lkml.org 
[lkml]   [2021]   [Feb]   [7]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
SubjectRe: [PATCH 6/9] perf report: Support instruction latency
From
Date


> On 05-Feb-2021, at 8:21 PM, Liang, Kan <kan.liang@linux.intel.com> wrote:
>
>
>
> On 2/5/2021 7:55 AM, Athira Rajeev wrote:
>>>> Because in other archs, the var2_w of ‘perf_sample_weight’ could be used to capture something else than the Local INSTR Latency.
>>>> Can we have some weak function to populate the header string ?
>>> I agree that the var2_w has different meanings among architectures. We should not force it to data->ins_lat.
>>>
>>> The patch as below should fix it. Does it work for you?
>> My point about weak function was actually for the arch specific header string. But I guess we should not force it to data->ins_lat
>
> Yes, I don't think PowerPC should force var2_w to data->ins_lat. I think you can create your own field.
>
>> as you mentioned. I checked the below patch defining an ‘arch_perf_parse_sample_weight' for powerpc and it works.
>> But one observation is that, for cases with kernel having support for PERF_SAMPLE_WEIGHT_STRUCT but missing arch specific support for ‘arch_perf_parse_sample_weight', it will report ‘Local Weight’ wrongly since weak function takes it as 64 bit. Not sure if that is a valid case to consider though.
>
> Currently, the PERF_SAMPLE_WEIGHT_STRUCT is only enabled on X86 by default.
> https://lore.kernel.org/lkml/1612296553-21962-6-git-send-email-kan.liang@linux.intel.com/
>
> For PowerPC, the PERF_SAMPLE_WEIGHT is still the default setting. There is no way to set PERF_SAMPLE_WEIGHT_STRUCT via perf tool.
> I don't think the above case will happen.

Yes.

I tested with kernel changes from perf/core branch of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip.git
And perf tools side changes from tmp.perf/core branch of git://git.kernel.org/pub/scm/linux/kernel/git/acme/linux.git along with the above change.
The default setting for powerpc works with out breaking anything and verified using “perf mem record <workload>”

Tested-by: Athira Rajeev <atrajeev@linux.vnet.ibm.com>

Thanks
Athira Rajeev
>
> Thanks,
> Kan

\
 
 \ /
  Last update: 2021-02-07 17:47    [W:0.084 / U:10.556 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site