memcg: coalesce uncharge during unmap/truncate

[pandora-kernel.git] / mm / memcontrol.c

diff --git a/mm/memcontrol.c b/mm/memcontrol.c
index 7b5b108..a730c91 100644 (file)
--- a/mm/memcontrol.c
+++ b/mm/memcontrol.c
@@ -1827,6 +1827,50 @@ void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *mem)
        css_put(&mem->css);
 }
 
+static void
+__do_uncharge(struct mem_cgroup *mem, const enum charge_type ctype)
+{
+       struct memcg_batch_info *batch = NULL;
+       bool uncharge_memsw = true;
+       /* If swapout, usage of swap doesn't decrease */
+       if (!do_swap_account || ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT)
+               uncharge_memsw = false;
+       /*
+        * do_batch > 0 when unmapping pages or inode invalidate/truncate.
+        * In those cases, all pages freed continously can be expected to be in
+        * the same cgroup and we have chance to coalesce uncharges.
+        * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE)
+        * because we want to do uncharge as soon as possible.
+        */
+       if (!current->memcg_batch.do_batch || test_thread_flag(TIF_MEMDIE))
+               goto direct_uncharge;
+
+       batch = &current->memcg_batch;
+       /*
+        * In usual, we do css_get() when we remember memcg pointer.
+        * But in this case, we keep res->usage until end of a series of
+        * uncharges. Then, it's ok to ignore memcg's refcnt.
+        */
+       if (!batch->memcg)
+               batch->memcg = mem;
+       /*
+        * In typical case, batch->memcg == mem. This means we can
+        * merge a series of uncharges to an uncharge of res_counter.
+        * If not, we uncharge res_counter ony by one.
+        */
+       if (batch->memcg != mem)
+               goto direct_uncharge;
+       /* remember freed charge and uncharge it later */
+       batch->bytes += PAGE_SIZE;
+       if (uncharge_memsw)
+               batch->memsw_bytes += PAGE_SIZE;
+       return;
+direct_uncharge:
+       res_counter_uncharge(&mem->res, PAGE_SIZE);
+       if (uncharge_memsw)
+               res_counter_uncharge(&mem->memsw, PAGE_SIZE);
+       return;
+}
 
 /*
  * uncharge if !page_mapped(page)
@@ -1875,12 +1919,8 @@ __mem_cgroup_uncharge_common(struct page *page, enum charge_type ctype)
                break;
        }
 
-       if (!mem_cgroup_is_root(mem)) {
-               res_counter_uncharge(&mem->res, PAGE_SIZE);
-               if (do_swap_account &&
-                               (ctype != MEM_CGROUP_CHARGE_TYPE_SWAPOUT))
-                       res_counter_uncharge(&mem->memsw, PAGE_SIZE);
-       }
+       if (!mem_cgroup_is_root(mem))
+               __do_uncharge(mem, ctype);
        if (ctype == MEM_CGROUP_CHARGE_TYPE_SWAPOUT)
                mem_cgroup_swap_statistics(mem, true);
        mem_cgroup_charge_statistics(mem, pc, false);
@@ -1926,6 +1966,50 @@ void mem_cgroup_uncharge_cache_page(struct page *page)
        __mem_cgroup_uncharge_common(page, MEM_CGROUP_CHARGE_TYPE_CACHE);
 }
 
+/*
+ * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate.
+ * In that cases, pages are freed continuously and we can expect pages
+ * are in the same memcg. All these calls itself limits the number of
+ * pages freed at once, then uncharge_start/end() is called properly.
+ * This may be called prural(2) times in a context,
+ */
+
+void mem_cgroup_uncharge_start(void)
+{
+       current->memcg_batch.do_batch++;
+       /* We can do nest. */
+       if (current->memcg_batch.do_batch == 1) {
+               current->memcg_batch.memcg = NULL;
+               current->memcg_batch.bytes = 0;
+               current->memcg_batch.memsw_bytes = 0;
+       }
+}
+
+void mem_cgroup_uncharge_end(void)
+{
+       struct memcg_batch_info *batch = &current->memcg_batch;
+
+       if (!batch->do_batch)
+               return;
+
+       batch->do_batch--;
+       if (batch->do_batch) /* If stacked, do nothing. */
+               return;
+
+       if (!batch->memcg)
+               return;
+       /*
+        * This "batch->memcg" is valid without any css_get/put etc...
+        * bacause we hide charges behind us.
+        */
+       if (batch->bytes)
+               res_counter_uncharge(&batch->memcg->res, batch->bytes);
+       if (batch->memsw_bytes)
+               res_counter_uncharge(&batch->memcg->memsw, batch->memsw_bytes);
+       /* forget this pointer (for sanity check) */
+       batch->memcg = NULL;
+}
+
 #ifdef CONFIG_SWAP
 /*
  * called after __delete_from_swap_cache() and drop "page" account.