1 /* memcontrol.c - Memory Controller
3 * Copyright IBM Corporation, 2007
4 * Author Balbir Singh <balbir@linux.vnet.ibm.com>
6 * Copyright 2007 OpenVZ SWsoft Inc
7 * Author: Pavel Emelianov <xemul@openvz.org>
10 * Copyright (C) 2009 Nokia Corporation
11 * Author: Kirill A. Shutemov
13 * This program is free software; you can redistribute it and/or modify
14 * it under the terms of the GNU General Public License as published by
15 * the Free Software Foundation; either version 2 of the License, or
16 * (at your option) any later version.
18 * This program is distributed in the hope that it will be useful,
19 * but WITHOUT ANY WARRANTY; without even the implied warranty of
20 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
21 * GNU General Public License for more details.
24 #include <linux/res_counter.h>
25 #include <linux/memcontrol.h>
26 #include <linux/cgroup.h>
28 #include <linux/hugetlb.h>
29 #include <linux/pagemap.h>
30 #include <linux/smp.h>
31 #include <linux/page-flags.h>
32 #include <linux/backing-dev.h>
33 #include <linux/bit_spinlock.h>
34 #include <linux/rcupdate.h>
35 #include <linux/limits.h>
36 #include <linux/export.h>
37 #include <linux/mutex.h>
38 #include <linux/rbtree.h>
39 #include <linux/slab.h>
40 #include <linux/swap.h>
41 #include <linux/swapops.h>
42 #include <linux/spinlock.h>
43 #include <linux/eventfd.h>
44 #include <linux/sort.h>
46 #include <linux/seq_file.h>
47 #include <linux/vmalloc.h>
48 #include <linux/mm_inline.h>
49 #include <linux/page_cgroup.h>
50 #include <linux/cpu.h>
51 #include <linux/oom.h>
54 #include <net/tcp_memcontrol.h>
56 #include <asm/uaccess.h>
58 #include <trace/events/vmscan.h>
60 struct cgroup_subsys mem_cgroup_subsys __read_mostly
;
61 #define MEM_CGROUP_RECLAIM_RETRIES 5
62 struct mem_cgroup
*root_mem_cgroup __read_mostly
;
64 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
65 /* Turned on only when memory cgroup is enabled && really_do_swap_account = 1 */
66 int do_swap_account __read_mostly
;
68 /* for remember boot option*/
69 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP_ENABLED
70 static int really_do_swap_account __initdata
= 1;
72 static int really_do_swap_account __initdata
= 0;
76 #define do_swap_account (0)
81 * Statistics for memory cgroup.
83 enum mem_cgroup_stat_index
{
85 * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss.
87 MEM_CGROUP_STAT_CACHE
, /* # of pages charged as cache */
88 MEM_CGROUP_STAT_RSS
, /* # of pages charged as anon rss */
89 MEM_CGROUP_STAT_FILE_MAPPED
, /* # of pages charged as file rss */
90 MEM_CGROUP_STAT_SWAPOUT
, /* # of pages, swapped out */
91 MEM_CGROUP_STAT_DATA
, /* end of data requires synchronization */
92 MEM_CGROUP_STAT_NSTATS
,
95 enum mem_cgroup_events_index
{
96 MEM_CGROUP_EVENTS_PGPGIN
, /* # of pages paged in */
97 MEM_CGROUP_EVENTS_PGPGOUT
, /* # of pages paged out */
98 MEM_CGROUP_EVENTS_COUNT
, /* # of pages paged in/out */
99 MEM_CGROUP_EVENTS_PGFAULT
, /* # of page-faults */
100 MEM_CGROUP_EVENTS_PGMAJFAULT
, /* # of major page-faults */
101 MEM_CGROUP_EVENTS_NSTATS
,
104 * Per memcg event counter is incremented at every pagein/pageout. With THP,
105 * it will be incremated by the number of pages. This counter is used for
106 * for trigger some periodic events. This is straightforward and better
107 * than using jiffies etc. to handle periodic memcg event.
109 enum mem_cgroup_events_target
{
110 MEM_CGROUP_TARGET_THRESH
,
111 MEM_CGROUP_TARGET_SOFTLIMIT
,
112 MEM_CGROUP_TARGET_NUMAINFO
,
115 #define THRESHOLDS_EVENTS_TARGET (128)
116 #define SOFTLIMIT_EVENTS_TARGET (1024)
117 #define NUMAINFO_EVENTS_TARGET (1024)
119 struct mem_cgroup_stat_cpu
{
120 long count
[MEM_CGROUP_STAT_NSTATS
];
121 unsigned long events
[MEM_CGROUP_EVENTS_NSTATS
];
122 unsigned long targets
[MEM_CGROUP_NTARGETS
];
125 struct mem_cgroup_reclaim_iter
{
126 /* css_id of the last scanned hierarchy member */
128 /* scan generation, increased every round-trip */
129 unsigned int generation
;
133 * per-zone information in memory controller.
135 struct mem_cgroup_per_zone
{
136 struct lruvec lruvec
;
137 unsigned long lru_size
[NR_LRU_LISTS
];
139 struct mem_cgroup_reclaim_iter reclaim_iter
[DEF_PRIORITY
+ 1];
141 struct rb_node tree_node
; /* RB tree node */
142 unsigned long long usage_in_excess
;/* Set to the value by which */
143 /* the soft limit is exceeded*/
145 struct mem_cgroup
*memcg
; /* Back pointer, we cannot */
146 /* use container_of */
149 struct mem_cgroup_per_node
{
150 struct mem_cgroup_per_zone zoneinfo
[MAX_NR_ZONES
];
153 struct mem_cgroup_lru_info
{
154 struct mem_cgroup_per_node
*nodeinfo
[MAX_NUMNODES
];
158 * Cgroups above their limits are maintained in a RB-Tree, independent of
159 * their hierarchy representation
162 struct mem_cgroup_tree_per_zone
{
163 struct rb_root rb_root
;
167 struct mem_cgroup_tree_per_node
{
168 struct mem_cgroup_tree_per_zone rb_tree_per_zone
[MAX_NR_ZONES
];
171 struct mem_cgroup_tree
{
172 struct mem_cgroup_tree_per_node
*rb_tree_per_node
[MAX_NUMNODES
];
175 static struct mem_cgroup_tree soft_limit_tree __read_mostly
;
177 struct mem_cgroup_threshold
{
178 struct eventfd_ctx
*eventfd
;
183 struct mem_cgroup_threshold_ary
{
184 /* An array index points to threshold just below usage. */
185 int current_threshold
;
186 /* Size of entries[] */
188 /* Array of thresholds */
189 struct mem_cgroup_threshold entries
[0];
192 struct mem_cgroup_thresholds
{
193 /* Primary thresholds array */
194 struct mem_cgroup_threshold_ary
*primary
;
196 * Spare threshold array.
197 * This is needed to make mem_cgroup_unregister_event() "never fail".
198 * It must be able to store at least primary->size - 1 entries.
200 struct mem_cgroup_threshold_ary
*spare
;
204 struct mem_cgroup_eventfd_list
{
205 struct list_head list
;
206 struct eventfd_ctx
*eventfd
;
209 static void mem_cgroup_threshold(struct mem_cgroup
*memcg
);
210 static void mem_cgroup_oom_notify(struct mem_cgroup
*memcg
);
213 * The memory controller data structure. The memory controller controls both
214 * page cache and RSS per cgroup. We would eventually like to provide
215 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
216 * to help the administrator determine what knobs to tune.
218 * TODO: Add a water mark for the memory controller. Reclaim will begin when
219 * we hit the water mark. May be even add a low water mark, such that
220 * no reclaim occurs from a cgroup at it's low water mark, this is
221 * a feature that will be implemented much later in the future.
224 struct cgroup_subsys_state css
;
226 * the counter to account for memory usage
228 struct res_counter res
;
232 * the counter to account for mem+swap usage.
234 struct res_counter memsw
;
237 * rcu_freeing is used only when freeing struct mem_cgroup,
238 * so put it into a union to avoid wasting more memory.
239 * It must be disjoint from the css field. It could be
240 * in a union with the res field, but res plays a much
241 * larger part in mem_cgroup life than memsw, and might
242 * be of interest, even at time of free, when debugging.
243 * So share rcu_head with the less interesting memsw.
245 struct rcu_head rcu_freeing
;
247 * But when using vfree(), that cannot be done at
248 * interrupt time, so we must then queue the work.
250 struct work_struct work_freeing
;
254 * Per cgroup active and inactive list, similar to the
255 * per zone LRU lists.
257 struct mem_cgroup_lru_info info
;
258 int last_scanned_node
;
260 nodemask_t scan_nodes
;
261 atomic_t numainfo_events
;
262 atomic_t numainfo_updating
;
265 * Should the accounting and control be hierarchical, per subtree?
275 /* OOM-Killer disable */
276 int oom_kill_disable
;
278 /* set when res.limit == memsw.limit */
279 bool memsw_is_minimum
;
281 /* protect arrays of thresholds */
282 struct mutex thresholds_lock
;
284 /* thresholds for memory usage. RCU-protected */
285 struct mem_cgroup_thresholds thresholds
;
287 /* thresholds for mem+swap usage. RCU-protected */
288 struct mem_cgroup_thresholds memsw_thresholds
;
290 /* For oom notifier event fd */
291 struct list_head oom_notify
;
294 * Should we move charges of a task when a task is moved into this
295 * mem_cgroup ? And what type of charges should we move ?
297 unsigned long move_charge_at_immigrate
;
299 * set > 0 if pages under this cgroup are moving to other cgroup.
301 atomic_t moving_account
;
302 /* taken only while moving_account > 0 */
303 spinlock_t move_lock
;
307 struct mem_cgroup_stat_cpu
*stat
;
309 * used when a cpu is offlined or other synchronizations
310 * See mem_cgroup_read_stat().
312 struct mem_cgroup_stat_cpu nocpu_base
;
313 spinlock_t pcp_counter_lock
;
316 struct tcp_memcontrol tcp_mem
;
320 /* Stuffs for move charges at task migration. */
322 * Types of charges to be moved. "move_charge_at_immitgrate" is treated as a
323 * left-shifted bitmap of these types.
326 MOVE_CHARGE_TYPE_ANON
, /* private anonymous page and swap of it */
327 MOVE_CHARGE_TYPE_FILE
, /* file page(including tmpfs) and swap of it */
331 /* "mc" and its members are protected by cgroup_mutex */
332 static struct move_charge_struct
{
333 spinlock_t lock
; /* for from, to */
334 struct mem_cgroup
*from
;
335 struct mem_cgroup
*to
;
336 unsigned long precharge
;
337 unsigned long moved_charge
;
338 unsigned long moved_swap
;
339 struct task_struct
*moving_task
; /* a task moving charges */
340 wait_queue_head_t waitq
; /* a waitq for other context */
342 .lock
= __SPIN_LOCK_UNLOCKED(mc
.lock
),
343 .waitq
= __WAIT_QUEUE_HEAD_INITIALIZER(mc
.waitq
),
346 static bool move_anon(void)
348 return test_bit(MOVE_CHARGE_TYPE_ANON
,
349 &mc
.to
->move_charge_at_immigrate
);
352 static bool move_file(void)
354 return test_bit(MOVE_CHARGE_TYPE_FILE
,
355 &mc
.to
->move_charge_at_immigrate
);
359 * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft
360 * limit reclaim to prevent infinite loops, if they ever occur.
362 #define MEM_CGROUP_MAX_RECLAIM_LOOPS (100)
363 #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS (2)
366 MEM_CGROUP_CHARGE_TYPE_CACHE
= 0,
367 MEM_CGROUP_CHARGE_TYPE_MAPPED
,
368 MEM_CGROUP_CHARGE_TYPE_SHMEM
, /* used by page migration of shmem */
369 MEM_CGROUP_CHARGE_TYPE_FORCE
, /* used by force_empty */
370 MEM_CGROUP_CHARGE_TYPE_SWAPOUT
, /* for accounting swapcache */
371 MEM_CGROUP_CHARGE_TYPE_DROP
, /* a page was unused swap cache */
375 /* for encoding cft->private value on file */
378 #define _OOM_TYPE (2)
379 #define MEMFILE_PRIVATE(x, val) (((x) << 16) | (val))
380 #define MEMFILE_TYPE(val) (((val) >> 16) & 0xffff)
381 #define MEMFILE_ATTR(val) ((val) & 0xffff)
382 /* Used for OOM nofiier */
383 #define OOM_CONTROL (0)
386 * Reclaim flags for mem_cgroup_hierarchical_reclaim
388 #define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0
389 #define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT)
390 #define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1
391 #define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT)
393 static void mem_cgroup_get(struct mem_cgroup
*memcg
);
394 static void mem_cgroup_put(struct mem_cgroup
*memcg
);
396 /* Writing them here to avoid exposing memcg's inner layout */
397 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_KMEM
398 #include <net/sock.h>
401 static bool mem_cgroup_is_root(struct mem_cgroup
*memcg
);
402 void sock_update_memcg(struct sock
*sk
)
404 if (mem_cgroup_sockets_enabled
) {
405 struct mem_cgroup
*memcg
;
407 BUG_ON(!sk
->sk_prot
->proto_cgroup
);
409 /* Socket cloning can throw us here with sk_cgrp already
410 * filled. It won't however, necessarily happen from
411 * process context. So the test for root memcg given
412 * the current task's memcg won't help us in this case.
414 * Respecting the original socket's memcg is a better
415 * decision in this case.
