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>
54 #include <net/tcp_memcontrol.h>
56 #include <asm/uaccess.h>
58 #include <trace/events/vmscan.h>
60 struct cgroup_subsys mem_cgroup_subsys __read_mostly
;
61 #define MEM_CGROUP_RECLAIM_RETRIES 5
62 static struct mem_cgroup
*root_mem_cgroup __read_mostly
;
64 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
65 /* Turned on only when memory cgroup is enabled && really_do_swap_account = 1 */
66 int do_swap_account __read_mostly
;
68 /* for remember boot option*/
69 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP_ENABLED
70 static int really_do_swap_account __initdata
= 1;
72 static int really_do_swap_account __initdata
= 0;
76 #define do_swap_account 0
81 * Statistics for memory cgroup.
83 enum mem_cgroup_stat_index
{
85 * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss.
87 MEM_CGROUP_STAT_CACHE
, /* # of pages charged as cache */
88 MEM_CGROUP_STAT_RSS
, /* # of pages charged as anon rss */
89 MEM_CGROUP_STAT_FILE_MAPPED
, /* # of pages charged as file rss */
90 MEM_CGROUP_STAT_SWAPOUT
, /* # of pages, swapped out */
91 MEM_CGROUP_STAT_NSTATS
,
94 enum mem_cgroup_events_index
{
95 MEM_CGROUP_EVENTS_PGPGIN
, /* # of pages paged in */
96 MEM_CGROUP_EVENTS_PGPGOUT
, /* # of pages paged out */
97 MEM_CGROUP_EVENTS_PGFAULT
, /* # of page-faults */
98 MEM_CGROUP_EVENTS_PGMAJFAULT
, /* # of major page-faults */
99 MEM_CGROUP_EVENTS_NSTATS
,
102 * Per memcg event counter is incremented at every pagein/pageout. With THP,
103 * it will be incremated by the number of pages. This counter is used for
104 * for trigger some periodic events. This is straightforward and better
105 * than using jiffies etc. to handle periodic memcg event.
107 enum mem_cgroup_events_target
{
108 MEM_CGROUP_TARGET_THRESH
,
109 MEM_CGROUP_TARGET_SOFTLIMIT
,
110 MEM_CGROUP_TARGET_NUMAINFO
,
113 #define THRESHOLDS_EVENTS_TARGET 128
114 #define SOFTLIMIT_EVENTS_TARGET 1024
115 #define NUMAINFO_EVENTS_TARGET 1024
117 struct mem_cgroup_stat_cpu
{
118 long count
[MEM_CGROUP_STAT_NSTATS
];
119 unsigned long events
[MEM_CGROUP_EVENTS_NSTATS
];
120 unsigned long nr_page_events
;
121 unsigned long targets
[MEM_CGROUP_NTARGETS
];
124 struct mem_cgroup_reclaim_iter
{
125 /* css_id of the last scanned hierarchy member */
127 /* scan generation, increased every round-trip */
128 unsigned int generation
;
132 * per-zone information in memory controller.
134 struct mem_cgroup_per_zone
{
135 struct lruvec lruvec
;
136 unsigned long lru_size
[NR_LRU_LISTS
];
138 struct mem_cgroup_reclaim_iter reclaim_iter
[DEF_PRIORITY
+ 1];
140 struct rb_node tree_node
; /* RB tree node */
141 unsigned long long usage_in_excess
;/* Set to the value by which */
142 /* the soft limit is exceeded*/
144 struct mem_cgroup
*memcg
; /* Back pointer, we cannot */
145 /* use container_of */
148 struct mem_cgroup_per_node
{
149 struct mem_cgroup_per_zone zoneinfo
[MAX_NR_ZONES
];
152 struct mem_cgroup_lru_info
{
153 struct mem_cgroup_per_node
*nodeinfo
[MAX_NUMNODES
];
157 * Cgroups above their limits are maintained in a RB-Tree, independent of
158 * their hierarchy representation
161 struct mem_cgroup_tree_per_zone
{
162 struct rb_root rb_root
;
166 struct mem_cgroup_tree_per_node
{
167 struct mem_cgroup_tree_per_zone rb_tree_per_zone
[MAX_NR_ZONES
];
170 struct mem_cgroup_tree
{
171 struct mem_cgroup_tree_per_node
*rb_tree_per_node
[MAX_NUMNODES
];
174 static struct mem_cgroup_tree soft_limit_tree __read_mostly
;
176 struct mem_cgroup_threshold
{
177 struct eventfd_ctx
*eventfd
;
182 struct mem_cgroup_threshold_ary
{
183 /* An array index points to threshold just below or equal to usage. */
184 int current_threshold
;
185 /* Size of entries[] */
187 /* Array of thresholds */
188 struct mem_cgroup_threshold entries
[0];
191 struct mem_cgroup_thresholds
{
192 /* Primary thresholds array */
193 struct mem_cgroup_threshold_ary
*primary
;
195 * Spare threshold array.
196 * This is needed to make mem_cgroup_unregister_event() "never fail".
197 * It must be able to store at least primary->size - 1 entries.
199 struct mem_cgroup_threshold_ary
*spare
;
203 struct mem_cgroup_eventfd_list
{
204 struct list_head list
;
205 struct eventfd_ctx
*eventfd
;
208 static void mem_cgroup_threshold(struct mem_cgroup
*memcg
);
209 static void mem_cgroup_oom_notify(struct mem_cgroup
*memcg
);
212 * The memory controller data structure. The memory controller controls both
213 * page cache and RSS per cgroup. We would eventually like to provide
214 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
215 * to help the administrator determine what knobs to tune.
217 * TODO: Add a water mark for the memory controller. Reclaim will begin when
218 * we hit the water mark. May be even add a low water mark, such that
219 * no reclaim occurs from a cgroup at it's low water mark, this is
220 * a feature that will be implemented much later in the future.
223 struct cgroup_subsys_state css
;
225 * the counter to account for memory usage
227 struct res_counter res
;
231 * the counter to account for mem+swap usage.
233 struct res_counter memsw
;
236 * rcu_freeing is used only when freeing struct mem_cgroup,
237 * so put it into a union to avoid wasting more memory.
238 * It must be disjoint from the css field. It could be
239 * in a union with the res field, but res plays a much
240 * larger part in mem_cgroup life than memsw, and might
241 * be of interest, even at time of free, when debugging.
242 * So share rcu_head with the less interesting memsw.
244 struct rcu_head rcu_freeing
;
246 * But when using vfree(), that cannot be done at
247 * interrupt time, so we must then queue the work.
249 struct work_struct work_freeing
;
253 * Per cgroup active and inactive list, similar to the
254 * per zone LRU lists.
256 struct mem_cgroup_lru_info info
;
257 int last_scanned_node
;
259 nodemask_t scan_nodes
;
260 atomic_t numainfo_events
;
261 atomic_t numainfo_updating
;
264 * Should the accounting and control be hierarchical, per subtree?
274 /* OOM-Killer disable */
275 int oom_kill_disable
;
277 /* set when res.limit == memsw.limit */
278 bool memsw_is_minimum
;
280 /* protect arrays of thresholds */
281 struct mutex thresholds_lock
;
283 /* thresholds for memory usage. RCU-protected */
284 struct mem_cgroup_thresholds thresholds
;
286 /* thresholds for mem+swap usage. RCU-protected */
287 struct mem_cgroup_thresholds memsw_thresholds
;
289 /* For oom notifier event fd */
290 struct list_head oom_notify
;
293 * Should we move charges of a task when a task is moved into this
294 * mem_cgroup ? And what type of charges should we move ?
296 unsigned long move_charge_at_immigrate
;
298 * set > 0 if pages under this cgroup are moving to other cgroup.
300 atomic_t moving_account
;
301 /* taken only while moving_account > 0 */
302 spinlock_t move_lock
;
306 struct mem_cgroup_stat_cpu __percpu
*stat
;
308 * used when a cpu is offlined or other synchronizations
309 * See mem_cgroup_read_stat().
311 struct mem_cgroup_stat_cpu nocpu_base
;
312 spinlock_t pcp_counter_lock
;
315 struct tcp_memcontrol tcp_mem
;
319 /* Stuffs for move charges at task migration. */
321 * Types of charges to be moved. "move_charge_at_immitgrate" is treated as a
322 * left-shifted bitmap of these types.
325 MOVE_CHARGE_TYPE_ANON
, /* private anonymous page and swap of it */
326 MOVE_CHARGE_TYPE_FILE
, /* file page(including tmpfs) and swap of it */
330 /* "mc" and its members are protected by cgroup_mutex */
331 static struct move_charge_struct
{
332 spinlock_t lock
; /* for from, to */
333 struct mem_cgroup
*from
;
334 struct mem_cgroup
*to
;
335 unsigned long precharge
;
336 unsigned long moved_charge
;
337 unsigned long moved_swap
;
338 struct task_struct
*moving_task
; /* a task moving charges */
339 wait_queue_head_t waitq
; /* a waitq for other context */
341 .lock
= __SPIN_LOCK_UNLOCKED(mc
.lock
),
342 .waitq
= __WAIT_QUEUE_HEAD_INITIALIZER(mc
.waitq
),
345 static bool move_anon(void)
347 return test_bit(MOVE_CHARGE_TYPE_ANON
,
348 &mc
.to
->move_charge_at_immigrate
);
351 static bool move_file(void)
353 return test_bit(MOVE_CHARGE_TYPE_FILE
,
354 &mc
.to
->move_charge_at_immigrate
);
358 * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft
359 * limit reclaim to prevent infinite loops, if they ever occur.
361 #define MEM_CGROUP_MAX_RECLAIM_LOOPS 100
362 #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2
365 MEM_CGROUP_CHARGE_TYPE_CACHE
= 0,
366 MEM_CGROUP_CHARGE_TYPE_MAPPED
,
367 MEM_CGROUP_CHARGE_TYPE_SHMEM
, /* used by page migration of shmem */
368 MEM_CGROUP_CHARGE_TYPE_FORCE
, /* used by force_empty */
369 MEM_CGROUP_CHARGE_TYPE_SWAPOUT
, /* for accounting swapcache */
370 MEM_CGROUP_CHARGE_TYPE_DROP
, /* a page was unused swap cache */
374 /* for encoding cft->private value on file */
377 #define _OOM_TYPE (2)
378 #define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val))
379 #define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff)
380 #define MEMFILE_ATTR(val) ((val) & 0xffff)
381 /* Used for OOM nofiier */
382 #define OOM_CONTROL (0)
385 * Reclaim flags for mem_cgroup_hierarchical_reclaim
387 #define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0
388 #define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT)
389 #define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1
390 #define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT)
392 static void mem_cgroup_get(struct mem_cgroup
*memcg
);
393 static void mem_cgroup_put(struct mem_cgroup
*memcg
);
395 /* Writing them here to avoid exposing memcg's inner layout */
396 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_KMEM
397 #include <net/sock.h>
400 static bool mem_cgroup_is_root(struct mem_cgroup
*memcg
);
401 void sock_update_memcg(struct sock
*sk
)
403 if (mem_cgroup_sockets_enabled
) {
404 struct mem_cgroup
*memcg
;
406 BUG_ON(!sk
->sk_prot
->proto_cgroup
);
408 /* Socket cloning can throw us here with sk_cgrp already
409 * filled. It won't however, necessarily happen from
410 * process context. So the test for root memcg given
411 * the current task's memcg won't help us in this case.
413 * Respecting the original socket's memcg is a better
414 * decision in this case.
417 BUG_ON(mem_cgroup_is_root(sk
->sk_cgrp
->memcg
));
418 mem_cgroup_get(sk
->sk_cgrp
->memcg
);
423 memcg
= mem_cgroup_from_task(current
);
424 if (!mem_cgroup_is_root(memcg
)) {
425 mem_cgroup_get(memcg
);
426 sk
->sk_cgrp
= sk
->sk_prot
->proto_cgroup(memcg
);
431 EXPORT_SYMBOL(sock_update_memcg
);
433 void sock_release_memcg(struct sock
*sk
)
435 if (mem_cgroup_sockets_enabled
&& sk
->sk_cgrp
) {
436 struct mem_cgroup
*memcg
;
437 WARN_ON(!sk
->sk_cgrp
->memcg
);
438 memcg
= sk
->sk_cgrp
->memcg
;
439 mem_cgroup_put(memcg
);
444 struct cg_proto
*tcp_proto_cgroup(struct mem_cgroup
*memcg
)
446 if (!memcg
|| mem_cgroup_is_root(memcg
))
449 return &memcg
->tcp_mem
.cg_proto
;
451 EXPORT_SYMBOL(tcp_proto_cgroup
);
452 #endif /* CONFIG_INET */
453 #endif /* CONFIG_CGROUP_MEM_RES_CTLR_KMEM */
455 static void drain_all_stock_async(struct mem_cgroup
*memcg
);
457 static struct mem_cgroup_per_zone
*
458 mem_cgroup_zoneinfo(struct mem_cgroup
*memcg
, int nid
, int zid
)
460 return &memcg
->info
.nodeinfo
[nid
]->zoneinfo
[zid
];
463 struct cgroup_subsys_state
*mem_cgroup_css(struct mem_cgroup
*memcg
)
468 static struct mem_cgroup_per_zone
*
469 page_cgroup_zoneinfo(struct mem_cgroup
*memcg
, struct page
*page
)
471 int nid
= page_to_nid(page
);
472 int zid
= page_zonenum(page
);
474 return mem_cgroup_zoneinfo(memcg
, nid
, zid
);
477 static struct mem_cgroup_tree_per_zone
*
478 soft_limit_tree_node_zone(int nid
, int zid
)
480 return &soft_limit_tree
.rb_tree_per_node
[nid
]->rb_tree_per_zone
[zid
];
483 static struct mem_cgroup_tree_per_zone
*
484 soft_limit_tree_from_page(struct page
*page
)
486 int nid
= page_to_nid(page
);
487 int zid
= page_zonenum(page
);
489 return &soft_limit_tree
.rb_tree_per_node
[nid
]->rb_tree_per_zone
[zid
];
493 __mem_cgroup_insert_exceeded(struct mem_cgroup
*memcg
,
494 struct mem_cgroup_per_zone
*mz
,
495 struct mem_cgroup_tree_per_zone
*mctz
,
496 unsigned long long new_usage_in_excess
)
498 struct rb_node
**p
= &mctz
->rb_root
.rb_node
;
499 struct rb_node
*parent
= NULL
;
500 struct mem_cgroup_per_zone
*mz_node
;
505 mz
->usage_in_excess
= new_usage_in_excess
;
506 if (!mz
->usage_in_excess
)
510 mz_node
= rb_entry(parent
, struct mem_cgroup_per_zone
,
512 if (mz
->usage_in_excess
< mz_node
->usage_in_excess
)
515 * We can't avoid mem cgroups that are over their soft
516 * limit by the same amount
518 else if (mz
->usage_in_excess
>= mz_node
->usage_in_excess
)
521 rb_link_node(&mz
->tree_node
, parent
, p
);
522 rb_insert_color(&mz
->tree_node
, &mctz
->rb_root
);
527 __mem_cgroup_remove_exceeded(struct mem_cgroup
*memcg
,
528 struct mem_cgroup_per_zone
*mz
,
529 struct mem_cgroup_tree_per_zone
*mctz
)
533 rb_erase(&mz
->tree_node
, &mctz
->rb_root
);
538 mem_cgroup_remove_exceeded(struct mem_cgroup
*memcg
,
539 struct mem_cgroup_per_zone
*mz
,
540 struct mem_cgroup_tree_per_zone
*mctz
)
542 spin_lock(&mctz
->lock
);
543 __mem_cgroup_remove_exceeded(memcg
, mz
, mctz
);
544 spin_unlock(&mctz
->lock
);
548 static void mem_cgroup_update_tree(struct mem_cgroup
*memcg
, struct page
*page
)
550 unsigned long long excess
;
551 struct mem_cgroup_per_zone
*mz
;
552 struct mem_cgroup_tree_per_zone
*mctz
;
553 int nid
= page_to_nid(page
);
554 int zid
= page_zonenum(page
);
555 mctz
= soft_limit_tree_from_page(page
);
558 * Necessary to update all ancestors when hierarchy is used.
559 * because their event counter is not touched.
561 for (; memcg
; memcg
= parent_mem_cgroup(memcg
)) {
562 mz
= mem_cgroup_zoneinfo(memcg
, nid
, zid
);
563 excess
= res_counter_soft_limit_excess(&memcg
->res
);
565 * We have to update the tree if mz is on RB-tree or
566 * mem is over its softlimit.
568 if (excess
|| mz
->on_tree
) {
569 spin_lock(&mctz
->lock
);
570 /* if on-tree, remove it */
572 __mem_cgroup_remove_exceeded(memcg
, mz
, mctz
);
574 * Insert again. mz->usage_in_excess will be updated.
575 * If excess is 0, no tree ops.
577 __mem_cgroup_insert_exceeded(memcg
, mz
, mctz
, excess
);
578 spin_unlock(&mctz
->lock
);
583 static void mem_cgroup_remove_from_trees(struct mem_cgroup
*memcg
)
586 struct mem_cgroup_per_zone
*mz
;
587 struct mem_cgroup_tree_per_zone
*mctz
;
589 for_each_node(node
) {
590 for (zone
= 0; zone
< MAX_NR_ZONES
; zone
++) {
591 mz
= mem_cgroup_zoneinfo(memcg
, node
, zone
);
592 mctz
= soft_limit_tree_node_zone(node
, zone
);
593 mem_cgroup_remove_exceeded(memcg
, mz
, mctz
);
598 static struct mem_cgroup_per_zone
*
599 __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone
*mctz
)
601 struct rb_node
*rightmost
= NULL
;
602 struct mem_cgroup_per_zone
*mz
;
606 rightmost
= rb_last(&mctz
->rb_root
);
608 goto done
; /* Nothing to reclaim from */
610 mz
= rb_entry(rightmost
, struct mem_cgroup_per_zone
, tree_node
);
612 * Remove the node now but someone else can add it back,
613 * we will to add it back at the end of reclaim to its correct
614 * position in the tree.
616 __mem_cgroup_remove_exceeded(mz
->memcg
, mz
, mctz
);
617 if (!res_counter_soft_limit_excess(&mz
->memcg
->res
) ||
618 !css_tryget(&mz
->memcg
->css
))
624 static struct mem_cgroup_per_zone
*
625 mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone
*mctz
)
627 struct mem_cgroup_per_zone
*mz
;
629 spin_lock(&mctz
->lock
);
630 mz
= __mem_cgroup_largest_soft_limit_node(mctz
);
631 spin_unlock(&mctz
->lock
);
636 * Implementation Note: reading percpu statistics for memcg.
638 * Both of vmstat[] and percpu_counter has threshold and do periodic
639 * synchronization to implement "quick" read. There are trade-off between
640 * reading cost and precision of value. Then, we may have a chance to implement
641 * a periodic synchronizion of counter in memcg's counter.
643 * But this _read() function is used for user interface now. The user accounts
644 * memory usage by memory cgroup and he _always_ requires exact value because
645 * he accounts memory. Even if we provide quick-and-fuzzy read, we always
646 * have to visit all online cpus and make sum. So, for now, unnecessary
647 * synchronization is not implemented. (just implemented for cpu hotplug)
649 * If there are kernel internal actions which can make use of some not-exact
650 * value, and reading all cpu value can be performance bottleneck in some
651 * common workload, threashold and synchonization as vmstat[] should be
654 static long mem_cgroup_read_stat(struct mem_cgroup
*memcg
,
655 enum mem_cgroup_stat_index idx
)
661 for_each_online_cpu(cpu
)
662 val
+= per_cpu(memcg
->stat
->count
[idx
], cpu
);
663 #ifdef CONFIG_HOTPLUG_CPU
664 spin_lock(&memcg
->pcp_counter_lock
);
665 val
+= memcg
->nocpu_base
.count
[idx
];
666 spin_unlock(&memcg
->pcp_counter_lock
);
672 static void mem_cgroup_swap_statistics(struct mem_cgroup
*memcg
,
675 int val
= (charge
) ? 1 : -1;
676 this_cpu_add(memcg
->stat
->count
[MEM_CGROUP_STAT_SWAPOUT
], val
);
679 static unsigned long mem_cgroup_read_events(struct mem_cgroup
*memcg
,
680 enum mem_cgroup_events_index idx
)
682 unsigned long val
= 0;
685 for_each_online_cpu(cpu
)
686 val
+= per_cpu(memcg
->stat
->events
[idx
], cpu
);
687 #ifdef CONFIG_HOTPLUG_CPU
688 spin_lock(&memcg
->pcp_counter_lock
);
689 val
+= memcg
->nocpu_base
.events
[idx
];
690 spin_unlock(&memcg
->pcp_counter_lock
);
695 static void mem_cgroup_charge_statistics(struct mem_cgroup
*memcg
,
696 bool anon
, int nr_pages
)
701 * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is
702 * counted as CACHE even if it's on ANON LRU.
