Re: [PATCH 1/2] drm/xe/shrinker: use async RPM get

[sashiko-bot@kernel.org]

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [Low] xe: missing trace event in xe_pm_runtime_get_async() creates unbalanced tracing
--

commit 1b04fd7832712be0fc1a084098ad02baf3e5f01f
Author: Matthew Auld <matthew.auld@intel.com>

drm/xe/shrinker: use async RPM get

Replaces the custom PM worker with pm_runtime_get() using RPM_ASYNC to
queue a resume request.

> diff --git a/drivers/gpu/drm/xe/xe_pm.c b/drivers/gpu/drm/xe/xe_pm.c
> index 4a745c3a933b4..ba63a26444957 100644
> --- a/drivers/gpu/drm/xe/xe_pm.c
> +++ b/drivers/gpu/drm/xe/xe_pm.c

[ ... ]

> @@ -803,6 +803,18 @@ void xe_pm_runtime_get(struct xe_device *xe)
>  	pm_runtime_resume(xe->drm.dev);
>  }
>  
> +/**
> + * xe_pm_runtime_get_async - Get a runtime_pm reference and queue up a resume
> + * @xe: xe device instance
> + *
> + * Grabs a runtime PM reference and requests an asynchronous resume without
> + * blocking. Safe to call from contexts such as reclaim.
> + */
> +void xe_pm_runtime_get_async(struct xe_device *xe)
> +{
> +	pm_runtime_get(xe->drm.dev);
> +}

[Severity: Low]
Does this new function need a trace_xe_pm_runtime_get() call?

Other synchronous PM wrappers in this driver include a trace event.
Without it here, tools tracking RPM refcounts might log 'put' events
without corresponding 'get' events from the async path.

[ ... ]

> diff --git a/drivers/gpu/drm/xe/xe_shrinker.c b/drivers/gpu/drm/xe/xe_shrinker.c
> index deb4378c1ec15..518f77acf1dd3 100644
> --- a/drivers/gpu/drm/xe/xe_shrinker.c
> +++ b/drivers/gpu/drm/xe/xe_shrinker.c

[ ... ]

> @@ -66,7 +64,8 @@ static bool __xe_shrinker_runtime_pm_get(struct xe_shrinker *shrinker)
>  		return true;
>  	}
>  
> -	queue_work(xe->unordered_wq, &shrinker->pm_worker);
> +	xe_pm_runtime_get_async(xe);
> +	xe_pm_runtime_put(xe);
>  
>  	return false;
>  }

[Severity: Low]
During memory reclaim under pressure, if xe_pm_runtime_get_if_active()
fails, the driver falls back to this path.

Because xe_pm_runtime_get_async() is missing a trace, this sequence will
produce a trace_xe_pm_runtime_put() without a corresponding get trace,
unlike the pm_worker thread it replaced. Is this tracing regression
intentional?

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260925115521.1018940-3-matthew.auld@intel.com?part=1