418 BUG_ON(mem_cgroup_is_root(sk
->sk_cgrp
->memcg
));
419 mem_cgroup_get(sk
->sk_cgrp
->memcg
);
424 memcg
= mem_cgroup_from_task(current
);
425 if (!mem_cgroup_is_root(memcg
)) {
426 mem_cgroup_get(memcg
);
427 sk
->sk_cgrp
= sk
->sk_prot
->proto_cgroup(memcg
);
432 EXPORT_SYMBOL(sock_update_memcg
);
434 void sock_release_memcg(struct sock
*sk
)
436 if (mem_cgroup_sockets_enabled
&& sk
->sk_cgrp
) {
437 struct mem_cgroup
*memcg
;
438 WARN_ON(!sk
->sk_cgrp
->memcg
);
439 memcg
= sk
->sk_cgrp
->memcg
;
440 mem_cgroup_put(memcg
);
445 struct cg_proto
*tcp_proto_cgroup(struct mem_cgroup
*memcg
)
447 if (!memcg
|| mem_cgroup_is_root(memcg
))
450 return &memcg
->tcp_mem
.cg_proto
;
452 EXPORT_SYMBOL(tcp_proto_cgroup
);
453 #endif /* CONFIG_INET */
454 #endif /* CONFIG_CGROUP_MEM_RES_CTLR_KMEM */
456 static void drain_all_stock_async(struct mem_cgroup
*memcg
);
458 static struct mem_cgroup_per_zone
*
459 mem_cgroup_zoneinfo(struct mem_cgroup
*memcg
, int nid
, int zid
)
461 return &memcg
->info
.nodeinfo
[nid
]->zoneinfo
[zid
];
464 struct cgroup_subsys_state
*mem_cgroup_css(struct mem_cgroup
*memcg
)
469 static struct mem_cgroup_per_zone
*
470 page_cgroup_zoneinfo(struct mem_cgroup
*memcg
, struct page
*page
)
472 int nid
= page_to_nid(page
);
473 int zid
= page_zonenum(page
);
475 return mem_cgroup_zoneinfo(memcg
, nid
, zid
);
478 static struct mem_cgroup_tree_per_zone
*
479 soft_limit_tree_node_zone(int nid
, int zid
)
481 return &soft_limit_tree
.rb_tree_per_node
[nid
]->rb_tree_per_zone
[zid
];
484 static struct mem_cgroup_tree_per_zone
*
485 soft_limit_tree_from_page(struct page
*page
)
487 int nid
= page_to_nid(page
);
488 int zid
= page_zonenum(page
);
490 return &soft_limit_tree
.rb_tree_per_node
[nid
]->rb_tree_per_zone
[zid
];
494 __mem_cgroup_insert_exceeded(struct mem_cgroup
*memcg
,
495 struct mem_cgroup_per_zone
*mz
,
496 struct mem_cgroup_tree_per_zone
*mctz
,
497 unsigned long long new_usage_in_excess
)
499 struct rb_node
**p
= &mctz
->rb_root
.rb_node
;
500 struct rb_node
*parent
= NULL
;
501 struct mem_cgroup_per_zone
*mz_node
;
506 mz
->usage_in_excess
= new_usage_in_excess
;
507 if (!mz
->usage_in_excess
)
511 mz_node
= rb_entry(parent
, struct mem_cgroup_per_zone
,
513 if (mz
->usage_in_excess
< mz_node
->usage_in_excess
)
516 * We can't avoid mem cgroups that are over their soft
517 * limit by the same amount
519 else if (mz
->usage_in_excess
>= mz_node
->usage_in_excess
)
522 rb_link_node(&mz
->tree_node
, parent
, p
);
523 rb_insert_color(&mz
->tree_node
, &mctz
->rb_root
);
528 __mem_cgroup_remove_exceeded(struct mem_cgroup
*memcg
,
529 struct mem_cgroup_per_zone
*mz
,
530 struct mem_cgroup_tree_per_zone
*mctz
)
534 rb_erase(&mz
->tree_node
, &mctz
->rb_root
);
539 mem_cgroup_remove_exceeded(struct mem_cgroup
*memcg
,
540 struct mem_cgroup_per_zone
*mz
,
541 struct mem_cgroup_tree_per_zone
*mctz
)
543 spin_lock(&mctz
->lock
);
544 __mem_cgroup_remove_exceeded(memcg
, mz
, mctz
);
545 spin_unlock(&mctz
->lock
);
549 static void mem_cgroup_update_tree(struct mem_cgroup
*memcg
, struct page
*page
)
551 unsigned long long excess
;
552 struct mem_cgroup_per_zone
*mz
;
553 struct mem_cgroup_tree_per_zone
*mctz
;
554 int nid
= page_to_nid(page
);
555 int zid
= page_zonenum(page
);
556 mctz
= soft_limit_tree_from_page(page
);
559 * Necessary to update all ancestors when hierarchy is used.
560 * because their event counter is not touched.
562 for (; memcg
; memcg
= parent_mem_cgroup(memcg
)) {
563 mz
= mem_cgroup_zoneinfo(memcg
, nid
, zid
);
564 excess
= res_counter_soft_limit_excess(&memcg
->res
);
566 * We have to update the tree if mz is on RB-tree or
567 * mem is over its softlimit.
569 if (excess
|| mz
->on_tree
) {
570 spin_lock(&mctz
->lock
);
571 /* if on-tree, remove it */
573 __mem_cgroup_remove_exceeded(memcg
, mz
, mctz
);
575 * Insert again. mz->usage_in_excess will be updated.
576 * If excess is 0, no tree ops.
578 __mem_cgroup_insert_exceeded(memcg
, mz
, mctz
, excess
);
579 spin_unlock(&mctz
->lock
);
584 static void mem_cgroup_remove_from_trees(struct mem_cgroup
*memcg
)
587 struct mem_cgroup_per_zone
*mz
;
588 struct mem_cgroup_tree_per_zone
*mctz
;
590 for_each_node(node
) {
591 for (zone
= 0; zone
< MAX_NR_ZONES
; zone
++) {
592 mz
= mem_cgroup_zoneinfo(memcg
, node
, zone
);
593 mctz
= soft_limit_tree_node_zone(node
, zone
);
594 mem_cgroup_remove_exceeded(memcg
, mz
, mctz
);
599 static struct mem_cgroup_per_zone
*
600 __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone
*mctz
)
602 struct rb_node
*rightmost
= NULL
;
603 struct mem_cgroup_per_zone
*mz
;
607 rightmost
= rb_last(&mctz
->rb_root
);
609 goto done
; /* Nothing to reclaim from */
611 mz
= rb_entry(rightmost
, struct mem_cgroup_per_zone
, tree_node
);
613 * Remove the node now but someone else can add it back,
614 * we will to add it back at the end of reclaim to its correct
615 * position in the tree.
617 __mem_cgroup_remove_exceeded(mz
->memcg
, mz
, mctz
);
618 if (!res_counter_soft_limit_excess(&mz
->memcg
->res
) ||
619 !css_tryget(&mz
->memcg
->css
))
625 static struct mem_cgroup_per_zone
*
626 mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone
*mctz
)
628 struct mem_cgroup_per_zone
*mz
;
630 spin_lock(&mctz
->lock
);
631 mz
= __mem_cgroup_largest_soft_limit_node(mctz
);
632 spin_unlock(&mctz
->lock
);
637 * Implementation Note: reading percpu statistics for memcg.
639 * Both of vmstat[] and percpu_counter has threshold and do periodic
640 * synchronization to implement "quick" read. There are trade-off between
641 * reading cost and precision of value. Then, we may have a chance to implement
642 * a periodic synchronizion of counter in memcg's counter.
644 * But this _read() function is used for user interface now. The user accounts
645 * memory usage by memory cgroup and he _always_ requires exact value because
646 * he accounts memory. Even if we provide quick-and-fuzzy read, we always
647 * have to visit all online cpus and make sum. So, for now, unnecessary
648 * synchronization is not implemented. (just implemented for cpu hotplug)
650 * If there are kernel internal actions which can make use of some not-exact
651 * value, and reading all cpu value can be performance bottleneck in some
652 * common workload, threashold and synchonization as vmstat[] should be
655 static long mem_cgroup_read_stat(struct mem_cgroup
*memcg
,
656 enum mem_cgroup_stat_index idx
)
662 for_each_online_cpu(cpu
)
663 val
+= per_cpu(memcg
->stat
->count
[idx
], cpu
);
664 #ifdef CONFIG_HOTPLUG_CPU
665 spin_lock(&memcg
->pcp_counter_lock
);
666 val
+= memcg
->nocpu_base
.count
[idx
];
667 spin_unlock(&memcg
->pcp_counter_lock
);
673 static void mem_cgroup_swap_statistics(struct mem_cgroup
*memcg
,
676 int val
= (charge
) ? 1 : -1;
677 this_cpu_add(memcg
->stat
->count
[MEM_CGROUP_STAT_SWAPOUT
], val
);
680 static unsigned long mem_cgroup_read_events(struct mem_cgroup
*memcg
,
681 enum mem_cgroup_events_index idx
)
683 unsigned long val
= 0;
686 for_each_online_cpu(cpu
)
687 val
+= per_cpu(memcg
->stat
->events
[idx
], cpu
);
688 #ifdef CONFIG_HOTPLUG_CPU
689 spin_lock(&memcg
->pcp_counter_lock
);
690 val
+= memcg
->nocpu_base
.events
[idx
];
691 spin_unlock(&memcg
->pcp_counter_lock
);
696 static void mem_cgroup_charge_statistics(struct mem_cgroup
*memcg
,
697 bool anon
, int nr_pages
)
702 * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is
703 * counted as CACHE even if it's on ANON LRU.
706 __this_cpu_add(memcg
->stat
->count
[MEM_CGROUP_STAT_RSS
],
709 __this_cpu_add(memcg
->stat
->count
[MEM_CGROUP_STAT_CACHE
],
712 /* pagein of a big page is an event. So, ignore page size */
714 __this_cpu_inc(memcg
->stat
->events
[MEM_CGROUP_EVENTS_PGPGIN
]);
716 __this_cpu_inc(memcg
->stat
->events
[MEM_CGROUP_EVENTS_PGPGOUT
]);
717 nr_pages
= -nr_pages
; /* for event */
720 __this_cpu_add(memcg
->stat
->events
[MEM_CGROUP_EVENTS_COUNT
], nr_pages
);
726 mem_cgroup_zone_nr_lru_pages(struct mem_cgroup
*memcg
, int nid
, int zid
,
727 unsigned int lru_mask
)
729 struct mem_cgroup_per_zone
*mz
;
731 unsigned long ret
= 0;
733 mz
= mem_cgroup_zoneinfo(memcg
, nid
, zid
);
736 if (BIT(lru
) & lru_mask
)
737 ret
+= mz
->lru_size
[lru
];
743 mem_cgroup_node_nr_lru_pages(struct mem_cgroup
*memcg
,
744 int nid
, unsigned int lru_mask
)
749 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++)
750 total
+= mem_cgroup_zone_nr_lru_pages(memcg
,
756 static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup
*memcg
,
757 unsigned int lru_mask
)
762 for_each_node_state(nid
, N_HIGH_MEMORY
)
763 total
+= mem_cgroup_node_nr_lru_pages(memcg
, nid
, lru_mask
);
767 static bool mem_cgroup_event_ratelimit(struct mem_cgroup
*memcg
,
768 enum mem_cgroup_events_target target
)
770 unsigned long val
, next
;
772 val
= __this_cpu_read(memcg
->stat
->events
[MEM_CGROUP_EVENTS_COUNT
]);
773 next
= __this_cpu_read(memcg
->stat
->targets
[target
]);
774 /* from time_after() in jiffies.h */
775 if ((long)next
- (long)val
< 0) {
777 case MEM_CGROUP_TARGET_THRESH
:
778 next
= val
+ THRESHOLDS_EVENTS_TARGET
;
780 case MEM_CGROUP_TARGET_SOFTLIMIT
:
781 next
= val
+ SOFTLIMIT_EVENTS_TARGET
;
783 case MEM_CGROUP_TARGET_NUMAINFO
:
784 next
= val
+ NUMAINFO_EVENTS_TARGET
;
789 __this_cpu_write(memcg
->stat
->targets
[target
], next
);
796 * Check events in order.
799 static void memcg_check_events(struct mem_cgroup
*memcg
, struct page
*page
)
802 /* threshold event is triggered in finer grain than soft limit */
803 if (unlikely(mem_cgroup_event_ratelimit(memcg
,
804 MEM_CGROUP_TARGET_THRESH
))) {
806 bool do_numainfo __maybe_unused
;
808 do_softlimit
= mem_cgroup_event_ratelimit(memcg
,
809 MEM_CGROUP_TARGET_SOFTLIMIT
);
811 do_numainfo
= mem_cgroup_event_ratelimit(memcg
,
812 MEM_CGROUP_TARGET_NUMAINFO
);
816 mem_cgroup_threshold(memcg
);
817 if (unlikely(do_softlimit
))
818 mem_cgroup_update_tree(memcg
, page
);
820 if (unlikely(do_numainfo
))
821 atomic_inc(&memcg
->numainfo_events
);
827 struct mem_cgroup
*mem_cgroup_from_cont(struct cgroup
*cont
)
829 return container_of(cgroup_subsys_state(cont
,
830 mem_cgroup_subsys_id
), struct mem_cgroup
,
834 struct mem_cgroup
*mem_cgroup_from_task(struct task_struct
*p
)
837 * mm_update_next_owner() may clear mm->owner to NULL
838 * if it races with swapoff, page migration, etc.
839 * So this can be called with p == NULL.
844 return container_of(task_subsys_state(p
, mem_cgroup_subsys_id
),
845 struct mem_cgroup
, css
);
848 struct mem_cgroup
*try_get_mem_cgroup_from_mm(struct mm_struct
*mm
)
850 struct mem_cgroup
*memcg
= NULL
;
855 * Because we have no locks, mm->owner's may be being moved to other
856 * cgroup. We use css_tryget() here even if this looks
857 * pessimistic (rather than adding locks here).
861 memcg
= mem_cgroup_from_task(rcu_dereference(mm
->owner
));
862 if (unlikely(!memcg
))
864 } while (!css_tryget(&memcg
->css
));
870 * mem_cgroup_iter - iterate over memory cgroup hierarchy
871 * @root: hierarchy root
872 * @prev: previously returned memcg, NULL on first invocation
873 * @reclaim: cookie for shared reclaim walks, NULL for full walks
875 * Returns references to children of the hierarchy below @root, or
876 * @root itself, or %NULL after a full round-trip.
878 * Caller must pass the return value in @prev on subsequent
879 * invocations for reference counting, or use mem_cgroup_iter_break()
880 * to cancel a hierarchy walk before the round-trip is complete.
882 * Reclaimers can specify a zone and a priority level in @reclaim to
883 * divide up the memcgs in the hierarchy among all concurrent
884 * reclaimers operating on the same zone and priority.
886 struct mem_cgroup
*mem_cgroup_iter(struct mem_cgroup
*root
,
887 struct mem_cgroup
*prev
,
888 struct mem_cgroup_reclaim_cookie
*reclaim
)
890 struct mem_cgroup
*memcg
= NULL
;
893 if (mem_cgroup_disabled())
897 root
= root_mem_cgroup
;
899 if (prev
&& !reclaim
)
900 id
= css_id(&prev
->css
);
902 if (prev
&& prev
!= root
)
905 if (!root
->use_hierarchy
&& root
!= root_mem_cgroup
) {
912 struct mem_cgroup_reclaim_iter
*uninitialized_var(iter
);
913 struct cgroup_subsys_state
*css
;
916 int nid
= zone_to_nid(reclaim
->zone
);
917 int zid
= zone_idx(reclaim
->zone
);
918 struct mem_cgroup_per_zone
*mz
;
920 mz
= mem_cgroup_zoneinfo(root
, nid
, zid
);
921 iter
= &mz
->reclaim_iter
[reclaim
->priority
];
922 if (prev
&& reclaim
->generation
!= iter
->generation
)
928 css
= css_get_next(&mem_cgroup_subsys
, id
+ 1, &root
->css
, &id
);
930 if (css
== &root
->css
|| css_tryget(css
))
931 memcg
= container_of(css
,
932 struct mem_cgroup
, css
);
941 else if (!prev
&& memcg
)
942 reclaim
->generation
= iter
->generation
;
952 * mem_cgroup_iter_break - abort a hierarchy walk prematurely
953 * @root: hierarchy root
954 * @prev: last visited hierarchy member as returned by mem_cgroup_iter()
956 void mem_cgroup_iter_break(struct mem_cgroup
*root
,
957 struct mem_cgroup
*prev
)
960 root
= root_mem_cgroup
;
961 if (prev
&& prev
!= root
)
966 * Iteration constructs for visiting all cgroups (under a tree). If
967 * loops are exited prematurely (break), mem_cgroup_iter_break() must
968 * be used for reference counting.