705 __this_cpu_add(memcg
->stat
->count
[MEM_CGROUP_STAT_RSS
],
708 __this_cpu_add(memcg
->stat
->count
[MEM_CGROUP_STAT_CACHE
],
711 /* pagein of a big page is an event. So, ignore page size */
713 __this_cpu_inc(memcg
->stat
->events
[MEM_CGROUP_EVENTS_PGPGIN
]);
715 __this_cpu_inc(memcg
->stat
->events
[MEM_CGROUP_EVENTS_PGPGOUT
]);
716 nr_pages
= -nr_pages
; /* for event */
719 __this_cpu_add(memcg
->stat
->nr_page_events
, nr_pages
);
725 mem_cgroup_get_lruvec_size(struct lruvec
*lruvec
, enum lru_list lru
)
727 struct mem_cgroup_per_zone
*mz
;
729 mz
= container_of(lruvec
, struct mem_cgroup_per_zone
, lruvec
);
730 return mz
->lru_size
[lru
];
734 mem_cgroup_zone_nr_lru_pages(struct mem_cgroup
*memcg
, int nid
, int zid
,
735 unsigned int lru_mask
)
737 struct mem_cgroup_per_zone
*mz
;
739 unsigned long ret
= 0;
741 mz
= mem_cgroup_zoneinfo(memcg
, nid
, zid
);
744 if (BIT(lru
) & lru_mask
)
745 ret
+= mz
->lru_size
[lru
];
751 mem_cgroup_node_nr_lru_pages(struct mem_cgroup
*memcg
,
752 int nid
, unsigned int lru_mask
)
757 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++)
758 total
+= mem_cgroup_zone_nr_lru_pages(memcg
,
764 static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup
*memcg
,
765 unsigned int lru_mask
)
770 for_each_node_state(nid
, N_HIGH_MEMORY
)
771 total
+= mem_cgroup_node_nr_lru_pages(memcg
, nid
, lru_mask
);
775 static bool mem_cgroup_event_ratelimit(struct mem_cgroup
*memcg
,
776 enum mem_cgroup_events_target target
)
778 unsigned long val
, next
;
780 val
= __this_cpu_read(memcg
->stat
->nr_page_events
);
781 next
= __this_cpu_read(memcg
->stat
->targets
[target
]);
782 /* from time_after() in jiffies.h */
783 if ((long)next
- (long)val
< 0) {
785 case MEM_CGROUP_TARGET_THRESH
:
786 next
= val
+ THRESHOLDS_EVENTS_TARGET
;
788 case MEM_CGROUP_TARGET_SOFTLIMIT
:
789 next
= val
+ SOFTLIMIT_EVENTS_TARGET
;
791 case MEM_CGROUP_TARGET_NUMAINFO
:
792 next
= val
+ NUMAINFO_EVENTS_TARGET
;
797 __this_cpu_write(memcg
->stat
->targets
[target
], next
);
804 * Check events in order.
807 static void memcg_check_events(struct mem_cgroup
*memcg
, struct page
*page
)
810 /* threshold event is triggered in finer grain than soft limit */
811 if (unlikely(mem_cgroup_event_ratelimit(memcg
,
812 MEM_CGROUP_TARGET_THRESH
))) {
814 bool do_numainfo __maybe_unused
;
816 do_softlimit
= mem_cgroup_event_ratelimit(memcg
,
817 MEM_CGROUP_TARGET_SOFTLIMIT
);
819 do_numainfo
= mem_cgroup_event_ratelimit(memcg
,
820 MEM_CGROUP_TARGET_NUMAINFO
);
824 mem_cgroup_threshold(memcg
);
825 if (unlikely(do_softlimit
))
826 mem_cgroup_update_tree(memcg
, page
);
828 if (unlikely(do_numainfo
))
829 atomic_inc(&memcg
->numainfo_events
);
835 struct mem_cgroup
*mem_cgroup_from_cont(struct cgroup
*cont
)
837 return container_of(cgroup_subsys_state(cont
,
838 mem_cgroup_subsys_id
), struct mem_cgroup
,
842 struct mem_cgroup
*mem_cgroup_from_task(struct task_struct
*p
)
845 * mm_update_next_owner() may clear mm->owner to NULL
846 * if it races with swapoff, page migration, etc.
847 * So this can be called with p == NULL.
852 return container_of(task_subsys_state(p
, mem_cgroup_subsys_id
),
853 struct mem_cgroup
, css
);
856 struct mem_cgroup
*try_get_mem_cgroup_from_mm(struct mm_struct
*mm
)
858 struct mem_cgroup
*memcg
= NULL
;
863 * Because we have no locks, mm->owner's may be being moved to other
864 * cgroup. We use css_tryget() here even if this looks
865 * pessimistic (rather than adding locks here).
869 memcg
= mem_cgroup_from_task(rcu_dereference(mm
->owner
));
870 if (unlikely(!memcg
))
872 } while (!css_tryget(&memcg
->css
));
878 * mem_cgroup_iter - iterate over memory cgroup hierarchy
879 * @root: hierarchy root
880 * @prev: previously returned memcg, NULL on first invocation
881 * @reclaim: cookie for shared reclaim walks, NULL for full walks
883 * Returns references to children of the hierarchy below @root, or
884 * @root itself, or %NULL after a full round-trip.
886 * Caller must pass the return value in @prev on subsequent
887 * invocations for reference counting, or use mem_cgroup_iter_break()
888 * to cancel a hierarchy walk before the round-trip is complete.
890 * Reclaimers can specify a zone and a priority level in @reclaim to
891 * divide up the memcgs in the hierarchy among all concurrent
892 * reclaimers operating on the same zone and priority.
894 struct mem_cgroup
*mem_cgroup_iter(struct mem_cgroup
*root
,
895 struct mem_cgroup
*prev
,
896 struct mem_cgroup_reclaim_cookie
*reclaim
)
898 struct mem_cgroup
*memcg
= NULL
;
901 if (mem_cgroup_disabled())
905 root
= root_mem_cgroup
;
907 if (prev
&& !reclaim
)
908 id
= css_id(&prev
->css
);
910 if (prev
&& prev
!= root
)
913 if (!root
->use_hierarchy
&& root
!= root_mem_cgroup
) {
920 struct mem_cgroup_reclaim_iter
*uninitialized_var(iter
);
921 struct cgroup_subsys_state
*css
;
924 int nid
= zone_to_nid(reclaim
->zone
);
925 int zid
= zone_idx(reclaim
->zone
);
926 struct mem_cgroup_per_zone
*mz
;
928 mz
= mem_cgroup_zoneinfo(root
, nid
, zid
);
929 iter
= &mz
->reclaim_iter
[reclaim
->priority
];
930 if (prev
&& reclaim
->generation
!= iter
->generation
)
936 css
= css_get_next(&mem_cgroup_subsys
, id
+ 1, &root
->css
, &id
);
938 if (css
== &root
->css
|| css_tryget(css
))
939 memcg
= container_of(css
,
940 struct mem_cgroup
, css
);
949 else if (!prev
&& memcg
)
950 reclaim
->generation
= iter
->generation
;
960 * mem_cgroup_iter_break - abort a hierarchy walk prematurely
961 * @root: hierarchy root
962 * @prev: last visited hierarchy member as returned by mem_cgroup_iter()
964 void mem_cgroup_iter_break(struct mem_cgroup
*root
,
965 struct mem_cgroup
*prev
)
968 root
= root_mem_cgroup
;
969 if (prev
&& prev
!= root
)
974 * Iteration constructs for visiting all cgroups (under a tree). If
975 * loops are exited prematurely (break), mem_cgroup_iter_break() must
976 * be used for reference counting.
978 #define for_each_mem_cgroup_tree(iter, root) \
979 for (iter = mem_cgroup_iter(root, NULL, NULL); \
981 iter = mem_cgroup_iter(root, iter, NULL))
983 #define for_each_mem_cgroup(iter) \
984 for (iter = mem_cgroup_iter(NULL, NULL, NULL); \
986 iter = mem_cgroup_iter(NULL, iter, NULL))
988 static inline bool mem_cgroup_is_root(struct mem_cgroup
*memcg
)
990 return (memcg
== root_mem_cgroup
);
993 void mem_cgroup_count_vm_event(struct mm_struct
*mm
, enum vm_event_item idx
)
995 struct mem_cgroup
*memcg
;
1001 memcg
= mem_cgroup_from_task(rcu_dereference(mm
->owner
));
1002 if (unlikely(!memcg
))
1007 this_cpu_inc(memcg
->stat
->events
[MEM_CGROUP_EVENTS_PGFAULT
]);
1010 this_cpu_inc(memcg
->stat
->events
[MEM_CGROUP_EVENTS_PGMAJFAULT
]);
1018 EXPORT_SYMBOL(mem_cgroup_count_vm_event
);
1021 * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg
1022 * @zone: zone of the wanted lruvec
1023 * @mem: memcg of the wanted lruvec
1025 * Returns the lru list vector holding pages for the given @zone and
1026 * @mem. This can be the global zone lruvec, if the memory controller
1029 struct lruvec
*mem_cgroup_zone_lruvec(struct zone
*zone
,
1030 struct mem_cgroup
*memcg
)
1032 struct mem_cgroup_per_zone
*mz
;
1034 if (mem_cgroup_disabled())
1035 return &zone
->lruvec
;
1037 mz
= mem_cgroup_zoneinfo(memcg
, zone_to_nid(zone
), zone_idx(zone
));
1042 * Following LRU functions are allowed to be used without PCG_LOCK.
1043 * Operations are called by routine of global LRU independently from memcg.
1044 * What we have to take care of here is validness of pc->mem_cgroup.
1046 * Changes to pc->mem_cgroup happens when
1049 * In typical case, "charge" is done before add-to-lru. Exception is SwapCache.
1050 * It is added to LRU before charge.
1051 * If PCG_USED bit is not set, page_cgroup is not added to this private LRU.
1052 * When moving account, the page is not on LRU. It's isolated.
1056 * mem_cgroup_lru_add_list - account for adding an lru page and return lruvec
1057 * @zone: zone of the page
1061 * This function accounts for @page being added to @lru, and returns
1062 * the lruvec for the given @zone and the memcg @page is charged to.
1064 * The callsite is then responsible for physically linking the page to
1065 * the returned lruvec->lists[@lru].
1067 struct lruvec
*mem_cgroup_lru_add_list(struct zone
*zone
, struct page
*page
,
1070 struct mem_cgroup_per_zone
*mz
;
1071 struct mem_cgroup
*memcg
;
1072 struct page_cgroup
*pc
;
1074 if (mem_cgroup_disabled())
1075 return &zone
->lruvec
;
1077 pc
= lookup_page_cgroup(page
);
1078 memcg
= pc
->mem_cgroup
;
1081 * Surreptitiously switch any uncharged page to root:
1082 * an uncharged page off lru does nothing to secure
1083 * its former mem_cgroup from sudden removal.
1085 * Our caller holds lru_lock, and PageCgroupUsed is updated
1086 * under page_cgroup lock: between them, they make all uses
1087 * of pc->mem_cgroup safe.
1089 if (!PageCgroupUsed(pc
) && memcg
!= root_mem_cgroup
)
1090 pc
->mem_cgroup
= memcg
= root_mem_cgroup
;
1092 mz
= page_cgroup_zoneinfo(memcg
, page
);
1093 /* compound_order() is stabilized through lru_lock */
1094 mz
->lru_size
[lru
] += 1 << compound_order(page
);
1099 * mem_cgroup_lru_del_list - account for removing an lru page
1103 * This function accounts for @page being removed from @lru.
1105 * The callsite is then responsible for physically unlinking
1108 void mem_cgroup_lru_del_list(struct page
*page
, enum lru_list lru
)
1110 struct mem_cgroup_per_zone
*mz
;
1111 struct mem_cgroup
*memcg
;
1112 struct page_cgroup
*pc
;
1114 if (mem_cgroup_disabled())
1117 pc
= lookup_page_cgroup(page
);
1118 memcg
= pc
->mem_cgroup
;
1120 mz
= page_cgroup_zoneinfo(memcg
, page
);
1121 /* huge page split is done under lru_lock. so, we have no races. */
1122 VM_BUG_ON(mz
->lru_size
[lru
] < (1 << compound_order(page
)));
1123 mz
->lru_size
[lru
] -= 1 << compound_order(page
);
1127 * mem_cgroup_lru_move_lists - account for moving a page between lrus
1128 * @zone: zone of the page
1130 * @from: current lru
1133 * This function accounts for @page being moved between the lrus @from
1134 * and @to, and returns the lruvec for the given @zone and the memcg
1135 * @page is charged to.
1137 * The callsite is then responsible for physically relinking
1138 * @page->lru to the returned lruvec->lists[@to].
1140 struct lruvec
*mem_cgroup_lru_move_lists(struct zone
*zone
,
1145 /* XXX: Optimize this, especially for @from == @to */
1146 mem_cgroup_lru_del_list(page
, from
);
1147 return mem_cgroup_lru_add_list(zone
, page
, to
);
1151 * Checks whether given mem is same or in the root_mem_cgroup's
1154 bool __mem_cgroup_same_or_subtree(const struct mem_cgroup
*root_memcg
,
1155 struct mem_cgroup
*memcg
)
1157 if (root_memcg
== memcg
)
1159 if (!root_memcg
->use_hierarchy
)
1161 return css_is_ancestor(&memcg
->css
, &root_memcg
->css
);
1164 static bool mem_cgroup_same_or_subtree(const struct mem_cgroup
*root_memcg
,
1165 struct mem_cgroup
*memcg
)
1170 ret
= __mem_cgroup_same_or_subtree(root_memcg
, memcg
);
1175 int task_in_mem_cgroup(struct task_struct
*task
, const struct mem_cgroup
*memcg
)
1178 struct mem_cgroup
*curr
= NULL
;
1179 struct task_struct
*p
;
1181 p
= find_lock_task_mm(task
);
1183 curr
= try_get_mem_cgroup_from_mm(p
->mm
);
1187 * All threads may have already detached their mm's, but the oom
1188 * killer still needs to detect if they have already been oom
1189 * killed to prevent needlessly killing additional tasks.
1192 curr
= mem_cgroup_from_task(task
);
1194 css_get(&curr
->css
);
1200 * We should check use_hierarchy of "memcg" not "curr". Because checking
1201 * use_hierarchy of "curr" here make this function true if hierarchy is
1202 * enabled in "curr" and "curr" is a child of "memcg" in *cgroup*
1203 * hierarchy(even if use_hierarchy is disabled in "memcg").
1205 ret
= mem_cgroup_same_or_subtree(memcg
, curr
);
1206 css_put(&curr
->css
);
1210 int mem_cgroup_inactive_anon_is_low(struct lruvec
*lruvec
)
1212 unsigned long inactive_ratio
;
1213 unsigned long inactive
;
1214 unsigned long active
;
1217 inactive
= mem_cgroup_get_lruvec_size(lruvec
, LRU_INACTIVE_ANON
);
1218 active
= mem_cgroup_get_lruvec_size(lruvec
, LRU_ACTIVE_ANON
);
1220 gb
= (inactive
+ active
) >> (30 - PAGE_SHIFT
);
1222 inactive_ratio
= int_sqrt(10 * gb
);
1226 return inactive
* inactive_ratio
< active
;
1229 int mem_cgroup_inactive_file_is_low(struct lruvec
*lruvec
)
1231 unsigned long active
;
1232 unsigned long inactive
;
1234 inactive
= mem_cgroup_get_lruvec_size(lruvec
, LRU_INACTIVE_FILE
);
1235 active
= mem_cgroup_get_lruvec_size(lruvec
, LRU_ACTIVE_FILE
);
1237 return (active
> inactive
);
1240 struct zone_reclaim_stat
*
1241 mem_cgroup_get_reclaim_stat_from_page(struct page
*page
)
1243 struct page_cgroup
*pc
;
1244 struct mem_cgroup_per_zone
*mz
;
1246 if (mem_cgroup_disabled())
1249 pc
= lookup_page_cgroup(page
);
1250 if (!PageCgroupUsed(pc
))
1252 /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */
1254 mz
= page_cgroup_zoneinfo(pc
->mem_cgroup
, page
);
1255 return &mz
->lruvec
.reclaim_stat
;
1258 #define mem_cgroup_from_res_counter(counter, member) \
1259 container_of(counter, struct mem_cgroup, member)
1262 * mem_cgroup_margin - calculate chargeable space of a memory cgroup
1263 * @mem: the memory cgroup
1265 * Returns the maximum amount of memory @mem can be charged with, in
1268 static unsigned long mem_cgroup_margin(struct mem_cgroup
*memcg
)
1270 unsigned long long margin
;
1272 margin
= res_counter_margin(&memcg
->res
);
1273 if (do_swap_account
)
1274 margin
= min(margin
, res_counter_margin(&memcg
->memsw
));
1275 return margin
>> PAGE_SHIFT
;
1278 int mem_cgroup_swappiness(struct mem_cgroup
*memcg
)
1280 struct cgroup
*cgrp
= memcg
->css
.cgroup
;
1283 if (cgrp
->parent
== NULL
)
1284 return vm_swappiness
;
1286 return memcg
->swappiness
;
1290 * memcg->moving_account is used for checking possibility that some thread is
1291 * calling move_account(). When a thread on CPU-A starts moving pages under
1292 * a memcg, other threads should check memcg->moving_account under
1293 * rcu_read_lock(), like this:
1297 * memcg->moving_account+1 if (memcg->mocing_account)
1299 * synchronize_rcu() update something.
1304 /* for quick checking without looking up memcg */
1305 atomic_t memcg_moving __read_mostly
;
1307 static void mem_cgroup_start_move(struct mem_cgroup
*memcg
)
1309 atomic_inc(&memcg_moving
);
1310 atomic_inc(&memcg
->moving_account
);
1314 static void mem_cgroup_end_move(struct mem_cgroup
*memcg
)
1317 * Now, mem_cgroup_clear_mc() may call this function with NULL.
1318 * We check NULL in callee rather than caller.
1321 atomic_dec(&memcg_moving
);
1322 atomic_dec(&memcg
->moving_account
);
1327 * 2 routines for checking "mem" is under move_account() or not.
1329 * mem_cgroup_stolen() - checking whether a cgroup is mc.from or not. This
1330 * is used for avoiding races in accounting. If true,
1331 * pc->mem_cgroup may be overwritten.
1333 * mem_cgroup_under_move() - checking a cgroup is mc.from or mc.to or
1334 * under hierarchy of moving cgroups. This is for
1335 * waiting at hith-memory prressure caused by "move".
1338 static bool mem_cgroup_stolen(struct mem_cgroup
*memcg
)
1340 VM_BUG_ON(!rcu_read_lock_held());
1341 return atomic_read(&memcg
->moving_account
) > 0;
1344 static bool mem_cgroup_under_move(struct mem_cgroup
*memcg
)
1346 struct mem_cgroup
*from
;
1347 struct mem_cgroup
*to
;
1350 * Unlike task_move routines, we access mc.to, mc.from not under
1351 * mutual exclusion by cgroup_mutex. Here, we take spinlock instead.
1353 spin_lock(&mc
.lock
);
1359 ret
= mem_cgroup_same_or_subtree(memcg
, from
)
1360 || mem_cgroup_same_or_subtree(memcg
, to
);
1362 spin_unlock(&mc
.lock
);
1366 static bool mem_cgroup_wait_acct_move(struct mem_cgroup
*memcg
)
1368 if (mc
.moving_task
&& current
!= mc
.moving_task
) {
1369 if (mem_cgroup_under_move(memcg
)) {
1371 prepare_to_wait(&mc
.waitq
, &wait
, TASK_INTERRUPTIBLE
);
1372 /* moving charge context might have finished. */
1375 finish_wait(&mc
.waitq
, &wait
);
1383 * Take this lock when
1384 * - a code tries to modify page's memcg while it's USED.
1385 * - a code tries to modify page state accounting in a memcg.
1386 * see mem_cgroup_stolen(), too.
1388 static void move_lock_mem_cgroup(struct mem_cgroup
*memcg
,
1389 unsigned long *flags
)
1391 spin_lock_irqsave(&memcg
->move_lock
, *flags
);
1394 static void move_unlock_mem_cgroup(struct mem_cgroup
*memcg
,
1395 unsigned long *flags
)
1397 spin_unlock_irqrestore(&memcg
->move_lock
, *flags
);
1401 * mem_cgroup_print_oom_info: Called from OOM with tasklist_lock held in read mode.
1402 * @memcg: The memory cgroup that went over limit
1403 * @p: Task that is going to be killed
1405 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1408 void mem_cgroup_print_oom_info(struct mem_cgroup
*memcg
, struct task_struct
*p
)
1410 struct cgroup
*task_cgrp
;
1411 struct cgroup
*mem_cgrp
;
1413 * Need a buffer in BSS, can't rely on allocations. The code relies
1414 * on the assumption that OOM is serialized for memory controller.
1415 * If this assumption is broken, revisit this code.
