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/export.h>
37 #include <linux/mutex.h>
38 #include <linux/rbtree.h>
39 #include <linux/slab.h>
40 #include <linux/swap.h>
41 #include <linux/swapops.h>
42 #include <linux/spinlock.h>
43 #include <linux/eventfd.h>
44 #include <linux/sort.h>
46 #include <linux/seq_file.h>
47 #include <linux/vmalloc.h>
48 #include <linux/mm_inline.h>
49 #include <linux/page_cgroup.h>
50 #include <linux/cpu.h>
51 #include <linux/oom.h>
55 #include <net/tcp_memcontrol.h>
57 #include <asm/uaccess.h>
59 #include <trace/events/vmscan.h>
61 struct cgroup_subsys mem_cgroup_subsys __read_mostly
;
62 #define MEM_CGROUP_RECLAIM_RETRIES 5
63 static struct mem_cgroup
*root_mem_cgroup __read_mostly
;
65 #ifdef CONFIG_MEMCG_SWAP
66 /* Turned on only when memory cgroup is enabled && really_do_swap_account = 1 */
67 int do_swap_account __read_mostly
;
69 /* for remember boot option*/
70 #ifdef CONFIG_MEMCG_SWAP_ENABLED
71 static int really_do_swap_account __initdata
= 1;
73 static int really_do_swap_account __initdata
= 0;
77 #define do_swap_account 0
82 * Statistics for memory cgroup.
84 enum mem_cgroup_stat_index
{
86 * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss.
88 MEM_CGROUP_STAT_CACHE
, /* # of pages charged as cache */
89 MEM_CGROUP_STAT_RSS
, /* # of pages charged as anon rss */
90 MEM_CGROUP_STAT_FILE_MAPPED
, /* # of pages charged as file rss */
91 MEM_CGROUP_STAT_SWAP
, /* # of pages, swapped out */
92 MEM_CGROUP_STAT_NSTATS
,
95 static const char * const mem_cgroup_stat_names
[] = {
102 enum mem_cgroup_events_index
{
103 MEM_CGROUP_EVENTS_PGPGIN
, /* # of pages paged in */
104 MEM_CGROUP_EVENTS_PGPGOUT
, /* # of pages paged out */
105 MEM_CGROUP_EVENTS_PGFAULT
, /* # of page-faults */
106 MEM_CGROUP_EVENTS_PGMAJFAULT
, /* # of major page-faults */
107 MEM_CGROUP_EVENTS_NSTATS
,
110 static const char * const mem_cgroup_events_names
[] = {
118 * Per memcg event counter is incremented at every pagein/pageout. With THP,
119 * it will be incremated by the number of pages. This counter is used for
120 * for trigger some periodic events. This is straightforward and better
121 * than using jiffies etc. to handle periodic memcg event.
123 enum mem_cgroup_events_target
{
124 MEM_CGROUP_TARGET_THRESH
,
125 MEM_CGROUP_TARGET_SOFTLIMIT
,
126 MEM_CGROUP_TARGET_NUMAINFO
,
129 #define THRESHOLDS_EVENTS_TARGET 128
130 #define SOFTLIMIT_EVENTS_TARGET 1024
131 #define NUMAINFO_EVENTS_TARGET 1024
133 struct mem_cgroup_stat_cpu
{
134 long count
[MEM_CGROUP_STAT_NSTATS
];
135 unsigned long events
[MEM_CGROUP_EVENTS_NSTATS
];
136 unsigned long nr_page_events
;
137 unsigned long targets
[MEM_CGROUP_NTARGETS
];
140 struct mem_cgroup_reclaim_iter
{
141 /* css_id of the last scanned hierarchy member */
143 /* scan generation, increased every round-trip */
144 unsigned int generation
;
148 * per-zone information in memory controller.
150 struct mem_cgroup_per_zone
{
151 struct lruvec lruvec
;
152 unsigned long lru_size
[NR_LRU_LISTS
];
154 struct mem_cgroup_reclaim_iter reclaim_iter
[DEF_PRIORITY
+ 1];
156 struct rb_node tree_node
; /* RB tree node */
157 unsigned long long usage_in_excess
;/* Set to the value by which */
158 /* the soft limit is exceeded*/
160 struct mem_cgroup
*memcg
; /* Back pointer, we cannot */
161 /* use container_of */
164 struct mem_cgroup_per_node
{
165 struct mem_cgroup_per_zone zoneinfo
[MAX_NR_ZONES
];
168 struct mem_cgroup_lru_info
{
169 struct mem_cgroup_per_node
*nodeinfo
[MAX_NUMNODES
];
173 * Cgroups above their limits are maintained in a RB-Tree, independent of
174 * their hierarchy representation
177 struct mem_cgroup_tree_per_zone
{
178 struct rb_root rb_root
;
182 struct mem_cgroup_tree_per_node
{
183 struct mem_cgroup_tree_per_zone rb_tree_per_zone
[MAX_NR_ZONES
];
186 struct mem_cgroup_tree
{
187 struct mem_cgroup_tree_per_node
*rb_tree_per_node
[MAX_NUMNODES
];
190 static struct mem_cgroup_tree soft_limit_tree __read_mostly
;
192 struct mem_cgroup_threshold
{
193 struct eventfd_ctx
*eventfd
;
198 struct mem_cgroup_threshold_ary
{
199 /* An array index points to threshold just below or equal to usage. */
200 int current_threshold
;
201 /* Size of entries[] */
203 /* Array of thresholds */
204 struct mem_cgroup_threshold entries
[0];
207 struct mem_cgroup_thresholds
{
208 /* Primary thresholds array */
209 struct mem_cgroup_threshold_ary
*primary
;
211 * Spare threshold array.
212 * This is needed to make mem_cgroup_unregister_event() "never fail".
213 * It must be able to store at least primary->size - 1 entries.
215 struct mem_cgroup_threshold_ary
*spare
;
219 struct mem_cgroup_eventfd_list
{
220 struct list_head list
;
221 struct eventfd_ctx
*eventfd
;
224 static void mem_cgroup_threshold(struct mem_cgroup
*memcg
);
225 static void mem_cgroup_oom_notify(struct mem_cgroup
*memcg
);
228 * The memory controller data structure. The memory controller controls both
229 * page cache and RSS per cgroup. We would eventually like to provide
230 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
231 * to help the administrator determine what knobs to tune.
233 * TODO: Add a water mark for the memory controller. Reclaim will begin when
234 * we hit the water mark. May be even add a low water mark, such that
235 * no reclaim occurs from a cgroup at it's low water mark, this is
236 * a feature that will be implemented much later in the future.
239 struct cgroup_subsys_state css
;
241 * the counter to account for memory usage
243 struct res_counter res
;
247 * the counter to account for mem+swap usage.
249 struct res_counter memsw
;
252 * rcu_freeing is used only when freeing struct mem_cgroup,
253 * so put it into a union to avoid wasting more memory.
254 * It must be disjoint from the css field. It could be
255 * in a union with the res field, but res plays a much
256 * larger part in mem_cgroup life than memsw, and might
257 * be of interest, even at time of free, when debugging.
258 * So share rcu_head with the less interesting memsw.
260 struct rcu_head rcu_freeing
;
262 * We also need some space for a worker in deferred freeing.
263 * By the time we call it, rcu_freeing is no longer in use.
265 struct work_struct work_freeing
;
269 * Per cgroup active and inactive list, similar to the
270 * per zone LRU lists.
272 struct mem_cgroup_lru_info info
;
273 int last_scanned_node
;
275 nodemask_t scan_nodes
;
276 atomic_t numainfo_events
;
277 atomic_t numainfo_updating
;
280 * Should the accounting and control be hierarchical, per subtree?
290 /* OOM-Killer disable */
291 int oom_kill_disable
;
293 /* set when res.limit == memsw.limit */
294 bool memsw_is_minimum
;
296 /* protect arrays of thresholds */
297 struct mutex thresholds_lock
;
299 /* thresholds for memory usage. RCU-protected */
300 struct mem_cgroup_thresholds thresholds
;
302 /* thresholds for mem+swap usage. RCU-protected */
303 struct mem_cgroup_thresholds memsw_thresholds
;
305 /* For oom notifier event fd */
306 struct list_head oom_notify
;
309 * Should we move charges of a task when a task is moved into this
310 * mem_cgroup ? And what type of charges should we move ?
312 unsigned long move_charge_at_immigrate
;
314 * set > 0 if pages under this cgroup are moving to other cgroup.
316 atomic_t moving_account
;
317 /* taken only while moving_account > 0 */
318 spinlock_t move_lock
;
322 struct mem_cgroup_stat_cpu __percpu
*stat
;
324 * used when a cpu is offlined or other synchronizations
325 * See mem_cgroup_read_stat().
327 struct mem_cgroup_stat_cpu nocpu_base
;
328 spinlock_t pcp_counter_lock
;
330 #if defined(CONFIG_MEMCG_KMEM) && defined(CONFIG_INET)
331 struct tcp_memcontrol tcp_mem
;
335 /* Stuffs for move charges at task migration. */
337 * Types of charges to be moved. "move_charge_at_immitgrate" is treated as a
338 * left-shifted bitmap of these types.
341 MOVE_CHARGE_TYPE_ANON
, /* private anonymous page and swap of it */
342 MOVE_CHARGE_TYPE_FILE
, /* file page(including tmpfs) and swap of it */
346 /* "mc" and its members are protected by cgroup_mutex */
347 static struct move_charge_struct
{
348 spinlock_t lock
; /* for from, to */
349 struct mem_cgroup
*from
;
350 struct mem_cgroup
*to
;
351 unsigned long precharge
;
352 unsigned long moved_charge
;
353 unsigned long moved_swap
;
354 struct task_struct
*moving_task
; /* a task moving charges */
355 wait_queue_head_t waitq
; /* a waitq for other context */
357 .lock
= __SPIN_LOCK_UNLOCKED(mc
.lock
),
358 .waitq
= __WAIT_QUEUE_HEAD_INITIALIZER(mc
.waitq
),
361 static bool move_anon(void)
363 return test_bit(MOVE_CHARGE_TYPE_ANON
,
364 &mc
.to
->move_charge_at_immigrate
);
367 static bool move_file(void)
369 return test_bit(MOVE_CHARGE_TYPE_FILE
,
370 &mc
.to
->move_charge_at_immigrate
);
374 * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft
375 * limit reclaim to prevent infinite loops, if they ever occur.
377 #define MEM_CGROUP_MAX_RECLAIM_LOOPS 100
378 #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2
381 MEM_CGROUP_CHARGE_TYPE_CACHE
= 0,
382 MEM_CGROUP_CHARGE_TYPE_ANON
,
383 MEM_CGROUP_CHARGE_TYPE_SWAPOUT
, /* for accounting swapcache */
384 MEM_CGROUP_CHARGE_TYPE_DROP
, /* a page was unused swap cache */
388 /* for encoding cft->private value on file */
391 #define _OOM_TYPE (2)
392 #define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val))
393 #define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff)
394 #define MEMFILE_ATTR(val) ((val) & 0xffff)
395 /* Used for OOM nofiier */
396 #define OOM_CONTROL (0)
399 * Reclaim flags for mem_cgroup_hierarchical_reclaim
401 #define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0
402 #define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT)
403 #define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1
404 #define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT)
406 static void mem_cgroup_get(struct mem_cgroup
*memcg
);
407 static void mem_cgroup_put(struct mem_cgroup
*memcg
);
410 struct mem_cgroup
*mem_cgroup_from_css(struct cgroup_subsys_state
*s
)
412 return container_of(s
, struct mem_cgroup
, css
);
415 /* Writing them here to avoid exposing memcg's inner layout */
416 #if defined(CONFIG_INET) && defined(CONFIG_MEMCG_KMEM)
418 static bool mem_cgroup_is_root(struct mem_cgroup
*memcg
);
419 void sock_update_memcg(struct sock
*sk
)
421 if (mem_cgroup_sockets_enabled
) {
422 struct mem_cgroup
*memcg
;
423 struct cg_proto
*cg_proto
;
425 BUG_ON(!sk
->sk_prot
->proto_cgroup
);
427 /* Socket cloning can throw us here with sk_cgrp already
428 * filled. It won't however, necessarily happen from
429 * process context. So the test for root memcg given
430 * the current task's memcg won't help us in this case.
432 * Respecting the original socket's memcg is a better
433 * decision in this case.
436 BUG_ON(mem_cgroup_is_root(sk
->sk_cgrp
->memcg
));
437 mem_cgroup_get(sk
->sk_cgrp
->memcg
);
442 memcg
= mem_cgroup_from_task(current
);
443 cg_proto
= sk
->sk_prot
->proto_cgroup(memcg
);
444 if (!mem_cgroup_is_root(memcg
) && memcg_proto_active(cg_proto
)) {
445 mem_cgroup_get(memcg
);
446 sk
->sk_cgrp
= cg_proto
;
451 EXPORT_SYMBOL(sock_update_memcg
);
453 void sock_release_memcg(struct sock
*sk
)
455 if (mem_cgroup_sockets_enabled
&& sk
->sk_cgrp
) {
456 struct mem_cgroup
*memcg
;
457 WARN_ON(!sk
->sk_cgrp
->memcg
);
458 memcg
= sk
->sk_cgrp
->memcg
;
459 mem_cgroup_put(memcg
);
463 struct cg_proto
*tcp_proto_cgroup(struct mem_cgroup
*memcg
)
465 if (!memcg
|| mem_cgroup_is_root(memcg
))
468 return &memcg
->tcp_mem
.cg_proto
;
470 EXPORT_SYMBOL(tcp_proto_cgroup
);
472 static void disarm_sock_keys(struct mem_cgroup
*memcg
)
474 if (!memcg_proto_activated(&memcg
->tcp_mem
.cg_proto
))
476 static_key_slow_dec(&memcg_socket_limit_enabled
);
479 static void disarm_sock_keys(struct mem_cgroup
*memcg
)
484 static void drain_all_stock_async(struct mem_cgroup
*memcg
);
486 static struct mem_cgroup_per_zone
*
487 mem_cgroup_zoneinfo(struct mem_cgroup
*memcg
, int nid
, int zid
)
489 return &memcg
->info
.nodeinfo
[nid
]->zoneinfo
[zid
];
492 struct cgroup_subsys_state
*mem_cgroup_css(struct mem_cgroup
*memcg
)
497 static struct mem_cgroup_per_zone
*
498 page_cgroup_zoneinfo(struct mem_cgroup
*memcg
, struct page
*page
)
500 int nid
= page_to_nid(page
);
501 int zid
= page_zonenum(page
);
503 return mem_cgroup_zoneinfo(memcg
, nid
, zid
);
506 static struct mem_cgroup_tree_per_zone
*
507 soft_limit_tree_node_zone(int nid
, int zid
)
509 return &soft_limit_tree
.rb_tree_per_node
[nid
]->rb_tree_per_zone
[zid
];
512 static struct mem_cgroup_tree_per_zone
*
513 soft_limit_tree_from_page(struct page
*page
)
515 int nid
= page_to_nid(page
);
516 int zid
= page_zonenum(page
);
518 return &soft_limit_tree
.rb_tree_per_node
[nid
]->rb_tree_per_zone
[zid
];
522 __mem_cgroup_insert_exceeded(struct mem_cgroup
*memcg
,
523 struct mem_cgroup_per_zone
*mz
,
524 struct mem_cgroup_tree_per_zone
*mctz
,
525 unsigned long long new_usage_in_excess
)
527 struct rb_node
**p
= &mctz
->rb_root
.rb_node
;
528 struct rb_node
*parent
= NULL
;
529 struct mem_cgroup_per_zone
*mz_node
;
534 mz
->usage_in_excess
= new_usage_in_excess
;
535 if (!mz
->usage_in_excess
)
539 mz_node
= rb_entry(parent
, struct mem_cgroup_per_zone
,
541 if (mz
->usage_in_excess
< mz_node
->usage_in_excess
)
544 * We can't avoid mem cgroups that are over their soft
545 * limit by the same amount
547 else if (mz
->usage_in_excess
>= mz_node
->usage_in_excess
)
550 rb_link_node(&mz
->tree_node
, parent
, p
);
551 rb_insert_color(&mz
->tree_node
, &mctz
->rb_root
);
556 __mem_cgroup_remove_exceeded(struct mem_cgroup
*memcg
,
557 struct mem_cgroup_per_zone
*mz
,
558 struct mem_cgroup_tree_per_zone
*mctz
)
562 rb_erase(&mz
->tree_node
, &mctz
->rb_root
);
567 mem_cgroup_remove_exceeded(struct mem_cgroup
*memcg
,
568 struct mem_cgroup_per_zone
*mz
,
569 struct mem_cgroup_tree_per_zone
*mctz
)
571 spin_lock(&mctz
->lock
);
572 __mem_cgroup_remove_exceeded(memcg
, mz
, mctz
);
573 spin_unlock(&mctz
->lock
);
577 static void mem_cgroup_update_tree(struct mem_cgroup
*memcg
, struct page
*page
)
579 unsigned long long excess
;
580 struct mem_cgroup_per_zone
*mz
;
581 struct mem_cgroup_tree_per_zone
*mctz
;
582 int nid
= page_to_nid(page
);
583 int zid
= page_zonenum(page
);
584 mctz
= soft_limit_tree_from_page(page
);
587 * Necessary to update all ancestors when hierarchy is used.
588 * because their event counter is not touched.
590 for (; memcg
; memcg
= parent_mem_cgroup(memcg
)) {
591 mz
= mem_cgroup_zoneinfo(memcg
, nid
, zid
);
592 excess
= res_counter_soft_limit_excess(&memcg
->res
);
594 * We have to update the tree if mz is on RB-tree or
595 * mem is over its softlimit.
597 if (excess
|| mz
->on_tree
) {
598 spin_lock(&mctz
->lock
);
599 /* if on-tree, remove it */
601 __mem_cgroup_remove_exceeded(memcg
, mz
, mctz
);
603 * Insert again. mz->usage_in_excess will be updated.
604 * If excess is 0, no tree ops.
606 __mem_cgroup_insert_exceeded(memcg
, mz
, mctz
, excess
);
607 spin_unlock(&mctz
->lock
);
612 static void mem_cgroup_remove_from_trees(struct mem_cgroup
*memcg
)
615 struct mem_cgroup_per_zone
*mz
;
616 struct mem_cgroup_tree_per_zone
*mctz
;
618 for_each_node(node
) {
619 for (zone
= 0; zone
< MAX_NR_ZONES
; zone
++) {
620 mz
= mem_cgroup_zoneinfo(memcg
, node
, zone
);
621 mctz
= soft_limit_tree_node_zone(node
, zone
);
622 mem_cgroup_remove_exceeded(memcg
, mz
, mctz
);
627 static struct mem_cgroup_per_zone
*
628 __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone
*mctz
)
630 struct rb_node
*rightmost
= NULL
;
631 struct mem_cgroup_per_zone
*mz
;
635 rightmost
= rb_last(&mctz
->rb_root
);
637 goto done
; /* Nothing to reclaim from */
639 mz
= rb_entry(rightmost
, struct mem_cgroup_per_zone
, tree_node
);
641 * Remove the node now but someone else can add it back,
642 * we will to add it back at the end of reclaim to its correct
643 * position in the tree.
645 __mem_cgroup_remove_exceeded(mz
->memcg
, mz
, mctz
);
646 if (!res_counter_soft_limit_excess(&mz
->memcg
->res
) ||
647 !css_tryget(&mz
->memcg
->css
))
653 static struct mem_cgroup_per_zone
*
654 mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone
*mctz
)
656 struct mem_cgroup_per_zone
*mz
;
658 spin_lock(&mctz
->lock
);
659 mz
= __mem_cgroup_largest_soft_limit_node(mctz
);
660 spin_unlock(&mctz
->lock
);
665 * Implementation Note: reading percpu statistics for memcg.
667 * Both of vmstat[] and percpu_counter has threshold and do periodic
668 * synchronization to implement "quick" read. There are trade-off between
669 * reading cost and precision of value. Then, we may have a chance to implement
670 * a periodic synchronizion of counter in memcg's counter.
672 * But this _read() function is used for user interface now. The user accounts
673 * memory usage by memory cgroup and he _always_ requires exact value because
674 * he accounts memory. Even if we provide quick-and-fuzzy read, we always
675 * have to visit all online cpus and make sum. So, for now, unnecessary
676 * synchronization is not implemented. (just implemented for cpu hotplug)
678 * If there are kernel internal actions which can make use of some not-exact
679 * value, and reading all cpu value can be performance bottleneck in some
680 * common workload, threashold and synchonization as vmstat[] should be
683 static long mem_cgroup_read_stat(struct mem_cgroup
*memcg
,
684 enum mem_cgroup_stat_index idx
)
690 for_each_online_cpu(cpu
)
691 val
+= per_cpu(memcg
->stat
->count
[idx
], cpu
);
692 #ifdef CONFIG_HOTPLUG_CPU
693 spin_lock(&memcg
->pcp_counter_lock
);
694 val
+= memcg
->nocpu_base
.count
[idx
];
695 spin_unlock(&memcg
->pcp_counter_lock
);
701 static void mem_cgroup_swap_statistics(struct mem_cgroup
*memcg
,
704 int val
= (charge
) ? 1 : -1;
705 this_cpu_add(memcg
->stat
->count
[MEM_CGROUP_STAT_SWAP
], val
);
708 static unsigned long mem_cgroup_read_events(struct mem_cgroup
*memcg
,
709 enum mem_cgroup_events_index idx
)
711 unsigned long val
= 0;
714 for_each_online_cpu(cpu
)
715 val
+= per_cpu(memcg
->stat
->events
[idx
], cpu
);
716 #ifdef CONFIG_HOTPLUG_CPU
717 spin_lock(&memcg
->pcp_counter_lock
);
718 val
+= memcg
->nocpu_base
.events
[idx
];
719 spin_unlock(&memcg
->pcp_counter_lock
);
724 static void mem_cgroup_charge_statistics(struct mem_cgroup
*memcg
,
725 bool anon
, int nr_pages
)
730 * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is
731 * counted as CACHE even if it's on ANON LRU.
