Memory controller: memory accounting
authorBalbir Singh <balbir@linux.vnet.ibm.com>
Thu, 7 Feb 2008 08:13:53 +0000 (00:13 -0800)
committerLinus Torvalds <torvalds@woody.linux-foundation.org>
Thu, 7 Feb 2008 16:42:18 +0000 (08:42 -0800)
Add the accounting hooks.  The accounting is carried out for RSS and Page
Cache (unmapped) pages.  There is now a common limit and accounting for both.
The RSS accounting is accounted at page_add_*_rmap() and page_remove_rmap()
time.  Page cache is accounted at add_to_page_cache(),
__delete_from_page_cache().  Swap cache is also accounted for.

Each page's page_cgroup is protected with the last bit of the
page_cgroup pointer, this makes handling of race conditions involving
simultaneous mappings of a page easier.  A reference count is kept in the
page_cgroup to deal with cases where a page might be unmapped from the RSS
of all tasks, but still lives in the page cache.

Credits go to Vaidyanathan Srinivasan for helping with reference counting work
of the page cgroup.  Almost all of the page cache accounting code has help
from Vaidyanathan Srinivasan.

[hugh@veritas.com: fix swapoff breakage]
[akpm@linux-foundation.org: fix locking]
Signed-off-by: Vaidyanathan Srinivasan <svaidy@linux.vnet.ibm.com>
Signed-off-by: Balbir Singh <balbir@linux.vnet.ibm.com>
Cc: Pavel Emelianov <xemul@openvz.org>
Cc: Paul Menage <menage@google.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Cc: Nick Piggin <nickpiggin@yahoo.com.au>
Cc: Kirill Korotaev <dev@sw.ru>
Cc: Herbert Poetzl <herbert@13thfloor.at>
Cc: David Rientjes <rientjes@google.com>
Cc: <Valdis.Kletnieks@vt.edu>
Signed-off-by: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
include/linux/memcontrol.h
mm/filemap.c
mm/memcontrol.c
mm/memory.c
mm/migrate.c
mm/page_alloc.c
mm/rmap.c
mm/swap_state.c
mm/swapfile.c

index 7d1f119c796efb42e470572d2e05dd607d267408..f5b47efab48b8d9d602f8b65df20e5334144e822 100644 (file)
@@ -30,6 +30,13 @@ extern void mm_free_cgroup(struct mm_struct *mm);
 extern void page_assign_page_cgroup(struct page *page,
                                        struct page_cgroup *pc);
 extern struct page_cgroup *page_get_page_cgroup(struct page *page);
+extern int mem_cgroup_charge(struct page *page, struct mm_struct *mm);
+extern void mem_cgroup_uncharge(struct page_cgroup *pc);
+
+static inline void mem_cgroup_uncharge_page(struct page *page)
+{
+       mem_cgroup_uncharge(page_get_page_cgroup(page));
+}
 
 #else /* CONFIG_CGROUP_MEM_CONT */
 static inline void mm_init_cgroup(struct mm_struct *mm,
@@ -51,6 +58,19 @@ static inline struct page_cgroup *page_get_page_cgroup(struct page *page)
        return NULL;
 }
 
+static inline int mem_cgroup_charge(struct page *page, struct mm_struct *mm)
+{
+       return 0;
+}
+
+static inline void mem_cgroup_uncharge(struct page_cgroup *pc)
+{
+}
+
+static inline void mem_cgroup_uncharge_page(struct page *page)
+{
+}
+
 #endif /* CONFIG_CGROUP_MEM_CONT */
 
 #endif /* _LINUX_MEMCONTROL_H */
index 81fb9bff0d4f9022db02f9d3194ebc5ccd0c350e..b7a01e9279534d43d13c1c99aa6f20ac5b1c930c 100644 (file)
@@ -33,6 +33,7 @@
 #include <linux/syscalls.h>
 #include <linux/cpuset.h>
 #include <linux/hardirq.h> /* for BUG_ON(!in_atomic()) only */
+#include <linux/memcontrol.h>
 #include "internal.h"
 
