On 05/08/2015 09:50 PM, John Ferlan wrote:
On 05/05/2015 02:03 PM, Laine Stump wrote:
> All the way back at the end of 2009, commit d78554d8 added a check
> that prevented scsi disks from having a bus != 0 due to problems that
> caused (noted in the comments). At that time, -device wasn't supported
> by qemu, so the -drive parameter had to identify each disk by bus
> type (the "if" option), bus#, and unit#.
>
> Since that time qemu has added the ability to provide such details
> about a disk in a -device parameter, which will contain an id, and
> that id can be referenced in -drive rather than giving
> type/bus/unit. Since no bus= option is present in the -drive string in
> this case, the problem with a non-0 bus is presumably no longer
> present, so this patch makes the restriction active only when -device
> isn't used.
>
> qemuxml2argv-disk-scsi-lun-passthrough has been modified to set non-0
> buses to assure that the restriction is no longer valid.
> ---
> src/qemu/qemu_command.c | 13 +++++++++----
> .../qemuxml2argv-disk-scsi-lun-passthrough.args | 12 ++++++------
> .../qemuxml2argv-disk-scsi-lun-passthrough.xml | 4 ++--
> 3 files changed, 17 insertions(+), 12 deletions(-)
>
Seems reasonable to me... ACK
I think the comment in the code was unintentionally misleading (at least
to me). After testing this and looking into it further, I think I was
wrong, and that bus *should* always be limited to 0 (see the comments in
virDomainDiskDefAssignAddress()), so I'm dropping this patch.