FAILED: patch "[PATCH] drm/xe: Wait on external BO kernel fences in exec IOCTL" failed to apply to 6.12-stable tree

FAILED: patch "[PATCH] drm/xe: Wait on external BO kernel fences in exec IOCTL" failed to apply to 6.12-stable tree

[gregkh]

The patch below does not apply to the 6.12-stable tree.
If someone wants it applied there, or to any other stable or longterm
tree, then please email the backport, including the original git commit
id to <stable@vger.kernel.org>.

To reproduce the conflict and resubmit, you may use the following commands:

git fetch https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/ linux-6.12.y
git checkout FETCH_HEAD
git cherry-pick -x af80e2bfde9312c76b60cf9274248dce0410b30d
# <resolve conflicts, build, test, etc.>
git commit -s
git send-email --to '<stable@vger.kernel.org>' --in-reply-to '2026072911-surplus-properly-117b@gregkh' --subject-prefix 'PATCH 6.12.y' 'HEAD^..'

Possible dependencies:



thanks,

greg k-h

------------------ original commit in Linus's tree ------------------

From af80e2bfde9312c76b60cf9274248dce0410b30d Mon Sep 17 00:00:00 2001
From: Matthew Brost <matthew.brost@intel.com>
Date: Thu, 2 Jul 2026 14:58:05 -0700
Subject: [PATCH] drm/xe: Wait on external BO kernel fences in exec IOCTL
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Before arming a user job, xe_exec_ioctl() only added the VM's
dma-resv KERNEL slot as a dependency. That slot covers rebinds and
the kernel operations of the VM's private BOs, but not external BOs
(bo->vm == NULL), which carry their kernel operations (evictions,
moves, ...) in their own dma-resv KERNEL slot.

The DMA_RESV_USAGE_KERNEL slot is the cross-driver contract for
memory management operations that must complete before the BO or its
backing store may be used: any accessor is required to wait on the
KERNEL fences before touching the resv. By skipping the external BOs'
KERNEL slots, the exec path violated that contract and could schedule
a user job while a kernel operation on an external BO mapped by the VM
was still in flight, racing against it and potentially reading or
writing memory that was being moved.

Replace the VM-only dependency with an iteration over every object
locked by the exec, adding each object's KERNEL slot as a job
dependency. This covers the VM resv (rebinds and private BOs) as well
as every external BO, mirroring the drm_gpuvm_resv_add_fence() call
that later publishes the job fence to the same set of objects.
Long-running mode continues to skip this, as before.

Fixes: dd08ebf6c352 ("drm/xe: Introduce a new DRM driver for Intel GPUs")
Cc: stable@vger.kernel.org
Assisted-by: GitHub_Copilot:claude-opus-4.8
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patch.msgid.link/20260702215805.4011228-1-matthew.brost@intel.com
(cherry picked from commit a6b842acf3ddd1efc53a56de9260cfa718fb35e7)
Signed-off-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>

diff --git a/drivers/gpu/drm/xe/xe_exec.c b/drivers/gpu/drm/xe/xe_exec.c
index e05dabfcd43c..d5293bc33a67 100644
--- a/drivers/gpu/drm/xe/xe_exec.c
+++ b/drivers/gpu/drm/xe/xe_exec.c
@@ -292,13 +292,23 @@ int xe_exec_ioctl(struct drm_device *dev, void *data, struct drm_file *file)
 		goto err_exec;
 	}
 
-	/* Wait behind rebinds */
+	/*
+	 * Wait behind rebinds and any kernel operations (evictions, defrag
+	 * moves, ...) on the VM and all external BOs. The VM's private BOs
+	 * carry their kernel ops in the VM dma-resv KERNEL slot, while each
+	 * external BO carries them in its own dma-resv KERNEL slot; both are
+	 * covered by iterating every object locked by the exec, mirroring the
+	 * drm_gpuvm_resv_add_fence() below.
+	 */
 	if (!xe_vm_in_lr_mode(vm)) {
-		err = xe_sched_job_add_deps(job,
-					    xe_vm_resv(vm),
-					    DMA_RESV_USAGE_KERNEL);
-		if (err)
-			goto err_put_job;
+		struct drm_gem_object *obj;
+
+		drm_exec_for_each_locked_object(exec, obj) {
+			err = xe_sched_job_add_deps(job, obj->resv,
+						    DMA_RESV_USAGE_KERNEL);
+			if (err)
+				goto err_put_job;
+		}
 	}
 
 	for (i = 0; i < num_syncs && !err; i++)

