Re: [PATCH] dax: Fix missed PMD wakeups

[Jan Kara]

On Wed 03-07-19 20:27:28, Matthew Wilcox wrote:
> On Wed, Jul 03, 2019 at 02:28:41PM -0700, Dan Williams wrote:
> > On Wed, Jul 3, 2019 at 12:53 PM Matthew Wilcox <willy@xxxxxxxxxxxxx> wrote:
> > > @@ -211,7 +215,8 @@ static void *get_unlocked_entry(struct xa_state *xas)
> > >         for (;;) {
> > >                 entry = xas_find_conflict(xas);
> > >                 if (!entry || WARN_ON_ONCE(!xa_is_value(entry)) ||
> > > -                               !dax_is_locked(entry))
> > > +                               !dax_is_locked(entry) ||
> > > +                               dax_entry_order(entry) < xas_get_order(xas))
> > 
> > Doesn't this potentially allow a locked entry to be returned for a
> > caller that expects all value entries are unlocked?
> 
> It only allows locked entries to be returned for callers which pass in
> an xas which refers to a PMD entry.  This is fine for grab_mapping_entry()
> because it checks size_flag & is_pte_entry.
> 
> dax_layout_busy_page() only uses 0-order.
> __dax_invalidate_entry() only uses 0-order.
> dax_writeback_one() needs an extra fix:
> 
>                 /* Did a PMD entry get split? */
>                 if (dax_is_locked(entry))
>                         goto put_unlocked;
> 
> dax_insert_pfn_mkwrite() checks for a mismatch of pte vs pmd.
> 
> So I think we're good for all current users.

Agreed but it is an ugly trap. As I already said, I'd rather pay the
unnecessary cost of waiting for pte entry and have an easy to understand
interface. If we ever have a real world use case that would care for this
optimization, we will need to refactor functions to make this possible and
still keep the interfaces sane. For example get_unlocked_entry() could
return special "error code" indicating that there's no entry with matching
order in xarray but there's a conflict with it. That would be much less
error-prone interface.

								Honza

-- 
Jan Kara <jack@xxxxxxxx>
SUSE Labs, CR