Re: [resend][PATCH] mm: Restore zone->all_unreclaimable to independence word

[KOSAKI Motohiro]

> On Thu, 14 Jan 2010 15:19:59 -0800
> Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx> wrote:
> 
> > On Thu, 14 Jan 2010 16:32:29 +0800
> > Wu Fengguang <fengguang.wu@xxxxxxxxx> wrote:
> > 
> > > On Thu, Jan 14, 2010 at 03:14:10PM +0800, KOSAKI Motohiro wrote:
> > > > > On Thu, 14 Jan 2010, KOSAKI Motohiro wrote:
> > > > > 
> > > > > > commit e815af95 (change all_unreclaimable zone member to flags) chage
> > > > > > all_unreclaimable member to bit flag. but It have undesireble side
> > > > > > effect.
> > > > > > free_one_page() is one of most hot path in linux kernel and increasing
> > > > > > atomic ops in it can reduce kernel performance a bit.
> > > > > > 
> > > > > > Thus, this patch revert such commit partially. at least
> > > > > > all_unreclaimable shouldn't share memory word with other zone flags.
> > > > > > 
> > > > > 
> > > > > I still think you need to quantify this; saying you don't have a large 
> > > > > enough of a machine that will benefit from it isn't really a rationale for 
> > > > > the lack of any data supporting your claim.  We should be basing VM 
> > > > > changes on data, not on speculation that there's a measurable impact 
> > > > > here.
> > > > > 
> > > > > Perhaps you could ask a colleague or another hacker to run a benchmark for 
> > > > > you so that the changelog is complete?
> > > > 
> > > > ok, fair. although I dislike current unnecessary atomic-ops.
> > > > I'll pending this patch until get good data.
> > > 
> > > I think it's a reasonable expectation to help large boxes.
> > > 
> > > What we can do now, is to measure if it hurts mainline SMP
> > > boxes. If not, we are set on doing the patch :)
> > 
> > yup, the effects of the change might be hard to measure.  Not that one
> > shouldn't try!
> > 
> > But sometimes we just have to do a best-effort change based upon theory
> > and past experience.
> > 
> > Speaking of which...
> > 
> > : --- a/include/linux/mmzone.h
> > : +++ b/include/linux/mmzone.h
> > : @@ -341,6 +341,7 @@ struct zone {
> > :  
> > :  	unsigned long		pages_scanned;	   /* since last reclaim */
> > :  	unsigned long		flags;		   /* zone flags, see below */
> > : +	int                     all_unreclaimable; /* All pages pinned */
> > :  
> > :  	/* Zone statistics */
> > :  	atomic_long_t		vm_stat[NR_VM_ZONE_STAT_ITEMS];
> > 
> > Was that the best place to put the field?  It adds four bytes of
> > padding to the zone, hence is suboptimal from a cache utilisation point
> > of view.
> > 
> > It might also be that we can place this field closed in memory to other
> > fields which are being manipulated at the same time as
> > all_unreclaimable, hm?
> > 
> How about the same line where zone->lock is ?

Sure. page allocator obviously touch zone->lock at first.
Incremental patch is here.


---
 include/linux/mmzone.h |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)

diff --git a/include/linux/mmzone.h b/include/linux/mmzone.h
index 4f0c6f1..0df3749 100644
--- a/include/linux/mmzone.h
+++ b/include/linux/mmzone.h
@@ -314,6 +314,7 @@ struct zone {
 	 * free areas of different sizes
 	 */
 	spinlock_t		lock;
+	int                     all_unreclaimable; /* All pages pinned */
 #ifdef CONFIG_MEMORY_HOTPLUG
 	/* see spanned/present_pages for more description */
 	seqlock_t		span_seqlock;
@@ -341,7 +342,6 @@ struct zone {
 
 	unsigned long		pages_scanned;	   /* since last reclaim */
 	unsigned long		flags;		   /* zone flags, see below */
-	int                     all_unreclaimable; /* All pages pinned */
 
 	/* Zone statistics */
 	atomic_long_t		vm_stat[NR_VM_ZONE_STAT_ITEMS];
-- 
1.6.5.2





--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/