Re: [PATCH v5] drm/xe/shrinker: Take a runtime PM ref before shrinking non-system memory

[sashiko-bot@kernel.org]

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

Pre-existing issues:
- [High] Unbounded spinlock-held scan in LRU list iteration due to skipping entries without advancing the scan counter.
--

commit 63284e08ce6cb6db797af1075279023e72aafb0e
Author: Shuicheng Lin <shuicheng.lin@intel.com>
Subject: drm/xe/shrinker: Take a runtime PM ref before shrinking non-system memory

This commit updates the shrinker to take a runtime PM reference before walking 
non-system memory LRU lists. This prevents invalidating GPU mappings while the 
device is suspended, and also stops the walk once the scan target is met.

> diff --git a/drivers/gpu/drm/xe/xe_shrinker.c b/drivers/gpu/drm/xe/xe_shrinker.c
> --- a/drivers/gpu/drm/xe/xe_shrinker.c
> +++ b/drivers/gpu/drm/xe/xe_shrinker.c

[ ... ]

> @@ -71,16 +73,31 @@ static s64 __xe_shrinker_walk(struct xe_device *xe,
>  		ttm_bo_lru_for_each_reserved_guarded(&curs, man, &arg, ttm_bo) {
>  			if (!ttm_bo_shrink_suitable(ttm_bo, ctx))
>  				continue;

[Severity: High]
This isn't a bug introduced by this patch, but is there a risk of an
unbounded spinlock-held scan here?

During memory reclaim, __xe_shrinker_walk() iterates over the GPU LRU lists
using ttm_bo_lru_for_each_reserved_guarded(), which acquires the lru_lock
spinlock. If an item is not suitable for shrinking (e.g., pinned or busy),
the loop skips it using continue without incrementing the *scanned counter.

Since *scanned is only incremented inside xe_bo_shrink(), skipping entries
bypasses the bound check entirely. Under high memory pressure with many
pinned or busy buffer objects, could this create an unbounded loop that
leads to soft lockups or CPU stalls?

>  
>  			lret = xe_bo_shrink(ctx, ttm_bo, flags, scanned);
> -			if (lret < 0)
> -				return lret;
> +			if (lret < 0) {
> +				freed = lret;
> +				goto out;
> +			}
>  
>  			freed += lret;
>  			if (*scanned >= to_scan)
>  				break;
>  		}

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260901213331.458982-1-shuicheng.lin@intel.com?part=1