Re: [PATCH 00/27] Move LRU page reclaim from zones to nodes v7

[Michal Hocko]

On Thu 23-06-16 13:33:47, Mel Gorman wrote:
> On Thu, Jun 23, 2016 at 01:27:14PM +0200, Michal Hocko wrote:
> > On Thu 23-06-16 11:26:48, Mel Gorman wrote:
> > > On Tue, Jun 21, 2016 at 03:15:39PM +0100, Mel Gorman wrote:
> > > > The bulk of the updates are in response to review from Vlastimil Babka
> > > > and received a lot more testing than v6.
> > > > 
> > > 
> > > Hi Andrew,
> > > 
> > > Please drop these patches again from mmotm.
> > > 
> > > There has been a number of odd conflicts resulting in at least one major
> > > bug where a node-counter is used on a zone that will result in random
> > > behaviour. Some of the additional feedback is non-trivial and all of it
> > > will need to be resolved against the OOM detection rework and the huge
> > > tmpfs implementation.
> > 
> > FWIW I haven't spotted any obvious misbehaving wrt. the OOM detection
> > rework. You have kept the per-zone counters which are used for the retry
> > logic so I think we should be safe. I am still reading through the
> > series though.
> > 
> 
> The main snag is NR_FILE_DIRTY and NR_WRITEBACK in should_reclaim_retry.
> It currently is a random number generator if it reads a zone stat
> instead of the node one. In some configurations, it even reads values
> after the stats array.

OK, I haven't spotted that. As I've said I haven't seen the whole series
yet. I have just seen that the counters are there and assumed they are
used properly where appropriate.

-- 
Michal Hocko
SUSE Labs