQ
122 130
123config INLINE_SPIN_LOCK_IRQSAVE 131config INLINE_SPIN_LOCK_IRQSAVE
124 def_bool !DEBUG_SPINLOCK && !GENERIC_LOCKBREAK && \ 132 def_bool y
125 ARCH_INLINE_SPIN_LOCK_IRQSAVE 133 depends on !GENERIC_LOCKBREAK && ARCH_INLINE_SPIN_LOCK_IRQSAVE
126
127config UNINLINE_SPIN_UNLOCK
128 bool
129 134
130config INLINE_SPIN_UNLOCK_BH 135config INLINE_SPIN_UNLOCK_BH
131 def_bool !DEBUG_SPINLOCK && ARCH_INLINE_SPIN_UNLOCK_BH 136 def_bool y
137 depends on ARCH_INLINE_SPIN_UNLOCK_BH
132 138
133config INLINE_SPIN_UNLOCK_IRQ 139config INLINE_SPIN_UNLOCK_IRQ
134 def_bool !DEBUG_SPINLOCK && (!PREEMPT || ARCH_INLINE_SPIN_UNLOCK_BH) 140 def_bool y
141 depends on !PREEMPT || ARCH_INLINE_SPIN_UNLOCK_BH
135 142
136config INLINE_SPIN_UNLOCK_IRQRESTORE 143config INLINE_SPIN_UNLOCK_IRQRESTORE
137 def_bool !DEBUG_SPINLOCK && ARCH_INLINE_SPIN_UNLOCK_IRQRESTORE 144 def_bool y
145 depends on ARCH_INLINE_SPIN_UNLOCK_IRQRESTORE
138 146
139 147
140config INLINE_READ_TRYLOCK 148config INLINE_READ_TRYLOCK
141 def_bool !DEBUG_SPINLOCK && ARCH_INLINE_READ_TRYLOCK 149 def_bool y
150 depends on ARCH_INLINE_READ_TRYLOCK
142 151
143config INLINE_READ_LOCK 152config INLINE_READ_LOCK
144 def_bool !DEBUG_SPINLOCK && !GENERIC_LOCKBREAK && ARCH_INLINE_READ_LOCK 153 def_bool y
154 depends on !GENERIC_LOCKBREAK && ARCH_INLINE_READ_LOCK
145 155
146config INLINE_READ_LOCK_BH 156config INLINE_READ_LOCK_BH
147 def_bool !DEBUG_SPINLOCK && !GENERIC_LOCKBREAK && \ 157 def_bool y
148 ARCH_INLINE_READ_LOCK_BH 158 depends on !GENERIC_LOCKBREAK && ARCH_INLINE_READ_LOCK_BH
149 159
150config INLINE_READ_LOCK_IRQ 160config INLINE_READ_LOCK_IRQ
151 def_bool !DEBUG_SPINLOCK && !GENERIC_LOCKBREAK && \ 161 def_bool y
152 ARCH_INLINE_READ_LOCK_IRQ 162 depends on !GENERIC_LOCKBREAK && ARCH_INLINE_READ_LOCK_IRQ
153 163
154config INLINE_READ_LOCK_IRQSAVE 164config INLINE_READ_LOCK_IRQSAVE
155 def_bool !DEBUG_SPINLOCK && !GENERIC_LOCKBREAK && \ 165 def_bool y
156 ARCH_INLINE_READ_LOCK_IRQSAVE 166 depends on !GENERIC_LOCKBREAK && ARCH_INLINE_READ_LOCK_IRQSAVE
157 167
158config INLINE_READ_UNLOCK 168config INLINE_READ_UNLOCK
159 def_bool !DEBUG_SPINLOCK && (!PREEMPT || ARCH_INLINE_READ_UNLOCK) 169 def_bool y
170 depends on !PREEMPT || ARCH_INLINE_READ_UNLOCK
160 171
161config INLINE_READ_UNLOCK_BH 172config INLINE_READ_UNLOCK_BH
162 def_bool !DEBUG_SPINLOCK && ARCH_INLINE_READ_UNLOCK_BH 173 def_bool y
174 depends on ARCH_INLINE_READ_UNLOCK_BH
163 175
164config INLINE_READ_UNLOCK_IRQ 176config INLINE_READ_UNLOCK_IRQ
165 def_bool !DEBUG_SPINLOCK && (!PREEMPT || ARCH_INLINE_READ_UNLOCK_BH) 177 def_bool y
178 depends on !PREEMPT || ARCH_INLINE_READ_UNLOCK_BH
166 179
167config INLINE_READ_UNLOCK_IRQRESTORE 180config INLINE_READ_UNLOCK_IRQRESTORE
168 def_bool !DEBUG_SPINLOCK && ARCH_INLINE_READ_UNLOCK_IRQRESTORE 181 def_bool y
182 depends on ARCH_INLINE_READ_UNLOCK_IRQRESTORE
169 183
170 184
171config INLINE_WRITE_TRYLOCK 185config INLINE_WRITE_TRYLOCK
172 def_bool !DEBUG_SPINLOCK && ARCH_INLINE_WRITE_TRYLOCK 186 def_bool y
187 depends on ARCH_INLINE_WRITE_TRYLOCK
173 188
174config INLINE_WRITE_LOCK 189config INLINE_WRITE_LOCK
175 def_bool !DEBUG_SPINLOCK && !GENERIC_LOCKBREAK && ARCH_INLINE_WRITE_LOCK 190 def_bool y
191 depends on !GENERIC_LOCKBREAK && ARCH_INLINE_WRITE_LOCK
176 192
177config INLINE_WRITE_LOCK_BH 193config INLINE_WRITE_LOCK_BH
178 def_bool !DEBUG_SPINLOCK && !GENERIC_LOCKBREAK && \ 194 def_bool y
179 ARCH_INLINE_WRITE_LOCK_BH 195 depends on !GENERIC_LOCKBREAK && ARCH_INLINE_WRITE_LOCK_BH
180 196
181config INLINE_WRITE_LOCK_IRQ 197config INLINE_WRITE_LOCK_IRQ
182 def_bool !DEBUG_SPINLOCK && !GENERIC_LOCKBREAK && \ 198 def_bool y
183 ARCH_INLINE_WRITE_LOCK_IRQ 199 depends on !GENERIC_LOCKBREAK && ARCH_INLINE_WRITE_LOCK_IRQ
184 200
185config INLINE_WRITE_LOCK_IRQSAVE 201config INLINE_WRITE_LOCK_IRQSAVE
186 def_bool !DEBUG_SPINLOCK && !GENERIC_LOCKBREAK && \ 202 def_bool y
187 ARCH_INLINE_WRITE_LOCK_IRQSAVE 203 depends on !GENERIC_LOCKBREAK && ARCH_INLINE_WRITE_LOCK_IRQSAVE
188 204
189config INLINE_WRITE_UNLOCK 205config INLINE_WRITE_UNLOCK
190 def_bool !DEBUG_SPINLOCK && (!PREEMPT || ARCH_INLINE_WRITE_UNLOCK) 206 def_bool y
207 depends on !PREEMPT || ARCH_INLINE_WRITE_UNLOCK
191 208
192config INLINE_WRITE_UNLOCK_BH 209config INLINE_WRITE_UNLOCK_BH
193 def_bool !DEBUG_SPINLOCK && ARCH_INLINE_WRITE_UNLOCK_BH 210 def_bool y
211 depends on ARCH_INLINE_WRITE_UNLOCK_BH
194 212
195config INLINE_WRITE_UNLOCK_IRQ 213config INLINE_WRITE_UNLOCK_IRQ
196 def_bool !DEBUG_SPINLOCK && (!PREEMPT || ARCH_INLINE_WRITE_UNLOCK_BH) 214 def_bool y
215 depends on !PREEMPT || ARCH_INLINE_WRITE_UNLOCK_BH
197 216
198config INLINE_WRITE_UNLOCK_IRQRESTORE 217config INLINE_WRITE_UNLOCK_IRQRESTORE
199 def_bool !DEBUG_SPINLOCK && ARCH_INLINE_WRITE_UNLOCK_IRQRESTORE 218 def_bool y
219 depends on ARCH_INLINE_WRITE_UNLOCK_IRQRESTORE
220
221endif
200 222
201config MUTEX_SPIN_ON_OWNER 223config MUTEX_SPIN_ON_OWNER
202 def_bool SMP && !DEBUG_MUTEXES 224 def_bool y
225 depends on SMP && !DEBUG_MUTEXES
diff --git a/kernel/Makefile b/kernel/Makefile
index d3611c8a6b8d..0dfeca4324ee 100644
--- a/kernel/Makefile
+++ b/kernel/Makefile
@@ -10,7 +10,7 @@ obj-y = fork.o exec_domain.o panic.o printk.o \
10 kthread.o wait.o kfifo.o sys_ni.o posix-cpu-timers.o mutex.o \ 10 kthread.o wait.o kfifo.o sys_ni.o posix-cpu-timers.o mutex.o \
11 hrtimer.o rwsem.o nsproxy.o srcu.o semaphore.o \ 11 hrtimer.o rwsem.o nsproxy.o srcu.o semaphore.o \
12 notifier.o ksysfs.o cred.o \ 12 notifier.o ksysfs.o cred.o \
13 async.o range.o groups.o lglock.o 13 async.o range.o groups.o lglock.o smpboot.o
14 14
15ifdef CONFIG_FUNCTION_TRACER 15ifdef CONFIG_FUNCTION_TRACER
16# Do not trace debug files and internal ftrace files 16# Do not trace debug files and internal ftrace files
@@ -46,7 +46,6 @@ obj-$(CONFIG_DEBUG_RT_MUTEXES) += rtmutex-debug.o
46obj-$(CONFIG_RT_MUTEX_TESTER) += rtmutex-tester.o 46obj-$(CONFIG_RT_MUTEX_TESTER) += rtmutex-tester.o
47obj-$(CONFIG_GENERIC_ISA_DMA) += dma.o 47obj-$(CONFIG_GENERIC_ISA_DMA) += dma.o
48obj-$(CONFIG_SMP) += smp.o 48obj-$(CONFIG_SMP) += smp.o
49obj-$(CONFIG_SMP) += smpboot.o
50ifneq ($(CONFIG_SMP),y) 49ifneq ($(CONFIG_SMP),y)
51obj-y += up.o 50obj-y += up.o
52endif 51endif
@@ -99,7 +98,7 @@ obj-$(CONFIG_COMPAT_BINFMT_ELF) += elfcore.o
99obj-$(CONFIG_BINFMT_ELF_FDPIC) += elfcore.o 98obj-$(CONFIG_BINFMT_ELF_FDPIC) += elfcore.o
100obj-$(CONFIG_FUNCTION_TRACER) += trace/ 99obj-$(CONFIG_FUNCTION_TRACER) += trace/
101obj-$(CONFIG_TRACING) += trace/ 100obj-$(CONFIG_TRACING) += trace/
102obj-$(CONFIG_X86_DS) += trace/ 101obj-$(CONFIG_TRACE_CLOCK) += trace/
103obj-$(CONFIG_RING_BUFFER) += trace/ 102obj-$(CONFIG_RING_BUFFER) += trace/
104obj-$(CONFIG_TRACEPOINTS) += trace/ 103obj-$(CONFIG_TRACEPOINTS) += trace/
105obj-$(CONFIG_IRQ_WORK) += irq_work.o 104obj-$(CONFIG_IRQ_WORK) += irq_work.o
diff --git a/kernel/acct.c b/kernel/acct.c
index 02e6167a53b0..051e071a06e7 100644
--- a/kernel/acct.c
+++ b/kernel/acct.c
@@ -193,7 +193,7 @@ static void acct_file_reopen(struct bsd_acct_struct *acct, struct file *file,
193 } 193 }
194} 194}
195 195
196static int acct_on(char *name) 196static int acct_on(struct filename *pathname)
197{ 197{
198 struct file *file; 198 struct file *file;
199 struct vfsmount *mnt; 199 struct vfsmount *mnt;
@@ -201,7 +201,7 @@ static int acct_on(char *name)
201 struct bsd_acct_struct *acct = NULL; 201 struct bsd_acct_struct *acct = NULL;
202 202
203 /* Difference from BSD - they don't do O_APPEND */ 203 /* Difference from BSD - they don't do O_APPEND */
204 file = filp_open(name, O_WRONLY|O_APPEND|O_LARGEFILE, 0); 204 file = file_open_name(pathname, O_WRONLY|O_APPEND|O_LARGEFILE, 0);
205 if (IS_ERR(file)) 205 if (IS_ERR(file))
206 return PTR_ERR(file); 206 return PTR_ERR(file);
207 207
@@ -260,7 +260,7 @@ SYSCALL_DEFINE1(acct, const char __user *, name)
260 return -EPERM; 260 return -EPERM;
261 261
262 if (name) { 262 if (name) {
263 char *tmp = getname(name); 263 struct filename *tmp = getname(name);
264 if (IS_ERR(tmp)) 264 if (IS_ERR(tmp))
265 return (PTR_ERR(tmp)); 265 return (PTR_ERR(tmp));
266 error = acct_on(tmp); 266 error = acct_on(tmp);
@@ -507,8 +507,8 @@ static void do_acct_process(struct bsd_acct_struct *acct,
507 do_div(elapsed, AHZ); 507 do_div(elapsed, AHZ);
508 ac.ac_btime = get_seconds() - elapsed; 508 ac.ac_btime = get_seconds() - elapsed;
509 /* we really need to bite the bullet and change layout */ 509 /* we really need to bite the bullet and change layout */
510 ac.ac_uid = orig_cred->uid; 510 ac.ac_uid = from_kuid_munged(file->f_cred->user_ns, orig_cred->uid);
511 ac.ac_gid = orig_cred->gid; 511 ac.ac_gid = from_kgid_munged(file->f_cred->user_ns, orig_cred->gid);
512#if ACCT_VERSION==2 512#if ACCT_VERSION==2
513 ac.ac_ahz = AHZ; 513 ac.ac_ahz = AHZ;
514#endif 514#endif
diff --git a/kernel/audit.c b/kernel/audit.c
index ea3b7b6191c7..40414e9143db 100644
--- a/kernel/audit.c
+++ b/kernel/audit.c
@@ -61,6 +61,7 @@
61#include <linux/netlink.h> 61#include <linux/netlink.h>
62#include <linux/freezer.h> 62#include <linux/freezer.h>
63#include <linux/tty.h> 63#include <linux/tty.h>
64#include <linux/pid_namespace.h>
64 65
65#include "audit.h" 66#include "audit.h"
66 67
@@ -87,11 +88,11 @@ static int audit_failure = AUDIT_FAIL_PRINTK;
87 88
88/* 89/*
89 * If audit records are to be written to the netlink socket, audit_pid 90 * If audit records are to be written to the netlink socket, audit_pid
90 * contains the pid of the auditd process and audit_nlk_pid contains 91 * contains the pid of the auditd process and audit_nlk_portid contains
91 * the pid to use to send netlink messages to that process. 92 * the portid to use to send netlink messages to that process.
92 */ 93 */
93int audit_pid; 94int audit_pid;
94static int audit_nlk_pid; 95static int audit_nlk_portid;
95 96
96/* If audit_rate_limit is non-zero, limit the rate of sending audit records 97/* If audit_rate_limit is non-zero, limit the rate of sending audit records
97 * to that number per second. This prevents DoS attacks, but results in 98 * to that number per second. This prevents DoS attacks, but results in
@@ -104,7 +105,7 @@ static int audit_backlog_wait_time = 60 * HZ;
104static int audit_backlog_wait_overflow = 0; 105static int audit_backlog_wait_overflow = 0;
105 106
106/* The identity of the user shutting down the audit system. */ 107/* The identity of the user shutting down the audit system. */
107uid_t audit_sig_uid = -1; 108kuid_t audit_sig_uid = INVALID_UID;
108pid_t audit_sig_pid = -1; 109pid_t audit_sig_pid = -1;
109u32 audit_sig_sid = 0; 110u32 audit_sig_sid = 0;
110 111
@@ -264,7 +265,7 @@ void audit_log_lost(const char *message)
264} 265}
265 266
266static int audit_log_config_change(char *function_name, int new, int old, 267static int audit_log_config_change(char *function_name, int new, int old,
267 uid_t loginuid, u32 sessionid, u32 sid, 268 kuid_t loginuid, u32 sessionid, u32 sid,
268 int allow_changes) 269 int allow_changes)
269{ 270{
270 struct audit_buffer *ab; 271 struct audit_buffer *ab;
@@ -272,7 +273,7 @@ static int audit_log_config_change(char *function_name, int new, int old,
272 273
273 ab = audit_log_start(NULL, GFP_KERNEL, AUDIT_CONFIG_CHANGE); 274 ab = audit_log_start(NULL, GFP_KERNEL, AUDIT_CONFIG_CHANGE);
274 audit_log_format(ab, "%s=%d old=%d auid=%u ses=%u", function_name, new, 275 audit_log_format(ab, "%s=%d old=%d auid=%u ses=%u", function_name, new,
275 old, loginuid, sessionid); 276 old, from_kuid(&init_user_ns, loginuid), sessionid);
276 if (sid) { 277 if (sid) {
277 char *ctx = NULL; 278 char *ctx = NULL;
278 u32 len; 279 u32 len;
@@ -292,7 +293,7 @@ static int audit_log_config_change(char *function_name, int new, int old,
292} 293}
293 294
294static int audit_do_config_change(char *function_name, int *to_change, 295static int audit_do_config_change(char *function_name, int *to_change,
295 int new, uid_t loginuid, u32 sessionid, 296 int new, kuid_t loginuid, u32 sessionid,
296 u32 sid) 297 u32 sid)
297{ 298{
298 int allow_changes, rc = 0, old = *to_change; 299 int allow_changes, rc = 0, old = *to_change;
@@ -319,21 +320,21 @@ static int audit_do_config_change(char *function_name, int *to_change,
319 return rc; 320 return rc;
320} 321}
321 322
322static int audit_set_rate_limit(int limit, uid_t loginuid, u32 sessionid, 323static int audit_set_rate_limit(int limit, kuid_t loginuid, u32 sessionid,
323 u32 sid) 324 u32 sid)
324{ 325{
325 return audit_do_config_change("audit_rate_limit", &audit_rate_limit, 326 return audit_do_config_change("audit_rate_limit", &audit_rate_limit,
326 limit, loginuid, sessionid, sid); 327 limit, loginuid, sessionid, sid);
327} 328}
328 329
329static int audit_set_backlog_limit(int limit, uid_t loginuid, u32 sessionid, 330static int audit_set_backlog_limit(int limit, kuid_t loginuid, u32 sessionid,
330 u32 sid) 331 u32 sid)
331{ 332{
332 return audit_do_config_change("audit_backlog_limit", &audit_backlog_limit, 333 return audit_do_config_change("audit_backlog_limit", &audit_backlog_limit,
333 limit, loginuid, sessionid, sid); 334 limit, loginuid, sessionid, sid);
334} 335}
335 336
336static int audit_set_enabled(int state, uid_t loginuid, u32 sessionid, u32 sid) 337static int audit_set_enabled(int state, kuid_t loginuid, u32 sessionid, u32 sid)
337{ 338{
338 int rc; 339 int rc;
339 if (state < AUDIT_OFF || state > AUDIT_LOCKED) 340 if (state < AUDIT_OFF || state > AUDIT_LOCKED)
@@ -348,7 +349,7 @@ static int audit_set_enabled(int state, uid_t loginuid, u32 sessionid, u32 sid)
348 return rc; 349 return rc;
349} 350}
350 351
351static int audit_set_failure(int state, uid_t loginuid, u32 sessionid, u32 sid) 352static int audit_set_failure(int state, kuid_t loginuid, u32 sessionid, u32 sid)
352{ 353{
353 if (state != AUDIT_FAIL_SILENT 354 if (state != AUDIT_FAIL_SILENT
354 && state != AUDIT_FAIL_PRINTK 355 && state != AUDIT_FAIL_PRINTK
@@ -401,7 +402,7 @@ static void kauditd_send_skb(struct sk_buff *skb)
401 int err; 402 int err;
402 /* take a reference in case we can't send it and we want to hold it */ 403 /* take a reference in case we can't send it and we want to hold it */
403 skb_get(skb); 404 skb_get(skb);
404 err = netlink_unicast(audit_sock, skb, audit_nlk_pid, 0); 405 err = netlink_unicast(audit_sock, skb, audit_nlk_portid, 0);
405 if (err < 0) { 406 if (err < 0) {
406 BUG_ON(err != -ECONNREFUSED); /* Shouldn't happen */ 407 BUG_ON(err != -ECONNREFUSED); /* Shouldn't happen */
407 printk(KERN_ERR "audit: *NO* daemon at audit_pid=%d\n", audit_pid); 408 printk(KERN_ERR "audit: *NO* daemon at audit_pid=%d\n", audit_pid);
@@ -467,24 +468,6 @@ static int kauditd_thread(void *dummy)
467 return 0; 468 return 0;
468} 469}
469 470
470static int audit_prepare_user_tty(pid_t pid, uid_t loginuid, u32 sessionid)
471{
472 struct task_struct *tsk;
473 int err;
474
475 rcu_read_lock();
476 tsk = find_task_by_vpid(pid);
477 if (!tsk) {
478 rcu_read_unlock();
479 return -ESRCH;
480 }
481 get_task_struct(tsk);
482 rcu_read_unlock();
483 err = tty_audit_push_task(tsk, loginuid, sessionid);
484 put_task_struct(tsk);
485 return err;
486}
487
488int audit_send_list(void *_dest) 471int audit_send_list(void *_dest)
489{ 472{
490 struct audit_netlink_list *dest = _dest; 473 struct audit_netlink_list *dest = _dest;
@@ -588,6 +571,11 @@ static int audit_netlink_ok(struct sk_buff *skb, u16 msg_type)
588{ 571{
589 int err = 0; 572 int err = 0;
590 573
574 /* Only support the initial namespaces for now. */
575 if ((current_user_ns() != &init_user_ns) ||
576 (task_active_pid_ns(current) != &init_pid_ns))
577 return -EPERM;
578
591 switch (msg_type) { 579 switch (msg_type) {
592 case AUDIT_GET: 580 case AUDIT_GET:
593 case AUDIT_LIST: 581 case AUDIT_LIST:
@@ -619,8 +607,7 @@ static int audit_netlink_ok(struct sk_buff *skb, u16 msg_type)
619} 607}
620 608
621static int audit_log_common_recv_msg(struct audit_buffer **ab, u16 msg_type, 609static int audit_log_common_recv_msg(struct audit_buffer **ab, u16 msg_type,
622 u32 pid, u32 uid, uid_t auid, u32 ses, 610 kuid_t auid, u32 ses, u32 sid)
623 u32 sid)
624{ 611{
625 int rc = 0; 612 int rc = 0;
626 char *ctx = NULL; 613 char *ctx = NULL;
@@ -633,7 +620,9 @@ static int audit_log_common_recv_msg(struct audit_buffer **ab, u16 msg_type,
633 620
634 *ab = audit_log_start(NULL, GFP_KERNEL, msg_type); 621 *ab = audit_log_start(NULL, GFP_KERNEL, msg_type);
635 audit_log_format(*ab, "pid=%d uid=%u auid=%u ses=%u", 622 audit_log_format(*ab, "pid=%d uid=%u auid=%u ses=%u",
636 pid, uid, auid, ses); 623 task_tgid_vnr(current),
624 from_kuid(&init_user_ns, current_uid()),
625 from_kuid(&init_user_ns, auid), ses);
637 if (sid) { 626 if (sid) {
638 rc = security_secid_to_secctx(sid, &ctx, &len); 627 rc = security_secid_to_secctx(sid, &ctx, &len);
639 if (rc) 628 if (rc)
@@ -649,13 +638,13 @@ static int audit_log_common_recv_msg(struct audit_buffer **ab, u16 msg_type,
649 638
650static int audit_receive_msg(struct sk_buff *skb, struct nlmsghdr *nlh) 639static int audit_receive_msg(struct sk_buff *skb, struct nlmsghdr *nlh)
651{ 640{
652 u32 uid, pid, seq, sid; 641 u32 seq, sid;
653 void *data; 642 void *data;
654 struct audit_status *status_get, status_set; 643 struct audit_status *status_get, status_set;
655 int err; 644 int err;
656 struct audit_buffer *ab; 645 struct audit_buffer *ab;
657 u16 msg_type = nlh->nlmsg_type; 646 u16 msg_type = nlh->nlmsg_type;
658 uid_t loginuid; /* loginuid of sender */ 647 kuid_t loginuid; /* loginuid of sender */
659 u32 sessionid; 648 u32 sessionid;
660 struct audit_sig_info *sig_data; 649 struct audit_sig_info *sig_data;
661 char *ctx = NULL; 650 char *ctx = NULL;
@@ -675,8 +664,6 @@ static int audit_receive_msg(struct sk_buff *skb, struct nlmsghdr *nlh)
675 return err; 664 return err;
676 } 665 }
677 666
678 pid = NETLINK_CREDS(skb)->pid;
679 uid = NETLINK_CREDS(skb)->uid;
680 loginuid = audit_get_loginuid(current); 667 loginuid = audit_get_loginuid(current);
681 sessionid = audit_get_sessionid(current); 668 sessionid = audit_get_sessionid(current);
682 security_task_getsecid(current, &sid); 669 security_task_getsecid(current, &sid);
@@ -692,7 +679,7 @@ static int audit_receive_msg(struct sk_buff *skb, struct nlmsghdr *nlh)
692 status_set.backlog_limit = audit_backlog_limit; 679 status_set.backlog_limit = audit_backlog_limit;
693 status_set.lost = atomic_read(&audit_lost); 680 status_set.lost = atomic_read(&audit_lost);
694 status_set.backlog = skb_queue_len(&audit_skb_queue); 681 status_set.backlog = skb_queue_len(&audit_skb_queue);
695 audit_send_reply(NETLINK_CB(skb).pid, seq, AUDIT_GET, 0, 0, 682 audit_send_reply(NETLINK_CB(skb).portid, seq, AUDIT_GET, 0, 0,
696 &status_set, sizeof(status_set)); 683 &status_set, sizeof(status_set));
697 break; 684 break;
698 case AUDIT_SET: 685 case AUDIT_SET:
@@ -720,7 +707,7 @@ static int audit_receive_msg(struct sk_buff *skb, struct nlmsghdr *nlh)
720 sessionid, sid, 1); 707 sessionid, sid, 1);
721 708
722 audit_pid = new_pid; 709 audit_pid = new_pid;
723 audit_nlk_pid = NETLINK_CB(skb).pid; 710 audit_nlk_portid = NETLINK_CB(skb).portid;
724 } 711 }
725 if (status_get->mask & AUDIT_STATUS_RATE_LIMIT) { 712 if (status_get->mask & AUDIT_STATUS_RATE_LIMIT) {
726 err = audit_set_rate_limit(status_get->rate_limit, 713 err = audit_set_rate_limit(status_get->rate_limit,
@@ -738,16 +725,16 @@ static int audit_receive_msg(struct sk_buff *skb, struct nlmsghdr *nlh)
738 if (!audit_enabled && msg_type != AUDIT_USER_AVC) 725 if (!audit_enabled && msg_type != AUDIT_USER_AVC)
739 return 0; 726 return 0;
740 727
741 err = audit_filter_user(&NETLINK_CB(skb)); 728 err = audit_filter_user();
742 if (err == 1) { 729 if (err == 1) {
743 err = 0; 730 err = 0;
744 if (msg_type == AUDIT_USER_TTY) { 731 if (msg_type == AUDIT_USER_TTY) {
745 err = audit_prepare_user_tty(pid, loginuid, 732 err = tty_audit_push_task(current, loginuid,
746 sessionid); 733 sessionid);
747 if (err) 734 if (err)
748 break; 735 break;
749 } 736 }
750 audit_log_common_recv_msg(&ab, msg_type, pid, uid, 737 audit_log_common_recv_msg(&ab, msg_type,
751 loginuid, sessionid, sid); 738 loginuid, sessionid, sid);
752 739
753 if (msg_type != AUDIT_USER_TTY) 740 if (msg_type != AUDIT_USER_TTY)
@@ -763,7 +750,7 @@ static int audit_receive_msg(struct sk_buff *skb, struct nlmsghdr *nlh)
763 size--; 750 size--;
764 audit_log_n_untrustedstring(ab, data, size); 751 audit_log_n_untrustedstring(ab, data, size);
765 } 752 }
766 audit_set_pid(ab, pid); 753 audit_set_pid(ab, NETLINK_CB(skb).portid);
767 audit_log_end(ab); 754 audit_log_end(ab);
768 } 755 }
769 break; 756 break;
@@ -772,8 +759,8 @@ static int audit_receive_msg(struct sk_buff *skb, struct nlmsghdr *nlh)
772 if (nlmsg_len(nlh) < sizeof(struct audit_rule)) 759 if (nlmsg_len(nlh) < sizeof(struct audit_rule))
773 return -EINVAL; 760 return -EINVAL;
774 if (audit_enabled == AUDIT_LOCKED) { 761 if (audit_enabled == AUDIT_LOCKED) {
775 audit_log_common_recv_msg(&ab, AUDIT_CONFIG_CHANGE, pid, 762 audit_log_common_recv_msg(&ab, AUDIT_CONFIG_CHANGE,
776 uid, loginuid, sessionid, sid); 763 loginuid, sessionid, sid);
777 764
778 audit_log_format(ab, " audit_enabled=%d res=0", 765 audit_log_format(ab, " audit_enabled=%d res=0",
779 audit_enabled); 766 audit_enabled);
@@ -782,8 +769,8 @@ static int audit_receive_msg(struct sk_buff *skb, struct nlmsghdr *nlh)
782 } 769 }
783 /* fallthrough */ 770 /* fallthrough */
784 case AUDIT_LIST: 771 case AUDIT_LIST:
785 err = audit_receive_filter(msg_type, NETLINK_CB(skb).pid, 772 err = audit_receive_filter(msg_type, NETLINK_CB(skb).portid,
786 uid, seq, data, nlmsg_len(nlh), 773 seq, data, nlmsg_len(nlh),
787 loginuid, sessionid, sid); 774 loginuid, sessionid, sid);
788 break; 775 break;
789 case AUDIT_ADD_RULE: 776 case AUDIT_ADD_RULE:
@@ -791,8 +778,8 @@ static int audit_receive_msg(struct sk_buff *skb, struct nlmsghdr *nlh)
791 if (nlmsg_len(nlh) < sizeof(struct audit_rule_data)) 778 if (nlmsg_len(nlh) < sizeof(struct audit_rule_data))
792 return -EINVAL; 779 return -EINVAL;
793 if (audit_enabled == AUDIT_LOCKED) { 780 if (audit_enabled == AUDIT_LOCKED) {
794 audit_log_common_recv_msg(&ab, AUDIT_CONFIG_CHANGE, pid, 781 audit_log_common_recv_msg(&ab, AUDIT_CONFIG_CHANGE,
795 uid, loginuid, sessionid, sid); 782 loginuid, sessionid, sid);
796 783
797 audit_log_format(ab, " audit_enabled=%d res=0", 784 audit_log_format(ab, " audit_enabled=%d res=0",
798 audit_enabled); 785 audit_enabled);
@@ -801,15 +788,15 @@ static int audit_receive_msg(struct sk_buff *skb, struct nlmsghdr *nlh)
801 } 788 }
802 /* fallthrough */ 789 /* fallthrough */
803 case AUDIT_LIST_RULES: 790 case AUDIT_LIST_RULES:
804 err = audit_receive_filter(msg_type, NETLINK_CB(skb).pid, 791 err = audit_receive_filter(msg_type, NETLINK_CB(skb).portid,
805 uid, seq, data, nlmsg_len(nlh), 792 seq, data, nlmsg_len(nlh),
806 loginuid, sessionid, sid); 793 loginuid, sessionid, sid);
807 break; 794 break;
808 case AUDIT_TRIM: 795 case AUDIT_TRIM:
809 audit_trim_trees(); 796 audit_trim_trees();
810 797
811 audit_log_common_recv_msg(&ab, AUDIT_CONFIG_CHANGE, pid, 798 audit_log_common_recv_msg(&ab, AUDIT_CONFIG_CHANGE,
812 uid, loginuid, sessionid, sid); 799 loginuid, sessionid, sid);
813 800
814 audit_log_format(ab, " op=trim res=1"); 801 audit_log_format(ab, " op=trim res=1");
815 audit_log_end(ab); 802 audit_log_end(ab);
@@ -840,8 +827,8 @@ static int audit_receive_msg(struct sk_buff *skb, struct nlmsghdr *nlh)
840 /* OK, here comes... */ 827 /* OK, here comes... */
841 err = audit_tag_tree(old, new); 828 err = audit_tag_tree(old, new);
842 829
843 audit_log_common_recv_msg(&ab, AUDIT_CONFIG_CHANGE, pid, 830 audit_log_common_recv_msg(&ab, AUDIT_CONFIG_CHANGE,
844 uid, loginuid, sessionid, sid); 831 loginuid, sessionid, sid);
845 832
846 audit_log_format(ab, " op=make_equiv old="); 833 audit_log_format(ab, " op=make_equiv old=");
847 audit_log_untrustedstring(ab, old); 834 audit_log_untrustedstring(ab, old);
@@ -866,53 +853,41 @@ static int audit_receive_msg(struct sk_buff *skb, struct nlmsghdr *nlh)
866 security_release_secctx(ctx, len); 853 security_release_secctx(ctx, len);
867 return -ENOMEM; 854 return -ENOMEM;
868 } 855 }
869 sig_data->uid = audit_sig_uid; 856 sig_data->uid = from_kuid(&init_user_ns, audit_sig_uid);
870 sig_data->pid = audit_sig_pid; 857 sig_data->pid = audit_sig_pid;
871 if (audit_sig_sid) { 858 if (audit_sig_sid) {
872 memcpy(sig_data->ctx, ctx, len); 859 memcpy(sig_data->ctx, ctx, len);
873 security_release_secctx(ctx, len); 860 security_release_secctx(ctx, len);
874 } 861 }
875 audit_send_reply(NETLINK_CB(skb).pid, seq, AUDIT_SIGNAL_INFO, 862 audit_send_reply(NETLINK_CB(skb).portid, seq, AUDIT_SIGNAL_INFO,
876 0, 0, sig_data, sizeof(*sig_data) + len); 863 0, 0, sig_data, sizeof(*sig_data) + len);
877 kfree(sig_data); 864 kfree(sig_data);
878 break; 865 break;
879 case AUDIT_TTY_GET: { 866 case AUDIT_TTY_GET: {
880 struct audit_tty_status s; 867 struct audit_tty_status s;
881 struct task_struct *tsk; 868 struct task_struct *tsk = current;
882 unsigned long flags; 869
883 870 spin_lock_irq(&tsk->sighand->siglock);
884 rcu_read_lock(); 871 s.enabled = tsk->signal->audit_tty != 0;
885 tsk = find_task_by_vpid(pid); 872 spin_unlock_irq(&tsk->sighand->siglock);
886 if (tsk && lock_task_sighand(tsk, &flags)) { 873
887 s.enabled = tsk->signal->audit_tty != 0; 874 audit_send_reply(NETLINK_CB(skb).portid, seq,
888 unlock_task_sighand(tsk, &flags); 875 AUDIT_TTY_GET, 0, 0, &s, sizeof(s));
889 } else
890 err = -ESRCH;
891 rcu_read_unlock();
892
893 if (!err)
894 audit_send_reply(NETLINK_CB(skb).pid, seq,
895 AUDIT_TTY_GET, 0, 0, &s, sizeof(s));
896 break; 876 break;
897 } 877 }
898 case AUDIT_TTY_SET: { 878 case AUDIT_TTY_SET: {
899 struct audit_tty_status *s; 879 struct audit_tty_status *s;
900 struct task_struct *tsk; 880 struct task_struct *tsk = current;
901 unsigned long flags;
902 881
903 if (nlh->nlmsg_len < sizeof(struct audit_tty_status)) 882 if (nlh->nlmsg_len < sizeof(struct audit_tty_status))
904 return -EINVAL; 883 return -EINVAL;
905 s = data; 884 s = data;
906 if (s->enabled != 0 && s->enabled != 1) 885 if (s->enabled != 0 && s->enabled != 1)
907 return -EINVAL; 886 return -EINVAL;
908 rcu_read_lock(); 887
909 tsk = find_task_by_vpid(pid); 888 spin_lock_irq(&tsk->sighand->siglock);
910 if (tsk && lock_task_sighand(tsk, &flags)) { 889 tsk->signal->audit_tty = s->enabled != 0;
911 tsk->signal->audit_tty = s->enabled != 0; 890 spin_unlock_irq(&tsk->sighand->siglock);
912 unlock_task_sighand(tsk, &flags);
913 } else
914 err = -ESRCH;
915 rcu_read_unlock();
916 break; 891 break;
917 } 892 }
918 default: 893 default:
@@ -971,8 +946,7 @@ static int __init audit_init(void)
971 946
972 printk(KERN_INFO "audit: initializing netlink socket (%s)\n", 947 printk(KERN_INFO "audit: initializing netlink socket (%s)\n",
973 audit_default ? "enabled" : "disabled"); 948 audit_default ? "enabled" : "disabled");
974 audit_sock = netlink_kernel_create(&init_net, NETLINK_AUDIT, 949 audit_sock = netlink_kernel_create(&init_net, NETLINK_AUDIT, &cfg);
975 THIS_MODULE, &cfg);
976 if (!audit_sock) 950 if (!audit_sock)
977 audit_panic("cannot initialize netlink socket"); 951 audit_panic("cannot initialize netlink socket");
978 else 952 else
@@ -1466,6 +1440,8 @@ void audit_log_link_denied(const char *operation, struct path *link)
1466 1440
1467 ab = audit_log_start(current->audit_context, GFP_KERNEL, 1441 ab = audit_log_start(current->audit_context, GFP_KERNEL,
1468 AUDIT_ANOM_LINK); 1442 AUDIT_ANOM_LINK);
1443 if (!ab)
1444 return;
1469 audit_log_format(ab, "op=%s action=denied", operation); 1445 audit_log_format(ab, "op=%s action=denied", operation);
1470 audit_log_format(ab, " pid=%d comm=", current->pid); 1446 audit_log_format(ab, " pid=%d comm=", current->pid);
1471 audit_log_untrustedstring(ab, current->comm); 1447 audit_log_untrustedstring(ab, current->comm);
diff --git a/kernel/audit.h b/kernel/audit.h
index 816766803371..d51cba868e1b 100644
--- a/kernel/audit.h
+++ b/kernel/audit.h
@@ -74,10 +74,15 @@ static inline int audit_hash_ino(u32 ino)
74 return (ino & (AUDIT_INODE_BUCKETS-1)); 74 return (ino & (AUDIT_INODE_BUCKETS-1));
75} 75}
76 76
77/* Indicates that audit should log the full pathname. */
78#define AUDIT_NAME_FULL -1
79
77extern int audit_match_class(int class, unsigned syscall); 80extern int audit_match_class(int class, unsigned syscall);
78extern int audit_comparator(const u32 left, const u32 op, const u32 right); 81extern int audit_comparator(const u32 left, const u32 op, const u32 right);
79extern int audit_compare_dname_path(const char *dname, const char *path, 82extern int audit_uid_comparator(kuid_t left, u32 op, kuid_t right);
80 int *dirlen); 83extern int audit_gid_comparator(kgid_t left, u32 op, kgid_t right);
84extern int parent_len(const char *path);
85extern int audit_compare_dname_path(const char *dname, const char *path, int plen);
81extern struct sk_buff * audit_make_reply(int pid, int seq, int type, 86extern struct sk_buff * audit_make_reply(int pid, int seq, int type,
82 int done, int multi, 87 int done, int multi,
83 const void *payload, int size); 88 const void *payload, int size);
@@ -144,7 +149,7 @@ extern void audit_kill_trees(struct list_head *);
144extern char *audit_unpack_string(void **, size_t *, size_t); 149extern char *audit_unpack_string(void **, size_t *, size_t);
145 150
146extern pid_t audit_sig_pid; 151extern pid_t audit_sig_pid;
147extern uid_t audit_sig_uid; 152extern kuid_t audit_sig_uid;
148extern u32 audit_sig_sid; 153extern u32 audit_sig_sid;
149 154
150#ifdef CONFIG_AUDITSYSCALL 155#ifdef CONFIG_AUDITSYSCALL
diff --git a/kernel/audit_watch.c b/kernel/audit_watch.c
index 3823281401b5..9a9ae6e3d290 100644
--- a/kernel/audit_watch.c
+++ b/kernel/audit_watch.c
@@ -241,7 +241,7 @@ static void audit_watch_log_rule_change(struct audit_krule *r, struct audit_watc
241 struct audit_buffer *ab; 241 struct audit_buffer *ab;
242 ab = audit_log_start(NULL, GFP_NOFS, AUDIT_CONFIG_CHANGE); 242 ab = audit_log_start(NULL, GFP_NOFS, AUDIT_CONFIG_CHANGE);
243 audit_log_format(ab, "auid=%u ses=%u op=", 243 audit_log_format(ab, "auid=%u ses=%u op=",
244 audit_get_loginuid(current), 244 from_kuid(&init_user_ns, audit_get_loginuid(current)),
245 audit_get_sessionid(current)); 245 audit_get_sessionid(current));
246 audit_log_string(ab, op); 246 audit_log_string(ab, op);
247 audit_log_format(ab, " path="); 247 audit_log_format(ab, " path=");
@@ -265,7 +265,8 @@ static void audit_update_watch(struct audit_parent *parent,
265 /* Run all of the watches on this parent looking for the one that 265 /* Run all of the watches on this parent looking for the one that
266 * matches the given dname */ 266 * matches the given dname */
267 list_for_each_entry_safe(owatch, nextw, &parent->watches, wlist) { 267 list_for_each_entry_safe(owatch, nextw, &parent->watches, wlist) {
268 if (audit_compare_dname_path(dname, owatch->path, NULL)) 268 if (audit_compare_dname_path(dname, owatch->path,
269 AUDIT_NAME_FULL))
269 continue; 270 continue;
270 271
271 /* If the update involves invalidating rules, do the inode-based 272 /* If the update involves invalidating rules, do the inode-based
diff --git a/kernel/auditfilter.c b/kernel/auditfilter.c
index a6c3f1abd206..7f19f23d38a3 100644
--- a/kernel/auditfilter.c
+++ b/kernel/auditfilter.c
@@ -342,6 +342,8 @@ static struct audit_entry *audit_rule_to_entry(struct audit_rule *rule)
342 342
343 f->type = rule->fields[i] & ~(AUDIT_NEGATE|AUDIT_OPERATORS); 343 f->type = rule->fields[i] & ~(AUDIT_NEGATE|AUDIT_OPERATORS);
344 f->val = rule->values[i]; 344 f->val = rule->values[i];
345 f->uid = INVALID_UID;
346 f->gid = INVALID_GID;
345 347
346 err = -EINVAL; 348 err = -EINVAL;
347 if (f->op == Audit_bad) 349 if (f->op == Audit_bad)
@@ -350,16 +352,32 @@ static struct audit_entry *audit_rule_to_entry(struct audit_rule *rule)
350 switch(f->type) { 352 switch(f->type) {
351 default: 353 default:
352 goto exit_free; 354 goto exit_free;
353 case AUDIT_PID:
354 case AUDIT_UID: 355 case AUDIT_UID:
355 case AUDIT_EUID: 356 case AUDIT_EUID:
356 case AUDIT_SUID: 357 case AUDIT_SUID:
357 case AUDIT_FSUID: 358 case AUDIT_FSUID:
359 case AUDIT_LOGINUID:
360 /* bit ops not implemented for uid comparisons */
361 if (f->op == Audit_bitmask || f->op == Audit_bittest)
362 goto exit_free;
363
364 f->uid = make_kuid(current_user_ns(), f->val);
365 if (!uid_valid(f->uid))
366 goto exit_free;
367 break;
358 case AUDIT_GID: 368 case AUDIT_GID:
359 case AUDIT_EGID: 369 case AUDIT_EGID:
360 case AUDIT_SGID: 370 case AUDIT_SGID:
361 case AUDIT_FSGID: 371 case AUDIT_FSGID:
362 case AUDIT_LOGINUID: 372 /* bit ops not implemented for gid comparisons */
373 if (f->op == Audit_bitmask || f->op == Audit_bittest)
374 goto exit_free;
375
376 f->gid = make_kgid(current_user_ns(), f->val);
377 if (!gid_valid(f->gid))
378 goto exit_free;
379 break;
380 case AUDIT_PID:
363 case AUDIT_PERS: 381 case AUDIT_PERS:
364 case AUDIT_MSGTYPE: 382 case AUDIT_MSGTYPE:
365 case AUDIT_PPID: 383 case AUDIT_PPID:
@@ -437,19 +455,39 @@ static struct audit_entry *audit_data_to_entry(struct audit_rule_data *data,
437 455
438 f->type = data->fields[i]; 456 f->type = data->fields[i];
439 f->val = data->values[i]; 457 f->val = data->values[i];
458 f->uid = INVALID_UID;
459 f->gid = INVALID_GID;
440 f->lsm_str = NULL; 460 f->lsm_str = NULL;
441 f->lsm_rule = NULL; 461 f->lsm_rule = NULL;
442 switch(f->type) { 462 switch(f->type) {
443 case AUDIT_PID:
444 case AUDIT_UID: 463 case AUDIT_UID:
445 case AUDIT_EUID: 464 case AUDIT_EUID:
446 case AUDIT_SUID: 465 case AUDIT_SUID:
447 case AUDIT_FSUID: 466 case AUDIT_FSUID:
467 case AUDIT_LOGINUID:
468 case AUDIT_OBJ_UID:
469 /* bit ops not implemented for uid comparisons */
470 if (f->op == Audit_bitmask || f->op == Audit_bittest)
471 goto exit_free;
472
473 f->uid = make_kuid(current_user_ns(), f->val);
474 if (!uid_valid(f->uid))
475 goto exit_free;
476 break;
448 case AUDIT_GID: 477 case AUDIT_GID:
449 case AUDIT_EGID: 478 case AUDIT_EGID:
450 case AUDIT_SGID: 479 case AUDIT_SGID:
451 case AUDIT_FSGID: 480 case AUDIT_FSGID:
452 case AUDIT_LOGINUID: 481 case AUDIT_OBJ_GID:
482 /* bit ops not implemented for gid comparisons */
483 if (f->op == Audit_bitmask || f->op == Audit_bittest)
484 goto exit_free;
485
486 f->gid = make_kgid(current_user_ns(), f->val);
487 if (!gid_valid(f->gid))
488 goto exit_free;
489 break;
490 case AUDIT_PID:
453 case AUDIT_PERS: 491 case AUDIT_PERS:
454 case AUDIT_MSGTYPE: 492 case AUDIT_MSGTYPE:
455 case AUDIT_PPID: 493 case AUDIT_PPID:
@@ -461,8 +499,6 @@ static struct audit_entry *audit_data_to_entry(struct audit_rule_data *data,
461 case AUDIT_ARG1: 499 case AUDIT_ARG1:
462 case AUDIT_ARG2: 500 case AUDIT_ARG2:
463 case AUDIT_ARG3: 501 case AUDIT_ARG3:
464 case AUDIT_OBJ_UID:
465 case AUDIT_OBJ_GID:
466 break; 502 break;
467 case AUDIT_ARCH: 503 case AUDIT_ARCH:
468 entry->rule.arch_f = f; 504 entry->rule.arch_f = f;
@@ -707,6 +743,23 @@ static int audit_compare_rule(struct audit_krule *a, struct audit_krule *b)
707 if (strcmp(a->filterkey, b->filterkey)) 743 if (strcmp(a->filterkey, b->filterkey))
708 return 1; 744 return 1;
709 break; 745 break;
746 case AUDIT_UID:
747 case AUDIT_EUID:
748 case AUDIT_SUID:
749 case AUDIT_FSUID:
750 case AUDIT_LOGINUID:
751 case AUDIT_OBJ_UID:
752 if (!uid_eq(a->fields[i].uid, b->fields[i].uid))
753 return 1;
754 break;
755 case AUDIT_GID:
756 case AUDIT_EGID:
757 case AUDIT_SGID:
758 case AUDIT_FSGID:
759 case AUDIT_OBJ_GID:
760 if (!gid_eq(a->fields[i].gid, b->fields[i].gid))
761 return 1;
762 break;
710 default: 763 default:
711 if (a->fields[i].val != b->fields[i].val) 764 if (a->fields[i].val != b->fields[i].val)
712 return 1; 765 return 1;
@@ -1056,7 +1109,7 @@ static void audit_list_rules(int pid, int seq, struct sk_buff_head *q)
1056} 1109}
1057 1110
1058/* Log rule additions and removals */ 1111/* Log rule additions and removals */
1059static void audit_log_rule_change(uid_t loginuid, u32 sessionid, u32 sid, 1112static void audit_log_rule_change(kuid_t loginuid, u32 sessionid, u32 sid,
1060 char *action, struct audit_krule *rule, 1113 char *action, struct audit_krule *rule,
1061 int res) 1114 int res)
1062{ 1115{
@@ -1068,7 +1121,8 @@ static void audit_log_rule_change(uid_t loginuid, u32 sessionid, u32 sid,
1068 ab = audit_log_start(NULL, GFP_KERNEL, AUDIT_CONFIG_CHANGE); 1121 ab = audit_log_start(NULL, GFP_KERNEL, AUDIT_CONFIG_CHANGE);
1069 if (!ab) 1122 if (!ab)
1070 return; 1123 return;
1071 audit_log_format(ab, "auid=%u ses=%u", loginuid, sessionid); 1124 audit_log_format(ab, "auid=%u ses=%u",
1125 from_kuid(&init_user_ns, loginuid), sessionid);
1072 if (sid) { 1126 if (sid) {
1073 char *ctx = NULL; 1127 char *ctx = NULL;
1074 u32 len; 1128 u32 len;
@@ -1098,8 +1152,8 @@ static void audit_log_rule_change(uid_t loginuid, u32 sessionid, u32 sid,
1098 * @sessionid: sessionid for netlink audit message 1152 * @sessionid: sessionid for netlink audit message
1099 * @sid: SE Linux Security ID of sender 1153 * @sid: SE Linux Security ID of sender
1100 */ 1154 */
1101int audit_receive_filter(int type, int pid, int uid, int seq, void *data, 1155int audit_receive_filter(int type, int pid, int seq, void *data,
1102 size_t datasz, uid_t loginuid, u32 sessionid, u32 sid) 1156 size_t datasz, kuid_t loginuid, u32 sessionid, u32 sid)
1103{ 1157{
1104 struct task_struct *tsk; 1158 struct task_struct *tsk;
1105 struct audit_netlink_list *dest; 1159 struct audit_netlink_list *dest;
@@ -1198,46 +1252,110 @@ int audit_comparator(u32 left, u32 op, u32 right)
1198 } 1252 }
1199} 1253}
1200 1254
1201/* Compare given dentry name with last component in given path, 1255int audit_uid_comparator(kuid_t left, u32 op, kuid_t right)
1202 * return of 0 indicates a match. */
1203int audit_compare_dname_path(const char *dname, const char *path,
1204 int *dirlen)
1205{ 1256{
1206 int dlen, plen; 1257 switch (op) {
1207 const char *p; 1258 case Audit_equal:
1259 return uid_eq(left, right);
1260 case Audit_not_equal:
1261 return !uid_eq(left, right);
1262 case Audit_lt:
1263 return uid_lt(left, right);
1264 case Audit_le:
1265 return uid_lte(left, right);
1266 case Audit_gt:
1267 return uid_gt(left, right);
1268 case Audit_ge:
1269 return uid_gte(left, right);
1270 case Audit_bitmask:
1271 case Audit_bittest:
1272 default:
1273 BUG();
1274 return 0;
1275 }
1276}
1208 1277
1209 if (!dname || !path) 1278int audit_gid_comparator(kgid_t left, u32 op, kgid_t right)
1210 return 1; 1279{
1280 switch (op) {
1281 case Audit_equal:
1282 return gid_eq(left, right);
1283 case Audit_not_equal:
1284 return !gid_eq(left, right);
1285 case Audit_lt:
1286 return gid_lt(left, right);
1287 case Audit_le:
1288 return gid_lte(left, right);
1289 case Audit_gt:
1290 return gid_gt(left, right);
1291 case Audit_ge:
1292 return gid_gte(left, right);
1293 case Audit_bitmask:
1294 case Audit_bittest:
1295 default:
1296 BUG();
1297 return 0;
1298 }
1299}
1300
1301/**
1302 * parent_len - find the length of the parent portion of a pathname
1303 * @path: pathname of which to determine length
1304 */
1305int parent_len(const char *path)
1306{
1307 int plen;
1308 const char *p;
1211 1309
1212 dlen = strlen(dname);
1213 plen = strlen(path); 1310 plen = strlen(path);
1214 if (plen < dlen) 1311
1215 return 1; 1312 if (plen == 0)
1313 return plen;
1216 1314
1217 /* disregard trailing slashes */ 1315 /* disregard trailing slashes */
1218 p = path + plen - 1; 1316 p = path + plen - 1;
1219 while ((*p == '/') && (p > path)) 1317 while ((*p == '/') && (p > path))
1220 p--; 1318 p--;
1221 1319
1222 /* find last path component */ 1320 /* walk backward until we find the next slash or hit beginning */
1223 p = p - dlen + 1; 1321 while ((*p != '/') && (p > path))
1224 if (p < path) 1322 p--;
1323
1324 /* did we find a slash? Then increment to include it in path */
1325 if (*p == '/')
1326 p++;
1327
1328 return p - path;
1329}
1330
1331/**
1332 * audit_compare_dname_path - compare given dentry name with last component in
1333 * given path. Return of 0 indicates a match.
1334 * @dname: dentry name that we're comparing
1335 * @path: full pathname that we're comparing
1336 * @parentlen: length of the parent if known. Passing in AUDIT_NAME_FULL
1337 * here indicates that we must compute this value.
1338 */
1339int audit_compare_dname_path(const char *dname, const char *path, int parentlen)
1340{
1341 int dlen, pathlen;
1342 const char *p;
1343
1344 dlen = strlen(dname);
1345 pathlen = strlen(path);
1346 if (pathlen < dlen)
1225 return 1; 1347 return 1;
1226 else if (p > path) {
1227 if (*--p != '/')
1228 return 1;
1229 else
1230 p++;
1231 }
1232 1348
1233 /* return length of path's directory component */ 1349 parentlen = parentlen == AUDIT_NAME_FULL ? parent_len(path) : parentlen;
1234 if (dirlen) 1350 if (pathlen - parentlen != dlen)
1235 *dirlen = p - path; 1351 return 1;
1352
1353 p = path + parentlen;
1354
1236 return strncmp(p, dname, dlen); 1355 return strncmp(p, dname, dlen);
1237} 1356}
1238 1357
1239static int audit_filter_user_rules(struct netlink_skb_parms *cb, 1358static int audit_filter_user_rules(struct audit_krule *rule,
1240 struct audit_krule *rule,
1241 enum audit_state *state) 1359 enum audit_state *state)
1242{ 1360{
1243 int i; 1361 int i;
@@ -1249,17 +1367,17 @@ static int audit_filter_user_rules(struct netlink_skb_parms *cb,
1249 1367
1250 switch (f->type) { 1368 switch (f->type) {
1251 case AUDIT_PID: 1369 case AUDIT_PID:
1252 result = audit_comparator(cb->creds.pid, f->op, f->val); 1370 result = audit_comparator(task_pid_vnr(current), f->op, f->val);
1253 break; 1371 break;
1254 case AUDIT_UID: 1372 case AUDIT_UID:
1255 result = audit_comparator(cb->creds.uid, f->op, f->val); 1373 result = audit_uid_comparator(current_uid(), f->op, f->uid);
1256 break; 1374 break;
1257 case AUDIT_GID: 1375 case AUDIT_GID:
1258 result = audit_comparator(cb->creds.gid, f->op, f->val); 1376 result = audit_gid_comparator(current_gid(), f->op, f->gid);
1259 break; 1377 break;
1260 case AUDIT_LOGINUID: 1378 case AUDIT_LOGINUID:
1261 result = audit_comparator(audit_get_loginuid(current), 1379 result = audit_uid_comparator(audit_get_loginuid(current),
1262 f->op, f->val); 1380 f->op, f->uid);
1263 break; 1381 break;
1264 case AUDIT_SUBJ_USER: 1382 case AUDIT_SUBJ_USER:
1265 case AUDIT_SUBJ_ROLE: 1383 case AUDIT_SUBJ_ROLE:
@@ -1287,7 +1405,7 @@ static int audit_filter_user_rules(struct netlink_skb_parms *cb,
1287 return 1; 1405 return 1;
1288} 1406}
1289 1407
1290int audit_filter_user(struct netlink_skb_parms *cb) 1408int audit_filter_user(void)
1291{ 1409{
1292 enum audit_state state = AUDIT_DISABLED; 1410 enum audit_state state = AUDIT_DISABLED;
1293 struct audit_entry *e; 1411 struct audit_entry *e;
@@ -1295,7 +1413,7 @@ int audit_filter_user(struct netlink_skb_parms *cb)
1295 1413
1296 rcu_read_lock(); 1414 rcu_read_lock();
1297 list_for_each_entry_rcu(e, &audit_filter_list[AUDIT_FILTER_USER], list) { 1415 list_for_each_entry_rcu(e, &audit_filter_list[AUDIT_FILTER_USER], list) {
1298 if (audit_filter_user_rules(cb, &e->rule, &state)) { 1416 if (audit_filter_user_rules(&e->rule, &state)) {
1299 if (state == AUDIT_DISABLED) 1417 if (state == AUDIT_DISABLED)
1300 ret = 0; 1418 ret = 0;
1301 break; 1419 break;
diff --git a/kernel/auditsc.c b/kernel/auditsc.c
index 4b96415527b8..2f186ed80c40 100644
--- a/kernel/auditsc.c
+++ b/kernel/auditsc.c
@@ -81,9 +81,6 @@
81 * a name dynamically and also add those to the list anchored by names_list. */ 81 * a name dynamically and also add those to the list anchored by names_list. */
82#define AUDIT_NAMES 5 82#define AUDIT_NAMES 5
83 83
84/* Indicates that audit should log the full pathname. */
85#define AUDIT_NAME_FULL -1
86
87/* no execve audit message should be longer than this (userspace limits) */ 84/* no execve audit message should be longer than this (userspace limits) */
88#define MAX_EXECVE_AUDIT_LEN 7500 85#define MAX_EXECVE_AUDIT_LEN 7500
89 86
@@ -106,27 +103,29 @@ struct audit_cap_data {
106 * we don't let putname() free it (instead we free all of the saved 103 * we don't let putname() free it (instead we free all of the saved
107 * pointers at syscall exit time). 104 * pointers at syscall exit time).
108 * 105 *
109 * Further, in fs/namei.c:path_lookup() we store the inode and device. */ 106 * Further, in fs/namei.c:path_lookup() we store the inode and device.
107 */
110struct audit_names { 108struct audit_names {
111 struct list_head list; /* audit_context->names_list */ 109 struct list_head list; /* audit_context->names_list */
112 const char *name; 110 struct filename *name;
113 unsigned long ino; 111 unsigned long ino;
114 dev_t dev; 112 dev_t dev;
115 umode_t mode; 113 umode_t mode;
116 uid_t uid; 114 kuid_t uid;
117 gid_t gid; 115 kgid_t gid;
118 dev_t rdev; 116 dev_t rdev;
119 u32 osid; 117 u32 osid;
120 struct audit_cap_data fcap; 118 struct audit_cap_data fcap;
121 unsigned int fcap_ver; 119 unsigned int fcap_ver;
122 int name_len; /* number of name's characters to log */ 120 int name_len; /* number of name's characters to log */
123 bool name_put; /* call __putname() for this name */ 121 unsigned char type; /* record type */
122 bool name_put; /* call __putname() for this name */
124 /* 123 /*
125 * This was an allocated audit_names and not from the array of 124 * This was an allocated audit_names and not from the array of
126 * names allocated in the task audit context. Thus this name 125 * names allocated in the task audit context. Thus this name
127 * should be freed on syscall exit 126 * should be freed on syscall exit
128 */ 127 */
129 bool should_free; 128 bool should_free;
130}; 129};
131 130
132struct audit_aux_data { 131struct audit_aux_data {
@@ -149,8 +148,8 @@ struct audit_aux_data_execve {
149struct audit_aux_data_pids { 148struct audit_aux_data_pids {
150 struct audit_aux_data d; 149 struct audit_aux_data d;
151 pid_t target_pid[AUDIT_AUX_PIDS]; 150 pid_t target_pid[AUDIT_AUX_PIDS];
152 uid_t target_auid[AUDIT_AUX_PIDS]; 151 kuid_t target_auid[AUDIT_AUX_PIDS];
153 uid_t target_uid[AUDIT_AUX_PIDS]; 152 kuid_t target_uid[AUDIT_AUX_PIDS];
154 unsigned int target_sessionid[AUDIT_AUX_PIDS]; 153 unsigned int target_sessionid[AUDIT_AUX_PIDS];
155 u32 target_sid[AUDIT_AUX_PIDS]; 154 u32 target_sid[AUDIT_AUX_PIDS];
156 char target_comm[AUDIT_AUX_PIDS][TASK_COMM_LEN]; 155 char target_comm[AUDIT_AUX_PIDS][TASK_COMM_LEN];
@@ -208,14 +207,14 @@ struct audit_context {
208 size_t sockaddr_len; 207 size_t sockaddr_len;
209 /* Save things to print about task_struct */ 208 /* Save things to print about task_struct */
210 pid_t pid, ppid; 209 pid_t pid, ppid;
211 uid_t uid, euid, suid, fsuid; 210 kuid_t uid, euid, suid, fsuid;
212 gid_t gid, egid, sgid, fsgid; 211 kgid_t gid, egid, sgid, fsgid;
213 unsigned long personality; 212 unsigned long personality;
214 int arch; 213 int arch;
215 214
216 pid_t target_pid; 215 pid_t target_pid;
217 uid_t target_auid; 216 kuid_t target_auid;
218 uid_t target_uid; 217 kuid_t target_uid;
219 unsigned int target_sessionid; 218 unsigned int target_sessionid;
220 u32 target_sid; 219 u32 target_sid;
221 char target_comm[TASK_COMM_LEN]; 220 char target_comm[TASK_COMM_LEN];
@@ -231,8 +230,8 @@ struct audit_context {
231 long args[6]; 230 long args[6];
232 } socketcall; 231 } socketcall;
233 struct { 232 struct {
234 uid_t uid; 233 kuid_t uid;
235 gid_t gid; 234 kgid_t gid;
236 umode_t mode; 235 umode_t mode;
237 u32 osid; 236 u32 osid;
238 int has_perm; 237 int has_perm;
@@ -464,37 +463,47 @@ static int match_tree_refs(struct audit_context *ctx, struct audit_tree *tree)
464 return 0; 463 return 0;
465} 464}
466 465
467static int audit_compare_id(uid_t uid1, 466static int audit_compare_uid(kuid_t uid,
468 struct audit_names *name, 467 struct audit_names *name,
469 unsigned long name_offset, 468 struct audit_field *f,
470 struct audit_field *f, 469 struct audit_context *ctx)
471 struct audit_context *ctx)
472{ 470{
473 struct audit_names *n; 471 struct audit_names *n;
474 unsigned long addr;
475 uid_t uid2;
476 int rc; 472 int rc;
477 473
478 BUILD_BUG_ON(sizeof(uid_t) != sizeof(gid_t));
479
480 if (name) { 474 if (name) {
481 addr = (unsigned long)name; 475 rc = audit_uid_comparator(uid, f->op, name->uid);
482 addr += name_offset;
483
484 uid2 = *(uid_t *)addr;
485 rc = audit_comparator(uid1, f->op, uid2);
486 if (rc) 476 if (rc)
487 return rc; 477 return rc;
488 } 478 }
489 479
490 if (ctx) { 480 if (ctx) {
491 list_for_each_entry(n, &ctx->names_list, list) { 481 list_for_each_entry(n, &ctx->names_list, list) {
492 addr = (unsigned long)n; 482 rc = audit_uid_comparator(uid, f->op, n->uid);
493 addr += name_offset; 483 if (rc)
494 484 return rc;
495 uid2 = *(uid_t *)addr; 485 }
486 }
487 return 0;
488}
496 489
497 rc = audit_comparator(uid1, f->op, uid2); 490static int audit_compare_gid(kgid_t gid,
491 struct audit_names *name,
492 struct audit_field *f,
493 struct audit_context *ctx)
494{
495 struct audit_names *n;
496 int rc;
497
498 if (name) {
499 rc = audit_gid_comparator(gid, f->op, name->gid);
500 if (rc)
501 return rc;
502 }
503
504 if (ctx) {
505 list_for_each_entry(n, &ctx->names_list, list) {
506 rc = audit_gid_comparator(gid, f->op, n->gid);
498 if (rc) 507 if (rc)
499 return rc; 508 return rc;
500 } 509 }
@@ -511,80 +520,62 @@ static int audit_field_compare(struct task_struct *tsk,
511 switch (f->val) { 520 switch (f->val) {
512 /* process to file object comparisons */ 521 /* process to file object comparisons */
513 case AUDIT_COMPARE_UID_TO_OBJ_UID: 522 case AUDIT_COMPARE_UID_TO_OBJ_UID:
514 return audit_compare_id(cred->uid, 523 return audit_compare_uid(cred->uid, name, f, ctx);
515 name, offsetof(struct audit_names, uid),
516 f, ctx);
517 case AUDIT_COMPARE_GID_TO_OBJ_GID: 524 case AUDIT_COMPARE_GID_TO_OBJ_GID:
518 return audit_compare_id(cred->gid, 525 return audit_compare_gid(cred->gid, name, f, ctx);
519 name, offsetof(struct audit_names, gid),
520 f, ctx);
521 case AUDIT_COMPARE_EUID_TO_OBJ_UID: 526 case AUDIT_COMPARE_EUID_TO_OBJ_UID:
522 return audit_compare_id(cred->euid, 527 return audit_compare_uid(cred->euid, name, f, ctx);
523 name, offsetof(struct audit_names, uid),
524 f, ctx);
525 case AUDIT_COMPARE_EGID_TO_OBJ_GID: 528 case AUDIT_COMPARE_EGID_TO_OBJ_GID:
526 return audit_compare_id(cred->egid, 529 return audit_compare_gid(cred->egid, name, f, ctx);
527 name, offsetof(struct audit_names, gid),
528 f, ctx);
529 case AUDIT_COMPARE_AUID_TO_OBJ_UID: 530 case AUDIT_COMPARE_AUID_TO_OBJ_UID:
530 return audit_compare_id(tsk->loginuid, 531 return audit_compare_uid(tsk->loginuid, name, f, ctx);
531 name, offsetof(struct audit_names, uid),
532 f, ctx);
533 case AUDIT_COMPARE_SUID_TO_OBJ_UID: 532 case AUDIT_COMPARE_SUID_TO_OBJ_UID:
534 return audit_compare_id(cred->suid, 533 return audit_compare_uid(cred->suid, name, f, ctx);
535 name, offsetof(struct audit_names, uid),
536 f, ctx);
537 case AUDIT_COMPARE_SGID_TO_OBJ_GID: 534 case AUDIT_COMPARE_SGID_TO_OBJ_GID:
538 return audit_compare_id(cred->sgid, 535 return audit_compare_gid(cred->sgid, name, f, ctx);
539 name, offsetof(struct audit_names, gid),
540 f, ctx);
541 case AUDIT_COMPARE_FSUID_TO_OBJ_UID: 536 case AUDIT_COMPARE_FSUID_TO_OBJ_UID:
542 return audit_compare_id(cred->fsuid, 537 return audit_compare_uid(cred->fsuid, name, f, ctx);
543 name, offsetof(struct audit_names, uid),
544 f, ctx);
545 case AUDIT_COMPARE_FSGID_TO_OBJ_GID: 538 case AUDIT_COMPARE_FSGID_TO_OBJ_GID:
546 return audit_compare_id(cred->fsgid, 539 return audit_compare_gid(cred->fsgid, name, f, ctx);
547 name, offsetof(struct audit_names, gid),
548 f, ctx);
549 /* uid comparisons */ 540 /* uid comparisons */
550 case AUDIT_COMPARE_UID_TO_AUID: 541 case AUDIT_COMPARE_UID_TO_AUID:
551 return audit_comparator(cred->uid, f->op, tsk->loginuid); 542 return audit_uid_comparator(cred->uid, f->op, tsk->loginuid);
552 case AUDIT_COMPARE_UID_TO_EUID: 543 case AUDIT_COMPARE_UID_TO_EUID:
553 return audit_comparator(cred->uid, f->op, cred->euid); 544 return audit_uid_comparator(cred->uid, f->op, cred->euid);
554 case AUDIT_COMPARE_UID_TO_SUID: 545 case AUDIT_COMPARE_UID_TO_SUID:
555 return audit_comparator(cred->uid, f->op, cred->suid); 546 return audit_uid_comparator(cred->uid, f->op, cred->suid);
556 case AUDIT_COMPARE_UID_TO_FSUID: 547 case AUDIT_COMPARE_UID_TO_FSUID:
557 return audit_comparator(cred->uid, f->op, cred->fsuid); 548 return audit_uid_comparator(cred->uid, f->op, cred->fsuid);
558 /* auid comparisons */ 549 /* auid comparisons */
559 case AUDIT_COMPARE_AUID_TO_EUID: 550 case AUDIT_COMPARE_AUID_TO_EUID:
560 return audit_comparator(tsk->loginuid, f->op, cred->euid); 551 return audit_uid_comparator(tsk->loginuid, f->op, cred->euid);
561 case AUDIT_COMPARE_AUID_TO_SUID: 552 case AUDIT_COMPARE_AUID_TO_SUID:
562 return audit_comparator(tsk->loginuid, f->op, cred->suid); 553 return audit_uid_comparator(tsk->loginuid, f->op, cred->suid);
563 case AUDIT_COMPARE_AUID_TO_FSUID: 554 case AUDIT_COMPARE_AUID_TO_FSUID:
564 return audit_comparator(tsk->loginuid, f->op, cred->fsuid); 555 return audit_uid_comparator(tsk->loginuid, f->op, cred->fsuid);
565 /* euid comparisons */ 556 /* euid comparisons */
566 case AUDIT_COMPARE_EUID_TO_SUID: 557 case AUDIT_COMPARE_EUID_TO_SUID:
567 return audit_comparator(cred->euid, f->op, cred->suid); 558 return audit_uid_comparator(cred->euid, f->op, cred->suid);
568 case AUDIT_COMPARE_EUID_TO_FSUID: 559 case AUDIT_COMPARE_EUID_TO_FSUID:
569 return audit_comparator(cred->euid, f->op, cred->fsuid); 560 return audit_uid_comparator(cred->euid, f->op, cred->fsuid);
570 /* suid comparisons */ 561 /* suid comparisons */
571 case AUDIT_COMPARE_SUID_TO_FSUID: 562 case AUDIT_COMPARE_SUID_TO_FSUID:
572 return audit_comparator(cred->suid, f->op, cred->fsuid); 563 return audit_uid_comparator(cred->suid, f->op, cred->fsuid);
573 /* gid comparisons */ 564 /* gid comparisons */
574 case AUDIT_COMPARE_GID_TO_EGID: 565 case AUDIT_COMPARE_GID_TO_EGID:
575 return audit_comparator(cred->gid, f->op, cred->egid); 566 return audit_gid_comparator(cred->gid, f->op, cred->egid);
576 case AUDIT_COMPARE_GID_TO_SGID: 567 case AUDIT_COMPARE_GID_TO_SGID:
577 return audit_comparator(cred->gid, f->op, cred->sgid); 568 return audit_gid_comparator(cred->gid, f->op, cred->sgid);
578 case AUDIT_COMPARE_GID_TO_FSGID: 569 case AUDIT_COMPARE_GID_TO_FSGID:
579 return audit_comparator(cred->gid, f->op, cred->fsgid); 570 return audit_gid_comparator(cred->gid, f->op, cred->fsgid);
580 /* egid comparisons */ 571 /* egid comparisons */
581 case AUDIT_COMPARE_EGID_TO_SGID: 572 case AUDIT_COMPARE_EGID_TO_SGID:
582 return audit_comparator(cred->egid, f->op, cred->sgid); 573 return audit_gid_comparator(cred->egid, f->op, cred->sgid);
583 case AUDIT_COMPARE_EGID_TO_FSGID: 574 case AUDIT_COMPARE_EGID_TO_FSGID:
584 return audit_comparator(cred->egid, f->op, cred->fsgid); 575 return audit_gid_comparator(cred->egid, f->op, cred->fsgid);
585 /* sgid comparison */ 576 /* sgid comparison */
586 case AUDIT_COMPARE_SGID_TO_FSGID: 577 case AUDIT_COMPARE_SGID_TO_FSGID:
587 return audit_comparator(cred->sgid, f->op, cred->fsgid); 578 return audit_gid_comparator(cred->sgid, f->op, cred->fsgid);
588 default: 579 default:
589 WARN(1, "Missing AUDIT_COMPARE define. Report as a bug\n"); 580 WARN(1, "Missing AUDIT_COMPARE define. Report as a bug\n");
590 return 0; 581 return 0;
@@ -630,28 +621,28 @@ static int audit_filter_rules(struct task_struct *tsk,
630 } 621 }
631 break; 622 break;
632 case AUDIT_UID: 623 case AUDIT_UID:
633 result = audit_comparator(cred->uid, f->op, f->val); 624 result = audit_uid_comparator(cred->uid, f->op, f->uid);
634 break; 625 break;
635 case AUDIT_EUID: 626 case AUDIT_EUID:
636 result = audit_comparator(cred->euid, f->op, f->val); 627 result = audit_uid_comparator(cred->euid, f->op, f->uid);
637 break; 628 break;
638 case AUDIT_SUID: 629 case AUDIT_SUID:
639 result = audit_comparator(cred->suid, f->op, f->val); 630 result = audit_uid_comparator(cred->suid, f->op, f->uid);
640 break; 631 break;
641 case AUDIT_FSUID: 632 case AUDIT_FSUID:
642 result = audit_comparator(cred->fsuid, f->op, f->val); 633 result = audit_uid_comparator(cred->fsuid, f->op, f->uid);
643 break; 634 break;
644 case AUDIT_GID: 635 case AUDIT_GID:
645 result = audit_comparator(cred->gid, f->op, f->val); 636 result = audit_gid_comparator(cred->gid, f->op, f->gid);
646 break; 637 break;
647 case AUDIT_EGID: 638 case AUDIT_EGID:
648 result = audit_comparator(cred->egid, f->op, f->val); 639 result = audit_gid_comparator(cred->egid, f->op, f->gid);
649 break; 640 break;
650 case AUDIT_SGID: 641 case AUDIT_SGID:
651 result = audit_comparator(cred->sgid, f->op, f->val); 642 result = audit_gid_comparator(cred->sgid, f->op, f->gid);
652 break; 643 break;
653 case AUDIT_FSGID: 644 case AUDIT_FSGID:
654 result = audit_comparator(cred->fsgid, f->op, f->val); 645 result = audit_gid_comparator(cred->fsgid, f->op, f->gid);
655 break; 646 break;
656 case AUDIT_PERS: 647 case AUDIT_PERS:
657 result = audit_comparator(tsk->personality, f->op, f->val); 648 result = audit_comparator(tsk->personality, f->op, f->val);
@@ -717,10 +708,10 @@ static int audit_filter_rules(struct task_struct *tsk,
717 break; 708 break;
718 case AUDIT_OBJ_UID: 709 case AUDIT_OBJ_UID:
719 if (name) { 710 if (name) {
720 result = audit_comparator(name->uid, f->op, f->val); 711 result = audit_uid_comparator(name->uid, f->op, f->uid);
721 } else if (ctx) { 712 } else if (ctx) {
722 list_for_each_entry(n, &ctx->names_list, list) { 713 list_for_each_entry(n, &ctx->names_list, list) {
723 if (audit_comparator(n->uid, f->op, f->val)) { 714 if (audit_uid_comparator(n->uid, f->op, f->uid)) {
724 ++result; 715 ++result;
725 break; 716 break;
726 } 717 }
@@ -729,10 +720,10 @@ static int audit_filter_rules(struct task_struct *tsk,
729 break; 720 break;
730 case AUDIT_OBJ_GID: 721 case AUDIT_OBJ_GID:
731 if (name) { 722 if (name) {
732 result = audit_comparator(name->gid, f->op, f->val); 723 result = audit_gid_comparator(name->gid, f->op, f->gid);
733 } else if (ctx) { 724 } else if (ctx) {
734 list_for_each_entry(n, &ctx->names_list, list) { 725 list_for_each_entry(n, &ctx->names_list, list) {
735 if (audit_comparator(n->gid, f->op, f->val)) { 726 if (audit_gid_comparator(n->gid, f->op, f->gid)) {
736 ++result; 727 ++result;
737 break; 728 break;
738 } 729 }
@@ -750,7 +741,7 @@ static int audit_filter_rules(struct task_struct *tsk,
750 case AUDIT_LOGINUID: 741 case AUDIT_LOGINUID:
751 result = 0; 742 result = 0;
752 if (ctx) 743 if (ctx)
753 result = audit_comparator(tsk->loginuid, f->op, f->val); 744 result = audit_uid_comparator(tsk->loginuid, f->op, f->uid);
754 break; 745 break;
755 case AUDIT_SUBJ_USER: 746 case AUDIT_SUBJ_USER:
756 case AUDIT_SUBJ_ROLE: 747 case AUDIT_SUBJ_ROLE:
@@ -1006,7 +997,7 @@ static inline void audit_free_names(struct audit_context *context)
1006 context->ino_count); 997 context->ino_count);
1007 list_for_each_entry(n, &context->names_list, list) { 998 list_for_each_entry(n, &context->names_list, list) {
1008 printk(KERN_ERR "names[%d] = %p = %s\n", i, 999 printk(KERN_ERR "names[%d] = %p = %s\n", i,
1009 n->name, n->name ?: "(null)"); 1000 n->name, n->name->name ?: "(null)");
1010 } 1001 }
1011 dump_stack(); 1002 dump_stack();
1012 return; 1003 return;
@@ -1154,13 +1145,43 @@ error_path:
1154 1145
1155EXPORT_SYMBOL(audit_log_task_context); 1146EXPORT_SYMBOL(audit_log_task_context);
1156 1147
1157static void audit_log_task_info(struct audit_buffer *ab, struct task_struct *tsk) 1148void audit_log_task_info(struct audit_buffer *ab, struct task_struct *tsk)
1158{ 1149{
1150 const struct cred *cred;
1159 char name[sizeof(tsk->comm)]; 1151 char name[sizeof(tsk->comm)];
1160 struct mm_struct *mm = tsk->mm; 1152 struct mm_struct *mm = tsk->mm;
1161 struct vm_area_struct *vma; 1153 char *tty;
1154
1155 if (!ab)
1156 return;
1162 1157
1163 /* tsk == current */ 1158 /* tsk == current */
1159 cred = current_cred();
1160
1161 spin_lock_irq(&tsk->sighand->siglock);
1162 if (tsk->signal && tsk->signal->tty && tsk->signal->tty->name)
1163 tty = tsk->signal->tty->name;
1164 else
1165 tty = "(none)";
1166 spin_unlock_irq(&tsk->sighand->siglock);
1167
1168
1169 audit_log_format(ab,
1170 " ppid=%ld pid=%d auid=%u uid=%u gid=%u"
1171 " euid=%u suid=%u fsuid=%u"
1172 " egid=%u sgid=%u fsgid=%u ses=%u tty=%s",
1173 sys_getppid(),
1174 tsk->pid,
1175 from_kuid(&init_user_ns, tsk->loginuid),
1176 from_kuid(&init_user_ns, cred->uid),
1177 from_kgid(&init_user_ns, cred->gid),
1178 from_kuid(&init_user_ns, cred->euid),
1179 from_kuid(&init_user_ns, cred->suid),
1180 from_kuid(&init_user_ns, cred->fsuid),
1181 from_kgid(&init_user_ns, cred->egid),
1182 from_kgid(&init_user_ns, cred->sgid),
1183 from_kgid(&init_user_ns, cred->fsgid),
1184 tsk->sessionid, tty);
1164 1185
1165 get_task_comm(name, tsk); 1186 get_task_comm(name, tsk);
1166 audit_log_format(ab, " comm="); 1187 audit_log_format(ab, " comm=");
@@ -1168,23 +1189,17 @@ static void audit_log_task_info(struct audit_buffer *ab, struct task_struct *tsk
1168 1189
1169 if (mm) { 1190 if (mm) {
1170 down_read(&mm->mmap_sem); 1191 down_read(&mm->mmap_sem);
1171 vma = mm->mmap; 1192 if (mm->exe_file)
1172 while (vma) { 1193 audit_log_d_path(ab, " exe=", &mm->exe_file->f_path);
1173 if ((vma->vm_flags & VM_EXECUTABLE) &&
1174 vma->vm_file) {
1175 audit_log_d_path(ab, " exe=",
1176 &vma->vm_file->f_path);
1177 break;
1178 }
1179 vma = vma->vm_next;
1180 }
1181 up_read(&mm->mmap_sem); 1194 up_read(&mm->mmap_sem);
1182 } 1195 }
1183 audit_log_task_context(ab); 1196 audit_log_task_context(ab);
1184} 1197}
1185 1198
1199EXPORT_SYMBOL(audit_log_task_info);
1200
1186static int audit_log_pid_context(struct audit_context *context, pid_t pid, 1201static int audit_log_pid_context(struct audit_context *context, pid_t pid,
1187 uid_t auid, uid_t uid, unsigned int sessionid, 1202 kuid_t auid, kuid_t uid, unsigned int sessionid,
1188 u32 sid, char *comm) 1203 u32 sid, char *comm)
1189{ 1204{
1190 struct audit_buffer *ab; 1205 struct audit_buffer *ab;
@@ -1196,8 +1211,9 @@ static int audit_log_pid_context(struct audit_context *context, pid_t pid,
1196 if (!ab) 1211 if (!ab)
1197 return rc; 1212 return rc;
1198 1213
1199 audit_log_format(ab, "opid=%d oauid=%d ouid=%d oses=%d", pid, auid, 1214 audit_log_format(ab, "opid=%d oauid=%d ouid=%d oses=%d", pid,
1200 uid, sessionid); 1215 from_kuid(&init_user_ns, auid),
1216 from_kuid(&init_user_ns, uid), sessionid);
1201 if (security_secid_to_secctx(sid, &ctx, &len)) { 1217 if (security_secid_to_secctx(sid, &ctx, &len)) {
1202 audit_log_format(ab, " obj=(none)"); 1218 audit_log_format(ab, " obj=(none)");
1203 rc = 1; 1219 rc = 1;
@@ -1447,7 +1463,9 @@ static void show_special(struct audit_context *context, int *call_panic)
1447 u32 osid = context->ipc.osid; 1463 u32 osid = context->ipc.osid;
1448 1464
1449 audit_log_format(ab, "ouid=%u ogid=%u mode=%#ho", 1465 audit_log_format(ab, "ouid=%u ogid=%u mode=%#ho",
1450 context->ipc.uid, context->ipc.gid, context->ipc.mode); 1466 from_kuid(&init_user_ns, context->ipc.uid),
1467 from_kgid(&init_user_ns, context->ipc.gid),
1468 context->ipc.mode);
1451 if (osid) { 1469 if (osid) {
1452 char *ctx = NULL; 1470 char *ctx = NULL;
1453 u32 len; 1471 u32 len;
@@ -1536,7 +1554,7 @@ static void audit_log_name(struct audit_context *context, struct audit_names *n,
1536 case AUDIT_NAME_FULL: 1554 case AUDIT_NAME_FULL:
1537 /* log the full path */ 1555 /* log the full path */
1538 audit_log_format(ab, " name="); 1556 audit_log_format(ab, " name=");
1539 audit_log_untrustedstring(ab, n->name); 1557 audit_log_untrustedstring(ab, n->name->name);
1540 break; 1558 break;
1541 case 0: 1559 case 0:
1542 /* name was specified as a relative path and the 1560 /* name was specified as a relative path and the
@@ -1546,7 +1564,7 @@ static void audit_log_name(struct audit_context *context, struct audit_names *n,
1546 default: 1564 default:
1547 /* log the name's directory component */ 1565 /* log the name's directory component */
1548 audit_log_format(ab, " name="); 1566 audit_log_format(ab, " name=");
1549 audit_log_n_untrustedstring(ab, n->name, 1567 audit_log_n_untrustedstring(ab, n->name->name,
1550 n->name_len); 1568 n->name_len);
1551 } 1569 }
1552 } else 1570 } else
@@ -1560,8 +1578,8 @@ static void audit_log_name(struct audit_context *context, struct audit_names *n,
1560 MAJOR(n->dev), 1578 MAJOR(n->dev),
1561 MINOR(n->dev), 1579 MINOR(n->dev),
1562 n->mode, 1580 n->mode,
1563 n->uid, 1581 from_kuid(&init_user_ns, n->uid),
1564 n->gid, 1582 from_kgid(&init_user_ns, n->gid),
1565 MAJOR(n->rdev), 1583 MAJOR(n->rdev),
1566 MINOR(n->rdev)); 1584 MINOR(n->rdev));
1567 } 1585 }
@@ -1585,26 +1603,12 @@ static void audit_log_name(struct audit_context *context, struct audit_names *n,
1585 1603
1586static void audit_log_exit(struct audit_context *context, struct task_struct *tsk) 1604static void audit_log_exit(struct audit_context *context, struct task_struct *tsk)
1587{ 1605{
1588 const struct cred *cred;
1589 int i, call_panic = 0; 1606 int i, call_panic = 0;
1590 struct audit_buffer *ab; 1607 struct audit_buffer *ab;
1591 struct audit_aux_data *aux; 1608 struct audit_aux_data *aux;
1592 const char *tty;
1593 struct audit_names *n; 1609 struct audit_names *n;
1594 1610
1595 /* tsk == current */ 1611 /* tsk == current */
1596 context->pid = tsk->pid;
1597 if (!context->ppid)
1598 context->ppid = sys_getppid();
1599 cred = current_cred();
1600 context->uid = cred->uid;
1601 context->gid = cred->gid;
1602 context->euid = cred->euid;
1603 context->suid = cred->suid;
1604 context->fsuid = cred->fsuid;
1605 context->egid = cred->egid;
1606 context->sgid = cred->sgid;
1607 context->fsgid = cred->fsgid;
1608 context->personality = tsk->personality; 1612 context->personality = tsk->personality;
1609 1613
1610 ab = audit_log_start(context, GFP_KERNEL, AUDIT_SYSCALL); 1614 ab = audit_log_start(context, GFP_KERNEL, AUDIT_SYSCALL);
@@ -1619,32 +1623,13 @@ static void audit_log_exit(struct audit_context *context, struct task_struct *ts
1619 (context->return_valid==AUDITSC_SUCCESS)?"yes":"no", 1623 (context->return_valid==AUDITSC_SUCCESS)?"yes":"no",
1620 context->return_code); 1624 context->return_code);
1621 1625
1622 spin_lock_irq(&tsk->sighand->siglock);
1623 if (tsk->signal && tsk->signal->tty && tsk->signal->tty->name)
1624 tty = tsk->signal->tty->name;
1625 else
1626 tty = "(none)";
1627 spin_unlock_irq(&tsk->sighand->siglock);
1628
1629 audit_log_format(ab, 1626 audit_log_format(ab,
1630 " a0=%lx a1=%lx a2=%lx a3=%lx items=%d" 1627 " a0=%lx a1=%lx a2=%lx a3=%lx items=%d",
1631 " ppid=%d pid=%d auid=%u uid=%u gid=%u" 1628 context->argv[0],
1632 " euid=%u suid=%u fsuid=%u" 1629 context->argv[1],
1633 " egid=%u sgid=%u fsgid=%u tty=%s ses=%u", 1630 context->argv[2],
1634 context->argv[0], 1631 context->argv[3],
1635 context->argv[1], 1632 context->name_count);
1636 context->argv[2],
1637 context->argv[3],
1638 context->name_count,
1639 context->ppid,
1640 context->pid,
1641 tsk->loginuid,
1642 context->uid,
1643 context->gid,
1644 context->euid, context->suid, context->fsuid,
1645 context->egid, context->sgid, context->fsgid, tty,
1646 tsk->sessionid);
1647
1648 1633
1649 audit_log_task_info(ab, tsk); 1634 audit_log_task_info(ab, tsk);
1650 audit_log_key(ab, context->filterkey); 1635 audit_log_key(ab, context->filterkey);
@@ -2009,7 +1994,8 @@ retry:
2009#endif 1994#endif
2010} 1995}
2011 1996
2012static struct audit_names *audit_alloc_name(struct audit_context *context) 1997static struct audit_names *audit_alloc_name(struct audit_context *context,
1998 unsigned char type)
2013{ 1999{
2014 struct audit_names *aname; 2000 struct audit_names *aname;
2015 2001
@@ -2024,6 +2010,7 @@ static struct audit_names *audit_alloc_name(struct audit_context *context)
2024 } 2010 }
2025 2011
2026 aname->ino = (unsigned long)-1; 2012 aname->ino = (unsigned long)-1;
2013 aname->type = type;
2027 list_add_tail(&aname->list, &context->names_list); 2014 list_add_tail(&aname->list, &context->names_list);
2028 2015
2029 context->name_count++; 2016 context->name_count++;
@@ -2034,13 +2021,36 @@ static struct audit_names *audit_alloc_name(struct audit_context *context)
2034} 2021}
2035 2022
2036/** 2023/**
2024 * audit_reusename - fill out filename with info from existing entry
2025 * @uptr: userland ptr to pathname
2026 *
2027 * Search the audit_names list for the current audit context. If there is an
2028 * existing entry with a matching "uptr" then return the filename
2029 * associated with that audit_name. If not, return NULL.
2030 */
2031struct filename *
2032__audit_reusename(const __user char *uptr)
2033{
2034 struct audit_context *context = current->audit_context;
2035 struct audit_names *n;
2036
2037 list_for_each_entry(n, &context->names_list, list) {
2038 if (!n->name)
2039 continue;
2040 if (n->name->uptr == uptr)
2041 return n->name;
2042 }
2043 return NULL;
2044}
2045
2046/**
2037 * audit_getname - add a name to the list 2047 * audit_getname - add a name to the list
2038 * @name: name to add 2048 * @name: name to add
2039 * 2049 *
2040 * Add a name to the list of audit names for this context. 2050 * Add a name to the list of audit names for this context.
2041 * Called from fs/namei.c:getname(). 2051 * Called from fs/namei.c:getname().
2042 */ 2052 */
2043void __audit_getname(const char *name) 2053void __audit_getname(struct filename *name)
2044{ 2054{
2045 struct audit_context *context = current->audit_context; 2055 struct audit_context *context = current->audit_context;
2046 struct audit_names *n; 2056 struct audit_names *n;
@@ -2054,13 +2064,19 @@ void __audit_getname(const char *name)
2054 return; 2064 return;
2055 } 2065 }
2056 2066
2057 n = audit_alloc_name(context); 2067#if AUDIT_DEBUG
2068 /* The filename _must_ have a populated ->name */
2069 BUG_ON(!name->name);
2070#endif
2071
2072 n = audit_alloc_name(context, AUDIT_TYPE_UNKNOWN);
2058 if (!n) 2073 if (!n)
2059 return; 2074 return;
2060 2075
2061 n->name = name; 2076 n->name = name;
2062 n->name_len = AUDIT_NAME_FULL; 2077 n->name_len = AUDIT_NAME_FULL;
2063 n->name_put = true; 2078 n->name_put = true;
2079 name->aname = n;
2064 2080
2065 if (!context->pwd.dentry) 2081 if (!context->pwd.dentry)
2066 get_fs_pwd(current->fs, &context->pwd); 2082 get_fs_pwd(current->fs, &context->pwd);
@@ -2073,7 +2089,7 @@ void __audit_getname(const char *name)
2073 * then we delay the putname until syscall exit. 2089 * then we delay the putname until syscall exit.
2074 * Called from include/linux/fs.h:putname(). 2090 * Called from include/linux/fs.h:putname().
2075 */ 2091 */
2076void audit_putname(const char *name) 2092void audit_putname(struct filename *name)
2077{ 2093{
2078 struct audit_context *context = current->audit_context; 2094 struct audit_context *context = current->audit_context;
2079 2095
@@ -2088,7 +2104,7 @@ void audit_putname(const char *name)
2088 2104
2089 list_for_each_entry(n, &context->names_list, list) 2105 list_for_each_entry(n, &context->names_list, list)
2090 printk(KERN_ERR "name[%d] = %p = %s\n", i, 2106 printk(KERN_ERR "name[%d] = %p = %s\n", i,
2091 n->name, n->name ?: "(null)"); 2107 n->name, n->name->name ?: "(null)");
2092 } 2108 }
2093#endif 2109#endif
2094 __putname(name); 2110 __putname(name);
@@ -2102,8 +2118,8 @@ void audit_putname(const char *name)
2102 " put_count=%d\n", 2118 " put_count=%d\n",
2103 __FILE__, __LINE__, 2119 __FILE__, __LINE__,
2104 context->serial, context->major, 2120 context->serial, context->major,
2105 context->in_syscall, name, context->name_count, 2121 context->in_syscall, name->name,
2106 context->put_count); 2122 context->name_count, context->put_count);
2107 dump_stack(); 2123 dump_stack();
2108 } 2124 }
2109 } 2125 }
@@ -2146,13 +2162,13 @@ static void audit_copy_inode(struct audit_names *name, const struct dentry *dent
2146} 2162}
2147 2163
2148/** 2164/**
2149 * audit_inode - store the inode and device from a lookup 2165 * __audit_inode - store the inode and device from a lookup
2150 * @name: name being audited 2166 * @name: name being audited
2151 * @dentry: dentry being audited 2167 * @dentry: dentry being audited
2152 * 2168 * @parent: does this dentry represent the parent?
2153 * Called from fs/namei.c:path_lookup().
2154 */ 2169 */
2155void __audit_inode(const char *name, const struct dentry *dentry) 2170void __audit_inode(struct filename *name, const struct dentry *dentry,
2171 unsigned int parent)
2156{ 2172{
2157 struct audit_context *context = current->audit_context; 2173 struct audit_context *context = current->audit_context;
2158 const struct inode *inode = dentry->d_inode; 2174 const struct inode *inode = dentry->d_inode;
@@ -2161,24 +2177,69 @@ void __audit_inode(const char *name, const struct dentry *dentry)
2161 if (!context->in_syscall) 2177 if (!context->in_syscall)
2162 return; 2178 return;
2163 2179
2180 if (!name)
2181 goto out_alloc;
2182
2183#if AUDIT_DEBUG
2184 /* The struct filename _must_ have a populated ->name */
2185 BUG_ON(!name->name);
2186#endif
2187 /*
2188 * If we have a pointer to an audit_names entry already, then we can
2189 * just use it directly if the type is correct.
2190 */
2191 n = name->aname;
2192 if (n) {
2193 if (parent) {
2194 if (n->type == AUDIT_TYPE_PARENT ||
2195 n->type == AUDIT_TYPE_UNKNOWN)
2196 goto out;
2197 } else {
2198 if (n->type != AUDIT_TYPE_PARENT)
2199 goto out;
2200 }
2201 }
2202
2164 list_for_each_entry_reverse(n, &context->names_list, list) { 2203 list_for_each_entry_reverse(n, &context->names_list, list) {
2165 if (n->name && (n->name == name)) 2204 /* does the name pointer match? */
2166 goto out; 2205 if (!n->name || n->name->name != name->name)
2206 continue;
2207
2208 /* match the correct record type */
2209 if (parent) {
2210 if (n->type == AUDIT_TYPE_PARENT ||
2211 n->type == AUDIT_TYPE_UNKNOWN)
2212 goto out;
2213 } else {
2214 if (n->type != AUDIT_TYPE_PARENT)
2215 goto out;
2216 }
2167 } 2217 }
2168 2218
2169 /* unable to find the name from a previous getname() */ 2219out_alloc:
2170 n = audit_alloc_name(context); 2220 /* unable to find the name from a previous getname(). Allocate a new
2221 * anonymous entry.
2222 */
2223 n = audit_alloc_name(context, AUDIT_TYPE_NORMAL);
2171 if (!n) 2224 if (!n)
2172 return; 2225 return;
2173out: 2226out:
2227 if (parent) {
2228 n->name_len = n->name ? parent_len(n->name->name) : AUDIT_NAME_FULL;
2229 n->type = AUDIT_TYPE_PARENT;
2230 } else {
2231 n->name_len = AUDIT_NAME_FULL;
2232 n->type = AUDIT_TYPE_NORMAL;
2233 }
2174 handle_path(dentry); 2234 handle_path(dentry);
2175 audit_copy_inode(n, dentry, inode); 2235 audit_copy_inode(n, dentry, inode);
2176} 2236}
2177 2237
2178/** 2238/**
2179 * audit_inode_child - collect inode info for created/removed objects 2239 * __audit_inode_child - collect inode info for created/removed objects
2180 * @dentry: dentry being audited
2181 * @parent: inode of dentry parent 2240 * @parent: inode of dentry parent
2241 * @dentry: dentry being audited
2242 * @type: AUDIT_TYPE_* value that we're looking for
2182 * 2243 *
2183 * For syscalls that create or remove filesystem objects, audit_inode 2244 * For syscalls that create or remove filesystem objects, audit_inode
2184 * can only collect information for the filesystem object's parent. 2245 * can only collect information for the filesystem object's parent.
@@ -2188,15 +2249,14 @@ out:
2188 * must be hooked prior, in order to capture the target inode during 2249 * must be hooked prior, in order to capture the target inode during
2189 * unsuccessful attempts. 2250 * unsuccessful attempts.
2190 */ 2251 */
2191void __audit_inode_child(const struct dentry *dentry, 2252void __audit_inode_child(const struct inode *parent,
2192 const struct inode *parent) 2253 const struct dentry *dentry,
2254 const unsigned char type)
2193{ 2255{
2194 struct audit_context *context = current->audit_context; 2256 struct audit_context *context = current->audit_context;
2195 const char *found_parent = NULL, *found_child = NULL;
2196 const struct inode *inode = dentry->d_inode; 2257 const struct inode *inode = dentry->d_inode;
2197 const char *dname = dentry->d_name.name; 2258 const char *dname = dentry->d_name.name;
2198 struct audit_names *n; 2259 struct audit_names *n, *found_parent = NULL, *found_child = NULL;
2199 int dirlen = 0;
2200 2260
2201 if (!context->in_syscall) 2261 if (!context->in_syscall)
2202 return; 2262 return;
@@ -2204,62 +2264,65 @@ void __audit_inode_child(const struct dentry *dentry,
2204 if (inode) 2264 if (inode)
2205 handle_one(inode); 2265 handle_one(inode);
2206 2266
2207 /* parent is more likely, look for it first */ 2267 /* look for a parent entry first */
2208 list_for_each_entry(n, &context->names_list, list) { 2268 list_for_each_entry(n, &context->names_list, list) {
2209 if (!n->name) 2269 if (!n->name || n->type != AUDIT_TYPE_PARENT)
2210 continue; 2270 continue;
2211 2271
2212 if (n->ino == parent->i_ino && 2272 if (n->ino == parent->i_ino &&
2213 !audit_compare_dname_path(dname, n->name, &dirlen)) { 2273 !audit_compare_dname_path(dname, n->name->name, n->name_len)) {
2214 n->name_len = dirlen; /* update parent data in place */ 2274 found_parent = n;
2215 found_parent = n->name; 2275 break;
2216 goto add_names;
2217 } 2276 }
2218 } 2277 }
2219 2278
2220 /* no matching parent, look for matching child */ 2279 /* is there a matching child entry? */
2221 list_for_each_entry(n, &context->names_list, list) { 2280 list_for_each_entry(n, &context->names_list, list) {
2222 if (!n->name) 2281 /* can only match entries that have a name */
2282 if (!n->name || n->type != type)
2223 continue; 2283 continue;
2224 2284
2225 /* strcmp() is the more likely scenario */ 2285 /* if we found a parent, make sure this one is a child of it */
2226 if (!strcmp(dname, n->name) || 2286 if (found_parent && (n->name != found_parent->name))
2227 !audit_compare_dname_path(dname, n->name, &dirlen)) { 2287 continue;
2228 if (inode) 2288
2229 audit_copy_inode(n, NULL, inode); 2289 if (!strcmp(dname, n->name->name) ||
2230 else 2290 !audit_compare_dname_path(dname, n->name->name,
2231 n->ino = (unsigned long)-1; 2291 found_parent ?
2232 found_child = n->name; 2292 found_parent->name_len :
2233 goto add_names; 2293 AUDIT_NAME_FULL)) {
2294 found_child = n;
2295 break;
2234 } 2296 }
2235 } 2297 }
2236 2298
2237add_names:
2238 if (!found_parent) { 2299 if (!found_parent) {
2239 n = audit_alloc_name(context); 2300 /* create a new, "anonymous" parent record */
2301 n = audit_alloc_name(context, AUDIT_TYPE_PARENT);
2240 if (!n) 2302 if (!n)
2241 return; 2303 return;
2242 audit_copy_inode(n, NULL, parent); 2304 audit_copy_inode(n, NULL, parent);
2243 } 2305 }
2244 2306
2245 if (!found_child) { 2307 if (!found_child) {
2246 n = audit_alloc_name(context); 2308 found_child = audit_alloc_name(context, type);
2247 if (!n) 2309 if (!found_child)
2248 return; 2310 return;
2249 2311
2250 /* Re-use the name belonging to the slot for a matching parent 2312 /* Re-use the name belonging to the slot for a matching parent
2251 * directory. All names for this context are relinquished in 2313 * directory. All names for this context are relinquished in
2252 * audit_free_names() */ 2314 * audit_free_names() */
2253 if (found_parent) { 2315 if (found_parent) {
2254 n->name = found_parent; 2316 found_child->name = found_parent->name;
2255 n->name_len = AUDIT_NAME_FULL; 2317 found_child->name_len = AUDIT_NAME_FULL;
2256 /* don't call __putname() */ 2318 /* don't call __putname() */
2257 n->name_put = false; 2319 found_child->name_put = false;
2258 } 2320 }
2259
2260 if (inode)
2261 audit_copy_inode(n, NULL, inode);
2262 } 2321 }
2322 if (inode)
2323 audit_copy_inode(found_child, dentry, inode);
2324 else
2325 found_child->ino = (unsigned long)-1;
2263} 2326}
2264EXPORT_SYMBOL_GPL(__audit_inode_child); 2327EXPORT_SYMBOL_GPL(__audit_inode_child);
2265 2328
@@ -2299,14 +2362,14 @@ static atomic_t session_id = ATOMIC_INIT(0);
2299 * 2362 *
2300 * Called (set) from fs/proc/base.c::proc_loginuid_write(). 2363 * Called (set) from fs/proc/base.c::proc_loginuid_write().
2301 */ 2364 */
2302int audit_set_loginuid(uid_t loginuid) 2365int audit_set_loginuid(kuid_t loginuid)
2303{ 2366{
2304 struct task_struct *task = current; 2367 struct task_struct *task = current;
2305 struct audit_context *context = task->audit_context; 2368 struct audit_context *context = task->audit_context;
2306 unsigned int sessionid; 2369 unsigned int sessionid;
2307 2370
2308#ifdef CONFIG_AUDIT_LOGINUID_IMMUTABLE 2371#ifdef CONFIG_AUDIT_LOGINUID_IMMUTABLE
2309 if (task->loginuid != -1) 2372 if (uid_valid(task->loginuid))
2310 return -EPERM; 2373 return -EPERM;
2311#else /* CONFIG_AUDIT_LOGINUID_IMMUTABLE */ 2374#else /* CONFIG_AUDIT_LOGINUID_IMMUTABLE */
2312 if (!capable(CAP_AUDIT_CONTROL)) 2375 if (!capable(CAP_AUDIT_CONTROL))
@@ -2322,8 +2385,10 @@ int audit_set_loginuid(uid_t loginuid)
2322 audit_log_format(ab, "login pid=%d uid=%u " 2385 audit_log_format(ab, "login pid=%d uid=%u "
2323 "old auid=%u new auid=%u" 2386 "old auid=%u new auid=%u"
2324 " old ses=%u new ses=%u", 2387 " old ses=%u new ses=%u",
2325 task->pid, task_uid(task), 2388 task->pid,
2326 task->loginuid, loginuid, 2389 from_kuid(&init_user_ns, task_uid(task)),
2390 from_kuid(&init_user_ns, task->loginuid),
2391 from_kuid(&init_user_ns, loginuid),
2327 task->sessionid, sessionid); 2392 task->sessionid, sessionid);
2328 audit_log_end(ab); 2393 audit_log_end(ab);
2329 } 2394 }
@@ -2546,12 +2611,12 @@ int __audit_signal_info(int sig, struct task_struct *t)
2546 struct audit_aux_data_pids *axp; 2611 struct audit_aux_data_pids *axp;
2547 struct task_struct *tsk = current; 2612 struct task_struct *tsk = current;
2548 struct audit_context *ctx = tsk->audit_context; 2613 struct audit_context *ctx = tsk->audit_context;
2549 uid_t uid = current_uid(), t_uid = task_uid(t); 2614 kuid_t uid = current_uid(), t_uid = task_uid(t);
2550 2615
2551 if (audit_pid && t->tgid == audit_pid) { 2616 if (audit_pid && t->tgid == audit_pid) {
2552 if (sig == SIGTERM || sig == SIGHUP || sig == SIGUSR1 || sig == SIGUSR2) { 2617 if (sig == SIGTERM || sig == SIGHUP || sig == SIGUSR1 || sig == SIGUSR2) {
2553 audit_sig_pid = tsk->pid; 2618 audit_sig_pid = tsk->pid;
2554 if (tsk->loginuid != -1) 2619 if (uid_valid(tsk->loginuid))
2555 audit_sig_uid = tsk->loginuid; 2620 audit_sig_uid = tsk->loginuid;
2556 else 2621 else
2557 audit_sig_uid = uid; 2622 audit_sig_uid = uid;
@@ -2672,8 +2737,8 @@ void __audit_mmap_fd(int fd, int flags)
2672 2737
2673static void audit_log_abend(struct audit_buffer *ab, char *reason, long signr) 2738static void audit_log_abend(struct audit_buffer *ab, char *reason, long signr)
2674{ 2739{
2675 uid_t auid, uid; 2740 kuid_t auid, uid;
2676 gid_t gid; 2741 kgid_t gid;
2677 unsigned int sessionid; 2742 unsigned int sessionid;
2678 2743
2679 auid = audit_get_loginuid(current); 2744 auid = audit_get_loginuid(current);
@@ -2681,7 +2746,10 @@ static void audit_log_abend(struct audit_buffer *ab, char *reason, long signr)
2681 current_uid_gid(&uid, &gid); 2746 current_uid_gid(&uid, &gid);
2682 2747
2683 audit_log_format(ab, "auid=%u uid=%u gid=%u ses=%u", 2748 audit_log_format(ab, "auid=%u uid=%u gid=%u ses=%u",
2684 auid, uid, gid, sessionid); 2749 from_kuid(&init_user_ns, auid),
2750 from_kuid(&init_user_ns, uid),
2751 from_kgid(&init_user_ns, gid),
2752 sessionid);
2685 audit_log_task_context(ab); 2753 audit_log_task_context(ab);
2686 audit_log_format(ab, " pid=%d comm=", current->pid); 2754 audit_log_format(ab, " pid=%d comm=", current->pid);
2687 audit_log_untrustedstring(ab, current->comm); 2755 audit_log_untrustedstring(ab, current->comm);
diff --git a/kernel/cgroup.c b/kernel/cgroup.c
index 79818507e444..13774b3b39aa 100644
--- a/kernel/cgroup.c
+++ b/kernel/cgroup.c
@@ -88,11 +88,12 @@ static DEFINE_MUTEX(cgroup_root_mutex);
88 88
89/* 89/*
90 * Generate an array of cgroup subsystem pointers. At boot time, this is 90 * Generate an array of cgroup subsystem pointers. At boot time, this is
91 * populated up to CGROUP_BUILTIN_SUBSYS_COUNT, and modular subsystems are 91 * populated with the built in subsystems, and modular subsystems are
92 * registered after that. The mutable section of this array is protected by 92 * registered after that. The mutable section of this array is protected by
93 * cgroup_mutex. 93 * cgroup_mutex.