 /*
@@ -118,6 +119,7 @@ void __remove_from_page_cache(struct page *page)
 {
        struct address_space *mapping = page->mapping;
 
+       mem_cgroup_uncharge_page(page);
        radix_tree_delete(&mapping->page_tree, page->index);
        page->mapping = NULL;
        mapping->nrpages--;
@@ -461,6 +463,11 @@ int add_to_page_cache(struct page *page, struct address_space *mapping,
        int error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
 
        if (error == 0) {
+
+               error = mem_cgroup_charge(page, current->mm);
+               if (error)
+                       goto out;
+
                write_lock_irq(&mapping->tree_lock);
                error = radix_tree_insert(&mapping->page_tree, offset, page);
                if (!error) {
@@ -470,10 +477,13 @@ int add_to_page_cache(struct page *page, struct address_space *mapping,
                        page->index = offset;
                        mapping->nrpages++;
                        __inc_zone_page_state(page, NR_FILE_PAGES);
-               }
+               } else
+                       mem_cgroup_uncharge_page(page);
+
                write_unlock_irq(&mapping->tree_lock);
                radix_tree_preload_end();
        }
+out:
        return error;
 }
 EXPORT_SYMBOL(add_to_page_cache);
index 4d4805eb37c77b06a98a9d06690982420734e077..ebca767292dc2ad2aac09d47e1c819e600fa02ce 100644 (file)
@@ -21,6 +21,9 @@
 #include <linux/memcontrol.h>
 #include <linux/cgroup.h>
 #include <linux/mm.h>
+#include <linux/page-flags.h>
+#include <linux/bit_spinlock.h>
+#include <linux/rcupdate.h>
 
 struct cgroup_subsys mem_cgroup_subsys;
 
@@ -31,7 +34,9 @@ struct cgroup_subsys mem_cgroup_subsys;
  * to help the administrator determine what knobs to tune.
  *
  * TODO: Add a water mark for the memory controller. Reclaim will begin when
- * we hit the water mark.
+ * we hit the water mark. May be even add a low water mark, such that
+ * no reclaim occurs from a cgroup at it's low water mark, this is
+ * a feature that will be implemented much later in the future.
  */
 struct mem_cgroup {
        struct cgroup_subsys_state css;
@@ -48,6 +53,14 @@ struct mem_cgroup {
        struct list_head inactive_list;
 };
 
+/*
+ * We use the lower bit of the page->page_cgroup pointer as a bit spin
+ * lock. We need to ensure that page->page_cgroup is atleast two
+ * byte aligned (based on comments from Nick Piggin)
+ */
+#define PAGE_CGROUP_LOCK_BIT   0x0
+#define PAGE_CGROUP_LOCK               (1 << PAGE_CGROUP_LOCK_BIT)
+
 /*
  * A page_cgroup page is associated with every page descriptor. The
  * page_cgroup helps us identify information about the cgroup
@@ -56,6 +69,8 @@ struct page_cgroup {
        struct list_head lru;           /* per cgroup LRU list */
        struct page *page;
        struct mem_cgroup *mem_cgroup;
+       atomic_t ref_cnt;               /* Helpful when pages move b/w  */
+                                       /* mapped and cached states     */
 };
 
 
@@ -88,14 +103,157 @@ void mm_free_cgroup(struct mm_struct *mm)
        css_put(&mm->mem_cgroup->css);
 }
 
+static inline int page_cgroup_locked(struct page *page)
+{
+       return bit_spin_is_locked(PAGE_CGROUP_LOCK_BIT,
+                                       &page->page_cgroup);
+}
+
 void page_assign_page_cgroup(struct page *page, struct page_cgroup *pc)
 {
-       page->page_cgroup = (unsigned long)pc;
+       int locked;
+
+       /*
+        * While resetting the page_cgroup we might not hold the
+        * page_cgroup lock. free_hot_cold_page() is an example
+        * of such a scenario
+        */
+       if (pc)
+               VM_BUG_ON(!page_cgroup_locked(page));
+       locked = (page->page_cgroup & PAGE_CGROUP_LOCK);
+       page->page_cgroup = ((unsigned long)pc | locked);
 }
 
 struct page_cgroup *page_get_page_cgroup(struct page *page)
 {
-       return page->page_cgroup;
+       return (struct page_cgroup *)
+               (page->page_cgroup & ~PAGE_CGROUP_LOCK);
+}
+
+void __always_inline lock_page_cgroup(struct page *page)
+{
+       bit_spin_lock(PAGE_CGROUP_LOCK_BIT, &page->page_cgroup);
+       VM_BUG_ON(!page_cgroup_locked(page));
+}
+
+void __always_inline unlock_page_cgroup(struct page *page)
+{
+       bit_spin_unlock(PAGE_CGROUP_LOCK_BIT, &page->page_cgroup);
+}
+
+/*
+ * Charge the memory controller for page usage.
+ * Return
+ * 0 if the charge was successful
+ * < 0 if the cgroup is over its limit
+ */
+int mem_cgroup_charge(struct page *page, struct mm_struct *mm)
+{
+       struct mem_cgroup *mem;
+       struct page_cgroup *pc, *race_pc;
+
+       /*
+        * Should page_cgroup's go to their own slab?
+        * One could optimize the performance of the charging routine
+        * by saving a bit in the page_flags and using it as a lock
+        * to see if the cgroup page already has a page_cgroup associated
+        * with it
+        */
+       lock_page_cgroup(page);
+       pc = page_get_page_cgroup(page);
+       /*
+        * The page_cgroup exists and the page has already been accounted
+        */
+       if (pc) {
+               atomic_inc(&pc->ref_cnt);
+               goto done;
+       }
+
+       unlock_page_cgroup(page);
+
+       pc = kzalloc(sizeof(struct page_cgroup), GFP_KERNEL);
+       if (pc == NULL)
+               goto err;
+
+       rcu_read_lock();
+       /*
+        * 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).
+        */
+       if (!mm)
+               mm = &init_mm;
+
+       mem = rcu_dereference(mm->mem_cgroup);
+       /*
+        * For every charge from the cgroup, increment reference
+        * count
+        */
+       css_get(&mem->css);
+       rcu_read_unlock();
+
+       /*
+        * If we created the page_cgroup, we should free it on exceeding
+        * the cgroup limit.
+        */
+       if (res_counter_charge(&mem->res, 1)) {
+               css_put(&mem->css);
+               goto free_pc;
+       }
+
+       lock_page_cgroup(page);
+       /*
+        * Check if somebody else beat us to allocating the page_cgroup
+        */
+       race_pc = page_get_page_cgroup(page);
+       if (race_pc) {
+               kfree(pc);
+               pc = race_pc;
+               atomic_inc(&pc->ref_cnt);
+               res_counter_uncharge(&mem->res, 1);
+               css_put(&mem->css);
+               goto done;
+       }
+
+       atomic_set(&pc->ref_cnt, 1);
+       pc->mem_cgroup = mem;
+       pc->page = page;
+       page_assign_page_cgroup(page, pc);
+
+done:
+       unlock_page_cgroup(page);
+       return 0;
+free_pc:
+       kfree(pc);
+       return -ENOMEM;
+err:
+       unlock_page_cgroup(page);
+       return -ENOMEM;
+}
+
+/*
+ * Uncharging is always a welcome operation, we never complain, simply
+ * uncharge.
+ */
+void mem_cgroup_uncharge(struct page_cgroup *pc)
+{
+       struct mem_cgroup *mem;
+       struct page *page;
+
+       if (!pc)
+               return;
+
+       if (atomic_dec_and_test(&pc->ref_cnt)) {
+               page = pc->page;
+               lock_page_cgroup(page);
+               mem = pc->mem_cgroup;
+               css_put(&mem->css);
+               page_assign_page_cgroup(page, NULL);
+               unlock_page_cgroup(page);
+               res_counter_uncharge(&mem->res, 1);
+               kfree(pc);
+       }
 }
 
 static ssize_t mem_cgroup_read(struct cgroup *cont, struct cftype *cft,
@@ -150,6 +308,8 @@ mem_cgroup_create(struct cgroup_subsys *ss, struct cgroup *cont)
                return NULL;
 
