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 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_DATA
, /* end of data requires synchronization */
92 MEM_CGROUP_STAT_NSTATS
,
95 enum mem_cgroup_events_index
{
96 MEM_CGROUP_EVENTS_PGPGIN
, /* # of pages paged in */
97 MEM_CGROUP_EVENTS_PGPGOUT
, /* # of pages paged out */
98 MEM_CGROUP_EVENTS_COUNT
, /* # of pages paged in/out */
99 MEM_CGROUP_EVENTS_PGFAULT
, /* # of page-faults */
100 MEM_CGROUP_EVENTS_PGMAJFAULT
, /* # of major page-faults */
101 MEM_CGROUP_EVENTS_NSTATS
,
104 * Per memcg event counter is incremented at every pagein/pageout. With THP,
105 * it will be incremated by the number of pages. This counter is used for
106 * for trigger some periodic events. This is straightforward and better
107 * than using jiffies etc. to handle periodic memcg event.
109 enum mem_cgroup_events_target
{
110 MEM_CGROUP_TARGET_THRESH
,
111 MEM_CGROUP_TARGET_SOFTLIMIT
,
112 MEM_CGROUP_TARGET_NUMAINFO
,
115 #define THRESHOLDS_EVENTS_TARGET (128)
116 #define SOFTLIMIT_EVENTS_TARGET (1024)
117 #define NUMAINFO_EVENTS_TARGET (1024)
119 struct mem_cgroup_stat_cpu
{
120 long count
[MEM_CGROUP_STAT_NSTATS
];
121 unsigned long events
[MEM_CGROUP_EVENTS_NSTATS
];
122 unsigned long targets
[MEM_CGROUP_NTARGETS
];
125 struct mem_cgroup_reclaim_iter
{
126 /* css_id of the last scanned hierarchy member */
128 /* scan generation, increased every round-trip */
129 unsigned int generation
;
133 * per-zone information in memory controller.
135 struct mem_cgroup_per_zone
{
136 struct lruvec lruvec
;
137 unsigned long lru_size
[NR_LRU_LISTS
];
139 struct mem_cgroup_reclaim_iter reclaim_iter
[DEF_PRIORITY
+ 1];
141 struct zone_reclaim_stat reclaim_stat
;
142 struct rb_node tree_node
; /* RB tree node */
143 unsigned long long usage_in_excess
;/* Set to the value by which */
144 /* the soft limit is exceeded*/
146 struct mem_cgroup
*memcg
; /* Back pointer, we cannot */
147 /* use container_of */
150 struct mem_cgroup_per_node
{
151 struct mem_cgroup_per_zone zoneinfo
[MAX_NR_ZONES
];
154 struct mem_cgroup_lru_info
{
155 struct mem_cgroup_per_node
*nodeinfo
[MAX_NUMNODES
];
159 * Cgroups above their limits are maintained in a RB-Tree, independent of
160 * their hierarchy representation
163 struct mem_cgroup_tree_per_zone
{
164 struct rb_root rb_root
;
168 struct mem_cgroup_tree_per_node
{
169 struct mem_cgroup_tree_per_zone rb_tree_per_zone
[MAX_NR_ZONES
];
172 struct mem_cgroup_tree
{
173 struct mem_cgroup_tree_per_node
*rb_tree_per_node
[MAX_NUMNODES
];
176 static struct mem_cgroup_tree soft_limit_tree __read_mostly
;
178 struct mem_cgroup_threshold
{
179 struct eventfd_ctx
*eventfd
;
184 struct mem_cgroup_threshold_ary
{
185 /* An array index points to threshold just below usage. */
186 int current_threshold
;
187 /* Size of entries[] */
189 /* Array of thresholds */
190 struct mem_cgroup_threshold entries
[0];
193 struct mem_cgroup_thresholds
{
194 /* Primary thresholds array */
195 struct mem_cgroup_threshold_ary
*primary
;
197 * Spare threshold array.
198 * This is needed to make mem_cgroup_unregister_event() "never fail".
199 * It must be able to store at least primary->size - 1 entries.
201 struct mem_cgroup_threshold_ary
*spare
;
205 struct mem_cgroup_eventfd_list
{
206 struct list_head list
;
207 struct eventfd_ctx
*eventfd
;
210 static void mem_cgroup_threshold(struct mem_cgroup
*memcg
);
211 static void mem_cgroup_oom_notify(struct mem_cgroup
*memcg
);
214 * The memory controller data structure. The memory controller controls both
215 * page cache and RSS per cgroup. We would eventually like to provide
216 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
217 * to help the administrator determine what knobs to tune.
219 * TODO: Add a water mark for the memory controller. Reclaim will begin when
220 * we hit the water mark. May be even add a low water mark, such that
221 * no reclaim occurs from a cgroup at it's low water mark, this is
222 * a feature that will be implemented much later in the future.
225 struct cgroup_subsys_state css
;
227 * the counter to account for memory usage
229 struct res_counter res
;
233 * the counter to account for mem+swap usage.
235 struct res_counter memsw
;
238 * rcu_freeing is used only when freeing struct mem_cgroup,
239 * so put it into a union to avoid wasting more memory.
240 * It must be disjoint from the css field. It could be
241 * in a union with the res field, but res plays a much
242 * larger part in mem_cgroup life than memsw, and might
243 * be of interest, even at time of free, when debugging.
244 * So share rcu_head with the less interesting memsw.
246 struct rcu_head rcu_freeing
;
248 * But when using vfree(), that cannot be done at
249 * interrupt time, so we must then queue the work.
251 struct work_struct work_freeing
;
255 * Per cgroup active and inactive list, similar to the
256 * per zone LRU lists.
258 struct mem_cgroup_lru_info info
;
259 int last_scanned_node
;
261 nodemask_t scan_nodes
;
262 atomic_t numainfo_events
;
263 atomic_t numainfo_updating
;
266 * Should the accounting and control be hierarchical, per subtree?
276 /* OOM-Killer disable */
277 int oom_kill_disable
;
279 /* set when res.limit == memsw.limit */
280 bool memsw_is_minimum
;
282 /* protect arrays of thresholds */
283 struct mutex thresholds_lock
;
285 /* thresholds for memory usage. RCU-protected */
286 struct mem_cgroup_thresholds thresholds
;
288 /* thresholds for mem+swap usage. RCU-protected */
289 struct mem_cgroup_thresholds memsw_thresholds
;
291 /* For oom notifier event fd */
292 struct list_head oom_notify
;
295 * Should we move charges of a task when a task is moved into this
296 * mem_cgroup ? And what type of charges should we move ?
298 unsigned long move_charge_at_immigrate
;
300 * set > 0 if pages under this cgroup are moving to other cgroup.
302 atomic_t moving_account
;
303 /* taken only while moving_account > 0 */
304 spinlock_t move_lock
;
308 struct mem_cgroup_stat_cpu
*stat
;
310 * used when a cpu is offlined or other synchronizations
311 * See mem_cgroup_read_stat().
313 struct mem_cgroup_stat_cpu nocpu_base
;
314 spinlock_t pcp_counter_lock
;
317 struct tcp_memcontrol tcp_mem
;
321 /* Stuffs for move charges at task migration. */
323 * Types of charges to be moved. "move_charge_at_immitgrate" is treated as a
324 * left-shifted bitmap of these types.
327 MOVE_CHARGE_TYPE_ANON
, /* private anonymous page and swap of it */
328 MOVE_CHARGE_TYPE_FILE
, /* file page(including tmpfs) and swap of it */
332 /* "mc" and its members are protected by cgroup_mutex */
333 static struct move_charge_struct
{
334 spinlock_t lock
; /* for from, to */
335 struct mem_cgroup
*from
;
336 struct mem_cgroup
*to
;
337 unsigned long precharge
;
338 unsigned long moved_charge
;
339 unsigned long moved_swap
;
340 struct task_struct
*moving_task
; /* a task moving charges */
341 wait_queue_head_t waitq
; /* a waitq for other context */
343 .lock
= __SPIN_LOCK_UNLOCKED(mc
.lock
),
344 .waitq
= __WAIT_QUEUE_HEAD_INITIALIZER(mc
.waitq
),
347 static bool move_anon(void)
349 return test_bit(MOVE_CHARGE_TYPE_ANON
,
350 &mc
.to
->move_charge_at_immigrate
);
353 static bool move_file(void)
355 return test_bit(MOVE_CHARGE_TYPE_FILE
,
356 &mc
.to
->move_charge_at_immigrate
);
360 * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft
361 * limit reclaim to prevent infinite loops, if they ever occur.
363 #define MEM_CGROUP_MAX_RECLAIM_LOOPS (100)
364 #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS (2)
367 MEM_CGROUP_CHARGE_TYPE_CACHE
= 0,
368 MEM_CGROUP_CHARGE_TYPE_MAPPED
,
369 MEM_CGROUP_CHARGE_TYPE_SHMEM
, /* used by page migration of shmem */
370 MEM_CGROUP_CHARGE_TYPE_FORCE
, /* used by force_empty */
371 MEM_CGROUP_CHARGE_TYPE_SWAPOUT
, /* for accounting swapcache */
372 MEM_CGROUP_CHARGE_TYPE_DROP
, /* a page was unused swap cache */
376 /* for encoding cft->private value on file */
379 #define _OOM_TYPE (2)
380 #define MEMFILE_PRIVATE(x, val) (((x) << 16) | (val))
381 #define MEMFILE_TYPE(val) (((val) >> 16) & 0xffff)
382 #define MEMFILE_ATTR(val) ((val) & 0xffff)
383 /* Used for OOM nofiier */
384 #define OOM_CONTROL (0)
387 * Reclaim flags for mem_cgroup_hierarchical_reclaim
389 #define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0
390 #define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT)
391 #define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1
392 #define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT)
394 static void mem_cgroup_get(struct mem_cgroup
*memcg
);
395 static void mem_cgroup_put(struct mem_cgroup
*memcg
);
397 /* Writing them here to avoid exposing memcg's inner layout */
398 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_KMEM
399 #include <net/sock.h>
402 static bool mem_cgroup_is_root(struct mem_cgroup
*memcg
);
403 void sock_update_memcg(struct sock
*sk
)
405 if (mem_cgroup_sockets_enabled
) {
406 struct mem_cgroup
*memcg
;
408 BUG_ON(!sk
->sk_prot
->proto_cgroup
);
410 /* Socket cloning can throw us here with sk_cgrp already
411 * filled. It won't however, necessarily happen from
412 * process context. So the test for root memcg given
413 * the current task's memcg won't help us in this case.
415 * Respecting the original socket's memcg is a better
416 * decision in this case.
419 BUG_ON(mem_cgroup_is_root(sk
->sk_cgrp
->memcg
));
420 mem_cgroup_get(sk
->sk_cgrp
->memcg
);
425 memcg
= mem_cgroup_from_task(current
);
426 if (!mem_cgroup_is_root(memcg
)) {
427 mem_cgroup_get(memcg
);
428 sk
->sk_cgrp
= sk
->sk_prot
->proto_cgroup(memcg
);
433 EXPORT_SYMBOL(sock_update_memcg
);
435 void sock_release_memcg(struct sock
*sk
)
437 if (mem_cgroup_sockets_enabled
&& sk
->sk_cgrp
) {
438 struct mem_cgroup
*memcg
;
439 WARN_ON(!sk
->sk_cgrp
->memcg
);
440 memcg
= sk
->sk_cgrp
->memcg
;
441 mem_cgroup_put(memcg
);
446 struct cg_proto
*tcp_proto_cgroup(struct mem_cgroup
*memcg
)
448 if (!memcg
|| mem_cgroup_is_root(memcg
))
451 return &memcg
->tcp_mem
.cg_proto
;
453 EXPORT_SYMBOL(tcp_proto_cgroup
);
454 #endif /* CONFIG_INET */
455 #endif /* CONFIG_CGROUP_MEM_RES_CTLR_KMEM */
457 static void drain_all_stock_async(struct mem_cgroup
*memcg
);
459 static struct mem_cgroup_per_zone
*
460 mem_cgroup_zoneinfo(struct mem_cgroup
*memcg
, int nid
, int zid
)
462 return &memcg
->info
.nodeinfo
[nid
]->zoneinfo
[zid
];
465 struct cgroup_subsys_state
*mem_cgroup_css(struct mem_cgroup
*memcg
)
470 static struct mem_cgroup_per_zone
*
471 page_cgroup_zoneinfo(struct mem_cgroup
*memcg
, struct page
*page
)
473 int nid
= page_to_nid(page
);
474 int zid
= page_zonenum(page
);
476 return mem_cgroup_zoneinfo(memcg
, nid
, zid
);
479 static struct mem_cgroup_tree_per_zone
*
480 soft_limit_tree_node_zone(int nid
, int zid
)
482 return &soft_limit_tree
.rb_tree_per_node
[nid
]->rb_tree_per_zone
[zid
];
485 static struct mem_cgroup_tree_per_zone
*
486 soft_limit_tree_from_page(struct page
*page
)
488 int nid
= page_to_nid(page
);
489 int zid
= page_zonenum(page
);
491 return &soft_limit_tree
.rb_tree_per_node
[nid
]->rb_tree_per_zone
[zid
];
495 __mem_cgroup_insert_exceeded(struct mem_cgroup
*memcg
,
496 struct mem_cgroup_per_zone
*mz
,
497 struct mem_cgroup_tree_per_zone
*mctz
,
498 unsigned long long new_usage_in_excess
)
500 struct rb_node
**p
= &mctz
->rb_root
.rb_node
;
501 struct rb_node
*parent
= NULL
;
502 struct mem_cgroup_per_zone
*mz_node
;
507 mz
->usage_in_excess
= new_usage_in_excess
;
508 if (!mz
->usage_in_excess
)
512 mz_node
= rb_entry(parent
, struct mem_cgroup_per_zone
,
514 if (mz
->usage_in_excess
< mz_node
->usage_in_excess
)
517 * We can't avoid mem cgroups that are over their soft
518 * limit by the same amount
520 else if (mz
->usage_in_excess
>= mz_node
->usage_in_excess
)
523 rb_link_node(&mz
->tree_node
, parent
, p
);
524 rb_insert_color(&mz
->tree_node
, &mctz
->rb_root
);
529 __mem_cgroup_remove_exceeded(struct mem_cgroup
*memcg
,
530 struct mem_cgroup_per_zone
*mz
,
531 struct mem_cgroup_tree_per_zone
*mctz
)
535 rb_erase(&mz
->tree_node
, &mctz
->rb_root
);
540 mem_cgroup_remove_exceeded(struct mem_cgroup
*memcg
,
541 struct mem_cgroup_per_zone
*mz
,
542 struct mem_cgroup_tree_per_zone
*mctz
)
544 spin_lock(&mctz
->lock
);
545 __mem_cgroup_remove_exceeded(memcg
, mz
, mctz
);
546 spin_unlock(&mctz
->lock
);
550 static void mem_cgroup_update_tree(struct mem_cgroup
*memcg
, struct page
*page
)
552 unsigned long long excess
;
553 struct mem_cgroup_per_zone
*mz
;
554 struct mem_cgroup_tree_per_zone
*mctz
;
555 int nid
= page_to_nid(page
);
556 int zid
= page_zonenum(page
);
557 mctz
= soft_limit_tree_from_page(page
);
560 * Necessary to update all ancestors when hierarchy is used.
561 * because their event counter is not touched.
563 for (; memcg
; memcg
= parent_mem_cgroup(memcg
)) {
564 mz
= mem_cgroup_zoneinfo(memcg
, nid
, zid
);
565 excess
= res_counter_soft_limit_excess(&memcg
->res
);
567 * We have to update the tree if mz is on RB-tree or
568 * mem is over its softlimit.
570 if (excess
|| mz
->on_tree
) {
571 spin_lock(&mctz
->lock
);
572 /* if on-tree, remove it */
574 __mem_cgroup_remove_exceeded(memcg
, mz
, mctz
);
576 * Insert again. mz->usage_in_excess will be updated.
577 * If excess is 0, no tree ops.
579 __mem_cgroup_insert_exceeded(memcg
, mz
, mctz
, excess
);
580 spin_unlock(&mctz
->lock
);
585 static void mem_cgroup_remove_from_trees(struct mem_cgroup
*memcg
)
588 struct mem_cgroup_per_zone
*mz
;
589 struct mem_cgroup_tree_per_zone
*mctz
;
591 for_each_node(node
) {
592 for (zone
= 0; zone
< MAX_NR_ZONES
; zone
++) {
593 mz
= mem_cgroup_zoneinfo(memcg
, node
, zone
);
594 mctz
= soft_limit_tree_node_zone(node
, zone
);
595 mem_cgroup_remove_exceeded(memcg
, mz
, mctz
);
600 static struct mem_cgroup_per_zone
*
601 __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone
*mctz
)
603 struct rb_node
*rightmost
= NULL
;
604 struct mem_cgroup_per_zone
*mz
;
608 rightmost
= rb_last(&mctz
->rb_root
);
610 goto done
; /* Nothing to reclaim from */
612 mz
= rb_entry(rightmost
, struct mem_cgroup_per_zone
, tree_node
);
614 * Remove the node now but someone else can add it back,
615 * we will to add it back at the end of reclaim to its correct
616 * position in the tree.
618 __mem_cgroup_remove_exceeded(mz
->memcg
, mz
, mctz
);
619 if (!res_counter_soft_limit_excess(&mz
->memcg
->res
) ||
620 !css_tryget(&mz
->memcg
->css
))
626 static struct mem_cgroup_per_zone
*
627 mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone
*mctz
)
629 struct mem_cgroup_per_zone
*mz
;
631 spin_lock(&mctz
->lock
);
632 mz
= __mem_cgroup_largest_soft_limit_node(mctz
);
633 spin_unlock(&mctz
->lock
);
638 * Implementation Note: reading percpu statistics for memcg.
640 * Both of vmstat[] and percpu_counter has threshold and do periodic
641 * synchronization to implement "quick" read. There are trade-off between
642 * reading cost and precision of value. Then, we may have a chance to implement
643 * a periodic synchronizion of counter in memcg's counter.
645 * But this _read() function is used for user interface now. The user accounts
646 * memory usage by memory cgroup and he _always_ requires exact value because
647 * he accounts memory. Even if we provide quick-and-fuzzy read, we always
648 * have to visit all online cpus and make sum. So, for now, unnecessary
649 * synchronization is not implemented. (just implemented for cpu hotplug)
651 * If there are kernel internal actions which can make use of some not-exact
652 * value, and reading all cpu value can be performance bottleneck in some
653 * common workload, threashold and synchonization as vmstat[] should be
656 static long mem_cgroup_read_stat(struct mem_cgroup
*memcg
,
657 enum mem_cgroup_stat_index idx
)
663 for_each_online_cpu(cpu
)
664 val
+= per_cpu(memcg
->stat
->count
[idx
], cpu
);
665 #ifdef CONFIG_HOTPLUG_CPU
666 spin_lock(&memcg
->pcp_counter_lock
);
667 val
+= memcg
->nocpu_base
.count
[idx
];
668 spin_unlock(&memcg
->pcp_counter_lock
);
674 static void mem_cgroup_swap_statistics(struct mem_cgroup
*memcg
,
677 int val
= (charge
) ? 1 : -1;
678 this_cpu_add(memcg
->stat
->count
[MEM_CGROUP_STAT_SWAPOUT
], val
);
681 static unsigned long mem_cgroup_read_events(struct mem_cgroup
*memcg
,
682 enum mem_cgroup_events_index idx
)
684 unsigned long val
= 0;
687 for_each_online_cpu(cpu
)
688 val
+= per_cpu(memcg
->stat
->events
[idx
], cpu
);
689 #ifdef CONFIG_HOTPLUG_CPU
690 spin_lock(&memcg
->pcp_counter_lock
);
691 val
+= memcg
->nocpu_base
.events
[idx
];
692 spin_unlock(&memcg
->pcp_counter_lock
);
697 static void mem_cgroup_charge_statistics(struct mem_cgroup
*memcg
,
698 bool anon
, int nr_pages
)
703 * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is
704 * counted as CACHE even if it's on ANON LRU.
707 __this_cpu_add(memcg
->stat
->count
[MEM_CGROUP_STAT_RSS
],
710 __this_cpu_add(memcg
->stat
->count
[MEM_CGROUP_STAT_CACHE
],
713 /* pagein of a big page is an event. So, ignore page size */
715 __this_cpu_inc(memcg
->stat
->events
[MEM_CGROUP_EVENTS_PGPGIN
]);
717 __this_cpu_inc(memcg
->stat
->events
[MEM_CGROUP_EVENTS_PGPGOUT
]);
718 nr_pages
= -nr_pages
; /* for event */
721 __this_cpu_add(memcg
->stat
->events
[MEM_CGROUP_EVENTS_COUNT
], nr_pages
);
727 mem_cgroup_zone_nr_lru_pages(struct mem_cgroup
*memcg
, int nid
, int zid
,
728 unsigned int lru_mask
)
730 struct mem_cgroup_per_zone
*mz
;
732 unsigned long ret
= 0;
734 mz
= mem_cgroup_zoneinfo(memcg
, nid
, zid
);
737 if (BIT(lru
) & lru_mask
)
738 ret
+= mz
->lru_size
[lru
];
744 mem_cgroup_node_nr_lru_pages(struct mem_cgroup
*memcg
,
745 int nid
, unsigned int lru_mask
)
750 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++)
751 total
+= mem_cgroup_zone_nr_lru_pages(memcg
,
757 static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup
*memcg
,
758 unsigned int lru_mask
)
763 for_each_node_state(nid
, N_HIGH_MEMORY
)
764 total
+= mem_cgroup_node_nr_lru_pages(memcg
, nid
, lru_mask
);
768 static bool mem_cgroup_event_ratelimit(struct mem_cgroup
*memcg
,
769 enum mem_cgroup_events_target target
)
771 unsigned long val
, next
;
773 val
= __this_cpu_read(memcg
->stat
->events
[MEM_CGROUP_EVENTS_COUNT
]);
774 next
= __this_cpu_read(memcg
->stat
->targets
[target
]);
775 /* from time_after() in jiffies.h */
776 if ((long)next
- (long)val
< 0) {
778 case MEM_CGROUP_TARGET_THRESH
:
779 next
= val
+ THRESHOLDS_EVENTS_TARGET
;
781 case MEM_CGROUP_TARGET_SOFTLIMIT
:
782 next
= val
+ SOFTLIMIT_EVENTS_TARGET
;
784 case MEM_CGROUP_TARGET_NUMAINFO
:
785 next
= val
+ NUMAINFO_EVENTS_TARGET
;
790 __this_cpu_write(memcg
->stat
->targets
[target
], next
);
797 * Check events in order.