94 */ 94 */
95#define SUBSYS(_x) &_x ## _subsys, 95#define SUBSYS(_x) [_x ## _subsys_id] = &_x ## _subsys,
96#define IS_SUBSYS_ENABLED(option) IS_BUILTIN(option)
96static struct cgroup_subsys *subsys[CGROUP_SUBSYS_COUNT] = { 97static struct cgroup_subsys *subsys[CGROUP_SUBSYS_COUNT] = {
97#include <linux/cgroup_subsys.h> 98#include <linux/cgroup_subsys.h>
98}; 99};
@@ -111,13 +112,13 @@ struct cgroupfs_root {
111 * The bitmask of subsystems intended to be attached to this 112 * The bitmask of subsystems intended to be attached to this
112 * hierarchy 113 * hierarchy
113 */ 114 */
114 unsigned long subsys_bits; 115 unsigned long subsys_mask;
115 116
116 /* Unique id for this hierarchy. */ 117 /* Unique id for this hierarchy. */
117 int hierarchy_id; 118 int hierarchy_id;
118 119
119 /* The bitmask of subsystems currently attached to this hierarchy */ 120 /* The bitmask of subsystems currently attached to this hierarchy */
120 unsigned long actual_subsys_bits; 121 unsigned long actual_subsys_mask;
121 122
122 /* A list running through the attached subsystems */ 123 /* A list running through the attached subsystems */
123 struct list_head subsys_list; 124 struct list_head subsys_list;
@@ -276,7 +277,8 @@ inline int cgroup_is_removed(const struct cgroup *cgrp)
276 277
277/* bits in struct cgroupfs_root flags field */ 278/* bits in struct cgroupfs_root flags field */
278enum { 279enum {
279 ROOT_NOPREFIX, /* mounted subsystems have no named prefix */ 280 ROOT_NOPREFIX, /* mounted subsystems have no named prefix */
281 ROOT_XATTR, /* supports extended attributes */
280}; 282};
281 283
282static int cgroup_is_releasable(const struct cgroup *cgrp) 284static int cgroup_is_releasable(const struct cgroup *cgrp)
@@ -556,7 +558,7 @@ static struct css_set *find_existing_css_set(
556 * won't change, so no need for locking. 558 * won't change, so no need for locking.
557 */ 559 */
558 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { 560 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
559 if (root->subsys_bits & (1UL << i)) { 561 if (root->subsys_mask & (1UL << i)) {
560 /* Subsystem is in this hierarchy. So we want 562 /* Subsystem is in this hierarchy. So we want
561 * the subsystem state from the new 563 * the subsystem state from the new
562 * cgroup */ 564 * cgroup */
@@ -824,7 +826,8 @@ EXPORT_SYMBOL_GPL(cgroup_unlock);
824static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode); 826static int cgroup_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode);
825static struct dentry *cgroup_lookup(struct inode *, struct dentry *, unsigned int); 827static struct dentry *cgroup_lookup(struct inode *, struct dentry *, unsigned int);
826static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry); 828static int cgroup_rmdir(struct inode *unused_dir, struct dentry *dentry);
827static int cgroup_populate_dir(struct cgroup *cgrp); 829static int cgroup_populate_dir(struct cgroup *cgrp, bool base_files,
830 unsigned long subsys_mask);
828static const struct inode_operations cgroup_dir_inode_operations; 831static const struct inode_operations cgroup_dir_inode_operations;
829static const struct file_operations proc_cgroupstats_operations; 832static const struct file_operations proc_cgroupstats_operations;
830 833
@@ -912,15 +915,19 @@ static void cgroup_diput(struct dentry *dentry, struct inode *inode)
912 */ 915 */
913 BUG_ON(!list_empty(&cgrp->pidlists)); 916 BUG_ON(!list_empty(&cgrp->pidlists));
914 917
918 simple_xattrs_free(&cgrp->xattrs);
919
915 kfree_rcu(cgrp, rcu_head); 920 kfree_rcu(cgrp, rcu_head);
916 } else { 921 } else {
917 struct cfent *cfe = __d_cfe(dentry); 922 struct cfent *cfe = __d_cfe(dentry);
918 struct cgroup *cgrp = dentry->d_parent->d_fsdata; 923 struct cgroup *cgrp = dentry->d_parent->d_fsdata;
924 struct cftype *cft = cfe->type;
919 925
920 WARN_ONCE(!list_empty(&cfe->node) && 926 WARN_ONCE(!list_empty(&cfe->node) &&
921 cgrp != &cgrp->root->top_cgroup, 927 cgrp != &cgrp->root->top_cgroup,
922 "cfe still linked for %s\n", cfe->type->name); 928 "cfe still linked for %s\n", cfe->type->name);
923 kfree(cfe); 929 kfree(cfe);
930 simple_xattrs_free(&cft->xattrs);
924 } 931 }
925 iput(inode); 932 iput(inode);
926} 933}
@@ -963,12 +970,29 @@ static int cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft)
963 return -ENOENT; 970 return -ENOENT;
964} 971}
965 972
966static void cgroup_clear_directory(struct dentry *dir) 973/**
974 * cgroup_clear_directory - selective removal of base and subsystem files
975 * @dir: directory containing the files
976 * @base_files: true if the base files should be removed
977 * @subsys_mask: mask of the subsystem ids whose files should be removed
978 */
979static void cgroup_clear_directory(struct dentry *dir, bool base_files,
980 unsigned long subsys_mask)
967{ 981{
968 struct cgroup *cgrp = __d_cgrp(dir); 982 struct cgroup *cgrp = __d_cgrp(dir);
983 struct cgroup_subsys *ss;
969 984
970 while (!list_empty(&cgrp->files)) 985 for_each_subsys(cgrp->root, ss) {
971 cgroup_rm_file(cgrp, NULL); 986 struct cftype_set *set;
987 if (!test_bit(ss->subsys_id, &subsys_mask))
988 continue;
989 list_for_each_entry(set, &ss->cftsets, node)
990 cgroup_rm_file(cgrp, set->cfts);
991 }
992 if (base_files) {
993 while (!list_empty(&cgrp->files))
994 cgroup_rm_file(cgrp, NULL);
995 }
972} 996}
973 997
974/* 998/*
@@ -977,8 +1001,9 @@ static void cgroup_clear_directory(struct dentry *dir)
977static void cgroup_d_remove_dir(struct dentry *dentry) 1001static void cgroup_d_remove_dir(struct dentry *dentry)
978{ 1002{
979 struct dentry *parent; 1003 struct dentry *parent;
1004 struct cgroupfs_root *root = dentry->d_sb->s_fs_info;
980 1005
981 cgroup_clear_directory(dentry); 1006 cgroup_clear_directory(dentry, true, root->subsys_mask);
982 1007
983 parent = dentry->d_parent; 1008 parent = dentry->d_parent;
984 spin_lock(&parent->d_lock); 1009 spin_lock(&parent->d_lock);
@@ -1022,22 +1047,22 @@ void cgroup_release_and_wakeup_rmdir(struct cgroup_subsys_state *css)
1022 * returns an error, no reference counts are touched. 1047 * returns an error, no reference counts are touched.
1023 */ 1048 */
1024static int rebind_subsystems(struct cgroupfs_root *root, 1049static int rebind_subsystems(struct cgroupfs_root *root,
1025 unsigned long final_bits) 1050 unsigned long final_subsys_mask)
1026{ 1051{
1027 unsigned long added_bits, removed_bits; 1052 unsigned long added_mask, removed_mask;
1028 struct cgroup *cgrp = &root->top_cgroup; 1053 struct cgroup *cgrp = &root->top_cgroup;
1029 int i; 1054 int i;
1030 1055
1031 BUG_ON(!mutex_is_locked(&cgroup_mutex)); 1056 BUG_ON(!mutex_is_locked(&cgroup_mutex));
1032 BUG_ON(!mutex_is_locked(&cgroup_root_mutex)); 1057 BUG_ON(!mutex_is_locked(&cgroup_root_mutex));
1033 1058
1034 removed_bits = root->actual_subsys_bits & ~final_bits; 1059 removed_mask = root->actual_subsys_mask & ~final_subsys_mask;
1035 added_bits = final_bits & ~root->actual_subsys_bits; 1060 added_mask = final_subsys_mask & ~root->actual_subsys_mask;
1036 /* Check that any added subsystems are currently free */ 1061 /* Check that any added subsystems are currently free */
1037 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { 1062 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1038 unsigned long bit = 1UL << i; 1063 unsigned long bit = 1UL << i;
1039 struct cgroup_subsys *ss = subsys[i]; 1064 struct cgroup_subsys *ss = subsys[i];
1040 if (!(bit & added_bits)) 1065 if (!(bit & added_mask))
1041 continue; 1066 continue;
1042 /* 1067 /*
1043 * Nobody should tell us to do a subsys that doesn't exist: 1068 * Nobody should tell us to do a subsys that doesn't exist:
@@ -1062,7 +1087,7 @@ static int rebind_subsystems(struct cgroupfs_root *root,
1062 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) { 1087 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1063 struct cgroup_subsys *ss = subsys[i]; 1088 struct cgroup_subsys *ss = subsys[i];
1064 unsigned long bit = 1UL << i; 1089 unsigned long bit = 1UL << i;
1065 if (bit & added_bits) { 1090 if (bit & added_mask) {
1066 /* We're binding this subsystem to this hierarchy */ 1091 /* We're binding this subsystem to this hierarchy */
1067 BUG_ON(ss == NULL); 1092 BUG_ON(ss == NULL);
1068 BUG_ON(cgrp->subsys[i]); 1093 BUG_ON(cgrp->subsys[i]);
@@ -1075,7 +1100,7 @@ static int rebind_subsystems(struct cgroupfs_root *root,
1075 if (ss->bind) 1100 if (ss->bind)
1076 ss->bind(cgrp); 1101 ss->bind(cgrp);
1077 /* refcount was already taken, and we're keeping it */ 1102 /* refcount was already taken, and we're keeping it */
1078 } else if (bit & removed_bits) { 1103 } else if (bit & removed_mask) {
1079 /* We're removing this subsystem */ 1104 /* We're removing this subsystem */
1080 BUG_ON(ss == NULL); 1105 BUG_ON(ss == NULL);
1081 BUG_ON(cgrp->subsys[i] != dummytop->subsys[i]); 1106 BUG_ON(cgrp->subsys[i] != dummytop->subsys[i]);
@@ -1088,7 +1113,7 @@ static int rebind_subsystems(struct cgroupfs_root *root,
1088 list_move(&ss->sibling, &rootnode.subsys_list); 1113 list_move(&ss->sibling, &rootnode.subsys_list);
1089 /* subsystem is now free - drop reference on module */ 1114 /* subsystem is now free - drop reference on module */
1090 module_put(ss->module); 1115 module_put(ss->module);
1091 } else if (bit & final_bits) { 1116 } else if (bit & final_subsys_mask) {
1092 /* Subsystem state should already exist */ 1117 /* Subsystem state should already exist */
1093 BUG_ON(ss == NULL); 1118 BUG_ON(ss == NULL);
1094 BUG_ON(!cgrp->subsys[i]); 1119 BUG_ON(!cgrp->subsys[i]);
@@ -1105,7 +1130,7 @@ static int rebind_subsystems(struct cgroupfs_root *root,
1105 BUG_ON(cgrp->subsys[i]); 1130 BUG_ON(cgrp->subsys[i]);
1106 } 1131 }
1107 } 1132 }
1108 root->subsys_bits = root->actual_subsys_bits = final_bits; 1133 root->subsys_mask = root->actual_subsys_mask = final_subsys_mask;
1109 synchronize_rcu(); 1134 synchronize_rcu();
1110 1135
1111 return 0; 1136 return 0;
@@ -1121,6 +1146,8 @@ static int cgroup_show_options(struct seq_file *seq, struct dentry *dentry)
1121 seq_printf(seq, ",%s", ss->name); 1146 seq_printf(seq, ",%s", ss->name);
1122 if (test_bit(ROOT_NOPREFIX, &root->flags)) 1147 if (test_bit(ROOT_NOPREFIX, &root->flags))
1123 seq_puts(seq, ",noprefix"); 1148 seq_puts(seq, ",noprefix");
1149 if (test_bit(ROOT_XATTR, &root->flags))
1150 seq_puts(seq, ",xattr");
1124 if (strlen(root->release_agent_path)) 1151 if (strlen(root->release_agent_path))
1125 seq_printf(seq, ",release_agent=%s", root->release_agent_path); 1152 seq_printf(seq, ",release_agent=%s", root->release_agent_path);
1126 if (clone_children(&root->top_cgroup)) 1153 if (clone_children(&root->top_cgroup))
@@ -1132,7 +1159,7 @@ static int cgroup_show_options(struct seq_file *seq, struct dentry *dentry)
1132} 1159}
1133 1160
1134struct cgroup_sb_opts { 1161struct cgroup_sb_opts {
1135 unsigned long subsys_bits; 1162 unsigned long subsys_mask;
1136 unsigned long flags; 1163 unsigned long flags;
1137 char *release_agent; 1164 char *release_agent;
1138 bool clone_children; 1165 bool clone_children;
@@ -1189,6 +1216,10 @@ static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts)
1189 opts->clone_children = true; 1216 opts->clone_children = true;
1190 continue; 1217 continue;
1191 } 1218 }
1219 if (!strcmp(token, "xattr")) {
1220 set_bit(ROOT_XATTR, &opts->flags);
1221 continue;
1222 }
1192 if (!strncmp(token, "release_agent=", 14)) { 1223 if (!strncmp(token, "release_agent=", 14)) {
1193 /* Specifying two release agents is forbidden */ 1224 /* Specifying two release agents is forbidden */
1194 if (opts->release_agent) 1225 if (opts->release_agent)
@@ -1237,7 +1268,7 @@ static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts)
1237 /* Mutually exclusive option 'all' + subsystem name */ 1268 /* Mutually exclusive option 'all' + subsystem name */
1238 if (all_ss) 1269 if (all_ss)
1239 return -EINVAL; 1270 return -EINVAL;
1240 set_bit(i, &opts->subsys_bits); 1271 set_bit(i, &opts->subsys_mask);
1241 one_ss = true; 1272 one_ss = true;
1242 1273
1243 break; 1274 break;
@@ -1258,7 +1289,7 @@ static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts)
1258 continue; 1289 continue;
1259 if (ss->disabled) 1290 if (ss->disabled)
1260 continue; 1291 continue;
1261 set_bit(i, &opts->subsys_bits); 1292 set_bit(i, &opts->subsys_mask);
1262 } 1293 }
1263 } 1294 }
1264 1295
@@ -1270,19 +1301,19 @@ static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts)
1270 * the cpuset subsystem. 1301 * the cpuset subsystem.
1271 */ 1302 */
1272 if (test_bit(ROOT_NOPREFIX, &opts->flags) && 1303 if (test_bit(ROOT_NOPREFIX, &opts->flags) &&
1273 (opts->subsys_bits & mask)) 1304 (opts->subsys_mask & mask))
1274 return -EINVAL; 1305 return -EINVAL;
1275 1306
1276 1307
1277 /* Can't specify "none" and some subsystems */ 1308 /* Can't specify "none" and some subsystems */
1278 if (opts->subsys_bits && opts->none) 1309 if (opts->subsys_mask && opts->none)
1279 return -EINVAL; 1310 return -EINVAL;
1280 1311
1281 /* 1312 /*
1282 * We either have to specify by name or by subsystems. (So all 1313 * We either have to specify by name or by subsystems. (So all
1283 * empty hierarchies must have a name). 1314 * empty hierarchies must have a name).
1284 */ 1315 */
1285 if (!opts->subsys_bits && !opts->name) 1316 if (!opts->subsys_mask && !opts->name)
1286 return -EINVAL; 1317 return -EINVAL;
1287 1318
1288 /* 1319 /*
@@ -1291,10 +1322,10 @@ static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts)
1291 * take duplicate reference counts on a subsystem that's already used, 1322 * take duplicate reference counts on a subsystem that's already used,
1292 * but rebind_subsystems handles this case. 1323 * but rebind_subsystems handles this case.
1293 */ 1324 */
1294 for (i = CGROUP_BUILTIN_SUBSYS_COUNT; i < CGROUP_SUBSYS_COUNT; i++) { 1325 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1295 unsigned long bit = 1UL << i; 1326 unsigned long bit = 1UL << i;
1296 1327
1297 if (!(bit & opts->subsys_bits)) 1328 if (!(bit & opts->subsys_mask))
1298 continue; 1329 continue;
1299 if (!try_module_get(subsys[i]->module)) { 1330 if (!try_module_get(subsys[i]->module)) {
1300 module_pin_failed = true; 1331 module_pin_failed = true;
@@ -1307,11 +1338,11 @@ static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts)
1307 * raced with a module_delete call, and to the user this is 1338 * raced with a module_delete call, and to the user this is
1308 * essentially a "subsystem doesn't exist" case. 1339 * essentially a "subsystem doesn't exist" case.
1309 */ 1340 */
1310 for (i--; i >= CGROUP_BUILTIN_SUBSYS_COUNT; i--) { 1341 for (i--; i >= 0; i--) {
1311 /* drop refcounts only on the ones we took */ 1342 /* drop refcounts only on the ones we took */
1312 unsigned long bit = 1UL << i; 1343 unsigned long bit = 1UL << i;
1313 1344
1314 if (!(bit & opts->subsys_bits)) 1345 if (!(bit & opts->subsys_mask))
1315 continue; 1346 continue;
1316 module_put(subsys[i]->module); 1347 module_put(subsys[i]->module);
1317 } 1348 }
@@ -1321,13 +1352,13 @@ static int parse_cgroupfs_options(char *data, struct cgroup_sb_opts *opts)
1321 return 0; 1352 return 0;
1322} 1353}
1323 1354
1324static void drop_parsed_module_refcounts(unsigned long subsys_bits) 1355static void drop_parsed_module_refcounts(unsigned long subsys_mask)
1325{ 1356{
1326 int i; 1357 int i;
1327 for (i = CGROUP_BUILTIN_SUBSYS_COUNT; i < CGROUP_SUBSYS_COUNT; i++) { 1358 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
1328 unsigned long bit = 1UL << i; 1359 unsigned long bit = 1UL << i;
1329 1360
1330 if (!(bit & subsys_bits)) 1361 if (!(bit & subsys_mask))
1331 continue; 1362 continue;
1332 module_put(subsys[i]->module); 1363 module_put(subsys[i]->module);
1333 } 1364 }
@@ -1339,6 +1370,7 @@ static int cgroup_remount(struct super_block *sb, int *flags, char *data)
1339 struct cgroupfs_root *root = sb->s_fs_info; 1370 struct cgroupfs_root *root = sb->s_fs_info;
1340 struct cgroup *cgrp = &root->top_cgroup; 1371 struct cgroup *cgrp = &root->top_cgroup;
1341 struct cgroup_sb_opts opts; 1372 struct cgroup_sb_opts opts;
1373 unsigned long added_mask, removed_mask;
1342 1374
1343 mutex_lock(&cgrp->dentry->d_inode->i_mutex); 1375 mutex_lock(&cgrp->dentry->d_inode->i_mutex);
1344 mutex_lock(&cgroup_mutex); 1376 mutex_lock(&cgroup_mutex);
@@ -1350,27 +1382,31 @@ static int cgroup_remount(struct super_block *sb, int *flags, char *data)
1350 goto out_unlock; 1382 goto out_unlock;
1351 1383
1352 /* See feature-removal-schedule.txt */ 1384 /* See feature-removal-schedule.txt */
1353 if (opts.subsys_bits != root->actual_subsys_bits || opts.release_agent) 1385 if (opts.subsys_mask != root->actual_subsys_mask || opts.release_agent)
1354 pr_warning("cgroup: option changes via remount are deprecated (pid=%d comm=%s)\n", 1386 pr_warning("cgroup: option changes via remount are deprecated (pid=%d comm=%s)\n",
1355 task_tgid_nr(current), current->comm); 1387 task_tgid_nr(current), current->comm);
1356 1388
1389 added_mask = opts.subsys_mask & ~root->subsys_mask;
1390 removed_mask = root->subsys_mask & ~opts.subsys_mask;
1391
1357 /* Don't allow flags or name to change at remount */ 1392 /* Don't allow flags or name to change at remount */
1358 if (opts.flags != root->flags || 1393 if (opts.flags != root->flags ||
1359 (opts.name && strcmp(opts.name, root->name))) { 1394 (opts.name && strcmp(opts.name, root->name))) {
1360 ret = -EINVAL; 1395 ret = -EINVAL;
1361 drop_parsed_module_refcounts(opts.subsys_bits); 1396 drop_parsed_module_refcounts(opts.subsys_mask);
1362 goto out_unlock; 1397 goto out_unlock;
1363 } 1398 }
1364 1399
1365 ret = rebind_subsystems(root, opts.subsys_bits); 1400 ret = rebind_subsystems(root, opts.subsys_mask);
1366 if (ret) { 1401 if (ret) {
1367 drop_parsed_module_refcounts(opts.subsys_bits); 1402 drop_parsed_module_refcounts(opts.subsys_mask);
1368 goto out_unlock; 1403 goto out_unlock;
1369 } 1404 }
1370 1405
1371 /* clear out any existing files and repopulate subsystem files */ 1406 /* clear out any existing files and repopulate subsystem files */
1372 cgroup_clear_directory(cgrp->dentry); 1407 cgroup_clear_directory(cgrp->dentry, false, removed_mask);
1373 cgroup_populate_dir(cgrp); 1408 /* re-populate subsystem files */
1409 cgroup_populate_dir(cgrp, false, added_mask);
1374 1410
1375 if (opts.release_agent) 1411 if (opts.release_agent)
1376 strcpy(root->release_agent_path, opts.release_agent); 1412 strcpy(root->release_agent_path, opts.release_agent);
@@ -1401,6 +1437,7 @@ static void init_cgroup_housekeeping(struct cgroup *cgrp)
1401 mutex_init(&cgrp->pidlist_mutex); 1437 mutex_init(&cgrp->pidlist_mutex);
1402 INIT_LIST_HEAD(&cgrp->event_list); 1438 INIT_LIST_HEAD(&cgrp->event_list);
1403 spin_lock_init(&cgrp->event_list_lock); 1439 spin_lock_init(&cgrp->event_list_lock);
1440 simple_xattrs_init(&cgrp->xattrs);
1404} 1441}
1405 1442
1406static void init_cgroup_root(struct cgroupfs_root *root) 1443static void init_cgroup_root(struct cgroupfs_root *root)
@@ -1455,8 +1492,8 @@ static int cgroup_test_super(struct super_block *sb, void *data)
1455 * If we asked for subsystems (or explicitly for no 1492 * If we asked for subsystems (or explicitly for no
1456 * subsystems) then they must match 1493 * subsystems) then they must match
1457 */ 1494 */
1458 if ((opts->subsys_bits || opts->none) 1495 if ((opts->subsys_mask || opts->none)
1459 && (opts->subsys_bits != root->subsys_bits)) 1496 && (opts->subsys_mask != root->subsys_mask))
1460 return 0; 1497 return 0;
1461 1498
1462 return 1; 1499 return 1;
@@ -1466,7 +1503,7 @@ static struct cgroupfs_root *cgroup_root_from_opts(struct cgroup_sb_opts *opts)
1466{ 1503{
1467 struct cgroupfs_root *root; 1504 struct cgroupfs_root *root;
1468 1505
1469 if (!opts->subsys_bits && !opts->none) 1506 if (!opts->subsys_mask && !opts->none)
1470 return NULL; 1507 return NULL;
1471 1508
1472 root = kzalloc(sizeof(*root), GFP_KERNEL); 1509 root = kzalloc(sizeof(*root), GFP_KERNEL);
@@ -1479,7 +1516,7 @@ static struct cgroupfs_root *cgroup_root_from_opts(struct cgroup_sb_opts *opts)
1479 } 1516 }
1480 init_cgroup_root(root); 1517 init_cgroup_root(root);
1481 1518
1482 root->subsys_bits = opts->subsys_bits; 1519 root->subsys_mask = opts->subsys_mask;
1483 root->flags = opts->flags; 1520 root->flags = opts->flags;
1484 if (opts->release_agent) 1521 if (opts->release_agent)
1485 strcpy(root->release_agent_path, opts->release_agent); 1522 strcpy(root->release_agent_path, opts->release_agent);
@@ -1511,7 +1548,7 @@ static int cgroup_set_super(struct super_block *sb, void *data)
1511 if (!opts->new_root) 1548 if (!opts->new_root)
1512 return -EINVAL; 1549 return -EINVAL;
1513 1550
1514 BUG_ON(!opts->subsys_bits && !opts->none); 1551 BUG_ON(!opts->subsys_mask && !opts->none);
1515 1552
1516 ret = set_anon_super(sb, NULL); 1553 ret = set_anon_super(sb, NULL);
1517 if (ret) 1554 if (ret)
@@ -1629,7 +1666,7 @@ static struct dentry *cgroup_mount(struct file_system_type *fs_type,
1629 if (ret) 1666 if (ret)
1630 goto unlock_drop; 1667 goto unlock_drop;
1631 1668
1632 ret = rebind_subsystems(root, root->subsys_bits); 1669 ret = rebind_subsystems(root, root->subsys_mask);
1633 if (ret == -EBUSY) { 1670 if (ret == -EBUSY) {
1634 free_cg_links(&tmp_cg_links); 1671 free_cg_links(&tmp_cg_links);
1635 goto unlock_drop; 1672 goto unlock_drop;
@@ -1669,7 +1706,7 @@ static struct dentry *cgroup_mount(struct file_system_type *fs_type,
1669 BUG_ON(root->number_of_cgroups != 1); 1706 BUG_ON(root->number_of_cgroups != 1);
1670 1707
1671 cred = override_creds(&init_cred); 1708 cred = override_creds(&init_cred);
1672 cgroup_populate_dir(root_cgrp); 1709 cgroup_populate_dir(root_cgrp, true, root->subsys_mask);
1673 revert_creds(cred); 1710 revert_creds(cred);
1674 mutex_unlock(&cgroup_root_mutex); 1711 mutex_unlock(&cgroup_root_mutex);
1675 mutex_unlock(&cgroup_mutex); 1712 mutex_unlock(&cgroup_mutex);
@@ -1681,7 +1718,7 @@ static struct dentry *cgroup_mount(struct file_system_type *fs_type,
1681 */ 1718 */
1682 cgroup_drop_root(opts.new_root); 1719 cgroup_drop_root(opts.new_root);
1683 /* no subsys rebinding, so refcounts don't change */ 1720 /* no subsys rebinding, so refcounts don't change */
1684 drop_parsed_module_refcounts(opts.subsys_bits); 1721 drop_parsed_module_refcounts(opts.subsys_mask);
1685 } 1722 }
1686 1723
1687 kfree(opts.release_agent); 1724 kfree(opts.release_agent);
@@ -1695,7 +1732,7 @@ static struct dentry *cgroup_mount(struct file_system_type *fs_type,
1695 drop_new_super: 1732 drop_new_super:
1696 deactivate_locked_super(sb); 1733 deactivate_locked_super(sb);
1697 drop_modules: 1734 drop_modules:
1698 drop_parsed_module_refcounts(opts.subsys_bits); 1735 drop_parsed_module_refcounts(opts.subsys_mask);
1699 out_err: 1736 out_err:
1700 kfree(opts.release_agent); 1737 kfree(opts.release_agent);
1701 kfree(opts.name); 1738 kfree(opts.name);
@@ -1745,6 +1782,8 @@ static void cgroup_kill_sb(struct super_block *sb) {
1745 mutex_unlock(&cgroup_root_mutex); 1782 mutex_unlock(&cgroup_root_mutex);
1746 mutex_unlock(&cgroup_mutex); 1783 mutex_unlock(&cgroup_mutex);
1747 1784
1785 simple_xattrs_free(&cgrp->xattrs);
1786
1748 kill_litter_super(sb); 1787 kill_litter_super(sb);
1749 cgroup_drop_root(root); 1788 cgroup_drop_root(root);
1750} 1789}
@@ -2551,6 +2590,64 @@ static int cgroup_rename(struct inode *old_dir, struct dentry *old_dentry,
2551 return simple_rename(old_dir, old_dentry, new_dir, new_dentry); 2590 return simple_rename(old_dir, old_dentry, new_dir, new_dentry);
2552} 2591}
2553 2592
2593static struct simple_xattrs *__d_xattrs(struct dentry *dentry)
2594{
2595 if (S_ISDIR(dentry->d_inode->i_mode))
2596 return &__d_cgrp(dentry)->xattrs;
2597 else
2598 return &__d_cft(dentry)->xattrs;
2599}
2600
2601static inline int xattr_enabled(struct dentry *dentry)
2602{
2603 struct cgroupfs_root *root = dentry->d_sb->s_fs_info;
2604 return test_bit(ROOT_XATTR, &root->flags);
2605}
2606
2607static bool is_valid_xattr(const char *name)
2608{
2609 if (!strncmp(name, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN) ||
2610 !strncmp(name, XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN))
2611 return true;
2612 return false;
2613}
2614
2615static int cgroup_setxattr(struct dentry *dentry, const char *name,
2616 const void *val, size_t size, int flags)
2617{
2618 if (!xattr_enabled(dentry))
2619 return -EOPNOTSUPP;
2620 if (!is_valid_xattr(name))
2621 return -EINVAL;
2622 return simple_xattr_set(__d_xattrs(dentry), name, val, size, flags);
2623}
2624
2625static int cgroup_removexattr(struct dentry *dentry, const char *name)
2626{
2627 if (!xattr_enabled(dentry))
2628 return -EOPNOTSUPP;
2629 if (!is_valid_xattr(name))
2630 return -EINVAL;
2631 return simple_xattr_remove(__d_xattrs(dentry), name);
2632}
2633
2634static ssize_t cgroup_getxattr(struct dentry *dentry, const char *name,
2635 void *buf, size_t size)
2636{
2637 if (!xattr_enabled(dentry))
2638 return -EOPNOTSUPP;
2639 if (!is_valid_xattr(name))
2640 return -EINVAL;
2641 return simple_xattr_get(__d_xattrs(dentry), name, buf, size);
2642}
2643
2644static ssize_t cgroup_listxattr(struct dentry *dentry, char *buf, size_t size)
2645{
2646 if (!xattr_enabled(dentry))
2647 return -EOPNOTSUPP;
2648 return simple_xattr_list(__d_xattrs(dentry), buf, size);
2649}
2650
2554static const struct file_operations cgroup_file_operations = { 2651static const struct file_operations cgroup_file_operations = {
2555 .read = cgroup_file_read, 2652 .read = cgroup_file_read,
2556 .write = cgroup_file_write, 2653 .write = cgroup_file_write,
@@ -2559,11 +2656,22 @@ static const struct file_operations cgroup_file_operations = {
2559 .release = cgroup_file_release, 2656 .release = cgroup_file_release,
2560}; 2657};
2561 2658
2659static const struct inode_operations cgroup_file_inode_operations = {
2660 .setxattr = cgroup_setxattr,
2661 .getxattr = cgroup_getxattr,
2662 .listxattr = cgroup_listxattr,
2663 .removexattr = cgroup_removexattr,
2664};
2665
2562static const struct inode_operations cgroup_dir_inode_operations = { 2666static const struct inode_operations cgroup_dir_inode_operations = {
2563 .lookup = cgroup_lookup, 2667 .lookup = cgroup_lookup,
2564 .mkdir = cgroup_mkdir, 2668 .mkdir = cgroup_mkdir,
2565 .rmdir = cgroup_rmdir, 2669 .rmdir = cgroup_rmdir,
2566 .rename = cgroup_rename, 2670 .rename = cgroup_rename,
2671 .setxattr = cgroup_setxattr,
2672 .getxattr = cgroup_getxattr,
2673 .listxattr = cgroup_listxattr,
2674 .removexattr = cgroup_removexattr,
2567}; 2675};
2568 2676
2569static struct dentry *cgroup_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags) 2677static struct dentry *cgroup_lookup(struct inode *dir, struct dentry *dentry, unsigned int flags)
@@ -2611,6 +2719,7 @@ static int cgroup_create_file(struct dentry *dentry, umode_t mode,
2611 } else if (S_ISREG(mode)) { 2719 } else if (S_ISREG(mode)) {
2612 inode->i_size = 0; 2720 inode->i_size = 0;
2613 inode->i_fop = &cgroup_file_operations; 2721 inode->i_fop = &cgroup_file_operations;
2722 inode->i_op = &cgroup_file_inode_operations;
2614 } 2723 }
2615 d_instantiate(dentry, inode); 2724 d_instantiate(dentry, inode);
2616 dget(dentry); /* Extra count - pin the dentry in core */ 2725 dget(dentry); /* Extra count - pin the dentry in core */
@@ -2671,7 +2780,7 @@ static umode_t cgroup_file_mode(const struct cftype *cft)
2671} 2780}
2672 2781
2673static int cgroup_add_file(struct cgroup *cgrp, struct cgroup_subsys *subsys, 2782static int cgroup_add_file(struct cgroup *cgrp, struct cgroup_subsys *subsys,
2674 const struct cftype *cft) 2783 struct cftype *cft)
2675{ 2784{
2676 struct dentry *dir = cgrp->dentry; 2785 struct dentry *dir = cgrp->dentry;
2677 struct cgroup *parent = __d_cgrp(dir); 2786 struct cgroup *parent = __d_cgrp(dir);
@@ -2681,6 +2790,8 @@ static int cgroup_add_file(struct cgroup *cgrp, struct cgroup_subsys *subsys,
2681 umode_t mode; 2790 umode_t mode;
2682 char name[MAX_CGROUP_TYPE_NAMELEN + MAX_CFTYPE_NAME + 2] = { 0 }; 2791 char name[MAX_CGROUP_TYPE_NAMELEN + MAX_CFTYPE_NAME + 2] = { 0 };
2683 2792
2793 simple_xattrs_init(&cft->xattrs);
2794
2684 /* does @cft->flags tell us to skip creation on @cgrp? */ 2795 /* does @cft->flags tell us to skip creation on @cgrp? */
2685 if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgrp->parent) 2796 if ((cft->flags & CFTYPE_NOT_ON_ROOT) && !cgrp->parent)
2686 return 0; 2797 return 0;
@@ -2721,9 +2832,9 @@ out:
2721} 2832}
2722 2833
2723static int cgroup_addrm_files(struct cgroup *cgrp, struct cgroup_subsys *subsys, 2834static int cgroup_addrm_files(struct cgroup *cgrp, struct cgroup_subsys *subsys,
2724 const struct cftype cfts[], bool is_add) 2835 struct cftype cfts[], bool is_add)
2725{ 2836{
2726 const struct cftype *cft; 2837 struct cftype *cft;
2727 int err, ret = 0; 2838 int err, ret = 0;
2728 2839
2729 for (cft = cfts; cft->name[0] != '\0'; cft++) { 2840 for (cft = cfts; cft->name[0] != '\0'; cft++) {
@@ -2757,7 +2868,7 @@ static void cgroup_cfts_prepare(void)
2757} 2868}
2758 2869
2759static void cgroup_cfts_commit(struct cgroup_subsys *ss, 2870static void cgroup_cfts_commit(struct cgroup_subsys *ss,
2760 const struct cftype *cfts, bool is_add) 2871 struct cftype *cfts, bool is_add)
2761 __releases(&cgroup_mutex) __releases(&cgroup_cft_mutex) 2872 __releases(&cgroup_mutex) __releases(&cgroup_cft_mutex)
2762{ 2873{
2763 LIST_HEAD(pending); 2874 LIST_HEAD(pending);
@@ -2808,7 +2919,7 @@ static void cgroup_cfts_commit(struct cgroup_subsys *ss,
2808 * function currently returns 0 as long as @cfts registration is successful 2919 * function currently returns 0 as long as @cfts registration is successful
2809 * even if some file creation attempts on existing cgroups fail. 2920 * even if some file creation attempts on existing cgroups fail.
2810 */ 2921 */
2811int cgroup_add_cftypes(struct cgroup_subsys *ss, const struct cftype *cfts) 2922int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
2812{ 2923{
2813 struct cftype_set *set; 2924 struct cftype_set *set;
2814 2925
@@ -2838,7 +2949,7 @@ EXPORT_SYMBOL_GPL(cgroup_add_cftypes);
2838 * Returns 0 on successful unregistration, -ENOENT if @cfts is not 2949 * Returns 0 on successful unregistration, -ENOENT if @cfts is not
2839 * registered with @ss. 2950 * registered with @ss.
2840 */ 2951 */
2841int cgroup_rm_cftypes(struct cgroup_subsys *ss, const struct cftype *cfts) 2952int cgroup_rm_cftypes(struct cgroup_subsys *ss, struct cftype *cfts)
2842{ 2953{
2843 struct cftype_set *set; 2954 struct cftype_set *set;
2844 2955
@@ -3843,18 +3954,29 @@ static struct cftype files[] = {
3843 { } /* terminate */ 3954 { } /* terminate */
3844}; 3955};
3845 3956
3846static int cgroup_populate_dir(struct cgroup *cgrp) 3957/**
3958 * cgroup_populate_dir - selectively creation of files in a directory
3959 * @cgrp: target cgroup
3960 * @base_files: true if the base files should be added
3961 * @subsys_mask: mask of the subsystem ids whose files should be added
3962 */
3963static int cgroup_populate_dir(struct cgroup *cgrp, bool base_files,
3964 unsigned long subsys_mask)
3847{ 3965{
3848 int err; 3966 int err;
3849 struct cgroup_subsys *ss; 3967 struct cgroup_subsys *ss;
3850 3968
3851 err = cgroup_addrm_files(cgrp, NULL, files, true); 3969 if (base_files) {
3852 if (err < 0) 3970 err = cgroup_addrm_files(cgrp, NULL, files, true);
3853 return err; 3971 if (err < 0)
3972 return err;
3973 }
3854 3974
3855 /* process cftsets of each subsystem */ 3975 /* process cftsets of each subsystem */
3856 for_each_subsys(cgrp->root, ss) { 3976 for_each_subsys(cgrp->root, ss) {
3857 struct cftype_set *set; 3977 struct cftype_set *set;
3978 if (!test_bit(ss->subsys_id, &subsys_mask))
3979 continue;
3858 3980
3859 list_for_each_entry(set, &ss->cftsets, node) 3981 list_for_each_entry(set, &ss->cftsets, node)
3860 cgroup_addrm_files(cgrp, ss, set->cfts, true); 3982 cgroup_addrm_files(cgrp, ss, set->cfts, true);
@@ -3954,8 +4076,9 @@ static long cgroup_create(struct cgroup *parent, struct dentry *dentry,
3954 set_bit(CGRP_CLONE_CHILDREN, &cgrp->flags); 4076 set_bit(CGRP_CLONE_CHILDREN, &cgrp->flags);
3955 4077
3956 for_each_subsys(root, ss) { 4078 for_each_subsys(root, ss) {
3957 struct cgroup_subsys_state *css = ss->create(cgrp); 4079 struct cgroup_subsys_state *css;
3958 4080
4081 css = ss->create(cgrp);
3959 if (IS_ERR(css)) { 4082 if (IS_ERR(css)) {
3960 err = PTR_ERR(css); 4083 err = PTR_ERR(css);
3961 goto err_destroy; 4084 goto err_destroy;
@@ -3969,6 +4092,15 @@ static long cgroup_create(struct cgroup *parent, struct dentry *dentry,
3969 /* At error, ->destroy() callback has to free assigned ID. */ 4092 /* At error, ->destroy() callback has to free assigned ID. */
3970 if (clone_children(parent) && ss->post_clone) 4093 if (clone_children(parent) && ss->post_clone)
3971 ss->post_clone(cgrp); 4094 ss->post_clone(cgrp);
4095
4096 if (ss->broken_hierarchy && !ss->warned_broken_hierarchy &&
4097 parent->parent) {
4098 pr_warning("cgroup: %s (%d) created nested cgroup for controller \"%s\" which has incomplete hierarchy support. Nested cgroups may change behavior in the future.\n",
4099 current->comm, current->pid, ss->name);
4100 if (!strcmp(ss->name, "memory"))
4101 pr_warning("cgroup: \"memory\" requires setting use_hierarchy to 1 on the root.\n");
4102 ss->warned_broken_hierarchy = true;
4103 }
3972 } 4104 }
3973 4105
3974 list_add(&cgrp->sibling, &cgrp->parent->children); 4106 list_add(&cgrp->sibling, &cgrp->parent->children);
@@ -3988,7 +4120,7 @@ static long cgroup_create(struct cgroup *parent, struct dentry *dentry,
3988 4120
3989 list_add_tail(&cgrp->allcg_node, &root->allcg_list); 4121 list_add_tail(&cgrp->allcg_node, &root->allcg_list);
3990 4122
3991 err = cgroup_populate_dir(cgrp); 4123 err = cgroup_populate_dir(cgrp, true, root->subsys_mask);
3992 /* If err < 0, we have a half-filled directory - oh well ;) */ 4124 /* If err < 0, we have a half-filled directory - oh well ;) */
3993 4125
3994 mutex_unlock(&cgroup_mutex); 4126 mutex_unlock(&cgroup_mutex);
@@ -4321,8 +4453,7 @@ int __init_or_module cgroup_load_subsys(struct cgroup_subsys *ss)
4321 * since cgroup_init_subsys will have already taken care of it. 4453 * since cgroup_init_subsys will have already taken care of it.
4322 */ 4454 */
4323 if (ss->module == NULL) { 4455 if (ss->module == NULL) {
4324 /* a few sanity checks */ 4456 /* a sanity check */
4325 BUG_ON(ss->subsys_id >= CGROUP_BUILTIN_SUBSYS_COUNT);
4326 BUG_ON(subsys[ss->subsys_id] != ss); 4457 BUG_ON(subsys[ss->subsys_id] != ss);
4327 return 0; 4458 return 0;
4328 } 4459 }
@@ -4330,24 +4461,8 @@ int __init_or_module cgroup_load_subsys(struct cgroup_subsys *ss)
4330 /* init base cftset */ 4461 /* init base cftset */
4331 cgroup_init_cftsets(ss); 4462 cgroup_init_cftsets(ss);
4332 4463
4333 /*
4334 * need to register a subsys id before anything else - for example,
4335 * init_cgroup_css needs it.
4336 */
4337 mutex_lock(&cgroup_mutex); 4464 mutex_lock(&cgroup_mutex);
4338 /* find the first empty slot in the array */ 4465 subsys[ss->subsys_id] = ss;
4339 for (i = CGROUP_BUILTIN_SUBSYS_COUNT; i < CGROUP_SUBSYS_COUNT; i++) {
4340 if (subsys[i] == NULL)
4341 break;
4342 }
4343 if (i == CGROUP_SUBSYS_COUNT) {
4344 /* maximum number of subsystems already registered! */
4345 mutex_unlock(&cgroup_mutex);
4346 return -EBUSY;
4347 }
4348 /* assign ourselves the subsys_id */
4349 ss->subsys_id = i;
4350 subsys[i] = ss;
4351 4466
4352 /* 4467 /*
4353 * no ss->create seems to need anything important in the ss struct, so 4468 * no ss->create seems to need anything important in the ss struct, so
@@ -4356,7 +4471,7 @@ int __init_or_module cgroup_load_subsys(struct cgroup_subsys *ss)
4356 css = ss->create(dummytop); 4471 css = ss->create(dummytop);
4357 if (IS_ERR(css)) { 4472 if (IS_ERR(css)) {
4358 /* failure case - need to deassign the subsys[] slot. */ 4473 /* failure case - need to deassign the subsys[] slot. */
4359 subsys[i] = NULL; 4474 subsys[ss->subsys_id] = NULL;
4360 mutex_unlock(&cgroup_mutex); 4475 mutex_unlock(&cgroup_mutex);
4361 return PTR_ERR(css); 4476 return PTR_ERR(css);
4362 } 4477 }
@@ -4372,7 +4487,7 @@ int __init_or_module cgroup_load_subsys(struct cgroup_subsys *ss)
4372 if (ret) { 4487 if (ret) {
4373 dummytop->subsys[ss->subsys_id] = NULL; 4488 dummytop->subsys[ss->subsys_id] = NULL;
4374 ss->destroy(dummytop); 4489 ss->destroy(dummytop);
4375 subsys[i] = NULL; 4490 subsys[ss->subsys_id] = NULL;
4376 mutex_unlock(&cgroup_mutex); 4491 mutex_unlock(&cgroup_mutex);
4377 return ret; 4492 return ret;
4378 } 4493 }
@@ -4439,7 +4554,6 @@ void cgroup_unload_subsys(struct cgroup_subsys *ss)
4439 4554
4440 mutex_lock(&cgroup_mutex); 4555 mutex_lock(&cgroup_mutex);
4441 /* deassign the subsys_id */ 4556 /* deassign the subsys_id */
4442 BUG_ON(ss->subsys_id < CGROUP_BUILTIN_SUBSYS_COUNT);
4443 subsys[ss->subsys_id] = NULL; 4557 subsys[ss->subsys_id] = NULL;
4444 4558
4445 /* remove subsystem from rootnode's list of subsystems */ 4559 /* remove subsystem from rootnode's list of subsystems */
@@ -4502,10 +4616,13 @@ int __init cgroup_init_early(void)
4502 for (i = 0; i < CSS_SET_TABLE_SIZE; i++) 4616 for (i = 0; i < CSS_SET_TABLE_SIZE; i++)
4503 INIT_HLIST_HEAD(&css_set_table[i]); 4617 INIT_HLIST_HEAD(&css_set_table[i]);
4504 4618
4505 /* at bootup time, we don't worry about modular subsystems */ 4619 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
4506 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4507 struct cgroup_subsys *ss = subsys[i]; 4620 struct cgroup_subsys *ss = subsys[i];
4508 4621
4622 /* at bootup time, we don't worry about modular subsystems */
4623 if (!ss || ss->module)
4624 continue;
4625
4509 BUG_ON(!ss->name); 4626 BUG_ON(!ss->name);
4510 BUG_ON(strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN); 4627 BUG_ON(strlen(ss->name) > MAX_CGROUP_TYPE_NAMELEN);
4511 BUG_ON(!ss->create); 4628 BUG_ON(!ss->create);
@@ -4538,9 +4655,12 @@ int __init cgroup_init(void)
4538 if (err) 4655 if (err)
4539 return err; 4656 return err;
4540 4657
4541 /* at bootup time, we don't worry about modular subsystems */ 4658 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
4542 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4543 struct cgroup_subsys *ss = subsys[i]; 4659 struct cgroup_subsys *ss = subsys[i];
4660
4661 /* at bootup time, we don't worry about modular subsystems */
4662 if (!ss || ss->module)
4663 continue;
4544 if (!ss->early_init) 4664 if (!ss->early_init)
4545 cgroup_init_subsys(ss); 4665 cgroup_init_subsys(ss);
4546 if (ss->use_id) 4666 if (ss->use_id)
@@ -4735,13 +4855,16 @@ void cgroup_fork_callbacks(struct task_struct *child)
4735{ 4855{
4736 if (need_forkexit_callback) { 4856 if (need_forkexit_callback) {
4737 int i; 4857 int i;
4738 /* 4858 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
4739 * forkexit callbacks are only supported for builtin
4740 * subsystems, and the builtin section of the subsys array is
4741 * immutable, so we don't need to lock the subsys array here.
4742 */
4743 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4744 struct cgroup_subsys *ss = subsys[i]; 4859 struct cgroup_subsys *ss = subsys[i];
4860
4861 /*
4862 * forkexit callbacks are only supported for
4863 * builtin subsystems.
4864 */
4865 if (!ss || ss->module)
4866 continue;
4867
4745 if (ss->fork) 4868 if (ss->fork)
4746 ss->fork(child); 4869 ss->fork(child);
4747 } 4870 }
@@ -4846,12 +4969,13 @@ void cgroup_exit(struct task_struct *tsk, int run_callbacks)
4846 tsk->cgroups = &init_css_set; 4969 tsk->cgroups = &init_css_set;
4847 4970
4848 if (run_callbacks && need_forkexit_callback) { 4971 if (run_callbacks && need_forkexit_callback) {
4849 /* 4972 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
4850 * modular subsystems can't use callbacks, so no need to lock
4851 * the subsys array
4852 */
4853 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
4854 struct cgroup_subsys *ss = subsys[i]; 4973 struct cgroup_subsys *ss = subsys[i];
4974
4975 /* modular subsystems can't use callbacks */
4976 if (!ss || ss->module)
4977 continue;
4978
4855 if (ss->exit) { 4979 if (ss->exit) {
4856 struct cgroup *old_cgrp = 4980 struct cgroup *old_cgrp =
4857 rcu_dereference_raw(cg->subsys[i])->cgroup; 4981 rcu_dereference_raw(cg->subsys[i])->cgroup;
@@ -5037,13 +5161,17 @@ static int __init cgroup_disable(char *str)
5037 while ((token = strsep(&str, ",")) != NULL) { 5161 while ((token = strsep(&str, ",")) != NULL) {
5038 if (!*token) 5162 if (!*token)
5039 continue; 5163 continue;
5040 /* 5164 for (i = 0; i < CGROUP_SUBSYS_COUNT; i++) {
5041 * cgroup_disable, being at boot time, can't know about module
5042 * subsystems, so we don't worry about them.
5043 */
5044 for (i = 0; i < CGROUP_BUILTIN_SUBSYS_COUNT; i++) {
5045 struct cgroup_subsys *ss = subsys[i]; 5165 struct cgroup_subsys *ss = subsys[i];
5046 5166
5167 /*
5168 * cgroup_disable, being at boot time, can't
5169 * know about module subsystems, so we don't
5170 * worry about them.
5171 */
5172 if (!ss || ss->module)
5173 continue;
5174
5047 if (!strcmp(token, ss->name)) { 5175 if (!strcmp(token, ss->name)) {
5048 ss->disabled = 1; 5176 ss->disabled = 1;
5049 printk(KERN_INFO "Disabling %s control group" 5177 printk(KERN_INFO "Disabling %s control group"
diff --git a/kernel/cgroup_freezer.c b/kernel/cgroup_freezer.c
index 3649fc6b3eaa..b1724ce98981 100644
--- a/kernel/cgroup_freezer.c
+++ b/kernel/cgroup_freezer.c
@@ -373,4 +373,12 @@ struct cgroup_subsys freezer_subsys = {
373 .can_attach = freezer_can_attach, 373 .can_attach = freezer_can_attach,
374 .fork = freezer_fork, 374 .fork = freezer_fork,
375 .base_cftypes = files, 375 .base_cftypes = files,
376
377 /*
378 * freezer subsys doesn't handle hierarchy at all. Frozen state
379 * should be inherited through the hierarchy - if a parent is
380 * frozen, all its children should be frozen. Fix it and remove
381 * the following.
382 */
383 .broken_hierarchy = true,
376}; 384};
diff --git a/kernel/cpu.c b/kernel/cpu.c
index 14d32588cccd..42bd331ee0ab 100644
--- a/kernel/cpu.c
+++ b/kernel/cpu.c
@@ -80,6 +80,10 @@ void put_online_cpus(void)
80 if (cpu_hotplug.active_writer == current) 80 if (cpu_hotplug.active_writer == current)
81 return; 81 return;
82 mutex_lock(&cpu_hotplug.lock); 82 mutex_lock(&cpu_hotplug.lock);
83
84 if (WARN_ON(!cpu_hotplug.refcount))
85 cpu_hotplug.refcount++; /* try to fix things up */
86
83 if (!--cpu_hotplug.refcount && unlikely(cpu_hotplug.active_writer)) 87 if (!--cpu_hotplug.refcount && unlikely(cpu_hotplug.active_writer))
84 wake_up_process(cpu_hotplug.active_writer); 88 wake_up_process(cpu_hotplug.active_writer);
85 mutex_unlock(&cpu_hotplug.lock); 89 mutex_unlock(&cpu_hotplug.lock);
@@ -280,12 +284,13 @@ static int __ref _cpu_down(unsigned int cpu, int tasks_frozen)
280 __func__, cpu); 284 __func__, cpu);
281 goto out_release; 285 goto out_release;
282 } 286 }
287 smpboot_park_threads(cpu);
283 288
284 err = __stop_machine(take_cpu_down, &tcd_param, cpumask_of(cpu)); 289 err = __stop_machine(take_cpu_down, &tcd_param, cpumask_of(cpu));
285 if (err) { 290 if (err) {
286 /* CPU didn't die: tell everyone. Can't complain. */ 291 /* CPU didn't die: tell everyone. Can't complain. */
292 smpboot_unpark_threads(cpu);
287 cpu_notify_nofail(CPU_DOWN_FAILED | mod, hcpu); 293 cpu_notify_nofail(CPU_DOWN_FAILED | mod, hcpu);
288
289 goto out_release; 294 goto out_release;
290 } 295 }
291 BUG_ON(cpu_online(cpu)); 296 BUG_ON(cpu_online(cpu));
@@ -354,6 +359,10 @@ static int __cpuinit _cpu_up(unsigned int cpu, int tasks_frozen)
354 goto out; 359 goto out;
355 } 360 }
356 361
362 ret = smpboot_create_threads(cpu);
363 if (ret)
364 goto out;
365
357 ret = __cpu_notify(CPU_UP_PREPARE | mod, hcpu, -1, &nr_calls); 366 ret = __cpu_notify(CPU_UP_PREPARE | mod, hcpu, -1, &nr_calls);
358 if (ret) { 367 if (ret) {
359 nr_calls--; 368 nr_calls--;
@@ -368,6 +377,9 @@ static int __cpuinit _cpu_up(unsigned int cpu, int tasks_frozen)
368 goto out_notify; 377 goto out_notify;
369 BUG_ON(!cpu_online(cpu)); 378 BUG_ON(!cpu_online(cpu));
370 379
380 /* Wake the per cpu threads */
381 smpboot_unpark_threads(cpu);
382
371 /* Now call notifier in preparation. */ 383 /* Now call notifier in preparation. */
372 cpu_notify(CPU_ONLINE | mod, hcpu); 384 cpu_notify(CPU_ONLINE | mod, hcpu);
373 385
@@ -439,14 +451,6 @@ EXPORT_SYMBOL_GPL(cpu_up);
439#ifdef CONFIG_PM_SLEEP_SMP 451#ifdef CONFIG_PM_SLEEP_SMP
440static cpumask_var_t frozen_cpus; 452static cpumask_var_t frozen_cpus;
441 453
442void __weak arch_disable_nonboot_cpus_begin(void)
443{
444}
445
446void __weak arch_disable_nonboot_cpus_end(void)
447{
448}
449
450int disable_nonboot_cpus(void) 454int disable_nonboot_cpus(void)
451{ 455{
452 int cpu, first_cpu, error = 0; 456 int cpu, first_cpu, error = 0;
@@ -458,7 +462,6 @@ int disable_nonboot_cpus(void)
458 * with the userspace trying to use the CPU hotplug at the same time 462 * with the userspace trying to use the CPU hotplug at the same time
459 */ 463 */
460 cpumask_clear(frozen_cpus); 464 cpumask_clear(frozen_cpus);
461 arch_disable_nonboot_cpus_begin();
462 465
463 printk("Disabling non-boot CPUs ...\n"); 466 printk("Disabling non-boot CPUs ...\n");
464 for_each_online_cpu(cpu) { 467 for_each_online_cpu(cpu) {
@@ -474,8 +477,6 @@ int disable_nonboot_cpus(void)
474 } 477 }
475 } 478 }
476 479
477 arch_disable_nonboot_cpus_end();
478
479 if (!error) { 480 if (!error) {
480 BUG_ON(num_online_cpus() > 1); 481 BUG_ON(num_online_cpus() > 1);
481 /* Make sure the CPUs won't be enabled by someone else */ 482 /* Make sure the CPUs won't be enabled by someone else */
diff --git a/kernel/cred.c b/kernel/cred.c
index de728ac50d82..48cea3da6d05 100644
--- a/kernel/cred.c
+++ b/kernel/cred.c
@@ -799,9 +799,15 @@ static void dump_invalid_creds(const struct cred *cred, const char *label,
799 atomic_read(&cred->usage), 799 atomic_read(&cred->usage),
800 read_cred_subscribers(cred)); 800 read_cred_subscribers(cred));
801 printk(KERN_ERR "CRED: ->*uid = { %d,%d,%d,%d }\n", 801 printk(KERN_ERR "CRED: ->*uid = { %d,%d,%d,%d }\n",
802 cred->uid, cred->euid, cred->suid, cred->fsuid); 802 from_kuid_munged(&init_user_ns, cred->uid),
803 from_kuid_munged(&init_user_ns, cred->euid),
804 from_kuid_munged(&init_user_ns, cred->suid),
805 from_kuid_munged(&init_user_ns, cred->fsuid));
803 printk(KERN_ERR "CRED: ->*gid = { %d,%d,%d,%d }\n", 806 printk(KERN_ERR "CRED: ->*gid = { %d,%d,%d,%d }\n",
804 cred->gid, cred->egid, cred->sgid, cred->fsgid); 807 from_kgid_munged(&init_user_ns, cred->gid),
808 from_kgid_munged(&init_user_ns, cred->egid),
809 from_kgid_munged(&init_user_ns, cred->sgid),
810 from_kgid_munged(&init_user_ns, cred->fsgid));
805#ifdef CONFIG_SECURITY 811#ifdef CONFIG_SECURITY
806 printk(KERN_ERR "CRED: ->security is %p\n", cred->security); 812 printk(KERN_ERR "CRED: ->security is %p\n", cred->security);
807 if ((unsigned long) cred->security >= PAGE_SIZE && 813 if ((unsigned long) cred->security >= PAGE_SIZE &&
diff --git a/kernel/debug/debug_core.c b/kernel/debug/debug_core.c
index 0557f24c6bca..9a61738cefc8 100644
--- a/kernel/debug/debug_core.c
+++ b/kernel/debug/debug_core.c
@@ -672,6 +672,10 @@ kgdb_handle_exception(int evector, int signo, int ecode, struct pt_regs *regs)
672{ 672{
673 struct kgdb_state kgdb_var; 673 struct kgdb_state kgdb_var;
674 struct kgdb_state *ks = &kgdb_var; 674 struct kgdb_state *ks = &kgdb_var;
675 int ret = 0;
676
677 if (arch_kgdb_ops.enable_nmi)
678 arch_kgdb_ops.enable_nmi(0);
675 679
676 ks->cpu = raw_smp_processor_id(); 680 ks->cpu = raw_smp_processor_id();
677 ks->ex_vector = evector; 681 ks->ex_vector = evector;
@@ -681,13 +685,33 @@ kgdb_handle_exception(int evector, int signo, int ecode, struct pt_regs *regs)
681 ks->linux_regs = regs; 685 ks->linux_regs = regs;
682 686
683 if (kgdb_reenter_check(ks)) 687 if (kgdb_reenter_check(ks))
684 return 0; /* Ouch, double exception ! */ 688 goto out; /* Ouch, double exception ! */
685 if (kgdb_info[ks->cpu].enter_kgdb != 0) 689 if (kgdb_info[ks->cpu].enter_kgdb != 0)
686 return 0; 690 goto out;
687 691
688 return kgdb_cpu_enter(ks, regs, DCPU_WANT_MASTER); 692 ret = kgdb_cpu_enter(ks, regs, DCPU_WANT_MASTER);
693out:
694 if (arch_kgdb_ops.enable_nmi)
695 arch_kgdb_ops.enable_nmi(1);
696 return ret;
689} 697}
690 698
699/*
700 * GDB places a breakpoint at this function to know dynamically
701 * loaded objects. It's not defined static so that only one instance with this
702 * name exists in the kernel.
703 */
704
705static int module_event(struct notifier_block *self, unsigned long val,
706 void *data)
707{
708 return 0;
709}
710
711static struct notifier_block dbg_module_load_nb = {
712 .notifier_call = module_event,
713};
714
691int kgdb_nmicallback(int cpu, void *regs) 715int kgdb_nmicallback(int cpu, void *regs)
692{ 716{
693#ifdef CONFIG_SMP 717#ifdef CONFIG_SMP
@@ -816,6 +840,7 @@ static void kgdb_register_callbacks(void)
816 kgdb_arch_init(); 840 kgdb_arch_init();
817 if (!dbg_is_early) 841 if (!dbg_is_early)
818 kgdb_arch_late(); 842 kgdb_arch_late();
843 register_module_notifier(&dbg_module_load_nb);
819 register_reboot_notifier(&dbg_reboot_notifier); 844 register_reboot_notifier(&dbg_reboot_notifier);
820 atomic_notifier_chain_register(&panic_notifier_list, 845 atomic_notifier_chain_register(&panic_notifier_list,
821 &kgdb_panic_event_nb); 846 &kgdb_panic_event_nb);
@@ -839,6 +864,7 @@ static void kgdb_unregister_callbacks(void)
839 if (kgdb_io_module_registered) { 864 if (kgdb_io_module_registered) {
840 kgdb_io_module_registered = 0; 865 kgdb_io_module_registered = 0;
841 unregister_reboot_notifier(&dbg_reboot_notifier); 866 unregister_reboot_notifier(&dbg_reboot_notifier);
867 unregister_module_notifier(&dbg_module_load_nb);
842 atomic_notifier_chain_unregister(&panic_notifier_list, 868 atomic_notifier_chain_unregister(&panic_notifier_list,
843 &kgdb_panic_event_nb); 869 &kgdb_panic_event_nb);
844 kgdb_arch_exit(); 870 kgdb_arch_exit();
diff --git a/kernel/debug/kdb/kdb_bt.c b/kernel/debug/kdb/kdb_bt.c
index 07c9bbb94a0b..b03e0e814e43 100644
--- a/kernel/debug/kdb/kdb_bt.c
+++ b/kernel/debug/kdb/kdb_bt.c
@@ -129,6 +129,8 @@ kdb_bt(int argc, const char **argv)
129 } 129 }
130 /* Now the inactive tasks */ 130 /* Now the inactive tasks */
131 kdb_do_each_thread(g, p) { 131 kdb_do_each_thread(g, p) {
132 if (KDB_FLAG(CMD_INTERRUPT))
133 return 0;
132 if (task_curr(p)) 134 if (task_curr(p))
133 continue; 135 continue;
134 if (kdb_bt1(p, mask, argcount, btaprompt)) 136 if (kdb_bt1(p, mask, argcount, btaprompt))
diff --git a/kernel/debug/kdb/kdb_io.c b/kernel/debug/kdb/kdb_io.c
index 0a69d2adc4f3..14ff4849262c 100644
--- a/kernel/debug/kdb/kdb_io.c
+++ b/kernel/debug/kdb/kdb_io.c
@@ -552,6 +552,7 @@ int vkdb_printf(const char *fmt, va_list ap)
552{ 552{
553 int diag; 553 int diag;
554 int linecount; 554 int linecount;
555 int colcount;
555 int logging, saved_loglevel = 0; 556 int logging, saved_loglevel = 0;
556 int saved_trap_printk; 557 int saved_trap_printk;
557 int got_printf_lock = 0; 558 int got_printf_lock = 0;
@@ -584,6 +585,10 @@ int vkdb_printf(const char *fmt, va_list ap)
584 if (diag || linecount <= 1) 585 if (diag || linecount <= 1)
585 linecount = 24; 586 linecount = 24;
586 587
588 diag = kdbgetintenv("COLUMNS", &colcount);
589 if (diag || colcount <= 1)
590 colcount = 80;
591
587 diag = kdbgetintenv("LOGGING", &logging); 592 diag = kdbgetintenv("LOGGING", &logging);
588 if (diag) 593 if (diag)
589 logging = 0; 594 logging = 0;
@@ -690,7 +695,7 @@ kdb_printit:
690 gdbstub_msg_write(kdb_buffer, retlen); 695 gdbstub_msg_write(kdb_buffer, retlen);
691 } else { 696 } else {
692 if (dbg_io_ops && !dbg_io_ops->is_console) { 697 if (dbg_io_ops && !dbg_io_ops->is_console) {
693 len = strlen(kdb_buffer); 698 len = retlen;
694 cp = kdb_buffer; 699 cp = kdb_buffer;
695 while (len--) { 700 while (len--) {
696 dbg_io_ops->write_char(*cp); 701 dbg_io_ops->write_char(*cp);
@@ -709,11 +714,29 @@ kdb_printit:
709 printk(KERN_INFO "%s", kdb_buffer); 714 printk(KERN_INFO "%s", kdb_buffer);
710 } 715 }
711 716
712 if (KDB_STATE(PAGER) && strchr(kdb_buffer, '\n')) 717 if (KDB_STATE(PAGER)) {
713 kdb_nextline++; 718 /*
719 * Check printed string to decide how to bump the
720 * kdb_nextline to control when the more prompt should
721 * show up.
722 */
723 int got = 0;
724 len = retlen;
725 while (len--) {
726 if (kdb_buffer[len] == '\n') {
727 kdb_nextline++;
728 got = 0;
729 } else if (kdb_buffer[len] == '\r') {
730 got = 0;
731 } else {
732 got++;
733 }
734 }
735 kdb_nextline += got / (colcount + 1);
736 }
714 737
715 /* check for having reached the LINES number of printed lines */ 738 /* check for having reached the LINES number of printed lines */
716 if (kdb_nextline == linecount) { 739 if (kdb_nextline >= linecount) {
717 char buf1[16] = ""; 740 char buf1[16] = "";
718 741
719 /* Watch out for recursion here. Any routine that calls 742 /* Watch out for recursion here. Any routine that calls
@@ -765,7 +788,7 @@ kdb_printit:
765 kdb_grepping_flag = 0; 788 kdb_grepping_flag = 0;
766 kdb_printf("\n"); 789 kdb_printf("\n");
767 } else if (buf1[0] == ' ') { 790 } else if (buf1[0] == ' ') {
768 kdb_printf("\n"); 791 kdb_printf("\r");
769 suspend_grep = 1; /* for this recursion */ 792 suspend_grep = 1; /* for this recursion */
770 } else if (buf1[0] == '\n') { 793 } else if (buf1[0] == '\n') {
771 kdb_nextline = linecount - 1; 794 kdb_nextline = linecount - 1;
diff --git a/kernel/debug/kdb/kdb_main.c b/kernel/debug/kdb/kdb_main.c
index 31df1706b9a9..4d5f8d5612f3 100644
--- a/kernel/debug/kdb/kdb_main.c
+++ b/kernel/debug/kdb/kdb_main.c
@@ -21,6 +21,7 @@
21#include <linux/smp.h> 21#include <linux/smp.h>
22#include <linux/utsname.h> 22#include <linux/utsname.h>
23#include <linux/vmalloc.h> 23#include <linux/vmalloc.h>
24#include <linux/atomic.h>
24#include <linux/module.h> 25#include <linux/module.h>
25#include <linux/mm.h> 26#include <linux/mm.h>
26#include <linux/init.h> 27#include <linux/init.h>
@@ -2100,6 +2101,8 @@ static int kdb_dmesg(int argc, const char **argv)
2100 } 2101 }
2101 if (!lines--) 2102 if (!lines--)
2102 break; 2103 break;
2104 if (KDB_FLAG(CMD_INTERRUPT))
2105 return 0;
2103 2106
2104 kdb_printf("%.*s\n", (int)len - 1, buf); 2107 kdb_printf("%.*s\n", (int)len - 1, buf);
2105 } 2108 }
@@ -2107,6 +2110,32 @@ static int kdb_dmesg(int argc, const char **argv)
2107 return 0; 2110 return 0;
2108} 2111}
2109#endif /* CONFIG_PRINTK */ 2112#endif /* CONFIG_PRINTK */
2113
2114/* Make sure we balance enable/disable calls, must disable first. */
2115static atomic_t kdb_nmi_disabled;
2116
2117static int kdb_disable_nmi(int argc, const char *argv[])
2118{
2119 if (atomic_read(&kdb_nmi_disabled))
2120 return 0;
2121 atomic_set(&kdb_nmi_disabled, 1);
2122 arch_kgdb_ops.enable_nmi(0);
2123 return 0;
2124}
2125
2126static int kdb_param_enable_nmi(const char *val, const struct kernel_param *kp)
2127{
2128 if (!atomic_add_unless(&kdb_nmi_disabled, -1, 0))
2129 return -EINVAL;
2130 arch_kgdb_ops.enable_nmi(1);
2131 return 0;
2132}
2133
2134static const struct kernel_param_ops kdb_param_ops_enable_nmi = {
2135 .set = kdb_param_enable_nmi,
2136};
2137module_param_cb(enable_nmi, &kdb_param_ops_enable_nmi, NULL, 0600);
2138
2110/* 2139/*
2111 * kdb_cpu - This function implements the 'cpu' command. 2140 * kdb_cpu - This function implements the 'cpu' command.
2112 * cpu [<cpunum>] 2141 * cpu [<cpunum>]
@@ -2851,6 +2880,10 @@ static void __init kdb_inittab(void)
2851 kdb_register_repeat("dmesg", kdb_dmesg, "[lines]", 2880 kdb_register_repeat("dmesg", kdb_dmesg, "[lines]",
2852 "Display syslog buffer", 0, KDB_REPEAT_NONE); 2881 "Display syslog buffer", 0, KDB_REPEAT_NONE);
2853#endif 2882#endif
2883 if (arch_kgdb_ops.enable_nmi) {
2884 kdb_register_repeat("disable_nmi", kdb_disable_nmi, "",
2885 "Disable NMI entry to KDB", 0, KDB_REPEAT_NONE);
2886 }
2854 kdb_register_repeat("defcmd", kdb_defcmd, "name \"usage\" \"help\"", 2887 kdb_register_repeat("defcmd", kdb_defcmd, "name \"usage\" \"help\"",
2855 "Define a set of commands, down to endefcmd", 0, KDB_REPEAT_NONE); 2888 "Define a set of commands, down to endefcmd", 0, KDB_REPEAT_NONE);
2856 kdb_register_repeat("kill", kdb_kill, "<-signal> <pid>", 2889 kdb_register_repeat("kill", kdb_kill, "<-signal> <pid>",
diff --git a/kernel/events/callchain.c b/kernel/events/callchain.c
index 98d4597f43d6..c77206184b8b 100644
--- a/kernel/events/callchain.c
+++ b/kernel/events/callchain.c
@@ -159,6 +159,11 @@ perf_callchain(struct perf_event *event, struct pt_regs *regs)
159 int rctx; 159 int rctx;
160 struct perf_callchain_entry *entry; 160 struct perf_callchain_entry *entry;
161 161
162 int kernel = !event->attr.exclude_callchain_kernel;
163 int user = !event->attr.exclude_callchain_user;
164
165 if (!kernel && !user)
166 return NULL;
162 167
163 entry = get_callchain_entry(&rctx); 168 entry = get_callchain_entry(&rctx);
164 if (rctx == -1) 169 if (rctx == -1)
@@ -169,24 +174,29 @@ perf_callchain(struct perf_event *event, struct pt_regs *regs)
169 174
170 entry->nr = 0; 175 entry->nr = 0;
171 176
172 if (!user_mode(regs)) { 177 if (kernel && !user_mode(regs)) {
173 perf_callchain_store(entry, PERF_CONTEXT_KERNEL); 178 perf_callchain_store(entry, PERF_CONTEXT_KERNEL);
174 perf_callchain_kernel(entry, regs); 179 perf_callchain_kernel(entry, regs);
175 if (current->mm)
176 regs = task_pt_regs(current);
177 else
178 regs = NULL;
179 } 180 }
180 181
181 if (regs) { 182 if (user) {
182 /* 183 if (!user_mode(regs)) {
183 * Disallow cross-task user callchains. 184 if (current->mm)
184 */ 185 regs = task_pt_regs(current);
185 if (event->ctx->task && event->ctx->task != current) 186 else
186 goto exit_put; 187 regs = NULL;
187 188 }
188 perf_callchain_store(entry, PERF_CONTEXT_USER); 189
189 perf_callchain_user(entry, regs); 190 if (regs) {
191 /*
192 * Disallow cross-task user callchains.
193 */
194 if (event->ctx->task && event->ctx->task != current)
195 goto exit_put;
196
197 perf_callchain_store(entry, PERF_CONTEXT_USER);
198 perf_callchain_user(entry, regs);
199 }
190 } 200 }
191 201
192exit_put: 202exit_put:
diff --git a/kernel/events/core.c b/kernel/events/core.c
index 7fee567153f0..dbccf83c134d 100644
--- a/kernel/events/core.c
+++ b/kernel/events/core.c
@@ -36,6 +36,7 @@
36#include <linux/perf_event.h> 36#include <linux/perf_event.h>
37#include <linux/ftrace_event.h> 37#include <linux/ftrace_event.h>
38#include <linux/hw_breakpoint.h> 38#include <linux/hw_breakpoint.h>
39#include <linux/mm_types.h>
39 40
40#include "internal.h" 41#include "internal.h"
41 42
@@ -371,6 +372,8 @@ void perf_cgroup_switch(struct task_struct *task, int mode)
371 372
372 list_for_each_entry_rcu(pmu, &pmus, entry) { 373 list_for_each_entry_rcu(pmu, &pmus, entry) {
373 cpuctx = this_cpu_ptr(pmu->pmu_cpu_context); 374 cpuctx = this_cpu_ptr(pmu->pmu_cpu_context);
375 if (cpuctx->unique_pmu != pmu)
376 continue; /* ensure we process each cpuctx once */
374 377
375 /* 378 /*
376 * perf_cgroup_events says at least one 379 * perf_cgroup_events says at least one
@@ -394,9 +397,10 @@ void perf_cgroup_switch(struct task_struct *task, int mode)
394 397
395 if (mode & PERF_CGROUP_SWIN) { 398 if (mode & PERF_CGROUP_SWIN) {
396 WARN_ON_ONCE(cpuctx->cgrp); 399 WARN_ON_ONCE(cpuctx->cgrp);
397 /* set cgrp before ctxsw in to 400 /*
398 * allow event_filter_match() to not 401 * set cgrp before ctxsw in to allow
399 * have to pass task around 402 * event_filter_match() to not have to pass
403 * task around
400 */ 404 */
401 cpuctx->cgrp = perf_cgroup_from_task(task); 405 cpuctx->cgrp = perf_cgroup_from_task(task);
402 cpu_ctx_sched_in(cpuctx, EVENT_ALL, task); 406 cpu_ctx_sched_in(cpuctx, EVENT_ALL, task);
@@ -467,14 +471,13 @@ static inline int perf_cgroup_connect(int fd, struct perf_event *event,
467{ 471{
468 struct perf_cgroup *cgrp; 472 struct perf_cgroup *cgrp;
469 struct cgroup_subsys_state *css; 473 struct cgroup_subsys_state *css;
470 struct file *file; 474 struct fd f = fdget(fd);
471 int ret = 0, fput_needed; 475 int ret = 0;
472 476
473 file = fget_light(fd, &fput_needed); 477 if (!f.file)
474 if (!file)
475 return -EBADF; 478 return -EBADF;
476 479
477 css = cgroup_css_from_dir(file, perf_subsys_id); 480 css = cgroup_css_from_dir(f.file, perf_subsys_id);
478 if (IS_ERR(css)) { 481 if (IS_ERR(css)) {
479 ret = PTR_ERR(css); 482 ret = PTR_ERR(css);
480 goto out; 483 goto out;
@@ -500,7 +503,7 @@ static inline int perf_cgroup_connect(int fd, struct perf_event *event,
500 ret = -EINVAL; 503 ret = -EINVAL;
501 } 504 }
502out: 505out:
503 fput_light(file, fput_needed); 506 fdput(f);
504 return ret; 507 return ret;
505} 508}
506 509
@@ -3233,21 +3236,18 @@ unlock:
3233 3236
3234static const struct file_operations perf_fops; 3237static const struct file_operations perf_fops;
3235 3238
3236static struct file *perf_fget_light(int fd, int *fput_needed) 3239static inline int perf_fget_light(int fd, struct fd *p)
3237{ 3240{
3238 struct file *file; 3241 struct fd f = fdget(fd);
3239 3242 if (!f.file)
3240 file = fget_light(fd, fput_needed); 3243 return -EBADF;
3241 if (!file)
3242 return ERR_PTR(-EBADF);
3243 3244
3244 if (file->f_op != &perf_fops) { 3245 if (f.file->f_op != &perf_fops) {
3245 fput_light(file, *fput_needed); 3246 fdput(f);
3246 *fput_needed = 0; 3247 return -EBADF;
3247 return ERR_PTR(-EBADF);
3248 } 3248 }
3249 3249 *p = f;
3250 return file; 3250 return 0;
3251} 3251}
3252 3252
3253static int perf_event_set_output(struct perf_event *event, 3253static int perf_event_set_output(struct perf_event *event,
@@ -3279,22 +3279,19 @@ static long perf_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
3279 3279
3280 case PERF_EVENT_IOC_SET_OUTPUT: 3280 case PERF_EVENT_IOC_SET_OUTPUT:
3281 { 3281 {
3282 struct file *output_file = NULL;
3283 struct perf_event *output_event = NULL;
3284 int fput_needed = 0;
3285 int ret; 3282 int ret;
3286
3287 if (arg != -1) { 3283 if (arg != -1) {
3288 output_file = perf_fget_light(arg, &fput_needed); 3284 struct perf_event *output_event;
3289 if (IS_ERR(output_file)) 3285 struct fd output;
3290 return PTR_ERR(output_file); 3286 ret = perf_fget_light(arg, &output);
3291 output_event = output_file->private_data; 3287 if (ret)
3288 return ret;
3289 output_event = output.file->private_data;
3290 ret = perf_event_set_output(event, output_event);
3291 fdput(output);
3292 } else {
3293 ret = perf_event_set_output(event, NULL);
3292 } 3294 }
3293
3294 ret = perf_event_set_output(event, output_event);
3295 if (output_event)
3296 fput_light(output_file, fput_needed);
3297
3298 return ret; 3295 return ret;
3299 } 3296 }
3300 3297
@@ -3677,7 +3674,7 @@ unlock:
3677 atomic_inc(&event->mmap_count); 3674 atomic_inc(&event->mmap_count);
3678 mutex_unlock(&event->mmap_mutex); 3675 mutex_unlock(&event->mmap_mutex);
3679 3676
3680 vma->vm_flags |= VM_RESERVED; 3677 vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
3681 vma->vm_ops = &perf_mmap_vmops; 3678 vma->vm_ops = &perf_mmap_vmops;
3682 3679
3683 return ret; 3680 return ret;
@@ -3764,6 +3761,132 @@ int perf_unregister_guest_info_callbacks(struct perf_guest_info_callbacks *cbs)
3764} 3761}
3765EXPORT_SYMBOL_GPL(perf_unregister_guest_info_callbacks); 3762EXPORT_SYMBOL_GPL(perf_unregister_guest_info_callbacks);
3766 3763
3764static void
3765perf_output_sample_regs(struct perf_output_handle *handle,
3766 struct pt_regs *regs, u64 mask)
3767{
3768 int bit;
3769
3770 for_each_set_bit(bit, (const unsigned long *) &mask,
3771 sizeof(mask) * BITS_PER_BYTE) {
3772 u64 val;
3773
3774 val = perf_reg_value(regs, bit);
3775 perf_output_put(handle, val);
3776 }
3777}
3778
3779static void perf_sample_regs_user(struct perf_regs_user *regs_user,
3780 struct pt_regs *regs)
3781{
3782 if (!user_mode(regs)) {
3783 if (current->mm)
3784 regs = task_pt_regs(current);
3785 else
3786 regs = NULL;
3787 }
3788
3789 if (regs) {
3790 regs_user->regs = regs;
3791 regs_user->abi = perf_reg_abi(current);
3792 }
3793}
3794
3795/*
3796 * Get remaining task size from user stack pointer.
3797 *
3798 * It'd be better to take stack vma map and limit this more
3799 * precisly, but there's no way to get it safely under interrupt,
3800 * so using TASK_SIZE as limit.
3801 */
3802static u64 perf_ustack_task_size(struct pt_regs *regs)
3803{
3804 unsigned long addr = perf_user_stack_pointer(regs);
3805
3806 if (!addr || addr >= TASK_SIZE)
3807 return 0;
3808
3809 return TASK_SIZE - addr;
3810}
3811
3812static u16
3813perf_sample_ustack_size(u16 stack_size, u16 header_size,
3814 struct pt_regs *regs)
3815{
3816 u64 task_size;
3817
3818 /* No regs, no stack pointer, no dump. */
3819 if (!regs)
3820 return 0;
3821
3822 /*
3823 * Check if we fit in with the requested stack size into the:
3824 * - TASK_SIZE
3825 * If we don't, we limit the size to the TASK_SIZE.
3826 *
3827 * - remaining sample size
3828 * If we don't, we customize the stack size to
3829 * fit in to the remaining sample size.
3830 */
3831
3832 task_size = min((u64) USHRT_MAX, perf_ustack_task_size(regs));
3833 stack_size = min(stack_size, (u16) task_size);
3834
3835 /* Current header size plus static size and dynamic size. */
3836 header_size += 2 * sizeof(u64);
3837
3838 /* Do we fit in with the current stack dump size? */
3839 if ((u16) (header_size + stack_size) < header_size) {
3840 /*
3841 * If we overflow the maximum size for the sample,
3842 * we customize the stack dump size to fit in.
3843 */
3844 stack_size = USHRT_MAX - header_size - sizeof(u64);
3845 stack_size = round_up(stack_size, sizeof(u64));
3846 }
3847
3848 return stack_size;
3849}
3850
3851static void
3852perf_output_sample_ustack(struct perf_output_handle *handle, u64 dump_size,
3853 struct pt_regs *regs)
3854{
3855 /* Case of a kernel thread, nothing to dump */
3856 if (!regs) {
3857 u64 size = 0;
3858 perf_output_put(handle, size);
3859 } else {
3860 unsigned long sp;
3861 unsigned int rem;
3862 u64 dyn_size;
3863
3864 /*
3865 * We dump:
3866 * static size
3867 * - the size requested by user or the best one we can fit
3868 * in to the sample max size
3869 * data
3870 * - user stack dump data
3871 * dynamic size
3872 * - the actual dumped size
3873 */
3874
3875 /* Static size. */
3876 perf_output_put(handle, dump_size);
3877
3878 /* Data. */
3879 sp = perf_user_stack_pointer(regs);
3880 rem = __output_copy_user(handle, (void *) sp, dump_size);
3881 dyn_size = dump_size - rem;
3882
3883 perf_output_skip(handle, rem);
3884
3885 /* Dynamic size. */
3886 perf_output_put(handle, dyn_size);
3887 }
3888}
3889
3767static void __perf_event_header__init_id(struct perf_event_header *header, 3890static void __perf_event_header__init_id(struct perf_event_header *header,
3768 struct perf_sample_data *data, 3891 struct perf_sample_data *data,
3769 struct perf_event *event) 3892 struct perf_event *event)
@@ -4024,6 +4147,28 @@ void perf_output_sample(struct perf_output_handle *handle,
4024 perf_output_put(handle, nr); 4147 perf_output_put(handle, nr);
4025 } 4148 }
4026 } 4149 }
4150
4151 if (sample_type & PERF_SAMPLE_REGS_USER) {
4152 u64 abi = data->regs_user.abi;
4153
4154 /*
4155 * If there are no regs to dump, notice it through
4156 * first u64 being zero (PERF_SAMPLE_REGS_ABI_NONE).
4157 */
4158 perf_output_put(handle, abi);
4159
4160 if (abi) {
4161 u64 mask = event->attr.sample_regs_user;
4162 perf_output_sample_regs(handle,
4163 data->regs_user.regs,
4164 mask);
4165 }
4166 }
4167
4168 if (sample_type & PERF_SAMPLE_STACK_USER)
4169 perf_output_sample_ustack(handle,
4170 data->stack_user_size,
4171 data->regs_user.regs);
4027} 4172}
4028 4173
4029void perf_prepare_sample(struct perf_event_header *header, 4174void perf_prepare_sample(struct perf_event_header *header,
@@ -4075,6 +4220,49 @@ void perf_prepare_sample(struct perf_event_header *header,
4075 } 4220 }
4076 header->size += size; 4221 header->size += size;
4077 } 4222 }
4223
4224 if (sample_type & PERF_SAMPLE_REGS_USER) {
4225 /* regs dump ABI info */
4226 int size = sizeof(u64);
4227
4228 perf_sample_regs_user(&data->regs_user, regs);
4229
4230 if (data->regs_user.regs) {
4231 u64 mask = event->attr.sample_regs_user;
4232 size += hweight64(mask) * sizeof(u64);
4233 }
4234
4235 header->size += size;
4236 }
4237
4238 if (sample_type & PERF_SAMPLE_STACK_USER) {
4239 /*
4240 * Either we need PERF_SAMPLE_STACK_USER bit to be allways
4241 * processed as the last one or have additional check added
4242 * in case new sample type is added, because we could eat
4243 * up the rest of the sample size.
4244 */
4245 struct perf_regs_user *uregs = &data->regs_user;
4246 u16 stack_size = event->attr.sample_stack_user;
4247 u16 size = sizeof(u64);
4248
4249 if (!uregs->abi)
4250 perf_sample_regs_user(uregs, regs);
4251
4252 stack_size = perf_sample_ustack_size(stack_size, header->size,
4253 uregs->regs);
4254
4255 /*
4256 * If there is something to dump, add space for the dump
4257 * itself and for the field that tells the dynamic size,
4258 * which is how many have been actually dumped.
4259 */
4260 if (stack_size)
4261 size += sizeof(u64) + stack_size;
4262
4263 data->stack_user_size = stack_size;
4264 header->size += size;
4265 }
4078} 4266}
4079 4267
4080static void perf_event_output(struct perf_event *event, 4268static void perf_event_output(struct perf_event *event,
@@ -4227,7 +4415,7 @@ static void perf_event_task_event(struct perf_task_event *task_event)
4227 rcu_read_lock(); 4415 rcu_read_lock();
4228 list_for_each_entry_rcu(pmu, &pmus, entry) { 4416 list_for_each_entry_rcu(pmu, &pmus, entry) {
4229 cpuctx = get_cpu_ptr(pmu->pmu_cpu_context); 4417 cpuctx = get_cpu_ptr(pmu->pmu_cpu_context);
4230 if (cpuctx->active_pmu != pmu) 4418 if (cpuctx->unique_pmu != pmu)
4231 goto next; 4419 goto next;
4232 perf_event_task_ctx(&cpuctx->ctx, task_event); 4420 perf_event_task_ctx(&cpuctx->ctx, task_event);
4233 4421
@@ -4373,7 +4561,7 @@ static void perf_event_comm_event(struct perf_comm_event *comm_event)
4373 rcu_read_lock(); 4561 rcu_read_lock();
4374 list_for_each_entry_rcu(pmu, &pmus, entry) { 4562 list_for_each_entry_rcu(pmu, &pmus, entry) {
4375 cpuctx = get_cpu_ptr(pmu->pmu_cpu_context); 4563 cpuctx = get_cpu_ptr(pmu->pmu_cpu_context);
4376 if (cpuctx->active_pmu != pmu) 4564 if (cpuctx->unique_pmu != pmu)
4377 goto next; 4565 goto next;
4378 perf_event_comm_ctx(&cpuctx->ctx, comm_event); 4566 perf_event_comm_ctx(&cpuctx->ctx, comm_event);
4379 4567
@@ -4569,7 +4757,7 @@ got_name:
4569 rcu_read_lock(); 4757 rcu_read_lock();
4570 list_for_each_entry_rcu(pmu, &pmus, entry) { 4758 list_for_each_entry_rcu(pmu, &pmus, entry) {
4571 cpuctx = get_cpu_ptr(pmu->pmu_cpu_context); 4759 cpuctx = get_cpu_ptr(pmu->pmu_cpu_context);
4572 if (cpuctx->active_pmu != pmu) 4760 if (cpuctx->unique_pmu != pmu)
4573 goto next; 4761 goto next;
4574 perf_event_mmap_ctx(&cpuctx->ctx, mmap_event, 4762 perf_event_mmap_ctx(&cpuctx->ctx, mmap_event,
4575 vma->vm_flags & VM_EXEC); 4763 vma->vm_flags & VM_EXEC);
@@ -5670,8 +5858,8 @@ static void update_pmu_context(struct pmu *pmu, struct pmu *old_pmu)
5670 5858
5671 cpuctx = per_cpu_ptr(pmu->pmu_cpu_context, cpu); 5859 cpuctx = per_cpu_ptr(pmu->pmu_cpu_context, cpu);
5672 5860
5673 if (cpuctx->active_pmu == old_pmu) 5861 if (cpuctx->unique_pmu == old_pmu)
5674 cpuctx->active_pmu = pmu; 5862 cpuctx->unique_pmu = pmu;
5675 } 5863 }
5676} 5864}
5677 5865
@@ -5806,7 +5994,7 @@ skip_type:
5806 cpuctx->ctx.pmu = pmu; 5994 cpuctx->ctx.pmu = pmu;
5807 cpuctx->jiffies_interval = 1; 5995 cpuctx->jiffies_interval = 1;
5808 INIT_LIST_HEAD(&cpuctx->rotation_list); 5996 INIT_LIST_HEAD(&cpuctx->rotation_list);
5809 cpuctx->active_pmu = pmu; 5997 cpuctx->unique_pmu = pmu;
5810 } 5998 }
5811 5999
5812got_cpu_context: 6000got_cpu_context:
@@ -6151,6 +6339,28 @@ static int perf_copy_attr(struct perf_event_attr __user *uattr,
6151 attr->branch_sample_type = mask; 6339 attr->branch_sample_type = mask;
6152 } 6340 }
6153 } 6341 }
6342
6343 if (attr->sample_type & PERF_SAMPLE_REGS_USER) {
6344 ret = perf_reg_validate(attr->sample_regs_user);
6345 if (ret)
6346 return ret;
6347 }
6348
6349 if (attr->sample_type & PERF_SAMPLE_STACK_USER) {
6350 if (!arch_perf_have_user_stack_dump())
6351 return -ENOSYS;
6352
6353 /*
6354 * We have __u32 type for the size, but so far
6355 * we can only use __u16 as maximum due to the
6356 * __u16 sample size limit.
6357 */
6358 if (attr->sample_stack_user >= USHRT_MAX)
6359 ret = -EINVAL;
6360 else if (!IS_ALIGNED(attr->sample_stack_user, sizeof(u64)))
6361 ret = -EINVAL;
6362 }
6363
6154out: 6364out:
6155 return ret; 6365 return ret;
6156 6366
@@ -6229,12 +6439,11 @@ SYSCALL_DEFINE5(perf_event_open,
6229 struct perf_event_attr attr; 6439 struct perf_event_attr attr;
6230 struct perf_event_context *ctx; 6440 struct perf_event_context *ctx;
6231 struct file *event_file = NULL; 6441 struct file *event_file = NULL;
6232 struct file *group_file = NULL; 6442 struct fd group = {NULL, 0};
6233 struct task_struct *task = NULL; 6443 struct task_struct *task = NULL;
6234 struct pmu *pmu; 6444 struct pmu *pmu;
6235 int event_fd; 6445 int event_fd;
6236 int move_group = 0; 6446 int move_group = 0;
6237 int fput_needed = 0;
6238 int err; 6447 int err;
6239 6448
6240 /* for future expandability... */ 6449 /* for future expandability... */
@@ -6264,17 +6473,15 @@ SYSCALL_DEFINE5(perf_event_open,
6264 if ((flags & PERF_FLAG_PID_CGROUP) && (pid == -1 || cpu == -1)) 6473 if ((flags & PERF_FLAG_PID_CGROUP) && (pid == -1 || cpu == -1))
6265 return -EINVAL; 6474 return -EINVAL;
6266 6475
6267 event_fd = get_unused_fd_flags(O_RDWR); 6476 event_fd = get_unused_fd();
6268 if (event_fd < 0) 6477 if (event_fd < 0)
6269 return event_fd; 6478 return event_fd;
6270 6479
6271 if (group_fd != -1) { 6480 if (group_fd != -1) {
6272 group_file = perf_fget_light(group_fd, &fput_needed); 6481 err = perf_fget_light(group_fd, &group);
6273 if (IS_ERR(group_file)) { 6482 if (err)
6274 err = PTR_ERR(group_file);
6275 goto err_fd; 6483 goto err_fd;
6276 } 6484 group_leader = group.file->private_data;
6277 group_leader = group_file->private_data;
6278 if (flags & PERF_FLAG_FD_OUTPUT) 6485 if (flags & PERF_FLAG_FD_OUTPUT)
6279 output_event = group_leader; 6486 output_event = group_leader;
6280 if (flags & PERF_FLAG_FD_NO_GROUP) 6487 if (flags & PERF_FLAG_FD_NO_GROUP)
@@ -6450,7 +6657,7 @@ SYSCALL_DEFINE5(perf_event_open,
6450 * of the group leader will find the pointer to itself in 6657 * of the group leader will find the pointer to itself in
6451 * perf_group_detach(). 6658 * perf_group_detach().
6452 */ 6659 */
6453 fput_light(group_file, fput_needed); 6660 fdput(group);
6454 fd_install(event_fd, event_file); 6661 fd_install(event_fd, event_file);
6455 return event_fd; 6662 return event_fd;
6456 6663
@@ -6464,7 +6671,7 @@ err_task:
6464 if (task) 6671 if (task)
6465 put_task_struct(task); 6672 put_task_struct(task);
6466err_group_fd: 6673err_group_fd:
6467 fput_light(group_file, fput_needed); 6674 fdput(group);
6468err_fd: 6675err_fd:
6469 put_unused_fd(event_fd); 6676 put_unused_fd(event_fd);
6470 return err; 6677 return err;
@@ -7289,5 +7496,12 @@ struct cgroup_subsys perf_subsys = {
7289 .destroy = perf_cgroup_destroy, 7496 .destroy = perf_cgroup_destroy,
7290 .exit = perf_cgroup_exit, 7497 .exit = perf_cgroup_exit,
7291 .attach = perf_cgroup_attach, 7498 .attach = perf_cgroup_attach,
7499
7500 /*
7501 * perf_event cgroup doesn't handle nesting correctly.
7502 * ctx->nr_cgroups adjustments should be propagated through the
7503 * cgroup hierarchy. Fix it and remove the following.
7504 */
7505 .broken_hierarchy = true,
7292}; 7506};
7293#endif /* CONFIG_CGROUP_PERF */ 7507#endif /* CONFIG_CGROUP_PERF */
diff --git a/kernel/events/internal.h b/kernel/events/internal.h
index a096c19f2c2a..d56a64c99a8b 100644
--- a/kernel/events/internal.h
+++ b/kernel/events/internal.h
@@ -2,6 +2,7 @@
2#define _KERNEL_EVENTS_INTERNAL_H 2#define _KERNEL_EVENTS_INTERNAL_H
3 3
4#include <linux/hardirq.h> 4#include <linux/hardirq.h>
5#include <linux/uaccess.h>
5 6
6/* Buffer handling */ 7/* Buffer handling */
7 8
@@ -76,30 +77,53 @@ static inline unsigned long perf_data_size(struct ring_buffer *rb)
76 return rb->nr_pages << (PAGE_SHIFT + page_order(rb)); 77 return rb->nr_pages << (PAGE_SHIFT + page_order(rb));
77} 78}
78 79
79static inline void 80#define DEFINE_OUTPUT_COPY(func_name, memcpy_func) \
80__output_copy(struct perf_output_handle *handle, 81static inline unsigned int \
81 const void *buf, unsigned int len) 82func_name(struct perf_output_handle *handle, \
83 const void *buf, unsigned int len) \
84{ \
85 unsigned long size, written; \
86 \
87 do { \
88 size = min_t(unsigned long, handle->size, len); \
89 \
90 written = memcpy_func(handle->addr, buf, size); \
91 \
92 len -= written; \
93 handle->addr += written; \
94 buf += written; \
95 handle->size -= written; \
96 if (!handle->size) { \
97 struct ring_buffer *rb = handle->rb; \
98 \
99 handle->page++; \
100 handle->page &= rb->nr_pages - 1; \
101 handle->addr = rb->data_pages[handle->page]; \
102 handle->size = PAGE_SIZE << page_order(rb); \
103 } \
104 } while (len && written == size); \
105 \
106 return len; \
107}
108
109static inline int memcpy_common(void *dst, const void *src, size_t n)
82{ 110{
83 do { 111 memcpy(dst, src, n);
84 unsigned long size = min_t(unsigned long, handle->size, len); 112 return n;
85
86 memcpy(handle->addr, buf, size);
87
88 len -= size;
89 handle->addr += size;
90 buf += size;
91 handle->size -= size;
92 if (!handle->size) {
93 struct ring_buffer *rb = handle->rb;
94
95 handle->page++;
96 handle->page &= rb->nr_pages - 1;
97 handle->addr = rb->data_pages[handle->page];
98 handle->size = PAGE_SIZE << page_order(rb);
99 }
100 } while (len);
101} 113}
102 114
115DEFINE_OUTPUT_COPY(__output_copy, memcpy_common)
116
117#define MEMCPY_SKIP(dst, src, n) (n)
118
119DEFINE_OUTPUT_COPY(__output_skip, MEMCPY_SKIP)
120
121#ifndef arch_perf_out_copy_user
122#define arch_perf_out_copy_user __copy_from_user_inatomic
123#endif
124
125DEFINE_OUTPUT_COPY(__output_copy_user, arch_perf_out_copy_user)
126
103/* Callchain handling */ 127/* Callchain handling */
104extern struct perf_callchain_entry * 128extern struct perf_callchain_entry *
105perf_callchain(struct perf_event *event, struct pt_regs *regs); 129perf_callchain(struct perf_event *event, struct pt_regs *regs);
@@ -134,4 +158,20 @@ static inline void put_recursion_context(int *recursion, int rctx)
134 recursion[rctx]--; 158 recursion[rctx]--;
135} 159}
136 160
161#ifdef CONFIG_HAVE_PERF_USER_STACK_DUMP
162static inline bool arch_perf_have_user_stack_dump(void)
163{
164 return true;
165}
166
167#define perf_user_stack_pointer(regs) user_stack_pointer(regs)
168#else
169static inline bool arch_perf_have_user_stack_dump(void)
170{
171 return false;
172}
173
174#define perf_user_stack_pointer(regs) 0
175#endif /* CONFIG_HAVE_PERF_USER_STACK_DUMP */
176
137#endif /* _KERNEL_EVENTS_INTERNAL_H */ 177#endif /* _KERNEL_EVENTS_INTERNAL_H */
diff --git a/kernel/events/ring_buffer.c b/kernel/events/ring_buffer.c
index 6ddaba43fb7a..23cb34ff3973 100644
--- a/kernel/events/ring_buffer.c
+++ b/kernel/events/ring_buffer.c
@@ -182,10 +182,16 @@ out:
182 return -ENOSPC; 182 return -ENOSPC;
183} 183}
184 184
185void perf_output_copy(struct perf_output_handle *handle, 185unsigned int perf_output_copy(struct perf_output_handle *handle,
186 const void *buf, unsigned int len) 186 const void *buf, unsigned int len)
187{ 187{
188 __output_copy(handle, buf, len); 188 return __output_copy(handle, buf, len);
189}
190
191unsigned int perf_output_skip(struct perf_output_handle *handle,
192 unsigned int len)
193{
194 return __output_skip(handle, NULL, len);
189} 195}
190 196
191void perf_output_end(struct perf_output_handle *handle) 197void perf_output_end(struct perf_output_handle *handle)
diff --git a/kernel/events/uprobes.c b/kernel/events/uprobes.c
index c08a22d02f72..98256bc71ee1 100644
--- a/kernel/events/uprobes.c
+++ b/kernel/events/uprobes.c
@@ -141,10 +141,14 @@ static int __replace_page(struct vm_area_struct *vma, unsigned long addr,
141 spinlock_t *ptl; 141 spinlock_t *ptl;
142 pte_t *ptep; 142 pte_t *ptep;
143 int err; 143 int err;
144 /* For mmu_notifiers */
145 const unsigned long mmun_start = addr;
146 const unsigned long mmun_end = addr + PAGE_SIZE;
144 147
145 /* For try_to_free_swap() and munlock_vma_page() below */ 148 /* For try_to_free_swap() and munlock_vma_page() below */
146 lock_page(page); 149 lock_page(page);
147 150
151 mmu_notifier_invalidate_range_start(mm, mmun_start, mmun_end);
148 err = -EAGAIN; 152 err = -EAGAIN;
149 ptep = page_check_address(page, mm, addr, &ptl, 0); 153 ptep = page_check_address(page, mm, addr, &ptl, 0);
150 if (!ptep) 154 if (!ptep)
@@ -173,6 +177,7 @@ static int __replace_page(struct vm_area_struct *vma, unsigned long addr,
173 177
174 err = 0; 178 err = 0;
175 unlock: 179 unlock:
180 mmu_notifier_invalidate_range_end(mm, mmun_start, mmun_end);
176 unlock_page(page); 181 unlock_page(page);
177 return err; 182 return err;
178} 183}
@@ -280,12 +285,10 @@ static int read_opcode(struct mm_struct *mm, unsigned long vaddr, uprobe_opcode_
280 if (ret <= 0) 285 if (ret <= 0)
281 return ret; 286 return ret;
282 287
283 lock_page(page);
284 vaddr_new = kmap_atomic(page); 288 vaddr_new = kmap_atomic(page);
285 vaddr &= ~PAGE_MASK; 289 vaddr &= ~PAGE_MASK;
286 memcpy(opcode, vaddr_new + vaddr, UPROBE_SWBP_INSN_SIZE); 290 memcpy(opcode, vaddr_new + vaddr, UPROBE_SWBP_INSN_SIZE);
287 kunmap_atomic(vaddr_new); 291 kunmap_atomic(vaddr_new);
288 unlock_page(page);
289 292
290 put_page(page); 293 put_page(page);
291 294
@@ -334,7 +337,7 @@ int __weak set_swbp(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned
334 */ 337 */
335 result = is_swbp_at_addr(mm, vaddr); 338 result = is_swbp_at_addr(mm, vaddr);
336 if (result == 1) 339 if (result == 1)
337 return -EEXIST; 340 return 0;
338 341
339 if (result) 342 if (result)
340 return result; 343 return result;
@@ -347,24 +350,22 @@ int __weak set_swbp(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned
347 * @mm: the probed process address space. 350 * @mm: the probed process address space.
348 * @auprobe: arch specific probepoint information. 351 * @auprobe: arch specific probepoint information.
349 * @vaddr: the virtual address to insert the opcode. 352 * @vaddr: the virtual address to insert the opcode.
350 * @verify: if true, verify existance of breakpoint instruction.
351 * 353 *
352 * For mm @mm, restore the original opcode (opcode) at @vaddr. 354 * For mm @mm, restore the original opcode (opcode) at @vaddr.
353 * Return 0 (success) or a negative errno. 355 * Return 0 (success) or a negative errno.
354 */ 356 */
355int __weak 357int __weak
356set_orig_insn(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr, bool verify) 358set_orig_insn(struct arch_uprobe *auprobe, struct mm_struct *mm, unsigned long vaddr)
357{ 359{
358 if (verify) { 360 int result;
359 int result; 361
362 result = is_swbp_at_addr(mm, vaddr);
363 if (!result)
364 return -EINVAL;
360 365
361 result = is_swbp_at_addr(mm, vaddr); 366 if (result != 1)
362 if (!result) 367 return result;
363 return -EINVAL;
364 368
365 if (result != 1)
366 return result;
367 }
368 return write_opcode(auprobe, mm, vaddr, *(uprobe_opcode_t *)auprobe->insn); 369 return write_opcode(auprobe, mm, vaddr, *(uprobe_opcode_t *)auprobe->insn);
369} 370}
370 371
@@ -415,11 +416,10 @@ static struct uprobe *__find_uprobe(struct inode *inode, loff_t offset)
415static struct uprobe *find_uprobe(struct inode *inode, loff_t offset) 416static struct uprobe *find_uprobe(struct inode *inode, loff_t offset)
416{ 417{
417 struct uprobe *uprobe; 418 struct uprobe *uprobe;
418 unsigned long flags;
419 419
420 spin_lock_irqsave(&uprobes_treelock, flags); 420 spin_lock(&uprobes_treelock);
421 uprobe = __find_uprobe(inode, offset); 421 uprobe = __find_uprobe(inode, offset);
422 spin_unlock_irqrestore(&uprobes_treelock, flags); 422 spin_unlock(&uprobes_treelock);
423 423
424 return uprobe; 424 return uprobe;
425} 425}
@@ -466,12 +466,11 @@ static struct uprobe *__insert_uprobe(struct uprobe *uprobe)
466 */ 466 */
467static struct uprobe *insert_uprobe(struct uprobe *uprobe) 467static struct uprobe *insert_uprobe(struct uprobe *uprobe)
468{ 468{
469 unsigned long flags;
470 struct uprobe *u; 469 struct uprobe *u;
471 470
472 spin_lock_irqsave(&uprobes_treelock, flags); 471 spin_lock(&uprobes_treelock);
473 u = __insert_uprobe(uprobe); 472 u = __insert_uprobe(uprobe);
474 spin_unlock_irqrestore(&uprobes_treelock, flags); 473 spin_unlock(&uprobes_treelock);
475 474
476 /* For now assume that the instruction need not be single-stepped */ 475 /* For now assume that the instruction need not be single-stepped */
477 uprobe->flags |= UPROBE_SKIP_SSTEP; 476 uprobe->flags |= UPROBE_SKIP_SSTEP;
@@ -649,6 +648,7 @@ static int
649install_breakpoint(struct uprobe *uprobe, struct mm_struct *mm, 648install_breakpoint(struct uprobe *uprobe, struct mm_struct *mm,
650 struct vm_area_struct *vma, unsigned long vaddr) 649 struct vm_area_struct *vma, unsigned long vaddr)
651{ 650{
651 bool first_uprobe;
652 int ret; 652 int ret;
653 653
654 /* 654 /*
@@ -659,7 +659,7 @@ install_breakpoint(struct uprobe *uprobe, struct mm_struct *mm,
659 * Hence behave as if probe already existed. 659 * Hence behave as if probe already existed.
660 */ 660 */
661 if (!uprobe->consumers) 661 if (!uprobe->consumers)
662 return -EEXIST; 662 return 0;
663 663
664 if (!(uprobe->flags & UPROBE_COPY_INSN)) { 664 if (!(uprobe->flags & UPROBE_COPY_INSN)) {
665 ret = copy_insn(uprobe, vma->vm_file); 665 ret = copy_insn(uprobe, vma->vm_file);
@@ -681,17 +681,18 @@ install_breakpoint(struct uprobe *uprobe, struct mm_struct *mm,
681 } 681 }
682 682
683 /* 683 /*
684 * Ideally, should be updating the probe count after the breakpoint 684 * set MMF_HAS_UPROBES in advance for uprobe_pre_sstep_notifier(),
685 * has been successfully inserted. However a thread could hit the 685 * the task can hit this breakpoint right after __replace_page().
