1 /* memcontrol.c - Memory Controller
3 * Copyright IBM Corporation, 2007
4 * Author Balbir Singh <balbir@linux.vnet.ibm.com>
6 * Copyright 2007 OpenVZ SWsoft Inc
7 * Author: Pavel Emelianov <xemul@openvz.org>
10 * Copyright (C) 2009 Nokia Corporation
11 * Author: Kirill A. Shutemov
13 * This program is free software; you can redistribute it and/or modify
14 * it under the terms of the GNU General Public License as published by
15 * the Free Software Foundation; either version 2 of the License, or
16 * (at your option) any later version.
18 * This program is distributed in the hope that it will be useful,
19 * but WITHOUT ANY WARRANTY; without even the implied warranty of
20 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
21 * GNU General Public License for more details.
24 #include <linux/res_counter.h>
25 #include <linux/memcontrol.h>
26 #include <linux/cgroup.h>
28 #include <linux/hugetlb.h>
29 #include <linux/pagemap.h>
30 #include <linux/smp.h>
31 #include <linux/page-flags.h>
32 #include <linux/backing-dev.h>
33 #include <linux/bit_spinlock.h>
34 #include <linux/rcupdate.h>
35 #include <linux/limits.h>
36 #include <linux/mutex.h>
37 #include <linux/rbtree.h>
38 #include <linux/slab.h>
39 #include <linux/swap.h>
40 #include <linux/swapops.h>
41 #include <linux/spinlock.h>
42 #include <linux/eventfd.h>
43 #include <linux/sort.h>
45 #include <linux/seq_file.h>
46 #include <linux/vmalloc.h>
47 #include <linux/mm_inline.h>
48 #include <linux/page_cgroup.h>
49 #include <linux/cpu.h>
50 #include <linux/oom.h>
53 #include <asm/uaccess.h>
55 #include <trace/events/vmscan.h>
57 struct cgroup_subsys mem_cgroup_subsys __read_mostly
;
58 #define MEM_CGROUP_RECLAIM_RETRIES 5
59 struct mem_cgroup
*root_mem_cgroup __read_mostly
;
61 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
62 /* Turned on only when memory cgroup is enabled && really_do_swap_account = 1 */
63 int do_swap_account __read_mostly
;
64 static int really_do_swap_account __initdata
= 1; /* for remember boot option*/
66 #define do_swap_account (0)
70 * Per memcg event counter is incremented at every pagein/pageout. This counter
71 * is used for trigger some periodic events. This is straightforward and better
72 * than using jiffies etc. to handle periodic memcg event.
74 * These values will be used as !((event) & ((1 <<(thresh)) - 1))
76 #define THRESHOLDS_EVENTS_THRESH (7) /* once in 128 */
77 #define SOFTLIMIT_EVENTS_THRESH (10) /* once in 1024 */
80 * Statistics for memory cgroup.
82 enum mem_cgroup_stat_index
{
84 * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss.
86 MEM_CGROUP_STAT_CACHE
, /* # of pages charged as cache */
87 MEM_CGROUP_STAT_RSS
, /* # of pages charged as anon rss */
88 MEM_CGROUP_STAT_FILE_MAPPED
, /* # of pages charged as file rss */
89 MEM_CGROUP_STAT_PGPGIN_COUNT
, /* # of pages paged in */
90 MEM_CGROUP_STAT_PGPGOUT_COUNT
, /* # of pages paged out */
91 MEM_CGROUP_STAT_SWAPOUT
, /* # of pages, swapped out */
92 MEM_CGROUP_EVENTS
, /* incremented at every pagein/pageout */
94 MEM_CGROUP_STAT_NSTATS
,
97 struct mem_cgroup_stat_cpu
{
98 s64 count
[MEM_CGROUP_STAT_NSTATS
];
102 * per-zone information in memory controller.
104 struct mem_cgroup_per_zone
{
106 * spin_lock to protect the per cgroup LRU
108 struct list_head lists
[NR_LRU_LISTS
];
109 unsigned long count
[NR_LRU_LISTS
];
111 struct zone_reclaim_stat reclaim_stat
;
112 struct rb_node tree_node
; /* RB tree node */
113 unsigned long long usage_in_excess
;/* Set to the value by which */
114 /* the soft limit is exceeded*/
116 struct mem_cgroup
*mem
; /* Back pointer, we cannot */
117 /* use container_of */
119 /* Macro for accessing counter */
120 #define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)])
122 struct mem_cgroup_per_node
{
123 struct mem_cgroup_per_zone zoneinfo
[MAX_NR_ZONES
];
126 struct mem_cgroup_lru_info
{
127 struct mem_cgroup_per_node
*nodeinfo
[MAX_NUMNODES
];
131 * Cgroups above their limits are maintained in a RB-Tree, independent of
132 * their hierarchy representation
135 struct mem_cgroup_tree_per_zone
{
136 struct rb_root rb_root
;
140 struct mem_cgroup_tree_per_node
{
141 struct mem_cgroup_tree_per_zone rb_tree_per_zone
[MAX_NR_ZONES
];
144 struct mem_cgroup_tree
{
145 struct mem_cgroup_tree_per_node
*rb_tree_per_node
[MAX_NUMNODES
];
148 static struct mem_cgroup_tree soft_limit_tree __read_mostly
;
150 struct mem_cgroup_threshold
{
151 struct eventfd_ctx
*eventfd
;
156 struct mem_cgroup_threshold_ary
{
157 /* An array index points to threshold just below usage. */
158 int current_threshold
;
159 /* Size of entries[] */
161 /* Array of thresholds */
162 struct mem_cgroup_threshold entries
[0];
165 struct mem_cgroup_thresholds
{
166 /* Primary thresholds array */
167 struct mem_cgroup_threshold_ary
*primary
;
169 * Spare threshold array.
170 * This is needed to make mem_cgroup_unregister_event() "never fail".
171 * It must be able to store at least primary->size - 1 entries.
173 struct mem_cgroup_threshold_ary
*spare
;
177 struct mem_cgroup_eventfd_list
{
178 struct list_head list
;
179 struct eventfd_ctx
*eventfd
;
182 static void mem_cgroup_threshold(struct mem_cgroup
*mem
);
183 static void mem_cgroup_oom_notify(struct mem_cgroup
*mem
);
186 * The memory controller data structure. The memory controller controls both
187 * page cache and RSS per cgroup. We would eventually like to provide
188 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
189 * to help the administrator determine what knobs to tune.
191 * TODO: Add a water mark for the memory controller. Reclaim will begin when
192 * we hit the water mark. May be even add a low water mark, such that
193 * no reclaim occurs from a cgroup at it's low water mark, this is
194 * a feature that will be implemented much later in the future.
197 struct cgroup_subsys_state css
;
199 * the counter to account for memory usage
201 struct res_counter res
;
203 * the counter to account for mem+swap usage.
205 struct res_counter memsw
;
207 * Per cgroup active and inactive list, similar to the
208 * per zone LRU lists.
210 struct mem_cgroup_lru_info info
;
213 protect against reclaim related member.
215 spinlock_t reclaim_param_lock
;
218 * While reclaiming in a hierarchy, we cache the last child we
221 int last_scanned_child
;
223 * Should the accounting and control be hierarchical, per subtree?
229 unsigned int swappiness
;
230 /* OOM-Killer disable */
231 int oom_kill_disable
;
233 /* set when res.limit == memsw.limit */
234 bool memsw_is_minimum
;
236 /* protect arrays of thresholds */
237 struct mutex thresholds_lock
;
239 /* thresholds for memory usage. RCU-protected */
240 struct mem_cgroup_thresholds thresholds
;
242 /* thresholds for mem+swap usage. RCU-protected */
243 struct mem_cgroup_thresholds memsw_thresholds
;
245 /* For oom notifier event fd */
246 struct list_head oom_notify
;
249 * Should we move charges of a task when a task is moved into this
250 * mem_cgroup ? And what type of charges should we move ?
252 unsigned long move_charge_at_immigrate
;
256 struct mem_cgroup_stat_cpu
*stat
;
259 /* Stuffs for move charges at task migration. */
261 * Types of charges to be moved. "move_charge_at_immitgrate" is treated as a
262 * left-shifted bitmap of these types.
265 MOVE_CHARGE_TYPE_ANON
, /* private anonymous page and swap of it */
266 MOVE_CHARGE_TYPE_FILE
, /* file page(including tmpfs) and swap of it */
270 /* "mc" and its members are protected by cgroup_mutex */
271 static struct move_charge_struct
{
272 spinlock_t lock
; /* for from, to */
273 struct mem_cgroup
*from
;
274 struct mem_cgroup
*to
;
275 unsigned long precharge
;
276 unsigned long moved_charge
;
277 unsigned long moved_swap
;
278 struct task_struct
*moving_task
; /* a task moving charges */
279 struct mm_struct
*mm
;
280 wait_queue_head_t waitq
; /* a waitq for other context */
282 .lock
= __SPIN_LOCK_UNLOCKED(mc
.lock
),
283 .waitq
= __WAIT_QUEUE_HEAD_INITIALIZER(mc
.waitq
),
286 static bool move_anon(void)
288 return test_bit(MOVE_CHARGE_TYPE_ANON
,
289 &mc
.to
->move_charge_at_immigrate
);
292 static bool move_file(void)
294 return test_bit(MOVE_CHARGE_TYPE_FILE
,
295 &mc
.to
->move_charge_at_immigrate
);
299 * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft
300 * limit reclaim to prevent infinite loops, if they ever occur.
302 #define MEM_CGROUP_MAX_RECLAIM_LOOPS (100)
303 #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS (2)
306 MEM_CGROUP_CHARGE_TYPE_CACHE
= 0,
307 MEM_CGROUP_CHARGE_TYPE_MAPPED
,
308 MEM_CGROUP_CHARGE_TYPE_SHMEM
, /* used by page migration of shmem */
309 MEM_CGROUP_CHARGE_TYPE_FORCE
, /* used by force_empty */
310 MEM_CGROUP_CHARGE_TYPE_SWAPOUT
, /* for accounting swapcache */
311 MEM_CGROUP_CHARGE_TYPE_DROP
, /* a page was unused swap cache */
315 /* only for here (for easy reading.) */
316 #define PCGF_CACHE (1UL << PCG_CACHE)
317 #define PCGF_USED (1UL << PCG_USED)
318 #define PCGF_LOCK (1UL << PCG_LOCK)
319 /* Not used, but added here for completeness */
320 #define PCGF_ACCT (1UL << PCG_ACCT)
322 /* for encoding cft->private value on file */
325 #define _OOM_TYPE (2)
326 #define MEMFILE_PRIVATE(x, val) (((x) << 16) | (val))
327 #define MEMFILE_TYPE(val) (((val) >> 16) & 0xffff)
328 #define MEMFILE_ATTR(val) ((val) & 0xffff)
329 /* Used for OOM nofiier */
330 #define OOM_CONTROL (0)
333 * Reclaim flags for mem_cgroup_hierarchical_reclaim
335 #define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0
336 #define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT)
337 #define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1
338 #define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT)
339 #define MEM_CGROUP_RECLAIM_SOFT_BIT 0x2
340 #define MEM_CGROUP_RECLAIM_SOFT (1 << MEM_CGROUP_RECLAIM_SOFT_BIT)
342 static void mem_cgroup_get(struct mem_cgroup
*mem
);
343 static void mem_cgroup_put(struct mem_cgroup
*mem
);
344 static struct mem_cgroup
*parent_mem_cgroup(struct mem_cgroup
*mem
);
345 static void drain_all_stock_async(void);
347 static struct mem_cgroup_per_zone
*
348 mem_cgroup_zoneinfo(struct mem_cgroup
*mem
, int nid
, int zid
)
350 return &mem
->info
.nodeinfo
[nid
]->zoneinfo
[zid
];
353 struct cgroup_subsys_state
*mem_cgroup_css(struct mem_cgroup
*mem
)
358 static struct mem_cgroup_per_zone
*
359 page_cgroup_zoneinfo(struct page_cgroup
*pc
)
361 struct mem_cgroup
*mem
= pc
->mem_cgroup
;
362 int nid
= page_cgroup_nid(pc
);
363 int zid
= page_cgroup_zid(pc
);
368 return mem_cgroup_zoneinfo(mem
, nid
, zid
);
371 static struct mem_cgroup_tree_per_zone
*
372 soft_limit_tree_node_zone(int nid
, int zid
)
374 return &soft_limit_tree
.rb_tree_per_node
[nid
]->rb_tree_per_zone
[zid
];
377 static struct mem_cgroup_tree_per_zone
*
378 soft_limit_tree_from_page(struct page
*page
)
380 int nid
= page_to_nid(page
);
381 int zid
= page_zonenum(page
);
383 return &soft_limit_tree
.rb_tree_per_node
[nid
]->rb_tree_per_zone
[zid
];
387 __mem_cgroup_insert_exceeded(struct mem_cgroup
*mem
,
388 struct mem_cgroup_per_zone
*mz
,
389 struct mem_cgroup_tree_per_zone
*mctz
,
390 unsigned long long new_usage_in_excess
)
392 struct rb_node
**p
= &mctz
->rb_root
.rb_node
;
393 struct rb_node
*parent
= NULL
;
394 struct mem_cgroup_per_zone
*mz_node
;
399 mz
->usage_in_excess
= new_usage_in_excess
;
400 if (!mz
->usage_in_excess
)
404 mz_node
= rb_entry(parent
, struct mem_cgroup_per_zone
,
406 if (mz
->usage_in_excess
< mz_node
->usage_in_excess
)
409 * We can't avoid mem cgroups that are over their soft
410 * limit by the same amount
412 else if (mz
->usage_in_excess
>= mz_node
->usage_in_excess
)
415 rb_link_node(&mz
->tree_node
, parent
, p
);
416 rb_insert_color(&mz
->tree_node
, &mctz
->rb_root
);
421 __mem_cgroup_remove_exceeded(struct mem_cgroup
*mem
,
422 struct mem_cgroup_per_zone
*mz
,
423 struct mem_cgroup_tree_per_zone
*mctz
)
427 rb_erase(&mz
->tree_node
, &mctz
->rb_root
);
432 mem_cgroup_remove_exceeded(struct mem_cgroup
*mem
,
433 struct mem_cgroup_per_zone
*mz
,
434 struct mem_cgroup_tree_per_zone
*mctz
)
436 spin_lock(&mctz
->lock
);
437 __mem_cgroup_remove_exceeded(mem
, mz
, mctz
);
438 spin_unlock(&mctz
->lock
);
442 static void mem_cgroup_update_tree(struct mem_cgroup
*mem
, struct page
*page
)
444 unsigned long long excess
;
445 struct mem_cgroup_per_zone
*mz
;
446 struct mem_cgroup_tree_per_zone
*mctz
;
447 int nid
= page_to_nid(page
);
448 int zid
= page_zonenum(page
);
449 mctz
= soft_limit_tree_from_page(page
);
452 * Necessary to update all ancestors when hierarchy is used.
453 * because their event counter is not touched.
455 for (; mem
; mem
= parent_mem_cgroup(mem
)) {
456 mz
= mem_cgroup_zoneinfo(mem
, nid
, zid
);
457 excess
= res_counter_soft_limit_excess(&mem
->res
);
459 * We have to update the tree if mz is on RB-tree or
460 * mem is over its softlimit.
462 if (excess
|| mz
->on_tree
) {
463 spin_lock(&mctz
->lock
);
464 /* if on-tree, remove it */
466 __mem_cgroup_remove_exceeded(mem
, mz
, mctz
);
468 * Insert again. mz->usage_in_excess will be updated.
469 * If excess is 0, no tree ops.
471 __mem_cgroup_insert_exceeded(mem
, mz
, mctz
, excess
);
472 spin_unlock(&mctz
->lock
);
477 static void mem_cgroup_remove_from_trees(struct mem_cgroup
*mem
)
480 struct mem_cgroup_per_zone
*mz
;
481 struct mem_cgroup_tree_per_zone
*mctz
;
483 for_each_node_state(node
, N_POSSIBLE
) {
484 for (zone
= 0; zone
< MAX_NR_ZONES
; zone
++) {
485 mz
= mem_cgroup_zoneinfo(mem
, node
, zone
);
486 mctz
= soft_limit_tree_node_zone(node
, zone
);
487 mem_cgroup_remove_exceeded(mem
, mz
, mctz
);
492 static inline unsigned long mem_cgroup_get_excess(struct mem_cgroup
*mem
)
494 return res_counter_soft_limit_excess(&mem
->res
) >> PAGE_SHIFT
;
497 static struct mem_cgroup_per_zone
*
498 __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone
*mctz
)
500 struct rb_node
*rightmost
= NULL
;
501 struct mem_cgroup_per_zone
*mz
;
505 rightmost
= rb_last(&mctz
->rb_root
);
507 goto done
; /* Nothing to reclaim from */
509 mz
= rb_entry(rightmost
, struct mem_cgroup_per_zone
, tree_node
);
511 * Remove the node now but someone else can add it back,
512 * we will to add it back at the end of reclaim to its correct
513 * position in the tree.