970 #define for_each_mem_cgroup_tree(iter, root) \
971 for (iter = mem_cgroup_iter(root, NULL, NULL); \
973 iter = mem_cgroup_iter(root, iter, NULL))
975 #define for_each_mem_cgroup(iter) \
976 for (iter = mem_cgroup_iter(NULL, NULL, NULL); \
978 iter = mem_cgroup_iter(NULL, iter, NULL))
980 static inline bool mem_cgroup_is_root(struct mem_cgroup
*memcg
)
982 return (memcg
== root_mem_cgroup
);
985 void mem_cgroup_count_vm_event(struct mm_struct
*mm
, enum vm_event_item idx
)
987 struct mem_cgroup
*memcg
;
993 memcg
= mem_cgroup_from_task(rcu_dereference(mm
->owner
));
994 if (unlikely(!memcg
))
999 this_cpu_inc(memcg
->stat
->events
[MEM_CGROUP_EVENTS_PGFAULT
]);
1002 this_cpu_inc(memcg
->stat
->events
[MEM_CGROUP_EVENTS_PGMAJFAULT
]);
1010 EXPORT_SYMBOL(mem_cgroup_count_vm_event
);
1013 * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg
1014 * @zone: zone of the wanted lruvec
1015 * @mem: memcg of the wanted lruvec
1017 * Returns the lru list vector holding pages for the given @zone and
1018 * @mem. This can be the global zone lruvec, if the memory controller
1021 struct lruvec
*mem_cgroup_zone_lruvec(struct zone
*zone
,
1022 struct mem_cgroup
*memcg
)
1024 struct mem_cgroup_per_zone
*mz
;
1026 if (mem_cgroup_disabled())
1027 return &zone
->lruvec
;
1029 mz
= mem_cgroup_zoneinfo(memcg
, zone_to_nid(zone
), zone_idx(zone
));
1034 * Following LRU functions are allowed to be used without PCG_LOCK.
1035 * Operations are called by routine of global LRU independently from memcg.
1036 * What we have to take care of here is validness of pc->mem_cgroup.
1038 * Changes to pc->mem_cgroup happens when
1041 * In typical case, "charge" is done before add-to-lru. Exception is SwapCache.
1042 * It is added to LRU before charge.
1043 * If PCG_USED bit is not set, page_cgroup is not added to this private LRU.
1044 * When moving account, the page is not on LRU. It's isolated.
1048 * mem_cgroup_lru_add_list - account for adding an lru page and return lruvec
1049 * @zone: zone of the page
1053 * This function accounts for @page being added to @lru, and returns
1054 * the lruvec for the given @zone and the memcg @page is charged to.
1056 * The callsite is then responsible for physically linking the page to
1057 * the returned lruvec->lists[@lru].
1059 struct lruvec
*mem_cgroup_lru_add_list(struct zone
*zone
, struct page
*page
,
1062 struct mem_cgroup_per_zone
*mz
;
1063 struct mem_cgroup
*memcg
;
1064 struct page_cgroup
*pc
;
1066 if (mem_cgroup_disabled())
1067 return &zone
->lruvec
;
1069 pc
= lookup_page_cgroup(page
);
1070 memcg
= pc
->mem_cgroup
;
1073 * Surreptitiously switch any uncharged page to root:
1074 * an uncharged page off lru does nothing to secure
1075 * its former mem_cgroup from sudden removal.
1077 * Our caller holds lru_lock, and PageCgroupUsed is updated
1078 * under page_cgroup lock: between them, they make all uses
1079 * of pc->mem_cgroup safe.
1081 if (!PageCgroupUsed(pc
) && memcg
!= root_mem_cgroup
)
1082 pc
->mem_cgroup
= memcg
= root_mem_cgroup
;
1084 mz
= page_cgroup_zoneinfo(memcg
, page
);
1085 /* compound_order() is stabilized through lru_lock */
1086 mz
->lru_size
[lru
] += 1 << compound_order(page
);
1091 * mem_cgroup_lru_del_list - account for removing an lru page
1095 * This function accounts for @page being removed from @lru.
1097 * The callsite is then responsible for physically unlinking
1100 void mem_cgroup_lru_del_list(struct page
*page
, enum lru_list lru
)
1102 struct mem_cgroup_per_zone
*mz
;
1103 struct mem_cgroup
*memcg
;
1104 struct page_cgroup
*pc
;
1106 if (mem_cgroup_disabled())
1109 pc
= lookup_page_cgroup(page
);
1110 memcg
= pc
->mem_cgroup
;
1112 mz
= page_cgroup_zoneinfo(memcg
, page
);
1113 /* huge page split is done under lru_lock. so, we have no races. */
1114 VM_BUG_ON(mz
->lru_size
[lru
] < (1 << compound_order(page
)));
1115 mz
->lru_size
[lru
] -= 1 << compound_order(page
);
1118 void mem_cgroup_lru_del(struct page
*page
)
1120 mem_cgroup_lru_del_list(page
, page_lru(page
));
1124 * mem_cgroup_lru_move_lists - account for moving a page between lrus
1125 * @zone: zone of the page
1127 * @from: current lru
1130 * This function accounts for @page being moved between the lrus @from
1131 * and @to, and returns the lruvec for the given @zone and the memcg
1132 * @page is charged to.
1134 * The callsite is then responsible for physically relinking
1135 * @page->lru to the returned lruvec->lists[@to].
1137 struct lruvec
*mem_cgroup_lru_move_lists(struct zone
*zone
,
1142 /* XXX: Optimize this, especially for @from == @to */
1143 mem_cgroup_lru_del_list(page
, from
);
1144 return mem_cgroup_lru_add_list(zone
, page
, to
);
1148 * Checks whether given mem is same or in the root_mem_cgroup's
1151 bool __mem_cgroup_same_or_subtree(const struct mem_cgroup
*root_memcg
,
1152 struct mem_cgroup
*memcg
)
1154 if (root_memcg
== memcg
)
1156 if (!root_memcg
->use_hierarchy
)
1158 return css_is_ancestor(&memcg
->css
, &root_memcg
->css
);
1161 static bool mem_cgroup_same_or_subtree(const struct mem_cgroup
*root_memcg
,
1162 struct mem_cgroup
*memcg
)
1167 ret
= __mem_cgroup_same_or_subtree(root_memcg
, memcg
);
1172 int task_in_mem_cgroup(struct task_struct
*task
, const struct mem_cgroup
*memcg
)
1175 struct mem_cgroup
*curr
= NULL
;
1176 struct task_struct
*p
;
1178 p
= find_lock_task_mm(task
);
1180 curr
= try_get_mem_cgroup_from_mm(p
->mm
);
1184 * All threads may have already detached their mm's, but the oom
1185 * killer still needs to detect if they have already been oom
1186 * killed to prevent needlessly killing additional tasks.
1189 curr
= mem_cgroup_from_task(task
);
1191 css_get(&curr
->css
);
1197 * We should check use_hierarchy of "memcg" not "curr". Because checking
1198 * use_hierarchy of "curr" here make this function true if hierarchy is
1199 * enabled in "curr" and "curr" is a child of "memcg" in *cgroup*
1200 * hierarchy(even if use_hierarchy is disabled in "memcg").
1202 ret
= mem_cgroup_same_or_subtree(memcg
, curr
);
1203 css_put(&curr
->css
);
1207 int mem_cgroup_inactive_anon_is_low(struct mem_cgroup
*memcg
, struct zone
*zone
)
1209 unsigned long inactive_ratio
;
1210 int nid
= zone_to_nid(zone
);
1211 int zid
= zone_idx(zone
);
1212 unsigned long inactive
;
1213 unsigned long active
;
1216 inactive
= mem_cgroup_zone_nr_lru_pages(memcg
, nid
, zid
,
1217 BIT(LRU_INACTIVE_ANON
));
1218 active
= mem_cgroup_zone_nr_lru_pages(memcg
, nid
, zid
,
1219 BIT(LRU_ACTIVE_ANON
));
1221 gb
= (inactive
+ active
) >> (30 - PAGE_SHIFT
);
1223 inactive_ratio
= int_sqrt(10 * gb
);
1227 return inactive
* inactive_ratio
< active
;
1230 int mem_cgroup_inactive_file_is_low(struct mem_cgroup
*memcg
, struct zone
*zone
)
1232 unsigned long active
;
1233 unsigned long inactive
;
1234 int zid
= zone_idx(zone
);
1235 int nid
= zone_to_nid(zone
);
1237 inactive
= mem_cgroup_zone_nr_lru_pages(memcg
, nid
, zid
,
1238 BIT(LRU_INACTIVE_FILE
));
1239 active
= mem_cgroup_zone_nr_lru_pages(memcg
, nid
, zid
,
1240 BIT(LRU_ACTIVE_FILE
));
1242 return (active
> inactive
);
1245 struct zone_reclaim_stat
*
1246 mem_cgroup_get_reclaim_stat_from_page(struct page
*page
)
1248 struct page_cgroup
*pc
;
1249 struct mem_cgroup_per_zone
*mz
;
1251 if (mem_cgroup_disabled())
1254 pc
= lookup_page_cgroup(page
);
1255 if (!PageCgroupUsed(pc
))
1257 /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */
1259 mz
= page_cgroup_zoneinfo(pc
->mem_cgroup
, page
);
1260 return &mz
->lruvec
.reclaim_stat
;
1263 #define mem_cgroup_from_res_counter(counter, member) \
1264 container_of(counter, struct mem_cgroup, member)
1267 * mem_cgroup_margin - calculate chargeable space of a memory cgroup
1268 * @mem: the memory cgroup
1270 * Returns the maximum amount of memory @mem can be charged with, in
1273 static unsigned long mem_cgroup_margin(struct mem_cgroup
*memcg
)
1275 unsigned long long margin
;
1277 margin
= res_counter_margin(&memcg
->res
);
1278 if (do_swap_account
)
1279 margin
= min(margin
, res_counter_margin(&memcg
->memsw
));
1280 return margin
>> PAGE_SHIFT
;
1283 int mem_cgroup_swappiness(struct mem_cgroup
*memcg
)
1285 struct cgroup
*cgrp
= memcg
->css
.cgroup
;
1288 if (cgrp
->parent
== NULL
)
1289 return vm_swappiness
;
1291 return memcg
->swappiness
;
1295 * memcg->moving_account is used for checking possibility that some thread is
1296 * calling move_account(). When a thread on CPU-A starts moving pages under
1297 * a memcg, other threads should check memcg->moving_account under
1298 * rcu_read_lock(), like this:
1302 * memcg->moving_account+1 if (memcg->mocing_account)
1304 * synchronize_rcu() update something.
1309 /* for quick checking without looking up memcg */
1310 atomic_t memcg_moving __read_mostly
;
1312 static void mem_cgroup_start_move(struct mem_cgroup
*memcg
)
1314 atomic_inc(&memcg_moving
);
1315 atomic_inc(&memcg
->moving_account
);
1319 static void mem_cgroup_end_move(struct mem_cgroup
*memcg
)
1322 * Now, mem_cgroup_clear_mc() may call this function with NULL.
1323 * We check NULL in callee rather than caller.
1326 atomic_dec(&memcg_moving
);
1327 atomic_dec(&memcg
->moving_account
);
1332 * 2 routines for checking "mem" is under move_account() or not.
1334 * mem_cgroup_stolen() - checking whether a cgroup is mc.from or not. This
1335 * is used for avoiding races in accounting. If true,
1336 * pc->mem_cgroup may be overwritten.
1338 * mem_cgroup_under_move() - checking a cgroup is mc.from or mc.to or
1339 * under hierarchy of moving cgroups. This is for
1340 * waiting at hith-memory prressure caused by "move".
1343 static bool mem_cgroup_stolen(struct mem_cgroup
*memcg
)
1345 VM_BUG_ON(!rcu_read_lock_held());
1346 return atomic_read(&memcg
->moving_account
) > 0;
1349 static bool mem_cgroup_under_move(struct mem_cgroup
*memcg
)
1351 struct mem_cgroup
*from
;
1352 struct mem_cgroup
*to
;
1355 * Unlike task_move routines, we access mc.to, mc.from not under
1356 * mutual exclusion by cgroup_mutex. Here, we take spinlock instead.
1358 spin_lock(&mc
.lock
);
1364 ret
= mem_cgroup_same_or_subtree(memcg
, from
)
1365 || mem_cgroup_same_or_subtree(memcg
, to
);
1367 spin_unlock(&mc
.lock
);
1371 static bool mem_cgroup_wait_acct_move(struct mem_cgroup
*memcg
)
1373 if (mc
.moving_task
&& current
!= mc
.moving_task
) {
1374 if (mem_cgroup_under_move(memcg
)) {
1376 prepare_to_wait(&mc
.waitq
, &wait
, TASK_INTERRUPTIBLE
);
1377 /* moving charge context might have finished. */
1380 finish_wait(&mc
.waitq
, &wait
);
1388 * Take this lock when
1389 * - a code tries to modify page's memcg while it's USED.
1390 * - a code tries to modify page state accounting in a memcg.
1391 * see mem_cgroup_stolen(), too.
1393 static void move_lock_mem_cgroup(struct mem_cgroup
*memcg
,
1394 unsigned long *flags
)
1396 spin_lock_irqsave(&memcg
->move_lock
, *flags
);
1399 static void move_unlock_mem_cgroup(struct mem_cgroup
*memcg
,
1400 unsigned long *flags
)
1402 spin_unlock_irqrestore(&memcg
->move_lock
, *flags
);
1406 * mem_cgroup_print_oom_info: Called from OOM with tasklist_lock held in read mode.
1407 * @memcg: The memory cgroup that went over limit
1408 * @p: Task that is going to be killed
1410 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1413 void mem_cgroup_print_oom_info(struct mem_cgroup
*memcg
, struct task_struct
*p
)
1415 struct cgroup
*task_cgrp
;
1416 struct cgroup
*mem_cgrp
;
1418 * Need a buffer in BSS, can't rely on allocations. The code relies
1419 * on the assumption that OOM is serialized for memory controller.
1420 * If this assumption is broken, revisit this code.
1422 static char memcg_name
[PATH_MAX
];
1430 mem_cgrp
= memcg
->css
.cgroup
;
1431 task_cgrp
= task_cgroup(p
, mem_cgroup_subsys_id
);
1433 ret
= cgroup_path(task_cgrp
, memcg_name
, PATH_MAX
);
1436 * Unfortunately, we are unable to convert to a useful name
1437 * But we'll still print out the usage information
1444 printk(KERN_INFO
"Task in %s killed", memcg_name
);
1447 ret
= cgroup_path(mem_cgrp
, memcg_name
, PATH_MAX
);
1455 * Continues from above, so we don't need an KERN_ level
1457 printk(KERN_CONT
" as a result of limit of %s\n", memcg_name
);
1460 printk(KERN_INFO
"memory: usage %llukB, limit %llukB, failcnt %llu\n",
1461 res_counter_read_u64(&memcg
->res
, RES_USAGE
) >> 10,
1462 res_counter_read_u64(&memcg
->res
, RES_LIMIT
) >> 10,
1463 res_counter_read_u64(&memcg
->res
, RES_FAILCNT
));
1464 printk(KERN_INFO
"memory+swap: usage %llukB, limit %llukB, "
1466 res_counter_read_u64(&memcg
->memsw
, RES_USAGE
) >> 10,
1467 res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
) >> 10,
1468 res_counter_read_u64(&memcg
->memsw
, RES_FAILCNT
));
1472 * This function returns the number of memcg under hierarchy tree. Returns
1473 * 1(self count) if no children.
1475 static int mem_cgroup_count_children(struct mem_cgroup
*memcg
)
1478 struct mem_cgroup
*iter
;
1480 for_each_mem_cgroup_tree(iter
, memcg
)
1486 * Return the memory (and swap, if configured) limit for a memcg.