686 * breakpoint we just inserted even before the probe count is
687 * incremented. If this is the first breakpoint placed, breakpoint
688 * notifier might ignore uprobes and pass the trap to the thread.
689 * Hence increment before and decrement on failure.
690 */ 686 */
691 atomic_inc(&mm->uprobes_state.count); 687 first_uprobe = !test_bit(MMF_HAS_UPROBES, &mm->flags);
688 if (first_uprobe)
689 set_bit(MMF_HAS_UPROBES, &mm->flags);
690
692 ret = set_swbp(&uprobe->arch, mm, vaddr); 691 ret = set_swbp(&uprobe->arch, mm, vaddr);
693 if (ret) 692 if (!ret)
694 atomic_dec(&mm->uprobes_state.count); 693 clear_bit(MMF_RECALC_UPROBES, &mm->flags);
694 else if (first_uprobe)
695 clear_bit(MMF_HAS_UPROBES, &mm->flags);
695 696
696 return ret; 697 return ret;
697} 698}
@@ -699,8 +700,12 @@ install_breakpoint(struct uprobe *uprobe, struct mm_struct *mm,
699static void 700static void
700remove_breakpoint(struct uprobe *uprobe, struct mm_struct *mm, unsigned long vaddr) 701remove_breakpoint(struct uprobe *uprobe, struct mm_struct *mm, unsigned long vaddr)
701{ 702{
702 if (!set_orig_insn(&uprobe->arch, mm, vaddr, true)) 703 /* can happen if uprobe_register() fails */
703 atomic_dec(&mm->uprobes_state.count); 704 if (!test_bit(MMF_HAS_UPROBES, &mm->flags))
705 return;
706
707 set_bit(MMF_RECALC_UPROBES, &mm->flags);
708 set_orig_insn(&uprobe->arch, mm, vaddr);
704} 709}
705 710
706/* 711/*
@@ -710,11 +715,9 @@ remove_breakpoint(struct uprobe *uprobe, struct mm_struct *mm, unsigned long vad
710 */ 715 */
711static void delete_uprobe(struct uprobe *uprobe) 716static void delete_uprobe(struct uprobe *uprobe)
712{ 717{
713 unsigned long flags; 718 spin_lock(&uprobes_treelock);
714
715 spin_lock_irqsave(&uprobes_treelock, flags);
716 rb_erase(&uprobe->rb_node, &uprobes_tree); 719 rb_erase(&uprobe->rb_node, &uprobes_tree);
717 spin_unlock_irqrestore(&uprobes_treelock, flags); 720 spin_unlock(&uprobes_treelock);
718 iput(uprobe->inode); 721 iput(uprobe->inode);
719 put_uprobe(uprobe); 722 put_uprobe(uprobe);
720 atomic_dec(&uprobe_events); 723 atomic_dec(&uprobe_events);
@@ -737,7 +740,6 @@ static struct map_info *
737build_map_info(struct address_space *mapping, loff_t offset, bool is_register) 740build_map_info(struct address_space *mapping, loff_t offset, bool is_register)
738{ 741{
739 unsigned long pgoff = offset >> PAGE_SHIFT; 742 unsigned long pgoff = offset >> PAGE_SHIFT;
740 struct prio_tree_iter iter;
741 struct vm_area_struct *vma; 743 struct vm_area_struct *vma;
742 struct map_info *curr = NULL; 744 struct map_info *curr = NULL;
743 struct map_info *prev = NULL; 745 struct map_info *prev = NULL;
@@ -746,7 +748,7 @@ build_map_info(struct address_space *mapping, loff_t offset, bool is_register)
746 748
747 again: 749 again:
748 mutex_lock(&mapping->i_mmap_mutex); 750 mutex_lock(&mapping->i_mmap_mutex);
749 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) { 751 vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) {
750 if (!valid_vma(vma, is_register)) 752 if (!valid_vma(vma, is_register))
751 continue; 753 continue;
752 754
@@ -831,17 +833,11 @@ static int register_for_each_vma(struct uprobe *uprobe, bool is_register)
831 vaddr_to_offset(vma, info->vaddr) != uprobe->offset) 833 vaddr_to_offset(vma, info->vaddr) != uprobe->offset)
832 goto unlock; 834 goto unlock;
833 835
834 if (is_register) { 836 if (is_register)
835 err = install_breakpoint(uprobe, mm, vma, info->vaddr); 837 err = install_breakpoint(uprobe, mm, vma, info->vaddr);
836 /* 838 else
837 * We can race against uprobe_mmap(), see the
838 * comment near uprobe_hash().
839 */
840 if (err == -EEXIST)
841 err = 0;
842 } else {
843 remove_breakpoint(uprobe, mm, info->vaddr); 839 remove_breakpoint(uprobe, mm, info->vaddr);
844 } 840
845 unlock: 841 unlock:
846 up_write(&mm->mmap_sem); 842 up_write(&mm->mmap_sem);
847 free: 843 free:
@@ -908,7 +904,8 @@ int uprobe_register(struct inode *inode, loff_t offset, struct uprobe_consumer *
908 } 904 }
909 905
910 mutex_unlock(uprobes_hash(inode)); 906 mutex_unlock(uprobes_hash(inode));
911 put_uprobe(uprobe); 907 if (uprobe)
908 put_uprobe(uprobe);
912 909
913 return ret; 910 return ret;
914} 911}
@@ -978,7 +975,6 @@ static void build_probe_list(struct inode *inode,
978 struct list_head *head) 975 struct list_head *head)
979{ 976{
980 loff_t min, max; 977 loff_t min, max;
981 unsigned long flags;
982 struct rb_node *n, *t; 978 struct rb_node *n, *t;
983 struct uprobe *u; 979 struct uprobe *u;
984 980
@@ -986,7 +982,7 @@ static void build_probe_list(struct inode *inode,
986 min = vaddr_to_offset(vma, start); 982 min = vaddr_to_offset(vma, start);
987 max = min + (end - start) - 1; 983 max = min + (end - start) - 1;
988 984
989 spin_lock_irqsave(&uprobes_treelock, flags); 985 spin_lock(&uprobes_treelock);
990 n = find_node_in_range(inode, min, max); 986 n = find_node_in_range(inode, min, max);
991 if (n) { 987 if (n) {
992 for (t = n; t; t = rb_prev(t)) { 988 for (t = n; t; t = rb_prev(t)) {
@@ -1004,27 +1000,20 @@ static void build_probe_list(struct inode *inode,
1004 atomic_inc(&u->ref); 1000 atomic_inc(&u->ref);
1005 } 1001 }
1006 } 1002 }
1007 spin_unlock_irqrestore(&uprobes_treelock, flags); 1003 spin_unlock(&uprobes_treelock);
1008} 1004}
1009 1005
1010/* 1006/*
1011 * Called from mmap_region. 1007 * Called from mmap_region/vma_adjust with mm->mmap_sem acquired.
1012 * called with mm->mmap_sem acquired.
1013 * 1008 *
1014 * Return -ve no if we fail to insert probes and we cannot 1009 * Currently we ignore all errors and always return 0, the callers
1015 * bail-out. 1010 * can't handle the failure anyway.
1016 * Return 0 otherwise. i.e:
1017 *
1018 * - successful insertion of probes
1019 * - (or) no possible probes to be inserted.
1020 * - (or) insertion of probes failed but we can bail-out.
1021 */ 1011 */
1022int uprobe_mmap(struct vm_area_struct *vma) 1012int uprobe_mmap(struct vm_area_struct *vma)
1023{ 1013{
1024 struct list_head tmp_list; 1014 struct list_head tmp_list;
1025 struct uprobe *uprobe, *u; 1015 struct uprobe *uprobe, *u;
1026 struct inode *inode; 1016 struct inode *inode;
1027 int ret, count;
1028 1017
1029 if (!atomic_read(&uprobe_events) || !valid_vma(vma, true)) 1018 if (!atomic_read(&uprobe_events) || !valid_vma(vma, true))
1030 return 0; 1019 return 0;
@@ -1036,44 +1025,35 @@ int uprobe_mmap(struct vm_area_struct *vma)
1036 mutex_lock(uprobes_mmap_hash(inode)); 1025 mutex_lock(uprobes_mmap_hash(inode));
1037 build_probe_list(inode, vma, vma->vm_start, vma->vm_end, &tmp_list); 1026 build_probe_list(inode, vma, vma->vm_start, vma->vm_end, &tmp_list);
1038 1027
1039 ret = 0;
1040 count = 0;
1041
1042 list_for_each_entry_safe(uprobe, u, &tmp_list, pending_list) { 1028 list_for_each_entry_safe(uprobe, u, &tmp_list, pending_list) {
1043 if (!ret) { 1029 if (!fatal_signal_pending(current)) {
1044 unsigned long vaddr = offset_to_vaddr(vma, uprobe->offset); 1030 unsigned long vaddr = offset_to_vaddr(vma, uprobe->offset);
1045 1031 install_breakpoint(uprobe, vma->vm_mm, vma, vaddr);
1046 ret = install_breakpoint(uprobe, vma->vm_mm, vma, vaddr);
1047 /*
1048 * We can race against uprobe_register(), see the
1049 * comment near uprobe_hash().
1050 */
1051 if (ret == -EEXIST) {
1052 ret = 0;
1053
1054 if (!is_swbp_at_addr(vma->vm_mm, vaddr))
1055 continue;
1056
1057 /*
1058 * Unable to insert a breakpoint, but
1059 * breakpoint lies underneath. Increment the
1060 * probe count.
1061 */
1062 atomic_inc(&vma->vm_mm->uprobes_state.count);
1063 }
1064
1065 if (!ret)
1066 count++;
1067 } 1032 }
1068 put_uprobe(uprobe); 1033 put_uprobe(uprobe);
1069 } 1034 }
1070
1071 mutex_unlock(uprobes_mmap_hash(inode)); 1035 mutex_unlock(uprobes_mmap_hash(inode));
1072 1036
1073 if (ret) 1037 return 0;
1074 atomic_sub(count, &vma->vm_mm->uprobes_state.count); 1038}
1075 1039
1076 return ret; 1040static bool
1041vma_has_uprobes(struct vm_area_struct *vma, unsigned long start, unsigned long end)
1042{
1043 loff_t min, max;
1044 struct inode *inode;
1045 struct rb_node *n;
1046
1047 inode = vma->vm_file->f_mapping->host;
1048
1049 min = vaddr_to_offset(vma, start);
1050 max = min + (end - start) - 1;
1051
1052 spin_lock(&uprobes_treelock);
1053 n = find_node_in_range(inode, min, max);
1054 spin_unlock(&uprobes_treelock);
1055
1056 return !!n;
1077} 1057}
1078 1058
1079/* 1059/*
@@ -1081,37 +1061,18 @@ int uprobe_mmap(struct vm_area_struct *vma)
1081 */ 1061 */
1082void uprobe_munmap(struct vm_area_struct *vma, unsigned long start, unsigned long end) 1062void uprobe_munmap(struct vm_area_struct *vma, unsigned long start, unsigned long end)
1083{ 1063{
1084 struct list_head tmp_list;
1085 struct uprobe *uprobe, *u;
1086 struct inode *inode;
1087
1088 if (!atomic_read(&uprobe_events) || !valid_vma(vma, false)) 1064 if (!atomic_read(&uprobe_events) || !valid_vma(vma, false))
1089 return; 1065 return;
1090 1066
1091 if (!atomic_read(&vma->vm_mm->mm_users)) /* called by mmput() ? */ 1067 if (!atomic_read(&vma->vm_mm->mm_users)) /* called by mmput() ? */
1092 return; 1068 return;
1093 1069
1094 if (!atomic_read(&vma->vm_mm->uprobes_state.count)) 1070 if (!test_bit(MMF_HAS_UPROBES, &vma->vm_mm->flags) ||
1095 return; 1071 test_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags))
1096
1097 inode = vma->vm_file->f_mapping->host;
1098 if (!inode)
1099 return; 1072 return;
1100 1073
1101 mutex_lock(uprobes_mmap_hash(inode)); 1074 if (vma_has_uprobes(vma, start, end))
1102 build_probe_list(inode, vma, start, end, &tmp_list); 1075 set_bit(MMF_RECALC_UPROBES, &vma->vm_mm->flags);
1103
1104 list_for_each_entry_safe(uprobe, u, &tmp_list, pending_list) {
1105 unsigned long vaddr = offset_to_vaddr(vma, uprobe->offset);
1106 /*
1107 * An unregister could have removed the probe before
1108 * unmap. So check before we decrement the count.
1109 */
1110 if (is_swbp_at_addr(vma->vm_mm, vaddr) == 1)
1111 atomic_dec(&vma->vm_mm->uprobes_state.count);
1112 put_uprobe(uprobe);
1113 }
1114 mutex_unlock(uprobes_mmap_hash(inode));
1115} 1076}
1116 1077
1117/* Slot allocation for XOL */ 1078/* Slot allocation for XOL */
@@ -1213,13 +1174,15 @@ void uprobe_clear_state(struct mm_struct *mm)
1213 kfree(area); 1174 kfree(area);
1214} 1175}
1215 1176
1216/* 1177void uprobe_dup_mmap(struct mm_struct *oldmm, struct mm_struct *newmm)
1217 * uprobe_reset_state - Free the area allocated for slots.
1218 */
1219void uprobe_reset_state(struct mm_struct *mm)
1220{ 1178{
1221 mm->uprobes_state.xol_area = NULL; 1179 newmm->uprobes_state.xol_area = NULL;
1222 atomic_set(&mm->uprobes_state.count, 0); 1180
1181 if (test_bit(MMF_HAS_UPROBES, &oldmm->flags)) {
1182 set_bit(MMF_HAS_UPROBES, &newmm->flags);
1183 /* unconditionally, dup_mmap() skips VM_DONTCOPY vmas */
1184 set_bit(MMF_RECALC_UPROBES, &newmm->flags);
1185 }
1223} 1186}
1224 1187
1225/* 1188/*
@@ -1437,6 +1400,25 @@ static bool can_skip_sstep(struct uprobe *uprobe, struct pt_regs *regs)
1437 return false; 1400 return false;
1438} 1401}
1439 1402
1403static void mmf_recalc_uprobes(struct mm_struct *mm)
1404{
1405 struct vm_area_struct *vma;
1406
1407 for (vma = mm->mmap; vma; vma = vma->vm_next) {
1408 if (!valid_vma(vma, false))
1409 continue;
1410 /*
1411 * This is not strictly accurate, we can race with
1412 * uprobe_unregister() and see the already removed
1413 * uprobe if delete_uprobe() was not yet called.
1414 */
1415 if (vma_has_uprobes(vma, vma->vm_start, vma->vm_end))
1416 return;
1417 }
1418
1419 clear_bit(MMF_HAS_UPROBES, &mm->flags);
1420}
1421
1440static struct uprobe *find_active_uprobe(unsigned long bp_vaddr, int *is_swbp) 1422static struct uprobe *find_active_uprobe(unsigned long bp_vaddr, int *is_swbp)
1441{ 1423{
1442 struct mm_struct *mm = current->mm; 1424 struct mm_struct *mm = current->mm;
@@ -1458,11 +1440,24 @@ static struct uprobe *find_active_uprobe(unsigned long bp_vaddr, int *is_swbp)
1458 } else { 1440 } else {
1459 *is_swbp = -EFAULT; 1441 *is_swbp = -EFAULT;
1460 } 1442 }
1443
1444 if (!uprobe && test_and_clear_bit(MMF_RECALC_UPROBES, &mm->flags))
1445 mmf_recalc_uprobes(mm);
1461 up_read(&mm->mmap_sem); 1446 up_read(&mm->mmap_sem);
1462 1447
1463 return uprobe; 1448 return uprobe;
1464} 1449}
1465 1450
1451void __weak arch_uprobe_enable_step(struct arch_uprobe *arch)
1452{
1453 user_enable_single_step(current);
1454}
1455
1456void __weak arch_uprobe_disable_step(struct arch_uprobe *arch)
1457{
1458 user_disable_single_step(current);
1459}
1460
1466/* 1461/*
1467 * Run handler and ask thread to singlestep. 1462 * Run handler and ask thread to singlestep.
1468 * Ensure all non-fatal signals cannot interrupt thread while it singlesteps. 1463 * Ensure all non-fatal signals cannot interrupt thread while it singlesteps.
@@ -1509,7 +1504,7 @@ static void handle_swbp(struct pt_regs *regs)
1509 1504
1510 utask->state = UTASK_SSTEP; 1505 utask->state = UTASK_SSTEP;
1511 if (!pre_ssout(uprobe, regs, bp_vaddr)) { 1506 if (!pre_ssout(uprobe, regs, bp_vaddr)) {
1512 user_enable_single_step(current); 1507 arch_uprobe_enable_step(&uprobe->arch);
1513 return; 1508 return;
1514 } 1509 }
1515 1510
@@ -1518,17 +1513,15 @@ cleanup_ret:
1518 utask->active_uprobe = NULL; 1513 utask->active_uprobe = NULL;
1519 utask->state = UTASK_RUNNING; 1514 utask->state = UTASK_RUNNING;
1520 } 1515 }
1521 if (uprobe) { 1516 if (!(uprobe->flags & UPROBE_SKIP_SSTEP))
1522 if (!(uprobe->flags & UPROBE_SKIP_SSTEP))
1523 1517
1524 /* 1518 /*
1525 * cannot singlestep; cannot skip instruction; 1519 * cannot singlestep; cannot skip instruction;
1526 * re-execute the instruction. 1520 * re-execute the instruction.
1527 */ 1521 */
1528 instruction_pointer_set(regs, bp_vaddr); 1522 instruction_pointer_set(regs, bp_vaddr);
1529 1523
1530 put_uprobe(uprobe); 1524 put_uprobe(uprobe);
1531 }
1532} 1525}
1533 1526
1534/* 1527/*
@@ -1547,10 +1540,10 @@ static void handle_singlestep(struct uprobe_task *utask, struct pt_regs *regs)
1547 else 1540 else
1548 WARN_ON_ONCE(1); 1541 WARN_ON_ONCE(1);
1549 1542
1543 arch_uprobe_disable_step(&uprobe->arch);
1550 put_uprobe(uprobe); 1544 put_uprobe(uprobe);
1551 utask->active_uprobe = NULL; 1545 utask->active_uprobe = NULL;
1552 utask->state = UTASK_RUNNING; 1546 utask->state = UTASK_RUNNING;
1553 user_disable_single_step(current);
1554 xol_free_insn_slot(current); 1547 xol_free_insn_slot(current);
1555 1548
1556 spin_lock_irq(&current->sighand->siglock); 1549 spin_lock_irq(&current->sighand->siglock);
@@ -1589,8 +1582,7 @@ int uprobe_pre_sstep_notifier(struct pt_regs *regs)
1589{ 1582{
1590 struct uprobe_task *utask; 1583 struct uprobe_task *utask;
1591 1584
1592 if (!current->mm || !atomic_read(&current->mm->uprobes_state.count)) 1585 if (!current->mm || !test_bit(MMF_HAS_UPROBES, &current->mm->flags))
1593 /* task is currently not uprobed */
1594 return 0; 1586 return 0;
1595 1587
1596 utask = current->utask; 1588 utask = current->utask;
diff --git a/kernel/exit.c b/kernel/exit.c
index f65345f9e5bb..346616c0092c 100644
--- a/kernel/exit.c
+++ b/kernel/exit.c
@@ -457,108 +457,13 @@ void daemonize(const char *name, ...)
457 /* Become as one with the init task */ 457 /* Become as one with the init task */
458 458
459 daemonize_fs_struct(); 459 daemonize_fs_struct();
460 exit_files(current); 460 daemonize_descriptors();
461 current->files = init_task.files;
462 atomic_inc(&current->files->count);
463 461
464 reparent_to_kthreadd(); 462 reparent_to_kthreadd();
465} 463}
466 464
467EXPORT_SYMBOL(daemonize); 465EXPORT_SYMBOL(daemonize);
468 466
469static void close_files(struct files_struct * files)
470{
471 int i, j;
472 struct fdtable *fdt;
473
474 j = 0;
475
476 /*
477 * It is safe to dereference the fd table without RCU or
478 * ->file_lock because this is the last reference to the
479 * files structure. But use RCU to shut RCU-lockdep up.
480 */
481 rcu_read_lock();
482 fdt = files_fdtable(files);
483 rcu_read_unlock();
484 for (;;) {
485 unsigned long set;
486 i = j * BITS_PER_LONG;
487 if (i >= fdt->max_fds)
488 break;
489 set = fdt->open_fds[j++];
490 while (set) {
491 if (set & 1) {
492 struct file * file = xchg(&fdt->fd[i], NULL);
493 if (file) {
494 filp_close(file, files);
495 cond_resched();
496 }
497 }
498 i++;
499 set >>= 1;
500 }
501 }
502}
503
504struct files_struct *get_files_struct(struct task_struct *task)
505{
506 struct files_struct *files;
507
508 task_lock(task);
509 files = task->files;
510 if (files)
511 atomic_inc(&files->count);
512 task_unlock(task);
513
514 return files;
515}
516
517void put_files_struct(struct files_struct *files)
518{
519 struct fdtable *fdt;
520
521 if (atomic_dec_and_test(&files->count)) {
522 close_files(files);
523 /*
524 * Free the fd and fdset arrays if we expanded them.
525 * If the fdtable was embedded, pass files for freeing
526 * at the end of the RCU grace period. Otherwise,
527 * you can free files immediately.
528 */
529 rcu_read_lock();
530 fdt = files_fdtable(files);
531 if (fdt != &files->fdtab)
532 kmem_cache_free(files_cachep, files);
533 free_fdtable(fdt);
534 rcu_read_unlock();
535 }
536}
537
538void reset_files_struct(struct files_struct *files)
539{
540 struct task_struct *tsk = current;
541 struct files_struct *old;
542
543 old = tsk->files;
544 task_lock(tsk);
545 tsk->files = files;
546 task_unlock(tsk);
547 put_files_struct(old);
548}
549
550void exit_files(struct task_struct *tsk)
551{
552 struct files_struct * files = tsk->files;
553
554 if (files) {
555 task_lock(tsk);
556 tsk->files = NULL;
557 task_unlock(tsk);
558 put_files_struct(files);
559 }
560}
561
562#ifdef CONFIG_MM_OWNER 467#ifdef CONFIG_MM_OWNER
563/* 468/*
564 * A task is exiting. If it owned this mm, find a new owner for the mm. 469 * A task is exiting. If it owned this mm, find a new owner for the mm.
@@ -1046,6 +951,9 @@ void do_exit(long code)
1046 if (tsk->splice_pipe) 951 if (tsk->splice_pipe)
1047 __free_pipe_info(tsk->splice_pipe); 952 __free_pipe_info(tsk->splice_pipe);
1048 953
954 if (tsk->task_frag.page)
955 put_page(tsk->task_frag.page);
956
1049 validate_creds_for_do_exit(tsk); 957 validate_creds_for_do_exit(tsk);
1050 958
1051 preempt_disable(); 959 preempt_disable();
diff --git a/kernel/fork.c b/kernel/fork.c
index 2c8857e12855..8b20ab7d3aa2 100644
--- a/kernel/fork.c
+++ b/kernel/fork.c
@@ -330,6 +330,7 @@ static struct task_struct *dup_task_struct(struct task_struct *orig)
330 tsk->btrace_seq = 0; 330 tsk->btrace_seq = 0;
331#endif 331#endif
332 tsk->splice_pipe = NULL; 332 tsk->splice_pipe = NULL;
333 tsk->task_frag.page = NULL;
333 334
334 account_kernel_stack(ti, 1); 335 account_kernel_stack(ti, 1);
335 336
@@ -353,6 +354,7 @@ static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
353 354
354 down_write(&oldmm->mmap_sem); 355 down_write(&oldmm->mmap_sem);
355 flush_cache_dup_mm(oldmm); 356 flush_cache_dup_mm(oldmm);
357 uprobe_dup_mmap(oldmm, mm);
356 /* 358 /*
357 * Not linked in yet - no deadlock potential: 359 * Not linked in yet - no deadlock potential:
358 */ 360 */
@@ -421,7 +423,12 @@ static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
421 mapping->i_mmap_writable++; 423 mapping->i_mmap_writable++;
422 flush_dcache_mmap_lock(mapping); 424 flush_dcache_mmap_lock(mapping);
423 /* insert tmp into the share list, just after mpnt */ 425 /* insert tmp into the share list, just after mpnt */
424 vma_prio_tree_add(tmp, mpnt); 426 if (unlikely(tmp->vm_flags & VM_NONLINEAR))
427 vma_nonlinear_insert(tmp,
428 &mapping->i_mmap_nonlinear);
429 else
430 vma_interval_tree_insert_after(tmp, mpnt,
431 &mapping->i_mmap);
425 flush_dcache_mmap_unlock(mapping); 432 flush_dcache_mmap_unlock(mapping);
426 mutex_unlock(&mapping->i_mmap_mutex); 433 mutex_unlock(&mapping->i_mmap_mutex);
427 } 434 }
@@ -454,9 +461,6 @@ static int dup_mmap(struct mm_struct *mm, struct mm_struct *oldmm)
454 461
455 if (retval) 462 if (retval)
456 goto out; 463 goto out;
457
458 if (file)
459 uprobe_mmap(tmp);
460 } 464 }
461 /* a new mm has just been created */ 465 /* a new mm has just been created */
462 arch_dup_mmap(oldmm, mm); 466 arch_dup_mmap(oldmm, mm);
@@ -623,26 +627,6 @@ void mmput(struct mm_struct *mm)
623} 627}
624EXPORT_SYMBOL_GPL(mmput); 628EXPORT_SYMBOL_GPL(mmput);
625 629
626/*
627 * We added or removed a vma mapping the executable. The vmas are only mapped
628 * during exec and are not mapped with the mmap system call.
629 * Callers must hold down_write() on the mm's mmap_sem for these
630 */
631void added_exe_file_vma(struct mm_struct *mm)
632{
633 mm->num_exe_file_vmas++;
634}
635
636void removed_exe_file_vma(struct mm_struct *mm)
637{
638 mm->num_exe_file_vmas--;
639 if ((mm->num_exe_file_vmas == 0) && mm->exe_file) {
640 fput(mm->exe_file);
641 mm->exe_file = NULL;
642 }
643
644}
645
646void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file) 630void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
647{ 631{
648 if (new_exe_file) 632 if (new_exe_file)
@@ -650,15 +634,13 @@ void set_mm_exe_file(struct mm_struct *mm, struct file *new_exe_file)
650 if (mm->exe_file) 634 if (mm->exe_file)
651 fput(mm->exe_file); 635 fput(mm->exe_file);
652 mm->exe_file = new_exe_file; 636 mm->exe_file = new_exe_file;
653 mm->num_exe_file_vmas = 0;
654} 637}
655 638
656struct file *get_mm_exe_file(struct mm_struct *mm) 639struct file *get_mm_exe_file(struct mm_struct *mm)
657{ 640{
658 struct file *exe_file; 641 struct file *exe_file;
659 642
660 /* We need mmap_sem to protect against races with removal of 643 /* We need mmap_sem to protect against races with removal of exe_file */
661 * VM_EXECUTABLE vmas */
662 down_read(&mm->mmap_sem); 644 down_read(&mm->mmap_sem);
663 exe_file = mm->exe_file; 645 exe_file = mm->exe_file;
664 if (exe_file) 646 if (exe_file)
@@ -839,8 +821,6 @@ struct mm_struct *dup_mm(struct task_struct *tsk)
839#ifdef CONFIG_TRANSPARENT_HUGEPAGE 821#ifdef CONFIG_TRANSPARENT_HUGEPAGE
840 mm->pmd_huge_pte = NULL; 822 mm->pmd_huge_pte = NULL;
841#endif 823#endif
842 uprobe_reset_state(mm);
843
844 if (!mm_init(mm, tsk)) 824 if (!mm_init(mm, tsk))
845 goto fail_nomem; 825 goto fail_nomem;
846 826
@@ -1081,7 +1061,6 @@ static int copy_signal(unsigned long clone_flags, struct task_struct *tsk)
1081 init_rwsem(&sig->group_rwsem); 1061 init_rwsem(&sig->group_rwsem);
1082#endif 1062#endif
1083 1063
1084 sig->oom_adj = current->signal->oom_adj;
1085 sig->oom_score_adj = current->signal->oom_score_adj; 1064 sig->oom_score_adj = current->signal->oom_score_adj;
1086 sig->oom_score_adj_min = current->signal->oom_score_adj_min; 1065 sig->oom_score_adj_min = current->signal->oom_score_adj_min;
1087 1066
@@ -1280,11 +1259,7 @@ static struct task_struct *copy_process(unsigned long clone_flags,
1280#endif 1259#endif
1281#ifdef CONFIG_TRACE_IRQFLAGS 1260#ifdef CONFIG_TRACE_IRQFLAGS
1282 p->irq_events = 0; 1261 p->irq_events = 0;
1283#ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1284 p->hardirqs_enabled = 1;
1285#else
1286 p->hardirqs_enabled = 0; 1262 p->hardirqs_enabled = 0;
1287#endif
1288 p->hardirq_enable_ip = 0; 1263 p->hardirq_enable_ip = 0;
1289 p->hardirq_enable_event = 0; 1264 p->hardirq_enable_event = 0;
1290 p->hardirq_disable_ip = _THIS_IP_; 1265 p->hardirq_disable_ip = _THIS_IP_;
@@ -1609,7 +1584,7 @@ long do_fork(unsigned long clone_flags,
1609 * requested, no event is reported; otherwise, report if the event 1584 * requested, no event is reported; otherwise, report if the event
1610 * for the type of forking is enabled. 1585 * for the type of forking is enabled.
1611 */ 1586 */
1612 if (likely(user_mode(regs)) && !(clone_flags & CLONE_UNTRACED)) { 1587 if (!(clone_flags & CLONE_UNTRACED) && likely(user_mode(regs))) {
1613 if (clone_flags & CLONE_VFORK) 1588 if (clone_flags & CLONE_VFORK)
1614 trace = PTRACE_EVENT_VFORK; 1589 trace = PTRACE_EVENT_VFORK;
1615 else if ((clone_flags & CSIGNAL) != SIGCHLD) 1590 else if ((clone_flags & CSIGNAL) != SIGCHLD)
@@ -1659,6 +1634,17 @@ long do_fork(unsigned long clone_flags,
1659 return nr; 1634 return nr;
1660} 1635}
1661 1636
1637#ifdef CONFIG_GENERIC_KERNEL_THREAD
1638/*
1639 * Create a kernel thread.
1640 */
1641pid_t kernel_thread(int (*fn)(void *), void *arg, unsigned long flags)
1642{
1643 return do_fork(flags|CLONE_VM|CLONE_UNTRACED, (unsigned long)fn, NULL,
1644 (unsigned long)arg, NULL, NULL);
1645}
1646#endif
1647
1662#ifndef ARCH_MIN_MMSTRUCT_ALIGN 1648#ifndef ARCH_MIN_MMSTRUCT_ALIGN
1663#define ARCH_MIN_MMSTRUCT_ALIGN 0 1649#define ARCH_MIN_MMSTRUCT_ALIGN 0
1664#endif 1650#endif
diff --git a/kernel/irq/chip.c b/kernel/irq/chip.c
index eebd6d5cfb44..57d86d07221e 100644
--- a/kernel/irq/chip.c
+++ b/kernel/irq/chip.c
@@ -671,6 +671,7 @@ irq_set_chip_and_handler_name(unsigned int irq, struct irq_chip *chip,
671 irq_set_chip(irq, chip); 671 irq_set_chip(irq, chip);
672 __irq_set_handler(irq, handle, 0, name); 672 __irq_set_handler(irq, handle, 0, name);
673} 673}
674EXPORT_SYMBOL_GPL(irq_set_chip_and_handler_name);
674 675
675void irq_modify_status(unsigned int irq, unsigned long clr, unsigned long set) 676void irq_modify_status(unsigned int irq, unsigned long clr, unsigned long set)
676{ 677{
diff --git a/kernel/irq/dummychip.c b/kernel/irq/dummychip.c
index b5fcd96c7102..988dc58e8847 100644
--- a/kernel/irq/dummychip.c
+++ b/kernel/irq/dummychip.c
@@ -6,6 +6,7 @@
6 */ 6 */
7#include <linux/interrupt.h> 7#include <linux/interrupt.h>
8#include <linux/irq.h> 8#include <linux/irq.h>
9#include <linux/export.h>
9 10
10#include "internals.h" 11#include "internals.h"
11 12
@@ -57,3 +58,4 @@ struct irq_chip dummy_irq_chip = {
57 .irq_mask = noop, 58 .irq_mask = noop,
58 .irq_unmask = noop, 59 .irq_unmask = noop,
59}; 60};
61EXPORT_SYMBOL_GPL(dummy_irq_chip);
diff --git a/kernel/irq/irqdomain.c b/kernel/irq/irqdomain.c
index 49a77727db42..4e69e24d3d7d 100644
--- a/kernel/irq/irqdomain.c
+++ b/kernel/irq/irqdomain.c
@@ -148,7 +148,8 @@ static unsigned int irq_domain_legacy_revmap(struct irq_domain *domain,
148 * @host_data: Controller private data pointer 148 * @host_data: Controller private data pointer
149 * 149 *
150 * Allocates a legacy irq_domain if irq_base is positive or a linear 150 * Allocates a legacy irq_domain if irq_base is positive or a linear
151 * domain otherwise. 151 * domain otherwise. For the legacy domain, IRQ descriptors will also
152 * be allocated.
152 * 153 *
153 * This is intended to implement the expected behaviour for most 154 * This is intended to implement the expected behaviour for most
154 * interrupt controllers which is that a linear mapping should 155 * interrupt controllers which is that a linear mapping should
@@ -162,11 +163,33 @@ struct irq_domain *irq_domain_add_simple(struct device_node *of_node,
162 const struct irq_domain_ops *ops, 163 const struct irq_domain_ops *ops,
163 void *host_data) 164 void *host_data)
164{ 165{
165 if (first_irq > 0) 166 if (first_irq > 0) {
166 return irq_domain_add_legacy(of_node, size, first_irq, 0, 167 int irq_base;
168
169 if (IS_ENABLED(CONFIG_SPARSE_IRQ)) {
170 /*
171 * Set the descriptor allocator to search for a
172 * 1-to-1 mapping, such as irq_alloc_desc_at().
173 * Use of_node_to_nid() which is defined to
174 * numa_node_id() on platforms that have no custom
175 * implementation.
176 */
177 irq_base = irq_alloc_descs(first_irq, first_irq, size,
178 of_node_to_nid(of_node));
179 if (irq_base < 0) {
180 WARN(1, "Cannot allocate irq_descs @ IRQ%d, assuming pre-allocated\n",
181 first_irq);
182 irq_base = first_irq;
183 }
184 } else
185 irq_base = first_irq;
186
187 return irq_domain_add_legacy(of_node, size, irq_base, 0,
167 ops, host_data); 188 ops, host_data);
168 else 189 }
169 return irq_domain_add_linear(of_node, size, ops, host_data); 190
191 /* A linear domain is the default */
192 return irq_domain_add_linear(of_node, size, ops, host_data);
170} 193}
171 194
172/** 195/**
diff --git a/kernel/jump_label.c b/kernel/jump_label.c
index 43049192b5ec..60f48fa0fd0d 100644
--- a/kernel/jump_label.c
+++ b/kernel/jump_label.c
@@ -118,6 +118,7 @@ void jump_label_rate_limit(struct static_key_deferred *key,
118 key->timeout = rl; 118 key->timeout = rl;
119 INIT_DELAYED_WORK(&key->work, jump_label_update_timeout); 119 INIT_DELAYED_WORK(&key->work, jump_label_update_timeout);
120} 120}
121EXPORT_SYMBOL_GPL(jump_label_rate_limit);
121 122
122static int addr_conflict(struct jump_entry *entry, void *start, void *end) 123static int addr_conflict(struct jump_entry *entry, void *start, void *end)
123{ 124{
diff --git a/kernel/kexec.c b/kernel/kexec.c
index 0668d58d6413..5e4bd7864c5d 100644
--- a/kernel/kexec.c
+++ b/kernel/kexec.c
@@ -21,7 +21,6 @@
21#include <linux/hardirq.h> 21#include <linux/hardirq.h>
22#include <linux/elf.h> 22#include <linux/elf.h>
23#include <linux/elfcore.h> 23#include <linux/elfcore.h>
24#include <generated/utsrelease.h>
25#include <linux/utsname.h> 24#include <linux/utsname.h>
26#include <linux/numa.h> 25#include <linux/numa.h>
27#include <linux/suspend.h> 26#include <linux/suspend.h>
diff --git a/kernel/kmod.c b/kernel/kmod.c
index 6f99aead66c6..1c317e386831 100644
--- a/kernel/kmod.c
+++ b/kernel/kmod.c
@@ -37,6 +37,7 @@
37#include <linux/notifier.h> 37#include <linux/notifier.h>
38#include <linux/suspend.h> 38#include <linux/suspend.h>
39#include <linux/rwsem.h> 39#include <linux/rwsem.h>
40#include <linux/ptrace.h>
40#include <asm/uaccess.h> 41#include <asm/uaccess.h>
41 42
42#include <trace/events/module.h> 43#include <trace/events/module.h>
@@ -221,11 +222,13 @@ static int ____call_usermodehelper(void *data)
221 retval = kernel_execve(sub_info->path, 222 retval = kernel_execve(sub_info->path,
222 (const char *const *)sub_info->argv, 223 (const char *const *)sub_info->argv,
223 (const char *const *)sub_info->envp); 224 (const char *const *)sub_info->envp);
225 if (!retval)
226 return 0;
224 227
225 /* Exec failed? */ 228 /* Exec failed? */
226fail: 229fail:
227 sub_info->retval = retval; 230 sub_info->retval = retval;
228 return 0; 231 do_exit(0);
229} 232}
230 233
231static int call_helper(void *data) 234static int call_helper(void *data)
@@ -292,7 +295,7 @@ static int wait_for_helper(void *data)
292 } 295 }
293 296
294 umh_complete(sub_info); 297 umh_complete(sub_info);
295 return 0; 298 do_exit(0);
296} 299}
297 300
298/* This is run by khelper thread */ 301/* This is run by khelper thread */
diff --git a/kernel/kprobes.c b/kernel/kprobes.c
index c62b8546cc90..098f396aa409 100644
--- a/kernel/kprobes.c
+++ b/kernel/kprobes.c
@@ -561,9 +561,9 @@ static __kprobes void kprobe_optimizer(struct work_struct *work)
561{ 561{
562 LIST_HEAD(free_list); 562 LIST_HEAD(free_list);
563 563
564 mutex_lock(&kprobe_mutex);
564 /* Lock modules while optimizing kprobes */ 565 /* Lock modules while optimizing kprobes */
565 mutex_lock(&module_mutex); 566 mutex_lock(&module_mutex);
566 mutex_lock(&kprobe_mutex);
567 567
568 /* 568 /*
569 * Step 1: Unoptimize kprobes and collect cleaned (unused and disarmed) 569 * Step 1: Unoptimize kprobes and collect cleaned (unused and disarmed)
@@ -586,8 +586,8 @@ static __kprobes void kprobe_optimizer(struct work_struct *work)
586 /* Step 4: Free cleaned kprobes after quiesence period */ 586 /* Step 4: Free cleaned kprobes after quiesence period */
587 do_free_cleaned_kprobes(&free_list); 587 do_free_cleaned_kprobes(&free_list);
588 588
589 mutex_unlock(&kprobe_mutex);
590 mutex_unlock(&module_mutex); 589 mutex_unlock(&module_mutex);
590 mutex_unlock(&kprobe_mutex);
591 591
592 /* Step 5: Kick optimizer again if needed */ 592 /* Step 5: Kick optimizer again if needed */
593 if (!list_empty(&optimizing_list) || !list_empty(&unoptimizing_list)) 593 if (!list_empty(&optimizing_list) || !list_empty(&unoptimizing_list))
@@ -759,20 +759,32 @@ static __kprobes void try_to_optimize_kprobe(struct kprobe *p)
759 struct kprobe *ap; 759 struct kprobe *ap;
760 struct optimized_kprobe *op; 760 struct optimized_kprobe *op;
761 761
762 /* Impossible to optimize ftrace-based kprobe */
763 if (kprobe_ftrace(p))
764 return;
765
766 /* For preparing optimization, jump_label_text_reserved() is called */
767 jump_label_lock();
768 mutex_lock(&text_mutex);
769
762 ap = alloc_aggr_kprobe(p); 770 ap = alloc_aggr_kprobe(p);
763 if (!ap) 771 if (!ap)
764 return; 772 goto out;
765 773
766 op = container_of(ap, struct optimized_kprobe, kp); 774 op = container_of(ap, struct optimized_kprobe, kp);
767 if (!arch_prepared_optinsn(&op->optinsn)) { 775 if (!arch_prepared_optinsn(&op->optinsn)) {
768 /* If failed to setup optimizing, fallback to kprobe */ 776 /* If failed to setup optimizing, fallback to kprobe */
769 arch_remove_optimized_kprobe(op); 777 arch_remove_optimized_kprobe(op);
770 kfree(op); 778 kfree(op);
771 return; 779 goto out;
772 } 780 }
773 781
774 init_aggr_kprobe(ap, p); 782 init_aggr_kprobe(ap, p);
775 optimize_kprobe(ap); 783 optimize_kprobe(ap); /* This just kicks optimizer thread */
784
785out:
786 mutex_unlock(&text_mutex);
787 jump_label_unlock();
776} 788}
777 789
778#ifdef CONFIG_SYSCTL 790#ifdef CONFIG_SYSCTL
@@ -907,9 +919,64 @@ static __kprobes struct kprobe *alloc_aggr_kprobe(struct kprobe *p)
907} 919}
908#endif /* CONFIG_OPTPROBES */ 920#endif /* CONFIG_OPTPROBES */
909 921
922#ifdef KPROBES_CAN_USE_FTRACE
923static struct ftrace_ops kprobe_ftrace_ops __read_mostly = {
924 .func = kprobe_ftrace_handler,
925 .flags = FTRACE_OPS_FL_SAVE_REGS,
926};
927static int kprobe_ftrace_enabled;
928
929/* Must ensure p->addr is really on ftrace */
930static int __kprobes prepare_kprobe(struct kprobe *p)
931{
932 if (!kprobe_ftrace(p))
933 return arch_prepare_kprobe(p);
934
935 return arch_prepare_kprobe_ftrace(p);
936}
937
938/* Caller must lock kprobe_mutex */
939static void __kprobes arm_kprobe_ftrace(struct kprobe *p)
940{
941 int ret;
942
943 ret = ftrace_set_filter_ip(&kprobe_ftrace_ops,
944 (unsigned long)p->addr, 0, 0);
945 WARN(ret < 0, "Failed to arm kprobe-ftrace at %p (%d)\n", p->addr, ret);
946 kprobe_ftrace_enabled++;
947 if (kprobe_ftrace_enabled == 1) {
948 ret = register_ftrace_function(&kprobe_ftrace_ops);
949 WARN(ret < 0, "Failed to init kprobe-ftrace (%d)\n", ret);
950 }
951}
952
953/* Caller must lock kprobe_mutex */
954static void __kprobes disarm_kprobe_ftrace(struct kprobe *p)
955{
956 int ret;
957
958 kprobe_ftrace_enabled--;
959 if (kprobe_ftrace_enabled == 0) {
960 ret = unregister_ftrace_function(&kprobe_ftrace_ops);
961 WARN(ret < 0, "Failed to init kprobe-ftrace (%d)\n", ret);
962 }
963 ret = ftrace_set_filter_ip(&kprobe_ftrace_ops,
964 (unsigned long)p->addr, 1, 0);
965 WARN(ret < 0, "Failed to disarm kprobe-ftrace at %p (%d)\n", p->addr, ret);
966}
967#else /* !KPROBES_CAN_USE_FTRACE */
968#define prepare_kprobe(p) arch_prepare_kprobe(p)
969#define arm_kprobe_ftrace(p) do {} while (0)
970#define disarm_kprobe_ftrace(p) do {} while (0)
971#endif
972
910/* Arm a kprobe with text_mutex */ 973/* Arm a kprobe with text_mutex */
911static void __kprobes arm_kprobe(struct kprobe *kp) 974static void __kprobes arm_kprobe(struct kprobe *kp)
912{ 975{
976 if (unlikely(kprobe_ftrace(kp))) {
977 arm_kprobe_ftrace(kp);
978 return;
979 }
913 /* 980 /*
914 * Here, since __arm_kprobe() doesn't use stop_machine(), 981 * Here, since __arm_kprobe() doesn't use stop_machine(),
915 * this doesn't cause deadlock on text_mutex. So, we don't 982 * this doesn't cause deadlock on text_mutex. So, we don't
@@ -921,11 +988,15 @@ static void __kprobes arm_kprobe(struct kprobe *kp)
921} 988}
922 989
923/* Disarm a kprobe with text_mutex */ 990/* Disarm a kprobe with text_mutex */
924static void __kprobes disarm_kprobe(struct kprobe *kp) 991static void __kprobes disarm_kprobe(struct kprobe *kp, bool reopt)
925{ 992{
993 if (unlikely(kprobe_ftrace(kp))) {
994 disarm_kprobe_ftrace(kp);
995 return;
996 }
926 /* Ditto */ 997 /* Ditto */
927 mutex_lock(&text_mutex); 998 mutex_lock(&text_mutex);
928 __disarm_kprobe(kp, true); 999 __disarm_kprobe(kp, reopt);
929 mutex_unlock(&text_mutex); 1000 mutex_unlock(&text_mutex);
930} 1001}
931 1002
@@ -1144,12 +1215,6 @@ static int __kprobes add_new_kprobe(struct kprobe *ap, struct kprobe *p)
1144 if (p->post_handler && !ap->post_handler) 1215 if (p->post_handler && !ap->post_handler)
1145 ap->post_handler = aggr_post_handler; 1216 ap->post_handler = aggr_post_handler;
1146 1217
1147 if (kprobe_disabled(ap) && !kprobe_disabled(p)) {
1148 ap->flags &= ~KPROBE_FLAG_DISABLED;
1149 if (!kprobes_all_disarmed)
1150 /* Arm the breakpoint again. */
1151 __arm_kprobe(ap);
1152 }
1153 return 0; 1218 return 0;
1154} 1219}
1155 1220
@@ -1189,11 +1254,22 @@ static int __kprobes register_aggr_kprobe(struct kprobe *orig_p,
1189 int ret = 0; 1254 int ret = 0;
1190 struct kprobe *ap = orig_p; 1255 struct kprobe *ap = orig_p;
1191 1256
1257 /* For preparing optimization, jump_label_text_reserved() is called */
1258 jump_label_lock();
1259 /*
1260 * Get online CPUs to avoid text_mutex deadlock.with stop machine,
1261 * which is invoked by unoptimize_kprobe() in add_new_kprobe()
1262 */
1263 get_online_cpus();
1264 mutex_lock(&text_mutex);
1265
1192 if (!kprobe_aggrprobe(orig_p)) { 1266 if (!kprobe_aggrprobe(orig_p)) {
1193 /* If orig_p is not an aggr_kprobe, create new aggr_kprobe. */ 1267 /* If orig_p is not an aggr_kprobe, create new aggr_kprobe. */
1194 ap = alloc_aggr_kprobe(orig_p); 1268 ap = alloc_aggr_kprobe(orig_p);
1195 if (!ap) 1269 if (!ap) {
1196 return -ENOMEM; 1270 ret = -ENOMEM;
1271 goto out;
1272 }
1197 init_aggr_kprobe(ap, orig_p); 1273 init_aggr_kprobe(ap, orig_p);
1198 } else if (kprobe_unused(ap)) 1274 } else if (kprobe_unused(ap))
1199 /* This probe is going to die. Rescue it */ 1275 /* This probe is going to die. Rescue it */
@@ -1213,7 +1289,7 @@ static int __kprobes register_aggr_kprobe(struct kprobe *orig_p,
1213 * free aggr_probe. It will be used next time, or 1289 * free aggr_probe. It will be used next time, or
1214 * freed by unregister_kprobe. 1290 * freed by unregister_kprobe.
1215 */ 1291 */
1216 return ret; 1292 goto out;
1217 1293
1218 /* Prepare optimized instructions if possible. */ 1294 /* Prepare optimized instructions if possible. */
1219 prepare_optimized_kprobe(ap); 1295 prepare_optimized_kprobe(ap);
@@ -1228,7 +1304,20 @@ static int __kprobes register_aggr_kprobe(struct kprobe *orig_p,
1228 1304
1229 /* Copy ap's insn slot to p */ 1305 /* Copy ap's insn slot to p */
1230 copy_kprobe(ap, p); 1306 copy_kprobe(ap, p);
1231 return add_new_kprobe(ap, p); 1307 ret = add_new_kprobe(ap, p);
1308
1309out:
1310 mutex_unlock(&text_mutex);
1311 put_online_cpus();
1312 jump_label_unlock();
1313
1314 if (ret == 0 && kprobe_disabled(ap) && !kprobe_disabled(p)) {
1315 ap->flags &= ~KPROBE_FLAG_DISABLED;
1316 if (!kprobes_all_disarmed)
1317 /* Arm the breakpoint again. */
1318 arm_kprobe(ap);
1319 }
1320 return ret;
1232} 1321}
1233 1322
1234static int __kprobes in_kprobes_functions(unsigned long addr) 1323static int __kprobes in_kprobes_functions(unsigned long addr)
@@ -1313,71 +1402,96 @@ static inline int check_kprobe_rereg(struct kprobe *p)
1313 return ret; 1402 return ret;
1314} 1403}
1315 1404
1316int __kprobes register_kprobe(struct kprobe *p) 1405static __kprobes int check_kprobe_address_safe(struct kprobe *p,
1406 struct module **probed_mod)
1317{ 1407{
1318 int ret = 0; 1408 int ret = 0;
1319 struct kprobe *old_p; 1409 unsigned long ftrace_addr;
1320 struct module *probed_mod;
1321 kprobe_opcode_t *addr;
1322
1323 addr = kprobe_addr(p);
1324 if (IS_ERR(addr))
1325 return PTR_ERR(addr);
1326 p->addr = addr;
1327 1410
1328 ret = check_kprobe_rereg(p); 1411 /*
1329 if (ret) 1412 * If the address is located on a ftrace nop, set the
1330 return ret; 1413 * breakpoint to the following instruction.
1414 */
1415 ftrace_addr = ftrace_location((unsigned long)p->addr);
1416 if (ftrace_addr) {
1417#ifdef KPROBES_CAN_USE_FTRACE
1418 /* Given address is not on the instruction boundary */
1419 if ((unsigned long)p->addr != ftrace_addr)
1420 return -EILSEQ;
1421 p->flags |= KPROBE_FLAG_FTRACE;
1422#else /* !KPROBES_CAN_USE_FTRACE */
1423 return -EINVAL;
1424#endif
1425 }
1331 1426
1332 jump_label_lock(); 1427 jump_label_lock();
1333 preempt_disable(); 1428 preempt_disable();
1429
1430 /* Ensure it is not in reserved area nor out of text */
1334 if (!kernel_text_address((unsigned long) p->addr) || 1431 if (!kernel_text_address((unsigned long) p->addr) ||
1335 in_kprobes_functions((unsigned long) p->addr) || 1432 in_kprobes_functions((unsigned long) p->addr) ||
1336 ftrace_text_reserved(p->addr, p->addr) ||
1337 jump_label_text_reserved(p->addr, p->addr)) { 1433 jump_label_text_reserved(p->addr, p->addr)) {
1338 ret = -EINVAL; 1434 ret = -EINVAL;
1339 goto cannot_probe; 1435 goto out;
1340 } 1436 }
1341 1437
1342 /* User can pass only KPROBE_FLAG_DISABLED to register_kprobe */ 1438 /* Check if are we probing a module */
1343 p->flags &= KPROBE_FLAG_DISABLED; 1439 *probed_mod = __module_text_address((unsigned long) p->addr);
1344 1440 if (*probed_mod) {
1345 /*
1346 * Check if are we probing a module.
1347 */
1348 probed_mod = __module_text_address((unsigned long) p->addr);
1349 if (probed_mod) {
1350 /* Return -ENOENT if fail. */
1351 ret = -ENOENT;
1352 /* 1441 /*
1353 * We must hold a refcount of the probed module while updating 1442 * We must hold a refcount of the probed module while updating
1354 * its code to prohibit unexpected unloading. 1443 * its code to prohibit unexpected unloading.
1355 */ 1444 */
1356 if (unlikely(!try_module_get(probed_mod))) 1445 if (unlikely(!try_module_get(*probed_mod))) {
1357 goto cannot_probe; 1446 ret = -ENOENT;
1447 goto out;
1448 }
1358 1449
1359 /* 1450 /*
1360 * If the module freed .init.text, we couldn't insert 1451 * If the module freed .init.text, we couldn't insert
1361 * kprobes in there. 1452 * kprobes in there.
1362 */ 1453 */
1363 if (within_module_init((unsigned long)p->addr, probed_mod) && 1454 if (within_module_init((unsigned long)p->addr, *probed_mod) &&
1364 probed_mod->state != MODULE_STATE_COMING) { 1455 (*probed_mod)->state != MODULE_STATE_COMING) {
1365 module_put(probed_mod); 1456 module_put(*probed_mod);
1366 goto cannot_probe; 1457 *probed_mod = NULL;
1458 ret = -ENOENT;
1367 } 1459 }
1368 /* ret will be updated by following code */
1369 } 1460 }
1461out:
1370 preempt_enable(); 1462 preempt_enable();
1371 jump_label_unlock(); 1463 jump_label_unlock();
1372 1464
1465 return ret;
1466}
1467
1468int __kprobes register_kprobe(struct kprobe *p)
1469{
1470 int ret;
1471 struct kprobe *old_p;
1472 struct module *probed_mod;
1473 kprobe_opcode_t *addr;
1474
1475 /* Adjust probe address from symbol */
1476 addr = kprobe_addr(p);
1477 if (IS_ERR(addr))
1478 return PTR_ERR(addr);
1479 p->addr = addr;
1480
1481 ret = check_kprobe_rereg(p);
1482 if (ret)
1483 return ret;
1484
1485 /* User can pass only KPROBE_FLAG_DISABLED to register_kprobe */
1486 p->flags &= KPROBE_FLAG_DISABLED;
1373 p->nmissed = 0; 1487 p->nmissed = 0;
1374 INIT_LIST_HEAD(&p->list); 1488 INIT_LIST_HEAD(&p->list);
1375 mutex_lock(&kprobe_mutex);
1376 1489
1377 jump_label_lock(); /* needed to call jump_label_text_reserved() */ 1490 ret = check_kprobe_address_safe(p, &probed_mod);
1491 if (ret)
1492 return ret;
1378 1493
1379 get_online_cpus(); /* For avoiding text_mutex deadlock. */ 1494 mutex_lock(&kprobe_mutex);
1380 mutex_lock(&text_mutex);
1381 1495
1382 old_p = get_kprobe(p->addr); 1496 old_p = get_kprobe(p->addr);
1383 if (old_p) { 1497 if (old_p) {
@@ -1386,7 +1500,9 @@ int __kprobes register_kprobe(struct kprobe *p)
1386 goto out; 1500 goto out;
1387 } 1501 }
1388 1502
1389 ret = arch_prepare_kprobe(p); 1503 mutex_lock(&text_mutex); /* Avoiding text modification */
1504 ret = prepare_kprobe(p);
1505 mutex_unlock(&text_mutex);
1390 if (ret) 1506 if (ret)
1391 goto out; 1507 goto out;
1392 1508
@@ -1395,26 +1511,18 @@ int __kprobes register_kprobe(struct kprobe *p)
1395 &kprobe_table[hash_ptr(p->addr, KPROBE_HASH_BITS)]); 1511 &kprobe_table[hash_ptr(p->addr, KPROBE_HASH_BITS)]);
1396 1512
1397 if (!kprobes_all_disarmed && !kprobe_disabled(p)) 1513 if (!kprobes_all_disarmed && !kprobe_disabled(p))
1398 __arm_kprobe(p); 1514 arm_kprobe(p);
1399 1515
1400 /* Try to optimize kprobe */ 1516 /* Try to optimize kprobe */
1401 try_to_optimize_kprobe(p); 1517 try_to_optimize_kprobe(p);
1402 1518
1403out: 1519out:
1404 mutex_unlock(&text_mutex);
1405 put_online_cpus();
1406 jump_label_unlock();
1407 mutex_unlock(&kprobe_mutex); 1520 mutex_unlock(&kprobe_mutex);
1408 1521
1409 if (probed_mod) 1522 if (probed_mod)
1410 module_put(probed_mod); 1523 module_put(probed_mod);
1411 1524
1412 return ret; 1525 return ret;
1413
1414cannot_probe:
1415 preempt_enable();
1416 jump_label_unlock();
1417 return ret;
1418} 1526}
1419EXPORT_SYMBOL_GPL(register_kprobe); 1527EXPORT_SYMBOL_GPL(register_kprobe);
1420 1528
@@ -1451,7 +1559,7 @@ static struct kprobe *__kprobes __disable_kprobe(struct kprobe *p)
1451 1559
1452 /* Try to disarm and disable this/parent probe */ 1560 /* Try to disarm and disable this/parent probe */
1453 if (p == orig_p || aggr_kprobe_disabled(orig_p)) { 1561 if (p == orig_p || aggr_kprobe_disabled(orig_p)) {
1454 disarm_kprobe(orig_p); 1562 disarm_kprobe(orig_p, true);
1455 orig_p->flags |= KPROBE_FLAG_DISABLED; 1563 orig_p->flags |= KPROBE_FLAG_DISABLED;
1456 } 1564 }
1457 } 1565 }
@@ -2049,10 +2157,11 @@ static void __kprobes report_probe(struct seq_file *pi, struct kprobe *p,
2049 2157
2050 if (!pp) 2158 if (!pp)
2051 pp = p; 2159 pp = p;
2052 seq_printf(pi, "%s%s%s\n", 2160 seq_printf(pi, "%s%s%s%s\n",
2053 (kprobe_gone(p) ? "[GONE]" : ""), 2161 (kprobe_gone(p) ? "[GONE]" : ""),
2054 ((kprobe_disabled(p) && !kprobe_gone(p)) ? "[DISABLED]" : ""), 2162 ((kprobe_disabled(p) && !kprobe_gone(p)) ? "[DISABLED]" : ""),
2055 (kprobe_optimized(pp) ? "[OPTIMIZED]" : "")); 2163 (kprobe_optimized(pp) ? "[OPTIMIZED]" : ""),
2164 (kprobe_ftrace(pp) ? "[FTRACE]" : ""));
2056} 2165}
2057 2166
2058static void __kprobes *kprobe_seq_start(struct seq_file *f, loff_t *pos) 2167static void __kprobes *kprobe_seq_start(struct seq_file *f, loff_t *pos)
@@ -2131,14 +2240,12 @@ static void __kprobes arm_all_kprobes(void)
2131 goto already_enabled; 2240 goto already_enabled;
2132 2241
2133 /* Arming kprobes doesn't optimize kprobe itself */ 2242 /* Arming kprobes doesn't optimize kprobe itself */
2134 mutex_lock(&text_mutex);
2135 for (i = 0; i < KPROBE_TABLE_SIZE; i++) { 2243 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2136 head = &kprobe_table[i]; 2244 head = &kprobe_table[i];
2137 hlist_for_each_entry_rcu(p, node, head, hlist) 2245 hlist_for_each_entry_rcu(p, node, head, hlist)
2138 if (!kprobe_disabled(p)) 2246 if (!kprobe_disabled(p))
2139 __arm_kprobe(p); 2247 arm_kprobe(p);
2140 } 2248 }
2141 mutex_unlock(&text_mutex);
2142 2249
2143 kprobes_all_disarmed = false; 2250 kprobes_all_disarmed = false;
2144 printk(KERN_INFO "Kprobes globally enabled\n"); 2251 printk(KERN_INFO "Kprobes globally enabled\n");
@@ -2166,15 +2273,13 @@ static void __kprobes disarm_all_kprobes(void)
2166 kprobes_all_disarmed = true; 2273 kprobes_all_disarmed = true;
2167 printk(KERN_INFO "Kprobes globally disabled\n"); 2274 printk(KERN_INFO "Kprobes globally disabled\n");
2168 2275
2169 mutex_lock(&text_mutex);
2170 for (i = 0; i < KPROBE_TABLE_SIZE; i++) { 2276 for (i = 0; i < KPROBE_TABLE_SIZE; i++) {
2171 head = &kprobe_table[i]; 2277 head = &kprobe_table[i];
2172 hlist_for_each_entry_rcu(p, node, head, hlist) { 2278 hlist_for_each_entry_rcu(p, node, head, hlist) {
2173 if (!arch_trampoline_kprobe(p) && !kprobe_disabled(p)) 2279 if (!arch_trampoline_kprobe(p) && !kprobe_disabled(p))
2174 __disarm_kprobe(p, false); 2280 disarm_kprobe(p, false);
2175 } 2281 }
2176 } 2282 }
2177 mutex_unlock(&text_mutex);
2178 mutex_unlock(&kprobe_mutex); 2283 mutex_unlock(&kprobe_mutex);
2179 2284
2180 /* Wait for disarming all kprobes by optimizer */ 2285 /* Wait for disarming all kprobes by optimizer */
diff --git a/kernel/kthread.c b/kernel/kthread.c
index b579af57ea10..29fb60caecb5 100644
--- a/kernel/kthread.c
+++ b/kernel/kthread.c
@@ -16,6 +16,7 @@
16#include <linux/mutex.h> 16#include <linux/mutex.h>
17#include <linux/slab.h> 17#include <linux/slab.h>
18#include <linux/freezer.h> 18#include <linux/freezer.h>
19#include <linux/ptrace.h>
19#include <trace/events/sched.h> 20#include <trace/events/sched.h>
20 21
21static DEFINE_SPINLOCK(kthread_create_lock); 22static DEFINE_SPINLOCK(kthread_create_lock);
@@ -37,11 +38,20 @@ struct kthread_create_info
37}; 38};
38 39
39struct kthread { 40struct kthread {
40 int should_stop; 41 unsigned long flags;
42 unsigned int cpu;
41 void *data; 43 void *data;
44 struct completion parked;
42 struct completion exited; 45 struct completion exited;
43}; 46};
44 47
48enum KTHREAD_BITS {
49 KTHREAD_IS_PER_CPU = 0,
50 KTHREAD_SHOULD_STOP,
51 KTHREAD_SHOULD_PARK,
52 KTHREAD_IS_PARKED,
53};
54
45#define to_kthread(tsk) \ 55#define to_kthread(tsk) \
46 container_of((tsk)->vfork_done, struct kthread, exited) 56 container_of((tsk)->vfork_done, struct kthread, exited)
47 57
@@ -52,13 +62,29 @@ struct kthread {
52 * and this will return true. You should then return, and your return 62 * and this will return true. You should then return, and your return
53 * value will be passed through to kthread_stop(). 63 * value will be passed through to kthread_stop().
54 */ 64 */
55int kthread_should_stop(void) 65bool kthread_should_stop(void)
56{ 66{
57 return to_kthread(current)->should_stop; 67 return test_bit(KTHREAD_SHOULD_STOP, &to_kthread(current)->flags);
58} 68}
59EXPORT_SYMBOL(kthread_should_stop); 69EXPORT_SYMBOL(kthread_should_stop);
60 70
61/** 71/**
72 * kthread_should_park - should this kthread park now?
73 *
74 * When someone calls kthread_park() on your kthread, it will be woken
75 * and this will return true. You should then do the necessary
76 * cleanup and call kthread_parkme()
77 *
78 * Similar to kthread_should_stop(), but this keeps the thread alive
79 * and in a park position. kthread_unpark() "restarts" the thread and
80 * calls the thread function again.
81 */
82bool kthread_should_park(void)
83{
84 return test_bit(KTHREAD_SHOULD_PARK, &to_kthread(current)->flags);
85}
86
87/**
62 * kthread_freezable_should_stop - should this freezable kthread return now? 88 * kthread_freezable_should_stop - should this freezable kthread return now?
63 * @was_frozen: optional out parameter, indicates whether %current was frozen 89 * @was_frozen: optional out parameter, indicates whether %current was frozen
64 * 90 *
@@ -96,6 +122,24 @@ void *kthread_data(struct task_struct *task)
96 return to_kthread(task)->data; 122 return to_kthread(task)->data;
97} 123}
98 124
125static void __kthread_parkme(struct kthread *self)
126{
127 __set_current_state(TASK_INTERRUPTIBLE);
128 while (test_bit(KTHREAD_SHOULD_PARK, &self->flags)) {
129 if (!test_and_set_bit(KTHREAD_IS_PARKED, &self->flags))
130 complete(&self->parked);
131 schedule();
132 __set_current_state(TASK_INTERRUPTIBLE);
133 }
134 clear_bit(KTHREAD_IS_PARKED, &self->flags);
135 __set_current_state(TASK_RUNNING);
136}
137
138void kthread_parkme(void)
139{
140 __kthread_parkme(to_kthread(current));
141}
142
99static int kthread(void *_create) 143static int kthread(void *_create)
100{ 144{
101 /* Copy data: it's on kthread's stack */ 145 /* Copy data: it's on kthread's stack */
@@ -105,9 +149,10 @@ static int kthread(void *_create)
105 struct kthread self; 149 struct kthread self;
106 int ret; 150 int ret;
107 151
108 self.should_stop = 0; 152 self.flags = 0;
109 self.data = data; 153 self.data = data;
110 init_completion(&self.exited); 154 init_completion(&self.exited);
155 init_completion(&self.parked);
111 current->vfork_done = &self.exited; 156 current->vfork_done = &self.exited;
112 157
113 /* OK, tell user we're spawned, wait for stop or wakeup */ 158 /* OK, tell user we're spawned, wait for stop or wakeup */
@@ -117,9 +162,11 @@ static int kthread(void *_create)
117 schedule(); 162 schedule();
118 163
119 ret = -EINTR; 164 ret = -EINTR;
120 if (!self.should_stop)
121 ret = threadfn(data);
122 165
166 if (!test_bit(KTHREAD_SHOULD_STOP, &self.flags)) {
167 __kthread_parkme(&self);
168 ret = threadfn(data);
169 }
123 /* we can't just return, we must preserve "self" on stack */ 170 /* we can't just return, we must preserve "self" on stack */
124 do_exit(ret); 171 do_exit(ret);
125} 172}
@@ -172,8 +219,7 @@ static void create_kthread(struct kthread_create_info *create)
172 * Returns a task_struct or ERR_PTR(-ENOMEM). 219 * Returns a task_struct or ERR_PTR(-ENOMEM).
173 */ 220 */
174struct task_struct *kthread_create_on_node(int (*threadfn)(void *data), 221struct task_struct *kthread_create_on_node(int (*threadfn)(void *data),
175 void *data, 222 void *data, int node,
176 int node,
177 const char namefmt[], 223 const char namefmt[],
178 ...) 224 ...)
179{ 225{
@@ -210,6 +256,13 @@ struct task_struct *kthread_create_on_node(int (*threadfn)(void *data),
210} 256}
211EXPORT_SYMBOL(kthread_create_on_node); 257EXPORT_SYMBOL(kthread_create_on_node);
212 258
259static void __kthread_bind(struct task_struct *p, unsigned int cpu)
260{
261 /* It's safe because the task is inactive. */
262 do_set_cpus_allowed(p, cpumask_of(cpu));
263 p->flags |= PF_THREAD_BOUND;
264}
265
213/** 266/**
214 * kthread_bind - bind a just-created kthread to a cpu. 267 * kthread_bind - bind a just-created kthread to a cpu.
215 * @p: thread created by kthread_create(). 268 * @p: thread created by kthread_create().
@@ -226,14 +279,112 @@ void kthread_bind(struct task_struct *p, unsigned int cpu)
226 WARN_ON(1); 279 WARN_ON(1);
227 return; 280 return;
228 } 281 }
229 282 __kthread_bind(p, cpu);
230 /* It's safe because the task is inactive. */
231 do_set_cpus_allowed(p, cpumask_of(cpu));
232 p->flags |= PF_THREAD_BOUND;
233} 283}
234EXPORT_SYMBOL(kthread_bind); 284EXPORT_SYMBOL(kthread_bind);
235 285
236/** 286/**
287 * kthread_create_on_cpu - Create a cpu bound kthread
288 * @threadfn: the function to run until signal_pending(current).
289 * @data: data ptr for @threadfn.
290 * @cpu: The cpu on which the thread should be bound,
291 * @namefmt: printf-style name for the thread. Format is restricted
292 * to "name.*%u". Code fills in cpu number.
293 *
294 * Description: This helper function creates and names a kernel thread
295 * The thread will be woken and put into park mode.
296 */
297struct task_struct *kthread_create_on_cpu(int (*threadfn)(void *data),
298 void *data, unsigned int cpu,
299 const char *namefmt)
300{
301 struct task_struct *p;
302
303 p = kthread_create_on_node(threadfn, data, cpu_to_node(cpu), namefmt,
304 cpu);
305 if (IS_ERR(p))
306 return p;
307 set_bit(KTHREAD_IS_PER_CPU, &to_kthread(p)->flags);
308 to_kthread(p)->cpu = cpu;
309 /* Park the thread to get it out of TASK_UNINTERRUPTIBLE state */
310 kthread_park(p);
311 return p;
312}
313
314static struct kthread *task_get_live_kthread(struct task_struct *k)
315{
316 struct kthread *kthread;
317
318 get_task_struct(k);
319 kthread = to_kthread(k);
320 /* It might have exited */
321 barrier();
322 if (k->vfork_done != NULL)
323 return kthread;
324 return NULL;
325}
326
327/**
328 * kthread_unpark - unpark a thread created by kthread_create().
329 * @k: thread created by kthread_create().
330 *
331 * Sets kthread_should_park() for @k to return false, wakes it, and
332 * waits for it to return. If the thread is marked percpu then its
333 * bound to the cpu again.
334 */
335void kthread_unpark(struct task_struct *k)
336{
337 struct kthread *kthread = task_get_live_kthread(k);
338
339 if (kthread) {
340 clear_bit(KTHREAD_SHOULD_PARK, &kthread->flags);
341 /*
342 * We clear the IS_PARKED bit here as we don't wait
343 * until the task has left the park code. So if we'd
344 * park before that happens we'd see the IS_PARKED bit
345 * which might be about to be cleared.
346 */
347 if (test_and_clear_bit(KTHREAD_IS_PARKED, &kthread->flags)) {
348 if (test_bit(KTHREAD_IS_PER_CPU, &kthread->flags))
349 __kthread_bind(k, kthread->cpu);
350 wake_up_process(k);
351 }
352 }
353 put_task_struct(k);
354}
355
356/**
357 * kthread_park - park a thread created by kthread_create().
358 * @k: thread created by kthread_create().
359 *
360 * Sets kthread_should_park() for @k to return true, wakes it, and
361 * waits for it to return. This can also be called after kthread_create()
362 * instead of calling wake_up_process(): the thread will park without
363 * calling threadfn().
364 *
365 * Returns 0 if the thread is parked, -ENOSYS if the thread exited.
366 * If called by the kthread itself just the park bit is set.
367 */
368int kthread_park(struct task_struct *k)
369{
370 struct kthread *kthread = task_get_live_kthread(k);
371 int ret = -ENOSYS;
372
373 if (kthread) {
374 if (!test_bit(KTHREAD_IS_PARKED, &kthread->flags)) {
375 set_bit(KTHREAD_SHOULD_PARK, &kthread->flags);
376 if (k != current) {
377 wake_up_process(k);
378 wait_for_completion(&kthread->parked);
379 }
380 }
381 ret = 0;
382 }
383 put_task_struct(k);
384 return ret;
385}
386
387/**
237 * kthread_stop - stop a thread created by kthread_create(). 388 * kthread_stop - stop a thread created by kthread_create().
238 * @k: thread created by kthread_create(). 389 * @k: thread created by kthread_create().
239 * 390 *
@@ -250,16 +401,13 @@ EXPORT_SYMBOL(kthread_bind);
250 */ 401 */
251int kthread_stop(struct task_struct *k) 402int kthread_stop(struct task_struct *k)
252{ 403{
253 struct kthread *kthread; 404 struct kthread *kthread = task_get_live_kthread(k);
254 int ret; 405 int ret;
255 406
256 trace_sched_kthread_stop(k); 407 trace_sched_kthread_stop(k);
257 get_task_struct(k); 408 if (kthread) {
258 409 set_bit(KTHREAD_SHOULD_STOP, &kthread->flags);
259 kthread = to_kthread(k); 410 clear_bit(KTHREAD_SHOULD_PARK, &kthread->flags);
260 barrier(); /* it might have exited */
261 if (k->vfork_done != NULL) {
262 kthread->should_stop = 1;
263 wake_up_process(k); 411 wake_up_process(k);
264 wait_for_completion(&kthread->exited); 412 wait_for_completion(&kthread->exited);
265 } 413 }
diff --git a/kernel/lockdep.c b/kernel/lockdep.c
index ea9ee4518c35..7981e5b2350d 100644
--- a/kernel/lockdep.c
+++ b/kernel/lockdep.c
@@ -2998,6 +2998,42 @@ EXPORT_SYMBOL_GPL(lockdep_init_map);
2998 2998
2999struct lock_class_key __lockdep_no_validate__; 2999struct lock_class_key __lockdep_no_validate__;
3000 3000
3001static int
3002print_lock_nested_lock_not_held(struct task_struct *curr,
3003 struct held_lock *hlock,
3004 unsigned long ip)
3005{
3006 if (!debug_locks_off())
3007 return 0;
3008 if (debug_locks_silent)
3009 return 0;
3010
3011 printk("\n");
3012 printk("==================================\n");
3013 printk("[ BUG: Nested lock was not taken ]\n");
3014 print_kernel_ident();
3015 printk("----------------------------------\n");
3016
3017 printk("%s/%d is trying to lock:\n", curr->comm, task_pid_nr(curr));
3018 print_lock(hlock);
3019
3020 printk("\nbut this task is not holding:\n");
3021 printk("%s\n", hlock->nest_lock->name);
3022
3023 printk("\nstack backtrace:\n");
3024 dump_stack();
3025
3026 printk("\nother info that might help us debug this:\n");
3027 lockdep_print_held_locks(curr);
3028
3029 printk("\nstack backtrace:\n");
3030 dump_stack();
3031
3032 return 0;
3033}
3034
3035static int __lock_is_held(struct lockdep_map *lock);
3036
3001/* 3037/*
3002 * This gets called for every mutex_lock*()/spin_lock*() operation. 3038 * This gets called for every mutex_lock*()/spin_lock*() operation.
3003 * We maintain the dependency maps and validate the locking attempt: 3039 * We maintain the dependency maps and validate the locking attempt:
@@ -3139,6 +3175,9 @@ static int __lock_acquire(struct lockdep_map *lock, unsigned int subclass,
3139 } 3175 }
3140 chain_key = iterate_chain_key(chain_key, id); 3176 chain_key = iterate_chain_key(chain_key, id);
3141 3177
3178 if (nest_lock && !__lock_is_held(nest_lock))
3179 return print_lock_nested_lock_not_held(curr, hlock, ip);
3180
3142 if (!validate_chain(curr, lock, hlock, chain_head, chain_key)) 3181 if (!validate_chain(curr, lock, hlock, chain_head, chain_key))
3143 return 0; 3182 return 0;
3144 3183
diff --git a/kernel/pid.c b/kernel/pid.c
index e86b291ad834..aebd4f5aaf41 100644
--- a/kernel/pid.c
+++ b/kernel/pid.c
@@ -479,6 +479,7 @@ pid_t pid_nr_ns(struct pid *pid, struct pid_namespace *ns)
479 } 479 }
480 return nr; 480 return nr;
481} 481}
482EXPORT_SYMBOL_GPL(pid_nr_ns);
482 483
483pid_t pid_vnr(struct pid *pid) 484pid_t pid_vnr(struct pid *pid)
484{ 485{
diff --git a/kernel/pid_namespace.c b/kernel/pid_namespace.c
index 6144bab8fd8e..478bad2745e3 100644
--- a/kernel/pid_namespace.c
+++ b/kernel/pid_namespace.c
@@ -16,6 +16,7 @@
16#include <linux/slab.h> 16#include <linux/slab.h>
17#include <linux/proc_fs.h> 17#include <linux/proc_fs.h>
18#include <linux/reboot.h> 18#include <linux/reboot.h>
19#include <linux/export.h>
19 20
20#define BITS_PER_PAGE (PAGE_SIZE*8) 21#define BITS_PER_PAGE (PAGE_SIZE*8)
21 22
@@ -144,6 +145,7 @@ void free_pid_ns(struct kref *kref)
144 if (parent != NULL) 145 if (parent != NULL)
145 put_pid_ns(parent); 146 put_pid_ns(parent);
146} 147}
148EXPORT_SYMBOL_GPL(free_pid_ns);
147 149
148void zap_pid_ns_processes(struct pid_namespace *pid_ns) 150void zap_pid_ns_processes(struct pid_namespace *pid_ns)
149{ 151{
diff --git a/kernel/power/Kconfig b/kernel/power/Kconfig
index a70518c9d82f..5dfdc9ea180b 100644
--- a/kernel/power/Kconfig
+++ b/kernel/power/Kconfig
@@ -263,6 +263,10 @@ config PM_GENERIC_DOMAINS
263 bool 263 bool
264 depends on PM 264 depends on PM
265 265
266config PM_GENERIC_DOMAINS_SLEEP
267 def_bool y
268 depends on PM_SLEEP && PM_GENERIC_DOMAINS
269
266config PM_GENERIC_DOMAINS_RUNTIME 270config PM_GENERIC_DOMAINS_RUNTIME
267 def_bool y 271 def_bool y
268 depends on PM_RUNTIME && PM_GENERIC_DOMAINS 272 depends on PM_RUNTIME && PM_GENERIC_DOMAINS
diff --git a/kernel/power/poweroff.c b/kernel/power/poweroff.c
index d52359374e85..68197a4e8fc9 100644
--- a/kernel/power/poweroff.c
+++ b/kernel/power/poweroff.c
@@ -37,7 +37,7 @@ static struct sysrq_key_op sysrq_poweroff_op = {
37 .enable_mask = SYSRQ_ENABLE_BOOT, 37 .enable_mask = SYSRQ_ENABLE_BOOT,
38}; 38};
39 39
40static int pm_sysrq_init(void) 40static int __init pm_sysrq_init(void)
41{ 41{
42 register_sysrq_key('o', &sysrq_poweroff_op); 42 register_sysrq_key('o', &sysrq_poweroff_op);
43 return 0; 43 return 0;
diff --git a/kernel/power/process.c b/kernel/power/process.c
index 19db29f67558..87da817f9e13 100644
--- a/kernel/power/process.c
+++ b/kernel/power/process.c
@@ -79,7 +79,7 @@ static int try_to_freeze_tasks(bool user_only)
79 79
80 /* 80 /*
81 * We need to retry, but first give the freezing tasks some 81 * We need to retry, but first give the freezing tasks some
82 * time to enter the regrigerator. 82 * time to enter the refrigerator.
83 */ 83 */
84 msleep(10); 84 msleep(10);
85 } 85 }
diff --git a/kernel/power/qos.c b/kernel/power/qos.c
index 6a031e684026..846bd42c7ed1 100644
--- a/kernel/power/qos.c
+++ b/kernel/power/qos.c
@@ -139,6 +139,7 @@ static inline int pm_qos_get_value(struct pm_qos_constraints *c)
139 default: 139 default:
140 /* runtime check for not using enum */ 140 /* runtime check for not using enum */
141 BUG(); 141 BUG();
142 return PM_QOS_DEFAULT_VALUE;
142 } 143 }
143} 144}
144 145
diff --git a/kernel/ptrace.c b/kernel/ptrace.c
index a232bb59d93f..1f5e55dda955 100644
--- a/kernel/ptrace.c
+++ b/kernel/ptrace.c
@@ -180,7 +180,8 @@ static int ptrace_has_cap(struct user_namespace *ns, unsigned int mode)
180 return has_ns_capability(current, ns, CAP_SYS_PTRACE); 180 return has_ns_capability(current, ns, CAP_SYS_PTRACE);
181} 181}
182 182
183int __ptrace_may_access(struct task_struct *task, unsigned int mode) 183/* Returns 0 on success, -errno on denial. */
184static int __ptrace_may_access(struct task_struct *task, unsigned int mode)
184{ 185{
185 const struct cred *cred = current_cred(), *tcred; 186 const struct cred *cred = current_cred(), *tcred;
186 187
diff --git a/kernel/rcupdate.c b/kernel/rcupdate.c
index 4e6a61b15e86..29ca1c6da594 100644
--- a/kernel/rcupdate.c
+++ b/kernel/rcupdate.c
@@ -45,6 +45,7 @@
45#include <linux/mutex.h> 45#include <linux/mutex.h>
46#include <linux/export.h> 46#include <linux/export.h>
47#include <linux/hardirq.h> 47#include <linux/hardirq.h>
48#include <linux/delay.h>
48 49
49#define CREATE_TRACE_POINTS 50#define CREATE_TRACE_POINTS
50#include <trace/events/rcu.h> 51#include <trace/events/rcu.h>
@@ -81,6 +82,9 @@ void __rcu_read_unlock(void)
81 } else { 82 } else {
82 barrier(); /* critical section before exit code. */ 83 barrier(); /* critical section before exit code. */
83 t->rcu_read_lock_nesting = INT_MIN; 84 t->rcu_read_lock_nesting = INT_MIN;
85#ifdef CONFIG_PROVE_RCU_DELAY
86 udelay(10); /* Make preemption more probable. */
87#endif /* #ifdef CONFIG_PROVE_RCU_DELAY */
84 barrier(); /* assign before ->rcu_read_unlock_special load */ 88 barrier(); /* assign before ->rcu_read_unlock_special load */
85 if (unlikely(ACCESS_ONCE(t->rcu_read_unlock_special))) 89 if (unlikely(ACCESS_ONCE(t->rcu_read_unlock_special)))
86 rcu_read_unlock_special(t); 90 rcu_read_unlock_special(t);
diff --git a/kernel/rcutiny.c b/kernel/rcutiny.c
index 547b1fe5b052..e4c6a598d6f7 100644
--- a/kernel/rcutiny.c
+++ b/kernel/rcutiny.c
@@ -56,25 +56,28 @@ static void __call_rcu(struct rcu_head *head,
56static long long rcu_dynticks_nesting = DYNTICK_TASK_EXIT_IDLE; 56static long long rcu_dynticks_nesting = DYNTICK_TASK_EXIT_IDLE;
57 57
58/* Common code for rcu_idle_enter() and rcu_irq_exit(), see kernel/rcutree.c. */ 58/* Common code for rcu_idle_enter() and rcu_irq_exit(), see kernel/rcutree.c. */
59static void rcu_idle_enter_common(long long oldval) 59static void rcu_idle_enter_common(long long newval)
60{ 60{
61 if (rcu_dynticks_nesting) { 61 if (newval) {
62 RCU_TRACE(trace_rcu_dyntick("--=", 62 RCU_TRACE(trace_rcu_dyntick("--=",
63 oldval, rcu_dynticks_nesting)); 63 rcu_dynticks_nesting, newval));
64 rcu_dynticks_nesting = newval;
64 return; 65 return;
65 } 66 }
66 RCU_TRACE(trace_rcu_dyntick("Start", oldval, rcu_dynticks_nesting)); 67 RCU_TRACE(trace_rcu_dyntick("Start", rcu_dynticks_nesting, newval));
67 if (!is_idle_task(current)) { 68 if (!is_idle_task(current)) {
68 struct task_struct *idle = idle_task(smp_processor_id()); 69 struct task_struct *idle = idle_task(smp_processor_id());
69 70
70 RCU_TRACE(trace_rcu_dyntick("Error on entry: not idle task", 71 RCU_TRACE(trace_rcu_dyntick("Error on entry: not idle task",
71 oldval, rcu_dynticks_nesting)); 72 rcu_dynticks_nesting, newval));
72 ftrace_dump(DUMP_ALL); 73 ftrace_dump(DUMP_ALL);
73 WARN_ONCE(1, "Current pid: %d comm: %s / Idle pid: %d comm: %s", 74 WARN_ONCE(1, "Current pid: %d comm: %s / Idle pid: %d comm: %s",
74 current->pid, current->comm, 75 current->pid, current->comm,
75 idle->pid, idle->comm); /* must be idle task! */ 76 idle->pid, idle->comm); /* must be idle task! */
76 } 77 }
77 rcu_sched_qs(0); /* implies rcu_bh_qsctr_inc(0) */ 78 rcu_sched_qs(0); /* implies rcu_bh_qsctr_inc(0) */
79 barrier();
80 rcu_dynticks_nesting = newval;
78} 81}
79 82
80/* 83/*
@@ -84,17 +87,16 @@ static void rcu_idle_enter_common(long long oldval)
84void rcu_idle_enter(void) 87void rcu_idle_enter(void)
85{ 88{
86 unsigned long flags; 89 unsigned long flags;
87 long long oldval; 90 long long newval;
88 91
89 local_irq_save(flags); 92 local_irq_save(flags);
90 oldval = rcu_dynticks_nesting;
91 WARN_ON_ONCE((rcu_dynticks_nesting & DYNTICK_TASK_NEST_MASK) == 0); 93 WARN_ON_ONCE((rcu_dynticks_nesting & DYNTICK_TASK_NEST_MASK) == 0);
92 if ((rcu_dynticks_nesting & DYNTICK_TASK_NEST_MASK) == 94 if ((rcu_dynticks_nesting & DYNTICK_TASK_NEST_MASK) ==
93 DYNTICK_TASK_NEST_VALUE) 95 DYNTICK_TASK_NEST_VALUE)
94 rcu_dynticks_nesting = 0; 96 newval = 0;
95 else 97 else
96 rcu_dynticks_nesting -= DYNTICK_TASK_NEST_VALUE; 98 newval = rcu_dynticks_nesting - DYNTICK_TASK_NEST_VALUE;
97 rcu_idle_enter_common(oldval); 99 rcu_idle_enter_common(newval);
98 local_irq_restore(flags); 100 local_irq_restore(flags);
99} 101}
100EXPORT_SYMBOL_GPL(rcu_idle_enter); 102EXPORT_SYMBOL_GPL(rcu_idle_enter);
@@ -105,15 +107,15 @@ EXPORT_SYMBOL_GPL(rcu_idle_enter);
105void rcu_irq_exit(void) 107void rcu_irq_exit(void)
106{ 108{
107 unsigned long flags; 109 unsigned long flags;
108 long long oldval; 110 long long newval;
109 111
110 local_irq_save(flags); 112 local_irq_save(flags);
111 oldval = rcu_dynticks_nesting; 113 newval = rcu_dynticks_nesting - 1;
112 rcu_dynticks_nesting--; 114 WARN_ON_ONCE(newval < 0);
113 WARN_ON_ONCE(rcu_dynticks_nesting < 0); 115 rcu_idle_enter_common(newval);
114 rcu_idle_enter_common(oldval);
115 local_irq_restore(flags); 116 local_irq_restore(flags);
116} 117}
118EXPORT_SYMBOL_GPL(rcu_irq_exit);
117 119
118/* Common code for rcu_idle_exit() and rcu_irq_enter(), see kernel/rcutree.c. */ 120/* Common code for rcu_idle_exit() and rcu_irq_enter(), see kernel/rcutree.c. */
119static void rcu_idle_exit_common(long long oldval) 121static void rcu_idle_exit_common(long long oldval)
@@ -171,6 +173,7 @@ void rcu_irq_enter(void)
171 rcu_idle_exit_common(oldval); 173 rcu_idle_exit_common(oldval);
172 local_irq_restore(flags); 174 local_irq_restore(flags);
173} 175}
176EXPORT_SYMBOL_GPL(rcu_irq_enter);
174 177
175#ifdef CONFIG_DEBUG_LOCK_ALLOC 178#ifdef CONFIG_DEBUG_LOCK_ALLOC
176 179
diff --git a/kernel/rcutiny_plugin.h b/kernel/rcutiny_plugin.h
index 918fd1e8509c..3d0190282204 100644
--- a/kernel/rcutiny_plugin.h
+++ b/kernel/rcutiny_plugin.h
@@ -278,7 +278,7 @@ static int rcu_boost(void)
278 rcu_preempt_ctrlblk.exp_tasks == NULL) 278 rcu_preempt_ctrlblk.exp_tasks == NULL)
279 return 0; /* Nothing to boost. */ 279 return 0; /* Nothing to boost. */
280 280
281 raw_local_irq_save(flags); 281 local_irq_save(flags);
282 282
283 /* 283 /*
284 * Recheck with irqs disabled: all tasks in need of boosting 284 * Recheck with irqs disabled: all tasks in need of boosting
@@ -287,7 +287,7 @@ static int rcu_boost(void)
287 */ 287 */
288 if (rcu_preempt_ctrlblk.boost_tasks == NULL && 288 if (rcu_preempt_ctrlblk.boost_tasks == NULL &&
289 rcu_preempt_ctrlblk.exp_tasks == NULL) { 289 rcu_preempt_ctrlblk.exp_tasks == NULL) {
290 raw_local_irq_restore(flags); 290 local_irq_restore(flags);
291 return 0; 291 return 0;
292 } 292 }
293 293
@@ -317,7 +317,7 @@ static int rcu_boost(void)
317 t = container_of(tb, struct task_struct, rcu_node_entry); 317 t = container_of(tb, struct task_struct, rcu_node_entry);
318 rt_mutex_init_proxy_locked(&mtx, t); 318 rt_mutex_init_proxy_locked(&mtx, t);
319 t->rcu_boost_mutex = &mtx; 319 t->rcu_boost_mutex = &mtx;
320 raw_local_irq_restore(flags); 320 local_irq_restore(flags);
321 rt_mutex_lock(&mtx); 321 rt_mutex_lock(&mtx);
322 rt_mutex_unlock(&mtx); /* Keep lockdep happy. */ 322 rt_mutex_unlock(&mtx); /* Keep lockdep happy. */
323 323
@@ -991,9 +991,9 @@ static void rcu_trace_sub_qlen(struct rcu_ctrlblk *rcp, int n)
991{ 991{
992 unsigned long flags; 992 unsigned long flags;
993 993
994 raw_local_irq_save(flags); 994 local_irq_save(flags);
995 rcp->qlen -= n; 995 rcp->qlen -= n;
996 raw_local_irq_restore(flags); 996 local_irq_restore(flags);
997} 997}
998 998
999/* 999/*
diff --git a/kernel/rcutorture.c b/kernel/rcutorture.c
index 25b15033c61f..aaa7b9f3532a 100644
--- a/kernel/rcutorture.c
+++ b/kernel/rcutorture.c
@@ -53,10 +53,11 @@ MODULE_AUTHOR("Paul E. McKenney <paulmck@us.ibm.com> and Josh Triplett <josh@fre
53 53
54static int nreaders = -1; /* # reader threads, defaults to 2*ncpus */ 54static int nreaders = -1; /* # reader threads, defaults to 2*ncpus */
55static int nfakewriters = 4; /* # fake writer threads */ 55static int nfakewriters = 4; /* # fake writer threads */
56static int stat_interval; /* Interval between stats, in seconds. */ 56static int stat_interval = 60; /* Interval between stats, in seconds. */
57 /* Defaults to "only at end of test". */ 57 /* Zero means "only at end of test". */
58static bool verbose; /* Print more debug info. */ 58static bool verbose; /* Print more debug info. */
59static bool test_no_idle_hz; /* Test RCU's support for tickless idle CPUs. */ 59static bool test_no_idle_hz = true;
60 /* Test RCU support for tickless idle CPUs. */
60static int shuffle_interval = 3; /* Interval between shuffles (in sec)*/ 61static int shuffle_interval = 3; /* Interval between shuffles (in sec)*/
61static int stutter = 5; /* Start/stop testing interval (in sec) */ 62static int stutter = 5; /* Start/stop testing interval (in sec) */
62static int irqreader = 1; /* RCU readers from irq (timers). */ 63static int irqreader = 1; /* RCU readers from irq (timers). */
@@ -119,11 +120,11 @@ MODULE_PARM_DESC(torture_type, "Type of RCU to torture (rcu, rcu_bh, srcu)");
119 120
120#define TORTURE_FLAG "-torture:" 121#define TORTURE_FLAG "-torture:"
121#define PRINTK_STRING(s) \ 122#define PRINTK_STRING(s) \
122 do { printk(KERN_ALERT "%s" TORTURE_FLAG s "\n", torture_type); } while (0) 123 do { pr_alert("%s" TORTURE_FLAG s "\n", torture_type); } while (0)
123#define VERBOSE_PRINTK_STRING(s) \ 124#define VERBOSE_PRINTK_STRING(s) \
124 do { if (verbose) printk(KERN_ALERT "%s" TORTURE_FLAG s "\n", torture_type); } while (0) 125 do { if (verbose) pr_alert("%s" TORTURE_FLAG s "\n", torture_type); } while (0)
125#define VERBOSE_PRINTK_ERRSTRING(s) \ 126#define VERBOSE_PRINTK_ERRSTRING(s) \
126 do { if (verbose) printk(KERN_ALERT "%s" TORTURE_FLAG "!!! " s "\n", torture_type); } while (0) 127 do { if (verbose) pr_alert("%s" TORTURE_FLAG "!!! " s "\n", torture_type); } while (0)
127 128
128static char printk_buf[4096]; 129static char printk_buf[4096];
129 130
@@ -176,8 +177,14 @@ static long n_rcu_torture_boosts;
176static long n_rcu_torture_timers; 177static long n_rcu_torture_timers;
177static long n_offline_attempts; 178static long n_offline_attempts;
178static long n_offline_successes; 179static long n_offline_successes;
180static unsigned long sum_offline;
181static int min_offline = -1;
182static int max_offline;
179static long n_online_attempts; 183static long n_online_attempts;
180static long n_online_successes; 184static long n_online_successes;
185static unsigned long sum_online;
186static int min_online = -1;
187static int max_online;
181static long n_barrier_attempts; 188static long n_barrier_attempts;
182static long n_barrier_successes; 189static long n_barrier_successes;
183static struct list_head rcu_torture_removed; 190static struct list_head rcu_torture_removed;
@@ -235,7 +242,7 @@ rcutorture_shutdown_notify(struct notifier_block *unused1,
235 if (fullstop == FULLSTOP_DONTSTOP) 242 if (fullstop == FULLSTOP_DONTSTOP)
236 fullstop = FULLSTOP_SHUTDOWN; 243 fullstop = FULLSTOP_SHUTDOWN;
237 else 244 else
238 printk(KERN_WARNING /* but going down anyway, so... */ 245 pr_warn(/* but going down anyway, so... */
239 "Concurrent 'rmmod rcutorture' and shutdown illegal!\n"); 246 "Concurrent 'rmmod rcutorture' and shutdown illegal!\n");
240 mutex_unlock(&fullstop_mutex); 247 mutex_unlock(&fullstop_mutex);
241 return NOTIFY_DONE; 248 return NOTIFY_DONE;
@@ -248,7 +255,7 @@ rcutorture_shutdown_notify(struct notifier_block *unused1,
248static void rcutorture_shutdown_absorb(char *title) 255static void rcutorture_shutdown_absorb(char *title)
249{ 256{
250 if (ACCESS_ONCE(fullstop) == FULLSTOP_SHUTDOWN) { 257 if (ACCESS_ONCE(fullstop) == FULLSTOP_SHUTDOWN) {
251 printk(KERN_NOTICE 258 pr_notice(
252 "rcutorture thread %s parking due to system shutdown\n", 259 "rcutorture thread %s parking due to system shutdown\n",
253 title); 260 title);
254 schedule_timeout_uninterruptible(MAX_SCHEDULE_TIMEOUT); 261 schedule_timeout_uninterruptible(MAX_SCHEDULE_TIMEOUT);
@@ -1214,11 +1221,13 @@ rcu_torture_printk(char *page)
1214 n_rcu_torture_boost_failure, 1221 n_rcu_torture_boost_failure,
1215 n_rcu_torture_boosts, 1222 n_rcu_torture_boosts,
1216 n_rcu_torture_timers); 1223 n_rcu_torture_timers);
1217 cnt += sprintf(&page[cnt], "onoff: %ld/%ld:%ld/%ld ", 1224 cnt += sprintf(&page[cnt],
1218 n_online_successes, 1225 "onoff: %ld/%ld:%ld/%ld %d,%d:%d,%d %lu:%lu (HZ=%d) ",
1219 n_online_attempts, 1226 n_online_successes, n_online_attempts,
1220 n_offline_successes, 1227 n_offline_successes, n_offline_attempts,
1221 n_offline_attempts); 1228 min_online, max_online,
1229 min_offline, max_offline,
1230 sum_online, sum_offline, HZ);
1222 cnt += sprintf(&page[cnt], "barrier: %ld/%ld:%ld", 1231 cnt += sprintf(&page[cnt], "barrier: %ld/%ld:%ld",
1223 n_barrier_successes, 1232 n_barrier_successes,
1224 n_barrier_attempts, 1233 n_barrier_attempts,
@@ -1267,7 +1276,7 @@ rcu_torture_stats_print(void)
1267 int cnt; 1276 int cnt;
1268 1277
1269 cnt = rcu_torture_printk(printk_buf); 1278 cnt = rcu_torture_printk(printk_buf);
1270 printk(KERN_ALERT "%s", printk_buf); 1279 pr_alert("%s", printk_buf);
1271} 1280}
1272 1281
1273/* 1282/*
@@ -1380,20 +1389,20 @@ rcu_torture_stutter(void *arg)
1380static inline void 1389static inline void
1381rcu_torture_print_module_parms(struct rcu_torture_ops *cur_ops, char *tag) 1390rcu_torture_print_module_parms(struct rcu_torture_ops *cur_ops, char *tag)
1382{ 1391{
1383 printk(KERN_ALERT "%s" TORTURE_FLAG 1392 pr_alert("%s" TORTURE_FLAG
1384 "--- %s: nreaders=%d nfakewriters=%d " 1393 "--- %s: nreaders=%d nfakewriters=%d "
1385 "stat_interval=%d verbose=%d test_no_idle_hz=%d " 1394 "stat_interval=%d verbose=%d test_no_idle_hz=%d "
1386 "shuffle_interval=%d stutter=%d irqreader=%d " 1395 "shuffle_interval=%d stutter=%d irqreader=%d "
1387 "fqs_duration=%d fqs_holdoff=%d fqs_stutter=%d " 1396 "fqs_duration=%d fqs_holdoff=%d fqs_stutter=%d "
1388 "test_boost=%d/%d test_boost_interval=%d " 1397 "test_boost=%d/%d test_boost_interval=%d "
1389 "test_boost_duration=%d shutdown_secs=%d " 1398 "test_boost_duration=%d shutdown_secs=%d "
1390 "onoff_interval=%d onoff_holdoff=%d\n", 1399 "onoff_interval=%d onoff_holdoff=%d\n",
1391 torture_type, tag, nrealreaders, nfakewriters, 1400 torture_type, tag, nrealreaders, nfakewriters,
1392 stat_interval, verbose, test_no_idle_hz, shuffle_interval, 1401 stat_interval, verbose, test_no_idle_hz, shuffle_interval,
1393 stutter, irqreader, fqs_duration, fqs_holdoff, fqs_stutter, 1402 stutter, irqreader, fqs_duration, fqs_holdoff, fqs_stutter,
1394 test_boost, cur_ops->can_boost, 1403 test_boost, cur_ops->can_boost,
1395 test_boost_interval, test_boost_duration, shutdown_secs, 1404 test_boost_interval, test_boost_duration, shutdown_secs,
1396 onoff_interval, onoff_holdoff); 1405 onoff_interval, onoff_holdoff);
1397} 1406}
1398 1407
1399static struct notifier_block rcutorture_shutdown_nb = { 1408static struct notifier_block rcutorture_shutdown_nb = {
@@ -1460,9 +1469,9 @@ rcu_torture_shutdown(void *arg)
1460 !kthread_should_stop()) { 1469 !kthread_should_stop()) {
1461 delta = shutdown_time - jiffies_snap; 1470 delta = shutdown_time - jiffies_snap;
1462 if (verbose) 1471 if (verbose)
1463 printk(KERN_ALERT "%s" TORTURE_FLAG 1472 pr_alert("%s" TORTURE_FLAG
1464 "rcu_torture_shutdown task: %lu jiffies remaining\n", 1473 "rcu_torture_shutdown task: %lu jiffies remaining\n",
1465 torture_type, delta); 1474 torture_type, delta);
1466 schedule_timeout_interruptible(delta); 1475 schedule_timeout_interruptible(delta);
1467 jiffies_snap = ACCESS_ONCE(jiffies); 1476 jiffies_snap = ACCESS_ONCE(jiffies);
1468 } 1477 }
@@ -1490,8 +1499,10 @@ static int __cpuinit
1490rcu_torture_onoff(void *arg) 1499rcu_torture_onoff(void *arg)
1491{ 1500{
1492 int cpu; 1501 int cpu;
1502 unsigned long delta;
1493 int maxcpu = -1; 1503 int maxcpu = -1;
1494 DEFINE_RCU_RANDOM(rand); 1504 DEFINE_RCU_RANDOM(rand);
1505 unsigned long starttime;
1495 1506
1496 VERBOSE_PRINTK_STRING("rcu_torture_onoff task started"); 1507 VERBOSE_PRINTK_STRING("rcu_torture_onoff task started");
1497 for_each_online_cpu(cpu) 1508 for_each_online_cpu(cpu)
@@ -1506,29 +1517,51 @@ rcu_torture_onoff(void *arg)
1506 cpu = (rcu_random(&rand) >> 4) % (maxcpu + 1); 1517 cpu = (rcu_random(&rand) >> 4) % (maxcpu + 1);
1507 if (cpu_online(cpu) && cpu_is_hotpluggable(cpu)) { 1518 if (cpu_online(cpu) && cpu_is_hotpluggable(cpu)) {
1508 if (verbose) 1519 if (verbose)
1509 printk(KERN_ALERT "%s" TORTURE_FLAG 1520 pr_alert("%s" TORTURE_FLAG
1510 "rcu_torture_onoff task: offlining %d\n", 1521 "rcu_torture_onoff task: offlining %d\n",
1511 torture_type, cpu); 1522 torture_type, cpu);
1523 starttime = jiffies;
1512 n_offline_attempts++; 1524 n_offline_attempts++;
1513 if (cpu_down(cpu) == 0) { 1525 if (cpu_down(cpu) == 0) {
1514 if (verbose) 1526 if (verbose)
1515 printk(KERN_ALERT "%s" TORTURE_FLAG 1527 pr_alert("%s" TORTURE_FLAG
1516 "rcu_torture_onoff task: offlined %d\n", 1528 "rcu_torture_onoff task: offlined %d\n",
1517 torture_type, cpu); 1529 torture_type, cpu);
1518 n_offline_successes++; 1530 n_offline_successes++;
1531 delta = jiffies - starttime;
1532 sum_offline += delta;
1533 if (min_offline < 0) {
1534 min_offline = delta;
1535 max_offline = delta;
1536 }
1537 if (min_offline > delta)
1538 min_offline = delta;
1539 if (max_offline < delta)
1540 max_offline = delta;
1519 } 1541 }
1520 } else if (cpu_is_hotpluggable(cpu)) { 1542 } else if (cpu_is_hotpluggable(cpu)) {
1521 if (verbose) 1543 if (verbose)
1522 printk(KERN_ALERT "%s" TORTURE_FLAG 1544 pr_alert("%s" TORTURE_FLAG
1523 "rcu_torture_onoff task: onlining %d\n", 1545 "rcu_torture_onoff task: onlining %d\n",
1524 torture_type, cpu); 1546 torture_type, cpu);
1547 starttime = jiffies;
1525 n_online_attempts++; 1548 n_online_attempts++;
1526 if (cpu_up(cpu) == 0) { 1549 if (cpu_up(cpu) == 0) {
1527 if (verbose) 1550 if (verbose)
1528 printk(KERN_ALERT "%s" TORTURE_FLAG 1551 pr_alert("%s" TORTURE_FLAG
1529 "rcu_torture_onoff task: onlined %d\n", 1552 "rcu_torture_onoff task: onlined %d\n",
1530 torture_type, cpu); 1553 torture_type, cpu);
1531 n_online_successes++; 1554 n_online_successes++;
1555 delta = jiffies - starttime;
1556 sum_online += delta;
1557 if (min_online < 0) {
1558 min_online = delta;
1559 max_online = delta;
1560 }
1561 if (min_online > delta)
1562 min_online = delta;
1563 if (max_online < delta)
1564 max_online = delta;
1532 } 1565 }
1533 } 1566 }
1534 schedule_timeout_interruptible(onoff_interval * HZ); 1567 schedule_timeout_interruptible(onoff_interval * HZ);
@@ -1593,14 +1626,14 @@ static int __cpuinit rcu_torture_stall(void *args)
1593 if (!kthread_should_stop()) { 1626 if (!kthread_should_stop()) {
1594 stop_at = get_seconds() + stall_cpu; 1627 stop_at = get_seconds() + stall_cpu;
1595 /* RCU CPU stall is expected behavior in following code. */ 1628 /* RCU CPU stall is expected behavior in following code. */
1596 printk(KERN_ALERT "rcu_torture_stall start.\n"); 1629 pr_alert("rcu_torture_stall start.\n");
1597 rcu_read_lock(); 1630 rcu_read_lock();
1598 preempt_disable(); 1631 preempt_disable();
1599 while (ULONG_CMP_LT(get_seconds(), stop_at)) 1632 while (ULONG_CMP_LT(get_seconds(), stop_at))
1600 continue; /* Induce RCU CPU stall warning. */ 1633 continue; /* Induce RCU CPU stall warning. */
1601 preempt_enable(); 1634 preempt_enable();
1602 rcu_read_unlock(); 1635 rcu_read_unlock();
1603 printk(KERN_ALERT "rcu_torture_stall end.\n"); 1636 pr_alert("rcu_torture_stall end.\n");
1604 } 1637 }
1605 rcutorture_shutdown_absorb("rcu_torture_stall"); 1638 rcutorture_shutdown_absorb("rcu_torture_stall");
1606 while (!kthread_should_stop()) 1639 while (!kthread_should_stop())
@@ -1716,12 +1749,12 @@ static int rcu_torture_barrier_init(void)
1716 if (n_barrier_cbs == 0) 1749 if (n_barrier_cbs == 0)
1717 return 0; 1750 return 0;
1718 if (cur_ops->call == NULL || cur_ops->cb_barrier == NULL) { 1751 if (cur_ops->call == NULL || cur_ops->cb_barrier == NULL) {
1719 printk(KERN_ALERT "%s" TORTURE_FLAG 1752 pr_alert("%s" TORTURE_FLAG
1720 " Call or barrier ops missing for %s,\n", 1753 " Call or barrier ops missing for %s,\n",
1721 torture_type, cur_ops->name); 1754 torture_type, cur_ops->name);
1722 printk(KERN_ALERT "%s" TORTURE_FLAG 1755 pr_alert("%s" TORTURE_FLAG
1723 " RCU barrier testing omitted from run.\n", 1756 " RCU barrier testing omitted from run.\n",
1724 torture_type); 1757 torture_type);
1725 return 0; 1758 return 0;
1726 } 1759 }
1727 atomic_set(&barrier_cbs_count, 0); 1760 atomic_set(&barrier_cbs_count, 0);
@@ -1814,7 +1847,7 @@ rcu_torture_cleanup(void)
1814 mutex_lock(&fullstop_mutex); 1847 mutex_lock(&fullstop_mutex);
1815 rcutorture_record_test_transition(); 1848 rcutorture_record_test_transition();
1816 if (fullstop == FULLSTOP_SHUTDOWN) { 1849 if (fullstop == FULLSTOP_SHUTDOWN) {
1817 printk(KERN_WARNING /* but going down anyway, so... */ 1850 pr_warn(/* but going down anyway, so... */
1818 "Concurrent 'rmmod rcutorture' and shutdown illegal!\n"); 1851 "Concurrent 'rmmod rcutorture' and shutdown illegal!\n");
1819 mutex_unlock(&fullstop_mutex); 1852 mutex_unlock(&fullstop_mutex);
1820 schedule_timeout_uninterruptible(10); 1853 schedule_timeout_uninterruptible(10);
@@ -1938,17 +1971,17 @@ rcu_torture_init(void)
1938 break; 1971 break;
1939 } 1972 }
1940 if (i == ARRAY_SIZE(torture_ops)) { 1973 if (i == ARRAY_SIZE(torture_ops)) {
1941 printk(KERN_ALERT "rcu-torture: invalid torture type: \"%s\"\n", 1974 pr_alert("rcu-torture: invalid torture type: \"%s\"\n",
1942 torture_type); 1975 torture_type);
1943 printk(KERN_ALERT "rcu-torture types:"); 1976 pr_alert("rcu-torture types:");
1944 for (i = 0; i < ARRAY_SIZE(torture_ops); i++) 1977 for (i = 0; i < ARRAY_SIZE(torture_ops); i++)
1945 printk(KERN_ALERT " %s", torture_ops[i]->name); 1978 pr_alert(" %s", torture_ops[i]->name);
1946 printk(KERN_ALERT "\n"); 1979 pr_alert("\n");
1947 mutex_unlock(&fullstop_mutex); 1980 mutex_unlock(&fullstop_mutex);
1948 return -EINVAL; 1981 return -EINVAL;
1949 } 1982 }
1950 if (cur_ops->fqs == NULL && fqs_duration != 0) { 1983 if (cur_ops->fqs == NULL && fqs_duration != 0) {
1951 printk(KERN_ALERT "rcu-torture: ->fqs NULL and non-zero fqs_duration, fqs disabled.\n"); 1984 pr_alert("rcu-torture: ->fqs NULL and non-zero fqs_duration, fqs disabled.\n");
1952 fqs_duration = 0; 1985 fqs_duration = 0;
1953 } 1986 }
1954 if (cur_ops->init) 1987 if (cur_ops->init)
@@ -1996,14 +2029,15 @@ rcu_torture_init(void)
1996 /* Start up the kthreads. */ 2029 /* Start up the kthreads. */
1997 2030
1998 VERBOSE_PRINTK_STRING("Creating rcu_torture_writer task"); 2031 VERBOSE_PRINTK_STRING("Creating rcu_torture_writer task");
1999 writer_task = kthread_run(rcu_torture_writer, NULL, 2032 writer_task = kthread_create(rcu_torture_writer, NULL,
2000 "rcu_torture_writer"); 2033 "rcu_torture_writer");
2001 if (IS_ERR(writer_task)) { 2034 if (IS_ERR(writer_task)) {
2002 firsterr = PTR_ERR(writer_task); 2035 firsterr = PTR_ERR(writer_task);
2003 VERBOSE_PRINTK_ERRSTRING("Failed to create writer"); 2036 VERBOSE_PRINTK_ERRSTRING("Failed to create writer");
2004 writer_task = NULL; 2037 writer_task = NULL;
2005 goto unwind; 2038 goto unwind;
2006 } 2039 }
2040 wake_up_process(writer_task);
2007 fakewriter_tasks = kzalloc(nfakewriters * sizeof(fakewriter_tasks[0]), 2041 fakewriter_tasks = kzalloc(nfakewriters * sizeof(fakewriter_tasks[0]),
2008 GFP_KERNEL); 2042 GFP_KERNEL);
2009 if (fakewriter_tasks == NULL) { 2043 if (fakewriter_tasks == NULL) {
@@ -2118,14 +2152,15 @@ rcu_torture_init(void)
2118 } 2152 }
2119 if (shutdown_secs > 0) { 2153 if (shutdown_secs > 0) {
2120 shutdown_time = jiffies + shutdown_secs * HZ; 2154 shutdown_time = jiffies + shutdown_secs * HZ;
2121 shutdown_task = kthread_run(rcu_torture_shutdown, NULL, 2155 shutdown_task = kthread_create(rcu_torture_shutdown, NULL,
2122 "rcu_torture_shutdown"); 2156 "rcu_torture_shutdown");
2123 if (IS_ERR(shutdown_task)) { 2157 if (IS_ERR(shutdown_task)) {
2124 firsterr = PTR_ERR(shutdown_task); 2158 firsterr = PTR_ERR(shutdown_task);
2125 VERBOSE_PRINTK_ERRSTRING("Failed to create shutdown"); 2159 VERBOSE_PRINTK_ERRSTRING("Failed to create shutdown");
2126 shutdown_task = NULL; 2160 shutdown_task = NULL;
2127 goto unwind; 2161 goto unwind;
2128 } 2162 }
2163 wake_up_process(shutdown_task);
2129 } 2164 }
2130 i = rcu_torture_onoff_init(); 2165 i = rcu_torture_onoff_init();
2131 if (i != 0) { 2166 if (i != 0) {
diff --git a/kernel/rcutree.c b/kernel/rcutree.c
index f280e542e3e9..74df86bd9204 100644
--- a/kernel/rcutree.c
+++ b/kernel/rcutree.c
@@ -52,6 +52,7 @@
52#include <linux/prefetch.h> 52#include <linux/prefetch.h>
53#include <linux/delay.h> 53#include <linux/delay.h>
54#include <linux/stop_machine.h> 54#include <linux/stop_machine.h>
55#include <linux/random.h>
55 56
56#include "rcutree.h" 57#include "rcutree.h"
57#include <trace/events/rcu.h> 58#include <trace/events/rcu.h>
@@ -61,6 +62,7 @@
61/* Data structures. */ 62/* Data structures. */
62 63
63static struct lock_class_key rcu_node_class[RCU_NUM_LVLS]; 64static struct lock_class_key rcu_node_class[RCU_NUM_LVLS];
65static struct lock_class_key rcu_fqs_class[RCU_NUM_LVLS];
64 66
65#define RCU_STATE_INITIALIZER(sname, cr) { \ 67#define RCU_STATE_INITIALIZER(sname, cr) { \
66 .level = { &sname##_state.node[0] }, \ 68 .level = { &sname##_state.node[0] }, \
@@ -72,7 +74,7 @@ static struct lock_class_key rcu_node_class[RCU_NUM_LVLS];
72 .orphan_nxttail = &sname##_state.orphan_nxtlist, \ 74 .orphan_nxttail = &sname##_state.orphan_nxtlist, \
73 .orphan_donetail = &sname##_state.orphan_donelist, \ 75 .orphan_donetail = &sname##_state.orphan_donelist, \
74 .barrier_mutex = __MUTEX_INITIALIZER(sname##_state.barrier_mutex), \ 76 .barrier_mutex = __MUTEX_INITIALIZER(sname##_state.barrier_mutex), \
75 .fqslock = __RAW_SPIN_LOCK_UNLOCKED(&sname##_state.fqslock), \ 77 .onoff_mutex = __MUTEX_INITIALIZER(sname##_state.onoff_mutex), \
76 .name = #sname, \ 78 .name = #sname, \
77} 79}
78 80
@@ -88,7 +90,7 @@ LIST_HEAD(rcu_struct_flavors);
88 90
89/* Increase (but not decrease) the CONFIG_RCU_FANOUT_LEAF at boot time. */ 91/* Increase (but not decrease) the CONFIG_RCU_FANOUT_LEAF at boot time. */
90static int rcu_fanout_leaf = CONFIG_RCU_FANOUT_LEAF; 92static int rcu_fanout_leaf = CONFIG_RCU_FANOUT_LEAF;
91module_param(rcu_fanout_leaf, int, 0); 93module_param(rcu_fanout_leaf, int, 0444);
92int rcu_num_lvls __read_mostly = RCU_NUM_LVLS; 94int rcu_num_lvls __read_mostly = RCU_NUM_LVLS;
93static int num_rcu_lvl[] = { /* Number of rcu_nodes at specified level. */ 95static int num_rcu_lvl[] = { /* Number of rcu_nodes at specified level. */
94 NUM_RCU_LVL_0, 96 NUM_RCU_LVL_0,
@@ -133,13 +135,12 @@ static int rcu_scheduler_fully_active __read_mostly;
133 */ 135 */
134static DEFINE_PER_CPU(struct task_struct *, rcu_cpu_kthread_task); 136static DEFINE_PER_CPU(struct task_struct *, rcu_cpu_kthread_task);
135DEFINE_PER_CPU(unsigned int, rcu_cpu_kthread_status); 137DEFINE_PER_CPU(unsigned int, rcu_cpu_kthread_status);
136DEFINE_PER_CPU(int, rcu_cpu_kthread_cpu);
137DEFINE_PER_CPU(unsigned int, rcu_cpu_kthread_loops); 138DEFINE_PER_CPU(unsigned int, rcu_cpu_kthread_loops);
138DEFINE_PER_CPU(char, rcu_cpu_has_work); 139DEFINE_PER_CPU(char, rcu_cpu_has_work);
139 140
140#endif /* #ifdef CONFIG_RCU_BOOST */ 141#endif /* #ifdef CONFIG_RCU_BOOST */
141 142
142static void rcu_node_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu); 143static void rcu_boost_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu);
143static void invoke_rcu_core(void); 144static void invoke_rcu_core(void);
144static void invoke_rcu_callbacks(struct rcu_state *rsp, struct rcu_data *rdp); 145static void invoke_rcu_callbacks(struct rcu_state *rsp, struct rcu_data *rdp);
145 146
@@ -175,8 +176,6 @@ void rcu_sched_qs(int cpu)
175{ 176{
176 struct rcu_data *rdp = &per_cpu(rcu_sched_data, cpu); 177 struct rcu_data *rdp = &per_cpu(rcu_sched_data, cpu);
177 178
178 rdp->passed_quiesce_gpnum = rdp->gpnum;
179 barrier();
180 if (rdp->passed_quiesce == 0) 179 if (rdp->passed_quiesce == 0)
181 trace_rcu_grace_period("rcu_sched", rdp->gpnum, "cpuqs"); 180 trace_rcu_grace_period("rcu_sched", rdp->gpnum, "cpuqs");
182 rdp->passed_quiesce = 1; 181 rdp->passed_quiesce = 1;
@@ -186,8 +185,6 @@ void rcu_bh_qs(int cpu)
186{ 185{
187 struct rcu_data *rdp = &per_cpu(rcu_bh_data, cpu); 186 struct rcu_data *rdp = &per_cpu(rcu_bh_data, cpu);
188 187
189 rdp->passed_quiesce_gpnum = rdp->gpnum;
190 barrier();
191 if (rdp->passed_quiesce == 0) 188 if (rdp->passed_quiesce == 0)
192 trace_rcu_grace_period("rcu_bh", rdp->gpnum, "cpuqs"); 189 trace_rcu_grace_period("rcu_bh", rdp->gpnum, "cpuqs");
193 rdp->passed_quiesce = 1; 190 rdp->passed_quiesce = 1;
@@ -210,15 +207,18 @@ EXPORT_SYMBOL_GPL(rcu_note_context_switch);
210DEFINE_PER_CPU(struct rcu_dynticks, rcu_dynticks) = { 207DEFINE_PER_CPU(struct rcu_dynticks, rcu_dynticks) = {
211 .dynticks_nesting = DYNTICK_TASK_EXIT_IDLE, 208 .dynticks_nesting = DYNTICK_TASK_EXIT_IDLE,
212 .dynticks = ATOMIC_INIT(1), 209 .dynticks = ATOMIC_INIT(1),
210#if defined(CONFIG_RCU_USER_QS) && !defined(CONFIG_RCU_USER_QS_FORCE)
211 .ignore_user_qs = true,
212#endif
213}; 213};
214 214
215static int blimit = 10; /* Maximum callbacks per rcu_do_batch. */ 215static int blimit = 10; /* Maximum callbacks per rcu_do_batch. */
216static int qhimark = 10000; /* If this many pending, ignore blimit. */ 216static int qhimark = 10000; /* If this many pending, ignore blimit. */
217static int qlowmark = 100; /* Once only this many pending, use blimit. */ 217static int qlowmark = 100; /* Once only this many pending, use blimit. */
218 218
219module_param(blimit, int, 0); 219module_param(blimit, int, 0444);
220module_param(qhimark, int, 0); 220module_param(qhimark, int, 0444);
221module_param(qlowmark, int, 0); 221module_param(qlowmark, int, 0444);
222 222
223int rcu_cpu_stall_suppress __read_mostly; /* 1 = suppress stall warnings. */ 223int rcu_cpu_stall_suppress __read_mostly; /* 1 = suppress stall warnings. */
224int rcu_cpu_stall_timeout __read_mostly = CONFIG_RCU_CPU_STALL_TIMEOUT; 224int rcu_cpu_stall_timeout __read_mostly = CONFIG_RCU_CPU_STALL_TIMEOUT;
@@ -226,7 +226,14 @@ int rcu_cpu_stall_timeout __read_mostly = CONFIG_RCU_CPU_STALL_TIMEOUT;
226module_param(rcu_cpu_stall_suppress, int, 0644); 226module_param(rcu_cpu_stall_suppress, int, 0644);
227module_param(rcu_cpu_stall_timeout, int, 0644); 227module_param(rcu_cpu_stall_timeout, int, 0644);
228 228
229static void force_quiescent_state(struct rcu_state *rsp, int relaxed); 229static ulong jiffies_till_first_fqs = RCU_JIFFIES_TILL_FORCE_QS;
230static ulong jiffies_till_next_fqs = RCU_JIFFIES_TILL_FORCE_QS;
231
232module_param(jiffies_till_first_fqs, ulong, 0644);
233module_param(jiffies_till_next_fqs, ulong, 0644);
234
235static void force_qs_rnp(struct rcu_state *rsp, int (*f)(struct rcu_data *));
236static void force_quiescent_state(struct rcu_state *rsp);
230static int rcu_pending(int cpu); 237static int rcu_pending(int cpu);
231 238
232/* 239/*
@@ -252,7 +259,7 @@ EXPORT_SYMBOL_GPL(rcu_batches_completed_bh);
252 */ 259 */
253void rcu_bh_force_quiescent_state(void) 260void rcu_bh_force_quiescent_state(void)
254{ 261{
255 force_quiescent_state(&rcu_bh_state, 0); 262 force_quiescent_state(&rcu_bh_state);
256} 263}
257EXPORT_SYMBOL_GPL(rcu_bh_force_quiescent_state); 264EXPORT_SYMBOL_GPL(rcu_bh_force_quiescent_state);
258 265
@@ -286,7 +293,7 @@ EXPORT_SYMBOL_GPL(rcutorture_record_progress);
286 */ 293 */
287void rcu_sched_force_quiescent_state(void) 294void rcu_sched_force_quiescent_state(void)
288{ 295{
289 force_quiescent_state(&rcu_sched_state, 0); 296 force_quiescent_state(&rcu_sched_state);
290} 297}
291EXPORT_SYMBOL_GPL(rcu_sched_force_quiescent_state); 298EXPORT_SYMBOL_GPL(rcu_sched_force_quiescent_state);
292 299
@@ -305,7 +312,9 @@ cpu_has_callbacks_ready_to_invoke(struct rcu_data *rdp)
305static int 312static int
306cpu_needs_another_gp(struct rcu_state *rsp, struct rcu_data *rdp) 313cpu_needs_another_gp(struct rcu_state *rsp, struct rcu_data *rdp)
307{ 314{
308 return *rdp->nxttail[RCU_DONE_TAIL] && !rcu_gp_in_progress(rsp); 315 return *rdp->nxttail[RCU_DONE_TAIL +
316 ACCESS_ONCE(rsp->completed) != rdp->completed] &&
317 !rcu_gp_in_progress(rsp);
309} 318}
310 319
311/* 320/*
@@ -317,45 +326,17 @@ static struct rcu_node *rcu_get_root(struct rcu_state *rsp)
317} 326}
318 327
319/* 328/*
320 * If the specified CPU is offline, tell the caller that it is in 329 * rcu_eqs_enter_common - current CPU is moving towards extended quiescent state
321 * a quiescent state. Otherwise, whack it with a reschedule IPI.
