Re: [PATCH] dax: Fix missed PMD wakeups

[Matthew Wilcox]

On Thu, Jul 04, 2019 at 04:27:14PM -0700, Dan Williams wrote:
> On Thu, Jul 4, 2019 at 12:14 PM Matthew Wilcox <willy@xxxxxxxxxxxxx> wrote:
> >
> > On Thu, Jul 04, 2019 at 06:54:50PM +0200, Jan Kara wrote:
> > > On Wed 03-07-19 20:27:28, Matthew Wilcox wrote:
> > > > 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.
> >
> > This is an internal interface.  I think it's already a pretty gnarly
> > interface to use by definition -- it's going to sleep and might return
> > almost anything.  There's not much scope for returning an error indicator
> > either; value entries occupy half of the range (all odd numbers between 1
> > and ULONG_MAX inclusive), plus NULL.  We could use an internal entry, but
> > I don't think that makes the interface any easier to use than returning
> > a locked entry.
> >
> > I think this iteration of the patch makes it a little clearer.  What do you
> > think?
> >
> 
> Not much clearer to me. get_unlocked_entry() is now misnamed and this

misnamed?  You'd rather it was called "try_get_unlocked_entry()"?

> arrangement allows for mismatches of @order argument vs @xas
> configuration.

> Can you describe, or even better demonstrate with
> numbers, why it's better to carry this complication than just
> converging the waitqueues between the types?

You've got the reproducer ;-)  It seems quite wrong to make a page fault
stall just because another task is working on a different page in the
same 2MB chunk.