Re: [Xen-devel] Xen on POWER

[George Dunlap <dunlapg@xxxxxxxxx>]

On Fri, Mar 9, 2018 at 11:24 AM, Andre Przywara <andre.przywara@xxxxxxx> wrote:
> Hi,
>
> On 09/03/18 10:38, awokd wrote:
>> On Fri, March 9, 2018 10:03 am, Andrew Cooper wrote:
>>> On 09/03/2018 09:37, awokd wrote:
>>
>>>
>>> Xen currently has x86 and ARM as supported architectures, so there is a
>>> reasonable split between common and arch-specific code.  As a start, you'd
>>> need to implement enough of the arch stubs to make Power9 build, then work
>>> on trying to get it to boot.
>>>
>>> As a guess, the ARM port is going to be far closer to what you're
>>> looking to do than the x86 side.
>>
>> Thanks for your reply, that helps. Would it make sense to target 4.8.3
>> stable to mostly avoid trying to hit a moving target, then port "up"? I'll
>> investigate how ARM support was added too, that's relatively fresh and
>> could be used as a model.
>>
>>> Porting is one thing, but as you identify below, not  letting it wither
>>> is very different.  I can't predict how much ongoing maintenance would be
>>> required, but it certainly can't be left alone once "complete".
>>
>> Assuming it would be a similar workload to maintaining ARM, any idea if
>> we're talking multiple FTEs here or just a fraction of one? Trying to get
>> a sense of if it's something that could be maintained in one person's
>> off-hours.
>
> If you are serious about it, you need a team. Which is about to stay
> around! At least two people, who both know the architecture *and* Xen
> well. And it will probably take them more than a year to get something
> into a state where you can rely on it in a security sensitive
> environment. It gets even worse since you probably need more advanced
> features like hardware passthrough for Qubes OS.

I'm all for encouraging people to jump into Xen, but I agree with
Andre here, that you should really count the cost.  I doubt this is
the sort of thing a single person could really write and maintain
unpaid on evenings and weekends.  Remember also that you'd be porting
not only Xen to Power9, but also Linux (or some other operating
system) to run as a Xen dom0.

It might be worth looking around to see what organizations / vendors
might be willing to take up and/or fund such an effort.  Who did the
Power9 port?  Would any Power9 vendors / chipmakers benefit
significantly from Xen on Power9?

Failing that, if there's large desire for such a thing in the
open-source world (e.g., to get something Qubes could use), a
Kickstarter campaign to allow you (and maybe someone else) to work on
it full-time might be a reasonable approach.

Wish you luck in any case! :-)

 -George

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxxx
https://lists.xenproject.org/mailman/listinfo/xen-devel