Re: [RFC 2/2] drm/amdgpu: Drop null SDMA instance checks from SDMA vfuncs

["Timur Kristóf" <timur.kristof@gmail.com>]

On 2026. június 22., hétfő 16:36:49 közép-európai nyári idő Tvrtko Ursulin 
wrote:
> Commit message TBD - Can these be NULL?

I've looked it up, burst_nop was added 11 years ago in commit 18111de0dfc3.
I have the following thoughts on this topic:

1. For newer SDMA IP block versions, I think burst_nop would be supported on 
all firmware versions and the check shouldn't be necessary at all. (Eg. I don't 
think there was ever any SDMA v5.x+ FW without this feature.) We should find 
out which IP block versions exactly are the ones that always supported it and 
we should just remove the check from those.

2. I don't understand why the burst_nop flag is per-instance. Is it really 
possible that not all SDMA instances support it? We should find out and if 
possible move the flag to amdgpu_sdma instead of amdgpu_sdma_instance.

3. To actually answer the question in the commit message, I don't think they 
can be NULL, unless an invalid (not SDMA) amdgpu_ring pointer is passed. I 
would rather just assert that the ring type is correct.

Best regards,
Timur