Does Perf support AMD cpu perfectly?

Does Perf support AMD cpu perfectly?

[Chulmin Kim]

Hi all.

I'm on the way of monitoring AMD machine.

I wonder Perf is supporting AMD PMU Events perfectly or not.

It seems not working correctly despite the same command works in intel
machines perfectly.

command : (perf stat -e rc0 -e instructions ./blah)
result : the value for "rc0" almost equals to that of "instruction" in
Intel machine, while it does not in AMD machine


Other events such as cache events seem weird also.

Is there any material mentioning this issue?



Thanks!

Re: Does Perf support AMD cpu perfectly?

[Chulmin Kim]

Update for this issue.

- rc0 and instruction events show the similar result.
- still, the count of L3 cache miss event is 1/10 of what i expected.
(LLC-load-misses)



2012-11-26 오후 8:24, Chulmin Kim 쓴 글:
> Hi all.
> 
> I'm on the way of monitoring AMD machine.
> 
> I wonder Perf is supporting AMD PMU Events perfectly or not.
> 
> It seems not working correctly despite the same command works in intel
> machines perfectly.
> 
> command : (perf stat -e rc0 -e instructions ./blah)
> result : the value for "rc0" almost equals to that of "instruction" in
> Intel machine, while it does not in AMD machine
> 
> 
> Other events such as cache events seem weird also.
> 
> Is there any material mentioning this issue?
> 
> 
> 
> Thanks!
> 
> --
> To unsubscribe from this list: send the line "unsubscribe linux-perf-users" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> 

Re: Does Perf support AMD cpu perfectly?

[Chulmin Kim]

Update for this issue.

I'm investigating the kernel source.

I found that Perf for AMD seems imperfect in linux 3.0 (my machine's
linux version).

For AMD family 15h, there is a commented line as follow.
 - /* nortbridge counters not yet implemented: */  (line 541,
perf_event_amd.c )


According to my further exploration, linux 3.2 does not support it either.

Does any of you know or have the patch for it?


2012-11-26 오후 9:15, Chulmin Kim 쓴 글:
> Update for this issue.
> 
> - rc0 and instruction events show the similar result.
> - still, the count of L3 cache miss event is 1/10 of what i expected.
> (LLC-load-misses)
> 
> 
> 
> 2012-11-26 오후 8:24, Chulmin Kim 쓴 글:
>> Hi all.
>>
>> I'm on the way of monitoring AMD machine.
>>
>> I wonder Perf is supporting AMD PMU Events perfectly or not.
>>
>> It seems not working correctly despite the same command works in intel
>> machines perfectly.
>>
>> command : (perf stat -e rc0 -e instructions ./blah)
>> result : the value for "rc0" almost equals to that of "instruction" in
>> Intel machine, while it does not in AMD machine
>>
>>
>> Other events such as cache events seem weird also.
>>
>> Is there any material mentioning this issue?
>>
>>
>>
>> Thanks!
>>
>> --
>> To unsubscribe from this list: send the line "unsubscribe linux-perf-users" in
>> the body of a message to majordomo@vger.kernel.org
>> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>>
> --
> To unsubscribe from this list: send the line "unsubscribe linux-perf-users" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> 

Re: Does Perf support AMD cpu perfectly?

[Chulmin Kim]

Sorry for the posting with no patience.

I've found the patch. (https://patchwork.kernel.org/patch/1722691/)

It's pretty new. (released on 10,nov,2012)

If you have something to mention about this issue,

don't hesitate to contact me.

Thanks.

2012-11-26 오후 10:45, Chulmin Kim 쓴 글:
> Update for this issue.
> 
> I'm investigating the kernel source.
> 
> I found that Perf for AMD seems imperfect in linux 3.0 (my machine's
> linux version).
> 
> For AMD family 15h, there is a commented line as follow.
>   - /* nortbridge counters not yet implemented: */  (line 541,
> perf_event_amd.c )
> 
> 
> According to my further exploration, linux 3.2 does not support it either.
> 
> Does any of you know or have the patch for it?
> 
> 
> 2012-11-26 오후 9:15, Chulmin Kim 쓴 글:
>> Update for this issue.
>>
>> - rc0 and instruction events show the similar result.
>> - still, the count of L3 cache miss event is 1/10 of what i expected.
>> (LLC-load-misses)
>>
>>
>>
>> 2012-11-26 오후 8:24, Chulmin Kim 쓴 글:
>>> Hi all.
>>>
>>> I'm on the way of monitoring AMD machine.
>>>
>>> I wonder Perf is supporting AMD PMU Events perfectly or not.
>>>
>>> It seems not working correctly despite the same command works in intel
>>> machines perfectly.
>>>
>>> command : (perf stat -e rc0 -e instructions ./blah)
>>> result : the value for "rc0" almost equals to that of "instruction" in
>>> Intel machine, while it does not in AMD machine
>>>
>>>
>>> Other events such as cache events seem weird also.
>>>
>>> Is there any material mentioning this issue?
>>>
>>>
>>>
>>> Thanks!
>>>
>>> --
>>> To unsubscribe from this list: send the line "unsubscribe linux-perf-users" in
>>> the body of a message to majordomo@vger.kernel.org
>>> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>>>
>> --
>> To unsubscribe from this list: send the line "unsubscribe linux-perf-users" in
>> the body of a message to majordomo@vger.kernel.org
>> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>>
> --
> To unsubscribe from this list: send the line "unsubscribe linux-perf-users" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> 

Re: Does Perf support AMD cpu perfectly?