800 static void memcg_check_events(struct mem_cgroup
*memcg
, struct page
*page
)
803 /* threshold event is triggered in finer grain than soft limit */
804 if (unlikely(mem_cgroup_event_ratelimit(memcg
,
805 MEM_CGROUP_TARGET_THRESH
))) {
807 bool do_numainfo __maybe_unused
;
809 do_softlimit
= mem_cgroup_event_ratelimit(memcg
,
810 MEM_CGROUP_TARGET_SOFTLIMIT
);
812 do_numainfo
= mem_cgroup_event_ratelimit(memcg
,
813 MEM_CGROUP_TARGET_NUMAINFO
);
817 mem_cgroup_threshold(memcg
);
818 if (unlikely(do_softlimit
))
819 mem_cgroup_update_tree(memcg
, page
);
821 if (unlikely(do_numainfo
))
822 atomic_inc(&memcg
->numainfo_events
);
828 struct mem_cgroup
*mem_cgroup_from_cont(struct cgroup
*cont
)
830 return container_of(cgroup_subsys_state(cont
,
831 mem_cgroup_subsys_id
), struct mem_cgroup
,
835 struct mem_cgroup
*mem_cgroup_from_task(struct task_struct
*p
)
838 * mm_update_next_owner() may clear mm->owner to NULL
839 * if it races with swapoff, page migration, etc.
840 * So this can be called with p == NULL.
845 return container_of(task_subsys_state(p
, mem_cgroup_subsys_id
),
846 struct mem_cgroup
, css
);
849 struct mem_cgroup
*try_get_mem_cgroup_from_mm(struct mm_struct
*mm
)
851 struct mem_cgroup
*memcg
= NULL
;
856 * Because we have no locks, mm->owner's may be being moved to other
857 * cgroup. We use css_tryget() here even if this looks
858 * pessimistic (rather than adding locks here).
862 memcg
= mem_cgroup_from_task(rcu_dereference(mm
->owner
));
863 if (unlikely(!memcg
))
865 } while (!css_tryget(&memcg
->css
));
871 * mem_cgroup_iter - iterate over memory cgroup hierarchy
872 * @root: hierarchy root
873 * @prev: previously returned memcg, NULL on first invocation
874 * @reclaim: cookie for shared reclaim walks, NULL for full walks
876 * Returns references to children of the hierarchy below @root, or
877 * @root itself, or %NULL after a full round-trip.
879 * Caller must pass the return value in @prev on subsequent
880 * invocations for reference counting, or use mem_cgroup_iter_break()
881 * to cancel a hierarchy walk before the round-trip is complete.
883 * Reclaimers can specify a zone and a priority level in @reclaim to
884 * divide up the memcgs in the hierarchy among all concurrent
885 * reclaimers operating on the same zone and priority.
887 struct mem_cgroup
*mem_cgroup_iter(struct mem_cgroup
*root
,
888 struct mem_cgroup
*prev
,
889 struct mem_cgroup_reclaim_cookie
*reclaim
)
891 struct mem_cgroup
*memcg
= NULL
;
894 if (mem_cgroup_disabled())
898 root
= root_mem_cgroup
;
900 if (prev
&& !reclaim
)
901 id
= css_id(&prev
->css
);
903 if (prev
&& prev
!= root
)
906 if (!root
->use_hierarchy
&& root
!= root_mem_cgroup
) {
913 struct mem_cgroup_reclaim_iter
*uninitialized_var(iter
);
914 struct cgroup_subsys_state
*css
;
917 int nid
= zone_to_nid(reclaim
->zone
);
918 int zid
= zone_idx(reclaim
->zone
);
919 struct mem_cgroup_per_zone
*mz
;
921 mz
= mem_cgroup_zoneinfo(root
, nid
, zid
);
922 iter
= &mz
->reclaim_iter
[reclaim
->priority
];
923 if (prev
&& reclaim
->generation
!= iter
->generation
)
929 css
= css_get_next(&mem_cgroup_subsys
, id
+ 1, &root
->css
, &id
);
931 if (css
== &root
->css
|| css_tryget(css
))
932 memcg
= container_of(css
,
933 struct mem_cgroup
, css
);
942 else if (!prev
&& memcg
)
943 reclaim
->generation
= iter
->generation
;
953 * mem_cgroup_iter_break - abort a hierarchy walk prematurely
954 * @root: hierarchy root
955 * @prev: last visited hierarchy member as returned by mem_cgroup_iter()
957 void mem_cgroup_iter_break(struct mem_cgroup
*root
,
958 struct mem_cgroup
*prev
)
961 root
= root_mem_cgroup
;
962 if (prev
&& prev
!= root
)
967 * Iteration constructs for visiting all cgroups (under a tree). If
968 * loops are exited prematurely (break), mem_cgroup_iter_break() must
969 * be used for reference counting.
971 #define for_each_mem_cgroup_tree(iter, root) \
972 for (iter = mem_cgroup_iter(root, NULL, NULL); \
974 iter = mem_cgroup_iter(root, iter, NULL))
976 #define for_each_mem_cgroup(iter) \
977 for (iter = mem_cgroup_iter(NULL, NULL, NULL); \
979 iter = mem_cgroup_iter(NULL, iter, NULL))
981 static inline bool mem_cgroup_is_root(struct mem_cgroup
*memcg
)
983 return (memcg
== root_mem_cgroup
);
986 void mem_cgroup_count_vm_event(struct mm_struct
*mm
, enum vm_event_item idx
)
988 struct mem_cgroup
*memcg
;
994 memcg
= mem_cgroup_from_task(rcu_dereference(mm
->owner
));
995 if (unlikely(!memcg
))
1000 this_cpu_inc(memcg
->stat
->events
[MEM_CGROUP_EVENTS_PGFAULT
]);
1003 this_cpu_inc(memcg
->stat
->events
[MEM_CGROUP_EVENTS_PGMAJFAULT
]);
1011 EXPORT_SYMBOL(mem_cgroup_count_vm_event
);
1014 * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg
1015 * @zone: zone of the wanted lruvec
1016 * @mem: memcg of the wanted lruvec
1018 * Returns the lru list vector holding pages for the given @zone and
1019 * @mem. This can be the global zone lruvec, if the memory controller
1022 struct lruvec
*mem_cgroup_zone_lruvec(struct zone
*zone
,
1023 struct mem_cgroup
*memcg
)
1025 struct mem_cgroup_per_zone
*mz
;
1027 if (mem_cgroup_disabled())
1028 return &zone
->lruvec
;
1030 mz
= mem_cgroup_zoneinfo(memcg
, zone_to_nid(zone
), zone_idx(zone
));
1035 * Following LRU functions are allowed to be used without PCG_LOCK.
1036 * Operations are called by routine of global LRU independently from memcg.
1037 * What we have to take care of here is validness of pc->mem_cgroup.
1039 * Changes to pc->mem_cgroup happens when
1042 * In typical case, "charge" is done before add-to-lru. Exception is SwapCache.
1043 * It is added to LRU before charge.
1044 * If PCG_USED bit is not set, page_cgroup is not added to this private LRU.
1045 * When moving account, the page is not on LRU. It's isolated.
1049 * mem_cgroup_lru_add_list - account for adding an lru page and return lruvec
1050 * @zone: zone of the page
1054 * This function accounts for @page being added to @lru, and returns
1055 * the lruvec for the given @zone and the memcg @page is charged to.
1057 * The callsite is then responsible for physically linking the page to
1058 * the returned lruvec->lists[@lru].
1060 struct lruvec
*mem_cgroup_lru_add_list(struct zone
*zone
, struct page
*page
,
1063 struct mem_cgroup_per_zone
*mz
;
1064 struct mem_cgroup
*memcg
;
1065 struct page_cgroup
*pc
;
1067 if (mem_cgroup_disabled())
1068 return &zone
->lruvec
;
1070 pc
= lookup_page_cgroup(page
);
1071 memcg
= pc
->mem_cgroup
;
1074 * Surreptitiously switch any uncharged page to root:
1075 * an uncharged page off lru does nothing to secure
1076 * its former mem_cgroup from sudden removal.
1078 * Our caller holds lru_lock, and PageCgroupUsed is updated
1079 * under page_cgroup lock: between them, they make all uses
1080 * of pc->mem_cgroup safe.
1082 if (!PageCgroupUsed(pc
) && memcg
!= root_mem_cgroup
)
1083 pc
->mem_cgroup
= memcg
= root_mem_cgroup
;
1085 mz
= page_cgroup_zoneinfo(memcg
, page
);
1086 /* compound_order() is stabilized through lru_lock */
1087 mz
->lru_size
[lru
] += 1 << compound_order(page
);
1092 * mem_cgroup_lru_del_list - account for removing an lru page
1096 * This function accounts for @page being removed from @lru.
1098 * The callsite is then responsible for physically unlinking
1101 void mem_cgroup_lru_del_list(struct page
*page
, enum lru_list lru
)
1103 struct mem_cgroup_per_zone
*mz
;
1104 struct mem_cgroup
*memcg
;
1105 struct page_cgroup
*pc
;
1107 if (mem_cgroup_disabled())
1110 pc
= lookup_page_cgroup(page
);
1111 memcg
= pc
->mem_cgroup
;
1113 mz
= page_cgroup_zoneinfo(memcg
, page
);
1114 /* huge page split is done under lru_lock. so, we have no races. */
1115 VM_BUG_ON(mz
->lru_size
[lru
] < (1 << compound_order(page
)));
1116 mz
->lru_size
[lru
] -= 1 << compound_order(page
);
1119 void mem_cgroup_lru_del(struct page
*page
)
1121 mem_cgroup_lru_del_list(page
, page_lru(page
));
1125 * mem_cgroup_lru_move_lists - account for moving a page between lrus
1126 * @zone: zone of the page
1128 * @from: current lru
1131 * This function accounts for @page being moved between the lrus @from
1132 * and @to, and returns the lruvec for the given @zone and the memcg
1133 * @page is charged to.
1135 * The callsite is then responsible for physically relinking
1136 * @page->lru to the returned lruvec->lists[@to].
1138 struct lruvec
*mem_cgroup_lru_move_lists(struct zone
*zone
,
1143 /* XXX: Optimize this, especially for @from == @to */
1144 mem_cgroup_lru_del_list(page
, from
);
1145 return mem_cgroup_lru_add_list(zone
, page
, to
);
1149 * Checks whether given mem is same or in the root_mem_cgroup's
1152 bool __mem_cgroup_same_or_subtree(const struct mem_cgroup
*root_memcg
,
1153 struct mem_cgroup
*memcg
)
1155 if (root_memcg
== memcg
)
1157 if (!root_memcg
->use_hierarchy
)
1159 return css_is_ancestor(&memcg
->css
, &root_memcg
->css
);
1162 static bool mem_cgroup_same_or_subtree(const struct mem_cgroup
*root_memcg
,
1163 struct mem_cgroup
*memcg
)
1168 ret
= __mem_cgroup_same_or_subtree(root_memcg
, memcg
);
1173 int task_in_mem_cgroup(struct task_struct
*task
, const struct mem_cgroup
*memcg
)
1176 struct mem_cgroup
*curr
= NULL
;
1177 struct task_struct
*p
;
1179 p
= find_lock_task_mm(task
);
1181 curr
= try_get_mem_cgroup_from_mm(p
->mm
);
1185 * All threads may have already detached their mm's, but the oom
1186 * killer still needs to detect if they have already been oom
1187 * killed to prevent needlessly killing additional tasks.
1190 curr
= mem_cgroup_from_task(task
);
1192 css_get(&curr
->css
);
1198 * We should check use_hierarchy of "memcg" not "curr". Because checking
1199 * use_hierarchy of "curr" here make this function true if hierarchy is
1200 * enabled in "curr" and "curr" is a child of "memcg" in *cgroup*
1201 * hierarchy(even if use_hierarchy is disabled in "memcg").
1203 ret
= mem_cgroup_same_or_subtree(memcg
, curr
);
1204 css_put(&curr
->css
);
1208 int mem_cgroup_inactive_anon_is_low(struct mem_cgroup
*memcg
, struct zone
*zone
)
1210 unsigned long inactive_ratio
;
1211 int nid
= zone_to_nid(zone
);
1212 int zid
= zone_idx(zone
);
1213 unsigned long inactive
;
1214 unsigned long active
;
1217 inactive
= mem_cgroup_zone_nr_lru_pages(memcg
, nid
, zid
,
1218 BIT(LRU_INACTIVE_ANON
));
1219 active
= mem_cgroup_zone_nr_lru_pages(memcg
, nid
, zid
,
1220 BIT(LRU_ACTIVE_ANON
));
1222 gb
= (inactive
+ active
) >> (30 - PAGE_SHIFT
);
1224 inactive_ratio
= int_sqrt(10 * gb
);
1228 return inactive
* inactive_ratio
< active
;
1231 int mem_cgroup_inactive_file_is_low(struct mem_cgroup
*memcg
, struct zone
*zone
)
1233 unsigned long active
;
1234 unsigned long inactive
;
1235 int zid
= zone_idx(zone
);
1236 int nid
= zone_to_nid(zone
);
1238 inactive
= mem_cgroup_zone_nr_lru_pages(memcg
, nid
, zid
,
1239 BIT(LRU_INACTIVE_FILE
));
1240 active
= mem_cgroup_zone_nr_lru_pages(memcg
, nid
, zid
,
1241 BIT(LRU_ACTIVE_FILE
));
1243 return (active
> inactive
);
1246 struct zone_reclaim_stat
*mem_cgroup_get_reclaim_stat(struct mem_cgroup
*memcg
,
1249 int nid
= zone_to_nid(zone
);
1250 int zid
= zone_idx(zone
);
1251 struct mem_cgroup_per_zone
*mz
= mem_cgroup_zoneinfo(memcg
, nid
, zid
);
1253 return &mz
->reclaim_stat
;
1256 struct zone_reclaim_stat
*
1257 mem_cgroup_get_reclaim_stat_from_page(struct page
*page
)
1259 struct page_cgroup
*pc
;
1260 struct mem_cgroup_per_zone
*mz
;
1262 if (mem_cgroup_disabled())
1265 pc
= lookup_page_cgroup(page
);
1266 if (!PageCgroupUsed(pc
))
1268 /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */
1270 mz
= page_cgroup_zoneinfo(pc
->mem_cgroup
, page
);
1271 return &mz
->reclaim_stat
;
1274 #define mem_cgroup_from_res_counter(counter, member) \
1275 container_of(counter, struct mem_cgroup, member)
1278 * mem_cgroup_margin - calculate chargeable space of a memory cgroup
1279 * @mem: the memory cgroup
1281 * Returns the maximum amount of memory @mem can be charged with, in
1284 static unsigned long mem_cgroup_margin(struct mem_cgroup
*memcg
)
1286 unsigned long long margin
;
1288 margin
= res_counter_margin(&memcg
->res
);
1289 if (do_swap_account
)
1290 margin
= min(margin
, res_counter_margin(&memcg
->memsw
));
1291 return margin
>> PAGE_SHIFT
;
1294 int mem_cgroup_swappiness(struct mem_cgroup
*memcg
)
1296 struct cgroup
*cgrp
= memcg
->css
.cgroup
;
1299 if (cgrp
->parent
== NULL
)
1300 return vm_swappiness
;
1302 return memcg
->swappiness
;
1306 * memcg->moving_account is used for checking possibility that some thread is
1307 * calling move_account(). When a thread on CPU-A starts moving pages under
1308 * a memcg, other threads should check memcg->moving_account under
1309 * rcu_read_lock(), like this:
1313 * memcg->moving_account+1 if (memcg->mocing_account)
1315 * synchronize_rcu() update something.
1320 /* for quick checking without looking up memcg */
1321 atomic_t memcg_moving __read_mostly
;
1323 static void mem_cgroup_start_move(struct mem_cgroup
*memcg
)
1325 atomic_inc(&memcg_moving
);
1326 atomic_inc(&memcg
->moving_account
);
1330 static void mem_cgroup_end_move(struct mem_cgroup
*memcg
)
1333 * Now, mem_cgroup_clear_mc() may call this function with NULL.
1334 * We check NULL in callee rather than caller.
1337 atomic_dec(&memcg_moving
);
1338 atomic_dec(&memcg
->moving_account
);
1343 * 2 routines for checking "mem" is under move_account() or not.
1345 * mem_cgroup_stolen() - checking whether a cgroup is mc.from or not. This
1346 * is used for avoiding races in accounting. If true,
1347 * pc->mem_cgroup may be overwritten.
1349 * mem_cgroup_under_move() - checking a cgroup is mc.from or mc.to or
1350 * under hierarchy of moving cgroups. This is for
1351 * waiting at hith-memory prressure caused by "move".
1354 static bool mem_cgroup_stolen(struct mem_cgroup
*memcg
)
1356 VM_BUG_ON(!rcu_read_lock_held());
1357 return atomic_read(&memcg
->moving_account
) > 0;
1360 static bool mem_cgroup_under_move(struct mem_cgroup
*memcg
)
1362 struct mem_cgroup
*from
;
1363 struct mem_cgroup
*to
;
1366 * Unlike task_move routines, we access mc.to, mc.from not under
1367 * mutual exclusion by cgroup_mutex. Here, we take spinlock instead.
1369 spin_lock(&mc
.lock
);
1375 ret
= mem_cgroup_same_or_subtree(memcg
, from
)
1376 || mem_cgroup_same_or_subtree(memcg
, to
);
1378 spin_unlock(&mc
.lock
);
1382 static bool mem_cgroup_wait_acct_move(struct mem_cgroup
*memcg
)
1384 if (mc
.moving_task
&& current
!= mc
.moving_task
) {
1385 if (mem_cgroup_under_move(memcg
)) {
1387 prepare_to_wait(&mc
.waitq
, &wait
, TASK_INTERRUPTIBLE
);
1388 /* moving charge context might have finished. */
1391 finish_wait(&mc
.waitq
, &wait
);
1399 * Take this lock when
1400 * - a code tries to modify page's memcg while it's USED.
1401 * - a code tries to modify page state accounting in a memcg.
1402 * see mem_cgroup_stolen(), too.
1404 static void move_lock_mem_cgroup(struct mem_cgroup
*memcg
,
1405 unsigned long *flags
)
1407 spin_lock_irqsave(&memcg
->move_lock
, *flags
);
1410 static void move_unlock_mem_cgroup(struct mem_cgroup
*memcg
,
1411 unsigned long *flags
)
1413 spin_unlock_irqrestore(&memcg
->move_lock
, *flags
);
1417 * mem_cgroup_print_oom_info: Called from OOM with tasklist_lock held in read mode.
1418 * @memcg: The memory cgroup that went over limit
1419 * @p: Task that is going to be killed
1421 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1424 void mem_cgroup_print_oom_info(struct mem_cgroup
*memcg
, struct task_struct
*p
)
1426 struct cgroup
*task_cgrp
;
1427 struct cgroup
*mem_cgrp
;
1429 * Need a buffer in BSS, can't rely on allocations. The code relies
1430 * on the assumption that OOM is serialized for memory controller.
1431 * If this assumption is broken, revisit this code.
1433 static char memcg_name
[PATH_MAX
];
1441 mem_cgrp
= memcg
->css
.cgroup
;
1442 task_cgrp
= task_cgroup(p
, mem_cgroup_subsys_id
);
1444 ret
= cgroup_path(task_cgrp
, memcg_name
, PATH_MAX
);
1447 * Unfortunately, we are unable to convert to a useful name
1448 * But we'll still print out the usage information
1455 printk(KERN_INFO
"Task in %s killed", memcg_name
);
1458 ret
= cgroup_path(mem_cgrp
, memcg_name
, PATH_MAX
);
1466 * Continues from above, so we don't need an KERN_ level
1468 printk(KERN_CONT
" as a result of limit of %s\n", memcg_name
);
1471 printk(KERN_INFO
"memory: usage %llukB, limit %llukB, failcnt %llu\n",
1472 res_counter_read_u64(&memcg
->res
, RES_USAGE
) >> 10,
1473 res_counter_read_u64(&memcg
->res
, RES_LIMIT
) >> 10,
1474 res_counter_read_u64(&memcg
->res
, RES_FAILCNT
));
1475 printk(KERN_INFO
"memory+swap: usage %llukB, limit %llukB, "
1477 res_counter_read_u64(&memcg
->memsw
, RES_USAGE
) >> 10,
1478 res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
) >> 10,
1479 res_counter_read_u64(&memcg
->memsw
, RES_FAILCNT
));
1483 * This function returns the number of memcg under hierarchy tree. Returns
1484 * 1(self count) if no children.
