Currently, we try to report to the shrinker the precise number of
objects (pages) that are available to be reaped at this moment. This
requires searching all objects with allocated pages to see if they
fulfill the search criteria, and this count is performed quite
frequently. (The shrinker tries to free ~128 pages on each invocation,
before which we count all the objects; counting takes longer than
unbinding the objects!) If we take the pragmatic view that with
sufficient desire, all objects are eventually reapable (they become
inactive, or no longer used as framebuffer etc), we can simply return
the count of pinned pages maintained during get_pages/put_pages rather
than walk the lists every time.
The downside is that we may (slightly) over-report the number of
objects/pages we could shrink and so penalize ourselves by shrinking
more than required. This is mitigated by keeping the order in which we
shrink objects such that we avoid penalizing active and frequently used
objects, and if memory is so tight that we need to free them we would
need to anyway.
v2: Only expose shrinkable objects to the shrinker; a small reduction in
not considering stolen and foreign objects.
v3: Restore the tracking from a "backup" copy from before the gem/ split
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Cc: Matthew Auld <matthew.auld@intel.com>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190530203500.26272-2-chris@chris-wilson.co.uk
}
mutex_unlock(&i915->ggtt.vm.mutex);
- if (obj->mm.madv == I915_MADV_WILLNEED) {
+ if (i915_gem_object_is_shrinkable(obj) &&
+ obj->mm.madv == I915_MADV_WILLNEED) {
struct list_head *list;
spin_lock(&i915->mm.obj_lock);
return kmem_cache_free(global.slab_objects, obj);
}
-/* some bookkeeping */
-static void i915_gem_info_add_obj(struct drm_i915_private *i915,
- u64 size)
-{
- spin_lock(&i915->mm.object_stat_lock);
- i915->mm.object_count++;
- i915->mm.object_memory += size;
- spin_unlock(&i915->mm.object_stat_lock);
-}
-
-static void i915_gem_info_remove_obj(struct drm_i915_private *i915,
- u64 size)
-{
- spin_lock(&i915->mm.object_stat_lock);
- i915->mm.object_count--;
- i915->mm.object_memory -= size;
- spin_unlock(&i915->mm.object_stat_lock);
-}
-
static void
frontbuffer_retire(struct i915_active_request *active,
struct i915_request *request)
obj->mm.madv = I915_MADV_WILLNEED;
INIT_RADIX_TREE(&obj->mm.get_page.radix, GFP_KERNEL | __GFP_NOWARN);
mutex_init(&obj->mm.get_page.lock);
-
- i915_gem_info_add_obj(to_i915(obj->base.dev), obj->base.size);
}
/**
static bool discard_backing_storage(struct drm_i915_gem_object *obj)
{
- /* If we are the last user of the backing storage (be it shmemfs
+ /*
+ * If we are the last user of the backing storage (be it shmemfs
* pages or stolen etc), we know that the pages are going to be
* immediately released. In this case, we can then skip copying
* back the contents from the GPU.
*/
+ if (!i915_gem_object_is_shrinkable(obj))
+ return false;
if (obj->mm.madv != I915_MADV_WILLNEED)
return false;
GEM_BUG_ON(!list_empty(&obj->vma.list));
GEM_BUG_ON(!RB_EMPTY_ROOT(&obj->vma.tree));
- /* This serializes freeing with the shrinker. Since the free
+ /*
+ * This serializes freeing with the shrinker. Since the free
* is delayed, first by RCU then by the workqueue, we want the
* shrinker to be able to free pages of unreferenced objects,
* or else we may oom whilst there are plenty of deferred
* freed objects.