515 __mem_cgroup_remove_exceeded(mz
->mem
, mz
, mctz
);
516 if (!res_counter_soft_limit_excess(&mz
->mem
->res
) ||
517 !css_tryget(&mz
->mem
->css
))
523 static struct mem_cgroup_per_zone
*
524 mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone
*mctz
)
526 struct mem_cgroup_per_zone
*mz
;
528 spin_lock(&mctz
->lock
);
529 mz
= __mem_cgroup_largest_soft_limit_node(mctz
);
530 spin_unlock(&mctz
->lock
);
534 static s64
mem_cgroup_read_stat(struct mem_cgroup
*mem
,
535 enum mem_cgroup_stat_index idx
)
540 for_each_possible_cpu(cpu
)
541 val
+= per_cpu(mem
->stat
->count
[idx
], cpu
);
545 static s64
mem_cgroup_local_usage(struct mem_cgroup
*mem
)
549 ret
= mem_cgroup_read_stat(mem
, MEM_CGROUP_STAT_RSS
);
550 ret
+= mem_cgroup_read_stat(mem
, MEM_CGROUP_STAT_CACHE
);
554 static void mem_cgroup_swap_statistics(struct mem_cgroup
*mem
,
557 int val
= (charge
) ? 1 : -1;
558 this_cpu_add(mem
->stat
->count
[MEM_CGROUP_STAT_SWAPOUT
], val
);
561 static void mem_cgroup_charge_statistics(struct mem_cgroup
*mem
,
562 struct page_cgroup
*pc
,
565 int val
= (charge
) ? 1 : -1;
569 if (PageCgroupCache(pc
))
570 __this_cpu_add(mem
->stat
->count
[MEM_CGROUP_STAT_CACHE
], val
);
572 __this_cpu_add(mem
->stat
->count
[MEM_CGROUP_STAT_RSS
], val
);
575 __this_cpu_inc(mem
->stat
->count
[MEM_CGROUP_STAT_PGPGIN_COUNT
]);
577 __this_cpu_inc(mem
->stat
->count
[MEM_CGROUP_STAT_PGPGOUT_COUNT
]);
578 __this_cpu_inc(mem
->stat
->count
[MEM_CGROUP_EVENTS
]);
583 static unsigned long mem_cgroup_get_local_zonestat(struct mem_cgroup
*mem
,
587 struct mem_cgroup_per_zone
*mz
;
590 for_each_online_node(nid
)
591 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++) {
592 mz
= mem_cgroup_zoneinfo(mem
, nid
, zid
);
593 total
+= MEM_CGROUP_ZSTAT(mz
, idx
);
598 static bool __memcg_event_check(struct mem_cgroup
*mem
, int event_mask_shift
)
602 val
= this_cpu_read(mem
->stat
->count
[MEM_CGROUP_EVENTS
]);
604 return !(val
& ((1 << event_mask_shift
) - 1));
608 * Check events in order.
611 static void memcg_check_events(struct mem_cgroup
*mem
, struct page
*page
)
613 /* threshold event is triggered in finer grain than soft limit */
614 if (unlikely(__memcg_event_check(mem
, THRESHOLDS_EVENTS_THRESH
))) {
615 mem_cgroup_threshold(mem
);
616 if (unlikely(__memcg_event_check(mem
, SOFTLIMIT_EVENTS_THRESH
)))
617 mem_cgroup_update_tree(mem
, page
);
621 static struct mem_cgroup
*mem_cgroup_from_cont(struct cgroup
*cont
)
623 return container_of(cgroup_subsys_state(cont
,
624 mem_cgroup_subsys_id
), struct mem_cgroup
,
628 struct mem_cgroup
*mem_cgroup_from_task(struct task_struct
*p
)
631 * mm_update_next_owner() may clear mm->owner to NULL
632 * if it races with swapoff, page migration, etc.
633 * So this can be called with p == NULL.
638 return container_of(task_subsys_state(p
, mem_cgroup_subsys_id
),
639 struct mem_cgroup
, css
);
642 static struct mem_cgroup
*try_get_mem_cgroup_from_mm(struct mm_struct
*mm
)
644 struct mem_cgroup
*mem
= NULL
;
649 * Because we have no locks, mm->owner's may be being moved to other
650 * cgroup. We use css_tryget() here even if this looks
651 * pessimistic (rather than adding locks here).
655 mem
= mem_cgroup_from_task(rcu_dereference(mm
->owner
));
658 } while (!css_tryget(&mem
->css
));
664 * Call callback function against all cgroup under hierarchy tree.
666 static int mem_cgroup_walk_tree(struct mem_cgroup
*root
, void *data
,
667 int (*func
)(struct mem_cgroup
*, void *))
669 int found
, ret
, nextid
;
670 struct cgroup_subsys_state
*css
;
671 struct mem_cgroup
*mem
;
673 if (!root
->use_hierarchy
)
674 return (*func
)(root
, data
);
682 css
= css_get_next(&mem_cgroup_subsys
, nextid
, &root
->css
,
684 if (css
&& css_tryget(css
))
685 mem
= container_of(css
, struct mem_cgroup
, css
);
689 ret
= (*func
)(mem
, data
);
693 } while (!ret
&& css
);
698 static inline bool mem_cgroup_is_root(struct mem_cgroup
*mem
)
700 return (mem
== root_mem_cgroup
);
704 * Following LRU functions are allowed to be used without PCG_LOCK.
705 * Operations are called by routine of global LRU independently from memcg.
706 * What we have to take care of here is validness of pc->mem_cgroup.
708 * Changes to pc->mem_cgroup happens when
711 * In typical case, "charge" is done before add-to-lru. Exception is SwapCache.
712 * It is added to LRU before charge.
713 * If PCG_USED bit is not set, page_cgroup is not added to this private LRU.
714 * When moving account, the page is not on LRU. It's isolated.
717 void mem_cgroup_del_lru_list(struct page
*page
, enum lru_list lru
)
719 struct page_cgroup
*pc
;
720 struct mem_cgroup_per_zone
*mz
;
722 if (mem_cgroup_disabled())
724 pc
= lookup_page_cgroup(page
);
725 /* can happen while we handle swapcache. */
726 if (!TestClearPageCgroupAcctLRU(pc
))
728 VM_BUG_ON(!pc
->mem_cgroup
);
730 * We don't check PCG_USED bit. It's cleared when the "page" is finally
731 * removed from global LRU.
733 mz
= page_cgroup_zoneinfo(pc
);
734 MEM_CGROUP_ZSTAT(mz
, lru
) -= 1;
735 if (mem_cgroup_is_root(pc
->mem_cgroup
))
737 VM_BUG_ON(list_empty(&pc
->lru
));
738 list_del_init(&pc
->lru
);
742 void mem_cgroup_del_lru(struct page
*page
)
744 mem_cgroup_del_lru_list(page
, page_lru(page
));
747 void mem_cgroup_rotate_lru_list(struct page
*page
, enum lru_list lru
)
749 struct mem_cgroup_per_zone
*mz
;
750 struct page_cgroup
*pc
;
752 if (mem_cgroup_disabled())
755 pc
= lookup_page_cgroup(page
);
757 * Used bit is set without atomic ops but after smp_wmb().
758 * For making pc->mem_cgroup visible, insert smp_rmb() here.
761 /* unused or root page is not rotated. */
762 if (!PageCgroupUsed(pc
) || mem_cgroup_is_root(pc
->mem_cgroup
))
764 mz
= page_cgroup_zoneinfo(pc
);
765 list_move(&pc
->lru
, &mz
->lists
[lru
]);
768 void mem_cgroup_add_lru_list(struct page
*page
, enum lru_list lru
)
770 struct page_cgroup
*pc
;
771 struct mem_cgroup_per_zone
*mz
;
773 if (mem_cgroup_disabled())
775 pc
= lookup_page_cgroup(page
);
776 VM_BUG_ON(PageCgroupAcctLRU(pc
));
778 * Used bit is set without atomic ops but after smp_wmb().
779 * For making pc->mem_cgroup visible, insert smp_rmb() here.
782 if (!PageCgroupUsed(pc
))
785 mz
= page_cgroup_zoneinfo(pc
);
786 MEM_CGROUP_ZSTAT(mz
, lru
) += 1;
787 SetPageCgroupAcctLRU(pc
);
788 if (mem_cgroup_is_root(pc
->mem_cgroup
))
790 list_add(&pc
->lru
, &mz
->lists
[lru
]);
794 * At handling SwapCache, pc->mem_cgroup may be changed while it's linked to
795 * lru because the page may.be reused after it's fully uncharged (because of
796 * SwapCache behavior).To handle that, unlink page_cgroup from LRU when charge
797 * it again. This function is only used to charge SwapCache. It's done under
798 * lock_page and expected that zone->lru_lock is never held.
800 static void mem_cgroup_lru_del_before_commit_swapcache(struct page
*page
)
803 struct zone
*zone
= page_zone(page
);
804 struct page_cgroup
*pc
= lookup_page_cgroup(page
);
806 spin_lock_irqsave(&zone
->lru_lock
, flags
);
808 * Forget old LRU when this page_cgroup is *not* used. This Used bit
809 * is guarded by lock_page() because the page is SwapCache.
811 if (!PageCgroupUsed(pc
))
812 mem_cgroup_del_lru_list(page
, page_lru(page
));
813 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
816 static void mem_cgroup_lru_add_after_commit_swapcache(struct page
*page
)
819 struct zone
*zone
= page_zone(page
);
820 struct page_cgroup
*pc
= lookup_page_cgroup(page
);
822 spin_lock_irqsave(&zone
->lru_lock
, flags
);
823 /* link when the page is linked to LRU but page_cgroup isn't */
824 if (PageLRU(page
) && !PageCgroupAcctLRU(pc
))
825 mem_cgroup_add_lru_list(page
, page_lru(page
));
826 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
830 void mem_cgroup_move_lists(struct page
*page
,
831 enum lru_list from
, enum lru_list to
)
833 if (mem_cgroup_disabled())
835 mem_cgroup_del_lru_list(page
, from
);
836 mem_cgroup_add_lru_list(page
, to
);
839 int task_in_mem_cgroup(struct task_struct
*task
, const struct mem_cgroup
*mem
)
842 struct mem_cgroup
*curr
= NULL
;
843 struct task_struct
*p
;
845 p
= find_lock_task_mm(task
);
848 curr
= try_get_mem_cgroup_from_mm(p
->mm
);
853 * We should check use_hierarchy of "mem" not "curr". Because checking
854 * use_hierarchy of "curr" here make this function true if hierarchy is
855 * enabled in "curr" and "curr" is a child of "mem" in *cgroup*
856 * hierarchy(even if use_hierarchy is disabled in "mem").
858 if (mem
->use_hierarchy
)
859 ret
= css_is_ancestor(&curr
->css
, &mem
->css
);
866 static int calc_inactive_ratio(struct mem_cgroup
*memcg
, unsigned long *present_pages
)
868 unsigned long active
;
869 unsigned long inactive
;
871 unsigned long inactive_ratio
;
873 inactive
= mem_cgroup_get_local_zonestat(memcg
, LRU_INACTIVE_ANON
);
874 active
= mem_cgroup_get_local_zonestat(memcg
, LRU_ACTIVE_ANON
);
876 gb
= (inactive
+ active
) >> (30 - PAGE_SHIFT
);
878 inactive_ratio
= int_sqrt(10 * gb
);
883 present_pages
[0] = inactive
;
884 present_pages
[1] = active
;
887 return inactive_ratio
;
890 int mem_cgroup_inactive_anon_is_low(struct mem_cgroup
*memcg
)
892 unsigned long active
;
893 unsigned long inactive
;
894 unsigned long present_pages
[2];
895 unsigned long inactive_ratio
;
897 inactive_ratio
= calc_inactive_ratio(memcg
, present_pages
);
899 inactive
= present_pages
[0];
900 active
= present_pages
[1];
902 if (inactive
* inactive_ratio
< active
)
908 int mem_cgroup_inactive_file_is_low(struct mem_cgroup
*memcg
)
910 unsigned long active
;
911 unsigned long inactive
;
913 inactive
= mem_cgroup_get_local_zonestat(memcg
, LRU_INACTIVE_FILE
);
914 active
= mem_cgroup_get_local_zonestat(memcg
, LRU_ACTIVE_FILE
);
916 return (active
> inactive
);
919 unsigned long mem_cgroup_zone_nr_pages(struct mem_cgroup
*memcg
,
923 int nid
= zone_to_nid(zone
);
924 int zid
= zone_idx(zone
);
925 struct mem_cgroup_per_zone
*mz
= mem_cgroup_zoneinfo(memcg
, nid
, zid
);
927 return MEM_CGROUP_ZSTAT(mz
, lru
);
930 struct zone_reclaim_stat
*mem_cgroup_get_reclaim_stat(struct mem_cgroup
*memcg
,
933 int nid
= zone_to_nid(zone
);
934 int zid
= zone_idx(zone
);
935 struct mem_cgroup_per_zone
*mz
= mem_cgroup_zoneinfo(memcg
, nid
, zid
);
937 return &mz
->reclaim_stat
;
940 struct zone_reclaim_stat
*
941 mem_cgroup_get_reclaim_stat_from_page(struct page
*page
)
943 struct page_cgroup
*pc
;
944 struct mem_cgroup_per_zone
*mz
;
946 if (mem_cgroup_disabled())
949 pc
= lookup_page_cgroup(page
);
951 * Used bit is set without atomic ops but after smp_wmb().
952 * For making pc->mem_cgroup visible, insert smp_rmb() here.
955 if (!PageCgroupUsed(pc
))
958 mz
= page_cgroup_zoneinfo(pc
);
962 return &mz
->reclaim_stat
;
965 unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan
,
966 struct list_head
*dst
,
967 unsigned long *scanned
, int order
,
968 int mode
, struct zone
*z
,
969 struct mem_cgroup
*mem_cont
,
970 int active
, int file
)
972 unsigned long nr_taken
= 0;
976 struct list_head
*src
;
977 struct page_cgroup
*pc
, *tmp
;
978 int nid
= zone_to_nid(z
);
979 int zid
= zone_idx(z
);
980 struct mem_cgroup_per_zone
*mz
;
981 int lru
= LRU_FILE
* file
+ active
;
985 mz
= mem_cgroup_zoneinfo(mem_cont
, nid
, zid
);
986 src
= &mz
->lists
[lru
];
989 list_for_each_entry_safe_reverse(pc
, tmp
, src
, lru
) {
990 if (scan
>= nr_to_scan
)
994 if (unlikely(!PageCgroupUsed(pc
)))
996 if (unlikely(!PageLRU(page
)))
1000 ret
= __isolate_lru_page(page
, mode
, file
);
1003 list_move(&page
->lru
, dst
);
1004 mem_cgroup_del_lru(page
);
1008 /* we don't affect global LRU but rotate in our LRU */
1009 mem_cgroup_rotate_lru_list(page
, page_lru(page
));
1018 trace_mm_vmscan_memcg_isolate(0, nr_to_scan
, scan
, nr_taken
,
1024 #define mem_cgroup_from_res_counter(counter, member) \
1025 container_of(counter, struct mem_cgroup, member)
1027 static bool mem_cgroup_check_under_limit(struct mem_cgroup
*mem
)
1029 if (do_swap_account
) {
1030 if (res_counter_check_under_limit(&mem
->res
) &&
1031 res_counter_check_under_limit(&mem
->memsw
))
1034 if (res_counter_check_under_limit(&mem
->res
))
1039 static unsigned int get_swappiness(struct mem_cgroup
*memcg
)
1041 struct cgroup
*cgrp
= memcg
->css
.cgroup
;
1042 unsigned int swappiness
;
1045 if (cgrp
->parent
== NULL
)
1046 return vm_swappiness
;
1048 spin_lock(&memcg
->reclaim_param_lock
);
1049 swappiness
= memcg
->swappiness
;
1050 spin_unlock(&memcg
->reclaim_param_lock
);
1055 /* A routine for testing mem is not under move_account */
1057 static bool mem_cgroup_under_move(struct mem_cgroup
*mem
)
1059 struct mem_cgroup
*from
;
1060 struct mem_cgroup
*to
;
1063 * Unlike task_move routines, we access mc.to, mc.from not under
1064 * mutual exclusion by cgroup_mutex. Here, we take spinlock instead.
1066 spin_lock(&mc
.lock
);
1071 if (from
== mem
|| to
== mem
1072 || (mem
->use_hierarchy
&& css_is_ancestor(&from
->css
, &mem
->css
))
1073 || (mem
->use_hierarchy
&& css_is_ancestor(&to
->css
, &mem
->css
)))
1076 spin_unlock(&mc
.lock
);
1080 static bool mem_cgroup_wait_acct_move(struct mem_cgroup
*mem
)
1082 if (mc
.moving_task
&& current
!= mc
.moving_task
) {
1083 if (mem_cgroup_under_move(mem
)) {
1085 prepare_to_wait(&mc
.waitq
, &wait
, TASK_INTERRUPTIBLE
);
1086 /* moving charge context might have finished. */
1089 finish_wait(&mc
.waitq
, &wait
);
1096 static int mem_cgroup_count_children_cb(struct mem_cgroup
*mem
, void *data
)
1104 * mem_cgroup_print_oom_info: Called from OOM with tasklist_lock held in read mode.
1105 * @memcg: The memory cgroup that went over limit
1106 * @p: Task that is going to be killed
1108 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1111 void mem_cgroup_print_oom_info(struct mem_cgroup
*memcg
, struct task_struct
*p
)
1113 struct cgroup
*task_cgrp
;
1114 struct cgroup
*mem_cgrp
;
1116 * Need a buffer in BSS, can't rely on allocations. The code relies
1117 * on the assumption that OOM is serialized for memory controller.
1118 * If this assumption is broken, revisit this code.
1120 static char memcg_name
[PATH_MAX
];
1129 mem_cgrp
= memcg
->css
.cgroup
;
1130 task_cgrp
= task_cgroup(p
, mem_cgroup_subsys_id
);
1132 ret
= cgroup_path(task_cgrp
, memcg_name
, PATH_MAX
);
1135 * Unfortunately, we are unable to convert to a useful name
1136 * But we'll still print out the usage information
1143 printk(KERN_INFO
"Task in %s killed", memcg_name
);
1146 ret
= cgroup_path(mem_cgrp
, memcg_name
, PATH_MAX
);
1154 * Continues from above, so we don't need an KERN_ level
1156 printk(KERN_CONT
" as a result of limit of %s\n", memcg_name
);
1159 printk(KERN_INFO
"memory: usage %llukB, limit %llukB, failcnt %llu\n",
1160 res_counter_read_u64(&memcg
->res
, RES_USAGE
) >> 10,
1161 res_counter_read_u64(&memcg
->res
, RES_LIMIT
) >> 10,
1162 res_counter_read_u64(&memcg
->res
, RES_FAILCNT
));
1163 printk(KERN_INFO
"memory+swap: usage %llukB, limit %llukB, "
1165 res_counter_read_u64(&memcg
->memsw
, RES_USAGE
) >> 10,
1166 res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
) >> 10,
1167 res_counter_read_u64(&memcg
->memsw
, RES_FAILCNT
));
1171 * This function returns the number of memcg under hierarchy tree. Returns
1172 * 1(self count) if no children.