        res_counter_init(&mem->res);
+       INIT_LIST_HEAD(&mem->active_list);
+       INIT_LIST_HEAD(&mem->inactive_list);
        return &mem->css;
 }
 
index 9d073fa0a2d02631f53ba3117e80b32efc2b9ae9..0ba224ea6ba4b723c50a8c231c53fbf4e56d2cc6 100644 (file)
@@ -50,6 +50,7 @@
 #include <linux/delayacct.h>
 #include <linux/init.h>
 #include <linux/writeback.h>
+#include <linux/memcontrol.h>
 
 #include <asm/pgalloc.h>
 #include <asm/uaccess.h>
@@ -1144,16 +1145,20 @@ static int insert_page(struct mm_struct *mm, unsigned long addr, struct page *pa
 {
        int retval;
        pte_t *pte;
-       spinlock_t *ptl;  
+       spinlock_t *ptl;
+
+       retval = mem_cgroup_charge(page, mm);
+       if (retval)
+               goto out;
 
        retval = -EINVAL;
        if (PageAnon(page))
-               goto out;
+               goto out_uncharge;
        retval = -ENOMEM;
        flush_dcache_page(page);
        pte = get_locked_pte(mm, addr, &ptl);
        if (!pte)
-               goto out;
+               goto out_uncharge;
        retval = -EBUSY;
        if (!pte_none(*pte))
                goto out_unlock;
@@ -1165,8 +1170,12 @@ static int insert_page(struct mm_struct *mm, unsigned long addr, struct page *pa
        set_pte_at(mm, addr, pte, mk_pte(page, prot));
 
        retval = 0;
+       pte_unmap_unlock(pte, ptl);
+       return retval;
 out_unlock:
        pte_unmap_unlock(pte, ptl);
+out_uncharge:
+       mem_cgroup_uncharge_page(page);
 out:
        return retval;
 }
@@ -1641,6 +1650,9 @@ gotten:
        cow_user_page(new_page, old_page, address, vma);
        __SetPageUptodate(new_page);
 
+       if (mem_cgroup_charge(new_page, mm))
+               goto oom_free_new;
+
        /*
         * Re-check the pte - we dropped the lock
         */
@@ -1672,7 +1684,9 @@ gotten:
                /* Free the old page.. */
                new_page = old_page;
                ret |= VM_FAULT_WRITE;
-       }
+       } else
+               mem_cgroup_uncharge_page(new_page);
+
        if (new_page)
                page_cache_release(new_page);
        if (old_page)
@@ -1696,6 +1710,8 @@ unlock:
                put_page(dirty_page);
        }
        return ret;
+oom_free_new:
+       __free_page(new_page);
 oom:
        if (old_page)
                page_cache_release(old_page);
@@ -2036,6 +2052,12 @@ static int do_swap_page(struct mm_struct *mm, struct vm_area_struct *vma,
                count_vm_event(PGMAJFAULT);
        }
 
+       if (mem_cgroup_charge(page, mm)) {
+               delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
+               ret = VM_FAULT_OOM;
+               goto out;
+       }
+
        mark_page_accessed(page);
        lock_page(page);
        delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
@@ -2073,8 +2095,10 @@ static int do_swap_page(struct mm_struct *mm, struct vm_area_struct *vma,
        if (write_access) {
                /* XXX: We could OR the do_wp_page code with this one? */
                if (do_wp_page(mm, vma, address,
-                               page_table, pmd, ptl, pte) & VM_FAULT_OOM)
+                               page_table, pmd, ptl, pte) & VM_FAULT_OOM) {
+                       mem_cgroup_uncharge_page(page);
                        ret = VM_FAULT_OOM;
+               }
                goto out;
        }
 
@@ -2085,6 +2109,7 @@ unlock:
 out:
        return ret;
 out_nomap:
+       mem_cgroup_uncharge_page(page);
        pte_unmap_unlock(page_table, ptl);
        unlock_page(page);
        page_cache_release(page);
@@ -2114,6 +2139,9 @@ static int do_anonymous_page(struct mm_struct *mm, struct vm_area_struct *vma,
                goto oom;
        __SetPageUptodate(page);
 