1417 static char memcg_name
[PATH_MAX
];
1425 mem_cgrp
= memcg
->css
.cgroup
;
1426 task_cgrp
= task_cgroup(p
, mem_cgroup_subsys_id
);
1428 ret
= cgroup_path(task_cgrp
, memcg_name
, PATH_MAX
);
1431 * Unfortunately, we are unable to convert to a useful name
1432 * But we'll still print out the usage information
1439 printk(KERN_INFO
"Task in %s killed", memcg_name
);
1442 ret
= cgroup_path(mem_cgrp
, memcg_name
, PATH_MAX
);
1450 * Continues from above, so we don't need an KERN_ level
1452 printk(KERN_CONT
" as a result of limit of %s\n", memcg_name
);
1455 printk(KERN_INFO
"memory: usage %llukB, limit %llukB, failcnt %llu\n",
1456 res_counter_read_u64(&memcg
->res
, RES_USAGE
) >> 10,
1457 res_counter_read_u64(&memcg
->res
, RES_LIMIT
) >> 10,
1458 res_counter_read_u64(&memcg
->res
, RES_FAILCNT
));
1459 printk(KERN_INFO
"memory+swap: usage %llukB, limit %llukB, "
1461 res_counter_read_u64(&memcg
->memsw
, RES_USAGE
) >> 10,
1462 res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
) >> 10,
1463 res_counter_read_u64(&memcg
->memsw
, RES_FAILCNT
));
1467 * This function returns the number of memcg under hierarchy tree. Returns
1468 * 1(self count) if no children.
1470 static int mem_cgroup_count_children(struct mem_cgroup
*memcg
)
1473 struct mem_cgroup
*iter
;
1475 for_each_mem_cgroup_tree(iter
, memcg
)
1481 * Return the memory (and swap, if configured) limit for a memcg.
1483 u64
mem_cgroup_get_limit(struct mem_cgroup
*memcg
)
1488 limit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
1489 limit
+= total_swap_pages
<< PAGE_SHIFT
;
1491 memsw
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
1493 * If memsw is finite and limits the amount of swap space available
1494 * to this memcg, return that limit.
1496 return min(limit
, memsw
);
1499 static unsigned long mem_cgroup_reclaim(struct mem_cgroup
*memcg
,
1501 unsigned long flags
)
1503 unsigned long total
= 0;
1504 bool noswap
= false;
1507 if (flags
& MEM_CGROUP_RECLAIM_NOSWAP
)
1509 if (!(flags
& MEM_CGROUP_RECLAIM_SHRINK
) && memcg
->memsw_is_minimum
)
1512 for (loop
= 0; loop
< MEM_CGROUP_MAX_RECLAIM_LOOPS
; loop
++) {
1514 drain_all_stock_async(memcg
);
1515 total
+= try_to_free_mem_cgroup_pages(memcg
, gfp_mask
, noswap
);
1517 * Allow limit shrinkers, which are triggered directly
1518 * by userspace, to catch signals and stop reclaim
1519 * after minimal progress, regardless of the margin.
1521 if (total
&& (flags
& MEM_CGROUP_RECLAIM_SHRINK
))
1523 if (mem_cgroup_margin(memcg
))
1526 * If nothing was reclaimed after two attempts, there
1527 * may be no reclaimable pages in this hierarchy.
1536 * test_mem_cgroup_node_reclaimable
1537 * @mem: the target memcg
1538 * @nid: the node ID to be checked.
1539 * @noswap : specify true here if the user wants flle only information.
1541 * This function returns whether the specified memcg contains any
1542 * reclaimable pages on a node. Returns true if there are any reclaimable
1543 * pages in the node.
1545 static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup
*memcg
,
1546 int nid
, bool noswap
)
1548 if (mem_cgroup_node_nr_lru_pages(memcg
, nid
, LRU_ALL_FILE
))
1550 if (noswap
|| !total_swap_pages
)
1552 if (mem_cgroup_node_nr_lru_pages(memcg
, nid
, LRU_ALL_ANON
))
1557 #if MAX_NUMNODES > 1
1560 * Always updating the nodemask is not very good - even if we have an empty
1561 * list or the wrong list here, we can start from some node and traverse all
1562 * nodes based on the zonelist. So update the list loosely once per 10 secs.
1565 static void mem_cgroup_may_update_nodemask(struct mem_cgroup
*memcg
)
1569 * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET
1570 * pagein/pageout changes since the last update.
1572 if (!atomic_read(&memcg
->numainfo_events
))
1574 if (atomic_inc_return(&memcg
->numainfo_updating
) > 1)
1577 /* make a nodemask where this memcg uses memory from */
1578 memcg
->scan_nodes
= node_states
[N_HIGH_MEMORY
];
1580 for_each_node_mask(nid
, node_states
[N_HIGH_MEMORY
]) {
1582 if (!test_mem_cgroup_node_reclaimable(memcg
, nid
, false))
1583 node_clear(nid
, memcg
->scan_nodes
);
1586 atomic_set(&memcg
->numainfo_events
, 0);
1587 atomic_set(&memcg
->numainfo_updating
, 0);
1591 * Selecting a node where we start reclaim from. Because what we need is just
1592 * reducing usage counter, start from anywhere is O,K. Considering
1593 * memory reclaim from current node, there are pros. and cons.
1595 * Freeing memory from current node means freeing memory from a node which
1596 * we'll use or we've used. So, it may make LRU bad. And if several threads
1597 * hit limits, it will see a contention on a node. But freeing from remote
1598 * node means more costs for memory reclaim because of memory latency.
1600 * Now, we use round-robin. Better algorithm is welcomed.
1602 int mem_cgroup_select_victim_node(struct mem_cgroup
*memcg
)
1606 mem_cgroup_may_update_nodemask(memcg
);
1607 node
= memcg
->last_scanned_node
;
1609 node
= next_node(node
, memcg
->scan_nodes
);
1610 if (node
== MAX_NUMNODES
)
1611 node
= first_node(memcg
->scan_nodes
);
1613 * We call this when we hit limit, not when pages are added to LRU.
1614 * No LRU may hold pages because all pages are UNEVICTABLE or
1615 * memcg is too small and all pages are not on LRU. In that case,
1616 * we use curret node.
1618 if (unlikely(node
== MAX_NUMNODES
))
1619 node
= numa_node_id();
1621 memcg
->last_scanned_node
= node
;
1626 * Check all nodes whether it contains reclaimable pages or not.
1627 * For quick scan, we make use of scan_nodes. This will allow us to skip
1628 * unused nodes. But scan_nodes is lazily updated and may not cotain
1629 * enough new information. We need to do double check.
1631 static bool mem_cgroup_reclaimable(struct mem_cgroup
*memcg
, bool noswap
)
1636 * quick check...making use of scan_node.
1637 * We can skip unused nodes.
1639 if (!nodes_empty(memcg
->scan_nodes
)) {
1640 for (nid
= first_node(memcg
->scan_nodes
);
1642 nid
= next_node(nid
, memcg
->scan_nodes
)) {
1644 if (test_mem_cgroup_node_reclaimable(memcg
, nid
, noswap
))
1649 * Check rest of nodes.
1651 for_each_node_state(nid
, N_HIGH_MEMORY
) {
1652 if (node_isset(nid
, memcg
->scan_nodes
))
1654 if (test_mem_cgroup_node_reclaimable(memcg
, nid
, noswap
))
1661 int mem_cgroup_select_victim_node(struct mem_cgroup
*memcg
)
1666 static bool mem_cgroup_reclaimable(struct mem_cgroup
*memcg
, bool noswap
)
1668 return test_mem_cgroup_node_reclaimable(memcg
, 0, noswap
);
1672 static int mem_cgroup_soft_reclaim(struct mem_cgroup
*root_memcg
,
1675 unsigned long *total_scanned
)
1677 struct mem_cgroup
*victim
= NULL
;
1680 unsigned long excess
;
1681 unsigned long nr_scanned
;
1682 struct mem_cgroup_reclaim_cookie reclaim
= {
1687 excess
= res_counter_soft_limit_excess(&root_memcg
->res
) >> PAGE_SHIFT
;
1690 victim
= mem_cgroup_iter(root_memcg
, victim
, &reclaim
);
1695 * If we have not been able to reclaim
1696 * anything, it might because there are
1697 * no reclaimable pages under this hierarchy
1702 * We want to do more targeted reclaim.
1703 * excess >> 2 is not to excessive so as to
1704 * reclaim too much, nor too less that we keep
1705 * coming back to reclaim from this cgroup
1707 if (total
>= (excess
>> 2) ||
1708 (loop
> MEM_CGROUP_MAX_RECLAIM_LOOPS
))
1713 if (!mem_cgroup_reclaimable(victim
, false))
1715 total
+= mem_cgroup_shrink_node_zone(victim
, gfp_mask
, false,
1717 *total_scanned
+= nr_scanned
;
1718 if (!res_counter_soft_limit_excess(&root_memcg
->res
))
1721 mem_cgroup_iter_break(root_memcg
, victim
);
1726 * Check OOM-Killer is already running under our hierarchy.
1727 * If someone is running, return false.
1728 * Has to be called with memcg_oom_lock
1730 static bool mem_cgroup_oom_lock(struct mem_cgroup
*memcg
)
1732 struct mem_cgroup
*iter
, *failed
= NULL
;
1734 for_each_mem_cgroup_tree(iter
, memcg
) {
1735 if (iter
->oom_lock
) {
1737 * this subtree of our hierarchy is already locked
1738 * so we cannot give a lock.
1741 mem_cgroup_iter_break(memcg
, iter
);
1744 iter
->oom_lock
= true;
1751 * OK, we failed to lock the whole subtree so we have to clean up
1752 * what we set up to the failing subtree
1754 for_each_mem_cgroup_tree(iter
, memcg
) {
1755 if (iter
== failed
) {
1756 mem_cgroup_iter_break(memcg
, iter
);
1759 iter
->oom_lock
= false;
1765 * Has to be called with memcg_oom_lock
1767 static int mem_cgroup_oom_unlock(struct mem_cgroup
*memcg
)
1769 struct mem_cgroup
*iter
;
1771 for_each_mem_cgroup_tree(iter
, memcg
)
1772 iter
->oom_lock
= false;
1776 static void mem_cgroup_mark_under_oom(struct mem_cgroup
*memcg
)
1778 struct mem_cgroup
*iter
;
1780 for_each_mem_cgroup_tree(iter
, memcg
)
1781 atomic_inc(&iter
->under_oom
);
1784 static void mem_cgroup_unmark_under_oom(struct mem_cgroup
*memcg
)
1786 struct mem_cgroup
*iter
;
1789 * When a new child is created while the hierarchy is under oom,
1790 * mem_cgroup_oom_lock() may not be called. We have to use
1791 * atomic_add_unless() here.
1793 for_each_mem_cgroup_tree(iter
, memcg
)
1794 atomic_add_unless(&iter
->under_oom
, -1, 0);
1797 static DEFINE_SPINLOCK(memcg_oom_lock
);
1798 static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq
);
1800 struct oom_wait_info
{
1801 struct mem_cgroup
*memcg
;
1805 static int memcg_oom_wake_function(wait_queue_t
*wait
,
1806 unsigned mode
, int sync
, void *arg
)
1808 struct mem_cgroup
*wake_memcg
= (struct mem_cgroup
*)arg
;
1809 struct mem_cgroup
*oom_wait_memcg
;
1810 struct oom_wait_info
*oom_wait_info
;
1812 oom_wait_info
= container_of(wait
, struct oom_wait_info
, wait
);
1813 oom_wait_memcg
= oom_wait_info
->memcg
;
1816 * Both of oom_wait_info->memcg and wake_memcg are stable under us.
1817 * Then we can use css_is_ancestor without taking care of RCU.
1819 if (!mem_cgroup_same_or_subtree(oom_wait_memcg
, wake_memcg
)
1820 && !mem_cgroup_same_or_subtree(wake_memcg
, oom_wait_memcg
))
1822 return autoremove_wake_function(wait
, mode
, sync
, arg
);
1825 static void memcg_wakeup_oom(struct mem_cgroup
*memcg
)
1827 /* for filtering, pass "memcg" as argument. */
1828 __wake_up(&memcg_oom_waitq
, TASK_NORMAL
, 0, memcg
);
1831 static void memcg_oom_recover(struct mem_cgroup
*memcg
)
1833 if (memcg
&& atomic_read(&memcg
->under_oom
))
1834 memcg_wakeup_oom(memcg
);
1838 * try to call OOM killer. returns false if we should exit memory-reclaim loop.
1840 static bool mem_cgroup_handle_oom(struct mem_cgroup
*memcg
, gfp_t mask
,
1843 struct oom_wait_info owait
;
1844 bool locked
, need_to_kill
;
1846 owait
.memcg
= memcg
;
1847 owait
.wait
.flags
= 0;
1848 owait
.wait
.func
= memcg_oom_wake_function
;
1849 owait
.wait
.private = current
;
1850 INIT_LIST_HEAD(&owait
.wait
.task_list
);
1851 need_to_kill
= true;
1852 mem_cgroup_mark_under_oom(memcg
);
1854 /* At first, try to OOM lock hierarchy under memcg.*/
1855 spin_lock(&memcg_oom_lock
);
1856 locked
= mem_cgroup_oom_lock(memcg
);
1858 * Even if signal_pending(), we can't quit charge() loop without
1859 * accounting. So, UNINTERRUPTIBLE is appropriate. But SIGKILL
1860 * under OOM is always welcomed, use TASK_KILLABLE here.
1862 prepare_to_wait(&memcg_oom_waitq
, &owait
.wait
, TASK_KILLABLE
);
1863 if (!locked
|| memcg
->oom_kill_disable
)
1864 need_to_kill
= false;
1866 mem_cgroup_oom_notify(memcg
);
1867 spin_unlock(&memcg_oom_lock
);
1870 finish_wait(&memcg_oom_waitq
, &owait
.wait
);
1871 mem_cgroup_out_of_memory(memcg
, mask
, order
);
1874 finish_wait(&memcg_oom_waitq
, &owait
.wait
);
1876 spin_lock(&memcg_oom_lock
);
1878 mem_cgroup_oom_unlock(memcg
);
1879 memcg_wakeup_oom(memcg
);
1880 spin_unlock(&memcg_oom_lock
);
1882 mem_cgroup_unmark_under_oom(memcg
);
1884 if (test_thread_flag(TIF_MEMDIE
) || fatal_signal_pending(current
))
1886 /* Give chance to dying process */
1887 schedule_timeout_uninterruptible(1);
1892 * Currently used to update mapped file statistics, but the routine can be
1893 * generalized to update other statistics as well.
1895 * Notes: Race condition
1897 * We usually use page_cgroup_lock() for accessing page_cgroup member but
1898 * it tends to be costly. But considering some conditions, we doesn't need
1899 * to do so _always_.
1901 * Considering "charge", lock_page_cgroup() is not required because all
1902 * file-stat operations happen after a page is attached to radix-tree. There
1903 * are no race with "charge".
1905 * Considering "uncharge", we know that memcg doesn't clear pc->mem_cgroup
1906 * at "uncharge" intentionally. So, we always see valid pc->mem_cgroup even
1907 * if there are race with "uncharge". Statistics itself is properly handled
1910 * Considering "move", this is an only case we see a race. To make the race
1911 * small, we check mm->moving_account and detect there are possibility of race
1912 * If there is, we take a lock.
1915 void __mem_cgroup_begin_update_page_stat(struct page
*page
,
1916 bool *locked
, unsigned long *flags
)
1918 struct mem_cgroup
*memcg
;
1919 struct page_cgroup
*pc
;
1921 pc
= lookup_page_cgroup(page
);
1923 memcg
= pc
->mem_cgroup
;
1924 if (unlikely(!memcg
|| !PageCgroupUsed(pc
)))
1927 * If this memory cgroup is not under account moving, we don't
1928 * need to take move_lock_page_cgroup(). Because we already hold
1929 * rcu_read_lock(), any calls to move_account will be delayed until
1930 * rcu_read_unlock() if mem_cgroup_stolen() == true.
1932 if (!mem_cgroup_stolen(memcg
))
1935 move_lock_mem_cgroup(memcg
, flags
);
1936 if (memcg
!= pc
->mem_cgroup
|| !PageCgroupUsed(pc
)) {
1937 move_unlock_mem_cgroup(memcg
, flags
);
1943 void __mem_cgroup_end_update_page_stat(struct page
*page
, unsigned long *flags
)
1945 struct page_cgroup
*pc
= lookup_page_cgroup(page
);
1948 * It's guaranteed that pc->mem_cgroup never changes while
1949 * lock is held because a routine modifies pc->mem_cgroup
1950 * should take move_lock_page_cgroup().
1952 move_unlock_mem_cgroup(pc
->mem_cgroup
, flags
);
1955 void mem_cgroup_update_page_stat(struct page
*page
,
1956 enum mem_cgroup_page_stat_item idx
, int val
)
1958 struct mem_cgroup
*memcg
;
1959 struct page_cgroup
*pc
= lookup_page_cgroup(page
);
1960 unsigned long uninitialized_var(flags
);
1962 if (mem_cgroup_disabled())
1965 memcg
= pc
->mem_cgroup
;
1966 if (unlikely(!memcg
|| !PageCgroupUsed(pc
)))
1970 case MEMCG_NR_FILE_MAPPED
:
1971 idx
= MEM_CGROUP_STAT_FILE_MAPPED
;
1977 this_cpu_add(memcg
->stat
->count
[idx
], val
);
1981 * size of first charge trial. "32" comes from vmscan.c's magic value.
1982 * TODO: maybe necessary to use big numbers in big irons.
1984 #define CHARGE_BATCH 32U
1985 struct memcg_stock_pcp
{
1986 struct mem_cgroup
*cached
; /* this never be root cgroup */
1987 unsigned int nr_pages
;
1988 struct work_struct work
;
1989 unsigned long flags
;
1990 #define FLUSHING_CACHED_CHARGE 0
1992 static DEFINE_PER_CPU(struct memcg_stock_pcp
, memcg_stock
);
1993 static DEFINE_MUTEX(percpu_charge_mutex
);
1996 * Try to consume stocked charge on this cpu. If success, one page is consumed
1997 * from local stock and true is returned. If the stock is 0 or charges from a
1998 * cgroup which is not current target, returns false. This stock will be
2001 static bool consume_stock(struct mem_cgroup
*memcg
)
2003 struct memcg_stock_pcp
*stock
;
2006 stock
= &get_cpu_var(memcg_stock
);
2007 if (memcg
== stock
->cached
&& stock
->nr_pages
)
2009 else /* need to call res_counter_charge */
2011 put_cpu_var(memcg_stock
);
2016 * Returns stocks cached in percpu to res_counter and reset cached information.
2018 static void drain_stock(struct memcg_stock_pcp
*stock
)
2020 struct mem_cgroup
*old
= stock
->cached
;
2022 if (stock
->nr_pages
) {
2023 unsigned long bytes
= stock
->nr_pages
* PAGE_SIZE
;
2025 res_counter_uncharge(&old
->res
, bytes
);
2026 if (do_swap_account
)
2027 res_counter_uncharge(&old
->memsw
, bytes
);
2028 stock
->nr_pages
= 0;
2030 stock
->cached
= NULL
;
2034 * This must be called under preempt disabled or must be called by
2035 * a thread which is pinned to local cpu.
2037 static void drain_local_stock(struct work_struct
*dummy
)
2039 struct memcg_stock_pcp
*stock
= &__get_cpu_var(memcg_stock
);
2041 clear_bit(FLUSHING_CACHED_CHARGE
, &stock
->flags
);
2045 * Cache charges(val) which is from res_counter, to local per_cpu area.
2046 * This will be consumed by consume_stock() function, later.
2048 static void refill_stock(struct mem_cgroup
*memcg
, unsigned int nr_pages
)
2050 struct memcg_stock_pcp
*stock
= &get_cpu_var(memcg_stock
);
2052 if (stock
->cached
!= memcg
) { /* reset if necessary */
2054 stock
->cached
= memcg
;
2056 stock
->nr_pages
+= nr_pages
;
2057 put_cpu_var(memcg_stock
);
2061 * Drains all per-CPU charge caches for given root_memcg resp. subtree
2062 * of the hierarchy under it. sync flag says whether we should block
2063 * until the work is done.
2065 static void drain_all_stock(struct mem_cgroup
*root_memcg
, bool sync
)
2069 /* Notify other cpus that system-wide "drain" is running */
2072 for_each_online_cpu(cpu
) {
2073 struct memcg_stock_pcp
*stock
= &per_cpu(memcg_stock
, cpu
);
2074 struct mem_cgroup
*memcg
;
2076 memcg
= stock
->cached
;
2077 if (!memcg
|| !stock
->nr_pages
)
2079 if (!mem_cgroup_same_or_subtree(root_memcg
, memcg
))
2081 if (!test_and_set_bit(FLUSHING_CACHED_CHARGE
, &stock
->flags
)) {
2083 drain_local_stock(&stock
->work
);
2085 schedule_work_on(cpu
, &stock
->work
);
2093 for_each_online_cpu(cpu
) {
2094 struct memcg_stock_pcp
*stock
= &per_cpu(memcg_stock
, cpu
);
2095 if (test_bit(FLUSHING_CACHED_CHARGE
, &stock
->flags
))
2096 flush_work(&stock
->work
);
2103 * Tries to drain stocked charges in other cpus. This function is asynchronous
2104 * and just put a work per cpu for draining localy on each cpu. Caller can
2105 * expects some charges will be back to res_counter later but cannot wait for
2108 static void drain_all_stock_async(struct mem_cgroup
*root_memcg
)
2111 * If someone calls draining, avoid adding more kworker runs.
2113 if (!mutex_trylock(&percpu_charge_mutex
))
2115 drain_all_stock(root_memcg
, false);
2116 mutex_unlock(&percpu_charge_mutex
);
2119 /* This is a synchronous drain interface. */
2120 static void drain_all_stock_sync(struct mem_cgroup
*root_memcg
)
2122 /* called when force_empty is called */
2123 mutex_lock(&percpu_charge_mutex
);
2124 drain_all_stock(root_memcg
, true);
2125 mutex_unlock(&percpu_charge_mutex
);
2129 * This function drains percpu counter value from DEAD cpu and
2130 * move it to local cpu. Note that this function can be preempted.