1486 static int mem_cgroup_count_children(struct mem_cgroup
*memcg
)
1489 struct mem_cgroup
*iter
;
1491 for_each_mem_cgroup_tree(iter
, memcg
)
1497 * Return the memory (and swap, if configured) limit for a memcg.
1499 u64
mem_cgroup_get_limit(struct mem_cgroup
*memcg
)
1504 limit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
1505 limit
+= total_swap_pages
<< PAGE_SHIFT
;
1507 memsw
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
1509 * If memsw is finite and limits the amount of swap space available
1510 * to this memcg, return that limit.
1512 return min(limit
, memsw
);
1515 static unsigned long mem_cgroup_reclaim(struct mem_cgroup
*memcg
,
1517 unsigned long flags
)
1519 unsigned long total
= 0;
1520 bool noswap
= false;
1523 if (flags
& MEM_CGROUP_RECLAIM_NOSWAP
)
1525 if (!(flags
& MEM_CGROUP_RECLAIM_SHRINK
) && memcg
->memsw_is_minimum
)
1528 for (loop
= 0; loop
< MEM_CGROUP_MAX_RECLAIM_LOOPS
; loop
++) {
1530 drain_all_stock_async(memcg
);
1531 total
+= try_to_free_mem_cgroup_pages(memcg
, gfp_mask
, noswap
);
1533 * Allow limit shrinkers, which are triggered directly
1534 * by userspace, to catch signals and stop reclaim
1535 * after minimal progress, regardless of the margin.
1537 if (total
&& (flags
& MEM_CGROUP_RECLAIM_SHRINK
))
1539 if (mem_cgroup_margin(memcg
))
1542 * If nothing was reclaimed after two attempts, there
1543 * may be no reclaimable pages in this hierarchy.
1552 * test_mem_cgroup_node_reclaimable
1553 * @mem: the target memcg
1554 * @nid: the node ID to be checked.
1555 * @noswap : specify true here if the user wants flle only information.
1557 * This function returns whether the specified memcg contains any
1558 * reclaimable pages on a node. Returns true if there are any reclaimable
1559 * pages in the node.
1561 static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup
*memcg
,
1562 int nid
, bool noswap
)
1564 if (mem_cgroup_node_nr_lru_pages(memcg
, nid
, LRU_ALL_FILE
))
1566 if (noswap
|| !total_swap_pages
)
1568 if (mem_cgroup_node_nr_lru_pages(memcg
, nid
, LRU_ALL_ANON
))
1573 #if MAX_NUMNODES > 1
1576 * Always updating the nodemask is not very good - even if we have an empty
1577 * list or the wrong list here, we can start from some node and traverse all
1578 * nodes based on the zonelist. So update the list loosely once per 10 secs.
1581 static void mem_cgroup_may_update_nodemask(struct mem_cgroup
*memcg
)
1585 * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET
1586 * pagein/pageout changes since the last update.
1588 if (!atomic_read(&memcg
->numainfo_events
))
1590 if (atomic_inc_return(&memcg
->numainfo_updating
) > 1)
1593 /* make a nodemask where this memcg uses memory from */
1594 memcg
->scan_nodes
= node_states
[N_HIGH_MEMORY
];
1596 for_each_node_mask(nid
, node_states
[N_HIGH_MEMORY
]) {
1598 if (!test_mem_cgroup_node_reclaimable(memcg
, nid
, false))
1599 node_clear(nid
, memcg
->scan_nodes
);
1602 atomic_set(&memcg
->numainfo_events
, 0);
1603 atomic_set(&memcg
->numainfo_updating
, 0);
1607 * Selecting a node where we start reclaim from. Because what we need is just
1608 * reducing usage counter, start from anywhere is O,K. Considering
1609 * memory reclaim from current node, there are pros. and cons.
1611 * Freeing memory from current node means freeing memory from a node which
1612 * we'll use or we've used. So, it may make LRU bad. And if several threads
1613 * hit limits, it will see a contention on a node. But freeing from remote
1614 * node means more costs for memory reclaim because of memory latency.
1616 * Now, we use round-robin. Better algorithm is welcomed.
1618 int mem_cgroup_select_victim_node(struct mem_cgroup
*memcg
)
1622 mem_cgroup_may_update_nodemask(memcg
);
1623 node
= memcg
->last_scanned_node
;
1625 node
= next_node(node
, memcg
->scan_nodes
);
1626 if (node
== MAX_NUMNODES
)
1627 node
= first_node(memcg
->scan_nodes
);
1629 * We call this when we hit limit, not when pages are added to LRU.
1630 * No LRU may hold pages because all pages are UNEVICTABLE or
1631 * memcg is too small and all pages are not on LRU. In that case,
1632 * we use curret node.
1634 if (unlikely(node
== MAX_NUMNODES
))
1635 node
= numa_node_id();
1637 memcg
->last_scanned_node
= node
;
1642 * Check all nodes whether it contains reclaimable pages or not.
1643 * For quick scan, we make use of scan_nodes. This will allow us to skip
1644 * unused nodes. But scan_nodes is lazily updated and may not cotain
1645 * enough new information. We need to do double check.
1647 bool mem_cgroup_reclaimable(struct mem_cgroup
*memcg
, bool noswap
)
1652 * quick check...making use of scan_node.
1653 * We can skip unused nodes.
1655 if (!nodes_empty(memcg
->scan_nodes
)) {
1656 for (nid
= first_node(memcg
->scan_nodes
);
1658 nid
= next_node(nid
, memcg
->scan_nodes
)) {
1660 if (test_mem_cgroup_node_reclaimable(memcg
, nid
, noswap
))
1665 * Check rest of nodes.
1667 for_each_node_state(nid
, N_HIGH_MEMORY
) {
1668 if (node_isset(nid
, memcg
->scan_nodes
))
1670 if (test_mem_cgroup_node_reclaimable(memcg
, nid
, noswap
))
1677 int mem_cgroup_select_victim_node(struct mem_cgroup
*memcg
)
1682 bool mem_cgroup_reclaimable(struct mem_cgroup
*memcg
, bool noswap
)
1684 return test_mem_cgroup_node_reclaimable(memcg
, 0, noswap
);
1688 static int mem_cgroup_soft_reclaim(struct mem_cgroup
*root_memcg
,
1691 unsigned long *total_scanned
)
1693 struct mem_cgroup
*victim
= NULL
;
1696 unsigned long excess
;
1697 unsigned long nr_scanned
;
1698 struct mem_cgroup_reclaim_cookie reclaim
= {
1703 excess
= res_counter_soft_limit_excess(&root_memcg
->res
) >> PAGE_SHIFT
;
1706 victim
= mem_cgroup_iter(root_memcg
, victim
, &reclaim
);
1711 * If we have not been able to reclaim
1712 * anything, it might because there are
1713 * no reclaimable pages under this hierarchy
1718 * We want to do more targeted reclaim.
1719 * excess >> 2 is not to excessive so as to
1720 * reclaim too much, nor too less that we keep
1721 * coming back to reclaim from this cgroup
1723 if (total
>= (excess
>> 2) ||
1724 (loop
> MEM_CGROUP_MAX_RECLAIM_LOOPS
))
1729 if (!mem_cgroup_reclaimable(victim
, false))
1731 total
+= mem_cgroup_shrink_node_zone(victim
, gfp_mask
, false,
1733 *total_scanned
+= nr_scanned
;
1734 if (!res_counter_soft_limit_excess(&root_memcg
->res
))
1737 mem_cgroup_iter_break(root_memcg
, victim
);
1742 * Check OOM-Killer is already running under our hierarchy.
1743 * If someone is running, return false.
1744 * Has to be called with memcg_oom_lock
1746 static bool mem_cgroup_oom_lock(struct mem_cgroup
*memcg
)
1748 struct mem_cgroup
*iter
, *failed
= NULL
;
1750 for_each_mem_cgroup_tree(iter
, memcg
) {
1751 if (iter
->oom_lock
) {
1753 * this subtree of our hierarchy is already locked
1754 * so we cannot give a lock.
1757 mem_cgroup_iter_break(memcg
, iter
);
1760 iter
->oom_lock
= true;
1767 * OK, we failed to lock the whole subtree so we have to clean up
1768 * what we set up to the failing subtree
1770 for_each_mem_cgroup_tree(iter
, memcg
) {
1771 if (iter
== failed
) {
1772 mem_cgroup_iter_break(memcg
, iter
);
1775 iter
->oom_lock
= false;
1781 * Has to be called with memcg_oom_lock
1783 static int mem_cgroup_oom_unlock(struct mem_cgroup
*memcg
)
1785 struct mem_cgroup
*iter
;
1787 for_each_mem_cgroup_tree(iter
, memcg
)
1788 iter
->oom_lock
= false;
1792 static void mem_cgroup_mark_under_oom(struct mem_cgroup
*memcg
)
1794 struct mem_cgroup
*iter
;
1796 for_each_mem_cgroup_tree(iter
, memcg
)
1797 atomic_inc(&iter
->under_oom
);
1800 static void mem_cgroup_unmark_under_oom(struct mem_cgroup
*memcg
)
1802 struct mem_cgroup
*iter
;
1805 * When a new child is created while the hierarchy is under oom,
1806 * mem_cgroup_oom_lock() may not be called. We have to use
1807 * atomic_add_unless() here.
1809 for_each_mem_cgroup_tree(iter
, memcg
)
1810 atomic_add_unless(&iter
->under_oom
, -1, 0);
1813 static DEFINE_SPINLOCK(memcg_oom_lock
);
1814 static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq
);
1816 struct oom_wait_info
{
1817 struct mem_cgroup
*memcg
;
1821 static int memcg_oom_wake_function(wait_queue_t
*wait
,
1822 unsigned mode
, int sync
, void *arg
)
1824 struct mem_cgroup
*wake_memcg
= (struct mem_cgroup
*)arg
;
1825 struct mem_cgroup
*oom_wait_memcg
;
1826 struct oom_wait_info
*oom_wait_info
;
1828 oom_wait_info
= container_of(wait
, struct oom_wait_info
, wait
);
1829 oom_wait_memcg
= oom_wait_info
->memcg
;
1832 * Both of oom_wait_info->memcg and wake_memcg are stable under us.
1833 * Then we can use css_is_ancestor without taking care of RCU.
1835 if (!mem_cgroup_same_or_subtree(oom_wait_memcg
, wake_memcg
)
1836 && !mem_cgroup_same_or_subtree(wake_memcg
, oom_wait_memcg
))
1838 return autoremove_wake_function(wait
, mode
, sync
, arg
);
1841 static void memcg_wakeup_oom(struct mem_cgroup
*memcg
)
1843 /* for filtering, pass "memcg" as argument. */
1844 __wake_up(&memcg_oom_waitq
, TASK_NORMAL
, 0, memcg
);
1847 static void memcg_oom_recover(struct mem_cgroup
*memcg
)
1849 if (memcg
&& atomic_read(&memcg
->under_oom
))
1850 memcg_wakeup_oom(memcg
);
1854 * try to call OOM killer. returns false if we should exit memory-reclaim loop.
1856 bool mem_cgroup_handle_oom(struct mem_cgroup
*memcg
, gfp_t mask
, int order
)
1858 struct oom_wait_info owait
;
1859 bool locked
, need_to_kill
;
1861 owait
.memcg
= memcg
;
1862 owait
.wait
.flags
= 0;
1863 owait
.wait
.func
= memcg_oom_wake_function
;
1864 owait
.wait
.private = current
;
1865 INIT_LIST_HEAD(&owait
.wait
.task_list
);
1866 need_to_kill
= true;
1867 mem_cgroup_mark_under_oom(memcg
);
1869 /* At first, try to OOM lock hierarchy under memcg.*/
1870 spin_lock(&memcg_oom_lock
);
1871 locked
= mem_cgroup_oom_lock(memcg
);
1873 * Even if signal_pending(), we can't quit charge() loop without
1874 * accounting. So, UNINTERRUPTIBLE is appropriate. But SIGKILL
1875 * under OOM is always welcomed, use TASK_KILLABLE here.
1877 prepare_to_wait(&memcg_oom_waitq
, &owait
.wait
, TASK_KILLABLE
);
1878 if (!locked
|| memcg
->oom_kill_disable
)
1879 need_to_kill
= false;
1881 mem_cgroup_oom_notify(memcg
);
1882 spin_unlock(&memcg_oom_lock
);
1885 finish_wait(&memcg_oom_waitq
, &owait
.wait
);
1886 mem_cgroup_out_of_memory(memcg
, mask
, order
);
1889 finish_wait(&memcg_oom_waitq
, &owait
.wait
);
1891 spin_lock(&memcg_oom_lock
);
1893 mem_cgroup_oom_unlock(memcg
);
1894 memcg_wakeup_oom(memcg
);
1895 spin_unlock(&memcg_oom_lock
);
1897 mem_cgroup_unmark_under_oom(memcg
);
1899 if (test_thread_flag(TIF_MEMDIE
) || fatal_signal_pending(current
))
1901 /* Give chance to dying process */
1902 schedule_timeout_uninterruptible(1);
1907 * Currently used to update mapped file statistics, but the routine can be
1908 * generalized to update other statistics as well.
1910 * Notes: Race condition
1912 * We usually use page_cgroup_lock() for accessing page_cgroup member but
1913 * it tends to be costly. But considering some conditions, we doesn't need
1914 * to do so _always_.
1916 * Considering "charge", lock_page_cgroup() is not required because all
1917 * file-stat operations happen after a page is attached to radix-tree. There
1918 * are no race with "charge".
1920 * Considering "uncharge", we know that memcg doesn't clear pc->mem_cgroup
1921 * at "uncharge" intentionally. So, we always see valid pc->mem_cgroup even
1922 * if there are race with "uncharge". Statistics itself is properly handled
1925 * Considering "move", this is an only case we see a race. To make the race
1926 * small, we check mm->moving_account and detect there are possibility of race
1927 * If there is, we take a lock.
1930 void __mem_cgroup_begin_update_page_stat(struct page
*page
,
1931 bool *locked
, unsigned long *flags
)
1933 struct mem_cgroup
*memcg
;
1934 struct page_cgroup
*pc
;
1936 pc
= lookup_page_cgroup(page
);
1938 memcg
= pc
->mem_cgroup
;
1939 if (unlikely(!memcg
|| !PageCgroupUsed(pc
)))
1942 * If this memory cgroup is not under account moving, we don't
1943 * need to take move_lock_page_cgroup(). Because we already hold
1944 * rcu_read_lock(), any calls to move_account will be delayed until
1945 * rcu_read_unlock() if mem_cgroup_stolen() == true.
1947 if (!mem_cgroup_stolen(memcg
))
1950 move_lock_mem_cgroup(memcg
, flags
);
1951 if (memcg
!= pc
->mem_cgroup
|| !PageCgroupUsed(pc
)) {
1952 move_unlock_mem_cgroup(memcg
, flags
);
1958 void __mem_cgroup_end_update_page_stat(struct page
*page
, unsigned long *flags
)
1960 struct page_cgroup
*pc
= lookup_page_cgroup(page
);
1963 * It's guaranteed that pc->mem_cgroup never changes while
1964 * lock is held because a routine modifies pc->mem_cgroup
1965 * should take move_lock_page_cgroup().
1967 move_unlock_mem_cgroup(pc
->mem_cgroup
, flags
);
1970 void mem_cgroup_update_page_stat(struct page
*page
,
1971 enum mem_cgroup_page_stat_item idx
, int val
)
1973 struct mem_cgroup
*memcg
;
1974 struct page_cgroup
*pc
= lookup_page_cgroup(page
);
1975 unsigned long uninitialized_var(flags
);
1977 if (mem_cgroup_disabled())
1980 memcg
= pc
->mem_cgroup
;
1981 if (unlikely(!memcg
|| !PageCgroupUsed(pc
)))
1985 case MEMCG_NR_FILE_MAPPED
:
1986 idx
= MEM_CGROUP_STAT_FILE_MAPPED
;
1992 this_cpu_add(memcg
->stat
->count
[idx
], val
);
1996 * size of first charge trial. "32" comes from vmscan.c's magic value.
1997 * TODO: maybe necessary to use big numbers in big irons.
1999 #define CHARGE_BATCH 32U
2000 struct memcg_stock_pcp
{
2001 struct mem_cgroup
*cached
; /* this never be root cgroup */
2002 unsigned int nr_pages
;
2003 struct work_struct work
;
2004 unsigned long flags
;
2005 #define FLUSHING_CACHED_CHARGE (0)
2007 static DEFINE_PER_CPU(struct memcg_stock_pcp
, memcg_stock
);
2008 static DEFINE_MUTEX(percpu_charge_mutex
);
2011 * Try to consume stocked charge on this cpu. If success, one page is consumed
2012 * from local stock and true is returned. If the stock is 0 or charges from a
2013 * cgroup which is not current target, returns false. This stock will be
2016 static bool consume_stock(struct mem_cgroup
*memcg
)
2018 struct memcg_stock_pcp
*stock
;
2021 stock
= &get_cpu_var(memcg_stock
);
2022 if (memcg
== stock
->cached
&& stock
->nr_pages
)
2024 else /* need to call res_counter_charge */
2026 put_cpu_var(memcg_stock
);
2031 * Returns stocks cached in percpu to res_counter and reset cached information.
2033 static void drain_stock(struct memcg_stock_pcp
*stock
)
2035 struct mem_cgroup
*old
= stock
->cached
;
2037 if (stock
->nr_pages
) {
2038 unsigned long bytes
= stock
->nr_pages
* PAGE_SIZE
;
2040 res_counter_uncharge(&old
->res
, bytes
);
2041 if (do_swap_account
)
2042 res_counter_uncharge(&old
->memsw
, bytes
);
2043 stock
->nr_pages
= 0;
2045 stock
->cached
= NULL
;
2049 * This must be called under preempt disabled or must be called by
2050 * a thread which is pinned to local cpu.
2052 static void drain_local_stock(struct work_struct
*dummy
)
2054 struct memcg_stock_pcp
*stock
= &__get_cpu_var(memcg_stock
);
2056 clear_bit(FLUSHING_CACHED_CHARGE
, &stock
->flags
);
2060 * Cache charges(val) which is from res_counter, to local per_cpu area.
2061 * This will be consumed by consume_stock() function, later.
2063 static void refill_stock(struct mem_cgroup
*memcg
, unsigned int nr_pages
)
2065 struct memcg_stock_pcp
*stock
= &get_cpu_var(memcg_stock
);
2067 if (stock
->cached
!= memcg
) { /* reset if necessary */
2069 stock
->cached
= memcg
;
2071 stock
->nr_pages
+= nr_pages
;
2072 put_cpu_var(memcg_stock
);
2076 * Drains all per-CPU charge caches for given root_memcg resp. subtree
2077 * of the hierarchy under it. sync flag says whether we should block
2078 * until the work is done.
2080 static void drain_all_stock(struct mem_cgroup
*root_memcg
, bool sync
)
2084 /* Notify other cpus that system-wide "drain" is running */
2087 for_each_online_cpu(cpu
) {
2088 struct memcg_stock_pcp
*stock
= &per_cpu(memcg_stock
, cpu
);
2089 struct mem_cgroup
*memcg
;
2091 memcg
= stock
->cached
;
2092 if (!memcg
|| !stock
->nr_pages
)
2094 if (!mem_cgroup_same_or_subtree(root_memcg
, memcg
))
2096 if (!test_and_set_bit(FLUSHING_CACHED_CHARGE
, &stock
->flags
)) {
2098 drain_local_stock(&stock
->work
);
2100 schedule_work_on(cpu
, &stock
->work
);
2108 for_each_online_cpu(cpu
) {
2109 struct memcg_stock_pcp
*stock
= &per_cpu(memcg_stock
, cpu
);
2110 if (test_bit(FLUSHING_CACHED_CHARGE
, &stock
->flags
))
2111 flush_work(&stock
->work
);
2118 * Tries to drain stocked charges in other cpus. This function is asynchronous
2119 * and just put a work per cpu for draining localy on each cpu. Caller can
2120 * expects some charges will be back to res_counter later but cannot wait for
2123 static void drain_all_stock_async(struct mem_cgroup
*root_memcg
)
2126 * If someone calls draining, avoid adding more kworker runs.
2128 if (!mutex_trylock(&percpu_charge_mutex
))
2130 drain_all_stock(root_memcg
, false);
2131 mutex_unlock(&percpu_charge_mutex
);
2134 /* This is a synchronous drain interface. */
2135 static void drain_all_stock_sync(struct mem_cgroup
*root_memcg
)
2137 /* called when force_empty is called */
2138 mutex_lock(&percpu_charge_mutex
);
2139 drain_all_stock(root_memcg
, true);
2140 mutex_unlock(&percpu_charge_mutex
);
2144 * This function drains percpu counter value from DEAD cpu and
2145 * move it to local cpu. Note that this function can be preempted.
