Re: [PATCH] tracing, perf: add more power related events

[Ingo Molnar <mingo@elte.hu>]

* Thomas Renninger <trenn@suse.de> wrote:

> On Friday 03 September 2010 21:55:55 Ingo Molnar wrote:
> > 
> > * Thomas Renninger <trenn@suse.de> wrote:
> > 
> > > Hi Jean,
> > > 
> > > Summary at the beginning if people stop reading on the details
> > > below...:
> > > Nice work!
> > > You have my vote to get this merged.
> > > I think it's up to Ingo to push it.
> > 
> > Well, the patch that got posted:
> > 
> >   [PATCH] tracing, perf: add more power related events
> > 
> > Only adds tracepoints to power.h, but doesnt actually use them anywhere 
> > ...
> > 
> > Is there a companion patch i missed?
> Yep.
> Jean has nicely made up a tiny website he refered to in his mail:
> http://www.omappedia.com/wiki/Power_Management_Debug_and_Profiling

Ah, that wasnt in the first patch. (Please re-start discussion threads 
when something is re-submitted)

Nice website. Looks like these new tracepoints are really useful.

> There he shows the whole big patch and states:

Generally it's nice to post patches to lkml too, so that people have 
something to discuss.

> Since the patch touches generic files (in include/trace/events/power.h), it is split in parts:
>     * generic code: submitted to the LKML, cf. http://marc.info/?l=linux-kernel&m=128195697205096&w=4 ,
>     * ARM/OMAP specific code for the already existing events: to be submitted to l-o, cf. http://www.spinics.net/lists/linux-omap/msg35357.html,
>     * ARM/OMAP specific code for the new events, to be submitted once the generic code is accepted. 
> 
> > Also, it would be nice to add it to x86 and to OMAP at once, and 
> > synchronize up all the power events
>
> I can add sleep state events to x86 acpi parts, but even these are, as 
> said, not that interesting as they might be on ARM ultra mobile use 
> cases.

Yeah - but as long as they are at least attempted to be cross-platform, 
that would be fine. The ACPI folks are Cc:-ed so they can object if they 
dont like something.

> > - and check how existing tools like 
> > powertop make use of such events - and sync it all up.
>
> Does powertop use these events?

It certainly does quite a few things wrt. power usage analysis. Arjan is 
on the Cc:, maybe he has an opinion about these tracepoints.

> > We dont want to have inconsistent, architecture-dependent events in 
> > the end, for obvious reasons.
>
> Jean also shows screenshots at the link above
> how perf timechart or the the corresponding
> gui app PyTimechart still works together with these.
> I can't see how the new events should break an existing app, they should
> just ignore events they are not aware of.

Yeah, extensions rarely break anything. What i mentioned was 
'inconsistent, architecture-dependent events', which is not a fact but a 
possible danger: as these new events are only for OMAP right now. They 
might be OK, they might be not OK - i dont know, that's why i suggested 
to make them cross-platform in practice too.

So it would be nice to get some feedback from the ACPI folks - but if 
you guys can also do an ACPI patch that would be even better!

Thanks,

	Ingo