2132 static void mem_cgroup_drain_pcp_counter(struct mem_cgroup
*memcg
, int cpu
)
2136 spin_lock(&memcg
->pcp_counter_lock
);
2137 for (i
= 0; i
< MEM_CGROUP_STAT_NSTATS
; i
++) {
2138 long x
= per_cpu(memcg
->stat
->count
[i
], cpu
);
2140 per_cpu(memcg
->stat
->count
[i
], cpu
) = 0;
2141 memcg
->nocpu_base
.count
[i
] += x
;
2143 for (i
= 0; i
< MEM_CGROUP_EVENTS_NSTATS
; i
++) {
2144 unsigned long x
= per_cpu(memcg
->stat
->events
[i
], cpu
);
2146 per_cpu(memcg
->stat
->events
[i
], cpu
) = 0;
2147 memcg
->nocpu_base
.events
[i
] += x
;
2149 spin_unlock(&memcg
->pcp_counter_lock
);
2152 static int __cpuinit
memcg_cpu_hotplug_callback(struct notifier_block
*nb
,
2153 unsigned long action
,
2156 int cpu
= (unsigned long)hcpu
;
2157 struct memcg_stock_pcp
*stock
;
2158 struct mem_cgroup
*iter
;
2160 if (action
== CPU_ONLINE
)
2163 if (action
!= CPU_DEAD
&& action
!= CPU_DEAD_FROZEN
)
2166 for_each_mem_cgroup(iter
)
2167 mem_cgroup_drain_pcp_counter(iter
, cpu
);
2169 stock
= &per_cpu(memcg_stock
, cpu
);
2175 /* See __mem_cgroup_try_charge() for details */
2177 CHARGE_OK
, /* success */
2178 CHARGE_RETRY
, /* need to retry but retry is not bad */
2179 CHARGE_NOMEM
, /* we can't do more. return -ENOMEM */
2180 CHARGE_WOULDBLOCK
, /* GFP_WAIT wasn't set and no enough res. */
2181 CHARGE_OOM_DIE
, /* the current is killed because of OOM */
2184 static int mem_cgroup_do_charge(struct mem_cgroup
*memcg
, gfp_t gfp_mask
,
2185 unsigned int nr_pages
, bool oom_check
)
2187 unsigned long csize
= nr_pages
* PAGE_SIZE
;
2188 struct mem_cgroup
*mem_over_limit
;
2189 struct res_counter
*fail_res
;
2190 unsigned long flags
= 0;
2193 ret
= res_counter_charge(&memcg
->res
, csize
, &fail_res
);
2196 if (!do_swap_account
)
2198 ret
= res_counter_charge(&memcg
->memsw
, csize
, &fail_res
);
2202 res_counter_uncharge(&memcg
->res
, csize
);
2203 mem_over_limit
= mem_cgroup_from_res_counter(fail_res
, memsw
);
2204 flags
|= MEM_CGROUP_RECLAIM_NOSWAP
;
2206 mem_over_limit
= mem_cgroup_from_res_counter(fail_res
, res
);
2208 * nr_pages can be either a huge page (HPAGE_PMD_NR), a batch
2209 * of regular pages (CHARGE_BATCH), or a single regular page (1).
2211 * Never reclaim on behalf of optional batching, retry with a
2212 * single page instead.
2214 if (nr_pages
== CHARGE_BATCH
)
2215 return CHARGE_RETRY
;
2217 if (!(gfp_mask
& __GFP_WAIT
))
2218 return CHARGE_WOULDBLOCK
;
2220 ret
= mem_cgroup_reclaim(mem_over_limit
, gfp_mask
, flags
);
2221 if (mem_cgroup_margin(mem_over_limit
) >= nr_pages
)
2222 return CHARGE_RETRY
;
2224 * Even though the limit is exceeded at this point, reclaim
2225 * may have been able to free some pages. Retry the charge
2226 * before killing the task.
2228 * Only for regular pages, though: huge pages are rather
2229 * unlikely to succeed so close to the limit, and we fall back
2230 * to regular pages anyway in case of failure.
2232 if (nr_pages
== 1 && ret
)
2233 return CHARGE_RETRY
;
2236 * At task move, charge accounts can be doubly counted. So, it's
2237 * better to wait until the end of task_move if something is going on.
2239 if (mem_cgroup_wait_acct_move(mem_over_limit
))
2240 return CHARGE_RETRY
;
2242 /* If we don't need to call oom-killer at el, return immediately */
2244 return CHARGE_NOMEM
;
2246 if (!mem_cgroup_handle_oom(mem_over_limit
, gfp_mask
, get_order(csize
)))
2247 return CHARGE_OOM_DIE
;
2249 return CHARGE_RETRY
;
2253 * __mem_cgroup_try_charge() does
2254 * 1. detect memcg to be charged against from passed *mm and *ptr,
2255 * 2. update res_counter
2256 * 3. call memory reclaim if necessary.
2258 * In some special case, if the task is fatal, fatal_signal_pending() or
2259 * has TIF_MEMDIE, this function returns -EINTR while writing root_mem_cgroup
2260 * to *ptr. There are two reasons for this. 1: fatal threads should quit as soon
2261 * as possible without any hazards. 2: all pages should have a valid
2262 * pc->mem_cgroup. If mm is NULL and the caller doesn't pass a valid memcg
2263 * pointer, that is treated as a charge to root_mem_cgroup.
2265 * So __mem_cgroup_try_charge() will return
2266 * 0 ... on success, filling *ptr with a valid memcg pointer.
2267 * -ENOMEM ... charge failure because of resource limits.
2268 * -EINTR ... if thread is fatal. *ptr is filled with root_mem_cgroup.
2270 * Unlike the exported interface, an "oom" parameter is added. if oom==true,
2271 * the oom-killer can be invoked.
2273 static int __mem_cgroup_try_charge(struct mm_struct
*mm
,
2275 unsigned int nr_pages
,
2276 struct mem_cgroup
**ptr
,
2279 unsigned int batch
= max(CHARGE_BATCH
, nr_pages
);
2280 int nr_oom_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
2281 struct mem_cgroup
*memcg
= NULL
;
2285 * Unlike gloval-vm's OOM-kill, we're not in memory shortage
2286 * in system level. So, allow to go ahead dying process in addition to
2289 if (unlikely(test_thread_flag(TIF_MEMDIE
)
2290 || fatal_signal_pending(current
)))
2294 * We always charge the cgroup the mm_struct belongs to.
2295 * The mm_struct's mem_cgroup changes on task migration if the
2296 * thread group leader migrates. It's possible that mm is not
2297 * set, if so charge the init_mm (happens for pagecache usage).
2300 *ptr
= root_mem_cgroup
;
2302 if (*ptr
) { /* css should be a valid one */
2304 VM_BUG_ON(css_is_removed(&memcg
->css
));
2305 if (mem_cgroup_is_root(memcg
))
2307 if (nr_pages
== 1 && consume_stock(memcg
))
2309 css_get(&memcg
->css
);
2311 struct task_struct
*p
;
2314 p
= rcu_dereference(mm
->owner
);
2316 * Because we don't have task_lock(), "p" can exit.
2317 * In that case, "memcg" can point to root or p can be NULL with
2318 * race with swapoff. Then, we have small risk of mis-accouning.
2319 * But such kind of mis-account by race always happens because
2320 * we don't have cgroup_mutex(). It's overkill and we allo that
2322 * (*) swapoff at el will charge against mm-struct not against
2323 * task-struct. So, mm->owner can be NULL.
2325 memcg
= mem_cgroup_from_task(p
);
2327 memcg
= root_mem_cgroup
;
2328 if (mem_cgroup_is_root(memcg
)) {
2332 if (nr_pages
== 1 && consume_stock(memcg
)) {
2334 * It seems dagerous to access memcg without css_get().
2335 * But considering how consume_stok works, it's not
2336 * necessary. If consume_stock success, some charges
2337 * from this memcg are cached on this cpu. So, we
2338 * don't need to call css_get()/css_tryget() before
2339 * calling consume_stock().
2344 /* after here, we may be blocked. we need to get refcnt */
2345 if (!css_tryget(&memcg
->css
)) {
2355 /* If killed, bypass charge */
2356 if (fatal_signal_pending(current
)) {
2357 css_put(&memcg
->css
);
2362 if (oom
&& !nr_oom_retries
) {
2364 nr_oom_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
2367 ret
= mem_cgroup_do_charge(memcg
, gfp_mask
, batch
, oom_check
);
2371 case CHARGE_RETRY
: /* not in OOM situation but retry */
2373 css_put(&memcg
->css
);
2376 case CHARGE_WOULDBLOCK
: /* !__GFP_WAIT */
2377 css_put(&memcg
->css
);
2379 case CHARGE_NOMEM
: /* OOM routine works */
2381 css_put(&memcg
->css
);
2384 /* If oom, we never return -ENOMEM */
2387 case CHARGE_OOM_DIE
: /* Killed by OOM Killer */
2388 css_put(&memcg
->css
);
2391 } while (ret
!= CHARGE_OK
);
2393 if (batch
> nr_pages
)
2394 refill_stock(memcg
, batch
- nr_pages
);
2395 css_put(&memcg
->css
);
2403 *ptr
= root_mem_cgroup
;
2408 * Somemtimes we have to undo a charge we got by try_charge().
2409 * This function is for that and do uncharge, put css's refcnt.
2410 * gotten by try_charge().
2412 static void __mem_cgroup_cancel_charge(struct mem_cgroup
*memcg
,
2413 unsigned int nr_pages
)
2415 if (!mem_cgroup_is_root(memcg
)) {
2416 unsigned long bytes
= nr_pages
* PAGE_SIZE
;
2418 res_counter_uncharge(&memcg
->res
, bytes
);
2419 if (do_swap_account
)
2420 res_counter_uncharge(&memcg
->memsw
, bytes
);
2425 * Cancel chrages in this cgroup....doesn't propagate to parent cgroup.
2426 * This is useful when moving usage to parent cgroup.
2428 static void __mem_cgroup_cancel_local_charge(struct mem_cgroup
*memcg
,
2429 unsigned int nr_pages
)
2431 unsigned long bytes
= nr_pages
* PAGE_SIZE
;
2433 if (mem_cgroup_is_root(memcg
))
2436 res_counter_uncharge_until(&memcg
->res
, memcg
->res
.parent
, bytes
);
2437 if (do_swap_account
)
2438 res_counter_uncharge_until(&memcg
->memsw
,
2439 memcg
->memsw
.parent
, bytes
);
2443 * A helper function to get mem_cgroup from ID. must be called under
2444 * rcu_read_lock(). The caller must check css_is_removed() or some if
2445 * it's concern. (dropping refcnt from swap can be called against removed
2448 static struct mem_cgroup
*mem_cgroup_lookup(unsigned short id
)
2450 struct cgroup_subsys_state
*css
;
2452 /* ID 0 is unused ID */
2455 css
= css_lookup(&mem_cgroup_subsys
, id
);
2458 return container_of(css
, struct mem_cgroup
, css
);
2461 struct mem_cgroup
*try_get_mem_cgroup_from_page(struct page
*page
)
2463 struct mem_cgroup
*memcg
= NULL
;
2464 struct page_cgroup
*pc
;
2468 VM_BUG_ON(!PageLocked(page
));
2470 pc
= lookup_page_cgroup(page
);
2471 lock_page_cgroup(pc
);
2472 if (PageCgroupUsed(pc
)) {
2473 memcg
= pc
->mem_cgroup
;
2474 if (memcg
&& !css_tryget(&memcg
->css
))
2476 } else if (PageSwapCache(page
)) {
2477 ent
.val
= page_private(page
);
2478 id
= lookup_swap_cgroup_id(ent
);
2480 memcg
= mem_cgroup_lookup(id
);
2481 if (memcg
&& !css_tryget(&memcg
->css
))
2485 unlock_page_cgroup(pc
);
2489 static void __mem_cgroup_commit_charge(struct mem_cgroup
*memcg
,
2491 unsigned int nr_pages
,
2492 enum charge_type ctype
,
2495 struct page_cgroup
*pc
= lookup_page_cgroup(page
);
2496 struct zone
*uninitialized_var(zone
);
2497 bool was_on_lru
= false;
2500 lock_page_cgroup(pc
);
2501 if (unlikely(PageCgroupUsed(pc
))) {
2502 unlock_page_cgroup(pc
);
2503 __mem_cgroup_cancel_charge(memcg
, nr_pages
);
2507 * we don't need page_cgroup_lock about tail pages, becase they are not
2508 * accessed by any other context at this point.
2512 * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page
2513 * may already be on some other mem_cgroup's LRU. Take care of it.
2516 zone
= page_zone(page
);
2517 spin_lock_irq(&zone
->lru_lock
);
2518 if (PageLRU(page
)) {
2520 del_page_from_lru_list(zone
, page
, page_lru(page
));
2525 pc
->mem_cgroup
= memcg
;
2527 * We access a page_cgroup asynchronously without lock_page_cgroup().
2528 * Especially when a page_cgroup is taken from a page, pc->mem_cgroup
2529 * is accessed after testing USED bit. To make pc->mem_cgroup visible
2530 * before USED bit, we need memory barrier here.
2531 * See mem_cgroup_add_lru_list(), etc.
2534 SetPageCgroupUsed(pc
);
2538 VM_BUG_ON(PageLRU(page
));
2540 add_page_to_lru_list(zone
, page
, page_lru(page
));
2542 spin_unlock_irq(&zone
->lru_lock
);
2545 if (ctype
== MEM_CGROUP_CHARGE_TYPE_MAPPED
)
2550 mem_cgroup_charge_statistics(memcg
, anon
, nr_pages
);
2551 unlock_page_cgroup(pc
);
2554 * "charge_statistics" updated event counter. Then, check it.
2555 * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree.
2556 * if they exceeds softlimit.
2558 memcg_check_events(memcg
, page
);
2561 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
2563 #define PCGF_NOCOPY_AT_SPLIT (1 << PCG_LOCK | 1 << PCG_MIGRATION)
2565 * Because tail pages are not marked as "used", set it. We're under
2566 * zone->lru_lock, 'splitting on pmd' and compound_lock.
2567 * charge/uncharge will be never happen and move_account() is done under
2568 * compound_lock(), so we don't have to take care of races.
2570 void mem_cgroup_split_huge_fixup(struct page
*head
)
2572 struct page_cgroup
*head_pc
= lookup_page_cgroup(head
);
2573 struct page_cgroup
*pc
;
2576 if (mem_cgroup_disabled())
2578 for (i
= 1; i
< HPAGE_PMD_NR
; i
++) {
2580 pc
->mem_cgroup
= head_pc
->mem_cgroup
;
2581 smp_wmb();/* see __commit_charge() */
2582 pc
->flags
= head_pc
->flags
& ~PCGF_NOCOPY_AT_SPLIT
;
2585 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
2588 * mem_cgroup_move_account - move account of the page
2590 * @nr_pages: number of regular pages (>1 for huge pages)
2591 * @pc: page_cgroup of the page.
2592 * @from: mem_cgroup which the page is moved from.
2593 * @to: mem_cgroup which the page is moved to. @from != @to.
2595 * The caller must confirm following.
2596 * - page is not on LRU (isolate_page() is useful.)
2597 * - compound_lock is held when nr_pages > 1
2599 * This function doesn't do "charge" to new cgroup and doesn't do "uncharge"
2602 static int mem_cgroup_move_account(struct page
*page
,
2603 unsigned int nr_pages
,
2604 struct page_cgroup
*pc
,
2605 struct mem_cgroup
*from
,
2606 struct mem_cgroup
*to
)
2608 unsigned long flags
;
2610 bool anon
= PageAnon(page
);
2612 VM_BUG_ON(from
== to
);
2613 VM_BUG_ON(PageLRU(page
));
2615 * The page is isolated from LRU. So, collapse function
2616 * will not handle this page. But page splitting can happen.
2617 * Do this check under compound_page_lock(). The caller should
2621 if (nr_pages
> 1 && !PageTransHuge(page
))
2624 lock_page_cgroup(pc
);
2627 if (!PageCgroupUsed(pc
) || pc
->mem_cgroup
!= from
)
2630 move_lock_mem_cgroup(from
, &flags
);
2632 if (!anon
&& page_mapped(page
)) {
2633 /* Update mapped_file data for mem_cgroup */
2635 __this_cpu_dec(from
->stat
->count
[MEM_CGROUP_STAT_FILE_MAPPED
]);
2636 __this_cpu_inc(to
->stat
->count
[MEM_CGROUP_STAT_FILE_MAPPED
]);
2639 mem_cgroup_charge_statistics(from
, anon
, -nr_pages
);
2641 /* caller should have done css_get */
2642 pc
->mem_cgroup
= to
;
2643 mem_cgroup_charge_statistics(to
, anon
, nr_pages
);
2645 * We charges against "to" which may not have any tasks. Then, "to"
2646 * can be under rmdir(). But in current implementation, caller of
2647 * this function is just force_empty() and move charge, so it's
2648 * guaranteed that "to" is never removed. So, we don't check rmdir
2651 move_unlock_mem_cgroup(from
, &flags
);
2654 unlock_page_cgroup(pc
);
2658 memcg_check_events(to
, page
);
2659 memcg_check_events(from
, page
);
2665 * move charges to its parent.
2668 static int mem_cgroup_move_parent(struct page
*page
,
2669 struct page_cgroup
*pc
,
2670 struct mem_cgroup
*child
,
2673 struct mem_cgroup
*parent
;
2674 unsigned int nr_pages
;
2675 unsigned long uninitialized_var(flags
);
2679 if (mem_cgroup_is_root(child
))
2683 if (!get_page_unless_zero(page
))
2685 if (isolate_lru_page(page
))
2688 nr_pages
= hpage_nr_pages(page
);
2690 parent
= parent_mem_cgroup(child
);
2692 * If no parent, move charges to root cgroup.
2695 parent
= root_mem_cgroup
;
2698 flags
= compound_lock_irqsave(page
);
2700 ret
= mem_cgroup_move_account(page
, nr_pages
,
2703 __mem_cgroup_cancel_local_charge(child
, nr_pages
);
2706 compound_unlock_irqrestore(page
, flags
);
2707 putback_lru_page(page
);
2715 * Charge the memory controller for page usage.
2717 * 0 if the charge was successful
2718 * < 0 if the cgroup is over its limit
2720 static int mem_cgroup_charge_common(struct page
*page
, struct mm_struct
*mm
,
2721 gfp_t gfp_mask
, enum charge_type ctype
)
2723 struct mem_cgroup
*memcg
= NULL
;
2724 unsigned int nr_pages
= 1;
2728 if (PageTransHuge(page
)) {
2729 nr_pages
<<= compound_order(page
);
2730 VM_BUG_ON(!PageTransHuge(page
));
2732 * Never OOM-kill a process for a huge page. The
2733 * fault handler will fall back to regular pages.
2738 ret
= __mem_cgroup_try_charge(mm
, gfp_mask
, nr_pages
, &memcg
, oom
);
2741 __mem_cgroup_commit_charge(memcg
, page
, nr_pages
, ctype
, false);
2745 int mem_cgroup_newpage_charge(struct page
*page
,
2746 struct mm_struct
*mm
, gfp_t gfp_mask
)
2748 if (mem_cgroup_disabled())
2750 VM_BUG_ON(page_mapped(page
));
2751 VM_BUG_ON(page
->mapping
&& !PageAnon(page
));
2753 return mem_cgroup_charge_common(page
, mm
, gfp_mask
,
2754 MEM_CGROUP_CHARGE_TYPE_MAPPED
);
2758 __mem_cgroup_commit_charge_swapin(struct page
*page
, struct mem_cgroup
*ptr
,
2759 enum charge_type ctype
);
2761 int mem_cgroup_cache_charge(struct page
*page
, struct mm_struct
*mm
,
2764 struct mem_cgroup
*memcg
= NULL
;
2765 enum charge_type type
= MEM_CGROUP_CHARGE_TYPE_CACHE
;
2768 if (mem_cgroup_disabled())
2770 if (PageCompound(page
))
2775 if (!page_is_file_cache(page
))
2776 type
= MEM_CGROUP_CHARGE_TYPE_SHMEM
;
2778 if (!PageSwapCache(page
))
2779 ret
= mem_cgroup_charge_common(page
, mm
, gfp_mask
, type
);
2780 else { /* page is swapcache/shmem */
2781 ret
= mem_cgroup_try_charge_swapin(mm
, page
, gfp_mask
, &memcg
);
2783 __mem_cgroup_commit_charge_swapin(page
, memcg
, type
);
2789 * While swap-in, try_charge -> commit or cancel, the page is locked.
2790 * And when try_charge() successfully returns, one refcnt to memcg without
2791 * struct page_cgroup is acquired. This refcnt will be consumed by
2792 * "commit()" or removed by "cancel()"
2794 int mem_cgroup_try_charge_swapin(struct mm_struct
*mm
,
2796 gfp_t mask
, struct mem_cgroup
**memcgp
)
2798 struct mem_cgroup
*memcg
;
2803 if (mem_cgroup_disabled())
2806 if (!do_swap_account
)
2809 * A racing thread's fault, or swapoff, may have already updated
2810 * the pte, and even removed page from swap cache: in those cases
2811 * do_swap_page()'s pte_same() test will fail; but there's also a
2812 * KSM case which does need to charge the page.
