Hi, I accidentally replied only to the sender. Here is the reply to the list.
On Fri, 6 Sep 2024, Thomas Siegmund wrote: > We at SEPPmail have been using OpenBSD as the operating system for our > SEPPmail mail gateway since > the company was founded. > For two years we have been developing and operating a cloud service for mail > processing, > encryption/decryption, spam filtering and so on. > > Yesterday we had an incident at Hetzner, a German data center operator. They > upgraded their QEMU > version (we assume to 8.2.3, but they are not willing to tell us the exact > version) and this caused > all OpenBSD 7.5 appliances to fail to boot (short boot up to initialize the > Virtio network interface > or the Vioscsi driver and then the appliance just shuts down). > We found out that this was caused by the Virtio drivers if_vio and vioscsi. > By undoing one of your > commits > https://github.com/openbsd/src/commit/cdd248411fe303b936d5a056fde97097bd7015f0 > "virtio: Set > DRIVER_OK earlier” we were able to get our service up and running again. > Hetzner themselves found out that it could be a problem related to the QEMU > commit > https://gitlab.com/qemu-project/qemu/-/commit/fcbb086ae590e910614fe5b8bf76e264f71ef304. > > Since I am not familiar with the OpenBSD kernel and the virtio system, I am > not able to figure out > what is really causing the crash/shutdown. > Maybe you can figure it out and this information will be helpful. I have seen a qemu crash related to virtio with openbsd on some qemu 9.1 rc, I think rc2. This commit fixed it: https://gitlab.com/qemu-project/qemu/-/commit/a8e63ff289d137197ad7a701a587cc432872d798 I have also tried an older version and that had the same issue, I believe that was some 9.0.2 package from debian. The fix is in qemu 9.1 release. I think it likely that this is your issue. Maybe you can ask Hetzner if they can backport the fix to their build? I don't think there is a bug in OpenBSD in this. The OpenBSD commit you mentioned fixes various problems, both with qemu and with other hypervisors. We definitely don't want to revert that commit. Cheers, Stefan