734 __this_cpu_add(memcg
->stat
->count
[MEM_CGROUP_STAT_RSS
],
737 __this_cpu_add(memcg
->stat
->count
[MEM_CGROUP_STAT_CACHE
],
740 /* pagein of a big page is an event. So, ignore page size */
742 __this_cpu_inc(memcg
->stat
->events
[MEM_CGROUP_EVENTS_PGPGIN
]);
744 __this_cpu_inc(memcg
->stat
->events
[MEM_CGROUP_EVENTS_PGPGOUT
]);
745 nr_pages
= -nr_pages
; /* for event */
748 __this_cpu_add(memcg
->stat
->nr_page_events
, nr_pages
);
754 mem_cgroup_get_lru_size(struct lruvec
*lruvec
, enum lru_list lru
)
756 struct mem_cgroup_per_zone
*mz
;
758 mz
= container_of(lruvec
, struct mem_cgroup_per_zone
, lruvec
);
759 return mz
->lru_size
[lru
];
763 mem_cgroup_zone_nr_lru_pages(struct mem_cgroup
*memcg
, int nid
, int zid
,
764 unsigned int lru_mask
)
766 struct mem_cgroup_per_zone
*mz
;
768 unsigned long ret
= 0;
770 mz
= mem_cgroup_zoneinfo(memcg
, nid
, zid
);
773 if (BIT(lru
) & lru_mask
)
774 ret
+= mz
->lru_size
[lru
];
780 mem_cgroup_node_nr_lru_pages(struct mem_cgroup
*memcg
,
781 int nid
, unsigned int lru_mask
)
786 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++)
787 total
+= mem_cgroup_zone_nr_lru_pages(memcg
,
793 static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup
*memcg
,
794 unsigned int lru_mask
)
799 for_each_node_state(nid
, N_HIGH_MEMORY
)
800 total
+= mem_cgroup_node_nr_lru_pages(memcg
, nid
, lru_mask
);
804 static bool mem_cgroup_event_ratelimit(struct mem_cgroup
*memcg
,
805 enum mem_cgroup_events_target target
)
807 unsigned long val
, next
;
809 val
= __this_cpu_read(memcg
->stat
->nr_page_events
);
810 next
= __this_cpu_read(memcg
->stat
->targets
[target
]);
811 /* from time_after() in jiffies.h */
812 if ((long)next
- (long)val
< 0) {
814 case MEM_CGROUP_TARGET_THRESH
:
815 next
= val
+ THRESHOLDS_EVENTS_TARGET
;
817 case MEM_CGROUP_TARGET_SOFTLIMIT
:
818 next
= val
+ SOFTLIMIT_EVENTS_TARGET
;
820 case MEM_CGROUP_TARGET_NUMAINFO
:
821 next
= val
+ NUMAINFO_EVENTS_TARGET
;
826 __this_cpu_write(memcg
->stat
->targets
[target
], next
);
833 * Check events in order.
836 static void memcg_check_events(struct mem_cgroup
*memcg
, struct page
*page
)
839 /* threshold event is triggered in finer grain than soft limit */
840 if (unlikely(mem_cgroup_event_ratelimit(memcg
,
841 MEM_CGROUP_TARGET_THRESH
))) {
843 bool do_numainfo __maybe_unused
;
845 do_softlimit
= mem_cgroup_event_ratelimit(memcg
,
846 MEM_CGROUP_TARGET_SOFTLIMIT
);
848 do_numainfo
= mem_cgroup_event_ratelimit(memcg
,
849 MEM_CGROUP_TARGET_NUMAINFO
);
853 mem_cgroup_threshold(memcg
);
854 if (unlikely(do_softlimit
))
855 mem_cgroup_update_tree(memcg
, page
);
857 if (unlikely(do_numainfo
))
858 atomic_inc(&memcg
->numainfo_events
);
864 struct mem_cgroup
*mem_cgroup_from_cont(struct cgroup
*cont
)
866 return mem_cgroup_from_css(
867 cgroup_subsys_state(cont
, mem_cgroup_subsys_id
));
870 struct mem_cgroup
*mem_cgroup_from_task(struct task_struct
*p
)
873 * mm_update_next_owner() may clear mm->owner to NULL
874 * if it races with swapoff, page migration, etc.
875 * So this can be called with p == NULL.
880 return mem_cgroup_from_css(task_subsys_state(p
, mem_cgroup_subsys_id
));
883 struct mem_cgroup
*try_get_mem_cgroup_from_mm(struct mm_struct
*mm
)
885 struct mem_cgroup
*memcg
= NULL
;
890 * Because we have no locks, mm->owner's may be being moved to other
891 * cgroup. We use css_tryget() here even if this looks
892 * pessimistic (rather than adding locks here).
896 memcg
= mem_cgroup_from_task(rcu_dereference(mm
->owner
));
897 if (unlikely(!memcg
))
899 } while (!css_tryget(&memcg
->css
));
905 * mem_cgroup_iter - iterate over memory cgroup hierarchy
906 * @root: hierarchy root
907 * @prev: previously returned memcg, NULL on first invocation
908 * @reclaim: cookie for shared reclaim walks, NULL for full walks
910 * Returns references to children of the hierarchy below @root, or
911 * @root itself, or %NULL after a full round-trip.
913 * Caller must pass the return value in @prev on subsequent
914 * invocations for reference counting, or use mem_cgroup_iter_break()
915 * to cancel a hierarchy walk before the round-trip is complete.
917 * Reclaimers can specify a zone and a priority level in @reclaim to
918 * divide up the memcgs in the hierarchy among all concurrent
919 * reclaimers operating on the same zone and priority.
921 struct mem_cgroup
*mem_cgroup_iter(struct mem_cgroup
*root
,
922 struct mem_cgroup
*prev
,
923 struct mem_cgroup_reclaim_cookie
*reclaim
)
925 struct mem_cgroup
*memcg
= NULL
;
928 if (mem_cgroup_disabled())
932 root
= root_mem_cgroup
;
934 if (prev
&& !reclaim
)
935 id
= css_id(&prev
->css
);
937 if (prev
&& prev
!= root
)
940 if (!root
->use_hierarchy
&& root
!= root_mem_cgroup
) {
947 struct mem_cgroup_reclaim_iter
*uninitialized_var(iter
);
948 struct cgroup_subsys_state
*css
;
951 int nid
= zone_to_nid(reclaim
->zone
);
952 int zid
= zone_idx(reclaim
->zone
);
953 struct mem_cgroup_per_zone
*mz
;
955 mz
= mem_cgroup_zoneinfo(root
, nid
, zid
);
956 iter
= &mz
->reclaim_iter
[reclaim
->priority
];
957 if (prev
&& reclaim
->generation
!= iter
->generation
)
963 css
= css_get_next(&mem_cgroup_subsys
, id
+ 1, &root
->css
, &id
);
965 if (css
== &root
->css
|| css_tryget(css
))
966 memcg
= mem_cgroup_from_css(css
);
975 else if (!prev
&& memcg
)
976 reclaim
->generation
= iter
->generation
;
986 * mem_cgroup_iter_break - abort a hierarchy walk prematurely
987 * @root: hierarchy root
988 * @prev: last visited hierarchy member as returned by mem_cgroup_iter()
990 void mem_cgroup_iter_break(struct mem_cgroup
*root
,
991 struct mem_cgroup
*prev
)
994 root
= root_mem_cgroup
;
995 if (prev
&& prev
!= root
)
1000 * Iteration constructs for visiting all cgroups (under a tree). If
1001 * loops are exited prematurely (break), mem_cgroup_iter_break() must
1002 * be used for reference counting.
1004 #define for_each_mem_cgroup_tree(iter, root) \
1005 for (iter = mem_cgroup_iter(root, NULL, NULL); \
1007 iter = mem_cgroup_iter(root, iter, NULL))
1009 #define for_each_mem_cgroup(iter) \
1010 for (iter = mem_cgroup_iter(NULL, NULL, NULL); \
1012 iter = mem_cgroup_iter(NULL, iter, NULL))
1014 static inline bool mem_cgroup_is_root(struct mem_cgroup
*memcg
)
1016 return (memcg
== root_mem_cgroup
);
1019 void mem_cgroup_count_vm_event(struct mm_struct
*mm
, enum vm_event_item idx
)
1021 struct mem_cgroup
*memcg
;
1027 memcg
= mem_cgroup_from_task(rcu_dereference(mm
->owner
));
1028 if (unlikely(!memcg
))
1033 this_cpu_inc(memcg
->stat
->events
[MEM_CGROUP_EVENTS_PGFAULT
]);
1036 this_cpu_inc(memcg
->stat
->events
[MEM_CGROUP_EVENTS_PGMAJFAULT
]);
1044 EXPORT_SYMBOL(mem_cgroup_count_vm_event
);
1047 * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg
1048 * @zone: zone of the wanted lruvec
1049 * @memcg: memcg of the wanted lruvec
1051 * Returns the lru list vector holding pages for the given @zone and
1052 * @mem. This can be the global zone lruvec, if the memory controller
1055 struct lruvec
*mem_cgroup_zone_lruvec(struct zone
*zone
,
1056 struct mem_cgroup
*memcg
)
1058 struct mem_cgroup_per_zone
*mz
;
1060 if (mem_cgroup_disabled())
1061 return &zone
->lruvec
;
1063 mz
= mem_cgroup_zoneinfo(memcg
, zone_to_nid(zone
), zone_idx(zone
));
1068 * Following LRU functions are allowed to be used without PCG_LOCK.
1069 * Operations are called by routine of global LRU independently from memcg.
1070 * What we have to take care of here is validness of pc->mem_cgroup.
1072 * Changes to pc->mem_cgroup happens when
1075 * In typical case, "charge" is done before add-to-lru. Exception is SwapCache.
1076 * It is added to LRU before charge.
1077 * If PCG_USED bit is not set, page_cgroup is not added to this private LRU.
1078 * When moving account, the page is not on LRU. It's isolated.
1082 * mem_cgroup_page_lruvec - return lruvec for adding an lru page
1084 * @zone: zone of the page
1086 struct lruvec
*mem_cgroup_page_lruvec(struct page
*page
, struct zone
*zone
)
1088 struct mem_cgroup_per_zone
*mz
;
1089 struct mem_cgroup
*memcg
;
1090 struct page_cgroup
*pc
;
1092 if (mem_cgroup_disabled())
1093 return &zone
->lruvec
;
1095 pc
= lookup_page_cgroup(page
);
1096 memcg
= pc
->mem_cgroup
;
1099 * Surreptitiously switch any uncharged offlist page to root:
1100 * an uncharged page off lru does nothing to secure
1101 * its former mem_cgroup from sudden removal.
1103 * Our caller holds lru_lock, and PageCgroupUsed is updated
1104 * under page_cgroup lock: between them, they make all uses
1105 * of pc->mem_cgroup safe.
1107 if (!PageLRU(page
) && !PageCgroupUsed(pc
) && memcg
!= root_mem_cgroup
)
1108 pc
->mem_cgroup
= memcg
= root_mem_cgroup
;
1110 mz
= page_cgroup_zoneinfo(memcg
, page
);
1115 * mem_cgroup_update_lru_size - account for adding or removing an lru page
1116 * @lruvec: mem_cgroup per zone lru vector
1117 * @lru: index of lru list the page is sitting on
1118 * @nr_pages: positive when adding or negative when removing
1120 * This function must be called when a page is added to or removed from an
1123 void mem_cgroup_update_lru_size(struct lruvec
*lruvec
, enum lru_list lru
,
1126 struct mem_cgroup_per_zone
*mz
;
1127 unsigned long *lru_size
;
1129 if (mem_cgroup_disabled())
1132 mz
= container_of(lruvec
, struct mem_cgroup_per_zone
, lruvec
);
1133 lru_size
= mz
->lru_size
+ lru
;
1134 *lru_size
+= nr_pages
;
1135 VM_BUG_ON((long)(*lru_size
) < 0);
1139 * Checks whether given mem is same or in the root_mem_cgroup's
1142 bool __mem_cgroup_same_or_subtree(const struct mem_cgroup
*root_memcg
,
1143 struct mem_cgroup
*memcg
)
1145 if (root_memcg
== memcg
)
1147 if (!root_memcg
->use_hierarchy
|| !memcg
)
1149 return css_is_ancestor(&memcg
->css
, &root_memcg
->css
);
1152 static bool mem_cgroup_same_or_subtree(const struct mem_cgroup
*root_memcg
,
1153 struct mem_cgroup
*memcg
)
1158 ret
= __mem_cgroup_same_or_subtree(root_memcg
, memcg
);
1163 int task_in_mem_cgroup(struct task_struct
*task
, const struct mem_cgroup
*memcg
)
1166 struct mem_cgroup
*curr
= NULL
;
1167 struct task_struct
*p
;
1169 p
= find_lock_task_mm(task
);
1171 curr
= try_get_mem_cgroup_from_mm(p
->mm
);
1175 * All threads may have already detached their mm's, but the oom
1176 * killer still needs to detect if they have already been oom
1177 * killed to prevent needlessly killing additional tasks.
1180 curr
= mem_cgroup_from_task(task
);
1182 css_get(&curr
->css
);
1188 * We should check use_hierarchy of "memcg" not "curr". Because checking
1189 * use_hierarchy of "curr" here make this function true if hierarchy is
1190 * enabled in "curr" and "curr" is a child of "memcg" in *cgroup*
1191 * hierarchy(even if use_hierarchy is disabled in "memcg").
1193 ret
= mem_cgroup_same_or_subtree(memcg
, curr
);
1194 css_put(&curr
->css
);
1198 int mem_cgroup_inactive_anon_is_low(struct lruvec
*lruvec
)
1200 unsigned long inactive_ratio
;
1201 unsigned long inactive
;
1202 unsigned long active
;
1205 inactive
= mem_cgroup_get_lru_size(lruvec
, LRU_INACTIVE_ANON
);
1206 active
= mem_cgroup_get_lru_size(lruvec
, LRU_ACTIVE_ANON
);
1208 gb
= (inactive
+ active
) >> (30 - PAGE_SHIFT
);
1210 inactive_ratio
= int_sqrt(10 * gb
);
1214 return inactive
* inactive_ratio
< active
;
1217 int mem_cgroup_inactive_file_is_low(struct lruvec
*lruvec
)
1219 unsigned long active
;
1220 unsigned long inactive
;
1222 inactive
= mem_cgroup_get_lru_size(lruvec
, LRU_INACTIVE_FILE
);
1223 active
= mem_cgroup_get_lru_size(lruvec
, LRU_ACTIVE_FILE
);
1225 return (active
> inactive
);
1228 #define mem_cgroup_from_res_counter(counter, member) \
1229 container_of(counter, struct mem_cgroup, member)
1232 * mem_cgroup_margin - calculate chargeable space of a memory cgroup
1233 * @memcg: the memory cgroup
1235 * Returns the maximum amount of memory @mem can be charged with, in
1238 static unsigned long mem_cgroup_margin(struct mem_cgroup
*memcg
)
1240 unsigned long long margin
;
1242 margin
= res_counter_margin(&memcg
->res
);
1243 if (do_swap_account
)
1244 margin
= min(margin
, res_counter_margin(&memcg
->memsw
));
1245 return margin
>> PAGE_SHIFT
;
1248 int mem_cgroup_swappiness(struct mem_cgroup
*memcg
)
1250 struct cgroup
*cgrp
= memcg
->css
.cgroup
;
1253 if (cgrp
->parent
== NULL
)
1254 return vm_swappiness
;
1256 return memcg
->swappiness
;
1260 * memcg->moving_account is used for checking possibility that some thread is
1261 * calling move_account(). When a thread on CPU-A starts moving pages under
1262 * a memcg, other threads should check memcg->moving_account under
1263 * rcu_read_lock(), like this:
1267 * memcg->moving_account+1 if (memcg->mocing_account)
1269 * synchronize_rcu() update something.
1274 /* for quick checking without looking up memcg */
1275 atomic_t memcg_moving __read_mostly
;
1277 static void mem_cgroup_start_move(struct mem_cgroup
*memcg
)
1279 atomic_inc(&memcg_moving
);
1280 atomic_inc(&memcg
->moving_account
);
1284 static void mem_cgroup_end_move(struct mem_cgroup
*memcg
)
1287 * Now, mem_cgroup_clear_mc() may call this function with NULL.
1288 * We check NULL in callee rather than caller.
1291 atomic_dec(&memcg_moving
);
1292 atomic_dec(&memcg
->moving_account
);
1297 * 2 routines for checking "mem" is under move_account() or not.
1299 * mem_cgroup_stolen() - checking whether a cgroup is mc.from or not. This
1300 * is used for avoiding races in accounting. If true,
1301 * pc->mem_cgroup may be overwritten.
1303 * mem_cgroup_under_move() - checking a cgroup is mc.from or mc.to or
1304 * under hierarchy of moving cgroups. This is for
1305 * waiting at hith-memory prressure caused by "move".
1308 static bool mem_cgroup_stolen(struct mem_cgroup
*memcg
)
1310 VM_BUG_ON(!rcu_read_lock_held());
1311 return atomic_read(&memcg
->moving_account
) > 0;
1314 static bool mem_cgroup_under_move(struct mem_cgroup
*memcg
)
1316 struct mem_cgroup
*from
;
1317 struct mem_cgroup
*to
;
1320 * Unlike task_move routines, we access mc.to, mc.from not under
1321 * mutual exclusion by cgroup_mutex. Here, we take spinlock instead.
1323 spin_lock(&mc
.lock
);
1329 ret
= mem_cgroup_same_or_subtree(memcg
, from
)
1330 || mem_cgroup_same_or_subtree(memcg
, to
);
1332 spin_unlock(&mc
.lock
);
1336 static bool mem_cgroup_wait_acct_move(struct mem_cgroup
*memcg
)
1338 if (mc
.moving_task
&& current
!= mc
.moving_task
) {
1339 if (mem_cgroup_under_move(memcg
)) {
1341 prepare_to_wait(&mc
.waitq
, &wait
, TASK_INTERRUPTIBLE
);
1342 /* moving charge context might have finished. */
1345 finish_wait(&mc
.waitq
, &wait
);
1353 * Take this lock when
1354 * - a code tries to modify page's memcg while it's USED.
1355 * - a code tries to modify page state accounting in a memcg.
1356 * see mem_cgroup_stolen(), too.
1358 static void move_lock_mem_cgroup(struct mem_cgroup
*memcg
,
1359 unsigned long *flags
)
1361 spin_lock_irqsave(&memcg
->move_lock
, *flags
);
1364 static void move_unlock_mem_cgroup(struct mem_cgroup
*memcg
,
1365 unsigned long *flags
)
1367 spin_unlock_irqrestore(&memcg
->move_lock
, *flags
);
1371 * mem_cgroup_print_oom_info: Called from OOM with tasklist_lock held in read mode.
1372 * @memcg: The memory cgroup that went over limit
1373 * @p: Task that is going to be killed
1375 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1378 void mem_cgroup_print_oom_info(struct mem_cgroup
*memcg
, struct task_struct
*p
)
1380 struct cgroup
*task_cgrp
;
1381 struct cgroup
*mem_cgrp
;
1383 * Need a buffer in BSS, can't rely on allocations. The code relies
1384 * on the assumption that OOM is serialized for memory controller.
1385 * If this assumption is broken, revisit this code.
1387 static char memcg_name
[PATH_MAX
];
1395 mem_cgrp
= memcg
->css
.cgroup
;
1396 task_cgrp
= task_cgroup(p
, mem_cgroup_subsys_id
);
1398 ret
= cgroup_path(task_cgrp
, memcg_name
, PATH_MAX
);
1401 * Unfortunately, we are unable to convert to a useful name
1402 * But we'll still print out the usage information
1409 printk(KERN_INFO
"Task in %s killed", memcg_name
);
1412 ret
= cgroup_path(mem_cgrp
, memcg_name
, PATH_MAX
);
1420 * Continues from above, so we don't need an KERN_ level
1422 printk(KERN_CONT
" as a result of limit of %s\n", memcg_name
);
1425 printk(KERN_INFO
"memory: usage %llukB, limit %llukB, failcnt %llu\n",
1426 res_counter_read_u64(&memcg
->res
, RES_USAGE
) >> 10,
1427 res_counter_read_u64(&memcg
->res
, RES_LIMIT
) >> 10,
1428 res_counter_read_u64(&memcg
->res
, RES_FAILCNT
));
1429 printk(KERN_INFO
"memory+swap: usage %llukB, limit %llukB, "
1431 res_counter_read_u64(&memcg
->memsw
, RES_USAGE
) >> 10,
1432 res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
) >> 10,
1433 res_counter_read_u64(&memcg
->memsw
, RES_FAILCNT
));
1437 * This function returns the number of memcg under hierarchy tree. Returns
1438 * 1(self count) if no children.
