Re: [RFC PATCH 1/1] smp: Add tracepoints for functions called with smp_call_function*()

[Peter Zijlstra]

On Thu, May 04, 2023 at 12:59:41PM +0100, Valentin Schneider wrote:

> With that said, I suppose this could still be helpful for e.g. osnoise to
> hook into and point the finger at which CPU/context sent the problematic
> IPI. Or more generally, as Leonardo suggested, to measure CSD IPI delivery
> times.
> 
> One thing though is that trace_ipi_send_cpu*() is not used solely for
> CSD's, cf. irq_work_raise() or smp_send_reschedule(). We might want to
> split that into e.g. trace_csd_queue_cpu*() + trace_ipi_send*().
> 
> Something like so...
> 

Yeah, looks about right I suppose.. it generates more events, but given
we need the csd that's inevitable.