1488 u64
mem_cgroup_get_limit(struct mem_cgroup
*memcg
)
1493 limit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
1494 limit
+= total_swap_pages
<< PAGE_SHIFT
;
1496 memsw
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
1498 * If memsw is finite and limits the amount of swap space available
1499 * to this memcg, return that limit.
1501 return min(limit
, memsw
);
1504 static unsigned long mem_cgroup_reclaim(struct mem_cgroup
*memcg
,
1506 unsigned long flags
)
1508 unsigned long total
= 0;
1509 bool noswap
= false;
1512 if (flags
& MEM_CGROUP_RECLAIM_NOSWAP
)
1514 if (!(flags
& MEM_CGROUP_RECLAIM_SHRINK
) && memcg
->memsw_is_minimum
)
1517 for (loop
= 0; loop
< MEM_CGROUP_MAX_RECLAIM_LOOPS
; loop
++) {
1519 drain_all_stock_async(memcg
);
1520 total
+= try_to_free_mem_cgroup_pages(memcg
, gfp_mask
, noswap
);
1522 * Allow limit shrinkers, which are triggered directly
1523 * by userspace, to catch signals and stop reclaim
1524 * after minimal progress, regardless of the margin.
1526 if (total
&& (flags
& MEM_CGROUP_RECLAIM_SHRINK
))
1528 if (mem_cgroup_margin(memcg
))
1531 * If nothing was reclaimed after two attempts, there
1532 * may be no reclaimable pages in this hierarchy.
1541 * test_mem_cgroup_node_reclaimable
1542 * @mem: the target memcg
1543 * @nid: the node ID to be checked.
1544 * @noswap : specify true here if the user wants flle only information.
1546 * This function returns whether the specified memcg contains any
1547 * reclaimable pages on a node. Returns true if there are any reclaimable
1548 * pages in the node.
1550 static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup
*memcg
,
1551 int nid
, bool noswap
)
1553 if (mem_cgroup_node_nr_lru_pages(memcg
, nid
, LRU_ALL_FILE
))
1555 if (noswap
|| !total_swap_pages
)
1557 if (mem_cgroup_node_nr_lru_pages(memcg
, nid
, LRU_ALL_ANON
))
1562 #if MAX_NUMNODES > 1
1565 * Always updating the nodemask is not very good - even if we have an empty
1566 * list or the wrong list here, we can start from some node and traverse all
1567 * nodes based on the zonelist. So update the list loosely once per 10 secs.
1570 static void mem_cgroup_may_update_nodemask(struct mem_cgroup
*memcg
)
1574 * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET
1575 * pagein/pageout changes since the last update.
1577 if (!atomic_read(&memcg
->numainfo_events
))
1579 if (atomic_inc_return(&memcg
->numainfo_updating
) > 1)
1582 /* make a nodemask where this memcg uses memory from */
1583 memcg
->scan_nodes
= node_states
[N_HIGH_MEMORY
];
1585 for_each_node_mask(nid
, node_states
[N_HIGH_MEMORY
]) {
1587 if (!test_mem_cgroup_node_reclaimable(memcg
, nid
, false))
1588 node_clear(nid
, memcg
->scan_nodes
);
1591 atomic_set(&memcg
->numainfo_events
, 0);
1592 atomic_set(&memcg
->numainfo_updating
, 0);
1596 * Selecting a node where we start reclaim from. Because what we need is just
1597 * reducing usage counter, start from anywhere is O,K. Considering
1598 * memory reclaim from current node, there are pros. and cons.
1600 * Freeing memory from current node means freeing memory from a node which
1601 * we'll use or we've used. So, it may make LRU bad. And if several threads
1602 * hit limits, it will see a contention on a node. But freeing from remote
1603 * node means more costs for memory reclaim because of memory latency.
1605 * Now, we use round-robin. Better algorithm is welcomed.
1607 int mem_cgroup_select_victim_node(struct mem_cgroup
*memcg
)
1611 mem_cgroup_may_update_nodemask(memcg
);
1612 node
= memcg
->last_scanned_node
;
1614 node
= next_node(node
, memcg
->scan_nodes
);
1615 if (node
== MAX_NUMNODES
)
1616 node
= first_node(memcg
->scan_nodes
);
1618 * We call this when we hit limit, not when pages are added to LRU.
1619 * No LRU may hold pages because all pages are UNEVICTABLE or
1620 * memcg is too small and all pages are not on LRU. In that case,
1621 * we use curret node.
1623 if (unlikely(node
== MAX_NUMNODES
))
1624 node
= numa_node_id();
1626 memcg
->last_scanned_node
= node
;
1631 * Check all nodes whether it contains reclaimable pages or not.
1632 * For quick scan, we make use of scan_nodes. This will allow us to skip
1633 * unused nodes. But scan_nodes is lazily updated and may not cotain
1634 * enough new information. We need to do double check.
1636 bool mem_cgroup_reclaimable(struct mem_cgroup
*memcg
, bool noswap
)
1641 * quick check...making use of scan_node.
1642 * We can skip unused nodes.
1644 if (!nodes_empty(memcg
->scan_nodes
)) {
1645 for (nid
= first_node(memcg
->scan_nodes
);
1647 nid
= next_node(nid
, memcg
->scan_nodes
)) {
1649 if (test_mem_cgroup_node_reclaimable(memcg
, nid
, noswap
))
1654 * Check rest of nodes.
1656 for_each_node_state(nid
, N_HIGH_MEMORY
) {
1657 if (node_isset(nid
, memcg
->scan_nodes
))
1659 if (test_mem_cgroup_node_reclaimable(memcg
, nid
, noswap
))
1666 int mem_cgroup_select_victim_node(struct mem_cgroup
*memcg
)
1671 bool mem_cgroup_reclaimable(struct mem_cgroup
*memcg
, bool noswap
)
1673 return test_mem_cgroup_node_reclaimable(memcg
, 0, noswap
);
1677 static int mem_cgroup_soft_reclaim(struct mem_cgroup
*root_memcg
,
1680 unsigned long *total_scanned
)
1682 struct mem_cgroup
*victim
= NULL
;
1685 unsigned long excess
;
1686 unsigned long nr_scanned
;
1687 struct mem_cgroup_reclaim_cookie reclaim
= {
1692 excess
= res_counter_soft_limit_excess(&root_memcg
->res
) >> PAGE_SHIFT
;
1695 victim
= mem_cgroup_iter(root_memcg
, victim
, &reclaim
);
1700 * If we have not been able to reclaim
1701 * anything, it might because there are
1702 * no reclaimable pages under this hierarchy
1707 * We want to do more targeted reclaim.
1708 * excess >> 2 is not to excessive so as to
1709 * reclaim too much, nor too less that we keep
1710 * coming back to reclaim from this cgroup
1712 if (total
>= (excess
>> 2) ||
1713 (loop
> MEM_CGROUP_MAX_RECLAIM_LOOPS
))
1718 if (!mem_cgroup_reclaimable(victim
, false))
1720 total
+= mem_cgroup_shrink_node_zone(victim
, gfp_mask
, false,
1722 *total_scanned
+= nr_scanned
;
1723 if (!res_counter_soft_limit_excess(&root_memcg
->res
))
1726 mem_cgroup_iter_break(root_memcg
, victim
);
1731 * Check OOM-Killer is already running under our hierarchy.
1732 * If someone is running, return false.
1733 * Has to be called with memcg_oom_lock
1735 static bool mem_cgroup_oom_lock(struct mem_cgroup
*memcg
)
1737 struct mem_cgroup
*iter
, *failed
= NULL
;
1739 for_each_mem_cgroup_tree(iter
, memcg
) {
1740 if (iter
->oom_lock
) {
1742 * this subtree of our hierarchy is already locked
1743 * so we cannot give a lock.
1746 mem_cgroup_iter_break(memcg
, iter
);
1749 iter
->oom_lock
= true;
1756 * OK, we failed to lock the whole subtree so we have to clean up
1757 * what we set up to the failing subtree
1759 for_each_mem_cgroup_tree(iter
, memcg
) {
1760 if (iter
== failed
) {
1761 mem_cgroup_iter_break(memcg
, iter
);
1764 iter
->oom_lock
= false;
1770 * Has to be called with memcg_oom_lock
1772 static int mem_cgroup_oom_unlock(struct mem_cgroup
*memcg
)
1774 struct mem_cgroup
*iter
;
1776 for_each_mem_cgroup_tree(iter
, memcg
)
1777 iter
->oom_lock
= false;
1781 static void mem_cgroup_mark_under_oom(struct mem_cgroup
*memcg
)
1783 struct mem_cgroup
*iter
;
1785 for_each_mem_cgroup_tree(iter
, memcg
)
1786 atomic_inc(&iter
->under_oom
);
1789 static void mem_cgroup_unmark_under_oom(struct mem_cgroup
*memcg
)
1791 struct mem_cgroup
*iter
;
1794 * When a new child is created while the hierarchy is under oom,
1795 * mem_cgroup_oom_lock() may not be called. We have to use
1796 * atomic_add_unless() here.
1798 for_each_mem_cgroup_tree(iter
, memcg
)
1799 atomic_add_unless(&iter
->under_oom
, -1, 0);
1802 static DEFINE_SPINLOCK(memcg_oom_lock
);
1803 static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq
);
1805 struct oom_wait_info
{
1806 struct mem_cgroup
*memcg
;
1810 static int memcg_oom_wake_function(wait_queue_t
*wait
,
1811 unsigned mode
, int sync
, void *arg
)
1813 struct mem_cgroup
*wake_memcg
= (struct mem_cgroup
*)arg
;
1814 struct mem_cgroup
*oom_wait_memcg
;
1815 struct oom_wait_info
*oom_wait_info
;
1817 oom_wait_info
= container_of(wait
, struct oom_wait_info
, wait
);
1818 oom_wait_memcg
= oom_wait_info
->memcg
;
1821 * Both of oom_wait_info->memcg and wake_memcg are stable under us.
1822 * Then we can use css_is_ancestor without taking care of RCU.
1824 if (!mem_cgroup_same_or_subtree(oom_wait_memcg
, wake_memcg
)
1825 && !mem_cgroup_same_or_subtree(wake_memcg
, oom_wait_memcg
))
1827 return autoremove_wake_function(wait
, mode
, sync
, arg
);
1830 static void memcg_wakeup_oom(struct mem_cgroup
*memcg
)
1832 /* for filtering, pass "memcg" as argument. */
1833 __wake_up(&memcg_oom_waitq
, TASK_NORMAL
, 0, memcg
);
1836 static void memcg_oom_recover(struct mem_cgroup
*memcg
)
1838 if (memcg
&& atomic_read(&memcg
->under_oom
))
1839 memcg_wakeup_oom(memcg
);
1843 * try to call OOM killer. returns false if we should exit memory-reclaim loop.
1845 bool mem_cgroup_handle_oom(struct mem_cgroup
*memcg
, gfp_t mask
, int order
)
1847 struct oom_wait_info owait
;
1848 bool locked
, need_to_kill
;
1850 owait
.memcg
= memcg
;
1851 owait
.wait
.flags
= 0;
1852 owait
.wait
.func
= memcg_oom_wake_function
;
1853 owait
.wait
.private = current
;
1854 INIT_LIST_HEAD(&owait
.wait
.task_list
);
1855 need_to_kill
= true;
1856 mem_cgroup_mark_under_oom(memcg
);
1858 /* At first, try to OOM lock hierarchy under memcg.*/
1859 spin_lock(&memcg_oom_lock
);
1860 locked
= mem_cgroup_oom_lock(memcg
);
1862 * Even if signal_pending(), we can't quit charge() loop without
1863 * accounting. So, UNINTERRUPTIBLE is appropriate. But SIGKILL
1864 * under OOM is always welcomed, use TASK_KILLABLE here.
1866 prepare_to_wait(&memcg_oom_waitq
, &owait
.wait
, TASK_KILLABLE
);
1867 if (!locked
|| memcg
->oom_kill_disable
)
1868 need_to_kill
= false;
1870 mem_cgroup_oom_notify(memcg
);
1871 spin_unlock(&memcg_oom_lock
);
1874 finish_wait(&memcg_oom_waitq
, &owait
.wait
);
1875 mem_cgroup_out_of_memory(memcg
, mask
, order
);
1878 finish_wait(&memcg_oom_waitq
, &owait
.wait
);
1880 spin_lock(&memcg_oom_lock
);
1882 mem_cgroup_oom_unlock(memcg
);
1883 memcg_wakeup_oom(memcg
);
1884 spin_unlock(&memcg_oom_lock
);
1886 mem_cgroup_unmark_under_oom(memcg
);
1888 if (test_thread_flag(TIF_MEMDIE
) || fatal_signal_pending(current
))
1890 /* Give chance to dying process */
1891 schedule_timeout_uninterruptible(1);
1896 * Currently used to update mapped file statistics, but the routine can be
1897 * generalized to update other statistics as well.
1899 * Notes: Race condition
1901 * We usually use page_cgroup_lock() for accessing page_cgroup member but
1902 * it tends to be costly. But considering some conditions, we doesn't need
1903 * to do so _always_.
1905 * Considering "charge", lock_page_cgroup() is not required because all
1906 * file-stat operations happen after a page is attached to radix-tree. There
1907 * are no race with "charge".
1909 * Considering "uncharge", we know that memcg doesn't clear pc->mem_cgroup
1910 * at "uncharge" intentionally. So, we always see valid pc->mem_cgroup even
1911 * if there are race with "uncharge". Statistics itself is properly handled
1914 * Considering "move", this is an only case we see a race. To make the race
1915 * small, we check mm->moving_account and detect there are possibility of race
1916 * If there is, we take a lock.
1919 void __mem_cgroup_begin_update_page_stat(struct page
*page
,
1920 bool *locked
, unsigned long *flags
)
1922 struct mem_cgroup
*memcg
;
1923 struct page_cgroup
*pc
;
1925 pc
= lookup_page_cgroup(page
);
1927 memcg
= pc
->mem_cgroup
;
1928 if (unlikely(!memcg
|| !PageCgroupUsed(pc
)))
1931 * If this memory cgroup is not under account moving, we don't
1932 * need to take move_lock_page_cgroup(). Because we already hold
1933 * rcu_read_lock(), any calls to move_account will be delayed until
1934 * rcu_read_unlock() if mem_cgroup_stolen() == true.
1936 if (!mem_cgroup_stolen(memcg
))
1939 move_lock_mem_cgroup(memcg
, flags
);
1940 if (memcg
!= pc
->mem_cgroup
|| !PageCgroupUsed(pc
)) {
1941 move_unlock_mem_cgroup(memcg
, flags
);
1947 void __mem_cgroup_end_update_page_stat(struct page
*page
, unsigned long *flags
)
1949 struct page_cgroup
*pc
= lookup_page_cgroup(page
);
1952 * It's guaranteed that pc->mem_cgroup never changes while
1953 * lock is held because a routine modifies pc->mem_cgroup
1954 * should take move_lock_page_cgroup().
