Re: [PATCH 07/25] Check in advance if the zonelist needsadditional filtering

[Mel Gorman]

On Tue, Apr 21, 2009 at 03:52:48PM +0900, KOSAKI Motohiro wrote:
> > Zonelist are filtered based on nodemasks for memory policies normally.
> > It can be additionally filters on cpusets if they exist as well as
> > noting when zones are full. These simple checks are expensive enough to
> > be noticed in profiles. This patch checks in advance if zonelist
> > filtering will ever be needed. If not, then the bulk of the checks are
> > skipped.
> > 
> > Signed-off-by: Mel Gorman <mel@xxxxxxxxx>
> > ---
> >  include/linux/cpuset.h |    2 ++
> >  mm/page_alloc.c        |   37 ++++++++++++++++++++++++++-----------
> >  2 files changed, 28 insertions(+), 11 deletions(-)
> > 
> > diff --git a/include/linux/cpuset.h b/include/linux/cpuset.h
> > index a5740fc..978e2f1 100644
> > --- a/include/linux/cpuset.h
> > +++ b/include/linux/cpuset.h
> > @@ -97,6 +97,8 @@ static inline void set_mems_allowed(nodemask_t nodemask)
> >  
> >  #else /* !CONFIG_CPUSETS */
> >  
> > +#define number_of_cpusets (0)
> > +
> >  static inline int cpuset_init(void) { return 0; }
> >  static inline void cpuset_init_smp(void) {}
> >  
> > diff --git a/mm/page_alloc.c b/mm/page_alloc.c
> > index c8465d0..3613ba4 100644
> > --- a/mm/page_alloc.c
> > +++ b/mm/page_alloc.c
> > @@ -1137,7 +1137,11 @@ failed:
> >  #define ALLOC_WMARK_HIGH	0x08 /* use pages_high watermark */
> >  #define ALLOC_HARDER		0x10 /* try to alloc harder */
> >  #define ALLOC_HIGH		0x20 /* __GFP_HIGH set */
> > +#ifdef CONFIG_CPUSETS
> >  #define ALLOC_CPUSET		0x40 /* check for correct cpuset */
> > +#else
> > +#define ALLOC_CPUSET		0x00
> > +#endif /* CONFIG_CPUSETS */
> >  
> >  #ifdef CONFIG_FAIL_PAGE_ALLOC
> >  
> > @@ -1401,6 +1405,7 @@ get_page_from_freelist(gfp_t gfp_mask, nodemask_t *nodemask, unsigned int order,
> >  	nodemask_t *allowednodes = NULL;/* zonelist_cache approximation */
> >  	int zlc_active = 0;		/* set if using zonelist_cache */
> >  	int did_zlc_setup = 0;		/* just call zlc_setup() one time */
> > +	int zonelist_filter = 0;
> >  
> >  	(void)first_zones_zonelist(zonelist, high_zoneidx, nodemask,
> >  							&preferred_zone);
> > @@ -1411,6 +1416,10 @@ get_page_from_freelist(gfp_t gfp_mask, nodemask_t *nodemask, unsigned int order,
> >  
> >  	VM_BUG_ON(order >= MAX_ORDER);
> >  
> > +	/* Determine in advance if the zonelist needs filtering */
> > +	if ((alloc_flags & ALLOC_CPUSET) && unlikely(number_of_cpusets > 1))
> > +		zonelist_filter = 1;
> > +
> >  zonelist_scan:
> >  	/*
> >  	 * Scan zonelist, looking for a zone with enough free.
> > @@ -1418,12 +1427,16 @@ zonelist_scan:
> >  	 */
> >  	for_each_zone_zonelist_nodemask(zone, z, zonelist,
> >  						high_zoneidx, nodemask) {
> > -		if (NUMA_BUILD && zlc_active &&
> > -			!zlc_zone_worth_trying(zonelist, z, allowednodes))
> > -				continue;
> > -		if ((alloc_flags & ALLOC_CPUSET) &&
> > -			!cpuset_zone_allowed_softwall(zone, gfp_mask))
> > -				goto try_next_zone;
> > +
> > +		/* Ignore the additional zonelist filter checks if possible */
> > +		if (zonelist_filter) {
> > +			if (NUMA_BUILD && zlc_active &&
> > +				!zlc_zone_worth_trying(zonelist, z, allowednodes))
> > +					continue;
> > +			if ((alloc_flags & ALLOC_CPUSET) &&
> > +				!cpuset_zone_allowed_softwall(zone, gfp_mask))
> > +					goto try_next_zone;
> > +		}
> 
> if number_of_cpusets==1, old code call zlc_zone_worth_trying(). but your one never call.
> it seems regression.
> 

True, but once fixed, the patch becomes a lot less useful. The intention was
to avoid the zlc_setup() function which is pretty heavy and hits on HIGHMEM
machines quite easily but I did it wrong. I should have made the
decision to only call zlc_setup() when there were online NUMA nodes to
care about.

I'll drop this patch and try again.

> 
> >  
> >  		if (!(alloc_flags & ALLOC_NO_WATERMARKS)) {
> >  			unsigned long mark;
> > @@ -1445,13 +1458,15 @@ zonelist_scan:
> >  		if (page)
> >  			break;
> >  this_zone_full:
> > -		if (NUMA_BUILD)
> > +		if (NUMA_BUILD && zonelist_filter)
> >  			zlc_mark_zone_full(zonelist, z);
> >  try_next_zone:
> > -		if (NUMA_BUILD && !did_zlc_setup) {
> > -			/* we do zlc_setup after the first zone is tried */
> > -			allowednodes = zlc_setup(zonelist, alloc_flags);
> > -			zlc_active = 1;
> > +		if (NUMA_BUILD && zonelist_filter) {
> > +			if (!did_zlc_setup) {
> > +				/* do zlc_setup after the first zone is tried */
> > +				allowednodes = zlc_setup(zonelist, alloc_flags);
> > +				zlc_active = 1;
> > +			}
> >  			did_zlc_setup = 1;
> >  		}
> >  	}
> > -- 
> > 1.5.6.5
> > 
> > --
> > To unsubscribe, send a message with 'unsubscribe linux-mm' in
> > the body to majordomo@xxxxxxxxxx  For more info on Linux MM,
> > see: http://www.linux-mm.org/ .
> > Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>
> 
> 
> 

-- 
Mel Gorman
Part-time Phd Student                          Linux Technology Center
University of Limerick                         IBM Dublin Software Lab
--
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/