Re: [PATCH 3/3] gnttab: GTF_sub_page is a v2-only flag

[Jan Beulich <jbeulich@suse.com>]

On 18.02.2021 15:22, Andrew Cooper wrote:
> On 17/02/2021 10:46, Jan Beulich wrote:
>> Prior to its introduction, v1 entries weren't checked for this flag, and
>> the flag also has been meaningless for v1 entries. Therefore it also
>> shouldn't be checked. (The only consistent alternative would be to also
>> check for all currently undefined flags to be clear.)
> 
> We recently had a similar discussion for the stable libs.
> 
> Whatever we do, an unexpected corner case needs to break.  Checking for
> all undefined flags up front is far cleaner - absolutely nothing good
> can come for a guest which set GTF_sub_page with v1, and is expecting it
> to work, and this way, we do all breaking in one go, rather than
> breaking $N times in the future as new flags get added.

Except that there doesn't need to be any breaking: v1 could continue
to ignore all originally undefined flags. v1 and v2 could continue to
ignore all flags presently undefined. See also Julien's question
about GTF_sub_page on a transitive grant entry. That's presently an
ignored setting as well (and, as an implementation detail, in fact
getting forced to be set, but with a range covering the entire page).

Retroactively starting to check (and reject) undefined flags isn't
going to be nice; nevertheless I wanted to put this up as an at
least theoretical alternative. Perhaps a topic for the next community
call, if I don't forget by the time an agenda page appears.

Jan