2147 static void mem_cgroup_drain_pcp_counter(struct mem_cgroup
*memcg
, int cpu
)
2151 spin_lock(&memcg
->pcp_counter_lock
);
2152 for (i
= 0; i
< MEM_CGROUP_STAT_DATA
; i
++) {
2153 long x
= per_cpu(memcg
->stat
->count
[i
], cpu
);
2155 per_cpu(memcg
->stat
->count
[i
], cpu
) = 0;
2156 memcg
->nocpu_base
.count
[i
] += x
;
2158 for (i
= 0; i
< MEM_CGROUP_EVENTS_NSTATS
; i
++) {
2159 unsigned long x
= per_cpu(memcg
->stat
->events
[i
], cpu
);
2161 per_cpu(memcg
->stat
->events
[i
], cpu
) = 0;
2162 memcg
->nocpu_base
.events
[i
] += x
;
2164 spin_unlock(&memcg
->pcp_counter_lock
);
2167 static int __cpuinit
memcg_cpu_hotplug_callback(struct notifier_block
*nb
,
2168 unsigned long action
,
2171 int cpu
= (unsigned long)hcpu
;
2172 struct memcg_stock_pcp
*stock
;
2173 struct mem_cgroup
*iter
;
2175 if (action
== CPU_ONLINE
)
2178 if (action
!= CPU_DEAD
&& action
!= CPU_DEAD_FROZEN
)
2181 for_each_mem_cgroup(iter
)
2182 mem_cgroup_drain_pcp_counter(iter
, cpu
);
2184 stock
= &per_cpu(memcg_stock
, cpu
);
2190 /* See __mem_cgroup_try_charge() for details */
2192 CHARGE_OK
, /* success */
2193 CHARGE_RETRY
, /* need to retry but retry is not bad */
2194 CHARGE_NOMEM
, /* we can't do more. return -ENOMEM */
2195 CHARGE_WOULDBLOCK
, /* GFP_WAIT wasn't set and no enough res. */
2196 CHARGE_OOM_DIE
, /* the current is killed because of OOM */
2199 static int mem_cgroup_do_charge(struct mem_cgroup
*memcg
, gfp_t gfp_mask
,
2200 unsigned int nr_pages
, bool oom_check
)
2202 unsigned long csize
= nr_pages
* PAGE_SIZE
;
2203 struct mem_cgroup
*mem_over_limit
;
2204 struct res_counter
*fail_res
;
2205 unsigned long flags
= 0;
2208 ret
= res_counter_charge(&memcg
->res
, csize
, &fail_res
);
2211 if (!do_swap_account
)
2213 ret
= res_counter_charge(&memcg
->memsw
, csize
, &fail_res
);
2217 res_counter_uncharge(&memcg
->res
, csize
);
2218 mem_over_limit
= mem_cgroup_from_res_counter(fail_res
, memsw
);
2219 flags
|= MEM_CGROUP_RECLAIM_NOSWAP
;
2221 mem_over_limit
= mem_cgroup_from_res_counter(fail_res
, res
);
2223 * nr_pages can be either a huge page (HPAGE_PMD_NR), a batch
2224 * of regular pages (CHARGE_BATCH), or a single regular page (1).
2226 * Never reclaim on behalf of optional batching, retry with a
2227 * single page instead.
2229 if (nr_pages
== CHARGE_BATCH
)
2230 return CHARGE_RETRY
;
2232 if (!(gfp_mask
& __GFP_WAIT
))
2233 return CHARGE_WOULDBLOCK
;
2235 ret
= mem_cgroup_reclaim(mem_over_limit
, gfp_mask
, flags
);
2236 if (mem_cgroup_margin(mem_over_limit
) >= nr_pages
)
2237 return CHARGE_RETRY
;
2239 * Even though the limit is exceeded at this point, reclaim
2240 * may have been able to free some pages. Retry the charge
2241 * before killing the task.
2243 * Only for regular pages, though: huge pages are rather
2244 * unlikely to succeed so close to the limit, and we fall back
2245 * to regular pages anyway in case of failure.
2247 if (nr_pages
== 1 && ret
)
2248 return CHARGE_RETRY
;
2251 * At task move, charge accounts can be doubly counted. So, it's
2252 * better to wait until the end of task_move if something is going on.
2254 if (mem_cgroup_wait_acct_move(mem_over_limit
))
2255 return CHARGE_RETRY
;
2257 /* If we don't need to call oom-killer at el, return immediately */
2259 return CHARGE_NOMEM
;
2261 if (!mem_cgroup_handle_oom(mem_over_limit
, gfp_mask
, get_order(csize
)))
2262 return CHARGE_OOM_DIE
;
2264 return CHARGE_RETRY
;
2268 * __mem_cgroup_try_charge() does
2269 * 1. detect memcg to be charged against from passed *mm and *ptr,
2270 * 2. update res_counter
2271 * 3. call memory reclaim if necessary.
2273 * In some special case, if the task is fatal, fatal_signal_pending() or
2274 * has TIF_MEMDIE, this function returns -EINTR while writing root_mem_cgroup
2275 * to *ptr. There are two reasons for this. 1: fatal threads should quit as soon
2276 * as possible without any hazards. 2: all pages should have a valid
2277 * pc->mem_cgroup. If mm is NULL and the caller doesn't pass a valid memcg
2278 * pointer, that is treated as a charge to root_mem_cgroup.
2280 * So __mem_cgroup_try_charge() will return
2281 * 0 ... on success, filling *ptr with a valid memcg pointer.
2282 * -ENOMEM ... charge failure because of resource limits.
2283 * -EINTR ... if thread is fatal. *ptr is filled with root_mem_cgroup.
2285 * Unlike the exported interface, an "oom" parameter is added. if oom==true,
2286 * the oom-killer can be invoked.
2288 static int __mem_cgroup_try_charge(struct mm_struct
*mm
,
2290 unsigned int nr_pages
,
2291 struct mem_cgroup
**ptr
,
2294 unsigned int batch
= max(CHARGE_BATCH
, nr_pages
);
2295 int nr_oom_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
2296 struct mem_cgroup
*memcg
= NULL
;
2300 * Unlike gloval-vm's OOM-kill, we're not in memory shortage
2301 * in system level. So, allow to go ahead dying process in addition to
2304 if (unlikely(test_thread_flag(TIF_MEMDIE
)
2305 || fatal_signal_pending(current
)))
2309 * We always charge the cgroup the mm_struct belongs to.
2310 * The mm_struct's mem_cgroup changes on task migration if the
2311 * thread group leader migrates. It's possible that mm is not
2312 * set, if so charge the init_mm (happens for pagecache usage).
2315 *ptr
= root_mem_cgroup
;
2317 if (*ptr
) { /* css should be a valid one */
2319 VM_BUG_ON(css_is_removed(&memcg
->css
));
2320 if (mem_cgroup_is_root(memcg
))
2322 if (nr_pages
== 1 && consume_stock(memcg
))
2324 css_get(&memcg
->css
);
2326 struct task_struct
*p
;
2329 p
= rcu_dereference(mm
->owner
);
2331 * Because we don't have task_lock(), "p" can exit.
2332 * In that case, "memcg" can point to root or p can be NULL with
2333 * race with swapoff. Then, we have small risk of mis-accouning.
2334 * But such kind of mis-account by race always happens because
2335 * we don't have cgroup_mutex(). It's overkill and we allo that
2337 * (*) swapoff at el will charge against mm-struct not against
2338 * task-struct. So, mm->owner can be NULL.
2340 memcg
= mem_cgroup_from_task(p
);
2342 memcg
= root_mem_cgroup
;
2343 if (mem_cgroup_is_root(memcg
)) {
2347 if (nr_pages
== 1 && consume_stock(memcg
)) {
2349 * It seems dagerous to access memcg without css_get().
2350 * But considering how consume_stok works, it's not
2351 * necessary. If consume_stock success, some charges
2352 * from this memcg are cached on this cpu. So, we
2353 * don't need to call css_get()/css_tryget() before
2354 * calling consume_stock().
2359 /* after here, we may be blocked. we need to get refcnt */
2360 if (!css_tryget(&memcg
->css
)) {
2370 /* If killed, bypass charge */
2371 if (fatal_signal_pending(current
)) {
2372 css_put(&memcg
->css
);
2377 if (oom
&& !nr_oom_retries
) {
2379 nr_oom_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
2382 ret
= mem_cgroup_do_charge(memcg
, gfp_mask
, batch
, oom_check
);
2386 case CHARGE_RETRY
: /* not in OOM situation but retry */
2388 css_put(&memcg
->css
);
2391 case CHARGE_WOULDBLOCK
: /* !__GFP_WAIT */
2392 css_put(&memcg
->css
);
2394 case CHARGE_NOMEM
: /* OOM routine works */
2396 css_put(&memcg
->css
);
2399 /* If oom, we never return -ENOMEM */
2402 case CHARGE_OOM_DIE
: /* Killed by OOM Killer */
2403 css_put(&memcg
->css
);
2406 } while (ret
!= CHARGE_OK
);
2408 if (batch
> nr_pages
)
2409 refill_stock(memcg
, batch
- nr_pages
);
2410 css_put(&memcg
->css
);
2418 *ptr
= root_mem_cgroup
;
2423 * Somemtimes we have to undo a charge we got by try_charge().
2424 * This function is for that and do uncharge, put css's refcnt.
2425 * gotten by try_charge().
2427 static void __mem_cgroup_cancel_charge(struct mem_cgroup
*memcg
,
2428 unsigned int nr_pages
)
2430 if (!mem_cgroup_is_root(memcg
)) {
2431 unsigned long bytes
= nr_pages
* PAGE_SIZE
;
2433 res_counter_uncharge(&memcg
->res
, bytes
);
2434 if (do_swap_account
)
2435 res_counter_uncharge(&memcg
->memsw
, bytes
);
2440 * A helper function to get mem_cgroup from ID. must be called under
2441 * rcu_read_lock(). The caller must check css_is_removed() or some if
2442 * it's concern. (dropping refcnt from swap can be called against removed
2445 static struct mem_cgroup
*mem_cgroup_lookup(unsigned short id
)
2447 struct cgroup_subsys_state
*css
;
2449 /* ID 0 is unused ID */
2452 css
= css_lookup(&mem_cgroup_subsys
, id
);
2455 return container_of(css
, struct mem_cgroup
, css
);
2458 struct mem_cgroup
*try_get_mem_cgroup_from_page(struct page
*page
)
2460 struct mem_cgroup
*memcg
= NULL
;
2461 struct page_cgroup
*pc
;
2465 VM_BUG_ON(!PageLocked(page
));
2467 pc
= lookup_page_cgroup(page
);
2468 lock_page_cgroup(pc
);
2469 if (PageCgroupUsed(pc
)) {
2470 memcg
= pc
->mem_cgroup
;
2471 if (memcg
&& !css_tryget(&memcg
->css
))
2473 } else if (PageSwapCache(page
)) {
2474 ent
.val
= page_private(page
);
2475 id
= lookup_swap_cgroup_id(ent
);
2477 memcg
= mem_cgroup_lookup(id
);
2478 if (memcg
&& !css_tryget(&memcg
->css
))
2482 unlock_page_cgroup(pc
);
2486 static void __mem_cgroup_commit_charge(struct mem_cgroup
*memcg
,
2488 unsigned int nr_pages
,
2489 enum charge_type ctype
,
2492 struct page_cgroup
*pc
= lookup_page_cgroup(page
);
2493 struct zone
*uninitialized_var(zone
);
2494 bool was_on_lru
= false;
2497 lock_page_cgroup(pc
);
2498 if (unlikely(PageCgroupUsed(pc
))) {
2499 unlock_page_cgroup(pc
);
2500 __mem_cgroup_cancel_charge(memcg
, nr_pages
);
2504 * we don't need page_cgroup_lock about tail pages, becase they are not
2505 * accessed by any other context at this point.
2509 * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page
2510 * may already be on some other mem_cgroup's LRU. Take care of it.
2513 zone
= page_zone(page
);
2514 spin_lock_irq(&zone
->lru_lock
);
2515 if (PageLRU(page
)) {
2517 del_page_from_lru_list(zone
, page
, page_lru(page
));
2522 pc
->mem_cgroup
= memcg
;
2524 * We access a page_cgroup asynchronously without lock_page_cgroup().
2525 * Especially when a page_cgroup is taken from a page, pc->mem_cgroup
2526 * is accessed after testing USED bit. To make pc->mem_cgroup visible
2527 * before USED bit, we need memory barrier here.
2528 * See mem_cgroup_add_lru_list(), etc.
2531 SetPageCgroupUsed(pc
);
2535 VM_BUG_ON(PageLRU(page
));
2537 add_page_to_lru_list(zone
, page
, page_lru(page
));
2539 spin_unlock_irq(&zone
->lru_lock
);
2542 if (ctype
== MEM_CGROUP_CHARGE_TYPE_MAPPED
)
2547 mem_cgroup_charge_statistics(memcg
, anon
, nr_pages
);
2548 unlock_page_cgroup(pc
);
2551 * "charge_statistics" updated event counter. Then, check it.
2552 * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree.
2553 * if they exceeds softlimit.
2555 memcg_check_events(memcg
, page
);
2558 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
2560 #define PCGF_NOCOPY_AT_SPLIT ((1 << PCG_LOCK) | (1 << PCG_MIGRATION))
2562 * Because tail pages are not marked as "used", set it. We're under
2563 * zone->lru_lock, 'splitting on pmd' and compound_lock.
2564 * charge/uncharge will be never happen and move_account() is done under
2565 * compound_lock(), so we don't have to take care of races.
2567 void mem_cgroup_split_huge_fixup(struct page
*head
)
2569 struct page_cgroup
*head_pc
= lookup_page_cgroup(head
);
2570 struct page_cgroup
*pc
;
2573 if (mem_cgroup_disabled())
2575 for (i
= 1; i
< HPAGE_PMD_NR
; i
++) {
2577 pc
->mem_cgroup
= head_pc
->mem_cgroup
;
2578 smp_wmb();/* see __commit_charge() */
2579 pc
->flags
= head_pc
->flags
& ~PCGF_NOCOPY_AT_SPLIT
;
2582 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
2585 * mem_cgroup_move_account - move account of the page
2587 * @nr_pages: number of regular pages (>1 for huge pages)
2588 * @pc: page_cgroup of the page.
2589 * @from: mem_cgroup which the page is moved from.
2590 * @to: mem_cgroup which the page is moved to. @from != @to.
2591 * @uncharge: whether we should call uncharge and css_put against @from.
2593 * The caller must confirm following.
2594 * - page is not on LRU (isolate_page() is useful.)
2595 * - compound_lock is held when nr_pages > 1
2597 * This function doesn't do "charge" nor css_get to new cgroup. It should be
2598 * done by a caller(__mem_cgroup_try_charge would be useful). If @uncharge is
2599 * true, this function does "uncharge" from old cgroup, but it doesn't if
2600 * @uncharge is false, so a caller should 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
,
2609 unsigned long flags
;
2611 bool anon
= PageAnon(page
);
2613 VM_BUG_ON(from
== to
);
2614 VM_BUG_ON(PageLRU(page
));
2616 * The page is isolated from LRU. So, collapse function
2617 * will not handle this page. But page splitting can happen.
2618 * Do this check under compound_page_lock(). The caller should
2622 if (nr_pages
> 1 && !PageTransHuge(page
))
2625 lock_page_cgroup(pc
);
2628 if (!PageCgroupUsed(pc
) || pc
->mem_cgroup
!= from
)
2631 move_lock_mem_cgroup(from
, &flags
);
2633 if (!anon
&& page_mapped(page
)) {
2634 /* Update mapped_file data for mem_cgroup */
2636 __this_cpu_dec(from
->stat
->count
[MEM_CGROUP_STAT_FILE_MAPPED
]);
2637 __this_cpu_inc(to
->stat
->count
[MEM_CGROUP_STAT_FILE_MAPPED
]);
2640 mem_cgroup_charge_statistics(from
, anon
, -nr_pages
);
2642 /* This is not "cancel", but cancel_charge does all we need. */
2643 __mem_cgroup_cancel_charge(from
, nr_pages
);
2645 /* caller should have done css_get */
2646 pc
->mem_cgroup
= to
;
2647 mem_cgroup_charge_statistics(to
, anon
, nr_pages
);
2649 * We charges against "to" which may not have any tasks. Then, "to"
2650 * can be under rmdir(). But in current implementation, caller of
2651 * this function is just force_empty() and move charge, so it's
2652 * guaranteed that "to" is never removed. So, we don't check rmdir
2655 move_unlock_mem_cgroup(from
, &flags
);
2658 unlock_page_cgroup(pc
);
2662 memcg_check_events(to
, page
);
2663 memcg_check_events(from
, page
);
2669 * move charges to its parent.
2672 static int mem_cgroup_move_parent(struct page
*page
,
2673 struct page_cgroup
*pc
,
2674 struct mem_cgroup
*child
,
2677 struct cgroup
*cg
= child
->css
.cgroup
;
2678 struct cgroup
*pcg
= cg
->parent
;
2679 struct mem_cgroup
*parent
;
2680 unsigned int nr_pages
;
2681 unsigned long uninitialized_var(flags
);
2689 if (!get_page_unless_zero(page
))
2691 if (isolate_lru_page(page
))
2694 nr_pages
= hpage_nr_pages(page
);
2696 parent
= mem_cgroup_from_cont(pcg
);
2697 ret
= __mem_cgroup_try_charge(NULL
, gfp_mask
, nr_pages
, &parent
, false);
2702 flags
= compound_lock_irqsave(page
);
2704 ret
= mem_cgroup_move_account(page
, nr_pages
, pc
, child
, parent
, true);
2706 __mem_cgroup_cancel_charge(parent
, nr_pages
);
2709 compound_unlock_irqrestore(page
, flags
);
2711 putback_lru_page(page
);
2719 * Charge the memory controller for page usage.
2721 * 0 if the charge was successful
2722 * < 0 if the cgroup is over its limit
2724 static int mem_cgroup_charge_common(struct page
*page
, struct mm_struct
*mm
,
2725 gfp_t gfp_mask
, enum charge_type ctype
)
2727 struct mem_cgroup
*memcg
= NULL
;
2728 unsigned int nr_pages
= 1;
2732 if (PageTransHuge(page
)) {
2733 nr_pages
<<= compound_order(page
);
2734 VM_BUG_ON(!PageTransHuge(page
));
2736 * Never OOM-kill a process for a huge page. The
2737 * fault handler will fall back to regular pages.
2742 ret
= __mem_cgroup_try_charge(mm
, gfp_mask
, nr_pages
, &memcg
, oom
);
2745 __mem_cgroup_commit_charge(memcg
, page
, nr_pages
, ctype
, false);
2749 int mem_cgroup_newpage_charge(struct page
*page
,
2750 struct mm_struct
*mm
, gfp_t gfp_mask
)
2752 if (mem_cgroup_disabled())
2754 VM_BUG_ON(page_mapped(page
));
2755 VM_BUG_ON(page
->mapping
&& !PageAnon(page
));
2757 return mem_cgroup_charge_common(page
, mm
, gfp_mask
,
2758 MEM_CGROUP_CHARGE_TYPE_MAPPED
);
2762 __mem_cgroup_commit_charge_swapin(struct page
*page
, struct mem_cgroup
*ptr
,
2763 enum charge_type ctype
);
2765 int mem_cgroup_cache_charge(struct page
*page
, struct mm_struct
*mm
,
2768 struct mem_cgroup
*memcg
= NULL
;
2769 enum charge_type type
= MEM_CGROUP_CHARGE_TYPE_CACHE
;
2772 if (mem_cgroup_disabled())
2774 if (PageCompound(page
))
2779 if (!page_is_file_cache(page
))
2780 type
= MEM_CGROUP_CHARGE_TYPE_SHMEM
;
2782 if (!PageSwapCache(page
))
2783 ret
= mem_cgroup_charge_common(page
, mm
, gfp_mask
, type
);
2784 else { /* page is swapcache/shmem */
2785 ret
= mem_cgroup_try_charge_swapin(mm
, page
, gfp_mask
, &memcg
);
2787 __mem_cgroup_commit_charge_swapin(page
, memcg
, type
);
2793 * While swap-in, try_charge -> commit or cancel, the page is locked.
2794 * And when try_charge() successfully returns, one refcnt to memcg without
2795 * struct page_cgroup is acquired. This refcnt will be consumed by
2796 * "commit()" or removed by "cancel()"
2798 int mem_cgroup_try_charge_swapin(struct mm_struct
*mm
,
2800 gfp_t mask
, struct mem_cgroup
**memcgp
)
2802 struct mem_cgroup
*memcg
;
2807 if (mem_cgroup_disabled())
2810 if (!do_swap_account
)
2813 * A racing thread's fault, or swapoff, may have already updated
2814 * the pte, and even removed page from swap cache: in those cases
2815 * do_swap_page()'s pte_same() test will fail; but there's also a
2816 * KSM case which does need to charge the page.
2818 if (!PageSwapCache(page
))
2820 memcg
= try_get_mem_cgroup_from_page(page
);
2824 ret
= __mem_cgroup_try_charge(NULL
, mask
, 1, memcgp
, true);
2825 css_put(&memcg
->css
);
2832 ret
= __mem_cgroup_try_charge(mm
, mask
, 1, memcgp
, true);
2839 __mem_cgroup_commit_charge_swapin(struct page
*page
, struct mem_cgroup
*memcg
,
2840 enum charge_type ctype
)
2842 if (mem_cgroup_disabled())
2846 cgroup_exclude_rmdir(&memcg
->css
);
2848 __mem_cgroup_commit_charge(memcg
, page
, 1, ctype
, true);
2850 * Now swap is on-memory. This means this page may be
2851 * counted both as mem and swap....double count.