[Jiri Olsa]

On Mon, Nov 26, 2012 at 10:45:16PM +0900, Chulmin Kim wrote:
> Update for this issue.
> 
> I'm investigating the kernel source.
> 
> I found that Perf for AMD seems imperfect in linux 3.0 (my machine's
> linux version).
> 
> For AMD family 15h, there is a commented line as follow.
>  - /* nortbridge counters not yet implemented: */  (line 541,
> perf_event_amd.c )

there were some updates for 15h models recently,
you probably need to try latest sources

jirka

Re: Does Perf support AMD cpu perfectly?

[Chulmin Kim]

2012-11-26 오후 11:05, Jiri Olsa 쓴 글:
> On Mon, Nov 26, 2012 at 10:45:16PM +0900, Chulmin Kim wrote:
>> Update for this issue.
>>
>> I'm investigating the kernel source.
>>
>> I found that Perf for AMD seems imperfect in linux 3.0 (my machine's
>> linux version).
>>
>> For AMD family 15h, there is a commented line as follow.
>>   - /* nortbridge counters not yet implemented: */  (line 541,
>> perf_event_amd.c )
>
> there were some updates for 15h models recently,
> you probably need to try latest sources
>

Thanks for your advice.

I installed v3.6.7 which is the latest stable kernel version.

However, the result of Perf shows that the northbridge event is not yet 
supported as shown below.
root@ubuntu:~# /bin/perf stat -a -x ' ' -C 0 -e rc0 -e r4008307e1 sleep 1
3956883 raw 0xc0
<not supported> raw 0x4008307e1:HG

The recent patches for this issue might not have been included yet.

I should add the patchset I mentioned in this thread.
Will report it soon. :)


> jirka
>

Re: Does Perf support AMD cpu perfectly?

[Chulmin Kim]

Update for this issue.

The patchset works correctly.

Now the problem is the number of the events to be monitored.

I will start a new thread for this issue.

Thanks.



2012-11-26 오후 10:53, Chulmin Kim 쓴 글:
> Sorry for the posting with no patience.
> 
> I've found the patch. (https://patchwork.kernel.org/patch/1722691/)
> 
> It's pretty new. (released on 10,nov,2012)
> 
> If you have something to mention about this issue,
> 
> don't hesitate to contact me.
> 
> Thanks.
> 
> 2012-11-26 오후 10:45, Chulmin Kim 쓴 글:
>> Update for this issue.
>>
>> I'm investigating the kernel source.
>>
>> I found that Perf for AMD seems imperfect in linux 3.0 (my machine's
>> linux version).
>>
>> For AMD family 15h, there is a commented line as follow.
>>    - /* nortbridge counters not yet implemented: */  (line 541,
>> perf_event_amd.c )
>>
>>
>> According to my further exploration, linux 3.2 does not support it either.
>>
>> Does any of you know or have the patch for it?
>>
>>
>> 2012-11-26 오후 9:15, Chulmin Kim 쓴 글:
>>> Update for this issue.
>>>
>>> - rc0 and instruction events show the similar result.
>>> - still, the count of L3 cache miss event is 1/10 of what i expected.
>>> (LLC-load-misses)
>>>
>>>
>>>
>>> 2012-11-26 오후 8:24, Chulmin Kim 쓴 글:
>>>> Hi all.
>>>>
>>>> I'm on the way of monitoring AMD machine.
>>>>
>>>> I wonder Perf is supporting AMD PMU Events perfectly or not.
>>>>
>>>> It seems not working correctly despite the same command works in intel
>>>> machines perfectly.
>>>>
>>>> command : (perf stat -e rc0 -e instructions ./blah)
>>>> result : the value for "rc0" almost equals to that of "instruction" in
>>>> Intel machine, while it does not in AMD machine
>>>>
>>>>
>>>> Other events such as cache events seem weird also.
>>>>
>>>> Is there any material mentioning this issue?
>>>>
>>>>
>>>>
>>>> Thanks!
>>>>
>>>> --
>>>> To unsubscribe from this list: send the line "unsubscribe linux-perf-users" in
>>>> the body of a message to majordomo@vger.kernel.org
>>>> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>>>>
>>> --
>>> To unsubscribe from this list: send the line "unsubscribe linux-perf-users" in
>>> the body of a message to majordomo@vger.kernel.org
>>> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>>>
>> --
>> To unsubscribe from this list: send the line "unsubscribe linux-perf-users" in
>> the body of a message to majordomo@vger.kernel.org
>> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>>
> --
> To unsubscribe from this list: send the line "unsubscribe linux-perf-users" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>