[PATCH 03/17] drm/i915: Limit searching for PIN_HIGH

[Chris Wilson]

To no surprise (since we've flip-flopped over the use of PIN_HIGH a few
times), doing a search by address over a pathologically fragmented
address space is exceeding slow. To protect ourselves from nearly
unbounded latency (think searching a million holes while under
struct_mutex), limit the search for the highest available hole and
fallback to best-fit if it fails.

In the pathologically fragmented case, such as igt/gem_ctx_thrash, the
effect is dramatic, bringing the runtime down from hours to seconds
(depending on how many other slow searches you hit, e.g. alloc_iova()
and alloc_vmap_area() both degrade to a slow rbtree walk after their
small cache is exhausted). For the real world, the number of search
steps is unlikely to be significant as we should only need to search
once per new context.

Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
Cc: Joonas Lahtinen <joonas.lahtinen at linux.intel.com>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen at linux.intel.com>
---
 drivers/gpu/drm/i915/i915_gem_gtt.c | 11 ++++++++++-
 1 file changed, 10 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/i915/i915_gem_gtt.c b/drivers/gpu/drm/i915/i915_gem_gtt.c
index 996ab2ad6c45..fcf3a973925b 100644
--- a/drivers/gpu/drm/i915/i915_gem_gtt.c
+++ b/drivers/gpu/drm/i915/i915_gem_gtt.c
@@ -3988,7 +3988,7 @@ int i915_gem_gtt_insert(struct i915_address_space *vm,
 
 	mode = DRM_MM_INSERT_BEST;
 	if (flags & PIN_HIGH)
-		mode = DRM_MM_INSERT_HIGH;
+		mode = DRM_MM_INSERT_HIGHEST;
 	if (flags & PIN_MAPPABLE)
 		mode = DRM_MM_INSERT_LOW;
 
@@ -4008,6 +4008,15 @@ int i915_gem_gtt_insert(struct i915_address_space *vm,
 	if (err != -ENOSPC)
 		return err;
 
+	if (mode & DRM_MM_INSERT_ONCE) {
+		err = drm_mm_insert_node_in_range(&vm->mm, node,
+						  size, alignment, color,
+						  start, end,
+						  DRM_MM_INSERT_BEST);
+		if (err != -ENOSPC)
+			return err;
+	}
+
 	if (flags & PIN_NOEVICT)
 		return -ENOSPC;
 
-- 
2.17.0