Re: [PATCH v1 03/10] perf/x86/uncore: do not assume PCI fixed ctrs have more than 32 bits

[Stephane Eranian]

Yan,

In fact, I realized I was not even using all those fixed counter
changes, thanks to the
pmu bypass from patch 1. So I will drop all of this in V2. It will
make the patch simpler.

Thanks for catching this.


On Mon, Feb 10, 2014 at 4:11 AM, Yan, Zheng <zheng.z.yan@xxxxxxxxx> wrote:
> On 02/03/2014 08:55 PM, Stephane Eranian wrote:
>> The current code assumes all PCI fixed counters implement more than
>> 32-bit hardware counters. The actual width is then round up to 64 to
>> enable base + 8 * idx calculations.
>>
>> Not all PMUs necessarily implement counters with more than 32-bits.
>> The patch makes the uncore_pci_perf_ctr() function dynamically
>> determine the actual bits width of a pci perf counter.
>>
>> This patch paves the way for handling more than one uncore fixed
>> counter per PMU box.
>>
>> Signed-off-by: Stephane Eranian <eranian@xxxxxxxxxx>
>> ---
>>  arch/x86/kernel/cpu/perf_event_intel_uncore.h |   31 ++++++++++++++++---------
>>  1 file changed, 20 insertions(+), 11 deletions(-)
>>
>> diff --git a/arch/x86/kernel/cpu/perf_event_intel_uncore.h b/arch/x86/kernel/cpu/perf_event_intel_uncore.h
>> index 77dc9a5..f5549cf 100644
>> --- a/arch/x86/kernel/cpu/perf_event_intel_uncore.h
>> +++ b/arch/x86/kernel/cpu/perf_event_intel_uncore.h
>> @@ -548,10 +548,29 @@ unsigned uncore_pci_event_ctl(struct intel_uncore_box *box, int idx)
>>       return idx * 4 + box->pmu->type->event_ctl;
>>  }
>>
>> +static inline int uncore_perf_ctr_bits(struct intel_uncore_box *box)
>> +{
>> +     return box->pmu->type->perf_ctr_bits;
>> +}
>> +
>> +static inline int uncore_fixed_ctr_bits(struct intel_uncore_box *box)
>> +{
>> +     return box->pmu->type->fixed_ctr_bits;
>> +}
>> +
>>  static inline
>>  unsigned uncore_pci_perf_ctr(struct intel_uncore_box *box, int idx)
>>  {
>> -     return idx * 8 + box->pmu->type->perf_ctr;
>> +     int bits, bytes;
>> +
>> +     if (idx == UNCORE_PMC_IDX_FIXED)
>> +             bits = uncore_fixed_ctr_bits(box);
>> +     else
>> +             bits = uncore_perf_ctr_bits(box);
>> +
>> +     bytes = round_up(bits, 8);
>
> should this be "round_up(bits, 32) / 8" ?
>
> Regards
> Yan, Zheng
>
>> +
>> +     return idx * bytes + box->pmu->type->perf_ctr;
>>  }
>>
>>  static inline unsigned uncore_msr_box_offset(struct intel_uncore_box *box)
>> @@ -633,16 +652,6 @@ unsigned uncore_perf_ctr(struct intel_uncore_box *box, int idx)
>>               return uncore_msr_perf_ctr(box, idx);
>>  }
>>
>> -static inline int uncore_perf_ctr_bits(struct intel_uncore_box *box)
>> -{
>> -     return box->pmu->type->perf_ctr_bits;
>> -}
>> -
>> -static inline int uncore_fixed_ctr_bits(struct intel_uncore_box *box)
>> -{
>> -     return box->pmu->type->fixed_ctr_bits;
>> -}
>> -
>>  static inline int uncore_num_counters(struct intel_uncore_box *box)
>>  {
>>       return box->pmu->type->num_counters;
>>
>
--
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/