2814 if (!PageSwapCache(page
))
2816 memcg
= try_get_mem_cgroup_from_page(page
);
2820 ret
= __mem_cgroup_try_charge(NULL
, mask
, 1, memcgp
, true);
2821 css_put(&memcg
->css
);
2828 ret
= __mem_cgroup_try_charge(mm
, mask
, 1, memcgp
, true);
2835 __mem_cgroup_commit_charge_swapin(struct page
*page
, struct mem_cgroup
*memcg
,
2836 enum charge_type ctype
)
2838 if (mem_cgroup_disabled())
2842 cgroup_exclude_rmdir(&memcg
->css
);
2844 __mem_cgroup_commit_charge(memcg
, page
, 1, ctype
, true);
2846 * Now swap is on-memory. This means this page may be
2847 * counted both as mem and swap....double count.
2848 * Fix it by uncharging from memsw. Basically, this SwapCache is stable
2849 * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page()
2850 * may call delete_from_swap_cache() before reach here.
2852 if (do_swap_account
&& PageSwapCache(page
)) {
2853 swp_entry_t ent
= {.val
= page_private(page
)};
2854 mem_cgroup_uncharge_swap(ent
);
2857 * At swapin, we may charge account against cgroup which has no tasks.
2858 * So, rmdir()->pre_destroy() can be called while we do this charge.
2859 * In that case, we need to call pre_destroy() again. check it here.
2861 cgroup_release_and_wakeup_rmdir(&memcg
->css
);
2864 void mem_cgroup_commit_charge_swapin(struct page
*page
,
2865 struct mem_cgroup
*memcg
)
2867 __mem_cgroup_commit_charge_swapin(page
, memcg
,
2868 MEM_CGROUP_CHARGE_TYPE_MAPPED
);
2871 void mem_cgroup_cancel_charge_swapin(struct mem_cgroup
*memcg
)
2873 if (mem_cgroup_disabled())
2877 __mem_cgroup_cancel_charge(memcg
, 1);
2880 static void mem_cgroup_do_uncharge(struct mem_cgroup
*memcg
,
2881 unsigned int nr_pages
,
2882 const enum charge_type ctype
)
2884 struct memcg_batch_info
*batch
= NULL
;
2885 bool uncharge_memsw
= true;
2887 /* If swapout, usage of swap doesn't decrease */
2888 if (!do_swap_account
|| ctype
== MEM_CGROUP_CHARGE_TYPE_SWAPOUT
)
2889 uncharge_memsw
= false;
2891 batch
= ¤t
->memcg_batch
;
2893 * In usual, we do css_get() when we remember memcg pointer.
2894 * But in this case, we keep res->usage until end of a series of
2895 * uncharges. Then, it's ok to ignore memcg's refcnt.
2898 batch
->memcg
= memcg
;
2900 * do_batch > 0 when unmapping pages or inode invalidate/truncate.
2901 * In those cases, all pages freed continuously can be expected to be in
2902 * the same cgroup and we have chance to coalesce uncharges.
2903 * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE)
2904 * because we want to do uncharge as soon as possible.
2907 if (!batch
->do_batch
|| test_thread_flag(TIF_MEMDIE
))
2908 goto direct_uncharge
;
2911 goto direct_uncharge
;
2914 * In typical case, batch->memcg == mem. This means we can
2915 * merge a series of uncharges to an uncharge of res_counter.
2916 * If not, we uncharge res_counter ony by one.
2918 if (batch
->memcg
!= memcg
)
2919 goto direct_uncharge
;
2920 /* remember freed charge and uncharge it later */
2923 batch
->memsw_nr_pages
++;
2926 res_counter_uncharge(&memcg
->res
, nr_pages
* PAGE_SIZE
);
2928 res_counter_uncharge(&memcg
->memsw
, nr_pages
* PAGE_SIZE
);
2929 if (unlikely(batch
->memcg
!= memcg
))
2930 memcg_oom_recover(memcg
);
2934 * uncharge if !page_mapped(page)
2936 static struct mem_cgroup
*
2937 __mem_cgroup_uncharge_common(struct page
*page
, enum charge_type ctype
)
2939 struct mem_cgroup
*memcg
= NULL
;
2940 unsigned int nr_pages
= 1;
2941 struct page_cgroup
*pc
;
2944 if (mem_cgroup_disabled())
2947 if (PageSwapCache(page
))
2950 if (PageTransHuge(page
)) {
2951 nr_pages
<<= compound_order(page
);
2952 VM_BUG_ON(!PageTransHuge(page
));
2955 * Check if our page_cgroup is valid
2957 pc
= lookup_page_cgroup(page
);
2958 if (unlikely(!PageCgroupUsed(pc
)))
2961 lock_page_cgroup(pc
);
2963 memcg
= pc
->mem_cgroup
;
2965 if (!PageCgroupUsed(pc
))
2968 anon
= PageAnon(page
);
2971 case MEM_CGROUP_CHARGE_TYPE_MAPPED
:
2973 * Generally PageAnon tells if it's the anon statistics to be
2974 * updated; but sometimes e.g. mem_cgroup_uncharge_page() is
2975 * used before page reached the stage of being marked PageAnon.
2979 case MEM_CGROUP_CHARGE_TYPE_DROP
:
2980 /* See mem_cgroup_prepare_migration() */
2981 if (page_mapped(page
) || PageCgroupMigration(pc
))
2984 case MEM_CGROUP_CHARGE_TYPE_SWAPOUT
:
2985 if (!PageAnon(page
)) { /* Shared memory */
2986 if (page
->mapping
&& !page_is_file_cache(page
))
2988 } else if (page_mapped(page
)) /* Anon */
2995 mem_cgroup_charge_statistics(memcg
, anon
, -nr_pages
);
2997 ClearPageCgroupUsed(pc
);
2999 * pc->mem_cgroup is not cleared here. It will be accessed when it's
3000 * freed from LRU. This is safe because uncharged page is expected not
3001 * to be reused (freed soon). Exception is SwapCache, it's handled by
3002 * special functions.
3005 unlock_page_cgroup(pc
);
3007 * even after unlock, we have memcg->res.usage here and this memcg
3008 * will never be freed.
3010 memcg_check_events(memcg
, page
);
3011 if (do_swap_account
&& ctype
== MEM_CGROUP_CHARGE_TYPE_SWAPOUT
) {
3012 mem_cgroup_swap_statistics(memcg
, true);
3013 mem_cgroup_get(memcg
);
3015 if (!mem_cgroup_is_root(memcg
))
3016 mem_cgroup_do_uncharge(memcg
, nr_pages
, ctype
);
3021 unlock_page_cgroup(pc
);
3025 void mem_cgroup_uncharge_page(struct page
*page
)
3028 if (page_mapped(page
))
3030 VM_BUG_ON(page
->mapping
&& !PageAnon(page
));
3031 __mem_cgroup_uncharge_common(page
, MEM_CGROUP_CHARGE_TYPE_MAPPED
);
3034 void mem_cgroup_uncharge_cache_page(struct page
*page
)
3036 VM_BUG_ON(page_mapped(page
));
3037 VM_BUG_ON(page
->mapping
);
3038 __mem_cgroup_uncharge_common(page
, MEM_CGROUP_CHARGE_TYPE_CACHE
);
3042 * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate.
3043 * In that cases, pages are freed continuously and we can expect pages
3044 * are in the same memcg. All these calls itself limits the number of
3045 * pages freed at once, then uncharge_start/end() is called properly.
3046 * This may be called prural(2) times in a context,
3049 void mem_cgroup_uncharge_start(void)
3051 current
->memcg_batch
.do_batch
++;
3052 /* We can do nest. */
3053 if (current
->memcg_batch
.do_batch
== 1) {
3054 current
->memcg_batch
.memcg
= NULL
;
3055 current
->memcg_batch
.nr_pages
= 0;
3056 current
->memcg_batch
.memsw_nr_pages
= 0;
3060 void mem_cgroup_uncharge_end(void)
3062 struct memcg_batch_info
*batch
= ¤t
->memcg_batch
;
3064 if (!batch
->do_batch
)
3068 if (batch
->do_batch
) /* If stacked, do nothing. */
3074 * This "batch->memcg" is valid without any css_get/put etc...
3075 * bacause we hide charges behind us.
3077 if (batch
->nr_pages
)
3078 res_counter_uncharge(&batch
->memcg
->res
,
3079 batch
->nr_pages
* PAGE_SIZE
);
3080 if (batch
->memsw_nr_pages
)
3081 res_counter_uncharge(&batch
->memcg
->memsw
,
3082 batch
->memsw_nr_pages
* PAGE_SIZE
);
3083 memcg_oom_recover(batch
->memcg
);
3084 /* forget this pointer (for sanity check) */
3085 batch
->memcg
= NULL
;
3090 * called after __delete_from_swap_cache() and drop "page" account.
3091 * memcg information is recorded to swap_cgroup of "ent"
3094 mem_cgroup_uncharge_swapcache(struct page
*page
, swp_entry_t ent
, bool swapout
)
3096 struct mem_cgroup
*memcg
;
3097 int ctype
= MEM_CGROUP_CHARGE_TYPE_SWAPOUT
;
3099 if (!swapout
) /* this was a swap cache but the swap is unused ! */
3100 ctype
= MEM_CGROUP_CHARGE_TYPE_DROP
;
3102 memcg
= __mem_cgroup_uncharge_common(page
, ctype
);
3105 * record memcg information, if swapout && memcg != NULL,
3106 * mem_cgroup_get() was called in uncharge().
3108 if (do_swap_account
&& swapout
&& memcg
)
3109 swap_cgroup_record(ent
, css_id(&memcg
->css
));
3113 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
3115 * called from swap_entry_free(). remove record in swap_cgroup and
3116 * uncharge "memsw" account.
3118 void mem_cgroup_uncharge_swap(swp_entry_t ent
)
3120 struct mem_cgroup
*memcg
;
3123 if (!do_swap_account
)
3126 id
= swap_cgroup_record(ent
, 0);
3128 memcg
= mem_cgroup_lookup(id
);
3131 * We uncharge this because swap is freed.
3132 * This memcg can be obsolete one. We avoid calling css_tryget
3134 if (!mem_cgroup_is_root(memcg
))
3135 res_counter_uncharge(&memcg
->memsw
, PAGE_SIZE
);
3136 mem_cgroup_swap_statistics(memcg
, false);
3137 mem_cgroup_put(memcg
);
3143 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
3144 * @entry: swap entry to be moved
3145 * @from: mem_cgroup which the entry is moved from
3146 * @to: mem_cgroup which the entry is moved to
3148 * It succeeds only when the swap_cgroup's record for this entry is the same
3149 * as the mem_cgroup's id of @from.
3151 * Returns 0 on success, -EINVAL on failure.
3153 * The caller must have charged to @to, IOW, called res_counter_charge() about
3154 * both res and memsw, and called css_get().
3156 static int mem_cgroup_move_swap_account(swp_entry_t entry
,
3157 struct mem_cgroup
*from
, struct mem_cgroup
*to
)
3159 unsigned short old_id
, new_id
;
3161 old_id
= css_id(&from
->css
);
3162 new_id
= css_id(&to
->css
);
3164 if (swap_cgroup_cmpxchg(entry
, old_id
, new_id
) == old_id
) {
3165 mem_cgroup_swap_statistics(from
, false);
3166 mem_cgroup_swap_statistics(to
, true);
3168 * This function is only called from task migration context now.
3169 * It postpones res_counter and refcount handling till the end
3170 * of task migration(mem_cgroup_clear_mc()) for performance
3171 * improvement. But we cannot postpone mem_cgroup_get(to)
3172 * because if the process that has been moved to @to does
3173 * swap-in, the refcount of @to might be decreased to 0.
3181 static inline int mem_cgroup_move_swap_account(swp_entry_t entry
,
3182 struct mem_cgroup
*from
, struct mem_cgroup
*to
)
3189 * Before starting migration, account PAGE_SIZE to mem_cgroup that the old
3192 int mem_cgroup_prepare_migration(struct page
*page
,
3193 struct page
*newpage
, struct mem_cgroup
**memcgp
, gfp_t gfp_mask
)
3195 struct mem_cgroup
*memcg
= NULL
;
3196 struct page_cgroup
*pc
;
3197 enum charge_type ctype
;
3202 VM_BUG_ON(PageTransHuge(page
));
3203 if (mem_cgroup_disabled())
3206 pc
= lookup_page_cgroup(page
);
3207 lock_page_cgroup(pc
);
3208 if (PageCgroupUsed(pc
)) {
3209 memcg
= pc
->mem_cgroup
;
3210 css_get(&memcg
->css
);
3212 * At migrating an anonymous page, its mapcount goes down
3213 * to 0 and uncharge() will be called. But, even if it's fully
3214 * unmapped, migration may fail and this page has to be
3215 * charged again. We set MIGRATION flag here and delay uncharge
3216 * until end_migration() is called
3218 * Corner Case Thinking
3220 * When the old page was mapped as Anon and it's unmap-and-freed
3221 * while migration was ongoing.
3222 * If unmap finds the old page, uncharge() of it will be delayed
3223 * until end_migration(). If unmap finds a new page, it's
3224 * uncharged when it make mapcount to be 1->0. If unmap code
3225 * finds swap_migration_entry, the new page will not be mapped
3226 * and end_migration() will find it(mapcount==0).
3229 * When the old page was mapped but migraion fails, the kernel
3230 * remaps it. A charge for it is kept by MIGRATION flag even
3231 * if mapcount goes down to 0. We can do remap successfully
3232 * without charging it again.
3235 * The "old" page is under lock_page() until the end of
3236 * migration, so, the old page itself will not be swapped-out.
3237 * If the new page is swapped out before end_migraton, our
3238 * hook to usual swap-out path will catch the event.
3241 SetPageCgroupMigration(pc
);
3243 unlock_page_cgroup(pc
);
3245 * If the page is not charged at this point,
3252 ret
= __mem_cgroup_try_charge(NULL
, gfp_mask
, 1, memcgp
, false);
3253 css_put(&memcg
->css
);/* drop extra refcnt */
3255 if (PageAnon(page
)) {
3256 lock_page_cgroup(pc
);
3257 ClearPageCgroupMigration(pc
);
3258 unlock_page_cgroup(pc
);
3260 * The old page may be fully unmapped while we kept it.
3262 mem_cgroup_uncharge_page(page
);
3264 /* we'll need to revisit this error code (we have -EINTR) */
3268 * We charge new page before it's used/mapped. So, even if unlock_page()
3269 * is called before end_migration, we can catch all events on this new
3270 * page. In the case new page is migrated but not remapped, new page's
3271 * mapcount will be finally 0 and we call uncharge in end_migration().
3274 ctype
= MEM_CGROUP_CHARGE_TYPE_MAPPED
;
3275 else if (page_is_file_cache(page
))
3276 ctype
= MEM_CGROUP_CHARGE_TYPE_CACHE
;
3278 ctype
= MEM_CGROUP_CHARGE_TYPE_SHMEM
;
3279 __mem_cgroup_commit_charge(memcg
, newpage
, 1, ctype
, false);
3283 /* remove redundant charge if migration failed*/
3284 void mem_cgroup_end_migration(struct mem_cgroup
*memcg
,
3285 struct page
*oldpage
, struct page
*newpage
, bool migration_ok
)
3287 struct page
*used
, *unused
;
3288 struct page_cgroup
*pc
;
3293 /* blocks rmdir() */
3294 cgroup_exclude_rmdir(&memcg
->css
);
3295 if (!migration_ok
) {
3303 * We disallowed uncharge of pages under migration because mapcount
3304 * of the page goes down to zero, temporarly.
3305 * Clear the flag and check the page should be charged.
3307 pc
= lookup_page_cgroup(oldpage
);
3308 lock_page_cgroup(pc
);
3309 ClearPageCgroupMigration(pc
);
3310 unlock_page_cgroup(pc
);
3311 anon
= PageAnon(used
);
3312 __mem_cgroup_uncharge_common(unused
,
3313 anon
? MEM_CGROUP_CHARGE_TYPE_MAPPED
3314 : MEM_CGROUP_CHARGE_TYPE_CACHE
);
3317 * If a page is a file cache, radix-tree replacement is very atomic
3318 * and we can skip this check. When it was an Anon page, its mapcount
3319 * goes down to 0. But because we added MIGRATION flage, it's not
3320 * uncharged yet. There are several case but page->mapcount check
3321 * and USED bit check in mem_cgroup_uncharge_page() will do enough
3322 * check. (see prepare_charge() also)
3325 mem_cgroup_uncharge_page(used
);
3327 * At migration, we may charge account against cgroup which has no
3329 * So, rmdir()->pre_destroy() can be called while we do this charge.
3330 * In that case, we need to call pre_destroy() again. check it here.
3332 cgroup_release_and_wakeup_rmdir(&memcg
->css
);
3336 * At replace page cache, newpage is not under any memcg but it's on
3337 * LRU. So, this function doesn't touch res_counter but handles LRU
3338 * in correct way. Both pages are locked so we cannot race with uncharge.
3340 void mem_cgroup_replace_page_cache(struct page
*oldpage
,
3341 struct page
*newpage
)
3343 struct mem_cgroup
*memcg
= NULL
;
3344 struct page_cgroup
*pc
;
3345 enum charge_type type
= MEM_CGROUP_CHARGE_TYPE_CACHE
;
3347 if (mem_cgroup_disabled())
3350 pc
= lookup_page_cgroup(oldpage
);
3351 /* fix accounting on old pages */
3352 lock_page_cgroup(pc
);
3353 if (PageCgroupUsed(pc
)) {
3354 memcg
= pc
->mem_cgroup
;
3355 mem_cgroup_charge_statistics(memcg
, false, -1);
3356 ClearPageCgroupUsed(pc
);
3358 unlock_page_cgroup(pc
);
3361 * When called from shmem_replace_page(), in some cases the
3362 * oldpage has already been charged, and in some cases not.
3367 if (PageSwapBacked(oldpage
))
3368 type
= MEM_CGROUP_CHARGE_TYPE_SHMEM
;
3371 * Even if newpage->mapping was NULL before starting replacement,
3372 * the newpage may be on LRU(or pagevec for LRU) already. We lock
3373 * LRU while we overwrite pc->mem_cgroup.
3375 __mem_cgroup_commit_charge(memcg
, newpage
, 1, type
, true);
3378 #ifdef CONFIG_DEBUG_VM
3379 static struct page_cgroup
*lookup_page_cgroup_used(struct page
*page
)
3381 struct page_cgroup
*pc
;
3383 pc
= lookup_page_cgroup(page
);
3385 * Can be NULL while feeding pages into the page allocator for
3386 * the first time, i.e. during boot or memory hotplug;
3387 * or when mem_cgroup_disabled().
3389 if (likely(pc
) && PageCgroupUsed(pc
))
3394 bool mem_cgroup_bad_page_check(struct page
*page
)
3396 if (mem_cgroup_disabled())
3399 return lookup_page_cgroup_used(page
) != NULL
;
3402 void mem_cgroup_print_bad_page(struct page
*page
)
3404 struct page_cgroup
*pc
;
3406 pc
= lookup_page_cgroup_used(page
);
3408 printk(KERN_ALERT
"pc:%p pc->flags:%lx pc->mem_cgroup:%p\n",
3409 pc
, pc
->flags
, pc
->mem_cgroup
);
3414 static DEFINE_MUTEX(set_limit_mutex
);
3416 static int mem_cgroup_resize_limit(struct mem_cgroup
*memcg
,
3417 unsigned long long val
)
3420 u64 memswlimit
, memlimit
;
3422 int children
= mem_cgroup_count_children(memcg
);
3423 u64 curusage
, oldusage
;
3427 * For keeping hierarchical_reclaim simple, how long we should retry
3428 * is depends on callers. We set our retry-count to be function
3429 * of # of children which we should visit in this loop.
3431 retry_count
= MEM_CGROUP_RECLAIM_RETRIES
* children
;
3433 oldusage
= res_counter_read_u64(&memcg
->res
, RES_USAGE
);
3436 while (retry_count
) {
3437 if (signal_pending(current
)) {
3442 * Rather than hide all in some function, I do this in
3443 * open coded manner. You see what this really does.
3444 * We have to guarantee memcg->res.limit < memcg->memsw.limit.
3446 mutex_lock(&set_limit_mutex
);
3447 memswlimit
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
3448 if (memswlimit
< val
) {
3450 mutex_unlock(&set_limit_mutex
);
3454 memlimit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
3458 ret
= res_counter_set_limit(&memcg
->res
, val
);
3460 if (memswlimit
== val
)
3461 memcg
->memsw_is_minimum
= true;
3463 memcg
->memsw_is_minimum
= false;
3465 mutex_unlock(&set_limit_mutex
);
3470 mem_cgroup_reclaim(memcg
, GFP_KERNEL
,
3471 MEM_CGROUP_RECLAIM_SHRINK
);
3472 curusage
= res_counter_read_u64(&memcg
->res
, RES_USAGE
);
3473 /* Usage is reduced ? */
3474 if (curusage
>= oldusage
)
3477 oldusage
= curusage
;
3479 if (!ret
&& enlarge
)
3480 memcg_oom_recover(memcg
);
3485 static int mem_cgroup_resize_memsw_limit(struct mem_cgroup
*memcg
,
3486 unsigned long long val
)
3489 u64 memlimit
, memswlimit
, oldusage
, curusage
;
3490 int children
= mem_cgroup_count_children(memcg
);
3494 /* see mem_cgroup_resize_res_limit */
3495 retry_count
= children
* MEM_CGROUP_RECLAIM_RETRIES
;
3496 oldusage
= res_counter_read_u64(&memcg
->memsw
, RES_USAGE
);
3497 while (retry_count
) {
3498 if (signal_pending(current
)) {
3503 * Rather than hide all in some function, I do this in
3504 * open coded manner. You see what this really does.
