Re: [PATCH] drm/xe: Tie page count tracepoints to TTM accounting functions

[Matthew Brost <matthew.brost@intel.com>]

On Wed, Jan 07, 2026 at 01:01:12PM -0700, Summers, Stuart wrote:
> On Tue, 2026-01-06 at 16:21 -0800, Matthew Brost wrote:
> > Page accounting can change via the shrinker without calling
> > xe_ttm_tt_unpopulate(), but those paths already update accounting
> > through the xe_ttm_tt_account_*() helpers.
> 
> I see this is getting also called through the xe_bo_pin/unpin functions
> with a check for ttm_tt_is_populated(). Does that mean after this
> change we'd get a double accounting in those cases?
> 

Ah, good catch. It would actually be the inverse of that - we'd not
account for pinned pages in global page counter. Let me fix this.

Matt

> Thanks,
> Stuart
> 
> > 
> > Move the page count tracepoints into xe_ttm_tt_account_add() and
> > xe_ttm_tt_account_subtract() so accounting updates are recorded
> > consistently, regardless of whether pages are populated, unpopulated,
> > or reclaimed via the shrinker.
> > 
> > This avoids missing page count updates and keeps global accounting
> > balanced across all TT lifecycle paths.
> > 
> > Cc: stable@vger.kernel.org
> > Fixes: ce3d39fae3d3 ("drm/xe/bo: add GPU memory trace points")
> > Signed-off-by: Matthew Brost <matthew.brost@intel.com>
> > ---
> >  drivers/gpu/drm/xe/xe_bo.c | 7 +++++--
> >  1 file changed, 5 insertions(+), 2 deletions(-)
> > 
> > diff --git a/drivers/gpu/drm/xe/xe_bo.c b/drivers/gpu/drm/xe/xe_bo.c
> > index 8b6474cd3eaf..33afaee38f48 100644
> > --- a/drivers/gpu/drm/xe/xe_bo.c
> > +++ b/drivers/gpu/drm/xe/xe_bo.c
> > @@ -432,6 +432,9 @@ struct sg_table *xe_bo_sg(struct xe_bo *bo)
> >         return xe_tt->sg;
> >  }
> >  
> > +static void update_global_total_pages(struct ttm_device *ttm_dev,
> > +                                     long num_pages);
> > +
> >  /*
> >   * Account ttm pages against the device shrinker's shrinkable and
> >   * purgeable counts.
> > @@ -440,6 +443,7 @@ static void xe_ttm_tt_account_add(struct
> > xe_device *xe, struct ttm_tt *tt)
> >  {
> >         struct xe_ttm_tt *xe_tt = container_of(tt, struct xe_ttm_tt,
> > ttm);
> >  
> > +       update_global_total_pages(&xe->ttm, tt->num_pages);
> >         if (xe_tt->purgeable)
> >                 xe_shrinker_mod_pages(xe->mem.shrinker, 0, tt-
> > >num_pages);
> >         else
> > @@ -450,6 +454,7 @@ static void xe_ttm_tt_account_subtract(struct
> > xe_device *xe, struct ttm_tt *tt)
> >  {
> >         struct xe_ttm_tt *xe_tt = container_of(tt, struct xe_ttm_tt,
> > ttm);
> >  
> > +       update_global_total_pages(&xe->ttm, -(long)tt->num_pages);
> >         if (xe_tt->purgeable)
> >                 xe_shrinker_mod_pages(xe->mem.shrinker, 0, -(long)tt-
> > >num_pages);
> >         else
> > @@ -575,7 +580,6 @@ static int xe_ttm_tt_populate(struct ttm_device
> > *ttm_dev, struct ttm_tt *tt,
> >  
> >         xe_tt->purgeable = false;
> >         xe_ttm_tt_account_add(ttm_to_xe_device(ttm_dev), tt);
> > -       update_global_total_pages(ttm_dev, tt->num_pages);
> >  
> >         return 0;
> >  }
> > @@ -592,7 +596,6 @@ static void xe_ttm_tt_unpopulate(struct
> > ttm_device *ttm_dev, struct ttm_tt *tt)
> >  
> >         ttm_pool_free(&ttm_dev->pool, tt);
> >         xe_ttm_tt_account_subtract(xe, tt);
> > -       update_global_total_pages(ttm_dev, -(long)tt->num_pages);
> >  }
> >  
> >  static void xe_ttm_tt_destroy(struct ttm_device *ttm_dev, struct
> > ttm_tt *tt)
>