summaryrefslogtreecommitdiff
path: root/kernel
diff options
context:
space:
mode:
authorStephane Eranian <eranian@google.com>2011-08-25 15:58:03 +0200
committerIngo Molnar <mingo@elte.hu>2011-08-29 12:28:33 +0200
commita8d757ef076f0f95f13a918808824058de25b3eb (patch)
tree3c1151ef886d9b72d0a7b7b267d9f37c72d5f475 /kernel
parentc6a389f123b9f68d605bb7e0f9b32ec1e3e14132 (diff)
perf events: Fix slow and broken cgroup context switch code
The current cgroup context switch code was incorrect leading to bogus counts. Furthermore, as soon as there was an active cgroup event on a CPU, the context switch cost on that CPU would increase by a significant amount as demonstrated by a simple ping/pong example: $ ./pong Both processes pinned to CPU1, running for 10s 10684.51 ctxsw/s Now start a cgroup perf stat: $ perf stat -e cycles,cycles -A -a -G test -C 1 -- sleep 100 $ ./pong Both processes pinned to CPU1, running for 10s 6674.61 ctxsw/s That's a 37% penalty. Note that pong is not even in the monitored cgroup. The results shown by perf stat are bogus: $ perf stat -e cycles,cycles -A -a -G test -C 1 -- sleep 100 Performance counter stats for 'sleep 100': CPU1 <not counted> cycles test CPU1 16,984,189,138 cycles # 0.000 GHz The second 'cycles' event should report a count @ CPU clock (here 2.4GHz) as it is counting across all cgroups. The patch below fixes the bogus accounting and bypasses any cgroup switches in case the outgoing and incoming tasks are in the same cgroup. With this patch the same test now yields: $ ./pong Both processes pinned to CPU1, running for 10s 10775.30 ctxsw/s Start perf stat with cgroup: $ perf stat -e cycles,cycles -A -a -G test -C 1 -- sleep 10 Run pong outside the cgroup: $ /pong Both processes pinned to CPU1, running for 10s 10687.80 ctxsw/s The penalty is now less than 2%. And the results for perf stat are correct: $ perf stat -e cycles,cycles -A -a -G test -C 1 -- sleep 10 Performance counter stats for 'sleep 10': CPU1 <not counted> cycles test # 0.000 GHz CPU1 23,933,981,448 cycles # 0.000 GHz Now perf stat reports the correct counts for for the non cgroup event. If we run pong inside the cgroup, then we also get the correct counts: $ perf stat -e cycles,cycles -A -a -G test -C 1 -- sleep 10 Performance counter stats for 'sleep 10': CPU1 22,297,726,205 cycles test # 0.000 GHz CPU1 23,933,981,448 cycles # 0.000 GHz 10.001457237 seconds time elapsed Signed-off-by: Stephane Eranian <eranian@google.com> Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Link: http://lkml.kernel.org/r/20110825135803.GA4697@quad Signed-off-by: Ingo Molnar <mingo@elte.hu>
Diffstat (limited to 'kernel')
-rw-r--r--kernel/events/core.c63
-rw-r--r--kernel/sched.c2
2 files changed, 54 insertions, 11 deletions
diff --git a/kernel/events/core.c b/kernel/events/core.c
index b8785e26ee1c..45847fbb599a 100644
--- a/kernel/events/core.c
+++ b/kernel/events/core.c
@@ -399,14 +399,54 @@ void perf_cgroup_switch(struct task_struct *task, int mode)
local_irq_restore(flags);
}
-static inline void perf_cgroup_sched_out(struct task_struct *task)
+static inline void perf_cgroup_sched_out(struct task_struct *task,
+ struct task_struct *next)
{
- perf_cgroup_switch(task, PERF_CGROUP_SWOUT);
+ struct perf_cgroup *cgrp1;
+ struct perf_cgroup *cgrp2 = NULL;
+
+ /*
+ * we come here when we know perf_cgroup_events > 0
+ */
+ cgrp1 = perf_cgroup_from_task(task);
+
+ /*
+ * next is NULL when called from perf_event_enable_on_exec()
+ * that will systematically cause a cgroup_switch()
+ */
+ if (next)
+ cgrp2 = perf_cgroup_from_task(next);
+
+ /*
+ * only schedule out current cgroup events if we know
+ * that we are switching to a different cgroup. Otherwise,
+ * do no touch the cgroup events.
