Skip to contentSkip to conversation

A couple exec IOCTL fixes

PATCH In review Matthew Brost 2 participants 7 messages Nov 5, 2024

Subsystems (2)
lore
Share & tools
lore

Matthew Brost

author maintainer
#0 lore raw

cover A couple exec IOCTL fixes

Noticed a couple of bug in exec IOCTL while working on something else.

Matt

Matthew Brost (2):
drm/xe: Fix possible exec queue leak in exec IOCTL
drm/xe: Drop VM dma-resv lock on xe_sync_in_fence_get failure in exec
IOCTL

 drivers/gpu/drm/xe/xe_exec.c | 13 +++++++++----
 1 file changed, 9 insertions(+), 4 deletions(-)
Signature · 1 line
2.34.1
author maintainer
#1 lore raw

PATCH 1/2 drm/xe: Fix possible exec queue leak in exec IOCTL

In a couple of places after an exec queue is looked up the exec IOCTL returns on input errors without dropping the exec queue ref. Fix this ensuring the exec queue ref is dropped on input error.

Fixes dd08ebf6c352 ("drm/xe: Introduce a new DRM driver for Intel GPUs")
Cc stable <[email protected]>
---
 drivers/gpu/drm/xe/xe_exec.c | 12 ++++++++----
 1 file changed, 8 insertions(+), 4 deletions(-)
drivers/gpu/drm/xe/xe_exec.c +8 -4
index f23ac1e2ed88..6de12f91b865 100644
@@ -132,12 +132,16 @@ int xe_exec_ioctl(struct drm_device *dev, void *data, struct drm_file *file)
 	if (XE_IOCTL_DBG(xe, !q))
 		return -ENOENT;
 
-	if (XE_IOCTL_DBG(xe, q->flags & EXEC_QUEUE_FLAG_VM))
-		return -EINVAL;
+	if (XE_IOCTL_DBG(xe, q->flags & EXEC_QUEUE_FLAG_VM)) {
+		err = -EINVAL;
+		goto err_exec_queue;
+	}
 
 	if (XE_IOCTL_DBG(xe, args->num_batch_buffer &&
-			 q->width != args->num_batch_buffer))
-		return -EINVAL;
+			 q->width != args->num_batch_buffer)) {
+		err = -EINVAL;
+		goto err_exec_queue;
+	}
 
 	if (XE_IOCTL_DBG(xe, q->ops->reset_status(q))) {
 		err = -ECANCELED;
Signature · 1 line
2.34.1
maintainer Reviewed-by on v1 1/2
#2 lore raw

Re: [PATCH 1/2] drm/xe: Fix possible exec queue leak in exec IOCTL

Quoted 23 lines · Matthew Brost

In a couple of places after an exec queue is looked up the exec IOCTL returns on input errors without dropping the exec queue ref. Fix this ensuring the exec queue ref is dropped on input error.

Fixes dd08ebf6c352 ("drm/xe: Introduce a new DRM driver for Intel GPUs")
Cc stable <[email protected]>
---
 drivers/gpu/drm/xe/xe_exec.c | 12 ++++++++----
 1 file changed, 8 insertions(+), 4 deletions(-)
drivers/gpu/drm/xe/xe_exec.c +1 -2
index f23ac1e2ed88..6de12f91b865 100644
@@ -132,12 +132,16 @@ int xe_exec_ioctl(struct drm_device *dev, void *data, struct drm_file *file)
 	if (XE_IOCTL_DBG(xe, !q))
 		return -ENOENT;
 
-	if (XE_IOCTL_DBG(xe, q->flags & EXEC_QUEUE_FLAG_VM))
-		return -EINVAL;
+	if (XE_IOCTL_DBG(xe, q->flags & EXEC_QUEUE_FLAG_VM)) {

drivers/gpu/drm/xe/xe_exec.c int xe_exec_ioctl(struct drm_device *dev, void *data, struct drm_file *file) · line 143

Quoted 10 lines
drivers/gpu/drm/xe/xe_exec.c +6 -2
+		err = -EINVAL;
+		goto err_exec_queue;
+	}
 
 	if (XE_IOCTL_DBG(xe, args->num_batch_buffer &&
-			 q->width != args->num_batch_buffer))
-		return -EINVAL;
+			 q->width != args->num_batch_buffer)) {
+		err = -EINVAL;
+		goto err_exec_queue;

this function is so huge... it probably deserves some splits... but the patch is correct

Quoted 7 lines
drivers/gpu/drm/xe/xe_exec.c +1 -0
+	}
 
 	if (XE_IOCTL_DBG(xe, q->ops->reset_status(q))) {
 		err = -ECANCELED;
Signature · 1 line
2.34.1
author maintainer
#3 lore raw

Re: [PATCH 1/2] drm/xe: Fix possible exec queue leak in exec IOCTL

Quoted 28 lines · Rodrigo Vivi
Quoted 27 lines · Matthew Brost

In a couple of places after an exec queue is looked up the exec IOCTL returns on input errors without dropping the exec queue ref. Fix this ensuring the exec queue ref is dropped on input error.