322 * Grace periods can end up waiting on an offline CPU when that
323 * CPU is in the process of coming online -- it will be added to the
324 * rcu_node bitmasks before it actually makes it online. The same thing
325 * can happen while a CPU is in the process of coming online. Because this
326 * race is quite rare, we check for it after detecting that the grace
327 * period has been delayed rather than checking each and every CPU
328 * each and every time we start a new grace period.
329 */
330static int rcu_implicit_offline_qs(struct rcu_data *rdp)
331{
332 /*
333 * If the CPU is offline for more than a jiffy, it is in a quiescent
334 * state. We can trust its state not to change because interrupts
335 * are disabled. The reason for the jiffy's worth of slack is to
336 * handle CPUs initializing on the way up and finding their way
337 * to the idle loop on the way down.
338 */
339 if (cpu_is_offline(rdp->cpu) &&
340 ULONG_CMP_LT(rdp->rsp->gp_start + 2, jiffies)) {
341 trace_rcu_fqs(rdp->rsp->name, rdp->gpnum, rdp->cpu, "ofl");
342 rdp->offline_fqs++;
343 return 1;
344 }
345 return 0;
346}
347
348/*
349 * rcu_idle_enter_common - inform RCU that current CPU is moving towards idle
350 * 330 *
351 * If the new value of the ->dynticks_nesting counter now is zero, 331 * If the new value of the ->dynticks_nesting counter now is zero,
352 * we really have entered idle, and must do the appropriate accounting. 332 * we really have entered idle, and must do the appropriate accounting.
353 * The caller must have disabled interrupts. 333 * The caller must have disabled interrupts.
354 */ 334 */
355static void rcu_idle_enter_common(struct rcu_dynticks *rdtp, long long oldval) 335static void rcu_eqs_enter_common(struct rcu_dynticks *rdtp, long long oldval,
336 bool user)
356{ 337{
357 trace_rcu_dyntick("Start", oldval, 0); 338 trace_rcu_dyntick("Start", oldval, 0);
358 if (!is_idle_task(current)) { 339 if (!user && !is_idle_task(current)) {
359 struct task_struct *idle = idle_task(smp_processor_id()); 340 struct task_struct *idle = idle_task(smp_processor_id());
360 341
361 trace_rcu_dyntick("Error on entry: not idle task", oldval, 0); 342 trace_rcu_dyntick("Error on entry: not idle task", oldval, 0);
@@ -372,7 +353,7 @@ static void rcu_idle_enter_common(struct rcu_dynticks *rdtp, long long oldval)
372 WARN_ON_ONCE(atomic_read(&rdtp->dynticks) & 0x1); 353 WARN_ON_ONCE(atomic_read(&rdtp->dynticks) & 0x1);
373 354
374 /* 355 /*
375 * The idle task is not permitted to enter the idle loop while 356 * It is illegal to enter an extended quiescent state while
376 * in an RCU read-side critical section. 357 * in an RCU read-side critical section.
377 */ 358 */
378 rcu_lockdep_assert(!lock_is_held(&rcu_lock_map), 359 rcu_lockdep_assert(!lock_is_held(&rcu_lock_map),
@@ -383,6 +364,25 @@ static void rcu_idle_enter_common(struct rcu_dynticks *rdtp, long long oldval)
383 "Illegal idle entry in RCU-sched read-side critical section."); 364 "Illegal idle entry in RCU-sched read-side critical section.");
384} 365}
385 366
367/*
368 * Enter an RCU extended quiescent state, which can be either the
369 * idle loop or adaptive-tickless usermode execution.
370 */
371static void rcu_eqs_enter(bool user)
372{
373 long long oldval;
374 struct rcu_dynticks *rdtp;
375
376 rdtp = &__get_cpu_var(rcu_dynticks);
377 oldval = rdtp->dynticks_nesting;
378 WARN_ON_ONCE((oldval & DYNTICK_TASK_NEST_MASK) == 0);
379 if ((oldval & DYNTICK_TASK_NEST_MASK) == DYNTICK_TASK_NEST_VALUE)
380 rdtp->dynticks_nesting = 0;
381 else
382 rdtp->dynticks_nesting -= DYNTICK_TASK_NEST_VALUE;
383 rcu_eqs_enter_common(rdtp, oldval, user);
384}
385
386/** 386/**
387 * rcu_idle_enter - inform RCU that current CPU is entering idle 387 * rcu_idle_enter - inform RCU that current CPU is entering idle
388 * 388 *
@@ -398,21 +398,70 @@ static void rcu_idle_enter_common(struct rcu_dynticks *rdtp, long long oldval)
398void rcu_idle_enter(void) 398void rcu_idle_enter(void)
399{ 399{
400 unsigned long flags; 400 unsigned long flags;
401 long long oldval; 401
402 local_irq_save(flags);
403 rcu_eqs_enter(false);
404 local_irq_restore(flags);
405}
406EXPORT_SYMBOL_GPL(rcu_idle_enter);
407
408#ifdef CONFIG_RCU_USER_QS
409/**
410 * rcu_user_enter - inform RCU that we are resuming userspace.
411 *
412 * Enter RCU idle mode right before resuming userspace. No use of RCU
413 * is permitted between this call and rcu_user_exit(). This way the
414 * CPU doesn't need to maintain the tick for RCU maintenance purposes
415 * when the CPU runs in userspace.
416 */
417void rcu_user_enter(void)
418{
419 unsigned long flags;
402 struct rcu_dynticks *rdtp; 420 struct rcu_dynticks *rdtp;
403 421
422 /*
423 * Some contexts may involve an exception occuring in an irq,
424 * leading to that nesting:
425 * rcu_irq_enter() rcu_user_exit() rcu_user_exit() rcu_irq_exit()
426 * This would mess up the dyntick_nesting count though. And rcu_irq_*()
427 * helpers are enough to protect RCU uses inside the exception. So
428 * just return immediately if we detect we are in an IRQ.
429 */
430 if (in_interrupt())
431 return;
432
433 WARN_ON_ONCE(!current->mm);
434
404 local_irq_save(flags); 435 local_irq_save(flags);
405 rdtp = &__get_cpu_var(rcu_dynticks); 436 rdtp = &__get_cpu_var(rcu_dynticks);
406 oldval = rdtp->dynticks_nesting; 437 if (!rdtp->ignore_user_qs && !rdtp->in_user) {
407 WARN_ON_ONCE((oldval & DYNTICK_TASK_NEST_MASK) == 0); 438 rdtp->in_user = true;
408 if ((oldval & DYNTICK_TASK_NEST_MASK) == DYNTICK_TASK_NEST_VALUE) 439 rcu_eqs_enter(true);
409 rdtp->dynticks_nesting = 0; 440 }
410 else
411 rdtp->dynticks_nesting -= DYNTICK_TASK_NEST_VALUE;
412 rcu_idle_enter_common(rdtp, oldval);
413 local_irq_restore(flags); 441 local_irq_restore(flags);
414} 442}
415EXPORT_SYMBOL_GPL(rcu_idle_enter); 443
444/**
445 * rcu_user_enter_after_irq - inform RCU that we are going to resume userspace
446 * after the current irq returns.
447 *
448 * This is similar to rcu_user_enter() but in the context of a non-nesting
449 * irq. After this call, RCU enters into idle mode when the interrupt
450 * returns.
451 */
452void rcu_user_enter_after_irq(void)
453{
454 unsigned long flags;
455 struct rcu_dynticks *rdtp;
456
457 local_irq_save(flags);
458 rdtp = &__get_cpu_var(rcu_dynticks);
459 /* Ensure this irq is interrupting a non-idle RCU state. */
460 WARN_ON_ONCE(!(rdtp->dynticks_nesting & DYNTICK_TASK_MASK));
461 rdtp->dynticks_nesting = 1;
462 local_irq_restore(flags);
463}
464#endif /* CONFIG_RCU_USER_QS */
416 465
417/** 466/**
418 * rcu_irq_exit - inform RCU that current CPU is exiting irq towards idle 467 * rcu_irq_exit - inform RCU that current CPU is exiting irq towards idle
@@ -444,18 +493,19 @@ void rcu_irq_exit(void)
444 if (rdtp->dynticks_nesting) 493 if (rdtp->dynticks_nesting)
445 trace_rcu_dyntick("--=", oldval, rdtp->dynticks_nesting); 494 trace_rcu_dyntick("--=", oldval, rdtp->dynticks_nesting);
446 else 495 else
447 rcu_idle_enter_common(rdtp, oldval); 496 rcu_eqs_enter_common(rdtp, oldval, true);
448 local_irq_restore(flags); 497 local_irq_restore(flags);
449} 498}
450 499
451/* 500/*
452 * rcu_idle_exit_common - inform RCU that current CPU is moving away from idle 501 * rcu_eqs_exit_common - current CPU moving away from extended quiescent state
453 * 502 *
454 * If the new value of the ->dynticks_nesting counter was previously zero, 503 * If the new value of the ->dynticks_nesting counter was previously zero,
455 * we really have exited idle, and must do the appropriate accounting. 504 * we really have exited idle, and must do the appropriate accounting.
456 * The caller must have disabled interrupts. 505 * The caller must have disabled interrupts.
457 */ 506 */
458static void rcu_idle_exit_common(struct rcu_dynticks *rdtp, long long oldval) 507static void rcu_eqs_exit_common(struct rcu_dynticks *rdtp, long long oldval,
508 int user)
459{ 509{
460 smp_mb__before_atomic_inc(); /* Force ordering w/previous sojourn. */ 510 smp_mb__before_atomic_inc(); /* Force ordering w/previous sojourn. */
461 atomic_inc(&rdtp->dynticks); 511 atomic_inc(&rdtp->dynticks);
@@ -464,7 +514,7 @@ static void rcu_idle_exit_common(struct rcu_dynticks *rdtp, long long oldval)
464 WARN_ON_ONCE(!(atomic_read(&rdtp->dynticks) & 0x1)); 514 WARN_ON_ONCE(!(atomic_read(&rdtp->dynticks) & 0x1));
465 rcu_cleanup_after_idle(smp_processor_id()); 515 rcu_cleanup_after_idle(smp_processor_id());
466 trace_rcu_dyntick("End", oldval, rdtp->dynticks_nesting); 516 trace_rcu_dyntick("End", oldval, rdtp->dynticks_nesting);
467 if (!is_idle_task(current)) { 517 if (!user && !is_idle_task(current)) {
468 struct task_struct *idle = idle_task(smp_processor_id()); 518 struct task_struct *idle = idle_task(smp_processor_id());
469 519
470 trace_rcu_dyntick("Error on exit: not idle task", 520 trace_rcu_dyntick("Error on exit: not idle task",
@@ -476,6 +526,25 @@ static void rcu_idle_exit_common(struct rcu_dynticks *rdtp, long long oldval)
476 } 526 }
477} 527}
478 528
529/*
530 * Exit an RCU extended quiescent state, which can be either the
531 * idle loop or adaptive-tickless usermode execution.
532 */
533static void rcu_eqs_exit(bool user)
534{
535 struct rcu_dynticks *rdtp;
536 long long oldval;
537
538 rdtp = &__get_cpu_var(rcu_dynticks);
539 oldval = rdtp->dynticks_nesting;
540 WARN_ON_ONCE(oldval < 0);
541 if (oldval & DYNTICK_TASK_NEST_MASK)
542 rdtp->dynticks_nesting += DYNTICK_TASK_NEST_VALUE;
543 else
544 rdtp->dynticks_nesting = DYNTICK_TASK_EXIT_IDLE;
545 rcu_eqs_exit_common(rdtp, oldval, user);
546}
547
479/** 548/**
480 * rcu_idle_exit - inform RCU that current CPU is leaving idle 549 * rcu_idle_exit - inform RCU that current CPU is leaving idle
481 * 550 *
@@ -490,21 +559,67 @@ static void rcu_idle_exit_common(struct rcu_dynticks *rdtp, long long oldval)
490void rcu_idle_exit(void) 559void rcu_idle_exit(void)
491{ 560{
492 unsigned long flags; 561 unsigned long flags;
562
563 local_irq_save(flags);
564 rcu_eqs_exit(false);
565 local_irq_restore(flags);
566}
567EXPORT_SYMBOL_GPL(rcu_idle_exit);
568
569#ifdef CONFIG_RCU_USER_QS
570/**
571 * rcu_user_exit - inform RCU that we are exiting userspace.
572 *
573 * Exit RCU idle mode while entering the kernel because it can
574 * run a RCU read side critical section anytime.
575 */
576void rcu_user_exit(void)
577{
578 unsigned long flags;
493 struct rcu_dynticks *rdtp; 579 struct rcu_dynticks *rdtp;
494 long long oldval; 580
581 /*
582 * Some contexts may involve an exception occuring in an irq,
583 * leading to that nesting:
584 * rcu_irq_enter() rcu_user_exit() rcu_user_exit() rcu_irq_exit()
585 * This would mess up the dyntick_nesting count though. And rcu_irq_*()
586 * helpers are enough to protect RCU uses inside the exception. So
587 * just return immediately if we detect we are in an IRQ.
588 */
589 if (in_interrupt())
590 return;
495 591
496 local_irq_save(flags); 592 local_irq_save(flags);
497 rdtp = &__get_cpu_var(rcu_dynticks); 593 rdtp = &__get_cpu_var(rcu_dynticks);
498 oldval = rdtp->dynticks_nesting; 594 if (rdtp->in_user) {
499 WARN_ON_ONCE(oldval < 0); 595 rdtp->in_user = false;
500 if (oldval & DYNTICK_TASK_NEST_MASK) 596 rcu_eqs_exit(true);
501 rdtp->dynticks_nesting += DYNTICK_TASK_NEST_VALUE; 597 }
502 else
503 rdtp->dynticks_nesting = DYNTICK_TASK_EXIT_IDLE;
504 rcu_idle_exit_common(rdtp, oldval);
505 local_irq_restore(flags); 598 local_irq_restore(flags);
506} 599}
507EXPORT_SYMBOL_GPL(rcu_idle_exit); 600
601/**
602 * rcu_user_exit_after_irq - inform RCU that we won't resume to userspace
603 * idle mode after the current non-nesting irq returns.
604 *
605 * This is similar to rcu_user_exit() but in the context of an irq.
606 * This is called when the irq has interrupted a userspace RCU idle mode
607 * context. When the current non-nesting interrupt returns after this call,
608 * the CPU won't restore the RCU idle mode.
609 */
610void rcu_user_exit_after_irq(void)
611{
612 unsigned long flags;
613 struct rcu_dynticks *rdtp;
614
615 local_irq_save(flags);
616 rdtp = &__get_cpu_var(rcu_dynticks);
617 /* Ensure we are interrupting an RCU idle mode. */
618 WARN_ON_ONCE(rdtp->dynticks_nesting & DYNTICK_TASK_NEST_MASK);
619 rdtp->dynticks_nesting += DYNTICK_TASK_EXIT_IDLE;
620 local_irq_restore(flags);
621}
622#endif /* CONFIG_RCU_USER_QS */
508 623
509/** 624/**
510 * rcu_irq_enter - inform RCU that current CPU is entering irq away from idle 625 * rcu_irq_enter - inform RCU that current CPU is entering irq away from idle
@@ -539,7 +654,7 @@ void rcu_irq_enter(void)
539 if (oldval) 654 if (oldval)
540 trace_rcu_dyntick("++=", oldval, rdtp->dynticks_nesting); 655 trace_rcu_dyntick("++=", oldval, rdtp->dynticks_nesting);
541 else 656 else
542 rcu_idle_exit_common(rdtp, oldval); 657 rcu_eqs_exit_common(rdtp, oldval, true);
543 local_irq_restore(flags); 658 local_irq_restore(flags);
544} 659}
545 660
@@ -603,6 +718,21 @@ int rcu_is_cpu_idle(void)
603} 718}
604EXPORT_SYMBOL(rcu_is_cpu_idle); 719EXPORT_SYMBOL(rcu_is_cpu_idle);
605 720
721#ifdef CONFIG_RCU_USER_QS
722void rcu_user_hooks_switch(struct task_struct *prev,
723 struct task_struct *next)
724{
725 struct rcu_dynticks *rdtp;
726
727 /* Interrupts are disabled in context switch */
728 rdtp = &__get_cpu_var(rcu_dynticks);
729 if (!rdtp->ignore_user_qs) {
730 clear_tsk_thread_flag(prev, TIF_NOHZ);
731 set_tsk_thread_flag(next, TIF_NOHZ);
732 }
733}
734#endif /* #ifdef CONFIG_RCU_USER_QS */
735
606#if defined(CONFIG_PROVE_RCU) && defined(CONFIG_HOTPLUG_CPU) 736#if defined(CONFIG_PROVE_RCU) && defined(CONFIG_HOTPLUG_CPU)
607 737
608/* 738/*
@@ -673,7 +803,7 @@ static int dyntick_save_progress_counter(struct rcu_data *rdp)
673 * Return true if the specified CPU has passed through a quiescent 803 * Return true if the specified CPU has passed through a quiescent
674 * state by virtue of being in or having passed through an dynticks 804 * state by virtue of being in or having passed through an dynticks
675 * idle state since the last call to dyntick_save_progress_counter() 805 * idle state since the last call to dyntick_save_progress_counter()
676 * for this same CPU. 806 * for this same CPU, or by virtue of having been offline.
677 */ 807 */
678static int rcu_implicit_dynticks_qs(struct rcu_data *rdp) 808static int rcu_implicit_dynticks_qs(struct rcu_data *rdp)
679{ 809{
@@ -697,8 +827,26 @@ static int rcu_implicit_dynticks_qs(struct rcu_data *rdp)
697 return 1; 827 return 1;
698 } 828 }
699 829
700 /* Go check for the CPU being offline. */ 830 /*
701 return rcu_implicit_offline_qs(rdp); 831 * Check for the CPU being offline, but only if the grace period
832 * is old enough. We don't need to worry about the CPU changing
833 * state: If we see it offline even once, it has been through a
834 * quiescent state.
835 *
836 * The reason for insisting that the grace period be at least
837 * one jiffy old is that CPUs that are not quite online and that
838 * have just gone offline can still execute RCU read-side critical
839 * sections.
840 */
841 if (ULONG_CMP_GE(rdp->rsp->gp_start + 2, jiffies))
842 return 0; /* Grace period is not old enough. */
843 barrier();
844 if (cpu_is_offline(rdp->cpu)) {
845 trace_rcu_fqs(rdp->rsp->name, rdp->gpnum, rdp->cpu, "ofl");
846 rdp->offline_fqs++;
847 return 1;
848 }
849 return 0;
702} 850}
703 851
704static int jiffies_till_stall_check(void) 852static int jiffies_till_stall_check(void)
@@ -755,14 +903,15 @@ static void print_other_cpu_stall(struct rcu_state *rsp)
755 rcu_for_each_leaf_node(rsp, rnp) { 903 rcu_for_each_leaf_node(rsp, rnp) {
756 raw_spin_lock_irqsave(&rnp->lock, flags); 904 raw_spin_lock_irqsave(&rnp->lock, flags);
757 ndetected += rcu_print_task_stall(rnp); 905 ndetected += rcu_print_task_stall(rnp);
906 if (rnp->qsmask != 0) {
907 for (cpu = 0; cpu <= rnp->grphi - rnp->grplo; cpu++)
908 if (rnp->qsmask & (1UL << cpu)) {
909 print_cpu_stall_info(rsp,
910 rnp->grplo + cpu);
911 ndetected++;
912 }
913 }
758 raw_spin_unlock_irqrestore(&rnp->lock, flags); 914 raw_spin_unlock_irqrestore(&rnp->lock, flags);
759 if (rnp->qsmask == 0)
760 continue;
761 for (cpu = 0; cpu <= rnp->grphi - rnp->grplo; cpu++)
762 if (rnp->qsmask & (1UL << cpu)) {
763 print_cpu_stall_info(rsp, rnp->grplo + cpu);
764 ndetected++;
765 }
766 } 915 }
767 916
768 /* 917 /*
@@ -782,11 +931,11 @@ static void print_other_cpu_stall(struct rcu_state *rsp)
782 else if (!trigger_all_cpu_backtrace()) 931 else if (!trigger_all_cpu_backtrace())
783 dump_stack(); 932 dump_stack();
784 933
785 /* If so configured, complain about tasks blocking the grace period. */ 934 /* Complain about tasks blocking the grace period. */
786 935
787 rcu_print_detail_task_stall(rsp); 936 rcu_print_detail_task_stall(rsp);
788 937
789 force_quiescent_state(rsp, 0); /* Kick them all. */ 938 force_quiescent_state(rsp); /* Kick them all. */
790} 939}
791 940
792static void print_cpu_stall(struct rcu_state *rsp) 941static void print_cpu_stall(struct rcu_state *rsp)
@@ -827,7 +976,8 @@ static void check_cpu_stall(struct rcu_state *rsp, struct rcu_data *rdp)
827 j = ACCESS_ONCE(jiffies); 976 j = ACCESS_ONCE(jiffies);
828 js = ACCESS_ONCE(rsp->jiffies_stall); 977 js = ACCESS_ONCE(rsp->jiffies_stall);
829 rnp = rdp->mynode; 978 rnp = rdp->mynode;
830 if ((ACCESS_ONCE(rnp->qsmask) & rdp->grpmask) && ULONG_CMP_GE(j, js)) { 979 if (rcu_gp_in_progress(rsp) &&
980 (ACCESS_ONCE(rnp->qsmask) & rdp->grpmask) && ULONG_CMP_GE(j, js)) {
831 981
832 /* We haven't checked in, so go dump stack. */ 982 /* We haven't checked in, so go dump stack. */
833 print_cpu_stall(rsp); 983 print_cpu_stall(rsp);
@@ -889,12 +1039,8 @@ static void __note_new_gpnum(struct rcu_state *rsp, struct rcu_node *rnp, struct
889 */ 1039 */
890 rdp->gpnum = rnp->gpnum; 1040 rdp->gpnum = rnp->gpnum;
891 trace_rcu_grace_period(rsp->name, rdp->gpnum, "cpustart"); 1041 trace_rcu_grace_period(rsp->name, rdp->gpnum, "cpustart");
892 if (rnp->qsmask & rdp->grpmask) { 1042 rdp->passed_quiesce = 0;
893 rdp->qs_pending = 1; 1043 rdp->qs_pending = !!(rnp->qsmask & rdp->grpmask);
894 rdp->passed_quiesce = 0;
895 } else {
896 rdp->qs_pending = 0;
897 }
898 zero_cpu_stall_ticks(rdp); 1044 zero_cpu_stall_ticks(rdp);
899 } 1045 }
900} 1046}
@@ -974,10 +1120,13 @@ __rcu_process_gp_end(struct rcu_state *rsp, struct rcu_node *rnp, struct rcu_dat
974 * our behalf. Catch up with this state to avoid noting 1120 * our behalf. Catch up with this state to avoid noting
975 * spurious new grace periods. If another grace period 1121 * spurious new grace periods. If another grace period
976 * has started, then rnp->gpnum will have advanced, so 1122 * has started, then rnp->gpnum will have advanced, so
977 * we will detect this later on. 1123 * we will detect this later on. Of course, any quiescent
1124 * states we found for the old GP are now invalid.
978 */ 1125 */
979 if (ULONG_CMP_LT(rdp->gpnum, rdp->completed)) 1126 if (ULONG_CMP_LT(rdp->gpnum, rdp->completed)) {
980 rdp->gpnum = rdp->completed; 1127 rdp->gpnum = rdp->completed;
1128 rdp->passed_quiesce = 0;
1129 }
981 1130
982 /* 1131 /*
983 * If RCU does not need a quiescent state from this CPU, 1132 * If RCU does not need a quiescent state from this CPU,
@@ -1021,97 +1170,56 @@ rcu_start_gp_per_cpu(struct rcu_state *rsp, struct rcu_node *rnp, struct rcu_dat
1021 /* Prior grace period ended, so advance callbacks for current CPU. */ 1170 /* Prior grace period ended, so advance callbacks for current CPU. */
1022 __rcu_process_gp_end(rsp, rnp, rdp); 1171 __rcu_process_gp_end(rsp, rnp, rdp);
1023 1172
1024 /*
1025 * Because this CPU just now started the new grace period, we know
1026 * that all of its callbacks will be covered by this upcoming grace
1027 * period, even the ones that were registered arbitrarily recently.
1028 * Therefore, advance all outstanding callbacks to RCU_WAIT_TAIL.
1029 *
1030 * Other CPUs cannot be sure exactly when the grace period started.
1031 * Therefore, their recently registered callbacks must pass through
1032 * an additional RCU_NEXT_READY stage, so that they will be handled
1033 * by the next RCU grace period.
1034 */
1035 rdp->nxttail[RCU_NEXT_READY_TAIL] = rdp->nxttail[RCU_NEXT_TAIL];
1036 rdp->nxttail[RCU_WAIT_TAIL] = rdp->nxttail[RCU_NEXT_TAIL];
1037
1038 /* Set state so that this CPU will detect the next quiescent state. */ 1173 /* Set state so that this CPU will detect the next quiescent state. */
1039 __note_new_gpnum(rsp, rnp, rdp); 1174 __note_new_gpnum(rsp, rnp, rdp);
1040} 1175}
1041 1176
1042/* 1177/*
1043 * Start a new RCU grace period if warranted, re-initializing the hierarchy 1178 * Initialize a new grace period.
1044 * in preparation for detecting the next grace period. The caller must hold
1045 * the root node's ->lock, which is released before return. Hard irqs must
1046 * be disabled.
1047 *
1048 * Note that it is legal for a dying CPU (which is marked as offline) to
1049 * invoke this function. This can happen when the dying CPU reports its
1050 * quiescent state.
1051 */ 1179 */
1052static void 1180static int rcu_gp_init(struct rcu_state *rsp)
1053rcu_start_gp(struct rcu_state *rsp, unsigned long flags)
1054 __releases(rcu_get_root(rsp)->lock)
1055{ 1181{
1056 struct rcu_data *rdp = this_cpu_ptr(rsp->rda); 1182 struct rcu_data *rdp;
1057 struct rcu_node *rnp = rcu_get_root(rsp); 1183 struct rcu_node *rnp = rcu_get_root(rsp);
1058 1184
1059 if (!rcu_scheduler_fully_active || 1185 raw_spin_lock_irq(&rnp->lock);
1060 !cpu_needs_another_gp(rsp, rdp)) { 1186 rsp->gp_flags = 0; /* Clear all flags: New grace period. */
1061 /*
1062 * Either the scheduler hasn't yet spawned the first
1063 * non-idle task or this CPU does not need another
1064 * grace period. Either way, don't start a new grace
1065 * period.
1066 */
1067 raw_spin_unlock_irqrestore(&rnp->lock, flags);
1068 return;
1069 }
1070 1187
1071 if (rsp->fqs_active) { 1188 if (rcu_gp_in_progress(rsp)) {
1072 /* 1189 /* Grace period already in progress, don't start another. */
1073 * This CPU needs a grace period, but force_quiescent_state() 1190 raw_spin_unlock_irq(&rnp->lock);
1074 * is running. Tell it to start one on this CPU's behalf. 1191 return 0;
1075 */
1076 rsp->fqs_need_gp = 1;
1077 raw_spin_unlock_irqrestore(&rnp->lock, flags);
1078 return;
1079 } 1192 }
1080 1193
1081 /* Advance to a new grace period and initialize state. */ 1194 /* Advance to a new grace period and initialize state. */
1082 rsp->gpnum++; 1195 rsp->gpnum++;
1083 trace_rcu_grace_period(rsp->name, rsp->gpnum, "start"); 1196 trace_rcu_grace_period(rsp->name, rsp->gpnum, "start");
1084 WARN_ON_ONCE(rsp->fqs_state == RCU_GP_INIT);
1085 rsp->fqs_state = RCU_GP_INIT; /* Hold off force_quiescent_state. */
1086 rsp->jiffies_force_qs = jiffies + RCU_JIFFIES_TILL_FORCE_QS;
1087 record_gp_stall_check_time(rsp); 1197 record_gp_stall_check_time(rsp);
1088 raw_spin_unlock(&rnp->lock); /* leave irqs disabled. */ 1198 raw_spin_unlock_irq(&rnp->lock);
1089 1199
1090 /* Exclude any concurrent CPU-hotplug operations. */ 1200 /* Exclude any concurrent CPU-hotplug operations. */
1091 raw_spin_lock(&rsp->onofflock); /* irqs already disabled. */ 1201 mutex_lock(&rsp->onoff_mutex);
1092 1202
1093 /* 1203 /*
1094 * Set the quiescent-state-needed bits in all the rcu_node 1204 * Set the quiescent-state-needed bits in all the rcu_node
1095 * structures for all currently online CPUs in breadth-first 1205 * structures for all currently online CPUs in breadth-first order,
1096 * order, starting from the root rcu_node structure. This 1206 * starting from the root rcu_node structure, relying on the layout
1097 * operation relies on the layout of the hierarchy within the 1207 * of the tree within the rsp->node[] array. Note that other CPUs
1098 * rsp->node[] array. Note that other CPUs will access only 1208 * will access only the leaves of the hierarchy, thus seeing that no
1099 * the leaves of the hierarchy, which still indicate that no
1100 * grace period is in progress, at least until the corresponding 1209 * grace period is in progress, at least until the corresponding
1101 * leaf node has been initialized. In addition, we have excluded 1210 * leaf node has been initialized. In addition, we have excluded
1102 * CPU-hotplug operations. 1211 * CPU-hotplug operations.
1103 * 1212 *
1104 * Note that the grace period cannot complete until we finish 1213 * The grace period cannot complete until the initialization
1105 * the initialization process, as there will be at least one 1214 * process finishes, because this kthread handles both.
1106 * qsmask bit set in the root node until that time, namely the
1107 * one corresponding to this CPU, due to the fact that we have
1108 * irqs disabled.
1109 */ 1215 */
1110 rcu_for_each_node_breadth_first(rsp, rnp) { 1216 rcu_for_each_node_breadth_first(rsp, rnp) {
1111 raw_spin_lock(&rnp->lock); /* irqs already disabled. */ 1217 raw_spin_lock_irq(&rnp->lock);
1218 rdp = this_cpu_ptr(rsp->rda);
1112 rcu_preempt_check_blocked_tasks(rnp); 1219 rcu_preempt_check_blocked_tasks(rnp);
1113 rnp->qsmask = rnp->qsmaskinit; 1220 rnp->qsmask = rnp->qsmaskinit;
1114 rnp->gpnum = rsp->gpnum; 1221 rnp->gpnum = rsp->gpnum;
1222 WARN_ON_ONCE(rnp->completed != rsp->completed);
1115 rnp->completed = rsp->completed; 1223 rnp->completed = rsp->completed;
1116 if (rnp == rdp->mynode) 1224 if (rnp == rdp->mynode)
1117 rcu_start_gp_per_cpu(rsp, rnp, rdp); 1225 rcu_start_gp_per_cpu(rsp, rnp, rdp);
@@ -1119,37 +1227,54 @@ rcu_start_gp(struct rcu_state *rsp, unsigned long flags)
1119 trace_rcu_grace_period_init(rsp->name, rnp->gpnum, 1227 trace_rcu_grace_period_init(rsp->name, rnp->gpnum,
1120 rnp->level, rnp->grplo, 1228 rnp->level, rnp->grplo,
1121 rnp->grphi, rnp->qsmask); 1229 rnp->grphi, rnp->qsmask);
1122 raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ 1230 raw_spin_unlock_irq(&rnp->lock);
1231#ifdef CONFIG_PROVE_RCU_DELAY
1232 if ((random32() % (rcu_num_nodes * 8)) == 0)
1233 schedule_timeout_uninterruptible(2);
1234#endif /* #ifdef CONFIG_PROVE_RCU_DELAY */
1235 cond_resched();
1123 } 1236 }
1124 1237
1125 rnp = rcu_get_root(rsp); 1238 mutex_unlock(&rsp->onoff_mutex);
1126 raw_spin_lock(&rnp->lock); /* irqs already disabled. */ 1239 return 1;
1127 rsp->fqs_state = RCU_SIGNAL_INIT; /* force_quiescent_state now OK. */
1128 raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */
1129 raw_spin_unlock_irqrestore(&rsp->onofflock, flags);
1130} 1240}
1131 1241
1132/* 1242/*
1133 * Report a full set of quiescent states to the specified rcu_state 1243 * Do one round of quiescent-state forcing.
1134 * data structure. This involves cleaning up after the prior grace
1135 * period and letting rcu_start_gp() start up the next grace period
1136 * if one is needed. Note that the caller must hold rnp->lock, as
1137 * required by rcu_start_gp(), which will release it.
1138 */ 1244 */
1139static void rcu_report_qs_rsp(struct rcu_state *rsp, unsigned long flags) 1245int rcu_gp_fqs(struct rcu_state *rsp, int fqs_state_in)
1140 __releases(rcu_get_root(rsp)->lock)
1141{ 1246{
1142 unsigned long gp_duration; 1247 int fqs_state = fqs_state_in;
1143 struct rcu_node *rnp = rcu_get_root(rsp); 1248 struct rcu_node *rnp = rcu_get_root(rsp);
1144 struct rcu_data *rdp = this_cpu_ptr(rsp->rda);
1145 1249
1146 WARN_ON_ONCE(!rcu_gp_in_progress(rsp)); 1250 rsp->n_force_qs++;
1251 if (fqs_state == RCU_SAVE_DYNTICK) {
1252 /* Collect dyntick-idle snapshots. */
1253 force_qs_rnp(rsp, dyntick_save_progress_counter);
1254 fqs_state = RCU_FORCE_QS;
1255 } else {
1256 /* Handle dyntick-idle and offline CPUs. */
1257 force_qs_rnp(rsp, rcu_implicit_dynticks_qs);
1258 }
1259 /* Clear flag to prevent immediate re-entry. */
1260 if (ACCESS_ONCE(rsp->gp_flags) & RCU_GP_FLAG_FQS) {
1261 raw_spin_lock_irq(&rnp->lock);
1262 rsp->gp_flags &= ~RCU_GP_FLAG_FQS;
1263 raw_spin_unlock_irq(&rnp->lock);
1264 }
1265 return fqs_state;
1266}
1147 1267
1148 /* 1268/*
1149 * Ensure that all grace-period and pre-grace-period activity 1269 * Clean up after the old grace period.
1150 * is seen before the assignment to rsp->completed. 1270 */
1151 */ 1271static void rcu_gp_cleanup(struct rcu_state *rsp)
1152 smp_mb(); /* See above block comment. */ 1272{
1273 unsigned long gp_duration;
1274 struct rcu_data *rdp;
1275 struct rcu_node *rnp = rcu_get_root(rsp);
1276
1277 raw_spin_lock_irq(&rnp->lock);
1153 gp_duration = jiffies - rsp->gp_start; 1278 gp_duration = jiffies - rsp->gp_start;
1154 if (gp_duration > rsp->gp_max) 1279 if (gp_duration > rsp->gp_max)
1155 rsp->gp_max = gp_duration; 1280 rsp->gp_max = gp_duration;
@@ -1161,35 +1286,149 @@ static void rcu_report_qs_rsp(struct rcu_state *rsp, unsigned long flags)
1161 * they can do to advance the grace period. It is therefore 1286 * they can do to advance the grace period. It is therefore
1162 * safe for us to drop the lock in order to mark the grace 1287 * safe for us to drop the lock in order to mark the grace
1163 * period as completed in all of the rcu_node structures. 1288 * period as completed in all of the rcu_node structures.
1164 *
1165 * But if this CPU needs another grace period, it will take
1166 * care of this while initializing the next grace period.
1167 * We use RCU_WAIT_TAIL instead of the usual RCU_DONE_TAIL
1168 * because the callbacks have not yet been advanced: Those
1169 * callbacks are waiting on the grace period that just now
1170 * completed.
1171 */ 1289 */
1172 if (*rdp->nxttail[RCU_WAIT_TAIL] == NULL) { 1290 raw_spin_unlock_irq(&rnp->lock);
1173 raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */
1174 1291
1175 /* 1292 /*
1176 * Propagate new ->completed value to rcu_node structures 1293 * Propagate new ->completed value to rcu_node structures so
1177 * so that other CPUs don't have to wait until the start 1294 * that other CPUs don't have to wait until the start of the next
1178 * of the next grace period to process their callbacks. 1295 * grace period to process their callbacks. This also avoids
1179 */ 1296 * some nasty RCU grace-period initialization races by forcing
1180 rcu_for_each_node_breadth_first(rsp, rnp) { 1297 * the end of the current grace period to be completely recorded in
1181 raw_spin_lock(&rnp->lock); /* irqs already disabled. */ 1298 * all of the rcu_node structures before the beginning of the next
1182 rnp->completed = rsp->gpnum; 1299 * grace period is recorded in any of the rcu_node structures.
1183 raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ 1300 */
1184 } 1301 rcu_for_each_node_breadth_first(rsp, rnp) {
1185 rnp = rcu_get_root(rsp); 1302 raw_spin_lock_irq(&rnp->lock);
1186 raw_spin_lock(&rnp->lock); /* irqs already disabled. */ 1303 rnp->completed = rsp->gpnum;
1304 raw_spin_unlock_irq(&rnp->lock);
1305 cond_resched();
1187 } 1306 }
1307 rnp = rcu_get_root(rsp);
1308 raw_spin_lock_irq(&rnp->lock);
1188 1309
1189 rsp->completed = rsp->gpnum; /* Declare the grace period complete. */ 1310 rsp->completed = rsp->gpnum; /* Declare grace period done. */
1190 trace_rcu_grace_period(rsp->name, rsp->completed, "end"); 1311 trace_rcu_grace_period(rsp->name, rsp->completed, "end");
1191 rsp->fqs_state = RCU_GP_IDLE; 1312 rsp->fqs_state = RCU_GP_IDLE;
1192 rcu_start_gp(rsp, flags); /* releases root node's rnp->lock. */ 1313 rdp = this_cpu_ptr(rsp->rda);
1314 if (cpu_needs_another_gp(rsp, rdp))
1315 rsp->gp_flags = 1;
1316 raw_spin_unlock_irq(&rnp->lock);
1317}
1318
1319/*
1320 * Body of kthread that handles grace periods.
1321 */
1322static int __noreturn rcu_gp_kthread(void *arg)
1323{
1324 int fqs_state;
1325 unsigned long j;
1326 int ret;
1327 struct rcu_state *rsp = arg;
1328 struct rcu_node *rnp = rcu_get_root(rsp);
1329
1330 for (;;) {
1331
1332 /* Handle grace-period start. */
1333 for (;;) {
1334 wait_event_interruptible(rsp->gp_wq,
1335 rsp->gp_flags &
1336 RCU_GP_FLAG_INIT);
1337 if ((rsp->gp_flags & RCU_GP_FLAG_INIT) &&
1338 rcu_gp_init(rsp))
1339 break;
1340 cond_resched();
1341 flush_signals(current);
1342 }
1343
1344 /* Handle quiescent-state forcing. */
1345 fqs_state = RCU_SAVE_DYNTICK;
1346 j = jiffies_till_first_fqs;
1347 if (j > HZ) {
1348 j = HZ;
1349 jiffies_till_first_fqs = HZ;
1350 }
1351 for (;;) {
1352 rsp->jiffies_force_qs = jiffies + j;
1353 ret = wait_event_interruptible_timeout(rsp->gp_wq,
1354 (rsp->gp_flags & RCU_GP_FLAG_FQS) ||
1355 (!ACCESS_ONCE(rnp->qsmask) &&
1356 !rcu_preempt_blocked_readers_cgp(rnp)),
1357 j);
1358 /* If grace period done, leave loop. */
1359 if (!ACCESS_ONCE(rnp->qsmask) &&
1360 !rcu_preempt_blocked_readers_cgp(rnp))
1361 break;
1362 /* If time for quiescent-state forcing, do it. */
1363 if (ret == 0 || (rsp->gp_flags & RCU_GP_FLAG_FQS)) {
1364 fqs_state = rcu_gp_fqs(rsp, fqs_state);
1365 cond_resched();
1366 } else {
1367 /* Deal with stray signal. */
1368 cond_resched();
1369 flush_signals(current);
1370 }
1371 j = jiffies_till_next_fqs;
1372 if (j > HZ) {
1373 j = HZ;
1374 jiffies_till_next_fqs = HZ;
1375 } else if (j < 1) {
1376 j = 1;
1377 jiffies_till_next_fqs = 1;
1378 }
1379 }
1380
1381 /* Handle grace-period end. */
1382 rcu_gp_cleanup(rsp);
1383 }
1384}
1385
1386/*
1387 * Start a new RCU grace period if warranted, re-initializing the hierarchy
1388 * in preparation for detecting the next grace period. The caller must hold
1389 * the root node's ->lock, which is released before return. Hard irqs must
1390 * be disabled.
1391 *
1392 * Note that it is legal for a dying CPU (which is marked as offline) to
1393 * invoke this function. This can happen when the dying CPU reports its
1394 * quiescent state.
1395 */
1396static void
1397rcu_start_gp(struct rcu_state *rsp, unsigned long flags)
1398 __releases(rcu_get_root(rsp)->lock)
1399{
1400 struct rcu_data *rdp = this_cpu_ptr(rsp->rda);
1401 struct rcu_node *rnp = rcu_get_root(rsp);
1402
1403 if (!rsp->gp_kthread ||
1404 !cpu_needs_another_gp(rsp, rdp)) {
1405 /*
1406 * Either we have not yet spawned the grace-period
1407 * task or this CPU does not need another grace period.
1408 * Either way, don't start a new grace period.
1409 */
1410 raw_spin_unlock_irqrestore(&rnp->lock, flags);
1411 return;
1412 }
1413
1414 rsp->gp_flags = RCU_GP_FLAG_INIT;
1415 raw_spin_unlock_irqrestore(&rnp->lock, flags);
1416 wake_up(&rsp->gp_wq);
1417}
1418
1419/*
1420 * Report a full set of quiescent states to the specified rcu_state
1421 * data structure. This involves cleaning up after the prior grace
1422 * period and letting rcu_start_gp() start up the next grace period
1423 * if one is needed. Note that the caller must hold rnp->lock, as
1424 * required by rcu_start_gp(), which will release it.
1425 */
1426static void rcu_report_qs_rsp(struct rcu_state *rsp, unsigned long flags)
1427 __releases(rcu_get_root(rsp)->lock)
1428{
1429 WARN_ON_ONCE(!rcu_gp_in_progress(rsp));
1430 raw_spin_unlock_irqrestore(&rcu_get_root(rsp)->lock, flags);
1431 wake_up(&rsp->gp_wq); /* Memory barrier implied by wake_up() path. */
1193} 1432}
1194 1433
1195/* 1434/*
@@ -1258,7 +1497,7 @@ rcu_report_qs_rnp(unsigned long mask, struct rcu_state *rsp,
1258 * based on quiescent states detected in an earlier grace period! 1497 * based on quiescent states detected in an earlier grace period!
1259 */ 1498 */
1260static void 1499static void
1261rcu_report_qs_rdp(int cpu, struct rcu_state *rsp, struct rcu_data *rdp, long lastgp) 1500rcu_report_qs_rdp(int cpu, struct rcu_state *rsp, struct rcu_data *rdp)
1262{ 1501{
1263 unsigned long flags; 1502 unsigned long flags;
1264 unsigned long mask; 1503 unsigned long mask;
@@ -1266,7 +1505,8 @@ rcu_report_qs_rdp(int cpu, struct rcu_state *rsp, struct rcu_data *rdp, long las
1266 1505
1267 rnp = rdp->mynode; 1506 rnp = rdp->mynode;
1268 raw_spin_lock_irqsave(&rnp->lock, flags); 1507 raw_spin_lock_irqsave(&rnp->lock, flags);
1269 if (lastgp != rnp->gpnum || rnp->completed == rnp->gpnum) { 1508 if (rdp->passed_quiesce == 0 || rdp->gpnum != rnp->gpnum ||
1509 rnp->completed == rnp->gpnum) {
1270 1510
1271 /* 1511 /*
1272 * The grace period in which this quiescent state was 1512 * The grace period in which this quiescent state was
@@ -1325,7 +1565,7 @@ rcu_check_quiescent_state(struct rcu_state *rsp, struct rcu_data *rdp)
1325 * Tell RCU we are done (but rcu_report_qs_rdp() will be the 1565 * Tell RCU we are done (but rcu_report_qs_rdp() will be the
1326 * judge of that). 1566 * judge of that).
1327 */ 1567 */
1328 rcu_report_qs_rdp(rdp->cpu, rsp, rdp, rdp->passed_quiesce_gpnum); 1568 rcu_report_qs_rdp(rdp->cpu, rsp, rdp);
1329} 1569}
1330 1570
1331#ifdef CONFIG_HOTPLUG_CPU 1571#ifdef CONFIG_HOTPLUG_CPU
@@ -1390,17 +1630,6 @@ static void rcu_adopt_orphan_cbs(struct rcu_state *rsp)
1390 int i; 1630 int i;
1391 struct rcu_data *rdp = __this_cpu_ptr(rsp->rda); 1631 struct rcu_data *rdp = __this_cpu_ptr(rsp->rda);
1392 1632
1393 /*
1394 * If there is an rcu_barrier() operation in progress, then
1395 * only the task doing that operation is permitted to adopt
1396 * callbacks. To do otherwise breaks rcu_barrier() and friends
1397 * by causing them to fail to wait for the callbacks in the
1398 * orphanage.
1399 */
1400 if (rsp->rcu_barrier_in_progress &&
1401 rsp->rcu_barrier_in_progress != current)
1402 return;
1403
1404 /* Do the accounting first. */ 1633 /* Do the accounting first. */
1405 rdp->qlen_lazy += rsp->qlen_lazy; 1634 rdp->qlen_lazy += rsp->qlen_lazy;
1406 rdp->qlen += rsp->qlen; 1635 rdp->qlen += rsp->qlen;
@@ -1455,9 +1684,8 @@ static void rcu_cleanup_dying_cpu(struct rcu_state *rsp)
1455 * The CPU has been completely removed, and some other CPU is reporting 1684 * The CPU has been completely removed, and some other CPU is reporting
1456 * this fact from process context. Do the remainder of the cleanup, 1685 * this fact from process context. Do the remainder of the cleanup,
1457 * including orphaning the outgoing CPU's RCU callbacks, and also 1686 * including orphaning the outgoing CPU's RCU callbacks, and also
1458 * adopting them, if there is no _rcu_barrier() instance running. 1687 * adopting them. There can only be one CPU hotplug operation at a time,
1459 * There can only be one CPU hotplug operation at a time, so no other 1688 * so no other CPU can be attempting to update rcu_cpu_kthread_task.
1460 * CPU can be attempting to update rcu_cpu_kthread_task.
1461 */ 1689 */
1462static void rcu_cleanup_dead_cpu(int cpu, struct rcu_state *rsp) 1690static void rcu_cleanup_dead_cpu(int cpu, struct rcu_state *rsp)
1463{ 1691{
@@ -1468,12 +1696,12 @@ static void rcu_cleanup_dead_cpu(int cpu, struct rcu_state *rsp)
1468 struct rcu_node *rnp = rdp->mynode; /* Outgoing CPU's rdp & rnp. */ 1696 struct rcu_node *rnp = rdp->mynode; /* Outgoing CPU's rdp & rnp. */
1469 1697
1470 /* Adjust any no-longer-needed kthreads. */ 1698 /* Adjust any no-longer-needed kthreads. */
1471 rcu_stop_cpu_kthread(cpu); 1699 rcu_boost_kthread_setaffinity(rnp, -1);
1472 rcu_node_kthread_setaffinity(rnp, -1);
1473 1700
1474 /* Remove the dead CPU from the bitmasks in the rcu_node hierarchy. */ 1701 /* Remove the dead CPU from the bitmasks in the rcu_node hierarchy. */
1475 1702
1476 /* Exclude any attempts to start a new grace period. */ 1703 /* Exclude any attempts to start a new grace period. */
1704 mutex_lock(&rsp->onoff_mutex);
1477 raw_spin_lock_irqsave(&rsp->onofflock, flags); 1705 raw_spin_lock_irqsave(&rsp->onofflock, flags);
1478 1706
1479 /* Orphan the dead CPU's callbacks, and adopt them if appropriate. */ 1707 /* Orphan the dead CPU's callbacks, and adopt them if appropriate. */
@@ -1515,14 +1743,14 @@ static void rcu_cleanup_dead_cpu(int cpu, struct rcu_state *rsp)
1515 WARN_ONCE(rdp->qlen != 0 || rdp->nxtlist != NULL, 1743 WARN_ONCE(rdp->qlen != 0 || rdp->nxtlist != NULL,
1516 "rcu_cleanup_dead_cpu: Callbacks on offline CPU %d: qlen=%lu, nxtlist=%p\n", 1744 "rcu_cleanup_dead_cpu: Callbacks on offline CPU %d: qlen=%lu, nxtlist=%p\n",
1517 cpu, rdp->qlen, rdp->nxtlist); 1745 cpu, rdp->qlen, rdp->nxtlist);
1746 init_callback_list(rdp);
1747 /* Disallow further callbacks on this CPU. */
1748 rdp->nxttail[RCU_NEXT_TAIL] = NULL;
1749 mutex_unlock(&rsp->onoff_mutex);
1518} 1750}
1519 1751
1520#else /* #ifdef CONFIG_HOTPLUG_CPU */ 1752#else /* #ifdef CONFIG_HOTPLUG_CPU */
1521 1753
1522static void rcu_adopt_orphan_cbs(struct rcu_state *rsp)
1523{
1524}
1525
1526static void rcu_cleanup_dying_cpu(struct rcu_state *rsp) 1754static void rcu_cleanup_dying_cpu(struct rcu_state *rsp)
1527{ 1755{
1528} 1756}
@@ -1687,6 +1915,7 @@ static void force_qs_rnp(struct rcu_state *rsp, int (*f)(struct rcu_data *))
1687 struct rcu_node *rnp; 1915 struct rcu_node *rnp;
1688 1916
1689 rcu_for_each_leaf_node(rsp, rnp) { 1917 rcu_for_each_leaf_node(rsp, rnp) {
1918 cond_resched();
1690 mask = 0; 1919 mask = 0;
1691 raw_spin_lock_irqsave(&rnp->lock, flags); 1920 raw_spin_lock_irqsave(&rnp->lock, flags);
1692 if (!rcu_gp_in_progress(rsp)) { 1921 if (!rcu_gp_in_progress(rsp)) {
@@ -1723,72 +1952,39 @@ static void force_qs_rnp(struct rcu_state *rsp, int (*f)(struct rcu_data *))
1723 * Force quiescent states on reluctant CPUs, and also detect which 1952 * Force quiescent states on reluctant CPUs, and also detect which
1724 * CPUs are in dyntick-idle mode. 1953 * CPUs are in dyntick-idle mode.
1725 */ 1954 */
1726static void force_quiescent_state(struct rcu_state *rsp, int relaxed) 1955static void force_quiescent_state(struct rcu_state *rsp)
1727{ 1956{
1728 unsigned long flags; 1957 unsigned long flags;
1729 struct rcu_node *rnp = rcu_get_root(rsp); 1958 bool ret;
1730 1959 struct rcu_node *rnp;
1731 trace_rcu_utilization("Start fqs"); 1960 struct rcu_node *rnp_old = NULL;
1732 if (!rcu_gp_in_progress(rsp)) { 1961
1733 trace_rcu_utilization("End fqs"); 1962 /* Funnel through hierarchy to reduce memory contention. */
1734 return; /* No grace period in progress, nothing to force. */ 1963 rnp = per_cpu_ptr(rsp->rda, raw_smp_processor_id())->mynode;
1735 } 1964 for (; rnp != NULL; rnp = rnp->parent) {
1736 if (!raw_spin_trylock_irqsave(&rsp->fqslock, flags)) { 1965 ret = (ACCESS_ONCE(rsp->gp_flags) & RCU_GP_FLAG_FQS) ||
1737 rsp->n_force_qs_lh++; /* Inexact, can lose counts. Tough! */ 1966 !raw_spin_trylock(&rnp->fqslock);
1738 trace_rcu_utilization("End fqs"); 1967 if (rnp_old != NULL)
1739 return; /* Someone else is already on the job. */ 1968 raw_spin_unlock(&rnp_old->fqslock);
1740 } 1969 if (ret) {
1741 if (relaxed && ULONG_CMP_GE(rsp->jiffies_force_qs, jiffies)) 1970 rsp->n_force_qs_lh++;
1742 goto unlock_fqs_ret; /* no emergency and done recently. */ 1971 return;
1743 rsp->n_force_qs++; 1972 }
1744 raw_spin_lock(&rnp->lock); /* irqs already disabled */ 1973 rnp_old = rnp;
1745 rsp->jiffies_force_qs = jiffies + RCU_JIFFIES_TILL_FORCE_QS;
1746 if(!rcu_gp_in_progress(rsp)) {
1747 rsp->n_force_qs_ngp++;
1748 raw_spin_unlock(&rnp->lock); /* irqs remain disabled */
1749 goto unlock_fqs_ret; /* no GP in progress, time updated. */
1750 }
1751 rsp->fqs_active = 1;
1752 switch (rsp->fqs_state) {
1753 case RCU_GP_IDLE:
1754 case RCU_GP_INIT:
1755
1756 break; /* grace period idle or initializing, ignore. */
1757
1758 case RCU_SAVE_DYNTICK:
1759
1760 raw_spin_unlock(&rnp->lock); /* irqs remain disabled */
1761
1762 /* Record dyntick-idle state. */
1763 force_qs_rnp(rsp, dyntick_save_progress_counter);
1764 raw_spin_lock(&rnp->lock); /* irqs already disabled */
1765 if (rcu_gp_in_progress(rsp))
1766 rsp->fqs_state = RCU_FORCE_QS;
1767 break;
1768
1769 case RCU_FORCE_QS:
1770
1771 /* Check dyntick-idle state, send IPI to laggarts. */
1772 raw_spin_unlock(&rnp->lock); /* irqs remain disabled */
1773 force_qs_rnp(rsp, rcu_implicit_dynticks_qs);
1774
1775 /* Leave state in case more forcing is required. */
1776
1777 raw_spin_lock(&rnp->lock); /* irqs already disabled */
1778 break;
1779 } 1974 }
1780 rsp->fqs_active = 0; 1975 /* rnp_old == rcu_get_root(rsp), rnp == NULL. */
1781 if (rsp->fqs_need_gp) { 1976
1782 raw_spin_unlock(&rsp->fqslock); /* irqs remain disabled */ 1977 /* Reached the root of the rcu_node tree, acquire lock. */
1783 rsp->fqs_need_gp = 0; 1978 raw_spin_lock_irqsave(&rnp_old->lock, flags);
1784 rcu_start_gp(rsp, flags); /* releases rnp->lock */ 1979 raw_spin_unlock(&rnp_old->fqslock);
1785 trace_rcu_utilization("End fqs"); 1980 if (ACCESS_ONCE(rsp->gp_flags) & RCU_GP_FLAG_FQS) {
1786 return; 1981 rsp->n_force_qs_lh++;
1982 raw_spin_unlock_irqrestore(&rnp_old->lock, flags);
1983 return; /* Someone beat us to it. */
1787 } 1984 }
1788 raw_spin_unlock(&rnp->lock); /* irqs remain disabled */ 1985 rsp->gp_flags |= RCU_GP_FLAG_FQS;
1789unlock_fqs_ret: 1986 raw_spin_unlock_irqrestore(&rnp_old->lock, flags);
1790 raw_spin_unlock_irqrestore(&rsp->fqslock, flags); 1987 wake_up(&rsp->gp_wq); /* Memory barrier implied by wake_up() path. */
1791 trace_rcu_utilization("End fqs");
1792} 1988}
1793 1989
1794/* 1990/*
@@ -1805,13 +2001,6 @@ __rcu_process_callbacks(struct rcu_state *rsp)
1805 WARN_ON_ONCE(rdp->beenonline == 0); 2001 WARN_ON_ONCE(rdp->beenonline == 0);
1806 2002
1807 /* 2003 /*
1808 * If an RCU GP has gone long enough, go check for dyntick
1809 * idle CPUs and, if needed, send resched IPIs.
1810 */
1811 if (ULONG_CMP_LT(ACCESS_ONCE(rsp->jiffies_force_qs), jiffies))
1812 force_quiescent_state(rsp, 1);
1813
1814 /*
1815 * Advance callbacks in response to end of earlier grace 2004 * Advance callbacks in response to end of earlier grace
1816 * period that some other CPU ended. 2005 * period that some other CPU ended.
1817 */ 2006 */
@@ -1838,6 +2027,8 @@ static void rcu_process_callbacks(struct softirq_action *unused)
1838{ 2027{
1839 struct rcu_state *rsp; 2028 struct rcu_state *rsp;
1840 2029
2030 if (cpu_is_offline(smp_processor_id()))
2031 return;
1841 trace_rcu_utilization("Start RCU core"); 2032 trace_rcu_utilization("Start RCU core");
1842 for_each_rcu_flavor(rsp) 2033 for_each_rcu_flavor(rsp)
1843 __rcu_process_callbacks(rsp); 2034 __rcu_process_callbacks(rsp);
@@ -1909,12 +2100,11 @@ static void __call_rcu_core(struct rcu_state *rsp, struct rcu_data *rdp,
1909 rdp->blimit = LONG_MAX; 2100 rdp->blimit = LONG_MAX;
1910 if (rsp->n_force_qs == rdp->n_force_qs_snap && 2101 if (rsp->n_force_qs == rdp->n_force_qs_snap &&
1911 *rdp->nxttail[RCU_DONE_TAIL] != head) 2102 *rdp->nxttail[RCU_DONE_TAIL] != head)
1912 force_quiescent_state(rsp, 0); 2103 force_quiescent_state(rsp);
1913 rdp->n_force_qs_snap = rsp->n_force_qs; 2104 rdp->n_force_qs_snap = rsp->n_force_qs;
1914 rdp->qlen_last_fqs_check = rdp->qlen; 2105 rdp->qlen_last_fqs_check = rdp->qlen;
1915 } 2106 }
1916 } else if (ULONG_CMP_LT(ACCESS_ONCE(rsp->jiffies_force_qs), jiffies)) 2107 }
1917 force_quiescent_state(rsp, 1);
1918} 2108}
1919 2109
1920static void 2110static void
@@ -1929,8 +2119,6 @@ __call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu),
1929 head->func = func; 2119 head->func = func;
1930 head->next = NULL; 2120 head->next = NULL;
1931 2121
1932 smp_mb(); /* Ensure RCU update seen before callback registry. */
1933
1934 /* 2122 /*
1935 * Opportunistically note grace-period endings and beginnings. 2123 * Opportunistically note grace-period endings and beginnings.
1936 * Note that we might see a beginning right after we see an 2124 * Note that we might see a beginning right after we see an
@@ -1941,6 +2129,12 @@ __call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu),
1941 rdp = this_cpu_ptr(rsp->rda); 2129 rdp = this_cpu_ptr(rsp->rda);
1942 2130
1943 /* Add the callback to our list. */ 2131 /* Add the callback to our list. */
2132 if (unlikely(rdp->nxttail[RCU_NEXT_TAIL] == NULL)) {
2133 /* _call_rcu() is illegal on offline CPU; leak the callback. */
2134 WARN_ON_ONCE(1);
2135 local_irq_restore(flags);
2136 return;
2137 }
1944 ACCESS_ONCE(rdp->qlen)++; 2138 ACCESS_ONCE(rdp->qlen)++;
1945 if (lazy) 2139 if (lazy)
1946 rdp->qlen_lazy++; 2140 rdp->qlen_lazy++;
@@ -2195,17 +2389,7 @@ static int __rcu_pending(struct rcu_state *rsp, struct rcu_data *rdp)
2195 /* Is the RCU core waiting for a quiescent state from this CPU? */ 2389 /* Is the RCU core waiting for a quiescent state from this CPU? */
2196 if (rcu_scheduler_fully_active && 2390 if (rcu_scheduler_fully_active &&
2197 rdp->qs_pending && !rdp->passed_quiesce) { 2391 rdp->qs_pending && !rdp->passed_quiesce) {
2198
2199 /*
2200 * If force_quiescent_state() coming soon and this CPU
2201 * needs a quiescent state, and this is either RCU-sched
2202 * or RCU-bh, force a local reschedule.
2203 */
2204 rdp->n_rp_qs_pending++; 2392 rdp->n_rp_qs_pending++;
2205 if (!rdp->preemptible &&
2206 ULONG_CMP_LT(ACCESS_ONCE(rsp->jiffies_force_qs) - 1,
2207 jiffies))
2208 set_need_resched();
2209 } else if (rdp->qs_pending && rdp->passed_quiesce) { 2393 } else if (rdp->qs_pending && rdp->passed_quiesce) {
2210 rdp->n_rp_report_qs++; 2394 rdp->n_rp_report_qs++;
2211 return 1; 2395 return 1;
@@ -2235,13 +2419,6 @@ static int __rcu_pending(struct rcu_state *rsp, struct rcu_data *rdp)
2235 return 1; 2419 return 1;
2236 } 2420 }
2237 2421
2238 /* Has an RCU GP gone long enough to send resched IPIs &c? */
2239 if (rcu_gp_in_progress(rsp) &&
2240 ULONG_CMP_LT(ACCESS_ONCE(rsp->jiffies_force_qs), jiffies)) {
2241 rdp->n_rp_need_fqs++;
2242 return 1;
2243 }
2244
2245 /* nothing to do */ 2422 /* nothing to do */
2246 rdp->n_rp_need_nothing++; 2423 rdp->n_rp_need_nothing++;
2247 return 0; 2424 return 0;
@@ -2326,13 +2503,10 @@ static void rcu_barrier_func(void *type)
2326static void _rcu_barrier(struct rcu_state *rsp) 2503static void _rcu_barrier(struct rcu_state *rsp)
2327{ 2504{
2328 int cpu; 2505 int cpu;
2329 unsigned long flags;
2330 struct rcu_data *rdp; 2506 struct rcu_data *rdp;
2331 struct rcu_data rd;
2332 unsigned long snap = ACCESS_ONCE(rsp->n_barrier_done); 2507 unsigned long snap = ACCESS_ONCE(rsp->n_barrier_done);
2333 unsigned long snap_done; 2508 unsigned long snap_done;
2334 2509
2335 init_rcu_head_on_stack(&rd.barrier_head);
2336 _rcu_barrier_trace(rsp, "Begin", -1, snap); 2510 _rcu_barrier_trace(rsp, "Begin", -1, snap);
2337 2511
2338 /* Take mutex to serialize concurrent rcu_barrier() requests. */ 2512 /* Take mutex to serialize concurrent rcu_barrier() requests. */
@@ -2372,70 +2546,30 @@ static void _rcu_barrier(struct rcu_state *rsp)
2372 /* 2546 /*
2373 * Initialize the count to one rather than to zero in order to 2547 * Initialize the count to one rather than to zero in order to
2374 * avoid a too-soon return to zero in case of a short grace period 2548 * avoid a too-soon return to zero in case of a short grace period
2375 * (or preemption of this task). Also flag this task as doing 2549 * (or preemption of this task). Exclude CPU-hotplug operations
2376 * an rcu_barrier(). This will prevent anyone else from adopting 2550 * to ensure that no offline CPU has callbacks queued.
2377 * orphaned callbacks, which could cause otherwise failure if a
2378 * CPU went offline and quickly came back online. To see this,
2379 * consider the following sequence of events:
2380 *
2381 * 1. We cause CPU 0 to post an rcu_barrier_callback() callback.
2382 * 2. CPU 1 goes offline, orphaning its callbacks.
2383 * 3. CPU 0 adopts CPU 1's orphaned callbacks.
2384 * 4. CPU 1 comes back online.
2385 * 5. We cause CPU 1 to post an rcu_barrier_callback() callback.
2386 * 6. Both rcu_barrier_callback() callbacks are invoked, awakening
2387 * us -- but before CPU 1's orphaned callbacks are invoked!!!
2388 */ 2551 */
2389 init_completion(&rsp->barrier_completion); 2552 init_completion(&rsp->barrier_completion);
2390 atomic_set(&rsp->barrier_cpu_count, 1); 2553 atomic_set(&rsp->barrier_cpu_count, 1);
2391 raw_spin_lock_irqsave(&rsp->onofflock, flags); 2554 get_online_cpus();
2392 rsp->rcu_barrier_in_progress = current;
2393 raw_spin_unlock_irqrestore(&rsp->onofflock, flags);
2394 2555
2395 /* 2556 /*
2396 * Force every CPU with callbacks to register a new callback 2557 * Force each CPU with callbacks to register a new callback.
2397 * that will tell us when all the preceding callbacks have 2558 * When that callback is invoked, we will know that all of the
2398 * been invoked. If an offline CPU has callbacks, wait for 2559 * corresponding CPU's preceding callbacks have been invoked.
2399 * it to either come back online or to finish orphaning those
2400 * callbacks.
2401 */ 2560 */
2402 for_each_possible_cpu(cpu) { 2561 for_each_online_cpu(cpu) {
2403 preempt_disable();
2404 rdp = per_cpu_ptr(rsp->rda, cpu); 2562 rdp = per_cpu_ptr(rsp->rda, cpu);
2405 if (cpu_is_offline(cpu)) { 2563 if (ACCESS_ONCE(rdp->qlen)) {
2406 _rcu_barrier_trace(rsp, "Offline", cpu,
2407 rsp->n_barrier_done);
2408 preempt_enable();
2409 while (cpu_is_offline(cpu) && ACCESS_ONCE(rdp->qlen))
2410 schedule_timeout_interruptible(1);
2411 } else if (ACCESS_ONCE(rdp->qlen)) {
2412 _rcu_barrier_trace(rsp, "OnlineQ", cpu, 2564 _rcu_barrier_trace(rsp, "OnlineQ", cpu,
2413 rsp->n_barrier_done); 2565 rsp->n_barrier_done);
2414 smp_call_function_single(cpu, rcu_barrier_func, rsp, 1); 2566 smp_call_function_single(cpu, rcu_barrier_func, rsp, 1);
2415 preempt_enable();
2416 } else { 2567 } else {
2417 _rcu_barrier_trace(rsp, "OnlineNQ", cpu, 2568 _rcu_barrier_trace(rsp, "OnlineNQ", cpu,
2418 rsp->n_barrier_done); 2569 rsp->n_barrier_done);
2419 preempt_enable();
2420 } 2570 }
2421 } 2571 }
2422 2572 put_online_cpus();
2423 /*
2424 * Now that all online CPUs have rcu_barrier_callback() callbacks
2425 * posted, we can adopt all of the orphaned callbacks and place
2426 * an rcu_barrier_callback() callback after them. When that is done,
2427 * we are guaranteed to have an rcu_barrier_callback() callback
2428 * following every callback that could possibly have been
2429 * registered before _rcu_barrier() was called.
2430 */
2431 raw_spin_lock_irqsave(&rsp->onofflock, flags);
2432 rcu_adopt_orphan_cbs(rsp);
2433 rsp->rcu_barrier_in_progress = NULL;
2434 raw_spin_unlock_irqrestore(&rsp->onofflock, flags);
2435 atomic_inc(&rsp->barrier_cpu_count);
2436 smp_mb__after_atomic_inc(); /* Ensure atomic_inc() before callback. */
2437 rd.rsp = rsp;
2438 rsp->call(&rd.barrier_head, rcu_barrier_callback);
2439 2573
2440 /* 2574 /*
2441 * Now that we have an rcu_barrier_callback() callback on each 2575 * Now that we have an rcu_barrier_callback() callback on each
@@ -2456,8 +2590,6 @@ static void _rcu_barrier(struct rcu_state *rsp)
2456 2590
2457 /* Other rcu_barrier() invocations can now safely proceed. */ 2591 /* Other rcu_barrier() invocations can now safely proceed. */
2458 mutex_unlock(&rsp->barrier_mutex); 2592 mutex_unlock(&rsp->barrier_mutex);
2459
2460 destroy_rcu_head_on_stack(&rd.barrier_head);
2461} 2593}
2462 2594
2463/** 2595/**
@@ -2497,6 +2629,9 @@ rcu_boot_init_percpu_data(int cpu, struct rcu_state *rsp)
2497 rdp->dynticks = &per_cpu(rcu_dynticks, cpu); 2629 rdp->dynticks = &per_cpu(rcu_dynticks, cpu);
2498 WARN_ON_ONCE(rdp->dynticks->dynticks_nesting != DYNTICK_TASK_EXIT_IDLE); 2630 WARN_ON_ONCE(rdp->dynticks->dynticks_nesting != DYNTICK_TASK_EXIT_IDLE);
2499 WARN_ON_ONCE(atomic_read(&rdp->dynticks->dynticks) != 1); 2631 WARN_ON_ONCE(atomic_read(&rdp->dynticks->dynticks) != 1);
2632#ifdef CONFIG_RCU_USER_QS
2633 WARN_ON_ONCE(rdp->dynticks->in_user);
2634#endif
2500 rdp->cpu = cpu; 2635 rdp->cpu = cpu;
2501 rdp->rsp = rsp; 2636 rdp->rsp = rsp;
2502 raw_spin_unlock_irqrestore(&rnp->lock, flags); 2637 raw_spin_unlock_irqrestore(&rnp->lock, flags);
@@ -2516,6 +2651,9 @@ rcu_init_percpu_data(int cpu, struct rcu_state *rsp, int preemptible)
2516 struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu); 2651 struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu);
2517 struct rcu_node *rnp = rcu_get_root(rsp); 2652 struct rcu_node *rnp = rcu_get_root(rsp);
2518 2653
2654 /* Exclude new grace periods. */
2655 mutex_lock(&rsp->onoff_mutex);
2656
2519 /* Set up local state, ensuring consistent view of global state. */ 2657 /* Set up local state, ensuring consistent view of global state. */
2520 raw_spin_lock_irqsave(&rnp->lock, flags); 2658 raw_spin_lock_irqsave(&rnp->lock, flags);
2521 rdp->beenonline = 1; /* We have now been online. */ 2659 rdp->beenonline = 1; /* We have now been online. */
@@ -2523,20 +2661,13 @@ rcu_init_percpu_data(int cpu, struct rcu_state *rsp, int preemptible)
2523 rdp->qlen_last_fqs_check = 0; 2661 rdp->qlen_last_fqs_check = 0;
2524 rdp->n_force_qs_snap = rsp->n_force_qs; 2662 rdp->n_force_qs_snap = rsp->n_force_qs;
2525 rdp->blimit = blimit; 2663 rdp->blimit = blimit;
2664 init_callback_list(rdp); /* Re-enable callbacks on this CPU. */
2526 rdp->dynticks->dynticks_nesting = DYNTICK_TASK_EXIT_IDLE; 2665 rdp->dynticks->dynticks_nesting = DYNTICK_TASK_EXIT_IDLE;
2527 atomic_set(&rdp->dynticks->dynticks, 2666 atomic_set(&rdp->dynticks->dynticks,
2528 (atomic_read(&rdp->dynticks->dynticks) & ~0x1) + 1); 2667 (atomic_read(&rdp->dynticks->dynticks) & ~0x1) + 1);
2529 rcu_prepare_for_idle_init(cpu); 2668 rcu_prepare_for_idle_init(cpu);
2530 raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ 2669 raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */
2531 2670
2532 /*
2533 * A new grace period might start here. If so, we won't be part
2534 * of it, but that is OK, as we are currently in a quiescent state.
2535 */
2536
2537 /* Exclude any attempts to start a new GP on large systems. */
2538 raw_spin_lock(&rsp->onofflock); /* irqs already disabled. */
2539
2540 /* Add CPU to rcu_node bitmasks. */ 2671 /* Add CPU to rcu_node bitmasks. */
2541 rnp = rdp->mynode; 2672 rnp = rdp->mynode;
2542 mask = rdp->grpmask; 2673 mask = rdp->grpmask;
@@ -2555,14 +2686,14 @@ rcu_init_percpu_data(int cpu, struct rcu_state *rsp, int preemptible)
2555 rdp->completed = rnp->completed; 2686 rdp->completed = rnp->completed;
2556 rdp->passed_quiesce = 0; 2687 rdp->passed_quiesce = 0;
2557 rdp->qs_pending = 0; 2688 rdp->qs_pending = 0;
2558 rdp->passed_quiesce_gpnum = rnp->gpnum - 1;
2559 trace_rcu_grace_period(rsp->name, rdp->gpnum, "cpuonl"); 2689 trace_rcu_grace_period(rsp->name, rdp->gpnum, "cpuonl");
2560 } 2690 }
2561 raw_spin_unlock(&rnp->lock); /* irqs already disabled. */ 2691 raw_spin_unlock(&rnp->lock); /* irqs already disabled. */
2562 rnp = rnp->parent; 2692 rnp = rnp->parent;
2563 } while (rnp != NULL && !(rnp->qsmaskinit & mask)); 2693 } while (rnp != NULL && !(rnp->qsmaskinit & mask));
2694 local_irq_restore(flags);
2564 2695
2565 raw_spin_unlock_irqrestore(&rsp->onofflock, flags); 2696 mutex_unlock(&rsp->onoff_mutex);
2566} 2697}
2567 2698
2568static void __cpuinit rcu_prepare_cpu(int cpu) 2699static void __cpuinit rcu_prepare_cpu(int cpu)
@@ -2594,12 +2725,10 @@ static int __cpuinit rcu_cpu_notify(struct notifier_block *self,
2594 break; 2725 break;
2595 case CPU_ONLINE: 2726 case CPU_ONLINE:
2596 case CPU_DOWN_FAILED: 2727 case CPU_DOWN_FAILED:
2597 rcu_node_kthread_setaffinity(rnp, -1); 2728 rcu_boost_kthread_setaffinity(rnp, -1);
2598 rcu_cpu_kthread_setrt(cpu, 1);
2599 break; 2729 break;
2600 case CPU_DOWN_PREPARE: 2730 case CPU_DOWN_PREPARE:
2601 rcu_node_kthread_setaffinity(rnp, cpu); 2731 rcu_boost_kthread_setaffinity(rnp, cpu);
2602 rcu_cpu_kthread_setrt(cpu, 0);
2603 break; 2732 break;
2604 case CPU_DYING: 2733 case CPU_DYING:
2605 case CPU_DYING_FROZEN: 2734 case CPU_DYING_FROZEN:
@@ -2627,6 +2756,28 @@ static int __cpuinit rcu_cpu_notify(struct notifier_block *self,
2627} 2756}
2628 2757
2629/* 2758/*
2759 * Spawn the kthread that handles this RCU flavor's grace periods.
2760 */
2761static int __init rcu_spawn_gp_kthread(void)
2762{
2763 unsigned long flags;
2764 struct rcu_node *rnp;
2765 struct rcu_state *rsp;
2766 struct task_struct *t;
2767
2768 for_each_rcu_flavor(rsp) {
2769 t = kthread_run(rcu_gp_kthread, rsp, rsp->name);
2770 BUG_ON(IS_ERR(t));
2771 rnp = rcu_get_root(rsp);
2772 raw_spin_lock_irqsave(&rnp->lock, flags);
2773 rsp->gp_kthread = t;
2774 raw_spin_unlock_irqrestore(&rnp->lock, flags);
2775 }
2776 return 0;
2777}
2778early_initcall(rcu_spawn_gp_kthread);
2779
2780/*
2630 * This function is invoked towards the end of the scheduler's initialization 2781 * This function is invoked towards the end of the scheduler's initialization
2631 * process. Before this is called, the idle task might contain 2782 * process. Before this is called, the idle task might contain
2632 * RCU read-side critical sections (during which time, this idle 2783 * RCU read-side critical sections (during which time, this idle
@@ -2661,7 +2812,7 @@ static void __init rcu_init_levelspread(struct rcu_state *rsp)
2661 int cprv; 2812 int cprv;
2662 int i; 2813 int i;
2663 2814
2664 cprv = NR_CPUS; 2815 cprv = nr_cpu_ids;
2665 for (i = rcu_num_lvls - 1; i >= 0; i--) { 2816 for (i = rcu_num_lvls - 1; i >= 0; i--) {
2666 ccur = rsp->levelcnt[i]; 2817 ccur = rsp->levelcnt[i];
2667 rsp->levelspread[i] = (cprv + ccur - 1) / ccur; 2818 rsp->levelspread[i] = (cprv + ccur - 1) / ccur;
@@ -2676,10 +2827,14 @@ static void __init rcu_init_levelspread(struct rcu_state *rsp)
2676static void __init rcu_init_one(struct rcu_state *rsp, 2827static void __init rcu_init_one(struct rcu_state *rsp,
2677 struct rcu_data __percpu *rda) 2828 struct rcu_data __percpu *rda)
2678{ 2829{
2679 static char *buf[] = { "rcu_node_level_0", 2830 static char *buf[] = { "rcu_node_0",
2680 "rcu_node_level_1", 2831 "rcu_node_1",
2681 "rcu_node_level_2", 2832 "rcu_node_2",
2682 "rcu_node_level_3" }; /* Match MAX_RCU_LVLS */ 2833 "rcu_node_3" }; /* Match MAX_RCU_LVLS */
2834 static char *fqs[] = { "rcu_node_fqs_0",
2835 "rcu_node_fqs_1",
2836 "rcu_node_fqs_2",
2837 "rcu_node_fqs_3" }; /* Match MAX_RCU_LVLS */
2683 int cpustride = 1; 2838 int cpustride = 1;
2684 int i; 2839 int i;
2685 int j; 2840 int j;
@@ -2704,7 +2859,11 @@ static void __init rcu_init_one(struct rcu_state *rsp,
2704 raw_spin_lock_init(&rnp->lock); 2859 raw_spin_lock_init(&rnp->lock);
2705 lockdep_set_class_and_name(&rnp->lock, 2860 lockdep_set_class_and_name(&rnp->lock,
2706 &rcu_node_class[i], buf[i]); 2861 &rcu_node_class[i], buf[i]);
2707 rnp->gpnum = 0; 2862 raw_spin_lock_init(&rnp->fqslock);
2863 lockdep_set_class_and_name(&rnp->fqslock,
2864 &rcu_fqs_class[i], fqs[i]);
2865 rnp->gpnum = rsp->gpnum;
2866 rnp->completed = rsp->completed;
2708 rnp->qsmask = 0; 2867 rnp->qsmask = 0;
2709 rnp->qsmaskinit = 0; 2868 rnp->qsmaskinit = 0;
2710 rnp->grplo = j * cpustride; 2869 rnp->grplo = j * cpustride;
@@ -2727,6 +2886,7 @@ static void __init rcu_init_one(struct rcu_state *rsp,
2727 } 2886 }
2728 2887
2729 rsp->rda = rda; 2888 rsp->rda = rda;
2889 init_waitqueue_head(&rsp->gp_wq);
2730 rnp = rsp->level[rcu_num_lvls - 1]; 2890 rnp = rsp->level[rcu_num_lvls - 1];
2731 for_each_possible_cpu(i) { 2891 for_each_possible_cpu(i) {
2732 while (i > rnp->grphi) 2892 while (i > rnp->grphi)
@@ -2750,7 +2910,8 @@ static void __init rcu_init_geometry(void)
2750 int rcu_capacity[MAX_RCU_LVLS + 1]; 2910 int rcu_capacity[MAX_RCU_LVLS + 1];
2751 2911
2752 /* If the compile-time values are accurate, just leave. */ 2912 /* If the compile-time values are accurate, just leave. */
2753 if (rcu_fanout_leaf == CONFIG_RCU_FANOUT_LEAF) 2913 if (rcu_fanout_leaf == CONFIG_RCU_FANOUT_LEAF &&
2914 nr_cpu_ids == NR_CPUS)
2754 return; 2915 return;
2755 2916
2756 /* 2917 /*
diff --git a/kernel/rcutree.h b/kernel/rcutree.h
index 4d29169f2124..a240f032848e 100644
--- a/kernel/rcutree.h
+++ b/kernel/rcutree.h
@@ -102,6 +102,10 @@ struct rcu_dynticks {
102 /* idle-period nonlazy_posted snapshot. */ 102 /* idle-period nonlazy_posted snapshot. */
103 int tick_nohz_enabled_snap; /* Previously seen value from sysfs. */ 103 int tick_nohz_enabled_snap; /* Previously seen value from sysfs. */
104#endif /* #ifdef CONFIG_RCU_FAST_NO_HZ */ 104#endif /* #ifdef CONFIG_RCU_FAST_NO_HZ */
105#ifdef CONFIG_RCU_USER_QS
106 bool ignore_user_qs; /* Treat userspace as extended QS or not */
107 bool in_user; /* Is the CPU in userland from RCU POV? */
108#endif
105}; 109};
106 110
107/* RCU's kthread states for tracing. */ 111/* RCU's kthread states for tracing. */
@@ -196,12 +200,7 @@ struct rcu_node {
196 /* Refused to boost: not sure why, though. */ 200 /* Refused to boost: not sure why, though. */
197 /* This can happen due to race conditions. */ 201 /* This can happen due to race conditions. */
198#endif /* #ifdef CONFIG_RCU_BOOST */ 202#endif /* #ifdef CONFIG_RCU_BOOST */
199 struct task_struct *node_kthread_task; 203 raw_spinlock_t fqslock ____cacheline_internodealigned_in_smp;
200 /* kthread that takes care of this rcu_node */
201 /* structure, for example, awakening the */
202 /* per-CPU kthreads as needed. */
203 unsigned int node_kthread_status;
204 /* State of node_kthread_task for tracing. */
205} ____cacheline_internodealigned_in_smp; 204} ____cacheline_internodealigned_in_smp;
206 205
207/* 206/*
@@ -245,8 +244,6 @@ struct rcu_data {
245 /* in order to detect GP end. */ 244 /* in order to detect GP end. */
246 unsigned long gpnum; /* Highest gp number that this CPU */ 245 unsigned long gpnum; /* Highest gp number that this CPU */
247 /* is aware of having started. */ 246 /* is aware of having started. */
248 unsigned long passed_quiesce_gpnum;
249 /* gpnum at time of quiescent state. */
250 bool passed_quiesce; /* User-mode/idle loop etc. */ 247 bool passed_quiesce; /* User-mode/idle loop etc. */
251 bool qs_pending; /* Core waits for quiesc state. */ 248 bool qs_pending; /* Core waits for quiesc state. */
252 bool beenonline; /* CPU online at least once. */ 249 bool beenonline; /* CPU online at least once. */
@@ -312,11 +309,13 @@ struct rcu_data {
312 unsigned long n_rp_cpu_needs_gp; 309 unsigned long n_rp_cpu_needs_gp;
313 unsigned long n_rp_gp_completed; 310 unsigned long n_rp_gp_completed;
314 unsigned long n_rp_gp_started; 311 unsigned long n_rp_gp_started;
315 unsigned long n_rp_need_fqs;
316 unsigned long n_rp_need_nothing; 312 unsigned long n_rp_need_nothing;
317 313
318 /* 6) _rcu_barrier() callback. */ 314 /* 6) _rcu_barrier() and OOM callbacks. */
319 struct rcu_head barrier_head; 315 struct rcu_head barrier_head;
316#ifdef CONFIG_RCU_FAST_NO_HZ
317 struct rcu_head oom_head;
318#endif /* #ifdef CONFIG_RCU_FAST_NO_HZ */
320 319
321 int cpu; 320 int cpu;
322 struct rcu_state *rsp; 321 struct rcu_state *rsp;
@@ -375,20 +374,17 @@ struct rcu_state {
375 374
376 u8 fqs_state ____cacheline_internodealigned_in_smp; 375 u8 fqs_state ____cacheline_internodealigned_in_smp;
377 /* Force QS state. */ 376 /* Force QS state. */
378 u8 fqs_active; /* force_quiescent_state() */
379 /* is running. */
380 u8 fqs_need_gp; /* A CPU was prevented from */
381 /* starting a new grace */
382 /* period because */
383 /* force_quiescent_state() */
384 /* was running. */
385 u8 boost; /* Subject to priority boost. */ 377 u8 boost; /* Subject to priority boost. */
386 unsigned long gpnum; /* Current gp number. */ 378 unsigned long gpnum; /* Current gp number. */
387 unsigned long completed; /* # of last completed gp. */ 379 unsigned long completed; /* # of last completed gp. */
380 struct task_struct *gp_kthread; /* Task for grace periods. */
381 wait_queue_head_t gp_wq; /* Where GP task waits. */
382 int gp_flags; /* Commands for GP task. */
388 383
389 /* End of fields guarded by root rcu_node's lock. */ 384 /* End of fields guarded by root rcu_node's lock. */
390 385
391 raw_spinlock_t onofflock; /* exclude on/offline and */ 386 raw_spinlock_t onofflock ____cacheline_internodealigned_in_smp;
387 /* exclude on/offline and */
392 /* starting new GP. */ 388 /* starting new GP. */
393 struct rcu_head *orphan_nxtlist; /* Orphaned callbacks that */ 389 struct rcu_head *orphan_nxtlist; /* Orphaned callbacks that */
394 /* need a grace period. */ 390 /* need a grace period. */
@@ -398,16 +394,17 @@ struct rcu_state {
398 struct rcu_head **orphan_donetail; /* Tail of above. */ 394 struct rcu_head **orphan_donetail; /* Tail of above. */
399 long qlen_lazy; /* Number of lazy callbacks. */ 395 long qlen_lazy; /* Number of lazy callbacks. */
400 long qlen; /* Total number of callbacks. */ 396 long qlen; /* Total number of callbacks. */
401 struct task_struct *rcu_barrier_in_progress; 397 /* End of fields guarded by onofflock. */
402 /* Task doing rcu_barrier(), */ 398
403 /* or NULL if no barrier. */ 399 struct mutex onoff_mutex; /* Coordinate hotplug & GPs. */
400
404 struct mutex barrier_mutex; /* Guards barrier fields. */ 401 struct mutex barrier_mutex; /* Guards barrier fields. */
405 atomic_t barrier_cpu_count; /* # CPUs waiting on. */ 402 atomic_t barrier_cpu_count; /* # CPUs waiting on. */
406 struct completion barrier_completion; /* Wake at barrier end. */ 403 struct completion barrier_completion; /* Wake at barrier end. */
407 unsigned long n_barrier_done; /* ++ at start and end of */ 404 unsigned long n_barrier_done; /* ++ at start and end of */
408 /* _rcu_barrier(). */ 405 /* _rcu_barrier(). */
409 raw_spinlock_t fqslock; /* Only one task forcing */ 406 /* End of fields guarded by barrier_mutex. */
410 /* quiescent states. */ 407
411 unsigned long jiffies_force_qs; /* Time at which to invoke */ 408 unsigned long jiffies_force_qs; /* Time at which to invoke */
412 /* force_quiescent_state(). */ 409 /* force_quiescent_state(). */
413 unsigned long n_force_qs; /* Number of calls to */ 410 unsigned long n_force_qs; /* Number of calls to */
@@ -426,6 +423,10 @@ struct rcu_state {
426 struct list_head flavors; /* List of RCU flavors. */ 423 struct list_head flavors; /* List of RCU flavors. */
427}; 424};
428 425
426/* Values for rcu_state structure's gp_flags field. */
427#define RCU_GP_FLAG_INIT 0x1 /* Need grace-period initialization. */
428#define RCU_GP_FLAG_FQS 0x2 /* Need grace-period quiescent-state forcing. */
429
429extern struct list_head rcu_struct_flavors; 430extern struct list_head rcu_struct_flavors;
430#define for_each_rcu_flavor(rsp) \ 431#define for_each_rcu_flavor(rsp) \
431 list_for_each_entry((rsp), &rcu_struct_flavors, flavors) 432 list_for_each_entry((rsp), &rcu_struct_flavors, flavors)
@@ -468,7 +469,6 @@ static int rcu_preempt_blocked_readers_cgp(struct rcu_node *rnp);
468#ifdef CONFIG_HOTPLUG_CPU 469#ifdef CONFIG_HOTPLUG_CPU
469static void rcu_report_unblock_qs_rnp(struct rcu_node *rnp, 470static void rcu_report_unblock_qs_rnp(struct rcu_node *rnp,
470 unsigned long flags); 471 unsigned long flags);
471static void rcu_stop_cpu_kthread(int cpu);
472#endif /* #ifdef CONFIG_HOTPLUG_CPU */ 472#endif /* #ifdef CONFIG_HOTPLUG_CPU */
473static void rcu_print_detail_task_stall(struct rcu_state *rsp); 473static void rcu_print_detail_task_stall(struct rcu_state *rsp);
474static int rcu_print_task_stall(struct rcu_node *rnp); 474static int rcu_print_task_stall(struct rcu_node *rnp);
@@ -491,15 +491,9 @@ static void invoke_rcu_callbacks_kthread(void);
491static bool rcu_is_callbacks_kthread(void); 491static bool rcu_is_callbacks_kthread(void);
492#ifdef CONFIG_RCU_BOOST 492#ifdef CONFIG_RCU_BOOST
493static void rcu_preempt_do_callbacks(void); 493static void rcu_preempt_do_callbacks(void);
494static void rcu_boost_kthread_setaffinity(struct rcu_node *rnp,
495 cpumask_var_t cm);
496static int __cpuinit rcu_spawn_one_boost_kthread(struct rcu_state *rsp, 494static int __cpuinit rcu_spawn_one_boost_kthread(struct rcu_state *rsp,
497 struct rcu_node *rnp, 495 struct rcu_node *rnp);
498 int rnp_index);
499static void invoke_rcu_node_kthread(struct rcu_node *rnp);
500static void rcu_yield(void (*f)(unsigned long), unsigned long arg);
501#endif /* #ifdef CONFIG_RCU_BOOST */ 496#endif /* #ifdef CONFIG_RCU_BOOST */
502static void rcu_cpu_kthread_setrt(int cpu, int to_rt);
503static void __cpuinit rcu_prepare_kthreads(int cpu); 497static void __cpuinit rcu_prepare_kthreads(int cpu);
504static void rcu_prepare_for_idle_init(int cpu); 498static void rcu_prepare_for_idle_init(int cpu);
505static void rcu_cleanup_after_idle(int cpu); 499static void rcu_cleanup_after_idle(int cpu);
diff --git a/kernel/rcutree_plugin.h b/kernel/rcutree_plugin.h
index 7f3244c0df01..f92115488187 100644
--- a/kernel/rcutree_plugin.h
+++ b/kernel/rcutree_plugin.h
@@ -25,6 +25,8 @@
25 */ 25 */
26 26
27#include <linux/delay.h> 27#include <linux/delay.h>
28#include <linux/oom.h>
29#include <linux/smpboot.h>
28 30
29#define RCU_KTHREAD_PRIO 1 31#define RCU_KTHREAD_PRIO 1
30 32
@@ -118,7 +120,7 @@ EXPORT_SYMBOL_GPL(rcu_batches_completed);
118 */ 120 */
119void rcu_force_quiescent_state(void) 121void rcu_force_quiescent_state(void)
120{ 122{
121 force_quiescent_state(&rcu_preempt_state, 0); 123 force_quiescent_state(&rcu_preempt_state);
122} 124}
123EXPORT_SYMBOL_GPL(rcu_force_quiescent_state); 125EXPORT_SYMBOL_GPL(rcu_force_quiescent_state);
124 126
@@ -136,8 +138,6 @@ static void rcu_preempt_qs(int cpu)
136{ 138{
137 struct rcu_data *rdp = &per_cpu(rcu_preempt_data, cpu); 139 struct rcu_data *rdp = &per_cpu(rcu_preempt_data, cpu);
138 140
139 rdp->passed_quiesce_gpnum = rdp->gpnum;
140 barrier();
141 if (rdp->passed_quiesce == 0) 141 if (rdp->passed_quiesce == 0)
142 trace_rcu_grace_period("rcu_preempt", rdp->gpnum, "cpuqs"); 142 trace_rcu_grace_period("rcu_preempt", rdp->gpnum, "cpuqs");
143 rdp->passed_quiesce = 1; 143 rdp->passed_quiesce = 1;
@@ -422,9 +422,11 @@ static void rcu_print_detail_task_stall_rnp(struct rcu_node *rnp)
422 unsigned long flags; 422 unsigned long flags;
423 struct task_struct *t; 423 struct task_struct *t;
424 424
425 if (!rcu_preempt_blocked_readers_cgp(rnp))
426 return;
427 raw_spin_lock_irqsave(&rnp->lock, flags); 425 raw_spin_lock_irqsave(&rnp->lock, flags);
426 if (!rcu_preempt_blocked_readers_cgp(rnp)) {
427 raw_spin_unlock_irqrestore(&rnp->lock, flags);
428 return;
429 }
428 t = list_entry(rnp->gp_tasks, 430 t = list_entry(rnp->gp_tasks,
429 struct task_struct, rcu_node_entry); 431 struct task_struct, rcu_node_entry);
430 list_for_each_entry_continue(t, &rnp->blkd_tasks, rcu_node_entry) 432 list_for_each_entry_continue(t, &rnp->blkd_tasks, rcu_node_entry)
@@ -584,17 +586,23 @@ static int rcu_preempt_offline_tasks(struct rcu_state *rsp,
584 raw_spin_unlock(&rnp_root->lock); /* irqs still disabled */ 586 raw_spin_unlock(&rnp_root->lock); /* irqs still disabled */
585 } 587 }
586 588
589 rnp->gp_tasks = NULL;
590 rnp->exp_tasks = NULL;
587#ifdef CONFIG_RCU_BOOST 591#ifdef CONFIG_RCU_BOOST
588 /* In case root is being boosted and leaf is not. */ 592 rnp->boost_tasks = NULL;
593 /*
594 * In case root is being boosted and leaf was not. Make sure
595 * that we boost the tasks blocking the current grace period
596 * in this case.
