[PATCH] maple_tree: Fix mas_next() when already on the last node entry

[Liam Howlett]

* Andrew Morton <akpm at linux-foundation.org> [220517 13:35]:
> On Tue, 17 May 2022 17:34:58 +0200 Sven Schnelle <svens at linux.ibm.com> wrote:
> 
> > Liam Howlett <liam.howlett at oracle.com> writes:
> > 
> > > It is possible to return the metadata as the next entry if the last node
> > > entry is already in the maple state and the limit is not reached.  Check
> > > for this condition in mas_next_nentry() where the node end is returned.
> > >
> > > Signed-off-by: Liam R. Howlett <Liam.Howlett at oracle.com>
> > 
> > Thanks, that matches my observation from the initial report that we're
> > returing metadata. I just applied the patch to next-20220516 and i'm no
> > longer able to trigger the crash. So feel free to add my:
> > 
> > Tested-by: Sven Schnelle <svens at linux.ibm.com>
> 
> Thanks.
> 
> > However, as Heiko already wrote in another mail i would also like to
> > request that the maple tree code isn't merged with the next merge
> > window. These patches touch a lot of critical infrastructure, and i would like
> > to have it in next for at least one development cycle, so we can be sure
> > that we've seen and fixed most of the issues.
> 
> Agree, it isn't looking good.  If Linus does an -rc8 (unlikely) then
> perhaps we'll be in better shape.
> 

Yeah, it needs more soak time.  I've been testing on a lot of platforms
and have a number of testcases, but it's difficult to cover this
critical infrastructure with the variations of hardware and
randomization of addresses.