On 12/4/2024 10:00, Alex Deucher wrote:
On Wed, Dec 4, 2024 at 10:56 AM Lazar, Lijo <lijo.lazar@xxxxxxx> wrote:
On 12/4/2024 7:51 PM, Alex Deucher wrote:
On Wed, Dec 4, 2024 at 12:47 AM Lazar, Lijo <lijo.lazar@xxxxxxx> wrote:
On 12/4/2024 10:44 AM, Mario Limonciello wrote:
+enum amdgpu_ucode_required {
+ AMDGPU_UCODE_NOT_REQUIRED,
+ AMDGPU_UCODE_REQUIRED,
Couldn't this be handled in another API instead of having to flag every
load? By default, every ucode is required and if optional may be skipped
with amdgpu_ucode_request_optional() API?
I guess this would be a smaller patch, but 6 eggs one hand, half dozen
in the other?
I thought only ISP and gpu_info (no longer there for newer SOCs) fall
into the optional ones so far. The usage is rare, similar to the
nowarn() API usage.
Also, as far as I know, the cap microcode is a must whenever used. That
is not optional.
The cap firmware is definitely optional. Some customers use it, some don't.
I thought optional is something that can be ignored even if FW is not
found and then driver load proceeds.
What is the expected driver action if we classify cap firmware as
optional and then it fails on a customer system that expects it?
I guess if the customer expects it, they can make sure it's there.
I'm not sure how you can have both without it being optional. For
customers that don't use it, requiring it would break them if it
wasn't present.
Besides those, there were a few others I marked as optional too. We
have a few cases that we do a "fallback" firmware. If that one isn't
found try to get another filename.
I marked the first one as optional and the fallback one as required.
So effectively if both are missing they'll get exactly one message in
the logs about missing firmware.
Alex
Thanks,
Lijo
Alex
Thanks,
Lijo
Alex - what's your take?