memcg: use native word page statistics counters
authorJohannes Weiner <hannes@cmpxchg.org>
Wed, 23 Mar 2011 23:42:38 +0000 (16:42 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Thu, 24 Mar 2011 02:46:31 +0000 (19:46 -0700)
The statistic counters are in units of pages, there is no reason to make
them 64-bit wide on 32-bit machines.

Make them native words.  Since they are signed, this leaves 31 bit on
32-bit machines, which can represent roughly 8TB assuming a page size of
4k.

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

index d884f758c0e36e11b29abef24781712a6de01523..e5759b51f37e9a7708cf7df946a6eb11dbae1688 100644 (file)
@@ -73,15 +73,6 @@ static int really_do_swap_account __initdata = 0;
 #define do_swap_account                (0)
 #endif
 
-/*
- * Per memcg event counter is incremented at every pagein/pageout. This counter
- * is used for trigger some periodic events. This is straightforward and better
- * than using jiffies etc. to handle periodic memcg event.
- *
- * These values will be used as !((event) & ((1 <<(thresh)) - 1))
- */
-#define THRESHOLDS_EVENTS_THRESH (7) /* once in 128 */
-#define SOFTLIMIT_EVENTS_THRESH (10) /* once in 1024 */
 
 /*
  * Statistics for memory cgroup.
@@ -105,10 +96,24 @@ enum mem_cgroup_events_index {
        MEM_CGROUP_EVENTS_COUNT,        /* # of pages paged in/out */
        MEM_CGROUP_EVENTS_NSTATS,
 };
+/*
+ * Per memcg event counter is incremented at every pagein/pageout. With THP,
+ * it will be incremated by the number of pages. This counter is used for
+ * for trigger some periodic events. This is straightforward and better
+ * than using jiffies etc. to handle periodic memcg event.
+ */
+enum mem_cgroup_events_target {
+       MEM_CGROUP_TARGET_THRESH,
+       MEM_CGROUP_TARGET_SOFTLIMIT,
+       MEM_CGROUP_NTARGETS,
+};
+#define THRESHOLDS_EVENTS_TARGET (128)
+#define SOFTLIMIT_EVENTS_TARGET (1024)
 
 struct mem_cgroup_stat_cpu {
-       s64 count[MEM_CGROUP_STAT_NSTATS];
+       long count[MEM_CGROUP_STAT_NSTATS];
        unsigned long events[MEM_CGROUP_EVENTS_NSTATS];
+       unsigned long targets[MEM_CGROUP_NTARGETS];
 };
 
 /*
@@ -546,11 +551,11 @@ mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone *mctz)
  * common workload, threashold and synchonization as vmstat[] should be
  * implemented.
  */
-static s64 mem_cgroup_read_stat(struct mem_cgroup *mem,
-               enum mem_cgroup_stat_index idx)
+static long mem_cgroup_read_stat(struct mem_cgroup *mem,
+                                enum mem_cgroup_stat_index idx)
 {
+       long val = 0;
        int cpu;
-       s64 val = 0;
 
        get_online_cpus();
        for_each_online_cpu(cpu)
@@ -564,9 +569,9 @@ static s64 mem_cgroup_read_stat(struct mem_cgroup *mem,
        return val;
 }
 
-static s64 mem_cgroup_local_usage(struct mem_cgroup *mem)
+static long mem_cgroup_local_usage(struct mem_cgroup *mem)
 {
-       s64 ret;
+       long ret;
 
        ret = mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_RSS);
        ret += mem_cgroup_read_stat(mem, MEM_CGROUP_STAT_CACHE);
@@ -634,13 +639,34 @@ static unsigned long mem_cgroup_get_local_zonestat(struct mem_cgroup *mem,
        return total;
 }
 
-static bool __memcg_event_check(struct mem_cgroup *mem, int event_mask_shift)
+static bool __memcg_event_check(struct mem_cgroup *mem, int target)
+{
+       unsigned long val, next;
+
+       val = this_cpu_read(mem->stat->events[MEM_CGROUP_EVENTS_COUNT]);
+       next = this_cpu_read(mem->stat->targets[target]);
+       /* from time_after() in jiffies.h */
+       return ((long)next - (long)val < 0);
+}
+
+static void __mem_cgroup_target_update(struct mem_cgroup *mem, int target)
 {
-       unsigned long val;
+       unsigned long val, next;
 
        val = this_cpu_read(mem->stat->events[MEM_CGROUP_EVENTS_COUNT]);
 
