Re: QEMU-KVM offers OPAL firmware interface? OpenBSD guest support?

[Cédric Le Goater]

Hello,

On 8/27/21 7:00 PM, Greg Kurz wrote:
> On Fri, 27 Aug 2021 18:48:04 +0200
> Mark Kettenis <mark.kettenis@xs4all.nl> wrote:
> 
>>> Date: Fri, 27 Aug 2021 18:02:59 +0200
>>> From: Greg Kurz <groug@kaod.org>
>>>
>>> On Fri, 27 Aug 2021 15:12:31 +0000
>>> Joseph <joseph.mayer@protonmail.com> wrote:
>>>
>>>> On Friday, August 27th, 2021 at 11:01 PM, Greg Kurz <groug@kaod.org> wrote:
>>>>> Linux knows how to drive both powernv and pseries platforms.
>>>> ..
>>>>> OpenBSD might have to implement proper guest-side pseries
>>>>> support to run as a guest under an hypervisor on POWER. I don't
>>>>> know OpenBSD but this likely a huge effort.
>>>>>
>>>>> More details in the "Linux on POWER Architecture Reference":
>>>>> https://openpowerfoundation.org/wp-content/uploads/2020/07/LoPAR-20200611.pdf
>>>>> and under the arch/powerpc/platforms/pseries/ directory in the linux
>>>>> kernel sources.
>>>>
>>>> Hi Greg,
>>>>
>>>> Thanks for following up.
>>>>
>>>> (Meanwhile I posted this Q to KVM-PPC and QEMU-PPC too. On the latter it's 
>>>> https://lists.nongnu.org/archive/html/qemu-ppc/2021-08/msg00416.html and 
>>>> on the further it didn't register yet.)
>>>>
>>>> First I believe KVM-QEMU and "PowerKVM" are different - the latter was
>>>> a KVM fork maintained by IBM. The further is just the KVM and QEMU
>>>> repo. Did IBM their contributions so PowerKVM was upstreamed, you tell
>>>> me.
>>>>
>>>
>>> Yes you're right that PowerKVM was an IBM internal fork of KVM and
>>> QEMU, but it was discontinued ages ago. All the relevant bits have
>>> been merged upstream and all development since then happens upstream.
>>> So it doesn't make a big difference now to use PowerKVM instead of
>>> QEMU-KVM on POWER.
>>>
>>>> Are you saying that KVM-QEMU has all relevant Power9 support
>>>> already, for a Linux host OpenBSD as a guest on bare metal
>>>> ("powernv" mode)?
>>>>
>>>
>>> There's some confusion here. In bare metal, you just have a single
>>> instance of the OS running unvirtualized directly on the host.
>>> KVM-QEMU isn't involved at all in this case. According to the OpenBSD
>>> statement, you can already install and run it in this mode on an
>>> OpenPOWER system.
>>
>> Correct.
>>
>>>> In this case why would there be any relevance in OpenBSD implementing
>>>> pseries.
>>>>
>>>
>>> If you want to also run OpenBSD inside a VM, then OpenBSD must
>>> implement proper support to be able to run in the paravirtualized
>>> PAPR environment provided by KVM-QEMU on POWER. The OpenBSD statement
>>> seem to indicate this is missing. Nothing special "should" be needed
>>> on the KVM-QEMU side.
>>
>> Indeed.  OpenBSD/powerpc64 currently does not implement PAPR
>> environment support.  I've looked at it at some point and I don't
>> think adding support for it would be impossible.  But I only have
>> powernv hardware.  I suppose I could run KVM-QEMU on Linux on that but
>> my time is limited.
>>
> 
> Yeah, KVM-QEMU can run on powernv hardware. So you could use it
> as a development environment to implement PAPR support in OpenBSD.
> This is clearly not something achievable if you only have limited
> time though.
> 
>> It is possible to use QEMU to emulate a powernv machine and in
>> principle you can run OpenBSD in that environment on a Linux host.
>> But that's emulation and not virtualization.
> 
> Yeah, no KVM in this case and it will be extremely slow.

KVM is supported under the QEMU PowerNV machine. I would not say it 
is fast (clearly not) but guests start under the emulated hypervisor.
The emulated hypervisor it self can sustain a network bandwidth of 
10MB/s running on a small laptop which is quite fast for emulation.

Anyhow, the QEMU PowerNV is a dev/test platform and not a production 
one 

We do use the pseries XIVE emulation in prod on some P9 systems.

>> It doesn't seem to work
>> though and the installer hangs after printing its first message.

Can you send the command line ? 

>> I suspect QEMU's emulation of the XIVE interrupt controller isn't 100%
>> faithful.  

It is for pseries. PowerNV takes some shortcuts but it has enough support.  

>> It's also really slow when running on an amd64 machine, 

Compared to real HW. Yes :)

>> so I didn't investigate further.

Here are some docs :

  https://qemu.readthedocs.io/en/latest/system/ppc/powernv.html

But if you have a real POWER9 system available, I would just install 
the latest debian, ubunutu or fedora distros to run KVM guests. All 
is merged. Working your way through the PAPR interface is going to 
take some time, though. see the pseries platform under Linux. 

I thought the BSD folks were working on POWER9 baremetal support,
PowerNV platform, on top of OPAL/skiboot. Is that completed ? 

Thanks,

C.