On Thu, Sep 10, 2020 at 03:28:03PM +0200, Gerald Schaefer wrote:
> On Thu, 10 Sep 2020 10:02:33 -0300
> Jason Gunthorpe <jgg@xxxxxxxx> wrote:
>
> > On Thu, Sep 10, 2020 at 11:39:25AM +0200, Alexander Gordeev wrote:
> >
> > > As Gerald mentioned, it is very difficult to explain in a clear way.
> > > Hopefully, one could make sense ot of it.
> >
> > I would say the page table API requires this invariant:
> >
> > pud = pud_offset(p4d, addr);
> > do {
> > WARN_ON(pud != pud_offset(p4d, addr);
> > next = pud_addr_end(addr, end);
> > } while (pud++, addr = next, addr != end);
> >
> > ie pud++ is supposed to be a shortcut for
> > pud_offset(p4d, next)
> >
> > While S390 does not follow this. Fixing addr_end brings it into
> > alignment by preventing pud++ from happening.
> >
> > The only currently known side effect is that gup_fast crashes, but it
> > sure is an unexpected thing.
>
> It only is unexpected in a "top-level folding" world, see my other reply.
> Consider it an optimization, which was possible because of how our dynamic
> folding works, and e.g. because we can determine the correct pagetable
> level from a pXd value in pXd_offset.
No, I disagree. The page walker API the arch presents has to have well
defined semantics. For instance, there is an effort to define tests
and invarients for the page table accesses to bring this understanding
and uniformity:
mm/debug_vm_pgtable.c
If we fix S390 using the pX_addr_end() change then the above should be
updated with an invariant to check it. I've added Anshuman for some
thoughts..
For better or worse, that invariant does exclude arches from using
other folding techniques.
The other solution would be to address the other side of != and adjust
the pud++
eg replcae pud++ with something like:
pud = pud_next_entry(p4d, pud, next)
Such that:
pud_next_entry(p4d, pud, next) === pud_offset(p4d, next)
In which case the invarient changes to 'callers can never do pointer
arithmetic on the result of pXX_offset()' which is a bit harder to
enforce.
Jason
