linux-kernel - Re: [PATCH] drm/vkms: Add information on how to benchmark

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Date: Tue, 27 Feb 2024 13:52:59 -0300
From: Arthur Grillo <arthurgrillo@...eup.net>
To: Pekka Paalanen <pekka.paalanen@...labora.com>,
 Rodrigo Siqueira <rodrigosiqueiramelo@...il.com>,
 Melissa Wen <melissa.srw@...il.com>, Maíra Canal
 <mairacanal@...eup.net>, Haneen Mohammed <hamohammed.sa@...il.com>,
 Daniel Vetter <daniel@...ll.ch>,
 Maarten Lankhorst <maarten.lankhorst@...ux.intel.com>,
 Maxime Ripard <mripard@...nel.org>, Thomas Zimmermann <tzimmermann@...e.de>,
 David Airlie <airlied@...il.com>, Jonathan Corbet <corbet@....net>,
 dri-devel@...ts.freedesktop.org, linux-doc@...r.kernel.org,
 linux-kernel@...r.kernel.org
Subject: Re: [PATCH] drm/vkms: Add information on how to benchmark



On 27/02/24 12:02, Louis Chauvet wrote:
> Le 27/02/24 - 15:26, Pekka Paalanen a écrit :
>> On Tue, 27 Feb 2024 09:29:58 -0300
>> Arthur Grillo <arthurgrillo@...eup.net> wrote:
>>
>>> On 27/02/24 08:55, Pekka Paalanen wrote:
>>>> On Tue, 27 Feb 2024 08:44:52 -0300
>>>> Arthur Grillo <arthurgrillo@...eup.net> wrote:
>>>>   
>>>>> On 27/02/24 06:19, Pekka Paalanen wrote:  
>>>>>> On Mon, 26 Feb 2024 17:42:11 -0300
>>>>>> Arthur Grillo <arthurgrillo@...eup.net> wrote:
>>>>>>     
>>>>>>> Now that we have a defined benchmark for testing the driver, add
>>>>>>> documentation on how to run it.
>>>>>>>
>>>>>>> Signed-off-by: Arthur Grillo <arthurgrillo@...eup.net>
>>>>>>> ---
>>>>>>>  Documentation/gpu/vkms.rst | 6 ++++++
>>>>>>>  1 file changed, 6 insertions(+)
>>>>>>>
>>>>>>> diff --git a/Documentation/gpu/vkms.rst b/Documentation/gpu/vkms.rst
>>>>>>> index ba04ac7c2167..6d07f79f77ff 100644
>>>>>>> --- a/Documentation/gpu/vkms.rst
>>>>>>> +++ b/Documentation/gpu/vkms.rst
>>>>>>> @@ -89,6 +89,12 @@ You can also run subtests if you do not want to run the entire test::
>>>>>>>    sudo ./build/tests/kms_flip --run-subtest basic-plain-flip --device "sys:/sys/devices/platform/vkms"
>>>>>>>    sudo IGT_DEVICE="sys:/sys/devices/platform/vkms" ./build/tests/kms_flip --run-subtest basic-plain-flip
>>>>>>>  
>>>>>>> +If you are developing features that may affect performance, you can run the kms_fb_stress    
>>>>>>
>>>>>> s/can/must/
>>>>>>     
>>>>>>> +benchmark::    
>>>>>>
>>>>>> before and after, and report the numbers.    
>>>>>
>>>>> Did you mean to write the benchmarks logs here?  
>>>>
>>>> I mean people should be required tell their before and after numbers in
>>>> either commit message (my preference) or in series cover letter (if
>>>> benchmarking commits is not useful).
>>>>
>>>> With the addition of YUV support in VKMS, maybe the benchmark needs to  
>>>
>>> With the upcoming addition, I've sent a patch to arbitrarily change the
>>> formats on the benchmark via command-line options. It's not adding a new
>>> case, but maybe just this could already help.
>>>
>>> https://lore.kernel.org/all/20240226-kms_fb_stress-dev-v1-0-1c14942b1244@riseup.net/
>>
>> In that case you would need to document exactly what command line
>> options to use, and ask people to report the numbers of each test
>> case.
>>
>> That works. Alternatively or additionally, the test cases could be
>> built in to the benchmark, and it just reports numbers for all of them
>> in a single invocation. Then people running the standard benchmark do
>> not need to worry about getting the command line options right, or
>> running multiple cases. And reviewers do not need to ask to re-run with
>> the correct options.
>>
>> I suppose rotations might get added, too.
>>
>> Or maybe you'd provide a script that covers all the standard
>> performance test cases?
> 
> I agree with Pekka, it would be nice to have a simple "bench everything" 
> tool. Like kms_rotation is a test for all rotations, kms_plane for color 
> conversions... kms_fb_test can run a few combinations (rgb+norotation, 
> rgb+yuv, rgb+rotation...) and report a "global result" (this way it's easy 
> to spot a regression not related directly to your changes).
> 
> I don't know the IGT benchmark API, but I think there is a way to create 
> "sub-benchmarks" so you can run a specific benchmark when developping and 
> the whole thing before pushing your series.
> 
> Kind regards,
> Louis Chauvet
> 

Thank you guys for the ideas!

I'm going to work on something along on those lines, having combinations
of formats, rotations and whatever property that might be useful for
stress testing the driver. Maybe query the driver for the supported
properties (e.g. what formats and rotations it supports) and do many/all
possible combinations.

Best Regards,
~Arthur Grillo


>>
>> Thanks,
>> pq
>>
>>>> start printing YUV numbers separately as a new case.
>>>>
>>>>
>>>> Thanks,
>>>> pq
>>>>   
>>>>>  
>>>>>>     
>>>>>>> +
>>>>>>> +  sudo ./build/benchmarks/kms_fb_stress --device "sys:/sys/devices/platform/vkms"
>>>>>>> +  sudo IGT_DEVICE="sys:/sys/devices/platform/vkms" ./build/benchmarks/kms_fb_stress    
>>>>>>
>>>>>> Do people need to run both commands?    
>>>>>
>>>>> No, they don't, just two options.
>>>>>
>>>>> Best Regards,
>>>>> ~Arthur Grillo
>>>>>  
>>>>>>
>>>>>> Anyway, a good idea.
>>>>>>
>>>>>> Acked-by: Pekka Paalanen <pekka.paalanen@...labora.com>
>>>>>>
>>>>>>
>>>>>> Thanks,
>>>>>> pq
>>>>>>     
>>>>>>> +
>>>>>>>  TODO
>>>>>>>  ====
>>>>>>>  
>>>>>>>
>>>>>>> ---
>>>>>>> base-commit: eeb8e8d9f124f279e80ae679f4ba6e822ce4f95f
>>>>>>> change-id: 20240226-bench-vkms-5b8b7aab255e
>>>>>>>
>>>>>>> Best regards,    
>>>>>>     
>>>>   
>>
> 
> 
>