drm/i915: Pull i915_vma_pin under the vm->mutex

Replace the struct_mutex requirement for pinning the i915_vma with the
local vm->mutex instead. Note that the vm->mutex is tainted by the
shrinker (we require unbinding from inside fs-reclaim) and so we cannot
allocate while holding that mutex. Instead we have to preallocate
workers to do allocate and apply the PTE updates after we have we
reserved their slot in the drm_mm (using fences to order the PTE writes
with the GPU work and with later unbind).

In adding the asynchronous vma binding, one subtle requirement is to
avoid coupling the binding fence into the backing object->resv. That is
the asynchronous binding only applies to the vma timeline itself and not
to the pages as that is a more global timeline (the binding of one vma
does not need to be ordered with another vma, nor does the implicit GEM
fencing depend on a vma, only on writes to the backing store). Keeping
the vma binding distinct from the backing store timelines is verified by
a number of async gem_exec_fence and gem_exec_schedule tests. The way we
do this is quite simple, we keep the fence for the vma binding separate
and only wait on it as required, and never add it to the obj->resv
itself.

Another consequence in reducing the locking around the vma is the
destruction of the vma is no longer globally serialised by struct_mutex.
A natural solution would be to add a kref to i915_vma, but that requires
decoupling the reference cycles, possibly by introducing a new
i915_mm_pages object that is own by both obj->mm and vma->pages.
However, we have not taken that route due to the overshadowing lmem/ttm
discussions, and instead play a series of complicated games with
trylocks to (hopefully) ensure that only one destruction path is called!

v2: Add some commentary, and some helpers to reduce patch churn.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20191004134015.13204-4-chris@chris-wilson.co.uk

1 files changed, 33 insertions, 50 deletions

diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index fa8e028ac0b5..fc5c618f6c19 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -62,20 +62,31 @@
 #include "intel_pm.h"
 
 static int
-insert_mappable_node(struct i915_ggtt *ggtt,
-                     struct drm_mm_node *node, u32 size)
+insert_mappable_node(struct i915_ggtt *ggtt, struct drm_mm_node *node, u32 size)
 {
+	int err;
+
+	err = mutex_lock_interruptible(&ggtt->vm.mutex);
+	if (err)
+		return err;
+
 	memset(node, 0, sizeof(*node));
-	return drm_mm_insert_node_in_range(&ggtt->vm.mm, node,
-					   size, 0, I915_COLOR_UNEVICTABLE,
-					   0, ggtt->mappable_end,
-					   DRM_MM_INSERT_LOW);
+	err = drm_mm_insert_node_in_range(&ggtt->vm.mm, node,
+					  size, 0, I915_COLOR_UNEVICTABLE,
+					  0, ggtt->mappable_end,
+					  DRM_MM_INSERT_LOW);
+
+	mutex_unlock(&ggtt->vm.mutex);
+
+	return err;
 }
 
 static void
-remove_mappable_node(struct drm_mm_node *node)
+remove_mappable_node(struct i915_ggtt *ggtt, struct drm_mm_node *node)
 {
+	mutex_lock(&ggtt->vm.mutex);
 	drm_mm_remove_node(node);
+	mutex_unlock(&ggtt->vm.mutex);
 }
 
 int
@@ -87,7 +98,8 @@ i915_gem_get_aperture_ioctl(struct drm_device *dev, void *data,
 	struct i915_vma *vma;
 	u64 pinned;
 
-	mutex_lock(&ggtt->vm.mutex);
+	if (mutex_lock_interruptible(&ggtt->vm.mutex))
+		return -EINTR;
 
 	pinned = ggtt->vm.reserved;
 	list_for_each_entry(vma, &ggtt->vm.bound_list, vm_link)
@@ -109,20 +121,24 @@ int i915_gem_object_unbind(struct drm_i915_gem_object *obj,
 	LIST_HEAD(still_in_list);
 	int ret = 0;
 
-	lockdep_assert_held(&obj->base.dev->struct_mutex);
-
 	spin_lock(&obj->vma.lock);
 	while (!ret && (vma = list_first_entry_or_null(&obj->vma.list,
 						       struct i915_vma,
 						       obj_link))) {
+		struct i915_address_space *vm = vma->vm;
+
+		ret = -EBUSY;
+		if (!i915_vm_tryopen(vm))
+			break;
+
 		list_move_tail(&vma->obj_link, &still_in_list);
 		spin_unlock(&obj->vma.lock);
 
-		ret = -EBUSY;
 		if (flags & I915_GEM_OBJECT_UNBIND_ACTIVE ||
 		    !i915_vma_is_active(vma))
 			ret = i915_vma_unbind(vma);
 
+		i915_vm_close(vm);
 		spin_lock(&obj->vma.lock);
 	}
 	list_splice(&still_in_list, &obj->vma.list);
@@ -338,10 +354,6 @@ i915_gem_gtt_pread(struct drm_i915_gem_object *obj,
 	u64 remain, offset;
 	int ret;
 
-	ret = mutex_lock_interruptible(&i915->drm.struct_mutex);
-	if (ret)
-		return ret;
-
 	wakeref = intel_runtime_pm_get(&i915->runtime_pm);
 	vma = ERR_PTR(-ENODEV);
 	if (!i915_gem_object_is_tiled(obj))
@@ -355,12 +367,10 @@ i915_gem_gtt_pread(struct drm_i915_gem_object *obj,
 	} else {
 		ret = insert_mappable_node(ggtt, &node, PAGE_SIZE);
 		if (ret)
-			goto out_unlock;
+			goto out_rpm;
 		GEM_BUG_ON(!drm_mm_node_allocated(&node));
 	}
 