*/
- if (i915_gem_object_has_pages(obj)) {
+ if (i915_gem_object_has_pages(obj) &&
+ i915_gem_object_is_shrinkable(obj)) {
spin_lock(&i915->mm.obj_lock);
list_del_init(&obj->mm.link);
spin_unlock(&i915->mm.obj_lock);
reservation_object_fini(&obj->__builtin_resv);
drm_gem_object_release(&obj->base);
- i915_gem_info_remove_obj(i915, obj->base.size);
bitmap_free(obj->bit_17);
i915_gem_object_free(obj);
}
GEM_BUG_ON(!HAS_PAGE_SIZES(i915, obj->mm.page_sizes.sg));
- spin_lock(&i915->mm.obj_lock);
- list_add(&obj->mm.link, &i915->mm.unbound_list);
- spin_unlock(&i915->mm.obj_lock);
+ if (i915_gem_object_is_shrinkable(obj)) {
+ spin_lock(&i915->mm.obj_lock);
+ i915->mm.shrink_count++;
+ i915->mm.shrink_memory += obj->base.size;
+ list_add(&obj->mm.link, &i915->mm.unbound_list);
+ spin_unlock(&i915->mm.obj_lock);
+ }
}
int ____i915_gem_object_get_pages(struct drm_i915_gem_object *obj)
if (IS_ERR_OR_NULL(pages))
return pages;
- spin_lock(&i915->mm.obj_lock);
- list_del(&obj->mm.link);
- spin_unlock(&i915->mm.obj_lock);
+ if (i915_gem_object_is_shrinkable(obj)) {
+ spin_lock(&i915->mm.obj_lock);
+ list_del(&obj->mm.link);
+ i915->mm.shrink_count--;
+ i915->mm.shrink_memory -= obj->base.size;
+ spin_unlock(&i915->mm.obj_lock);
+ }
if (obj->mm.mapping) {
void *ptr;
{
struct drm_i915_private *i915 =
container_of(shrinker, struct drm_i915_private, mm.shrinker);
- struct drm_i915_gem_object *obj;
- unsigned long num_objects = 0;
- unsigned long count = 0;
+ unsigned long num_objects;
+ unsigned long count;
- spin_lock(&i915->mm.obj_lock);
- list_for_each_entry(obj, &i915->mm.unbound_list, mm.link)
- if (can_release_pages(obj)) {
- count += obj->base.size >> PAGE_SHIFT;
- num_objects++;
- }
+ count = READ_ONCE(i915->mm.shrink_memory) >> PAGE_SHIFT;
+ num_objects = READ_ONCE(i915->mm.shrink_count);
- list_for_each_entry(obj, &i915->mm.bound_list, mm.link)
- if (!i915_gem_object_is_active(obj) && can_release_pages(obj)) {
- 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.
+ /*
+ * Update our preferred vmscan batch size for the next pass.
* Our rough guess for an effective batch size is roughly 2
* available GEM objects worth of pages. That is we don't want
* the shrinker to fire, until it is worth the cost of freeing an
mutex_unlock(&ggtt->vm.mutex);
spin_lock(&dev_priv->mm.obj_lock);
- list_move_tail(&obj->mm.link, &dev_priv->mm.bound_list);
+ GEM_BUG_ON(i915_gem_object_is_shrinkable(obj));
obj->bind_count++;
spin_unlock(&dev_priv->mm.obj_lock);
unsigned long total, count, n;
int ret;
- total = READ_ONCE(dev_priv->mm.object_count);
+ total = READ_ONCE(dev_priv->mm.shrink_count);
objects = kvmalloc_array(total, sizeof(*objects), GFP_KERNEL);
if (!objects)
return -ENOMEM;
char buf[80];
int ret;
- seq_printf(m, "%u objects, %llu bytes\n",
- dev_priv->mm.object_count,
- dev_priv->mm.object_memory);
+ seq_printf(m, "%u shrinkable objects, %llu bytes\n",
+ dev_priv->mm.shrink_count,
+ dev_priv->mm.shrink_memory);
size = count = 0;
mapped_size = mapped_count = 0;
return 0;
}
-static int i915_gem_gtt_info(struct seq_file *m, void *data)
-{
- struct drm_info_node *node = m->private;
- struct drm_i915_private *dev_priv = node_to_i915(node);
- struct drm_device *dev = &dev_priv->drm;
- struct drm_i915_gem_object **objects;
- struct drm_i915_gem_object *obj;
- u64 total_obj_size, total_gtt_size;
- unsigned long nobject, n;
- int count, ret;
-
- nobject = READ_ONCE(dev_priv->mm.object_count);
- objects = kvmalloc_array(nobject, sizeof(*objects), GFP_KERNEL);
- if (!objects)
- return -ENOMEM;
-
- ret = mutex_lock_interruptible(&dev->struct_mutex);
- if (ret)
- return ret;
-
- count = 0;
- spin_lock(&dev_priv->mm.obj_lock);
