Re: [Xen-devel] [PATCH 09/14] xen: Introduce HAS_M2P config and use to protect mfn_to_gmfn call

[Julien Grall <julien.grall@xxxxxxx>]

Hi,

Answering to myself.

On 10/05/2019 15:34, Julien Grall wrote:
> On 10/05/2019 15:19, Jan Beulich wrote:
> > > > > On 10.05.19 at 16:04, <julien.grall@xxxxxxx> wrote:
> > > On 10/05/2019 14:45, Jan Beulich wrote:
> > > > > > > On 10.05.19 at 15:41, <julien.grall@xxxxxxx> wrote:
> > > > > The point here, we keep within the hypervisor the idea of what's valid or
> > > > > invalid. This allows us more flexibility on the value here (imagine we 
> > > > > decide to
> > > > > change the value of GFN_INVALID in the future...).
> > > > Exactly, and hence INVALID_GFN should not become visible to the
> > > > outside. Hence my request to use an all-ones value here.
> > > It is only visible if you put an exact value in the documentation. Your
> > > suggestion is to put a exactly value and I would rather not see that.
> > I did specifically suggest to _not_ store INVALID_GFN here, but to
> > store 64-bit bits of ones. Note the difference between the two on
> > 32-bit Arm.
> Your point of having an exact value is only useful if you want to toolstack to 
> silently ignore the missing frame and avoid a call.
> The former is pretty much wrong as if you were trying to read the frame then 
> most likely you wanted to access it. So a message makes sense here.
> For the latter, avoiding the call is only going to save you a couple of cycles 
> in a likely cold path.
> You really don't need to give an exact (including say all ones). You only need 
> to say that the address return may not be mappable. The toolstack will try to 
> map it and fail. That's not a big deal.
> Anyway, I will wait and see what's the view from the tools maintainer.
I had a discussion with Ian on IRC regarding the value here. After some debate 
we agreed that specifying a single value would be best.
At the moment, Xen only supports 48-bits address that could be covered by 36-bit 
MFN. Newer Arm revision supports up to 52-bit address, but that's only with 64KB 
page granularity. Even if we were supporting 64-bit address, it would only cover 
48-bit, so all ones should still be invalid. So I am happy with the all ones 
version.
I will introduce a define so the tools can use it rather than an hardcoded 
value.

Cheers,

--
Julien Grall

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