1440 static int mem_cgroup_count_children(struct mem_cgroup
*memcg
)
1443 struct mem_cgroup
*iter
;
1445 for_each_mem_cgroup_tree(iter
, memcg
)
1451 * Return the memory (and swap, if configured) limit for a memcg.
1453 static u64
mem_cgroup_get_limit(struct mem_cgroup
*memcg
)
1458 limit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
1459 limit
+= total_swap_pages
<< PAGE_SHIFT
;
1461 memsw
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
1463 * If memsw is finite and limits the amount of swap space available
1464 * to this memcg, return that limit.
1466 return min(limit
, memsw
);
1469 void mem_cgroup_out_of_memory(struct mem_cgroup
*memcg
, gfp_t gfp_mask
,
1472 struct mem_cgroup
*iter
;
1473 unsigned long chosen_points
= 0;
1474 unsigned long totalpages
;
1475 unsigned int points
= 0;
1476 struct task_struct
*chosen
= NULL
;
1479 * If current has a pending SIGKILL, then automatically select it. The
1480 * goal is to allow it to allocate so that it may quickly exit and free
1483 if (fatal_signal_pending(current
)) {
1484 set_thread_flag(TIF_MEMDIE
);
1488 check_panic_on_oom(CONSTRAINT_MEMCG
, gfp_mask
, order
, NULL
);
1489 totalpages
= mem_cgroup_get_limit(memcg
) >> PAGE_SHIFT
? : 1;
1490 for_each_mem_cgroup_tree(iter
, memcg
) {
1491 struct cgroup
*cgroup
= iter
->css
.cgroup
;
1492 struct cgroup_iter it
;
1493 struct task_struct
*task
;
1495 cgroup_iter_start(cgroup
, &it
);
1496 while ((task
= cgroup_iter_next(cgroup
, &it
))) {
1497 switch (oom_scan_process_thread(task
, totalpages
, NULL
,
1499 case OOM_SCAN_SELECT
:
1501 put_task_struct(chosen
);
1503 chosen_points
= ULONG_MAX
;
1504 get_task_struct(chosen
);
1506 case OOM_SCAN_CONTINUE
:
1508 case OOM_SCAN_ABORT
:
1509 cgroup_iter_end(cgroup
, &it
);
1510 mem_cgroup_iter_break(memcg
, iter
);
1512 put_task_struct(chosen
);
1517 points
= oom_badness(task
, memcg
, NULL
, totalpages
);
1518 if (points
> chosen_points
) {
1520 put_task_struct(chosen
);
1522 chosen_points
= points
;
1523 get_task_struct(chosen
);
1526 cgroup_iter_end(cgroup
, &it
);
1531 points
= chosen_points
* 1000 / totalpages
;
1532 oom_kill_process(chosen
, gfp_mask
, order
, points
, totalpages
, memcg
,
1533 NULL
, "Memory cgroup out of memory");
1536 static unsigned long mem_cgroup_reclaim(struct mem_cgroup
*memcg
,
1538 unsigned long flags
)
1540 unsigned long total
= 0;
1541 bool noswap
= false;
1544 if (flags
& MEM_CGROUP_RECLAIM_NOSWAP
)
1546 if (!(flags
& MEM_CGROUP_RECLAIM_SHRINK
) && memcg
->memsw_is_minimum
)
1549 for (loop
= 0; loop
< MEM_CGROUP_MAX_RECLAIM_LOOPS
; loop
++) {
1551 drain_all_stock_async(memcg
);
1552 total
+= try_to_free_mem_cgroup_pages(memcg
, gfp_mask
, noswap
);
1554 * Allow limit shrinkers, which are triggered directly
1555 * by userspace, to catch signals and stop reclaim
1556 * after minimal progress, regardless of the margin.
1558 if (total
&& (flags
& MEM_CGROUP_RECLAIM_SHRINK
))
1560 if (mem_cgroup_margin(memcg
))
1563 * If nothing was reclaimed after two attempts, there
1564 * may be no reclaimable pages in this hierarchy.
1573 * test_mem_cgroup_node_reclaimable
1574 * @memcg: the target memcg
1575 * @nid: the node ID to be checked.
1576 * @noswap : specify true here if the user wants flle only information.
1578 * This function returns whether the specified memcg contains any
1579 * reclaimable pages on a node. Returns true if there are any reclaimable
1580 * pages in the node.
1582 static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup
*memcg
,
1583 int nid
, bool noswap
)
1585 if (mem_cgroup_node_nr_lru_pages(memcg
, nid
, LRU_ALL_FILE
))
1587 if (noswap
|| !total_swap_pages
)
1589 if (mem_cgroup_node_nr_lru_pages(memcg
, nid
, LRU_ALL_ANON
))
1594 #if MAX_NUMNODES > 1
1597 * Always updating the nodemask is not very good - even if we have an empty
1598 * list or the wrong list here, we can start from some node and traverse all
1599 * nodes based on the zonelist. So update the list loosely once per 10 secs.
1602 static void mem_cgroup_may_update_nodemask(struct mem_cgroup
*memcg
)
1606 * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET
1607 * pagein/pageout changes since the last update.
1609 if (!atomic_read(&memcg
->numainfo_events
))
1611 if (atomic_inc_return(&memcg
->numainfo_updating
) > 1)
1614 /* make a nodemask where this memcg uses memory from */
1615 memcg
->scan_nodes
= node_states
[N_HIGH_MEMORY
];
1617 for_each_node_mask(nid
, node_states
[N_HIGH_MEMORY
]) {
1619 if (!test_mem_cgroup_node_reclaimable(memcg
, nid
, false))
1620 node_clear(nid
, memcg
->scan_nodes
);
1623 atomic_set(&memcg
->numainfo_events
, 0);
1624 atomic_set(&memcg
->numainfo_updating
, 0);
1628 * Selecting a node where we start reclaim from. Because what we need is just
1629 * reducing usage counter, start from anywhere is O,K. Considering
1630 * memory reclaim from current node, there are pros. and cons.
1632 * Freeing memory from current node means freeing memory from a node which
1633 * we'll use or we've used. So, it may make LRU bad. And if several threads
1634 * hit limits, it will see a contention on a node. But freeing from remote
1635 * node means more costs for memory reclaim because of memory latency.
1637 * Now, we use round-robin. Better algorithm is welcomed.
1639 int mem_cgroup_select_victim_node(struct mem_cgroup
*memcg
)
1643 mem_cgroup_may_update_nodemask(memcg
);
1644 node
= memcg
->last_scanned_node
;
1646 node
= next_node(node
, memcg
->scan_nodes
);
1647 if (node
== MAX_NUMNODES
)
1648 node
= first_node(memcg
->scan_nodes
);
1650 * We call this when we hit limit, not when pages are added to LRU.
1651 * No LRU may hold pages because all pages are UNEVICTABLE or
1652 * memcg is too small and all pages are not on LRU. In that case,
1653 * we use curret node.
1655 if (unlikely(node
== MAX_NUMNODES
))
1656 node
= numa_node_id();
1658 memcg
->last_scanned_node
= node
;
1663 * Check all nodes whether it contains reclaimable pages or not.
1664 * For quick scan, we make use of scan_nodes. This will allow us to skip
1665 * unused nodes. But scan_nodes is lazily updated and may not cotain
1666 * enough new information. We need to do double check.
1668 static bool mem_cgroup_reclaimable(struct mem_cgroup
*memcg
, bool noswap
)
1673 * quick check...making use of scan_node.
1674 * We can skip unused nodes.
1676 if (!nodes_empty(memcg
->scan_nodes
)) {
1677 for (nid
= first_node(memcg
->scan_nodes
);
1679 nid
= next_node(nid
, memcg
->scan_nodes
)) {
1681 if (test_mem_cgroup_node_reclaimable(memcg
, nid
, noswap
))
1686 * Check rest of nodes.
1688 for_each_node_state(nid
, N_HIGH_MEMORY
) {
1689 if (node_isset(nid
, memcg
->scan_nodes
))
1691 if (test_mem_cgroup_node_reclaimable(memcg
, nid
, noswap
))
1698 int mem_cgroup_select_victim_node(struct mem_cgroup
*memcg
)
1703 static bool mem_cgroup_reclaimable(struct mem_cgroup
*memcg
, bool noswap
)
1705 return test_mem_cgroup_node_reclaimable(memcg
, 0, noswap
);
1709 static int mem_cgroup_soft_reclaim(struct mem_cgroup
*root_memcg
,
1712 unsigned long *total_scanned
)
1714 struct mem_cgroup
*victim
= NULL
;
1717 unsigned long excess
;
1718 unsigned long nr_scanned
;
1719 struct mem_cgroup_reclaim_cookie reclaim
= {
1724 excess
= res_counter_soft_limit_excess(&root_memcg
->res
) >> PAGE_SHIFT
;
1727 victim
= mem_cgroup_iter(root_memcg
, victim
, &reclaim
);
1732 * If we have not been able to reclaim
1733 * anything, it might because there are
1734 * no reclaimable pages under this hierarchy
1739 * We want to do more targeted reclaim.
1740 * excess >> 2 is not to excessive so as to
1741 * reclaim too much, nor too less that we keep
1742 * coming back to reclaim from this cgroup
1744 if (total
>= (excess
>> 2) ||
1745 (loop
> MEM_CGROUP_MAX_RECLAIM_LOOPS
))
1750 if (!mem_cgroup_reclaimable(victim
, false))
1752 total
+= mem_cgroup_shrink_node_zone(victim
, gfp_mask
, false,
1754 *total_scanned
+= nr_scanned
;
1755 if (!res_counter_soft_limit_excess(&root_memcg
->res
))
1758 mem_cgroup_iter_break(root_memcg
, victim
);
1763 * Check OOM-Killer is already running under our hierarchy.
1764 * If someone is running, return false.
1765 * Has to be called with memcg_oom_lock
1767 static bool mem_cgroup_oom_lock(struct mem_cgroup
*memcg
)
1769 struct mem_cgroup
*iter
, *failed
= NULL
;
1771 for_each_mem_cgroup_tree(iter
, memcg
) {
1772 if (iter
->oom_lock
) {
1774 * this subtree of our hierarchy is already locked
1775 * so we cannot give a lock.
1778 mem_cgroup_iter_break(memcg
, iter
);
1781 iter
->oom_lock
= true;
1788 * OK, we failed to lock the whole subtree so we have to clean up
1789 * what we set up to the failing subtree
1791 for_each_mem_cgroup_tree(iter
, memcg
) {
1792 if (iter
== failed
) {
1793 mem_cgroup_iter_break(memcg
, iter
);
1796 iter
->oom_lock
= false;
1802 * Has to be called with memcg_oom_lock
1804 static int mem_cgroup_oom_unlock(struct mem_cgroup
*memcg
)
1806 struct mem_cgroup
*iter
;
1808 for_each_mem_cgroup_tree(iter
, memcg
)
1809 iter
->oom_lock
= false;
1813 static void mem_cgroup_mark_under_oom(struct mem_cgroup
*memcg
)
1815 struct mem_cgroup
*iter
;
1817 for_each_mem_cgroup_tree(iter
, memcg
)
1818 atomic_inc(&iter
->under_oom
);
1821 static void mem_cgroup_unmark_under_oom(struct mem_cgroup
*memcg
)
1823 struct mem_cgroup
*iter
;
1826 * When a new child is created while the hierarchy is under oom,
1827 * mem_cgroup_oom_lock() may not be called. We have to use
1828 * atomic_add_unless() here.
1830 for_each_mem_cgroup_tree(iter
, memcg
)
1831 atomic_add_unless(&iter
->under_oom
, -1, 0);
1834 static DEFINE_SPINLOCK(memcg_oom_lock
);
1835 static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq
);
1837 struct oom_wait_info
{
1838 struct mem_cgroup
*memcg
;
1842 static int memcg_oom_wake_function(wait_queue_t
*wait
,
1843 unsigned mode
, int sync
, void *arg
)
1845 struct mem_cgroup
*wake_memcg
= (struct mem_cgroup
*)arg
;
1846 struct mem_cgroup
*oom_wait_memcg
;
1847 struct oom_wait_info
*oom_wait_info
;
1849 oom_wait_info
= container_of(wait
, struct oom_wait_info
, wait
);
1850 oom_wait_memcg
= oom_wait_info
->memcg
;
1853 * Both of oom_wait_info->memcg and wake_memcg are stable under us.
1854 * Then we can use css_is_ancestor without taking care of RCU.
1856 if (!mem_cgroup_same_or_subtree(oom_wait_memcg
, wake_memcg
)
1857 && !mem_cgroup_same_or_subtree(wake_memcg
, oom_wait_memcg
))
1859 return autoremove_wake_function(wait
, mode
, sync
, arg
);
1862 static void memcg_wakeup_oom(struct mem_cgroup
*memcg
)
1864 /* for filtering, pass "memcg" as argument. */
1865 __wake_up(&memcg_oom_waitq
, TASK_NORMAL
, 0, memcg
);
1868 static void memcg_oom_recover(struct mem_cgroup
*memcg
)
1870 if (memcg
&& atomic_read(&memcg
->under_oom
))
1871 memcg_wakeup_oom(memcg
);
1875 * try to call OOM killer. returns false if we should exit memory-reclaim loop.
1877 static bool mem_cgroup_handle_oom(struct mem_cgroup
*memcg
, gfp_t mask
,
1880 struct oom_wait_info owait
;
1881 bool locked
, need_to_kill
;
1883 owait
.memcg
= memcg
;
1884 owait
.wait
.flags
= 0;
1885 owait
.wait
.func
= memcg_oom_wake_function
;
1886 owait
.wait
.private = current
;
1887 INIT_LIST_HEAD(&owait
.wait
.task_list
);
1888 need_to_kill
= true;
1889 mem_cgroup_mark_under_oom(memcg
);
1891 /* At first, try to OOM lock hierarchy under memcg.*/
1892 spin_lock(&memcg_oom_lock
);
1893 locked
= mem_cgroup_oom_lock(memcg
);
1895 * Even if signal_pending(), we can't quit charge() loop without
1896 * accounting. So, UNINTERRUPTIBLE is appropriate. But SIGKILL
1897 * under OOM is always welcomed, use TASK_KILLABLE here.
1899 prepare_to_wait(&memcg_oom_waitq
, &owait
.wait
, TASK_KILLABLE
);
1900 if (!locked
|| memcg
->oom_kill_disable
)
1901 need_to_kill
= false;
1903 mem_cgroup_oom_notify(memcg
);
1904 spin_unlock(&memcg_oom_lock
);
1907 finish_wait(&memcg_oom_waitq
, &owait
.wait
);
1908 mem_cgroup_out_of_memory(memcg
, mask
, order
);
1911 finish_wait(&memcg_oom_waitq
, &owait
.wait
);
1913 spin_lock(&memcg_oom_lock
);
1915 mem_cgroup_oom_unlock(memcg
);
1916 memcg_wakeup_oom(memcg
);
1917 spin_unlock(&memcg_oom_lock
);
1919 mem_cgroup_unmark_under_oom(memcg
);
1921 if (test_thread_flag(TIF_MEMDIE
) || fatal_signal_pending(current
))
1923 /* Give chance to dying process */
1924 schedule_timeout_uninterruptible(1);
1929 * Currently used to update mapped file statistics, but the routine can be
1930 * generalized to update other statistics as well.
1932 * Notes: Race condition
1934 * We usually use page_cgroup_lock() for accessing page_cgroup member but
1935 * it tends to be costly. But considering some conditions, we doesn't need
1936 * to do so _always_.
1938 * Considering "charge", lock_page_cgroup() is not required because all
1939 * file-stat operations happen after a page is attached to radix-tree. There
1940 * are no race with "charge".
1942 * Considering "uncharge", we know that memcg doesn't clear pc->mem_cgroup
1943 * at "uncharge" intentionally. So, we always see valid pc->mem_cgroup even
1944 * if there are race with "uncharge". Statistics itself is properly handled
1947 * Considering "move", this is an only case we see a race. To make the race
1948 * small, we check mm->moving_account and detect there are possibility of race
1949 * If there is, we take a lock.
1952 void __mem_cgroup_begin_update_page_stat(struct page
*page
,
1953 bool *locked
, unsigned long *flags
)
1955 struct mem_cgroup
*memcg
;
1956 struct page_cgroup
*pc
;
1958 pc
= lookup_page_cgroup(page
);
1960 memcg
= pc
->mem_cgroup
;
1961 if (unlikely(!memcg
|| !PageCgroupUsed(pc
)))
1964 * If this memory cgroup is not under account moving, we don't
1965 * need to take move_lock_mem_cgroup(). Because we already hold
1966 * rcu_read_lock(), any calls to move_account will be delayed until
1967 * rcu_read_unlock() if mem_cgroup_stolen() == true.
1969 if (!mem_cgroup_stolen(memcg
))
1972 move_lock_mem_cgroup(memcg
, flags
);
1973 if (memcg
!= pc
->mem_cgroup
|| !PageCgroupUsed(pc
)) {
1974 move_unlock_mem_cgroup(memcg
, flags
);
1980 void __mem_cgroup_end_update_page_stat(struct page
*page
, unsigned long *flags
)
1982 struct page_cgroup
*pc
= lookup_page_cgroup(page
);
1985 * It's guaranteed that pc->mem_cgroup never changes while
1986 * lock is held because a routine modifies pc->mem_cgroup
1987 * should take move_lock_mem_cgroup().
1989 move_unlock_mem_cgroup(pc
->mem_cgroup
, flags
);
1992 void mem_cgroup_update_page_stat(struct page
*page
,
1993 enum mem_cgroup_page_stat_item idx
, int val
)
1995 struct mem_cgroup
*memcg
;
1996 struct page_cgroup
*pc
= lookup_page_cgroup(page
);
1997 unsigned long uninitialized_var(flags
);
1999 if (mem_cgroup_disabled())
2002 memcg
= pc
->mem_cgroup
;
2003 if (unlikely(!memcg
|| !PageCgroupUsed(pc
)))
2007 case MEMCG_NR_FILE_MAPPED
:
2008 idx
= MEM_CGROUP_STAT_FILE_MAPPED
;
2014 this_cpu_add(memcg
->stat
->count
[idx
], val
);
2018 * size of first charge trial. "32" comes from vmscan.c's magic value.
2019 * TODO: maybe necessary to use big numbers in big irons.
2021 #define CHARGE_BATCH 32U
2022 struct memcg_stock_pcp
{
2023 struct mem_cgroup
*cached
; /* this never be root cgroup */
2024 unsigned int nr_pages
;
2025 struct work_struct work
;
2026 unsigned long flags
;
2027 #define FLUSHING_CACHED_CHARGE 0
2029 static DEFINE_PER_CPU(struct memcg_stock_pcp
, memcg_stock
);
2030 static DEFINE_MUTEX(percpu_charge_mutex
);
2033 * Try to consume stocked charge on this cpu. If success, one page is consumed
2034 * from local stock and true is returned. If the stock is 0 or charges from a
2035 * cgroup which is not current target, returns false. This stock will be
2038 static bool consume_stock(struct mem_cgroup
*memcg
)
2040 struct memcg_stock_pcp
*stock
;
2043 stock
= &get_cpu_var(memcg_stock
);
2044 if (memcg
== stock
->cached
&& stock
->nr_pages
)
2046 else /* need to call res_counter_charge */
2048 put_cpu_var(memcg_stock
);
2053 * Returns stocks cached in percpu to res_counter and reset cached information.
2055 static void drain_stock(struct memcg_stock_pcp
*stock
)
2057 struct mem_cgroup
*old
= stock
->cached
;
2059 if (stock
->nr_pages
) {
2060 unsigned long bytes
= stock
->nr_pages
* PAGE_SIZE
;
2062 res_counter_uncharge(&old
->res
, bytes
);
2063 if (do_swap_account
)
2064 res_counter_uncharge(&old
->memsw
, bytes
);
2065 stock
->nr_pages
= 0;
2067 stock
->cached
= NULL
;
2071 * This must be called under preempt disabled or must be called by
2072 * a thread which is pinned to local cpu.
2074 static void drain_local_stock(struct work_struct
*dummy
)
2076 struct memcg_stock_pcp
*stock
= &__get_cpu_var(memcg_stock
);
2078 clear_bit(FLUSHING_CACHED_CHARGE
, &stock
->flags
);
2082 * Cache charges(val) which is from res_counter, to local per_cpu area.
2083 * This will be consumed by consume_stock() function, later.
2085 static void refill_stock(struct mem_cgroup
*memcg
, unsigned int nr_pages
)
2087 struct memcg_stock_pcp
*stock
= &get_cpu_var(memcg_stock
);
2089 if (stock
->cached
!= memcg
) { /* reset if necessary */
2091 stock
->cached
= memcg
;
2093 stock
->nr_pages
+= nr_pages
;
2094 put_cpu_var(memcg_stock
);
2098 * Drains all per-CPU charge caches for given root_memcg resp. subtree
2099 * of the hierarchy under it. sync flag says whether we should block
2100 * until the work is done.
