Re: [PATCH 1/1] drm/i915/dmc: fix assert_dmc_loaded WARN during async firmware load

[James Xiong <james.xiong@intel.com>]

On 5/7/26 07:40, Gustavo Sousa wrote:
> James Xiong <james.xiong@intel.com> writes:
>
>> During driver probe, DMC firmware is loaded asynchronously via a
>> workqueue. There is a race between parse_dmc_fw() setting the payload
>> pointer (making has_dmc_id_fw() return true) and intel_dmc_load_program()
>> writing the firmware to hardware registers. If the probe thread calls
>> intel_dmc_enable_pipe() -> assert_dmc_loaded() in this window, it sees
>> parsed payload but stale HW registers, triggering a ~20% intermittent
>> WARNING on ADL-N warm boot.
>>
>> Fix this by adding a 'loaded' flag to struct intel_dmc, set with
>> WRITE_ONCE() after intel_dmc_load_program() completes. Check it with
>> READ_ONCE() in intel_dmc_enable_pipe() and intel_dmc_disable_pipe()
>> as an additional guard before has_dmc_id_fw().
> Making intel_dmc_enable_pipe() bail if the DMC is not yet loaded doesn't
> look like the correct solution here.
>
> Does the warning come from intel_modeset_setup_hw_state()'s call stack?
> If so, I wonder if adding a call to intel_dmc_wait_fw_load() before
> iterating the crtcs would be a saner approach.
>
> --
> Gustavo Sousa

Yes, the warning comes from intel_modeset_setup_hw_state() →
intel_dmc_enable_pipe() → assert_dmc_loaded().

You're right, adding intel_dmc_wait_fw_load() before the CRTC loop
is a much saner approach. I've updated the patch to do exactly that
— the loaded flag machinery is dropped entirely, and instead
intel_modeset_setup_hw_state() now calls intel_dmc_wait_fw_load(display)
before for_each_intel_crtc(), following the same pattern as intel_flipq_init().

Will send v2

James