1174 static int mem_cgroup_count_children(struct mem_cgroup
*mem
)
1177 mem_cgroup_walk_tree(mem
, &num
, mem_cgroup_count_children_cb
);
1182 * Return the memory (and swap, if configured) limit for a memcg.
1184 u64
mem_cgroup_get_limit(struct mem_cgroup
*memcg
)
1189 limit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
) +
1191 memsw
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
1193 * If memsw is finite and limits the amount of swap space available
1194 * to this memcg, return that limit.
1196 return min(limit
, memsw
);
1200 * Visit the first child (need not be the first child as per the ordering
1201 * of the cgroup list, since we track last_scanned_child) of @mem and use
1202 * that to reclaim free pages from.
1204 static struct mem_cgroup
*
1205 mem_cgroup_select_victim(struct mem_cgroup
*root_mem
)
1207 struct mem_cgroup
*ret
= NULL
;
1208 struct cgroup_subsys_state
*css
;
1211 if (!root_mem
->use_hierarchy
) {
1212 css_get(&root_mem
->css
);
1218 nextid
= root_mem
->last_scanned_child
+ 1;
1219 css
= css_get_next(&mem_cgroup_subsys
, nextid
, &root_mem
->css
,
1221 if (css
&& css_tryget(css
))
1222 ret
= container_of(css
, struct mem_cgroup
, css
);
1225 /* Updates scanning parameter */
1226 spin_lock(&root_mem
->reclaim_param_lock
);
1228 /* this means start scan from ID:1 */
1229 root_mem
->last_scanned_child
= 0;
1231 root_mem
->last_scanned_child
= found
;
1232 spin_unlock(&root_mem
->reclaim_param_lock
);
1239 * Scan the hierarchy if needed to reclaim memory. We remember the last child
1240 * we reclaimed from, so that we don't end up penalizing one child extensively
1241 * based on its position in the children list.
1243 * root_mem is the original ancestor that we've been reclaim from.
1245 * We give up and return to the caller when we visit root_mem twice.
1246 * (other groups can be removed while we're walking....)
1248 * If shrink==true, for avoiding to free too much, this returns immedieately.
1250 static int mem_cgroup_hierarchical_reclaim(struct mem_cgroup
*root_mem
,
1253 unsigned long reclaim_options
)
1255 struct mem_cgroup
*victim
;
1258 bool noswap
= reclaim_options
& MEM_CGROUP_RECLAIM_NOSWAP
;
1259 bool shrink
= reclaim_options
& MEM_CGROUP_RECLAIM_SHRINK
;
1260 bool check_soft
= reclaim_options
& MEM_CGROUP_RECLAIM_SOFT
;
1261 unsigned long excess
= mem_cgroup_get_excess(root_mem
);
1263 /* If memsw_is_minimum==1, swap-out is of-no-use. */
1264 if (root_mem
->memsw_is_minimum
)
1268 victim
= mem_cgroup_select_victim(root_mem
);
1269 if (victim
== root_mem
) {
1272 drain_all_stock_async();
1275 * If we have not been able to reclaim
1276 * anything, it might because there are
1277 * no reclaimable pages under this hierarchy
1279 if (!check_soft
|| !total
) {
1280 css_put(&victim
->css
);
1284 * We want to do more targetted reclaim.
1285 * excess >> 2 is not to excessive so as to
1286 * reclaim too much, nor too less that we keep
1287 * coming back to reclaim from this cgroup
1289 if (total
>= (excess
>> 2) ||
1290 (loop
> MEM_CGROUP_MAX_RECLAIM_LOOPS
)) {
1291 css_put(&victim
->css
);
1296 if (!mem_cgroup_local_usage(victim
)) {
1297 /* this cgroup's local usage == 0 */
1298 css_put(&victim
->css
);
1301 /* we use swappiness of local cgroup */
1303 ret
= mem_cgroup_shrink_node_zone(victim
, gfp_mask
,
1304 noswap
, get_swappiness(victim
), zone
);
1306 ret
= try_to_free_mem_cgroup_pages(victim
, gfp_mask
,
1307 noswap
, get_swappiness(victim
));
1308 css_put(&victim
->css
);
1310 * At shrinking usage, we can't check we should stop here or
1311 * reclaim more. It's depends on callers. last_scanned_child
1312 * will work enough for keeping fairness under tree.
1318 if (res_counter_check_under_soft_limit(&root_mem
->res
))
1320 } else if (mem_cgroup_check_under_limit(root_mem
))
1326 static int mem_cgroup_oom_lock_cb(struct mem_cgroup
*mem
, void *data
)
1328 int *val
= (int *)data
;
1331 * Logically, we can stop scanning immediately when we find
1332 * a memcg is already locked. But condidering unlock ops and
1333 * creation/removal of memcg, scan-all is simple operation.
1335 x
= atomic_inc_return(&mem
->oom_lock
);
1336 *val
= max(x
, *val
);
1340 * Check OOM-Killer is already running under our hierarchy.
1341 * If someone is running, return false.
1343 static bool mem_cgroup_oom_lock(struct mem_cgroup
*mem
)
1347 mem_cgroup_walk_tree(mem
, &lock_count
, mem_cgroup_oom_lock_cb
);
1349 if (lock_count
== 1)
1354 static int mem_cgroup_oom_unlock_cb(struct mem_cgroup
*mem
, void *data
)
1357 * When a new child is created while the hierarchy is under oom,
1358 * mem_cgroup_oom_lock() may not be called. We have to use
1359 * atomic_add_unless() here.
1361 atomic_add_unless(&mem
->oom_lock
, -1, 0);
1365 static void mem_cgroup_oom_unlock(struct mem_cgroup
*mem
)
1367 mem_cgroup_walk_tree(mem
, NULL
, mem_cgroup_oom_unlock_cb
);
1370 static DEFINE_MUTEX(memcg_oom_mutex
);
1371 static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq
);
1373 struct oom_wait_info
{
1374 struct mem_cgroup
*mem
;
1378 static int memcg_oom_wake_function(wait_queue_t
*wait
,
1379 unsigned mode
, int sync
, void *arg
)
1381 struct mem_cgroup
*wake_mem
= (struct mem_cgroup
*)arg
;
1382 struct oom_wait_info
*oom_wait_info
;
1384 oom_wait_info
= container_of(wait
, struct oom_wait_info
, wait
);
1386 if (oom_wait_info
->mem
== wake_mem
)
1388 /* if no hierarchy, no match */
1389 if (!oom_wait_info
->mem
->use_hierarchy
|| !wake_mem
->use_hierarchy
)
1392 * Both of oom_wait_info->mem and wake_mem are stable under us.
1393 * Then we can use css_is_ancestor without taking care of RCU.
1395 if (!css_is_ancestor(&oom_wait_info
->mem
->css
, &wake_mem
->css
) &&
1396 !css_is_ancestor(&wake_mem
->css
, &oom_wait_info
->mem
->css
))
1400 return autoremove_wake_function(wait
, mode
, sync
, arg
);
1403 static void memcg_wakeup_oom(struct mem_cgroup
*mem
)
1405 /* for filtering, pass "mem" as argument. */
1406 __wake_up(&memcg_oom_waitq
, TASK_NORMAL
, 0, mem
);
1409 static void memcg_oom_recover(struct mem_cgroup
*mem
)
1411 if (mem
&& atomic_read(&mem
->oom_lock
))
1412 memcg_wakeup_oom(mem
);
1416 * try to call OOM killer. returns false if we should exit memory-reclaim loop.
1418 bool mem_cgroup_handle_oom(struct mem_cgroup
*mem
, gfp_t mask
)
1420 struct oom_wait_info owait
;
1421 bool locked
, need_to_kill
;
1424 owait
.wait
.flags
= 0;
1425 owait
.wait
.func
= memcg_oom_wake_function
;
1426 owait
.wait
.private = current
;
1427 INIT_LIST_HEAD(&owait
.wait
.task_list
);
1428 need_to_kill
= true;
1429 /* At first, try to OOM lock hierarchy under mem.*/
1430 mutex_lock(&memcg_oom_mutex
);
1431 locked
= mem_cgroup_oom_lock(mem
);
1433 * Even if signal_pending(), we can't quit charge() loop without
1434 * accounting. So, UNINTERRUPTIBLE is appropriate. But SIGKILL
1435 * under OOM is always welcomed, use TASK_KILLABLE here.
1437 prepare_to_wait(&memcg_oom_waitq
, &owait
.wait
, TASK_KILLABLE
);
1438 if (!locked
|| mem
->oom_kill_disable
)
1439 need_to_kill
= false;
1441 mem_cgroup_oom_notify(mem
);
1442 mutex_unlock(&memcg_oom_mutex
);
1445 finish_wait(&memcg_oom_waitq
, &owait
.wait
);
1446 mem_cgroup_out_of_memory(mem
, mask
);
1449 finish_wait(&memcg_oom_waitq
, &owait
.wait
);
1451 mutex_lock(&memcg_oom_mutex
);
1452 mem_cgroup_oom_unlock(mem
);
1453 memcg_wakeup_oom(mem
);
1454 mutex_unlock(&memcg_oom_mutex
);
1456 if (test_thread_flag(TIF_MEMDIE
) || fatal_signal_pending(current
))
1458 /* Give chance to dying process */
1459 schedule_timeout(1);
1464 * Currently used to update mapped file statistics, but the routine can be
1465 * generalized to update other statistics as well.
1467 void mem_cgroup_update_file_mapped(struct page
*page
, int val
)
1469 struct mem_cgroup
*mem
;
1470 struct page_cgroup
*pc
;
1472 pc
= lookup_page_cgroup(page
);
1476 lock_page_cgroup(pc
);
1477 mem
= pc
->mem_cgroup
;
1478 if (!mem
|| !PageCgroupUsed(pc
))
1482 * Preemption is already disabled. We can use __this_cpu_xxx
1485 __this_cpu_inc(mem
->stat
->count
[MEM_CGROUP_STAT_FILE_MAPPED
]);
1486 SetPageCgroupFileMapped(pc
);
1488 __this_cpu_dec(mem
->stat
->count
[MEM_CGROUP_STAT_FILE_MAPPED
]);
1489 ClearPageCgroupFileMapped(pc
);
1493 unlock_page_cgroup(pc
);
1497 * size of first charge trial. "32" comes from vmscan.c's magic value.
1498 * TODO: maybe necessary to use big numbers in big irons.
1500 #define CHARGE_SIZE (32 * PAGE_SIZE)
1501 struct memcg_stock_pcp
{
1502 struct mem_cgroup
*cached
; /* this never be root cgroup */
1504 struct work_struct work
;
1506 static DEFINE_PER_CPU(struct memcg_stock_pcp
, memcg_stock
);
1507 static atomic_t memcg_drain_count
;
1510 * Try to consume stocked charge on this cpu. If success, PAGE_SIZE is consumed
1511 * from local stock and true is returned. If the stock is 0 or charges from a
1512 * cgroup which is not current target, returns false. This stock will be
1515 static bool consume_stock(struct mem_cgroup
*mem
)
1517 struct memcg_stock_pcp
*stock
;
1520 stock
= &get_cpu_var(memcg_stock
);
1521 if (mem
== stock
->cached
&& stock
->charge
)
1522 stock
->charge
-= PAGE_SIZE
;
1523 else /* need to call res_counter_charge */
1525 put_cpu_var(memcg_stock
);
1530 * Returns stocks cached in percpu to res_counter and reset cached information.
1532 static void drain_stock(struct memcg_stock_pcp
*stock
)
1534 struct mem_cgroup
*old
= stock
->cached
;
1536 if (stock
->charge
) {
1537 res_counter_uncharge(&old
->res
, stock
->charge
);
1538 if (do_swap_account
)
1539 res_counter_uncharge(&old
->memsw
, stock
->charge
);
1541 stock
->cached
= NULL
;
1546 * This must be called under preempt disabled or must be called by
1547 * a thread which is pinned to local cpu.
1549 static void drain_local_stock(struct work_struct
*dummy
)
1551 struct memcg_stock_pcp
*stock
= &__get_cpu_var(memcg_stock
);
1556 * Cache charges(val) which is from res_counter, to local per_cpu area.
1557 * This will be consumed by consume_stock() function, later.
1559 static void refill_stock(struct mem_cgroup
*mem
, int val
)
1561 struct memcg_stock_pcp
*stock
= &get_cpu_var(memcg_stock
);
1563 if (stock
->cached
!= mem
) { /* reset if necessary */
1565 stock
->cached
= mem
;
1567 stock
->charge
+= val
;
1568 put_cpu_var(memcg_stock
);
1572 * Tries to drain stocked charges in other cpus. This function is asynchronous
1573 * and just put a work per cpu for draining localy on each cpu. Caller can
1574 * expects some charges will be back to res_counter later but cannot wait for
1577 static void drain_all_stock_async(void)
1580 /* This function is for scheduling "drain" in asynchronous way.
1581 * The result of "drain" is not directly handled by callers. Then,
1582 * if someone is calling drain, we don't have to call drain more.
1583 * Anyway, WORK_STRUCT_PENDING check in queue_work_on() will catch if
1584 * there is a race. We just do loose check here.
1586 if (atomic_read(&memcg_drain_count
))
1588 /* Notify other cpus that system-wide "drain" is running */
1589 atomic_inc(&memcg_drain_count
);
1591 for_each_online_cpu(cpu
) {
1592 struct memcg_stock_pcp
*stock
= &per_cpu(memcg_stock
, cpu
);
1593 schedule_work_on(cpu
, &stock
->work
);
1596 atomic_dec(&memcg_drain_count
);
1597 /* We don't wait for flush_work */
1600 /* This is a synchronous drain interface. */
1601 static void drain_all_stock_sync(void)
1603 /* called when force_empty is called */
1604 atomic_inc(&memcg_drain_count
);
1605 schedule_on_each_cpu(drain_local_stock
);
1606 atomic_dec(&memcg_drain_count
);
1609 static int __cpuinit
memcg_stock_cpu_callback(struct notifier_block
*nb
,
1610 unsigned long action
,
1613 int cpu
= (unsigned long)hcpu
;
1614 struct memcg_stock_pcp
*stock
;
1616 if (action
!= CPU_DEAD
)
1618 stock
= &per_cpu(memcg_stock
, cpu
);
1624 /* See __mem_cgroup_try_charge() for details */
1626 CHARGE_OK
, /* success */
1627 CHARGE_RETRY
, /* need to retry but retry is not bad */
1628 CHARGE_NOMEM
, /* we can't do more. return -ENOMEM */
1629 CHARGE_WOULDBLOCK
, /* GFP_WAIT wasn't set and no enough res. */
1630 CHARGE_OOM_DIE
, /* the current is killed because of OOM */
1633 static int __mem_cgroup_do_charge(struct mem_cgroup
*mem
, gfp_t gfp_mask
,
1634 int csize
, bool oom_check
)
1636 struct mem_cgroup
*mem_over_limit
;
1637 struct res_counter
*fail_res
;
1638 unsigned long flags
= 0;
1641 ret
= res_counter_charge(&mem
->res
, csize
, &fail_res
);
1644 if (!do_swap_account
)
1646 ret
= res_counter_charge(&mem
->memsw
, csize
, &fail_res
);
1650 mem_over_limit
= mem_cgroup_from_res_counter(fail_res
, memsw
);
1651 flags
|= MEM_CGROUP_RECLAIM_NOSWAP
;
1653 mem_over_limit
= mem_cgroup_from_res_counter(fail_res
, res
);
1655 if (csize
> PAGE_SIZE
) /* change csize and retry */
1656 return CHARGE_RETRY
;
1658 if (!(gfp_mask
& __GFP_WAIT
))
1659 return CHARGE_WOULDBLOCK
;
1661 ret
= mem_cgroup_hierarchical_reclaim(mem_over_limit
, NULL
,
1664 * try_to_free_mem_cgroup_pages() might not give us a full
1665 * picture of reclaim. Some pages are reclaimed and might be
1666 * moved to swap cache or just unmapped from the cgroup.
1667 * Check the limit again to see if the reclaim reduced the
1668 * current usage of the cgroup before giving up
1670 if (ret
|| mem_cgroup_check_under_limit(mem_over_limit
))
1671 return CHARGE_RETRY
;
1674 * At task move, charge accounts can be doubly counted. So, it's
1675 * better to wait until the end of task_move if something is going on.
1677 if (mem_cgroup_wait_acct_move(mem_over_limit
))
1678 return CHARGE_RETRY
;
1680 /* If we don't need to call oom-killer at el, return immediately */
1682 return CHARGE_NOMEM
;
1684 if (!mem_cgroup_handle_oom(mem_over_limit
, gfp_mask
))
1685 return CHARGE_OOM_DIE
;
1687 return CHARGE_RETRY
;
1691 * Unlike exported interface, "oom" parameter is added. if oom==true,
1692 * oom-killer can be invoked.
1694 static int __mem_cgroup_try_charge(struct mm_struct
*mm
,
1695 gfp_t gfp_mask
, struct mem_cgroup
**memcg
, bool oom
)
1697 int nr_oom_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
1698 struct mem_cgroup
*mem
= NULL
;
1700 int csize
= CHARGE_SIZE
;
1703 * Unlike gloval-vm's OOM-kill, we're not in memory shortage
1704 * in system level. So, allow to go ahead dying process in addition to
1707 if (unlikely(test_thread_flag(TIF_MEMDIE
)
1708 || fatal_signal_pending(current
)))
1712 * We always charge the cgroup the mm_struct belongs to.
1713 * The mm_struct's mem_cgroup changes on task migration if the
1714 * thread group leader migrates. It's possible that mm is not
1715 * set, if so charge the init_mm (happens for pagecache usage).
