aboutsummaryrefslogtreecommitdiffstats
path: root/kernel
diff options
context:
space:
mode:
Diffstat (limited to 'kernel')
-rw-r--r--kernel/Makefile6
-rw-r--r--kernel/cgroup.c6
-rw-r--r--kernel/configs.c1
-rw-r--r--kernel/cpuset.c4
-rw-r--r--kernel/debug/debug_core.c139
-rw-r--r--kernel/debug/debug_core.h1
-rw-r--r--kernel/debug/kdb/kdb_debugger.c3
-rw-r--r--kernel/debug/kdb/kdb_io.c2
-rw-r--r--kernel/debug/kdb/kdb_main.c18
-rw-r--r--kernel/debug/kdb/kdb_private.h48
-rw-r--r--kernel/early_res.c590
-rw-r--r--kernel/exit.c4
-rw-r--r--kernel/futex.c66
-rw-r--r--kernel/futex_compat.c2
-rw-r--r--kernel/gcov/fs.c1
-rw-r--r--kernel/hrtimer.c13
-rw-r--r--kernel/hung_task.c4
-rw-r--r--kernel/hw_breakpoint.c75
-rw-r--r--kernel/irq/Kconfig53
-rw-r--r--kernel/irq/Makefile3
-rw-r--r--kernel/irq/autoprobe.c15
-rw-r--r--kernel/irq/chip.c378
-rw-r--r--kernel/irq/dummychip.c68
-rw-r--r--kernel/irq/handle.c341
-rw-r--r--kernel/irq/internals.h39
-rw-r--r--kernel/irq/irqdesc.c395
-rw-r--r--kernel/irq/manage.c87
-rw-r--r--kernel/irq/migration.c12
-rw-r--r--kernel/irq/numa_migrate.c120
-rw-r--r--kernel/irq/proc.c26
-rw-r--r--kernel/irq/resend.c5
-rw-r--r--kernel/irq/spurious.c8
-rw-r--r--kernel/irq_work.c164
-rw-r--r--kernel/jump_label.c429
-rw-r--r--kernel/kfifo.c2
-rw-r--r--kernel/kprobes.c27
-rw-r--r--kernel/lockdep.c51
-rw-r--r--kernel/module.c10
-rw-r--r--kernel/perf_event.c2596
-rw-r--r--kernel/pid.c3
-rw-r--r--kernel/pm_qos_params.c1
-rw-r--r--kernel/power/Kconfig17
-rw-r--r--kernel/power/hibernate.c25
-rw-r--r--kernel/power/main.c29
-rw-r--r--kernel/power/power.h10
-rw-r--r--kernel/power/process.c11
-rw-r--r--kernel/power/snapshot.c13
-rw-r--r--kernel/power/swap.c300
-rw-r--r--kernel/printk.c4
-rw-r--r--kernel/profile.c1
-rw-r--r--kernel/rcupdate.c8
-rw-r--r--kernel/rcutiny.c33
-rw-r--r--kernel/rcutiny_plugin.h582
-rw-r--r--kernel/rcutorture.c17
-rw-r--r--kernel/rcutree.c92
-rw-r--r--kernel/rcutree.h20
-rw-r--r--kernel/rcutree_plugin.h47
-rw-r--r--kernel/rcutree_trace.c12
-rw-r--r--kernel/rtmutex-tester.c6
-rw-r--r--kernel/sched.c309
-rw-r--r--kernel/sched_fair.c81
-rw-r--r--kernel/sched_features.h5
-rw-r--r--kernel/sched_rt.c40
-rw-r--r--kernel/sched_stoptask.c108
-rw-r--r--kernel/signal.c8
-rw-r--r--kernel/smp.c17
-rw-r--r--kernel/softirq.c75
-rw-r--r--kernel/srcu.c2
-rw-r--r--kernel/stop_machine.c8
-rw-r--r--kernel/sys_ni.c1
-rw-r--r--kernel/sysctl.c2
-rw-r--r--kernel/sysctl_check.c9
-rw-r--r--kernel/test_kprobes.c12
-rw-r--r--kernel/time/ntp.c14
-rw-r--r--kernel/timer.c7
-rw-r--r--kernel/trace/Kconfig7
-rw-r--r--kernel/trace/blktrace.c16
-rw-r--r--kernel/trace/ftrace.c129
-rw-r--r--kernel/trace/ring_buffer.c24
-rw-r--r--kernel/trace/trace.c2
-rw-r--r--kernel/trace/trace.h4
-rw-r--r--kernel/trace/trace_event_perf.c28
-rw-r--r--kernel/trace/trace_events.c61
-rw-r--r--kernel/trace/trace_functions_graph.c209
-rw-r--r--kernel/trace/trace_irqsoff.c152
-rw-r--r--kernel/trace/trace_kdb.c1
-rw-r--r--kernel/trace/trace_sched_wakeup.c256
-rw-r--r--kernel/trace/trace_stack.c1
-rw-r--r--kernel/trace/trace_workqueue.c10
-rw-r--r--kernel/tracepoint.c14
-rw-r--r--kernel/watchdog.c43
-rw-r--r--kernel/workqueue.c310
92 files changed, 5756 insertions, 3252 deletions
diff --git a/kernel/Makefile b/kernel/Makefile
index 0b72d1a74be0..0b5ff083fa22 100644
--- a/kernel/Makefile
+++ b/kernel/Makefile
@@ -10,8 +10,7 @@ obj-y = sched.o 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 pm_qos_params.o sched_clock.o cred.o \ 12 notifier.o ksysfs.o pm_qos_params.o sched_clock.o cred.o \
13 async.o range.o 13 async.o range.o jump_label.o
14obj-$(CONFIG_HAVE_EARLY_RES) += early_res.o
15obj-y += groups.o 14obj-y += groups.o
16 15
17ifdef CONFIG_FUNCTION_TRACER 16ifdef CONFIG_FUNCTION_TRACER
@@ -23,6 +22,7 @@ CFLAGS_REMOVE_rtmutex-debug.o = -pg
23CFLAGS_REMOVE_cgroup-debug.o = -pg 22CFLAGS_REMOVE_cgroup-debug.o = -pg
24CFLAGS_REMOVE_sched_clock.o = -pg 23CFLAGS_REMOVE_sched_clock.o = -pg
25CFLAGS_REMOVE_perf_event.o = -pg 24CFLAGS_REMOVE_perf_event.o = -pg
25CFLAGS_REMOVE_irq_work.o = -pg
26endif 26endif
27 27
28obj-$(CONFIG_FREEZER) += freezer.o 28obj-$(CONFIG_FREEZER) += freezer.o
@@ -86,6 +86,7 @@ obj-$(CONFIG_TREE_RCU) += rcutree.o
86obj-$(CONFIG_TREE_PREEMPT_RCU) += rcutree.o 86obj-$(CONFIG_TREE_PREEMPT_RCU) += rcutree.o
87obj-$(CONFIG_TREE_RCU_TRACE) += rcutree_trace.o 87obj-$(CONFIG_TREE_RCU_TRACE) += rcutree_trace.o
88obj-$(CONFIG_TINY_RCU) += rcutiny.o 88obj-$(CONFIG_TINY_RCU) += rcutiny.o
89obj-$(CONFIG_TINY_PREEMPT_RCU) += rcutiny.o
89obj-$(CONFIG_RELAY) += relay.o 90obj-$(CONFIG_RELAY) += relay.o
90obj-$(CONFIG_SYSCTL) += utsname_sysctl.o 91obj-$(CONFIG_SYSCTL) += utsname_sysctl.o
91obj-$(CONFIG_TASK_DELAY_ACCT) += delayacct.o 92obj-$(CONFIG_TASK_DELAY_ACCT) += delayacct.o
@@ -100,6 +101,7 @@ obj-$(CONFIG_TRACING) += trace/
100obj-$(CONFIG_X86_DS) += trace/ 101obj-$(CONFIG_X86_DS) += trace/
101obj-$(CONFIG_RING_BUFFER) += trace/ 102obj-$(CONFIG_RING_BUFFER) += trace/
102obj-$(CONFIG_SMP) += sched_cpupri.o 103obj-$(CONFIG_SMP) += sched_cpupri.o
104obj-$(CONFIG_IRQ_WORK) += irq_work.o
103obj-$(CONFIG_PERF_EVENTS) += perf_event.o 105obj-$(CONFIG_PERF_EVENTS) += perf_event.o
104obj-$(CONFIG_HAVE_HW_BREAKPOINT) += hw_breakpoint.o 106obj-$(CONFIG_HAVE_HW_BREAKPOINT) += hw_breakpoint.o
105obj-$(CONFIG_USER_RETURN_NOTIFIER) += user-return-notifier.o 107obj-$(CONFIG_USER_RETURN_NOTIFIER) += user-return-notifier.o
diff --git a/kernel/cgroup.c b/kernel/cgroup.c
index c9483d8f6140..7b69b8d0313d 100644
--- a/kernel/cgroup.c
+++ b/kernel/cgroup.c
@@ -52,7 +52,6 @@
52#include <linux/cgroupstats.h> 52#include <linux/cgroupstats.h>
53#include <linux/hash.h> 53#include <linux/hash.h>
54#include <linux/namei.h> 54#include <linux/namei.h>
55#include <linux/smp_lock.h>
56#include <linux/pid_namespace.h> 55#include <linux/pid_namespace.h>
57#include <linux/idr.h> 56#include <linux/idr.h>
58#include <linux/vmalloc.h> /* TODO: replace with more sophisticated array */ 57#include <linux/vmalloc.h> /* TODO: replace with more sophisticated array */
@@ -138,7 +137,7 @@ struct css_id {
138 * is called after synchronize_rcu(). But for safe use, css_is_removed() 137 * is called after synchronize_rcu(). But for safe use, css_is_removed()
139 * css_tryget() should be used for avoiding race. 138 * css_tryget() should be used for avoiding race.
140 */ 139 */
141 struct cgroup_subsys_state *css; 140 struct cgroup_subsys_state __rcu *css;
142 /* 141 /*
143 * ID of this css. 142 * ID of this css.
144 */ 143 */
@@ -1222,7 +1221,6 @@ static int cgroup_remount(struct super_block *sb, int *flags, char *data)
1222 struct cgroup *cgrp = &root->top_cgroup; 1221 struct cgroup *cgrp = &root->top_cgroup;
1223 struct cgroup_sb_opts opts; 1222 struct cgroup_sb_opts opts;
1224 1223
1225 lock_kernel();
1226 mutex_lock(&cgrp->dentry->d_inode->i_mutex); 1224 mutex_lock(&cgrp->dentry->d_inode->i_mutex);
1227 mutex_lock(&cgroup_mutex); 1225 mutex_lock(&cgroup_mutex);
1228 1226
@@ -1255,7 +1253,6 @@ static int cgroup_remount(struct super_block *sb, int *flags, char *data)
1255 kfree(opts.name); 1253 kfree(opts.name);
1256 mutex_unlock(&cgroup_mutex); 1254 mutex_unlock(&cgroup_mutex);
1257 mutex_unlock(&cgrp->dentry->d_inode->i_mutex); 1255 mutex_unlock(&cgrp->dentry->d_inode->i_mutex);
1258 unlock_kernel();
1259 return ret; 1256 return ret;
1260} 1257}
1261 1258
@@ -1568,7 +1565,6 @@ static int cgroup_get_sb(struct file_system_type *fs_type,
1568 out_err: 1565 out_err:
1569 kfree(opts.release_agent); 1566 kfree(opts.release_agent);
1570 kfree(opts.name); 1567 kfree(opts.name);
1571
1572 return ret; 1568 return ret;
1573} 1569}
1574 1570
diff --git a/kernel/configs.c b/kernel/configs.c
index abaee684ecbf..b4066b44a99d 100644
--- a/kernel/configs.c
+++ b/kernel/configs.c
@@ -66,6 +66,7 @@ ikconfig_read_current(struct file *file, char __user *buf,
66static const struct file_operations ikconfig_file_ops = { 66static const struct file_operations ikconfig_file_ops = {
67 .owner = THIS_MODULE, 67 .owner = THIS_MODULE,
68 .read = ikconfig_read_current, 68 .read = ikconfig_read_current,
69 .llseek = default_llseek,
69}; 70};
70 71
71static int __init ikconfig_init(void) 72static int __init ikconfig_init(void)
diff --git a/kernel/cpuset.c b/kernel/cpuset.c
index b23c0979bbe7..51b143e2a07a 100644
--- a/kernel/cpuset.c
+++ b/kernel/cpuset.c
@@ -1397,7 +1397,7 @@ static int cpuset_can_attach(struct cgroup_subsys *ss, struct cgroup *cont,
1397 if (tsk->flags & PF_THREAD_BOUND) 1397 if (tsk->flags & PF_THREAD_BOUND)
1398 return -EINVAL; 1398 return -EINVAL;
1399 1399
1400 ret = security_task_setscheduler(tsk, 0, NULL); 1400 ret = security_task_setscheduler(tsk);
1401 if (ret) 1401 if (ret)
1402 return ret; 1402 return ret;
1403 if (threadgroup) { 1403 if (threadgroup) {
@@ -1405,7 +1405,7 @@ static int cpuset_can_attach(struct cgroup_subsys *ss, struct cgroup *cont,
1405 1405
1406 rcu_read_lock(); 1406 rcu_read_lock();
1407 list_for_each_entry_rcu(c, &tsk->thread_group, thread_group) { 1407 list_for_each_entry_rcu(c, &tsk->thread_group, thread_group) {
1408 ret = security_task_setscheduler(c, 0, NULL); 1408 ret = security_task_setscheduler(c);
1409 if (ret) { 1409 if (ret) {
1410 rcu_read_unlock(); 1410 rcu_read_unlock();
1411 return ret; 1411 return ret;
diff --git a/kernel/debug/debug_core.c b/kernel/debug/debug_core.c
index de407c78178d..fec596da9bd0 100644
--- a/kernel/debug/debug_core.c
+++ b/kernel/debug/debug_core.c
@@ -47,6 +47,7 @@
47#include <linux/pid.h> 47#include <linux/pid.h>
48#include <linux/smp.h> 48#include <linux/smp.h>
49#include <linux/mm.h> 49#include <linux/mm.h>
50#include <linux/rcupdate.h>
50 51
51#include <asm/cacheflush.h> 52#include <asm/cacheflush.h>
52#include <asm/byteorder.h> 53#include <asm/byteorder.h>
@@ -109,13 +110,15 @@ static struct kgdb_bkpt kgdb_break[KGDB_MAX_BREAKPOINTS] = {
109 */ 110 */
110atomic_t kgdb_active = ATOMIC_INIT(-1); 111atomic_t kgdb_active = ATOMIC_INIT(-1);
111EXPORT_SYMBOL_GPL(kgdb_active); 112EXPORT_SYMBOL_GPL(kgdb_active);
113static DEFINE_RAW_SPINLOCK(dbg_master_lock);
114static DEFINE_RAW_SPINLOCK(dbg_slave_lock);
112 115
113/* 116/*
114 * We use NR_CPUs not PERCPU, in case kgdb is used to debug early 117 * We use NR_CPUs not PERCPU, in case kgdb is used to debug early
115 * bootup code (which might not have percpu set up yet): 118 * bootup code (which might not have percpu set up yet):
116 */ 119 */
117static atomic_t passive_cpu_wait[NR_CPUS]; 120static atomic_t masters_in_kgdb;
118static atomic_t cpu_in_kgdb[NR_CPUS]; 121static atomic_t slaves_in_kgdb;
119static atomic_t kgdb_break_tasklet_var; 122static atomic_t kgdb_break_tasklet_var;
120atomic_t kgdb_setting_breakpoint; 123atomic_t kgdb_setting_breakpoint;
121 124
@@ -457,26 +460,32 @@ static int kgdb_reenter_check(struct kgdb_state *ks)
457 return 1; 460 return 1;
458} 461}
459 462
460static void dbg_cpu_switch(int cpu, int next_cpu) 463static void dbg_touch_watchdogs(void)
461{ 464{
462 /* Mark the cpu we are switching away from as a slave when it 465 touch_softlockup_watchdog_sync();
463 * holds the kgdb_active token. This must be done so that the 466 clocksource_touch_watchdog();
464 * that all the cpus wait in for the debug core will not enter 467 rcu_cpu_stall_reset();
465 * again as the master. */
466 if (cpu == atomic_read(&kgdb_active)) {
467 kgdb_info[cpu].exception_state |= DCPU_IS_SLAVE;
468 kgdb_info[cpu].exception_state &= ~DCPU_WANT_MASTER;
469 }
470 kgdb_info[next_cpu].exception_state |= DCPU_NEXT_MASTER;
471} 468}
472 469
473static int kgdb_cpu_enter(struct kgdb_state *ks, struct pt_regs *regs) 470static int kgdb_cpu_enter(struct kgdb_state *ks, struct pt_regs *regs,
471 int exception_state)
474{ 472{
475 unsigned long flags; 473 unsigned long flags;
476 int sstep_tries = 100; 474 int sstep_tries = 100;
477 int error; 475 int error;
478 int i, cpu; 476 int cpu;
479 int trace_on = 0; 477 int trace_on = 0;
478 int online_cpus = num_online_cpus();
479
480 kgdb_info[ks->cpu].enter_kgdb++;
481 kgdb_info[ks->cpu].exception_state |= exception_state;
482
483 if (exception_state == DCPU_WANT_MASTER)
484 atomic_inc(&masters_in_kgdb);
485 else
486 atomic_inc(&slaves_in_kgdb);
487 kgdb_disable_hw_debug(ks->linux_regs);
488
480acquirelock: 489acquirelock:
481 /* 490 /*
482 * Interrupts will be restored by the 'trap return' code, except when 491 * Interrupts will be restored by the 'trap return' code, except when
@@ -489,14 +498,15 @@ acquirelock:
489 kgdb_info[cpu].task = current; 498 kgdb_info[cpu].task = current;
490 kgdb_info[cpu].ret_state = 0; 499 kgdb_info[cpu].ret_state = 0;
491 kgdb_info[cpu].irq_depth = hardirq_count() >> HARDIRQ_SHIFT; 500 kgdb_info[cpu].irq_depth = hardirq_count() >> HARDIRQ_SHIFT;
492 /*
493 * Make sure the above info reaches the primary CPU before
494 * our cpu_in_kgdb[] flag setting does:
495 */
496 atomic_inc(&cpu_in_kgdb[cpu]);
497 501
498 if (exception_level == 1) 502 /* Make sure the above info reaches the primary CPU */
503 smp_mb();
504
505 if (exception_level == 1) {
506 if (raw_spin_trylock(&dbg_master_lock))
507 atomic_xchg(&kgdb_active, cpu);
499 goto cpu_master_loop; 508 goto cpu_master_loop;
509 }
500 510
501 /* 511 /*
502 * CPU will loop if it is a slave or request to become a kgdb 512 * CPU will loop if it is a slave or request to become a kgdb
@@ -508,10 +518,12 @@ cpu_loop:
508 kgdb_info[cpu].exception_state &= ~DCPU_NEXT_MASTER; 518 kgdb_info[cpu].exception_state &= ~DCPU_NEXT_MASTER;
509 goto cpu_master_loop; 519 goto cpu_master_loop;
510 } else if (kgdb_info[cpu].exception_state & DCPU_WANT_MASTER) { 520 } else if (kgdb_info[cpu].exception_state & DCPU_WANT_MASTER) {
511 if (atomic_cmpxchg(&kgdb_active, -1, cpu) == cpu) 521 if (raw_spin_trylock(&dbg_master_lock)) {
522 atomic_xchg(&kgdb_active, cpu);
512 break; 523 break;
524 }
513 } else if (kgdb_info[cpu].exception_state & DCPU_IS_SLAVE) { 525 } else if (kgdb_info[cpu].exception_state & DCPU_IS_SLAVE) {
514 if (!atomic_read(&passive_cpu_wait[cpu])) 526 if (!raw_spin_is_locked(&dbg_slave_lock))
515 goto return_normal; 527 goto return_normal;
516 } else { 528 } else {
517return_normal: 529return_normal:
@@ -522,9 +534,12 @@ return_normal:
522 arch_kgdb_ops.correct_hw_break(); 534 arch_kgdb_ops.correct_hw_break();
523 if (trace_on) 535 if (trace_on)
524 tracing_on(); 536 tracing_on();
525 atomic_dec(&cpu_in_kgdb[cpu]); 537 kgdb_info[cpu].exception_state &=
526 touch_softlockup_watchdog_sync(); 538 ~(DCPU_WANT_MASTER | DCPU_IS_SLAVE);
527 clocksource_touch_watchdog(); 539 kgdb_info[cpu].enter_kgdb--;
540 smp_mb__before_atomic_dec();
541 atomic_dec(&slaves_in_kgdb);
542 dbg_touch_watchdogs();
528 local_irq_restore(flags); 543 local_irq_restore(flags);
529 return 0; 544 return 0;
530 } 545 }
@@ -541,8 +556,8 @@ return_normal:
541 (kgdb_info[cpu].task && 556 (kgdb_info[cpu].task &&
542 kgdb_info[cpu].task->pid != kgdb_sstep_pid) && --sstep_tries) { 557 kgdb_info[cpu].task->pid != kgdb_sstep_pid) && --sstep_tries) {
543 atomic_set(&kgdb_active, -1); 558 atomic_set(&kgdb_active, -1);
544 touch_softlockup_watchdog_sync(); 559 raw_spin_unlock(&dbg_master_lock);
545 clocksource_touch_watchdog(); 560 dbg_touch_watchdogs();
546 local_irq_restore(flags); 561 local_irq_restore(flags);
547 562
548 goto acquirelock; 563 goto acquirelock;
@@ -563,16 +578,12 @@ return_normal:
563 if (dbg_io_ops->pre_exception) 578 if (dbg_io_ops->pre_exception)
564 dbg_io_ops->pre_exception(); 579 dbg_io_ops->pre_exception();
565 580
566 kgdb_disable_hw_debug(ks->linux_regs);
567
568 /* 581 /*
569 * Get the passive CPU lock which will hold all the non-primary 582 * Get the passive CPU lock which will hold all the non-primary
570 * CPU in a spin state while the debugger is active 583 * CPU in a spin state while the debugger is active
571 */ 584 */
572 if (!kgdb_single_step) { 585 if (!kgdb_single_step)
573 for (i = 0; i < NR_CPUS; i++) 586 raw_spin_lock(&dbg_slave_lock);
574 atomic_inc(&passive_cpu_wait[i]);
575 }
576 587
577#ifdef CONFIG_SMP 588#ifdef CONFIG_SMP
578 /* Signal the other CPUs to enter kgdb_wait() */ 589 /* Signal the other CPUs to enter kgdb_wait() */
@@ -583,10 +594,9 @@ return_normal:
583 /* 594 /*
584 * Wait for the other CPUs to be notified and be waiting for us: 595 * Wait for the other CPUs to be notified and be waiting for us:
585 */ 596 */
586 for_each_online_cpu(i) { 597 while (kgdb_do_roundup && (atomic_read(&masters_in_kgdb) +
587 while (kgdb_do_roundup && !atomic_read(&cpu_in_kgdb[i])) 598 atomic_read(&slaves_in_kgdb)) != online_cpus)
588 cpu_relax(); 599 cpu_relax();
589 }
590 600
591 /* 601 /*
592 * At this point the primary processor is completely 602 * At this point the primary processor is completely
@@ -615,7 +625,8 @@ cpu_master_loop:
615 if (error == DBG_PASS_EVENT) { 625 if (error == DBG_PASS_EVENT) {
616 dbg_kdb_mode = !dbg_kdb_mode; 626 dbg_kdb_mode = !dbg_kdb_mode;
617 } else if (error == DBG_SWITCH_CPU_EVENT) { 627 } else if (error == DBG_SWITCH_CPU_EVENT) {
618 dbg_cpu_switch(cpu, dbg_switch_cpu); 628 kgdb_info[dbg_switch_cpu].exception_state |=
629 DCPU_NEXT_MASTER;
619 goto cpu_loop; 630 goto cpu_loop;
620 } else { 631 } else {
621 kgdb_info[cpu].ret_state = error; 632 kgdb_info[cpu].ret_state = error;
@@ -627,24 +638,11 @@ cpu_master_loop:
627 if (dbg_io_ops->post_exception) 638 if (dbg_io_ops->post_exception)
628 dbg_io_ops->post_exception(); 639 dbg_io_ops->post_exception();
629 640
630 atomic_dec(&cpu_in_kgdb[ks->cpu]);
631
632 if (!kgdb_single_step) { 641 if (!kgdb_single_step) {
633 for (i = NR_CPUS-1; i >= 0; i--) 642 raw_spin_unlock(&dbg_slave_lock);
634 atomic_dec(&passive_cpu_wait[i]); 643 /* Wait till all the CPUs have quit from the debugger. */
635 /* 644 while (kgdb_do_roundup && atomic_read(&slaves_in_kgdb))
636 * Wait till all the CPUs have quit from the debugger, 645 cpu_relax();
637 * but allow a CPU that hit an exception and is
638 * waiting to become the master to remain in the debug
639 * core.
640 */
641 for_each_online_cpu(i) {
642 while (kgdb_do_roundup &&
643 atomic_read(&cpu_in_kgdb[i]) &&
644 !(kgdb_info[i].exception_state &
645 DCPU_WANT_MASTER))
646 cpu_relax();
647 }
648 } 646 }
649 647
650kgdb_restore: 648kgdb_restore:
@@ -655,12 +653,20 @@ kgdb_restore:
655 else 653 else
656 kgdb_sstep_pid = 0; 654 kgdb_sstep_pid = 0;
657 } 655 }
656 if (arch_kgdb_ops.correct_hw_break)
657 arch_kgdb_ops.correct_hw_break();
658 if (trace_on) 658 if (trace_on)
659 tracing_on(); 659 tracing_on();
660
661 kgdb_info[cpu].exception_state &=
662 ~(DCPU_WANT_MASTER | DCPU_IS_SLAVE);
663 kgdb_info[cpu].enter_kgdb--;
664 smp_mb__before_atomic_dec();
665 atomic_dec(&masters_in_kgdb);
660 /* Free kgdb_active */ 666 /* Free kgdb_active */
661 atomic_set(&kgdb_active, -1); 667 atomic_set(&kgdb_active, -1);
662 touch_softlockup_watchdog_sync(); 668 raw_spin_unlock(&dbg_master_lock);
663 clocksource_touch_watchdog(); 669 dbg_touch_watchdogs();
664 local_irq_restore(flags); 670 local_irq_restore(flags);
665 671
666 return kgdb_info[cpu].ret_state; 672 return kgdb_info[cpu].ret_state;
@@ -678,7 +684,6 @@ kgdb_handle_exception(int evector, int signo, int ecode, struct pt_regs *regs)
678{ 684{
679 struct kgdb_state kgdb_var; 685 struct kgdb_state kgdb_var;
680 struct kgdb_state *ks = &kgdb_var; 686 struct kgdb_state *ks = &kgdb_var;
681 int ret;
682 687
683 ks->cpu = raw_smp_processor_id(); 688 ks->cpu = raw_smp_processor_id();
684 ks->ex_vector = evector; 689 ks->ex_vector = evector;
@@ -689,11 +694,10 @@ kgdb_handle_exception(int evector, int signo, int ecode, struct pt_regs *regs)
689 694
690 if (kgdb_reenter_check(ks)) 695 if (kgdb_reenter_check(ks))
691 return 0; /* Ouch, double exception ! */ 696 return 0; /* Ouch, double exception ! */
692 kgdb_info[ks->cpu].exception_state |= DCPU_WANT_MASTER; 697 if (kgdb_info[ks->cpu].enter_kgdb != 0)
693 ret = kgdb_cpu_enter(ks, regs); 698 return 0;
694 kgdb_info[ks->cpu].exception_state &= ~(DCPU_WANT_MASTER | 699
695 DCPU_IS_SLAVE); 700 return kgdb_cpu_enter(ks, regs, DCPU_WANT_MASTER);
696 return ret;
697} 701}
698 702
699int kgdb_nmicallback(int cpu, void *regs) 703int kgdb_nmicallback(int cpu, void *regs)
@@ -706,12 +710,9 @@ int kgdb_nmicallback(int cpu, void *regs)
706 ks->cpu = cpu; 710 ks->cpu = cpu;
707 ks->linux_regs = regs; 711 ks->linux_regs = regs;
708 712
709 if (!atomic_read(&cpu_in_kgdb[cpu]) && 713 if (kgdb_info[ks->cpu].enter_kgdb == 0 &&
710 atomic_read(&kgdb_active) != -1 && 714 raw_spin_is_locked(&dbg_master_lock)) {
711 atomic_read(&kgdb_active) != cpu) { 715 kgdb_cpu_enter(ks, regs, DCPU_IS_SLAVE);
712 kgdb_info[cpu].exception_state |= DCPU_IS_SLAVE;
713 kgdb_cpu_enter(ks, regs);
714 kgdb_info[cpu].exception_state &= ~DCPU_IS_SLAVE;
715 return 0; 716 return 0;
716 } 717 }
717#endif 718#endif
diff --git a/kernel/debug/debug_core.h b/kernel/debug/debug_core.h
index c5d753d80f67..3494c28a7e7a 100644
--- a/kernel/debug/debug_core.h
+++ b/kernel/debug/debug_core.h
@@ -40,6 +40,7 @@ struct debuggerinfo_struct {
40 int exception_state; 40 int exception_state;
41 int ret_state; 41 int ret_state;
42 int irq_depth; 42 int irq_depth;
43 int enter_kgdb;
43}; 44};
44 45
45extern struct debuggerinfo_struct kgdb_info[]; 46extern struct debuggerinfo_struct kgdb_info[];
diff --git a/kernel/debug/kdb/kdb_debugger.c b/kernel/debug/kdb/kdb_debugger.c
index bf6e8270e957..dd0b1b7dd02c 100644
--- a/kernel/debug/kdb/kdb_debugger.c
+++ b/kernel/debug/kdb/kdb_debugger.c
@@ -86,7 +86,7 @@ int kdb_stub(struct kgdb_state *ks)
86 } 86 }
87 /* Set initial kdb state variables */ 87 /* Set initial kdb state variables */
88 KDB_STATE_CLEAR(KGDB_TRANS); 88 KDB_STATE_CLEAR(KGDB_TRANS);
89 kdb_initial_cpu = ks->cpu; 89 kdb_initial_cpu = atomic_read(&kgdb_active);
90 kdb_current_task = kgdb_info[ks->cpu].task; 90 kdb_current_task = kgdb_info[ks->cpu].task;
91 kdb_current_regs = kgdb_info[ks->cpu].debuggerinfo; 91 kdb_current_regs = kgdb_info[ks->cpu].debuggerinfo;
92 /* Remove any breakpoints as needed by kdb and clear single step */ 92 /* Remove any breakpoints as needed by kdb and clear single step */
@@ -105,7 +105,6 @@ int kdb_stub(struct kgdb_state *ks)
105 ks->pass_exception = 1; 105 ks->pass_exception = 1;
106 KDB_FLAG_SET(CATASTROPHIC); 106 KDB_FLAG_SET(CATASTROPHIC);
107 } 107 }
108 kdb_initial_cpu = ks->cpu;
109 if (KDB_STATE(SSBPT) && reason == KDB_REASON_SSTEP) { 108 if (KDB_STATE(SSBPT) && reason == KDB_REASON_SSTEP) {
110 KDB_STATE_CLEAR(SSBPT); 109 KDB_STATE_CLEAR(SSBPT);
111 KDB_STATE_CLEAR(DOING_SS); 110 KDB_STATE_CLEAR(DOING_SS);
diff --git a/kernel/debug/kdb/kdb_io.c b/kernel/debug/kdb/kdb_io.c
index c9b7f4f90bba..96fdaac46a80 100644
--- a/kernel/debug/kdb/kdb_io.c
+++ b/kernel/debug/kdb/kdb_io.c
@@ -823,4 +823,4 @@ int kdb_printf(const char *fmt, ...)
823 823
824 return r; 824 return r;
825} 825}
826 826EXPORT_SYMBOL_GPL(kdb_printf);
diff --git a/kernel/debug/kdb/kdb_main.c b/kernel/debug/kdb/kdb_main.c
index caf057a3de0e..d7bda21a106b 100644
--- a/kernel/debug/kdb/kdb_main.c
+++ b/kernel/debug/kdb/kdb_main.c
@@ -1749,13 +1749,13 @@ static int kdb_go(int argc, const char **argv)
1749 int nextarg; 1749 int nextarg;
1750 long offset; 1750 long offset;
1751 1751
1752 if (raw_smp_processor_id() != kdb_initial_cpu) {
1753 kdb_printf("go must execute on the entry cpu, "
1754 "please use \"cpu %d\" and then execute go\n",
1755 kdb_initial_cpu);
1756 return KDB_BADCPUNUM;
1757 }
1752 if (argc == 1) { 1758 if (argc == 1) {
1753 if (raw_smp_processor_id() != kdb_initial_cpu) {
1754 kdb_printf("go <address> must be issued from the "
1755 "initial cpu, do cpu %d first\n",
1756 kdb_initial_cpu);
1757 return KDB_ARGCOUNT;
1758 }
1759 nextarg = 1; 1759 nextarg = 1;
1760 diag = kdbgetaddrarg(argc, argv, &nextarg, 1760 diag = kdbgetaddrarg(argc, argv, &nextarg,
1761 &addr, &offset, NULL); 1761 &addr, &offset, NULL);
@@ -2783,6 +2783,8 @@ int kdb_register_repeat(char *cmd,
2783 2783
2784 return 0; 2784 return 0;
2785} 2785}
2786EXPORT_SYMBOL_GPL(kdb_register_repeat);
2787
2786 2788
2787/* 2789/*
2788 * kdb_register - Compatibility register function for commands that do 2790 * kdb_register - Compatibility register function for commands that do
@@ -2805,6 +2807,7 @@ int kdb_register(char *cmd,
2805 return kdb_register_repeat(cmd, func, usage, help, minlen, 2807 return kdb_register_repeat(cmd, func, usage, help, minlen,
2806 KDB_REPEAT_NONE); 2808 KDB_REPEAT_NONE);
2807} 2809}
2810EXPORT_SYMBOL_GPL(kdb_register);
2808 2811
2809/* 2812/*
2810 * kdb_unregister - This function is used to unregister a kernel 2813 * kdb_unregister - This function is used to unregister a kernel
@@ -2823,7 +2826,7 @@ int kdb_unregister(char *cmd)
2823 /* 2826 /*
2824 * find the command. 2827 * find the command.
2825 */ 2828 */
2826 for (i = 0, kp = kdb_commands; i < kdb_max_commands; i++, kp++) { 2829 for_each_kdbcmd(kp, i) {
2827 if (kp->cmd_name && (strcmp(kp->cmd_name, cmd) == 0)) { 2830 if (kp->cmd_name && (strcmp(kp->cmd_name, cmd) == 0)) {
2828 kp->cmd_name = NULL; 2831 kp->cmd_name = NULL;
2829 return 0; 2832 return 0;
@@ -2833,6 +2836,7 @@ int kdb_unregister(char *cmd)
2833 /* Couldn't find it. */ 2836 /* Couldn't find it. */
2834 return 1; 2837 return 1;
2835} 2838}
2839EXPORT_SYMBOL_GPL(kdb_unregister);
2836 2840
2837/* Initialize the kdb command table. */ 2841/* Initialize the kdb command table. */
2838static void __init kdb_inittab(void) 2842static void __init kdb_inittab(void)
diff --git a/kernel/debug/kdb/kdb_private.h b/kernel/debug/kdb/kdb_private.h
index be775f7e81e0..35d69ed1dfb5 100644
--- a/kernel/debug/kdb/kdb_private.h
+++ b/kernel/debug/kdb/kdb_private.h
@@ -15,29 +15,6 @@
15#include <linux/kgdb.h> 15#include <linux/kgdb.h>
16#include "../debug_core.h" 16#include "../debug_core.h"
17 17
18/* Kernel Debugger Error codes. Must not overlap with command codes. */
19#define KDB_NOTFOUND (-1)
20#define KDB_ARGCOUNT (-2)
21#define KDB_BADWIDTH (-3)
22#define KDB_BADRADIX (-4)
23#define KDB_NOTENV (-5)
24#define KDB_NOENVVALUE (-6)
25#define KDB_NOTIMP (-7)
26#define KDB_ENVFULL (-8)
27#define KDB_ENVBUFFULL (-9)
28#define KDB_TOOMANYBPT (-10)
29#define KDB_TOOMANYDBREGS (-11)
30#define KDB_DUPBPT (-12)
31#define KDB_BPTNOTFOUND (-13)
32#define KDB_BADMODE (-14)
33#define KDB_BADINT (-15)
34#define KDB_INVADDRFMT (-16)
35#define KDB_BADREG (-17)
36#define KDB_BADCPUNUM (-18)
37#define KDB_BADLENGTH (-19)
38#define KDB_NOBP (-20)
39#define KDB_BADADDR (-21)
40
41/* Kernel Debugger Command codes. Must not overlap with error codes. */ 18/* Kernel Debugger Command codes. Must not overlap with error codes. */
42#define KDB_CMD_GO (-1001) 19#define KDB_CMD_GO (-1001)
43#define KDB_CMD_CPU (-1002) 20#define KDB_CMD_CPU (-1002)
@@ -93,17 +70,6 @@
93 */ 70 */
94#define KDB_MAXBPT 16 71#define KDB_MAXBPT 16
95 72
96/* Maximum number of arguments to a function */
97#define KDB_MAXARGS 16
98
99typedef enum {
100 KDB_REPEAT_NONE = 0, /* Do not repeat this command */
101 KDB_REPEAT_NO_ARGS, /* Repeat the command without arguments */
102 KDB_REPEAT_WITH_ARGS, /* Repeat the command including its arguments */
103} kdb_repeat_t;
104
105typedef int (*kdb_func_t)(int, const char **);
106
107/* Symbol table format returned by kallsyms. */ 73/* Symbol table format returned by kallsyms. */
108typedef struct __ksymtab { 74typedef struct __ksymtab {
109 unsigned long value; /* Address of symbol */ 75 unsigned long value; /* Address of symbol */
@@ -123,11 +89,6 @@ extern int kallsyms_symbol_next(char *prefix_name, int flag);
123extern int kallsyms_symbol_complete(char *prefix_name, int max_len); 89extern int kallsyms_symbol_complete(char *prefix_name, int max_len);
124 90
125/* Exported Symbols for kernel loadable modules to use. */ 91/* Exported Symbols for kernel loadable modules to use. */
126extern int kdb_register(char *, kdb_func_t, char *, char *, short);
127extern int kdb_register_repeat(char *, kdb_func_t, char *, char *,
128 short, kdb_repeat_t);
129extern int kdb_unregister(char *);
130
131extern int kdb_getarea_size(void *, unsigned long, size_t); 92extern int kdb_getarea_size(void *, unsigned long, size_t);
132extern int kdb_putarea_size(unsigned long, void *, size_t); 93extern int kdb_putarea_size(unsigned long, void *, size_t);
133 94
@@ -144,6 +105,7 @@ extern int kdb_getword(unsigned long *, unsigned long, size_t);
144extern int kdb_putword(unsigned long, unsigned long, size_t); 105extern int kdb_putword(unsigned long, unsigned long, size_t);
145 106
146extern int kdbgetularg(const char *, unsigned long *); 107extern int kdbgetularg(const char *, unsigned long *);
108extern int kdbgetu64arg(const char *, u64 *);
147extern char *kdbgetenv(const char *); 109extern char *kdbgetenv(const char *);
148extern int kdbgetaddrarg(int, const char **, int*, unsigned long *, 110extern int kdbgetaddrarg(int, const char **, int*, unsigned long *,
149 long *, char **); 111 long *, char **);
@@ -255,14 +217,6 @@ extern void kdb_ps1(const struct task_struct *p);
255extern void kdb_print_nameval(const char *name, unsigned long val); 217extern void kdb_print_nameval(const char *name, unsigned long val);
256extern void kdb_send_sig_info(struct task_struct *p, struct siginfo *info); 218extern void kdb_send_sig_info(struct task_struct *p, struct siginfo *info);
257extern void kdb_meminfo_proc_show(void); 219extern void kdb_meminfo_proc_show(void);
258#ifdef CONFIG_KALLSYMS
259extern const char *kdb_walk_kallsyms(loff_t *pos);
260#else /* ! CONFIG_KALLSYMS */
261static inline const char *kdb_walk_kallsyms(loff_t *pos)
262{
263 return NULL;
264}
265#endif /* ! CONFIG_KALLSYMS */
266extern char *kdb_getstr(char *, size_t, char *); 220extern char *kdb_getstr(char *, size_t, char *);
267 221
268/* Defines for kdb_symbol_print */ 222/* Defines for kdb_symbol_print */
diff --git a/kernel/early_res.c b/kernel/early_res.c
deleted file mode 100644
index 7bfae887f211..000000000000
--- a/kernel/early_res.c
+++ /dev/null
@@ -1,590 +0,0 @@
1/*
2 * early_res, could be used to replace bootmem
3 */
4#include <linux/kernel.h>
5#include <linux/types.h>
6#include <linux/init.h>
7#include <linux/bootmem.h>
8#include <linux/mm.h>
9#include <linux/early_res.h>
10#include <linux/slab.h>
11#include <linux/kmemleak.h>
12
13/*
14 * Early reserved memory areas.
15 */
16/*
17 * need to make sure this one is bigger enough before
18 * find_fw_memmap_area could be used
19 */
20#define MAX_EARLY_RES_X 32
21
22struct early_res {
23 u64 start, end;
24 char name[15];
25 char overlap_ok;
26};
27static struct early_res early_res_x[MAX_EARLY_RES_X] __initdata;
28
29static int max_early_res __initdata = MAX_EARLY_RES_X;
30static struct early_res *early_res __initdata = &early_res_x[0];
31static int early_res_count __initdata;
32
33static int __init find_overlapped_early(u64 start, u64 end)
34{
35 int i;
36 struct early_res *r;
37
38 for (i = 0; i < max_early_res && early_res[i].end; i++) {
39 r = &early_res[i];
40 if (end > r->start && start < r->end)
41 break;
42 }
43
44 return i;
45}
46
47/*
48 * Drop the i-th range from the early reservation map,
49 * by copying any higher ranges down one over it, and
50 * clearing what had been the last slot.
51 */
52static void __init drop_range(int i)
53{
54 int j;
55
56 for (j = i + 1; j < max_early_res && early_res[j].end; j++)
57 ;
58
59 memmove(&early_res[i], &early_res[i + 1],
60 (j - 1 - i) * sizeof(struct early_res));
61
62 early_res[j - 1].end = 0;
63 early_res_count--;
64}
65
66static void __init drop_range_partial(int i, u64 start, u64 end)
67{
68 u64 common_start, common_end;
69 u64 old_start, old_end;
70
71 old_start = early_res[i].start;
72 old_end = early_res[i].end;
73 common_start = max(old_start, start);
74 common_end = min(old_end, end);
75
76 /* no overlap ? */
77 if (common_start >= common_end)
78 return;
79
80 if (old_start < common_start) {
81 /* make head segment */
82 early_res[i].end = common_start;
83 if (old_end > common_end) {
84 char name[15];
85
86 /*
87 * Save a local copy of the name, since the
88 * early_res array could get resized inside
89 * reserve_early_without_check() ->
90 * __check_and_double_early_res(), which would
91 * make the current name pointer invalid.
92 */
93 strncpy(name, early_res[i].name,
94 sizeof(early_res[i].name) - 1);
95 /* add another for left over on tail */
96 reserve_early_without_check(common_end, old_end, name);
97 }
98 return;
99 } else {
100 if (old_end > common_end) {
101 /* reuse the entry for tail left */
102 early_res[i].start = common_end;
103 return;
104 }
105 /* all covered */
106 drop_range(i);
107 }
108}
109
110/*
111 * Split any existing ranges that:
112 * 1) are marked 'overlap_ok', and
113 * 2) overlap with the stated range [start, end)
114 * into whatever portion (if any) of the existing range is entirely
115 * below or entirely above the stated range. Drop the portion
116 * of the existing range that overlaps with the stated range,
117 * which will allow the caller of this routine to then add that
118 * stated range without conflicting with any existing range.
119 */
120static void __init drop_overlaps_that_are_ok(u64 start, u64 end)
121{
122 int i;
123 struct early_res *r;
124 u64 lower_start, lower_end;
125 u64 upper_start, upper_end;
126 char name[15];
127
128 for (i = 0; i < max_early_res && early_res[i].end; i++) {
129 r = &early_res[i];
130
131 /* Continue past non-overlapping ranges */
132 if (end <= r->start || start >= r->end)
133 continue;
134
135 /*
136 * Leave non-ok overlaps as is; let caller
137 * panic "Overlapping early reservations"
138 * when it hits this overlap.
139 */
140 if (!r->overlap_ok)
141 return;
142
143 /*
144 * We have an ok overlap. We will drop it from the early
145 * reservation map, and add back in any non-overlapping
146 * portions (lower or upper) as separate, overlap_ok,
147 * non-overlapping ranges.
148 */
149
150 /* 1. Note any non-overlapping (lower or upper) ranges. */
151 strncpy(name, r->name, sizeof(name) - 1);
152
153 lower_start = lower_end = 0;
154 upper_start = upper_end = 0;
155 if (r->start < start) {
156 lower_start = r->start;
157 lower_end = start;
158 }
159 if (r->end > end) {
160 upper_start = end;
161 upper_end = r->end;
162 }
163
164 /* 2. Drop the original ok overlapping range */
165 drop_range(i);
166
167 i--; /* resume for-loop on copied down entry */
168
169 /* 3. Add back in any non-overlapping ranges. */
170 if (lower_end)
171 reserve_early_overlap_ok(lower_start, lower_end, name);
172 if (upper_end)
173 reserve_early_overlap_ok(upper_start, upper_end, name);
174 }
175}
176
177static void __init __reserve_early(u64 start, u64 end, char *name,
178 int overlap_ok)
179{
180 int i;
181 struct early_res *r;
182
183 i = find_overlapped_early(start, end);
184 if (i >= max_early_res)
185 panic("Too many early reservations");
186 r = &early_res[i];
187 if (r->end)
188 panic("Overlapping early reservations "
189 "%llx-%llx %s to %llx-%llx %s\n",
190 start, end - 1, name ? name : "", r->start,
191 r->end - 1, r->name);
192 r->start = start;
193 r->end = end;
194 r->overlap_ok = overlap_ok;
195 if (name)
196 strncpy(r->name, name, sizeof(r->name) - 1);
197 early_res_count++;
198}
199
200/*
201 * A few early reservtations come here.
202 *
203 * The 'overlap_ok' in the name of this routine does -not- mean it
204 * is ok for these reservations to overlap an earlier reservation.
205 * Rather it means that it is ok for subsequent reservations to
206 * overlap this one.
207 *
208 * Use this entry point to reserve early ranges when you are doing
209 * so out of "Paranoia", reserving perhaps more memory than you need,
210 * just in case, and don't mind a subsequent overlapping reservation
211 * that is known to be needed.
212 *
213 * The drop_overlaps_that_are_ok() call here isn't really needed.
214 * It would be needed if we had two colliding 'overlap_ok'
215 * reservations, so that the second such would not panic on the
216 * overlap with the first. We don't have any such as of this
217 * writing, but might as well tolerate such if it happens in
218 * the future.
219 */
220void __init reserve_early_overlap_ok(u64 start, u64 end, char *name)
221{
222 drop_overlaps_that_are_ok(start, end);
223 __reserve_early(start, end, name, 1);
224}
225
226static void __init __check_and_double_early_res(u64 ex_start, u64 ex_end)
227{
228 u64 start, end, size, mem;
229 struct early_res *new;
230
231 /* do we have enough slots left ? */
232 if ((max_early_res - early_res_count) > max(max_early_res/8, 2))
233 return;
234
235 /* double it */
236 mem = -1ULL;
237 size = sizeof(struct early_res) * max_early_res * 2;
238 if (early_res == early_res_x)
239 start = 0;
240 else
241 start = early_res[0].end;
242 end = ex_start;
243 if (start + size < end)
244 mem = find_fw_memmap_area(start, end, size,
245 sizeof(struct early_res));
246 if (mem == -1ULL) {
247 start = ex_end;
248 end = get_max_mapped();
249 if (start + size < end)
250 mem = find_fw_memmap_area(start, end, size,
251 sizeof(struct early_res));
252 }
253 if (mem == -1ULL)
254 panic("can not find more space for early_res array");
255
256 new = __va(mem);
257 /* save the first one for own */
258 new[0].start = mem;
259 new[0].end = mem + size;
260 new[0].overlap_ok = 0;
261 /* copy old to new */
262 if (early_res == early_res_x) {
263 memcpy(&new[1], &early_res[0],
264 sizeof(struct early_res) * max_early_res);
265 memset(&new[max_early_res+1], 0,
266 sizeof(struct early_res) * (max_early_res - 1));
267 early_res_count++;
268 } else {
269 memcpy(&new[1], &early_res[1],
270 sizeof(struct early_res) * (max_early_res - 1));
271 memset(&new[max_early_res], 0,
272 sizeof(struct early_res) * max_early_res);
273 }
274 memset(&early_res[0], 0, sizeof(struct early_res) * max_early_res);
275 early_res = new;
276 max_early_res *= 2;
277 printk(KERN_DEBUG "early_res array is doubled to %d at [%llx - %llx]\n",
278 max_early_res, mem, mem + size - 1);
279}
280
281/*
282 * Most early reservations come here.
283 *
284 * We first have drop_overlaps_that_are_ok() drop any pre-existing
285 * 'overlap_ok' ranges, so that we can then reserve this memory
286 * range without risk of panic'ing on an overlapping overlap_ok
287 * early reservation.
288 */
289void __init reserve_early(u64 start, u64 end, char *name)
290{
291 if (start >= end)
292 return;
293
294 __check_and_double_early_res(start, end);
295
296 drop_overlaps_that_are_ok(start, end);
297 __reserve_early(start, end, name, 0);
298}
299
300void __init reserve_early_without_check(u64 start, u64 end, char *name)
301{
302 struct early_res *r;
303
304 if (start >= end)
305 return;
306
307 __check_and_double_early_res(start, end);
308
309 r = &early_res[early_res_count];
310
311 r->start = start;
312 r->end = end;
313 r->overlap_ok = 0;
314 if (name)
315 strncpy(r->name, name, sizeof(r->name) - 1);
316 early_res_count++;
317}
318
319void __init free_early(u64 start, u64 end)
320{
321 struct early_res *r;
322 int i;
323
324 kmemleak_free_part(__va(start), end - start);
325
326 i = find_overlapped_early(start, end);
327 r = &early_res[i];
328 if (i >= max_early_res || r->end != end || r->start != start)
329 panic("free_early on not reserved area: %llx-%llx!",
330 start, end - 1);
331
332 drop_range(i);
333}
334
335void __init free_early_partial(u64 start, u64 end)
336{
337 struct early_res *r;
338 int i;
339
340 kmemleak_free_part(__va(start), end - start);
341
342 if (start == end)
343 return;
344
345 if (WARN_ONCE(start > end, " wrong range [%#llx, %#llx]\n", start, end))
346 return;
347
348try_next:
349 i = find_overlapped_early(start, end);
350 if (i >= max_early_res)
351 return;
352
353 r = &early_res[i];
354 /* hole ? */
355 if (r->end >= end && r->start <= start) {
356 drop_range_partial(i, start, end);
357 return;
358 }
359
360 drop_range_partial(i, start, end);
361 goto try_next;
362}
363
364#ifdef CONFIG_NO_BOOTMEM
365static void __init subtract_early_res(struct range *range, int az)
366{
367 int i, count;
368 u64 final_start, final_end;
369 int idx = 0;
370
371 count = 0;
372 for (i = 0; i < max_early_res && early_res[i].end; i++)
373 count++;
374
375 /* need to skip first one ?*/
376 if (early_res != early_res_x)
377 idx = 1;
378
379#define DEBUG_PRINT_EARLY_RES 1
380
381#if DEBUG_PRINT_EARLY_RES
382 printk(KERN_INFO "Subtract (%d early reservations)\n", count);
383#endif
384 for (i = idx; i < count; i++) {
385 struct early_res *r = &early_res[i];
386#if DEBUG_PRINT_EARLY_RES
387 printk(KERN_INFO " #%d [%010llx - %010llx] %15s\n", i,
388 r->start, r->end, r->name);
389#endif
390 final_start = PFN_DOWN(r->start);
391 final_end = PFN_UP(r->end);
392 if (final_start >= final_end)
393 continue;
394 subtract_range(range, az, final_start, final_end);
395 }
396
397}
398
399int __init get_free_all_memory_range(struct range **rangep, int nodeid)
400{
401 int i, count;
402 u64 start = 0, end;
403 u64 size;
404 u64 mem;
405 struct range *range;
406 int nr_range;
407
408 count = 0;
409 for (i = 0; i < max_early_res && early_res[i].end; i++)
410 count++;
411
412 count *= 2;
413
414 size = sizeof(struct range) * count;
415 end = get_max_mapped();
416#ifdef MAX_DMA32_PFN
417 if (end > (MAX_DMA32_PFN << PAGE_SHIFT))
418 start = MAX_DMA32_PFN << PAGE_SHIFT;
419#endif
420 mem = find_fw_memmap_area(start, end, size, sizeof(struct range));
421 if (mem == -1ULL)
422 panic("can not find more space for range free");
423
424 range = __va(mem);
425 /* use early_node_map[] and early_res to get range array at first */
426 memset(range, 0, size);
427 nr_range = 0;
428
429 /* need to go over early_node_map to find out good range for node */
430 nr_range = add_from_early_node_map(range, count, nr_range, nodeid);
431#ifdef CONFIG_X86_32
432 subtract_range(range, count, max_low_pfn, -1ULL);
433#endif
434 subtract_early_res(range, count);
435 nr_range = clean_sort_range(range, count);
436
437 /* need to clear it ? */
438 if (nodeid == MAX_NUMNODES) {
439 memset(&early_res[0], 0,
440 sizeof(struct early_res) * max_early_res);
441 early_res = NULL;
442 max_early_res = 0;
443 }
444
445 *rangep = range;
446 return nr_range;
447}
448#else
449void __init early_res_to_bootmem(u64 start, u64 end)
450{
451 int i, count;
452 u64 final_start, final_end;
453 int idx = 0;
454
455 count = 0;
456 for (i = 0; i < max_early_res && early_res[i].end; i++)
457 count++;
458
459 /* need to skip first one ?*/
460 if (early_res != early_res_x)
461 idx = 1;
462
463 printk(KERN_INFO "(%d/%d early reservations) ==> bootmem [%010llx - %010llx]\n",
464 count - idx, max_early_res, start, end);
465 for (i = idx; i < count; i++) {
466 struct early_res *r = &early_res[i];
467 printk(KERN_INFO " #%d [%010llx - %010llx] %16s", i,
468 r->start, r->end, r->name);
469 final_start = max(start, r->start);
470 final_end = min(end, r->end);
471 if (final_start >= final_end) {
472 printk(KERN_CONT "\n");
473 continue;
474 }
475 printk(KERN_CONT " ==> [%010llx - %010llx]\n",
476 final_start, final_end);
477 reserve_bootmem_generic(final_start, final_end - final_start,
478 BOOTMEM_DEFAULT);
479 }
480 /* clear them */
481 memset(&early_res[0], 0, sizeof(struct early_res) * max_early_res);
482 early_res = NULL;
483 max_early_res = 0;
484 early_res_count = 0;
485}
486#endif
487
488/* Check for already reserved areas */
489static inline int __init bad_addr(u64 *addrp, u64 size, u64 align)
490{
491 int i;
492 u64 addr = *addrp;
493 int changed = 0;
494 struct early_res *r;
495again:
496 i = find_overlapped_early(addr, addr + size);
497 r = &early_res[i];
498 if (i < max_early_res && r->end) {
499 *addrp = addr = round_up(r->end, align);
500 changed = 1;
501 goto again;
502 }
503 return changed;
504}
505
506/* Check for already reserved areas */
507static inline int __init bad_addr_size(u64 *addrp, u64 *sizep, u64 align)
508{
509 int i;
510 u64 addr = *addrp, last;
511 u64 size = *sizep;
512 int changed = 0;
513again:
514 last = addr + size;
515 for (i = 0; i < max_early_res && early_res[i].end; i++) {
516 struct early_res *r = &early_res[i];
517 if (last > r->start && addr < r->start) {
518 size = r->start - addr;
519 changed = 1;
520 goto again;
521 }
522 if (last > r->end && addr < r->end) {
523 addr = round_up(r->end, align);
524 size = last - addr;
525 changed = 1;
526 goto again;
527 }
528 if (last <= r->end && addr >= r->start) {
529 (*sizep)++;
530 return 0;
531 }
532 }
533 if (changed) {
534 *addrp = addr;
535 *sizep = size;
536 }
537 return changed;
538}
539
540/*
541 * Find a free area with specified alignment in a specific range.
542 * only with the area.between start to end is active range from early_node_map
543 * so they are good as RAM
544 */
545u64 __init find_early_area(u64 ei_start, u64 ei_last, u64 start, u64 end,
546 u64 size, u64 align)
547{
548 u64 addr, last;
549
550 addr = round_up(ei_start, align);
551 if (addr < start)
552 addr = round_up(start, align);
553 if (addr >= ei_last)
554 goto out;
555 while (bad_addr(&addr, size, align) && addr+size <= ei_last)
556 ;
557 last = addr + size;
558 if (last > ei_last)
559 goto out;
560 if (last > end)
561 goto out;
562
563 return addr;
564
565out:
566 return -1ULL;
567}
568
569u64 __init find_early_area_size(u64 ei_start, u64 ei_last, u64 start,
570 u64 *sizep, u64 align)
571{
572 u64 addr, last;
573
574 addr = round_up(ei_start, align);
575 if (addr < start)
576 addr = round_up(start, align);
577 if (addr >= ei_last)
578 goto out;
579 *sizep = ei_last - addr;
580 while (bad_addr_size(&addr, sizep, align) && addr + *sizep <= ei_last)
581 ;
582 last = addr + *sizep;
583 if (last > ei_last)
584 goto out;
585
586 return addr;
587
588out:
589 return -1ULL;
590}
diff --git a/kernel/exit.c b/kernel/exit.c
index 03120229db28..e2bdf37f9fde 100644
--- a/kernel/exit.c
+++ b/kernel/exit.c
@@ -149,9 +149,7 @@ static void delayed_put_task_struct(struct rcu_head *rhp)
149{ 149{
150 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu); 150 struct task_struct *tsk = container_of(rhp, struct task_struct, rcu);
151 151
152#ifdef CONFIG_PERF_EVENTS 152 perf_event_delayed_put(tsk);
153 WARN_ON_ONCE(tsk->perf_event_ctxp);
154#endif
155 trace_sched_process_free(tsk); 153 trace_sched_process_free(tsk);
156 put_task_struct(tsk); 154 put_task_struct(tsk);
157} 155}
diff --git a/kernel/futex.c b/kernel/futex.c
index 6a3a5fa1526d..a118bf160e0b 100644
--- a/kernel/futex.c
+++ b/kernel/futex.c
@@ -91,6 +91,7 @@ struct futex_pi_state {
91 91
92/** 92/**
93 * struct futex_q - The hashed futex queue entry, one per waiting task 93 * struct futex_q - The hashed futex queue entry, one per waiting task
94 * @list: priority-sorted list of tasks waiting on this futex
94 * @task: the task waiting on the futex 95 * @task: the task waiting on the futex
95 * @lock_ptr: the hash bucket lock 96 * @lock_ptr: the hash bucket lock
96 * @key: the key the futex is hashed on 97 * @key: the key the futex is hashed on
@@ -104,7 +105,7 @@ struct futex_pi_state {
104 * 105 *
105 * A futex_q has a woken state, just like tasks have TASK_RUNNING. 106 * A futex_q has a woken state, just like tasks have TASK_RUNNING.
106 * It is considered woken when plist_node_empty(&q->list) || q->lock_ptr == 0. 107 * It is considered woken when plist_node_empty(&q->list) || q->lock_ptr == 0.
107 * The order of wakup is always to make the first condition true, then 108 * The order of wakeup is always to make the first condition true, then
108 * the second. 109 * the second.
109 * 110 *
110 * PI futexes are typically woken before they are removed from the hash list via 111 * PI futexes are typically woken before they are removed from the hash list via
@@ -295,7 +296,7 @@ void put_futex_key(int fshared, union futex_key *key)
295 * Slow path to fixup the fault we just took in the atomic write 296 * Slow path to fixup the fault we just took in the atomic write
296 * access to @uaddr. 297 * access to @uaddr.
297 * 298 *
298 * We have no generic implementation of a non destructive write to the 299 * We have no generic implementation of a non-destructive write to the
299 * user address. We know that we faulted in the atomic pagefault 300 * user address. We know that we faulted in the atomic pagefault
300 * disabled section so we can as well avoid the #PF overhead by 301 * disabled section so we can as well avoid the #PF overhead by
301 * calling get_user_pages() right away. 302 * calling get_user_pages() right away.
@@ -515,7 +516,7 @@ lookup_pi_state(u32 uval, struct futex_hash_bucket *hb,
515 */ 516 */
516 pi_state = this->pi_state; 517 pi_state = this->pi_state;
517 /* 518 /*
518 * Userspace might have messed up non PI and PI futexes 519 * Userspace might have messed up non-PI and PI futexes
519 */ 520 */
520 if (unlikely(!pi_state)) 521 if (unlikely(!pi_state))
521 return -EINVAL; 522 return -EINVAL;
@@ -736,8 +737,8 @@ static void wake_futex(struct futex_q *q)
736 737
737 /* 738 /*
738 * We set q->lock_ptr = NULL _before_ we wake up the task. If 739 * We set q->lock_ptr = NULL _before_ we wake up the task. If
739 * a non futex wake up happens on another CPU then the task 740 * a non-futex wake up happens on another CPU then the task
740 * might exit and p would dereference a non existing task 741 * might exit and p would dereference a non-existing task
741 * struct. Prevent this by holding a reference on p across the 742 * struct. Prevent this by holding a reference on p across the
742 * wake up. 743 * wake up.
743 */ 744 */
@@ -1131,11 +1132,13 @@ static int futex_proxy_trylock_atomic(u32 __user *pifutex,
1131 1132
1132/** 1133/**
1133 * futex_requeue() - Requeue waiters from uaddr1 to uaddr2 1134 * futex_requeue() - Requeue waiters from uaddr1 to uaddr2
1134 * uaddr1: source futex user address 1135 * @uaddr1: source futex user address
1135 * uaddr2: target futex user address 1136 * @fshared: 0 for a PROCESS_PRIVATE futex, 1 for PROCESS_SHARED
1136 * nr_wake: number of waiters to wake (must be 1 for requeue_pi) 1137 * @uaddr2: target futex user address
1137 * nr_requeue: number of waiters to requeue (0-INT_MAX) 1138 * @nr_wake: number of waiters to wake (must be 1 for requeue_pi)
1138 * requeue_pi: if we are attempting to requeue from a non-pi futex to a 1139 * @nr_requeue: number of waiters to requeue (0-INT_MAX)
1140 * @cmpval: @uaddr1 expected value (or %NULL)
1141 * @requeue_pi: if we are attempting to requeue from a non-pi futex to a
1139 * pi futex (pi to pi requeue is not supported) 1142 * pi futex (pi to pi requeue is not supported)
1140 * 1143 *
1141 * Requeue waiters on uaddr1 to uaddr2. In the requeue_pi case, try to acquire 1144 * Requeue waiters on uaddr1 to uaddr2. In the requeue_pi case, try to acquire
@@ -1360,10 +1363,10 @@ out:
1360 1363
1361/* The key must be already stored in q->key. */ 1364/* The key must be already stored in q->key. */
1362static inline struct futex_hash_bucket *queue_lock(struct futex_q *q) 1365static inline struct futex_hash_bucket *queue_lock(struct futex_q *q)
1366 __acquires(&hb->lock)
1363{ 1367{
1364 struct futex_hash_bucket *hb; 1368 struct futex_hash_bucket *hb;
1365 1369
1366 get_futex_key_refs(&q->key);
1367 hb = hash_futex(&q->key); 1370 hb = hash_futex(&q->key);
1368 q->lock_ptr = &hb->lock; 1371 q->lock_ptr = &hb->lock;
1369 1372
@@ -1373,9 +1376,9 @@ static inline struct futex_hash_bucket *queue_lock(struct futex_q *q)
1373 1376
1374static inline void 1377static inline void
1375queue_unlock(struct futex_q *q, struct futex_hash_bucket *hb) 1378queue_unlock(struct futex_q *q, struct futex_hash_bucket *hb)
1379 __releases(&hb->lock)
1376{ 1380{
1377 spin_unlock(&hb->lock); 1381 spin_unlock(&hb->lock);
1378 drop_futex_key_refs(&q->key);
1379} 1382}
1380 1383
1381/** 1384/**
@@ -1391,6 +1394,7 @@ queue_unlock(struct futex_q *q, struct futex_hash_bucket *hb)
1391 * an example). 1394 * an example).
1392 */ 1395 */
1393static inline void queue_me(struct futex_q *q, struct futex_hash_bucket *hb) 1396static inline void queue_me(struct futex_q *q, struct futex_hash_bucket *hb)
1397 __releases(&hb->lock)
1394{ 1398{
1395 int prio; 1399 int prio;
1396 1400
@@ -1471,6 +1475,7 @@ retry:
1471 * and dropped here. 1475 * and dropped here.
1472 */ 1476 */
1473static void unqueue_me_pi(struct futex_q *q) 1477static void unqueue_me_pi(struct futex_q *q)
1478 __releases(q->lock_ptr)
1474{ 1479{
1475 WARN_ON(plist_node_empty(&q->list)); 1480 WARN_ON(plist_node_empty(&q->list));
1476 plist_del(&q->list, &q->list.plist); 1481 plist_del(&q->list, &q->list.plist);
@@ -1480,8 +1485,6 @@ static void unqueue_me_pi(struct futex_q *q)
1480 q->pi_state = NULL; 1485 q->pi_state = NULL;
1481 1486
1482 spin_unlock(q->lock_ptr); 1487 spin_unlock(q->lock_ptr);
1483
1484 drop_futex_key_refs(&q->key);
1485} 1488}
1486 1489
1487/* 1490/*
@@ -1812,7 +1815,10 @@ static int futex_wait(u32 __user *uaddr, int fshared,
1812 } 1815 }
1813 1816
1814retry: 1817retry:
1815 /* Prepare to wait on uaddr. */ 1818 /*
1819 * Prepare to wait on uaddr. On success, holds hb lock and increments
1820 * q.key refs.
1821 */
1816 ret = futex_wait_setup(uaddr, val, fshared, &q, &hb); 1822 ret = futex_wait_setup(uaddr, val, fshared, &q, &hb);
1817 if (ret) 1823 if (ret)
1818 goto out; 1824 goto out;
@@ -1822,28 +1828,27 @@ retry:
1822 1828
1823 /* If we were woken (and unqueued), we succeeded, whatever. */ 1829 /* If we were woken (and unqueued), we succeeded, whatever. */
1824 ret = 0; 1830 ret = 0;
1831 /* unqueue_me() drops q.key ref */
1825 if (!unqueue_me(&q)) 1832 if (!unqueue_me(&q))
1826 goto out_put_key; 1833 goto out;
1827 ret = -ETIMEDOUT; 1834 ret = -ETIMEDOUT;
1828 if (to && !to->task) 1835 if (to && !to->task)
1829 goto out_put_key; 1836 goto out;
1830 1837
1831 /* 1838 /*
1832 * We expect signal_pending(current), but we might be the 1839 * We expect signal_pending(current), but we might be the
1833 * victim of a spurious wakeup as well. 1840 * victim of a spurious wakeup as well.
1834 */ 1841 */
1835 if (!signal_pending(current)) { 1842 if (!signal_pending(current))
1836 put_futex_key(fshared, &q.key);
1837 goto retry; 1843 goto retry;
1838 }
1839 1844
1840 ret = -ERESTARTSYS; 1845 ret = -ERESTARTSYS;
1841 if (!abs_time) 1846 if (!abs_time)
1842 goto out_put_key; 1847 goto out;
1843 1848
1844 restart = &current_thread_info()->restart_block; 1849 restart = &current_thread_info()->restart_block;
1845 restart->fn = futex_wait_restart; 1850 restart->fn = futex_wait_restart;
1846 restart->futex.uaddr = (u32 *)uaddr; 1851 restart->futex.uaddr = uaddr;
1847 restart->futex.val = val; 1852 restart->futex.val = val;
1848 restart->futex.time = abs_time->tv64; 1853 restart->futex.time = abs_time->tv64;
1849 restart->futex.bitset = bitset; 1854 restart->futex.bitset = bitset;
@@ -1856,8 +1861,6 @@ retry:
1856 1861
1857 ret = -ERESTART_RESTARTBLOCK; 1862 ret = -ERESTART_RESTARTBLOCK;
1858 1863
1859out_put_key:
1860 put_futex_key(fshared, &q.key);
1861out: 1864out:
1862 if (to) { 1865 if (to) {
1863 hrtimer_cancel(&to->timer); 1866 hrtimer_cancel(&to->timer);
@@ -1869,7 +1872,7 @@ out:
1869 1872
1870static long futex_wait_restart(struct restart_block *restart) 1873static long futex_wait_restart(struct restart_block *restart)
1871{ 1874{
1872 u32 __user *uaddr = (u32 __user *)restart->futex.uaddr; 1875 u32 __user *uaddr = restart->futex.uaddr;
1873 int fshared = 0; 1876 int fshared = 0;
1874 ktime_t t, *tp = NULL; 1877 ktime_t t, *tp = NULL;
1875 1878
@@ -2236,7 +2239,10 @@ static int futex_wait_requeue_pi(u32 __user *uaddr, int fshared,
2236 q.rt_waiter = &rt_waiter; 2239 q.rt_waiter = &rt_waiter;
2237 q.requeue_pi_key = &key2; 2240 q.requeue_pi_key = &key2;
2238 2241
2239 /* Prepare to wait on uaddr. */ 2242 /*
2243 * Prepare to wait on uaddr. On success, increments q.key (key1) ref
2244 * count.
2245 */
2240 ret = futex_wait_setup(uaddr, val, fshared, &q, &hb); 2246 ret = futex_wait_setup(uaddr, val, fshared, &q, &hb);
2241 if (ret) 2247 if (ret)
2242 goto out_key2; 2248 goto out_key2;
@@ -2254,7 +2260,9 @@ static int futex_wait_requeue_pi(u32 __user *uaddr, int fshared,
2254 * In order for us to be here, we know our q.key == key2, and since 2260 * In order for us to be here, we know our q.key == key2, and since
2255 * we took the hb->lock above, we also know that futex_requeue() has 2261 * we took the hb->lock above, we also know that futex_requeue() has
2256 * completed and we no longer have to concern ourselves with a wakeup 2262 * completed and we no longer have to concern ourselves with a wakeup
2257 * race with the atomic proxy lock acquition by the requeue code. 2263 * race with the atomic proxy lock acquisition by the requeue code. The
2264 * futex_requeue dropped our key1 reference and incremented our key2
2265 * reference count.
2258 */ 2266 */
2259 2267
2260 /* Check if the requeue code acquired the second futex for us. */ 2268 /* Check if the requeue code acquired the second futex for us. */
@@ -2458,7 +2466,7 @@ retry:
2458 */ 2466 */
2459static inline int fetch_robust_entry(struct robust_list __user **entry, 2467static inline int fetch_robust_entry(struct robust_list __user **entry,
2460 struct robust_list __user * __user *head, 2468 struct robust_list __user * __user *head,
2461 int *pi) 2469 unsigned int *pi)
2462{ 2470{
2463 unsigned long uentry; 2471 unsigned long uentry;
2464 2472
@@ -2647,7 +2655,7 @@ static int __init futex_init(void)
2647 * of the complex code paths. Also we want to prevent 2655 * of the complex code paths. Also we want to prevent
2648 * registration of robust lists in that case. NULL is 2656 * registration of robust lists in that case. NULL is
2649 * guaranteed to fault and we get -EFAULT on functional 2657 * guaranteed to fault and we get -EFAULT on functional
2650 * implementation, the non functional ones will return 2658 * implementation, the non-functional ones will return
2651 * -ENOSYS. 2659 * -ENOSYS.
2652 */ 2660 */
2653 curval = cmpxchg_futex_value_locked(NULL, 0, 0); 2661 curval = cmpxchg_futex_value_locked(NULL, 0, 0);
diff --git a/kernel/futex_compat.c b/kernel/futex_compat.c
index d49afb2395e5..06da4dfc339b 100644
--- a/kernel/futex_compat.c
+++ b/kernel/futex_compat.c
@@ -19,7 +19,7 @@
19 */ 19 */
20static inline int 20static inline int
21fetch_robust_entry(compat_uptr_t *uentry, struct robust_list __user **entry, 21fetch_robust_entry(compat_uptr_t *uentry, struct robust_list __user **entry,
22 compat_uptr_t __user *head, int *pi) 22 compat_uptr_t __user *head, unsigned int *pi)
23{ 23{
24 if (get_user(*uentry, head)) 24 if (get_user(*uentry, head))
25 return -EFAULT; 25 return -EFAULT;
diff --git a/kernel/gcov/fs.c b/kernel/gcov/fs.c
index f83972b16564..9bd0934f6c33 100644
--- a/kernel/gcov/fs.c
+++ b/kernel/gcov/fs.c
@@ -561,6 +561,7 @@ static ssize_t reset_read(struct file *file, char __user *addr, size_t len,
561static const struct file_operations gcov_reset_fops = { 561static const struct file_operations gcov_reset_fops = {
562 .write = reset_write, 562 .write = reset_write,
563 .read = reset_read, 563 .read = reset_read,
564 .llseek = noop_llseek,
564}; 565};
565 566
566/* 567/*
diff --git a/kernel/hrtimer.c b/kernel/hrtimer.c
index 1decafbb6b1a..72206cf5c6cf 100644
--- a/kernel/hrtimer.c
+++ b/kernel/hrtimer.c
@@ -931,6 +931,7 @@ static inline int
931remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base) 931remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base)
932{ 932{
933 if (hrtimer_is_queued(timer)) { 933 if (hrtimer_is_queued(timer)) {
934 unsigned long state;
934 int reprogram; 935 int reprogram;
935 936
936 /* 937 /*
@@ -944,8 +945,13 @@ remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base)
944 debug_deactivate(timer); 945 debug_deactivate(timer);
945 timer_stats_hrtimer_clear_start_info(timer); 946 timer_stats_hrtimer_clear_start_info(timer);
946 reprogram = base->cpu_base == &__get_cpu_var(hrtimer_bases); 947 reprogram = base->cpu_base == &__get_cpu_var(hrtimer_bases);
947 __remove_hrtimer(timer, base, HRTIMER_STATE_INACTIVE, 948 /*
948 reprogram); 949 * We must preserve the CALLBACK state flag here,
950 * otherwise we could move the timer base in
951 * switch_hrtimer_base.
952 */
953 state = timer->state & HRTIMER_STATE_CALLBACK;
954 __remove_hrtimer(timer, base, state, reprogram);
949 return 1; 955 return 1;
950 } 956 }
951 return 0; 957 return 0;
@@ -1231,6 +1237,9 @@ static void __run_hrtimer(struct hrtimer *timer, ktime_t *now)
1231 BUG_ON(timer->state != HRTIMER_STATE_CALLBACK); 1237 BUG_ON(timer->state != HRTIMER_STATE_CALLBACK);
1232 enqueue_hrtimer(timer, base); 1238 enqueue_hrtimer(timer, base);
1233 } 1239 }
1240
1241 WARN_ON_ONCE(!(timer->state & HRTIMER_STATE_CALLBACK));
1242
1234 timer->state &= ~HRTIMER_STATE_CALLBACK; 1243 timer->state &= ~HRTIMER_STATE_CALLBACK;
1235} 1244}
1236 1245
diff --git a/kernel/hung_task.c b/kernel/hung_task.c
index 0c642d51aac2..53ead174da2f 100644
--- a/kernel/hung_task.c
+++ b/kernel/hung_task.c
@@ -98,7 +98,7 @@ static void check_hung_task(struct task_struct *t, unsigned long timeout)
98 printk(KERN_ERR "\"echo 0 > /proc/sys/kernel/hung_task_timeout_secs\"" 98 printk(KERN_ERR "\"echo 0 > /proc/sys/kernel/hung_task_timeout_secs\""
99 " disables this message.\n"); 99 " disables this message.\n");
100 sched_show_task(t); 100 sched_show_task(t);
101 __debug_show_held_locks(t); 101 debug_show_held_locks(t);
102 102
103 touch_nmi_watchdog(); 103 touch_nmi_watchdog();
104 104
@@ -111,7 +111,7 @@ static void check_hung_task(struct task_struct *t, unsigned long timeout)
111 * periodically exit the critical section and enter a new one. 111 * periodically exit the critical section and enter a new one.
112 * 112 *
113 * For preemptible RCU it is sufficient to call rcu_read_unlock in order 113 * For preemptible RCU it is sufficient to call rcu_read_unlock in order
114 * exit the grace period. For classic RCU, a reschedule is required. 114 * to exit the grace period. For classic RCU, a reschedule is required.
115 */ 115 */
116static void rcu_lock_break(struct task_struct *g, struct task_struct *t) 116static void rcu_lock_break(struct task_struct *g, struct task_struct *t)
117{ 117{
diff --git a/kernel/hw_breakpoint.c b/kernel/hw_breakpoint.c
index c7c2aed9e2dc..2c9120f0afca 100644
--- a/kernel/hw_breakpoint.c
+++ b/kernel/hw_breakpoint.c
@@ -113,12 +113,12 @@ static unsigned int max_task_bp_pinned(int cpu, enum bp_type_idx type)
113 */ 113 */
114static int task_bp_pinned(struct perf_event *bp, enum bp_type_idx type) 114static int task_bp_pinned(struct perf_event *bp, enum bp_type_idx type)
115{ 115{
116 struct perf_event_context *ctx = bp->ctx; 116 struct task_struct *tsk = bp->hw.bp_target;
117 struct perf_event *iter; 117 struct perf_event *iter;
118 int count = 0; 118 int count = 0;
119 119
120 list_for_each_entry(iter, &bp_task_head, hw.bp_list) { 120 list_for_each_entry(iter, &bp_task_head, hw.bp_list) {
121 if (iter->ctx == ctx && find_slot_idx(iter) == type) 121 if (iter->hw.bp_target == tsk && find_slot_idx(iter) == type)
122 count += hw_breakpoint_weight(iter); 122 count += hw_breakpoint_weight(iter);
123 } 123 }
124 124
@@ -134,7 +134,7 @@ fetch_bp_busy_slots(struct bp_busy_slots *slots, struct perf_event *bp,
134 enum bp_type_idx type) 134 enum bp_type_idx type)
135{ 135{
136 int cpu = bp->cpu; 136 int cpu = bp->cpu;
137 struct task_struct *tsk = bp->ctx->task; 137 struct task_struct *tsk = bp->hw.bp_target;
138 138
139 if (cpu >= 0) { 139 if (cpu >= 0) {
140 slots->pinned = per_cpu(nr_cpu_bp_pinned[type], cpu); 140 slots->pinned = per_cpu(nr_cpu_bp_pinned[type], cpu);
@@ -213,7 +213,7 @@ toggle_bp_slot(struct perf_event *bp, bool enable, enum bp_type_idx type,
213 int weight) 213 int weight)
214{ 214{
215 int cpu = bp->cpu; 215 int cpu = bp->cpu;
216 struct task_struct *tsk = bp->ctx->task; 216 struct task_struct *tsk = bp->hw.bp_target;
217 217
218 /* Pinned counter cpu profiling */ 218 /* Pinned counter cpu profiling */
219 if (!tsk) { 219 if (!tsk) {
@@ -433,8 +433,7 @@ register_user_hw_breakpoint(struct perf_event_attr *attr,
433 perf_overflow_handler_t triggered, 433 perf_overflow_handler_t triggered,
434 struct task_struct *tsk) 434 struct task_struct *tsk)
435{ 435{
436 return perf_event_create_kernel_counter(attr, -1, task_pid_vnr(tsk), 436 return perf_event_create_kernel_counter(attr, -1, tsk, triggered);
437 triggered);
438} 437}
439EXPORT_SYMBOL_GPL(register_user_hw_breakpoint); 438EXPORT_SYMBOL_GPL(register_user_hw_breakpoint);
440 439
@@ -516,7 +515,7 @@ register_wide_hw_breakpoint(struct perf_event_attr *attr,
516 get_online_cpus(); 515 get_online_cpus();
517 for_each_online_cpu(cpu) { 516 for_each_online_cpu(cpu) {
518 pevent = per_cpu_ptr(cpu_events, cpu); 517 pevent = per_cpu_ptr(cpu_events, cpu);
519 bp = perf_event_create_kernel_counter(attr, cpu, -1, triggered); 518 bp = perf_event_create_kernel_counter(attr, cpu, NULL, triggered);
520 519
521 *pevent = bp; 520 *pevent = bp;
522 521
@@ -566,6 +565,61 @@ static struct notifier_block hw_breakpoint_exceptions_nb = {
566 .priority = 0x7fffffff 565 .priority = 0x7fffffff
567}; 566};
568 567
568static void bp_perf_event_destroy(struct perf_event *event)
569{
570 release_bp_slot(event);
571}
572
573static int hw_breakpoint_event_init(struct perf_event *bp)
574{
575 int err;
576
577 if (bp->attr.type != PERF_TYPE_BREAKPOINT)
578 return -ENOENT;
579
580 err = register_perf_hw_breakpoint(bp);
581 if (err)
582 return err;
583
584 bp->destroy = bp_perf_event_destroy;
585
586 return 0;
587}
588
589static int hw_breakpoint_add(struct perf_event *bp, int flags)
590{
591 if (!(flags & PERF_EF_START))
592 bp->hw.state = PERF_HES_STOPPED;
593
594 return arch_install_hw_breakpoint(bp);
595}
596
597static void hw_breakpoint_del(struct perf_event *bp, int flags)
598{
599 arch_uninstall_hw_breakpoint(bp);
600}
601
602static void hw_breakpoint_start(struct perf_event *bp, int flags)
603{
604 bp->hw.state = 0;
605}
606
607static void hw_breakpoint_stop(struct perf_event *bp, int flags)
608{
609 bp->hw.state = PERF_HES_STOPPED;
610}
611
612static struct pmu perf_breakpoint = {
613 .task_ctx_nr = perf_sw_context, /* could eventually get its own */
614
615 .event_init = hw_breakpoint_event_init,
616 .add = hw_breakpoint_add,
617 .del = hw_breakpoint_del,
618 .start = hw_breakpoint_start,
619 .stop = hw_breakpoint_stop,
620 .read = hw_breakpoint_pmu_read,
621};
622
569static int __init init_hw_breakpoint(void) 623static int __init init_hw_breakpoint(void)
570{ 624{
571 unsigned int **task_bp_pinned; 625 unsigned int **task_bp_pinned;
@@ -587,6 +641,8 @@ static int __init init_hw_breakpoint(void)
587 641
588 constraints_initialized = 1; 642 constraints_initialized = 1;
589 643
644 perf_pmu_register(&perf_breakpoint);
645
590 return register_die_notifier(&hw_breakpoint_exceptions_nb); 646 return register_die_notifier(&hw_breakpoint_exceptions_nb);
591 647
592 err_alloc: 648 err_alloc:
@@ -602,8 +658,3 @@ static int __init init_hw_breakpoint(void)
602core_initcall(init_hw_breakpoint); 658core_initcall(init_hw_breakpoint);
603 659
604 660
605struct pmu perf_ops_bp = {
606 .enable = arch_install_hw_breakpoint,
607 .disable = arch_uninstall_hw_breakpoint,
608 .read = hw_breakpoint_pmu_read,
609};
diff --git a/kernel/irq/Kconfig b/kernel/irq/Kconfig
new file mode 100644
index 000000000000..31d766bf5d2e
--- /dev/null
+++ b/kernel/irq/Kconfig
@@ -0,0 +1,53 @@
1config HAVE_GENERIC_HARDIRQS
2 def_bool n
3
4if HAVE_GENERIC_HARDIRQS
5menu "IRQ subsystem"
6#
7# Interrupt subsystem related configuration options
8#
9config GENERIC_HARDIRQS
10 def_bool y
11
12config GENERIC_HARDIRQS_NO__DO_IRQ
13 def_bool y
14
15# Select this to disable the deprecated stuff
16config GENERIC_HARDIRQS_NO_DEPRECATED
17 def_bool n
18
19# Options selectable by the architecture code
20config HAVE_SPARSE_IRQ
21 def_bool n
22
23config GENERIC_IRQ_PROBE
24 def_bool n
25
26config GENERIC_PENDING_IRQ
27 def_bool n
28
29config AUTO_IRQ_AFFINITY
30 def_bool n
31
32config IRQ_PER_CPU
33 def_bool n
34
35config HARDIRQS_SW_RESEND
36 def_bool n
37
38config SPARSE_IRQ
39 bool "Support sparse irq numbering"
40 depends on HAVE_SPARSE_IRQ
41 ---help---
42
43 Sparse irq numbering is useful for distro kernels that want
44 to define a high CONFIG_NR_CPUS value but still want to have
45 low kernel memory footprint on smaller machines.