+       if (mem_cgroup_charge(page, mm))
+               goto oom_free_page;
+
        entry = mk_pte(page, vma->vm_page_prot);
        entry = maybe_mkwrite(pte_mkdirty(entry), vma);
 
@@ -2131,8 +2159,11 @@ unlock:
        pte_unmap_unlock(page_table, ptl);
        return 0;
 release:
+       mem_cgroup_uncharge_page(page);
        page_cache_release(page);
        goto unlock;
+oom_free_page:
+       __free_page(page);
 oom:
        return VM_FAULT_OOM;
 }
@@ -2246,6 +2277,11 @@ static int __do_fault(struct mm_struct *mm, struct vm_area_struct *vma,
 
        }
 
+       if (mem_cgroup_charge(page, mm)) {
+               ret = VM_FAULT_OOM;
+               goto out;
+       }
+
        page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
 
        /*
@@ -2281,6 +2317,7 @@ static int __do_fault(struct mm_struct *mm, struct vm_area_struct *vma,
                /* no need to invalidate: a not-present page won't be cached */
                update_mmu_cache(vma, address, entry);
        } else {
+               mem_cgroup_uncharge_page(page);
                if (anon)
                        page_cache_release(page);
                else
index 857a987e36904a5850a4d663c6d9e41e494ffe1f..417bbda14e5b749b9c8512ac1b44ebcb31c346f1 100644 (file)
@@ -29,6 +29,7 @@
 #include <linux/mempolicy.h>
 #include <linux/vmalloc.h>
 #include <linux/security.h>
+#include <linux/memcontrol.h>
 
 #include "internal.h"
 
@@ -152,6 +153,11 @@ static void remove_migration_pte(struct vm_area_struct *vma,
                return;
        }
 
+       if (mem_cgroup_charge(new, mm)) {
+               pte_unmap(ptep);
+               return;
+       }
+
        ptl = pte_lockptr(mm, pmd);
        spin_lock(ptl);
        pte = *ptep;
index 37576b822f06c98c2d3342af1cdecbe0e5a8f25b..26a54a17dc9f5dfc0191da8f0dd774476ab8eee0 100644 (file)
@@ -43,6 +43,7 @@
 #include <linux/backing-dev.h>
 #include <linux/fault-inject.h>
 #include <linux/page-isolation.h>
+#include <linux/memcontrol.h>
 
 #include <asm/tlbflush.h>
 #include <asm/div64.h>
@@ -987,6 +988,7 @@ static void free_hot_cold_page(struct page *page, int cold)
 
        if (!PageHighMem(page))
                debug_check_no_locks_freed(page_address(page), PAGE_SIZE);
+       VM_BUG_ON(page_get_page_cgroup(page));
        arch_free_page(page, 0);
        kernel_map_pages(page, 1, 0);
 
@@ -2525,6 +2527,7 @@ void __meminit memmap_init_zone(unsigned long size, int nid, unsigned long zone,
                set_page_links(page, zone, nid, pfn);
                init_page_count(page);
                reset_page_mapcount(page);
+               page_assign_page_cgroup(page, NULL);
                SetPageReserved(page);
 
                /*
index 57ad276900c94903a2febd53e1c0d995958a38a9..4a3487921effd04b4ab3c3cdbddf216259366d10 100644 (file)
--- a/mm/rmap.c
+++ b/mm/rmap.c
@@ -48,6 +48,7 @@
 #include <linux/rcupdate.h>
 #include <linux/module.h>
 #include <linux/kallsyms.h>
+#include <linux/memcontrol.h>
 
 #include <asm/tlbflush.h>
 
@@ -554,8 +555,14 @@ void page_add_anon_rmap(struct page *page,
        VM_BUG_ON(address < vma->vm_start || address >= vma->vm_end);
        if (atomic_inc_and_test(&page->_mapcount))
                __page_set_anon_rmap(page, vma, address);
-       else
+       else {
                __page_check_anon_rmap(page, vma, address);
+               /*
+                * We unconditionally charged during prepare, we uncharge here
+                * This takes care of balancing the reference counts
+                */
+               mem_cgroup_uncharge_page(page);
+       }
 }
 
