memcg: cpu hotplug aware quick acount_move detection
KAMEZAWA Hiroyuki [Wed, 27 Oct 2010 22:33:42 +0000 (15:33 -0700)]
An event counter MEM_CGROUP_ON_MOVE is used for quick check whether file
stat update can be done in async manner or not.  Now, it use percpu
counter and for_each_possible_cpu to update.

This patch replaces for_each_possible_cpu to for_each_online_cpu and adds
necessary synchronization logic at CPU HOTPLUG.

[akpm@linux-foundation.org: coding-style fixes]
Signed-off-by: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
Cc: Balbir Singh <balbir@in.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 31a1d3b..52840ad 100644 (file)
@@ -1132,11 +1132,14 @@ static unsigned int get_swappiness(struct mem_cgroup *memcg)
 static void mem_cgroup_start_move(struct mem_cgroup *mem)
 {
        int cpu;
-       /* Because this is for moving account, reuse mc.lock */
-       spin_lock(&mc.lock);
-       for_each_possible_cpu(cpu)
+
+       get_online_cpus();
+       spin_lock(&mem->pcp_counter_lock);
+       for_each_online_cpu(cpu)
                per_cpu(mem->stat->count[MEM_CGROUP_ON_MOVE], cpu) += 1;
-       spin_unlock(&mc.lock);
+       mem->nocpu_base.count[MEM_CGROUP_ON_MOVE] += 1;
+       spin_unlock(&mem->pcp_counter_lock);
+       put_online_cpus();
 
        synchronize_rcu();
 }
@@ -1147,10 +1150,13 @@ static void mem_cgroup_end_move(struct mem_cgroup *mem)
 
        if (!mem)
                return;
-       spin_lock(&mc.lock);
-       for_each_possible_cpu(cpu)
+       get_online_cpus();
+       spin_lock(&mem->pcp_counter_lock);
+       for_each_online_cpu(cpu)
                per_cpu(mem->stat->count[MEM_CGROUP_ON_MOVE], cpu) -= 1;
-       spin_unlock(&mc.lock);
+       mem->nocpu_base.count[MEM_CGROUP_ON_MOVE] -= 1;
+       spin_unlock(&mem->pcp_counter_lock);
+       put_online_cpus();
 }
 /*
  * 2 routines for checking "mem" is under move_account() or not.
@@ -1751,6 +1757,17 @@ static void mem_cgroup_drain_pcp_counter(struct mem_cgroup *mem, int cpu)
                per_cpu(mem->stat->count[i], cpu) = 0;
                mem->nocpu_base.count[i] += x;
        }
+       /* need to clear ON_MOVE value, works as a kind of lock. */
+       per_cpu(mem->stat->count[MEM_CGROUP_ON_MOVE], cpu) = 0;
+       spin_unlock(&mem->pcp_counter_lock);
+}
+
+static void synchronize_mem_cgroup_on_move(struct mem_cgroup *mem, int cpu)
+{
+       int idx = MEM_CGROUP_ON_MOVE;
+
+       spin_lock(&mem->pcp_counter_lock);
+       per_cpu(mem->stat->count[idx], cpu) = mem->nocpu_base.count[idx];
        spin_unlock(&mem->pcp_counter_lock);
 }
 
@@ -1762,6 +1779,12 @@ static int __cpuinit memcg_cpu_hotplug_callback(struct notifier_block *nb,
        struct memcg_stock_pcp *stock;
        struct mem_cgroup *iter;
 
+       if ((action == CPU_ONLINE)) {
+               for_each_mem_cgroup_all(iter)
+                       synchronize_mem_cgroup_on_move(iter, cpu);
+               return NOTIFY_OK;
+       }
+
        if ((action != CPU_DEAD) || action != CPU_DEAD_FROZEN)
                return NOTIFY_OK;