aboutsummaryrefslogtreecommitdiffstats
path: root/kernel/perf_counter.c
diff options
context:
space:
mode:
authorPeter Zijlstra <a.p.zijlstra@chello.nl>2009-08-07 13:49:01 -0400
committerIngo Molnar <mingo@elte.hu>2009-08-09 06:54:46 -0400
commit3a80b4a3539696f4b0574876326860323035a302 (patch)
tree9d2901259a2b25b181997f89495452fd7f06827b /kernel/perf_counter.c
parent3a43ce68ae1758fa6a839386025ef45acb6baa22 (diff)
perf_counter: Fix a race on perf_counter_ctx
While extending perfcounters with BTS hw-tracing, Markus Metzger managed to trigger this warning: [ 995.557128] WARNING: at kernel/perf_counter.c:1191 __perf_counter_task_sched_out+0x48/0x6b() triggers because commit 9f498cc5be7e013d8d6e4c616980ed0ffc8680d2 (perf_counter: Full task tracing) removed clearing of tsk->perf_counter_ctxp out from under ctx->lock which introduced a race (against perf_lock_task_context). Move it back and deal with the exit notification by explicitly passing along the former task context. Reported-by: Markus T Metzger <markus.t.metzger@intel.com> Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Cc: Paul Mackerras <paulus@samba.org> LKML-Reference: <1249667341.17467.5.camel@twins> Signed-off-by: Ingo Molnar <mingo@elte.hu>
Diffstat (limited to 'kernel/perf_counter.c')
-rw-r--r--kernel/perf_counter.c30
1 files changed, 15 insertions, 15 deletions
diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c
index 002310540417..546e62d62941 100644
--- a/kernel/perf_counter.c
+++ b/kernel/perf_counter.c
@@ -2850,7 +2850,8 @@ perf_counter_read_event(struct perf_counter *counter,
2850 */ 2850 */
2851 2851
2852struct perf_task_event { 2852struct perf_task_event {
2853 struct task_struct *task; 2853 struct task_struct *task;
2854 struct perf_counter_context *task_ctx;
2854 2855
2855 struct { 2856 struct {
2856 struct perf_event_header header; 2857 struct perf_event_header header;
@@ -2910,24 +2911,23 @@ static void perf_counter_task_ctx(struct perf_counter_context *ctx,
2910static void perf_counter_task_event(struct perf_task_event *task_event) 2911static void perf_counter_task_event(struct perf_task_event *task_event)
2911{ 2912{
2912 struct perf_cpu_context *cpuctx; 2913 struct perf_cpu_context *cpuctx;
2913 struct perf_counter_context *ctx; 2914 struct perf_counter_context *ctx = task_event->task_ctx;
2914 2915
2915 cpuctx = &get_cpu_var(perf_cpu_context); 2916 cpuctx = &get_cpu_var(perf_cpu_context);
2916 perf_counter_task_ctx(&cpuctx->ctx, task_event); 2917 perf_counter_task_ctx(&cpuctx->ctx, task_event);
2917 put_cpu_var(perf_cpu_context); 2918 put_cpu_var(perf_cpu_context);
2918 2919
2919 rcu_read_lock(); 2920 rcu_read_lock();
2920 /* 2921 if (!ctx)
2921 * doesn't really matter which of the child contexts the 2922 ctx = rcu_dereference(task_event->task->perf_counter_ctxp);
2922 * events ends up in.
2923 */
2924 ctx = rcu_dereference(current->perf_counter_ctxp);
2925 if (ctx) 2923 if (ctx)
2926 perf_counter_task_ctx(ctx, task_event); 2924 perf_counter_task_ctx(ctx, task_event);
2927 rcu_read_unlock(); 2925 rcu_read_unlock();
2928} 2926}
2929 2927
2930static void perf_counter_task(struct task_struct *task, int new) 2928static void perf_counter_task(struct task_struct *task,
2929 struct perf_counter_context *task_ctx,
2930 int new)
2931{ 2931{
2932 struct perf_task_event task_event; 2932 struct perf_task_event task_event;
2933 2933
@@ -2937,8 +2937,9 @@ static void perf_counter_task(struct task_struct *task, int new)
2937 return; 2937 return;
2938 2938
2939 task_event = (struct perf_task_event){ 2939 task_event = (struct perf_task_event){
2940 .task = task, 2940 .task = task,
2941 .event = { 2941 .task_ctx = task_ctx,
2942 .event = {
2942 .header = { 2943 .header = {
2943 .type = new ? PERF_EVENT_FORK : PERF_EVENT_EXIT, 2944 .type = new ? PERF_EVENT_FORK : PERF_EVENT_EXIT,
2944 .misc = 0, 2945 .misc = 0,
@@ -2956,7 +2957,7 @@ static void perf_counter_task(struct task_struct *task, int new)
2956 2957
2957void perf_counter_fork(struct task_struct *task) 2958void perf_counter_fork(struct task_struct *task)
2958{ 2959{
2959 perf_counter_task(task, 1); 2960 perf_counter_task(task, NULL, 1);
2960} 2961}
2961 2962
2962/* 2963/*
@@ -4310,7 +4311,7 @@ void perf_counter_exit_task(struct task_struct *child)
4310 unsigned long flags; 4311 unsigned long flags;
4311 4312
4312 if (likely(!child->perf_counter_ctxp)) { 4313 if (likely(!child->perf_counter_ctxp)) {
4313 perf_counter_task(child, 0); 4314 perf_counter_task(child, NULL, 0);
4314 return; 4315 return;
4315 } 4316 }
4316 4317
@@ -4330,6 +4331,7 @@ void perf_counter_exit_task(struct task_struct *child)
4330 * incremented the context's refcount before we do put_ctx below. 4331 * incremented the context's refcount before we do put_ctx below.
4331 */ 4332 */
4332 spin_lock(&child_ctx->lock); 4333 spin_lock(&child_ctx->lock);
4334 child->perf_counter_ctxp = NULL;
4333 /* 4335 /*
4334 * If this context is a clone; unclone it so it can't get 4336 * If this context is a clone; unclone it so it can't get
4335 * swapped to another process while we're removing all 4337 * swapped to another process while we're removing all
@@ -4343,9 +4345,7 @@ void perf_counter_exit_task(struct task_struct *child)
4343 * won't get any samples after PERF_EVENT_EXIT. We can however still 4345 * won't get any samples after PERF_EVENT_EXIT. We can however still
4344 * get a few PERF_EVENT_READ events. 4346 * get a few PERF_EVENT_READ events.
4345 */ 4347 */
4346 perf_counter_task(child, 0); 4348 perf_counter_task(child, child_ctx, 0);
4347
4348 child->perf_counter_ctxp = NULL;
4349 4349
4350 /* 4350 /*
4351 * We can recurse on the same lock type through: 4351 * We can recurse on the same lock type through: