drm/i915: Ignore applying the self-relocation BIAS if no relocations
authorChris Wilson <chris@chris-wilson.co.uk>
Thu, 21 Jun 2018 07:32:05 +0000 (08:32 +0100)
committerChris Wilson <chris@chris-wilson.co.uk>
Thu, 21 Jun 2018 09:32:11 +0000 (10:32 +0100)
We only need to apply the BIAS for self-relocations into the batchbuffer
iff the execobject has any relocations.

This suppresses some warnings we may get with a full gtt (so the batch
object has wound up at 0 from a previous invocation), but doesn't fix
the underlying problem of how we tried to move a pinned batch vma (how
we have a pinned user vma outside of execbuf, I do not know, though this
being on an aliasing ppgtt means it could be a spurious pinning via the
global gtt). One step at a time...

References: https://bugs.freedesktop.org/show_bug.cgi?id=106744#c1
Testcase: igt/gem_exec_gttfill # byt (sporadic)
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180621073205.26701-1-chris@chris-wilson.co.uk
drivers/gpu/drm/i915/i915_gem_execbuffer.c

index 60dc2a865f5f960ce0ca46d89fa4e8eb66a0d652..437441f4af4129fa81f9e82d11e9594359a6f236 100644 (file)
@@ -534,7 +534,8 @@ eb_add_vma(struct i915_execbuffer *eb,
         * paranoia do it everywhere.
         */
        if (i == batch_idx) {
-               if (!(eb->flags[i] & EXEC_OBJECT_PINNED))
+               if (entry->relocation_count &&
+                   !(eb->flags[i] & EXEC_OBJECT_PINNED))
                        eb->flags[i] |= __EXEC_OBJECT_NEEDS_BIAS;
                if (eb->reloc_cache.has_fence)
                        eb->flags[i] |= EXEC_OBJECT_NEEDS_FENCE;