Skip to contentSkip to conversation

drm/gem: modernize locks to use scoped_guard()

PATCH In review Biren Pandya 3 participants 5 messages Jun 16 – Jun 22

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 drm/gem: modernize locks to use scoped_guard()

Several GEM core functions manually managed mutex_lock() and mutex_unlock() over single scopes or error paths. This adds boilerplate and carries the risk of lock leaks if error paths are refactored.

Modernize these locks by deploying the <linux/cleanup.h> scoped_guard() macro. This ensures that the locks are reliably dropped when the block exits, cleanly removing goto out_unlock paths and tightening the lifecycle.

Compiled locally, but requires IGT validation by the DRM CI.

---
 drivers/gpu/drm/drm_gem.c | 66 ++++++++++++++++++---------------------
 1 file changed, 30 insertions(+), 36 deletions(-)
drivers/gpu/drm/drm_gem.c +30 -36
index 891c3bff5ae0..d3a061d42ba7 100644
@@ -346,13 +346,13 @@ void drm_gem_object_handle_put_unlocked(struct drm_gem_object *obj)
 	 * we checked for a name.
 	 */
 
-	mutex_lock(&dev->object_name_lock);
-	if (--obj->handle_count == 0) {
-		drm_gem_object_handle_free(obj);
-		drm_gem_object_exported_dma_buf_free(obj);
-		final = true;
+	scoped_guard(mutex, &dev->object_name_lock) {
+		if (--obj->handle_count == 0) {
+			drm_gem_object_handle_free(obj);
+			drm_gem_object_exported_dma_buf_free(obj);
+			final = true;
+		}
 	}
-	mutex_unlock(&dev->object_name_lock);
 
 	if (final)
 		drm_gem_object_put(obj);
@@ -374,11 +374,8 @@ drm_gem_object_release_handle(int id, void *ptr, void *data)
 	if (obj->funcs->close)
 		obj->funcs->close(obj, file_priv);
 
-	mutex_lock(&file_priv->prime.lock);
-
-	drm_prime_remove_buf_handle(&file_priv->prime, id);
-
-	mutex_unlock(&file_priv->prime.lock);
+	scoped_guard(mutex, &file_priv->prime.lock)
+		drm_prime_remove_buf_handle(&file_priv->prime, id);
 
 	drm_vma_node_revoke(&obj->vma_node, file_priv);
 
@@ -1021,37 +1018,34 @@ int drm_gem_change_handle_ioctl(struct drm_device *dev, void *data,
 		goto out;
 	}
 
-	mutex_lock(&file_priv->prime.lock);
+	scoped_guard(mutex, &file_priv->prime.lock) {
+		spin_lock(&file_priv->table_lock);
+		ret = idr_alloc(&file_priv->object_idr, obj, handle, handle + 1,
+				GFP_NOWAIT);
+		spin_unlock(&file_priv->table_lock);
 
-	spin_lock(&file_priv->table_lock);
-	ret = idr_alloc(&file_priv->object_idr, obj, handle, handle + 1,
-			GFP_NOWAIT);
-	spin_unlock(&file_priv->table_lock);
+		if (ret < 0)
+			break;
 
-	if (ret < 0)
-		goto out_unlock;
+		if (obj->dma_buf) {
+			ret = drm_prime_add_buf_handle(&file_priv->prime, obj->dma_buf,
+						       handle);
+			if (ret < 0) {
+				spin_lock(&file_priv->table_lock);
+				idr_remove(&file_priv->object_idr, handle);
+				spin_unlock(&file_priv->table_lock);
+				break;
+			}
 
-	if (obj->dma_buf) {
-		ret = drm_prime_add_buf_handle(&file_priv->prime, obj->dma_buf,
-					       handle);
-		if (ret < 0) {
-			spin_lock(&file_priv->table_lock);
-			idr_remove(&file_priv->object_idr, handle);
-			spin_unlock(&file_priv->table_lock);
-			goto out_unlock;
+			drm_prime_remove_buf_handle(&file_priv->prime, args->handle);
 		}
 
