Re: [V3][PATCH 0/7] perf, x86: Implement AMD IBS

[Ingo Molnar]

* Robert Richter <robert.richter@xxxxxxx> wrote:

> On 04.10.11 04:54:09, Ingo Molnar wrote:
> > So the whole IBS thing looks quite unintegrated to me - and that's 
> > partly because the hw is admittedly weird. The way we could perhaps 
> > live with it upstream is two conditions:
> > 
> >  - Testable IBS user-space code a bit more prominently integrated
> >    than having to go down into a cellar with no working lights and 
> >    finding the code on display in tools/perf/Documentation/examples/ 
> >    on the bottom of a locked filing cabinet stuck in a disused
> >    lavatory with a sign on the door saying 'Beware of the Leopard.'
> 
> With Lin Ming's patch
> 
>  [PATCH v2 6/6] perf tool: Parse general/raw events from sysfs
> 
> you can use perf record to collect IBS samples, e.g.:
> 
>  # perf record -e ibs_fetch:r0 -c 100000 -a bash -c someload
>  [ perf record: Woken up 3 times to write data ]
>  [ perf record: Captured and wrote 2.004 MB perf.data (~87572 samples) ]
> 
> With Stephane's patch
> 
>  [PATCH] perf: make perf.data more self-descriptive (v8)
> 
> I could extend perf report to add an IBS parser (I need to know the
> pmu name/type value mapping).
> 
> Do you have something like that in mind?
> 
> As all these patches are still off-tree I would rather prefer to 
> decouple the IBS patches from a parser implementation and add it 
> later as a single patch series. Hmm?

Yeah, i'd like to have something that i can try out in practice on an 
AMD box without having to apply any other patches.

The perf.data extension patch from Stephane is now in perf/core. You 
could add Lin Ming's patch to your tree - they really interdepend. 
That way it could be reviewed and handled as a coherent unit offering 
some real, testable functionality.

> >  - Only root/privileged users should be able to access it. Right now
> >    i think it's root-only due to percpu restrictions, but wanted to 
> >    mention it that this is an explicit requirement.
> 
> Yes, IBS is restricted to percpu. And percpu requires root 
> privileges like all other pmu events. What is your point here?

My point is that this should not just be an incidental side-effect of 
hardware limitations but should stay so until deeper integration is 
achieved. The list i gave is similar for other future PMU additions.

Anyway, if perf report [and, in latest perf/core, perf top] can do 
IBS output that would be a sufficient level of integration i think.

Thanks,

	Ingo
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/