-       return !(val & ((1 << event_mask_shift) - 1));
+       switch (target) {
+       case MEM_CGROUP_TARGET_THRESH:
+               next = val + THRESHOLDS_EVENTS_TARGET;
+               break;
+       case MEM_CGROUP_TARGET_SOFTLIMIT:
+               next = val + SOFTLIMIT_EVENTS_TARGET;
+               break;
+       default:
+               return;
+       }
+
+       this_cpu_write(mem->stat->targets[target], next);
 }
 
 /*
@@ -650,10 +676,15 @@ static bool __memcg_event_check(struct mem_cgroup *mem, int event_mask_shift)
 static void memcg_check_events(struct mem_cgroup *mem, struct page *page)
 {
        /* threshold event is triggered in finer grain than soft limit */
-       if (unlikely(__memcg_event_check(mem, THRESHOLDS_EVENTS_THRESH))) {
+       if (unlikely(__memcg_event_check(mem, MEM_CGROUP_TARGET_THRESH))) {
                mem_cgroup_threshold(mem);
-               if (unlikely(__memcg_event_check(mem, SOFTLIMIT_EVENTS_THRESH)))
+               __mem_cgroup_target_update(mem, MEM_CGROUP_TARGET_THRESH);
+               if (unlikely(__memcg_event_check(mem,
+                       MEM_CGROUP_TARGET_SOFTLIMIT))){
                        mem_cgroup_update_tree(mem, page);
+                       __mem_cgroup_target_update(mem,
+                               MEM_CGROUP_TARGET_SOFTLIMIT);
+               }
        }
 }
 
@@ -1787,7 +1818,7 @@ static void mem_cgroup_drain_pcp_counter(struct mem_cgroup *mem, int cpu)
 
        spin_lock(&mem->pcp_counter_lock);
        for (i = 0; i < MEM_CGROUP_STAT_DATA; i++) {
-               s64 x = per_cpu(mem->stat->count[i], cpu);
+               long x = per_cpu(mem->stat->count[i], cpu);
 
                per_cpu(mem->stat->count[i], cpu) = 0;
                mem->nocpu_base.count[i] += x;
@@ -3499,13 +3530,13 @@ static int mem_cgroup_hierarchy_write(struct cgroup *cont, struct cftype *cft,
 }
 
 
-static u64 mem_cgroup_get_recursive_idx_stat(struct mem_cgroup *mem,
-                               enum mem_cgroup_stat_index idx)
+static unsigned long mem_cgroup_recursive_stat(struct mem_cgroup *mem,
+                                              enum mem_cgroup_stat_index idx)
 {
        struct mem_cgroup *iter;
-       s64 val = 0;
+       long val = 0;
 
-       /* each per cpu's value can be minus.Then, use s64 */
+       /* Per-cpu values can be negative, use a signed accumulator */
        for_each_mem_cgroup_tree(iter, mem)
                val += mem_cgroup_read_stat(iter, idx);
 
@@ -3525,12 +3556,11 @@ static inline u64 mem_cgroup_usage(struct mem_cgroup *mem, bool swap)
                        return res_counter_read_u64(&mem->memsw, RES_USAGE);
        }
 
-       val = mem_cgroup_get_recursive_idx_stat(mem, MEM_CGROUP_STAT_CACHE);
-       val += mem_cgroup_get_recursive_idx_stat(mem, MEM_CGROUP_STAT_RSS);
+       val = mem_cgroup_recursive_stat(mem, MEM_CGROUP_STAT_CACHE);
+       val += mem_cgroup_recursive_stat(mem, MEM_CGROUP_STAT_RSS);
 
        if (swap)
-               val += mem_cgroup_get_recursive_idx_stat(mem,
-                               MEM_CGROUP_STAT_SWAPOUT);
+               val += mem_cgroup_recursive_stat(mem, MEM_CGROUP_STAT_SWAPOUT);
 
        return val << PAGE_SHIFT;
 }