[PATCH 6.12.y 1/2] drm/exec: Remove the index parameter from drm_exec_for_each_locked_obj[_reverse]

[Sasha Levin]

From: Thomas Hellström <thomas.hellstrom@linux.intel.com>

[ Upstream commit ce44b78512e9102aea54ff6b6e521d6c8de9f31c ]

Nobody makes any use of it. Possible internal future users can
instead use the _index variable. External users shouldn't use
it since the array it's pointing into is internal drm_exec state.

v2:
- Use a unique id for the loop variable (Christian)

Assisted-by: GitHub Copilot:claude-sonnet-4.6
Signed-off-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Link: https://patch.msgid.link/20260520101616.41284-2-thomas.hellstrom@linux.intel.com
Stable-dep-of: af80e2bfde93 ("drm/xe: Wait on external BO kernel fences in exec IOCTL")
Signed-off-by: Sasha Levin <sashal@kernel.org>
---
 drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c |  9 +++------
 drivers/gpu/drm/drm_exec.c             |  6 ++----
 drivers/gpu/drm/drm_gpuvm.c            |  3 +--
 drivers/gpu/drm/xe/xe_vm.c             |  3 +--
 include/drm/drm_exec.h                 | 20 ++++++++++++--------
 5 files changed, 19 insertions(+), 22 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
index e018a807e4e3..eb94e7891c2a 100644
--- a/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
+++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_cs.c
@@ -843,7 +843,6 @@ static int amdgpu_cs_parser_bos(struct amdgpu_cs_parser *p,
 	struct amdgpu_vm *vm = &fpriv->vm;
 	struct amdgpu_bo_list_entry *e;
 	struct drm_gem_object *obj;
-	unsigned long index;
 	unsigned int i;
 	int r;
 
@@ -964,7 +963,7 @@ static int amdgpu_cs_parser_bos(struct amdgpu_cs_parser *p,
 		goto out_free_user_pages;
 	}
 
-	drm_exec_for_each_locked_object(&p->exec, index, obj) {
+	drm_exec_for_each_locked_object(&p->exec, obj) {
 		r = amdgpu_cs_bo_validate(p, gem_to_amdgpu_bo(obj));
 		if (unlikely(r))
 			goto out_free_user_pages;
@@ -1202,7 +1201,6 @@ static int amdgpu_cs_sync_rings(struct amdgpu_cs_parser *p)
 	struct drm_gpu_scheduler *sched;
 	struct drm_gem_object *obj;
 	struct dma_fence *fence;
-	unsigned long index;
 	unsigned int i;
 	int r;
 
@@ -1213,7 +1211,7 @@ static int amdgpu_cs_sync_rings(struct amdgpu_cs_parser *p)
 		return r;
 	}
 
-	drm_exec_for_each_locked_object(&p->exec, index, obj) {
+	drm_exec_for_each_locked_object(&p->exec, obj) {
 		struct amdgpu_bo *bo = gem_to_amdgpu_bo(obj);
 
 		struct dma_resv *resv = bo->tbo.base.resv;
@@ -1281,7 +1279,6 @@ static int amdgpu_cs_submit(struct amdgpu_cs_parser *p,
 	struct amdgpu_vm *vm = &fpriv->vm;
 	struct amdgpu_bo_list_entry *e;
 	struct drm_gem_object *gobj;
-	unsigned long index;
 	unsigned int i;
 	uint64_t seq;
 	int r;
@@ -1332,7 +1329,7 @@ static int amdgpu_cs_submit(struct amdgpu_cs_parser *p,
 	}
 
 	p->fence = dma_fence_get(&leader->base.s_fence->finished);
-	drm_exec_for_each_locked_object(&p->exec, index, gobj) {
+	drm_exec_for_each_locked_object(&p->exec, gobj) {
 
 		ttm_bo_move_to_lru_tail_unlocked(&gem_to_amdgpu_bo(gobj)->tbo);
 
diff --git a/drivers/gpu/drm/drm_exec.c b/drivers/gpu/drm/drm_exec.c
index 18e366cc4993..6c1351cd7f15 100644
--- a/drivers/gpu/drm/drm_exec.c
+++ b/drivers/gpu/drm/drm_exec.c
@@ -22,7 +22,6 @@
  *
  *	struct drm_gem_object *obj;
  *	struct drm_exec exec;
- *	unsigned long index;
  *	int ret;
  *
  *	drm_exec_init(&exec, DRM_EXEC_INTERRUPTIBLE_WAIT);
@@ -38,7 +37,7 @@
  *			goto error;
  *	}
  *
- *	drm_exec_for_each_locked_object(&exec, index, obj) {
+ *	drm_exec_for_each_locked_object(&exec, obj) {
  *		dma_resv_add_fence(obj->resv, fence, DMA_RESV_USAGE_READ);
  *		...
  *	}
@@ -54,9 +53,8 @@
 static void drm_exec_unlock_all(struct drm_exec *exec)
 {
 	struct drm_gem_object *obj;
-	unsigned long index;
 
