Sorry for breaking thread.
I resend.
On Mon, May 30, 2011 at 02:38:31PM +0200, Michal Hocko wrote:
> 56de7263 (mm: compaction: direct compact when a high-order allocation
> fails) introduced a check for cc->order == -1 in compact_finished. We
> should continue compacting in that case because the request came from
> userspace and there is no particular order to compact for.
>
> The check is, however, done after zone_watermark_ok which uses order as
> a right hand argument for shifts. Not only watermark check is pointless
> if we can break out without it but it also uses 1 << -1 which is not
> well defined (at least from C standard). Let's move the -1 check above
> zone_watermark_ok.
>
> Signed-off-by: Michal Hocko <mhocko@xxxxxxx>
> Cc: Mel Gorman <mgorman@xxxxxxx>
> ---
> compaction.c | 14 +++++++-------
> 1 file changed, 7 insertions(+), 7 deletions(-)
> Index: linus_tree/mm/compaction.c
> ===================================================================
> --- linus_tree.orig/mm/compaction.c 2011-05-30 14:19:58.000000000 +0200
> +++ linus_tree/mm/compaction.c 2011-05-30 14:20:40.000000000 +0200
> @@ -420,13 +420,6 @@ static int compact_finished(struct zone
> if (cc->free_pfn <= cc->migrate_pfn)
> return COMPACT_COMPLETE;
>
> - /* Compaction run is not finished if the watermark is not met */
> - watermark = low_wmark_pages(zone);
> - watermark += (1 << cc->order);
> -
> - if (!zone_watermark_ok(zone, cc->order, watermark, 0, 0))
> - return COMPACT_CONTINUE;
> -
> /*
> * order == -1 is expected when compacting via
> * /proc/sys/vm/compact_memory
> @@ -434,6 +427,13 @@ static int compact_finished(struct zone
> if (cc->order == -1)
> return COMPACT_CONTINUE;
>
> + /* Compaction run is not finished if the watermark is not met */
> + watermark = low_wmark_pages(zone);
> + watermark += (1 << cc->order);
> +
> + if (!zone_watermark_ok(zone, cc->order, watermark, 0, 0))
> + return COMPACT_CONTINUE;
> +
> /* Direct compactor: Is a suitable page free? */
> for (order = cc->order; order < MAX_ORDER; order++) {
> /* Job done if page is free of the right migratetype */
It looks good to me.
Let's think about another place, compaction_suitable.
It has same problem so we can move the check right before zone_watermark_ok.
As I look it more, I thought we need free pages for compaction so we would
be better to give up early if we can't get enough free pages. But I changed
my mind. It's a totally user request and we can get free pages in migration
progress(ex, other big memory hogger might free his big rss).
So my conclusion is that we should do *best effort* than early give up.
If you agree with me, how about resending patch with compaction_suitable fix?
> --
> Michal Hocko
> SUSE Labs
> SUSE LINUX s.r.o.
> Lihovarska 1060/12
> 190 00 Praha 9
> Czech Republic
>
> --
> 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/ .
> Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/
> Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>
--
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/ .
Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/
Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>