1956 move_unlock_mem_cgroup(pc
->mem_cgroup
, flags
);
1959 void mem_cgroup_update_page_stat(struct page
*page
,
1960 enum mem_cgroup_page_stat_item idx
, int val
)
1962 struct mem_cgroup
*memcg
;
1963 struct page_cgroup
*pc
= lookup_page_cgroup(page
);
1964 unsigned long uninitialized_var(flags
);
1966 if (mem_cgroup_disabled())
1969 memcg
= pc
->mem_cgroup
;
1970 if (unlikely(!memcg
|| !PageCgroupUsed(pc
)))
1974 case MEMCG_NR_FILE_MAPPED
:
1975 idx
= MEM_CGROUP_STAT_FILE_MAPPED
;
1981 this_cpu_add(memcg
->stat
->count
[idx
], val
);
1985 * size of first charge trial. "32" comes from vmscan.c's magic value.
1986 * TODO: maybe necessary to use big numbers in big irons.
1988 #define CHARGE_BATCH 32U
1989 struct memcg_stock_pcp
{
1990 struct mem_cgroup
*cached
; /* this never be root cgroup */
1991 unsigned int nr_pages
;
1992 struct work_struct work
;
1993 unsigned long flags
;
1994 #define FLUSHING_CACHED_CHARGE (0)
1996 static DEFINE_PER_CPU(struct memcg_stock_pcp
, memcg_stock
);
1997 static DEFINE_MUTEX(percpu_charge_mutex
);
2000 * Try to consume stocked charge on this cpu. If success, one page is consumed
2001 * from local stock and true is returned. If the stock is 0 or charges from a
2002 * cgroup which is not current target, returns false. This stock will be
2005 static bool consume_stock(struct mem_cgroup
*memcg
)
2007 struct memcg_stock_pcp
*stock
;
2010 stock
= &get_cpu_var(memcg_stock
);
2011 if (memcg
== stock
->cached
&& stock
->nr_pages
)
2013 else /* need to call res_counter_charge */
2015 put_cpu_var(memcg_stock
);
2020 * Returns stocks cached in percpu to res_counter and reset cached information.
2022 static void drain_stock(struct memcg_stock_pcp
*stock
)
2024 struct mem_cgroup
*old
= stock
->cached
;
2026 if (stock
->nr_pages
) {
2027 unsigned long bytes
= stock
->nr_pages
* PAGE_SIZE
;
2029 res_counter_uncharge(&old
->res
, bytes
);
2030 if (do_swap_account
)
2031 res_counter_uncharge(&old
->memsw
, bytes
);
2032 stock
->nr_pages
= 0;
2034 stock
->cached
= NULL
;
2038 * This must be called under preempt disabled or must be called by
2039 * a thread which is pinned to local cpu.
2041 static void drain_local_stock(struct work_struct
*dummy
)
2043 struct memcg_stock_pcp
*stock
= &__get_cpu_var(memcg_stock
);
2045 clear_bit(FLUSHING_CACHED_CHARGE
, &stock
->flags
);
2049 * Cache charges(val) which is from res_counter, to local per_cpu area.
2050 * This will be consumed by consume_stock() function, later.
2052 static void refill_stock(struct mem_cgroup
*memcg
, unsigned int nr_pages
)
2054 struct memcg_stock_pcp
*stock
= &get_cpu_var(memcg_stock
);
2056 if (stock
->cached
!= memcg
) { /* reset if necessary */
2058 stock
->cached
= memcg
;
2060 stock
->nr_pages
+= nr_pages
;
2061 put_cpu_var(memcg_stock
);
2065 * Drains all per-CPU charge caches for given root_memcg resp. subtree
2066 * of the hierarchy under it. sync flag says whether we should block
2067 * until the work is done.
2069 static void drain_all_stock(struct mem_cgroup
*root_memcg
, bool sync
)
2073 /* Notify other cpus that system-wide "drain" is running */
2076 for_each_online_cpu(cpu
) {
2077 struct memcg_stock_pcp
*stock
= &per_cpu(memcg_stock
, cpu
);
2078 struct mem_cgroup
*memcg
;
2080 memcg
= stock
->cached
;
2081 if (!memcg
|| !stock
->nr_pages
)
2083 if (!mem_cgroup_same_or_subtree(root_memcg
, memcg
))
2085 if (!test_and_set_bit(FLUSHING_CACHED_CHARGE
, &stock
->flags
)) {
2087 drain_local_stock(&stock
->work
);
2089 schedule_work_on(cpu
, &stock
->work
);
2097 for_each_online_cpu(cpu
) {
2098 struct memcg_stock_pcp
*stock
= &per_cpu(memcg_stock
, cpu
);
2099 if (test_bit(FLUSHING_CACHED_CHARGE
, &stock
->flags
))
2100 flush_work(&stock
->work
);
2107 * Tries to drain stocked charges in other cpus. This function is asynchronous
2108 * and just put a work per cpu for draining localy on each cpu. Caller can
2109 * expects some charges will be back to res_counter later but cannot wait for
2112 static void drain_all_stock_async(struct mem_cgroup
*root_memcg
)
2115 * If someone calls draining, avoid adding more kworker runs.
2117 if (!mutex_trylock(&percpu_charge_mutex
))
2119 drain_all_stock(root_memcg
, false);
2120 mutex_unlock(&percpu_charge_mutex
);
2123 /* This is a synchronous drain interface. */
2124 static void drain_all_stock_sync(struct mem_cgroup
*root_memcg
)
2126 /* called when force_empty is called */
2127 mutex_lock(&percpu_charge_mutex
);
2128 drain_all_stock(root_memcg
, true);
2129 mutex_unlock(&percpu_charge_mutex
);
2133 * This function drains percpu counter value from DEAD cpu and
2134 * move it to local cpu. Note that this function can be preempted.
2136 static void mem_cgroup_drain_pcp_counter(struct mem_cgroup
*memcg
, int cpu
)
2140 spin_lock(&memcg
->pcp_counter_lock
);
2141 for (i
= 0; i
< MEM_CGROUP_STAT_DATA
; i
++) {
2142 long x
= per_cpu(memcg
->stat
->count
[i
], cpu
);
2144 per_cpu(memcg
->stat
->count
[i
], cpu
) = 0;
2145 memcg
->nocpu_base
.count
[i
] += x
;
2147 for (i
= 0; i
< MEM_CGROUP_EVENTS_NSTATS
; i
++) {
2148 unsigned long x
= per_cpu(memcg
->stat
->events
[i
], cpu
);
2150 per_cpu(memcg
->stat
->events
[i
], cpu
) = 0;
2151 memcg
->nocpu_base
.events
[i
] += x
;
2153 spin_unlock(&memcg
->pcp_counter_lock
);
2156 static int __cpuinit
memcg_cpu_hotplug_callback(struct notifier_block
*nb
,
2157 unsigned long action
,
2160 int cpu
= (unsigned long)hcpu
;
2161 struct memcg_stock_pcp
*stock
;
2162 struct mem_cgroup
*iter
;
2164 if (action
== CPU_ONLINE
)
2167 if (action
!= CPU_DEAD
&& action
!= CPU_DEAD_FROZEN
)
2170 for_each_mem_cgroup(iter
)
2171 mem_cgroup_drain_pcp_counter(iter
, cpu
);
2173 stock
= &per_cpu(memcg_stock
, cpu
);
2179 /* See __mem_cgroup_try_charge() for details */
2181 CHARGE_OK
, /* success */
2182 CHARGE_RETRY
, /* need to retry but retry is not bad */
2183 CHARGE_NOMEM
, /* we can't do more. return -ENOMEM */
2184 CHARGE_WOULDBLOCK
, /* GFP_WAIT wasn't set and no enough res. */
2185 CHARGE_OOM_DIE
, /* the current is killed because of OOM */
2188 static int mem_cgroup_do_charge(struct mem_cgroup
*memcg
, gfp_t gfp_mask
,
2189 unsigned int nr_pages
, bool oom_check
)
2191 unsigned long csize
= nr_pages
* PAGE_SIZE
;
2192 struct mem_cgroup
*mem_over_limit
;
2193 struct res_counter
*fail_res
;
2194 unsigned long flags
= 0;
2197 ret
= res_counter_charge(&memcg
->res
, csize
, &fail_res
);
2200 if (!do_swap_account
)
2202 ret
= res_counter_charge(&memcg
->memsw
, csize
, &fail_res
);
2206 res_counter_uncharge(&memcg
->res
, csize
);
2207 mem_over_limit
= mem_cgroup_from_res_counter(fail_res
, memsw
);
2208 flags
|= MEM_CGROUP_RECLAIM_NOSWAP
;
2210 mem_over_limit
= mem_cgroup_from_res_counter(fail_res
, res
);
2212 * nr_pages can be either a huge page (HPAGE_PMD_NR), a batch
2213 * of regular pages (CHARGE_BATCH), or a single regular page (1).
2215 * Never reclaim on behalf of optional batching, retry with a
2216 * single page instead.
2218 if (nr_pages
== CHARGE_BATCH
)
2219 return CHARGE_RETRY
;
2221 if (!(gfp_mask
& __GFP_WAIT
))
2222 return CHARGE_WOULDBLOCK
;
2224 ret
= mem_cgroup_reclaim(mem_over_limit
, gfp_mask
, flags
);
2225 if (mem_cgroup_margin(mem_over_limit
) >= nr_pages
)
2226 return CHARGE_RETRY
;
2228 * Even though the limit is exceeded at this point, reclaim
2229 * may have been able to free some pages. Retry the charge
2230 * before killing the task.
2232 * Only for regular pages, though: huge pages are rather
2233 * unlikely to succeed so close to the limit, and we fall back
2234 * to regular pages anyway in case of failure.
2236 if (nr_pages
== 1 && ret
)
2237 return CHARGE_RETRY
;
2240 * At task move, charge accounts can be doubly counted. So, it's
2241 * better to wait until the end of task_move if something is going on.
2243 if (mem_cgroup_wait_acct_move(mem_over_limit
))
2244 return CHARGE_RETRY
;
2246 /* If we don't need to call oom-killer at el, return immediately */
2248 return CHARGE_NOMEM
;
2250 if (!mem_cgroup_handle_oom(mem_over_limit
, gfp_mask
, get_order(csize
)))
2251 return CHARGE_OOM_DIE
;
2253 return CHARGE_RETRY
;
2257 * __mem_cgroup_try_charge() does
2258 * 1. detect memcg to be charged against from passed *mm and *ptr,
2259 * 2. update res_counter
2260 * 3. call memory reclaim if necessary.
2262 * In some special case, if the task is fatal, fatal_signal_pending() or
2263 * has TIF_MEMDIE, this function returns -EINTR while writing root_mem_cgroup
2264 * to *ptr. There are two reasons for this. 1: fatal threads should quit as soon
2265 * as possible without any hazards. 2: all pages should have a valid
2266 * pc->mem_cgroup. If mm is NULL and the caller doesn't pass a valid memcg
2267 * pointer, that is treated as a charge to root_mem_cgroup.
2269 * So __mem_cgroup_try_charge() will return
2270 * 0 ... on success, filling *ptr with a valid memcg pointer.
2271 * -ENOMEM ... charge failure because of resource limits.
2272 * -EINTR ... if thread is fatal. *ptr is filled with root_mem_cgroup.
2274 * Unlike the exported interface, an "oom" parameter is added. if oom==true,
2275 * the oom-killer can be invoked.
2277 static int __mem_cgroup_try_charge(struct mm_struct
*mm
,
2279 unsigned int nr_pages
,
2280 struct mem_cgroup
**ptr
,
2283 unsigned int batch
= max(CHARGE_BATCH
, nr_pages
);
2284 int nr_oom_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
2285 struct mem_cgroup
*memcg
= NULL
;
2289 * Unlike gloval-vm's OOM-kill, we're not in memory shortage
2290 * in system level. So, allow to go ahead dying process in addition to
2293 if (unlikely(test_thread_flag(TIF_MEMDIE
)
2294 || fatal_signal_pending(current
)))
2298 * We always charge the cgroup the mm_struct belongs to.
2299 * The mm_struct's mem_cgroup changes on task migration if the
2300 * thread group leader migrates. It's possible that mm is not
2301 * set, if so charge the init_mm (happens for pagecache usage).
2304 *ptr
= root_mem_cgroup
;
2306 if (*ptr
) { /* css should be a valid one */
2308 VM_BUG_ON(css_is_removed(&memcg
->css
));
2309 if (mem_cgroup_is_root(memcg
))
2311 if (nr_pages
== 1 && consume_stock(memcg
))
2313 css_get(&memcg
->css
);
2315 struct task_struct
*p
;
2318 p
= rcu_dereference(mm
->owner
);
2320 * Because we don't have task_lock(), "p" can exit.
2321 * In that case, "memcg" can point to root or p can be NULL with
2322 * race with swapoff. Then, we have small risk of mis-accouning.
2323 * But such kind of mis-account by race always happens because
2324 * we don't have cgroup_mutex(). It's overkill and we allo that
2326 * (*) swapoff at el will charge against mm-struct not against
2327 * task-struct. So, mm->owner can be NULL.
2329 memcg
= mem_cgroup_from_task(p
);
2331 memcg
= root_mem_cgroup
;
2332 if (mem_cgroup_is_root(memcg
)) {
2336 if (nr_pages
== 1 && consume_stock(memcg
)) {
2338 * It seems dagerous to access memcg without css_get().
2339 * But considering how consume_stok works, it's not
2340 * necessary. If consume_stock success, some charges
2341 * from this memcg are cached on this cpu. So, we
2342 * don't need to call css_get()/css_tryget() before
2343 * calling consume_stock().
2348 /* after here, we may be blocked. we need to get refcnt */
2349 if (!css_tryget(&memcg
->css
)) {
2359 /* If killed, bypass charge */
2360 if (fatal_signal_pending(current
)) {
2361 css_put(&memcg
->css
);
2366 if (oom
&& !nr_oom_retries
) {
2368 nr_oom_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
2371 ret
= mem_cgroup_do_charge(memcg
, gfp_mask
, batch
, oom_check
);
2375 case CHARGE_RETRY
: /* not in OOM situation but retry */
2377 css_put(&memcg
->css
);
2380 case CHARGE_WOULDBLOCK
: /* !__GFP_WAIT */
2381 css_put(&memcg
->css
);
2383 case CHARGE_NOMEM
: /* OOM routine works */
2385 css_put(&memcg
->css
);
2388 /* If oom, we never return -ENOMEM */
2391 case CHARGE_OOM_DIE
: /* Killed by OOM Killer */
2392 css_put(&memcg
->css
);
2395 } while (ret
!= CHARGE_OK
);
2397 if (batch
> nr_pages
)
2398 refill_stock(memcg
, batch
- nr_pages
);
2399 css_put(&memcg
->css
);
2407 *ptr
= root_mem_cgroup
;
2412 * Somemtimes we have to undo a charge we got by try_charge().
2413 * This function is for that and do uncharge, put css's refcnt.
2414 * gotten by try_charge().
