Merge tag 'fixes-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/arm...
[firefly-linux-kernel-4.4.55.git] / mm / memcontrol.c
index 3be1afab85235d0c2bbc9bd2aeeefa96f4ea2ee9..795e525afaba8914f3f8863de6a7299d84898ada 100644 (file)
@@ -379,7 +379,6 @@ static bool move_file(void)
 enum charge_type {
        MEM_CGROUP_CHARGE_TYPE_CACHE = 0,
        MEM_CGROUP_CHARGE_TYPE_ANON,
-       MEM_CGROUP_CHARGE_TYPE_SHMEM,   /* used by page migration of shmem */
        MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */
        MEM_CGROUP_CHARGE_TYPE_DROP,    /* a page was unused swap cache */
        NR_CHARGE_TYPE,
@@ -406,6 +405,12 @@ enum charge_type {
 static void mem_cgroup_get(struct mem_cgroup *memcg);
 static void mem_cgroup_put(struct mem_cgroup *memcg);
 
+static inline
+struct mem_cgroup *mem_cgroup_from_css(struct cgroup_subsys_state *s)
+{
+       return container_of(s, struct mem_cgroup, css);
+}
+
 /* Writing them here to avoid exposing memcg's inner layout */
 #ifdef CONFIG_MEMCG_KMEM
 #include <net/sock.h>
@@ -863,9 +868,8 @@ static void memcg_check_events(struct mem_cgroup *memcg, struct page *page)
 
 struct mem_cgroup *mem_cgroup_from_cont(struct cgroup *cont)
 {
-       return container_of(cgroup_subsys_state(cont,
-                               mem_cgroup_subsys_id), struct mem_cgroup,
-                               css);
+       return mem_cgroup_from_css(
+               cgroup_subsys_state(cont, mem_cgroup_subsys_id));
 }
 
 struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
@@ -878,8 +882,7 @@ struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p)
        if (unlikely(!p))
                return NULL;
 
-       return container_of(task_subsys_state(p, mem_cgroup_subsys_id),
-                               struct mem_cgroup, css);
+       return mem_cgroup_from_css(task_subsys_state(p, mem_cgroup_subsys_id));
 }
 
 struct mem_cgroup *try_get_mem_cgroup_from_mm(struct mm_struct *mm)
@@ -965,8 +968,7 @@ struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root,
                css = css_get_next(&mem_cgroup_subsys, id + 1, &root->css, &id);
                if (css) {
                        if (css == &root->css || css_tryget(css))
-                               memcg = container_of(css,
-                                                    struct mem_cgroup, css);
+                               memcg = mem_cgroup_from_css(css);
                } else
                        id = 0;
                rcu_read_unlock();
@@ -2334,7 +2336,7 @@ static int __mem_cgroup_try_charge(struct mm_struct *mm,
         * We always charge the cgroup the mm_struct belongs to.
         * The mm_struct's mem_cgroup changes on task migration if the
         * thread group leader migrates. It's possible that mm is not
-        * set, if so charge the init_mm (happens for pagecache usage).
+        * set, if so charge the root memcg (happens for pagecache usage).
         */
        if (!*ptr && !mm)
                *ptr = root_mem_cgroup;
@@ -2495,7 +2497,7 @@ static struct mem_cgroup *mem_cgroup_lookup(unsigned short id)
        css = css_lookup(&mem_cgroup_subsys, id);
        if (!css)
                return NULL;
-       return container_of(css, struct mem_cgroup, css);
+       return mem_cgroup_from_css(css);
 }
 
 struct mem_cgroup *try_get_mem_cgroup_from_page(struct page *page)
@@ -2539,11 +2541,7 @@ static void __mem_cgroup_commit_charge(struct mem_cgroup *memcg,
        bool anon;
 
        lock_page_cgroup(pc);
-       if (unlikely(PageCgroupUsed(pc))) {
-               unlock_page_cgroup(pc);
-               __mem_cgroup_cancel_charge(memcg, nr_pages);
-               return;
-       }
+       VM_BUG_ON(PageCgroupUsed(pc));
        /*
         * we don't need page_cgroup_lock about tail pages, becase they are not
         * accessed by any other context at this point.
@@ -2802,27 +2800,26 @@ int mem_cgroup_newpage_charge(struct page *page,
  * struct page_cgroup is acquired. This refcnt will be consumed by
  * "commit()" or removed by "cancel()"
  */
-int mem_cgroup_try_charge_swapin(struct mm_struct *mm,
-                                struct page *page,
-                                gfp_t mask, struct mem_cgroup **memcgp)
+static int __mem_cgroup_try_charge_swapin(struct mm_struct *mm,
+                                         struct page *page,
+                                         gfp_t mask,
+                                         struct mem_cgroup **memcgp)
 {
        struct mem_cgroup *memcg;
+       struct page_cgroup *pc;
        int ret;
 
