Re: Kernel modules on 10.1

[Manuel Bouyer <bouyer%antioche.eu.org@localhost>]

On Tue, Mar 18, 2025 at 05:28:32PM +0000, Stephen Borrill wrote:
> On Tue, 18 Mar 2025, Greg Troxel wrote:
> > Stephen Borrill <netbsd%precedence.co.uk@localhost> writes:
> > 
> > > > I would say it's a bug in 10 that *if* KDTRACE_HOOKS changes the module
> > > > ABI, that it isn't enabled in XEN3_DOMU.  Quick grepping makes me think
> > > > it's enabled there in current but not 10.
> > > 
> > > Yes, it looks like KDTRACE should be in all kernel configurations that
> > > are ever likely to load the solaris module.
> > > 
> > > I wlil build and test and then submit a patch to be pulled up.
> > 
> > Sounds good.
> > 
> > > > However, on a netbsd-10 domU, modload zfs works ok.
> > > 
> > > It doesn't on a PV kernel, but it will with a GENERIC kernel running
> > > as PVHVM.
> > 
> > Indeed, my 10 systems are pvhvm, both i386 and amd64, and I forgot about
> > that.
> > 
> > You have found a legit bug so I don't meant to say you shouldn't be
> > using PV, but I am curious what leads you to run PV instead of PVHVM.
> > I found that PVHVM was about twice as fast.
> 
> 1) PVHVM wasn't an option before 10
> 2) All the existing VMs have disk layouts without a GPT or MBR, so no
> bootloader (kernel loaded from domU root by XenServer using pygrub)
> 3) PVHVM breaks CDs (see other email from earlier), so you can't actually
> install.

PVH would solve issues 2 and 3

On the other hand, on older hardware, I found PV to be faster than PVH or
PVHVM

-- 
Manuel Bouyer <bouyer%antioche.eu.org@localhost>
     NetBSD: 26 ans d'experience feront toujours la difference
--