46
47 ( Sparse irqs can also be beneficial on NUMA boxes, as they spread
48 out the interrupt descriptors in a more NUMA-friendly way. )
49
50 If you don't know what to do here, say N.
51
52endmenu
53endif
diff --git a/kernel/irq/Makefile b/kernel/irq/Makefile
index 7d047808419d..54329cd7b3ee 100644
--- a/kernel/irq/Makefile
+++ b/kernel/irq/Makefile
@@ -1,7 +1,6 @@
1 1
2obj-y := handle.o manage.o spurious.o resend.o chip.o devres.o 2obj-y := irqdesc.o handle.o manage.o spurious.o resend.o chip.o dummychip.o devres.o
3obj-$(CONFIG_GENERIC_IRQ_PROBE) += autoprobe.o 3obj-$(CONFIG_GENERIC_IRQ_PROBE) += autoprobe.o
4obj-$(CONFIG_PROC_FS) += proc.o 4obj-$(CONFIG_PROC_FS) += proc.o
5obj-$(CONFIG_GENERIC_PENDING_IRQ) += migration.o 5obj-$(CONFIG_GENERIC_PENDING_IRQ) += migration.o
6obj-$(CONFIG_NUMA_IRQ_DESC) += numa_migrate.o
7obj-$(CONFIG_PM_SLEEP) += pm.o 6obj-$(CONFIG_PM_SLEEP) += pm.o
diff --git a/kernel/irq/autoprobe.c b/kernel/irq/autoprobe.c
index 2295a31ef110..505798f86c36 100644
--- a/kernel/irq/autoprobe.c
+++ b/kernel/irq/autoprobe.c
@@ -57,9 +57,10 @@ unsigned long probe_irq_on(void)
57 * Some chips need to know about probing in 57 * Some chips need to know about probing in
58 * progress: 58 * progress:
59 */ 59 */
60 if (desc->chip->set_type) 60 if (desc->irq_data.chip->irq_set_type)
61 desc->chip->set_type(i, IRQ_TYPE_PROBE); 61 desc->irq_data.chip->irq_set_type(&desc->irq_data,
62 desc->chip->startup(i); 62 IRQ_TYPE_PROBE);
63 desc->irq_data.chip->irq_startup(&desc->irq_data);
63 } 64 }
64 raw_spin_unlock_irq(&desc->lock); 65 raw_spin_unlock_irq(&desc->lock);
65 } 66 }
@@ -76,7 +77,7 @@ unsigned long probe_irq_on(void)
76 raw_spin_lock_irq(&desc->lock); 77 raw_spin_lock_irq(&desc->lock);
77 if (!desc->action && !(desc->status & IRQ_NOPROBE)) { 78 if (!desc->action && !(desc->status & IRQ_NOPROBE)) {
78 desc->status |= IRQ_AUTODETECT | IRQ_WAITING; 79 desc->status |= IRQ_AUTODETECT | IRQ_WAITING;
79 if (desc->chip->startup(i)) 80 if (desc->irq_data.chip->irq_startup(&desc->irq_data))
80 desc->status |= IRQ_PENDING; 81 desc->status |= IRQ_PENDING;
81 } 82 }
82 raw_spin_unlock_irq(&desc->lock); 83 raw_spin_unlock_irq(&desc->lock);
@@ -98,7 +99,7 @@ unsigned long probe_irq_on(void)
98 /* It triggered already - consider it spurious. */ 99 /* It triggered already - consider it spurious. */
99 if (!(status & IRQ_WAITING)) { 100 if (!(status & IRQ_WAITING)) {
100 desc->status = status & ~IRQ_AUTODETECT; 101 desc->status = status & ~IRQ_AUTODETECT;
101 desc->chip->shutdown(i); 102 desc->irq_data.chip->irq_shutdown(&desc->irq_data);
102 } else 103 } else
103 if (i < 32) 104 if (i < 32)
104 mask |= 1 << i; 105 mask |= 1 << i;
@@ -137,7 +138,7 @@ unsigned int probe_irq_mask(unsigned long val)
137 mask |= 1 << i; 138 mask |= 1 << i;
138 139
139 desc->status = status & ~IRQ_AUTODETECT; 140 desc->status = status & ~IRQ_AUTODETECT;
140 desc->chip->shutdown(i); 141 desc->irq_data.chip->irq_shutdown(&desc->irq_data);
141 } 142 }
142 raw_spin_unlock_irq(&desc->lock); 143 raw_spin_unlock_irq(&desc->lock);
143 } 144 }
@@ -181,7 +182,7 @@ int probe_irq_off(unsigned long val)
181 nr_of_irqs++; 182 nr_of_irqs++;
182 } 183 }
183 desc->status = status & ~IRQ_AUTODETECT; 184 desc->status = status & ~IRQ_AUTODETECT;
184 desc->chip->shutdown(i); 185 desc->irq_data.chip->irq_shutdown(&desc->irq_data);
185 } 186 }
186 raw_spin_unlock_irq(&desc->lock); 187 raw_spin_unlock_irq(&desc->lock);
187 } 188 }
diff --git a/kernel/irq/chip.c b/kernel/irq/chip.c
index b7091d5ca2f8..baa5c4acad83 100644
--- a/kernel/irq/chip.c
+++ b/kernel/irq/chip.c
@@ -18,108 +18,6 @@
18 18
19#include "internals.h" 19#include "internals.h"
20 20
21static void dynamic_irq_init_x(unsigned int irq, bool keep_chip_data)
22{
23 struct irq_desc *desc;
24 unsigned long flags;
25
26 desc = irq_to_desc(irq);
27 if (!desc) {
28 WARN(1, KERN_ERR "Trying to initialize invalid IRQ%d\n", irq);
29 return;
30 }
31
32 /* Ensure we don't have left over values from a previous use of this irq */
33 raw_spin_lock_irqsave(&desc->lock, flags);
34 desc->status = IRQ_DISABLED;
35 desc->chip = &no_irq_chip;
36 desc->handle_irq = handle_bad_irq;
37 desc->depth = 1;
38 desc->msi_desc = NULL;
39 desc->handler_data = NULL;
40 if (!keep_chip_data)
41 desc->chip_data = NULL;
42 desc->action = NULL;
43 desc->irq_count = 0;
44 desc->irqs_unhandled = 0;
45#ifdef CONFIG_SMP
46 cpumask_setall(desc->affinity);
47#ifdef CONFIG_GENERIC_PENDING_IRQ
48 cpumask_clear(desc->pending_mask);
49#endif
50#endif
51 raw_spin_unlock_irqrestore(&desc->lock, flags);
52}
53
54/**
55 * dynamic_irq_init - initialize a dynamically allocated irq
56 * @irq: irq number to initialize
57 */
58void dynamic_irq_init(unsigned int irq)
59{
60 dynamic_irq_init_x(irq, false);
61}
62
63/**
64 * dynamic_irq_init_keep_chip_data - initialize a dynamically allocated irq
65 * @irq: irq number to initialize
66 *
67 * does not set irq_to_desc(irq)->chip_data to NULL
68 */
69void dynamic_irq_init_keep_chip_data(unsigned int irq)
70{
71 dynamic_irq_init_x(irq, true);
72}
73
74static void dynamic_irq_cleanup_x(unsigned int irq, bool keep_chip_data)
75{
76 struct irq_desc *desc = irq_to_desc(irq);
77 unsigned long flags;
78
79 if (!desc) {
80 WARN(1, KERN_ERR "Trying to cleanup invalid IRQ%d\n", irq);
81 return;
82 }
83
84 raw_spin_lock_irqsave(&desc->lock, flags);
85 if (desc->action) {
86 raw_spin_unlock_irqrestore(&desc->lock, flags);
87 WARN(1, KERN_ERR "Destroying IRQ%d without calling free_irq\n",
88 irq);
89 return;
90 }
91 desc->msi_desc = NULL;
92 desc->handler_data = NULL;
93 if (!keep_chip_data)
94 desc->chip_data = NULL;
95 desc->handle_irq = handle_bad_irq;
96 desc->chip = &no_irq_chip;
97 desc->name = NULL;
98 clear_kstat_irqs(desc);
99 raw_spin_unlock_irqrestore(&desc->lock, flags);
100}
101
102/**
103 * dynamic_irq_cleanup - cleanup a dynamically allocated irq
104 * @irq: irq number to initialize
105 */
106void dynamic_irq_cleanup(unsigned int irq)
107{
108 dynamic_irq_cleanup_x(irq, false);
109}
110
111/**
112 * dynamic_irq_cleanup_keep_chip_data - cleanup a dynamically allocated irq
113 * @irq: irq number to initialize
114 *
115 * does not set irq_to_desc(irq)->chip_data to NULL
116 */
117void dynamic_irq_cleanup_keep_chip_data(unsigned int irq)
118{
119 dynamic_irq_cleanup_x(irq, true);
120}
121
122
123/** 21/**
124 * set_irq_chip - set the irq chip for an irq 22 * set_irq_chip - set the irq chip for an irq
125 * @irq: irq number 23 * @irq: irq number
@@ -140,7 +38,7 @@ int set_irq_chip(unsigned int irq, struct irq_chip *chip)
140 38
141 raw_spin_lock_irqsave(&desc->lock, flags); 39 raw_spin_lock_irqsave(&desc->lock, flags);
142 irq_chip_set_defaults(chip); 40 irq_chip_set_defaults(chip);
143 desc->chip = chip; 41 desc->irq_data.chip = chip;
144 raw_spin_unlock_irqrestore(&desc->lock, flags); 42 raw_spin_unlock_irqrestore(&desc->lock, flags);
145 43
146 return 0; 44 return 0;
@@ -193,7 +91,7 @@ int set_irq_data(unsigned int irq, void *data)
193 } 91 }
194 92
195 raw_spin_lock_irqsave(&desc->lock, flags); 93 raw_spin_lock_irqsave(&desc->lock, flags);
196 desc->handler_data = data; 94 desc->irq_data.handler_data = data;
197 raw_spin_unlock_irqrestore(&desc->lock, flags); 95 raw_spin_unlock_irqrestore(&desc->lock, flags);
198 return 0; 96 return 0;
199} 97}
@@ -218,7 +116,7 @@ int set_irq_msi(unsigned int irq, struct msi_desc *entry)
218 } 116 }
219 117
220 raw_spin_lock_irqsave(&desc->lock, flags); 118 raw_spin_lock_irqsave(&desc->lock, flags);
221 desc->msi_desc = entry; 119 desc->irq_data.msi_desc = entry;
222 if (entry) 120 if (entry)
223 entry->irq = irq; 121 entry->irq = irq;
224 raw_spin_unlock_irqrestore(&desc->lock, flags); 122 raw_spin_unlock_irqrestore(&desc->lock, flags);
@@ -243,19 +141,27 @@ int set_irq_chip_data(unsigned int irq, void *data)
243 return -EINVAL; 141 return -EINVAL;
244 } 142 }
245 143
246 if (!desc->chip) { 144 if (!desc->irq_data.chip) {
247 printk(KERN_ERR "BUG: bad set_irq_chip_data(IRQ#%d)\n", irq); 145 printk(KERN_ERR "BUG: bad set_irq_chip_data(IRQ#%d)\n", irq);
248 return -EINVAL; 146 return -EINVAL;
249 } 147 }
250 148
251 raw_spin_lock_irqsave(&desc->lock, flags); 149 raw_spin_lock_irqsave(&desc->lock, flags);
252 desc->chip_data = data; 150 desc->irq_data.chip_data = data;
253 raw_spin_unlock_irqrestore(&desc->lock, flags); 151 raw_spin_unlock_irqrestore(&desc->lock, flags);
254 152
255 return 0; 153 return 0;
256} 154}
257EXPORT_SYMBOL(set_irq_chip_data); 155EXPORT_SYMBOL(set_irq_chip_data);
258 156
157struct irq_data *irq_get_irq_data(unsigned int irq)
158{
159 struct irq_desc *desc = irq_to_desc(irq);
160
161 return desc ? &desc->irq_data : NULL;
162}
163EXPORT_SYMBOL_GPL(irq_get_irq_data);
164
259/** 165/**
260 * set_irq_nested_thread - Set/Reset the IRQ_NESTED_THREAD flag of an irq 166 * set_irq_nested_thread - Set/Reset the IRQ_NESTED_THREAD flag of an irq
261 * 167 *
@@ -287,93 +193,216 @@ EXPORT_SYMBOL_GPL(set_irq_nested_thread);
287/* 193/*
288 * default enable function 194 * default enable function
289 */ 195 */
290static void default_enable(unsigned int irq) 196static void default_enable(struct irq_data *data)
291{ 197{
292 struct irq_desc *desc = irq_to_desc(irq); 198 struct irq_desc *desc = irq_data_to_desc(data);
293 199
294 desc->chip->unmask(irq); 200 desc->irq_data.chip->irq_unmask(&desc->irq_data);
295 desc->status &= ~IRQ_MASKED; 201 desc->status &= ~IRQ_MASKED;
296} 202}
297 203
298/* 204/*
299 * default disable function 205 * default disable function
300 */ 206 */
301static void default_disable(unsigned int irq) 207static void default_disable(struct irq_data *data)
302{ 208{
303} 209}
304 210
305/* 211/*
306 * default startup function 212 * default startup function
307 */ 213 */
308static unsigned int default_startup(unsigned int irq) 214static unsigned int default_startup(struct irq_data *data)
309{ 215{
310 struct irq_desc *desc = irq_to_desc(irq); 216 struct irq_desc *desc = irq_data_to_desc(data);
311 217
312 desc->chip->enable(irq); 218 desc->irq_data.chip->irq_enable(data);
313 return 0; 219 return 0;
314} 220}
315 221
316/* 222/*
317 * default shutdown function 223 * default shutdown function
318 */ 224 */
319static void default_shutdown(unsigned int irq) 225static void default_shutdown(struct irq_data *data)
320{ 226{
321 struct irq_desc *desc = irq_to_desc(irq); 227 struct irq_desc *desc = irq_data_to_desc(data);
322 228
323 desc->chip->mask(irq); 229 desc->irq_data.chip->irq_mask(&desc->irq_data);
324 desc->status |= IRQ_MASKED; 230 desc->status |= IRQ_MASKED;
325} 231}
326 232
233#ifndef CONFIG_GENERIC_HARDIRQS_NO_DEPRECATED
234/* Temporary migration helpers */
235static void compat_irq_mask(struct irq_data *data)
236{
237 data->chip->mask(data->irq);
238}
239
240static void compat_irq_unmask(struct irq_data *data)
241{
242 data->chip->unmask(data->irq);
243}
244
245static void compat_irq_ack(struct irq_data *data)
246{
247 data->chip->ack(data->irq);
248}
249
250static void compat_irq_mask_ack(struct irq_data *data)
251{
252 data->chip->mask_ack(data->irq);
253}
254
255static void compat_irq_eoi(struct irq_data *data)
256{
257 data->chip->eoi(data->irq);
258}
259
260static void compat_irq_enable(struct irq_data *data)
261{
262 data->chip->enable(data->irq);
263}
264
265static void compat_irq_disable(struct irq_data *data)
266{
267 data->chip->disable(data->irq);
268}
269
270static void compat_irq_shutdown(struct irq_data *data)
271{
272 data->chip->shutdown(data->irq);
273}
274
275static unsigned int compat_irq_startup(struct irq_data *data)
276{
277 return data->chip->startup(data->irq);
278}
279
280static int compat_irq_set_affinity(struct irq_data *data,
281 const struct cpumask *dest, bool force)
282{
283 return data->chip->set_affinity(data->irq, dest);
284}
285
286static int compat_irq_set_type(struct irq_data *data, unsigned int type)
287{
288 return data->chip->set_type(data->irq, type);
289}
290
291static int compat_irq_set_wake(struct irq_data *data, unsigned int on)
292{
293 return data->chip->set_wake(data->irq, on);
294}
295
296static int compat_irq_retrigger(struct irq_data *data)
297{
298 return data->chip->retrigger(data->irq);
299}
300
301static void compat_bus_lock(struct irq_data *data)
302{
303 data->chip->bus_lock(data->irq);
304}
305
306static void compat_bus_sync_unlock(struct irq_data *data)
307{
308 data->chip->bus_sync_unlock(data->irq);
309}
310#endif
311
327/* 312/*
328 * Fixup enable/disable function pointers 313 * Fixup enable/disable function pointers
329 */ 314 */
330void irq_chip_set_defaults(struct irq_chip *chip) 315void irq_chip_set_defaults(struct irq_chip *chip)
331{ 316{
332 if (!chip->enable) 317#ifndef CONFIG_GENERIC_HARDIRQS_NO_DEPRECATED
333 chip->enable = default_enable;
334 if (!chip->disable)
335 chip->disable = default_disable;
336 if (!chip->startup)
337 chip->startup = default_startup;
338 /* 318 /*
339 * We use chip->disable, when the user provided its own. When 319 * Compat fixup functions need to be before we set the
340 * we have default_disable set for chip->disable, then we need 320 * defaults for enable/disable/startup/shutdown
321 */
322 if (chip->enable)
323 chip->irq_enable = compat_irq_enable;
324 if (chip->disable)
325 chip->irq_disable = compat_irq_disable;
326 if (chip->shutdown)
327 chip->irq_shutdown = compat_irq_shutdown;
328 if (chip->startup)
329 chip->irq_startup = compat_irq_startup;
330#endif
331 /*
332 * The real defaults
333 */
334 if (!chip->irq_enable)
335 chip->irq_enable = default_enable;
336 if (!chip->irq_disable)
337 chip->irq_disable = default_disable;
338 if (!chip->irq_startup)
339 chip->irq_startup = default_startup;
340 /*
341 * We use chip->irq_disable, when the user provided its own. When
342 * we have default_disable set for chip->irq_disable, then we need
341 * to use default_shutdown, otherwise the irq line is not 343 * to use default_shutdown, otherwise the irq line is not
342 * disabled on free_irq(): 344 * disabled on free_irq():
343 */ 345 */
344 if (!chip->shutdown) 346 if (!chip->irq_shutdown)
345 chip->shutdown = chip->disable != default_disable ? 347 chip->irq_shutdown = chip->irq_disable != default_disable ?
346 chip->disable : default_shutdown; 348 chip->irq_disable : default_shutdown;
347 if (!chip->name) 349
348 chip->name = chip->typename; 350#ifndef CONFIG_GENERIC_HARDIRQS_NO_DEPRECATED
349 if (!chip->end) 351 if (!chip->end)
350 chip->end = dummy_irq_chip.end; 352 chip->end = dummy_irq_chip.end;
353
354 /*
355 * Now fix up the remaining compat handlers
356 */
357 if (chip->bus_lock)
358 chip->irq_bus_lock = compat_bus_lock;
359 if (chip->bus_sync_unlock)
360 chip->irq_bus_sync_unlock = compat_bus_sync_unlock;
361 if (chip->mask)
362 chip->irq_mask = compat_irq_mask;
363 if (chip->unmask)
364 chip->irq_unmask = compat_irq_unmask;
365 if (chip->ack)
366 chip->irq_ack = compat_irq_ack;
367 if (chip->mask_ack)
368 chip->irq_mask_ack = compat_irq_mask_ack;
369 if (chip->eoi)
370 chip->irq_eoi = compat_irq_eoi;
371 if (chip->set_affinity)
372 chip->irq_set_affinity = compat_irq_set_affinity;
373 if (chip->set_type)
374 chip->irq_set_type = compat_irq_set_type;
375 if (chip->set_wake)
376 chip->irq_set_wake = compat_irq_set_wake;
377 if (chip->retrigger)
378 chip->irq_retrigger = compat_irq_retrigger;
379#endif
351} 380}
352 381
353static inline void mask_ack_irq(struct irq_desc *desc, int irq) 382static inline void mask_ack_irq(struct irq_desc *desc)
354{ 383{
355 if (desc->chip->mask_ack) 384 if (desc->irq_data.chip->irq_mask_ack)
356 desc->chip->mask_ack(irq); 385 desc->irq_data.chip->irq_mask_ack(&desc->irq_data);
357 else { 386 else {
358 desc->chip->mask(irq); 387 desc->irq_data.chip->irq_mask(&desc->irq_data);
359 if (desc->chip->ack) 388 if (desc->irq_data.chip->irq_ack)
360 desc->chip->ack(irq); 389 desc->irq_data.chip->irq_ack(&desc->irq_data);
361 } 390 }
362 desc->status |= IRQ_MASKED; 391 desc->status |= IRQ_MASKED;
363} 392}
364 393
365static inline void mask_irq(struct irq_desc *desc, int irq) 394static inline void mask_irq(struct irq_desc *desc)
366{ 395{
367 if (desc->chip->mask) { 396 if (desc->irq_data.chip->irq_mask) {
368 desc->chip->mask(irq); 397 desc->irq_data.chip->irq_mask(&desc->irq_data);
369 desc->status |= IRQ_MASKED; 398 desc->status |= IRQ_MASKED;
370 } 399 }
371} 400}
372 401
373static inline void unmask_irq(struct irq_desc *desc, int irq) 402static inline void unmask_irq(struct irq_desc *desc)
374{ 403{
375 if (desc->chip->unmask) { 404 if (desc->irq_data.chip->irq_unmask) {
376 desc->chip->unmask(irq); 405 desc->irq_data.chip->irq_unmask(&desc->irq_data);
377 desc->status &= ~IRQ_MASKED; 406 desc->status &= ~IRQ_MASKED;
378 } 407 }
379} 408}
@@ -476,7 +505,7 @@ handle_level_irq(unsigned int irq, struct irq_desc *desc)
476 irqreturn_t action_ret; 505 irqreturn_t action_ret;
477 506
478 raw_spin_lock(&desc->lock); 507 raw_spin_lock(&desc->lock);
479 mask_ack_irq(desc, irq); 508 mask_ack_irq(desc);
480 509
481 if (unlikely(desc->status & IRQ_INPROGRESS)) 510 if (unlikely(desc->status & IRQ_INPROGRESS))
482 goto out_unlock; 511 goto out_unlock;
@@ -502,7 +531,7 @@ handle_level_irq(unsigned int irq, struct irq_desc *desc)
502 desc->status &= ~IRQ_INPROGRESS; 531 desc->status &= ~IRQ_INPROGRESS;
503 532
504 if (!(desc->status & (IRQ_DISABLED | IRQ_ONESHOT))) 533 if (!(desc->status & (IRQ_DISABLED | IRQ_ONESHOT)))
505 unmask_irq(desc, irq); 534 unmask_irq(desc);
506out_unlock: 535out_unlock:
507 raw_spin_unlock(&desc->lock); 536 raw_spin_unlock(&desc->lock);
508} 537}
@@ -539,7 +568,7 @@ handle_fasteoi_irq(unsigned int irq, struct irq_desc *desc)
539 action = desc->action; 568 action = desc->action;
540 if (unlikely(!action || (desc->status & IRQ_DISABLED))) { 569 if (unlikely(!action || (desc->status & IRQ_DISABLED))) {
541 desc->status |= IRQ_PENDING; 570 desc->status |= IRQ_PENDING;
542 mask_irq(desc, irq); 571 mask_irq(desc);
543 goto out; 572 goto out;
544 } 573 }
545 574
@@ -554,7 +583,7 @@ handle_fasteoi_irq(unsigned int irq, struct irq_desc *desc)
554 raw_spin_lock(&desc->lock); 583 raw_spin_lock(&desc->lock);
555 desc->status &= ~IRQ_INPROGRESS; 584 desc->status &= ~IRQ_INPROGRESS;
556out: 585out:
557 desc->chip->eoi(irq); 586 desc->irq_data.chip->irq_eoi(&desc->irq_data);
558 587
559 raw_spin_unlock(&desc->lock); 588 raw_spin_unlock(&desc->lock);
560} 589}
@@ -590,14 +619,13 @@ handle_edge_irq(unsigned int irq, struct irq_desc *desc)
590 if (unlikely((desc->status & (IRQ_INPROGRESS | IRQ_DISABLED)) || 619 if (unlikely((desc->status & (IRQ_INPROGRESS | IRQ_DISABLED)) ||
591 !desc->action)) { 620 !desc->action)) {
592 desc->status |= (IRQ_PENDING | IRQ_MASKED); 621 desc->status |= (IRQ_PENDING | IRQ_MASKED);
593 mask_ack_irq(desc, irq); 622 mask_ack_irq(desc);
594 goto out_unlock; 623 goto out_unlock;
595 } 624 }
596 kstat_incr_irqs_this_cpu(irq, desc); 625 kstat_incr_irqs_this_cpu(irq, desc);
597 626
598 /* Start handling the irq */ 627 /* Start handling the irq */
599 if (desc->chip->ack) 628 desc->irq_data.chip->irq_ack(&desc->irq_data);
600 desc->chip->ack(irq);
601 629
602 /* Mark the IRQ currently in progress.*/ 630 /* Mark the IRQ currently in progress.*/
603 desc->status |= IRQ_INPROGRESS; 631 desc->status |= IRQ_INPROGRESS;
@@ -607,7 +635,7 @@ handle_edge_irq(unsigned int irq, struct irq_desc *desc)
607 irqreturn_t action_ret; 635 irqreturn_t action_ret;
608 636
609 if (unlikely(!action)) { 637 if (unlikely(!action)) {
610 mask_irq(desc, irq); 638 mask_irq(desc);
611 goto out_unlock; 639 goto out_unlock;
612 } 640 }
613 641
@@ -619,7 +647,7 @@ handle_edge_irq(unsigned int irq, struct irq_desc *desc)
619 if (unlikely((desc->status & 647 if (unlikely((desc->status &
620 (IRQ_PENDING | IRQ_MASKED | IRQ_DISABLED)) == 648 (IRQ_PENDING | IRQ_MASKED | IRQ_DISABLED)) ==
621 (IRQ_PENDING | IRQ_MASKED))) { 649 (IRQ_PENDING | IRQ_MASKED))) {
622 unmask_irq(desc, irq); 650 unmask_irq(desc);
623 } 651 }
624 652
625 desc->status &= ~IRQ_PENDING; 653 desc->status &= ~IRQ_PENDING;
@@ -650,15 +678,15 @@ handle_percpu_irq(unsigned int irq, struct irq_desc *desc)
650 678
651 kstat_incr_irqs_this_cpu(irq, desc); 679 kstat_incr_irqs_this_cpu(irq, desc);
652 680
653 if (desc->chip->ack) 681 if (desc->irq_data.chip->irq_ack)
654 desc->chip->ack(irq); 682 desc->irq_data.chip->irq_ack(&desc->irq_data);
655 683
656 action_ret = handle_IRQ_event(irq, desc->action); 684 action_ret = handle_IRQ_event(irq, desc->action);
657 if (!noirqdebug) 685 if (!noirqdebug)
658 note_interrupt(irq, desc, action_ret); 686 note_interrupt(irq, desc, action_ret);
659 687
660 if (desc->chip->eoi) 688 if (desc->irq_data.chip->irq_eoi)
661 desc->chip->eoi(irq); 689 desc->irq_data.chip->irq_eoi(&desc->irq_data);
662} 690}
663 691
664void 692void
@@ -676,7 +704,7 @@ __set_irq_handler(unsigned int irq, irq_flow_handler_t handle, int is_chained,
676 704
677 if (!handle) 705 if (!handle)
678 handle = handle_bad_irq; 706 handle = handle_bad_irq;
679 else if (desc->chip == &no_irq_chip) { 707 else if (desc->irq_data.chip == &no_irq_chip) {
680 printk(KERN_WARNING "Trying to install %sinterrupt handler " 708 printk(KERN_WARNING "Trying to install %sinterrupt handler "
681 "for IRQ%d\n", is_chained ? "chained " : "", irq); 709 "for IRQ%d\n", is_chained ? "chained " : "", irq);
682 /* 710 /*
@@ -686,16 +714,16 @@ __set_irq_handler(unsigned int irq, irq_flow_handler_t handle, int is_chained,
686 * prevent us to setup the interrupt at all. Switch it to 714 * prevent us to setup the interrupt at all. Switch it to
687 * dummy_irq_chip for easy transition. 715 * dummy_irq_chip for easy transition.
688 */ 716 */
689 desc->chip = &dummy_irq_chip; 717 desc->irq_data.chip = &dummy_irq_chip;
690 } 718 }
691 719
692 chip_bus_lock(irq, desc); 720 chip_bus_lock(desc);
693 raw_spin_lock_irqsave(&desc->lock, flags); 721 raw_spin_lock_irqsave(&desc->lock, flags);
694 722
695 /* Uninstall? */ 723 /* Uninstall? */
696 if (handle == handle_bad_irq) { 724 if (handle == handle_bad_irq) {
697 if (desc->chip != &no_irq_chip) 725 if (desc->irq_data.chip != &no_irq_chip)
698 mask_ack_irq(desc, irq); 726 mask_ack_irq(desc);
699 desc->status |= IRQ_DISABLED; 727 desc->status |= IRQ_DISABLED;
700 desc->depth = 1; 728 desc->depth = 1;
701 } 729 }
@@ -706,10 +734,10 @@ __set_irq_handler(unsigned int irq, irq_flow_handler_t handle, int is_chained,
706 desc->status &= ~IRQ_DISABLED; 734 desc->status &= ~IRQ_DISABLED;
707 desc->status |= IRQ_NOREQUEST | IRQ_NOPROBE; 735 desc->status |= IRQ_NOREQUEST | IRQ_NOPROBE;
708 desc->depth = 0; 736 desc->depth = 0;
709 desc->chip->startup(irq); 737 desc->irq_data.chip->irq_startup(&desc->irq_data);
710 } 738 }
711 raw_spin_unlock_irqrestore(&desc->lock, flags); 739 raw_spin_unlock_irqrestore(&desc->lock, flags);
712 chip_bus_sync_unlock(irq, desc); 740 chip_bus_sync_unlock(desc);
713} 741}
714EXPORT_SYMBOL_GPL(__set_irq_handler); 742EXPORT_SYMBOL_GPL(__set_irq_handler);
715 743
@@ -729,32 +757,20 @@ set_irq_chip_and_handler_name(unsigned int irq, struct irq_chip *chip,
729 __set_irq_handler(irq, handle, 0, name); 757 __set_irq_handler(irq, handle, 0, name);
730} 758}
731 759
732void set_irq_noprobe(unsigned int irq) 760void irq_modify_status(unsigned int irq, unsigned long clr, unsigned long set)
733{ 761{
734 struct irq_desc *desc = irq_to_desc(irq); 762 struct irq_desc *desc = irq_to_desc(irq);
735 unsigned long flags; 763 unsigned long flags;
736 764
737 if (!desc) { 765 if (!desc)
738 printk(KERN_ERR "Trying to mark IRQ%d non-probeable\n", irq);
739 return; 766 return;
740 }
741
742 raw_spin_lock_irqsave(&desc->lock, flags);
743 desc->status |= IRQ_NOPROBE;
744 raw_spin_unlock_irqrestore(&desc->lock, flags);
745}
746
747void set_irq_probe(unsigned int irq)
748{
749 struct irq_desc *desc = irq_to_desc(irq);
750 unsigned long flags;
751 767
752 if (!desc) { 768 /* Sanitize flags */
753 printk(KERN_ERR "Trying to mark IRQ%d probeable\n", irq); 769 set &= IRQF_MODIFY_MASK;
754 return; 770 clr &= IRQF_MODIFY_MASK;
755 }
756 771
757 raw_spin_lock_irqsave(&desc->lock, flags); 772 raw_spin_lock_irqsave(&desc->lock, flags);
758 desc->status &= ~IRQ_NOPROBE; 773 desc->status &= ~clr;
774 desc->status |= set;
759 raw_spin_unlock_irqrestore(&desc->lock, flags); 775 raw_spin_unlock_irqrestore(&desc->lock, flags);
760} 776}
diff --git a/kernel/irq/dummychip.c b/kernel/irq/dummychip.c
new file mode 100644
index 000000000000..20dc5474947e
--- /dev/null
+++ b/kernel/irq/dummychip.c
@@ -0,0 +1,68 @@
1/*
2 * Copyright (C) 1992, 1998-2006 Linus Torvalds, Ingo Molnar
3 * Copyright (C) 2005-2006, Thomas Gleixner, Russell King
4 *
5 * This file contains the dummy interrupt chip implementation
6 */
7#include <linux/interrupt.h>
8#include <linux/irq.h>
9
10#include "internals.h"
11
12/*
13 * What should we do if we get a hw irq event on an illegal vector?
14 * Each architecture has to answer this themself.
15 */
16static void ack_bad(struct irq_data *data)
17{
18 struct irq_desc *desc = irq_data_to_desc(data);
19
20 print_irq_desc(data->irq, desc);
21 ack_bad_irq(data->irq);
22}
23
24/*
25 * NOP functions
26 */
27static void noop(struct irq_data *data) { }
28
29static unsigned int noop_ret(struct irq_data *data)
30{
31 return 0;
32}
33
34#ifndef CONFIG_GENERIC_HARDIRQS_NO_DEPRECATED
35static void compat_noop(unsigned int irq) { }
36#define END_INIT .end = compat_noop
37#else
38#define END_INIT
39#endif
40
41/*
42 * Generic no controller implementation
43 */
44struct irq_chip no_irq_chip = {
45 .name = "none",
46 .irq_startup = noop_ret,
47 .irq_shutdown = noop,
48 .irq_enable = noop,
49 .irq_disable = noop,
50 .irq_ack = ack_bad,
51 END_INIT
52};
53
54/*
55 * Generic dummy implementation which can be used for
56 * real dumb interrupt sources
57 */
58struct irq_chip dummy_irq_chip = {
59 .name = "dummy",
60 .irq_startup = noop_ret,
61 .irq_shutdown = noop,
62 .irq_enable = noop,
63 .irq_disable = noop,
64 .irq_ack = noop,
65 .irq_mask = noop,
66 .irq_unmask = noop,
67 END_INIT
68};
diff --git a/kernel/irq/handle.c b/kernel/irq/handle.c
index 27e5c6911223..e2347eb63306 100644
--- a/kernel/irq/handle.c
+++ b/kernel/irq/handle.c
@@ -11,24 +11,15 @@
11 */ 11 */
12 12
13#include <linux/irq.h> 13#include <linux/irq.h>
14#include <linux/sched.h>
15#include <linux/slab.h>
16#include <linux/module.h>
17#include <linux/random.h> 14#include <linux/random.h>
15#include <linux/sched.h>
18#include <linux/interrupt.h> 16#include <linux/interrupt.h>
19#include <linux/kernel_stat.h> 17#include <linux/kernel_stat.h>
20#include <linux/rculist.h> 18
21#include <linux/hash.h>
22#include <linux/radix-tree.h>
23#include <trace/events/irq.h> 19#include <trace/events/irq.h>
24 20
25#include "internals.h" 21#include "internals.h"
26 22
27/*
28 * lockdep: we want to handle all irq_desc locks as a single lock-class:
29 */
30struct lock_class_key irq_desc_lock_class;
31
32/** 23/**
33 * handle_bad_irq - handle spurious and unhandled irqs 24 * handle_bad_irq - handle spurious and unhandled irqs
34 * @irq: the interrupt number 25 * @irq: the interrupt number
@@ -43,304 +34,6 @@ void handle_bad_irq(unsigned int irq, struct irq_desc *desc)
43 ack_bad_irq(irq); 34 ack_bad_irq(irq);
44} 35}
45 36
46#if defined(CONFIG_SMP) && defined(CONFIG_GENERIC_HARDIRQS)
47static void __init init_irq_default_affinity(void)
48{
49 alloc_cpumask_var(&irq_default_affinity, GFP_NOWAIT);
50 cpumask_setall(irq_default_affinity);
51}
52#else
53static void __init init_irq_default_affinity(void)
54{
55}
56#endif
57
58/*
59 * Linux has a controller-independent interrupt architecture.
60 * Every controller has a 'controller-template', that is used
61 * by the main code to do the right thing. Each driver-visible
62 * interrupt source is transparently wired to the appropriate
63 * controller. Thus drivers need not be aware of the
64 * interrupt-controller.
65 *
66 * The code is designed to be easily extended with new/different
67 * interrupt controllers, without having to do assembly magic or
68 * having to touch the generic code.
69 *
70 * Controller mappings for all interrupt sources:
71 */
72int nr_irqs = NR_IRQS;
73EXPORT_SYMBOL_GPL(nr_irqs);
74
75#ifdef CONFIG_SPARSE_IRQ
76
77static struct irq_desc irq_desc_init = {
78 .irq = -1,
79 .status = IRQ_DISABLED,
80 .chip = &no_irq_chip,
81 .handle_irq = handle_bad_irq,
82 .depth = 1,
83 .lock = __RAW_SPIN_LOCK_UNLOCKED(irq_desc_init.lock),
84};
85
86void __ref init_kstat_irqs(struct irq_desc *desc, int node, int nr)
87{
88 void *ptr;
89
90 ptr = kzalloc_node(nr * sizeof(*desc->kstat_irqs),
91 GFP_ATOMIC, node);
92
93 /*
94 * don't overwite if can not get new one
95 * init_copy_kstat_irqs() could still use old one
96 */
97 if (ptr) {
98 printk(KERN_DEBUG " alloc kstat_irqs on node %d\n", node);
99 desc->kstat_irqs = ptr;
100 }
101}
102
103static void init_one_irq_desc(int irq, struct irq_desc *desc, int node)
104{
105 memcpy(desc, &irq_desc_init, sizeof(struct irq_desc));
106
107 raw_spin_lock_init(&desc->lock);
108 desc->irq = irq;
109#ifdef CONFIG_SMP
110 desc->node = node;
111#endif
112 lockdep_set_class(&desc->lock, &irq_desc_lock_class);
113 init_kstat_irqs(desc, node, nr_cpu_ids);
114 if (!desc->kstat_irqs) {
115 printk(KERN_ERR "can not alloc kstat_irqs\n");
116 BUG_ON(1);
117 }
118 if (!alloc_desc_masks(desc, node, false)) {
119 printk(KERN_ERR "can not alloc irq_desc cpumasks\n");
120 BUG_ON(1);
121 }
122 init_desc_masks(desc);
123 arch_init_chip_data(desc, node);
124}
125
126/*
127 * Protect the sparse_irqs:
128 */
129DEFINE_RAW_SPINLOCK(sparse_irq_lock);
130
131static RADIX_TREE(irq_desc_tree, GFP_ATOMIC);
132
133static void set_irq_desc(unsigned int irq, struct irq_desc *desc)
134{
135 radix_tree_insert(&irq_desc_tree, irq, desc);
136}
137
138struct irq_desc *irq_to_desc(unsigned int irq)
139{
140 return radix_tree_lookup(&irq_desc_tree, irq);
141}
142
143void replace_irq_desc(unsigned int irq, struct irq_desc *desc)
144{
145 void **ptr;
146
147 ptr = radix_tree_lookup_slot(&irq_desc_tree, irq);
148 if (ptr)
149 radix_tree_replace_slot(ptr, desc);
150}
151
152static struct irq_desc irq_desc_legacy[NR_IRQS_LEGACY] __cacheline_aligned_in_smp = {
153 [0 ... NR_IRQS_LEGACY-1] = {
154 .irq = -1,
155 .status = IRQ_DISABLED,
156 .chip = &no_irq_chip,
157 .handle_irq = handle_bad_irq,
158 .depth = 1,
159 .lock = __RAW_SPIN_LOCK_UNLOCKED(irq_desc_init.lock),
160 }
161};
162
163static unsigned int *kstat_irqs_legacy;
164
165int __init early_irq_init(void)
166{
167 struct irq_desc *desc;
168 int legacy_count;
169 int node;
170 int i;
171
172 init_irq_default_affinity();
173
174 /* initialize nr_irqs based on nr_cpu_ids */
175 arch_probe_nr_irqs();
176 printk(KERN_INFO "NR_IRQS:%d nr_irqs:%d\n", NR_IRQS, nr_irqs);
177
178 desc = irq_desc_legacy;
179 legacy_count = ARRAY_SIZE(irq_desc_legacy);
180 node = first_online_node;
181
182 /* allocate based on nr_cpu_ids */
183 kstat_irqs_legacy = kzalloc_node(NR_IRQS_LEGACY * nr_cpu_ids *
184 sizeof(int), GFP_NOWAIT, node);
185
186 for (i = 0; i < legacy_count; i++) {
187 desc[i].irq = i;
188#ifdef CONFIG_SMP
189 desc[i].node = node;
190#endif
191 desc[i].kstat_irqs = kstat_irqs_legacy + i * nr_cpu_ids;
192 lockdep_set_class(&desc[i].lock, &irq_desc_lock_class);
193 alloc_desc_masks(&desc[i], node, true);
194 init_desc_masks(&desc[i]);
195 set_irq_desc(i, &desc[i]);
196 }
197
198 return arch_early_irq_init();
199}
200
201struct irq_desc * __ref irq_to_desc_alloc_node(unsigned int irq, int node)
202{
203 struct irq_desc *desc;
204 unsigned long flags;
205
206 if (irq >= nr_irqs) {
207 WARN(1, "irq (%d) >= nr_irqs (%d) in irq_to_desc_alloc\n",
208 irq, nr_irqs);
209 return NULL;
210 }
211
212 desc = irq_to_desc(irq);
213 if (desc)
214 return desc;
215
216 raw_spin_lock_irqsave(&sparse_irq_lock, flags);
217
218 /* We have to check it to avoid races with another CPU */
219 desc = irq_to_desc(irq);
220 if (desc)
221 goto out_unlock;
222
223 desc = kzalloc_node(sizeof(*desc), GFP_ATOMIC, node);
224
225 printk(KERN_DEBUG " alloc irq_desc for %d on node %d\n", irq, node);
226 if (!desc) {
227 printk(KERN_ERR "can not alloc irq_desc\n");
228 BUG_ON(1);
229 }
230 init_one_irq_desc(irq, desc, node);
231
232 set_irq_desc(irq, desc);
233
234out_unlock:
235 raw_spin_unlock_irqrestore(&sparse_irq_lock, flags);
236
237 return desc;
238}
239
240#else /* !CONFIG_SPARSE_IRQ */
241
242struct irq_desc irq_desc[NR_IRQS] __cacheline_aligned_in_smp = {
243 [0 ... NR_IRQS-1] = {
244 .status = IRQ_DISABLED,
245 .chip = &no_irq_chip,
246 .handle_irq = handle_bad_irq,
247 .depth = 1,
248 .lock = __RAW_SPIN_LOCK_UNLOCKED(irq_desc->lock),
249 }
250};
251
252static unsigned int kstat_irqs_all[NR_IRQS][NR_CPUS];
253int __init early_irq_init(void)
254{
255 struct irq_desc *desc;
256 int count;
257 int i;
258
259 init_irq_default_affinity();
260
261 printk(KERN_INFO "NR_IRQS:%d\n", NR_IRQS);
262
263 desc = irq_desc;
264 count = ARRAY_SIZE(irq_desc);
265
266 for (i = 0; i < count; i++) {
267 desc[i].irq = i;
268 alloc_desc_masks(&desc[i], 0, true);
269 init_desc_masks(&desc[i]);
270 desc[i].kstat_irqs = kstat_irqs_all[i];
271 }
272 return arch_early_irq_init();
273}
274
275struct irq_desc *irq_to_desc(unsigned int irq)
276{
277 return (irq < NR_IRQS) ? irq_desc + irq : NULL;
278}
279
280struct irq_desc *irq_to_desc_alloc_node(unsigned int irq, int node)
281{
282 return irq_to_desc(irq);
283}
284#endif /* !CONFIG_SPARSE_IRQ */
285
286void clear_kstat_irqs(struct irq_desc *desc)
287{
288 memset(desc->kstat_irqs, 0, nr_cpu_ids * sizeof(*(desc->kstat_irqs)));
289}
290
291/*
292 * What should we do if we get a hw irq event on an illegal vector?
293 * Each architecture has to answer this themself.
294 */
295static void ack_bad(unsigned int irq)
296{
297 struct irq_desc *desc = irq_to_desc(irq);
298
299 print_irq_desc(irq, desc);
300 ack_bad_irq(irq);
301}
302
303/*
304 * NOP functions
305 */
306static void noop(unsigned int irq)
307{
308}
309
310static unsigned int noop_ret(unsigned int irq)
311{
312 return 0;
313}
314
315/*
316 * Generic no controller implementation
317 */
318struct irq_chip no_irq_chip = {
319 .name = "none",
320 .startup = noop_ret,
321 .shutdown = noop,
322 .enable = noop,
323 .disable = noop,
324 .ack = ack_bad,
325 .end = noop,
326};
327
328/*
329 * Generic dummy implementation which can be used for
330 * real dumb interrupt sources
331 */
332struct irq_chip dummy_irq_chip = {
333 .name = "dummy",
334 .startup = noop_ret,
335 .shutdown = noop,
336 .enable = noop,
337 .disable = noop,
338 .ack = noop,
339 .mask = noop,
340 .unmask = noop,
341 .end = noop,
342};
343
344/* 37/*
345 * Special, empty irq handler: 38 * Special, empty irq handler:
346 */ 39 */
@@ -457,20 +150,20 @@ unsigned int __do_IRQ(unsigned int irq)
457 /* 150 /*
458 * No locking required for CPU-local interrupts: 151 * No locking required for CPU-local interrupts:
459 */ 152 */
460 if (desc->chip->ack) 153 if (desc->irq_data.chip->ack)
461 desc->chip->ack(irq); 154 desc->irq_data.chip->ack(irq);
462 if (likely(!(desc->status & IRQ_DISABLED))) { 155 if (likely(!(desc->status & IRQ_DISABLED))) {
463 action_ret = handle_IRQ_event(irq, desc->action); 156 action_ret = handle_IRQ_event(irq, desc->action);
464 if (!noirqdebug) 157 if (!noirqdebug)
465 note_interrupt(irq, desc, action_ret); 158 note_interrupt(irq, desc, action_ret);
466 } 159 }
467 desc->chip->end(irq); 160 desc->irq_data.chip->end(irq);
468 return 1; 161 return 1;
469 } 162 }
470 163
471 raw_spin_lock(&desc->lock); 164 raw_spin_lock(&desc->lock);
472 if (desc->chip->ack) 165 if (desc->irq_data.chip->ack)
473 desc->chip->ack(irq); 166 desc->irq_data.chip->ack(irq);
474 /* 167 /*
475 * REPLAY is when Linux resends an IRQ that was dropped earlier 168 * REPLAY is when Linux resends an IRQ that was dropped earlier
476 * WAITING is used by probe to mark irqs that are being tested 169 * WAITING is used by probe to mark irqs that are being tested
@@ -530,27 +223,9 @@ out:
530 * The ->end() handler has to deal with interrupts which got 223 * The ->end() handler has to deal with interrupts which got
531 * disabled while the handler was running. 224 * disabled while the handler was running.
532 */ 225 */
533 desc->chip->end(irq); 226 desc->irq_data.chip->end(irq);
534 raw_spin_unlock(&desc->lock); 227 raw_spin_unlock(&desc->lock);
535 228
536 return 1; 229 return 1;
537} 230}
538#endif 231#endif
539
540void early_init_irq_lock_class(void)
541{
542 struct irq_desc *desc;
543 int i;
544
545 for_each_irq_desc(i, desc) {
546 lockdep_set_class(&desc->lock, &irq_desc_lock_class);
547 }
548}
549
550unsigned int kstat_irqs_cpu(unsigned int irq, int cpu)
551{
552 struct irq_desc *desc = irq_to_desc(irq);
553 return desc ? desc->kstat_irqs[cpu] : 0;
554}
555EXPORT_SYMBOL(kstat_irqs_cpu);
556
diff --git a/kernel/irq/internals.h b/kernel/irq/internals.h
index c63f3bc88f0b..4571ae7e085a 100644
--- a/kernel/irq/internals.h
+++ b/kernel/irq/internals.h
@@ -1,9 +1,12 @@
1/* 1/*
2 * IRQ subsystem internal functions and variables: 2 * IRQ subsystem internal functions and variables:
3 */ 3 */
4#include <linux/irqdesc.h>
4 5
5extern int noirqdebug; 6extern int noirqdebug;
6 7
8#define irq_data_to_desc(data) container_of(data, struct irq_desc, irq_data)
9
7/* Set default functions for irq_chip structures: */ 10/* Set default functions for irq_chip structures: */
8extern void irq_chip_set_defaults(struct irq_chip *chip); 11extern void irq_chip_set_defaults(struct irq_chip *chip);
9 12
@@ -15,21 +18,19 @@ extern int __irq_set_trigger(struct irq_desc *desc, unsigned int irq,
15extern void __disable_irq(struct irq_desc *desc, unsigned int irq, bool susp); 18extern void __disable_irq(struct irq_desc *desc, unsigned int irq, bool susp);
16extern void __enable_irq(struct irq_desc *desc, unsigned int irq, bool resume); 19extern void __enable_irq(struct irq_desc *desc, unsigned int irq, bool resume);
17 20
18extern struct lock_class_key irq_desc_lock_class;
19extern void init_kstat_irqs(struct irq_desc *desc, int node, int nr); 21extern void init_kstat_irqs(struct irq_desc *desc, int node, int nr);
20extern void clear_kstat_irqs(struct irq_desc *desc);
21extern raw_spinlock_t sparse_irq_lock;
22 22
23#ifdef CONFIG_SPARSE_IRQ 23/* Resending of interrupts :*/
24void replace_irq_desc(unsigned int irq, struct irq_desc *desc); 24void check_irq_resend(struct irq_desc *desc, unsigned int irq);
25#endif
26 25
27#ifdef CONFIG_PROC_FS 26#ifdef CONFIG_PROC_FS
28extern void register_irq_proc(unsigned int irq, struct irq_desc *desc); 27extern void register_irq_proc(unsigned int irq, struct irq_desc *desc);
28extern void unregister_irq_proc(unsigned int irq, struct irq_desc *desc);
29extern void register_handler_proc(unsigned int irq, struct irqaction *action); 29extern void register_handler_proc(unsigned int irq, struct irqaction *action);
30extern void unregister_handler_proc(unsigned int irq, struct irqaction *action); 30extern void unregister_handler_proc(unsigned int irq, struct irqaction *action);
31#else 31#else
32static inline void register_irq_proc(unsigned int irq, struct irq_desc *desc) { } 32static inline void register_irq_proc(unsigned int irq, struct irq_desc *desc) { }
33static inline void unregister_irq_proc(unsigned int irq, struct irq_desc *desc) { }
33static inline void register_handler_proc(unsigned int irq, 34static inline void register_handler_proc(unsigned int irq,
34 struct irqaction *action) { } 35 struct irqaction *action) { }
35static inline void unregister_handler_proc(unsigned int irq, 36static inline void unregister_handler_proc(unsigned int irq,
@@ -40,17 +41,27 @@ extern int irq_select_affinity_usr(unsigned int irq);
40 41
41extern void irq_set_thread_affinity(struct irq_desc *desc); 42extern void irq_set_thread_affinity(struct irq_desc *desc);
42 43
44#ifndef CONFIG_GENERIC_HARDIRQS_NO_DEPRECATED
45static inline void irq_end(unsigned int irq, struct irq_desc *desc)
46{
47 if (desc->irq_data.chip && desc->irq_data.chip->end)
48 desc->irq_data.chip->end(irq);
49}
50#else
51static inline void irq_end(unsigned int irq, struct irq_desc *desc) { }
52#endif
53
43/* Inline functions for support of irq chips on slow busses */ 54/* Inline functions for support of irq chips on slow busses */
44static inline void chip_bus_lock(unsigned int irq, struct irq_desc *desc) 55static inline void chip_bus_lock(struct irq_desc *desc)
45{ 56{
46 if (unlikely(desc->chip->bus_lock)) 57 if (unlikely(desc->irq_data.chip->irq_bus_lock))
47 desc->chip->bus_lock(irq); 58 desc->irq_data.chip->irq_bus_lock(&desc->irq_data);
48} 59}
49 60
50static inline void chip_bus_sync_unlock(unsigned int irq, struct irq_desc *desc) 61static inline void chip_bus_sync_unlock(struct irq_desc *desc)
51{ 62{
52 if (unlikely(desc->chip->bus_sync_unlock)) 63 if (unlikely(desc->irq_data.chip->irq_bus_sync_unlock))
53 desc->chip->bus_sync_unlock(irq); 64 desc->irq_data.chip->irq_bus_sync_unlock(&desc->irq_data);
54} 65}
55 66
56/* 67/*
@@ -67,8 +78,8 @@ static inline void print_irq_desc(unsigned int irq, struct irq_desc *desc)
67 irq, desc, desc->depth, desc->irq_count, desc->irqs_unhandled); 78 irq, desc, desc->depth, desc->irq_count, desc->irqs_unhandled);
68 printk("->handle_irq(): %p, ", desc->handle_irq); 79 printk("->handle_irq(): %p, ", desc->handle_irq);
69 print_symbol("%s\n", (unsigned long)desc->handle_irq); 80 print_symbol("%s\n", (unsigned long)desc->handle_irq);
70 printk("->chip(): %p, ", desc->chip); 81 printk("->irq_data.chip(): %p, ", desc->irq_data.chip);
71 print_symbol("%s\n", (unsigned long)desc->chip); 82 print_symbol("%s\n", (unsigned long)desc->irq_data.chip);
72 printk("->action(): %p\n", desc->action); 83 printk("->action(): %p\n", desc->action);
73 if (desc->action) { 84 if (desc->action) {
74 printk("->action->handler(): %p, ", desc->action->handler); 85 printk("->action->handler(): %p, ", desc->action->handler);
diff --git a/kernel/irq/irqdesc.c b/kernel/irq/irqdesc.c
new file mode 100644
index 000000000000..9d917ff72675
--- /dev/null
+++ b/kernel/irq/irqdesc.c
@@ -0,0 +1,395 @@
1/*
2 * Copyright (C) 1992, 1998-2006 Linus Torvalds, Ingo Molnar
3 * Copyright (C) 2005-2006, Thomas Gleixner, Russell King
4 *
5 * This file contains the interrupt descriptor management code
6 *
7 * Detailed information is available in Documentation/DocBook/genericirq
8 *
9 */
10#include <linux/irq.h>
11#include <linux/slab.h>
12#include <linux/module.h>
13#include <linux/interrupt.h>
14#include <linux/kernel_stat.h>
15#include <linux/radix-tree.h>
16#include <linux/bitmap.h>
17
18#include "internals.h"
19
20/*
21 * lockdep: we want to handle all irq_desc locks as a single lock-class:
22 */
23static struct lock_class_key irq_desc_lock_class;
24
25#if defined(CONFIG_SMP) && defined(CONFIG_GENERIC_HARDIRQS)
26static void __init init_irq_default_affinity(void)
27{
28 alloc_cpumask_var(&irq_default_affinity, GFP_NOWAIT);
29 cpumask_setall(irq_default_affinity);
30}
31#else
32static void __init init_irq_default_affinity(void)
33{
34}
35#endif
36
37#ifdef CONFIG_SMP
38static int alloc_masks(struct irq_desc *desc, gfp_t gfp, int node)
39{
40 if (!zalloc_cpumask_var_node(&desc->irq_data.affinity, gfp, node))
41 return -ENOMEM;
42
43#ifdef CONFIG_GENERIC_PENDING_IRQ
44 if (!zalloc_cpumask_var_node(&desc->pending_mask, gfp, node)) {
45 free_cpumask_var(desc->irq_data.affinity);
46 return -ENOMEM;
47 }
48#endif
49 return 0;
50}
51
52static void desc_smp_init(struct irq_desc *desc, int node)
53{
54 desc->irq_data.node = node;
55 cpumask_copy(desc->irq_data.affinity, irq_default_affinity);
56#ifdef CONFIG_GENERIC_PENDING_IRQ
57 cpumask_clear(desc->pending_mask);
58#endif
59}
60
61static inline int desc_node(struct irq_desc *desc)
62{
63 return desc->irq_data.node;
64}
65
66#else
67static inline int
68alloc_masks(struct irq_desc *desc, gfp_t gfp, int node) { return 0; }
69static inline void desc_smp_init(struct irq_desc *desc, int node) { }
70static inline int desc_node(struct irq_desc *desc) { return 0; }
71#endif
72
73static void desc_set_defaults(unsigned int irq, struct irq_desc *desc, int node)
74{
75 desc->irq_data.irq = irq;
76 desc->irq_data.chip = &no_irq_chip;
77 desc->irq_data.chip_data = NULL;
78 desc->irq_data.handler_data = NULL;
79 desc->irq_data.msi_desc = NULL;
80 desc->status = IRQ_DEFAULT_INIT_FLAGS;
81 desc->handle_irq = handle_bad_irq;
82 desc->depth = 1;
83 desc->irq_count = 0;
84 desc->irqs_unhandled = 0;
85 desc->name = NULL;
86 memset(desc->kstat_irqs, 0, nr_cpu_ids * sizeof(*(desc->kstat_irqs)));
87 desc_smp_init(desc, node);
88}
89
90int nr_irqs = NR_IRQS;
91EXPORT_SYMBOL_GPL(nr_irqs);
92
93static DEFINE_MUTEX(sparse_irq_lock);
94static DECLARE_BITMAP(allocated_irqs, NR_IRQS);
95
96#ifdef CONFIG_SPARSE_IRQ
97
98static RADIX_TREE(irq_desc_tree, GFP_KERNEL);
99
100static void irq_insert_desc(unsigned int irq, struct irq_desc *desc)
101{
102 radix_tree_insert(&irq_desc_tree, irq, desc);
103}
104
105struct irq_desc *irq_to_desc(unsigned int irq)
106{
107 return radix_tree_lookup(&irq_desc_tree, irq);
108}
109
110static void delete_irq_desc(unsigned int irq)
111{
112 radix_tree_delete(&irq_desc_tree, irq);
113}
114
115#ifdef CONFIG_SMP
116static void free_masks(struct irq_desc *desc)
117{
118#ifdef CONFIG_GENERIC_PENDING_IRQ
119 free_cpumask_var(desc->pending_mask);
120#endif
121 free_cpumask_var(desc->irq_data.affinity);
122}
123#else
124static inline void free_masks(struct irq_desc *desc) { }
125#endif
126
127static struct irq_desc *alloc_desc(int irq, int node)
128{
129 struct irq_desc *desc;
130 gfp_t gfp = GFP_KERNEL;
131
132 desc = kzalloc_node(sizeof(*desc), gfp, node);
133 if (!desc)
134 return NULL;
135 /* allocate based on nr_cpu_ids */
136 desc->kstat_irqs = kzalloc_node(nr_cpu_ids * sizeof(*desc->kstat_irqs),
137 gfp, node);
138 if (!desc->kstat_irqs)
139 goto err_desc;
140
141 if (alloc_masks(desc, gfp, node))
142 goto err_kstat;
143
144 raw_spin_lock_init(&desc->lock);
145 lockdep_set_class(&desc->lock, &irq_desc_lock_class);
146
147 desc_set_defaults(irq, desc, node);
148
149 return desc;
150
151err_kstat:
152 kfree(desc->kstat_irqs);
153err_desc:
154 kfree(desc);
155 return NULL;
156}
157
158static void free_desc(unsigned int irq)
159{
160 struct irq_desc *desc = irq_to_desc(irq);
161
162 unregister_irq_proc(irq, desc);
163
164 mutex_lock(&sparse_irq_lock);
165 delete_irq_desc(irq);
166 mutex_unlock(&sparse_irq_lock);
167
168 free_masks(desc);
169 kfree(desc->kstat_irqs);
170 kfree(desc);
171}
172
173static int alloc_descs(unsigned int start, unsigned int cnt, int node)
174{
175 struct irq_desc *desc;
176 int i;
177
178 for (i = 0; i < cnt; i++) {
179 desc = alloc_desc(start + i, node);
180 if (!desc)
181 goto err;
182 mutex_lock(&sparse_irq_lock);
183 irq_insert_desc(start + i, desc);
184 mutex_unlock(&sparse_irq_lock);
185 }
186 return start;
187
188err:
189 for (i--; i >= 0; i--)
190 free_desc(start + i);
191
192 mutex_lock(&sparse_irq_lock);
193 bitmap_clear(allocated_irqs, start, cnt);
194 mutex_unlock(&sparse_irq_lock);
195 return -ENOMEM;
196}
197
198struct irq_desc * __ref irq_to_desc_alloc_node(unsigned int irq, int node)
199{
200 int res = irq_alloc_descs(irq, irq, 1, node);
201
202 if (res == -EEXIST || res == irq)
203 return irq_to_desc(irq);
204 return NULL;
205}
206
207int __init early_irq_init(void)
208{
209 int i, initcnt, node = first_online_node;
210 struct irq_desc *desc;
211
212 init_irq_default_affinity();
213
214 /* Let arch update nr_irqs and return the nr of preallocated irqs */
215 initcnt = arch_probe_nr_irqs();
216 printk(KERN_INFO "NR_IRQS:%d nr_irqs:%d %d\n", NR_IRQS, nr_irqs, initcnt);
217
218 for (i = 0; i < initcnt; i++) {
219 desc = alloc_desc(i, node);
220 set_bit(i, allocated_irqs);
221 irq_insert_desc(i, desc);
222 }
223 return arch_early_irq_init();
224}
225
226#else /* !CONFIG_SPARSE_IRQ */
227
228struct irq_desc irq_desc[NR_IRQS] __cacheline_aligned_in_smp = {
229 [0 ... NR_IRQS-1] = {
230 .status = IRQ_DEFAULT_INIT_FLAGS,
231 .handle_irq = handle_bad_irq,
232 .depth = 1,
233 .lock = __RAW_SPIN_LOCK_UNLOCKED(irq_desc->lock),
234 }
235};
236
237static unsigned int kstat_irqs_all[NR_IRQS][NR_CPUS];
238int __init early_irq_init(void)
239{
240 int count, i, node = first_online_node;
241 struct irq_desc *desc;
242
243 init_irq_default_affinity();
244
245 printk(KERN_INFO "NR_IRQS:%d\n", NR_IRQS);
246
247 desc = irq_desc;
248 count = ARRAY_SIZE(irq_desc);
249
250 for (i = 0; i < count; i++) {
251 desc[i].irq_data.irq = i;
252 desc[i].irq_data.chip = &no_irq_chip;
253 desc[i].kstat_irqs = kstat_irqs_all[i];
254 alloc_masks(desc + i, GFP_KERNEL, node);
255 desc_smp_init(desc + i, node);
256 lockdep_set_class(&desc[i].lock, &irq_desc_lock_class);
257 }
258 return arch_early_irq_init();
259}
260
261struct irq_desc *irq_to_desc(unsigned int irq)
262{
263 return (irq < NR_IRQS) ? irq_desc + irq : NULL;
264}
265
266struct irq_desc *irq_to_desc_alloc_node(unsigned int irq, int node)
267{
268 return irq_to_desc(irq);
269}
270
271static void free_desc(unsigned int irq)
272{
273 dynamic_irq_cleanup(irq);
274}
275
276static inline int alloc_descs(unsigned int start, unsigned int cnt, int node)
277{
278 return start;
279}
280#endif /* !CONFIG_SPARSE_IRQ */
281
282/* Dynamic interrupt handling */
283
284/**
285 * irq_free_descs - free irq descriptors
286 * @from: Start of descriptor range
287 * @cnt: Number of consecutive irqs to free
288 */
289void irq_free_descs(unsigned int from, unsigned int cnt)
290{
291 int i;
292
293 if (from >= nr_irqs || (from + cnt) > nr_irqs)
294 return;
295
296 for (i = 0; i < cnt; i++)
297 free_desc(from + i);
298
299 mutex_lock(&sparse_irq_lock);
300 bitmap_clear(allocated_irqs, from, cnt);
301 mutex_unlock(&sparse_irq_lock);
302}
303
304/**
305 * irq_alloc_descs - allocate and initialize a range of irq descriptors
306 * @irq: Allocate for specific irq number if irq >= 0
307 * @from: Start the search from this irq number
308 * @cnt: Number of consecutive irqs to allocate.
309 * @node: Preferred node on which the irq descriptor should be allocated
310 *
311 * Returns the first irq number or error code
312 */
313int __ref
314irq_alloc_descs(int irq, unsigned int from, unsigned int cnt, int node)
315{
316 int start, ret;
317
318 if (!cnt)
319 return -EINVAL;
320
321 mutex_lock(&sparse_irq_lock);
322
323 start = bitmap_find_next_zero_area(allocated_irqs, nr_irqs, from, cnt, 0);
324 ret = -EEXIST;
325 if (irq >=0 && start != irq)
326 goto err;
327
328 ret = -ENOMEM;
329 if (start >= nr_irqs)
330 goto err;
331
332 bitmap_set(allocated_irqs, start, cnt);
333 mutex_unlock(&sparse_irq_lock);
334 return alloc_descs(start, cnt, node);
335
336err:
337 mutex_unlock(&sparse_irq_lock);
338 return ret;
339}
340
341/**
342 * irq_reserve_irqs - mark irqs allocated
343 * @from: mark from irq number
344 * @cnt: number of irqs to mark
345 *
346 * Returns 0 on success or an appropriate error code
347 */
348int irq_reserve_irqs(unsigned int from, unsigned int cnt)
349{
350 unsigned int start;
351 int ret = 0;
352
353 if (!cnt || (from + cnt) > nr_irqs)
354 return -EINVAL;
355
356 mutex_lock(&sparse_irq_lock);
357 start = bitmap_find_next_zero_area(allocated_irqs, nr_irqs, from, cnt, 0);
358 if (start == from)
359 bitmap_set(allocated_irqs, start, cnt);
360 else
361 ret = -EEXIST;
362 mutex_unlock(&sparse_irq_lock);
363 return ret;
364}
365
366/**
367 * irq_get_next_irq - get next allocated irq number
368 * @offset: where to start the search
369 *
370 * Returns next irq number after offset or nr_irqs if none is found.
371 */
372unsigned int irq_get_next_irq(unsigned int offset)
373{
374 return find_next_bit(allocated_irqs, nr_irqs, offset);
375}
376
377/**
378 * dynamic_irq_cleanup - cleanup a dynamically allocated irq
379 * @irq: irq number to initialize
380 */
381void dynamic_irq_cleanup(unsigned int irq)
382{
383 struct irq_desc *desc = irq_to_desc(irq);
384 unsigned long flags;
385
386 raw_spin_lock_irqsave(&desc->lock, flags);
387 desc_set_defaults(irq, desc, desc_node(desc));
388 raw_spin_unlock_irqrestore(&desc->lock, flags);
389}
390
391unsigned int kstat_irqs_cpu(unsigned int irq, int cpu)
392{
393 struct irq_desc *desc = irq_to_desc(irq);
394 return desc ? desc->kstat_irqs[cpu] : 0;
395}
diff --git a/kernel/irq/manage.c b/kernel/irq/manage.c
index c3003e9d91a3..644e8d5fa367 100644
--- a/kernel/irq/manage.c
+++ b/kernel/irq/manage.c
@@ -73,8 +73,8 @@ int irq_can_set_affinity(unsigned int irq)
73{ 73{
74 struct irq_desc *desc = irq_to_desc(irq); 74 struct irq_desc *desc = irq_to_desc(irq);
75 75
76 if (CHECK_IRQ_PER_CPU(desc->status) || !desc->chip || 76 if (CHECK_IRQ_PER_CPU(desc->status) || !desc->irq_data.chip ||
77 !desc->chip->set_affinity) 77 !desc->irq_data.chip->irq_set_affinity)
78 return 0; 78 return 0;
79 79
80 return 1; 80 return 1;
@@ -109,17 +109,18 @@ void irq_set_thread_affinity(struct irq_desc *desc)
109int irq_set_affinity(unsigned int irq, const struct cpumask *cpumask) 109int irq_set_affinity(unsigned int irq, const struct cpumask *cpumask)
110{ 110{
111 struct irq_desc *desc = irq_to_desc(irq); 111 struct irq_desc *desc = irq_to_desc(irq);
112 struct irq_chip *chip = desc->irq_data.chip;
112 unsigned long flags; 113 unsigned long flags;
113 114
114 if (!desc->chip->set_affinity) 115 if (!chip->irq_set_affinity)
115 return -EINVAL; 116 return -EINVAL;
116 117
117 raw_spin_lock_irqsave(&desc->lock, flags); 118 raw_spin_lock_irqsave(&desc->lock, flags);
118 119
119#ifdef CONFIG_GENERIC_PENDING_IRQ 120#ifdef CONFIG_GENERIC_PENDING_IRQ
120 if (desc->status & IRQ_MOVE_PCNTXT) { 121 if (desc->status & IRQ_MOVE_PCNTXT) {
121 if (!desc->chip->set_affinity(irq, cpumask)) { 122 if (!chip->irq_set_affinity(&desc->irq_data, cpumask, false)) {
122 cpumask_copy(desc->affinity, cpumask); 123 cpumask_copy(desc->irq_data.affinity, cpumask);
123 irq_set_thread_affinity(desc); 124 irq_set_thread_affinity(desc);
124 } 125 }
125 } 126 }
@@ -128,8 +129,8 @@ int irq_set_affinity(unsigned int irq, const struct cpumask *cpumask)
128 cpumask_copy(desc->pending_mask, cpumask); 129 cpumask_copy(desc->pending_mask, cpumask);
129 } 130 }
130#else 131#else
131 if (!desc->chip->set_affinity(irq, cpumask)) { 132 if (!chip->irq_set_affinity(&desc->irq_data, cpumask, false)) {
132 cpumask_copy(desc->affinity, cpumask); 133 cpumask_copy(desc->irq_data.affinity, cpumask);
133 irq_set_thread_affinity(desc); 134 irq_set_thread_affinity(desc);
134 } 135 }
135#endif 136#endif
@@ -168,16 +169,16 @@ static int setup_affinity(unsigned int irq, struct irq_desc *desc)
168 * one of the targets is online. 169 * one of the targets is online.
169 */ 170 */
170 if (desc->status & (IRQ_AFFINITY_SET | IRQ_NO_BALANCING)) { 171 if (desc->status & (IRQ_AFFINITY_SET | IRQ_NO_BALANCING)) {
171 if (cpumask_any_and(desc->affinity, cpu_online_mask) 172 if (cpumask_any_and(desc->irq_data.affinity, cpu_online_mask)
172 < nr_cpu_ids) 173 < nr_cpu_ids)
173 goto set_affinity; 174 goto set_affinity;
174 else 175 else
175 desc->status &= ~IRQ_AFFINITY_SET; 176 desc->status &= ~IRQ_AFFINITY_SET;
176 } 177 }
177 178
178 cpumask_and(desc->affinity, cpu_online_mask, irq_default_affinity); 179 cpumask_and(desc->irq_data.affinity, cpu_online_mask, irq_default_affinity);
179set_affinity: 180set_affinity:
180 desc->chip->set_affinity(irq, desc->affinity); 181 desc->irq_data.chip->irq_set_affinity(&desc->irq_data, desc->irq_data.affinity, false);
181 182
182 return 0; 183 return 0;
183} 184}
@@ -223,7 +224,7 @@ void __disable_irq(struct irq_desc *desc, unsigned int irq, bool suspend)
223 224
224 if (!desc->depth++) { 225 if (!desc->depth++) {
225 desc->status |= IRQ_DISABLED; 226 desc->status |= IRQ_DISABLED;
226 desc->chip->disable(irq); 227 desc->irq_data.chip->irq_disable(&desc->irq_data);
227 } 228 }
228} 229}
229 230
@@ -246,11 +247,11 @@ void disable_irq_nosync(unsigned int irq)
246 if (!desc) 247 if (!desc)
247 return; 248 return;
248 249
249 chip_bus_lock(irq, desc); 250 chip_bus_lock(desc);
250 raw_spin_lock_irqsave(&desc->lock, flags); 251 raw_spin_lock_irqsave(&desc->lock, flags);
251 __disable_irq(desc, irq, false); 252 __disable_irq(desc, irq, false);
252 raw_spin_unlock_irqrestore(&desc->lock, flags); 253 raw_spin_unlock_irqrestore(&desc->lock, flags);
253 chip_bus_sync_unlock(irq, desc); 254 chip_bus_sync_unlock(desc);
254} 255}
255EXPORT_SYMBOL(disable_irq_nosync); 256EXPORT_SYMBOL(disable_irq_nosync);
256 257
@@ -313,7 +314,7 @@ void __enable_irq(struct irq_desc *desc, unsigned int irq, bool resume)
313 * IRQ line is re-enabled. 314 * IRQ line is re-enabled.
314 * 315 *
315 * This function may be called from IRQ context only when 316 * This function may be called from IRQ context only when
316 * desc->chip->bus_lock and desc->chip->bus_sync_unlock are NULL ! 317 * desc->irq_data.chip->bus_lock and desc->chip->bus_sync_unlock are NULL !
317 */ 318 */
318void enable_irq(unsigned int irq) 319void enable_irq(unsigned int irq)
319{ 320{
@@ -323,11 +324,11 @@ void enable_irq(unsigned int irq)
323 if (!desc) 324 if (!desc)
324 return; 325 return;
325 326
326 chip_bus_lock(irq, desc); 327 chip_bus_lock(desc);
327 raw_spin_lock_irqsave(&desc->lock, flags); 328 raw_spin_lock_irqsave(&desc->lock, flags);
328 __enable_irq(desc, irq, false); 329 __enable_irq(desc, irq, false);
329 raw_spin_unlock_irqrestore(&desc->lock, flags); 330 raw_spin_unlock_irqrestore(&desc->lock, flags);
330 chip_bus_sync_unlock(irq, desc); 331 chip_bus_sync_unlock(desc);
331} 332}
332EXPORT_SYMBOL(enable_irq); 333EXPORT_SYMBOL(enable_irq);
333 334
@@ -336,8 +337,8 @@ static int set_irq_wake_real(unsigned int irq, unsigned int on)
336 struct irq_desc *desc = irq_to_desc(irq); 337 struct irq_desc *desc = irq_to_desc(irq);
337 int ret = -ENXIO; 338 int ret = -ENXIO;
338 339
339 if (desc->chip->set_wake) 340 if (desc->irq_data.chip->irq_set_wake)
340 ret = desc->chip->set_wake(irq, on); 341 ret = desc->irq_data.chip->irq_set_wake(&desc->irq_data, on);
341 342
342 return ret; 343 return ret;
343} 344}
@@ -429,12 +430,12 @@ void compat_irq_chip_set_default_handler(struct irq_desc *desc)
429} 430}
430 431
431int __irq_set_trigger(struct irq_desc *desc, unsigned int irq, 432int __irq_set_trigger(struct irq_desc *desc, unsigned int irq,
432 unsigned long flags) 433 unsigned long flags)
433{ 434{
434 int ret; 435 int ret;
435 struct irq_chip *chip = desc->chip; 436 struct irq_chip *chip = desc->irq_data.chip;
436 437
437 if (!chip || !chip->set_type) { 438 if (!chip || !chip->irq_set_type) {
438 /* 439 /*
439 * IRQF_TRIGGER_* but the PIC does not support multiple 440 * IRQF_TRIGGER_* but the PIC does not support multiple
440 * flow-types? 441 * flow-types?
@@ -445,11 +446,11 @@ int __irq_set_trigger(struct irq_desc *desc, unsigned int irq,
445 } 446 }
446 447
447 /* caller masked out all except trigger mode flags */ 448 /* caller masked out all except trigger mode flags */
448 ret = chip->set_type(irq, flags); 449 ret = chip->irq_set_type(&desc->irq_data, flags);
449 450
450 if (ret) 451 if (ret)
451 pr_err("setting trigger mode %d for irq %u failed (%pF)\n", 452 pr_err("setting trigger mode %lu for irq %u failed (%pF)\n",
452 (int)flags, irq, chip->set_type); 453 flags, irq, chip->irq_set_type);
453 else { 454 else {
454 if (flags & (IRQ_TYPE_LEVEL_LOW | IRQ_TYPE_LEVEL_HIGH)) 455 if (flags & (IRQ_TYPE_LEVEL_LOW | IRQ_TYPE_LEVEL_HIGH))
455 flags |= IRQ_LEVEL; 456 flags |= IRQ_LEVEL;
@@ -457,8 +458,8 @@ int __irq_set_trigger(struct irq_desc *desc, unsigned int irq,
457 desc->status &= ~(IRQ_LEVEL | IRQ_TYPE_SENSE_MASK); 458 desc->status &= ~(IRQ_LEVEL | IRQ_TYPE_SENSE_MASK);
458 desc->status |= flags; 459 desc->status |= flags;
459 460
460 if (chip != desc->chip) 461 if (chip != desc->irq_data.chip)
461 irq_chip_set_defaults(desc->chip); 462 irq_chip_set_defaults(desc->irq_data.chip);
462 } 463 }
463 464
464 return ret; 465 return ret;
@@ -507,7 +508,7 @@ static int irq_wait_for_interrupt(struct irqaction *action)
507static void irq_finalize_oneshot(unsigned int irq, struct irq_desc *desc) 508static void irq_finalize_oneshot(unsigned int irq, struct irq_desc *desc)
508{ 509{
509again: 510again:
510 chip_bus_lock(irq, desc); 511 chip_bus_lock(desc);
511 raw_spin_lock_irq(&desc->lock); 512 raw_spin_lock_irq(&desc->lock);
512 513
513 /* 514 /*
@@ -521,17 +522,17 @@ again:
521 */ 522 */
522 if (unlikely(desc->status & IRQ_INPROGRESS)) { 523 if (unlikely(desc->status & IRQ_INPROGRESS)) {
523 raw_spin_unlock_irq(&desc->lock); 524 raw_spin_unlock_irq(&desc->lock);
524 chip_bus_sync_unlock(irq, desc); 525 chip_bus_sync_unlock(desc);
525 cpu_relax(); 526 cpu_relax();
526 goto again; 527 goto again;
527 } 528 }
528 529
529 if (!(desc->status & IRQ_DISABLED) && (desc->status & IRQ_MASKED)) { 530 if (!(desc->status & IRQ_DISABLED) && (desc->status & IRQ_MASKED)) {
530 desc->status &= ~IRQ_MASKED; 531 desc->status &= ~IRQ_MASKED;
531 desc->chip->unmask(irq); 532 desc->irq_data.chip->irq_unmask(&desc->irq_data);
532 } 533 }
533 raw_spin_unlock_irq(&desc->lock); 534 raw_spin_unlock_irq(&desc->lock);
534 chip_bus_sync_unlock(irq, desc); 535 chip_bus_sync_unlock(desc);
535} 536}
536 537
537#ifdef CONFIG_SMP 538#ifdef CONFIG_SMP
@@ -556,7 +557,7 @@ irq_thread_check_affinity(struct irq_desc *desc, struct irqaction *action)
556 } 557 }
557 558
558 raw_spin_lock_irq(&desc->lock); 559 raw_spin_lock_irq(&desc->lock);
559 cpumask_copy(mask, desc->affinity); 560 cpumask_copy(mask, desc->irq_data.affinity);
560 raw_spin_unlock_irq(&desc->lock); 561 raw_spin_unlock_irq(&desc->lock);
561 562
562 set_cpus_allowed_ptr(current, mask); 563 set_cpus_allowed_ptr(current, mask);
@@ -657,7 +658,7 @@ __setup_irq(unsigned int irq, struct irq_desc *desc, struct irqaction *new)
657 if (!desc) 658 if (!desc)
658 return -EINVAL; 659 return -EINVAL;
659 660
660 if (desc->chip == &no_irq_chip) 661 if (desc->irq_data.chip == &no_irq_chip)
661 return -ENOSYS; 662 return -ENOSYS;
662 /* 663 /*
663 * Some drivers like serial.c use request_irq() heavily, 664 * Some drivers like serial.c use request_irq() heavily,
@@ -752,7 +753,7 @@ __setup_irq(unsigned int irq, struct irq_desc *desc, struct irqaction *new)
752 } 753 }
753 754
754 if (!shared) { 755 if (!shared) {
755 irq_chip_set_defaults(desc->chip); 756 irq_chip_set_defaults(desc->irq_data.chip);
756 757
757 init_waitqueue_head(&desc->wait_for_threads); 758 init_waitqueue_head(&desc->wait_for_threads);
758 759
@@ -779,7 +780,7 @@ __setup_irq(unsigned int irq, struct irq_desc *desc, struct irqaction *new)
779 if (!(desc->status & IRQ_NOAUTOEN)) { 780 if (!(desc->status & IRQ_NOAUTOEN)) {
780 desc->depth = 0; 781 desc->depth = 0;
781 desc->status &= ~IRQ_DISABLED; 782 desc->status &= ~IRQ_DISABLED;
782 desc->chip->startup(irq); 783 desc->irq_data.chip->irq_startup(&desc->irq_data);
783 } else 784 } else
784 /* Undo nested disables: */ 785 /* Undo nested disables: */
785 desc->depth = 1; 786 desc->depth = 1;
@@ -912,17 +913,17 @@ static struct irqaction *__free_irq(unsigned int irq, void *dev_id)
912 913
913 /* Currently used only by UML, might disappear one day: */ 914 /* Currently used only by UML, might disappear one day: */
914#ifdef CONFIG_IRQ_RELEASE_METHOD 915#ifdef CONFIG_IRQ_RELEASE_METHOD
915 if (desc->chip->release) 916 if (desc->irq_data.chip->release)
916 desc->chip->release(irq, dev_id); 917 desc->irq_data.chip->release(irq, dev_id);
917#endif 918#endif
918 919
919 /* If this was the last handler, shut down the IRQ line: */ 920 /* If this was the last handler, shut down the IRQ line: */
920 if (!desc->action) { 921 if (!desc->action) {
921 desc->status |= IRQ_DISABLED; 922 desc->status |= IRQ_DISABLED;
922 if (desc->chip->shutdown) 923 if (desc->irq_data.chip->irq_shutdown)
923 desc->chip->shutdown(irq); 924 desc->irq_data.chip->irq_shutdown(&desc->irq_data);
924 else 925 else
925 desc->chip->disable(irq); 926 desc->irq_data.chip->irq_disable(&desc->irq_data);
926 } 927 }
927 928
928#ifdef CONFIG_SMP 929#ifdef CONFIG_SMP
@@ -997,9 +998,9 @@ void free_irq(unsigned int irq, void *dev_id)
997 if (!desc) 998 if (!desc)
998 return; 999 return;
999 1000
1000 chip_bus_lock(irq, desc); 1001 chip_bus_lock(desc);
1001 kfree(__free_irq(irq, dev_id)); 1002 kfree(__free_irq(irq, dev_id));
1002 chip_bus_sync_unlock(irq, desc); 1003 chip_bus_sync_unlock(desc);
1003} 1004}
1004EXPORT_SYMBOL(free_irq); 1005EXPORT_SYMBOL(free_irq);
1005 1006
@@ -1086,9 +1087,9 @@ int request_threaded_irq(unsigned int irq, irq_handler_t handler,
1086 action->name = devname; 1087 action->name = devname;
1087 action->dev_id = dev_id; 1088 action->dev_id = dev_id;
1088 1089
1089 chip_bus_lock(irq, desc); 1090 chip_bus_lock(desc);
1090 retval = __setup_irq(irq, desc, action); 1091 retval = __setup_irq(irq, desc, action);
1091 chip_bus_sync_unlock(irq, desc); 1092 chip_bus_sync_unlock(desc);
1092 1093
1093 if (retval) 1094 if (retval)
1094 kfree(action); 1095 kfree(action);
diff --git a/kernel/irq/migration.c b/kernel/irq/migration.c
index 241962280836..1d2541940480 100644
--- a/kernel/irq/migration.c
+++ b/kernel/irq/migration.c
@@ -7,6 +7,7 @@
7void move_masked_irq(int irq) 7void move_masked_irq(int irq)
8{ 8{
9 struct irq_desc *desc = irq_to_desc(irq); 9 struct irq_desc *desc = irq_to_desc(irq);
10 struct irq_chip *chip = desc->irq_data.chip;
10 11
11 if (likely(!(desc->status & IRQ_MOVE_PENDING))) 12 if (likely(!(desc->status & IRQ_MOVE_PENDING)))
12 return; 13 return;
@@ -24,7 +25,7 @@ void move_masked_irq(int irq)
24 if (unlikely(cpumask_empty(desc->pending_mask))) 25 if (unlikely(cpumask_empty(desc->pending_mask)))
25 return; 26 return;
26 27
27 if (!desc->chip->set_affinity) 28 if (!chip->irq_set_affinity)
28 return; 29 return;
29 30
30 assert_raw_spin_locked(&desc->lock); 31 assert_raw_spin_locked(&desc->lock);
@@ -43,8 +44,9 @@ void move_masked_irq(int irq)
43 */ 44 */
44 if (likely(cpumask_any_and(desc->pending_mask, cpu_online_mask) 45 if (likely(cpumask_any_and(desc->pending_mask, cpu_online_mask)
45 < nr_cpu_ids)) 46 < nr_cpu_ids))
46 if (!desc->chip->set_affinity(irq, desc->pending_mask)) { 47 if (!chip->irq_set_affinity(&desc->irq_data,
47 cpumask_copy(desc->affinity, desc->pending_mask); 48 desc->pending_mask, false)) {
49 cpumask_copy(desc->irq_data.affinity, desc->pending_mask);
48 irq_set_thread_affinity(desc); 50 irq_set_thread_affinity(desc);
49 } 51 }
50 52
@@ -61,8 +63,8 @@ void move_native_irq(int irq)
61 if (unlikely(desc->status & IRQ_DISABLED)) 63 if (unlikely(desc->status & IRQ_DISABLED))
62 return; 64 return;
63 65
64 desc->chip->mask(irq); 66 desc->irq_data.chip->irq_mask(&desc->irq_data);
65 move_masked_irq(irq); 67 move_masked_irq(irq);
66 desc->chip->unmask(irq); 68 desc->irq_data.chip->irq_unmask(&desc->irq_data);
67} 69}
68 70
diff --git a/kernel/irq/numa_migrate.c b/kernel/irq/numa_migrate.c
deleted file mode 100644
index 65d3845665ac..000000000000
--- a/kernel/irq/numa_migrate.c
+++ /dev/null
@@ -1,120 +0,0 @@
1/*
2 * NUMA irq-desc migration code
3 *
4 * Migrate IRQ data structures (irq_desc, chip_data, etc.) over to
5 * the new "home node" of the IRQ.