2852 * Fix it by uncharging from memsw. Basically, this SwapCache is stable
2853 * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page()
2854 * may call delete_from_swap_cache() before reach here.
2856 if (do_swap_account
&& PageSwapCache(page
)) {
2857 swp_entry_t ent
= {.val
= page_private(page
)};
2858 mem_cgroup_uncharge_swap(ent
);
2861 * At swapin, we may charge account against cgroup which has no tasks.
2862 * So, rmdir()->pre_destroy() can be called while we do this charge.
2863 * In that case, we need to call pre_destroy() again. check it here.
2865 cgroup_release_and_wakeup_rmdir(&memcg
->css
);
2868 void mem_cgroup_commit_charge_swapin(struct page
*page
,
2869 struct mem_cgroup
*memcg
)
2871 __mem_cgroup_commit_charge_swapin(page
, memcg
,
2872 MEM_CGROUP_CHARGE_TYPE_MAPPED
);
2875 void mem_cgroup_cancel_charge_swapin(struct mem_cgroup
*memcg
)
2877 if (mem_cgroup_disabled())
2881 __mem_cgroup_cancel_charge(memcg
, 1);
2884 static void mem_cgroup_do_uncharge(struct mem_cgroup
*memcg
,
2885 unsigned int nr_pages
,
2886 const enum charge_type ctype
)
2888 struct memcg_batch_info
*batch
= NULL
;
2889 bool uncharge_memsw
= true;
2891 /* If swapout, usage of swap doesn't decrease */
2892 if (!do_swap_account
|| ctype
== MEM_CGROUP_CHARGE_TYPE_SWAPOUT
)
2893 uncharge_memsw
= false;
2895 batch
= ¤t
->memcg_batch
;
2897 * In usual, we do css_get() when we remember memcg pointer.
2898 * But in this case, we keep res->usage until end of a series of
2899 * uncharges. Then, it's ok to ignore memcg's refcnt.
2902 batch
->memcg
= memcg
;
2904 * do_batch > 0 when unmapping pages or inode invalidate/truncate.
2905 * In those cases, all pages freed continuously can be expected to be in
2906 * the same cgroup and we have chance to coalesce uncharges.
2907 * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE)
2908 * because we want to do uncharge as soon as possible.
2911 if (!batch
->do_batch
|| test_thread_flag(TIF_MEMDIE
))
2912 goto direct_uncharge
;
2915 goto direct_uncharge
;
2918 * In typical case, batch->memcg == mem. This means we can
2919 * merge a series of uncharges to an uncharge of res_counter.
2920 * If not, we uncharge res_counter ony by one.
2922 if (batch
->memcg
!= memcg
)
2923 goto direct_uncharge
;
2924 /* remember freed charge and uncharge it later */
2927 batch
->memsw_nr_pages
++;
2930 res_counter_uncharge(&memcg
->res
, nr_pages
* PAGE_SIZE
);
2932 res_counter_uncharge(&memcg
->memsw
, nr_pages
* PAGE_SIZE
);
2933 if (unlikely(batch
->memcg
!= memcg
))
2934 memcg_oom_recover(memcg
);
2938 * uncharge if !page_mapped(page)
2940 static struct mem_cgroup
*
2941 __mem_cgroup_uncharge_common(struct page
*page
, enum charge_type ctype
)
2943 struct mem_cgroup
*memcg
= NULL
;
2944 unsigned int nr_pages
= 1;
2945 struct page_cgroup
*pc
;
2948 if (mem_cgroup_disabled())
2951 if (PageSwapCache(page
))
2954 if (PageTransHuge(page
)) {
2955 nr_pages
<<= compound_order(page
);
2956 VM_BUG_ON(!PageTransHuge(page
));
2959 * Check if our page_cgroup is valid
2961 pc
= lookup_page_cgroup(page
);
2962 if (unlikely(!PageCgroupUsed(pc
)))
2965 lock_page_cgroup(pc
);
2967 memcg
= pc
->mem_cgroup
;
2969 if (!PageCgroupUsed(pc
))
2972 anon
= PageAnon(page
);
2975 case MEM_CGROUP_CHARGE_TYPE_MAPPED
:
2977 * Generally PageAnon tells if it's the anon statistics to be
2978 * updated; but sometimes e.g. mem_cgroup_uncharge_page() is
2979 * used before page reached the stage of being marked PageAnon.
2983 case MEM_CGROUP_CHARGE_TYPE_DROP
:
2984 /* See mem_cgroup_prepare_migration() */
2985 if (page_mapped(page
) || PageCgroupMigration(pc
))
2988 case MEM_CGROUP_CHARGE_TYPE_SWAPOUT
:
2989 if (!PageAnon(page
)) { /* Shared memory */
2990 if (page
->mapping
&& !page_is_file_cache(page
))
2992 } else if (page_mapped(page
)) /* Anon */
2999 mem_cgroup_charge_statistics(memcg
, anon
, -nr_pages
);
3001 ClearPageCgroupUsed(pc
);
3003 * pc->mem_cgroup is not cleared here. It will be accessed when it's
3004 * freed from LRU. This is safe because uncharged page is expected not
3005 * to be reused (freed soon). Exception is SwapCache, it's handled by
3006 * special functions.
3009 unlock_page_cgroup(pc
);
3011 * even after unlock, we have memcg->res.usage here and this memcg
3012 * will never be freed.
3014 memcg_check_events(memcg
, page
);
3015 if (do_swap_account
&& ctype
== MEM_CGROUP_CHARGE_TYPE_SWAPOUT
) {
3016 mem_cgroup_swap_statistics(memcg
, true);
3017 mem_cgroup_get(memcg
);
3019 if (!mem_cgroup_is_root(memcg
))
3020 mem_cgroup_do_uncharge(memcg
, nr_pages
, ctype
);
3025 unlock_page_cgroup(pc
);
3029 void mem_cgroup_uncharge_page(struct page
*page
)
3032 if (page_mapped(page
))
3034 VM_BUG_ON(page
->mapping
&& !PageAnon(page
));
3035 __mem_cgroup_uncharge_common(page
, MEM_CGROUP_CHARGE_TYPE_MAPPED
);
3038 void mem_cgroup_uncharge_cache_page(struct page
*page
)
3040 VM_BUG_ON(page_mapped(page
));
3041 VM_BUG_ON(page
->mapping
);
3042 __mem_cgroup_uncharge_common(page
, MEM_CGROUP_CHARGE_TYPE_CACHE
);
3046 * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate.
3047 * In that cases, pages are freed continuously and we can expect pages
3048 * are in the same memcg. All these calls itself limits the number of
3049 * pages freed at once, then uncharge_start/end() is called properly.
3050 * This may be called prural(2) times in a context,
3053 void mem_cgroup_uncharge_start(void)
3055 current
->memcg_batch
.do_batch
++;
3056 /* We can do nest. */
3057 if (current
->memcg_batch
.do_batch
== 1) {
3058 current
->memcg_batch
.memcg
= NULL
;
3059 current
->memcg_batch
.nr_pages
= 0;
3060 current
->memcg_batch
.memsw_nr_pages
= 0;
3064 void mem_cgroup_uncharge_end(void)
3066 struct memcg_batch_info
*batch
= ¤t
->memcg_batch
;
3068 if (!batch
->do_batch
)
3072 if (batch
->do_batch
) /* If stacked, do nothing. */
3078 * This "batch->memcg" is valid without any css_get/put etc...
3079 * bacause we hide charges behind us.
3081 if (batch
->nr_pages
)
3082 res_counter_uncharge(&batch
->memcg
->res
,
3083 batch
->nr_pages
* PAGE_SIZE
);
3084 if (batch
->memsw_nr_pages
)
3085 res_counter_uncharge(&batch
->memcg
->memsw
,
3086 batch
->memsw_nr_pages
* PAGE_SIZE
);
3087 memcg_oom_recover(batch
->memcg
);
3088 /* forget this pointer (for sanity check) */
3089 batch
->memcg
= NULL
;
3094 * called after __delete_from_swap_cache() and drop "page" account.
3095 * memcg information is recorded to swap_cgroup of "ent"
3098 mem_cgroup_uncharge_swapcache(struct page
*page
, swp_entry_t ent
, bool swapout
)
3100 struct mem_cgroup
*memcg
;
3101 int ctype
= MEM_CGROUP_CHARGE_TYPE_SWAPOUT
;
3103 if (!swapout
) /* this was a swap cache but the swap is unused ! */
3104 ctype
= MEM_CGROUP_CHARGE_TYPE_DROP
;
3106 memcg
= __mem_cgroup_uncharge_common(page
, ctype
);
3109 * record memcg information, if swapout && memcg != NULL,
3110 * mem_cgroup_get() was called in uncharge().
3112 if (do_swap_account
&& swapout
&& memcg
)
3113 swap_cgroup_record(ent
, css_id(&memcg
->css
));
3117 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
3119 * called from swap_entry_free(). remove record in swap_cgroup and
3120 * uncharge "memsw" account.
3122 void mem_cgroup_uncharge_swap(swp_entry_t ent
)
3124 struct mem_cgroup
*memcg
;
3127 if (!do_swap_account
)
3130 id
= swap_cgroup_record(ent
, 0);
3132 memcg
= mem_cgroup_lookup(id
);
3135 * We uncharge this because swap is freed.
3136 * This memcg can be obsolete one. We avoid calling css_tryget
3138 if (!mem_cgroup_is_root(memcg
))
3139 res_counter_uncharge(&memcg
->memsw
, PAGE_SIZE
);
3140 mem_cgroup_swap_statistics(memcg
, false);
3141 mem_cgroup_put(memcg
);
3147 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
3148 * @entry: swap entry to be moved
3149 * @from: mem_cgroup which the entry is moved from
3150 * @to: mem_cgroup which the entry is moved to
3152 * It succeeds only when the swap_cgroup's record for this entry is the same
3153 * as the mem_cgroup's id of @from.
3155 * Returns 0 on success, -EINVAL on failure.
3157 * The caller must have charged to @to, IOW, called res_counter_charge() about
3158 * both res and memsw, and called css_get().
3160 static int mem_cgroup_move_swap_account(swp_entry_t entry
,
3161 struct mem_cgroup
*from
, struct mem_cgroup
*to
)
3163 unsigned short old_id
, new_id
;
3165 old_id
= css_id(&from
->css
);
3166 new_id
= css_id(&to
->css
);
3168 if (swap_cgroup_cmpxchg(entry
, old_id
, new_id
) == old_id
) {
3169 mem_cgroup_swap_statistics(from
, false);
3170 mem_cgroup_swap_statistics(to
, true);
3172 * This function is only called from task migration context now.
3173 * It postpones res_counter and refcount handling till the end
3174 * of task migration(mem_cgroup_clear_mc()) for performance
3175 * improvement. But we cannot postpone mem_cgroup_get(to)
3176 * because if the process that has been moved to @to does
3177 * swap-in, the refcount of @to might be decreased to 0.
3185 static inline int mem_cgroup_move_swap_account(swp_entry_t entry
,
3186 struct mem_cgroup
*from
, struct mem_cgroup
*to
)
3193 * Before starting migration, account PAGE_SIZE to mem_cgroup that the old
3196 int mem_cgroup_prepare_migration(struct page
*page
,
3197 struct page
*newpage
, struct mem_cgroup
**memcgp
, gfp_t gfp_mask
)
3199 struct mem_cgroup
*memcg
= NULL
;
3200 struct page_cgroup
*pc
;
3201 enum charge_type ctype
;
3206 VM_BUG_ON(PageTransHuge(page
));
3207 if (mem_cgroup_disabled())
3210 pc
= lookup_page_cgroup(page
);
3211 lock_page_cgroup(pc
);
3212 if (PageCgroupUsed(pc
)) {
3213 memcg
= pc
->mem_cgroup
;
3214 css_get(&memcg
->css
);
3216 * At migrating an anonymous page, its mapcount goes down
3217 * to 0 and uncharge() will be called. But, even if it's fully
3218 * unmapped, migration may fail and this page has to be
3219 * charged again. We set MIGRATION flag here and delay uncharge
3220 * until end_migration() is called
3222 * Corner Case Thinking
3224 * When the old page was mapped as Anon and it's unmap-and-freed
3225 * while migration was ongoing.
3226 * If unmap finds the old page, uncharge() of it will be delayed
3227 * until end_migration(). If unmap finds a new page, it's
3228 * uncharged when it make mapcount to be 1->0. If unmap code
3229 * finds swap_migration_entry, the new page will not be mapped
3230 * and end_migration() will find it(mapcount==0).
3233 * When the old page was mapped but migraion fails, the kernel
3234 * remaps it. A charge for it is kept by MIGRATION flag even
3235 * if mapcount goes down to 0. We can do remap successfully
3236 * without charging it again.
3239 * The "old" page is under lock_page() until the end of
3240 * migration, so, the old page itself will not be swapped-out.
3241 * If the new page is swapped out before end_migraton, our
3242 * hook to usual swap-out path will catch the event.
3245 SetPageCgroupMigration(pc
);
3247 unlock_page_cgroup(pc
);
3249 * If the page is not charged at this point,
3256 ret
= __mem_cgroup_try_charge(NULL
, gfp_mask
, 1, memcgp
, false);
3257 css_put(&memcg
->css
);/* drop extra refcnt */
3259 if (PageAnon(page
)) {
3260 lock_page_cgroup(pc
);
3261 ClearPageCgroupMigration(pc
);
3262 unlock_page_cgroup(pc
);
3264 * The old page may be fully unmapped while we kept it.
3266 mem_cgroup_uncharge_page(page
);
3268 /* we'll need to revisit this error code (we have -EINTR) */
3272 * We charge new page before it's used/mapped. So, even if unlock_page()
3273 * is called before end_migration, we can catch all events on this new
3274 * page. In the case new page is migrated but not remapped, new page's
3275 * mapcount will be finally 0 and we call uncharge in end_migration().
3278 ctype
= MEM_CGROUP_CHARGE_TYPE_MAPPED
;
3279 else if (page_is_file_cache(page
))
3280 ctype
= MEM_CGROUP_CHARGE_TYPE_CACHE
;
3282 ctype
= MEM_CGROUP_CHARGE_TYPE_SHMEM
;
3283 __mem_cgroup_commit_charge(memcg
, newpage
, 1, ctype
, false);
3287 /* remove redundant charge if migration failed*/
3288 void mem_cgroup_end_migration(struct mem_cgroup
*memcg
,
3289 struct page
*oldpage
, struct page
*newpage
, bool migration_ok
)
3291 struct page
*used
, *unused
;
3292 struct page_cgroup
*pc
;
3297 /* blocks rmdir() */
3298 cgroup_exclude_rmdir(&memcg
->css
);
3299 if (!migration_ok
) {
3307 * We disallowed uncharge of pages under migration because mapcount
3308 * of the page goes down to zero, temporarly.
3309 * Clear the flag and check the page should be charged.
3311 pc
= lookup_page_cgroup(oldpage
);
3312 lock_page_cgroup(pc
);
3313 ClearPageCgroupMigration(pc
);
3314 unlock_page_cgroup(pc
);
3315 anon
= PageAnon(used
);
3316 __mem_cgroup_uncharge_common(unused
,
3317 anon
? MEM_CGROUP_CHARGE_TYPE_MAPPED
3318 : MEM_CGROUP_CHARGE_TYPE_CACHE
);
3321 * If a page is a file cache, radix-tree replacement is very atomic
3322 * and we can skip this check. When it was an Anon page, its mapcount
3323 * goes down to 0. But because we added MIGRATION flage, it's not
3324 * uncharged yet. There are several case but page->mapcount check
3325 * and USED bit check in mem_cgroup_uncharge_page() will do enough
3326 * check. (see prepare_charge() also)
3329 mem_cgroup_uncharge_page(used
);
3331 * At migration, we may charge account against cgroup which has no
3333 * So, rmdir()->pre_destroy() can be called while we do this charge.
3334 * In that case, we need to call pre_destroy() again. check it here.
3336 cgroup_release_and_wakeup_rmdir(&memcg
->css
);
3340 * At replace page cache, newpage is not under any memcg but it's on
3341 * LRU. So, this function doesn't touch res_counter but handles LRU
3342 * in correct way. Both pages are locked so we cannot race with uncharge.
3344 void mem_cgroup_replace_page_cache(struct page
*oldpage
,
3345 struct page
*newpage
)
3347 struct mem_cgroup
*memcg
= NULL
;
3348 struct page_cgroup
*pc
;
3349 enum charge_type type
= MEM_CGROUP_CHARGE_TYPE_CACHE
;
3351 if (mem_cgroup_disabled())
3354 pc
= lookup_page_cgroup(oldpage
);
3355 /* fix accounting on old pages */
3356 lock_page_cgroup(pc
);
3357 if (PageCgroupUsed(pc
)) {
3358 memcg
= pc
->mem_cgroup
;
3359 mem_cgroup_charge_statistics(memcg
, false, -1);
3360 ClearPageCgroupUsed(pc
);
3362 unlock_page_cgroup(pc
);
3365 * When called from shmem_replace_page(), in some cases the
3366 * oldpage has already been charged, and in some cases not.
3371 if (PageSwapBacked(oldpage
))
3372 type
= MEM_CGROUP_CHARGE_TYPE_SHMEM
;
3375 * Even if newpage->mapping was NULL before starting replacement,
3376 * the newpage may be on LRU(or pagevec for LRU) already. We lock
3377 * LRU while we overwrite pc->mem_cgroup.
3379 __mem_cgroup_commit_charge(memcg
, newpage
, 1, type
, true);
3382 #ifdef CONFIG_DEBUG_VM
3383 static struct page_cgroup
*lookup_page_cgroup_used(struct page
*page
)
3385 struct page_cgroup
*pc
;
3387 pc
= lookup_page_cgroup(page
);
3389 * Can be NULL while feeding pages into the page allocator for
3390 * the first time, i.e. during boot or memory hotplug;
3391 * or when mem_cgroup_disabled().
3393 if (likely(pc
) && PageCgroupUsed(pc
))
3398 bool mem_cgroup_bad_page_check(struct page
*page
)
3400 if (mem_cgroup_disabled())
3403 return lookup_page_cgroup_used(page
) != NULL
;
3406 void mem_cgroup_print_bad_page(struct page
*page
)
3408 struct page_cgroup
*pc
;
3410 pc
= lookup_page_cgroup_used(page
);
3412 printk(KERN_ALERT
"pc:%p pc->flags:%lx pc->mem_cgroup:%p\n",
3413 pc
, pc
->flags
, pc
->mem_cgroup
);
3418 static DEFINE_MUTEX(set_limit_mutex
);
3420 static int mem_cgroup_resize_limit(struct mem_cgroup
*memcg
,
3421 unsigned long long val
)
3424 u64 memswlimit
, memlimit
;
3426 int children
= mem_cgroup_count_children(memcg
);
3427 u64 curusage
, oldusage
;
3431 * For keeping hierarchical_reclaim simple, how long we should retry
3432 * is depends on callers. We set our retry-count to be function
3433 * of # of children which we should visit in this loop.
3435 retry_count
= MEM_CGROUP_RECLAIM_RETRIES
* children
;
3437 oldusage
= res_counter_read_u64(&memcg
->res
, RES_USAGE
);
3440 while (retry_count
) {
3441 if (signal_pending(current
)) {
3446 * Rather than hide all in some function, I do this in
3447 * open coded manner. You see what this really does.
3448 * We have to guarantee memcg->res.limit < memcg->memsw.limit.
3450 mutex_lock(&set_limit_mutex
);
3451 memswlimit
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
3452 if (memswlimit
< val
) {
3454 mutex_unlock(&set_limit_mutex
);
3458 memlimit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
3462 ret
= res_counter_set_limit(&memcg
->res
, val
);
3464 if (memswlimit
== val
)
3465 memcg
->memsw_is_minimum
= true;
3467 memcg
->memsw_is_minimum
= false;
3469 mutex_unlock(&set_limit_mutex
);
3474 mem_cgroup_reclaim(memcg
, GFP_KERNEL
,
3475 MEM_CGROUP_RECLAIM_SHRINK
);
3476 curusage
= res_counter_read_u64(&memcg
->res
, RES_USAGE
);
3477 /* Usage is reduced ? */
3478 if (curusage
>= oldusage
)
3481 oldusage
= curusage
;
3483 if (!ret
&& enlarge
)
3484 memcg_oom_recover(memcg
);
3489 static int mem_cgroup_resize_memsw_limit(struct mem_cgroup
*memcg
,
3490 unsigned long long val
)
3493 u64 memlimit
, memswlimit
, oldusage
, curusage
;
3494 int children
= mem_cgroup_count_children(memcg
);
3498 /* see mem_cgroup_resize_res_limit */
3499 retry_count
= children
* MEM_CGROUP_RECLAIM_RETRIES
;
3500 oldusage
= res_counter_read_u64(&memcg
->memsw
, RES_USAGE
);
3501 while (retry_count
) {
3502 if (signal_pending(current
)) {
3507 * Rather than hide all in some function, I do this in
3508 * open coded manner. You see what this really does.
3509 * We have to guarantee memcg->res.limit < memcg->memsw.limit.
