Re: [PATCH] PM: runtime: Fix check_flush_dependency warning from pm_runtime_remove

[Rafael J. Wysocki (Intel)]

On Wed, Nov 4, 2026 at 1:05 PM Zhou Minqiang <zhouminqiang2@xxxxxxxxxx> wrote:
>
> When a WQ_MEM_RECLAIM workqueue (e.g. iscsi_ctrl) triggers device removal
> via device_del -> pm_runtime_remove, the unconditional flush_work inside
> pm_runtime_remove can trigger a check_flush_dependency warning:
>
>   CPU1: pm_wq worker                 CPU2: iscsi_ctrl wq
>   --------                           --------
>
>   pm_runtime_work(starget)           pm_runtime_remove(starget)
>     spin_lock_irq(&starget->lock)      __pm_runtime_disable(starget, false)
>     request_pending = false              spin_lock_irq(&starget->lock)
>                                          (blocked, CPU1 holds the lock)
>
>     rpm_suspend(starget, RPM_NOWAIT)
>       status = RPM_SUSPENDING
>       rpm_callback(cb, starget)
>         spin_unlock_irq(&starget->lock)
>                                          (acquires lock)
>                                          disable_depth 0→1
>                                          __pm_runtime_barrier():
>                                            request_pending == false
>                                              → skip cancel_work_sync
>                                            status == RPM_SUSPENDING
>                                              → enter wait_queue, sleep
>                                            spin_unlock_irq(&starget->lock)
>         cb(starget) → 0
>         spin_lock_irq(&starget->lock)
>       status = RPM_SUSPENDED
>       wake_up_all(&wait_queue)
>                                            (woken up, waiting for lock)
>       spin_unlock(&starget->lock)          spin_lock_irq(&starget->lock)
>                                              status == RPM_SUSPENDED
>       spin_lock(&parent->lock)                → exit wait_queue
>       rpm_idle(parent, RPM_ASYNC)        barrier returns
>       spin_unlock(&parent->lock)       flush_work(&starget->work)
>       spin_lock(&starget->lock)          __flush_work(from_cancel=false)
>       spin_unlock_irq(&starget->lock)      check_flush_dependency()
>                                             ← WARNING!
>
> Replace flush_work with cancel_work_sync in pm_runtime_remove.
> cancel_work_sync internally calls __flush_work with from_cancel=true,
> which causes check_flush_dependency to return early before the warning.
> This preserves the use-after-free fix from commit 29ab76827761
> ("PM: runtime: Fix a race condition related to device removal")
> without the warning.
>
> Fixes: 29ab76827761 ("PM: runtime: Fix a race condition related to device removal")
> Signed-off-by: Zhou Minqiang <zhouminqiang2@xxxxxxxxxx>
> ---
>  drivers/base/power/runtime.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/drivers/base/power/runtime.c b/drivers/base/power/runtime.c
> index 335288e8b5b3..31537b745c81 100644
> --- a/drivers/base/power/runtime.c
> +++ b/drivers/base/power/runtime.c
> @@ -1895,7 +1895,7 @@ void pm_runtime_reinit(struct device *dev)
>  void pm_runtime_remove(struct device *dev)
>  {
>         __pm_runtime_disable(dev, false);
> -       flush_work(&dev->power.work);
> +       cancel_work_sync(&dev->power.work);
>         pm_runtime_reinit(dev);
>  }
>
> --

Applied as 7.4-rc material, thanks!