mm: memcontrol: let mem_cgroup_move_account() have effect only if MMU enabled
authorChen Gang <762976180@qq.com>
Tue, 14 Apr 2015 22:47:35 +0000 (15:47 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Tue, 14 Apr 2015 23:49:04 +0000 (16:49 -0700)
When !MMU, it will report warning. The related warning with allmodconfig
under c6x:

    CC      mm/memcontrol.o
  mm/memcontrol.c:2802:12: warning: 'mem_cgroup_move_account' defined but not used [-Wunused-function]
   static int mem_cgroup_move_account(struct page *page,
              ^

Signed-off-by: Chen Gang <gang.chen.5i5j@gmail.com>
Acked-by: Michal Hocko <mhocko@suse.cz>
Acked-by: 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 68d4890fc4bdd40006a3542900b46c18fb9eb445..f227786e73db2fa7dd190cdba993bc0d53c42460 100644 (file)
@@ -2785,92 +2785,6 @@ void mem_cgroup_split_huge_fixup(struct page *head)
 }
 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
 
-/**
- * mem_cgroup_move_account - move account of the page
- * @page: the page
- * @nr_pages: number of regular pages (>1 for huge pages)
- * @from: mem_cgroup which the page is moved from.
- * @to:        mem_cgroup which the page is moved to. @from != @to.
- *
- * The caller must confirm following.
- * - page is not on LRU (isolate_page() is useful.)
- * - compound_lock is held when nr_pages > 1
- *
- * This function doesn't do "charge" to new cgroup and doesn't do "uncharge"
- * from old cgroup.
- */
-static int mem_cgroup_move_account(struct page *page,
-                                  unsigned int nr_pages,
-                                  struct mem_cgroup *from,
-                                  struct mem_cgroup *to)
-{
-       unsigned long flags;
-       int ret;
-
-       VM_BUG_ON(from == to);
-       VM_BUG_ON_PAGE(PageLRU(page), page);
-       /*
-        * The page is isolated from LRU. So, collapse function
-        * will not handle this page. But page splitting can happen.
-        * Do this check under compound_page_lock(). The caller should
-        * hold it.
-        */
-       ret = -EBUSY;
-       if (nr_pages > 1 && !PageTransHuge(page))
-               goto out;
-
-       /*
-        * Prevent mem_cgroup_migrate() from looking at page->mem_cgroup
-        * of its source page while we change it: page migration takes
-        * both pages off the LRU, but page cache replacement doesn't.
-        */
-       if (!trylock_page(page))
-               goto out;
-
-       ret = -EINVAL;
-       if (page->mem_cgroup != from)
-               goto out_unlock;
-
-       spin_lock_irqsave(&from->move_lock, flags);
-
-       if (!PageAnon(page) && page_mapped(page)) {
-               __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED],
-                              nr_pages);
-               __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED],
-                              nr_pages);
-       }
-
-       if (PageWriteback(page)) {
-               __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_WRITEBACK],
-                              nr_pages);
-               __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_WRITEBACK],
-                              nr_pages);
-       }
-
-       /*
-        * It is safe to change page->mem_cgroup here because the page
-        * is referenced, charged, and isolated - we can't race with
-        * uncharging, charging, migration, or LRU putback.
-        */
-
-       /* caller should have done css_get */
-       page->mem_cgroup = to;
-       spin_unlock_irqrestore(&from->move_lock, flags);
-
-       ret = 0;
-
-       local_irq_disable();
-       mem_cgroup_charge_statistics(to, page, nr_pages);
-       memcg_check_events(to, page);
-       mem_cgroup_charge_statistics(from, page, -nr_pages);
-       memcg_check_events(from, page);
-       local_irq_enable();
-out_unlock:
-       unlock_page(page);
-out:
-       return ret;
-}
-
 #ifdef CONFIG_MEMCG_SWAP
 static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg,
                                         bool charge)
@@ -4822,6 +4736,92 @@ static struct page *mc_handle_file_pte(struct vm_area_struct *vma,
        return page;
 }
 
+/**
+ * mem_cgroup_move_account - move account of the page
+ * @page: the page
+ * @nr_pages: number of regular pages (>1 for huge pages)
+ * @from: mem_cgroup which the page is moved from.
+ * @to:        mem_cgroup which the page is moved to. @from != @to.
+ *
+ * The caller must confirm following.
+ * - page is not on LRU (isolate_page() is useful.)
+ * - compound_lock is held when nr_pages > 1
+ *
+ * This function doesn't do "charge" to new cgroup and doesn't do "uncharge"
+ * from old cgroup.
+ */
+static int mem_cgroup_move_account(struct page *page,
+                                  unsigned int nr_pages,
+                                  struct mem_cgroup *from,
+                                  struct mem_cgroup *to)
+{
+       unsigned long flags;
+       int ret;
+
+       VM_BUG_ON(from == to);
+       VM_BUG_ON_PAGE(PageLRU(page), page);
+       /*
+        * The page is isolated from LRU. So, collapse function
+        * will not handle this page. But page splitting can happen.
+        * Do this check under compound_page_lock(). The caller should
+        * hold it.
+        */
+       ret = -EBUSY;
+       if (nr_pages > 1 && !PageTransHuge(page))
+               goto out;
+
+       /*
+        * Prevent mem_cgroup_migrate() from looking at page->mem_cgroup
+        * of its source page while we change it: page migration takes
+        * both pages off the LRU, but page cache replacement doesn't.
+        */
+       if (!trylock_page(page))
+               goto out;
+
+       ret = -EINVAL;
+       if (page->mem_cgroup != from)
+               goto out_unlock;
+
+       spin_lock_irqsave(&from->move_lock, flags);
+
+       if (!PageAnon(page) && page_mapped(page)) {
+               __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_FILE_MAPPED],
+                              nr_pages);
+               __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_FILE_MAPPED],
+                              nr_pages);
+       }
+
+       if (PageWriteback(page)) {
+               __this_cpu_sub(from->stat->count[MEM_CGROUP_STAT_WRITEBACK],
+                              nr_pages);
+               __this_cpu_add(to->stat->count[MEM_CGROUP_STAT_WRITEBACK],
+                              nr_pages);
+       }
+
+       /*
+        * It is safe to change page->mem_cgroup here because the page
+        * is referenced, charged, and isolated - we can't race with
+        * uncharging, charging, migration, or LRU putback.
+        */
+
+       /* caller should have done css_get */
+       page->mem_cgroup = to;
+       spin_unlock_irqrestore(&from->move_lock, flags);
+
+       ret = 0;
+
+       local_irq_disable();
+       mem_cgroup_charge_statistics(to, page, nr_pages);
+       memcg_check_events(to, page);
+       mem_cgroup_charge_statistics(from, page, -nr_pages);
+       memcg_check_events(from, page);
+       local_irq_enable();
+out_unlock:
+       unlock_page(page);
+out:
+       return ret;
+}
+
 static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma,
                unsigned long addr, pte_t ptent, union mc_target *target)
 {