LSI 1068 SAS scsihw

[dblacknc]

I have a unique FreeBSD VM that will only work with the LSI 1068 SAS qemu virtual device - mptsas1068 or in the build that ships with PVE 8.0, lsisas1068.

I patched PVE to add the device and the VM is working well for my purposes. I'm not sure how stable this device is though, as I also tried booting an Ubuntu 22.04 VM using it for the SCSI adapter, and the mptsas driver paniced late in the boot process. The default LSI SCSI device isn't exactly new code either.

Is there any interest in/willingness to add the LSI 1068 SAS driver to PVE? Happy to provide a complete proposed patch. It's a little different than what I see assumed in the code for current (virtual) SCSI devices: Each of the eight ports per controller can host two LUNs/drives, for a max of 16 per controller. It also requires at least in Linux that each disk has a WWN specified for them to be detected. (FreeBSD doesn't.) I haven't yet looked into whether/how automatically or at least easily adding WWNs to virtual SCSI disks is in PVE. I know they can manually be added.

 

[fiona]

Hi,

I have a unique FreeBSD VM that will only work with the LSI 1068 SAS qemu virtual device - mptsas1068 or in the build that ships with PVE 8.0, lsisas1068.

I patched PVE to add the device and the VM is working well for my purposes. I'm not sure how stable this device is though, as I also tried booting an Ubuntu 22.04 VM using it for the SCSI adapter, and the mptsas driver paniced late in the boot process. The default LSI SCSI device isn't exactly new code either.

yes, but our configuration files are not versioned, so we cannot change the default for backwards compatibility reasons (e.g. for restoring an old backup where the option is not explicitly set, it still needs to use the old default).

Is there any interest in/willingness to add the LSI 1068 SAS driver to PVE? Happy to provide a complete proposed patch.

I'm not aware of anybody else requesting support for this yet and each new feature adds a maintenance burden. You can always add a feature request on the https://bugzilla.proxmox.com/ first and see if anybody else would be interested. If there is enough interest, see https://pve.proxmox.com/wiki/Developer_Documentation for how to submit patches. Then you'd need to sign the harmony CLA too: https://pve.proxmox.com/wiki/Developer_Documentation#Software_License_and_Copyright

It's a little different than what I see assumed in the code for current (virtual) SCSI devices: Each of the eight ports per controller can host two LUNs/drives, for a max of 16 per controller. It also requires at least in Linux that each disk has a WWN specified for them to be detected. (FreeBSD doesn't.) I haven't yet looked into whether/how automatically or at least easily adding WWNs to virtual SCSI disks is in PVE. I know they can manually be added.

It's not yet exposed, as described here: https://bugzilla.proxmox.com/show_bug.cgi?id=4792 and sounds like a prerequisite for us to accept the change if the controller doesn't work with Linux otherwise.

 

[RolandK]

yes please !

i'd be also very happy about addition of that virtual device.

it's in qemu since 2.6.0 ( https://listman.redhat.com/archives/libvir-list/2016-February/msg00575.html ) , so i wonder why it's not already available

it's one of the most popular sas controller / chips in older servers generation and i'm sure this would also simplify physical to virtual migrations from older machines/servers.

 

[fiona]

it's one of the most popular sas controller / chips in older servers generation and i'm sure this would also simplify physical to virtual migrations from older machines/servers.

Well, I'd only recommend using an emulated controller if it really is needed by the VM. Hardware emulation is more complex and I'd expect less performant than just using a VirtIO SCSI controller.

 

[RolandK]

yes, sure.

for example we have some very ancient legacy VMs (centos5) which were migrated from xenserver , and i remember i struggled with virtual scsi controllers for a while to get them boot and run without problem - and i guess i would have loved availability of lsi1068 because that is well known to me