Re: [PATCH v2 3/7] x86: extend update_intpte() to support atomic get-and-update

[Jan Beulich <jbeulich@xxxxxxxx>]

On 16.09.2026 10:23, Kevin Lampis wrote:
>> The "flag set, pointer non-NULL" case wouldn't hit the "return 0" path
> 
> Correct.
> 
> But what about the "flag not set" case. When PTE_UPDATE_SWAP is not set then
> UPDATE_ENTRY may return 0.
> 
> If UPDATE_ENTRY() returns 0 then mod_l1_entry() will give
> the wrong ol1e value to put_page_from_l1e().
> 
>     static int mod_l1_entry( ... )
>     {
>         ...
>         ol1e = UPDATE_ENTRY(l1, pl1e, ol1e, ...);
>         ...
>         put_page_from_l1e(ol1e, pt_dom);
>         ...
>     }
> 
> I think mod_l1_entry may need to store a copy of ol1e incase UPDATE_ENTRY()
> returns 0. Or add some extra if/else around every call to UPDATE_ENTRY() to
> only assign ol1e if we know we're on the new PTE_UPDATE_SWAP path.

Possibly, yes. Or even have a separate helper for the SWAP case. What may
be possible with the returning of 0 is

    ol1e = UPDATE_ENTRY(l1, pl1e, ol1e, ...) ?: ol1e;

when the return value is relevant in both cases. Question is: Is it, in the
first place when SWAP isn't set?

Jan