This is just an RFC to reduce some of the more obvious stack usage in page
reclaim. It's a bit rushed and I haven't tested this yet but am sending
it out as there may be others working on similar material and would rather
avoid overlap. I built on some of Kosaki Motohiro's work.
On X86 bit, stack usage figures (generated using a modified bloat-o-meter
that uses checkstack.pl as its input) change in the following ways after
the series of patches.
add/remove: 2/0 grow/shrink: 0/4 up/down: 804/-1688 (-884)
function old new delta
putback_lru_pages - 676 +676
update_isolated_counts - 128 +128
do_try_to_free_pages 172 128 -44
kswapd 1324 1168 -156
shrink_page_list 1616 1224 -392
shrink_zone 2320 1224 -1096
There are some growths there but critically they are no longer in the path
that would call writepages. In the main path, there is about 1K of stack
lopped off giving a small amount of breathing room.
KOSAKI Motohiro (3):
vmscan: kill prev_priority completely
vmscan: move priority variable into scan_control
vmscan: simplify shrink_inactive_list()
Mel Gorman (7):
vmscan: Remove useless loop at end of do_try_to_free_pages
vmscan: Remove unnecessary temporary vars in do_try_to_free_pages
vmscan: Split shrink_zone to reduce stack usage
vmscan: Remove unnecessary temporary variables in shrink_zone()
vmscan: Setup pagevec as late as possible in shrink_inactive_list()
vmscan: Setup pagevec as late as possible in shrink_page_list()
vmscan: Update isolated page counters outside of main path in
shrink_inactive_list()
include/linux/mmzone.h | 15 --
mm/page_alloc.c | 2 -
mm/vmscan.c | 447 +++++++++++++++++++++++-------------------------
mm/vmstat.c | 2 -
4 files changed, 210 insertions(+), 256 deletions(-)
--
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>