Re: [PATCH 2/2] drm/panthor: Implement evicted status for GEM objects

[Nicolas Frattaroli]

On Monday, 20 April 2026 18:17:35 Central European Summer Time Boris Brezillon wrote:
> On Mon, 20 Apr 2026 17:47:00 +0200
> Nicolas Frattaroli <nicolas.frattaroli@xxxxxxxxxxxxx> wrote:
> 
> > For fdinfo to be able to fill its evicted counter with data, panthor
> > needs to keep track of whether a GEM object has ever been reclaimed.
> > Just checking whether the pages are resident isn't enough, as newly
> > allocated objects also won't be resident.
> > 
> > Do this with a new atomic_t member on panthor_gem_object. It's increased
> > when an object gets evicted by the shrinker. While it's allowed to wrap
> > around to below zero and assume a value less than a previous observed
> > value, the reclaim counter will never return to 0 for any particular
> > object once it's been reclaimed at least once.
> > 
> > Use this new member to then set the appropriate DRM_GEM_OBJECT_EVICTED
> > status flag for fdinfo, and use it in the gems debugfs. It's possible to
> > distinguish evicted non-resident pages from newly allocated non-resident
> > pages by checking whether reclaimed_count is != 0.
> > 
> > Signed-off-by: Nicolas Frattaroli <nicolas.frattaroli@xxxxxxxxxxxxx>
> > ---
> >  drivers/gpu/drm/panthor/panthor_gem.c | 10 ++++++++++
> >  drivers/gpu/drm/panthor/panthor_gem.h | 11 +++++++++++
> >  2 files changed, 21 insertions(+)
> > 
> > diff --git a/drivers/gpu/drm/panthor/panthor_gem.c b/drivers/gpu/drm/panthor/panthor_gem.c
> > index 69cef05b6ef7..4b761b39565d 100644
> > --- a/drivers/gpu/drm/panthor/panthor_gem.c
> > +++ b/drivers/gpu/drm/panthor/panthor_gem.c
> > @@ -687,6 +687,10 @@ static void panthor_gem_evict_locked(struct panthor_gem_object *bo)
> >  	if (drm_WARN_ON_ONCE(bo->base.dev, !bo->backing.pages))
> >  		return;
> >  
> > +	/* Don't ever wrap around as far as 0, jump from INT_MIN to 1 */
> > +	if (!atomic_inc_unless_negative(&bo->reclaimed_count))
> > +		atomic_set(&bo->reclaimed_count, 1);
> 
> Can't we just go
> 
> 	atomic_add_unless(&bo->reclaimed_count, 1, INT_MAX);
> 
> here, to handle the INT_MAX saturation?

Yeah, I was torn between the two. My way does keep a somewhat
cyclical nature, so once it wraps, things could still detect
that it's increasing from one observance to the next.

But now that I think about it again, it's not a property of this
count that's worth keeping, I think. Nothing relies on this
behaviour currently, and at best it'll trip up any code that uses
it for a < comparison later down the road.

> > +
> >  	panthor_gem_dev_map_cleanup_locked(bo);
> >  	panthor_gem_backing_cleanup_locked(bo);
> >  	panthor_gem_update_reclaim_state_locked(bo, NULL);
> > @@ -788,6 +792,8 @@ static enum drm_gem_object_status panthor_gem_status(struct drm_gem_object *obj)
> >  
> >  	if (drm_gem_is_imported(&bo->base) || bo->backing.pages)
> >  		res |= DRM_GEM_OBJECT_RESIDENT;
> > +	else if (atomic_read(&bo->reclaimed_count))
> > +		res |= DRM_GEM_OBJECT_EVICTED;
> >  
> >  	return res;
> >  }
> > @@ -1595,6 +1601,7 @@ static void panthor_gem_debugfs_print_flag_names(struct seq_file *m)
> >  	static const char * const gem_state_flags_names[] = {
> >  		[PANTHOR_DEBUGFS_GEM_STATE_IMPORTED_BIT] = "imported",
> >  		[PANTHOR_DEBUGFS_GEM_STATE_EXPORTED_BIT] = "exported",
> > +		[PANTHOR_DEBUGFS_GEM_STATE_EVICTED_BIT] = "evicted",
> >  	};
> >  
> >  	static const char * const gem_usage_flags_names[] = {
> > @@ -1648,6 +1655,9 @@ static void panthor_gem_debugfs_bo_print(struct panthor_gem_object *bo,
> >  
> >  	if (drm_gem_is_imported(&bo->base))
> >  		gem_state_flags |= PANTHOR_DEBUGFS_GEM_STATE_FLAG_IMPORTED;
> > +	else if (!resident_size && atomic_read(&bo->reclaimed_count))
> > +		gem_state_flags |= PANTHOR_DEBUGFS_GEM_STATE_FLAG_EVICTED;
> 
> I think it'd be interesting to know the number of times a BO got
> evicted.