1720 if (*memcg
) { /* css should be a valid one */
1722 VM_BUG_ON(css_is_removed(&mem
->css
));
1723 if (mem_cgroup_is_root(mem
))
1725 if (consume_stock(mem
))
1729 struct task_struct
*p
;
1732 p
= rcu_dereference(mm
->owner
);
1735 * because we don't have task_lock(), "p" can exit while
1736 * we're here. In that case, "mem" can point to root
1737 * cgroup but never be NULL. (and task_struct itself is freed
1738 * by RCU, cgroup itself is RCU safe.) Then, we have small
1739 * risk here to get wrong cgroup. But such kind of mis-account
1740 * by race always happens because we don't have cgroup_mutex().
1741 * It's overkill and we allow that small race, here.
1743 mem
= mem_cgroup_from_task(p
);
1745 if (mem_cgroup_is_root(mem
)) {
1749 if (consume_stock(mem
)) {
1751 * It seems dagerous to access memcg without css_get().
1752 * But considering how consume_stok works, it's not
1753 * necessary. If consume_stock success, some charges
1754 * from this memcg are cached on this cpu. So, we
1755 * don't need to call css_get()/css_tryget() before
1756 * calling consume_stock().
1761 /* after here, we may be blocked. we need to get refcnt */
1762 if (!css_tryget(&mem
->css
)) {
1772 /* If killed, bypass charge */
1773 if (fatal_signal_pending(current
)) {
1779 if (oom
&& !nr_oom_retries
) {
1781 nr_oom_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
1784 ret
= __mem_cgroup_do_charge(mem
, gfp_mask
, csize
, oom_check
);
1789 case CHARGE_RETRY
: /* not in OOM situation but retry */
1794 case CHARGE_WOULDBLOCK
: /* !__GFP_WAIT */
1797 case CHARGE_NOMEM
: /* OOM routine works */
1802 /* If oom, we never return -ENOMEM */
1805 case CHARGE_OOM_DIE
: /* Killed by OOM Killer */
1809 } while (ret
!= CHARGE_OK
);
1811 if (csize
> PAGE_SIZE
)
1812 refill_stock(mem
, csize
- PAGE_SIZE
);
1826 * Somemtimes we have to undo a charge we got by try_charge().
1827 * This function is for that and do uncharge, put css's refcnt.
1828 * gotten by try_charge().
1830 static void __mem_cgroup_cancel_charge(struct mem_cgroup
*mem
,
1831 unsigned long count
)
1833 if (!mem_cgroup_is_root(mem
)) {
1834 res_counter_uncharge(&mem
->res
, PAGE_SIZE
* count
);
1835 if (do_swap_account
)
1836 res_counter_uncharge(&mem
->memsw
, PAGE_SIZE
* count
);
1840 static void mem_cgroup_cancel_charge(struct mem_cgroup
*mem
)
1842 __mem_cgroup_cancel_charge(mem
, 1);
1846 * A helper function to get mem_cgroup from ID. must be called under
1847 * rcu_read_lock(). The caller must check css_is_removed() or some if
1848 * it's concern. (dropping refcnt from swap can be called against removed
1851 static struct mem_cgroup
*mem_cgroup_lookup(unsigned short id
)
1853 struct cgroup_subsys_state
*css
;
1855 /* ID 0 is unused ID */
1858 css
= css_lookup(&mem_cgroup_subsys
, id
);
1861 return container_of(css
, struct mem_cgroup
, css
);
1864 struct mem_cgroup
*try_get_mem_cgroup_from_page(struct page
*page
)
1866 struct mem_cgroup
*mem
= NULL
;
1867 struct page_cgroup
*pc
;
1871 VM_BUG_ON(!PageLocked(page
));
1873 pc
= lookup_page_cgroup(page
);
1874 lock_page_cgroup(pc
);
1875 if (PageCgroupUsed(pc
)) {
1876 mem
= pc
->mem_cgroup
;
1877 if (mem
&& !css_tryget(&mem
->css
))
1879 } else if (PageSwapCache(page
)) {
1880 ent
.val
= page_private(page
);
1881 id
= lookup_swap_cgroup(ent
);
1883 mem
= mem_cgroup_lookup(id
);
1884 if (mem
&& !css_tryget(&mem
->css
))
1888 unlock_page_cgroup(pc
);
1893 * commit a charge got by __mem_cgroup_try_charge() and makes page_cgroup to be
1894 * USED state. If already USED, uncharge and return.
1897 static void __mem_cgroup_commit_charge(struct mem_cgroup
*mem
,
1898 struct page_cgroup
*pc
,
1899 enum charge_type ctype
)
1901 /* try_charge() can return NULL to *memcg, taking care of it. */
1905 lock_page_cgroup(pc
);
1906 if (unlikely(PageCgroupUsed(pc
))) {
1907 unlock_page_cgroup(pc
);
1908 mem_cgroup_cancel_charge(mem
);
1912 pc
->mem_cgroup
= mem
;
1914 * We access a page_cgroup asynchronously without lock_page_cgroup().
1915 * Especially when a page_cgroup is taken from a page, pc->mem_cgroup
1916 * is accessed after testing USED bit. To make pc->mem_cgroup visible
1917 * before USED bit, we need memory barrier here.
1918 * See mem_cgroup_add_lru_list(), etc.
1922 case MEM_CGROUP_CHARGE_TYPE_CACHE
:
1923 case MEM_CGROUP_CHARGE_TYPE_SHMEM
:
1924 SetPageCgroupCache(pc
);
1925 SetPageCgroupUsed(pc
);
1927 case MEM_CGROUP_CHARGE_TYPE_MAPPED
:
1928 ClearPageCgroupCache(pc
);
1929 SetPageCgroupUsed(pc
);
1935 mem_cgroup_charge_statistics(mem
, pc
, true);
1937 unlock_page_cgroup(pc
);
1939 * "charge_statistics" updated event counter. Then, check it.
1940 * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree.
1941 * if they exceeds softlimit.
1943 memcg_check_events(mem
, pc
->page
);
1947 * __mem_cgroup_move_account - move account of the page
1948 * @pc: page_cgroup of the page.
1949 * @from: mem_cgroup which the page is moved from.
1950 * @to: mem_cgroup which the page is moved to. @from != @to.
1951 * @uncharge: whether we should call uncharge and css_put against @from.
1953 * The caller must confirm following.
1954 * - page is not on LRU (isolate_page() is useful.)
1955 * - the pc is locked, used, and ->mem_cgroup points to @from.
1957 * This function doesn't do "charge" nor css_get to new cgroup. It should be
1958 * done by a caller(__mem_cgroup_try_charge would be usefull). If @uncharge is
1959 * true, this function does "uncharge" from old cgroup, but it doesn't if
1960 * @uncharge is false, so a caller should do "uncharge".
1963 static void __mem_cgroup_move_account(struct page_cgroup
*pc
,
1964 struct mem_cgroup
*from
, struct mem_cgroup
*to
, bool uncharge
)
1966 VM_BUG_ON(from
== to
);
1967 VM_BUG_ON(PageLRU(pc
->page
));
1968 VM_BUG_ON(!PageCgroupLocked(pc
));
1969 VM_BUG_ON(!PageCgroupUsed(pc
));
1970 VM_BUG_ON(pc
->mem_cgroup
!= from
);
1972 if (PageCgroupFileMapped(pc
)) {
1973 /* Update mapped_file data for mem_cgroup */
1975 __this_cpu_dec(from
->stat
->count
[MEM_CGROUP_STAT_FILE_MAPPED
]);
1976 __this_cpu_inc(to
->stat
->count
[MEM_CGROUP_STAT_FILE_MAPPED
]);
1979 mem_cgroup_charge_statistics(from
, pc
, false);
1981 /* This is not "cancel", but cancel_charge does all we need. */
1982 mem_cgroup_cancel_charge(from
);
1984 /* caller should have done css_get */
1985 pc
->mem_cgroup
= to
;
1986 mem_cgroup_charge_statistics(to
, pc
, true);
1988 * We charges against "to" which may not have any tasks. Then, "to"
1989 * can be under rmdir(). But in current implementation, caller of
1990 * this function is just force_empty() and move charge, so it's
1991 * garanteed that "to" is never removed. So, we don't check rmdir
1997 * check whether the @pc is valid for moving account and call
1998 * __mem_cgroup_move_account()
2000 static int mem_cgroup_move_account(struct page_cgroup
*pc
,
2001 struct mem_cgroup
*from
, struct mem_cgroup
*to
, bool uncharge
)
2004 lock_page_cgroup(pc
);
2005 if (PageCgroupUsed(pc
) && pc
->mem_cgroup
== from
) {
2006 __mem_cgroup_move_account(pc
, from
, to
, uncharge
);
2009 unlock_page_cgroup(pc
);
2013 memcg_check_events(to
, pc
->page
);
2014 memcg_check_events(from
, pc
->page
);
2019 * move charges to its parent.
2022 static int mem_cgroup_move_parent(struct page_cgroup
*pc
,
2023 struct mem_cgroup
*child
,
2026 struct page
*page
= pc
->page
;
2027 struct cgroup
*cg
= child
->css
.cgroup
;
2028 struct cgroup
*pcg
= cg
->parent
;
2029 struct mem_cgroup
*parent
;
2037 if (!get_page_unless_zero(page
))
2039 if (isolate_lru_page(page
))
2042 parent
= mem_cgroup_from_cont(pcg
);
2043 ret
= __mem_cgroup_try_charge(NULL
, gfp_mask
, &parent
, false);
2047 ret
= mem_cgroup_move_account(pc
, child
, parent
, true);
2049 mem_cgroup_cancel_charge(parent
);
2051 putback_lru_page(page
);
2059 * Charge the memory controller for page usage.
2061 * 0 if the charge was successful
2062 * < 0 if the cgroup is over its limit
2064 static int mem_cgroup_charge_common(struct page
*page
, struct mm_struct
*mm
,
2065 gfp_t gfp_mask
, enum charge_type ctype
)
2067 struct mem_cgroup
*mem
= NULL
;
2068 struct page_cgroup
*pc
;
2071 pc
= lookup_page_cgroup(page
);
2072 /* can happen at boot */
2077 ret
= __mem_cgroup_try_charge(mm
, gfp_mask
, &mem
, true);
2081 __mem_cgroup_commit_charge(mem
, pc
, ctype
);
2085 int mem_cgroup_newpage_charge(struct page
*page
,
2086 struct mm_struct
*mm
, gfp_t gfp_mask
)
2088 if (mem_cgroup_disabled())
2090 if (PageCompound(page
))
2093 * If already mapped, we don't have to account.
2094 * If page cache, page->mapping has address_space.
2095 * But page->mapping may have out-of-use anon_vma pointer,
2096 * detecit it by PageAnon() check. newly-mapped-anon's page->mapping
2099 if (page_mapped(page
) || (page
->mapping
&& !PageAnon(page
)))
2103 return mem_cgroup_charge_common(page
, mm
, gfp_mask
,
2104 MEM_CGROUP_CHARGE_TYPE_MAPPED
);
2108 __mem_cgroup_commit_charge_swapin(struct page
*page
, struct mem_cgroup
*ptr
,
2109 enum charge_type ctype
);
2111 int mem_cgroup_cache_charge(struct page
*page
, struct mm_struct
*mm
,
2116 if (mem_cgroup_disabled())
2118 if (PageCompound(page
))
2121 * Corner case handling. This is called from add_to_page_cache()
2122 * in usual. But some FS (shmem) precharges this page before calling it
2123 * and call add_to_page_cache() with GFP_NOWAIT.
2125 * For GFP_NOWAIT case, the page may be pre-charged before calling
2126 * add_to_page_cache(). (See shmem.c) check it here and avoid to call
2127 * charge twice. (It works but has to pay a bit larger cost.)
2128 * And when the page is SwapCache, it should take swap information
2129 * into account. This is under lock_page() now.
2131 if (!(gfp_mask
& __GFP_WAIT
)) {
2132 struct page_cgroup
*pc
;
2134 pc
= lookup_page_cgroup(page
);
2137 lock_page_cgroup(pc
);
2138 if (PageCgroupUsed(pc
)) {
2139 unlock_page_cgroup(pc
);
2142 unlock_page_cgroup(pc
);
2148 if (page_is_file_cache(page
))
2149 return mem_cgroup_charge_common(page
, mm
, gfp_mask
,
2150 MEM_CGROUP_CHARGE_TYPE_CACHE
);
2153 if (PageSwapCache(page
)) {
2154 struct mem_cgroup
*mem
= NULL
;
2156 ret
= mem_cgroup_try_charge_swapin(mm
, page
, gfp_mask
, &mem
);
2158 __mem_cgroup_commit_charge_swapin(page
, mem
,
2159 MEM_CGROUP_CHARGE_TYPE_SHMEM
);
2161 ret
= mem_cgroup_charge_common(page
, mm
, gfp_mask
,
2162 MEM_CGROUP_CHARGE_TYPE_SHMEM
);
2168 * While swap-in, try_charge -> commit or cancel, the page is locked.
2169 * And when try_charge() successfully returns, one refcnt to memcg without
2170 * struct page_cgroup is acquired. This refcnt will be consumed by
2171 * "commit()" or removed by "cancel()"
2173 int mem_cgroup_try_charge_swapin(struct mm_struct
*mm
,
2175 gfp_t mask
, struct mem_cgroup
**ptr
)
2177 struct mem_cgroup
*mem
;
2180 if (mem_cgroup_disabled())
2183 if (!do_swap_account
)
2186 * A racing thread's fault, or swapoff, may have already updated
2187 * the pte, and even removed page from swap cache: in those cases
2188 * do_swap_page()'s pte_same() test will fail; but there's also a
2189 * KSM case which does need to charge the page.
2191 if (!PageSwapCache(page
))
2193 mem
= try_get_mem_cgroup_from_page(page
);
2197 ret
= __mem_cgroup_try_charge(NULL
, mask
, ptr
, true);
2203 return __mem_cgroup_try_charge(mm
, mask
, ptr
, true);
2207 __mem_cgroup_commit_charge_swapin(struct page
*page
, struct mem_cgroup
*ptr
,
2208 enum charge_type ctype
)
2210 struct page_cgroup
*pc
;
2212 if (mem_cgroup_disabled())
2216 cgroup_exclude_rmdir(&ptr
->css
);
2217 pc
= lookup_page_cgroup(page
);
2218 mem_cgroup_lru_del_before_commit_swapcache(page
);
2219 __mem_cgroup_commit_charge(ptr
, pc
, ctype
);
2220 mem_cgroup_lru_add_after_commit_swapcache(page
);
2222 * Now swap is on-memory. This means this page may be
2223 * counted both as mem and swap....double count.
2224 * Fix it by uncharging from memsw. Basically, this SwapCache is stable
2225 * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page()
2226 * may call delete_from_swap_cache() before reach here.
2228 if (do_swap_account
&& PageSwapCache(page
)) {
2229 swp_entry_t ent
= {.val
= page_private(page
)};
2231 struct mem_cgroup
*memcg
;
2233 id
= swap_cgroup_record(ent
, 0);
2235 memcg
= mem_cgroup_lookup(id
);
2238 * This recorded memcg can be obsolete one. So, avoid
2239 * calling css_tryget
2241 if (!mem_cgroup_is_root(memcg
))
2242 res_counter_uncharge(&memcg
->memsw
, PAGE_SIZE
);
2243 mem_cgroup_swap_statistics(memcg
, false);
2244 mem_cgroup_put(memcg
);
2249 * At swapin, we may charge account against cgroup which has no tasks.
2250 * So, rmdir()->pre_destroy() can be called while we do this charge.
2251 * In that case, we need to call pre_destroy() again. check it here.
2253 cgroup_release_and_wakeup_rmdir(&ptr
->css
);
2256 void mem_cgroup_commit_charge_swapin(struct page
*page
, struct mem_cgroup
*ptr
)
2258 __mem_cgroup_commit_charge_swapin(page
, ptr
,
2259 MEM_CGROUP_CHARGE_TYPE_MAPPED
);
2262 void mem_cgroup_cancel_charge_swapin(struct mem_cgroup
*mem
)
2264 if (mem_cgroup_disabled())
2268 mem_cgroup_cancel_charge(mem
);
2272 __do_uncharge(struct mem_cgroup
*mem
, const enum charge_type ctype
)
2274 struct memcg_batch_info
*batch
= NULL
;
2275 bool uncharge_memsw
= true;
2276 /* If swapout, usage of swap doesn't decrease */
2277 if (!do_swap_account
|| ctype
== MEM_CGROUP_CHARGE_TYPE_SWAPOUT
)
2278 uncharge_memsw
= false;
2280 batch
= ¤t
->memcg_batch
;
2282 * In usual, we do css_get() when we remember memcg pointer.
2283 * But in this case, we keep res->usage until end of a series of
2284 * uncharges. Then, it's ok to ignore memcg's refcnt.
2289 * do_batch > 0 when unmapping pages or inode invalidate/truncate.
2290 * In those cases, all pages freed continously can be expected to be in
2291 * the same cgroup and we have chance to coalesce uncharges.
2292 * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE)
2293 * because we want to do uncharge as soon as possible.
2296 if (!batch
->do_batch
|| test_thread_flag(TIF_MEMDIE
))
2297 goto direct_uncharge
;
2300 * In typical case, batch->memcg == mem. This means we can
2301 * merge a series of uncharges to an uncharge of res_counter.
2302 * If not, we uncharge res_counter ony by one.
