[PATCH] drm/i915/display: Don't drop the async-put delay on self-requeue

[PATCH] drm/i915/display: Don't drop the async-put delay on self-requeue

[Wentao Guan]

cancel_async_put_work() also clears async_put_next_delay, so moving it
ahead of the requeue in intel_display_power_put_async_work() makes the
requeue always use a zero delay. That discards the delay accumulated by
concurrent async puts, including the 17 ms DC-state anti-toggle delay
requested for POWER_DOMAIN_DC_OFF.

Use cancel_delayed_work() directly to cancel the stale instance without
clearing the delay, and clear the delay only after the requeue has
consumed it.

Fixes: 30ca6365bb42 ("drm/i915: Fix assert on pending async-put power domain work when it requeues itself")
Cc: <stable@vger.kernel.org> # v6.11+
Signed-off-by: Wentao Guan <guanwentao@uniontech.com>
---
 drivers/gpu/drm/i915/display/intel_display_power.c | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)

diff --git a/drivers/gpu/drm/i915/display/intel_display_power.c b/drivers/gpu/drm/i915/display/intel_display_power.c
index b000ec001ccfc..7da6eda46e9e0 100644
--- a/drivers/gpu/drm/i915/display/intel_display_power.c
+++ b/drivers/gpu/drm/i915/display/intel_display_power.c
@@ -873,7 +873,7 @@ intel_display_power_put_async_work(struct work_struct *work)
 	 * Cancel the work that got queued after this one got dequeued,
 	 * since here we released the corresponding async-put reference.
 	 */
-	cancel_async_put_work(power_domains, false);
+	cancel_delayed_work(&power_domains->async_put_work);
 
 	/* Requeue the work if more domains were async put meanwhile. */
 	if (!bitmap_empty(power_domains->async_put_domains[1].bits, POWER_DOMAIN_NUM)) {
@@ -885,9 +885,10 @@ intel_display_power_put_async_work(struct work_struct *work)
 		queue_async_put_domains_work(power_domains,
 					     fetch_and_zero(&new_work_wakeref),
 					     power_domains->async_put_next_delay);
-		power_domains->async_put_next_delay = 0;
 	}
 
+	power_domains->async_put_next_delay = 0;
+
 out_verify:
 	verify_async_put_domains_state(power_domains);
 
-- 
2.47.3

Re: [PATCH] drm/i915/display: Don't drop the async-put delay on self-requeue

[Jani Nikula]

On Wed, 04 Nov 2026, Wentao Guan <guanwentao@uniontech.com> wrote:
> cancel_async_put_work() also clears async_put_next_delay, so moving it
> ahead of the requeue in intel_display_power_put_async_work() makes the
> requeue always use a zero delay. That discards the delay accumulated by
> concurrent async puts, including the 17 ms DC-state anti-toggle delay
> requested for POWER_DOMAIN_DC_OFF.
>
> Use cancel_delayed_work() directly to cancel the stale instance without
> clearing the delay, and clear the delay only after the requeue has
> consumed it.
>
> Fixes: 30ca6365bb42 ("drm/i915: Fix assert on pending async-put power domain work when it requeues itself")
> Cc: <stable@vger.kernel.org> # v6.11+
> Signed-off-by: Wentao Guan <guanwentao@uniontech.com>

How did you find this?

Is there a bug report about this?

How did you test this?


BR,
Jani.

> ---
>  drivers/gpu/drm/i915/display/intel_display_power.c | 5 +++--
>  1 file changed, 3 insertions(+), 2 deletions(-)
>
> diff --git a/drivers/gpu/drm/i915/display/intel_display_power.c b/drivers/gpu/drm/i915/display/intel_display_power.c
> index b000ec001ccfc..7da6eda46e9e0 100644
> --- a/drivers/gpu/drm/i915/display/intel_display_power.c
> +++ b/drivers/gpu/drm/i915/display/intel_display_power.c
> @@ -873,7 +873,7 @@ intel_display_power_put_async_work(struct work_struct *work)
>  	 * Cancel the work that got queued after this one got dequeued,
>  	 * since here we released the corresponding async-put reference.
>  	 */
> -	cancel_async_put_work(power_domains, false);
> +	cancel_delayed_work(&power_domains->async_put_work);
>  
>  	/* Requeue the work if more domains were async put meanwhile. */
>  	if (!bitmap_empty(power_domains->async_put_domains[1].bits, POWER_DOMAIN_NUM)) {
> @@ -885,9 +885,10 @@ intel_display_power_put_async_work(struct work_struct *work)
>  		queue_async_put_domains_work(power_domains,
>  					     fetch_and_zero(&new_work_wakeref),
>  					     power_domains->async_put_next_delay);
> -		power_domains->async_put_next_delay = 0;
>  	}
>  
> +	power_domains->async_put_next_delay = 0;
> +
>  out_verify:
>  	verify_async_put_domains_state(power_domains);

-- 
Jani Nikula, Intel

Re: [PATCH] drm/i915/display: Don't drop the async-put delay on self-requeue

[Wentao Guan]

Hi,

> On Wed, 04 Nov 2026, Wentao Guan <guanwentao@uniontech.com> wrote:
> > cancel_async_put_work() also clears async_put_next_delay, so moving it
> > ahead of the requeue in intel_display_power_put_async_work() makes the
> > requeue always use a zero delay. That discards the delay accumulated by
> > concurrent async puts, including the 17 ms DC-state anti-toggle delay
> > requested for POWER_DOMAIN_DC_OFF.
> >
> > Use cancel_delayed_work() directly to cancel the stale instance without
> > clearing the delay, and clear the delay only after the requeue has
> > consumed it.
> >
> > Fixes: 30ca6365bb42 ("drm/i915: Fix assert on pending async-put power domain work when it requeues itself")
> > Cc: <stable@vger.kernel.org> # v6.11+
> > Signed-off-by: Wentao Guan <guanwentao@uniontech.com>
> 
> How did you find this?
During a stable bugfix backport work, found by github copilot max(gpt-6.1-sol[ReasoningEffort=xhigh]).

> Is there a bug report about this?
Just see: https://github.com/deepin-community/kernel/pull/2239#discussion_r4463718883

> How did you test this?
Review again, build again, integation by anothor one.
It is not the original bug we met, the bug sovled by backport many i915 update commits as them in mainline.

BRs
Wentao Guan