Re: [RFC PATCH v2 16/22] i386/xen: handle VCPUOP_register_vcpu_info

[David Woodhouse <dwmw2@infradead.org>]

[-- Attachment #1: Type: text/plain, Size: 1693 bytes --]

On Wed, 2022-12-14 at 10:28 +0000, Paul Durrant wrote:
> On 13/12/2022 00:13, David Woodhouse wrote:
> > On Mon, 2022-12-12 at 14:58 +0000, Paul Durrant wrote:
> > > On 09/12/2022 09:56, David Woodhouse wrote:
> > > > 
> > > > @@ -1811,6 +1811,9 @@ int kvm_arch_init_vcpu(CPUState *cs)
> > > >            has_msr_hv_hypercall = true;
> > > >        }
> > > >    
> > > > +    env->xen_vcpu_info_gpa = UINT64_MAX;
> > > > +    env->xen_vcpu_info_default_gpa = UINT64_MAX;
> > > 
> > > 
> > > There was an INVALID_GPA definition for shared info. Looks like we
> > > could use it here too.
> > 
> > There was, and I started trying to use it, but it fell foul of the "is
> > this going to live in target/ or hw/ and who can include what from
> > where?" and I decided to just use UINT64_MAX for now and keep typing.
> > 
> > That will work out in the end, I'm sure.
> 
> Hopefully 
> https://lore.kernel.org/lkml/20221209023622.274715-1-yu.c.zhang@linux.intel.com/
>  
> will help.

Those are kernel-internal; not in uapi headers. Although maybe they
*should* be uapi, at least for the KVM/Xen support because they are
actually part of the userspace ABI.

The kernel returns GFN_INVALID when queried about the shared info page
if it isn't set, or GPA_INVALID when queried about vcpu_info etc.

(Those are the same numerically but semantically subtly different, and
it hurts my brain that GFN_INVALID != GPA_INVALID >> PAGE_SHIFT.)

Userspace can also *set* those fields to Gxx_INVALID. Unlike the Xen
APIs which don't allow them to be turned off, we implement
SHUTDOWN_soft_reset in the userspace VMM so it needs to be able to turn
the shinfo areas off.


[-- Attachment #2: smime.p7s --]
[-- Type: application/pkcs7-signature, Size: 5965 bytes --]