597 */
589 raw_spin_lock(&rnp_root->lock); /* irqs already disabled */ 598 raw_spin_lock(&rnp_root->lock); /* irqs already disabled */
590 if (rnp_root->boost_tasks != NULL && 599 if (rnp_root->boost_tasks != NULL &&
591 rnp_root->boost_tasks != rnp_root->gp_tasks) 600 rnp_root->boost_tasks != rnp_root->gp_tasks &&
601 rnp_root->boost_tasks != rnp_root->exp_tasks)
592 rnp_root->boost_tasks = rnp_root->gp_tasks; 602 rnp_root->boost_tasks = rnp_root->gp_tasks;
593 raw_spin_unlock(&rnp_root->lock); /* irqs still disabled */ 603 raw_spin_unlock(&rnp_root->lock); /* irqs still disabled */
594#endif /* #ifdef CONFIG_RCU_BOOST */ 604#endif /* #ifdef CONFIG_RCU_BOOST */
595 605
596 rnp->gp_tasks = NULL;
597 rnp->exp_tasks = NULL;
598 return retval; 606 return retval;
599} 607}
600 608
@@ -676,7 +684,7 @@ void synchronize_rcu(void)
676EXPORT_SYMBOL_GPL(synchronize_rcu); 684EXPORT_SYMBOL_GPL(synchronize_rcu);
677 685
678static DECLARE_WAIT_QUEUE_HEAD(sync_rcu_preempt_exp_wq); 686static DECLARE_WAIT_QUEUE_HEAD(sync_rcu_preempt_exp_wq);
679static long sync_rcu_preempt_exp_count; 687static unsigned long sync_rcu_preempt_exp_count;
680static DEFINE_MUTEX(sync_rcu_preempt_exp_mutex); 688static DEFINE_MUTEX(sync_rcu_preempt_exp_mutex);
681 689
682/* 690/*
@@ -791,7 +799,7 @@ void synchronize_rcu_expedited(void)
791 unsigned long flags; 799 unsigned long flags;
792 struct rcu_node *rnp; 800 struct rcu_node *rnp;
793 struct rcu_state *rsp = &rcu_preempt_state; 801 struct rcu_state *rsp = &rcu_preempt_state;
794 long snap; 802 unsigned long snap;
795 int trycount = 0; 803 int trycount = 0;
796 804
797 smp_mb(); /* Caller's modifications seen first by other CPUs. */ 805 smp_mb(); /* Caller's modifications seen first by other CPUs. */
@@ -799,33 +807,47 @@ void synchronize_rcu_expedited(void)
799 smp_mb(); /* Above access cannot bleed into critical section. */ 807 smp_mb(); /* Above access cannot bleed into critical section. */
800 808
801 /* 809 /*
810 * Block CPU-hotplug operations. This means that any CPU-hotplug
811 * operation that finds an rcu_node structure with tasks in the
812 * process of being boosted will know that all tasks blocking
813 * this expedited grace period will already be in the process of
814 * being boosted. This simplifies the process of moving tasks
815 * from leaf to root rcu_node structures.
816 */
817 get_online_cpus();
818
819 /*
802 * Acquire lock, falling back to synchronize_rcu() if too many 820 * Acquire lock, falling back to synchronize_rcu() if too many
803 * lock-acquisition failures. Of course, if someone does the 821 * lock-acquisition failures. Of course, if someone does the
804 * expedited grace period for us, just leave. 822 * expedited grace period for us, just leave.
805 */ 823 */
806 while (!mutex_trylock(&sync_rcu_preempt_exp_mutex)) { 824 while (!mutex_trylock(&sync_rcu_preempt_exp_mutex)) {
825 if (ULONG_CMP_LT(snap,
826 ACCESS_ONCE(sync_rcu_preempt_exp_count))) {
827 put_online_cpus();
828 goto mb_ret; /* Others did our work for us. */
829 }
807 if (trycount++ < 10) { 830 if (trycount++ < 10) {
808 udelay(trycount * num_online_cpus()); 831 udelay(trycount * num_online_cpus());
809 } else { 832 } else {
833 put_online_cpus();
810 synchronize_rcu(); 834 synchronize_rcu();
811 return; 835 return;
812 } 836 }
813 if ((ACCESS_ONCE(sync_rcu_preempt_exp_count) - snap) > 0)
814 goto mb_ret; /* Others did our work for us. */
815 } 837 }
816 if ((ACCESS_ONCE(sync_rcu_preempt_exp_count) - snap) > 0) 838 if (ULONG_CMP_LT(snap, ACCESS_ONCE(sync_rcu_preempt_exp_count))) {
839 put_online_cpus();
817 goto unlock_mb_ret; /* Others did our work for us. */ 840 goto unlock_mb_ret; /* Others did our work for us. */
841 }
818 842
819 /* force all RCU readers onto ->blkd_tasks lists. */ 843 /* force all RCU readers onto ->blkd_tasks lists. */
820 synchronize_sched_expedited(); 844 synchronize_sched_expedited();
821 845
822 raw_spin_lock_irqsave(&rsp->onofflock, flags);
823
824 /* Initialize ->expmask for all non-leaf rcu_node structures. */ 846 /* Initialize ->expmask for all non-leaf rcu_node structures. */
825 rcu_for_each_nonleaf_node_breadth_first(rsp, rnp) { 847 rcu_for_each_nonleaf_node_breadth_first(rsp, rnp) {
826 raw_spin_lock(&rnp->lock); /* irqs already disabled. */ 848 raw_spin_lock_irqsave(&rnp->lock, flags);
827 rnp->expmask = rnp->qsmaskinit; 849 rnp->expmask = rnp->qsmaskinit;
828 raw_spin_unlock(&rnp->lock); /* irqs remain disabled. */ 850 raw_spin_unlock_irqrestore(&rnp->lock, flags);
829 } 851 }
830 852
831 /* Snapshot current state of ->blkd_tasks lists. */ 853 /* Snapshot current state of ->blkd_tasks lists. */
@@ -834,7 +856,7 @@ void synchronize_rcu_expedited(void)
834 if (NUM_RCU_NODES > 1) 856 if (NUM_RCU_NODES > 1)
835 sync_rcu_preempt_exp_init(rsp, rcu_get_root(rsp)); 857 sync_rcu_preempt_exp_init(rsp, rcu_get_root(rsp));
836 858
837 raw_spin_unlock_irqrestore(&rsp->onofflock, flags); 859 put_online_cpus();
838 860
839 /* Wait for snapshotted ->blkd_tasks lists to drain. */ 861 /* Wait for snapshotted ->blkd_tasks lists to drain. */
840 rnp = rcu_get_root(rsp); 862 rnp = rcu_get_root(rsp);
@@ -1069,6 +1091,16 @@ static void rcu_initiate_boost_trace(struct rcu_node *rnp)
1069 1091
1070#endif /* #else #ifdef CONFIG_RCU_TRACE */ 1092#endif /* #else #ifdef CONFIG_RCU_TRACE */
1071 1093
1094static void rcu_wake_cond(struct task_struct *t, int status)
1095{
1096 /*
1097 * If the thread is yielding, only wake it when this
1098 * is invoked from idle
1099 */
1100 if (status != RCU_KTHREAD_YIELDING || is_idle_task(current))
1101 wake_up_process(t);
1102}
1103
1072/* 1104/*
1073 * Carry out RCU priority boosting on the task indicated by ->exp_tasks 1105 * Carry out RCU priority boosting on the task indicated by ->exp_tasks
1074 * or ->boost_tasks, advancing the pointer to the next task in the 1106 * or ->boost_tasks, advancing the pointer to the next task in the
@@ -1141,17 +1173,6 @@ static int rcu_boost(struct rcu_node *rnp)
1141} 1173}
1142 1174
1143/* 1175/*
1144 * Timer handler to initiate waking up of boost kthreads that
1145 * have yielded the CPU due to excessive numbers of tasks to
1146 * boost. We wake up the per-rcu_node kthread, which in turn
1147 * will wake up the booster kthread.
1148 */
1149static void rcu_boost_kthread_timer(unsigned long arg)
1150{
1151 invoke_rcu_node_kthread((struct rcu_node *)arg);
1152}
1153
1154/*
1155 * Priority-boosting kthread. One per leaf rcu_node and one for the 1176 * Priority-boosting kthread. One per leaf rcu_node and one for the
1156 * root rcu_node. 1177 * root rcu_node.
1157 */ 1178 */
@@ -1174,8 +1195,9 @@ static int rcu_boost_kthread(void *arg)
1174 else 1195 else
1175 spincnt = 0; 1196 spincnt = 0;
1176 if (spincnt > 10) { 1197 if (spincnt > 10) {
1198 rnp->boost_kthread_status = RCU_KTHREAD_YIELDING;
1177 trace_rcu_utilization("End boost kthread@rcu_yield"); 1199 trace_rcu_utilization("End boost kthread@rcu_yield");
1178 rcu_yield(rcu_boost_kthread_timer, (unsigned long)rnp); 1200 schedule_timeout_interruptible(2);
1179 trace_rcu_utilization("Start boost kthread@rcu_yield"); 1201 trace_rcu_utilization("Start boost kthread@rcu_yield");
1180 spincnt = 0; 1202 spincnt = 0;
1181 } 1203 }
@@ -1191,9 +1213,9 @@ static int rcu_boost_kthread(void *arg)
1191 * kthread to start boosting them. If there is an expedited grace 1213 * kthread to start boosting them. If there is an expedited grace
1192 * period in progress, it is always time to boost. 1214 * period in progress, it is always time to boost.
1193 * 1215 *
1194 * The caller must hold rnp->lock, which this function releases, 1216 * The caller must hold rnp->lock, which this function releases.
1195 * but irqs remain disabled. The ->boost_kthread_task is immortal, 1217 * The ->boost_kthread_task is immortal, so we don't need to worry
1196 * so we don't need to worry about it going away. 1218 * about it going away.
1197 */ 1219 */
1198static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags) 1220static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags)
1199{ 1221{
@@ -1213,8 +1235,8 @@ static void rcu_initiate_boost(struct rcu_node *rnp, unsigned long flags)
1213 rnp->boost_tasks = rnp->gp_tasks; 1235 rnp->boost_tasks = rnp->gp_tasks;
1214 raw_spin_unlock_irqrestore(&rnp->lock, flags); 1236 raw_spin_unlock_irqrestore(&rnp->lock, flags);
1215 t = rnp->boost_kthread_task; 1237 t = rnp->boost_kthread_task;
1216 if (t != NULL) 1238 if (t)
1217 wake_up_process(t); 1239 rcu_wake_cond(t, rnp->boost_kthread_status);
1218 } else { 1240 } else {
1219 rcu_initiate_boost_trace(rnp); 1241 rcu_initiate_boost_trace(rnp);
1220 raw_spin_unlock_irqrestore(&rnp->lock, flags); 1242 raw_spin_unlock_irqrestore(&rnp->lock, flags);
@@ -1231,8 +1253,10 @@ static void invoke_rcu_callbacks_kthread(void)
1231 local_irq_save(flags); 1253 local_irq_save(flags);
1232 __this_cpu_write(rcu_cpu_has_work, 1); 1254 __this_cpu_write(rcu_cpu_has_work, 1);
1233 if (__this_cpu_read(rcu_cpu_kthread_task) != NULL && 1255 if (__this_cpu_read(rcu_cpu_kthread_task) != NULL &&
1234 current != __this_cpu_read(rcu_cpu_kthread_task)) 1256 current != __this_cpu_read(rcu_cpu_kthread_task)) {
1235 wake_up_process(__this_cpu_read(rcu_cpu_kthread_task)); 1257 rcu_wake_cond(__this_cpu_read(rcu_cpu_kthread_task),
1258 __this_cpu_read(rcu_cpu_kthread_status));
1259 }
1236 local_irq_restore(flags); 1260 local_irq_restore(flags);
1237} 1261}
1238 1262
@@ -1245,21 +1269,6 @@ static bool rcu_is_callbacks_kthread(void)
1245 return __get_cpu_var(rcu_cpu_kthread_task) == current; 1269 return __get_cpu_var(rcu_cpu_kthread_task) == current;
1246} 1270}
1247 1271
1248/*
1249 * Set the affinity of the boost kthread. The CPU-hotplug locks are
1250 * held, so no one should be messing with the existence of the boost
1251 * kthread.
1252 */
1253static void rcu_boost_kthread_setaffinity(struct rcu_node *rnp,
1254 cpumask_var_t cm)
1255{
1256 struct task_struct *t;
1257
1258 t = rnp->boost_kthread_task;
1259 if (t != NULL)
1260 set_cpus_allowed_ptr(rnp->boost_kthread_task, cm);
1261}
1262
1263#define RCU_BOOST_DELAY_JIFFIES DIV_ROUND_UP(CONFIG_RCU_BOOST_DELAY * HZ, 1000) 1272#define RCU_BOOST_DELAY_JIFFIES DIV_ROUND_UP(CONFIG_RCU_BOOST_DELAY * HZ, 1000)
1264 1273
1265/* 1274/*
@@ -1276,15 +1285,19 @@ static void rcu_preempt_boost_start_gp(struct rcu_node *rnp)
1276 * Returns zero if all is well, a negated errno otherwise. 1285 * Returns zero if all is well, a negated errno otherwise.
1277 */ 1286 */
1278static int __cpuinit rcu_spawn_one_boost_kthread(struct rcu_state *rsp, 1287static int __cpuinit rcu_spawn_one_boost_kthread(struct rcu_state *rsp,
1279 struct rcu_node *rnp, 1288 struct rcu_node *rnp)
1280 int rnp_index)
1281{ 1289{
1290 int rnp_index = rnp - &rsp->node[0];
1282 unsigned long flags; 1291 unsigned long flags;
1283 struct sched_param sp; 1292 struct sched_param sp;
1284 struct task_struct *t; 1293 struct task_struct *t;
1285 1294
1286 if (&rcu_preempt_state != rsp) 1295 if (&rcu_preempt_state != rsp)
1287 return 0; 1296 return 0;
1297
1298 if (!rcu_scheduler_fully_active || rnp->qsmaskinit == 0)
1299 return 0;
1300
1288 rsp->boost = 1; 1301 rsp->boost = 1;
1289 if (rnp->boost_kthread_task != NULL) 1302 if (rnp->boost_kthread_task != NULL)
1290 return 0; 1303 return 0;
@@ -1301,25 +1314,6 @@ static int __cpuinit rcu_spawn_one_boost_kthread(struct rcu_state *rsp,
1301 return 0; 1314 return 0;
1302} 1315}
1303 1316
1304#ifdef CONFIG_HOTPLUG_CPU
1305
1306/*
1307 * Stop the RCU's per-CPU kthread when its CPU goes offline,.
1308 */
1309static void rcu_stop_cpu_kthread(int cpu)
1310{
1311 struct task_struct *t;
1312
1313 /* Stop the CPU's kthread. */
1314 t = per_cpu(rcu_cpu_kthread_task, cpu);
1315 if (t != NULL) {
1316 per_cpu(rcu_cpu_kthread_task, cpu) = NULL;
1317 kthread_stop(t);
1318 }
1319}
1320
1321#endif /* #ifdef CONFIG_HOTPLUG_CPU */
1322
1323static void rcu_kthread_do_work(void) 1317static void rcu_kthread_do_work(void)
1324{ 1318{
1325 rcu_do_batch(&rcu_sched_state, &__get_cpu_var(rcu_sched_data)); 1319 rcu_do_batch(&rcu_sched_state, &__get_cpu_var(rcu_sched_data));
@@ -1327,112 +1321,22 @@ static void rcu_kthread_do_work(void)
1327 rcu_preempt_do_callbacks(); 1321 rcu_preempt_do_callbacks();
1328} 1322}
1329 1323
1330/* 1324static void rcu_cpu_kthread_setup(unsigned int cpu)
1331 * Wake up the specified per-rcu_node-structure kthread.
1332 * Because the per-rcu_node kthreads are immortal, we don't need
1333 * to do anything to keep them alive.
1334 */
1335static void invoke_rcu_node_kthread(struct rcu_node *rnp)
1336{
1337 struct task_struct *t;
1338
1339 t = rnp->node_kthread_task;
1340 if (t != NULL)
1341 wake_up_process(t);
1342}
1343
1344/*
1345 * Set the specified CPU's kthread to run RT or not, as specified by
1346 * the to_rt argument. The CPU-hotplug locks are held, so the task
1347 * is not going away.
1348 */
1349static void rcu_cpu_kthread_setrt(int cpu, int to_rt)
1350{ 1325{
1351 int policy;
1352 struct sched_param sp; 1326 struct sched_param sp;
1353 struct task_struct *t;
1354 1327
1355 t = per_cpu(rcu_cpu_kthread_task, cpu); 1328 sp.sched_priority = RCU_KTHREAD_PRIO;
1356 if (t == NULL) 1329 sched_setscheduler_nocheck(current, SCHED_FIFO, &sp);
1357 return;
1358 if (to_rt) {
1359 policy = SCHED_FIFO;
1360 sp.sched_priority = RCU_KTHREAD_PRIO;
1361 } else {
1362 policy = SCHED_NORMAL;
1363 sp.sched_priority = 0;
1364 }
1365 sched_setscheduler_nocheck(t, policy, &sp);
1366} 1330}
1367 1331
1368/* 1332static void rcu_cpu_kthread_park(unsigned int cpu)
1369 * Timer handler to initiate the waking up of per-CPU kthreads that
1370 * have yielded the CPU due to excess numbers of RCU callbacks.
1371 * We wake up the per-rcu_node kthread, which in turn will wake up
1372 * the booster kthread.
1373 */
1374static void rcu_cpu_kthread_timer(unsigned long arg)
1375{ 1333{
1376 struct rcu_data *rdp = per_cpu_ptr(rcu_state->rda, arg); 1334 per_cpu(rcu_cpu_kthread_status, cpu) = RCU_KTHREAD_OFFCPU;
1377 struct rcu_node *rnp = rdp->mynode;
1378
1379 atomic_or(rdp->grpmask, &rnp->wakemask);
1380 invoke_rcu_node_kthread(rnp);
1381} 1335}
1382 1336
1383/* 1337static int rcu_cpu_kthread_should_run(unsigned int cpu)
1384 * Drop to non-real-time priority and yield, but only after posting a
1385 * timer that will cause us to regain our real-time priority if we
1386 * remain preempted. Either way, we restore our real-time priority
1387 * before returning.
1388 */
1389static void rcu_yield(void (*f)(unsigned long), unsigned long arg)
1390{ 1338{
1391 struct sched_param sp; 1339 return __get_cpu_var(rcu_cpu_has_work);
1392 struct timer_list yield_timer;
1393 int prio = current->rt_priority;
1394
1395 setup_timer_on_stack(&yield_timer, f, arg);
1396 mod_timer(&yield_timer, jiffies + 2);
1397 sp.sched_priority = 0;
1398 sched_setscheduler_nocheck(current, SCHED_NORMAL, &sp);
1399 set_user_nice(current, 19);
1400 schedule();
1401 set_user_nice(current, 0);
1402 sp.sched_priority = prio;
1403 sched_setscheduler_nocheck(current, SCHED_FIFO, &sp);
1404 del_timer(&yield_timer);
1405}
1406
1407/*
1408 * Handle cases where the rcu_cpu_kthread() ends up on the wrong CPU.
1409 * This can happen while the corresponding CPU is either coming online
1410 * or going offline. We cannot wait until the CPU is fully online
1411 * before starting the kthread, because the various notifier functions
1412 * can wait for RCU grace periods. So we park rcu_cpu_kthread() until
1413 * the corresponding CPU is online.
1414 *
1415 * Return 1 if the kthread needs to stop, 0 otherwise.
1416 *
1417 * Caller must disable bh. This function can momentarily enable it.
1418 */
1419static int rcu_cpu_kthread_should_stop(int cpu)
1420{
1421 while (cpu_is_offline(cpu) ||
1422 !cpumask_equal(&current->cpus_allowed, cpumask_of(cpu)) ||
1423 smp_processor_id() != cpu) {
1424 if (kthread_should_stop())
1425 return 1;
1426 per_cpu(rcu_cpu_kthread_status, cpu) = RCU_KTHREAD_OFFCPU;
1427 per_cpu(rcu_cpu_kthread_cpu, cpu) = raw_smp_processor_id();
1428 local_bh_enable();
1429 schedule_timeout_uninterruptible(1);
1430 if (!cpumask_equal(&current->cpus_allowed, cpumask_of(cpu)))
1431 set_cpus_allowed_ptr(current, cpumask_of(cpu));
1432 local_bh_disable();
1433 }
1434 per_cpu(rcu_cpu_kthread_cpu, cpu) = cpu;
1435 return 0;
1436} 1340}
1437 1341
1438/* 1342/*
@@ -1440,138 +1344,35 @@ static int rcu_cpu_kthread_should_stop(int cpu)
1440 * RCU softirq used in flavors and configurations of RCU that do not 1344 * RCU softirq used in flavors and configurations of RCU that do not
1441 * support RCU priority boosting. 1345 * support RCU priority boosting.
1442 */ 1346 */
1443static int rcu_cpu_kthread(void *arg) 1347static void rcu_cpu_kthread(unsigned int cpu)
1444{ 1348{
1445 int cpu = (int)(long)arg; 1349 unsigned int *statusp = &__get_cpu_var(rcu_cpu_kthread_status);
1446 unsigned long flags; 1350 char work, *workp = &__get_cpu_var(rcu_cpu_has_work);
1447 int spincnt = 0; 1351 int spincnt;
1448 unsigned int *statusp = &per_cpu(rcu_cpu_kthread_status, cpu);
1449 char work;
1450 char *workp = &per_cpu(rcu_cpu_has_work, cpu);
1451 1352
1452 trace_rcu_utilization("Start CPU kthread@init"); 1353 for (spincnt = 0; spincnt < 10; spincnt++) {
1453 for (;;) {
1454 *statusp = RCU_KTHREAD_WAITING;
1455 trace_rcu_utilization("End CPU kthread@rcu_wait");
1456 rcu_wait(*workp != 0 || kthread_should_stop());
1457 trace_rcu_utilization("Start CPU kthread@rcu_wait"); 1354 trace_rcu_utilization("Start CPU kthread@rcu_wait");
1458 local_bh_disable(); 1355 local_bh_disable();
1459 if (rcu_cpu_kthread_should_stop(cpu)) {
1460 local_bh_enable();
1461 break;
1462 }
1463 *statusp = RCU_KTHREAD_RUNNING; 1356 *statusp = RCU_KTHREAD_RUNNING;
1464 per_cpu(rcu_cpu_kthread_loops, cpu)++; 1357 this_cpu_inc(rcu_cpu_kthread_loops);
1465 local_irq_save(flags); 1358 local_irq_disable();
1466 work = *workp; 1359 work = *workp;
1467 *workp = 0; 1360 *workp = 0;
1468 local_irq_restore(flags); 1361 local_irq_enable();
1469 if (work) 1362 if (work)
1470 rcu_kthread_do_work(); 1363 rcu_kthread_do_work();
1471 local_bh_enable(); 1364 local_bh_enable();
1472 if (*workp != 0) 1365 if (*workp == 0) {
1473 spincnt++; 1366 trace_rcu_utilization("End CPU kthread@rcu_wait");
1474 else 1367 *statusp = RCU_KTHREAD_WAITING;
1475 spincnt = 0; 1368 return;
1476 if (spincnt > 10) {
1477 *statusp = RCU_KTHREAD_YIELDING;
1478 trace_rcu_utilization("End CPU kthread@rcu_yield");
1479 rcu_yield(rcu_cpu_kthread_timer, (unsigned long)cpu);
1480 trace_rcu_utilization("Start CPU kthread@rcu_yield");
1481 spincnt = 0;
1482 }
1483 }
1484 *statusp = RCU_KTHREAD_STOPPED;
1485 trace_rcu_utilization("End CPU kthread@term");
1486 return 0;
1487}
1488
1489/*
1490 * Spawn a per-CPU kthread, setting up affinity and priority.
1491 * Because the CPU hotplug lock is held, no other CPU will be attempting
1492 * to manipulate rcu_cpu_kthread_task. There might be another CPU
1493 * attempting to access it during boot, but the locking in kthread_bind()
1494 * will enforce sufficient ordering.
1495 *
1496 * Please note that we cannot simply refuse to wake up the per-CPU
1497 * kthread because kthreads are created in TASK_UNINTERRUPTIBLE state,
1498 * which can result in softlockup complaints if the task ends up being
1499 * idle for more than a couple of minutes.
1500 *
1501 * However, please note also that we cannot bind the per-CPU kthread to its
1502 * CPU until that CPU is fully online. We also cannot wait until the
1503 * CPU is fully online before we create its per-CPU kthread, as this would
1504 * deadlock the system when CPU notifiers tried waiting for grace
1505 * periods. So we bind the per-CPU kthread to its CPU only if the CPU
1506 * is online. If its CPU is not yet fully online, then the code in
1507 * rcu_cpu_kthread() will wait until it is fully online, and then do
1508 * the binding.
1509 */
1510static int __cpuinit rcu_spawn_one_cpu_kthread(int cpu)
1511{
1512 struct sched_param sp;
1513 struct task_struct *t;
1514
1515 if (!rcu_scheduler_fully_active ||
1516 per_cpu(rcu_cpu_kthread_task, cpu) != NULL)
1517 return 0;
1518 t = kthread_create_on_node(rcu_cpu_kthread,
1519 (void *)(long)cpu,
1520 cpu_to_node(cpu),
1521 "rcuc/%d", cpu);
1522 if (IS_ERR(t))
1523 return PTR_ERR(t);
1524 if (cpu_online(cpu))
1525 kthread_bind(t, cpu);
1526 per_cpu(rcu_cpu_kthread_cpu, cpu) = cpu;
1527 WARN_ON_ONCE(per_cpu(rcu_cpu_kthread_task, cpu) != NULL);
1528 sp.sched_priority = RCU_KTHREAD_PRIO;
1529 sched_setscheduler_nocheck(t, SCHED_FIFO, &sp);
1530 per_cpu(rcu_cpu_kthread_task, cpu) = t;
1531 wake_up_process(t); /* Get to TASK_INTERRUPTIBLE quickly. */
1532 return 0;
1533}
1534
1535/*
1536 * Per-rcu_node kthread, which is in charge of waking up the per-CPU
1537 * kthreads when needed. We ignore requests to wake up kthreads
1538 * for offline CPUs, which is OK because force_quiescent_state()
1539 * takes care of this case.
1540 */
1541static int rcu_node_kthread(void *arg)
1542{
1543 int cpu;
1544 unsigned long flags;
1545 unsigned long mask;
1546 struct rcu_node *rnp = (struct rcu_node *)arg;
1547 struct sched_param sp;
1548 struct task_struct *t;
1549
1550 for (;;) {
1551 rnp->node_kthread_status = RCU_KTHREAD_WAITING;
1552 rcu_wait(atomic_read(&rnp->wakemask) != 0);
1553 rnp->node_kthread_status = RCU_KTHREAD_RUNNING;
1554 raw_spin_lock_irqsave(&rnp->lock, flags);
1555 mask = atomic_xchg(&rnp->wakemask, 0);
1556 rcu_initiate_boost(rnp, flags); /* releases rnp->lock. */
1557 for (cpu = rnp->grplo; cpu <= rnp->grphi; cpu++, mask >>= 1) {
1558 if ((mask & 0x1) == 0)
1559 continue;
1560 preempt_disable();
1561 t = per_cpu(rcu_cpu_kthread_task, cpu);
1562 if (!cpu_online(cpu) || t == NULL) {
1563 preempt_enable();
1564 continue;
1565 }
1566 per_cpu(rcu_cpu_has_work, cpu) = 1;
1567 sp.sched_priority = RCU_KTHREAD_PRIO;
1568 sched_setscheduler_nocheck(t, SCHED_FIFO, &sp);
1569 preempt_enable();
1570 } 1369 }
1571 } 1370 }
1572 /* NOTREACHED */ 1371 *statusp = RCU_KTHREAD_YIELDING;
1573 rnp->node_kthread_status = RCU_KTHREAD_STOPPED; 1372 trace_rcu_utilization("Start CPU kthread@rcu_yield");
1574 return 0; 1373 schedule_timeout_interruptible(2);
1374 trace_rcu_utilization("End CPU kthread@rcu_yield");
1375 *statusp = RCU_KTHREAD_WAITING;
1575} 1376}
1576 1377
1577/* 1378/*
@@ -1583,17 +1384,17 @@ static int rcu_node_kthread(void *arg)
1583 * no outgoing CPU. If there are no CPUs left in the affinity set, 1384 * no outgoing CPU. If there are no CPUs left in the affinity set,
1584 * this function allows the kthread to execute on any CPU. 1385 * this function allows the kthread to execute on any CPU.
1585 */ 1386 */
1586static void rcu_node_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu) 1387static void rcu_boost_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu)
1587{ 1388{
1389 struct task_struct *t = rnp->boost_kthread_task;
1390 unsigned long mask = rnp->qsmaskinit;
1588 cpumask_var_t cm; 1391 cpumask_var_t cm;
1589 int cpu; 1392 int cpu;
1590 unsigned long mask = rnp->qsmaskinit;
1591 1393
1592 if (rnp->node_kthread_task == NULL) 1394 if (!t)
1593 return; 1395 return;
1594 if (!alloc_cpumask_var(&cm, GFP_KERNEL)) 1396 if (!zalloc_cpumask_var(&cm, GFP_KERNEL))
1595 return; 1397 return;
1596 cpumask_clear(cm);
1597 for (cpu = rnp->grplo; cpu <= rnp->grphi; cpu++, mask >>= 1) 1398 for (cpu = rnp->grplo; cpu <= rnp->grphi; cpu++, mask >>= 1)
1598 if ((mask & 0x1) && cpu != outgoingcpu) 1399 if ((mask & 0x1) && cpu != outgoingcpu)
1599 cpumask_set_cpu(cpu, cm); 1400 cpumask_set_cpu(cpu, cm);
@@ -1603,62 +1404,36 @@ static void rcu_node_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu)
1603 cpumask_clear_cpu(cpu, cm); 1404 cpumask_clear_cpu(cpu, cm);
1604 WARN_ON_ONCE(cpumask_weight(cm) == 0); 1405 WARN_ON_ONCE(cpumask_weight(cm) == 0);
1605 } 1406 }
1606 set_cpus_allowed_ptr(rnp->node_kthread_task, cm); 1407 set_cpus_allowed_ptr(t, cm);
1607 rcu_boost_kthread_setaffinity(rnp, cm);
1608 free_cpumask_var(cm); 1408 free_cpumask_var(cm);
1609} 1409}
1610 1410
1611/* 1411static struct smp_hotplug_thread rcu_cpu_thread_spec = {
1612 * Spawn a per-rcu_node kthread, setting priority and affinity. 1412 .store = &rcu_cpu_kthread_task,
1613 * Called during boot before online/offline can happen, or, if 1413 .thread_should_run = rcu_cpu_kthread_should_run,
1614 * during runtime, with the main CPU-hotplug locks held. So only 1414 .thread_fn = rcu_cpu_kthread,
1615 * one of these can be executing at a time. 1415 .thread_comm = "rcuc/%u",
1616 */ 1416 .setup = rcu_cpu_kthread_setup,
1617static int __cpuinit rcu_spawn_one_node_kthread(struct rcu_state *rsp, 1417 .park = rcu_cpu_kthread_park,
1618 struct rcu_node *rnp) 1418};
1619{
1620 unsigned long flags;
1621 int rnp_index = rnp - &rsp->node[0];
1622 struct sched_param sp;
1623 struct task_struct *t;
1624
1625 if (!rcu_scheduler_fully_active ||
1626 rnp->qsmaskinit == 0)
1627 return 0;
1628 if (rnp->node_kthread_task == NULL) {
1629 t = kthread_create(rcu_node_kthread, (void *)rnp,
1630 "rcun/%d", rnp_index);
1631 if (IS_ERR(t))
1632 return PTR_ERR(t);
1633 raw_spin_lock_irqsave(&rnp->lock, flags);
1634 rnp->node_kthread_task = t;
1635 raw_spin_unlock_irqrestore(&rnp->lock, flags);
1636 sp.sched_priority = 99;
1637 sched_setscheduler_nocheck(t, SCHED_FIFO, &sp);
1638 wake_up_process(t); /* get to TASK_INTERRUPTIBLE quickly. */
1639 }
1640 return rcu_spawn_one_boost_kthread(rsp, rnp, rnp_index);
1641}
1642 1419
1643/* 1420/*
1644 * Spawn all kthreads -- called as soon as the scheduler is running. 1421 * Spawn all kthreads -- called as soon as the scheduler is running.
1645 */ 1422 */
1646static int __init rcu_spawn_kthreads(void) 1423static int __init rcu_spawn_kthreads(void)
1647{ 1424{
1648 int cpu;
1649 struct rcu_node *rnp; 1425 struct rcu_node *rnp;
1426 int cpu;
1650 1427
1651 rcu_scheduler_fully_active = 1; 1428 rcu_scheduler_fully_active = 1;
1652 for_each_possible_cpu(cpu) { 1429 for_each_possible_cpu(cpu)
1653 per_cpu(rcu_cpu_has_work, cpu) = 0; 1430 per_cpu(rcu_cpu_has_work, cpu) = 0;
1654 if (cpu_online(cpu)) 1431 BUG_ON(smpboot_register_percpu_thread(&rcu_cpu_thread_spec));
1655 (void)rcu_spawn_one_cpu_kthread(cpu);
1656 }
1657 rnp = rcu_get_root(rcu_state); 1432 rnp = rcu_get_root(rcu_state);
1658 (void)rcu_spawn_one_node_kthread(rcu_state, rnp); 1433 (void)rcu_spawn_one_boost_kthread(rcu_state, rnp);
1659 if (NUM_RCU_NODES > 1) { 1434 if (NUM_RCU_NODES > 1) {
1660 rcu_for_each_leaf_node(rcu_state, rnp) 1435 rcu_for_each_leaf_node(rcu_state, rnp)
1661 (void)rcu_spawn_one_node_kthread(rcu_state, rnp); 1436 (void)rcu_spawn_one_boost_kthread(rcu_state, rnp);
1662 } 1437 }
1663 return 0; 1438 return 0;
1664} 1439}
@@ -1670,11 +1445,8 @@ static void __cpuinit rcu_prepare_kthreads(int cpu)
1670 struct rcu_node *rnp = rdp->mynode; 1445 struct rcu_node *rnp = rdp->mynode;
1671 1446
1672 /* Fire up the incoming CPU's kthread and leaf rcu_node kthread. */ 1447 /* Fire up the incoming CPU's kthread and leaf rcu_node kthread. */
1673 if (rcu_scheduler_fully_active) { 1448 if (rcu_scheduler_fully_active)
1674 (void)rcu_spawn_one_cpu_kthread(cpu); 1449 (void)rcu_spawn_one_boost_kthread(rcu_state, rnp);
1675 if (rnp->node_kthread_task == NULL)
1676 (void)rcu_spawn_one_node_kthread(rcu_state, rnp);
1677 }
1678} 1450}
1679 1451
1680#else /* #ifdef CONFIG_RCU_BOOST */ 1452#else /* #ifdef CONFIG_RCU_BOOST */
@@ -1698,19 +1470,7 @@ static void rcu_preempt_boost_start_gp(struct rcu_node *rnp)
1698{ 1470{
1699} 1471}
1700 1472
1701#ifdef CONFIG_HOTPLUG_CPU 1473static void rcu_boost_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu)
1702
1703static void rcu_stop_cpu_kthread(int cpu)
1704{
1705}
1706
1707#endif /* #ifdef CONFIG_HOTPLUG_CPU */
1708
1709static void rcu_node_kthread_setaffinity(struct rcu_node *rnp, int outgoingcpu)
1710{
1711}
1712
1713static void rcu_cpu_kthread_setrt(int cpu, int to_rt)
1714{ 1474{
1715} 1475}
1716 1476
@@ -1997,6 +1757,26 @@ static void rcu_prepare_for_idle(int cpu)
1997 if (!tne) 1757 if (!tne)
1998 return; 1758 return;
1999 1759
1760 /* Adaptive-tick mode, where usermode execution is idle to RCU. */
1761 if (!is_idle_task(current)) {
1762 rdtp->dyntick_holdoff = jiffies - 1;
1763 if (rcu_cpu_has_nonlazy_callbacks(cpu)) {
1764 trace_rcu_prep_idle("User dyntick with callbacks");
1765 rdtp->idle_gp_timer_expires =
1766 round_up(jiffies + RCU_IDLE_GP_DELAY,
1767 RCU_IDLE_GP_DELAY);
1768 } else if (rcu_cpu_has_callbacks(cpu)) {
1769 rdtp->idle_gp_timer_expires =
1770 round_jiffies(jiffies + RCU_IDLE_LAZY_GP_DELAY);
1771 trace_rcu_prep_idle("User dyntick with lazy callbacks");
1772 } else {
1773 return;
1774 }
1775 tp = &rdtp->idle_gp_timer;
1776 mod_timer_pinned(tp, rdtp->idle_gp_timer_expires);
1777 return;
1778 }
1779
2000 /* 1780 /*
2001 * If this is an idle re-entry, for example, due to use of 1781 * If this is an idle re-entry, for example, due to use of
2002 * RCU_NONIDLE() or the new idle-loop tracing API within the idle 1782 * RCU_NONIDLE() or the new idle-loop tracing API within the idle
@@ -2075,16 +1855,16 @@ static void rcu_prepare_for_idle(int cpu)
2075#ifdef CONFIG_TREE_PREEMPT_RCU 1855#ifdef CONFIG_TREE_PREEMPT_RCU
2076 if (per_cpu(rcu_preempt_data, cpu).nxtlist) { 1856 if (per_cpu(rcu_preempt_data, cpu).nxtlist) {
2077 rcu_preempt_qs(cpu); 1857 rcu_preempt_qs(cpu);
2078 force_quiescent_state(&rcu_preempt_state, 0); 1858 force_quiescent_state(&rcu_preempt_state);
2079 } 1859 }
2080#endif /* #ifdef CONFIG_TREE_PREEMPT_RCU */ 1860#endif /* #ifdef CONFIG_TREE_PREEMPT_RCU */
2081 if (per_cpu(rcu_sched_data, cpu).nxtlist) { 1861 if (per_cpu(rcu_sched_data, cpu).nxtlist) {
2082 rcu_sched_qs(cpu); 1862 rcu_sched_qs(cpu);
2083 force_quiescent_state(&rcu_sched_state, 0); 1863 force_quiescent_state(&rcu_sched_state);
2084 } 1864 }
2085 if (per_cpu(rcu_bh_data, cpu).nxtlist) { 1865 if (per_cpu(rcu_bh_data, cpu).nxtlist) {
2086 rcu_bh_qs(cpu); 1866 rcu_bh_qs(cpu);
2087 force_quiescent_state(&rcu_bh_state, 0); 1867 force_quiescent_state(&rcu_bh_state);
2088 } 1868 }
2089 1869
2090 /* 1870 /*
@@ -2112,6 +1892,88 @@ static void rcu_idle_count_callbacks_posted(void)
2112 __this_cpu_add(rcu_dynticks.nonlazy_posted, 1); 1892 __this_cpu_add(rcu_dynticks.nonlazy_posted, 1);
2113} 1893}
2114 1894
1895/*
1896 * Data for flushing lazy RCU callbacks at OOM time.
1897 */
1898static atomic_t oom_callback_count;
1899static DECLARE_WAIT_QUEUE_HEAD(oom_callback_wq);
1900
1901/*
1902 * RCU OOM callback -- decrement the outstanding count and deliver the
1903 * wake-up if we are the last one.
1904 */
1905static void rcu_oom_callback(struct rcu_head *rhp)
1906{
1907 if (atomic_dec_and_test(&oom_callback_count))
1908 wake_up(&oom_callback_wq);
1909}
1910
1911/*
1912 * Post an rcu_oom_notify callback on the current CPU if it has at
1913 * least one lazy callback. This will unnecessarily post callbacks
1914 * to CPUs that already have a non-lazy callback at the end of their
1915 * callback list, but this is an infrequent operation, so accept some
1916 * extra overhead to keep things simple.
1917 */
1918static void rcu_oom_notify_cpu(void *unused)
1919{
1920 struct rcu_state *rsp;
1921 struct rcu_data *rdp;
1922
1923 for_each_rcu_flavor(rsp) {
1924 rdp = __this_cpu_ptr(rsp->rda);
1925 if (rdp->qlen_lazy != 0) {
1926 atomic_inc(&oom_callback_count);
1927 rsp->call(&rdp->oom_head, rcu_oom_callback);
1928 }
1929 }
1930}
1931
1932/*
1933 * If low on memory, ensure that each CPU has a non-lazy callback.
1934 * This will wake up CPUs that have only lazy callbacks, in turn
1935 * ensuring that they free up the corresponding memory in a timely manner.
1936 * Because an uncertain amount of memory will be freed in some uncertain
1937 * timeframe, we do not claim to have freed anything.
1938 */
1939static int rcu_oom_notify(struct notifier_block *self,
1940 unsigned long notused, void *nfreed)
1941{
1942 int cpu;
1943
1944 /* Wait for callbacks from earlier instance to complete. */
1945 wait_event(oom_callback_wq, atomic_read(&oom_callback_count) == 0);
1946
1947 /*
1948 * Prevent premature wakeup: ensure that all increments happen
1949 * before there is a chance of the counter reaching zero.
1950 */
1951 atomic_set(&oom_callback_count, 1);
1952
1953 get_online_cpus();
1954 for_each_online_cpu(cpu) {
1955 smp_call_function_single(cpu, rcu_oom_notify_cpu, NULL, 1);
1956 cond_resched();
1957 }
1958 put_online_cpus();
1959
1960 /* Unconditionally decrement: no need to wake ourselves up. */
1961 atomic_dec(&oom_callback_count);
1962
1963 return NOTIFY_OK;
1964}
1965
1966static struct notifier_block rcu_oom_nb = {
1967 .notifier_call = rcu_oom_notify
1968};
1969
1970static int __init rcu_register_oom_notifier(void)
1971{
1972 register_oom_notifier(&rcu_oom_nb);
1973 return 0;
1974}
1975early_initcall(rcu_register_oom_notifier);
1976
2115#endif /* #else #if !defined(CONFIG_RCU_FAST_NO_HZ) */ 1977#endif /* #else #if !defined(CONFIG_RCU_FAST_NO_HZ) */
2116 1978
2117#ifdef CONFIG_RCU_CPU_STALL_INFO 1979#ifdef CONFIG_RCU_CPU_STALL_INFO
@@ -2122,11 +1984,15 @@ static void print_cpu_stall_fast_no_hz(char *cp, int cpu)
2122{ 1984{
2123 struct rcu_dynticks *rdtp = &per_cpu(rcu_dynticks, cpu); 1985 struct rcu_dynticks *rdtp = &per_cpu(rcu_dynticks, cpu);
2124 struct timer_list *tltp = &rdtp->idle_gp_timer; 1986 struct timer_list *tltp = &rdtp->idle_gp_timer;
1987 char c;
2125 1988
2126 sprintf(cp, "drain=%d %c timer=%lu", 1989 c = rdtp->dyntick_holdoff == jiffies ? 'H' : '.';
2127 rdtp->dyntick_drain, 1990 if (timer_pending(tltp))
2128 rdtp->dyntick_holdoff == jiffies ? 'H' : '.', 1991 sprintf(cp, "drain=%d %c timer=%lu",
2129 timer_pending(tltp) ? tltp->expires - jiffies : -1); 1992 rdtp->dyntick_drain, c, tltp->expires - jiffies);
1993 else
1994 sprintf(cp, "drain=%d %c timer not pending",
1995 rdtp->dyntick_drain, c);
2130} 1996}
2131 1997
2132#else /* #ifdef CONFIG_RCU_FAST_NO_HZ */ 1998#else /* #ifdef CONFIG_RCU_FAST_NO_HZ */
@@ -2194,11 +2060,10 @@ static void zero_cpu_stall_ticks(struct rcu_data *rdp)
2194/* Increment ->ticks_this_gp for all flavors of RCU. */ 2060/* Increment ->ticks_this_gp for all flavors of RCU. */
2195static void increment_cpu_stall_ticks(void) 2061static void increment_cpu_stall_ticks(void)
2196{ 2062{
2197 __get_cpu_var(rcu_sched_data).ticks_this_gp++; 2063 struct rcu_state *rsp;
2198 __get_cpu_var(rcu_bh_data).ticks_this_gp++; 2064
2199#ifdef CONFIG_TREE_PREEMPT_RCU 2065 for_each_rcu_flavor(rsp)
2200 __get_cpu_var(rcu_preempt_data).ticks_this_gp++; 2066 __this_cpu_ptr(rsp->rda)->ticks_this_gp++;
2201#endif /* #ifdef CONFIG_TREE_PREEMPT_RCU */
2202} 2067}
2203 2068
2204#else /* #ifdef CONFIG_RCU_CPU_STALL_INFO */ 2069#else /* #ifdef CONFIG_RCU_CPU_STALL_INFO */
diff --git a/kernel/rcutree_trace.c b/kernel/rcutree_trace.c
index abffb486e94e..693513bc50e6 100644
--- a/kernel/rcutree_trace.c
+++ b/kernel/rcutree_trace.c
@@ -51,8 +51,8 @@ static int show_rcubarrier(struct seq_file *m, void *unused)
51 struct rcu_state *rsp; 51 struct rcu_state *rsp;
52 52
53 for_each_rcu_flavor(rsp) 53 for_each_rcu_flavor(rsp)
54 seq_printf(m, "%s: %c bcc: %d nbd: %lu\n", 54 seq_printf(m, "%s: bcc: %d nbd: %lu\n",
55 rsp->name, rsp->rcu_barrier_in_progress ? 'B' : '.', 55 rsp->name,
56 atomic_read(&rsp->barrier_cpu_count), 56 atomic_read(&rsp->barrier_cpu_count),
57 rsp->n_barrier_done); 57 rsp->n_barrier_done);
58 return 0; 58 return 0;
@@ -86,12 +86,11 @@ static void print_one_rcu_data(struct seq_file *m, struct rcu_data *rdp)
86{ 86{
87 if (!rdp->beenonline) 87 if (!rdp->beenonline)
88 return; 88 return;
89 seq_printf(m, "%3d%cc=%lu g=%lu pq=%d pgp=%lu qp=%d", 89 seq_printf(m, "%3d%cc=%lu g=%lu pq=%d qp=%d",
90 rdp->cpu, 90 rdp->cpu,
91 cpu_is_offline(rdp->cpu) ? '!' : ' ', 91 cpu_is_offline(rdp->cpu) ? '!' : ' ',
92 rdp->completed, rdp->gpnum, 92 rdp->completed, rdp->gpnum,
93 rdp->passed_quiesce, rdp->passed_quiesce_gpnum, 93 rdp->passed_quiesce, rdp->qs_pending);
94 rdp->qs_pending);
95 seq_printf(m, " dt=%d/%llx/%d df=%lu", 94 seq_printf(m, " dt=%d/%llx/%d df=%lu",
96 atomic_read(&rdp->dynticks->dynticks), 95 atomic_read(&rdp->dynticks->dynticks),
97 rdp->dynticks->dynticks_nesting, 96 rdp->dynticks->dynticks_nesting,
@@ -108,11 +107,10 @@ static void print_one_rcu_data(struct seq_file *m, struct rcu_data *rdp)
108 rdp->nxttail[RCU_WAIT_TAIL]], 107 rdp->nxttail[RCU_WAIT_TAIL]],
109 ".D"[&rdp->nxtlist != rdp->nxttail[RCU_DONE_TAIL]]); 108 ".D"[&rdp->nxtlist != rdp->nxttail[RCU_DONE_TAIL]]);
110#ifdef CONFIG_RCU_BOOST 109#ifdef CONFIG_RCU_BOOST
111 seq_printf(m, " kt=%d/%c/%d ktl=%x", 110 seq_printf(m, " kt=%d/%c ktl=%x",
112 per_cpu(rcu_cpu_has_work, rdp->cpu), 111 per_cpu(rcu_cpu_has_work, rdp->cpu),
113 convert_kthread_status(per_cpu(rcu_cpu_kthread_status, 112 convert_kthread_status(per_cpu(rcu_cpu_kthread_status,
114 rdp->cpu)), 113 rdp->cpu)),
115 per_cpu(rcu_cpu_kthread_cpu, rdp->cpu),
116 per_cpu(rcu_cpu_kthread_loops, rdp->cpu) & 0xffff); 114 per_cpu(rcu_cpu_kthread_loops, rdp->cpu) & 0xffff);
117#endif /* #ifdef CONFIG_RCU_BOOST */ 115#endif /* #ifdef CONFIG_RCU_BOOST */
118 seq_printf(m, " b=%ld", rdp->blimit); 116 seq_printf(m, " b=%ld", rdp->blimit);
@@ -150,12 +148,11 @@ static void print_one_rcu_data_csv(struct seq_file *m, struct rcu_data *rdp)
150{ 148{
151 if (!rdp->beenonline) 149 if (!rdp->beenonline)
152 return; 150 return;
153 seq_printf(m, "%d,%s,%lu,%lu,%d,%lu,%d", 151 seq_printf(m, "%d,%s,%lu,%lu,%d,%d",
154 rdp->cpu, 152 rdp->cpu,
155 cpu_is_offline(rdp->cpu) ? "\"N\"" : "\"Y\"", 153 cpu_is_offline(rdp->cpu) ? "\"N\"" : "\"Y\"",
156 rdp->completed, rdp->gpnum, 154 rdp->completed, rdp->gpnum,
157 rdp->passed_quiesce, rdp->passed_quiesce_gpnum, 155 rdp->passed_quiesce, rdp->qs_pending);
158 rdp->qs_pending);
159 seq_printf(m, ",%d,%llx,%d,%lu", 156 seq_printf(m, ",%d,%llx,%d,%lu",
160 atomic_read(&rdp->dynticks->dynticks), 157 atomic_read(&rdp->dynticks->dynticks),
161 rdp->dynticks->dynticks_nesting, 158 rdp->dynticks->dynticks_nesting,
@@ -186,7 +183,7 @@ static int show_rcudata_csv(struct seq_file *m, void *unused)
186 int cpu; 183 int cpu;
187 struct rcu_state *rsp; 184 struct rcu_state *rsp;
188 185
189 seq_puts(m, "\"CPU\",\"Online?\",\"c\",\"g\",\"pq\",\"pgp\",\"pq\","); 186 seq_puts(m, "\"CPU\",\"Online?\",\"c\",\"g\",\"pq\",\"pq\",");
190 seq_puts(m, "\"dt\",\"dt nesting\",\"dt NMI nesting\",\"df\","); 187 seq_puts(m, "\"dt\",\"dt nesting\",\"dt NMI nesting\",\"df\",");
191 seq_puts(m, "\"of\",\"qll\",\"ql\",\"qs\""); 188 seq_puts(m, "\"of\",\"qll\",\"ql\",\"qs\"");
192#ifdef CONFIG_RCU_BOOST 189#ifdef CONFIG_RCU_BOOST
@@ -386,10 +383,9 @@ static void print_one_rcu_pending(struct seq_file *m, struct rcu_data *rdp)
386 rdp->n_rp_report_qs, 383 rdp->n_rp_report_qs,
387 rdp->n_rp_cb_ready, 384 rdp->n_rp_cb_ready,
388 rdp->n_rp_cpu_needs_gp); 385 rdp->n_rp_cpu_needs_gp);
389 seq_printf(m, "gpc=%ld gps=%ld nf=%ld nn=%ld\n", 386 seq_printf(m, "gpc=%ld gps=%ld nn=%ld\n",
390 rdp->n_rp_gp_completed, 387 rdp->n_rp_gp_completed,
391 rdp->n_rp_gp_started, 388 rdp->n_rp_gp_started,
392 rdp->n_rp_need_fqs,
393 rdp->n_rp_need_nothing); 389 rdp->n_rp_need_nothing);
394} 390}
395 391
diff --git a/kernel/resource.c b/kernel/resource.c
index 34d45886ee84..73f35d4b30b9 100644
--- a/kernel/resource.c
+++ b/kernel/resource.c
@@ -763,6 +763,7 @@ static void __init __reserve_region_with_split(struct resource *root,
763 struct resource *parent = root; 763 struct resource *parent = root;
764 struct resource *conflict; 764 struct resource *conflict;
765 struct resource *res = kzalloc(sizeof(*res), GFP_ATOMIC); 765 struct resource *res = kzalloc(sizeof(*res), GFP_ATOMIC);
766 struct resource *next_res = NULL;
766 767
767 if (!res) 768 if (!res)
768 return; 769 return;
@@ -772,21 +773,46 @@ static void __init __reserve_region_with_split(struct resource *root,
772 res->end = end; 773 res->end = end;
773 res->flags = IORESOURCE_BUSY; 774 res->flags = IORESOURCE_BUSY;
774 775
775 conflict = __request_resource(parent, res); 776 while (1) {
776 if (!conflict)
777 return;
778 777
779 /* failed, split and try again */ 778 conflict = __request_resource(parent, res);
780 kfree(res); 779 if (!conflict) {
780 if (!next_res)
781 break;
782 res = next_res;
783 next_res = NULL;
784 continue;
785 }
781 786
782 /* conflict covered whole area */ 787 /* conflict covered whole area */
783 if (conflict->start <= start && conflict->end >= end) 788 if (conflict->start <= res->start &&
784 return; 789 conflict->end >= res->end) {
790 kfree(res);
791 WARN_ON(next_res);
792 break;
793 }
794
795 /* failed, split and try again */
796 if (conflict->start > res->start) {
797 end = res->end;
798 res->end = conflict->start - 1;
799 if (conflict->end < end) {
800 next_res = kzalloc(sizeof(*next_res),
801 GFP_ATOMIC);
802 if (!next_res) {
803 kfree(res);
804 break;
805 }
806 next_res->name = name;
807 next_res->start = conflict->end + 1;
808 next_res->end = end;
809 next_res->flags = IORESOURCE_BUSY;
810 }
811 } else {
812 res->start = conflict->end + 1;
813 }
814 }
785 815
786 if (conflict->start > start)
787 __reserve_region_with_split(root, start, conflict->start-1, name);
788 if (conflict->end < end)
789 __reserve_region_with_split(root, conflict->end+1, end, name);
790} 816}
791 817
792void __init reserve_region_with_split(struct resource *root, 818void __init reserve_region_with_split(struct resource *root,
diff --git a/kernel/sched/Makefile b/kernel/sched/Makefile
index 173ea52f3af0..f06d249e103b 100644
--- a/kernel/sched/Makefile
+++ b/kernel/sched/Makefile
@@ -11,7 +11,7 @@ ifneq ($(CONFIG_SCHED_OMIT_FRAME_POINTER),y)
11CFLAGS_core.o := $(PROFILING) -fno-omit-frame-pointer 11CFLAGS_core.o := $(PROFILING) -fno-omit-frame-pointer
12endif 12endif
13 13
14obj-y += core.o clock.o idle_task.o fair.o rt.o stop_task.o 14obj-y += core.o clock.o cputime.o idle_task.o fair.o rt.o stop_task.o
15obj-$(CONFIG_SMP) += cpupri.o 15obj-$(CONFIG_SMP) += cpupri.o
16obj-$(CONFIG_SCHED_AUTOGROUP) += auto_group.o 16obj-$(CONFIG_SCHED_AUTOGROUP) += auto_group.o
17obj-$(CONFIG_SCHEDSTATS) += stats.o 17obj-$(CONFIG_SCHEDSTATS) += stats.o
diff --git a/kernel/sched/core.c b/kernel/sched/core.c
index 649c9f876cb1..2d8927fda712 100644
--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -505,7 +505,7 @@ static inline void init_hrtick(void)
505#ifdef CONFIG_SMP 505#ifdef CONFIG_SMP
506 506
507#ifndef tsk_is_polling 507#ifndef tsk_is_polling
508#define tsk_is_polling(t) test_tsk_thread_flag(t, TIF_POLLING_NRFLAG) 508#define tsk_is_polling(t) 0
509#endif 509#endif
510 510
511void resched_task(struct task_struct *p) 511void resched_task(struct task_struct *p)
@@ -740,126 +740,6 @@ void deactivate_task(struct rq *rq, struct task_struct *p, int flags)
740 dequeue_task(rq, p, flags); 740 dequeue_task(rq, p, flags);
741} 741}
742 742
743#ifdef CONFIG_IRQ_TIME_ACCOUNTING
744
745/*
746 * There are no locks covering percpu hardirq/softirq time.
747 * They are only modified in account_system_vtime, on corresponding CPU
748 * with interrupts disabled. So, writes are safe.
749 * They are read and saved off onto struct rq in update_rq_clock().
750 * This may result in other CPU reading this CPU's irq time and can
751 * race with irq/account_system_vtime on this CPU. We would either get old
752 * or new value with a side effect of accounting a slice of irq time to wrong
753 * task when irq is in progress while we read rq->clock. That is a worthy
754 * compromise in place of having locks on each irq in account_system_time.
755 */
756static DEFINE_PER_CPU(u64, cpu_hardirq_time);
757static DEFINE_PER_CPU(u64, cpu_softirq_time);
758
759static DEFINE_PER_CPU(u64, irq_start_time);
760static int sched_clock_irqtime;
761
762void enable_sched_clock_irqtime(void)
763{
764 sched_clock_irqtime = 1;
765}
766
767void disable_sched_clock_irqtime(void)
768{
769 sched_clock_irqtime = 0;
770}
771
772#ifndef CONFIG_64BIT
773static DEFINE_PER_CPU(seqcount_t, irq_time_seq);
774
775static inline void irq_time_write_begin(void)
776{
777 __this_cpu_inc(irq_time_seq.sequence);
778 smp_wmb();
779}
780
781static inline void irq_time_write_end(void)
782{
783 smp_wmb();
784 __this_cpu_inc(irq_time_seq.sequence);
785}
786
787static inline u64 irq_time_read(int cpu)
788{
789 u64 irq_time;
790 unsigned seq;
791
792 do {
793 seq = read_seqcount_begin(&per_cpu(irq_time_seq, cpu));
794 irq_time = per_cpu(cpu_softirq_time, cpu) +
795 per_cpu(cpu_hardirq_time, cpu);
796 } while (read_seqcount_retry(&per_cpu(irq_time_seq, cpu), seq));
797
798 return irq_time;
799}
800#else /* CONFIG_64BIT */
801static inline void irq_time_write_begin(void)
802{
803}
804
805static inline void irq_time_write_end(void)
806{
807}
808
809static inline u64 irq_time_read(int cpu)
810{
811 return per_cpu(cpu_softirq_time, cpu) + per_cpu(cpu_hardirq_time, cpu);
812}
813#endif /* CONFIG_64BIT */
814
815/*
816 * Called before incrementing preempt_count on {soft,}irq_enter
817 * and before decrementing preempt_count on {soft,}irq_exit.
818 */
819void account_system_vtime(struct task_struct *curr)
820{
821 unsigned long flags;
822 s64 delta;
823 int cpu;
824
825 if (!sched_clock_irqtime)
826 return;
827
828 local_irq_save(flags);
829
830 cpu = smp_processor_id();
831 delta = sched_clock_cpu(cpu) - __this_cpu_read(irq_start_time);
832 __this_cpu_add(irq_start_time, delta);
833
834 irq_time_write_begin();
835 /*
836 * We do not account for softirq time from ksoftirqd here.
837 * We want to continue accounting softirq time to ksoftirqd thread
838 * in that case, so as not to confuse scheduler with a special task
839 * that do not consume any time, but still wants to run.
840 */
841 if (hardirq_count())
842 __this_cpu_add(cpu_hardirq_time, delta);
843 else if (in_serving_softirq() && curr != this_cpu_ksoftirqd())
844 __this_cpu_add(cpu_softirq_time, delta);
845
846 irq_time_write_end();
847 local_irq_restore(flags);
848}
849EXPORT_SYMBOL_GPL(account_system_vtime);
850
851#endif /* CONFIG_IRQ_TIME_ACCOUNTING */
852
853#ifdef CONFIG_PARAVIRT
854static inline u64 steal_ticks(u64 steal)
855{
856 if (unlikely(steal > NSEC_PER_SEC))
857 return div_u64(steal, TICK_NSEC);
858
859 return __iter_div_u64_rem(steal, TICK_NSEC, &steal);
860}
861#endif
862
863static void update_rq_clock_task(struct rq *rq, s64 delta) 743static void update_rq_clock_task(struct rq *rq, s64 delta)
864{ 744{
865/* 745/*
@@ -920,43 +800,6 @@ static void update_rq_clock_task(struct rq *rq, s64 delta)
920#endif 800#endif
921} 801}
922 802
923#ifdef CONFIG_IRQ_TIME_ACCOUNTING
924static int irqtime_account_hi_update(void)
925{
926 u64 *cpustat = kcpustat_this_cpu->cpustat;
927 unsigned long flags;
928 u64 latest_ns;
929 int ret = 0;
930
931 local_irq_save(flags);
932 latest_ns = this_cpu_read(cpu_hardirq_time);
933 if (nsecs_to_cputime64(latest_ns) > cpustat[CPUTIME_IRQ])
934 ret = 1;
935 local_irq_restore(flags);
936 return ret;
937}
938
939static int irqtime_account_si_update(void)
940{
941 u64 *cpustat = kcpustat_this_cpu->cpustat;
942 unsigned long flags;
943 u64 latest_ns;
944 int ret = 0;
945
946 local_irq_save(flags);
947 latest_ns = this_cpu_read(cpu_softirq_time);
948 if (nsecs_to_cputime64(latest_ns) > cpustat[CPUTIME_SOFTIRQ])
949 ret = 1;
950 local_irq_restore(flags);
951 return ret;
952}
953
954#else /* CONFIG_IRQ_TIME_ACCOUNTING */
955
956#define sched_clock_irqtime (0)
957
958#endif
959
960void sched_set_stop_task(int cpu, struct task_struct *stop) 803void sched_set_stop_task(int cpu, struct task_struct *stop)
961{ 804{
962 struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 }; 805 struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 };
@@ -1518,25 +1361,6 @@ static void ttwu_queue_remote(struct task_struct *p, int cpu)
1518 smp_send_reschedule(cpu); 1361 smp_send_reschedule(cpu);
1519} 1362}
1520 1363
1521#ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1522static int ttwu_activate_remote(struct task_struct *p, int wake_flags)
1523{
1524 struct rq *rq;
1525 int ret = 0;
1526
1527 rq = __task_rq_lock(p);
1528 if (p->on_cpu) {
1529 ttwu_activate(rq, p, ENQUEUE_WAKEUP);
1530 ttwu_do_wakeup(rq, p, wake_flags);
1531 ret = 1;
1532 }
1533 __task_rq_unlock(rq);
1534
1535 return ret;
1536
1537}
1538#endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */
1539
1540bool cpus_share_cache(int this_cpu, int that_cpu) 1364bool cpus_share_cache(int this_cpu, int that_cpu)
1541{ 1365{
1542 return per_cpu(sd_llc_id, this_cpu) == per_cpu(sd_llc_id, that_cpu); 1366 return per_cpu(sd_llc_id, this_cpu) == per_cpu(sd_llc_id, that_cpu);
@@ -1597,21 +1421,8 @@ try_to_wake_up(struct task_struct *p, unsigned int state, int wake_flags)
1597 * If the owning (remote) cpu is still in the middle of schedule() with 1421 * If the owning (remote) cpu is still in the middle of schedule() with
1598 * this task as prev, wait until its done referencing the task. 1422 * this task as prev, wait until its done referencing the task.
1599 */ 1423 */
1600 while (p->on_cpu) { 1424 while (p->on_cpu)
1601#ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1602 /*
1603 * In case the architecture enables interrupts in
1604 * context_switch(), we cannot busy wait, since that
1605 * would lead to deadlocks when an interrupt hits and
1606 * tries to wake up @prev. So bail and do a complete
1607 * remote wakeup.
1608 */
1609 if (ttwu_activate_remote(p, wake_flags))
1610 goto stat;
1611#else
1612 cpu_relax(); 1425 cpu_relax();
1613#endif
1614 }
1615 /* 1426 /*
1616 * Pairs with the smp_wmb() in finish_lock_switch(). 1427 * Pairs with the smp_wmb() in finish_lock_switch().
1617 */ 1428 */
@@ -1953,14 +1764,9 @@ static void finish_task_switch(struct rq *rq, struct task_struct *prev)
1953 * Manfred Spraul <manfred@colorfullife.com> 1764 * Manfred Spraul <manfred@colorfullife.com>
1954 */ 1765 */
1955 prev_state = prev->state; 1766 prev_state = prev->state;
1767 vtime_task_switch(prev);
1956 finish_arch_switch(prev); 1768 finish_arch_switch(prev);
1957#ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1958 local_irq_disable();
1959#endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */
1960 perf_event_task_sched_in(prev, current); 1769 perf_event_task_sched_in(prev, current);
1961#ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1962 local_irq_enable();
1963#endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */
1964 finish_lock_switch(rq, prev); 1770 finish_lock_switch(rq, prev);
1965 finish_arch_post_lock_switch(); 1771 finish_arch_post_lock_switch();
1966 1772
@@ -2081,6 +1887,7 @@ context_switch(struct rq *rq, struct task_struct *prev,
2081#endif 1887#endif
2082 1888
2083 /* Here we just switch the register state and the stack. */ 1889 /* Here we just switch the register state and the stack. */
1890 rcu_switch(prev, next);
2084 switch_to(prev, next, prev); 1891 switch_to(prev, next, prev);
2085 1892
2086 barrier(); 1893 barrier();
@@ -2809,404 +2616,6 @@ unsigned long long task_sched_runtime(struct task_struct *p)
2809 return ns; 2616 return ns;
2810} 2617}
2811 2618
2812#ifdef CONFIG_CGROUP_CPUACCT
2813struct cgroup_subsys cpuacct_subsys;
2814struct cpuacct root_cpuacct;
2815#endif
2816
2817static inline void task_group_account_field(struct task_struct *p, int index,
2818 u64 tmp)
2819{
2820#ifdef CONFIG_CGROUP_CPUACCT
2821 struct kernel_cpustat *kcpustat;
2822 struct cpuacct *ca;
2823#endif
2824 /*
2825 * Since all updates are sure to touch the root cgroup, we
2826 * get ourselves ahead and touch it first. If the root cgroup
2827 * is the only cgroup, then nothing else should be necessary.
2828 *
2829 */
2830 __get_cpu_var(kernel_cpustat).cpustat[index] += tmp;
2831
2832#ifdef CONFIG_CGROUP_CPUACCT
2833 if (unlikely(!cpuacct_subsys.active))
2834 return;
2835
2836 rcu_read_lock();
2837 ca = task_ca(p);
2838 while (ca && (ca != &root_cpuacct)) {
2839 kcpustat = this_cpu_ptr(ca->cpustat);
2840 kcpustat->cpustat[index] += tmp;
2841 ca = parent_ca(ca);
2842 }
2843 rcu_read_unlock();
2844#endif
2845}
2846
2847
2848/*
2849 * Account user cpu time to a process.
2850 * @p: the process that the cpu time gets accounted to
2851 * @cputime: the cpu time spent in user space since the last update
2852 * @cputime_scaled: cputime scaled by cpu frequency
2853 */
2854void account_user_time(struct task_struct *p, cputime_t cputime,
2855 cputime_t cputime_scaled)
2856{
2857 int index;
2858
2859 /* Add user time to process. */
2860 p->utime += cputime;
2861 p->utimescaled += cputime_scaled;
2862 account_group_user_time(p, cputime);
2863
2864 index = (TASK_NICE(p) > 0) ? CPUTIME_NICE : CPUTIME_USER;
2865
2866 /* Add user time to cpustat. */
2867 task_group_account_field(p, index, (__force u64) cputime);
2868
2869 /* Account for user time used */
2870 acct_update_integrals(p);
2871}
2872
2873/*
2874 * Account guest cpu time to a process.
2875 * @p: the process that the cpu time gets accounted to
2876 * @cputime: the cpu time spent in virtual machine since the last update
2877 * @cputime_scaled: cputime scaled by cpu frequency
2878 */
2879static void account_guest_time(struct task_struct *p, cputime_t cputime,
2880 cputime_t cputime_scaled)
2881{
2882 u64 *cpustat = kcpustat_this_cpu->cpustat;
2883
2884 /* Add guest time to process. */
2885 p->utime += cputime;
2886 p->utimescaled += cputime_scaled;
2887 account_group_user_time(p, cputime);
2888 p->gtime += cputime;
2889
2890 /* Add guest time to cpustat. */
2891 if (TASK_NICE(p) > 0) {
2892 cpustat[CPUTIME_NICE] += (__force u64) cputime;
2893 cpustat[CPUTIME_GUEST_NICE] += (__force u64) cputime;
2894 } else {
2895 cpustat[CPUTIME_USER] += (__force u64) cputime;
2896 cpustat[CPUTIME_GUEST] += (__force u64) cputime;
2897 }
2898}
2899
2900/*
2901 * Account system cpu time to a process and desired cpustat field
2902 * @p: the process that the cpu time gets accounted to
2903 * @cputime: the cpu time spent in kernel space since the last update
2904 * @cputime_scaled: cputime scaled by cpu frequency
2905 * @target_cputime64: pointer to cpustat field that has to be updated
2906 */
2907static inline
2908void __account_system_time(struct task_struct *p, cputime_t cputime,
2909 cputime_t cputime_scaled, int index)
2910{
2911 /* Add system time to process. */
2912 p->stime += cputime;
2913 p->stimescaled += cputime_scaled;
2914 account_group_system_time(p, cputime);
2915
2916 /* Add system time to cpustat. */
2917 task_group_account_field(p, index, (__force u64) cputime);
2918
2919 /* Account for system time used */
2920 acct_update_integrals(p);
2921}
2922
2923/*
2924 * Account system cpu time to a process.
2925 * @p: the process that the cpu time gets accounted to
2926 * @hardirq_offset: the offset to subtract from hardirq_count()
2927 * @cputime: the cpu time spent in kernel space since the last update
2928 * @cputime_scaled: cputime scaled by cpu frequency
2929 */
2930void account_system_time(struct task_struct *p, int hardirq_offset,
2931 cputime_t cputime, cputime_t cputime_scaled)
2932{
2933 int index;
2934
2935 if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) {
2936 account_guest_time(p, cputime, cputime_scaled);
2937 return;
2938 }
2939
2940 if (hardirq_count() - hardirq_offset)
2941 index = CPUTIME_IRQ;
2942 else if (in_serving_softirq())
2943 index = CPUTIME_SOFTIRQ;
2944 else
2945 index = CPUTIME_SYSTEM;
2946
2947 __account_system_time(p, cputime, cputime_scaled, index);
2948}
2949
2950/*
2951 * Account for involuntary wait time.
2952 * @cputime: the cpu time spent in involuntary wait
2953 */
2954void account_steal_time(cputime_t cputime)
2955{
2956 u64 *cpustat = kcpustat_this_cpu->cpustat;
2957
2958 cpustat[CPUTIME_STEAL] += (__force u64) cputime;
2959}
2960
2961/*
2962 * Account for idle time.
2963 * @cputime: the cpu time spent in idle wait
2964 */
2965void account_idle_time(cputime_t cputime)
2966{
2967 u64 *cpustat = kcpustat_this_cpu->cpustat;
2968 struct rq *rq = this_rq();
2969
2970 if (atomic_read(&rq->nr_iowait) > 0)
2971 cpustat[CPUTIME_IOWAIT] += (__force u64) cputime;
2972 else
2973 cpustat[CPUTIME_IDLE] += (__force u64) cputime;
2974}
2975
2976static __always_inline bool steal_account_process_tick(void)
2977{
2978#ifdef CONFIG_PARAVIRT
2979 if (static_key_false(&paravirt_steal_enabled)) {
2980 u64 steal, st = 0;
2981
2982 steal = paravirt_steal_clock(smp_processor_id());
2983 steal -= this_rq()->prev_steal_time;
2984
2985 st = steal_ticks(steal);
2986 this_rq()->prev_steal_time += st * TICK_NSEC;
2987
2988 account_steal_time(st);
2989 return st;
2990 }
2991#endif
2992 return false;
2993}
2994
2995#ifndef CONFIG_VIRT_CPU_ACCOUNTING
2996
2997#ifdef CONFIG_IRQ_TIME_ACCOUNTING
2998/*
2999 * Account a tick to a process and cpustat
3000 * @p: the process that the cpu time gets accounted to
3001 * @user_tick: is the tick from userspace
3002 * @rq: the pointer to rq
3003 *
3004 * Tick demultiplexing follows the order
3005 * - pending hardirq update
3006 * - pending softirq update
3007 * - user_time
3008 * - idle_time
3009 * - system time
3010 * - check for guest_time
3011 * - else account as system_time
3012 *
3013 * Check for hardirq is done both for system and user time as there is
3014 * no timer going off while we are on hardirq and hence we may never get an
3015 * opportunity to update it solely in system time.
3016 * p->stime and friends are only updated on system time and not on irq
3017 * softirq as those do not count in task exec_runtime any more.
3018 */
3019static void irqtime_account_process_tick(struct task_struct *p, int user_tick,
3020 struct rq *rq)
3021{
3022 cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy);
3023 u64 *cpustat = kcpustat_this_cpu->cpustat;
3024
3025 if (steal_account_process_tick())
3026 return;
3027
3028 if (irqtime_account_hi_update()) {
3029 cpustat[CPUTIME_IRQ] += (__force u64) cputime_one_jiffy;
3030 } else if (irqtime_account_si_update()) {
3031 cpustat[CPUTIME_SOFTIRQ] += (__force u64) cputime_one_jiffy;
3032 } else if (this_cpu_ksoftirqd() == p) {
3033 /*
3034 * ksoftirqd time do not get accounted in cpu_softirq_time.
3035 * So, we have to handle it separately here.
3036 * Also, p->stime needs to be updated for ksoftirqd.
3037 */
3038 __account_system_time(p, cputime_one_jiffy, one_jiffy_scaled,
3039 CPUTIME_SOFTIRQ);
3040 } else if (user_tick) {
3041 account_user_time(p, cputime_one_jiffy, one_jiffy_scaled);
3042 } else if (p == rq->idle) {
3043 account_idle_time(cputime_one_jiffy);
3044 } else if (p->flags & PF_VCPU) { /* System time or guest time */
3045 account_guest_time(p, cputime_one_jiffy, one_jiffy_scaled);
3046 } else {
3047 __account_system_time(p, cputime_one_jiffy, one_jiffy_scaled,
3048 CPUTIME_SYSTEM);
3049 }
3050}
3051
3052static void irqtime_account_idle_ticks(int ticks)
3053{
3054 int i;
3055 struct rq *rq = this_rq();
3056
3057 for (i = 0; i < ticks; i++)
3058 irqtime_account_process_tick(current, 0, rq);
3059}
3060#else /* CONFIG_IRQ_TIME_ACCOUNTING */
3061static void irqtime_account_idle_ticks(int ticks) {}
3062static void irqtime_account_process_tick(struct task_struct *p, int user_tick,
3063 struct rq *rq) {}
3064#endif /* CONFIG_IRQ_TIME_ACCOUNTING */
3065
3066/*
3067 * Account a single tick of cpu time.
3068 * @p: the process that the cpu time gets accounted to
3069 * @user_tick: indicates if the tick is a user or a system tick
3070 */
3071void account_process_tick(struct task_struct *p, int user_tick)
3072{
3073 cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy);
3074 struct rq *rq = this_rq();
3075
3076 if (sched_clock_irqtime) {
3077 irqtime_account_process_tick(p, user_tick, rq);
3078 return;
3079 }
3080
3081 if (steal_account_process_tick())
3082 return;
3083
3084 if (user_tick)
3085 account_user_time(p, cputime_one_jiffy, one_jiffy_scaled);
3086 else if ((p != rq->idle) || (irq_count() != HARDIRQ_OFFSET))
3087 account_system_time(p, HARDIRQ_OFFSET, cputime_one_jiffy,
3088 one_jiffy_scaled);
3089 else
3090 account_idle_time(cputime_one_jiffy);
3091}
3092
3093/*
3094 * Account multiple ticks of steal time.
3095 * @p: the process from which the cpu time has been stolen
3096 * @ticks: number of stolen ticks
3097 */
3098void account_steal_ticks(unsigned long ticks)
3099{
3100 account_steal_time(jiffies_to_cputime(ticks));
3101}
3102
3103/*
3104 * Account multiple ticks of idle time.
3105 * @ticks: number of stolen ticks
3106 */
3107void account_idle_ticks(unsigned long ticks)
3108{
3109
3110 if (sched_clock_irqtime) {
3111 irqtime_account_idle_ticks(ticks);
3112 return;
3113 }
3114
3115 account_idle_time(jiffies_to_cputime(ticks));
3116}
3117
3118#endif
3119
3120/*
3121 * Use precise platform statistics if available:
3122 */
3123#ifdef CONFIG_VIRT_CPU_ACCOUNTING
3124void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st)
3125{
3126 *ut = p->utime;
3127 *st = p->stime;
3128}
3129
3130void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st)
3131{
3132 struct task_cputime cputime;
3133
3134 thread_group_cputime(p, &cputime);
3135
3136 *ut = cputime.utime;
3137 *st = cputime.stime;
3138}
3139#else
3140
3141#ifndef nsecs_to_cputime
3142# define nsecs_to_cputime(__nsecs) nsecs_to_jiffies(__nsecs)
3143#endif
3144
3145static cputime_t scale_utime(cputime_t utime, cputime_t rtime, cputime_t total)
3146{
3147 u64 temp = (__force u64) rtime;
3148
3149 temp *= (__force u64) utime;
3150
3151 if (sizeof(cputime_t) == 4)
3152 temp = div_u64(temp, (__force u32) total);
3153 else
3154 temp = div64_u64(temp, (__force u64) total);
3155
3156 return (__force cputime_t) temp;
3157}
3158
3159void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st)
3160{
3161 cputime_t rtime, utime = p->utime, total = utime + p->stime;
3162
3163 /*
3164 * Use CFS's precise accounting:
3165 */
3166 rtime = nsecs_to_cputime(p->se.sum_exec_runtime);
3167
3168 if (total)
3169 utime = scale_utime(utime, rtime, total);
3170 else
3171 utime = rtime;
3172
3173 /*
3174 * Compare with previous values, to keep monotonicity:
3175 */
3176 p->prev_utime = max(p->prev_utime, utime);
3177 p->prev_stime = max(p->prev_stime, rtime - p->prev_utime);
3178
3179 *ut = p->prev_utime;
3180 *st = p->prev_stime;
3181}
3182
3183/*
3184 * Must be called with siglock held.
3185 */
3186void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st)
3187{
3188 struct signal_struct *sig = p->signal;
3189 struct task_cputime cputime;
3190 cputime_t rtime, utime, total;
3191
3192 thread_group_cputime(p, &cputime);
3193
3194 total = cputime.utime + cputime.stime;
3195 rtime = nsecs_to_cputime(cputime.sum_exec_runtime);
3196
3197 if (total)
3198 utime = scale_utime(cputime.utime, rtime, total);
3199 else
3200 utime = rtime;
3201
3202 sig->prev_utime = max(sig->prev_utime, utime);
3203 sig->prev_stime = max(sig->prev_stime, rtime - sig->prev_utime);
3204
3205 *ut = sig->prev_utime;
3206 *st = sig->prev_stime;
3207}
3208#endif
3209
3210/* 2619/*
3211 * This function gets called by the timer code, with HZ frequency. 2620 * This function gets called by the timer code, with HZ frequency.
3212 * We call it with interrupts disabled. 2621 * We call it with interrupts disabled.
@@ -3367,6 +2776,40 @@ pick_next_task(struct rq *rq)
3367 2776
3368/* 2777/*
3369 * __schedule() is the main scheduler function. 2778 * __schedule() is the main scheduler function.
2779 *
2780 * The main means of driving the scheduler and thus entering this function are:
2781 *
2782 * 1. Explicit blocking: mutex, semaphore, waitqueue, etc.
2783 *
2784 * 2. TIF_NEED_RESCHED flag is checked on interrupt and userspace return
2785 * paths. For example, see arch/x86/entry_64.S.
2786 *
2787 * To drive preemption between tasks, the scheduler sets the flag in timer
2788 * interrupt handler scheduler_tick().
2789 *
2790 * 3. Wakeups don't really cause entry into schedule(). They add a
2791 * task to the run-queue and that's it.
2792 *
2793 * Now, if the new task added to the run-queue preempts the current
2794 * task, then the wakeup sets TIF_NEED_RESCHED and schedule() gets
2795 * called on the nearest possible occasion:
2796 *
2797 * - If the kernel is preemptible (CONFIG_PREEMPT=y):
2798 *
2799 * - in syscall or exception context, at the next outmost
2800 * preempt_enable(). (this might be as soon as the wake_up()'s
2801 * spin_unlock()!)
2802 *
2803 * - in IRQ context, return from interrupt-handler to
2804 * preemptible context
2805 *
2806 * - If the kernel is not preemptible (CONFIG_PREEMPT is not set)
2807 * then at the next:
2808 *
2809 * - cond_resched() call
2810 * - explicit schedule() call
2811 * - return from syscall or exception to user-space
2812 * - return from interrupt-handler to user-space
3370 */ 2813 */
3371static void __sched __schedule(void) 2814static void __sched __schedule(void)
3372{ 2815{
@@ -3468,6 +2911,21 @@ asmlinkage void __sched schedule(void)
3468} 2911}
3469EXPORT_SYMBOL(schedule); 2912EXPORT_SYMBOL(schedule);
3470 2913
2914#ifdef CONFIG_RCU_USER_QS
2915asmlinkage void __sched schedule_user(void)
2916{
2917 /*
2918 * If we come here after a random call to set_need_resched(),
2919 * or we have been woken up remotely but the IPI has not yet arrived,
2920 * we haven't yet exited the RCU idle mode. Do it here manually until
2921 * we find a better solution.
2922 */
2923 rcu_user_exit();
2924 schedule();
2925 rcu_user_enter();
2926}
2927#endif
2928
3471/** 2929/**
3472 * schedule_preempt_disabled - called with preemption disabled 2930 * schedule_preempt_disabled - called with preemption disabled
3473 * 2931 *
@@ -3569,6 +3027,7 @@ asmlinkage void __sched preempt_schedule_irq(void)
3569 /* Catch callers which need to be fixed */ 3027 /* Catch callers which need to be fixed */
3570 BUG_ON(ti->preempt_count || !irqs_disabled()); 3028 BUG_ON(ti->preempt_count || !irqs_disabled());
3571 3029
3030 rcu_user_exit();
3572 do { 3031 do {
3573 add_preempt_count(PREEMPT_ACTIVE); 3032 add_preempt_count(PREEMPT_ACTIVE);
3574 local_irq_enable(); 3033 local_irq_enable();
@@ -4868,13 +4327,6 @@ again:
4868 */ 4327 */
4869 if (preempt && rq != p_rq) 4328 if (preempt && rq != p_rq)
4870 resched_task(p_rq->curr); 4329 resched_task(p_rq->curr);
4871 } else {
4872 /*
4873 * We might have set it in task_yield_fair(), but are
4874 * not going to schedule(), so don't want to skip
4875 * the next update.
4876 */
4877 rq->skip_clock_update = 0;
4878 } 4330 }
4879 4331
4880out: 4332out:
@@ -5416,16 +4868,25 @@ static void sd_free_ctl_entry(struct ctl_table **tablep)
5416 *tablep = NULL; 4868 *tablep = NULL;
5417} 4869}
5418 4870
4871static int min_load_idx = 0;
4872static int max_load_idx = CPU_LOAD_IDX_MAX;
4873
5419static void 4874static void
5420set_table_entry(struct ctl_table *entry, 4875set_table_entry(struct ctl_table *entry,
5421 const char *procname, void *data, int maxlen, 4876 const char *procname, void *data, int maxlen,
5422 umode_t mode, proc_handler *proc_handler) 4877 umode_t mode, proc_handler *proc_handler,
4878 bool load_idx)
5423{ 4879{
5424 entry->procname = procname; 4880 entry->procname = procname;
5425 entry->data = data; 4881 entry->data = data;
5426 entry->maxlen = maxlen; 4882 entry->maxlen = maxlen;
5427 entry->mode = mode; 4883 entry->mode = mode;
5428 entry->proc_handler = proc_handler; 4884 entry->proc_handler = proc_handler;
4885
4886 if (load_idx) {
4887 entry->extra1 = &min_load_idx;
4888 entry->extra2 = &max_load_idx;
4889 }
5429} 4890}
5430 4891
5431static struct ctl_table * 4892static struct ctl_table *
@@ -5437,30 +4898,30 @@ sd_alloc_ctl_domain_table(struct sched_domain *sd)
5437 return NULL; 4898 return NULL;
5438 4899
5439 set_table_entry(&table[0], "min_interval", &sd->min_interval, 4900 set_table_entry(&table[0], "min_interval", &sd->min_interval,
5440 sizeof(long), 0644, proc_doulongvec_minmax); 4901 sizeof(long), 0644, proc_doulongvec_minmax, false);
5441 set_table_entry(&table[1], "max_interval", &sd->max_interval, 4902 set_table_entry(&table[1], "max_interval", &sd->max_interval,
5442 sizeof(long), 0644, proc_doulongvec_minmax); 4903 sizeof(long), 0644, proc_doulongvec_minmax, false);
5443 set_table_entry(&table[2], "busy_idx", &sd->busy_idx, 4904 set_table_entry(&table[2], "busy_idx", &sd->busy_idx,
5444 sizeof(int), 0644, proc_dointvec_minmax); 4905 sizeof(int), 0644, proc_dointvec_minmax, true);
5445 set_table_entry(&table[3], "idle_idx", &sd->idle_idx, 4906 set_table_entry(&table[3], "idle_idx", &sd->idle_idx,
5446 sizeof(int), 0644, proc_dointvec_minmax); 4907 sizeof(int), 0644, proc_dointvec_minmax, true);
5447 set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx, 4908 set_table_entry(&table[4], "newidle_idx", &sd->newidle_idx,
5448 sizeof(int), 0644, proc_dointvec_minmax); 4909 sizeof(int), 0644, proc_dointvec_minmax, true);
5449 set_table_entry(&table[5], "wake_idx", &sd->wake_idx, 4910 set_table_entry(&table[5], "wake_idx", &sd->wake_idx,
5450 sizeof(int), 0644, proc_dointvec_minmax); 4911 sizeof(int), 0644, proc_dointvec_minmax, true);
5451 set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx, 4912 set_table_entry(&table[6], "forkexec_idx", &sd->forkexec_idx,
5452 sizeof(int), 0644, proc_dointvec_minmax); 4913 sizeof(int), 0644, proc_dointvec_minmax, true);
5453 set_table_entry(&table[7], "busy_factor", &sd->busy_factor, 4914 set_table_entry(&table[7], "busy_factor", &sd->busy_factor,
5454 sizeof(int), 0644, proc_dointvec_minmax); 4915 sizeof(int), 0644, proc_dointvec_minmax, false);
5455 set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct, 4916 set_table_entry(&table[8], "imbalance_pct", &sd->imbalance_pct,
5456 sizeof(int), 0644, proc_dointvec_minmax); 4917 sizeof(int), 0644, proc_dointvec_minmax, false);
5457 set_table_entry(&table[9], "cache_nice_tries", 4918 set_table_entry(&table[9], "cache_nice_tries",
5458 &sd->cache_nice_tries, 4919 &sd->cache_nice_tries,
5459 sizeof(int), 0644, proc_dointvec_minmax); 4920 sizeof(int), 0644, proc_dointvec_minmax, false);
5460 set_table_entry(&table[10], "flags", &sd->flags, 4921 set_table_entry(&table[10], "flags", &sd->flags,
5461 sizeof(int), 0644, proc_dointvec_minmax); 4922 sizeof(int), 0644, proc_dointvec_minmax, false);
5462 set_table_entry(&table[11], "name", sd->name, 4923 set_table_entry(&table[11], "name", sd->name,
5463 CORENAME_MAX_SIZE, 0444, proc_dostring); 4924 CORENAME_MAX_SIZE, 0444, proc_dostring, false);
5464 /* &table[12] is terminator */ 4925 /* &table[12] is terminator */
5465 4926
5466 return table; 4927 return table;
@@ -5604,7 +5065,9 @@ migration_call(struct notifier_block *nfb, unsigned long action, void *hcpu)
5604 migrate_tasks(cpu); 5065 migrate_tasks(cpu);
5605 BUG_ON(rq->nr_running != 1); /* the migration thread */ 5066 BUG_ON(rq->nr_running != 1); /* the migration thread */
5606 raw_spin_unlock_irqrestore(&rq->lock, flags); 5067 raw_spin_unlock_irqrestore(&rq->lock, flags);
5068 break;
5607 5069
5070 case CPU_DEAD:
5608 calc_load_migrate(rq); 5071 calc_load_migrate(rq);
5609 break; 5072 break;
5610#endif 5073#endif
@@ -6537,7 +6000,6 @@ sd_numa_init(struct sched_domain_topology_level *tl, int cpu)
6537 | 0*SD_BALANCE_FORK 6000 | 0*SD_BALANCE_FORK
6538 | 0*SD_BALANCE_WAKE 6001 | 0*SD_BALANCE_WAKE
6539 | 0*SD_WAKE_AFFINE 6002 | 0*SD_WAKE_AFFINE
6540 | 0*SD_PREFER_LOCAL
6541 | 0*SD_SHARE_CPUPOWER 6003 | 0*SD_SHARE_CPUPOWER
6542 | 0*SD_SHARE_PKG_RESOURCES 6004 | 0*SD_SHARE_PKG_RESOURCES
6543 | 1*SD_SERIALIZE 6005 | 1*SD_SERIALIZE
@@ -6660,6 +6122,17 @@ static void sched_init_numa(void)
6660 * numbers. 6122 * numbers.
6661 */ 6123 */
6662 6124
6125 /*
6126 * Here, we should temporarily reset sched_domains_numa_levels to 0.
6127 * If it fails to allocate memory for array sched_domains_numa_masks[][],
6128 * the array will contain less then 'level' members. This could be
6129 * dangerous when we use it to iterate array sched_domains_numa_masks[][]
6130 * in other functions.
6131 *
6132 * We reset it to 'level' at the end of this function.
6133 */
6134 sched_domains_numa_levels = 0;
6135
6663 sched_domains_numa_masks = kzalloc(sizeof(void *) * level, GFP_KERNEL); 6136 sched_domains_numa_masks = kzalloc(sizeof(void *) * level, GFP_KERNEL);
6664 if (!sched_domains_numa_masks) 6137 if (!sched_domains_numa_masks)
6665 return; 6138 return;
@@ -6714,11 +6187,68 @@ static void sched_init_numa(void)
6714 } 6187 }
6715 6188
6716 sched_domain_topology = tl; 6189 sched_domain_topology = tl;
6190
6191 sched_domains_numa_levels = level;
6192}
6193
6194static void sched_domains_numa_masks_set(int cpu)
6195{
6196 int i, j;
6197 int node = cpu_to_node(cpu);
6198
6199 for (i = 0; i < sched_domains_numa_levels; i++) {
6200 for (j = 0; j < nr_node_ids; j++) {
6201 if (node_distance(j, node) <= sched_domains_numa_distance[i])
6202 cpumask_set_cpu(cpu, sched_domains_numa_masks[i][j]);
6203 }
6204 }
6205}
6206
6207static void sched_domains_numa_masks_clear(int cpu)
6208{
6209 int i, j;
6210 for (i = 0; i < sched_domains_numa_levels; i++) {
6211 for (j = 0; j < nr_node_ids; j++)
6212 cpumask_clear_cpu(cpu, sched_domains_numa_masks[i][j]);
6213 }
6214}
6215
6216/*
6217 * Update sched_domains_numa_masks[level][node] array when new cpus
6218 * are onlined.
6219 */
6220static int sched_domains_numa_masks_update(struct notifier_block *nfb,
6221 unsigned long action,
6222 void *hcpu)
6223{
6224 int cpu = (long)hcpu;
6225
6226 switch (action & ~CPU_TASKS_FROZEN) {
6227 case CPU_ONLINE:
6228 sched_domains_numa_masks_set(cpu);
6229 break;
6230
6231 case CPU_DEAD:
6232 sched_domains_numa_masks_clear(cpu);
6233 break;
6234
6235 default:
6236 return NOTIFY_DONE;
6237 }
6238
6239 return NOTIFY_OK;
6717} 6240}
6718#else 6241#else
6719static inline void sched_init_numa(void) 6242static inline void sched_init_numa(void)
6720{ 6243{
6721} 6244}
6245
6246static int sched_domains_numa_masks_update(struct notifier_block *nfb,
6247 unsigned long action,
6248 void *hcpu)
6249{
6250 return 0;
6251}
6722#endif /* CONFIG_NUMA */ 6252#endif /* CONFIG_NUMA */
6723 6253
6724static int __sdt_alloc(const struct cpumask *cpu_map) 6254static int __sdt_alloc(const struct cpumask *cpu_map)
@@ -7167,6 +6697,7 @@ void __init sched_init_smp(void)
7167 mutex_unlock(&sched_domains_mutex); 6697 mutex_unlock(&sched_domains_mutex);
7168 put_online_cpus(); 6698 put_online_cpus();
7169 6699
6700 hotcpu_notifier(sched_domains_numa_masks_update, CPU_PRI_SCHED_ACTIVE);
7170 hotcpu_notifier(cpuset_cpu_active, CPU_PRI_CPUSET_ACTIVE); 6701 hotcpu_notifier(cpuset_cpu_active, CPU_PRI_CPUSET_ACTIVE);
7171 hotcpu_notifier(cpuset_cpu_inactive, CPU_PRI_CPUSET_INACTIVE); 6702 hotcpu_notifier(cpuset_cpu_inactive, CPU_PRI_CPUSET_INACTIVE);
7172 6703
@@ -8335,6 +7866,8 @@ struct cgroup_subsys cpu_cgroup_subsys = {
8335 * (balbir@in.ibm.com). 7866 * (balbir@in.ibm.com).
8336 */ 7867 */
8337 7868
7869struct cpuacct root_cpuacct;
7870
8338/* create a new cpu accounting group */ 7871/* create a new cpu accounting group */
8339static struct cgroup_subsys_state *cpuacct_create(struct cgroup *cgrp) 7872static struct cgroup_subsys_state *cpuacct_create(struct cgroup *cgrp)
8340{ 7873{
diff --git a/kernel/sched/cputime.c b/kernel/sched/cputime.c
new file mode 100644
index 000000000000..81b763ba58a6
--- /dev/null
+++ b/kernel/sched/cputime.c
@@ -0,0 +1,530 @@
1#include <linux/export.h>
2#include <linux/sched.h>
3#include <linux/tsacct_kern.h>
4#include <linux/kernel_stat.h>
5#include <linux/static_key.h>
6#include "sched.h"
7
8
9#ifdef CONFIG_IRQ_TIME_ACCOUNTING
10
11/*
12 * There are no locks covering percpu hardirq/softirq time.
13 * They are only modified in vtime_account, on corresponding CPU
14 * with interrupts disabled. So, writes are safe.
15 * They are read and saved off onto struct rq in update_rq_clock().
16 * This may result in other CPU reading this CPU's irq time and can
17 * race with irq/vtime_account on this CPU. We would either get old
18 * or new value with a side effect of accounting a slice of irq time to wrong
19 * task when irq is in progress while we read rq->clock. That is a worthy
20 * compromise in place of having locks on each irq in account_system_time.
21 */
22DEFINE_PER_CPU(u64, cpu_hardirq_time);
23DEFINE_PER_CPU(u64, cpu_softirq_time);
24
25static DEFINE_PER_CPU(u64, irq_start_time);
26static int sched_clock_irqtime;
27
28void enable_sched_clock_irqtime(void)
29{
30 sched_clock_irqtime = 1;
31}
32
33void disable_sched_clock_irqtime(void)
34{
35 sched_clock_irqtime = 0;
36}
37
38#ifndef CONFIG_64BIT
39DEFINE_PER_CPU(seqcount_t, irq_time_seq);
40#endif /* CONFIG_64BIT */
41
42/*
43 * Called before incrementing preempt_count on {soft,}irq_enter
44 * and before decrementing preempt_count on {soft,}irq_exit.
45 */
46void vtime_account(struct task_struct *curr)
47{
48 unsigned long flags;
49 s64 delta;
50 int cpu;
51
52 if (!sched_clock_irqtime)
53 return;
54
55 local_irq_save(flags);
56
57 cpu = smp_processor_id();
58 delta = sched_clock_cpu(cpu) - __this_cpu_read(irq_start_time);
59 __this_cpu_add(irq_start_time, delta);
60
61 irq_time_write_begin();
62 /*
63 * We do not account for softirq time from ksoftirqd here.
64 * We want to continue accounting softirq time to ksoftirqd thread
65 * in that case, so as not to confuse scheduler with a special task
66 * that do not consume any time, but still wants to run.
67 */
68 if (hardirq_count())
69 __this_cpu_add(cpu_hardirq_time, delta);
70 else if (in_serving_softirq() && curr != this_cpu_ksoftirqd())
71 __this_cpu_add(cpu_softirq_time, delta);
72
73 irq_time_write_end();
74 local_irq_restore(flags);
75}
76EXPORT_SYMBOL_GPL(vtime_account);
77
78static int irqtime_account_hi_update(void)
79{
80 u64 *cpustat = kcpustat_this_cpu->cpustat;
81 unsigned long flags;
82 u64 latest_ns;
83 int ret = 0;
84
85 local_irq_save(flags);
86 latest_ns = this_cpu_read(cpu_hardirq_time);
87 if (nsecs_to_cputime64(latest_ns) > cpustat[CPUTIME_IRQ])
88 ret = 1;
89 local_irq_restore(flags);
90 return ret;
91}
92
93static int irqtime_account_si_update(void)
94{
95 u64 *cpustat = kcpustat_this_cpu->cpustat;
96 unsigned long flags;
97 u64 latest_ns;
98 int ret = 0;
99
100 local_irq_save(flags);
101 latest_ns = this_cpu_read(cpu_softirq_time);
102 if (nsecs_to_cputime64(latest_ns) > cpustat[CPUTIME_SOFTIRQ])
103 ret = 1;
104 local_irq_restore(flags);
105 return ret;
106}
107
108#else /* CONFIG_IRQ_TIME_ACCOUNTING */
109
110#define sched_clock_irqtime (0)
111
112#endif /* !CONFIG_IRQ_TIME_ACCOUNTING */
113
114static inline void task_group_account_field(struct task_struct *p, int index,
115 u64 tmp)
116{
117#ifdef CONFIG_CGROUP_CPUACCT
118 struct kernel_cpustat *kcpustat;
119 struct cpuacct *ca;
120#endif
121 /*
122 * Since all updates are sure to touch the root cgroup, we
123 * get ourselves ahead and touch it first. If the root cgroup
124 * is the only cgroup, then nothing else should be necessary.
125 *
126 */
127 __get_cpu_var(kernel_cpustat).cpustat[index] += tmp;
128
129#ifdef CONFIG_CGROUP_CPUACCT
130 if (unlikely(!cpuacct_subsys.active))
131 return;
132
133 rcu_read_lock();
134 ca = task_ca(p);
135 while (ca && (ca != &root_cpuacct)) {
136 kcpustat = this_cpu_ptr(ca->cpustat);
137 kcpustat->cpustat[index] += tmp;
138 ca = parent_ca(ca);
139 }
140 rcu_read_unlock();
141#endif
142}
143
144/*
145 * Account user cpu time to a process.
146 * @p: the process that the cpu time gets accounted to
147 * @cputime: the cpu time spent in user space since the last update
148 * @cputime_scaled: cputime scaled by cpu frequency
149 */
150void account_user_time(struct task_struct *p, cputime_t cputime,
151 cputime_t cputime_scaled)
152{
153 int index;
154
155 /* Add user time to process. */
156 p->utime += cputime;
157 p->utimescaled += cputime_scaled;
158 account_group_user_time(p, cputime);
159
160 index = (TASK_NICE(p) > 0) ? CPUTIME_NICE : CPUTIME_USER;
161
162 /* Add user time to cpustat. */
163 task_group_account_field(p, index, (__force u64) cputime);
164
165 /* Account for user time used */
166 acct_update_integrals(p);
167}
168
169/*
170 * Account guest cpu time to a process.
171 * @p: the process that the cpu time gets accounted to
172 * @cputime: the cpu time spent in virtual machine since the last update
173 * @cputime_scaled: cputime scaled by cpu frequency
174 */
175static void account_guest_time(struct task_struct *p, cputime_t cputime,
176 cputime_t cputime_scaled)
177{
178 u64 *cpustat = kcpustat_this_cpu->cpustat;
179
180 /* Add guest time to process. */
181 p->utime += cputime;
182 p->utimescaled += cputime_scaled;
183 account_group_user_time(p, cputime);
184 p->gtime += cputime;
185
186 /* Add guest time to cpustat. */
187 if (TASK_NICE(p) > 0) {
188 cpustat[CPUTIME_NICE] += (__force u64) cputime;
189 cpustat[CPUTIME_GUEST_NICE] += (__force u64) cputime;
190 } else {
191 cpustat[CPUTIME_USER] += (__force u64) cputime;
192 cpustat[CPUTIME_GUEST] += (__force u64) cputime;
193 }
194}
195
196/*
197 * Account system cpu time to a process and desired cpustat field
198 * @p: the process that the cpu time gets accounted to
199 * @cputime: the cpu time spent in kernel space since the last update
200 * @cputime_scaled: cputime scaled by cpu frequency
201 * @target_cputime64: pointer to cpustat field that has to be updated
202 */
203static inline
204void __account_system_time(struct task_struct *p, cputime_t cputime,
205 cputime_t cputime_scaled, int index)
206{
207 /* Add system time to process. */
208 p->stime += cputime;
209 p->stimescaled += cputime_scaled;
210 account_group_system_time(p, cputime);
211
212 /* Add system time to cpustat. */
213 task_group_account_field(p, index, (__force u64) cputime);
214
215 /* Account for system time used */
216 acct_update_integrals(p);
217}
218
219/*
220 * Account system cpu time to a process.
221 * @p: the process that the cpu time gets accounted to
222 * @hardirq_offset: the offset to subtract from hardirq_count()
223 * @cputime: the cpu time spent in kernel space since the last update
224 * @cputime_scaled: cputime scaled by cpu frequency
225 */
226void account_system_time(struct task_struct *p, int hardirq_offset,
227 cputime_t cputime, cputime_t cputime_scaled)
228{
229 int index;
230
231 if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) {
232 account_guest_time(p, cputime, cputime_scaled);
233 return;
234 }
235
236 if (hardirq_count() - hardirq_offset)
237 index = CPUTIME_IRQ;
238 else if (in_serving_softirq())
239 index = CPUTIME_SOFTIRQ;
240 else
241 index = CPUTIME_SYSTEM;
242
243 __account_system_time(p, cputime, cputime_scaled, index);
244}
245
246/*
247 * Account for involuntary wait time.
248 * @cputime: the cpu time spent in involuntary wait
249 */
250void account_steal_time(cputime_t cputime)
251{
252 u64 *cpustat = kcpustat_this_cpu->cpustat;
253
254 cpustat[CPUTIME_STEAL] += (__force u64) cputime;
255}
256
257/*
258 * Account for idle time.
259 * @cputime: the cpu time spent in idle wait
260 */
261void account_idle_time(cputime_t cputime)
262{
263 u64 *cpustat = kcpustat_this_cpu->cpustat;
264 struct rq *rq = this_rq();
265
266 if (atomic_read(&rq->nr_iowait) > 0)
267 cpustat[CPUTIME_IOWAIT] += (__force u64) cputime;
268 else
269 cpustat[CPUTIME_IDLE] += (__force u64) cputime;
270}
271
272static __always_inline bool steal_account_process_tick(void)
273{
274#ifdef CONFIG_PARAVIRT
275 if (static_key_false(&paravirt_steal_enabled)) {
276 u64 steal, st = 0;
277
278 steal = paravirt_steal_clock(smp_processor_id());
279 steal -= this_rq()->prev_steal_time;
280
281 st = steal_ticks(steal);
282 this_rq()->prev_steal_time += st * TICK_NSEC;
283
284 account_steal_time(st);
285 return st;
286 }
287#endif
288 return false;
289}
290
291#ifndef CONFIG_VIRT_CPU_ACCOUNTING
292
293#ifdef CONFIG_IRQ_TIME_ACCOUNTING
294/*
295 * Account a tick to a process and cpustat
296 * @p: the process that the cpu time gets accounted to
297 * @user_tick: is the tick from userspace
298 * @rq: the pointer to rq
299 *
300 * Tick demultiplexing follows the order
301 * - pending hardirq update
302 * - pending softirq update
303 * - user_time
304 * - idle_time
305 * - system time
306 * - check for guest_time
307 * - else account as system_time
308 *
309 * Check for hardirq is done both for system and user time as there is
310 * no timer going off while we are on hardirq and hence we may never get an
311 * opportunity to update it solely in system time.
