[Intel-gfx] [PATCH 1/3] drm/i915: Track user GTT faulting per-vma

[Chris Wilson]

We don't wish to refault the entire object (other vma) when unbinding
one partial vma. To do this track which vma have been faulted into the
user's address space.

Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/i915_gem.c           | 31 ++++++++++++++++++++++---------
 drivers/gpu/drm/i915/i915_gem_evict.c     |  2 +-
 drivers/gpu/drm/i915/i915_gem_fence_reg.c |  2 +-
 drivers/gpu/drm/i915/i915_vma.c           |  3 ++-
 drivers/gpu/drm/i915/i915_vma.h           |  6 ++++++
 5 files changed, 32 insertions(+), 12 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index 36d838677982..cbac1b0ea0ae 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -1972,6 +1972,8 @@ int i915_gem_fault(struct vm_fault *vmf)
 			       (ggtt->mappable_base + vma->node.start) >> PAGE_SHIFT,
 			       min_t(u64, vma->size, area->vm_end - area->vm_start),
 			       &ggtt->mappable);
+	if (ret == 0)
+		vma->flags |= I915_VMA_USERFAULT;
 
 err_unpin:
 	__i915_vma_unpin(vma);
@@ -2024,6 +2026,22 @@ int i915_gem_fault(struct vm_fault *vmf)
 	return ret;
 }
 
+static void __i915_gem_object_release_mmap(struct drm_i915_gem_object *obj)
+{
+	struct i915_vma *vma;
+
+	list_del_init(&obj->userfault_link);
+	drm_vma_node_unmap(&obj->base.vma_node,
+			   obj->base.dev->anon_inode->i_mapping);
+
+	list_for_each_entry(vma, &obj->vma_list, obj_link) {
+		if (!i915_vma_is_ggtt(vma))
+			break;
+
+		vma->flags &= ~I915_VMA_USERFAULT;
+	}
+}
+
 /**
  * i915_gem_release_mmap - remove physical page mappings
  * @obj: obj in question
@@ -2057,9 +2075,7 @@ i915_gem_release_mmap(struct drm_i915_gem_object *obj)
 	if (list_empty(&obj->userfault_link))
 		goto out;
 
-	list_del_init(&obj->userfault_link);
-	drm_vma_node_unmap(&obj->base.vma_node,
-			   obj->base.dev->anon_inode->i_mapping);
+	__i915_gem_object_release_mmap(obj);
 
 	/* Ensure that the CPU's PTE are revoked and there are not outstanding
 	 * memory transactions from userspace before we return. The TLB
@@ -2087,11 +2103,8 @@ void i915_gem_runtime_suspend(struct drm_i915_private *dev_priv)
 	 */
 
 	list_for_each_entry_safe(obj, on,
-				 &dev_priv->mm.userfault_list, userfault_link) {
-		list_del_init(&obj->userfault_link);
-		drm_vma_node_unmap(&obj->base.vma_node,
-				   obj->base.dev->anon_inode->i_mapping);
-	}
+				 &dev_priv->mm.userfault_list, userfault_link)
+		__i915_gem_object_release_mmap(obj);
 
 	/* The fence will be lost when the device powers down. If any were
 	 * in use by hardware (i.e. they are pinned), we should not be powering
@@ -2114,7 +2127,7 @@ void i915_gem_runtime_suspend(struct drm_i915_private *dev_priv)
 		if (!reg->vma)
 			continue;
 
-		GEM_BUG_ON(!list_empty(&reg->vma->obj->userfault_link));
+		GEM_BUG_ON(i915_vma_has_userfault(reg->vma));
 		reg->dirty = true;
 	}
 }
diff --git a/drivers/gpu/drm/i915/i915_gem_evict.c b/drivers/gpu/drm/i915/i915_gem_evict.c
index a193f1b36c67..f4ea181d46fb 100644
--- a/drivers/gpu/drm/i915/i915_gem_evict.c
+++ b/drivers/gpu/drm/i915/i915_gem_evict.c
@@ -82,7 +82,7 @@ mark_free(struct drm_mm_scan *scan,
 	if (i915_vma_is_pinned(vma))
 		return false;
 
-	if (flags & PIN_NONFAULT && !list_empty(&vma->obj->userfault_link))
+	if (flags & PIN_NONFAULT && i915_vma_has_userfault(vma))
 		return false;
 
 	list_add(&vma->evict_link, unwind);
diff --git a/drivers/gpu/drm/i915/i915_gem_fence_reg.c b/drivers/gpu/drm/i915/i915_gem_fence_reg.c
index 5fe2cd8c8f28..52d5696f9e49 100644
--- a/drivers/gpu/drm/i915/i915_gem_fence_reg.c
+++ b/drivers/gpu/drm/i915/i915_gem_fence_reg.c
@@ -405,7 +405,7 @@ void i915_gem_restore_fences(struct drm_i915_private *dev_priv)
 		 */
 		if (vma && !i915_gem_object_is_tiled(vma->obj)) {
 			GEM_BUG_ON(!reg->dirty);
-			GEM_BUG_ON(!list_empty(&vma->obj->userfault_link));
+			GEM_BUG_ON(i915_vma_has_userfault(vma));
 
 			list_move(&reg->link, &dev_priv->mm.fence_list);
 			vma->fence = NULL;
diff --git a/drivers/gpu/drm/i915/i915_vma.c b/drivers/gpu/drm/i915/i915_vma.c
index 958be0a95960..fd671d493ca6 100644
--- a/drivers/gpu/drm/i915/i915_vma.c
+++ b/drivers/gpu/drm/i915/i915_vma.c
@@ -705,7 +705,8 @@ int i915_vma_unbind(struct i915_vma *vma)
 			return ret;
 
 		/* Force a pagefault for domain tracking on next user access */
-		i915_gem_release_mmap(obj);
+		if (i915_vma_has_userfault(vma))
+			i915_gem_release_mmap(obj);
 
 		__i915_vma_iounmap(vma);
 		vma->flags &= ~I915_VMA_CAN_FENCE;
diff --git a/drivers/gpu/drm/i915/i915_vma.h b/drivers/gpu/drm/i915/i915_vma.h
index 4a673fc1a432..78afbac2bca8 100644
--- a/drivers/gpu/drm/i915/i915_vma.h
+++ b/drivers/gpu/drm/i915/i915_vma.h
@@ -81,6 +81,7 @@ struct i915_vma {
 #define I915_VMA_GGTT		BIT(8)
 #define I915_VMA_CAN_FENCE	BIT(9)
 #define I915_VMA_CLOSED		BIT(10)
+#define I915_VMA_USERFAULT	BIT(11)
 
 	unsigned int active;
 	struct i915_gem_active last_read[I915_NUM_ENGINES];
@@ -143,6 +144,11 @@ static inline bool i915_vma_is_closed(const struct i915_vma *vma)
 	return vma->flags & I915_VMA_CLOSED;
 }
 
+static inline bool i915_vma_has_userfault(const struct i915_vma *vma)
+{
+	return vma->flags & I915_VMA_USERFAULT;
+}
+
 static inline unsigned int i915_vma_get_active(const struct i915_vma *vma)
 {
 	return vma->active;
-- 
2.13.1