2304 if (batch
->memcg
!= mem
)
2305 goto direct_uncharge
;
2306 /* remember freed charge and uncharge it later */
2307 batch
->bytes
+= PAGE_SIZE
;
2309 batch
->memsw_bytes
+= PAGE_SIZE
;
2312 res_counter_uncharge(&mem
->res
, PAGE_SIZE
);
2314 res_counter_uncharge(&mem
->memsw
, PAGE_SIZE
);
2315 if (unlikely(batch
->memcg
!= mem
))
2316 memcg_oom_recover(mem
);
2321 * uncharge if !page_mapped(page)
2323 static struct mem_cgroup
*
2324 __mem_cgroup_uncharge_common(struct page
*page
, enum charge_type ctype
)
2326 struct page_cgroup
*pc
;
2327 struct mem_cgroup
*mem
= NULL
;
2329 if (mem_cgroup_disabled())
2332 if (PageSwapCache(page
))
2336 * Check if our page_cgroup is valid
2338 pc
= lookup_page_cgroup(page
);
2339 if (unlikely(!pc
|| !PageCgroupUsed(pc
)))
2342 lock_page_cgroup(pc
);
2344 mem
= pc
->mem_cgroup
;
2346 if (!PageCgroupUsed(pc
))
2350 case MEM_CGROUP_CHARGE_TYPE_MAPPED
:
2351 case MEM_CGROUP_CHARGE_TYPE_DROP
:
2352 /* See mem_cgroup_prepare_migration() */
2353 if (page_mapped(page
) || PageCgroupMigration(pc
))
2356 case MEM_CGROUP_CHARGE_TYPE_SWAPOUT
:
2357 if (!PageAnon(page
)) { /* Shared memory */
2358 if (page
->mapping
&& !page_is_file_cache(page
))
2360 } else if (page_mapped(page
)) /* Anon */
2367 mem_cgroup_charge_statistics(mem
, pc
, false);
2369 ClearPageCgroupUsed(pc
);
2371 * pc->mem_cgroup is not cleared here. It will be accessed when it's
2372 * freed from LRU. This is safe because uncharged page is expected not
2373 * to be reused (freed soon). Exception is SwapCache, it's handled by
2374 * special functions.
2377 unlock_page_cgroup(pc
);
2379 * even after unlock, we have mem->res.usage here and this memcg
2380 * will never be freed.
2382 memcg_check_events(mem
, page
);
2383 if (do_swap_account
&& ctype
== MEM_CGROUP_CHARGE_TYPE_SWAPOUT
) {
2384 mem_cgroup_swap_statistics(mem
, true);
2385 mem_cgroup_get(mem
);
2387 if (!mem_cgroup_is_root(mem
))
2388 __do_uncharge(mem
, ctype
);
2393 unlock_page_cgroup(pc
);
2397 void mem_cgroup_uncharge_page(struct page
*page
)
2400 if (page_mapped(page
))
2402 if (page
->mapping
&& !PageAnon(page
))
2404 __mem_cgroup_uncharge_common(page
, MEM_CGROUP_CHARGE_TYPE_MAPPED
);
2407 void mem_cgroup_uncharge_cache_page(struct page
*page
)
2409 VM_BUG_ON(page_mapped(page
));
2410 VM_BUG_ON(page
->mapping
);
2411 __mem_cgroup_uncharge_common(page
, MEM_CGROUP_CHARGE_TYPE_CACHE
);
2415 * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate.
2416 * In that cases, pages are freed continuously and we can expect pages
2417 * are in the same memcg. All these calls itself limits the number of
2418 * pages freed at once, then uncharge_start/end() is called properly.
2419 * This may be called prural(2) times in a context,
2422 void mem_cgroup_uncharge_start(void)
2424 current
->memcg_batch
.do_batch
++;
2425 /* We can do nest. */
2426 if (current
->memcg_batch
.do_batch
== 1) {
2427 current
->memcg_batch
.memcg
= NULL
;
2428 current
->memcg_batch
.bytes
= 0;
2429 current
->memcg_batch
.memsw_bytes
= 0;
2433 void mem_cgroup_uncharge_end(void)
2435 struct memcg_batch_info
*batch
= ¤t
->memcg_batch
;
2437 if (!batch
->do_batch
)
2441 if (batch
->do_batch
) /* If stacked, do nothing. */
2447 * This "batch->memcg" is valid without any css_get/put etc...
2448 * bacause we hide charges behind us.
2451 res_counter_uncharge(&batch
->memcg
->res
, batch
->bytes
);
2452 if (batch
->memsw_bytes
)
2453 res_counter_uncharge(&batch
->memcg
->memsw
, batch
->memsw_bytes
);
2454 memcg_oom_recover(batch
->memcg
);
2455 /* forget this pointer (for sanity check) */
2456 batch
->memcg
= NULL
;
2461 * called after __delete_from_swap_cache() and drop "page" account.
2462 * memcg information is recorded to swap_cgroup of "ent"
2465 mem_cgroup_uncharge_swapcache(struct page
*page
, swp_entry_t ent
, bool swapout
)
2467 struct mem_cgroup
*memcg
;
2468 int ctype
= MEM_CGROUP_CHARGE_TYPE_SWAPOUT
;
2470 if (!swapout
) /* this was a swap cache but the swap is unused ! */
2471 ctype
= MEM_CGROUP_CHARGE_TYPE_DROP
;
2473 memcg
= __mem_cgroup_uncharge_common(page
, ctype
);
2476 * record memcg information, if swapout && memcg != NULL,
2477 * mem_cgroup_get() was called in uncharge().
2479 if (do_swap_account
&& swapout
&& memcg
)
2480 swap_cgroup_record(ent
, css_id(&memcg
->css
));
2484 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
2486 * called from swap_entry_free(). remove record in swap_cgroup and
2487 * uncharge "memsw" account.
2489 void mem_cgroup_uncharge_swap(swp_entry_t ent
)
2491 struct mem_cgroup
*memcg
;
2494 if (!do_swap_account
)
2497 id
= swap_cgroup_record(ent
, 0);
2499 memcg
= mem_cgroup_lookup(id
);
2502 * We uncharge this because swap is freed.
2503 * This memcg can be obsolete one. We avoid calling css_tryget
2505 if (!mem_cgroup_is_root(memcg
))
2506 res_counter_uncharge(&memcg
->memsw
, PAGE_SIZE
);
2507 mem_cgroup_swap_statistics(memcg
, false);
2508 mem_cgroup_put(memcg
);
2514 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
2515 * @entry: swap entry to be moved
2516 * @from: mem_cgroup which the entry is moved from
2517 * @to: mem_cgroup which the entry is moved to
2518 * @need_fixup: whether we should fixup res_counters and refcounts.
2520 * It succeeds only when the swap_cgroup's record for this entry is the same
2521 * as the mem_cgroup's id of @from.
2523 * Returns 0 on success, -EINVAL on failure.
2525 * The caller must have charged to @to, IOW, called res_counter_charge() about
2526 * both res and memsw, and called css_get().
2528 static int mem_cgroup_move_swap_account(swp_entry_t entry
,
2529 struct mem_cgroup
*from
, struct mem_cgroup
*to
, bool need_fixup
)
2531 unsigned short old_id
, new_id
;
2533 old_id
= css_id(&from
->css
);
2534 new_id
= css_id(&to
->css
);
2536 if (swap_cgroup_cmpxchg(entry
, old_id
, new_id
) == old_id
) {
2537 mem_cgroup_swap_statistics(from
, false);
2538 mem_cgroup_swap_statistics(to
, true);
2540 * This function is only called from task migration context now.
2541 * It postpones res_counter and refcount handling till the end
2542 * of task migration(mem_cgroup_clear_mc()) for performance
2543 * improvement. But we cannot postpone mem_cgroup_get(to)
2544 * because if the process that has been moved to @to does
2545 * swap-in, the refcount of @to might be decreased to 0.
2549 if (!mem_cgroup_is_root(from
))
2550 res_counter_uncharge(&from
->memsw
, PAGE_SIZE
);
2551 mem_cgroup_put(from
);
2553 * we charged both to->res and to->memsw, so we should
2556 if (!mem_cgroup_is_root(to
))
2557 res_counter_uncharge(&to
->res
, PAGE_SIZE
);
2564 static inline int mem_cgroup_move_swap_account(swp_entry_t entry
,
2565 struct mem_cgroup
*from
, struct mem_cgroup
*to
, bool need_fixup
)
2572 * Before starting migration, account PAGE_SIZE to mem_cgroup that the old
2575 int mem_cgroup_prepare_migration(struct page
*page
,
2576 struct page
*newpage
, struct mem_cgroup
**ptr
)
2578 struct page_cgroup
*pc
;
2579 struct mem_cgroup
*mem
= NULL
;
2580 enum charge_type ctype
;
2583 if (mem_cgroup_disabled())
2586 pc
= lookup_page_cgroup(page
);
2587 lock_page_cgroup(pc
);
2588 if (PageCgroupUsed(pc
)) {
2589 mem
= pc
->mem_cgroup
;
2592 * At migrating an anonymous page, its mapcount goes down
2593 * to 0 and uncharge() will be called. But, even if it's fully
2594 * unmapped, migration may fail and this page has to be
2595 * charged again. We set MIGRATION flag here and delay uncharge
2596 * until end_migration() is called
2598 * Corner Case Thinking
2600 * When the old page was mapped as Anon and it's unmap-and-freed
2601 * while migration was ongoing.
2602 * If unmap finds the old page, uncharge() of it will be delayed
2603 * until end_migration(). If unmap finds a new page, it's
2604 * uncharged when it make mapcount to be 1->0. If unmap code
2605 * finds swap_migration_entry, the new page will not be mapped
2606 * and end_migration() will find it(mapcount==0).
2609 * When the old page was mapped but migraion fails, the kernel
2610 * remaps it. A charge for it is kept by MIGRATION flag even
2611 * if mapcount goes down to 0. We can do remap successfully
2612 * without charging it again.
2615 * The "old" page is under lock_page() until the end of
2616 * migration, so, the old page itself will not be swapped-out.
2617 * If the new page is swapped out before end_migraton, our
2618 * hook to usual swap-out path will catch the event.
2621 SetPageCgroupMigration(pc
);
2623 unlock_page_cgroup(pc
);
2625 * If the page is not charged at this point,
2632 ret
= __mem_cgroup_try_charge(NULL
, GFP_KERNEL
, ptr
, false);
2633 css_put(&mem
->css
);/* drop extra refcnt */
2634 if (ret
|| *ptr
== NULL
) {
2635 if (PageAnon(page
)) {
2636 lock_page_cgroup(pc
);
2637 ClearPageCgroupMigration(pc
);
2638 unlock_page_cgroup(pc
);
2640 * The old page may be fully unmapped while we kept it.
2642 mem_cgroup_uncharge_page(page
);
2647 * We charge new page before it's used/mapped. So, even if unlock_page()
2648 * is called before end_migration, we can catch all events on this new
2649 * page. In the case new page is migrated but not remapped, new page's
2650 * mapcount will be finally 0 and we call uncharge in end_migration().
2652 pc
= lookup_page_cgroup(newpage
);
2654 ctype
= MEM_CGROUP_CHARGE_TYPE_MAPPED
;
2655 else if (page_is_file_cache(page
))
2656 ctype
= MEM_CGROUP_CHARGE_TYPE_CACHE
;
2658 ctype
= MEM_CGROUP_CHARGE_TYPE_SHMEM
;
2659 __mem_cgroup_commit_charge(mem
, pc
, ctype
);
2663 /* remove redundant charge if migration failed*/
2664 void mem_cgroup_end_migration(struct mem_cgroup
*mem
,
2665 struct page
*oldpage
, struct page
*newpage
)
2667 struct page
*used
, *unused
;
2668 struct page_cgroup
*pc
;
2672 /* blocks rmdir() */
2673 cgroup_exclude_rmdir(&mem
->css
);
2674 /* at migration success, oldpage->mapping is NULL. */
2675 if (oldpage
->mapping
) {
2683 * We disallowed uncharge of pages under migration because mapcount
2684 * of the page goes down to zero, temporarly.
2685 * Clear the flag and check the page should be charged.
2687 pc
= lookup_page_cgroup(oldpage
);
2688 lock_page_cgroup(pc
);
2689 ClearPageCgroupMigration(pc
);
2690 unlock_page_cgroup(pc
);
2692 __mem_cgroup_uncharge_common(unused
, MEM_CGROUP_CHARGE_TYPE_FORCE
);
2695 * If a page is a file cache, radix-tree replacement is very atomic
2696 * and we can skip this check. When it was an Anon page, its mapcount
2697 * goes down to 0. But because we added MIGRATION flage, it's not
2698 * uncharged yet. There are several case but page->mapcount check
2699 * and USED bit check in mem_cgroup_uncharge_page() will do enough
2700 * check. (see prepare_charge() also)
2703 mem_cgroup_uncharge_page(used
);
2705 * At migration, we may charge account against cgroup which has no
2707 * So, rmdir()->pre_destroy() can be called while we do this charge.
2708 * In that case, we need to call pre_destroy() again. check it here.
2710 cgroup_release_and_wakeup_rmdir(&mem
->css
);
2714 * A call to try to shrink memory usage on charge failure at shmem's swapin.
2715 * Calling hierarchical_reclaim is not enough because we should update
2716 * last_oom_jiffies to prevent pagefault_out_of_memory from invoking global OOM.
2717 * Moreover considering hierarchy, we should reclaim from the mem_over_limit,
2718 * not from the memcg which this page would be charged to.
2719 * try_charge_swapin does all of these works properly.
2721 int mem_cgroup_shmem_charge_fallback(struct page
*page
,
2722 struct mm_struct
*mm
,
2725 struct mem_cgroup
*mem
= NULL
;
2728 if (mem_cgroup_disabled())
2731 ret
= mem_cgroup_try_charge_swapin(mm
, page
, gfp_mask
, &mem
);
2733 mem_cgroup_cancel_charge_swapin(mem
); /* it does !mem check */
2738 static DEFINE_MUTEX(set_limit_mutex
);
2740 static int mem_cgroup_resize_limit(struct mem_cgroup
*memcg
,
2741 unsigned long long val
)
2744 u64 memswlimit
, memlimit
;
2746 int children
= mem_cgroup_count_children(memcg
);
2747 u64 curusage
, oldusage
;
2751 * For keeping hierarchical_reclaim simple, how long we should retry
2752 * is depends on callers. We set our retry-count to be function
2753 * of # of children which we should visit in this loop.
2755 retry_count
= MEM_CGROUP_RECLAIM_RETRIES
* children
;
2757 oldusage
= res_counter_read_u64(&memcg
->res
, RES_USAGE
);
2760 while (retry_count
) {
2761 if (signal_pending(current
)) {
2766 * Rather than hide all in some function, I do this in
2767 * open coded manner. You see what this really does.
2768 * We have to guarantee mem->res.limit < mem->memsw.limit.
2770 mutex_lock(&set_limit_mutex
);
2771 memswlimit
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
2772 if (memswlimit
< val
) {
2774 mutex_unlock(&set_limit_mutex
);
2778 memlimit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
2782 ret
= res_counter_set_limit(&memcg
->res
, val
);
2784 if (memswlimit
== val
)
2785 memcg
->memsw_is_minimum
= true;
2787 memcg
->memsw_is_minimum
= false;
2789 mutex_unlock(&set_limit_mutex
);
2794 mem_cgroup_hierarchical_reclaim(memcg
, NULL
, GFP_KERNEL
,
2795 MEM_CGROUP_RECLAIM_SHRINK
);
2796 curusage
= res_counter_read_u64(&memcg
->res
, RES_USAGE
);
2797 /* Usage is reduced ? */
2798 if (curusage
>= oldusage
)
2801 oldusage
= curusage
;
2803 if (!ret
&& enlarge
)
2804 memcg_oom_recover(memcg
);
2809 static int mem_cgroup_resize_memsw_limit(struct mem_cgroup
*memcg
,
2810 unsigned long long val
)
2813 u64 memlimit
, memswlimit
, oldusage
, curusage
;
2814 int children
= mem_cgroup_count_children(memcg
);
2818 /* see mem_cgroup_resize_res_limit */
2819 retry_count
= children
* MEM_CGROUP_RECLAIM_RETRIES
;
2820 oldusage
= res_counter_read_u64(&memcg
->memsw
, RES_USAGE
);
2821 while (retry_count
) {
2822 if (signal_pending(current
)) {
2827 * Rather than hide all in some function, I do this in
2828 * open coded manner. You see what this really does.
2829 * We have to guarantee mem->res.limit < mem->memsw.limit.
2831 mutex_lock(&set_limit_mutex
);
2832 memlimit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
2833 if (memlimit
> val
) {
2835 mutex_unlock(&set_limit_mutex
);
2838 memswlimit
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
2839 if (memswlimit
< val
)
2841 ret
= res_counter_set_limit(&memcg
->memsw
, val
);
2843 if (memlimit
== val
)
2844 memcg
->memsw_is_minimum
= true;
2846 memcg
->memsw_is_minimum
= false;
2848 mutex_unlock(&set_limit_mutex
);
2853 mem_cgroup_hierarchical_reclaim(memcg
, NULL
, GFP_KERNEL
,
2854 MEM_CGROUP_RECLAIM_NOSWAP
|
2855 MEM_CGROUP_RECLAIM_SHRINK
);
2856 curusage
= res_counter_read_u64(&memcg
->memsw
, RES_USAGE
);
2857 /* Usage is reduced ? */
2858 if (curusage
>= oldusage
)
2861 oldusage
= curusage
;
2863 if (!ret
&& enlarge
)
2864 memcg_oom_recover(memcg
);
2868 unsigned long mem_cgroup_soft_limit_reclaim(struct zone
*zone
, int order
,
2871 unsigned long nr_reclaimed
= 0;
2872 struct mem_cgroup_per_zone
*mz
, *next_mz
= NULL
;
2873 unsigned long reclaimed
;
2875 struct mem_cgroup_tree_per_zone
*mctz
;
2876 unsigned long long excess
;
2881 mctz
= soft_limit_tree_node_zone(zone_to_nid(zone
), zone_idx(zone
));
2883 * This loop can run a while, specially if mem_cgroup's continuously
2884 * keep exceeding their soft limit and putting the system under
2891 mz
= mem_cgroup_largest_soft_limit_node(mctz
);
2895 reclaimed
= mem_cgroup_hierarchical_reclaim(mz
->mem
, zone
,
2897 MEM_CGROUP_RECLAIM_SOFT
);
2898 nr_reclaimed
+= reclaimed
;
2899 spin_lock(&mctz
->lock
);
2902 * If we failed to reclaim anything from this memory cgroup
2903 * it is time to move on to the next cgroup
2909 * Loop until we find yet another one.
