Re: [RFC PATCH 1/1] OvmfPkg: Introduce LSI 53C895A SCSI Controller Driver

["Laszlo Ersek" <lersek@redhat.com>]

On 06/17/20 04:09, Gary Lin wrote:
> On Mon, Jun 15, 2020 at 01:27:22PM +0200, Laszlo Ersek wrote:
>> Hello Gary,
>>
> Hi Laszlo,
> 
>> On 06/12/20 12:04, Gary Lin wrote:
>>> This commit introduces the driver for LSI 53C895A SCSI Controller
>>> which, so that OVMF can access the devices attached to the emulated
>>> "lsi" SCSI controller.
>>>
>>> Cc: Jordan Justen <jordan.l.justen@intel.com>
>>> Cc: Laszlo Ersek <lersek@redhat.com>
>>> Cc: Ard Biesheuvel <ard.biesheuvel@arm.com>
>>> Signed-off-by: Gary Lin <glin@suse.com>
>>> ---
>>> Although lsi is now considered obsolete in QEMU, I wrote this driver
>>> mainly for learning the UEFI SCSI driver and ... FUN! The majority of
>>> the code is actually borrowed from VirtioScsci and MptScsi, and I
>>> mainly focus on LsiScsiPassThru().
>>>
>>> If it's fine to add this driver into OvmfPkg, I'll start to split this
>>> patch into smaller pieces to make it easier to review.
>>
>> (1) Do we have an official deprecation notice for this SCSI controller
>> in QEMU?
>>
>> If we do, then (AIUI) the controller will be removed in QEMU in one or
>> two releases, so this code would become effectively dead in the mid
>> term. I wouldn't like to review and/or carry code that's soon to be
>> dead.
>>
> I just vaguely remember that virtio-scsi is the new default over lsi and
> it's not recommended to use lsi except for the old OS without virtio-scsi
> driver.

So the question is then whether the old OS that supports LSI but not
virtio-scsi is capable of booting on UEFI in the first place.

> 
>> (2) If there is no official deprecation notice in QEMU, then I agree it
>> makes sense to include this driver. In that case, I have another
>> question:
>>
>> Do you intend this driver for production purposes? I.e., do you expect
>> users or "layered products" (libvirt, proxmox, openstack, ...) to use
>> this SCSI controller for some well-defined purpose? (The MPT SCSI and PV
>> SCSI drivers had a clear product-oriented modivation:
>>
>>   https://edk2.groups.io/g/devel/message/55620
>>   http://mid.mail-archive.com/a96b6b74-c35d-e291-2122-9d77f1d5f89c@oracle.com
>> )
>>
>> (2a) If this driver is not meant for a production environment, then
>> LSI_SCSI_ENABLE should be FALSE by default (and I'll do a lighter
>> review).
>>
>> (2b) If the driver is meant for production, then LSI_SCSI_ENABLE should
>> indeed be TRUE, and I'll have to be more diligent in reviewing this.
>>
> I kind of wonder if the any serious use for lsi+ovmf. If the OS is so
> old and only supports lsi, seabios probably serves it better.

Right, my point exactly!

> Anyway,
> I'll try to gather more information from my colleagues to see if they
> got any feedback about lsi from our customers. If not, I'll set
> LSI_SCSI_ENABLE to FALSE by default.

Sounds like a good approach!

> 
>> For either (2a) or (2b), please do split up the driver into smaller
>> patches, and please also add yourself to Maintainers.txt as the
>> designated reviewer of the new driver.
>>
> Sure. Will do that.

Thanks!
Laszlo