aboutsummaryrefslogtreecommitdiffstats
path: root/kernel/perf_event.c
diff options
context:
space:
mode:
authorStephane Eranian <eranian@google.com>2011-02-18 07:40:01 -0500
committerIngo Molnar <mingo@elte.hu>2011-02-23 05:35:46 -0500
commit3f7cce3c18188a067d463749168bdda5abc5b0f7 (patch)
tree9c161a970a19176f26dc29811dc9ae4d64bfe8a8 /kernel/perf_event.c
parentc97cf42219b7b6037d2f96c27a5f114f2383f828 (diff)
perf_events: Fix rcu and locking issues with cgroup support
This patches ensures that we do not end up calling perf_cgroup_from_task() when there is no cgroup event. This avoids potential RCU and locking issues. The change in perf_cgroup_set_timestamp() ensures we check against ctx->nr_cgroups. It also avoids calling perf_clock() tiwce in a row. It also ensures we do need to grab ctx->lock before calling the function. We drop update_cgrp_time() from task_clock_event_read() because it is not needed. This also avoids having to deal with perf_cgroup_from_task(). Thanks to Peter Zijlstra for his help on this. Signed-off-by: Stephane Eranian <eranian@gmail.com> Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> LKML-Reference: <4d5e76b8.815bdf0a.7ac3.774f@mx.google.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
Diffstat (limited to 'kernel/perf_event.c')
-rw-r--r--kernel/perf_event.c40
1 files changed, 29 insertions, 11 deletions
diff --git a/kernel/perf_event.c b/kernel/perf_event.c
index a0a6987fabc4..dadeaea4b3fc 100644
--- a/kernel/perf_event.c
+++ b/kernel/perf_event.c
@@ -201,6 +201,11 @@ __get_cpu_context(struct perf_event_context *ctx)
201 201
202#ifdef CONFIG_CGROUP_PERF 202#ifdef CONFIG_CGROUP_PERF
203 203
204/*
205 * Must ensure cgroup is pinned (css_get) before calling
206 * this function. In other words, we cannot call this function
207 * if there is no cgroup event for the current CPU context.
208 */
204static inline struct perf_cgroup * 209static inline struct perf_cgroup *
205perf_cgroup_from_task(struct task_struct *task) 210perf_cgroup_from_task(struct task_struct *task)
206{ 211{
@@ -268,28 +273,41 @@ static inline void update_cgrp_time_from_cpuctx(struct perf_cpu_context *cpuctx)
268 273
269static inline void update_cgrp_time_from_event(struct perf_event *event) 274static inline void update_cgrp_time_from_event(struct perf_event *event)
270{ 275{
271 struct perf_cgroup *cgrp = perf_cgroup_from_task(current); 276 struct perf_cgroup *cgrp;
277
272 /* 278 /*
273 * do not update time when cgroup is not active 279 * ensure we access cgroup data only when needed and
280 * when we know the cgroup is pinned (css_get)
274 */ 281 */
275 if (!event->cgrp || cgrp != event->cgrp) 282 if (!is_cgroup_event(event))
276 return; 283 return;
277 284
278 __update_cgrp_time(event->cgrp); 285 cgrp = perf_cgroup_from_task(current);
286 /*
287 * Do not update time when cgroup is not active
288 */
289 if (cgrp == event->cgrp)
290 __update_cgrp_time(event->cgrp);
279} 291}
280 292
281static inline void 293static inline void
282perf_cgroup_set_timestamp(struct task_struct *task, u64 now) 294perf_cgroup_set_timestamp(struct task_struct *task,
295 struct perf_event_context *ctx)
283{ 296{
284 struct perf_cgroup *cgrp; 297 struct perf_cgroup *cgrp;
285 struct perf_cgroup_info *info; 298 struct perf_cgroup_info *info;
286 299
287 if (!task) 300 /*
301 * ctx->lock held by caller
302 * ensure we do not access cgroup data
303 * unless we have the cgroup pinned (css_get)
304 */
305 if (!task || !ctx->nr_cgroups)
288 return; 306 return;
289 307
290 cgrp = perf_cgroup_from_task(task); 308 cgrp = perf_cgroup_from_task(task);
291 info = this_cpu_ptr(cgrp->info); 309 info = this_cpu_ptr(cgrp->info);
292 info->timestamp = now; 310 info->timestamp = ctx->timestamp;
293} 311}
294 312
295#define PERF_CGROUP_SWOUT 0x1 /* cgroup switch out every event */ 313#define PERF_CGROUP_SWOUT 0x1 /* cgroup switch out every event */
@@ -494,7 +512,8 @@ static inline int perf_cgroup_connect(pid_t pid, struct perf_event *event,
494} 512}
495 513
496static inline void 514static inline void
497perf_cgroup_set_timestamp(struct task_struct *task, u64 now) 515perf_cgroup_set_timestamp(struct task_struct *task,
516 struct perf_event_context *ctx)
498{ 517{
499} 518}
500 519
@@ -1613,7 +1632,7 @@ static int __perf_event_enable(void *info)
1613 /* 1632 /*
1614 * set current task's cgroup time reference point 1633 * set current task's cgroup time reference point
1615 */ 1634 */
1616 perf_cgroup_set_timestamp(current, perf_clock()); 1635 perf_cgroup_set_timestamp(current, ctx);
1617 1636
1618 __perf_event_mark_enabled(event, ctx); 1637 __perf_event_mark_enabled(event, ctx);
1619 1638
@@ -2048,7 +2067,7 @@ ctx_sched_in(struct perf_event_context *ctx,
2048 2067
2049 now = perf_clock(); 2068 now = perf_clock();
2050 ctx->timestamp = now; 2069 ctx->timestamp = now;
2051 perf_cgroup_set_timestamp(task, now); 2070 perf_cgroup_set_timestamp(task, ctx);
2052 /* 2071 /*
2053 * First go through the list and put on any pinned groups 2072 * First go through the list and put on any pinned groups
2054 * in order to give them the best chance of going on. 2073 * in order to give them the best chance of going on.
@@ -5795,7 +5814,6 @@ static void task_clock_event_read(struct perf_event *event)
5795 5814
5796 if (!in_nmi()) { 5815 if (!in_nmi()) {
5797 update_context_time(event->ctx); 5816 update_context_time(event->ctx);
5798 update_cgrp_time_from_event(event);
5799 time = event->ctx->time; 5817 time = event->ctx->time;
5800 } else { 5818 } else {
5801 u64 now = perf_clock(); 5819 u64 now = perf_clock();