On 05/14/2015 12:25 PM, Laine Stump wrote:
On 05/05/2015 02:03 PM, Laine Stump wrote:
libvirt enforces bus (channel to qemu) == 0 for those qemu binaries that don't support the channel argument to scsi disk devices, but still adds "channel=0" in those cases. Apparently nobody with a qemu old enough to not support channel ever uses these devices, because they otherwise should get an error.
This patch only adds channel when the scsi-disk.channel capability is set for the qemu binary. Most of the change in the patch is updating patches to either 1) remove channel=0 from the .args file of the test, or 2) reposition channel=0 to precede the bus arg (because this makes the code cleaner/simpler) and change the DO_TEST() invocation for the test to add QEMU_CAPS_SCSI_DISK_CHANNEL.
I'm uncommitted about whether or not it is worthwhile to push this patch. On one hand, I'm fairly certain this is what is correct; on the other hand nobody has ever complained about it, and at this point almost everyone is using new enough qemu that it supports channel. --- src/qemu/qemu_command.c | 7 +++--- .../qemuxml2argv-disk-drive-network-iscsi-lun.args | 2 +- .../qemuxml2argv-disk-scsi-disk-split.args | 8 +++---- .../qemuxml2argv-disk-scsi-disk-vpd.args | 2 +- .../qemuxml2argv-disk-scsi-disk-wwn.args | 4 ++-- .../qemuxml2argv-disk-scsi-lun-passthrough.args | 4 ++-- .../qemuxml2argv-disk-scsi-megasas.args | 2 +- .../qemuxml2argv-disk-scsi-virtio-scsi.args | 2 +- .../qemuxml2argv-disk-scsi-vscsi.args | 2 +- .../qemuxml2argv-disk-virtio-scsi-ccw.args | 2 +- .../qemuxml2argv-disk-virtio-scsi-cmd_per_lun.args | 2 +- .../qemuxml2argv-disk-virtio-scsi-max_sectors.args | 2 +- .../qemuxml2argv-disk-virtio-scsi-num_queues.args | 2 +- .../qemuxml2argv-pseries-vio-user-assigned.args | 2 +- .../qemuxml2argvdata/qemuxml2argv-pseries-vio.args | 2 +- tests/qemuxml2argvtest.c | 25 +++++++++++++--------- 16 files changed, 38 insertions(+), 32 deletions(-)
I'm in agreement with you vis-a-vis whether it's worthwhile. Nothing's broken, so the axiom may apply. However, we don't know if it's broken "under the covers" since we don't know the affect of adding channel=# to a qemu that doesn't support it since we don't seem to have one of those today.
OTOH since you added the QEMU_CAPS_SCSI_DISK_CHANNEL bit in and that alone didn't break anything, then it would appear this change should be OK. Other than changing the order/placement of channel (which seems to be a non-issue), it seems these changes are fine.
Should I assume you've started guests with all these changes? And perhaps restart libvirtd to make sure noone disappears? I did restart libvirtd with a running guest with SCSI disk, no problems, and then destroyed and restart the guest (so it uses the different argument ordering), no problems. I don't have a reasonable way to test
On 05/08/2015 09:44 PM, John Ferlan wrote: the case when channel= isn't supported, since that has even been backported to RHEL6.
quasi-ACK, your call depending on your comfort level. I'm planning on pushing this later today, if there are no objections.
Along with RHEL6/CentOS6 supporting channel, RHEL5 is old enough that it doesn't even support -device, so I have nothing of the proper vintage to test for non-support of channel, and this makes me doubt that anybody anywhere has a system that would be affected by this patch, so I'm dropping it too.