Status of virtio-scsi?

[rterzi]

I've run into a problem with Virtio-SCSI: VirtualBox 7.0.14 on RHEL 9.3, with RHEL 9.3 guest. I thought paravirtualized drivers were supposed to deliver improved performance/less overhead by reducing the amount of emulation needed. But apparently I didn't do enough research:

The current user manual in section 5.1 says "Currently virtio-scsi controller support is experimental." (However this is in the text and not in a warning call-out box, so somewhat easy to miss.)

However, Section 14.1, Experimental Features, does not list virtio-scsi as experimental.

this inconsistency appears to be a documentation bug So, What's the status of virtio-scsi and which section should be updated?

----

In case anyone asks, the error that I saw on the guest VM:

Code: Select all

Feb 23 03:59:08 msw kernel: virtio_scsi virtio1: request:id 825 is not a head!
Feb 23 03:59:39 msw kernel: sd 0:0:0:0: [sda] tag#716 FAILED Result: hostbyte=DID_BAD_TARGET driverbyte=DRIVER_OK cmd_age=0s

After that sda was essentially offline and the guest was unusable because it lost the root disk. It still responded pings.

However, there was nothing in the guest's VBox.log or the host's syslog, so I don't know quite how to debug/collect enough info for a bug report. Though googling the log messages brought me to other mentions of problems with virtio-scsi.

[HulkySulky]

Good question. I too would like to know if they are people out there who use virtio-scsi on a regular basis, or if it's still experimental. I briefly played with it back in 6.1, but experienced too many crashes and forgot about it.

[klaus]

It is still somewhat experimental (and therefore the remark in the documentation is appropriate). The code is still quite new, with a lot of changes in the generic support for virtio 1.0 and later with VirtualBox 7.0. What existed in 6.1 had a significantly different codebase, and not all of the changes were in the 'right direction'.

The testing with virtio-scsi drivers other than the Linux ones (and even with the Linux ones over the full range of kernels which could deal with it) was so far limited.

[rterzi]

It is still somewhat experimental (and therefore the remark in the documentation is appropriate).

@klaus - Thank you for your reply.

So should virtio-scsi be listed in the User Manual under Known Limitations / Experimental Features (Section 14.1)?

It is not listed there.

The only mention is in section 5.1, " Currently virtio-scsi controller support is experimental.". This is NOT in a warning call-out box but sandwiched between warning call-out boxes where it is easy to miss.

I figured I should ask before filing a doc bug.

[klaus]

I wrote "It is somewhat experimental" which was meant as a hint that it's from my perspective correct to not list it as "fully experimental". It has been seen working, but there might be a need for more polishing to make all possible guest OSes happy.