6 */
7
8#include <linux/irq.h>
9#include <linux/slab.h>
10#include <linux/module.h>
11#include <linux/random.h>
12#include <linux/interrupt.h>
13#include <linux/kernel_stat.h>
14
15#include "internals.h"
16
17static void init_copy_kstat_irqs(struct irq_desc *old_desc,
18 struct irq_desc *desc,
19 int node, int nr)
20{
21 init_kstat_irqs(desc, node, nr);
22
23 if (desc->kstat_irqs != old_desc->kstat_irqs)
24 memcpy(desc->kstat_irqs, old_desc->kstat_irqs,
25 nr * sizeof(*desc->kstat_irqs));
26}
27
28static void free_kstat_irqs(struct irq_desc *old_desc, struct irq_desc *desc)
29{
30 if (old_desc->kstat_irqs == desc->kstat_irqs)
31 return;
32
33 kfree(old_desc->kstat_irqs);
34 old_desc->kstat_irqs = NULL;
35}
36
37static bool init_copy_one_irq_desc(int irq, struct irq_desc *old_desc,
38 struct irq_desc *desc, int node)
39{
40 memcpy(desc, old_desc, sizeof(struct irq_desc));
41 if (!alloc_desc_masks(desc, node, false)) {
42 printk(KERN_ERR "irq %d: can not get new irq_desc cpumask "
43 "for migration.\n", irq);
44 return false;
45 }
46 raw_spin_lock_init(&desc->lock);
47 desc->node = node;
48 lockdep_set_class(&desc->lock, &irq_desc_lock_class);
49 init_copy_kstat_irqs(old_desc, desc, node, nr_cpu_ids);
50 init_copy_desc_masks(old_desc, desc);
51 arch_init_copy_chip_data(old_desc, desc, node);
52 return true;
53}
54
55static void free_one_irq_desc(struct irq_desc *old_desc, struct irq_desc *desc)
56{
57 free_kstat_irqs(old_desc, desc);
58 free_desc_masks(old_desc, desc);
59 arch_free_chip_data(old_desc, desc);
60}
61
62static struct irq_desc *__real_move_irq_desc(struct irq_desc *old_desc,
63 int node)
64{
65 struct irq_desc *desc;
66 unsigned int irq;
67 unsigned long flags;
68
69 irq = old_desc->irq;
70
71 raw_spin_lock_irqsave(&sparse_irq_lock, flags);
72
73 /* We have to check it to avoid races with another CPU */
74 desc = irq_to_desc(irq);
75
76 if (desc && old_desc != desc)
77 goto out_unlock;
78
79 desc = kzalloc_node(sizeof(*desc), GFP_ATOMIC, node);
80 if (!desc) {
81 printk(KERN_ERR "irq %d: can not get new irq_desc "
82 "for migration.\n", irq);
83 /* still use old one */
84 desc = old_desc;
85 goto out_unlock;
86 }
87 if (!init_copy_one_irq_desc(irq, old_desc, desc, node)) {
88 /* still use old one */
89 kfree(desc);
90 desc = old_desc;
91 goto out_unlock;
92 }
93
94 replace_irq_desc(irq, desc);
95 raw_spin_unlock_irqrestore(&sparse_irq_lock, flags);
96
97 /* free the old one */
98 free_one_irq_desc(old_desc, desc);
99 kfree(old_desc);
100
101 return desc;
102
103out_unlock:
104 raw_spin_unlock_irqrestore(&sparse_irq_lock, flags);
105
106 return desc;
107}
108
109struct irq_desc *move_irq_desc(struct irq_desc *desc, int node)
110{
111 /* those static or target node is -1, do not move them */
112 if (desc->irq < NR_IRQS_LEGACY || node == -1)
113 return desc;
114
115 if (desc->node != node)
116 desc = __real_move_irq_desc(desc, node);
117
118 return desc;
119}
120
diff --git a/kernel/irq/proc.c b/kernel/irq/proc.c
index 09a2ee540bd2..01b1d3a88983 100644
--- a/kernel/irq/proc.c
+++ b/kernel/irq/proc.c
@@ -21,7 +21,7 @@ static struct proc_dir_entry *root_irq_dir;
21static int irq_affinity_proc_show(struct seq_file *m, void *v) 21static int irq_affinity_proc_show(struct seq_file *m, void *v)
22{ 22{
23 struct irq_desc *desc = irq_to_desc((long)m->private); 23 struct irq_desc *desc = irq_to_desc((long)m->private);
24 const struct cpumask *mask = desc->affinity; 24 const struct cpumask *mask = desc->irq_data.affinity;
25 25
26#ifdef CONFIG_GENERIC_PENDING_IRQ 26#ifdef CONFIG_GENERIC_PENDING_IRQ
27 if (desc->status & IRQ_MOVE_PENDING) 27 if (desc->status & IRQ_MOVE_PENDING)
@@ -65,7 +65,7 @@ static ssize_t irq_affinity_proc_write(struct file *file,
65 cpumask_var_t new_value; 65 cpumask_var_t new_value;
66 int err; 66 int err;
67 67
68 if (!irq_to_desc(irq)->chip->set_affinity || no_irq_affinity || 68 if (!irq_to_desc(irq)->irq_data.chip->irq_set_affinity || no_irq_affinity ||
69 irq_balancing_disabled(irq)) 69 irq_balancing_disabled(irq))
70 return -EIO; 70 return -EIO;
71 71
@@ -185,7 +185,7 @@ static int irq_node_proc_show(struct seq_file *m, void *v)
185{ 185{
186 struct irq_desc *desc = irq_to_desc((long) m->private); 186 struct irq_desc *desc = irq_to_desc((long) m->private);
187 187
188 seq_printf(m, "%d\n", desc->node); 188 seq_printf(m, "%d\n", desc->irq_data.node);
189 return 0; 189 return 0;
190} 190}
191 191
@@ -269,7 +269,7 @@ void register_irq_proc(unsigned int irq, struct irq_desc *desc)
269{ 269{
270 char name [MAX_NAMELEN]; 270 char name [MAX_NAMELEN];
271 271
272 if (!root_irq_dir || (desc->chip == &no_irq_chip) || desc->dir) 272 if (!root_irq_dir || (desc->irq_data.chip == &no_irq_chip) || desc->dir)
273 return; 273 return;
274 274
275 memset(name, 0, MAX_NAMELEN); 275 memset(name, 0, MAX_NAMELEN);
@@ -297,6 +297,24 @@ void register_irq_proc(unsigned int irq, struct irq_desc *desc)
297 &irq_spurious_proc_fops, (void *)(long)irq); 297 &irq_spurious_proc_fops, (void *)(long)irq);
298} 298}
299 299
300void unregister_irq_proc(unsigned int irq, struct irq_desc *desc)
301{
302 char name [MAX_NAMELEN];
303
304 if (!root_irq_dir || !desc->dir)
305 return;
306#ifdef CONFIG_SMP
307 remove_proc_entry("smp_affinity", desc->dir);
308 remove_proc_entry("affinity_hint", desc->dir);
309 remove_proc_entry("node", desc->dir);
310#endif
311 remove_proc_entry("spurious", desc->dir);
312
313 memset(name, 0, MAX_NAMELEN);
314 sprintf(name, "%u", irq);
315 remove_proc_entry(name, root_irq_dir);
316}
317
300#undef MAX_NAMELEN 318#undef MAX_NAMELEN
301 319
302void unregister_handler_proc(unsigned int irq, struct irqaction *action) 320void unregister_handler_proc(unsigned int irq, struct irqaction *action)
diff --git a/kernel/irq/resend.c b/kernel/irq/resend.c
index 090c3763f3a2..891115a929aa 100644
--- a/kernel/irq/resend.c
+++ b/kernel/irq/resend.c
@@ -60,7 +60,7 @@ void check_irq_resend(struct irq_desc *desc, unsigned int irq)
60 /* 60 /*
61 * Make sure the interrupt is enabled, before resending it: 61 * Make sure the interrupt is enabled, before resending it:
62 */ 62 */
63 desc->chip->enable(irq); 63 desc->irq_data.chip->irq_enable(&desc->irq_data);
64 64
65 /* 65 /*
66 * We do not resend level type interrupts. Level type 66 * We do not resend level type interrupts. Level type
@@ -70,7 +70,8 @@ void check_irq_resend(struct irq_desc *desc, unsigned int irq)
70 if ((status & (IRQ_LEVEL | IRQ_PENDING | IRQ_REPLAY)) == IRQ_PENDING) { 70 if ((status & (IRQ_LEVEL | IRQ_PENDING | IRQ_REPLAY)) == IRQ_PENDING) {
71 desc->status = (status & ~IRQ_PENDING) | IRQ_REPLAY; 71 desc->status = (status & ~IRQ_PENDING) | IRQ_REPLAY;
72 72
73 if (!desc->chip->retrigger || !desc->chip->retrigger(irq)) { 73 if (!desc->irq_data.chip->irq_retrigger ||
74 !desc->irq_data.chip->irq_retrigger(&desc->irq_data)) {
74#ifdef CONFIG_HARDIRQS_SW_RESEND 75#ifdef CONFIG_HARDIRQS_SW_RESEND
75 /* Set it pending and activate the softirq: */ 76 /* Set it pending and activate the softirq: */
76 set_bit(irq, irqs_resend); 77 set_bit(irq, irqs_resend);
diff --git a/kernel/irq/spurious.c b/kernel/irq/spurious.c
index 89fb90ae534f..3089d3b9d5f3 100644
--- a/kernel/irq/spurious.c
+++ b/kernel/irq/spurious.c
@@ -14,6 +14,8 @@
14#include <linux/moduleparam.h> 14#include <linux/moduleparam.h>
15#include <linux/timer.h> 15#include <linux/timer.h>
16 16
17#include "internals.h"
18
17static int irqfixup __read_mostly; 19static int irqfixup __read_mostly;
18 20
19#define POLL_SPURIOUS_IRQ_INTERVAL (HZ/10) 21#define POLL_SPURIOUS_IRQ_INTERVAL (HZ/10)
@@ -78,8 +80,8 @@ static int try_one_irq(int irq, struct irq_desc *desc)
78 * If we did actual work for the real IRQ line we must let the 80 * If we did actual work for the real IRQ line we must let the
79 * IRQ controller clean up too 81 * IRQ controller clean up too
80 */ 82 */
81 if (work && desc->chip && desc->chip->end) 83 if (work)
82 desc->chip->end(irq); 84 irq_end(irq, desc);
83 raw_spin_unlock(&desc->lock); 85 raw_spin_unlock(&desc->lock);
84 86
85 return ok; 87 return ok;
@@ -254,7 +256,7 @@ void note_interrupt(unsigned int irq, struct irq_desc *desc,
254 printk(KERN_EMERG "Disabling IRQ #%d\n", irq); 256 printk(KERN_EMERG "Disabling IRQ #%d\n", irq);
255 desc->status |= IRQ_DISABLED | IRQ_SPURIOUS_DISABLED; 257 desc->status |= IRQ_DISABLED | IRQ_SPURIOUS_DISABLED;
256 desc->depth++; 258 desc->depth++;
257 desc->chip->disable(irq); 259 desc->irq_data.chip->irq_disable(&desc->irq_data);
258 260
259 mod_timer(&poll_spurious_irq_timer, 261 mod_timer(&poll_spurious_irq_timer,
260 jiffies + POLL_SPURIOUS_IRQ_INTERVAL); 262 jiffies + POLL_SPURIOUS_IRQ_INTERVAL);
diff --git a/kernel/irq_work.c b/kernel/irq_work.c
new file mode 100644
index 000000000000..f16763ff8481
--- /dev/null
+++ b/kernel/irq_work.c
@@ -0,0 +1,164 @@
1/*
2 * Copyright (C) 2010 Red Hat, Inc., Peter Zijlstra <pzijlstr@redhat.com>
3 *
4 * Provides a framework for enqueueing and running callbacks from hardirq
5 * context. The enqueueing is NMI-safe.
6 */
7
8#include <linux/kernel.h>
9#include <linux/module.h>
10#include <linux/irq_work.h>
11#include <linux/hardirq.h>
12
13/*
14 * An entry can be in one of four states:
15 *
16 * free NULL, 0 -> {claimed} : free to be used
17 * claimed NULL, 3 -> {pending} : claimed to be enqueued
18 * pending next, 3 -> {busy} : queued, pending callback
19 * busy NULL, 2 -> {free, claimed} : callback in progress, can be claimed
20 *
21 * We use the lower two bits of the next pointer to keep PENDING and BUSY
22 * flags.
23 */
24
25#define IRQ_WORK_PENDING 1UL
26#define IRQ_WORK_BUSY 2UL
27#define IRQ_WORK_FLAGS 3UL
28
29static inline bool irq_work_is_set(struct irq_work *entry, int flags)
30{
31 return (unsigned long)entry->next & flags;
32}
33
34static inline struct irq_work *irq_work_next(struct irq_work *entry)
35{
36 unsigned long next = (unsigned long)entry->next;
37 next &= ~IRQ_WORK_FLAGS;
38 return (struct irq_work *)next;
39}
40
41static inline struct irq_work *next_flags(struct irq_work *entry, int flags)
42{
43 unsigned long next = (unsigned long)entry;
44 next |= flags;
45 return (struct irq_work *)next;
46}
47
48static DEFINE_PER_CPU(struct irq_work *, irq_work_list);
49
50/*
51 * Claim the entry so that no one else will poke at it.
52 */
53static bool irq_work_claim(struct irq_work *entry)
54{
55 struct irq_work *next, *nflags;
56
57 do {
58 next = entry->next;
59 if ((unsigned long)next & IRQ_WORK_PENDING)
60 return false;
61 nflags = next_flags(next, IRQ_WORK_FLAGS);
62 } while (cmpxchg(&entry->next, next, nflags) != next);
63
64 return true;
65}
66
67
68void __weak arch_irq_work_raise(void)
69{
70 /*
71 * Lame architectures will get the timer tick callback
72 */
73}
74
75/*
76 * Queue the entry and raise the IPI if needed.
77 */
78static void __irq_work_queue(struct irq_work *entry)
79{
80 struct irq_work **head, *next;
81
82 head = &get_cpu_var(irq_work_list);
83
84 do {
85 next = *head;
86 /* Can assign non-atomic because we keep the flags set. */
87 entry->next = next_flags(next, IRQ_WORK_FLAGS);
88 } while (cmpxchg(head, next, entry) != next);
89
90 /* The list was empty, raise self-interrupt to start processing. */
91 if (!irq_work_next(entry))
92 arch_irq_work_raise();
93
94 put_cpu_var(irq_work_list);
95}
96
97/*
98 * Enqueue the irq_work @entry, returns true on success, failure when the
99 * @entry was already enqueued by someone else.
100 *
101 * Can be re-enqueued while the callback is still in progress.
102 */
103bool irq_work_queue(struct irq_work *entry)
104{
105 if (!irq_work_claim(entry)) {
106 /*
107 * Already enqueued, can't do!
108 */
109 return false;
110 }
111
112 __irq_work_queue(entry);
113 return true;
114}
115EXPORT_SYMBOL_GPL(irq_work_queue);
116
117/*
118 * Run the irq_work entries on this cpu. Requires to be ran from hardirq
119 * context with local IRQs disabled.
120 */
121void irq_work_run(void)
122{
123 struct irq_work *list, **head;
124
125 head = &__get_cpu_var(irq_work_list);
126 if (*head == NULL)
127 return;
128
129 BUG_ON(!in_irq());
130 BUG_ON(!irqs_disabled());
131
132 list = xchg(head, NULL);
133 while (list != NULL) {
134 struct irq_work *entry = list;
135
136 list = irq_work_next(list);
137
138 /*
139 * Clear the PENDING bit, after this point the @entry
140 * can be re-used.
141 */
142 entry->next = next_flags(NULL, IRQ_WORK_BUSY);
143 entry->func(entry);
144 /*
145 * Clear the BUSY bit and return to the free state if
146 * no-one else claimed it meanwhile.
147 */
148 cmpxchg(&entry->next, next_flags(NULL, IRQ_WORK_BUSY), NULL);
149 }
150}
151EXPORT_SYMBOL_GPL(irq_work_run);
152
153/*
154 * Synchronize against the irq_work @entry, ensures the entry is not
155 * currently in use.
156 */
157void irq_work_sync(struct irq_work *entry)
158{
159 WARN_ON_ONCE(irqs_disabled());
160
161 while (irq_work_is_set(entry, IRQ_WORK_BUSY))
162 cpu_relax();
163}
164EXPORT_SYMBOL_GPL(irq_work_sync);
diff --git a/kernel/jump_label.c b/kernel/jump_label.c
new file mode 100644
index 000000000000..7be868bf25c6
--- /dev/null
+++ b/kernel/jump_label.c
@@ -0,0 +1,429 @@
1/*
2 * jump label support
3 *
4 * Copyright (C) 2009 Jason Baron <jbaron@redhat.com>
5 *
6 */
7#include <linux/jump_label.h>
8#include <linux/memory.h>
9#include <linux/uaccess.h>
10#include <linux/module.h>
11#include <linux/list.h>
12#include <linux/jhash.h>
13#include <linux/slab.h>
14#include <linux/sort.h>
15#include <linux/err.h>
16
17#ifdef HAVE_JUMP_LABEL
18
19#define JUMP_LABEL_HASH_BITS 6
20#define JUMP_LABEL_TABLE_SIZE (1 << JUMP_LABEL_HASH_BITS)
21static struct hlist_head jump_label_table[JUMP_LABEL_TABLE_SIZE];
22
23/* mutex to protect coming/going of the the jump_label table */
24static DEFINE_MUTEX(jump_label_mutex);
25
26struct jump_label_entry {
27 struct hlist_node hlist;
28 struct jump_entry *table;
29 int nr_entries;
30 /* hang modules off here */
31 struct hlist_head modules;
32 unsigned long key;
33};
34
35struct jump_label_module_entry {
36 struct hlist_node hlist;
37 struct jump_entry *table;
38 int nr_entries;
39 struct module *mod;
40};
41
42static int jump_label_cmp(const void *a, const void *b)
43{
44 const struct jump_entry *jea = a;
45 const struct jump_entry *jeb = b;
46
47 if (jea->key < jeb->key)
48 return -1;
49
50 if (jea->key > jeb->key)
51 return 1;
52
53 return 0;
54}
55
56static void
57sort_jump_label_entries(struct jump_entry *start, struct jump_entry *stop)
58{
59 unsigned long size;
60
61 size = (((unsigned long)stop - (unsigned long)start)
62 / sizeof(struct jump_entry));
63 sort(start, size, sizeof(struct jump_entry), jump_label_cmp, NULL);
64}
65
66static struct jump_label_entry *get_jump_label_entry(jump_label_t key)
67{
68 struct hlist_head *head;
69 struct hlist_node *node;
70 struct jump_label_entry *e;
71 u32 hash = jhash((void *)&key, sizeof(jump_label_t), 0);
72
73 head = &jump_label_table[hash & (JUMP_LABEL_TABLE_SIZE - 1)];
74 hlist_for_each_entry(e, node, head, hlist) {
75 if (key == e->key)
76 return e;
77 }
78 return NULL;
79}
80
81static struct jump_label_entry *
82add_jump_label_entry(jump_label_t key, int nr_entries, struct jump_entry *table)
83{
84 struct hlist_head *head;
85 struct jump_label_entry *e;
86 u32 hash;
87
88 e = get_jump_label_entry(key);
89 if (e)
90 return ERR_PTR(-EEXIST);
91
92 e = kmalloc(sizeof(struct jump_label_entry), GFP_KERNEL);
93 if (!e)
94 return ERR_PTR(-ENOMEM);
95
96 hash = jhash((void *)&key, sizeof(jump_label_t), 0);
97 head = &jump_label_table[hash & (JUMP_LABEL_TABLE_SIZE - 1)];
98 e->key = key;
99 e->table = table;
100 e->nr_entries = nr_entries;
101 INIT_HLIST_HEAD(&(e->modules));
102 hlist_add_head(&e->hlist, head);
103 return e;
104}
105
106static int
107build_jump_label_hashtable(struct jump_entry *start, struct jump_entry *stop)
108{
109 struct jump_entry *iter, *iter_begin;
110 struct jump_label_entry *entry;
111 int count;
112
113 sort_jump_label_entries(start, stop);
114 iter = start;
115 while (iter < stop) {
116 entry = get_jump_label_entry(iter->key);
117 if (!entry) {
118 iter_begin = iter;
119 count = 0;
120 while ((iter < stop) &&
121 (iter->key == iter_begin->key)) {
122 iter++;
123 count++;
124 }
125 entry = add_jump_label_entry(iter_begin->key,
126 count, iter_begin);
127 if (IS_ERR(entry))
128 return PTR_ERR(entry);
129 } else {
130 WARN_ONCE(1, KERN_ERR "build_jump_hashtable: unexpected entry!\n");
131 return -1;
132 }
133 }
134 return 0;
135}
136
137/***
138 * jump_label_update - update jump label text
139 * @key - key value associated with a a jump label
140 * @type - enum set to JUMP_LABEL_ENABLE or JUMP_LABEL_DISABLE
141 *
142 * Will enable/disable the jump for jump label @key, depending on the
143 * value of @type.
144 *
145 */
146
147void jump_label_update(unsigned long key, enum jump_label_type type)
148{
149 struct jump_entry *iter;
150 struct jump_label_entry *entry;
151 struct hlist_node *module_node;
152 struct jump_label_module_entry *e_module;
153 int count;
154
155 mutex_lock(&jump_label_mutex);
156 entry = get_jump_label_entry((jump_label_t)key);
157 if (entry) {
158 count = entry->nr_entries;
159 iter = entry->table;
160 while (count--) {
161 if (kernel_text_address(iter->code))
162 arch_jump_label_transform(iter, type);
163 iter++;
164 }
165 /* eanble/disable jump labels in modules */
166 hlist_for_each_entry(e_module, module_node, &(entry->modules),
167 hlist) {
168 count = e_module->nr_entries;
169 iter = e_module->table;
170 while (count--) {
171 if (kernel_text_address(iter->code))
172 arch_jump_label_transform(iter, type);
173 iter++;
174 }
175 }
176 }
177 mutex_unlock(&jump_label_mutex);
178}
179
180static int addr_conflict(struct jump_entry *entry, void *start, void *end)
181{
182 if (entry->code <= (unsigned long)end &&
183 entry->code + JUMP_LABEL_NOP_SIZE > (unsigned long)start)
184 return 1;
185
186 return 0;
187}
188
189#ifdef CONFIG_MODULES
190
191static int module_conflict(void *start, void *end)
192{
193 struct hlist_head *head;
194 struct hlist_node *node, *node_next, *module_node, *module_node_next;
195 struct jump_label_entry *e;
196 struct jump_label_module_entry *e_module;
197 struct jump_entry *iter;
198 int i, count;
199 int conflict = 0;
200
201 for (i = 0; i < JUMP_LABEL_TABLE_SIZE; i++) {
202 head = &jump_label_table[i];
203 hlist_for_each_entry_safe(e, node, node_next, head, hlist) {
204 hlist_for_each_entry_safe(e_module, module_node,
205 module_node_next,
206 &(e->modules), hlist) {
207 count = e_module->nr_entries;
208 iter = e_module->table;
209 while (count--) {
210 if (addr_conflict(iter, start, end)) {
211 conflict = 1;
212 goto out;
213 }
214 iter++;
215 }
216 }
217 }
218 }
219out:
220 return conflict;
221}
222
223#endif
224
225/***
226 * jump_label_text_reserved - check if addr range is reserved
227 * @start: start text addr
228 * @end: end text addr
229 *
230 * checks if the text addr located between @start and @end
231 * overlaps with any of the jump label patch addresses. Code
232 * that wants to modify kernel text should first verify that
233 * it does not overlap with any of the jump label addresses.
234 *
235 * returns 1 if there is an overlap, 0 otherwise
236 */
237int jump_label_text_reserved(void *start, void *end)
238{
239 struct jump_entry *iter;
240 struct jump_entry *iter_start = __start___jump_table;
241 struct jump_entry *iter_stop = __start___jump_table;
242 int conflict = 0;
243
244 mutex_lock(&jump_label_mutex);
245 iter = iter_start;
246 while (iter < iter_stop) {
247 if (addr_conflict(iter, start, end)) {
248 conflict = 1;
249 goto out;
250 }
251 iter++;
252 }
253
254 /* now check modules */
255#ifdef CONFIG_MODULES
256 conflict = module_conflict(start, end);
257#endif
258out:
259 mutex_unlock(&jump_label_mutex);
260 return conflict;
261}
262
263static __init int init_jump_label(void)
264{
265 int ret;
266 struct jump_entry *iter_start = __start___jump_table;
267 struct jump_entry *iter_stop = __stop___jump_table;
268 struct jump_entry *iter;
269
270 mutex_lock(&jump_label_mutex);
271 ret = build_jump_label_hashtable(__start___jump_table,
272 __stop___jump_table);
273 iter = iter_start;
274 while (iter < iter_stop) {
275 arch_jump_label_text_poke_early(iter->code);
276 iter++;
277 }
278 mutex_unlock(&jump_label_mutex);
279 return ret;
280}
281early_initcall(init_jump_label);
282
283#ifdef CONFIG_MODULES
284
285static struct jump_label_module_entry *
286add_jump_label_module_entry(struct jump_label_entry *entry,
287 struct jump_entry *iter_begin,
288 int count, struct module *mod)
289{
290 struct jump_label_module_entry *e;
291
292 e = kmalloc(sizeof(struct jump_label_module_entry), GFP_KERNEL);
293 if (!e)
294 return ERR_PTR(-ENOMEM);
295 e->mod = mod;
296 e->nr_entries = count;
297 e->table = iter_begin;
298 hlist_add_head(&e->hlist, &entry->modules);
299 return e;
300}
301
302static int add_jump_label_module(struct module *mod)
303{
304 struct jump_entry *iter, *iter_begin;
305 struct jump_label_entry *entry;
306 struct jump_label_module_entry *module_entry;
307 int count;
308
309 /* if the module doesn't have jump label entries, just return */
310 if (!mod->num_jump_entries)
311 return 0;
312
313 sort_jump_label_entries(mod->jump_entries,
314 mod->jump_entries + mod->num_jump_entries);
315 iter = mod->jump_entries;
316 while (iter < mod->jump_entries + mod->num_jump_entries) {
317 entry = get_jump_label_entry(iter->key);
318 iter_begin = iter;
319 count = 0;
320 while ((iter < mod->jump_entries + mod->num_jump_entries) &&
321 (iter->key == iter_begin->key)) {
322 iter++;
323 count++;
324 }
325 if (!entry) {
326 entry = add_jump_label_entry(iter_begin->key, 0, NULL);
327 if (IS_ERR(entry))
328 return PTR_ERR(entry);
329 }
330 module_entry = add_jump_label_module_entry(entry, iter_begin,
331 count, mod);
332 if (IS_ERR(module_entry))
333 return PTR_ERR(module_entry);
334 }
335 return 0;
336}
337
338static void remove_jump_label_module(struct module *mod)
339{
340 struct hlist_head *head;
341 struct hlist_node *node, *node_next, *module_node, *module_node_next;
342 struct jump_label_entry *e;
343 struct jump_label_module_entry *e_module;
344 int i;
345
346 /* if the module doesn't have jump label entries, just return */
347 if (!mod->num_jump_entries)
348 return;
349
350 for (i = 0; i < JUMP_LABEL_TABLE_SIZE; i++) {
351 head = &jump_label_table[i];
352 hlist_for_each_entry_safe(e, node, node_next, head, hlist) {
353 hlist_for_each_entry_safe(e_module, module_node,
354 module_node_next,
355 &(e->modules), hlist) {
356 if (e_module->mod == mod) {
357 hlist_del(&e_module->hlist);
358 kfree(e_module);
359 }
360 }
361 if (hlist_empty(&e->modules) && (e->nr_entries == 0)) {
362 hlist_del(&e->hlist);
363 kfree(e);
364 }
365 }
366 }
367}
368
369static int
370jump_label_module_notify(struct notifier_block *self, unsigned long val,
371 void *data)
372{
373 struct module *mod = data;
374 int ret = 0;
375
376 switch (val) {
377 case MODULE_STATE_COMING:
378 mutex_lock(&jump_label_mutex);
379 ret = add_jump_label_module(mod);
380 if (ret)
381 remove_jump_label_module(mod);
382 mutex_unlock(&jump_label_mutex);
383 break;
384 case MODULE_STATE_GOING:
385 mutex_lock(&jump_label_mutex);
386 remove_jump_label_module(mod);
387 mutex_unlock(&jump_label_mutex);
388 break;
389 }
390 return ret;
391}
392
393/***
394 * apply_jump_label_nops - patch module jump labels with arch_get_jump_label_nop()
395 * @mod: module to patch
396 *
397 * Allow for run-time selection of the optimal nops. Before the module
398 * loads patch these with arch_get_jump_label_nop(), which is specified by
399 * the arch specific jump label code.
400 */
401void jump_label_apply_nops(struct module *mod)
402{
403 struct jump_entry *iter;
404
405 /* if the module doesn't have jump label entries, just return */
406 if (!mod->num_jump_entries)
407 return;
408
409 iter = mod->jump_entries;
410 while (iter < mod->jump_entries + mod->num_jump_entries) {
411 arch_jump_label_text_poke_early(iter->code);
412 iter++;
413 }
414}
415
416struct notifier_block jump_label_module_nb = {
417 .notifier_call = jump_label_module_notify,
418 .priority = 0,
419};
420
421static __init int init_jump_label_module(void)
422{
423 return register_module_notifier(&jump_label_module_nb);
424}
425early_initcall(init_jump_label_module);
426
427#endif /* CONFIG_MODULES */
428
429#endif
diff --git a/kernel/kfifo.c b/kernel/kfifo.c
index 6b5580c57644..01a0700e873f 100644
--- a/kernel/kfifo.c
+++ b/kernel/kfifo.c
@@ -365,8 +365,6 @@ static unsigned int setup_sgl(struct __kfifo *fifo, struct scatterlist *sgl,
365 n = setup_sgl_buf(sgl, fifo->data + off, nents, l); 365 n = setup_sgl_buf(sgl, fifo->data + off, nents, l);
366 n += setup_sgl_buf(sgl + n, fifo->data, nents - n, len - l); 366 n += setup_sgl_buf(sgl + n, fifo->data, nents - n, len - l);
367 367
368 if (n)
369 sg_mark_end(sgl + n - 1);
370 return n; 368 return n;
371} 369}
372 370
diff --git a/kernel/kprobes.c b/kernel/kprobes.c
index 282035f3ae96..56a891914273 100644
--- a/kernel/kprobes.c
+++ b/kernel/kprobes.c
@@ -47,6 +47,7 @@
47#include <linux/memory.h> 47#include <linux/memory.h>
48#include <linux/ftrace.h> 48#include <linux/ftrace.h>
49#include <linux/cpu.h> 49#include <linux/cpu.h>
50#include <linux/jump_label.h>
50 51
51#include <asm-generic/sections.h> 52#include <asm-generic/sections.h>
52#include <asm/cacheflush.h> 53#include <asm/cacheflush.h>
@@ -399,7 +400,7 @@ static inline int kprobe_optready(struct kprobe *p)
399 * Return an optimized kprobe whose optimizing code replaces 400 * Return an optimized kprobe whose optimizing code replaces
400 * instructions including addr (exclude breakpoint). 401 * instructions including addr (exclude breakpoint).
401 */ 402 */
402struct kprobe *__kprobes get_optimized_kprobe(unsigned long addr) 403static struct kprobe *__kprobes get_optimized_kprobe(unsigned long addr)
403{ 404{
404 int i; 405 int i;
405 struct kprobe *p = NULL; 406 struct kprobe *p = NULL;
@@ -831,6 +832,7 @@ void __kprobes recycle_rp_inst(struct kretprobe_instance *ri,
831 832
832void __kprobes kretprobe_hash_lock(struct task_struct *tsk, 833void __kprobes kretprobe_hash_lock(struct task_struct *tsk,
833 struct hlist_head **head, unsigned long *flags) 834 struct hlist_head **head, unsigned long *flags)
835__acquires(hlist_lock)
834{ 836{
835 unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS); 837 unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS);
836 spinlock_t *hlist_lock; 838 spinlock_t *hlist_lock;
@@ -842,6 +844,7 @@ void __kprobes kretprobe_hash_lock(struct task_struct *tsk,
842 844
843static void __kprobes kretprobe_table_lock(unsigned long hash, 845static void __kprobes kretprobe_table_lock(unsigned long hash,
844 unsigned long *flags) 846 unsigned long *flags)
847__acquires(hlist_lock)
845{ 848{
846 spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash); 849 spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash);
847 spin_lock_irqsave(hlist_lock, *flags); 850 spin_lock_irqsave(hlist_lock, *flags);
@@ -849,6 +852,7 @@ static void __kprobes kretprobe_table_lock(unsigned long hash,
849 852
850void __kprobes kretprobe_hash_unlock(struct task_struct *tsk, 853void __kprobes kretprobe_hash_unlock(struct task_struct *tsk,
851 unsigned long *flags) 854 unsigned long *flags)
855__releases(hlist_lock)
852{ 856{
853 unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS); 857 unsigned long hash = hash_ptr(tsk, KPROBE_HASH_BITS);
854 spinlock_t *hlist_lock; 858 spinlock_t *hlist_lock;
@@ -857,7 +861,9 @@ void __kprobes kretprobe_hash_unlock(struct task_struct *tsk,
857 spin_unlock_irqrestore(hlist_lock, *flags); 861 spin_unlock_irqrestore(hlist_lock, *flags);
858} 862}
859 863
860void __kprobes kretprobe_table_unlock(unsigned long hash, unsigned long *flags) 864static void __kprobes kretprobe_table_unlock(unsigned long hash,
865 unsigned long *flags)
866__releases(hlist_lock)
861{ 867{
862 spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash); 868 spinlock_t *hlist_lock = kretprobe_table_lock_ptr(hash);
863 spin_unlock_irqrestore(hlist_lock, *flags); 869 spin_unlock_irqrestore(hlist_lock, *flags);
@@ -1141,7 +1147,8 @@ int __kprobes register_kprobe(struct kprobe *p)
1141 preempt_disable(); 1147 preempt_disable();
1142 if (!kernel_text_address((unsigned long) p->addr) || 1148 if (!kernel_text_address((unsigned long) p->addr) ||
1143 in_kprobes_functions((unsigned long) p->addr) || 1149 in_kprobes_functions((unsigned long) p->addr) ||
1144 ftrace_text_reserved(p->addr, p->addr)) { 1150 ftrace_text_reserved(p->addr, p->addr) ||
1151 jump_label_text_reserved(p->addr, p->addr)) {
1145 preempt_enable(); 1152 preempt_enable();
1146 return -EINVAL; 1153 return -EINVAL;
1147 } 1154 }
@@ -1339,18 +1346,19 @@ int __kprobes register_jprobes(struct jprobe **jps, int num)
1339 if (num <= 0) 1346 if (num <= 0)
1340 return -EINVAL; 1347 return -EINVAL;
1341 for (i = 0; i < num; i++) { 1348 for (i = 0; i < num; i++) {
1342 unsigned long addr; 1349 unsigned long addr, offset;
1343 jp = jps[i]; 1350 jp = jps[i];
1344 addr = arch_deref_entry_point(jp->entry); 1351 addr = arch_deref_entry_point(jp->entry);
1345 1352
1346 if (!kernel_text_address(addr)) 1353 /* Verify probepoint is a function entry point */
1347 ret = -EINVAL; 1354 if (kallsyms_lookup_size_offset(addr, NULL, &offset) &&
1348 else { 1355 offset == 0) {
1349 /* Todo: Verify probepoint is a function entry point */
1350 jp->kp.pre_handler = setjmp_pre_handler; 1356 jp->kp.pre_handler = setjmp_pre_handler;
1351 jp->kp.break_handler = longjmp_break_handler; 1357 jp->kp.break_handler = longjmp_break_handler;
1352 ret = register_kprobe(&jp->kp); 1358 ret = register_kprobe(&jp->kp);
1353 } 1359 } else
1360 ret = -EINVAL;
1361
1354 if (ret < 0) { 1362 if (ret < 0) {
1355 if (i > 0) 1363 if (i > 0)
1356 unregister_jprobes(jps, i); 1364 unregister_jprobes(jps, i);
@@ -1992,6 +2000,7 @@ static ssize_t write_enabled_file_bool(struct file *file,
1992static const struct file_operations fops_kp = { 2000static const struct file_operations fops_kp = {
1993 .read = read_enabled_file_bool, 2001 .read = read_enabled_file_bool,
1994 .write = write_enabled_file_bool, 2002 .write = write_enabled_file_bool,
2003 .llseek = default_llseek,
1995}; 2004};
1996 2005
1997static int __kprobes debugfs_kprobe_init(void) 2006static int __kprobes debugfs_kprobe_init(void)
diff --git a/kernel/lockdep.c b/kernel/lockdep.c
index f2852a510232..42ba65dff7d9 100644
--- a/kernel/lockdep.c
+++ b/kernel/lockdep.c
@@ -639,6 +639,16 @@ look_up_lock_class(struct lockdep_map *lock, unsigned int subclass)
639 } 639 }
640#endif 640#endif
641 641
642 if (unlikely(subclass >= MAX_LOCKDEP_SUBCLASSES)) {
643 debug_locks_off();
644 printk(KERN_ERR
645 "BUG: looking up invalid subclass: %u\n", subclass);
646 printk(KERN_ERR
647 "turning off the locking correctness validator.\n");
648 dump_stack();
649 return NULL;
650 }
651
642 /* 652 /*
643 * Static locks do not have their class-keys yet - for them the key 653 * Static locks do not have their class-keys yet - for them the key
644 * is the lock object itself: 654 * is the lock object itself:
@@ -774,7 +784,9 @@ out_unlock_set:
774 raw_local_irq_restore(flags); 784 raw_local_irq_restore(flags);
775 785
776 if (!subclass || force) 786 if (!subclass || force)
777 lock->class_cache = class; 787 lock->class_cache[0] = class;
788 else if (subclass < NR_LOCKDEP_CACHING_CLASSES)
789 lock->class_cache[subclass] = class;
778 790
779 if (DEBUG_LOCKS_WARN_ON(class->subclass != subclass)) 791 if (DEBUG_LOCKS_WARN_ON(class->subclass != subclass))
780 return NULL; 792 return NULL;
@@ -2679,7 +2691,11 @@ static int mark_lock(struct task_struct *curr, struct held_lock *this,
2679void lockdep_init_map(struct lockdep_map *lock, const char *name, 2691void lockdep_init_map(struct lockdep_map *lock, const char *name,
2680 struct lock_class_key *key, int subclass) 2692 struct lock_class_key *key, int subclass)
2681{ 2693{
2682 lock->class_cache = NULL; 2694 int i;
2695
2696 for (i = 0; i < NR_LOCKDEP_CACHING_CLASSES; i++)
2697 lock->class_cache[i] = NULL;
2698
2683#ifdef CONFIG_LOCK_STAT 2699#ifdef CONFIG_LOCK_STAT
2684 lock->cpu = raw_smp_processor_id(); 2700 lock->cpu = raw_smp_processor_id();
2685#endif 2701#endif
@@ -2739,21 +2755,13 @@ static int __lock_acquire(struct lockdep_map *lock, unsigned int subclass,
2739 if (DEBUG_LOCKS_WARN_ON(!irqs_disabled())) 2755 if (DEBUG_LOCKS_WARN_ON(!irqs_disabled()))
2740 return 0; 2756 return 0;
2741 2757
2742 if (unlikely(subclass >= MAX_LOCKDEP_SUBCLASSES)) {
2743 debug_locks_off();
2744 printk("BUG: MAX_LOCKDEP_SUBCLASSES too low!\n");
2745 printk("turning off the locking correctness validator.\n");
2746 dump_stack();
2747 return 0;
2748 }
2749
2750 if (lock->key == &__lockdep_no_validate__) 2758 if (lock->key == &__lockdep_no_validate__)
2751 check = 1; 2759 check = 1;
2752 2760
2753 if (!subclass) 2761 if (subclass < NR_LOCKDEP_CACHING_CLASSES)
2754 class = lock->class_cache; 2762 class = lock->class_cache[subclass];
2755 /* 2763 /*
2756 * Not cached yet or subclass? 2764 * Not cached?
2757 */ 2765 */
2758 if (unlikely(!class)) { 2766 if (unlikely(!class)) {
2759 class = register_lock_class(lock, subclass, 0); 2767 class = register_lock_class(lock, subclass, 0);
@@ -2918,7 +2926,7 @@ static int match_held_lock(struct held_lock *hlock, struct lockdep_map *lock)
2918 return 1; 2926 return 1;
2919 2927
2920 if (hlock->references) { 2928 if (hlock->references) {
2921 struct lock_class *class = lock->class_cache; 2929 struct lock_class *class = lock->class_cache[0];
2922 2930
2923 if (!class) 2931 if (!class)
2924 class = look_up_lock_class(lock, 0); 2932 class = look_up_lock_class(lock, 0);
@@ -3559,7 +3567,12 @@ void lockdep_reset_lock(struct lockdep_map *lock)
3559 if (list_empty(head)) 3567 if (list_empty(head))
3560 continue; 3568 continue;
3561 list_for_each_entry_safe(class, next, head, hash_entry) { 3569 list_for_each_entry_safe(class, next, head, hash_entry) {
3562 if (unlikely(class == lock->class_cache)) { 3570 int match = 0;
3571
3572 for (j = 0; j < NR_LOCKDEP_CACHING_CLASSES; j++)
3573 match |= class == lock->class_cache[j];
3574
3575 if (unlikely(match)) {
3563 if (debug_locks_off_graph_unlock()) 3576 if (debug_locks_off_graph_unlock())
3564 WARN_ON(1); 3577 WARN_ON(1);
3565 goto out_restore; 3578 goto out_restore;
@@ -3775,7 +3788,7 @@ EXPORT_SYMBOL_GPL(debug_show_all_locks);
3775 * Careful: only use this function if you are sure that 3788 * Careful: only use this function if you are sure that
3776 * the task cannot run in parallel! 3789 * the task cannot run in parallel!
3777 */ 3790 */
3778void __debug_show_held_locks(struct task_struct *task) 3791void debug_show_held_locks(struct task_struct *task)
3779{ 3792{
3780 if (unlikely(!debug_locks)) { 3793 if (unlikely(!debug_locks)) {
3781 printk("INFO: lockdep is turned off.\n"); 3794 printk("INFO: lockdep is turned off.\n");
@@ -3783,12 +3796,6 @@ void __debug_show_held_locks(struct task_struct *task)
3783 } 3796 }
3784 lockdep_print_held_locks(task); 3797 lockdep_print_held_locks(task);
3785} 3798}
3786EXPORT_SYMBOL_GPL(__debug_show_held_locks);
3787
3788void debug_show_held_locks(struct task_struct *task)
3789{
3790 __debug_show_held_locks(task);
3791}
3792EXPORT_SYMBOL_GPL(debug_show_held_locks); 3799EXPORT_SYMBOL_GPL(debug_show_held_locks);
3793 3800
3794void lockdep_sys_exit(void) 3801void lockdep_sys_exit(void)
diff --git a/kernel/module.c b/kernel/module.c
index d0b5f8db11b4..2df46301a7a4 100644
--- a/kernel/module.c
+++ b/kernel/module.c
@@ -55,6 +55,7 @@
55#include <linux/async.h> 55#include <linux/async.h>
56#include <linux/percpu.h> 56#include <linux/percpu.h>
57#include <linux/kmemleak.h> 57#include <linux/kmemleak.h>
58#include <linux/jump_label.h>
58 59
59#define CREATE_TRACE_POINTS 60#define CREATE_TRACE_POINTS
60#include <trace/events/module.h> 61#include <trace/events/module.h>
@@ -1537,6 +1538,7 @@ static int __unlink_module(void *_mod)
1537{ 1538{
1538 struct module *mod = _mod; 1539 struct module *mod = _mod;
1539 list_del(&mod->list); 1540 list_del(&mod->list);
1541 module_bug_cleanup(mod);
1540 return 0; 1542 return 0;
1541} 1543}
1542 1544
@@ -2308,6 +2310,11 @@ static void find_module_sections(struct module *mod, struct load_info *info)
2308 sizeof(*mod->tracepoints), 2310 sizeof(*mod->tracepoints),
2309 &mod->num_tracepoints); 2311 &mod->num_tracepoints);
2310#endif 2312#endif
2313#ifdef HAVE_JUMP_LABEL
2314 mod->jump_entries = section_objs(info, "__jump_table",
2315 sizeof(*mod->jump_entries),
2316 &mod->num_jump_entries);
2317#endif
2311#ifdef CONFIG_EVENT_TRACING 2318#ifdef CONFIG_EVENT_TRACING
2312 mod->trace_events = section_objs(info, "_ftrace_events", 2319 mod->trace_events = section_objs(info, "_ftrace_events",
2313 sizeof(*mod->trace_events), 2320 sizeof(*mod->trace_events),
@@ -2625,6 +2632,7 @@ static struct module *load_module(void __user *umod,
2625 if (err < 0) 2632 if (err < 0)
2626 goto ddebug; 2633 goto ddebug;
2627 2634
2635 module_bug_finalize(info.hdr, info.sechdrs, mod);
2628 list_add_rcu(&mod->list, &modules); 2636 list_add_rcu(&mod->list, &modules);
2629 mutex_unlock(&module_mutex); 2637 mutex_unlock(&module_mutex);
2630 2638
@@ -2650,6 +2658,8 @@ static struct module *load_module(void __user *umod,
2650 mutex_lock(&module_mutex); 2658 mutex_lock(&module_mutex);
2651 /* Unlink carefully: kallsyms could be walking list. */ 2659 /* Unlink carefully: kallsyms could be walking list. */
2652 list_del_rcu(&mod->list); 2660 list_del_rcu(&mod->list);
2661 module_bug_cleanup(mod);
2662
2653 ddebug: 2663 ddebug:
2654 if (!mod->taints) 2664 if (!mod->taints)
2655 dynamic_debug_remove(info.debug); 2665 dynamic_debug_remove(info.debug);
diff --git a/kernel/perf_event.c b/kernel/perf_event.c
index db5b56064687..f309e8014c78 100644
--- a/kernel/perf_event.c
+++ b/kernel/perf_event.c
@@ -31,24 +31,18 @@
31#include <linux/kernel_stat.h> 31#include <linux/kernel_stat.h>
32#include <linux/perf_event.h> 32#include <linux/perf_event.h>
33#include <linux/ftrace_event.h> 33#include <linux/ftrace_event.h>
34#include <linux/hw_breakpoint.h>
35 34
36#include <asm/irq_regs.h> 35#include <asm/irq_regs.h>
37 36
38/* 37atomic_t perf_task_events __read_mostly;
39 * Each CPU has a list of per CPU events:
40 */
41static DEFINE_PER_CPU(struct perf_cpu_context, perf_cpu_context);
42
43int perf_max_events __read_mostly = 1;
44static int perf_reserved_percpu __read_mostly;
45static int perf_overcommit __read_mostly = 1;
46
47static atomic_t nr_events __read_mostly;
48static atomic_t nr_mmap_events __read_mostly; 38static atomic_t nr_mmap_events __read_mostly;
49static atomic_t nr_comm_events __read_mostly; 39static atomic_t nr_comm_events __read_mostly;
50static atomic_t nr_task_events __read_mostly; 40static atomic_t nr_task_events __read_mostly;
51 41
42static LIST_HEAD(pmus);
43static DEFINE_MUTEX(pmus_lock);
44static struct srcu_struct pmus_srcu;
45
52/* 46/*
53 * perf event paranoia level: 47 * perf event paranoia level:
54 * -1 - not paranoid at all 48 * -1 - not paranoid at all
@@ -67,36 +61,43 @@ int sysctl_perf_event_sample_rate __read_mostly = 100000;
67 61
68static atomic64_t perf_event_id; 62static atomic64_t perf_event_id;
69 63
70/* 64void __weak perf_event_print_debug(void) { }
71 * Lock for (sysadmin-configurable) event reservations:
72 */
73static DEFINE_SPINLOCK(perf_resource_lock);
74 65
75/* 66extern __weak const char *perf_pmu_name(void)
76 * Architecture provided APIs - weak aliases:
77 */
78extern __weak const struct pmu *hw_perf_event_init(struct perf_event *event)
79{ 67{
80 return NULL; 68 return "pmu";
81} 69}
82 70
83void __weak hw_perf_disable(void) { barrier(); } 71void perf_pmu_disable(struct pmu *pmu)
84void __weak hw_perf_enable(void) { barrier(); } 72{
85 73 int *count = this_cpu_ptr(pmu->pmu_disable_count);
86void __weak perf_event_print_debug(void) { } 74 if (!(*count)++)
87 75 pmu->pmu_disable(pmu);
88static DEFINE_PER_CPU(int, perf_disable_count); 76}
89 77
90void perf_disable(void) 78void perf_pmu_enable(struct pmu *pmu)
91{ 79{
92 if (!__get_cpu_var(perf_disable_count)++) 80 int *count = this_cpu_ptr(pmu->pmu_disable_count);
93 hw_perf_disable(); 81 if (!--(*count))
82 pmu->pmu_enable(pmu);
94} 83}
95 84
96void perf_enable(void) 85static DEFINE_PER_CPU(struct list_head, rotation_list);
86
87/*
88 * perf_pmu_rotate_start() and perf_rotate_context() are fully serialized
89 * because they're strictly cpu affine and rotate_start is called with IRQs
90 * disabled, while rotate_context is called from IRQ context.
91 */
92static void perf_pmu_rotate_start(struct pmu *pmu)
97{ 93{
98 if (!--__get_cpu_var(perf_disable_count)) 94 struct perf_cpu_context *cpuctx = this_cpu_ptr(pmu->pmu_cpu_context);
99 hw_perf_enable(); 95 struct list_head *head = &__get_cpu_var(rotation_list);
96
97 WARN_ON(!irqs_disabled());
98
99 if (list_empty(&cpuctx->rotation_list))
100 list_add(&cpuctx->rotation_list, head);
100} 101}
101 102
102static void get_ctx(struct perf_event_context *ctx) 103static void get_ctx(struct perf_event_context *ctx)
@@ -151,13 +152,13 @@ static u64 primary_event_id(struct perf_event *event)
151 * the context could get moved to another task. 152 * the context could get moved to another task.
152 */ 153 */
153static struct perf_event_context * 154static struct perf_event_context *
154perf_lock_task_context(struct task_struct *task, unsigned long *flags) 155perf_lock_task_context(struct task_struct *task, int ctxn, unsigned long *flags)
155{ 156{
156 struct perf_event_context *ctx; 157 struct perf_event_context *ctx;
157 158
158 rcu_read_lock(); 159 rcu_read_lock();
159 retry: 160retry:
160 ctx = rcu_dereference(task->perf_event_ctxp); 161 ctx = rcu_dereference(task->perf_event_ctxp[ctxn]);
161 if (ctx) { 162 if (ctx) {
162 /* 163 /*
163 * If this context is a clone of another, it might 164 * If this context is a clone of another, it might
@@ -170,7 +171,7 @@ perf_lock_task_context(struct task_struct *task, unsigned long *flags)
170 * can't get swapped on us any more. 171 * can't get swapped on us any more.
171 */ 172 */
172 raw_spin_lock_irqsave(&ctx->lock, *flags); 173 raw_spin_lock_irqsave(&ctx->lock, *flags);
173 if (ctx != rcu_dereference(task->perf_event_ctxp)) { 174 if (ctx != rcu_dereference(task->perf_event_ctxp[ctxn])) {
174 raw_spin_unlock_irqrestore(&ctx->lock, *flags); 175 raw_spin_unlock_irqrestore(&ctx->lock, *flags);
175 goto retry; 176 goto retry;
176 } 177 }
@@ -189,12 +190,13 @@ perf_lock_task_context(struct task_struct *task, unsigned long *flags)
189 * can't get swapped to another task. This also increments its 190 * can't get swapped to another task. This also increments its
190 * reference count so that the context can't get freed. 191 * reference count so that the context can't get freed.
191 */ 192 */
192static struct perf_event_context *perf_pin_task_context(struct task_struct *task) 193static struct perf_event_context *
194perf_pin_task_context(struct task_struct *task, int ctxn)
193{ 195{
194 struct perf_event_context *ctx; 196 struct perf_event_context *ctx;
195 unsigned long flags; 197 unsigned long flags;
196 198
197 ctx = perf_lock_task_context(task, &flags); 199 ctx = perf_lock_task_context(task, ctxn, &flags);
198 if (ctx) { 200 if (ctx) {
199 ++ctx->pin_count; 201 ++ctx->pin_count;
200 raw_spin_unlock_irqrestore(&ctx->lock, flags); 202 raw_spin_unlock_irqrestore(&ctx->lock, flags);
@@ -302,6 +304,8 @@ list_add_event(struct perf_event *event, struct perf_event_context *ctx)
302 } 304 }
303 305
304 list_add_rcu(&event->event_entry, &ctx->event_list); 306 list_add_rcu(&event->event_entry, &ctx->event_list);
307 if (!ctx->nr_events)
308 perf_pmu_rotate_start(ctx->pmu);
305 ctx->nr_events++; 309 ctx->nr_events++;
306 if (event->attr.inherit_stat) 310 if (event->attr.inherit_stat)
307 ctx->nr_stat++; 311 ctx->nr_stat++;
@@ -311,7 +315,12 @@ static void perf_group_attach(struct perf_event *event)
311{ 315{
312 struct perf_event *group_leader = event->group_leader; 316 struct perf_event *group_leader = event->group_leader;
313 317
314 WARN_ON_ONCE(event->attach_state & PERF_ATTACH_GROUP); 318 /*
319 * We can have double attach due to group movement in perf_event_open.
320 */
321 if (event->attach_state & PERF_ATTACH_GROUP)
322 return;
323
315 event->attach_state |= PERF_ATTACH_GROUP; 324 event->attach_state |= PERF_ATTACH_GROUP;
316 325
317 if (group_leader == event) 326 if (group_leader == event)
@@ -408,8 +417,8 @@ event_filter_match(struct perf_event *event)
408 return event->cpu == -1 || event->cpu == smp_processor_id(); 417 return event->cpu == -1 || event->cpu == smp_processor_id();
409} 418}
410 419
411static void 420static int
412event_sched_out(struct perf_event *event, 421__event_sched_out(struct perf_event *event,
413 struct perf_cpu_context *cpuctx, 422 struct perf_cpu_context *cpuctx,
414 struct perf_event_context *ctx) 423 struct perf_event_context *ctx)
415{ 424{
@@ -428,15 +437,14 @@ event_sched_out(struct perf_event *event,
428 } 437 }
429 438
430 if (event->state != PERF_EVENT_STATE_ACTIVE) 439 if (event->state != PERF_EVENT_STATE_ACTIVE)
431 return; 440 return 0;
432 441
433 event->state = PERF_EVENT_STATE_INACTIVE; 442 event->state = PERF_EVENT_STATE_INACTIVE;
434 if (event->pending_disable) { 443 if (event->pending_disable) {
435 event->pending_disable = 0; 444 event->pending_disable = 0;
436 event->state = PERF_EVENT_STATE_OFF; 445 event->state = PERF_EVENT_STATE_OFF;
437 } 446 }
438 event->tstamp_stopped = ctx->time; 447 event->pmu->del(event, 0);
439 event->pmu->disable(event);
440 event->oncpu = -1; 448 event->oncpu = -1;
441 449
442 if (!is_software_event(event)) 450 if (!is_software_event(event))
@@ -444,6 +452,19 @@ event_sched_out(struct perf_event *event,
444 ctx->nr_active--; 452 ctx->nr_active--;
445 if (event->attr.exclusive || !cpuctx->active_oncpu) 453 if (event->attr.exclusive || !cpuctx->active_oncpu)
446 cpuctx->exclusive = 0; 454 cpuctx->exclusive = 0;
455 return 1;
456}
457
458static void
459event_sched_out(struct perf_event *event,
460 struct perf_cpu_context *cpuctx,
461 struct perf_event_context *ctx)
462{
463 int ret;
464
465 ret = __event_sched_out(event, cpuctx, ctx);
466 if (ret)
467 event->tstamp_stopped = ctx->time;
447} 468}
448 469
449static void 470static void
@@ -466,6 +487,12 @@ group_sched_out(struct perf_event *group_event,
466 cpuctx->exclusive = 0; 487 cpuctx->exclusive = 0;
467} 488}
468 489
490static inline struct perf_cpu_context *
491__get_cpu_context(struct perf_event_context *ctx)
492{
493 return this_cpu_ptr(ctx->pmu->pmu_cpu_context);
494}
495
469/* 496/*
470 * Cross CPU call to remove a performance event 497 * Cross CPU call to remove a performance event
471 * 498 *
@@ -474,9 +501,9 @@ group_sched_out(struct perf_event *group_event,
474 */ 501 */
475static void __perf_event_remove_from_context(void *info) 502static void __perf_event_remove_from_context(void *info)
476{ 503{
477 struct perf_cpu_context *cpuctx = &__get_cpu_var(perf_cpu_context);
478 struct perf_event *event = info; 504 struct perf_event *event = info;
479 struct perf_event_context *ctx = event->ctx; 505 struct perf_event_context *ctx = event->ctx;
506 struct perf_cpu_context *cpuctx = __get_cpu_context(ctx);
480 507
481 /* 508 /*
482 * If this is a task context, we need to check whether it is 509 * If this is a task context, we need to check whether it is
@@ -487,27 +514,11 @@ static void __perf_event_remove_from_context(void *info)
487 return; 514 return;
488 515
489 raw_spin_lock(&ctx->lock); 516 raw_spin_lock(&ctx->lock);
490 /*
491 * Protect the list operation against NMI by disabling the
492 * events on a global level.
493 */
494 perf_disable();
495 517
496 event_sched_out(event, cpuctx, ctx); 518 event_sched_out(event, cpuctx, ctx);
497 519
498 list_del_event(event, ctx); 520 list_del_event(event, ctx);
499 521
500 if (!ctx->task) {
501 /*
502 * Allow more per task events with respect to the
503 * reservation:
504 */
505 cpuctx->max_pertask =
506 min(perf_max_events - ctx->nr_events,
507 perf_max_events - perf_reserved_percpu);
508 }
509
510 perf_enable();
511 raw_spin_unlock(&ctx->lock); 522 raw_spin_unlock(&ctx->lock);
512} 523}
513 524
@@ -572,8 +583,8 @@ retry:
572static void __perf_event_disable(void *info) 583static void __perf_event_disable(void *info)
573{ 584{
574 struct perf_event *event = info; 585 struct perf_event *event = info;
575 struct perf_cpu_context *cpuctx = &__get_cpu_var(perf_cpu_context);
576 struct perf_event_context *ctx = event->ctx; 586 struct perf_event_context *ctx = event->ctx;
587 struct perf_cpu_context *cpuctx = __get_cpu_context(ctx);
577 588
578 /* 589 /*
579 * If this is a per-task event, need to check whether this 590 * If this is a per-task event, need to check whether this
@@ -628,7 +639,7 @@ void perf_event_disable(struct perf_event *event)
628 return; 639 return;
629 } 640 }
630 641
631 retry: 642retry:
632 task_oncpu_function_call(task, __perf_event_disable, event); 643 task_oncpu_function_call(task, __perf_event_disable, event);
633 644
634 raw_spin_lock_irq(&ctx->lock); 645 raw_spin_lock_irq(&ctx->lock);
@@ -653,7 +664,7 @@ void perf_event_disable(struct perf_event *event)
653} 664}
654 665
655static int 666static int
656event_sched_in(struct perf_event *event, 667__event_sched_in(struct perf_event *event,
657 struct perf_cpu_context *cpuctx, 668 struct perf_cpu_context *cpuctx,
658 struct perf_event_context *ctx) 669 struct perf_event_context *ctx)
659{ 670{
@@ -667,14 +678,12 @@ event_sched_in(struct perf_event *event,
667 */ 678 */
668 smp_wmb(); 679 smp_wmb();
669 680
670 if (event->pmu->enable(event)) { 681 if (event->pmu->add(event, PERF_EF_START)) {
671 event->state = PERF_EVENT_STATE_INACTIVE; 682 event->state = PERF_EVENT_STATE_INACTIVE;
672 event->oncpu = -1; 683 event->oncpu = -1;
673 return -EAGAIN; 684 return -EAGAIN;
674 } 685 }
675 686
676 event->tstamp_running += ctx->time - event->tstamp_stopped;
677
678 if (!is_software_event(event)) 687 if (!is_software_event(event))
679 cpuctx->active_oncpu++; 688 cpuctx->active_oncpu++;
680 ctx->nr_active++; 689 ctx->nr_active++;
@@ -685,28 +694,56 @@ event_sched_in(struct perf_event *event,
685 return 0; 694 return 0;
686} 695}
687 696
697static inline int
698event_sched_in(struct perf_event *event,
699 struct perf_cpu_context *cpuctx,
700 struct perf_event_context *ctx)
701{
702 int ret = __event_sched_in(event, cpuctx, ctx);
703 if (ret)
704 return ret;
705 event->tstamp_running += ctx->time - event->tstamp_stopped;
706 return 0;
707}
708
709static void
710group_commit_event_sched_in(struct perf_event *group_event,
711 struct perf_cpu_context *cpuctx,
712 struct perf_event_context *ctx)
713{
714 struct perf_event *event;
715 u64 now = ctx->time;
716
717 group_event->tstamp_running += now - group_event->tstamp_stopped;
718 /*
719 * Schedule in siblings as one group (if any):
720 */
721 list_for_each_entry(event, &group_event->sibling_list, group_entry) {
722 event->tstamp_running += now - event->tstamp_stopped;
723 }
724}
725
688static int 726static int
689group_sched_in(struct perf_event *group_event, 727group_sched_in(struct perf_event *group_event,
690 struct perf_cpu_context *cpuctx, 728 struct perf_cpu_context *cpuctx,
691 struct perf_event_context *ctx) 729 struct perf_event_context *ctx)
692{ 730{
693 struct perf_event *event, *partial_group = NULL; 731 struct perf_event *event, *partial_group = NULL;
694 const struct pmu *pmu = group_event->pmu; 732 struct pmu *pmu = group_event->pmu;
695 bool txn = false;
696 733
697 if (group_event->state == PERF_EVENT_STATE_OFF) 734 if (group_event->state == PERF_EVENT_STATE_OFF)
698 return 0; 735 return 0;
699 736
700 /* Check if group transaction availabe */ 737 pmu->start_txn(pmu);
701 if (pmu->start_txn)
702 txn = true;
703 738
704 if (txn) 739 /*
705 pmu->start_txn(pmu); 740 * use __event_sched_in() to delay updating tstamp_running
706 741 * until the transaction is committed. In case of failure
707 if (event_sched_in(group_event, cpuctx, ctx)) { 742 * we will keep an unmodified tstamp_running which is a
708 if (txn) 743 * requirement to get correct timing information
709 pmu->cancel_txn(pmu); 744 */
745 if (__event_sched_in(group_event, cpuctx, ctx)) {
746 pmu->cancel_txn(pmu);
710 return -EAGAIN; 747 return -EAGAIN;
711 } 748 }
712 749
@@ -714,29 +751,33 @@ group_sched_in(struct perf_event *group_event,
714 * Schedule in siblings as one group (if any): 751 * Schedule in siblings as one group (if any):
715 */ 752 */
716 list_for_each_entry(event, &group_event->sibling_list, group_entry) { 753 list_for_each_entry(event, &group_event->sibling_list, group_entry) {
717 if (event_sched_in(event, cpuctx, ctx)) { 754 if (__event_sched_in(event, cpuctx, ctx)) {
718 partial_group = event; 755 partial_group = event;
719 goto group_error; 756 goto group_error;
720 } 757 }
721 } 758 }
722 759
723 if (!txn || !pmu->commit_txn(pmu)) 760 if (!pmu->commit_txn(pmu)) {
761 /* commit tstamp_running */
762 group_commit_event_sched_in(group_event, cpuctx, ctx);
724 return 0; 763 return 0;
725 764 }
726group_error: 765group_error:
727 /* 766 /*
728 * Groups can be scheduled in as one unit only, so undo any 767 * Groups can be scheduled in as one unit only, so undo any
729 * partial group before returning: 768 * partial group before returning:
769 *
770 * use __event_sched_out() to avoid updating tstamp_stopped
771 * because the event never actually ran
730 */ 772 */
731 list_for_each_entry(event, &group_event->sibling_list, group_entry) { 773 list_for_each_entry(event, &group_event->sibling_list, group_entry) {
732 if (event == partial_group) 774 if (event == partial_group)
733 break; 775 break;
734 event_sched_out(event, cpuctx, ctx); 776 __event_sched_out(event, cpuctx, ctx);
735 } 777 }
736 event_sched_out(group_event, cpuctx, ctx); 778 __event_sched_out(group_event, cpuctx, ctx);
737 779
738 if (txn) 780 pmu->cancel_txn(pmu);
739 pmu->cancel_txn(pmu);
740 781
741 return -EAGAIN; 782 return -EAGAIN;
742} 783}
@@ -789,10 +830,10 @@ static void add_event_to_ctx(struct perf_event *event,
789 */ 830 */
790static void __perf_install_in_context(void *info) 831static void __perf_install_in_context(void *info)
791{ 832{
792 struct perf_cpu_context *cpuctx = &__get_cpu_var(perf_cpu_context);
793 struct perf_event *event = info; 833 struct perf_event *event = info;
794 struct perf_event_context *ctx = event->ctx; 834 struct perf_event_context *ctx = event->ctx;
795 struct perf_event *leader = event->group_leader; 835 struct perf_event *leader = event->group_leader;
836 struct perf_cpu_context *cpuctx = __get_cpu_context(ctx);
796 int err; 837 int err;
797 838
798 /* 839 /*
@@ -812,12 +853,6 @@ static void __perf_install_in_context(void *info)
812 ctx->is_active = 1; 853 ctx->is_active = 1;
813 update_context_time(ctx); 854 update_context_time(ctx);
814 855
815 /*
816 * Protect the list operation against NMI by disabling the
817 * events on a global level. NOP for non NMI based events.
818 */
819 perf_disable();
820
821 add_event_to_ctx(event, ctx); 856 add_event_to_ctx(event, ctx);
822 857
823 if (event->cpu != -1 && event->cpu != smp_processor_id()) 858 if (event->cpu != -1 && event->cpu != smp_processor_id())
@@ -855,12 +890,7 @@ static void __perf_install_in_context(void *info)
855 } 890 }
856 } 891 }
857 892
858 if (!err && !ctx->task && cpuctx->max_pertask) 893unlock:
859 cpuctx->max_pertask--;
860
861 unlock:
862 perf_enable();
863
864 raw_spin_unlock(&ctx->lock); 894 raw_spin_unlock(&ctx->lock);
865} 895}
866 896
@@ -883,6 +913,8 @@ perf_install_in_context(struct perf_event_context *ctx,
883{ 913{
884 struct task_struct *task = ctx->task; 914 struct task_struct *task = ctx->task;
885 915
916 event->ctx = ctx;
917
886 if (!task) { 918 if (!task) {
887 /* 919 /*
888 * Per cpu events are installed via an smp call and 920 * Per cpu events are installed via an smp call and
@@ -931,10 +963,12 @@ static void __perf_event_mark_enabled(struct perf_event *event,
931 963
932 event->state = PERF_EVENT_STATE_INACTIVE; 964 event->state = PERF_EVENT_STATE_INACTIVE;
933 event->tstamp_enabled = ctx->time - event->total_time_enabled; 965 event->tstamp_enabled = ctx->time - event->total_time_enabled;
934 list_for_each_entry(sub, &event->sibling_list, group_entry) 966 list_for_each_entry(sub, &event->sibling_list, group_entry) {
935 if (sub->state >= PERF_EVENT_STATE_INACTIVE) 967 if (sub->state >= PERF_EVENT_STATE_INACTIVE) {
936 sub->tstamp_enabled = 968 sub->tstamp_enabled =
937 ctx->time - sub->total_time_enabled; 969 ctx->time - sub->total_time_enabled;
970 }
971 }
938} 972}
939 973
940/* 974/*
@@ -943,9 +977,9 @@ static void __perf_event_mark_enabled(struct perf_event *event,
943static void __perf_event_enable(void *info) 977static void __perf_event_enable(void *info)
944{ 978{
945 struct perf_event *event = info; 979 struct perf_event *event = info;
946 struct perf_cpu_context *cpuctx = &__get_cpu_var(perf_cpu_context);
947 struct perf_event_context *ctx = event->ctx; 980 struct perf_event_context *ctx = event->ctx;
948 struct perf_event *leader = event->group_leader; 981 struct perf_event *leader = event->group_leader;
982 struct perf_cpu_context *cpuctx = __get_cpu_context(ctx);
949 int err; 983 int err;
950 984
951 /* 985 /*
@@ -979,12 +1013,10 @@ static void __perf_event_enable(void *info)
979 if (!group_can_go_on(event, cpuctx, 1)) { 1013 if (!group_can_go_on(event, cpuctx, 1)) {
980 err = -EEXIST; 1014 err = -EEXIST;
981 } else { 1015 } else {
982 perf_disable();
983 if (event == leader) 1016 if (event == leader)
984 err = group_sched_in(event, cpuctx, ctx); 1017 err = group_sched_in(event, cpuctx, ctx);
985 else 1018 else
986 err = event_sched_in(event, cpuctx, ctx); 1019 err = event_sched_in(event, cpuctx, ctx);
987 perf_enable();
988 } 1020 }
989 1021
990 if (err) { 1022 if (err) {
@@ -1000,7 +1032,7 @@ static void __perf_event_enable(void *info)
1000 } 1032 }
1001 } 1033 }
1002 1034
1003 unlock: 1035unlock:
1004 raw_spin_unlock(&ctx->lock); 1036 raw_spin_unlock(&ctx->lock);
1005} 1037}
1006 1038
@@ -1041,7 +1073,7 @@ void perf_event_enable(struct perf_event *event)
1041 if (event->state == PERF_EVENT_STATE_ERROR) 1073 if (event->state == PERF_EVENT_STATE_ERROR)
1042 event->state = PERF_EVENT_STATE_OFF; 1074 event->state = PERF_EVENT_STATE_OFF;
1043 1075
1044 retry: 1076retry:
1045 raw_spin_unlock_irq(&ctx->lock); 1077 raw_spin_unlock_irq(&ctx->lock);
1046 task_oncpu_function_call(task, __perf_event_enable, event); 1078 task_oncpu_function_call(task, __perf_event_enable, event);
1047 1079
@@ -1061,7 +1093,7 @@ void perf_event_enable(struct perf_event *event)
1061 if (event->state == PERF_EVENT_STATE_OFF) 1093 if (event->state == PERF_EVENT_STATE_OFF)
1062 __perf_event_mark_enabled(event, ctx); 1094 __perf_event_mark_enabled(event, ctx);
1063 1095
1064 out: 1096out:
1065 raw_spin_unlock_irq(&ctx->lock); 1097 raw_spin_unlock_irq(&ctx->lock);
1066} 1098}
1067 1099
@@ -1092,26 +1124,26 @@ static void ctx_sched_out(struct perf_event_context *ctx,
1092 struct perf_event *event; 1124 struct perf_event *event;
1093 1125
1094 raw_spin_lock(&ctx->lock); 1126 raw_spin_lock(&ctx->lock);
1127 perf_pmu_disable(ctx->pmu);
1095 ctx->is_active = 0; 1128 ctx->is_active = 0;
1096 if (likely(!ctx->nr_events)) 1129 if (likely(!ctx->nr_events))
1097 goto out; 1130 goto out;
1098 update_context_time(ctx); 1131 update_context_time(ctx);
1099 1132
1100 perf_disable();
1101 if (!ctx->nr_active) 1133 if (!ctx->nr_active)
1102 goto out_enable; 1134 goto out;
1103 1135
1104 if (event_type & EVENT_PINNED) 1136 if (event_type & EVENT_PINNED) {
1105 list_for_each_entry(event, &ctx->pinned_groups, group_entry) 1137 list_for_each_entry(event, &ctx->pinned_groups, group_entry)
1106 group_sched_out(event, cpuctx, ctx); 1138 group_sched_out(event, cpuctx, ctx);
1139 }
1107 1140
1108 if (event_type & EVENT_FLEXIBLE) 1141 if (event_type & EVENT_FLEXIBLE) {
1109 list_for_each_entry(event, &ctx->flexible_groups, group_entry) 1142 list_for_each_entry(event, &ctx->flexible_groups, group_entry)
1110 group_sched_out(event, cpuctx, ctx); 1143 group_sched_out(event, cpuctx, ctx);
1111 1144 }
1112 out_enable: 1145out:
1113 perf_enable(); 1146 perf_pmu_enable(ctx->pmu);
1114 out:
1115 raw_spin_unlock(&ctx->lock); 1147 raw_spin_unlock(&ctx->lock);
1116} 1148}
1117 1149
@@ -1209,34 +1241,25 @@ static void perf_event_sync_stat(struct perf_event_context *ctx,
1209 } 1241 }
1210} 1242}
1211 1243
1212/* 1244void perf_event_context_sched_out(struct task_struct *task, int ctxn,
1213 * Called from scheduler to remove the events of the current task, 1245 struct task_struct *next)
1214 * with interrupts disabled.
1215 *
1216 * We stop each event and update the event value in event->count.
1217 *
1218 * This does not protect us against NMI, but disable()
1219 * sets the disabled bit in the control field of event _before_
1220 * accessing the event control register. If a NMI hits, then it will
1221 * not restart the event.
1222 */
1223void perf_event_task_sched_out(struct task_struct *task,
1224 struct task_struct *next)
1225{ 1246{
1226 struct perf_cpu_context *cpuctx = &__get_cpu_var(perf_cpu_context); 1247 struct perf_event_context *ctx = task->perf_event_ctxp[ctxn];
1227 struct perf_event_context *ctx = task->perf_event_ctxp;
1228 struct perf_event_context *next_ctx; 1248 struct perf_event_context *next_ctx;
1229 struct perf_event_context *parent; 1249 struct perf_event_context *parent;
1250 struct perf_cpu_context *cpuctx;
1230 int do_switch = 1; 1251 int do_switch = 1;
1231 1252
1232 perf_sw_event(PERF_COUNT_SW_CONTEXT_SWITCHES, 1, 1, NULL, 0); 1253 if (likely(!ctx))
1254 return;
1233 1255
1234 if (likely(!ctx || !cpuctx->task_ctx)) 1256 cpuctx = __get_cpu_context(ctx);
1257 if (!cpuctx->task_ctx)
1235 return; 1258 return;
1236 1259
1237 rcu_read_lock(); 1260 rcu_read_lock();
1238 parent = rcu_dereference(ctx->parent_ctx); 1261 parent = rcu_dereference(ctx->parent_ctx);
1239 next_ctx = next->perf_event_ctxp; 1262 next_ctx = next->perf_event_ctxp[ctxn];
1240 if (parent && next_ctx && 1263 if (parent && next_ctx &&
1241 rcu_dereference(next_ctx->parent_ctx) == parent) { 1264 rcu_dereference(next_ctx->parent_ctx) == parent) {
1242 /* 1265 /*
@@ -1255,8 +1278,8 @@ void perf_event_task_sched_out(struct task_struct *task,
1255 * XXX do we need a memory barrier of sorts 1278 * XXX do we need a memory barrier of sorts
1256 * wrt to rcu_dereference() of perf_event_ctxp 1279 * wrt to rcu_dereference() of perf_event_ctxp
1257 */ 1280 */
1258 task->perf_event_ctxp = next_ctx; 1281 task->perf_event_ctxp[ctxn] = next_ctx;
1259 next->perf_event_ctxp = ctx; 1282 next->perf_event_ctxp[ctxn] = ctx;
1260 ctx->task = next; 1283 ctx->task = next;
1261 next_ctx->task = task; 1284 next_ctx->task = task;
1262 do_switch = 0; 1285 do_switch = 0;
@@ -1274,10 +1297,35 @@ void perf_event_task_sched_out(struct task_struct *task,
1274 } 1297 }
1275} 1298}
1276 1299
1300#define for_each_task_context_nr(ctxn) \
1301 for ((ctxn) = 0; (ctxn) < perf_nr_task_contexts; (ctxn)++)
1302
1303/*
1304 * Called from scheduler to remove the events of the current task,
1305 * with interrupts disabled.
1306 *
1307 * We stop each event and update the event value in event->count.
1308 *
1309 * This does not protect us against NMI, but disable()
1310 * sets the disabled bit in the control field of event _before_
1311 * accessing the event control register. If a NMI hits, then it will
1312 * not restart the event.
1313 */
1314void __perf_event_task_sched_out(struct task_struct *task,
1315 struct task_struct *next)
1316{
1317 int ctxn;
1318
1319 perf_sw_event(PERF_COUNT_SW_CONTEXT_SWITCHES, 1, 1, NULL, 0);
1320
1321 for_each_task_context_nr(ctxn)
1322 perf_event_context_sched_out(task, ctxn, next);
1323}
1324
1277static void task_ctx_sched_out(struct perf_event_context *ctx, 1325static void task_ctx_sched_out(struct perf_event_context *ctx,
1278 enum event_type_t event_type) 1326 enum event_type_t event_type)
1279{ 1327{
1280 struct perf_cpu_context *cpuctx = &__get_cpu_var(perf_cpu_context); 1328 struct perf_cpu_context *cpuctx = __get_cpu_context(ctx);
1281 1329
1282 if (!cpuctx->task_ctx) 1330 if (!cpuctx->task_ctx)
1283 return; 1331 return;
@@ -1292,14 +1340,6 @@ static void task_ctx_sched_out(struct perf_event_context *ctx,
1292/* 1340/*
1293 * Called with IRQs disabled 1341 * Called with IRQs disabled
1294 */ 1342 */
1295static void __perf_event_task_sched_out(struct perf_event_context *ctx)
1296{
1297 task_ctx_sched_out(ctx, EVENT_ALL);
1298}
1299
1300/*
1301 * Called with IRQs disabled
1302 */
1303static void cpu_ctx_sched_out(struct perf_cpu_context *cpuctx, 1343static void cpu_ctx_sched_out(struct perf_cpu_context *cpuctx,
1304 enum event_type_t event_type) 1344 enum event_type_t event_type)
1305{ 1345{
@@ -1350,9 +1390,10 @@ ctx_flexible_sched_in(struct perf_event_context *ctx,
1350 if (event->cpu != -1 && event->cpu != smp_processor_id()) 1390 if (event->cpu != -1 && event->cpu != smp_processor_id())
1351 continue; 1391 continue;
1352 1392
1353 if (group_can_go_on(event, cpuctx, can_add_hw)) 1393 if (group_can_go_on(event, cpuctx, can_add_hw)) {
1354 if (group_sched_in(event, cpuctx, ctx)) 1394 if (group_sched_in(event, cpuctx, ctx))
1355 can_add_hw = 0; 1395 can_add_hw = 0;
1396 }
1356 } 1397 }
1357} 1398}
1358 1399
@@ -1368,8 +1409,6 @@ ctx_sched_in(struct perf_event_context *ctx,
1368 1409
1369 ctx->timestamp = perf_clock(); 1410 ctx->timestamp = perf_clock();
1370 1411
1371 perf_disable();
1372
1373 /* 1412 /*
1374 * First go through the list and put on any pinned groups 1413 * First go through the list and put on any pinned groups
1375 * in order to give them the best chance of going on. 1414 * in order to give them the best chance of going on.
@@ -1381,8 +1420,7 @@ ctx_sched_in(struct perf_event_context *ctx,
1381 if (event_type & EVENT_FLEXIBLE) 1420 if (event_type & EVENT_FLEXIBLE)
1382 ctx_flexible_sched_in(ctx, cpuctx); 1421 ctx_flexible_sched_in(ctx, cpuctx);
1383 1422
1384 perf_enable(); 1423out:
1385 out:
1386 raw_spin_unlock(&ctx->lock); 1424 raw_spin_unlock(&ctx->lock);
1387} 1425}
1388 1426
@@ -1394,43 +1432,28 @@ static void cpu_ctx_sched_in(struct perf_cpu_context *cpuctx,
1394 ctx_sched_in(ctx, cpuctx, event_type); 1432 ctx_sched_in(ctx, cpuctx, event_type);
1395} 1433}
1396 1434
1397static void task_ctx_sched_in(struct task_struct *task, 1435static void task_ctx_sched_in(struct perf_event_context *ctx,
1398 enum event_type_t event_type) 1436 enum event_type_t event_type)
1399{ 1437{
1400 struct perf_cpu_context *cpuctx = &__get_cpu_var(perf_cpu_context); 1438 struct perf_cpu_context *cpuctx;
1401 struct perf_event_context *ctx = task->perf_event_ctxp;
1402 1439
1403 if (likely(!ctx)) 1440 cpuctx = __get_cpu_context(ctx);
1404 return;
1405 if (cpuctx->task_ctx == ctx) 1441 if (cpuctx->task_ctx == ctx)
1406 return; 1442 return;
1443
1407 ctx_sched_in(ctx, cpuctx, event_type); 1444 ctx_sched_in(ctx, cpuctx, event_type);
1408 cpuctx->task_ctx = ctx; 1445 cpuctx->task_ctx = ctx;
1409} 1446}
1410/*
1411 * Called from scheduler to add the events of the current task
1412 * with interrupts disabled.
1413 *
1414 * We restore the event value and then enable it.
1415 *
1416 * This does not protect us against NMI, but enable()
1417 * sets the enabled bit in the control field of event _before_
1418 * accessing the event control register. If a NMI hits, then it will
1419 * keep the event running.
1420 */
1421void perf_event_task_sched_in(struct task_struct *task)
1422{
1423 struct perf_cpu_context *cpuctx = &__get_cpu_var(perf_cpu_context);
1424 struct perf_event_context *ctx = task->perf_event_ctxp;
1425 1447
1426 if (likely(!ctx)) 1448void perf_event_context_sched_in(struct perf_event_context *ctx)
1427 return; 1449{
1450 struct perf_cpu_context *cpuctx;
1428 1451
1452 cpuctx = __get_cpu_context(ctx);
1429 if (cpuctx->task_ctx == ctx) 1453 if (cpuctx->task_ctx == ctx)
1430 return; 1454 return;
1431 1455
1432 perf_disable(); 1456 perf_pmu_disable(ctx->pmu);
1433
1434 /* 1457 /*
1435 * We want to keep the following priority order: 1458 * We want to keep the following priority order:
1436 * cpu pinned (that don't need to move), task pinned, 1459 * cpu pinned (that don't need to move), task pinned,
@@ -1444,7 +1467,37 @@ void perf_event_task_sched_in(struct task_struct *task)
1444 1467
1445 cpuctx->task_ctx = ctx; 1468 cpuctx->task_ctx = ctx;
1446 1469
1447 perf_enable(); 1470 /*
1471 * Since these rotations are per-cpu, we need to ensure the
1472 * cpu-context we got scheduled on is actually rotating.
1473 */
1474 perf_pmu_rotate_start(ctx->pmu);
1475 perf_pmu_enable(ctx->pmu);
1476}
1477
1478/*
1479 * Called from scheduler to add the events of the current task
1480 * with interrupts disabled.
1481 *
1482 * We restore the event value and then enable it.
1483 *
1484 * This does not protect us against NMI, but enable()
1485 * sets the enabled bit in the control field of event _before_
1486 * accessing the event control register. If a NMI hits, then it will
1487 * keep the event running.
1488 */
1489void __perf_event_task_sched_in(struct task_struct *task)
1490{
1491 struct perf_event_context *ctx;
1492 int ctxn;
1493
1494 for_each_task_context_nr(ctxn) {
1495 ctx = task->perf_event_ctxp[ctxn];
1496 if (likely(!ctx))
1497 continue;
1498
1499 perf_event_context_sched_in(ctx);
1500 }
1448} 1501}
1449 1502
1450#define MAX_INTERRUPTS (~0ULL) 1503#define MAX_INTERRUPTS (~0ULL)
@@ -1524,22 +1577,6 @@ do { \
1524 return div64_u64(dividend, divisor); 1577 return div64_u64(dividend, divisor);
1525} 1578}
1526 1579
1527static void perf_event_stop(struct perf_event *event)
1528{
1529 if (!event->pmu->stop)
1530 return event->pmu->disable(event);
1531
1532 return event->pmu->stop(event);
1533}
1534
1535static int perf_event_start(struct perf_event *event)
1536{
1537 if (!event->pmu->start)
1538 return event->pmu->enable(event);
1539
1540 return event->pmu->start(event);
1541}
1542
1543static void perf_adjust_period(struct perf_event *event, u64 nsec, u64 count) 1580static void perf_adjust_period(struct perf_event *event, u64 nsec, u64 count)
1544{ 1581{
1545 struct hw_perf_event *hwc = &event->hw; 1582 struct hw_perf_event *hwc = &event->hw;
@@ -1559,15 +1596,13 @@ static void perf_adjust_period(struct perf_event *event, u64 nsec, u64 count)
1559 hwc->sample_period = sample_period; 1596 hwc->sample_period = sample_period;
1560 1597
1561 if (local64_read(&hwc->period_left) > 8*sample_period) { 1598 if (local64_read(&hwc->period_left) > 8*sample_period) {
1562 perf_disable(); 1599 event->pmu->stop(event, PERF_EF_UPDATE);
1563 perf_event_stop(event);
1564 local64_set(&hwc->period_left, 0); 1600 local64_set(&hwc->period_left, 0);
1565 perf_event_start(event); 1601 event->pmu->start(event, PERF_EF_RELOAD);
1566 perf_enable();
1567 } 1602 }
1568} 1603}
1569 1604
1570static void perf_ctx_adjust_freq(struct perf_event_context *ctx) 1605static void perf_ctx_adjust_freq(struct perf_event_context *ctx, u64 period)
1571{ 1606{
1572 struct perf_event *event; 1607 struct perf_event *event;
1573 struct hw_perf_event *hwc; 1608 struct hw_perf_event *hwc;
@@ -1592,23 +1627,19 @@ static void perf_ctx_adjust_freq(struct perf_event_context *ctx)
1592 */ 1627 */
1593 if (interrupts == MAX_INTERRUPTS) { 1628 if (interrupts == MAX_INTERRUPTS) {
1594 perf_log_throttle(event, 1); 1629 perf_log_throttle(event, 1);
1595 perf_disable(); 1630 event->pmu->start(event, 0);
1596 event->pmu->unthrottle(event);
1597 perf_enable();
1598 } 1631 }
1599 1632
1600 if (!event->attr.freq || !event->attr.sample_freq) 1633 if (!event->attr.freq || !event->attr.sample_freq)
1601 continue; 1634 continue;
1602 1635
1603 perf_disable();
1604 event->pmu->read(event); 1636 event->pmu->read(event);
1605 now = local64_read(&event->count); 1637 now = local64_read(&event->count);
1606 delta = now - hwc->freq_count_stamp; 1638 delta = now - hwc->freq_count_stamp;
1607 hwc->freq_count_stamp = now; 1639 hwc->freq_count_stamp = now;
1608 1640
1609 if (delta > 0) 1641 if (delta > 0)
1610 perf_adjust_period(event, TICK_NSEC, delta); 1642 perf_adjust_period(event, period, delta);
1611 perf_enable();
1612 } 1643 }
1613 raw_spin_unlock(&ctx->lock); 1644 raw_spin_unlock(&ctx->lock);
1614} 1645}
@@ -1626,32 +1657,38 @@ static void rotate_ctx(struct perf_event_context *ctx)
1626 raw_spin_unlock(&ctx->lock); 1657 raw_spin_unlock(&ctx->lock);
1627} 1658}
1628 1659
1629void perf_event_task_tick(struct task_struct *curr) 1660/*
1661 * perf_pmu_rotate_start() and perf_rotate_context() are fully serialized
1662 * because they're strictly cpu affine and rotate_start is called with IRQs
1663 * disabled, while rotate_context is called from IRQ context.
