[libvirt-users] This QEMU doesn't support the LSI 53C895A SCSI controller

[Gionatan Danti]

Hi all,
trying to edit a domain xml to enable an LSI SCSI controller I get the
following error:

error: unsupported configuration: This QEMU doesn't support the LSI
53C895A SCSI controller

It is my understanding that the error is raised due to RedHat disabling
this controller in its own qemu-kvm builds. This seems an unfortunate
decision, as it makes harder to migrate from VMWare (which uses LSI SCSI
and SAS adapters) to RH+KVM.

Can anyone elaborate on this decision? It is possible to enable LSI
controller support in RedHat 7.x? Should I open a BZ against it?

Thanks.

[Andrea Bolognani]

Post by Gionatan Danti
Hi all,
trying to edit a domain xml to enable an LSI SCSI controller I get the
error: unsupported configuration: This QEMU doesn't support the LSI
53C895A SCSI controller
It is my understanding that the error is raised due to RedHat disabling
this controller in its own qemu-kvm builds. This seems an unfortunate
decision, as it makes harder to migrate from VMWare (which uses LSI SCSI
and SAS adapters) to RH+KVM.
Can anyone elaborate on this decision? It is possible to enable LSI
controller support in RedHat 7.x? Should I open a BZ against it?

Your assessment looks correct, and the controller is indeed compiled
out downstream. Filing a BZ sounds like a reasonable next step, but
you might also want to investigate virt-v2v, which I believe will
take care of switching to the more performant virtio-scsi (including
installing the necessary drivers) for you when moving guests off
VMware.

[Gionatan Danti]

Post by Andrea Bolognani
Your assessment looks correct, and the controller is indeed compiled
out downstream. Filing a BZ sounds like a reasonable next step, but
you might also want to investigate virt-v2v, which I believe will
take care of switching to the more performant virtio-scsi (including
installing the necessary drivers) for you when moving guests off
VMware.

Hi Andrea, thanks for the reply.

From my understanding, virt-v2v use virt-win-reg to open the disk image
and install new drivers in the registry. I am doing a very similar thing
by manually using virt-win-reg to add the mergeide.reg file to enable
booting from any IDE controller.

That said, both virt-v2v and manual virt-win-reg have a significant
shortcoming: when used on non-cleanly-shutdown images, the act of
opening the NTFS filesystem alters the journal/filesystem state. In one
planned lab test I found some file to be corrupted after running
virt-win-reg, and I traced back the cause to the unclean NTFS
shutdown/mounting.

Adding a supported controller will prevent any issue: even with an
unclean shutdown, the machine will boot and replay its NTFS journal. If
needed, it can also run the proper Windows filesystem checking tool
(chkdsk) which is, as far I know, better positioned to correct any
filesystem problem.

Am I missing something?
Thanks.

[Andrea Bolognani]

Post by Gionatan Danti

Post by Andrea Bolognani
Your assessment looks correct, and the controller is indeed compiled
out downstream. Filing a BZ sounds like a reasonable next step, but
you might also want to investigate virt-v2v, which I believe will
take care of switching to the more performant virtio-scsi (including
installing the necessary drivers) for you when moving guests off
VMware.

Hi Andrea, thanks for the reply.
From my understanding, virt-v2v use virt-win-reg to open the disk image
and install new drivers in the registry. I am doing a very similar thing
by manually using virt-win-reg to add the mergeide.reg file to enable
booting from any IDE controller.
That said, both virt-v2v and manual virt-win-reg have a significant
shortcoming: when used on non-cleanly-shutdown images, the act of
opening the NTFS filesystem alters the journal/filesystem state. In one
planned lab test I found some file to be corrupted after running
virt-win-reg, and I traced back the cause to the unclean NTFS
shutdown/mounting.
Adding a supported controller will prevent any issue: even with an
unclean shutdown, the machine will boot and replay its NTFS journal. If
needed, it can also run the proper Windows filesystem checking tool
(chkdsk) which is, as far I know, better positioned to correct any
filesystem problem.
Am I missing something?

I have very little knowledge of how virt-v2v achieves its goals and
even less when it comes to running Windows guests, so I'm afraid I'm
in no position to help you.

I'd suggest asking on the libguestfs mailing list.