2416 static void __mem_cgroup_cancel_charge(struct mem_cgroup
*memcg
,
2417 unsigned int nr_pages
)
2419 if (!mem_cgroup_is_root(memcg
)) {
2420 unsigned long bytes
= nr_pages
* PAGE_SIZE
;
2422 res_counter_uncharge(&memcg
->res
, bytes
);
2423 if (do_swap_account
)
2424 res_counter_uncharge(&memcg
->memsw
, bytes
);
2429 * A helper function to get mem_cgroup from ID. must be called under
2430 * rcu_read_lock(). The caller must check css_is_removed() or some if
2431 * it's concern. (dropping refcnt from swap can be called against removed
2434 static struct mem_cgroup
*mem_cgroup_lookup(unsigned short id
)
2436 struct cgroup_subsys_state
*css
;
2438 /* ID 0 is unused ID */
2441 css
= css_lookup(&mem_cgroup_subsys
, id
);
2444 return container_of(css
, struct mem_cgroup
, css
);
2447 struct mem_cgroup
*try_get_mem_cgroup_from_page(struct page
*page
)
2449 struct mem_cgroup
*memcg
= NULL
;
2450 struct page_cgroup
*pc
;
2454 VM_BUG_ON(!PageLocked(page
));
2456 pc
= lookup_page_cgroup(page
);
2457 lock_page_cgroup(pc
);
2458 if (PageCgroupUsed(pc
)) {
2459 memcg
= pc
->mem_cgroup
;
2460 if (memcg
&& !css_tryget(&memcg
->css
))
2462 } else if (PageSwapCache(page
)) {
2463 ent
.val
= page_private(page
);
2464 id
= lookup_swap_cgroup_id(ent
);
2466 memcg
= mem_cgroup_lookup(id
);
2467 if (memcg
&& !css_tryget(&memcg
->css
))
2471 unlock_page_cgroup(pc
);
2475 static void __mem_cgroup_commit_charge(struct mem_cgroup
*memcg
,
2477 unsigned int nr_pages
,
2478 enum charge_type ctype
,
2481 struct page_cgroup
*pc
= lookup_page_cgroup(page
);
2482 struct zone
*uninitialized_var(zone
);
2483 bool was_on_lru
= false;
2486 lock_page_cgroup(pc
);
2487 if (unlikely(PageCgroupUsed(pc
))) {
2488 unlock_page_cgroup(pc
);
2489 __mem_cgroup_cancel_charge(memcg
, nr_pages
);
2493 * we don't need page_cgroup_lock about tail pages, becase they are not
2494 * accessed by any other context at this point.
2498 * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page
2499 * may already be on some other mem_cgroup's LRU. Take care of it.
2502 zone
= page_zone(page
);
2503 spin_lock_irq(&zone
->lru_lock
);
2504 if (PageLRU(page
)) {
2506 del_page_from_lru_list(zone
, page
, page_lru(page
));
2511 pc
->mem_cgroup
= memcg
;
2513 * We access a page_cgroup asynchronously without lock_page_cgroup().
2514 * Especially when a page_cgroup is taken from a page, pc->mem_cgroup
2515 * is accessed after testing USED bit. To make pc->mem_cgroup visible
2516 * before USED bit, we need memory barrier here.
2517 * See mem_cgroup_add_lru_list(), etc.
2520 SetPageCgroupUsed(pc
);
2524 VM_BUG_ON(PageLRU(page
));
2526 add_page_to_lru_list(zone
, page
, page_lru(page
));
2528 spin_unlock_irq(&zone
->lru_lock
);
2531 if (ctype
== MEM_CGROUP_CHARGE_TYPE_MAPPED
)
2536 mem_cgroup_charge_statistics(memcg
, anon
, nr_pages
);
2537 unlock_page_cgroup(pc
);
2540 * "charge_statistics" updated event counter. Then, check it.
2541 * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree.
2542 * if they exceeds softlimit.
2544 memcg_check_events(memcg
, page
);
2547 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
2549 #define PCGF_NOCOPY_AT_SPLIT ((1 << PCG_LOCK) | (1 << PCG_MIGRATION))
2551 * Because tail pages are not marked as "used", set it. We're under
2552 * zone->lru_lock, 'splitting on pmd' and compound_lock.
2553 * charge/uncharge will be never happen and move_account() is done under
2554 * compound_lock(), so we don't have to take care of races.
2556 void mem_cgroup_split_huge_fixup(struct page
*head
)
2558 struct page_cgroup
*head_pc
= lookup_page_cgroup(head
);
2559 struct page_cgroup
*pc
;
2562 if (mem_cgroup_disabled())
2564 for (i
= 1; i
< HPAGE_PMD_NR
; i
++) {
2566 pc
->mem_cgroup
= head_pc
->mem_cgroup
;
2567 smp_wmb();/* see __commit_charge() */
2568 pc
->flags
= head_pc
->flags
& ~PCGF_NOCOPY_AT_SPLIT
;
2571 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
2574 * mem_cgroup_move_account - move account of the page
2576 * @nr_pages: number of regular pages (>1 for huge pages)
2577 * @pc: page_cgroup of the page.
2578 * @from: mem_cgroup which the page is moved from.
2579 * @to: mem_cgroup which the page is moved to. @from != @to.
2580 * @uncharge: whether we should call uncharge and css_put against @from.
2582 * The caller must confirm following.
2583 * - page is not on LRU (isolate_page() is useful.)
2584 * - compound_lock is held when nr_pages > 1
2586 * This function doesn't do "charge" nor css_get to new cgroup. It should be
2587 * done by a caller(__mem_cgroup_try_charge would be useful). If @uncharge is
2588 * true, this function does "uncharge" from old cgroup, but it doesn't if
2589 * @uncharge is false, so a caller should do "uncharge".
2591 static int mem_cgroup_move_account(struct page
*page
,
2592 unsigned int nr_pages
,
2593 struct page_cgroup
*pc
,
2594 struct mem_cgroup
*from
,
2595 struct mem_cgroup
*to
,
2598 unsigned long flags
;
2600 bool anon
= PageAnon(page
);
2602 VM_BUG_ON(from
== to
);
2603 VM_BUG_ON(PageLRU(page
));
2605 * The page is isolated from LRU. So, collapse function
2606 * will not handle this page. But page splitting can happen.
2607 * Do this check under compound_page_lock(). The caller should
2611 if (nr_pages
> 1 && !PageTransHuge(page
))
2614 lock_page_cgroup(pc
);
2617 if (!PageCgroupUsed(pc
) || pc
->mem_cgroup
!= from
)
2620 move_lock_mem_cgroup(from
, &flags
);
2622 if (!anon
&& page_mapped(page
)) {
2623 /* Update mapped_file data for mem_cgroup */
2625 __this_cpu_dec(from
->stat
->count
[MEM_CGROUP_STAT_FILE_MAPPED
]);
2626 __this_cpu_inc(to
->stat
->count
[MEM_CGROUP_STAT_FILE_MAPPED
]);
2629 mem_cgroup_charge_statistics(from
, anon
, -nr_pages
);
2631 /* This is not "cancel", but cancel_charge does all we need. */
2632 __mem_cgroup_cancel_charge(from
, nr_pages
);
2634 /* caller should have done css_get */
2635 pc
->mem_cgroup
= to
;
2636 mem_cgroup_charge_statistics(to
, anon
, nr_pages
);
2638 * We charges against "to" which may not have any tasks. Then, "to"
2639 * can be under rmdir(). But in current implementation, caller of
2640 * this function is just force_empty() and move charge, so it's
2641 * guaranteed that "to" is never removed. So, we don't check rmdir
2644 move_unlock_mem_cgroup(from
, &flags
);
2647 unlock_page_cgroup(pc
);
2651 memcg_check_events(to
, page
);
2652 memcg_check_events(from
, page
);
2658 * move charges to its parent.
2661 static int mem_cgroup_move_parent(struct page
*page
,
2662 struct page_cgroup
*pc
,
2663 struct mem_cgroup
*child
,
2666 struct cgroup
*cg
= child
->css
.cgroup
;
2667 struct cgroup
*pcg
= cg
->parent
;
2668 struct mem_cgroup
*parent
;
2669 unsigned int nr_pages
;
2670 unsigned long uninitialized_var(flags
);
2678 if (!get_page_unless_zero(page
))
2680 if (isolate_lru_page(page
))
2683 nr_pages
= hpage_nr_pages(page
);
2685 parent
= mem_cgroup_from_cont(pcg
);
2686 ret
= __mem_cgroup_try_charge(NULL
, gfp_mask
, nr_pages
, &parent
, false);
2691 flags
= compound_lock_irqsave(page
);
2693 ret
= mem_cgroup_move_account(page
, nr_pages
, pc
, child
, parent
, true);
2695 __mem_cgroup_cancel_charge(parent
, nr_pages
);
2698 compound_unlock_irqrestore(page
, flags
);
2700 putback_lru_page(page
);
2708 * Charge the memory controller for page usage.
2710 * 0 if the charge was successful
2711 * < 0 if the cgroup is over its limit
2713 static int mem_cgroup_charge_common(struct page
*page
, struct mm_struct
*mm
,
2714 gfp_t gfp_mask
, enum charge_type ctype
)
2716 struct mem_cgroup
*memcg
= NULL
;
2717 unsigned int nr_pages
= 1;
2721 if (PageTransHuge(page
)) {
2722 nr_pages
<<= compound_order(page
);
2723 VM_BUG_ON(!PageTransHuge(page
));
2725 * Never OOM-kill a process for a huge page. The
2726 * fault handler will fall back to regular pages.
2731 ret
= __mem_cgroup_try_charge(mm
, gfp_mask
, nr_pages
, &memcg
, oom
);
2734 __mem_cgroup_commit_charge(memcg
, page
, nr_pages
, ctype
, false);
2738 int mem_cgroup_newpage_charge(struct page
*page
,
2739 struct mm_struct
*mm
, gfp_t gfp_mask
)
2741 if (mem_cgroup_disabled())
2743 VM_BUG_ON(page_mapped(page
));
2744 VM_BUG_ON(page
->mapping
&& !PageAnon(page
));
2746 return mem_cgroup_charge_common(page
, mm
, gfp_mask
,
2747 MEM_CGROUP_CHARGE_TYPE_MAPPED
);
2751 __mem_cgroup_commit_charge_swapin(struct page
*page
, struct mem_cgroup
*ptr
,
2752 enum charge_type ctype
);
2754 int mem_cgroup_cache_charge(struct page
*page
, struct mm_struct
*mm
,
2757 struct mem_cgroup
*memcg
= NULL
;
2758 enum charge_type type
= MEM_CGROUP_CHARGE_TYPE_CACHE
;
2761 if (mem_cgroup_disabled())
2763 if (PageCompound(page
))
2768 if (!page_is_file_cache(page
))
2769 type
= MEM_CGROUP_CHARGE_TYPE_SHMEM
;
2771 if (!PageSwapCache(page
))
2772 ret
= mem_cgroup_charge_common(page
, mm
, gfp_mask
, type
);
2773 else { /* page is swapcache/shmem */
2774 ret
= mem_cgroup_try_charge_swapin(mm
, page
, gfp_mask
, &memcg
);
2776 __mem_cgroup_commit_charge_swapin(page
, memcg
, type
);
2782 * While swap-in, try_charge -> commit or cancel, the page is locked.
2783 * And when try_charge() successfully returns, one refcnt to memcg without
2784 * struct page_cgroup is acquired. This refcnt will be consumed by
2785 * "commit()" or removed by "cancel()"
2787 int mem_cgroup_try_charge_swapin(struct mm_struct
*mm
,
2789 gfp_t mask
, struct mem_cgroup
**memcgp
)
2791 struct mem_cgroup
*memcg
;
2796 if (mem_cgroup_disabled())
2799 if (!do_swap_account
)
2802 * A racing thread's fault, or swapoff, may have already updated
2803 * the pte, and even removed page from swap cache: in those cases
2804 * do_swap_page()'s pte_same() test will fail; but there's also a
2805 * KSM case which does need to charge the page.
2807 if (!PageSwapCache(page
))
2809 memcg
= try_get_mem_cgroup_from_page(page
);
2813 ret
= __mem_cgroup_try_charge(NULL
, mask
, 1, memcgp
, true);
2814 css_put(&memcg
->css
);
2821 ret
= __mem_cgroup_try_charge(mm
, mask
, 1, memcgp
, true);
2828 __mem_cgroup_commit_charge_swapin(struct page
*page
, struct mem_cgroup
*memcg
,
2829 enum charge_type ctype
)
2831 if (mem_cgroup_disabled())
2835 cgroup_exclude_rmdir(&memcg
->css
);
2837 __mem_cgroup_commit_charge(memcg
, page
, 1, ctype
, true);
2839 * Now swap is on-memory. This means this page may be
2840 * counted both as mem and swap....double count.
2841 * Fix it by uncharging from memsw. Basically, this SwapCache is stable
2842 * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page()
2843 * may call delete_from_swap_cache() before reach here.
2845 if (do_swap_account
&& PageSwapCache(page
)) {
2846 swp_entry_t ent
= {.val
= page_private(page
)};
2847 mem_cgroup_uncharge_swap(ent
);
2850 * At swapin, we may charge account against cgroup which has no tasks.
2851 * So, rmdir()->pre_destroy() can be called while we do this charge.
2852 * In that case, we need to call pre_destroy() again. check it here.
2854 cgroup_release_and_wakeup_rmdir(&memcg
->css
);
2857 void mem_cgroup_commit_charge_swapin(struct page
*page
,
2858 struct mem_cgroup
*memcg
)
2860 __mem_cgroup_commit_charge_swapin(page
, memcg
,
2861 MEM_CGROUP_CHARGE_TYPE_MAPPED
);
2864 void mem_cgroup_cancel_charge_swapin(struct mem_cgroup
*memcg
)
2866 if (mem_cgroup_disabled())
2870 __mem_cgroup_cancel_charge(memcg
, 1);
2873 static void mem_cgroup_do_uncharge(struct mem_cgroup
*memcg
,
2874 unsigned int nr_pages
,
2875 const enum charge_type ctype
)
2877 struct memcg_batch_info
*batch
= NULL
;
2878 bool uncharge_memsw
= true;
2880 /* If swapout, usage of swap doesn't decrease */
2881 if (!do_swap_account
|| ctype
== MEM_CGROUP_CHARGE_TYPE_SWAPOUT
)
2882 uncharge_memsw
= false;
2884 batch
= ¤t
->memcg_batch
;
2886 * In usual, we do css_get() when we remember memcg pointer.
2887 * But in this case, we keep res->usage until end of a series of
2888 * uncharges. Then, it's ok to ignore memcg's refcnt.
2891 batch
->memcg
= memcg
;
2893 * do_batch > 0 when unmapping pages or inode invalidate/truncate.
2894 * In those cases, all pages freed continuously can be expected to be in
2895 * the same cgroup and we have chance to coalesce uncharges.
2896 * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE)
2897 * because we want to do uncharge as soon as possible.
2900 if (!batch
->do_batch
|| test_thread_flag(TIF_MEMDIE
))
2901 goto direct_uncharge
;
2904 goto direct_uncharge
;
2907 * In typical case, batch->memcg == mem. This means we can
2908 * merge a series of uncharges to an uncharge of res_counter.
2909 * If not, we uncharge res_counter ony by one.
