drm/i915: Skip waking the device to service pwrite
authorChris Wilson <chris@chris-wilson.co.uk>
Thu, 19 Oct 2017 06:37:33 +0000 (07:37 +0100)
committerChris Wilson <chris@chris-wilson.co.uk>
Thu, 19 Oct 2017 12:56:38 +0000 (13:56 +0100)
If the device is in runtime suspend, resuming takes time and reduces our
powersaving. If this was for a small write into an object, that resume
will take longer than any savings in using the indirect GGTT access to
avoid the cpu cache.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Link: https://patchwork.freedesktop.org/patch/msgid/20171019063733.31620-1-chris@chris-wilson.co.uk
drivers/gpu/drm/i915/i915_gem.c

index d699ea3ab80be3d35e40072ec86d0fd40a7924ec..026cb52ece0b7e34db950c3d9a2128d46c290383 100644 (file)
@@ -1240,7 +1240,23 @@ i915_gem_gtt_pwrite_fast(struct drm_i915_gem_object *obj,
        if (ret)
                return ret;
 
-       intel_runtime_pm_get(i915);
+       if (i915_gem_object_has_struct_page(obj)) {
+               /*
+                * Avoid waking the device up if we can fallback, as
+                * waking/resuming is very slow (worst-case 10-100 ms
+                * depending on PCI sleeps and our own resume time).
+                * This easily dwarfs any performance advantage from
+                * using the cache bypass of indirect GGTT access.
+                */
+               if (!intel_runtime_pm_get_if_in_use(i915)) {
+                       ret = -EFAULT;
+                       goto out_unlock;
+               }
+       } else {
+               /* No backing pages, no fallback, we must force GGTT access */
+               intel_runtime_pm_get(i915);
+       }
+
        vma = i915_gem_object_ggtt_pin(obj, NULL, 0, 0,
                                       PIN_MAPPABLE |
                                       PIN_NONFAULT |
@@ -1257,7 +1273,7 @@ i915_gem_gtt_pwrite_fast(struct drm_i915_gem_object *obj,
        if (IS_ERR(vma)) {
                ret = insert_mappable_node(ggtt, &node, PAGE_SIZE);
                if (ret)
-                       goto out_unlock;
+                       goto out_rpm;
                GEM_BUG_ON(!node.allocated);
        }
 
@@ -1320,8 +1336,9 @@ out_unpin:
        } else {
                i915_vma_unpin(vma);
        }
-out_unlock:
+out_rpm:
        intel_runtime_pm_put(i915);
+out_unlock:
        mutex_unlock(&i915->drm.struct_mutex);
        return ret;
 }