1664 */
1665static void perf_rotate_context(struct perf_cpu_context *cpuctx)
1630{ 1666{
1631 struct perf_cpu_context *cpuctx; 1667 u64 interval = (u64)cpuctx->jiffies_interval * TICK_NSEC;
1632 struct perf_event_context *ctx; 1668 struct perf_event_context *ctx = NULL;
1633 int rotate = 0; 1669 int rotate = 0, remove = 1;
1634
1635 if (!atomic_read(&nr_events))
1636 return;
1637 1670
1638 cpuctx = &__get_cpu_var(perf_cpu_context); 1671 if (cpuctx->ctx.nr_events) {
1639 if (cpuctx->ctx.nr_events && 1672 remove = 0;
1640 cpuctx->ctx.nr_events != cpuctx->ctx.nr_active) 1673 if (cpuctx->ctx.nr_events != cpuctx->ctx.nr_active)
1641 rotate = 1; 1674 rotate = 1;
1675 }
1642 1676
1643 ctx = curr->perf_event_ctxp; 1677 ctx = cpuctx->task_ctx;
1644 if (ctx && ctx->nr_events && ctx->nr_events != ctx->nr_active) 1678 if (ctx && ctx->nr_events) {
1645 rotate = 1; 1679 remove = 0;
1680 if (ctx->nr_events != ctx->nr_active)
1681 rotate = 1;
1682 }
1646 1683
1647 perf_ctx_adjust_freq(&cpuctx->ctx); 1684 perf_pmu_disable(cpuctx->ctx.pmu);
1685 perf_ctx_adjust_freq(&cpuctx->ctx, interval);
1648 if (ctx) 1686 if (ctx)
1649 perf_ctx_adjust_freq(ctx); 1687 perf_ctx_adjust_freq(ctx, interval);
1650 1688
1651 if (!rotate) 1689 if (!rotate)
1652 return; 1690 goto done;
1653 1691
1654 perf_disable();
1655 cpu_ctx_sched_out(cpuctx, EVENT_FLEXIBLE); 1692 cpu_ctx_sched_out(cpuctx, EVENT_FLEXIBLE);
1656 if (ctx) 1693 if (ctx)
1657 task_ctx_sched_out(ctx, EVENT_FLEXIBLE); 1694 task_ctx_sched_out(ctx, EVENT_FLEXIBLE);
@@ -1662,8 +1699,27 @@ void perf_event_task_tick(struct task_struct *curr)
1662 1699
1663 cpu_ctx_sched_in(cpuctx, EVENT_FLEXIBLE); 1700 cpu_ctx_sched_in(cpuctx, EVENT_FLEXIBLE);
1664 if (ctx) 1701 if (ctx)
1665 task_ctx_sched_in(curr, EVENT_FLEXIBLE); 1702 task_ctx_sched_in(ctx, EVENT_FLEXIBLE);
1666 perf_enable(); 1703
1704done:
1705 if (remove)
1706 list_del_init(&cpuctx->rotation_list);
1707
1708 perf_pmu_enable(cpuctx->ctx.pmu);
1709}
1710
1711void perf_event_task_tick(void)
1712{
1713 struct list_head *head = &__get_cpu_var(rotation_list);
1714 struct perf_cpu_context *cpuctx, *tmp;
1715
1716 WARN_ON(!irqs_disabled());
1717
1718 list_for_each_entry_safe(cpuctx, tmp, head, rotation_list) {
1719 if (cpuctx->jiffies_interval == 1 ||
1720 !(jiffies % cpuctx->jiffies_interval))
1721 perf_rotate_context(cpuctx);
1722 }
1667} 1723}
1668 1724
1669static int event_enable_on_exec(struct perf_event *event, 1725static int event_enable_on_exec(struct perf_event *event,
@@ -1685,20 +1741,18 @@ static int event_enable_on_exec(struct perf_event *event,
1685 * Enable all of a task's events that have been marked enable-on-exec. 1741 * Enable all of a task's events that have been marked enable-on-exec.
1686 * This expects task == current. 1742 * This expects task == current.
1687 */ 1743 */
1688static void perf_event_enable_on_exec(struct task_struct *task) 1744static void perf_event_enable_on_exec(struct perf_event_context *ctx)
1689{ 1745{
1690 struct perf_event_context *ctx;
1691 struct perf_event *event; 1746 struct perf_event *event;
1692 unsigned long flags; 1747 unsigned long flags;
1693 int enabled = 0; 1748 int enabled = 0;
1694 int ret; 1749 int ret;
1695 1750
1696 local_irq_save(flags); 1751 local_irq_save(flags);
1697 ctx = task->perf_event_ctxp;
1698 if (!ctx || !ctx->nr_events) 1752 if (!ctx || !ctx->nr_events)
1699 goto out; 1753 goto out;
1700 1754
1701 __perf_event_task_sched_out(ctx); 1755 task_ctx_sched_out(ctx, EVENT_ALL);
1702 1756
1703 raw_spin_lock(&ctx->lock); 1757 raw_spin_lock(&ctx->lock);
1704 1758
@@ -1722,8 +1776,8 @@ static void perf_event_enable_on_exec(struct task_struct *task)
1722 1776
1723 raw_spin_unlock(&ctx->lock); 1777 raw_spin_unlock(&ctx->lock);
1724 1778
1725 perf_event_task_sched_in(task); 1779 perf_event_context_sched_in(ctx);
1726 out: 1780out:
1727 local_irq_restore(flags); 1781 local_irq_restore(flags);
1728} 1782}
1729 1783
@@ -1732,9 +1786,9 @@ static void perf_event_enable_on_exec(struct task_struct *task)
1732 */ 1786 */
1733static void __perf_event_read(void *info) 1787static void __perf_event_read(void *info)
1734{ 1788{
1735 struct perf_cpu_context *cpuctx = &__get_cpu_var(perf_cpu_context);
1736 struct perf_event *event = info; 1789 struct perf_event *event = info;
1737 struct perf_event_context *ctx = event->ctx; 1790 struct perf_event_context *ctx = event->ctx;
1791 struct perf_cpu_context *cpuctx = __get_cpu_context(ctx);
1738 1792
1739 /* 1793 /*
1740 * If this is a task context, we need to check whether it is 1794 * If this is a task context, we need to check whether it is
@@ -1773,7 +1827,13 @@ static u64 perf_event_read(struct perf_event *event)
1773 unsigned long flags; 1827 unsigned long flags;
1774 1828
1775 raw_spin_lock_irqsave(&ctx->lock, flags); 1829 raw_spin_lock_irqsave(&ctx->lock, flags);
1776 update_context_time(ctx); 1830 /*
1831 * may read while context is not active
1832 * (e.g., thread is blocked), in that case
1833 * we cannot update context time
1834 */
1835 if (ctx->is_active)
1836 update_context_time(ctx);
1777 update_event_times(event); 1837 update_event_times(event);
1778 raw_spin_unlock_irqrestore(&ctx->lock, flags); 1838 raw_spin_unlock_irqrestore(&ctx->lock, flags);
1779 } 1839 }
@@ -1782,11 +1842,219 @@ static u64 perf_event_read(struct perf_event *event)
1782} 1842}
1783 1843
1784/* 1844/*
1785 * Initialize the perf_event context in a task_struct: 1845 * Callchain support
1786 */ 1846 */
1847
1848struct callchain_cpus_entries {
1849 struct rcu_head rcu_head;
1850 struct perf_callchain_entry *cpu_entries[0];
1851};
1852
1853static DEFINE_PER_CPU(int, callchain_recursion[PERF_NR_CONTEXTS]);
1854static atomic_t nr_callchain_events;
1855static DEFINE_MUTEX(callchain_mutex);
1856struct callchain_cpus_entries *callchain_cpus_entries;
1857
1858
1859__weak void perf_callchain_kernel(struct perf_callchain_entry *entry,
1860 struct pt_regs *regs)
1861{
1862}
1863
1864__weak void perf_callchain_user(struct perf_callchain_entry *entry,
1865 struct pt_regs *regs)
1866{
1867}
1868
1869static void release_callchain_buffers_rcu(struct rcu_head *head)
1870{
1871 struct callchain_cpus_entries *entries;
1872 int cpu;
1873
1874 entries = container_of(head, struct callchain_cpus_entries, rcu_head);
1875
1876 for_each_possible_cpu(cpu)
1877 kfree(entries->cpu_entries[cpu]);
1878
1879 kfree(entries);
1880}
1881
1882static void release_callchain_buffers(void)
1883{
1884 struct callchain_cpus_entries *entries;
1885
1886 entries = callchain_cpus_entries;
1887 rcu_assign_pointer(callchain_cpus_entries, NULL);
1888 call_rcu(&entries->rcu_head, release_callchain_buffers_rcu);
1889}
1890
1891static int alloc_callchain_buffers(void)
1892{
1893 int cpu;
1894 int size;
1895 struct callchain_cpus_entries *entries;
1896
1897 /*
1898 * We can't use the percpu allocation API for data that can be
1899 * accessed from NMI. Use a temporary manual per cpu allocation
1900 * until that gets sorted out.
1901 */
1902 size = sizeof(*entries) + sizeof(struct perf_callchain_entry *) *
1903 num_possible_cpus();
1904
1905 entries = kzalloc(size, GFP_KERNEL);
1906 if (!entries)
1907 return -ENOMEM;
1908
1909 size = sizeof(struct perf_callchain_entry) * PERF_NR_CONTEXTS;
1910
1911 for_each_possible_cpu(cpu) {
1912 entries->cpu_entries[cpu] = kmalloc_node(size, GFP_KERNEL,
1913 cpu_to_node(cpu));
1914 if (!entries->cpu_entries[cpu])
1915 goto fail;
1916 }
1917
1918 rcu_assign_pointer(callchain_cpus_entries, entries);
1919
1920 return 0;
1921
1922fail:
1923 for_each_possible_cpu(cpu)
1924 kfree(entries->cpu_entries[cpu]);
1925 kfree(entries);
1926
1927 return -ENOMEM;
1928}
1929
1930static int get_callchain_buffers(void)
1931{
1932 int err = 0;
1933 int count;
1934
1935 mutex_lock(&callchain_mutex);
1936
1937 count = atomic_inc_return(&nr_callchain_events);
1938 if (WARN_ON_ONCE(count < 1)) {
1939 err = -EINVAL;
1940 goto exit;
1941 }
1942
1943 if (count > 1) {
1944 /* If the allocation failed, give up */
1945 if (!callchain_cpus_entries)
1946 err = -ENOMEM;
1947 goto exit;
1948 }
1949
1950 err = alloc_callchain_buffers();
1951 if (err)
1952 release_callchain_buffers();
1953exit:
1954 mutex_unlock(&callchain_mutex);
1955
1956 return err;
1957}
1958
1959static void put_callchain_buffers(void)
1960{
1961 if (atomic_dec_and_mutex_lock(&nr_callchain_events, &callchain_mutex)) {
1962 release_callchain_buffers();
1963 mutex_unlock(&callchain_mutex);
1964 }
1965}
1966
1967static int get_recursion_context(int *recursion)
1968{
1969 int rctx;
1970
1971 if (in_nmi())
1972 rctx = 3;
1973 else if (in_irq())
1974 rctx = 2;
1975 else if (in_softirq())
1976 rctx = 1;
1977 else
1978 rctx = 0;
1979
1980 if (recursion[rctx])
1981 return -1;
1982
1983 recursion[rctx]++;
1984 barrier();
1985
1986 return rctx;
1987}
1988
1989static inline void put_recursion_context(int *recursion, int rctx)
1990{
1991 barrier();
1992 recursion[rctx]--;
1993}
1994
1995static struct perf_callchain_entry *get_callchain_entry(int *rctx)
1996{
1997 int cpu;
1998 struct callchain_cpus_entries *entries;
1999
2000 *rctx = get_recursion_context(__get_cpu_var(callchain_recursion));
2001 if (*rctx == -1)
2002 return NULL;
2003
2004 entries = rcu_dereference(callchain_cpus_entries);
2005 if (!entries)
2006 return NULL;
2007
2008 cpu = smp_processor_id();
2009
2010 return &entries->cpu_entries[cpu][*rctx];
2011}
2012
1787static void 2013static void
1788__perf_event_init_context(struct perf_event_context *ctx, 2014put_callchain_entry(int rctx)
1789 struct task_struct *task) 2015{
2016 put_recursion_context(__get_cpu_var(callchain_recursion), rctx);
2017}
2018
2019static struct perf_callchain_entry *perf_callchain(struct pt_regs *regs)
2020{
2021 int rctx;
2022 struct perf_callchain_entry *entry;
2023
2024
2025 entry = get_callchain_entry(&rctx);
2026 if (rctx == -1)
2027 return NULL;
2028
2029 if (!entry)
2030 goto exit_put;
2031
2032 entry->nr = 0;
2033
2034 if (!user_mode(regs)) {
2035 perf_callchain_store(entry, PERF_CONTEXT_KERNEL);
2036 perf_callchain_kernel(entry, regs);
2037 if (current->mm)
2038 regs = task_pt_regs(current);
2039 else
2040 regs = NULL;
2041 }
2042
2043 if (regs) {
2044 perf_callchain_store(entry, PERF_CONTEXT_USER);
2045 perf_callchain_user(entry, regs);
2046 }
2047
2048exit_put:
2049 put_callchain_entry(rctx);
2050
2051 return entry;
2052}
2053
2054/*
2055 * Initialize the perf_event context in a task_struct:
2056 */
2057static void __perf_event_init_context(struct perf_event_context *ctx)
1790{ 2058{
1791 raw_spin_lock_init(&ctx->lock); 2059 raw_spin_lock_init(&ctx->lock);
1792 mutex_init(&ctx->mutex); 2060 mutex_init(&ctx->mutex);
@@ -1794,45 +2062,38 @@ __perf_event_init_context(struct perf_event_context *ctx,
1794 INIT_LIST_HEAD(&ctx->flexible_groups); 2062 INIT_LIST_HEAD(&ctx->flexible_groups);
1795 INIT_LIST_HEAD(&ctx->event_list); 2063 INIT_LIST_HEAD(&ctx->event_list);
1796 atomic_set(&ctx->refcount, 1); 2064 atomic_set(&ctx->refcount, 1);
1797 ctx->task = task;
1798} 2065}
1799 2066
1800static struct perf_event_context *find_get_context(pid_t pid, int cpu) 2067static struct perf_event_context *
2068alloc_perf_context(struct pmu *pmu, struct task_struct *task)
1801{ 2069{
1802 struct perf_event_context *ctx; 2070 struct perf_event_context *ctx;
1803 struct perf_cpu_context *cpuctx;
1804 struct task_struct *task;
1805 unsigned long flags;
1806 int err;
1807
1808 if (pid == -1 && cpu != -1) {
1809 /* Must be root to operate on a CPU event: */
1810 if (perf_paranoid_cpu() && !capable(CAP_SYS_ADMIN))
1811 return ERR_PTR(-EACCES);
1812 2071
1813 if (cpu < 0 || cpu >= nr_cpumask_bits) 2072 ctx = kzalloc(sizeof(struct perf_event_context), GFP_KERNEL);
1814 return ERR_PTR(-EINVAL); 2073 if (!ctx)
2074 return NULL;
1815 2075
1816 /* 2076 __perf_event_init_context(ctx);
1817 * We could be clever and allow to attach a event to an 2077 if (task) {
1818 * offline CPU and activate it when the CPU comes up, but 2078 ctx->task = task;
1819 * that's for later. 2079 get_task_struct(task);
1820 */ 2080 }
1821 if (!cpu_online(cpu)) 2081 ctx->pmu = pmu;
1822 return ERR_PTR(-ENODEV);
1823 2082
1824 cpuctx = &per_cpu(perf_cpu_context, cpu); 2083 return ctx;
1825 ctx = &cpuctx->ctx; 2084}
1826 get_ctx(ctx);
1827 2085
1828 return ctx; 2086static struct task_struct *
1829 } 2087find_lively_task_by_vpid(pid_t vpid)
2088{
2089 struct task_struct *task;
2090 int err;
1830 2091
1831 rcu_read_lock(); 2092 rcu_read_lock();
1832 if (!pid) 2093 if (!vpid)
1833 task = current; 2094 task = current;
1834 else 2095 else
1835 task = find_task_by_vpid(pid); 2096 task = find_task_by_vpid(vpid);
1836 if (task) 2097 if (task)
1837 get_task_struct(task); 2098 get_task_struct(task);
1838 rcu_read_unlock(); 2099 rcu_read_unlock();
@@ -1852,36 +2113,78 @@ static struct perf_event_context *find_get_context(pid_t pid, int cpu)
1852 if (!ptrace_may_access(task, PTRACE_MODE_READ)) 2113 if (!ptrace_may_access(task, PTRACE_MODE_READ))
1853 goto errout; 2114 goto errout;
1854 2115
1855 retry: 2116 return task;
1856 ctx = perf_lock_task_context(task, &flags); 2117errout:
2118 put_task_struct(task);
2119 return ERR_PTR(err);
2120
2121}
2122
2123static struct perf_event_context *
2124find_get_context(struct pmu *pmu, struct task_struct *task, int cpu)
2125{
2126 struct perf_event_context *ctx;
2127 struct perf_cpu_context *cpuctx;
2128 unsigned long flags;
2129 int ctxn, err;
2130
2131 if (!task && cpu != -1) {
2132 /* Must be root to operate on a CPU event: */
2133 if (perf_paranoid_cpu() && !capable(CAP_SYS_ADMIN))
2134 return ERR_PTR(-EACCES);
2135
2136 if (cpu < 0 || cpu >= nr_cpumask_bits)
2137 return ERR_PTR(-EINVAL);
2138
2139 /*
2140 * We could be clever and allow to attach a event to an
2141 * offline CPU and activate it when the CPU comes up, but
2142 * that's for later.
2143 */
2144 if (!cpu_online(cpu))
2145 return ERR_PTR(-ENODEV);
2146
2147 cpuctx = per_cpu_ptr(pmu->pmu_cpu_context, cpu);
2148 ctx = &cpuctx->ctx;
2149 get_ctx(ctx);
2150
2151 return ctx;
2152 }
2153
2154 err = -EINVAL;
2155 ctxn = pmu->task_ctx_nr;
2156 if (ctxn < 0)
2157 goto errout;
2158
2159retry:
2160 ctx = perf_lock_task_context(task, ctxn, &flags);
1857 if (ctx) { 2161 if (ctx) {
1858 unclone_ctx(ctx); 2162 unclone_ctx(ctx);
1859 raw_spin_unlock_irqrestore(&ctx->lock, flags); 2163 raw_spin_unlock_irqrestore(&ctx->lock, flags);
1860 } 2164 }
1861 2165
1862 if (!ctx) { 2166 if (!ctx) {
1863 ctx = kzalloc(sizeof(struct perf_event_context), GFP_KERNEL); 2167 ctx = alloc_perf_context(pmu, task);
1864 err = -ENOMEM; 2168 err = -ENOMEM;
1865 if (!ctx) 2169 if (!ctx)
1866 goto errout; 2170 goto errout;
1867 __perf_event_init_context(ctx, task); 2171
1868 get_ctx(ctx); 2172 get_ctx(ctx);
1869 if (cmpxchg(&task->perf_event_ctxp, NULL, ctx)) { 2173
2174 if (cmpxchg(&task->perf_event_ctxp[ctxn], NULL, ctx)) {
1870 /* 2175 /*
1871 * We raced with some other task; use 2176 * We raced with some other task; use
1872 * the context they set. 2177 * the context they set.
1873 */ 2178 */
2179 put_task_struct(task);
1874 kfree(ctx); 2180 kfree(ctx);
1875 goto retry; 2181 goto retry;
1876 } 2182 }
1877 get_task_struct(task);
1878 } 2183 }
1879 2184
1880 put_task_struct(task);
1881 return ctx; 2185 return ctx;
1882 2186
1883 errout: 2187errout:
1884 put_task_struct(task);
1885 return ERR_PTR(err); 2188 return ERR_PTR(err);
1886} 2189}
1887 2190
@@ -1898,21 +2201,23 @@ static void free_event_rcu(struct rcu_head *head)
1898 kfree(event); 2201 kfree(event);
1899} 2202}
1900 2203
1901static void perf_pending_sync(struct perf_event *event);
1902static void perf_buffer_put(struct perf_buffer *buffer); 2204static void perf_buffer_put(struct perf_buffer *buffer);
1903 2205
1904static void free_event(struct perf_event *event) 2206static void free_event(struct perf_event *event)
1905{ 2207{
1906 perf_pending_sync(event); 2208 irq_work_sync(&event->pending);
1907 2209
1908 if (!event->parent) { 2210 if (!event->parent) {
1909 atomic_dec(&nr_events); 2211 if (event->attach_state & PERF_ATTACH_TASK)
2212 jump_label_dec(&perf_task_events);
1910 if (event->attr.mmap || event->attr.mmap_data) 2213 if (event->attr.mmap || event->attr.mmap_data)
1911 atomic_dec(&nr_mmap_events); 2214 atomic_dec(&nr_mmap_events);
1912 if (event->attr.comm) 2215 if (event->attr.comm)
1913 atomic_dec(&nr_comm_events); 2216 atomic_dec(&nr_comm_events);
1914 if (event->attr.task) 2217 if (event->attr.task)
1915 atomic_dec(&nr_task_events); 2218 atomic_dec(&nr_task_events);
2219 if (event->attr.sample_type & PERF_SAMPLE_CALLCHAIN)
2220 put_callchain_buffers();
1916 } 2221 }
1917 2222
1918 if (event->buffer) { 2223 if (event->buffer) {
@@ -1923,7 +2228,9 @@ static void free_event(struct perf_event *event)
1923 if (event->destroy) 2228 if (event->destroy)
1924 event->destroy(event); 2229 event->destroy(event);
1925 2230
1926 put_ctx(event->ctx); 2231 if (event->ctx)
2232 put_ctx(event->ctx);
2233
1927 call_rcu(&event->rcu_head, free_event_rcu); 2234 call_rcu(&event->rcu_head, free_event_rcu);
1928} 2235}
1929 2236
@@ -2202,15 +2509,13 @@ static void perf_event_for_each(struct perf_event *event,
2202static int perf_event_period(struct perf_event *event, u64 __user *arg) 2509static int perf_event_period(struct perf_event *event, u64 __user *arg)
2203{ 2510{
2204 struct perf_event_context *ctx = event->ctx; 2511 struct perf_event_context *ctx = event->ctx;
2205 unsigned long size;
2206 int ret = 0; 2512 int ret = 0;
2207 u64 value; 2513 u64 value;
2208 2514
2209 if (!event->attr.sample_period) 2515 if (!event->attr.sample_period)
2210 return -EINVAL; 2516 return -EINVAL;
2211 2517
2212 size = copy_from_user(&value, arg, sizeof(value)); 2518 if (copy_from_user(&value, arg, sizeof(value)))
2213 if (size != sizeof(value))
2214 return -EFAULT; 2519 return -EFAULT;
2215 2520
2216 if (!value) 2521 if (!value)
@@ -2344,6 +2649,9 @@ int perf_event_task_disable(void)
2344 2649
2345static int perf_event_index(struct perf_event *event) 2650static int perf_event_index(struct perf_event *event)
2346{ 2651{
2652 if (event->hw.state & PERF_HES_STOPPED)
2653 return 0;
2654
2347 if (event->state != PERF_EVENT_STATE_ACTIVE) 2655 if (event->state != PERF_EVENT_STATE_ACTIVE)
2348 return 0; 2656 return 0;
2349 2657
@@ -2847,16 +3155,7 @@ void perf_event_wakeup(struct perf_event *event)
2847 } 3155 }
2848} 3156}
2849 3157
2850/* 3158static void perf_pending_event(struct irq_work *entry)
2851 * Pending wakeups
2852 *
2853 * Handle the case where we need to wakeup up from NMI (or rq->lock) context.
2854 *
2855 * The NMI bit means we cannot possibly take locks. Therefore, maintain a
2856 * single linked list and use cmpxchg() to add entries lockless.
2857 */
2858
2859static void perf_pending_event(struct perf_pending_entry *entry)
2860{ 3159{
2861 struct perf_event *event = container_of(entry, 3160 struct perf_event *event = container_of(entry,
2862 struct perf_event, pending); 3161 struct perf_event, pending);
@@ -2872,99 +3171,6 @@ static void perf_pending_event(struct perf_pending_entry *entry)
2872 } 3171 }
2873} 3172}
2874 3173
2875#define PENDING_TAIL ((struct perf_pending_entry *)-1UL)
2876
2877static DEFINE_PER_CPU(struct perf_pending_entry *, perf_pending_head) = {
2878 PENDING_TAIL,
2879};
2880
2881static void perf_pending_queue(struct perf_pending_entry *entry,
2882 void (*func)(struct perf_pending_entry *))
2883{
2884 struct perf_pending_entry **head;
2885
2886 if (cmpxchg(&entry->next, NULL, PENDING_TAIL) != NULL)
2887 return;
2888
2889 entry->func = func;
2890
2891 head = &get_cpu_var(perf_pending_head);
2892
2893 do {
2894 entry->next = *head;
2895 } while (cmpxchg(head, entry->next, entry) != entry->next);
2896
2897 set_perf_event_pending();
2898
2899 put_cpu_var(perf_pending_head);
2900}
2901
2902static int __perf_pending_run(void)
2903{
2904 struct perf_pending_entry *list;
2905 int nr = 0;
2906
2907 list = xchg(&__get_cpu_var(perf_pending_head), PENDING_TAIL);
2908 while (list != PENDING_TAIL) {
2909 void (*func)(struct perf_pending_entry *);
2910 struct perf_pending_entry *entry = list;
2911
2912 list = list->next;
2913
2914 func = entry->func;
2915 entry->next = NULL;
2916 /*
2917 * Ensure we observe the unqueue before we issue the wakeup,
2918 * so that we won't be waiting forever.
2919 * -- see perf_not_pending().
2920 */
2921 smp_wmb();
2922
2923 func(entry);
2924 nr++;
2925 }
2926
2927 return nr;
2928}
2929
2930static inline int perf_not_pending(struct perf_event *event)
2931{
2932 /*
2933 * If we flush on whatever cpu we run, there is a chance we don't
2934 * need to wait.
2935 */
2936 get_cpu();
2937 __perf_pending_run();
2938 put_cpu();
2939
2940 /*
2941 * Ensure we see the proper queue state before going to sleep
2942 * so that we do not miss the wakeup. -- see perf_pending_handle()
2943 */
2944 smp_rmb();
2945 return event->pending.next == NULL;
2946}
2947
2948static void perf_pending_sync(struct perf_event *event)
2949{
2950 wait_event(event->waitq, perf_not_pending(event));
2951}
2952
2953void perf_event_do_pending(void)
2954{
2955 __perf_pending_run();
2956}
2957
2958/*
2959 * Callchain support -- arch specific
2960 */
2961
2962__weak struct perf_callchain_entry *perf_callchain(struct pt_regs *regs)
2963{
2964 return NULL;
2965}
2966
2967
2968/* 3174/*
2969 * We assume there is only KVM supporting the callbacks. 3175 * We assume there is only KVM supporting the callbacks.
2970 * Later on, we might change it to a list if there is 3176 * Later on, we might change it to a list if there is
@@ -3014,8 +3220,7 @@ static void perf_output_wakeup(struct perf_output_handle *handle)
3014 3220
3015 if (handle->nmi) { 3221 if (handle->nmi) {
3016 handle->event->pending_wakeup = 1; 3222 handle->event->pending_wakeup = 1;
3017 perf_pending_queue(&handle->event->pending, 3223 irq_work_queue(&handle->event->pending);
3018 perf_pending_event);
3019 } else 3224 } else
3020 perf_event_wakeup(handle->event); 3225 perf_event_wakeup(handle->event);
3021} 3226}
@@ -3071,7 +3276,7 @@ again:
3071 if (handle->wakeup != local_read(&buffer->wakeup)) 3276 if (handle->wakeup != local_read(&buffer->wakeup))
3072 perf_output_wakeup(handle); 3277 perf_output_wakeup(handle);
3073 3278
3074 out: 3279out:
3075 preempt_enable(); 3280 preempt_enable();
3076} 3281}
3077 3282
@@ -3459,14 +3664,20 @@ static void perf_event_output(struct perf_event *event, int nmi,
3459 struct perf_output_handle handle; 3664 struct perf_output_handle handle;
3460 struct perf_event_header header; 3665 struct perf_event_header header;
3461 3666
3667 /* protect the callchain buffers */
3668 rcu_read_lock();
3669
3462 perf_prepare_sample(&header, data, event, regs); 3670 perf_prepare_sample(&header, data, event, regs);
3463 3671
3464 if (perf_output_begin(&handle, event, header.size, nmi, 1)) 3672 if (perf_output_begin(&handle, event, header.size, nmi, 1))
3465 return; 3673 goto exit;
3466 3674
3467 perf_output_sample(&handle, &header, data, event); 3675 perf_output_sample(&handle, &header, data, event);
3468 3676
3469 perf_output_end(&handle); 3677 perf_output_end(&handle);
3678
3679exit:
3680 rcu_read_unlock();
3470} 3681}
3471 3682
3472/* 3683/*
@@ -3580,16 +3791,27 @@ static void perf_event_task_ctx(struct perf_event_context *ctx,
3580static void perf_event_task_event(struct perf_task_event *task_event) 3791static void perf_event_task_event(struct perf_task_event *task_event)
3581{ 3792{
3582 struct perf_cpu_context *cpuctx; 3793 struct perf_cpu_context *cpuctx;
3583 struct perf_event_context *ctx = task_event->task_ctx; 3794 struct perf_event_context *ctx;
3795 struct pmu *pmu;
3796 int ctxn;
3584 3797
3585 rcu_read_lock(); 3798 rcu_read_lock();
3586 cpuctx = &get_cpu_var(perf_cpu_context); 3799 list_for_each_entry_rcu(pmu, &pmus, entry) {
3587 perf_event_task_ctx(&cpuctx->ctx, task_event); 3800 cpuctx = get_cpu_ptr(pmu->pmu_cpu_context);
3588 if (!ctx) 3801 perf_event_task_ctx(&cpuctx->ctx, task_event);
3589 ctx = rcu_dereference(current->perf_event_ctxp); 3802
3590 if (ctx) 3803 ctx = task_event->task_ctx;
3591 perf_event_task_ctx(ctx, task_event); 3804 if (!ctx) {
3592 put_cpu_var(perf_cpu_context); 3805 ctxn = pmu->task_ctx_nr;
3806 if (ctxn < 0)
3807 goto next;
3808 ctx = rcu_dereference(current->perf_event_ctxp[ctxn]);
3809 }
3810 if (ctx)
3811 perf_event_task_ctx(ctx, task_event);
3812next:
3813 put_cpu_ptr(pmu->pmu_cpu_context);
3814 }
3593 rcu_read_unlock(); 3815 rcu_read_unlock();
3594} 3816}
3595 3817
@@ -3694,8 +3916,10 @@ static void perf_event_comm_event(struct perf_comm_event *comm_event)
3694{ 3916{
3695 struct perf_cpu_context *cpuctx; 3917 struct perf_cpu_context *cpuctx;
3696 struct perf_event_context *ctx; 3918 struct perf_event_context *ctx;
3697 unsigned int size;
3698 char comm[TASK_COMM_LEN]; 3919 char comm[TASK_COMM_LEN];
3920 unsigned int size;
3921 struct pmu *pmu;
3922 int ctxn;
3699 3923
3700 memset(comm, 0, sizeof(comm)); 3924 memset(comm, 0, sizeof(comm));
3701 strlcpy(comm, comm_event->task->comm, sizeof(comm)); 3925 strlcpy(comm, comm_event->task->comm, sizeof(comm));
@@ -3707,21 +3931,36 @@ static void perf_event_comm_event(struct perf_comm_event *comm_event)
3707 comm_event->event_id.header.size = sizeof(comm_event->event_id) + size; 3931 comm_event->event_id.header.size = sizeof(comm_event->event_id) + size;
3708 3932
3709 rcu_read_lock(); 3933 rcu_read_lock();
3710 cpuctx = &get_cpu_var(perf_cpu_context); 3934 list_for_each_entry_rcu(pmu, &pmus, entry) {
3711 perf_event_comm_ctx(&cpuctx->ctx, comm_event); 3935 cpuctx = get_cpu_ptr(pmu->pmu_cpu_context);
3712 ctx = rcu_dereference(current->perf_event_ctxp); 3936 perf_event_comm_ctx(&cpuctx->ctx, comm_event);
3713 if (ctx) 3937
3714 perf_event_comm_ctx(ctx, comm_event); 3938 ctxn = pmu->task_ctx_nr;
3715 put_cpu_var(perf_cpu_context); 3939 if (ctxn < 0)
3940 goto next;
3941
3942 ctx = rcu_dereference(current->perf_event_ctxp[ctxn]);
3943 if (ctx)
3944 perf_event_comm_ctx(ctx, comm_event);
3945next:
3946 put_cpu_ptr(pmu->pmu_cpu_context);
3947 }
3716 rcu_read_unlock(); 3948 rcu_read_unlock();
3717} 3949}
3718 3950
3719void perf_event_comm(struct task_struct *task) 3951void perf_event_comm(struct task_struct *task)
3720{ 3952{
3721 struct perf_comm_event comm_event; 3953 struct perf_comm_event comm_event;
3954 struct perf_event_context *ctx;
3955 int ctxn;
3722 3956
3723 if (task->perf_event_ctxp) 3957 for_each_task_context_nr(ctxn) {
3724 perf_event_enable_on_exec(task); 3958 ctx = task->perf_event_ctxp[ctxn];
3959 if (!ctx)
3960 continue;
3961
3962 perf_event_enable_on_exec(ctx);
3963 }
3725 3964
3726 if (!atomic_read(&nr_comm_events)) 3965 if (!atomic_read(&nr_comm_events))
3727 return; 3966 return;
@@ -3823,6 +4062,8 @@ static void perf_event_mmap_event(struct perf_mmap_event *mmap_event)
3823 char tmp[16]; 4062 char tmp[16];
3824 char *buf = NULL; 4063 char *buf = NULL;
3825 const char *name; 4064 const char *name;
4065 struct pmu *pmu;
4066 int ctxn;
3826 4067
3827 memset(tmp, 0, sizeof(tmp)); 4068 memset(tmp, 0, sizeof(tmp));
3828 4069
@@ -3875,12 +4116,23 @@ got_name:
3875 mmap_event->event_id.header.size = sizeof(mmap_event->event_id) + size; 4116 mmap_event->event_id.header.size = sizeof(mmap_event->event_id) + size;
3876 4117
3877 rcu_read_lock(); 4118 rcu_read_lock();
3878 cpuctx = &get_cpu_var(perf_cpu_context); 4119 list_for_each_entry_rcu(pmu, &pmus, entry) {
3879 perf_event_mmap_ctx(&cpuctx->ctx, mmap_event, vma->vm_flags & VM_EXEC); 4120 cpuctx = get_cpu_ptr(pmu->pmu_cpu_context);
3880 ctx = rcu_dereference(current->perf_event_ctxp); 4121 perf_event_mmap_ctx(&cpuctx->ctx, mmap_event,
3881 if (ctx) 4122 vma->vm_flags & VM_EXEC);
3882 perf_event_mmap_ctx(ctx, mmap_event, vma->vm_flags & VM_EXEC); 4123
3883 put_cpu_var(perf_cpu_context); 4124 ctxn = pmu->task_ctx_nr;
4125 if (ctxn < 0)
4126 goto next;
4127
4128 ctx = rcu_dereference(current->perf_event_ctxp[ctxn]);
4129 if (ctx) {
4130 perf_event_mmap_ctx(ctx, mmap_event,
4131 vma->vm_flags & VM_EXEC);
4132 }
4133next:
4134 put_cpu_ptr(pmu->pmu_cpu_context);
4135 }
3884 rcu_read_unlock(); 4136 rcu_read_unlock();
3885 4137
3886 kfree(buf); 4138 kfree(buf);
@@ -3962,8 +4214,6 @@ static int __perf_event_overflow(struct perf_event *event, int nmi,
3962 struct hw_perf_event *hwc = &event->hw; 4214 struct hw_perf_event *hwc = &event->hw;
3963 int ret = 0; 4215 int ret = 0;
3964 4216
3965 throttle = (throttle && event->pmu->unthrottle != NULL);
3966
3967 if (!throttle) { 4217 if (!throttle) {
3968 hwc->interrupts++; 4218 hwc->interrupts++;
3969 } else { 4219 } else {
@@ -4006,8 +4256,7 @@ static int __perf_event_overflow(struct perf_event *event, int nmi,
4006 event->pending_kill = POLL_HUP; 4256 event->pending_kill = POLL_HUP;
4007 if (nmi) { 4257 if (nmi) {
4008 event->pending_disable = 1; 4258 event->pending_disable = 1;
4009 perf_pending_queue(&event->pending, 4259 irq_work_queue(&event->pending);
4010 perf_pending_event);
4011 } else 4260 } else
4012 perf_event_disable(event); 4261 perf_event_disable(event);
4013 } 4262 }
@@ -4031,6 +4280,17 @@ int perf_event_overflow(struct perf_event *event, int nmi,
4031 * Generic software event infrastructure 4280 * Generic software event infrastructure
4032 */ 4281 */
4033 4282
4283struct swevent_htable {
4284 struct swevent_hlist *swevent_hlist;
4285 struct mutex hlist_mutex;
4286 int hlist_refcount;
4287
4288 /* Recursion avoidance in each contexts */
4289 int recursion[PERF_NR_CONTEXTS];
4290};
4291
4292static DEFINE_PER_CPU(struct swevent_htable, swevent_htable);
4293
4034/* 4294/*
4035 * We directly increment event->count and keep a second value in 4295 * We directly increment event->count and keep a second value in
4036 * event->hw.period_left to count intervals. This period event 4296 * event->hw.period_left to count intervals. This period event
@@ -4088,7 +4348,7 @@ static void perf_swevent_overflow(struct perf_event *event, u64 overflow,
4088 } 4348 }
4089} 4349}
4090 4350
4091static void perf_swevent_add(struct perf_event *event, u64 nr, 4351static void perf_swevent_event(struct perf_event *event, u64 nr,
4092 int nmi, struct perf_sample_data *data, 4352 int nmi, struct perf_sample_data *data,
4093 struct pt_regs *regs) 4353 struct pt_regs *regs)
4094{ 4354{
@@ -4114,6 +4374,9 @@ static void perf_swevent_add(struct perf_event *event, u64 nr,
4114static int perf_exclude_event(struct perf_event *event, 4374static int perf_exclude_event(struct perf_event *event,
4115 struct pt_regs *regs) 4375 struct pt_regs *regs)
4116{ 4376{
4377 if (event->hw.state & PERF_HES_STOPPED)
4378 return 0;
4379
4117 if (regs) { 4380 if (regs) {
4118 if (event->attr.exclude_user && user_mode(regs)) 4381 if (event->attr.exclude_user && user_mode(regs))
4119 return 1; 4382 return 1;
@@ -4160,11 +4423,11 @@ __find_swevent_head(struct swevent_hlist *hlist, u64 type, u32 event_id)
4160 4423
4161/* For the read side: events when they trigger */ 4424/* For the read side: events when they trigger */
4162static inline struct hlist_head * 4425static inline struct hlist_head *
4163find_swevent_head_rcu(struct perf_cpu_context *ctx, u64 type, u32 event_id) 4426find_swevent_head_rcu(struct swevent_htable *swhash, u64 type, u32 event_id)
4164{ 4427{
4165 struct swevent_hlist *hlist; 4428 struct swevent_hlist *hlist;
4166 4429
4167 hlist = rcu_dereference(ctx->swevent_hlist); 4430 hlist = rcu_dereference(swhash->swevent_hlist);
4168 if (!hlist) 4431 if (!hlist)
4169 return NULL; 4432 return NULL;
4170 4433
@@ -4173,7 +4436,7 @@ find_swevent_head_rcu(struct perf_cpu_context *ctx, u64 type, u32 event_id)
4173 4436
4174/* For the event head insertion and removal in the hlist */ 4437/* For the event head insertion and removal in the hlist */
4175static inline struct hlist_head * 4438static inline struct hlist_head *
4176find_swevent_head(struct perf_cpu_context *ctx, struct perf_event *event) 4439find_swevent_head(struct swevent_htable *swhash, struct perf_event *event)
4177{ 4440{
4178 struct swevent_hlist *hlist; 4441 struct swevent_hlist *hlist;
4179 u32 event_id = event->attr.config; 4442 u32 event_id = event->attr.config;
@@ -4184,7 +4447,7 @@ find_swevent_head(struct perf_cpu_context *ctx, struct perf_event *event)
4184 * and release. Which makes the protected version suitable here. 4447 * and release. Which makes the protected version suitable here.
4185 * The context lock guarantees that. 4448 * The context lock guarantees that.
4186 */ 4449 */
4187 hlist = rcu_dereference_protected(ctx->swevent_hlist, 4450 hlist = rcu_dereference_protected(swhash->swevent_hlist,
4188 lockdep_is_held(&event->ctx->lock)); 4451 lockdep_is_held(&event->ctx->lock));
4189 if (!hlist) 4452 if (!hlist)
4190 return NULL; 4453 return NULL;
@@ -4197,23 +4460,19 @@ static void do_perf_sw_event(enum perf_type_id type, u32 event_id,
4197 struct perf_sample_data *data, 4460 struct perf_sample_data *data,
4198 struct pt_regs *regs) 4461 struct pt_regs *regs)
4199{ 4462{
4200 struct perf_cpu_context *cpuctx; 4463 struct swevent_htable *swhash = &__get_cpu_var(swevent_htable);
4201 struct perf_event *event; 4464 struct perf_event *event;
4202 struct hlist_node *node; 4465 struct hlist_node *node;
4203 struct hlist_head *head; 4466 struct hlist_head *head;
4204 4467
4205 cpuctx = &__get_cpu_var(perf_cpu_context);
4206
4207 rcu_read_lock(); 4468 rcu_read_lock();
4208 4469 head = find_swevent_head_rcu(swhash, type, event_id);
4209 head = find_swevent_head_rcu(cpuctx, type, event_id);
4210
4211 if (!head) 4470 if (!head)
4212 goto end; 4471 goto end;
4213 4472
4214 hlist_for_each_entry_rcu(event, node, head, hlist_entry) { 4473 hlist_for_each_entry_rcu(event, node, head, hlist_entry) {
4215 if (perf_swevent_match(event, type, event_id, data, regs)) 4474 if (perf_swevent_match(event, type, event_id, data, regs))
4216 perf_swevent_add(event, nr, nmi, data, regs); 4475 perf_swevent_event(event, nr, nmi, data, regs);
4217 } 4476 }
4218end: 4477end:
4219 rcu_read_unlock(); 4478 rcu_read_unlock();
@@ -4221,33 +4480,17 @@ end:
4221 4480
4222int perf_swevent_get_recursion_context(void) 4481int perf_swevent_get_recursion_context(void)
4223{ 4482{
4224 struct perf_cpu_context *cpuctx = &__get_cpu_var(perf_cpu_context); 4483 struct swevent_htable *swhash = &__get_cpu_var(swevent_htable);
4225 int rctx;
4226
4227 if (in_nmi())
4228 rctx = 3;
4229 else if (in_irq())
4230 rctx = 2;
4231 else if (in_softirq())
4232 rctx = 1;
4233 else
4234 rctx = 0;
4235
4236 if (cpuctx->recursion[rctx])
4237 return -1;
4238
4239 cpuctx->recursion[rctx]++;
4240 barrier();
4241 4484
4242 return rctx; 4485 return get_recursion_context(swhash->recursion);
4243} 4486}
4244EXPORT_SYMBOL_GPL(perf_swevent_get_recursion_context); 4487EXPORT_SYMBOL_GPL(perf_swevent_get_recursion_context);
4245 4488
4246void inline perf_swevent_put_recursion_context(int rctx) 4489void inline perf_swevent_put_recursion_context(int rctx)
4247{ 4490{
4248 struct perf_cpu_context *cpuctx = &__get_cpu_var(perf_cpu_context); 4491 struct swevent_htable *swhash = &__get_cpu_var(swevent_htable);
4249 barrier(); 4492
4250 cpuctx->recursion[rctx]--; 4493 put_recursion_context(swhash->recursion, rctx);
4251} 4494}
4252 4495
4253void __perf_sw_event(u32 event_id, u64 nr, int nmi, 4496void __perf_sw_event(u32 event_id, u64 nr, int nmi,
@@ -4273,20 +4516,20 @@ static void perf_swevent_read(struct perf_event *event)
4273{ 4516{
4274} 4517}
4275 4518
4276static int perf_swevent_enable(struct perf_event *event) 4519static int perf_swevent_add(struct perf_event *event, int flags)
4277{ 4520{
4521 struct swevent_htable *swhash = &__get_cpu_var(swevent_htable);
4278 struct hw_perf_event *hwc = &event->hw; 4522 struct hw_perf_event *hwc = &event->hw;
4279 struct perf_cpu_context *cpuctx;
4280 struct hlist_head *head; 4523 struct hlist_head *head;
4281 4524
4282 cpuctx = &__get_cpu_var(perf_cpu_context);
4283
4284 if (hwc->sample_period) { 4525 if (hwc->sample_period) {
4285 hwc->last_period = hwc->sample_period; 4526 hwc->last_period = hwc->sample_period;
4286 perf_swevent_set_period(event); 4527 perf_swevent_set_period(event);
4287 } 4528 }
4288 4529
4289 head = find_swevent_head(cpuctx, event); 4530 hwc->state = !(flags & PERF_EF_START);
4531
4532 head = find_swevent_head(swhash, event);
4290 if (WARN_ON_ONCE(!head)) 4533 if (WARN_ON_ONCE(!head))
4291 return -EINVAL; 4534 return -EINVAL;
4292 4535
@@ -4295,202 +4538,27 @@ static int perf_swevent_enable(struct perf_event *event)
4295 return 0; 4538 return 0;
4296} 4539}
4297 4540
4298static void perf_swevent_disable(struct perf_event *event) 4541static void perf_swevent_del(struct perf_event *event, int flags)
4299{ 4542{
4300 hlist_del_rcu(&event->hlist_entry); 4543 hlist_del_rcu(&event->hlist_entry);
4301} 4544}
4302 4545
4303static void perf_swevent_void(struct perf_event *event) 4546static void perf_swevent_start(struct perf_event *event, int flags)
4304{
4305}
4306
4307static int perf_swevent_int(struct perf_event *event)
4308{
4309 return 0;
4310}
4311
4312static const struct pmu perf_ops_generic = {
4313 .enable = perf_swevent_enable,
4314 .disable = perf_swevent_disable,
4315 .start = perf_swevent_int,
4316 .stop = perf_swevent_void,
4317 .read = perf_swevent_read,
4318 .unthrottle = perf_swevent_void, /* hwc->interrupts already reset */
4319};
4320
4321/*
4322 * hrtimer based swevent callback
4323 */
4324
4325static enum hrtimer_restart perf_swevent_hrtimer(struct hrtimer *hrtimer)
4326{ 4547{
4327 enum hrtimer_restart ret = HRTIMER_RESTART; 4548 event->hw.state = 0;
4328 struct perf_sample_data data;
4329 struct pt_regs *regs;
4330 struct perf_event *event;
4331 u64 period;
4332
4333 event = container_of(hrtimer, struct perf_event, hw.hrtimer);
4334 event->pmu->read(event);
4335
4336 perf_sample_data_init(&data, 0);
4337 data.period = event->hw.last_period;
4338 regs = get_irq_regs();
4339
4340 if (regs && !perf_exclude_event(event, regs)) {
4341 if (!(event->attr.exclude_idle && current->pid == 0))
4342 if (perf_event_overflow(event, 0, &data, regs))
4343 ret = HRTIMER_NORESTART;
4344 }
4345
4346 period = max_t(u64, 10000, event->hw.sample_period);
4347 hrtimer_forward_now(hrtimer, ns_to_ktime(period));
4348
4349 return ret;
4350} 4549}
4351 4550
4352static void perf_swevent_start_hrtimer(struct perf_event *event) 4551static void perf_swevent_stop(struct perf_event *event, int flags)
4353{ 4552{
4354 struct hw_perf_event *hwc = &event->hw; 4553 event->hw.state = PERF_HES_STOPPED;
4355
4356 hrtimer_init(&hwc->hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
4357 hwc->hrtimer.function = perf_swevent_hrtimer;
4358 if (hwc->sample_period) {
4359 u64 period;
4360
4361 if (hwc->remaining) {
4362 if (hwc->remaining < 0)
4363 period = 10000;
4364 else
4365 period = hwc->remaining;
4366 hwc->remaining = 0;
4367 } else {
4368 period = max_t(u64, 10000, hwc->sample_period);
4369 }
4370 __hrtimer_start_range_ns(&hwc->hrtimer,
4371 ns_to_ktime(period), 0,
4372 HRTIMER_MODE_REL, 0);
4373 }
4374}
4375
4376static void perf_swevent_cancel_hrtimer(struct perf_event *event)
4377{
4378 struct hw_perf_event *hwc = &event->hw;
4379
4380 if (hwc->sample_period) {
4381 ktime_t remaining = hrtimer_get_remaining(&hwc->hrtimer);
4382 hwc->remaining = ktime_to_ns(remaining);
4383
4384 hrtimer_cancel(&hwc->hrtimer);
4385 }
4386}
4387
4388/*
4389 * Software event: cpu wall time clock
4390 */
4391
4392static void cpu_clock_perf_event_update(struct perf_event *event)
4393{
4394 int cpu = raw_smp_processor_id();
4395 s64 prev;
4396 u64 now;
4397
4398 now = cpu_clock(cpu);
4399 prev = local64_xchg(&event->hw.prev_count, now);
4400 local64_add(now - prev, &event->count);
4401}
4402
4403static int cpu_clock_perf_event_enable(struct perf_event *event)
4404{
4405 struct hw_perf_event *hwc = &event->hw;
4406 int cpu = raw_smp_processor_id();
4407
4408 local64_set(&hwc->prev_count, cpu_clock(cpu));
4409 perf_swevent_start_hrtimer(event);
4410
4411 return 0;
4412}
4413
4414static void cpu_clock_perf_event_disable(struct perf_event *event)
4415{
4416 perf_swevent_cancel_hrtimer(event);
4417 cpu_clock_perf_event_update(event);
4418}
4419
4420static void cpu_clock_perf_event_read(struct perf_event *event)
4421{
4422 cpu_clock_perf_event_update(event);
4423}
4424
4425static const struct pmu perf_ops_cpu_clock = {
4426 .enable = cpu_clock_perf_event_enable,
4427 .disable = cpu_clock_perf_event_disable,
4428 .read = cpu_clock_perf_event_read,
4429};
4430
4431/*
4432 * Software event: task time clock
4433 */
4434
4435static void task_clock_perf_event_update(struct perf_event *event, u64 now)
4436{
4437 u64 prev;
4438 s64 delta;
4439
4440 prev = local64_xchg(&event->hw.prev_count, now);
4441 delta = now - prev;
4442 local64_add(delta, &event->count);
4443}
4444
4445static int task_clock_perf_event_enable(struct perf_event *event)
4446{
4447 struct hw_perf_event *hwc = &event->hw;
4448 u64 now;
4449
4450 now = event->ctx->time;
4451
4452 local64_set(&hwc->prev_count, now);
4453
4454 perf_swevent_start_hrtimer(event);
4455
4456 return 0;
4457}
4458
4459static void task_clock_perf_event_disable(struct perf_event *event)
4460{
4461 perf_swevent_cancel_hrtimer(event);
4462 task_clock_perf_event_update(event, event->ctx->time);
4463
4464}
4465
4466static void task_clock_perf_event_read(struct perf_event *event)
4467{
4468 u64 time;
4469
4470 if (!in_nmi()) {
4471 update_context_time(event->ctx);
4472 time = event->ctx->time;
4473 } else {
4474 u64 now = perf_clock();
4475 u64 delta = now - event->ctx->timestamp;
4476 time = event->ctx->time + delta;
4477 }
4478
4479 task_clock_perf_event_update(event, time);
4480} 4554}
4481 4555
4482static const struct pmu perf_ops_task_clock = {
4483 .enable = task_clock_perf_event_enable,
4484 .disable = task_clock_perf_event_disable,
4485 .read = task_clock_perf_event_read,
4486};
4487
4488/* Deref the hlist from the update side */ 4556/* Deref the hlist from the update side */
4489static inline struct swevent_hlist * 4557static inline struct swevent_hlist *
4490swevent_hlist_deref(struct perf_cpu_context *cpuctx) 4558swevent_hlist_deref(struct swevent_htable *swhash)
4491{ 4559{
4492 return rcu_dereference_protected(cpuctx->swevent_hlist, 4560 return rcu_dereference_protected(swhash->swevent_hlist,
4493 lockdep_is_held(&cpuctx->hlist_mutex)); 4561 lockdep_is_held(&swhash->hlist_mutex));
4494} 4562}
4495 4563
4496static void swevent_hlist_release_rcu(struct rcu_head *rcu_head) 4564static void swevent_hlist_release_rcu(struct rcu_head *rcu_head)
@@ -4501,27 +4569,27 @@ static void swevent_hlist_release_rcu(struct rcu_head *rcu_head)
4501 kfree(hlist); 4569 kfree(hlist);
4502} 4570}
4503 4571
4504static void swevent_hlist_release(struct perf_cpu_context *cpuctx) 4572static void swevent_hlist_release(struct swevent_htable *swhash)
4505{ 4573{
4506 struct swevent_hlist *hlist = swevent_hlist_deref(cpuctx); 4574 struct swevent_hlist *hlist = swevent_hlist_deref(swhash);
4507 4575
4508 if (!hlist) 4576 if (!hlist)
4509 return; 4577 return;
4510 4578
4511 rcu_assign_pointer(cpuctx->swevent_hlist, NULL); 4579 rcu_assign_pointer(swhash->swevent_hlist, NULL);
4512 call_rcu(&hlist->rcu_head, swevent_hlist_release_rcu); 4580 call_rcu(&hlist->rcu_head, swevent_hlist_release_rcu);
4513} 4581}
4514 4582
4515static void swevent_hlist_put_cpu(struct perf_event *event, int cpu) 4583static void swevent_hlist_put_cpu(struct perf_event *event, int cpu)
4516{ 4584{
4517 struct perf_cpu_context *cpuctx = &per_cpu(perf_cpu_context, cpu); 4585 struct swevent_htable *swhash = &per_cpu(swevent_htable, cpu);
4518 4586
4519 mutex_lock(&cpuctx->hlist_mutex); 4587 mutex_lock(&swhash->hlist_mutex);
4520 4588
4521 if (!--cpuctx->hlist_refcount) 4589 if (!--swhash->hlist_refcount)
4522 swevent_hlist_release(cpuctx); 4590 swevent_hlist_release(swhash);
4523 4591
4524 mutex_unlock(&cpuctx->hlist_mutex); 4592 mutex_unlock(&swhash->hlist_mutex);
4525} 4593}
4526 4594
4527static void swevent_hlist_put(struct perf_event *event) 4595static void swevent_hlist_put(struct perf_event *event)
@@ -4539,12 +4607,12 @@ static void swevent_hlist_put(struct perf_event *event)
4539 4607
4540static int swevent_hlist_get_cpu(struct perf_event *event, int cpu) 4608static int swevent_hlist_get_cpu(struct perf_event *event, int cpu)
4541{ 4609{
4542 struct perf_cpu_context *cpuctx = &per_cpu(perf_cpu_context, cpu); 4610 struct swevent_htable *swhash = &per_cpu(swevent_htable, cpu);
4543 int err = 0; 4611 int err = 0;
4544 4612
4545 mutex_lock(&cpuctx->hlist_mutex); 4613 mutex_lock(&swhash->hlist_mutex);
4546 4614
4547 if (!swevent_hlist_deref(cpuctx) && cpu_online(cpu)) { 4615 if (!swevent_hlist_deref(swhash) && cpu_online(cpu)) {
4548 struct swevent_hlist *hlist; 4616 struct swevent_hlist *hlist;
4549 4617
4550 hlist = kzalloc(sizeof(*hlist), GFP_KERNEL); 4618 hlist = kzalloc(sizeof(*hlist), GFP_KERNEL);
@@ -4552,11 +4620,11 @@ static int swevent_hlist_get_cpu(struct perf_event *event, int cpu)
4552 err = -ENOMEM; 4620 err = -ENOMEM;
4553 goto exit; 4621 goto exit;
4554 } 4622 }
4555 rcu_assign_pointer(cpuctx->swevent_hlist, hlist); 4623 rcu_assign_pointer(swhash->swevent_hlist, hlist);
4556 } 4624 }
4557 cpuctx->hlist_refcount++; 4625 swhash->hlist_refcount++;
4558 exit: 4626exit:
4559 mutex_unlock(&cpuctx->hlist_mutex); 4627 mutex_unlock(&swhash->hlist_mutex);
4560 4628
4561 return err; 4629 return err;
4562} 4630}
@@ -4580,7 +4648,7 @@ static int swevent_hlist_get(struct perf_event *event)
4580 put_online_cpus(); 4648 put_online_cpus();
4581 4649
4582 return 0; 4650 return 0;
4583 fail: 4651fail:
4584 for_each_possible_cpu(cpu) { 4652 for_each_possible_cpu(cpu) {
4585 if (cpu == failed_cpu) 4653 if (cpu == failed_cpu)
4586 break; 4654 break;
@@ -4591,17 +4659,64 @@ static int swevent_hlist_get(struct perf_event *event)
4591 return err; 4659 return err;
4592} 4660}
4593 4661
4594#ifdef CONFIG_EVENT_TRACING 4662atomic_t perf_swevent_enabled[PERF_COUNT_SW_MAX];
4663
4664static void sw_perf_event_destroy(struct perf_event *event)
4665{
4666 u64 event_id = event->attr.config;
4667
4668 WARN_ON(event->parent);
4669
4670 jump_label_dec(&perf_swevent_enabled[event_id]);
4671 swevent_hlist_put(event);
4672}
4673
4674static int perf_swevent_init(struct perf_event *event)
4675{
4676 int event_id = event->attr.config;
4677
4678 if (event->attr.type != PERF_TYPE_SOFTWARE)
4679 return -ENOENT;
4680
4681 switch (event_id) {
4682 case PERF_COUNT_SW_CPU_CLOCK:
4683 case PERF_COUNT_SW_TASK_CLOCK:
4684 return -ENOENT;
4685
4686 default:
4687 break;
4688 }
4689
4690 if (event_id > PERF_COUNT_SW_MAX)
4691 return -ENOENT;
4595 4692
4596static const struct pmu perf_ops_tracepoint = { 4693 if (!event->parent) {
4597 .enable = perf_trace_enable, 4694 int err;
4598 .disable = perf_trace_disable, 4695
4599 .start = perf_swevent_int, 4696 err = swevent_hlist_get(event);
4600 .stop = perf_swevent_void, 4697 if (err)
4698 return err;
4699
4700 jump_label_inc(&perf_swevent_enabled[event_id]);
4701 event->destroy = sw_perf_event_destroy;
4702 }
4703
4704 return 0;
4705}
4706
4707static struct pmu perf_swevent = {
4708 .task_ctx_nr = perf_sw_context,
4709
4710 .event_init = perf_swevent_init,
4711 .add = perf_swevent_add,
4712 .del = perf_swevent_del,
4713 .start = perf_swevent_start,
4714 .stop = perf_swevent_stop,
4601 .read = perf_swevent_read, 4715 .read = perf_swevent_read,
4602 .unthrottle = perf_swevent_void,
4603}; 4716};
4604 4717
4718#ifdef CONFIG_EVENT_TRACING
4719
4605static int perf_tp_filter_match(struct perf_event *event, 4720static int perf_tp_filter_match(struct perf_event *event,
4606 struct perf_sample_data *data) 4721 struct perf_sample_data *data)
4607{ 4722{
@@ -4645,7 +4760,7 @@ void perf_tp_event(u64 addr, u64 count, void *record, int entry_size,
4645 4760
4646 hlist_for_each_entry_rcu(event, node, head, hlist_entry) { 4761 hlist_for_each_entry_rcu(event, node, head, hlist_entry) {
4647 if (perf_tp_event_match(event, &data, regs)) 4762 if (perf_tp_event_match(event, &data, regs))
4648 perf_swevent_add(event, count, 1, &data, regs); 4763 perf_swevent_event(event, count, 1, &data, regs);
4649 } 4764 }
4650 4765
4651 perf_swevent_put_recursion_context(rctx); 4766 perf_swevent_put_recursion_context(rctx);
@@ -4657,10 +4772,13 @@ static void tp_perf_event_destroy(struct perf_event *event)
4657 perf_trace_destroy(event); 4772 perf_trace_destroy(event);
4658} 4773}
4659 4774
4660static const struct pmu *tp_perf_event_init(struct perf_event *event) 4775static int perf_tp_event_init(struct perf_event *event)
4661{ 4776{
4662 int err; 4777 int err;
4663 4778
4779 if (event->attr.type != PERF_TYPE_TRACEPOINT)
4780 return -ENOENT;
4781
4664 /* 4782 /*
4665 * Raw tracepoint data is a severe data leak, only allow root to 4783 * Raw tracepoint data is a severe data leak, only allow root to
4666 * have these. 4784 * have these.
@@ -4668,15 +4786,31 @@ static const struct pmu *tp_perf_event_init(struct perf_event *event)
4668 if ((event->attr.sample_type & PERF_SAMPLE_RAW) && 4786 if ((event->attr.sample_type & PERF_SAMPLE_RAW) &&
4669 perf_paranoid_tracepoint_raw() && 4787 perf_paranoid_tracepoint_raw() &&
4670 !capable(CAP_SYS_ADMIN)) 4788 !capable(CAP_SYS_ADMIN))
4671 return ERR_PTR(-EPERM); 4789 return -EPERM;
4672 4790
4673 err = perf_trace_init(event); 4791 err = perf_trace_init(event);
4674 if (err) 4792 if (err)
4675 return NULL; 4793 return err;
4676 4794
4677 event->destroy = tp_perf_event_destroy; 4795 event->destroy = tp_perf_event_destroy;
4678 4796
4679 return &perf_ops_tracepoint; 4797 return 0;
4798}
4799
4800static struct pmu perf_tracepoint = {
4801 .task_ctx_nr = perf_sw_context,
4802
4803 .event_init = perf_tp_event_init,
4804 .add = perf_trace_add,
4805 .del = perf_trace_del,
4806 .start = perf_swevent_start,
4807 .stop = perf_swevent_stop,
4808 .read = perf_swevent_read,
4809};
4810
4811static inline void perf_tp_register(void)
4812{
4813 perf_pmu_register(&perf_tracepoint);
4680} 4814}
4681 4815
4682static int perf_event_set_filter(struct perf_event *event, void __user *arg) 4816static int perf_event_set_filter(struct perf_event *event, void __user *arg)
@@ -4704,9 +4838,8 @@ static void perf_event_free_filter(struct perf_event *event)
4704 4838
4705#else 4839#else
4706 4840
4707static const struct pmu *tp_perf_event_init(struct perf_event *event) 4841static inline void perf_tp_register(void)
4708{ 4842{
4709 return NULL;
4710} 4843}
4711 4844
4712static int perf_event_set_filter(struct perf_event *event, void __user *arg) 4845static int perf_event_set_filter(struct perf_event *event, void __user *arg)
@@ -4721,105 +4854,389 @@ static void perf_event_free_filter(struct perf_event *event)
4721#endif /* CONFIG_EVENT_TRACING */ 4854#endif /* CONFIG_EVENT_TRACING */
4722 4855
4723#ifdef CONFIG_HAVE_HW_BREAKPOINT 4856#ifdef CONFIG_HAVE_HW_BREAKPOINT
4724static void bp_perf_event_destroy(struct perf_event *event) 4857void perf_bp_event(struct perf_event *bp, void *data)
4725{ 4858{
4726 release_bp_slot(event); 4859 struct perf_sample_data sample;
4860 struct pt_regs *regs = data;
4861
4862 perf_sample_data_init(&sample, bp->attr.bp_addr);
4863
4864 if (!bp->hw.state && !perf_exclude_event(bp, regs))
4865 perf_swevent_event(bp, 1, 1, &sample, regs);
4727} 4866}
4867#endif
4868
4869/*
4870 * hrtimer based swevent callback
4871 */
4728 4872
4729static const struct pmu *bp_perf_event_init(struct perf_event *bp) 4873static enum hrtimer_restart perf_swevent_hrtimer(struct hrtimer *hrtimer)
4730{ 4874{
4731 int err; 4875 enum hrtimer_restart ret = HRTIMER_RESTART;
4876 struct perf_sample_data data;
4877 struct pt_regs *regs;
4878 struct perf_event *event;
4879 u64 period;
4732 4880
4733 err = register_perf_hw_breakpoint(bp); 4881 event = container_of(hrtimer, struct perf_event, hw.hrtimer);
4734 if (err) 4882 event->pmu->read(event);
4735 return ERR_PTR(err); 4883
4884 perf_sample_data_init(&data, 0);
4885 data.period = event->hw.last_period;
4886 regs = get_irq_regs();
4887
4888 if (regs && !perf_exclude_event(event, regs)) {
4889 if (!(event->attr.exclude_idle && current->pid == 0))
4890 if (perf_event_overflow(event, 0, &data, regs))
4891 ret = HRTIMER_NORESTART;
4892 }
4736 4893
4737 bp->destroy = bp_perf_event_destroy; 4894 period = max_t(u64, 10000, event->hw.sample_period);
4895 hrtimer_forward_now(hrtimer, ns_to_ktime(period));
4738 4896
4739 return &perf_ops_bp; 4897 return ret;
4740} 4898}
4741 4899
4742void perf_bp_event(struct perf_event *bp, void *data) 4900static void perf_swevent_start_hrtimer(struct perf_event *event)
4743{ 4901{
4744 struct perf_sample_data sample; 4902 struct hw_perf_event *hwc = &event->hw;
4745 struct pt_regs *regs = data;
4746 4903
4747 perf_sample_data_init(&sample, bp->attr.bp_addr); 4904 hrtimer_init(&hwc->hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
4905 hwc->hrtimer.function = perf_swevent_hrtimer;
4906 if (hwc->sample_period) {
4907 s64 period = local64_read(&hwc->period_left);
4908
4909 if (period) {
4910 if (period < 0)
4911 period = 10000;
4748 4912
4749 if (!perf_exclude_event(bp, regs)) 4913 local64_set(&hwc->period_left, 0);
4750 perf_swevent_add(bp, 1, 1, &sample, regs); 4914 } else {
4915 period = max_t(u64, 10000, hwc->sample_period);
4916 }
4917 __hrtimer_start_range_ns(&hwc->hrtimer,
4918 ns_to_ktime(period), 0,
4919 HRTIMER_MODE_REL_PINNED, 0);
4920 }
4751} 4921}
4752#else 4922
4753static const struct pmu *bp_perf_event_init(struct perf_event *bp) 4923static void perf_swevent_cancel_hrtimer(struct perf_event *event)
4754{ 4924{
4755 return NULL; 4925 struct hw_perf_event *hwc = &event->hw;
4926
4927 if (hwc->sample_period) {
4928 ktime_t remaining = hrtimer_get_remaining(&hwc->hrtimer);
4929 local64_set(&hwc->period_left, ktime_to_ns(remaining));
4930
4931 hrtimer_cancel(&hwc->hrtimer);
4932 }
4756} 4933}
4757 4934
4758void perf_bp_event(struct perf_event *bp, void *regs) 4935/*
4936 * Software event: cpu wall time clock
4937 */
4938
4939static void cpu_clock_event_update(struct perf_event *event)
4759{ 4940{
4941 s64 prev;
4942 u64 now;
4943
4944 now = local_clock();
4945 prev = local64_xchg(&event->hw.prev_count, now);
4946 local64_add(now - prev, &event->count);
4760} 4947}
4761#endif
4762 4948
4763atomic_t perf_swevent_enabled[PERF_COUNT_SW_MAX]; 4949static void cpu_clock_event_start(struct perf_event *event, int flags)
4950{
4951 local64_set(&event->hw.prev_count, local_clock());
4952 perf_swevent_start_hrtimer(event);
4953}
4764 4954
4765static void sw_perf_event_destroy(struct perf_event *event) 4955static void cpu_clock_event_stop(struct perf_event *event, int flags)
4766{ 4956{
4767 u64 event_id = event->attr.config; 4957 perf_swevent_cancel_hrtimer(event);
4958 cpu_clock_event_update(event);
4959}
4768 4960
4769 WARN_ON(event->parent); 4961static int cpu_clock_event_add(struct perf_event *event, int flags)
4962{
4963 if (flags & PERF_EF_START)
4964 cpu_clock_event_start(event, flags);
4770 4965
4771 atomic_dec(&perf_swevent_enabled[event_id]); 4966 return 0;
4772 swevent_hlist_put(event);
4773} 4967}
4774 4968
4775static const struct pmu *sw_perf_event_init(struct perf_event *event) 4969static void cpu_clock_event_del(struct perf_event *event, int flags)
4776{ 4970{
4777 const struct pmu *pmu = NULL; 4971 cpu_clock_event_stop(event, flags);
4778 u64 event_id = event->attr.config; 4972}
4973
4974static void cpu_clock_event_read(struct perf_event *event)
4975{
4976 cpu_clock_event_update(event);
4977}
4978
4979static int cpu_clock_event_init(struct perf_event *event)
4980{
4981 if (event->attr.type != PERF_TYPE_SOFTWARE)
4982 return -ENOENT;
4983
4984 if (event->attr.config != PERF_COUNT_SW_CPU_CLOCK)
4985 return -ENOENT;
4986
4987 return 0;
4988}
4989
4990static struct pmu perf_cpu_clock = {
4991 .task_ctx_nr = perf_sw_context,
4992
4993 .event_init = cpu_clock_event_init,
4994 .add = cpu_clock_event_add,
4995 .del = cpu_clock_event_del,
4996 .start = cpu_clock_event_start,
4997 .stop = cpu_clock_event_stop,
4998 .read = cpu_clock_event_read,
4999};
5000
5001/*
5002 * Software event: task time clock
5003 */
5004
5005static void task_clock_event_update(struct perf_event *event, u64 now)
5006{
5007 u64 prev;
5008 s64 delta;
5009
5010 prev = local64_xchg(&event->hw.prev_count, now);
5011 delta = now - prev;
5012 local64_add(delta, &event->count);
5013}
5014
5015static void task_clock_event_start(struct perf_event *event, int flags)
5016{
5017 local64_set(&event->hw.prev_count, event->ctx->time);
5018 perf_swevent_start_hrtimer(event);
5019}
4779 5020
5021static void task_clock_event_stop(struct perf_event *event, int flags)
5022{
5023 perf_swevent_cancel_hrtimer(event);
5024 task_clock_event_update(event, event->ctx->time);
5025}
5026
5027static int task_clock_event_add(struct perf_event *event, int flags)
5028{
5029 if (flags & PERF_EF_START)
5030 task_clock_event_start(event, flags);
5031
5032 return 0;
5033}
5034
5035static void task_clock_event_del(struct perf_event *event, int flags)
5036{
5037 task_clock_event_stop(event, PERF_EF_UPDATE);
5038}
5039
5040static void task_clock_event_read(struct perf_event *event)
5041{
5042 u64 time;
5043
5044 if (!in_nmi()) {
5045 update_context_time(event->ctx);
5046 time = event->ctx->time;
5047 } else {
5048 u64 now = perf_clock();
5049 u64 delta = now - event->ctx->timestamp;
5050 time = event->ctx->time + delta;
5051 }
5052
5053 task_clock_event_update(event, time);
5054}
5055
5056static int task_clock_event_init(struct perf_event *event)
5057{
5058 if (event->attr.type != PERF_TYPE_SOFTWARE)
5059 return -ENOENT;
5060
5061 if (event->attr.config != PERF_COUNT_SW_TASK_CLOCK)
5062 return -ENOENT;
5063
5064 return 0;
5065}
5066
5067static struct pmu perf_task_clock = {
5068 .task_ctx_nr = perf_sw_context,
5069
5070 .event_init = task_clock_event_init,
5071 .add = task_clock_event_add,
5072 .del = task_clock_event_del,
5073 .start = task_clock_event_start,
5074 .stop = task_clock_event_stop,
5075 .read = task_clock_event_read,
5076};
5077
5078static void perf_pmu_nop_void(struct pmu *pmu)
5079{
5080}
5081
5082static int perf_pmu_nop_int(struct pmu *pmu)
5083{
5084 return 0;
5085}
5086
5087static void perf_pmu_start_txn(struct pmu *pmu)
5088{
5089 perf_pmu_disable(pmu);
5090}
5091
5092static int perf_pmu_commit_txn(struct pmu *pmu)
5093{
5094 perf_pmu_enable(pmu);
5095 return 0;
5096}
5097
5098static void perf_pmu_cancel_txn(struct pmu *pmu)
5099{
5100 perf_pmu_enable(pmu);
5101}
5102
5103/*
5104 * Ensures all contexts with the same task_ctx_nr have the same
5105 * pmu_cpu_context too.
5106 */
5107static void *find_pmu_context(int ctxn)
5108{
5109 struct pmu *pmu;
5110
5111 if (ctxn < 0)
5112 return NULL;
5113
5114 list_for_each_entry(pmu, &pmus, entry) {
5115 if (pmu->task_ctx_nr == ctxn)
5116 return pmu->pmu_cpu_context;
5117 }
5118
5119 return NULL;
5120}
5121
5122static void free_pmu_context(void * __percpu cpu_context)
5123{
5124 struct pmu *pmu;
5125
5126 mutex_lock(&pmus_lock);
4780 /* 5127 /*
4781 * Software events (currently) can't in general distinguish 5128 * Like a real lame refcount.
4782 * between user, kernel and hypervisor events.
4783 * However, context switches and cpu migrations are considered
4784 * to be kernel events, and page faults are never hypervisor
4785 * events.
4786 */ 5129 */
4787 switch (event_id) { 5130 list_for_each_entry(pmu, &pmus, entry) {
4788 case PERF_COUNT_SW_CPU_CLOCK: 5131 if (pmu->pmu_cpu_context == cpu_context)
4789 pmu = &perf_ops_cpu_clock; 5132 goto out;
5133 }
4790 5134
4791 break; 5135 free_percpu(cpu_context);
4792 case PERF_COUNT_SW_TASK_CLOCK: 5136out:
4793 /* 5137 mutex_unlock(&pmus_lock);
4794 * If the user instantiates this as a per-cpu event, 5138}
4795 * use the cpu_clock event instead.
4796 */
4797 if (event->ctx->task)
4798 pmu = &perf_ops_task_clock;
4799 else
4800 pmu = &perf_ops_cpu_clock;
4801 5139
4802 break; 5140int perf_pmu_register(struct pmu *pmu)
4803 case PERF_COUNT_SW_PAGE_FAULTS: 5141{
4804 case PERF_COUNT_SW_PAGE_FAULTS_MIN: 5142 int cpu, ret;
4805 case PERF_COUNT_SW_PAGE_FAULTS_MAJ: 5143
4806 case PERF_COUNT_SW_CONTEXT_SWITCHES: 5144 mutex_lock(&pmus_lock);
4807 case PERF_COUNT_SW_CPU_MIGRATIONS: 5145 ret = -ENOMEM;
4808 case PERF_COUNT_SW_ALIGNMENT_FAULTS: 5146 pmu->pmu_disable_count = alloc_percpu(int);
4809 case PERF_COUNT_SW_EMULATION_FAULTS: 5147 if (!pmu->pmu_disable_count)
4810 if (!event->parent) { 5148 goto unlock;
4811 int err; 5149
4812 5150 pmu->pmu_cpu_context = find_pmu_context(pmu->task_ctx_nr);
4813 err = swevent_hlist_get(event); 5151 if (pmu->pmu_cpu_context)
4814 if (err) 5152 goto got_cpu_context;
4815 return ERR_PTR(err); 5153
5154 pmu->pmu_cpu_context = alloc_percpu(struct perf_cpu_context);
5155 if (!pmu->pmu_cpu_context)
5156 goto free_pdc;
5157
5158 for_each_possible_cpu(cpu) {
5159 struct perf_cpu_context *cpuctx;
4816 5160
4817 atomic_inc(&perf_swevent_enabled[event_id]); 5161 cpuctx = per_cpu_ptr(pmu->pmu_cpu_context, cpu);
4818 event->destroy = sw_perf_event_destroy; 5162 __perf_event_init_context(&cpuctx->ctx);
5163 cpuctx->ctx.type = cpu_context;
5164 cpuctx->ctx.pmu = pmu;
5165 cpuctx->jiffies_interval = 1;
5166 INIT_LIST_HEAD(&cpuctx->rotation_list);
5167 }
5168
5169got_cpu_context:
5170 if (!pmu->start_txn) {
5171 if (pmu->pmu_enable) {
5172 /*
5173 * If we have pmu_enable/pmu_disable calls, install
5174 * transaction stubs that use that to try and batch
5175 * hardware accesses.
5176 */
5177 pmu->start_txn = perf_pmu_start_txn;
5178 pmu->commit_txn = perf_pmu_commit_txn;
5179 pmu->cancel_txn = perf_pmu_cancel_txn;
5180 } else {
5181 pmu->start_txn = perf_pmu_nop_void;
5182 pmu->commit_txn = perf_pmu_nop_int;
5183 pmu->cancel_txn = perf_pmu_nop_void;
5184 }
5185 }
5186
5187 if (!pmu->pmu_enable) {
5188 pmu->pmu_enable = perf_pmu_nop_void;
5189 pmu->pmu_disable = perf_pmu_nop_void;
5190 }
5191
5192 list_add_rcu(&pmu->entry, &pmus);
5193 ret = 0;
5194unlock:
5195 mutex_unlock(&pmus_lock);
5196
5197 return ret;
5198
5199free_pdc:
5200 free_percpu(pmu->pmu_disable_count);
5201 goto unlock;
5202}
5203
5204void perf_pmu_unregister(struct pmu *pmu)
5205{
5206 mutex_lock(&pmus_lock);
5207 list_del_rcu(&pmu->entry);
5208 mutex_unlock(&pmus_lock);
5209
5210 /*
5211 * We dereference the pmu list under both SRCU and regular RCU, so
5212 * synchronize against both of those.
5213 */
5214 synchronize_srcu(&pmus_srcu);
5215 synchronize_rcu();
5216
5217 free_percpu(pmu->pmu_disable_count);
5218 free_pmu_context(pmu->pmu_cpu_context);
5219}
5220
5221struct pmu *perf_init_event(struct perf_event *event)
5222{
5223 struct pmu *pmu = NULL;
5224 int idx;
5225
5226 idx = srcu_read_lock(&pmus_srcu);
5227 list_for_each_entry_rcu(pmu, &pmus, entry) {
5228 int ret = pmu->event_init(event);
5229 if (!ret)
5230 goto unlock;
5231
5232 if (ret != -ENOENT) {
5233 pmu = ERR_PTR(ret);
5234 goto unlock;
4819 } 5235 }
4820 pmu = &perf_ops_generic;
4821 break;
4822 } 5236 }
5237 pmu = ERR_PTR(-ENOENT);
5238unlock:
5239 srcu_read_unlock(&pmus_srcu, idx);
4823 5240
4824 return pmu; 5241 return pmu;
4825} 5242}
@@ -4828,20 +5245,18 @@ static const struct pmu *sw_perf_event_init(struct perf_event *event)
4828 * Allocate and initialize a event structure 5245 * Allocate and initialize a event structure
4829 */ 5246 */
4830static struct perf_event * 5247static struct perf_event *
4831perf_event_alloc(struct perf_event_attr *attr, 5248perf_event_alloc(struct perf_event_attr *attr, int cpu,
4832 int cpu, 5249 struct task_struct *task,
4833 struct perf_event_context *ctx, 5250 struct perf_event *group_leader,
4834 struct perf_event *group_leader, 5251 struct perf_event *parent_event,
4835 struct perf_event *parent_event, 5252 perf_overflow_handler_t overflow_handler)
4836 perf_overflow_handler_t overflow_handler, 5253{
4837 gfp_t gfpflags) 5254 struct pmu *pmu;
4838{
4839 const struct pmu *pmu;
4840 struct perf_event *event; 5255 struct perf_event *event;
4841 struct hw_perf_event *hwc; 5256 struct hw_perf_event *hwc;
4842 long err; 5257 long err;
4843 5258
4844 event = kzalloc(sizeof(*event), gfpflags); 5259 event = kzalloc(sizeof(*event), GFP_KERNEL);
4845 if (!event) 5260 if (!event)
4846 return ERR_PTR(-ENOMEM); 5261 return ERR_PTR(-ENOMEM);
4847 5262
@@ -4859,6 +5274,7 @@ perf_event_alloc(struct perf_event_attr *attr,
4859 INIT_LIST_HEAD(&event->event_entry); 5274 INIT_LIST_HEAD(&event->event_entry);
4860 INIT_LIST_HEAD(&event->sibling_list); 5275 INIT_LIST_HEAD(&event->sibling_list);
4861 init_waitqueue_head(&event->waitq); 5276 init_waitqueue_head(&event->waitq);
5277 init_irq_work(&event->pending, perf_pending_event);
4862 5278
4863 mutex_init(&event->mmap_mutex); 5279 mutex_init(&event->mmap_mutex);
4864 5280
@@ -4866,7 +5282,6 @@ perf_event_alloc(struct perf_event_attr *attr,
4866 event->attr = *attr; 5282 event->attr = *attr;
4867 event->group_leader = group_leader; 5283 event->group_leader = group_leader;
4868 event->pmu = NULL; 5284 event->pmu = NULL;
4869 event->ctx = ctx;
4870 event->oncpu = -1; 5285 event->oncpu = -1;
4871 5286
4872 event->parent = parent_event; 5287 event->parent = parent_event;
@@ -4876,6 +5291,17 @@ perf_event_alloc(struct perf_event_attr *attr,
4876 5291
4877 event->state = PERF_EVENT_STATE_INACTIVE; 5292 event->state = PERF_EVENT_STATE_INACTIVE;
4878 5293
5294 if (task) {
5295 event->attach_state = PERF_ATTACH_TASK;
5296#ifdef CONFIG_HAVE_HW_BREAKPOINT
5297 /*
5298 * hw_breakpoint is a bit difficult here..
