drm/xe: Fix fault mode invalidation with unbind

Fix fault mode invalidation racing with unbind leading to the
PTE zapping potentially traversing an invalid page-table tree.
Do this by holding the notifier lock across PTE zapping. This
might transfer any contention waiting on the notifier seqlock
read side to the notifier lock read side, but that shouldn't be
a major problem.

At the same time get rid of the open-coded invalidation in the bind
code by relying on the notifier even when the vma bind is not
yet committed.

Finally let userptr invalidation call a dedicated xe_vm function
performing a full invalidation.

Fixes: e8babb2 ("drm/xe: Convert multiple bind ops into single job")
Cc: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Cc: Matthew Brost <matthew.brost@intel.com>
Cc: Matthew Auld <matthew.auld@intel.com>
Cc: <stable@vger.kernel.org> # v6.12+
Signed-off-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Reviewed-by: Matthew Brost <matthew.brost@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20250228073058.59510-4-thomas.hellstrom@linux.intel.com
(cherry picked from commit 100a5b8)
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>

Author: Thomas Hellström

drivers/gpu/drm/xe/xe_pt.c

@@ -1213,42 +1213,22 @@ static int vma_check_userptr(struct xe_vm *vm, struct xe_vma *vma,
 		return 0;
 
 	uvma = to_userptr_vma(vma);
-	notifier_seq = uvma->userptr.notifier_seq;
+	if (xe_pt_userptr_inject_eagain(uvma))
+		xe_vma_userptr_force_invalidate(uvma);
 
-	if (uvma->userptr.initial_bind && !xe_vm_in_fault_mode(vm))
-		return 0;
+	notifier_seq = uvma->userptr.notifier_seq;
 
 	if (!mmu_interval_read_retry(&uvma->userptr.notifier,
-				     notifier_seq) &&
-	    !xe_pt_userptr_inject_eagain(uvma))
+				     notifier_seq))
 		return 0;
 
-	if (xe_vm_in_fault_mode(vm)) {
+	if (xe_vm_in_fault_mode(vm))
 		return -EAGAIN;
-	} else {
-		spin_lock(&vm->userptr.invalidated_lock);
-		list_move_tail(&uvma->userptr.invalidate_link,
-			       &vm->userptr.invalidated);
-		spin_unlock(&vm->userptr.invalidated_lock);
-
-		if (xe_vm_in_preempt_fence_mode(vm)) {
-			struct dma_resv_iter cursor;
-			struct dma_fence *fence;
-			long err;
-
-			dma_resv_iter_begin(&cursor, xe_vm_resv(vm),
-					    DMA_RESV_USAGE_BOOKKEEP);
-			dma_resv_for_each_fence_unlocked(&cursor, fence)
-				dma_fence_enable_sw_signaling(fence);
-			dma_resv_iter_end(&cursor);
-
-			err = dma_resv_wait_timeout(xe_vm_resv(vm),
-						    DMA_RESV_USAGE_BOOKKEEP,
-						    false, MAX_SCHEDULE_TIMEOUT);
-			XE_WARN_ON(err <= 0);
-		}
-	}
 
+	/*
+	 * Just continue the operation since exec or rebind worker
+	 * will take care of rebinding.
+	 */
 	return 0;
 }
 

drivers/gpu/drm/xe/xe_vm.c

@@ -579,51 +579,26 @@ static void preempt_rebind_work_func(struct work_struct *w)
 	trace_xe_vm_rebind_worker_exit(vm);
 }
 
-static bool vma_userptr_invalidate(struct mmu_interval_notifier *mni,
-				   const struct mmu_notifier_range *range,
-				   unsigned long cur_seq)
+static void __vma_userptr_invalidate(struct xe_vm *vm, struct xe_userptr_vma *uvma)
 {
-	struct xe_userptr *userptr = container_of(mni, typeof(*userptr), notifier);
-	struct xe_userptr_vma *uvma = container_of(userptr, typeof(*uvma), userptr);
+	struct xe_userptr *userptr = &uvma->userptr;
 	struct xe_vma *vma = &uvma->vma;
-	struct xe_vm *vm = xe_vma_vm(vma);
 	struct dma_resv_iter cursor;
 	struct dma_fence *fence;
 	long err;
 
-	xe_assert(vm->xe, xe_vma_is_userptr(vma));
-	trace_xe_vma_userptr_invalidate(vma);
-
-	if (!mmu_notifier_range_blockable(range))
-		return false;
-
-	vm_dbg(&xe_vma_vm(vma)->xe->drm,
-	       "NOTIFIER: addr=0x%016llx, range=0x%016llx",
-		xe_vma_start(vma), xe_vma_size(vma));
-
-	down_write(&vm->userptr.notifier_lock);
-	mmu_interval_set_seq(mni, cur_seq);
-
-	/* No need to stop gpu access if the userptr is not yet bound. */
-	if (!userptr->initial_bind) {
-		up_write(&vm->userptr.notifier_lock);
-		return true;
-	}
-
 	/*
 	 * Tell exec and rebind worker they need to repin and rebind this
 	 * userptr.
 	 */
 	if (!xe_vm_in_fault_mode(vm) &&
-	    !(vma->gpuva.flags & XE_VMA_DESTROYED) && vma->tile_present) {
+	    !(vma->gpuva.flags & XE_VMA_DESTROYED)) {
 		spin_lock(&vm->userptr.invalidated_lock);
 		list_move_tail(&userptr->invalidate_link,
 			       &vm->userptr.invalidated);
 		spin_unlock(&vm->userptr.invalidated_lock);
 	}
 
