Hi all,
I'm posting this for the benefit of OpenBSD community members hosting
virtual machines on Vultr.
I encountered the same failure that Claus A. reported in December:
cd*.iso reboot loop (vultr, Skylake AVX MDS)
https://www.mail-archive.com/misc@openbsd.org/msg180415.html
To summarize, the VM halts with this kernel fault during boot, shortly
before running fsck:
kernel: privileged instruction fault trap, code=0
Stopped at mds_handler_skl_avx+0x33: clflush
__ALIGN_SIZE+0x500(%rid,%rax,8)
It then drops in to ddb.
As in Claus's case, the solution was to ask Vultr support to migrate the
guest to another hypervisor. The representative assigned to my case was
unable to explain the root cause.
This is the second incident in the past month that resulted in a lengthy
service outage caused by a Vultr misconfiguration. In the previous
incident, the representative told me, "OpenBSD has very special
configurations that are required on our end to work properly with our
virtualization software". It lowers my confidence in Vultr as a reliable
OpenBSD host.
