On Wed, 11 Nov 2026 09:28:16 +0800, Wayne Lin wrote:
> +     /* The poll re-queues itself, so stop it before the wq and dc go away. 
> */
> +     if (adev->dm.hdmi_frl_status_polling_wq) {
> +             
> cancel_delayed_work_sync(&adev->dm.hdmi_frl_status_polling_work);

This is the same change as patch 2/3 of a series I posted on Nov 10:
https://lore.kernel.org/all/20261110-frl-polling-wq-teardown-v1-0-0ff028d5c...@protonmail.com/

I'm fine with either version going in, but checking the wq pointer
isn't enough on its own. amdgpu_dm_init() creates the wq before
amdgpu_dm_initialize_drm_device(), and hdmi_frl_status_polling_work is
only initialized later, in amdgpu_dm_connector_init_helper(). If init
fails between those two points (the dmub_notify or delayed_hpd_wq
allocation, for example), the error path calls amdgpu_dm_fini(), which
runs cancel_delayed_work_sync() on a work item that was never
initialized. __flush_work() then hits WARN_ON(!work->func).

Patch 1/3 of my series fixes that by moving INIT_DELAYED_WORK() into
amdgpu_dm_init(), before the wq is created. Patch 3/3 (don't queue the
work on a NULL wq when the wq allocation failed) isn't covered by this
batch either.

If this patch is applied, I'll rebase 1/3 and 3/3 on top of it and drop
2/3 in v2. Should I wait for this batch to land, or send v2 against
amd-staging-drm-next now?

One small request: my series was on the list the day before this batch.
Would you consider taking my 2/3 in place of this patch, or, if you'd
rather keep this version, adding a Reported-by: and a Link: to my
posting? This is my first kernel series, so the credit would mean a lot
to me.

Thanks,
Eren

Reply via email to