-	mutex_unlock(&i915->drm.struct_mutex);
-
 	ret = i915_gem_object_lock_interruptible(obj);
 	if (ret)
 		goto out_unpin;
@@ -414,17 +424,14 @@ i915_gem_gtt_pread(struct drm_i915_gem_object *obj,
 
 	i915_gem_object_unlock_fence(obj, fence);
 out_unpin:
-	mutex_lock(&i915->drm.struct_mutex);
 	if (drm_mm_node_allocated(&node)) {
 		ggtt->vm.clear_range(&ggtt->vm, node.start, node.size);
-		remove_mappable_node(&node);
+		remove_mappable_node(ggtt, &node);
 	} else {
 		i915_vma_unpin(vma);
 	}
-out_unlock:
+out_rpm:
 	intel_runtime_pm_put(&i915->runtime_pm, wakeref);
-	mutex_unlock(&i915->drm.struct_mutex);
-
 	return ret;
 }
 
@@ -531,10 +538,6 @@ i915_gem_gtt_pwrite_fast(struct drm_i915_gem_object *obj,
 	void __user *user_data;
 	int ret;
 
-	ret = mutex_lock_interruptible(&i915->drm.struct_mutex);
-	if (ret)
-		return ret;
-
 	if (i915_gem_object_has_struct_page(obj)) {
 		/*
 		 * Avoid waking the device up if we can fallback, as
@@ -544,10 +547,8 @@ i915_gem_gtt_pwrite_fast(struct drm_i915_gem_object *obj,
 		 * using the cache bypass of indirect GGTT access.
 		 */
 		wakeref = intel_runtime_pm_get_if_in_use(rpm);
-		if (!wakeref) {
-			ret = -EFAULT;
-			goto out_unlock;
-		}
+		if (!wakeref)
+			return -EFAULT;
 	} else {
 		/* No backing pages, no fallback, we must force GGTT access */
 		wakeref = intel_runtime_pm_get(rpm);
@@ -569,8 +570,6 @@ i915_gem_gtt_pwrite_fast(struct drm_i915_gem_object *obj,
 		GEM_BUG_ON(!drm_mm_node_allocated(&node));
 	}
 
-	mutex_unlock(&i915->drm.struct_mutex);
-
 	ret = i915_gem_object_lock_interruptible(obj);
 	if (ret)
 		goto out_unpin;
@@ -634,18 +633,15 @@ i915_gem_gtt_pwrite_fast(struct drm_i915_gem_object *obj,
 
 	i915_gem_object_unlock_fence(obj, fence);
 out_unpin:
-	mutex_lock(&i915->drm.struct_mutex);
 	intel_gt_flush_ggtt_writes(ggtt->vm.gt);
 	if (drm_mm_node_allocated(&node)) {
 		ggtt->vm.clear_range(&ggtt->vm, node.start, node.size);
-		remove_mappable_node(&node);
+		remove_mappable_node(ggtt, &node);
 	} else {
 		i915_vma_unpin(vma);
 	}
 out_rpm:
 	intel_runtime_pm_put(rpm, wakeref);
-out_unlock:
-	mutex_unlock(&i915->drm.struct_mutex);
 	return ret;
 }
 
@@ -968,8 +964,6 @@ i915_gem_object_ggtt_pin(struct drm_i915_gem_object *obj,
 	struct i915_vma *vma;
 	int ret;
 
-	lockdep_assert_held(&obj->base.dev->struct_mutex);
-
 	if (i915_gem_object_never_bind_ggtt(obj))
 		return ERR_PTR(-ENODEV);
 
@@ -1019,13 +1013,6 @@ i915_gem_object_ggtt_pin(struct drm_i915_gem_object *obj,
 				return ERR_PTR(-ENOSPC);
 		}
 
-		WARN(i915_vma_is_pinned(vma),
-		     "bo is already pinned in ggtt with incorrect alignment:"
-		     " offset=%08x, req.alignment=%llx,"
-		     " req.map_and_fenceable=%d, vma->map_and_fenceable=%d\n",
-		     i915_ggtt_offset(vma), alignment,
-		     !!(flags & PIN_MAPPABLE),
-		     i915_vma_is_map_and_fenceable(vma));
 		ret = i915_vma_unbind(vma);
 		if (ret)
 			return ERR_PTR(ret);
@@ -1444,8 +1431,6 @@ err_unlock:
 	}
 
 	if (ret == -EIO) {
-		mutex_lock(&dev_priv->drm.struct_mutex);
-
 		/*
 		 * Allow engines or uC initialisation to fail by marking the GPU
 		 * as wedged. But we only want to do this when the GPU is angry,
@@ -1462,8 +1447,6 @@ err_unlock:
 		i915_gem_restore_gtt_mappings(dev_priv);
 		i915_gem_restore_fences(dev_priv);
 		intel_init_clock_gating(dev_priv);
-
-		mutex_unlock(&dev_priv->drm.struct_mutex);
 	}
 
 	i915_gem_drain_freed_objects(dev_priv);