On 11/21/2025 6:28 AM, Sean Christopherson wrote:
> On Wed, Sep 03, 2025, Dapeng Mi wrote:
>> From: dongsheng <dongsheng.x.zhang@intel.com>
>>
>> The current implementation mistakenly limits the width of fixed counters
>> to the width of GP counters. Corrects the logic to ensure fixed counters
>> are properly masked according to their own width.
>>
>> Opportunistically refine the GP counter bitwidth processing code.
>>
>> Signed-off-by: dongsheng <dongsheng.x.zhang@intel.com>
>> Co-developed-by: Dapeng Mi <dapeng1.mi@linux.intel.com>
>> Signed-off-by: Dapeng Mi <dapeng1.mi@linux.intel.com>
>> Tested-by: Yi Lai <yi1.lai@intel.com>
>> ---
>> x86/pmu.c | 8 +++-----
>> 1 file changed, 3 insertions(+), 5 deletions(-)
>>
>> diff --git a/x86/pmu.c b/x86/pmu.c
>> index 04946d10..44c728a5 100644
>> --- a/x86/pmu.c
>> +++ b/x86/pmu.c
>> @@ -556,18 +556,16 @@ static void check_counter_overflow(void)
>> int idx;
>>
>> cnt.count = overflow_preset;
>> - if (pmu_use_full_writes())
>> - cnt.count &= (1ull << pmu.gp_counter_width) - 1;
>> -
>> if (i == pmu.nr_gp_counters) {
>> if (!pmu.is_intel)
>> break;
>>
>> cnt.ctr = fixed_events[0].unit_sel;
>> - cnt.count = (&cnt);
> Per commit 7ec3b67a ("x86/pmu: Reset the expected count of the fixed counter 0
> when i386"), re-measuring for the fixed counter is necessary when running a 32-bit
> guests. I didn't see failures (spotted this by inspection), but I don't see any
> point in making this change without good reason.
Thanks. I didn't realized that the 2nd measure_for_overflow() is intended
to add ...
>
>> - cnt.count &= (1ull << pmu.gp_counter_width) - 1;
>> + cnt.count &= (1ull << pmu.fixed_counter_width) - 1;
>> } else {
>> cnt.ctr = MSR_GP_COUNTERx(i);
>> + if (pmu_use_full_writes())
>> + cnt.count &= (1ull << pmu.gp_counter_width) - 1;
>> }
>>
>> if (i % 2)
>> --
>> 2.34.1
>>