Re: [PATCH 4/4] mm/page_alloc : relieve zone->lock's pressure for memory free

[KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>]

> On Tue, 12 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.
> > > > 
> > > 
> > > Could you please elaborate on "a bit" in the changelog with some data?  If 
> > > it's so egregious, it should be easily be quantifiable.
> > 
> > Unfortunately I can't. atomic ops is mainly the issue of large machine. but
> > I can't access such machine now. but I'm sure we shouldn't take unnecessary
> > atomic ops.
> > 
> 
> e815af95 was intended to consolidate all bit flags into a single word 
> merely for space efficiency and cleanliness.  At that time, we only had 
> one member of struct zone that could be converted, and that was 
> all_unreclaimable.  That said, it was part of a larger patchset that 
> later added another zone flag meant to serialize the oom killer by 
> zonelist.  So no consideration was given at the time concerning any 
> penalty incurred by moving all_unreclaimable to an atomic op.

I agree ZONE_OOM_LOCKED have lots worth. 


> > That's fundamental space vs performance tradeoff thing. if we talked about
> > struct page or similar lots created struct, space efficient is very important.
> > but struct zone isn't such one.
> > 
> > Or, do you have strong argue to use bitops without space efficiency?
> 
> I'd suggest using a non-atomic variation within zone->flags that may still 
> be reordered so that it does not incur any performance penalty.  In other 
> words, instead of readding zone->all_unreclaimable, we should add 
> __zone_set_flag(), __zone_test_and_set_flag(), and __zone_clear_flag() 
> variants to wrap non-atomic bitops.

No. non-atomic ops assume the flags are protected by same lock. but
all_unreclaimable and ZONE_OOM_LOCKED don't have such lock. iow,
your opinion might cause ZONE_OOM_LOCKED lost.


--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>