lk/linux-kernel-test
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

perf: Fix duplicate events with multiple-pmu vs software events

Browse Source 
Because the multi-pmu bits can share contexts between struct pmu
instances we could get duplicate events by iterating the pmu list.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
This commit is contained in:
Peter Zijlstra
2010-12-07 14:18:20 +01:00
committed by Ingo Molnar
parent 6313e3c217
commit 5167695753
2 changed files with 30 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files

1
include/linux/perf_event.h
View File

@@ -887,6 +887,7 @@ struct perf_cpu_context {
int exclusive; int exclusive;
struct list_head rotation_list; struct list_head rotation_list;
int jiffies_interval; int jiffies_interval;
struct pmu *active_pmu;
}; };
struct perf_output_handle { struct perf_output_handle {

35
kernel/perf_event.c
View File

@@ -3824,6 +3824,8 @@ static void perf_event_task_event(struct perf_task_event *task_event)
rcu_read_lock(); rcu_read_lock();
list_for_each_entry_rcu(pmu, &pmus, entry) { list_for_each_entry_rcu(pmu, &pmus, entry) {
cpuctx = get_cpu_ptr(pmu->pmu_cpu_context); cpuctx = get_cpu_ptr(pmu->pmu_cpu_context);
if (cpuctx->active_pmu != pmu)
goto next;
perf_event_task_ctx(&cpuctx->ctx, task_event); perf_event_task_ctx(&cpuctx->ctx, task_event);
ctx = task_event->task_ctx; ctx = task_event->task_ctx;
@@ -3959,6 +3961,8 @@ static void perf_event_comm_event(struct perf_comm_event *comm_event)
rcu_read_lock(); rcu_read_lock();
list_for_each_entry_rcu(pmu, &pmus, entry) { list_for_each_entry_rcu(pmu, &pmus, entry) {
cpuctx = get_cpu_ptr(pmu->pmu_cpu_context); cpuctx = get_cpu_ptr(pmu->pmu_cpu_context);
if (cpuctx->active_pmu != pmu)
goto next;
perf_event_comm_ctx(&cpuctx->ctx, comm_event); perf_event_comm_ctx(&cpuctx->ctx, comm_event);
ctxn = pmu->task_ctx_nr; ctxn = pmu->task_ctx_nr;
@@ -4144,6 +4148,8 @@ got_name:
rcu_read_lock(); rcu_read_lock();
list_for_each_entry_rcu(pmu, &pmus, entry) { list_for_each_entry_rcu(pmu, &pmus, entry) {
cpuctx = get_cpu_ptr(pmu->pmu_cpu_context); cpuctx = get_cpu_ptr(pmu->pmu_cpu_context);
if (cpuctx->active_pmu != pmu)
goto next;
perf_event_mmap_ctx(&cpuctx->ctx, mmap_event, perf_event_mmap_ctx(&cpuctx->ctx, mmap_event,
vma->vm_flags & VM_EXEC); vma->vm_flags & VM_EXEC);
@@ -5145,20 +5151,36 @@ static void *find_pmu_context(int ctxn)
return NULL; return NULL;
} }
static void free_pmu_context(void * __percpu cpu_context) static void update_pmu_context(struct pmu *pmu, struct pmu *old_pmu)
{ {
struct pmu *pmu; int cpu;
for_each_possible_cpu(cpu) {
struct perf_cpu_context *cpuctx;
cpuctx = per_cpu_ptr(pmu->pmu_cpu_context, cpu);
if (cpuctx->active_pmu == old_pmu)
cpuctx->active_pmu = pmu;
}
}
static void free_pmu_context(struct pmu *pmu)
{
struct pmu *i;
mutex_lock(&pmus_lock); mutex_lock(&pmus_lock);
/* /*
* Like a real lame refcount. * Like a real lame refcount.
*/ */
list_for_each_entry(pmu, &pmus, entry) { list_for_each_entry(i, &pmus, entry) {
if (pmu->pmu_cpu_context == cpu_context) if (i->pmu_cpu_context == pmu->pmu_cpu_context) {
update_pmu_context(i, pmu);
goto out; goto out;
} }
}
free_percpu(cpu_context); free_percpu(pmu->pmu_cpu_context);
out: out:
mutex_unlock(&pmus_lock); mutex_unlock(&pmus_lock);
} }
@@ -5190,6 +5212,7 @@ int perf_pmu_register(struct pmu *pmu)
cpuctx->ctx.pmu = pmu; cpuctx->ctx.pmu = pmu;
cpuctx->jiffies_interval = 1; cpuctx->jiffies_interval = 1;
INIT_LIST_HEAD(&cpuctx->rotation_list); INIT_LIST_HEAD(&cpuctx->rotation_list);
cpuctx->active_pmu = pmu;
} }
got_cpu_context: got_cpu_context:
@@ -5241,7 +5264,7 @@ void perf_pmu_unregister(struct pmu *pmu)
synchronize_rcu(); synchronize_rcu();
free_percpu(pmu->pmu_disable_count); free_percpu(pmu->pmu_disable_count);
free_pmu_context(pmu->pmu_cpu_context); free_pmu_context(pmu);
} }
struct pmu *perf_init_event(struct perf_event *event) struct pmu *perf_init_event(struct perf_event *event)