Index | Thread | Search

From:
Kenneth Gober <kgober@gmail.com>
Subject:
Re: improve os names used in vmt(4)
To:
Jonathan Matthew <jonathan@d14n.org>
Cc:
tech@openbsd.org
Date:
Mon, 9 Nov 2026 02:33:30 -0500

Download raw body.

Thread
On Sun, Nov 8, 2026 at 8:52 PM Jonathan Matthew <jonathan@d14n.org> wrote:

> On Fri, Nov 06, 2026 at 04:02:58PM -0500, Kenneth Gober wrote:
> > On Fri, Nov 6, 2026 at 2:41 AM Jonathan Matthew <jonathan@d14n.org>
> wrote:
> >
> > > Forever ago, we started reporting some guest OS details to vmware.
> > > At the time, some of their hypervisor products didn't like getting an
> > > unknown guest OS name, so we pretended to be FreeBSD.  That doesn't
> seem
> > > to be the case any more, at least in the esx/vsphere product line.
> > > I don't have fusion or workstation, so it'd be great if someone could
> > > test there.
> > >
> > > While I was looking at this, I realised osversion should probably be
> > > removed.  'GENERIC.MP#110' or similar is not really meaningful so
> > > I think it'd be better to just report 'OpenBSD 8.0'.
> > >
> > > Index: vmt.c
> > > ===================================================================
> > > RCS file: /cvs/src/sys/dev/pv/vmt.c,v
> > > diff -u -p -r1.32 vmt.c
> > > --- vmt.c       24 May 2024 10:05:55 -0000      1.32
> > > +++ vmt.c       6 Nov 2026 05:46:36 -0000
> > > @@ -647,21 +647,15 @@ vmt_update_guest_info(struct vmt_softc *
> > >         }
> > >
> > >         if (sc->sc_set_guest_os == 0) {
> > > -               if (vm_rpc_send_rpci_tx(sc, "SetGuestInfo  %d %s %s
> %s",
> > > +               if (vm_rpc_send_rpci_tx(sc, "SetGuestInfo  %d %s %s",
> > >                     VM_GUEST_INFO_OS_NAME_FULL,
> > > -                   ostype, osrelease, osversion) != 0) {
> > > +                   ostype, osrelease) != 0) {
> > >                         DPRINTF("%s: unable to set full guest OS",
> > > DEVNAME(sc));
> > >                         sc->sc_rpc_error = 1;
> > >                 }
> > >
> > > -               /*
> > > -                * Host doesn't like it if we send an OS name it
> doesn't
> > > -                * recognise, so use the closest match, which happens
> > > -                * to be FreeBSD.
> > > -                */
> > > -
> > >                 if (vm_rpc_send_rpci_tx(sc, "SetGuestInfo  %d %s",
> > > -                   VM_GUEST_INFO_OS_NAME, "FreeBSD") != 0) {
> > > +                   VM_GUEST_INFO_OS_NAME, ostype) != 0) {
> > >                         DPRINTF("%s: unable to set guest OS",
> DEVNAME(sc));
> > >                         sc->sc_rpc_error = 1;
> > >                 }
> > >
> > >
> > With this patch, OpenBSD/amd64 VMs running under VMware ESXi 5.5 now
> > show "Guest OS: Other (32-bit)". Previously they showed "FreeBSD
> (32-bit)".
> > For whatever reason, both amd64 and i386 VMs display the same 32-bit tag.
> >
> > Aside from that small display change my OpenBSD 8.0/amd64 VM appears
> > to be running exactly as it was before. Still shuts down cleanly when
> > issuing
> > a shutdown command from vSphere Client.
> >
> > I did not test an i386 VM as I don't expect to see a different outcome
> > there.
>
> I'm pretty sure this only affects what shows up in the UI.  It gets
> reported well into the boot process so it can't affect what emulated
> hardware shows up.
>
> I think we only actually need to change the short OS name, and I suspect
> earlier versions of vsphere check for a "-64" suffix there to indicate
> the guest is 64 bit.  The vcenter 8.0 setup I'm looking at doesn't seem
> to care about that.
>
> Can you try this and see if your amd64 guest is still reported as 32 bit?
>
>
> Index: vmt.c
> ===================================================================
> RCS file: /cvs/src/sys/dev/pv/vmt.c,v
> diff -u -p -r1.32 vmt.c
> --- vmt.c       24 May 2024 10:05:55 -0000      1.32
> +++ vmt.c       7 Nov 2026 02:55:36 -0000
> @@ -654,14 +654,14 @@ vmt_update_guest_info(struct vmt_softc *
>                         sc->sc_rpc_error = 1;
>                 }
>
> -               /*
> -                * Host doesn't like it if we send an OS name it doesn't
> -                * recognise, so use the closest match, which happens
> -                * to be FreeBSD.
> -                */
> -
> -               if (vm_rpc_send_rpci_tx(sc, "SetGuestInfo  %d %s",
> -                   VM_GUEST_INFO_OS_NAME, "FreeBSD") != 0) {
> +               if (vm_rpc_send_rpci_tx(sc, "SetGuestInfo  %d %s%s",
> +                   VM_GUEST_INFO_OS_NAME, ostype,
> +#ifdef __LP64__
> +                   "-64"
> +#else
> +                   ""
> +#endif
> +                   ) != 0) {
>                         DPRINTF("%s: unable to set guest OS", DEVNAME(sc));
>                         sc->sc_rpc_error = 1;
>                 }
>

This did not work. ESXi reported "Other (32-bit)" with this diff.

I tried sending "FreeBSD-64" which was displayed as "FreeBSD (64-bit)"
so it is certainly true that the "-64" suffix works. But apparently it only
works (in ESXi 5.5 at least) when used with a 'known' prefix. I checked
and apparently "Other" is also considered a 'known' prefix, so "Other-64"
displays as "Other (64-bit)".

I suspect it's not worth the effort to fix this for VMware versions as old
as
mine. "Other (32-bit)" is still an improvement over "FreeBSD (32-bit)" (it's
now only half-wrong rather than completely wrong). I suppose it would be
possible to query the VMware host version and send a different string
depending on the result but I don't know when ESXi stopped being limited
to 'known' prefixes.

Also, I was able to test this diff on i386 and it displayed the expected
value "Other (32-bit)". So there's that at least.

-ken