On 08/09/2026 09:14, Timur Kristóf wrote:
On 2026. szeptember 8., kedd 9:41:48 közép-európai nyári idő Tvrtko Ursulin
wrote:
On 07/09/2026 21:33, Timur Kristóf wrote:
This prepares the code for implementing recovery for SDMA 4.0.
Reorganize the SDMA 4.0 code so that the functions that are
responsible for managing the SDMA engines take an instance ID
as an argument.
This makes it possible to manage the SDMA instances
independently of each other, which will enable us to
also use these functions to implement resetting and
recovering them independently.
Radom patch to drop a comment on the whole series. It looks clean and
reasonable to me. Once the pre-requisite series is reviewed I can do a
more detail pass and slap an r-b to it.
Thanks! If we agree this is the right approach, I would like to do the same
refactor for all the newer SDMA blocks as well for the sake of consistency.
This refactor is necessary because I want to reuse the same functions for
implementing per-instance SDMA soft reset (as opposed to duplicating a lot of
code which seems to me to be more error-prone).
One discussion point could be is do we maybe want the prototype for all
to be "struct amdgpu_sdma_instance *" instead of "struct amdgpu_device,
const u32 instance_id"?
I don't want to complicate this with the amdgpu_sdma_instance struct, because
these functions only need the instance ID, and only for the purpose of knowing
the register offset.
Fine by me to keep the signatures as you have them.
Regards,
Tvrtko
Could I also interest you in a related cleanup:
https://lore.kernel.org/amd-gfx/20260622143649.54632-1-tvrtko.ursulin@igalia
.com/#t
?
I've taken a quick look, I'd suggest a different approach to it.
I will reply on your commits.
Thanks & best regards,
Timur