-		drm_prime_remove_buf_handle(&file_priv->prime, args->handle);
-	}
-
-	ret = 0;
-
-	spin_lock(&file_priv->table_lock);
-	idr_remove(&file_priv->object_idr, args->handle);
-	spin_unlock(&file_priv->table_lock);
+		ret = 0;
 
-out_unlock:
-	mutex_unlock(&file_priv->prime.lock);
+		spin_lock(&file_priv->table_lock);
+		idr_remove(&file_priv->object_idr, args->handle);
+		spin_unlock(&file_priv->table_lock);
+	}
 out:
 	drm_gem_object_put(obj);
 

base-commit: 72c25183cac9bc584c9de21797a5883af44bcc7a

Signature · 1 line
2.50.1 (Apple Git-155)
#1 lore raw
Quoted 8 lines · Biren Pandya

Several GEM core functions manually managed mutex_lock() and mutex_unlock() over single scopes or error paths. This adds boilerplate and carries the risk of lock leaks if error paths are refactored.

Modernize these locks by deploying the <linux/cleanup.h> scoped_guard() macro. This ensures that the locks are reliably dropped when the block exits, cleanly removing goto out_unlock paths and tightening the lifecycle.

What's the reason for doing so in in drm_gem and not other areas in DRM ?

Quoted 46 lines

Compiled locally, but requires IGT validation by the DRM CI.

---
 drivers/gpu/drm/drm_gem.c | 66 ++++++++++++++++++---------------------
 1 file changed, 30 insertions(+), 36 deletions(-)
drivers/gpu/drm/drm_gem.c +8 -11
index 891c3bff5ae0..d3a061d42ba7 100644
@@ -346,13 +346,13 @@ void drm_gem_object_handle_put_unlocked(struct drm_gem_object *obj)
 	 * we checked for a name.
 	 */
 
-	mutex_lock(&dev->object_name_lock);
-	if (--obj->handle_count == 0) {
-		drm_gem_object_handle_free(obj);
-		drm_gem_object_exported_dma_buf_free(obj);
-		final = true;
+	scoped_guard(mutex, &dev->object_name_lock) {
+		if (--obj->handle_count == 0) {
+			drm_gem_object_handle_free(obj);
+			drm_gem_object_exported_dma_buf_free(obj);
+			final = true;
+		}
 	}
-	mutex_unlock(&dev->object_name_lock);
 
 	if (final)
 		drm_gem_object_put(obj);
@@ -374,11 +374,8 @@ drm_gem_object_release_handle(int id, void *ptr, void *data)
 	if (obj->funcs->close)
 		obj->funcs->close(obj, file_priv);
 
-	mutex_lock(&file_priv->prime.lock);
-
-	drm_prime_remove_buf_handle(&file_priv->prime, id);
-
-	mutex_unlock(&file_priv->prime.lock);
+	scoped_guard(mutex, &file_priv->prime.lock)
+		drm_prime_remove_buf_handle(&file_priv->prime, id);
 
 	drm_vma_node_revoke(&obj->vma_node, file_priv);
 

drivers/gpu/drm/drm_gem.c int drm_gem_change_handle_ioctl(struct drm_device *dev, void *data, · line 1025

Quoted 9 lines
drivers/gpu/drm/drm_gem.c +5 -1
@@ -1021,37 +1018,34 @@ int drm_gem_change_handle_ioctl(struct drm_device *dev, void *data,
 		goto out;
 	}
 
-	mutex_lock(&file_priv->prime.lock);
+	scoped_guard(mutex, &file_priv->prime.lock) {
+		spin_lock(&file_priv->table_lock);
+		ret = idr_alloc(&file_priv->object_idr, obj, handle, handle + 1,
+				GFP_NOWAIT);
+		spin_unlock(&file_priv->table_lock);

And why don't you use guards for the spinlock as well ?