5299 */
5300 if (attr->type == PERF_TYPE_BREAKPOINT)
5301 event->hw.bp_target = task;
5302#endif
5303 }
5304
4879 if (!overflow_handler && parent_event) 5305 if (!overflow_handler && parent_event)
4880 overflow_handler = parent_event->overflow_handler; 5306 overflow_handler = parent_event->overflow_handler;
4881 5307
@@ -4900,29 +5326,8 @@ perf_event_alloc(struct perf_event_attr *attr,
4900 if (attr->inherit && (attr->read_format & PERF_FORMAT_GROUP)) 5326 if (attr->inherit && (attr->read_format & PERF_FORMAT_GROUP))
4901 goto done; 5327 goto done;
4902 5328
4903 switch (attr->type) { 5329 pmu = perf_init_event(event);
4904 case PERF_TYPE_RAW:
4905 case PERF_TYPE_HARDWARE:
4906 case PERF_TYPE_HW_CACHE:
4907 pmu = hw_perf_event_init(event);
4908 break;
4909 5330
4910 case PERF_TYPE_SOFTWARE:
4911 pmu = sw_perf_event_init(event);
4912 break;
4913
4914 case PERF_TYPE_TRACEPOINT:
4915 pmu = tp_perf_event_init(event);
4916 break;
4917
4918 case PERF_TYPE_BREAKPOINT:
4919 pmu = bp_perf_event_init(event);
4920 break;
4921
4922
4923 default:
4924 break;
4925 }
4926done: 5331done:
4927 err = 0; 5332 err = 0;
4928 if (!pmu) 5333 if (!pmu)
@@ -4940,13 +5345,21 @@ done:
4940 event->pmu = pmu; 5345 event->pmu = pmu;
4941 5346
4942 if (!event->parent) { 5347 if (!event->parent) {
4943 atomic_inc(&nr_events); 5348 if (event->attach_state & PERF_ATTACH_TASK)
5349 jump_label_inc(&perf_task_events);
4944 if (event->attr.mmap || event->attr.mmap_data) 5350 if (event->attr.mmap || event->attr.mmap_data)
4945 atomic_inc(&nr_mmap_events); 5351 atomic_inc(&nr_mmap_events);
4946 if (event->attr.comm) 5352 if (event->attr.comm)
4947 atomic_inc(&nr_comm_events); 5353 atomic_inc(&nr_comm_events);
4948 if (event->attr.task) 5354 if (event->attr.task)
4949 atomic_inc(&nr_task_events); 5355 atomic_inc(&nr_task_events);
5356 if (event->attr.sample_type & PERF_SAMPLE_CALLCHAIN) {
5357 err = get_callchain_buffers();
5358 if (err) {
5359 free_event(event);
5360 return ERR_PTR(err);
5361 }
5362 }
4950 } 5363 }
4951 5364
4952 return event; 5365 return event;
@@ -5094,12 +5507,16 @@ SYSCALL_DEFINE5(perf_event_open,
5094 struct perf_event_attr __user *, attr_uptr, 5507 struct perf_event_attr __user *, attr_uptr,
5095 pid_t, pid, int, cpu, int, group_fd, unsigned long, flags) 5508 pid_t, pid, int, cpu, int, group_fd, unsigned long, flags)
5096{ 5509{
5097 struct perf_event *event, *group_leader = NULL, *output_event = NULL; 5510 struct perf_event *group_leader = NULL, *output_event = NULL;
5511 struct perf_event *event, *sibling;
5098 struct perf_event_attr attr; 5512 struct perf_event_attr attr;
5099 struct perf_event_context *ctx; 5513 struct perf_event_context *ctx;
5100 struct file *event_file = NULL; 5514 struct file *event_file = NULL;
5101 struct file *group_file = NULL; 5515 struct file *group_file = NULL;
5516 struct task_struct *task = NULL;
5517 struct pmu *pmu;
5102 int event_fd; 5518 int event_fd;
5519 int move_group = 0;
5103 int fput_needed = 0; 5520 int fput_needed = 0;
5104 int err; 5521 int err;
5105 5522
@@ -5125,20 +5542,11 @@ SYSCALL_DEFINE5(perf_event_open,
5125 if (event_fd < 0) 5542 if (event_fd < 0)
5126 return event_fd; 5543 return event_fd;
5127 5544
5128 /*
5129 * Get the target context (task or percpu):
5130 */
5131 ctx = find_get_context(pid, cpu);
5132 if (IS_ERR(ctx)) {
5133 err = PTR_ERR(ctx);
5134 goto err_fd;
5135 }
5136
5137 if (group_fd != -1) { 5545 if (group_fd != -1) {
5138 group_leader = perf_fget_light(group_fd, &fput_needed); 5546 group_leader = perf_fget_light(group_fd, &fput_needed);
5139 if (IS_ERR(group_leader)) { 5547 if (IS_ERR(group_leader)) {
5140 err = PTR_ERR(group_leader); 5548 err = PTR_ERR(group_leader);
5141 goto err_put_context; 5549 goto err_fd;
5142 } 5550 }
5143 group_file = group_leader->filp; 5551 group_file = group_leader->filp;
5144 if (flags & PERF_FLAG_FD_OUTPUT) 5552 if (flags & PERF_FLAG_FD_OUTPUT)
@@ -5147,6 +5555,58 @@ SYSCALL_DEFINE5(perf_event_open,
5147 group_leader = NULL; 5555 group_leader = NULL;
5148 } 5556 }
5149 5557
5558 if (pid != -1) {
5559 task = find_lively_task_by_vpid(pid);
5560 if (IS_ERR(task)) {
5561 err = PTR_ERR(task);
5562 goto err_group_fd;
5563 }
5564 }
5565
5566 event = perf_event_alloc(&attr, cpu, task, group_leader, NULL, NULL);
5567 if (IS_ERR(event)) {
5568 err = PTR_ERR(event);
5569 goto err_task;
5570 }
5571
5572 /*
5573 * Special case software events and allow them to be part of
5574 * any hardware group.
5575 */
5576 pmu = event->pmu;
5577
5578 if (group_leader &&
5579 (is_software_event(event) != is_software_event(group_leader))) {
5580 if (is_software_event(event)) {
5581 /*
5582 * If event and group_leader are not both a software
5583 * event, and event is, then group leader is not.
5584 *
5585 * Allow the addition of software events to !software
5586 * groups, this is safe because software events never
5587 * fail to schedule.
5588 */
5589 pmu = group_leader->pmu;
5590 } else if (is_software_event(group_leader) &&
5591 (group_leader->group_flags & PERF_GROUP_SOFTWARE)) {
5592 /*
5593 * In case the group is a pure software group, and we
5594 * try to add a hardware event, move the whole group to
5595 * the hardware context.
5596 */
5597 move_group = 1;
5598 }
5599 }
5600
5601 /*
5602 * Get the target context (task or percpu):
5603 */
5604 ctx = find_get_context(pmu, task, cpu);
5605 if (IS_ERR(ctx)) {
5606 err = PTR_ERR(ctx);
5607 goto err_alloc;
5608 }
5609
5150 /* 5610 /*
5151 * Look up the group leader (we will attach this event to it): 5611 * Look up the group leader (we will attach this event to it):
5152 */ 5612 */
@@ -5158,42 +5618,66 @@ SYSCALL_DEFINE5(perf_event_open,
5158 * becoming part of another group-sibling): 5618 * becoming part of another group-sibling):
5159 */ 5619 */
5160 if (group_leader->group_leader != group_leader) 5620 if (group_leader->group_leader != group_leader)
5161 goto err_put_context; 5621 goto err_context;
5162 /* 5622 /*
5163 * Do not allow to attach to a group in a different 5623 * Do not allow to attach to a group in a different
5164 * task or CPU context: 5624 * task or CPU context:
5165 */ 5625 */
5166 if (group_leader->ctx != ctx) 5626 if (move_group) {
5167 goto err_put_context; 5627 if (group_leader->ctx->type != ctx->type)
5628 goto err_context;
5629 } else {
5630 if (group_leader->ctx != ctx)
5631 goto err_context;
5632 }
5633
5168 /* 5634 /*
5169 * Only a group leader can be exclusive or pinned 5635 * Only a group leader can be exclusive or pinned
5170 */ 5636 */
5171 if (attr.exclusive || attr.pinned) 5637 if (attr.exclusive || attr.pinned)
5172 goto err_put_context; 5638 goto err_context;
5173 }
5174
5175 event = perf_event_alloc(&attr, cpu, ctx, group_leader,
5176 NULL, NULL, GFP_KERNEL);
5177 if (IS_ERR(event)) {
5178 err = PTR_ERR(event);
5179 goto err_put_context;
5180 } 5639 }
5181 5640
5182 if (output_event) { 5641 if (output_event) {
5183 err = perf_event_set_output(event, output_event); 5642 err = perf_event_set_output(event, output_event);
5184 if (err) 5643 if (err)
5185 goto err_free_put_context; 5644 goto err_context;
5186 } 5645 }
5187 5646
5188 event_file = anon_inode_getfile("[perf_event]", &perf_fops, event, O_RDWR); 5647 event_file = anon_inode_getfile("[perf_event]", &perf_fops, event, O_RDWR);
5189 if (IS_ERR(event_file)) { 5648 if (IS_ERR(event_file)) {
5190 err = PTR_ERR(event_file); 5649 err = PTR_ERR(event_file);
5191 goto err_free_put_context; 5650 goto err_context;
5651 }
5652
5653 if (move_group) {
5654 struct perf_event_context *gctx = group_leader->ctx;
5655
5656 mutex_lock(&gctx->mutex);
5657 perf_event_remove_from_context(group_leader);
5658 list_for_each_entry(sibling, &group_leader->sibling_list,
5659 group_entry) {
5660 perf_event_remove_from_context(sibling);
5661 put_ctx(gctx);
5662 }
5663 mutex_unlock(&gctx->mutex);
5664 put_ctx(gctx);
5192 } 5665 }
5193 5666
5194 event->filp = event_file; 5667 event->filp = event_file;
5195 WARN_ON_ONCE(ctx->parent_ctx); 5668 WARN_ON_ONCE(ctx->parent_ctx);
5196 mutex_lock(&ctx->mutex); 5669 mutex_lock(&ctx->mutex);
5670
5671 if (move_group) {
5672 perf_install_in_context(ctx, group_leader, cpu);
5673 get_ctx(ctx);
5674 list_for_each_entry(sibling, &group_leader->sibling_list,
5675 group_entry) {
5676 perf_install_in_context(ctx, sibling, cpu);
5677 get_ctx(ctx);
5678 }
5679 }
5680
5197 perf_install_in_context(ctx, event, cpu); 5681 perf_install_in_context(ctx, event, cpu);
5198 ++ctx->generation; 5682 ++ctx->generation;
5199 mutex_unlock(&ctx->mutex); 5683 mutex_unlock(&ctx->mutex);
@@ -5214,11 +5698,15 @@ SYSCALL_DEFINE5(perf_event_open,
5214 fd_install(event_fd, event_file); 5698 fd_install(event_fd, event_file);
5215 return event_fd; 5699 return event_fd;
5216 5700
5217err_free_put_context: 5701err_context:
5702 put_ctx(ctx);
5703err_alloc:
5218 free_event(event); 5704 free_event(event);
5219err_put_context: 5705err_task:
5706 if (task)
5707 put_task_struct(task);
5708err_group_fd:
5220 fput_light(group_file, fput_needed); 5709 fput_light(group_file, fput_needed);
5221 put_ctx(ctx);
5222err_fd: 5710err_fd:
5223 put_unused_fd(event_fd); 5711 put_unused_fd(event_fd);
5224 return err; 5712 return err;
@@ -5229,32 +5717,31 @@ err_fd:
5229 * 5717 *
5230 * @attr: attributes of the counter to create 5718 * @attr: attributes of the counter to create
5231 * @cpu: cpu in which the counter is bound 5719 * @cpu: cpu in which the counter is bound
5232 * @pid: task to profile 5720 * @task: task to profile (NULL for percpu)
5233 */ 5721 */
5234struct perf_event * 5722struct perf_event *
5235perf_event_create_kernel_counter(struct perf_event_attr *attr, int cpu, 5723perf_event_create_kernel_counter(struct perf_event_attr *attr, int cpu,
5236 pid_t pid, 5724 struct task_struct *task,
5237 perf_overflow_handler_t overflow_handler) 5725 perf_overflow_handler_t overflow_handler)
5238{ 5726{
5239 struct perf_event *event;
5240 struct perf_event_context *ctx; 5727 struct perf_event_context *ctx;
5728 struct perf_event *event;
5241 int err; 5729 int err;
5242 5730
5243 /* 5731 /*
5244 * Get the target context (task or percpu): 5732 * Get the target context (task or percpu):
5245 */ 5733 */
5246 5734
5247 ctx = find_get_context(pid, cpu); 5735 event = perf_event_alloc(attr, cpu, task, NULL, NULL, overflow_handler);
5248 if (IS_ERR(ctx)) {
5249 err = PTR_ERR(ctx);
5250 goto err_exit;
5251 }
5252
5253 event = perf_event_alloc(attr, cpu, ctx, NULL,
5254 NULL, overflow_handler, GFP_KERNEL);
5255 if (IS_ERR(event)) { 5736 if (IS_ERR(event)) {
5256 err = PTR_ERR(event); 5737 err = PTR_ERR(event);
5257 goto err_put_context; 5738 goto err;
5739 }
5740
5741 ctx = find_get_context(event->pmu, task, cpu);
5742 if (IS_ERR(ctx)) {
5743 err = PTR_ERR(ctx);
5744 goto err_free;
5258 } 5745 }
5259 5746
5260 event->filp = NULL; 5747 event->filp = NULL;
@@ -5272,112 +5759,13 @@ perf_event_create_kernel_counter(struct perf_event_attr *attr, int cpu,
5272 5759
5273 return event; 5760 return event;
5274 5761
5275 err_put_context: 5762err_free:
5276 put_ctx(ctx); 5763 free_event(event);
5277 err_exit: 5764err:
5278 return ERR_PTR(err); 5765 return ERR_PTR(err);
5279} 5766}
5280EXPORT_SYMBOL_GPL(perf_event_create_kernel_counter); 5767EXPORT_SYMBOL_GPL(perf_event_create_kernel_counter);
5281 5768
5282/*
5283 * inherit a event from parent task to child task:
5284 */
5285static struct perf_event *
5286inherit_event(struct perf_event *parent_event,
5287 struct task_struct *parent,
5288 struct perf_event_context *parent_ctx,
5289 struct task_struct *child,
5290 struct perf_event *group_leader,
5291 struct perf_event_context *child_ctx)
5292{
5293 struct perf_event *child_event;
5294
5295 /*
5296 * Instead of creating recursive hierarchies of events,
5297 * we link inherited events back to the original parent,
5298 * which has a filp for sure, which we use as the reference
5299 * count:
5300 */
5301 if (parent_event->parent)
5302 parent_event = parent_event->parent;
5303
5304 child_event = perf_event_alloc(&parent_event->attr,
5305 parent_event->cpu, child_ctx,
5306 group_leader, parent_event,
5307 NULL, GFP_KERNEL);
5308 if (IS_ERR(child_event))
5309 return child_event;
5310 get_ctx(child_ctx);
5311
5312 /*
5313 * Make the child state follow the state of the parent event,
5314 * not its attr.disabled bit. We hold the parent's mutex,
5315 * so we won't race with perf_event_{en, dis}able_family.
5316 */
5317 if (parent_event->state >= PERF_EVENT_STATE_INACTIVE)
5318 child_event->state = PERF_EVENT_STATE_INACTIVE;
5319 else
5320 child_event->state = PERF_EVENT_STATE_OFF;
5321
5322 if (parent_event->attr.freq) {
5323 u64 sample_period = parent_event->hw.sample_period;
5324 struct hw_perf_event *hwc = &child_event->hw;
5325
5326 hwc->sample_period = sample_period;
5327 hwc->last_period = sample_period;
5328
5329 local64_set(&hwc->period_left, sample_period);
5330 }
5331
5332 child_event->overflow_handler = parent_event->overflow_handler;
5333
5334 /*
5335 * Link it up in the child's context:
5336 */
5337 add_event_to_ctx(child_event, child_ctx);
5338
5339 /*
5340 * Get a reference to the parent filp - we will fput it
5341 * when the child event exits. This is safe to do because
5342 * we are in the parent and we know that the filp still
5343 * exists and has a nonzero count:
5344 */
5345 atomic_long_inc(&parent_event->filp->f_count);
5346
5347 /*
5348 * Link this into the parent event's child list
5349 */
5350 WARN_ON_ONCE(parent_event->ctx->parent_ctx);
5351 mutex_lock(&parent_event->child_mutex);
5352 list_add_tail(&child_event->child_list, &parent_event->child_list);
5353 mutex_unlock(&parent_event->child_mutex);
5354
5355 return child_event;
5356}
5357
5358static int inherit_group(struct perf_event *parent_event,
5359 struct task_struct *parent,
5360 struct perf_event_context *parent_ctx,
5361 struct task_struct *child,
5362 struct perf_event_context *child_ctx)
5363{
5364 struct perf_event *leader;
5365 struct perf_event *sub;
5366 struct perf_event *child_ctr;
5367
5368 leader = inherit_event(parent_event, parent, parent_ctx,
5369 child, NULL, child_ctx);
5370 if (IS_ERR(leader))
5371 return PTR_ERR(leader);
5372 list_for_each_entry(sub, &parent_event->sibling_list, group_entry) {
5373 child_ctr = inherit_event(sub, parent, parent_ctx,
5374 child, leader, child_ctx);
5375 if (IS_ERR(child_ctr))
5376 return PTR_ERR(child_ctr);
5377 }
5378 return 0;
5379}
5380
5381static void sync_child_event(struct perf_event *child_event, 5769static void sync_child_event(struct perf_event *child_event,
5382 struct task_struct *child) 5770 struct task_struct *child)
5383{ 5771{
@@ -5434,16 +5822,13 @@ __perf_event_exit_task(struct perf_event *child_event,
5434 } 5822 }
5435} 5823}
5436 5824
5437/* 5825static void perf_event_exit_task_context(struct task_struct *child, int ctxn)
5438 * When a child task exits, feed back event values to parent events.
5439 */
5440void perf_event_exit_task(struct task_struct *child)
5441{ 5826{
5442 struct perf_event *child_event, *tmp; 5827 struct perf_event *child_event, *tmp;
5443 struct perf_event_context *child_ctx; 5828 struct perf_event_context *child_ctx;
5444 unsigned long flags; 5829 unsigned long flags;
5445 5830
5446 if (likely(!child->perf_event_ctxp)) { 5831 if (likely(!child->perf_event_ctxp[ctxn])) {
5447 perf_event_task(child, NULL, 0); 5832 perf_event_task(child, NULL, 0);
5448 return; 5833 return;
5449 } 5834 }
@@ -5455,8 +5840,8 @@ void perf_event_exit_task(struct task_struct *child)
5455 * scheduled, so we are now safe from rescheduling changing 5840 * scheduled, so we are now safe from rescheduling changing
5456 * our context. 5841 * our context.
5457 */ 5842 */
5458 child_ctx = child->perf_event_ctxp; 5843 child_ctx = child->perf_event_ctxp[ctxn];
5459 __perf_event_task_sched_out(child_ctx); 5844 task_ctx_sched_out(child_ctx, EVENT_ALL);
5460 5845
5461 /* 5846 /*
5462 * Take the context lock here so that if find_get_context is 5847 * Take the context lock here so that if find_get_context is
@@ -5464,7 +5849,7 @@ void perf_event_exit_task(struct task_struct *child)
5464 * incremented the context's refcount before we do put_ctx below. 5849 * incremented the context's refcount before we do put_ctx below.
5465 */ 5850 */
5466 raw_spin_lock(&child_ctx->lock); 5851 raw_spin_lock(&child_ctx->lock);
5467 child->perf_event_ctxp = NULL; 5852 child->perf_event_ctxp[ctxn] = NULL;
5468 /* 5853 /*
5469 * If this context is a clone; unclone it so it can't get 5854 * If this context is a clone; unclone it so it can't get
5470 * swapped to another process while we're removing all 5855 * swapped to another process while we're removing all
@@ -5517,6 +5902,17 @@ again:
5517 put_ctx(child_ctx); 5902 put_ctx(child_ctx);
5518} 5903}
5519 5904
5905/*
5906 * When a child task exits, feed back event values to parent events.
5907 */
5908void perf_event_exit_task(struct task_struct *child)
5909{
5910 int ctxn;
5911
5912 for_each_task_context_nr(ctxn)
5913 perf_event_exit_task_context(child, ctxn);
5914}
5915
5520static void perf_free_event(struct perf_event *event, 5916static void perf_free_event(struct perf_event *event,
5521 struct perf_event_context *ctx) 5917 struct perf_event_context *ctx)
5522{ 5918{
@@ -5538,48 +5934,166 @@ static void perf_free_event(struct perf_event *event,
5538 5934
5539/* 5935/*
5540 * free an unexposed, unused context as created by inheritance by 5936 * free an unexposed, unused context as created by inheritance by
5541 * init_task below, used by fork() in case of fail. 5937 * perf_event_init_task below, used by fork() in case of fail.
5542 */ 5938 */
5543void perf_event_free_task(struct task_struct *task) 5939void perf_event_free_task(struct task_struct *task)
5544{ 5940{
5545 struct perf_event_context *ctx = task->perf_event_ctxp; 5941 struct perf_event_context *ctx;
5546 struct perf_event *event, *tmp; 5942 struct perf_event *event, *tmp;
5943 int ctxn;
5547 5944
5548 if (!ctx) 5945 for_each_task_context_nr(ctxn) {
5549 return; 5946 ctx = task->perf_event_ctxp[ctxn];
5947 if (!ctx)
5948 continue;
5550 5949
5551 mutex_lock(&ctx->mutex); 5950 mutex_lock(&ctx->mutex);
5552again: 5951again:
5553 list_for_each_entry_safe(event, tmp, &ctx->pinned_groups, group_entry) 5952 list_for_each_entry_safe(event, tmp, &ctx->pinned_groups,
5554 perf_free_event(event, ctx); 5953 group_entry)
5954 perf_free_event(event, ctx);
5555 5955
5556 list_for_each_entry_safe(event, tmp, &ctx->flexible_groups, 5956 list_for_each_entry_safe(event, tmp, &ctx->flexible_groups,
5557 group_entry) 5957 group_entry)
5558 perf_free_event(event, ctx); 5958 perf_free_event(event, ctx);
5559 5959
5560 if (!list_empty(&ctx->pinned_groups) || 5960 if (!list_empty(&ctx->pinned_groups) ||
5561 !list_empty(&ctx->flexible_groups)) 5961 !list_empty(&ctx->flexible_groups))
5562 goto again; 5962 goto again;
5563 5963
5564 mutex_unlock(&ctx->mutex); 5964 mutex_unlock(&ctx->mutex);
5565 5965
5566 put_ctx(ctx); 5966 put_ctx(ctx);
5967 }
5968}
5969
5970void perf_event_delayed_put(struct task_struct *task)
5971{
5972 int ctxn;
5973
5974 for_each_task_context_nr(ctxn)
5975 WARN_ON_ONCE(task->perf_event_ctxp[ctxn]);
5976}
5977
5978/*
5979 * inherit a event from parent task to child task:
5980 */
5981static struct perf_event *
5982inherit_event(struct perf_event *parent_event,
5983 struct task_struct *parent,
5984 struct perf_event_context *parent_ctx,
5985 struct task_struct *child,
5986 struct perf_event *group_leader,
5987 struct perf_event_context *child_ctx)
5988{
5989 struct perf_event *child_event;
5990 unsigned long flags;
5991
5992 /*
5993 * Instead of creating recursive hierarchies of events,
5994 * we link inherited events back to the original parent,
5995 * which has a filp for sure, which we use as the reference
5996 * count:
5997 */
5998 if (parent_event->parent)
5999 parent_event = parent_event->parent;
6000
6001 child_event = perf_event_alloc(&parent_event->attr,
6002 parent_event->cpu,
6003 child,
6004 group_leader, parent_event,
6005 NULL);
6006 if (IS_ERR(child_event))
6007 return child_event;
6008 get_ctx(child_ctx);
6009
6010 /*
6011 * Make the child state follow the state of the parent event,
6012 * not its attr.disabled bit. We hold the parent's mutex,
6013 * so we won't race with perf_event_{en, dis}able_family.
6014 */
6015 if (parent_event->state >= PERF_EVENT_STATE_INACTIVE)
6016 child_event->state = PERF_EVENT_STATE_INACTIVE;
6017 else
6018 child_event->state = PERF_EVENT_STATE_OFF;
6019
6020 if (parent_event->attr.freq) {
6021 u64 sample_period = parent_event->hw.sample_period;
6022 struct hw_perf_event *hwc = &child_event->hw;
6023
6024 hwc->sample_period = sample_period;
6025 hwc->last_period = sample_period;
6026
6027 local64_set(&hwc->period_left, sample_period);
6028 }
6029
6030 child_event->ctx = child_ctx;
6031 child_event->overflow_handler = parent_event->overflow_handler;
6032
6033 /*
6034 * Link it up in the child's context:
6035 */
6036 raw_spin_lock_irqsave(&child_ctx->lock, flags);
6037 add_event_to_ctx(child_event, child_ctx);
6038 raw_spin_unlock_irqrestore(&child_ctx->lock, flags);
6039
6040 /*
6041 * Get a reference to the parent filp - we will fput it
6042 * when the child event exits. This is safe to do because
6043 * we are in the parent and we know that the filp still
6044 * exists and has a nonzero count:
6045 */
6046 atomic_long_inc(&parent_event->filp->f_count);
6047
6048 /*
6049 * Link this into the parent event's child list
6050 */
6051 WARN_ON_ONCE(parent_event->ctx->parent_ctx);
6052 mutex_lock(&parent_event->child_mutex);
6053 list_add_tail(&child_event->child_list, &parent_event->child_list);
6054 mutex_unlock(&parent_event->child_mutex);
6055
6056 return child_event;
6057}
6058
6059static int inherit_group(struct perf_event *parent_event,
6060 struct task_struct *parent,
6061 struct perf_event_context *parent_ctx,
6062 struct task_struct *child,
6063 struct perf_event_context *child_ctx)
6064{
6065 struct perf_event *leader;
6066 struct perf_event *sub;
6067 struct perf_event *child_ctr;
6068
6069 leader = inherit_event(parent_event, parent, parent_ctx,
6070 child, NULL, child_ctx);
6071 if (IS_ERR(leader))
6072 return PTR_ERR(leader);
6073 list_for_each_entry(sub, &parent_event->sibling_list, group_entry) {
6074 child_ctr = inherit_event(sub, parent, parent_ctx,
6075 child, leader, child_ctx);
6076 if (IS_ERR(child_ctr))
6077 return PTR_ERR(child_ctr);
6078 }
6079 return 0;
5567} 6080}
5568 6081
5569static int 6082static int
5570inherit_task_group(struct perf_event *event, struct task_struct *parent, 6083inherit_task_group(struct perf_event *event, struct task_struct *parent,
5571 struct perf_event_context *parent_ctx, 6084 struct perf_event_context *parent_ctx,
5572 struct task_struct *child, 6085 struct task_struct *child, int ctxn,
5573 int *inherited_all) 6086 int *inherited_all)
5574{ 6087{
5575 int ret; 6088 int ret;
5576 struct perf_event_context *child_ctx = child->perf_event_ctxp; 6089 struct perf_event_context *child_ctx;
5577 6090
5578 if (!event->attr.inherit) { 6091 if (!event->attr.inherit) {
5579 *inherited_all = 0; 6092 *inherited_all = 0;
5580 return 0; 6093 return 0;
5581 } 6094 }
5582 6095
6096 child_ctx = child->perf_event_ctxp[ctxn];
5583 if (!child_ctx) { 6097 if (!child_ctx) {
5584 /* 6098 /*
5585 * This is executed from the parent task context, so 6099 * This is executed from the parent task context, so
@@ -5588,14 +6102,11 @@ inherit_task_group(struct perf_event *event, struct task_struct *parent,
5588 * child. 6102 * child.
5589 */ 6103 */
5590 6104
5591 child_ctx = kzalloc(sizeof(struct perf_event_context), 6105 child_ctx = alloc_perf_context(event->pmu, child);
5592 GFP_KERNEL);
5593 if (!child_ctx) 6106 if (!child_ctx)
5594 return -ENOMEM; 6107 return -ENOMEM;
5595 6108
5596 __perf_event_init_context(child_ctx, child); 6109 child->perf_event_ctxp[ctxn] = child_ctx;
5597 child->perf_event_ctxp = child_ctx;
5598 get_task_struct(child);
5599 } 6110 }
5600 6111
5601 ret = inherit_group(event, parent, parent_ctx, 6112 ret = inherit_group(event, parent, parent_ctx,
@@ -5607,11 +6118,10 @@ inherit_task_group(struct perf_event *event, struct task_struct *parent,
5607 return ret; 6118 return ret;
5608} 6119}
5609 6120
5610
5611/* 6121/*
5612 * Initialize the perf_event context in task_struct 6122 * Initialize the perf_event context in task_struct
5613 */ 6123 */
5614int perf_event_init_task(struct task_struct *child) 6124int perf_event_init_context(struct task_struct *child, int ctxn)
5615{ 6125{
5616 struct perf_event_context *child_ctx, *parent_ctx; 6126 struct perf_event_context *child_ctx, *parent_ctx;
5617 struct perf_event_context *cloned_ctx; 6127 struct perf_event_context *cloned_ctx;
@@ -5620,19 +6130,19 @@ int perf_event_init_task(struct task_struct *child)
5620 int inherited_all = 1; 6130 int inherited_all = 1;
5621 int ret = 0; 6131 int ret = 0;
5622 6132
5623 child->perf_event_ctxp = NULL; 6133 child->perf_event_ctxp[ctxn] = NULL;
5624 6134
5625 mutex_init(&child->perf_event_mutex); 6135 mutex_init(&child->perf_event_mutex);
5626 INIT_LIST_HEAD(&child->perf_event_list); 6136 INIT_LIST_HEAD(&child->perf_event_list);
5627 6137
5628 if (likely(!parent->perf_event_ctxp)) 6138 if (likely(!parent->perf_event_ctxp[ctxn]))
5629 return 0; 6139 return 0;
5630 6140
5631 /* 6141 /*
5632 * If the parent's context is a clone, pin it so it won't get 6142 * If the parent's context is a clone, pin it so it won't get
5633 * swapped under us. 6143 * swapped under us.
5634 */ 6144 */
5635 parent_ctx = perf_pin_task_context(parent); 6145 parent_ctx = perf_pin_task_context(parent, ctxn);
5636 6146
5637 /* 6147 /*
5638 * No need to check if parent_ctx != NULL here; since we saw 6148 * No need to check if parent_ctx != NULL here; since we saw
@@ -5652,20 +6162,20 @@ int perf_event_init_task(struct task_struct *child)
5652 * the list, not manipulating it: 6162 * the list, not manipulating it:
5653 */ 6163 */
5654 list_for_each_entry(event, &parent_ctx->pinned_groups, group_entry) { 6164 list_for_each_entry(event, &parent_ctx->pinned_groups, group_entry) {
5655 ret = inherit_task_group(event, parent, parent_ctx, child, 6165 ret = inherit_task_group(event, parent, parent_ctx,
5656 &inherited_all); 6166 child, ctxn, &inherited_all);
5657 if (ret) 6167 if (ret)
5658 break; 6168 break;
5659 } 6169 }
5660 6170
5661 list_for_each_entry(event, &parent_ctx->flexible_groups, group_entry) { 6171 list_for_each_entry(event, &parent_ctx->flexible_groups, group_entry) {
5662 ret = inherit_task_group(event, parent, parent_ctx, child, 6172 ret = inherit_task_group(event, parent, parent_ctx,
5663 &inherited_all); 6173 child, ctxn, &inherited_all);
5664 if (ret) 6174 if (ret)
5665 break; 6175 break;
5666 } 6176 }
5667 6177
5668 child_ctx = child->perf_event_ctxp; 6178 child_ctx = child->perf_event_ctxp[ctxn];
5669 6179
5670 if (child_ctx && inherited_all) { 6180 if (child_ctx && inherited_all) {
5671 /* 6181 /*
@@ -5694,63 +6204,98 @@ int perf_event_init_task(struct task_struct *child)
5694 return ret; 6204 return ret;
5695} 6205}
5696 6206
6207/*
6208 * Initialize the perf_event context in task_struct
6209 */
6210int perf_event_init_task(struct task_struct *child)
6211{
6212 int ctxn, ret;
6213
6214 for_each_task_context_nr(ctxn) {
6215 ret = perf_event_init_context(child, ctxn);
6216 if (ret)
6217 return ret;
6218 }
6219
6220 return 0;
6221}
6222
5697static void __init perf_event_init_all_cpus(void) 6223static void __init perf_event_init_all_cpus(void)
5698{ 6224{
6225 struct swevent_htable *swhash;
5699 int cpu; 6226 int cpu;
5700 struct perf_cpu_context *cpuctx;
5701 6227
5702 for_each_possible_cpu(cpu) { 6228 for_each_possible_cpu(cpu) {
5703 cpuctx = &per_cpu(perf_cpu_context, cpu); 6229 swhash = &per_cpu(swevent_htable, cpu);
5704 mutex_init(&cpuctx->hlist_mutex); 6230 mutex_init(&swhash->hlist_mutex);
5705 __perf_event_init_context(&cpuctx->ctx, NULL); 6231 INIT_LIST_HEAD(&per_cpu(rotation_list, cpu));
5706 } 6232 }
5707} 6233}
5708 6234
5709static void __cpuinit perf_event_init_cpu(int cpu) 6235static void __cpuinit perf_event_init_cpu(int cpu)
5710{ 6236{
5711 struct perf_cpu_context *cpuctx; 6237 struct swevent_htable *swhash = &per_cpu(swevent_htable, cpu);
5712
5713 cpuctx = &per_cpu(perf_cpu_context, cpu);
5714 6238
5715 spin_lock(&perf_resource_lock); 6239 mutex_lock(&swhash->hlist_mutex);
5716 cpuctx->max_pertask = perf_max_events - perf_reserved_percpu; 6240 if (swhash->hlist_refcount > 0) {
5717 spin_unlock(&perf_resource_lock);
5718
5719 mutex_lock(&cpuctx->hlist_mutex);
5720 if (cpuctx->hlist_refcount > 0) {
5721 struct swevent_hlist *hlist; 6241 struct swevent_hlist *hlist;
5722 6242
5723 hlist = kzalloc(sizeof(*hlist), GFP_KERNEL); 6243 hlist = kzalloc_node(sizeof(*hlist), GFP_KERNEL, cpu_to_node(cpu));
5724 WARN_ON_ONCE(!hlist); 6244 WARN_ON(!hlist);
5725 rcu_assign_pointer(cpuctx->swevent_hlist, hlist); 6245 rcu_assign_pointer(swhash->swevent_hlist, hlist);
5726 } 6246 }
5727 mutex_unlock(&cpuctx->hlist_mutex); 6247 mutex_unlock(&swhash->hlist_mutex);
5728} 6248}
5729 6249
5730#ifdef CONFIG_HOTPLUG_CPU 6250#ifdef CONFIG_HOTPLUG_CPU
5731static void __perf_event_exit_cpu(void *info) 6251static void perf_pmu_rotate_stop(struct pmu *pmu)
5732{ 6252{
5733 struct perf_cpu_context *cpuctx = &__get_cpu_var(perf_cpu_context); 6253 struct perf_cpu_context *cpuctx = this_cpu_ptr(pmu->pmu_cpu_context);
5734 struct perf_event_context *ctx = &cpuctx->ctx; 6254
6255 WARN_ON(!irqs_disabled());
6256
6257 list_del_init(&cpuctx->rotation_list);
6258}
6259
6260static void __perf_event_exit_context(void *__info)
6261{
6262 struct perf_event_context *ctx = __info;
5735 struct perf_event *event, *tmp; 6263 struct perf_event *event, *tmp;
5736 6264
6265 perf_pmu_rotate_stop(ctx->pmu);
6266
5737 list_for_each_entry_safe(event, tmp, &ctx->pinned_groups, group_entry) 6267 list_for_each_entry_safe(event, tmp, &ctx->pinned_groups, group_entry)
5738 __perf_event_remove_from_context(event); 6268 __perf_event_remove_from_context(event);
5739 list_for_each_entry_safe(event, tmp, &ctx->flexible_groups, group_entry) 6269 list_for_each_entry_safe(event, tmp, &ctx->flexible_groups, group_entry)
5740 __perf_event_remove_from_context(event); 6270 __perf_event_remove_from_context(event);
5741} 6271}
6272
6273static void perf_event_exit_cpu_context(int cpu)
6274{
6275 struct perf_event_context *ctx;
6276 struct pmu *pmu;
6277 int idx;
6278
6279 idx = srcu_read_lock(&pmus_srcu);
6280 list_for_each_entry_rcu(pmu, &pmus, entry) {
6281 ctx = &per_cpu_ptr(pmu->pmu_cpu_context, cpu)->ctx;
6282
6283 mutex_lock(&ctx->mutex);
6284 smp_call_function_single(cpu, __perf_event_exit_context, ctx, 1);
6285 mutex_unlock(&ctx->mutex);
6286 }
6287 srcu_read_unlock(&pmus_srcu, idx);
6288}
6289
5742static void perf_event_exit_cpu(int cpu) 6290static void perf_event_exit_cpu(int cpu)
5743{ 6291{
5744 struct perf_cpu_context *cpuctx = &per_cpu(perf_cpu_context, cpu); 6292 struct swevent_htable *swhash = &per_cpu(swevent_htable, cpu);
5745 struct perf_event_context *ctx = &cpuctx->ctx;
5746 6293
5747 mutex_lock(&cpuctx->hlist_mutex); 6294 mutex_lock(&swhash->hlist_mutex);
5748 swevent_hlist_release(cpuctx); 6295 swevent_hlist_release(swhash);
5749 mutex_unlock(&cpuctx->hlist_mutex); 6296 mutex_unlock(&swhash->hlist_mutex);
5750 6297
5751 mutex_lock(&ctx->mutex); 6298 perf_event_exit_cpu_context(cpu);
5752 smp_call_function_single(cpu, __perf_event_exit_cpu, NULL, 1);
5753 mutex_unlock(&ctx->mutex);
5754} 6299}
5755#else 6300#else
5756static inline void perf_event_exit_cpu(int cpu) { } 6301static inline void perf_event_exit_cpu(int cpu) { }
@@ -5780,118 +6325,13 @@ perf_cpu_notify(struct notifier_block *self, unsigned long action, void *hcpu)
5780 return NOTIFY_OK; 6325 return NOTIFY_OK;
5781} 6326}
5782 6327
5783/*
5784 * This has to have a higher priority than migration_notifier in sched.c.
5785 */
5786static struct notifier_block __cpuinitdata perf_cpu_nb = {
5787 .notifier_call = perf_cpu_notify,
5788 .priority = 20,
5789};
5790
5791void __init perf_event_init(void) 6328void __init perf_event_init(void)
5792{ 6329{
5793 perf_event_init_all_cpus(); 6330 perf_event_init_all_cpus();
5794 perf_cpu_notify(&perf_cpu_nb, (unsigned long)CPU_UP_PREPARE, 6331 init_srcu_struct(&pmus_srcu);
5795 (void *)(long)smp_processor_id()); 6332 perf_pmu_register(&perf_swevent);
5796 perf_cpu_notify(&perf_cpu_nb, (unsigned long)CPU_ONLINE, 6333 perf_pmu_register(&perf_cpu_clock);
5797 (void *)(long)smp_processor_id()); 6334 perf_pmu_register(&perf_task_clock);
5798 register_cpu_notifier(&perf_cpu_nb); 6335 perf_tp_register();
5799} 6336 perf_cpu_notifier(perf_cpu_notify);
5800
5801static ssize_t perf_show_reserve_percpu(struct sysdev_class *class,
5802 struct sysdev_class_attribute *attr,
5803 char *buf)
5804{
5805 return sprintf(buf, "%d\n", perf_reserved_percpu);
5806}
5807
5808static ssize_t
5809perf_set_reserve_percpu(struct sysdev_class *class,
5810 struct sysdev_class_attribute *attr,
5811 const char *buf,
5812 size_t count)
5813{
5814 struct perf_cpu_context *cpuctx;
5815 unsigned long val;
5816 int err, cpu, mpt;
5817
5818 err = strict_strtoul(buf, 10, &val);
5819 if (err)
5820 return err;
5821 if (val > perf_max_events)
5822 return -EINVAL;
5823
5824 spin_lock(&perf_resource_lock);
5825 perf_reserved_percpu = val;
5826 for_each_online_cpu(cpu) {
5827 cpuctx = &per_cpu(perf_cpu_context, cpu);
5828 raw_spin_lock_irq(&cpuctx->ctx.lock);
5829 mpt = min(perf_max_events - cpuctx->ctx.nr_events,
5830 perf_max_events - perf_reserved_percpu);
5831 cpuctx->max_pertask = mpt;
5832 raw_spin_unlock_irq(&cpuctx->ctx.lock);
5833 }
5834 spin_unlock(&perf_resource_lock);
5835
5836 return count;
5837}
5838
5839static ssize_t perf_show_overcommit(struct sysdev_class *class,
5840 struct sysdev_class_attribute *attr,
5841 char *buf)
5842{
5843 return sprintf(buf, "%d\n", perf_overcommit);
5844}
5845
5846static ssize_t
5847perf_set_overcommit(struct sysdev_class *class,
5848 struct sysdev_class_attribute *attr,
5849 const char *buf, size_t count)
5850{
5851 unsigned long val;
5852 int err;
5853
5854 err = strict_strtoul(buf, 10, &val);
5855 if (err)
5856 return err;
5857 if (val > 1)
5858 return -EINVAL;
5859
5860 spin_lock(&perf_resource_lock);
5861 perf_overcommit = val;
5862 spin_unlock(&perf_resource_lock);
5863
5864 return count;
5865}
5866
5867static SYSDEV_CLASS_ATTR(
5868 reserve_percpu,
5869 0644,
5870 perf_show_reserve_percpu,
5871 perf_set_reserve_percpu
5872 );
5873
5874static SYSDEV_CLASS_ATTR(
5875 overcommit,
5876 0644,
5877 perf_show_overcommit,
5878 perf_set_overcommit
5879 );
5880
5881static struct attribute *perfclass_attrs[] = {
5882 &attr_reserve_percpu.attr,
5883 &attr_overcommit.attr,
5884 NULL
5885};
5886
5887static struct attribute_group perfclass_attr_group = {
5888 .attrs = perfclass_attrs,
5889 .name = "perf_events",
5890};
5891
5892static int __init perf_event_sysfs_init(void)
5893{
5894 return sysfs_create_group(&cpu_sysdev_class.kset.kobj,
5895 &perfclass_attr_group);
5896} 6337}
5897device_initcall(perf_event_sysfs_init);
diff --git a/kernel/pid.c b/kernel/pid.c
index d55c6fb8d087..39b65b69584f 100644
--- a/kernel/pid.c
+++ b/kernel/pid.c
@@ -401,7 +401,7 @@ struct task_struct *pid_task(struct pid *pid, enum pid_type type)
401 struct task_struct *result = NULL; 401 struct task_struct *result = NULL;
402 if (pid) { 402 if (pid) {
403 struct hlist_node *first; 403 struct hlist_node *first;
404 first = rcu_dereference_check(pid->tasks[type].first, 404 first = rcu_dereference_check(hlist_first_rcu(&pid->tasks[type]),
405 rcu_read_lock_held() || 405 rcu_read_lock_held() ||
406 lockdep_tasklist_lock_is_held()); 406 lockdep_tasklist_lock_is_held());
407 if (first) 407 if (first)
@@ -416,6 +416,7 @@ EXPORT_SYMBOL(pid_task);
416 */ 416 */
417struct task_struct *find_task_by_pid_ns(pid_t nr, struct pid_namespace *ns) 417struct task_struct *find_task_by_pid_ns(pid_t nr, struct pid_namespace *ns)
418{ 418{
419 rcu_lockdep_assert(rcu_read_lock_held());
419 return pid_task(find_pid_ns(nr, ns), PIDTYPE_PID); 420 return pid_task(find_pid_ns(nr, ns), PIDTYPE_PID);
420} 421}
421 422
diff --git a/kernel/pm_qos_params.c b/kernel/pm_qos_params.c
index 645e541a45f6..a96b850ba08a 100644
--- a/kernel/pm_qos_params.c
+++ b/kernel/pm_qos_params.c
@@ -110,6 +110,7 @@ static const struct file_operations pm_qos_power_fops = {
110 .write = pm_qos_power_write, 110 .write = pm_qos_power_write,
111 .open = pm_qos_power_open, 111 .open = pm_qos_power_open,
112 .release = pm_qos_power_release, 112 .release = pm_qos_power_release,
113 .llseek = noop_llseek,
113}; 114};
114 115
115/* unlocked internal variant */ 116/* unlocked internal variant */
diff --git a/kernel/power/Kconfig b/kernel/power/Kconfig
index ca6066a6952e..29bff6117abc 100644
--- a/kernel/power/Kconfig
+++ b/kernel/power/Kconfig
@@ -86,6 +86,7 @@ config PM_SLEEP_SMP
86 depends on SMP 86 depends on SMP
87 depends on ARCH_SUSPEND_POSSIBLE || ARCH_HIBERNATION_POSSIBLE 87 depends on ARCH_SUSPEND_POSSIBLE || ARCH_HIBERNATION_POSSIBLE
88 depends on PM_SLEEP 88 depends on PM_SLEEP
89 select HOTPLUG
89 select HOTPLUG_CPU 90 select HOTPLUG_CPU
90 default y 91 default y
91 92
@@ -137,6 +138,8 @@ config SUSPEND_FREEZER
137config HIBERNATION 138config HIBERNATION
138 bool "Hibernation (aka 'suspend to disk')" 139 bool "Hibernation (aka 'suspend to disk')"
139 depends on PM && SWAP && ARCH_HIBERNATION_POSSIBLE 140 depends on PM && SWAP && ARCH_HIBERNATION_POSSIBLE
141 select LZO_COMPRESS
142 select LZO_DECOMPRESS
140 select SUSPEND_NVS if HAS_IOMEM 143 select SUSPEND_NVS if HAS_IOMEM
141 ---help--- 144 ---help---
142 Enable the suspend to disk (STD) functionality, which is usually 145 Enable the suspend to disk (STD) functionality, which is usually
@@ -242,3 +245,17 @@ config PM_OPS
242 bool 245 bool
243 depends on PM_SLEEP || PM_RUNTIME 246 depends on PM_SLEEP || PM_RUNTIME
244 default y 247 default y
248
249config PM_OPP
250 bool "Operating Performance Point (OPP) Layer library"
251 depends on PM
252 ---help---
253 SOCs have a standard set of tuples consisting of frequency and
254 voltage pairs that the device will support per voltage domain. This
255 is called Operating Performance Point or OPP. The actual definitions
256 of OPP varies over silicon within the same family of devices.
257
258 OPP layer organizes the data internally using device pointers
259 representing individual voltage domains and provides SOC
260 implementations a ready to use framework to manage OPPs.
261 For more information, read <file:Documentation/power/opp.txt>
diff --git a/kernel/power/hibernate.c b/kernel/power/hibernate.c
index 8dc31e02ae12..657272e91d0a 100644
--- a/kernel/power/hibernate.c
+++ b/kernel/power/hibernate.c
@@ -29,6 +29,7 @@
29#include "power.h" 29#include "power.h"
30 30
31 31
32static int nocompress = 0;
32static int noresume = 0; 33static int noresume = 0;
33static char resume_file[256] = CONFIG_PM_STD_PARTITION; 34static char resume_file[256] = CONFIG_PM_STD_PARTITION;
34dev_t swsusp_resume_device; 35dev_t swsusp_resume_device;
@@ -638,6 +639,8 @@ int hibernate(void)
638 639
639 if (hibernation_mode == HIBERNATION_PLATFORM) 640 if (hibernation_mode == HIBERNATION_PLATFORM)
640 flags |= SF_PLATFORM_MODE; 641 flags |= SF_PLATFORM_MODE;
642 if (nocompress)
643 flags |= SF_NOCOMPRESS_MODE;
641 pr_debug("PM: writing image.\n"); 644 pr_debug("PM: writing image.\n");
642 error = swsusp_write(flags); 645 error = swsusp_write(flags);
643 swsusp_free(); 646 swsusp_free();
@@ -705,7 +708,7 @@ static int software_resume(void)
705 goto Unlock; 708 goto Unlock;
706 } 709 }
707 710
708 pr_debug("PM: Checking image partition %s\n", resume_file); 711 pr_debug("PM: Checking hibernation image partition %s\n", resume_file);
709 712
710 /* Check if the device is there */ 713 /* Check if the device is there */
711 swsusp_resume_device = name_to_dev_t(resume_file); 714 swsusp_resume_device = name_to_dev_t(resume_file);
@@ -730,10 +733,10 @@ static int software_resume(void)
730 } 733 }
731 734
732 Check_image: 735 Check_image:
733 pr_debug("PM: Resume from partition %d:%d\n", 736 pr_debug("PM: Hibernation image partition %d:%d present\n",
734 MAJOR(swsusp_resume_device), MINOR(swsusp_resume_device)); 737 MAJOR(swsusp_resume_device), MINOR(swsusp_resume_device));
735 738
736 pr_debug("PM: Checking hibernation image.\n"); 739 pr_debug("PM: Looking for hibernation image.\n");
737 error = swsusp_check(); 740 error = swsusp_check();
738 if (error) 741 if (error)
739 goto Unlock; 742 goto Unlock;
@@ -765,14 +768,14 @@ static int software_resume(void)
765 goto Done; 768 goto Done;
766 } 769 }
767 770
768 pr_debug("PM: Reading hibernation image.\n"); 771 pr_debug("PM: Loading hibernation image.\n");
769 772
770 error = swsusp_read(&flags); 773 error = swsusp_read(&flags);
771 swsusp_close(FMODE_READ); 774 swsusp_close(FMODE_READ);
772 if (!error) 775 if (!error)
773 hibernation_restore(flags & SF_PLATFORM_MODE); 776 hibernation_restore(flags & SF_PLATFORM_MODE);
774 777
775 printk(KERN_ERR "PM: Restore failed, recovering.\n"); 778 printk(KERN_ERR "PM: Failed to load hibernation image, recovering.\n");
776 swsusp_free(); 779 swsusp_free();
777 thaw_processes(); 780 thaw_processes();
778 Done: 781 Done:
@@ -785,7 +788,7 @@ static int software_resume(void)
785 /* For success case, the suspend path will release the lock */ 788 /* For success case, the suspend path will release the lock */
786 Unlock: 789 Unlock:
787 mutex_unlock(&pm_mutex); 790 mutex_unlock(&pm_mutex);
788 pr_debug("PM: Resume from disk failed.\n"); 791 pr_debug("PM: Hibernation image not present or could not be loaded.\n");
789 return error; 792 return error;
790close_finish: 793close_finish:
791 swsusp_close(FMODE_READ); 794 swsusp_close(FMODE_READ);
@@ -1004,6 +1007,15 @@ static int __init resume_offset_setup(char *str)
1004 return 1; 1007 return 1;
1005} 1008}
1006 1009
1010static int __init hibernate_setup(char *str)
1011{
1012 if (!strncmp(str, "noresume", 8))
1013 noresume = 1;
1014 else if (!strncmp(str, "nocompress", 10))
1015 nocompress = 1;
1016 return 1;
1017}
1018
1007static int __init noresume_setup(char *str) 1019static int __init noresume_setup(char *str)
1008{ 1020{
1009 noresume = 1; 1021 noresume = 1;
@@ -1013,3 +1025,4 @@ static int __init noresume_setup(char *str)
1013__setup("noresume", noresume_setup); 1025__setup("noresume", noresume_setup);
1014__setup("resume_offset=", resume_offset_setup); 1026__setup("resume_offset=", resume_offset_setup);
1015__setup("resume=", resume_setup); 1027__setup("resume=", resume_setup);
1028__setup("hibernate=", hibernate_setup);
diff --git a/kernel/power/main.c b/kernel/power/main.c
index 62b0bc6e4983..7b5db6a8561e 100644
--- a/kernel/power/main.c
+++ b/kernel/power/main.c
@@ -237,18 +237,18 @@ static ssize_t wakeup_count_show(struct kobject *kobj,
237 struct kobj_attribute *attr, 237 struct kobj_attribute *attr,
238 char *buf) 238 char *buf)
239{ 239{
240 unsigned long val; 240 unsigned int val;
241 241
242 return pm_get_wakeup_count(&val) ? sprintf(buf, "%lu\n", val) : -EINTR; 242 return pm_get_wakeup_count(&val) ? sprintf(buf, "%u\n", val) : -EINTR;
243} 243}
244 244
245static ssize_t wakeup_count_store(struct kobject *kobj, 245static ssize_t wakeup_count_store(struct kobject *kobj,
246 struct kobj_attribute *attr, 246 struct kobj_attribute *attr,
247 const char *buf, size_t n) 247 const char *buf, size_t n)
248{ 248{
249 unsigned long val; 249 unsigned int val;
250 250
251 if (sscanf(buf, "%lu", &val) == 1) { 251 if (sscanf(buf, "%u", &val) == 1) {
252 if (pm_save_wakeup_count(val)) 252 if (pm_save_wakeup_count(val))
253 return n; 253 return n;
254 } 254 }
@@ -281,12 +281,30 @@ pm_trace_store(struct kobject *kobj, struct kobj_attribute *attr,
281} 281}
282 282
283power_attr(pm_trace); 283power_attr(pm_trace);
284
285static ssize_t pm_trace_dev_match_show(struct kobject *kobj,
286 struct kobj_attribute *attr,
287 char *buf)
288{
289 return show_trace_dev_match(buf, PAGE_SIZE);
290}
291
292static ssize_t
293pm_trace_dev_match_store(struct kobject *kobj, struct kobj_attribute *attr,
294 const char *buf, size_t n)
295{
296 return -EINVAL;
297}
298
299power_attr(pm_trace_dev_match);
300
284#endif /* CONFIG_PM_TRACE */ 301#endif /* CONFIG_PM_TRACE */
285 302
286static struct attribute * g[] = { 303static struct attribute * g[] = {
287 &state_attr.attr, 304 &state_attr.attr,
288#ifdef CONFIG_PM_TRACE 305#ifdef CONFIG_PM_TRACE
289 &pm_trace_attr.attr, 306 &pm_trace_attr.attr,
307 &pm_trace_dev_match_attr.attr,
290#endif 308#endif
291#ifdef CONFIG_PM_SLEEP 309#ifdef CONFIG_PM_SLEEP
292 &pm_async_attr.attr, 310 &pm_async_attr.attr,
@@ -308,7 +326,7 @@ EXPORT_SYMBOL_GPL(pm_wq);
308 326
309static int __init pm_start_workqueue(void) 327static int __init pm_start_workqueue(void)
310{ 328{
311 pm_wq = create_freezeable_workqueue("pm"); 329 pm_wq = alloc_workqueue("pm", WQ_FREEZEABLE, 0);
312 330
313 return pm_wq ? 0 : -ENOMEM; 331 return pm_wq ? 0 : -ENOMEM;
314} 332}
@@ -321,6 +339,7 @@ static int __init pm_init(void)
321 int error = pm_start_workqueue(); 339 int error = pm_start_workqueue();
322 if (error) 340 if (error)
323 return error; 341 return error;
342 hibernate_image_size_init();
324 power_kobj = kobject_create_and_add("power", NULL); 343 power_kobj = kobject_create_and_add("power", NULL);
325 if (!power_kobj) 344 if (!power_kobj)
326 return -ENOMEM; 345 return -ENOMEM;
diff --git a/kernel/power/power.h b/kernel/power/power.h
index 006270fe382d..03634be55f62 100644
--- a/kernel/power/power.h
+++ b/kernel/power/power.h
@@ -14,6 +14,9 @@ struct swsusp_info {
14} __attribute__((aligned(PAGE_SIZE))); 14} __attribute__((aligned(PAGE_SIZE)));
15 15
16#ifdef CONFIG_HIBERNATION 16#ifdef CONFIG_HIBERNATION
17/* kernel/power/snapshot.c */
18extern void __init hibernate_image_size_init(void);
19
17#ifdef CONFIG_ARCH_HIBERNATION_HEADER 20#ifdef CONFIG_ARCH_HIBERNATION_HEADER
18/* Maximum size of architecture specific data in a hibernation header */ 21/* Maximum size of architecture specific data in a hibernation header */
19#define MAX_ARCH_HEADER_SIZE (sizeof(struct new_utsname) + 4) 22#define MAX_ARCH_HEADER_SIZE (sizeof(struct new_utsname) + 4)
@@ -49,7 +52,11 @@ static inline char *check_image_kernel(struct swsusp_info *info)
49extern int hibernation_snapshot(int platform_mode); 52extern int hibernation_snapshot(int platform_mode);
50extern int hibernation_restore(int platform_mode); 53extern int hibernation_restore(int platform_mode);
51extern int hibernation_platform_enter(void); 54extern int hibernation_platform_enter(void);
52#endif 55
56#else /* !CONFIG_HIBERNATION */
57
58static inline void hibernate_image_size_init(void) {}
59#endif /* !CONFIG_HIBERNATION */
53 60
54extern int pfn_is_nosave(unsigned long); 61extern int pfn_is_nosave(unsigned long);
55 62
@@ -134,6 +141,7 @@ extern int swsusp_swap_in_use(void);
134 * the image header. 141 * the image header.
135 */ 142 */
136#define SF_PLATFORM_MODE 1 143#define SF_PLATFORM_MODE 1
144#define SF_NOCOMPRESS_MODE 2
137 145
138/* kernel/power/hibernate.c */ 146/* kernel/power/hibernate.c */
139extern int swsusp_check(void); 147extern int swsusp_check(void);
diff --git a/kernel/power/process.c b/kernel/power/process.c
index 028a99598f49..e50b4c1b2a0f 100644
--- a/kernel/power/process.c
+++ b/kernel/power/process.c
@@ -40,6 +40,7 @@ static int try_to_freeze_tasks(bool sig_only)
40 struct timeval start, end; 40 struct timeval start, end;
41 u64 elapsed_csecs64; 41 u64 elapsed_csecs64;
42 unsigned int elapsed_csecs; 42 unsigned int elapsed_csecs;
43 bool wakeup = false;
43 44
44 do_gettimeofday(&start); 45 do_gettimeofday(&start);
45 46
@@ -78,6 +79,11 @@ static int try_to_freeze_tasks(bool sig_only)
78 if (!todo || time_after(jiffies, end_time)) 79 if (!todo || time_after(jiffies, end_time))
79 break; 80 break;
80 81
82 if (!pm_check_wakeup_events()) {
83 wakeup = true;
84 break;
85 }
86
81 /* 87 /*
82 * We need to retry, but first give the freezing tasks some 88 * We need to retry, but first give the freezing tasks some
83 * time to enter the regrigerator. 89 * time to enter the regrigerator.
@@ -97,8 +103,9 @@ static int try_to_freeze_tasks(bool sig_only)
97 * but it cleans up leftover PF_FREEZE requests. 103 * but it cleans up leftover PF_FREEZE requests.
98 */ 104 */
99 printk("\n"); 105 printk("\n");
100 printk(KERN_ERR "Freezing of tasks failed after %d.%02d seconds " 106 printk(KERN_ERR "Freezing of tasks %s after %d.%02d seconds "
101 "(%d tasks refusing to freeze, wq_busy=%d):\n", 107 "(%d tasks refusing to freeze, wq_busy=%d):\n",
108 wakeup ? "aborted" : "failed",
102 elapsed_csecs / 100, elapsed_csecs % 100, 109 elapsed_csecs / 100, elapsed_csecs % 100,
103 todo - wq_busy, wq_busy); 110 todo - wq_busy, wq_busy);
104 111
@@ -107,7 +114,7 @@ static int try_to_freeze_tasks(bool sig_only)
107 read_lock(&tasklist_lock); 114 read_lock(&tasklist_lock);
108 do_each_thread(g, p) { 115 do_each_thread(g, p) {
109 task_lock(p); 116 task_lock(p);
110 if (freezing(p) && !freezer_should_skip(p)) 117 if (!wakeup && freezing(p) && !freezer_should_skip(p))
111 sched_show_task(p); 118 sched_show_task(p);
112 cancel_freezing(p); 119 cancel_freezing(p);
113 task_unlock(p); 120 task_unlock(p);
diff --git a/kernel/power/snapshot.c b/kernel/power/snapshot.c
index d3f795f01bbc..ac7eb109f196 100644
--- a/kernel/power/snapshot.c
+++ b/kernel/power/snapshot.c
@@ -46,7 +46,12 @@ static void swsusp_unset_page_forbidden(struct page *);
46 * size will not exceed N bytes, but if that is impossible, it will 46 * size will not exceed N bytes, but if that is impossible, it will
47 * try to create the smallest image possible. 47 * try to create the smallest image possible.
48 */ 48 */
49unsigned long image_size = 500 * 1024 * 1024; 49unsigned long image_size;
50
51void __init hibernate_image_size_init(void)
52{
53 image_size = ((totalram_pages * 2) / 5) * PAGE_SIZE;
54}
50 55
51/* List of PBEs needed for restoring the pages that were allocated before 56/* List of PBEs needed for restoring the pages that were allocated before
52 * the suspend and included in the suspend image, but have also been 57 * the suspend and included in the suspend image, but have also been
@@ -1318,12 +1323,14 @@ int hibernate_preallocate_memory(void)
1318 1323
1319 /* Compute the maximum number of saveable pages to leave in memory. */ 1324 /* Compute the maximum number of saveable pages to leave in memory. */
1320 max_size = (count - (size + PAGES_FOR_IO)) / 2 - 2 * SPARE_PAGES; 1325 max_size = (count - (size + PAGES_FOR_IO)) / 2 - 2 * SPARE_PAGES;
1326 /* Compute the desired number of image pages specified by image_size. */
1321 size = DIV_ROUND_UP(image_size, PAGE_SIZE); 1327 size = DIV_ROUND_UP(image_size, PAGE_SIZE);
1322 if (size > max_size) 1328 if (size > max_size)
1323 size = max_size; 1329 size = max_size;
1324 /* 1330 /*
1325 * If the maximum is not less than the current number of saveable pages 1331 * If the desired number of image pages is at least as large as the
1326 * in memory, allocate page frames for the image and we're done. 1332 * current number of saveable pages in memory, allocate page frames for
1333 * the image and we're done.
1327 */ 1334 */
1328 if (size >= saveable) { 1335 if (size >= saveable) {
1329 pages = preallocate_image_highmem(save_highmem); 1336 pages = preallocate_image_highmem(save_highmem);
diff --git a/kernel/power/swap.c b/kernel/power/swap.c
index e6a5bdf61a37..916eaa790399 100644
--- a/kernel/power/swap.c
+++ b/kernel/power/swap.c
@@ -24,10 +24,12 @@
24#include <linux/swapops.h> 24#include <linux/swapops.h>
25#include <linux/pm.h> 25#include <linux/pm.h>
26#include <linux/slab.h> 26#include <linux/slab.h>
27#include <linux/lzo.h>
28#include <linux/vmalloc.h>
27 29
28#include "power.h" 30#include "power.h"
29 31
30#define SWSUSP_SIG "S1SUSPEND" 32#define HIBERNATE_SIG "LINHIB0001"
31 33
32/* 34/*
33 * The swap map is a data structure used for keeping track of each page 35 * The swap map is a data structure used for keeping track of each page
@@ -193,7 +195,7 @@ static int mark_swapfiles(struct swap_map_handle *handle, unsigned int flags)
193 if (!memcmp("SWAP-SPACE",swsusp_header->sig, 10) || 195 if (!memcmp("SWAP-SPACE",swsusp_header->sig, 10) ||
194 !memcmp("SWAPSPACE2",swsusp_header->sig, 10)) { 196 !memcmp("SWAPSPACE2",swsusp_header->sig, 10)) {
195 memcpy(swsusp_header->orig_sig,swsusp_header->sig, 10); 197 memcpy(swsusp_header->orig_sig,swsusp_header->sig, 10);
196 memcpy(swsusp_header->sig,SWSUSP_SIG, 10); 198 memcpy(swsusp_header->sig, HIBERNATE_SIG, 10);
197 swsusp_header->image = handle->first_sector; 199 swsusp_header->image = handle->first_sector;
198 swsusp_header->flags = flags; 200 swsusp_header->flags = flags;
199 error = hib_bio_write_page(swsusp_resume_block, 201 error = hib_bio_write_page(swsusp_resume_block,
@@ -357,6 +359,18 @@ static int swap_writer_finish(struct swap_map_handle *handle,
357 return error; 359 return error;
358} 360}
359 361
362/* We need to remember how much compressed data we need to read. */
363#define LZO_HEADER sizeof(size_t)
364
365/* Number of pages/bytes we'll compress at one time. */
366#define LZO_UNC_PAGES 32
367#define LZO_UNC_SIZE (LZO_UNC_PAGES * PAGE_SIZE)
368
369/* Number of pages/bytes we need for compressed data (worst case). */
370#define LZO_CMP_PAGES DIV_ROUND_UP(lzo1x_worst_compress(LZO_UNC_SIZE) + \
371 LZO_HEADER, PAGE_SIZE)
372#define LZO_CMP_SIZE (LZO_CMP_PAGES * PAGE_SIZE)
373
360/** 374/**
361 * save_image - save the suspend image data 375 * save_image - save the suspend image data
362 */ 376 */
@@ -404,6 +418,137 @@ static int save_image(struct swap_map_handle *handle,
404 return ret; 418 return ret;
405} 419}
406 420
421
422/**
423 * save_image_lzo - Save the suspend image data compressed with LZO.
424 * @handle: Swap mam handle to use for saving the image.
425 * @snapshot: Image to read data from.
426 * @nr_to_write: Number of pages to save.
427 */
428static int save_image_lzo(struct swap_map_handle *handle,
429 struct snapshot_handle *snapshot,
430 unsigned int nr_to_write)
431{
432 unsigned int m;
433 int ret = 0;
434 int nr_pages;
435 int err2;
436 struct bio *bio;
437 struct timeval start;
438 struct timeval stop;
439 size_t off, unc_len, cmp_len;
440 unsigned char *unc, *cmp, *wrk, *page;
441
442 page = (void *)__get_free_page(__GFP_WAIT | __GFP_HIGH);
443 if (!page) {
444 printk(KERN_ERR "PM: Failed to allocate LZO page\n");
445 return -ENOMEM;
446 }
447
448 wrk = vmalloc(LZO1X_1_MEM_COMPRESS);
449 if (!wrk) {
450 printk(KERN_ERR "PM: Failed to allocate LZO workspace\n");
451 free_page((unsigned long)page);
452 return -ENOMEM;
453 }
454
455 unc = vmalloc(LZO_UNC_SIZE);
456 if (!unc) {
457 printk(KERN_ERR "PM: Failed to allocate LZO uncompressed\n");
458 vfree(wrk);
459 free_page((unsigned long)page);
460 return -ENOMEM;
461 }
462
463 cmp = vmalloc(LZO_CMP_SIZE);
464 if (!cmp) {
465 printk(KERN_ERR "PM: Failed to allocate LZO compressed\n");
466 vfree(unc);
467 vfree(wrk);
468 free_page((unsigned long)page);
469 return -ENOMEM;
470 }
471
472 printk(KERN_INFO
473 "PM: Compressing and saving image data (%u pages) ... ",
474 nr_to_write);
475 m = nr_to_write / 100;
476 if (!m)
477 m = 1;
478 nr_pages = 0;
479 bio = NULL;
480 do_gettimeofday(&start);
481 for (;;) {
482 for (off = 0; off < LZO_UNC_SIZE; off += PAGE_SIZE) {
483 ret = snapshot_read_next(snapshot);
484 if (ret < 0)
485 goto out_finish;
486
487 if (!ret)
488 break;
489
490 memcpy(unc + off, data_of(*snapshot), PAGE_SIZE);
491
492 if (!(nr_pages % m))
493 printk(KERN_CONT "\b\b\b\b%3d%%", nr_pages / m);
494 nr_pages++;
495 }
496
497 if (!off)
498 break;
499
500 unc_len = off;
501 ret = lzo1x_1_compress(unc, unc_len,
502 cmp + LZO_HEADER, &cmp_len, wrk);
503 if (ret < 0) {
504 printk(KERN_ERR "PM: LZO compression failed\n");
505 break;
506 }
507
508 if (unlikely(!cmp_len ||
509 cmp_len > lzo1x_worst_compress(unc_len))) {
510 printk(KERN_ERR "PM: Invalid LZO compressed length\n");
511 ret = -1;
512 break;
513 }
514
515 *(size_t *)cmp = cmp_len;
516
517 /*
518 * Given we are writing one page at a time to disk, we copy
519 * that much from the buffer, although the last bit will likely
520 * be smaller than full page. This is OK - we saved the length
521 * of the compressed data, so any garbage at the end will be
522 * discarded when we read it.
523 */
524 for (off = 0; off < LZO_HEADER + cmp_len; off += PAGE_SIZE) {
525 memcpy(page, cmp + off, PAGE_SIZE);
526
527 ret = swap_write_page(handle, page, &bio);
528 if (ret)
529 goto out_finish;
530 }
531 }
532
533out_finish:
534 err2 = hib_wait_on_bio_chain(&bio);
535 do_gettimeofday(&stop);
536 if (!ret)
537 ret = err2;
538 if (!ret)
539 printk(KERN_CONT "\b\b\b\bdone\n");
540 else
541 printk(KERN_CONT "\n");
542 swsusp_show_speed(&start, &stop, nr_to_write, "Wrote");
543
544 vfree(cmp);
545 vfree(unc);
546 vfree(wrk);
547 free_page((unsigned long)page);
548
549 return ret;
550}
551
407/** 552/**
408 * enough_swap - Make sure we have enough swap to save the image. 553 * enough_swap - Make sure we have enough swap to save the image.
409 * 554 *
@@ -411,12 +556,16 @@ static int save_image(struct swap_map_handle *handle,
411 * space avaiable from the resume partition. 556 * space avaiable from the resume partition.
412 */ 557 */
413 558
414static int enough_swap(unsigned int nr_pages) 559static int enough_swap(unsigned int nr_pages, unsigned int flags)
415{ 560{
416 unsigned int free_swap = count_swap_pages(root_swap, 1); 561 unsigned int free_swap = count_swap_pages(root_swap, 1);
562 unsigned int required;
417 563
418 pr_debug("PM: Free swap pages: %u\n", free_swap); 564 pr_debug("PM: Free swap pages: %u\n", free_swap);
419 return free_swap > nr_pages + PAGES_FOR_IO; 565
566 required = PAGES_FOR_IO + ((flags & SF_NOCOMPRESS_MODE) ?
567 nr_pages : (nr_pages * LZO_CMP_PAGES) / LZO_UNC_PAGES + 1);
568 return free_swap > required;
420} 569}
421 570
422/** 571/**
@@ -443,7 +592,7 @@ int swsusp_write(unsigned int flags)
443 printk(KERN_ERR "PM: Cannot get swap writer\n"); 592 printk(KERN_ERR "PM: Cannot get swap writer\n");
444 return error; 593 return error;
445 } 594 }
446 if (!enough_swap(pages)) { 595 if (!enough_swap(pages, flags)) {
447 printk(KERN_ERR "PM: Not enough free swap\n"); 596 printk(KERN_ERR "PM: Not enough free swap\n");
448 error = -ENOSPC; 597 error = -ENOSPC;
449 goto out_finish; 598 goto out_finish;
@@ -458,8 +607,11 @@ int swsusp_write(unsigned int flags)
458 } 607 }
459 header = (struct swsusp_info *)data_of(snapshot); 608 header = (struct swsusp_info *)data_of(snapshot);
460 error = swap_write_page(&handle, header, NULL); 609 error = swap_write_page(&handle, header, NULL);
461 if (!error) 610 if (!error) {
462 error = save_image(&handle, &snapshot, pages - 1); 611 error = (flags & SF_NOCOMPRESS_MODE) ?
612 save_image(&handle, &snapshot, pages - 1) :
613 save_image_lzo(&handle, &snapshot, pages - 1);
614 }
463out_finish: 615out_finish:
464 error = swap_writer_finish(&handle, flags, error); 616 error = swap_writer_finish(&handle, flags, error);
465 return error; 617 return error;
@@ -590,6 +742,127 @@ static int load_image(struct swap_map_handle *handle,
590} 742}
591 743
592/** 744/**
745 * load_image_lzo - Load compressed image data and decompress them with LZO.
746 * @handle: Swap map handle to use for loading data.
747 * @snapshot: Image to copy uncompressed data into.
748 * @nr_to_read: Number of pages to load.
749 */
750static int load_image_lzo(struct swap_map_handle *handle,
751 struct snapshot_handle *snapshot,
752 unsigned int nr_to_read)
753{
754 unsigned int m;
755 int error = 0;
756 struct timeval start;
757 struct timeval stop;
758 unsigned nr_pages;
759 size_t off, unc_len, cmp_len;
760 unsigned char *unc, *cmp, *page;
761
762 page = (void *)__get_free_page(__GFP_WAIT | __GFP_HIGH);
763 if (!page) {
764 printk(KERN_ERR "PM: Failed to allocate LZO page\n");
765 return -ENOMEM;
766 }
767
768 unc = vmalloc(LZO_UNC_SIZE);
769 if (!unc) {
770 printk(KERN_ERR "PM: Failed to allocate LZO uncompressed\n");
771 free_page((unsigned long)page);
772 return -ENOMEM;
773 }
774
775 cmp = vmalloc(LZO_CMP_SIZE);
776 if (!cmp) {
777 printk(KERN_ERR "PM: Failed to allocate LZO compressed\n");
778 vfree(unc);
779 free_page((unsigned long)page);
780 return -ENOMEM;
781 }
782
783 printk(KERN_INFO
784 "PM: Loading and decompressing image data (%u pages) ... ",
785 nr_to_read);
786 m = nr_to_read / 100;
787 if (!m)
788 m = 1;
789 nr_pages = 0;
790 do_gettimeofday(&start);
791
792 error = snapshot_write_next(snapshot);
793 if (error <= 0)
794 goto out_finish;
795
796 for (;;) {
797 error = swap_read_page(handle, page, NULL); /* sync */
798 if (error)
799 break;
800
801 cmp_len = *(size_t *)page;
802 if (unlikely(!cmp_len ||
803 cmp_len > lzo1x_worst_compress(LZO_UNC_SIZE))) {
804 printk(KERN_ERR "PM: Invalid LZO compressed length\n");
805 error = -1;
806 break;
807 }
808
809 memcpy(cmp, page, PAGE_SIZE);
810 for (off = PAGE_SIZE; off < LZO_HEADER + cmp_len; off += PAGE_SIZE) {
811 error = swap_read_page(handle, page, NULL); /* sync */
812 if (error)
813 goto out_finish;
814
815 memcpy(cmp + off, page, PAGE_SIZE);
816 }
817
818 unc_len = LZO_UNC_SIZE;
819 error = lzo1x_decompress_safe(cmp + LZO_HEADER, cmp_len,
820 unc, &unc_len);
821 if (error < 0) {
822 printk(KERN_ERR "PM: LZO decompression failed\n");
823 break;
824 }
825
826 if (unlikely(!unc_len ||
827 unc_len > LZO_UNC_SIZE ||
828 unc_len & (PAGE_SIZE - 1))) {
829 printk(KERN_ERR "PM: Invalid LZO uncompressed length\n");
830 error = -1;
831 break;
832 }
833
834 for (off = 0; off < unc_len; off += PAGE_SIZE) {
835 memcpy(data_of(*snapshot), unc + off, PAGE_SIZE);
836
837 if (!(nr_pages % m))
838 printk("\b\b\b\b%3d%%", nr_pages / m);
839 nr_pages++;
840
841 error = snapshot_write_next(snapshot);
842 if (error <= 0)
843 goto out_finish;
844 }
845 }
846
847out_finish:
848 do_gettimeofday(&stop);
849 if (!error) {
850 printk("\b\b\b\bdone\n");
851 snapshot_write_finalize(snapshot);
852 if (!snapshot_image_loaded(snapshot))
853 error = -ENODATA;
854 } else
855 printk("\n");
856 swsusp_show_speed(&start, &stop, nr_to_read, "Read");
857
858 vfree(cmp);
859 vfree(unc);
860 free_page((unsigned long)page);
861
862 return error;
863}
864
865/**
593 * swsusp_read - read the hibernation image. 866 * swsusp_read - read the hibernation image.
594 * @flags_p: flags passed by the "frozen" kernel in the image header should 867 * @flags_p: flags passed by the "frozen" kernel in the image header should
595 * be written into this memeory location 868 * be written into this memeory location
@@ -612,8 +885,11 @@ int swsusp_read(unsigned int *flags_p)
612 goto end; 885 goto end;
613 if (!error) 886 if (!error)
614 error = swap_read_page(&handle, header, NULL); 887 error = swap_read_page(&handle, header, NULL);
615 if (!error) 888 if (!error) {
616 error = load_image(&handle, &snapshot, header->pages - 1); 889 error = (*flags_p & SF_NOCOMPRESS_MODE) ?
890 load_image(&handle, &snapshot, header->pages - 1) :
891 load_image_lzo(&handle, &snapshot, header->pages - 1);
892 }
617 swap_reader_finish(&handle); 893 swap_reader_finish(&handle);
618end: 894end:
619 if (!error) 895 if (!error)
@@ -640,7 +916,7 @@ int swsusp_check(void)
640 if (error) 916 if (error)
641 goto put; 917 goto put;
642 918
643 if (!memcmp(SWSUSP_SIG, swsusp_header->sig, 10)) { 919 if (!memcmp(HIBERNATE_SIG, swsusp_header->sig, 10)) {
644 memcpy(swsusp_header->sig, swsusp_header->orig_sig, 10); 920 memcpy(swsusp_header->sig, swsusp_header->orig_sig, 10);
645 /* Reset swap signature now */ 921 /* Reset swap signature now */
646 error = hib_bio_write_page(swsusp_resume_block, 922 error = hib_bio_write_page(swsusp_resume_block,
@@ -653,13 +929,13 @@ put:
653 if (error) 929 if (error)
654 blkdev_put(hib_resume_bdev, FMODE_READ); 930 blkdev_put(hib_resume_bdev, FMODE_READ);
655 else 931 else
656 pr_debug("PM: Signature found, resuming\n"); 932 pr_debug("PM: Image signature found, resuming\n");
657 } else { 933 } else {
658 error = PTR_ERR(hib_resume_bdev); 934 error = PTR_ERR(hib_resume_bdev);
659 } 935 }
660 936
661 if (error) 937 if (error)
662 pr_debug("PM: Error %d checking image file\n", error); 938 pr_debug("PM: Image not found (code %d)\n", error);
663 939
664 return error; 940 return error;
665} 941}
diff --git a/kernel/printk.c b/kernel/printk.c
index 8fe465ac008a..2531017795f6 100644
--- a/kernel/printk.c
+++ b/kernel/printk.c
@@ -85,7 +85,7 @@ EXPORT_SYMBOL(oops_in_progress);
85 * provides serialisation for access to the entire console 85 * provides serialisation for access to the entire console
86 * driver system. 86 * driver system.
87 */ 87 */
88static DECLARE_MUTEX(console_sem); 88static DEFINE_SEMAPHORE(console_sem);
89struct console *console_drivers; 89struct console *console_drivers;
90EXPORT_SYMBOL_GPL(console_drivers); 90EXPORT_SYMBOL_GPL(console_drivers);
91 91
@@ -556,7 +556,7 @@ static void zap_locks(void)
556 /* If a crash is occurring, make sure we can't deadlock */ 556 /* If a crash is occurring, make sure we can't deadlock */
557 spin_lock_init(&logbuf_lock); 557 spin_lock_init(&logbuf_lock);
558 /* And make sure that we print immediately */ 558 /* And make sure that we print immediately */
559 init_MUTEX(&console_sem); 559 sema_init(&console_sem, 1);
560} 560}
561 561
562#if defined(CONFIG_PRINTK_TIME) 562#if defined(CONFIG_PRINTK_TIME)
diff --git a/kernel/profile.c b/kernel/profile.c
index b22a899934cc..66f841b7fbd3 100644
--- a/kernel/profile.c
+++ b/kernel/profile.c
@@ -555,6 +555,7 @@ static ssize_t write_profile(struct file *file, const char __user *buf,
555static const struct file_operations proc_profile_operations = { 555static const struct file_operations proc_profile_operations = {
556 .read = read_profile, 556 .read = read_profile,
557 .write = write_profile, 557 .write = write_profile,
558 .llseek = default_llseek,
558}; 559};
559 560
560#ifdef CONFIG_SMP 561#ifdef CONFIG_SMP
diff --git a/kernel/rcupdate.c b/kernel/rcupdate.c
index 4d169835fb36..a23a57a976d1 100644
--- a/kernel/rcupdate.c
+++ b/kernel/rcupdate.c
@@ -73,12 +73,14 @@ int debug_lockdep_rcu_enabled(void)
73EXPORT_SYMBOL_GPL(debug_lockdep_rcu_enabled); 73EXPORT_SYMBOL_GPL(debug_lockdep_rcu_enabled);
74 74
75/** 75/**
76 * rcu_read_lock_bh_held - might we be in RCU-bh read-side critical section? 76 * rcu_read_lock_bh_held() - might we be in RCU-bh read-side critical section?
77 * 77 *
78 * Check for bottom half being disabled, which covers both the 78 * Check for bottom half being disabled, which covers both the
79 * CONFIG_PROVE_RCU and not cases. Note that if someone uses 79 * CONFIG_PROVE_RCU and not cases. Note that if someone uses
80 * rcu_read_lock_bh(), but then later enables BH, lockdep (if enabled) 80 * rcu_read_lock_bh(), but then later enables BH, lockdep (if enabled)
81 * will show the situation. 81 * will show the situation. This is useful for debug checks in functions
82 * that require that they be called within an RCU read-side critical
83 * section.
82 * 84 *
83 * Check debug_lockdep_rcu_enabled() to prevent false positives during boot. 85 * Check debug_lockdep_rcu_enabled() to prevent false positives during boot.
84 */ 86 */
@@ -86,7 +88,7 @@ int rcu_read_lock_bh_held(void)
86{ 88{
87 if (!debug_lockdep_rcu_enabled()) 89 if (!debug_lockdep_rcu_enabled())
88 return 1; 90 return 1;
89 return in_softirq(); 91 return in_softirq() || irqs_disabled();
90} 92}
91EXPORT_SYMBOL_GPL(rcu_read_lock_bh_held); 93EXPORT_SYMBOL_GPL(rcu_read_lock_bh_held);
92 94
diff --git a/kernel/rcutiny.c b/kernel/rcutiny.c
index 196ec02f8be0..d806735342ac 100644
--- a/kernel/rcutiny.c
+++ b/kernel/rcutiny.c
@@ -59,6 +59,14 @@ int rcu_scheduler_active __read_mostly;
59EXPORT_SYMBOL_GPL(rcu_scheduler_active); 59EXPORT_SYMBOL_GPL(rcu_scheduler_active);
60#endif /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */ 60#endif /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */
61 61
62/* Forward declarations for rcutiny_plugin.h. */
63static void __rcu_process_callbacks(struct rcu_ctrlblk *rcp);
64static void __call_rcu(struct rcu_head *head,
65 void (*func)(struct rcu_head *rcu),
66 struct rcu_ctrlblk *rcp);
67
68#include "rcutiny_plugin.h"
69
62#ifdef CONFIG_NO_HZ 70#ifdef CONFIG_NO_HZ
63 71
64static long rcu_dynticks_nesting = 1; 72static long rcu_dynticks_nesting = 1;
@@ -140,6 +148,7 @@ void rcu_check_callbacks(int cpu, int user)
140 rcu_sched_qs(cpu); 148 rcu_sched_qs(cpu);
141 else if (!in_softirq()) 149 else if (!in_softirq())
142 rcu_bh_qs(cpu); 150 rcu_bh_qs(cpu);
151 rcu_preempt_check_callbacks();
143} 152}
144 153
145/* 154/*
@@ -162,6 +171,7 @@ static void __rcu_process_callbacks(struct rcu_ctrlblk *rcp)
162 *rcp->donetail = NULL; 171 *rcp->donetail = NULL;
163 if (rcp->curtail == rcp->donetail) 172 if (rcp->curtail == rcp->donetail)
164 rcp->curtail = &rcp->rcucblist; 173 rcp->curtail = &rcp->rcucblist;
174 rcu_preempt_remove_callbacks(rcp);
165 rcp->donetail = &rcp->rcucblist; 175 rcp->donetail = &rcp->rcucblist;
166 local_irq_restore(flags); 176 local_irq_restore(flags);
167 177
@@ -182,6 +192,7 @@ static void rcu_process_callbacks(struct softirq_action *unused)
182{ 192{
183 __rcu_process_callbacks(&rcu_sched_ctrlblk); 193 __rcu_process_callbacks(&rcu_sched_ctrlblk);
184 __rcu_process_callbacks(&rcu_bh_ctrlblk); 194 __rcu_process_callbacks(&rcu_bh_ctrlblk);
195 rcu_preempt_process_callbacks();
185} 196}
186 197
187/* 198/*
@@ -223,15 +234,15 @@ static void __call_rcu(struct rcu_head *head,
223} 234}
224 235
225/* 236/*
226 * Post an RCU callback to be invoked after the end of an RCU grace 237 * Post an RCU callback to be invoked after the end of an RCU-sched grace
227 * period. But since we have but one CPU, that would be after any 238 * period. But since we have but one CPU, that would be after any
228 * quiescent state. 239 * quiescent state.
229 */ 240 */
230void call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu)) 241void call_rcu_sched(struct rcu_head *head, void (*func)(struct rcu_head *rcu))
231{ 242{
232 __call_rcu(head, func, &rcu_sched_ctrlblk); 243 __call_rcu(head, func, &rcu_sched_ctrlblk);
233} 244}
234EXPORT_SYMBOL_GPL(call_rcu); 245EXPORT_SYMBOL_GPL(call_rcu_sched);
235 246
236/* 247/*
237 * Post an RCU bottom-half callback to be invoked after any subsequent 248 * Post an RCU bottom-half callback to be invoked after any subsequent
@@ -243,20 +254,6 @@ void call_rcu_bh(struct rcu_head *head, void (*func)(struct rcu_head *rcu))
243} 254}
244EXPORT_SYMBOL_GPL(call_rcu_bh); 255EXPORT_SYMBOL_GPL(call_rcu_bh);
245 256
246void rcu_barrier(void)
247{
248 struct rcu_synchronize rcu;
249
250 init_rcu_head_on_stack(&rcu.head);
251 init_completion(&rcu.completion);
252 /* Will wake me after RCU finished. */
253 call_rcu(&rcu.head, wakeme_after_rcu);
254 /* Wait for it. */
255 wait_for_completion(&rcu.completion);
256 destroy_rcu_head_on_stack(&rcu.head);
257}
258EXPORT_SYMBOL_GPL(rcu_barrier);
259
260void rcu_barrier_bh(void) 257void rcu_barrier_bh(void)
261{ 258{
262 struct rcu_synchronize rcu; 259 struct rcu_synchronize rcu;
@@ -289,5 +286,3 @@ void __init rcu_init(void)
289{ 286{
290 open_softirq(RCU_SOFTIRQ, rcu_process_callbacks); 287 open_softirq(RCU_SOFTIRQ, rcu_process_callbacks);
291} 288}
292
293#include "rcutiny_plugin.h"
diff --git a/kernel/rcutiny_plugin.h b/kernel/rcutiny_plugin.h
index d223a92bc742..6ceca4f745ff 100644
--- a/kernel/rcutiny_plugin.h
+++ b/kernel/rcutiny_plugin.h
@@ -1,7 +1,7 @@
1/* 1/*
2 * Read-Copy Update mechanism for mutual exclusion (tree-based version) 2 * Read-Copy Update mechanism for mutual exclusion, the Bloatwatch edition
3 * Internal non-public definitions that provide either classic 3 * Internal non-public definitions that provide either classic
4 * or preemptable semantics. 4 * or preemptible semantics.
