linux-kernel - Re: [PATCH v4] events/core: fix acoount failure for event's child_total_enable

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Message-ID: <20250325084320.GC604566@e132581.arm.com>
Date: Tue, 25 Mar 2025 08:43:20 +0000
From: Leo Yan <leo.yan@....com>
To: Yeoreum Yun <yeoreum.yun@....com>
Cc: mingo@...nel.org, peterz@...radead.org, acme@...nel.org,
	namhyung@...nel.org, Mark.Rutland@....com,
	alexander.shishkin@...ux.intel.com, jolsa@...nel.org,
	irogers@...gle.com, adrian.hunter@...el.com,
	kan.liang@...ux.intel.com, james.clark@...aro.org,
	linux-perf-users@...r.kernel.org, linux-kernel@...r.kernel.org
Subject: Re: [PATCH v4] events/core: fix acoount failure for event's
 child_total_enable_time at task exit

On Tue, Mar 25, 2025 at 07:39:03AM +0000, Yeoreum Yun wrote:

[...]

> > > After this patch, this problem is gone like:
> > >
> > > sudo ./perf stat -vvv -e armv8_pmuv3_0/event=0x08/ -e armv8_pmuv3_1/event=0x08/ -- stress-ng --pthread=2 -t 10s
> > > ...
> > > armv8_pmuv3_0/event=0x08/: 15396770398 32157963940 21898169000
> > > armv8_pmuv3_1/event=0x08/: 22428964974 32157963940 10259794940
> > >
> > >  Performance counter stats for 'stress-ng --pthread=2 -t 10s':
> > >
> > >     15,396,770,398      armv8_pmuv3_0/event=0x08/                                               (68.10%)
> > >     22,428,964,974      armv8_pmuv3_1/event=0x08/                                               (31.90%)
> > >
> > > Signed-off-by: Yeoreum Yun <yeoreum.yun@....com>
> > > Suggsted-by: Peter Zijlstra <peterz@...radead.org>
> >
> > /Suggsted-by/Suggested-by/
> 
> Thanks ;) I'll respin.

Given this patch is a fix, it is good to add a fix tag.

> > > ---
> > >  kernel/events/core.c | 18 +++++++++---------
> > >  1 file changed, 9 insertions(+), 9 deletions(-)
> > >
> > > diff --git a/kernel/events/core.c b/kernel/events/core.c
> > > index 6364319e2f88..058533a50493 100644
> > > --- a/kernel/events/core.c
> > > +++ b/kernel/events/core.c
> > > @@ -2407,6 +2407,7 @@ ctx_time_update_event(struct perf_event_context *ctx, struct perf_event *event)
> > >  #define DETACH_GROUP	0x01UL
> > >  #define DETACH_CHILD	0x02UL
> > >  #define DETACH_DEAD	0x04UL
> > > +#define DETACH_EXIT	0x08UL
> > >
> > >  /*
> > >   * Cross CPU call to remove a performance event
> > > @@ -2421,6 +2422,7 @@ __perf_remove_from_context(struct perf_event *event,
> > >  			   void *info)
> > >  {
> > >  	struct perf_event_pmu_context *pmu_ctx = event->pmu_ctx;
> > > +	enum perf_event_state state = PERF_EVENT_STATE_OFF;
> > >  	unsigned long flags = (unsigned long)info;
> > >
> > >  	ctx_time_update(cpuctx, ctx);
> > > @@ -2429,16 +2431,19 @@ __perf_remove_from_context(struct perf_event *event,
> > >  	 * Ensure event_sched_out() switches to OFF, at the very least
> > >  	 * this avoids raising perf_pending_task() at this time.
> > >  	 */
> > > -	if (flags & DETACH_DEAD)
> > > +	if (flags & DETACH_EXIT)
> > > +		state = PERF_EVENT_STATE_EXIT;
> > > +	if (flags & DETACH_DEAD) {
> > >  		event->pending_disable = 1;
> > > +		state = PERF_EVENT_STATE_DEAD;
> > > +	}
> > >  	event_sched_out(event, ctx);
> > > +	perf_event_set_state(event, min(event->state, state));
> >
> > Nitpick: can we move perf_event_set_state() before event_sched_out()?
> >
> > So the function handles the state machine ahead, then proceed for
> > other operations.
> 
> No It couldn't. IIUC, event_sched_out() disable pmu with ACTIVE state
> event only.
> If state is changed first from active state, it wouldn't be sched out by
> event_sched_out.

Indeed !  Please ignore my comment.

> > >  	if (flags & DETACH_GROUP)
> > >  		perf_group_detach(event);
> > >  	if (flags & DETACH_CHILD)
> > >  		perf_child_detach(event);
> > >  	list_del_event(event, ctx);
> > > -	if (flags & DETACH_DEAD)
> > > -		event->state = PERF_EVENT_STATE_DEAD;
> > >
> > >  	if (!pmu_ctx->nr_events) {
> > >  		pmu_ctx->rotate_necessary = 0;
> > > @@ -13424,12 +13429,7 @@ perf_event_exit_event(struct perf_event *event, struct perf_event_context *ctx)
> > >  		mutex_lock(&parent_event->child_mutex);
> > >  	}
> > >
> > > -	perf_remove_from_context(event, detach_flags);
> > > -
> > > -	raw_spin_lock_irq(&ctx->lock);
> > > -	if (event->state > PERF_EVENT_STATE_EXIT)
> > > -		perf_event_set_state(event, PERF_EVENT_STATE_EXIT);
> > > -	raw_spin_unlock_irq(&ctx->lock);
> > > +	perf_remove_from_context(event, detach_flags | DETACH_EXIT);

It is good to add a description in commit log for why remove the
code chunk for updating state in the function perf_event_exit_event().

As we discussed, it uses a central place __perf_remove_from_context()
to maintain the state when event exits,  this can avoid race
condition.  To support this, the 'DETACH_EXIT' flag is passed to
__perf_remove_from_context() instead.

Thanks,
Leo

> > >
> > >  	/*
> > >  	 * Child events can be freed.
> > > --
> > > LEVI:{C3F47F37-75D8-414A-A8BA-3980EC8A46D7}
> > >