3511 mutex_lock(&set_limit_mutex
);
3512 memlimit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
3513 if (memlimit
> val
) {
3515 mutex_unlock(&set_limit_mutex
);
3518 memswlimit
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
3519 if (memswlimit
< val
)
3521 ret
= res_counter_set_limit(&memcg
->memsw
, val
);
3523 if (memlimit
== val
)
3524 memcg
->memsw_is_minimum
= true;
3526 memcg
->memsw_is_minimum
= false;
3528 mutex_unlock(&set_limit_mutex
);
3533 mem_cgroup_reclaim(memcg
, GFP_KERNEL
,
3534 MEM_CGROUP_RECLAIM_NOSWAP
|
3535 MEM_CGROUP_RECLAIM_SHRINK
);
3536 curusage
= res_counter_read_u64(&memcg
->memsw
, RES_USAGE
);
3537 /* Usage is reduced ? */
3538 if (curusage
>= oldusage
)
3541 oldusage
= curusage
;
3543 if (!ret
&& enlarge
)
3544 memcg_oom_recover(memcg
);
3548 unsigned long mem_cgroup_soft_limit_reclaim(struct zone
*zone
, int order
,
3550 unsigned long *total_scanned
)
3552 unsigned long nr_reclaimed
= 0;
3553 struct mem_cgroup_per_zone
*mz
, *next_mz
= NULL
;
3554 unsigned long reclaimed
;
3556 struct mem_cgroup_tree_per_zone
*mctz
;
3557 unsigned long long excess
;
3558 unsigned long nr_scanned
;
3563 mctz
= soft_limit_tree_node_zone(zone_to_nid(zone
), zone_idx(zone
));
3565 * This loop can run a while, specially if mem_cgroup's continuously
3566 * keep exceeding their soft limit and putting the system under
3573 mz
= mem_cgroup_largest_soft_limit_node(mctz
);
3578 reclaimed
= mem_cgroup_soft_reclaim(mz
->memcg
, zone
,
3579 gfp_mask
, &nr_scanned
);
3580 nr_reclaimed
+= reclaimed
;
3581 *total_scanned
+= nr_scanned
;
3582 spin_lock(&mctz
->lock
);
3585 * If we failed to reclaim anything from this memory cgroup
3586 * it is time to move on to the next cgroup
3592 * Loop until we find yet another one.
3594 * By the time we get the soft_limit lock
3595 * again, someone might have aded the
3596 * group back on the RB tree. Iterate to
3597 * make sure we get a different mem.
3598 * mem_cgroup_largest_soft_limit_node returns
3599 * NULL if no other cgroup is present on
3603 __mem_cgroup_largest_soft_limit_node(mctz
);
3605 css_put(&next_mz
->memcg
->css
);
3606 else /* next_mz == NULL or other memcg */
3610 __mem_cgroup_remove_exceeded(mz
->memcg
, mz
, mctz
);
3611 excess
= res_counter_soft_limit_excess(&mz
->memcg
->res
);
3613 * One school of thought says that we should not add
3614 * back the node to the tree if reclaim returns 0.
3615 * But our reclaim could return 0, simply because due
3616 * to priority we are exposing a smaller subset of
3617 * memory to reclaim from. Consider this as a longer
3620 /* If excess == 0, no tree ops */
3621 __mem_cgroup_insert_exceeded(mz
->memcg
, mz
, mctz
, excess
);
3622 spin_unlock(&mctz
->lock
);
3623 css_put(&mz
->memcg
->css
);
3626 * Could not reclaim anything and there are no more
3627 * mem cgroups to try or we seem to be looping without
3628 * reclaiming anything.
3630 if (!nr_reclaimed
&&
3632 loop
> MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS
))
3634 } while (!nr_reclaimed
);
3636 css_put(&next_mz
->memcg
->css
);
3637 return nr_reclaimed
;
3641 * This routine traverse page_cgroup in given list and drop them all.
3642 * *And* this routine doesn't reclaim page itself, just removes page_cgroup.
3644 static int mem_cgroup_force_empty_list(struct mem_cgroup
*memcg
,
3645 int node
, int zid
, enum lru_list lru
)
3647 struct mem_cgroup_per_zone
*mz
;
3648 unsigned long flags
, loop
;
3649 struct list_head
*list
;
3654 zone
= &NODE_DATA(node
)->node_zones
[zid
];
3655 mz
= mem_cgroup_zoneinfo(memcg
, node
, zid
);
3656 list
= &mz
->lruvec
.lists
[lru
];
3658 loop
= mz
->lru_size
[lru
];
3659 /* give some margin against EBUSY etc...*/
3663 struct page_cgroup
*pc
;
3667 spin_lock_irqsave(&zone
->lru_lock
, flags
);
3668 if (list_empty(list
)) {
3669 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
3672 page
= list_entry(list
->prev
, struct page
, lru
);
3674 list_move(&page
->lru
, list
);
3676 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
3679 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
3681 pc
= lookup_page_cgroup(page
);
3683 ret
= mem_cgroup_move_parent(page
, pc
, memcg
, GFP_KERNEL
);
3684 if (ret
== -ENOMEM
|| ret
== -EINTR
)
3687 if (ret
== -EBUSY
|| ret
== -EINVAL
) {
3688 /* found lock contention or "pc" is obsolete. */
3695 if (!ret
&& !list_empty(list
))
3701 * make mem_cgroup's charge to be 0 if there is no task.
3702 * This enables deleting this mem_cgroup.
3704 static int mem_cgroup_force_empty(struct mem_cgroup
*memcg
, bool free_all
)
3707 int node
, zid
, shrink
;
3708 int nr_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
3709 struct cgroup
*cgrp
= memcg
->css
.cgroup
;
3711 css_get(&memcg
->css
);
3714 /* should free all ? */
3720 if (cgroup_task_count(cgrp
) || !list_empty(&cgrp
->children
))
3723 if (signal_pending(current
))
3725 /* This is for making all *used* pages to be on LRU. */
3726 lru_add_drain_all();
3727 drain_all_stock_sync(memcg
);
3729 mem_cgroup_start_move(memcg
);
3730 for_each_node_state(node
, N_HIGH_MEMORY
) {
3731 for (zid
= 0; !ret
&& zid
< MAX_NR_ZONES
; zid
++) {
3734 ret
= mem_cgroup_force_empty_list(memcg
,
3743 mem_cgroup_end_move(memcg
);
3744 memcg_oom_recover(memcg
);
3745 /* it seems parent cgroup doesn't have enough mem */
3749 /* "ret" should also be checked to ensure all lists are empty. */
3750 } while (res_counter_read_u64(&memcg
->res
, RES_USAGE
) > 0 || ret
);
3752 css_put(&memcg
->css
);
3756 /* returns EBUSY if there is a task or if we come here twice. */
3757 if (cgroup_task_count(cgrp
) || !list_empty(&cgrp
->children
) || shrink
) {
3761 /* we call try-to-free pages for make this cgroup empty */
3762 lru_add_drain_all();
3763 /* try to free all pages in this cgroup */
3765 while (nr_retries
&& res_counter_read_u64(&memcg
->res
, RES_USAGE
) > 0) {
3768 if (signal_pending(current
)) {
3772 progress
= try_to_free_mem_cgroup_pages(memcg
, GFP_KERNEL
,
3776 /* maybe some writeback is necessary */
3777 congestion_wait(BLK_RW_ASYNC
, HZ
/10);
3782 /* try move_account...there may be some *locked* pages. */
3786 int mem_cgroup_force_empty_write(struct cgroup
*cont
, unsigned int event
)
3788 return mem_cgroup_force_empty(mem_cgroup_from_cont(cont
), true);
3792 static u64
mem_cgroup_hierarchy_read(struct cgroup
*cont
, struct cftype
*cft
)
3794 return mem_cgroup_from_cont(cont
)->use_hierarchy
;
3797 static int mem_cgroup_hierarchy_write(struct cgroup
*cont
, struct cftype
*cft
,
3801 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
3802 struct cgroup
*parent
= cont
->parent
;
3803 struct mem_cgroup
*parent_memcg
= NULL
;
3806 parent_memcg
= mem_cgroup_from_cont(parent
);
3810 * If parent's use_hierarchy is set, we can't make any modifications
3811 * in the child subtrees. If it is unset, then the change can
3812 * occur, provided the current cgroup has no children.
3814 * For the root cgroup, parent_mem is NULL, we allow value to be
3815 * set if there are no children.
3817 if ((!parent_memcg
|| !parent_memcg
->use_hierarchy
) &&
3818 (val
== 1 || val
== 0)) {
3819 if (list_empty(&cont
->children
))
3820 memcg
->use_hierarchy
= val
;
3831 static unsigned long mem_cgroup_recursive_stat(struct mem_cgroup
*memcg
,
3832 enum mem_cgroup_stat_index idx
)
3834 struct mem_cgroup
*iter
;
3837 /* Per-cpu values can be negative, use a signed accumulator */
3838 for_each_mem_cgroup_tree(iter
, memcg
)
3839 val
+= mem_cgroup_read_stat(iter
, idx
);
3841 if (val
< 0) /* race ? */
3846 static inline u64
mem_cgroup_usage(struct mem_cgroup
*memcg
, bool swap
)
3850 if (!mem_cgroup_is_root(memcg
)) {
3852 return res_counter_read_u64(&memcg
->res
, RES_USAGE
);
3854 return res_counter_read_u64(&memcg
->memsw
, RES_USAGE
);
3857 val
= mem_cgroup_recursive_stat(memcg
, MEM_CGROUP_STAT_CACHE
);
3858 val
+= mem_cgroup_recursive_stat(memcg
, MEM_CGROUP_STAT_RSS
);
3861 val
+= mem_cgroup_recursive_stat(memcg
, MEM_CGROUP_STAT_SWAPOUT
);
3863 return val
<< PAGE_SHIFT
;
3866 static ssize_t
mem_cgroup_read(struct cgroup
*cont
, struct cftype
*cft
,
3867 struct file
*file
, char __user
*buf
,
3868 size_t nbytes
, loff_t
*ppos
)
3870 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
3873 int type
, name
, len
;
3875 type
= MEMFILE_TYPE(cft
->private);
3876 name
= MEMFILE_ATTR(cft
->private);
3878 if (!do_swap_account
&& type
== _MEMSWAP
)
3883 if (name
== RES_USAGE
)
3884 val
= mem_cgroup_usage(memcg
, false);
3886 val
= res_counter_read_u64(&memcg
->res
, name
);
3889 if (name
== RES_USAGE
)
3890 val
= mem_cgroup_usage(memcg
, true);
3892 val
= res_counter_read_u64(&memcg
->memsw
, name
);
3898 len
= scnprintf(str
, sizeof(str
), "%llu\n", (unsigned long long)val
);
3899 return simple_read_from_buffer(buf
, nbytes
, ppos
, str
, len
);
3902 * The user of this function is...
3905 static int mem_cgroup_write(struct cgroup
*cont
, struct cftype
*cft
,
3908 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
3910 unsigned long long val
;
3913 type
= MEMFILE_TYPE(cft
->private);
3914 name
= MEMFILE_ATTR(cft
->private);
3916 if (!do_swap_account
&& type
== _MEMSWAP
)
3921 if (mem_cgroup_is_root(memcg
)) { /* Can't set limit on root */
3925 /* This function does all necessary parse...reuse it */
3926 ret
= res_counter_memparse_write_strategy(buffer
, &val
);
3930 ret
= mem_cgroup_resize_limit(memcg
, val
);
3932 ret
= mem_cgroup_resize_memsw_limit(memcg
, val
);
3934 case RES_SOFT_LIMIT
:
3935 ret
= res_counter_memparse_write_strategy(buffer
, &val
);
3939 * For memsw, soft limits are hard to implement in terms
3940 * of semantics, for now, we support soft limits for
3941 * control without swap
3944 ret
= res_counter_set_soft_limit(&memcg
->res
, val
);
3949 ret
= -EINVAL
; /* should be BUG() ? */
3955 static void memcg_get_hierarchical_limit(struct mem_cgroup
*memcg
,
3956 unsigned long long *mem_limit
, unsigned long long *memsw_limit
)
3958 struct cgroup
*cgroup
;
3959 unsigned long long min_limit
, min_memsw_limit
, tmp
;
3961 min_limit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
3962 min_memsw_limit
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
3963 cgroup
= memcg
->css
.cgroup
;
3964 if (!memcg
->use_hierarchy
)
3967 while (cgroup
->parent
) {
3968 cgroup
= cgroup
->parent
;
3969 memcg
= mem_cgroup_from_cont(cgroup
);
3970 if (!memcg
->use_hierarchy
)
3972 tmp
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
3973 min_limit
= min(min_limit
, tmp
);
3974 tmp
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
3975 min_memsw_limit
= min(min_memsw_limit
, tmp
);
3978 *mem_limit
= min_limit
;
3979 *memsw_limit
= min_memsw_limit
;
3982 static int mem_cgroup_reset(struct cgroup
*cont
, unsigned int event
)
3984 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
3987 type
= MEMFILE_TYPE(event
);
3988 name
= MEMFILE_ATTR(event
);
3990 if (!do_swap_account
&& type
== _MEMSWAP
)
3996 res_counter_reset_max(&memcg
->res
);
3998 res_counter_reset_max(&memcg
->memsw
);
4002 res_counter_reset_failcnt(&memcg
->res
);
4004 res_counter_reset_failcnt(&memcg
->memsw
);
4011 static u64
mem_cgroup_move_charge_read(struct cgroup
*cgrp
,
4014 return mem_cgroup_from_cont(cgrp
)->move_charge_at_immigrate
;
4018 static int mem_cgroup_move_charge_write(struct cgroup
*cgrp
,
4019 struct cftype
*cft
, u64 val
)
4021 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4023 if (val
>= (1 << NR_MOVE_TYPE
))
4026 * We check this value several times in both in can_attach() and
4027 * attach(), so we need cgroup lock to prevent this value from being
4031 memcg
->move_charge_at_immigrate
= val
;
4037 static int mem_cgroup_move_charge_write(struct cgroup
*cgrp
,
4038 struct cftype
*cft
, u64 val
)
4045 /* For read statistics */
4063 struct mcs_total_stat
{
4064 s64 stat
[NR_MCS_STAT
];
4070 } memcg_stat_strings
[NR_MCS_STAT
] = {
4071 {"cache", "total_cache"},
4072 {"rss", "total_rss"},
4073 {"mapped_file", "total_mapped_file"},
4074 {"pgpgin", "total_pgpgin"},
4075 {"pgpgout", "total_pgpgout"},
4076 {"swap", "total_swap"},
4077 {"pgfault", "total_pgfault"},
4078 {"pgmajfault", "total_pgmajfault"},
4079 {"inactive_anon", "total_inactive_anon"},
4080 {"active_anon", "total_active_anon"},
4081 {"inactive_file", "total_inactive_file"},
4082 {"active_file", "total_active_file"},
4083 {"unevictable", "total_unevictable"}
4088 mem_cgroup_get_local_stat(struct mem_cgroup
*memcg
, struct mcs_total_stat
*s
)
4093 val
= mem_cgroup_read_stat(memcg
, MEM_CGROUP_STAT_CACHE
);
4094 s
->stat
[MCS_CACHE
] += val
* PAGE_SIZE
;
4095 val
= mem_cgroup_read_stat(memcg
, MEM_CGROUP_STAT_RSS
);
4096 s
->stat
[MCS_RSS
] += val
* PAGE_SIZE
;
4097 val
= mem_cgroup_read_stat(memcg
, MEM_CGROUP_STAT_FILE_MAPPED
);
4098 s
->stat
[MCS_FILE_MAPPED
] += val
* PAGE_SIZE
;
4099 val
= mem_cgroup_read_events(memcg
, MEM_CGROUP_EVENTS_PGPGIN
);
4100 s
->stat
[MCS_PGPGIN
] += val
;
4101 val
= mem_cgroup_read_events(memcg
, MEM_CGROUP_EVENTS_PGPGOUT
);
4102 s
->stat
[MCS_PGPGOUT
] += val
;
4103 if (do_swap_account
) {
4104 val
= mem_cgroup_read_stat(memcg
, MEM_CGROUP_STAT_SWAPOUT
);
4105 s
->stat
[MCS_SWAP
] += val
* PAGE_SIZE
;
4107 val
= mem_cgroup_read_events(memcg
, MEM_CGROUP_EVENTS_PGFAULT
);
4108 s
->stat
[MCS_PGFAULT
] += val
;
4109 val
= mem_cgroup_read_events(memcg
, MEM_CGROUP_EVENTS_PGMAJFAULT
);
4110 s
->stat
[MCS_PGMAJFAULT
] += val
;
4113 val
= mem_cgroup_nr_lru_pages(memcg
, BIT(LRU_INACTIVE_ANON
));
4114 s
->stat
[MCS_INACTIVE_ANON
] += val
* PAGE_SIZE
;
4115 val
= mem_cgroup_nr_lru_pages(memcg
, BIT(LRU_ACTIVE_ANON
));
4116 s
->stat
[MCS_ACTIVE_ANON
] += val
* PAGE_SIZE
;
4117 val
= mem_cgroup_nr_lru_pages(memcg
, BIT(LRU_INACTIVE_FILE
));
4118 s
->stat
[MCS_INACTIVE_FILE
] += val
* PAGE_SIZE
;
4119 val
= mem_cgroup_nr_lru_pages(memcg
, BIT(LRU_ACTIVE_FILE
));
4120 s
->stat
[MCS_ACTIVE_FILE
] += val
* PAGE_SIZE
;
4121 val
= mem_cgroup_nr_lru_pages(memcg
, BIT(LRU_UNEVICTABLE
));
4122 s
->stat
[MCS_UNEVICTABLE
] += val
* PAGE_SIZE
;
4126 mem_cgroup_get_total_stat(struct mem_cgroup
*memcg
, struct mcs_total_stat
*s
)
4128 struct mem_cgroup
*iter
;
4130 for_each_mem_cgroup_tree(iter
, memcg
)
4131 mem_cgroup_get_local_stat(iter
, s
);
4135 static int mem_control_numa_stat_show(struct seq_file
*m
, void *arg
)
4138 unsigned long total_nr
, file_nr
, anon_nr
, unevictable_nr
;
4139 unsigned long node_nr
;
4140 struct cgroup
*cont
= m
->private;
4141 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
4143 total_nr
= mem_cgroup_nr_lru_pages(memcg
, LRU_ALL
);
4144 seq_printf(m
, "total=%lu", total_nr
);
4145 for_each_node_state(nid
, N_HIGH_MEMORY
) {
4146 node_nr
= mem_cgroup_node_nr_lru_pages(memcg
, nid
, LRU_ALL
);
4147 seq_printf(m
, " N%d=%lu", nid
, node_nr
);
4151 file_nr
= mem_cgroup_nr_lru_pages(memcg
, LRU_ALL_FILE
);
4152 seq_printf(m
, "file=%lu", file_nr
);
4153 for_each_node_state(nid
, N_HIGH_MEMORY
) {
4154 node_nr
= mem_cgroup_node_nr_lru_pages(memcg
, nid
,
4156 seq_printf(m
, " N%d=%lu", nid
, node_nr
);
4160 anon_nr
= mem_cgroup_nr_lru_pages(memcg
, LRU_ALL_ANON
);
4161 seq_printf(m
, "anon=%lu", anon_nr
);
4162 for_each_node_state(nid
, N_HIGH_MEMORY
) {
4163 node_nr
= mem_cgroup_node_nr_lru_pages(memcg
, nid
,
4165 seq_printf(m
, " N%d=%lu", nid
, node_nr
);
4169 unevictable_nr
= mem_cgroup_nr_lru_pages(memcg
, BIT(LRU_UNEVICTABLE
));
4170 seq_printf(m
, "unevictable=%lu", unevictable_nr
);
4171 for_each_node_state(nid
, N_HIGH_MEMORY
) {
4172 node_nr
= mem_cgroup_node_nr_lru_pages(memcg
, nid
,
4173 BIT(LRU_UNEVICTABLE
));
4174 seq_printf(m
, " N%d=%lu", nid
, node_nr
);
4179 #endif /* CONFIG_NUMA */
4181 static int mem_control_stat_show(struct cgroup
*cont
, struct cftype
*cft
,
4182 struct cgroup_map_cb
*cb
)
4184 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
4185 struct mcs_total_stat mystat
;
4188 memset(&mystat
, 0, sizeof(mystat
));
4189 mem_cgroup_get_local_stat(memcg
, &mystat
);
4192 for (i
= 0; i
< NR_MCS_STAT
; i
++) {
4193 if (i
== MCS_SWAP
&& !do_swap_account
)
4195 cb
->fill(cb
, memcg_stat_strings
[i
].local_name
, mystat
.stat
[i
]);
4198 /* Hierarchical information */
4200 unsigned long long limit
, memsw_limit
;
4201 memcg_get_hierarchical_limit(memcg
, &limit
, &memsw_limit
);
4202 cb
->fill(cb
, "hierarchical_memory_limit", limit
);
4203 if (do_swap_account
)
4204 cb
->fill(cb
, "hierarchical_memsw_limit", memsw_limit
);
4207 memset(&mystat
, 0, sizeof(mystat
));
4208 mem_cgroup_get_total_stat(memcg
, &mystat
);
4209 for (i
= 0; i
< NR_MCS_STAT
; i
++) {
4210 if (i
== MCS_SWAP
&& !do_swap_account
)
4212 cb
->fill(cb
, memcg_stat_strings
[i
].total_name
, mystat
.stat
[i
]);
4215 #ifdef CONFIG_DEBUG_VM
4218 struct mem_cgroup_per_zone
*mz
;
4219 unsigned long recent_rotated
[2] = {0, 0};
4220 unsigned long recent_scanned
[2] = {0, 0};
4222 for_each_online_node(nid
)
4223 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++) {
4224 mz
= mem_cgroup_zoneinfo(memcg
, nid
, zid
);
4226 recent_rotated
[0] +=
4227 mz
->reclaim_stat
.recent_rotated
[0];
4228 recent_rotated
[1] +=
4229 mz
->reclaim_stat
.recent_rotated
[1];
4230 recent_scanned
[0] +=
4231 mz
->reclaim_stat
.recent_scanned
[0];
4232 recent_scanned
[1] +=
4233 mz
->reclaim_stat
.recent_scanned
[1];
4235 cb
->fill(cb
, "recent_rotated_anon", recent_rotated
[0]);
4236 cb
->fill(cb
, "recent_rotated_file", recent_rotated
[1]);
4237 cb
->fill(cb
, "recent_scanned_anon", recent_scanned
[0]);
4238 cb
->fill(cb
, "recent_scanned_file", recent_scanned
[1]);
4245 static u64
mem_cgroup_swappiness_read(struct cgroup
*cgrp
, struct cftype
*cft
)
4247 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4249 return mem_cgroup_swappiness(memcg
);
4252 static int mem_cgroup_swappiness_write(struct cgroup
*cgrp
, struct cftype
*cft
,
4255 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4256 struct mem_cgroup
*parent
;
4261 if (cgrp
->parent
== NULL
)
4264 parent
= mem_cgroup_from_cont(cgrp
->parent
);
4268 /* If under hierarchy, only empty-root can set this value */
4269 if ((parent
->use_hierarchy
) ||
4270 (memcg
->use_hierarchy
&& !list_empty(&cgrp
->children
))) {
4275 memcg
->swappiness
= val
;
4282 static void __mem_cgroup_threshold(struct mem_cgroup
*memcg
, bool swap
)
4284 struct mem_cgroup_threshold_ary
*t
;
4290 t
= rcu_dereference(memcg
->thresholds
.primary
);
4292 t
= rcu_dereference(memcg
->memsw_thresholds
.primary
);
4297 usage
= mem_cgroup_usage(memcg
, swap
);
4300 * current_threshold points to threshold just below usage.