3505 * We have to guarantee memcg->res.limit < memcg->memsw.limit.
3507 mutex_lock(&set_limit_mutex
);
3508 memlimit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
3509 if (memlimit
> val
) {
3511 mutex_unlock(&set_limit_mutex
);
3514 memswlimit
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
3515 if (memswlimit
< val
)
3517 ret
= res_counter_set_limit(&memcg
->memsw
, val
);
3519 if (memlimit
== val
)
3520 memcg
->memsw_is_minimum
= true;
3522 memcg
->memsw_is_minimum
= false;
3524 mutex_unlock(&set_limit_mutex
);
3529 mem_cgroup_reclaim(memcg
, GFP_KERNEL
,
3530 MEM_CGROUP_RECLAIM_NOSWAP
|
3531 MEM_CGROUP_RECLAIM_SHRINK
);
3532 curusage
= res_counter_read_u64(&memcg
->memsw
, RES_USAGE
);
3533 /* Usage is reduced ? */
3534 if (curusage
>= oldusage
)
3537 oldusage
= curusage
;
3539 if (!ret
&& enlarge
)
3540 memcg_oom_recover(memcg
);
3544 unsigned long mem_cgroup_soft_limit_reclaim(struct zone
*zone
, int order
,
3546 unsigned long *total_scanned
)
3548 unsigned long nr_reclaimed
= 0;
3549 struct mem_cgroup_per_zone
*mz
, *next_mz
= NULL
;
3550 unsigned long reclaimed
;
3552 struct mem_cgroup_tree_per_zone
*mctz
;
3553 unsigned long long excess
;
3554 unsigned long nr_scanned
;
3559 mctz
= soft_limit_tree_node_zone(zone_to_nid(zone
), zone_idx(zone
));
3561 * This loop can run a while, specially if mem_cgroup's continuously
3562 * keep exceeding their soft limit and putting the system under
3569 mz
= mem_cgroup_largest_soft_limit_node(mctz
);
3574 reclaimed
= mem_cgroup_soft_reclaim(mz
->memcg
, zone
,
3575 gfp_mask
, &nr_scanned
);
3576 nr_reclaimed
+= reclaimed
;
3577 *total_scanned
+= nr_scanned
;
3578 spin_lock(&mctz
->lock
);
3581 * If we failed to reclaim anything from this memory cgroup
3582 * it is time to move on to the next cgroup
3588 * Loop until we find yet another one.
3590 * By the time we get the soft_limit lock
3591 * again, someone might have aded the
3592 * group back on the RB tree. Iterate to
3593 * make sure we get a different mem.
3594 * mem_cgroup_largest_soft_limit_node returns
3595 * NULL if no other cgroup is present on
3599 __mem_cgroup_largest_soft_limit_node(mctz
);
3601 css_put(&next_mz
->memcg
->css
);
3602 else /* next_mz == NULL or other memcg */
3606 __mem_cgroup_remove_exceeded(mz
->memcg
, mz
, mctz
);
3607 excess
= res_counter_soft_limit_excess(&mz
->memcg
->res
);
3609 * One school of thought says that we should not add
3610 * back the node to the tree if reclaim returns 0.
3611 * But our reclaim could return 0, simply because due
3612 * to priority we are exposing a smaller subset of
3613 * memory to reclaim from. Consider this as a longer
3616 /* If excess == 0, no tree ops */
3617 __mem_cgroup_insert_exceeded(mz
->memcg
, mz
, mctz
, excess
);
3618 spin_unlock(&mctz
->lock
);
3619 css_put(&mz
->memcg
->css
);
3622 * Could not reclaim anything and there are no more
3623 * mem cgroups to try or we seem to be looping without
3624 * reclaiming anything.
3626 if (!nr_reclaimed
&&
3628 loop
> MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS
))
3630 } while (!nr_reclaimed
);
3632 css_put(&next_mz
->memcg
->css
);
3633 return nr_reclaimed
;
3637 * This routine traverse page_cgroup in given list and drop them all.
3638 * *And* this routine doesn't reclaim page itself, just removes page_cgroup.
3640 static int mem_cgroup_force_empty_list(struct mem_cgroup
*memcg
,
3641 int node
, int zid
, enum lru_list lru
)
3643 struct mem_cgroup_per_zone
*mz
;
3644 unsigned long flags
, loop
;
3645 struct list_head
*list
;
3650 zone
= &NODE_DATA(node
)->node_zones
[zid
];
3651 mz
= mem_cgroup_zoneinfo(memcg
, node
, zid
);
3652 list
= &mz
->lruvec
.lists
[lru
];
3654 loop
= mz
->lru_size
[lru
];
3655 /* give some margin against EBUSY etc...*/
3659 struct page_cgroup
*pc
;
3663 spin_lock_irqsave(&zone
->lru_lock
, flags
);
3664 if (list_empty(list
)) {
3665 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
3668 page
= list_entry(list
->prev
, struct page
, lru
);
3670 list_move(&page
->lru
, list
);
3672 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
3675 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
3677 pc
= lookup_page_cgroup(page
);
3679 ret
= mem_cgroup_move_parent(page
, pc
, memcg
, GFP_KERNEL
);
3680 if (ret
== -ENOMEM
|| ret
== -EINTR
)
3683 if (ret
== -EBUSY
|| ret
== -EINVAL
) {
3684 /* found lock contention or "pc" is obsolete. */
3691 if (!ret
&& !list_empty(list
))
3697 * make mem_cgroup's charge to be 0 if there is no task.
3698 * This enables deleting this mem_cgroup.
3700 static int mem_cgroup_force_empty(struct mem_cgroup
*memcg
, bool free_all
)
3703 int node
, zid
, shrink
;
3704 int nr_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
3705 struct cgroup
*cgrp
= memcg
->css
.cgroup
;
3707 css_get(&memcg
->css
);
3710 /* should free all ? */
3716 if (cgroup_task_count(cgrp
) || !list_empty(&cgrp
->children
))
3719 if (signal_pending(current
))
3721 /* This is for making all *used* pages to be on LRU. */
3722 lru_add_drain_all();
3723 drain_all_stock_sync(memcg
);
3725 mem_cgroup_start_move(memcg
);
3726 for_each_node_state(node
, N_HIGH_MEMORY
) {
3727 for (zid
= 0; !ret
&& zid
< MAX_NR_ZONES
; zid
++) {
3730 ret
= mem_cgroup_force_empty_list(memcg
,
3739 mem_cgroup_end_move(memcg
);
3740 memcg_oom_recover(memcg
);
3741 /* it seems parent cgroup doesn't have enough mem */
3745 /* "ret" should also be checked to ensure all lists are empty. */
3746 } while (res_counter_read_u64(&memcg
->res
, RES_USAGE
) > 0 || ret
);
3748 css_put(&memcg
->css
);
3752 /* returns EBUSY if there is a task or if we come here twice. */
3753 if (cgroup_task_count(cgrp
) || !list_empty(&cgrp
->children
) || shrink
) {
3757 /* we call try-to-free pages for make this cgroup empty */
3758 lru_add_drain_all();
3759 /* try to free all pages in this cgroup */
3761 while (nr_retries
&& res_counter_read_u64(&memcg
->res
, RES_USAGE
) > 0) {
3764 if (signal_pending(current
)) {
3768 progress
= try_to_free_mem_cgroup_pages(memcg
, GFP_KERNEL
,
3772 /* maybe some writeback is necessary */
3773 congestion_wait(BLK_RW_ASYNC
, HZ
/10);
3778 /* try move_account...there may be some *locked* pages. */
3782 static int mem_cgroup_force_empty_write(struct cgroup
*cont
, unsigned int event
)
3784 return mem_cgroup_force_empty(mem_cgroup_from_cont(cont
), true);
3788 static u64
mem_cgroup_hierarchy_read(struct cgroup
*cont
, struct cftype
*cft
)
3790 return mem_cgroup_from_cont(cont
)->use_hierarchy
;
3793 static int mem_cgroup_hierarchy_write(struct cgroup
*cont
, struct cftype
*cft
,
3797 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
3798 struct cgroup
*parent
= cont
->parent
;
3799 struct mem_cgroup
*parent_memcg
= NULL
;
3802 parent_memcg
= mem_cgroup_from_cont(parent
);
3806 * If parent's use_hierarchy is set, we can't make any modifications
3807 * in the child subtrees. If it is unset, then the change can
3808 * occur, provided the current cgroup has no children.
3810 * For the root cgroup, parent_mem is NULL, we allow value to be
3811 * set if there are no children.
3813 if ((!parent_memcg
|| !parent_memcg
->use_hierarchy
) &&
3814 (val
== 1 || val
== 0)) {
3815 if (list_empty(&cont
->children
))
3816 memcg
->use_hierarchy
= val
;
3827 static unsigned long mem_cgroup_recursive_stat(struct mem_cgroup
*memcg
,
3828 enum mem_cgroup_stat_index idx
)
3830 struct mem_cgroup
*iter
;
3833 /* Per-cpu values can be negative, use a signed accumulator */
3834 for_each_mem_cgroup_tree(iter
, memcg
)
3835 val
+= mem_cgroup_read_stat(iter
, idx
);
3837 if (val
< 0) /* race ? */
3842 static inline u64
mem_cgroup_usage(struct mem_cgroup
*memcg
, bool swap
)
3846 if (!mem_cgroup_is_root(memcg
)) {
3848 return res_counter_read_u64(&memcg
->res
, RES_USAGE
);
3850 return res_counter_read_u64(&memcg
->memsw
, RES_USAGE
);
3853 val
= mem_cgroup_recursive_stat(memcg
, MEM_CGROUP_STAT_CACHE
);
3854 val
+= mem_cgroup_recursive_stat(memcg
, MEM_CGROUP_STAT_RSS
);
3857 val
+= mem_cgroup_recursive_stat(memcg
, MEM_CGROUP_STAT_SWAPOUT
);
3859 return val
<< PAGE_SHIFT
;
3862 static ssize_t
mem_cgroup_read(struct cgroup
*cont
, struct cftype
*cft
,
3863 struct file
*file
, char __user
*buf
,
3864 size_t nbytes
, loff_t
*ppos
)
3866 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
3869 int type
, name
, len
;
3871 type
= MEMFILE_TYPE(cft
->private);
3872 name
= MEMFILE_ATTR(cft
->private);
3874 if (!do_swap_account
&& type
== _MEMSWAP
)
3879 if (name
== RES_USAGE
)
3880 val
= mem_cgroup_usage(memcg
, false);
3882 val
= res_counter_read_u64(&memcg
->res
, name
);
3885 if (name
== RES_USAGE
)
3886 val
= mem_cgroup_usage(memcg
, true);
3888 val
= res_counter_read_u64(&memcg
->memsw
, name
);
3894 len
= scnprintf(str
, sizeof(str
), "%llu\n", (unsigned long long)val
);
3895 return simple_read_from_buffer(buf
, nbytes
, ppos
, str
, len
);
3898 * The user of this function is...
3901 static int mem_cgroup_write(struct cgroup
*cont
, struct cftype
*cft
,
3904 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
3906 unsigned long long val
;
3909 type
= MEMFILE_TYPE(cft
->private);
3910 name
= MEMFILE_ATTR(cft
->private);
3912 if (!do_swap_account
&& type
== _MEMSWAP
)
3917 if (mem_cgroup_is_root(memcg
)) { /* Can't set limit on root */
3921 /* This function does all necessary parse...reuse it */
3922 ret
= res_counter_memparse_write_strategy(buffer
, &val
);
3926 ret
= mem_cgroup_resize_limit(memcg
, val
);
3928 ret
= mem_cgroup_resize_memsw_limit(memcg
, val
);
3930 case RES_SOFT_LIMIT
:
3931 ret
= res_counter_memparse_write_strategy(buffer
, &val
);
3935 * For memsw, soft limits are hard to implement in terms
3936 * of semantics, for now, we support soft limits for
3937 * control without swap
3940 ret
= res_counter_set_soft_limit(&memcg
->res
, val
);
3945 ret
= -EINVAL
; /* should be BUG() ? */
3951 static void memcg_get_hierarchical_limit(struct mem_cgroup
*memcg
,
3952 unsigned long long *mem_limit
, unsigned long long *memsw_limit
)
3954 struct cgroup
*cgroup
;
3955 unsigned long long min_limit
, min_memsw_limit
, tmp
;
3957 min_limit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
3958 min_memsw_limit
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
3959 cgroup
= memcg
->css
.cgroup
;
3960 if (!memcg
->use_hierarchy
)
3963 while (cgroup
->parent
) {
3964 cgroup
= cgroup
->parent
;
3965 memcg
= mem_cgroup_from_cont(cgroup
);
3966 if (!memcg
->use_hierarchy
)
3968 tmp
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
3969 min_limit
= min(min_limit
, tmp
);
3970 tmp
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
3971 min_memsw_limit
= min(min_memsw_limit
, tmp
);
3974 *mem_limit
= min_limit
;
3975 *memsw_limit
= min_memsw_limit
;
3978 static int mem_cgroup_reset(struct cgroup
*cont
, unsigned int event
)
3980 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
3983 type
= MEMFILE_TYPE(event
);
3984 name
= MEMFILE_ATTR(event
);
3986 if (!do_swap_account
&& type
== _MEMSWAP
)
3992 res_counter_reset_max(&memcg
->res
);
3994 res_counter_reset_max(&memcg
->memsw
);
3998 res_counter_reset_failcnt(&memcg
->res
);
4000 res_counter_reset_failcnt(&memcg
->memsw
);
4007 static u64
mem_cgroup_move_charge_read(struct cgroup
*cgrp
,
4010 return mem_cgroup_from_cont(cgrp
)->move_charge_at_immigrate
;
4014 static int mem_cgroup_move_charge_write(struct cgroup
*cgrp
,
4015 struct cftype
*cft
, u64 val
)
4017 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4019 if (val
>= (1 << NR_MOVE_TYPE
))
4022 * We check this value several times in both in can_attach() and
4023 * attach(), so we need cgroup lock to prevent this value from being
4027 memcg
->move_charge_at_immigrate
= val
;
4033 static int mem_cgroup_move_charge_write(struct cgroup
*cgrp
,
4034 struct cftype
*cft
, u64 val
)
4041 /* For read statistics */
4059 struct mcs_total_stat
{
4060 s64 stat
[NR_MCS_STAT
];
4063 static const char *memcg_stat_strings
[NR_MCS_STAT
] = {
4080 mem_cgroup_get_local_stat(struct mem_cgroup
*memcg
, struct mcs_total_stat
*s
)
4085 val
= mem_cgroup_read_stat(memcg
, MEM_CGROUP_STAT_CACHE
);
4086 s
->stat
[MCS_CACHE
] += val
* PAGE_SIZE
;
4087 val
= mem_cgroup_read_stat(memcg
, MEM_CGROUP_STAT_RSS
);
4088 s
->stat
[MCS_RSS
] += val
* PAGE_SIZE
;
4089 val
= mem_cgroup_read_stat(memcg
, MEM_CGROUP_STAT_FILE_MAPPED
);
4090 s
->stat
[MCS_FILE_MAPPED
] += val
* PAGE_SIZE
;
4091 val
= mem_cgroup_read_events(memcg
, MEM_CGROUP_EVENTS_PGPGIN
);
4092 s
->stat
[MCS_PGPGIN
] += val
;
4093 val
= mem_cgroup_read_events(memcg
, MEM_CGROUP_EVENTS_PGPGOUT
);
4094 s
->stat
[MCS_PGPGOUT
] += val
;
4095 if (do_swap_account
) {
4096 val
= mem_cgroup_read_stat(memcg
, MEM_CGROUP_STAT_SWAPOUT
);
4097 s
->stat
[MCS_SWAP
] += val
* PAGE_SIZE
;
4099 val
= mem_cgroup_read_events(memcg
, MEM_CGROUP_EVENTS_PGFAULT
);
4100 s
->stat
[MCS_PGFAULT
] += val
;
4101 val
= mem_cgroup_read_events(memcg
, MEM_CGROUP_EVENTS_PGMAJFAULT
);
4102 s
->stat
[MCS_PGMAJFAULT
] += val
;
4105 val
= mem_cgroup_nr_lru_pages(memcg
, BIT(LRU_INACTIVE_ANON
));
4106 s
->stat
[MCS_INACTIVE_ANON
] += val
* PAGE_SIZE
;
4107 val
= mem_cgroup_nr_lru_pages(memcg
, BIT(LRU_ACTIVE_ANON
));
4108 s
->stat
[MCS_ACTIVE_ANON
] += val
* PAGE_SIZE
;
4109 val
= mem_cgroup_nr_lru_pages(memcg
, BIT(LRU_INACTIVE_FILE
));
4110 s
->stat
[MCS_INACTIVE_FILE
] += val
* PAGE_SIZE
;
4111 val
= mem_cgroup_nr_lru_pages(memcg
, BIT(LRU_ACTIVE_FILE
));
4112 s
->stat
[MCS_ACTIVE_FILE
] += val
* PAGE_SIZE
;
4113 val
= mem_cgroup_nr_lru_pages(memcg
, BIT(LRU_UNEVICTABLE
));
4114 s
->stat
[MCS_UNEVICTABLE
] += val
* PAGE_SIZE
;
4118 mem_cgroup_get_total_stat(struct mem_cgroup
*memcg
, struct mcs_total_stat
*s
)
4120 struct mem_cgroup
*iter
;
4122 for_each_mem_cgroup_tree(iter
, memcg
)
4123 mem_cgroup_get_local_stat(iter
, s
);
4127 static int mem_control_numa_stat_show(struct cgroup
*cont
, struct cftype
*cft
,
4131 unsigned long total_nr
, file_nr
, anon_nr
, unevictable_nr
;
4132 unsigned long node_nr
;
4133 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
4135 total_nr
= mem_cgroup_nr_lru_pages(memcg
, LRU_ALL
);
4136 seq_printf(m
, "total=%lu", total_nr
);
4137 for_each_node_state(nid
, N_HIGH_MEMORY
) {
4138 node_nr
= mem_cgroup_node_nr_lru_pages(memcg
, nid
, LRU_ALL
);
4139 seq_printf(m
, " N%d=%lu", nid
, node_nr
);
4143 file_nr
= mem_cgroup_nr_lru_pages(memcg
, LRU_ALL_FILE
);
4144 seq_printf(m
, "file=%lu", file_nr
);
4145 for_each_node_state(nid
, N_HIGH_MEMORY
) {
4146 node_nr
= mem_cgroup_node_nr_lru_pages(memcg
, nid
,
4148 seq_printf(m
, " N%d=%lu", nid
, node_nr
);
4152 anon_nr
= mem_cgroup_nr_lru_pages(memcg
, LRU_ALL_ANON
);
4153 seq_printf(m
, "anon=%lu", anon_nr
);
4154 for_each_node_state(nid
, N_HIGH_MEMORY
) {
4155 node_nr
= mem_cgroup_node_nr_lru_pages(memcg
, nid
,
4157 seq_printf(m
, " N%d=%lu", nid
, node_nr
);
4161 unevictable_nr
= mem_cgroup_nr_lru_pages(memcg
, BIT(LRU_UNEVICTABLE
));
4162 seq_printf(m
, "unevictable=%lu", unevictable_nr
);
4163 for_each_node_state(nid
, N_HIGH_MEMORY
) {
4164 node_nr
= mem_cgroup_node_nr_lru_pages(memcg
, nid
,
4165 BIT(LRU_UNEVICTABLE
));
4166 seq_printf(m
, " N%d=%lu", nid
, node_nr
);
4171 #endif /* CONFIG_NUMA */
4173 static int mem_control_stat_show(struct cgroup
*cont
, struct cftype
*cft
,
4176 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
4177 struct mcs_total_stat mystat
;
4180 memset(&mystat
, 0, sizeof(mystat
));
4181 mem_cgroup_get_local_stat(memcg
, &mystat
);
4184 for (i
= 0; i
< NR_MCS_STAT
; i
++) {
4185 if (i
== MCS_SWAP
&& !do_swap_account
)
4187 seq_printf(m
, "%s %llu\n", memcg_stat_strings
[i
],
4188 (unsigned long long)mystat
.stat
[i
]);
4191 /* Hierarchical information */
4193 unsigned long long limit
, memsw_limit
;
4194 memcg_get_hierarchical_limit(memcg
, &limit
, &memsw_limit
);
4195 seq_printf(m
, "hierarchical_memory_limit %llu\n", limit
);
4196 if (do_swap_account
)
4197 seq_printf(m
, "hierarchical_memsw_limit %llu\n",
4201 memset(&mystat
, 0, sizeof(mystat
));
4202 mem_cgroup_get_total_stat(memcg
, &mystat
);
4203 for (i
= 0; i
< NR_MCS_STAT
; i
++) {
4204 if (i
== MCS_SWAP
&& !do_swap_account
)
4206 seq_printf(m
, "total_%s %llu\n", memcg_stat_strings
[i
],
4207 (unsigned long long)mystat
.stat
[i
]);
4210 #ifdef CONFIG_DEBUG_VM
4213 struct mem_cgroup_per_zone
*mz
;
4214 struct zone_reclaim_stat
*rstat
;
4215 unsigned long recent_rotated
[2] = {0, 0};
4216 unsigned long recent_scanned
[2] = {0, 0};
4218 for_each_online_node(nid
)
4219 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++) {
4220 mz
= mem_cgroup_zoneinfo(memcg
, nid
, zid
);
4221 rstat
= &mz
->lruvec
.reclaim_stat
;
4223 recent_rotated
[0] += rstat
->recent_rotated
[0];
4224 recent_rotated
[1] += rstat
->recent_rotated
[1];
4225 recent_scanned
[0] += rstat
->recent_scanned
[0];
4226 recent_scanned
[1] += rstat
->recent_scanned
[1];
4228 seq_printf(m
, "recent_rotated_anon %lu\n", recent_rotated
[0]);
4229 seq_printf(m
, "recent_rotated_file %lu\n", recent_rotated
[1]);
4230 seq_printf(m
, "recent_scanned_anon %lu\n", recent_scanned
[0]);
4231 seq_printf(m
, "recent_scanned_file %lu\n", recent_scanned
[1]);
4238 static u64
mem_cgroup_swappiness_read(struct cgroup
*cgrp
, struct cftype
*cft
)
4240 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4242 return mem_cgroup_swappiness(memcg
);
4245 static int mem_cgroup_swappiness_write(struct cgroup
*cgrp
, struct cftype
*cft
,
4248 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4249 struct mem_cgroup
*parent
;
4254 if (cgrp
->parent
== NULL
)
4257 parent
= mem_cgroup_from_cont(cgrp
->parent
);
4261 /* If under hierarchy, only empty-root can set this value */
4262 if ((parent
->use_hierarchy
) ||
4263 (memcg
->use_hierarchy
&& !list_empty(&cgrp
->children
))) {
4268 memcg
->swappiness
= val
;
4275 static void __mem_cgroup_threshold(struct mem_cgroup
*memcg
, bool swap
)
4277 struct mem_cgroup_threshold_ary
*t
;
4283 t
= rcu_dereference(memcg
->thresholds
.primary
);
4285 t
= rcu_dereference(memcg
->memsw_thresholds
.primary
);
4290 usage
= mem_cgroup_usage(memcg
, swap
);
4293 * current_threshold points to threshold just below or equal to usage.