-	drm_exec_for_each_locked_object_reverse(exec, index, obj) {
+	drm_exec_for_each_locked_object_reverse(exec, obj) {
 		dma_resv_unlock(obj->resv);
 		drm_gem_object_put(obj);
 	}
diff --git a/drivers/gpu/drm/drm_gpuvm.c b/drivers/gpu/drm/drm_gpuvm.c
index 2515cff33485..338a21a58ec5 100644
--- a/drivers/gpu/drm/drm_gpuvm.c
+++ b/drivers/gpu/drm/drm_gpuvm.c
@@ -1451,9 +1451,8 @@ drm_gpuvm_resv_add_fence(struct drm_gpuvm *gpuvm,
 			 enum dma_resv_usage extobj_usage)
 {
 	struct drm_gem_object *obj;
-	unsigned long index;
 
-	drm_exec_for_each_locked_object(exec, index, obj) {
+	drm_exec_for_each_locked_object(exec, obj) {
 		dma_resv_assert_held(obj->resv);
 		dma_resv_add_fence(obj->resv, fence,
 				   drm_gpuvm_is_extobj(gpuvm, obj) ?
diff --git a/drivers/gpu/drm/xe/xe_vm.c b/drivers/gpu/drm/xe/xe_vm.c
index 7ce6a9d4e5c1..deb4d0314db3 100644
--- a/drivers/gpu/drm/xe/xe_vm.c
+++ b/drivers/gpu/drm/xe/xe_vm.c
@@ -418,7 +418,6 @@ int xe_vm_validate_rebind(struct xe_vm *vm, struct drm_exec *exec,
 			  unsigned int num_fences)
 {
 	struct drm_gem_object *obj;
-	unsigned long index;
 	int ret;
 
 	do {
@@ -431,7 +430,7 @@ int xe_vm_validate_rebind(struct xe_vm *vm, struct drm_exec *exec,
 			return ret;
 	} while (!list_empty(&vm->gpuvm.evict.list));
 
-	drm_exec_for_each_locked_object(exec, index, obj) {
+	drm_exec_for_each_locked_object(exec, obj) {
 		ret = dma_resv_reserve_fences(obj->resv, num_fences);
 		if (ret)
 			return ret;
diff --git a/include/drm/drm_exec.h b/include/drm/drm_exec.h
index aa786b828a0a..dee6ebdbe416 100644
--- a/include/drm/drm_exec.h
+++ b/include/drm/drm_exec.h
@@ -65,31 +65,35 @@ drm_exec_obj(struct drm_exec *exec, unsigned long index)
 	return index < exec->num_objects ? exec->objects[index] : NULL;
 }
 
+/* Helper for drm_exec_for_each_locked_object(). Internal use only. */
+#define __drm_exec_for_each_locked_object(exec, obj, __index)		\
+	for (unsigned long __index = 0; ((obj) = drm_exec_obj(exec, __index)); ++__index)
 /**
  * drm_exec_for_each_locked_object - iterate over all the locked objects
  * @exec: drm_exec object
- * @index: unsigned long index for the iteration
  * @obj: the current GEM object
  *
  * Iterate over all the locked GEM objects inside the drm_exec object.
  */
-#define drm_exec_for_each_locked_object(exec, index, obj)		\
-	for ((index) = 0; ((obj) = drm_exec_obj(exec, index)); ++(index))
+#define drm_exec_for_each_locked_object(exec, obj)			\
+	__drm_exec_for_each_locked_object(exec, obj, __UNIQUE_ID(drm_exec))
 
