Re: [PATCH v3] tracing: Guard __DECLARE_TRACE() use of __DO_TRACE_CALL() with SRCU-fast

[Mathieu Desnoyers]

On 2025-12-12 21:18, Steven Rostedt wrote:
[...]
> Thus, I'm going to keep this a PREEMPT_RT only change. If someone can
> come in and convince us that the PREEMPT_RT way is also beneficial for
> the non-RT case then we can make it consistent again. Until then, this
> change is focusing on fixing PREEMPT_RT, and that's what the patch is
> going to be limited to.

Here is one additional thing to keep in mind: although
SRCU-fast is probably quite fast (as the name implies),
last time I tried using migrate disable in a fast path
I was surprised to see verbosity of the generated assembly,
and how slow it was compared to preempt disable.

So before using migrate disable on a fast path, at least on
non-preempt-RT configs, we should carefully consider the
performance impact of migrate disable.

Thanks,

Mathieu

--
Mathieu Desnoyers
EfficiOS Inc.
https://www.efficios.com