Re: perf: fuzzer triggered warning in intel_pmu_drain_pebs_nhm()

From: Peter Zijlstra
Date: Mon Jul 06 2015 - 06:55:39 EST


On Fri, Jul 03, 2015 at 08:08:27PM +0000, Liang, Kan wrote:
> If we cleared the last bit, we not only drain the buffer but also decrease
> the event->ctx->pmu, which is used to flush the PEBS buffer during
> context switches.
> We need to disable cpuc->pebs_enabled before changing
> event->ctx->pmu as below.
>

Indeed, mind sending a proper patch so I can press 'A' on it?

> diff --git a/arch/x86/kernel/cpu/perf_event_intel_ds.c
> b/arch/x86/kernel/cpu/perf_event_intel_ds.c
> index 71fc402..76285c1 100644
> --- a/arch/x86/kernel/cpu/perf_event_intel_ds.c
> +++ b/arch/x86/kernel/cpu/perf_event_intel_ds.c
> @@ -754,6 +754,11 @@ void intel_pmu_pebs_disable(struct
> perf_event *event)
> struct cpu_hw_events *cpuc = this_cpu_ptr(&cpu_hw_events);
> struct hw_perf_event *hwc = &event->hw;
> struct debug_store *ds = cpuc->ds;
> + bool large_pebs = ds->pebs_interrupt_threshold >
> + ds->pebs_buffer_base + x86_pmu.pebs_record_size;
> +
> + if (large_pebs)
> + intel_pmu_drain_pebs_buffer();
>
> cpuc->pebs_enabled &= ~(1ULL << hwc->idx);
>
> @@ -762,12 +767,8 @@ void intel_pmu_pebs_disable(struct
> perf_event *event)
> else if (event->hw.flags & PERF_X86_EVENT_PEBS_ST)
> cpuc->pebs_enabled &= ~(1ULL << 63);
>
> - if (ds->pebs_interrupt_threshold >
> - ds->pebs_buffer_base + x86_pmu.pebs_record_size) {
> - intel_pmu_drain_pebs_buffer();
> - if (!pebs_is_enabled(cpuc))
> - perf_sched_cb_dec(event->ctx->pmu);
> - }
> + if (large_pebs && !pebs_is_enabled(cpuc))
> + perf_sched_cb_dec(event->ctx->pmu);
>
> if (cpuc->enabled)
> wrmsrl(MSR_IA32_PEBS_ENABLE, cpuc->pebs_enabled);
>
>
>
> Thanks,
> Kan
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/