4294 * If it's not true, a threshold was crossed after last
4295 * call of __mem_cgroup_threshold().
4297 i
= t
->current_threshold
;
4300 * Iterate backward over array of thresholds starting from
4301 * current_threshold and check if a threshold is crossed.
4302 * If none of thresholds below usage is crossed, we read
4303 * only one element of the array here.
4305 for (; i
>= 0 && unlikely(t
->entries
[i
].threshold
> usage
); i
--)
4306 eventfd_signal(t
->entries
[i
].eventfd
, 1);
4308 /* i = current_threshold + 1 */
4312 * Iterate forward over array of thresholds starting from
4313 * current_threshold+1 and check if a threshold is crossed.
4314 * If none of thresholds above usage is crossed, we read
4315 * only one element of the array here.
4317 for (; i
< t
->size
&& unlikely(t
->entries
[i
].threshold
<= usage
); i
++)
4318 eventfd_signal(t
->entries
[i
].eventfd
, 1);
4320 /* Update current_threshold */
4321 t
->current_threshold
= i
- 1;
4326 static void mem_cgroup_threshold(struct mem_cgroup
*memcg
)
4329 __mem_cgroup_threshold(memcg
, false);
4330 if (do_swap_account
)
4331 __mem_cgroup_threshold(memcg
, true);
4333 memcg
= parent_mem_cgroup(memcg
);
4337 static int compare_thresholds(const void *a
, const void *b
)
4339 const struct mem_cgroup_threshold
*_a
= a
;
4340 const struct mem_cgroup_threshold
*_b
= b
;
4342 return _a
->threshold
- _b
->threshold
;
4345 static int mem_cgroup_oom_notify_cb(struct mem_cgroup
*memcg
)
4347 struct mem_cgroup_eventfd_list
*ev
;
4349 list_for_each_entry(ev
, &memcg
->oom_notify
, list
)
4350 eventfd_signal(ev
->eventfd
, 1);
4354 static void mem_cgroup_oom_notify(struct mem_cgroup
*memcg
)
4356 struct mem_cgroup
*iter
;
4358 for_each_mem_cgroup_tree(iter
, memcg
)
4359 mem_cgroup_oom_notify_cb(iter
);
4362 static int mem_cgroup_usage_register_event(struct cgroup
*cgrp
,
4363 struct cftype
*cft
, struct eventfd_ctx
*eventfd
, const char *args
)
4365 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4366 struct mem_cgroup_thresholds
*thresholds
;
4367 struct mem_cgroup_threshold_ary
*new;
4368 int type
= MEMFILE_TYPE(cft
->private);
4369 u64 threshold
, usage
;
4372 ret
= res_counter_memparse_write_strategy(args
, &threshold
);
4376 mutex_lock(&memcg
->thresholds_lock
);
4379 thresholds
= &memcg
->thresholds
;
4380 else if (type
== _MEMSWAP
)
4381 thresholds
= &memcg
->memsw_thresholds
;
4385 usage
= mem_cgroup_usage(memcg
, type
== _MEMSWAP
);
4387 /* Check if a threshold crossed before adding a new one */
4388 if (thresholds
->primary
)
4389 __mem_cgroup_threshold(memcg
, type
== _MEMSWAP
);
4391 size
= thresholds
->primary
? thresholds
->primary
->size
+ 1 : 1;
4393 /* Allocate memory for new array of thresholds */
4394 new = kmalloc(sizeof(*new) + size
* sizeof(struct mem_cgroup_threshold
),
4402 /* Copy thresholds (if any) to new array */
4403 if (thresholds
->primary
) {
4404 memcpy(new->entries
, thresholds
->primary
->entries
, (size
- 1) *
4405 sizeof(struct mem_cgroup_threshold
));
4408 /* Add new threshold */
4409 new->entries
[size
- 1].eventfd
= eventfd
;
4410 new->entries
[size
- 1].threshold
= threshold
;
4412 /* Sort thresholds. Registering of new threshold isn't time-critical */
4413 sort(new->entries
, size
, sizeof(struct mem_cgroup_threshold
),
4414 compare_thresholds
, NULL
);
4416 /* Find current threshold */
4417 new->current_threshold
= -1;
4418 for (i
= 0; i
< size
; i
++) {
4419 if (new->entries
[i
].threshold
<= usage
) {
4421 * new->current_threshold will not be used until
4422 * rcu_assign_pointer(), so it's safe to increment
4425 ++new->current_threshold
;
4430 /* Free old spare buffer and save old primary buffer as spare */
4431 kfree(thresholds
->spare
);
4432 thresholds
->spare
= thresholds
->primary
;
4434 rcu_assign_pointer(thresholds
->primary
, new);
4436 /* To be sure that nobody uses thresholds */
4440 mutex_unlock(&memcg
->thresholds_lock
);
4445 static void mem_cgroup_usage_unregister_event(struct cgroup
*cgrp
,
4446 struct cftype
*cft
, struct eventfd_ctx
*eventfd
)
4448 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4449 struct mem_cgroup_thresholds
*thresholds
;
4450 struct mem_cgroup_threshold_ary
*new;
4451 int type
= MEMFILE_TYPE(cft
->private);
4455 mutex_lock(&memcg
->thresholds_lock
);
4457 thresholds
= &memcg
->thresholds
;
4458 else if (type
== _MEMSWAP
)
4459 thresholds
= &memcg
->memsw_thresholds
;
4463 if (!thresholds
->primary
)
4466 usage
= mem_cgroup_usage(memcg
, type
== _MEMSWAP
);
4468 /* Check if a threshold crossed before removing */
4469 __mem_cgroup_threshold(memcg
, type
== _MEMSWAP
);
4471 /* Calculate new number of threshold */
4473 for (i
= 0; i
< thresholds
->primary
->size
; i
++) {
4474 if (thresholds
->primary
->entries
[i
].eventfd
!= eventfd
)
4478 new = thresholds
->spare
;
4480 /* Set thresholds array to NULL if we don't have thresholds */
4489 /* Copy thresholds and find current threshold */
4490 new->current_threshold
= -1;
4491 for (i
= 0, j
= 0; i
< thresholds
->primary
->size
; i
++) {
4492 if (thresholds
->primary
->entries
[i
].eventfd
== eventfd
)
4495 new->entries
[j
] = thresholds
->primary
->entries
[i
];
4496 if (new->entries
[j
].threshold
<= usage
) {
4498 * new->current_threshold will not be used
4499 * until rcu_assign_pointer(), so it's safe to increment
4502 ++new->current_threshold
;
4508 /* Swap primary and spare array */
4509 thresholds
->spare
= thresholds
->primary
;
4510 /* If all events are unregistered, free the spare array */
4512 kfree(thresholds
->spare
);
4513 thresholds
->spare
= NULL
;
4516 rcu_assign_pointer(thresholds
->primary
, new);
4518 /* To be sure that nobody uses thresholds */
4521 mutex_unlock(&memcg
->thresholds_lock
);
4524 static int mem_cgroup_oom_register_event(struct cgroup
*cgrp
,
4525 struct cftype
*cft
, struct eventfd_ctx
*eventfd
, const char *args
)
4527 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4528 struct mem_cgroup_eventfd_list
*event
;
4529 int type
= MEMFILE_TYPE(cft
->private);
4531 BUG_ON(type
!= _OOM_TYPE
);
4532 event
= kmalloc(sizeof(*event
), GFP_KERNEL
);
4536 spin_lock(&memcg_oom_lock
);
4538 event
->eventfd
= eventfd
;
4539 list_add(&event
->list
, &memcg
->oom_notify
);
4541 /* already in OOM ? */
4542 if (atomic_read(&memcg
->under_oom
))
4543 eventfd_signal(eventfd
, 1);
4544 spin_unlock(&memcg_oom_lock
);
4549 static void mem_cgroup_oom_unregister_event(struct cgroup
*cgrp
,
4550 struct cftype
*cft
, struct eventfd_ctx
*eventfd
)
4552 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4553 struct mem_cgroup_eventfd_list
*ev
, *tmp
;
4554 int type
= MEMFILE_TYPE(cft
->private);
4556 BUG_ON(type
!= _OOM_TYPE
);
4558 spin_lock(&memcg_oom_lock
);
4560 list_for_each_entry_safe(ev
, tmp
, &memcg
->oom_notify
, list
) {
4561 if (ev
->eventfd
== eventfd
) {
4562 list_del(&ev
->list
);
4567 spin_unlock(&memcg_oom_lock
);
4570 static int mem_cgroup_oom_control_read(struct cgroup
*cgrp
,
4571 struct cftype
*cft
, struct cgroup_map_cb
*cb
)
4573 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4575 cb
->fill(cb
, "oom_kill_disable", memcg
->oom_kill_disable
);
4577 if (atomic_read(&memcg
->under_oom
))
4578 cb
->fill(cb
, "under_oom", 1);
4580 cb
->fill(cb
, "under_oom", 0);
4584 static int mem_cgroup_oom_control_write(struct cgroup
*cgrp
,
4585 struct cftype
*cft
, u64 val
)
4587 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4588 struct mem_cgroup
*parent
;
4590 /* cannot set to root cgroup and only 0 and 1 are allowed */
4591 if (!cgrp
->parent
|| !((val
== 0) || (val
== 1)))
4594 parent
= mem_cgroup_from_cont(cgrp
->parent
);
4597 /* oom-kill-disable is a flag for subhierarchy. */
4598 if ((parent
->use_hierarchy
) ||
4599 (memcg
->use_hierarchy
&& !list_empty(&cgrp
->children
))) {
4603 memcg
->oom_kill_disable
= val
;
4605 memcg_oom_recover(memcg
);
4610 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_KMEM
4611 static int memcg_init_kmem(struct mem_cgroup
*memcg
, struct cgroup_subsys
*ss
)
4613 return mem_cgroup_sockets_init(memcg
, ss
);
4616 static void kmem_cgroup_destroy(struct mem_cgroup
*memcg
)
4618 mem_cgroup_sockets_destroy(memcg
);
4621 static int memcg_init_kmem(struct mem_cgroup
*memcg
, struct cgroup_subsys
*ss
)
4626 static void kmem_cgroup_destroy(struct mem_cgroup
*memcg
)
4631 static struct cftype mem_cgroup_files
[] = {
4633 .name
= "usage_in_bytes",
4634 .private = MEMFILE_PRIVATE(_MEM
, RES_USAGE
),
4635 .read
= mem_cgroup_read
,
4636 .register_event
= mem_cgroup_usage_register_event
,
4637 .unregister_event
= mem_cgroup_usage_unregister_event
,
4640 .name
= "max_usage_in_bytes",
4641 .private = MEMFILE_PRIVATE(_MEM
, RES_MAX_USAGE
),
4642 .trigger
= mem_cgroup_reset
,
4643 .read
= mem_cgroup_read
,
4646 .name
= "limit_in_bytes",
4647 .private = MEMFILE_PRIVATE(_MEM
, RES_LIMIT
),
4648 .write_string
= mem_cgroup_write
,
4649 .read
= mem_cgroup_read
,
4652 .name
= "soft_limit_in_bytes",
4653 .private = MEMFILE_PRIVATE(_MEM
, RES_SOFT_LIMIT
),
4654 .write_string
= mem_cgroup_write
,
4655 .read
= mem_cgroup_read
,
4659 .private = MEMFILE_PRIVATE(_MEM
, RES_FAILCNT
),
4660 .trigger
= mem_cgroup_reset
,
4661 .read
= mem_cgroup_read
,
4665 .read_seq_string
= mem_control_stat_show
,
4668 .name
= "force_empty",
4669 .trigger
= mem_cgroup_force_empty_write
,
4672 .name
= "use_hierarchy",
4673 .write_u64
= mem_cgroup_hierarchy_write
,
4674 .read_u64
= mem_cgroup_hierarchy_read
,
4677 .name
= "swappiness",
4678 .read_u64
= mem_cgroup_swappiness_read
,
4679 .write_u64
= mem_cgroup_swappiness_write
,
4682 .name
= "move_charge_at_immigrate",
4683 .read_u64
= mem_cgroup_move_charge_read
,
4684 .write_u64
= mem_cgroup_move_charge_write
,
4687 .name
= "oom_control",
4688 .read_map
= mem_cgroup_oom_control_read
,
4689 .write_u64
= mem_cgroup_oom_control_write
,
4690 .register_event
= mem_cgroup_oom_register_event
,
4691 .unregister_event
= mem_cgroup_oom_unregister_event
,
4692 .private = MEMFILE_PRIVATE(_OOM_TYPE
, OOM_CONTROL
),
4696 .name
= "numa_stat",
4697 .read_seq_string
= mem_control_numa_stat_show
,
4700 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
4702 .name
= "memsw.usage_in_bytes",
4703 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_USAGE
),
4704 .read
= mem_cgroup_read
,
4705 .register_event
= mem_cgroup_usage_register_event
,
4706 .unregister_event
= mem_cgroup_usage_unregister_event
,
4709 .name
= "memsw.max_usage_in_bytes",
4710 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_MAX_USAGE
),
4711 .trigger
= mem_cgroup_reset
,
4712 .read
= mem_cgroup_read
,
4715 .name
= "memsw.limit_in_bytes",
4716 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_LIMIT
),
4717 .write_string
= mem_cgroup_write
,
4718 .read
= mem_cgroup_read
,
4721 .name
= "memsw.failcnt",
4722 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_FAILCNT
),
4723 .trigger
= mem_cgroup_reset
,
4724 .read
= mem_cgroup_read
,
4727 { }, /* terminate */
4730 static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup
*memcg
, int node
)
4732 struct mem_cgroup_per_node
*pn
;
4733 struct mem_cgroup_per_zone
*mz
;
4734 int zone
, tmp
= node
;
4736 * This routine is called against possible nodes.
4737 * But it's BUG to call kmalloc() against offline node.
4739 * TODO: this routine can waste much memory for nodes which will
4740 * never be onlined. It's better to use memory hotplug callback
4743 if (!node_state(node
, N_NORMAL_MEMORY
))
4745 pn
= kzalloc_node(sizeof(*pn
), GFP_KERNEL
, tmp
);
4749 for (zone
= 0; zone
< MAX_NR_ZONES
; zone
++) {
4750 mz
= &pn
->zoneinfo
[zone
];
4751 lruvec_init(&mz
->lruvec
, &NODE_DATA(node
)->node_zones
[zone
]);
4752 mz
->usage_in_excess
= 0;
4753 mz
->on_tree
= false;
4756 memcg
->info
.nodeinfo
[node
] = pn
;
4760 static void free_mem_cgroup_per_zone_info(struct mem_cgroup
*memcg
, int node
)
4762 kfree(memcg
->info
.nodeinfo
[node
]);
4765 static struct mem_cgroup
*mem_cgroup_alloc(void)
4767 struct mem_cgroup
*memcg
;
4768 int size
= sizeof(struct mem_cgroup
);
4770 /* Can be very big if MAX_NUMNODES is very big */
4771 if (size
< PAGE_SIZE
)
4772 memcg
= kzalloc(size
, GFP_KERNEL
);
4774 memcg
= vzalloc(size
);
4779 memcg
->stat
= alloc_percpu(struct mem_cgroup_stat_cpu
);
4782 spin_lock_init(&memcg
->pcp_counter_lock
);
4786 if (size
< PAGE_SIZE
)
4794 * Helpers for freeing a vzalloc()ed mem_cgroup by RCU,
4795 * but in process context. The work_freeing structure is overlaid
4796 * on the rcu_freeing structure, which itself is overlaid on memsw.
4798 static void vfree_work(struct work_struct
*work
)
4800 struct mem_cgroup
*memcg
;
4802 memcg
= container_of(work
, struct mem_cgroup
, work_freeing
);
4805 static void vfree_rcu(struct rcu_head
*rcu_head
)
4807 struct mem_cgroup
*memcg
;
4809 memcg
= container_of(rcu_head
, struct mem_cgroup
, rcu_freeing
);
4810 INIT_WORK(&memcg
->work_freeing
, vfree_work
);
4811 schedule_work(&memcg
->work_freeing
);
4815 * At destroying mem_cgroup, references from swap_cgroup can remain.
4816 * (scanning all at force_empty is too costly...)
4818 * Instead of clearing all references at force_empty, we remember
4819 * the number of reference from swap_cgroup and free mem_cgroup when
4820 * it goes down to 0.
4822 * Removal of cgroup itself succeeds regardless of refs from swap.
4825 static void __mem_cgroup_free(struct mem_cgroup
*memcg
)
4829 mem_cgroup_remove_from_trees(memcg
);
4830 free_css_id(&mem_cgroup_subsys
, &memcg
->css
);
4833 free_mem_cgroup_per_zone_info(memcg
, node
);
4835 free_percpu(memcg
->stat
);
4836 if (sizeof(struct mem_cgroup
) < PAGE_SIZE
)
4837 kfree_rcu(memcg
, rcu_freeing
);
4839 call_rcu(&memcg
->rcu_freeing
, vfree_rcu
);
4842 static void mem_cgroup_get(struct mem_cgroup
*memcg
)
4844 atomic_inc(&memcg
->refcnt
);
4847 static void __mem_cgroup_put(struct mem_cgroup
*memcg
, int count
)
4849 if (atomic_sub_and_test(count
, &memcg
->refcnt
)) {
4850 struct mem_cgroup
*parent
= parent_mem_cgroup(memcg
);
4851 __mem_cgroup_free(memcg
);
4853 mem_cgroup_put(parent
);
4857 static void mem_cgroup_put(struct mem_cgroup
*memcg
)
4859 __mem_cgroup_put(memcg
, 1);
4863 * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled.