5 * 5 *
6 * This program is free software; you can redistribute it and/or modify 6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License as published by 7 * it under the terms of the GNU General Public License as published by
@@ -17,11 +17,587 @@
17 * along with this program; if not, write to the Free Software 17 * along with this program; if not, write to the Free Software
18 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA. 18 * Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
19 * 19 *
20 * Copyright IBM Corporation, 2009 20 * Copyright (c) 2010 Linaro
21 * 21 *
22 * Author: Paul E. McKenney <paulmck@linux.vnet.ibm.com> 22 * Author: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
23 */ 23 */
24 24
25#ifdef CONFIG_TINY_PREEMPT_RCU
26
27#include <linux/delay.h>
28
29/* Global control variables for preemptible RCU. */
30struct rcu_preempt_ctrlblk {
31 struct rcu_ctrlblk rcb; /* curtail: ->next ptr of last CB for GP. */
32 struct rcu_head **nexttail;
33 /* Tasks blocked in a preemptible RCU */
34 /* read-side critical section while an */
35 /* preemptible-RCU grace period is in */
36 /* progress must wait for a later grace */
37 /* period. This pointer points to the */
38 /* ->next pointer of the last task that */
39 /* must wait for a later grace period, or */
40 /* to &->rcb.rcucblist if there is no */
41 /* such task. */
42 struct list_head blkd_tasks;
43 /* Tasks blocked in RCU read-side critical */
44 /* section. Tasks are placed at the head */
45 /* of this list and age towards the tail. */
46 struct list_head *gp_tasks;
47 /* Pointer to the first task blocking the */
48 /* current grace period, or NULL if there */
49 /* is not such task. */
50 struct list_head *exp_tasks;
51 /* Pointer to first task blocking the */
52 /* current expedited grace period, or NULL */
53 /* if there is no such task. If there */
54 /* is no current expedited grace period, */
55 /* then there cannot be any such task. */
56 u8 gpnum; /* Current grace period. */
57 u8 gpcpu; /* Last grace period blocked by the CPU. */
58 u8 completed; /* Last grace period completed. */
59 /* If all three are equal, RCU is idle. */
60};
61
62static struct rcu_preempt_ctrlblk rcu_preempt_ctrlblk = {
63 .rcb.donetail = &rcu_preempt_ctrlblk.rcb.rcucblist,
64 .rcb.curtail = &rcu_preempt_ctrlblk.rcb.rcucblist,
65 .nexttail = &rcu_preempt_ctrlblk.rcb.rcucblist,
66 .blkd_tasks = LIST_HEAD_INIT(rcu_preempt_ctrlblk.blkd_tasks),
67};
68
69static int rcu_preempted_readers_exp(void);
70static void rcu_report_exp_done(void);
71
72/*
73 * Return true if the CPU has not yet responded to the current grace period.
74 */
75static int rcu_cpu_blocking_cur_gp(void)
76{
77 return rcu_preempt_ctrlblk.gpcpu != rcu_preempt_ctrlblk.gpnum;
78}
79
80/*
81 * Check for a running RCU reader. Because there is only one CPU,
82 * there can be but one running RCU reader at a time. ;-)
83 */
84static int rcu_preempt_running_reader(void)
85{
86 return current->rcu_read_lock_nesting;
87}
88
89/*
90 * Check for preempted RCU readers blocking any grace period.
91 * If the caller needs a reliable answer, it must disable hard irqs.
92 */
93static int rcu_preempt_blocked_readers_any(void)
94{
95 return !list_empty(&rcu_preempt_ctrlblk.blkd_tasks);
96}
97
98/*
99 * Check for preempted RCU readers blocking the current grace period.
100 * If the caller needs a reliable answer, it must disable hard irqs.
101 */
102static int rcu_preempt_blocked_readers_cgp(void)
103{
104 return rcu_preempt_ctrlblk.gp_tasks != NULL;
105}
106
107/*
108 * Return true if another preemptible-RCU grace period is needed.
109 */
110static int rcu_preempt_needs_another_gp(void)
111{
112 return *rcu_preempt_ctrlblk.rcb.curtail != NULL;
113}
114
115/*
116 * Return true if a preemptible-RCU grace period is in progress.
117 * The caller must disable hardirqs.
118 */
119static int rcu_preempt_gp_in_progress(void)
120{
121 return rcu_preempt_ctrlblk.completed != rcu_preempt_ctrlblk.gpnum;
122}
123
124/*
125 * Record a preemptible-RCU quiescent state for the specified CPU. Note
126 * that this just means that the task currently running on the CPU is
127 * in a quiescent state. There might be any number of tasks blocked
128 * while in an RCU read-side critical section.
129 *
130 * Unlike the other rcu_*_qs() functions, callers to this function
131 * must disable irqs in order to protect the assignment to
132 * ->rcu_read_unlock_special.
133 *
134 * Because this is a single-CPU implementation, the only way a grace
135 * period can end is if the CPU is in a quiescent state. The reason is
136 * that a blocked preemptible-RCU reader can exit its critical section
137 * only if the CPU is running it at the time. Therefore, when the
138 * last task blocking the current grace period exits its RCU read-side
139 * critical section, neither the CPU nor blocked tasks will be stopping
140 * the current grace period. (In contrast, SMP implementations
141 * might have CPUs running in RCU read-side critical sections that
142 * block later grace periods -- but this is not possible given only
143 * one CPU.)
144 */
145static void rcu_preempt_cpu_qs(void)
146{
147 /* Record both CPU and task as having responded to current GP. */
148 rcu_preempt_ctrlblk.gpcpu = rcu_preempt_ctrlblk.gpnum;
149 current->rcu_read_unlock_special &= ~RCU_READ_UNLOCK_NEED_QS;
150
151 /*
152 * If there is no GP, or if blocked readers are still blocking GP,
153 * then there is nothing more to do.
154 */
155 if (!rcu_preempt_gp_in_progress() || rcu_preempt_blocked_readers_cgp())
156 return;
157
158 /* Advance callbacks. */
159 rcu_preempt_ctrlblk.completed = rcu_preempt_ctrlblk.gpnum;
160 rcu_preempt_ctrlblk.rcb.donetail = rcu_preempt_ctrlblk.rcb.curtail;
161 rcu_preempt_ctrlblk.rcb.curtail = rcu_preempt_ctrlblk.nexttail;
162
163 /* If there are no blocked readers, next GP is done instantly. */
164 if (!rcu_preempt_blocked_readers_any())
165 rcu_preempt_ctrlblk.rcb.donetail = rcu_preempt_ctrlblk.nexttail;
166
167 /* If there are done callbacks, make RCU_SOFTIRQ process them. */
168 if (*rcu_preempt_ctrlblk.rcb.donetail != NULL)
169 raise_softirq(RCU_SOFTIRQ);
170}
171
172/*
173 * Start a new RCU grace period if warranted. Hard irqs must be disabled.
174 */
175static void rcu_preempt_start_gp(void)
176{
177 if (!rcu_preempt_gp_in_progress() && rcu_preempt_needs_another_gp()) {
178
179 /* Official start of GP. */
180 rcu_preempt_ctrlblk.gpnum++;
181
182 /* Any blocked RCU readers block new GP. */
183 if (rcu_preempt_blocked_readers_any())
184 rcu_preempt_ctrlblk.gp_tasks =
185 rcu_preempt_ctrlblk.blkd_tasks.next;
186
187 /* If there is no running reader, CPU is done with GP. */
188 if (!rcu_preempt_running_reader())
189 rcu_preempt_cpu_qs();
190 }
191}
192
193/*
194 * We have entered the scheduler, and the current task might soon be
195 * context-switched away from. If this task is in an RCU read-side
196 * critical section, we will no longer be able to rely on the CPU to
197 * record that fact, so we enqueue the task on the blkd_tasks list.
198 * If the task started after the current grace period began, as recorded
199 * by ->gpcpu, we enqueue at the beginning of the list. Otherwise
200 * before the element referenced by ->gp_tasks (or at the tail if
201 * ->gp_tasks is NULL) and point ->gp_tasks at the newly added element.
202 * The task will dequeue itself when it exits the outermost enclosing
203 * RCU read-side critical section. Therefore, the current grace period
204 * cannot be permitted to complete until the ->gp_tasks pointer becomes
205 * NULL.
206 *
207 * Caller must disable preemption.
208 */
209void rcu_preempt_note_context_switch(void)
210{
211 struct task_struct *t = current;
212 unsigned long flags;
213
214 local_irq_save(flags); /* must exclude scheduler_tick(). */
215 if (rcu_preempt_running_reader() &&
216 (t->rcu_read_unlock_special & RCU_READ_UNLOCK_BLOCKED) == 0) {
217
218 /* Possibly blocking in an RCU read-side critical section. */
219 t->rcu_read_unlock_special |= RCU_READ_UNLOCK_BLOCKED;
220
221 /*
222 * If this CPU has already checked in, then this task
223 * will hold up the next grace period rather than the
224 * current grace period. Queue the task accordingly.
225 * If the task is queued for the current grace period
226 * (i.e., this CPU has not yet passed through a quiescent
227 * state for the current grace period), then as long
228 * as that task remains queued, the current grace period
229 * cannot end.
230 */
231 list_add(&t->rcu_node_entry, &rcu_preempt_ctrlblk.blkd_tasks);
232 if (rcu_cpu_blocking_cur_gp())
233 rcu_preempt_ctrlblk.gp_tasks = &t->rcu_node_entry;
234 }
235
236 /*
237 * Either we were not in an RCU read-side critical section to
238 * begin with, or we have now recorded that critical section
239 * globally. Either way, we can now note a quiescent state
240 * for this CPU. Again, if we were in an RCU read-side critical
241 * section, and if that critical section was blocking the current
242 * grace period, then the fact that the task has been enqueued
243 * means that current grace period continues to be blocked.
244 */
245 rcu_preempt_cpu_qs();
246 local_irq_restore(flags);
247}
248
249/*
250 * Tiny-preemptible RCU implementation for rcu_read_lock().
251 * Just increment ->rcu_read_lock_nesting, shared state will be updated
252 * if we block.
253 */
254void __rcu_read_lock(void)
255{
256 current->rcu_read_lock_nesting++;
257 barrier(); /* needed if we ever invoke rcu_read_lock in rcutiny.c */
258}
259EXPORT_SYMBOL_GPL(__rcu_read_lock);
260
261/*
262 * Handle special cases during rcu_read_unlock(), such as needing to
263 * notify RCU core processing or task having blocked during the RCU
264 * read-side critical section.
265 */
266static void rcu_read_unlock_special(struct task_struct *t)
267{
268 int empty;
269 int empty_exp;
270 unsigned long flags;
271 struct list_head *np;
272 int special;
273
274 /*
275 * NMI handlers cannot block and cannot safely manipulate state.
276 * They therefore cannot possibly be special, so just leave.
277 */
278 if (in_nmi())
279 return;
280
281 local_irq_save(flags);
282
283 /*
284 * If RCU core is waiting for this CPU to exit critical section,
285 * let it know that we have done so.
286 */
287 special = t->rcu_read_unlock_special;
288 if (special & RCU_READ_UNLOCK_NEED_QS)
289 rcu_preempt_cpu_qs();
290
291 /* Hardware IRQ handlers cannot block. */
292 if (in_irq()) {
293 local_irq_restore(flags);
294 return;
295 }
296
297 /* Clean up if blocked during RCU read-side critical section. */
298 if (special & RCU_READ_UNLOCK_BLOCKED) {
299 t->rcu_read_unlock_special &= ~RCU_READ_UNLOCK_BLOCKED;
300
301 /*
302 * Remove this task from the ->blkd_tasks list and adjust
303 * any pointers that might have been referencing it.
304 */
305 empty = !rcu_preempt_blocked_readers_cgp();
306 empty_exp = rcu_preempt_ctrlblk.exp_tasks == NULL;
307 np = t->rcu_node_entry.next;
308 if (np == &rcu_preempt_ctrlblk.blkd_tasks)
309 np = NULL;
310 list_del(&t->rcu_node_entry);
311 if (&t->rcu_node_entry == rcu_preempt_ctrlblk.gp_tasks)
312 rcu_preempt_ctrlblk.gp_tasks = np;
313 if (&t->rcu_node_entry == rcu_preempt_ctrlblk.exp_tasks)
314 rcu_preempt_ctrlblk.exp_tasks = np;
315 INIT_LIST_HEAD(&t->rcu_node_entry);
316
317 /*
318 * If this was the last task on the current list, and if
319 * we aren't waiting on the CPU, report the quiescent state
320 * and start a new grace period if needed.
321 */
322 if (!empty && !rcu_preempt_blocked_readers_cgp()) {
323 rcu_preempt_cpu_qs();
324 rcu_preempt_start_gp();
325 }
326
327 /*
328 * If this was the last task on the expedited lists,
329 * then we need wake up the waiting task.
330 */
331 if (!empty_exp && rcu_preempt_ctrlblk.exp_tasks == NULL)
332 rcu_report_exp_done();
333 }
334 local_irq_restore(flags);
335}
336
337/*
338 * Tiny-preemptible RCU implementation for rcu_read_unlock().
339 * Decrement ->rcu_read_lock_nesting. If the result is zero (outermost
340 * rcu_read_unlock()) and ->rcu_read_unlock_special is non-zero, then
341 * invoke rcu_read_unlock_special() to clean up after a context switch
342 * in an RCU read-side critical section and other special cases.
343 */
344void __rcu_read_unlock(void)
345{
346 struct task_struct *t = current;
347
348 barrier(); /* needed if we ever invoke rcu_read_unlock in rcutiny.c */
349 --t->rcu_read_lock_nesting;
350 barrier(); /* decrement before load of ->rcu_read_unlock_special */
351 if (t->rcu_read_lock_nesting == 0 &&
352 unlikely(ACCESS_ONCE(t->rcu_read_unlock_special)))
353 rcu_read_unlock_special(t);
354#ifdef CONFIG_PROVE_LOCKING
355 WARN_ON_ONCE(t->rcu_read_lock_nesting < 0);
356#endif /* #ifdef CONFIG_PROVE_LOCKING */
357}
358EXPORT_SYMBOL_GPL(__rcu_read_unlock);
359
360/*
361 * Check for a quiescent state from the current CPU. When a task blocks,
362 * the task is recorded in the rcu_preempt_ctrlblk structure, which is
363 * checked elsewhere. This is called from the scheduling-clock interrupt.
364 *
365 * Caller must disable hard irqs.
366 */
367static void rcu_preempt_check_callbacks(void)
368{
369 struct task_struct *t = current;
370
371 if (rcu_preempt_gp_in_progress() &&
372 (!rcu_preempt_running_reader() ||
373 !rcu_cpu_blocking_cur_gp()))
374 rcu_preempt_cpu_qs();
375 if (&rcu_preempt_ctrlblk.rcb.rcucblist !=
376 rcu_preempt_ctrlblk.rcb.donetail)
377 raise_softirq(RCU_SOFTIRQ);
378 if (rcu_preempt_gp_in_progress() &&
379 rcu_cpu_blocking_cur_gp() &&
380 rcu_preempt_running_reader())
381 t->rcu_read_unlock_special |= RCU_READ_UNLOCK_NEED_QS;
382}
383
384/*
385 * TINY_PREEMPT_RCU has an extra callback-list tail pointer to
386 * update, so this is invoked from __rcu_process_callbacks() to
387 * handle that case. Of course, it is invoked for all flavors of
388 * RCU, but RCU callbacks can appear only on one of the lists, and
389 * neither ->nexttail nor ->donetail can possibly be NULL, so there
390 * is no need for an explicit check.
391 */
392static void rcu_preempt_remove_callbacks(struct rcu_ctrlblk *rcp)
393{
394 if (rcu_preempt_ctrlblk.nexttail == rcp->donetail)
395 rcu_preempt_ctrlblk.nexttail = &rcp->rcucblist;
396}
397
398/*
399 * Process callbacks for preemptible RCU.
400 */
401static void rcu_preempt_process_callbacks(void)
402{
403 __rcu_process_callbacks(&rcu_preempt_ctrlblk.rcb);
404}
405
406/*
407 * Queue a preemptible -RCU callback for invocation after a grace period.
408 */
409void call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu))
410{
411 unsigned long flags;
412
413 debug_rcu_head_queue(head);
414 head->func = func;
415 head->next = NULL;
416
417 local_irq_save(flags);
418 *rcu_preempt_ctrlblk.nexttail = head;
419 rcu_preempt_ctrlblk.nexttail = &head->next;
420 rcu_preempt_start_gp(); /* checks to see if GP needed. */
421 local_irq_restore(flags);
422}
423EXPORT_SYMBOL_GPL(call_rcu);
424
425void rcu_barrier(void)
426{
427 struct rcu_synchronize rcu;
428
429 init_rcu_head_on_stack(&rcu.head);
430 init_completion(&rcu.completion);
431 /* Will wake me after RCU finished. */
432 call_rcu(&rcu.head, wakeme_after_rcu);
433 /* Wait for it. */
434 wait_for_completion(&rcu.completion);
435 destroy_rcu_head_on_stack(&rcu.head);
436}
437EXPORT_SYMBOL_GPL(rcu_barrier);
438
439/*
440 * synchronize_rcu - wait until a grace period has elapsed.
441 *
442 * Control will return to the caller some time after a full grace
443 * period has elapsed, in other words after all currently executing RCU
444 * read-side critical sections have completed. RCU read-side critical
445 * sections are delimited by rcu_read_lock() and rcu_read_unlock(),
446 * and may be nested.
447 */
448void synchronize_rcu(void)
449{
450#ifdef CONFIG_DEBUG_LOCK_ALLOC
451 if (!rcu_scheduler_active)
452 return;
453#endif /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */
454
455 WARN_ON_ONCE(rcu_preempt_running_reader());
456 if (!rcu_preempt_blocked_readers_any())
457 return;
458
459 /* Once we get past the fastpath checks, same code as rcu_barrier(). */
460 rcu_barrier();
461}
462EXPORT_SYMBOL_GPL(synchronize_rcu);
463
464static DECLARE_WAIT_QUEUE_HEAD(sync_rcu_preempt_exp_wq);
465static unsigned long sync_rcu_preempt_exp_count;
466static DEFINE_MUTEX(sync_rcu_preempt_exp_mutex);
467
468/*
469 * Return non-zero if there are any tasks in RCU read-side critical
470 * sections blocking the current preemptible-RCU expedited grace period.
471 * If there is no preemptible-RCU expedited grace period currently in
472 * progress, returns zero unconditionally.
473 */
474static int rcu_preempted_readers_exp(void)
475{
476 return rcu_preempt_ctrlblk.exp_tasks != NULL;
477}
478
479/*
480 * Report the exit from RCU read-side critical section for the last task
481 * that queued itself during or before the current expedited preemptible-RCU
482 * grace period.
483 */
484static void rcu_report_exp_done(void)
485{
486 wake_up(&sync_rcu_preempt_exp_wq);
487}
488
489/*
490 * Wait for an rcu-preempt grace period, but expedite it. The basic idea
491 * is to rely in the fact that there is but one CPU, and that it is
492 * illegal for a task to invoke synchronize_rcu_expedited() while in a
493 * preemptible-RCU read-side critical section. Therefore, any such
494 * critical sections must correspond to blocked tasks, which must therefore
495 * be on the ->blkd_tasks list. So just record the current head of the
496 * list in the ->exp_tasks pointer, and wait for all tasks including and
497 * after the task pointed to by ->exp_tasks to drain.
498 */
499void synchronize_rcu_expedited(void)
500{
501 unsigned long flags;
502 struct rcu_preempt_ctrlblk *rpcp = &rcu_preempt_ctrlblk;
503 unsigned long snap;
504
505 barrier(); /* ensure prior action seen before grace period. */
506
507 WARN_ON_ONCE(rcu_preempt_running_reader());
508
509 /*
510 * Acquire lock so that there is only one preemptible RCU grace
511 * period in flight. Of course, if someone does the expedited
512 * grace period for us while we are acquiring the lock, just leave.
513 */
514 snap = sync_rcu_preempt_exp_count + 1;
515 mutex_lock(&sync_rcu_preempt_exp_mutex);
516 if (ULONG_CMP_LT(snap, sync_rcu_preempt_exp_count))
517 goto unlock_mb_ret; /* Others did our work for us. */
518
519 local_irq_save(flags);
520
521 /*
522 * All RCU readers have to already be on blkd_tasks because
523 * we cannot legally be executing in an RCU read-side critical
524 * section.
525 */
526
527 /* Snapshot current head of ->blkd_tasks list. */
528 rpcp->exp_tasks = rpcp->blkd_tasks.next;
529 if (rpcp->exp_tasks == &rpcp->blkd_tasks)
530 rpcp->exp_tasks = NULL;
531 local_irq_restore(flags);
532
533 /* Wait for tail of ->blkd_tasks list to drain. */
534 if (rcu_preempted_readers_exp())
535 wait_event(sync_rcu_preempt_exp_wq,
536 !rcu_preempted_readers_exp());
537
538 /* Clean up and exit. */
539 barrier(); /* ensure expedited GP seen before counter increment. */
540 sync_rcu_preempt_exp_count++;
541unlock_mb_ret:
542 mutex_unlock(&sync_rcu_preempt_exp_mutex);
543 barrier(); /* ensure subsequent action seen after grace period. */
544}
545EXPORT_SYMBOL_GPL(synchronize_rcu_expedited);
546
547/*
548 * Does preemptible RCU need the CPU to stay out of dynticks mode?
549 */
550int rcu_preempt_needs_cpu(void)
551{
552 if (!rcu_preempt_running_reader())
553 rcu_preempt_cpu_qs();
554 return rcu_preempt_ctrlblk.rcb.rcucblist != NULL;
555}
556
557/*
558 * Check for a task exiting while in a preemptible -RCU read-side
559 * critical section, clean up if so. No need to issue warnings,
560 * as debug_check_no_locks_held() already does this if lockdep
561 * is enabled.
562 */
563void exit_rcu(void)
564{
565 struct task_struct *t = current;
566
567 if (t->rcu_read_lock_nesting == 0)
568 return;
569 t->rcu_read_lock_nesting = 1;
570 rcu_read_unlock();
571}
572
573#else /* #ifdef CONFIG_TINY_PREEMPT_RCU */
574
575/*
576 * Because preemptible RCU does not exist, it never has any callbacks
577 * to check.
578 */
579static void rcu_preempt_check_callbacks(void)
580{
581}
582
583/*
584 * Because preemptible RCU does not exist, it never has any callbacks
585 * to remove.
586 */
587static void rcu_preempt_remove_callbacks(struct rcu_ctrlblk *rcp)
588{
589}
590
591/*
592 * Because preemptible RCU does not exist, it never has any callbacks
593 * to process.
594 */
595static void rcu_preempt_process_callbacks(void)
596{
597}
598
599#endif /* #else #ifdef CONFIG_TINY_PREEMPT_RCU */
600
25#ifdef CONFIG_DEBUG_LOCK_ALLOC 601#ifdef CONFIG_DEBUG_LOCK_ALLOC
26 602
27#include <linux/kernel_stat.h> 603#include <linux/kernel_stat.h>
diff --git a/kernel/rcutorture.c b/kernel/rcutorture.c
index 2e2726d790b9..9d8e8fb2515f 100644
--- a/kernel/rcutorture.c
+++ b/kernel/rcutorture.c
@@ -120,7 +120,7 @@ struct rcu_torture {
120}; 120};
121 121
122static LIST_HEAD(rcu_torture_freelist); 122static LIST_HEAD(rcu_torture_freelist);
123static struct rcu_torture *rcu_torture_current; 123static struct rcu_torture __rcu *rcu_torture_current;
124static long rcu_torture_current_version; 124static long rcu_torture_current_version;
125static struct rcu_torture rcu_tortures[10 * RCU_TORTURE_PIPE_LEN]; 125static struct rcu_torture rcu_tortures[10 * RCU_TORTURE_PIPE_LEN];
126static DEFINE_SPINLOCK(rcu_torture_lock); 126static DEFINE_SPINLOCK(rcu_torture_lock);
@@ -153,8 +153,10 @@ int rcutorture_runnable = RCUTORTURE_RUNNABLE_INIT;
153#define FULLSTOP_SHUTDOWN 1 /* System shutdown with rcutorture running. */ 153#define FULLSTOP_SHUTDOWN 1 /* System shutdown with rcutorture running. */
154#define FULLSTOP_RMMOD 2 /* Normal rmmod of rcutorture. */ 154#define FULLSTOP_RMMOD 2 /* Normal rmmod of rcutorture. */
155static int fullstop = FULLSTOP_RMMOD; 155static int fullstop = FULLSTOP_RMMOD;
156DEFINE_MUTEX(fullstop_mutex); /* Protect fullstop transitions and spawning */ 156/*
157 /* of kthreads. */ 157 * Protect fullstop transitions and spawning of kthreads.
158 */
159static DEFINE_MUTEX(fullstop_mutex);
158 160
159/* 161/*
160 * Detect and respond to a system shutdown. 162 * Detect and respond to a system shutdown.
@@ -303,6 +305,10 @@ static void rcu_read_delay(struct rcu_random_state *rrsp)
303 mdelay(longdelay_ms); 305 mdelay(longdelay_ms);
304 if (!(rcu_random(rrsp) % (nrealreaders * 2 * shortdelay_us))) 306 if (!(rcu_random(rrsp) % (nrealreaders * 2 * shortdelay_us)))
305 udelay(shortdelay_us); 307 udelay(shortdelay_us);
308#ifdef CONFIG_PREEMPT
309 if (!preempt_count() && !(rcu_random(rrsp) % (nrealreaders * 20000)))
310 preempt_schedule(); /* No QS if preempt_disable() in effect */
311#endif
306} 312}
307 313
308static void rcu_torture_read_unlock(int idx) __releases(RCU) 314static void rcu_torture_read_unlock(int idx) __releases(RCU)
@@ -536,6 +542,8 @@ static void srcu_read_delay(struct rcu_random_state *rrsp)
536 delay = rcu_random(rrsp) % (nrealreaders * 2 * longdelay * uspertick); 542 delay = rcu_random(rrsp) % (nrealreaders * 2 * longdelay * uspertick);
537 if (!delay) 543 if (!delay)
538 schedule_timeout_interruptible(longdelay); 544 schedule_timeout_interruptible(longdelay);
545 else
546 rcu_read_delay(rrsp);
539} 547}
540 548
541static void srcu_torture_read_unlock(int idx) __releases(&srcu_ctl) 549static void srcu_torture_read_unlock(int idx) __releases(&srcu_ctl)
@@ -731,7 +739,8 @@ rcu_torture_writer(void *arg)
731 continue; 739 continue;
732 rp->rtort_pipe_count = 0; 740 rp->rtort_pipe_count = 0;
733 udelay(rcu_random(&rand) & 0x3ff); 741 udelay(rcu_random(&rand) & 0x3ff);
734 old_rp = rcu_torture_current; 742 old_rp = rcu_dereference_check(rcu_torture_current,
743 current == writer_task);
735 rp->rtort_mbtest = 1; 744 rp->rtort_mbtest = 1;
736 rcu_assign_pointer(rcu_torture_current, rp); 745 rcu_assign_pointer(rcu_torture_current, rp);
737 smp_wmb(); /* Mods to old_rp must follow rcu_assign_pointer() */ 746 smp_wmb(); /* Mods to old_rp must follow rcu_assign_pointer() */
diff --git a/kernel/rcutree.c b/kernel/rcutree.c
index d5bc43976c5a..ccdc04c47981 100644
--- a/kernel/rcutree.c
+++ b/kernel/rcutree.c
@@ -143,6 +143,11 @@ module_param(blimit, int, 0);
143module_param(qhimark, int, 0); 143module_param(qhimark, int, 0);
144module_param(qlowmark, int, 0); 144module_param(qlowmark, int, 0);
145 145
146#ifdef CONFIG_RCU_CPU_STALL_DETECTOR
147int rcu_cpu_stall_suppress __read_mostly = RCU_CPU_STALL_SUPPRESS_INIT;
148module_param(rcu_cpu_stall_suppress, int, 0644);
149#endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
150
146static void force_quiescent_state(struct rcu_state *rsp, int relaxed); 151static void force_quiescent_state(struct rcu_state *rsp, int relaxed);
147static int rcu_pending(int cpu); 152static int rcu_pending(int cpu);
148 153
@@ -450,7 +455,7 @@ static int rcu_implicit_dynticks_qs(struct rcu_data *rdp)
450 455
451#ifdef CONFIG_RCU_CPU_STALL_DETECTOR 456#ifdef CONFIG_RCU_CPU_STALL_DETECTOR
452 457
453int rcu_cpu_stall_panicking __read_mostly; 458int rcu_cpu_stall_suppress __read_mostly;
454 459
455static void record_gp_stall_check_time(struct rcu_state *rsp) 460static void record_gp_stall_check_time(struct rcu_state *rsp)
456{ 461{
@@ -482,8 +487,11 @@ static void print_other_cpu_stall(struct rcu_state *rsp)
482 rcu_print_task_stall(rnp); 487 rcu_print_task_stall(rnp);
483 raw_spin_unlock_irqrestore(&rnp->lock, flags); 488 raw_spin_unlock_irqrestore(&rnp->lock, flags);
484 489
485 /* OK, time to rat on our buddy... */ 490 /*
486 491 * OK, time to rat on our buddy...
492 * See Documentation/RCU/stallwarn.txt for info on how to debug
493 * RCU CPU stall warnings.
494 */
487 printk(KERN_ERR "INFO: %s detected stalls on CPUs/tasks: {", 495 printk(KERN_ERR "INFO: %s detected stalls on CPUs/tasks: {",
488 rsp->name); 496 rsp->name);
489 rcu_for_each_leaf_node(rsp, rnp) { 497 rcu_for_each_leaf_node(rsp, rnp) {
@@ -512,6 +520,11 @@ static void print_cpu_stall(struct rcu_state *rsp)
512 unsigned long flags; 520 unsigned long flags;
513 struct rcu_node *rnp = rcu_get_root(rsp); 521 struct rcu_node *rnp = rcu_get_root(rsp);
514 522
523 /*
524 * OK, time to rat on ourselves...
525 * See Documentation/RCU/stallwarn.txt for info on how to debug
526 * RCU CPU stall warnings.
527 */
515 printk(KERN_ERR "INFO: %s detected stall on CPU %d (t=%lu jiffies)\n", 528 printk(KERN_ERR "INFO: %s detected stall on CPU %d (t=%lu jiffies)\n",
516 rsp->name, smp_processor_id(), jiffies - rsp->gp_start); 529 rsp->name, smp_processor_id(), jiffies - rsp->gp_start);
517 trigger_all_cpu_backtrace(); 530 trigger_all_cpu_backtrace();
@@ -530,11 +543,11 @@ static void check_cpu_stall(struct rcu_state *rsp, struct rcu_data *rdp)
530 long delta; 543 long delta;
531 struct rcu_node *rnp; 544 struct rcu_node *rnp;
532 545
533 if (rcu_cpu_stall_panicking) 546 if (rcu_cpu_stall_suppress)
534 return; 547 return;
535 delta = jiffies - rsp->jiffies_stall; 548 delta = jiffies - ACCESS_ONCE(rsp->jiffies_stall);
536 rnp = rdp->mynode; 549 rnp = rdp->mynode;
537 if ((rnp->qsmask & rdp->grpmask) && delta >= 0) { 550 if ((ACCESS_ONCE(rnp->qsmask) & rdp->grpmask) && delta >= 0) {
538 551
539 /* We haven't checked in, so go dump stack. */ 552 /* We haven't checked in, so go dump stack. */
540 print_cpu_stall(rsp); 553 print_cpu_stall(rsp);
@@ -548,10 +561,26 @@ static void check_cpu_stall(struct rcu_state *rsp, struct rcu_data *rdp)
548 561
549static int rcu_panic(struct notifier_block *this, unsigned long ev, void *ptr) 562static int rcu_panic(struct notifier_block *this, unsigned long ev, void *ptr)
550{ 563{
551 rcu_cpu_stall_panicking = 1; 564 rcu_cpu_stall_suppress = 1;
552 return NOTIFY_DONE; 565 return NOTIFY_DONE;
553} 566}
554 567
568/**
569 * rcu_cpu_stall_reset - prevent further stall warnings in current grace period
570 *
571 * Set the stall-warning timeout way off into the future, thus preventing
572 * any RCU CPU stall-warning messages from appearing in the current set of
573 * RCU grace periods.
574 *
575 * The caller must disable hard irqs.
576 */
577void rcu_cpu_stall_reset(void)
578{
579 rcu_sched_state.jiffies_stall = jiffies + ULONG_MAX / 2;
580 rcu_bh_state.jiffies_stall = jiffies + ULONG_MAX / 2;
581 rcu_preempt_stall_reset();
582}
583
555static struct notifier_block rcu_panic_block = { 584static struct notifier_block rcu_panic_block = {
556 .notifier_call = rcu_panic, 585 .notifier_call = rcu_panic,
557}; 586};
@@ -571,6 +600,10 @@ static void check_cpu_stall(struct rcu_state *rsp, struct rcu_data *rdp)
571{ 600{
572} 601}
573 602
603void rcu_cpu_stall_reset(void)
604{
605}
606
574static void __init check_cpu_stall_init(void) 607static void __init check_cpu_stall_init(void)
575{ 608{
576} 609}
@@ -712,7 +745,7 @@ static void
712rcu_start_gp(struct rcu_state *rsp, unsigned long flags) 745rcu_start_gp(struct rcu_state *rsp, unsigned long flags)
713 __releases(rcu_get_root(rsp)->lock) 746 __releases(rcu_get_root(rsp)->lock)
714{ 747{
715 struct rcu_data *rdp = rsp->rda[smp_processor_id()]; 748 struct rcu_data *rdp = this_cpu_ptr(rsp->rda);
716 struct rcu_node *rnp = rcu_get_root(rsp); 749 struct rcu_node *rnp = rcu_get_root(rsp);
717 750
718 if (!cpu_needs_another_gp(rsp, rdp) || rsp->fqs_active) { 751 if (!cpu_needs_another_gp(rsp, rdp) || rsp->fqs_active) {
@@ -960,7 +993,7 @@ rcu_check_quiescent_state(struct rcu_state *rsp, struct rcu_data *rdp)
960static void rcu_send_cbs_to_orphanage(struct rcu_state *rsp) 993static void rcu_send_cbs_to_orphanage(struct rcu_state *rsp)
961{ 994{
962 int i; 995 int i;
963 struct rcu_data *rdp = rsp->rda[smp_processor_id()]; 996 struct rcu_data *rdp = this_cpu_ptr(rsp->rda);
964 997
965 if (rdp->nxtlist == NULL) 998 if (rdp->nxtlist == NULL)
966 return; /* irqs disabled, so comparison is stable. */ 999 return; /* irqs disabled, so comparison is stable. */
@@ -971,6 +1004,7 @@ static void rcu_send_cbs_to_orphanage(struct rcu_state *rsp)
971 for (i = 0; i < RCU_NEXT_SIZE; i++) 1004 for (i = 0; i < RCU_NEXT_SIZE; i++)
972 rdp->nxttail[i] = &rdp->nxtlist; 1005 rdp->nxttail[i] = &rdp->nxtlist;
973 rsp->orphan_qlen += rdp->qlen; 1006 rsp->orphan_qlen += rdp->qlen;
1007 rdp->n_cbs_orphaned += rdp->qlen;
974 rdp->qlen = 0; 1008 rdp->qlen = 0;
975 raw_spin_unlock(&rsp->onofflock); /* irqs remain disabled. */ 1009 raw_spin_unlock(&rsp->onofflock); /* irqs remain disabled. */
976} 1010}
@@ -984,7 +1018,7 @@ static void rcu_adopt_orphan_cbs(struct rcu_state *rsp)
984 struct rcu_data *rdp; 1018 struct rcu_data *rdp;
985 1019
986 raw_spin_lock_irqsave(&rsp->onofflock, flags); 1020 raw_spin_lock_irqsave(&rsp->onofflock, flags);
987 rdp = rsp->rda[smp_processor_id()]; 1021 rdp = this_cpu_ptr(rsp->rda);
988 if (rsp->orphan_cbs_list == NULL) { 1022 if (rsp->orphan_cbs_list == NULL) {
989 raw_spin_unlock_irqrestore(&rsp->onofflock, flags); 1023 raw_spin_unlock_irqrestore(&rsp->onofflock, flags);
990 return; 1024 return;
@@ -992,6 +1026,7 @@ static void rcu_adopt_orphan_cbs(struct rcu_state *rsp)
992 *rdp->nxttail[RCU_NEXT_TAIL] = rsp->orphan_cbs_list; 1026 *rdp->nxttail[RCU_NEXT_TAIL] = rsp->orphan_cbs_list;
993 rdp->nxttail[RCU_NEXT_TAIL] = rsp->orphan_cbs_tail; 1027 rdp->nxttail[RCU_NEXT_TAIL] = rsp->orphan_cbs_tail;
994 rdp->qlen += rsp->orphan_qlen; 1028 rdp->qlen += rsp->orphan_qlen;
1029 rdp->n_cbs_adopted += rsp->orphan_qlen;
995 rsp->orphan_cbs_list = NULL; 1030 rsp->orphan_cbs_list = NULL;
996 rsp->orphan_cbs_tail = &rsp->orphan_cbs_list; 1031 rsp->orphan_cbs_tail = &rsp->orphan_cbs_list;
997 rsp->orphan_qlen = 0; 1032 rsp->orphan_qlen = 0;
@@ -1007,7 +1042,7 @@ static void __rcu_offline_cpu(int cpu, struct rcu_state *rsp)
1007 unsigned long flags; 1042 unsigned long flags;
1008 unsigned long mask; 1043 unsigned long mask;
1009 int need_report = 0; 1044 int need_report = 0;
1010 struct rcu_data *rdp = rsp->rda[cpu]; 1045 struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu);
1011 struct rcu_node *rnp; 1046 struct rcu_node *rnp;
1012 1047
1013 /* Exclude any attempts to start a new grace period. */ 1048 /* Exclude any attempts to start a new grace period. */
@@ -1123,6 +1158,7 @@ static void rcu_do_batch(struct rcu_state *rsp, struct rcu_data *rdp)
1123 1158
1124 /* Update count, and requeue any remaining callbacks. */ 1159 /* Update count, and requeue any remaining callbacks. */
1125 rdp->qlen -= count; 1160 rdp->qlen -= count;
1161 rdp->n_cbs_invoked += count;
1126 if (list != NULL) { 1162 if (list != NULL) {
1127 *tail = rdp->nxtlist; 1163 *tail = rdp->nxtlist;
1128 rdp->nxtlist = list; 1164 rdp->nxtlist = list;
@@ -1226,7 +1262,8 @@ static void force_qs_rnp(struct rcu_state *rsp, int (*f)(struct rcu_data *))
1226 cpu = rnp->grplo; 1262 cpu = rnp->grplo;
1227 bit = 1; 1263 bit = 1;
1228 for (; cpu <= rnp->grphi; cpu++, bit <<= 1) { 1264 for (; cpu <= rnp->grphi; cpu++, bit <<= 1) {
1229 if ((rnp->qsmask & bit) != 0 && f(rsp->rda[cpu])) 1265 if ((rnp->qsmask & bit) != 0 &&
1266 f(per_cpu_ptr(rsp->rda, cpu)))
1230 mask |= bit; 1267 mask |= bit;
1231 } 1268 }
1232 if (mask != 0) { 1269 if (mask != 0) {
@@ -1402,7 +1439,7 @@ __call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu),
1402 * a quiescent state betweentimes. 1439 * a quiescent state betweentimes.
1403 */ 1440 */
1404 local_irq_save(flags); 1441 local_irq_save(flags);
1405 rdp = rsp->rda[smp_processor_id()]; 1442 rdp = this_cpu_ptr(rsp->rda);
1406 rcu_process_gp_end(rsp, rdp); 1443 rcu_process_gp_end(rsp, rdp);
1407 check_for_new_grace_period(rsp, rdp); 1444 check_for_new_grace_period(rsp, rdp);
1408 1445
@@ -1701,7 +1738,7 @@ rcu_boot_init_percpu_data(int cpu, struct rcu_state *rsp)
1701{ 1738{
1702 unsigned long flags; 1739 unsigned long flags;
1703 int i; 1740 int i;
1704 struct rcu_data *rdp = rsp->rda[cpu]; 1741 struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu);
1705 struct rcu_node *rnp = rcu_get_root(rsp); 1742 struct rcu_node *rnp = rcu_get_root(rsp);
1706 1743
1707 /* Set up local state, ensuring consistent view of global state. */ 1744 /* Set up local state, ensuring consistent view of global state. */
@@ -1729,7 +1766,7 @@ rcu_init_percpu_data(int cpu, struct rcu_state *rsp, int preemptable)
1729{ 1766{
1730 unsigned long flags; 1767 unsigned long flags;
1731 unsigned long mask; 1768 unsigned long mask;
1732 struct rcu_data *rdp = rsp->rda[cpu]; 1769 struct rcu_data *rdp = per_cpu_ptr(rsp->rda, cpu);
1733 struct rcu_node *rnp = rcu_get_root(rsp); 1770 struct rcu_node *rnp = rcu_get_root(rsp);
1734 1771
1735 /* Set up local state, ensuring consistent view of global state. */ 1772 /* Set up local state, ensuring consistent view of global state. */
@@ -1865,7 +1902,8 @@ static void __init rcu_init_levelspread(struct rcu_state *rsp)
1865/* 1902/*
1866 * Helper function for rcu_init() that initializes one rcu_state structure. 1903 * Helper function for rcu_init() that initializes one rcu_state structure.
1867 */ 1904 */
1868static void __init rcu_init_one(struct rcu_state *rsp) 1905static void __init rcu_init_one(struct rcu_state *rsp,
1906 struct rcu_data __percpu *rda)
1869{ 1907{
1870 static char *buf[] = { "rcu_node_level_0", 1908 static char *buf[] = { "rcu_node_level_0",
1871 "rcu_node_level_1", 1909 "rcu_node_level_1",
@@ -1918,37 +1956,23 @@ static void __init rcu_init_one(struct rcu_state *rsp)
1918 } 1956 }
1919 } 1957 }
1920 1958
1959 rsp->rda = rda;
1921 rnp = rsp->level[NUM_RCU_LVLS - 1]; 1960 rnp = rsp->level[NUM_RCU_LVLS - 1];
1922 for_each_possible_cpu(i) { 1961 for_each_possible_cpu(i) {
1923 while (i > rnp->grphi) 1962 while (i > rnp->grphi)
1924 rnp++; 1963 rnp++;
1925 rsp->rda[i]->mynode = rnp; 1964 per_cpu_ptr(rsp->rda, i)->mynode = rnp;
1926 rcu_boot_init_percpu_data(i, rsp); 1965 rcu_boot_init_percpu_data(i, rsp);
1927 } 1966 }
1928} 1967}
1929 1968
1930/*
1931 * Helper macro for __rcu_init() and __rcu_init_preempt(). To be used
1932 * nowhere else! Assigns leaf node pointers into each CPU's rcu_data
1933 * structure.
1934 */
1935#define RCU_INIT_FLAVOR(rsp, rcu_data) \
1936do { \
1937 int i; \
1938 \
1939 for_each_possible_cpu(i) { \
1940 (rsp)->rda[i] = &per_cpu(rcu_data, i); \
1941 } \
1942 rcu_init_one(rsp); \
1943} while (0)
1944
1945void __init rcu_init(void) 1969void __init rcu_init(void)
1946{ 1970{
1947 int cpu; 1971 int cpu;
1948 1972
1949 rcu_bootup_announce(); 1973 rcu_bootup_announce();
1950 RCU_INIT_FLAVOR(&rcu_sched_state, rcu_sched_data); 1974 rcu_init_one(&rcu_sched_state, &rcu_sched_data);
1951 RCU_INIT_FLAVOR(&rcu_bh_state, rcu_bh_data); 1975 rcu_init_one(&rcu_bh_state, &rcu_bh_data);
1952 __rcu_init_preempt(); 1976 __rcu_init_preempt();
1953 open_softirq(RCU_SOFTIRQ, rcu_process_callbacks); 1977 open_softirq(RCU_SOFTIRQ, rcu_process_callbacks);
1954 1978
diff --git a/kernel/rcutree.h b/kernel/rcutree.h
index 14c040b18ed0..91d4170c5c13 100644
--- a/kernel/rcutree.h
+++ b/kernel/rcutree.h
@@ -202,6 +202,9 @@ struct rcu_data {
202 long qlen; /* # of queued callbacks */ 202 long qlen; /* # of queued callbacks */
203 long qlen_last_fqs_check; 203 long qlen_last_fqs_check;
204 /* qlen at last check for QS forcing */ 204 /* qlen at last check for QS forcing */
205 unsigned long n_cbs_invoked; /* count of RCU cbs invoked. */
206 unsigned long n_cbs_orphaned; /* RCU cbs sent to orphanage. */
207 unsigned long n_cbs_adopted; /* RCU cbs adopted from orphanage. */
205 unsigned long n_force_qs_snap; 208 unsigned long n_force_qs_snap;
206 /* did other CPU force QS recently? */ 209 /* did other CPU force QS recently? */
207 long blimit; /* Upper limit on a processed batch */ 210 long blimit; /* Upper limit on a processed batch */
@@ -254,19 +257,23 @@ struct rcu_data {
254#define RCU_STALL_DELAY_DELTA 0 257#define RCU_STALL_DELAY_DELTA 0
255#endif 258#endif
256 259
257#define RCU_SECONDS_TILL_STALL_CHECK (10 * HZ + RCU_STALL_DELAY_DELTA) 260#define RCU_SECONDS_TILL_STALL_CHECK (CONFIG_RCU_CPU_STALL_TIMEOUT * HZ + \
261 RCU_STALL_DELAY_DELTA)
258 /* for rsp->jiffies_stall */ 262 /* for rsp->jiffies_stall */
259#define RCU_SECONDS_TILL_STALL_RECHECK (30 * HZ + RCU_STALL_DELAY_DELTA) 263#define RCU_SECONDS_TILL_STALL_RECHECK (3 * RCU_SECONDS_TILL_STALL_CHECK + 30)
260 /* for rsp->jiffies_stall */ 264 /* for rsp->jiffies_stall */
261#define RCU_STALL_RAT_DELAY 2 /* Allow other CPUs time */ 265#define RCU_STALL_RAT_DELAY 2 /* Allow other CPUs time */
262 /* to take at least one */ 266 /* to take at least one */
263 /* scheduling clock irq */ 267 /* scheduling clock irq */
264 /* before ratting on them. */ 268 /* before ratting on them. */
265 269
266#endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */ 270#ifdef CONFIG_RCU_CPU_STALL_DETECTOR_RUNNABLE
271#define RCU_CPU_STALL_SUPPRESS_INIT 0
272#else
273#define RCU_CPU_STALL_SUPPRESS_INIT 1
274#endif
267 275
268#define ULONG_CMP_GE(a, b) (ULONG_MAX / 2 >= (a) - (b)) 276#endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
269#define ULONG_CMP_LT(a, b) (ULONG_MAX / 2 < (a) - (b))
270 277
271/* 278/*
272 * RCU global state, including node hierarchy. This hierarchy is 279 * RCU global state, including node hierarchy. This hierarchy is
@@ -283,7 +290,7 @@ struct rcu_state {
283 struct rcu_node *level[NUM_RCU_LVLS]; /* Hierarchy levels. */ 290 struct rcu_node *level[NUM_RCU_LVLS]; /* Hierarchy levels. */
284 u32 levelcnt[MAX_RCU_LVLS + 1]; /* # nodes in each level. */ 291 u32 levelcnt[MAX_RCU_LVLS + 1]; /* # nodes in each level. */
285 u8 levelspread[NUM_RCU_LVLS]; /* kids/node in each level. */ 292 u8 levelspread[NUM_RCU_LVLS]; /* kids/node in each level. */
286 struct rcu_data *rda[NR_CPUS]; /* array of rdp pointers. */ 293 struct rcu_data __percpu *rda; /* pointer of percu rcu_data. */
287 294
288 /* The following fields are guarded by the root rcu_node's lock. */ 295 /* The following fields are guarded by the root rcu_node's lock. */
289 296
@@ -365,6 +372,7 @@ static void rcu_report_unblock_qs_rnp(struct rcu_node *rnp,
365#ifdef CONFIG_RCU_CPU_STALL_DETECTOR 372#ifdef CONFIG_RCU_CPU_STALL_DETECTOR
366static void rcu_print_detail_task_stall(struct rcu_state *rsp); 373static void rcu_print_detail_task_stall(struct rcu_state *rsp);
367static void rcu_print_task_stall(struct rcu_node *rnp); 374static void rcu_print_task_stall(struct rcu_node *rnp);
375static void rcu_preempt_stall_reset(void);
368#endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */ 376#endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
369static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp); 377static void rcu_preempt_check_blocked_tasks(struct rcu_node *rnp);
370#ifdef CONFIG_HOTPLUG_CPU 378#ifdef CONFIG_HOTPLUG_CPU
diff --git a/kernel/rcutree_plugin.h b/kernel/rcutree_plugin.h
index 0e4f420245d9..71a4147473f9 100644
--- a/kernel/rcutree_plugin.h
+++ b/kernel/rcutree_plugin.h
@@ -57,7 +57,7 @@ static void __init rcu_bootup_announce_oddness(void)
57 printk(KERN_INFO 57 printk(KERN_INFO
58 "\tRCU-based detection of stalled CPUs is disabled.\n"); 58 "\tRCU-based detection of stalled CPUs is disabled.\n");
59#endif 59#endif
60#ifndef CONFIG_RCU_CPU_STALL_VERBOSE 60#if defined(CONFIG_TREE_PREEMPT_RCU) && !defined(CONFIG_RCU_CPU_STALL_VERBOSE)
61 printk(KERN_INFO "\tVerbose stalled-CPUs detection is disabled.\n"); 61 printk(KERN_INFO "\tVerbose stalled-CPUs detection is disabled.\n");
62#endif 62#endif
63#if NUM_RCU_LVL_4 != 0 63#if NUM_RCU_LVL_4 != 0
@@ -154,7 +154,7 @@ static void rcu_preempt_note_context_switch(int cpu)
154 (t->rcu_read_unlock_special & RCU_READ_UNLOCK_BLOCKED) == 0) { 154 (t->rcu_read_unlock_special & RCU_READ_UNLOCK_BLOCKED) == 0) {
155 155
156 /* Possibly blocking in an RCU read-side critical section. */ 156 /* Possibly blocking in an RCU read-side critical section. */
157 rdp = rcu_preempt_state.rda[cpu]; 157 rdp = per_cpu_ptr(rcu_preempt_state.rda, cpu);
158 rnp = rdp->mynode; 158 rnp = rdp->mynode;
159 raw_spin_lock_irqsave(&rnp->lock, flags); 159 raw_spin_lock_irqsave(&rnp->lock, flags);
160 t->rcu_read_unlock_special |= RCU_READ_UNLOCK_BLOCKED; 160 t->rcu_read_unlock_special |= RCU_READ_UNLOCK_BLOCKED;
@@ -201,7 +201,7 @@ static void rcu_preempt_note_context_switch(int cpu)
201 */ 201 */
202void __rcu_read_lock(void) 202void __rcu_read_lock(void)
203{ 203{
204 ACCESS_ONCE(current->rcu_read_lock_nesting)++; 204 current->rcu_read_lock_nesting++;
205 barrier(); /* needed if we ever invoke rcu_read_lock in rcutree.c */ 205 barrier(); /* needed if we ever invoke rcu_read_lock in rcutree.c */
206} 206}
207EXPORT_SYMBOL_GPL(__rcu_read_lock); 207EXPORT_SYMBOL_GPL(__rcu_read_lock);
@@ -344,7 +344,9 @@ void __rcu_read_unlock(void)
344 struct task_struct *t = current; 344 struct task_struct *t = current;
345 345
346 barrier(); /* needed if we ever invoke rcu_read_unlock in rcutree.c */ 346 barrier(); /* needed if we ever invoke rcu_read_unlock in rcutree.c */
347 if (--ACCESS_ONCE(t->rcu_read_lock_nesting) == 0 && 347 --t->rcu_read_lock_nesting;
348 barrier(); /* decrement before load of ->rcu_read_unlock_special */
349 if (t->rcu_read_lock_nesting == 0 &&
348 unlikely(ACCESS_ONCE(t->rcu_read_unlock_special))) 350 unlikely(ACCESS_ONCE(t->rcu_read_unlock_special)))
349 rcu_read_unlock_special(t); 351 rcu_read_unlock_special(t);
350#ifdef CONFIG_PROVE_LOCKING 352#ifdef CONFIG_PROVE_LOCKING
@@ -417,6 +419,16 @@ static void rcu_print_task_stall(struct rcu_node *rnp)
417 } 419 }
418} 420}
419 421
422/*
423 * Suppress preemptible RCU's CPU stall warnings by pushing the
424 * time of the next stall-warning message comfortably far into the
425 * future.
426 */
427static void rcu_preempt_stall_reset(void)
428{
429 rcu_preempt_state.jiffies_stall = jiffies + ULONG_MAX / 2;
430}
431
420#endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */ 432#endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
421 433
422/* 434/*
@@ -546,9 +558,11 @@ EXPORT_SYMBOL_GPL(call_rcu);
546 * 558 *
547 * Control will return to the caller some time after a full grace 559 * Control will return to the caller some time after a full grace
548 * period has elapsed, in other words after all currently executing RCU 560 * period has elapsed, in other words after all currently executing RCU
549 * read-side critical sections have completed. RCU read-side critical 561 * read-side critical sections have completed. Note, however, that
550 * sections are delimited by rcu_read_lock() and rcu_read_unlock(), 562 * upon return from synchronize_rcu(), the caller might well be executing
551 * and may be nested. 563 * concurrently with new RCU read-side critical sections that began while
564 * synchronize_rcu() was waiting. RCU read-side critical sections are
565 * delimited by rcu_read_lock() and rcu_read_unlock(), and may be nested.
552 */ 566 */
553void synchronize_rcu(void) 567void synchronize_rcu(void)
554{ 568{
@@ -771,7 +785,7 @@ static void rcu_preempt_send_cbs_to_orphanage(void)
771 */ 785 */
772static void __init __rcu_init_preempt(void) 786static void __init __rcu_init_preempt(void)
773{ 787{
774 RCU_INIT_FLAVOR(&rcu_preempt_state, rcu_preempt_data); 788 rcu_init_one(&rcu_preempt_state, &rcu_preempt_data);
775} 789}
776 790
777/* 791/*
@@ -865,6 +879,14 @@ static void rcu_print_task_stall(struct rcu_node *rnp)
865{ 879{
866} 880}
867 881
882/*
883 * Because preemptible RCU does not exist, there is no need to suppress
884 * its CPU stall warnings.
885 */
886static void rcu_preempt_stall_reset(void)
887{
888}
889
868#endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */ 890#endif /* #ifdef CONFIG_RCU_CPU_STALL_DETECTOR */
869 891
870/* 892/*
@@ -919,15 +941,6 @@ static void rcu_preempt_process_callbacks(void)
919} 941}
920 942
921/* 943/*
922 * In classic RCU, call_rcu() is just call_rcu_sched().
923 */
924void call_rcu(struct rcu_head *head, void (*func)(struct rcu_head *rcu))
925{
926 call_rcu_sched(head, func);
927}
928EXPORT_SYMBOL_GPL(call_rcu);
929
930/*
931 * Wait for an rcu-preempt grace period, but make it happen quickly. 944 * Wait for an rcu-preempt grace period, but make it happen quickly.
932 * But because preemptable RCU does not exist, map to rcu-sched. 945 * But because preemptable RCU does not exist, map to rcu-sched.
933 */ 946 */
diff --git a/kernel/rcutree_trace.c b/kernel/rcutree_trace.c
index 36c95b45738e..d15430b9d122 100644
--- a/kernel/rcutree_trace.c
+++ b/kernel/rcutree_trace.c
@@ -64,7 +64,9 @@ static void print_one_rcu_data(struct seq_file *m, struct rcu_data *rdp)
64 rdp->dynticks_fqs); 64 rdp->dynticks_fqs);
65#endif /* #ifdef CONFIG_NO_HZ */ 65#endif /* #ifdef CONFIG_NO_HZ */
66 seq_printf(m, " of=%lu ri=%lu", rdp->offline_fqs, rdp->resched_ipi); 66 seq_printf(m, " of=%lu ri=%lu", rdp->offline_fqs, rdp->resched_ipi);
67 seq_printf(m, " ql=%ld b=%ld\n", rdp->qlen, rdp->blimit); 67 seq_printf(m, " ql=%ld b=%ld", rdp->qlen, rdp->blimit);
68 seq_printf(m, " ci=%lu co=%lu ca=%lu\n",
69 rdp->n_cbs_invoked, rdp->n_cbs_orphaned, rdp->n_cbs_adopted);
68} 70}
69 71
70#define PRINT_RCU_DATA(name, func, m) \ 72#define PRINT_RCU_DATA(name, func, m) \
@@ -119,7 +121,9 @@ static void print_one_rcu_data_csv(struct seq_file *m, struct rcu_data *rdp)
119 rdp->dynticks_fqs); 121 rdp->dynticks_fqs);
120#endif /* #ifdef CONFIG_NO_HZ */ 122#endif /* #ifdef CONFIG_NO_HZ */
121 seq_printf(m, ",%lu,%lu", rdp->offline_fqs, rdp->resched_ipi); 123 seq_printf(m, ",%lu,%lu", rdp->offline_fqs, rdp->resched_ipi);
122 seq_printf(m, ",%ld,%ld\n", rdp->qlen, rdp->blimit); 124 seq_printf(m, ",%ld,%ld", rdp->qlen, rdp->blimit);
125 seq_printf(m, ",%lu,%lu,%lu\n",
126 rdp->n_cbs_invoked, rdp->n_cbs_orphaned, rdp->n_cbs_adopted);
123} 127}
124 128
125static int show_rcudata_csv(struct seq_file *m, void *unused) 129static int show_rcudata_csv(struct seq_file *m, void *unused)
@@ -128,7 +132,7 @@ static int show_rcudata_csv(struct seq_file *m, void *unused)
128#ifdef CONFIG_NO_HZ 132#ifdef CONFIG_NO_HZ
129 seq_puts(m, "\"dt\",\"dt nesting\",\"dn\",\"df\","); 133 seq_puts(m, "\"dt\",\"dt nesting\",\"dn\",\"df\",");
130#endif /* #ifdef CONFIG_NO_HZ */ 134#endif /* #ifdef CONFIG_NO_HZ */
131 seq_puts(m, "\"of\",\"ri\",\"ql\",\"b\"\n"); 135 seq_puts(m, "\"of\",\"ri\",\"ql\",\"b\",\"ci\",\"co\",\"ca\"\n");
132#ifdef CONFIG_TREE_PREEMPT_RCU 136#ifdef CONFIG_TREE_PREEMPT_RCU
133 seq_puts(m, "\"rcu_preempt:\"\n"); 137 seq_puts(m, "\"rcu_preempt:\"\n");
134 PRINT_RCU_DATA(rcu_preempt_data, print_one_rcu_data_csv, m); 138 PRINT_RCU_DATA(rcu_preempt_data, print_one_rcu_data_csv, m);
@@ -262,7 +266,7 @@ static void print_rcu_pendings(struct seq_file *m, struct rcu_state *rsp)
262 struct rcu_data *rdp; 266 struct rcu_data *rdp;
263 267
264 for_each_possible_cpu(cpu) { 268 for_each_possible_cpu(cpu) {
265 rdp = rsp->rda[cpu]; 269 rdp = per_cpu_ptr(rsp->rda, cpu);
266 if (rdp->beenonline) 270 if (rdp->beenonline)
267 print_one_rcu_pending(m, rdp); 271 print_one_rcu_pending(m, rdp);
268 } 272 }
diff --git a/kernel/rtmutex-tester.c b/kernel/rtmutex-tester.c
index a56f629b057a..66cb89bc5ef1 100644
--- a/kernel/rtmutex-tester.c
+++ b/kernel/rtmutex-tester.c
@@ -76,7 +76,9 @@ static int handle_op(struct test_thread_data *td, int lockwakeup)
76 } 76 }
77 77
78 if (!lockwakeup && td->bkl == 4) { 78 if (!lockwakeup && td->bkl == 4) {
79#ifdef CONFIG_LOCK_KERNEL
79 unlock_kernel(); 80 unlock_kernel();
81#endif
80 td->bkl = 0; 82 td->bkl = 0;
81 } 83 }
82 return 0; 84 return 0;
@@ -133,14 +135,18 @@ static int handle_op(struct test_thread_data *td, int lockwakeup)
133 if (td->bkl) 135 if (td->bkl)
134 return 0; 136 return 0;
135 td->bkl = 1; 137 td->bkl = 1;
138#ifdef CONFIG_LOCK_KERNEL
136 lock_kernel(); 139 lock_kernel();
140#endif
137 td->bkl = 4; 141 td->bkl = 4;
138 return 0; 142 return 0;
139 143
140 case RTTEST_UNLOCKBKL: 144 case RTTEST_UNLOCKBKL:
141 if (td->bkl != 4) 145 if (td->bkl != 4)
142 break; 146 break;
147#ifdef CONFIG_LOCK_KERNEL
143 unlock_kernel(); 148 unlock_kernel();
149#endif
144 td->bkl = 0; 150 td->bkl = 0;
145 return 0; 151 return 0;
146 152
diff --git a/kernel/sched.c b/kernel/sched.c
index dc85ceb90832..d42992bccdfa 100644
--- a/kernel/sched.c
+++ b/kernel/sched.c
@@ -426,9 +426,7 @@ struct root_domain {
426 */ 426 */
427 cpumask_var_t rto_mask; 427 cpumask_var_t rto_mask;
428 atomic_t rto_count; 428 atomic_t rto_count;
429#ifdef CONFIG_SMP
430 struct cpupri cpupri; 429 struct cpupri cpupri;
431#endif
432}; 430};
433 431
434/* 432/*
@@ -437,7 +435,7 @@ struct root_domain {
437 */ 435 */
438static struct root_domain def_root_domain; 436static struct root_domain def_root_domain;
439 437
440#endif 438#endif /* CONFIG_SMP */
441 439
442/* 440/*
443 * This is the main, per-CPU runqueue data structure. 441 * This is the main, per-CPU runqueue data structure.
@@ -488,11 +486,12 @@ struct rq {
488 */ 486 */
489 unsigned long nr_uninterruptible; 487 unsigned long nr_uninterruptible;
490 488
491 struct task_struct *curr, *idle; 489 struct task_struct *curr, *idle, *stop;
492 unsigned long next_balance; 490 unsigned long next_balance;
493 struct mm_struct *prev_mm; 491 struct mm_struct *prev_mm;
494 492
495 u64 clock; 493 u64 clock;
494 u64 clock_task;
496 495
497 atomic_t nr_iowait; 496 atomic_t nr_iowait;
498 497
@@ -520,6 +519,10 @@ struct rq {
520 u64 avg_idle; 519 u64 avg_idle;
521#endif 520#endif
522 521
522#ifdef CONFIG_IRQ_TIME_ACCOUNTING
523 u64 prev_irq_time;
524#endif
525
523 /* calc_load related fields */ 526 /* calc_load related fields */
524 unsigned long calc_load_update; 527 unsigned long calc_load_update;
525 long calc_load_active; 528 long calc_load_active;
@@ -643,10 +646,22 @@ static inline struct task_group *task_group(struct task_struct *p)
643 646
644#endif /* CONFIG_CGROUP_SCHED */ 647#endif /* CONFIG_CGROUP_SCHED */
645 648
649static u64 irq_time_cpu(int cpu);
650static void sched_irq_time_avg_update(struct rq *rq, u64 irq_time);
651
646inline void update_rq_clock(struct rq *rq) 652inline void update_rq_clock(struct rq *rq)
647{ 653{
648 if (!rq->skip_clock_update) 654 if (!rq->skip_clock_update) {
649 rq->clock = sched_clock_cpu(cpu_of(rq)); 655 int cpu = cpu_of(rq);
656 u64 irq_time;
657
658 rq->clock = sched_clock_cpu(cpu);
659 irq_time = irq_time_cpu(cpu);
660 if (rq->clock - irq_time > rq->clock_task)
661 rq->clock_task = rq->clock - irq_time;
662
663 sched_irq_time_avg_update(rq, irq_time);
664 }
650} 665}
651 666
652/* 667/*
@@ -723,7 +738,7 @@ sched_feat_write(struct file *filp, const char __user *ubuf,
723 size_t cnt, loff_t *ppos) 738 size_t cnt, loff_t *ppos)
724{ 739{
725 char buf[64]; 740 char buf[64];
726 char *cmp = buf; 741 char *cmp;
727 int neg = 0; 742 int neg = 0;
728 int i; 743 int i;
729 744
@@ -734,6 +749,7 @@ sched_feat_write(struct file *filp, const char __user *ubuf,
734 return -EFAULT; 749 return -EFAULT;
735 750
736 buf[cnt] = 0; 751 buf[cnt] = 0;
752 cmp = strstrip(buf);
737 753
738 if (strncmp(buf, "NO_", 3) == 0) { 754 if (strncmp(buf, "NO_", 3) == 0) {
739 neg = 1; 755 neg = 1;
@@ -741,9 +757,7 @@ sched_feat_write(struct file *filp, const char __user *ubuf,
741 } 757 }
742 758
743 for (i = 0; sched_feat_names[i]; i++) { 759 for (i = 0; sched_feat_names[i]; i++) {
744 int len = strlen(sched_feat_names[i]); 760 if (strcmp(cmp, sched_feat_names[i]) == 0) {
745
746 if (strncmp(cmp, sched_feat_names[i], len) == 0) {
747 if (neg) 761 if (neg)
748 sysctl_sched_features &= ~(1UL << i); 762 sysctl_sched_features &= ~(1UL << i);
749 else 763 else
@@ -1840,7 +1854,7 @@ static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
1840 1854
1841static const struct sched_class rt_sched_class; 1855static const struct sched_class rt_sched_class;
1842 1856
1843#define sched_class_highest (&rt_sched_class) 1857#define sched_class_highest (&stop_sched_class)
1844#define for_each_class(class) \ 1858#define for_each_class(class) \
1845 for (class = sched_class_highest; class; class = class->next) 1859 for (class = sched_class_highest; class; class = class->next)
1846 1860
@@ -1858,12 +1872,6 @@ static void dec_nr_running(struct rq *rq)
1858 1872
1859static void set_load_weight(struct task_struct *p) 1873static void set_load_weight(struct task_struct *p)
1860{ 1874{
1861 if (task_has_rt_policy(p)) {
1862 p->se.load.weight = 0;
1863 p->se.load.inv_weight = WMULT_CONST;
1864 return;
1865 }
1866
1867 /* 1875 /*
1868 * SCHED_IDLE tasks get minimal weight: 1876 * SCHED_IDLE tasks get minimal weight:
1869 */ 1877 */
@@ -1917,13 +1925,132 @@ static void deactivate_task(struct rq *rq, struct task_struct *p, int flags)
1917 dec_nr_running(rq); 1925 dec_nr_running(rq);
1918} 1926}
1919 1927
1928#ifdef CONFIG_IRQ_TIME_ACCOUNTING
1929
1930/*
1931 * There are no locks covering percpu hardirq/softirq time.
1932 * They are only modified in account_system_vtime, on corresponding CPU
1933 * with interrupts disabled. So, writes are safe.
1934 * They are read and saved off onto struct rq in update_rq_clock().
1935 * This may result in other CPU reading this CPU's irq time and can
1936 * race with irq/account_system_vtime on this CPU. We would either get old
1937 * or new value (or semi updated value on 32 bit) with a side effect of
1938 * accounting a slice of irq time to wrong task when irq is in progress
1939 * while we read rq->clock. That is a worthy compromise in place of having
1940 * locks on each irq in account_system_time.
1941 */
1942static DEFINE_PER_CPU(u64, cpu_hardirq_time);
1943static DEFINE_PER_CPU(u64, cpu_softirq_time);
1944
1945static DEFINE_PER_CPU(u64, irq_start_time);
1946static int sched_clock_irqtime;
1947
1948void enable_sched_clock_irqtime(void)
1949{
1950 sched_clock_irqtime = 1;
1951}
1952
1953void disable_sched_clock_irqtime(void)
1954{
1955 sched_clock_irqtime = 0;
1956}
1957
1958static u64 irq_time_cpu(int cpu)
1959{
1960 if (!sched_clock_irqtime)
1961 return 0;
1962
1963 return per_cpu(cpu_softirq_time, cpu) + per_cpu(cpu_hardirq_time, cpu);
1964}
1965
1966void account_system_vtime(struct task_struct *curr)
1967{
1968 unsigned long flags;
1969 int cpu;
1970 u64 now, delta;
1971
1972 if (!sched_clock_irqtime)
1973 return;
1974
1975 local_irq_save(flags);
1976
1977 cpu = smp_processor_id();
1978 now = sched_clock_cpu(cpu);
1979 delta = now - per_cpu(irq_start_time, cpu);
1980 per_cpu(irq_start_time, cpu) = now;
1981 /*
1982 * We do not account for softirq time from ksoftirqd here.
1983 * We want to continue accounting softirq time to ksoftirqd thread
1984 * in that case, so as not to confuse scheduler with a special task
1985 * that do not consume any time, but still wants to run.
1986 */
1987 if (hardirq_count())
1988 per_cpu(cpu_hardirq_time, cpu) += delta;
1989 else if (in_serving_softirq() && !(curr->flags & PF_KSOFTIRQD))
1990 per_cpu(cpu_softirq_time, cpu) += delta;
1991
1992 local_irq_restore(flags);
1993}
1994EXPORT_SYMBOL_GPL(account_system_vtime);
1995
1996static void sched_irq_time_avg_update(struct rq *rq, u64 curr_irq_time)
1997{
1998 if (sched_clock_irqtime && sched_feat(NONIRQ_POWER)) {
1999 u64 delta_irq = curr_irq_time - rq->prev_irq_time;
2000 rq->prev_irq_time = curr_irq_time;
2001 sched_rt_avg_update(rq, delta_irq);
2002 }
2003}
2004
2005#else
2006
2007static u64 irq_time_cpu(int cpu)
2008{
2009 return 0;
2010}
2011
2012static void sched_irq_time_avg_update(struct rq *rq, u64 curr_irq_time) { }
2013
2014#endif
2015
1920#include "sched_idletask.c" 2016#include "sched_idletask.c"
1921#include "sched_fair.c" 2017#include "sched_fair.c"
1922#include "sched_rt.c" 2018#include "sched_rt.c"
2019#include "sched_stoptask.c"
1923#ifdef CONFIG_SCHED_DEBUG 2020#ifdef CONFIG_SCHED_DEBUG
1924# include "sched_debug.c" 2021# include "sched_debug.c"
1925#endif 2022#endif
1926 2023
2024void sched_set_stop_task(int cpu, struct task_struct *stop)
2025{
2026 struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 };
2027 struct task_struct *old_stop = cpu_rq(cpu)->stop;
2028
2029 if (stop) {
2030 /*
2031 * Make it appear like a SCHED_FIFO task, its something
2032 * userspace knows about and won't get confused about.
2033 *
2034 * Also, it will make PI more or less work without too
2035 * much confusion -- but then, stop work should not
2036 * rely on PI working anyway.
2037 */
2038 sched_setscheduler_nocheck(stop, SCHED_FIFO, &param);
2039
2040 stop->sched_class = &stop_sched_class;
2041 }
2042
2043 cpu_rq(cpu)->stop = stop;
2044
2045 if (old_stop) {
2046 /*
2047 * Reset it back to a normal scheduling class so that
2048 * it can die in pieces.
2049 */
2050 old_stop->sched_class = &rt_sched_class;
2051 }
2052}
2053
1927/* 2054/*
1928 * __normal_prio - return the priority that is based on the static prio 2055 * __normal_prio - return the priority that is based on the static prio
1929 */ 2056 */
@@ -2003,6 +2130,9 @@ task_hot(struct task_struct *p, u64 now, struct sched_domain *sd)
2003 if (p->sched_class != &fair_sched_class) 2130 if (p->sched_class != &fair_sched_class)
2004 return 0; 2131 return 0;
2005 2132
2133 if (unlikely(p->policy == SCHED_IDLE))
2134 return 0;
2135
2006 /* 2136 /*
2007 * Buddy candidates are cache hot: 2137 * Buddy candidates are cache hot:
2008 */ 2138 */
@@ -2852,14 +2982,14 @@ context_switch(struct rq *rq, struct task_struct *prev,
2852 */ 2982 */
2853 arch_start_context_switch(prev); 2983 arch_start_context_switch(prev);
2854 2984
2855 if (likely(!mm)) { 2985 if (!mm) {
2856 next->active_mm = oldmm; 2986 next->active_mm = oldmm;
2857 atomic_inc(&oldmm->mm_count); 2987 atomic_inc(&oldmm->mm_count);
2858 enter_lazy_tlb(oldmm, next); 2988 enter_lazy_tlb(oldmm, next);
2859 } else 2989 } else
2860 switch_mm(oldmm, mm, next); 2990 switch_mm(oldmm, mm, next);
2861 2991
2862 if (likely(!prev->mm)) { 2992 if (!prev->mm) {
2863 prev->active_mm = NULL; 2993 prev->active_mm = NULL;
2864 rq->prev_mm = oldmm; 2994 rq->prev_mm = oldmm;
2865 } 2995 }
@@ -3248,7 +3378,7 @@ static u64 do_task_delta_exec(struct task_struct *p, struct rq *rq)
3248 3378
3249 if (task_current(rq, p)) { 3379 if (task_current(rq, p)) {
3250 update_rq_clock(rq); 3380 update_rq_clock(rq);
3251 ns = rq->clock - p->se.exec_start; 3381 ns = rq->clock_task - p->se.exec_start;
3252 if ((s64)ns < 0) 3382 if ((s64)ns < 0)
3253 ns = 0; 3383 ns = 0;
3254 } 3384 }
@@ -3397,7 +3527,7 @@ void account_system_time(struct task_struct *p, int hardirq_offset,
3397 tmp = cputime_to_cputime64(cputime); 3527 tmp = cputime_to_cputime64(cputime);
3398 if (hardirq_count() - hardirq_offset) 3528 if (hardirq_count() - hardirq_offset)
3399 cpustat->irq = cputime64_add(cpustat->irq, tmp); 3529 cpustat->irq = cputime64_add(cpustat->irq, tmp);
3400 else if (softirq_count()) 3530 else if (in_serving_softirq())
3401 cpustat->softirq = cputime64_add(cpustat->softirq, tmp); 3531 cpustat->softirq = cputime64_add(cpustat->softirq, tmp);
3402 else 3532 else
3403 cpustat->system = cputime64_add(cpustat->system, tmp); 3533 cpustat->system = cputime64_add(cpustat->system, tmp);
@@ -3584,7 +3714,7 @@ void scheduler_tick(void)
3584 curr->sched_class->task_tick(rq, curr, 0); 3714 curr->sched_class->task_tick(rq, curr, 0);
3585 raw_spin_unlock(&rq->lock); 3715 raw_spin_unlock(&rq->lock);
3586 3716
3587 perf_event_task_tick(curr); 3717 perf_event_task_tick();
3588 3718
3589#ifdef CONFIG_SMP 3719#ifdef CONFIG_SMP
3590 rq->idle_at_tick = idle_cpu(cpu); 3720 rq->idle_at_tick = idle_cpu(cpu);
@@ -3723,17 +3853,13 @@ pick_next_task(struct rq *rq)
3723 return p; 3853 return p;
3724 } 3854 }
3725 3855
3726 class = sched_class_highest; 3856 for_each_class(class) {
3727 for ( ; ; ) {
3728 p = class->pick_next_task(rq); 3857 p = class->pick_next_task(rq);
3729 if (p) 3858 if (p)
3730 return p; 3859 return p;
3731 /*
3732 * Will never be NULL as the idle class always
3733 * returns a non-NULL p:
3734 */
3735 class = class->next;
3736 } 3860 }
3861
3862 BUG(); /* the idle class will always have a runnable task */
3737} 3863}
3738 3864
3739/* 3865/*
@@ -4358,6 +4484,7 @@ void rt_mutex_setprio(struct task_struct *p, int prio)
4358 4484
4359 rq = task_rq_lock(p, &flags); 4485 rq = task_rq_lock(p, &flags);
4360 4486
4487 trace_sched_pi_setprio(p, prio);
4361 oldprio = p->prio; 4488 oldprio = p->prio;
4362 prev_class = p->sched_class; 4489 prev_class = p->sched_class;
4363 on_rq = p->se.on_rq; 4490 on_rq = p->se.on_rq;
@@ -4645,7 +4772,7 @@ recheck:
4645 } 4772 }
4646 4773
4647 if (user) { 4774 if (user) {
4648 retval = security_task_setscheduler(p, policy, param); 4775 retval = security_task_setscheduler(p);
4649 if (retval) 4776 if (retval)
4650 return retval; 4777 return retval;
4651 } 4778 }
@@ -4661,6 +4788,15 @@ recheck:
4661 */ 4788 */
4662 rq = __task_rq_lock(p); 4789 rq = __task_rq_lock(p);
4663 4790
4791 /*
4792 * Changing the policy of the stop threads its a very bad idea
4793 */
4794 if (p == rq->stop) {
4795 __task_rq_unlock(rq);
4796 raw_spin_unlock_irqrestore(&p->pi_lock, flags);
4797 return -EINVAL;
4798 }
4799
4664#ifdef CONFIG_RT_GROUP_SCHED 4800#ifdef CONFIG_RT_GROUP_SCHED
4665 if (user) { 4801 if (user) {
4666 /* 4802 /*
@@ -4887,13 +5023,13 @@ long sched_setaffinity(pid_t pid, const struct cpumask *in_mask)
4887 if (!check_same_owner(p) && !capable(CAP_SYS_NICE)) 5023 if (!check_same_owner(p) && !capable(CAP_SYS_NICE))
4888 goto out_unlock; 5024 goto out_unlock;
4889 5025
4890 retval = security_task_setscheduler(p, 0, NULL); 5026 retval = security_task_setscheduler(p);
4891 if (retval) 5027 if (retval)
4892 goto out_unlock; 5028 goto out_unlock;
4893 5029
4894 cpuset_cpus_allowed(p, cpus_allowed); 5030 cpuset_cpus_allowed(p, cpus_allowed);
4895 cpumask_and(new_mask, in_mask, cpus_allowed); 5031 cpumask_and(new_mask, in_mask, cpus_allowed);
4896 again: 5032again:
4897 retval = set_cpus_allowed_ptr(p, new_mask); 5033 retval = set_cpus_allowed_ptr(p, new_mask);
4898 5034
4899 if (!retval) { 5035 if (!retval) {
@@ -5337,7 +5473,19 @@ void __cpuinit init_idle(struct task_struct *idle, int cpu)
5337 idle->se.exec_start = sched_clock(); 5473 idle->se.exec_start = sched_clock();
5338 5474
5339 cpumask_copy(&idle->cpus_allowed, cpumask_of(cpu)); 5475 cpumask_copy(&idle->cpus_allowed, cpumask_of(cpu));
5476 /*
5477 * We're having a chicken and egg problem, even though we are
5478 * holding rq->lock, the cpu isn't yet set to this cpu so the
5479 * lockdep check in task_group() will fail.
5480 *
5481 * Similar case to sched_fork(). / Alternatively we could
5482 * use task_rq_lock() here and obtain the other rq->lock.
