On 11/10/22 18:22, Daniel Bristot de Oliveira wrote:

On 10/11/22 18:17, Valentin Schneider wrote:

quoted

Thinking out loud, it makes way more sense to record a cpumask in the
tracepoint, but perhaps we could have a postprocessing step to transform
those into N events each targeting a single CPU?

My approach on the tracers/rtla is to make the simple things in kernel, and beautify
things in user-space.

You could keep the tracepoint as a mask, and then make it pretty, like cpus=3-5,8
in user-space. For example with a trace-cmd/perf loadable plugin, libtracefs helper.

That's a nice idea, the one downside I see is that means registering an
event handler for all events with cpumasks rather than directly targeting
cpumask fields, but that doesn't look too horrible. I'll dig a bit in that
direction.

For rtla I was thinking to make a new tool to parse them. and make it pretty there.

-- Daniel