xl support for PVSCSI
xm/xend supports PVSCSI, so xl should support it aswell.
PVSCSI is useful for passing thru SCSI/FC/SAS tape drives, autoloaders, tape robots, backplane/chassis management devices, disk LUNs, or any other SCSI device to a Xen VM. In addition to passing thru single devices/LUNs PVSCSI can also passthru all devices/LUNs on a SCSI/FC/SAS host controller. PVSCSI can also be used for NPIV.
More information about PVSCSI:
PVSCSI requires backend driver in dom0 kernel, and frontend driver in guest kernel. Suse's xenlinux forward-ported patches contain PVSCSI drivers, also for HVM guests (unmodied_drivers). Konrad's git tree has initial port of PVSCSI to pvops. PVSCSI drivers should also be reviewed, cleaned up and upstreamed.
James Harper's Windows GPLPV drivers also include working PVSCSI frontend driver for Windows VMs. James is using it himself for passing thru a tape drive from dom0 to a Windows VM, and running a backup software in the VM.
James was semi-interested in helping with xl support for PVSCSI.
This is on our roadmap, but probably for the 4.4 release.
Well it's certainly true that there are no libxl/xl patches. But when we discussed it a few weeks ago, it seemed that while there was PVSCSI in the "classic Xen" kernels (e.g., SuSE's), there was no support in pvops kernels. This means that implementing the libxl/xl side of things would be of limited use.
Florian Heigl commented
My understanding would be there's no patches for *xl* yet.
I put points on this because it feels a very "clean" way of presenting devices to a domU. I can think of some more cases, i.e. NPIV assigned luns.
If I understand correctly, there are PVSCSI patches for pvops floating around, but they haven't been upstreamed yet; is that correct?