- list_for_each_entry(obj, &dev_priv->mm.bound_list, mm.link) {
- objects[count++] = obj;
- if (count == nobject)
- break;
- }
- spin_unlock(&dev_priv->mm.obj_lock);
-
- total_obj_size = total_gtt_size = 0;
- for (n = 0; n < count; n++) {
- obj = objects[n];
-
- seq_puts(m, " ");
- describe_obj(m, obj);
- seq_putc(m, '\n');
- total_obj_size += obj->base.size;
- total_gtt_size += i915_gem_obj_total_ggtt_size(obj);
- }
-
- mutex_unlock(&dev->struct_mutex);
-
- seq_printf(m, "Total %d objects, %llu bytes, %llu GTT size\n",
- count, total_obj_size, total_gtt_size);
- kvfree(objects);
-
- return 0;
-}
-
static int i915_gem_batch_pool_info(struct seq_file *m, void *data)
{
struct drm_i915_private *dev_priv = node_to_i915(m->private);
static const struct drm_info_list i915_debugfs_list[] = {
{"i915_capabilities", i915_capabilities, 0},
{"i915_gem_objects", i915_gem_object_info, 0},
- {"i915_gem_gtt", i915_gem_gtt_info, 0},
{"i915_gem_stolen", i915_gem_stolen_list_info },
{"i915_gem_fence_regs", i915_gem_fence_regs_info, 0},
{"i915_gem_interrupt", i915_interrupt_info, 0},
/** Bit 6 swizzling required for Y tiling */
u32 bit_6_swizzle_y;
- /* accounting, useful for userland debugging */
- spinlock_t object_stat_lock;
- u64 object_memory;
- u32 object_count;
+ /* shrinker accounting, also useful for userland debugging */
+ u64 shrink_memory;
+ u32 shrink_count;
};
#define I915_IDLE_ENGINES_TIMEOUT (200) /* in ms */
if (i915_gem_object_has_pages(obj)) {
struct list_head *list;
- spin_lock(&i915->mm.obj_lock);
- if (obj->mm.madv != I915_MADV_WILLNEED)
- list = &i915->mm.purge_list;
- else if (obj->bind_count)
- list = &i915->mm.bound_list;
- else
- list = &i915->mm.unbound_list;
- list_move_tail(&obj->mm.link, list);
- spin_unlock(&i915->mm.obj_lock);
+ if (i915_gem_object_is_shrinkable(obj)) {
+ spin_lock(&i915->mm.obj_lock);
+ if (obj->mm.madv != I915_MADV_WILLNEED)
+ list = &i915->mm.purge_list;
+ else if (obj->bind_count)
+ list = &i915->mm.bound_list;
+ else
+ list = &i915->mm.unbound_list;
+ list_move_tail(&obj->mm.link, list);
+ spin_unlock(&i915->mm.obj_lock);
+ }
}
/* if the object is no longer attached, discard its backing storage */
static void i915_gem_init__mm(struct drm_i915_private *i915)
{
- spin_lock_init(&i915->mm.object_stat_lock);
spin_lock_init(&i915->mm.obj_lock);
spin_lock_init(&i915->mm.free_lock);
i915_gem_drain_freed_objects(dev_priv);
GEM_BUG_ON(!llist_empty(&dev_priv->mm.free_list));
GEM_BUG_ON(atomic_read(&dev_priv->mm.free_count));
- WARN_ON(dev_priv->mm.object_count);
+ WARN_ON(dev_priv->mm.shrink_count);
cleanup_srcu_struct(&dev_priv->gpu_error.reset_backoff_srcu);
* so that we don't steal from recently used but inactive objects
* (unless we are forced to ofc!)
*/
- obj_bump_mru(obj);
+ if (i915_gem_object_is_shrinkable(obj))
+ obj_bump_mru(obj);
i915_gem_object_put(obj); /* and drop the active reference */
}
struct drm_i915_gem_object *obj = vma->obj;
spin_lock(&dev_priv->mm.obj_lock);
- list_move_tail(&obj->mm.link, &dev_priv->mm.bound_list);
- obj->bind_count++;
- spin_unlock(&dev_priv->mm.obj_lock);
+ if (i915_gem_object_is_shrinkable(obj))
+ list_move_tail(&obj->mm.link, &dev_priv->mm.bound_list);
+
+ obj->bind_count++;
assert_bind_count(obj);
+
+ spin_unlock(&dev_priv->mm.obj_lock);
}
return 0;
struct drm_i915_gem_object *obj = vma->obj;
spin_lock(&i915->mm.obj_lock);
+
+ GEM_BUG_ON(obj->bind_count == 0);
if (--obj->bind_count == 0 &&
+ i915_gem_object_is_shrinkable(obj) &&
obj->mm.madv == I915_MADV_WILLNEED)
list_move_tail(&obj->mm.link, &i915->mm.unbound_list);
+
spin_unlock(&i915->mm.obj_lock);
/*
projects / openwrt / staging / blogic.git / commitdiff