Intel-XE Archive on lore.kernel.org
 help / color / mirror / Atom feed
From: "Thomas Hellström" <thomas.hellstrom@linux.intel.com>
To: Matthew Auld <matthew.auld@intel.com>, intel-xe@lists.freedesktop.org
Subject: Re: [PATCH v2 2/3] drm/xe: share bo dma-resv with backup object
Date: Tue, 22 Apr 2025 09:01:11 +0200	[thread overview]
Message-ID: <52554ca70519b3567a68cc8bf3c5c7b819975e89.camel@linux.intel.com> (raw)
In-Reply-To: <9442833c-9bb8-4247-8c36-6ac5396dad86@intel.com>

On Thu, 2025-04-17 at 09:45 +0100, Matthew Auld wrote:
> On 16/04/2025 16:43, Thomas Hellström wrote:
> > On Wed, 2025-04-16 at 16:09 +0100, Matthew Auld wrote:
> > > We end up needing to grab both locks together anyway and keep
> > > them
> > > held
> > > until we complete the copy or add the fence. Plus the backup_obj
> > > is
> > > short lived and tied to the parent object, so seems reasonable to
> > > share
> > > the same dma-resv. This will simplify the locking here, and in
> > > follow
> > > up patches.
> > > 
> > > v2:
> > >    - Hold reference to the parent bo to be sure the shared dma-
> > > resv
> > > can't
> > >      go out of scope too soon. (Thomas)
> > > 
> > > Signed-off-by: Matthew Auld <matthew.auld@intel.com>
> > > Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
> > 
> > Another thought: What if someone releases the last "ordinary"
> > reference
> > on an object that has a backup object so that the only remaining
> > reference is the backup object. Can that happen?
> > 
> > I have a vague recollection of having to deal with that for i915...
> 
> I don't think that can happen in xe, since the xe_bo_unpin() has to
> be 
> explicitly called before the object is finally put(), and the unpin
> will 
> make sure to drop its reference on the backup object, if it's still 
> there, breaking the circular reference and starting the chain
> reaction.

OK, makes sense. 