312 * p->stime and friends are only updated on system time and not on irq
313 * softirq as those do not count in task exec_runtime any more.
314 */
315static void irqtime_account_process_tick(struct task_struct *p, int user_tick,
316 struct rq *rq)
317{
318 cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy);
319 u64 *cpustat = kcpustat_this_cpu->cpustat;
320
321 if (steal_account_process_tick())
322 return;
323
324 if (irqtime_account_hi_update()) {
325 cpustat[CPUTIME_IRQ] += (__force u64) cputime_one_jiffy;
326 } else if (irqtime_account_si_update()) {
327 cpustat[CPUTIME_SOFTIRQ] += (__force u64) cputime_one_jiffy;
328 } else if (this_cpu_ksoftirqd() == p) {
329 /*
330 * ksoftirqd time do not get accounted in cpu_softirq_time.
331 * So, we have to handle it separately here.
332 * Also, p->stime needs to be updated for ksoftirqd.
333 */
334 __account_system_time(p, cputime_one_jiffy, one_jiffy_scaled,
335 CPUTIME_SOFTIRQ);
336 } else if (user_tick) {
337 account_user_time(p, cputime_one_jiffy, one_jiffy_scaled);
338 } else if (p == rq->idle) {
339 account_idle_time(cputime_one_jiffy);
340 } else if (p->flags & PF_VCPU) { /* System time or guest time */
341 account_guest_time(p, cputime_one_jiffy, one_jiffy_scaled);
342 } else {
343 __account_system_time(p, cputime_one_jiffy, one_jiffy_scaled,
344 CPUTIME_SYSTEM);
345 }
346}
347
348static void irqtime_account_idle_ticks(int ticks)
349{
350 int i;
351 struct rq *rq = this_rq();
352
353 for (i = 0; i < ticks; i++)
354 irqtime_account_process_tick(current, 0, rq);
355}
356#else /* CONFIG_IRQ_TIME_ACCOUNTING */
357static void irqtime_account_idle_ticks(int ticks) {}
358static void irqtime_account_process_tick(struct task_struct *p, int user_tick,
359 struct rq *rq) {}
360#endif /* CONFIG_IRQ_TIME_ACCOUNTING */
361
362/*
363 * Account a single tick of cpu time.
364 * @p: the process that the cpu time gets accounted to
365 * @user_tick: indicates if the tick is a user or a system tick
366 */
367void account_process_tick(struct task_struct *p, int user_tick)
368{
369 cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy);
370 struct rq *rq = this_rq();
371
372 if (sched_clock_irqtime) {
373 irqtime_account_process_tick(p, user_tick, rq);
374 return;
375 }
376
377 if (steal_account_process_tick())
378 return;
379
380 if (user_tick)
381 account_user_time(p, cputime_one_jiffy, one_jiffy_scaled);
382 else if ((p != rq->idle) || (irq_count() != HARDIRQ_OFFSET))
383 account_system_time(p, HARDIRQ_OFFSET, cputime_one_jiffy,
384 one_jiffy_scaled);
385 else
386 account_idle_time(cputime_one_jiffy);
387}
388
389/*
390 * Account multiple ticks of steal time.
391 * @p: the process from which the cpu time has been stolen
392 * @ticks: number of stolen ticks
393 */
394void account_steal_ticks(unsigned long ticks)
395{
396 account_steal_time(jiffies_to_cputime(ticks));
397}
398
399/*
400 * Account multiple ticks of idle time.
401 * @ticks: number of stolen ticks
402 */
403void account_idle_ticks(unsigned long ticks)
404{
405
406 if (sched_clock_irqtime) {
407 irqtime_account_idle_ticks(ticks);
408 return;
409 }
410
411 account_idle_time(jiffies_to_cputime(ticks));
412}
413
414#endif
415
416/*
417 * Use precise platform statistics if available:
418 */
419#ifdef CONFIG_VIRT_CPU_ACCOUNTING
420void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st)
421{
422 *ut = p->utime;
423 *st = p->stime;
424}
425
426void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st)
427{
428 struct task_cputime cputime;
429
430 thread_group_cputime(p, &cputime);
431
432 *ut = cputime.utime;
433 *st = cputime.stime;
434}
435
436/*
437 * Archs that account the whole time spent in the idle task
438 * (outside irq) as idle time can rely on this and just implement
439 * vtime_account_system() and vtime_account_idle(). Archs that
440 * have other meaning of the idle time (s390 only includes the
441 * time spent by the CPU when it's in low power mode) must override
442 * vtime_account().
443 */
444#ifndef __ARCH_HAS_VTIME_ACCOUNT
445void vtime_account(struct task_struct *tsk)
446{
447 unsigned long flags;
448
449 local_irq_save(flags);
450
451 if (in_interrupt() || !is_idle_task(tsk))
452 vtime_account_system(tsk);
453 else
454 vtime_account_idle(tsk);
455
456 local_irq_restore(flags);
457}
458EXPORT_SYMBOL_GPL(vtime_account);
459#endif /* __ARCH_HAS_VTIME_ACCOUNT */
460
461#else
462
463#ifndef nsecs_to_cputime
464# define nsecs_to_cputime(__nsecs) nsecs_to_jiffies(__nsecs)
465#endif
466
467static cputime_t scale_utime(cputime_t utime, cputime_t rtime, cputime_t total)
468{
469 u64 temp = (__force u64) rtime;
470
471 temp *= (__force u64) utime;
472
473 if (sizeof(cputime_t) == 4)
474 temp = div_u64(temp, (__force u32) total);
475 else
476 temp = div64_u64(temp, (__force u64) total);
477
478 return (__force cputime_t) temp;
479}
480
481void task_times(struct task_struct *p, cputime_t *ut, cputime_t *st)
482{
483 cputime_t rtime, utime = p->utime, total = utime + p->stime;
484
485 /*
486 * Use CFS's precise accounting:
487 */
488 rtime = nsecs_to_cputime(p->se.sum_exec_runtime);
489
490 if (total)
491 utime = scale_utime(utime, rtime, total);
492 else
493 utime = rtime;
494
495 /*
496 * Compare with previous values, to keep monotonicity:
497 */
498 p->prev_utime = max(p->prev_utime, utime);
499 p->prev_stime = max(p->prev_stime, rtime - p->prev_utime);
500
501 *ut = p->prev_utime;
502 *st = p->prev_stime;
503}
504
505/*
506 * Must be called with siglock held.
507 */
508void thread_group_times(struct task_struct *p, cputime_t *ut, cputime_t *st)
509{
510 struct signal_struct *sig = p->signal;
511 struct task_cputime cputime;
512 cputime_t rtime, utime, total;
513
514 thread_group_cputime(p, &cputime);
515
516 total = cputime.utime + cputime.stime;
517 rtime = nsecs_to_cputime(cputime.sum_exec_runtime);
518
519 if (total)
520 utime = scale_utime(cputime.utime, rtime, total);
521 else
522 utime = rtime;
523
524 sig->prev_utime = max(sig->prev_utime, utime);
525 sig->prev_stime = max(sig->prev_stime, rtime - sig->prev_utime);
526
527 *ut = sig->prev_utime;
528 *st = sig->prev_stime;
529}
530#endif
diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index 96e2b18b6283..6b800a14b990 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -597,7 +597,7 @@ calc_delta_fair(unsigned long delta, struct sched_entity *se)
597/* 597/*
598 * The idea is to set a period in which each task runs once. 598 * The idea is to set a period in which each task runs once.
599 * 599 *
600 * When there are too many tasks (sysctl_sched_nr_latency) we have to stretch 600 * When there are too many tasks (sched_nr_latency) we have to stretch
601 * this period because otherwise the slices get too small. 601 * this period because otherwise the slices get too small.
602 * 602 *
603 * p = (nr <= nl) ? l : l*nr/nl 603 * p = (nr <= nl) ? l : l*nr/nl
@@ -2700,7 +2700,6 @@ select_task_rq_fair(struct task_struct *p, int sd_flag, int wake_flags)
2700 int prev_cpu = task_cpu(p); 2700 int prev_cpu = task_cpu(p);
2701 int new_cpu = cpu; 2701 int new_cpu = cpu;
2702 int want_affine = 0; 2702 int want_affine = 0;
2703 int want_sd = 1;
2704 int sync = wake_flags & WF_SYNC; 2703 int sync = wake_flags & WF_SYNC;
2705 2704
2706 if (p->nr_cpus_allowed == 1) 2705 if (p->nr_cpus_allowed == 1)
@@ -2718,48 +2717,21 @@ select_task_rq_fair(struct task_struct *p, int sd_flag, int wake_flags)
2718 continue; 2717 continue;
2719 2718
2720 /* 2719 /*
2721 * If power savings logic is enabled for a domain, see if we
2722 * are not overloaded, if so, don't balance wider.
2723 */
2724 if (tmp->flags & (SD_PREFER_LOCAL)) {
2725 unsigned long power = 0;
2726 unsigned long nr_running = 0;
2727 unsigned long capacity;
2728 int i;
2729
2730 for_each_cpu(i, sched_domain_span(tmp)) {
2731 power += power_of(i);
2732 nr_running += cpu_rq(i)->cfs.nr_running;
2733 }
2734
2735 capacity = DIV_ROUND_CLOSEST(power, SCHED_POWER_SCALE);
2736
2737 if (nr_running < capacity)
2738 want_sd = 0;
2739 }
2740
2741 /*
2742 * If both cpu and prev_cpu are part of this domain, 2720 * If both cpu and prev_cpu are part of this domain,
2743 * cpu is a valid SD_WAKE_AFFINE target. 2721 * cpu is a valid SD_WAKE_AFFINE target.
2744 */ 2722 */
2745 if (want_affine && (tmp->flags & SD_WAKE_AFFINE) && 2723 if (want_affine && (tmp->flags & SD_WAKE_AFFINE) &&
2746 cpumask_test_cpu(prev_cpu, sched_domain_span(tmp))) { 2724 cpumask_test_cpu(prev_cpu, sched_domain_span(tmp))) {
2747 affine_sd = tmp; 2725 affine_sd = tmp;
2748 want_affine = 0;
2749 }
2750
2751 if (!want_sd && !want_affine)
2752 break; 2726 break;
2727 }
2753 2728
2754 if (!(tmp->flags & sd_flag)) 2729 if (tmp->flags & sd_flag)
2755 continue;
2756
2757 if (want_sd)
2758 sd = tmp; 2730 sd = tmp;
2759 } 2731 }
2760 2732
2761 if (affine_sd) { 2733 if (affine_sd) {
2762 if (cpu == prev_cpu || wake_affine(affine_sd, p, sync)) 2734 if (cpu != prev_cpu && wake_affine(affine_sd, p, sync))
2763 prev_cpu = cpu; 2735 prev_cpu = cpu;
2764 2736
2765 new_cpu = select_idle_sibling(p, prev_cpu); 2737 new_cpu = select_idle_sibling(p, prev_cpu);
@@ -4295,7 +4267,7 @@ redo:
4295 goto out_balanced; 4267 goto out_balanced;
4296 } 4268 }
4297 4269
4298 BUG_ON(busiest == this_rq); 4270 BUG_ON(busiest == env.dst_rq);
4299 4271
4300 schedstat_add(sd, lb_imbalance[idle], env.imbalance); 4272 schedstat_add(sd, lb_imbalance[idle], env.imbalance);
4301 4273
@@ -4316,7 +4288,7 @@ redo:
4316 update_h_load(env.src_cpu); 4288 update_h_load(env.src_cpu);
4317more_balance: 4289more_balance:
4318 local_irq_save(flags); 4290 local_irq_save(flags);
4319 double_rq_lock(this_rq, busiest); 4291 double_rq_lock(env.dst_rq, busiest);
4320 4292
4321 /* 4293 /*
4322 * cur_ld_moved - load moved in current iteration 4294 * cur_ld_moved - load moved in current iteration
@@ -4324,7 +4296,7 @@ more_balance:
4324 */ 4296 */
4325 cur_ld_moved = move_tasks(&env); 4297 cur_ld_moved = move_tasks(&env);
4326 ld_moved += cur_ld_moved; 4298 ld_moved += cur_ld_moved;
4327 double_rq_unlock(this_rq, busiest); 4299 double_rq_unlock(env.dst_rq, busiest);
4328 local_irq_restore(flags); 4300 local_irq_restore(flags);
4329 4301
4330 if (env.flags & LBF_NEED_BREAK) { 4302 if (env.flags & LBF_NEED_BREAK) {
@@ -4360,8 +4332,7 @@ more_balance:
4360 if ((env.flags & LBF_SOME_PINNED) && env.imbalance > 0 && 4332 if ((env.flags & LBF_SOME_PINNED) && env.imbalance > 0 &&
4361 lb_iterations++ < max_lb_iterations) { 4333 lb_iterations++ < max_lb_iterations) {
4362 4334
4363 this_rq = cpu_rq(env.new_dst_cpu); 4335 env.dst_rq = cpu_rq(env.new_dst_cpu);
4364 env.dst_rq = this_rq;
4365 env.dst_cpu = env.new_dst_cpu; 4336 env.dst_cpu = env.new_dst_cpu;
4366 env.flags &= ~LBF_SOME_PINNED; 4337 env.flags &= ~LBF_SOME_PINNED;
4367 env.loop = 0; 4338 env.loop = 0;
@@ -4646,7 +4617,7 @@ static void nohz_balancer_kick(int cpu)
4646 return; 4617 return;
4647} 4618}
4648 4619
4649static inline void clear_nohz_tick_stopped(int cpu) 4620static inline void nohz_balance_exit_idle(int cpu)
4650{ 4621{
4651 if (unlikely(test_bit(NOHZ_TICK_STOPPED, nohz_flags(cpu)))) { 4622 if (unlikely(test_bit(NOHZ_TICK_STOPPED, nohz_flags(cpu)))) {
4652 cpumask_clear_cpu(cpu, nohz.idle_cpus_mask); 4623 cpumask_clear_cpu(cpu, nohz.idle_cpus_mask);
@@ -4686,28 +4657,23 @@ void set_cpu_sd_state_idle(void)
4686} 4657}
4687 4658
4688/* 4659/*
4689 * This routine will record that this cpu is going idle with tick stopped. 4660 * This routine will record that the cpu is going idle with tick stopped.
4690 * This info will be used in performing idle load balancing in the future. 4661 * This info will be used in performing idle load balancing in the future.
4691 */ 4662 */
4692void select_nohz_load_balancer(int stop_tick) 4663void nohz_balance_enter_idle(int cpu)
4693{ 4664{
4694 int cpu = smp_processor_id();
4695
4696 /* 4665 /*
4697 * If this cpu is going down, then nothing needs to be done. 4666 * If this cpu is going down, then nothing needs to be done.
4698 */ 4667 */
4699 if (!cpu_active(cpu)) 4668 if (!cpu_active(cpu))
4700 return; 4669 return;
4701 4670
4702 if (stop_tick) { 4671 if (test_bit(NOHZ_TICK_STOPPED, nohz_flags(cpu)))
4703 if (test_bit(NOHZ_TICK_STOPPED, nohz_flags(cpu))) 4672 return;
4704 return;
4705 4673
4706 cpumask_set_cpu(cpu, nohz.idle_cpus_mask); 4674 cpumask_set_cpu(cpu, nohz.idle_cpus_mask);
4707 atomic_inc(&nohz.nr_cpus); 4675 atomic_inc(&nohz.nr_cpus);
4708 set_bit(NOHZ_TICK_STOPPED, nohz_flags(cpu)); 4676 set_bit(NOHZ_TICK_STOPPED, nohz_flags(cpu));
4709 }
4710 return;
4711} 4677}
4712 4678
4713static int __cpuinit sched_ilb_notifier(struct notifier_block *nfb, 4679static int __cpuinit sched_ilb_notifier(struct notifier_block *nfb,
@@ -4715,7 +4681,7 @@ static int __cpuinit sched_ilb_notifier(struct notifier_block *nfb,
4715{ 4681{
4716 switch (action & ~CPU_TASKS_FROZEN) { 4682 switch (action & ~CPU_TASKS_FROZEN) {
4717 case CPU_DYING: 4683 case CPU_DYING:
4718 clear_nohz_tick_stopped(smp_processor_id()); 4684 nohz_balance_exit_idle(smp_processor_id());
4719 return NOTIFY_OK; 4685 return NOTIFY_OK;
4720 default: 4686 default:
4721 return NOTIFY_DONE; 4687 return NOTIFY_DONE;
@@ -4837,14 +4803,15 @@ static void nohz_idle_balance(int this_cpu, enum cpu_idle_type idle)
4837 if (need_resched()) 4803 if (need_resched())
4838 break; 4804 break;
4839 4805
4840 raw_spin_lock_irq(&this_rq->lock); 4806 rq = cpu_rq(balance_cpu);
4841 update_rq_clock(this_rq); 4807
4842 update_idle_cpu_load(this_rq); 4808 raw_spin_lock_irq(&rq->lock);
4843 raw_spin_unlock_irq(&this_rq->lock); 4809 update_rq_clock(rq);
4810 update_idle_cpu_load(rq);
4811 raw_spin_unlock_irq(&rq->lock);
4844 4812
4845 rebalance_domains(balance_cpu, CPU_IDLE); 4813 rebalance_domains(balance_cpu, CPU_IDLE);
4846 4814
4847 rq = cpu_rq(balance_cpu);
4848 if (time_after(this_rq->next_balance, rq->next_balance)) 4815 if (time_after(this_rq->next_balance, rq->next_balance))
4849 this_rq->next_balance = rq->next_balance; 4816 this_rq->next_balance = rq->next_balance;
4850 } 4817 }
@@ -4875,7 +4842,7 @@ static inline int nohz_kick_needed(struct rq *rq, int cpu)
4875 * busy tick after returning from idle, we will update the busy stats. 4842 * busy tick after returning from idle, we will update the busy stats.
4876 */ 4843 */
4877 set_cpu_sd_state_busy(); 4844 set_cpu_sd_state_busy();
4878 clear_nohz_tick_stopped(cpu); 4845 nohz_balance_exit_idle(cpu);
4879 4846
4880 /* 4847 /*
4881 * None are in tickless mode and hence no need for NOHZ idle load 4848 * None are in tickless mode and hence no need for NOHZ idle load
diff --git a/kernel/sched/features.h b/kernel/sched/features.h
index de00a486c5c6..eebefcad7027 100644
--- a/kernel/sched/features.h
+++ b/kernel/sched/features.h
@@ -12,14 +12,6 @@ SCHED_FEAT(GENTLE_FAIR_SLEEPERS, true)
12SCHED_FEAT(START_DEBIT, true) 12SCHED_FEAT(START_DEBIT, true)
13 13
14/* 14/*
15 * Based on load and program behaviour, see if it makes sense to place
16 * a newly woken task on the same cpu as the task that woke it --
17 * improve cache locality. Typically used with SYNC wakeups as
18 * generated by pipes and the like, see also SYNC_WAKEUPS.
19 */
20SCHED_FEAT(AFFINE_WAKEUPS, true)
21
22/*
23 * Prefer to schedule the task we woke last (assuming it failed 15 * Prefer to schedule the task we woke last (assuming it failed
24 * wakeup-preemption), since its likely going to consume data we 16 * wakeup-preemption), since its likely going to consume data we
25 * touched, increases cache locality. 17 * touched, increases cache locality.
@@ -42,7 +34,7 @@ SCHED_FEAT(CACHE_HOT_BUDDY, true)
42/* 34/*
43 * Use arch dependent cpu power functions 35 * Use arch dependent cpu power functions
44 */ 36 */
45SCHED_FEAT(ARCH_POWER, false) 37SCHED_FEAT(ARCH_POWER, true)
46 38
47SCHED_FEAT(HRTICK, false) 39SCHED_FEAT(HRTICK, false)
48SCHED_FEAT(DOUBLE_TICK, false) 40SCHED_FEAT(DOUBLE_TICK, false)
diff --git a/kernel/sched/rt.c b/kernel/sched/rt.c
index e0b7ba9c040f..418feb01344e 100644
--- a/kernel/sched/rt.c
+++ b/kernel/sched/rt.c
@@ -1632,11 +1632,6 @@ static int push_rt_task(struct rq *rq)
1632 if (!next_task) 1632 if (!next_task)
1633 return 0; 1633 return 0;
1634 1634
1635#ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
1636 if (unlikely(task_running(rq, next_task)))
1637 return 0;
1638#endif
1639
1640retry: 1635retry:
1641 if (unlikely(next_task == rq->curr)) { 1636 if (unlikely(next_task == rq->curr)) {
1642 WARN_ON(1); 1637 WARN_ON(1);
diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h
index 0848fa36c383..7a7db09cfabc 100644
--- a/kernel/sched/sched.h
+++ b/kernel/sched/sched.h
@@ -737,11 +737,7 @@ static inline void prepare_lock_switch(struct rq *rq, struct task_struct *next)
737 */ 737 */
738 next->on_cpu = 1; 738 next->on_cpu = 1;
739#endif 739#endif
740#ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
741 raw_spin_unlock_irq(&rq->lock);
742#else
743 raw_spin_unlock(&rq->lock); 740 raw_spin_unlock(&rq->lock);
744#endif
745} 741}
746 742
747static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev) 743static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev)
@@ -755,9 +751,7 @@ static inline void finish_lock_switch(struct rq *rq, struct task_struct *prev)
755 smp_wmb(); 751 smp_wmb();
756 prev->on_cpu = 0; 752 prev->on_cpu = 0;
757#endif 753#endif
758#ifndef __ARCH_WANT_INTERRUPTS_ON_CTXSW
759 local_irq_enable(); 754 local_irq_enable();
760#endif
761} 755}
762#endif /* __ARCH_WANT_UNLOCKED_CTXSW */ 756#endif /* __ARCH_WANT_UNLOCKED_CTXSW */
763 757
@@ -891,6 +885,9 @@ struct cpuacct {
891 struct kernel_cpustat __percpu *cpustat; 885 struct kernel_cpustat __percpu *cpustat;
892}; 886};
893 887
888extern struct cgroup_subsys cpuacct_subsys;
889extern struct cpuacct root_cpuacct;
890
894/* return cpu accounting group corresponding to this container */ 891/* return cpu accounting group corresponding to this container */
895static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp) 892static inline struct cpuacct *cgroup_ca(struct cgroup *cgrp)
896{ 893{
@@ -917,6 +914,16 @@ extern void cpuacct_charge(struct task_struct *tsk, u64 cputime);
917static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {} 914static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {}
918#endif 915#endif
919 916
917#ifdef CONFIG_PARAVIRT
918static inline u64 steal_ticks(u64 steal)
919{
920 if (unlikely(steal > NSEC_PER_SEC))
921 return div_u64(steal, TICK_NSEC);
922
923 return __iter_div_u64_rem(steal, TICK_NSEC, &steal);
924}
925#endif
926
920static inline void inc_nr_running(struct rq *rq) 927static inline void inc_nr_running(struct rq *rq)
921{ 928{
922 rq->nr_running++; 929 rq->nr_running++;
@@ -1156,3 +1163,53 @@ enum rq_nohz_flag_bits {
1156 1163
1157#define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags) 1164#define nohz_flags(cpu) (&cpu_rq(cpu)->nohz_flags)
1158#endif 1165#endif
1166
1167#ifdef CONFIG_IRQ_TIME_ACCOUNTING
1168
1169DECLARE_PER_CPU(u64, cpu_hardirq_time);
1170DECLARE_PER_CPU(u64, cpu_softirq_time);
1171
1172#ifndef CONFIG_64BIT
1173DECLARE_PER_CPU(seqcount_t, irq_time_seq);
1174
1175static inline void irq_time_write_begin(void)
1176{
1177 __this_cpu_inc(irq_time_seq.sequence);
1178 smp_wmb();
1179}
1180
1181static inline void irq_time_write_end(void)
1182{
1183 smp_wmb();
1184 __this_cpu_inc(irq_time_seq.sequence);
1185}
1186
1187static inline u64 irq_time_read(int cpu)
1188{
1189 u64 irq_time;
1190 unsigned seq;
1191
1192 do {
1193 seq = read_seqcount_begin(&per_cpu(irq_time_seq, cpu));
1194 irq_time = per_cpu(cpu_softirq_time, cpu) +
1195 per_cpu(cpu_hardirq_time, cpu);
1196 } while (read_seqcount_retry(&per_cpu(irq_time_seq, cpu), seq));
1197
1198 return irq_time;
1199}
1200#else /* CONFIG_64BIT */
1201static inline void irq_time_write_begin(void)
1202{
1203}
1204
1205static inline void irq_time_write_end(void)
1206{
1207}
1208
1209static inline u64 irq_time_read(int cpu)
1210{
1211 return per_cpu(cpu_softirq_time, cpu) + per_cpu(cpu_hardirq_time, cpu);
1212}
1213#endif /* CONFIG_64BIT */
1214#endif /* CONFIG_IRQ_TIME_ACCOUNTING */
1215
diff --git a/kernel/signal.c b/kernel/signal.c
index be4f856d52f8..0af8868525d6 100644
--- a/kernel/signal.c
+++ b/kernel/signal.c
@@ -17,6 +17,7 @@
17#include <linux/fs.h> 17#include <linux/fs.h>
18#include <linux/tty.h> 18#include <linux/tty.h>
19#include <linux/binfmts.h> 19#include <linux/binfmts.h>
20#include <linux/coredump.h>
20#include <linux/security.h> 21#include <linux/security.h>
21#include <linux/syscalls.h> 22#include <linux/syscalls.h>
22#include <linux/ptrace.h> 23#include <linux/ptrace.h>
@@ -1971,13 +1972,8 @@ static void ptrace_do_notify(int signr, int exit_code, int why)
1971void ptrace_notify(int exit_code) 1972void ptrace_notify(int exit_code)
1972{ 1973{
1973 BUG_ON((exit_code & (0x7f | ~0xffff)) != SIGTRAP); 1974 BUG_ON((exit_code & (0x7f | ~0xffff)) != SIGTRAP);
1974 if (unlikely(current->task_works)) { 1975 if (unlikely(current->task_works))
1975 if (test_and_clear_ti_thread_flag(current_thread_info(), 1976 task_work_run();
1976 TIF_NOTIFY_RESUME)) {
1977 smp_mb__after_clear_bit();
1978 task_work_run();
1979 }
1980 }
1981 1977
1982 spin_lock_irq(&current->sighand->siglock); 1978 spin_lock_irq(&current->sighand->siglock);
1983 ptrace_do_notify(SIGTRAP, exit_code, CLD_TRAPPED); 1979 ptrace_do_notify(SIGTRAP, exit_code, CLD_TRAPPED);
@@ -2198,13 +2194,8 @@ int get_signal_to_deliver(siginfo_t *info, struct k_sigaction *return_ka,
2198 struct signal_struct *signal = current->signal; 2194 struct signal_struct *signal = current->signal;
2199 int signr; 2195 int signr;
2200 2196
2201 if (unlikely(current->task_works)) { 2197 if (unlikely(current->task_works))
2202 if (test_and_clear_ti_thread_flag(current_thread_info(), 2198 task_work_run();
2203 TIF_NOTIFY_RESUME)) {
2204 smp_mb__after_clear_bit();
2205 task_work_run();
2206 }
2207 }
2208 2199
2209 if (unlikely(uprobe_deny_signal())) 2200 if (unlikely(uprobe_deny_signal()))
2210 return 0; 2201 return 0;
@@ -2369,7 +2360,7 @@ relock:
2369 * first and our do_group_exit call below will use 2360 * first and our do_group_exit call below will use
2370 * that value and ignore the one we pass it. 2361 * that value and ignore the one we pass it.
2371 */ 2362 */
2372 do_coredump(info->si_signo, info->si_signo, regs); 2363 do_coredump(info, regs);
2373 } 2364 }
2374 2365
2375 /* 2366 /*
diff --git a/kernel/smpboot.c b/kernel/smpboot.c
index 98f60c5caa1b..d6c5fc054242 100644
--- a/kernel/smpboot.c
+++ b/kernel/smpboot.c
@@ -1,14 +1,22 @@
1/* 1/*
2 * Common SMP CPU bringup/teardown functions 2 * Common SMP CPU bringup/teardown functions
3 */ 3 */
4#include <linux/cpu.h>
4#include <linux/err.h> 5#include <linux/err.h>
5#include <linux/smp.h> 6#include <linux/smp.h>
6#include <linux/init.h> 7#include <linux/init.h>
8#include <linux/list.h>
9#include <linux/slab.h>
7#include <linux/sched.h> 10#include <linux/sched.h>
11#include <linux/export.h>
8#include <linux/percpu.h> 12#include <linux/percpu.h>
13#include <linux/kthread.h>
14#include <linux/smpboot.h>
9 15
10#include "smpboot.h" 16#include "smpboot.h"
11 17
18#ifdef CONFIG_SMP
19
12#ifdef CONFIG_GENERIC_SMP_IDLE_THREAD 20#ifdef CONFIG_GENERIC_SMP_IDLE_THREAD
13/* 21/*
14 * For the hotplug case we keep the task structs around and reuse 22 * For the hotplug case we keep the task structs around and reuse
@@ -65,3 +73,228 @@ void __init idle_threads_init(void)
65 } 73 }
66} 74}
67#endif 75#endif
76
77#endif /* #ifdef CONFIG_SMP */
78
79static LIST_HEAD(hotplug_threads);
80static DEFINE_MUTEX(smpboot_threads_lock);
81
82struct smpboot_thread_data {
83 unsigned int cpu;
84 unsigned int status;
85 struct smp_hotplug_thread *ht;
86};
87
88enum {
89 HP_THREAD_NONE = 0,
90 HP_THREAD_ACTIVE,
91 HP_THREAD_PARKED,
92};
93
94/**
95 * smpboot_thread_fn - percpu hotplug thread loop function
96 * @data: thread data pointer
97 *
98 * Checks for thread stop and park conditions. Calls the necessary
99 * setup, cleanup, park and unpark functions for the registered
100 * thread.
101 *
102 * Returns 1 when the thread should exit, 0 otherwise.
103 */
104static int smpboot_thread_fn(void *data)
105{
106 struct smpboot_thread_data *td = data;
107 struct smp_hotplug_thread *ht = td->ht;
108
109 while (1) {
110 set_current_state(TASK_INTERRUPTIBLE);
111 preempt_disable();
112 if (kthread_should_stop()) {
113 set_current_state(TASK_RUNNING);
114 preempt_enable();
115 if (ht->cleanup)
116 ht->cleanup(td->cpu, cpu_online(td->cpu));
117 kfree(td);
118 return 0;
119 }
120
121 if (kthread_should_park()) {
122 __set_current_state(TASK_RUNNING);
123 preempt_enable();
124 if (ht->park && td->status == HP_THREAD_ACTIVE) {
125 BUG_ON(td->cpu != smp_processor_id());
126 ht->park(td->cpu);
127 td->status = HP_THREAD_PARKED;
128 }
129 kthread_parkme();
130 /* We might have been woken for stop */
131 continue;
132 }
133
134 BUG_ON(td->cpu != smp_processor_id());
135
136 /* Check for state change setup */
137 switch (td->status) {
138 case HP_THREAD_NONE:
139 preempt_enable();
140 if (ht->setup)
141 ht->setup(td->cpu);
142 td->status = HP_THREAD_ACTIVE;
143 preempt_disable();
144 break;
145 case HP_THREAD_PARKED:
146 preempt_enable();
147 if (ht->unpark)
148 ht->unpark(td->cpu);
149 td->status = HP_THREAD_ACTIVE;
150 preempt_disable();
151 break;
152 }
153
154 if (!ht->thread_should_run(td->cpu)) {
155 preempt_enable();
156 schedule();
157 } else {
158 set_current_state(TASK_RUNNING);
159 preempt_enable();
160 ht->thread_fn(td->cpu);
161 }
162 }
163}
164
165static int
166__smpboot_create_thread(struct smp_hotplug_thread *ht, unsigned int cpu)
167{
168 struct task_struct *tsk = *per_cpu_ptr(ht->store, cpu);
169 struct smpboot_thread_data *td;
170
171 if (tsk)
172 return 0;
173
174 td = kzalloc_node(sizeof(*td), GFP_KERNEL, cpu_to_node(cpu));
175 if (!td)
176 return -ENOMEM;
177 td->cpu = cpu;
178 td->ht = ht;
179
180 tsk = kthread_create_on_cpu(smpboot_thread_fn, td, cpu,
181 ht->thread_comm);
182 if (IS_ERR(tsk)) {
183 kfree(td);
184 return PTR_ERR(tsk);
185 }
186
187 get_task_struct(tsk);
188 *per_cpu_ptr(ht->store, cpu) = tsk;
189 return 0;
190}
191
192int smpboot_create_threads(unsigned int cpu)
193{
194 struct smp_hotplug_thread *cur;
195 int ret = 0;
196
197 mutex_lock(&smpboot_threads_lock);
198 list_for_each_entry(cur, &hotplug_threads, list) {
199 ret = __smpboot_create_thread(cur, cpu);
200 if (ret)
201 break;
202 }
203 mutex_unlock(&smpboot_threads_lock);
204 return ret;
205}
206
207static void smpboot_unpark_thread(struct smp_hotplug_thread *ht, unsigned int cpu)
208{
209 struct task_struct *tsk = *per_cpu_ptr(ht->store, cpu);
210
211 kthread_unpark(tsk);
212}
213
214void smpboot_unpark_threads(unsigned int cpu)
215{
216 struct smp_hotplug_thread *cur;
217
218 mutex_lock(&smpboot_threads_lock);
219 list_for_each_entry(cur, &hotplug_threads, list)
220 smpboot_unpark_thread(cur, cpu);
221 mutex_unlock(&smpboot_threads_lock);
222}
223
224static void smpboot_park_thread(struct smp_hotplug_thread *ht, unsigned int cpu)
225{
226 struct task_struct *tsk = *per_cpu_ptr(ht->store, cpu);
227
228 if (tsk)
229 kthread_park(tsk);
230}
231
232void smpboot_park_threads(unsigned int cpu)
233{
234 struct smp_hotplug_thread *cur;
235
236 mutex_lock(&smpboot_threads_lock);
237 list_for_each_entry_reverse(cur, &hotplug_threads, list)
238 smpboot_park_thread(cur, cpu);
239 mutex_unlock(&smpboot_threads_lock);
240}
241
242static void smpboot_destroy_threads(struct smp_hotplug_thread *ht)
243{
244 unsigned int cpu;
245
246 /* We need to destroy also the parked threads of offline cpus */
247 for_each_possible_cpu(cpu) {
248 struct task_struct *tsk = *per_cpu_ptr(ht->store, cpu);
249
250 if (tsk) {
251 kthread_stop(tsk);
252 put_task_struct(tsk);
253 *per_cpu_ptr(ht->store, cpu) = NULL;
254 }
255 }
256}
257
258/**
259 * smpboot_register_percpu_thread - Register a per_cpu thread related to hotplug
260 * @plug_thread: Hotplug thread descriptor
261 *
262 * Creates and starts the threads on all online cpus.
263 */
264int smpboot_register_percpu_thread(struct smp_hotplug_thread *plug_thread)
265{
266 unsigned int cpu;
267 int ret = 0;
268
269 mutex_lock(&smpboot_threads_lock);
270 for_each_online_cpu(cpu) {
271 ret = __smpboot_create_thread(plug_thread, cpu);
272 if (ret) {
273 smpboot_destroy_threads(plug_thread);
274 goto out;
275 }
276 smpboot_unpark_thread(plug_thread, cpu);
277 }
278 list_add(&plug_thread->list, &hotplug_threads);
279out:
280 mutex_unlock(&smpboot_threads_lock);
281 return ret;
282}
283EXPORT_SYMBOL_GPL(smpboot_register_percpu_thread);
284
285/**
286 * smpboot_unregister_percpu_thread - Unregister a per_cpu thread related to hotplug
287 * @plug_thread: Hotplug thread descriptor
288 *
289 * Stops all threads on all possible cpus.
290 */
291void smpboot_unregister_percpu_thread(struct smp_hotplug_thread *plug_thread)
292{
293 get_online_cpus();
294 mutex_lock(&smpboot_threads_lock);
295 list_del(&plug_thread->list);
296 smpboot_destroy_threads(plug_thread);
297 mutex_unlock(&smpboot_threads_lock);
298 put_online_cpus();
299}
300EXPORT_SYMBOL_GPL(smpboot_unregister_percpu_thread);
diff --git a/kernel/smpboot.h b/kernel/smpboot.h
index 6ef9433e1c70..72415a0eb955 100644
--- a/kernel/smpboot.h
+++ b/kernel/smpboot.h
@@ -13,4 +13,8 @@ static inline void idle_thread_set_boot_cpu(void) { }
13static inline void idle_threads_init(void) { } 13static inline void idle_threads_init(void) { }
14#endif 14#endif
15 15
16int smpboot_create_threads(unsigned int cpu);
17void smpboot_park_threads(unsigned int cpu);
18void smpboot_unpark_threads(unsigned int cpu);
19
16#endif 20#endif
diff --git a/kernel/softirq.c b/kernel/softirq.c
index b73e681df09e..cc96bdc0c2c9 100644
--- a/kernel/softirq.c
+++ b/kernel/softirq.c
@@ -23,6 +23,7 @@
23#include <linux/rcupdate.h> 23#include <linux/rcupdate.h>
24#include <linux/ftrace.h> 24#include <linux/ftrace.h>
25#include <linux/smp.h> 25#include <linux/smp.h>
26#include <linux/smpboot.h>
26#include <linux/tick.h> 27#include <linux/tick.h>
27 28
28#define CREATE_TRACE_POINTS 29#define CREATE_TRACE_POINTS
@@ -220,7 +221,7 @@ asmlinkage void __do_softirq(void)
220 current->flags &= ~PF_MEMALLOC; 221 current->flags &= ~PF_MEMALLOC;
221 222
222 pending = local_softirq_pending(); 223 pending = local_softirq_pending();
223 account_system_vtime(current); 224 vtime_account(current);
224 225
225 __local_bh_disable((unsigned long)__builtin_return_address(0), 226 __local_bh_disable((unsigned long)__builtin_return_address(0),
226 SOFTIRQ_OFFSET); 227 SOFTIRQ_OFFSET);
@@ -271,7 +272,7 @@ restart:
271 272
272 lockdep_softirq_exit(); 273 lockdep_softirq_exit();
273 274
274 account_system_vtime(current); 275 vtime_account(current);
275 __local_bh_enable(SOFTIRQ_OFFSET); 276 __local_bh_enable(SOFTIRQ_OFFSET);
276 tsk_restore_flags(current, old_flags, PF_MEMALLOC); 277 tsk_restore_flags(current, old_flags, PF_MEMALLOC);
277} 278}
@@ -340,7 +341,7 @@ static inline void invoke_softirq(void)
340 */ 341 */
341void irq_exit(void) 342void irq_exit(void)
342{ 343{
343 account_system_vtime(current); 344 vtime_account(current);
344 trace_hardirq_exit(); 345 trace_hardirq_exit();
345 sub_preempt_count(IRQ_EXIT_OFFSET); 346 sub_preempt_count(IRQ_EXIT_OFFSET);
346 if (!in_interrupt() && local_softirq_pending()) 347 if (!in_interrupt() && local_softirq_pending())
@@ -742,49 +743,22 @@ void __init softirq_init(void)
742 open_softirq(HI_SOFTIRQ, tasklet_hi_action); 743 open_softirq(HI_SOFTIRQ, tasklet_hi_action);
743} 744}
744 745
745static int run_ksoftirqd(void * __bind_cpu) 746static int ksoftirqd_should_run(unsigned int cpu)
746{ 747{
747 set_current_state(TASK_INTERRUPTIBLE); 748 return local_softirq_pending();
748 749}
749 while (!kthread_should_stop()) {
750 preempt_disable();
751 if (!local_softirq_pending()) {
752 schedule_preempt_disabled();
753 }
754
755 __set_current_state(TASK_RUNNING);
756
757 while (local_softirq_pending()) {
758 /* Preempt disable stops cpu going offline.
759 If already offline, we'll be on wrong CPU:
760 don't process */
761 if (cpu_is_offline((long)__bind_cpu))
762 goto wait_to_die;
763 local_irq_disable();
764 if (local_softirq_pending())
765 __do_softirq();
766 local_irq_enable();
767 sched_preempt_enable_no_resched();
768 cond_resched();
769 preempt_disable();
770 rcu_note_context_switch((long)__bind_cpu);
771 }
772 preempt_enable();
773 set_current_state(TASK_INTERRUPTIBLE);
774 }
775 __set_current_state(TASK_RUNNING);
776 return 0;
777 750
778wait_to_die: 751static void run_ksoftirqd(unsigned int cpu)
779 preempt_enable(); 752{
780 /* Wait for kthread_stop */ 753 local_irq_disable();
781 set_current_state(TASK_INTERRUPTIBLE); 754 if (local_softirq_pending()) {
782 while (!kthread_should_stop()) { 755 __do_softirq();
783 schedule(); 756 rcu_note_context_switch(cpu);
784 set_current_state(TASK_INTERRUPTIBLE); 757 local_irq_enable();
758 cond_resched();
759 return;
785 } 760 }
786 __set_current_state(TASK_RUNNING); 761 local_irq_enable();
787 return 0;
788} 762}
789 763
790#ifdef CONFIG_HOTPLUG_CPU 764#ifdef CONFIG_HOTPLUG_CPU
@@ -850,50 +824,14 @@ static int __cpuinit cpu_callback(struct notifier_block *nfb,
850 unsigned long action, 824 unsigned long action,
851 void *hcpu) 825 void *hcpu)
852{ 826{
853 int hotcpu = (unsigned long)hcpu;
854 struct task_struct *p;
855
856 switch (action) { 827 switch (action) {
857 case CPU_UP_PREPARE:
858 case CPU_UP_PREPARE_FROZEN:
859 p = kthread_create_on_node(run_ksoftirqd,
860 hcpu,
861 cpu_to_node(hotcpu),
862 "ksoftirqd/%d", hotcpu);
863 if (IS_ERR(p)) {
864 printk("ksoftirqd for %i failed\n", hotcpu);
865 return notifier_from_errno(PTR_ERR(p));
866 }
867 kthread_bind(p, hotcpu);
868 per_cpu(ksoftirqd, hotcpu) = p;
869 break;
870 case CPU_ONLINE:
871 case CPU_ONLINE_FROZEN:
872 wake_up_process(per_cpu(ksoftirqd, hotcpu));
873 break;
874#ifdef CONFIG_HOTPLUG_CPU 828#ifdef CONFIG_HOTPLUG_CPU
875 case CPU_UP_CANCELED:
876 case CPU_UP_CANCELED_FROZEN:
877 if (!per_cpu(ksoftirqd, hotcpu))
878 break;
879 /* Unbind so it can run. Fall thru. */
880 kthread_bind(per_cpu(ksoftirqd, hotcpu),
881 cpumask_any(cpu_online_mask));
882 case CPU_DEAD: 829 case CPU_DEAD:
883 case CPU_DEAD_FROZEN: { 830 case CPU_DEAD_FROZEN:
884 static const struct sched_param param = { 831 takeover_tasklets((unsigned long)hcpu);
885 .sched_priority = MAX_RT_PRIO-1
886 };
887
888 p = per_cpu(ksoftirqd, hotcpu);
889 per_cpu(ksoftirqd, hotcpu) = NULL;
890 sched_setscheduler_nocheck(p, SCHED_FIFO, &param);
891 kthread_stop(p);
892 takeover_tasklets(hotcpu);
893 break; 832 break;
894 }
895#endif /* CONFIG_HOTPLUG_CPU */ 833#endif /* CONFIG_HOTPLUG_CPU */
896 } 834 }
897 return NOTIFY_OK; 835 return NOTIFY_OK;
898} 836}
899 837
@@ -901,14 +839,19 @@ static struct notifier_block __cpuinitdata cpu_nfb = {
901 .notifier_call = cpu_callback 839 .notifier_call = cpu_callback
902}; 840};
903 841
842static struct smp_hotplug_thread softirq_threads = {
843 .store = &ksoftirqd,
844 .thread_should_run = ksoftirqd_should_run,
845 .thread_fn = run_ksoftirqd,
846 .thread_comm = "ksoftirqd/%u",
847};
848
904static __init int spawn_ksoftirqd(void) 849static __init int spawn_ksoftirqd(void)
905{ 850{
906 void *cpu = (void *)(long)smp_processor_id();
907 int err = cpu_callback(&cpu_nfb, CPU_UP_PREPARE, cpu);
908
909 BUG_ON(err != NOTIFY_OK);
910 cpu_callback(&cpu_nfb, CPU_ONLINE, cpu);
911 register_cpu_notifier(&cpu_nfb); 851 register_cpu_notifier(&cpu_nfb);
852
853 BUG_ON(smpboot_register_percpu_thread(&softirq_threads));
854
912 return 0; 855 return 0;
913} 856}
914early_initcall(spawn_ksoftirqd); 857early_initcall(spawn_ksoftirqd);
diff --git a/kernel/srcu.c b/kernel/srcu.c
index 2095be3318d5..97c465ebd844 100644
--- a/kernel/srcu.c
+++ b/kernel/srcu.c
@@ -379,7 +379,7 @@ void call_srcu(struct srcu_struct *sp, struct rcu_head *head,
379 rcu_batch_queue(&sp->batch_queue, head); 379 rcu_batch_queue(&sp->batch_queue, head);
380 if (!sp->running) { 380 if (!sp->running) {
381 sp->running = true; 381 sp->running = true;
382 queue_delayed_work(system_nrt_wq, &sp->work, 0); 382 schedule_delayed_work(&sp->work, 0);
383 } 383 }
384 spin_unlock_irqrestore(&sp->queue_lock, flags); 384 spin_unlock_irqrestore(&sp->queue_lock, flags);
385} 385}
@@ -631,7 +631,7 @@ static void srcu_reschedule(struct srcu_struct *sp)
631 } 631 }
632 632
633 if (pending) 633 if (pending)
634 queue_delayed_work(system_nrt_wq, &sp->work, SRCU_INTERVAL); 634 schedule_delayed_work(&sp->work, SRCU_INTERVAL);
635} 635}
636 636
637/* 637/*
diff --git a/kernel/sys.c b/kernel/sys.c
index 241507f23eca..c5cb5b99cb81 100644
--- a/kernel/sys.c
+++ b/kernel/sys.c
@@ -368,6 +368,7 @@ EXPORT_SYMBOL(unregister_reboot_notifier);
368void kernel_restart(char *cmd) 368void kernel_restart(char *cmd)
369{ 369{
370 kernel_restart_prepare(cmd); 370 kernel_restart_prepare(cmd);
371 disable_nonboot_cpus();
371 if (!cmd) 372 if (!cmd)
372 printk(KERN_EMERG "Restarting system.\n"); 373 printk(KERN_EMERG "Restarting system.\n");
373 else 374 else
@@ -1788,15 +1789,15 @@ SYSCALL_DEFINE1(umask, int, mask)
1788#ifdef CONFIG_CHECKPOINT_RESTORE 1789#ifdef CONFIG_CHECKPOINT_RESTORE
1789static int prctl_set_mm_exe_file(struct mm_struct *mm, unsigned int fd) 1790static int prctl_set_mm_exe_file(struct mm_struct *mm, unsigned int fd)
1790{ 1791{
1791 struct file *exe_file; 1792 struct fd exe;
1792 struct dentry *dentry; 1793 struct dentry *dentry;
1793 int err; 1794 int err;
1794 1795
1795 exe_file = fget(fd); 1796 exe = fdget(fd);
1796 if (!exe_file) 1797 if (!exe.file)
1797 return -EBADF; 1798 return -EBADF;
1798 1799
1799 dentry = exe_file->f_path.dentry; 1800 dentry = exe.file->f_path.dentry;
1800 1801
1801 /* 1802 /*
1802 * Because the original mm->exe_file points to executable file, make 1803 * Because the original mm->exe_file points to executable file, make
@@ -1805,7 +1806,7 @@ static int prctl_set_mm_exe_file(struct mm_struct *mm, unsigned int fd)
1805 */ 1806 */
1806 err = -EACCES; 1807 err = -EACCES;
1807 if (!S_ISREG(dentry->d_inode->i_mode) || 1808 if (!S_ISREG(dentry->d_inode->i_mode) ||
1808 exe_file->f_path.mnt->mnt_flags & MNT_NOEXEC) 1809 exe.file->f_path.mnt->mnt_flags & MNT_NOEXEC)
1809 goto exit; 1810 goto exit;
1810 1811
1811 err = inode_permission(dentry->d_inode, MAY_EXEC); 1812 err = inode_permission(dentry->d_inode, MAY_EXEC);
@@ -1839,12 +1840,12 @@ static int prctl_set_mm_exe_file(struct mm_struct *mm, unsigned int fd)
1839 goto exit_unlock; 1840 goto exit_unlock;
1840 1841
1841 err = 0; 1842 err = 0;
1842 set_mm_exe_file(mm, exe_file); 1843 set_mm_exe_file(mm, exe.file); /* this grabs a reference to exe.file */
1843exit_unlock: 1844exit_unlock:
1844 up_write(&mm->mmap_sem); 1845 up_write(&mm->mmap_sem);
1845 1846
1846exit: 1847exit:
1847 fput(exe_file); 1848 fdput(exe);
1848 return err; 1849 return err;
1849} 1850}
1850 1851
@@ -2204,7 +2205,7 @@ static int __orderly_poweroff(void)
2204 return -ENOMEM; 2205 return -ENOMEM;
2205 } 2206 }
2206 2207
2207 ret = call_usermodehelper_fns(argv[0], argv, envp, UMH_NO_WAIT, 2208 ret = call_usermodehelper_fns(argv[0], argv, envp, UMH_WAIT_EXEC,
2208 NULL, argv_cleanup, NULL); 2209 NULL, argv_cleanup, NULL);
2209 if (ret == -ENOMEM) 2210 if (ret == -ENOMEM)
2210 argv_free(argv); 2211 argv_free(argv);
diff --git a/kernel/sysctl.c b/kernel/sysctl.c
index 87174ef59161..26f65eaa01f9 100644
--- a/kernel/sysctl.c
+++ b/kernel/sysctl.c
@@ -97,10 +97,12 @@
97extern int sysctl_overcommit_memory; 97extern int sysctl_overcommit_memory;
98extern int sysctl_overcommit_ratio; 98extern int sysctl_overcommit_ratio;
99extern int max_threads; 99extern int max_threads;
100extern int core_uses_pid;
101extern int suid_dumpable; 100extern int suid_dumpable;
101#ifdef CONFIG_COREDUMP
102extern int core_uses_pid;
102extern char core_pattern[]; 103extern char core_pattern[];
103extern unsigned int core_pipe_limit; 104extern unsigned int core_pipe_limit;
105#endif
104extern int pid_max; 106extern int pid_max;
105extern int min_free_kbytes; 107extern int min_free_kbytes;
106extern int pid_max_min, pid_max_max; 108extern int pid_max_min, pid_max_max;
@@ -177,8 +179,10 @@ static int proc_dointvec_minmax_sysadmin(struct ctl_table *table, int write,
177 179
178static int proc_dointvec_minmax_coredump(struct ctl_table *table, int write, 180static int proc_dointvec_minmax_coredump(struct ctl_table *table, int write,
179 void __user *buffer, size_t *lenp, loff_t *ppos); 181 void __user *buffer, size_t *lenp, loff_t *ppos);
182#ifdef CONFIG_COREDUMP
180static int proc_dostring_coredump(struct ctl_table *table, int write, 183static int proc_dostring_coredump(struct ctl_table *table, int write,
181 void __user *buffer, size_t *lenp, loff_t *ppos); 184 void __user *buffer, size_t *lenp, loff_t *ppos);
185#endif
182 186
183#ifdef CONFIG_MAGIC_SYSRQ 187#ifdef CONFIG_MAGIC_SYSRQ
184/* Note: sysrq code uses it's own private copy */ 188/* Note: sysrq code uses it's own private copy */
@@ -307,7 +311,7 @@ static struct ctl_table kern_table[] = {
307 .extra2 = &max_sched_tunable_scaling, 311 .extra2 = &max_sched_tunable_scaling,
308 }, 312 },
309 { 313 {
310 .procname = "sched_migration_cost", 314 .procname = "sched_migration_cost_ns",
311 .data = &sysctl_sched_migration_cost, 315 .data = &sysctl_sched_migration_cost,
312 .maxlen = sizeof(unsigned int), 316 .maxlen = sizeof(unsigned int),
313 .mode = 0644, 317 .mode = 0644,
@@ -321,14 +325,14 @@ static struct ctl_table kern_table[] = {
321 .proc_handler = proc_dointvec, 325 .proc_handler = proc_dointvec,
322 }, 326 },
323 { 327 {
324 .procname = "sched_time_avg", 328 .procname = "sched_time_avg_ms",
325 .data = &sysctl_sched_time_avg, 329 .data = &sysctl_sched_time_avg,
326 .maxlen = sizeof(unsigned int), 330 .maxlen = sizeof(unsigned int),
327 .mode = 0644, 331 .mode = 0644,
328 .proc_handler = proc_dointvec, 332 .proc_handler = proc_dointvec,
329 }, 333 },
330 { 334 {
331 .procname = "sched_shares_window", 335 .procname = "sched_shares_window_ns",
332 .data = &sysctl_sched_shares_window, 336 .data = &sysctl_sched_shares_window,
333 .maxlen = sizeof(unsigned int), 337 .maxlen = sizeof(unsigned int),
334 .mode = 0644, 338 .mode = 0644,
@@ -404,6 +408,7 @@ static struct ctl_table kern_table[] = {
404 .mode = 0644, 408 .mode = 0644,
405 .proc_handler = proc_dointvec, 409 .proc_handler = proc_dointvec,
406 }, 410 },
411#ifdef CONFIG_COREDUMP
407 { 412 {
408 .procname = "core_uses_pid", 413 .procname = "core_uses_pid",
409 .data = &core_uses_pid, 414 .data = &core_uses_pid,
@@ -425,6 +430,7 @@ static struct ctl_table kern_table[] = {
425 .mode = 0644, 430 .mode = 0644,
426 .proc_handler = proc_dointvec, 431 .proc_handler = proc_dointvec,
427 }, 432 },
433#endif
428#ifdef CONFIG_PROC_SYSCTL 434#ifdef CONFIG_PROC_SYSCTL
429 { 435 {
430 .procname = "tainted", 436 .procname = "tainted",
@@ -1543,8 +1549,7 @@ static struct ctl_table fs_table[] = {
1543}; 1549};
1544 1550
1545static struct ctl_table debug_table[] = { 1551static struct ctl_table debug_table[] = {
1546#if defined(CONFIG_X86) || defined(CONFIG_PPC) || defined(CONFIG_SPARC) || \ 1552#ifdef CONFIG_SYSCTL_EXCEPTION_TRACE
1547 defined(CONFIG_S390) || defined(CONFIG_TILE)
1548 { 1553 {
1549 .procname = "exception-trace", 1554 .procname = "exception-trace",
1550 .data = &show_unhandled_signals, 1555 .data = &show_unhandled_signals,
@@ -2036,12 +2041,14 @@ int proc_dointvec_minmax(struct ctl_table *table, int write,
2036 2041
2037static void validate_coredump_safety(void) 2042static void validate_coredump_safety(void)
2038{ 2043{
2044#ifdef CONFIG_COREDUMP
2039 if (suid_dumpable == SUID_DUMPABLE_SAFE && 2045 if (suid_dumpable == SUID_DUMPABLE_SAFE &&
2040 core_pattern[0] != '/' && core_pattern[0] != '|') { 2046 core_pattern[0] != '/' && core_pattern[0] != '|') {
2041 printk(KERN_WARNING "Unsafe core_pattern used with "\ 2047 printk(KERN_WARNING "Unsafe core_pattern used with "\
2042 "suid_dumpable=2. Pipe handler or fully qualified "\ 2048 "suid_dumpable=2. Pipe handler or fully qualified "\
2043 "core dump path required.\n"); 2049 "core dump path required.\n");
2044 } 2050 }
2051#endif
2045} 2052}
2046 2053
2047static int proc_dointvec_minmax_coredump(struct ctl_table *table, int write, 2054static int proc_dointvec_minmax_coredump(struct ctl_table *table, int write,
@@ -2053,6 +2060,7 @@ static int proc_dointvec_minmax_coredump(struct ctl_table *table, int write,
2053 return error; 2060 return error;
2054} 2061}
2055 2062
2063#ifdef CONFIG_COREDUMP
2056static int proc_dostring_coredump(struct ctl_table *table, int write, 2064static int proc_dostring_coredump(struct ctl_table *table, int write,
2057 void __user *buffer, size_t *lenp, loff_t *ppos) 2065 void __user *buffer, size_t *lenp, loff_t *ppos)
2058{ 2066{
@@ -2061,6 +2069,7 @@ static int proc_dostring_coredump(struct ctl_table *table, int write,
2061 validate_coredump_safety(); 2069 validate_coredump_safety();
2062 return error; 2070 return error;
2063} 2071}
2072#endif
2064 2073
2065static int __do_proc_doulongvec_minmax(void *data, struct ctl_table *table, int write, 2074static int __do_proc_doulongvec_minmax(void *data, struct ctl_table *table, int write,
2066 void __user *buffer, 2075 void __user *buffer,
diff --git a/kernel/task_work.c b/kernel/task_work.c
index d320d44903bd..65bd3c92d6f3 100644
--- a/kernel/task_work.c
+++ b/kernel/task_work.c
@@ -2,26 +2,20 @@
2#include <linux/task_work.h> 2#include <linux/task_work.h>
3#include <linux/tracehook.h> 3#include <linux/tracehook.h>
4 4
5static struct callback_head work_exited; /* all we need is ->next == NULL */
6
5int 7int
6task_work_add(struct task_struct *task, struct callback_head *twork, bool notify) 8task_work_add(struct task_struct *task, struct callback_head *work, bool notify)
7{ 9{
8 struct callback_head *last, *first; 10 struct callback_head *head;
9 unsigned long flags;
10 11
11 /* 12 do {
12 * Not inserting the new work if the task has already passed 13 head = ACCESS_ONCE(task->task_works);
13 * exit_task_work() is the responisbility of callers. 14 if (unlikely(head == &work_exited))
14 */ 15 return -ESRCH;
15 raw_spin_lock_irqsave(&task->pi_lock, flags); 16 work->next = head;
16 last = task->task_works; 17 } while (cmpxchg(&task->task_works, head, work) != head);
17 first = last ? last->next : twork;
18 twork->next = first;
19 if (last)
20 last->next = twork;
21 task->task_works = twork;
22 raw_spin_unlock_irqrestore(&task->pi_lock, flags);
23 18
24 /* test_and_set_bit() implies mb(), see tracehook_notify_resume(). */
25 if (notify) 19 if (notify)
26 set_notify_resume(task); 20 set_notify_resume(task);
27 return 0; 21 return 0;
@@ -30,52 +24,69 @@ task_work_add(struct task_struct *task, struct callback_head *twork, bool notify
30struct callback_head * 24struct callback_head *
31task_work_cancel(struct task_struct *task, task_work_func_t func) 25task_work_cancel(struct task_struct *task, task_work_func_t func)
32{ 26{
27 struct callback_head **pprev = &task->task_works;
28 struct callback_head *work = NULL;
33 unsigned long flags; 29 unsigned long flags;
34 struct callback_head *last, *res = NULL; 30 /*
35 31 * If cmpxchg() fails we continue without updating pprev.
32 * Either we raced with task_work_add() which added the
33 * new entry before this work, we will find it again. Or
34 * we raced with task_work_run(), *pprev == NULL/exited.
35 */
36 raw_spin_lock_irqsave(&task->pi_lock, flags); 36 raw_spin_lock_irqsave(&task->pi_lock, flags);
37 last = task->task_works; 37 while ((work = ACCESS_ONCE(*pprev))) {
38 if (last) { 38 read_barrier_depends();
39 struct callback_head *q = last, *p = q->next; 39 if (work->func != func)
40 while (1) { 40 pprev = &work->next;
41 if (p->func == func) { 41 else if (cmpxchg(pprev, work, work->next) == work)
42 q->next = p->next; 42 break;
43 if (p == last)
44 task->task_works = q == p ? NULL : q;
45 res = p;
46 break;
47 }
48 if (p == last)
49 break;
50 q = p;
51 p = q->next;
52 }
53 } 43 }
54 raw_spin_unlock_irqrestore(&task->pi_lock, flags); 44 raw_spin_unlock_irqrestore(&task->pi_lock, flags);
55 return res; 45
46 return work;
56} 47}
57 48
58void task_work_run(void) 49void task_work_run(void)
59{ 50{
60 struct task_struct *task = current; 51 struct task_struct *task = current;
61 struct callback_head *p, *q; 52 struct callback_head *work, *head, *next;
53
54 for (;;) {
55 /*
56 * work->func() can do task_work_add(), do not set
57 * work_exited unless the list is empty.
58 */
59 do {
60 work = ACCESS_ONCE(task->task_works);
61 head = !work && (task->flags & PF_EXITING) ?
62 &work_exited : NULL;
63 } while (cmpxchg(&task->task_works, work, head) != work);
62 64
63 while (1) { 65 if (!work)
64 raw_spin_lock_irq(&task->pi_lock); 66 break;
65 p = task->task_works; 67 /*
66 task->task_works = NULL; 68 * Synchronize with task_work_cancel(). It can't remove
67 raw_spin_unlock_irq(&task->pi_lock); 69 * the first entry == work, cmpxchg(task_works) should
70 * fail, but it can play with *work and other entries.
71 */
72 raw_spin_unlock_wait(&task->pi_lock);
73 smp_mb();
68 74
69 if (unlikely(!p)) 75 /* Reverse the list to run the works in fifo order */
70 return; 76 head = NULL;
77 do {
78 next = work->next;
79 work->next = head;
80 head = work;
81 work = next;
82 } while (work);
71 83
72 q = p->next; /* head */ 84 work = head;
73 p->next = NULL; /* cut it */ 85 do {
74 while (q) { 86 next = work->next;
75 p = q->next; 87 work->func(work);
76 q->func(q); 88 work = next;
77 q = p;
78 cond_resched(); 89 cond_resched();
79 } 90 } while (work);
80 } 91 }
81} 92}
diff --git a/kernel/taskstats.c b/kernel/taskstats.c
index d0a32796550f..145bb4d3bd4d 100644
--- a/kernel/taskstats.c
+++ b/kernel/taskstats.c
@@ -27,6 +27,7 @@
27#include <linux/cgroup.h> 27#include <linux/cgroup.h>
28#include <linux/fs.h> 28#include <linux/fs.h>
29#include <linux/file.h> 29#include <linux/file.h>
30#include <linux/pid_namespace.h>
30#include <net/genetlink.h> 31#include <net/genetlink.h>
31#include <linux/atomic.h> 32#include <linux/atomic.h>
32 33
@@ -174,7 +175,9 @@ static void send_cpu_listeners(struct sk_buff *skb,
174 up_write(&listeners->sem); 175 up_write(&listeners->sem);
175} 176}
176 177
177static void fill_stats(struct task_struct *tsk, struct taskstats *stats) 178static void fill_stats(struct user_namespace *user_ns,
179 struct pid_namespace *pid_ns,
180 struct task_struct *tsk, struct taskstats *stats)
178{ 181{
179 memset(stats, 0, sizeof(*stats)); 182 memset(stats, 0, sizeof(*stats));
180 /* 183 /*
@@ -190,7 +193,7 @@ static void fill_stats(struct task_struct *tsk, struct taskstats *stats)
190 stats->version = TASKSTATS_VERSION; 193 stats->version = TASKSTATS_VERSION;
191 stats->nvcsw = tsk->nvcsw; 194 stats->nvcsw = tsk->nvcsw;
192 stats->nivcsw = tsk->nivcsw; 195 stats->nivcsw = tsk->nivcsw;
193 bacct_add_tsk(stats, tsk); 196 bacct_add_tsk(user_ns, pid_ns, stats, tsk);
194 197
195 /* fill in extended acct fields */ 198 /* fill in extended acct fields */
196 xacct_add_tsk(stats, tsk); 199 xacct_add_tsk(stats, tsk);
@@ -207,7 +210,7 @@ static int fill_stats_for_pid(pid_t pid, struct taskstats *stats)
207 rcu_read_unlock(); 210 rcu_read_unlock();
208 if (!tsk) 211 if (!tsk)
209 return -ESRCH; 212 return -ESRCH;
210 fill_stats(tsk, stats); 213 fill_stats(current_user_ns(), task_active_pid_ns(current), tsk, stats);
211 put_task_struct(tsk); 214 put_task_struct(tsk);
212 return 0; 215 return 0;
213} 216}
@@ -291,6 +294,12 @@ static int add_del_listener(pid_t pid, const struct cpumask *mask, int isadd)
291 if (!cpumask_subset(mask, cpu_possible_mask)) 294 if (!cpumask_subset(mask, cpu_possible_mask))
292 return -EINVAL; 295 return -EINVAL;
293 296
297 if (current_user_ns() != &init_user_ns)
298 return -EINVAL;
299
300 if (task_active_pid_ns(current) != &init_pid_ns)
301 return -EINVAL;
302
294 if (isadd == REGISTER) { 303 if (isadd == REGISTER) {
295 for_each_cpu(cpu, mask) { 304 for_each_cpu(cpu, mask) {
296 s = kmalloc_node(sizeof(struct listener), 305 s = kmalloc_node(sizeof(struct listener),
@@ -415,16 +424,15 @@ static int cgroupstats_user_cmd(struct sk_buff *skb, struct genl_info *info)
415 struct nlattr *na; 424 struct nlattr *na;
416 size_t size; 425 size_t size;
417 u32 fd; 426 u32 fd;
418 struct file *file; 427 struct fd f;
419 int fput_needed;
420 428
421 na = info->attrs[CGROUPSTATS_CMD_ATTR_FD]; 429 na = info->attrs[CGROUPSTATS_CMD_ATTR_FD];
422 if (!na) 430 if (!na)
423 return -EINVAL; 431 return -EINVAL;
424 432
425 fd = nla_get_u32(info->attrs[CGROUPSTATS_CMD_ATTR_FD]); 433 fd = nla_get_u32(info->attrs[CGROUPSTATS_CMD_ATTR_FD]);
426 file = fget_light(fd, &fput_needed); 434 f = fdget(fd);
427 if (!file) 435 if (!f.file)
428 return 0; 436 return 0;
429 437
430 size = nla_total_size(sizeof(struct cgroupstats)); 438 size = nla_total_size(sizeof(struct cgroupstats));
@@ -437,6 +445,7 @@ static int cgroupstats_user_cmd(struct sk_buff *skb, struct genl_info *info)
437 na = nla_reserve(rep_skb, CGROUPSTATS_TYPE_CGROUP_STATS, 445 na = nla_reserve(rep_skb, CGROUPSTATS_TYPE_CGROUP_STATS,
438 sizeof(struct cgroupstats)); 446 sizeof(struct cgroupstats));
439 if (na == NULL) { 447 if (na == NULL) {
448 nlmsg_free(rep_skb);
440 rc = -EMSGSIZE; 449 rc = -EMSGSIZE;
441 goto err; 450 goto err;
442 } 451 }
@@ -444,7 +453,7 @@ static int cgroupstats_user_cmd(struct sk_buff *skb, struct genl_info *info)
444 stats = nla_data(na); 453 stats = nla_data(na);
445 memset(stats, 0, sizeof(*stats)); 454 memset(stats, 0, sizeof(*stats));
446 455
447 rc = cgroupstats_build(stats, file->f_dentry); 456 rc = cgroupstats_build(stats, f.file->f_dentry);
448 if (rc < 0) { 457 if (rc < 0) {
449 nlmsg_free(rep_skb); 458 nlmsg_free(rep_skb);
450 goto err; 459 goto err;
@@ -453,7 +462,7 @@ static int cgroupstats_user_cmd(struct sk_buff *skb, struct genl_info *info)
453 rc = send_reply(rep_skb, info); 462 rc = send_reply(rep_skb, info);
454 463
455err: 464err:
456 fput_light(file, fput_needed); 465 fdput(f);
457 return rc; 466 return rc;
458} 467}
459 468
@@ -467,7 +476,7 @@ static int cmd_attr_register_cpumask(struct genl_info *info)
467 rc = parse(info->attrs[TASKSTATS_CMD_ATTR_REGISTER_CPUMASK], mask); 476 rc = parse(info->attrs[TASKSTATS_CMD_ATTR_REGISTER_CPUMASK], mask);
468 if (rc < 0) 477 if (rc < 0)
469 goto out; 478 goto out;
470 rc = add_del_listener(info->snd_pid, mask, REGISTER); 479 rc = add_del_listener(info->snd_portid, mask, REGISTER);
471out: 480out:
472 free_cpumask_var(mask); 481 free_cpumask_var(mask);
473 return rc; 482 return rc;
@@ -483,7 +492,7 @@ static int cmd_attr_deregister_cpumask(struct genl_info *info)
483 rc = parse(info->attrs[TASKSTATS_CMD_ATTR_DEREGISTER_CPUMASK], mask); 492 rc = parse(info->attrs[TASKSTATS_CMD_ATTR_DEREGISTER_CPUMASK], mask);
484 if (rc < 0) 493 if (rc < 0)
485 goto out; 494 goto out;
486 rc = add_del_listener(info->snd_pid, mask, DEREGISTER); 495 rc = add_del_listener(info->snd_portid, mask, DEREGISTER);
487out: 496out:
488 free_cpumask_var(mask); 497 free_cpumask_var(mask);
489 return rc; 498 return rc;
@@ -631,11 +640,12 @@ void taskstats_exit(struct task_struct *tsk, int group_dead)
631 if (rc < 0) 640 if (rc < 0)
632 return; 641 return;
633 642
634 stats = mk_reply(rep_skb, TASKSTATS_TYPE_PID, tsk->pid); 643 stats = mk_reply(rep_skb, TASKSTATS_TYPE_PID,
644 task_pid_nr_ns(tsk, &init_pid_ns));
635 if (!stats) 645 if (!stats)
636 goto err; 646 goto err;
637 647
638 fill_stats(tsk, stats); 648 fill_stats(&init_user_ns, &init_pid_ns, tsk, stats);
639 649
640 /* 650 /*
641 * Doesn't matter if tsk is the leader or the last group member leaving 651 * Doesn't matter if tsk is the leader or the last group member leaving
@@ -643,7 +653,8 @@ void taskstats_exit(struct task_struct *tsk, int group_dead)
643 if (!is_thread_group || !group_dead) 653 if (!is_thread_group || !group_dead)
644 goto send; 654 goto send;
645 655
646 stats = mk_reply(rep_skb, TASKSTATS_TYPE_TGID, tsk->tgid); 656 stats = mk_reply(rep_skb, TASKSTATS_TYPE_TGID,
657 task_tgid_nr_ns(tsk, &init_pid_ns));
647 if (!stats) 658 if (!stats)
648 goto err; 659 goto err;
649 660
diff --git a/kernel/time.c b/kernel/time.c
index ba744cf80696..d226c6a3fd28 100644
--- a/kernel/time.c
+++ b/kernel/time.c
@@ -30,7 +30,7 @@
30#include <linux/export.h> 30#include <linux/export.h>
31#include <linux/timex.h> 31#include <linux/timex.h>
32#include <linux/capability.h> 32#include <linux/capability.h>
33#include <linux/clocksource.h> 33#include <linux/timekeeper_internal.h>
34#include <linux/errno.h> 34#include <linux/errno.h>
35#include <linux/syscalls.h> 35#include <linux/syscalls.h>
36#include <linux/security.h> 36#include <linux/security.h>
diff --git a/kernel/time/Kconfig b/kernel/time/Kconfig
index fd42bd452b75..8601f0db1261 100644
--- a/kernel/time/Kconfig
+++ b/kernel/time/Kconfig
@@ -16,6 +16,10 @@ config ARCH_CLOCKSOURCE_DATA
16config GENERIC_TIME_VSYSCALL 16config GENERIC_TIME_VSYSCALL
17 bool 17 bool
18 18
19# Timekeeping vsyscall support
20config GENERIC_TIME_VSYSCALL_OLD
21 bool
22
19# ktime_t scalar 64bit nsec representation 23# ktime_t scalar 64bit nsec representation
20config KTIME_SCALAR 24config KTIME_SCALAR
21 bool 25 bool
diff --git a/kernel/time/alarmtimer.c b/kernel/time/alarmtimer.c
index aa27d391bfc8..f11d83b12949 100644
--- a/kernel/time/alarmtimer.c
+++ b/kernel/time/alarmtimer.c
@@ -37,7 +37,6 @@
37static struct alarm_base { 37static struct alarm_base {
38 spinlock_t lock; 38 spinlock_t lock;
39 struct timerqueue_head timerqueue; 39 struct timerqueue_head timerqueue;
40 struct hrtimer timer;
41 ktime_t (*gettime)(void); 40 ktime_t (*gettime)(void);
42 clockid_t base_clockid; 41 clockid_t base_clockid;
43} alarm_bases[ALARM_NUMTYPE]; 42} alarm_bases[ALARM_NUMTYPE];
@@ -46,6 +45,8 @@ static struct alarm_base {
46static ktime_t freezer_delta; 45static ktime_t freezer_delta;
47static DEFINE_SPINLOCK(freezer_delta_lock); 46static DEFINE_SPINLOCK(freezer_delta_lock);
48 47
48static struct wakeup_source *ws;
49
49#ifdef CONFIG_RTC_CLASS 50#ifdef CONFIG_RTC_CLASS
50/* rtc timer and device for setting alarm wakeups at suspend */ 51/* rtc timer and device for setting alarm wakeups at suspend */
51static struct rtc_timer rtctimer; 52static struct rtc_timer rtctimer;
@@ -130,50 +131,35 @@ static inline void alarmtimer_rtc_timer_init(void) { }
130 * @base: pointer to the base where the timer is being run 131 * @base: pointer to the base where the timer is being run
131 * @alarm: pointer to alarm being enqueued. 132 * @alarm: pointer to alarm being enqueued.
132 * 133 *
133 * Adds alarm to a alarm_base timerqueue and if necessary sets 134 * Adds alarm to a alarm_base timerqueue
134 * an hrtimer to run.
135 * 135 *
136 * Must hold base->lock when calling. 136 * Must hold base->lock when calling.
137 */ 137 */
138static void alarmtimer_enqueue(struct alarm_base *base, struct alarm *alarm) 138static void alarmtimer_enqueue(struct alarm_base *base, struct alarm *alarm)
139{ 139{
140 if (alarm->state & ALARMTIMER_STATE_ENQUEUED)
141 timerqueue_del(&base->timerqueue, &alarm->node);
142
140 timerqueue_add(&base->timerqueue, &alarm->node); 143 timerqueue_add(&base->timerqueue, &alarm->node);
141 alarm->state |= ALARMTIMER_STATE_ENQUEUED; 144 alarm->state |= ALARMTIMER_STATE_ENQUEUED;
142
143 if (&alarm->node == timerqueue_getnext(&base->timerqueue)) {
144 hrtimer_try_to_cancel(&base->timer);
145 hrtimer_start(&base->timer, alarm->node.expires,
146 HRTIMER_MODE_ABS);
147 }
148} 145}
149 146
150/** 147/**
151 * alarmtimer_remove - Removes an alarm timer from an alarm_base timerqueue 148 * alarmtimer_dequeue - Removes an alarm timer from an alarm_base timerqueue
152 * @base: pointer to the base where the timer is running 149 * @base: pointer to the base where the timer is running
153 * @alarm: pointer to alarm being removed 150 * @alarm: pointer to alarm being removed
154 * 151 *
155 * Removes alarm to a alarm_base timerqueue and if necessary sets 152 * Removes alarm to a alarm_base timerqueue
156 * a new timer to run.
157 * 153 *
158 * Must hold base->lock when calling. 154 * Must hold base->lock when calling.
159 */ 155 */
160static void alarmtimer_remove(struct alarm_base *base, struct alarm *alarm) 156static void alarmtimer_dequeue(struct alarm_base *base, struct alarm *alarm)
161{ 157{
162 struct timerqueue_node *next = timerqueue_getnext(&base->timerqueue);
163
164 if (!(alarm->state & ALARMTIMER_STATE_ENQUEUED)) 158 if (!(alarm->state & ALARMTIMER_STATE_ENQUEUED))
165 return; 159 return;
166 160
167 timerqueue_del(&base->timerqueue, &alarm->node); 161 timerqueue_del(&base->timerqueue, &alarm->node);
168 alarm->state &= ~ALARMTIMER_STATE_ENQUEUED; 162 alarm->state &= ~ALARMTIMER_STATE_ENQUEUED;
169
170 if (next == &alarm->node) {
171 hrtimer_try_to_cancel(&base->timer);
172 next = timerqueue_getnext(&base->timerqueue);
173 if (!next)
174 return;
175 hrtimer_start(&base->timer, next->expires, HRTIMER_MODE_ABS);
176 }
177} 163}
178 164
179 165
@@ -188,42 +174,23 @@ static void alarmtimer_remove(struct alarm_base *base, struct alarm *alarm)
188 */ 174 */
189static enum hrtimer_restart alarmtimer_fired(struct hrtimer *timer) 175static enum hrtimer_restart alarmtimer_fired(struct hrtimer *timer)
190{ 176{
191 struct alarm_base *base = container_of(timer, struct alarm_base, timer); 177 struct alarm *alarm = container_of(timer, struct alarm, timer);
192 struct timerqueue_node *next; 178 struct alarm_base *base = &alarm_bases[alarm->type];
193 unsigned long flags; 179 unsigned long flags;
194 ktime_t now;
195 int ret = HRTIMER_NORESTART; 180 int ret = HRTIMER_NORESTART;
196 int restart = ALARMTIMER_NORESTART; 181 int restart = ALARMTIMER_NORESTART;
197 182
198 spin_lock_irqsave(&base->lock, flags); 183 spin_lock_irqsave(&base->lock, flags);
199 now = base->gettime(); 184 alarmtimer_dequeue(base, alarm);
200 while ((next = timerqueue_getnext(&base->timerqueue))) { 185 spin_unlock_irqrestore(&base->lock, flags);
201 struct alarm *alarm;
202 ktime_t expired = next->expires;
203
204 if (expired.tv64 > now.tv64)
205 break;
206
207 alarm = container_of(next, struct alarm, node);
208
209 timerqueue_del(&base->timerqueue, &alarm->node);
210 alarm->state &= ~ALARMTIMER_STATE_ENQUEUED;
211
212 alarm->state |= ALARMTIMER_STATE_CALLBACK;
213 spin_unlock_irqrestore(&base->lock, flags);
214 if (alarm->function)
215 restart = alarm->function(alarm, now);
216 spin_lock_irqsave(&base->lock, flags);
217 alarm->state &= ~ALARMTIMER_STATE_CALLBACK;
218 186
219 if (restart != ALARMTIMER_NORESTART) { 187 if (alarm->function)
220 timerqueue_add(&base->timerqueue, &alarm->node); 188 restart = alarm->function(alarm, base->gettime());
221 alarm->state |= ALARMTIMER_STATE_ENQUEUED;
222 }
223 }
224 189
225 if (next) { 190 spin_lock_irqsave(&base->lock, flags);
226 hrtimer_set_expires(&base->timer, next->expires); 191 if (restart != ALARMTIMER_NORESTART) {
192 hrtimer_set_expires(&alarm->timer, alarm->node.expires);
193 alarmtimer_enqueue(base, alarm);
227 ret = HRTIMER_RESTART; 194 ret = HRTIMER_RESTART;
228 } 195 }
229 spin_unlock_irqrestore(&base->lock, flags); 196 spin_unlock_irqrestore(&base->lock, flags);
@@ -250,6 +217,7 @@ static int alarmtimer_suspend(struct device *dev)
250 unsigned long flags; 217 unsigned long flags;
251 struct rtc_device *rtc; 218 struct rtc_device *rtc;
252 int i; 219 int i;
220 int ret;
253 221
254 spin_lock_irqsave(&freezer_delta_lock, flags); 222 spin_lock_irqsave(&freezer_delta_lock, flags);
255 min = freezer_delta; 223 min = freezer_delta;
@@ -279,8 +247,10 @@ static int alarmtimer_suspend(struct device *dev)
279 if (min.tv64 == 0) 247 if (min.tv64 == 0)
280 return 0; 248 return 0;
281 249
282 /* XXX - Should we enforce a minimum sleep time? */ 250 if (ktime_to_ns(min) < 2 * NSEC_PER_SEC) {
283 WARN_ON(min.tv64 < NSEC_PER_SEC); 251 __pm_wakeup_event(ws, 2 * MSEC_PER_SEC);
252 return -EBUSY;
253 }
284 254
285 /* Setup an rtc timer to fire that far in the future */ 255 /* Setup an rtc timer to fire that far in the future */
286 rtc_timer_cancel(rtc, &rtctimer); 256 rtc_timer_cancel(rtc, &rtctimer);
@@ -288,9 +258,11 @@ static int alarmtimer_suspend(struct device *dev)
288 now = rtc_tm_to_ktime(tm); 258 now = rtc_tm_to_ktime(tm);
289 now = ktime_add(now, min); 259 now = ktime_add(now, min);
290 260
291 rtc_timer_start(rtc, &rtctimer, now, ktime_set(0, 0)); 261 /* Set alarm, if in the past reject suspend briefly to handle */
292 262 ret = rtc_timer_start(rtc, &rtctimer, now, ktime_set(0, 0));
293 return 0; 263 if (ret < 0)
264 __pm_wakeup_event(ws, MSEC_PER_SEC);
265 return ret;
294} 266}
295#else 267#else
296static int alarmtimer_suspend(struct device *dev) 268static int alarmtimer_suspend(struct device *dev)
@@ -324,6 +296,9 @@ void alarm_init(struct alarm *alarm, enum alarmtimer_type type,
324 enum alarmtimer_restart (*function)(struct alarm *, ktime_t)) 296 enum alarmtimer_restart (*function)(struct alarm *, ktime_t))
325{ 297{
326 timerqueue_init(&alarm->node); 298 timerqueue_init(&alarm->node);
299 hrtimer_init(&alarm->timer, alarm_bases[type].base_clockid,
300 HRTIMER_MODE_ABS);
301 alarm->timer.function = alarmtimer_fired;
327 alarm->function = function; 302 alarm->function = function;
328 alarm->type = type; 303 alarm->type = type;
329 alarm->state = ALARMTIMER_STATE_INACTIVE; 304 alarm->state = ALARMTIMER_STATE_INACTIVE;
@@ -334,17 +309,19 @@ void alarm_init(struct alarm *alarm, enum alarmtimer_type type,
334 * @alarm: ptr to alarm to set 309 * @alarm: ptr to alarm to set
335 * @start: time to run the alarm 310 * @start: time to run the alarm
336 */ 311 */
337void alarm_start(struct alarm *alarm, ktime_t start) 312int alarm_start(struct alarm *alarm, ktime_t start)
338{ 313{
339 struct alarm_base *base = &alarm_bases[alarm->type]; 314 struct alarm_base *base = &alarm_bases[alarm->type];
340 unsigned long flags; 315 unsigned long flags;
316 int ret;
341 317
342 spin_lock_irqsave(&base->lock, flags); 318 spin_lock_irqsave(&base->lock, flags);
343 if (alarmtimer_active(alarm))
344 alarmtimer_remove(base, alarm);
345 alarm->node.expires = start; 319 alarm->node.expires = start;
346 alarmtimer_enqueue(base, alarm); 320 alarmtimer_enqueue(base, alarm);
321 ret = hrtimer_start(&alarm->timer, alarm->node.expires,
322 HRTIMER_MODE_ABS);
347 spin_unlock_irqrestore(&base->lock, flags); 323 spin_unlock_irqrestore(&base->lock, flags);
324 return ret;
348} 325}
349 326
350/** 327/**
@@ -358,18 +335,12 @@ int alarm_try_to_cancel(struct alarm *alarm)
358{ 335{
359 struct alarm_base *base = &alarm_bases[alarm->type]; 336 struct alarm_base *base = &alarm_bases[alarm->type];
360 unsigned long flags; 337 unsigned long flags;
361 int ret = -1; 338 int ret;
362 spin_lock_irqsave(&base->lock, flags);
363
364 if (alarmtimer_callback_running(alarm))
365 goto out;
366 339
367 if (alarmtimer_is_queued(alarm)) { 340 spin_lock_irqsave(&base->lock, flags);
368 alarmtimer_remove(base, alarm); 341 ret = hrtimer_try_to_cancel(&alarm->timer);
369 ret = 1; 342 if (ret >= 0)
370 } else 343 alarmtimer_dequeue(base, alarm);
371 ret = 0;
372out:
373 spin_unlock_irqrestore(&base->lock, flags); 344 spin_unlock_irqrestore(&base->lock, flags);
374 return ret; 345 return ret;
375} 346}
@@ -802,10 +773,6 @@ static int __init alarmtimer_init(void)
802 for (i = 0; i < ALARM_NUMTYPE; i++) { 773 for (i = 0; i < ALARM_NUMTYPE; i++) {
803 timerqueue_init_head(&alarm_bases[i].timerqueue); 774 timerqueue_init_head(&alarm_bases[i].timerqueue);
804 spin_lock_init(&alarm_bases[i].lock); 775 spin_lock_init(&alarm_bases[i].lock);
805 hrtimer_init(&alarm_bases[i].timer,
806 alarm_bases[i].base_clockid,
807 HRTIMER_MODE_ABS);
808 alarm_bases[i].timer.function = alarmtimer_fired;
809 } 776 }
810 777
811 error = alarmtimer_rtc_interface_setup(); 778 error = alarmtimer_rtc_interface_setup();
@@ -821,6 +788,7 @@ static int __init alarmtimer_init(void)
821 error = PTR_ERR(pdev); 788 error = PTR_ERR(pdev);
822 goto out_drv; 789 goto out_drv;
823 } 790 }
791 ws = wakeup_source_register("alarmtimer");
824 return 0; 792 return 0;
825 793
826out_drv: 794out_drv:
diff --git a/kernel/time/clockevents.c b/kernel/time/clockevents.c
index 7e1ce012a851..30b6de0d977c 100644
--- a/kernel/time/clockevents.c
+++ b/kernel/time/clockevents.c
@@ -397,6 +397,30 @@ void clockevents_exchange_device(struct clock_event_device *old,
397 local_irq_restore(flags); 397 local_irq_restore(flags);
398} 398}
399 399
400/**
401 * clockevents_suspend - suspend clock devices
402 */
403void clockevents_suspend(void)
404{
405 struct clock_event_device *dev;
406
407 list_for_each_entry_reverse(dev, &clockevent_devices, list)
408 if (dev->suspend)
409 dev->suspend(dev);
410}
411
412/**
413 * clockevents_resume - resume clock devices
414 */
415void clockevents_resume(void)
416{
417 struct clock_event_device *dev;
418
419 list_for_each_entry(dev, &clockevent_devices, list)
420 if (dev->resume)
421 dev->resume(dev);
422}
423
400#ifdef CONFIG_GENERIC_CLOCKEVENTS 424#ifdef CONFIG_GENERIC_CLOCKEVENTS
401/** 425/**
402 * clockevents_notify - notification about relevant events 426 * clockevents_notify - notification about relevant events
diff --git a/kernel/time/jiffies.c b/kernel/time/jiffies.c
index 46da0537c10b..6629bf7b5285 100644
--- a/kernel/time/jiffies.c
+++ b/kernel/time/jiffies.c
@@ -37,7 +37,7 @@
37 * requested HZ value. It is also not recommended 37 * requested HZ value. It is also not recommended
38 * for "tick-less" systems. 38 * for "tick-less" systems.
39 */ 39 */
40#define NSEC_PER_JIFFY ((u32)((((u64)NSEC_PER_SEC)<<8)/SHIFTED_HZ)) 40#define NSEC_PER_JIFFY ((NSEC_PER_SEC+HZ/2)/HZ)
41 41
42/* Since jiffies uses a simple NSEC_PER_JIFFY multiplier 42/* Since jiffies uses a simple NSEC_PER_JIFFY multiplier
43 * conversion, the .shift value could be zero. However 43 * conversion, the .shift value could be zero. However
@@ -95,3 +95,33 @@ struct clocksource * __init __weak clocksource_default_clock(void)
95{ 95{
96 return &clocksource_jiffies; 96 return &clocksource_jiffies;
97} 97}
98
99struct clocksource refined_jiffies;
100
101int register_refined_jiffies(long cycles_per_second)
102{
103 u64 nsec_per_tick, shift_hz;
104 long cycles_per_tick;
105
106
107
108 refined_jiffies = clocksource_jiffies;
109 refined_jiffies.name = "refined-jiffies";
110 refined_jiffies.rating++;
111
112 /* Calc cycles per tick */
113 cycles_per_tick = (cycles_per_second + HZ/2)/HZ;
114 /* shift_hz stores hz<<8 for extra accuracy */
115 shift_hz = (u64)cycles_per_second << 8;
116 shift_hz += cycles_per_tick/2;
117 do_div(shift_hz, cycles_per_tick);
118 /* Calculate nsec_per_tick using shift_hz */
119 nsec_per_tick = (u64)NSEC_PER_SEC << 8;
120 nsec_per_tick += (u32)shift_hz/2;
121 do_div(nsec_per_tick, (u32)shift_hz);
122
123 refined_jiffies.mult = ((u32)nsec_per_tick) << JIFFIES_SHIFT;
124
125 clocksource_register(&refined_jiffies);
126 return 0;
127}
diff --git a/kernel/time/tick-sched.c b/kernel/time/tick-sched.c
index 3a9e5d5c1091..a40260885265 100644
--- a/kernel/time/tick-sched.c
+++ b/kernel/time/tick-sched.c
@@ -372,7 +372,7 @@ static ktime_t tick_nohz_stop_sched_tick(struct tick_sched *ts,
372 * the scheduler tick in nohz_restart_sched_tick. 372 * the scheduler tick in nohz_restart_sched_tick.
373 */ 373 */
374 if (!ts->tick_stopped) { 374 if (!ts->tick_stopped) {
375 select_nohz_load_balancer(1); 375 nohz_balance_enter_idle(cpu);
376 calc_load_enter_idle(); 376 calc_load_enter_idle();
377 377
378 ts->last_tick = hrtimer_get_expires(&ts->sched_timer); 378 ts->last_tick = hrtimer_get_expires(&ts->sched_timer);
@@ -436,7 +436,8 @@ static bool can_stop_idle_tick(int cpu, struct tick_sched *ts)
436 if (unlikely(local_softirq_pending() && cpu_online(cpu))) { 436 if (unlikely(local_softirq_pending() && cpu_online(cpu))) {
437 static int ratelimit; 437 static int ratelimit;
438 438
439 if (ratelimit < 10) { 439 if (ratelimit < 10 &&
440 (local_softirq_pending() & SOFTIRQ_STOP_IDLE_MASK)) {
440 printk(KERN_ERR "NOHZ: local_softirq_pending %02x\n", 441 printk(KERN_ERR "NOHZ: local_softirq_pending %02x\n",
441 (unsigned int) local_softirq_pending()); 442 (unsigned int) local_softirq_pending());
442 ratelimit++; 443 ratelimit++;
@@ -569,7 +570,6 @@ static void tick_nohz_restart(struct tick_sched *ts, ktime_t now)
569static void tick_nohz_restart_sched_tick(struct tick_sched *ts, ktime_t now) 570static void tick_nohz_restart_sched_tick(struct tick_sched *ts, ktime_t now)
570{ 571{
571 /* Update jiffies first */ 572 /* Update jiffies first */
572 select_nohz_load_balancer(0);
573 tick_do_update_jiffies64(now); 573 tick_do_update_jiffies64(now);
574 update_cpu_load_nohz(); 574 update_cpu_load_nohz();
575 575
@@ -835,7 +835,7 @@ static enum hrtimer_restart tick_sched_timer(struct hrtimer *timer)
835 */ 835 */
836 if (ts->tick_stopped) { 836 if (ts->tick_stopped) {
837 touch_softlockup_watchdog(); 837 touch_softlockup_watchdog();
838 if (idle_cpu(cpu)) 838 if (is_idle_task(current))
839 ts->idle_jiffies++; 839 ts->idle_jiffies++;
840 } 840 }
841 update_process_times(user_mode(regs)); 841 update_process_times(user_mode(regs));
diff --git a/kernel/time/timekeeping.c b/kernel/time/timekeeping.c
index 34e5eac81424..e424970bb562 100644
--- a/kernel/time/timekeeping.c
+++ b/kernel/time/timekeeping.c
@@ -8,6 +8,7 @@
8 * 8 *
9 */ 9 */
10 10
11#include <linux/timekeeper_internal.h>
11#include <linux/module.h> 12#include <linux/module.h>
12#include <linux/interrupt.h> 13#include <linux/interrupt.h>
13#include <linux/percpu.h> 14#include <linux/percpu.h>
@@ -21,61 +22,6 @@
21#include <linux/tick.h> 22#include <linux/tick.h>
22#include <linux/stop_machine.h> 23#include <linux/stop_machine.h>
23 24
24/* Structure holding internal timekeeping values. */
25struct timekeeper {
26 /* Current clocksource used for timekeeping. */
27 struct clocksource *clock;
28 /* NTP adjusted clock multiplier */
29 u32 mult;
30 /* The shift value of the current clocksource. */
31 u32 shift;
32 /* Number of clock cycles in one NTP interval. */
33 cycle_t cycle_interval;
34 /* Number of clock shifted nano seconds in one NTP interval. */
35 u64 xtime_interval;
36 /* shifted nano seconds left over when rounding cycle_interval */
37 s64 xtime_remainder;
38 /* Raw nano seconds accumulated per NTP interval. */
39 u32 raw_interval;
40
41 /* Current CLOCK_REALTIME time in seconds */
42 u64 xtime_sec;
43 /* Clock shifted nano seconds */
44 u64 xtime_nsec;
45
46 /* Difference between accumulated time and NTP time in ntp
47 * shifted nano seconds. */
48 s64 ntp_error;
49 /* Shift conversion between clock shifted nano seconds and
50 * ntp shifted nano seconds. */
51 u32 ntp_error_shift;
52
53 /*
54 * wall_to_monotonic is what we need to add to xtime (or xtime corrected
55 * for sub jiffie times) to get to monotonic time. Monotonic is pegged
56 * at zero at system boot time, so wall_to_monotonic will be negative,
57 * however, we will ALWAYS keep the tv_nsec part positive so we can use
58 * the usual normalization.
59 *
60 * wall_to_monotonic is moved after resume from suspend for the
61 * monotonic time not to jump. We need to add total_sleep_time to
62 * wall_to_monotonic to get the real boot based time offset.
63 *
64 * - wall_to_monotonic is no longer the boot time, getboottime must be
65 * used instead.
66 */
67 struct timespec wall_to_monotonic;
68 /* Offset clock monotonic -> clock realtime */
69 ktime_t offs_real;
70 /* time spent in suspend */
71 struct timespec total_sleep_time;
72 /* Offset clock monotonic -> clock boottime */
73 ktime_t offs_boot;
74 /* The raw monotonic time for the CLOCK_MONOTONIC_RAW posix clock. */
75 struct timespec raw_time;
76 /* Seqlock for all timekeeper values */
77 seqlock_t lock;
78};
79 25
80static struct timekeeper timekeeper; 26static struct timekeeper timekeeper;
81 27
@@ -96,15 +42,6 @@ static inline void tk_normalize_xtime(struct timekeeper *tk)
96 } 42 }
97} 43}
98 44
99static struct timespec tk_xtime(struct timekeeper *tk)
100{
101 struct timespec ts;
102
103 ts.tv_sec = tk->xtime_sec;
104 ts.tv_nsec = (long)(tk->xtime_nsec >> tk->shift);
105 return ts;
106}
107
108static void tk_set_xtime(struct timekeeper *tk, const struct timespec *ts) 45static void tk_set_xtime(struct timekeeper *tk, const struct timespec *ts)
109{ 46{
110 tk->xtime_sec = ts->tv_sec; 47 tk->xtime_sec = ts->tv_sec;
@@ -246,14 +183,11 @@ static inline s64 timekeeping_get_ns_raw(struct timekeeper *tk)
246/* must hold write on timekeeper.lock */ 183/* must hold write on timekeeper.lock */
247static void timekeeping_update(struct timekeeper *tk, bool clearntp) 184static void timekeeping_update(struct timekeeper *tk, bool clearntp)
248{ 185{
249 struct timespec xt;
250
251 if (clearntp) { 186 if (clearntp) {
252 tk->ntp_error = 0; 187 tk->ntp_error = 0;
253 ntp_clear(); 188 ntp_clear();
254 } 189 }
255 xt = tk_xtime(tk); 190 update_vsyscall(tk);
256 update_vsyscall(&xt, &tk->wall_to_monotonic, tk->clock, tk->mult);
257} 191}
258 192
259/** 193/**
@@ -303,10 +237,11 @@ void getnstimeofday(struct timespec *ts)
303 seq = read_seqbegin(&tk->lock); 237 seq = read_seqbegin(&tk->lock);
304 238
305 ts->tv_sec = tk->xtime_sec; 239 ts->tv_sec = tk->xtime_sec;
306 ts->tv_nsec = timekeeping_get_ns(tk); 240 nsecs = timekeeping_get_ns(tk);
307 241
308 } while (read_seqretry(&tk->lock, seq)); 242 } while (read_seqretry(&tk->lock, seq));
309 243
244 ts->tv_nsec = 0;
310 timespec_add_ns(ts, nsecs); 245 timespec_add_ns(ts, nsecs);
311} 246}
312EXPORT_SYMBOL(getnstimeofday); 247EXPORT_SYMBOL(getnstimeofday);
@@ -345,6 +280,7 @@ void ktime_get_ts(struct timespec *ts)
345{ 280{
346 struct timekeeper *tk = &timekeeper; 281 struct timekeeper *tk = &timekeeper;
347 struct timespec tomono; 282 struct timespec tomono;
283 s64 nsec;
348 unsigned int seq; 284 unsigned int seq;
349 285
350 WARN_ON(timekeeping_suspended); 286 WARN_ON(timekeeping_suspended);
@@ -352,13 +288,14 @@ void ktime_get_ts(struct timespec *ts)
352 do { 288 do {
353 seq = read_seqbegin(&tk->lock); 289 seq = read_seqbegin(&tk->lock);
354 ts->tv_sec = tk->xtime_sec; 290 ts->tv_sec = tk->xtime_sec;
355 ts->tv_nsec = timekeeping_get_ns(tk); 291 nsec = timekeeping_get_ns(tk);
356 tomono = tk->wall_to_monotonic; 292 tomono = tk->wall_to_monotonic;
357 293
358 } while (read_seqretry(&tk->lock, seq)); 294 } while (read_seqretry(&tk->lock, seq));
359 295
360 set_normalized_timespec(ts, ts->tv_sec + tomono.tv_sec, 296 ts->tv_sec += tomono.tv_sec;
361 ts->tv_nsec + tomono.tv_nsec); 297 ts->tv_nsec = 0;
298 timespec_add_ns(ts, nsec + tomono.tv_nsec);
362} 299}
363EXPORT_SYMBOL_GPL(ktime_get_ts); 300EXPORT_SYMBOL_GPL(ktime_get_ts);
364 301
@@ -773,6 +710,7 @@ static void timekeeping_resume(void)
773 710
774 read_persistent_clock(&ts); 711 read_persistent_clock(&ts);
775 712
713 clockevents_resume();
776 clocksource_resume(); 714 clocksource_resume();
777 715
778 write_seqlock_irqsave(&tk->lock, flags); 716 write_seqlock_irqsave(&tk->lock, flags);
@@ -832,6 +770,7 @@ static int timekeeping_suspend(void)
832 770
833 clockevents_notify(CLOCK_EVT_NOTIFY_SUSPEND, NULL); 771 clockevents_notify(CLOCK_EVT_NOTIFY_SUSPEND, NULL);
834 clocksource_suspend(); 772 clocksource_suspend();
773 clockevents_suspend();
835 774
836 return 0; 775 return 0;
837} 776}
@@ -1108,7 +1047,7 @@ static cycle_t logarithmic_accumulation(struct timekeeper *tk, cycle_t offset,
1108 accumulate_nsecs_to_secs(tk); 1047 accumulate_nsecs_to_secs(tk);
1109 1048
1110 /* Accumulate raw time */ 1049 /* Accumulate raw time */
1111 raw_nsecs = tk->raw_interval << shift; 1050 raw_nsecs = (u64)tk->raw_interval << shift;
1112 raw_nsecs += tk->raw_time.tv_nsec; 1051 raw_nsecs += tk->raw_time.tv_nsec;
1113 if (raw_nsecs >= NSEC_PER_SEC) { 1052 if (raw_nsecs >= NSEC_PER_SEC) {
1114 u64 raw_secs = raw_nsecs; 1053 u64 raw_secs = raw_nsecs;
@@ -1125,6 +1064,33 @@ static cycle_t logarithmic_accumulation(struct timekeeper *tk, cycle_t offset,
1125 return offset; 1064 return offset;
1126} 1065}
1127 1066
1067#ifdef CONFIG_GENERIC_TIME_VSYSCALL_OLD
1068static inline void old_vsyscall_fixup(struct timekeeper *tk)
1069{
1070 s64 remainder;
1071
1072 /*
1073 * Store only full nanoseconds into xtime_nsec after rounding
1074 * it up and add the remainder to the error difference.
1075 * XXX - This is necessary to avoid small 1ns inconsistnecies caused
1076 * by truncating the remainder in vsyscalls. However, it causes
1077 * additional work to be done in timekeeping_adjust(). Once
1078 * the vsyscall implementations are converted to use xtime_nsec
1079 * (shifted nanoseconds), and CONFIG_GENERIC_TIME_VSYSCALL_OLD
1080 * users are removed, this can be killed.
1081 */
1082 remainder = tk->xtime_nsec & ((1ULL << tk->shift) - 1);
1083 tk->xtime_nsec -= remainder;
1084 tk->xtime_nsec += 1ULL << tk->shift;
1085 tk->ntp_error += remainder << tk->ntp_error_shift;
1086
1087}
1088#else
1089#define old_vsyscall_fixup(tk)
1090#endif
1091
1092
1093
1128/** 1094/**
1129 * update_wall_time - Uses the current clocksource to increment the wall time 1095 * update_wall_time - Uses the current clocksource to increment the wall time
1130 * 1096 *
@@ -1136,7 +1102,6 @@ static void update_wall_time(void)
1136 cycle_t offset; 1102 cycle_t offset;
1137 int shift = 0, maxshift; 1103 int shift = 0, maxshift;
1138 unsigned long flags; 1104 unsigned long flags;
1139 s64 remainder;
1140 1105
1141 write_seqlock_irqsave(&tk->lock, flags); 1106 write_seqlock_irqsave(&tk->lock, flags);
1142 1107
@@ -1178,20 +1143,11 @@ static void update_wall_time(void)
1178 /* correct the clock when NTP error is too big */ 1143 /* correct the clock when NTP error is too big */
1179 timekeeping_adjust(tk, offset); 1144 timekeeping_adjust(tk, offset);
1180 1145
1181
1182 /* 1146 /*
1183 * Store only full nanoseconds into xtime_nsec after rounding 1147 * XXX This can be killed once everyone converts
1184 * it up and add the remainder to the error difference. 1148 * to the new update_vsyscall.
1185 * XXX - This is necessary to avoid small 1ns inconsistnecies caused 1149 */
1186 * by truncating the remainder in vsyscalls. However, it causes 1150 old_vsyscall_fixup(tk);
1187 * additional work to be done in timekeeping_adjust(). Once
1188 * the vsyscall implementations are converted to use xtime_nsec
1189 * (shifted nanoseconds), this can be killed.
1190 */
1191 remainder = tk->xtime_nsec & ((1ULL << tk->shift) - 1);
1192 tk->xtime_nsec -= remainder;
1193 tk->xtime_nsec += 1ULL << tk->shift;
1194 tk->ntp_error += remainder << tk->ntp_error_shift;
1195 1151
1196 /* 1152 /*
1197 * Finally, make sure that after the rounding 1153 * Finally, make sure that after the rounding
@@ -1244,6 +1200,7 @@ void get_monotonic_boottime(struct timespec *ts)
1244{ 1200{
1245 struct timekeeper *tk = &timekeeper; 1201 struct timekeeper *tk = &timekeeper;
1246 struct timespec tomono, sleep; 1202 struct timespec tomono, sleep;
1203 s64 nsec;
1247 unsigned int seq; 1204 unsigned int seq;
1248 1205
1249 WARN_ON(timekeeping_suspended); 1206 WARN_ON(timekeeping_suspended);
@@ -1251,14 +1208,15 @@ void get_monotonic_boottime(struct timespec *ts)
1251 do { 1208 do {
1252 seq = read_seqbegin(&tk->lock); 1209 seq = read_seqbegin(&tk->lock);
1253 ts->tv_sec = tk->xtime_sec; 1210 ts->tv_sec = tk->xtime_sec;
1254 ts->tv_nsec = timekeeping_get_ns(tk); 1211 nsec = timekeeping_get_ns(tk);
1255 tomono = tk->wall_to_monotonic; 1212 tomono = tk->wall_to_monotonic;
1256 sleep = tk->total_sleep_time; 1213 sleep = tk->total_sleep_time;
1257 1214
1258 } while (read_seqretry(&tk->lock, seq)); 1215 } while (read_seqretry(&tk->lock, seq));
1259 1216
1260 set_normalized_timespec(ts, ts->tv_sec + tomono.tv_sec + sleep.tv_sec, 1217 ts->tv_sec += tomono.tv_sec + sleep.tv_sec;
1261 ts->tv_nsec + tomono.tv_nsec + sleep.tv_nsec); 1218 ts->tv_nsec = 0;
1219 timespec_add_ns(ts, nsec + tomono.tv_nsec + sleep.tv_nsec);
1262} 1220}
1263EXPORT_SYMBOL_GPL(get_monotonic_boottime); 1221EXPORT_SYMBOL_GPL(get_monotonic_boottime);
1264 1222
diff --git a/kernel/timer.c b/kernel/timer.c
index 8c5e7b908c68..367d00858482 100644
--- a/kernel/timer.c
+++ b/kernel/timer.c
@@ -63,6 +63,7 @@ EXPORT_SYMBOL(jiffies_64);
63#define TVR_SIZE (1 << TVR_BITS) 63#define TVR_SIZE (1 << TVR_BITS)
64#define TVN_MASK (TVN_SIZE - 1) 64#define TVN_MASK (TVN_SIZE - 1)
65#define TVR_MASK (TVR_SIZE - 1) 65#define TVR_MASK (TVR_SIZE - 1)
66#define MAX_TVAL ((unsigned long)((1ULL << (TVR_BITS + 4*TVN_BITS)) - 1))
66 67
67struct tvec { 68struct tvec {
68 struct list_head vec[TVN_SIZE]; 69 struct list_head vec[TVN_SIZE];
@@ -92,24 +93,25 @@ static DEFINE_PER_CPU(struct tvec_base *, tvec_bases) = &boot_tvec_bases;
92/* Functions below help us manage 'deferrable' flag */ 93/* Functions below help us manage 'deferrable' flag */
93static inline unsigned int tbase_get_deferrable(struct tvec_base *base) 94static inline unsigned int tbase_get_deferrable(struct tvec_base *base)
94{ 95{
95 return ((unsigned int)(unsigned long)base & TBASE_DEFERRABLE_FLAG); 96 return ((unsigned int)(unsigned long)base & TIMER_DEFERRABLE);
96} 97}
97 98
98static inline struct tvec_base *tbase_get_base(struct tvec_base *base) 99static inline unsigned int tbase_get_irqsafe(struct tvec_base *base)
99{ 100{
100 return ((struct tvec_base *)((unsigned long)base & ~TBASE_DEFERRABLE_FLAG)); 101 return ((unsigned int)(unsigned long)base & TIMER_IRQSAFE);
101} 102}
102 103
103static inline void timer_set_deferrable(struct timer_list *timer) 104static inline struct tvec_base *tbase_get_base(struct tvec_base *base)
104{ 105{
105 timer->base = TBASE_MAKE_DEFERRED(timer->base); 106 return ((struct tvec_base *)((unsigned long)base & ~TIMER_FLAG_MASK));
106} 107}
107 108
108static inline void 109static inline void
109timer_set_base(struct timer_list *timer, struct tvec_base *new_base) 110timer_set_base(struct timer_list *timer, struct tvec_base *new_base)
110{ 111{
111 timer->base = (struct tvec_base *)((unsigned long)(new_base) | 112 unsigned long flags = (unsigned long)timer->base & TIMER_FLAG_MASK;
112 tbase_get_deferrable(timer->base)); 113
114 timer->base = (struct tvec_base *)((unsigned long)(new_base) | flags);
113} 115}
114 116
115static unsigned long round_jiffies_common(unsigned long j, int cpu, 117static unsigned long round_jiffies_common(unsigned long j, int cpu,
@@ -358,11 +360,12 @@ __internal_add_timer(struct tvec_base *base, struct timer_list *timer)
358 vec = base->tv1.vec + (base->timer_jiffies & TVR_MASK); 360 vec = base->tv1.vec + (base->timer_jiffies & TVR_MASK);
359 } else { 361 } else {
360 int i; 362 int i;
361 /* If the timeout is larger than 0xffffffff on 64-bit 363 /* If the timeout is larger than MAX_TVAL (on 64-bit
362 * architectures then we use the maximum timeout: 364 * architectures or with CONFIG_BASE_SMALL=1) then we
365 * use the maximum timeout.