Quoted 52 lines
drivers/gpu/drm/drm_gem.c +17 -24
-	spin_lock(&file_priv->table_lock);
-	ret = idr_alloc(&file_priv->object_idr, obj, handle, handle + 1,
-			GFP_NOWAIT);
-	spin_unlock(&file_priv->table_lock);
+		if (ret < 0)
+			break;
 
-	if (ret < 0)
-		goto out_unlock;
+		if (obj->dma_buf) {
+			ret = drm_prime_add_buf_handle(&file_priv->prime, obj->dma_buf,
+						       handle);
+			if (ret < 0) {
+				spin_lock(&file_priv->table_lock);
+				idr_remove(&file_priv->object_idr, handle);
+				spin_unlock(&file_priv->table_lock);
+				break;
+			}
 
-	if (obj->dma_buf) {
-		ret = drm_prime_add_buf_handle(&file_priv->prime, obj->dma_buf,
-					       handle);
-		if (ret < 0) {
-			spin_lock(&file_priv->table_lock);
-			idr_remove(&file_priv->object_idr, handle);
-			spin_unlock(&file_priv->table_lock);
-			goto out_unlock;
+			drm_prime_remove_buf_handle(&file_priv->prime, args->handle);
 		}
 
-		drm_prime_remove_buf_handle(&file_priv->prime, args->handle);
-	}
-
-	ret = 0;
-
-	spin_lock(&file_priv->table_lock);
-	idr_remove(&file_priv->object_idr, args->handle);
-	spin_unlock(&file_priv->table_lock);
+		ret = 0;
 
-out_unlock:
-	mutex_unlock(&file_priv->prime.lock);
+		spin_lock(&file_priv->table_lock);
+		idr_remove(&file_priv->object_idr, args->handle);
+		spin_unlock(&file_priv->table_lock);
+	}
 out:
 	drm_gem_object_put(obj);
 
 

base-commit: 72c25183cac9bc584c9de21797a5883af44bcc7a

Signature · 3 lines
Regards,

Laurent Pinchart
#2 lore raw
Quoted 6 lines · Laurent Pinchart
Quoted 4 lines · Biren Pandya

Several GEM core functions manually managed mutex_lock() and mutex_unlock() over single scopes or error paths. This adds boilerplate and carries the risk of lock leaks if error paths are refactored.

Quoted 7 lines
Quoted 4 lines

Modernize these locks by deploying the <linux/cleanup.h> scoped_guard() macro. This ensures that the locks are reliably dropped when the block exits, cleanly removing goto out_unlock paths and tightening the lifecycle.

What's the reason for doing so in in drm_gem and not other areas in DRM ?

Hi Laurent,

Thanks for taking a look.
No deeper reason than it being where I happened to start — I didn't mean to single it out, and I'd rather the treatment be consistent than piecemeal.

Quoted 4 lines
Quoted 4 lines
drivers/gpu/drm/drm_gem.c +0 -0
@@ -1021,37 +1018,34 @@ int drm_gem_change_handle_ioctl(struct drm_device *dev, void *data,
              goto out;
      }
 
Quoted 8 lines

drivers/gpu/drm/drm_gem.c int drm_gem_change_handle_ioctl(struct drm_device *dev, void *data, · line 1025

Quoted 6 lines
drivers/gpu/drm/drm_gem.c +5 -1
-     mutex_lock(&file_priv->prime.lock);
+     scoped_guard(mutex, &file_priv->prime.lock) {
+             spin_lock(&file_priv->table_lock);
+             ret = idr_alloc(&file_priv->object_idr, obj, handle, handle + 1,
+                             GFP_NOWAIT);
+             spin_unlock(&file_priv->table_lock);

And why don't you use guards for the spinlock as well ?

Fair point — the spinlocks here are equally good candidates; I only kept v1 to mutexes to keep it small.

That said, this is a pure cleanup with no functional change, so it's entirely your call whether it's worth carrying.
If you'd like it, I'll send a v2 that converts both the mutexes and the spinlocks in drm_gem.c consistently. If you'd prefer not to take cleanup-only churn, I'm happy to drop it — no problem either way.

