The function panthor_fw_unplug() will free the FW memory sections.
The problem is that there could still be pending FW events which are yet
not handled at this point. process_fw_events_work() can in this case try
to access said freed memory.

Simply call disable_work_sync() to both drain and prevent future
invocation of process_fw_events_work().

Signed-off-by: Ketil Johnsen <ketil.johnsen@arm.com>
Fixes: de85488138247 ("drm/panthor: Add the scheduler logical block")
---
v2:
- Followed Boris's advice and handle the race purely within the
  scheduler block (by adding a destroyed state)

v3:
- New approach, one single call to disable_work_sync()
---
 drivers/gpu/drm/panthor/panthor_sched.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/drivers/gpu/drm/panthor/panthor_sched.c b/drivers/gpu/drm/panthor/panthor_sched.c
index 0cc9055f4ee52..b7595beaa0205 100644
--- a/drivers/gpu/drm/panthor/panthor_sched.c
+++ b/drivers/gpu/drm/panthor/panthor_sched.c
@@ -3880,6 +3880,7 @@ void panthor_sched_unplug(struct panthor_device *ptdev)
 	struct panthor_scheduler *sched = ptdev->scheduler;
 
 	cancel_delayed_work_sync(&sched->tick_work);
+	disable_work_sync(&sched->fw_events_work);
 
 	mutex_lock(&sched->lock);
 	if (sched->pm.has_ref) {
-- 
2.47.2

On Mon, Oct 27, 2025 at 03:02:15PM +0100, Ketil Johnsen wrote:
> The function panthor_fw_unplug() will free the FW memory sections.
> The problem is that there could still be pending FW events which are yet
> not handled at this point. process_fw_events_work() can in this case try
> to access said freed memory.
> 
> Simply call disable_work_sync() to both drain and prevent future
> invocation of process_fw_events_work().
> 
> Signed-off-by: Ketil Johnsen <ketil.johnsen@arm.com>
> Fixes: de85488138247 ("drm/panthor: Add the scheduler logical block")

Pushed to drm-misc-next.

Best regards,
Liviu

> ---
> v2:
> - Followed Boris's advice and handle the race purely within the
>   scheduler block (by adding a destroyed state)
> 
> v3:
> - New approach, one single call to disable_work_sync()
> ---
>  drivers/gpu/drm/panthor/panthor_sched.c | 1 +
>  1 file changed, 1 insertion(+)
> 
> diff --git a/drivers/gpu/drm/panthor/panthor_sched.c b/drivers/gpu/drm/panthor/panthor_sched.c
> index 0cc9055f4ee52..b7595beaa0205 100644
> --- a/drivers/gpu/drm/panthor/panthor_sched.c
> +++ b/drivers/gpu/drm/panthor/panthor_sched.c
> @@ -3880,6 +3880,7 @@ void panthor_sched_unplug(struct panthor_device *ptdev)
>  	struct panthor_scheduler *sched = ptdev->scheduler;
>  
>  	cancel_delayed_work_sync(&sched->tick_work);
> +	disable_work_sync(&sched->fw_events_work);
>  
>  	mutex_lock(&sched->lock);
>  	if (sched->pm.has_ref) {
> -- 
> 2.47.2
> 

-- 
====================
| I would like to |
| fix the world,  |
| but they're not |
| giving me the   |
 \ source code!  /
  ---------------
    ¯\_(ツ)_/¯

On Mon, Oct 27, 2025 at 03:02:15PM +0100, Ketil Johnsen wrote:
> The function panthor_fw_unplug() will free the FW memory sections.
> The problem is that there could still be pending FW events which are yet
> not handled at this point. process_fw_events_work() can in this case try
> to access said freed memory.
> 
> Simply call disable_work_sync() to both drain and prevent future
> invocation of process_fw_events_work().
> 
> Signed-off-by: Ketil Johnsen <ketil.johnsen@arm.com>
> Fixes: de85488138247 ("drm/panthor: Add the scheduler logical block")

Reviewed-by: Liviu Dudau <liviu.dudau@arm.com>

Best regards,
Liviu

> ---
> v2:
> - Followed Boris's advice and handle the race purely within the
>   scheduler block (by adding a destroyed state)
> 
> v3:
> - New approach, one single call to disable_work_sync()
> ---
>  drivers/gpu/drm/panthor/panthor_sched.c | 1 +
>  1 file changed, 1 insertion(+)
> 
> diff --git a/drivers/gpu/drm/panthor/panthor_sched.c b/drivers/gpu/drm/panthor/panthor_sched.c
> index 0cc9055f4ee52..b7595beaa0205 100644
> --- a/drivers/gpu/drm/panthor/panthor_sched.c
> +++ b/drivers/gpu/drm/panthor/panthor_sched.c
> @@ -3880,6 +3880,7 @@ void panthor_sched_unplug(struct panthor_device *ptdev)
>  	struct panthor_scheduler *sched = ptdev->scheduler;
>  
>  	cancel_delayed_work_sync(&sched->tick_work);
> +	disable_work_sync(&sched->fw_events_work);
>  
>  	mutex_lock(&sched->lock);
>  	if (sched->pm.has_ref) {
> -- 
> 2.47.2
> 

-- 
====================
| I would like to |
| fix the world,  |
| but they're not |
| giving me the   |
 \ source code!  /
  ---------------
    ¯\_(ツ)_/¯

On Mon, 27 Oct 2025 15:02:15 +0100
Ketil Johnsen <ketil.johnsen@arm.com> wrote:

> The function panthor_fw_unplug() will free the FW memory sections.
> The problem is that there could still be pending FW events which are yet
> not handled at this point. process_fw_events_work() can in this case try
> to access said freed memory.
> 
> Simply call disable_work_sync() to both drain and prevent future
> invocation of process_fw_events_work().
> 
> Signed-off-by: Ketil Johnsen <ketil.johnsen@arm.com>
> Fixes: de85488138247 ("drm/panthor: Add the scheduler logical block")

Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>

Would you mind sending a patch on top of this one turning all
cancel_[delayed_]work_sync() happening in the unplug path int
disable_[delayed_]work_sync(), so we're sure we won't face other "work
in queued after being cancelled" issues in the future.

> ---
> v2:
> - Followed Boris's advice and handle the race purely within the
>   scheduler block (by adding a destroyed state)
> 
> v3:
> - New approach, one single call to disable_work_sync()
> ---
>  drivers/gpu/drm/panthor/panthor_sched.c | 1 +
>  1 file changed, 1 insertion(+)
> 
> diff --git a/drivers/gpu/drm/panthor/panthor_sched.c b/drivers/gpu/drm/panthor/panthor_sched.c
> index 0cc9055f4ee52..b7595beaa0205 100644
> --- a/drivers/gpu/drm/panthor/panthor_sched.c
> +++ b/drivers/gpu/drm/panthor/panthor_sched.c
> @@ -3880,6 +3880,7 @@ void panthor_sched_unplug(struct panthor_device *ptdev)
>  	struct panthor_scheduler *sched = ptdev->scheduler;
>  
>  	cancel_delayed_work_sync(&sched->tick_work);
> +	disable_work_sync(&sched->fw_events_work);
>  
>  	mutex_lock(&sched->lock);
>  	if (sched->pm.has_ref) {