On 09/12/2016 15:05, Chris Wilson wrote:
Some object retain an extra pin whilst they are active (e.g. contexts).
This excludes them from being considered for eviction unless we idle the
GPU. If before we look at the active list, before we retire we can
hopefully remove a few excess pins and reduce the amount of searching
required.
v2: Similar principle applies to evict_for_vma
Signed-off-by: Chris Wilson <chris@xxxxxxxxxxxxxxxxxx>
---
drivers/gpu/drm/i915/i915_gem_evict.c | 17 ++++++++++++++++-
1 file changed, 16 insertions(+), 1 deletion(-)
diff --git a/drivers/gpu/drm/i915/i915_gem_evict.c b/drivers/gpu/drm/i915/i915_gem_evict.c
index a3f7f1ca6f21..689338e22ba8 100644
--- a/drivers/gpu/drm/i915/i915_gem_evict.c
+++ b/drivers/gpu/drm/i915/i915_gem_evict.c
@@ -130,7 +130,14 @@ i915_gem_evict_something(struct i915_address_space *vm,
start, end,
flags & PIN_HIGH ? DRM_MM_INSERT_HIGH : 0);
- if (flags & PIN_NONBLOCK)
+ /* Retire before we search the active list. Although we have
+ * reasonable accuracy in our retirement lists, we may have
+ * a stray pin (preventing eviction) that can only be resolved by
+ * retiring.
+ */
+ if (!(flags & PIN_NONBLOCK))
+ i915_gem_retire_requests(dev_priv);
+ else
phases[1] = NULL;
search_again:
@@ -236,6 +243,14 @@ int i915_gem_evict_for_vma(struct i915_vma *target, unsigned int flags)
lockdep_assert_held(&target->vm->i915->drm.struct_mutex);
trace_i915_gem_evict_vma(target, flags);
+ /* Retire before we search the active list. Although we have
+ * reasonable accuracy in our retirement lists, we may have
+ * a stray pin (preventing eviction) that can only be resolved by
+ * retiring.
+ */
+ if (!(flags & PIN_NONBLOCK))
+ i915_gem_retire_requests(target->vm->i915);
+
check_color = target->vm->mm.color_adjust;
if (check_color) {
/* Expand search to cover neighbouring guard pages (or lack!) */
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@xxxxxxxxx>
Regards,
Tvrtko
_______________________________________________
Intel-gfx mailing list
Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