Fixes dd08ebf6c352 ("drm/xe: Introduce a new DRM driver for Intel GPUs")
Cc stable <[email protected]>
---
 drivers/gpu/drm/xe/xe_exec.c | 12 ++++++++----
 1 file changed, 8 insertions(+), 4 deletions(-)
drivers/gpu/drm/xe/xe_exec.c +4 -2
index f23ac1e2ed88..6de12f91b865 100644
@@ -132,12 +132,16 @@ int xe_exec_ioctl(struct drm_device *dev, void *data, struct drm_file *file)
 	if (XE_IOCTL_DBG(xe, !q))
 		return -ENOENT;
 
-	if (XE_IOCTL_DBG(xe, q->flags & EXEC_QUEUE_FLAG_VM))
-		return -EINVAL;
+	if (XE_IOCTL_DBG(xe, q->flags & EXEC_QUEUE_FLAG_VM)) {
+		err = -EINVAL;
+		goto err_exec_queue;
+	}
 
Quoted 10 lines

drivers/gpu/drm/xe/xe_exec.c int xe_exec_ioctl(struct drm_device *dev, void *data, struct drm_file *file) · line 143

Quoted 6 lines
drivers/gpu/drm/xe/xe_exec.c +3 -2
 	if (XE_IOCTL_DBG(xe, args->num_batch_buffer &&
-			 q->width != args->num_batch_buffer))
-		return -EINVAL;
+			 q->width != args->num_batch_buffer)) {
+		err = -EINVAL;
+		goto err_exec_queue;

this function is so huge... it probably deserves some splits... but the patch is correct

Thanks for the review. Yes, big function. Let me look at this, I see a couple of logical places to split.

But I think in general our exec IOCTL is pretty tidy...

mbrost@lstrano-desk:xe$ wc xe_exec.c
  354  1320 10399 xe_exec.c
mbrost@lstrano-desk:xe$ wc ../i915/gem/i915_gem_execbuffer.c
 3645 11483 95597 ../i915/gem/i915_gem_execbuffer.c

Matt

Quoted 9 lines
Quoted 7 lines
drivers/gpu/drm/xe/xe_exec.c +1 -0
+	}
 
 	if (XE_IOCTL_DBG(xe, q->ops->reset_status(q))) {
 		err = -ECANCELED;
Signature · 1 line
2.34.1
author maintainer
#4 lore raw

PATCH v2 1/1 drm/xe: Restore system memory GGTT mappings

In reply to Matthew Brost

GGTT mappings reside on the device and this state is lost during suspend / d3cold thus this state must be restored resume regardless if the BO is in system memory or VRAM.

---
 drivers/gpu/drm/xe/xe_bo.c       | 14 +++++++++++---
 drivers/gpu/drm/xe/xe_bo_evict.c |  1 -
 2 files changed, 11 insertions(+), 4 deletions(-)
drivers/gpu/drm/xe/xe_bo.c +11 -3
index 5b232f2951b1..d79d8ef5c7d5 100644
@@ -888,8 +888,8 @@ int xe_bo_evict_pinned(struct xe_bo *bo)
 	if (WARN_ON(!xe_bo_is_pinned(bo)))
 		return -EINVAL;
 
-	if (WARN_ON(!xe_bo_is_vram(bo)))
-		return -EINVAL;
+	if (!xe_bo_is_vram(bo))
+		return 0;
 
 	ret = ttm_bo_mem_space(&bo->ttm, &placement, &new_mem, &ctx);
 	if (ret)
@@ -939,6 +939,7 @@ int xe_bo_restore_pinned(struct xe_bo *bo)
 		.interruptible = false,
 	};
 	struct ttm_resource *new_mem;
+	struct ttm_place *place = &(bo->placements[0]);
 	int ret;
 
 	xe_bo_assert_held(bo);
@@ -952,6 +953,9 @@ int xe_bo_restore_pinned(struct xe_bo *bo)
 	if (WARN_ON(xe_bo_is_vram(bo) || !bo->ttm.ttm))
 		return -EINVAL;
 
