I come bearing data!
After a kind of long bisect, the bad commit was:
commit 1e886639791762e89b51aa0507f523c6a1448831
Author: Paolo Bonzini <pbonz...@redhat.com>
Date: Thu Jun 29 15:27:41 2017 +0200
vdi: make it thread-safe
And then, when I went to verify it on git master before reporting it
against qemu upstream, I couldn't reproduce it against
894fc4fd670aaf04a67dc7507739f914ff4bacf2.
Another round of bisecting points to
commit 050de36b13f7a841b7805391bca44f36370e86e4
Author: Paolo Bonzini <pbonz...@redhat.com>
Date: Thu Mar 25 12:29:39 2021 +0100
coroutine-lock: Reimplement CoRwlock to fix downgrade bug
as the first commit where it works reliably. Which is unfortunate, as
cherrypicking that looks a bit more invasive than is probably
reasonable.
- Rich
