Re: perf top -p broken for multithreaded processes since 5.19

[Adrian Hunter <adrian.hunter@intel.com>]

On 3/09/22 17:08, Arnaldo Carvalho de Melo wrote:
> Em Sat, Sep 03, 2022 at 10:14:25AM +0300, Adrian Hunter escreveu:
>> On 2/09/22 22:17, Arnaldo Carvalho de Melo wrote:
>>> Em Fri, Sep 02, 2022 at 05:50:22PM +0300, Adrian Hunter escreveu:
>>>> On 2/09/22 17:46, Tomáš Trnka wrote:
>>>>> Hello,
>>>>>
>>>>> A bug in perf v5.19 and newer completely breaks monitoring multithreaded
>>>>> processes using "perf top -p". The tool fails to start with "Failed to mmap
>>>>> with 22 (Invalid argument)". It still seems to work fine on single-threaded
>>>>> processes. "perf record" is also unaffected.
>>>>
>>>> It has been reported here:
>>>>
>>>> 	https://bugzilla.kernel.org/show_bug.cgi?id=216441
>>>
>>> If I do:
>>>
>>> ⬢[acme@toolbox perf-urgent]$ git log -2
>>> commit dfeb0bc60782471c293938e71b1a1117cfac2cb3 (HEAD -> perf/urgent)
>>> Author: Arnaldo Carvalho de Melo <acme@redhat.com>
>>> Date:   Fri Sep 2 16:15:39 2022 -0300
>>>
>>>     Revert "libperf evlist: Check nr_mmaps is correct"
>>>
>>>     This reverts commit 4ce47d842d4c16c07b135b8a7975b8f0672bcc0e.
>>>
>>>     Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
>>>
>>> commit 78cd283f6b8ab701cb35eafd5af8140560a88f16
>>> Author: Arnaldo Carvalho de Melo <acme@redhat.com>
>>> Date:   Fri Sep 2 16:13:41 2022 -0300
>>>
>>>     Revert "libperf evlist: Allow mixing per-thread and per-cpu mmaps"
>>>
>>>     This reverts commit ae4f8ae16a07896403c90305d4b9be27f657c1fc.
>>>
>>>     Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
>>> ⬢[acme@toolbox perf-urgent]$
>>>
>>> It works again, Tomáš can you please try doing this to see if this works
>>> for you?
>>>
>>
>> This is the fix I have so far.  I would like to test it some more though.
> 
> Ok, so I'll leave it for the next pull req, possibly after Linux
> Plumbers.
> 
> What do you think about reverting those two patches for v6.0 and then
> add this for v6.1?

That would break sideband collection with selected CPUs, so I would
prefer to avoid that.

The fix is a effectively a partial revert of "libperf evlist: Allow
mixing per-thread and per-cpu mmaps" anyway.  I just need a few days
to do more testing.