On Wed, 14 May 2025, Chuck Zmudzinski wrote:
On 5/14/2025 1:33 PM, Manuel Bouyer wrote:
On Wed, May 14, 2025 at 09:54:35AM -0400, Chuck Zmudzinski wrote:
Another odd thing I saw with this system (Intel Raptor Lake i5-14500). I first
installed NetBSD 10.1 as a Xen HVM DomU on Xen 4.19.2 distributed by Fedora
Linux, with a Fedora 41 Linux PV Dom0 before trying to boot NetBSD PV Dom0 on
the pkgsrc version of Xen 4.18. In that configuration, the NetBSD GENERIC
kernel did not detect hypervisor0 at mainbus0, and instead detected hyperv0 at
cpu0, which would only be valid if the hypervisor was Microsoft Hyper-V instead
of Xen! It still boots, but it uses the emulated Qemu devices for disk and
network I/O instead of the Xen PV devices. Perhaps this is happening because,
according to the information on the NetBSD port-xen HOWTO wiki page, some
versions of Xen distributed by Linux distros don't compile Xen with
CONFIG_PV_LINEAR_PT, which is required for NetBSD PV support. But I have not
checked if Xen 4.19.2 shipped by Fedora is compiled without
CONFIG_PV_LINEAR_PT. If Fedora's Xen 4.19.2 does have CONFIG_PV_LINEAR_PT,
though, then something strange is happening on my system with Xen because I
presume NetBSD/Xen HVM should detect hypervisor0, not hyperv0, when running on
Xen with the CONFIG_PV_LINEAR_PT option.
It's not related to CONFIG_PV_LINEAR_PT, it's probably because Viridian
is enabled by defaut. You should be able to disable it on a per-guest basis.
Ah, yes I think Viridian is a Microsoft thing and I usually have it on for
Windows and Linux HVM guests. I will try turning Viridian off for the NetBSD
HVM next time I try it.
There's a patch here for fixing NetBSD's mis-reading of the CPU leaves
which causes this:
https://mail-index.netbsd.org/port-xen/2025/04/25/msg010803.html
--
Stephen