-	up_write(&vm->userptr.notifier_lock);
-
 	/*
 	 * Preempt fences turn into schedule disables, pipeline these.
 	 * Note that even in fault mode, we need to wait for binds and
@@ -641,11 +616,35 @@ static bool vma_userptr_invalidate(struct mmu_interval_notifier *mni,
 				    false, MAX_SCHEDULE_TIMEOUT);
 	XE_WARN_ON(err <= 0);
 
-	if (xe_vm_in_fault_mode(vm)) {
+	if (xe_vm_in_fault_mode(vm) && userptr->initial_bind) {
 		err = xe_vm_invalidate_vma(vma);
 		XE_WARN_ON(err);
 	}
+}
 
+static bool vma_userptr_invalidate(struct mmu_interval_notifier *mni,
+				   const struct mmu_notifier_range *range,
+				   unsigned long cur_seq)
+{
+	struct xe_userptr_vma *uvma = container_of(mni, typeof(*uvma), userptr.notifier);
+	struct xe_vma *vma = &uvma->vma;
+	struct xe_vm *vm = xe_vma_vm(vma);
+
+	xe_assert(vm->xe, xe_vma_is_userptr(vma));
+	trace_xe_vma_userptr_invalidate(vma);
+
+	if (!mmu_notifier_range_blockable(range))
+		return false;
+
+	vm_dbg(&xe_vma_vm(vma)->xe->drm,
+	       "NOTIFIER: addr=0x%016llx, range=0x%016llx",
+		xe_vma_start(vma), xe_vma_size(vma));
+
+	down_write(&vm->userptr.notifier_lock);
+	mmu_interval_set_seq(mni, cur_seq);
+
+	__vma_userptr_invalidate(vm, uvma);
+	up_write(&vm->userptr.notifier_lock);
 	trace_xe_vma_userptr_invalidate_complete(vma);
 
 	return true;
@@ -655,6 +654,34 @@ static const struct mmu_interval_notifier_ops vma_userptr_notifier_ops = {
 	.invalidate = vma_userptr_invalidate,
 };
 
+#if IS_ENABLED(CONFIG_DRM_XE_USERPTR_INVAL_INJECT)
+/**
+ * xe_vma_userptr_force_invalidate() - force invalidate a userptr
+ * @uvma: The userptr vma to invalidate
+ *
+ * Perform a forced userptr invalidation for testing purposes.
+ */
+void xe_vma_userptr_force_invalidate(struct xe_userptr_vma *uvma)
+{
+	struct xe_vm *vm = xe_vma_vm(&uvma->vma);
+
+	/* Protect against concurrent userptr pinning */
+	lockdep_assert_held(&vm->lock);
+	/* Protect against concurrent notifiers */
+	lockdep_assert_held(&vm->userptr.notifier_lock);
+	/*
+	 * Protect against concurrent instances of this function and
+	 * the critical exec sections
+	 */
+	xe_vm_assert_held(vm);
+
+	if (!mmu_interval_read_retry(&uvma->userptr.notifier,
+				     uvma->userptr.notifier_seq))
+		uvma->userptr.notifier_seq -= 2;
+	__vma_userptr_invalidate(vm, uvma);
+}
+#endif
+
 int xe_vm_userptr_pin(struct xe_vm *vm)
 {
 	struct xe_userptr_vma *uvma, *next;

drivers/gpu/drm/xe/xe_vm.h

@@ -279,4 +279,12 @@ struct xe_vm_snapshot *xe_vm_snapshot_capture(struct xe_vm *vm);
 void xe_vm_snapshot_capture_delayed(struct xe_vm_snapshot *snap);
 void xe_vm_snapshot_print(struct xe_vm_snapshot *snap, struct drm_printer *p);
 void xe_vm_snapshot_free(struct xe_vm_snapshot *snap);
+
+#if IS_ENABLED(CONFIG_DRM_XE_USERPTR_INVAL_INJECT)
+void xe_vma_userptr_force_invalidate(struct xe_userptr_vma *uvma);
+#else
+static inline void xe_vma_userptr_force_invalidate(struct xe_userptr_vma *uvma)
+{
+}
+#endif
 #endif

drivers/gpu/drm/xe/xe_vm_types.h

@@ -227,8 +227,8 @@ struct xe_vm {
 		 * up for revalidation. Protected from access with the
 		 * @invalidated_lock. Removing items from the list
 		 * additionally requires @lock in write mode, and adding
-		 * items to the list requires the @userptr.notifer_lock in
-		 * write mode.
+		 * items to the list requires either the @userptr.notifer_lock in
+		 * write mode, OR @lock in write mode.
 		 */
 		struct list_head invalidated;
 	} userptr;