Re: [PATCH] x86/altp2m: p2m_altp2m_get_or_propagate() should honor ap2m->default_access

[George Dunlap <george.dunlap@xxxxxxxxx>]

On Tue, Feb 6, 2024 at 6:08 PM Petr Beneš <w1benny@xxxxxxxxx> wrote:
>
> From: Petr Beneš <w1benny@xxxxxxxxx>
>
> This patch addresses a behavior discrepancy in the handling of altp2m views,
> where upon the creation and subsequent EPT violation, the page access
> permissions were incorrectly inherited from the hostp2m instead of respecting
> the altp2m default_access.
>
> Previously, when a new altp2m view was established with restrictive
> default_access permissions and activated via xc_altp2m_switch_to_view(),
> it failed to trigger an event on the first access violation.  This behavior
> diverged from the intended mechanism, where the altp2m's default_access
> should dictate the initial permissions, ensuring proper event triggering on
> access violations.
>
> The correction involves modifying the handling mechanism to respect the
> altp2m view's default_access upon its activation, eliminating the need for
> setting memory access permissions for the entire altp2m range (e.g. within
> xen-access.c).  This change not only aligns the behavior with the expected
> access control logic but also results in a significant performance improvement
> by reducing the overhead associated with setting memory access permissions
> across the altp2m range.
>
> Signed-off-by: Petr Beneš <w1benny@xxxxxxxxx>

Thanks Petr, this looks like a great change.

Two things:

- Probably worth adjusting the comment at the top of
p2m_altp2m_get_or_propagate to mention that you use the altp2m
default_access when propagating from the host p2m

- This represents a change in behavior, so probably at least worth a
mention in CHANGELOG.md?

Tamas, I guess this is OK from an interface compatibility point of
view?  In theory it should always have been behaving this way.

 -George