Re: [XEN PATCH v1 1/1] x86/domctl: add gva_to_gfn command

[Tamas K Lengyel <tamas@xxxxxxxxxxxxx>]

On Mon, Mar 20, 2023 at 3:34 PM Andrew Cooper <andrew.cooper3@xxxxxxxxxx> wrote:
>
> On 20/03/2023 7:22 pm, Ковалёв Сергей wrote:
> >
> > 21.03.2023 1:51, Tamas K Lengyel wrote:
> >> On Mon, Mar 20, 2023 at 12:32 PM Ковалёв Сергей <valor@xxxxxxx
> >> <mailto:valor@xxxxxxx>> wrote:
> >>  >
> >>  > gva_to_gfn command used for fast address translation in LibVMI
> >> project.
> >>  > With such a command it is possible to perform address translation in
> >>  > single call instead of series of queries to get every page table.
> >>
> >> You have a couple assumptions here:
> >>   - Xen will always have a direct map of the entire guest memory -
> >> there are already plans to move away from that. Without that this
> >> approach won't have any advantage over doing the same mapping by LibVMI
> >
> > Thanks! I didn't know about the plan.
>
> To be clear, "not mapping the guest by default" is for speculative
> safety/hardening reasons.
>
> Xen will always need to be capable of mapping arbitrary parts of the
> guest, even if only transiently, so there's no relevant interaction with
> this new proposed hypercall.
>
>
> The truth is that Xen will always be able to do a single pagewalk faster
> than libvmi can do it (via mappings, or otherwise), but if libvmi does
> properly maintain a cache of mappings then it will be faster that
> repeated hypercalls into Xen.  Where the split lies depends heavily on
> the libvmi workload.
>
> I don't see a problem in principle with a hypercall like this - it is
> "just" a performance optimisation over capabilities that libvmi already
> has - but the version presented here is overly simplistic.

For debugging purposes sure it would be fine to have this hypercall but I wouldn't set it as the default for LibVMI. Oftentimes the lookup needs to be more nuanced then what Xen understands about paging. For example, on Windows guests you can have transition pages that don't have the present bit set yet are perfectly valid for introspection purposes (https://citeseerx.ist.psu.edu/document?repid=rep1&type=pdf&doi=3311ed0c63d4ca707c49256655e401f37f25ec50). Xen would need to be enlightened about this type of OS-specific tidbits for which I think LibVMI is a much better place to keep the logic for.

Tamas