Re: [PATCH v1 2/6] perf trace: Ignore thread hashing in summary

From: Arnaldo Carvalho de Melo <acme@kernel.org>
To: Ian Rogers <irogers@google.com>
Cc: Peter Zijlstra <peterz@infradead.org>,
	Ingo Molnar <mingo@redhat.com>,
	Namhyung Kim <namhyung@kernel.org>,
	Mark Rutland <mark.rutland@arm.com>,
	Alexander Shishkin <alexander.shishkin@linux.intel.com>,
	Jiri Olsa <jolsa@kernel.org>,
	Adrian Hunter <adrian.hunter@intel.com>,
	Oliver Upton <oliver.upton@linux.dev>,
	Yang Jihong <yangjihong1@huawei.com>,
	linux-kernel@vger.kernel.org, linux-perf-users@vger.kernel.org,
	bpf@vger.kernel.org
Subject: Re: [PATCH v1 2/6] perf trace: Ignore thread hashing in summary
Date: Fri, 16 Feb 2024 11:57:32 -0300	[thread overview]
Message-ID: <Zc933IYjWlkMo4D2@x1> (raw)
In-Reply-To: <CAP-5=fWweUBP_-SHfoADswizMER6axNw89JyG7Fo_qiC883fNw@mail.gmail.com>

On Wed, Feb 14, 2024 at 01:36:46PM -0800, Ian Rogers wrote:
> On Wed, Feb 14, 2024 at 1:15 PM Ian Rogers <irogers@google.com> wrote:
> > On Wed, Feb 14, 2024 at 10:27 AM Ian Rogers <irogers@google.com> wrote:
> > > On Wed, Feb 14, 2024 at 9:25 AM Arnaldo Carvalho de Melo
> > > <acme@kernel.org> wrote:
> > > > On Tue, Feb 13, 2024 at 10:37:04PM -0800, Ian Rogers wrote:
> > > > > Commit 91e467bc568f ("perf machine: Use hashtable for machine
> > > > > threads") made the iteration of thread tids unordered. The perf trace
> > > > > --summary output sorts and prints each hash bucket, rather than all
> > > > > threads globally. Change this behavior by turn all threads into a
> > > > > list, sort the list by number of trace events then by tids, finally
> > > > > print the list. This also allows the rbtree in threads to be not
> > > > > accessed outside of machine.

> > > > Can you please provide a refresh of the output that is changed by your patch?
> > >
> > > Hmm.. looks like perf trace record has broken and doesn't produce
> > > output in newer perfs. It works on 6.5 and so a bisect is necessary.
> >
> > Bisect result:
> > ```
> > 9925495d96efc14d885ba66c5696f664fe0e663c is the first bad commit
> > commit 9925495d96efc14d885ba66c5696f664fe0e663c
> > Author: Ian Rogers <irogers@google.com>
> > Date:   Thu Sep 14 14:19:45 2023 -0700
> >
> >    perf build: Default BUILD_BPF_SKEL, warn/disable for missing deps
> > ...
> > https://lore.kernel.org/r/20230914211948.814999-3-irogers@google.com
> > ```
> >
> > Now to do the bisect with BUILD_BPF_SKEL=1 on each make.
> 
> This looks better (how could I be at fault :-) ):
> ```
> 1836480429d173c01664a633b61e525b13d41a2a is the first bad commit
> commit 1836480429d173c01664a633b61e525b13d41a2a
> Author: Arnaldo Carvalho de Melo <acme@redhat.com>
> Date:   Wed Aug 16 13:53:26 2023 -0300
> 
>    perf bpf_skel augmented_raw_syscalls: Cap the socklen parameter
> using &= sizeof(saddr)
> ...
>    Cc: Adrian Hunter <adrian.hunter@intel.com>
>    Cc: Ian Rogers <irogers@google.com>
>    Cc: Jiri Olsa <jolsa@kernel.org>
>    Cc: Namhyung Kim <namhyung@kernel.org>
>    Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
> ```
> No LKML link.

So simple... ;-\

I've reproduced your steps and got to the same cset while testing on a
recent distro kernel (6.6.13-200.fc39.x86_64), scratching my head now
and trying to figure this out.

Wonder if trying to run on an older kernel the problem would appear.

Will try and add a perf test shell entry with a simple:

root@number:~# perf trace record sleep 0.001 && perf script | head | wc -l
[ perf record: Woken up 1 times to write data ]
[ perf record: Captured and wrote 0.034 MB perf.data ]
0
root@number:~#

Has to be 10 :-)

Thanks,

- Arnaldo