2911 if (batch
->memcg
!= memcg
)
2912 goto direct_uncharge
;
2913 /* remember freed charge and uncharge it later */
2916 batch
->memsw_nr_pages
++;
2919 res_counter_uncharge(&memcg
->res
, nr_pages
* PAGE_SIZE
);
2921 res_counter_uncharge(&memcg
->memsw
, nr_pages
* PAGE_SIZE
);
2922 if (unlikely(batch
->memcg
!= memcg
))
2923 memcg_oom_recover(memcg
);
2927 * uncharge if !page_mapped(page)
2929 static struct mem_cgroup
*
2930 __mem_cgroup_uncharge_common(struct page
*page
, enum charge_type ctype
)
2932 struct mem_cgroup
*memcg
= NULL
;
2933 unsigned int nr_pages
= 1;
2934 struct page_cgroup
*pc
;
2937 if (mem_cgroup_disabled())
2940 if (PageSwapCache(page
))
2943 if (PageTransHuge(page
)) {
2944 nr_pages
<<= compound_order(page
);
2945 VM_BUG_ON(!PageTransHuge(page
));
2948 * Check if our page_cgroup is valid
2950 pc
= lookup_page_cgroup(page
);
2951 if (unlikely(!PageCgroupUsed(pc
)))
2954 lock_page_cgroup(pc
);
2956 memcg
= pc
->mem_cgroup
;
2958 if (!PageCgroupUsed(pc
))
2961 anon
= PageAnon(page
);
2964 case MEM_CGROUP_CHARGE_TYPE_MAPPED
:
2966 * Generally PageAnon tells if it's the anon statistics to be
2967 * updated; but sometimes e.g. mem_cgroup_uncharge_page() is
2968 * used before page reached the stage of being marked PageAnon.
2972 case MEM_CGROUP_CHARGE_TYPE_DROP
:
2973 /* See mem_cgroup_prepare_migration() */
2974 if (page_mapped(page
) || PageCgroupMigration(pc
))
2977 case MEM_CGROUP_CHARGE_TYPE_SWAPOUT
:
2978 if (!PageAnon(page
)) { /* Shared memory */
2979 if (page
->mapping
&& !page_is_file_cache(page
))
2981 } else if (page_mapped(page
)) /* Anon */
2988 mem_cgroup_charge_statistics(memcg
, anon
, -nr_pages
);
2990 ClearPageCgroupUsed(pc
);
2992 * pc->mem_cgroup is not cleared here. It will be accessed when it's
2993 * freed from LRU. This is safe because uncharged page is expected not
2994 * to be reused (freed soon). Exception is SwapCache, it's handled by
2995 * special functions.
2998 unlock_page_cgroup(pc
);
3000 * even after unlock, we have memcg->res.usage here and this memcg
3001 * will never be freed.
3003 memcg_check_events(memcg
, page
);
3004 if (do_swap_account
&& ctype
== MEM_CGROUP_CHARGE_TYPE_SWAPOUT
) {
3005 mem_cgroup_swap_statistics(memcg
, true);
3006 mem_cgroup_get(memcg
);
3008 if (!mem_cgroup_is_root(memcg
))
3009 mem_cgroup_do_uncharge(memcg
, nr_pages
, ctype
);
3014 unlock_page_cgroup(pc
);
3018 void mem_cgroup_uncharge_page(struct page
*page
)
3021 if (page_mapped(page
))
3023 VM_BUG_ON(page
->mapping
&& !PageAnon(page
));
3024 __mem_cgroup_uncharge_common(page
, MEM_CGROUP_CHARGE_TYPE_MAPPED
);
3027 void mem_cgroup_uncharge_cache_page(struct page
*page
)
3029 VM_BUG_ON(page_mapped(page
));
3030 VM_BUG_ON(page
->mapping
);
3031 __mem_cgroup_uncharge_common(page
, MEM_CGROUP_CHARGE_TYPE_CACHE
);
3035 * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate.
3036 * In that cases, pages are freed continuously and we can expect pages
3037 * are in the same memcg. All these calls itself limits the number of
3038 * pages freed at once, then uncharge_start/end() is called properly.
3039 * This may be called prural(2) times in a context,
3042 void mem_cgroup_uncharge_start(void)
3044 current
->memcg_batch
.do_batch
++;
3045 /* We can do nest. */
3046 if (current
->memcg_batch
.do_batch
== 1) {
3047 current
->memcg_batch
.memcg
= NULL
;
3048 current
->memcg_batch
.nr_pages
= 0;
3049 current
->memcg_batch
.memsw_nr_pages
= 0;
3053 void mem_cgroup_uncharge_end(void)
3055 struct memcg_batch_info
*batch
= ¤t
->memcg_batch
;
3057 if (!batch
->do_batch
)
3061 if (batch
->do_batch
) /* If stacked, do nothing. */
3067 * This "batch->memcg" is valid without any css_get/put etc...
3068 * bacause we hide charges behind us.
3070 if (batch
->nr_pages
)
3071 res_counter_uncharge(&batch
->memcg
->res
,
3072 batch
->nr_pages
* PAGE_SIZE
);
3073 if (batch
->memsw_nr_pages
)
3074 res_counter_uncharge(&batch
->memcg
->memsw
,
3075 batch
->memsw_nr_pages
* PAGE_SIZE
);
3076 memcg_oom_recover(batch
->memcg
);
3077 /* forget this pointer (for sanity check) */
3078 batch
->memcg
= NULL
;
3083 * called after __delete_from_swap_cache() and drop "page" account.
3084 * memcg information is recorded to swap_cgroup of "ent"
3087 mem_cgroup_uncharge_swapcache(struct page
*page
, swp_entry_t ent
, bool swapout
)
3089 struct mem_cgroup
*memcg
;
3090 int ctype
= MEM_CGROUP_CHARGE_TYPE_SWAPOUT
;
3092 if (!swapout
) /* this was a swap cache but the swap is unused ! */
3093 ctype
= MEM_CGROUP_CHARGE_TYPE_DROP
;
3095 memcg
= __mem_cgroup_uncharge_common(page
, ctype
);
3098 * record memcg information, if swapout && memcg != NULL,
3099 * mem_cgroup_get() was called in uncharge().
3101 if (do_swap_account
&& swapout
&& memcg
)
3102 swap_cgroup_record(ent
, css_id(&memcg
->css
));
3106 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
3108 * called from swap_entry_free(). remove record in swap_cgroup and
3109 * uncharge "memsw" account.
3111 void mem_cgroup_uncharge_swap(swp_entry_t ent
)
3113 struct mem_cgroup
*memcg
;
3116 if (!do_swap_account
)
3119 id
= swap_cgroup_record(ent
, 0);
3121 memcg
= mem_cgroup_lookup(id
);
3124 * We uncharge this because swap is freed.
3125 * This memcg can be obsolete one. We avoid calling css_tryget
3127 if (!mem_cgroup_is_root(memcg
))
3128 res_counter_uncharge(&memcg
->memsw
, PAGE_SIZE
);
3129 mem_cgroup_swap_statistics(memcg
, false);
3130 mem_cgroup_put(memcg
);
3136 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
3137 * @entry: swap entry to be moved
3138 * @from: mem_cgroup which the entry is moved from
3139 * @to: mem_cgroup which the entry is moved to
3141 * It succeeds only when the swap_cgroup's record for this entry is the same
3142 * as the mem_cgroup's id of @from.
3144 * Returns 0 on success, -EINVAL on failure.
3146 * The caller must have charged to @to, IOW, called res_counter_charge() about
3147 * both res and memsw, and called css_get().
3149 static int mem_cgroup_move_swap_account(swp_entry_t entry
,
3150 struct mem_cgroup
*from
, struct mem_cgroup
*to
)
3152 unsigned short old_id
, new_id
;
3154 old_id
= css_id(&from
->css
);
3155 new_id
= css_id(&to
->css
);
3157 if (swap_cgroup_cmpxchg(entry
, old_id
, new_id
) == old_id
) {
3158 mem_cgroup_swap_statistics(from
, false);
3159 mem_cgroup_swap_statistics(to
, true);
3161 * This function is only called from task migration context now.
3162 * It postpones res_counter and refcount handling till the end
3163 * of task migration(mem_cgroup_clear_mc()) for performance
3164 * improvement. But we cannot postpone mem_cgroup_get(to)
3165 * because if the process that has been moved to @to does
3166 * swap-in, the refcount of @to might be decreased to 0.
3174 static inline int mem_cgroup_move_swap_account(swp_entry_t entry
,
3175 struct mem_cgroup
*from
, struct mem_cgroup
*to
)
3182 * Before starting migration, account PAGE_SIZE to mem_cgroup that the old
3185 int mem_cgroup_prepare_migration(struct page
*page
,
3186 struct page
*newpage
, struct mem_cgroup
**memcgp
, gfp_t gfp_mask
)
3188 struct mem_cgroup
*memcg
= NULL
;
3189 struct page_cgroup
*pc
;
3190 enum charge_type ctype
;
3195 VM_BUG_ON(PageTransHuge(page
));
3196 if (mem_cgroup_disabled())
3199 pc
= lookup_page_cgroup(page
);
3200 lock_page_cgroup(pc
);
3201 if (PageCgroupUsed(pc
)) {
3202 memcg
= pc
->mem_cgroup
;
3203 css_get(&memcg
->css
);
3205 * At migrating an anonymous page, its mapcount goes down
3206 * to 0 and uncharge() will be called. But, even if it's fully
3207 * unmapped, migration may fail and this page has to be
3208 * charged again. We set MIGRATION flag here and delay uncharge
3209 * until end_migration() is called
3211 * Corner Case Thinking
3213 * When the old page was mapped as Anon and it's unmap-and-freed
3214 * while migration was ongoing.
3215 * If unmap finds the old page, uncharge() of it will be delayed
3216 * until end_migration(). If unmap finds a new page, it's
3217 * uncharged when it make mapcount to be 1->0. If unmap code
3218 * finds swap_migration_entry, the new page will not be mapped
3219 * and end_migration() will find it(mapcount==0).
3222 * When the old page was mapped but migraion fails, the kernel
3223 * remaps it. A charge for it is kept by MIGRATION flag even
3224 * if mapcount goes down to 0. We can do remap successfully
3225 * without charging it again.
3228 * The "old" page is under lock_page() until the end of
3229 * migration, so, the old page itself will not be swapped-out.
3230 * If the new page is swapped out before end_migraton, our
3231 * hook to usual swap-out path will catch the event.
3234 SetPageCgroupMigration(pc
);
3236 unlock_page_cgroup(pc
);
3238 * If the page is not charged at this point,
3245 ret
= __mem_cgroup_try_charge(NULL
, gfp_mask
, 1, memcgp
, false);
3246 css_put(&memcg
->css
);/* drop extra refcnt */
3248 if (PageAnon(page
)) {
3249 lock_page_cgroup(pc
);
3250 ClearPageCgroupMigration(pc
);
3251 unlock_page_cgroup(pc
);
3253 * The old page may be fully unmapped while we kept it.
3255 mem_cgroup_uncharge_page(page
);
3257 /* we'll need to revisit this error code (we have -EINTR) */
3261 * We charge new page before it's used/mapped. So, even if unlock_page()
3262 * is called before end_migration, we can catch all events on this new
3263 * page. In the case new page is migrated but not remapped, new page's
3264 * mapcount will be finally 0 and we call uncharge in end_migration().
3267 ctype
= MEM_CGROUP_CHARGE_TYPE_MAPPED
;
3268 else if (page_is_file_cache(page
))
3269 ctype
= MEM_CGROUP_CHARGE_TYPE_CACHE
;
3271 ctype
= MEM_CGROUP_CHARGE_TYPE_SHMEM
;
3272 __mem_cgroup_commit_charge(memcg
, newpage
, 1, ctype
, false);
3276 /* remove redundant charge if migration failed*/
3277 void mem_cgroup_end_migration(struct mem_cgroup
*memcg
,
3278 struct page
*oldpage
, struct page
*newpage
, bool migration_ok
)
3280 struct page
*used
, *unused
;
3281 struct page_cgroup
*pc
;
3286 /* blocks rmdir() */
3287 cgroup_exclude_rmdir(&memcg
->css
);
3288 if (!migration_ok
) {
3296 * We disallowed uncharge of pages under migration because mapcount
3297 * of the page goes down to zero, temporarly.
3298 * Clear the flag and check the page should be charged.
3300 pc
= lookup_page_cgroup(oldpage
);
3301 lock_page_cgroup(pc
);
3302 ClearPageCgroupMigration(pc
);
3303 unlock_page_cgroup(pc
);
3304 anon
= PageAnon(used
);
3305 __mem_cgroup_uncharge_common(unused
,
3306 anon
? MEM_CGROUP_CHARGE_TYPE_MAPPED
3307 : MEM_CGROUP_CHARGE_TYPE_CACHE
);
3310 * If a page is a file cache, radix-tree replacement is very atomic
3311 * and we can skip this check. When it was an Anon page, its mapcount
3312 * goes down to 0. But because we added MIGRATION flage, it's not
3313 * uncharged yet. There are several case but page->mapcount check
3314 * and USED bit check in mem_cgroup_uncharge_page() will do enough
3315 * check. (see prepare_charge() also)
3318 mem_cgroup_uncharge_page(used
);
3320 * At migration, we may charge account against cgroup which has no
3322 * So, rmdir()->pre_destroy() can be called while we do this charge.
3323 * In that case, we need to call pre_destroy() again. check it here.
3325 cgroup_release_and_wakeup_rmdir(&memcg
->css
);
3329 * At replace page cache, newpage is not under any memcg but it's on
3330 * LRU. So, this function doesn't touch res_counter but handles LRU
3331 * in correct way. Both pages are locked so we cannot race with uncharge.
3333 void mem_cgroup_replace_page_cache(struct page
*oldpage
,
3334 struct page
*newpage
)
3336 struct mem_cgroup
*memcg
= NULL
;
3337 struct page_cgroup
*pc
;
3338 enum charge_type type
= MEM_CGROUP_CHARGE_TYPE_CACHE
;
3340 if (mem_cgroup_disabled())
3343 pc
= lookup_page_cgroup(oldpage
);
3344 /* fix accounting on old pages */
3345 lock_page_cgroup(pc
);
3346 if (PageCgroupUsed(pc
)) {
3347 memcg
= pc
->mem_cgroup
;
3348 mem_cgroup_charge_statistics(memcg
, false, -1);
3349 ClearPageCgroupUsed(pc
);
3351 unlock_page_cgroup(pc
);
3354 * When called from shmem_replace_page(), in some cases the
3355 * oldpage has already been charged, and in some cases not.
3360 if (PageSwapBacked(oldpage
))
3361 type
= MEM_CGROUP_CHARGE_TYPE_SHMEM
;
3364 * Even if newpage->mapping was NULL before starting replacement,
3365 * the newpage may be on LRU(or pagevec for LRU) already. We lock
3366 * LRU while we overwrite pc->mem_cgroup.
3368 __mem_cgroup_commit_charge(memcg
, newpage
, 1, type
, true);
3371 #ifdef CONFIG_DEBUG_VM
3372 static struct page_cgroup
*lookup_page_cgroup_used(struct page
*page
)
3374 struct page_cgroup
*pc
;
3376 pc
= lookup_page_cgroup(page
);
3378 * Can be NULL while feeding pages into the page allocator for
3379 * the first time, i.e. during boot or memory hotplug;
3380 * or when mem_cgroup_disabled().
3382 if (likely(pc
) && PageCgroupUsed(pc
))
3387 bool mem_cgroup_bad_page_check(struct page
*page
)
3389 if (mem_cgroup_disabled())
3392 return lookup_page_cgroup_used(page
) != NULL
;
3395 void mem_cgroup_print_bad_page(struct page
*page
)
3397 struct page_cgroup
*pc
;
3399 pc
= lookup_page_cgroup_used(page
);
3401 printk(KERN_ALERT
"pc:%p pc->flags:%lx pc->mem_cgroup:%p\n",
3402 pc
, pc
->flags
, pc
->mem_cgroup
);
3407 static DEFINE_MUTEX(set_limit_mutex
);
3409 static int mem_cgroup_resize_limit(struct mem_cgroup
*memcg
,
3410 unsigned long long val
)
3413 u64 memswlimit
, memlimit
;
3415 int children
= mem_cgroup_count_children(memcg
);
3416 u64 curusage
, oldusage
;
3420 * For keeping hierarchical_reclaim simple, how long we should retry
3421 * is depends on callers. We set our retry-count to be function
3422 * of # of children which we should visit in this loop.