5483 *
5484 * Silence PROVE_RCU
5485 */
5486 rcu_read_lock();
5340 __set_task_cpu(idle, cpu); 5487 __set_task_cpu(idle, cpu);
5488 rcu_read_unlock();
5341 5489
5342 rq->curr = rq->idle = idle; 5490 rq->curr = rq->idle = idle;
5343#if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW) 5491#if defined(CONFIG_SMP) && defined(__ARCH_WANT_UNLOCKED_CTXSW)
@@ -6514,6 +6662,7 @@ struct s_data {
6514 cpumask_var_t nodemask; 6662 cpumask_var_t nodemask;
6515 cpumask_var_t this_sibling_map; 6663 cpumask_var_t this_sibling_map;
6516 cpumask_var_t this_core_map; 6664 cpumask_var_t this_core_map;
6665 cpumask_var_t this_book_map;
6517 cpumask_var_t send_covered; 6666 cpumask_var_t send_covered;
6518 cpumask_var_t tmpmask; 6667 cpumask_var_t tmpmask;
6519 struct sched_group **sched_group_nodes; 6668 struct sched_group **sched_group_nodes;
@@ -6525,6 +6674,7 @@ enum s_alloc {
6525 sa_rootdomain, 6674 sa_rootdomain,
6526 sa_tmpmask, 6675 sa_tmpmask,
6527 sa_send_covered, 6676 sa_send_covered,
6677 sa_this_book_map,
6528 sa_this_core_map, 6678 sa_this_core_map,
6529 sa_this_sibling_map, 6679 sa_this_sibling_map,
6530 sa_nodemask, 6680 sa_nodemask,
@@ -6560,31 +6710,48 @@ cpu_to_cpu_group(int cpu, const struct cpumask *cpu_map,
6560#ifdef CONFIG_SCHED_MC 6710#ifdef CONFIG_SCHED_MC
6561static DEFINE_PER_CPU(struct static_sched_domain, core_domains); 6711static DEFINE_PER_CPU(struct static_sched_domain, core_domains);
6562static DEFINE_PER_CPU(struct static_sched_group, sched_group_core); 6712static DEFINE_PER_CPU(struct static_sched_group, sched_group_core);
6563#endif /* CONFIG_SCHED_MC */
6564 6713
6565#if defined(CONFIG_SCHED_MC) && defined(CONFIG_SCHED_SMT)
6566static int 6714static int
6567cpu_to_core_group(int cpu, const struct cpumask *cpu_map, 6715cpu_to_core_group(int cpu, const struct cpumask *cpu_map,
6568 struct sched_group **sg, struct cpumask *mask) 6716 struct sched_group **sg, struct cpumask *mask)
6569{ 6717{
6570 int group; 6718 int group;
6571 6719#ifdef CONFIG_SCHED_SMT
6572 cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map); 6720 cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map);
6573 group = cpumask_first(mask); 6721 group = cpumask_first(mask);
6722#else
6723 group = cpu;
6724#endif
6574 if (sg) 6725 if (sg)
6575 *sg = &per_cpu(sched_group_core, group).sg; 6726 *sg = &per_cpu(sched_group_core, group).sg;
6576 return group; 6727 return group;
6577} 6728}
6578#elif defined(CONFIG_SCHED_MC) 6729#endif /* CONFIG_SCHED_MC */
6730
6731/*
6732 * book sched-domains:
6733 */
6734#ifdef CONFIG_SCHED_BOOK
6735static DEFINE_PER_CPU(struct static_sched_domain, book_domains);
6736static DEFINE_PER_CPU(struct static_sched_group, sched_group_book);
6737
6579static int 6738static int
6580cpu_to_core_group(int cpu, const struct cpumask *cpu_map, 6739cpu_to_book_group(int cpu, const struct cpumask *cpu_map,
6581 struct sched_group **sg, struct cpumask *unused) 6740 struct sched_group **sg, struct cpumask *mask)
6582{ 6741{
6742 int group = cpu;
6743#ifdef CONFIG_SCHED_MC
6744 cpumask_and(mask, cpu_coregroup_mask(cpu), cpu_map);
6745 group = cpumask_first(mask);
6746#elif defined(CONFIG_SCHED_SMT)
6747 cpumask_and(mask, topology_thread_cpumask(cpu), cpu_map);
6748 group = cpumask_first(mask);
6749#endif
6583 if (sg) 6750 if (sg)
6584 *sg = &per_cpu(sched_group_core, cpu).sg; 6751 *sg = &per_cpu(sched_group_book, group).sg;
6585 return cpu; 6752 return group;
6586} 6753}
6587#endif 6754#endif /* CONFIG_SCHED_BOOK */
6588 6755
6589static DEFINE_PER_CPU(struct static_sched_domain, phys_domains); 6756static DEFINE_PER_CPU(struct static_sched_domain, phys_domains);
6590static DEFINE_PER_CPU(struct static_sched_group, sched_group_phys); 6757static DEFINE_PER_CPU(struct static_sched_group, sched_group_phys);
@@ -6594,7 +6761,10 @@ cpu_to_phys_group(int cpu, const struct cpumask *cpu_map,
6594 struct sched_group **sg, struct cpumask *mask) 6761 struct sched_group **sg, struct cpumask *mask)
6595{ 6762{
6596 int group; 6763 int group;
6597#ifdef CONFIG_SCHED_MC 6764#ifdef CONFIG_SCHED_BOOK
6765 cpumask_and(mask, cpu_book_mask(cpu), cpu_map);
6766 group = cpumask_first(mask);
6767#elif defined(CONFIG_SCHED_MC)
6598 cpumask_and(mask, cpu_coregroup_mask(cpu), cpu_map); 6768 cpumask_and(mask, cpu_coregroup_mask(cpu), cpu_map);
6599 group = cpumask_first(mask); 6769 group = cpumask_first(mask);
6600#elif defined(CONFIG_SCHED_SMT) 6770#elif defined(CONFIG_SCHED_SMT)
@@ -6855,6 +7025,9 @@ SD_INIT_FUNC(CPU)
6855#ifdef CONFIG_SCHED_MC 7025#ifdef CONFIG_SCHED_MC
6856 SD_INIT_FUNC(MC) 7026 SD_INIT_FUNC(MC)
6857#endif 7027#endif
7028#ifdef CONFIG_SCHED_BOOK
7029 SD_INIT_FUNC(BOOK)
7030#endif
6858 7031
6859static int default_relax_domain_level = -1; 7032static int default_relax_domain_level = -1;
6860 7033
@@ -6904,6 +7077,8 @@ static void __free_domain_allocs(struct s_data *d, enum s_alloc what,
6904 free_cpumask_var(d->tmpmask); /* fall through */ 7077 free_cpumask_var(d->tmpmask); /* fall through */
6905 case sa_send_covered: 7078 case sa_send_covered:
6906 free_cpumask_var(d->send_covered); /* fall through */ 7079 free_cpumask_var(d->send_covered); /* fall through */
7080 case sa_this_book_map:
7081 free_cpumask_var(d->this_book_map); /* fall through */
6907 case sa_this_core_map: 7082 case sa_this_core_map:
6908 free_cpumask_var(d->this_core_map); /* fall through */ 7083 free_cpumask_var(d->this_core_map); /* fall through */
6909 case sa_this_sibling_map: 7084 case sa_this_sibling_map:
@@ -6950,8 +7125,10 @@ static enum s_alloc __visit_domain_allocation_hell(struct s_data *d,
6950 return sa_nodemask; 7125 return sa_nodemask;
6951 if (!alloc_cpumask_var(&d->this_core_map, GFP_KERNEL)) 7126 if (!alloc_cpumask_var(&d->this_core_map, GFP_KERNEL))
6952 return sa_this_sibling_map; 7127 return sa_this_sibling_map;
6953 if (!alloc_cpumask_var(&d->send_covered, GFP_KERNEL)) 7128 if (!alloc_cpumask_var(&d->this_book_map, GFP_KERNEL))
6954 return sa_this_core_map; 7129 return sa_this_core_map;
7130 if (!alloc_cpumask_var(&d->send_covered, GFP_KERNEL))
7131 return sa_this_book_map;
6955 if (!alloc_cpumask_var(&d->tmpmask, GFP_KERNEL)) 7132 if (!alloc_cpumask_var(&d->tmpmask, GFP_KERNEL))
6956 return sa_send_covered; 7133 return sa_send_covered;
6957 d->rd = alloc_rootdomain(); 7134 d->rd = alloc_rootdomain();
@@ -7009,6 +7186,23 @@ static struct sched_domain *__build_cpu_sched_domain(struct s_data *d,
7009 return sd; 7186 return sd;
7010} 7187}
7011 7188
7189static struct sched_domain *__build_book_sched_domain(struct s_data *d,
7190 const struct cpumask *cpu_map, struct sched_domain_attr *attr,
7191 struct sched_domain *parent, int i)
7192{
7193 struct sched_domain *sd = parent;
7194#ifdef CONFIG_SCHED_BOOK
7195 sd = &per_cpu(book_domains, i).sd;
7196 SD_INIT(sd, BOOK);
7197 set_domain_attribute(sd, attr);
7198 cpumask_and(sched_domain_span(sd), cpu_map, cpu_book_mask(i));
7199 sd->parent = parent;
7200 parent->child = sd;
7201 cpu_to_book_group(i, cpu_map, &sd->groups, d->tmpmask);
7202#endif
7203 return sd;
7204}
7205
7012static struct sched_domain *__build_mc_sched_domain(struct s_data *d, 7206static struct sched_domain *__build_mc_sched_domain(struct s_data *d,
7013 const struct cpumask *cpu_map, struct sched_domain_attr *attr, 7207 const struct cpumask *cpu_map, struct sched_domain_attr *attr,
7014 struct sched_domain *parent, int i) 7208 struct sched_domain *parent, int i)
@@ -7066,6 +7260,15 @@ static void build_sched_groups(struct s_data *d, enum sched_domain_level l,
7066 d->send_covered, d->tmpmask); 7260 d->send_covered, d->tmpmask);
7067 break; 7261 break;
7068#endif 7262#endif
7263#ifdef CONFIG_SCHED_BOOK
7264 case SD_LV_BOOK: /* set up book groups */
7265 cpumask_and(d->this_book_map, cpu_map, cpu_book_mask(cpu));
7266 if (cpu == cpumask_first(d->this_book_map))
7267 init_sched_build_groups(d->this_book_map, cpu_map,
7268 &cpu_to_book_group,
7269 d->send_covered, d->tmpmask);
7270 break;
7271#endif
7069 case SD_LV_CPU: /* set up physical groups */ 7272 case SD_LV_CPU: /* set up physical groups */
7070 cpumask_and(d->nodemask, cpumask_of_node(cpu), cpu_map); 7273 cpumask_and(d->nodemask, cpumask_of_node(cpu), cpu_map);
7071 if (!cpumask_empty(d->nodemask)) 7274 if (!cpumask_empty(d->nodemask))
@@ -7113,12 +7316,14 @@ static int __build_sched_domains(const struct cpumask *cpu_map,
7113 7316
7114 sd = __build_numa_sched_domains(&d, cpu_map, attr, i); 7317 sd = __build_numa_sched_domains(&d, cpu_map, attr, i);
7115 sd = __build_cpu_sched_domain(&d, cpu_map, attr, sd, i); 7318 sd = __build_cpu_sched_domain(&d, cpu_map, attr, sd, i);
7319 sd = __build_book_sched_domain(&d, cpu_map, attr, sd, i);
7116 sd = __build_mc_sched_domain(&d, cpu_map, attr, sd, i); 7320 sd = __build_mc_sched_domain(&d, cpu_map, attr, sd, i);
7117 sd = __build_smt_sched_domain(&d, cpu_map, attr, sd, i); 7321 sd = __build_smt_sched_domain(&d, cpu_map, attr, sd, i);
7118 } 7322 }
7119 7323
7120 for_each_cpu(i, cpu_map) { 7324 for_each_cpu(i, cpu_map) {
7121 build_sched_groups(&d, SD_LV_SIBLING, cpu_map, i); 7325 build_sched_groups(&d, SD_LV_SIBLING, cpu_map, i);
7326 build_sched_groups(&d, SD_LV_BOOK, cpu_map, i);
7122 build_sched_groups(&d, SD_LV_MC, cpu_map, i); 7327 build_sched_groups(&d, SD_LV_MC, cpu_map, i);
7123 } 7328 }
7124 7329
@@ -7149,6 +7354,12 @@ static int __build_sched_domains(const struct cpumask *cpu_map,
7149 init_sched_groups_power(i, sd); 7354 init_sched_groups_power(i, sd);
7150 } 7355 }
7151#endif 7356#endif
7357#ifdef CONFIG_SCHED_BOOK
7358 for_each_cpu(i, cpu_map) {
7359 sd = &per_cpu(book_domains, i).sd;
7360 init_sched_groups_power(i, sd);
7361 }
7362#endif
7152 7363
7153 for_each_cpu(i, cpu_map) { 7364 for_each_cpu(i, cpu_map) {
7154 sd = &per_cpu(phys_domains, i).sd; 7365 sd = &per_cpu(phys_domains, i).sd;
@@ -7174,6 +7385,8 @@ static int __build_sched_domains(const struct cpumask *cpu_map,
7174 sd = &per_cpu(cpu_domains, i).sd; 7385 sd = &per_cpu(cpu_domains, i).sd;
7175#elif defined(CONFIG_SCHED_MC) 7386#elif defined(CONFIG_SCHED_MC)
7176 sd = &per_cpu(core_domains, i).sd; 7387 sd = &per_cpu(core_domains, i).sd;
7388#elif defined(CONFIG_SCHED_BOOK)
7389 sd = &per_cpu(book_domains, i).sd;
7177#else 7390#else
7178 sd = &per_cpu(phys_domains, i).sd; 7391 sd = &per_cpu(phys_domains, i).sd;
7179#endif 7392#endif
@@ -8078,9 +8291,9 @@ int alloc_fair_sched_group(struct task_group *tg, struct task_group *parent)
8078 8291
8079 return 1; 8292 return 1;
8080 8293
8081 err_free_rq: 8294err_free_rq:
8082 kfree(cfs_rq); 8295 kfree(cfs_rq);
8083 err: 8296err:
8084 return 0; 8297 return 0;
8085} 8298}
8086 8299
@@ -8168,9 +8381,9 @@ int alloc_rt_sched_group(struct task_group *tg, struct task_group *parent)
8168 8381
8169 return 1; 8382 return 1;
8170 8383
8171 err_free_rq: 8384err_free_rq:
8172 kfree(rt_rq); 8385 kfree(rt_rq);
8173 err: 8386err:
8174 return 0; 8387 return 0;
8175} 8388}
8176 8389
@@ -8528,7 +8741,7 @@ static int tg_set_bandwidth(struct task_group *tg,
8528 raw_spin_unlock(&rt_rq->rt_runtime_lock); 8741 raw_spin_unlock(&rt_rq->rt_runtime_lock);
8529 } 8742 }
8530 raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock); 8743 raw_spin_unlock_irq(&tg->rt_bandwidth.rt_runtime_lock);
8531 unlock: 8744unlock:
8532 read_unlock(&tasklist_lock); 8745 read_unlock(&tasklist_lock);
8533 mutex_unlock(&rt_constraints_mutex); 8746 mutex_unlock(&rt_constraints_mutex);
8534 8747
diff --git a/kernel/sched_fair.c b/kernel/sched_fair.c
index db3f674ca49d..933f3d1b62ea 100644
--- a/kernel/sched_fair.c
+++ b/kernel/sched_fair.c
@@ -25,7 +25,7 @@
25 25
26/* 26/*
27 * Targeted preemption latency for CPU-bound tasks: 27 * Targeted preemption latency for CPU-bound tasks:
28 * (default: 5ms * (1 + ilog(ncpus)), units: nanoseconds) 28 * (default: 6ms * (1 + ilog(ncpus)), units: nanoseconds)
29 * 29 *
30 * NOTE: this latency value is not the same as the concept of 30 * NOTE: this latency value is not the same as the concept of
31 * 'timeslice length' - timeslices in CFS are of variable length 31 * 'timeslice length' - timeslices in CFS are of variable length
@@ -52,7 +52,7 @@ enum sched_tunable_scaling sysctl_sched_tunable_scaling
52 52
53/* 53/*
54 * Minimal preemption granularity for CPU-bound tasks: 54 * Minimal preemption granularity for CPU-bound tasks:
55 * (default: 2 msec * (1 + ilog(ncpus)), units: nanoseconds) 55 * (default: 0.75 msec * (1 + ilog(ncpus)), units: nanoseconds)
56 */ 56 */
57unsigned int sysctl_sched_min_granularity = 750000ULL; 57unsigned int sysctl_sched_min_granularity = 750000ULL;
58unsigned int normalized_sysctl_sched_min_granularity = 750000ULL; 58unsigned int normalized_sysctl_sched_min_granularity = 750000ULL;
@@ -519,7 +519,7 @@ __update_curr(struct cfs_rq *cfs_rq, struct sched_entity *curr,
519static void update_curr(struct cfs_rq *cfs_rq) 519static void update_curr(struct cfs_rq *cfs_rq)
520{ 520{
521 struct sched_entity *curr = cfs_rq->curr; 521 struct sched_entity *curr = cfs_rq->curr;
522 u64 now = rq_of(cfs_rq)->clock; 522 u64 now = rq_of(cfs_rq)->clock_task;
523 unsigned long delta_exec; 523 unsigned long delta_exec;
524 524
525 if (unlikely(!curr)) 525 if (unlikely(!curr))
@@ -602,7 +602,7 @@ update_stats_curr_start(struct cfs_rq *cfs_rq, struct sched_entity *se)
602 /* 602 /*
603 * We are starting a new run period: 603 * We are starting a new run period:
604 */ 604 */
605 se->exec_start = rq_of(cfs_rq)->clock; 605 se->exec_start = rq_of(cfs_rq)->clock_task;
606} 606}
607 607
608/************************************************** 608/**************************************************
@@ -1764,6 +1764,10 @@ static void pull_task(struct rq *src_rq, struct task_struct *p,
1764 set_task_cpu(p, this_cpu); 1764 set_task_cpu(p, this_cpu);
1765 activate_task(this_rq, p, 0); 1765 activate_task(this_rq, p, 0);
1766 check_preempt_curr(this_rq, p, 0); 1766 check_preempt_curr(this_rq, p, 0);
1767
1768 /* re-arm NEWIDLE balancing when moving tasks */
1769 src_rq->avg_idle = this_rq->avg_idle = 2*sysctl_sched_migration_cost;
1770 this_rq->idle_stamp = 0;
1767} 1771}
1768 1772
1769/* 1773/*
@@ -1798,7 +1802,7 @@ int can_migrate_task(struct task_struct *p, struct rq *rq, int this_cpu,
1798 * 2) too many balance attempts have failed. 1802 * 2) too many balance attempts have failed.
1799 */ 1803 */
1800 1804
1801 tsk_cache_hot = task_hot(p, rq->clock, sd); 1805 tsk_cache_hot = task_hot(p, rq->clock_task, sd);
1802 if (!tsk_cache_hot || 1806 if (!tsk_cache_hot ||
1803 sd->nr_balance_failed > sd->cache_nice_tries) { 1807 sd->nr_balance_failed > sd->cache_nice_tries) {
1804#ifdef CONFIG_SCHEDSTATS 1808#ifdef CONFIG_SCHEDSTATS
@@ -2030,12 +2034,14 @@ struct sd_lb_stats {
2030 unsigned long this_load; 2034 unsigned long this_load;
2031 unsigned long this_load_per_task; 2035 unsigned long this_load_per_task;
2032 unsigned long this_nr_running; 2036 unsigned long this_nr_running;
2037 unsigned long this_has_capacity;
2033 2038
2034 /* Statistics of the busiest group */ 2039 /* Statistics of the busiest group */
2035 unsigned long max_load; 2040 unsigned long max_load;
2036 unsigned long busiest_load_per_task; 2041 unsigned long busiest_load_per_task;
2037 unsigned long busiest_nr_running; 2042 unsigned long busiest_nr_running;
2038 unsigned long busiest_group_capacity; 2043 unsigned long busiest_group_capacity;
2044 unsigned long busiest_has_capacity;
2039 2045
2040 int group_imb; /* Is there imbalance in this sd */ 2046 int group_imb; /* Is there imbalance in this sd */
2041#if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT) 2047#if defined(CONFIG_SCHED_MC) || defined(CONFIG_SCHED_SMT)
@@ -2058,6 +2064,7 @@ struct sg_lb_stats {
2058 unsigned long sum_weighted_load; /* Weighted load of group's tasks */ 2064 unsigned long sum_weighted_load; /* Weighted load of group's tasks */
2059 unsigned long group_capacity; 2065 unsigned long group_capacity;
2060 int group_imb; /* Is there an imbalance in the group ? */ 2066 int group_imb; /* Is there an imbalance in the group ? */
2067 int group_has_capacity; /* Is there extra capacity in the group? */
2061}; 2068};
2062 2069
2063/** 2070/**
@@ -2268,7 +2275,13 @@ unsigned long scale_rt_power(int cpu)
2268 u64 total, available; 2275 u64 total, available;
2269 2276
2270 total = sched_avg_period() + (rq->clock - rq->age_stamp); 2277 total = sched_avg_period() + (rq->clock - rq->age_stamp);
2271 available = total - rq->rt_avg; 2278
2279 if (unlikely(total < rq->rt_avg)) {
2280 /* Ensures that power won't end up being negative */
2281 available = 0;
2282 } else {
2283 available = total - rq->rt_avg;
2284 }
2272 2285
2273 if (unlikely((s64)total < SCHED_LOAD_SCALE)) 2286 if (unlikely((s64)total < SCHED_LOAD_SCALE))
2274 total = SCHED_LOAD_SCALE; 2287 total = SCHED_LOAD_SCALE;
@@ -2378,7 +2391,7 @@ static inline void update_sg_lb_stats(struct sched_domain *sd,
2378 int local_group, const struct cpumask *cpus, 2391 int local_group, const struct cpumask *cpus,
2379 int *balance, struct sg_lb_stats *sgs) 2392 int *balance, struct sg_lb_stats *sgs)
2380{ 2393{
2381 unsigned long load, max_cpu_load, min_cpu_load; 2394 unsigned long load, max_cpu_load, min_cpu_load, max_nr_running;
2382 int i; 2395 int i;
2383 unsigned int balance_cpu = -1, first_idle_cpu = 0; 2396 unsigned int balance_cpu = -1, first_idle_cpu = 0;
2384 unsigned long avg_load_per_task = 0; 2397 unsigned long avg_load_per_task = 0;
@@ -2389,6 +2402,7 @@ static inline void update_sg_lb_stats(struct sched_domain *sd,
2389 /* Tally up the load of all CPUs in the group */ 2402 /* Tally up the load of all CPUs in the group */
2390 max_cpu_load = 0; 2403 max_cpu_load = 0;
2391 min_cpu_load = ~0UL; 2404 min_cpu_load = ~0UL;
2405 max_nr_running = 0;
2392 2406
2393 for_each_cpu_and(i, sched_group_cpus(group), cpus) { 2407 for_each_cpu_and(i, sched_group_cpus(group), cpus) {
2394 struct rq *rq = cpu_rq(i); 2408 struct rq *rq = cpu_rq(i);
@@ -2406,8 +2420,10 @@ static inline void update_sg_lb_stats(struct sched_domain *sd,
2406 load = target_load(i, load_idx); 2420 load = target_load(i, load_idx);
2407 } else { 2421 } else {
2408 load = source_load(i, load_idx); 2422 load = source_load(i, load_idx);
2409 if (load > max_cpu_load) 2423 if (load > max_cpu_load) {
2410 max_cpu_load = load; 2424 max_cpu_load = load;
2425 max_nr_running = rq->nr_running;
2426 }
2411 if (min_cpu_load > load) 2427 if (min_cpu_load > load)
2412 min_cpu_load = load; 2428 min_cpu_load = load;
2413 } 2429 }
@@ -2447,13 +2463,15 @@ static inline void update_sg_lb_stats(struct sched_domain *sd,
2447 if (sgs->sum_nr_running) 2463 if (sgs->sum_nr_running)
2448 avg_load_per_task = sgs->sum_weighted_load / sgs->sum_nr_running; 2464 avg_load_per_task = sgs->sum_weighted_load / sgs->sum_nr_running;
2449 2465
2450 if ((max_cpu_load - min_cpu_load) > 2*avg_load_per_task) 2466 if ((max_cpu_load - min_cpu_load) > 2*avg_load_per_task && max_nr_running > 1)
2451 sgs->group_imb = 1; 2467 sgs->group_imb = 1;
2452 2468
2453 sgs->group_capacity = 2469 sgs->group_capacity = DIV_ROUND_CLOSEST(group->cpu_power, SCHED_LOAD_SCALE);
2454 DIV_ROUND_CLOSEST(group->cpu_power, SCHED_LOAD_SCALE);
2455 if (!sgs->group_capacity) 2470 if (!sgs->group_capacity)
2456 sgs->group_capacity = fix_small_capacity(sd, group); 2471 sgs->group_capacity = fix_small_capacity(sd, group);
2472
2473 if (sgs->group_capacity > sgs->sum_nr_running)
2474 sgs->group_has_capacity = 1;
2457} 2475}
2458 2476
2459/** 2477/**
@@ -2542,9 +2560,14 @@ static inline void update_sd_lb_stats(struct sched_domain *sd, int this_cpu,
2542 /* 2560 /*
2543 * In case the child domain prefers tasks go to siblings 2561 * In case the child domain prefers tasks go to siblings
2544 * first, lower the sg capacity to one so that we'll try 2562 * first, lower the sg capacity to one so that we'll try
2545 * and move all the excess tasks away. 2563 * and move all the excess tasks away. We lower the capacity
2564 * of a group only if the local group has the capacity to fit
2565 * these excess tasks, i.e. nr_running < group_capacity. The
2566 * extra check prevents the case where you always pull from the
2567 * heaviest group when it is already under-utilized (possible
2568 * with a large weight task outweighs the tasks on the system).
2546 */ 2569 */
2547 if (prefer_sibling) 2570 if (prefer_sibling && !local_group && sds->this_has_capacity)
2548 sgs.group_capacity = min(sgs.group_capacity, 1UL); 2571 sgs.group_capacity = min(sgs.group_capacity, 1UL);
2549 2572
2550 if (local_group) { 2573 if (local_group) {
@@ -2552,12 +2575,14 @@ static inline void update_sd_lb_stats(struct sched_domain *sd, int this_cpu,
2552 sds->this = sg; 2575 sds->this = sg;
2553 sds->this_nr_running = sgs.sum_nr_running; 2576 sds->this_nr_running = sgs.sum_nr_running;
2554 sds->this_load_per_task = sgs.sum_weighted_load; 2577 sds->this_load_per_task = sgs.sum_weighted_load;
2578 sds->this_has_capacity = sgs.group_has_capacity;
2555 } else if (update_sd_pick_busiest(sd, sds, sg, &sgs, this_cpu)) { 2579 } else if (update_sd_pick_busiest(sd, sds, sg, &sgs, this_cpu)) {
2556 sds->max_load = sgs.avg_load; 2580 sds->max_load = sgs.avg_load;
2557 sds->busiest = sg; 2581 sds->busiest = sg;
2558 sds->busiest_nr_running = sgs.sum_nr_running; 2582 sds->busiest_nr_running = sgs.sum_nr_running;
2559 sds->busiest_group_capacity = sgs.group_capacity; 2583 sds->busiest_group_capacity = sgs.group_capacity;
2560 sds->busiest_load_per_task = sgs.sum_weighted_load; 2584 sds->busiest_load_per_task = sgs.sum_weighted_load;
2585 sds->busiest_has_capacity = sgs.group_has_capacity;
2561 sds->group_imb = sgs.group_imb; 2586 sds->group_imb = sgs.group_imb;
2562 } 2587 }
2563 2588
@@ -2754,6 +2779,7 @@ static inline void calculate_imbalance(struct sd_lb_stats *sds, int this_cpu,
2754 return fix_small_imbalance(sds, this_cpu, imbalance); 2779 return fix_small_imbalance(sds, this_cpu, imbalance);
2755 2780
2756} 2781}
2782
2757/******* find_busiest_group() helpers end here *********************/ 2783/******* find_busiest_group() helpers end here *********************/
2758 2784
2759/** 2785/**
@@ -2805,6 +2831,11 @@ find_busiest_group(struct sched_domain *sd, int this_cpu,
2805 * 4) This group is more busy than the avg busieness at this 2831 * 4) This group is more busy than the avg busieness at this
2806 * sched_domain. 2832 * sched_domain.
2807 * 5) The imbalance is within the specified limit. 2833 * 5) The imbalance is within the specified limit.
2834 *
2835 * Note: when doing newidle balance, if the local group has excess
2836 * capacity (i.e. nr_running < group_capacity) and the busiest group
2837 * does not have any capacity, we force a load balance to pull tasks
2838 * to the local group. In this case, we skip past checks 3, 4 and 5.
2808 */ 2839 */
2809 if (!(*balance)) 2840 if (!(*balance))
2810 goto ret; 2841 goto ret;
@@ -2816,6 +2847,11 @@ find_busiest_group(struct sched_domain *sd, int this_cpu,
2816 if (!sds.busiest || sds.busiest_nr_running == 0) 2847 if (!sds.busiest || sds.busiest_nr_running == 0)
2817 goto out_balanced; 2848 goto out_balanced;
2818 2849
2850 /* SD_BALANCE_NEWIDLE trumps SMP nice when underutilized */
2851 if (idle == CPU_NEWLY_IDLE && sds.this_has_capacity &&
2852 !sds.busiest_has_capacity)
2853 goto force_balance;
2854
2819 if (sds.this_load >= sds.max_load) 2855 if (sds.this_load >= sds.max_load)
2820 goto out_balanced; 2856 goto out_balanced;
2821 2857
@@ -2827,6 +2863,7 @@ find_busiest_group(struct sched_domain *sd, int this_cpu,
2827 if (100 * sds.max_load <= sd->imbalance_pct * sds.this_load) 2863 if (100 * sds.max_load <= sd->imbalance_pct * sds.this_load)
2828 goto out_balanced; 2864 goto out_balanced;
2829 2865
2866force_balance:
2830 /* Looks like there is an imbalance. Compute it */ 2867 /* Looks like there is an imbalance. Compute it */
2831 calculate_imbalance(&sds, this_cpu, imbalance); 2868 calculate_imbalance(&sds, this_cpu, imbalance);
2832 return sds.busiest; 2869 return sds.busiest;
@@ -3031,7 +3068,14 @@ redo:
3031 3068
3032 if (!ld_moved) { 3069 if (!ld_moved) {
3033 schedstat_inc(sd, lb_failed[idle]); 3070 schedstat_inc(sd, lb_failed[idle]);
3034 sd->nr_balance_failed++; 3071 /*
3072 * Increment the failure counter only on periodic balance.
3073 * We do not want newidle balance, which can be very
3074 * frequent, pollute the failure counter causing
3075 * excessive cache_hot migrations and active balances.
3076 */
3077 if (idle != CPU_NEWLY_IDLE)
3078 sd->nr_balance_failed++;
3035 3079
3036 if (need_active_balance(sd, sd_idle, idle, cpu_of(busiest), 3080 if (need_active_balance(sd, sd_idle, idle, cpu_of(busiest),
3037 this_cpu)) { 3081 this_cpu)) {
@@ -3153,10 +3197,8 @@ static void idle_balance(int this_cpu, struct rq *this_rq)
3153 interval = msecs_to_jiffies(sd->balance_interval); 3197 interval = msecs_to_jiffies(sd->balance_interval);
3154 if (time_after(next_balance, sd->last_balance + interval)) 3198 if (time_after(next_balance, sd->last_balance + interval))
3155 next_balance = sd->last_balance + interval; 3199 next_balance = sd->last_balance + interval;
3156 if (pulled_task) { 3200 if (pulled_task)
3157 this_rq->idle_stamp = 0;
3158 break; 3201 break;
3159 }
3160 } 3202 }
3161 3203
3162 raw_spin_lock(&this_rq->lock); 3204 raw_spin_lock(&this_rq->lock);
@@ -3751,8 +3793,11 @@ static void task_fork_fair(struct task_struct *p)
3751 3793
3752 update_rq_clock(rq); 3794 update_rq_clock(rq);
3753 3795
3754 if (unlikely(task_cpu(p) != this_cpu)) 3796 if (unlikely(task_cpu(p) != this_cpu)) {
3797 rcu_read_lock();
3755 __set_task_cpu(p, this_cpu); 3798 __set_task_cpu(p, this_cpu);
3799 rcu_read_unlock();
3800 }
3756 3801
3757 update_curr(cfs_rq); 3802 update_curr(cfs_rq);
3758 3803
diff --git a/kernel/sched_features.h b/kernel/sched_features.h
index 83c66e8ad3ee..185f920ec1a2 100644
--- a/kernel/sched_features.h
+++ b/kernel/sched_features.h
@@ -61,3 +61,8 @@ SCHED_FEAT(ASYM_EFF_LOAD, 1)
61 * release the lock. Decreases scheduling overhead. 61 * release the lock. Decreases scheduling overhead.
62 */ 62 */
63SCHED_FEAT(OWNER_SPIN, 1) 63SCHED_FEAT(OWNER_SPIN, 1)
64
65/*
66 * Decrement CPU power based on irq activity
67 */
68SCHED_FEAT(NONIRQ_POWER, 1)
diff --git a/kernel/sched_rt.c b/kernel/sched_rt.c
index d10c80ebb67a..bea7d79f7e9c 100644
--- a/kernel/sched_rt.c
+++ b/kernel/sched_rt.c
@@ -609,7 +609,7 @@ static void update_curr_rt(struct rq *rq)
609 if (!task_has_rt_policy(curr)) 609 if (!task_has_rt_policy(curr))
610 return; 610 return;
611 611
612 delta_exec = rq->clock - curr->se.exec_start; 612 delta_exec = rq->clock_task - curr->se.exec_start;
613 if (unlikely((s64)delta_exec < 0)) 613 if (unlikely((s64)delta_exec < 0))
614 delta_exec = 0; 614 delta_exec = 0;
615 615
@@ -618,7 +618,7 @@ static void update_curr_rt(struct rq *rq)
618 curr->se.sum_exec_runtime += delta_exec; 618 curr->se.sum_exec_runtime += delta_exec;
619 account_group_exec_runtime(curr, delta_exec); 619 account_group_exec_runtime(curr, delta_exec);
620 620
621 curr->se.exec_start = rq->clock; 621 curr->se.exec_start = rq->clock_task;
622 cpuacct_charge(curr, delta_exec); 622 cpuacct_charge(curr, delta_exec);
623 623
624 sched_rt_avg_update(rq, delta_exec); 624 sched_rt_avg_update(rq, delta_exec);
@@ -960,18 +960,19 @@ select_task_rq_rt(struct rq *rq, struct task_struct *p, int sd_flag, int flags)
960 * runqueue. Otherwise simply start this RT task 960 * runqueue. Otherwise simply start this RT task
961 * on its current runqueue. 961 * on its current runqueue.
962 * 962 *
963 * We want to avoid overloading runqueues. Even if 963 * We want to avoid overloading runqueues. If the woken
964 * the RT task is of higher priority than the current RT task. 964 * task is a higher priority, then it will stay on this CPU
965 * RT tasks behave differently than other tasks. If 965 * and the lower prio task should be moved to another CPU.
966 * one gets preempted, we try to push it off to another queue. 966 * Even though this will probably make the lower prio task
967 * So trying to keep a preempting RT task on the same 967 * lose its cache, we do not want to bounce a higher task
968 * cache hot CPU will force the running RT task to 968 * around just because it gave up its CPU, perhaps for a
969 * a cold CPU. So we waste all the cache for the lower 969 * lock?
970 * RT task in hopes of saving some of a RT task 970 *
971 * that is just being woken and probably will have 971 * For equal prio tasks, we just let the scheduler sort it out.
972 * cold cache anyway.
973 */ 972 */
974 if (unlikely(rt_task(rq->curr)) && 973 if (unlikely(rt_task(rq->curr)) &&
974 (rq->curr->rt.nr_cpus_allowed < 2 ||
975 rq->curr->prio < p->prio) &&
975 (p->rt.nr_cpus_allowed > 1)) { 976 (p->rt.nr_cpus_allowed > 1)) {
976 int cpu = find_lowest_rq(p); 977 int cpu = find_lowest_rq(p);
977 978
@@ -1074,7 +1075,7 @@ static struct task_struct *_pick_next_task_rt(struct rq *rq)
1074 } while (rt_rq); 1075 } while (rt_rq);
1075 1076
1076 p = rt_task_of(rt_se); 1077 p = rt_task_of(rt_se);
1077 p->se.exec_start = rq->clock; 1078 p->se.exec_start = rq->clock_task;
1078 1079
1079 return p; 1080 return p;
1080} 1081}
@@ -1139,7 +1140,7 @@ static struct task_struct *pick_next_highest_task_rt(struct rq *rq, int cpu)
1139 for_each_leaf_rt_rq(rt_rq, rq) { 1140 for_each_leaf_rt_rq(rt_rq, rq) {
1140 array = &rt_rq->active; 1141 array = &rt_rq->active;
1141 idx = sched_find_first_bit(array->bitmap); 1142 idx = sched_find_first_bit(array->bitmap);
1142 next_idx: 1143next_idx:
1143 if (idx >= MAX_RT_PRIO) 1144 if (idx >= MAX_RT_PRIO)
1144 continue; 1145 continue;
1145 if (next && next->prio < idx) 1146 if (next && next->prio < idx)
@@ -1315,7 +1316,7 @@ static int push_rt_task(struct rq *rq)
1315 if (!next_task) 1316 if (!next_task)
1316 return 0; 1317 return 0;
1317 1318
1318 retry: 1319retry:
1319 if (unlikely(next_task == rq->curr)) { 1320 if (unlikely(next_task == rq->curr)) {
1320 WARN_ON(1); 1321 WARN_ON(1);
1321 return 0; 1322 return 0;
@@ -1463,7 +1464,7 @@ static int pull_rt_task(struct rq *this_rq)
1463 * but possible) 1464 * but possible)
1464 */ 1465 */
1465 } 1466 }
1466 skip: 1467skip:
1467 double_unlock_balance(this_rq, src_rq); 1468 double_unlock_balance(this_rq, src_rq);
1468 } 1469 }
1469 1470
@@ -1491,7 +1492,10 @@ static void task_woken_rt(struct rq *rq, struct task_struct *p)
1491 if (!task_running(rq, p) && 1492 if (!task_running(rq, p) &&
1492 !test_tsk_need_resched(rq->curr) && 1493 !test_tsk_need_resched(rq->curr) &&
1493 has_pushable_tasks(rq) && 1494 has_pushable_tasks(rq) &&
1494 p->rt.nr_cpus_allowed > 1) 1495 p->rt.nr_cpus_allowed > 1 &&
1496 rt_task(rq->curr) &&
1497 (rq->curr->rt.nr_cpus_allowed < 2 ||
1498 rq->curr->prio < p->prio))
1495 push_rt_tasks(rq); 1499 push_rt_tasks(rq);
1496} 1500}
1497 1501
@@ -1709,7 +1713,7 @@ static void set_curr_task_rt(struct rq *rq)
1709{ 1713{
1710 struct task_struct *p = rq->curr; 1714 struct task_struct *p = rq->curr;
1711 1715
1712 p->se.exec_start = rq->clock; 1716 p->se.exec_start = rq->clock_task;
1713 1717
1714 /* The running task is never eligible for pushing */ 1718 /* The running task is never eligible for pushing */
1715 dequeue_pushable_task(rq, p); 1719 dequeue_pushable_task(rq, p);
diff --git a/kernel/sched_stoptask.c b/kernel/sched_stoptask.c
new file mode 100644
index 000000000000..45bddc0c1048
--- /dev/null
+++ b/kernel/sched_stoptask.c
@@ -0,0 +1,108 @@
1/*
2 * stop-task scheduling class.
3 *
4 * The stop task is the highest priority task in the system, it preempts
5 * everything and will be preempted by nothing.
6 *
7 * See kernel/stop_machine.c
8 */
9
10#ifdef CONFIG_SMP
11static int
12select_task_rq_stop(struct rq *rq, struct task_struct *p,
13 int sd_flag, int flags)
14{
15 return task_cpu(p); /* stop tasks as never migrate */
16}
17#endif /* CONFIG_SMP */
18
19static void
20check_preempt_curr_stop(struct rq *rq, struct task_struct *p, int flags)
21{
22 resched_task(rq->curr); /* we preempt everything */
23}
24
25static struct task_struct *pick_next_task_stop(struct rq *rq)
26{
27 struct task_struct *stop = rq->stop;
28
29 if (stop && stop->state == TASK_RUNNING)
30 return stop;
31
32 return NULL;
33}
34
35static void
36enqueue_task_stop(struct rq *rq, struct task_struct *p, int flags)
37{
38}
39
40static void
41dequeue_task_stop(struct rq *rq, struct task_struct *p, int flags)
42{
43}
44
45static void yield_task_stop(struct rq *rq)
46{
47 BUG(); /* the stop task should never yield, its pointless. */
48}
49
50static void put_prev_task_stop(struct rq *rq, struct task_struct *prev)
51{
52}
53
54static void task_tick_stop(struct rq *rq, struct task_struct *curr, int queued)
55{
56}
57
58static void set_curr_task_stop(struct rq *rq)
59{
60}
61
62static void switched_to_stop(struct rq *rq, struct task_struct *p,
63 int running)
64{
65 BUG(); /* its impossible to change to this class */
66}
67
68static void prio_changed_stop(struct rq *rq, struct task_struct *p,
69 int oldprio, int running)
70{
71 BUG(); /* how!?, what priority? */
72}
73
74static unsigned int
75get_rr_interval_stop(struct rq *rq, struct task_struct *task)
76{
77 return 0;
78}
79
80/*
81 * Simple, special scheduling class for the per-CPU stop tasks:
82 */
83static const struct sched_class stop_sched_class = {
84 .next = &rt_sched_class,
85
86 .enqueue_task = enqueue_task_stop,
87 .dequeue_task = dequeue_task_stop,
88 .yield_task = yield_task_stop,
89
90 .check_preempt_curr = check_preempt_curr_stop,
91
92 .pick_next_task = pick_next_task_stop,
93 .put_prev_task = put_prev_task_stop,
94
95#ifdef CONFIG_SMP
96 .select_task_rq = select_task_rq_stop,
97#endif
98
99 .set_curr_task = set_curr_task_stop,
100 .task_tick = task_tick_stop,
101
102 .get_rr_interval = get_rr_interval_stop,
103
104 .prio_changed = prio_changed_stop,
105 .switched_to = switched_to_stop,
106
107 /* no .task_new for stop tasks */
108};
diff --git a/kernel/signal.c b/kernel/signal.c
index bded65187780..919562c3d6b7 100644
--- a/kernel/signal.c
+++ b/kernel/signal.c
@@ -2215,6 +2215,14 @@ int copy_siginfo_to_user(siginfo_t __user *to, siginfo_t *from)
2215#ifdef __ARCH_SI_TRAPNO 2215#ifdef __ARCH_SI_TRAPNO
2216 err |= __put_user(from->si_trapno, &to->si_trapno); 2216 err |= __put_user(from->si_trapno, &to->si_trapno);
2217#endif 2217#endif
2218#ifdef BUS_MCEERR_AO
2219 /*
2220 * Other callers might not initialize the si_lsb field,
2221 * so check explicitely for the right codes here.
2222 */
2223 if (from->si_code == BUS_MCEERR_AR || from->si_code == BUS_MCEERR_AO)
2224 err |= __put_user(from->si_addr_lsb, &to->si_addr_lsb);
2225#endif
2218 break; 2226 break;
2219 case __SI_CHLD: 2227 case __SI_CHLD:
2220 err |= __put_user(from->si_pid, &to->si_pid); 2228 err |= __put_user(from->si_pid, &to->si_pid);
diff --git a/kernel/smp.c b/kernel/smp.c
index 75c970c715d3..ed6aacfcb7ef 100644
--- a/kernel/smp.c
+++ b/kernel/smp.c
@@ -365,9 +365,10 @@ call:
365EXPORT_SYMBOL_GPL(smp_call_function_any); 365EXPORT_SYMBOL_GPL(smp_call_function_any);
366 366
367/** 367/**
368 * __smp_call_function_single(): Run a function on another CPU 368 * __smp_call_function_single(): Run a function on a specific CPU
369 * @cpu: The CPU to run on. 369 * @cpu: The CPU to run on.
370 * @data: Pre-allocated and setup data structure 370 * @data: Pre-allocated and setup data structure
371 * @wait: If true, wait until function has completed on specified CPU.
371 * 372 *
372 * Like smp_call_function_single(), but allow caller to pass in a 373 * Like smp_call_function_single(), but allow caller to pass in a
373 * pre-allocated data structure. Useful for embedding @data inside 374 * pre-allocated data structure. Useful for embedding @data inside
@@ -376,8 +377,10 @@ EXPORT_SYMBOL_GPL(smp_call_function_any);
376void __smp_call_function_single(int cpu, struct call_single_data *data, 377void __smp_call_function_single(int cpu, struct call_single_data *data,
377 int wait) 378 int wait)
378{ 379{
379 csd_lock(data); 380 unsigned int this_cpu;
381 unsigned long flags;
380 382
383 this_cpu = get_cpu();
381 /* 384 /*
382 * Can deadlock when called with interrupts disabled. 385 * Can deadlock when called with interrupts disabled.
383 * We allow cpu's that are not yet online though, as no one else can 386 * We allow cpu's that are not yet online though, as no one else can
@@ -387,7 +390,15 @@ void __smp_call_function_single(int cpu, struct call_single_data *data,
387 WARN_ON_ONCE(cpu_online(smp_processor_id()) && wait && irqs_disabled() 390 WARN_ON_ONCE(cpu_online(smp_processor_id()) && wait && irqs_disabled()
388 && !oops_in_progress); 391 && !oops_in_progress);
389 392
390 generic_exec_single(cpu, data, wait); 393 if (cpu == this_cpu) {
394 local_irq_save(flags);
395 data->func(data->info);
396 local_irq_restore(flags);
397 } else {
398 csd_lock(data);
399 generic_exec_single(cpu, data, wait);
400 }
401 put_cpu();
391} 402}
392 403
393/** 404/**
diff --git a/kernel/softirq.c b/kernel/softirq.c
index 07b4f1b1a73a..f02a9dfa19bc 100644
--- a/kernel/softirq.c
+++ b/kernel/softirq.c
@@ -67,7 +67,7 @@ char *softirq_to_name[NR_SOFTIRQS] = {
67 * to the pending events, so lets the scheduler to balance 67 * to the pending events, so lets the scheduler to balance
68 * the softirq load for us. 68 * the softirq load for us.
69 */ 69 */
70void wakeup_softirqd(void) 70static void wakeup_softirqd(void)
71{ 71{
72 /* Interrupts are disabled: no need to stop preemption */ 72 /* Interrupts are disabled: no need to stop preemption */
73 struct task_struct *tsk = __get_cpu_var(ksoftirqd); 73 struct task_struct *tsk = __get_cpu_var(ksoftirqd);
@@ -77,11 +77,21 @@ void wakeup_softirqd(void)
77} 77}
78 78
79/* 79/*
80 * preempt_count and SOFTIRQ_OFFSET usage:
81 * - preempt_count is changed by SOFTIRQ_OFFSET on entering or leaving
82 * softirq processing.
83 * - preempt_count is changed by SOFTIRQ_DISABLE_OFFSET (= 2 * SOFTIRQ_OFFSET)
84 * on local_bh_disable or local_bh_enable.
85 * This lets us distinguish between whether we are currently processing
86 * softirq and whether we just have bh disabled.
87 */
88
89/*
80 * This one is for softirq.c-internal use, 90 * This one is for softirq.c-internal use,
81 * where hardirqs are disabled legitimately: 91 * where hardirqs are disabled legitimately:
82 */ 92 */
83#ifdef CONFIG_TRACE_IRQFLAGS 93#ifdef CONFIG_TRACE_IRQFLAGS
84static void __local_bh_disable(unsigned long ip) 94static void __local_bh_disable(unsigned long ip, unsigned int cnt)
85{ 95{
86 unsigned long flags; 96 unsigned long flags;
87 97
@@ -95,32 +105,43 @@ static void __local_bh_disable(unsigned long ip)
95 * We must manually increment preempt_count here and manually 105 * We must manually increment preempt_count here and manually
96 * call the trace_preempt_off later. 106 * call the trace_preempt_off later.
97 */ 107 */
98 preempt_count() += SOFTIRQ_OFFSET; 108 preempt_count() += cnt;
99 /* 109 /*
100 * Were softirqs turned off above: 110 * Were softirqs turned off above:
101 */ 111 */
102 if (softirq_count() == SOFTIRQ_OFFSET) 112 if (softirq_count() == cnt)
103 trace_softirqs_off(ip); 113 trace_softirqs_off(ip);
104 raw_local_irq_restore(flags); 114 raw_local_irq_restore(flags);
105 115
106 if (preempt_count() == SOFTIRQ_OFFSET) 116 if (preempt_count() == cnt)
107 trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1)); 117 trace_preempt_off(CALLER_ADDR0, get_parent_ip(CALLER_ADDR1));
108} 118}
109#else /* !CONFIG_TRACE_IRQFLAGS */ 119#else /* !CONFIG_TRACE_IRQFLAGS */
110static inline void __local_bh_disable(unsigned long ip) 120static inline void __local_bh_disable(unsigned long ip, unsigned int cnt)
111{ 121{
112 add_preempt_count(SOFTIRQ_OFFSET); 122 add_preempt_count(cnt);
113 barrier(); 123 barrier();
114} 124}
115#endif /* CONFIG_TRACE_IRQFLAGS */ 125#endif /* CONFIG_TRACE_IRQFLAGS */
116 126
117void local_bh_disable(void) 127void local_bh_disable(void)
118{ 128{
119 __local_bh_disable((unsigned long)__builtin_return_address(0)); 129 __local_bh_disable((unsigned long)__builtin_return_address(0),
130 SOFTIRQ_DISABLE_OFFSET);
120} 131}
121 132
122EXPORT_SYMBOL(local_bh_disable); 133EXPORT_SYMBOL(local_bh_disable);
123 134
135static void __local_bh_enable(unsigned int cnt)
136{
137 WARN_ON_ONCE(in_irq());
138 WARN_ON_ONCE(!irqs_disabled());
139
140 if (softirq_count() == cnt)
141 trace_softirqs_on((unsigned long)__builtin_return_address(0));
142 sub_preempt_count(cnt);
143}
144
124/* 145/*
125 * Special-case - softirqs can safely be enabled in 146 * Special-case - softirqs can safely be enabled in
126 * cond_resched_softirq(), or by __do_softirq(), 147 * cond_resched_softirq(), or by __do_softirq(),
@@ -128,12 +149,7 @@ EXPORT_SYMBOL(local_bh_disable);
128 */ 149 */
129void _local_bh_enable(void) 150void _local_bh_enable(void)
130{ 151{
131 WARN_ON_ONCE(in_irq()); 152 __local_bh_enable(SOFTIRQ_DISABLE_OFFSET);
132 WARN_ON_ONCE(!irqs_disabled());
133
134 if (softirq_count() == SOFTIRQ_OFFSET)
135 trace_softirqs_on((unsigned long)__builtin_return_address(0));
136 sub_preempt_count(SOFTIRQ_OFFSET);
137} 153}
138 154
139EXPORT_SYMBOL(_local_bh_enable); 155EXPORT_SYMBOL(_local_bh_enable);
@@ -147,13 +163,13 @@ static inline void _local_bh_enable_ip(unsigned long ip)
147 /* 163 /*
148 * Are softirqs going to be turned on now: 164 * Are softirqs going to be turned on now:
149 */ 165 */
150 if (softirq_count() == SOFTIRQ_OFFSET) 166 if (softirq_count() == SOFTIRQ_DISABLE_OFFSET)
151 trace_softirqs_on(ip); 167 trace_softirqs_on(ip);
152 /* 168 /*
153 * Keep preemption disabled until we are done with 169 * Keep preemption disabled until we are done with
154 * softirq processing: 170 * softirq processing:
155 */ 171 */
156 sub_preempt_count(SOFTIRQ_OFFSET - 1); 172 sub_preempt_count(SOFTIRQ_DISABLE_OFFSET - 1);
157 173
158 if (unlikely(!in_interrupt() && local_softirq_pending())) 174 if (unlikely(!in_interrupt() && local_softirq_pending()))
159 do_softirq(); 175 do_softirq();
@@ -198,7 +214,8 @@ asmlinkage void __do_softirq(void)
198 pending = local_softirq_pending(); 214 pending = local_softirq_pending();
199 account_system_vtime(current); 215 account_system_vtime(current);
200 216
201 __local_bh_disable((unsigned long)__builtin_return_address(0)); 217 __local_bh_disable((unsigned long)__builtin_return_address(0),
218 SOFTIRQ_OFFSET);
202 lockdep_softirq_enter(); 219 lockdep_softirq_enter();
203 220
204 cpu = smp_processor_id(); 221 cpu = smp_processor_id();
@@ -245,7 +262,7 @@ restart:
245 lockdep_softirq_exit(); 262 lockdep_softirq_exit();
246 263
247 account_system_vtime(current); 264 account_system_vtime(current);
248 _local_bh_enable(); 265 __local_bh_enable(SOFTIRQ_OFFSET);
249} 266}
250 267
251#ifndef __ARCH_HAS_DO_SOFTIRQ 268#ifndef __ARCH_HAS_DO_SOFTIRQ
@@ -279,10 +296,16 @@ void irq_enter(void)
279 296
280 rcu_irq_enter(); 297 rcu_irq_enter();
281 if (idle_cpu(cpu) && !in_interrupt()) { 298 if (idle_cpu(cpu) && !in_interrupt()) {
282 __irq_enter(); 299 /*
300 * Prevent raise_softirq from needlessly waking up ksoftirqd
301 * here, as softirq will be serviced on return from interrupt.
302 */
303 local_bh_disable();
283 tick_check_idle(cpu); 304 tick_check_idle(cpu);
284 } else 305 _local_bh_enable();
285 __irq_enter(); 306 }
307
308 __irq_enter();
286} 309}
287 310
288#ifdef __ARCH_IRQ_EXIT_IRQS_DISABLED 311#ifdef __ARCH_IRQ_EXIT_IRQS_DISABLED
@@ -696,6 +719,7 @@ static int run_ksoftirqd(void * __bind_cpu)
696{ 719{
697 set_current_state(TASK_INTERRUPTIBLE); 720 set_current_state(TASK_INTERRUPTIBLE);
698 721
722 current->flags |= PF_KSOFTIRQD;
699 while (!kthread_should_stop()) { 723 while (!kthread_should_stop()) {
700 preempt_disable(); 724 preempt_disable();
701 if (!local_softirq_pending()) { 725 if (!local_softirq_pending()) {
@@ -886,17 +910,14 @@ int __init __weak early_irq_init(void)
886 return 0; 910 return 0;
887} 911}
888 912
913#ifdef CONFIG_GENERIC_HARDIRQS
889int __init __weak arch_probe_nr_irqs(void) 914int __init __weak arch_probe_nr_irqs(void)
890{ 915{
891 return 0; 916 return NR_IRQS_LEGACY;
892} 917}
893 918
894int __init __weak arch_early_irq_init(void) 919int __init __weak arch_early_irq_init(void)
895{ 920{
896 return 0; 921 return 0;
897} 922}
898 923#endif
899int __weak arch_init_chip_data(struct irq_desc *desc, int node)
900{
901 return 0;
902}
diff --git a/kernel/srcu.c b/kernel/srcu.c
index 2980da3fd509..c71e07500536 100644
--- a/kernel/srcu.c
+++ b/kernel/srcu.c
@@ -46,11 +46,9 @@ static int init_srcu_struct_fields(struct srcu_struct *sp)
46int __init_srcu_struct(struct srcu_struct *sp, const char *name, 46int __init_srcu_struct(struct srcu_struct *sp, const char *name,
47 struct lock_class_key *key) 47 struct lock_class_key *key)
48{ 48{
49#ifdef CONFIG_DEBUG_LOCK_ALLOC
50 /* Don't re-initialize a lock while it is held. */ 49 /* Don't re-initialize a lock while it is held. */
51 debug_check_no_locks_freed((void *)sp, sizeof(*sp)); 50 debug_check_no_locks_freed((void *)sp, sizeof(*sp));
52 lockdep_init_map(&sp->dep_map, name, key, 0); 51 lockdep_init_map(&sp->dep_map, name, key, 0);
53#endif /* #ifdef CONFIG_DEBUG_LOCK_ALLOC */
54 return init_srcu_struct_fields(sp); 52 return init_srcu_struct_fields(sp);
55} 53}
56EXPORT_SYMBOL_GPL(__init_srcu_struct); 54EXPORT_SYMBOL_GPL(__init_srcu_struct);
diff --git a/kernel/stop_machine.c b/kernel/stop_machine.c
index 4372ccb25127..090c28812ce1 100644
--- a/kernel/stop_machine.c
+++ b/kernel/stop_machine.c
@@ -287,11 +287,12 @@ repeat:
287 goto repeat; 287 goto repeat;
288} 288}
289 289
290extern void sched_set_stop_task(int cpu, struct task_struct *stop);
291
290/* manage stopper for a cpu, mostly lifted from sched migration thread mgmt */ 292/* manage stopper for a cpu, mostly lifted from sched migration thread mgmt */
291static int __cpuinit cpu_stop_cpu_callback(struct notifier_block *nfb, 293static int __cpuinit cpu_stop_cpu_callback(struct notifier_block *nfb,
292 unsigned long action, void *hcpu) 294 unsigned long action, void *hcpu)
293{ 295{
294 struct sched_param param = { .sched_priority = MAX_RT_PRIO - 1 };
295 unsigned int cpu = (unsigned long)hcpu; 296 unsigned int cpu = (unsigned long)hcpu;
296 struct cpu_stopper *stopper = &per_cpu(cpu_stopper, cpu); 297 struct cpu_stopper *stopper = &per_cpu(cpu_stopper, cpu);
297 struct task_struct *p; 298 struct task_struct *p;
@@ -304,13 +305,13 @@ static int __cpuinit cpu_stop_cpu_callback(struct notifier_block *nfb,
304 cpu); 305 cpu);
305 if (IS_ERR(p)) 306 if (IS_ERR(p))
306 return NOTIFY_BAD; 307 return NOTIFY_BAD;
307 sched_setscheduler_nocheck(p, SCHED_FIFO, &param);
308 get_task_struct(p); 308 get_task_struct(p);
309 kthread_bind(p, cpu);
310 sched_set_stop_task(cpu, p);
309 stopper->thread = p; 311 stopper->thread = p;
310 break; 312 break;
311 313
312 case CPU_ONLINE: 314 case CPU_ONLINE:
313 kthread_bind(stopper->thread, cpu);
314 /* strictly unnecessary, as first user will wake it */ 315 /* strictly unnecessary, as first user will wake it */
315 wake_up_process(stopper->thread); 316 wake_up_process(stopper->thread);
316 /* mark enabled */ 317 /* mark enabled */
@@ -325,6 +326,7 @@ static int __cpuinit cpu_stop_cpu_callback(struct notifier_block *nfb,
325 { 326 {
326 struct cpu_stop_work *work; 327 struct cpu_stop_work *work;
327 328
329 sched_set_stop_task(cpu, NULL);
328 /* kill the stopper */ 330 /* kill the stopper */
329 kthread_stop(stopper->thread); 331 kthread_stop(stopper->thread);
330 /* drain remaining works */ 332 /* drain remaining works */
diff --git a/kernel/sys_ni.c b/kernel/sys_ni.c
index bad369ec5403..c782fe9924c7 100644
--- a/kernel/sys_ni.c
+++ b/kernel/sys_ni.c
@@ -50,6 +50,7 @@ cond_syscall(compat_sys_sendmsg);
50cond_syscall(sys_recvmsg); 50cond_syscall(sys_recvmsg);
51cond_syscall(sys_recvmmsg); 51cond_syscall(sys_recvmmsg);
52cond_syscall(compat_sys_recvmsg); 52cond_syscall(compat_sys_recvmsg);
53cond_syscall(compat_sys_recv);
53cond_syscall(compat_sys_recvfrom); 54cond_syscall(compat_sys_recvfrom);
54cond_syscall(compat_sys_recvmmsg); 55cond_syscall(compat_sys_recvmmsg);
55cond_syscall(sys_socketcall); 56cond_syscall(sys_socketcall);
diff --git a/kernel/sysctl.c b/kernel/sysctl.c
index f88552c6d227..3a45c224770f 100644
--- a/kernel/sysctl.c
+++ b/kernel/sysctl.c
@@ -2485,7 +2485,7 @@ static int __do_proc_doulongvec_minmax(void *data, struct ctl_table *table, int
2485 kbuf[left] = 0; 2485 kbuf[left] = 0;
2486 } 2486 }
2487 2487
2488 for (; left && vleft--; i++, min++, max++, first=0) { 2488 for (; left && vleft--; i++, first = 0) {
2489 unsigned long val; 2489 unsigned long val;
2490 2490
2491 if (write) { 2491 if (write) {
diff --git a/kernel/sysctl_check.c b/kernel/sysctl_check.c
index 04cdcf72c827..10b90d8a03c4 100644
--- a/kernel/sysctl_check.c
+++ b/kernel/sysctl_check.c
@@ -143,15 +143,6 @@ int sysctl_check_table(struct nsproxy *namespaces, struct ctl_table *table)
143 if (!table->maxlen) 143 if (!table->maxlen)
144 set_fail(&fail, table, "No maxlen"); 144 set_fail(&fail, table, "No maxlen");
145 } 145 }
146 if ((table->proc_handler == proc_doulongvec_minmax) ||
147 (table->proc_handler == proc_doulongvec_ms_jiffies_minmax)) {
148 if (table->maxlen > sizeof (unsigned long)) {
149 if (!table->extra1)
150 set_fail(&fail, table, "No min");
151 if (!table->extra2)
152 set_fail(&fail, table, "No max");
153 }
154 }
155#ifdef CONFIG_PROC_SYSCTL 146#ifdef CONFIG_PROC_SYSCTL
156 if (table->procname && !table->proc_handler) 147 if (table->procname && !table->proc_handler)
157 set_fail(&fail, table, "No proc_handler"); 148 set_fail(&fail, table, "No proc_handler");
diff --git a/kernel/test_kprobes.c b/kernel/test_kprobes.c
index 4f104515a19b..f8b11a283171 100644
--- a/kernel/test_kprobes.c
+++ b/kernel/test_kprobes.c
@@ -115,7 +115,9 @@ static int test_kprobes(void)
115 int ret; 115 int ret;
116 struct kprobe *kps[2] = {&kp, &kp2}; 116 struct kprobe *kps[2] = {&kp, &kp2};
117 117
118 kp.addr = 0; /* addr should be cleard for reusing kprobe. */ 118 /* addr and flags should be cleard for reusing kprobe. */
119 kp.addr = NULL;
120 kp.flags = 0;
119 ret = register_kprobes(kps, 2); 121 ret = register_kprobes(kps, 2);
120 if (ret < 0) { 122 if (ret < 0) {
121 printk(KERN_ERR "Kprobe smoke test failed: " 123 printk(KERN_ERR "Kprobe smoke test failed: "
@@ -210,7 +212,9 @@ static int test_jprobes(void)
210 int ret; 212 int ret;
211 struct jprobe *jps[2] = {&jp, &jp2}; 213 struct jprobe *jps[2] = {&jp, &jp2};
212 214
213 jp.kp.addr = 0; /* addr should be cleard for reusing kprobe. */ 215 /* addr and flags should be cleard for reusing kprobe. */
216 jp.kp.addr = NULL;
217 jp.kp.flags = 0;
214 ret = register_jprobes(jps, 2); 218 ret = register_jprobes(jps, 2);
215 if (ret < 0) { 219 if (ret < 0) {
216 printk(KERN_ERR "Kprobe smoke test failed: " 220 printk(KERN_ERR "Kprobe smoke test failed: "
@@ -323,7 +327,9 @@ static int test_kretprobes(void)
323 int ret; 327 int ret;
324 struct kretprobe *rps[2] = {&rp, &rp2}; 328 struct kretprobe *rps[2] = {&rp, &rp2};
325 329
326 rp.kp.addr = 0; /* addr should be cleard for reusing kprobe. */ 330 /* addr and flags should be cleard for reusing kprobe. */
331 rp.kp.addr = NULL;
332 rp.kp.flags = 0;
327 ret = register_kretprobes(rps, 2); 333 ret = register_kretprobes(rps, 2);
328 if (ret < 0) { 334 if (ret < 0) {
329 printk(KERN_ERR "Kprobe smoke test failed: " 335 printk(KERN_ERR "Kprobe smoke test failed: "
diff --git a/kernel/time/ntp.c b/kernel/time/ntp.c
index c63116863a80..d2321891538f 100644
--- a/kernel/time/ntp.c
+++ b/kernel/time/ntp.c
@@ -149,10 +149,18 @@ static void ntp_update_offset(long offset)
149 time_reftime = get_seconds(); 149 time_reftime = get_seconds();
150 150
151 offset64 = offset; 151 offset64 = offset;
152 freq_adj = (offset64 * secs) << 152 freq_adj = ntp_update_offset_fll(offset64, secs);
153 (NTP_SCALE_SHIFT - 2 * (SHIFT_PLL + 2 + time_constant));
154 153
155 freq_adj += ntp_update_offset_fll(offset64, secs); 154 /*
155 * Clamp update interval to reduce PLL gain with low
156 * sampling rate (e.g. intermittent network connection)
157 * to avoid instability.
158 */
159 if (unlikely(secs > 1 << (SHIFT_PLL + 1 + time_constant)))
160 secs = 1 << (SHIFT_PLL + 1 + time_constant);
161
162 freq_adj += (offset64 * secs) <<
163 (NTP_SCALE_SHIFT - 2 * (SHIFT_PLL + 2 + time_constant));
156 164
157 freq_adj = min(freq_adj + time_freq, MAXFREQ_SCALED); 165 freq_adj = min(freq_adj + time_freq, MAXFREQ_SCALED);
158 166
diff --git a/kernel/timer.c b/kernel/timer.c
index 97bf05baade7..68a9ae7679b7 100644
--- a/kernel/timer.c
+++ b/kernel/timer.c
@@ -37,7 +37,7 @@
37#include <linux/delay.h> 37#include <linux/delay.h>
38#include <linux/tick.h> 38#include <linux/tick.h>
39#include <linux/kallsyms.h> 39#include <linux/kallsyms.h>
40#include <linux/perf_event.h> 40#include <linux/irq_work.h>
41#include <linux/sched.h> 41#include <linux/sched.h>
42#include <linux/slab.h> 42#include <linux/slab.h>
43 43
@@ -1279,7 +1279,10 @@ void update_process_times(int user_tick)
1279 run_local_timers(); 1279 run_local_timers();
1280 rcu_check_callbacks(cpu, user_tick); 1280 rcu_check_callbacks(cpu, user_tick);
1281 printk_tick(); 1281 printk_tick();
1282 perf_event_do_pending(); 1282#ifdef CONFIG_IRQ_WORK
1283 if (in_irq())
1284 irq_work_run();
1285#endif
1283 scheduler_tick(); 1286 scheduler_tick();
1284 run_posix_cpu_timers(p); 1287 run_posix_cpu_timers(p);
1285} 1288}
diff --git a/kernel/trace/Kconfig b/kernel/trace/Kconfig
index 538501c6ea50..e04b8bcdef88 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_C_RECORDMCOUNT
53 bool
54 help
55 C version of recordmcount available?
56
52config TRACER_MAX_TRACE 57config TRACER_MAX_TRACE
53 bool 58 bool
54 59
@@ -121,7 +126,7 @@ if FTRACE
121config FUNCTION_TRACER 126config FUNCTION_TRACER
122 bool "Kernel Function Tracer" 127 bool "Kernel Function Tracer"
123 depends on HAVE_FUNCTION_TRACER 128 depends on HAVE_FUNCTION_TRACER
124 select FRAME_POINTER 129 select FRAME_POINTER if (!ARM_UNWIND)
125 select KALLSYMS 130 select KALLSYMS
126 select GENERIC_TRACER 131 select GENERIC_TRACER
127 select CONTEXT_SWITCH_TRACER 132 select CONTEXT_SWITCH_TRACER
diff --git a/kernel/trace/blktrace.c b/kernel/trace/blktrace.c
index 959f8d6c8cc1..bc251ed66724 100644
--- a/kernel/trace/blktrace.c
+++ b/kernel/trace/blktrace.c
@@ -23,7 +23,6 @@
23#include <linux/mutex.h> 23#include <linux/mutex.h>
24#include <linux/slab.h> 24#include <linux/slab.h>
25#include <linux/debugfs.h> 25#include <linux/debugfs.h>
26#include <linux/smp_lock.h>
27#include <linux/time.h> 26#include <linux/time.h>
28#include <linux/uaccess.h> 27#include <linux/uaccess.h>
29 28
@@ -326,6 +325,7 @@ static const struct file_operations blk_dropped_fops = {
326 .owner = THIS_MODULE, 325 .owner = THIS_MODULE,
327 .open = blk_dropped_open, 326 .open = blk_dropped_open,
328 .read = blk_dropped_read, 327 .read = blk_dropped_read,
328 .llseek = default_llseek,
329}; 329};
330 330
331static int blk_msg_open(struct inode *inode, struct file *filp) 331static int blk_msg_open(struct inode *inode, struct file *filp)
@@ -365,6 +365,7 @@ static const struct file_operations blk_msg_fops = {
365 .owner = THIS_MODULE, 365 .owner = THIS_MODULE,
366 .open = blk_msg_open, 366 .open = blk_msg_open,
367 .write = blk_msg_write, 367 .write = blk_msg_write,
368 .llseek = noop_llseek,
368}; 369};
369 370
370/* 371/*
@@ -639,7 +640,6 @@ int blk_trace_ioctl(struct block_device *bdev, unsigned cmd, char __user *arg)
639 if (!q) 640 if (!q)
640 return -ENXIO; 641 return -ENXIO;
641 642
642 lock_kernel();
643 mutex_lock(&bdev->bd_mutex); 643 mutex_lock(&bdev->bd_mutex);
644 644
645 switch (cmd) { 645 switch (cmd) {
@@ -667,7 +667,6 @@ int blk_trace_ioctl(struct block_device *bdev, unsigned cmd, char __user *arg)
667 } 667 }
668 668
669 mutex_unlock(&bdev->bd_mutex); 669 mutex_unlock(&bdev->bd_mutex);
670 unlock_kernel();
671 return ret; 670 return ret;
672} 671}
673 672
@@ -1652,10 +1651,9 @@ static ssize_t sysfs_blk_trace_attr_show(struct device *dev,
1652 struct block_device *bdev; 1651 struct block_device *bdev;
1653 ssize_t ret = -ENXIO; 1652 ssize_t ret = -ENXIO;
1654 1653
1655 lock_kernel();
1656 bdev = bdget(part_devt(p)); 1654 bdev = bdget(part_devt(p));
1657 if (bdev == NULL) 1655 if (bdev == NULL)
1658 goto out_unlock_kernel; 1656 goto out;
1659 1657
1660 q = blk_trace_get_queue(bdev); 1658 q = blk_trace_get_queue(bdev);
1661 if (q == NULL) 1659 if (q == NULL)
@@ -1683,8 +1681,7 @@ out_unlock_bdev:
1683 mutex_unlock(&bdev->bd_mutex); 1681 mutex_unlock(&bdev->bd_mutex);
1684out_bdput: 1682out_bdput:
1685 bdput(bdev); 1683 bdput(bdev);
1686out_unlock_kernel: 1684out:
1687 unlock_kernel();
1688 return ret; 1685 return ret;
1689} 1686}
1690 1687
@@ -1714,11 +1711,10 @@ static ssize_t sysfs_blk_trace_attr_store(struct device *dev,
1714 1711
1715 ret = -ENXIO; 1712 ret = -ENXIO;
1716 1713
1717 lock_kernel();
1718 p = dev_to_part(dev); 1714 p = dev_to_part(dev);
1719 bdev = bdget(part_devt(p)); 1715 bdev = bdget(part_devt(p));
1720 if (bdev == NULL) 1716 if (bdev == NULL)
1721 goto out_unlock_kernel; 1717 goto out;
1722 1718
1723 q = blk_trace_get_queue(bdev); 1719 q = blk_trace_get_queue(bdev);
1724 if (q == NULL) 1720 if (q == NULL)
@@ -1753,8 +1749,6 @@ out_unlock_bdev:
1753 mutex_unlock(&bdev->bd_mutex); 1749 mutex_unlock(&bdev->bd_mutex);
1754out_bdput: 1750out_bdput:
1755 bdput(bdev); 1751 bdput(bdev);
1756out_unlock_kernel:
1757 unlock_kernel();
1758out: 1752out:
1759 return ret ? ret : count; 1753 return ret ? ret : count;
1760} 1754}
diff --git a/kernel/trace/ftrace.c b/kernel/trace/ftrace.c
index fa7ece649fe1..f3dadae83883 100644
--- a/kernel/trace/ftrace.c
+++ b/kernel/trace/ftrace.c
@@ -800,6 +800,7 @@ static const struct file_operations ftrace_profile_fops = {
800 .open = tracing_open_generic, 800 .open = tracing_open_generic,
801 .read = ftrace_profile_read, 801 .read = ftrace_profile_read,
802 .write = ftrace_profile_write, 802 .write = ftrace_profile_write,
803 .llseek = default_llseek,
803}; 804};
804 805
805/* used to initialize the real stat files */ 806/* used to initialize the real stat files */
@@ -884,10 +885,8 @@ enum {
884 FTRACE_ENABLE_CALLS = (1 << 0), 885 FTRACE_ENABLE_CALLS = (1 << 0),
885 FTRACE_DISABLE_CALLS = (1 << 1), 886 FTRACE_DISABLE_CALLS = (1 << 1),
886 FTRACE_UPDATE_TRACE_FUNC = (1 << 2), 887 FTRACE_UPDATE_TRACE_FUNC = (1 << 2),
887 FTRACE_ENABLE_MCOUNT = (1 << 3), 888 FTRACE_START_FUNC_RET = (1 << 3),
888 FTRACE_DISABLE_MCOUNT = (1 << 4), 889 FTRACE_STOP_FUNC_RET = (1 << 4),
889 FTRACE_START_FUNC_RET = (1 << 5),
890 FTRACE_STOP_FUNC_RET = (1 << 6),
891}; 890};
892 891
893static int ftrace_filtered; 892static int ftrace_filtered;
@@ -1226,8 +1225,6 @@ static void ftrace_shutdown(int command)
1226 1225
1227static void ftrace_startup_sysctl(void) 1226static void ftrace_startup_sysctl(void)
1228{ 1227{
1229 int command = FTRACE_ENABLE_MCOUNT;
1230
1231 if (unlikely(ftrace_disabled)) 1228 if (unlikely(ftrace_disabled))
1232 return; 1229 return;
1233 1230
@@ -1235,23 +1232,17 @@ static void ftrace_startup_sysctl(void)
1235 saved_ftrace_func = NULL; 1232 saved_ftrace_func = NULL;
1236 /* ftrace_start_up is true if we want ftrace running */ 1233 /* ftrace_start_up is true if we want ftrace running */
1237 if (ftrace_start_up) 1234 if (ftrace_start_up)
1238 command |= FTRACE_ENABLE_CALLS; 1235 ftrace_run_update_code(FTRACE_ENABLE_CALLS);
1239
1240 ftrace_run_update_code(command);
1241} 1236}
1242 1237
1243static void ftrace_shutdown_sysctl(void) 1238static void ftrace_shutdown_sysctl(void)
1244{ 1239{
1245 int command = FTRACE_DISABLE_MCOUNT;
1246
1247 if (unlikely(ftrace_disabled)) 1240 if (unlikely(ftrace_disabled))
1248 return; 1241 return;
1249 1242
1250 /* ftrace_start_up is true if ftrace is running */ 1243 /* ftrace_start_up is true if ftrace is running */
1251 if (ftrace_start_up) 1244 if (ftrace_start_up)
1252 command |= FTRACE_DISABLE_CALLS; 1245 ftrace_run_update_code(FTRACE_DISABLE_CALLS);
1253
1254 ftrace_run_update_code(command);
1255} 1246}
1256 1247
1257static cycle_t ftrace_update_time; 1248static cycle_t ftrace_update_time;
@@ -1368,24 +1359,29 @@ enum {
1368#define FTRACE_BUFF_MAX (KSYM_SYMBOL_LEN+4) /* room for wildcards */ 1359#define FTRACE_BUFF_MAX (KSYM_SYMBOL_LEN+4) /* room for wildcards */
1369 1360
1370struct ftrace_iterator { 1361struct ftrace_iterator {
1371 struct ftrace_page *pg; 1362 loff_t pos;
1372 int hidx; 1363 loff_t func_pos;
1373 int idx; 1364 struct ftrace_page *pg;
1374 unsigned flags; 1365 struct dyn_ftrace *func;
1375 struct trace_parser parser; 1366 struct ftrace_func_probe *probe;
1367 struct trace_parser parser;
1368 int hidx;
1369 int idx;
1370 unsigned flags;
1376}; 1371};
1377 1372
1378static void * 1373static void *
1379t_hash_next(struct seq_file *m, void *v, loff_t *pos) 1374t_hash_next(struct seq_file *m, loff_t *pos)
1380{ 1375{
1381 struct ftrace_iterator *iter = m->private; 1376 struct ftrace_iterator *iter = m->private;
1382 struct hlist_node *hnd = v; 1377 struct hlist_node *hnd = NULL;
1383 struct hlist_head *hhd; 1378 struct hlist_head *hhd;
1384 1379
1385 WARN_ON(!(iter->flags & FTRACE_ITER_HASH));
1386
1387 (*pos)++; 1380 (*pos)++;
1381 iter->pos = *pos;
1388 1382
1383 if (iter->probe)
1384 hnd = &iter->probe->node;
1389 retry: 1385 retry:
1390 if (iter->hidx >= FTRACE_FUNC_HASHSIZE) 1386 if (iter->hidx >= FTRACE_FUNC_HASHSIZE)
1391 return NULL; 1387 return NULL;
@@ -1408,7 +1404,12 @@ t_hash_next(struct seq_file *m, void *v, loff_t *pos)
1408 } 1404 }
1409 } 1405 }
1410 1406
1411 return hnd; 1407 if (WARN_ON_ONCE(!hnd))
1408 return NULL;
1409
1410 iter->probe = hlist_entry(hnd, struct ftrace_func_probe, node);
1411
1412 return iter;
1412} 1413}
1413 1414
1414static void *t_hash_start(struct seq_file *m, loff_t *pos) 1415static void *t_hash_start(struct seq_file *m, loff_t *pos)
@@ -1417,26 +1418,32 @@ static void *t_hash_start(struct seq_file *m, loff_t *pos)
1417 void *p = NULL; 1418 void *p = NULL;
1418 loff_t l; 1419 loff_t l;
1419 1420
1420 if (!(iter->flags & FTRACE_ITER_HASH)) 1421 if (iter->func_pos > *pos)
1421 *pos = 0; 1422 return NULL;
1422
1423 iter->flags |= FTRACE_ITER_HASH;
1424 1423
1425 iter->hidx = 0; 1424 iter->hidx = 0;
1426 for (l = 0; l <= *pos; ) { 1425 for (l = 0; l <= (*pos - iter->func_pos); ) {
1427 p = t_hash_next(m, p, &l); 1426 p = t_hash_next(m, &l);
1428 if (!p) 1427 if (!p)
1429 break; 1428 break;
1430 } 1429 }
1431 return p; 1430 if (!p)
1431 return NULL;
1432
1433 /* Only set this if we have an item */
1434 iter->flags |= FTRACE_ITER_HASH;
1435
1436 return iter;
1432} 1437}
1433 1438
1434static int t_hash_show(struct seq_file *m, void *v) 1439static int
1440t_hash_show(struct seq_file *m, struct ftrace_iterator *iter)
1435{ 1441{
1436 struct ftrace_func_probe *rec; 1442 struct ftrace_func_probe *rec;
1437 struct hlist_node *hnd = v;
1438 1443
1439 rec = hlist_entry(hnd, struct ftrace_func_probe, node); 1444 rec = iter->probe;
1445 if (WARN_ON_ONCE(!rec))
1446 return -EIO;
1440 1447
1441 if (rec->ops->print) 1448 if (rec->ops->print)
1442 return rec->ops->print(m, rec->ip, rec->ops, rec->data); 1449 return rec->ops->print(m, rec->ip, rec->ops, rec->data);
@@ -1457,12 +1464,13 @@ t_next(struct seq_file *m, void *v, loff_t *pos)
1457 struct dyn_ftrace *rec = NULL; 1464 struct dyn_ftrace *rec = NULL;
1458 1465
1459 if (iter->flags & FTRACE_ITER_HASH) 1466 if (iter->flags & FTRACE_ITER_HASH)
1460 return t_hash_next(m, v, pos); 1467 return t_hash_next(m, pos);
1461 1468
1462 (*pos)++; 1469 (*pos)++;
1470 iter->pos = *pos;
1463 1471
1464 if (iter->flags & FTRACE_ITER_PRINTALL) 1472 if (iter->flags & FTRACE_ITER_PRINTALL)
1465 return NULL; 1473 return t_hash_start(m, pos);
1466 1474
1467 retry: 1475 retry:
1468 if (iter->idx >= iter->pg->index) { 1476 if (iter->idx >= iter->pg->index) {
@@ -1491,7 +1499,20 @@ t_next(struct seq_file *m, void *v, loff_t *pos)
1491 } 1499 }
1492 } 1500 }
1493 1501
1494 return rec; 1502 if (!rec)
1503 return t_hash_start(m, pos);
1504
1505 iter->func_pos = *pos;
1506 iter->func = rec;
1507
1508 return iter;
1509}
1510
1511static void reset_iter_read(struct ftrace_iterator *iter)
1512{
1513 iter->pos = 0;
1514 iter->func_pos = 0;
1515 iter->flags &= ~(FTRACE_ITER_PRINTALL & FTRACE_ITER_HASH);
1495} 1516}
1496 1517
1497static void *t_start(struct seq_file *m, loff_t *pos) 1518static void *t_start(struct seq_file *m, loff_t *pos)
@@ -1502,6 +1523,12 @@ static void *t_start(struct seq_file *m, loff_t *pos)
1502 1523
1503 mutex_lock(&ftrace_lock); 1524 mutex_lock(&ftrace_lock);
1504 /* 1525 /*
1526 * If an lseek was done, then reset and start from beginning.
1527 */
1528 if (*pos < iter->pos)
1529 reset_iter_read(iter);
1530
1531 /*
1505 * For set_ftrace_filter reading, if we have the filter 1532 * For set_ftrace_filter reading, if we have the filter
1506 * off, we can short cut and just print out that all 1533 * off, we can short cut and just print out that all
1507 * functions are enabled. 1534 * functions are enabled.
@@ -1518,6 +1545,11 @@ static void *t_start(struct seq_file *m, loff_t *pos)
1518 if (iter->flags & FTRACE_ITER_HASH) 1545 if (iter->flags & FTRACE_ITER_HASH)
1519 return t_hash_start(m, pos); 1546 return t_hash_start(m, pos);
1520 1547
1548 /*
1549 * Unfortunately, we need to restart at ftrace_pages_start
1550 * every time we let go of the ftrace_mutex. This is because
1551 * those pointers can change without the lock.
1552 */
1521 iter->pg = ftrace_pages_start; 1553 iter->pg = ftrace_pages_start;
1522 iter->idx = 0; 1554 iter->idx = 0;
1523 for (l = 0; l <= *pos; ) { 1555 for (l = 0; l <= *pos; ) {
@@ -1526,10 +1558,14 @@ static void *t_start(struct seq_file *m, loff_t *pos)
1526 break; 1558 break;
1527 } 1559 }
1528 1560
1529 if (!p && iter->flags & FTRACE_ITER_FILTER) 1561 if (!p) {
1530 return t_hash_start(m, pos); 1562 if (iter->flags & FTRACE_ITER_FILTER)
1563 return t_hash_start(m, pos);
1564
1565 return NULL;
1566 }
1531 1567
1532 return p; 1568 return iter;
1533} 1569}
1534 1570
1535static void t_stop(struct seq_file *m, void *p) 1571static void t_stop(struct seq_file *m, void *p)
@@ -1540,16 +1576,18 @@ static void t_stop(struct seq_file *m, void *p)
1540static int t_show(struct seq_file *m, void *v) 1576static int t_show(struct seq_file *m, void *v)
1541{ 1577{
1542 struct ftrace_iterator *iter = m->private; 1578 struct ftrace_iterator *iter = m->private;
1543 struct dyn_ftrace *rec = v; 1579 struct dyn_ftrace *rec;
1544 1580
1545 if (iter->flags & FTRACE_ITER_HASH) 1581 if (iter->flags & FTRACE_ITER_HASH)
1546 return t_hash_show(m, v); 1582 return t_hash_show(m, iter);
1547 1583
1548 if (iter->flags & FTRACE_ITER_PRINTALL) { 1584 if (iter->flags & FTRACE_ITER_PRINTALL) {
1549 seq_printf(m, "#### all functions enabled ####\n"); 1585 seq_printf(m, "#### all functions enabled ####\n");
1550 return 0; 1586 return 0;
1551 } 1587 }
1552 1588
1589 rec = iter->func;
1590
1553 if (!rec) 1591 if (!rec)
1554 return 0; 1592 return 0;
1555 1593
@@ -1601,8 +1639,8 @@ ftrace_failures_open(struct inode *inode, struct file *file)
1601 1639
1602 ret = ftrace_avail_open(inode, file); 1640 ret = ftrace_avail_open(inode, file);
1603 if (!ret) { 1641 if (!ret) {
1604 m = (struct seq_file *)file->private_data; 1642 m = file->private_data;
1605 iter = (struct ftrace_iterator *)m->private; 1643 iter = m->private;
1606 iter->flags = FTRACE_ITER_FAILURES; 1644 iter->flags = FTRACE_ITER_FAILURES;
1607 } 1645 }
1608 1646
@@ -2418,7 +2456,7 @@ static const struct file_operations ftrace_filter_fops = {
2418 .open = ftrace_filter_open, 2456 .open = ftrace_filter_open,
2419 .read = seq_read, 2457 .read = seq_read,
2420 .write = ftrace_filter_write, 2458 .write = ftrace_filter_write,
2421 .llseek = no_llseek, 2459 .llseek = ftrace_regex_lseek,
2422 .release = ftrace_filter_release, 2460 .release = ftrace_filter_release,
2423}; 2461};
2424 2462
@@ -2632,6 +2670,7 @@ static const struct file_operations ftrace_graph_fops = {
2632 .read = seq_read, 2670 .read = seq_read,
2633 .write = ftrace_graph_write, 2671 .write = ftrace_graph_write,
2634 .release = ftrace_graph_release, 2672 .release = ftrace_graph_release,
2673 .llseek = seq_lseek,
2635}; 2674};
2636#endif /* CONFIG_FUNCTION_GRAPH_TRACER */ 2675#endif /* CONFIG_FUNCTION_GRAPH_TRACER */
2637 2676
diff --git a/kernel/trace/ring_buffer.c b/kernel/trace/ring_buffer.c
index 492197e2f86c..c3dab054d18e 100644
--- a/kernel/trace/ring_buffer.c
+++ b/kernel/trace/ring_buffer.c
@@ -405,7 +405,7 @@ static inline int test_time_stamp(u64 delta)
405#define BUF_MAX_DATA_SIZE (BUF_PAGE_SIZE - (sizeof(u32) * 2)) 405#define BUF_MAX_DATA_SIZE (BUF_PAGE_SIZE - (sizeof(u32) * 2))
406 406
407/* Max number of timestamps that can fit on a page */ 407/* Max number of timestamps that can fit on a page */
408#define RB_TIMESTAMPS_PER_PAGE (BUF_PAGE_SIZE / RB_LEN_TIME_STAMP) 408#define RB_TIMESTAMPS_PER_PAGE (BUF_PAGE_SIZE / RB_LEN_TIME_EXTEND)
409 409
410int ring_buffer_print_page_header(struct trace_seq *s) 410int ring_buffer_print_page_header(struct trace_seq *s)
411{ 411{
@@ -2606,6 +2606,19 @@ void ring_buffer_record_enable_cpu(struct ring_buffer *buffer, int cpu)
2606} 2606}
2607EXPORT_SYMBOL_GPL(ring_buffer_record_enable_cpu); 2607EXPORT_SYMBOL_GPL(ring_buffer_record_enable_cpu);
2608 2608
2609/*
2610 * The total entries in the ring buffer is the running counter
2611 * of entries entered into the ring buffer, minus the sum of
2612 * the entries read from the ring buffer and the number of
2613 * entries that were overwritten.