2102 static void drain_all_stock(struct mem_cgroup
*root_memcg
, bool sync
)
2106 /* Notify other cpus that system-wide "drain" is running */
2109 for_each_online_cpu(cpu
) {
2110 struct memcg_stock_pcp
*stock
= &per_cpu(memcg_stock
, cpu
);
2111 struct mem_cgroup
*memcg
;
2113 memcg
= stock
->cached
;
2114 if (!memcg
|| !stock
->nr_pages
)
2116 if (!mem_cgroup_same_or_subtree(root_memcg
, memcg
))
2118 if (!test_and_set_bit(FLUSHING_CACHED_CHARGE
, &stock
->flags
)) {
2120 drain_local_stock(&stock
->work
);
2122 schedule_work_on(cpu
, &stock
->work
);
2130 for_each_online_cpu(cpu
) {
2131 struct memcg_stock_pcp
*stock
= &per_cpu(memcg_stock
, cpu
);
2132 if (test_bit(FLUSHING_CACHED_CHARGE
, &stock
->flags
))
2133 flush_work(&stock
->work
);
2140 * Tries to drain stocked charges in other cpus. This function is asynchronous
2141 * and just put a work per cpu for draining localy on each cpu. Caller can
2142 * expects some charges will be back to res_counter later but cannot wait for
2145 static void drain_all_stock_async(struct mem_cgroup
*root_memcg
)
2148 * If someone calls draining, avoid adding more kworker runs.
2150 if (!mutex_trylock(&percpu_charge_mutex
))
2152 drain_all_stock(root_memcg
, false);
2153 mutex_unlock(&percpu_charge_mutex
);
2156 /* This is a synchronous drain interface. */
2157 static void drain_all_stock_sync(struct mem_cgroup
*root_memcg
)
2159 /* called when force_empty is called */
2160 mutex_lock(&percpu_charge_mutex
);
2161 drain_all_stock(root_memcg
, true);
2162 mutex_unlock(&percpu_charge_mutex
);
2166 * This function drains percpu counter value from DEAD cpu and
2167 * move it to local cpu. Note that this function can be preempted.
2169 static void mem_cgroup_drain_pcp_counter(struct mem_cgroup
*memcg
, int cpu
)
2173 spin_lock(&memcg
->pcp_counter_lock
);
2174 for (i
= 0; i
< MEM_CGROUP_STAT_NSTATS
; i
++) {
2175 long x
= per_cpu(memcg
->stat
->count
[i
], cpu
);
2177 per_cpu(memcg
->stat
->count
[i
], cpu
) = 0;
2178 memcg
->nocpu_base
.count
[i
] += x
;
2180 for (i
= 0; i
< MEM_CGROUP_EVENTS_NSTATS
; i
++) {
2181 unsigned long x
= per_cpu(memcg
->stat
->events
[i
], cpu
);
2183 per_cpu(memcg
->stat
->events
[i
], cpu
) = 0;
2184 memcg
->nocpu_base
.events
[i
] += x
;
2186 spin_unlock(&memcg
->pcp_counter_lock
);
2189 static int __cpuinit
memcg_cpu_hotplug_callback(struct notifier_block
*nb
,
2190 unsigned long action
,
2193 int cpu
= (unsigned long)hcpu
;
2194 struct memcg_stock_pcp
*stock
;
2195 struct mem_cgroup
*iter
;
2197 if (action
== CPU_ONLINE
)
2200 if (action
!= CPU_DEAD
&& action
!= CPU_DEAD_FROZEN
)
2203 for_each_mem_cgroup(iter
)
2204 mem_cgroup_drain_pcp_counter(iter
, cpu
);
2206 stock
= &per_cpu(memcg_stock
, cpu
);
2212 /* See __mem_cgroup_try_charge() for details */
2214 CHARGE_OK
, /* success */
2215 CHARGE_RETRY
, /* need to retry but retry is not bad */
2216 CHARGE_NOMEM
, /* we can't do more. return -ENOMEM */
2217 CHARGE_WOULDBLOCK
, /* GFP_WAIT wasn't set and no enough res. */
2218 CHARGE_OOM_DIE
, /* the current is killed because of OOM */
2221 static int mem_cgroup_do_charge(struct mem_cgroup
*memcg
, gfp_t gfp_mask
,
2222 unsigned int nr_pages
, bool oom_check
)
2224 unsigned long csize
= nr_pages
* PAGE_SIZE
;
2225 struct mem_cgroup
*mem_over_limit
;
2226 struct res_counter
*fail_res
;
2227 unsigned long flags
= 0;
2230 ret
= res_counter_charge(&memcg
->res
, csize
, &fail_res
);
2233 if (!do_swap_account
)
2235 ret
= res_counter_charge(&memcg
->memsw
, csize
, &fail_res
);
2239 res_counter_uncharge(&memcg
->res
, csize
);
2240 mem_over_limit
= mem_cgroup_from_res_counter(fail_res
, memsw
);
2241 flags
|= MEM_CGROUP_RECLAIM_NOSWAP
;
2243 mem_over_limit
= mem_cgroup_from_res_counter(fail_res
, res
);
2245 * nr_pages can be either a huge page (HPAGE_PMD_NR), a batch
2246 * of regular pages (CHARGE_BATCH), or a single regular page (1).
2248 * Never reclaim on behalf of optional batching, retry with a
2249 * single page instead.
2251 if (nr_pages
== CHARGE_BATCH
)
2252 return CHARGE_RETRY
;
2254 if (!(gfp_mask
& __GFP_WAIT
))
2255 return CHARGE_WOULDBLOCK
;
2257 ret
= mem_cgroup_reclaim(mem_over_limit
, gfp_mask
, flags
);
2258 if (mem_cgroup_margin(mem_over_limit
) >= nr_pages
)
2259 return CHARGE_RETRY
;
2261 * Even though the limit is exceeded at this point, reclaim
2262 * may have been able to free some pages. Retry the charge
2263 * before killing the task.
2265 * Only for regular pages, though: huge pages are rather
2266 * unlikely to succeed so close to the limit, and we fall back
2267 * to regular pages anyway in case of failure.
2269 if (nr_pages
== 1 && ret
)
2270 return CHARGE_RETRY
;
2273 * At task move, charge accounts can be doubly counted. So, it's
2274 * better to wait until the end of task_move if something is going on.
2276 if (mem_cgroup_wait_acct_move(mem_over_limit
))
2277 return CHARGE_RETRY
;
2279 /* If we don't need to call oom-killer at el, return immediately */
2281 return CHARGE_NOMEM
;
2283 if (!mem_cgroup_handle_oom(mem_over_limit
, gfp_mask
, get_order(csize
)))
2284 return CHARGE_OOM_DIE
;
2286 return CHARGE_RETRY
;
2290 * __mem_cgroup_try_charge() does
2291 * 1. detect memcg to be charged against from passed *mm and *ptr,
2292 * 2. update res_counter
2293 * 3. call memory reclaim if necessary.
2295 * In some special case, if the task is fatal, fatal_signal_pending() or
2296 * has TIF_MEMDIE, this function returns -EINTR while writing root_mem_cgroup
2297 * to *ptr. There are two reasons for this. 1: fatal threads should quit as soon
2298 * as possible without any hazards. 2: all pages should have a valid
2299 * pc->mem_cgroup. If mm is NULL and the caller doesn't pass a valid memcg
2300 * pointer, that is treated as a charge to root_mem_cgroup.
2302 * So __mem_cgroup_try_charge() will return
2303 * 0 ... on success, filling *ptr with a valid memcg pointer.
2304 * -ENOMEM ... charge failure because of resource limits.
2305 * -EINTR ... if thread is fatal. *ptr is filled with root_mem_cgroup.
2307 * Unlike the exported interface, an "oom" parameter is added. if oom==true,
2308 * the oom-killer can be invoked.
2310 static int __mem_cgroup_try_charge(struct mm_struct
*mm
,
2312 unsigned int nr_pages
,
2313 struct mem_cgroup
**ptr
,
2316 unsigned int batch
= max(CHARGE_BATCH
, nr_pages
);
2317 int nr_oom_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
2318 struct mem_cgroup
*memcg
= NULL
;
2322 * Unlike gloval-vm's OOM-kill, we're not in memory shortage
2323 * in system level. So, allow to go ahead dying process in addition to
2326 if (unlikely(test_thread_flag(TIF_MEMDIE
)
2327 || fatal_signal_pending(current
)))
2331 * We always charge the cgroup the mm_struct belongs to.
2332 * The mm_struct's mem_cgroup changes on task migration if the
2333 * thread group leader migrates. It's possible that mm is not
2334 * set, if so charge the root memcg (happens for pagecache usage).
2337 *ptr
= root_mem_cgroup
;
2339 if (*ptr
) { /* css should be a valid one */
2341 VM_BUG_ON(css_is_removed(&memcg
->css
));
2342 if (mem_cgroup_is_root(memcg
))
2344 if (nr_pages
== 1 && consume_stock(memcg
))
2346 css_get(&memcg
->css
);
2348 struct task_struct
*p
;
2351 p
= rcu_dereference(mm
->owner
);
2353 * Because we don't have task_lock(), "p" can exit.
2354 * In that case, "memcg" can point to root or p can be NULL with
2355 * race with swapoff. Then, we have small risk of mis-accouning.
2356 * But such kind of mis-account by race always happens because
2357 * we don't have cgroup_mutex(). It's overkill and we allo that
2359 * (*) swapoff at el will charge against mm-struct not against
2360 * task-struct. So, mm->owner can be NULL.
2362 memcg
= mem_cgroup_from_task(p
);
2364 memcg
= root_mem_cgroup
;
2365 if (mem_cgroup_is_root(memcg
)) {
2369 if (nr_pages
== 1 && consume_stock(memcg
)) {
2371 * It seems dagerous to access memcg without css_get().
2372 * But considering how consume_stok works, it's not
2373 * necessary. If consume_stock success, some charges
2374 * from this memcg are cached on this cpu. So, we
2375 * don't need to call css_get()/css_tryget() before
2376 * calling consume_stock().
2381 /* after here, we may be blocked. we need to get refcnt */
2382 if (!css_tryget(&memcg
->css
)) {
2392 /* If killed, bypass charge */
2393 if (fatal_signal_pending(current
)) {
2394 css_put(&memcg
->css
);
2399 if (oom
&& !nr_oom_retries
) {
2401 nr_oom_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
2404 ret
= mem_cgroup_do_charge(memcg
, gfp_mask
, batch
, oom_check
);
2408 case CHARGE_RETRY
: /* not in OOM situation but retry */
2410 css_put(&memcg
->css
);
2413 case CHARGE_WOULDBLOCK
: /* !__GFP_WAIT */
2414 css_put(&memcg
->css
);
2416 case CHARGE_NOMEM
: /* OOM routine works */
2418 css_put(&memcg
->css
);
2421 /* If oom, we never return -ENOMEM */
2424 case CHARGE_OOM_DIE
: /* Killed by OOM Killer */
2425 css_put(&memcg
->css
);
2428 } while (ret
!= CHARGE_OK
);
2430 if (batch
> nr_pages
)
2431 refill_stock(memcg
, batch
- nr_pages
);
2432 css_put(&memcg
->css
);
2440 *ptr
= root_mem_cgroup
;
2445 * Somemtimes we have to undo a charge we got by try_charge().
2446 * This function is for that and do uncharge, put css's refcnt.
2447 * gotten by try_charge().
2449 static void __mem_cgroup_cancel_charge(struct mem_cgroup
*memcg
,
2450 unsigned int nr_pages
)
2452 if (!mem_cgroup_is_root(memcg
)) {
2453 unsigned long bytes
= nr_pages
* PAGE_SIZE
;
2455 res_counter_uncharge(&memcg
->res
, bytes
);
2456 if (do_swap_account
)
2457 res_counter_uncharge(&memcg
->memsw
, bytes
);
2462 * Cancel chrages in this cgroup....doesn't propagate to parent cgroup.
2463 * This is useful when moving usage to parent cgroup.
2465 static void __mem_cgroup_cancel_local_charge(struct mem_cgroup
*memcg
,
2466 unsigned int nr_pages
)
2468 unsigned long bytes
= nr_pages
* PAGE_SIZE
;
2470 if (mem_cgroup_is_root(memcg
))
2473 res_counter_uncharge_until(&memcg
->res
, memcg
->res
.parent
, bytes
);
2474 if (do_swap_account
)
2475 res_counter_uncharge_until(&memcg
->memsw
,
2476 memcg
->memsw
.parent
, bytes
);
2480 * A helper function to get mem_cgroup from ID. must be called under
2481 * rcu_read_lock(). The caller must check css_is_removed() or some if
2482 * it's concern. (dropping refcnt from swap can be called against removed
2485 static struct mem_cgroup
*mem_cgroup_lookup(unsigned short id
)
2487 struct cgroup_subsys_state
*css
;
2489 /* ID 0 is unused ID */
2492 css
= css_lookup(&mem_cgroup_subsys
, id
);
2495 return mem_cgroup_from_css(css
);
2498 struct mem_cgroup
*try_get_mem_cgroup_from_page(struct page
*page
)
2500 struct mem_cgroup
*memcg
= NULL
;
2501 struct page_cgroup
*pc
;
2505 VM_BUG_ON(!PageLocked(page
));
2507 pc
= lookup_page_cgroup(page
);
2508 lock_page_cgroup(pc
);
2509 if (PageCgroupUsed(pc
)) {
2510 memcg
= pc
->mem_cgroup
;
2511 if (memcg
&& !css_tryget(&memcg
->css
))
2513 } else if (PageSwapCache(page
)) {
2514 ent
.val
= page_private(page
);
2515 id
= lookup_swap_cgroup_id(ent
);
2517 memcg
= mem_cgroup_lookup(id
);
2518 if (memcg
&& !css_tryget(&memcg
->css
))
2522 unlock_page_cgroup(pc
);
2526 static void __mem_cgroup_commit_charge(struct mem_cgroup
*memcg
,
2528 unsigned int nr_pages
,
2529 enum charge_type ctype
,
2532 struct page_cgroup
*pc
= lookup_page_cgroup(page
);
2533 struct zone
*uninitialized_var(zone
);
2534 struct lruvec
*lruvec
;
2535 bool was_on_lru
= false;
2538 lock_page_cgroup(pc
);
2539 VM_BUG_ON(PageCgroupUsed(pc
));
2541 * we don't need page_cgroup_lock about tail pages, becase they are not
2542 * accessed by any other context at this point.
2546 * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page
2547 * may already be on some other mem_cgroup's LRU. Take care of it.
2550 zone
= page_zone(page
);
2551 spin_lock_irq(&zone
->lru_lock
);
2552 if (PageLRU(page
)) {
2553 lruvec
= mem_cgroup_zone_lruvec(zone
, pc
->mem_cgroup
);
2555 del_page_from_lru_list(page
, lruvec
, page_lru(page
));
2560 pc
->mem_cgroup
= memcg
;
2562 * We access a page_cgroup asynchronously without lock_page_cgroup().
2563 * Especially when a page_cgroup is taken from a page, pc->mem_cgroup
2564 * is accessed after testing USED bit. To make pc->mem_cgroup visible
2565 * before USED bit, we need memory barrier here.
2566 * See mem_cgroup_add_lru_list(), etc.
2569 SetPageCgroupUsed(pc
);
2573 lruvec
= mem_cgroup_zone_lruvec(zone
, pc
->mem_cgroup
);
2574 VM_BUG_ON(PageLRU(page
));
2576 add_page_to_lru_list(page
, lruvec
, page_lru(page
));
2578 spin_unlock_irq(&zone
->lru_lock
);
2581 if (ctype
== MEM_CGROUP_CHARGE_TYPE_ANON
)
2586 mem_cgroup_charge_statistics(memcg
, anon
, nr_pages
);
2587 unlock_page_cgroup(pc
);
2590 * "charge_statistics" updated event counter. Then, check it.
2591 * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree.
2592 * if they exceeds softlimit.
2594 memcg_check_events(memcg
, page
);
2597 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
2599 #define PCGF_NOCOPY_AT_SPLIT (1 << PCG_LOCK | 1 << PCG_MIGRATION)
2601 * Because tail pages are not marked as "used", set it. We're under
2602 * zone->lru_lock, 'splitting on pmd' and compound_lock.
2603 * charge/uncharge will be never happen and move_account() is done under
2604 * compound_lock(), so we don't have to take care of races.
2606 void mem_cgroup_split_huge_fixup(struct page
*head
)
2608 struct page_cgroup
*head_pc
= lookup_page_cgroup(head
);
2609 struct page_cgroup
*pc
;
2612 if (mem_cgroup_disabled())
2614 for (i
= 1; i
< HPAGE_PMD_NR
; i
++) {
2616 pc
->mem_cgroup
= head_pc
->mem_cgroup
;
2617 smp_wmb();/* see __commit_charge() */
2618 pc
->flags
= head_pc
->flags
& ~PCGF_NOCOPY_AT_SPLIT
;
2621 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
2624 * mem_cgroup_move_account - move account of the page
2626 * @nr_pages: number of regular pages (>1 for huge pages)
2627 * @pc: page_cgroup of the page.
2628 * @from: mem_cgroup which the page is moved from.
2629 * @to: mem_cgroup which the page is moved to. @from != @to.
2631 * The caller must confirm following.
2632 * - page is not on LRU (isolate_page() is useful.)
2633 * - compound_lock is held when nr_pages > 1
2635 * This function doesn't do "charge" to new cgroup and doesn't do "uncharge"
2638 static int mem_cgroup_move_account(struct page
*page
,
2639 unsigned int nr_pages
,
2640 struct page_cgroup
*pc
,
2641 struct mem_cgroup
*from
,
2642 struct mem_cgroup
*to
)
2644 unsigned long flags
;
2646 bool anon
= PageAnon(page
);
2648 VM_BUG_ON(from
== to
);
2649 VM_BUG_ON(PageLRU(page
));
2651 * The page is isolated from LRU. So, collapse function
2652 * will not handle this page. But page splitting can happen.
2653 * Do this check under compound_page_lock(). The caller should
2657 if (nr_pages
> 1 && !PageTransHuge(page
))
2660 lock_page_cgroup(pc
);
2663 if (!PageCgroupUsed(pc
) || pc
->mem_cgroup
!= from
)
2666 move_lock_mem_cgroup(from
, &flags
);
2668 if (!anon
&& page_mapped(page
)) {
2669 /* Update mapped_file data for mem_cgroup */
2671 __this_cpu_dec(from
->stat
->count
[MEM_CGROUP_STAT_FILE_MAPPED
]);
2672 __this_cpu_inc(to
->stat
->count
[MEM_CGROUP_STAT_FILE_MAPPED
]);
2675 mem_cgroup_charge_statistics(from
, anon
, -nr_pages
);
2677 /* caller should have done css_get */
2678 pc
->mem_cgroup
= to
;
2679 mem_cgroup_charge_statistics(to
, anon
, nr_pages
);
2681 * We charges against "to" which may not have any tasks. Then, "to"
2682 * can be under rmdir(). But in current implementation, caller of
2683 * this function is just force_empty() and move charge, so it's
2684 * guaranteed that "to" is never removed. So, we don't check rmdir
2687 move_unlock_mem_cgroup(from
, &flags
);
2690 unlock_page_cgroup(pc
);
2694 memcg_check_events(to
, page
);
2695 memcg_check_events(from
, page
);
2701 * move charges to its parent.
2704 static int mem_cgroup_move_parent(struct page
*page
,
2705 struct page_cgroup
*pc
,
2706 struct mem_cgroup
*child
)
2708 struct mem_cgroup
*parent
;
2709 unsigned int nr_pages
;
2710 unsigned long uninitialized_var(flags
);
2714 if (mem_cgroup_is_root(child
))
2718 if (!get_page_unless_zero(page
))
2720 if (isolate_lru_page(page
))
2723 nr_pages
= hpage_nr_pages(page
);
2725 parent
= parent_mem_cgroup(child
);
2727 * If no parent, move charges to root cgroup.
2730 parent
= root_mem_cgroup
;
2733 flags
= compound_lock_irqsave(page
);
2735 ret
= mem_cgroup_move_account(page
, nr_pages
,
2738 __mem_cgroup_cancel_local_charge(child
, nr_pages
);
2741 compound_unlock_irqrestore(page
, flags
);
2742 putback_lru_page(page
);
2750 * Charge the memory controller for page usage.
2752 * 0 if the charge was successful
2753 * < 0 if the cgroup is over its limit
2755 static int mem_cgroup_charge_common(struct page
*page
, struct mm_struct
*mm
,
2756 gfp_t gfp_mask
, enum charge_type ctype
)
2758 struct mem_cgroup
*memcg
= NULL
;
2759 unsigned int nr_pages
= 1;
2763 if (PageTransHuge(page
)) {
2764 nr_pages
<<= compound_order(page
);
2765 VM_BUG_ON(!PageTransHuge(page
));
2767 * Never OOM-kill a process for a huge page. The
2768 * fault handler will fall back to regular pages.
2773 ret
= __mem_cgroup_try_charge(mm
, gfp_mask
, nr_pages
, &memcg
, oom
);
2776 __mem_cgroup_commit_charge(memcg
, page
, nr_pages
, ctype
, false);
2780 int mem_cgroup_newpage_charge(struct page
*page
,
2781 struct mm_struct
*mm
, gfp_t gfp_mask
)
2783 if (mem_cgroup_disabled())
2785 VM_BUG_ON(page_mapped(page
));
2786 VM_BUG_ON(page
->mapping
&& !PageAnon(page
));
2788 return mem_cgroup_charge_common(page
, mm
, gfp_mask
,
2789 MEM_CGROUP_CHARGE_TYPE_ANON
);
2793 * While swap-in, try_charge -> commit or cancel, the page is locked.