-       *memcgp = NULL;
-
-       if (mem_cgroup_disabled())
-               return 0;
-
-       if (!do_swap_account)
-               goto charge_cur_mm;
+       pc = lookup_page_cgroup(page);
        /*
-        * A racing thread's fault, or swapoff, may have already updated
-        * the pte, and even removed page from swap cache: in those cases
-        * do_swap_page()'s pte_same() test will fail; but there's also a
-        * KSM case which does need to charge the page.
+        * Every swap fault against a single page tries to charge the
+        * page, bail as early as possible.  shmem_unuse() encounters
+        * already charged pages, too.  The USED bit is protected by
+        * the page lock, which serializes swap cache removal, which
+        * in turn serializes uncharging.
         */
-       if (!PageSwapCache(page))
+       if (PageCgroupUsed(pc))
+               return 0;
+       if (!do_swap_account)
                goto charge_cur_mm;
        memcg = try_get_mem_cgroup_from_page(page);
        if (!memcg)
@@ -2834,14 +2831,35 @@ int mem_cgroup_try_charge_swapin(struct mm_struct *mm,
                ret = 0;
        return ret;
 charge_cur_mm:
-       if (unlikely(!mm))
-               mm = &init_mm;
        ret = __mem_cgroup_try_charge(mm, mask, 1, memcgp, true);
        if (ret == -EINTR)
                ret = 0;
        return ret;
 }
 
+int mem_cgroup_try_charge_swapin(struct mm_struct *mm, struct page *page,
+                                gfp_t gfp_mask, struct mem_cgroup **memcgp)
+{
+       *memcgp = NULL;
+       if (mem_cgroup_disabled())
+               return 0;
+       /*
+        * A racing thread's fault, or swapoff, may have already
+        * updated the pte, and even removed page from swap cache: in
+        * those cases unuse_pte()'s pte_same() test will fail; but
+        * there's also a KSM case which does need to charge the page.
+        */
+       if (!PageSwapCache(page)) {
+               int ret;
+
+               ret = __mem_cgroup_try_charge(mm, gfp_mask, 1, memcgp, true);
+               if (ret == -EINTR)
+                       ret = 0;
+               return ret;
+       }
+       return __mem_cgroup_try_charge_swapin(mm, page, gfp_mask, memcgp);
+}
+
 void mem_cgroup_cancel_charge_swapin(struct mem_cgroup *memcg)
 {
        if (mem_cgroup_disabled())
@@ -2900,15 +2918,11 @@ int mem_cgroup_cache_charge(struct page *page, struct mm_struct *mm,
        if (PageCompound(page))
                return 0;
 
-       if (unlikely(!mm))
-               mm = &init_mm;
-       if (!page_is_file_cache(page))
-               type = MEM_CGROUP_CHARGE_TYPE_SHMEM;
-
        if (!PageSwapCache(page))
                ret = mem_cgroup_charge_common(page, mm, gfp_mask, type);
        else { /* page is swapcache/shmem */
-               ret = mem_cgroup_try_charge_swapin(mm, page, gfp_mask, &memcg);
+               ret = __mem_cgroup_try_charge_swapin(mm, page,
+                                                    gfp_mask, &memcg);
                if (!ret)
                        __mem_cgroup_commit_charge_swapin(page, memcg, type);
        }
@@ -3310,10 +3324,8 @@ void mem_cgroup_prepare_migration(struct page *page, struct page *newpage,
         */
        if (PageAnon(page))
                ctype = MEM_CGROUP_CHARGE_TYPE_ANON;
-       else if (page_is_file_cache(page))
-               ctype = MEM_CGROUP_CHARGE_TYPE_CACHE;
        else
-               ctype = MEM_CGROUP_CHARGE_TYPE_SHMEM;
+               ctype = MEM_CGROUP_CHARGE_TYPE_CACHE;
        /*
         * The page is committed to the memcg, but it's not actually
         * charged to the res_counter since we plan on replacing the
@@ -3407,10 +3419,6 @@ void mem_cgroup_replace_page_cache(struct page *oldpage,
         */
        if (!memcg)
                return;
-
-       if (PageSwapBacked(oldpage))
-               type = MEM_CGROUP_CHARGE_TYPE_SHMEM;
-
        /*
         * Even if newpage->mapping was NULL before starting replacement,
         * the newpage may be on LRU(or pagevec for LRU) already. We lock