Agreed. Should I add that as a separate column? I think I'll keep the
state flag even if I add a column, because even if it's technically
redundant, it'll still be useful to see without having to compare
three different columns.

> > +
> >  	if (bo->base.dma_buf)
> >  		gem_state_flags |= PANTHOR_DEBUGFS_GEM_STATE_FLAG_EXPORTED;
> >  
> > diff --git a/drivers/gpu/drm/panthor/panthor_gem.h b/drivers/gpu/drm/panthor/panthor_gem.h
> > index ae0491d0b121..1ab573f03330 100644
> > --- a/drivers/gpu/drm/panthor/panthor_gem.h
> > +++ b/drivers/gpu/drm/panthor/panthor_gem.h
> > @@ -19,12 +19,16 @@ struct panthor_vm;
> >  enum panthor_debugfs_gem_state_flags {
> >  	PANTHOR_DEBUGFS_GEM_STATE_IMPORTED_BIT = 0,
> >  	PANTHOR_DEBUGFS_GEM_STATE_EXPORTED_BIT = 1,
> > +	PANTHOR_DEBUGFS_GEM_STATE_EVICTED_BIT = 2,
> >  
> >  	/** @PANTHOR_DEBUGFS_GEM_STATE_FLAG_IMPORTED: GEM BO is PRIME imported. */
> >  	PANTHOR_DEBUGFS_GEM_STATE_FLAG_IMPORTED = BIT(PANTHOR_DEBUGFS_GEM_STATE_IMPORTED_BIT),
> >  
> >  	/** @PANTHOR_DEBUGFS_GEM_STATE_FLAG_EXPORTED: GEM BO is PRIME exported. */
> >  	PANTHOR_DEBUGFS_GEM_STATE_FLAG_EXPORTED = BIT(PANTHOR_DEBUGFS_GEM_STATE_EXPORTED_BIT),
> > +
> > +	/** @PANTHOR_DEBUGFS_GEM_STATE_FLAG_EVICTED: GEM BO is evicted to swap. */
> > +	PANTHOR_DEBUGFS_GEM_STATE_FLAG_EVICTED = BIT(PANTHOR_DEBUGFS_GEM_STATE_EVICTED_BIT),
> >  };
> >  
> >  enum panthor_debugfs_gem_usage_flags {
> > @@ -172,6 +176,13 @@ struct panthor_gem_object {
> >  	/** @reclaim_state: Cached reclaim state */
> >  	enum panthor_gem_reclaim_state reclaim_state;
> >  
> > +	/**
> > +	 * @reclaimed_count: How many times object has been evicted to swap.
> 
> 	 *
> 
> > +	 * Never returns to 0 once incremented even on wrap-around, but may
> > +	 * become < 0 and < the previous value if wrap-around occurs.
> 
> With the saturation I suggested, I'd just add that when INT_MAX is
> reached, it will stay there.
> 
> > +	 */
> > +	atomic_t reclaimed_count;
> > +
> >  	/**
> >  	 * @exclusive_vm_root_gem: Root GEM of the exclusive VM this GEM object
> >  	 * is attached to.
> > 
> 
>