2794 * And when try_charge() successfully returns, one refcnt to memcg without
2795 * struct page_cgroup is acquired. This refcnt will be consumed by
2796 * "commit()" or removed by "cancel()"
2798 static int __mem_cgroup_try_charge_swapin(struct mm_struct
*mm
,
2801 struct mem_cgroup
**memcgp
)
2803 struct mem_cgroup
*memcg
;
2804 struct page_cgroup
*pc
;
2807 pc
= lookup_page_cgroup(page
);
2809 * Every swap fault against a single page tries to charge the
2810 * page, bail as early as possible. shmem_unuse() encounters
2811 * already charged pages, too. The USED bit is protected by
2812 * the page lock, which serializes swap cache removal, which
2813 * in turn serializes uncharging.
2815 if (PageCgroupUsed(pc
))
2817 if (!do_swap_account
)
2819 memcg
= try_get_mem_cgroup_from_page(page
);
2823 ret
= __mem_cgroup_try_charge(NULL
, mask
, 1, memcgp
, true);
2824 css_put(&memcg
->css
);
2829 ret
= __mem_cgroup_try_charge(mm
, mask
, 1, memcgp
, true);
2835 int mem_cgroup_try_charge_swapin(struct mm_struct
*mm
, struct page
*page
,
2836 gfp_t gfp_mask
, struct mem_cgroup
**memcgp
)
2839 if (mem_cgroup_disabled())
2842 * A racing thread's fault, or swapoff, may have already
2843 * updated the pte, and even removed page from swap cache: in
2844 * those cases unuse_pte()'s pte_same() test will fail; but
2845 * there's also a KSM case which does need to charge the page.
2847 if (!PageSwapCache(page
)) {
2850 ret
= __mem_cgroup_try_charge(mm
, gfp_mask
, 1, memcgp
, true);
2855 return __mem_cgroup_try_charge_swapin(mm
, page
, gfp_mask
, memcgp
);
2858 void mem_cgroup_cancel_charge_swapin(struct mem_cgroup
*memcg
)
2860 if (mem_cgroup_disabled())
2864 __mem_cgroup_cancel_charge(memcg
, 1);
2868 __mem_cgroup_commit_charge_swapin(struct page
*page
, struct mem_cgroup
*memcg
,
2869 enum charge_type ctype
)
2871 if (mem_cgroup_disabled())
2875 cgroup_exclude_rmdir(&memcg
->css
);
2877 __mem_cgroup_commit_charge(memcg
, page
, 1, ctype
, true);
2879 * Now swap is on-memory. This means this page may be
2880 * counted both as mem and swap....double count.
2881 * Fix it by uncharging from memsw. Basically, this SwapCache is stable
2882 * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page()
2883 * may call delete_from_swap_cache() before reach here.
2885 if (do_swap_account
&& PageSwapCache(page
)) {
2886 swp_entry_t ent
= {.val
= page_private(page
)};
2887 mem_cgroup_uncharge_swap(ent
);
2890 * At swapin, we may charge account against cgroup which has no tasks.
2891 * So, rmdir()->pre_destroy() can be called while we do this charge.
2892 * In that case, we need to call pre_destroy() again. check it here.
2894 cgroup_release_and_wakeup_rmdir(&memcg
->css
);
2897 void mem_cgroup_commit_charge_swapin(struct page
*page
,
2898 struct mem_cgroup
*memcg
)
2900 __mem_cgroup_commit_charge_swapin(page
, memcg
,
2901 MEM_CGROUP_CHARGE_TYPE_ANON
);
2904 int mem_cgroup_cache_charge(struct page
*page
, struct mm_struct
*mm
,
2907 struct mem_cgroup
*memcg
= NULL
;
2908 enum charge_type type
= MEM_CGROUP_CHARGE_TYPE_CACHE
;
2911 if (mem_cgroup_disabled())
2913 if (PageCompound(page
))
2916 if (!PageSwapCache(page
))
2917 ret
= mem_cgroup_charge_common(page
, mm
, gfp_mask
, type
);
2918 else { /* page is swapcache/shmem */
2919 ret
= __mem_cgroup_try_charge_swapin(mm
, page
,
2922 __mem_cgroup_commit_charge_swapin(page
, memcg
, type
);
2927 static void mem_cgroup_do_uncharge(struct mem_cgroup
*memcg
,
2928 unsigned int nr_pages
,
2929 const enum charge_type ctype
)
2931 struct memcg_batch_info
*batch
= NULL
;
2932 bool uncharge_memsw
= true;
2934 /* If swapout, usage of swap doesn't decrease */
2935 if (!do_swap_account
|| ctype
== MEM_CGROUP_CHARGE_TYPE_SWAPOUT
)
2936 uncharge_memsw
= false;
2938 batch
= ¤t
->memcg_batch
;
2940 * In usual, we do css_get() when we remember memcg pointer.
2941 * But in this case, we keep res->usage until end of a series of
2942 * uncharges. Then, it's ok to ignore memcg's refcnt.
2945 batch
->memcg
= memcg
;
2947 * do_batch > 0 when unmapping pages or inode invalidate/truncate.
2948 * In those cases, all pages freed continuously can be expected to be in
2949 * the same cgroup and we have chance to coalesce uncharges.
2950 * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE)
2951 * because we want to do uncharge as soon as possible.
2954 if (!batch
->do_batch
|| test_thread_flag(TIF_MEMDIE
))
2955 goto direct_uncharge
;
2958 goto direct_uncharge
;
2961 * In typical case, batch->memcg == mem. This means we can
2962 * merge a series of uncharges to an uncharge of res_counter.
2963 * If not, we uncharge res_counter ony by one.
2965 if (batch
->memcg
!= memcg
)
2966 goto direct_uncharge
;
2967 /* remember freed charge and uncharge it later */
2970 batch
->memsw_nr_pages
++;
2973 res_counter_uncharge(&memcg
->res
, nr_pages
* PAGE_SIZE
);
2975 res_counter_uncharge(&memcg
->memsw
, nr_pages
* PAGE_SIZE
);
2976 if (unlikely(batch
->memcg
!= memcg
))
2977 memcg_oom_recover(memcg
);
2981 * uncharge if !page_mapped(page)
2983 static struct mem_cgroup
*
2984 __mem_cgroup_uncharge_common(struct page
*page
, enum charge_type ctype
,
2987 struct mem_cgroup
*memcg
= NULL
;
2988 unsigned int nr_pages
= 1;
2989 struct page_cgroup
*pc
;
2992 if (mem_cgroup_disabled())
2995 VM_BUG_ON(PageSwapCache(page
));
2997 if (PageTransHuge(page
)) {
2998 nr_pages
<<= compound_order(page
);
2999 VM_BUG_ON(!PageTransHuge(page
));
3002 * Check if our page_cgroup is valid
3004 pc
= lookup_page_cgroup(page
);
3005 if (unlikely(!PageCgroupUsed(pc
)))
3008 lock_page_cgroup(pc
);
3010 memcg
= pc
->mem_cgroup
;
3012 if (!PageCgroupUsed(pc
))
3015 anon
= PageAnon(page
);
3018 case MEM_CGROUP_CHARGE_TYPE_ANON
:
3020 * Generally PageAnon tells if it's the anon statistics to be
3021 * updated; but sometimes e.g. mem_cgroup_uncharge_page() is
3022 * used before page reached the stage of being marked PageAnon.
3026 case MEM_CGROUP_CHARGE_TYPE_DROP
:
3027 /* See mem_cgroup_prepare_migration() */
3028 if (page_mapped(page
))
3031 * Pages under migration may not be uncharged. But
3032 * end_migration() /must/ be the one uncharging the
3033 * unused post-migration page and so it has to call
3034 * here with the migration bit still set. See the
3035 * res_counter handling below.
3037 if (!end_migration
&& PageCgroupMigration(pc
))
3040 case MEM_CGROUP_CHARGE_TYPE_SWAPOUT
:
3041 if (!PageAnon(page
)) { /* Shared memory */
3042 if (page
->mapping
&& !page_is_file_cache(page
))
3044 } else if (page_mapped(page
)) /* Anon */
3051 mem_cgroup_charge_statistics(memcg
, anon
, -nr_pages
);
3053 ClearPageCgroupUsed(pc
);
3055 * pc->mem_cgroup is not cleared here. It will be accessed when it's
3056 * freed from LRU. This is safe because uncharged page is expected not
3057 * to be reused (freed soon). Exception is SwapCache, it's handled by
3058 * special functions.
3061 unlock_page_cgroup(pc
);
3063 * even after unlock, we have memcg->res.usage here and this memcg
3064 * will never be freed.
3066 memcg_check_events(memcg
, page
);
3067 if (do_swap_account
&& ctype
== MEM_CGROUP_CHARGE_TYPE_SWAPOUT
) {
3068 mem_cgroup_swap_statistics(memcg
, true);
3069 mem_cgroup_get(memcg
);
3072 * Migration does not charge the res_counter for the
3073 * replacement page, so leave it alone when phasing out the
3074 * page that is unused after the migration.
3076 if (!end_migration
&& !mem_cgroup_is_root(memcg
))
3077 mem_cgroup_do_uncharge(memcg
, nr_pages
, ctype
);
3082 unlock_page_cgroup(pc
);
3086 void mem_cgroup_uncharge_page(struct page
*page
)
3089 if (page_mapped(page
))
3091 VM_BUG_ON(page
->mapping
&& !PageAnon(page
));
3092 if (PageSwapCache(page
))
3094 __mem_cgroup_uncharge_common(page
, MEM_CGROUP_CHARGE_TYPE_ANON
, false);
3097 void mem_cgroup_uncharge_cache_page(struct page
*page
)
3099 VM_BUG_ON(page_mapped(page
));
3100 VM_BUG_ON(page
->mapping
);
3101 __mem_cgroup_uncharge_common(page
, MEM_CGROUP_CHARGE_TYPE_CACHE
, false);
3105 * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate.
3106 * In that cases, pages are freed continuously and we can expect pages
3107 * are in the same memcg. All these calls itself limits the number of
3108 * pages freed at once, then uncharge_start/end() is called properly.
3109 * This may be called prural(2) times in a context,
3112 void mem_cgroup_uncharge_start(void)
3114 current
->memcg_batch
.do_batch
++;
3115 /* We can do nest. */
3116 if (current
->memcg_batch
.do_batch
== 1) {
3117 current
->memcg_batch
.memcg
= NULL
;
3118 current
->memcg_batch
.nr_pages
= 0;
3119 current
->memcg_batch
.memsw_nr_pages
= 0;
3123 void mem_cgroup_uncharge_end(void)
3125 struct memcg_batch_info
*batch
= ¤t
->memcg_batch
;
3127 if (!batch
->do_batch
)
3131 if (batch
->do_batch
) /* If stacked, do nothing. */
3137 * This "batch->memcg" is valid without any css_get/put etc...
3138 * bacause we hide charges behind us.
3140 if (batch
->nr_pages
)
3141 res_counter_uncharge(&batch
->memcg
->res
,
3142 batch
->nr_pages
* PAGE_SIZE
);
3143 if (batch
->memsw_nr_pages
)
3144 res_counter_uncharge(&batch
->memcg
->memsw
,
3145 batch
->memsw_nr_pages
* PAGE_SIZE
);
3146 memcg_oom_recover(batch
->memcg
);
3147 /* forget this pointer (for sanity check) */
3148 batch
->memcg
= NULL
;
3153 * called after __delete_from_swap_cache() and drop "page" account.
3154 * memcg information is recorded to swap_cgroup of "ent"
3157 mem_cgroup_uncharge_swapcache(struct page
*page
, swp_entry_t ent
, bool swapout
)
3159 struct mem_cgroup
*memcg
;
3160 int ctype
= MEM_CGROUP_CHARGE_TYPE_SWAPOUT
;
3162 if (!swapout
) /* this was a swap cache but the swap is unused ! */
3163 ctype
= MEM_CGROUP_CHARGE_TYPE_DROP
;
3165 memcg
= __mem_cgroup_uncharge_common(page
, ctype
, false);
3168 * record memcg information, if swapout && memcg != NULL,
3169 * mem_cgroup_get() was called in uncharge().
3171 if (do_swap_account
&& swapout
&& memcg
)
3172 swap_cgroup_record(ent
, css_id(&memcg
->css
));
3176 #ifdef CONFIG_MEMCG_SWAP
3178 * called from swap_entry_free(). remove record in swap_cgroup and
3179 * uncharge "memsw" account.
3181 void mem_cgroup_uncharge_swap(swp_entry_t ent
)
3183 struct mem_cgroup
*memcg
;
3186 if (!do_swap_account
)
3189 id
= swap_cgroup_record(ent
, 0);
3191 memcg
= mem_cgroup_lookup(id
);
3194 * We uncharge this because swap is freed.
3195 * This memcg can be obsolete one. We avoid calling css_tryget
3197 if (!mem_cgroup_is_root(memcg
))
3198 res_counter_uncharge(&memcg
->memsw
, PAGE_SIZE
);
3199 mem_cgroup_swap_statistics(memcg
, false);
3200 mem_cgroup_put(memcg
);
3206 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
3207 * @entry: swap entry to be moved
3208 * @from: mem_cgroup which the entry is moved from
3209 * @to: mem_cgroup which the entry is moved to
3211 * It succeeds only when the swap_cgroup's record for this entry is the same
3212 * as the mem_cgroup's id of @from.
3214 * Returns 0 on success, -EINVAL on failure.
3216 * The caller must have charged to @to, IOW, called res_counter_charge() about
3217 * both res and memsw, and called css_get().
3219 static int mem_cgroup_move_swap_account(swp_entry_t entry
,
3220 struct mem_cgroup
*from
, struct mem_cgroup
*to
)
3222 unsigned short old_id
, new_id
;
3224 old_id
= css_id(&from
->css
);
3225 new_id
= css_id(&to
->css
);
3227 if (swap_cgroup_cmpxchg(entry
, old_id
, new_id
) == old_id
) {
3228 mem_cgroup_swap_statistics(from
, false);
3229 mem_cgroup_swap_statistics(to
, true);
3231 * This function is only called from task migration context now.
3232 * It postpones res_counter and refcount handling till the end
3233 * of task migration(mem_cgroup_clear_mc()) for performance
3234 * improvement. But we cannot postpone mem_cgroup_get(to)
3235 * because if the process that has been moved to @to does
3236 * swap-in, the refcount of @to might be decreased to 0.
3244 static inline int mem_cgroup_move_swap_account(swp_entry_t entry
,
3245 struct mem_cgroup
*from
, struct mem_cgroup
*to
)
3252 * Before starting migration, account PAGE_SIZE to mem_cgroup that the old
3255 void mem_cgroup_prepare_migration(struct page
*page
, struct page
*newpage
,
3256 struct mem_cgroup
**memcgp
)
3258 struct mem_cgroup
*memcg
= NULL
;
3259 struct page_cgroup
*pc
;
3260 enum charge_type ctype
;
3264 VM_BUG_ON(PageTransHuge(page
));
3265 if (mem_cgroup_disabled())
3268 pc
= lookup_page_cgroup(page
);
3269 lock_page_cgroup(pc
);
3270 if (PageCgroupUsed(pc
)) {
3271 memcg
= pc
->mem_cgroup
;
3272 css_get(&memcg
->css
);
3274 * At migrating an anonymous page, its mapcount goes down
3275 * to 0 and uncharge() will be called. But, even if it's fully
3276 * unmapped, migration may fail and this page has to be
3277 * charged again. We set MIGRATION flag here and delay uncharge
3278 * until end_migration() is called
3280 * Corner Case Thinking
3282 * When the old page was mapped as Anon and it's unmap-and-freed
3283 * while migration was ongoing.
3284 * If unmap finds the old page, uncharge() of it will be delayed
3285 * until end_migration(). If unmap finds a new page, it's
3286 * uncharged when it make mapcount to be 1->0. If unmap code
3287 * finds swap_migration_entry, the new page will not be mapped
3288 * and end_migration() will find it(mapcount==0).
3291 * When the old page was mapped but migraion fails, the kernel
3292 * remaps it. A charge for it is kept by MIGRATION flag even
3293 * if mapcount goes down to 0. We can do remap successfully
3294 * without charging it again.
3297 * The "old" page is under lock_page() until the end of
3298 * migration, so, the old page itself will not be swapped-out.
3299 * If the new page is swapped out before end_migraton, our
3300 * hook to usual swap-out path will catch the event.
3303 SetPageCgroupMigration(pc
);
3305 unlock_page_cgroup(pc
);
3307 * If the page is not charged at this point,
3315 * We charge new page before it's used/mapped. So, even if unlock_page()
3316 * is called before end_migration, we can catch all events on this new
3317 * page. In the case new page is migrated but not remapped, new page's
3318 * mapcount will be finally 0 and we call uncharge in end_migration().
3321 ctype
= MEM_CGROUP_CHARGE_TYPE_ANON
;
3323 ctype
= MEM_CGROUP_CHARGE_TYPE_CACHE
;
3325 * The page is committed to the memcg, but it's not actually
3326 * charged to the res_counter since we plan on replacing the
3327 * old one and only one page is going to be left afterwards.
3329 __mem_cgroup_commit_charge(memcg
, newpage
, 1, ctype
, false);
3332 /* remove redundant charge if migration failed*/
3333 void mem_cgroup_end_migration(struct mem_cgroup
*memcg
,
3334 struct page
*oldpage
, struct page
*newpage
, bool migration_ok
)
3336 struct page
*used
, *unused
;
3337 struct page_cgroup
*pc
;
3342 /* blocks rmdir() */
3343 cgroup_exclude_rmdir(&memcg
->css
);
3344 if (!migration_ok
) {
3351 anon
= PageAnon(used
);
3352 __mem_cgroup_uncharge_common(unused
,
3353 anon
? MEM_CGROUP_CHARGE_TYPE_ANON
3354 : MEM_CGROUP_CHARGE_TYPE_CACHE
,
3356 css_put(&memcg
->css
);
3358 * We disallowed uncharge of pages under migration because mapcount
3359 * of the page goes down to zero, temporarly.
3360 * Clear the flag and check the page should be charged.
3362 pc
= lookup_page_cgroup(oldpage
);
3363 lock_page_cgroup(pc
);
3364 ClearPageCgroupMigration(pc
);
3365 unlock_page_cgroup(pc
);
3368 * If a page is a file cache, radix-tree replacement is very atomic
3369 * and we can skip this check. When it was an Anon page, its mapcount
3370 * goes down to 0. But because we added MIGRATION flage, it's not
3371 * uncharged yet. There are several case but page->mapcount check
3372 * and USED bit check in mem_cgroup_uncharge_page() will do enough
3373 * check. (see prepare_charge() also)
3376 mem_cgroup_uncharge_page(used
);
3378 * At migration, we may charge account against cgroup which has no
3380 * So, rmdir()->pre_destroy() can be called while we do this charge.
3381 * In that case, we need to call pre_destroy() again. check it here.
3383 cgroup_release_and_wakeup_rmdir(&memcg
->css
);
3387 * At replace page cache, newpage is not under any memcg but it's on
3388 * LRU. So, this function doesn't touch res_counter but handles LRU
3389 * in correct way. Both pages are locked so we cannot race with uncharge.
3391 void mem_cgroup_replace_page_cache(struct page
*oldpage
,
3392 struct page
*newpage
)
3394 struct mem_cgroup
*memcg
= NULL
;
3395 struct page_cgroup
*pc
;
3396 enum charge_type type
= MEM_CGROUP_CHARGE_TYPE_CACHE
;
3398 if (mem_cgroup_disabled())
3401 pc
= lookup_page_cgroup(oldpage
);
3402 /* fix accounting on old pages */
3403 lock_page_cgroup(pc
);
3404 if (PageCgroupUsed(pc
)) {
3405 memcg
= pc
->mem_cgroup
;
3406 mem_cgroup_charge_statistics(memcg
, false, -1);
3407 ClearPageCgroupUsed(pc
);
3409 unlock_page_cgroup(pc
);
3412 * When called from shmem_replace_page(), in some cases the
3413 * oldpage has already been charged, and in some cases not.
3418 * Even if newpage->mapping was NULL before starting replacement,
3419 * the newpage may be on LRU(or pagevec for LRU) already. We lock
3420 * LRU while we overwrite pc->mem_cgroup.
3422 __mem_cgroup_commit_charge(memcg
, newpage
, 1, type
, true);
3425 #ifdef CONFIG_DEBUG_VM
3426 static struct page_cgroup
*lookup_page_cgroup_used(struct page
*page
)
3428 struct page_cgroup
*pc
;
3430 pc
= lookup_page_cgroup(page
);
3432 * Can be NULL while feeding pages into the page allocator for
3433 * the first time, i.e. during boot or memory hotplug;
3434 * or when mem_cgroup_disabled().
3436 if (likely(pc
) && PageCgroupUsed(pc
))
3441 bool mem_cgroup_bad_page_check(struct page
*page
)
3443 if (mem_cgroup_disabled())
3446 return lookup_page_cgroup_used(page
) != NULL
;
3449 void mem_cgroup_print_bad_page(struct page
*page
)
3451 struct page_cgroup
*pc
;
3453 pc
= lookup_page_cgroup_used(page
);
3455 printk(KERN_ALERT
"pc:%p pc->flags:%lx pc->mem_cgroup:%p\n",
3456 pc
, pc
->flags
, pc
->mem_cgroup
);
3461 static DEFINE_MUTEX(set_limit_mutex
);
3463 static int mem_cgroup_resize_limit(struct mem_cgroup
*memcg
,
3464 unsigned long long val
)
3467 u64 memswlimit
, memlimit
;
3469 int children
= mem_cgroup_count_children(memcg
);
3470 u64 curusage
, oldusage
;
3474 * For keeping hierarchical_reclaim simple, how long we should retry
3475 * is depends on callers. We set our retry-count to be function
3476 * of # of children which we should visit in this loop.
