Re: [Patch v7 02/12] perf/x86/intel: Fix NULL event access and potential PEBS record loss

From: Mi, Dapeng

Date: Mon Oct 06 2025 - 04:03:57 EST



On 10/2/2025 3:09 PM, Oliver Sang wrote:
> hi, Dapeng,
>
> On Tue, Sep 30, 2025 at 02:19:25PM +0800, Mi, Dapeng wrote:
> [...]
>
>> Oops, it looks previous fix was incomplete. :(
>>
>> Oliver, could you please verify the new attached patch (Please apply this
>> patch on top of the whole patch series)? Thanks a lot for your effort. 
> we confirmed the patch fixed the warning we reported.
>
> Tested-by: kernel test robot <oliver.sang@xxxxxxxxx>
>
> =========================================================================================
> compiler/cpufreq_governor/kconfig/option_a/rootfs/tbox_group/test/testcase:
> gcc-12/performance/x86_64-rhel-9.4/Socket Activity/debian-12-x86_64-phoronix/lkp-csl-2sp7/stress-ng-1.11.0/phoronix-test-suite
>
> commit:
> 0c9567b36ae6f8 ("perf/x86: Remove redundant is_x86_event() prototype")
> a7138973beb1d1 ("perf/x86/intel: Fix NULL event access and potential PEBS record loss")
> 81248d31dd384c ("perf/x86/intel: Add counter group support for arch-PEBS")
> 54701e916f6782 ("Fixup: perf/x86/intel: Fix NULL event access waring from test robot")
>
> 0c9567b36ae6f83c a7138973beb1d124386472663cf 81248d31dd384c71c3b6a6af25a 54701e916f6782039c2ea8adb4b
> ---------------- --------------------------- --------------------------- ---------------------------
> fail:runs %reproduction fail:runs %reproduction fail:runs %reproduction fail:runs
> | | | | | | |
> :6 100% 6:6 100% 6:6 0% :20 dmesg.WARNING:at_arch/x86/events/intel/ds.c:#intel_pmu_drain_pebs_nhm

Oliver, thanks a lot for confirm this. :)



>
>
>> diff --git a/arch/x86/events/intel/ds.c b/arch/x86/events/intel/ds.c
>> index 65908880f424..3dedf7a0acf6 100644
>> --- a/arch/x86/events/intel/ds.c
>> +++ b/arch/x86/events/intel/ds.c
>> @@ -2781,9 +2781,11 @@ static void intel_pmu_drain_pebs_nhm(struct pt_regs
>> *iregs, struct perf_sample_d
>>  
>>          /* PEBS v3 has more accurate status bits */
>>          if (x86_pmu.intel_cap.pebs_format >= 3) {
>> -            for_each_set_bit(bit, (unsigned long *)&pebs_status, size)
>> +            for_each_set_bit(bit, (unsigned long *)&pebs_status, size) {
>>                  counts[bit]++;
>> -
>> +                if (counts[bit] && !events[bit])
>> +                    events[bit] = cpuc->events[bit];
>> +            }
>>              continue;
>>          }
>>  
>> @@ -2821,8 +2823,11 @@ static void intel_pmu_drain_pebs_nhm(struct pt_regs
>> *iregs, struct perf_sample_d
>>           * If collision happened, the record will be dropped.
>>           */
>>          if (pebs_status != (1ULL << bit)) {
>> -            for_each_set_bit(i, (unsigned long *)&pebs_status, size)
>> +            for_each_set_bit(i, (unsigned long *)&pebs_status, size) {
>>                  error[i]++;
>> +                if (error[i] && !events[i])
>> +                    events[i] = cpuc->events[i];
>> +            }
>>              continue;
>>          }
>>