Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > linux.kernel > #1581192 > unrolled thread

Re: [PATCH v9 0/8] perf/amd/iommu: Enable multi-IOMMU support

Back to article view | Back to linux.kernel

This discussion starts older than the indexed window; earlier articles aren't shown. The article labeled Started by below is the oldest one visible, not the original post.

Contents

  Re: [PATCH v9 0/8] perf/amd/iommu: Enable multi-IOMMU support Borislav Petkov <bp@alien8.de> - 2017-02-15 11:20 +0100
    Re: [PATCH v9 0/8] perf/amd/iommu: Enable multi-IOMMU support Jiri Olsa <jolsa@redhat.com> - 2017-02-15 11:50 +0100
      Re: [PATCH v9 0/8] perf/amd/iommu: Enable multi-IOMMU support Borislav Petkov <bp@alien8.de> - 2017-02-15 12:20 +0100

#1581192 — Re: [PATCH v9 0/8] perf/amd/iommu: Enable multi-IOMMU support

On Tue, Feb 07, 2017 at 02:40:28AM -0600, Suravee Suthikulpanit wrote:
> From: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
> 
> This patch series modifies the existing IOMMU and Perf drivers to support
> systems with multiple IOMMUs by allocating an amd_iommu PMU per IOMMU instance.
> This allows users to specify performance events and filters separately for each
> IOMMU.
> 
> This has been tested on the new family17h-based server w/ multiple IOMMUs.

Ok, so far so good.

There's just one thing:

$ perf stat -e amd_iommu_X/Y

says <not supported> and only doing the system wide tracing with -a does
it count events.

So, lemme ask perf tool people, can we guys make the -a thing default
when detect that we're running only uncore events which all should need
-a anyway?

Otherwise people who, like me, don't know that, would try to do
measurements and see the <not supported> and wonder...

Lemme add some more people to CC.

-- 
Regards/Gruss,
    Boris.

Good mailing practices for 400: avoid top-posting and trim the reply.

[toc] | [next] | [standalone]

#1581205

On Wed, Feb 15, 2017 at 11:13:23AM +0100, Borislav Petkov wrote:
> On Tue, Feb 07, 2017 at 02:40:28AM -0600, Suravee Suthikulpanit wrote:
> > From: Suravee Suthikulpanit <suravee.suthikulpanit@amd.com>
> > 
> > This patch series modifies the existing IOMMU and Perf drivers to support
> > systems with multiple IOMMUs by allocating an amd_iommu PMU per IOMMU instance.
> > This allows users to specify performance events and filters separately for each
> > IOMMU.
> > 
> > This has been tested on the new family17h-based server w/ multiple IOMMUs.
> 
> Ok, so far so good.
> 
> There's just one thing:
> 
> $ perf stat -e amd_iommu_X/Y
> 
> says <not supported> and only doing the system wide tracing with -a does
> it count events.

does it say unsupported when you omit -a? it should display error
and options like:

[jolsa@krava perf]$ ./perf stat -e 'cpu/cpu-cycles/'

 Usage: perf stat [<options>] [<command>]

    -a, --all-cpus        system-wide collection from all CPUs
    -A, --no-aggr         disable CPU count aggregation
    -B, --big-num         print large numbers with thousands' separators
    -C, --cpu <cpu>       list of cpus to monitor in system-wide
...


> 
> So, lemme ask perf tool people, can we guys make the -a thing default
> when detect that we're running only uncore events which all should need
> -a anyway?

it's possible, can't think of anything being hurt by this now..

jirka

[toc] | [prev] | [next] | [standalone]

#1581225

On Wed, Feb 15, 2017 at 11:44:24AM +0100, Jiri Olsa wrote:
> does it say unsupported when you omit -a?

Yeah, it does dump the event name like it was counting but then it says
<not supported>. For example:

   <not supported>      amd_iommu_0/...

> [jolsa@krava perf]$ ./perf stat -e 'cpu/cpu-cycles/'
	 ^^^^^

Haha, I chuckle everytime I see this hostname :-)))

> > So, lemme ask perf tool people, can we guys make the -a thing default
> > when detect that we're running only uncore events which all should need
> > -a anyway?
> 
> it's possible, can't think of anything being hurt by this now..

Right, so Peter meant something simple like: you parse all events, see
that they're all uncore and enable system_wide if so.

I get the feeling that we want to do that for uncore-only events
anyway...

-- 
Regards/Gruss,
    Boris.

Good mailing practices for 400: avoid top-posting and trim the reply.

[toc] | [prev] | [standalone]

Back to top | Article view | linux.kernel

csiph-web