2911 * By the time we get the soft_limit lock
2912 * again, someone might have aded the
2913 * group back on the RB tree. Iterate to
2914 * make sure we get a different mem.
2915 * mem_cgroup_largest_soft_limit_node returns
2916 * NULL if no other cgroup is present on
2920 __mem_cgroup_largest_soft_limit_node(mctz
);
2921 if (next_mz
== mz
) {
2922 css_put(&next_mz
->mem
->css
);
2924 } else /* next_mz == NULL or other memcg */
2928 __mem_cgroup_remove_exceeded(mz
->mem
, mz
, mctz
);
2929 excess
= res_counter_soft_limit_excess(&mz
->mem
->res
);
2931 * One school of thought says that we should not add
2932 * back the node to the tree if reclaim returns 0.
2933 * But our reclaim could return 0, simply because due
2934 * to priority we are exposing a smaller subset of
2935 * memory to reclaim from. Consider this as a longer
2938 /* If excess == 0, no tree ops */
2939 __mem_cgroup_insert_exceeded(mz
->mem
, mz
, mctz
, excess
);
2940 spin_unlock(&mctz
->lock
);
2941 css_put(&mz
->mem
->css
);
2944 * Could not reclaim anything and there are no more
2945 * mem cgroups to try or we seem to be looping without
2946 * reclaiming anything.
2948 if (!nr_reclaimed
&&
2950 loop
> MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS
))
2952 } while (!nr_reclaimed
);
2954 css_put(&next_mz
->mem
->css
);
2955 return nr_reclaimed
;
2959 * This routine traverse page_cgroup in given list and drop them all.
2960 * *And* this routine doesn't reclaim page itself, just removes page_cgroup.
2962 static int mem_cgroup_force_empty_list(struct mem_cgroup
*mem
,
2963 int node
, int zid
, enum lru_list lru
)
2966 struct mem_cgroup_per_zone
*mz
;
2967 struct page_cgroup
*pc
, *busy
;
2968 unsigned long flags
, loop
;
2969 struct list_head
*list
;
2972 zone
= &NODE_DATA(node
)->node_zones
[zid
];
2973 mz
= mem_cgroup_zoneinfo(mem
, node
, zid
);
2974 list
= &mz
->lists
[lru
];
2976 loop
= MEM_CGROUP_ZSTAT(mz
, lru
);
2977 /* give some margin against EBUSY etc...*/
2982 spin_lock_irqsave(&zone
->lru_lock
, flags
);
2983 if (list_empty(list
)) {
2984 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
2987 pc
= list_entry(list
->prev
, struct page_cgroup
, lru
);
2989 list_move(&pc
->lru
, list
);
2991 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
2994 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
2996 ret
= mem_cgroup_move_parent(pc
, mem
, GFP_KERNEL
);
3000 if (ret
== -EBUSY
|| ret
== -EINVAL
) {
3001 /* found lock contention or "pc" is obsolete. */
3008 if (!ret
&& !list_empty(list
))
3014 * make mem_cgroup's charge to be 0 if there is no task.
3015 * This enables deleting this mem_cgroup.
3017 static int mem_cgroup_force_empty(struct mem_cgroup
*mem
, bool free_all
)
3020 int node
, zid
, shrink
;
3021 int nr_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
3022 struct cgroup
*cgrp
= mem
->css
.cgroup
;
3027 /* should free all ? */
3033 if (cgroup_task_count(cgrp
) || !list_empty(&cgrp
->children
))
3036 if (signal_pending(current
))
3038 /* This is for making all *used* pages to be on LRU. */
3039 lru_add_drain_all();
3040 drain_all_stock_sync();
3042 for_each_node_state(node
, N_HIGH_MEMORY
) {
3043 for (zid
= 0; !ret
&& zid
< MAX_NR_ZONES
; zid
++) {
3046 ret
= mem_cgroup_force_empty_list(mem
,
3055 memcg_oom_recover(mem
);
3056 /* it seems parent cgroup doesn't have enough mem */
3060 /* "ret" should also be checked to ensure all lists are empty. */
3061 } while (mem
->res
.usage
> 0 || ret
);
3067 /* returns EBUSY if there is a task or if we come here twice. */
3068 if (cgroup_task_count(cgrp
) || !list_empty(&cgrp
->children
) || shrink
) {
3072 /* we call try-to-free pages for make this cgroup empty */
3073 lru_add_drain_all();
3074 /* try to free all pages in this cgroup */
3076 while (nr_retries
&& mem
->res
.usage
> 0) {
3079 if (signal_pending(current
)) {
3083 progress
= try_to_free_mem_cgroup_pages(mem
, GFP_KERNEL
,
3084 false, get_swappiness(mem
));
3087 /* maybe some writeback is necessary */
3088 congestion_wait(BLK_RW_ASYNC
, HZ
/10);
3093 /* try move_account...there may be some *locked* pages. */
3097 int mem_cgroup_force_empty_write(struct cgroup
*cont
, unsigned int event
)
3099 return mem_cgroup_force_empty(mem_cgroup_from_cont(cont
), true);
3103 static u64
mem_cgroup_hierarchy_read(struct cgroup
*cont
, struct cftype
*cft
)
3105 return mem_cgroup_from_cont(cont
)->use_hierarchy
;
3108 static int mem_cgroup_hierarchy_write(struct cgroup
*cont
, struct cftype
*cft
,
3112 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cont
);
3113 struct cgroup
*parent
= cont
->parent
;
3114 struct mem_cgroup
*parent_mem
= NULL
;
3117 parent_mem
= mem_cgroup_from_cont(parent
);
3121 * If parent's use_hierarchy is set, we can't make any modifications
3122 * in the child subtrees. If it is unset, then the change can
3123 * occur, provided the current cgroup has no children.
3125 * For the root cgroup, parent_mem is NULL, we allow value to be
3126 * set if there are no children.
3128 if ((!parent_mem
|| !parent_mem
->use_hierarchy
) &&
3129 (val
== 1 || val
== 0)) {
3130 if (list_empty(&cont
->children
))
3131 mem
->use_hierarchy
= val
;
3141 struct mem_cgroup_idx_data
{
3143 enum mem_cgroup_stat_index idx
;
3147 mem_cgroup_get_idx_stat(struct mem_cgroup
*mem
, void *data
)
3149 struct mem_cgroup_idx_data
*d
= data
;
3150 d
->val
+= mem_cgroup_read_stat(mem
, d
->idx
);
3155 mem_cgroup_get_recursive_idx_stat(struct mem_cgroup
*mem
,
3156 enum mem_cgroup_stat_index idx
, s64
*val
)
3158 struct mem_cgroup_idx_data d
;
3161 mem_cgroup_walk_tree(mem
, &d
, mem_cgroup_get_idx_stat
);
3165 static inline u64
mem_cgroup_usage(struct mem_cgroup
*mem
, bool swap
)
3169 if (!mem_cgroup_is_root(mem
)) {
3171 return res_counter_read_u64(&mem
->res
, RES_USAGE
);
3173 return res_counter_read_u64(&mem
->memsw
, RES_USAGE
);
3176 mem_cgroup_get_recursive_idx_stat(mem
, MEM_CGROUP_STAT_CACHE
, &idx_val
);
3178 mem_cgroup_get_recursive_idx_stat(mem
, MEM_CGROUP_STAT_RSS
, &idx_val
);
3182 mem_cgroup_get_recursive_idx_stat(mem
,
3183 MEM_CGROUP_STAT_SWAPOUT
, &idx_val
);
3187 return val
<< PAGE_SHIFT
;
3190 static u64
mem_cgroup_read(struct cgroup
*cont
, struct cftype
*cft
)
3192 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cont
);
3196 type
= MEMFILE_TYPE(cft
->private);
3197 name
= MEMFILE_ATTR(cft
->private);
3200 if (name
== RES_USAGE
)
3201 val
= mem_cgroup_usage(mem
, false);
3203 val
= res_counter_read_u64(&mem
->res
, name
);
3206 if (name
== RES_USAGE
)
3207 val
= mem_cgroup_usage(mem
, true);
3209 val
= res_counter_read_u64(&mem
->memsw
, name
);
3218 * The user of this function is...
3221 static int mem_cgroup_write(struct cgroup
*cont
, struct cftype
*cft
,
3224 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
3226 unsigned long long val
;
3229 type
= MEMFILE_TYPE(cft
->private);
3230 name
= MEMFILE_ATTR(cft
->private);
3233 if (mem_cgroup_is_root(memcg
)) { /* Can't set limit on root */
3237 /* This function does all necessary parse...reuse it */
3238 ret
= res_counter_memparse_write_strategy(buffer
, &val
);
3242 ret
= mem_cgroup_resize_limit(memcg
, val
);
3244 ret
= mem_cgroup_resize_memsw_limit(memcg
, val
);
3246 case RES_SOFT_LIMIT
:
3247 ret
= res_counter_memparse_write_strategy(buffer
, &val
);
3251 * For memsw, soft limits are hard to implement in terms
3252 * of semantics, for now, we support soft limits for
3253 * control without swap
3256 ret
= res_counter_set_soft_limit(&memcg
->res
, val
);
3261 ret
= -EINVAL
; /* should be BUG() ? */
3267 static void memcg_get_hierarchical_limit(struct mem_cgroup
*memcg
,
3268 unsigned long long *mem_limit
, unsigned long long *memsw_limit
)
3270 struct cgroup
*cgroup
;
3271 unsigned long long min_limit
, min_memsw_limit
, tmp
;
3273 min_limit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
3274 min_memsw_limit
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
3275 cgroup
= memcg
->css
.cgroup
;
3276 if (!memcg
->use_hierarchy
)
3279 while (cgroup
->parent
) {
3280 cgroup
= cgroup
->parent
;
3281 memcg
= mem_cgroup_from_cont(cgroup
);
3282 if (!memcg
->use_hierarchy
)
3284 tmp
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
3285 min_limit
= min(min_limit
, tmp
);
3286 tmp
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
3287 min_memsw_limit
= min(min_memsw_limit
, tmp
);
3290 *mem_limit
= min_limit
;
3291 *memsw_limit
= min_memsw_limit
;
3295 static int mem_cgroup_reset(struct cgroup
*cont
, unsigned int event
)
3297 struct mem_cgroup
*mem
;
3300 mem
= mem_cgroup_from_cont(cont
);
3301 type
= MEMFILE_TYPE(event
);
3302 name
= MEMFILE_ATTR(event
);
3306 res_counter_reset_max(&mem
->res
);
3308 res_counter_reset_max(&mem
->memsw
);
3312 res_counter_reset_failcnt(&mem
->res
);
3314 res_counter_reset_failcnt(&mem
->memsw
);
3321 static u64
mem_cgroup_move_charge_read(struct cgroup
*cgrp
,
3324 return mem_cgroup_from_cont(cgrp
)->move_charge_at_immigrate
;
3328 static int mem_cgroup_move_charge_write(struct cgroup
*cgrp
,
3329 struct cftype
*cft
, u64 val
)
3331 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cgrp
);
3333 if (val
>= (1 << NR_MOVE_TYPE
))
3336 * We check this value several times in both in can_attach() and
3337 * attach(), so we need cgroup lock to prevent this value from being
3341 mem
->move_charge_at_immigrate
= val
;
3347 static int mem_cgroup_move_charge_write(struct cgroup
*cgrp
,
3348 struct cftype
*cft
, u64 val
)
3355 /* For read statistics */
3371 struct mcs_total_stat
{
3372 s64 stat
[NR_MCS_STAT
];
3378 } memcg_stat_strings
[NR_MCS_STAT
] = {
3379 {"cache", "total_cache"},
3380 {"rss", "total_rss"},
3381 {"mapped_file", "total_mapped_file"},
3382 {"pgpgin", "total_pgpgin"},
3383 {"pgpgout", "total_pgpgout"},
3384 {"swap", "total_swap"},
3385 {"inactive_anon", "total_inactive_anon"},
3386 {"active_anon", "total_active_anon"},
3387 {"inactive_file", "total_inactive_file"},
3388 {"active_file", "total_active_file"},
3389 {"unevictable", "total_unevictable"}
3393 static int mem_cgroup_get_local_stat(struct mem_cgroup
*mem
, void *data
)
3395 struct mcs_total_stat
*s
= data
;
3399 val
= mem_cgroup_read_stat(mem
, MEM_CGROUP_STAT_CACHE
);
3400 s
->stat
[MCS_CACHE
] += val
* PAGE_SIZE
;
3401 val
= mem_cgroup_read_stat(mem
, MEM_CGROUP_STAT_RSS
);
3402 s
->stat
[MCS_RSS
] += val
* PAGE_SIZE
;
3403 val
= mem_cgroup_read_stat(mem
, MEM_CGROUP_STAT_FILE_MAPPED
);
3404 s
->stat
[MCS_FILE_MAPPED
] += val
* PAGE_SIZE
;
3405 val
= mem_cgroup_read_stat(mem
, MEM_CGROUP_STAT_PGPGIN_COUNT
);
3406 s
->stat
[MCS_PGPGIN
] += val
;
3407 val
= mem_cgroup_read_stat(mem
, MEM_CGROUP_STAT_PGPGOUT_COUNT
);
3408 s
->stat
[MCS_PGPGOUT
] += val
;
3409 if (do_swap_account
) {
3410 val
= mem_cgroup_read_stat(mem
, MEM_CGROUP_STAT_SWAPOUT
);
3411 s
->stat
[MCS_SWAP
] += val
* PAGE_SIZE
;
3415 val
= mem_cgroup_get_local_zonestat(mem
, LRU_INACTIVE_ANON
);
3416 s
->stat
[MCS_INACTIVE_ANON
] += val
* PAGE_SIZE
;
3417 val
= mem_cgroup_get_local_zonestat(mem
, LRU_ACTIVE_ANON
);
3418 s
->stat
[MCS_ACTIVE_ANON
] += val
* PAGE_SIZE
;
3419 val
= mem_cgroup_get_local_zonestat(mem
, LRU_INACTIVE_FILE
);
3420 s
->stat
[MCS_INACTIVE_FILE
] += val
* PAGE_SIZE
;
3421 val
= mem_cgroup_get_local_zonestat(mem
, LRU_ACTIVE_FILE
);
3422 s
->stat
[MCS_ACTIVE_FILE
] += val
* PAGE_SIZE
;
3423 val
= mem_cgroup_get_local_zonestat(mem
, LRU_UNEVICTABLE
);
3424 s
->stat
[MCS_UNEVICTABLE
] += val
* PAGE_SIZE
;
3429 mem_cgroup_get_total_stat(struct mem_cgroup
*mem
, struct mcs_total_stat
*s
)
3431 mem_cgroup_walk_tree(mem
, s
, mem_cgroup_get_local_stat
);
3434 static int mem_control_stat_show(struct cgroup
*cont
, struct cftype
*cft
,
3435 struct cgroup_map_cb
*cb
)
3437 struct mem_cgroup
*mem_cont
= mem_cgroup_from_cont(cont
);
3438 struct mcs_total_stat mystat
;
3441 memset(&mystat
, 0, sizeof(mystat
));
3442 mem_cgroup_get_local_stat(mem_cont
, &mystat
);
3444 for (i
= 0; i
< NR_MCS_STAT
; i
++) {
3445 if (i
== MCS_SWAP
&& !do_swap_account
)
3447 cb
->fill(cb
, memcg_stat_strings
[i
].local_name
, mystat
.stat
[i
]);
3450 /* Hierarchical information */
3452 unsigned long long limit
, memsw_limit
;
3453 memcg_get_hierarchical_limit(mem_cont
, &limit
, &memsw_limit
);
3454 cb
->fill(cb
, "hierarchical_memory_limit", limit
);
3455 if (do_swap_account
)
3456 cb
->fill(cb
, "hierarchical_memsw_limit", memsw_limit
);
3459 memset(&mystat
, 0, sizeof(mystat
));
3460 mem_cgroup_get_total_stat(mem_cont
, &mystat
);
3461 for (i
= 0; i
< NR_MCS_STAT
; i
++) {
3462 if (i
== MCS_SWAP
&& !do_swap_account
)
3464 cb
->fill(cb
, memcg_stat_strings
[i
].total_name
, mystat
.stat
[i
]);
3467 #ifdef CONFIG_DEBUG_VM
3468 cb
->fill(cb
, "inactive_ratio", calc_inactive_ratio(mem_cont
, NULL
));
3472 struct mem_cgroup_per_zone
*mz
;
3473 unsigned long recent_rotated
[2] = {0, 0};
3474 unsigned long recent_scanned
[2] = {0, 0};
3476 for_each_online_node(nid
)
3477 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++) {
3478 mz
= mem_cgroup_zoneinfo(mem_cont
, nid
, zid
);
3480 recent_rotated
[0] +=
3481 mz
->reclaim_stat
.recent_rotated
[0];
3482 recent_rotated
[1] +=
3483 mz
->reclaim_stat
.recent_rotated
[1];
3484 recent_scanned
[0] +=
3485 mz
->reclaim_stat
.recent_scanned
[0];
3486 recent_scanned
[1] +=
3487 mz
->reclaim_stat
.recent_scanned
[1];
3489 cb
->fill(cb
, "recent_rotated_anon", recent_rotated
[0]);
3490 cb
->fill(cb
, "recent_rotated_file", recent_rotated
[1]);
3491 cb
->fill(cb
, "recent_scanned_anon", recent_scanned
[0]);
3492 cb
->fill(cb
, "recent_scanned_file", recent_scanned
[1]);
3499 static u64
mem_cgroup_swappiness_read(struct cgroup
*cgrp
, struct cftype
*cft
)
3501 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
3503 return get_swappiness(memcg
);
3506 static int mem_cgroup_swappiness_write(struct cgroup
*cgrp
, struct cftype
*cft
,
3509 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
3510 struct mem_cgroup
*parent
;
3515 if (cgrp
->parent
== NULL
)
3518 parent
= mem_cgroup_from_cont(cgrp
->parent
);
3522 /* If under hierarchy, only empty-root can set this value */
3523 if ((parent
->use_hierarchy
) ||
3524 (memcg
->use_hierarchy
&& !list_empty(&cgrp
->children
))) {
3529 spin_lock(&memcg
->reclaim_param_lock
);
3530 memcg
->swappiness
= val
;
3531 spin_unlock(&memcg
->reclaim_param_lock
);
3538 static void __mem_cgroup_threshold(struct mem_cgroup
*memcg
, bool swap
)
3540 struct mem_cgroup_threshold_ary
*t
;
3546 t
= rcu_dereference(memcg
->thresholds
.primary
);
3548 t
= rcu_dereference(memcg
->memsw_thresholds
.primary
);
3553 usage
= mem_cgroup_usage(memcg
, swap
);
3556 * current_threshold points to threshold just below usage.
