Quoting Tvrtko Ursulin (2018-02-19 13:05:28)
> But I don't think it will fix the issue in GuC mode, at least it didn't
> in my testing. For some reason in GuC mode hangs still happen easily. Do
> you actually expect i915 may decide to move the VMA in this test, just so?
No. With EXEC_OBJECT_PINNED it has to be at the same location or else
the gem_execbuf will fail. There may still be a discrepancy if the
__submit_spin_batch() while the spinner is busy fails to capture the
busy spinner, but we have to delay longer than the GPU hang before
__submit_spin_batch() so incredibly unlikely.
Not sure what to think of the guc hangs yet.
Intel-gfx mailing list