memcg: reclaim when more than one page needed
authorSuleiman Souhlal <ssouhlal@FreeBSD.org>
Tue, 18 Dec 2012 22:21:41 +0000 (14:21 -0800)
committerLinus Torvalds <torvalds@linux-foundation.org>
Tue, 18 Dec 2012 23:02:12 +0000 (15:02 -0800)
mem_cgroup_do_charge() was written before kmem accounting, and expects
three cases: being called for 1 page, being called for a stock of 32
pages, or being called for a hugepage.  If we call for 2 or 3 pages (and
both the stack and several slabs used in process creation are such, at
least with the debug options I had), it assumed it's being called for
stock and just retried without reclaiming.

Fix that by passing down a minsize argument in addition to the csize.

And what to do about that (csize == PAGE_SIZE && ret) retry?  If it's
needed at all (and presumably is since it's there, perhaps to handle
races), then it should be extended to more than PAGE_SIZE, yet how far?
And should there be a retry count limit, of what?  For now retry up to
COSTLY_ORDER (as page_alloc.c does) and make sure not to do it if
__GFP_NORETRY.

v4: fixed nr pages calculation pointed out by Christoph Lameter.

Signed-off-by: Suleiman Souhlal <suleiman@google.com>
Signed-off-by: Glauber Costa <glommer@parallels.com>
Acked-by: Kamezawa Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Acked-by: Michal Hocko <mhocko@suse.cz>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: David Rientjes <rientjes@google.com>
Cc: Tejun Heo <tj@kernel.org>
Cc: Christoph Lameter <cl@linux-foundation.org>
Cc: Frederic Weisbecker <fweisbec@redhat.com>
Cc: Greg Thelen <gthelen@google.com>
Cc: JoonSoo Kim <js1304@gmail.com>
Cc: Mel Gorman <mel@csn.ul.ie>
Cc: Pekka Enberg <penberg@cs.helsinki.fi>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
mm/memcontrol.c

index 92887b286246f021c786a8ddcbc92e5ebe3bb2ef..20e3619870ae588c960fa7e4785d11956359c594 100644 (file)
@@ -2258,7 +2258,8 @@ enum {
 };
 
 static int mem_cgroup_do_charge(struct mem_cgroup *memcg, gfp_t gfp_mask,
-                               unsigned int nr_pages, bool oom_check)
+                               unsigned int nr_pages, unsigned int min_pages,
+                               bool oom_check)
 {
        unsigned long csize = nr_pages * PAGE_SIZE;
        struct mem_cgroup *mem_over_limit;
@@ -2281,18 +2282,18 @@ static int mem_cgroup_do_charge(struct mem_cgroup *memcg, gfp_t gfp_mask,
        } else
                mem_over_limit = mem_cgroup_from_res_counter(fail_res, res);
        /*
-        * nr_pages can be either a huge page (HPAGE_PMD_NR), a batch
-        * of regular pages (CHARGE_BATCH), or a single regular page (1).
-        *
         * Never reclaim on behalf of optional batching, retry with a
         * single page instead.
         */
-       if (nr_pages == CHARGE_BATCH)
+       if (nr_pages > min_pages)
                return CHARGE_RETRY;
 
        if (!(gfp_mask & __GFP_WAIT))
                return CHARGE_WOULDBLOCK;
 
+       if (gfp_mask & __GFP_NORETRY)
+               return CHARGE_NOMEM;
+
        ret = mem_cgroup_reclaim(mem_over_limit, gfp_mask, flags);
        if (mem_cgroup_margin(mem_over_limit) >= nr_pages)
                return CHARGE_RETRY;
@@ -2305,7 +2306,7 @@ static int mem_cgroup_do_charge(struct mem_cgroup *memcg, gfp_t gfp_mask,
         * unlikely to succeed so close to the limit, and we fall back
         * to regular pages anyway in case of failure.
         */
-       if (nr_pages == 1 && ret)
+       if (nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER) && ret)
                return CHARGE_RETRY;
 
        /*
@@ -2439,7 +2440,8 @@ again:
                        nr_oom_retries = MEM_CGROUP_RECLAIM_RETRIES;
                }
 
-               ret = mem_cgroup_do_charge(memcg, gfp_mask, batch, oom_check);
+               ret = mem_cgroup_do_charge(memcg, gfp_mask, batch, nr_pages,
+                   oom_check);
                switch (ret) {
                case CHARGE_OK:
                        break;