drm/i915/dgfx: Grab wakeref at i915_ttm_unmap_virtual

We had already grabbed the rpm wakeref at obj destruction path,
but it also required to grab the wakeref when object moves.
When i915_gem_object_release_mmap_offset() gets called by
i915_ttm_move_notify(), it will release the mmap offset without
grabbing the wakeref. We want to avoid that therefore,
grab the wakeref at i915_ttm_unmap_virtual() accordingly.

While doing that also changed the lmem_userfault_lock from
mutex to spinlock, as spinlock widely used for list.

Also changed if (obj->userfault_count) to
GEM_BUG_ON(!obj->userfault_count).

v2:
- Removed lmem_userfault_{list,lock} from intel_gt. [Matt Auld]

Fixes: ad74457a ("drm/i915/dgfx: Release mmap on rpm suspend")
Suggested-by: Matthew Auld <matthew.auld@intel.com>
Signed-off-by: Anshuman Gupta <anshuman.gupta@intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20221027092242.1476080-3-anshuman.gupta@intel.com

drivers/gpu/drm/i915/gem/i915_gem_mman.c

@@ -557,11 +557,13 @@ void i915_gem_object_runtime_pm_release_mmap_offset(struct drm_i915_gem_object *
 
 	drm_vma_node_unmap(&bo->base.vma_node, bdev->dev_mapping);
 
-	if (obj->userfault_count) {
-		/* rpm wakeref provide exclusive access */
-		list_del(&obj->userfault_link);
-		obj->userfault_count = 0;
-	}
+	/*
+	 * We have exclusive access here via runtime suspend. All other callers
+	 * must first grab the rpm wakeref.
+	 */
+	GEM_BUG_ON(!obj->userfault_count);
+	list_del(&obj->userfault_link);
+	obj->userfault_count = 0;
 }
 
 void i915_gem_object_release_mmap_offset(struct drm_i915_gem_object *obj)
@@ -587,13 +589,6 @@ void i915_gem_object_release_mmap_offset(struct drm_i915_gem_object *obj)
 		spin_lock(&obj->mmo.lock);
 	}
 	spin_unlock(&obj->mmo.lock);
-
-	if (obj->userfault_count) {
-		mutex_lock(&to_i915(obj->base.dev)->runtime_pm.lmem_userfault_lock);
-		list_del(&obj->userfault_link);
-		mutex_unlock(&to_i915(obj->base.dev)->runtime_pm.lmem_userfault_lock);
-		obj->userfault_count = 0;
-	}
 }
 
 static struct i915_mmap_offset *

drivers/gpu/drm/i915/gem/i915_gem_ttm.c

@@ -509,18 +509,9 @@ static int i915_ttm_shrink(struct drm_i915_gem_object *obj, unsigned int flags)
 static void i915_ttm_delete_mem_notify(struct ttm_buffer_object *bo)
 {
 	struct drm_i915_gem_object *obj = i915_ttm_to_gem(bo);
-	intel_wakeref_t wakeref = 0;
 
 	if (bo->resource && !i915_ttm_is_ghost_object(bo)) {
-		/* ttm_bo_release() already has dma_resv_lock */
-		if (i915_ttm_cpu_maps_iomem(bo->resource))
-			wakeref = intel_runtime_pm_get(&to_i915(obj->base.dev)->runtime_pm);
-
 		__i915_gem_object_pages_fini(obj);
-
-		if (wakeref)
-			intel_runtime_pm_put(&to_i915(obj->base.dev)->runtime_pm, wakeref);
-
 		i915_ttm_free_cached_io_rsgt(obj);
 	}
 }
@@ -1098,12 +1089,15 @@ static vm_fault_t vm_fault_ttm(struct vm_fault *vmf)
 	if (ret == VM_FAULT_RETRY && !(vmf->flags & FAULT_FLAG_RETRY_NOWAIT))
 		goto out_rpm;
 
-	/* ttm_bo_vm_reserve() already has dma_resv_lock */
+	/*
+	 * ttm_bo_vm_reserve() already has dma_resv_lock.
+	 * userfault_count is protected by dma_resv lock and rpm wakeref.
+	 */
 	if (ret == VM_FAULT_NOPAGE && wakeref && !obj->userfault_count) {
 		obj->userfault_count = 1;
-		mutex_lock(&to_i915(obj->base.dev)->runtime_pm.lmem_userfault_lock);
+		spin_lock(&to_i915(obj->base.dev)->runtime_pm.lmem_userfault_lock);
 		list_add(&obj->userfault_link, &to_i915(obj->base.dev)->runtime_pm.lmem_userfault_list);
-		mutex_unlock(&to_i915(obj->base.dev)->runtime_pm.lmem_userfault_lock);
+		spin_unlock(&to_i915(obj->base.dev)->runtime_pm.lmem_userfault_lock);
 	}
 
 	if (wakeref & CONFIG_DRM_I915_USERFAULT_AUTOSUSPEND)
@@ -1169,7 +1163,27 @@ static u64 i915_ttm_mmap_offset(struct drm_i915_gem_object *obj)
 
 static void i915_ttm_unmap_virtual(struct drm_i915_gem_object *obj)
 {
+	struct ttm_buffer_object *bo = i915_gem_to_ttm(obj);
+	intel_wakeref_t wakeref = 0;
+
+	assert_object_held_shared(obj);
+
+	if (i915_ttm_cpu_maps_iomem(bo->resource)) {
+		wakeref = intel_runtime_pm_get(&to_i915(obj->base.dev)->runtime_pm);
+
+		/* userfault_count is protected by obj lock and rpm wakeref. */
+		if (obj->userfault_count) {
+			spin_lock(&to_i915(obj->base.dev)->runtime_pm.lmem_userfault_lock);
+			list_del(&obj->userfault_link);
+			spin_unlock(&to_i915(obj->base.dev)->runtime_pm.lmem_userfault_lock);
+			obj->userfault_count = 0;
+		}
+	}
+
 	ttm_bo_unmap_virtual(i915_gem_to_ttm(obj));
+
+	if (wakeref)
+		intel_runtime_pm_put(&to_i915(obj->base.dev)->runtime_pm, wakeref);
 }
 
 static const struct drm_i915_gem_object_ops i915_gem_ttm_obj_ops = {

drivers/gpu/drm/i915/intel_runtime_pm.c

@@ -648,6 +648,6 @@ void intel_runtime_pm_init_early(struct intel_runtime_pm *rpm)
 
 	init_intel_runtime_pm_wakeref(rpm);
 	INIT_LIST_HEAD(&rpm->lmem_userfault_list);
-	mutex_init(&rpm->lmem_userfault_lock);
+	spin_lock_init(&rpm->lmem_userfault_lock);
 	intel_wakeref_auto_init(&rpm->userfault_wakeref, rpm);
 }

drivers/gpu/drm/i915/intel_runtime_pm.h

@@ -64,7 +64,7 @@ struct intel_runtime_pm {
 	 *  but instead has exclusive access by virtue of all other accesses requiring
 	 *  holding the runtime pm wakeref.
 	 */
-	struct mutex lmem_userfault_lock;
+	spinlock_t lmem_userfault_lock;
 
 	/*
 	 *  Keep list of userfaulted gem obj, which require to release their