3557 * If it's not true, a threshold was crossed after last
3558 * call of __mem_cgroup_threshold().
3560 i
= t
->current_threshold
;
3563 * Iterate backward over array of thresholds starting from
3564 * current_threshold and check if a threshold is crossed.
3565 * If none of thresholds below usage is crossed, we read
3566 * only one element of the array here.
3568 for (; i
>= 0 && unlikely(t
->entries
[i
].threshold
> usage
); i
--)
3569 eventfd_signal(t
->entries
[i
].eventfd
, 1);
3571 /* i = current_threshold + 1 */
3575 * Iterate forward over array of thresholds starting from
3576 * current_threshold+1 and check if a threshold is crossed.
3577 * If none of thresholds above usage is crossed, we read
3578 * only one element of the array here.
3580 for (; i
< t
->size
&& unlikely(t
->entries
[i
].threshold
<= usage
); i
++)
3581 eventfd_signal(t
->entries
[i
].eventfd
, 1);
3583 /* Update current_threshold */
3584 t
->current_threshold
= i
- 1;
3589 static void mem_cgroup_threshold(struct mem_cgroup
*memcg
)
3592 __mem_cgroup_threshold(memcg
, false);
3593 if (do_swap_account
)
3594 __mem_cgroup_threshold(memcg
, true);
3596 memcg
= parent_mem_cgroup(memcg
);
3600 static int compare_thresholds(const void *a
, const void *b
)
3602 const struct mem_cgroup_threshold
*_a
= a
;
3603 const struct mem_cgroup_threshold
*_b
= b
;
3605 return _a
->threshold
- _b
->threshold
;
3608 static int mem_cgroup_oom_notify_cb(struct mem_cgroup
*mem
, void *data
)
3610 struct mem_cgroup_eventfd_list
*ev
;
3612 list_for_each_entry(ev
, &mem
->oom_notify
, list
)
3613 eventfd_signal(ev
->eventfd
, 1);
3617 static void mem_cgroup_oom_notify(struct mem_cgroup
*mem
)
3619 mem_cgroup_walk_tree(mem
, NULL
, mem_cgroup_oom_notify_cb
);
3622 static int mem_cgroup_usage_register_event(struct cgroup
*cgrp
,
3623 struct cftype
*cft
, struct eventfd_ctx
*eventfd
, const char *args
)
3625 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
3626 struct mem_cgroup_thresholds
*thresholds
;
3627 struct mem_cgroup_threshold_ary
*new;
3628 int type
= MEMFILE_TYPE(cft
->private);
3629 u64 threshold
, usage
;
3632 ret
= res_counter_memparse_write_strategy(args
, &threshold
);
3636 mutex_lock(&memcg
->thresholds_lock
);
3639 thresholds
= &memcg
->thresholds
;
3640 else if (type
== _MEMSWAP
)
3641 thresholds
= &memcg
->memsw_thresholds
;
3645 usage
= mem_cgroup_usage(memcg
, type
== _MEMSWAP
);
3647 /* Check if a threshold crossed before adding a new one */
3648 if (thresholds
->primary
)
3649 __mem_cgroup_threshold(memcg
, type
== _MEMSWAP
);
3651 size
= thresholds
->primary
? thresholds
->primary
->size
+ 1 : 1;
3653 /* Allocate memory for new array of thresholds */
3654 new = kmalloc(sizeof(*new) + size
* sizeof(struct mem_cgroup_threshold
),
3662 /* Copy thresholds (if any) to new array */
3663 if (thresholds
->primary
) {
3664 memcpy(new->entries
, thresholds
->primary
->entries
, (size
- 1) *
3665 sizeof(struct mem_cgroup_threshold
));
3668 /* Add new threshold */
3669 new->entries
[size
- 1].eventfd
= eventfd
;
3670 new->entries
[size
- 1].threshold
= threshold
;
3672 /* Sort thresholds. Registering of new threshold isn't time-critical */
3673 sort(new->entries
, size
, sizeof(struct mem_cgroup_threshold
),
3674 compare_thresholds
, NULL
);
3676 /* Find current threshold */
3677 new->current_threshold
= -1;
3678 for (i
= 0; i
< size
; i
++) {
3679 if (new->entries
[i
].threshold
< usage
) {
3681 * new->current_threshold will not be used until
3682 * rcu_assign_pointer(), so it's safe to increment
3685 ++new->current_threshold
;
3689 /* Free old spare buffer and save old primary buffer as spare */
3690 kfree(thresholds
->spare
);
3691 thresholds
->spare
= thresholds
->primary
;
3693 rcu_assign_pointer(thresholds
->primary
, new);
3695 /* To be sure that nobody uses thresholds */
3699 mutex_unlock(&memcg
->thresholds_lock
);
3704 static void mem_cgroup_usage_unregister_event(struct cgroup
*cgrp
,
3705 struct cftype
*cft
, struct eventfd_ctx
*eventfd
)
3707 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
3708 struct mem_cgroup_thresholds
*thresholds
;
3709 struct mem_cgroup_threshold_ary
*new;
3710 int type
= MEMFILE_TYPE(cft
->private);
3714 mutex_lock(&memcg
->thresholds_lock
);
3716 thresholds
= &memcg
->thresholds
;
3717 else if (type
== _MEMSWAP
)
3718 thresholds
= &memcg
->memsw_thresholds
;
3723 * Something went wrong if we trying to unregister a threshold
3724 * if we don't have thresholds
3726 BUG_ON(!thresholds
);
3728 usage
= mem_cgroup_usage(memcg
, type
== _MEMSWAP
);
3730 /* Check if a threshold crossed before removing */
3731 __mem_cgroup_threshold(memcg
, type
== _MEMSWAP
);
3733 /* Calculate new number of threshold */
3735 for (i
= 0; i
< thresholds
->primary
->size
; i
++) {
3736 if (thresholds
->primary
->entries
[i
].eventfd
!= eventfd
)
3740 new = thresholds
->spare
;
3742 /* Set thresholds array to NULL if we don't have thresholds */
3751 /* Copy thresholds and find current threshold */
3752 new->current_threshold
= -1;
3753 for (i
= 0, j
= 0; i
< thresholds
->primary
->size
; i
++) {
3754 if (thresholds
->primary
->entries
[i
].eventfd
== eventfd
)
3757 new->entries
[j
] = thresholds
->primary
->entries
[i
];
3758 if (new->entries
[j
].threshold
< usage
) {
3760 * new->current_threshold will not be used
3761 * until rcu_assign_pointer(), so it's safe to increment
3764 ++new->current_threshold
;
3770 /* Swap primary and spare array */
3771 thresholds
->spare
= thresholds
->primary
;
3772 rcu_assign_pointer(thresholds
->primary
, new);
3774 /* To be sure that nobody uses thresholds */
3777 mutex_unlock(&memcg
->thresholds_lock
);
3780 static int mem_cgroup_oom_register_event(struct cgroup
*cgrp
,
3781 struct cftype
*cft
, struct eventfd_ctx
*eventfd
, const char *args
)
3783 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
3784 struct mem_cgroup_eventfd_list
*event
;
3785 int type
= MEMFILE_TYPE(cft
->private);
3787 BUG_ON(type
!= _OOM_TYPE
);
3788 event
= kmalloc(sizeof(*event
), GFP_KERNEL
);
3792 mutex_lock(&memcg_oom_mutex
);
3794 event
->eventfd
= eventfd
;
3795 list_add(&event
->list
, &memcg
->oom_notify
);
3797 /* already in OOM ? */
3798 if (atomic_read(&memcg
->oom_lock
))
3799 eventfd_signal(eventfd
, 1);
3800 mutex_unlock(&memcg_oom_mutex
);
3805 static void mem_cgroup_oom_unregister_event(struct cgroup
*cgrp
,
3806 struct cftype
*cft
, struct eventfd_ctx
*eventfd
)
3808 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cgrp
);
3809 struct mem_cgroup_eventfd_list
*ev
, *tmp
;
3810 int type
= MEMFILE_TYPE(cft
->private);
3812 BUG_ON(type
!= _OOM_TYPE
);
3814 mutex_lock(&memcg_oom_mutex
);
3816 list_for_each_entry_safe(ev
, tmp
, &mem
->oom_notify
, list
) {
3817 if (ev
->eventfd
== eventfd
) {
3818 list_del(&ev
->list
);
3823 mutex_unlock(&memcg_oom_mutex
);
3826 static int mem_cgroup_oom_control_read(struct cgroup
*cgrp
,
3827 struct cftype
*cft
, struct cgroup_map_cb
*cb
)
3829 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cgrp
);
3831 cb
->fill(cb
, "oom_kill_disable", mem
->oom_kill_disable
);
3833 if (atomic_read(&mem
->oom_lock
))
3834 cb
->fill(cb
, "under_oom", 1);
3836 cb
->fill(cb
, "under_oom", 0);
3840 static int mem_cgroup_oom_control_write(struct cgroup
*cgrp
,
3841 struct cftype
*cft
, u64 val
)
3843 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cgrp
);
3844 struct mem_cgroup
*parent
;
3846 /* cannot set to root cgroup and only 0 and 1 are allowed */
3847 if (!cgrp
->parent
|| !((val
== 0) || (val
== 1)))
3850 parent
= mem_cgroup_from_cont(cgrp
->parent
);
3853 /* oom-kill-disable is a flag for subhierarchy. */
3854 if ((parent
->use_hierarchy
) ||
3855 (mem
->use_hierarchy
&& !list_empty(&cgrp
->children
))) {
3859 mem
->oom_kill_disable
= val
;
3861 memcg_oom_recover(mem
);
3866 static struct cftype mem_cgroup_files
[] = {
3868 .name
= "usage_in_bytes",
3869 .private = MEMFILE_PRIVATE(_MEM
, RES_USAGE
),
3870 .read_u64
= mem_cgroup_read
,
3871 .register_event
= mem_cgroup_usage_register_event
,
3872 .unregister_event
= mem_cgroup_usage_unregister_event
,
3875 .name
= "max_usage_in_bytes",
3876 .private = MEMFILE_PRIVATE(_MEM
, RES_MAX_USAGE
),
3877 .trigger
= mem_cgroup_reset
,
3878 .read_u64
= mem_cgroup_read
,
3881 .name
= "limit_in_bytes",
3882 .private = MEMFILE_PRIVATE(_MEM
, RES_LIMIT
),
3883 .write_string
= mem_cgroup_write
,
3884 .read_u64
= mem_cgroup_read
,
3887 .name
= "soft_limit_in_bytes",
3888 .private = MEMFILE_PRIVATE(_MEM
, RES_SOFT_LIMIT
),
3889 .write_string
= mem_cgroup_write
,
3890 .read_u64
= mem_cgroup_read
,
3894 .private = MEMFILE_PRIVATE(_MEM
, RES_FAILCNT
),
3895 .trigger
= mem_cgroup_reset
,
3896 .read_u64
= mem_cgroup_read
,
3900 .read_map
= mem_control_stat_show
,
3903 .name
= "force_empty",
3904 .trigger
= mem_cgroup_force_empty_write
,
3907 .name
= "use_hierarchy",
3908 .write_u64
= mem_cgroup_hierarchy_write
,
3909 .read_u64
= mem_cgroup_hierarchy_read
,
3912 .name
= "swappiness",
3913 .read_u64
= mem_cgroup_swappiness_read
,
3914 .write_u64
= mem_cgroup_swappiness_write
,
3917 .name
= "move_charge_at_immigrate",
3918 .read_u64
= mem_cgroup_move_charge_read
,
3919 .write_u64
= mem_cgroup_move_charge_write
,
3922 .name
= "oom_control",
3923 .read_map
= mem_cgroup_oom_control_read
,
3924 .write_u64
= mem_cgroup_oom_control_write
,
3925 .register_event
= mem_cgroup_oom_register_event
,
3926 .unregister_event
= mem_cgroup_oom_unregister_event
,
3927 .private = MEMFILE_PRIVATE(_OOM_TYPE
, OOM_CONTROL
),
3931 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
3932 static struct cftype memsw_cgroup_files
[] = {
3934 .name
= "memsw.usage_in_bytes",
3935 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_USAGE
),
3936 .read_u64
= mem_cgroup_read
,
3937 .register_event
= mem_cgroup_usage_register_event
,
3938 .unregister_event
= mem_cgroup_usage_unregister_event
,
3941 .name
= "memsw.max_usage_in_bytes",
3942 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_MAX_USAGE
),
3943 .trigger
= mem_cgroup_reset
,
3944 .read_u64
= mem_cgroup_read
,
3947 .name
= "memsw.limit_in_bytes",
3948 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_LIMIT
),
3949 .write_string
= mem_cgroup_write
,
3950 .read_u64
= mem_cgroup_read
,
3953 .name
= "memsw.failcnt",
3954 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_FAILCNT
),
3955 .trigger
= mem_cgroup_reset
,
3956 .read_u64
= mem_cgroup_read
,
3960 static int register_memsw_files(struct cgroup
*cont
, struct cgroup_subsys
*ss
)
3962 if (!do_swap_account
)
3964 return cgroup_add_files(cont
, ss
, memsw_cgroup_files
,
3965 ARRAY_SIZE(memsw_cgroup_files
));
3968 static int register_memsw_files(struct cgroup
*cont
, struct cgroup_subsys
*ss
)
3974 static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup
*mem
, int node
)
3976 struct mem_cgroup_per_node
*pn
;
3977 struct mem_cgroup_per_zone
*mz
;
3979 int zone
, tmp
= node
;
3981 * This routine is called against possible nodes.
3982 * But it's BUG to call kmalloc() against offline node.
3984 * TODO: this routine can waste much memory for nodes which will
3985 * never be onlined. It's better to use memory hotplug callback
3988 if (!node_state(node
, N_NORMAL_MEMORY
))
3990 pn
= kmalloc_node(sizeof(*pn
), GFP_KERNEL
, tmp
);
3994 mem
->info
.nodeinfo
[node
] = pn
;
3995 memset(pn
, 0, sizeof(*pn
));
3997 for (zone
= 0; zone
< MAX_NR_ZONES
; zone
++) {
3998 mz
= &pn
->zoneinfo
[zone
];
4000 INIT_LIST_HEAD(&mz
->lists
[l
]);
4001 mz
->usage_in_excess
= 0;
4002 mz
->on_tree
= false;
4008 static void free_mem_cgroup_per_zone_info(struct mem_cgroup
*mem
, int node
)
4010 kfree(mem
->info
.nodeinfo
[node
]);
4013 static struct mem_cgroup
*mem_cgroup_alloc(void)
4015 struct mem_cgroup
*mem
;
4016 int size
= sizeof(struct mem_cgroup
);
4018 /* Can be very big if MAX_NUMNODES is very big */
4019 if (size
< PAGE_SIZE
)
4020 mem
= kmalloc(size
, GFP_KERNEL
);
4022 mem
= vmalloc(size
);
4027 memset(mem
, 0, size
);
4028 mem
->stat
= alloc_percpu(struct mem_cgroup_stat_cpu
);
4030 if (size
< PAGE_SIZE
)
4040 * At destroying mem_cgroup, references from swap_cgroup can remain.
4041 * (scanning all at force_empty is too costly...)
4043 * Instead of clearing all references at force_empty, we remember
4044 * the number of reference from swap_cgroup and free mem_cgroup when
4045 * it goes down to 0.
4047 * Removal of cgroup itself succeeds regardless of refs from swap.
4050 static void __mem_cgroup_free(struct mem_cgroup
*mem
)
4054 mem_cgroup_remove_from_trees(mem
);
4055 free_css_id(&mem_cgroup_subsys
, &mem
->css
);
4057 for_each_node_state(node
, N_POSSIBLE
)
4058 free_mem_cgroup_per_zone_info(mem
, node
);
4060 free_percpu(mem
->stat
);
4061 if (sizeof(struct mem_cgroup
) < PAGE_SIZE
)
4067 static void mem_cgroup_get(struct mem_cgroup
*mem
)
4069 atomic_inc(&mem
->refcnt
);
4072 static void __mem_cgroup_put(struct mem_cgroup
*mem
, int count
)
4074 if (atomic_sub_and_test(count
, &mem
->refcnt
)) {
4075 struct mem_cgroup
*parent
= parent_mem_cgroup(mem
);
4076 __mem_cgroup_free(mem
);
4078 mem_cgroup_put(parent
);
4082 static void mem_cgroup_put(struct mem_cgroup
*mem
)
4084 __mem_cgroup_put(mem
, 1);
4088 * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled.