3478 retry_count
= MEM_CGROUP_RECLAIM_RETRIES
* children
;
3480 oldusage
= res_counter_read_u64(&memcg
->res
, RES_USAGE
);
3483 while (retry_count
) {
3484 if (signal_pending(current
)) {
3489 * Rather than hide all in some function, I do this in
3490 * open coded manner. You see what this really does.
3491 * We have to guarantee memcg->res.limit <= memcg->memsw.limit.
3493 mutex_lock(&set_limit_mutex
);
3494 memswlimit
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
3495 if (memswlimit
< val
) {
3497 mutex_unlock(&set_limit_mutex
);
3501 memlimit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
3505 ret
= res_counter_set_limit(&memcg
->res
, val
);
3507 if (memswlimit
== val
)
3508 memcg
->memsw_is_minimum
= true;
3510 memcg
->memsw_is_minimum
= false;
3512 mutex_unlock(&set_limit_mutex
);
3517 mem_cgroup_reclaim(memcg
, GFP_KERNEL
,
3518 MEM_CGROUP_RECLAIM_SHRINK
);
3519 curusage
= res_counter_read_u64(&memcg
->res
, RES_USAGE
);
3520 /* Usage is reduced ? */
3521 if (curusage
>= oldusage
)
3524 oldusage
= curusage
;
3526 if (!ret
&& enlarge
)
3527 memcg_oom_recover(memcg
);
3532 static int mem_cgroup_resize_memsw_limit(struct mem_cgroup
*memcg
,
3533 unsigned long long val
)
3536 u64 memlimit
, memswlimit
, oldusage
, curusage
;
3537 int children
= mem_cgroup_count_children(memcg
);
3541 /* see mem_cgroup_resize_res_limit */
3542 retry_count
= children
* MEM_CGROUP_RECLAIM_RETRIES
;
3543 oldusage
= res_counter_read_u64(&memcg
->memsw
, RES_USAGE
);
3544 while (retry_count
) {
3545 if (signal_pending(current
)) {
3550 * Rather than hide all in some function, I do this in
3551 * open coded manner. You see what this really does.
3552 * We have to guarantee memcg->res.limit <= memcg->memsw.limit.
3554 mutex_lock(&set_limit_mutex
);
3555 memlimit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
3556 if (memlimit
> val
) {
3558 mutex_unlock(&set_limit_mutex
);
3561 memswlimit
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
3562 if (memswlimit
< val
)
3564 ret
= res_counter_set_limit(&memcg
->memsw
, val
);
3566 if (memlimit
== val
)
3567 memcg
->memsw_is_minimum
= true;
3569 memcg
->memsw_is_minimum
= false;
3571 mutex_unlock(&set_limit_mutex
);
3576 mem_cgroup_reclaim(memcg
, GFP_KERNEL
,
3577 MEM_CGROUP_RECLAIM_NOSWAP
|
3578 MEM_CGROUP_RECLAIM_SHRINK
);
3579 curusage
= res_counter_read_u64(&memcg
->memsw
, RES_USAGE
);
3580 /* Usage is reduced ? */
3581 if (curusage
>= oldusage
)
3584 oldusage
= curusage
;
3586 if (!ret
&& enlarge
)
3587 memcg_oom_recover(memcg
);
3591 unsigned long mem_cgroup_soft_limit_reclaim(struct zone
*zone
, int order
,
3593 unsigned long *total_scanned
)
3595 unsigned long nr_reclaimed
= 0;
3596 struct mem_cgroup_per_zone
*mz
, *next_mz
= NULL
;
3597 unsigned long reclaimed
;
3599 struct mem_cgroup_tree_per_zone
*mctz
;
3600 unsigned long long excess
;
3601 unsigned long nr_scanned
;
3606 mctz
= soft_limit_tree_node_zone(zone_to_nid(zone
), zone_idx(zone
));
3608 * This loop can run a while, specially if mem_cgroup's continuously
3609 * keep exceeding their soft limit and putting the system under
3616 mz
= mem_cgroup_largest_soft_limit_node(mctz
);
3621 reclaimed
= mem_cgroup_soft_reclaim(mz
->memcg
, zone
,
3622 gfp_mask
, &nr_scanned
);
3623 nr_reclaimed
+= reclaimed
;
3624 *total_scanned
+= nr_scanned
;
3625 spin_lock(&mctz
->lock
);
3628 * If we failed to reclaim anything from this memory cgroup
3629 * it is time to move on to the next cgroup
3635 * Loop until we find yet another one.
3637 * By the time we get the soft_limit lock
3638 * again, someone might have aded the
3639 * group back on the RB tree. Iterate to
3640 * make sure we get a different mem.
3641 * mem_cgroup_largest_soft_limit_node returns
3642 * NULL if no other cgroup is present on
3646 __mem_cgroup_largest_soft_limit_node(mctz
);
3648 css_put(&next_mz
->memcg
->css
);
3649 else /* next_mz == NULL or other memcg */
3653 __mem_cgroup_remove_exceeded(mz
->memcg
, mz
, mctz
);
3654 excess
= res_counter_soft_limit_excess(&mz
->memcg
->res
);
3656 * One school of thought says that we should not add
3657 * back the node to the tree if reclaim returns 0.
3658 * But our reclaim could return 0, simply because due
3659 * to priority we are exposing a smaller subset of
3660 * memory to reclaim from. Consider this as a longer
3663 /* If excess == 0, no tree ops */
3664 __mem_cgroup_insert_exceeded(mz
->memcg
, mz
, mctz
, excess
);
3665 spin_unlock(&mctz
->lock
);
3666 css_put(&mz
->memcg
->css
);
3669 * Could not reclaim anything and there are no more
3670 * mem cgroups to try or we seem to be looping without
3671 * reclaiming anything.
3673 if (!nr_reclaimed
&&
3675 loop
> MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS
))
3677 } while (!nr_reclaimed
);
3679 css_put(&next_mz
->memcg
->css
);
3680 return nr_reclaimed
;
3684 * Traverse a specified page_cgroup list and try to drop them all. This doesn't
3685 * reclaim the pages page themselves - it just removes the page_cgroups.
3686 * Returns true if some page_cgroups were not freed, indicating that the caller
3687 * must retry this operation.
3689 static bool mem_cgroup_force_empty_list(struct mem_cgroup
*memcg
,
3690 int node
, int zid
, enum lru_list lru
)
3692 struct mem_cgroup_per_zone
*mz
;
3693 unsigned long flags
, loop
;
3694 struct list_head
*list
;
3698 zone
= &NODE_DATA(node
)->node_zones
[zid
];
3699 mz
= mem_cgroup_zoneinfo(memcg
, node
, zid
);
3700 list
= &mz
->lruvec
.lists
[lru
];
3702 loop
= mz
->lru_size
[lru
];
3703 /* give some margin against EBUSY etc...*/
3707 struct page_cgroup
*pc
;
3710 spin_lock_irqsave(&zone
->lru_lock
, flags
);
3711 if (list_empty(list
)) {
3712 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
3715 page
= list_entry(list
->prev
, struct page
, lru
);
3717 list_move(&page
->lru
, list
);
3719 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
3722 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
3724 pc
= lookup_page_cgroup(page
);
3726 if (mem_cgroup_move_parent(page
, pc
, memcg
)) {
3727 /* found lock contention or "pc" is obsolete. */
3733 return !list_empty(list
);
3737 * make mem_cgroup's charge to be 0 if there is no task.
3738 * This enables deleting this mem_cgroup.
3740 static int mem_cgroup_force_empty(struct mem_cgroup
*memcg
, bool free_all
)
3743 int node
, zid
, shrink
;
3744 int nr_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
3745 struct cgroup
*cgrp
= memcg
->css
.cgroup
;
3747 css_get(&memcg
->css
);
3750 /* should free all ? */
3756 if (cgroup_task_count(cgrp
) || !list_empty(&cgrp
->children
))
3758 /* This is for making all *used* pages to be on LRU. */
3759 lru_add_drain_all();
3760 drain_all_stock_sync(memcg
);
3762 mem_cgroup_start_move(memcg
);
3763 for_each_node_state(node
, N_HIGH_MEMORY
) {
3764 for (zid
= 0; !ret
&& zid
< MAX_NR_ZONES
; zid
++) {
3767 ret
= mem_cgroup_force_empty_list(memcg
,
3776 mem_cgroup_end_move(memcg
);
3777 memcg_oom_recover(memcg
);
3779 /* "ret" should also be checked to ensure all lists are empty. */
3780 } while (res_counter_read_u64(&memcg
->res
, RES_USAGE
) > 0 || ret
);
3782 css_put(&memcg
->css
);
3786 /* returns EBUSY if there is a task or if we come here twice. */
3787 if (cgroup_task_count(cgrp
) || !list_empty(&cgrp
->children
) || shrink
) {
3791 /* we call try-to-free pages for make this cgroup empty */
3792 lru_add_drain_all();
3793 /* try to free all pages in this cgroup */
3795 while (nr_retries
&& res_counter_read_u64(&memcg
->res
, RES_USAGE
) > 0) {
3798 if (signal_pending(current
)) {
3802 progress
= try_to_free_mem_cgroup_pages(memcg
, GFP_KERNEL
,
3806 /* maybe some writeback is necessary */
3807 congestion_wait(BLK_RW_ASYNC
, HZ
/10);
3812 /* try move_account...there may be some *locked* pages. */
3816 static int mem_cgroup_force_empty_write(struct cgroup
*cont
, unsigned int event
)
3818 return mem_cgroup_force_empty(mem_cgroup_from_cont(cont
), true);
3822 static u64
mem_cgroup_hierarchy_read(struct cgroup
*cont
, struct cftype
*cft
)
3824 return mem_cgroup_from_cont(cont
)->use_hierarchy
;
3827 static int mem_cgroup_hierarchy_write(struct cgroup
*cont
, struct cftype
*cft
,
3831 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
3832 struct cgroup
*parent
= cont
->parent
;
3833 struct mem_cgroup
*parent_memcg
= NULL
;
3836 parent_memcg
= mem_cgroup_from_cont(parent
);
3840 if (memcg
->use_hierarchy
== val
)
3844 * If parent's use_hierarchy is set, we can't make any modifications
3845 * in the child subtrees. If it is unset, then the change can
3846 * occur, provided the current cgroup has no children.
3848 * For the root cgroup, parent_mem is NULL, we allow value to be
3849 * set if there are no children.
3851 if ((!parent_memcg
|| !parent_memcg
->use_hierarchy
) &&
3852 (val
== 1 || val
== 0)) {
3853 if (list_empty(&cont
->children
))
3854 memcg
->use_hierarchy
= val
;
3867 static unsigned long mem_cgroup_recursive_stat(struct mem_cgroup
*memcg
,
3868 enum mem_cgroup_stat_index idx
)
3870 struct mem_cgroup
*iter
;
3873 /* Per-cpu values can be negative, use a signed accumulator */
3874 for_each_mem_cgroup_tree(iter
, memcg
)
3875 val
+= mem_cgroup_read_stat(iter
, idx
);
3877 if (val
< 0) /* race ? */
3882 static inline u64
mem_cgroup_usage(struct mem_cgroup
*memcg
, bool swap
)
3886 if (!mem_cgroup_is_root(memcg
)) {
3888 return res_counter_read_u64(&memcg
->res
, RES_USAGE
);
3890 return res_counter_read_u64(&memcg
->memsw
, RES_USAGE
);
3893 val
= mem_cgroup_recursive_stat(memcg
, MEM_CGROUP_STAT_CACHE
);
3894 val
+= mem_cgroup_recursive_stat(memcg
, MEM_CGROUP_STAT_RSS
);
3897 val
+= mem_cgroup_recursive_stat(memcg
, MEM_CGROUP_STAT_SWAP
);
3899 return val
<< PAGE_SHIFT
;
3902 static ssize_t
mem_cgroup_read(struct cgroup
*cont
, struct cftype
*cft
,
3903 struct file
*file
, char __user
*buf
,
3904 size_t nbytes
, loff_t
*ppos
)
3906 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
3909 int type
, name
, len
;
3911 type
= MEMFILE_TYPE(cft
->private);
3912 name
= MEMFILE_ATTR(cft
->private);
3914 if (!do_swap_account
&& type
== _MEMSWAP
)
3919 if (name
== RES_USAGE
)
3920 val
= mem_cgroup_usage(memcg
, false);
3922 val
= res_counter_read_u64(&memcg
->res
, name
);
3925 if (name
== RES_USAGE
)
3926 val
= mem_cgroup_usage(memcg
, true);
3928 val
= res_counter_read_u64(&memcg
->memsw
, name
);
3934 len
= scnprintf(str
, sizeof(str
), "%llu\n", (unsigned long long)val
);
3935 return simple_read_from_buffer(buf
, nbytes
, ppos
, str
, len
);
3938 * The user of this function is...
3941 static int mem_cgroup_write(struct cgroup
*cont
, struct cftype
*cft
,
3944 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
3946 unsigned long long val
;
3949 type
= MEMFILE_TYPE(cft
->private);
3950 name
= MEMFILE_ATTR(cft
->private);
3952 if (!do_swap_account
&& type
== _MEMSWAP
)
3957 if (mem_cgroup_is_root(memcg
)) { /* Can't set limit on root */
3961 /* This function does all necessary parse...reuse it */
3962 ret
= res_counter_memparse_write_strategy(buffer
, &val
);
3966 ret
= mem_cgroup_resize_limit(memcg
, val
);
3968 ret
= mem_cgroup_resize_memsw_limit(memcg
, val
);
3970 case RES_SOFT_LIMIT
:
3971 ret
= res_counter_memparse_write_strategy(buffer
, &val
);
3975 * For memsw, soft limits are hard to implement in terms
3976 * of semantics, for now, we support soft limits for
3977 * control without swap
3980 ret
= res_counter_set_soft_limit(&memcg
->res
, val
);
3985 ret
= -EINVAL
; /* should be BUG() ? */
3991 static void memcg_get_hierarchical_limit(struct mem_cgroup
*memcg
,
3992 unsigned long long *mem_limit
, unsigned long long *memsw_limit
)
3994 struct cgroup
*cgroup
;
3995 unsigned long long min_limit
, min_memsw_limit
, tmp
;
3997 min_limit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
3998 min_memsw_limit
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
3999 cgroup
= memcg
->css
.cgroup
;
4000 if (!memcg
->use_hierarchy
)
4003 while (cgroup
->parent
) {
4004 cgroup
= cgroup
->parent
;
4005 memcg
= mem_cgroup_from_cont(cgroup
);
4006 if (!memcg
->use_hierarchy
)
4008 tmp
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
4009 min_limit
= min(min_limit
, tmp
);
4010 tmp
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
4011 min_memsw_limit
= min(min_memsw_limit
, tmp
);
4014 *mem_limit
= min_limit
;
4015 *memsw_limit
= min_memsw_limit
;
4018 static int mem_cgroup_reset(struct cgroup
*cont
, unsigned int event
)
4020 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
4023 type
= MEMFILE_TYPE(event
);
4024 name
= MEMFILE_ATTR(event
);
4026 if (!do_swap_account
&& type
== _MEMSWAP
)
4032 res_counter_reset_max(&memcg
->res
);
4034 res_counter_reset_max(&memcg
->memsw
);
4038 res_counter_reset_failcnt(&memcg
->res
);
4040 res_counter_reset_failcnt(&memcg
->memsw
);
4047 static u64
mem_cgroup_move_charge_read(struct cgroup
*cgrp
,
4050 return mem_cgroup_from_cont(cgrp
)->move_charge_at_immigrate
;
4054 static int mem_cgroup_move_charge_write(struct cgroup
*cgrp
,
4055 struct cftype
*cft
, u64 val
)
4057 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4059 if (val
>= (1 << NR_MOVE_TYPE
))
4062 * We check this value several times in both in can_attach() and
4063 * attach(), so we need cgroup lock to prevent this value from being
4067 memcg
->move_charge_at_immigrate
= val
;
4073 static int mem_cgroup_move_charge_write(struct cgroup
*cgrp
,
4074 struct cftype
*cft
, u64 val
)
4081 static int memcg_numa_stat_show(struct cgroup
*cont
, struct cftype
*cft
,
4085 unsigned long total_nr
, file_nr
, anon_nr
, unevictable_nr
;
4086 unsigned long node_nr
;
4087 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
4089 total_nr
= mem_cgroup_nr_lru_pages(memcg
, LRU_ALL
);
4090 seq_printf(m
, "total=%lu", total_nr
);
4091 for_each_node_state(nid
, N_HIGH_MEMORY
) {
4092 node_nr
= mem_cgroup_node_nr_lru_pages(memcg
, nid
, LRU_ALL
);
4093 seq_printf(m
, " N%d=%lu", nid
, node_nr
);
4097 file_nr
= mem_cgroup_nr_lru_pages(memcg
, LRU_ALL_FILE
);
4098 seq_printf(m
, "file=%lu", file_nr
);
4099 for_each_node_state(nid
, N_HIGH_MEMORY
) {
4100 node_nr
= mem_cgroup_node_nr_lru_pages(memcg
, nid
,
4102 seq_printf(m
, " N%d=%lu", nid
, node_nr
);
4106 anon_nr
= mem_cgroup_nr_lru_pages(memcg
, LRU_ALL_ANON
);
4107 seq_printf(m
, "anon=%lu", anon_nr
);
4108 for_each_node_state(nid
, N_HIGH_MEMORY
) {
4109 node_nr
= mem_cgroup_node_nr_lru_pages(memcg
, nid
,
4111 seq_printf(m
, " N%d=%lu", nid
, node_nr
);
4115 unevictable_nr
= mem_cgroup_nr_lru_pages(memcg
, BIT(LRU_UNEVICTABLE
));
4116 seq_printf(m
, "unevictable=%lu", unevictable_nr
);
4117 for_each_node_state(nid
, N_HIGH_MEMORY
) {
4118 node_nr
= mem_cgroup_node_nr_lru_pages(memcg
, nid
,
4119 BIT(LRU_UNEVICTABLE
));
4120 seq_printf(m
, " N%d=%lu", nid
, node_nr
);
4125 #endif /* CONFIG_NUMA */
4127 static const char * const mem_cgroup_lru_names
[] = {
4135 static inline void mem_cgroup_lru_names_not_uptodate(void)
4137 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names
) != NR_LRU_LISTS
);
4140 static int memcg_stat_show(struct cgroup
*cont
, struct cftype
*cft
,
4143 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
4144 struct mem_cgroup
*mi
;
4147 for (i
= 0; i
< MEM_CGROUP_STAT_NSTATS
; i
++) {
4148 if (i
== MEM_CGROUP_STAT_SWAP
&& !do_swap_account
)
4150 seq_printf(m
, "%s %ld\n", mem_cgroup_stat_names
[i
],
4151 mem_cgroup_read_stat(memcg
, i
) * PAGE_SIZE
);
4154 for (i
= 0; i
< MEM_CGROUP_EVENTS_NSTATS
; i
++)
4155 seq_printf(m
, "%s %lu\n", mem_cgroup_events_names
[i
],
4156 mem_cgroup_read_events(memcg
, i
));
4158 for (i
= 0; i
< NR_LRU_LISTS
; i
++)
4159 seq_printf(m
, "%s %lu\n", mem_cgroup_lru_names
[i
],
4160 mem_cgroup_nr_lru_pages(memcg
, BIT(i
)) * PAGE_SIZE
);
4162 /* Hierarchical information */
4164 unsigned long long limit
, memsw_limit
;
4165 memcg_get_hierarchical_limit(memcg
, &limit
, &memsw_limit
);
4166 seq_printf(m
, "hierarchical_memory_limit %llu\n", limit
);
4167 if (do_swap_account
)
4168 seq_printf(m
, "hierarchical_memsw_limit %llu\n",
4172 for (i
= 0; i
< MEM_CGROUP_STAT_NSTATS
; i
++) {
4175 if (i
== MEM_CGROUP_STAT_SWAP
&& !do_swap_account
)
4177 for_each_mem_cgroup_tree(mi
, memcg
)
4178 val
+= mem_cgroup_read_stat(mi
, i
) * PAGE_SIZE
;
4179 seq_printf(m
, "total_%s %lld\n", mem_cgroup_stat_names
[i
], val
);
4182 for (i
= 0; i
< MEM_CGROUP_EVENTS_NSTATS
; i
++) {
4183 unsigned long long val
= 0;
4185 for_each_mem_cgroup_tree(mi
, memcg
)
4186 val
+= mem_cgroup_read_events(mi
, i
);
4187 seq_printf(m
, "total_%s %llu\n",
4188 mem_cgroup_events_names
[i
], val
);
4191 for (i
= 0; i
< NR_LRU_LISTS
; i
++) {
4192 unsigned long long val
= 0;
4194 for_each_mem_cgroup_tree(mi
, memcg
)
4195 val
+= mem_cgroup_nr_lru_pages(mi
, BIT(i
)) * PAGE_SIZE
;
4196 seq_printf(m
, "total_%s %llu\n", mem_cgroup_lru_names
[i
], val
);
4199 #ifdef CONFIG_DEBUG_VM
4202 struct mem_cgroup_per_zone
*mz
;
4203 struct zone_reclaim_stat
*rstat
;
4204 unsigned long recent_rotated
[2] = {0, 0};
4205 unsigned long recent_scanned
[2] = {0, 0};
4207 for_each_online_node(nid
)
4208 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++) {
4209 mz
= mem_cgroup_zoneinfo(memcg
, nid
, zid
);
4210 rstat
= &mz
->lruvec
.reclaim_stat
;
4212 recent_rotated
[0] += rstat
->recent_rotated
[0];
4213 recent_rotated
[1] += rstat
->recent_rotated
[1];
4214 recent_scanned
[0] += rstat
->recent_scanned
[0];
4215 recent_scanned
[1] += rstat
->recent_scanned
[1];
4217 seq_printf(m
, "recent_rotated_anon %lu\n", recent_rotated
[0]);
4218 seq_printf(m
, "recent_rotated_file %lu\n", recent_rotated
[1]);
4219 seq_printf(m
, "recent_scanned_anon %lu\n", recent_scanned
[0]);
4220 seq_printf(m
, "recent_scanned_file %lu\n", recent_scanned
[1]);
4227 static u64
mem_cgroup_swappiness_read(struct cgroup
*cgrp
, struct cftype
*cft
)
4229 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4231 return mem_cgroup_swappiness(memcg
);
4234 static int mem_cgroup_swappiness_write(struct cgroup
*cgrp
, struct cftype
*cft
,
4237 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4238 struct mem_cgroup
*parent
;
4243 if (cgrp
->parent
== NULL
)
4246 parent
= mem_cgroup_from_cont(cgrp
->parent
);
4250 /* If under hierarchy, only empty-root can set this value */
4251 if ((parent
->use_hierarchy
) ||
4252 (memcg
->use_hierarchy
&& !list_empty(&cgrp
->children
))) {
4257 memcg
->swappiness
= val
;
4264 static void __mem_cgroup_threshold(struct mem_cgroup
*memcg
, bool swap
)
4266 struct mem_cgroup_threshold_ary
*t
;
4272 t
= rcu_dereference(memcg
->thresholds
.primary
);
4274 t
= rcu_dereference(memcg
->memsw_thresholds
.primary
);
4279 usage
= mem_cgroup_usage(memcg
, swap
);
4282 * current_threshold points to threshold just below or equal to usage.
