Re: [edk2-devel] [PATCH v3 00/13] OvmfPkg: Support booting from Fusion-MPT SCSI controllers

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

On 03/06/20 22:52, Liran Alon wrote:
> 
> Hi Lazlo,

("Laszlo")

> On 06/03/2020 22:14, Laszlo Ersek wrote:
>> Hi Nikita,
>>
>> On 03/04/20 20:22, Nikita Leshenko wrote:
>>> This series adds driver support for:
>>> - LSI53C1030
>>> - SAS1068
>>> - SAS1068E
>>>
>>> These controllers are widely supported by QEMU, VirtualBox and VMWare.
>>> This work is part of the more general agenda of enhancing OVMF boot
>>> device support to have feature parity with SeaBIOS.
>>>
>>> We have also developed support for PVSCSI which we will submit in a
>>> separate patch series.
>> I'd like to learn more of this general agenda ("feature parity with
>> SeaBIOS"). I have never felt the need for any SCSI controller offered by
>> QEMU other than virtio-scsi.
>>
>> Because you guys are contributing Fusion-MPT and PVSCSI to OVMF,
>> obviously such a practical need must exist ("feature parity with
>> SeaBIOS" is vague, I'm not really buying it :) ).
> 
> The motivation behind supporting booting from these devices is being
> able to run VMs that originally run on other hypervisors, such as VMware
> ESXi, on top of QEMU/KVM without any changes to the image (In contrast
> to approaches such as virt-v2v, Amazon CloudEndure or Google
> Velostrata). This technology was developed in Ravello Systems (Acquired
> by Oracle), which offered a product to run VMware-based images on top of
> any public cloud without any modification to the VMs.
> This is also why you would also see QEMU PVSCSI device emulation and
> SeaBIOS PVSCSI driver were contributed by Ravello as-well.
> 
> Similar work to Ravello was done by Virtuozzo for Hyper-V based VMs (See
> their enhancements to QEMU of emulating Hyper-V PV devices). They also
> have EDK2 drivers for these devices that are not upstream.

Thank you for the detailed explanation. Very helpful.

> 
> Having said that, it is true that these two patch-series provide feature
> parity with SeaBIOS. I do believe that it makes sense OVMF will support
> booting from any storage device that QEMU is able to emulate. Especially
> if SeaBIOS is already able to boot from that device as-well.

Yes, as long as a user can control relatively fine-grained what support
they'd like to include in the firmware binary.

> 
>>
>> So I have two requests:
>>
>> (1) please describe the actual use case (hypervisor, guest OS, maybe
>> performance, etc) for these drivers, in the associated bugzilla,
> 
> I don't think the description we have written in the relevant BugZilla
> tickets should change much. It's still true that these are contributed
> as part of feature-parity with SeaBIOS and allowing to boot from these
> devices. How one use this depends on it's own use-case. It could be to
> be able to boot VMware-based VMs as-is under QEMU/KVM, but it could also
> be used to boot a VM with a PVSCSI controller that was originally
> created with QEMU/KVM.

Right. All I'm going to do now is to link your message in the mailing
list archive in a new BZ comment (on both BZs).

> This is similar to upstream commit c137d9508169 ("OvmfPkg/QemuVideoDxe:
> VMWare SVGA device support") which added support for VMware-SVGA GPU.

Yes, I agree about the similarity.

(Two side comments about that commit:

- the commit explains the motivation in the message ("Drivers for this
device exist for some guest OSes which do not support Qemu's other
display adapters, so supporting it in OVMF is useful in conjunction with
those OSes").

- that commit ended up being reverted (even though not because the use
case got invalidated, but because it was replaced by a simpler solution
-- see commit range 7f3b0bad4bbb..d021868ccf49).)

> 
>>
>> (2) please make the inclusion of these drivers in the OVMF DSC and FDF
>> files dependent on a new build flag (-D). It's up to you whether you
>> want to gate PVSCSI and Fusion-MPT with the same flag, or if you want to
>> assign separate flags to them.
>>
>> It's fine if the default value is TRUE, for that flag (those flags). I'm
>> asking for them from a downstream perspective -- some distros follow a
>> "we ship it, we support it" model, and so they must be careful with
>> rebases to new usptream releases. I'd like to permit such downstreams to
>> easily disable these drivers, with just -D flags, without
>> downstream-only patches for the OVMF DSC and FDF files, that might need
>> repeated downstream rebasing and review.
>>
>> Again, it's perfectly fine if the upstream defaults are TRUE.
>>
>> ... If you feel tempted to point out the Xen paravirt drivers: you are
>> entirely right, but those are already covered by
>> <https://urldefense.com/v3/__https://bugzilla.tianocore.org/show_bug.cgi?id=2122__;!!GqivPVa7Brio!JJpn5PCAL9AX7xdt4qLhQJMF7L19-QEEbGF_o9bHTGsccK6a5AeDD0-_j3_UPCk$
>> >.
> 
> Funny. I was about to point out that the Xen PV drivers are not
> conditioned by any flag! :)

:)

> OK. I think it's appropriate we will add a separate flag for each device
> and make them both TRUE by default (As I think should be appropriate for
> any device built-in by default into QEMU).

Thank you very much!
Laszlo