[Intel-gfx] [PATCH] drm/i915/selftests: Provide full mb() around clflush
Chris Wilson
chris at chris-wilson.co.uk
Fri Jul 6 17:49:26 UTC 2018
clflush is an unserialised instruction and the IA manual strongly advises
you to serialise it with a mb. To be cautious, apply one before and one
after, so that it is serialised with both writes and reads without
worrying too much about the required direction.
Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
---
.../drm/i915/selftests/i915_gem_coherency.c | 21 ++++++++++++++++---
1 file changed, 18 insertions(+), 3 deletions(-)
diff --git a/drivers/gpu/drm/i915/selftests/i915_gem_coherency.c b/drivers/gpu/drm/i915/selftests/i915_gem_coherency.c
index 294c58aba2c1..df44c302a9fe 100644
--- a/drivers/gpu/drm/i915/selftests/i915_gem_coherency.c
+++ b/drivers/gpu/drm/i915/selftests/i915_gem_coherency.c
@@ -42,11 +42,21 @@ static int cpu_set(struct drm_i915_gem_object *obj,
page = i915_gem_object_get_page(obj, offset >> PAGE_SHIFT);
map = kmap_atomic(page);
- if (needs_clflush & CLFLUSH_BEFORE)
+
+ if (needs_clflush & CLFLUSH_BEFORE) {
+ mb();
clflush(map+offset_in_page(offset) / sizeof(*map));
+ mb();
+ }
+
map[offset_in_page(offset) / sizeof(*map)] = v;
- if (needs_clflush & CLFLUSH_AFTER)
+
+ if (needs_clflush & CLFLUSH_AFTER) {
+ mb();
clflush(map+offset_in_page(offset) / sizeof(*map));
+ mb();
+ }
+
kunmap_atomic(map);
i915_gem_obj_finish_shmem_access(obj);
@@ -68,8 +78,13 @@ static int cpu_get(struct drm_i915_gem_object *obj,
page = i915_gem_object_get_page(obj, offset >> PAGE_SHIFT);
map = kmap_atomic(page);
- if (needs_clflush & CLFLUSH_BEFORE)
+
+ if (needs_clflush & CLFLUSH_BEFORE) {
+ mb();
clflush(map+offset_in_page(offset) / sizeof(*map));
+ mb();
+ }
+
*v = map[offset_in_page(offset) / sizeof(*map)];
kunmap_atomic(map);
--
2.18.0
More information about the Intel-gfx
mailing list