 /*
@@ -586,6 +593,12 @@ void page_add_file_rmap(struct page *page)
 {
        if (atomic_inc_and_test(&page->_mapcount))
                __inc_zone_page_state(page, NR_FILE_MAPPED);
+       else
+               /*
+                * We unconditionally charged during prepare, we uncharge here
+                * This takes care of balancing the reference counts
+                */
+               mem_cgroup_uncharge_page(page);
 }
 
 #ifdef CONFIG_DEBUG_VM
@@ -646,6 +659,8 @@ void page_remove_rmap(struct page *page, struct vm_area_struct *vma)
                        page_clear_dirty(page);
                        set_page_dirty(page);
                }
+               mem_cgroup_uncharge_page(page);
+
                __dec_zone_page_state(page,
                                PageAnon(page) ? NR_ANON_PAGES : NR_FILE_MAPPED);
        }
index ec42f01a8d02669fb5bd2c73dcffefb5bb619f6c..f96e3ff1e7913472107791f35360c273ab0862a4 100644 (file)
@@ -17,6 +17,7 @@
 #include <linux/backing-dev.h>
 #include <linux/pagevec.h>
 #include <linux/migrate.h>
+#include <linux/memcontrol.h>
 
 #include <asm/pgtable.h>
 
@@ -76,6 +77,11 @@ int add_to_swap_cache(struct page *page, swp_entry_t entry, gfp_t gfp_mask)
        BUG_ON(PagePrivate(page));
        error = radix_tree_preload(gfp_mask);
        if (!error) {
+
+               error = mem_cgroup_charge(page, current->mm);
+               if (error)
+                       goto out;
+
                write_lock_irq(&swapper_space.tree_lock);
                error = radix_tree_insert(&swapper_space.page_tree,
                                                entry.val, page);
@@ -86,10 +92,13 @@ int add_to_swap_cache(struct page *page, swp_entry_t entry, gfp_t gfp_mask)
                        total_swapcache_pages++;
                        __inc_zone_page_state(page, NR_FILE_PAGES);
                        INC_CACHE_INFO(add_total);
+               } else {
+                       mem_cgroup_uncharge_page(page);
                }
                write_unlock_irq(&swapper_space.tree_lock);
                radix_tree_preload_end();
        }
+out:
        return error;
 }
 
@@ -104,6 +113,7 @@ void __delete_from_swap_cache(struct page *page)
        BUG_ON(PageWriteback(page));
        BUG_ON(PagePrivate(page));
 
+       mem_cgroup_uncharge_page(page);
        radix_tree_delete(&swapper_space.page_tree, page_private(page));
        set_page_private(page, 0);
        ClearPageSwapCache(page);
index afae7b1f680b5f4140d305d2a5be7fb476f50fd5..fddc4cc4149bce91491ee9d08aa8099969793236 100644 (file)
@@ -27,6 +27,7 @@
 #include <linux/mutex.h>
 #include <linux/capability.h>
 #include <linux/syscalls.h>
+#include <linux/memcontrol.h>
 
 #include <asm/pgtable.h>
 #include <asm/tlbflush.h>
@@ -506,9 +507,12 @@ unsigned int count_swap_pages(int type, int free)
  * just let do_wp_page work it out if a write is requested later - to
  * force COW, vm_page_prot omits write permission from any private vma.
  */
-static void unuse_pte(struct vm_area_struct *vma, pte_t *pte,
+static int unuse_pte(struct vm_area_struct *vma, pte_t *pte,
                unsigned long addr, swp_entry_t entry, struct page *page)
 {
+       if (mem_cgroup_charge(page, vma->vm_mm))
+               return -ENOMEM;
+
        inc_mm_counter(vma->vm_mm, anon_rss);
        get_page(page);
        set_pte_at(vma->vm_mm, addr, pte,
@@ -520,6 +524,7 @@ static void unuse_pte(struct vm_area_struct *vma, pte_t *pte,
         * immediately swapped out again after swapon.
         */
        activate_page(page);
+       return 1;
 }
 
