drm/i915: Throw away the active object retirement complexity
Remove the accumulated optimisations that we have for i915_vma_retire
and reduce it to the bare essential of tracking the active object
reference. This allows us to only use atomic operations, and so will be
able to avoid the struct_mutex requirement.
The principal loss here is the shrinker MRU bumping, so now if we have
to shrink, we will do so in much more random order and more likely to
try and shrink recently used objects. That is a nuisance, but shrinking
active objects is a second step we try to avoid and will always be a
system-wide performance issue.
The other loss is here is in the automatic pruning of the
reservation_object when idling. This is not as large an issue as upon
reservation_object introduction as now adding new fences into the object
replaces already signaled fences, keeping the array compact. But we do
lose the auto-expiration of stale fences and unused arrays. That may be
a noticeable problem for which we need to re-implement autopruning.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190621183801.23252-3-chris@chris-wilson.co.uk
diff --git a/drivers/gpu/drm/i915/i915_vma.c b/drivers/gpu/drm/i915/i915_vma.c
index 503f118..c13b86e 100644
--- a/drivers/gpu/drm/i915/i915_vma.c
+++ b/drivers/gpu/drm/i915/i915_vma.c
@@ -78,43 +78,11 @@ static void vma_print_allocator(struct i915_vma *vma, const char *reason)
#endif
-static void obj_bump_mru(struct drm_i915_gem_object *obj)
-{
- struct drm_i915_private *i915 = to_i915(obj->base.dev);
- unsigned long flags;
-
- spin_lock_irqsave(&i915->mm.obj_lock, flags);
- list_move_tail(&obj->mm.link, &i915->mm.shrink_list);
- spin_unlock_irqrestore(&i915->mm.obj_lock, flags);
-
- obj->mm.dirty = true; /* be paranoid */
-}
-
static void __i915_vma_retire(struct i915_active *ref)
{
struct i915_vma *vma = container_of(ref, typeof(*vma), active);
- struct drm_i915_gem_object *obj = vma->obj;
- GEM_BUG_ON(!i915_gem_object_is_active(obj));
- if (--obj->active_count)
- return;
-
- /* Prune the shared fence arrays iff completely idle (inc. external) */
- if (reservation_object_trylock(obj->base.resv)) {
- if (reservation_object_test_signaled_rcu(obj->base.resv, true))
- reservation_object_add_excl_fence(obj->base.resv, NULL);
- reservation_object_unlock(obj->base.resv);
- }
-
- /*
- * Bump our place on the bound list to keep it roughly in LRU order
- * so that we don't steal from recently used but inactive objects
- * (unless we are forced to ofc!)
- */
- if (i915_gem_object_is_shrinkable(obj))
- obj_bump_mru(obj);
-
- i915_gem_object_put(obj); /* and drop the active reference */
+ i915_vma_put(vma);
}
static struct i915_vma *
@@ -922,6 +890,7 @@ int i915_vma_move_to_active(struct i915_vma *vma,
unsigned int flags)
{
struct drm_i915_gem_object *obj = vma->obj;
+ int err;
assert_vma_held(vma);
assert_object_held(obj);
@@ -935,17 +904,13 @@ int i915_vma_move_to_active(struct i915_vma *vma,
* add the active reference first and queue for it to be dropped
* *last*.
*/
- if (!vma->active.count && !obj->active_count++)
- i915_gem_object_get(obj); /* once more for the active ref */
+ if (i915_active_acquire(&vma->active))
+ i915_vma_get(vma);
- if (unlikely(i915_active_ref(&vma->active, rq->fence.context, rq))) {
- if (!vma->active.count && !--obj->active_count)
- i915_gem_object_put(obj);
- return -ENOMEM;
- }
-
- GEM_BUG_ON(!i915_vma_is_active(vma));
- GEM_BUG_ON(!obj->active_count);
+ err = i915_active_ref(&vma->active, rq->fence.context, rq);
+ i915_active_release(&vma->active);
+ if (unlikely(err))
+ return err;
obj->write_domain = 0;
if (flags & EXEC_OBJECT_WRITE) {
@@ -957,11 +922,14 @@ int i915_vma_move_to_active(struct i915_vma *vma,
obj->read_domains = 0;
}
obj->read_domains |= I915_GEM_GPU_DOMAINS;
+ obj->mm.dirty = true;
if (flags & EXEC_OBJECT_NEEDS_FENCE)
__i915_active_request_set(&vma->last_fence, rq);
export_fence(vma, rq, flags);
+
+ GEM_BUG_ON(!i915_vma_is_active(vma));
return 0;
}