Re: [PATCH] drm/xe: Don't transfer shrinker pages for pinned bos on purgeable state change

["Thomas Hellström" <thomas.hellstrom@linux.intel.com>]

On Fri, 2026-10-09 at 07:13 +0000, sashiko-bot@kernel.org wrote:
> Thank you for your contribution! Sashiko AI review found 1 potential
> issue(s) to consider:
> - [High] drm/xe: Shrinker accounting underflow for display BOs on
> MADV_DONTNEED
> --
> 
> commit d5c4a3b2c7860c46686f096832d7c3684f902103
> Author: Thomas Hellström <thomas.hellstrom@linux.intel.com>
> 
>     drm/xe: Don't transfer shrinker pages for pinned bos on purgeable
> state change
>     
>     Pinned bos have their pages removed from the shrinker's
> shrinkable /
>     purgeable counts on pin, and added back on unpin according to the
>     bo's purgeable state at that time.
> 
> > diff --git a/drivers/gpu/drm/xe/xe_bo.c
> > b/drivers/gpu/drm/xe/xe_bo.c
> > index 6921b6967330d..a752e01efbd75 100644
> > --- a/drivers/gpu/drm/xe/xe_bo.c
> > +++ b/drivers/gpu/drm/xe/xe_bo.c
> [ ... ]
> > @@ -876,7 +878,8 @@ static void xe_bo_set_purgeable_shrinker(struct
> > xe_bo *bo,
> >  		return;
> >  
> >  	xe_tt = container_of(tt, struct xe_ttm_tt, ttm);
> > -	tt_pages = tt->num_pages;
> > +	/* Pinned bos are not accounted. Only update the purgeable
> > flag. */
> > +	tt_pages = xe_bo_is_pinned(bo) ? 0 : tt->num_pages;
> 
> [Severity: High]
> Could this introduce a regression leading to a shrinker accounting
> underflow
> for native display BOs?
> 
> Native display framebuffers are pinned for scanout via
> __xe_pin_fb_vma().
> Unlike imported BOs, this function calls ttm_bo_pin() directly
> without calling
> xe_ttm_tt_account_subtract():
> 
> drivers/gpu/drm/xe/display/xe_fb_pin.c:__xe_pin_fb_vma() {
>     ...
>     ret = ttm_bo_pin(&bo->ttm);
>     ...
> }

This is no longer the case. This issue is fixed in drm-tip.

/Thomas




> 
> Because of this, their pages appear to remain actively accounted in
> the
> shrinker's shrinkable_pages bucket while pinned.
> 
> If userspace sets MADV_DONTNEED on a display BO while it is pinned,
> won't this
> new check skip the page transfer from shrinkable_pages to
> purgeable_pages
> since xe_bo_is_pinned(bo) is true?
> 
> It seems this skips the page transfer but still sets xe_tt->purgeable
> = true.
> When the display BO is later unpinned via __xe_unpin_fb_vma(), it
> calls
> ttm_bo_unpin() directly without adding pages back to the shrinker:
> 
> drivers/gpu/drm/xe/display/xe_fb_pin.c:__xe_unpin_fb_vma() {
>     ...
>     ttm_bo_unpin(&vma->bo->ttm);
>     ...
> }
> 
> When the BO is eventually unpopulated or destroyed, wouldn't
> xe_ttm_tt_account_subtract() see xe_tt->purgeable == true and
> subtract the
> pages from the purgeable_pages bucket?
> 
> Since the pages were never transferred, does this cause
> purgeable_pages to
> underflow and shrinkable_pages to remain permanently inflated?