Re: [PATCH] x86/mem-sharing: short-circuit p2m_is_shared() when MEM_SHARING=n

[Tamas K Lengyel <tamas@xxxxxxxxxxxxx>]

On Thu, Apr 3, 2025 at 4:42 AM Jan Beulich <jbeulich@xxxxxxxx> wrote:
>
> Some of the uses of dom_cow aren't easily DCE-able (without extra
> #ifdef-ary), and hence it being constantly NULL when MEM_SHARING=n
> misguides Coverity into thinking that there may be a NULL deref in
>
>         if ( p2m_is_shared(t) )
>             d = dom_cow;
>
>         if ( get_page(page, d) )
>             return page;
>
> (in get_page_from_mfn_and_type()). Help the situation by making
> p2m_is_shared() be compile-time false when MEM_SHARING=n, thus also
> permitting the compiler to DCE some other code.
>
> Note that p2m_is_sharable() isn't used outside of mem_sharing.c, and
> hence P2M_SHARABLE_TYPES can simply be left undefined when
> MEM_SHARING=n.
>
> Coverity ID: 1645573
> Fixes: 79d91e178a1a ("dom_cow is needed for mem-sharing only")
> Signed-off-by: Jan Beulich <jbeulich@xxxxxxxx>

Reviewed-by: Tamas K Lengyel <tamas@xxxxxxxxxxxxx>

> ---
> Might be nice to also eliminate p2m_ram_shared (and for MEM_PAGING=n
> also the three paging types) entirely from such builds, to eliminate the
> risk of accidental use. Yet that would apparently also come at the price
> of more #ifdef-ary. Opinions?

I don't think the risk of accidental use is a concern. I wouldn't
touch them unless they lead to similar confusion with coverity or some
other tool.

Cheers,
Tamas