363 */ 366 */
364 if (idx > 0xffffffffUL) { 367 if (idx > MAX_TVAL) {
365 idx = 0xffffffffUL; 368 idx = MAX_TVAL;
366 expires = idx + base->timer_jiffies; 369 expires = idx + base->timer_jiffies;
367 } 370 }
368 i = (expires >> (TVR_BITS + 3 * TVN_BITS)) & TVN_MASK; 371 i = (expires >> (TVR_BITS + 3 * TVN_BITS)) & TVN_MASK;
@@ -563,16 +566,14 @@ static inline void debug_timer_assert_init(struct timer_list *timer)
563 debug_object_assert_init(timer, &timer_debug_descr); 566 debug_object_assert_init(timer, &timer_debug_descr);
564} 567}
565 568
566static void __init_timer(struct timer_list *timer, 569static void do_init_timer(struct timer_list *timer, unsigned int flags,
567 const char *name, 570 const char *name, struct lock_class_key *key);
568 struct lock_class_key *key);
569 571
570void init_timer_on_stack_key(struct timer_list *timer, 572void init_timer_on_stack_key(struct timer_list *timer, unsigned int flags,
571 const char *name, 573 const char *name, struct lock_class_key *key)
572 struct lock_class_key *key)
573{ 574{
574 debug_object_init_on_stack(timer, &timer_debug_descr); 575 debug_object_init_on_stack(timer, &timer_debug_descr);
575 __init_timer(timer, name, key); 576 do_init_timer(timer, flags, name, key);
576} 577}
577EXPORT_SYMBOL_GPL(init_timer_on_stack_key); 578EXPORT_SYMBOL_GPL(init_timer_on_stack_key);
578 579
@@ -613,12 +614,13 @@ static inline void debug_assert_init(struct timer_list *timer)
613 debug_timer_assert_init(timer); 614 debug_timer_assert_init(timer);
614} 615}
615 616
616static void __init_timer(struct timer_list *timer, 617static void do_init_timer(struct timer_list *timer, unsigned int flags,
617 const char *name, 618 const char *name, struct lock_class_key *key)
618 struct lock_class_key *key)
619{ 619{
620 struct tvec_base *base = __raw_get_cpu_var(tvec_bases);
621
620 timer->entry.next = NULL; 622 timer->entry.next = NULL;
621 timer->base = __raw_get_cpu_var(tvec_bases); 623 timer->base = (void *)((unsigned long)base | flags);
622 timer->slack = -1; 624 timer->slack = -1;
623#ifdef CONFIG_TIMER_STATS 625#ifdef CONFIG_TIMER_STATS
624 timer->start_site = NULL; 626 timer->start_site = NULL;
@@ -628,22 +630,10 @@ static void __init_timer(struct timer_list *timer,
628 lockdep_init_map(&timer->lockdep_map, name, key, 0); 630 lockdep_init_map(&timer->lockdep_map, name, key, 0);
629} 631}
630 632
631void setup_deferrable_timer_on_stack_key(struct timer_list *timer,
632 const char *name,
633 struct lock_class_key *key,
634 void (*function)(unsigned long),
635 unsigned long data)
636{
637 timer->function = function;
638 timer->data = data;
639 init_timer_on_stack_key(timer, name, key);
640 timer_set_deferrable(timer);
641}
642EXPORT_SYMBOL_GPL(setup_deferrable_timer_on_stack_key);
643
644/** 633/**
645 * init_timer_key - initialize a timer 634 * init_timer_key - initialize a timer
646 * @timer: the timer to be initialized 635 * @timer: the timer to be initialized
636 * @flags: timer flags
647 * @name: name of the timer 637 * @name: name of the timer
648 * @key: lockdep class key of the fake lock used for tracking timer 638 * @key: lockdep class key of the fake lock used for tracking timer
649 * sync lock dependencies 639 * sync lock dependencies
@@ -651,24 +641,14 @@ EXPORT_SYMBOL_GPL(setup_deferrable_timer_on_stack_key);
651 * init_timer_key() must be done to a timer prior calling *any* of the 641 * init_timer_key() must be done to a timer prior calling *any* of the
652 * other timer functions. 642 * other timer functions.
653 */ 643 */
654void init_timer_key(struct timer_list *timer, 644void init_timer_key(struct timer_list *timer, unsigned int flags,
655 const char *name, 645 const char *name, struct lock_class_key *key)
656 struct lock_class_key *key)
657{ 646{
658 debug_init(timer); 647 debug_init(timer);
659 __init_timer(timer, name, key); 648 do_init_timer(timer, flags, name, key);
660} 649}
661EXPORT_SYMBOL(init_timer_key); 650EXPORT_SYMBOL(init_timer_key);
662 651
663void init_timer_deferrable_key(struct timer_list *timer,
664 const char *name,
665 struct lock_class_key *key)
666{
667 init_timer_key(timer, name, key);
668 timer_set_deferrable(timer);
669}
670EXPORT_SYMBOL(init_timer_deferrable_key);
671
672static inline void detach_timer(struct timer_list *timer, bool clear_pending) 652static inline void detach_timer(struct timer_list *timer, bool clear_pending)
673{ 653{
674 struct list_head *entry = &timer->entry; 654 struct list_head *entry = &timer->entry;
@@ -686,7 +666,7 @@ detach_expired_timer(struct timer_list *timer, struct tvec_base *base)
686{ 666{
687 detach_timer(timer, true); 667 detach_timer(timer, true);
688 if (!tbase_get_deferrable(timer->base)) 668 if (!tbase_get_deferrable(timer->base))
689 timer->base->active_timers--; 669 base->active_timers--;
690} 670}
691 671
692static int detach_if_pending(struct timer_list *timer, struct tvec_base *base, 672static int detach_if_pending(struct timer_list *timer, struct tvec_base *base,
@@ -697,7 +677,7 @@ static int detach_if_pending(struct timer_list *timer, struct tvec_base *base,
697 677
698 detach_timer(timer, clear_pending); 678 detach_timer(timer, clear_pending);
699 if (!tbase_get_deferrable(timer->base)) { 679 if (!tbase_get_deferrable(timer->base)) {
700 timer->base->active_timers--; 680 base->active_timers--;
701 if (timer->expires == base->next_timer) 681 if (timer->expires == base->next_timer)
702 base->next_timer = base->timer_jiffies; 682 base->next_timer = base->timer_jiffies;
703 } 683 }
@@ -1029,14 +1009,14 @@ EXPORT_SYMBOL(try_to_del_timer_sync);
1029 * 1009 *
1030 * Synchronization rules: Callers must prevent restarting of the timer, 1010 * Synchronization rules: Callers must prevent restarting of the timer,
1031 * otherwise this function is meaningless. It must not be called from 1011 * otherwise this function is meaningless. It must not be called from
1032 * interrupt contexts. The caller must not hold locks which would prevent 1012 * interrupt contexts unless the timer is an irqsafe one. The caller must
1033 * completion of the timer's handler. The timer's handler must not call 1013 * not hold locks which would prevent completion of the timer's
1034 * add_timer_on(). Upon exit the timer is not queued and the handler is 1014 * handler. The timer's handler must not call add_timer_on(). Upon exit the
1035 * not running on any CPU. 1015 * timer is not queued and the handler is not running on any CPU.
1036 * 1016 *
1037 * Note: You must not hold locks that are held in interrupt context 1017 * Note: For !irqsafe timers, you must not hold locks that are held in
1038 * while calling this function. Even if the lock has nothing to do 1018 * interrupt context while calling this function. Even if the lock has
1039 * with the timer in question. Here's why: 1019 * nothing to do with the timer in question. Here's why:
1040 * 1020 *
1041 * CPU0 CPU1 1021 * CPU0 CPU1
1042 * ---- ---- 1022 * ---- ----
@@ -1073,7 +1053,7 @@ int del_timer_sync(struct timer_list *timer)
1073 * don't use it in hardirq context, because it 1053 * don't use it in hardirq context, because it
1074 * could lead to deadlock. 1054 * could lead to deadlock.
1075 */ 1055 */
1076 WARN_ON(in_irq()); 1056 WARN_ON(in_irq() && !tbase_get_irqsafe(timer->base));
1077 for (;;) { 1057 for (;;) {
1078 int ret = try_to_del_timer_sync(timer); 1058 int ret = try_to_del_timer_sync(timer);
1079 if (ret >= 0) 1059 if (ret >= 0)
@@ -1180,19 +1160,27 @@ static inline void __run_timers(struct tvec_base *base)
1180 while (!list_empty(head)) { 1160 while (!list_empty(head)) {
1181 void (*fn)(unsigned long); 1161 void (*fn)(unsigned long);
1182 unsigned long data; 1162 unsigned long data;
1163 bool irqsafe;
1183 1164
1184 timer = list_first_entry(head, struct timer_list,entry); 1165 timer = list_first_entry(head, struct timer_list,entry);
1185 fn = timer->function; 1166 fn = timer->function;
1186 data = timer->data; 1167 data = timer->data;
1168 irqsafe = tbase_get_irqsafe(timer->base);
1187 1169
1188 timer_stats_account_timer(timer); 1170 timer_stats_account_timer(timer);
1189 1171
1190 base->running_timer = timer; 1172 base->running_timer = timer;
1191 detach_expired_timer(timer, base); 1173 detach_expired_timer(timer, base);
1192 1174
1193 spin_unlock_irq(&base->lock); 1175 if (irqsafe) {
1194 call_timer_fn(timer, fn, data); 1176 spin_unlock(&base->lock);
1195 spin_lock_irq(&base->lock); 1177 call_timer_fn(timer, fn, data);
1178 spin_lock(&base->lock);
1179 } else {
1180 spin_unlock_irq(&base->lock);
1181 call_timer_fn(timer, fn, data);
1182 spin_lock_irq(&base->lock);
1183 }
1196 } 1184 }
1197 } 1185 }
1198 base->running_timer = NULL; 1186 base->running_timer = NULL;
@@ -1791,9 +1779,13 @@ static struct notifier_block __cpuinitdata timers_nb = {
1791 1779
1792void __init init_timers(void) 1780void __init init_timers(void)
1793{ 1781{
1794 int err = timer_cpu_notify(&timers_nb, (unsigned long)CPU_UP_PREPARE, 1782 int err;
1795 (void *)(long)smp_processor_id()); 1783
1784 /* ensure there are enough low bits for flags in timer->base pointer */
1785 BUILD_BUG_ON(__alignof__(struct tvec_base) & TIMER_FLAG_MASK);
1796 1786
1787 err = timer_cpu_notify(&timers_nb, (unsigned long)CPU_UP_PREPARE,
1788 (void *)(long)smp_processor_id());
1797 init_timer_stats(); 1789 init_timer_stats();
1798 1790
1799 BUG_ON(err != NOTIFY_OK); 1791 BUG_ON(err != NOTIFY_OK);
diff --git a/kernel/trace/Kconfig b/kernel/trace/Kconfig
index 8c4c07071cc5..4cea4f41c1d9 100644
--- a/kernel/trace/Kconfig
+++ b/kernel/trace/Kconfig
@@ -49,6 +49,11 @@ config HAVE_SYSCALL_TRACEPOINTS
49 help 49 help
50 See Documentation/trace/ftrace-design.txt 50 See Documentation/trace/ftrace-design.txt
51 51
52config HAVE_FENTRY
53 bool
54 help
55 Arch supports the gcc options -pg with -mfentry
56
52config HAVE_C_RECORDMCOUNT 57config HAVE_C_RECORDMCOUNT
53 bool 58 bool
54 help 59 help
@@ -57,8 +62,12 @@ config HAVE_C_RECORDMCOUNT
57config TRACER_MAX_TRACE 62config TRACER_MAX_TRACE
58 bool 63 bool
59 64
65config TRACE_CLOCK
66 bool
67
60config RING_BUFFER 68config RING_BUFFER
61 bool 69 bool
70 select TRACE_CLOCK
62 71
63config FTRACE_NMI_ENTER 72config FTRACE_NMI_ENTER
64 bool 73 bool
@@ -109,6 +118,7 @@ config TRACING
109 select NOP_TRACER 118 select NOP_TRACER
110 select BINARY_PRINTF 119 select BINARY_PRINTF
111 select EVENT_TRACING 120 select EVENT_TRACING
121 select TRACE_CLOCK
112 122
113config GENERIC_TRACER 123config GENERIC_TRACER
114 bool 124 bool
diff --git a/kernel/trace/Makefile b/kernel/trace/Makefile
index b831087c8200..d7e2068e4b71 100644
--- a/kernel/trace/Makefile
+++ b/kernel/trace/Makefile
@@ -5,10 +5,12 @@ ifdef CONFIG_FUNCTION_TRACER
5ORIG_CFLAGS := $(KBUILD_CFLAGS) 5ORIG_CFLAGS := $(KBUILD_CFLAGS)
6KBUILD_CFLAGS = $(subst -pg,,$(ORIG_CFLAGS)) 6KBUILD_CFLAGS = $(subst -pg,,$(ORIG_CFLAGS))
7 7
8ifdef CONFIG_FTRACE_SELFTEST
8# selftest needs instrumentation 9# selftest needs instrumentation
9CFLAGS_trace_selftest_dynamic.o = -pg 10CFLAGS_trace_selftest_dynamic.o = -pg
10obj-y += trace_selftest_dynamic.o 11obj-y += trace_selftest_dynamic.o
11endif 12endif
13endif
12 14
13# If unlikely tracing is enabled, do not trace these files 15# If unlikely tracing is enabled, do not trace these files
14ifdef CONFIG_TRACING_BRANCHES 16ifdef CONFIG_TRACING_BRANCHES
@@ -17,11 +19,7 @@ endif
17 19
18CFLAGS_trace_events_filter.o := -I$(src) 20CFLAGS_trace_events_filter.o := -I$(src)
19 21
20# 22obj-$(CONFIG_TRACE_CLOCK) += trace_clock.o
21# Make the trace clocks available generally: it's infrastructure
22# relied on by ptrace for example:
23#
24obj-y += trace_clock.o
25 23
26obj-$(CONFIG_FUNCTION_TRACER) += libftrace.o 24obj-$(CONFIG_FUNCTION_TRACER) += libftrace.o
27obj-$(CONFIG_RING_BUFFER) += ring_buffer.o 25obj-$(CONFIG_RING_BUFFER) += ring_buffer.o
diff --git a/kernel/trace/ftrace.c b/kernel/trace/ftrace.c
index b4f20fba09fc..9dcf15d38380 100644
--- a/kernel/trace/ftrace.c
+++ b/kernel/trace/ftrace.c
@@ -64,12 +64,20 @@
64 64
65#define FL_GLOBAL_CONTROL_MASK (FTRACE_OPS_FL_GLOBAL | FTRACE_OPS_FL_CONTROL) 65#define FL_GLOBAL_CONTROL_MASK (FTRACE_OPS_FL_GLOBAL | FTRACE_OPS_FL_CONTROL)
66 66
67static struct ftrace_ops ftrace_list_end __read_mostly = {
68 .func = ftrace_stub,
69 .flags = FTRACE_OPS_FL_RECURSION_SAFE,
70};
71
67/* ftrace_enabled is a method to turn ftrace on or off */ 72/* ftrace_enabled is a method to turn ftrace on or off */
68int ftrace_enabled __read_mostly; 73int ftrace_enabled __read_mostly;
69static int last_ftrace_enabled; 74static int last_ftrace_enabled;
70 75
71/* Quick disabling of function tracer. */ 76/* Quick disabling of function tracer. */
72int function_trace_stop; 77int function_trace_stop __read_mostly;
78
79/* Current function tracing op */
80struct ftrace_ops *function_trace_op __read_mostly = &ftrace_list_end;
73 81
74/* List for set_ftrace_pid's pids. */ 82/* List for set_ftrace_pid's pids. */
75LIST_HEAD(ftrace_pids); 83LIST_HEAD(ftrace_pids);
@@ -86,22 +94,43 @@ static int ftrace_disabled __read_mostly;
86 94
87static DEFINE_MUTEX(ftrace_lock); 95static DEFINE_MUTEX(ftrace_lock);
88 96
89static struct ftrace_ops ftrace_list_end __read_mostly = {
90 .func = ftrace_stub,
91};
92
93static struct ftrace_ops *ftrace_global_list __read_mostly = &ftrace_list_end; 97static struct ftrace_ops *ftrace_global_list __read_mostly = &ftrace_list_end;
94static struct ftrace_ops *ftrace_control_list __read_mostly = &ftrace_list_end; 98static struct ftrace_ops *ftrace_control_list __read_mostly = &ftrace_list_end;
95static struct ftrace_ops *ftrace_ops_list __read_mostly = &ftrace_list_end; 99static struct ftrace_ops *ftrace_ops_list __read_mostly = &ftrace_list_end;
96ftrace_func_t ftrace_trace_function __read_mostly = ftrace_stub; 100ftrace_func_t ftrace_trace_function __read_mostly = ftrace_stub;
97static ftrace_func_t __ftrace_trace_function_delay __read_mostly = ftrace_stub;
98ftrace_func_t __ftrace_trace_function __read_mostly = ftrace_stub;
99ftrace_func_t ftrace_pid_function __read_mostly = ftrace_stub; 101ftrace_func_t ftrace_pid_function __read_mostly = ftrace_stub;
100static struct ftrace_ops global_ops; 102static struct ftrace_ops global_ops;
101static struct ftrace_ops control_ops; 103static struct ftrace_ops control_ops;
102 104
103static void 105#if ARCH_SUPPORTS_FTRACE_OPS
104ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip); 106static void ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip,
107 struct ftrace_ops *op, struct pt_regs *regs);
108#else
109/* See comment below, where ftrace_ops_list_func is defined */
110static void ftrace_ops_no_ops(unsigned long ip, unsigned long parent_ip);
111#define ftrace_ops_list_func ((ftrace_func_t)ftrace_ops_no_ops)
112#endif
113
114/**
115 * ftrace_nr_registered_ops - return number of ops registered
116 *
117 * Returns the number of ftrace_ops registered and tracing functions
118 */
119int ftrace_nr_registered_ops(void)
120{
121 struct ftrace_ops *ops;
122 int cnt = 0;
123
124 mutex_lock(&ftrace_lock);
125
126 for (ops = ftrace_ops_list;
127 ops != &ftrace_list_end; ops = ops->next)
128 cnt++;
129
130 mutex_unlock(&ftrace_lock);
131
132 return cnt;
133}
105 134
106/* 135/*
107 * Traverse the ftrace_global_list, invoking all entries. The reason that we 136 * Traverse the ftrace_global_list, invoking all entries. The reason that we
@@ -112,29 +141,29 @@ ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip);
112 * 141 *
113 * Silly Alpha and silly pointer-speculation compiler optimizations! 142 * Silly Alpha and silly pointer-speculation compiler optimizations!
114 */ 143 */
115static void ftrace_global_list_func(unsigned long ip, 144static void
116 unsigned long parent_ip) 145ftrace_global_list_func(unsigned long ip, unsigned long parent_ip,
146 struct ftrace_ops *op, struct pt_regs *regs)
117{ 147{
118 struct ftrace_ops *op;
119
120 if (unlikely(trace_recursion_test(TRACE_GLOBAL_BIT))) 148 if (unlikely(trace_recursion_test(TRACE_GLOBAL_BIT)))
121 return; 149 return;
122 150
123 trace_recursion_set(TRACE_GLOBAL_BIT); 151 trace_recursion_set(TRACE_GLOBAL_BIT);
124 op = rcu_dereference_raw(ftrace_global_list); /*see above*/ 152 op = rcu_dereference_raw(ftrace_global_list); /*see above*/
125 while (op != &ftrace_list_end) { 153 while (op != &ftrace_list_end) {
126 op->func(ip, parent_ip); 154 op->func(ip, parent_ip, op, regs);
127 op = rcu_dereference_raw(op->next); /*see above*/ 155 op = rcu_dereference_raw(op->next); /*see above*/
128 }; 156 };
129 trace_recursion_clear(TRACE_GLOBAL_BIT); 157 trace_recursion_clear(TRACE_GLOBAL_BIT);
130} 158}
131 159
132static void ftrace_pid_func(unsigned long ip, unsigned long parent_ip) 160static void ftrace_pid_func(unsigned long ip, unsigned long parent_ip,
161 struct ftrace_ops *op, struct pt_regs *regs)
133{ 162{
134 if (!test_tsk_trace_trace(current)) 163 if (!test_tsk_trace_trace(current))
135 return; 164 return;
136 165
137 ftrace_pid_function(ip, parent_ip); 166 ftrace_pid_function(ip, parent_ip, op, regs);
138} 167}
139 168
140static void set_ftrace_pid_function(ftrace_func_t func) 169static void set_ftrace_pid_function(ftrace_func_t func)
@@ -153,25 +182,9 @@ static void set_ftrace_pid_function(ftrace_func_t func)
153void clear_ftrace_function(void) 182void clear_ftrace_function(void)
154{ 183{
155 ftrace_trace_function = ftrace_stub; 184 ftrace_trace_function = ftrace_stub;
156 __ftrace_trace_function = ftrace_stub;
157 __ftrace_trace_function_delay = ftrace_stub;
158 ftrace_pid_function = ftrace_stub; 185 ftrace_pid_function = ftrace_stub;
159} 186}
160 187
161#ifndef CONFIG_HAVE_FUNCTION_TRACE_MCOUNT_TEST
162/*
163 * For those archs that do not test ftrace_trace_stop in their
164 * mcount call site, we need to do it from C.
165 */
166static void ftrace_test_stop_func(unsigned long ip, unsigned long parent_ip)
167{
168 if (function_trace_stop)
169 return;
170
171 __ftrace_trace_function(ip, parent_ip);
172}
173#endif
174
175static void control_ops_disable_all(struct ftrace_ops *ops) 188static void control_ops_disable_all(struct ftrace_ops *ops)
176{ 189{
177 int cpu; 190 int cpu;
@@ -230,28 +243,27 @@ static void update_ftrace_function(void)
230 243
231 /* 244 /*
232 * If we are at the end of the list and this ops is 245 * If we are at the end of the list and this ops is
233 * not dynamic, then have the mcount trampoline call 246 * recursion safe and not dynamic and the arch supports passing ops,
234 * the function directly 247 * then have the mcount trampoline call the function directly.
235 */ 248 */
236 if (ftrace_ops_list == &ftrace_list_end || 249 if (ftrace_ops_list == &ftrace_list_end ||
237 (ftrace_ops_list->next == &ftrace_list_end && 250 (ftrace_ops_list->next == &ftrace_list_end &&
238 !(ftrace_ops_list->flags & FTRACE_OPS_FL_DYNAMIC))) 251 !(ftrace_ops_list->flags & FTRACE_OPS_FL_DYNAMIC) &&
252 (ftrace_ops_list->flags & FTRACE_OPS_FL_RECURSION_SAFE) &&
253 !FTRACE_FORCE_LIST_FUNC)) {
254 /* Set the ftrace_ops that the arch callback uses */
255 if (ftrace_ops_list == &global_ops)
256 function_trace_op = ftrace_global_list;
257 else
258 function_trace_op = ftrace_ops_list;
239 func = ftrace_ops_list->func; 259 func = ftrace_ops_list->func;
240 else 260 } else {
261 /* Just use the default ftrace_ops */
262 function_trace_op = &ftrace_list_end;
241 func = ftrace_ops_list_func; 263 func = ftrace_ops_list_func;
264 }
242 265
243#ifdef CONFIG_HAVE_FUNCTION_TRACE_MCOUNT_TEST
244 ftrace_trace_function = func; 266 ftrace_trace_function = func;
245#else
246#ifdef CONFIG_DYNAMIC_FTRACE
247 /* do not update till all functions have been modified */
248 __ftrace_trace_function_delay = func;
249#else
250 __ftrace_trace_function = func;
251#endif
252 ftrace_trace_function =
253 (func == ftrace_stub) ? func : ftrace_test_stop_func;
254#endif
255} 267}
256 268
257static void add_ftrace_ops(struct ftrace_ops **list, struct ftrace_ops *ops) 269static void add_ftrace_ops(struct ftrace_ops **list, struct ftrace_ops *ops)
@@ -325,6 +337,20 @@ static int __register_ftrace_function(struct ftrace_ops *ops)
325 if ((ops->flags & FL_GLOBAL_CONTROL_MASK) == FL_GLOBAL_CONTROL_MASK) 337 if ((ops->flags & FL_GLOBAL_CONTROL_MASK) == FL_GLOBAL_CONTROL_MASK)
326 return -EINVAL; 338 return -EINVAL;
327 339
340#ifndef ARCH_SUPPORTS_FTRACE_SAVE_REGS
341 /*
342 * If the ftrace_ops specifies SAVE_REGS, then it only can be used
343 * if the arch supports it, or SAVE_REGS_IF_SUPPORTED is also set.
344 * Setting SAVE_REGS_IF_SUPPORTED makes SAVE_REGS irrelevant.
345 */
346 if (ops->flags & FTRACE_OPS_FL_SAVE_REGS &&
347 !(ops->flags & FTRACE_OPS_FL_SAVE_REGS_IF_SUPPORTED))
348 return -EINVAL;
349
350 if (ops->flags & FTRACE_OPS_FL_SAVE_REGS_IF_SUPPORTED)
351 ops->flags |= FTRACE_OPS_FL_SAVE_REGS;
352#endif
353
328 if (!core_kernel_data((unsigned long)ops)) 354 if (!core_kernel_data((unsigned long)ops))
329 ops->flags |= FTRACE_OPS_FL_DYNAMIC; 355 ops->flags |= FTRACE_OPS_FL_DYNAMIC;
330 356
@@ -773,7 +799,8 @@ ftrace_profile_alloc(struct ftrace_profile_stat *stat, unsigned long ip)
773} 799}
774 800
775static void 801static void
776function_profile_call(unsigned long ip, unsigned long parent_ip) 802function_profile_call(unsigned long ip, unsigned long parent_ip,
803 struct ftrace_ops *ops, struct pt_regs *regs)
777{ 804{
778 struct ftrace_profile_stat *stat; 805 struct ftrace_profile_stat *stat;
779 struct ftrace_profile *rec; 806 struct ftrace_profile *rec;
@@ -803,7 +830,7 @@ function_profile_call(unsigned long ip, unsigned long parent_ip)
803#ifdef CONFIG_FUNCTION_GRAPH_TRACER 830#ifdef CONFIG_FUNCTION_GRAPH_TRACER
804static int profile_graph_entry(struct ftrace_graph_ent *trace) 831static int profile_graph_entry(struct ftrace_graph_ent *trace)
805{ 832{
806 function_profile_call(trace->func, 0); 833 function_profile_call(trace->func, 0, NULL, NULL);
807 return 1; 834 return 1;
808} 835}
809 836
@@ -863,6 +890,7 @@ static void unregister_ftrace_profiler(void)
863#else 890#else
864static struct ftrace_ops ftrace_profile_ops __read_mostly = { 891static struct ftrace_ops ftrace_profile_ops __read_mostly = {
865 .func = function_profile_call, 892 .func = function_profile_call,
893 .flags = FTRACE_OPS_FL_RECURSION_SAFE,
866}; 894};
867 895
868static int register_ftrace_profiler(void) 896static int register_ftrace_profiler(void)
@@ -1045,6 +1073,7 @@ static struct ftrace_ops global_ops = {
1045 .func = ftrace_stub, 1073 .func = ftrace_stub,
1046 .notrace_hash = EMPTY_HASH, 1074 .notrace_hash = EMPTY_HASH,
1047 .filter_hash = EMPTY_HASH, 1075 .filter_hash = EMPTY_HASH,
1076 .flags = FTRACE_OPS_FL_RECURSION_SAFE,
1048}; 1077};
1049 1078
1050static DEFINE_MUTEX(ftrace_regex_lock); 1079static DEFINE_MUTEX(ftrace_regex_lock);
@@ -1525,6 +1554,12 @@ static void __ftrace_hash_rec_update(struct ftrace_ops *ops,
1525 rec->flags++; 1554 rec->flags++;
1526 if (FTRACE_WARN_ON((rec->flags & ~FTRACE_FL_MASK) == FTRACE_REF_MAX)) 1555 if (FTRACE_WARN_ON((rec->flags & ~FTRACE_FL_MASK) == FTRACE_REF_MAX))
1527 return; 1556 return;
1557 /*
1558 * If any ops wants regs saved for this function
1559 * then all ops will get saved regs.
1560 */
1561 if (ops->flags & FTRACE_OPS_FL_SAVE_REGS)
1562 rec->flags |= FTRACE_FL_REGS;
1528 } else { 1563 } else {
1529 if (FTRACE_WARN_ON((rec->flags & ~FTRACE_FL_MASK) == 0)) 1564 if (FTRACE_WARN_ON((rec->flags & ~FTRACE_FL_MASK) == 0))
1530 return; 1565 return;
@@ -1616,18 +1651,59 @@ static int ftrace_check_record(struct dyn_ftrace *rec, int enable, int update)
1616 if (enable && (rec->flags & ~FTRACE_FL_MASK)) 1651 if (enable && (rec->flags & ~FTRACE_FL_MASK))
1617 flag = FTRACE_FL_ENABLED; 1652 flag = FTRACE_FL_ENABLED;
1618 1653
1654 /*
1655 * If enabling and the REGS flag does not match the REGS_EN, then
1656 * do not ignore this record. Set flags to fail the compare against
1657 * ENABLED.
1658 */
1659 if (flag &&
1660 (!(rec->flags & FTRACE_FL_REGS) != !(rec->flags & FTRACE_FL_REGS_EN)))
1661 flag |= FTRACE_FL_REGS;
1662
1619 /* If the state of this record hasn't changed, then do nothing */ 1663 /* If the state of this record hasn't changed, then do nothing */
1620 if ((rec->flags & FTRACE_FL_ENABLED) == flag) 1664 if ((rec->flags & FTRACE_FL_ENABLED) == flag)
1621 return FTRACE_UPDATE_IGNORE; 1665 return FTRACE_UPDATE_IGNORE;
1622 1666
1623 if (flag) { 1667 if (flag) {
1624 if (update) 1668 /* Save off if rec is being enabled (for return value) */
1669 flag ^= rec->flags & FTRACE_FL_ENABLED;
1670
1671 if (update) {
1625 rec->flags |= FTRACE_FL_ENABLED; 1672 rec->flags |= FTRACE_FL_ENABLED;
1626 return FTRACE_UPDATE_MAKE_CALL; 1673 if (flag & FTRACE_FL_REGS) {
1674 if (rec->flags & FTRACE_FL_REGS)
1675 rec->flags |= FTRACE_FL_REGS_EN;
1676 else
1677 rec->flags &= ~FTRACE_FL_REGS_EN;
1678 }
1679 }
1680
1681 /*
1682 * If this record is being updated from a nop, then
1683 * return UPDATE_MAKE_CALL.
1684 * Otherwise, if the EN flag is set, then return
1685 * UPDATE_MODIFY_CALL_REGS to tell the caller to convert
1686 * from the non-save regs, to a save regs function.
1687 * Otherwise,
1688 * return UPDATE_MODIFY_CALL to tell the caller to convert
1689 * from the save regs, to a non-save regs function.
1690 */
1691 if (flag & FTRACE_FL_ENABLED)
1692 return FTRACE_UPDATE_MAKE_CALL;
1693 else if (rec->flags & FTRACE_FL_REGS_EN)
1694 return FTRACE_UPDATE_MODIFY_CALL_REGS;
1695 else
1696 return FTRACE_UPDATE_MODIFY_CALL;
1627 } 1697 }
1628 1698
1629 if (update) 1699 if (update) {
1630 rec->flags &= ~FTRACE_FL_ENABLED; 1700 /* If there's no more users, clear all flags */
1701 if (!(rec->flags & ~FTRACE_FL_MASK))
1702 rec->flags = 0;
1703 else
1704 /* Just disable the record (keep REGS state) */
1705 rec->flags &= ~FTRACE_FL_ENABLED;
1706 }
1631 1707
1632 return FTRACE_UPDATE_MAKE_NOP; 1708 return FTRACE_UPDATE_MAKE_NOP;
1633} 1709}
@@ -1662,13 +1738,17 @@ int ftrace_test_record(struct dyn_ftrace *rec, int enable)
1662static int 1738static int
1663__ftrace_replace_code(struct dyn_ftrace *rec, int enable) 1739__ftrace_replace_code(struct dyn_ftrace *rec, int enable)
1664{ 1740{
1741 unsigned long ftrace_old_addr;
1665 unsigned long ftrace_addr; 1742 unsigned long ftrace_addr;
1666 int ret; 1743 int ret;
1667 1744
1668 ftrace_addr = (unsigned long)FTRACE_ADDR;
1669
1670 ret = ftrace_update_record(rec, enable); 1745 ret = ftrace_update_record(rec, enable);
1671 1746
1747 if (rec->flags & FTRACE_FL_REGS)
1748 ftrace_addr = (unsigned long)FTRACE_REGS_ADDR;
1749 else
1750 ftrace_addr = (unsigned long)FTRACE_ADDR;
1751
1672 switch (ret) { 1752 switch (ret) {
1673 case FTRACE_UPDATE_IGNORE: 1753 case FTRACE_UPDATE_IGNORE:
1674 return 0; 1754 return 0;
@@ -1678,6 +1758,15 @@ __ftrace_replace_code(struct dyn_ftrace *rec, int enable)
1678 1758
1679 case FTRACE_UPDATE_MAKE_NOP: 1759 case FTRACE_UPDATE_MAKE_NOP:
1680 return ftrace_make_nop(NULL, rec, ftrace_addr); 1760 return ftrace_make_nop(NULL, rec, ftrace_addr);
1761
1762 case FTRACE_UPDATE_MODIFY_CALL_REGS:
1763 case FTRACE_UPDATE_MODIFY_CALL:
1764 if (rec->flags & FTRACE_FL_REGS)
1765 ftrace_old_addr = (unsigned long)FTRACE_ADDR;
1766 else
1767 ftrace_old_addr = (unsigned long)FTRACE_REGS_ADDR;
1768
1769 return ftrace_modify_call(rec, ftrace_old_addr, ftrace_addr);
1681 } 1770 }
1682 1771
1683 return -1; /* unknow ftrace bug */ 1772 return -1; /* unknow ftrace bug */
@@ -1882,16 +1971,6 @@ static void ftrace_run_update_code(int command)
1882 */ 1971 */
1883 arch_ftrace_update_code(command); 1972 arch_ftrace_update_code(command);
1884 1973
1885#ifndef CONFIG_HAVE_FUNCTION_TRACE_MCOUNT_TEST
1886 /*
1887 * For archs that call ftrace_test_stop_func(), we must
1888 * wait till after we update all the function callers
1889 * before we update the callback. This keeps different
1890 * ops that record different functions from corrupting
1891 * each other.
1892 */
1893 __ftrace_trace_function = __ftrace_trace_function_delay;
1894#endif
1895 function_trace_stop--; 1974 function_trace_stop--;
1896 1975
1897 ret = ftrace_arch_code_modify_post_process(); 1976 ret = ftrace_arch_code_modify_post_process();
@@ -2441,8 +2520,9 @@ static int t_show(struct seq_file *m, void *v)
2441 2520
2442 seq_printf(m, "%ps", (void *)rec->ip); 2521 seq_printf(m, "%ps", (void *)rec->ip);
2443 if (iter->flags & FTRACE_ITER_ENABLED) 2522 if (iter->flags & FTRACE_ITER_ENABLED)
2444 seq_printf(m, " (%ld)", 2523 seq_printf(m, " (%ld)%s",
2445 rec->flags & ~FTRACE_FL_MASK); 2524 rec->flags & ~FTRACE_FL_MASK,
2525 rec->flags & FTRACE_FL_REGS ? " R" : "");
2446 seq_printf(m, "\n"); 2526 seq_printf(m, "\n");
2447 2527
2448 return 0; 2528 return 0;
@@ -2790,8 +2870,8 @@ static int __init ftrace_mod_cmd_init(void)
2790} 2870}
2791device_initcall(ftrace_mod_cmd_init); 2871device_initcall(ftrace_mod_cmd_init);
2792 2872
2793static void 2873static void function_trace_probe_call(unsigned long ip, unsigned long parent_ip,
2794function_trace_probe_call(unsigned long ip, unsigned long parent_ip) 2874 struct ftrace_ops *op, struct pt_regs *pt_regs)
2795{ 2875{
2796 struct ftrace_func_probe *entry; 2876 struct ftrace_func_probe *entry;
2797 struct hlist_head *hhd; 2877 struct hlist_head *hhd;
@@ -3162,8 +3242,27 @@ ftrace_notrace_write(struct file *file, const char __user *ubuf,
3162} 3242}
3163 3243
3164static int 3244static int
3165ftrace_set_regex(struct ftrace_ops *ops, unsigned char *buf, int len, 3245ftrace_match_addr(struct ftrace_hash *hash, unsigned long ip, int remove)
3166 int reset, int enable) 3246{
3247 struct ftrace_func_entry *entry;
3248
3249 if (!ftrace_location(ip))
3250 return -EINVAL;
3251
3252 if (remove) {
3253 entry = ftrace_lookup_ip(hash, ip);
3254 if (!entry)
3255 return -ENOENT;
3256 free_hash_entry(hash, entry);
3257 return 0;
3258 }
3259
3260 return add_hash_entry(hash, ip);
3261}
3262
3263static int
3264ftrace_set_hash(struct ftrace_ops *ops, unsigned char *buf, int len,
3265 unsigned long ip, int remove, int reset, int enable)
3167{ 3266{
3168 struct ftrace_hash **orig_hash; 3267 struct ftrace_hash **orig_hash;
3169 struct ftrace_hash *hash; 3268 struct ftrace_hash *hash;
@@ -3192,6 +3291,11 @@ ftrace_set_regex(struct ftrace_ops *ops, unsigned char *buf, int len,
3192 ret = -EINVAL; 3291 ret = -EINVAL;
3193 goto out_regex_unlock; 3292 goto out_regex_unlock;
3194 } 3293 }
3294 if (ip) {
3295 ret = ftrace_match_addr(hash, ip, remove);
3296 if (ret < 0)
3297 goto out_regex_unlock;
3298 }
3195 3299
3196 mutex_lock(&ftrace_lock); 3300 mutex_lock(&ftrace_lock);
3197 ret = ftrace_hash_move(ops, enable, orig_hash, hash); 3301 ret = ftrace_hash_move(ops, enable, orig_hash, hash);
@@ -3208,6 +3312,37 @@ ftrace_set_regex(struct ftrace_ops *ops, unsigned char *buf, int len,
3208 return ret; 3312 return ret;
3209} 3313}
3210 3314
3315static int
3316ftrace_set_addr(struct ftrace_ops *ops, unsigned long ip, int remove,
3317 int reset, int enable)
3318{
3319 return ftrace_set_hash(ops, 0, 0, ip, remove, reset, enable);
3320}
3321
3322/**
3323 * ftrace_set_filter_ip - set a function to filter on in ftrace by address
3324 * @ops - the ops to set the filter with
3325 * @ip - the address to add to or remove from the filter.
3326 * @remove - non zero to remove the ip from the filter
3327 * @reset - non zero to reset all filters before applying this filter.
3328 *
3329 * Filters denote which functions should be enabled when tracing is enabled
3330 * If @ip is NULL, it failes to update filter.
3331 */
3332int ftrace_set_filter_ip(struct ftrace_ops *ops, unsigned long ip,
3333 int remove, int reset)
3334{
3335 return ftrace_set_addr(ops, ip, remove, reset, 1);
3336}
3337EXPORT_SYMBOL_GPL(ftrace_set_filter_ip);
3338
3339static int
3340ftrace_set_regex(struct ftrace_ops *ops, unsigned char *buf, int len,
3341 int reset, int enable)
3342{
3343 return ftrace_set_hash(ops, buf, len, 0, 0, reset, enable);
3344}
3345
3211/** 3346/**
3212 * ftrace_set_filter - set a function to filter on in ftrace 3347 * ftrace_set_filter - set a function to filter on in ftrace
3213 * @ops - the ops to set the filter with 3348 * @ops - the ops to set the filter with
@@ -3912,6 +4047,7 @@ void __init ftrace_init(void)
3912 4047
3913static struct ftrace_ops global_ops = { 4048static struct ftrace_ops global_ops = {
3914 .func = ftrace_stub, 4049 .func = ftrace_stub,
4050 .flags = FTRACE_OPS_FL_RECURSION_SAFE,
3915}; 4051};
3916 4052
3917static int __init ftrace_nodyn_init(void) 4053static int __init ftrace_nodyn_init(void)
@@ -3942,10 +4078,9 @@ ftrace_ops_test(struct ftrace_ops *ops, unsigned long ip)
3942#endif /* CONFIG_DYNAMIC_FTRACE */ 4078#endif /* CONFIG_DYNAMIC_FTRACE */
3943 4079
3944static void 4080static void
3945ftrace_ops_control_func(unsigned long ip, unsigned long parent_ip) 4081ftrace_ops_control_func(unsigned long ip, unsigned long parent_ip,
4082 struct ftrace_ops *op, struct pt_regs *regs)
3946{ 4083{
3947 struct ftrace_ops *op;
3948
3949 if (unlikely(trace_recursion_test(TRACE_CONTROL_BIT))) 4084 if (unlikely(trace_recursion_test(TRACE_CONTROL_BIT)))
3950 return; 4085 return;
3951 4086
@@ -3959,7 +4094,7 @@ ftrace_ops_control_func(unsigned long ip, unsigned long parent_ip)
3959 while (op != &ftrace_list_end) { 4094 while (op != &ftrace_list_end) {
3960 if (!ftrace_function_local_disabled(op) && 4095 if (!ftrace_function_local_disabled(op) &&
3961 ftrace_ops_test(op, ip)) 4096 ftrace_ops_test(op, ip))
3962 op->func(ip, parent_ip); 4097 op->func(ip, parent_ip, op, regs);
3963 4098
3964 op = rcu_dereference_raw(op->next); 4099 op = rcu_dereference_raw(op->next);
3965 }; 4100 };
@@ -3969,13 +4104,18 @@ ftrace_ops_control_func(unsigned long ip, unsigned long parent_ip)
3969 4104
3970static struct ftrace_ops control_ops = { 4105static struct ftrace_ops control_ops = {
3971 .func = ftrace_ops_control_func, 4106 .func = ftrace_ops_control_func,
4107 .flags = FTRACE_OPS_FL_RECURSION_SAFE,
3972}; 4108};
3973 4109
3974static void 4110static inline void
3975ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip) 4111__ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip,
4112 struct ftrace_ops *ignored, struct pt_regs *regs)
3976{ 4113{
3977 struct ftrace_ops *op; 4114 struct ftrace_ops *op;
3978 4115
4116 if (function_trace_stop)
4117 return;
4118
3979 if (unlikely(trace_recursion_test(TRACE_INTERNAL_BIT))) 4119 if (unlikely(trace_recursion_test(TRACE_INTERNAL_BIT)))
3980 return; 4120 return;
3981 4121
@@ -3988,13 +4128,39 @@ ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip)
3988 op = rcu_dereference_raw(ftrace_ops_list); 4128 op = rcu_dereference_raw(ftrace_ops_list);
3989 while (op != &ftrace_list_end) { 4129 while (op != &ftrace_list_end) {
3990 if (ftrace_ops_test(op, ip)) 4130 if (ftrace_ops_test(op, ip))
3991 op->func(ip, parent_ip); 4131 op->func(ip, parent_ip, op, regs);
3992 op = rcu_dereference_raw(op->next); 4132 op = rcu_dereference_raw(op->next);
3993 }; 4133 };
3994 preempt_enable_notrace(); 4134 preempt_enable_notrace();
3995 trace_recursion_clear(TRACE_INTERNAL_BIT); 4135 trace_recursion_clear(TRACE_INTERNAL_BIT);
3996} 4136}
3997 4137
4138/*
4139 * Some archs only support passing ip and parent_ip. Even though
4140 * the list function ignores the op parameter, we do not want any
4141 * C side effects, where a function is called without the caller
4142 * sending a third parameter.
4143 * Archs are to support both the regs and ftrace_ops at the same time.
4144 * If they support ftrace_ops, it is assumed they support regs.
4145 * If call backs want to use regs, they must either check for regs
4146 * being NULL, or ARCH_SUPPORTS_FTRACE_SAVE_REGS.
4147 * Note, ARCH_SUPPORT_SAVE_REGS expects a full regs to be saved.
4148 * An architecture can pass partial regs with ftrace_ops and still
4149 * set the ARCH_SUPPORT_FTARCE_OPS.
4150 */
4151#if ARCH_SUPPORTS_FTRACE_OPS
4152static void ftrace_ops_list_func(unsigned long ip, unsigned long parent_ip,
4153 struct ftrace_ops *op, struct pt_regs *regs)
4154{
4155 __ftrace_ops_list_func(ip, parent_ip, NULL, regs);
4156}
4157#else
4158static void ftrace_ops_no_ops(unsigned long ip, unsigned long parent_ip)
4159{
4160 __ftrace_ops_list_func(ip, parent_ip, NULL, NULL);
4161}
4162#endif
4163
3998static void clear_ftrace_swapper(void) 4164static void clear_ftrace_swapper(void)
3999{ 4165{
4000 struct task_struct *p; 4166 struct task_struct *p;
diff --git a/kernel/trace/ring_buffer.c b/kernel/trace/ring_buffer.c
index 49491fa7daa2..b32ed0e385a5 100644
--- a/kernel/trace/ring_buffer.c
+++ b/kernel/trace/ring_buffer.c
@@ -2816,7 +2816,7 @@ EXPORT_SYMBOL_GPL(ring_buffer_record_enable);
2816 * to the buffer after this will fail and return NULL. 2816 * to the buffer after this will fail and return NULL.
2817 * 2817 *
2818 * This is different than ring_buffer_record_disable() as 2818 * This is different than ring_buffer_record_disable() as
2819 * it works like an on/off switch, where as the disable() verison 2819 * it works like an on/off switch, where as the disable() version
2820 * must be paired with a enable(). 2820 * must be paired with a enable().
2821 */ 2821 */
2822void ring_buffer_record_off(struct ring_buffer *buffer) 2822void ring_buffer_record_off(struct ring_buffer *buffer)
@@ -2839,7 +2839,7 @@ EXPORT_SYMBOL_GPL(ring_buffer_record_off);
2839 * ring_buffer_record_off(). 2839 * ring_buffer_record_off().
2840 * 2840 *
2841 * This is different than ring_buffer_record_enable() as 2841 * This is different than ring_buffer_record_enable() as
2842 * it works like an on/off switch, where as the enable() verison 2842 * it works like an on/off switch, where as the enable() version
2843 * must be paired with a disable(). 2843 * must be paired with a disable().
2844 */ 2844 */
2845void ring_buffer_record_on(struct ring_buffer *buffer) 2845void ring_buffer_record_on(struct ring_buffer *buffer)
diff --git a/kernel/trace/trace.c b/kernel/trace/trace.c
index 5c38c81496ce..31e4f55773f1 100644
--- a/kernel/trace/trace.c
+++ b/kernel/trace/trace.c
@@ -328,7 +328,7 @@ static DECLARE_WAIT_QUEUE_HEAD(trace_wait);
328unsigned long trace_flags = TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK | 328unsigned long trace_flags = TRACE_ITER_PRINT_PARENT | TRACE_ITER_PRINTK |
329 TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO | TRACE_ITER_SLEEP_TIME | 329 TRACE_ITER_ANNOTATE | TRACE_ITER_CONTEXT_INFO | TRACE_ITER_SLEEP_TIME |
330 TRACE_ITER_GRAPH_TIME | TRACE_ITER_RECORD_CMD | TRACE_ITER_OVERWRITE | 330 TRACE_ITER_GRAPH_TIME | TRACE_ITER_RECORD_CMD | TRACE_ITER_OVERWRITE |
331 TRACE_ITER_IRQ_INFO; 331 TRACE_ITER_IRQ_INFO | TRACE_ITER_MARKERS;
332 332
333static int trace_stop_count; 333static int trace_stop_count;
334static DEFINE_RAW_SPINLOCK(tracing_start_lock); 334static DEFINE_RAW_SPINLOCK(tracing_start_lock);
@@ -426,15 +426,15 @@ __setup("trace_buf_size=", set_buf_size);
426 426
427static int __init set_tracing_thresh(char *str) 427static int __init set_tracing_thresh(char *str)
428{ 428{
429 unsigned long threshhold; 429 unsigned long threshold;
430 int ret; 430 int ret;
431 431
432 if (!str) 432 if (!str)
433 return 0; 433 return 0;
434 ret = strict_strtoul(str, 0, &threshhold); 434 ret = strict_strtoul(str, 0, &threshold);
435 if (ret < 0) 435 if (ret < 0)
436 return 0; 436 return 0;
437 tracing_thresh = threshhold * 1000; 437 tracing_thresh = threshold * 1000;
438 return 1; 438 return 1;
439} 439}
440__setup("tracing_thresh=", set_tracing_thresh); 440__setup("tracing_thresh=", set_tracing_thresh);
@@ -470,6 +470,7 @@ static const char *trace_options[] = {
470 "overwrite", 470 "overwrite",
471 "disable_on_free", 471 "disable_on_free",
472 "irq-info", 472 "irq-info",
473 "markers",
473 NULL 474 NULL
474}; 475};
475 476
@@ -2060,7 +2061,8 @@ print_trace_header(struct seq_file *m, struct trace_iterator *iter)
2060 seq_puts(m, "# -----------------\n"); 2061 seq_puts(m, "# -----------------\n");
2061 seq_printf(m, "# | task: %.16s-%d " 2062 seq_printf(m, "# | task: %.16s-%d "
2062 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n", 2063 "(uid:%d nice:%ld policy:%ld rt_prio:%ld)\n",
2063 data->comm, data->pid, data->uid, data->nice, 2064 data->comm, data->pid,
2065 from_kuid_munged(seq_user_ns(m), data->uid), data->nice,
2064 data->policy, data->rt_priority); 2066 data->policy, data->rt_priority);
2065 seq_puts(m, "# -----------------\n"); 2067 seq_puts(m, "# -----------------\n");
2066 2068
@@ -3886,6 +3888,9 @@ tracing_mark_write(struct file *filp, const char __user *ubuf,
3886 if (tracing_disabled) 3888 if (tracing_disabled)
3887 return -EINVAL; 3889 return -EINVAL;
3888 3890
3891 if (!(trace_flags & TRACE_ITER_MARKERS))
3892 return -EINVAL;
3893
3889 if (cnt > TRACE_BUF_SIZE) 3894 if (cnt > TRACE_BUF_SIZE)
3890 cnt = TRACE_BUF_SIZE; 3895 cnt = TRACE_BUF_SIZE;
3891 3896
@@ -4195,12 +4200,6 @@ static void buffer_pipe_buf_release(struct pipe_inode_info *pipe,
4195 buf->private = 0; 4200 buf->private = 0;
4196} 4201}
4197 4202
4198static int buffer_pipe_buf_steal(struct pipe_inode_info *pipe,
4199 struct pipe_buffer *buf)
4200{
4201 return 1;
4202}
4203
4204static void buffer_pipe_buf_get(struct pipe_inode_info *pipe, 4203static void buffer_pipe_buf_get(struct pipe_inode_info *pipe,
4205 struct pipe_buffer *buf) 4204 struct pipe_buffer *buf)
4206{ 4205{
@@ -4216,7 +4215,7 @@ static const struct pipe_buf_operations buffer_pipe_buf_ops = {
4216 .unmap = generic_pipe_buf_unmap, 4215 .unmap = generic_pipe_buf_unmap,
4217 .confirm = generic_pipe_buf_confirm, 4216 .confirm = generic_pipe_buf_confirm,
4218 .release = buffer_pipe_buf_release, 4217 .release = buffer_pipe_buf_release,
4219 .steal = buffer_pipe_buf_steal, 4218 .steal = generic_pipe_buf_steal,
4220 .get = buffer_pipe_buf_get, 4219 .get = buffer_pipe_buf_get,
4221}; 4220};
4222 4221
diff --git a/kernel/trace/trace.h b/kernel/trace/trace.h
index 55e1f7f0db12..c15f528c1af4 100644
--- a/kernel/trace/trace.h
+++ b/kernel/trace/trace.h
@@ -147,7 +147,7 @@ struct trace_array_cpu {
147 unsigned long skipped_entries; 147 unsigned long skipped_entries;
148 cycle_t preempt_timestamp; 148 cycle_t preempt_timestamp;
149 pid_t pid; 149 pid_t pid;
150 uid_t uid; 150 kuid_t uid;
151 char comm[TASK_COMM_LEN]; 151 char comm[TASK_COMM_LEN];
152}; 152};
153 153
@@ -472,11 +472,11 @@ extern void trace_find_cmdline(int pid, char comm[]);
472 472
473#ifdef CONFIG_DYNAMIC_FTRACE 473#ifdef CONFIG_DYNAMIC_FTRACE
474extern unsigned long ftrace_update_tot_cnt; 474extern unsigned long ftrace_update_tot_cnt;
475#endif
475#define DYN_FTRACE_TEST_NAME trace_selftest_dynamic_test_func 476#define DYN_FTRACE_TEST_NAME trace_selftest_dynamic_test_func
476extern int DYN_FTRACE_TEST_NAME(void); 477extern int DYN_FTRACE_TEST_NAME(void);
477#define DYN_FTRACE_TEST_NAME2 trace_selftest_dynamic_test_func2 478#define DYN_FTRACE_TEST_NAME2 trace_selftest_dynamic_test_func2
478extern int DYN_FTRACE_TEST_NAME2(void); 479extern int DYN_FTRACE_TEST_NAME2(void);
479#endif
480 480
481extern int ring_buffer_expanded; 481extern int ring_buffer_expanded;
482extern bool tracing_selftest_disabled; 482extern bool tracing_selftest_disabled;
@@ -680,6 +680,7 @@ enum trace_iterator_flags {
680 TRACE_ITER_OVERWRITE = 0x200000, 680 TRACE_ITER_OVERWRITE = 0x200000,
681 TRACE_ITER_STOP_ON_FREE = 0x400000, 681 TRACE_ITER_STOP_ON_FREE = 0x400000,
682 TRACE_ITER_IRQ_INFO = 0x800000, 682 TRACE_ITER_IRQ_INFO = 0x800000,
683 TRACE_ITER_MARKERS = 0x1000000,
683}; 684};
684 685
685/* 686/*
diff --git a/kernel/trace/trace_event_perf.c b/kernel/trace/trace_event_perf.c
index 8a6d2ee2086c..84b1e045faba 100644
--- a/kernel/trace/trace_event_perf.c
+++ b/kernel/trace/trace_event_perf.c
@@ -258,7 +258,8 @@ EXPORT_SYMBOL_GPL(perf_trace_buf_prepare);
258 258
259#ifdef CONFIG_FUNCTION_TRACER 259#ifdef CONFIG_FUNCTION_TRACER
260static void 260static void
261perf_ftrace_function_call(unsigned long ip, unsigned long parent_ip) 261perf_ftrace_function_call(unsigned long ip, unsigned long parent_ip,
262 struct ftrace_ops *ops, struct pt_regs *pt_regs)
262{ 263{
263 struct ftrace_entry *entry; 264 struct ftrace_entry *entry;
264 struct hlist_head *head; 265 struct hlist_head *head;
diff --git a/kernel/trace/trace_events.c b/kernel/trace/trace_events.c
index 29111da1d100..d608d09d08c0 100644
--- a/kernel/trace/trace_events.c
+++ b/kernel/trace/trace_events.c
@@ -1199,6 +1199,31 @@ event_create_dir(struct ftrace_event_call *call, struct dentry *d_events,
1199 return 0; 1199 return 0;
1200} 1200}
1201 1201
1202static void event_remove(struct ftrace_event_call *call)
1203{
1204 ftrace_event_enable_disable(call, 0);
1205 if (call->event.funcs)
1206 __unregister_ftrace_event(&call->event);
1207 list_del(&call->list);
1208}
1209
1210static int event_init(struct ftrace_event_call *call)
1211{
1212 int ret = 0;
1213
1214 if (WARN_ON(!call->name))
1215 return -EINVAL;
1216
1217 if (call->class->raw_init) {
1218 ret = call->class->raw_init(call);
1219 if (ret < 0 && ret != -ENOSYS)
1220 pr_warn("Could not initialize trace events/%s\n",
1221 call->name);
1222 }
1223
1224 return ret;
1225}
1226
1202static int 1227static int
1203__trace_add_event_call(struct ftrace_event_call *call, struct module *mod, 1228__trace_add_event_call(struct ftrace_event_call *call, struct module *mod,
1204 const struct file_operations *id, 1229 const struct file_operations *id,
@@ -1209,19 +1234,9 @@ __trace_add_event_call(struct ftrace_event_call *call, struct module *mod,
1209 struct dentry *d_events; 1234 struct dentry *d_events;
1210 int ret; 1235 int ret;
1211 1236
1212 /* The linker may leave blanks */ 1237 ret = event_init(call);
1213 if (!call->name) 1238 if (ret < 0)
1214 return -EINVAL; 1239 return ret;
1215
1216 if (call->class->raw_init) {
1217 ret = call->class->raw_init(call);
1218 if (ret < 0) {
1219 if (ret != -ENOSYS)
1220 pr_warning("Could not initialize trace events/%s\n",
1221 call->name);
1222 return ret;
1223 }
1224 }
1225 1240
1226 d_events = event_trace_events_dir(); 1241 d_events = event_trace_events_dir();
1227 if (!d_events) 1242 if (!d_events)
@@ -1272,13 +1287,10 @@ static void remove_subsystem_dir(const char *name)
1272 */ 1287 */
1273static void __trace_remove_event_call(struct ftrace_event_call *call) 1288static void __trace_remove_event_call(struct ftrace_event_call *call)
1274{ 1289{
1275 ftrace_event_enable_disable(call, 0); 1290 event_remove(call);
1276 if (call->event.funcs)
1277 __unregister_ftrace_event(&call->event);
1278 debugfs_remove_recursive(call->dir);
1279 list_del(&call->list);
1280 trace_destroy_fields(call); 1291 trace_destroy_fields(call);
1281 destroy_preds(call); 1292 destroy_preds(call);
1293 debugfs_remove_recursive(call->dir);
1282 remove_subsystem_dir(call->class->system); 1294 remove_subsystem_dir(call->class->system);
1283} 1295}
1284 1296
@@ -1450,15 +1462,43 @@ static __init int setup_trace_event(char *str)
1450} 1462}
1451__setup("trace_event=", setup_trace_event); 1463__setup("trace_event=", setup_trace_event);
1452 1464
1465static __init int event_trace_enable(void)
1466{
1467 struct ftrace_event_call **iter, *call;
1468 char *buf = bootup_event_buf;
1469 char *token;
1470 int ret;
1471
1472 for_each_event(iter, __start_ftrace_events, __stop_ftrace_events) {
1473
1474 call = *iter;
1475 ret = event_init(call);
1476 if (!ret)
1477 list_add(&call->list, &ftrace_events);
1478 }
1479
1480 while (true) {
1481 token = strsep(&buf, ",");
1482
1483 if (!token)
1484 break;
1485 if (!*token)
1486 continue;
1487
1488 ret = ftrace_set_clr_event(token, 1);
1489 if (ret)
1490 pr_warn("Failed to enable trace event: %s\n", token);
1491 }
1492 return 0;
1493}
1494
1453static __init int event_trace_init(void) 1495static __init int event_trace_init(void)
1454{ 1496{
1455 struct ftrace_event_call **call; 1497 struct ftrace_event_call *call;
1456 struct dentry *d_tracer; 1498 struct dentry *d_tracer;
1457 struct dentry *entry; 1499 struct dentry *entry;
1458 struct dentry *d_events; 1500 struct dentry *d_events;
1459 int ret; 1501 int ret;
1460 char *buf = bootup_event_buf;
1461 char *token;
1462 1502
1463 d_tracer = tracing_init_dentry(); 1503 d_tracer = tracing_init_dentry();
1464 if (!d_tracer) 1504 if (!d_tracer)
@@ -1497,24 +1537,19 @@ static __init int event_trace_init(void)
1497 if (trace_define_common_fields()) 1537 if (trace_define_common_fields())
1498 pr_warning("tracing: Failed to allocate common fields"); 1538 pr_warning("tracing: Failed to allocate common fields");
1499 1539
1500 for_each_event(call, __start_ftrace_events, __stop_ftrace_events) { 1540 /*
1501 __trace_add_event_call(*call, NULL, &ftrace_event_id_fops, 1541 * Early initialization already enabled ftrace event.
1542 * Now it's only necessary to create the event directory.
1543 */
1544 list_for_each_entry(call, &ftrace_events, list) {
1545
1546 ret = event_create_dir(call, d_events,
1547 &ftrace_event_id_fops,
1502 &ftrace_enable_fops, 1548 &ftrace_enable_fops,
1503 &ftrace_event_filter_fops, 1549 &ftrace_event_filter_fops,
1504 &ftrace_event_format_fops); 1550 &ftrace_event_format_fops);
1505 } 1551 if (ret < 0)
1506 1552 event_remove(call);
1507 while (true) {
1508 token = strsep(&buf, ",");
1509
1510 if (!token)
1511 break;
1512 if (!*token)
1513 continue;
1514
1515 ret = ftrace_set_clr_event(token, 1);
1516 if (ret)
1517 pr_warning("Failed to enable trace event: %s\n", token);
1518 } 1553 }
1519 1554
1520 ret = register_module_notifier(&trace_module_nb); 1555 ret = register_module_notifier(&trace_module_nb);
@@ -1523,6 +1558,7 @@ static __init int event_trace_init(void)
1523 1558
1524 return 0; 1559 return 0;
1525} 1560}
1561core_initcall(event_trace_enable);
1526fs_initcall(event_trace_init); 1562fs_initcall(event_trace_init);
1527 1563
1528#ifdef CONFIG_FTRACE_STARTUP_TEST 1564#ifdef CONFIG_FTRACE_STARTUP_TEST
@@ -1646,9 +1682,11 @@ static __init void event_trace_self_tests(void)
1646 event_test_stuff(); 1682 event_test_stuff();
1647 1683
1648 ret = __ftrace_set_clr_event(NULL, system->name, NULL, 0); 1684 ret = __ftrace_set_clr_event(NULL, system->name, NULL, 0);
1649 if (WARN_ON_ONCE(ret)) 1685 if (WARN_ON_ONCE(ret)) {
1650 pr_warning("error disabling system %s\n", 1686 pr_warning("error disabling system %s\n",
1651 system->name); 1687 system->name);
1688 continue;
1689 }
1652 1690
1653 pr_cont("OK\n"); 1691 pr_cont("OK\n");
1654 } 1692 }
@@ -1681,7 +1719,8 @@ static __init void event_trace_self_tests(void)
1681static DEFINE_PER_CPU(atomic_t, ftrace_test_event_disable); 1719static DEFINE_PER_CPU(atomic_t, ftrace_test_event_disable);
1682 1720
1683static void 1721static void
1684function_test_events_call(unsigned long ip, unsigned long parent_ip) 1722function_test_events_call(unsigned long ip, unsigned long parent_ip,
1723 struct ftrace_ops *op, struct pt_regs *pt_regs)
1685{ 1724{
1686 struct ring_buffer_event *event; 1725 struct ring_buffer_event *event;
1687 struct ring_buffer *buffer; 1726 struct ring_buffer *buffer;
@@ -1720,6 +1759,7 @@ function_test_events_call(unsigned long ip, unsigned long parent_ip)
1720static struct ftrace_ops trace_ops __initdata = 1759static struct ftrace_ops trace_ops __initdata =
1721{ 1760{
1722 .func = function_test_events_call, 1761 .func = function_test_events_call,
1762 .flags = FTRACE_OPS_FL_RECURSION_SAFE,
1723}; 1763};
1724 1764
1725static __init void event_trace_self_test_with_function(void) 1765static __init void event_trace_self_test_with_function(void)
diff --git a/kernel/trace/trace_events_filter.c b/kernel/trace/trace_events_filter.c
index 431dba8b7542..c154797a7ff7 100644
--- a/kernel/trace/trace_events_filter.c
+++ b/kernel/trace/trace_events_filter.c
@@ -2002,7 +2002,7 @@ static int ftrace_function_set_regexp(struct ftrace_ops *ops, int filter,
2002static int __ftrace_function_set_filter(int filter, char *buf, int len, 2002static int __ftrace_function_set_filter(int filter, char *buf, int len,
2003 struct function_filter_data *data) 2003 struct function_filter_data *data)
2004{ 2004{
2005 int i, re_cnt, ret; 2005 int i, re_cnt, ret = -EINVAL;
2006 int *reset; 2006 int *reset;
2007 char **re; 2007 char **re;
2008 2008
diff --git a/kernel/trace/trace_functions.c b/kernel/trace/trace_functions.c
index a426f410c060..507a7a9630bf 100644
--- a/kernel/trace/trace_functions.c
+++ b/kernel/trace/trace_functions.c
@@ -13,7 +13,6 @@
13#include <linux/debugfs.h> 13#include <linux/debugfs.h>
14#include <linux/uaccess.h> 14#include <linux/uaccess.h>
15#include <linux/ftrace.h> 15#include <linux/ftrace.h>
16#include <linux/pstore.h>
17#include <linux/fs.h> 16#include <linux/fs.h>
18 17
19#include "trace.h" 18#include "trace.h"
@@ -49,7 +48,8 @@ static void function_trace_start(struct trace_array *tr)
49} 48}
50 49
51static void 50static void
52function_trace_call_preempt_only(unsigned long ip, unsigned long parent_ip) 51function_trace_call_preempt_only(unsigned long ip, unsigned long parent_ip,
52 struct ftrace_ops *op, struct pt_regs *pt_regs)
53{ 53{
54 struct trace_array *tr = func_trace; 54 struct trace_array *tr = func_trace;
55 struct trace_array_cpu *data; 55 struct trace_array_cpu *data;
@@ -75,16 +75,17 @@ function_trace_call_preempt_only(unsigned long ip, unsigned long parent_ip)
75 preempt_enable_notrace(); 75 preempt_enable_notrace();
76} 76}
77 77
78/* Our two options */ 78/* Our option */
79enum { 79enum {
80 TRACE_FUNC_OPT_STACK = 0x1, 80 TRACE_FUNC_OPT_STACK = 0x1,
81 TRACE_FUNC_OPT_PSTORE = 0x2,
82}; 81};
83 82
84static struct tracer_flags func_flags; 83static struct tracer_flags func_flags;
85 84
86static void 85static void
87function_trace_call(unsigned long ip, unsigned long parent_ip) 86function_trace_call(unsigned long ip, unsigned long parent_ip,
87 struct ftrace_ops *op, struct pt_regs *pt_regs)
88
88{ 89{
89 struct trace_array *tr = func_trace; 90 struct trace_array *tr = func_trace;
90 struct trace_array_cpu *data; 91 struct trace_array_cpu *data;
@@ -106,12 +107,6 @@ function_trace_call(unsigned long ip, unsigned long parent_ip)
106 disabled = atomic_inc_return(&data->disabled); 107 disabled = atomic_inc_return(&data->disabled);
107 108
108 if (likely(disabled == 1)) { 109 if (likely(disabled == 1)) {
109 /*
110 * So far tracing doesn't support multiple buffers, so
111 * we make an explicit call for now.
112 */
113 if (unlikely(func_flags.val & TRACE_FUNC_OPT_PSTORE))
114 pstore_ftrace_call(ip, parent_ip);
115 pc = preempt_count(); 110 pc = preempt_count();
116 trace_function(tr, ip, parent_ip, flags, pc); 111 trace_function(tr, ip, parent_ip, flags, pc);
117 } 112 }
@@ -121,7 +116,8 @@ function_trace_call(unsigned long ip, unsigned long parent_ip)
121} 116}
122 117
123static void 118static void
124function_stack_trace_call(unsigned long ip, unsigned long parent_ip) 119function_stack_trace_call(unsigned long ip, unsigned long parent_ip,
120 struct ftrace_ops *op, struct pt_regs *pt_regs)
125{ 121{
126 struct trace_array *tr = func_trace; 122 struct trace_array *tr = func_trace;
127 struct trace_array_cpu *data; 123 struct trace_array_cpu *data;
@@ -164,22 +160,19 @@ function_stack_trace_call(unsigned long ip, unsigned long parent_ip)
164static struct ftrace_ops trace_ops __read_mostly = 160static struct ftrace_ops trace_ops __read_mostly =
165{ 161{
166 .func = function_trace_call, 162 .func = function_trace_call,
167 .flags = FTRACE_OPS_FL_GLOBAL, 163 .flags = FTRACE_OPS_FL_GLOBAL | FTRACE_OPS_FL_RECURSION_SAFE,
168}; 164};
169 165
170static struct ftrace_ops trace_stack_ops __read_mostly = 166static struct ftrace_ops trace_stack_ops __read_mostly =
171{ 167{
172 .func = function_stack_trace_call, 168 .func = function_stack_trace_call,
173 .flags = FTRACE_OPS_FL_GLOBAL, 169 .flags = FTRACE_OPS_FL_GLOBAL | FTRACE_OPS_FL_RECURSION_SAFE,
174}; 170};
175 171
176static struct tracer_opt func_opts[] = { 172static struct tracer_opt func_opts[] = {
177#ifdef CONFIG_STACKTRACE 173#ifdef CONFIG_STACKTRACE
178 { TRACER_OPT(func_stack_trace, TRACE_FUNC_OPT_STACK) }, 174 { TRACER_OPT(func_stack_trace, TRACE_FUNC_OPT_STACK) },
179#endif 175#endif
180#ifdef CONFIG_PSTORE_FTRACE
181 { TRACER_OPT(func_pstore, TRACE_FUNC_OPT_PSTORE) },
182#endif
183 { } /* Always set a last empty entry */ 176 { } /* Always set a last empty entry */
184}; 177};
185 178
@@ -232,8 +225,6 @@ static int func_set_flag(u32 old_flags, u32 bit, int set)
232 } 225 }
233 226
234 break; 227 break;
235 case TRACE_FUNC_OPT_PSTORE:
236 break;
237 default: 228 default:
238 return -EINVAL; 229 return -EINVAL;
239 } 230 }
diff --git a/kernel/trace/trace_functions_graph.c b/kernel/trace/trace_functions_graph.c
index ce27c8ba8d31..99b4378393d5 100644
--- a/kernel/trace/trace_functions_graph.c
+++ b/kernel/trace/trace_functions_graph.c
@@ -143,7 +143,7 @@ ftrace_pop_return_trace(struct ftrace_graph_ret *trace, unsigned long *ret,
143 return; 143 return;
144 } 144 }
145 145
146#ifdef CONFIG_HAVE_FUNCTION_GRAPH_FP_TEST 146#if defined(CONFIG_HAVE_FUNCTION_GRAPH_FP_TEST) && !defined(CC_USING_FENTRY)
147 /* 147 /*
148 * The arch may choose to record the frame pointer used 148 * The arch may choose to record the frame pointer used
149 * and check it here to make sure that it is what we expect it 149 * and check it here to make sure that it is what we expect it
@@ -154,6 +154,9 @@ ftrace_pop_return_trace(struct ftrace_graph_ret *trace, unsigned long *ret,
154 * 154 *
155 * Currently, x86_32 with optimize for size (-Os) makes the latest 155 * Currently, x86_32 with optimize for size (-Os) makes the latest
156 * gcc do the above. 156 * gcc do the above.
157 *
158 * Note, -mfentry does not use frame pointers, and this test
159 * is not needed if CC_USING_FENTRY is set.
157 */ 160 */
158 if (unlikely(current->ret_stack[index].fp != frame_pointer)) { 161 if (unlikely(current->ret_stack[index].fp != frame_pointer)) {
159 ftrace_graph_stop(); 162 ftrace_graph_stop();
diff --git a/kernel/trace/trace_irqsoff.c b/kernel/trace/trace_irqsoff.c
index 99d20e920368..d98ee8283b29 100644
--- a/kernel/trace/trace_irqsoff.c
+++ b/kernel/trace/trace_irqsoff.c
@@ -136,7 +136,8 @@ static int func_prolog_dec(struct trace_array *tr,
136 * irqsoff uses its own tracer function to keep the overhead down: 136 * irqsoff uses its own tracer function to keep the overhead down:
137 */ 137 */
138static void 138static void
139irqsoff_tracer_call(unsigned long ip, unsigned long parent_ip) 139irqsoff_tracer_call(unsigned long ip, unsigned long parent_ip,
140 struct ftrace_ops *op, struct pt_regs *pt_regs)
140{ 141{
141 struct trace_array *tr = irqsoff_trace; 142 struct trace_array *tr = irqsoff_trace;
142 struct trace_array_cpu *data; 143 struct trace_array_cpu *data;
@@ -153,7 +154,7 @@ irqsoff_tracer_call(unsigned long ip, unsigned long parent_ip)
153static struct ftrace_ops trace_ops __read_mostly = 154static struct ftrace_ops trace_ops __read_mostly =
154{ 155{
155 .func = irqsoff_tracer_call, 156 .func = irqsoff_tracer_call,
156 .flags = FTRACE_OPS_FL_GLOBAL, 157 .flags = FTRACE_OPS_FL_GLOBAL | FTRACE_OPS_FL_RECURSION_SAFE,
157}; 158};
158#endif /* CONFIG_FUNCTION_TRACER */ 159#endif /* CONFIG_FUNCTION_TRACER */
159 160
diff --git a/kernel/trace/trace_sched_wakeup.c b/kernel/trace/trace_sched_wakeup.c
index ff791ea48b57..02170c00c413 100644
--- a/kernel/trace/trace_sched_wakeup.c
+++ b/kernel/trace/trace_sched_wakeup.c
@@ -108,7 +108,8 @@ out_enable:
108 * wakeup uses its own tracer function to keep the overhead down: 108 * wakeup uses its own tracer function to keep the overhead down:
109 */ 109 */
110static void 110static void
111wakeup_tracer_call(unsigned long ip, unsigned long parent_ip) 111wakeup_tracer_call(unsigned long ip, unsigned long parent_ip,
112 struct ftrace_ops *op, struct pt_regs *pt_regs)
112{ 113{
113 struct trace_array *tr = wakeup_trace; 114 struct trace_array *tr = wakeup_trace;
114 struct trace_array_cpu *data; 115 struct trace_array_cpu *data;
@@ -129,7 +130,7 @@ wakeup_tracer_call(unsigned long ip, unsigned long parent_ip)
129static struct ftrace_ops trace_ops __read_mostly = 130static struct ftrace_ops trace_ops __read_mostly =
130{ 131{
131 .func = wakeup_tracer_call, 132 .func = wakeup_tracer_call,
132 .flags = FTRACE_OPS_FL_GLOBAL, 133 .flags = FTRACE_OPS_FL_GLOBAL | FTRACE_OPS_FL_RECURSION_SAFE,
133}; 134};
134#endif /* CONFIG_FUNCTION_TRACER */ 135#endif /* CONFIG_FUNCTION_TRACER */
135 136
diff --git a/kernel/trace/trace_selftest.c b/kernel/trace/trace_selftest.c
index 288541f977fb..2c00a691a540 100644
--- a/kernel/trace/trace_selftest.c
+++ b/kernel/trace/trace_selftest.c
@@ -103,54 +103,67 @@ static inline void warn_failed_init_tracer(struct tracer *trace, int init_ret)
103 103
104static int trace_selftest_test_probe1_cnt; 104static int trace_selftest_test_probe1_cnt;
105static void trace_selftest_test_probe1_func(unsigned long ip, 105static void trace_selftest_test_probe1_func(unsigned long ip,
106 unsigned long pip) 106 unsigned long pip,
107 struct ftrace_ops *op,
108 struct pt_regs *pt_regs)
107{ 109{
108 trace_selftest_test_probe1_cnt++; 110 trace_selftest_test_probe1_cnt++;
109} 111}
110 112
111static int trace_selftest_test_probe2_cnt; 113static int trace_selftest_test_probe2_cnt;
112static void trace_selftest_test_probe2_func(unsigned long ip, 114static void trace_selftest_test_probe2_func(unsigned long ip,
113 unsigned long pip) 115 unsigned long pip,
116 struct ftrace_ops *op,
117 struct pt_regs *pt_regs)
114{ 118{
115 trace_selftest_test_probe2_cnt++; 119 trace_selftest_test_probe2_cnt++;
116} 120}
117 121
118static int trace_selftest_test_probe3_cnt; 122static int trace_selftest_test_probe3_cnt;
119static void trace_selftest_test_probe3_func(unsigned long ip, 123static void trace_selftest_test_probe3_func(unsigned long ip,
120 unsigned long pip) 124 unsigned long pip,
125 struct ftrace_ops *op,
126 struct pt_regs *pt_regs)
121{ 127{
122 trace_selftest_test_probe3_cnt++; 128 trace_selftest_test_probe3_cnt++;
123} 129}
124 130
125static int trace_selftest_test_global_cnt; 131static int trace_selftest_test_global_cnt;
126static void trace_selftest_test_global_func(unsigned long ip, 132static void trace_selftest_test_global_func(unsigned long ip,
127 unsigned long pip) 133 unsigned long pip,
134 struct ftrace_ops *op,
135 struct pt_regs *pt_regs)
128{ 136{
129 trace_selftest_test_global_cnt++; 137 trace_selftest_test_global_cnt++;
130} 138}
131 139
132static int trace_selftest_test_dyn_cnt; 140static int trace_selftest_test_dyn_cnt;
133static void trace_selftest_test_dyn_func(unsigned long ip, 141static void trace_selftest_test_dyn_func(unsigned long ip,
134 unsigned long pip) 142 unsigned long pip,
143 struct ftrace_ops *op,
144 struct pt_regs *pt_regs)
135{ 145{
136 trace_selftest_test_dyn_cnt++; 146 trace_selftest_test_dyn_cnt++;
137} 147}
138 148
139static struct ftrace_ops test_probe1 = { 149static struct ftrace_ops test_probe1 = {
140 .func = trace_selftest_test_probe1_func, 150 .func = trace_selftest_test_probe1_func,
151 .flags = FTRACE_OPS_FL_RECURSION_SAFE,
141}; 152};
142 153
143static struct ftrace_ops test_probe2 = { 154static struct ftrace_ops test_probe2 = {
144 .func = trace_selftest_test_probe2_func, 155 .func = trace_selftest_test_probe2_func,
156 .flags = FTRACE_OPS_FL_RECURSION_SAFE,
145}; 157};
146 158
147static struct ftrace_ops test_probe3 = { 159static struct ftrace_ops test_probe3 = {
148 .func = trace_selftest_test_probe3_func, 160 .func = trace_selftest_test_probe3_func,
161 .flags = FTRACE_OPS_FL_RECURSION_SAFE,
149}; 162};
150 163
151static struct ftrace_ops test_global = { 164static struct ftrace_ops test_global = {
152 .func = trace_selftest_test_global_func, 165 .func = trace_selftest_test_global_func,
153 .flags = FTRACE_OPS_FL_GLOBAL, 166 .flags = FTRACE_OPS_FL_GLOBAL | FTRACE_OPS_FL_RECURSION_SAFE,
154}; 167};
155 168
156static void print_counts(void) 169static void print_counts(void)
@@ -393,10 +406,253 @@ int trace_selftest_startup_dynamic_tracing(struct tracer *trace,
393 406
394 return ret; 407 return ret;
395} 408}
409
410static int trace_selftest_recursion_cnt;
411static void trace_selftest_test_recursion_func(unsigned long ip,
412 unsigned long pip,
413 struct ftrace_ops *op,
414 struct pt_regs *pt_regs)
415{
416 /*
417 * This function is registered without the recursion safe flag.
418 * The ftrace infrastructure should provide the recursion
419 * protection. If not, this will crash the kernel!
420 */
421 trace_selftest_recursion_cnt++;
422 DYN_FTRACE_TEST_NAME();
423}
424
425static void trace_selftest_test_recursion_safe_func(unsigned long ip,
426 unsigned long pip,
427 struct ftrace_ops *op,
428 struct pt_regs *pt_regs)
429{
430 /*
431 * We said we would provide our own recursion. By calling
432 * this function again, we should recurse back into this function
433 * and count again. But this only happens if the arch supports
434 * all of ftrace features and nothing else is using the function
435 * tracing utility.
436 */
437 if (trace_selftest_recursion_cnt++)
438 return;
439 DYN_FTRACE_TEST_NAME();
440}
441
442static struct ftrace_ops test_rec_probe = {
443 .func = trace_selftest_test_recursion_func,
444};
445
446static struct ftrace_ops test_recsafe_probe = {
447 .func = trace_selftest_test_recursion_safe_func,
448 .flags = FTRACE_OPS_FL_RECURSION_SAFE,
449};
450
451static int
452trace_selftest_function_recursion(void)
453{
454 int save_ftrace_enabled = ftrace_enabled;
455 int save_tracer_enabled = tracer_enabled;
456 char *func_name;
457 int len;
458 int ret;
459 int cnt;
460
461 /* The previous test PASSED */
462 pr_cont("PASSED\n");
463 pr_info("Testing ftrace recursion: ");
464
465
466 /* enable tracing, and record the filter function */
467 ftrace_enabled = 1;
468 tracer_enabled = 1;
469
470 /* Handle PPC64 '.' name */
471 func_name = "*" __stringify(DYN_FTRACE_TEST_NAME);
472 len = strlen(func_name);
473
474 ret = ftrace_set_filter(&test_rec_probe, func_name, len, 1);
475 if (ret) {
476 pr_cont("*Could not set filter* ");
477 goto out;
478 }
479
480 ret = register_ftrace_function(&test_rec_probe);
481 if (ret) {
482 pr_cont("*could not register callback* ");
483 goto out;
484 }
485
486 DYN_FTRACE_TEST_NAME();
487
488 unregister_ftrace_function(&test_rec_probe);
489
490 ret = -1;
491 if (trace_selftest_recursion_cnt != 1) {
492 pr_cont("*callback not called once (%d)* ",
493 trace_selftest_recursion_cnt);
494 goto out;
495 }
496
497 trace_selftest_recursion_cnt = 1;
498
499 pr_cont("PASSED\n");
500 pr_info("Testing ftrace recursion safe: ");
501
502 ret = ftrace_set_filter(&test_recsafe_probe, func_name, len, 1);
503 if (ret) {
504 pr_cont("*Could not set filter* ");
505 goto out;
506 }
507
508 ret = register_ftrace_function(&test_recsafe_probe);
509 if (ret) {
510 pr_cont("*could not register callback* ");
511 goto out;
512 }
513
514 DYN_FTRACE_TEST_NAME();
515
516 unregister_ftrace_function(&test_recsafe_probe);
517
518 /*
519 * If arch supports all ftrace features, and no other task
520 * was on the list, we should be fine.
521 */
522 if (!ftrace_nr_registered_ops() && !FTRACE_FORCE_LIST_FUNC)
523 cnt = 2; /* Should have recursed */
524 else
525 cnt = 1;
526
527 ret = -1;
528 if (trace_selftest_recursion_cnt != cnt) {
529 pr_cont("*callback not called expected %d times (%d)* ",
530 cnt, trace_selftest_recursion_cnt);
531 goto out;
532 }
533
534 ret = 0;
535out:
536 ftrace_enabled = save_ftrace_enabled;
537 tracer_enabled = save_tracer_enabled;
538
539 return ret;
540}
396#else 541#else
397# define trace_selftest_startup_dynamic_tracing(trace, tr, func) ({ 0; }) 542# define trace_selftest_startup_dynamic_tracing(trace, tr, func) ({ 0; })
543# define trace_selftest_function_recursion() ({ 0; })
398#endif /* CONFIG_DYNAMIC_FTRACE */ 544#endif /* CONFIG_DYNAMIC_FTRACE */
399 545
546static enum {
547 TRACE_SELFTEST_REGS_START,
548 TRACE_SELFTEST_REGS_FOUND,
549 TRACE_SELFTEST_REGS_NOT_FOUND,
550} trace_selftest_regs_stat;
551
552static void trace_selftest_test_regs_func(unsigned long ip,
553 unsigned long pip,
554 struct ftrace_ops *op,
555 struct pt_regs *pt_regs)
556{
557 if (pt_regs)
558 trace_selftest_regs_stat = TRACE_SELFTEST_REGS_FOUND;
559 else
560 trace_selftest_regs_stat = TRACE_SELFTEST_REGS_NOT_FOUND;
561}
562
563static struct ftrace_ops test_regs_probe = {
564 .func = trace_selftest_test_regs_func,
565 .flags = FTRACE_OPS_FL_RECURSION_SAFE | FTRACE_OPS_FL_SAVE_REGS,
566};
567
568static int
569trace_selftest_function_regs(void)
570{
571 int save_ftrace_enabled = ftrace_enabled;
572 int save_tracer_enabled = tracer_enabled;
573 char *func_name;
574 int len;
575 int ret;
576 int supported = 0;
577
578#ifdef ARCH_SUPPORTS_FTRACE_SAVE_REGS
579 supported = 1;
580#endif
581
582 /* The previous test PASSED */
583 pr_cont("PASSED\n");
584 pr_info("Testing ftrace regs%s: ",
585 !supported ? "(no arch support)" : "");
586
587 /* enable tracing, and record the filter function */
588 ftrace_enabled = 1;
589 tracer_enabled = 1;
590
591 /* Handle PPC64 '.' name */
592 func_name = "*" __stringify(DYN_FTRACE_TEST_NAME);
593 len = strlen(func_name);
594
595 ret = ftrace_set_filter(&test_regs_probe, func_name, len, 1);
596 /*
597 * If DYNAMIC_FTRACE is not set, then we just trace all functions.
598 * This test really doesn't care.
599 */
600 if (ret && ret != -ENODEV) {
601 pr_cont("*Could not set filter* ");
602 goto out;
603 }
604
605 ret = register_ftrace_function(&test_regs_probe);
606 /*
607 * Now if the arch does not support passing regs, then this should
608 * have failed.
609 */
610 if (!supported) {
611 if (!ret) {
612 pr_cont("*registered save-regs without arch support* ");
613 goto out;
614 }
615 test_regs_probe.flags |= FTRACE_OPS_FL_SAVE_REGS_IF_SUPPORTED;
616 ret = register_ftrace_function(&test_regs_probe);
617 }
618 if (ret) {
619 pr_cont("*could not register callback* ");
620 goto out;
621 }
622
623
624 DYN_FTRACE_TEST_NAME();
625
626 unregister_ftrace_function(&test_regs_probe);
627
628 ret = -1;
629
630 switch (trace_selftest_regs_stat) {
631 case TRACE_SELFTEST_REGS_START:
632 pr_cont("*callback never called* ");
633 goto out;
634
635 case TRACE_SELFTEST_REGS_FOUND:
636 if (supported)
637 break;
638 pr_cont("*callback received regs without arch support* ");
639 goto out;
640
641 case TRACE_SELFTEST_REGS_NOT_FOUND:
642 if (!supported)
643 break;
644 pr_cont("*callback received NULL regs* ");
645 goto out;
646 }
647
648 ret = 0;
649out:
650 ftrace_enabled = save_ftrace_enabled;
651 tracer_enabled = save_tracer_enabled;
652
653 return ret;
654}
655
400/* 656/*
401 * Simple verification test of ftrace function tracer. 657 * Simple verification test of ftrace function tracer.
402 * Enable ftrace, sleep 1/10 second, and then read the trace 658 * Enable ftrace, sleep 1/10 second, and then read the trace
@@ -442,7 +698,14 @@ trace_selftest_startup_function(struct tracer *trace, struct trace_array *tr)
442 698
443 ret = trace_selftest_startup_dynamic_tracing(trace, tr, 699 ret = trace_selftest_startup_dynamic_tracing(trace, tr,
444 DYN_FTRACE_TEST_NAME); 700 DYN_FTRACE_TEST_NAME);
701 if (ret)
702 goto out;
445 703
704 ret = trace_selftest_function_recursion();
705 if (ret)
706 goto out;
707
708 ret = trace_selftest_function_regs();
446 out: 709 out:
447 ftrace_enabled = save_ftrace_enabled; 710 ftrace_enabled = save_ftrace_enabled;
448 tracer_enabled = save_tracer_enabled; 711 tracer_enabled = save_tracer_enabled;
@@ -778,6 +1041,8 @@ static int trace_wakeup_test_thread(void *data)
778 set_current_state(TASK_INTERRUPTIBLE); 1041 set_current_state(TASK_INTERRUPTIBLE);
779 schedule(); 1042 schedule();
780 1043
1044 complete(x);
1045
781 /* we are awake, now wait to disappear */ 1046 /* we are awake, now wait to disappear */
782 while (!kthread_should_stop()) { 1047 while (!kthread_should_stop()) {
783 /* 1048 /*
@@ -821,24 +1086,21 @@ trace_selftest_startup_wakeup(struct tracer *trace, struct trace_array *tr)
821 /* reset the max latency */ 1086 /* reset the max latency */
822 tracing_max_latency = 0; 1087 tracing_max_latency = 0;
823 1088
824 /* sleep to let the RT thread sleep too */ 1089 while (p->on_rq) {
825 msleep(100); 1090 /*
1091 * Sleep to make sure the RT thread is asleep too.
1092 * On virtual machines we can't rely on timings,
1093 * but we want to make sure this test still works.
1094 */
1095 msleep(100);
1096 }
826 1097
827 /* 1098 init_completion(&isrt);
828 * Yes this is slightly racy. It is possible that for some
829 * strange reason that the RT thread we created, did not
830 * call schedule for 100ms after doing the completion,
831 * and we do a wakeup on a task that already is awake.
832 * But that is extremely unlikely, and the worst thing that
833 * happens in such a case, is that we disable tracing.
834 * Honestly, if this race does happen something is horrible
835 * wrong with the system.
836 */
837 1099
838 wake_up_process(p); 1100 wake_up_process(p);
839 1101
840 /* give a little time to let the thread wake up */ 1102 /* Wait for the task to wake up */
841 msleep(100); 1103 wait_for_completion(&isrt);
842 1104
843 /* stop the tracing. */ 1105 /* stop the tracing. */
844 tracing_stop(); 1106 tracing_stop();
diff --git a/kernel/trace/trace_stack.c b/kernel/trace/trace_stack.c
index d4545f49242e..0c1b165778e5 100644
--- a/kernel/trace/trace_stack.c
+++ b/kernel/trace/trace_stack.c
@@ -111,7 +111,8 @@ static inline void check_stack(void)
111} 111}
112 112
113static void 113static void
114stack_trace_call(unsigned long ip, unsigned long parent_ip) 114stack_trace_call(unsigned long ip, unsigned long parent_ip,
115 struct ftrace_ops *op, struct pt_regs *pt_regs)
115{ 116{
116 int cpu; 117 int cpu;
117 118
@@ -136,6 +137,7 @@ stack_trace_call(unsigned long ip, unsigned long parent_ip)
136static struct ftrace_ops trace_ops __read_mostly = 137static struct ftrace_ops trace_ops __read_mostly =
137{ 138{
138 .func = stack_trace_call, 139 .func = stack_trace_call,
140 .flags = FTRACE_OPS_FL_RECURSION_SAFE,
139}; 141};
140 142
141static ssize_t 143static ssize_t
diff --git a/kernel/trace/trace_syscalls.c b/kernel/trace/trace_syscalls.c
index 6b245f64c8dd..2485a7d09b11 100644
--- a/kernel/trace/trace_syscalls.c
+++ b/kernel/trace/trace_syscalls.c
@@ -487,7 +487,7 @@ int __init init_ftrace_syscalls(void)
487 487
488 return 0; 488 return 0;
489} 489}
490core_initcall(init_ftrace_syscalls); 490early_initcall(init_ftrace_syscalls);
491 491
492#ifdef CONFIG_PERF_EVENTS 492#ifdef CONFIG_PERF_EVENTS
493 493
diff --git a/kernel/tsacct.c b/kernel/tsacct.c
index 23b4d784ebdd..625df0b44690 100644
--- a/kernel/tsacct.c
+++ b/kernel/tsacct.c
@@ -26,7 +26,9 @@
26/* 26/*
27 * fill in basic accounting fields 27 * fill in basic accounting fields
28 */ 28 */
29void bacct_add_tsk(struct taskstats *stats, struct task_struct *tsk) 29void bacct_add_tsk(struct user_namespace *user_ns,
30 struct pid_namespace *pid_ns,
31 struct taskstats *stats, struct task_struct *tsk)
30{ 32{
31 const struct cred *tcred; 33 const struct cred *tcred;
32 struct timespec uptime, ts; 34 struct timespec uptime, ts;
@@ -55,13 +57,13 @@ void bacct_add_tsk(struct taskstats *stats, struct task_struct *tsk)
55 stats->ac_flag |= AXSIG; 57 stats->ac_flag |= AXSIG;
56 stats->ac_nice = task_nice(tsk); 58 stats->ac_nice = task_nice(tsk);
57 stats->ac_sched = tsk->policy; 59 stats->ac_sched = tsk->policy;
58 stats->ac_pid = tsk->pid; 60 stats->ac_pid = task_pid_nr_ns(tsk, pid_ns);
59 rcu_read_lock(); 61 rcu_read_lock();
60 tcred = __task_cred(tsk); 62 tcred = __task_cred(tsk);
61 stats->ac_uid = tcred->uid; 63 stats->ac_uid = from_kuid_munged(user_ns, tcred->uid);
62 stats->ac_gid = tcred->gid; 64 stats->ac_gid = from_kgid_munged(user_ns, tcred->gid);
63 stats->ac_ppid = pid_alive(tsk) ? 65 stats->ac_ppid = pid_alive(tsk) ?
64 rcu_dereference(tsk->real_parent)->tgid : 0; 66 task_tgid_nr_ns(rcu_dereference(tsk->real_parent), pid_ns) : 0;
65 rcu_read_unlock(); 67 rcu_read_unlock();
66 stats->ac_utime = cputime_to_usecs(tsk->utime); 68 stats->ac_utime = cputime_to_usecs(tsk->utime);
67 stats->ac_stime = cputime_to_usecs(tsk->stime); 69 stats->ac_stime = cputime_to_usecs(tsk->stime);
diff --git a/kernel/user.c b/kernel/user.c
index b815fefbe76f..750acffbe9ec 100644
--- a/kernel/user.c
+++ b/kernel/user.c
@@ -38,6 +38,14 @@ struct user_namespace init_user_ns = {
38 .count = 4294967295U, 38 .count = 4294967295U,
39 }, 39 },
40 }, 40 },
41 .projid_map = {
42 .nr_extents = 1,
43 .extent[0] = {
44 .first = 0,
45 .lower_first = 0,
46 .count = 4294967295U,
47 },
48 },
41 .kref = { 49 .kref = {
42 .refcount = ATOMIC_INIT(3), 50 .refcount = ATOMIC_INIT(3),
43 }, 51 },
diff --git a/kernel/user_namespace.c b/kernel/user_namespace.c
index 86602316422d..456a6b9fba34 100644
--- a/kernel/user_namespace.c
+++ b/kernel/user_namespace.c
@@ -19,6 +19,7 @@
19#include <linux/fs.h> 19#include <linux/fs.h>
20#include <linux/uaccess.h> 20#include <linux/uaccess.h>
21#include <linux/ctype.h> 21#include <linux/ctype.h>
22#include <linux/projid.h>
22 23
23static struct kmem_cache *user_ns_cachep __read_mostly; 24static struct kmem_cache *user_ns_cachep __read_mostly;
24 25
@@ -295,6 +296,75 @@ gid_t from_kgid_munged(struct user_namespace *targ, kgid_t kgid)
295} 296}
296EXPORT_SYMBOL(from_kgid_munged); 297EXPORT_SYMBOL(from_kgid_munged);
297 298
299/**
300 * make_kprojid - Map a user-namespace projid pair into a kprojid.
301 * @ns: User namespace that the projid is in
302 * @projid: Project identifier
303 *
304 * Maps a user-namespace uid pair into a kernel internal kuid,
305 * and returns that kuid.
306 *
307 * When there is no mapping defined for the user-namespace projid
308 * pair INVALID_PROJID is returned. Callers are expected to test
309 * for and handle handle INVALID_PROJID being returned. INVALID_PROJID
310 * may be tested for using projid_valid().
311 */
312kprojid_t make_kprojid(struct user_namespace *ns, projid_t projid)
313{
314 /* Map the uid to a global kernel uid */
315 return KPROJIDT_INIT(map_id_down(&ns->projid_map, projid));
316}
317EXPORT_SYMBOL(make_kprojid);
318
319/**
320 * from_kprojid - Create a projid from a kprojid user-namespace pair.
321 * @targ: The user namespace we want a projid in.
322 * @kprojid: The kernel internal project identifier to start with.
323 *
324 * Map @kprojid into the user-namespace specified by @targ and
325 * return the resulting projid.
326 *
327 * There is always a mapping into the initial user_namespace.
328 *
329 * If @kprojid has no mapping in @targ (projid_t)-1 is returned.
330 */
331projid_t from_kprojid(struct user_namespace *targ, kprojid_t kprojid)
332{
333 /* Map the uid from a global kernel uid */
334 return map_id_up(&targ->projid_map, __kprojid_val(kprojid));
335}
336EXPORT_SYMBOL(from_kprojid);
337
338/**
339 * from_kprojid_munged - Create a projiid from a kprojid user-namespace pair.
340 * @targ: The user namespace we want a projid in.
341 * @kprojid: The kernel internal projid to start with.
342 *
343 * Map @kprojid into the user-namespace specified by @targ and
344 * return the resulting projid.
345 *
346 * There is always a mapping into the initial user_namespace.
347 *
348 * Unlike from_kprojid from_kprojid_munged never fails and always
349 * returns a valid projid. This makes from_kprojid_munged
350 * appropriate for use in syscalls like stat and where
351 * failing the system call and failing to provide a valid projid are
352 * not an options.
353 *
354 * If @kprojid has no mapping in @targ OVERFLOW_PROJID is returned.
355 */
356projid_t from_kprojid_munged(struct user_namespace *targ, kprojid_t kprojid)
357{
358 projid_t projid;
359 projid = from_kprojid(targ, kprojid);
360
361 if (projid == (projid_t) -1)
362 projid = OVERFLOW_PROJID;
363 return projid;
364}
365EXPORT_SYMBOL(from_kprojid_munged);
366
367
298static int uid_m_show(struct seq_file *seq, void *v) 368static int uid_m_show(struct seq_file *seq, void *v)
299{ 369{
300 struct user_namespace *ns = seq->private; 370 struct user_namespace *ns = seq->private;
@@ -337,6 +407,27 @@ static int gid_m_show(struct seq_file *seq, void *v)
337 return 0; 407 return 0;
338} 408}
339 409
410static int projid_m_show(struct seq_file *seq, void *v)
411{
412 struct user_namespace *ns = seq->private;
413 struct uid_gid_extent *extent = v;
414 struct user_namespace *lower_ns;
415 projid_t lower;
416
417 lower_ns = seq_user_ns(seq);
418 if ((lower_ns == ns) && lower_ns->parent)
419 lower_ns = lower_ns->parent;
420
421 lower = from_kprojid(lower_ns, KPROJIDT_INIT(extent->lower_first));
422
423 seq_printf(seq, "%10u %10u %10u\n",
424 extent->first,
425 lower,
426 extent->count);
427
428 return 0;
429}
430
340static void *m_start(struct seq_file *seq, loff_t *ppos, struct uid_gid_map *map) 431static void *m_start(struct seq_file *seq, loff_t *ppos, struct uid_gid_map *map)
341{ 432{
342 struct uid_gid_extent *extent = NULL; 433 struct uid_gid_extent *extent = NULL;
@@ -362,6 +453,13 @@ static void *gid_m_start(struct seq_file *seq, loff_t *ppos)
362 return m_start(seq, ppos, &ns->gid_map); 453 return m_start(seq, ppos, &ns->gid_map);
363} 454}
364 455
456static void *projid_m_start(struct seq_file *seq, loff_t *ppos)
457{
458 struct user_namespace *ns = seq->private;
459
460 return m_start(seq, ppos, &ns->projid_map);
461}
462
365static void *m_next(struct seq_file *seq, void *v, loff_t *pos) 463static void *m_next(struct seq_file *seq, void *v, loff_t *pos)
366{ 464{
367 (*pos)++; 465 (*pos)++;
@@ -387,6 +485,13 @@ struct seq_operations proc_gid_seq_operations = {
387 .show = gid_m_show, 485 .show = gid_m_show,
388}; 486};
389 487
488struct seq_operations proc_projid_seq_operations = {
489 .start = projid_m_start,
490 .stop = m_stop,
491 .next = m_next,
492 .show = projid_m_show,
493};
494
390static DEFINE_MUTEX(id_map_mutex); 495static DEFINE_MUTEX(id_map_mutex);
391 496
392static ssize_t map_write(struct file *file, const char __user *buf, 497static ssize_t map_write(struct file *file, const char __user *buf,
@@ -434,7 +539,7 @@ static ssize_t map_write(struct file *file, const char __user *buf,
434 /* Require the appropriate privilege CAP_SETUID or CAP_SETGID 539 /* Require the appropriate privilege CAP_SETUID or CAP_SETGID
435 * over the user namespace in order to set the id mapping. 540 * over the user namespace in order to set the id mapping.
436 */ 541 */
437 if (!ns_capable(ns, cap_setid)) 542 if (cap_valid(cap_setid) && !ns_capable(ns, cap_setid))
438 goto out; 543 goto out;
439 544
440 /* Get a buffer */ 545 /* Get a buffer */
@@ -584,9 +689,30 @@ ssize_t proc_gid_map_write(struct file *file, const char __user *buf, size_t siz
584 &ns->gid_map, &ns->parent->gid_map); 689 &ns->gid_map, &ns->parent->gid_map);
585} 690}
586 691
692ssize_t proc_projid_map_write(struct file *file, const char __user *buf, size_t size, loff_t *ppos)
693{
694 struct seq_file *seq = file->private_data;
695 struct user_namespace *ns = seq->private;
696 struct user_namespace *seq_ns = seq_user_ns(seq);
697
698 if (!ns->parent)
699 return -EPERM;
700
701 if ((seq_ns != ns) && (seq_ns != ns->parent))
702 return -EPERM;
703
704 /* Anyone can set any valid project id no capability needed */
705 return map_write(file, buf, size, ppos, -1,
706 &ns->projid_map, &ns->parent->projid_map);
707}
708
587static bool new_idmap_permitted(struct user_namespace *ns, int cap_setid, 709static bool new_idmap_permitted(struct user_namespace *ns, int cap_setid,
588 struct uid_gid_map *new_map) 710 struct uid_gid_map *new_map)
589{ 711{
712 /* Allow anyone to set a mapping that doesn't require privilege */
713 if (!cap_valid(cap_setid))
714 return true;
715
590 /* Allow the specified ids if we have the appropriate capability 716 /* Allow the specified ids if we have the appropriate capability
591 * (CAP_SETUID or CAP_SETGID) over the parent user namespace. 717 * (CAP_SETUID or CAP_SETGID) over the parent user namespace.
