On 2/9/26 10:51, SHANMUGAM, SRINIVASAN wrote:
> [AMD Official Use Only - AMD Internal Distribution Only]
>
>> -----Original Message-----
>> From: Koenig, Christian <[email protected]>
>> Sent: Monday, February 9, 2026 3:11 PM
>> To: SHANMUGAM, SRINIVASAN <[email protected]>;
>> Deucher, Alexander <[email protected]>
>> Cc: [email protected]
>> Subject: Re: [PATCH] drm/amdgpu: Don't propagate errors from
>> amdgpu_vm_flush()
>>
>> On 2/8/26 07:27, Srinivasan Shanmugam wrote:
>>> amdgpu_vm_flush() only returns an error when amdgpu_fence_emit() fails.
>>
>> You need to take a step further amdgpu_fence_emit() can't fail either.
>>
>> The fallback wait inside that function blocks forever until the fence
>> signals and that
>> should never happen in the job submission path in the first place.
>
> Thanks Christian, understood.
>
> For amdgpu_fence_emit(), I will not add any recovery logic.
IIRC we only used the recovery logic for direct submission for low level
benchmark testing.
I think we completely dropped that, but I'm not 100% sure.
For now potentially just remove the error handling here:
r = dma_fence_wait(old, false);
dma_fence_put(old);
if (r)
return r;
Background is that dma_fence_wait() can only return an error if the second
parameter is true.
In the long term we could move this waiting completely out of the emit
function, but that is clearly something separate.
Regards,
christian.
> I will keep only a WARN_ON_ONCE() and return early if it ever triggers.
>
> Please let me know if this approach is OK, and if you would like this as a
> follow-up patch.
>
> Regards,
> Srini
>
>>
>> Regards,
>> Christian.
