Re: [PATCH V3 1/4] perf: Add PERF_RECORD_SWITCH to indicate context switches

[Arnaldo Carvalho de Melo <acme@kernel.org>]

Em Tue, Jul 07, 2015 at 10:25:52AM -0300, Arnaldo Carvalho de Melo escreveu:
> Em Tue, Jul 07, 2015 at 11:36:39AM +0300, Adrian Hunter escreveu:
> > To make the event useful for privileged users also, if the
> > context is cpu-wide then the event will also provide the
> > next or previous pid/tid.
> 
> <SNIP>
>   
> > +	/*
> > +	 * Records a context switch in or out (flagged by
> > +	 * PERF_RECORD_MISC_SWITCH_OUT).  next_prev_pid and next_prev_tid are
> > +	 * (u32)-1 unless the context is cpu-wide, in which case they are the
> 
> Why carry those extra 8 bytes for non priviledged users, all the time
> with -1?

> Can't userspace cope with this, i.e. we should be able to look for those
> fields when the context is CPU wide, and to not look for them otherwise,
> no?

To help userspace in places where all it has is the union perf_event, we
can reuse one bit in misc to state that, i.e.

  #define PERF_RECORD_MISC_SWITCH_NEXT_PREV_PID 14

For instance.

- Arnaldo
 
> > +	 * next (switching out) or previous (switching in) pid/tid.
> > +	 *
> > +	 * struct {
> > +	 *	struct perf_event_header	header;
> > +	 *	u32				next_prev_pid;
> > +	 *	u32				next_prev_tid;
> > +	 *	struct sample_id		sample_id;
> > +	 * };
> > +	 */
> > +	PERF_RECORD_SWITCH			= 14,
> > +