+	if (!mem_type_is_vram(place->mem_type))
+		return 0;
+
 	ret = ttm_bo_mem_space(&bo->ttm, &bo->placement, &new_mem, &ctx);
 	if (ret)
 		return ret;
@@ -1804,7 +1808,10 @@ int xe_bo_pin(struct xe_bo *bo)
 			place->fpfn = (xe_bo_addr(bo, 0, PAGE_SIZE) -
 				       vram_region_gpu_offset(bo->ttm.resource)) >> PAGE_SHIFT;
 			place->lpfn = place->fpfn + (bo->size >> PAGE_SHIFT);
+		}
 
+		if (mem_type_is_vram(place->mem_type) ||
+		    bo->flags & XE_BO_FLAG_GGTT) {
 			spin_lock(&xe->pinned.lock);
 			list_add_tail(&bo->pinned_link, &xe->pinned.kernel_bo_present);
 			spin_unlock(&xe->pinned.lock);
@@ -1865,7 +1872,8 @@ void xe_bo_unpin(struct xe_bo *bo)
 	    bo->flags & XE_BO_FLAG_INTERNAL_TEST)) {
 		struct ttm_place *place = &(bo->placements[0]);
 
-		if (mem_type_is_vram(place->mem_type)) {
+		if (mem_type_is_vram(place->mem_type) ||
+		    bo->flags & XE_BO_FLAG_GGTT) {
 			spin_lock(&xe->pinned.lock);
 			xe_assert(xe, !list_empty(&bo->pinned_link));
 			list_del_init(&bo->pinned_link);
drivers/gpu/drm/xe/xe_bo_evict.c +0 -1
index 541b49007d73..32043e1e5a86 100644
@@ -159,7 +159,6 @@ int xe_bo_restore_kernel(struct xe_device *xe)
 		 * should setup the iosys map.
 		 */
 		xe_assert(xe, !iosys_map_is_null(&bo->vmap));
-		xe_assert(xe, xe_bo_is_vram(bo));
 
 		xe_bo_put(bo);
 
Signature · 1 line
2.34.1
author maintainer
#5 lore raw

PATCH 2/2 drm/xe: Drop VM dma-resv lock on xe_sync_in_fence_get failure in exec IOCTL

In reply to Matthew Brost

Upon failure all locks need to be dropped before returning to the user.

Fixes 58480c1c912f ("drm/xe: Skip VMAs pin when requesting signal to the last XE_EXEC")
Cc stable <[email protected]>
---
 drivers/gpu/drm/xe/xe_exec.c | 1 +
 1 file changed, 1 insertion(+)
drivers/gpu/drm/xe/xe_exec.c +1 -0
index 6de12f91b865..756b492f13b0 100644
@@ -224,6 +224,7 @@ int xe_exec_ioctl(struct drm_device *dev, void *data, struct drm_file *file)
 			fence = xe_sync_in_fence_get(syncs, num_syncs, q, vm);
 			if (IS_ERR(fence)) {
 				err = PTR_ERR(fence);
+				xe_vm_unlock(vm);
 				goto err_unlock_list;
 			}
 			for (i = 0; i < num_syncs; i++)
Signature · 1 line
2.34.1
maintainer Reviewed-by on v1 2/2
#6 lore raw

Re: [PATCH 2/2] drm/xe: Drop VM dma-resv lock on xe_sync_in_fence_get failure in exec IOCTL

Quoted 5 lines · Matthew Brost

Upon failure all locks need to be dropped before returning to the user.

Fixes 58480c1c912f ("drm/xe: Skip VMAs pin when requesting signal to the last XE_EXEC")
Cc stable <[email protected]>
Quoted 19 lines
---
 drivers/gpu/drm/xe/xe_exec.c | 1 +
 1 file changed, 1 insertion(+)
drivers/gpu/drm/xe/xe_exec.c +1 -0
index 6de12f91b865..756b492f13b0 100644
@@ -224,6 +224,7 @@ int xe_exec_ioctl(struct drm_device *dev, void *data, struct drm_file *file)
 			fence = xe_sync_in_fence_get(syncs, num_syncs, q, vm);
 			if (IS_ERR(fence)) {
 				err = PTR_ERR(fence);
+				xe_vm_unlock(vm);
 				goto err_unlock_list;
 			}
 			for (i = 0; i < num_syncs; i++)
Signature · 1 line
2.34.1