Re: [PATCH v5 09/11] coresight: etm4x: Add flag to control single-shot restart

[Leo Yan <leo.yan@arm.com>]

On Tue, Nov 11, 2025 at 11:37:08AM +0000, Suzuki Kuruppassery Poulose wrote:
> On 11/11/2025 10:18, Leo Yan wrote:
> > On Tue, Nov 11, 2025 at 09:43:13AM +0000, Suzuki Kuruppassery Poulose wrote:
> > 
> > [...]
> > 
> > > >     if (!is_idle_task(current)) {
> > > >         if (config->ss_ctrl[i] || config->ss_pe_cmp[i])
> > > >             config->ss_status[i] &= ~TRCSSCSRn_STATUS;
> > > >     }
> > > 
> > > What happens when we do a system wide session and a CPU is running an
> > > idle task ?
> > 
> > For a system wide trace, the ETM device is enabled and disabled by
> > the perf when the session start and finish.
> > 
> > During the session:
> > 
> > - If CPU off states are not enabled, nothing will happen in idle
> >    threads, as no CPU PM notifier will be invoked.
> 
> What is a CPU is running the idle task, when the system wide session
> is activated ?

Just makes clear: if a CPU running in idle task, and it controls ETM,
it must be a CPU PM operation.  Otherwise, it is no reason for idle
thread to enable/disable ETM, no matter what the session mode is.

An edge case is in per-thread mode when switch between the profiled
process and the idle thread, the ETM operations always occur in
profiled process rather than idle threads.

> Basing our decision to tune the ETM control on the
> "active task being idle" feels fragile to me. This is purely a
> decision under ETM driver, specific to the session you are dealing
> with. Thus having it in the "config" is preferred and is cleaner.

It is fine for me to use "config" method.  Will spin in this way.

Thanks,
Leo