592 */ 718 */
diff --git a/kernel/watchdog.c b/kernel/watchdog.c
index 4b1dfba70f7c..9d4c8d5a1f53 100644
--- a/kernel/watchdog.c
+++ b/kernel/watchdog.c
@@ -22,6 +22,7 @@
22#include <linux/notifier.h> 22#include <linux/notifier.h>
23#include <linux/module.h> 23#include <linux/module.h>
24#include <linux/sysctl.h> 24#include <linux/sysctl.h>
25#include <linux/smpboot.h>
25 26
26#include <asm/irq_regs.h> 27#include <asm/irq_regs.h>
27#include <linux/kvm_para.h> 28#include <linux/kvm_para.h>
@@ -29,16 +30,18 @@
29 30
30int watchdog_enabled = 1; 31int watchdog_enabled = 1;
31int __read_mostly watchdog_thresh = 10; 32int __read_mostly watchdog_thresh = 10;
33static int __read_mostly watchdog_disabled;
32 34
33static DEFINE_PER_CPU(unsigned long, watchdog_touch_ts); 35static DEFINE_PER_CPU(unsigned long, watchdog_touch_ts);
34static DEFINE_PER_CPU(struct task_struct *, softlockup_watchdog); 36static DEFINE_PER_CPU(struct task_struct *, softlockup_watchdog);
35static DEFINE_PER_CPU(struct hrtimer, watchdog_hrtimer); 37static DEFINE_PER_CPU(struct hrtimer, watchdog_hrtimer);
36static DEFINE_PER_CPU(bool, softlockup_touch_sync); 38static DEFINE_PER_CPU(bool, softlockup_touch_sync);
37static DEFINE_PER_CPU(bool, soft_watchdog_warn); 39static DEFINE_PER_CPU(bool, soft_watchdog_warn);
40static DEFINE_PER_CPU(unsigned long, hrtimer_interrupts);
41static DEFINE_PER_CPU(unsigned long, soft_lockup_hrtimer_cnt);
38#ifdef CONFIG_HARDLOCKUP_DETECTOR 42#ifdef CONFIG_HARDLOCKUP_DETECTOR
39static DEFINE_PER_CPU(bool, hard_watchdog_warn); 43static DEFINE_PER_CPU(bool, hard_watchdog_warn);
40static DEFINE_PER_CPU(bool, watchdog_nmi_touch); 44static DEFINE_PER_CPU(bool, watchdog_nmi_touch);
41static DEFINE_PER_CPU(unsigned long, hrtimer_interrupts);
42static DEFINE_PER_CPU(unsigned long, hrtimer_interrupts_saved); 45static DEFINE_PER_CPU(unsigned long, hrtimer_interrupts_saved);
43static DEFINE_PER_CPU(struct perf_event *, watchdog_ev); 46static DEFINE_PER_CPU(struct perf_event *, watchdog_ev);
44#endif 47#endif
@@ -248,13 +251,15 @@ static void watchdog_overflow_callback(struct perf_event *event,
248 __this_cpu_write(hard_watchdog_warn, false); 251 __this_cpu_write(hard_watchdog_warn, false);
249 return; 252 return;
250} 253}
254#endif /* CONFIG_HARDLOCKUP_DETECTOR */
255
251static void watchdog_interrupt_count(void) 256static void watchdog_interrupt_count(void)
252{ 257{
253 __this_cpu_inc(hrtimer_interrupts); 258 __this_cpu_inc(hrtimer_interrupts);
254} 259}
255#else 260
256static inline void watchdog_interrupt_count(void) { return; } 261static int watchdog_nmi_enable(unsigned int cpu);
257#endif /* CONFIG_HARDLOCKUP_DETECTOR */ 262static void watchdog_nmi_disable(unsigned int cpu);
258 263
259/* watchdog kicker functions */ 264/* watchdog kicker functions */
260static enum hrtimer_restart watchdog_timer_fn(struct hrtimer *hrtimer) 265static enum hrtimer_restart watchdog_timer_fn(struct hrtimer *hrtimer)
@@ -327,49 +332,68 @@ static enum hrtimer_restart watchdog_timer_fn(struct hrtimer *hrtimer)
327 return HRTIMER_RESTART; 332 return HRTIMER_RESTART;
328} 333}
329 334
335static void watchdog_set_prio(unsigned int policy, unsigned int prio)
336{
337 struct sched_param param = { .sched_priority = prio };
330 338
331/* 339 sched_setscheduler(current, policy, &param);
332 * The watchdog thread - touches the timestamp. 340}
333 */ 341
334static int watchdog(void *unused) 342static void watchdog_enable(unsigned int cpu)
335{ 343{
336 struct sched_param param = { .sched_priority = 0 };
337 struct hrtimer *hrtimer = &__raw_get_cpu_var(watchdog_hrtimer); 344 struct hrtimer *hrtimer = &__raw_get_cpu_var(watchdog_hrtimer);
338 345
339 /* initialize timestamp */ 346 if (!watchdog_enabled) {
340 __touch_watchdog(); 347 kthread_park(current);
348 return;
349 }
350
351 /* Enable the perf event */
352 watchdog_nmi_enable(cpu);
341 353
342 /* kick off the timer for the hardlockup detector */ 354 /* kick off the timer for the hardlockup detector */
355 hrtimer_init(hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
356 hrtimer->function = watchdog_timer_fn;
357
343 /* done here because hrtimer_start can only pin to smp_processor_id() */ 358 /* done here because hrtimer_start can only pin to smp_processor_id() */
344 hrtimer_start(hrtimer, ns_to_ktime(get_sample_period()), 359 hrtimer_start(hrtimer, ns_to_ktime(get_sample_period()),
345 HRTIMER_MODE_REL_PINNED); 360 HRTIMER_MODE_REL_PINNED);
346 361
347 set_current_state(TASK_INTERRUPTIBLE); 362 /* initialize timestamp */
348 /* 363 watchdog_set_prio(SCHED_FIFO, MAX_RT_PRIO - 1);
349 * Run briefly (kicked by the hrtimer callback function) once every 364 __touch_watchdog();
350 * get_sample_period() seconds (4 seconds by default) to reset the 365}
351 * softlockup timestamp. If this gets delayed for more than
352 * 2*watchdog_thresh seconds then the debug-printout triggers in
353 * watchdog_timer_fn().
354 */
355 while (!kthread_should_stop()) {
356 __touch_watchdog();
357 schedule();
358 366
359 if (kthread_should_stop()) 367static void watchdog_disable(unsigned int cpu)
360 break; 368{
369 struct hrtimer *hrtimer = &__raw_get_cpu_var(watchdog_hrtimer);
361 370
362 set_current_state(TASK_INTERRUPTIBLE); 371 watchdog_set_prio(SCHED_NORMAL, 0);
363 } 372 hrtimer_cancel(hrtimer);
364 /* 373 /* disable the perf event */
365 * Drop the policy/priority elevation during thread exit to avoid a 374 watchdog_nmi_disable(cpu);
366 * scheduling latency spike.
367 */
368 __set_current_state(TASK_RUNNING);
369 sched_setscheduler(current, SCHED_NORMAL, &param);
370 return 0;
371} 375}
372 376
377static int watchdog_should_run(unsigned int cpu)
378{
379 return __this_cpu_read(hrtimer_interrupts) !=
380 __this_cpu_read(soft_lockup_hrtimer_cnt);
381}
382
383/*
384 * The watchdog thread function - touches the timestamp.
385 *
386 * It only runs once every get_sample_period() seconds (4 seconds by
387 * default) to reset the softlockup timestamp. If this gets delayed
388 * for more than 2*watchdog_thresh seconds then the debug-printout
389 * triggers in watchdog_timer_fn().
390 */
391static void watchdog(unsigned int cpu)
392{
393 __this_cpu_write(soft_lockup_hrtimer_cnt,
394 __this_cpu_read(hrtimer_interrupts));
395 __touch_watchdog();
396}
373 397
374#ifdef CONFIG_HARDLOCKUP_DETECTOR 398#ifdef CONFIG_HARDLOCKUP_DETECTOR
375/* 399/*
@@ -379,7 +403,7 @@ static int watchdog(void *unused)
379 */ 403 */
380static unsigned long cpu0_err; 404static unsigned long cpu0_err;
381 405
382static int watchdog_nmi_enable(int cpu) 406static int watchdog_nmi_enable(unsigned int cpu)
383{ 407{
384 struct perf_event_attr *wd_attr; 408 struct perf_event_attr *wd_attr;
385 struct perf_event *event = per_cpu(watchdog_ev, cpu); 409 struct perf_event *event = per_cpu(watchdog_ev, cpu);
@@ -433,7 +457,7 @@ out:
433 return 0; 457 return 0;
434} 458}
435 459
436static void watchdog_nmi_disable(int cpu) 460static void watchdog_nmi_disable(unsigned int cpu)
437{ 461{
438 struct perf_event *event = per_cpu(watchdog_ev, cpu); 462 struct perf_event *event = per_cpu(watchdog_ev, cpu);
439 463
@@ -447,107 +471,35 @@ static void watchdog_nmi_disable(int cpu)
447 return; 471 return;
448} 472}
449#else 473#else
450static int watchdog_nmi_enable(int cpu) { return 0; } 474static int watchdog_nmi_enable(unsigned int cpu) { return 0; }
451static void watchdog_nmi_disable(int cpu) { return; } 475static void watchdog_nmi_disable(unsigned int cpu) { return; }
452#endif /* CONFIG_HARDLOCKUP_DETECTOR */ 476#endif /* CONFIG_HARDLOCKUP_DETECTOR */
453 477
454/* prepare/enable/disable routines */ 478/* prepare/enable/disable routines */
455static void watchdog_prepare_cpu(int cpu)
456{
457 struct hrtimer *hrtimer = &per_cpu(watchdog_hrtimer, cpu);
458
459 WARN_ON(per_cpu(softlockup_watchdog, cpu));
460 hrtimer_init(hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
461 hrtimer->function = watchdog_timer_fn;
462}
463
464static int watchdog_enable(int cpu)
465{
466 struct task_struct *p = per_cpu(softlockup_watchdog, cpu);
467 int err = 0;
468
469 /* enable the perf event */
470 err = watchdog_nmi_enable(cpu);
471
472 /* Regardless of err above, fall through and start softlockup */
473
474 /* create the watchdog thread */
475 if (!p) {
476 struct sched_param param = { .sched_priority = MAX_RT_PRIO-1 };
477 p = kthread_create_on_node(watchdog, NULL, cpu_to_node(cpu), "watchdog/%d", cpu);
478 if (IS_ERR(p)) {
479 pr_err("softlockup watchdog for %i failed\n", cpu);
480 if (!err) {
481 /* if hardlockup hasn't already set this */
482 err = PTR_ERR(p);
483 /* and disable the perf event */
484 watchdog_nmi_disable(cpu);
485 }
486 goto out;
487 }
488 sched_setscheduler(p, SCHED_FIFO, &param);
489 kthread_bind(p, cpu);
490 per_cpu(watchdog_touch_ts, cpu) = 0;
491 per_cpu(softlockup_watchdog, cpu) = p;
492 wake_up_process(p);
493 }
494
495out:
496 return err;
497}
498
499static void watchdog_disable(int cpu)
500{
501 struct task_struct *p = per_cpu(softlockup_watchdog, cpu);
502 struct hrtimer *hrtimer = &per_cpu(watchdog_hrtimer, cpu);
503
504 /*
505 * cancel the timer first to stop incrementing the stats
506 * and waking up the kthread
507 */
508 hrtimer_cancel(hrtimer);
509
510 /* disable the perf event */
511 watchdog_nmi_disable(cpu);
512
513 /* stop the watchdog thread */
514 if (p) {
515 per_cpu(softlockup_watchdog, cpu) = NULL;
516 kthread_stop(p);
517 }
518}
519
520/* sysctl functions */ 479/* sysctl functions */
521#ifdef CONFIG_SYSCTL 480#ifdef CONFIG_SYSCTL
522static void watchdog_enable_all_cpus(void) 481static void watchdog_enable_all_cpus(void)
523{ 482{
524 int cpu; 483 unsigned int cpu;
525
526 watchdog_enabled = 0;
527
528 for_each_online_cpu(cpu)
529 if (!watchdog_enable(cpu))
530 /* if any cpu succeeds, watchdog is considered
531 enabled for the system */
532 watchdog_enabled = 1;
533
534 if (!watchdog_enabled)
535 pr_err("failed to be enabled on some cpus\n");
536 484
485 if (watchdog_disabled) {
486 watchdog_disabled = 0;
487 for_each_online_cpu(cpu)
488 kthread_unpark(per_cpu(softlockup_watchdog, cpu));
489 }
537} 490}
538 491
539static void watchdog_disable_all_cpus(void) 492static void watchdog_disable_all_cpus(void)
540{ 493{
541 int cpu; 494 unsigned int cpu;
542
543 for_each_online_cpu(cpu)
544 watchdog_disable(cpu);
545 495
546 /* if all watchdogs are disabled, then they are disabled for the system */ 496 if (!watchdog_disabled) {
547 watchdog_enabled = 0; 497 watchdog_disabled = 1;
498 for_each_online_cpu(cpu)
499 kthread_park(per_cpu(softlockup_watchdog, cpu));
500 }
548} 501}
549 502
550
551/* 503/*
552 * proc handler for /proc/sys/kernel/nmi_watchdog,watchdog_thresh 504 * proc handler for /proc/sys/kernel/nmi_watchdog,watchdog_thresh
553 */ 505 */
@@ -557,73 +509,36 @@ int proc_dowatchdog(struct ctl_table *table, int write,
557{ 509{
558 int ret; 510 int ret;
559 511
512 if (watchdog_disabled < 0)
513 return -ENODEV;
514
560 ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos); 515 ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos);
561 if (ret || !write) 516 if (ret || !write)
562 goto out; 517 return ret;
563 518
564 if (watchdog_enabled && watchdog_thresh) 519 if (watchdog_enabled && watchdog_thresh)
565 watchdog_enable_all_cpus(); 520 watchdog_enable_all_cpus();
566 else 521 else
567 watchdog_disable_all_cpus(); 522 watchdog_disable_all_cpus();
568 523
569out:
570 return ret; 524 return ret;
571} 525}
572#endif /* CONFIG_SYSCTL */ 526#endif /* CONFIG_SYSCTL */
573 527
574 528static struct smp_hotplug_thread watchdog_threads = {
575/* 529 .store = &softlockup_watchdog,
576 * Create/destroy watchdog threads as CPUs come and go: 530 .thread_should_run = watchdog_should_run,
577 */ 531 .thread_fn = watchdog,
578static int __cpuinit 532 .thread_comm = "watchdog/%u",
579cpu_callback(struct notifier_block *nfb, unsigned long action, void *hcpu) 533 .setup = watchdog_enable,
580{ 534 .park = watchdog_disable,
581 int hotcpu = (unsigned long)hcpu; 535 .unpark = watchdog_enable,
582
583 switch (action) {
584 case CPU_UP_PREPARE:
585 case CPU_UP_PREPARE_FROZEN:
586 watchdog_prepare_cpu(hotcpu);
587 break;
588 case CPU_ONLINE:
589 case CPU_ONLINE_FROZEN:
590 if (watchdog_enabled)
591 watchdog_enable(hotcpu);
592 break;
593#ifdef CONFIG_HOTPLUG_CPU
594 case CPU_UP_CANCELED:
595 case CPU_UP_CANCELED_FROZEN:
596 watchdog_disable(hotcpu);
597 break;
598 case CPU_DEAD:
599 case CPU_DEAD_FROZEN:
600 watchdog_disable(hotcpu);
601 break;
602#endif /* CONFIG_HOTPLUG_CPU */
603 }
604
605 /*
606 * hardlockup and softlockup are not important enough
607 * to block cpu bring up. Just always succeed and
608 * rely on printk output to flag problems.
609 */
610 return NOTIFY_OK;
611}
612
613static struct notifier_block __cpuinitdata cpu_nfb = {
614 .notifier_call = cpu_callback
615}; 536};
616 537
617void __init lockup_detector_init(void) 538void __init lockup_detector_init(void)
618{ 539{
619 void *cpu = (void *)(long)smp_processor_id(); 540 if (smpboot_register_percpu_thread(&watchdog_threads)) {
620 int err; 541 pr_err("Failed to create watchdog threads, disabled\n");
621 542 watchdog_disabled = -ENODEV;
622 err = cpu_callback(&cpu_nfb, CPU_UP_PREPARE, cpu); 543 }
623 WARN_ON(notifier_to_errno(err));
624
625 cpu_callback(&cpu_nfb, CPU_ONLINE, cpu);
626 register_cpu_notifier(&cpu_nfb);
627
628 return;
629} 544}
diff --git a/kernel/workqueue.c b/kernel/workqueue.c
index b80065a2450a..d951daa0ca9a 100644
--- a/kernel/workqueue.c
+++ b/kernel/workqueue.c
@@ -58,7 +58,7 @@ enum {
58 * be executing on any CPU. The gcwq behaves as an unbound one. 58 * be executing on any CPU. The gcwq behaves as an unbound one.
59 * 59 *
60 * Note that DISASSOCIATED can be flipped only while holding 60 * Note that DISASSOCIATED can be flipped only while holding
61 * managership of all pools on the gcwq to avoid changing binding 61 * assoc_mutex of all pools on the gcwq to avoid changing binding
62 * state while create_worker() is in progress. 62 * state while create_worker() is in progress.
63 */ 63 */
64 GCWQ_DISASSOCIATED = 1 << 0, /* cpu can't serve workers */ 64 GCWQ_DISASSOCIATED = 1 << 0, /* cpu can't serve workers */
@@ -73,11 +73,10 @@ enum {
73 WORKER_DIE = 1 << 1, /* die die die */ 73 WORKER_DIE = 1 << 1, /* die die die */
74 WORKER_IDLE = 1 << 2, /* is idle */ 74 WORKER_IDLE = 1 << 2, /* is idle */
75 WORKER_PREP = 1 << 3, /* preparing to run works */ 75 WORKER_PREP = 1 << 3, /* preparing to run works */
76 WORKER_REBIND = 1 << 5, /* mom is home, come back */
77 WORKER_CPU_INTENSIVE = 1 << 6, /* cpu intensive */ 76 WORKER_CPU_INTENSIVE = 1 << 6, /* cpu intensive */
78 WORKER_UNBOUND = 1 << 7, /* worker is unbound */ 77 WORKER_UNBOUND = 1 << 7, /* worker is unbound */
79 78
80 WORKER_NOT_RUNNING = WORKER_PREP | WORKER_REBIND | WORKER_UNBOUND | 79 WORKER_NOT_RUNNING = WORKER_PREP | WORKER_UNBOUND |
81 WORKER_CPU_INTENSIVE, 80 WORKER_CPU_INTENSIVE,
82 81
83 NR_WORKER_POOLS = 2, /* # worker pools per gcwq */ 82 NR_WORKER_POOLS = 2, /* # worker pools per gcwq */
@@ -126,7 +125,6 @@ enum {
126 125
127struct global_cwq; 126struct global_cwq;
128struct worker_pool; 127struct worker_pool;
129struct idle_rebind;
130 128
131/* 129/*
132 * The poor guys doing the actual heavy lifting. All on-duty workers 130 * The poor guys doing the actual heavy lifting. All on-duty workers
@@ -150,7 +148,6 @@ struct worker {
150 int id; /* I: worker id */ 148 int id; /* I: worker id */
151 149
152 /* for rebinding worker to CPU */ 150 /* for rebinding worker to CPU */
153 struct idle_rebind *idle_rebind; /* L: for idle worker */
154 struct work_struct rebind_work; /* L: for busy worker */ 151 struct work_struct rebind_work; /* L: for busy worker */
155}; 152};
156 153
@@ -160,13 +157,15 @@ struct worker_pool {
160 157
161 struct list_head worklist; /* L: list of pending works */ 158 struct list_head worklist; /* L: list of pending works */
162 int nr_workers; /* L: total number of workers */ 159 int nr_workers; /* L: total number of workers */
160
161 /* nr_idle includes the ones off idle_list for rebinding */
163 int nr_idle; /* L: currently idle ones */ 162 int nr_idle; /* L: currently idle ones */
164 163
165 struct list_head idle_list; /* X: list of idle workers */ 164 struct list_head idle_list; /* X: list of idle workers */
166 struct timer_list idle_timer; /* L: worker idle timeout */ 165 struct timer_list idle_timer; /* L: worker idle timeout */
167 struct timer_list mayday_timer; /* L: SOS timer for workers */ 166 struct timer_list mayday_timer; /* L: SOS timer for workers */
168 167
169 struct mutex manager_mutex; /* mutex manager should hold */ 168 struct mutex assoc_mutex; /* protect GCWQ_DISASSOCIATED */
170 struct ida worker_ida; /* L: for worker IDs */ 169 struct ida worker_ida; /* L: for worker IDs */
171}; 170};
172 171
@@ -184,9 +183,8 @@ struct global_cwq {
184 struct hlist_head busy_hash[BUSY_WORKER_HASH_SIZE]; 183 struct hlist_head busy_hash[BUSY_WORKER_HASH_SIZE];
185 /* L: hash of busy workers */ 184 /* L: hash of busy workers */
186 185
187 struct worker_pool pools[2]; /* normal and highpri pools */ 186 struct worker_pool pools[NR_WORKER_POOLS];
188 187 /* normal and highpri pools */
189 wait_queue_head_t rebind_hold; /* rebind hold wait */
190} ____cacheline_aligned_in_smp; 188} ____cacheline_aligned_in_smp;
191 189
192/* 190/*
@@ -269,17 +267,15 @@ struct workqueue_struct {
269}; 267};
270 268
271struct workqueue_struct *system_wq __read_mostly; 269struct workqueue_struct *system_wq __read_mostly;
272struct workqueue_struct *system_long_wq __read_mostly;
273struct workqueue_struct *system_nrt_wq __read_mostly;
274struct workqueue_struct *system_unbound_wq __read_mostly;
275struct workqueue_struct *system_freezable_wq __read_mostly;
276struct workqueue_struct *system_nrt_freezable_wq __read_mostly;
277EXPORT_SYMBOL_GPL(system_wq); 270EXPORT_SYMBOL_GPL(system_wq);
271struct workqueue_struct *system_highpri_wq __read_mostly;
272EXPORT_SYMBOL_GPL(system_highpri_wq);
273struct workqueue_struct *system_long_wq __read_mostly;
278EXPORT_SYMBOL_GPL(system_long_wq); 274EXPORT_SYMBOL_GPL(system_long_wq);
279EXPORT_SYMBOL_GPL(system_nrt_wq); 275struct workqueue_struct *system_unbound_wq __read_mostly;
280EXPORT_SYMBOL_GPL(system_unbound_wq); 276EXPORT_SYMBOL_GPL(system_unbound_wq);
277struct workqueue_struct *system_freezable_wq __read_mostly;
281EXPORT_SYMBOL_GPL(system_freezable_wq); 278EXPORT_SYMBOL_GPL(system_freezable_wq);
282EXPORT_SYMBOL_GPL(system_nrt_freezable_wq);
283 279
284#define CREATE_TRACE_POINTS 280#define CREATE_TRACE_POINTS
285#include <trace/events/workqueue.h> 281#include <trace/events/workqueue.h>
@@ -534,18 +530,24 @@ static int work_next_color(int color)
534} 530}
535 531
536/* 532/*
537 * A work's data points to the cwq with WORK_STRUCT_CWQ set while the 533 * While queued, %WORK_STRUCT_CWQ is set and non flag bits of a work's data
538 * work is on queue. Once execution starts, WORK_STRUCT_CWQ is 534 * contain the pointer to the queued cwq. Once execution starts, the flag
539 * cleared and the work data contains the cpu number it was last on. 535 * is cleared and the high bits contain OFFQ flags and CPU number.
540 * 536 *
541 * set_work_{cwq|cpu}() and clear_work_data() can be used to set the 537 * set_work_cwq(), set_work_cpu_and_clear_pending(), mark_work_canceling()
542 * cwq, cpu or clear work->data. These functions should only be 538 * and clear_work_data() can be used to set the cwq, cpu or clear
543 * called while the work is owned - ie. while the PENDING bit is set. 539 * work->data. These functions should only be called while the work is
540 * owned - ie. while the PENDING bit is set.
544 * 541 *
545 * get_work_[g]cwq() can be used to obtain the gcwq or cwq 542 * get_work_[g]cwq() can be used to obtain the gcwq or cwq corresponding to
546 * corresponding to a work. gcwq is available once the work has been 543 * a work. gcwq is available once the work has been queued anywhere after
547 * queued anywhere after initialization. cwq is available only from 544 * initialization until it is sync canceled. cwq is available only while
548 * queueing until execution starts. 545 * the work item is queued.
546 *
547 * %WORK_OFFQ_CANCELING is used to mark a work item which is being
548 * canceled. While being canceled, a work item may have its PENDING set
549 * but stay off timer and worklist for arbitrarily long and nobody should
550 * try to steal the PENDING bit.
549 */ 551 */
550static inline void set_work_data(struct work_struct *work, unsigned long data, 552static inline void set_work_data(struct work_struct *work, unsigned long data,
551 unsigned long flags) 553 unsigned long flags)
@@ -562,13 +564,22 @@ static void set_work_cwq(struct work_struct *work,
562 WORK_STRUCT_PENDING | WORK_STRUCT_CWQ | extra_flags); 564 WORK_STRUCT_PENDING | WORK_STRUCT_CWQ | extra_flags);
563} 565}
564 566
565static void set_work_cpu(struct work_struct *work, unsigned int cpu) 567static void set_work_cpu_and_clear_pending(struct work_struct *work,
568 unsigned int cpu)
566{ 569{
567 set_work_data(work, cpu << WORK_STRUCT_FLAG_BITS, WORK_STRUCT_PENDING); 570 /*
571 * The following wmb is paired with the implied mb in
572 * test_and_set_bit(PENDING) and ensures all updates to @work made
573 * here are visible to and precede any updates by the next PENDING
574 * owner.
575 */
576 smp_wmb();
577 set_work_data(work, (unsigned long)cpu << WORK_OFFQ_CPU_SHIFT, 0);
568} 578}
569 579
570static void clear_work_data(struct work_struct *work) 580static void clear_work_data(struct work_struct *work)
571{ 581{
582 smp_wmb(); /* see set_work_cpu_and_clear_pending() */
572 set_work_data(work, WORK_STRUCT_NO_CPU, 0); 583 set_work_data(work, WORK_STRUCT_NO_CPU, 0);
573} 584}
574 585
@@ -591,7 +602,7 @@ static struct global_cwq *get_work_gcwq(struct work_struct *work)
591 return ((struct cpu_workqueue_struct *) 602 return ((struct cpu_workqueue_struct *)
592 (data & WORK_STRUCT_WQ_DATA_MASK))->pool->gcwq; 603 (data & WORK_STRUCT_WQ_DATA_MASK))->pool->gcwq;
593 604
594 cpu = data >> WORK_STRUCT_FLAG_BITS; 605 cpu = data >> WORK_OFFQ_CPU_SHIFT;
595 if (cpu == WORK_CPU_NONE) 606 if (cpu == WORK_CPU_NONE)
596 return NULL; 607 return NULL;
597 608
@@ -599,6 +610,22 @@ static struct global_cwq *get_work_gcwq(struct work_struct *work)
599 return get_gcwq(cpu); 610 return get_gcwq(cpu);
600} 611}
601 612
613static void mark_work_canceling(struct work_struct *work)
614{
615 struct global_cwq *gcwq = get_work_gcwq(work);
616 unsigned long cpu = gcwq ? gcwq->cpu : WORK_CPU_NONE;
617
618 set_work_data(work, (cpu << WORK_OFFQ_CPU_SHIFT) | WORK_OFFQ_CANCELING,
619 WORK_STRUCT_PENDING);
620}
621
622static bool work_is_canceling(struct work_struct *work)
623{
624 unsigned long data = atomic_long_read(&work->data);
625
626 return !(data & WORK_STRUCT_CWQ) && (data & WORK_OFFQ_CANCELING);
627}
628
602/* 629/*
603 * Policy functions. These define the policies on how the global worker 630 * Policy functions. These define the policies on how the global worker
604 * pools are managed. Unless noted otherwise, these functions assume that 631 * pools are managed. Unless noted otherwise, these functions assume that
@@ -657,6 +684,13 @@ static bool too_many_workers(struct worker_pool *pool)
657 int nr_idle = pool->nr_idle + managing; /* manager is considered idle */ 684 int nr_idle = pool->nr_idle + managing; /* manager is considered idle */
658 int nr_busy = pool->nr_workers - nr_idle; 685 int nr_busy = pool->nr_workers - nr_idle;
659 686
687 /*
688 * nr_idle and idle_list may disagree if idle rebinding is in
689 * progress. Never return %true if idle_list is empty.
690 */
691 if (list_empty(&pool->idle_list))
692 return false;
693
660 return nr_idle > 2 && (nr_idle - 2) * MAX_IDLE_WORKERS_RATIO >= nr_busy; 694 return nr_idle > 2 && (nr_idle - 2) * MAX_IDLE_WORKERS_RATIO >= nr_busy;
661} 695}
662 696
@@ -903,6 +937,206 @@ static struct worker *find_worker_executing_work(struct global_cwq *gcwq,
903} 937}
904 938
905/** 939/**
940 * move_linked_works - move linked works to a list
941 * @work: start of series of works to be scheduled
942 * @head: target list to append @work to
943 * @nextp: out paramter for nested worklist walking
944 *
945 * Schedule linked works starting from @work to @head. Work series to
946 * be scheduled starts at @work and includes any consecutive work with
947 * WORK_STRUCT_LINKED set in its predecessor.
948 *
949 * If @nextp is not NULL, it's updated to point to the next work of
950 * the last scheduled work. This allows move_linked_works() to be
951 * nested inside outer list_for_each_entry_safe().
952 *
953 * CONTEXT:
954 * spin_lock_irq(gcwq->lock).
955 */
956static void move_linked_works(struct work_struct *work, struct list_head *head,
957 struct work_struct **nextp)
958{
959 struct work_struct *n;
960
961 /*
962 * Linked worklist will always end before the end of the list,
963 * use NULL for list head.
964 */
965 list_for_each_entry_safe_from(work, n, NULL, entry) {
966 list_move_tail(&work->entry, head);
967 if (!(*work_data_bits(work) & WORK_STRUCT_LINKED))
968 break;
969 }
970
971 /*
972 * If we're already inside safe list traversal and have moved
973 * multiple works to the scheduled queue, the next position
974 * needs to be updated.
975 */
976 if (nextp)
977 *nextp = n;
978}
979
980static void cwq_activate_delayed_work(struct work_struct *work)
981{
982 struct cpu_workqueue_struct *cwq = get_work_cwq(work);
983
984 trace_workqueue_activate_work(work);
985 move_linked_works(work, &cwq->pool->worklist, NULL);
986 __clear_bit(WORK_STRUCT_DELAYED_BIT, work_data_bits(work));
987 cwq->nr_active++;
988}
989
990static void cwq_activate_first_delayed(struct cpu_workqueue_struct *cwq)
991{
992 struct work_struct *work = list_first_entry(&cwq->delayed_works,
993 struct work_struct, entry);
994
995 cwq_activate_delayed_work(work);
996}
997
998/**
999 * cwq_dec_nr_in_flight - decrement cwq's nr_in_flight
1000 * @cwq: cwq of interest
1001 * @color: color of work which left the queue
1002 *
1003 * A work either has completed or is removed from pending queue,
1004 * decrement nr_in_flight of its cwq and handle workqueue flushing.
1005 *
1006 * CONTEXT:
1007 * spin_lock_irq(gcwq->lock).
1008 */
1009static void cwq_dec_nr_in_flight(struct cpu_workqueue_struct *cwq, int color)
1010{
1011 /* ignore uncolored works */
1012 if (color == WORK_NO_COLOR)
1013 return;
1014
1015 cwq->nr_in_flight[color]--;
1016
1017 cwq->nr_active--;
1018 if (!list_empty(&cwq->delayed_works)) {
1019 /* one down, submit a delayed one */
1020 if (cwq->nr_active < cwq->max_active)
1021 cwq_activate_first_delayed(cwq);
1022 }
1023
1024 /* is flush in progress and are we at the flushing tip? */
1025 if (likely(cwq->flush_color != color))
1026 return;
1027
1028 /* are there still in-flight works? */
1029 if (cwq->nr_in_flight[color])
1030 return;
1031
1032 /* this cwq is done, clear flush_color */
1033 cwq->flush_color = -1;
1034
1035 /*
1036 * If this was the last cwq, wake up the first flusher. It
1037 * will handle the rest.
1038 */
1039 if (atomic_dec_and_test(&cwq->wq->nr_cwqs_to_flush))
1040 complete(&cwq->wq->first_flusher->done);
1041}
1042
1043/**
1044 * try_to_grab_pending - steal work item from worklist and disable irq
1045 * @work: work item to steal
1046 * @is_dwork: @work is a delayed_work
1047 * @flags: place to store irq state
1048 *
1049 * Try to grab PENDING bit of @work. This function can handle @work in any
1050 * stable state - idle, on timer or on worklist. Return values are
1051 *
1052 * 1 if @work was pending and we successfully stole PENDING
1053 * 0 if @work was idle and we claimed PENDING
1054 * -EAGAIN if PENDING couldn't be grabbed at the moment, safe to busy-retry
1055 * -ENOENT if someone else is canceling @work, this state may persist
1056 * for arbitrarily long
1057 *
1058 * On >= 0 return, the caller owns @work's PENDING bit. To avoid getting
1059 * interrupted while holding PENDING and @work off queue, irq must be
1060 * disabled on entry. This, combined with delayed_work->timer being
1061 * irqsafe, ensures that we return -EAGAIN for finite short period of time.
1062 *
1063 * On successful return, >= 0, irq is disabled and the caller is
1064 * responsible for releasing it using local_irq_restore(*@flags).
1065 *
1066 * This function is safe to call from any context including IRQ handler.
1067 */
1068static int try_to_grab_pending(struct work_struct *work, bool is_dwork,
1069 unsigned long *flags)
1070{
1071 struct global_cwq *gcwq;
1072
1073 local_irq_save(*flags);
1074
1075 /* try to steal the timer if it exists */
1076 if (is_dwork) {
1077 struct delayed_work *dwork = to_delayed_work(work);
1078
1079 /*
1080 * dwork->timer is irqsafe. If del_timer() fails, it's
1081 * guaranteed that the timer is not queued anywhere and not
1082 * running on the local CPU.
1083 */
1084 if (likely(del_timer(&dwork->timer)))
1085 return 1;
1086 }
1087
1088 /* try to claim PENDING the normal way */
1089 if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work)))
1090 return 0;
1091
1092 /*
1093 * The queueing is in progress, or it is already queued. Try to
1094 * steal it from ->worklist without clearing WORK_STRUCT_PENDING.
1095 */
1096 gcwq = get_work_gcwq(work);
1097 if (!gcwq)
1098 goto fail;
1099
1100 spin_lock(&gcwq->lock);
1101 if (!list_empty(&work->entry)) {
1102 /*
1103 * This work is queued, but perhaps we locked the wrong gcwq.
1104 * In that case we must see the new value after rmb(), see
1105 * insert_work()->wmb().
1106 */
1107 smp_rmb();
1108 if (gcwq == get_work_gcwq(work)) {
1109 debug_work_deactivate(work);
1110
1111 /*
1112 * A delayed work item cannot be grabbed directly
1113 * because it might have linked NO_COLOR work items
1114 * which, if left on the delayed_list, will confuse
1115 * cwq->nr_active management later on and cause
1116 * stall. Make sure the work item is activated
1117 * before grabbing.
1118 */
1119 if (*work_data_bits(work) & WORK_STRUCT_DELAYED)
1120 cwq_activate_delayed_work(work);
1121
1122 list_del_init(&work->entry);
1123 cwq_dec_nr_in_flight(get_work_cwq(work),
1124 get_work_color(work));
1125
1126 spin_unlock(&gcwq->lock);
1127 return 1;
1128 }
1129 }
1130 spin_unlock(&gcwq->lock);
1131fail:
1132 local_irq_restore(*flags);
1133 if (work_is_canceling(work))
1134 return -ENOENT;
1135 cpu_relax();
1136 return -EAGAIN;
1137}
1138
1139/**
906 * insert_work - insert a work into gcwq 1140 * insert_work - insert a work into gcwq
907 * @cwq: cwq @work belongs to 1141 * @cwq: cwq @work belongs to
908 * @work: work to insert 1142 * @work: work to insert
@@ -982,7 +1216,15 @@ static void __queue_work(unsigned int cpu, struct workqueue_struct *wq,
982 struct cpu_workqueue_struct *cwq; 1216 struct cpu_workqueue_struct *cwq;
983 struct list_head *worklist; 1217 struct list_head *worklist;
984 unsigned int work_flags; 1218 unsigned int work_flags;
985 unsigned long flags; 1219 unsigned int req_cpu = cpu;
1220
1221 /*
1222 * While a work item is PENDING && off queue, a task trying to
1223 * steal the PENDING will busy-loop waiting for it to either get
1224 * queued or lose PENDING. Grabbing PENDING and queueing should
1225 * happen with IRQ disabled.
1226 */
1227 WARN_ON_ONCE(!irqs_disabled());
986 1228
987 debug_work_activate(work); 1229 debug_work_activate(work);
988 1230
@@ -995,21 +1237,22 @@ static void __queue_work(unsigned int cpu, struct workqueue_struct *wq,
995 if (!(wq->flags & WQ_UNBOUND)) { 1237 if (!(wq->flags & WQ_UNBOUND)) {
996 struct global_cwq *last_gcwq; 1238 struct global_cwq *last_gcwq;
997 1239
998 if (unlikely(cpu == WORK_CPU_UNBOUND)) 1240 if (cpu == WORK_CPU_UNBOUND)
999 cpu = raw_smp_processor_id(); 1241 cpu = raw_smp_processor_id();
1000 1242
1001 /* 1243 /*
1002 * It's multi cpu. If @wq is non-reentrant and @work 1244 * It's multi cpu. If @work was previously on a different
1003 * was previously on a different cpu, it might still 1245 * cpu, it might still be running there, in which case the
1004 * be running there, in which case the work needs to 1246 * work needs to be queued on that cpu to guarantee
1005 * be queued on that cpu to guarantee non-reentrance. 1247 * non-reentrancy.
1006 */ 1248 */
1007 gcwq = get_gcwq(cpu); 1249 gcwq = get_gcwq(cpu);
1008 if (wq->flags & WQ_NON_REENTRANT && 1250 last_gcwq = get_work_gcwq(work);
1009 (last_gcwq = get_work_gcwq(work)) && last_gcwq != gcwq) { 1251
1252 if (last_gcwq && last_gcwq != gcwq) {
1010 struct worker *worker; 1253 struct worker *worker;
1011 1254
1012 spin_lock_irqsave(&last_gcwq->lock, flags); 1255 spin_lock(&last_gcwq->lock);
1013 1256
1014 worker = find_worker_executing_work(last_gcwq, work); 1257 worker = find_worker_executing_work(last_gcwq, work);
1015 1258
@@ -1017,22 +1260,23 @@ static void __queue_work(unsigned int cpu, struct workqueue_struct *wq,
1017 gcwq = last_gcwq; 1260 gcwq = last_gcwq;
1018 else { 1261 else {
1019 /* meh... not running there, queue here */ 1262 /* meh... not running there, queue here */
1020 spin_unlock_irqrestore(&last_gcwq->lock, flags); 1263 spin_unlock(&last_gcwq->lock);
1021 spin_lock_irqsave(&gcwq->lock, flags); 1264 spin_lock(&gcwq->lock);
1022 } 1265 }
1023 } else 1266 } else {
1024 spin_lock_irqsave(&gcwq->lock, flags); 1267 spin_lock(&gcwq->lock);
1268 }
1025 } else { 1269 } else {
1026 gcwq = get_gcwq(WORK_CPU_UNBOUND); 1270 gcwq = get_gcwq(WORK_CPU_UNBOUND);
1027 spin_lock_irqsave(&gcwq->lock, flags); 1271 spin_lock(&gcwq->lock);
1028 } 1272 }
1029 1273
1030 /* gcwq determined, get cwq and queue */ 1274 /* gcwq determined, get cwq and queue */
1031 cwq = get_cwq(gcwq->cpu, wq); 1275 cwq = get_cwq(gcwq->cpu, wq);
1032 trace_workqueue_queue_work(cpu, cwq, work); 1276 trace_workqueue_queue_work(req_cpu, cwq, work);
1033 1277
1034 if (WARN_ON(!list_empty(&work->entry))) { 1278 if (WARN_ON(!list_empty(&work->entry))) {
1035 spin_unlock_irqrestore(&gcwq->lock, flags); 1279 spin_unlock(&gcwq->lock);
1036 return; 1280 return;
1037 } 1281 }
1038 1282
@@ -1050,79 +1294,110 @@ static void __queue_work(unsigned int cpu, struct workqueue_struct *wq,
1050 1294
1051 insert_work(cwq, work, worklist, work_flags); 1295 insert_work(cwq, work, worklist, work_flags);
1052 1296
1053 spin_unlock_irqrestore(&gcwq->lock, flags); 1297 spin_unlock(&gcwq->lock);
1054} 1298}
1055 1299
1056/** 1300/**
1057 * queue_work - queue work on a workqueue 1301 * queue_work_on - queue work on specific cpu
1302 * @cpu: CPU number to execute work on
1058 * @wq: workqueue to use 1303 * @wq: workqueue to use
1059 * @work: work to queue 1304 * @work: work to queue
1060 * 1305 *
1061 * Returns 0 if @work was already on a queue, non-zero otherwise. 1306 * Returns %false if @work was already on a queue, %true otherwise.
1062 * 1307 *
1063 * We queue the work to the CPU on which it was submitted, but if the CPU dies 1308 * We queue the work to a specific CPU, the caller must ensure it
1064 * it can be processed by another CPU. 1309 * can't go away.
1065 */ 1310 */
1066int queue_work(struct workqueue_struct *wq, struct work_struct *work) 1311bool queue_work_on(int cpu, struct workqueue_struct *wq,
1312 struct work_struct *work)
1067{ 1313{
1068 int ret; 1314 bool ret = false;
1315 unsigned long flags;
1069 1316
1070 ret = queue_work_on(get_cpu(), wq, work); 1317 local_irq_save(flags);
1071 put_cpu();
1072 1318
1319 if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work))) {
1320 __queue_work(cpu, wq, work);
1321 ret = true;
1322 }
1323
1324 local_irq_restore(flags);
1073 return ret; 1325 return ret;
1074} 1326}
1075EXPORT_SYMBOL_GPL(queue_work); 1327EXPORT_SYMBOL_GPL(queue_work_on);
1076 1328
1077/** 1329/**
1078 * queue_work_on - queue work on specific cpu 1330 * queue_work - queue work on a workqueue
1079 * @cpu: CPU number to execute work on
1080 * @wq: workqueue to use 1331 * @wq: workqueue to use
1081 * @work: work to queue 1332 * @work: work to queue
1082 * 1333 *
1083 * Returns 0 if @work was already on a queue, non-zero otherwise. 1334 * Returns %false if @work was already on a queue, %true otherwise.
1084 * 1335 *
1085 * We queue the work to a specific CPU, the caller must ensure it 1336 * We queue the work to the CPU on which it was submitted, but if the CPU dies
1086 * can't go away. 1337 * it can be processed by another CPU.
1087 */ 1338 */
1088int 1339bool queue_work(struct workqueue_struct *wq, struct work_struct *work)
1089queue_work_on(int cpu, struct workqueue_struct *wq, struct work_struct *work)
1090{ 1340{
1091 int ret = 0; 1341 return queue_work_on(WORK_CPU_UNBOUND, wq, work);
1092
1093 if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work))) {
1094 __queue_work(cpu, wq, work);
1095 ret = 1;
1096 }
1097 return ret;
1098} 1342}
1099EXPORT_SYMBOL_GPL(queue_work_on); 1343EXPORT_SYMBOL_GPL(queue_work);
1100 1344
1101static void delayed_work_timer_fn(unsigned long __data) 1345void delayed_work_timer_fn(unsigned long __data)
1102{ 1346{
1103 struct delayed_work *dwork = (struct delayed_work *)__data; 1347 struct delayed_work *dwork = (struct delayed_work *)__data;
1104 struct cpu_workqueue_struct *cwq = get_work_cwq(&dwork->work); 1348 struct cpu_workqueue_struct *cwq = get_work_cwq(&dwork->work);
1105 1349
1106 __queue_work(smp_processor_id(), cwq->wq, &dwork->work); 1350 /* should have been called from irqsafe timer with irq already off */
1351 __queue_work(dwork->cpu, cwq->wq, &dwork->work);
1107} 1352}
1353EXPORT_SYMBOL_GPL(delayed_work_timer_fn);
1108 1354
1109/** 1355static void __queue_delayed_work(int cpu, struct workqueue_struct *wq,
1110 * queue_delayed_work - queue work on a workqueue after delay 1356 struct delayed_work *dwork, unsigned long delay)
1111 * @wq: workqueue to use
1112 * @dwork: delayable work to queue
1113 * @delay: number of jiffies to wait before queueing
1114 *
1115 * Returns 0 if @work was already on a queue, non-zero otherwise.
1116 */
1117int queue_delayed_work(struct workqueue_struct *wq,
1118 struct delayed_work *dwork, unsigned long delay)
1119{ 1357{
1120 if (delay == 0) 1358 struct timer_list *timer = &dwork->timer;
1121 return queue_work(wq, &dwork->work); 1359 struct work_struct *work = &dwork->work;
1360 unsigned int lcpu;
1122 1361
1123 return queue_delayed_work_on(-1, wq, dwork, delay); 1362 WARN_ON_ONCE(timer->function != delayed_work_timer_fn ||
1363 timer->data != (unsigned long)dwork);
1364 BUG_ON(timer_pending(timer));
1365 BUG_ON(!list_empty(&work->entry));
1366
1367 timer_stats_timer_set_start_info(&dwork->timer);
1368
1369 /*
1370 * This stores cwq for the moment, for the timer_fn. Note that the
1371 * work's gcwq is preserved to allow reentrance detection for
1372 * delayed works.
1373 */
1374 if (!(wq->flags & WQ_UNBOUND)) {
1375 struct global_cwq *gcwq = get_work_gcwq(work);
1376
1377 /*
1378 * If we cannot get the last gcwq from @work directly,
1379 * select the last CPU such that it avoids unnecessarily
1380 * triggering non-reentrancy check in __queue_work().
1381 */
1382 lcpu = cpu;
1383 if (gcwq)
1384 lcpu = gcwq->cpu;
1385 if (lcpu == WORK_CPU_UNBOUND)
1386 lcpu = raw_smp_processor_id();
1387 } else {
1388 lcpu = WORK_CPU_UNBOUND;
1389 }
1390
1391 set_work_cwq(work, get_cwq(lcpu, wq), 0);
1392
1393 dwork->cpu = cpu;
1394 timer->expires = jiffies + delay;
1395
1396 if (unlikely(cpu != WORK_CPU_UNBOUND))
1397 add_timer_on(timer, cpu);
1398 else
1399 add_timer(timer);
1124} 1400}
1125EXPORT_SYMBOL_GPL(queue_delayed_work);
1126 1401
1127/** 1402/**
1128 * queue_delayed_work_on - queue work on specific CPU after delay 1403 * queue_delayed_work_on - queue work on specific CPU after delay
@@ -1131,53 +1406,100 @@ EXPORT_SYMBOL_GPL(queue_delayed_work);
1131 * @dwork: work to queue 1406 * @dwork: work to queue
1132 * @delay: number of jiffies to wait before queueing 1407 * @delay: number of jiffies to wait before queueing
1133 * 1408 *
1134 * Returns 0 if @work was already on a queue, non-zero otherwise. 1409 * Returns %false if @work was already on a queue, %true otherwise. If
1410 * @delay is zero and @dwork is idle, it will be scheduled for immediate
1411 * execution.
1135 */ 1412 */
1136int queue_delayed_work_on(int cpu, struct workqueue_struct *wq, 1413bool queue_delayed_work_on(int cpu, struct workqueue_struct *wq,
1137 struct delayed_work *dwork, unsigned long delay) 1414 struct delayed_work *dwork, unsigned long delay)
1138{ 1415{
1139 int ret = 0;
1140 struct timer_list *timer = &dwork->timer;
1141 struct work_struct *work = &dwork->work; 1416 struct work_struct *work = &dwork->work;
1417 bool ret = false;
1418 unsigned long flags;
1142 1419
1143 if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work))) { 1420 if (!delay)
1144 unsigned int lcpu; 1421 return queue_work_on(cpu, wq, &dwork->work);
1145 1422
1146 BUG_ON(timer_pending(timer)); 1423 /* read the comment in __queue_work() */
1147 BUG_ON(!list_empty(&work->entry)); 1424 local_irq_save(flags);
1148 1425
1149 timer_stats_timer_set_start_info(&dwork->timer); 1426 if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work))) {
1427 __queue_delayed_work(cpu, wq, dwork, delay);
1428 ret = true;
1429 }
1150 1430
1151 /* 1431 local_irq_restore(flags);
1152 * This stores cwq for the moment, for the timer_fn. 1432 return ret;
1153 * Note that the work's gcwq is preserved to allow 1433}
1154 * reentrance detection for delayed works. 1434EXPORT_SYMBOL_GPL(queue_delayed_work_on);
1155 */
1156 if (!(wq->flags & WQ_UNBOUND)) {
1157 struct global_cwq *gcwq = get_work_gcwq(work);
1158 1435
1159 if (gcwq && gcwq->cpu != WORK_CPU_UNBOUND) 1436/**
1160 lcpu = gcwq->cpu; 1437 * queue_delayed_work - queue work on a workqueue after delay
1161 else 1438 * @wq: workqueue to use
1162 lcpu = raw_smp_processor_id(); 1439 * @dwork: delayable work to queue
1163 } else 1440 * @delay: number of jiffies to wait before queueing
1164 lcpu = WORK_CPU_UNBOUND; 1441 *
1442 * Equivalent to queue_delayed_work_on() but tries to use the local CPU.
1443 */
1444bool queue_delayed_work(struct workqueue_struct *wq,
1445 struct delayed_work *dwork, unsigned long delay)
1446{
1447 return queue_delayed_work_on(WORK_CPU_UNBOUND, wq, dwork, delay);
1448}
1449EXPORT_SYMBOL_GPL(queue_delayed_work);
1165 1450
1166 set_work_cwq(work, get_cwq(lcpu, wq), 0); 1451/**
1452 * mod_delayed_work_on - modify delay of or queue a delayed work on specific CPU
1453 * @cpu: CPU number to execute work on
1454 * @wq: workqueue to use
1455 * @dwork: work to queue
1456 * @delay: number of jiffies to wait before queueing
1457 *
1458 * If @dwork is idle, equivalent to queue_delayed_work_on(); otherwise,
1459 * modify @dwork's timer so that it expires after @delay. If @delay is
1460 * zero, @work is guaranteed to be scheduled immediately regardless of its
1461 * current state.
1462 *
1463 * Returns %false if @dwork was idle and queued, %true if @dwork was
1464 * pending and its timer was modified.
1465 *
1466 * This function is safe to call from any context including IRQ handler.
1467 * See try_to_grab_pending() for details.
1468 */
1469bool mod_delayed_work_on(int cpu, struct workqueue_struct *wq,
1470 struct delayed_work *dwork, unsigned long delay)
1471{
1472 unsigned long flags;
1473 int ret;
1167 1474
1168 timer->expires = jiffies + delay; 1475 do {
1169 timer->data = (unsigned long)dwork; 1476 ret = try_to_grab_pending(&dwork->work, true, &flags);
1170 timer->function = delayed_work_timer_fn; 1477 } while (unlikely(ret == -EAGAIN));
1171 1478
1172 if (unlikely(cpu >= 0)) 1479 if (likely(ret >= 0)) {
1173 add_timer_on(timer, cpu); 1480 __queue_delayed_work(cpu, wq, dwork, delay);
1174 else 1481 local_irq_restore(flags);
1175 add_timer(timer);
1176 ret = 1;
1177 } 1482 }
1483
1484 /* -ENOENT from try_to_grab_pending() becomes %true */
1178 return ret; 1485 return ret;
1179} 1486}
1180EXPORT_SYMBOL_GPL(queue_delayed_work_on); 1487EXPORT_SYMBOL_GPL(mod_delayed_work_on);
1488
1489/**
1490 * mod_delayed_work - modify delay of or queue a delayed work
1491 * @wq: workqueue to use
1492 * @dwork: work to queue
1493 * @delay: number of jiffies to wait before queueing
1494 *
1495 * mod_delayed_work_on() on local CPU.
1496 */
1497bool mod_delayed_work(struct workqueue_struct *wq, struct delayed_work *dwork,
1498 unsigned long delay)
1499{
1500 return mod_delayed_work_on(WORK_CPU_UNBOUND, wq, dwork, delay);
1501}
1502EXPORT_SYMBOL_GPL(mod_delayed_work);
1181 1503
1182/** 1504/**
1183 * worker_enter_idle - enter idle state 1505 * worker_enter_idle - enter idle state
@@ -1305,37 +1627,21 @@ __acquires(&gcwq->lock)
1305 } 1627 }
1306} 1628}
1307 1629
1308struct idle_rebind {
1309 int cnt; /* # workers to be rebound */
1310 struct completion done; /* all workers rebound */
1311};
1312
1313/* 1630/*
1314 * Rebind an idle @worker to its CPU. During CPU onlining, this has to 1631 * Rebind an idle @worker to its CPU. worker_thread() will test
1315 * happen synchronously for idle workers. worker_thread() will test 1632 * list_empty(@worker->entry) before leaving idle and call this function.
1316 * %WORKER_REBIND before leaving idle and call this function.
1317 */ 1633 */
1318static void idle_worker_rebind(struct worker *worker) 1634static void idle_worker_rebind(struct worker *worker)
1319{ 1635{
1320 struct global_cwq *gcwq = worker->pool->gcwq; 1636 struct global_cwq *gcwq = worker->pool->gcwq;
1321 1637
1322 /* CPU must be online at this point */ 1638 /* CPU may go down again inbetween, clear UNBOUND only on success */
1323 WARN_ON(!worker_maybe_bind_and_lock(worker)); 1639 if (worker_maybe_bind_and_lock(worker))
1324 if (!--worker->idle_rebind->cnt) 1640 worker_clr_flags(worker, WORKER_UNBOUND);
1325 complete(&worker->idle_rebind->done);
1326 spin_unlock_irq(&worker->pool->gcwq->lock);
1327 1641
1328 /* we did our part, wait for rebind_workers() to finish up */ 1642 /* rebind complete, become available again */
1329 wait_event(gcwq->rebind_hold, !(worker->flags & WORKER_REBIND)); 1643 list_add(&worker->entry, &worker->pool->idle_list);
1330 1644 spin_unlock_irq(&gcwq->lock);
1331 /*
1332 * rebind_workers() shouldn't finish until all workers passed the
1333 * above WORKER_REBIND wait. Tell it when done.
1334 */
1335 spin_lock_irq(&worker->pool->gcwq->lock);
1336 if (!--worker->idle_rebind->cnt)
1337 complete(&worker->idle_rebind->done);
1338 spin_unlock_irq(&worker->pool->gcwq->lock);
1339} 1645}
1340 1646
1341/* 1647/*
@@ -1349,16 +1655,8 @@ static void busy_worker_rebind_fn(struct work_struct *work)
1349 struct worker *worker = container_of(work, struct worker, rebind_work); 1655 struct worker *worker = container_of(work, struct worker, rebind_work);
1350 struct global_cwq *gcwq = worker->pool->gcwq; 1656 struct global_cwq *gcwq = worker->pool->gcwq;
1351 1657
1352 worker_maybe_bind_and_lock(worker); 1658 if (worker_maybe_bind_and_lock(worker))
1353 1659 worker_clr_flags(worker, WORKER_UNBOUND);
1354 /*
1355 * %WORKER_REBIND must be cleared even if the above binding failed;
1356 * otherwise, we may confuse the next CPU_UP cycle or oops / get
1357 * stuck by calling idle_worker_rebind() prematurely. If CPU went
1358 * down again inbetween, %WORKER_UNBOUND would be set, so clearing
1359 * %WORKER_REBIND is always safe.
1360 */
1361 worker_clr_flags(worker, WORKER_REBIND);
1362 1660
1363 spin_unlock_irq(&gcwq->lock); 1661 spin_unlock_irq(&gcwq->lock);
1364} 1662}
@@ -1370,123 +1668,74 @@ static void busy_worker_rebind_fn(struct work_struct *work)
1370 * @gcwq->cpu is coming online. Rebind all workers to the CPU. Rebinding 1668 * @gcwq->cpu is coming online. Rebind all workers to the CPU. Rebinding
1371 * is different for idle and busy ones. 1669 * is different for idle and busy ones.
1372 * 1670 *
1373 * The idle ones should be rebound synchronously and idle rebinding should 1671 * Idle ones will be removed from the idle_list and woken up. They will
1374 * be complete before any worker starts executing work items with 1672 * add themselves back after completing rebind. This ensures that the
1375 * concurrency management enabled; otherwise, scheduler may oops trying to 1673 * idle_list doesn't contain any unbound workers when re-bound busy workers
1376 * wake up non-local idle worker from wq_worker_sleeping(). 1674 * try to perform local wake-ups for concurrency management.
1377 *
1378 * This is achieved by repeatedly requesting rebinding until all idle
1379 * workers are known to have been rebound under @gcwq->lock and holding all
1380 * idle workers from becoming busy until idle rebinding is complete.
1381 * 1675 *
1382 * Once idle workers are rebound, busy workers can be rebound as they 1676 * Busy workers can rebind after they finish their current work items.
1383 * finish executing their current work items. Queueing the rebind work at 1677 * Queueing the rebind work item at the head of the scheduled list is
1384 * the head of their scheduled lists is enough. Note that nr_running will 1678 * enough. Note that nr_running will be properly bumped as busy workers
1385 * be properbly bumped as busy workers rebind. 1679 * rebind.
1386 * 1680 *
1387 * On return, all workers are guaranteed to either be bound or have rebind 1681 * On return, all non-manager workers are scheduled for rebind - see
1388 * work item scheduled. 1682 * manage_workers() for the manager special case. Any idle worker
1683 * including the manager will not appear on @idle_list until rebind is
1684 * complete, making local wake-ups safe.
1389 */ 1685 */
1390static void rebind_workers(struct global_cwq *gcwq) 1686static void rebind_workers(struct global_cwq *gcwq)
1391 __releases(&gcwq->lock) __acquires(&gcwq->lock)
1392{ 1687{
1393 struct idle_rebind idle_rebind;
1394 struct worker_pool *pool; 1688 struct worker_pool *pool;
1395 struct worker *worker; 1689 struct worker *worker, *n;
1396 struct hlist_node *pos; 1690 struct hlist_node *pos;
1397 int i; 1691 int i;
1398 1692
1399 lockdep_assert_held(&gcwq->lock); 1693 lockdep_assert_held(&gcwq->lock);
1400 1694
1401 for_each_worker_pool(pool, gcwq) 1695 for_each_worker_pool(pool, gcwq)
1402 lockdep_assert_held(&pool->manager_mutex); 1696 lockdep_assert_held(&pool->assoc_mutex);
1403
1404 /*
1405 * Rebind idle workers. Interlocked both ways. We wait for
1406 * workers to rebind via @idle_rebind.done. Workers will wait for
1407 * us to finish up by watching %WORKER_REBIND.
1408 */
1409 init_completion(&idle_rebind.done);
1410retry:
1411 idle_rebind.cnt = 1;
1412 INIT_COMPLETION(idle_rebind.done);
1413 1697
1414 /* set REBIND and kick idle ones, we'll wait for these later */ 1698 /* dequeue and kick idle ones */
1415 for_each_worker_pool(pool, gcwq) { 1699 for_each_worker_pool(pool, gcwq) {
1416 list_for_each_entry(worker, &pool->idle_list, entry) { 1700 list_for_each_entry_safe(worker, n, &pool->idle_list, entry) {
1417 unsigned long worker_flags = worker->flags; 1701 /*
1418 1702 * idle workers should be off @pool->idle_list
1419 if (worker->flags & WORKER_REBIND) 1703 * until rebind is complete to avoid receiving
1420 continue; 1704 * premature local wake-ups.
1421 1705 */
1422 /* morph UNBOUND to REBIND atomically */ 1706 list_del_init(&worker->entry);
1423 worker_flags &= ~WORKER_UNBOUND;
1424 worker_flags |= WORKER_REBIND;
1425 ACCESS_ONCE(worker->flags) = worker_flags;
1426
1427 idle_rebind.cnt++;
1428 worker->idle_rebind = &idle_rebind;
1429 1707
1430 /* worker_thread() will call idle_worker_rebind() */ 1708 /*
1709 * worker_thread() will see the above dequeuing
1710 * and call idle_worker_rebind().
1711 */
1431 wake_up_process(worker->task); 1712 wake_up_process(worker->task);
1432 } 1713 }
1433 } 1714 }
1434 1715
1435 if (--idle_rebind.cnt) { 1716 /* rebind busy workers */
1436 spin_unlock_irq(&gcwq->lock);
1437 wait_for_completion(&idle_rebind.done);
1438 spin_lock_irq(&gcwq->lock);
1439 /* busy ones might have become idle while waiting, retry */
1440 goto retry;
1441 }
1442
1443 /* all idle workers are rebound, rebind busy workers */
1444 for_each_busy_worker(worker, i, pos, gcwq) { 1717 for_each_busy_worker(worker, i, pos, gcwq) {
1445 struct work_struct *rebind_work = &worker->rebind_work; 1718 struct work_struct *rebind_work = &worker->rebind_work;
1446 unsigned long worker_flags = worker->flags; 1719 struct workqueue_struct *wq;
1447
1448 /* morph UNBOUND to REBIND atomically */
1449 worker_flags &= ~WORKER_UNBOUND;
1450 worker_flags |= WORKER_REBIND;
1451 ACCESS_ONCE(worker->flags) = worker_flags;
1452 1720
1453 if (test_and_set_bit(WORK_STRUCT_PENDING_BIT, 1721 if (test_and_set_bit(WORK_STRUCT_PENDING_BIT,
1454 work_data_bits(rebind_work))) 1722 work_data_bits(rebind_work)))
1455 continue; 1723 continue;
1456 1724
1457 /* wq doesn't matter, use the default one */
1458 debug_work_activate(rebind_work); 1725 debug_work_activate(rebind_work);
1459 insert_work(get_cwq(gcwq->cpu, system_wq), rebind_work,
1460 worker->scheduled.next,
1461 work_color_to_flags(WORK_NO_COLOR));
1462 }
1463
1464 /*
1465 * All idle workers are rebound and waiting for %WORKER_REBIND to
1466 * be cleared inside idle_worker_rebind(). Clear and release.
1467 * Clearing %WORKER_REBIND from this foreign context is safe
1468 * because these workers are still guaranteed to be idle.
1469 *
1470 * We need to make sure all idle workers passed WORKER_REBIND wait
1471 * in idle_worker_rebind() before returning; otherwise, workers can
1472 * get stuck at the wait if hotplug cycle repeats.
1473 */
1474 idle_rebind.cnt = 1;
1475 INIT_COMPLETION(idle_rebind.done);
1476
1477 for_each_worker_pool(pool, gcwq) {
1478 list_for_each_entry(worker, &pool->idle_list, entry) {
1479 worker->flags &= ~WORKER_REBIND;
1480 idle_rebind.cnt++;
1481 }
1482 }
1483 1726
1484 wake_up_all(&gcwq->rebind_hold); 1727 /*
1728 * wq doesn't really matter but let's keep @worker->pool
1729 * and @cwq->pool consistent for sanity.
1730 */
1731 if (worker_pool_pri(worker->pool))
1732 wq = system_highpri_wq;
1733 else
1734 wq = system_wq;
1485 1735
1486 if (--idle_rebind.cnt) { 1736 insert_work(get_cwq(gcwq->cpu, wq), rebind_work,
1487 spin_unlock_irq(&gcwq->lock); 1737 worker->scheduled.next,
1488 wait_for_completion(&idle_rebind.done); 1738 work_color_to_flags(WORK_NO_COLOR));
1489 spin_lock_irq(&gcwq->lock);
1490 } 1739 }
1491} 1740}
1492 1741
@@ -1844,22 +2093,22 @@ static bool manage_workers(struct worker *worker)
1844 * grab %POOL_MANAGING_WORKERS to achieve this because that can 2093 * grab %POOL_MANAGING_WORKERS to achieve this because that can
1845 * lead to idle worker depletion (all become busy thinking someone 2094 * lead to idle worker depletion (all become busy thinking someone
1846 * else is managing) which in turn can result in deadlock under 2095 * else is managing) which in turn can result in deadlock under
1847 * extreme circumstances. Use @pool->manager_mutex to synchronize 2096 * extreme circumstances. Use @pool->assoc_mutex to synchronize
1848 * manager against CPU hotplug. 2097 * manager against CPU hotplug.
1849 * 2098 *
1850 * manager_mutex would always be free unless CPU hotplug is in 2099 * assoc_mutex would always be free unless CPU hotplug is in
1851 * progress. trylock first without dropping @gcwq->lock. 2100 * progress. trylock first without dropping @gcwq->lock.
1852 */ 2101 */
1853 if (unlikely(!mutex_trylock(&pool->manager_mutex))) { 2102 if (unlikely(!mutex_trylock(&pool->assoc_mutex))) {
1854 spin_unlock_irq(&pool->gcwq->lock); 2103 spin_unlock_irq(&pool->gcwq->lock);
1855 mutex_lock(&pool->manager_mutex); 2104 mutex_lock(&pool->assoc_mutex);
1856 /* 2105 /*
1857 * CPU hotplug could have happened while we were waiting 2106 * CPU hotplug could have happened while we were waiting
1858 * for manager_mutex. Hotplug itself can't handle us 2107 * for assoc_mutex. Hotplug itself can't handle us
1859 * because manager isn't either on idle or busy list, and 2108 * because manager isn't either on idle or busy list, and
1860 * @gcwq's state and ours could have deviated. 2109 * @gcwq's state and ours could have deviated.
1861 * 2110 *
1862 * As hotplug is now excluded via manager_mutex, we can 2111 * As hotplug is now excluded via assoc_mutex, we can
1863 * simply try to bind. It will succeed or fail depending 2112 * simply try to bind. It will succeed or fail depending
1864 * on @gcwq's current state. Try it and adjust 2113 * on @gcwq's current state. Try it and adjust
1865 * %WORKER_UNBOUND accordingly. 2114 * %WORKER_UNBOUND accordingly.
@@ -1882,112 +2131,11 @@ static bool manage_workers(struct worker *worker)
1882 ret |= maybe_create_worker(pool); 2131 ret |= maybe_create_worker(pool);
1883 2132
1884 pool->flags &= ~POOL_MANAGING_WORKERS; 2133 pool->flags &= ~POOL_MANAGING_WORKERS;
1885 mutex_unlock(&pool->manager_mutex); 2134 mutex_unlock(&pool->assoc_mutex);
1886 return ret; 2135 return ret;
1887} 2136}
1888 2137
1889/** 2138/**
1890 * move_linked_works - move linked works to a list
1891 * @work: start of series of works to be scheduled
1892 * @head: target list to append @work to
1893 * @nextp: out paramter for nested worklist walking
1894 *
1895 * Schedule linked works starting from @work to @head. Work series to
1896 * be scheduled starts at @work and includes any consecutive work with
1897 * WORK_STRUCT_LINKED set in its predecessor.
1898 *
1899 * If @nextp is not NULL, it's updated to point to the next work of
1900 * the last scheduled work. This allows move_linked_works() to be
1901 * nested inside outer list_for_each_entry_safe().
1902 *
1903 * CONTEXT:
1904 * spin_lock_irq(gcwq->lock).
1905 */
1906static void move_linked_works(struct work_struct *work, struct list_head *head,
1907 struct work_struct **nextp)
1908{
1909 struct work_struct *n;
1910
1911 /*
1912 * Linked worklist will always end before the end of the list,
1913 * use NULL for list head.
1914 */
1915 list_for_each_entry_safe_from(work, n, NULL, entry) {
1916 list_move_tail(&work->entry, head);
1917 if (!(*work_data_bits(work) & WORK_STRUCT_LINKED))
1918 break;
1919 }
1920
1921 /*
1922 * If we're already inside safe list traversal and have moved
1923 * multiple works to the scheduled queue, the next position
1924 * needs to be updated.
1925 */
1926 if (nextp)
1927 *nextp = n;
1928}
1929
1930static void cwq_activate_first_delayed(struct cpu_workqueue_struct *cwq)
1931{
1932 struct work_struct *work = list_first_entry(&cwq->delayed_works,
1933 struct work_struct, entry);
1934
1935 trace_workqueue_activate_work(work);
1936 move_linked_works(work, &cwq->pool->worklist, NULL);
1937 __clear_bit(WORK_STRUCT_DELAYED_BIT, work_data_bits(work));
1938 cwq->nr_active++;
1939}
1940
1941/**
1942 * cwq_dec_nr_in_flight - decrement cwq's nr_in_flight
1943 * @cwq: cwq of interest
1944 * @color: color of work which left the queue
1945 * @delayed: for a delayed work
1946 *
1947 * A work either has completed or is removed from pending queue,
1948 * decrement nr_in_flight of its cwq and handle workqueue flushing.
1949 *
1950 * CONTEXT:
1951 * spin_lock_irq(gcwq->lock).
1952 */
1953static void cwq_dec_nr_in_flight(struct cpu_workqueue_struct *cwq, int color,
1954 bool delayed)
1955{
1956 /* ignore uncolored works */
1957 if (color == WORK_NO_COLOR)
1958 return;
1959
1960 cwq->nr_in_flight[color]--;
1961
1962 if (!delayed) {
1963 cwq->nr_active--;
1964 if (!list_empty(&cwq->delayed_works)) {
1965 /* one down, submit a delayed one */
1966 if (cwq->nr_active < cwq->max_active)
1967 cwq_activate_first_delayed(cwq);
1968 }
1969 }
1970
1971 /* is flush in progress and are we at the flushing tip? */
1972 if (likely(cwq->flush_color != color))
1973 return;
1974
1975 /* are there still in-flight works? */
1976 if (cwq->nr_in_flight[color])
1977 return;
1978
1979 /* this cwq is done, clear flush_color */
1980 cwq->flush_color = -1;
1981
1982 /*
1983 * If this was the last cwq, wake up the first flusher. It
1984 * will handle the rest.
1985 */
1986 if (atomic_dec_and_test(&cwq->wq->nr_cwqs_to_flush))
1987 complete(&cwq->wq->first_flusher->done);
1988}
1989
1990/**
1991 * process_one_work - process single work 2139 * process_one_work - process single work
1992 * @worker: self 2140 * @worker: self
1993 * @work: work to process 2141 * @work: work to process
@@ -2030,7 +2178,7 @@ __acquires(&gcwq->lock)
2030 * necessary to avoid spurious warnings from rescuers servicing the 2178 * necessary to avoid spurious warnings from rescuers servicing the
2031 * unbound or a disassociated gcwq. 2179 * unbound or a disassociated gcwq.
2032 */ 2180 */
2033 WARN_ON_ONCE(!(worker->flags & (WORKER_UNBOUND | WORKER_REBIND)) && 2181 WARN_ON_ONCE(!(worker->flags & WORKER_UNBOUND) &&
2034 !(gcwq->flags & GCWQ_DISASSOCIATED) && 2182 !(gcwq->flags & GCWQ_DISASSOCIATED) &&
2035 raw_smp_processor_id() != gcwq->cpu); 2183 raw_smp_processor_id() != gcwq->cpu);
2036 2184
@@ -2046,15 +2194,13 @@ __acquires(&gcwq->lock)
2046 return; 2194 return;
2047 } 2195 }
2048 2196
2049 /* claim and process */ 2197 /* claim and dequeue */
2050 debug_work_deactivate(work); 2198 debug_work_deactivate(work);
2051 hlist_add_head(&worker->hentry, bwh); 2199 hlist_add_head(&worker->hentry, bwh);
2052 worker->current_work = work; 2200 worker->current_work = work;
2053 worker->current_cwq = cwq; 2201 worker->current_cwq = cwq;
2054 work_color = get_work_color(work); 2202 work_color = get_work_color(work);
2055 2203
2056 /* record the current cpu number in the work data and dequeue */
2057 set_work_cpu(work, gcwq->cpu);
2058 list_del_init(&work->entry); 2204 list_del_init(&work->entry);
2059 2205
2060 /* 2206 /*
@@ -2071,9 +2217,16 @@ __acquires(&gcwq->lock)
2071 if ((worker->flags & WORKER_UNBOUND) && need_more_worker(pool)) 2217 if ((worker->flags & WORKER_UNBOUND) && need_more_worker(pool))
2072 wake_up_worker(pool); 2218 wake_up_worker(pool);
2073 2219
2220 /*
2221 * Record the last CPU and clear PENDING which should be the last
2222 * update to @work. Also, do this inside @gcwq->lock so that
2223 * PENDING and queued state changes happen together while IRQ is
2224 * disabled.
2225 */
2226 set_work_cpu_and_clear_pending(work, gcwq->cpu);
2227
2074 spin_unlock_irq(&gcwq->lock); 2228 spin_unlock_irq(&gcwq->lock);
2075 2229
2076 work_clear_pending(work);
2077 lock_map_acquire_read(&cwq->wq->lockdep_map); 2230 lock_map_acquire_read(&cwq->wq->lockdep_map);
2078 lock_map_acquire(&lockdep_map); 2231 lock_map_acquire(&lockdep_map);
2079 trace_workqueue_execute_start(work); 2232 trace_workqueue_execute_start(work);
@@ -2087,11 +2240,9 @@ __acquires(&gcwq->lock)
2087 lock_map_release(&cwq->wq->lockdep_map); 2240 lock_map_release(&cwq->wq->lockdep_map);
2088 2241
2089 if (unlikely(in_atomic() || lockdep_depth(current) > 0)) { 2242 if (unlikely(in_atomic() || lockdep_depth(current) > 0)) {
2090 printk(KERN_ERR "BUG: workqueue leaked lock or atomic: " 2243 pr_err("BUG: workqueue leaked lock or atomic: %s/0x%08x/%d\n"
2091 "%s/0x%08x/%d\n", 2244 " last function: %pf\n",
2092 current->comm, preempt_count(), task_pid_nr(current)); 2245 current->comm, preempt_count(), task_pid_nr(current), f);
2093 printk(KERN_ERR " last function: ");
2094 print_symbol("%s\n", (unsigned long)f);
2095 debug_show_held_locks(current); 2246 debug_show_held_locks(current);
2096 dump_stack(); 2247 dump_stack();
2097 } 2248 }
@@ -2106,7 +2257,7 @@ __acquires(&gcwq->lock)
2106 hlist_del_init(&worker->hentry); 2257 hlist_del_init(&worker->hentry);
2107 worker->current_work = NULL; 2258 worker->current_work = NULL;
2108 worker->current_cwq = NULL; 2259 worker->current_cwq = NULL;
2109 cwq_dec_nr_in_flight(cwq, work_color, false); 2260 cwq_dec_nr_in_flight(cwq, work_color);
2110} 2261}
2111 2262
2112/** 2263/**
@@ -2151,18 +2302,17 @@ static int worker_thread(void *__worker)
2151woke_up: 2302woke_up:
2152 spin_lock_irq(&gcwq->lock); 2303 spin_lock_irq(&gcwq->lock);
2153 2304
2154 /* 2305 /* we are off idle list if destruction or rebind is requested */
2155 * DIE can be set only while idle and REBIND set while busy has 2306 if (unlikely(list_empty(&worker->entry))) {
2156 * @worker->rebind_work scheduled. Checking here is enough.
2157 */
2158 if (unlikely(worker->flags & (WORKER_REBIND | WORKER_DIE))) {
2159 spin_unlock_irq(&gcwq->lock); 2307 spin_unlock_irq(&gcwq->lock);
2160 2308
2309 /* if DIE is set, destruction is requested */
2161 if (worker->flags & WORKER_DIE) { 2310 if (worker->flags & WORKER_DIE) {
2162 worker->task->flags &= ~PF_WQ_WORKER; 2311 worker->task->flags &= ~PF_WQ_WORKER;
2163 return 0; 2312 return 0;
2164 } 2313 }
2165 2314
2315 /* otherwise, rebind */
2166 idle_worker_rebind(worker); 2316 idle_worker_rebind(worker);
2167 goto woke_up; 2317 goto woke_up;
2168 } 2318 }
@@ -2645,8 +2795,8 @@ reflush:
2645 2795
2646 if (++flush_cnt == 10 || 2796 if (++flush_cnt == 10 ||
2647 (flush_cnt % 100 == 0 && flush_cnt <= 1000)) 2797 (flush_cnt % 100 == 0 && flush_cnt <= 1000))
2648 pr_warning("workqueue %s: flush on destruction isn't complete after %u tries\n", 2798 pr_warn("workqueue %s: flush on destruction isn't complete after %u tries\n",
2649 wq->name, flush_cnt); 2799 wq->name, flush_cnt);
2650 goto reflush; 2800 goto reflush;
2651 } 2801 }
2652 2802
@@ -2657,8 +2807,7 @@ reflush:
2657} 2807}
2658EXPORT_SYMBOL_GPL(drain_workqueue); 2808EXPORT_SYMBOL_GPL(drain_workqueue);
2659 2809
2660static bool start_flush_work(struct work_struct *work, struct wq_barrier *barr, 2810static bool start_flush_work(struct work_struct *work, struct wq_barrier *barr)
2661 bool wait_executing)
2662{ 2811{
2663 struct worker *worker = NULL; 2812 struct worker *worker = NULL;
2664 struct global_cwq *gcwq; 2813 struct global_cwq *gcwq;
@@ -2680,13 +2829,12 @@ static bool start_flush_work(struct work_struct *work, struct wq_barrier *barr,
2680 cwq = get_work_cwq(work); 2829 cwq = get_work_cwq(work);
2681 if (unlikely(!cwq || gcwq != cwq->pool->gcwq)) 2830 if (unlikely(!cwq || gcwq != cwq->pool->gcwq))
2682 goto already_gone; 2831 goto already_gone;
2683 } else if (wait_executing) { 2832 } else {
2684 worker = find_worker_executing_work(gcwq, work); 2833 worker = find_worker_executing_work(gcwq, work);
2685 if (!worker) 2834 if (!worker)
2686 goto already_gone; 2835 goto already_gone;
2687 cwq = worker->current_cwq; 2836 cwq = worker->current_cwq;
2688 } else 2837 }
2689 goto already_gone;
2690 2838
2691 insert_wq_barrier(cwq, barr, work, worker); 2839 insert_wq_barrier(cwq, barr, work, worker);
2692 spin_unlock_irq(&gcwq->lock); 2840 spin_unlock_irq(&gcwq->lock);
@@ -2713,15 +2861,8 @@ already_gone:
2713 * flush_work - wait for a work to finish executing the last queueing instance 2861 * flush_work - wait for a work to finish executing the last queueing instance
2714 * @work: the work to flush 2862 * @work: the work to flush
2715 * 2863 *
2716 * Wait until @work has finished execution. This function considers 2864 * Wait until @work has finished execution. @work is guaranteed to be idle
2717 * only the last queueing instance of @work. If @work has been 2865 * on return if it hasn't been requeued since flush started.
2718 * enqueued across different CPUs on a non-reentrant workqueue or on
2719 * multiple workqueues, @work might still be executing on return on
2720 * some of the CPUs from earlier queueing.
2721 *
2722 * If @work was queued only on a non-reentrant, ordered or unbound
2723 * workqueue, @work is guaranteed to be idle on return if it hasn't
2724 * been requeued since flush started.
2725 * 2866 *
2726 * RETURNS: 2867 * RETURNS:
2727 * %true if flush_work() waited for the work to finish execution, 2868 * %true if flush_work() waited for the work to finish execution,
@@ -2734,140 +2875,36 @@ bool flush_work(struct work_struct *work)
2734 lock_map_acquire(&work->lockdep_map); 2875 lock_map_acquire(&work->lockdep_map);
2735 lock_map_release(&work->lockdep_map); 2876 lock_map_release(&work->lockdep_map);
2736 2877
2737 if (start_flush_work(work, &barr, true)) { 2878 if (start_flush_work(work, &barr)) {
2738 wait_for_completion(&barr.done);
2739 destroy_work_on_stack(&barr.work);
2740 return true;
2741 } else
2742 return false;
2743}
2744EXPORT_SYMBOL_GPL(flush_work);
2745
2746static bool wait_on_cpu_work(struct global_cwq *gcwq, struct work_struct *work)
2747{
2748 struct wq_barrier barr;
2749 struct worker *worker;
2750
2751 spin_lock_irq(&gcwq->lock);
2752
2753 worker = find_worker_executing_work(gcwq, work);
2754 if (unlikely(worker))
2755 insert_wq_barrier(worker->current_cwq, &barr, work, worker);
2756
2757 spin_unlock_irq(&gcwq->lock);
2758
2759 if (unlikely(worker)) {
2760 wait_for_completion(&barr.done); 2879 wait_for_completion(&barr.done);
2761 destroy_work_on_stack(&barr.work); 2880 destroy_work_on_stack(&barr.work);
2762 return true; 2881 return true;
2763 } else 2882 } else {
2764 return false; 2883 return false;
2765}
2766
2767static bool wait_on_work(struct work_struct *work)
2768{
2769 bool ret = false;
2770 int cpu;
2771
2772 might_sleep();
2773
2774 lock_map_acquire(&work->lockdep_map);
2775 lock_map_release(&work->lockdep_map);
2776
2777 for_each_gcwq_cpu(cpu)
2778 ret |= wait_on_cpu_work(get_gcwq(cpu), work);
2779 return ret;
2780}
2781
2782/**
2783 * flush_work_sync - wait until a work has finished execution
2784 * @work: the work to flush
2785 *
2786 * Wait until @work has finished execution. On return, it's
2787 * guaranteed that all queueing instances of @work which happened
2788 * before this function is called are finished. In other words, if
2789 * @work hasn't been requeued since this function was called, @work is
2790 * guaranteed to be idle on return.
2791 *
2792 * RETURNS:
2793 * %true if flush_work_sync() waited for the work to finish execution,
2794 * %false if it was already idle.
2795 */
2796bool flush_work_sync(struct work_struct *work)
2797{
2798 struct wq_barrier barr;
2799 bool pending, waited;
2800
2801 /* we'll wait for executions separately, queue barr only if pending */
2802 pending = start_flush_work(work, &barr, false);
2803
2804 /* wait for executions to finish */
2805 waited = wait_on_work(work);
2806
2807 /* wait for the pending one */
2808 if (pending) {
2809 wait_for_completion(&barr.done);
2810 destroy_work_on_stack(&barr.work);
2811 }
2812
2813 return pending || waited;
2814}
2815EXPORT_SYMBOL_GPL(flush_work_sync);
2816
2817/*
2818 * Upon a successful return (>= 0), the caller "owns" WORK_STRUCT_PENDING bit,
2819 * so this work can't be re-armed in any way.
2820 */
2821static int try_to_grab_pending(struct work_struct *work)
2822{
2823 struct global_cwq *gcwq;
2824 int ret = -1;
2825
2826 if (!test_and_set_bit(WORK_STRUCT_PENDING_BIT, work_data_bits(work)))
2827 return 0;
2828
2829 /*
2830 * The queueing is in progress, or it is already queued. Try to
2831 * steal it from ->worklist without clearing WORK_STRUCT_PENDING.
2832 */
2833 gcwq = get_work_gcwq(work);
2834 if (!gcwq)
2835 return ret;
2836
2837 spin_lock_irq(&gcwq->lock);
2838 if (!list_empty(&work->entry)) {
2839 /*
2840 * This work is queued, but perhaps we locked the wrong gcwq.
2841 * In that case we must see the new value after rmb(), see
2842 * insert_work()->wmb().
2843 */
2844 smp_rmb();
2845 if (gcwq == get_work_gcwq(work)) {
2846 debug_work_deactivate(work);
2847 list_del_init(&work->entry);
2848 cwq_dec_nr_in_flight(get_work_cwq(work),
2849 get_work_color(work),
2850 *work_data_bits(work) & WORK_STRUCT_DELAYED);
2851 ret = 1;
2852 }
2853 } 2884 }
2854 spin_unlock_irq(&gcwq->lock);
2855
2856 return ret;
2857} 2885}
2886EXPORT_SYMBOL_GPL(flush_work);
2858 2887
2859static bool __cancel_work_timer(struct work_struct *work, 2888static bool __cancel_work_timer(struct work_struct *work, bool is_dwork)
2860 struct timer_list* timer)
2861{ 2889{
2890 unsigned long flags;
2862 int ret; 2891 int ret;
2863 2892
2864 do { 2893 do {
2865 ret = (timer && likely(del_timer(timer))); 2894 ret = try_to_grab_pending(work, is_dwork, &flags);
2866 if (!ret) 2895 /*
2867 ret = try_to_grab_pending(work); 2896 * If someone else is canceling, wait for the same event it
2868 wait_on_work(work); 2897 * would be waiting for before retrying.
2898 */
2899 if (unlikely(ret == -ENOENT))
2900 flush_work(work);
2869 } while (unlikely(ret < 0)); 2901 } while (unlikely(ret < 0));
2870 2902
2903 /* tell other tasks trying to grab @work to back off */
2904 mark_work_canceling(work);
2905 local_irq_restore(flags);
2906
2907 flush_work(work);
2871 clear_work_data(work); 2908 clear_work_data(work);
2872 return ret; 2909 return ret;
2873} 2910}
@@ -2892,7 +2929,7 @@ static bool __cancel_work_timer(struct work_struct *work,
2892 */ 2929 */
2893bool cancel_work_sync(struct work_struct *work) 2930bool cancel_work_sync(struct work_struct *work)
2894{ 2931{
2895 return __cancel_work_timer(work, NULL); 2932 return __cancel_work_timer(work, false);
2896} 2933}
2897EXPORT_SYMBOL_GPL(cancel_work_sync); 2934EXPORT_SYMBOL_GPL(cancel_work_sync);
2898 2935
@@ -2910,33 +2947,44 @@ EXPORT_SYMBOL_GPL(cancel_work_sync);
2910 */ 2947 */
2911bool flush_delayed_work(struct delayed_work *dwork) 2948bool flush_delayed_work(struct delayed_work *dwork)
2912{ 2949{
2950 local_irq_disable();
2913 if (del_timer_sync(&dwork->timer)) 2951 if (del_timer_sync(&dwork->timer))
2914 __queue_work(raw_smp_processor_id(), 2952 __queue_work(dwork->cpu,
2915 get_work_cwq(&dwork->work)->wq, &dwork->work); 2953 get_work_cwq(&dwork->work)->wq, &dwork->work);
2954 local_irq_enable();
2916 return flush_work(&dwork->work); 2955 return flush_work(&dwork->work);
2917} 2956}
2918EXPORT_SYMBOL(flush_delayed_work); 2957EXPORT_SYMBOL(flush_delayed_work);
2919 2958
2920/** 2959/**
2921 * flush_delayed_work_sync - wait for a dwork to finish 2960 * cancel_delayed_work - cancel a delayed work
2922 * @dwork: the delayed work to flush 2961 * @dwork: delayed_work to cancel
2923 * 2962 *
2924 * Delayed timer is cancelled and the pending work is queued for 2963 * Kill off a pending delayed_work. Returns %true if @dwork was pending
2925 * execution immediately. Other than timer handling, its behavior 2964 * and canceled; %false if wasn't pending. Note that the work callback
2926 * is identical to flush_work_sync(). 2965 * function may still be running on return, unless it returns %true and the
2966 * work doesn't re-arm itself. Explicitly flush or use
2967 * cancel_delayed_work_sync() to wait on it.
2927 * 2968 *
2928 * RETURNS: 2969 * This function is safe to call from any context including IRQ handler.
2929 * %true if flush_work_sync() waited for the work to finish execution,
2930 * %false if it was already idle.
2931 */ 2970 */
2932bool flush_delayed_work_sync(struct delayed_work *dwork) 2971bool cancel_delayed_work(struct delayed_work *dwork)
2933{ 2972{
2934 if (del_timer_sync(&dwork->timer)) 2973 unsigned long flags;
2935 __queue_work(raw_smp_processor_id(), 2974 int ret;
2936 get_work_cwq(&dwork->work)->wq, &dwork->work); 2975
2937 return flush_work_sync(&dwork->work); 2976 do {
2977 ret = try_to_grab_pending(&dwork->work, true, &flags);
2978 } while (unlikely(ret == -EAGAIN));
2979
2980 if (unlikely(ret < 0))
2981 return false;
2982
2983 set_work_cpu_and_clear_pending(&dwork->work, work_cpu(&dwork->work));
2984 local_irq_restore(flags);
2985 return true;
2938} 2986}
2939EXPORT_SYMBOL(flush_delayed_work_sync); 2987EXPORT_SYMBOL(cancel_delayed_work);
2940 2988
2941/** 2989/**
2942 * cancel_delayed_work_sync - cancel a delayed work and wait for it to finish 2990 * cancel_delayed_work_sync - cancel a delayed work and wait for it to finish
@@ -2949,54 +2997,39 @@ EXPORT_SYMBOL(flush_delayed_work_sync);
2949 */ 2997 */
2950bool cancel_delayed_work_sync(struct delayed_work *dwork) 2998bool cancel_delayed_work_sync(struct delayed_work *dwork)
2951{ 2999{
2952 return __cancel_work_timer(&dwork->work, &dwork->timer); 3000 return __cancel_work_timer(&dwork->work, true);
2953} 3001}
2954EXPORT_SYMBOL(cancel_delayed_work_sync); 3002EXPORT_SYMBOL(cancel_delayed_work_sync);
2955 3003
2956/** 3004/**
2957 * schedule_work - put work task in global workqueue
2958 * @work: job to be done
2959 *
2960 * Returns zero if @work was already on the kernel-global workqueue and
2961 * non-zero otherwise.
2962 *
2963 * This puts a job in the kernel-global workqueue if it was not already
2964 * queued and leaves it in the same position on the kernel-global
2965 * workqueue otherwise.
2966 */
2967int schedule_work(struct work_struct *work)
2968{
2969 return queue_work(system_wq, work);
2970}
2971EXPORT_SYMBOL(schedule_work);
2972
2973/*
2974 * schedule_work_on - put work task on a specific cpu 3005 * schedule_work_on - put work task on a specific cpu
2975 * @cpu: cpu to put the work task on 3006 * @cpu: cpu to put the work task on
2976 * @work: job to be done 3007 * @work: job to be done
2977 * 3008 *
2978 * This puts a job on a specific cpu 3009 * This puts a job on a specific cpu
2979 */ 3010 */
2980int schedule_work_on(int cpu, struct work_struct *work) 3011bool schedule_work_on(int cpu, struct work_struct *work)
2981{ 3012{
2982 return queue_work_on(cpu, system_wq, work); 3013 return queue_work_on(cpu, system_wq, work);
2983} 3014}
2984EXPORT_SYMBOL(schedule_work_on); 3015EXPORT_SYMBOL(schedule_work_on);
2985 3016
2986/** 3017/**
2987 * schedule_delayed_work - put work task in global workqueue after delay 3018 * schedule_work - put work task in global workqueue
2988 * @dwork: job to be done 3019 * @work: job to be done
2989 * @delay: number of jiffies to wait or 0 for immediate execution
2990 * 3020 *
2991 * After waiting for a given time this puts a job in the kernel-global 3021 * Returns %false if @work was already on the kernel-global workqueue and
2992 * workqueue. 3022 * %true otherwise.
3023 *
3024 * This puts a job in the kernel-global workqueue if it was not already
3025 * queued and leaves it in the same position on the kernel-global
3026 * workqueue otherwise.
2993 */ 3027 */
2994int schedule_delayed_work(struct delayed_work *dwork, 3028bool schedule_work(struct work_struct *work)
2995 unsigned long delay)
2996{ 3029{
2997 return queue_delayed_work(system_wq, dwork, delay); 3030 return queue_work(system_wq, work);
2998} 3031}
2999EXPORT_SYMBOL(schedule_delayed_work); 3032EXPORT_SYMBOL(schedule_work);
3000 3033
3001/** 3034/**
3002 * schedule_delayed_work_on - queue work in global workqueue on CPU after delay 3035 * schedule_delayed_work_on - queue work in global workqueue on CPU after delay
@@ -3007,14 +3040,28 @@ EXPORT_SYMBOL(schedule_delayed_work);
3007 * After waiting for a given time this puts a job in the kernel-global 3040 * After waiting for a given time this puts a job in the kernel-global
3008 * workqueue on the specified CPU. 3041 * workqueue on the specified CPU.
3009 */ 3042 */
3010int schedule_delayed_work_on(int cpu, 3043bool schedule_delayed_work_on(int cpu, struct delayed_work *dwork,
3011 struct delayed_work *dwork, unsigned long delay) 3044 unsigned long delay)
3012{ 3045{
3013 return queue_delayed_work_on(cpu, system_wq, dwork, delay); 3046 return queue_delayed_work_on(cpu, system_wq, dwork, delay);
3014} 3047}
3015EXPORT_SYMBOL(schedule_delayed_work_on); 3048EXPORT_SYMBOL(schedule_delayed_work_on);
3016 3049
3017/** 3050/**
3051 * schedule_delayed_work - put work task in global workqueue after delay
3052 * @dwork: job to be done
3053 * @delay: number of jiffies to wait or 0 for immediate execution
3054 *
3055 * After waiting for a given time this puts a job in the kernel-global
3056 * workqueue.
3057 */
3058bool schedule_delayed_work(struct delayed_work *dwork, unsigned long delay)
3059{
3060 return queue_delayed_work(system_wq, dwork, delay);
3061}
3062EXPORT_SYMBOL(schedule_delayed_work);
3063
3064/**
3018 * schedule_on_each_cpu - execute a function synchronously on each online CPU 3065 * schedule_on_each_cpu - execute a function synchronously on each online CPU
3019 * @func: the function to call 3066 * @func: the function to call
3020 * 3067 *
@@ -3161,9 +3208,8 @@ static int wq_clamp_max_active(int max_active, unsigned int flags,
3161 int lim = flags & WQ_UNBOUND ? WQ_UNBOUND_MAX_ACTIVE : WQ_MAX_ACTIVE; 3208 int lim = flags & WQ_UNBOUND ? WQ_UNBOUND_MAX_ACTIVE : WQ_MAX_ACTIVE;
3162 3209
3163 if (max_active < 1 || max_active > lim) 3210 if (max_active < 1 || max_active > lim)
3164 printk(KERN_WARNING "workqueue: max_active %d requested for %s " 3211 pr_warn("workqueue: max_active %d requested for %s is out of range, clamping between %d and %d\n",
3165 "is out of range, clamping between %d and %d\n", 3212 max_active, name, 1, lim);
3166 max_active, name, 1, lim);
3167 3213
3168 return clamp_val(max_active, 1, lim); 3214 return clamp_val(max_active, 1, lim);
3169} 3215}
@@ -3319,6 +3365,26 @@ void destroy_workqueue(struct workqueue_struct *wq)
3319EXPORT_SYMBOL_GPL(destroy_workqueue); 3365EXPORT_SYMBOL_GPL(destroy_workqueue);
3320 3366
3321/** 3367/**
3368 * cwq_set_max_active - adjust max_active of a cwq
3369 * @cwq: target cpu_workqueue_struct
3370 * @max_active: new max_active value.
3371 *
3372 * Set @cwq->max_active to @max_active and activate delayed works if
3373 * increased.
3374 *
3375 * CONTEXT:
3376 * spin_lock_irq(gcwq->lock).
3377 */
3378static void cwq_set_max_active(struct cpu_workqueue_struct *cwq, int max_active)
3379{
3380 cwq->max_active = max_active;
3381
3382 while (!list_empty(&cwq->delayed_works) &&
3383 cwq->nr_active < cwq->max_active)
3384 cwq_activate_first_delayed(cwq);
3385}
3386
3387/**
3322 * workqueue_set_max_active - adjust max_active of a workqueue 3388 * workqueue_set_max_active - adjust max_active of a workqueue
3323 * @wq: target workqueue 3389 * @wq: target workqueue
3324 * @max_active: new max_active value. 3390 * @max_active: new max_active value.
@@ -3345,7 +3411,7 @@ void workqueue_set_max_active(struct workqueue_struct *wq, int max_active)
3345 3411
3346 if (!(wq->flags & WQ_FREEZABLE) || 3412 if (!(wq->flags & WQ_FREEZABLE) ||
3347 !(gcwq->flags & GCWQ_FREEZING)) 3413 !(gcwq->flags & GCWQ_FREEZING))
3348 get_cwq(gcwq->cpu, wq)->max_active = max_active; 3414 cwq_set_max_active(get_cwq(gcwq->cpu, wq), max_active);
3349 3415
3350 spin_unlock_irq(&gcwq->lock); 3416 spin_unlock_irq(&gcwq->lock);
3351 } 3417 }
@@ -3440,23 +3506,23 @@ EXPORT_SYMBOL_GPL(work_busy);
3440 */ 3506 */
3441 3507
3442/* claim manager positions of all pools */ 3508/* claim manager positions of all pools */
3443static void gcwq_claim_management_and_lock(struct global_cwq *gcwq) 3509static void gcwq_claim_assoc_and_lock(struct global_cwq *gcwq)
3444{ 3510{
3445 struct worker_pool *pool; 3511 struct worker_pool *pool;
3446 3512
3447 for_each_worker_pool(pool, gcwq) 3513 for_each_worker_pool(pool, gcwq)
3448 mutex_lock_nested(&pool->manager_mutex, pool - gcwq->pools); 3514 mutex_lock_nested(&pool->assoc_mutex, pool - gcwq->pools);
3449 spin_lock_irq(&gcwq->lock); 3515 spin_lock_irq(&gcwq->lock);
3450} 3516}
3451 3517
3452/* release manager positions */ 3518/* release manager positions */
3453static void gcwq_release_management_and_unlock(struct global_cwq *gcwq) 3519static void gcwq_release_assoc_and_unlock(struct global_cwq *gcwq)
3454{ 3520{
3455 struct worker_pool *pool; 3521 struct worker_pool *pool;
3456 3522
3457 spin_unlock_irq(&gcwq->lock); 3523 spin_unlock_irq(&gcwq->lock);
3458 for_each_worker_pool(pool, gcwq) 3524 for_each_worker_pool(pool, gcwq)
3459 mutex_unlock(&pool->manager_mutex); 3525 mutex_unlock(&pool->assoc_mutex);
3460} 3526}
3461 3527
3462static void gcwq_unbind_fn(struct work_struct *work) 3528static void gcwq_unbind_fn(struct work_struct *work)
@@ -3469,7 +3535,7 @@ static void gcwq_unbind_fn(struct work_struct *work)
3469 3535
3470 BUG_ON(gcwq->cpu != smp_processor_id()); 3536 BUG_ON(gcwq->cpu != smp_processor_id());
3471 3537
3472 gcwq_claim_management_and_lock(gcwq); 3538 gcwq_claim_assoc_and_lock(gcwq);
3473 3539
3474 /* 3540 /*
3475 * We've claimed all manager positions. Make all workers unbound 3541 * We've claimed all manager positions. Make all workers unbound
@@ -3486,7 +3552,7 @@ static void gcwq_unbind_fn(struct work_struct *work)
3486 3552
3487 gcwq->flags |= GCWQ_DISASSOCIATED; 3553 gcwq->flags |= GCWQ_DISASSOCIATED;
3488 3554
3489 gcwq_release_management_and_unlock(gcwq); 3555 gcwq_release_assoc_and_unlock(gcwq);
3490 3556
3491 /* 3557 /*
3492 * Call schedule() so that we cross rq->lock and thus can guarantee 3558 * Call schedule() so that we cross rq->lock and thus can guarantee
@@ -3514,7 +3580,7 @@ static void gcwq_unbind_fn(struct work_struct *work)
3514 * Workqueues should be brought up before normal priority CPU notifiers. 3580 * Workqueues should be brought up before normal priority CPU notifiers.
3515 * This will be registered high priority CPU notifier. 3581 * This will be registered high priority CPU notifier.
3516 */ 3582 */
3517static int __devinit workqueue_cpu_up_callback(struct notifier_block *nfb, 3583static int __cpuinit workqueue_cpu_up_callback(struct notifier_block *nfb,
3518 unsigned long action, 3584 unsigned long action,
3519 void *hcpu) 3585 void *hcpu)
3520{ 3586{
@@ -3542,10 +3608,10 @@ static int __devinit workqueue_cpu_up_callback(struct notifier_block *nfb,
3542 3608
3543 case CPU_DOWN_FAILED: 3609 case CPU_DOWN_FAILED:
3544 case CPU_ONLINE: 3610 case CPU_ONLINE:
3545 gcwq_claim_management_and_lock(gcwq); 3611 gcwq_claim_assoc_and_lock(gcwq);
3546 gcwq->flags &= ~GCWQ_DISASSOCIATED; 3612 gcwq->flags &= ~GCWQ_DISASSOCIATED;
3547 rebind_workers(gcwq); 3613 rebind_workers(gcwq);
3548 gcwq_release_management_and_unlock(gcwq); 3614 gcwq_release_assoc_and_unlock(gcwq);
3549 break; 3615 break;
3550 } 3616 }
3551 return NOTIFY_OK; 3617 return NOTIFY_OK;
@@ -3555,7 +3621,7 @@ static int __devinit workqueue_cpu_up_callback(struct notifier_block *nfb,
3555 * Workqueues should be brought down after normal priority CPU notifiers. 3621 * Workqueues should be brought down after normal priority CPU notifiers.
3556 * This will be registered as low priority CPU notifier. 3622 * This will be registered as low priority CPU notifier.
3557 */ 3623 */
3558static int __devinit workqueue_cpu_down_callback(struct notifier_block *nfb, 3624static int __cpuinit workqueue_cpu_down_callback(struct notifier_block *nfb,
3559 unsigned long action, 3625 unsigned long action,
3560 void *hcpu) 3626 void *hcpu)
3561{ 3627{
@@ -3566,7 +3632,7 @@ static int __devinit workqueue_cpu_down_callback(struct notifier_block *nfb,
3566 case CPU_DOWN_PREPARE: 3632 case CPU_DOWN_PREPARE:
3567 /* unbinding should happen on the local CPU */ 3633 /* unbinding should happen on the local CPU */
3568 INIT_WORK_ONSTACK(&unbind_work, gcwq_unbind_fn); 3634 INIT_WORK_ONSTACK(&unbind_work, gcwq_unbind_fn);
3569 schedule_work_on(cpu, &unbind_work); 3635 queue_work_on(cpu, system_highpri_wq, &unbind_work);
3570 flush_work(&unbind_work); 3636 flush_work(&unbind_work);
3571 break; 3637 break;
3572 } 3638 }
@@ -3576,18 +3642,17 @@ static int __devinit workqueue_cpu_down_callback(struct notifier_block *nfb,
3576#ifdef CONFIG_SMP 3642#ifdef CONFIG_SMP
3577 3643
3578struct work_for_cpu { 3644struct work_for_cpu {
3579 struct completion completion; 3645 struct work_struct work;
3580 long (*fn)(void *); 3646 long (*fn)(void *);
3581 void *arg; 3647 void *arg;
3582 long ret; 3648 long ret;
3583}; 3649};
3584 3650
3585static int do_work_for_cpu(void *_wfc) 3651static void work_for_cpu_fn(struct work_struct *work)
3586{ 3652{
3587 struct work_for_cpu *wfc = _wfc; 3653 struct work_for_cpu *wfc = container_of(work, struct work_for_cpu, work);
3654
3588 wfc->ret = wfc->fn(wfc->arg); 3655 wfc->ret = wfc->fn(wfc->arg);
3589 complete(&wfc->completion);
3590 return 0;
3591} 3656}
3592 3657
3593/** 3658/**
@@ -3602,19 +3667,11 @@ static int do_work_for_cpu(void *_wfc)
3602 */ 3667 */
3603long work_on_cpu(unsigned int cpu, long (*fn)(void *), void *arg) 3668long work_on_cpu(unsigned int cpu, long (*fn)(void *), void *arg)
3604{ 3669{
3605 struct task_struct *sub_thread; 3670 struct work_for_cpu wfc = { .fn = fn, .arg = arg };
3606 struct work_for_cpu wfc = {
3607 .completion = COMPLETION_INITIALIZER_ONSTACK(wfc.completion),
3608 .fn = fn,
3609 .arg = arg,
3610 };
3611 3671
3612 sub_thread = kthread_create(do_work_for_cpu, &wfc, "work_for_cpu"); 3672 INIT_WORK_ONSTACK(&wfc.work, work_for_cpu_fn);
3613 if (IS_ERR(sub_thread)) 3673 schedule_work_on(cpu, &wfc.work);
3614 return PTR_ERR(sub_thread); 3674 flush_work(&wfc.work);
3615 kthread_bind(sub_thread, cpu);
3616 wake_up_process(sub_thread);
3617 wait_for_completion(&wfc.completion);
3618 return wfc.ret; 3675 return wfc.ret;
3619} 3676}
3620EXPORT_SYMBOL_GPL(work_on_cpu); 3677EXPORT_SYMBOL_GPL(work_on_cpu);
@@ -3744,11 +3801,7 @@ void thaw_workqueues(void)
3744 continue; 3801 continue;
3745 3802
3746 /* restore max_active and repopulate worklist */ 3803 /* restore max_active and repopulate worklist */
3747 cwq->max_active = wq->saved_max_active; 3804 cwq_set_max_active(cwq, wq->saved_max_active);
3748
3749 while (!list_empty(&cwq->delayed_works) &&
3750 cwq->nr_active < cwq->max_active)
3751 cwq_activate_first_delayed(cwq);
3752 } 3805 }
3753 3806
3754 for_each_worker_pool(pool, gcwq) 3807 for_each_worker_pool(pool, gcwq)
@@ -3768,8 +3821,12 @@ static int __init init_workqueues(void)
3768 unsigned int cpu; 3821 unsigned int cpu;
3769 int i; 3822 int i;
3770 3823
3824 /* make sure we have enough bits for OFFQ CPU number */
3825 BUILD_BUG_ON((1LU << (BITS_PER_LONG - WORK_OFFQ_CPU_SHIFT)) <
3826 WORK_CPU_LAST);
3827
3771 cpu_notifier(workqueue_cpu_up_callback, CPU_PRI_WORKQUEUE_UP); 3828 cpu_notifier(workqueue_cpu_up_callback, CPU_PRI_WORKQUEUE_UP);
3772 cpu_notifier(workqueue_cpu_down_callback, CPU_PRI_WORKQUEUE_DOWN); 3829 hotcpu_notifier(workqueue_cpu_down_callback, CPU_PRI_WORKQUEUE_DOWN);
3773 3830
3774 /* initialize gcwqs */ 3831 /* initialize gcwqs */
3775 for_each_gcwq_cpu(cpu) { 3832 for_each_gcwq_cpu(cpu) {
@@ -3795,11 +3852,9 @@ static int __init init_workqueues(void)
3795 setup_timer(&pool->mayday_timer, gcwq_mayday_timeout, 3852 setup_timer(&pool->mayday_timer, gcwq_mayday_timeout,
3796 (unsigned long)pool); 3853 (unsigned long)pool);
3797 3854
3798 mutex_init(&pool->manager_mutex); 3855 mutex_init(&pool->assoc_mutex);
3799 ida_init(&pool->worker_ida); 3856 ida_init(&pool->worker_ida);
3800 } 3857 }
3801
3802 init_waitqueue_head(&gcwq->rebind_hold);
3803 } 3858 }
3804 3859
3805 /* create the initial worker */ 3860 /* create the initial worker */
@@ -3822,17 +3877,14 @@ static int __init init_workqueues(void)
3822 } 3877 }
3823 3878
3824 system_wq = alloc_workqueue("events", 0, 0); 3879 system_wq = alloc_workqueue("events", 0, 0);
3880 system_highpri_wq = alloc_workqueue("events_highpri", WQ_HIGHPRI, 0);
3825 system_long_wq = alloc_workqueue("events_long", 0, 0); 3881 system_long_wq = alloc_workqueue("events_long", 0, 0);
3826 system_nrt_wq = alloc_workqueue("events_nrt", WQ_NON_REENTRANT, 0);
3827 system_unbound_wq = alloc_workqueue("events_unbound", WQ_UNBOUND, 3882 system_unbound_wq = alloc_workqueue("events_unbound", WQ_UNBOUND,
3828 WQ_UNBOUND_MAX_ACTIVE); 3883 WQ_UNBOUND_MAX_ACTIVE);
3829 system_freezable_wq = alloc_workqueue("events_freezable", 3884 system_freezable_wq = alloc_workqueue("events_freezable",
3830 WQ_FREEZABLE, 0); 3885 WQ_FREEZABLE, 0);
3831 system_nrt_freezable_wq = alloc_workqueue("events_nrt_freezable", 3886 BUG_ON(!system_wq || !system_highpri_wq || !system_long_wq ||
3832 WQ_NON_REENTRANT | WQ_FREEZABLE, 0); 3887 !system_unbound_wq || !system_freezable_wq);
3833 BUG_ON(!system_wq || !system_long_wq || !system_nrt_wq ||
3834 !system_unbound_wq || !system_freezable_wq ||
3835 !system_nrt_freezable_wq);
3836 return 0; 3888 return 0;
3837} 3889}
3838early_initcall(init_workqueues); 3890early_initcall(init_workqueues);
generated by cgit v1.2.2 (git 2.25.0) at 2025-02-25 17:48:12 -0500