Re: Quartz64: vmm: No vgic found

  Andrew Turner wrote:

> 
> 
> > On 19 Jan 2026, at 18:03, Roman Bogorodskiy <[email protected]> wrote:
> > 
> >  Mark Johnston wrote:
> > 
> >> On Sat, Jan 17, 2026 at 06:19:47PM +0100, Roman Bogorodskiy wrote:
> >>>  Mark Johnston wrote:
> >>> 
> >>>> On Thu, Jan 15, 2026 at 06:42:10PM +0100, Roman Bogorodskiy wrote:
> >>>>> But bhyve does not start:
> >>>>> [...]
> >>>>> Does the board need something in addition to GICv3 that is missing that
> >>>>> is needed to make vmm(4) work, or is that a configuration or vmm(4)
> >>>>> issue?
> >>>> 
> >>>> The vgic driver is refusing to attach, apparently because we don't
> >>>> detect virtualization support when probing the controller.  When booting
> >>>> from a FDT we look for an "interrupts" property to determine this, but
> >>>> it's not clear to me whether that's totally right.  Can you share the
> >>>> device tree you're using with this board?
> >>> 
> >>> Not sure what's the right way to share the device tree, so I used the
> >>> command I found on wiki (https://wiki.freebsd.org/FlattenedDeviceTree),
> >>> that is: sysctl -b hw.fdt.dtb | dtc -I dtb
> >>> 
> >>> It's lengthy, so I uploaded it here:
> >>> 
> >>> https://people.freebsd.org/~novel/misc/quartz64_devicetree.txt
> >> 
> >> So, from what I can tell, gic_v3_ofw_bus_attach() should be attaching a
> >> child vgic device.  Are you loading vmm.ko after the kernel has booted?
> >> Does anything change if you load it from loader.conf, or compile vmm
> >> into the kernel?
> > 
> > I was loading vmm.ko manually, i.e. by running `kldload vmm`.
> > Nothing changes when I load it via vmm_load="YES" in /boot/loader.conf:
> > 
> > vmm: No vgic found
> > module_register_init: MOD_LOAD (vmm, 0xffff0000014391f0, 0) error 19
> > 
> > I'll test adding that to the kernel once I have an established procedure
> > of building a kernel for this board.
> 
> Can you try the patch in https://reviews.freebsd.org/D54791? It fixes an issue where the vgic driver isn’t attached if there is no ITS on the board. This appears to be the case with the Quartz64.
> 
> Andrew

This does not fix the issue for my board, behaviour remains the same.

Roman