]> git.proxmox.com Git - mirror_ubuntu-disco-kernel.git/commitdiff
memcg: modify accounting function for supporting THP better
authorKAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Thu, 20 Jan 2011 22:44:23 +0000 (14:44 -0800)
committerLinus Torvalds <torvalds@linux-foundation.org>
Fri, 21 Jan 2011 01:02:05 +0000 (17:02 -0800)
mem_cgroup_charge_statisics() was designed for charging a page but now, we
have transparent hugepage.  To fix problems (in following patch) it's
required to change the function to get the number of pages as its
arguments.

The new function gets following as argument.
  - type of page rather than 'pc'
  - size of page which is accounted.

Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: Balbir Singh <balbir@linux.vnet.ibm.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
mm/memcontrol.c

index 8ab8410314363871da25fa9f613578f5ebe01783..6d59a2bd520a0fba52a47a1ceb7558252eb07ba7 100644 (file)
@@ -600,23 +600,22 @@ static void mem_cgroup_swap_statistics(struct mem_cgroup *mem,
 }
 
 static void mem_cgroup_charge_statistics(struct mem_cgroup *mem,
-                                        struct page_cgroup *pc,
-                                        bool charge)
+                                        bool file, int nr_pages)
 {
-       int val = (charge) ? 1 : -1;
-
        preempt_disable();
 
-       if (PageCgroupCache(pc))
-               __this_cpu_add(mem->stat->count[MEM_CGROUP_STAT_CACHE], val);
+       if (file)
+               __this_cpu_add(mem->stat->count[MEM_CGROUP_STAT_CACHE], nr_pages);
        else
-               __this_cpu_add(mem->stat->count[MEM_CGROUP_STAT_RSS], val);
+               __this_cpu_add(mem->stat->count[MEM_CGROUP_STAT_RSS], nr_pages);
 
-       if (charge)
+       /* pagein of a big page is an event. So, ignore page size */
+       if (nr_pages > 0)
                __this_cpu_inc(mem->stat->count[MEM_CGROUP_STAT_PGPGIN_COUNT]);
        else
                __this_cpu_inc(mem->stat->count[MEM_CGROUP_STAT_PGPGOUT_COUNT]);
-       __this_cpu_inc(mem->stat->count[MEM_CGROUP_EVENTS]);
+
+       __this_cpu_add(mem->stat->count[MEM_CGROUP_EVENTS], nr_pages);
 
        preempt_enable();
 }
@@ -2115,7 +2114,7 @@ static void ____mem_cgroup_commit_charge(struct mem_cgroup *mem,
                break;
        }
 
-       mem_cgroup_charge_statistics(mem, pc, true);
+       mem_cgroup_charge_statistics(mem, PageCgroupCache(pc), 1);
 }
 
 static void __mem_cgroup_commit_charge(struct mem_cgroup *mem,
@@ -2186,14 +2185,14 @@ static void __mem_cgroup_move_account(struct page_cgroup *pc,
                __this_cpu_inc(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED]);
                preempt_enable();
        }
-       mem_cgroup_charge_statistics(from, pc, false);
+       mem_cgroup_charge_statistics(from, PageCgroupCache(pc), -1);
        if (uncharge)
                /* This is not "cancel", but cancel_charge does all we need. */
                mem_cgroup_cancel_charge(from, PAGE_SIZE);
 
        /* caller should have done css_get */
        pc->mem_cgroup = to;
-       mem_cgroup_charge_statistics(to, pc, true);
+       mem_cgroup_charge_statistics(to, PageCgroupCache(pc), 1);
        /*
         * We charges against "to" which may not have any tasks. Then, "to"
         * can be under rmdir(). But in current implementation, caller of
@@ -2597,7 +2596,7 @@ __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype)
        }
 
        for (i = 0; i < count; i++)
-               mem_cgroup_charge_statistics(mem, pc + i, false);
+               mem_cgroup_charge_statistics(mem, PageCgroupCache(pc), -1);
 
        ClearPageCgroupUsed(pc);
        /*