Re: [PATCH v5 13/17] perf: Support deferred user callchains

[Namhyung Kim <namhyung@kernel.org>]

Hi Mathieu,

On Fri, May 09, 2025 at 08:23:49AM -0400, Mathieu Desnoyers wrote:
> On 2025-05-08 14:54, Steven Rostedt wrote:
> > On Thu, 8 May 2025 14:49:59 -0400
> > Mathieu Desnoyers <mathieu.desnoyers@efficios.com> wrote:
> > 
> > > AFAIR, the cookie method generates the cookie by combining the cpu
> > > number with a per-cpu count.
> > > 
> > > This ensures that there are not two cookies emitted at the same time
> > > from two CPUs that have the same value by accident.
> > > 
> > > How would the timestamp method prevent this ?
> > 
> > Do we care? It only needs to be unique per pid doesn't it?
> 
> Is it possible to have many threads writing into the same
> ring buffer in that scenario ? Are all event records stamped
> with their associated PID ? As long as we have enough information
> to know which thread was associated with the timestamp cookie
> on both ends (request for callchain and saving the user callchain
> on return to userspace), we should be OK.

Yep, basically perf sets PERF_SAMPLE_TID (and sample_id_all) which makes
every records come with PID/TID.. 

Thanks,
Namhyung