4090 static struct mem_cgroup
*parent_mem_cgroup(struct mem_cgroup
*mem
)
4092 if (!mem
->res
.parent
)
4094 return mem_cgroup_from_res_counter(mem
->res
.parent
, res
);
4097 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
4098 static void __init
enable_swap_cgroup(void)
4100 if (!mem_cgroup_disabled() && really_do_swap_account
)
4101 do_swap_account
= 1;
4104 static void __init
enable_swap_cgroup(void)
4109 static int mem_cgroup_soft_limit_tree_init(void)
4111 struct mem_cgroup_tree_per_node
*rtpn
;
4112 struct mem_cgroup_tree_per_zone
*rtpz
;
4113 int tmp
, node
, zone
;
4115 for_each_node_state(node
, N_POSSIBLE
) {
4117 if (!node_state(node
, N_NORMAL_MEMORY
))
4119 rtpn
= kzalloc_node(sizeof(*rtpn
), GFP_KERNEL
, tmp
);
4123 soft_limit_tree
.rb_tree_per_node
[node
] = rtpn
;
4125 for (zone
= 0; zone
< MAX_NR_ZONES
; zone
++) {
4126 rtpz
= &rtpn
->rb_tree_per_zone
[zone
];
4127 rtpz
->rb_root
= RB_ROOT
;
4128 spin_lock_init(&rtpz
->lock
);
4134 static struct cgroup_subsys_state
* __ref
4135 mem_cgroup_create(struct cgroup_subsys
*ss
, struct cgroup
*cont
)
4137 struct mem_cgroup
*mem
, *parent
;
4138 long error
= -ENOMEM
;
4141 mem
= mem_cgroup_alloc();
4143 return ERR_PTR(error
);
4145 for_each_node_state(node
, N_POSSIBLE
)
4146 if (alloc_mem_cgroup_per_zone_info(mem
, node
))
4150 if (cont
->parent
== NULL
) {
4152 enable_swap_cgroup();
4154 root_mem_cgroup
= mem
;
4155 if (mem_cgroup_soft_limit_tree_init())
4157 for_each_possible_cpu(cpu
) {
4158 struct memcg_stock_pcp
*stock
=
4159 &per_cpu(memcg_stock
, cpu
);
4160 INIT_WORK(&stock
->work
, drain_local_stock
);
4162 hotcpu_notifier(memcg_stock_cpu_callback
, 0);
4164 parent
= mem_cgroup_from_cont(cont
->parent
);
4165 mem
->use_hierarchy
= parent
->use_hierarchy
;
4166 mem
->oom_kill_disable
= parent
->oom_kill_disable
;
4169 if (parent
&& parent
->use_hierarchy
) {
4170 res_counter_init(&mem
->res
, &parent
->res
);
4171 res_counter_init(&mem
->memsw
, &parent
->memsw
);
4173 * We increment refcnt of the parent to ensure that we can
4174 * safely access it on res_counter_charge/uncharge.
4175 * This refcnt will be decremented when freeing this
4176 * mem_cgroup(see mem_cgroup_put).
4178 mem_cgroup_get(parent
);
4180 res_counter_init(&mem
->res
, NULL
);
4181 res_counter_init(&mem
->memsw
, NULL
);
4183 mem
->last_scanned_child
= 0;
4184 spin_lock_init(&mem
->reclaim_param_lock
);
4185 INIT_LIST_HEAD(&mem
->oom_notify
);
4188 mem
->swappiness
= get_swappiness(parent
);
4189 atomic_set(&mem
->refcnt
, 1);
4190 mem
->move_charge_at_immigrate
= 0;
4191 mutex_init(&mem
->thresholds_lock
);
4194 __mem_cgroup_free(mem
);
4195 root_mem_cgroup
= NULL
;
4196 return ERR_PTR(error
);
4199 static int mem_cgroup_pre_destroy(struct cgroup_subsys
*ss
,
4200 struct cgroup
*cont
)
4202 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cont
);
4204 return mem_cgroup_force_empty(mem
, false);
4207 static void mem_cgroup_destroy(struct cgroup_subsys
*ss
,
4208 struct cgroup
*cont
)
4210 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cont
);
4212 mem_cgroup_put(mem
);
4215 static int mem_cgroup_populate(struct cgroup_subsys
*ss
,
4216 struct cgroup
*cont
)
4220 ret
= cgroup_add_files(cont
, ss
, mem_cgroup_files
,
4221 ARRAY_SIZE(mem_cgroup_files
));
4224 ret
= register_memsw_files(cont
, ss
);
4229 /* Handlers for move charge at task migration. */
4230 #define PRECHARGE_COUNT_AT_ONCE 256
4231 static int mem_cgroup_do_precharge(unsigned long count
)
4234 int batch_count
= PRECHARGE_COUNT_AT_ONCE
;
4235 struct mem_cgroup
*mem
= mc
.to
;
4237 if (mem_cgroup_is_root(mem
)) {
4238 mc
.precharge
+= count
;
4239 /* we don't need css_get for root */
4242 /* try to charge at once */
4244 struct res_counter
*dummy
;
4246 * "mem" cannot be under rmdir() because we've already checked
4247 * by cgroup_lock_live_cgroup() that it is not removed and we
4248 * are still under the same cgroup_mutex. So we can postpone
4251 if (res_counter_charge(&mem
->res
, PAGE_SIZE
* count
, &dummy
))
4253 if (do_swap_account
&& res_counter_charge(&mem
->memsw
,
4254 PAGE_SIZE
* count
, &dummy
)) {
4255 res_counter_uncharge(&mem
->res
, PAGE_SIZE
* count
);
4258 mc
.precharge
+= count
;
4262 /* fall back to one by one charge */
4264 if (signal_pending(current
)) {
4268 if (!batch_count
--) {
4269 batch_count
= PRECHARGE_COUNT_AT_ONCE
;
4272 ret
= __mem_cgroup_try_charge(NULL
, GFP_KERNEL
, &mem
, false);
4274 /* mem_cgroup_clear_mc() will do uncharge later */
4282 * is_target_pte_for_mc - check a pte whether it is valid for move charge
4283 * @vma: the vma the pte to be checked belongs
4284 * @addr: the address corresponding to the pte to be checked
4285 * @ptent: the pte to be checked
4286 * @target: the pointer the target page or swap ent will be stored(can be NULL)
4289 * 0(MC_TARGET_NONE): if the pte is not a target for move charge.
4290 * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
4291 * move charge. if @target is not NULL, the page is stored in target->page
4292 * with extra refcnt got(Callers should handle it).
4293 * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
4294 * target for charge migration. if @target is not NULL, the entry is stored
4297 * Called with pte lock held.
4304 enum mc_target_type
{
4305 MC_TARGET_NONE
, /* not used */
4310 static struct page
*mc_handle_present_pte(struct vm_area_struct
*vma
,
4311 unsigned long addr
, pte_t ptent
)
4313 struct page
*page
= vm_normal_page(vma
, addr
, ptent
);
4315 if (!page
|| !page_mapped(page
))
4317 if (PageAnon(page
)) {
4318 /* we don't move shared anon */
4319 if (!move_anon() || page_mapcount(page
) > 2)
4321 } else if (!move_file())
4322 /* we ignore mapcount for file pages */
4324 if (!get_page_unless_zero(page
))
4330 static struct page
*mc_handle_swap_pte(struct vm_area_struct
*vma
,
4331 unsigned long addr
, pte_t ptent
, swp_entry_t
*entry
)
4334 struct page
*page
= NULL
;
4335 swp_entry_t ent
= pte_to_swp_entry(ptent
);
4337 if (!move_anon() || non_swap_entry(ent
))
4339 usage_count
= mem_cgroup_count_swap_user(ent
, &page
);
4340 if (usage_count
> 1) { /* we don't move shared anon */
4345 if (do_swap_account
)
4346 entry
->val
= ent
.val
;
4351 static struct page
*mc_handle_file_pte(struct vm_area_struct
*vma
,
4352 unsigned long addr
, pte_t ptent
, swp_entry_t
*entry
)
4354 struct page
*page
= NULL
;
4355 struct inode
*inode
;
4356 struct address_space
*mapping
;
4359 if (!vma
->vm_file
) /* anonymous vma */
4364 inode
= vma
->vm_file
->f_path
.dentry
->d_inode
;
4365 mapping
= vma
->vm_file
->f_mapping
;
4366 if (pte_none(ptent
))
4367 pgoff
= linear_page_index(vma
, addr
);
4368 else /* pte_file(ptent) is true */
4369 pgoff
= pte_to_pgoff(ptent
);
4371 /* page is moved even if it's not RSS of this task(page-faulted). */
4372 if (!mapping_cap_swap_backed(mapping
)) { /* normal file */
4373 page
= find_get_page(mapping
, pgoff
);
4374 } else { /* shmem/tmpfs file. we should take account of swap too. */
4376 mem_cgroup_get_shmem_target(inode
, pgoff
, &page
, &ent
);
4377 if (do_swap_account
)
4378 entry
->val
= ent
.val
;
4384 static int is_target_pte_for_mc(struct vm_area_struct
*vma
,
4385 unsigned long addr
, pte_t ptent
, union mc_target
*target
)
4387 struct page
*page
= NULL
;
4388 struct page_cgroup
*pc
;
4390 swp_entry_t ent
= { .val
= 0 };
4392 if (pte_present(ptent
))
4393 page
= mc_handle_present_pte(vma
, addr
, ptent
);
4394 else if (is_swap_pte(ptent
))
4395 page
= mc_handle_swap_pte(vma
, addr
, ptent
, &ent
);
4396 else if (pte_none(ptent
) || pte_file(ptent
))
4397 page
= mc_handle_file_pte(vma
, addr
, ptent
, &ent
);
4399 if (!page
&& !ent
.val
)
4402 pc
= lookup_page_cgroup(page
);
4404 * Do only loose check w/o page_cgroup lock.
4405 * mem_cgroup_move_account() checks the pc is valid or not under
4408 if (PageCgroupUsed(pc
) && pc
->mem_cgroup
== mc
.from
) {
4409 ret
= MC_TARGET_PAGE
;
4411 target
->page
= page
;
4413 if (!ret
|| !target
)
4416 /* There is a swap entry and a page doesn't exist or isn't charged */
4417 if (ent
.val
&& !ret
&&
4418 css_id(&mc
.from
->css
) == lookup_swap_cgroup(ent
)) {
4419 ret
= MC_TARGET_SWAP
;
4426 static int mem_cgroup_count_precharge_pte_range(pmd_t
*pmd
,
4427 unsigned long addr
, unsigned long end
,
4428 struct mm_walk
*walk
)
4430 struct vm_area_struct
*vma
= walk
->private;
4434 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
4435 for (; addr
!= end
; pte
++, addr
+= PAGE_SIZE
)
4436 if (is_target_pte_for_mc(vma
, addr
, *pte
, NULL
))
4437 mc
.precharge
++; /* increment precharge temporarily */
4438 pte_unmap_unlock(pte
- 1, ptl
);
4444 static unsigned long mem_cgroup_count_precharge(struct mm_struct
*mm
)
4446 unsigned long precharge
;
4447 struct vm_area_struct
*vma
;
4449 /* We've already held the mmap_sem */
4450 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
4451 struct mm_walk mem_cgroup_count_precharge_walk
= {
4452 .pmd_entry
= mem_cgroup_count_precharge_pte_range
,
4456 if (is_vm_hugetlb_page(vma
))
4458 walk_page_range(vma
->vm_start
, vma
->vm_end
,
4459 &mem_cgroup_count_precharge_walk
);
4462 precharge
= mc
.precharge
;
4468 static int mem_cgroup_precharge_mc(struct mm_struct
*mm
)
4470 return mem_cgroup_do_precharge(mem_cgroup_count_precharge(mm
));
4473 static void mem_cgroup_clear_mc(void)
4475 struct mem_cgroup
*from
= mc
.from
;
4476 struct mem_cgroup
*to
= mc
.to
;
4478 /* we must uncharge all the leftover precharges from mc.to */
4480 __mem_cgroup_cancel_charge(mc
.to
, mc
.precharge
);
4484 * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
4485 * we must uncharge here.
4487 if (mc
.moved_charge
) {
4488 __mem_cgroup_cancel_charge(mc
.from
, mc
.moved_charge
);
4489 mc
.moved_charge
= 0;
4491 /* we must fixup refcnts and charges */
4492 if (mc
.moved_swap
) {
4493 /* uncharge swap account from the old cgroup */
4494 if (!mem_cgroup_is_root(mc
.from
))
4495 res_counter_uncharge(&mc
.from
->memsw
,
4496 PAGE_SIZE
* mc
.moved_swap
);
4497 __mem_cgroup_put(mc
.from
, mc
.moved_swap
);
4499 if (!mem_cgroup_is_root(mc
.to
)) {
4501 * we charged both to->res and to->memsw, so we should
4504 res_counter_uncharge(&mc
.to
->res
,
4505 PAGE_SIZE
* mc
.moved_swap
);
4507 /* we've already done mem_cgroup_get(mc.to) */
4512 up_read(&mc
.mm
->mmap_sem
);
4515 spin_lock(&mc
.lock
);
4518 spin_unlock(&mc
.lock
);
4519 mc
.moving_task
= NULL
;
4521 memcg_oom_recover(from
);
4522 memcg_oom_recover(to
);
4523 wake_up_all(&mc
.waitq
);
4526 static int mem_cgroup_can_attach(struct cgroup_subsys
*ss
,
4527 struct cgroup
*cgroup
,
4528 struct task_struct
*p
,
4532 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cgroup
);
4534 if (mem
->move_charge_at_immigrate
) {
4535 struct mm_struct
*mm
;
4536 struct mem_cgroup
*from
= mem_cgroup_from_task(p
);
4538 VM_BUG_ON(from
== mem
);
4540 mm
= get_task_mm(p
);
4543 /* We move charges only when we move a owner of the mm */
4544 if (mm
->owner
== p
) {
4546 * We do all the move charge works under one mmap_sem to
4547 * avoid deadlock with down_write(&mmap_sem)
4548 * -> try_charge() -> if (mc.moving_task) -> sleep.
4550 down_read(&mm
->mmap_sem
);
4554 VM_BUG_ON(mc
.precharge
);
4555 VM_BUG_ON(mc
.moved_charge
);
4556 VM_BUG_ON(mc
.moved_swap
);
4557 VM_BUG_ON(mc
.moving_task
);
4560 spin_lock(&mc
.lock
);
4564 mc
.moved_charge
= 0;
4566 spin_unlock(&mc
.lock
);
4567 mc
.moving_task
= current
;
4570 ret
= mem_cgroup_precharge_mc(mm
);
4572 mem_cgroup_clear_mc();
4573 /* We call up_read() and mmput() in clear_mc(). */
4580 static void mem_cgroup_cancel_attach(struct cgroup_subsys
*ss
,
4581 struct cgroup
*cgroup
,
4582 struct task_struct
*p
,
4585 mem_cgroup_clear_mc();
4588 static int mem_cgroup_move_charge_pte_range(pmd_t
*pmd
,
4589 unsigned long addr
, unsigned long end
,
4590 struct mm_walk
*walk
)
4593 struct vm_area_struct
*vma
= walk
->private;
4598 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
4599 for (; addr
!= end
; addr
+= PAGE_SIZE
) {
4600 pte_t ptent
= *(pte
++);
4601 union mc_target target
;
4604 struct page_cgroup
*pc
;
4610 type
= is_target_pte_for_mc(vma
, addr
, ptent
, &target
);
4612 case MC_TARGET_PAGE
:
4614 if (isolate_lru_page(page
))
4616 pc
= lookup_page_cgroup(page
);
4617 if (!mem_cgroup_move_account(pc
,
4618 mc
.from
, mc
.to
, false)) {
4620 /* we uncharge from mc.from later. */
4623 putback_lru_page(page
);
4624 put
: /* is_target_pte_for_mc() gets the page */
4627 case MC_TARGET_SWAP
:
4629 if (!mem_cgroup_move_swap_account(ent
,
4630 mc
.from
, mc
.to
, false)) {
4632 /* we fixup refcnts and charges later. */
4640 pte_unmap_unlock(pte
- 1, ptl
);
4645 * We have consumed all precharges we got in can_attach().
4646 * We try charge one by one, but don't do any additional
4647 * charges to mc.to if we have failed in charge once in attach()
4650 ret
= mem_cgroup_do_precharge(1);
4658 static void mem_cgroup_move_charge(struct mm_struct
*mm
)
4660 struct vm_area_struct
*vma
;
4662 lru_add_drain_all();
4663 /* We've already held the mmap_sem */
4664 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
4666 struct mm_walk mem_cgroup_move_charge_walk
= {
4667 .pmd_entry
= mem_cgroup_move_charge_pte_range
,
4671 if (is_vm_hugetlb_page(vma
))
4673 ret
= walk_page_range(vma
->vm_start
, vma
->vm_end
,
4674 &mem_cgroup_move_charge_walk
);
4677 * means we have consumed all precharges and failed in
4678 * doing additional charge. Just abandon here.
4684 static void mem_cgroup_move_task(struct cgroup_subsys
*ss
,
4685 struct cgroup
*cont
,
4686 struct cgroup
*old_cont
,
4687 struct task_struct
*p
,
4691 /* no need to move charge */
4694 mem_cgroup_move_charge(mc
.mm
);
4695 mem_cgroup_clear_mc();
4697 #else /* !CONFIG_MMU */
4698 static int mem_cgroup_can_attach(struct cgroup_subsys
*ss
,
4699 struct cgroup
*cgroup
,
4700 struct task_struct
*p
,
4705 static void mem_cgroup_cancel_attach(struct cgroup_subsys
*ss
,
4706 struct cgroup
*cgroup
,
4707 struct task_struct
*p
,
4711 static void mem_cgroup_move_task(struct cgroup_subsys
*ss
,
4712 struct cgroup
*cont
,
4713 struct cgroup
*old_cont
,
4714 struct task_struct
*p
,
4720 struct cgroup_subsys mem_cgroup_subsys
= {
4722 .subsys_id
= mem_cgroup_subsys_id
,
4723 .create
= mem_cgroup_create
,
4724 .pre_destroy
= mem_cgroup_pre_destroy
,
4725 .destroy
= mem_cgroup_destroy
,
4726 .populate
= mem_cgroup_populate
,
4727 .can_attach
= mem_cgroup_can_attach
,
4728 .cancel_attach
= mem_cgroup_cancel_attach
,
4729 .attach
= mem_cgroup_move_task
,
4734 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
4736 static int __init
disable_swap_account(char *s
)
4738 really_do_swap_account
= 0;
4741 __setup("noswapaccount", disable_swap_account
);