2024-06-13 15:25:08

by Thomas Weißschuh

Subject: [PATCH v2] leds: triggers: flush pending brightness before activating trigger

The race fixed in timer_trig_activate() between a blocking
set_brightness() call and trigger->activate() can affect any trigger.
So move the call to flush_work() into led_trigger_set() where it can
avoid the race for all triggers.

Fixes: 0db37915d912 ("leds: avoid races with workqueue")
Fixes: 8c0f693c6eff ("leds: avoid flush_work in atomic context")
Cc: [email protected]
Tested-by: Dustin L. Howett <[email protected]>
Signed-off-by: Thomas Weißschuh <[email protected]>
---
Changes in v2:
- Cc stable
- Add Dustin's Tested-by
- Rebase against led/for-leds-next
- Always execute flush_work(), similar to synchronize_rcu()
- Link to v1: https://lore.kernel.org/r/[email protected]
---

Commit 7abae7a11fc9 ("leds: trigger: Call synchronize_rcu() before calling trig->activate()")
alone also solves the issue that the LED stays completely off after
enabling the trigger.
But there is a recognizable timeframe where it is disabled in between.

Also from a correctness perspective, this seems like a coincidence and
the call to flush_work() is the correct fix for that race.
---
drivers/leds/led-triggers.c | 6 ++++++
drivers/leds/trigger/ledtrig-timer.c | 5 -----
2 files changed, 6 insertions(+), 5 deletions(-)

diff --git a/drivers/leds/led-triggers.c b/drivers/leds/led-triggers.c
index 59deadb86335..78eb20093b2c 100644
--- a/drivers/leds/led-triggers.c
+++ b/drivers/leds/led-triggers.c
@@ -201,6 +201,12 @@ int led_trigger_set(struct led_classdev *led_cdev, struct led_trigger *trig)
*/
synchronize_rcu();

+ /*
+ * If "set brightness to 0" is pending in workqueue,
+ * we don't want that to be reordered after ->activate()
+ */
+ flush_work(&led_cdev->set_brightness_work);
+
ret = 0;
if (trig->activate)
ret = trig->activate(led_cdev);
diff --git a/drivers/leds/trigger/ledtrig-timer.c b/drivers/leds/trigger/ledtrig-timer.c
index b4688d1d9d2b..1d213c999d40 100644
--- a/drivers/leds/trigger/ledtrig-timer.c
+++ b/drivers/leds/trigger/ledtrig-timer.c
@@ -110,11 +110,6 @@ static int timer_trig_activate(struct led_classdev *led_cdev)
led_cdev->flags &= ~LED_INIT_DEFAULT_TRIGGER;
}

- /*
- * If "set brightness to 0" is pending in workqueue, we don't
- * want that to be reordered after blink_set()
- */
- flush_work(&led_cdev->set_brightness_work);
led_blink_set(led_cdev, &led_cdev->blink_delay_on,
&led_cdev->blink_delay_off);


---
base-commit: 005408af25d5550e1bd22a18bf371651969c17ee
change-id: 20240603-led-trigger-flush-3bef303eb902

Best regards,
--
Thomas Weißschuh <[email protected]>

2024-06-13 16:49:18

by Lee Jones

Subject: Re: (subset) [PATCH v2] leds: triggers: flush pending brightness before activating trigger

On Thu, 13 Jun 2024 17:24:51 +0200, Thomas Weißschuh wrote:
> The race fixed in timer_trig_activate() between a blocking
> set_brightness() call and trigger->activate() can affect any trigger.
> So move the call to flush_work() into led_trigger_set() where it can
> avoid the race for all triggers.
>
>

Applied, thanks!

[1/1] leds: triggers: flush pending brightness before activating trigger
commit: ad5851de0eb12e088a570092242e421cd00691c5

--
Lee Jones [李琼斯]