Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > linux.kernel > #1442003

Re: [PATCH 02/34] mm, vmscan: move lru_lock to the node

Show all headers | View raw

On Tue, Jul 12, 2016 at 12:18:05PM +0100, Mel Gorman wrote:
> On Tue, Jul 12, 2016 at 09:06:04PM +1000, Balbir Singh wrote:
> > > diff --git a/Documentation/cgroup-v1/memory.txt b/Documentation/cgroup-v1/memory.txt
> > > index b14abf217239..946e69103cdd 100644
> > > --- a/Documentation/cgroup-v1/memory.txt
> > > +++ b/Documentation/cgroup-v1/memory.txt
> > > @@ -267,11 +267,11 @@ When oom event notifier is registered, event will be delivered.
> > >     Other lock order is following:
> > >     PG_locked.
> > >     mm->page_table_lock
> > > -       zone->lru_lock
> > > +       zone_lru_lock
> > 
> > zone_lru_lock is a little confusing, can't we just call it
> > node_lru_lock?
> > 
> 
> It's a matter of perspective. People familiar with the VM already expect
> a zone lock so will be looking for it. I can do a rename if you insist
> but it may not actually help.

I don't want to insist, but zone_ in the name can be confusing, as to
leading us to think that the lru_lock is still in the zone

If the rest of the reviewers are fine with, we don't need to rename

> 
> > > @@ -496,7 +496,6 @@ struct zone {
> > >  	/* Write-intensive fields used by page reclaim */
> > >  
> > >  	/* Fields commonly accessed by the page reclaim scanner */
> > > -	spinlock_t		lru_lock;
> > >  	struct lruvec		lruvec;
> > >  
> > >  	/*
> > > @@ -690,6 +689,9 @@ typedef struct pglist_data {
> > >  	/* Number of pages migrated during the rate limiting time interval */
> > >  	unsigned long numabalancing_migrate_nr_pages;
> > >  #endif
> > > +	/* Write-intensive fields used by page reclaim */
> > > +	ZONE_PADDING(_pad1_)a
> > 
> > I thought this was to have zone->lock and zone->lru_lock in different
> > cachelines, do we still need the padding here?
> > 
> 
> The zone padding current keeps the page lock wait tables, page allocator
> lists, compaction and vmstats on separate cache lines. They're still
> fine.
> 
> The node padding may not be necessary. It currently ensures that zonelists
> and numa balancing are separate from the LRU lock but there is no guarantee
> the current arrangement is optimal. It would depend on both the kernel
> config and the workload but it may be necessary in the future to split
> node into read-mostly sections and then different write-intensive sections
> similar to what has happened to struct zone in the past.
>

Fair enough

Balbir Singh. 

Back to linux.kernel | Previous | Next — Previous in thread | Next in thread | Find similar | Unroll thread

Thread

Re: [PATCH 02/34] mm, vmscan: move lru_lock to the node Balbir Singh <bsingharora@gmail.com> - 2016-07-12 13:10 +0200
  Re: [PATCH 02/34] mm, vmscan: move lru_lock to the node Mel Gorman <mgorman@techsingularity.net> - 2016-07-12 13:20 +0200
    Re: [PATCH 02/34] mm, vmscan: move lru_lock to the node Balbir Singh <bsingharora@gmail.com> - 2016-07-13 08:00 +0200
      Re: [PATCH 02/34] mm, vmscan: move lru_lock to the node Vlastimil Babka <vbabka@suse.cz> - 2016-07-13 10:50 +0200

csiph-web