lkml.org 
[lkml]   [2022]   [Apr]   [26]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH 2/3] perf: arm-spe: Fix SPE events with phys addresses
On Mon, Apr 25, 2022 at 10:12:36AM +0100, James Clark wrote:

[...]

> >> diff --git a/tools/perf/util/arm-spe.c b/tools/perf/util/arm-spe.c
> >> index 151cc38a171c..1a80151baed9 100644
> >> --- a/tools/perf/util/arm-spe.c
> >> +++ b/tools/perf/util/arm-spe.c
> >> @@ -1033,7 +1033,8 @@ arm_spe_synth_events(struct arm_spe *spe, struct perf_session *session)
> >> memset(&attr, 0, sizeof(struct perf_event_attr));
> >> attr.size = sizeof(struct perf_event_attr);
> >> attr.type = PERF_TYPE_HARDWARE;
> >> - attr.sample_type = evsel->core.attr.sample_type & PERF_SAMPLE_MASK;
> >> + attr.sample_type = evsel->core.attr.sample_type &
> >> + (PERF_SAMPLE_MASK | PERF_SAMPLE_PHYS_ADDR);
> >
> > I verified this patch and I can confirm the physical address can be
> > dumped successfully.
> >
> > I have a more general question, seems to me, we need to change the
> > macro PERF_SAMPLE_MASK in the file util/event.h as below, so
> > here doesn't need to 'or' the flag PERF_SAMPLE_PHYS_ADDR anymore.
> >
> > @Arnaldo, @Jiri, could you confirm if this is the right way to move
> > forward? I am not sure why PERF_SAMPLE_MASK doesn't contain the bit
> > PERF_SAMPLE_PHYS_ADDR in current code.
>
> I think there is a reason that PERF_SAMPLE_MASK is a subset of all the
> bits. This comment below suggests it. Is it so the mask only includes fields
> that are 64bits? That makes the __evsel__sample_size() function a simple
> multiplication of a count of all the fields that are 64bits.

After reading code, seems the conclusion "a count of all the fields
that are 64bits" is not valid. PERF_SAMPLE_MASK contains bits
PERF_SAMPLE_IP and PERF_SAMPLE_TID, the corresponding fields 'pid'
and 'tid' in the structure perf_sample are u32 type.

> static int
> perf_event__check_size(union perf_event *event, unsigned int sample_size)
> {
> /*
> * The evsel's sample_size is based on PERF_SAMPLE_MASK which includes
> * up to PERF_SAMPLE_PERIOD. After that overflow() must be used to
> * check the format does not go past the end of the event.
> */
> if (sample_size + sizeof(event->header) > event->header.size)
> return -EFAULT;

Okay, thanks for sharing the info, it does show that it's deliberately to
not contain all fields in PERF_SAMPLE_MASK. If so, this patch is fine
for me:

Reviewed-by: Leo Yan <leo.yan@linaro.org>


> return 0;
> }
>
> Having said that, the mask was updated once to add PERF_SAMPLE_IDENTIFIER to
> it, so that comment is slightly out of date now.
>
>
> >
> > diff --git a/tools/perf/util/event.h b/tools/perf/util/event.h
> > index cdd72e05fd28..c905ac32ebad 100644
> > --- a/tools/perf/util/event.h
> > +++ b/tools/perf/util/event.h
> > @@ -39,7 +39,7 @@ struct perf_event_attr;
> > PERF_SAMPLE_TIME | PERF_SAMPLE_ADDR | \
> > PERF_SAMPLE_ID | PERF_SAMPLE_STREAM_ID | \
> > PERF_SAMPLE_CPU | PERF_SAMPLE_PERIOD | \
> > - PERF_SAMPLE_IDENTIFIER)
> > + PERF_SAMPLE_IDENTIFIER | PERF_SAMPLE_PHYS_ADDR)
> >
> > Thanks,
> > Leo
> >
> >> attr.sample_type |= PERF_SAMPLE_IP | PERF_SAMPLE_TID |
> >> PERF_SAMPLE_PERIOD | PERF_SAMPLE_DATA_SRC |
> >> PERF_SAMPLE_WEIGHT | PERF_SAMPLE_ADDR;
> >> --
> >> 2.25.1
> >>

\
 
 \ /
  Last update: 2022-04-26 15:21    [W:0.518 / U:0.032 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site