[libvirt] Fwd: [PATCH] qemu: Add cmd_per_lun, max_sectors to virtio-scsi

Nicholas A. Bellinger nab at linux-iscsi.org
Thu Jul 3 02:08:35 UTC 2014

On Wed, 2014-07-02 at 18:48 -0600, Eric Blake wrote:
> On 07/02/2014 06:28 PM, Nicholas A. Bellinger wrote:
> >> QEMU is not the only hypervisor that libvirt targets, so tieing libvirt
> >> names to QEMU names is a non-goal. We pick the names that make most sense
> >> in the context of libvirt.
> >>
> > 
> > Not sure I follow..  virtio-scsi is specific to QEMU/KVM, and per the
> > comment in the original patch:
> > 
> >   'Currently it only supports attribute <code>queues</code> (<span class="since">1.0.5</span>, QEMU and KVM only)'
> > 
> > would seem to indicate the parameter names are only used in the context
> > of QEMU/KVM, no..?
> Just because qemu is the only hypervisor driver that _currently_ uses
> the feature doesn't preclude the libxl hypervisor from _also_ gaining
> support for the feature in a future libvirt release, at which point the
> documentation would mention the new version number for the additional
> use of the feature.  Again, the name qemu chose is not necessarily the
> best name compared to what it might map to in libxl or any other
> hypervisor, so libvirt tries to pick names that are consistent with
> other libvirt terms, even if they don't match underlying qemu names.

It's worth mentioning that these names are coming directly from the
Linux SCSI subsystem tunables that are read from virtio-scsi config
space and set during drivers/scsi/virtio_scsi.c:virtscsi_probe() init
time, and did *not* originate with QEMU. 

> > 
> > If the virtio-scsi parameters are intended to be used across
> > hypervisors, then matching them to QEMU's own names doesn't really
> > matter.  But if they are specific to virtio-scsi and only used by
> > QEMU/KVM instances, then renaming them to something arbitrary to libvirt
> > is pointless and confusing.
> virtio is not necessarily a qemu-only concept.

The point is that if a potential port of virtio-scsi to Xen ever
happens, it will use the same parameter names that every Linux SCSI LLD
has been using for the last 20 years.  Eg: They will be the same
parameters names that the virtio-scsi LLD pulls out of configuration
space, regardless of if it's KVM or XEN.


More information about the libvir-list mailing list