Thanks,
Biren

#3 lore raw
Quoted 34 lines · biren pandya
Quoted 5 lines · Laurent Pinchart
Quoted 4 lines · Biren Pandya

Several GEM core functions manually managed mutex_lock() and mutex_unlock() over single scopes or error paths. This adds boilerplate and carries the risk of lock leaks if error paths are refactored.

Quoted 7 lines
Quoted 4 lines

Modernize these locks by deploying the <linux/cleanup.h> scoped_guard() macro. This ensures that the locks are reliably dropped when the block exits, cleanly removing goto out_unlock paths and tightening the lifecycle.

What's the reason for doing so in in drm_gem and not other areas in DRM ?

Hi Laurent,

Thanks for taking a look.
No deeper reason than it being where I happened to start — I didn't mean to single it out, and I'd rather the treatment be consistent than piecemeal.

Quoted 4 lines
Quoted 4 lines
drivers/gpu/drm/drm_gem.c +0 -0
@@ -1021,37 +1018,34 @@ int drm_gem_change_handle_ioctl(struct drm_device *dev, void *data,
              goto out;
      }
 
Quoted 8 lines

drivers/gpu/drm/drm_gem.c int drm_gem_change_handle_ioctl(struct drm_device *dev, void *data, · line 1025

Quoted 6 lines
drivers/gpu/drm/drm_gem.c +5 -1
-     mutex_lock(&file_priv->prime.lock);
+     scoped_guard(mutex, &file_priv->prime.lock) {
+             spin_lock(&file_priv->table_lock);
+             ret = idr_alloc(&file_priv->object_idr, obj, handle, handle + 1,
+                             GFP_NOWAIT);
+             spin_unlock(&file_priv->table_lock);

And why don't you use guards for the spinlock as well ?

Quoted 8 lines

Fair point — the spinlocks here are equally good candidates; I only kept v1 to mutexes to keep it small.

That said, this is a pure cleanup with no functional change, so it's entirely your call whether it's worth carrying.
If you'd like it, I'll send a v2 that converts both the mutexes and the spinlocks in drm_gem.c consistently. If you'd prefer not to take cleanup-only churn, I'm happy to drop it — no problem either way.

I'll let the maintainers of that code decide (I'm not one of them).

Signature · 3 lines
Regards,

Laurent Pinchart
#4 lore raw
In reply to Biren Pandya

Hi

Am 16.06.26 um 20:19 schrieb Biren Pandya:

drivers/gpu/drm/drm_gem.c int drm_gem_change_handle_ioctl(struct drm_device *dev, void *data,

Quoted 55 lines

Several GEM core functions manually managed mutex_lock() and mutex_unlock() over single scopes or error paths. This adds boilerplate and carries the risk of lock leaks if error paths are refactored.

Modernize these locks by deploying the <linux/cleanup.h> scoped_guard() macro. This ensures that the locks are reliably dropped when the block exits, cleanly removing goto out_unlock paths and tightening the lifecycle.

Compiled locally, but requires IGT validation by the DRM CI.

---
  drivers/gpu/drm/drm_gem.c | 66 ++++++++++++++++++---------------------
  1 file changed, 30 insertions(+), 36 deletions(-)
drivers/gpu/drm/drm_gem.c +8 -11
index 891c3bff5ae0..d3a061d42ba7 100644
@@ -346,13 +346,13 @@ void drm_gem_object_handle_put_unlocked(struct drm_gem_object *obj)
  	 * we checked for a name.
  	 */
 
-	mutex_lock(&dev->object_name_lock);
-	if (--obj->handle_count == 0) {
-		drm_gem_object_handle_free(obj);
-		drm_gem_object_exported_dma_buf_free(obj);
-		final = true;
+	scoped_guard(mutex, &dev->object_name_lock) {
+		if (--obj->handle_count == 0) {
+			drm_gem_object_handle_free(obj);
+			drm_gem_object_exported_dma_buf_free(obj);
+			final = true;
+		}
  	}
-	mutex_unlock(&dev->object_name_lock);
 
  	if (final)
  		drm_gem_object_put(obj);
@@ -374,11 +374,8 @@ drm_gem_object_release_handle(int id, void *ptr, void *data)
  	if (obj->funcs->close)
  		obj->funcs->close(obj, file_priv);
 