4301 * If it's not true, a threshold was crossed after last
4302 * call of __mem_cgroup_threshold().
4304 i
= t
->current_threshold
;
4307 * Iterate backward over array of thresholds starting from
4308 * current_threshold and check if a threshold is crossed.
4309 * If none of thresholds below usage is crossed, we read
4310 * only one element of the array here.
4312 for (; i
>= 0 && unlikely(t
->entries
[i
].threshold
> usage
); i
--)
4313 eventfd_signal(t
->entries
[i
].eventfd
, 1);
4315 /* i = current_threshold + 1 */
4319 * Iterate forward over array of thresholds starting from
4320 * current_threshold+1 and check if a threshold is crossed.
4321 * If none of thresholds above usage is crossed, we read
4322 * only one element of the array here.
4324 for (; i
< t
->size
&& unlikely(t
->entries
[i
].threshold
<= usage
); i
++)
4325 eventfd_signal(t
->entries
[i
].eventfd
, 1);
4327 /* Update current_threshold */
4328 t
->current_threshold
= i
- 1;
4333 static void mem_cgroup_threshold(struct mem_cgroup
*memcg
)
4336 __mem_cgroup_threshold(memcg
, false);
4337 if (do_swap_account
)
4338 __mem_cgroup_threshold(memcg
, true);
4340 memcg
= parent_mem_cgroup(memcg
);
4344 static int compare_thresholds(const void *a
, const void *b
)
4346 const struct mem_cgroup_threshold
*_a
= a
;
4347 const struct mem_cgroup_threshold
*_b
= b
;
4349 return _a
->threshold
- _b
->threshold
;
4352 static int mem_cgroup_oom_notify_cb(struct mem_cgroup
*memcg
)
4354 struct mem_cgroup_eventfd_list
*ev
;
4356 list_for_each_entry(ev
, &memcg
->oom_notify
, list
)
4357 eventfd_signal(ev
->eventfd
, 1);
4361 static void mem_cgroup_oom_notify(struct mem_cgroup
*memcg
)
4363 struct mem_cgroup
*iter
;
4365 for_each_mem_cgroup_tree(iter
, memcg
)
4366 mem_cgroup_oom_notify_cb(iter
);
4369 static int mem_cgroup_usage_register_event(struct cgroup
*cgrp
,
4370 struct cftype
*cft
, struct eventfd_ctx
*eventfd
, const char *args
)
4372 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4373 struct mem_cgroup_thresholds
*thresholds
;
4374 struct mem_cgroup_threshold_ary
*new;
4375 int type
= MEMFILE_TYPE(cft
->private);
4376 u64 threshold
, usage
;
4379 ret
= res_counter_memparse_write_strategy(args
, &threshold
);
4383 mutex_lock(&memcg
->thresholds_lock
);
4386 thresholds
= &memcg
->thresholds
;
4387 else if (type
== _MEMSWAP
)
4388 thresholds
= &memcg
->memsw_thresholds
;
4392 usage
= mem_cgroup_usage(memcg
, type
== _MEMSWAP
);
4394 /* Check if a threshold crossed before adding a new one */
4395 if (thresholds
->primary
)
4396 __mem_cgroup_threshold(memcg
, type
== _MEMSWAP
);
4398 size
= thresholds
->primary
? thresholds
->primary
->size
+ 1 : 1;
4400 /* Allocate memory for new array of thresholds */
4401 new = kmalloc(sizeof(*new) + size
* sizeof(struct mem_cgroup_threshold
),
4409 /* Copy thresholds (if any) to new array */
4410 if (thresholds
->primary
) {
4411 memcpy(new->entries
, thresholds
->primary
->entries
, (size
- 1) *
4412 sizeof(struct mem_cgroup_threshold
));
4415 /* Add new threshold */
4416 new->entries
[size
- 1].eventfd
= eventfd
;
4417 new->entries
[size
- 1].threshold
= threshold
;
4419 /* Sort thresholds. Registering of new threshold isn't time-critical */
4420 sort(new->entries
, size
, sizeof(struct mem_cgroup_threshold
),
4421 compare_thresholds
, NULL
);
4423 /* Find current threshold */
4424 new->current_threshold
= -1;
4425 for (i
= 0; i
< size
; i
++) {
4426 if (new->entries
[i
].threshold
< usage
) {
4428 * new->current_threshold will not be used until
4429 * rcu_assign_pointer(), so it's safe to increment
4432 ++new->current_threshold
;
4436 /* Free old spare buffer and save old primary buffer as spare */
4437 kfree(thresholds
->spare
);
4438 thresholds
->spare
= thresholds
->primary
;
4440 rcu_assign_pointer(thresholds
->primary
, new);
4442 /* To be sure that nobody uses thresholds */
4446 mutex_unlock(&memcg
->thresholds_lock
);
4451 static void mem_cgroup_usage_unregister_event(struct cgroup
*cgrp
,
4452 struct cftype
*cft
, struct eventfd_ctx
*eventfd
)
4454 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4455 struct mem_cgroup_thresholds
*thresholds
;
4456 struct mem_cgroup_threshold_ary
*new;
4457 int type
= MEMFILE_TYPE(cft
->private);
4461 mutex_lock(&memcg
->thresholds_lock
);
4463 thresholds
= &memcg
->thresholds
;
4464 else if (type
== _MEMSWAP
)
4465 thresholds
= &memcg
->memsw_thresholds
;
4469 if (!thresholds
->primary
)
4472 usage
= mem_cgroup_usage(memcg
, type
== _MEMSWAP
);
4474 /* Check if a threshold crossed before removing */
4475 __mem_cgroup_threshold(memcg
, type
== _MEMSWAP
);
4477 /* Calculate new number of threshold */
4479 for (i
= 0; i
< thresholds
->primary
->size
; i
++) {
4480 if (thresholds
->primary
->entries
[i
].eventfd
!= eventfd
)
4484 new = thresholds
->spare
;
4486 /* Set thresholds array to NULL if we don't have thresholds */
4495 /* Copy thresholds and find current threshold */
4496 new->current_threshold
= -1;
4497 for (i
= 0, j
= 0; i
< thresholds
->primary
->size
; i
++) {
4498 if (thresholds
->primary
->entries
[i
].eventfd
== eventfd
)
4501 new->entries
[j
] = thresholds
->primary
->entries
[i
];
4502 if (new->entries
[j
].threshold
< usage
) {
4504 * new->current_threshold will not be used
4505 * until rcu_assign_pointer(), so it's safe to increment
4508 ++new->current_threshold
;
4514 /* Swap primary and spare array */
4515 thresholds
->spare
= thresholds
->primary
;
4516 /* If all events are unregistered, free the spare array */
4518 kfree(thresholds
->spare
);
4519 thresholds
->spare
= NULL
;
4522 rcu_assign_pointer(thresholds
->primary
, new);
4524 /* To be sure that nobody uses thresholds */
4527 mutex_unlock(&memcg
->thresholds_lock
);
4530 static int mem_cgroup_oom_register_event(struct cgroup
*cgrp
,
4531 struct cftype
*cft
, struct eventfd_ctx
*eventfd
, const char *args
)
4533 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4534 struct mem_cgroup_eventfd_list
*event
;
4535 int type
= MEMFILE_TYPE(cft
->private);
4537 BUG_ON(type
!= _OOM_TYPE
);
4538 event
= kmalloc(sizeof(*event
), GFP_KERNEL
);
4542 spin_lock(&memcg_oom_lock
);
4544 event
->eventfd
= eventfd
;
4545 list_add(&event
->list
, &memcg
->oom_notify
);
4547 /* already in OOM ? */
4548 if (atomic_read(&memcg
->under_oom
))
4549 eventfd_signal(eventfd
, 1);
4550 spin_unlock(&memcg_oom_lock
);
4555 static void mem_cgroup_oom_unregister_event(struct cgroup
*cgrp
,
4556 struct cftype
*cft
, struct eventfd_ctx
*eventfd
)
4558 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4559 struct mem_cgroup_eventfd_list
*ev
, *tmp
;
4560 int type
= MEMFILE_TYPE(cft
->private);
4562 BUG_ON(type
!= _OOM_TYPE
);
4564 spin_lock(&memcg_oom_lock
);
4566 list_for_each_entry_safe(ev
, tmp
, &memcg
->oom_notify
, list
) {
4567 if (ev
->eventfd
== eventfd
) {
4568 list_del(&ev
->list
);
4573 spin_unlock(&memcg_oom_lock
);
4576 static int mem_cgroup_oom_control_read(struct cgroup
*cgrp
,
4577 struct cftype
*cft
, struct cgroup_map_cb
*cb
)
4579 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4581 cb
->fill(cb
, "oom_kill_disable", memcg
->oom_kill_disable
);
4583 if (atomic_read(&memcg
->under_oom
))
4584 cb
->fill(cb
, "under_oom", 1);
4586 cb
->fill(cb
, "under_oom", 0);
4590 static int mem_cgroup_oom_control_write(struct cgroup
*cgrp
,
4591 struct cftype
*cft
, u64 val
)
4593 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4594 struct mem_cgroup
*parent
;
4596 /* cannot set to root cgroup and only 0 and 1 are allowed */
4597 if (!cgrp
->parent
|| !((val
== 0) || (val
== 1)))
4600 parent
= mem_cgroup_from_cont(cgrp
->parent
);
4603 /* oom-kill-disable is a flag for subhierarchy. */
4604 if ((parent
->use_hierarchy
) ||
4605 (memcg
->use_hierarchy
&& !list_empty(&cgrp
->children
))) {
4609 memcg
->oom_kill_disable
= val
;
4611 memcg_oom_recover(memcg
);
4617 static const struct file_operations mem_control_numa_stat_file_operations
= {
4619 .llseek
= seq_lseek
,
4620 .release
= single_release
,
4623 static int mem_control_numa_stat_open(struct inode
*unused
, struct file
*file
)
4625 struct cgroup
*cont
= file
->f_dentry
->d_parent
->d_fsdata
;
4627 file
->f_op
= &mem_control_numa_stat_file_operations
;
4628 return single_open(file
, mem_control_numa_stat_show
, cont
);
4630 #endif /* CONFIG_NUMA */
4632 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_KMEM
4633 static int memcg_init_kmem(struct mem_cgroup
*memcg
, struct cgroup_subsys
*ss
)
4635 return mem_cgroup_sockets_init(memcg
, ss
);
4638 static void kmem_cgroup_destroy(struct mem_cgroup
*memcg
)
4640 mem_cgroup_sockets_destroy(memcg
);
4643 static int memcg_init_kmem(struct mem_cgroup
*memcg
, struct cgroup_subsys
*ss
)
4648 static void kmem_cgroup_destroy(struct mem_cgroup
*memcg
)
4653 static struct cftype mem_cgroup_files
[] = {
4655 .name
= "usage_in_bytes",
4656 .private = MEMFILE_PRIVATE(_MEM
, RES_USAGE
),
4657 .read
= mem_cgroup_read
,
4658 .register_event
= mem_cgroup_usage_register_event
,
4659 .unregister_event
= mem_cgroup_usage_unregister_event
,
4662 .name
= "max_usage_in_bytes",
4663 .private = MEMFILE_PRIVATE(_MEM
, RES_MAX_USAGE
),
4664 .trigger
= mem_cgroup_reset
,
4665 .read
= mem_cgroup_read
,
4668 .name
= "limit_in_bytes",
4669 .private = MEMFILE_PRIVATE(_MEM
, RES_LIMIT
),
4670 .write_string
= mem_cgroup_write
,
4671 .read
= mem_cgroup_read
,
4674 .name
= "soft_limit_in_bytes",
4675 .private = MEMFILE_PRIVATE(_MEM
, RES_SOFT_LIMIT
),
4676 .write_string
= mem_cgroup_write
,
4677 .read
= mem_cgroup_read
,
4681 .private = MEMFILE_PRIVATE(_MEM
, RES_FAILCNT
),
4682 .trigger
= mem_cgroup_reset
,
4683 .read
= mem_cgroup_read
,
4687 .read_map
= mem_control_stat_show
,
4690 .name
= "force_empty",
4691 .trigger
= mem_cgroup_force_empty_write
,
4694 .name
= "use_hierarchy",
4695 .write_u64
= mem_cgroup_hierarchy_write
,
4696 .read_u64
= mem_cgroup_hierarchy_read
,
4699 .name
= "swappiness",
4700 .read_u64
= mem_cgroup_swappiness_read
,
4701 .write_u64
= mem_cgroup_swappiness_write
,
4704 .name
= "move_charge_at_immigrate",
4705 .read_u64
= mem_cgroup_move_charge_read
,
4706 .write_u64
= mem_cgroup_move_charge_write
,
4709 .name
= "oom_control",
4710 .read_map
= mem_cgroup_oom_control_read
,
4711 .write_u64
= mem_cgroup_oom_control_write
,
4712 .register_event
= mem_cgroup_oom_register_event
,
4713 .unregister_event
= mem_cgroup_oom_unregister_event
,
4714 .private = MEMFILE_PRIVATE(_OOM_TYPE
, OOM_CONTROL
),
4718 .name
= "numa_stat",
4719 .open
= mem_control_numa_stat_open
,
4723 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
4725 .name
= "memsw.usage_in_bytes",
4726 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_USAGE
),
4727 .read
= mem_cgroup_read
,
4728 .register_event
= mem_cgroup_usage_register_event
,
4729 .unregister_event
= mem_cgroup_usage_unregister_event
,
4732 .name
= "memsw.max_usage_in_bytes",
4733 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_MAX_USAGE
),
4734 .trigger
= mem_cgroup_reset
,
4735 .read
= mem_cgroup_read
,
4738 .name
= "memsw.limit_in_bytes",
4739 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_LIMIT
),
4740 .write_string
= mem_cgroup_write
,
4741 .read
= mem_cgroup_read
,
4744 .name
= "memsw.failcnt",
4745 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_FAILCNT
),
4746 .trigger
= mem_cgroup_reset
,
4747 .read
= mem_cgroup_read
,
4750 { }, /* terminate */
4753 static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup
*memcg
, int node
)
4755 struct mem_cgroup_per_node
*pn
;
4756 struct mem_cgroup_per_zone
*mz
;
4758 int zone
, tmp
= node
;
4760 * This routine is called against possible nodes.
4761 * But it's BUG to call kmalloc() against offline node.
4763 * TODO: this routine can waste much memory for nodes which will
4764 * never be onlined. It's better to use memory hotplug callback
4767 if (!node_state(node
, N_NORMAL_MEMORY
))
4769 pn
= kzalloc_node(sizeof(*pn
), GFP_KERNEL
, tmp
);
4773 for (zone
= 0; zone
< MAX_NR_ZONES
; zone
++) {
4774 mz
= &pn
->zoneinfo
[zone
];
4776 INIT_LIST_HEAD(&mz
->lruvec
.lists
[lru
]);
4777 mz
->usage_in_excess
= 0;
4778 mz
->on_tree
= false;
4781 memcg
->info
.nodeinfo
[node
] = pn
;
4785 static void free_mem_cgroup_per_zone_info(struct mem_cgroup
*memcg
, int node
)
4787 kfree(memcg
->info
.nodeinfo
[node
]);
4790 static struct mem_cgroup
*mem_cgroup_alloc(void)
4792 struct mem_cgroup
*memcg
;
4793 int size
= sizeof(struct mem_cgroup
);
4795 /* Can be very big if MAX_NUMNODES is very big */
4796 if (size
< PAGE_SIZE
)
4797 memcg
= kzalloc(size
, GFP_KERNEL
);
4799 memcg
= vzalloc(size
);
4804 memcg
->stat
= alloc_percpu(struct mem_cgroup_stat_cpu
);
4807 spin_lock_init(&memcg
->pcp_counter_lock
);
4811 if (size
< PAGE_SIZE
)
4819 * Helpers for freeing a vzalloc()ed mem_cgroup by RCU,
4820 * but in process context. The work_freeing structure is overlaid
4821 * on the rcu_freeing structure, which itself is overlaid on memsw.
4823 static void vfree_work(struct work_struct
*work
)
4825 struct mem_cgroup
*memcg
;
4827 memcg
= container_of(work
, struct mem_cgroup
, work_freeing
);
4830 static void vfree_rcu(struct rcu_head
*rcu_head
)
4832 struct mem_cgroup
*memcg
;
4834 memcg
= container_of(rcu_head
, struct mem_cgroup
, rcu_freeing
);
4835 INIT_WORK(&memcg
->work_freeing
, vfree_work
);
4836 schedule_work(&memcg
->work_freeing
);
4840 * At destroying mem_cgroup, references from swap_cgroup can remain.
4841 * (scanning all at force_empty is too costly...)
4843 * Instead of clearing all references at force_empty, we remember
4844 * the number of reference from swap_cgroup and free mem_cgroup when
4845 * it goes down to 0.
4847 * Removal of cgroup itself succeeds regardless of refs from swap.
4850 static void __mem_cgroup_free(struct mem_cgroup
*memcg
)
4854 mem_cgroup_remove_from_trees(memcg
);
4855 free_css_id(&mem_cgroup_subsys
, &memcg
->css
);
4858 free_mem_cgroup_per_zone_info(memcg
, node
);
4860 free_percpu(memcg
->stat
);
4861 if (sizeof(struct mem_cgroup
) < PAGE_SIZE
)
4862 kfree_rcu(memcg
, rcu_freeing
);
4864 call_rcu(&memcg
->rcu_freeing
, vfree_rcu
);
4867 static void mem_cgroup_get(struct mem_cgroup
*memcg
)
4869 atomic_inc(&memcg
->refcnt
);
4872 static void __mem_cgroup_put(struct mem_cgroup
*memcg
, int count
)
4874 if (atomic_sub_and_test(count
, &memcg
->refcnt
)) {
4875 struct mem_cgroup
*parent
= parent_mem_cgroup(memcg
);
4876 __mem_cgroup_free(memcg
);
4878 mem_cgroup_put(parent
);
4882 static void mem_cgroup_put(struct mem_cgroup
*memcg
)
4884 __mem_cgroup_put(memcg
, 1);
4888 * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled.
