[PATCH] accel/amdxdna: fix usage_count leak when autosuspend_delay is negative

[PATCH] accel/amdxdna: fix usage_count leak when autosuspend_delay is negative

[Guangshuo Li]

amdxdna_pm_init() calls pm_runtime_use_autosuspend(), but
amdxdna_pm_fini() does not call the matching
pm_runtime_dont_use_autosuspend() when tearing down runtime PM.

If autosuspend_delay is set to a negative value while autosuspend is
enabled, the runtime PM core increments usage_count to prevent runtime
suspend. Without calling pm_runtime_dont_use_autosuspend() during
driver teardown, this reference is not dropped and usage_count remains
unbalanced.

Add the missing pm_runtime_dont_use_autosuspend() call in
amdxdna_pm_fini() before restoring the runtime PM usage reference and
forbidding runtime PM.

This issue was found by manual code inspection.

Fixes: 063db451832b ("accel/amdxdna: Enhance runtime power management")
Signed-off-by: Guangshuo Li <lgs201920130244@gmail.com>
---
 drivers/accel/amdxdna/amdxdna_pm.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/drivers/accel/amdxdna/amdxdna_pm.c b/drivers/accel/amdxdna/amdxdna_pm.c
index b1fafddd7ad5..7dcedb7c105f 100644
--- a/drivers/accel/amdxdna/amdxdna_pm.c
+++ b/drivers/accel/amdxdna/amdxdna_pm.c
@@ -73,6 +73,7 @@ void amdxdna_pm_fini(struct amdxdna_dev *xdna)
 {
 	struct device *dev = xdna->ddev.dev;
 
+	pm_runtime_dont_use_autosuspend(dev);
 	pm_runtime_get_noresume(dev);
 	pm_runtime_forbid(dev);
 }
-- 
2.43.0

Re: [PATCH] accel/amdxdna: fix usage_count leak when autosuspend_delay is negative

[sashiko-bot]

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:

Pre-existing issues:
- [High] Deadlock during device teardown due to synchronous runtime PM resume while holding dev_lock
--

--- Patch [1]: [PATCH] accel/amdxdna: fix usage_count leak when autosuspend_delay is negative ---
commit 31ff9d16b70b74f4b276ff67df887f4acb154424
Author: Guangshuo Li <lgs201920130244@gmail.com>

accel/amdxdna: fix usage_count leak when autosuspend_delay is negative

This commit fixes a usage_count leak when autosuspend_delay is negative
by adding a missing pm_runtime_dont_use_autosuspend() call during driver
teardown in amdxdna_pm_fini().

> diff --git a/drivers/accel/amdxdna/amdxdna_pm.c b/drivers/accel/amdxdna/amdxdna_pm.c
> index b1fafddd7ad59..7dcedb7c105f2 100644
> --- a/drivers/accel/amdxdna/amdxdna_pm.c
> +++ b/drivers/accel/amdxdna/amdxdna_pm.c
> @@ -73,6 +73,7 @@ void amdxdna_pm_fini(struct amdxdna_dev *xdna)
>  {
>  	struct device *dev = xdna->ddev.dev;
>  
> +	pm_runtime_dont_use_autosuspend(dev);
>  	pm_runtime_get_noresume(dev);
>  	pm_runtime_forbid(dev);

[Severity: High]
This is a pre-existing issue, but does calling pm_runtime_forbid() here
cause a self-deadlock during device teardown?

If the device is in the RPM_SUSPENDED state when driver unbind occurs,
amdxdna_remove() acquires dev_lock before starting the teardown chain:

drivers/accel/amdxdna/amdxdna_pci_drv.c:amdxdna_remove() {
    mutex_lock(&xdna->dev_lock);
    ...
    xdna->dev_info->ops->fini(xdna);
    ...
}

The teardown chain eventually reaches amdxdna_pm_fini(). Calling
pm_runtime_forbid() triggers a synchronous resume, which in turn invokes the
driver's resume callback. The resume callback then attempts to acquire the
exact same lock:

drivers/accel/amdxdna/amdxdna_pm.c:amdxdna_pm_resume() {
    guard(mutex)(&xdna->dev_lock);
    ...
}

Could this sequence cause the unbind thread to hang indefinitely waiting for
xdna->dev_lock?

>  }

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260807143200.2500342-1-lgs201920130244@gmail.com?part=1