-	mutex_lock(&file_priv->prime.lock);
-
-	drm_prime_remove_buf_handle(&file_priv->prime, id);
-
-	mutex_unlock(&file_priv->prime.lock);
+	scoped_guard(mutex, &file_priv->prime.lock)
+		drm_prime_remove_buf_handle(&file_priv->prime, id);
 
  	drm_vma_node_revoke(&obj->vma_node, file_priv);
 
@@ -1021,37 +1018,34 @@ int drm_gem_change_handle_ioctl(struct drm_device *dev, void *data,

There have been a number of revisions to this function recently and you want to make sure that you have the latest.  At least the code in v7.1 looks different from yours.

drivers/gpu/drm/drm_gem.c int drm_gem_change_handle_ioctl(struct drm_device *dev, void *data, · line 1022

Quoted 6 lines
drivers/gpu/drm/drm_gem.c +2 -1
  		goto out;
  	}
 
-	mutex_lock(&file_priv->prime.lock);
+	scoped_guard(mutex, &file_priv->prime.lock) {
+		spin_lock(&file_priv->table_lock);

Could this spin lock also use a guard?

Best regards
Thomas

Quoted 55 lines
drivers/gpu/drm/drm_gem.c +20 -24
+		ret = idr_alloc(&file_priv->object_idr, obj, handle, handle + 1,
+				GFP_NOWAIT);
+		spin_unlock(&file_priv->table_lock);
 
-	spin_lock(&file_priv->table_lock);
-	ret = idr_alloc(&file_priv->object_idr, obj, handle, handle + 1,
-			GFP_NOWAIT);
-	spin_unlock(&file_priv->table_lock);
+		if (ret < 0)
+			break;
 
-	if (ret < 0)
-		goto out_unlock;
+		if (obj->dma_buf) {
+			ret = drm_prime_add_buf_handle(&file_priv->prime, obj->dma_buf,
+						       handle);
+			if (ret < 0) {
+				spin_lock(&file_priv->table_lock);
+				idr_remove(&file_priv->object_idr, handle);
+				spin_unlock(&file_priv->table_lock);
+				break;
+			}
 
-	if (obj->dma_buf) {
-		ret = drm_prime_add_buf_handle(&file_priv->prime, obj->dma_buf,
-					       handle);
-		if (ret < 0) {
-			spin_lock(&file_priv->table_lock);
-			idr_remove(&file_priv->object_idr, handle);
-			spin_unlock(&file_priv->table_lock);
-			goto out_unlock;
+			drm_prime_remove_buf_handle(&file_priv->prime, args->handle);
  		}
 
-		drm_prime_remove_buf_handle(&file_priv->prime, args->handle);
-	}
-
-	ret = 0;
-
-	spin_lock(&file_priv->table_lock);
-	idr_remove(&file_priv->object_idr, args->handle);
-	spin_unlock(&file_priv->table_lock);
+		ret = 0;
 
-out_unlock:
-	mutex_unlock(&file_priv->prime.lock);
+		spin_lock(&file_priv->table_lock);
+		idr_remove(&file_priv->object_idr, args->handle);
+		spin_unlock(&file_priv->table_lock);
+	}
  out:
  	drm_gem_object_put(obj);
 

base-commit: 72c25183cac9bc584c9de21797a5883af44bcc7a

Signature · 6 lines
--
Thomas Zimmermann
Graphics Driver Developer
SUSE Software Solutions Germany GmbH
Frankenstr. 146, 90461 Nürnberg, Germany, www.suse.com
GF: Jochen Jaser, Andrew McDonald, Werner Knoblich, (HRB 36809, AG Nürnberg)