Currently, we reject attempting to pin a large bo into the mappable
aperture, but only after trying to create the vma. Under debug kernels,
repeatedly creating and freeing that vma for an oversized bo consumes
one-third of the runtime for pwrite/pread tests is spent on
kmalloc/kfree tracking. If we move the rejection to before creating that
vma, we lose some accuracy of checking against the fence_size as opposed
to object size, though the fence can never be smaller than the object.
Note that the vma creation itself will reject an attempt to create a vma
larger than the GTT so we can remove one redundant test.
Signed-off-by: Chris Wilson <chris@xxxxxxxxxxxxxxxxxx>
---
drivers/gpu/drm/i915/i915_gem.c | 65 ++++++++++++++++++++++-------------------
1 file changed, 35 insertions(+), 30 deletions(-)
diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index 792be4e2cc43..ed4b175403fc 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -4023,42 +4023,47 @@ i915_gem_object_ggtt_pin(struct drm_i915_gem_object *obj,
lockdep_assert_held(&obj->base.dev->struct_mutex);
+ if (!view && flags & PIN_MAPPABLE) {
+ /* If the required space is larger than the available
+ * aperture, we will not able to find a slot for the
+ * object and unbinding the object now will be in
+ * vain. Worse, doing so may cause us to ping-pong
+ * the object in and out of the Global GTT and
+ * waste a lot of cycles under the mutex.
+ */
+ if (obj->base.size > dev_priv->ggtt.mappable_end)
+ return ERR_PTR(-E2BIG);
+
+ /* If NONBLOCK is set the caller is optimistically
+ * trying to cache the full object within the mappable
+ * aperture, and *must* have a fallback in place for
+ * situations where we cannot bind the object. We
+ * can be a little more lax here and use the fallback
+ * more often to avoid costly migrations of ourselves
+ * and other objects within the aperture.
+ *
+ * Half-the-aperture is used as a simple heuristic.
+ * More interesting would to do search for a free
+ * block prior to making the commitment to unbind.
+ * That caters for the self-harm case, and with a
+ * little more heuristics (e.g. NOFAULT, NOEVICT)
+ * we could try to minimise harm to others.
+ */
+ if (flags & PIN_NONBLOCK &&
+ obj->base.size > dev_priv->ggtt.mappable_end / 2)
+ return ERR_PTR(-ENOSPC);
+ }
+
vma = i915_vma_instance(obj, vm, view);
if (unlikely(IS_ERR(vma)))
return vma;
if (i915_vma_misplaced(vma, size, alignment, flags)) {
- if (flags & PIN_NONBLOCK &&
- (i915_vma_is_pinned(vma) || i915_vma_is_active(vma)))
- return ERR_PTR(-ENOSPC);
+ if (flags & PIN_NONBLOCK) {
+ if (i915_vma_is_pinned(vma) || i915_vma_is_active(vma))
+ return ERR_PTR(-ENOSPC);
- if (flags & PIN_MAPPABLE) {
- /* If the required space is larger than the available
- * aperture, we will not able to find a slot for the
- * object and unbinding the object now will be in
- * vain. Worse, doing so may cause us to ping-pong
- * the object in and out of the Global GTT and
- * waste a lot of cycles under the mutex.
- */
- if (vma->fence_size > dev_priv->ggtt.mappable_end)
- return ERR_PTR(-E2BIG);
-
- /* If NONBLOCK is set the caller is optimistically
- * trying to cache the full object within the mappable
- * aperture, and *must* have a fallback in place for
- * situations where we cannot bind the object. We
- * can be a little more lax here and use the fallback
- * more often to avoid costly migrations of ourselves
- * and other objects within the aperture.
- *
- * Half-the-aperture is used as a simple heuristic.
- * More interesting would to do search for a free
- * block prior to making the commitment to unbind.
- * That caters for the self-harm case, and with a
- * little more heuristics (e.g. NOFAULT, NOEVICT)
- * we could try to minimise harm to others.
- */
- if (flags & PIN_NONBLOCK &&
+ if (flags & PIN_MAPPABLE &&
vma->fence_size > dev_priv->ggtt.mappable_end / 2)
return ERR_PTR(-ENOSPC);
}
--
2.14.2
_______________________________________________
Intel-gfx mailing list
Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
