[PATCH] drm/i915/rc6: Disable RPG during workload execution

[Nilawar, Badal]

On 24-10-2024 23:04, Rodrigo Vivi wrote:
> On Thu, Oct 24, 2024 at 10:28:38AM -0400, Gupta, Anshuman wrote:
>>
>>
>>> -----Original Message-----
>>> From: Nilawar, Badal <badal.nilawar at intel.com>
>>> Sent: Wednesday, October 23, 2024 9:42 PM
>>> To: Vivi, Rodrigo <rodrigo.vivi at intel.com>
>>> Cc: Andi Shyti <andi.shyti at linux.intel.com>; intel-gfx at lists.freedesktop.org;
>>> Gupta, Anshuman <anshuman.gupta at intel.com>;
>>> chris.p.wilson at linux.intel.com
>>> Subject: Re: [PATCH] drm/i915/rc6: Disable RPG during workload execution
>>>
>>>
>>>
>>> On 23-10-2024 21:31, Nilawar, Badal wrote:
>>>>
>>>>
>>>> On 23-10-2024 20:18, Rodrigo Vivi wrote:
>>>>> On Wed, Oct 23, 2024 at 11:03:57AM +0530, Nilawar, Badal wrote:
>>>>>>
>>>>>>
>>>>>> On 22-10-2024 22:39, Rodrigo Vivi wrote:
>>>>>>> On Tue, Oct 22, 2024 at 03:28:43PM +0200, Andi Shyti wrote:
>>>>>>>> Hi Badal,
>>>>>>>>
>>>>>>>> On Tue, Oct 22, 2024 at 06:52:26PM +0530, Badal Nilawar wrote:
>>>>>>>>> Encountering forcewake errors related to render power gating;
>>>>>>>>
>>>>>>>> Can you please expand your explanation here?
>>>>>>>
>>>>>>> yeap. More explanation please. All platforms? really?
>>>>>>
>>>>>> We are seeing Render forcewake timeouts on ADLP, ADLM, ADLN, TWL,
>>>>>> DG1, rpl.
>>>>>
>>>>> Is this a regression? or a new issue?
>>>>
>>>> This is old issue, first reported year back.
>>>>
>>>>>
>>>>> Is this happening with Xe on these platforms? or i915 only?
>>>>
>>>> i915 only. This is not reported on Xe kmd.
>>>>
>>>>>
>>>>>> Issue disappears after disabling RPG. Instead of fully disabling RPG
>>>>>> I am disabling it during active submissions i.e. during unpark.
>>>>>> For MTL and ARL RPG is already disabled permanently.
>> IMO this patch should be extended for MTL and ARL as well.
>> Don't disable the RPG completely, only disable it during workload submission.
>> That should save power on both MTL and ARL platforms, and right thing to do.
>> And patch should add the Fixes tag accordingly to the commit which disables the RPG on MTL.
>> @Vivi, Rodrigo what is your opinion on above ? it seems both MTL and ADL issues are same signature.
>> MTL issue got disappear as RPG is disabled globally but that will burn power.
>> If window does not have this issue then it is always difficult to get proper SV support.
>> But OS like window may not catch these kind of issues as they don't reload the their graphics driver like
>> Linux reload module multiple times during selftest execution. Even chrome-os does not do that.
>> Not a real world use case.
> 
> So we should change the tests without impacting the real world use case.
> 
> Let's try to grab the forcewake_all when using these self-tests, except the rc6 one.

Sure, I will try forcewake_all on affected tests first.

Thanks,
Badal

> 
>> Thanks,
>> Anshuman.
>>>>>
>>>>> uhm. Interesting. Why that is disabled on these platforms?
>>>>
>>>>   From commit log its temporary wa to avoid fw timeouts.
>>>>
>>>> perhaps we should be
>>>>> doing the same for all GuC enabled platforms?
>>>>
>>>> I think so as temporary Wa.
>>>
>>> Correction, DG1 we are not seeing this. I think we can go with platform check.
>>>
>>> Regards,
>>> Badal
>>