Reviewed-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>



> 
> > 
> > /Thomas
> > 
> > 
> > > ---
> > >   drivers/gpu/drm/xe/xe_bo.c       | 28 +++++++++++++------------
> > > ---
> > >   drivers/gpu/drm/xe/xe_bo_types.h |  2 ++
> > >   2 files changed, 15 insertions(+), 15 deletions(-)
> > > 
> > > diff --git a/drivers/gpu/drm/xe/xe_bo.c
> > > b/drivers/gpu/drm/xe/xe_bo.c
> > > index c337790c81ae..79adaee5a0e9 100644
> > > --- a/drivers/gpu/drm/xe/xe_bo.c
> > > +++ b/drivers/gpu/drm/xe/xe_bo.c
> > > @@ -1120,13 +1120,15 @@ int xe_bo_evict_pinned(struct xe_bo *bo)
> > >   	if (bo->flags & XE_BO_FLAG_PINNED_NORESTORE)
> > >   		goto out_unlock_bo;
> > >   
> > > -	backup = xe_bo_create_locked(xe, NULL, NULL, bo->size,
> > > ttm_bo_type_kernel,
> > > -				     XE_BO_FLAG_SYSTEM |
> > > XE_BO_FLAG_NEEDS_CPU_ACCESS |
> > > -				     XE_BO_FLAG_PINNED);
> > > +	backup = ___xe_bo_create_locked(xe, NULL, NULL, bo-
> > > > ttm.base.resv, NULL, bo->size,
> > > +					DRM_XE_GEM_CPU_CACHING_W
> > > B,
> > > ttm_bo_type_kernel,
> > > +					XE_BO_FLAG_SYSTEM |
> > > XE_BO_FLAG_NEEDS_CPU_ACCESS |
> > > +					XE_BO_FLAG_PINNED);
> > >   	if (IS_ERR(backup)) {
> > >   		ret = PTR_ERR(backup);
> > >   		goto out_unlock_bo;
> > >   	}
> > > +	backup->parent_obj = xe_bo_get(bo); /* Released by
> > > bo_destroy */
> > >   
> > >   	if (xe_bo_is_user(bo) || (bo->flags &
> > > XE_BO_FLAG_PINNED_LATE_RESTORE)) {
> > >   		struct xe_migrate *migrate;
> > > @@ -1177,7 +1179,6 @@ int xe_bo_evict_pinned(struct xe_bo *bo)
> > >   
> > >   out_backup:
> > >   	xe_bo_vunmap(backup);
> > > -	xe_bo_unlock(backup);
> > >   	if (ret)
> > >   		xe_bo_put(backup);
> > >   out_unlock_bo:
> > > @@ -1212,17 +1213,12 @@ int xe_bo_restore_pinned(struct xe_bo
> > > *bo)
> > >   	if (!backup)
> > >   		return 0;
> > >   
> > > -	xe_bo_lock(backup, false);
> > > +	xe_bo_lock(bo, false);
> > >   
> > >   	ret = ttm_bo_validate(&backup->ttm, &backup->placement,
> > > &ctx);
> > >   	if (ret)
> > >   		goto out_backup;
> > >   
> > > -	if (WARN_ON(!dma_resv_trylock(bo->ttm.base.resv))) {
> > > -		ret = -EBUSY;
> > > -		goto out_backup;
> > > -	}
> > > -
> > >   	if (xe_bo_is_user(bo) || (bo->flags &
> > > XE_BO_FLAG_PINNED_LATE_RESTORE)) {
> > >   		struct xe_migrate *migrate;
> > >   		struct dma_fence *fence;
> > > @@ -1271,15 +1267,14 @@ int xe_bo_restore_pinned(struct xe_bo
> > > *bo)
> > >   
> > >   	bo->backup_obj = NULL;
> > >   
> > > +out_backup:
> > > +	xe_bo_vunmap(backup);
> > > +	if (!bo->backup_obj)
> > > +		xe_bo_put(backup);
> > >   out_unlock_bo:
> > >   	if (unmap)
> > >   		xe_bo_vunmap(bo);
> > >   	xe_bo_unlock(bo);
> > > -out_backup:
> > > -	xe_bo_vunmap(backup);
> > > -	xe_bo_unlock(backup);
> > > -	if (!bo->backup_obj)
> > > -		xe_bo_put(backup);
> > >   	return ret;
> > >   }
> > >   
> > > @@ -1532,6 +1527,9 @@ static void xe_ttm_bo_destroy(struct
> > > ttm_buffer_object *ttm_bo)
> > >   	if (bo->vm && xe_bo_is_user(bo))
> > >   		xe_vm_put(bo->vm);
> > >   
> > > +	if (bo->parent_obj)
> > > +		xe_bo_put(bo->parent_obj);
> > > +
> > >   	mutex_lock(&xe->mem_access.vram_userfault.lock);
> > >   	if (!list_empty(&bo->vram_userfault_link))
> > >   		list_del(&bo->vram_userfault_link);
> > > diff --git a/drivers/gpu/drm/xe/xe_bo_types.h
> > > b/drivers/gpu/drm/xe/xe_bo_types.h
> > > index 81396181aaea..eb5e83c5f233 100644
> > > --- a/drivers/gpu/drm/xe/xe_bo_types.h
> > > +++ b/drivers/gpu/drm/xe/xe_bo_types.h
> > > @@ -30,6 +30,8 @@ struct xe_bo {
> > >   	struct ttm_buffer_object ttm;
> > >   	/** @backup_obj: The backup object when pinned and
> > > suspended
> > > (vram only) */
> > >   	struct xe_bo *backup_obj;
> > > +	/** @parent_obj: Ref to parent bo if this a backup_obj
> > > */
> > > +	struct xe_bo *parent_obj;
> > >   	/** @size: Size of this buffer object */
> > >   	size_t size;
> > >   	/** @flags: flags for this buffer object */
> > 
> 


  reply	other threads:[~2025-04-22  7:01 UTC|newest]

Thread overview: 18+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2025-04-16 15:09 [PATCH v2 0/3] PM notifier Matthew Auld
2025-04-16 15:09 ` [PATCH v2 1/3] drm/xe: evict user memory in " Matthew Auld
2025-04-16 15:09 ` [PATCH v2 2/3] drm/xe: share bo dma-resv with backup object Matthew Auld
2025-04-16 15:43   ` Thomas Hellström
2025-04-17  8:45     ` Matthew Auld
2025-04-22  7:01       ` Thomas Hellström [this message]
2025-04-16 15:09 ` [PATCH v2 3/3] drm/xe: handle pinned memory in PM notifier Matthew Auld
2025-04-16 15:45   ` Thomas Hellström
2025-04-16 16:23 ` ✓ Xe.CI.BAT: success for PM notifier (rev2) Patchwork
2025-04-16 23:44 ` ✗ Xe.CI.Full: failure " Patchwork
2025-04-22 11:31 ` ✓ CI.Patch_applied: success for PM notifier (rev3) Patchwork
2025-04-22 11:31 ` ✓ CI.checkpatch: " Patchwork
2025-04-22 11:33 ` ✓ CI.KUnit: " Patchwork
2025-04-22 11:41 ` ✓ CI.Build: " Patchwork
2025-04-22 11:43 ` ✗ CI.Hooks: failure " Patchwork
2025-04-22 11:44 ` ✓ CI.checksparse: success " Patchwork
2025-04-22 12:04 ` ✓ Xe.CI.BAT: " Patchwork
2025-04-22 14:29 ` ✗ Xe.CI.Full: failure " Patchwork

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=52554ca70519b3567a68cc8bf3c5c7b819975e89.camel@linux.intel.com \
    --to=thomas.hellstrom@linux.intel.com \
    --cc=intel-xe@lists.freedesktop.org \
    --cc=matthew.auld@intel.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox