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_ON_MOVE
, /* someone is moving account between groups */
93 MEM_CGROUP_STAT_NSTATS
,
96 enum mem_cgroup_events_index
{
97 MEM_CGROUP_EVENTS_PGPGIN
, /* # of pages paged in */
98 MEM_CGROUP_EVENTS_PGPGOUT
, /* # of pages paged out */
99 MEM_CGROUP_EVENTS_COUNT
, /* # of pages paged in/out */
100 MEM_CGROUP_EVENTS_PGFAULT
, /* # of page-faults */
101 MEM_CGROUP_EVENTS_PGMAJFAULT
, /* # of major page-faults */
102 MEM_CGROUP_EVENTS_NSTATS
,
105 * Per memcg event counter is incremented at every pagein/pageout. With THP,
106 * it will be incremated by the number of pages. This counter is used for
107 * for trigger some periodic events. This is straightforward and better
108 * than using jiffies etc. to handle periodic memcg event.
110 enum mem_cgroup_events_target
{
111 MEM_CGROUP_TARGET_THRESH
,
112 MEM_CGROUP_TARGET_SOFTLIMIT
,
113 MEM_CGROUP_TARGET_NUMAINFO
,
116 #define THRESHOLDS_EVENTS_TARGET (128)
117 #define SOFTLIMIT_EVENTS_TARGET (1024)
118 #define NUMAINFO_EVENTS_TARGET (1024)
120 struct mem_cgroup_stat_cpu
{
121 long count
[MEM_CGROUP_STAT_NSTATS
];
122 unsigned long events
[MEM_CGROUP_EVENTS_NSTATS
];
123 unsigned long targets
[MEM_CGROUP_NTARGETS
];
126 struct mem_cgroup_reclaim_iter
{
127 /* css_id of the last scanned hierarchy member */
129 /* scan generation, increased every round-trip */
130 unsigned int generation
;
134 * per-zone information in memory controller.
136 struct mem_cgroup_per_zone
{
137 struct lruvec lruvec
;
138 unsigned long count
[NR_LRU_LISTS
];
140 struct mem_cgroup_reclaim_iter reclaim_iter
[DEF_PRIORITY
+ 1];
142 struct zone_reclaim_stat reclaim_stat
;
143 struct rb_node tree_node
; /* RB tree node */
144 unsigned long long usage_in_excess
;/* Set to the value by which */
145 /* the soft limit is exceeded*/
147 struct mem_cgroup
*mem
; /* Back pointer, we cannot */
148 /* use container_of */
150 /* Macro for accessing counter */
151 #define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)])
153 struct mem_cgroup_per_node
{
154 struct mem_cgroup_per_zone zoneinfo
[MAX_NR_ZONES
];
157 struct mem_cgroup_lru_info
{
158 struct mem_cgroup_per_node
*nodeinfo
[MAX_NUMNODES
];
162 * Cgroups above their limits are maintained in a RB-Tree, independent of
163 * their hierarchy representation
166 struct mem_cgroup_tree_per_zone
{
167 struct rb_root rb_root
;
171 struct mem_cgroup_tree_per_node
{
172 struct mem_cgroup_tree_per_zone rb_tree_per_zone
[MAX_NR_ZONES
];
175 struct mem_cgroup_tree
{
176 struct mem_cgroup_tree_per_node
*rb_tree_per_node
[MAX_NUMNODES
];
179 static struct mem_cgroup_tree soft_limit_tree __read_mostly
;
181 struct mem_cgroup_threshold
{
182 struct eventfd_ctx
*eventfd
;
187 struct mem_cgroup_threshold_ary
{
188 /* An array index points to threshold just below usage. */
189 int current_threshold
;
190 /* Size of entries[] */
192 /* Array of thresholds */
193 struct mem_cgroup_threshold entries
[0];
196 struct mem_cgroup_thresholds
{
197 /* Primary thresholds array */
198 struct mem_cgroup_threshold_ary
*primary
;
200 * Spare threshold array.
201 * This is needed to make mem_cgroup_unregister_event() "never fail".
202 * It must be able to store at least primary->size - 1 entries.
204 struct mem_cgroup_threshold_ary
*spare
;
208 struct mem_cgroup_eventfd_list
{
209 struct list_head list
;
210 struct eventfd_ctx
*eventfd
;
213 static void mem_cgroup_threshold(struct mem_cgroup
*memcg
);
214 static void mem_cgroup_oom_notify(struct mem_cgroup
*memcg
);
217 * The memory controller data structure. The memory controller controls both
218 * page cache and RSS per cgroup. We would eventually like to provide
219 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
220 * to help the administrator determine what knobs to tune.
222 * TODO: Add a water mark for the memory controller. Reclaim will begin when
223 * we hit the water mark. May be even add a low water mark, such that
224 * no reclaim occurs from a cgroup at it's low water mark, this is
225 * a feature that will be implemented much later in the future.
228 struct cgroup_subsys_state css
;
230 * the counter to account for memory usage
232 struct res_counter res
;
234 * the counter to account for mem+swap usage.
236 struct res_counter memsw
;
238 * Per cgroup active and inactive list, similar to the
239 * per zone LRU lists.
241 struct mem_cgroup_lru_info info
;
242 int last_scanned_node
;
244 nodemask_t scan_nodes
;
245 atomic_t numainfo_events
;
246 atomic_t numainfo_updating
;
249 * Should the accounting and control be hierarchical, per subtree?
259 /* OOM-Killer disable */
260 int oom_kill_disable
;
262 /* set when res.limit == memsw.limit */
263 bool memsw_is_minimum
;
265 /* protect arrays of thresholds */
266 struct mutex thresholds_lock
;
268 /* thresholds for memory usage. RCU-protected */
269 struct mem_cgroup_thresholds thresholds
;
271 /* thresholds for mem+swap usage. RCU-protected */
272 struct mem_cgroup_thresholds memsw_thresholds
;
274 /* For oom notifier event fd */
275 struct list_head oom_notify
;
278 * Should we move charges of a task when a task is moved into this
279 * mem_cgroup ? And what type of charges should we move ?
281 unsigned long move_charge_at_immigrate
;
285 struct mem_cgroup_stat_cpu
*stat
;
287 * used when a cpu is offlined or other synchronizations
288 * See mem_cgroup_read_stat().
290 struct mem_cgroup_stat_cpu nocpu_base
;
291 spinlock_t pcp_counter_lock
;
294 struct tcp_memcontrol tcp_mem
;
298 /* Stuffs for move charges at task migration. */
300 * Types of charges to be moved. "move_charge_at_immitgrate" is treated as a
301 * left-shifted bitmap of these types.
304 MOVE_CHARGE_TYPE_ANON
, /* private anonymous page and swap of it */
305 MOVE_CHARGE_TYPE_FILE
, /* file page(including tmpfs) and swap of it */
309 /* "mc" and its members are protected by cgroup_mutex */
310 static struct move_charge_struct
{
311 spinlock_t lock
; /* for from, to */
312 struct mem_cgroup
*from
;
313 struct mem_cgroup
*to
;
314 unsigned long precharge
;
315 unsigned long moved_charge
;
316 unsigned long moved_swap
;
317 struct task_struct
*moving_task
; /* a task moving charges */
318 wait_queue_head_t waitq
; /* a waitq for other context */
320 .lock
= __SPIN_LOCK_UNLOCKED(mc
.lock
),
321 .waitq
= __WAIT_QUEUE_HEAD_INITIALIZER(mc
.waitq
),
324 static bool move_anon(void)
326 return test_bit(MOVE_CHARGE_TYPE_ANON
,
327 &mc
.to
->move_charge_at_immigrate
);
330 static bool move_file(void)
332 return test_bit(MOVE_CHARGE_TYPE_FILE
,
333 &mc
.to
->move_charge_at_immigrate
);
337 * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft
338 * limit reclaim to prevent infinite loops, if they ever occur.
340 #define MEM_CGROUP_MAX_RECLAIM_LOOPS (100)
341 #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS (2)
344 MEM_CGROUP_CHARGE_TYPE_CACHE
= 0,
345 MEM_CGROUP_CHARGE_TYPE_MAPPED
,
346 MEM_CGROUP_CHARGE_TYPE_SHMEM
, /* used by page migration of shmem */
347 MEM_CGROUP_CHARGE_TYPE_FORCE
, /* used by force_empty */
348 MEM_CGROUP_CHARGE_TYPE_SWAPOUT
, /* for accounting swapcache */
349 MEM_CGROUP_CHARGE_TYPE_DROP
, /* a page was unused swap cache */
353 /* for encoding cft->private value on file */
356 #define _OOM_TYPE (2)
357 #define MEMFILE_PRIVATE(x, val) (((x) << 16) | (val))
358 #define MEMFILE_TYPE(val) (((val) >> 16) & 0xffff)
359 #define MEMFILE_ATTR(val) ((val) & 0xffff)
360 /* Used for OOM nofiier */
361 #define OOM_CONTROL (0)
364 * Reclaim flags for mem_cgroup_hierarchical_reclaim
366 #define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0
367 #define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT)
368 #define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1
369 #define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT)
371 static void mem_cgroup_get(struct mem_cgroup
*memcg
);
372 static void mem_cgroup_put(struct mem_cgroup
*memcg
);
374 /* Writing them here to avoid exposing memcg's inner layout */
375 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_KMEM
377 #include <net/sock.h>
380 static bool mem_cgroup_is_root(struct mem_cgroup
*memcg
);
381 void sock_update_memcg(struct sock
*sk
)
383 if (static_branch(&memcg_socket_limit_enabled
)) {
384 struct mem_cgroup
*memcg
;
386 BUG_ON(!sk
->sk_prot
->proto_cgroup
);
388 /* Socket cloning can throw us here with sk_cgrp already
389 * filled. It won't however, necessarily happen from
390 * process context. So the test for root memcg given
391 * the current task's memcg won't help us in this case.
393 * Respecting the original socket's memcg is a better
394 * decision in this case.
397 BUG_ON(mem_cgroup_is_root(sk
->sk_cgrp
->memcg
));
398 mem_cgroup_get(sk
->sk_cgrp
->memcg
);
403 memcg
= mem_cgroup_from_task(current
);
404 if (!mem_cgroup_is_root(memcg
)) {
405 mem_cgroup_get(memcg
);
406 sk
->sk_cgrp
= sk
->sk_prot
->proto_cgroup(memcg
);
411 EXPORT_SYMBOL(sock_update_memcg
);
413 void sock_release_memcg(struct sock
*sk
)
415 if (static_branch(&memcg_socket_limit_enabled
) && sk
->sk_cgrp
) {
416 struct mem_cgroup
*memcg
;
417 WARN_ON(!sk
->sk_cgrp
->memcg
);
418 memcg
= sk
->sk_cgrp
->memcg
;
419 mem_cgroup_put(memcg
);
423 struct cg_proto
*tcp_proto_cgroup(struct mem_cgroup
*memcg
)
425 if (!memcg
|| mem_cgroup_is_root(memcg
))
428 return &memcg
->tcp_mem
.cg_proto
;
430 EXPORT_SYMBOL(tcp_proto_cgroup
);
431 #endif /* CONFIG_INET */
432 #endif /* CONFIG_CGROUP_MEM_RES_CTLR_KMEM */
434 static void drain_all_stock_async(struct mem_cgroup
*memcg
);
436 static struct mem_cgroup_per_zone
*
437 mem_cgroup_zoneinfo(struct mem_cgroup
*memcg
, int nid
, int zid
)
439 return &memcg
->info
.nodeinfo
[nid
]->zoneinfo
[zid
];
442 struct cgroup_subsys_state
*mem_cgroup_css(struct mem_cgroup
*memcg
)
447 static struct mem_cgroup_per_zone
*
448 page_cgroup_zoneinfo(struct mem_cgroup
*memcg
, struct page
*page
)
450 int nid
= page_to_nid(page
);
451 int zid
= page_zonenum(page
);
453 return mem_cgroup_zoneinfo(memcg
, nid
, zid
);
456 static struct mem_cgroup_tree_per_zone
*
457 soft_limit_tree_node_zone(int nid
, int zid
)
459 return &soft_limit_tree
.rb_tree_per_node
[nid
]->rb_tree_per_zone
[zid
];
462 static struct mem_cgroup_tree_per_zone
*
463 soft_limit_tree_from_page(struct page
*page
)
465 int nid
= page_to_nid(page
);
466 int zid
= page_zonenum(page
);
468 return &soft_limit_tree
.rb_tree_per_node
[nid
]->rb_tree_per_zone
[zid
];
472 __mem_cgroup_insert_exceeded(struct mem_cgroup
*memcg
,
473 struct mem_cgroup_per_zone
*mz
,
474 struct mem_cgroup_tree_per_zone
*mctz
,
475 unsigned long long new_usage_in_excess
)
477 struct rb_node
**p
= &mctz
->rb_root
.rb_node
;
478 struct rb_node
*parent
= NULL
;
479 struct mem_cgroup_per_zone
*mz_node
;
484 mz
->usage_in_excess
= new_usage_in_excess
;
485 if (!mz
->usage_in_excess
)
489 mz_node
= rb_entry(parent
, struct mem_cgroup_per_zone
,
491 if (mz
->usage_in_excess
< mz_node
->usage_in_excess
)
494 * We can't avoid mem cgroups that are over their soft
495 * limit by the same amount
497 else if (mz
->usage_in_excess
>= mz_node
->usage_in_excess
)
500 rb_link_node(&mz
->tree_node
, parent
, p
);
501 rb_insert_color(&mz
->tree_node
, &mctz
->rb_root
);
506 __mem_cgroup_remove_exceeded(struct mem_cgroup
*memcg
,
507 struct mem_cgroup_per_zone
*mz
,
508 struct mem_cgroup_tree_per_zone
*mctz
)
512 rb_erase(&mz
->tree_node
, &mctz
->rb_root
);
517 mem_cgroup_remove_exceeded(struct mem_cgroup
*memcg
,
518 struct mem_cgroup_per_zone
*mz
,
519 struct mem_cgroup_tree_per_zone
*mctz
)
521 spin_lock(&mctz
->lock
);
522 __mem_cgroup_remove_exceeded(memcg
, mz
, mctz
);
523 spin_unlock(&mctz
->lock
);
527 static void mem_cgroup_update_tree(struct mem_cgroup
*memcg
, struct page
*page
)
529 unsigned long long excess
;
530 struct mem_cgroup_per_zone
*mz
;
531 struct mem_cgroup_tree_per_zone
*mctz
;
532 int nid
= page_to_nid(page
);
533 int zid
= page_zonenum(page
);
534 mctz
= soft_limit_tree_from_page(page
);
537 * Necessary to update all ancestors when hierarchy is used.
538 * because their event counter is not touched.
540 for (; memcg
; memcg
= parent_mem_cgroup(memcg
)) {
541 mz
= mem_cgroup_zoneinfo(memcg
, nid
, zid
);
542 excess
= res_counter_soft_limit_excess(&memcg
->res
);
544 * We have to update the tree if mz is on RB-tree or
545 * mem is over its softlimit.
547 if (excess
|| mz
->on_tree
) {
548 spin_lock(&mctz
->lock
);
549 /* if on-tree, remove it */
551 __mem_cgroup_remove_exceeded(memcg
, mz
, mctz
);
553 * Insert again. mz->usage_in_excess will be updated.
554 * If excess is 0, no tree ops.
556 __mem_cgroup_insert_exceeded(memcg
, mz
, mctz
, excess
);
557 spin_unlock(&mctz
->lock
);
562 static void mem_cgroup_remove_from_trees(struct mem_cgroup
*memcg
)
565 struct mem_cgroup_per_zone
*mz
;
566 struct mem_cgroup_tree_per_zone
*mctz
;
568 for_each_node_state(node
, N_POSSIBLE
) {
569 for (zone
= 0; zone
< MAX_NR_ZONES
; zone
++) {
570 mz
= mem_cgroup_zoneinfo(memcg
, node
, zone
);
571 mctz
= soft_limit_tree_node_zone(node
, zone
);
572 mem_cgroup_remove_exceeded(memcg
, mz
, mctz
);
577 static struct mem_cgroup_per_zone
*
578 __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone
*mctz
)
580 struct rb_node
*rightmost
= NULL
;
581 struct mem_cgroup_per_zone
*mz
;
585 rightmost
= rb_last(&mctz
->rb_root
);
587 goto done
; /* Nothing to reclaim from */
589 mz
= rb_entry(rightmost
, struct mem_cgroup_per_zone
, tree_node
);
591 * Remove the node now but someone else can add it back,
592 * we will to add it back at the end of reclaim to its correct
593 * position in the tree.
595 __mem_cgroup_remove_exceeded(mz
->mem
, mz
, mctz
);
596 if (!res_counter_soft_limit_excess(&mz
->mem
->res
) ||
597 !css_tryget(&mz
->mem
->css
))
603 static struct mem_cgroup_per_zone
*
604 mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone
*mctz
)
606 struct mem_cgroup_per_zone
*mz
;
608 spin_lock(&mctz
->lock
);
609 mz
= __mem_cgroup_largest_soft_limit_node(mctz
);
610 spin_unlock(&mctz
->lock
);
615 * Implementation Note: reading percpu statistics for memcg.
617 * Both of vmstat[] and percpu_counter has threshold and do periodic
618 * synchronization to implement "quick" read. There are trade-off between
619 * reading cost and precision of value. Then, we may have a chance to implement
620 * a periodic synchronizion of counter in memcg's counter.
622 * But this _read() function is used for user interface now. The user accounts
623 * memory usage by memory cgroup and he _always_ requires exact value because
624 * he accounts memory. Even if we provide quick-and-fuzzy read, we always
625 * have to visit all online cpus and make sum. So, for now, unnecessary
626 * synchronization is not implemented. (just implemented for cpu hotplug)
628 * If there are kernel internal actions which can make use of some not-exact
629 * value, and reading all cpu value can be performance bottleneck in some
630 * common workload, threashold and synchonization as vmstat[] should be
633 static long mem_cgroup_read_stat(struct mem_cgroup
*memcg
,
634 enum mem_cgroup_stat_index idx
)
640 for_each_online_cpu(cpu
)
641 val
+= per_cpu(memcg
->stat
->count
[idx
], cpu
);
642 #ifdef CONFIG_HOTPLUG_CPU
643 spin_lock(&memcg
->pcp_counter_lock
);
644 val
+= memcg
->nocpu_base
.count
[idx
];
645 spin_unlock(&memcg
->pcp_counter_lock
);
651 static void mem_cgroup_swap_statistics(struct mem_cgroup
*memcg
,
654 int val
= (charge
) ? 1 : -1;
655 this_cpu_add(memcg
->stat
->count
[MEM_CGROUP_STAT_SWAPOUT
], val
);
658 void mem_cgroup_pgfault(struct mem_cgroup
*memcg
, int val
)
660 this_cpu_add(memcg
->stat
->events
[MEM_CGROUP_EVENTS_PGFAULT
], val
);
663 void mem_cgroup_pgmajfault(struct mem_cgroup
*memcg
, int val
)
665 this_cpu_add(memcg
->stat
->events
[MEM_CGROUP_EVENTS_PGMAJFAULT
], val
);
668 static unsigned long mem_cgroup_read_events(struct mem_cgroup
*memcg
,
669 enum mem_cgroup_events_index idx
)
671 unsigned long val
= 0;
674 for_each_online_cpu(cpu
)
675 val
+= per_cpu(memcg
->stat
->events
[idx
], cpu
);
676 #ifdef CONFIG_HOTPLUG_CPU
677 spin_lock(&memcg
->pcp_counter_lock
);
678 val
+= memcg
->nocpu_base
.events
[idx
];
679 spin_unlock(&memcg
->pcp_counter_lock
);
684 static void mem_cgroup_charge_statistics(struct mem_cgroup
*memcg
,
685 bool file
, int nr_pages
)
690 __this_cpu_add(memcg
->stat
->count
[MEM_CGROUP_STAT_CACHE
],
693 __this_cpu_add(memcg
->stat
->count
[MEM_CGROUP_STAT_RSS
],
696 /* pagein of a big page is an event. So, ignore page size */
698 __this_cpu_inc(memcg
->stat
->events
[MEM_CGROUP_EVENTS_PGPGIN
]);
700 __this_cpu_inc(memcg
->stat
->events
[MEM_CGROUP_EVENTS_PGPGOUT
]);
701 nr_pages
= -nr_pages
; /* for event */
704 __this_cpu_add(memcg
->stat
->events
[MEM_CGROUP_EVENTS_COUNT
], nr_pages
);
710 mem_cgroup_zone_nr_lru_pages(struct mem_cgroup
*memcg
, int nid
, int zid
,
711 unsigned int lru_mask
)
713 struct mem_cgroup_per_zone
*mz
;
715 unsigned long ret
= 0;
717 mz
= mem_cgroup_zoneinfo(memcg
, nid
, zid
);
720 if (BIT(l
) & lru_mask
)
721 ret
+= MEM_CGROUP_ZSTAT(mz
, l
);
727 mem_cgroup_node_nr_lru_pages(struct mem_cgroup
*memcg
,
728 int nid
, unsigned int lru_mask
)
733 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++)
734 total
+= mem_cgroup_zone_nr_lru_pages(memcg
,
740 static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup
*memcg
,
741 unsigned int lru_mask
)
746 for_each_node_state(nid
, N_HIGH_MEMORY
)
747 total
+= mem_cgroup_node_nr_lru_pages(memcg
, nid
, lru_mask
);
751 static bool __memcg_event_check(struct mem_cgroup
*memcg
, int target
)
753 unsigned long val
, next
;
755 val
= __this_cpu_read(memcg
->stat
->events
[MEM_CGROUP_EVENTS_COUNT
]);
756 next
= __this_cpu_read(memcg
->stat
->targets
[target
]);
757 /* from time_after() in jiffies.h */
758 return ((long)next
- (long)val
< 0);
761 static void __mem_cgroup_target_update(struct mem_cgroup
*memcg
, int target
)
763 unsigned long val
, next
;
765 val
= __this_cpu_read(memcg
->stat
->events
[MEM_CGROUP_EVENTS_COUNT
]);
768 case MEM_CGROUP_TARGET_THRESH
:
769 next
= val
+ THRESHOLDS_EVENTS_TARGET
;
771 case MEM_CGROUP_TARGET_SOFTLIMIT
:
772 next
= val
+ SOFTLIMIT_EVENTS_TARGET
;
774 case MEM_CGROUP_TARGET_NUMAINFO
:
775 next
= val
+ NUMAINFO_EVENTS_TARGET
;
781 __this_cpu_write(memcg
->stat
->targets
[target
], next
);
785 * Check events in order.
788 static void memcg_check_events(struct mem_cgroup
*memcg
, struct page
*page
)
791 /* threshold event is triggered in finer grain than soft limit */
792 if (unlikely(__memcg_event_check(memcg
, MEM_CGROUP_TARGET_THRESH
))) {
793 mem_cgroup_threshold(memcg
);
794 __mem_cgroup_target_update(memcg
, MEM_CGROUP_TARGET_THRESH
);
795 if (unlikely(__memcg_event_check(memcg
,
796 MEM_CGROUP_TARGET_SOFTLIMIT
))) {
797 mem_cgroup_update_tree(memcg
, page
);
798 __mem_cgroup_target_update(memcg
,
799 MEM_CGROUP_TARGET_SOFTLIMIT
);
802 if (unlikely(__memcg_event_check(memcg
,
803 MEM_CGROUP_TARGET_NUMAINFO
))) {
804 atomic_inc(&memcg
->numainfo_events
);
805 __mem_cgroup_target_update(memcg
,
806 MEM_CGROUP_TARGET_NUMAINFO
);
813 struct mem_cgroup
*mem_cgroup_from_cont(struct cgroup
*cont
)
815 return container_of(cgroup_subsys_state(cont
,
816 mem_cgroup_subsys_id
), struct mem_cgroup
,
820 struct mem_cgroup
*mem_cgroup_from_task(struct task_struct
*p
)
823 * mm_update_next_owner() may clear mm->owner to NULL
824 * if it races with swapoff, page migration, etc.
825 * So this can be called with p == NULL.
830 return container_of(task_subsys_state(p
, mem_cgroup_subsys_id
),
831 struct mem_cgroup
, css
);
834 struct mem_cgroup
*try_get_mem_cgroup_from_mm(struct mm_struct
*mm
)
836 struct mem_cgroup
*memcg
= NULL
;
841 * Because we have no locks, mm->owner's may be being moved to other
842 * cgroup. We use css_tryget() here even if this looks
843 * pessimistic (rather than adding locks here).
847 memcg
= mem_cgroup_from_task(rcu_dereference(mm
->owner
));
848 if (unlikely(!memcg
))
850 } while (!css_tryget(&memcg
->css
));
856 * mem_cgroup_iter - iterate over memory cgroup hierarchy
857 * @root: hierarchy root
858 * @prev: previously returned memcg, NULL on first invocation
859 * @reclaim: cookie for shared reclaim walks, NULL for full walks
861 * Returns references to children of the hierarchy below @root, or
862 * @root itself, or %NULL after a full round-trip.
864 * Caller must pass the return value in @prev on subsequent
865 * invocations for reference counting, or use mem_cgroup_iter_break()
866 * to cancel a hierarchy walk before the round-trip is complete.
868 * Reclaimers can specify a zone and a priority level in @reclaim to
869 * divide up the memcgs in the hierarchy among all concurrent
870 * reclaimers operating on the same zone and priority.
872 struct mem_cgroup
*mem_cgroup_iter(struct mem_cgroup
*root
,
873 struct mem_cgroup
*prev
,
874 struct mem_cgroup_reclaim_cookie
*reclaim
)
876 struct mem_cgroup
*memcg
= NULL
;
879 if (mem_cgroup_disabled())
883 root
= root_mem_cgroup
;
885 if (prev
&& !reclaim
)
886 id
= css_id(&prev
->css
);
888 if (prev
&& prev
!= root
)
891 if (!root
->use_hierarchy
&& root
!= root_mem_cgroup
) {
898 struct mem_cgroup_reclaim_iter
*uninitialized_var(iter
);
899 struct cgroup_subsys_state
*css
;
902 int nid
= zone_to_nid(reclaim
->zone
);
903 int zid
= zone_idx(reclaim
->zone
);
904 struct mem_cgroup_per_zone
*mz
;
906 mz
= mem_cgroup_zoneinfo(root
, nid
, zid
);
907 iter
= &mz
->reclaim_iter
[reclaim
->priority
];
908 if (prev
&& reclaim
->generation
!= iter
->generation
)
914 css
= css_get_next(&mem_cgroup_subsys
, id
+ 1, &root
->css
, &id
);
916 if (css
== &root
->css
|| css_tryget(css
))
917 memcg
= container_of(css
,
918 struct mem_cgroup
, css
);
927 else if (!prev
&& memcg
)
928 reclaim
->generation
= iter
->generation
;
938 * mem_cgroup_iter_break - abort a hierarchy walk prematurely
939 * @root: hierarchy root
940 * @prev: last visited hierarchy member as returned by mem_cgroup_iter()
942 void mem_cgroup_iter_break(struct mem_cgroup
*root
,
943 struct mem_cgroup
*prev
)
946 root
= root_mem_cgroup
;
947 if (prev
&& prev
!= root
)
952 * Iteration constructs for visiting all cgroups (under a tree). If
953 * loops are exited prematurely (break), mem_cgroup_iter_break() must
954 * be used for reference counting.
956 #define for_each_mem_cgroup_tree(iter, root) \
957 for (iter = mem_cgroup_iter(root, NULL, NULL); \
959 iter = mem_cgroup_iter(root, iter, NULL))
961 #define for_each_mem_cgroup(iter) \
962 for (iter = mem_cgroup_iter(NULL, NULL, NULL); \
964 iter = mem_cgroup_iter(NULL, iter, NULL))
966 static inline bool mem_cgroup_is_root(struct mem_cgroup
*memcg
)
968 return (memcg
== root_mem_cgroup
);
971 void mem_cgroup_count_vm_event(struct mm_struct
*mm
, enum vm_event_item idx
)
973 struct mem_cgroup
*memcg
;
979 memcg
= mem_cgroup_from_task(rcu_dereference(mm
->owner
));
980 if (unlikely(!memcg
))
985 mem_cgroup_pgmajfault(memcg
, 1);
988 mem_cgroup_pgfault(memcg
, 1);
996 EXPORT_SYMBOL(mem_cgroup_count_vm_event
);
999 * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg
1000 * @zone: zone of the wanted lruvec
1001 * @mem: memcg of the wanted lruvec
1003 * Returns the lru list vector holding pages for the given @zone and
1004 * @mem. This can be the global zone lruvec, if the memory controller
1007 struct lruvec
*mem_cgroup_zone_lruvec(struct zone
*zone
,
1008 struct mem_cgroup
*memcg
)
1010 struct mem_cgroup_per_zone
*mz
;
1012 if (mem_cgroup_disabled())
1013 return &zone
->lruvec
;
1015 mz
= mem_cgroup_zoneinfo(memcg
, zone_to_nid(zone
), zone_idx(zone
));
1020 * Following LRU functions are allowed to be used without PCG_LOCK.
1021 * Operations are called by routine of global LRU independently from memcg.
1022 * What we have to take care of here is validness of pc->mem_cgroup.
1024 * Changes to pc->mem_cgroup happens when
1027 * In typical case, "charge" is done before add-to-lru. Exception is SwapCache.
1028 * It is added to LRU before charge.
1029 * If PCG_USED bit is not set, page_cgroup is not added to this private LRU.
1030 * When moving account, the page is not on LRU. It's isolated.
1034 * mem_cgroup_lru_add_list - account for adding an lru page and return lruvec
1035 * @zone: zone of the page
1039 * This function accounts for @page being added to @lru, and returns
1040 * the lruvec for the given @zone and the memcg @page is charged to.
1042 * The callsite is then responsible for physically linking the page to
1043 * the returned lruvec->lists[@lru].
1045 struct lruvec
*mem_cgroup_lru_add_list(struct zone
*zone
, struct page
*page
,
1048 struct mem_cgroup_per_zone
*mz
;
1049 struct mem_cgroup
*memcg
;
1050 struct page_cgroup
*pc
;
1052 if (mem_cgroup_disabled())
1053 return &zone
->lruvec
;
1055 pc
= lookup_page_cgroup(page
);
1056 VM_BUG_ON(PageCgroupAcctLRU(pc
));
1059 * SetPageLRU SetPageCgroupUsed
1061 * PageCgroupUsed && add to memcg LRU PageLRU && add to memcg LRU
1063 * Ensure that one of the two sides adds the page to the memcg
1064 * LRU during a race.
1068 * If the page is uncharged, it may be freed soon, but it
1069 * could also be swap cache (readahead, swapoff) that needs to
1070 * be reclaimable in the future. root_mem_cgroup will babysit
1071 * it for the time being.
1073 if (PageCgroupUsed(pc
)) {
1074 /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */
1076 memcg
= pc
->mem_cgroup
;
1077 SetPageCgroupAcctLRU(pc
);
1079 memcg
= root_mem_cgroup
;
1080 mz
= page_cgroup_zoneinfo(memcg
, page
);
1081 /* compound_order() is stabilized through lru_lock */
1082 MEM_CGROUP_ZSTAT(mz
, lru
) += 1 << compound_order(page
);
1087 * mem_cgroup_lru_del_list - account for removing an lru page
1091 * This function accounts for @page being removed from @lru.
1093 * The callsite is then responsible for physically unlinking
1096 void mem_cgroup_lru_del_list(struct page
*page
, enum lru_list lru
)
1098 struct mem_cgroup_per_zone
*mz
;
1099 struct mem_cgroup
*memcg
;
1100 struct page_cgroup
*pc
;
1102 if (mem_cgroup_disabled())
1105 pc
= lookup_page_cgroup(page
);
1107 * root_mem_cgroup babysits uncharged LRU pages, but
1108 * PageCgroupUsed is cleared when the page is about to get
1109 * freed. PageCgroupAcctLRU remembers whether the
1110 * LRU-accounting happened against pc->mem_cgroup or
1113 if (TestClearPageCgroupAcctLRU(pc
)) {
1114 VM_BUG_ON(!pc
->mem_cgroup
);
1115 memcg
= pc
->mem_cgroup
;
1117 memcg
= root_mem_cgroup
;
1118 mz
= page_cgroup_zoneinfo(memcg
, page
);
1119 /* huge page split is done under lru_lock. so, we have no races. */
1120 MEM_CGROUP_ZSTAT(mz
, lru
) -= 1 << compound_order(page
);
1123 void mem_cgroup_lru_del(struct page
*page
)
1125 mem_cgroup_lru_del_list(page
, page_lru(page
));
1129 * mem_cgroup_lru_move_lists - account for moving a page between lrus
1130 * @zone: zone of the page
1132 * @from: current lru
1135 * This function accounts for @page being moved between the lrus @from
1136 * and @to, and returns the lruvec for the given @zone and the memcg
1137 * @page is charged to.
1139 * The callsite is then responsible for physically relinking
1140 * @page->lru to the returned lruvec->lists[@to].
1142 struct lruvec
*mem_cgroup_lru_move_lists(struct zone
*zone
,
1147 /* XXX: Optimize this, especially for @from == @to */
1148 mem_cgroup_lru_del_list(page
, from
);
1149 return mem_cgroup_lru_add_list(zone
, page
, to
);
1153 * At handling SwapCache and other FUSE stuff, pc->mem_cgroup may be changed
1154 * while it's linked to lru because the page may be reused after it's fully
1155 * uncharged. To handle that, unlink page_cgroup from LRU when charge it again.
1156 * It's done under lock_page and expected that zone->lru_lock isnever held.
1158 static void mem_cgroup_lru_del_before_commit(struct page
*page
)
1161 unsigned long flags
;
1162 struct zone
*zone
= page_zone(page
);
1163 struct page_cgroup
*pc
= lookup_page_cgroup(page
);
1166 * Doing this check without taking ->lru_lock seems wrong but this
1167 * is safe. Because if page_cgroup's USED bit is unset, the page
1168 * will not be added to any memcg's LRU. If page_cgroup's USED bit is
1169 * set, the commit after this will fail, anyway.
1170 * This all charge/uncharge is done under some mutual execustion.
1171 * So, we don't need to taking care of changes in USED bit.
1173 if (likely(!PageLRU(page
)))
1176 spin_lock_irqsave(&zone
->lru_lock
, flags
);
1177 lru
= page_lru(page
);
1179 * The uncharged page could still be registered to the LRU of
1180 * the stale pc->mem_cgroup.
1182 * As pc->mem_cgroup is about to get overwritten, the old LRU
1183 * accounting needs to be taken care of. Let root_mem_cgroup
1184 * babysit the page until the new memcg is responsible for it.
1186 * The PCG_USED bit is guarded by lock_page() as the page is
1187 * swapcache/pagecache.
1189 if (PageLRU(page
) && PageCgroupAcctLRU(pc
) && !PageCgroupUsed(pc
)) {
1190 del_page_from_lru_list(zone
, page
, lru
);
1191 add_page_to_lru_list(zone
, page
, lru
);
1193 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
1196 static void mem_cgroup_lru_add_after_commit(struct page
*page
)
1199 unsigned long flags
;
1200 struct zone
*zone
= page_zone(page
);
1201 struct page_cgroup
*pc
= lookup_page_cgroup(page
);
1204 * SetPageLRU SetPageCgroupUsed
1206 * PageCgroupUsed && add to memcg LRU PageLRU && add to memcg LRU
1208 * Ensure that one of the two sides adds the page to the memcg
1209 * LRU during a race.
1212 /* taking care of that the page is added to LRU while we commit it */
1213 if (likely(!PageLRU(page
)))
1215 spin_lock_irqsave(&zone
->lru_lock
, flags
);
1216 lru
= page_lru(page
);
1218 * If the page is not on the LRU, someone will soon put it
1219 * there. If it is, and also already accounted for on the
1220 * memcg-side, it must be on the right lruvec as setting
1221 * pc->mem_cgroup and PageCgroupUsed is properly ordered.
1222 * Otherwise, root_mem_cgroup has been babysitting the page
1223 * during the charge. Move it to the new memcg now.
1225 if (PageLRU(page
) && !PageCgroupAcctLRU(pc
)) {
1226 del_page_from_lru_list(zone
, page
, lru
);
1227 add_page_to_lru_list(zone
, page
, lru
);
1229 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
1233 * Checks whether given mem is same or in the root_mem_cgroup's
1236 static bool mem_cgroup_same_or_subtree(const struct mem_cgroup
*root_memcg
,
1237 struct mem_cgroup
*memcg
)
1239 if (root_memcg
!= memcg
) {
1240 return (root_memcg
->use_hierarchy
&&
1241 css_is_ancestor(&memcg
->css
, &root_memcg
->css
));
1247 int task_in_mem_cgroup(struct task_struct
*task
, const struct mem_cgroup
*memcg
)
1250 struct mem_cgroup
*curr
= NULL
;
1251 struct task_struct
*p
;
1253 p
= find_lock_task_mm(task
);
1256 curr
= try_get_mem_cgroup_from_mm(p
->mm
);
1261 * We should check use_hierarchy of "memcg" not "curr". Because checking
1262 * use_hierarchy of "curr" here make this function true if hierarchy is
1263 * enabled in "curr" and "curr" is a child of "memcg" in *cgroup*
1264 * hierarchy(even if use_hierarchy is disabled in "memcg").
1266 ret
= mem_cgroup_same_or_subtree(memcg
, curr
);
1267 css_put(&curr
->css
);
1271 int mem_cgroup_inactive_anon_is_low(struct mem_cgroup
*memcg
, struct zone
*zone
)
1273 unsigned long inactive_ratio
;
1274 int nid
= zone_to_nid(zone
);
1275 int zid
= zone_idx(zone
);
1276 unsigned long inactive
;
1277 unsigned long active
;
1280 inactive
= mem_cgroup_zone_nr_lru_pages(memcg
, nid
, zid
,
1281 BIT(LRU_INACTIVE_ANON
));
1282 active
= mem_cgroup_zone_nr_lru_pages(memcg
, nid
, zid
,
1283 BIT(LRU_ACTIVE_ANON
));
1285 gb
= (inactive
+ active
) >> (30 - PAGE_SHIFT
);
1287 inactive_ratio
= int_sqrt(10 * gb
);
1291 return inactive
* inactive_ratio
< active
;
1294 int mem_cgroup_inactive_file_is_low(struct mem_cgroup
*memcg
, struct zone
*zone
)
1296 unsigned long active
;
1297 unsigned long inactive
;
1298 int zid
= zone_idx(zone
);
1299 int nid
= zone_to_nid(zone
);
1301 inactive
= mem_cgroup_zone_nr_lru_pages(memcg
, nid
, zid
,
1302 BIT(LRU_INACTIVE_FILE
));
1303 active
= mem_cgroup_zone_nr_lru_pages(memcg
, nid
, zid
,
1304 BIT(LRU_ACTIVE_FILE
));
1306 return (active
> inactive
);
1309 struct zone_reclaim_stat
*mem_cgroup_get_reclaim_stat(struct mem_cgroup
*memcg
,
1312 int nid
= zone_to_nid(zone
);
1313 int zid
= zone_idx(zone
);
1314 struct mem_cgroup_per_zone
*mz
= mem_cgroup_zoneinfo(memcg
, nid
, zid
);
1316 return &mz
->reclaim_stat
;
1319 struct zone_reclaim_stat
*
1320 mem_cgroup_get_reclaim_stat_from_page(struct page
*page
)
1322 struct page_cgroup
*pc
;
1323 struct mem_cgroup_per_zone
*mz
;
1325 if (mem_cgroup_disabled())
1328 pc
= lookup_page_cgroup(page
);
1329 if (!PageCgroupUsed(pc
))
1331 /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */
1333 mz
= page_cgroup_zoneinfo(pc
->mem_cgroup
, page
);
1334 return &mz
->reclaim_stat
;
1337 #define mem_cgroup_from_res_counter(counter, member) \
1338 container_of(counter, struct mem_cgroup, member)
1341 * mem_cgroup_margin - calculate chargeable space of a memory cgroup
1342 * @mem: the memory cgroup
1344 * Returns the maximum amount of memory @mem can be charged with, in
1347 static unsigned long mem_cgroup_margin(struct mem_cgroup
*memcg
)
1349 unsigned long long margin
;
1351 margin
= res_counter_margin(&memcg
->res
);
1352 if (do_swap_account
)
1353 margin
= min(margin
, res_counter_margin(&memcg
->memsw
));
1354 return margin
>> PAGE_SHIFT
;
1357 int mem_cgroup_swappiness(struct mem_cgroup
*memcg
)
1359 struct cgroup
*cgrp
= memcg
->css
.cgroup
;
1362 if (cgrp
->parent
== NULL
)
1363 return vm_swappiness
;
1365 return memcg
->swappiness
;
1368 static void mem_cgroup_start_move(struct mem_cgroup
*memcg
)
1373 spin_lock(&memcg
->pcp_counter_lock
);
1374 for_each_online_cpu(cpu
)
1375 per_cpu(memcg
->stat
->count
[MEM_CGROUP_ON_MOVE
], cpu
) += 1;
1376 memcg
->nocpu_base
.count
[MEM_CGROUP_ON_MOVE
] += 1;
1377 spin_unlock(&memcg
->pcp_counter_lock
);
1383 static void mem_cgroup_end_move(struct mem_cgroup
*memcg
)
1390 spin_lock(&memcg
->pcp_counter_lock
);
1391 for_each_online_cpu(cpu
)
1392 per_cpu(memcg
->stat
->count
[MEM_CGROUP_ON_MOVE
], cpu
) -= 1;
1393 memcg
->nocpu_base
.count
[MEM_CGROUP_ON_MOVE
] -= 1;
1394 spin_unlock(&memcg
->pcp_counter_lock
);
1398 * 2 routines for checking "mem" is under move_account() or not.
1400 * mem_cgroup_stealed() - checking a cgroup is mc.from or not. This is used
1401 * for avoiding race in accounting. If true,
1402 * pc->mem_cgroup may be overwritten.
1404 * mem_cgroup_under_move() - checking a cgroup is mc.from or mc.to or
1405 * under hierarchy of moving cgroups. This is for
1406 * waiting at hith-memory prressure caused by "move".
1409 static bool mem_cgroup_stealed(struct mem_cgroup
*memcg
)
1411 VM_BUG_ON(!rcu_read_lock_held());
1412 return this_cpu_read(memcg
->stat
->count
[MEM_CGROUP_ON_MOVE
]) > 0;
1415 static bool mem_cgroup_under_move(struct mem_cgroup
*memcg
)
1417 struct mem_cgroup
*from
;
1418 struct mem_cgroup
*to
;
1421 * Unlike task_move routines, we access mc.to, mc.from not under
1422 * mutual exclusion by cgroup_mutex. Here, we take spinlock instead.
1424 spin_lock(&mc
.lock
);
1430 ret
= mem_cgroup_same_or_subtree(memcg
, from
)
1431 || mem_cgroup_same_or_subtree(memcg
, to
);
1433 spin_unlock(&mc
.lock
);
1437 static bool mem_cgroup_wait_acct_move(struct mem_cgroup
*memcg
)
1439 if (mc
.moving_task
&& current
!= mc
.moving_task
) {
1440 if (mem_cgroup_under_move(memcg
)) {
1442 prepare_to_wait(&mc
.waitq
, &wait
, TASK_INTERRUPTIBLE
);
1443 /* moving charge context might have finished. */
1446 finish_wait(&mc
.waitq
, &wait
);
1454 * mem_cgroup_print_oom_info: Called from OOM with tasklist_lock held in read mode.
1455 * @memcg: The memory cgroup that went over limit
1456 * @p: Task that is going to be killed
1458 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1461 void mem_cgroup_print_oom_info(struct mem_cgroup
*memcg
, struct task_struct
*p
)
1463 struct cgroup
*task_cgrp
;
1464 struct cgroup
*mem_cgrp
;
1466 * Need a buffer in BSS, can't rely on allocations. The code relies
1467 * on the assumption that OOM is serialized for memory controller.
1468 * If this assumption is broken, revisit this code.
1470 static char memcg_name
[PATH_MAX
];
1479 mem_cgrp
= memcg
->css
.cgroup
;
1480 task_cgrp
= task_cgroup(p
, mem_cgroup_subsys_id
);
1482 ret
= cgroup_path(task_cgrp
, memcg_name
, PATH_MAX
);
1485 * Unfortunately, we are unable to convert to a useful name
1486 * But we'll still print out the usage information
1493 printk(KERN_INFO
"Task in %s killed", memcg_name
);
1496 ret
= cgroup_path(mem_cgrp
, memcg_name
, PATH_MAX
);
1504 * Continues from above, so we don't need an KERN_ level
1506 printk(KERN_CONT
" as a result of limit of %s\n", memcg_name
);
1509 printk(KERN_INFO
"memory: usage %llukB, limit %llukB, failcnt %llu\n",
1510 res_counter_read_u64(&memcg
->res
, RES_USAGE
) >> 10,
1511 res_counter_read_u64(&memcg
->res
, RES_LIMIT
) >> 10,
1512 res_counter_read_u64(&memcg
->res
, RES_FAILCNT
));
1513 printk(KERN_INFO
"memory+swap: usage %llukB, limit %llukB, "
1515 res_counter_read_u64(&memcg
->memsw
, RES_USAGE
) >> 10,
1516 res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
) >> 10,
1517 res_counter_read_u64(&memcg
->memsw
, RES_FAILCNT
));
1521 * This function returns the number of memcg under hierarchy tree. Returns
1522 * 1(self count) if no children.
1524 static int mem_cgroup_count_children(struct mem_cgroup
*memcg
)
1527 struct mem_cgroup
*iter
;
1529 for_each_mem_cgroup_tree(iter
, memcg
)
1535 * Return the memory (and swap, if configured) limit for a memcg.
1537 u64
mem_cgroup_get_limit(struct mem_cgroup
*memcg
)
1542 limit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
1543 limit
+= total_swap_pages
<< PAGE_SHIFT
;
1545 memsw
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
1547 * If memsw is finite and limits the amount of swap space available
1548 * to this memcg, return that limit.
1550 return min(limit
, memsw
);
1553 static unsigned long mem_cgroup_reclaim(struct mem_cgroup
*memcg
,
1555 unsigned long flags
)
1557 unsigned long total
= 0;
1558 bool noswap
= false;
1561 if (flags
& MEM_CGROUP_RECLAIM_NOSWAP
)
1563 if (!(flags
& MEM_CGROUP_RECLAIM_SHRINK
) && memcg
->memsw_is_minimum
)
1566 for (loop
= 0; loop
< MEM_CGROUP_MAX_RECLAIM_LOOPS
; loop
++) {
1568 drain_all_stock_async(memcg
);
1569 total
+= try_to_free_mem_cgroup_pages(memcg
, gfp_mask
, noswap
);
1571 * Allow limit shrinkers, which are triggered directly
1572 * by userspace, to catch signals and stop reclaim
1573 * after minimal progress, regardless of the margin.
1575 if (total
&& (flags
& MEM_CGROUP_RECLAIM_SHRINK
))
1577 if (mem_cgroup_margin(memcg
))
1580 * If nothing was reclaimed after two attempts, there
1581 * may be no reclaimable pages in this hierarchy.
1590 * test_mem_cgroup_node_reclaimable
1591 * @mem: the target memcg
1592 * @nid: the node ID to be checked.
1593 * @noswap : specify true here if the user wants flle only information.
1595 * This function returns whether the specified memcg contains any
1596 * reclaimable pages on a node. Returns true if there are any reclaimable
1597 * pages in the node.
1599 static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup
*memcg
,
1600 int nid
, bool noswap
)
1602 if (mem_cgroup_node_nr_lru_pages(memcg
, nid
, LRU_ALL_FILE
))
1604 if (noswap
|| !total_swap_pages
)
1606 if (mem_cgroup_node_nr_lru_pages(memcg
, nid
, LRU_ALL_ANON
))
1611 #if MAX_NUMNODES > 1
1614 * Always updating the nodemask is not very good - even if we have an empty
1615 * list or the wrong list here, we can start from some node and traverse all
1616 * nodes based on the zonelist. So update the list loosely once per 10 secs.
1619 static void mem_cgroup_may_update_nodemask(struct mem_cgroup
*memcg
)
1623 * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET
1624 * pagein/pageout changes since the last update.
1626 if (!atomic_read(&memcg
->numainfo_events
))
1628 if (atomic_inc_return(&memcg
->numainfo_updating
) > 1)
1631 /* make a nodemask where this memcg uses memory from */
1632 memcg
->scan_nodes
= node_states
[N_HIGH_MEMORY
];
1634 for_each_node_mask(nid
, node_states
[N_HIGH_MEMORY
]) {
1636 if (!test_mem_cgroup_node_reclaimable(memcg
, nid
, false))
1637 node_clear(nid
, memcg
->scan_nodes
);
1640 atomic_set(&memcg
->numainfo_events
, 0);
1641 atomic_set(&memcg
->numainfo_updating
, 0);
1645 * Selecting a node where we start reclaim from. Because what we need is just
1646 * reducing usage counter, start from anywhere is O,K. Considering
1647 * memory reclaim from current node, there are pros. and cons.
1649 * Freeing memory from current node means freeing memory from a node which
1650 * we'll use or we've used. So, it may make LRU bad. And if several threads
1651 * hit limits, it will see a contention on a node. But freeing from remote
1652 * node means more costs for memory reclaim because of memory latency.
1654 * Now, we use round-robin. Better algorithm is welcomed.
1656 int mem_cgroup_select_victim_node(struct mem_cgroup
*memcg
)
1660 mem_cgroup_may_update_nodemask(memcg
);
1661 node
= memcg
->last_scanned_node
;
1663 node
= next_node(node
, memcg
->scan_nodes
);
1664 if (node
== MAX_NUMNODES
)
1665 node
= first_node(memcg
->scan_nodes
);
1667 * We call this when we hit limit, not when pages are added to LRU.
1668 * No LRU may hold pages because all pages are UNEVICTABLE or
1669 * memcg is too small and all pages are not on LRU. In that case,
1670 * we use curret node.
1672 if (unlikely(node
== MAX_NUMNODES
))
1673 node
= numa_node_id();
1675 memcg
->last_scanned_node
= node
;
1680 * Check all nodes whether it contains reclaimable pages or not.
1681 * For quick scan, we make use of scan_nodes. This will allow us to skip
1682 * unused nodes. But scan_nodes is lazily updated and may not cotain
1683 * enough new information. We need to do double check.
1685 bool mem_cgroup_reclaimable(struct mem_cgroup
*memcg
, bool noswap
)
1690 * quick check...making use of scan_node.
1691 * We can skip unused nodes.
1693 if (!nodes_empty(memcg
->scan_nodes
)) {
1694 for (nid
= first_node(memcg
->scan_nodes
);
1696 nid
= next_node(nid
, memcg
->scan_nodes
)) {
1698 if (test_mem_cgroup_node_reclaimable(memcg
, nid
, noswap
))
1703 * Check rest of nodes.
1705 for_each_node_state(nid
, N_HIGH_MEMORY
) {
1706 if (node_isset(nid
, memcg
->scan_nodes
))
1708 if (test_mem_cgroup_node_reclaimable(memcg
, nid
, noswap
))
1715 int mem_cgroup_select_victim_node(struct mem_cgroup
*memcg
)
1720 bool mem_cgroup_reclaimable(struct mem_cgroup
*memcg
, bool noswap
)
1722 return test_mem_cgroup_node_reclaimable(memcg
, 0, noswap
);
1726 static int mem_cgroup_soft_reclaim(struct mem_cgroup
*root_memcg
,
1729 unsigned long *total_scanned
)
1731 struct mem_cgroup
*victim
= NULL
;
1734 unsigned long excess
;
1735 unsigned long nr_scanned
;
1736 struct mem_cgroup_reclaim_cookie reclaim
= {
1741 excess
= res_counter_soft_limit_excess(&root_memcg
->res
) >> PAGE_SHIFT
;
1744 victim
= mem_cgroup_iter(root_memcg
, victim
, &reclaim
);
1749 * If we have not been able to reclaim
1750 * anything, it might because there are
1751 * no reclaimable pages under this hierarchy
1756 * We want to do more targeted reclaim.
1757 * excess >> 2 is not to excessive so as to
1758 * reclaim too much, nor too less that we keep
1759 * coming back to reclaim from this cgroup
1761 if (total
>= (excess
>> 2) ||
1762 (loop
> MEM_CGROUP_MAX_RECLAIM_LOOPS
))
1767 if (!mem_cgroup_reclaimable(victim
, false))
1769 total
+= mem_cgroup_shrink_node_zone(victim
, gfp_mask
, false,
1771 *total_scanned
+= nr_scanned
;
1772 if (!res_counter_soft_limit_excess(&root_memcg
->res
))
1775 mem_cgroup_iter_break(root_memcg
, victim
);
1780 * Check OOM-Killer is already running under our hierarchy.
1781 * If someone is running, return false.
1782 * Has to be called with memcg_oom_lock
1784 static bool mem_cgroup_oom_lock(struct mem_cgroup
*memcg
)
1786 struct mem_cgroup
*iter
, *failed
= NULL
;
1788 for_each_mem_cgroup_tree(iter
, memcg
) {
1789 if (iter
->oom_lock
) {
1791 * this subtree of our hierarchy is already locked
1792 * so we cannot give a lock.
1795 mem_cgroup_iter_break(memcg
, iter
);
1798 iter
->oom_lock
= true;
1805 * OK, we failed to lock the whole subtree so we have to clean up
1806 * what we set up to the failing subtree
1808 for_each_mem_cgroup_tree(iter
, memcg
) {
1809 if (iter
== failed
) {
1810 mem_cgroup_iter_break(memcg
, iter
);
1813 iter
->oom_lock
= false;
1819 * Has to be called with memcg_oom_lock
1821 static int mem_cgroup_oom_unlock(struct mem_cgroup
*memcg
)
1823 struct mem_cgroup
*iter
;
1825 for_each_mem_cgroup_tree(iter
, memcg
)
1826 iter
->oom_lock
= false;
1830 static void mem_cgroup_mark_under_oom(struct mem_cgroup
*memcg
)
1832 struct mem_cgroup
*iter
;
1834 for_each_mem_cgroup_tree(iter
, memcg
)
1835 atomic_inc(&iter
->under_oom
);
1838 static void mem_cgroup_unmark_under_oom(struct mem_cgroup
*memcg
)
1840 struct mem_cgroup
*iter
;
1843 * When a new child is created while the hierarchy is under oom,
1844 * mem_cgroup_oom_lock() may not be called. We have to use
1845 * atomic_add_unless() here.
1847 for_each_mem_cgroup_tree(iter
, memcg
)
1848 atomic_add_unless(&iter
->under_oom
, -1, 0);
1851 static DEFINE_SPINLOCK(memcg_oom_lock
);
1852 static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq
);
1854 struct oom_wait_info
{
1855 struct mem_cgroup
*mem
;
1859 static int memcg_oom_wake_function(wait_queue_t
*wait
,
1860 unsigned mode
, int sync
, void *arg
)
1862 struct mem_cgroup
*wake_memcg
= (struct mem_cgroup
*)arg
,
1864 struct oom_wait_info
*oom_wait_info
;
1866 oom_wait_info
= container_of(wait
, struct oom_wait_info
, wait
);
1867 oom_wait_memcg
= oom_wait_info
->mem
;
1870 * Both of oom_wait_info->mem and wake_mem are stable under us.
1871 * Then we can use css_is_ancestor without taking care of RCU.
1873 if (!mem_cgroup_same_or_subtree(oom_wait_memcg
, wake_memcg
)
1874 && !mem_cgroup_same_or_subtree(wake_memcg
, oom_wait_memcg
))
1876 return autoremove_wake_function(wait
, mode
, sync
, arg
);
1879 static void memcg_wakeup_oom(struct mem_cgroup
*memcg
)
1881 /* for filtering, pass "memcg" as argument. */
1882 __wake_up(&memcg_oom_waitq
, TASK_NORMAL
, 0, memcg
);
1885 static void memcg_oom_recover(struct mem_cgroup
*memcg
)
1887 if (memcg
&& atomic_read(&memcg
->under_oom
))
1888 memcg_wakeup_oom(memcg
);
1892 * try to call OOM killer. returns false if we should exit memory-reclaim loop.
1894 bool mem_cgroup_handle_oom(struct mem_cgroup
*memcg
, gfp_t mask
)
1896 struct oom_wait_info owait
;
1897 bool locked
, need_to_kill
;
1900 owait
.wait
.flags
= 0;
1901 owait
.wait
.func
= memcg_oom_wake_function
;
1902 owait
.wait
.private = current
;
1903 INIT_LIST_HEAD(&owait
.wait
.task_list
);
1904 need_to_kill
= true;
1905 mem_cgroup_mark_under_oom(memcg
);
1907 /* At first, try to OOM lock hierarchy under memcg.*/
1908 spin_lock(&memcg_oom_lock
);
1909 locked
= mem_cgroup_oom_lock(memcg
);
1911 * Even if signal_pending(), we can't quit charge() loop without
1912 * accounting. So, UNINTERRUPTIBLE is appropriate. But SIGKILL
1913 * under OOM is always welcomed, use TASK_KILLABLE here.
1915 prepare_to_wait(&memcg_oom_waitq
, &owait
.wait
, TASK_KILLABLE
);
1916 if (!locked
|| memcg
->oom_kill_disable
)
1917 need_to_kill
= false;
1919 mem_cgroup_oom_notify(memcg
);
1920 spin_unlock(&memcg_oom_lock
);
1923 finish_wait(&memcg_oom_waitq
, &owait
.wait
);
1924 mem_cgroup_out_of_memory(memcg
, mask
);
1927 finish_wait(&memcg_oom_waitq
, &owait
.wait
);
1929 spin_lock(&memcg_oom_lock
);
1931 mem_cgroup_oom_unlock(memcg
);
1932 memcg_wakeup_oom(memcg
);
1933 spin_unlock(&memcg_oom_lock
);
1935 mem_cgroup_unmark_under_oom(memcg
);
1937 if (test_thread_flag(TIF_MEMDIE
) || fatal_signal_pending(current
))
1939 /* Give chance to dying process */
1940 schedule_timeout_uninterruptible(1);
1945 * Currently used to update mapped file statistics, but the routine can be
1946 * generalized to update other statistics as well.
1948 * Notes: Race condition
1950 * We usually use page_cgroup_lock() for accessing page_cgroup member but
1951 * it tends to be costly. But considering some conditions, we doesn't need
1952 * to do so _always_.
1954 * Considering "charge", lock_page_cgroup() is not required because all
1955 * file-stat operations happen after a page is attached to radix-tree. There
1956 * are no race with "charge".
1958 * Considering "uncharge", we know that memcg doesn't clear pc->mem_cgroup
1959 * at "uncharge" intentionally. So, we always see valid pc->mem_cgroup even
1960 * if there are race with "uncharge". Statistics itself is properly handled
1963 * Considering "move", this is an only case we see a race. To make the race
1964 * small, we check MEM_CGROUP_ON_MOVE percpu value and detect there are
1965 * possibility of race condition. If there is, we take a lock.
1968 void mem_cgroup_update_page_stat(struct page
*page
,
1969 enum mem_cgroup_page_stat_item idx
, int val
)
1971 struct mem_cgroup
*memcg
;
1972 struct page_cgroup
*pc
= lookup_page_cgroup(page
);
1973 bool need_unlock
= false;
1974 unsigned long uninitialized_var(flags
);
1980 memcg
= pc
->mem_cgroup
;
1981 if (unlikely(!memcg
|| !PageCgroupUsed(pc
)))
1983 /* pc->mem_cgroup is unstable ? */
1984 if (unlikely(mem_cgroup_stealed(memcg
)) || PageTransHuge(page
)) {
1985 /* take a lock against to access pc->mem_cgroup */
1986 move_lock_page_cgroup(pc
, &flags
);
1988 memcg
= pc
->mem_cgroup
;
1989 if (!memcg
|| !PageCgroupUsed(pc
))
1994 case MEMCG_NR_FILE_MAPPED
:
1996 SetPageCgroupFileMapped(pc
);
1997 else if (!page_mapped(page
))
1998 ClearPageCgroupFileMapped(pc
);
1999 idx
= MEM_CGROUP_STAT_FILE_MAPPED
;
2005 this_cpu_add(memcg
->stat
->count
[idx
], val
);
2008 if (unlikely(need_unlock
))
2009 move_unlock_page_cgroup(pc
, &flags
);
2013 EXPORT_SYMBOL(mem_cgroup_update_page_stat
);
2016 * size of first charge trial. "32" comes from vmscan.c's magic value.
2017 * TODO: maybe necessary to use big numbers in big irons.
2019 #define CHARGE_BATCH 32U
2020 struct memcg_stock_pcp
{
2021 struct mem_cgroup
*cached
; /* this never be root cgroup */
2022 unsigned int nr_pages
;
2023 struct work_struct work
;
2024 unsigned long flags
;
2025 #define FLUSHING_CACHED_CHARGE (0)
2027 static DEFINE_PER_CPU(struct memcg_stock_pcp
, memcg_stock
);
2028 static DEFINE_MUTEX(percpu_charge_mutex
);
2031 * Try to consume stocked charge on this cpu. If success, one page is consumed
2032 * from local stock and true is returned. If the stock is 0 or charges from a
2033 * cgroup which is not current target, returns false. This stock will be
2036 static bool consume_stock(struct mem_cgroup
*memcg
)
2038 struct memcg_stock_pcp
*stock
;
2041 stock
= &get_cpu_var(memcg_stock
);
2042 if (memcg
== stock
->cached
&& stock
->nr_pages
)
2044 else /* need to call res_counter_charge */
2046 put_cpu_var(memcg_stock
);
2051 * Returns stocks cached in percpu to res_counter and reset cached information.
2053 static void drain_stock(struct memcg_stock_pcp
*stock
)
2055 struct mem_cgroup
*old
= stock
->cached
;
2057 if (stock
->nr_pages
) {
2058 unsigned long bytes
= stock
->nr_pages
* PAGE_SIZE
;
2060 res_counter_uncharge(&old
->res
, bytes
);
2061 if (do_swap_account
)
2062 res_counter_uncharge(&old
->memsw
, bytes
);
2063 stock
->nr_pages
= 0;
2065 stock
->cached
= NULL
;
2069 * This must be called under preempt disabled or must be called by
2070 * a thread which is pinned to local cpu.
2072 static void drain_local_stock(struct work_struct
*dummy
)
2074 struct memcg_stock_pcp
*stock
= &__get_cpu_var(memcg_stock
);
2076 clear_bit(FLUSHING_CACHED_CHARGE
, &stock
->flags
);
2080 * Cache charges(val) which is from res_counter, to local per_cpu area.
2081 * This will be consumed by consume_stock() function, later.
2083 static void refill_stock(struct mem_cgroup
*memcg
, unsigned int nr_pages
)
2085 struct memcg_stock_pcp
*stock
= &get_cpu_var(memcg_stock
);
2087 if (stock
->cached
!= memcg
) { /* reset if necessary */
2089 stock
->cached
= memcg
;
2091 stock
->nr_pages
+= nr_pages
;
2092 put_cpu_var(memcg_stock
);
2096 * Drains all per-CPU charge caches for given root_memcg resp. subtree
2097 * of the hierarchy under it. sync flag says whether we should block
2098 * until the work is done.
2100 static void drain_all_stock(struct mem_cgroup
*root_memcg
, bool sync
)
2104 /* Notify other cpus that system-wide "drain" is running */
2107 for_each_online_cpu(cpu
) {
2108 struct memcg_stock_pcp
*stock
= &per_cpu(memcg_stock
, cpu
);
2109 struct mem_cgroup
*memcg
;
2111 memcg
= stock
->cached
;
2112 if (!memcg
|| !stock
->nr_pages
)
2114 if (!mem_cgroup_same_or_subtree(root_memcg
, memcg
))
2116 if (!test_and_set_bit(FLUSHING_CACHED_CHARGE
, &stock
->flags
)) {
2118 drain_local_stock(&stock
->work
);
2120 schedule_work_on(cpu
, &stock
->work
);
2128 for_each_online_cpu(cpu
) {
2129 struct memcg_stock_pcp
*stock
= &per_cpu(memcg_stock
, cpu
);
2130 if (test_bit(FLUSHING_CACHED_CHARGE
, &stock
->flags
))
2131 flush_work(&stock
->work
);
2138 * Tries to drain stocked charges in other cpus. This function is asynchronous
2139 * and just put a work per cpu for draining localy on each cpu. Caller can
2140 * expects some charges will be back to res_counter later but cannot wait for
2143 static void drain_all_stock_async(struct mem_cgroup
*root_memcg
)
2146 * If someone calls draining, avoid adding more kworker runs.
2148 if (!mutex_trylock(&percpu_charge_mutex
))
2150 drain_all_stock(root_memcg
, false);
2151 mutex_unlock(&percpu_charge_mutex
);
2154 /* This is a synchronous drain interface. */
2155 static void drain_all_stock_sync(struct mem_cgroup
*root_memcg
)
2157 /* called when force_empty is called */
2158 mutex_lock(&percpu_charge_mutex
);
2159 drain_all_stock(root_memcg
, true);
2160 mutex_unlock(&percpu_charge_mutex
);
2164 * This function drains percpu counter value from DEAD cpu and
2165 * move it to local cpu. Note that this function can be preempted.
2167 static void mem_cgroup_drain_pcp_counter(struct mem_cgroup
*memcg
, int cpu
)
2171 spin_lock(&memcg
->pcp_counter_lock
);
2172 for (i
= 0; i
< MEM_CGROUP_STAT_DATA
; i
++) {
2173 long x
= per_cpu(memcg
->stat
->count
[i
], cpu
);
2175 per_cpu(memcg
->stat
->count
[i
], cpu
) = 0;
2176 memcg
->nocpu_base
.count
[i
] += x
;
2178 for (i
= 0; i
< MEM_CGROUP_EVENTS_NSTATS
; i
++) {
2179 unsigned long x
= per_cpu(memcg
->stat
->events
[i
], cpu
);
2181 per_cpu(memcg
->stat
->events
[i
], cpu
) = 0;
2182 memcg
->nocpu_base
.events
[i
] += x
;
2184 /* need to clear ON_MOVE value, works as a kind of lock. */
2185 per_cpu(memcg
->stat
->count
[MEM_CGROUP_ON_MOVE
], cpu
) = 0;
2186 spin_unlock(&memcg
->pcp_counter_lock
);
2189 static void synchronize_mem_cgroup_on_move(struct mem_cgroup
*memcg
, int cpu
)
2191 int idx
= MEM_CGROUP_ON_MOVE
;
2193 spin_lock(&memcg
->pcp_counter_lock
);
2194 per_cpu(memcg
->stat
->count
[idx
], cpu
) = memcg
->nocpu_base
.count
[idx
];
2195 spin_unlock(&memcg
->pcp_counter_lock
);
2198 static int __cpuinit
memcg_cpu_hotplug_callback(struct notifier_block
*nb
,
2199 unsigned long action
,
2202 int cpu
= (unsigned long)hcpu
;
2203 struct memcg_stock_pcp
*stock
;
2204 struct mem_cgroup
*iter
;
2206 if ((action
== CPU_ONLINE
)) {
2207 for_each_mem_cgroup(iter
)
2208 synchronize_mem_cgroup_on_move(iter
, cpu
);
2212 if ((action
!= CPU_DEAD
) || action
!= CPU_DEAD_FROZEN
)
2215 for_each_mem_cgroup(iter
)
2216 mem_cgroup_drain_pcp_counter(iter
, cpu
);
2218 stock
= &per_cpu(memcg_stock
, cpu
);
2224 /* See __mem_cgroup_try_charge() for details */
2226 CHARGE_OK
, /* success */
2227 CHARGE_RETRY
, /* need to retry but retry is not bad */
2228 CHARGE_NOMEM
, /* we can't do more. return -ENOMEM */
2229 CHARGE_WOULDBLOCK
, /* GFP_WAIT wasn't set and no enough res. */
2230 CHARGE_OOM_DIE
, /* the current is killed because of OOM */
2233 static int mem_cgroup_do_charge(struct mem_cgroup
*memcg
, gfp_t gfp_mask
,
2234 unsigned int nr_pages
, bool oom_check
)
2236 unsigned long csize
= nr_pages
* PAGE_SIZE
;
2237 struct mem_cgroup
*mem_over_limit
;
2238 struct res_counter
*fail_res
;
2239 unsigned long flags
= 0;
2242 ret
= res_counter_charge(&memcg
->res
, csize
, &fail_res
);
2245 if (!do_swap_account
)
2247 ret
= res_counter_charge(&memcg
->memsw
, csize
, &fail_res
);
2251 res_counter_uncharge(&memcg
->res
, csize
);
2252 mem_over_limit
= mem_cgroup_from_res_counter(fail_res
, memsw
);
2253 flags
|= MEM_CGROUP_RECLAIM_NOSWAP
;
2255 mem_over_limit
= mem_cgroup_from_res_counter(fail_res
, res
);
2257 * nr_pages can be either a huge page (HPAGE_PMD_NR), a batch
2258 * of regular pages (CHARGE_BATCH), or a single regular page (1).
2260 * Never reclaim on behalf of optional batching, retry with a
2261 * single page instead.
2263 if (nr_pages
== CHARGE_BATCH
)
2264 return CHARGE_RETRY
;
2266 if (!(gfp_mask
& __GFP_WAIT
))
2267 return CHARGE_WOULDBLOCK
;
2269 ret
= mem_cgroup_reclaim(mem_over_limit
, gfp_mask
, flags
);
2270 if (mem_cgroup_margin(mem_over_limit
) >= nr_pages
)
2271 return CHARGE_RETRY
;
2273 * Even though the limit is exceeded at this point, reclaim
2274 * may have been able to free some pages. Retry the charge
2275 * before killing the task.
2277 * Only for regular pages, though: huge pages are rather
2278 * unlikely to succeed so close to the limit, and we fall back
2279 * to regular pages anyway in case of failure.
2281 if (nr_pages
== 1 && ret
)
2282 return CHARGE_RETRY
;
2285 * At task move, charge accounts can be doubly counted. So, it's
2286 * better to wait until the end of task_move if something is going on.
2288 if (mem_cgroup_wait_acct_move(mem_over_limit
))
2289 return CHARGE_RETRY
;
2291 /* If we don't need to call oom-killer at el, return immediately */
2293 return CHARGE_NOMEM
;
2295 if (!mem_cgroup_handle_oom(mem_over_limit
, gfp_mask
))
2296 return CHARGE_OOM_DIE
;
2298 return CHARGE_RETRY
;
2302 * Unlike exported interface, "oom" parameter is added. if oom==true,
2303 * oom-killer can be invoked.
2305 static int __mem_cgroup_try_charge(struct mm_struct
*mm
,
2307 unsigned int nr_pages
,
2308 struct mem_cgroup
**ptr
,
2311 unsigned int batch
= max(CHARGE_BATCH
, nr_pages
);
2312 int nr_oom_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
2313 struct mem_cgroup
*memcg
= NULL
;
2317 * Unlike gloval-vm's OOM-kill, we're not in memory shortage
2318 * in system level. So, allow to go ahead dying process in addition to
2321 if (unlikely(test_thread_flag(TIF_MEMDIE
)
2322 || fatal_signal_pending(current
)))
2326 * We always charge the cgroup the mm_struct belongs to.
2327 * The mm_struct's mem_cgroup changes on task migration if the
2328 * thread group leader migrates. It's possible that mm is not
2329 * set, if so charge the init_mm (happens for pagecache usage).
2334 if (*ptr
) { /* css should be a valid one */
2336 VM_BUG_ON(css_is_removed(&memcg
->css
));
2337 if (mem_cgroup_is_root(memcg
))
2339 if (nr_pages
== 1 && consume_stock(memcg
))
2341 css_get(&memcg
->css
);
2343 struct task_struct
*p
;
2346 p
= rcu_dereference(mm
->owner
);
2348 * Because we don't have task_lock(), "p" can exit.
2349 * In that case, "memcg" can point to root or p can be NULL with
2350 * race with swapoff. Then, we have small risk of mis-accouning.
2351 * But such kind of mis-account by race always happens because
2352 * we don't have cgroup_mutex(). It's overkill and we allo that
2354 * (*) swapoff at el will charge against mm-struct not against
2355 * task-struct. So, mm->owner can be NULL.
2357 memcg
= mem_cgroup_from_task(p
);
2358 if (!memcg
|| mem_cgroup_is_root(memcg
)) {
2362 if (nr_pages
== 1 && consume_stock(memcg
)) {
2364 * It seems dagerous to access memcg without css_get().
2365 * But considering how consume_stok works, it's not
2366 * necessary. If consume_stock success, some charges
2367 * from this memcg are cached on this cpu. So, we
2368 * don't need to call css_get()/css_tryget() before
2369 * calling consume_stock().
2374 /* after here, we may be blocked. we need to get refcnt */
2375 if (!css_tryget(&memcg
->css
)) {
2385 /* If killed, bypass charge */
2386 if (fatal_signal_pending(current
)) {
2387 css_put(&memcg
->css
);
2392 if (oom
&& !nr_oom_retries
) {
2394 nr_oom_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
2397 ret
= mem_cgroup_do_charge(memcg
, gfp_mask
, batch
, oom_check
);
2401 case CHARGE_RETRY
: /* not in OOM situation but retry */
2403 css_put(&memcg
->css
);
2406 case CHARGE_WOULDBLOCK
: /* !__GFP_WAIT */
2407 css_put(&memcg
->css
);
2409 case CHARGE_NOMEM
: /* OOM routine works */
2411 css_put(&memcg
->css
);
2414 /* If oom, we never return -ENOMEM */
2417 case CHARGE_OOM_DIE
: /* Killed by OOM Killer */
2418 css_put(&memcg
->css
);
2421 } while (ret
!= CHARGE_OK
);
2423 if (batch
> nr_pages
)
2424 refill_stock(memcg
, batch
- nr_pages
);
2425 css_put(&memcg
->css
);
2438 * Somemtimes we have to undo a charge we got by try_charge().
2439 * This function is for that and do uncharge, put css's refcnt.
2440 * gotten by try_charge().
2442 static void __mem_cgroup_cancel_charge(struct mem_cgroup
*memcg
,
2443 unsigned int nr_pages
)
2445 if (!mem_cgroup_is_root(memcg
)) {
2446 unsigned long bytes
= nr_pages
* PAGE_SIZE
;
2448 res_counter_uncharge(&memcg
->res
, bytes
);
2449 if (do_swap_account
)
2450 res_counter_uncharge(&memcg
->memsw
, bytes
);
2455 * A helper function to get mem_cgroup from ID. must be called under
2456 * rcu_read_lock(). The caller must check css_is_removed() or some if
2457 * it's concern. (dropping refcnt from swap can be called against removed
2460 static struct mem_cgroup
*mem_cgroup_lookup(unsigned short id
)
2462 struct cgroup_subsys_state
*css
;
2464 /* ID 0 is unused ID */
2467 css
= css_lookup(&mem_cgroup_subsys
, id
);
2470 return container_of(css
, struct mem_cgroup
, css
);
2473 struct mem_cgroup
*try_get_mem_cgroup_from_page(struct page
*page
)
2475 struct mem_cgroup
*memcg
= NULL
;
2476 struct page_cgroup
*pc
;
2480 VM_BUG_ON(!PageLocked(page
));
2482 pc
= lookup_page_cgroup(page
);
2483 lock_page_cgroup(pc
);
2484 if (PageCgroupUsed(pc
)) {
2485 memcg
= pc
->mem_cgroup
;
2486 if (memcg
&& !css_tryget(&memcg
->css
))
2488 } else if (PageSwapCache(page
)) {
2489 ent
.val
= page_private(page
);
2490 id
= lookup_swap_cgroup(ent
);
2492 memcg
= mem_cgroup_lookup(id
);
2493 if (memcg
&& !css_tryget(&memcg
->css
))
2497 unlock_page_cgroup(pc
);
2501 static void __mem_cgroup_commit_charge(struct mem_cgroup
*memcg
,
2503 unsigned int nr_pages
,
2504 struct page_cgroup
*pc
,
2505 enum charge_type ctype
)
2507 lock_page_cgroup(pc
);
2508 if (unlikely(PageCgroupUsed(pc
))) {
2509 unlock_page_cgroup(pc
);
2510 __mem_cgroup_cancel_charge(memcg
, nr_pages
);
2514 * we don't need page_cgroup_lock about tail pages, becase they are not
2515 * accessed by any other context at this point.
2517 pc
->mem_cgroup
= memcg
;
2519 * We access a page_cgroup asynchronously without lock_page_cgroup().
2520 * Especially when a page_cgroup is taken from a page, pc->mem_cgroup
2521 * is accessed after testing USED bit. To make pc->mem_cgroup visible
2522 * before USED bit, we need memory barrier here.
2523 * See mem_cgroup_add_lru_list(), etc.
2527 case MEM_CGROUP_CHARGE_TYPE_CACHE
:
2528 case MEM_CGROUP_CHARGE_TYPE_SHMEM
:
2529 SetPageCgroupCache(pc
);
2530 SetPageCgroupUsed(pc
);
2532 case MEM_CGROUP_CHARGE_TYPE_MAPPED
:
2533 ClearPageCgroupCache(pc
);
2534 SetPageCgroupUsed(pc
);
2540 mem_cgroup_charge_statistics(memcg
, PageCgroupCache(pc
), nr_pages
);
2541 unlock_page_cgroup(pc
);
2543 * "charge_statistics" updated event counter. Then, check it.
2544 * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree.
2545 * if they exceeds softlimit.
2547 memcg_check_events(memcg
, page
);
2550 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
2552 #define PCGF_NOCOPY_AT_SPLIT ((1 << PCG_LOCK) | (1 << PCG_MOVE_LOCK) |\
2553 (1 << PCG_ACCT_LRU) | (1 << PCG_MIGRATION))
2555 * Because tail pages are not marked as "used", set it. We're under
2556 * zone->lru_lock, 'splitting on pmd' and compund_lock.
2558 void mem_cgroup_split_huge_fixup(struct page
*head
, struct page
*tail
)
2560 struct page_cgroup
*head_pc
= lookup_page_cgroup(head
);
2561 struct page_cgroup
*tail_pc
= lookup_page_cgroup(tail
);
2562 unsigned long flags
;
2564 if (mem_cgroup_disabled())
2567 * We have no races with charge/uncharge but will have races with
2568 * page state accounting.
2570 move_lock_page_cgroup(head_pc
, &flags
);
2572 tail_pc
->mem_cgroup
= head_pc
->mem_cgroup
;
2573 smp_wmb(); /* see __commit_charge() */
2574 if (PageCgroupAcctLRU(head_pc
)) {
2576 struct mem_cgroup_per_zone
*mz
;
2579 * LRU flags cannot be copied because we need to add tail
2580 *.page to LRU by generic call and our hook will be called.
2581 * We hold lru_lock, then, reduce counter directly.
2583 lru
= page_lru(head
);
2584 mz
= page_cgroup_zoneinfo(head_pc
->mem_cgroup
, head
);
2585 MEM_CGROUP_ZSTAT(mz
, lru
) -= 1;
2587 tail_pc
->flags
= head_pc
->flags
& ~PCGF_NOCOPY_AT_SPLIT
;
2588 move_unlock_page_cgroup(head_pc
, &flags
);
2593 * mem_cgroup_move_account - move account of the page
2595 * @nr_pages: number of regular pages (>1 for huge pages)
2596 * @pc: page_cgroup of the page.
2597 * @from: mem_cgroup which the page is moved from.
2598 * @to: mem_cgroup which the page is moved to. @from != @to.
2599 * @uncharge: whether we should call uncharge and css_put against @from.
2601 * The caller must confirm following.
2602 * - page is not on LRU (isolate_page() is useful.)
2603 * - compound_lock is held when nr_pages > 1
2605 * This function doesn't do "charge" nor css_get to new cgroup. It should be
2606 * done by a caller(__mem_cgroup_try_charge would be useful). If @uncharge is
2607 * true, this function does "uncharge" from old cgroup, but it doesn't if
2608 * @uncharge is false, so a caller should do "uncharge".
2610 static int mem_cgroup_move_account(struct page
*page
,
2611 unsigned int nr_pages
,
2612 struct page_cgroup
*pc
,
2613 struct mem_cgroup
*from
,
2614 struct mem_cgroup
*to
,
2617 unsigned long flags
;
2620 VM_BUG_ON(from
== to
);
2621 VM_BUG_ON(PageLRU(page
));
2623 * The page is isolated from LRU. So, collapse function
2624 * will not handle this page. But page splitting can happen.
2625 * Do this check under compound_page_lock(). The caller should
2629 if (nr_pages
> 1 && !PageTransHuge(page
))
2632 lock_page_cgroup(pc
);
2635 if (!PageCgroupUsed(pc
) || pc
->mem_cgroup
!= from
)
2638 move_lock_page_cgroup(pc
, &flags
);
2640 if (PageCgroupFileMapped(pc
)) {
2641 /* Update mapped_file data for mem_cgroup */
2643 __this_cpu_dec(from
->stat
->count
[MEM_CGROUP_STAT_FILE_MAPPED
]);
2644 __this_cpu_inc(to
->stat
->count
[MEM_CGROUP_STAT_FILE_MAPPED
]);
2647 mem_cgroup_charge_statistics(from
, PageCgroupCache(pc
), -nr_pages
);
2649 /* This is not "cancel", but cancel_charge does all we need. */
2650 __mem_cgroup_cancel_charge(from
, nr_pages
);
2652 /* caller should have done css_get */
2653 pc
->mem_cgroup
= to
;
2654 mem_cgroup_charge_statistics(to
, PageCgroupCache(pc
), nr_pages
);
2656 * We charges against "to" which may not have any tasks. Then, "to"
2657 * can be under rmdir(). But in current implementation, caller of
2658 * this function is just force_empty() and move charge, so it's
2659 * guaranteed that "to" is never removed. So, we don't check rmdir
2662 move_unlock_page_cgroup(pc
, &flags
);
2665 unlock_page_cgroup(pc
);
2669 memcg_check_events(to
, page
);
2670 memcg_check_events(from
, page
);
2676 * move charges to its parent.
2679 static int mem_cgroup_move_parent(struct page
*page
,
2680 struct page_cgroup
*pc
,
2681 struct mem_cgroup
*child
,
2684 struct cgroup
*cg
= child
->css
.cgroup
;
2685 struct cgroup
*pcg
= cg
->parent
;
2686 struct mem_cgroup
*parent
;
2687 unsigned int nr_pages
;
2688 unsigned long uninitialized_var(flags
);
2696 if (!get_page_unless_zero(page
))
2698 if (isolate_lru_page(page
))
2701 nr_pages
= hpage_nr_pages(page
);
2703 parent
= mem_cgroup_from_cont(pcg
);
2704 ret
= __mem_cgroup_try_charge(NULL
, gfp_mask
, nr_pages
, &parent
, false);
2709 flags
= compound_lock_irqsave(page
);
2711 ret
= mem_cgroup_move_account(page
, nr_pages
, pc
, child
, parent
, true);
2713 __mem_cgroup_cancel_charge(parent
, nr_pages
);
2716 compound_unlock_irqrestore(page
, flags
);
2718 putback_lru_page(page
);
2726 * Charge the memory controller for page usage.
2728 * 0 if the charge was successful
2729 * < 0 if the cgroup is over its limit
2731 static int mem_cgroup_charge_common(struct page
*page
, struct mm_struct
*mm
,
2732 gfp_t gfp_mask
, enum charge_type ctype
)
2734 struct mem_cgroup
*memcg
= NULL
;
2735 unsigned int nr_pages
= 1;
2736 struct page_cgroup
*pc
;
2740 if (PageTransHuge(page
)) {
2741 nr_pages
<<= compound_order(page
);
2742 VM_BUG_ON(!PageTransHuge(page
));
2744 * Never OOM-kill a process for a huge page. The
2745 * fault handler will fall back to regular pages.
2750 pc
= lookup_page_cgroup(page
);
2751 BUG_ON(!pc
); /* XXX: remove this and move pc lookup into commit */
2753 ret
= __mem_cgroup_try_charge(mm
, gfp_mask
, nr_pages
, &memcg
, oom
);
2757 __mem_cgroup_commit_charge(memcg
, page
, nr_pages
, pc
, ctype
);
2761 int mem_cgroup_newpage_charge(struct page
*page
,
2762 struct mm_struct
*mm
, gfp_t gfp_mask
)
2764 if (mem_cgroup_disabled())
2767 * If already mapped, we don't have to account.
2768 * If page cache, page->mapping has address_space.
2769 * But page->mapping may have out-of-use anon_vma pointer,
2770 * detecit it by PageAnon() check. newly-mapped-anon's page->mapping
2773 if (page_mapped(page
) || (page
->mapping
&& !PageAnon(page
)))
2777 return mem_cgroup_charge_common(page
, mm
, gfp_mask
,
2778 MEM_CGROUP_CHARGE_TYPE_MAPPED
);
2782 __mem_cgroup_commit_charge_swapin(struct page
*page
, struct mem_cgroup
*ptr
,
2783 enum charge_type ctype
);
2786 __mem_cgroup_commit_charge_lrucare(struct page
*page
, struct mem_cgroup
*memcg
,
2787 enum charge_type ctype
)
2789 struct page_cgroup
*pc
= lookup_page_cgroup(page
);
2791 * In some case, SwapCache, FUSE(splice_buf->radixtree), the page
2792 * is already on LRU. It means the page may on some other page_cgroup's
2793 * LRU. Take care of it.
2795 mem_cgroup_lru_del_before_commit(page
);
2796 __mem_cgroup_commit_charge(memcg
, page
, 1, pc
, ctype
);
2797 mem_cgroup_lru_add_after_commit(page
);
2801 int mem_cgroup_cache_charge(struct page
*page
, struct mm_struct
*mm
,
2804 struct mem_cgroup
*memcg
= NULL
;
2807 if (mem_cgroup_disabled())
2809 if (PageCompound(page
))
2815 if (page_is_file_cache(page
)) {
2816 ret
= __mem_cgroup_try_charge(mm
, gfp_mask
, 1, &memcg
, true);
2821 * FUSE reuses pages without going through the final
2822 * put that would remove them from the LRU list, make
2823 * sure that they get relinked properly.
2825 __mem_cgroup_commit_charge_lrucare(page
, memcg
,
2826 MEM_CGROUP_CHARGE_TYPE_CACHE
);
2830 if (PageSwapCache(page
)) {
2831 ret
= mem_cgroup_try_charge_swapin(mm
, page
, gfp_mask
, &memcg
);
2833 __mem_cgroup_commit_charge_swapin(page
, memcg
,
2834 MEM_CGROUP_CHARGE_TYPE_SHMEM
);
2836 ret
= mem_cgroup_charge_common(page
, mm
, gfp_mask
,
2837 MEM_CGROUP_CHARGE_TYPE_SHMEM
);
2843 * While swap-in, try_charge -> commit or cancel, the page is locked.
2844 * And when try_charge() successfully returns, one refcnt to memcg without
2845 * struct page_cgroup is acquired. This refcnt will be consumed by
2846 * "commit()" or removed by "cancel()"
2848 int mem_cgroup_try_charge_swapin(struct mm_struct
*mm
,
2850 gfp_t mask
, struct mem_cgroup
**ptr
)
2852 struct mem_cgroup
*memcg
;
2857 if (mem_cgroup_disabled())
2860 if (!do_swap_account
)
2863 * A racing thread's fault, or swapoff, may have already updated
2864 * the pte, and even removed page from swap cache: in those cases
2865 * do_swap_page()'s pte_same() test will fail; but there's also a
2866 * KSM case which does need to charge the page.
2868 if (!PageSwapCache(page
))
2870 memcg
= try_get_mem_cgroup_from_page(page
);
2874 ret
= __mem_cgroup_try_charge(NULL
, mask
, 1, ptr
, true);
2875 css_put(&memcg
->css
);
2880 return __mem_cgroup_try_charge(mm
, mask
, 1, ptr
, true);
2884 __mem_cgroup_commit_charge_swapin(struct page
*page
, struct mem_cgroup
*ptr
,
2885 enum charge_type ctype
)
2887 if (mem_cgroup_disabled())
2891 cgroup_exclude_rmdir(&ptr
->css
);
2893 __mem_cgroup_commit_charge_lrucare(page
, ptr
, ctype
);
2895 * Now swap is on-memory. This means this page may be
2896 * counted both as mem and swap....double count.
2897 * Fix it by uncharging from memsw. Basically, this SwapCache is stable
2898 * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page()
2899 * may call delete_from_swap_cache() before reach here.
2901 if (do_swap_account
&& PageSwapCache(page
)) {
2902 swp_entry_t ent
= {.val
= page_private(page
)};
2904 struct mem_cgroup
*memcg
;
2906 id
= swap_cgroup_record(ent
, 0);
2908 memcg
= mem_cgroup_lookup(id
);
2911 * This recorded memcg can be obsolete one. So, avoid
2912 * calling css_tryget
2914 if (!mem_cgroup_is_root(memcg
))
2915 res_counter_uncharge(&memcg
->memsw
, PAGE_SIZE
);
2916 mem_cgroup_swap_statistics(memcg
, false);
2917 mem_cgroup_put(memcg
);
2922 * At swapin, we may charge account against cgroup which has no tasks.
2923 * So, rmdir()->pre_destroy() can be called while we do this charge.
2924 * In that case, we need to call pre_destroy() again. check it here.
2926 cgroup_release_and_wakeup_rmdir(&ptr
->css
);
2929 void mem_cgroup_commit_charge_swapin(struct page
*page
, struct mem_cgroup
*ptr
)
2931 __mem_cgroup_commit_charge_swapin(page
, ptr
,
2932 MEM_CGROUP_CHARGE_TYPE_MAPPED
);
2935 void mem_cgroup_cancel_charge_swapin(struct mem_cgroup
*memcg
)
2937 if (mem_cgroup_disabled())
2941 __mem_cgroup_cancel_charge(memcg
, 1);
2944 static void mem_cgroup_do_uncharge(struct mem_cgroup
*memcg
,
2945 unsigned int nr_pages
,
2946 const enum charge_type ctype
)
2948 struct memcg_batch_info
*batch
= NULL
;
2949 bool uncharge_memsw
= true;
2951 /* If swapout, usage of swap doesn't decrease */
2952 if (!do_swap_account
|| ctype
== MEM_CGROUP_CHARGE_TYPE_SWAPOUT
)
2953 uncharge_memsw
= false;
2955 batch
= ¤t
->memcg_batch
;
2957 * In usual, we do css_get() when we remember memcg pointer.
2958 * But in this case, we keep res->usage until end of a series of
2959 * uncharges. Then, it's ok to ignore memcg's refcnt.
2962 batch
->memcg
= memcg
;
2964 * do_batch > 0 when unmapping pages or inode invalidate/truncate.
2965 * In those cases, all pages freed continuously can be expected to be in
2966 * the same cgroup and we have chance to coalesce uncharges.
2967 * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE)
2968 * because we want to do uncharge as soon as possible.
2971 if (!batch
->do_batch
|| test_thread_flag(TIF_MEMDIE
))
2972 goto direct_uncharge
;
2975 goto direct_uncharge
;
2978 * In typical case, batch->memcg == mem. This means we can
2979 * merge a series of uncharges to an uncharge of res_counter.
2980 * If not, we uncharge res_counter ony by one.
2982 if (batch
->memcg
!= memcg
)
2983 goto direct_uncharge
;
2984 /* remember freed charge and uncharge it later */
2987 batch
->memsw_nr_pages
++;
2990 res_counter_uncharge(&memcg
->res
, nr_pages
* PAGE_SIZE
);
2992 res_counter_uncharge(&memcg
->memsw
, nr_pages
* PAGE_SIZE
);
2993 if (unlikely(batch
->memcg
!= memcg
))
2994 memcg_oom_recover(memcg
);
2999 * uncharge if !page_mapped(page)
3001 static struct mem_cgroup
*
3002 __mem_cgroup_uncharge_common(struct page
*page
, enum charge_type ctype
)
3004 struct mem_cgroup
*memcg
= NULL
;
3005 unsigned int nr_pages
= 1;
3006 struct page_cgroup
*pc
;
3008 if (mem_cgroup_disabled())
3011 if (PageSwapCache(page
))
3014 if (PageTransHuge(page
)) {
3015 nr_pages
<<= compound_order(page
);
3016 VM_BUG_ON(!PageTransHuge(page
));
3019 * Check if our page_cgroup is valid
3021 pc
= lookup_page_cgroup(page
);
3022 if (unlikely(!pc
|| !PageCgroupUsed(pc
)))
3025 lock_page_cgroup(pc
);
3027 memcg
= pc
->mem_cgroup
;
3029 if (!PageCgroupUsed(pc
))
3033 case MEM_CGROUP_CHARGE_TYPE_MAPPED
:
3034 case MEM_CGROUP_CHARGE_TYPE_DROP
:
3035 /* See mem_cgroup_prepare_migration() */
3036 if (page_mapped(page
) || PageCgroupMigration(pc
))
3039 case MEM_CGROUP_CHARGE_TYPE_SWAPOUT
:
3040 if (!PageAnon(page
)) { /* Shared memory */
3041 if (page
->mapping
&& !page_is_file_cache(page
))
3043 } else if (page_mapped(page
)) /* Anon */
3050 mem_cgroup_charge_statistics(memcg
, PageCgroupCache(pc
), -nr_pages
);
3052 ClearPageCgroupUsed(pc
);
3054 * pc->mem_cgroup is not cleared here. It will be accessed when it's
3055 * freed from LRU. This is safe because uncharged page is expected not
3056 * to be reused (freed soon). Exception is SwapCache, it's handled by
3057 * special functions.
3060 unlock_page_cgroup(pc
);
3062 * even after unlock, we have memcg->res.usage here and this memcg
3063 * will never be freed.
3065 memcg_check_events(memcg
, page
);
3066 if (do_swap_account
&& ctype
== MEM_CGROUP_CHARGE_TYPE_SWAPOUT
) {
3067 mem_cgroup_swap_statistics(memcg
, true);
3068 mem_cgroup_get(memcg
);
3070 if (!mem_cgroup_is_root(memcg
))
3071 mem_cgroup_do_uncharge(memcg
, nr_pages
, ctype
);
3076 unlock_page_cgroup(pc
);
3080 void mem_cgroup_uncharge_page(struct page
*page
)
3083 if (page_mapped(page
))
3085 if (page
->mapping
&& !PageAnon(page
))
3087 __mem_cgroup_uncharge_common(page
, MEM_CGROUP_CHARGE_TYPE_MAPPED
);
3090 void mem_cgroup_uncharge_cache_page(struct page
*page
)
3092 VM_BUG_ON(page_mapped(page
));
3093 VM_BUG_ON(page
->mapping
);
3094 __mem_cgroup_uncharge_common(page
, MEM_CGROUP_CHARGE_TYPE_CACHE
);
3098 * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate.
3099 * In that cases, pages are freed continuously and we can expect pages
3100 * are in the same memcg. All these calls itself limits the number of
3101 * pages freed at once, then uncharge_start/end() is called properly.
3102 * This may be called prural(2) times in a context,
3105 void mem_cgroup_uncharge_start(void)
3107 current
->memcg_batch
.do_batch
++;
3108 /* We can do nest. */
3109 if (current
->memcg_batch
.do_batch
== 1) {
3110 current
->memcg_batch
.memcg
= NULL
;
3111 current
->memcg_batch
.nr_pages
= 0;
3112 current
->memcg_batch
.memsw_nr_pages
= 0;
3116 void mem_cgroup_uncharge_end(void)
3118 struct memcg_batch_info
*batch
= ¤t
->memcg_batch
;
3120 if (!batch
->do_batch
)
3124 if (batch
->do_batch
) /* If stacked, do nothing. */
3130 * This "batch->memcg" is valid without any css_get/put etc...
3131 * bacause we hide charges behind us.
3133 if (batch
->nr_pages
)
3134 res_counter_uncharge(&batch
->memcg
->res
,
3135 batch
->nr_pages
* PAGE_SIZE
);
3136 if (batch
->memsw_nr_pages
)
3137 res_counter_uncharge(&batch
->memcg
->memsw
,
3138 batch
->memsw_nr_pages
* PAGE_SIZE
);
3139 memcg_oom_recover(batch
->memcg
);
3140 /* forget this pointer (for sanity check) */
3141 batch
->memcg
= NULL
;
3146 * called after __delete_from_swap_cache() and drop "page" account.
3147 * memcg information is recorded to swap_cgroup of "ent"
3150 mem_cgroup_uncharge_swapcache(struct page
*page
, swp_entry_t ent
, bool swapout
)
3152 struct mem_cgroup
*memcg
;
3153 int ctype
= MEM_CGROUP_CHARGE_TYPE_SWAPOUT
;
3155 if (!swapout
) /* this was a swap cache but the swap is unused ! */
3156 ctype
= MEM_CGROUP_CHARGE_TYPE_DROP
;
3158 memcg
= __mem_cgroup_uncharge_common(page
, ctype
);
3161 * record memcg information, if swapout && memcg != NULL,
3162 * mem_cgroup_get() was called in uncharge().
3164 if (do_swap_account
&& swapout
&& memcg
)
3165 swap_cgroup_record(ent
, css_id(&memcg
->css
));
3169 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
3171 * called from swap_entry_free(). remove record in swap_cgroup and
3172 * uncharge "memsw" account.
3174 void mem_cgroup_uncharge_swap(swp_entry_t ent
)
3176 struct mem_cgroup
*memcg
;
3179 if (!do_swap_account
)
3182 id
= swap_cgroup_record(ent
, 0);
3184 memcg
= mem_cgroup_lookup(id
);
3187 * We uncharge this because swap is freed.
3188 * This memcg can be obsolete one. We avoid calling css_tryget
3190 if (!mem_cgroup_is_root(memcg
))
3191 res_counter_uncharge(&memcg
->memsw
, PAGE_SIZE
);
3192 mem_cgroup_swap_statistics(memcg
, false);
3193 mem_cgroup_put(memcg
);
3199 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
3200 * @entry: swap entry to be moved
3201 * @from: mem_cgroup which the entry is moved from
3202 * @to: mem_cgroup which the entry is moved to
3203 * @need_fixup: whether we should fixup res_counters and refcounts.
3205 * It succeeds only when the swap_cgroup's record for this entry is the same
3206 * as the mem_cgroup's id of @from.
3208 * Returns 0 on success, -EINVAL on failure.
3210 * The caller must have charged to @to, IOW, called res_counter_charge() about
3211 * both res and memsw, and called css_get().
3213 static int mem_cgroup_move_swap_account(swp_entry_t entry
,
3214 struct mem_cgroup
*from
, struct mem_cgroup
*to
, bool need_fixup
)
3216 unsigned short old_id
, new_id
;
3218 old_id
= css_id(&from
->css
);
3219 new_id
= css_id(&to
->css
);
3221 if (swap_cgroup_cmpxchg(entry
, old_id
, new_id
) == old_id
) {
3222 mem_cgroup_swap_statistics(from
, false);
3223 mem_cgroup_swap_statistics(to
, true);
3225 * This function is only called from task migration context now.
3226 * It postpones res_counter and refcount handling till the end
3227 * of task migration(mem_cgroup_clear_mc()) for performance
3228 * improvement. But we cannot postpone mem_cgroup_get(to)
3229 * because if the process that has been moved to @to does
3230 * swap-in, the refcount of @to might be decreased to 0.
3234 if (!mem_cgroup_is_root(from
))
3235 res_counter_uncharge(&from
->memsw
, PAGE_SIZE
);
3236 mem_cgroup_put(from
);
3238 * we charged both to->res and to->memsw, so we should
3241 if (!mem_cgroup_is_root(to
))
3242 res_counter_uncharge(&to
->res
, PAGE_SIZE
);
3249 static inline int mem_cgroup_move_swap_account(swp_entry_t entry
,
3250 struct mem_cgroup
*from
, struct mem_cgroup
*to
, bool need_fixup
)
3257 * Before starting migration, account PAGE_SIZE to mem_cgroup that the old
3260 int mem_cgroup_prepare_migration(struct page
*page
,
3261 struct page
*newpage
, struct mem_cgroup
**ptr
, gfp_t gfp_mask
)
3263 struct mem_cgroup
*memcg
= NULL
;
3264 struct page_cgroup
*pc
;
3265 enum charge_type ctype
;
3270 VM_BUG_ON(PageTransHuge(page
));
3271 if (mem_cgroup_disabled())
3274 pc
= lookup_page_cgroup(page
);
3275 lock_page_cgroup(pc
);
3276 if (PageCgroupUsed(pc
)) {
3277 memcg
= pc
->mem_cgroup
;
3278 css_get(&memcg
->css
);
3280 * At migrating an anonymous page, its mapcount goes down
3281 * to 0 and uncharge() will be called. But, even if it's fully
3282 * unmapped, migration may fail and this page has to be
3283 * charged again. We set MIGRATION flag here and delay uncharge
3284 * until end_migration() is called
3286 * Corner Case Thinking
3288 * When the old page was mapped as Anon and it's unmap-and-freed
3289 * while migration was ongoing.
3290 * If unmap finds the old page, uncharge() of it will be delayed
3291 * until end_migration(). If unmap finds a new page, it's
3292 * uncharged when it make mapcount to be 1->0. If unmap code
3293 * finds swap_migration_entry, the new page will not be mapped
3294 * and end_migration() will find it(mapcount==0).
3297 * When the old page was mapped but migraion fails, the kernel
3298 * remaps it. A charge for it is kept by MIGRATION flag even
3299 * if mapcount goes down to 0. We can do remap successfully
3300 * without charging it again.
3303 * The "old" page is under lock_page() until the end of
3304 * migration, so, the old page itself will not be swapped-out.
3305 * If the new page is swapped out before end_migraton, our
3306 * hook to usual swap-out path will catch the event.
3309 SetPageCgroupMigration(pc
);
3311 unlock_page_cgroup(pc
);
3313 * If the page is not charged at this point,
3320 ret
= __mem_cgroup_try_charge(NULL
, gfp_mask
, 1, ptr
, false);
3321 css_put(&memcg
->css
);/* drop extra refcnt */
3322 if (ret
|| *ptr
== NULL
) {
3323 if (PageAnon(page
)) {
3324 lock_page_cgroup(pc
);
3325 ClearPageCgroupMigration(pc
);
3326 unlock_page_cgroup(pc
);
3328 * The old page may be fully unmapped while we kept it.
3330 mem_cgroup_uncharge_page(page
);
3335 * We charge new page before it's used/mapped. So, even if unlock_page()
3336 * is called before end_migration, we can catch all events on this new
3337 * page. In the case new page is migrated but not remapped, new page's
3338 * mapcount will be finally 0 and we call uncharge in end_migration().
3340 pc
= lookup_page_cgroup(newpage
);
3342 ctype
= MEM_CGROUP_CHARGE_TYPE_MAPPED
;
3343 else if (page_is_file_cache(page
))
3344 ctype
= MEM_CGROUP_CHARGE_TYPE_CACHE
;
3346 ctype
= MEM_CGROUP_CHARGE_TYPE_SHMEM
;
3347 __mem_cgroup_commit_charge(memcg
, page
, 1, pc
, ctype
);
3351 /* remove redundant charge if migration failed*/
3352 void mem_cgroup_end_migration(struct mem_cgroup
*memcg
,
3353 struct page
*oldpage
, struct page
*newpage
, bool migration_ok
)
3355 struct page
*used
, *unused
;
3356 struct page_cgroup
*pc
;
3360 /* blocks rmdir() */
3361 cgroup_exclude_rmdir(&memcg
->css
);
3362 if (!migration_ok
) {
3370 * We disallowed uncharge of pages under migration because mapcount
3371 * of the page goes down to zero, temporarly.
3372 * Clear the flag and check the page should be charged.
3374 pc
= lookup_page_cgroup(oldpage
);
3375 lock_page_cgroup(pc
);
3376 ClearPageCgroupMigration(pc
);
3377 unlock_page_cgroup(pc
);
3379 __mem_cgroup_uncharge_common(unused
, MEM_CGROUP_CHARGE_TYPE_FORCE
);
3382 * If a page is a file cache, radix-tree replacement is very atomic
3383 * and we can skip this check. When it was an Anon page, its mapcount
3384 * goes down to 0. But because we added MIGRATION flage, it's not
3385 * uncharged yet. There are several case but page->mapcount check
3386 * and USED bit check in mem_cgroup_uncharge_page() will do enough
3387 * check. (see prepare_charge() also)
3390 mem_cgroup_uncharge_page(used
);
3392 * At migration, we may charge account against cgroup which has no
3394 * So, rmdir()->pre_destroy() can be called while we do this charge.
3395 * In that case, we need to call pre_destroy() again. check it here.
3397 cgroup_release_and_wakeup_rmdir(&memcg
->css
);
3401 * At replace page cache, newpage is not under any memcg but it's on
3402 * LRU. So, this function doesn't touch res_counter but handles LRU
3403 * in correct way. Both pages are locked so we cannot race with uncharge.
3405 void mem_cgroup_replace_page_cache(struct page
*oldpage
,
3406 struct page
*newpage
)
3408 struct mem_cgroup
*memcg
;
3409 struct page_cgroup
*pc
;
3411 enum charge_type type
= MEM_CGROUP_CHARGE_TYPE_CACHE
;
3412 unsigned long flags
;
3414 if (mem_cgroup_disabled())
3417 pc
= lookup_page_cgroup(oldpage
);
3418 /* fix accounting on old pages */
3419 lock_page_cgroup(pc
);
3420 memcg
= pc
->mem_cgroup
;
3421 mem_cgroup_charge_statistics(memcg
, PageCgroupCache(pc
), -1);
3422 ClearPageCgroupUsed(pc
);
3423 unlock_page_cgroup(pc
);
3425 if (PageSwapBacked(oldpage
))
3426 type
= MEM_CGROUP_CHARGE_TYPE_SHMEM
;
3428 zone
= page_zone(newpage
);
3429 pc
= lookup_page_cgroup(newpage
);
3431 * Even if newpage->mapping was NULL before starting replacement,
3432 * the newpage may be on LRU(or pagevec for LRU) already. We lock
3433 * LRU while we overwrite pc->mem_cgroup.
3435 spin_lock_irqsave(&zone
->lru_lock
, flags
);
3436 if (PageLRU(newpage
))
3437 del_page_from_lru_list(zone
, newpage
, page_lru(newpage
));
3438 __mem_cgroup_commit_charge(memcg
, newpage
, 1, pc
, type
);
3439 if (PageLRU(newpage
))
3440 add_page_to_lru_list(zone
, newpage
, page_lru(newpage
));
3441 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
3444 #ifdef CONFIG_DEBUG_VM
3445 static struct page_cgroup
*lookup_page_cgroup_used(struct page
*page
)
3447 struct page_cgroup
*pc
;
3449 pc
= lookup_page_cgroup(page
);
3450 if (likely(pc
) && PageCgroupUsed(pc
))
3455 bool mem_cgroup_bad_page_check(struct page
*page
)
3457 if (mem_cgroup_disabled())
3460 return lookup_page_cgroup_used(page
) != NULL
;
3463 void mem_cgroup_print_bad_page(struct page
*page
)
3465 struct page_cgroup
*pc
;
3467 pc
= lookup_page_cgroup_used(page
);
3472 printk(KERN_ALERT
"pc:%p pc->flags:%lx pc->mem_cgroup:%p",
3473 pc
, pc
->flags
, pc
->mem_cgroup
);
3475 path
= kmalloc(PATH_MAX
, GFP_KERNEL
);
3478 ret
= cgroup_path(pc
->mem_cgroup
->css
.cgroup
,
3483 printk(KERN_CONT
"(%s)\n",
3484 (ret
< 0) ? "cannot get the path" : path
);
3490 static DEFINE_MUTEX(set_limit_mutex
);
3492 static int mem_cgroup_resize_limit(struct mem_cgroup
*memcg
,
3493 unsigned long long val
)
3496 u64 memswlimit
, memlimit
;
3498 int children
= mem_cgroup_count_children(memcg
);
3499 u64 curusage
, oldusage
;
3503 * For keeping hierarchical_reclaim simple, how long we should retry
3504 * is depends on callers. We set our retry-count to be function
3505 * of # of children which we should visit in this loop.
3507 retry_count
= MEM_CGROUP_RECLAIM_RETRIES
* children
;
3509 oldusage
= res_counter_read_u64(&memcg
->res
, RES_USAGE
);
3512 while (retry_count
) {
3513 if (signal_pending(current
)) {
3518 * Rather than hide all in some function, I do this in
3519 * open coded manner. You see what this really does.
3520 * We have to guarantee memcg->res.limit < memcg->memsw.limit.
3522 mutex_lock(&set_limit_mutex
);
3523 memswlimit
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
3524 if (memswlimit
< val
) {
3526 mutex_unlock(&set_limit_mutex
);
3530 memlimit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
3534 ret
= res_counter_set_limit(&memcg
->res
, val
);
3536 if (memswlimit
== val
)
3537 memcg
->memsw_is_minimum
= true;
3539 memcg
->memsw_is_minimum
= false;
3541 mutex_unlock(&set_limit_mutex
);
3546 mem_cgroup_reclaim(memcg
, GFP_KERNEL
,
3547 MEM_CGROUP_RECLAIM_SHRINK
);
3548 curusage
= res_counter_read_u64(&memcg
->res
, RES_USAGE
);
3549 /* Usage is reduced ? */
3550 if (curusage
>= oldusage
)
3553 oldusage
= curusage
;
3555 if (!ret
&& enlarge
)
3556 memcg_oom_recover(memcg
);
3561 static int mem_cgroup_resize_memsw_limit(struct mem_cgroup
*memcg
,
3562 unsigned long long val
)
3565 u64 memlimit
, memswlimit
, oldusage
, curusage
;
3566 int children
= mem_cgroup_count_children(memcg
);
3570 /* see mem_cgroup_resize_res_limit */
3571 retry_count
= children
* MEM_CGROUP_RECLAIM_RETRIES
;
3572 oldusage
= res_counter_read_u64(&memcg
->memsw
, RES_USAGE
);
3573 while (retry_count
) {
3574 if (signal_pending(current
)) {
3579 * Rather than hide all in some function, I do this in
3580 * open coded manner. You see what this really does.
3581 * We have to guarantee memcg->res.limit < memcg->memsw.limit.
3583 mutex_lock(&set_limit_mutex
);
3584 memlimit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
3585 if (memlimit
> val
) {
3587 mutex_unlock(&set_limit_mutex
);
3590 memswlimit
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
3591 if (memswlimit
< val
)
3593 ret
= res_counter_set_limit(&memcg
->memsw
, val
);
3595 if (memlimit
== val
)
3596 memcg
->memsw_is_minimum
= true;
3598 memcg
->memsw_is_minimum
= false;
3600 mutex_unlock(&set_limit_mutex
);
3605 mem_cgroup_reclaim(memcg
, GFP_KERNEL
,
3606 MEM_CGROUP_RECLAIM_NOSWAP
|
3607 MEM_CGROUP_RECLAIM_SHRINK
);
3608 curusage
= res_counter_read_u64(&memcg
->memsw
, RES_USAGE
);
3609 /* Usage is reduced ? */
3610 if (curusage
>= oldusage
)
3613 oldusage
= curusage
;
3615 if (!ret
&& enlarge
)
3616 memcg_oom_recover(memcg
);
3620 unsigned long mem_cgroup_soft_limit_reclaim(struct zone
*zone
, int order
,
3622 unsigned long *total_scanned
)
3624 unsigned long nr_reclaimed
= 0;
3625 struct mem_cgroup_per_zone
*mz
, *next_mz
= NULL
;
3626 unsigned long reclaimed
;
3628 struct mem_cgroup_tree_per_zone
*mctz
;
3629 unsigned long long excess
;
3630 unsigned long nr_scanned
;
3635 mctz
= soft_limit_tree_node_zone(zone_to_nid(zone
), zone_idx(zone
));
3637 * This loop can run a while, specially if mem_cgroup's continuously
3638 * keep exceeding their soft limit and putting the system under
3645 mz
= mem_cgroup_largest_soft_limit_node(mctz
);
3650 reclaimed
= mem_cgroup_soft_reclaim(mz
->mem
, zone
,
3651 gfp_mask
, &nr_scanned
);
3652 nr_reclaimed
+= reclaimed
;
3653 *total_scanned
+= nr_scanned
;
3654 spin_lock(&mctz
->lock
);
3657 * If we failed to reclaim anything from this memory cgroup
3658 * it is time to move on to the next cgroup
3664 * Loop until we find yet another one.
3666 * By the time we get the soft_limit lock
3667 * again, someone might have aded the
3668 * group back on the RB tree. Iterate to
3669 * make sure we get a different mem.
3670 * mem_cgroup_largest_soft_limit_node returns
3671 * NULL if no other cgroup is present on
3675 __mem_cgroup_largest_soft_limit_node(mctz
);
3677 css_put(&next_mz
->mem
->css
);
3678 else /* next_mz == NULL or other memcg */
3682 __mem_cgroup_remove_exceeded(mz
->mem
, mz
, mctz
);
3683 excess
= res_counter_soft_limit_excess(&mz
->mem
->res
);
3685 * One school of thought says that we should not add
3686 * back the node to the tree if reclaim returns 0.
3687 * But our reclaim could return 0, simply because due
3688 * to priority we are exposing a smaller subset of
3689 * memory to reclaim from. Consider this as a longer
3692 /* If excess == 0, no tree ops */
3693 __mem_cgroup_insert_exceeded(mz
->mem
, mz
, mctz
, excess
);
3694 spin_unlock(&mctz
->lock
);
3695 css_put(&mz
->mem
->css
);
3698 * Could not reclaim anything and there are no more
3699 * mem cgroups to try or we seem to be looping without
3700 * reclaiming anything.
3702 if (!nr_reclaimed
&&
3704 loop
> MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS
))
3706 } while (!nr_reclaimed
);
3708 css_put(&next_mz
->mem
->css
);
3709 return nr_reclaimed
;
3713 * This routine traverse page_cgroup in given list and drop them all.
3714 * *And* this routine doesn't reclaim page itself, just removes page_cgroup.
3716 static int mem_cgroup_force_empty_list(struct mem_cgroup
*memcg
,
3717 int node
, int zid
, enum lru_list lru
)
3719 struct mem_cgroup_per_zone
*mz
;
3720 unsigned long flags
, loop
;
3721 struct list_head
*list
;
3726 zone
= &NODE_DATA(node
)->node_zones
[zid
];
3727 mz
= mem_cgroup_zoneinfo(memcg
, node
, zid
);
3728 list
= &mz
->lruvec
.lists
[lru
];
3730 loop
= MEM_CGROUP_ZSTAT(mz
, lru
);
3731 /* give some margin against EBUSY etc...*/
3735 struct page_cgroup
*pc
;
3739 spin_lock_irqsave(&zone
->lru_lock
, flags
);
3740 if (list_empty(list
)) {
3741 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
3744 page
= list_entry(list
->prev
, struct page
, lru
);
3746 list_move(&page
->lru
, list
);
3748 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
3751 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
3753 pc
= lookup_page_cgroup(page
);
3755 ret
= mem_cgroup_move_parent(page
, pc
, memcg
, GFP_KERNEL
);
3759 if (ret
== -EBUSY
|| ret
== -EINVAL
) {
3760 /* found lock contention or "pc" is obsolete. */
3767 if (!ret
&& !list_empty(list
))
3773 * make mem_cgroup's charge to be 0 if there is no task.
3774 * This enables deleting this mem_cgroup.
3776 static int mem_cgroup_force_empty(struct mem_cgroup
*memcg
, bool free_all
)
3779 int node
, zid
, shrink
;
3780 int nr_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
3781 struct cgroup
*cgrp
= memcg
->css
.cgroup
;
3783 css_get(&memcg
->css
);
3786 /* should free all ? */
3792 if (cgroup_task_count(cgrp
) || !list_empty(&cgrp
->children
))
3795 if (signal_pending(current
))
3797 /* This is for making all *used* pages to be on LRU. */
3798 lru_add_drain_all();
3799 drain_all_stock_sync(memcg
);
3801 mem_cgroup_start_move(memcg
);
3802 for_each_node_state(node
, N_HIGH_MEMORY
) {
3803 for (zid
= 0; !ret
&& zid
< MAX_NR_ZONES
; zid
++) {
3806 ret
= mem_cgroup_force_empty_list(memcg
,
3815 mem_cgroup_end_move(memcg
);
3816 memcg_oom_recover(memcg
);
3817 /* it seems parent cgroup doesn't have enough mem */
3821 /* "ret" should also be checked to ensure all lists are empty. */
3822 } while (memcg
->res
.usage
> 0 || ret
);
3824 css_put(&memcg
->css
);
3828 /* returns EBUSY if there is a task or if we come here twice. */
3829 if (cgroup_task_count(cgrp
) || !list_empty(&cgrp
->children
) || shrink
) {
3833 /* we call try-to-free pages for make this cgroup empty */
3834 lru_add_drain_all();
3835 /* try to free all pages in this cgroup */
3837 while (nr_retries
&& memcg
->res
.usage
> 0) {
3840 if (signal_pending(current
)) {
3844 progress
= try_to_free_mem_cgroup_pages(memcg
, GFP_KERNEL
,
3848 /* maybe some writeback is necessary */
3849 congestion_wait(BLK_RW_ASYNC
, HZ
/10);
3854 /* try move_account...there may be some *locked* pages. */
3858 int mem_cgroup_force_empty_write(struct cgroup
*cont
, unsigned int event
)
3860 return mem_cgroup_force_empty(mem_cgroup_from_cont(cont
), true);
3864 static u64
mem_cgroup_hierarchy_read(struct cgroup
*cont
, struct cftype
*cft
)
3866 return mem_cgroup_from_cont(cont
)->use_hierarchy
;
3869 static int mem_cgroup_hierarchy_write(struct cgroup
*cont
, struct cftype
*cft
,
3873 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
3874 struct cgroup
*parent
= cont
->parent
;
3875 struct mem_cgroup
*parent_memcg
= NULL
;
3878 parent_memcg
= mem_cgroup_from_cont(parent
);
3882 * If parent's use_hierarchy is set, we can't make any modifications
3883 * in the child subtrees. If it is unset, then the change can
3884 * occur, provided the current cgroup has no children.
3886 * For the root cgroup, parent_mem is NULL, we allow value to be
3887 * set if there are no children.
3889 if ((!parent_memcg
|| !parent_memcg
->use_hierarchy
) &&
3890 (val
== 1 || val
== 0)) {
3891 if (list_empty(&cont
->children
))
3892 memcg
->use_hierarchy
= val
;
3903 static unsigned long mem_cgroup_recursive_stat(struct mem_cgroup
*memcg
,
3904 enum mem_cgroup_stat_index idx
)
3906 struct mem_cgroup
*iter
;
3909 /* Per-cpu values can be negative, use a signed accumulator */
3910 for_each_mem_cgroup_tree(iter
, memcg
)
3911 val
+= mem_cgroup_read_stat(iter
, idx
);
3913 if (val
< 0) /* race ? */
3918 static inline u64
mem_cgroup_usage(struct mem_cgroup
*memcg
, bool swap
)
3922 if (!mem_cgroup_is_root(memcg
)) {
3924 return res_counter_read_u64(&memcg
->res
, RES_USAGE
);
3926 return res_counter_read_u64(&memcg
->memsw
, RES_USAGE
);
3929 val
= mem_cgroup_recursive_stat(memcg
, MEM_CGROUP_STAT_CACHE
);
3930 val
+= mem_cgroup_recursive_stat(memcg
, MEM_CGROUP_STAT_RSS
);
3933 val
+= mem_cgroup_recursive_stat(memcg
, MEM_CGROUP_STAT_SWAPOUT
);
3935 return val
<< PAGE_SHIFT
;
3938 static u64
mem_cgroup_read(struct cgroup
*cont
, struct cftype
*cft
)
3940 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
3944 type
= MEMFILE_TYPE(cft
->private);
3945 name
= MEMFILE_ATTR(cft
->private);
3948 if (name
== RES_USAGE
)
3949 val
= mem_cgroup_usage(memcg
, false);
3951 val
= res_counter_read_u64(&memcg
->res
, name
);
3954 if (name
== RES_USAGE
)
3955 val
= mem_cgroup_usage(memcg
, true);
3957 val
= res_counter_read_u64(&memcg
->memsw
, name
);
3966 * The user of this function is...
3969 static int mem_cgroup_write(struct cgroup
*cont
, struct cftype
*cft
,
3972 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
3974 unsigned long long val
;
3977 type
= MEMFILE_TYPE(cft
->private);
3978 name
= MEMFILE_ATTR(cft
->private);
3981 if (mem_cgroup_is_root(memcg
)) { /* Can't set limit on root */
3985 /* This function does all necessary parse...reuse it */
3986 ret
= res_counter_memparse_write_strategy(buffer
, &val
);
3990 ret
= mem_cgroup_resize_limit(memcg
, val
);
3992 ret
= mem_cgroup_resize_memsw_limit(memcg
, val
);
3994 case RES_SOFT_LIMIT
:
3995 ret
= res_counter_memparse_write_strategy(buffer
, &val
);
3999 * For memsw, soft limits are hard to implement in terms
4000 * of semantics, for now, we support soft limits for
4001 * control without swap
4004 ret
= res_counter_set_soft_limit(&memcg
->res
, val
);
4009 ret
= -EINVAL
; /* should be BUG() ? */
4015 static void memcg_get_hierarchical_limit(struct mem_cgroup
*memcg
,
4016 unsigned long long *mem_limit
, unsigned long long *memsw_limit
)
4018 struct cgroup
*cgroup
;
4019 unsigned long long min_limit
, min_memsw_limit
, tmp
;
4021 min_limit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
4022 min_memsw_limit
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
4023 cgroup
= memcg
->css
.cgroup
;
4024 if (!memcg
->use_hierarchy
)
4027 while (cgroup
->parent
) {
4028 cgroup
= cgroup
->parent
;
4029 memcg
= mem_cgroup_from_cont(cgroup
);
4030 if (!memcg
->use_hierarchy
)
4032 tmp
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
4033 min_limit
= min(min_limit
, tmp
);
4034 tmp
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
4035 min_memsw_limit
= min(min_memsw_limit
, tmp
);
4038 *mem_limit
= min_limit
;
4039 *memsw_limit
= min_memsw_limit
;
4043 static int mem_cgroup_reset(struct cgroup
*cont
, unsigned int event
)
4045 struct mem_cgroup
*memcg
;
4048 memcg
= mem_cgroup_from_cont(cont
);
4049 type
= MEMFILE_TYPE(event
);
4050 name
= MEMFILE_ATTR(event
);
4054 res_counter_reset_max(&memcg
->res
);
4056 res_counter_reset_max(&memcg
->memsw
);
4060 res_counter_reset_failcnt(&memcg
->res
);
4062 res_counter_reset_failcnt(&memcg
->memsw
);
4069 static u64
mem_cgroup_move_charge_read(struct cgroup
*cgrp
,
4072 return mem_cgroup_from_cont(cgrp
)->move_charge_at_immigrate
;
4076 static int mem_cgroup_move_charge_write(struct cgroup
*cgrp
,
4077 struct cftype
*cft
, u64 val
)
4079 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4081 if (val
>= (1 << NR_MOVE_TYPE
))
4084 * We check this value several times in both in can_attach() and
4085 * attach(), so we need cgroup lock to prevent this value from being
4089 memcg
->move_charge_at_immigrate
= val
;
4095 static int mem_cgroup_move_charge_write(struct cgroup
*cgrp
,
4096 struct cftype
*cft
, u64 val
)
4103 /* For read statistics */
4121 struct mcs_total_stat
{
4122 s64 stat
[NR_MCS_STAT
];
4128 } memcg_stat_strings
[NR_MCS_STAT
] = {
4129 {"cache", "total_cache"},
4130 {"rss", "total_rss"},
4131 {"mapped_file", "total_mapped_file"},
4132 {"pgpgin", "total_pgpgin"},
4133 {"pgpgout", "total_pgpgout"},
4134 {"swap", "total_swap"},
4135 {"pgfault", "total_pgfault"},
4136 {"pgmajfault", "total_pgmajfault"},
4137 {"inactive_anon", "total_inactive_anon"},
4138 {"active_anon", "total_active_anon"},
4139 {"inactive_file", "total_inactive_file"},
4140 {"active_file", "total_active_file"},
4141 {"unevictable", "total_unevictable"}
4146 mem_cgroup_get_local_stat(struct mem_cgroup
*memcg
, struct mcs_total_stat
*s
)
4151 val
= mem_cgroup_read_stat(memcg
, MEM_CGROUP_STAT_CACHE
);
4152 s
->stat
[MCS_CACHE
] += val
* PAGE_SIZE
;
4153 val
= mem_cgroup_read_stat(memcg
, MEM_CGROUP_STAT_RSS
);
4154 s
->stat
[MCS_RSS
] += val
* PAGE_SIZE
;
4155 val
= mem_cgroup_read_stat(memcg
, MEM_CGROUP_STAT_FILE_MAPPED
);
4156 s
->stat
[MCS_FILE_MAPPED
] += val
* PAGE_SIZE
;
4157 val
= mem_cgroup_read_events(memcg
, MEM_CGROUP_EVENTS_PGPGIN
);
4158 s
->stat
[MCS_PGPGIN
] += val
;
4159 val
= mem_cgroup_read_events(memcg
, MEM_CGROUP_EVENTS_PGPGOUT
);
4160 s
->stat
[MCS_PGPGOUT
] += val
;
4161 if (do_swap_account
) {
4162 val
= mem_cgroup_read_stat(memcg
, MEM_CGROUP_STAT_SWAPOUT
);
4163 s
->stat
[MCS_SWAP
] += val
* PAGE_SIZE
;
4165 val
= mem_cgroup_read_events(memcg
, MEM_CGROUP_EVENTS_PGFAULT
);
4166 s
->stat
[MCS_PGFAULT
] += val
;
4167 val
= mem_cgroup_read_events(memcg
, MEM_CGROUP_EVENTS_PGMAJFAULT
);
4168 s
->stat
[MCS_PGMAJFAULT
] += val
;
4171 val
= mem_cgroup_nr_lru_pages(memcg
, BIT(LRU_INACTIVE_ANON
));
4172 s
->stat
[MCS_INACTIVE_ANON
] += val
* PAGE_SIZE
;
4173 val
= mem_cgroup_nr_lru_pages(memcg
, BIT(LRU_ACTIVE_ANON
));
4174 s
->stat
[MCS_ACTIVE_ANON
] += val
* PAGE_SIZE
;
4175 val
= mem_cgroup_nr_lru_pages(memcg
, BIT(LRU_INACTIVE_FILE
));
4176 s
->stat
[MCS_INACTIVE_FILE
] += val
* PAGE_SIZE
;
4177 val
= mem_cgroup_nr_lru_pages(memcg
, BIT(LRU_ACTIVE_FILE
));
4178 s
->stat
[MCS_ACTIVE_FILE
] += val
* PAGE_SIZE
;
4179 val
= mem_cgroup_nr_lru_pages(memcg
, BIT(LRU_UNEVICTABLE
));
4180 s
->stat
[MCS_UNEVICTABLE
] += val
* PAGE_SIZE
;
4184 mem_cgroup_get_total_stat(struct mem_cgroup
*memcg
, struct mcs_total_stat
*s
)
4186 struct mem_cgroup
*iter
;
4188 for_each_mem_cgroup_tree(iter
, memcg
)
4189 mem_cgroup_get_local_stat(iter
, s
);
4193 static int mem_control_numa_stat_show(struct seq_file
*m
, void *arg
)
4196 unsigned long total_nr
, file_nr
, anon_nr
, unevictable_nr
;
4197 unsigned long node_nr
;
4198 struct cgroup
*cont
= m
->private;
4199 struct mem_cgroup
*mem_cont
= mem_cgroup_from_cont(cont
);
4201 total_nr
= mem_cgroup_nr_lru_pages(mem_cont
, LRU_ALL
);
4202 seq_printf(m
, "total=%lu", total_nr
);
4203 for_each_node_state(nid
, N_HIGH_MEMORY
) {
4204 node_nr
= mem_cgroup_node_nr_lru_pages(mem_cont
, nid
, LRU_ALL
);
4205 seq_printf(m
, " N%d=%lu", nid
, node_nr
);
4209 file_nr
= mem_cgroup_nr_lru_pages(mem_cont
, LRU_ALL_FILE
);
4210 seq_printf(m
, "file=%lu", file_nr
);
4211 for_each_node_state(nid
, N_HIGH_MEMORY
) {
4212 node_nr
= mem_cgroup_node_nr_lru_pages(mem_cont
, nid
,
4214 seq_printf(m
, " N%d=%lu", nid
, node_nr
);
4218 anon_nr
= mem_cgroup_nr_lru_pages(mem_cont
, LRU_ALL_ANON
);
4219 seq_printf(m
, "anon=%lu", anon_nr
);
4220 for_each_node_state(nid
, N_HIGH_MEMORY
) {
4221 node_nr
= mem_cgroup_node_nr_lru_pages(mem_cont
, nid
,
4223 seq_printf(m
, " N%d=%lu", nid
, node_nr
);
4227 unevictable_nr
= mem_cgroup_nr_lru_pages(mem_cont
, BIT(LRU_UNEVICTABLE
));
4228 seq_printf(m
, "unevictable=%lu", unevictable_nr
);
4229 for_each_node_state(nid
, N_HIGH_MEMORY
) {
4230 node_nr
= mem_cgroup_node_nr_lru_pages(mem_cont
, nid
,
4231 BIT(LRU_UNEVICTABLE
));
4232 seq_printf(m
, " N%d=%lu", nid
, node_nr
);
4237 #endif /* CONFIG_NUMA */
4239 static int mem_control_stat_show(struct cgroup
*cont
, struct cftype
*cft
,
4240 struct cgroup_map_cb
*cb
)
4242 struct mem_cgroup
*mem_cont
= mem_cgroup_from_cont(cont
);
4243 struct mcs_total_stat mystat
;
4246 memset(&mystat
, 0, sizeof(mystat
));
4247 mem_cgroup_get_local_stat(mem_cont
, &mystat
);
4250 for (i
= 0; i
< NR_MCS_STAT
; i
++) {
4251 if (i
== MCS_SWAP
&& !do_swap_account
)
4253 cb
->fill(cb
, memcg_stat_strings
[i
].local_name
, mystat
.stat
[i
]);
4256 /* Hierarchical information */
4258 unsigned long long limit
, memsw_limit
;
4259 memcg_get_hierarchical_limit(mem_cont
, &limit
, &memsw_limit
);
4260 cb
->fill(cb
, "hierarchical_memory_limit", limit
);
4261 if (do_swap_account
)
4262 cb
->fill(cb
, "hierarchical_memsw_limit", memsw_limit
);
4265 memset(&mystat
, 0, sizeof(mystat
));
4266 mem_cgroup_get_total_stat(mem_cont
, &mystat
);
4267 for (i
= 0; i
< NR_MCS_STAT
; i
++) {
4268 if (i
== MCS_SWAP
&& !do_swap_account
)
4270 cb
->fill(cb
, memcg_stat_strings
[i
].total_name
, mystat
.stat
[i
]);
4273 #ifdef CONFIG_DEBUG_VM
4276 struct mem_cgroup_per_zone
*mz
;
4277 unsigned long recent_rotated
[2] = {0, 0};
4278 unsigned long recent_scanned
[2] = {0, 0};
4280 for_each_online_node(nid
)
4281 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++) {
4282 mz
= mem_cgroup_zoneinfo(mem_cont
, nid
, zid
);
4284 recent_rotated
[0] +=
4285 mz
->reclaim_stat
.recent_rotated
[0];
4286 recent_rotated
[1] +=
4287 mz
->reclaim_stat
.recent_rotated
[1];
4288 recent_scanned
[0] +=
4289 mz
->reclaim_stat
.recent_scanned
[0];
4290 recent_scanned
[1] +=
4291 mz
->reclaim_stat
.recent_scanned
[1];
4293 cb
->fill(cb
, "recent_rotated_anon", recent_rotated
[0]);
4294 cb
->fill(cb
, "recent_rotated_file", recent_rotated
[1]);
4295 cb
->fill(cb
, "recent_scanned_anon", recent_scanned
[0]);
4296 cb
->fill(cb
, "recent_scanned_file", recent_scanned
[1]);
4303 static u64
mem_cgroup_swappiness_read(struct cgroup
*cgrp
, struct cftype
*cft
)
4305 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4307 return mem_cgroup_swappiness(memcg
);
4310 static int mem_cgroup_swappiness_write(struct cgroup
*cgrp
, struct cftype
*cft
,
4313 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4314 struct mem_cgroup
*parent
;
4319 if (cgrp
->parent
== NULL
)
4322 parent
= mem_cgroup_from_cont(cgrp
->parent
);
4326 /* If under hierarchy, only empty-root can set this value */
4327 if ((parent
->use_hierarchy
) ||
4328 (memcg
->use_hierarchy
&& !list_empty(&cgrp
->children
))) {
4333 memcg
->swappiness
= val
;
4340 static void __mem_cgroup_threshold(struct mem_cgroup
*memcg
, bool swap
)
4342 struct mem_cgroup_threshold_ary
*t
;
4348 t
= rcu_dereference(memcg
->thresholds
.primary
);
4350 t
= rcu_dereference(memcg
->memsw_thresholds
.primary
);
4355 usage
= mem_cgroup_usage(memcg
, swap
);
4358 * current_threshold points to threshold just below usage.
4359 * If it's not true, a threshold was crossed after last
4360 * call of __mem_cgroup_threshold().
4362 i
= t
->current_threshold
;
4365 * Iterate backward over array of thresholds starting from
4366 * current_threshold and check if a threshold is crossed.
4367 * If none of thresholds below usage is crossed, we read
4368 * only one element of the array here.
4370 for (; i
>= 0 && unlikely(t
->entries
[i
].threshold
> usage
); i
--)
4371 eventfd_signal(t
->entries
[i
].eventfd
, 1);
4373 /* i = current_threshold + 1 */
4377 * Iterate forward over array of thresholds starting from
4378 * current_threshold+1 and check if a threshold is crossed.
4379 * If none of thresholds above usage is crossed, we read
4380 * only one element of the array here.
4382 for (; i
< t
->size
&& unlikely(t
->entries
[i
].threshold
<= usage
); i
++)
4383 eventfd_signal(t
->entries
[i
].eventfd
, 1);
4385 /* Update current_threshold */
4386 t
->current_threshold
= i
- 1;
4391 static void mem_cgroup_threshold(struct mem_cgroup
*memcg
)
4394 __mem_cgroup_threshold(memcg
, false);
4395 if (do_swap_account
)
4396 __mem_cgroup_threshold(memcg
, true);
4398 memcg
= parent_mem_cgroup(memcg
);
4402 static int compare_thresholds(const void *a
, const void *b
)
4404 const struct mem_cgroup_threshold
*_a
= a
;
4405 const struct mem_cgroup_threshold
*_b
= b
;
4407 return _a
->threshold
- _b
->threshold
;
4410 static int mem_cgroup_oom_notify_cb(struct mem_cgroup
*memcg
)
4412 struct mem_cgroup_eventfd_list
*ev
;
4414 list_for_each_entry(ev
, &memcg
->oom_notify
, list
)
4415 eventfd_signal(ev
->eventfd
, 1);
4419 static void mem_cgroup_oom_notify(struct mem_cgroup
*memcg
)
4421 struct mem_cgroup
*iter
;
4423 for_each_mem_cgroup_tree(iter
, memcg
)
4424 mem_cgroup_oom_notify_cb(iter
);
4427 static int mem_cgroup_usage_register_event(struct cgroup
*cgrp
,
4428 struct cftype
*cft
, struct eventfd_ctx
*eventfd
, const char *args
)
4430 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4431 struct mem_cgroup_thresholds
*thresholds
;
4432 struct mem_cgroup_threshold_ary
*new;
4433 int type
= MEMFILE_TYPE(cft
->private);
4434 u64 threshold
, usage
;
4437 ret
= res_counter_memparse_write_strategy(args
, &threshold
);
4441 mutex_lock(&memcg
->thresholds_lock
);
4444 thresholds
= &memcg
->thresholds
;
4445 else if (type
== _MEMSWAP
)
4446 thresholds
= &memcg
->memsw_thresholds
;
4450 usage
= mem_cgroup_usage(memcg
, type
== _MEMSWAP
);
4452 /* Check if a threshold crossed before adding a new one */
4453 if (thresholds
->primary
)
4454 __mem_cgroup_threshold(memcg
, type
== _MEMSWAP
);
4456 size
= thresholds
->primary
? thresholds
->primary
->size
+ 1 : 1;
4458 /* Allocate memory for new array of thresholds */
4459 new = kmalloc(sizeof(*new) + size
* sizeof(struct mem_cgroup_threshold
),
4467 /* Copy thresholds (if any) to new array */
4468 if (thresholds
->primary
) {
4469 memcpy(new->entries
, thresholds
->primary
->entries
, (size
- 1) *
4470 sizeof(struct mem_cgroup_threshold
));
4473 /* Add new threshold */
4474 new->entries
[size
- 1].eventfd
= eventfd
;
4475 new->entries
[size
- 1].threshold
= threshold
;
4477 /* Sort thresholds. Registering of new threshold isn't time-critical */
4478 sort(new->entries
, size
, sizeof(struct mem_cgroup_threshold
),
4479 compare_thresholds
, NULL
);
4481 /* Find current threshold */
4482 new->current_threshold
= -1;
4483 for (i
= 0; i
< size
; i
++) {
4484 if (new->entries
[i
].threshold
< usage
) {
4486 * new->current_threshold will not be used until
4487 * rcu_assign_pointer(), so it's safe to increment
4490 ++new->current_threshold
;
4494 /* Free old spare buffer and save old primary buffer as spare */
4495 kfree(thresholds
->spare
);
4496 thresholds
->spare
= thresholds
->primary
;
4498 rcu_assign_pointer(thresholds
->primary
, new);
4500 /* To be sure that nobody uses thresholds */
4504 mutex_unlock(&memcg
->thresholds_lock
);
4509 static void mem_cgroup_usage_unregister_event(struct cgroup
*cgrp
,
4510 struct cftype
*cft
, struct eventfd_ctx
*eventfd
)
4512 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4513 struct mem_cgroup_thresholds
*thresholds
;
4514 struct mem_cgroup_threshold_ary
*new;
4515 int type
= MEMFILE_TYPE(cft
->private);
4519 mutex_lock(&memcg
->thresholds_lock
);
4521 thresholds
= &memcg
->thresholds
;
4522 else if (type
== _MEMSWAP
)
4523 thresholds
= &memcg
->memsw_thresholds
;
4528 * Something went wrong if we trying to unregister a threshold
4529 * if we don't have thresholds
4531 BUG_ON(!thresholds
);
4533 usage
= mem_cgroup_usage(memcg
, type
== _MEMSWAP
);
4535 /* Check if a threshold crossed before removing */
4536 __mem_cgroup_threshold(memcg
, type
== _MEMSWAP
);
4538 /* Calculate new number of threshold */
4540 for (i
= 0; i
< thresholds
->primary
->size
; i
++) {
4541 if (thresholds
->primary
->entries
[i
].eventfd
!= eventfd
)
4545 new = thresholds
->spare
;
4547 /* Set thresholds array to NULL if we don't have thresholds */
4556 /* Copy thresholds and find current threshold */
4557 new->current_threshold
= -1;
4558 for (i
= 0, j
= 0; i
< thresholds
->primary
->size
; i
++) {
4559 if (thresholds
->primary
->entries
[i
].eventfd
== eventfd
)
4562 new->entries
[j
] = thresholds
->primary
->entries
[i
];
4563 if (new->entries
[j
].threshold
< usage
) {
4565 * new->current_threshold will not be used
4566 * until rcu_assign_pointer(), so it's safe to increment
4569 ++new->current_threshold
;
4575 /* Swap primary and spare array */
4576 thresholds
->spare
= thresholds
->primary
;
4577 rcu_assign_pointer(thresholds
->primary
, new);
4579 /* To be sure that nobody uses thresholds */
4582 mutex_unlock(&memcg
->thresholds_lock
);
4585 static int mem_cgroup_oom_register_event(struct cgroup
*cgrp
,
4586 struct cftype
*cft
, struct eventfd_ctx
*eventfd
, const char *args
)
4588 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4589 struct mem_cgroup_eventfd_list
*event
;
4590 int type
= MEMFILE_TYPE(cft
->private);
4592 BUG_ON(type
!= _OOM_TYPE
);
4593 event
= kmalloc(sizeof(*event
), GFP_KERNEL
);
4597 spin_lock(&memcg_oom_lock
);
4599 event
->eventfd
= eventfd
;
4600 list_add(&event
->list
, &memcg
->oom_notify
);
4602 /* already in OOM ? */
4603 if (atomic_read(&memcg
->under_oom
))
4604 eventfd_signal(eventfd
, 1);
4605 spin_unlock(&memcg_oom_lock
);
4610 static void mem_cgroup_oom_unregister_event(struct cgroup
*cgrp
,
4611 struct cftype
*cft
, struct eventfd_ctx
*eventfd
)
4613 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4614 struct mem_cgroup_eventfd_list
*ev
, *tmp
;
4615 int type
= MEMFILE_TYPE(cft
->private);
4617 BUG_ON(type
!= _OOM_TYPE
);
4619 spin_lock(&memcg_oom_lock
);
4621 list_for_each_entry_safe(ev
, tmp
, &memcg
->oom_notify
, list
) {
4622 if (ev
->eventfd
== eventfd
) {
4623 list_del(&ev
->list
);
4628 spin_unlock(&memcg_oom_lock
);
4631 static int mem_cgroup_oom_control_read(struct cgroup
*cgrp
,
4632 struct cftype
*cft
, struct cgroup_map_cb
*cb
)
4634 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4636 cb
->fill(cb
, "oom_kill_disable", memcg
->oom_kill_disable
);
4638 if (atomic_read(&memcg
->under_oom
))
4639 cb
->fill(cb
, "under_oom", 1);
4641 cb
->fill(cb
, "under_oom", 0);
4645 static int mem_cgroup_oom_control_write(struct cgroup
*cgrp
,
4646 struct cftype
*cft
, u64 val
)
4648 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4649 struct mem_cgroup
*parent
;
4651 /* cannot set to root cgroup and only 0 and 1 are allowed */
4652 if (!cgrp
->parent
|| !((val
== 0) || (val
== 1)))
4655 parent
= mem_cgroup_from_cont(cgrp
->parent
);
4658 /* oom-kill-disable is a flag for subhierarchy. */
4659 if ((parent
->use_hierarchy
) ||
4660 (memcg
->use_hierarchy
&& !list_empty(&cgrp
->children
))) {
4664 memcg
->oom_kill_disable
= val
;
4666 memcg_oom_recover(memcg
);
4672 static const struct file_operations mem_control_numa_stat_file_operations
= {
4674 .llseek
= seq_lseek
,
4675 .release
= single_release
,
4678 static int mem_control_numa_stat_open(struct inode
*unused
, struct file
*file
)
4680 struct cgroup
*cont
= file
->f_dentry
->d_parent
->d_fsdata
;
4682 file
->f_op
= &mem_control_numa_stat_file_operations
;
4683 return single_open(file
, mem_control_numa_stat_show
, cont
);
4685 #endif /* CONFIG_NUMA */
4687 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_KMEM
4688 static int register_kmem_files(struct cgroup
*cont
, struct cgroup_subsys
*ss
)
4691 * Part of this would be better living in a separate allocation
4692 * function, leaving us with just the cgroup tree population work.
4693 * We, however, depend on state such as network's proto_list that
4694 * is only initialized after cgroup creation. I found the less
4695 * cumbersome way to deal with it to defer it all to populate time
4697 return mem_cgroup_sockets_init(cont
, ss
);
4700 static void kmem_cgroup_destroy(struct cgroup_subsys
*ss
,
4701 struct cgroup
*cont
)
4703 mem_cgroup_sockets_destroy(cont
, ss
);
4706 static int register_kmem_files(struct cgroup
*cont
, struct cgroup_subsys
*ss
)
4711 static void kmem_cgroup_destroy(struct cgroup_subsys
*ss
,
4712 struct cgroup
*cont
)
4717 static struct cftype mem_cgroup_files
[] = {
4719 .name
= "usage_in_bytes",
4720 .private = MEMFILE_PRIVATE(_MEM
, RES_USAGE
),
4721 .read_u64
= mem_cgroup_read
,
4722 .register_event
= mem_cgroup_usage_register_event
,
4723 .unregister_event
= mem_cgroup_usage_unregister_event
,
4726 .name
= "max_usage_in_bytes",
4727 .private = MEMFILE_PRIVATE(_MEM
, RES_MAX_USAGE
),
4728 .trigger
= mem_cgroup_reset
,
4729 .read_u64
= mem_cgroup_read
,
4732 .name
= "limit_in_bytes",
4733 .private = MEMFILE_PRIVATE(_MEM
, RES_LIMIT
),
4734 .write_string
= mem_cgroup_write
,
4735 .read_u64
= mem_cgroup_read
,
4738 .name
= "soft_limit_in_bytes",
4739 .private = MEMFILE_PRIVATE(_MEM
, RES_SOFT_LIMIT
),
4740 .write_string
= mem_cgroup_write
,
4741 .read_u64
= mem_cgroup_read
,
4745 .private = MEMFILE_PRIVATE(_MEM
, RES_FAILCNT
),
4746 .trigger
= mem_cgroup_reset
,
4747 .read_u64
= mem_cgroup_read
,
4751 .read_map
= mem_control_stat_show
,
4754 .name
= "force_empty",
4755 .trigger
= mem_cgroup_force_empty_write
,
4758 .name
= "use_hierarchy",
4759 .write_u64
= mem_cgroup_hierarchy_write
,
4760 .read_u64
= mem_cgroup_hierarchy_read
,
4763 .name
= "swappiness",
4764 .read_u64
= mem_cgroup_swappiness_read
,
4765 .write_u64
= mem_cgroup_swappiness_write
,
4768 .name
= "move_charge_at_immigrate",
4769 .read_u64
= mem_cgroup_move_charge_read
,
4770 .write_u64
= mem_cgroup_move_charge_write
,
4773 .name
= "oom_control",
4774 .read_map
= mem_cgroup_oom_control_read
,
4775 .write_u64
= mem_cgroup_oom_control_write
,
4776 .register_event
= mem_cgroup_oom_register_event
,
4777 .unregister_event
= mem_cgroup_oom_unregister_event
,
4778 .private = MEMFILE_PRIVATE(_OOM_TYPE
, OOM_CONTROL
),
4782 .name
= "numa_stat",
4783 .open
= mem_control_numa_stat_open
,
4789 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
4790 static struct cftype memsw_cgroup_files
[] = {
4792 .name
= "memsw.usage_in_bytes",
4793 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_USAGE
),
4794 .read_u64
= mem_cgroup_read
,
4795 .register_event
= mem_cgroup_usage_register_event
,
4796 .unregister_event
= mem_cgroup_usage_unregister_event
,
4799 .name
= "memsw.max_usage_in_bytes",
4800 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_MAX_USAGE
),
4801 .trigger
= mem_cgroup_reset
,
4802 .read_u64
= mem_cgroup_read
,
4805 .name
= "memsw.limit_in_bytes",
4806 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_LIMIT
),
4807 .write_string
= mem_cgroup_write
,
4808 .read_u64
= mem_cgroup_read
,
4811 .name
= "memsw.failcnt",
4812 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_FAILCNT
),
4813 .trigger
= mem_cgroup_reset
,
4814 .read_u64
= mem_cgroup_read
,
4818 static int register_memsw_files(struct cgroup
*cont
, struct cgroup_subsys
*ss
)
4820 if (!do_swap_account
)
4822 return cgroup_add_files(cont
, ss
, memsw_cgroup_files
,
4823 ARRAY_SIZE(memsw_cgroup_files
));
4826 static int register_memsw_files(struct cgroup
*cont
, struct cgroup_subsys
*ss
)
4832 static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup
*memcg
, int node
)
4834 struct mem_cgroup_per_node
*pn
;
4835 struct mem_cgroup_per_zone
*mz
;
4837 int zone
, tmp
= node
;
4839 * This routine is called against possible nodes.
4840 * But it's BUG to call kmalloc() against offline node.
4842 * TODO: this routine can waste much memory for nodes which will
4843 * never be onlined. It's better to use memory hotplug callback
4846 if (!node_state(node
, N_NORMAL_MEMORY
))
4848 pn
= kzalloc_node(sizeof(*pn
), GFP_KERNEL
, tmp
);
4852 for (zone
= 0; zone
< MAX_NR_ZONES
; zone
++) {
4853 mz
= &pn
->zoneinfo
[zone
];
4855 INIT_LIST_HEAD(&mz
->lruvec
.lists
[l
]);
4856 mz
->usage_in_excess
= 0;
4857 mz
->on_tree
= false;
4860 memcg
->info
.nodeinfo
[node
] = pn
;
4864 static void free_mem_cgroup_per_zone_info(struct mem_cgroup
*memcg
, int node
)
4866 kfree(memcg
->info
.nodeinfo
[node
]);
4869 static struct mem_cgroup
*mem_cgroup_alloc(void)
4871 struct mem_cgroup
*mem
;
4872 int size
= sizeof(struct mem_cgroup
);
4874 /* Can be very big if MAX_NUMNODES is very big */
4875 if (size
< PAGE_SIZE
)
4876 mem
= kzalloc(size
, GFP_KERNEL
);
4878 mem
= vzalloc(size
);
4883 mem
->stat
= alloc_percpu(struct mem_cgroup_stat_cpu
);
4886 spin_lock_init(&mem
->pcp_counter_lock
);
4890 if (size
< PAGE_SIZE
)
4898 * At destroying mem_cgroup, references from swap_cgroup can remain.