4283 * If it's not true, a threshold was crossed after last
4284 * call of __mem_cgroup_threshold().
4286 i
= t
->current_threshold
;
4289 * Iterate backward over array of thresholds starting from
4290 * current_threshold and check if a threshold is crossed.
4291 * If none of thresholds below usage is crossed, we read
4292 * only one element of the array here.
4294 for (; i
>= 0 && unlikely(t
->entries
[i
].threshold
> usage
); i
--)
4295 eventfd_signal(t
->entries
[i
].eventfd
, 1);
4297 /* i = current_threshold + 1 */
4301 * Iterate forward over array of thresholds starting from
4302 * current_threshold+1 and check if a threshold is crossed.
4303 * If none of thresholds above usage is crossed, we read
4304 * only one element of the array here.
4306 for (; i
< t
->size
&& unlikely(t
->entries
[i
].threshold
<= usage
); i
++)
4307 eventfd_signal(t
->entries
[i
].eventfd
, 1);
4309 /* Update current_threshold */
4310 t
->current_threshold
= i
- 1;
4315 static void mem_cgroup_threshold(struct mem_cgroup
*memcg
)
4318 __mem_cgroup_threshold(memcg
, false);
4319 if (do_swap_account
)
4320 __mem_cgroup_threshold(memcg
, true);
4322 memcg
= parent_mem_cgroup(memcg
);
4326 static int compare_thresholds(const void *a
, const void *b
)
4328 const struct mem_cgroup_threshold
*_a
= a
;
4329 const struct mem_cgroup_threshold
*_b
= b
;
4331 return _a
->threshold
- _b
->threshold
;
4334 static int mem_cgroup_oom_notify_cb(struct mem_cgroup
*memcg
)
4336 struct mem_cgroup_eventfd_list
*ev
;
4338 list_for_each_entry(ev
, &memcg
->oom_notify
, list
)
4339 eventfd_signal(ev
->eventfd
, 1);
4343 static void mem_cgroup_oom_notify(struct mem_cgroup
*memcg
)
4345 struct mem_cgroup
*iter
;
4347 for_each_mem_cgroup_tree(iter
, memcg
)
4348 mem_cgroup_oom_notify_cb(iter
);
4351 static int mem_cgroup_usage_register_event(struct cgroup
*cgrp
,
4352 struct cftype
*cft
, struct eventfd_ctx
*eventfd
, const char *args
)
4354 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4355 struct mem_cgroup_thresholds
*thresholds
;
4356 struct mem_cgroup_threshold_ary
*new;
4357 int type
= MEMFILE_TYPE(cft
->private);
4358 u64 threshold
, usage
;
4361 ret
= res_counter_memparse_write_strategy(args
, &threshold
);
4365 mutex_lock(&memcg
->thresholds_lock
);
4368 thresholds
= &memcg
->thresholds
;
4369 else if (type
== _MEMSWAP
)
4370 thresholds
= &memcg
->memsw_thresholds
;
4374 usage
= mem_cgroup_usage(memcg
, type
== _MEMSWAP
);
4376 /* Check if a threshold crossed before adding a new one */
4377 if (thresholds
->primary
)
4378 __mem_cgroup_threshold(memcg
, type
== _MEMSWAP
);
4380 size
= thresholds
->primary
? thresholds
->primary
->size
+ 1 : 1;
4382 /* Allocate memory for new array of thresholds */
4383 new = kmalloc(sizeof(*new) + size
* sizeof(struct mem_cgroup_threshold
),
4391 /* Copy thresholds (if any) to new array */
4392 if (thresholds
->primary
) {
4393 memcpy(new->entries
, thresholds
->primary
->entries
, (size
- 1) *
4394 sizeof(struct mem_cgroup_threshold
));
4397 /* Add new threshold */
4398 new->entries
[size
- 1].eventfd
= eventfd
;
4399 new->entries
[size
- 1].threshold
= threshold
;
4401 /* Sort thresholds. Registering of new threshold isn't time-critical */
4402 sort(new->entries
, size
, sizeof(struct mem_cgroup_threshold
),
4403 compare_thresholds
, NULL
);
4405 /* Find current threshold */
4406 new->current_threshold
= -1;
4407 for (i
= 0; i
< size
; i
++) {
4408 if (new->entries
[i
].threshold
<= usage
) {
4410 * new->current_threshold will not be used until
4411 * rcu_assign_pointer(), so it's safe to increment
4414 ++new->current_threshold
;
4419 /* Free old spare buffer and save old primary buffer as spare */
4420 kfree(thresholds
->spare
);
4421 thresholds
->spare
= thresholds
->primary
;
4423 rcu_assign_pointer(thresholds
->primary
, new);
4425 /* To be sure that nobody uses thresholds */
4429 mutex_unlock(&memcg
->thresholds_lock
);
4434 static void mem_cgroup_usage_unregister_event(struct cgroup
*cgrp
,
4435 struct cftype
*cft
, struct eventfd_ctx
*eventfd
)
4437 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4438 struct mem_cgroup_thresholds
*thresholds
;
4439 struct mem_cgroup_threshold_ary
*new;
4440 int type
= MEMFILE_TYPE(cft
->private);
4444 mutex_lock(&memcg
->thresholds_lock
);
4446 thresholds
= &memcg
->thresholds
;
4447 else if (type
== _MEMSWAP
)
4448 thresholds
= &memcg
->memsw_thresholds
;
4452 if (!thresholds
->primary
)
4455 usage
= mem_cgroup_usage(memcg
, type
== _MEMSWAP
);
4457 /* Check if a threshold crossed before removing */
4458 __mem_cgroup_threshold(memcg
, type
== _MEMSWAP
);
4460 /* Calculate new number of threshold */
4462 for (i
= 0; i
< thresholds
->primary
->size
; i
++) {
4463 if (thresholds
->primary
->entries
[i
].eventfd
!= eventfd
)
4467 new = thresholds
->spare
;
4469 /* Set thresholds array to NULL if we don't have thresholds */
4478 /* Copy thresholds and find current threshold */
4479 new->current_threshold
= -1;
4480 for (i
= 0, j
= 0; i
< thresholds
->primary
->size
; i
++) {
4481 if (thresholds
->primary
->entries
[i
].eventfd
== eventfd
)
4484 new->entries
[j
] = thresholds
->primary
->entries
[i
];
4485 if (new->entries
[j
].threshold
<= usage
) {
4487 * new->current_threshold will not be used
4488 * until rcu_assign_pointer(), so it's safe to increment
4491 ++new->current_threshold
;
4497 /* Swap primary and spare array */
4498 thresholds
->spare
= thresholds
->primary
;
4499 /* If all events are unregistered, free the spare array */
4501 kfree(thresholds
->spare
);
4502 thresholds
->spare
= NULL
;
4505 rcu_assign_pointer(thresholds
->primary
, new);
4507 /* To be sure that nobody uses thresholds */
4510 mutex_unlock(&memcg
->thresholds_lock
);
4513 static int mem_cgroup_oom_register_event(struct cgroup
*cgrp
,
4514 struct cftype
*cft
, struct eventfd_ctx
*eventfd
, const char *args
)
4516 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4517 struct mem_cgroup_eventfd_list
*event
;
4518 int type
= MEMFILE_TYPE(cft
->private);
4520 BUG_ON(type
!= _OOM_TYPE
);
4521 event
= kmalloc(sizeof(*event
), GFP_KERNEL
);
4525 spin_lock(&memcg_oom_lock
);
4527 event
->eventfd
= eventfd
;
4528 list_add(&event
->list
, &memcg
->oom_notify
);
4530 /* already in OOM ? */
4531 if (atomic_read(&memcg
->under_oom
))
4532 eventfd_signal(eventfd
, 1);
4533 spin_unlock(&memcg_oom_lock
);
4538 static void mem_cgroup_oom_unregister_event(struct cgroup
*cgrp
,
4539 struct cftype
*cft
, struct eventfd_ctx
*eventfd
)
4541 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4542 struct mem_cgroup_eventfd_list
*ev
, *tmp
;
4543 int type
= MEMFILE_TYPE(cft
->private);
4545 BUG_ON(type
!= _OOM_TYPE
);
4547 spin_lock(&memcg_oom_lock
);
4549 list_for_each_entry_safe(ev
, tmp
, &memcg
->oom_notify
, list
) {
4550 if (ev
->eventfd
== eventfd
) {
4551 list_del(&ev
->list
);
4556 spin_unlock(&memcg_oom_lock
);
4559 static int mem_cgroup_oom_control_read(struct cgroup
*cgrp
,
4560 struct cftype
*cft
, struct cgroup_map_cb
*cb
)
4562 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4564 cb
->fill(cb
, "oom_kill_disable", memcg
->oom_kill_disable
);
4566 if (atomic_read(&memcg
->under_oom
))
4567 cb
->fill(cb
, "under_oom", 1);
4569 cb
->fill(cb
, "under_oom", 0);
4573 static int mem_cgroup_oom_control_write(struct cgroup
*cgrp
,
4574 struct cftype
*cft
, u64 val
)
4576 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4577 struct mem_cgroup
*parent
;
4579 /* cannot set to root cgroup and only 0 and 1 are allowed */
4580 if (!cgrp
->parent
|| !((val
== 0) || (val
== 1)))
4583 parent
= mem_cgroup_from_cont(cgrp
->parent
);
4586 /* oom-kill-disable is a flag for subhierarchy. */
4587 if ((parent
->use_hierarchy
) ||
4588 (memcg
->use_hierarchy
&& !list_empty(&cgrp
->children
))) {
4592 memcg
->oom_kill_disable
= val
;
4594 memcg_oom_recover(memcg
);
4599 #ifdef CONFIG_MEMCG_KMEM
4600 static int memcg_init_kmem(struct mem_cgroup
*memcg
, struct cgroup_subsys
*ss
)
4602 return mem_cgroup_sockets_init(memcg
, ss
);
4605 static void kmem_cgroup_destroy(struct mem_cgroup
*memcg
)
4607 mem_cgroup_sockets_destroy(memcg
);
4610 static int memcg_init_kmem(struct mem_cgroup
*memcg
, struct cgroup_subsys
*ss
)
4615 static void kmem_cgroup_destroy(struct mem_cgroup
*memcg
)
4620 static struct cftype mem_cgroup_files
[] = {
4622 .name
= "usage_in_bytes",
4623 .private = MEMFILE_PRIVATE(_MEM
, RES_USAGE
),
4624 .read
= mem_cgroup_read
,
4625 .register_event
= mem_cgroup_usage_register_event
,
4626 .unregister_event
= mem_cgroup_usage_unregister_event
,
4629 .name
= "max_usage_in_bytes",
4630 .private = MEMFILE_PRIVATE(_MEM
, RES_MAX_USAGE
),
4631 .trigger
= mem_cgroup_reset
,
4632 .read
= mem_cgroup_read
,
4635 .name
= "limit_in_bytes",
4636 .private = MEMFILE_PRIVATE(_MEM
, RES_LIMIT
),
4637 .write_string
= mem_cgroup_write
,
4638 .read
= mem_cgroup_read
,
4641 .name
= "soft_limit_in_bytes",
4642 .private = MEMFILE_PRIVATE(_MEM
, RES_SOFT_LIMIT
),
4643 .write_string
= mem_cgroup_write
,
4644 .read
= mem_cgroup_read
,
4648 .private = MEMFILE_PRIVATE(_MEM
, RES_FAILCNT
),
4649 .trigger
= mem_cgroup_reset
,
4650 .read
= mem_cgroup_read
,
4654 .read_seq_string
= memcg_stat_show
,
4657 .name
= "force_empty",
4658 .trigger
= mem_cgroup_force_empty_write
,
4661 .name
= "use_hierarchy",
4662 .write_u64
= mem_cgroup_hierarchy_write
,
4663 .read_u64
= mem_cgroup_hierarchy_read
,
4666 .name
= "swappiness",
4667 .read_u64
= mem_cgroup_swappiness_read
,
4668 .write_u64
= mem_cgroup_swappiness_write
,
4671 .name
= "move_charge_at_immigrate",
4672 .read_u64
= mem_cgroup_move_charge_read
,
4673 .write_u64
= mem_cgroup_move_charge_write
,
4676 .name
= "oom_control",
4677 .read_map
= mem_cgroup_oom_control_read
,
4678 .write_u64
= mem_cgroup_oom_control_write
,
4679 .register_event
= mem_cgroup_oom_register_event
,
4680 .unregister_event
= mem_cgroup_oom_unregister_event
,
4681 .private = MEMFILE_PRIVATE(_OOM_TYPE
, OOM_CONTROL
),
4685 .name
= "numa_stat",
4686 .read_seq_string
= memcg_numa_stat_show
,
4689 #ifdef CONFIG_MEMCG_SWAP
4691 .name
= "memsw.usage_in_bytes",
4692 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_USAGE
),
4693 .read
= mem_cgroup_read
,
4694 .register_event
= mem_cgroup_usage_register_event
,
4695 .unregister_event
= mem_cgroup_usage_unregister_event
,
4698 .name
= "memsw.max_usage_in_bytes",
4699 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_MAX_USAGE
),
4700 .trigger
= mem_cgroup_reset
,
4701 .read
= mem_cgroup_read
,
4704 .name
= "memsw.limit_in_bytes",
4705 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_LIMIT
),
4706 .write_string
= mem_cgroup_write
,
4707 .read
= mem_cgroup_read
,
4710 .name
= "memsw.failcnt",
4711 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_FAILCNT
),
4712 .trigger
= mem_cgroup_reset
,
4713 .read
= mem_cgroup_read
,
4716 { }, /* terminate */
4719 static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup
*memcg
, int node
)
4721 struct mem_cgroup_per_node
*pn
;
4722 struct mem_cgroup_per_zone
*mz
;
4723 int zone
, tmp
= node
;
4725 * This routine is called against possible nodes.
4726 * But it's BUG to call kmalloc() against offline node.
4728 * TODO: this routine can waste much memory for nodes which will
4729 * never be onlined. It's better to use memory hotplug callback
4732 if (!node_state(node
, N_NORMAL_MEMORY
))
4734 pn
= kzalloc_node(sizeof(*pn
), GFP_KERNEL
, tmp
);
4738 for (zone
= 0; zone
< MAX_NR_ZONES
; zone
++) {
4739 mz
= &pn
->zoneinfo
[zone
];
4740 lruvec_init(&mz
->lruvec
, &NODE_DATA(node
)->node_zones
[zone
]);
4741 mz
->usage_in_excess
= 0;
4742 mz
->on_tree
= false;
4745 memcg
->info
.nodeinfo
[node
] = pn
;
4749 static void free_mem_cgroup_per_zone_info(struct mem_cgroup
*memcg
, int node
)
4751 kfree(memcg
->info
.nodeinfo
[node
]);
4754 static struct mem_cgroup
*mem_cgroup_alloc(void)
4756 struct mem_cgroup
*memcg
;
4757 int size
= sizeof(struct mem_cgroup
);
4759 /* Can be very big if MAX_NUMNODES is very big */
4760 if (size
< PAGE_SIZE
)
4761 memcg
= kzalloc(size
, GFP_KERNEL
);
4763 memcg
= vzalloc(size
);
4768 memcg
->stat
= alloc_percpu(struct mem_cgroup_stat_cpu
);
4771 spin_lock_init(&memcg
->pcp_counter_lock
);
4775 if (size
< PAGE_SIZE
)
4783 * Helpers for freeing a kmalloc()ed/vzalloc()ed mem_cgroup by RCU,
4784 * but in process context. The work_freeing structure is overlaid
4785 * on the rcu_freeing structure, which itself is overlaid on memsw.
4787 static void free_work(struct work_struct
*work
)
4789 struct mem_cgroup
*memcg
;
4790 int size
= sizeof(struct mem_cgroup
);
4792 memcg
= container_of(work
, struct mem_cgroup
, work_freeing
);
4794 * We need to make sure that (at least for now), the jump label
4795 * destruction code runs outside of the cgroup lock. This is because
4796 * get_online_cpus(), which is called from the static_branch update,
4797 * can't be called inside the cgroup_lock. cpusets are the ones
4798 * enforcing this dependency, so if they ever change, we might as well.
4800 * schedule_work() will guarantee this happens. Be careful if you need
4801 * to move this code around, and make sure it is outside
4804 disarm_sock_keys(memcg
);
4805 if (size
< PAGE_SIZE
)
4811 static void free_rcu(struct rcu_head
*rcu_head
)
4813 struct mem_cgroup
*memcg
;
4815 memcg
= container_of(rcu_head
, struct mem_cgroup
, rcu_freeing
);
4816 INIT_WORK(&memcg
->work_freeing
, free_work
);
4817 schedule_work(&memcg
->work_freeing
);
4821 * At destroying mem_cgroup, references from swap_cgroup can remain.
4822 * (scanning all at force_empty is too costly...)
4824 * Instead of clearing all references at force_empty, we remember
4825 * the number of reference from swap_cgroup and free mem_cgroup when
4826 * it goes down to 0.
4828 * Removal of cgroup itself succeeds regardless of refs from swap.
4831 static void __mem_cgroup_free(struct mem_cgroup
*memcg
)
4835 mem_cgroup_remove_from_trees(memcg
);
4836 free_css_id(&mem_cgroup_subsys
, &memcg
->css
);
4839 free_mem_cgroup_per_zone_info(memcg
, node
);
4841 free_percpu(memcg
->stat
);
4842 call_rcu(&memcg
->rcu_freeing
, free_rcu
);
4845 static void mem_cgroup_get(struct mem_cgroup
*memcg
)
4847 atomic_inc(&memcg
->refcnt
);
4850 static void __mem_cgroup_put(struct mem_cgroup
*memcg
, int count
)
4852 if (atomic_sub_and_test(count
, &memcg
->refcnt
)) {
4853 struct mem_cgroup
*parent
= parent_mem_cgroup(memcg
);
4854 __mem_cgroup_free(memcg
);
4856 mem_cgroup_put(parent
);
4860 static void mem_cgroup_put(struct mem_cgroup
*memcg
)
4862 __mem_cgroup_put(memcg
, 1);
4866 * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled.
