Re: [Qemu-devel] q35 and sysbus devices

[Marcel Apfelbaum]

On 03/22/2017 10:46 PM, Laszlo Ersek wrote:
> On 03/22/17 21:31, Eduardo Habkost wrote:
> > Hi,
> >
> > I am investigating the current status of has_dynamic_sysbus and
> > sysbus device support on each of QEMU's machine types. The good
> > news is that almost all has_dynamic_sysbus=1 machines have their
> > own internal (often short) whitelist of supported sysbus device
> > types, and automatically reject unsupported devices.
> >
> > ...except for q35.
> >
> > q35 currently accepts all sys-bus-device subtypes on "-device",
> > and today this includes the following 23 devices:
> >
> > * allwinner-ahci
> > * amd-iommu
> > * cfi.pflash01
> > * esp
> > * fw_cfg_io
> > * fw_cfg_mem
> > * generic-sdhci
> > * hpet
> > * intel-iommu
> > * ioapic
> > * isabus-bridge
> > * kvmclock
> > * kvm-ioapic
> > * kvmvapic
> > * SUNW,fdtwo
> > * sysbus-ahci
> > * sysbus-fdc
> > * sysbus-ohci
> > * unimplemented-device
> > * virtio-mmio
> > * xen-backend
> > * xen-sysdev
> >
> > My question is: do all those devices really make sense to be used
> > with "-device" on q35?
>
> I think fw_cfg_io and fw_cfg_mem should be board-only devices (no
> -device switch).
>
> Regarding cfi.pflash01, I think originally it would have been nice to
> specify pflash chips with the modern (non-legacy) syntax, that is,
> separate -drive if=none,file=... backend options combined with -device
> cfi.pflash01,drive=... frontend options. However, that ship has sailed,
> even libvirt uses -drive if=pflash for these, and given the purpose we
> use pflash chips for, on Q35, I don't see much benefit in exposing
> cfi.pflash01 with a naked -device *now*.
>
> Re: virtio-mmio, I don't think that should be available on Q35 at all.
>
> I can't comment on the rest.

Hi Eduardo,
Thanks for finding these problems.

We should ping all maintainers of the above devices, the best way to do it
is to add the "cannot_instantiate_with_device_add_yet = true" and ask 
maintainers
to agree (or not) on that.

Thanks,
Marcel

> Thanks
> Laszlo
>
> > Should we make q35 validate dynamic sysbus
> > devices against a whitelist, like the other has_dynamic_sysbus
> > machines?
> >
> > I'm asking this because I will resume work on the
> > "query-device-slots" command, which will report supported sysbus
> > devices too. And I don't want the new command to report any
> > devices that it shouldn't.