2614 */
2615static inline unsigned long
2616rb_num_of_entries(struct ring_buffer_per_cpu *cpu_buffer)
2617{
2618 return local_read(&cpu_buffer->entries) -
2619 (local_read(&cpu_buffer->overrun) + cpu_buffer->read);
2620}
2621
2609/** 2622/**
2610 * ring_buffer_entries_cpu - get the number of entries in a cpu buffer 2623 * ring_buffer_entries_cpu - get the number of entries in a cpu buffer
2611 * @buffer: The ring buffer 2624 * @buffer: The ring buffer
@@ -2614,16 +2627,13 @@ EXPORT_SYMBOL_GPL(ring_buffer_record_enable_cpu);
2614unsigned long ring_buffer_entries_cpu(struct ring_buffer *buffer, int cpu) 2627unsigned long ring_buffer_entries_cpu(struct ring_buffer *buffer, int cpu)
2615{ 2628{
2616 struct ring_buffer_per_cpu *cpu_buffer; 2629 struct ring_buffer_per_cpu *cpu_buffer;
2617 unsigned long ret;
2618 2630
2619 if (!cpumask_test_cpu(cpu, buffer->cpumask)) 2631 if (!cpumask_test_cpu(cpu, buffer->cpumask))
2620 return 0; 2632 return 0;
2621 2633
2622 cpu_buffer = buffer->buffers[cpu]; 2634 cpu_buffer = buffer->buffers[cpu];
2623 ret = (local_read(&cpu_buffer->entries) - local_read(&cpu_buffer->overrun))
2624 - cpu_buffer->read;
2625 2635
2626 return ret; 2636 return rb_num_of_entries(cpu_buffer);
2627} 2637}
2628EXPORT_SYMBOL_GPL(ring_buffer_entries_cpu); 2638EXPORT_SYMBOL_GPL(ring_buffer_entries_cpu);
2629 2639
@@ -2684,8 +2694,7 @@ unsigned long ring_buffer_entries(struct ring_buffer *buffer)
2684 /* if you care about this being correct, lock the buffer */ 2694 /* if you care about this being correct, lock the buffer */
2685 for_each_buffer_cpu(buffer, cpu) { 2695 for_each_buffer_cpu(buffer, cpu) {
2686 cpu_buffer = buffer->buffers[cpu]; 2696 cpu_buffer = buffer->buffers[cpu];
2687 entries += (local_read(&cpu_buffer->entries) - 2697 entries += rb_num_of_entries(cpu_buffer);
2688 local_read(&cpu_buffer->overrun)) - cpu_buffer->read;
2689 } 2698 }
2690 2699
2691 return entries; 2700 return entries;
@@ -3965,6 +3974,7 @@ static const struct file_operations rb_simple_fops = {
3965 .open = tracing_open_generic, 3974 .open = tracing_open_generic,
3966 .read = rb_simple_read, 3975 .read = rb_simple_read,
3967 .write = rb_simple_write, 3976 .write = rb_simple_write,
3977 .llseek = default_llseek,
3968}; 3978};
3969 3979
3970 3980
diff --git a/kernel/trace/trace.c b/kernel/trace/trace.c
index 9ec59f541156..001bcd2ccf4a 100644
--- a/kernel/trace/trace.c
+++ b/kernel/trace/trace.c
@@ -2196,7 +2196,7 @@ int tracing_open_generic(struct inode *inode, struct file *filp)
2196 2196
2197static int tracing_release(struct inode *inode, struct file *file) 2197static int tracing_release(struct inode *inode, struct file *file)
2198{ 2198{
2199 struct seq_file *m = (struct seq_file *)file->private_data; 2199 struct seq_file *m = file->private_data;
2200 struct trace_iterator *iter; 2200 struct trace_iterator *iter;
2201 int cpu; 2201 int cpu;
2202 2202
diff --git a/kernel/trace/trace.h b/kernel/trace/trace.h
index d39b3c5454a5..9021f8c0c0c3 100644
--- a/kernel/trace/trace.h
+++ b/kernel/trace/trace.h
@@ -343,6 +343,10 @@ void trace_function(struct trace_array *tr,
343 unsigned long ip, 343 unsigned long ip,
344 unsigned long parent_ip, 344 unsigned long parent_ip,
345 unsigned long flags, int pc); 345 unsigned long flags, int pc);
346void trace_graph_function(struct trace_array *tr,
347 unsigned long ip,
348 unsigned long parent_ip,
349 unsigned long flags, int pc);
346void trace_default_header(struct seq_file *m); 350void trace_default_header(struct seq_file *m);
347void print_trace_header(struct seq_file *m, struct trace_iterator *iter); 351void print_trace_header(struct seq_file *m, struct trace_iterator *iter);
348int trace_empty(struct trace_iterator *iter); 352int trace_empty(struct trace_iterator *iter);
diff --git a/kernel/trace/trace_event_perf.c b/kernel/trace/trace_event_perf.c
index 31cc4cb0dbf2..39c059ca670e 100644
--- a/kernel/trace/trace_event_perf.c
+++ b/kernel/trace/trace_event_perf.c
@@ -9,7 +9,7 @@
9#include <linux/kprobes.h> 9#include <linux/kprobes.h>
10#include "trace.h" 10#include "trace.h"
11 11
12static char *perf_trace_buf[4]; 12static char __percpu *perf_trace_buf[PERF_NR_CONTEXTS];
13 13
14/* 14/*
15 * Force it to be aligned to unsigned long to avoid misaligned accesses 15 * Force it to be aligned to unsigned long to avoid misaligned accesses
@@ -24,7 +24,7 @@ static int total_ref_count;
24static int perf_trace_event_init(struct ftrace_event_call *tp_event, 24static int perf_trace_event_init(struct ftrace_event_call *tp_event,
25 struct perf_event *p_event) 25 struct perf_event *p_event)
26{ 26{
27 struct hlist_head *list; 27 struct hlist_head __percpu *list;
28 int ret = -ENOMEM; 28 int ret = -ENOMEM;
29 int cpu; 29 int cpu;
30 30
@@ -42,11 +42,11 @@ static int perf_trace_event_init(struct ftrace_event_call *tp_event,
42 tp_event->perf_events = list; 42 tp_event->perf_events = list;
43 43
44 if (!total_ref_count) { 44 if (!total_ref_count) {
45 char *buf; 45 char __percpu *buf;
46 int i; 46 int i;
47 47
48 for (i = 0; i < 4; i++) { 48 for (i = 0; i < PERF_NR_CONTEXTS; i++) {
49 buf = (char *)alloc_percpu(perf_trace_t); 49 buf = (char __percpu *)alloc_percpu(perf_trace_t);
50 if (!buf) 50 if (!buf)
51 goto fail; 51 goto fail;
52 52
@@ -65,7 +65,7 @@ fail:
65 if (!total_ref_count) { 65 if (!total_ref_count) {
66 int i; 66 int i;
67 67
68 for (i = 0; i < 4; i++) { 68 for (i = 0; i < PERF_NR_CONTEXTS; i++) {
69 free_percpu(perf_trace_buf[i]); 69 free_percpu(perf_trace_buf[i]);
70 perf_trace_buf[i] = NULL; 70 perf_trace_buf[i] = NULL;
71 } 71 }
@@ -101,22 +101,26 @@ int perf_trace_init(struct perf_event *p_event)
101 return ret; 101 return ret;
102} 102}
103 103
104int perf_trace_enable(struct perf_event *p_event) 104int perf_trace_add(struct perf_event *p_event, int flags)
105{ 105{
106 struct ftrace_event_call *tp_event = p_event->tp_event; 106 struct ftrace_event_call *tp_event = p_event->tp_event;
107 struct hlist_head __percpu *pcpu_list;
107 struct hlist_head *list; 108 struct hlist_head *list;
108 109
109 list = tp_event->perf_events; 110 pcpu_list = tp_event->perf_events;
110 if (WARN_ON_ONCE(!list)) 111 if (WARN_ON_ONCE(!pcpu_list))
111 return -EINVAL; 112 return -EINVAL;
112 113
113 list = this_cpu_ptr(list); 114 if (!(flags & PERF_EF_START))
115 p_event->hw.state = PERF_HES_STOPPED;
116
117 list = this_cpu_ptr(pcpu_list);
114 hlist_add_head_rcu(&p_event->hlist_entry, list); 118 hlist_add_head_rcu(&p_event->hlist_entry, list);
115 119
116 return 0; 120 return 0;
117} 121}
118 122
119void perf_trace_disable(struct perf_event *p_event) 123void perf_trace_del(struct perf_event *p_event, int flags)
120{ 124{
121 hlist_del_rcu(&p_event->hlist_entry); 125 hlist_del_rcu(&p_event->hlist_entry);
122} 126}
@@ -142,7 +146,7 @@ void perf_trace_destroy(struct perf_event *p_event)
142 tp_event->perf_events = NULL; 146 tp_event->perf_events = NULL;
143 147
144 if (!--total_ref_count) { 148 if (!--total_ref_count) {
145 for (i = 0; i < 4; i++) { 149 for (i = 0; i < PERF_NR_CONTEXTS; i++) {
146 free_percpu(perf_trace_buf[i]); 150 free_percpu(perf_trace_buf[i]);
147 perf_trace_buf[i] = NULL; 151 perf_trace_buf[i] = NULL;
148 } 152 }
diff --git a/kernel/trace/trace_events.c b/kernel/trace/trace_events.c
index 4c758f146328..0725eeab1937 100644
--- a/kernel/trace/trace_events.c
+++ b/kernel/trace/trace_events.c
@@ -600,21 +600,29 @@ out:
600 600
601enum { 601enum {
602 FORMAT_HEADER = 1, 602 FORMAT_HEADER = 1,
603 FORMAT_PRINTFMT = 2, 603 FORMAT_FIELD_SEPERATOR = 2,
604 FORMAT_PRINTFMT = 3,
604}; 605};
605 606
606static void *f_next(struct seq_file *m, void *v, loff_t *pos) 607static void *f_next(struct seq_file *m, void *v, loff_t *pos)
607{ 608{
608 struct ftrace_event_call *call = m->private; 609 struct ftrace_event_call *call = m->private;
609 struct ftrace_event_field *field; 610 struct ftrace_event_field *field;
610 struct list_head *head; 611 struct list_head *common_head = &ftrace_common_fields;
612 struct list_head *head = trace_get_fields(call);
611 613
612 (*pos)++; 614 (*pos)++;
613 615
614 switch ((unsigned long)v) { 616 switch ((unsigned long)v) {
615 case FORMAT_HEADER: 617 case FORMAT_HEADER:
616 head = &ftrace_common_fields; 618 if (unlikely(list_empty(common_head)))
619 return NULL;
620
621 field = list_entry(common_head->prev,
622 struct ftrace_event_field, link);
623 return field;
617 624
625 case FORMAT_FIELD_SEPERATOR:
618 if (unlikely(list_empty(head))) 626 if (unlikely(list_empty(head)))
619 return NULL; 627 return NULL;
620 628
@@ -626,31 +634,10 @@ static void *f_next(struct seq_file *m, void *v, loff_t *pos)
626 return NULL; 634 return NULL;
627 } 635 }
628 636
629 head = trace_get_fields(call);
630
631 /*
632 * To separate common fields from event fields, the
633 * LSB is set on the first event field. Clear it in case.
634 */
635 v = (void *)((unsigned long)v & ~1L);
636
637 field = v; 637 field = v;
638 /* 638 if (field->link.prev == common_head)
639 * If this is a common field, and at the end of the list, then 639 return (void *)FORMAT_FIELD_SEPERATOR;
640 * continue with main list. 640 else if (field->link.prev == head)
641 */
642 if (field->link.prev == &ftrace_common_fields) {
643 if (unlikely(list_empty(head)))
644 return NULL;
645 field = list_entry(head->prev, struct ftrace_event_field, link);
646 /* Set the LSB to notify f_show to print an extra newline */
647 field = (struct ftrace_event_field *)
648 ((unsigned long)field | 1);
649 return field;
650 }
651
652 /* If we are done tell f_show to print the format */
653 if (field->link.prev == head)
654 return (void *)FORMAT_PRINTFMT; 641 return (void *)FORMAT_PRINTFMT;
655 642
656 field = list_entry(field->link.prev, struct ftrace_event_field, link); 643 field = list_entry(field->link.prev, struct ftrace_event_field, link);
@@ -688,22 +675,16 @@ static int f_show(struct seq_file *m, void *v)
688 seq_printf(m, "format:\n"); 675 seq_printf(m, "format:\n");
689 return 0; 676 return 0;
690 677
678 case FORMAT_FIELD_SEPERATOR:
679 seq_putc(m, '\n');
680 return 0;
681
691 case FORMAT_PRINTFMT: 682 case FORMAT_PRINTFMT:
692 seq_printf(m, "\nprint fmt: %s\n", 683 seq_printf(m, "\nprint fmt: %s\n",
693 call->print_fmt); 684 call->print_fmt);
694 return 0; 685 return 0;
695 } 686 }
696 687
697 /*
698 * To separate common fields from event fields, the
699 * LSB is set on the first event field. Clear it and
700 * print a newline if it is set.
701 */
702 if ((unsigned long)v & 1) {
703 seq_putc(m, '\n');
704 v = (void *)((unsigned long)v & ~1L);
705 }
706
707 field = v; 688 field = v;
708 689
709 /* 690 /*
@@ -951,6 +932,7 @@ static const struct file_operations ftrace_enable_fops = {
951 .open = tracing_open_generic, 932 .open = tracing_open_generic,
952 .read = event_enable_read, 933 .read = event_enable_read,
953 .write = event_enable_write, 934 .write = event_enable_write,
935 .llseek = default_llseek,
954}; 936};
955 937
956static const struct file_operations ftrace_event_format_fops = { 938static const struct file_operations ftrace_event_format_fops = {
@@ -963,29 +945,34 @@ static const struct file_operations ftrace_event_format_fops = {
963static const struct file_operations ftrace_event_id_fops = { 945static const struct file_operations ftrace_event_id_fops = {
964 .open = tracing_open_generic, 946 .open = tracing_open_generic,
965 .read = event_id_read, 947 .read = event_id_read,
948 .llseek = default_llseek,
966}; 949};
967 950
968static const struct file_operations ftrace_event_filter_fops = { 951static const struct file_operations ftrace_event_filter_fops = {
969 .open = tracing_open_generic, 952 .open = tracing_open_generic,
970 .read = event_filter_read, 953 .read = event_filter_read,
971 .write = event_filter_write, 954 .write = event_filter_write,
955 .llseek = default_llseek,
972}; 956};
973 957
974static const struct file_operations ftrace_subsystem_filter_fops = { 958static const struct file_operations ftrace_subsystem_filter_fops = {
975 .open = tracing_open_generic, 959 .open = tracing_open_generic,
976 .read = subsystem_filter_read, 960 .read = subsystem_filter_read,
977 .write = subsystem_filter_write, 961 .write = subsystem_filter_write,
962 .llseek = default_llseek,
978}; 963};
979 964
980static const struct file_operations ftrace_system_enable_fops = { 965static const struct file_operations ftrace_system_enable_fops = {
981 .open = tracing_open_generic, 966 .open = tracing_open_generic,
982 .read = system_enable_read, 967 .read = system_enable_read,
983 .write = system_enable_write, 968 .write = system_enable_write,
969 .llseek = default_llseek,
984}; 970};
985 971
986static const struct file_operations ftrace_show_header_fops = { 972static const struct file_operations ftrace_show_header_fops = {
987 .open = tracing_open_generic, 973 .open = tracing_open_generic,
988 .read = show_header, 974 .read = show_header,
975 .llseek = default_llseek,
989}; 976};
990 977
991static struct dentry *event_trace_events_dir(void) 978static struct dentry *event_trace_events_dir(void)
diff --git a/kernel/trace/trace_functions_graph.c b/kernel/trace/trace_functions_graph.c
index 6f233698518e..76b05980225c 100644
--- a/kernel/trace/trace_functions_graph.c
+++ b/kernel/trace/trace_functions_graph.c
@@ -15,15 +15,19 @@
15#include "trace.h" 15#include "trace.h"
16#include "trace_output.h" 16#include "trace_output.h"
17 17
18/* When set, irq functions will be ignored */
19static int ftrace_graph_skip_irqs;
20
18struct fgraph_cpu_data { 21struct fgraph_cpu_data {
19 pid_t last_pid; 22 pid_t last_pid;
20 int depth; 23 int depth;
24 int depth_irq;
21 int ignore; 25 int ignore;
22 unsigned long enter_funcs[FTRACE_RETFUNC_DEPTH]; 26 unsigned long enter_funcs[FTRACE_RETFUNC_DEPTH];
23}; 27};
24 28
25struct fgraph_data { 29struct fgraph_data {
26 struct fgraph_cpu_data *cpu_data; 30 struct fgraph_cpu_data __percpu *cpu_data;
27 31
28 /* Place to preserve last processed entry. */ 32 /* Place to preserve last processed entry. */
29 struct ftrace_graph_ent_entry ent; 33 struct ftrace_graph_ent_entry ent;
@@ -41,6 +45,7 @@ struct fgraph_data {
41#define TRACE_GRAPH_PRINT_PROC 0x8 45#define TRACE_GRAPH_PRINT_PROC 0x8
42#define TRACE_GRAPH_PRINT_DURATION 0x10 46#define TRACE_GRAPH_PRINT_DURATION 0x10
43#define TRACE_GRAPH_PRINT_ABS_TIME 0x20 47#define TRACE_GRAPH_PRINT_ABS_TIME 0x20
48#define TRACE_GRAPH_PRINT_IRQS 0x40
44 49
45static struct tracer_opt trace_opts[] = { 50static struct tracer_opt trace_opts[] = {
46 /* Display overruns? (for self-debug purpose) */ 51 /* Display overruns? (for self-debug purpose) */
@@ -55,13 +60,15 @@ static struct tracer_opt trace_opts[] = {
55 { TRACER_OPT(funcgraph-duration, TRACE_GRAPH_PRINT_DURATION) }, 60 { TRACER_OPT(funcgraph-duration, TRACE_GRAPH_PRINT_DURATION) },
56 /* Display absolute time of an entry */ 61 /* Display absolute time of an entry */
57 { TRACER_OPT(funcgraph-abstime, TRACE_GRAPH_PRINT_ABS_TIME) }, 62 { TRACER_OPT(funcgraph-abstime, TRACE_GRAPH_PRINT_ABS_TIME) },
63 /* Display interrupts */
64 { TRACER_OPT(funcgraph-irqs, TRACE_GRAPH_PRINT_IRQS) },
58 { } /* Empty entry */ 65 { } /* Empty entry */
59}; 66};
60 67
61static struct tracer_flags tracer_flags = { 68static struct tracer_flags tracer_flags = {
62 /* Don't display overruns and proc by default */ 69 /* Don't display overruns and proc by default */
63 .val = TRACE_GRAPH_PRINT_CPU | TRACE_GRAPH_PRINT_OVERHEAD | 70 .val = TRACE_GRAPH_PRINT_CPU | TRACE_GRAPH_PRINT_OVERHEAD |
64 TRACE_GRAPH_PRINT_DURATION, 71 TRACE_GRAPH_PRINT_DURATION | TRACE_GRAPH_PRINT_IRQS,
65 .opts = trace_opts 72 .opts = trace_opts
66}; 73};
67 74
@@ -204,6 +211,14 @@ int __trace_graph_entry(struct trace_array *tr,
204 return 1; 211 return 1;
205} 212}
206 213
214static inline int ftrace_graph_ignore_irqs(void)
215{
216 if (!ftrace_graph_skip_irqs)
217 return 0;
218
219 return in_irq();
220}
221
207int trace_graph_entry(struct ftrace_graph_ent *trace) 222int trace_graph_entry(struct ftrace_graph_ent *trace)
208{ 223{
209 struct trace_array *tr = graph_array; 224 struct trace_array *tr = graph_array;
@@ -218,7 +233,8 @@ int trace_graph_entry(struct ftrace_graph_ent *trace)
218 return 0; 233 return 0;
219 234
220 /* trace it when it is-nested-in or is a function enabled. */ 235 /* trace it when it is-nested-in or is a function enabled. */
221 if (!(trace->depth || ftrace_graph_addr(trace->func))) 236 if (!(trace->depth || ftrace_graph_addr(trace->func)) ||
237 ftrace_graph_ignore_irqs())
222 return 0; 238 return 0;
223 239
224 local_irq_save(flags); 240 local_irq_save(flags);
@@ -246,6 +262,34 @@ int trace_graph_thresh_entry(struct ftrace_graph_ent *trace)
246 return trace_graph_entry(trace); 262 return trace_graph_entry(trace);
247} 263}
248 264
265static void
266__trace_graph_function(struct trace_array *tr,
267 unsigned long ip, unsigned long flags, int pc)
268{
269 u64 time = trace_clock_local();
270 struct ftrace_graph_ent ent = {
271 .func = ip,
272 .depth = 0,
273 };
274 struct ftrace_graph_ret ret = {
275 .func = ip,
276 .depth = 0,
277 .calltime = time,
278 .rettime = time,
279 };
280
281 __trace_graph_entry(tr, &ent, flags, pc);
282 __trace_graph_return(tr, &ret, flags, pc);
283}
284
285void
286trace_graph_function(struct trace_array *tr,
287 unsigned long ip, unsigned long parent_ip,
288 unsigned long flags, int pc)
289{
290 __trace_graph_function(tr, ip, flags, pc);
291}
292
249void __trace_graph_return(struct trace_array *tr, 293void __trace_graph_return(struct trace_array *tr,
250 struct ftrace_graph_ret *trace, 294 struct ftrace_graph_ret *trace,
251 unsigned long flags, 295 unsigned long flags,
@@ -649,8 +693,9 @@ trace_print_graph_duration(unsigned long long duration, struct trace_seq *s)
649 693
650 /* Print nsecs (we don't want to exceed 7 numbers) */ 694 /* Print nsecs (we don't want to exceed 7 numbers) */
651 if (len < 7) { 695 if (len < 7) {
652 snprintf(nsecs_str, min(sizeof(nsecs_str), 8UL - len), "%03lu", 696 size_t slen = min_t(size_t, sizeof(nsecs_str), 8UL - len);
653 nsecs_rem); 697
698 snprintf(nsecs_str, slen, "%03lu", nsecs_rem);
654 ret = trace_seq_printf(s, ".%s", nsecs_str); 699 ret = trace_seq_printf(s, ".%s", nsecs_str);
655 if (!ret) 700 if (!ret)
656 return TRACE_TYPE_PARTIAL_LINE; 701 return TRACE_TYPE_PARTIAL_LINE;
@@ -855,6 +900,108 @@ print_graph_prologue(struct trace_iterator *iter, struct trace_seq *s,
855 return 0; 900 return 0;
856} 901}
857 902
903/*
904 * Entry check for irq code
905 *
906 * returns 1 if
907 * - we are inside irq code
908 * - we just extered irq code
909 *
910 * retunns 0 if
911 * - funcgraph-interrupts option is set
912 * - we are not inside irq code
913 */
914static int
915check_irq_entry(struct trace_iterator *iter, u32 flags,
916 unsigned long addr, int depth)
917{
918 int cpu = iter->cpu;
919 int *depth_irq;
920 struct fgraph_data *data = iter->private;
921
922 /*
923 * If we are either displaying irqs, or we got called as
924 * a graph event and private data does not exist,
925 * then we bypass the irq check.
926 */
927 if ((flags & TRACE_GRAPH_PRINT_IRQS) ||
928 (!data))
929 return 0;
930
931 depth_irq = &(per_cpu_ptr(data->cpu_data, cpu)->depth_irq);
932
933 /*
934 * We are inside the irq code
935 */
936 if (*depth_irq >= 0)
937 return 1;
938
939 if ((addr < (unsigned long)__irqentry_text_start) ||
940 (addr >= (unsigned long)__irqentry_text_end))
941 return 0;
942
943 /*
944 * We are entering irq code.
945 */
946 *depth_irq = depth;
947 return 1;
948}
949
950/*
951 * Return check for irq code
952 *
953 * returns 1 if
954 * - we are inside irq code
955 * - we just left irq code
956 *
957 * returns 0 if
958 * - funcgraph-interrupts option is set
959 * - we are not inside irq code
960 */
961static int
962check_irq_return(struct trace_iterator *iter, u32 flags, int depth)
963{
964 int cpu = iter->cpu;
965 int *depth_irq;
966 struct fgraph_data *data = iter->private;
967
968 /*
969 * If we are either displaying irqs, or we got called as
970 * a graph event and private data does not exist,
971 * then we bypass the irq check.
972 */
973 if ((flags & TRACE_GRAPH_PRINT_IRQS) ||
974 (!data))
975 return 0;
976
977 depth_irq = &(per_cpu_ptr(data->cpu_data, cpu)->depth_irq);
978
979 /*
980 * We are not inside the irq code.
981 */
982 if (*depth_irq == -1)
983 return 0;
984
985 /*
986 * We are inside the irq code, and this is returning entry.
987 * Let's not trace it and clear the entry depth, since
988 * we are out of irq code.
989 *
990 * This condition ensures that we 'leave the irq code' once
991 * we are out of the entry depth. Thus protecting us from
992 * the RETURN entry loss.
993 */
994 if (*depth_irq >= depth) {
995 *depth_irq = -1;
996 return 1;
997 }
998
999 /*
1000 * We are inside the irq code, and this is not the entry.
1001 */
1002 return 1;
1003}
1004
858static enum print_line_t 1005static enum print_line_t
859print_graph_entry(struct ftrace_graph_ent_entry *field, struct trace_seq *s, 1006print_graph_entry(struct ftrace_graph_ent_entry *field, struct trace_seq *s,
860 struct trace_iterator *iter, u32 flags) 1007 struct trace_iterator *iter, u32 flags)
@@ -865,6 +1012,9 @@ print_graph_entry(struct ftrace_graph_ent_entry *field, struct trace_seq *s,
865 static enum print_line_t ret; 1012 static enum print_line_t ret;
866 int cpu = iter->cpu; 1013 int cpu = iter->cpu;
867 1014
1015 if (check_irq_entry(iter, flags, call->func, call->depth))
1016 return TRACE_TYPE_HANDLED;
1017
868 if (print_graph_prologue(iter, s, TRACE_GRAPH_ENT, call->func, flags)) 1018 if (print_graph_prologue(iter, s, TRACE_GRAPH_ENT, call->func, flags))
869 return TRACE_TYPE_PARTIAL_LINE; 1019 return TRACE_TYPE_PARTIAL_LINE;
870 1020
@@ -902,6 +1052,9 @@ print_graph_return(struct ftrace_graph_ret *trace, struct trace_seq *s,
902 int ret; 1052 int ret;
903 int i; 1053 int i;
904 1054
1055 if (check_irq_return(iter, flags, trace->depth))
1056 return TRACE_TYPE_HANDLED;
1057
905 if (data) { 1058 if (data) {
906 struct fgraph_cpu_data *cpu_data; 1059 struct fgraph_cpu_data *cpu_data;
907 int cpu = iter->cpu; 1060 int cpu = iter->cpu;
@@ -1054,7 +1207,7 @@ print_graph_comment(struct trace_seq *s, struct trace_entry *ent,
1054 1207
1055 1208
1056enum print_line_t 1209enum print_line_t
1057print_graph_function_flags(struct trace_iterator *iter, u32 flags) 1210__print_graph_function_flags(struct trace_iterator *iter, u32 flags)
1058{ 1211{
1059 struct ftrace_graph_ent_entry *field; 1212 struct ftrace_graph_ent_entry *field;
1060 struct fgraph_data *data = iter->private; 1213 struct fgraph_data *data = iter->private;
@@ -1117,7 +1270,18 @@ print_graph_function_flags(struct trace_iterator *iter, u32 flags)
1117static enum print_line_t 1270static enum print_line_t
1118print_graph_function(struct trace_iterator *iter) 1271print_graph_function(struct trace_iterator *iter)
1119{ 1272{
1120 return print_graph_function_flags(iter, tracer_flags.val); 1273 return __print_graph_function_flags(iter, tracer_flags.val);
1274}
1275
1276enum print_line_t print_graph_function_flags(struct trace_iterator *iter,
1277 u32 flags)
1278{
1279 if (trace_flags & TRACE_ITER_LATENCY_FMT)
1280 flags |= TRACE_GRAPH_PRINT_DURATION;
1281 else
1282 flags |= TRACE_GRAPH_PRINT_ABS_TIME;
1283
1284 return __print_graph_function_flags(iter, flags);
1121} 1285}
1122 1286
1123static enum print_line_t 1287static enum print_line_t
@@ -1149,7 +1313,7 @@ static void print_lat_header(struct seq_file *s, u32 flags)
1149 seq_printf(s, "#%.*s|||| / \n", size, spaces); 1313 seq_printf(s, "#%.*s|||| / \n", size, spaces);
1150} 1314}
1151 1315
1152void print_graph_headers_flags(struct seq_file *s, u32 flags) 1316static void __print_graph_headers_flags(struct seq_file *s, u32 flags)
1153{ 1317{
1154 int lat = trace_flags & TRACE_ITER_LATENCY_FMT; 1318 int lat = trace_flags & TRACE_ITER_LATENCY_FMT;
1155 1319
@@ -1190,6 +1354,23 @@ void print_graph_headers(struct seq_file *s)
1190 print_graph_headers_flags(s, tracer_flags.val); 1354 print_graph_headers_flags(s, tracer_flags.val);
1191} 1355}
1192 1356
1357void print_graph_headers_flags(struct seq_file *s, u32 flags)
1358{
1359 struct trace_iterator *iter = s->private;
1360
1361 if (trace_flags & TRACE_ITER_LATENCY_FMT) {
1362 /* print nothing if the buffers are empty */
1363 if (trace_empty(iter))
1364 return;
1365
1366 print_trace_header(s, iter);
1367 flags |= TRACE_GRAPH_PRINT_DURATION;
1368 } else
1369 flags |= TRACE_GRAPH_PRINT_ABS_TIME;
1370
1371 __print_graph_headers_flags(s, flags);
1372}
1373
1193void graph_trace_open(struct trace_iterator *iter) 1374void graph_trace_open(struct trace_iterator *iter)
1194{ 1375{
1195 /* pid and depth on the last trace processed */ 1376 /* pid and depth on the last trace processed */
@@ -1210,9 +1391,12 @@ void graph_trace_open(struct trace_iterator *iter)
1210 pid_t *pid = &(per_cpu_ptr(data->cpu_data, cpu)->last_pid); 1391 pid_t *pid = &(per_cpu_ptr(data->cpu_data, cpu)->last_pid);
1211 int *depth = &(per_cpu_ptr(data->cpu_data, cpu)->depth); 1392 int *depth = &(per_cpu_ptr(data->cpu_data, cpu)->depth);
1212 int *ignore = &(per_cpu_ptr(data->cpu_data, cpu)->ignore); 1393 int *ignore = &(per_cpu_ptr(data->cpu_data, cpu)->ignore);
1394 int *depth_irq = &(per_cpu_ptr(data->cpu_data, cpu)->depth_irq);
1395
1213 *pid = -1; 1396 *pid = -1;
1214 *depth = 0; 1397 *depth = 0;
1215 *ignore = 0; 1398 *ignore = 0;
1399 *depth_irq = -1;
1216 } 1400 }
1217 1401
1218 iter->private = data; 1402 iter->private = data;
@@ -1235,6 +1419,14 @@ void graph_trace_close(struct trace_iterator *iter)
1235 } 1419 }
1236} 1420}
1237 1421
1422static int func_graph_set_flag(u32 old_flags, u32 bit, int set)
1423{
1424 if (bit == TRACE_GRAPH_PRINT_IRQS)
1425 ftrace_graph_skip_irqs = !set;
1426
1427 return 0;
1428}
1429
1238static struct trace_event_functions graph_functions = { 1430static struct trace_event_functions graph_functions = {
1239 .trace = print_graph_function_event, 1431 .trace = print_graph_function_event,
1240}; 1432};
@@ -1261,6 +1453,7 @@ static struct tracer graph_trace __read_mostly = {
1261 .print_line = print_graph_function, 1453 .print_line = print_graph_function,
1262 .print_header = print_graph_headers, 1454 .print_header = print_graph_headers,
1263 .flags = &tracer_flags, 1455 .flags = &tracer_flags,
1456 .set_flag = func_graph_set_flag,
1264#ifdef CONFIG_FTRACE_SELFTEST 1457#ifdef CONFIG_FTRACE_SELFTEST
1265 .selftest = trace_selftest_startup_function_graph, 1458 .selftest = trace_selftest_startup_function_graph,
1266#endif 1459#endif
diff --git a/kernel/trace/trace_irqsoff.c b/kernel/trace/trace_irqsoff.c
index 73a6b0601f2e..5cf8c602b880 100644
--- a/kernel/trace/trace_irqsoff.c
+++ b/kernel/trace/trace_irqsoff.c
@@ -87,14 +87,22 @@ static __cacheline_aligned_in_smp unsigned long max_sequence;
87 87
88#ifdef CONFIG_FUNCTION_TRACER 88#ifdef CONFIG_FUNCTION_TRACER
89/* 89/*
90 * irqsoff uses its own tracer function to keep the overhead down: 90 * Prologue for the preempt and irqs off function tracers.
91 *
92 * Returns 1 if it is OK to continue, and data->disabled is
93 * incremented.
94 * 0 if the trace is to be ignored, and data->disabled
95 * is kept the same.
96 *
97 * Note, this function is also used outside this ifdef but
98 * inside the #ifdef of the function graph tracer below.
99 * This is OK, since the function graph tracer is
100 * dependent on the function tracer.
91 */ 101 */
92static void 102static int func_prolog_dec(struct trace_array *tr,
93irqsoff_tracer_call(unsigned long ip, unsigned long parent_ip) 103 struct trace_array_cpu **data,
104 unsigned long *flags)
94{ 105{
95 struct trace_array *tr = irqsoff_trace;
96 struct trace_array_cpu *data;
97 unsigned long flags;
98 long disabled; 106 long disabled;
99 int cpu; 107 int cpu;
100 108
@@ -106,18 +114,38 @@ irqsoff_tracer_call(unsigned long ip, unsigned long parent_ip)
106 */ 114 */
107 cpu = raw_smp_processor_id(); 115 cpu = raw_smp_processor_id();
108 if (likely(!per_cpu(tracing_cpu, cpu))) 116 if (likely(!per_cpu(tracing_cpu, cpu)))
109 return; 117 return 0;
110 118
111 local_save_flags(flags); 119 local_save_flags(*flags);
112 /* slight chance to get a false positive on tracing_cpu */ 120 /* slight chance to get a false positive on tracing_cpu */
113 if (!irqs_disabled_flags(flags)) 121 if (!irqs_disabled_flags(*flags))
114 return; 122 return 0;
115 123
116 data = tr->data[cpu]; 124 *data = tr->data[cpu];
117 disabled = atomic_inc_return(&data->disabled); 125 disabled = atomic_inc_return(&(*data)->disabled);
118 126
119 if (likely(disabled == 1)) 127 if (likely(disabled == 1))
120 trace_function(tr, ip, parent_ip, flags, preempt_count()); 128 return 1;
129
130 atomic_dec(&(*data)->disabled);
131
132 return 0;
133}
134
135/*
136 * irqsoff uses its own tracer function to keep the overhead down:
137 */
138static void
139irqsoff_tracer_call(unsigned long ip, unsigned long parent_ip)
140{
141 struct trace_array *tr = irqsoff_trace;
142 struct trace_array_cpu *data;
143 unsigned long flags;
144
145 if (!func_prolog_dec(tr, &data, &flags))
146 return;
147
148 trace_function(tr, ip, parent_ip, flags, preempt_count());
121 149
122 atomic_dec(&data->disabled); 150 atomic_dec(&data->disabled);
123} 151}
@@ -155,30 +183,16 @@ static int irqsoff_graph_entry(struct ftrace_graph_ent *trace)
155 struct trace_array *tr = irqsoff_trace; 183 struct trace_array *tr = irqsoff_trace;
156 struct trace_array_cpu *data; 184 struct trace_array_cpu *data;
157 unsigned long flags; 185 unsigned long flags;
158 long disabled;
159 int ret; 186 int ret;
160 int cpu;
161 int pc; 187 int pc;
162 188
163 cpu = raw_smp_processor_id(); 189 if (!func_prolog_dec(tr, &data, &flags))
164 if (likely(!per_cpu(tracing_cpu, cpu)))
165 return 0; 190 return 0;
166 191
167 local_save_flags(flags); 192 pc = preempt_count();
168 /* slight chance to get a false positive on tracing_cpu */ 193 ret = __trace_graph_entry(tr, trace, flags, pc);
169 if (!irqs_disabled_flags(flags))
170 return 0;
171
172 data = tr->data[cpu];
173 disabled = atomic_inc_return(&data->disabled);
174
175 if (likely(disabled == 1)) {
176 pc = preempt_count();
177 ret = __trace_graph_entry(tr, trace, flags, pc);
178 } else
179 ret = 0;
180
181 atomic_dec(&data->disabled); 194 atomic_dec(&data->disabled);
195
182 return ret; 196 return ret;
183} 197}
184 198
@@ -187,27 +201,13 @@ static void irqsoff_graph_return(struct ftrace_graph_ret *trace)
187 struct trace_array *tr = irqsoff_trace; 201 struct trace_array *tr = irqsoff_trace;
188 struct trace_array_cpu *data; 202 struct trace_array_cpu *data;
189 unsigned long flags; 203 unsigned long flags;
190 long disabled;
191 int cpu;
192 int pc; 204 int pc;
193 205
194 cpu = raw_smp_processor_id(); 206 if (!func_prolog_dec(tr, &data, &flags))
195 if (likely(!per_cpu(tracing_cpu, cpu)))
196 return; 207 return;
197 208
198 local_save_flags(flags); 209 pc = preempt_count();
199 /* slight chance to get a false positive on tracing_cpu */ 210 __trace_graph_return(tr, trace, flags, pc);
200 if (!irqs_disabled_flags(flags))
201 return;
202
203 data = tr->data[cpu];
204 disabled = atomic_inc_return(&data->disabled);
205
206 if (likely(disabled == 1)) {
207 pc = preempt_count();
208 __trace_graph_return(tr, trace, flags, pc);
209 }
210
211 atomic_dec(&data->disabled); 211 atomic_dec(&data->disabled);
212} 212}
213 213
@@ -229,75 +229,33 @@ static void irqsoff_trace_close(struct trace_iterator *iter)
229 229
230static enum print_line_t irqsoff_print_line(struct trace_iterator *iter) 230static enum print_line_t irqsoff_print_line(struct trace_iterator *iter)
231{ 231{
232 u32 flags = GRAPH_TRACER_FLAGS;
233
234 if (trace_flags & TRACE_ITER_LATENCY_FMT)
235 flags |= TRACE_GRAPH_PRINT_DURATION;
236 else
237 flags |= TRACE_GRAPH_PRINT_ABS_TIME;
238
239 /* 232 /*
240 * In graph mode call the graph tracer output function, 233 * In graph mode call the graph tracer output function,
241 * otherwise go with the TRACE_FN event handler 234 * otherwise go with the TRACE_FN event handler
242 */ 235 */
243 if (is_graph()) 236 if (is_graph())
244 return print_graph_function_flags(iter, flags); 237 return print_graph_function_flags(iter, GRAPH_TRACER_FLAGS);
245 238
246 return TRACE_TYPE_UNHANDLED; 239 return TRACE_TYPE_UNHANDLED;
247} 240}
248 241
249static void irqsoff_print_header(struct seq_file *s) 242static void irqsoff_print_header(struct seq_file *s)
250{ 243{
251 if (is_graph()) { 244 if (is_graph())
252 struct trace_iterator *iter = s->private; 245 print_graph_headers_flags(s, GRAPH_TRACER_FLAGS);
253 u32 flags = GRAPH_TRACER_FLAGS; 246 else
254
255 if (trace_flags & TRACE_ITER_LATENCY_FMT) {
256 /* print nothing if the buffers are empty */
257 if (trace_empty(iter))
258 return;
259
260 print_trace_header(s, iter);
261 flags |= TRACE_GRAPH_PRINT_DURATION;
262 } else
263 flags |= TRACE_GRAPH_PRINT_ABS_TIME;
264
265 print_graph_headers_flags(s, flags);
266 } else
267 trace_default_header(s); 247 trace_default_header(s);
268} 248}
269 249
270static void 250static void
271trace_graph_function(struct trace_array *tr,
272 unsigned long ip, unsigned long flags, int pc)
273{
274 u64 time = trace_clock_local();
275 struct ftrace_graph_ent ent = {
276 .func = ip,
277 .depth = 0,
278 };
279 struct ftrace_graph_ret ret = {
280 .func = ip,
281 .depth = 0,
282 .calltime = time,
283 .rettime = time,
284 };
285
286 __trace_graph_entry(tr, &ent, flags, pc);
287 __trace_graph_return(tr, &ret, flags, pc);
288}
289
290static void
291__trace_function(struct trace_array *tr, 251__trace_function(struct trace_array *tr,
292 unsigned long ip, unsigned long parent_ip, 252 unsigned long ip, unsigned long parent_ip,
293 unsigned long flags, int pc) 253 unsigned long flags, int pc)
294{ 254{
295 if (!is_graph()) 255 if (is_graph())
256 trace_graph_function(tr, ip, parent_ip, flags, pc);
257 else
296 trace_function(tr, ip, parent_ip, flags, pc); 258 trace_function(tr, ip, parent_ip, flags, pc);
297 else {
298 trace_graph_function(tr, parent_ip, flags, pc);
299 trace_graph_function(tr, ip, flags, pc);
300 }
301} 259}
302 260
303#else 261#else
diff --git a/kernel/trace/trace_kdb.c b/kernel/trace/trace_kdb.c
index 7b8ecd751d93..3c5c5dfea0b3 100644
--- a/kernel/trace/trace_kdb.c
+++ b/kernel/trace/trace_kdb.c
@@ -13,7 +13,6 @@
13#include <linux/kdb.h> 13#include <linux/kdb.h>
14#include <linux/ftrace.h> 14#include <linux/ftrace.h>
15 15
16#include "../debug/kdb/kdb_private.h"
17#include "trace.h" 16#include "trace.h"
18#include "trace_output.h" 17#include "trace_output.h"
19 18
diff --git a/kernel/trace/trace_sched_wakeup.c b/kernel/trace/trace_sched_wakeup.c
index 4086eae6e81b..7319559ed59f 100644
--- a/kernel/trace/trace_sched_wakeup.c
+++ b/kernel/trace/trace_sched_wakeup.c
@@ -31,48 +31,98 @@ static int wakeup_rt;
31static arch_spinlock_t wakeup_lock = 31static arch_spinlock_t wakeup_lock =
32 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED; 32 (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
33 33
34static void wakeup_reset(struct trace_array *tr);
34static void __wakeup_reset(struct trace_array *tr); 35static void __wakeup_reset(struct trace_array *tr);
36static int wakeup_graph_entry(struct ftrace_graph_ent *trace);
37static void wakeup_graph_return(struct ftrace_graph_ret *trace);
35 38
36static int save_lat_flag; 39static int save_lat_flag;
37 40
41#define TRACE_DISPLAY_GRAPH 1
42
43static struct tracer_opt trace_opts[] = {
44#ifdef CONFIG_FUNCTION_GRAPH_TRACER
45 /* display latency trace as call graph */
46 { TRACER_OPT(display-graph, TRACE_DISPLAY_GRAPH) },
47#endif
48 { } /* Empty entry */
49};
50
51static struct tracer_flags tracer_flags = {
52 .val = 0,
53 .opts = trace_opts,
54};
55
56#define is_graph() (tracer_flags.val & TRACE_DISPLAY_GRAPH)
57
38#ifdef CONFIG_FUNCTION_TRACER 58#ifdef CONFIG_FUNCTION_TRACER
59
39/* 60/*
40 * irqsoff uses its own tracer function to keep the overhead down: 61 * Prologue for the wakeup function tracers.
62 *
63 * Returns 1 if it is OK to continue, and preemption
64 * is disabled and data->disabled is incremented.
65 * 0 if the trace is to be ignored, and preemption
66 * is not disabled and data->disabled is
67 * kept the same.
68 *
69 * Note, this function is also used outside this ifdef but
70 * inside the #ifdef of the function graph tracer below.
71 * This is OK, since the function graph tracer is
72 * dependent on the function tracer.
41 */ 73 */
42static void 74static int
43wakeup_tracer_call(unsigned long ip, unsigned long parent_ip) 75func_prolog_preempt_disable(struct trace_array *tr,
76 struct trace_array_cpu **data,
77 int *pc)
44{ 78{
45 struct trace_array *tr = wakeup_trace;
46 struct trace_array_cpu *data;
47 unsigned long flags;
48 long disabled; 79 long disabled;
49 int cpu; 80 int cpu;
50 int pc;
51 81
52 if (likely(!wakeup_task)) 82 if (likely(!wakeup_task))
53 return; 83 return 0;
54 84
55 pc = preempt_count(); 85 *pc = preempt_count();
56 preempt_disable_notrace(); 86 preempt_disable_notrace();
57 87
58 cpu = raw_smp_processor_id(); 88 cpu = raw_smp_processor_id();
59 if (cpu != wakeup_current_cpu) 89 if (cpu != wakeup_current_cpu)
60 goto out_enable; 90 goto out_enable;
61 91
62 data = tr->data[cpu]; 92 *data = tr->data[cpu];
63 disabled = atomic_inc_return(&data->disabled); 93 disabled = atomic_inc_return(&(*data)->disabled);
64 if (unlikely(disabled != 1)) 94 if (unlikely(disabled != 1))
65 goto out; 95 goto out;
66 96
67 local_irq_save(flags); 97 return 1;
68 98
69 trace_function(tr, ip, parent_ip, flags, pc); 99out:
100 atomic_dec(&(*data)->disabled);
101
102out_enable:
103 preempt_enable_notrace();
104 return 0;
105}
70 106
107/*
108 * wakeup uses its own tracer function to keep the overhead down:
109 */
110static void
111wakeup_tracer_call(unsigned long ip, unsigned long parent_ip)
112{
113 struct trace_array *tr = wakeup_trace;
114 struct trace_array_cpu *data;
115 unsigned long flags;
116 int pc;
117
118 if (!func_prolog_preempt_disable(tr, &data, &pc))
119 return;
120
121 local_irq_save(flags);
122 trace_function(tr, ip, parent_ip, flags, pc);
71 local_irq_restore(flags); 123 local_irq_restore(flags);
72 124
73 out:
74 atomic_dec(&data->disabled); 125 atomic_dec(&data->disabled);
75 out_enable:
76 preempt_enable_notrace(); 126 preempt_enable_notrace();
77} 127}
78 128
@@ -82,6 +132,156 @@ static struct ftrace_ops trace_ops __read_mostly =
82}; 132};
83#endif /* CONFIG_FUNCTION_TRACER */ 133#endif /* CONFIG_FUNCTION_TRACER */
84 134
135static int start_func_tracer(int graph)
136{
137 int ret;
138
139 if (!graph)
140 ret = register_ftrace_function(&trace_ops);
141 else
142 ret = register_ftrace_graph(&wakeup_graph_return,
143 &wakeup_graph_entry);
144
145 if (!ret && tracing_is_enabled())
146 tracer_enabled = 1;
147 else
148 tracer_enabled = 0;
149
150 return ret;
151}
152
153static void stop_func_tracer(int graph)
154{
155 tracer_enabled = 0;
156
157 if (!graph)
158 unregister_ftrace_function(&trace_ops);
159 else
160 unregister_ftrace_graph();
161}
162
163#ifdef CONFIG_FUNCTION_GRAPH_TRACER
164static int wakeup_set_flag(u32 old_flags, u32 bit, int set)
165{
166
167 if (!(bit & TRACE_DISPLAY_GRAPH))
168 return -EINVAL;
169
170 if (!(is_graph() ^ set))
171 return 0;
172
173 stop_func_tracer(!set);
174
175 wakeup_reset(wakeup_trace);
176 tracing_max_latency = 0;
177
178 return start_func_tracer(set);
179}
180
181static int wakeup_graph_entry(struct ftrace_graph_ent *trace)
182{
183 struct trace_array *tr = wakeup_trace;
184 struct trace_array_cpu *data;
185 unsigned long flags;
186 int pc, ret = 0;
187
188 if (!func_prolog_preempt_disable(tr, &data, &pc))
189 return 0;
190
191 local_save_flags(flags);
192 ret = __trace_graph_entry(tr, trace, flags, pc);
193 atomic_dec(&data->disabled);
194 preempt_enable_notrace();
195
196 return ret;
197}
198
199static void wakeup_graph_return(struct ftrace_graph_ret *trace)
200{
201 struct trace_array *tr = wakeup_trace;
202 struct trace_array_cpu *data;
203 unsigned long flags;
204 int pc;
205
206 if (!func_prolog_preempt_disable(tr, &data, &pc))
207 return;
208
209 local_save_flags(flags);
210 __trace_graph_return(tr, trace, flags, pc);
211 atomic_dec(&data->disabled);
212
213 preempt_enable_notrace();
214 return;
215}
216
217static void wakeup_trace_open(struct trace_iterator *iter)
218{
219 if (is_graph())
220 graph_trace_open(iter);
221}
222
223static void wakeup_trace_close(struct trace_iterator *iter)
224{
225 if (iter->private)
226 graph_trace_close(iter);
227}
228
229#define GRAPH_TRACER_FLAGS (TRACE_GRAPH_PRINT_PROC)
230
231static enum print_line_t wakeup_print_line(struct trace_iterator *iter)
232{
233 /*
234 * In graph mode call the graph tracer output function,
235 * otherwise go with the TRACE_FN event handler
236 */
237 if (is_graph())
238 return print_graph_function_flags(iter, GRAPH_TRACER_FLAGS);
239
240 return TRACE_TYPE_UNHANDLED;
241}
242
243static void wakeup_print_header(struct seq_file *s)
244{
245 if (is_graph())
246 print_graph_headers_flags(s, GRAPH_TRACER_FLAGS);
247 else
248 trace_default_header(s);
249}
250
251static void
252__trace_function(struct trace_array *tr,
253 unsigned long ip, unsigned long parent_ip,
254 unsigned long flags, int pc)
255{
256 if (is_graph())
257 trace_graph_function(tr, ip, parent_ip, flags, pc);
258 else
259 trace_function(tr, ip, parent_ip, flags, pc);
260}
261#else
262#define __trace_function trace_function
263
264static int wakeup_set_flag(u32 old_flags, u32 bit, int set)
265{
266 return -EINVAL;
267}
268
269static int wakeup_graph_entry(struct ftrace_graph_ent *trace)
270{
271 return -1;
272}
273
274static enum print_line_t wakeup_print_line(struct trace_iterator *iter)
275{
276 return TRACE_TYPE_UNHANDLED;
277}
278
279static void wakeup_graph_return(struct ftrace_graph_ret *trace) { }
280static void wakeup_print_header(struct seq_file *s) { }
281static void wakeup_trace_open(struct trace_iterator *iter) { }
282static void wakeup_trace_close(struct trace_iterator *iter) { }
283#endif /* CONFIG_FUNCTION_GRAPH_TRACER */
284
85/* 285/*
86 * Should this new latency be reported/recorded? 286 * Should this new latency be reported/recorded?
87 */ 287 */
@@ -152,7 +352,7 @@ probe_wakeup_sched_switch(void *ignore,
152 /* The task we are waiting for is waking up */ 352 /* The task we are waiting for is waking up */
153 data = wakeup_trace->data[wakeup_cpu]; 353 data = wakeup_trace->data[wakeup_cpu];
154 354
155 trace_function(wakeup_trace, CALLER_ADDR0, CALLER_ADDR1, flags, pc); 355 __trace_function(wakeup_trace, CALLER_ADDR0, CALLER_ADDR1, flags, pc);
156 tracing_sched_switch_trace(wakeup_trace, prev, next, flags, pc); 356 tracing_sched_switch_trace(wakeup_trace, prev, next, flags, pc);
157 357
158 T0 = data->preempt_timestamp; 358 T0 = data->preempt_timestamp;
@@ -252,7 +452,7 @@ probe_wakeup(void *ignore, struct task_struct *p, int success)
252 * is not called by an assembly function (where as schedule is) 452 * is not called by an assembly function (where as schedule is)
253 * it should be safe to use it here. 453 * it should be safe to use it here.
254 */ 454 */
255 trace_function(wakeup_trace, CALLER_ADDR1, CALLER_ADDR2, flags, pc); 455 __trace_function(wakeup_trace, CALLER_ADDR1, CALLER_ADDR2, flags, pc);
256 456
257out_locked: 457out_locked:
258 arch_spin_unlock(&wakeup_lock); 458 arch_spin_unlock(&wakeup_lock);
@@ -303,12 +503,8 @@ static void start_wakeup_tracer(struct trace_array *tr)
303 */ 503 */
304 smp_wmb(); 504 smp_wmb();
305 505
306 register_ftrace_function(&trace_ops); 506 if (start_func_tracer(is_graph()))
307 507 printk(KERN_ERR "failed to start wakeup tracer\n");
308 if (tracing_is_enabled())
309 tracer_enabled = 1;
310 else
311 tracer_enabled = 0;
312 508
313 return; 509 return;
314fail_deprobe_wake_new: 510fail_deprobe_wake_new:
@@ -320,7 +516,7 @@ fail_deprobe:
320static void stop_wakeup_tracer(struct trace_array *tr) 516static void stop_wakeup_tracer(struct trace_array *tr)
321{ 517{
322 tracer_enabled = 0; 518 tracer_enabled = 0;
323 unregister_ftrace_function(&trace_ops); 519 stop_func_tracer(is_graph());
324 unregister_trace_sched_switch(probe_wakeup_sched_switch, NULL); 520 unregister_trace_sched_switch(probe_wakeup_sched_switch, NULL);
325 unregister_trace_sched_wakeup_new(probe_wakeup, NULL); 521 unregister_trace_sched_wakeup_new(probe_wakeup, NULL);
326 unregister_trace_sched_wakeup(probe_wakeup, NULL); 522 unregister_trace_sched_wakeup(probe_wakeup, NULL);
@@ -379,9 +575,15 @@ static struct tracer wakeup_tracer __read_mostly =
379 .start = wakeup_tracer_start, 575 .start = wakeup_tracer_start,
380 .stop = wakeup_tracer_stop, 576 .stop = wakeup_tracer_stop,
381 .print_max = 1, 577 .print_max = 1,
578 .print_header = wakeup_print_header,
579 .print_line = wakeup_print_line,
580 .flags = &tracer_flags,
581 .set_flag = wakeup_set_flag,
382#ifdef CONFIG_FTRACE_SELFTEST 582#ifdef CONFIG_FTRACE_SELFTEST
383 .selftest = trace_selftest_startup_wakeup, 583 .selftest = trace_selftest_startup_wakeup,
384#endif 584#endif
585 .open = wakeup_trace_open,
586 .close = wakeup_trace_close,
385 .use_max_tr = 1, 587 .use_max_tr = 1,
386}; 588};
387 589
@@ -394,9 +596,15 @@ static struct tracer wakeup_rt_tracer __read_mostly =
394 .stop = wakeup_tracer_stop, 596 .stop = wakeup_tracer_stop,
395 .wait_pipe = poll_wait_pipe, 597 .wait_pipe = poll_wait_pipe,
396 .print_max = 1, 598 .print_max = 1,
599 .print_header = wakeup_print_header,
600 .print_line = wakeup_print_line,
601 .flags = &tracer_flags,
602 .set_flag = wakeup_set_flag,
397#ifdef CONFIG_FTRACE_SELFTEST 603#ifdef CONFIG_FTRACE_SELFTEST
398 .selftest = trace_selftest_startup_wakeup, 604 .selftest = trace_selftest_startup_wakeup,
399#endif 605#endif
606 .open = wakeup_trace_open,
607 .close = wakeup_trace_close,
400 .use_max_tr = 1, 608 .use_max_tr = 1,
401}; 609};
402 610
diff --git a/kernel/trace/trace_stack.c b/kernel/trace/trace_stack.c
index a6b7e0e0f3eb..4c5dead0c239 100644
--- a/kernel/trace/trace_stack.c
+++ b/kernel/trace/trace_stack.c
@@ -195,6 +195,7 @@ static const struct file_operations stack_max_size_fops = {
195 .open = tracing_open_generic, 195 .open = tracing_open_generic,
196 .read = stack_max_size_read, 196 .read = stack_max_size_read,
197 .write = stack_max_size_write, 197 .write = stack_max_size_write,
198 .llseek = default_llseek,
198}; 199};
199 200
200static void * 201static void *
diff --git a/kernel/trace/trace_workqueue.c b/kernel/trace/trace_workqueue.c
index a7cc3793baf6..209b379a4721 100644
--- a/kernel/trace/trace_workqueue.c
+++ b/kernel/trace/trace_workqueue.c
@@ -263,6 +263,11 @@ int __init trace_workqueue_early_init(void)
263{ 263{
264 int ret, cpu; 264 int ret, cpu;
265 265
266 for_each_possible_cpu(cpu) {
267 spin_lock_init(&workqueue_cpu_stat(cpu)->lock);
268 INIT_LIST_HEAD(&workqueue_cpu_stat(cpu)->list);
269 }
270
266 ret = register_trace_workqueue_insertion(probe_workqueue_insertion, NULL); 271 ret = register_trace_workqueue_insertion(probe_workqueue_insertion, NULL);
267 if (ret) 272 if (ret)
268 goto out; 273 goto out;
@@ -279,11 +284,6 @@ int __init trace_workqueue_early_init(void)
279 if (ret) 284 if (ret)
280 goto no_creation; 285 goto no_creation;
281 286
282 for_each_possible_cpu(cpu) {
283 spin_lock_init(&workqueue_cpu_stat(cpu)->lock);
284 INIT_LIST_HEAD(&workqueue_cpu_stat(cpu)->list);
285 }
286
287 return 0; 287 return 0;
288 288
289no_creation: 289no_creation:
diff --git a/kernel/tracepoint.c b/kernel/tracepoint.c
index c77f3eceea25..e95ee7f31d43 100644
--- a/kernel/tracepoint.c
+++ b/kernel/tracepoint.c
@@ -25,6 +25,7 @@
25#include <linux/err.h> 25#include <linux/err.h>
26#include <linux/slab.h> 26#include <linux/slab.h>
27#include <linux/sched.h> 27#include <linux/sched.h>
28#include <linux/jump_label.h>
28 29
29extern struct tracepoint __start___tracepoints[]; 30extern struct tracepoint __start___tracepoints[];
30extern struct tracepoint __stop___tracepoints[]; 31extern struct tracepoint __stop___tracepoints[];
@@ -263,7 +264,13 @@ static void set_tracepoint(struct tracepoint_entry **entry,
263 * is used. 264 * is used.
264 */ 265 */
265 rcu_assign_pointer(elem->funcs, (*entry)->funcs); 266 rcu_assign_pointer(elem->funcs, (*entry)->funcs);
266 elem->state = active; 267 if (!elem->state && active) {
268 jump_label_enable(&elem->state);
269 elem->state = active;
270 } else if (elem->state && !active) {
271 jump_label_disable(&elem->state);
272 elem->state = active;
273 }
267} 274}
268 275
269/* 276/*
@@ -277,7 +284,10 @@ static void disable_tracepoint(struct tracepoint *elem)
277 if (elem->unregfunc && elem->state) 284 if (elem->unregfunc && elem->state)
278 elem->unregfunc(); 285 elem->unregfunc();
279 286
280 elem->state = 0; 287 if (elem->state) {
288 jump_label_disable(&elem->state);
289 elem->state = 0;
290 }
281 rcu_assign_pointer(elem->funcs, NULL); 291 rcu_assign_pointer(elem->funcs, NULL);
282} 292}
283 293
diff --git a/kernel/watchdog.c b/kernel/watchdog.c
index 7f9c3c52ecc1..bafba687a6d8 100644
--- a/kernel/watchdog.c
+++ b/kernel/watchdog.c
@@ -43,7 +43,6 @@ static DEFINE_PER_CPU(unsigned long, hrtimer_interrupts_saved);
43static DEFINE_PER_CPU(struct perf_event *, watchdog_ev); 43static DEFINE_PER_CPU(struct perf_event *, watchdog_ev);
44#endif 44#endif
45 45
46static int __read_mostly did_panic;
47static int __initdata no_watchdog; 46static int __initdata no_watchdog;
48 47
49 48
@@ -187,18 +186,6 @@ static int is_softlockup(unsigned long touch_ts)
187 return 0; 186 return 0;
188} 187}
189 188
190static int
191watchdog_panic(struct notifier_block *this, unsigned long event, void *ptr)
192{
193 did_panic = 1;
194
195 return NOTIFY_DONE;
196}
197
198static struct notifier_block panic_block = {
199 .notifier_call = watchdog_panic,
200};
201
202#ifdef CONFIG_HARDLOCKUP_DETECTOR 189#ifdef CONFIG_HARDLOCKUP_DETECTOR
203static struct perf_event_attr wd_hw_attr = { 190static struct perf_event_attr wd_hw_attr = {
204 .type = PERF_TYPE_HARDWARE, 191 .type = PERF_TYPE_HARDWARE,
@@ -209,7 +196,7 @@ static struct perf_event_attr wd_hw_attr = {
209}; 196};
210 197
211/* Callback function for perf event subsystem */ 198/* Callback function for perf event subsystem */
212void watchdog_overflow_callback(struct perf_event *event, int nmi, 199static void watchdog_overflow_callback(struct perf_event *event, int nmi,
213 struct perf_sample_data *data, 200 struct perf_sample_data *data,
214 struct pt_regs *regs) 201 struct pt_regs *regs)
215{ 202{
@@ -371,14 +358,14 @@ static int watchdog_nmi_enable(int cpu)
371 /* Try to register using hardware perf events */ 358 /* Try to register using hardware perf events */
372 wd_attr = &wd_hw_attr; 359 wd_attr = &wd_hw_attr;
373 wd_attr->sample_period = hw_nmi_get_sample_period(); 360 wd_attr->sample_period = hw_nmi_get_sample_period();
374 event = perf_event_create_kernel_counter(wd_attr, cpu, -1, watchdog_overflow_callback); 361 event = perf_event_create_kernel_counter(wd_attr, cpu, NULL, watchdog_overflow_callback);
375 if (!IS_ERR(event)) { 362 if (!IS_ERR(event)) {
376 printk(KERN_INFO "NMI watchdog enabled, takes one hw-pmu counter.\n"); 363 printk(KERN_INFO "NMI watchdog enabled, takes one hw-pmu counter.\n");
377 goto out_save; 364 goto out_save;
378 } 365 }
379 366
380 printk(KERN_ERR "NMI watchdog failed to create perf event on cpu%i: %p\n", cpu, event); 367 printk(KERN_ERR "NMI watchdog failed to create perf event on cpu%i: %p\n", cpu, event);
381 return -1; 368 return PTR_ERR(event);
382 369
383 /* success path */ 370 /* success path */
384out_save: 371out_save:
@@ -422,17 +409,19 @@ static int watchdog_prepare_cpu(int cpu)
422static int watchdog_enable(int cpu) 409static int watchdog_enable(int cpu)
423{ 410{
424 struct task_struct *p = per_cpu(softlockup_watchdog, cpu); 411 struct task_struct *p = per_cpu(softlockup_watchdog, cpu);
412 int err;
425 413
426 /* enable the perf event */ 414 /* enable the perf event */
427 if (watchdog_nmi_enable(cpu) != 0) 415 err = watchdog_nmi_enable(cpu);
428 return -1; 416 if (err)
417 return err;
429 418
430 /* create the watchdog thread */ 419 /* create the watchdog thread */
431 if (!p) { 420 if (!p) {
432 p = kthread_create(watchdog, (void *)(unsigned long)cpu, "watchdog/%d", cpu); 421 p = kthread_create(watchdog, (void *)(unsigned long)cpu, "watchdog/%d", cpu);
433 if (IS_ERR(p)) { 422 if (IS_ERR(p)) {
434 printk(KERN_ERR "softlockup watchdog for %i failed\n", cpu); 423 printk(KERN_ERR "softlockup watchdog for %i failed\n", cpu);
435 return -1; 424 return PTR_ERR(p);
436 } 425 }
437 kthread_bind(p, cpu); 426 kthread_bind(p, cpu);
438 per_cpu(watchdog_touch_ts, cpu) = 0; 427 per_cpu(watchdog_touch_ts, cpu) = 0;
@@ -484,6 +473,9 @@ static void watchdog_disable_all_cpus(void)
484{ 473{
485 int cpu; 474 int cpu;
486 475
476 if (no_watchdog)
477 return;
478
487 for_each_online_cpu(cpu) 479 for_each_online_cpu(cpu)
488 watchdog_disable(cpu); 480 watchdog_disable(cpu);
489 481
@@ -526,17 +518,16 @@ static int __cpuinit
526cpu_callback(struct notifier_block *nfb, unsigned long action, void *hcpu) 518cpu_callback(struct notifier_block *nfb, unsigned long action, void *hcpu)
527{ 519{
528 int hotcpu = (unsigned long)hcpu; 520 int hotcpu = (unsigned long)hcpu;
521 int err = 0;
529 522
530 switch (action) { 523 switch (action) {
531 case CPU_UP_PREPARE: 524 case CPU_UP_PREPARE:
532 case CPU_UP_PREPARE_FROZEN: 525 case CPU_UP_PREPARE_FROZEN:
533 if (watchdog_prepare_cpu(hotcpu)) 526 err = watchdog_prepare_cpu(hotcpu);
534 return NOTIFY_BAD;
535 break; 527 break;
536 case CPU_ONLINE: 528 case CPU_ONLINE:
537 case CPU_ONLINE_FROZEN: 529 case CPU_ONLINE_FROZEN:
538 if (watchdog_enable(hotcpu)) 530 err = watchdog_enable(hotcpu);
539 return NOTIFY_BAD;
540 break; 531 break;
541#ifdef CONFIG_HOTPLUG_CPU 532#ifdef CONFIG_HOTPLUG_CPU
542 case CPU_UP_CANCELED: 533 case CPU_UP_CANCELED:
@@ -549,7 +540,7 @@ cpu_callback(struct notifier_block *nfb, unsigned long action, void *hcpu)
549 break; 540 break;
550#endif /* CONFIG_HOTPLUG_CPU */ 541#endif /* CONFIG_HOTPLUG_CPU */
551 } 542 }
552 return NOTIFY_OK; 543 return notifier_from_errno(err);
553} 544}
554 545
555static struct notifier_block __cpuinitdata cpu_nfb = { 546static struct notifier_block __cpuinitdata cpu_nfb = {
@@ -565,13 +556,11 @@ static int __init spawn_watchdog_task(void)
565 return 0; 556 return 0;
566 557
567 err = cpu_callback(&cpu_nfb, CPU_UP_PREPARE, cpu); 558 err = cpu_callback(&cpu_nfb, CPU_UP_PREPARE, cpu);
568 WARN_ON(err == NOTIFY_BAD); 559 WARN_ON(notifier_to_errno(err));
569 560
570 cpu_callback(&cpu_nfb, CPU_ONLINE, cpu); 561 cpu_callback(&cpu_nfb, CPU_ONLINE, cpu);
571 register_cpu_notifier(&cpu_nfb); 562 register_cpu_notifier(&cpu_nfb);
572 563
573 atomic_notifier_chain_register(&panic_notifier_list, &panic_block);
574
575 return 0; 564 return 0;
576} 565}
577early_initcall(spawn_watchdog_task); 566early_initcall(spawn_watchdog_task);
diff --git a/kernel/workqueue.c b/kernel/workqueue.c
index f77afd939229..30acdb74cc23 100644
--- a/kernel/workqueue.c
+++ b/kernel/workqueue.c
@@ -42,9 +42,6 @@
42#include <linux/lockdep.h> 42#include <linux/lockdep.h>
43#include <linux/idr.h> 43#include <linux/idr.h>
44 44
45#define CREATE_TRACE_POINTS
46#include <trace/events/workqueue.h>
47
48#include "workqueue_sched.h" 45#include "workqueue_sched.h"
49 46
50enum { 47enum {
@@ -257,6 +254,9 @@ EXPORT_SYMBOL_GPL(system_long_wq);
257EXPORT_SYMBOL_GPL(system_nrt_wq); 254EXPORT_SYMBOL_GPL(system_nrt_wq);
258EXPORT_SYMBOL_GPL(system_unbound_wq); 255EXPORT_SYMBOL_GPL(system_unbound_wq);
259 256
257#define CREATE_TRACE_POINTS
258#include <trace/events/workqueue.h>
259
260#define for_each_busy_worker(worker, i, pos, gcwq) \ 260#define for_each_busy_worker(worker, i, pos, gcwq) \
261 for (i = 0; i < BUSY_WORKER_HASH_SIZE; i++) \ 261 for (i = 0; i < BUSY_WORKER_HASH_SIZE; i++) \
262 hlist_for_each_entry(worker, pos, &gcwq->busy_hash[i], hentry) 262 hlist_for_each_entry(worker, pos, &gcwq->busy_hash[i], hentry)
@@ -310,21 +310,6 @@ static inline int __next_wq_cpu(int cpu, const struct cpumask *mask,
310 (cpu) < WORK_CPU_NONE; \ 310 (cpu) < WORK_CPU_NONE; \
311 (cpu) = __next_wq_cpu((cpu), cpu_possible_mask, (wq))) 311 (cpu) = __next_wq_cpu((cpu), cpu_possible_mask, (wq)))
312 312
313#ifdef CONFIG_LOCKDEP
314/**
315 * in_workqueue_context() - in context of specified workqueue?
316 * @wq: the workqueue of interest
317 *
318 * Checks lockdep state to see if the current task is executing from
319 * within a workqueue item. This function exists only if lockdep is
320 * enabled.
321 */
322int in_workqueue_context(struct workqueue_struct *wq)
323{
324 return lock_is_held(&wq->lockdep_map);
325}
326#endif
327
328#ifdef CONFIG_DEBUG_OBJECTS_WORK 313#ifdef CONFIG_DEBUG_OBJECTS_WORK
329 314
330static struct debug_obj_descr work_debug_descr; 315static struct debug_obj_descr work_debug_descr;
@@ -604,7 +589,9 @@ static bool keep_working(struct global_cwq *gcwq)
604{ 589{
605 atomic_t *nr_running = get_gcwq_nr_running(gcwq->cpu); 590 atomic_t *nr_running = get_gcwq_nr_running(gcwq->cpu);
606 591
607 return !list_empty(&gcwq->worklist) && atomic_read(nr_running) <= 1; 592 return !list_empty(&gcwq->worklist) &&
593 (atomic_read(nr_running) <= 1 ||
594 gcwq->flags & GCWQ_HIGHPRI_PENDING);
608} 595}
609 596
610/* Do we need a new worker? Called from manager. */ 597/* Do we need a new worker? Called from manager. */
@@ -997,6 +984,7 @@ static void __queue_work(unsigned int cpu, struct workqueue_struct *wq,
997 984
998 /* gcwq determined, get cwq and queue */ 985 /* gcwq determined, get cwq and queue */
999 cwq = get_cwq(gcwq->cpu, wq); 986 cwq = get_cwq(gcwq->cpu, wq);
987 trace_workqueue_queue_work(cpu, cwq, work);
1000 988
1001 BUG_ON(!list_empty(&work->entry)); 989 BUG_ON(!list_empty(&work->entry));
1002 990
@@ -1004,6 +992,7 @@ static void __queue_work(unsigned int cpu, struct workqueue_struct *wq,
1004 work_flags = work_color_to_flags(cwq->work_color); 992 work_flags = work_color_to_flags(cwq->work_color);
1005 993
1006 if (likely(cwq->nr_active < cwq->max_active)) { 994 if (likely(cwq->nr_active < cwq->max_active)) {
995 trace_workqueue_activate_work(work);
1007 cwq->nr_active++; 996 cwq->nr_active++;
1008 worklist = gcwq_determine_ins_pos(gcwq, cwq); 997 worklist = gcwq_determine_ins_pos(gcwq, cwq);
1009 } else { 998 } else {
@@ -1679,6 +1668,7 @@ static void cwq_activate_first_delayed(struct cpu_workqueue_struct *cwq)
1679 struct work_struct, entry); 1668 struct work_struct, entry);
1680 struct list_head *pos = gcwq_determine_ins_pos(cwq->gcwq, cwq); 1669 struct list_head *pos = gcwq_determine_ins_pos(cwq->gcwq, cwq);
1681 1670
1671 trace_workqueue_activate_work(work);
1682 move_linked_works(work, pos, NULL); 1672 move_linked_works(work, pos, NULL);
1683 __clear_bit(WORK_STRUCT_DELAYED_BIT, work_data_bits(work)); 1673 __clear_bit(WORK_STRUCT_DELAYED_BIT, work_data_bits(work));
1684 cwq->nr_active++; 1674 cwq->nr_active++;
@@ -2326,27 +2316,17 @@ out_unlock:
2326} 2316}
2327EXPORT_SYMBOL_GPL(flush_workqueue); 2317EXPORT_SYMBOL_GPL(flush_workqueue);
2328 2318
2329/** 2319static bool start_flush_work(struct work_struct *work, struct wq_barrier *barr,
2330 * flush_work - block until a work_struct's callback has terminated 2320 bool wait_executing)
2331 * @work: the work which is to be flushed
2332 *
2333 * Returns false if @work has already terminated.
2334 *
2335 * It is expected that, prior to calling flush_work(), the caller has
2336 * arranged for the work to not be requeued, otherwise it doesn't make
2337 * sense to use this function.
2338 */
2339int flush_work(struct work_struct *work)
2340{ 2321{
2341 struct worker *worker = NULL; 2322 struct worker *worker = NULL;
2342 struct global_cwq *gcwq; 2323 struct global_cwq *gcwq;
2343 struct cpu_workqueue_struct *cwq; 2324 struct cpu_workqueue_struct *cwq;
2344 struct wq_barrier barr;
2345 2325
2346 might_sleep(); 2326 might_sleep();
2347 gcwq = get_work_gcwq(work); 2327 gcwq = get_work_gcwq(work);
2348 if (!gcwq) 2328 if (!gcwq)
2349 return 0; 2329 return false;
2350 2330
2351 spin_lock_irq(&gcwq->lock); 2331 spin_lock_irq(&gcwq->lock);
2352 if (!list_empty(&work->entry)) { 2332 if (!list_empty(&work->entry)) {
@@ -2359,28 +2339,127 @@ int flush_work(struct work_struct *work)
2359 cwq = get_work_cwq(work); 2339 cwq = get_work_cwq(work);
2360 if (unlikely(!cwq || gcwq != cwq->gcwq)) 2340 if (unlikely(!cwq || gcwq != cwq->gcwq))
2361 goto already_gone; 2341 goto already_gone;
2362 } else { 2342 } else if (wait_executing) {
2363 worker = find_worker_executing_work(gcwq, work); 2343 worker = find_worker_executing_work(gcwq, work);
2364 if (!worker) 2344 if (!worker)
2365 goto already_gone; 2345 goto already_gone;
2366 cwq = worker->current_cwq; 2346 cwq = worker->current_cwq;
2367 } 2347 } else
2348 goto already_gone;
2368 2349
2369 insert_wq_barrier(cwq, &barr, work, worker); 2350 insert_wq_barrier(cwq, barr, work, worker);
2370 spin_unlock_irq(&gcwq->lock); 2351 spin_unlock_irq(&gcwq->lock);
2371 2352
2372 lock_map_acquire(&cwq->wq->lockdep_map); 2353 lock_map_acquire(&cwq->wq->lockdep_map);
2373 lock_map_release(&cwq->wq->lockdep_map); 2354 lock_map_release(&cwq->wq->lockdep_map);
2374 2355 return true;
2375 wait_for_completion(&barr.done);
2376 destroy_work_on_stack(&barr.work);
2377 return 1;
2378already_gone: 2356already_gone:
2379 spin_unlock_irq(&gcwq->lock); 2357 spin_unlock_irq(&gcwq->lock);
2380 return 0; 2358 return false;
2359}
2360
2361/**
2362 * flush_work - wait for a work to finish executing the last queueing instance
2363 * @work: the work to flush
2364 *
2365 * Wait until @work has finished execution. This function considers
2366 * only the last queueing instance of @work. If @work has been
2367 * enqueued across different CPUs on a non-reentrant workqueue or on
2368 * multiple workqueues, @work might still be executing on return on
2369 * some of the CPUs from earlier queueing.
2370 *
2371 * If @work was queued only on a non-reentrant, ordered or unbound
2372 * workqueue, @work is guaranteed to be idle on return if it hasn't
2373 * been requeued since flush started.
2374 *
2375 * RETURNS:
2376 * %true if flush_work() waited for the work to finish execution,
2377 * %false if it was already idle.
2378 */
2379bool flush_work(struct work_struct *work)
2380{
2381 struct wq_barrier barr;
2382
2383 if (start_flush_work(work, &barr, true)) {
2384 wait_for_completion(&barr.done);
2385 destroy_work_on_stack(&barr.work);
2386 return true;
2387 } else
2388 return false;
2381} 2389}
2382EXPORT_SYMBOL_GPL(flush_work); 2390EXPORT_SYMBOL_GPL(flush_work);
2383 2391
2392static bool wait_on_cpu_work(struct global_cwq *gcwq, struct work_struct *work)
2393{
2394 struct wq_barrier barr;
2395 struct worker *worker;
2396
2397 spin_lock_irq(&gcwq->lock);
2398
2399 worker = find_worker_executing_work(gcwq, work);
2400 if (unlikely(worker))
2401 insert_wq_barrier(worker->current_cwq, &barr, work, worker);
2402
2403 spin_unlock_irq(&gcwq->lock);
2404
2405 if (unlikely(worker)) {
2406 wait_for_completion(&barr.done);
2407 destroy_work_on_stack(&barr.work);
2408 return true;
2409 } else
2410 return false;
2411}
2412
2413static bool wait_on_work(struct work_struct *work)
2414{
2415 bool ret = false;
2416 int cpu;
2417
2418 might_sleep();
2419
2420 lock_map_acquire(&work->lockdep_map);
2421 lock_map_release(&work->lockdep_map);
2422
2423 for_each_gcwq_cpu(cpu)
2424 ret |= wait_on_cpu_work(get_gcwq(cpu), work);
2425 return ret;
2426}
2427
2428/**
2429 * flush_work_sync - wait until a work has finished execution
2430 * @work: the work to flush
2431 *
2432 * Wait until @work has finished execution. On return, it's
2433 * guaranteed that all queueing instances of @work which happened
2434 * before this function is called are finished. In other words, if
2435 * @work hasn't been requeued since this function was called, @work is
2436 * guaranteed to be idle on return.
2437 *
2438 * RETURNS:
2439 * %true if flush_work_sync() waited for the work to finish execution,
2440 * %false if it was already idle.
2441 */
2442bool flush_work_sync(struct work_struct *work)
2443{
2444 struct wq_barrier barr;
2445 bool pending, waited;
2446
2447 /* we'll wait for executions separately, queue barr only if pending */
2448 pending = start_flush_work(work, &barr, false);
2449
2450 /* wait for executions to finish */
2451 waited = wait_on_work(work);
2452
2453 /* wait for the pending one */
2454 if (pending) {
2455 wait_for_completion(&barr.done);
2456 destroy_work_on_stack(&barr.work);
2457 }
2458
2459 return pending || waited;
2460}
2461EXPORT_SYMBOL_GPL(flush_work_sync);
2462
2384/* 2463/*
2385 * Upon a successful return (>= 0), the caller "owns" WORK_STRUCT_PENDING bit, 2464 * Upon a successful return (>= 0), the caller "owns" WORK_STRUCT_PENDING bit,
2386 * so this work can't be re-armed in any way. 2465 * so this work can't be re-armed in any way.
@@ -2423,39 +2502,7 @@ static int try_to_grab_pending(struct work_struct *work)
2423 return ret; 2502 return ret;
2424} 2503}
2425 2504
2426static void wait_on_cpu_work(struct global_cwq *gcwq, struct work_struct *work) 2505static bool __cancel_work_timer(struct work_struct *work,
2427{
2428 struct wq_barrier barr;
2429 struct worker *worker;
2430
2431 spin_lock_irq(&gcwq->lock);
2432
2433 worker = find_worker_executing_work(gcwq, work);
2434 if (unlikely(worker))
2435 insert_wq_barrier(worker->current_cwq, &barr, work, worker);
2436
2437 spin_unlock_irq(&gcwq->lock);
2438
2439 if (unlikely(worker)) {
2440 wait_for_completion(&barr.done);
2441 destroy_work_on_stack(&barr.work);
2442 }
2443}
2444
2445static void wait_on_work(struct work_struct *work)
2446{
2447 int cpu;
2448
2449 might_sleep();
2450
2451 lock_map_acquire(&work->lockdep_map);
2452 lock_map_release(&work->lockdep_map);
2453
2454 for_each_gcwq_cpu(cpu)
2455 wait_on_cpu_work(get_gcwq(cpu), work);
2456}
2457
2458static int __cancel_work_timer(struct work_struct *work,
2459 struct timer_list* timer) 2506 struct timer_list* timer)
2460{ 2507{
2461 int ret; 2508 int ret;
@@ -2472,42 +2519,81 @@ static int __cancel_work_timer(struct work_struct *work,
2472} 2519}
2473 2520
2474/** 2521/**
2475 * cancel_work_sync - block until a work_struct's callback has terminated 2522 * cancel_work_sync - cancel a work and wait for it to finish
2476 * @work: the work which is to be flushed 2523 * @work: the work to cancel
2477 *
2478 * Returns true if @work was pending.
2479 * 2524 *
2480 * cancel_work_sync() will cancel the work if it is queued. If the work's 2525 * Cancel @work and wait for its execution to finish. This function
2481 * callback appears to be running, cancel_work_sync() will block until it 2526 * can be used even if the work re-queues itself or migrates to
2482 * has completed. 2527 * another workqueue. On return from this function, @work is
2483 * 2528 * guaranteed to be not pending or executing on any CPU.
2484 * It is possible to use this function if the work re-queues itself. It can
2485 * cancel the work even if it migrates to another workqueue, however in that
2486 * case it only guarantees that work->func() has completed on the last queued
2487 * workqueue.
2488 * 2529 *
2489 * cancel_work_sync(&delayed_work->work) should be used only if ->timer is not 2530 * cancel_work_sync(&delayed_work->work) must not be used for
2490 * pending, otherwise it goes into a busy-wait loop until the timer expires. 2531 * delayed_work's. Use cancel_delayed_work_sync() instead.
2491 * 2532 *
2492 * The caller must ensure that workqueue_struct on which this work was last 2533 * The caller must ensure that the workqueue on which @work was last
2493 * queued can't be destroyed before this function returns. 2534 * queued can't be destroyed before this function returns.
2535 *
2536 * RETURNS:
2537 * %true if @work was pending, %false otherwise.
2494 */ 2538 */
2495int cancel_work_sync(struct work_struct *work) 2539bool cancel_work_sync(struct work_struct *work)
2496{ 2540{
2497 return __cancel_work_timer(work, NULL); 2541 return __cancel_work_timer(work, NULL);
2498} 2542}
2499EXPORT_SYMBOL_GPL(cancel_work_sync); 2543EXPORT_SYMBOL_GPL(cancel_work_sync);
2500 2544
2501/** 2545/**
2502 * cancel_delayed_work_sync - reliably kill off a delayed work. 2546 * flush_delayed_work - wait for a dwork to finish executing the last queueing
2503 * @dwork: the delayed work struct 2547 * @dwork: the delayed work to flush
2548 *
2549 * Delayed timer is cancelled and the pending work is queued for
2550 * immediate execution. Like flush_work(), this function only
2551 * considers the last queueing instance of @dwork.
2552 *
2553 * RETURNS:
2554 * %true if flush_work() waited for the work to finish execution,
2555 * %false if it was already idle.
2556 */
2557bool flush_delayed_work(struct delayed_work *dwork)
2558{
2559 if (del_timer_sync(&dwork->timer))
2560 __queue_work(raw_smp_processor_id(),
2561 get_work_cwq(&dwork->work)->wq, &dwork->work);
2562 return flush_work(&dwork->work);
2563}
2564EXPORT_SYMBOL(flush_delayed_work);
2565
2566/**
2567 * flush_delayed_work_sync - wait for a dwork to finish
2568 * @dwork: the delayed work to flush
2504 * 2569 *
2505 * Returns true if @dwork was pending. 2570 * Delayed timer is cancelled and the pending work is queued for
2571 * execution immediately. Other than timer handling, its behavior
2572 * is identical to flush_work_sync().
2506 * 2573 *
2507 * It is possible to use this function if @dwork rearms itself via queue_work() 2574 * RETURNS:
2508 * or queue_delayed_work(). See also the comment for cancel_work_sync(). 2575 * %true if flush_work_sync() waited for the work to finish execution,
2576 * %false if it was already idle.
2509 */ 2577 */
2510int cancel_delayed_work_sync(struct delayed_work *dwork) 2578bool flush_delayed_work_sync(struct delayed_work *dwork)
2579{
2580 if (del_timer_sync(&dwork->timer))
2581 __queue_work(raw_smp_processor_id(),
2582 get_work_cwq(&dwork->work)->wq, &dwork->work);
2583 return flush_work_sync(&dwork->work);
2584}
2585EXPORT_SYMBOL(flush_delayed_work_sync);
2586
2587/**
2588 * cancel_delayed_work_sync - cancel a delayed work and wait for it to finish
2589 * @dwork: the delayed work cancel
2590 *
2591 * This is cancel_work_sync() for delayed works.
2592 *
2593 * RETURNS:
2594 * %true if @dwork was pending, %false otherwise.
2595 */
2596bool cancel_delayed_work_sync(struct delayed_work *dwork)
2511{ 2597{
2512 return __cancel_work_timer(&dwork->work, &dwork->timer); 2598 return __cancel_work_timer(&dwork->work, &dwork->timer);
2513} 2599}
@@ -2559,23 +2645,6 @@ int schedule_delayed_work(struct delayed_work *dwork,
2559EXPORT_SYMBOL(schedule_delayed_work); 2645EXPORT_SYMBOL(schedule_delayed_work);
2560 2646
2561/** 2647/**
2562 * flush_delayed_work - block until a dwork_struct's callback has terminated
2563 * @dwork: the delayed work which is to be flushed
2564 *
2565 * Any timeout is cancelled, and any pending work is run immediately.
2566 */
2567void flush_delayed_work(struct delayed_work *dwork)
2568{
2569 if (del_timer_sync(&dwork->timer)) {
2570 __queue_work(get_cpu(), get_work_cwq(&dwork->work)->wq,
2571 &dwork->work);
2572 put_cpu();
2573 }
2574 flush_work(&dwork->work);
2575}
2576EXPORT_SYMBOL(flush_delayed_work);
2577
2578/**
2579 * schedule_delayed_work_on - queue work in global workqueue on CPU after delay 2648 * schedule_delayed_work_on - queue work in global workqueue on CPU after delay
2580 * @cpu: cpu to use 2649 * @cpu: cpu to use
2581 * @dwork: job to be done 2650 * @dwork: job to be done
@@ -2592,13 +2661,15 @@ int schedule_delayed_work_on(int cpu,
2592EXPORT_SYMBOL(schedule_delayed_work_on); 2661EXPORT_SYMBOL(schedule_delayed_work_on);
2593 2662
2594/** 2663/**
2595 * schedule_on_each_cpu - call a function on each online CPU from keventd 2664 * schedule_on_each_cpu - execute a function synchronously on each online CPU
2596 * @func: the function to call 2665 * @func: the function to call
2597 * 2666 *
2598 * Returns zero on success. 2667 * schedule_on_each_cpu() executes @func on each online CPU using the
2599 * Returns -ve errno on failure. 2668 * system workqueue and blocks until all CPUs have completed.
2600 *
2601 * schedule_on_each_cpu() is very slow. 2669 * schedule_on_each_cpu() is very slow.
2670 *
2671 * RETURNS:
2672 * 0 on success, -errno on failure.
2602 */ 2673 */
2603int schedule_on_each_cpu(work_func_t func) 2674int schedule_on_each_cpu(work_func_t func)
2604{ 2675{
@@ -2764,6 +2835,13 @@ struct workqueue_struct *__alloc_workqueue_key(const char *name,
2764 unsigned int cpu; 2835 unsigned int cpu;
2765 2836
2766 /* 2837 /*
2838 * Workqueues which may be used during memory reclaim should
2839 * have a rescuer to guarantee forward progress.
2840 */
2841 if (flags & WQ_MEM_RECLAIM)
2842 flags |= WQ_RESCUER;
2843
2844 /*
2767 * Unbound workqueues aren't concurrency managed and should be 2845 * Unbound workqueues aren't concurrency managed and should be
2768 * dispatched to workers immediately. 2846 * dispatched to workers immediately.
2769 */ 2847 */