4868 struct mem_cgroup
*parent_mem_cgroup(struct mem_cgroup
*memcg
)
4870 if (!memcg
->res
.parent
)
4872 return mem_cgroup_from_res_counter(memcg
->res
.parent
, res
);
4874 EXPORT_SYMBOL(parent_mem_cgroup
);
4876 #ifdef CONFIG_MEMCG_SWAP
4877 static void __init
enable_swap_cgroup(void)
4879 if (!mem_cgroup_disabled() && really_do_swap_account
)
4880 do_swap_account
= 1;
4883 static void __init
enable_swap_cgroup(void)
4888 static int mem_cgroup_soft_limit_tree_init(void)
4890 struct mem_cgroup_tree_per_node
*rtpn
;
4891 struct mem_cgroup_tree_per_zone
*rtpz
;
4892 int tmp
, node
, zone
;
4894 for_each_node(node
) {
4896 if (!node_state(node
, N_NORMAL_MEMORY
))
4898 rtpn
= kzalloc_node(sizeof(*rtpn
), GFP_KERNEL
, tmp
);
4902 soft_limit_tree
.rb_tree_per_node
[node
] = rtpn
;
4904 for (zone
= 0; zone
< MAX_NR_ZONES
; zone
++) {
4905 rtpz
= &rtpn
->rb_tree_per_zone
[zone
];
4906 rtpz
->rb_root
= RB_ROOT
;
4907 spin_lock_init(&rtpz
->lock
);
4913 for_each_node(node
) {
4914 if (!soft_limit_tree
.rb_tree_per_node
[node
])
4916 kfree(soft_limit_tree
.rb_tree_per_node
[node
]);
4917 soft_limit_tree
.rb_tree_per_node
[node
] = NULL
;
4923 static struct cgroup_subsys_state
* __ref
4924 mem_cgroup_create(struct cgroup
*cont
)
4926 struct mem_cgroup
*memcg
, *parent
;
4927 long error
= -ENOMEM
;
4930 memcg
= mem_cgroup_alloc();
4932 return ERR_PTR(error
);
4935 if (alloc_mem_cgroup_per_zone_info(memcg
, node
))
4939 if (cont
->parent
== NULL
) {
4941 enable_swap_cgroup();
4943 if (mem_cgroup_soft_limit_tree_init())
4945 root_mem_cgroup
= memcg
;
4946 for_each_possible_cpu(cpu
) {
4947 struct memcg_stock_pcp
*stock
=
4948 &per_cpu(memcg_stock
, cpu
);
4949 INIT_WORK(&stock
->work
, drain_local_stock
);
4951 hotcpu_notifier(memcg_cpu_hotplug_callback
, 0);
4953 parent
= mem_cgroup_from_cont(cont
->parent
);
4954 memcg
->use_hierarchy
= parent
->use_hierarchy
;
4955 memcg
->oom_kill_disable
= parent
->oom_kill_disable
;
4958 if (parent
&& parent
->use_hierarchy
) {
4959 res_counter_init(&memcg
->res
, &parent
->res
);
4960 res_counter_init(&memcg
->memsw
, &parent
->memsw
);
4962 * We increment refcnt of the parent to ensure that we can
4963 * safely access it on res_counter_charge/uncharge.
4964 * This refcnt will be decremented when freeing this
4965 * mem_cgroup(see mem_cgroup_put).
4967 mem_cgroup_get(parent
);
4969 res_counter_init(&memcg
->res
, NULL
);
4970 res_counter_init(&memcg
->memsw
, NULL
);
4972 * Deeper hierachy with use_hierarchy == false doesn't make
4973 * much sense so let cgroup subsystem know about this
4974 * unfortunate state in our controller.
4976 if (parent
&& parent
!= root_mem_cgroup
)
4977 mem_cgroup_subsys
.broken_hierarchy
= true;
4979 memcg
->last_scanned_node
= MAX_NUMNODES
;
4980 INIT_LIST_HEAD(&memcg
->oom_notify
);
4983 memcg
->swappiness
= mem_cgroup_swappiness(parent
);
4984 atomic_set(&memcg
->refcnt
, 1);
4985 memcg
->move_charge_at_immigrate
= 0;
4986 mutex_init(&memcg
->thresholds_lock
);
4987 spin_lock_init(&memcg
->move_lock
);
4989 error
= memcg_init_kmem(memcg
, &mem_cgroup_subsys
);
4992 * We call put now because our (and parent's) refcnts
4993 * are already in place. mem_cgroup_put() will internally
4994 * call __mem_cgroup_free, so return directly
4996 mem_cgroup_put(memcg
);
4997 return ERR_PTR(error
);
5001 __mem_cgroup_free(memcg
);
5002 return ERR_PTR(error
);
5005 static int mem_cgroup_pre_destroy(struct cgroup
*cont
)
5007 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
5009 return mem_cgroup_force_empty(memcg
, false);
5012 static void mem_cgroup_destroy(struct cgroup
*cont
)
5014 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
5016 kmem_cgroup_destroy(memcg
);
5018 mem_cgroup_put(memcg
);
5022 /* Handlers for move charge at task migration. */
5023 #define PRECHARGE_COUNT_AT_ONCE 256
5024 static int mem_cgroup_do_precharge(unsigned long count
)
5027 int batch_count
= PRECHARGE_COUNT_AT_ONCE
;
5028 struct mem_cgroup
*memcg
= mc
.to
;
5030 if (mem_cgroup_is_root(memcg
)) {
5031 mc
.precharge
+= count
;
5032 /* we don't need css_get for root */
5035 /* try to charge at once */
5037 struct res_counter
*dummy
;
5039 * "memcg" cannot be under rmdir() because we've already checked
5040 * by cgroup_lock_live_cgroup() that it is not removed and we
5041 * are still under the same cgroup_mutex. So we can postpone
5044 if (res_counter_charge(&memcg
->res
, PAGE_SIZE
* count
, &dummy
))
5046 if (do_swap_account
&& res_counter_charge(&memcg
->memsw
,
5047 PAGE_SIZE
* count
, &dummy
)) {
5048 res_counter_uncharge(&memcg
->res
, PAGE_SIZE
* count
);
5051 mc
.precharge
+= count
;
5055 /* fall back to one by one charge */
5057 if (signal_pending(current
)) {
5061 if (!batch_count
--) {
5062 batch_count
= PRECHARGE_COUNT_AT_ONCE
;
5065 ret
= __mem_cgroup_try_charge(NULL
,
5066 GFP_KERNEL
, 1, &memcg
, false);
5068 /* mem_cgroup_clear_mc() will do uncharge later */
5076 * get_mctgt_type - get target type of moving charge
5077 * @vma: the vma the pte to be checked belongs
5078 * @addr: the address corresponding to the pte to be checked
5079 * @ptent: the pte to be checked
5080 * @target: the pointer the target page or swap ent will be stored(can be NULL)
5083 * 0(MC_TARGET_NONE): if the pte is not a target for move charge.
5084 * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
5085 * move charge. if @target is not NULL, the page is stored in target->page
5086 * with extra refcnt got(Callers should handle it).
5087 * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
5088 * target for charge migration. if @target is not NULL, the entry is stored
5091 * Called with pte lock held.
5098 enum mc_target_type
{
5104 static struct page
*mc_handle_present_pte(struct vm_area_struct
*vma
,
5105 unsigned long addr
, pte_t ptent
)
5107 struct page
*page
= vm_normal_page(vma
, addr
, ptent
);
5109 if (!page
|| !page_mapped(page
))
5111 if (PageAnon(page
)) {
5112 /* we don't move shared anon */
5115 } else if (!move_file())
5116 /* we ignore mapcount for file pages */
5118 if (!get_page_unless_zero(page
))
5125 static struct page
*mc_handle_swap_pte(struct vm_area_struct
*vma
,
5126 unsigned long addr
, pte_t ptent
, swp_entry_t
*entry
)
5128 struct page
*page
= NULL
;
5129 swp_entry_t ent
= pte_to_swp_entry(ptent
);
5131 if (!move_anon() || non_swap_entry(ent
))
5134 * Because lookup_swap_cache() updates some statistics counter,
5135 * we call find_get_page() with swapper_space directly.
5137 page
= find_get_page(&swapper_space
, ent
.val
);
5138 if (do_swap_account
)
5139 entry
->val
= ent
.val
;
5144 static struct page
*mc_handle_swap_pte(struct vm_area_struct
*vma
,
5145 unsigned long addr
, pte_t ptent
, swp_entry_t
*entry
)
5151 static struct page
*mc_handle_file_pte(struct vm_area_struct
*vma
,
5152 unsigned long addr
, pte_t ptent
, swp_entry_t
*entry
)
5154 struct page
*page
= NULL
;
5155 struct address_space
*mapping
;
5158 if (!vma
->vm_file
) /* anonymous vma */
5163 mapping
= vma
->vm_file
->f_mapping
;
5164 if (pte_none(ptent
))
5165 pgoff
= linear_page_index(vma
, addr
);
5166 else /* pte_file(ptent) is true */
5167 pgoff
= pte_to_pgoff(ptent
);
5169 /* page is moved even if it's not RSS of this task(page-faulted). */
5170 page
= find_get_page(mapping
, pgoff
);
5173 /* shmem/tmpfs may report page out on swap: account for that too. */
5174 if (radix_tree_exceptional_entry(page
)) {
5175 swp_entry_t swap
= radix_to_swp_entry(page
);
5176 if (do_swap_account
)
5178 page
= find_get_page(&swapper_space
, swap
.val
);
5184 static enum mc_target_type
get_mctgt_type(struct vm_area_struct
*vma
,
5185 unsigned long addr
, pte_t ptent
, union mc_target
*target
)
5187 struct page
*page
= NULL
;
5188 struct page_cgroup
*pc
;
5189 enum mc_target_type ret
= MC_TARGET_NONE
;
5190 swp_entry_t ent
= { .val
= 0 };
5192 if (pte_present(ptent
))
5193 page
= mc_handle_present_pte(vma
, addr
, ptent
);
5194 else if (is_swap_pte(ptent
))
5195 page
= mc_handle_swap_pte(vma
, addr
, ptent
, &ent
);
5196 else if (pte_none(ptent
) || pte_file(ptent
))
5197 page
= mc_handle_file_pte(vma
, addr
, ptent
, &ent
);
5199 if (!page
&& !ent
.val
)
5202 pc
= lookup_page_cgroup(page
);
5204 * Do only loose check w/o page_cgroup lock.
5205 * mem_cgroup_move_account() checks the pc is valid or not under
5208 if (PageCgroupUsed(pc
) && pc
->mem_cgroup
== mc
.from
) {
5209 ret
= MC_TARGET_PAGE
;
5211 target
->page
= page
;
5213 if (!ret
|| !target
)
5216 /* There is a swap entry and a page doesn't exist or isn't charged */
5217 if (ent
.val
&& !ret
&&
5218 css_id(&mc
.from
->css
) == lookup_swap_cgroup_id(ent
)) {
5219 ret
= MC_TARGET_SWAP
;
5226 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
5228 * We don't consider swapping or file mapped pages because THP does not
5229 * support them for now.
5230 * Caller should make sure that pmd_trans_huge(pmd) is true.
5232 static enum mc_target_type
get_mctgt_type_thp(struct vm_area_struct
*vma
,
5233 unsigned long addr
, pmd_t pmd
, union mc_target
*target
)
5235 struct page
*page
= NULL
;
5236 struct page_cgroup
*pc
;
5237 enum mc_target_type ret
= MC_TARGET_NONE
;
5239 page
= pmd_page(pmd
);
5240 VM_BUG_ON(!page
|| !PageHead(page
));
5243 pc
= lookup_page_cgroup(page
);
5244 if (PageCgroupUsed(pc
) && pc
->mem_cgroup
== mc
.from
) {
5245 ret
= MC_TARGET_PAGE
;
5248 target
->page
= page
;
5254 static inline enum mc_target_type
get_mctgt_type_thp(struct vm_area_struct
*vma
,
5255 unsigned long addr
, pmd_t pmd
, union mc_target
*target
)
5257 return MC_TARGET_NONE
;
5261 static int mem_cgroup_count_precharge_pte_range(pmd_t
*pmd
,
5262 unsigned long addr
, unsigned long end
,
5263 struct mm_walk
*walk
)
5265 struct vm_area_struct
*vma
= walk
->private;
5269 if (pmd_trans_huge_lock(pmd
, vma
) == 1) {
5270 if (get_mctgt_type_thp(vma
, addr
, *pmd
, NULL
) == MC_TARGET_PAGE
)
5271 mc
.precharge
+= HPAGE_PMD_NR
;
5272 spin_unlock(&vma
->vm_mm
->page_table_lock
);
5276 if (pmd_trans_unstable(pmd
))
5278 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
5279 for (; addr
!= end
; pte
++, addr
+= PAGE_SIZE
)
5280 if (get_mctgt_type(vma
, addr
, *pte
, NULL
))
5281 mc
.precharge
++; /* increment precharge temporarily */
5282 pte_unmap_unlock(pte
- 1, ptl
);
5288 static unsigned long mem_cgroup_count_precharge(struct mm_struct
*mm
)
5290 unsigned long precharge
;
5291 struct vm_area_struct
*vma
;
5293 down_read(&mm
->mmap_sem
);
5294 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
5295 struct mm_walk mem_cgroup_count_precharge_walk
= {
5296 .pmd_entry
= mem_cgroup_count_precharge_pte_range
,
5300 if (is_vm_hugetlb_page(vma
))
5302 walk_page_range(vma
->vm_start
, vma
->vm_end
,
5303 &mem_cgroup_count_precharge_walk
);
5305 up_read(&mm
->mmap_sem
);
5307 precharge
= mc
.precharge
;
5313 static int mem_cgroup_precharge_mc(struct mm_struct
*mm
)
5315 unsigned long precharge
= mem_cgroup_count_precharge(mm
);
5317 VM_BUG_ON(mc
.moving_task
);
5318 mc
.moving_task
= current
;
5319 return mem_cgroup_do_precharge(precharge
);
5322 /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */
5323 static void __mem_cgroup_clear_mc(void)
5325 struct mem_cgroup
*from
= mc
.from
;
5326 struct mem_cgroup
*to
= mc
.to
;
5328 /* we must uncharge all the leftover precharges from mc.to */
5330 __mem_cgroup_cancel_charge(mc
.to
, mc
.precharge
);
5334 * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
5335 * we must uncharge here.
5337 if (mc
.moved_charge
) {
5338 __mem_cgroup_cancel_charge(mc
.from
, mc
.moved_charge
);
5339 mc
.moved_charge
= 0;
5341 /* we must fixup refcnts and charges */
5342 if (mc
.moved_swap
) {
5343 /* uncharge swap account from the old cgroup */
5344 if (!mem_cgroup_is_root(mc
.from
))
5345 res_counter_uncharge(&mc
.from
->memsw
,
5346 PAGE_SIZE
* mc
.moved_swap
);
5347 __mem_cgroup_put(mc
.from
, mc
.moved_swap
);
5349 if (!mem_cgroup_is_root(mc
.to
)) {
5351 * we charged both to->res and to->memsw, so we should
5354 res_counter_uncharge(&mc
.to
->res
,
5355 PAGE_SIZE
* mc
.moved_swap
);
5357 /* we've already done mem_cgroup_get(mc.to) */
5360 memcg_oom_recover(from
);
5361 memcg_oom_recover(to
);
5362 wake_up_all(&mc
.waitq
);
5365 static void mem_cgroup_clear_mc(void)
5367 struct mem_cgroup
*from
= mc
.from
;
5370 * we must clear moving_task before waking up waiters at the end of
5373 mc
.moving_task
= NULL
;
5374 __mem_cgroup_clear_mc();
5375 spin_lock(&mc
.lock
);
5378 spin_unlock(&mc
.lock
);
5379 mem_cgroup_end_move(from
);
5382 static int mem_cgroup_can_attach(struct cgroup
*cgroup
,
5383 struct cgroup_taskset
*tset
)
5385 struct task_struct
*p
= cgroup_taskset_first(tset
);
5387 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgroup
);
5389 if (memcg
->move_charge_at_immigrate
) {
5390 struct mm_struct
*mm
;
5391 struct mem_cgroup
*from
= mem_cgroup_from_task(p
);
5393 VM_BUG_ON(from
== memcg
);
5395 mm
= get_task_mm(p
);
5398 /* We move charges only when we move a owner of the mm */
5399 if (mm
->owner
== p
) {
5402 VM_BUG_ON(mc
.precharge
);
5403 VM_BUG_ON(mc
.moved_charge
);
5404 VM_BUG_ON(mc
.moved_swap
);
5405 mem_cgroup_start_move(from
);
5406 spin_lock(&mc
.lock
);
5409 spin_unlock(&mc
.lock
);
5410 /* We set mc.moving_task later */
5412 ret
= mem_cgroup_precharge_mc(mm
);
5414 mem_cgroup_clear_mc();
5421 static void mem_cgroup_cancel_attach(struct cgroup
*cgroup
,
5422 struct cgroup_taskset
*tset
)
5424 mem_cgroup_clear_mc();
5427 static int mem_cgroup_move_charge_pte_range(pmd_t
*pmd
,
5428 unsigned long addr
, unsigned long end
,
5429 struct mm_walk
*walk
)
5432 struct vm_area_struct
*vma
= walk
->private;
5435 enum mc_target_type target_type
;
5436 union mc_target target
;
5438 struct page_cgroup
*pc
;
5441 * We don't take compound_lock() here but no race with splitting thp
5443 * - if pmd_trans_huge_lock() returns 1, the relevant thp is not
5444 * under splitting, which means there's no concurrent thp split,
5445 * - if another thread runs into split_huge_page() just after we
5446 * entered this if-block, the thread must wait for page table lock
5447 * to be unlocked in __split_huge_page_splitting(), where the main
5448 * part of thp split is not executed yet.
5450 if (pmd_trans_huge_lock(pmd
, vma
) == 1) {
5451 if (mc
.precharge
< HPAGE_PMD_NR
) {
5452 spin_unlock(&vma
->vm_mm
->page_table_lock
);
5455 target_type
= get_mctgt_type_thp(vma
, addr
, *pmd
, &target
);
5456 if (target_type
== MC_TARGET_PAGE
) {
5458 if (!isolate_lru_page(page
)) {
5459 pc
= lookup_page_cgroup(page
);
5460 if (!mem_cgroup_move_account(page
, HPAGE_PMD_NR
,
5461 pc
, mc
.from
, mc
.to
)) {
5462 mc
.precharge
-= HPAGE_PMD_NR
;
5463 mc
.moved_charge
+= HPAGE_PMD_NR
;
5465 putback_lru_page(page
);
5469 spin_unlock(&vma
->vm_mm
->page_table_lock
);
5473 if (pmd_trans_unstable(pmd
))
5476 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
5477 for (; addr
!= end
; addr
+= PAGE_SIZE
) {
5478 pte_t ptent
= *(pte
++);
5484 switch (get_mctgt_type(vma
, addr
, ptent
, &target
)) {
5485 case MC_TARGET_PAGE
:
5487 if (isolate_lru_page(page
))
5489 pc
= lookup_page_cgroup(page
);
5490 if (!mem_cgroup_move_account(page
, 1, pc
,
5493 /* we uncharge from mc.from later. */
5496 putback_lru_page(page
);
5497 put
: /* get_mctgt_type() gets the page */
5500 case MC_TARGET_SWAP
:
5502 if (!mem_cgroup_move_swap_account(ent
, mc
.from
, mc
.to
)) {
5504 /* we fixup refcnts and charges later. */
5512 pte_unmap_unlock(pte
- 1, ptl
);
5517 * We have consumed all precharges we got in can_attach().
5518 * We try charge one by one, but don't do any additional
5519 * charges to mc.to if we have failed in charge once in attach()
5522 ret
= mem_cgroup_do_precharge(1);
5530 static void mem_cgroup_move_charge(struct mm_struct
*mm
)
5532 struct vm_area_struct
*vma
;
5534 lru_add_drain_all();
5536 if (unlikely(!down_read_trylock(&mm
->mmap_sem
))) {
5538 * Someone who are holding the mmap_sem might be waiting in
5539 * waitq. So we cancel all extra charges, wake up all waiters,
5540 * and retry. Because we cancel precharges, we might not be able
5541 * to move enough charges, but moving charge is a best-effort
5542 * feature anyway, so it wouldn't be a big problem.
5544 __mem_cgroup_clear_mc();
5548 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
5550 struct mm_walk mem_cgroup_move_charge_walk
= {
5551 .pmd_entry
= mem_cgroup_move_charge_pte_range
,
5555 if (is_vm_hugetlb_page(vma
))
5557 ret
= walk_page_range(vma
->vm_start
, vma
->vm_end
,
5558 &mem_cgroup_move_charge_walk
);
5561 * means we have consumed all precharges and failed in
5562 * doing additional charge. Just abandon here.
5566 up_read(&mm
->mmap_sem
);
5569 static void mem_cgroup_move_task(struct cgroup
*cont
,
5570 struct cgroup_taskset
*tset
)
5572 struct task_struct
*p
= cgroup_taskset_first(tset
);
5573 struct mm_struct
*mm
= get_task_mm(p
);
5577 mem_cgroup_move_charge(mm
);
5581 mem_cgroup_clear_mc();
5583 #else /* !CONFIG_MMU */
5584 static int mem_cgroup_can_attach(struct cgroup
*cgroup
,
5585 struct cgroup_taskset
*tset
)
5589 static void mem_cgroup_cancel_attach(struct cgroup
*cgroup
,
5590 struct cgroup_taskset
*tset
)
5593 static void mem_cgroup_move_task(struct cgroup
*cont
,
5594 struct cgroup_taskset
*tset
)
5599 struct cgroup_subsys mem_cgroup_subsys
= {
5601 .subsys_id
= mem_cgroup_subsys_id
,
5602 .create
= mem_cgroup_create
,
5603 .pre_destroy
= mem_cgroup_pre_destroy
,
5604 .destroy
= mem_cgroup_destroy
,
5605 .can_attach
= mem_cgroup_can_attach
,
5606 .cancel_attach
= mem_cgroup_cancel_attach
,
5607 .attach
= mem_cgroup_move_task
,
5608 .base_cftypes
= mem_cgroup_files
,
5611 .__DEPRECATED_clear_css_refs
= true,
5614 #ifdef CONFIG_MEMCG_SWAP
5615 static int __init
enable_swap_account(char *s
)
5617 /* consider enabled if no parameter or 1 is given */
5618 if (!strcmp(s
, "1"))
5619 really_do_swap_account
= 1;
5620 else if (!strcmp(s
, "0"))
5621 really_do_swap_account
= 0;
5624 __setup("swapaccount=", enable_swap_account
);