+ */
+ if (cgrp1 != cgrp2)
+ perf_cgroup_switch(task, PERF_CGROUP_SWOUT);
}
-static inline void perf_cgroup_sched_in(struct task_struct *task)
+static inline void perf_cgroup_sched_in(struct task_struct *prev,
+ struct task_struct *task)
{
- perf_cgroup_switch(task, PERF_CGROUP_SWIN);
+ struct perf_cgroup *cgrp1;
+ struct perf_cgroup *cgrp2 = NULL;
+
+ /*
+ * we come here when we know perf_cgroup_events > 0
+ */
+ cgrp1 = perf_cgroup_from_task(task);
+
+ /* prev can never be NULL */
+ cgrp2 = perf_cgroup_from_task(prev);
+
+ /*
+ * only need to schedule in cgroup events if we are changing
+ * cgroup during ctxsw. Cgroup events were not scheduled
+ * out of ctxsw out if that was not the case.
+ */
+ if (cgrp1 != cgrp2)
+ perf_cgroup_switch(task, PERF_CGROUP_SWIN);
}
static inline int perf_cgroup_connect(int fd, struct perf_event *event,
@@ -518,11 +558,13 @@ static inline void update_cgrp_time_from_cpuctx(struct perf_cpu_context *cpuctx)
{
}
-static inline void perf_cgroup_sched_out(struct task_struct *task)
+static inline void perf_cgroup_sched_out(struct task_struct *task,
+ struct task_struct *next)
{
}
-static inline void perf_cgroup_sched_in(struct task_struct *task)
+static inline void perf_cgroup_sched_in(struct task_struct *prev,
+ struct task_struct *task)
{
}
@@ -1988,7 +2030,7 @@ void __perf_event_task_sched_out(struct task_struct *task,
* cgroup event are system-wide mode only
*/
if (atomic_read(&__get_cpu_var(perf_cgroup_events)))
- perf_cgroup_sched_out(task);
+ perf_cgroup_sched_out(task, next);
}
static void task_ctx_sched_out(struct perf_event_context *ctx)
@@ -2153,7 +2195,8 @@ static void perf_event_context_sched_in(struct perf_event_context *ctx,
* accessing the event control register. If a NMI hits, then it will
* keep the event running.
*/
-void __perf_event_task_sched_in(struct task_struct *task)
+void __perf_event_task_sched_in(struct task_struct *prev,
+ struct task_struct *task)
{
struct perf_event_context *ctx;
int ctxn;
@@ -2171,7 +2214,7 @@ void __perf_event_task_sched_in(struct task_struct *task)
* cgroup event are system-wide mode only
*/
if (atomic_read(&__get_cpu_var(perf_cgroup_events)))
- perf_cgroup_sched_in(task);
+ perf_cgroup_sched_in(prev, task);
}
static u64 perf_calculate_period(struct perf_event *event, u64 nsec, u64 count)
@@ -2427,7 +2470,7 @@ static void perf_event_enable_on_exec(struct perf_event_context *ctx)
* ctxswin cgroup events which are already scheduled
* in.
*/
- perf_cgroup_sched_out(current);
+ perf_cgroup_sched_out(current, NULL);
raw_spin_lock(&ctx->lock);
task_ctx_sched_out(ctx);
diff --git a/kernel/sched.c b/kernel/sched.c
index ccacdbdecf45..0408cdc6d572 100644
--- a/kernel/sched.c
+++ b/kernel/sched.c
@@ -3065,7 +3065,7 @@ static void finish_task_switch(struct rq *rq, struct task_struct *prev)
#ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
local_irq_disable();
#endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */
- perf_event_task_sched_in(current);
+ perf_event_task_sched_in(prev, current);
#ifdef __ARCH_WANT_INTERRUPTS_ON_CTXSW
local_irq_enable();
#endif /* __ARCH_WANT_INTERRUPTS_ON_CTXSW */