+/* Helper for drm_exec_for_each_locked_object_reverse(). Internal use only. */
+#define __drm_exec_for_each_locked_object_reverse(exec, obj, __index)	\
+	for (unsigned long __index = (exec)->num_objects - 1;		\
+	     ((obj) = drm_exec_obj(exec, __index)); --__index)
 /**
  * drm_exec_for_each_locked_object_reverse - iterate over all the locked
  * objects in reverse locking order
  * @exec: drm_exec object
- * @index: unsigned long index for the iteration
  * @obj: the current GEM object
  *
  * Iterate over all the locked GEM objects inside the drm_exec object in
- * reverse locking order. Note that @index may go below zero and wrap,
+ * reverse locking order. Note that the internal index may wrap around,
  * but that will be caught by drm_exec_obj(), returning a NULL object.
  */
-#define drm_exec_for_each_locked_object_reverse(exec, index, obj)	\
-	for ((index) = (exec)->num_objects - 1;				\
-	     ((obj) = drm_exec_obj(exec, index)); --(index))
+#define drm_exec_for_each_locked_object_reverse(exec, obj)		\
+	__drm_exec_for_each_locked_object_reverse(exec, obj, __UNIQUE_ID(drm_exec))
 
 /**
  * drm_exec_until_all_locked - loop until all GEM objects are locked
-- 
2.53.0

[PATCH 6.12.y 2/2] drm/xe: Wait on external BO kernel fences in exec IOCTL

[Sasha Levin]

From: Matthew Brost <matthew.brost@intel.com>

[ Upstream commit af80e2bfde9312c76b60cf9274248dce0410b30d ]

Before arming a user job, xe_exec_ioctl() only added the VM's
dma-resv KERNEL slot as a dependency. That slot covers rebinds and
the kernel operations of the VM's private BOs, but not external BOs
(bo->vm == NULL), which carry their kernel operations (evictions,
moves, ...) in their own dma-resv KERNEL slot.

The DMA_RESV_USAGE_KERNEL slot is the cross-driver contract for
memory management operations that must complete before the BO or its
backing store may be used: any accessor is required to wait on the
KERNEL fences before touching the resv. By skipping the external BOs'
KERNEL slots, the exec path violated that contract and could schedule
a user job while a kernel operation on an external BO mapped by the VM
was still in flight, racing against it and potentially reading or
writing memory that was being moved.

Replace the VM-only dependency with an iteration over every object
locked by the exec, adding each object's KERNEL slot as a job
dependency. This covers the VM resv (rebinds and private BOs) as well
as every external BO, mirroring the drm_gpuvm_resv_add_fence() call
that later publishes the job fence to the same set of objects.
Long-running mode continues to skip this, as before.

Fixes: dd08ebf6c352 ("drm/xe: Introduce a new DRM driver for Intel GPUs")
Cc: stable@vger.kernel.org
Assisted-by: GitHub_Copilot:claude-opus-4.8
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patch.msgid.link/20260702215805.4011228-1-matthew.brost@intel.com
(cherry picked from commit a6b842acf3ddd1efc53a56de9260cfa718fb35e7)
Signed-off-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
---
 drivers/gpu/drm/xe/xe_exec.c | 22 ++++++++++++++++------
 1 file changed, 16 insertions(+), 6 deletions(-)

diff --git a/drivers/gpu/drm/xe/xe_exec.c b/drivers/gpu/drm/xe/xe_exec.c
index 886d03ccf744..04a88c16d835 100644
--- a/drivers/gpu/drm/xe/xe_exec.c
+++ b/drivers/gpu/drm/xe/xe_exec.c
@@ -270,13 +270,23 @@ int xe_exec_ioctl(struct drm_device *dev, void *data, struct drm_file *file)
 		goto err_exec;
 	}
 
-	/* Wait behind rebinds */
+	/*
+	 * Wait behind rebinds and any kernel operations (evictions, defrag
+	 * moves, ...) on the VM and all external BOs. The VM's private BOs
+	 * carry their kernel ops in the VM dma-resv KERNEL slot, while each
+	 * external BO carries them in its own dma-resv KERNEL slot; both are
+	 * covered by iterating every object locked by the exec, mirroring the
+	 * drm_gpuvm_resv_add_fence() below.
+	 */
 	if (!xe_vm_in_lr_mode(vm)) {
-		err = xe_sched_job_add_deps(job,
-					    xe_vm_resv(vm),
-					    DMA_RESV_USAGE_KERNEL);
-		if (err)
-			goto err_put_job;
+		struct drm_gem_object *obj;
+
+		drm_exec_for_each_locked_object(exec, obj) {
+			err = xe_sched_job_add_deps(job, obj->resv,
+						    DMA_RESV_USAGE_KERNEL);
+			if (err)
+				goto err_put_job;
+		}
 	}
 
 	for (i = 0; i < num_syncs && !err; i++)
-- 
2.53.0