4899 * (scanning all at force_empty is too costly...)
4901 * Instead of clearing all references at force_empty, we remember
4902 * the number of reference from swap_cgroup and free mem_cgroup when
4903 * it goes down to 0.
4905 * Removal of cgroup itself succeeds regardless of refs from swap.
4908 static void __mem_cgroup_free(struct mem_cgroup
*memcg
)
4912 mem_cgroup_remove_from_trees(memcg
);
4913 free_css_id(&mem_cgroup_subsys
, &memcg
->css
);
4915 for_each_node_state(node
, N_POSSIBLE
)
4916 free_mem_cgroup_per_zone_info(memcg
, node
);
4918 free_percpu(memcg
->stat
);
4919 if (sizeof(struct mem_cgroup
) < PAGE_SIZE
)
4925 static void mem_cgroup_get(struct mem_cgroup
*memcg
)
4927 atomic_inc(&memcg
->refcnt
);
4930 static void __mem_cgroup_put(struct mem_cgroup
*memcg
, int count
)
4932 if (atomic_sub_and_test(count
, &memcg
->refcnt
)) {
4933 struct mem_cgroup
*parent
= parent_mem_cgroup(memcg
);
4934 __mem_cgroup_free(memcg
);
4936 mem_cgroup_put(parent
);
4940 static void mem_cgroup_put(struct mem_cgroup
*memcg
)
4942 __mem_cgroup_put(memcg
, 1);
4946 * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled.
4948 struct mem_cgroup
*parent_mem_cgroup(struct mem_cgroup
*memcg
)
4950 if (!memcg
->res
.parent
)
4952 return mem_cgroup_from_res_counter(memcg
->res
.parent
, res
);
4954 EXPORT_SYMBOL(parent_mem_cgroup
);
4956 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
4957 static void __init
enable_swap_cgroup(void)
4959 if (!mem_cgroup_disabled() && really_do_swap_account
)
4960 do_swap_account
= 1;
4963 static void __init
enable_swap_cgroup(void)
4968 static int mem_cgroup_soft_limit_tree_init(void)
4970 struct mem_cgroup_tree_per_node
*rtpn
;
4971 struct mem_cgroup_tree_per_zone
*rtpz
;
4972 int tmp
, node
, zone
;
4974 for_each_node_state(node
, N_POSSIBLE
) {
4976 if (!node_state(node
, N_NORMAL_MEMORY
))
4978 rtpn
= kzalloc_node(sizeof(*rtpn
), GFP_KERNEL
, tmp
);
4982 soft_limit_tree
.rb_tree_per_node
[node
] = rtpn
;
4984 for (zone
= 0; zone
< MAX_NR_ZONES
; zone
++) {
4985 rtpz
= &rtpn
->rb_tree_per_zone
[zone
];
4986 rtpz
->rb_root
= RB_ROOT
;
4987 spin_lock_init(&rtpz
->lock
);
4993 static struct cgroup_subsys_state
* __ref
4994 mem_cgroup_create(struct cgroup_subsys
*ss
, struct cgroup
*cont
)
4996 struct mem_cgroup
*memcg
, *parent
;
4997 long error
= -ENOMEM
;
5000 memcg
= mem_cgroup_alloc();
5002 return ERR_PTR(error
);
5004 for_each_node_state(node
, N_POSSIBLE
)
5005 if (alloc_mem_cgroup_per_zone_info(memcg
, node
))
5009 if (cont
->parent
== NULL
) {
5011 enable_swap_cgroup();
5013 if (mem_cgroup_soft_limit_tree_init())
5015 root_mem_cgroup
= memcg
;
5016 for_each_possible_cpu(cpu
) {
5017 struct memcg_stock_pcp
*stock
=
5018 &per_cpu(memcg_stock
, cpu
);
5019 INIT_WORK(&stock
->work
, drain_local_stock
);
5021 hotcpu_notifier(memcg_cpu_hotplug_callback
, 0);
5023 parent
= mem_cgroup_from_cont(cont
->parent
);
5024 memcg
->use_hierarchy
= parent
->use_hierarchy
;
5025 memcg
->oom_kill_disable
= parent
->oom_kill_disable
;
5028 if (parent
&& parent
->use_hierarchy
) {
5029 res_counter_init(&memcg
->res
, &parent
->res
);
5030 res_counter_init(&memcg
->memsw
, &parent
->memsw
);
5032 * We increment refcnt of the parent to ensure that we can
5033 * safely access it on res_counter_charge/uncharge.
5034 * This refcnt will be decremented when freeing this
5035 * mem_cgroup(see mem_cgroup_put).
5037 mem_cgroup_get(parent
);
5039 res_counter_init(&memcg
->res
, NULL
);
5040 res_counter_init(&memcg
->memsw
, NULL
);
5042 memcg
->last_scanned_node
= MAX_NUMNODES
;
5043 INIT_LIST_HEAD(&memcg
->oom_notify
);
5046 memcg
->swappiness
= mem_cgroup_swappiness(parent
);
5047 atomic_set(&memcg
->refcnt
, 1);
5048 memcg
->move_charge_at_immigrate
= 0;
5049 mutex_init(&memcg
->thresholds_lock
);
5052 __mem_cgroup_free(memcg
);
5053 return ERR_PTR(error
);
5056 static int mem_cgroup_pre_destroy(struct cgroup_subsys
*ss
,
5057 struct cgroup
*cont
)
5059 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
5061 return mem_cgroup_force_empty(memcg
, false);
5064 static void mem_cgroup_destroy(struct cgroup_subsys
*ss
,
5065 struct cgroup
*cont
)
5067 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
5069 kmem_cgroup_destroy(ss
, cont
);
5071 mem_cgroup_put(memcg
);
5074 static int mem_cgroup_populate(struct cgroup_subsys
*ss
,
5075 struct cgroup
*cont
)
5079 ret
= cgroup_add_files(cont
, ss
, mem_cgroup_files
,
5080 ARRAY_SIZE(mem_cgroup_files
));
5083 ret
= register_memsw_files(cont
, ss
);
5086 ret
= register_kmem_files(cont
, ss
);
5092 /* Handlers for move charge at task migration. */
5093 #define PRECHARGE_COUNT_AT_ONCE 256
5094 static int mem_cgroup_do_precharge(unsigned long count
)
5097 int batch_count
= PRECHARGE_COUNT_AT_ONCE
;
5098 struct mem_cgroup
*memcg
= mc
.to
;
5100 if (mem_cgroup_is_root(memcg
)) {
5101 mc
.precharge
+= count
;
5102 /* we don't need css_get for root */
5105 /* try to charge at once */
5107 struct res_counter
*dummy
;
5109 * "memcg" cannot be under rmdir() because we've already checked
5110 * by cgroup_lock_live_cgroup() that it is not removed and we
5111 * are still under the same cgroup_mutex. So we can postpone
5114 if (res_counter_charge(&memcg
->res
, PAGE_SIZE
* count
, &dummy
))
5116 if (do_swap_account
&& res_counter_charge(&memcg
->memsw
,
5117 PAGE_SIZE
* count
, &dummy
)) {
5118 res_counter_uncharge(&memcg
->res
, PAGE_SIZE
* count
);
5121 mc
.precharge
+= count
;
5125 /* fall back to one by one charge */
5127 if (signal_pending(current
)) {
5131 if (!batch_count
--) {
5132 batch_count
= PRECHARGE_COUNT_AT_ONCE
;
5135 ret
= __mem_cgroup_try_charge(NULL
,
5136 GFP_KERNEL
, 1, &memcg
, false);
5138 /* mem_cgroup_clear_mc() will do uncharge later */
5146 * is_target_pte_for_mc - check a pte whether it is valid for move charge
5147 * @vma: the vma the pte to be checked belongs
5148 * @addr: the address corresponding to the pte to be checked
5149 * @ptent: the pte to be checked
5150 * @target: the pointer the target page or swap ent will be stored(can be NULL)
5153 * 0(MC_TARGET_NONE): if the pte is not a target for move charge.
5154 * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
5155 * move charge. if @target is not NULL, the page is stored in target->page
5156 * with extra refcnt got(Callers should handle it).
5157 * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
5158 * target for charge migration. if @target is not NULL, the entry is stored
5161 * Called with pte lock held.
5168 enum mc_target_type
{
5169 MC_TARGET_NONE
, /* not used */
5174 static struct page
*mc_handle_present_pte(struct vm_area_struct
*vma
,
5175 unsigned long addr
, pte_t ptent
)
5177 struct page
*page
= vm_normal_page(vma
, addr
, ptent
);
5179 if (!page
|| !page_mapped(page
))
5181 if (PageAnon(page
)) {
5182 /* we don't move shared anon */
5183 if (!move_anon() || page_mapcount(page
) > 2)
5185 } else if (!move_file())
5186 /* we ignore mapcount for file pages */
5188 if (!get_page_unless_zero(page
))
5194 static struct page
*mc_handle_swap_pte(struct vm_area_struct
*vma
,
5195 unsigned long addr
, pte_t ptent
, swp_entry_t
*entry
)
5198 struct page
*page
= NULL
;
5199 swp_entry_t ent
= pte_to_swp_entry(ptent
);
5201 if (!move_anon() || non_swap_entry(ent
))
5203 usage_count
= mem_cgroup_count_swap_user(ent
, &page
);
5204 if (usage_count
> 1) { /* we don't move shared anon */
5209 if (do_swap_account
)
5210 entry
->val
= ent
.val
;
5215 static struct page
*mc_handle_file_pte(struct vm_area_struct
*vma
,
5216 unsigned long addr
, pte_t ptent
, swp_entry_t
*entry
)
5218 struct page
*page
= NULL
;
5219 struct inode
*inode
;
5220 struct address_space
*mapping
;
5223 if (!vma
->vm_file
) /* anonymous vma */
5228 inode
= vma
->vm_file
->f_path
.dentry
->d_inode
;
5229 mapping
= vma
->vm_file
->f_mapping
;
5230 if (pte_none(ptent
))
5231 pgoff
= linear_page_index(vma
, addr
);
5232 else /* pte_file(ptent) is true */
5233 pgoff
= pte_to_pgoff(ptent
);
5235 /* page is moved even if it's not RSS of this task(page-faulted). */
5236 page
= find_get_page(mapping
, pgoff
);
5239 /* shmem/tmpfs may report page out on swap: account for that too. */
5240 if (radix_tree_exceptional_entry(page
)) {
5241 swp_entry_t swap
= radix_to_swp_entry(page
);
5242 if (do_swap_account
)
5244 page
= find_get_page(&swapper_space
, swap
.val
);
5250 static int is_target_pte_for_mc(struct vm_area_struct
*vma
,
5251 unsigned long addr
, pte_t ptent
, union mc_target
*target
)
5253 struct page
*page
= NULL
;
5254 struct page_cgroup
*pc
;
5256 swp_entry_t ent
= { .val
= 0 };
5258 if (pte_present(ptent
))
5259 page
= mc_handle_present_pte(vma
, addr
, ptent
);
5260 else if (is_swap_pte(ptent
))
5261 page
= mc_handle_swap_pte(vma
, addr
, ptent
, &ent
);
5262 else if (pte_none(ptent
) || pte_file(ptent
))
5263 page
= mc_handle_file_pte(vma
, addr
, ptent
, &ent
);
5265 if (!page
&& !ent
.val
)
5268 pc
= lookup_page_cgroup(page
);
5270 * Do only loose check w/o page_cgroup lock.
5271 * mem_cgroup_move_account() checks the pc is valid or not under
5274 if (PageCgroupUsed(pc
) && pc
->mem_cgroup
== mc
.from
) {
5275 ret
= MC_TARGET_PAGE
;
5277 target
->page
= page
;
5279 if (!ret
|| !target
)
5282 /* There is a swap entry and a page doesn't exist or isn't charged */
5283 if (ent
.val
&& !ret
&&
5284 css_id(&mc
.from
->css
) == lookup_swap_cgroup(ent
)) {
5285 ret
= MC_TARGET_SWAP
;
5292 static int mem_cgroup_count_precharge_pte_range(pmd_t
*pmd
,
5293 unsigned long addr
, unsigned long end
,
5294 struct mm_walk
*walk
)
5296 struct vm_area_struct
*vma
= walk
->private;
5300 split_huge_page_pmd(walk
->mm
, pmd
);
5302 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
5303 for (; addr
!= end
; pte
++, addr
+= PAGE_SIZE
)
5304 if (is_target_pte_for_mc(vma
, addr
, *pte
, NULL
))
5305 mc
.precharge
++; /* increment precharge temporarily */
5306 pte_unmap_unlock(pte
- 1, ptl
);
5312 static unsigned long mem_cgroup_count_precharge(struct mm_struct
*mm
)
5314 unsigned long precharge
;
5315 struct vm_area_struct
*vma
;
5317 down_read(&mm
->mmap_sem
);
5318 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
5319 struct mm_walk mem_cgroup_count_precharge_walk
= {
5320 .pmd_entry
= mem_cgroup_count_precharge_pte_range
,
5324 if (is_vm_hugetlb_page(vma
))
5326 walk_page_range(vma
->vm_start
, vma
->vm_end
,
5327 &mem_cgroup_count_precharge_walk
);
5329 up_read(&mm
->mmap_sem
);
5331 precharge
= mc
.precharge
;
5337 static int mem_cgroup_precharge_mc(struct mm_struct
*mm
)
5339 unsigned long precharge
= mem_cgroup_count_precharge(mm
);
5341 VM_BUG_ON(mc
.moving_task
);
5342 mc
.moving_task
= current
;
5343 return mem_cgroup_do_precharge(precharge
);
5346 /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */
5347 static void __mem_cgroup_clear_mc(void)
5349 struct mem_cgroup
*from
= mc
.from
;
5350 struct mem_cgroup
*to
= mc
.to
;
5352 /* we must uncharge all the leftover precharges from mc.to */
5354 __mem_cgroup_cancel_charge(mc
.to
, mc
.precharge
);
5358 * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
5359 * we must uncharge here.
5361 if (mc
.moved_charge
) {
5362 __mem_cgroup_cancel_charge(mc
.from
, mc
.moved_charge
);
5363 mc
.moved_charge
= 0;
5365 /* we must fixup refcnts and charges */
5366 if (mc
.moved_swap
) {
5367 /* uncharge swap account from the old cgroup */
5368 if (!mem_cgroup_is_root(mc
.from
))
5369 res_counter_uncharge(&mc
.from
->memsw
,
5370 PAGE_SIZE
* mc
.moved_swap
);
5371 __mem_cgroup_put(mc
.from
, mc
.moved_swap
);
5373 if (!mem_cgroup_is_root(mc
.to
)) {
5375 * we charged both to->res and to->memsw, so we should
5378 res_counter_uncharge(&mc
.to
->res
,
5379 PAGE_SIZE
* mc
.moved_swap
);
5381 /* we've already done mem_cgroup_get(mc.to) */
5384 memcg_oom_recover(from
);
5385 memcg_oom_recover(to
);
5386 wake_up_all(&mc
.waitq
);
5389 static void mem_cgroup_clear_mc(void)
5391 struct mem_cgroup
*from
= mc
.from
;
5394 * we must clear moving_task before waking up waiters at the end of
5397 mc
.moving_task
= NULL
;
5398 __mem_cgroup_clear_mc();
5399 spin_lock(&mc
.lock
);
5402 spin_unlock(&mc
.lock
);
5403 mem_cgroup_end_move(from
);
5406 static int mem_cgroup_can_attach(struct cgroup_subsys
*ss
,
5407 struct cgroup
*cgroup
,
5408 struct cgroup_taskset
*tset
)
5410 struct task_struct
*p
= cgroup_taskset_first(tset
);
5412 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgroup
);
5414 if (memcg
->move_charge_at_immigrate
) {
5415 struct mm_struct
*mm
;
5416 struct mem_cgroup
*from
= mem_cgroup_from_task(p
);
5418 VM_BUG_ON(from
== memcg
);
5420 mm
= get_task_mm(p
);
5423 /* We move charges only when we move a owner of the mm */
5424 if (mm
->owner
== p
) {
5427 VM_BUG_ON(mc
.precharge
);
5428 VM_BUG_ON(mc
.moved_charge
);
5429 VM_BUG_ON(mc
.moved_swap
);
5430 mem_cgroup_start_move(from
);
5431 spin_lock(&mc
.lock
);
5434 spin_unlock(&mc
.lock
);
5435 /* We set mc.moving_task later */
5437 ret
= mem_cgroup_precharge_mc(mm
);
5439 mem_cgroup_clear_mc();
5446 static void mem_cgroup_cancel_attach(struct cgroup_subsys
*ss
,
5447 struct cgroup
*cgroup
,
5448 struct cgroup_taskset
*tset
)
5450 mem_cgroup_clear_mc();
5453 static int mem_cgroup_move_charge_pte_range(pmd_t
*pmd
,
5454 unsigned long addr
, unsigned long end
,
5455 struct mm_walk
*walk
)
5458 struct vm_area_struct
*vma
= walk
->private;
5462 split_huge_page_pmd(walk
->mm
, pmd
);
5464 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
5465 for (; addr
!= end
; addr
+= PAGE_SIZE
) {
5466 pte_t ptent
= *(pte
++);
5467 union mc_target target
;
5470 struct page_cgroup
*pc
;
5476 type
= is_target_pte_for_mc(vma
, addr
, ptent
, &target
);
5478 case MC_TARGET_PAGE
:
5480 if (isolate_lru_page(page
))
5482 pc
= lookup_page_cgroup(page
);
5483 if (!mem_cgroup_move_account(page
, 1, pc
,
5484 mc
.from
, mc
.to
, false)) {
5486 /* we uncharge from mc.from later. */
5489 putback_lru_page(page
);
5490 put
: /* is_target_pte_for_mc() gets the page */
5493 case MC_TARGET_SWAP
:
5495 if (!mem_cgroup_move_swap_account(ent
,
5496 mc
.from
, mc
.to
, false)) {
5498 /* we fixup refcnts and charges later. */
5506 pte_unmap_unlock(pte
- 1, ptl
);
5511 * We have consumed all precharges we got in can_attach().
5512 * We try charge one by one, but don't do any additional
5513 * charges to mc.to if we have failed in charge once in attach()
5516 ret
= mem_cgroup_do_precharge(1);
5524 static void mem_cgroup_move_charge(struct mm_struct
*mm
)
5526 struct vm_area_struct
*vma
;
5528 lru_add_drain_all();
5530 if (unlikely(!down_read_trylock(&mm
->mmap_sem
))) {
5532 * Someone who are holding the mmap_sem might be waiting in
5533 * waitq. So we cancel all extra charges, wake up all waiters,
5534 * and retry. Because we cancel precharges, we might not be able
5535 * to move enough charges, but moving charge is a best-effort
5536 * feature anyway, so it wouldn't be a big problem.
5538 __mem_cgroup_clear_mc();
5542 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
5544 struct mm_walk mem_cgroup_move_charge_walk
= {
5545 .pmd_entry
= mem_cgroup_move_charge_pte_range
,
5549 if (is_vm_hugetlb_page(vma
))
5551 ret
= walk_page_range(vma
->vm_start
, vma
->vm_end
,
5552 &mem_cgroup_move_charge_walk
);
5555 * means we have consumed all precharges and failed in
5556 * doing additional charge. Just abandon here.
5560 up_read(&mm
->mmap_sem
);
5563 static void mem_cgroup_move_task(struct cgroup_subsys
*ss
,
5564 struct cgroup
*cont
,
5565 struct cgroup_taskset
*tset
)
5567 struct task_struct
*p
= cgroup_taskset_first(tset
);
5568 struct mm_struct
*mm
= get_task_mm(p
);
5572 mem_cgroup_move_charge(mm
);
5577 mem_cgroup_clear_mc();
5579 #else /* !CONFIG_MMU */
5580 static int mem_cgroup_can_attach(struct cgroup_subsys
*ss
,
5581 struct cgroup
*cgroup
,
5582 struct cgroup_taskset
*tset
)
5586 static void mem_cgroup_cancel_attach(struct cgroup_subsys
*ss
,
5587 struct cgroup
*cgroup
,
5588 struct cgroup_taskset
*tset
)
5591 static void mem_cgroup_move_task(struct cgroup_subsys
*ss
,
5592 struct cgroup
*cont
,
5593 struct cgroup_taskset
*tset
)
5598 struct cgroup_subsys mem_cgroup_subsys
= {
5600 .subsys_id
= mem_cgroup_subsys_id
,
5601 .create
= mem_cgroup_create
,
5602 .pre_destroy
= mem_cgroup_pre_destroy
,
5603 .destroy
= mem_cgroup_destroy
,
5604 .populate
= mem_cgroup_populate
,
5605 .can_attach
= mem_cgroup_can_attach
,
5606 .cancel_attach
= mem_cgroup_cancel_attach
,
5607 .attach
= mem_cgroup_move_task
,
5612 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
5613 static int __init
enable_swap_account(char *s
)
5615 /* consider enabled if no parameter or 1 is given */
5616 if (!strcmp(s
, "1"))
5617 really_do_swap_account
= 1;
5618 else if (!strcmp(s
, "0"))
5619 really_do_swap_account
= 0;
5622 __setup("swapaccount=", enable_swap_account
);