Hey Alex,
Agree, we are moving it above, Christian also had the same feedback.
- Shashank
On 2/4/2022 7:44 PM, Deucher, Alexander wrote:
[Public]
Seems like this functionality should be moved up into the callers.
Maybe add new IP callbacks (dump_reset_registers()) so that each IP can
specify what registers are relevant for a reset debugging and then we
can walk the IP list and call the callback before we call the asic_reset
callbacks.
Alex
------------------------------------------------------------------------
*From:* amd-gfx <amd-gfx-bounces@xxxxxxxxxxxxxxxxxxxxx> on behalf of
Deucher, Alexander <Alexander.Deucher@xxxxxxx>
*Sent:* Friday, February 4, 2022 1:41 PM
*To:* Sharma, Shashank <Shashank.Sharma@xxxxxxx>; Lazar, Lijo
<Lijo.Lazar@xxxxxxx>; amd-gfx@xxxxxxxxxxxxxxxxxxxxx
<amd-gfx@xxxxxxxxxxxxxxxxxxxxx>
*Cc:* Somalapuram, Amaranath <Amaranath.Somalapuram@xxxxxxx>; Koenig,
Christian <Christian.Koenig@xxxxxxx>
*Subject:* Re: [PATCH 4/4] drm/amdgpu/nv: add navi GPU reset handler
[Public]
[Public]
In the suspend and hibernate cases, we don't care. In most cases the
power rail will be cut once the system enters suspend so it doesn't
really matter. That's why we call the asic reset callback directly
rather than going through the whole recovery process. The device is
already quiescent at this point we just want to make sure the device is
in a known state when we come out of suspend (in case suspend overall
fails).
Alex
------------------------------------------------------------------------
*From:* Sharma, Shashank <Shashank.Sharma@xxxxxxx>
*Sent:* Friday, February 4, 2022 12:22 PM
*To:* Lazar, Lijo <Lijo.Lazar@xxxxxxx>; amd-gfx@xxxxxxxxxxxxxxxxxxxxx
<amd-gfx@xxxxxxxxxxxxxxxxxxxxx>
*Cc:* Deucher, Alexander <Alexander.Deucher@xxxxxxx>; Somalapuram,
Amaranath <Amaranath.Somalapuram@xxxxxxx>; Koenig, Christian
<Christian.Koenig@xxxxxxx>
*Subject:* Re: [PATCH 4/4] drm/amdgpu/nv: add navi GPU reset handler
On 2/4/2022 6:20 PM, Lazar, Lijo wrote:
[AMD Official Use Only]
One more thing
In suspend-reset case, won't this cause to schedule a work item on suspend? I don't know if that is a good idea, ideally we would like to clean up all work items before going to suspend.
Thanks,
Lijo
Again, this opens scope for discussion. What if there is a GPU error
during suspend-reset, which is very probable case.
- Shashank
-----Original Message-----
From: Sharma, Shashank <Shashank.Sharma@xxxxxxx>
Sent: Friday, February 4, 2022 10:47 PM
To: Lazar, Lijo <Lijo.Lazar@xxxxxxx>; amd-gfx@xxxxxxxxxxxxxxxxxxxxx
Cc: Deucher, Alexander <Alexander.Deucher@xxxxxxx>; Somalapuram, Amaranath <Amaranath.Somalapuram@xxxxxxx>; Koenig, Christian <Christian.Koenig@xxxxxxx>
Subject: Re: [PATCH 4/4] drm/amdgpu/nv: add navi GPU reset handler
On 2/4/2022 6:11 PM, Lazar, Lijo wrote:
BTW, since this is already providing a set of values it would be useful to provide one more field as the reset reason - RAS error recovery, GPU hung recovery or something else.
Adding this additional parameter instead of blocking something in kernel, seems like a better idea. The app can filter out and read what it is interested into.
- Shashank
