FAILED: patch "[PATCH] drm/amdgpu/userq: actually validate userq BOs" failed to apply to 7.2-stable tree

The patch below does not apply to the 7.2-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 <[email protected]>.

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-7.2.y
git checkout FETCH_HEAD
git cherry-pick -x 32d0771ef4ca69c30a1db3aefa34dd177cd2a047
# <resolve conflicts, build, test, etc.>
git commit -s
git send-email --to '<[email protected]>' --in-reply-to '2026102137-thermos-preaching-3697@gregkh' --subject-prefix 'PATCH 7.2.y' 'HEAD^..'

Possible dependencies:

thanks,

greg k-h

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

From 32d0771ef4ca69c30a1db3aefa34dd177cd2a047 Mon Sep 17 00:00:00 2001
From: Benjamin Cheng <[email protected]>
Date: Fri, 9 Oct 2026 09:31:42 -0400
Subject: [PATCH] drm/amdgpu/userq: actually validate userq BOs
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

This was supposed to loop over the individual evicted list, as the kernel and always_valid evicted lists were already handled in amdgpu_vm_validate.

Without this, BOs that were created and evicted before a userq was made would never end up in an eviction fence rearm.

Also add the missing ref to the BO before drm_exec locks and refs it.

index 9316423f9340..b3dc5dd49a64 100644
@@ -980,19 +980,21 @@ amdgpu_userq_bo_validate(struct amdgpu_device *adev, struct drm_exec *exec,
 {
 	struct ttm_operation_ctx ctx = { false, false };
 	struct amdgpu_bo_va *bo_va;
-	struct amdgpu_bo *bo;
+	struct amdgpu_bo *bo, *tmp;
 	int ret;
 
 	spin_lock(&vm->individual_lock);
-	while (!list_empty(&vm->always_valid.evicted)) {
-		bo_va = list_first_entry(&vm->always_valid.evicted,
+	while (!list_empty(&vm->individual.evicted)) {
+		bo_va = list_first_entry(&vm->individual.evicted,
 					 struct amdgpu_bo_va,
 					 base.vm_status);
+		bo = bo_va->base.bo;
+		tmp = amdgpu_bo_ref(bo);
 		spin_unlock(&vm->individual_lock);
 
-		bo = bo_va->base.bo;
 		ret = drm_exec_prepare_obj(exec, &bo->tbo.base,
 					   TTM_NUM_MOVE_FENCES + 1);
+		amdgpu_bo_unref(&tmp);
 		if (unlikely(ret))
 			return ret;
 

From: Prike Liang <[email protected]>

[ Upstream commit 50b24a20be8c0d35f89f7d5c39533f69d84fe48e ]

It fixes both the move overflow and the eviction fence add for evicting these per-vm BOs.

Stable-dep-of: 32d0771ef4ca ("drm/amdgpu/userq: actually validate userq BOs")

---
 drivers/gpu/drm/amd/amdgpu/amdgpu_userq.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

index 0c3fa2d60555a..9352efa98bfa9 100644
@@ -948,7 +948,8 @@ amdgpu_userq_bo_validate(struct amdgpu_device *adev, struct drm_exec *exec,
 		spin_unlock(&vm->individual_lock);
 
 		bo = bo_va->base.bo;
-		ret = drm_exec_prepare_obj(exec, &bo->tbo.base, 2);
+		ret = drm_exec_prepare_obj(exec, &bo->tbo.base,
+					   TTM_NUM_MOVE_FENCES + 1);
 		if (unlikely(ret))
 			return ret;
 

2.53.0

From: Benjamin Cheng <[email protected]>

[ Upstream commit 32d0771ef4ca69c30a1db3aefa34dd177cd2a047 ]

This was supposed to loop over the individual evicted list, as the kernel and always_valid evicted lists were already handled in amdgpu_vm_validate.

Without this, BOs that were created and evicted before a userq was made would never end up in an eviction fence rearm.

Also add the missing ref to the BO before drm_exec locks and refs it.

---
 drivers/gpu/drm/amd/amdgpu/amdgpu_userq.c | 10 ++++++----
 1 file changed, 6 insertions(+), 4 deletions(-)

index 9352efa98bfa9..f014faf7eb1cd 100644
@@ -937,19 +937,21 @@ amdgpu_userq_bo_validate(struct amdgpu_device *adev, struct drm_exec *exec,
 {
 	struct ttm_operation_ctx ctx = { false, false };
 	struct amdgpu_bo_va *bo_va;
-	struct amdgpu_bo *bo;
+	struct amdgpu_bo *bo, *tmp;
 	int ret;
 
 	spin_lock(&vm->individual_lock);
-	while (!list_empty(&vm->always_valid.evicted)) {
-		bo_va = list_first_entry(&vm->always_valid.evicted,
+	while (!list_empty(&vm->individual.evicted)) {
+		bo_va = list_first_entry(&vm->individual.evicted,
 					 struct amdgpu_bo_va,
 					 base.vm_status);
+		bo = bo_va->base.bo;
+		tmp = amdgpu_bo_ref(bo);
 		spin_unlock(&vm->individual_lock);
 
-		bo = bo_va->base.bo;
 		ret = drm_exec_prepare_obj(exec, &bo->tbo.base,
 					   TTM_NUM_MOVE_FENCES + 1);
+		amdgpu_bo_unref(&tmp);
 		if (unlikely(ret))
 			return ret;
 

2.53.0