4865 struct mem_cgroup
*parent_mem_cgroup(struct mem_cgroup
*memcg
)
4867 if (!memcg
->res
.parent
)
4869 return mem_cgroup_from_res_counter(memcg
->res
.parent
, res
);
4871 EXPORT_SYMBOL(parent_mem_cgroup
);
4873 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
4874 static void __init
enable_swap_cgroup(void)
4876 if (!mem_cgroup_disabled() && really_do_swap_account
)
4877 do_swap_account
= 1;
4880 static void __init
enable_swap_cgroup(void)
4885 static int mem_cgroup_soft_limit_tree_init(void)
4887 struct mem_cgroup_tree_per_node
*rtpn
;
4888 struct mem_cgroup_tree_per_zone
*rtpz
;
4889 int tmp
, node
, zone
;
4891 for_each_node(node
) {
4893 if (!node_state(node
, N_NORMAL_MEMORY
))
4895 rtpn
= kzalloc_node(sizeof(*rtpn
), GFP_KERNEL
, tmp
);
4899 soft_limit_tree
.rb_tree_per_node
[node
] = rtpn
;
4901 for (zone
= 0; zone
< MAX_NR_ZONES
; zone
++) {
4902 rtpz
= &rtpn
->rb_tree_per_zone
[zone
];
4903 rtpz
->rb_root
= RB_ROOT
;
4904 spin_lock_init(&rtpz
->lock
);
4910 for_each_node(node
) {
4911 if (!soft_limit_tree
.rb_tree_per_node
[node
])
4913 kfree(soft_limit_tree
.rb_tree_per_node
[node
]);
4914 soft_limit_tree
.rb_tree_per_node
[node
] = NULL
;
4920 static struct cgroup_subsys_state
* __ref
4921 mem_cgroup_create(struct cgroup
*cont
)
4923 struct mem_cgroup
*memcg
, *parent
;
4924 long error
= -ENOMEM
;
4927 memcg
= mem_cgroup_alloc();
4929 return ERR_PTR(error
);
4932 if (alloc_mem_cgroup_per_zone_info(memcg
, node
))
4936 if (cont
->parent
== NULL
) {
4938 enable_swap_cgroup();
4940 if (mem_cgroup_soft_limit_tree_init())
4942 root_mem_cgroup
= memcg
;
4943 for_each_possible_cpu(cpu
) {
4944 struct memcg_stock_pcp
*stock
=
4945 &per_cpu(memcg_stock
, cpu
);
4946 INIT_WORK(&stock
->work
, drain_local_stock
);
4948 hotcpu_notifier(memcg_cpu_hotplug_callback
, 0);
4950 parent
= mem_cgroup_from_cont(cont
->parent
);
4951 memcg
->use_hierarchy
= parent
->use_hierarchy
;
4952 memcg
->oom_kill_disable
= parent
->oom_kill_disable
;
4955 if (parent
&& parent
->use_hierarchy
) {
4956 res_counter_init(&memcg
->res
, &parent
->res
);
4957 res_counter_init(&memcg
->memsw
, &parent
->memsw
);
4959 * We increment refcnt of the parent to ensure that we can
4960 * safely access it on res_counter_charge/uncharge.
4961 * This refcnt will be decremented when freeing this
4962 * mem_cgroup(see mem_cgroup_put).
4964 mem_cgroup_get(parent
);
4966 res_counter_init(&memcg
->res
, NULL
);
4967 res_counter_init(&memcg
->memsw
, NULL
);
4969 memcg
->last_scanned_node
= MAX_NUMNODES
;
4970 INIT_LIST_HEAD(&memcg
->oom_notify
);
4973 memcg
->swappiness
= mem_cgroup_swappiness(parent
);
4974 atomic_set(&memcg
->refcnt
, 1);
4975 memcg
->move_charge_at_immigrate
= 0;
4976 mutex_init(&memcg
->thresholds_lock
);
4977 spin_lock_init(&memcg
->move_lock
);
4979 error
= memcg_init_kmem(memcg
, &mem_cgroup_subsys
);
4982 * We call put now because our (and parent's) refcnts
4983 * are already in place. mem_cgroup_put() will internally
4984 * call __mem_cgroup_free, so return directly
4986 mem_cgroup_put(memcg
);
4987 return ERR_PTR(error
);
4991 __mem_cgroup_free(memcg
);
4992 return ERR_PTR(error
);
4995 static int mem_cgroup_pre_destroy(struct cgroup
*cont
)
4997 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
4999 return mem_cgroup_force_empty(memcg
, false);
5002 static void mem_cgroup_destroy(struct cgroup
*cont
)
5004 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
5006 kmem_cgroup_destroy(memcg
);
5008 mem_cgroup_put(memcg
);
5012 /* Handlers for move charge at task migration. */
5013 #define PRECHARGE_COUNT_AT_ONCE 256
5014 static int mem_cgroup_do_precharge(unsigned long count
)
5017 int batch_count
= PRECHARGE_COUNT_AT_ONCE
;
5018 struct mem_cgroup
*memcg
= mc
.to
;
5020 if (mem_cgroup_is_root(memcg
)) {
5021 mc
.precharge
+= count
;
5022 /* we don't need css_get for root */
5025 /* try to charge at once */
5027 struct res_counter
*dummy
;
5029 * "memcg" cannot be under rmdir() because we've already checked
5030 * by cgroup_lock_live_cgroup() that it is not removed and we
5031 * are still under the same cgroup_mutex. So we can postpone
5034 if (res_counter_charge(&memcg
->res
, PAGE_SIZE
* count
, &dummy
))
5036 if (do_swap_account
&& res_counter_charge(&memcg
->memsw
,
5037 PAGE_SIZE
* count
, &dummy
)) {
5038 res_counter_uncharge(&memcg
->res
, PAGE_SIZE
* count
);
5041 mc
.precharge
+= count
;
5045 /* fall back to one by one charge */
5047 if (signal_pending(current
)) {
5051 if (!batch_count
--) {
5052 batch_count
= PRECHARGE_COUNT_AT_ONCE
;
5055 ret
= __mem_cgroup_try_charge(NULL
,
5056 GFP_KERNEL
, 1, &memcg
, false);
5058 /* mem_cgroup_clear_mc() will do uncharge later */
5066 * get_mctgt_type - get target type of moving charge
5067 * @vma: the vma the pte to be checked belongs
5068 * @addr: the address corresponding to the pte to be checked
5069 * @ptent: the pte to be checked
5070 * @target: the pointer the target page or swap ent will be stored(can be NULL)
5073 * 0(MC_TARGET_NONE): if the pte is not a target for move charge.
5074 * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
5075 * move charge. if @target is not NULL, the page is stored in target->page
5076 * with extra refcnt got(Callers should handle it).
5077 * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
5078 * target for charge migration. if @target is not NULL, the entry is stored
5081 * Called with pte lock held.
5088 enum mc_target_type
{
5094 static struct page
*mc_handle_present_pte(struct vm_area_struct
*vma
,
5095 unsigned long addr
, pte_t ptent
)
5097 struct page
*page
= vm_normal_page(vma
, addr
, ptent
);
5099 if (!page
|| !page_mapped(page
))
5101 if (PageAnon(page
)) {
5102 /* we don't move shared anon */
5105 } else if (!move_file())
5106 /* we ignore mapcount for file pages */
5108 if (!get_page_unless_zero(page
))
5115 static struct page
*mc_handle_swap_pte(struct vm_area_struct
*vma
,
5116 unsigned long addr
, pte_t ptent
, swp_entry_t
*entry
)
5118 struct page
*page
= NULL
;
5119 swp_entry_t ent
= pte_to_swp_entry(ptent
);
5121 if (!move_anon() || non_swap_entry(ent
))
5124 * Because lookup_swap_cache() updates some statistics counter,
5125 * we call find_get_page() with swapper_space directly.
5127 page
= find_get_page(&swapper_space
, ent
.val
);
5128 if (do_swap_account
)
5129 entry
->val
= ent
.val
;
5134 static struct page
*mc_handle_swap_pte(struct vm_area_struct
*vma
,
5135 unsigned long addr
, pte_t ptent
, swp_entry_t
*entry
)
5141 static struct page
*mc_handle_file_pte(struct vm_area_struct
*vma
,
5142 unsigned long addr
, pte_t ptent
, swp_entry_t
*entry
)
5144 struct page
*page
= NULL
;
5145 struct address_space
*mapping
;
5148 if (!vma
->vm_file
) /* anonymous vma */
5153 mapping
= vma
->vm_file
->f_mapping
;
5154 if (pte_none(ptent
))
5155 pgoff
= linear_page_index(vma
, addr
);
5156 else /* pte_file(ptent) is true */
5157 pgoff
= pte_to_pgoff(ptent
);
5159 /* page is moved even if it's not RSS of this task(page-faulted). */
5160 page
= find_get_page(mapping
, pgoff
);
5163 /* shmem/tmpfs may report page out on swap: account for that too. */
5164 if (radix_tree_exceptional_entry(page
)) {
5165 swp_entry_t swap
= radix_to_swp_entry(page
);
5166 if (do_swap_account
)
5168 page
= find_get_page(&swapper_space
, swap
.val
);
5174 static enum mc_target_type
get_mctgt_type(struct vm_area_struct
*vma
,
5175 unsigned long addr
, pte_t ptent
, union mc_target
*target
)
5177 struct page
*page
= NULL
;
5178 struct page_cgroup
*pc
;
5179 enum mc_target_type ret
= MC_TARGET_NONE
;
5180 swp_entry_t ent
= { .val
= 0 };
5182 if (pte_present(ptent
))
5183 page
= mc_handle_present_pte(vma
, addr
, ptent
);
5184 else if (is_swap_pte(ptent
))
5185 page
= mc_handle_swap_pte(vma
, addr
, ptent
, &ent
);
5186 else if (pte_none(ptent
) || pte_file(ptent
))
5187 page
= mc_handle_file_pte(vma
, addr
, ptent
, &ent
);
5189 if (!page
&& !ent
.val
)
5192 pc
= lookup_page_cgroup(page
);
5194 * Do only loose check w/o page_cgroup lock.
5195 * mem_cgroup_move_account() checks the pc is valid or not under
5198 if (PageCgroupUsed(pc
) && pc
->mem_cgroup
== mc
.from
) {
5199 ret
= MC_TARGET_PAGE
;
5201 target
->page
= page
;
5203 if (!ret
|| !target
)
5206 /* There is a swap entry and a page doesn't exist or isn't charged */
5207 if (ent
.val
&& !ret
&&
5208 css_id(&mc
.from
->css
) == lookup_swap_cgroup_id(ent
)) {
5209 ret
= MC_TARGET_SWAP
;
5216 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
5218 * We don't consider swapping or file mapped pages because THP does not
5219 * support them for now.
5220 * Caller should make sure that pmd_trans_huge(pmd) is true.
5222 static enum mc_target_type
get_mctgt_type_thp(struct vm_area_struct
*vma
,
5223 unsigned long addr
, pmd_t pmd
, union mc_target
*target
)
5225 struct page
*page
= NULL
;
5226 struct page_cgroup
*pc
;
5227 enum mc_target_type ret
= MC_TARGET_NONE
;
5229 page
= pmd_page(pmd
);
5230 VM_BUG_ON(!page
|| !PageHead(page
));
5233 pc
= lookup_page_cgroup(page
);
5234 if (PageCgroupUsed(pc
) && pc
->mem_cgroup
== mc
.from
) {
5235 ret
= MC_TARGET_PAGE
;
5238 target
->page
= page
;
5244 static inline enum mc_target_type
get_mctgt_type_thp(struct vm_area_struct
*vma
,
5245 unsigned long addr
, pmd_t pmd
, union mc_target
*target
)
5247 return MC_TARGET_NONE
;
5251 static int mem_cgroup_count_precharge_pte_range(pmd_t
*pmd
,
5252 unsigned long addr
, unsigned long end
,
5253 struct mm_walk
*walk
)
5255 struct vm_area_struct
*vma
= walk
->private;
5259 if (pmd_trans_huge_lock(pmd
, vma
) == 1) {
5260 if (get_mctgt_type_thp(vma
, addr
, *pmd
, NULL
) == MC_TARGET_PAGE
)
5261 mc
.precharge
+= HPAGE_PMD_NR
;
5262 spin_unlock(&vma
->vm_mm
->page_table_lock
);
5266 if (pmd_trans_unstable(pmd
))
5268 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
5269 for (; addr
!= end
; pte
++, addr
+= PAGE_SIZE
)
5270 if (get_mctgt_type(vma
, addr
, *pte
, NULL
))
5271 mc
.precharge
++; /* increment precharge temporarily */
5272 pte_unmap_unlock(pte
- 1, ptl
);
5278 static unsigned long mem_cgroup_count_precharge(struct mm_struct
*mm
)
5280 unsigned long precharge
;
5281 struct vm_area_struct
*vma
;
5283 down_read(&mm
->mmap_sem
);
5284 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
5285 struct mm_walk mem_cgroup_count_precharge_walk
= {
5286 .pmd_entry
= mem_cgroup_count_precharge_pte_range
,
5290 if (is_vm_hugetlb_page(vma
))
5292 walk_page_range(vma
->vm_start
, vma
->vm_end
,
5293 &mem_cgroup_count_precharge_walk
);
5295 up_read(&mm
->mmap_sem
);
5297 precharge
= mc
.precharge
;
5303 static int mem_cgroup_precharge_mc(struct mm_struct
*mm
)
5305 unsigned long precharge
= mem_cgroup_count_precharge(mm
);
5307 VM_BUG_ON(mc
.moving_task
);
5308 mc
.moving_task
= current
;
5309 return mem_cgroup_do_precharge(precharge
);
5312 /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */
5313 static void __mem_cgroup_clear_mc(void)
5315 struct mem_cgroup
*from
= mc
.from
;
5316 struct mem_cgroup
*to
= mc
.to
;
5318 /* we must uncharge all the leftover precharges from mc.to */
5320 __mem_cgroup_cancel_charge(mc
.to
, mc
.precharge
);
5324 * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
5325 * we must uncharge here.
5327 if (mc
.moved_charge
) {
5328 __mem_cgroup_cancel_charge(mc
.from
, mc
.moved_charge
);
5329 mc
.moved_charge
= 0;
5331 /* we must fixup refcnts and charges */
5332 if (mc
.moved_swap
) {
5333 /* uncharge swap account from the old cgroup */
5334 if (!mem_cgroup_is_root(mc
.from
))
5335 res_counter_uncharge(&mc
.from
->memsw
,
5336 PAGE_SIZE
* mc
.moved_swap
);
5337 __mem_cgroup_put(mc
.from
, mc
.moved_swap
);
5339 if (!mem_cgroup_is_root(mc
.to
)) {
5341 * we charged both to->res and to->memsw, so we should
5344 res_counter_uncharge(&mc
.to
->res
,
5345 PAGE_SIZE
* mc
.moved_swap
);
5347 /* we've already done mem_cgroup_get(mc.to) */
5350 memcg_oom_recover(from
);
5351 memcg_oom_recover(to
);
5352 wake_up_all(&mc
.waitq
);
5355 static void mem_cgroup_clear_mc(void)
5357 struct mem_cgroup
*from
= mc
.from
;
5360 * we must clear moving_task before waking up waiters at the end of
5363 mc
.moving_task
= NULL
;
5364 __mem_cgroup_clear_mc();
5365 spin_lock(&mc
.lock
);
5368 spin_unlock(&mc
.lock
);
5369 mem_cgroup_end_move(from
);
5372 static int mem_cgroup_can_attach(struct cgroup
*cgroup
,
5373 struct cgroup_taskset
*tset
)
5375 struct task_struct
*p
= cgroup_taskset_first(tset
);
5377 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgroup
);
5379 if (memcg
->move_charge_at_immigrate
) {
5380 struct mm_struct
*mm
;
5381 struct mem_cgroup
*from
= mem_cgroup_from_task(p
);
5383 VM_BUG_ON(from
== memcg
);
5385 mm
= get_task_mm(p
);
5388 /* We move charges only when we move a owner of the mm */
5389 if (mm
->owner
== p
) {
5392 VM_BUG_ON(mc
.precharge
);
5393 VM_BUG_ON(mc
.moved_charge
);
5394 VM_BUG_ON(mc
.moved_swap
);
5395 mem_cgroup_start_move(from
);
5396 spin_lock(&mc
.lock
);
5399 spin_unlock(&mc
.lock
);
5400 /* We set mc.moving_task later */
5402 ret
= mem_cgroup_precharge_mc(mm
);
5404 mem_cgroup_clear_mc();
5411 static void mem_cgroup_cancel_attach(struct cgroup
*cgroup
,
5412 struct cgroup_taskset
*tset
)
5414 mem_cgroup_clear_mc();
5417 static int mem_cgroup_move_charge_pte_range(pmd_t
*pmd
,
5418 unsigned long addr
, unsigned long end
,
5419 struct mm_walk
*walk
)
5422 struct vm_area_struct
*vma
= walk
->private;
5425 enum mc_target_type target_type
;
5426 union mc_target target
;
5428 struct page_cgroup
*pc
;
5431 * We don't take compound_lock() here but no race with splitting thp
5433 * - if pmd_trans_huge_lock() returns 1, the relevant thp is not
5434 * under splitting, which means there's no concurrent thp split,
5435 * - if another thread runs into split_huge_page() just after we
5436 * entered this if-block, the thread must wait for page table lock
5437 * to be unlocked in __split_huge_page_splitting(), where the main
5438 * part of thp split is not executed yet.
5440 if (pmd_trans_huge_lock(pmd
, vma
) == 1) {
5441 if (mc
.precharge
< HPAGE_PMD_NR
) {
5442 spin_unlock(&vma
->vm_mm
->page_table_lock
);
5445 target_type
= get_mctgt_type_thp(vma
, addr
, *pmd
, &target
);
5446 if (target_type
== MC_TARGET_PAGE
) {
5448 if (!isolate_lru_page(page
)) {
5449 pc
= lookup_page_cgroup(page
);
5450 if (!mem_cgroup_move_account(page
, HPAGE_PMD_NR
,
5451 pc
, mc
.from
, mc
.to
)) {
5452 mc
.precharge
-= HPAGE_PMD_NR
;
5453 mc
.moved_charge
+= HPAGE_PMD_NR
;
5455 putback_lru_page(page
);
5459 spin_unlock(&vma
->vm_mm
->page_table_lock
);
5463 if (pmd_trans_unstable(pmd
))
5466 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
5467 for (; addr
!= end
; addr
+= PAGE_SIZE
) {
5468 pte_t ptent
= *(pte
++);
5474 switch (get_mctgt_type(vma
, addr
, ptent
, &target
)) {
5475 case MC_TARGET_PAGE
:
5477 if (isolate_lru_page(page
))
5479 pc
= lookup_page_cgroup(page
);
5480 if (!mem_cgroup_move_account(page
, 1, pc
,
5483 /* we uncharge from mc.from later. */
5486 putback_lru_page(page
);
5487 put
: /* get_mctgt_type() gets the page */
5490 case MC_TARGET_SWAP
:
5492 if (!mem_cgroup_move_swap_account(ent
, mc
.from
, mc
.to
)) {
5494 /* we fixup refcnts and charges later. */
5502 pte_unmap_unlock(pte
- 1, ptl
);
5507 * We have consumed all precharges we got in can_attach().
5508 * We try charge one by one, but don't do any additional
5509 * charges to mc.to if we have failed in charge once in attach()
5512 ret
= mem_cgroup_do_precharge(1);
5520 static void mem_cgroup_move_charge(struct mm_struct
*mm
)
5522 struct vm_area_struct
*vma
;
5524 lru_add_drain_all();
5526 if (unlikely(!down_read_trylock(&mm
->mmap_sem
))) {
5528 * Someone who are holding the mmap_sem might be waiting in
5529 * waitq. So we cancel all extra charges, wake up all waiters,
5530 * and retry. Because we cancel precharges, we might not be able
5531 * to move enough charges, but moving charge is a best-effort
5532 * feature anyway, so it wouldn't be a big problem.
5534 __mem_cgroup_clear_mc();
5538 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
5540 struct mm_walk mem_cgroup_move_charge_walk
= {
5541 .pmd_entry
= mem_cgroup_move_charge_pte_range
,
5545 if (is_vm_hugetlb_page(vma
))
5547 ret
= walk_page_range(vma
->vm_start
, vma
->vm_end
,
5548 &mem_cgroup_move_charge_walk
);
5551 * means we have consumed all precharges and failed in
5552 * doing additional charge. Just abandon here.
5556 up_read(&mm
->mmap_sem
);
5559 static void mem_cgroup_move_task(struct cgroup
*cont
,
5560 struct cgroup_taskset
*tset
)
5562 struct task_struct
*p
= cgroup_taskset_first(tset
);
5563 struct mm_struct
*mm
= get_task_mm(p
);
5567 mem_cgroup_move_charge(mm
);
5571 mem_cgroup_clear_mc();
5573 #else /* !CONFIG_MMU */
5574 static int mem_cgroup_can_attach(struct cgroup
*cgroup
,
5575 struct cgroup_taskset
*tset
)
5579 static void mem_cgroup_cancel_attach(struct cgroup
*cgroup
,
5580 struct cgroup_taskset
*tset
)
5583 static void mem_cgroup_move_task(struct cgroup
*cont
,
5584 struct cgroup_taskset
*tset
)
5589 struct cgroup_subsys mem_cgroup_subsys
= {
5591 .subsys_id
= mem_cgroup_subsys_id
,
5592 .create
= mem_cgroup_create
,
5593 .pre_destroy
= mem_cgroup_pre_destroy
,
5594 .destroy
= mem_cgroup_destroy
,
5595 .can_attach
= mem_cgroup_can_attach
,
5596 .cancel_attach
= mem_cgroup_cancel_attach
,
5597 .attach
= mem_cgroup_move_task
,
5598 .base_cftypes
= mem_cgroup_files
,
5601 .__DEPRECATED_clear_css_refs
= true,
5604 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
5605 static int __init
enable_swap_account(char *s
)
5607 /* consider enabled if no parameter or 1 is given */
5608 if (!strcmp(s
, "1"))
5609 really_do_swap_account
= 1;
5610 else if (!strcmp(s
, "0"))
5611 really_do_swap_account
= 0;
5614 __setup("swapaccount=", enable_swap_account
);