Re: [PATCH v2] SUPPORT.md: extend security support for x86 hosts to 12 TiB of memory

[Julien Grall <julien@xxxxxxx>]

Hi Jan,

On 25/05/2022 12:11, Jan Beulich wrote:
> On 25.05.2022 12:58, Julien Grall wrote:
> > On 25/05/2022 10:21, Jan Beulich wrote:
> > > --- a/SUPPORT.md
> > > +++ b/SUPPORT.md
> > > @@ -50,7 +50,7 @@ For the Cortex A57 r0p0 - r1p1, see Erra
> > >    
> > >    ### Physical Memory
> > >    
> > > -    Status, x86: Supported up to 8 TiB. Hosts with more memory are supported, but not security supported.
> > > +    Status, x86: Supported up to 12 TiB. Hosts with more memory are supported, 
> > > but not security supported.
> > >        Status, Arm32: Supported up to 12 GiB
> > >        Status, Arm64: Supported up to 2 TiB
> > >    
> > > @@ -121,6 +121,17 @@ ARM only has one guest type at the momen
> > >    
> > >        Status: Supported
> > >    
> > > +## Guest Limits
> > > +
> > > +### Memory
> > > +
> > > +    Status, x86: Supported up to 8 TiB
> > > +    Status, Arm64: Supported up to 1 TiB
> > > +    Status, Arm32: Supported up to 32 GiB
> > IIRC, the max the architecture would allow us is 16 Gib. Here we are
> > limited with how much physical memory is supported by Xen. So this wants
> > to be 12 GiB.
> Hmm, while I don't know where I took the 32 from, it was you who
> suggested (in reply to v1) I put 16 here.
Hmmm... I am pretty sure I wrote 16 in v1 [1].

> Though yes, with the
> host limit now set to just 12, putting more here would be odd.
> I didn't cross check the numbers, I'll admit.
>
> > > +
> > > +Guests with more memory, but less than 16 TiB, are supported,
> > > +but not security supported.
> > On Arm32, we definitely can't support up to 16 TiB. On Arm64, we would
> > need some work to support it. So I would move this sentence in the
> > "Status, x86" section.
> Sure, I can do that. Would have been nice if you had said so right
> on v1. 
I only spotted this oddity now. Sorry.

> As to Arm64 though - the host limit is 2 TiB. Going beyond
> that being impossible (without becoming at least unsupported), is
> the uniform upper bound of 16 TiB really a problem here (IOW do
> guests really only function up to 1 TiB)? 
The guest memory layout has only been defined for up to 1TB. Hopefully 
this is the only place where the assumption is baked. But as I don't 
have such machine in hand, I can't easily confirm it.
So I think we need to clarify that the 16TB limit only applies to x86.

> For Arm32 it would be
> even less of an issue, as hosts with more than 12 GiB are
> unsupported. 
This is only obvious if you know that Xen doesn't support memory 
overcommitting. Admittely, this could be considered as basic Xen knowledge.
Cheers,

[1] 
https://lore.kernel.org/xen-devel/6ec0e3d9-374c-1caa-9889-f091dcf894e3@xxxxxxx/
--
Julien Grall