 static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
@@ -529,7 +534,7 @@ static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
        pte_t swp_pte = swp_entry_to_pte(entry);
        pte_t *pte;
        spinlock_t *ptl;
-       int found = 0;
+       int ret = 0;
 
        pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl);
        do {
@@ -538,13 +543,12 @@ static int unuse_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
                 * Test inline before going to call unuse_pte.
                 */
                if (unlikely(pte_same(*pte, swp_pte))) {
-                       unuse_pte(vma, pte++, addr, entry, page);
-                       found = 1;
+                       ret = unuse_pte(vma, pte++, addr, entry, page);
                        break;
                }
        } while (pte++, addr += PAGE_SIZE, addr != end);
        pte_unmap_unlock(pte - 1, ptl);
-       return found;
+       return ret;
 }
 
 static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud,
@@ -553,14 +557,16 @@ static inline int unuse_pmd_range(struct vm_area_struct *vma, pud_t *pud,
 {
        pmd_t *pmd;
        unsigned long next;
+       int ret;
 
        pmd = pmd_offset(pud, addr);
        do {
                next = pmd_addr_end(addr, end);
                if (pmd_none_or_clear_bad(pmd))
                        continue;
-               if (unuse_pte_range(vma, pmd, addr, next, entry, page))
-                       return 1;
+               ret = unuse_pte_range(vma, pmd, addr, next, entry, page);
+               if (ret)
+                       return ret;
        } while (pmd++, addr = next, addr != end);
        return 0;
 }
@@ -571,14 +577,16 @@ static inline int unuse_pud_range(struct vm_area_struct *vma, pgd_t *pgd,
 {
        pud_t *pud;
        unsigned long next;
+       int ret;
 
        pud = pud_offset(pgd, addr);
        do {
                next = pud_addr_end(addr, end);
                if (pud_none_or_clear_bad(pud))
                        continue;
-               if (unuse_pmd_range(vma, pud, addr, next, entry, page))
-                       return 1;
+               ret = unuse_pmd_range(vma, pud, addr, next, entry, page);
+               if (ret)
+                       return ret;
        } while (pud++, addr = next, addr != end);
        return 0;
 }
@@ -588,6 +596,7 @@ static int unuse_vma(struct vm_area_struct *vma,
 {
        pgd_t *pgd;
        unsigned long addr, end, next;
+       int ret;
 
        if (page->mapping) {
                addr = page_address_in_vma(page, vma);
@@ -605,8 +614,9 @@ static int unuse_vma(struct vm_area_struct *vma,
                next = pgd_addr_end(addr, end);
                if (pgd_none_or_clear_bad(pgd))
                        continue;
-               if (unuse_pud_range(vma, pgd, addr, next, entry, page))
-                       return 1;
+               ret = unuse_pud_range(vma, pgd, addr, next, entry, page);
+               if (ret)
+                       return ret;
        } while (pgd++, addr = next, addr != end);
        return 0;
 }
@@ -615,6 +625,7 @@ static int unuse_mm(struct mm_struct *mm,
                                swp_entry_t entry, struct page *page)
 {
        struct vm_area_struct *vma;
+       int ret = 0;
 
        if (!down_read_trylock(&mm->mmap_sem)) {
                /*
@@ -627,15 +638,11 @@ static int unuse_mm(struct mm_struct *mm,
                lock_page(page);
        }
        for (vma = mm->mmap; vma; vma = vma->vm_next) {
-               if (vma->anon_vma && unuse_vma(vma, entry, page))
+               if (vma->anon_vma && (ret = unuse_vma(vma, entry, page)))
                        break;
        }
        up_read(&mm->mmap_sem);
-       /*
-        * Currently unuse_mm cannot fail, but leave error handling
-        * at call sites for now, since we change it from time to time.
-        */
-       return 0;
+       return (ret < 0)? ret: 0;
 }
 
 /*