Hi Marc-André, Akihiko, On Sun, 15 Nov 2026 11:58:52 +0400, Marc-André Lureau wrote: > Could you fix vhost-user-gpu in the same series? > > Should we review & merge your series before this one?
>From the testing side, I should clarify the scope of my earlier Tested-by for v3 in light of Akihiko's comments on 2/8-4/8. My November 13 tests covered the original four reported failures, stop/cont, guest boots and resets with a D-Bus listener attached, including five consecutive system_reset/boot cycles, and finalization of never-realized instances. They did not race a guest-initiated reset with QMP stop/cont, or stop the control queue while GET_DISPLAY_INFO or GET_EDID was waiting for the frontend's reply. The passing results do not cover those two cases raised on 2/8 and 4/8. For 3/8, the GL cleanup on the vCPU thread is the same path that caused the D-Bus tex_id assertion in my v1 testing. Folding 3/8 and its main-loop fix together makes sense so the intermediate commit is usable too. Thanks for catching these gaps. I can retest the follow-up once the reset changes are settled, including the reset/stop race and stopping the control queue with a frontend reply pending. Dmitry
