drm/i915: Track the purgeable objects on a separate eviction list

Currently the purgeable objects, I915_MADV_DONTNEED, are mixed in the
normal bound/unbound lists. Every shrinker pass starts with an attempt
to purge from this set of unneeded objects, which entails us doing a
walk over both lists looking for any candidates. If there are none, and
since we are shrinking we can reasonably assume that the lists are
full!, this becomes a very slow futile walk.

If we separate out the purgeable objects into own list, this search then
becomes its own phase that is preferentially handled during shrinking.
Instead the cost becomes that we then need to filter the purgeable list
if we want to distinguish between bound and unbound objects.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Cc: Matthew Auld <matthew.william.auld@gmail.com>
Reviewed-by: Matthew Auld <matthew.william.auld@gmail.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190530203500.26272-1-chris@chris-wilson.co.uk

5 files changed, 31 insertions, 21 deletions

diff --git a/drivers/gpu/drm/i915/gem/i915_gem_domain.c b/drivers/gpu/drm/i915/gem/i915_gem_domain.c
index cce96e6c6e52..52b73e90c9f4 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_domain.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_domain.c
@@ -462,7 +462,6 @@ err_unpin_global:
 static void i915_gem_object_bump_inactive_ggtt(struct drm_i915_gem_object *obj)
 {
 	struct drm_i915_private *i915 = to_i915(obj->base.dev);
-	struct list_head *list;
 	struct i915_vma *vma;
 
 	GEM_BUG_ON(!i915_gem_object_has_pinned_pages(obj));
@@ -476,10 +475,15 @@ static void i915_gem_object_bump_inactive_ggtt(struct drm_i915_gem_object *obj)
 	}
 	mutex_unlock(&i915->ggtt.vm.mutex);
 
-	spin_lock(&i915->mm.obj_lock);
-	list = obj->bind_count ? &i915->mm.bound_list : &i915->mm.unbound_list;
-	list_move_tail(&obj->mm.link, list);
-	spin_unlock(&i915->mm.obj_lock);
+	if (obj->mm.madv == I915_MADV_WILLNEED) {
+		struct list_head *list;
+
+		spin_lock(&i915->mm.obj_lock);
+		list = obj->bind_count ?
+			&i915->mm.bound_list : &i915->mm.unbound_list;
+		list_move_tail(&obj->mm.link, list);
+		spin_unlock(&i915->mm.obj_lock);
+	}
 }
 
 void
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_object.c b/drivers/gpu/drm/i915/gem/i915_gem_object.c
index 55e79fdb81aa..1ec60be06755 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_object.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_object.c
@@ -333,9 +333,18 @@ void i915_gem_free_object(struct drm_gem_object *gem_obj)
 	if (obj->mm.quirked)
 		__i915_gem_object_unpin_pages(obj);
 
-	if (discard_backing_storage(obj))
+	if (discard_backing_storage(obj)) {
+		struct drm_i915_private *i915 = to_i915(obj->base.dev);
+
 		obj->mm.madv = I915_MADV_DONTNEED;
 
+		if (i915_gem_object_has_pages(obj)) {
+			spin_lock(&i915->mm.obj_lock);
+			list_move_tail(&obj->mm.link, &i915->mm.purge_list);
+			spin_unlock(&i915->mm.obj_lock);
+		}
+	}
+
 	/*
 	 * Before we free the object, make sure any pure RCU-only
 	 * read-side critical sections are complete, e.g.
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_pm.c b/drivers/gpu/drm/i915/gem/i915_gem_pm.c
index 11890e96ed65..89bb6d822f6e 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_pm.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_pm.c
@@ -164,6 +164,7 @@ void i915_gem_suspend_late(struct drm_i915_private *i915)
 	struct list_head *phases[] = {
 		&i915->mm.unbound_list,
 		&i915->mm.bound_list,
+		&i915->mm.purge_list,
 		NULL
 	}, **phase;
 
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_shmem.c b/drivers/gpu/drm/i915/gem/i915_gem_shmem.c
index 665f22ebf8e8..19d9ecdb2894 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_shmem.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_shmem.c
@@ -80,9 +80,7 @@ rebuild_st:
 	sg_page_sizes = 0;
 	for (i = 0; i < page_count; i++) {
 		const unsigned int shrink[] = {
-			(I915_SHRINK_BOUND |
-			 I915_SHRINK_UNBOUND |
-			 I915_SHRINK_PURGEABLE),
+			I915_SHRINK_BOUND | I915_SHRINK_UNBOUND,
 			0,
 		}, *s = shrink;
 		gfp_t gfp = noreclaim;
diff --git a/drivers/gpu/drm/i915/gem/i915_gem_shrinker.c b/drivers/gpu/drm/i915/gem/i915_gem_shrinker.c
index cd42299f019a..6a93e326abf3 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_shrinker.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_shrinker.c
@@ -144,6 +144,7 @@ i915_gem_shrink(struct drm_i915_private *i915,
 		struct list_head *list;
 		unsigned int bit;
 	} phases[] = {
+		{ &i915->mm.purge_list, ~0u },
 		{ &i915->mm.unbound_list, I915_SHRINK_UNBOUND },
 		{ &i915->mm.bound_list, I915_SHRINK_BOUND },
 		{ NULL, 0 },
@@ -226,10 +227,6 @@ i915_gem_shrink(struct drm_i915_private *i915,
 						       mm.link))) {
 			list_move_tail(&obj->mm.link, &still_in_list);
 
-			if (flags & I915_SHRINK_PURGEABLE &&
-			    obj->mm.madv != I915_MADV_DONTNEED)
-				continue;
-
 			if (flags & I915_SHRINK_VMAPS &&
 			    !is_vmalloc_addr(obj->mm.mapping))
 				continue;
@@ -239,6 +236,10 @@ i915_gem_shrink(struct drm_i915_private *i915,
 			     i915_gem_object_is_framebuffer(obj)))
 				continue;
 
+			if (!(flags & I915_SHRINK_BOUND) &&
+			    READ_ONCE(obj->bind_count))
+				continue;
+
 			if (!can_release_pages(obj))
 				continue;
 
@@ -324,6 +325,11 @@ i915_gem_shrinker_count(struct shrinker *shrinker, struct shrink_control *sc)
 			count += obj->base.size >> PAGE_SHIFT;
 			num_objects++;
 		}
+	list_for_each_entry(obj, &i915->mm.purge_list, mm.link)
+		if (!i915_gem_object_is_active(obj) && can_release_pages(obj)) {
+			count += obj->base.size >> PAGE_SHIFT;
+			num_objects++;
+		}
 	spin_unlock(&i915->mm.obj_lock);
 
 	/* Update our preferred vmscan batch size for the next pass.
@@ -361,15 +367,7 @@ i915_gem_shrinker_scan(struct shrinker *shrinker, struct shrink_control *sc)
 				&sc->nr_scanned,
 				I915_SHRINK_BOUND |
 				I915_SHRINK_UNBOUND |
-				I915_SHRINK_PURGEABLE |
 				I915_SHRINK_WRITEBACK);
-	if (sc->nr_scanned < sc->nr_to_scan)
-		freed += i915_gem_shrink(i915,
-					 sc->nr_to_scan - sc->nr_scanned,
-					 &sc->nr_scanned,
-					 I915_SHRINK_BOUND |
-					 I915_SHRINK_UNBOUND |
-					 I915_SHRINK_WRITEBACK);
 	if (sc->nr_scanned < sc->nr_to_scan && current_is_kswapd()) {
 		intel_wakeref_t wakeref;