VMA pinning to GGTT is now commited asynchronously in CHV / BXT+VDT
environments to avoid lock inversion among reservation_ww and cpu_hotplug
locks, the latter acquired from stop_machine(). Then,
vma->resource->page_sizes_gtt the test uses as shift count may still be
not populated (equal 0) after i915_vma_pin() returns. Wait for VMA bind
completion in those cases to avoid shift-out-of-bounds kernel warnings and
the test case failing with -EBADSLT.
v2: Explain why VMA pinning is commited asynchronously on CHV/BXT+VTD
(Krzysztof),
- use more precise wording in commit description.
Reviewed-by: Sebastian Brzezinka <[email protected]>
Reviewed-by: Krzysztof Karas <[email protected]>
Signed-off-by: Janusz Krzysztofik <[email protected]>
---
drivers/gpu/drm/i915/selftests/i915_gem_gtt.c | 4 ++++
1 file changed, 4 insertions(+)
diff --git a/drivers/gpu/drm/i915/selftests/i915_gem_gtt.c
b/drivers/gpu/drm/i915/selftests/i915_gem_gtt.c
index 7ab4c4e602648..0a86e48575394 100644
--- a/drivers/gpu/drm/i915/selftests/i915_gem_gtt.c
+++ b/drivers/gpu/drm/i915/selftests/i915_gem_gtt.c
@@ -1118,6 +1118,10 @@ static int misaligned_case(struct i915_address_space
*vm, struct intel_memory_re
goto err_put;
}
+ /* make sure page_sizes_gtt has been populated before use */
+ if (i915_is_ggtt(vm) && intel_vm_no_concurrent_access_wa(vm->i915))
+ i915_vma_wait_for_bind(vma);
+
expected_vma_size = round_up(size, 1 <<
(ffs(vma->resource->page_sizes_gtt) - 1));
expected_node_size = expected_vma_size;
--
2.51.0
