drm/i915/gem: Split the context's obj:vma lut into its own mutex

Rather than reuse the common ctx->mutex for locking the execbuffer LUT,
split it into its own lock to avoid being taken [as part of ctx->mutex]
at inappropriate times. In particular to avoid the inversion from taking
the timeline->mutex for the whole execbuf submission in the next patch.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Andi Shyti <andi.shyti@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20200703004306.11117-1-chris@chris-wilson.co.uk

drivers/gpu/drm/i915/gem/i915_gem_context.c

@@ -101,8 +101,7 @@ static void lut_close(struct i915_gem_context *ctx)
 	struct radix_tree_iter iter;
 	void __rcu **slot;
 
-	lockdep_assert_held(&ctx->mutex);
+	mutex_lock(&ctx->lut_mutex);
-
 	rcu_read_lock();
 	radix_tree_for_each_slot(slot, &ctx->handles_vma, &iter, 0) {
 		struct i915_vma *vma = rcu_dereference_raw(*slot);
@@ -135,6 +134,7 @@ static void lut_close(struct i915_gem_context *ctx)
 		i915_gem_object_put(obj);
 	}
 	rcu_read_unlock();
+	mutex_unlock(&ctx->lut_mutex);
 }
 
 static struct intel_context *
@@ -342,6 +342,7 @@ static void i915_gem_context_free(struct i915_gem_context *ctx)
 	spin_unlock(&ctx->i915->gem.contexts.lock);
 
 	mutex_destroy(&ctx->engines_mutex);
+	mutex_destroy(&ctx->lut_mutex);
 
 	if (ctx->timeline)
 		intel_timeline_put(ctx->timeline);
@@ -725,6 +726,7 @@ __create_context(struct drm_i915_private *i915)
 	RCU_INIT_POINTER(ctx->engines, e);
 
 	INIT_RADIX_TREE(&ctx->handles_vma, GFP_KERNEL);
+	mutex_init(&ctx->lut_mutex);
 
 	/* NB: Mark all slices as needing a remap so that when the context first
 	 * loads it will restore whatever remap state already exists. If there
@@ -1312,11 +1314,11 @@ static int set_ppgtt(struct drm_i915_file_private *file_priv,
 	if (vm == rcu_access_pointer(ctx->vm))
 		goto unlock;
 
+	old = __set_ppgtt(ctx, vm);
+
 	/* Teardown the existing obj:vma cache, it will have to be rebuilt. */
 	lut_close(ctx);
 
-	old = __set_ppgtt(ctx, vm);
-
 	/*
 	 * We need to flush any requests using the current ppgtt before
 	 * we release it as the requests do not hold a reference themselves,
@@ -1330,6 +1332,7 @@ static int set_ppgtt(struct drm_i915_file_private *file_priv,
 	if (err) {
 		i915_vm_close(__set_ppgtt(ctx, old));
 		i915_vm_close(old);
+		lut_close(ctx); /* force a rebuild of the old obj:vma cache */
 	}
 
 unlock:

drivers/gpu/drm/i915/gem/i915_gem_context_types.h

@@ -170,6 +170,7 @@ struct i915_gem_context {
 	 * per vm, which may be one per context or shared with the global GTT)
 	 */
 	struct radix_tree_root handles_vma;
+	struct mutex lut_mutex;
 
 	/**
 	 * @name: arbitrary name, used for user debug

drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c

@@ -782,10 +782,15 @@ static int __eb_add_lut(struct i915_execbuffer *eb,
 
 	/* Check that the context hasn't been closed in the meantime */
 	err = -EINTR;
-	if (!mutex_lock_interruptible(&ctx->mutex)) {
+	if (!mutex_lock_interruptible(&ctx->lut_mutex)) {
-		err = -ENOENT;
+		struct i915_address_space *vm = rcu_access_pointer(ctx->vm);
-		if (likely(!i915_gem_context_is_closed(ctx)))
+
+		if (unlikely(vm && vma->vm != vm))
+			err = -EAGAIN; /* user racing with ctx set-vm */
+		else if (likely(!i915_gem_context_is_closed(ctx)))
 			err = radix_tree_insert(&ctx->handles_vma, handle, vma);
+		else
+			err = -ENOENT;
 		if (err == 0) { /* And nor has this handle */
 			struct drm_i915_gem_object *obj = vma->obj;
 
@@ -798,7 +803,7 @@ static int __eb_add_lut(struct i915_execbuffer *eb,
 			}
 			spin_unlock(&obj->lut_lock);
 		}
-		mutex_unlock(&ctx->mutex);
+		mutex_unlock(&ctx->lut_mutex);
 	}
 	if (unlikely(err))
 		goto err;
@@ -814,6 +819,8 @@ static int __eb_add_lut(struct i915_execbuffer *eb,
 
 static struct i915_vma *eb_lookup_vma(struct i915_execbuffer *eb, u32 handle)
 {
+	struct i915_address_space *vm = eb->context->vm;
+
 	do {
 		struct drm_i915_gem_object *obj;
 		struct i915_vma *vma;
@@ -821,7 +828,7 @@ static struct i915_vma *eb_lookup_vma(struct i915_execbuffer *eb, u32 handle)
 
 		rcu_read_lock();
 		vma = radix_tree_lookup(&eb->gem_context->handles_vma, handle);
-		if (likely(vma))
+		if (likely(vma && vma->vm == vm))
 			vma = i915_vma_tryget(vma);
 		rcu_read_unlock();
 		if (likely(vma))
@@ -831,7 +838,7 @@ static struct i915_vma *eb_lookup_vma(struct i915_execbuffer *eb, u32 handle)
 		if (unlikely(!obj))
 			return ERR_PTR(-ENOENT);
 
-		vma = i915_vma_instance(obj, eb->context->vm, NULL);
+		vma = i915_vma_instance(obj, vm, NULL);
 		if (IS_ERR(vma)) {
 			i915_gem_object_put(obj);
 			return vma;

drivers/gpu/drm/i915/gem/i915_gem_object.c

@@ -143,14 +143,14 @@ void i915_gem_close_object(struct drm_gem_object *gem, struct drm_file *file)
 		 * vma, in the same fd namespace, by virtue of flink/open.
 		 */
 
-		mutex_lock(&ctx->mutex);
+		mutex_lock(&ctx->lut_mutex);
 		vma = radix_tree_delete(&ctx->handles_vma, lut->handle);
 		if (vma) {
 			GEM_BUG_ON(vma->obj != obj);
 			GEM_BUG_ON(!atomic_read(&vma->open_count));
 			i915_vma_close(vma);
 		}
-		mutex_unlock(&ctx->mutex);
+		mutex_unlock(&ctx->lut_mutex);
 
 		i915_gem_context_put(lut->ctx);
 		i915_lut_handle_free(lut);

drivers/gpu/drm/i915/gem/selftests/mock_context.c

@@ -23,6 +23,8 @@ mock_context(struct drm_i915_private *i915,
 	INIT_LIST_HEAD(&ctx->link);
 	ctx->i915 = i915;
 
+	mutex_init(&ctx->mutex);
+
 	spin_lock_init(&ctx->stale.lock);
 	INIT_LIST_HEAD(&ctx->stale.engines);
 
@@ -35,7 +37,7 @@ mock_context(struct drm_i915_private *i915,
 	RCU_INIT_POINTER(ctx->engines, e);
 
 	INIT_RADIX_TREE(&ctx->handles_vma, GFP_KERNEL);
-	mutex_init(&ctx->mutex);
+	mutex_init(&ctx->lut_mutex);
 
 	if (name) {
 		struct i915_ppgtt *ppgtt;