Re: [edk2-devel] [Patch 1/2] OvmfPkg: Change default to disable MptScsi and PvScsi

["dann frazier" <dann.frazier@canonical.com>]

On Wed, Dec 07, 2022 at 11:22:10AM -0500, James Bottomley wrote:
> On Wed, 2022-12-07 at 17:04 +0100, Ard Biesheuvel wrote:
> > On Wed, 7 Dec 2022 at 17:02, Gerd Hoffmann <kraxel@redhat.com> wrote:
> > > 
> > > On Wed, Dec 07, 2022 at 09:14:39AM -0500, James Bottomley wrote:
> > > > On Wed, 2022-12-07 at 15:09 +0100, Ard Biesheuvel wrote:
> > > > > So at some point, these drivers will be removed rather than
> > > > > kept
> > > > > alive by the core team unless someone steps up.
> > > > 
> > > > How important is keeping them alive?
> > > 
> > > Most common use case is probably bootimg images created on other
> > > hypervisors on qemu.  Otherwise there is little reason to use
> > > something which is not virtio-scsi.
> > > 
> > > > I can volunteer to "maintain" them which I anticipate won't be
> > > > much effort (plus I'm used to looking after obsolete SCSI
> > > > equipment).  The hardware is obsolete, so the mechanics of their
> > > > emulation isn't going to change, the only potential risk is
> > > > changes in the guest to host transmission layer that breaks
> > > > something.
> > > 
> > 
> > Thanks James, that would be very helpful.
> > 
> > > Yes, I don't expect it being much effort, but knowing oldish scsi
> > > stuff certainly helps understanding the driver code if needed.  If
> > > you want step up sent a patch updating Maintainers.txt accordingly.
> > > 
> > 
> > Having the informed opinion of a domain expert should allow us to
> > diagnose issued related to these drivers with more confidence, and
> > also give us insight in how obsolete those drivers actually are.
> > 
> > I can send the patch if you prefer.
> 
> Sure, who can resist someone else doing all the work.
> 
> I note we do have a maintained LSI driver: OvmfPkg/LsiScsiDxe.  It
> seems to be based on the 53c896 which is really only a marginal subset
> of the 1030 ... if I'm remembering correctly the 1030 did Low Voltage
> Differential (so a faster SCSI Parallel bus), but since that's a SCSI
> Bus protocol, it should have no real impact on the utility of the
> emulation.  Is the LsiScsiDxe usable by Debian?

I tried it, but it doesn't seem to enumerate any blk devices.
The driver loads and reports that it is managing a device:

Shell> drivers
            T   D
D           Y C I
R           P F A
V  VERSION  E G G #D #C DRIVER NAME                         IMAGE NAME
== ======== = = = == == =================================== ==========
[...]
6E 00000010 D - -  1  - LSI 53C895A SCSI Controller Driver  LsiScsiDxe
[...]

But no blk devices. Using the same VM config and just swapping the
controller from lsilogic to virtio-scsi, yields the expected blk
devices.

To be clear, I'm unaware of OVMF ever working with this device in
Debian.

  -dann