Re: Possible dcache BUG

[Marcelo Tosatti <marcelo.tosatti@cyclades.com>]

On Tue, Aug 24, 2004 at 01:00:27PM -0700, Andrew Morton wrote:
> Marcelo Tosatti <marcelo.tosatti@cyclades.com> wrote:
> >
> > I dont fully understand the all_unreclaimable logic yet.
> 
> 1) bk revtool include/linux/mmzone.h
> 2) double-click on declaration of all_unreclaimable
> 3) read changelog ;)

Will do, but my question is still unanswered, why stop IO throttling when all_unreclaimable
is set? 

Doesnt make sense to me right now.

OK, will RTFS.

> 
> > --- mm/vmscan.c.orig	2004-08-24 16:48:09.467086840 -0300
> > +++ mm/vmscan.c	2004-08-24 16:51:55.304754296 -0300
> > @@ -878,7 +878,8 @@
> >  		if (zone->prev_priority > sc->priority)
> >  			zone->prev_priority = sc->priority;
> >  
> > -		if (zone->all_unreclaimable && sc->priority != DEF_PRIORITY)
> > +		if (zone->all_unreclaimable && 
> > +				(sc->priority < DEF_PRIORITY && sc->priority > 0))
> >  			continue;	/* Let kswapd poll it */
> >  
> >  		shrink_zone(zone, sc);
> > @@ -1054,7 +1055,8 @@
> >  		for (i = 0; i <= end_zone; i++) {
> >  			struct zone *zone = pgdat->node_zones + i;
> >  
> > -			if (zone->all_unreclaimable && priority != DEF_PRIORITY)
> > +			if (zone->all_unreclaimable && 
> > +					(priority < DEF_PRIORITY && priority > 0))
> >  				continue;
> >  
> >  			if (nr_pages == 0) {	/* Not software suspend */
> 
> Does anyone understand _why_ all_unreclaimable is getting set?
> 
> If not, it's too early to be writing patches...

As I wrote down in the first email, kswapd does

                        if (zone->pages_scanned > zone->present_pages * 2)
                                zone->all_unreclaimable = 1;

Sure, it makes perfect sense to happen when we can't unreclaim pages
from the zone.

Its not something hard to understand. What is your point?

I suppose your question is not "_why_ all_unreclaimable is getting set?" but 
"maybe it should not be getting set?". 

Anyway, will RTFS.