On Thu, 11 Feb 2016 17:18:32 +0100 Sascha Silbe <si...@linux.vnet.ibm.com> wrote:
> Cornelia Huck <cornelia.h...@de.ibm.com> writes: > > > On Thu, 11 Feb 2016 10:01:35 +0100 > > Markus Armbruster <arm...@redhat.com> wrote: > > > >> Sascha Silbe <si...@linux.vnet.ibm.com> writes: > > > >> > This leaves out > >> > virtio-{gpu,input,input-hid,input-host,keyboard,mouse,tablet} because > >> > they're currently only implemented using PCI, so there's no immediate > >> > value in having them. It would nevertheless make sense to include them > >> > so they can get used already and will start to Just Work⢠on s390x > >> > once a CCW implementation appears. I can post the corresponding patch > >> > if there's any interest. [For laughs and giggles, I have wired up all of these devices for ccw. Excluding input-host (for which I did not have a suitable evdev), I can specify the various devices on the commandline and get some devices in the guest which do... nothing :) I won't pursue this further for now, as I currently don't have a convincing use case beyond "because we can".] > >> > >> I guess that's for the virtio people to decide. I'm cc'ing some. > > > > What would the error look like if one decided to use e.g. virtio-gpu on > > s390x? If the error is more specific (i.e. virtio-gpu-ccw does not > > exist vs. virtio-gpu does not exist), I think adding the aliases makes > > sense: the user sees what is actually missing. > > Interesting point. Indeed, if we already define the matching -ccw > aliases, the error message may be slightly more useful: > > silbe@oc4731375738:~/recoverable/qemu$ s390x-softmmu/qemu-system-s390x > -device virtio-gpu > qemu-system-s390x: -device virtio-gpu: 'virtio-gpu-ccw' is not a valid device > model name > > Though we should probably at least add a comment to the alias list > mentioning that this is intentional. We might even want to adjust > qdev_get_device_class() to print a more specific error message in this > case. qemu-system-s390x: -device virtio-gpu: 'virtio-gpu' (alias 'virtio-gpu-ccw') is not a valid device model name would make it obvious that some alias expansion had been going on. I think that would be useful.