[Intel-gfx] [PATCH i-g-t] i915/gem_exec_balancer: Beware the migratory fence

[Chris Wilson]

If the object needs to be migrated, it may will need GPU relocs and so
have an exclusive fence showing up in the write domain.

Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
---
 tests/i915/gem_exec_balancer.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/tests/i915/gem_exec_balancer.c b/tests/i915/gem_exec_balancer.c
index b2074486a..65c3f7bea 100644
--- a/tests/i915/gem_exec_balancer.c
+++ b/tests/i915/gem_exec_balancer.c
@@ -836,10 +836,10 @@ static void busy(int i915)
 		do_ioctl(i915, DRM_IOCTL_I915_GEM_BUSY, &busy);
 		igt_assert_eq_u32(busy.busy, 1u << (class + 16));
 
-		/* Queued(read): expected class */
+		/* Queued(read, maybe write if being migrated): expected class */
 		busy.handle = spin[1]->handle;
 		do_ioctl(i915, DRM_IOCTL_I915_GEM_BUSY, &busy);
-		igt_assert_eq_u32(busy.busy, 1u << (class + 16));
+		igt_assert_eq_u32(busy.busy & 0xffff << 16, 1u << (class + 16));
 
 		/* Queued(write): expected class */
 		busy.handle = scratch;
-- 
2.23.0