Perf doesn't accept i915:i915_gem_request_wait_begin as an event

Perf doesn't accept i915:i915_gem_request_wait_begin as an event

[Fabian Korak]

Hi,

I'm currently trying to get the examples at http://dri.freedesktop.org/wiki/IntelPerformanceTuning to run, however when I try

perf record -f -g -e i915:i915_gem_request_wait_begin -c 1 openarena

I get the message "invalid or unsupported event: 'i915:i915_gem_request_wait_begin' ". It also isn't listed within perf list.
I'm currently on an up-to-date Fedora Core 14 rawhide with kernel 2.6.34-0.4.rc0.git2. Intel_gpu_tools and libdwarf is installed.
 I talked to a few guys on #xorg-devel yesterday and they told me to install kernel-debuginfo, which didn't change anything.
They also told me to run

nm /lib/modules/2.6.34-0.4.rc0.git2.fc14.x86_64/kernel/drivers/gpu/drm/i915/i915.ko | grep i915_gem

the output of that can be found here: http://www.pastebin.org/102547

Any help would be appreciated.

thx Fabian Korak

Re: Perf doesn't accept i915:i915_gem_request_wait_begin as an event

[Corey Ashford]

On 03/07/2010 06:00 AM, Fabian Korak wrote:
> Hi,
>
> I'm currently trying to get the examples at http://dri.freedesktop.org/wiki/IntelPerformanceTuning to run, however when I try
>
> perf record -f -g -e i915:i915_gem_request_wait_begin -c 1 openarena
>
> I get the message "invalid or unsupported event: 'i915:i915_gem_request_wait_begin' ". It also isn't listed within perf list.
> I'm currently on an up-to-date Fedora Core 14 rawhide with kernel 2.6.34-0.4.rc0.git2. Intel_gpu_tools and libdwarf is installed.
>   I talked to a few guys on #xorg-devel yesterday and they told me to install kernel-debuginfo, which didn't change anything.
> They also told me to run
>
> nm /lib/modules/2.6.34-0.4.rc0.git2.fc14.x86_64/kernel/drivers/gpu/drm/i915/i915.ko | grep i915_gem
>
> the output of that can be found here: http://www.pastebin.org/102547
>
> Any help would be appreciated.
>
> thx Fabian Korak
> --
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at  http://www.tux.org/lkml/

I find the example given on that web page quite confusing because perf 
currently doesn't have support for arch-dependent events like that, 
except via "raw code" events which use the syntax -e r<hex_code>.
I wonder if they locally modified perf to handle those events, and then 
forgot that they had made that modification before posting up those 
instructions.

I posted a patch last Wednesday that adds arch-dep symbolic event symbol 
support which uses an external library (like libpfm4), but so far I 
haven't received any feedback on it.

-- 
Regards,

- Corey

Corey Ashford
Software Engineer
IBM Linux Technology Center, Linux Toolchain
Beaverton, OR
cjashfor@us.ibm.com

Re: Perf doesn't accept i915:i915_gem_request_wait_begin as an event

[Arjan van de Ven]

On Sun, 07 Mar 2010 11:01:39 -0800
Corey Ashford <cjashfor@linux.vnet.ibm.com> wrote:
> 
> I find the example given on that web page quite confusing because
> perf currently doesn't have support for arch-dependent events like
> that, except via "raw code" events which use the syntax -e
> r<hex_code>. I wonder if they locally modified perf to handle those
> events, and then forgot that they had made that modification before
> posting up those instructions.

these are just standard trace points...


-- 
Arjan van de Ven 	Intel Open Source Technology Centre
For development, discussion and tips for power savings, 
visit http://www.lesswatts.org

Re: Perf doesn't accept i915:i915_gem_request_wait_begin as an event

[Corey Ashford]

On 03/07/2010 11:04 AM, Arjan van de Ven wrote:
> On Sun, 07 Mar 2010 11:01:39 -0800
> Corey Ashford<cjashfor@linux.vnet.ibm.com>  wrote:
>>
>> I find the example given on that web page quite confusing because
>> perf currently doesn't have support for arch-dependent events like
>> that, except via "raw code" events which use the syntax -e
>> r<hex_code>. I wonder if they locally modified perf to handle those
>> events, and then forgot that they had made that modification before
>> posting up those instructions.
>
> these are just standard trace points...

Ah, I see.  Thanks for the correction.

- Corey