On Fri, 23 Mar 2012, Stephen Rothwell wrote:

Changes since 20120322:

The akpm tree lost lots of patches that turned up elsewhere (mostly in
Linus' tree).

I'm amused to notice that your merging process is as impatient as akpm,
and just cannot wait for Mel to remove lumpy reclaim from mm/vmscan.c.

It has chosen to do so itself, by duplicating Konstantin's "mm: forbid
lumpy-reclaim in shrink_active_list()" (we had expected that to be held
back, but it's gone in anyway, oh well) into shrink_inactive_list(),
overriding the set_reclaim_mode() a few lines above with a spurious
reset_reclaim_mode(sc).

I expect that will sort itself out automatically once you get an update
from akpm, but something to beware of meanwhile (probably just a matter
of deleting Konstantin's patch from your trove now it's in Linus's tree).

Corrective patch below for illustration, but would need to be applied
with care: I wouldn't be surprised if it chose to do exactly the wrong
thing when applied automatically.

Hugh

--- linux-next/mm/vmscan.c.orig	2012-03-23 01:29:33.312011679 -0700
+++ linux-next/mm/vmscan.c	2012-03-23 03:00:08.688142469 -0700

@@ -1528,8 +1528,6 @@ shrink_inactive_list(unsigned long nr_to
 
 	lru_add_drain();
 
-	reset_reclaim_mode(sc);
-
 	if (!sc->may_unmap)
 		isolate_mode |= ISOLATE_UNMAPPED;
 	if (!sc->may_writepage)