Re: [PATCH V6] perf: Reset the dirty counter to prevent the leak for an RDPMC task

[Liang, Kan]

On 5/12/2021 3:35 AM, Peter Zijlstra wrote:
> On Tue, May 11, 2021 at 05:42:54PM -0400, Liang, Kan wrote:
> > diff --git a/kernel/events/core.c b/kernel/events/core.c
> > index 1574b70..8216acc 100644
> > --- a/kernel/events/core.c
> > +++ b/kernel/events/core.c
> > @@ -3851,7 +3851,7 @@ static void perf_event_context_sched_in(struct
> > perf_event_context *ctx,
> >   		cpu_ctx_sched_out(cpuctx, EVENT_FLEXIBLE);
> >   	perf_event_sched_in(cpuctx, ctx, task);
> >
> > -	if (cpuctx->sched_cb_usage && pmu->sched_task)
> > +	if (pmu->sched_task && (cpuctx->sched_cb_usage ||
> > atomic_read(&pmu->sched_cb_usages)))
> >   		pmu->sched_task(cpuctx->task_ctx, true);
>
> Aside from the obvious whitespace issues; I think this should work.

Thanks. The whitespace should be caused by the copy/paste. I will fix it 
in the V7.

I did more tests. For some cases, I can still observe the dirty counter 
for the first RDPMC read. I think we still have to clear the dirty 
counters in the x86_pmu_event_mapped() for the first RDPMC read.
I have to disable the the interrupts to prevent the preemption.


 static void x86_pmu_event_mapped(struct perf_event *event, struct 
mm_struct *mm)
 {
+	unsigned long flags;
+
 	if (!(event->hw.flags & PERF_X86_EVENT_RDPMC_ALLOWED))
 		return;

 	/*
+	 * Enable sched_task() for the RDPMC task,
+	 * and clear the existing dirty counters.
+	 */
+	if (x86_pmu.sched_task && event->hw.target) {
+		atomic_inc(&event->pmu->sched_cb_usages);
+		local_irq_save(flags);
+		x86_pmu_clear_dirty_counters();
+		local_irq_restore(flags);
+	}
+
+	/*
 	 * This function relies on not being called concurrently in two
 	 * tasks in the same mm.  Otherwise one task could observe
 	 * perf_rdpmc_allowed > 1 and return all the way back to

Thanks,
Kan