4890 struct mem_cgroup
*parent_mem_cgroup(struct mem_cgroup
*memcg
)
4892 if (!memcg
->res
.parent
)
4894 return mem_cgroup_from_res_counter(memcg
->res
.parent
, res
);
4896 EXPORT_SYMBOL(parent_mem_cgroup
);
4898 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
4899 static void __init
enable_swap_cgroup(void)
4901 if (!mem_cgroup_disabled() && really_do_swap_account
)
4902 do_swap_account
= 1;
4905 static void __init
enable_swap_cgroup(void)
4910 static int mem_cgroup_soft_limit_tree_init(void)
4912 struct mem_cgroup_tree_per_node
*rtpn
;
4913 struct mem_cgroup_tree_per_zone
*rtpz
;
4914 int tmp
, node
, zone
;
4916 for_each_node(node
) {
4918 if (!node_state(node
, N_NORMAL_MEMORY
))
4920 rtpn
= kzalloc_node(sizeof(*rtpn
), GFP_KERNEL
, tmp
);
4924 soft_limit_tree
.rb_tree_per_node
[node
] = rtpn
;
4926 for (zone
= 0; zone
< MAX_NR_ZONES
; zone
++) {
4927 rtpz
= &rtpn
->rb_tree_per_zone
[zone
];
4928 rtpz
->rb_root
= RB_ROOT
;
4929 spin_lock_init(&rtpz
->lock
);
4935 for_each_node(node
) {
4936 if (!soft_limit_tree
.rb_tree_per_node
[node
])
4938 kfree(soft_limit_tree
.rb_tree_per_node
[node
]);
4939 soft_limit_tree
.rb_tree_per_node
[node
] = NULL
;
4945 static struct cgroup_subsys_state
* __ref
4946 mem_cgroup_create(struct cgroup
*cont
)
4948 struct mem_cgroup
*memcg
, *parent
;
4949 long error
= -ENOMEM
;
4952 memcg
= mem_cgroup_alloc();
4954 return ERR_PTR(error
);
4957 if (alloc_mem_cgroup_per_zone_info(memcg
, node
))
4961 if (cont
->parent
== NULL
) {
4963 enable_swap_cgroup();
4965 if (mem_cgroup_soft_limit_tree_init())
4967 root_mem_cgroup
= memcg
;
4968 for_each_possible_cpu(cpu
) {
4969 struct memcg_stock_pcp
*stock
=
4970 &per_cpu(memcg_stock
, cpu
);
4971 INIT_WORK(&stock
->work
, drain_local_stock
);
4973 hotcpu_notifier(memcg_cpu_hotplug_callback
, 0);
4975 parent
= mem_cgroup_from_cont(cont
->parent
);
4976 memcg
->use_hierarchy
= parent
->use_hierarchy
;
4977 memcg
->oom_kill_disable
= parent
->oom_kill_disable
;
4980 if (parent
&& parent
->use_hierarchy
) {
4981 res_counter_init(&memcg
->res
, &parent
->res
);
4982 res_counter_init(&memcg
->memsw
, &parent
->memsw
);
4984 * We increment refcnt of the parent to ensure that we can
4985 * safely access it on res_counter_charge/uncharge.
4986 * This refcnt will be decremented when freeing this
4987 * mem_cgroup(see mem_cgroup_put).
4989 mem_cgroup_get(parent
);
4991 res_counter_init(&memcg
->res
, NULL
);
4992 res_counter_init(&memcg
->memsw
, NULL
);
4994 memcg
->last_scanned_node
= MAX_NUMNODES
;
4995 INIT_LIST_HEAD(&memcg
->oom_notify
);
4998 memcg
->swappiness
= mem_cgroup_swappiness(parent
);
4999 atomic_set(&memcg
->refcnt
, 1);
5000 memcg
->move_charge_at_immigrate
= 0;
5001 mutex_init(&memcg
->thresholds_lock
);
5002 spin_lock_init(&memcg
->move_lock
);
5004 error
= memcg_init_kmem(memcg
, &mem_cgroup_subsys
);
5007 * We call put now because our (and parent's) refcnts
5008 * are already in place. mem_cgroup_put() will internally
5009 * call __mem_cgroup_free, so return directly
5011 mem_cgroup_put(memcg
);
5012 return ERR_PTR(error
);
5016 __mem_cgroup_free(memcg
);
5017 return ERR_PTR(error
);
5020 static int mem_cgroup_pre_destroy(struct cgroup
*cont
)
5022 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
5024 return mem_cgroup_force_empty(memcg
, false);
5027 static void mem_cgroup_destroy(struct cgroup
*cont
)
5029 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
5031 kmem_cgroup_destroy(memcg
);
5033 mem_cgroup_put(memcg
);
5037 /* Handlers for move charge at task migration. */
5038 #define PRECHARGE_COUNT_AT_ONCE 256
5039 static int mem_cgroup_do_precharge(unsigned long count
)
5042 int batch_count
= PRECHARGE_COUNT_AT_ONCE
;
5043 struct mem_cgroup
*memcg
= mc
.to
;
5045 if (mem_cgroup_is_root(memcg
)) {
5046 mc
.precharge
+= count
;
5047 /* we don't need css_get for root */
5050 /* try to charge at once */
5052 struct res_counter
*dummy
;
5054 * "memcg" cannot be under rmdir() because we've already checked
5055 * by cgroup_lock_live_cgroup() that it is not removed and we
5056 * are still under the same cgroup_mutex. So we can postpone
5059 if (res_counter_charge(&memcg
->res
, PAGE_SIZE
* count
, &dummy
))
5061 if (do_swap_account
&& res_counter_charge(&memcg
->memsw
,
5062 PAGE_SIZE
* count
, &dummy
)) {
5063 res_counter_uncharge(&memcg
->res
, PAGE_SIZE
* count
);
5066 mc
.precharge
+= count
;
5070 /* fall back to one by one charge */
5072 if (signal_pending(current
)) {
5076 if (!batch_count
--) {
5077 batch_count
= PRECHARGE_COUNT_AT_ONCE
;
5080 ret
= __mem_cgroup_try_charge(NULL
,
5081 GFP_KERNEL
, 1, &memcg
, false);
5083 /* mem_cgroup_clear_mc() will do uncharge later */
5091 * get_mctgt_type - get target type of moving charge
5092 * @vma: the vma the pte to be checked belongs
5093 * @addr: the address corresponding to the pte to be checked
5094 * @ptent: the pte to be checked
5095 * @target: the pointer the target page or swap ent will be stored(can be NULL)
5098 * 0(MC_TARGET_NONE): if the pte is not a target for move charge.
5099 * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
5100 * move charge. if @target is not NULL, the page is stored in target->page
5101 * with extra refcnt got(Callers should handle it).
5102 * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
5103 * target for charge migration. if @target is not NULL, the entry is stored
5106 * Called with pte lock held.
5113 enum mc_target_type
{
5119 static struct page
*mc_handle_present_pte(struct vm_area_struct
*vma
,
5120 unsigned long addr
, pte_t ptent
)
5122 struct page
*page
= vm_normal_page(vma
, addr
, ptent
);
5124 if (!page
|| !page_mapped(page
))
5126 if (PageAnon(page
)) {
5127 /* we don't move shared anon */
5130 } else if (!move_file())
5131 /* we ignore mapcount for file pages */
5133 if (!get_page_unless_zero(page
))
5140 static struct page
*mc_handle_swap_pte(struct vm_area_struct
*vma
,
5141 unsigned long addr
, pte_t ptent
, swp_entry_t
*entry
)
5143 struct page
*page
= NULL
;
5144 swp_entry_t ent
= pte_to_swp_entry(ptent
);
5146 if (!move_anon() || non_swap_entry(ent
))
5149 * Because lookup_swap_cache() updates some statistics counter,
5150 * we call find_get_page() with swapper_space directly.
5152 page
= find_get_page(&swapper_space
, ent
.val
);
5153 if (do_swap_account
)
5154 entry
->val
= ent
.val
;
5159 static struct page
*mc_handle_swap_pte(struct vm_area_struct
*vma
,
5160 unsigned long addr
, pte_t ptent
, swp_entry_t
*entry
)
5166 static struct page
*mc_handle_file_pte(struct vm_area_struct
*vma
,
5167 unsigned long addr
, pte_t ptent
, swp_entry_t
*entry
)
5169 struct page
*page
= NULL
;
5170 struct inode
*inode
;
5171 struct address_space
*mapping
;
5174 if (!vma
->vm_file
) /* anonymous vma */
5179 inode
= vma
->vm_file
->f_path
.dentry
->d_inode
;
5180 mapping
= vma
->vm_file
->f_mapping
;
5181 if (pte_none(ptent
))
5182 pgoff
= linear_page_index(vma
, addr
);
5183 else /* pte_file(ptent) is true */
5184 pgoff
= pte_to_pgoff(ptent
);
5186 /* page is moved even if it's not RSS of this task(page-faulted). */
5187 page
= find_get_page(mapping
, pgoff
);
5190 /* shmem/tmpfs may report page out on swap: account for that too. */
5191 if (radix_tree_exceptional_entry(page
)) {
5192 swp_entry_t swap
= radix_to_swp_entry(page
);
5193 if (do_swap_account
)
5195 page
= find_get_page(&swapper_space
, swap
.val
);
5201 static enum mc_target_type
get_mctgt_type(struct vm_area_struct
*vma
,
5202 unsigned long addr
, pte_t ptent
, union mc_target
*target
)
5204 struct page
*page
= NULL
;
5205 struct page_cgroup
*pc
;
5206 enum mc_target_type ret
= MC_TARGET_NONE
;
5207 swp_entry_t ent
= { .val
= 0 };
5209 if (pte_present(ptent
))
5210 page
= mc_handle_present_pte(vma
, addr
, ptent
);
5211 else if (is_swap_pte(ptent
))
5212 page
= mc_handle_swap_pte(vma
, addr
, ptent
, &ent
);
5213 else if (pte_none(ptent
) || pte_file(ptent
))
5214 page
= mc_handle_file_pte(vma
, addr
, ptent
, &ent
);
5216 if (!page
&& !ent
.val
)
5219 pc
= lookup_page_cgroup(page
);
5221 * Do only loose check w/o page_cgroup lock.
5222 * mem_cgroup_move_account() checks the pc is valid or not under
5225 if (PageCgroupUsed(pc
) && pc
->mem_cgroup
== mc
.from
) {
5226 ret
= MC_TARGET_PAGE
;
5228 target
->page
= page
;
5230 if (!ret
|| !target
)
5233 /* There is a swap entry and a page doesn't exist or isn't charged */
5234 if (ent
.val
&& !ret
&&
5235 css_id(&mc
.from
->css
) == lookup_swap_cgroup_id(ent
)) {
5236 ret
= MC_TARGET_SWAP
;
5243 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
5245 * We don't consider swapping or file mapped pages because THP does not
5246 * support them for now.
5247 * Caller should make sure that pmd_trans_huge(pmd) is true.
5249 static enum mc_target_type
get_mctgt_type_thp(struct vm_area_struct
*vma
,
5250 unsigned long addr
, pmd_t pmd
, union mc_target
*target
)
5252 struct page
*page
= NULL
;
5253 struct page_cgroup
*pc
;
5254 enum mc_target_type ret
= MC_TARGET_NONE
;
5256 page
= pmd_page(pmd
);
5257 VM_BUG_ON(!page
|| !PageHead(page
));
5260 pc
= lookup_page_cgroup(page
);
5261 if (PageCgroupUsed(pc
) && pc
->mem_cgroup
== mc
.from
) {
5262 ret
= MC_TARGET_PAGE
;
5265 target
->page
= page
;
5271 static inline enum mc_target_type
get_mctgt_type_thp(struct vm_area_struct
*vma
,
5272 unsigned long addr
, pmd_t pmd
, union mc_target
*target
)
5274 return MC_TARGET_NONE
;
5278 static int mem_cgroup_count_precharge_pte_range(pmd_t
*pmd
,
5279 unsigned long addr
, unsigned long end
,
5280 struct mm_walk
*walk
)
5282 struct vm_area_struct
*vma
= walk
->private;
5286 if (pmd_trans_huge_lock(pmd
, vma
) == 1) {
5287 if (get_mctgt_type_thp(vma
, addr
, *pmd
, NULL
) == MC_TARGET_PAGE
)
5288 mc
.precharge
+= HPAGE_PMD_NR
;
5289 spin_unlock(&vma
->vm_mm
->page_table_lock
);
5293 if (pmd_trans_unstable(pmd
))
5295 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
5296 for (; addr
!= end
; pte
++, addr
+= PAGE_SIZE
)
5297 if (get_mctgt_type(vma
, addr
, *pte
, NULL
))
5298 mc
.precharge
++; /* increment precharge temporarily */
5299 pte_unmap_unlock(pte
- 1, ptl
);
5305 static unsigned long mem_cgroup_count_precharge(struct mm_struct
*mm
)
5307 unsigned long precharge
;
5308 struct vm_area_struct
*vma
;
5310 down_read(&mm
->mmap_sem
);
5311 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
5312 struct mm_walk mem_cgroup_count_precharge_walk
= {
5313 .pmd_entry
= mem_cgroup_count_precharge_pte_range
,
5317 if (is_vm_hugetlb_page(vma
))
5319 walk_page_range(vma
->vm_start
, vma
->vm_end
,
5320 &mem_cgroup_count_precharge_walk
);
5322 up_read(&mm
->mmap_sem
);
5324 precharge
= mc
.precharge
;
5330 static int mem_cgroup_precharge_mc(struct mm_struct
*mm
)
5332 unsigned long precharge
= mem_cgroup_count_precharge(mm
);
5334 VM_BUG_ON(mc
.moving_task
);
5335 mc
.moving_task
= current
;
5336 return mem_cgroup_do_precharge(precharge
);
5339 /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */
5340 static void __mem_cgroup_clear_mc(void)
5342 struct mem_cgroup
*from
= mc
.from
;
5343 struct mem_cgroup
*to
= mc
.to
;
5345 /* we must uncharge all the leftover precharges from mc.to */
5347 __mem_cgroup_cancel_charge(mc
.to
, mc
.precharge
);
5351 * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
5352 * we must uncharge here.
5354 if (mc
.moved_charge
) {
5355 __mem_cgroup_cancel_charge(mc
.from
, mc
.moved_charge
);
5356 mc
.moved_charge
= 0;
5358 /* we must fixup refcnts and charges */
5359 if (mc
.moved_swap
) {
5360 /* uncharge swap account from the old cgroup */
5361 if (!mem_cgroup_is_root(mc
.from
))
5362 res_counter_uncharge(&mc
.from
->memsw
,
5363 PAGE_SIZE
* mc
.moved_swap
);
5364 __mem_cgroup_put(mc
.from
, mc
.moved_swap
);
5366 if (!mem_cgroup_is_root(mc
.to
)) {
5368 * we charged both to->res and to->memsw, so we should
5371 res_counter_uncharge(&mc
.to
->res
,
5372 PAGE_SIZE
* mc
.moved_swap
);
5374 /* we've already done mem_cgroup_get(mc.to) */
5377 memcg_oom_recover(from
);
5378 memcg_oom_recover(to
);
5379 wake_up_all(&mc
.waitq
);
5382 static void mem_cgroup_clear_mc(void)
5384 struct mem_cgroup
*from
= mc
.from
;
5387 * we must clear moving_task before waking up waiters at the end of
5390 mc
.moving_task
= NULL
;
5391 __mem_cgroup_clear_mc();
5392 spin_lock(&mc
.lock
);
5395 spin_unlock(&mc
.lock
);
5396 mem_cgroup_end_move(from
);
5399 static int mem_cgroup_can_attach(struct cgroup
*cgroup
,
5400 struct cgroup_taskset
*tset
)
5402 struct task_struct
*p
= cgroup_taskset_first(tset
);
5404 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgroup
);
5406 if (memcg
->move_charge_at_immigrate
) {
5407 struct mm_struct
*mm
;
5408 struct mem_cgroup
*from
= mem_cgroup_from_task(p
);
5410 VM_BUG_ON(from
== memcg
);
5412 mm
= get_task_mm(p
);
5415 /* We move charges only when we move a owner of the mm */
5416 if (mm
->owner
== p
) {
5419 VM_BUG_ON(mc
.precharge
);
5420 VM_BUG_ON(mc
.moved_charge
);
5421 VM_BUG_ON(mc
.moved_swap
);
5422 mem_cgroup_start_move(from
);
5423 spin_lock(&mc
.lock
);
5426 spin_unlock(&mc
.lock
);
5427 /* We set mc.moving_task later */
5429 ret
= mem_cgroup_precharge_mc(mm
);
5431 mem_cgroup_clear_mc();
5438 static void mem_cgroup_cancel_attach(struct cgroup
*cgroup
,
5439 struct cgroup_taskset
*tset
)
5441 mem_cgroup_clear_mc();
5444 static int mem_cgroup_move_charge_pte_range(pmd_t
*pmd
,
5445 unsigned long addr
, unsigned long end
,
5446 struct mm_walk
*walk
)
5449 struct vm_area_struct
*vma
= walk
->private;
5452 enum mc_target_type target_type
;
5453 union mc_target target
;
5455 struct page_cgroup
*pc
;
5458 * We don't take compound_lock() here but no race with splitting thp
5460 * - if pmd_trans_huge_lock() returns 1, the relevant thp is not
5461 * under splitting, which means there's no concurrent thp split,
5462 * - if another thread runs into split_huge_page() just after we
5463 * entered this if-block, the thread must wait for page table lock
5464 * to be unlocked in __split_huge_page_splitting(), where the main
5465 * part of thp split is not executed yet.
5467 if (pmd_trans_huge_lock(pmd
, vma
) == 1) {
5468 if (mc
.precharge
< HPAGE_PMD_NR
) {
5469 spin_unlock(&vma
->vm_mm
->page_table_lock
);
5472 target_type
= get_mctgt_type_thp(vma
, addr
, *pmd
, &target
);
5473 if (target_type
== MC_TARGET_PAGE
) {
5475 if (!isolate_lru_page(page
)) {
5476 pc
= lookup_page_cgroup(page
);
5477 if (!mem_cgroup_move_account(page
, HPAGE_PMD_NR
,
5480 mc
.precharge
-= HPAGE_PMD_NR
;
5481 mc
.moved_charge
+= HPAGE_PMD_NR
;
5483 putback_lru_page(page
);
5487 spin_unlock(&vma
->vm_mm
->page_table_lock
);
5491 if (pmd_trans_unstable(pmd
))
5494 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
5495 for (; addr
!= end
; addr
+= PAGE_SIZE
) {
5496 pte_t ptent
= *(pte
++);
5502 switch (get_mctgt_type(vma
, addr
, ptent
, &target
)) {
5503 case MC_TARGET_PAGE
:
5505 if (isolate_lru_page(page
))
5507 pc
= lookup_page_cgroup(page
);
5508 if (!mem_cgroup_move_account(page
, 1, pc
,
5509 mc
.from
, mc
.to
, false)) {
5511 /* we uncharge from mc.from later. */
5514 putback_lru_page(page
);
5515 put
: /* get_mctgt_type() gets the page */
5518 case MC_TARGET_SWAP
:
5520 if (!mem_cgroup_move_swap_account(ent
, mc
.from
, mc
.to
)) {
5522 /* we fixup refcnts and charges later. */
5530 pte_unmap_unlock(pte
- 1, ptl
);
5535 * We have consumed all precharges we got in can_attach().
5536 * We try charge one by one, but don't do any additional
5537 * charges to mc.to if we have failed in charge once in attach()
5540 ret
= mem_cgroup_do_precharge(1);
5548 static void mem_cgroup_move_charge(struct mm_struct
*mm
)
5550 struct vm_area_struct
*vma
;
5552 lru_add_drain_all();
5554 if (unlikely(!down_read_trylock(&mm
->mmap_sem
))) {
5556 * Someone who are holding the mmap_sem might be waiting in
5557 * waitq. So we cancel all extra charges, wake up all waiters,
5558 * and retry. Because we cancel precharges, we might not be able
5559 * to move enough charges, but moving charge is a best-effort
5560 * feature anyway, so it wouldn't be a big problem.
5562 __mem_cgroup_clear_mc();
5566 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
5568 struct mm_walk mem_cgroup_move_charge_walk
= {
5569 .pmd_entry
= mem_cgroup_move_charge_pte_range
,
5573 if (is_vm_hugetlb_page(vma
))
5575 ret
= walk_page_range(vma
->vm_start
, vma
->vm_end
,
5576 &mem_cgroup_move_charge_walk
);
5579 * means we have consumed all precharges and failed in
5580 * doing additional charge. Just abandon here.
5584 up_read(&mm
->mmap_sem
);
5587 static void mem_cgroup_move_task(struct cgroup
*cont
,
5588 struct cgroup_taskset
*tset
)
5590 struct task_struct
*p
= cgroup_taskset_first(tset
);
5591 struct mm_struct
*mm
= get_task_mm(p
);
5595 mem_cgroup_move_charge(mm
);
5599 mem_cgroup_clear_mc();
5601 #else /* !CONFIG_MMU */
5602 static int mem_cgroup_can_attach(struct cgroup
*cgroup
,
5603 struct cgroup_taskset
*tset
)
5607 static void mem_cgroup_cancel_attach(struct cgroup
*cgroup
,
5608 struct cgroup_taskset
*tset
)
5611 static void mem_cgroup_move_task(struct cgroup
*cont
,
5612 struct cgroup_taskset
*tset
)
5617 struct cgroup_subsys mem_cgroup_subsys
= {
5619 .subsys_id
= mem_cgroup_subsys_id
,
5620 .create
= mem_cgroup_create
,
5621 .pre_destroy
= mem_cgroup_pre_destroy
,
5622 .destroy
= mem_cgroup_destroy
,
5623 .can_attach
= mem_cgroup_can_attach
,
5624 .cancel_attach
= mem_cgroup_cancel_attach
,
5625 .attach
= mem_cgroup_move_task
,
5626 .base_cftypes
= mem_cgroup_files
,
5629 .__DEPRECATED_clear_css_refs
= true,
5632 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
5633 static int __init
enable_swap_account(char *s
)
5635 /* consider enabled if no parameter or 1 is given */
5636 if (!strcmp(s
, "1"))
5637 really_do_swap_account
= 1;
5638 else if (!strcmp(s
, "0"))
5639 really_do_swap_account
= 0;
5642 __setup("swapaccount=", enable_swap_account
);