3424 retry_count
= MEM_CGROUP_RECLAIM_RETRIES
* children
;
3426 oldusage
= res_counter_read_u64(&memcg
->res
, RES_USAGE
);
3429 while (retry_count
) {
3430 if (signal_pending(current
)) {
3435 * Rather than hide all in some function, I do this in
3436 * open coded manner. You see what this really does.
3437 * We have to guarantee memcg->res.limit < memcg->memsw.limit.
3439 mutex_lock(&set_limit_mutex
);
3440 memswlimit
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
3441 if (memswlimit
< val
) {
3443 mutex_unlock(&set_limit_mutex
);
3447 memlimit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
3451 ret
= res_counter_set_limit(&memcg
->res
, val
);
3453 if (memswlimit
== val
)
3454 memcg
->memsw_is_minimum
= true;
3456 memcg
->memsw_is_minimum
= false;
3458 mutex_unlock(&set_limit_mutex
);
3463 mem_cgroup_reclaim(memcg
, GFP_KERNEL
,
3464 MEM_CGROUP_RECLAIM_SHRINK
);
3465 curusage
= res_counter_read_u64(&memcg
->res
, RES_USAGE
);
3466 /* Usage is reduced ? */
3467 if (curusage
>= oldusage
)
3470 oldusage
= curusage
;
3472 if (!ret
&& enlarge
)
3473 memcg_oom_recover(memcg
);
3478 static int mem_cgroup_resize_memsw_limit(struct mem_cgroup
*memcg
,
3479 unsigned long long val
)
3482 u64 memlimit
, memswlimit
, oldusage
, curusage
;
3483 int children
= mem_cgroup_count_children(memcg
);
3487 /* see mem_cgroup_resize_res_limit */
3488 retry_count
= children
* MEM_CGROUP_RECLAIM_RETRIES
;
3489 oldusage
= res_counter_read_u64(&memcg
->memsw
, RES_USAGE
);
3490 while (retry_count
) {
3491 if (signal_pending(current
)) {
3496 * Rather than hide all in some function, I do this in
3497 * open coded manner. You see what this really does.
3498 * We have to guarantee memcg->res.limit < memcg->memsw.limit.
3500 mutex_lock(&set_limit_mutex
);
3501 memlimit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
3502 if (memlimit
> val
) {
3504 mutex_unlock(&set_limit_mutex
);
3507 memswlimit
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
3508 if (memswlimit
< val
)
3510 ret
= res_counter_set_limit(&memcg
->memsw
, val
);
3512 if (memlimit
== val
)
3513 memcg
->memsw_is_minimum
= true;
3515 memcg
->memsw_is_minimum
= false;
3517 mutex_unlock(&set_limit_mutex
);
3522 mem_cgroup_reclaim(memcg
, GFP_KERNEL
,
3523 MEM_CGROUP_RECLAIM_NOSWAP
|
3524 MEM_CGROUP_RECLAIM_SHRINK
);
3525 curusage
= res_counter_read_u64(&memcg
->memsw
, RES_USAGE
);
3526 /* Usage is reduced ? */
3527 if (curusage
>= oldusage
)
3530 oldusage
= curusage
;
3532 if (!ret
&& enlarge
)
3533 memcg_oom_recover(memcg
);
3537 unsigned long mem_cgroup_soft_limit_reclaim(struct zone
*zone
, int order
,
3539 unsigned long *total_scanned
)
3541 unsigned long nr_reclaimed
= 0;
3542 struct mem_cgroup_per_zone
*mz
, *next_mz
= NULL
;
3543 unsigned long reclaimed
;
3545 struct mem_cgroup_tree_per_zone
*mctz
;
3546 unsigned long long excess
;
3547 unsigned long nr_scanned
;
3552 mctz
= soft_limit_tree_node_zone(zone_to_nid(zone
), zone_idx(zone
));
3554 * This loop can run a while, specially if mem_cgroup's continuously
3555 * keep exceeding their soft limit and putting the system under
3562 mz
= mem_cgroup_largest_soft_limit_node(mctz
);
3567 reclaimed
= mem_cgroup_soft_reclaim(mz
->memcg
, zone
,
3568 gfp_mask
, &nr_scanned
);
3569 nr_reclaimed
+= reclaimed
;
3570 *total_scanned
+= nr_scanned
;
3571 spin_lock(&mctz
->lock
);
3574 * If we failed to reclaim anything from this memory cgroup
3575 * it is time to move on to the next cgroup
3581 * Loop until we find yet another one.
3583 * By the time we get the soft_limit lock
3584 * again, someone might have aded the
3585 * group back on the RB tree. Iterate to
3586 * make sure we get a different mem.
3587 * mem_cgroup_largest_soft_limit_node returns
3588 * NULL if no other cgroup is present on
3592 __mem_cgroup_largest_soft_limit_node(mctz
);
3594 css_put(&next_mz
->memcg
->css
);
3595 else /* next_mz == NULL or other memcg */
3599 __mem_cgroup_remove_exceeded(mz
->memcg
, mz
, mctz
);
3600 excess
= res_counter_soft_limit_excess(&mz
->memcg
->res
);
3602 * One school of thought says that we should not add
3603 * back the node to the tree if reclaim returns 0.
3604 * But our reclaim could return 0, simply because due
3605 * to priority we are exposing a smaller subset of
3606 * memory to reclaim from. Consider this as a longer
3609 /* If excess == 0, no tree ops */
3610 __mem_cgroup_insert_exceeded(mz
->memcg
, mz
, mctz
, excess
);
3611 spin_unlock(&mctz
->lock
);
3612 css_put(&mz
->memcg
->css
);
3615 * Could not reclaim anything and there are no more
3616 * mem cgroups to try or we seem to be looping without
3617 * reclaiming anything.
3619 if (!nr_reclaimed
&&
3621 loop
> MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS
))
3623 } while (!nr_reclaimed
);
3625 css_put(&next_mz
->memcg
->css
);
3626 return nr_reclaimed
;
3630 * This routine traverse page_cgroup in given list and drop them all.
3631 * *And* this routine doesn't reclaim page itself, just removes page_cgroup.
3633 static int mem_cgroup_force_empty_list(struct mem_cgroup
*memcg
,
3634 int node
, int zid
, enum lru_list lru
)
3636 struct mem_cgroup_per_zone
*mz
;
3637 unsigned long flags
, loop
;
3638 struct list_head
*list
;
3643 zone
= &NODE_DATA(node
)->node_zones
[zid
];
3644 mz
= mem_cgroup_zoneinfo(memcg
, node
, zid
);
3645 list
= &mz
->lruvec
.lists
[lru
];
3647 loop
= mz
->lru_size
[lru
];
3648 /* give some margin against EBUSY etc...*/
3652 struct page_cgroup
*pc
;
3656 spin_lock_irqsave(&zone
->lru_lock
, flags
);
3657 if (list_empty(list
)) {
3658 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
3661 page
= list_entry(list
->prev
, struct page
, lru
);
3663 list_move(&page
->lru
, list
);
3665 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
3668 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
3670 pc
= lookup_page_cgroup(page
);
3672 ret
= mem_cgroup_move_parent(page
, pc
, memcg
, GFP_KERNEL
);
3673 if (ret
== -ENOMEM
|| ret
== -EINTR
)
3676 if (ret
== -EBUSY
|| ret
== -EINVAL
) {
3677 /* found lock contention or "pc" is obsolete. */
3684 if (!ret
&& !list_empty(list
))
3690 * make mem_cgroup's charge to be 0 if there is no task.
3691 * This enables deleting this mem_cgroup.
3693 static int mem_cgroup_force_empty(struct mem_cgroup
*memcg
, bool free_all
)
3696 int node
, zid
, shrink
;
3697 int nr_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
3698 struct cgroup
*cgrp
= memcg
->css
.cgroup
;
3700 css_get(&memcg
->css
);
3703 /* should free all ? */
3709 if (cgroup_task_count(cgrp
) || !list_empty(&cgrp
->children
))
3712 if (signal_pending(current
))
3714 /* This is for making all *used* pages to be on LRU. */
3715 lru_add_drain_all();
3716 drain_all_stock_sync(memcg
);
3718 mem_cgroup_start_move(memcg
);
3719 for_each_node_state(node
, N_HIGH_MEMORY
) {
3720 for (zid
= 0; !ret
&& zid
< MAX_NR_ZONES
; zid
++) {
3723 ret
= mem_cgroup_force_empty_list(memcg
,
3732 mem_cgroup_end_move(memcg
);
3733 memcg_oom_recover(memcg
);
3734 /* it seems parent cgroup doesn't have enough mem */
3738 /* "ret" should also be checked to ensure all lists are empty. */
3739 } while (res_counter_read_u64(&memcg
->res
, RES_USAGE
) > 0 || ret
);
3741 css_put(&memcg
->css
);
3745 /* returns EBUSY if there is a task or if we come here twice. */
3746 if (cgroup_task_count(cgrp
) || !list_empty(&cgrp
->children
) || shrink
) {
3750 /* we call try-to-free pages for make this cgroup empty */
3751 lru_add_drain_all();
3752 /* try to free all pages in this cgroup */
3754 while (nr_retries
&& res_counter_read_u64(&memcg
->res
, RES_USAGE
) > 0) {
3757 if (signal_pending(current
)) {
3761 progress
= try_to_free_mem_cgroup_pages(memcg
, GFP_KERNEL
,
3765 /* maybe some writeback is necessary */
3766 congestion_wait(BLK_RW_ASYNC
, HZ
/10);
3771 /* try move_account...there may be some *locked* pages. */
3775 int mem_cgroup_force_empty_write(struct cgroup
*cont
, unsigned int event
)
3777 return mem_cgroup_force_empty(mem_cgroup_from_cont(cont
), true);
3781 static u64
mem_cgroup_hierarchy_read(struct cgroup
*cont
, struct cftype
*cft
)
3783 return mem_cgroup_from_cont(cont
)->use_hierarchy
;
3786 static int mem_cgroup_hierarchy_write(struct cgroup
*cont
, struct cftype
*cft
,
3790 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
3791 struct cgroup
*parent
= cont
->parent
;
3792 struct mem_cgroup
*parent_memcg
= NULL
;
3795 parent_memcg
= mem_cgroup_from_cont(parent
);
3799 * If parent's use_hierarchy is set, we can't make any modifications
3800 * in the child subtrees. If it is unset, then the change can
3801 * occur, provided the current cgroup has no children.
3803 * For the root cgroup, parent_mem is NULL, we allow value to be
3804 * set if there are no children.
3806 if ((!parent_memcg
|| !parent_memcg
->use_hierarchy
) &&
3807 (val
== 1 || val
== 0)) {
3808 if (list_empty(&cont
->children
))
3809 memcg
->use_hierarchy
= val
;
3820 static unsigned long mem_cgroup_recursive_stat(struct mem_cgroup
*memcg
,
3821 enum mem_cgroup_stat_index idx
)
3823 struct mem_cgroup
*iter
;
3826 /* Per-cpu values can be negative, use a signed accumulator */
3827 for_each_mem_cgroup_tree(iter
, memcg
)
3828 val
+= mem_cgroup_read_stat(iter
, idx
);
3830 if (val
< 0) /* race ? */
3835 static inline u64
mem_cgroup_usage(struct mem_cgroup
*memcg
, bool swap
)
3839 if (!mem_cgroup_is_root(memcg
)) {
3841 return res_counter_read_u64(&memcg
->res
, RES_USAGE
);
3843 return res_counter_read_u64(&memcg
->memsw
, RES_USAGE
);
3846 val
= mem_cgroup_recursive_stat(memcg
, MEM_CGROUP_STAT_CACHE
);
3847 val
+= mem_cgroup_recursive_stat(memcg
, MEM_CGROUP_STAT_RSS
);
3850 val
+= mem_cgroup_recursive_stat(memcg
, MEM_CGROUP_STAT_SWAPOUT
);
3852 return val
<< PAGE_SHIFT
;
3855 static ssize_t
mem_cgroup_read(struct cgroup
*cont
, struct cftype
*cft
,
3856 struct file
*file
, char __user
*buf
,
3857 size_t nbytes
, loff_t
*ppos
)
3859 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
3862 int type
, name
, len
;
3864 type
= MEMFILE_TYPE(cft
->private);
3865 name
= MEMFILE_ATTR(cft
->private);
3867 if (!do_swap_account
&& type
== _MEMSWAP
)
3872 if (name
== RES_USAGE
)
3873 val
= mem_cgroup_usage(memcg
, false);
3875 val
= res_counter_read_u64(&memcg
->res
, name
);
3878 if (name
== RES_USAGE
)
3879 val
= mem_cgroup_usage(memcg
, true);
3881 val
= res_counter_read_u64(&memcg
->memsw
, name
);
3887 len
= scnprintf(str
, sizeof(str
), "%llu\n", (unsigned long long)val
);
3888 return simple_read_from_buffer(buf
, nbytes
, ppos
, str
, len
);
3891 * The user of this function is...
3894 static int mem_cgroup_write(struct cgroup
*cont
, struct cftype
*cft
,
3897 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
3899 unsigned long long val
;
3902 type
= MEMFILE_TYPE(cft
->private);
3903 name
= MEMFILE_ATTR(cft
->private);
3905 if (!do_swap_account
&& type
== _MEMSWAP
)
3910 if (mem_cgroup_is_root(memcg
)) { /* Can't set limit on root */
3914 /* This function does all necessary parse...reuse it */
3915 ret
= res_counter_memparse_write_strategy(buffer
, &val
);
3919 ret
= mem_cgroup_resize_limit(memcg
, val
);
3921 ret
= mem_cgroup_resize_memsw_limit(memcg
, val
);
3923 case RES_SOFT_LIMIT
:
3924 ret
= res_counter_memparse_write_strategy(buffer
, &val
);
3928 * For memsw, soft limits are hard to implement in terms
3929 * of semantics, for now, we support soft limits for
3930 * control without swap
3933 ret
= res_counter_set_soft_limit(&memcg
->res
, val
);
3938 ret
= -EINVAL
; /* should be BUG() ? */
3944 static void memcg_get_hierarchical_limit(struct mem_cgroup
*memcg
,
3945 unsigned long long *mem_limit
, unsigned long long *memsw_limit
)
3947 struct cgroup
*cgroup
;
3948 unsigned long long min_limit
, min_memsw_limit
, tmp
;
3950 min_limit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
3951 min_memsw_limit
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
3952 cgroup
= memcg
->css
.cgroup
;
3953 if (!memcg
->use_hierarchy
)
3956 while (cgroup
->parent
) {
3957 cgroup
= cgroup
->parent
;
3958 memcg
= mem_cgroup_from_cont(cgroup
);
3959 if (!memcg
->use_hierarchy
)
3961 tmp
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
3962 min_limit
= min(min_limit
, tmp
);
3963 tmp
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
3964 min_memsw_limit
= min(min_memsw_limit
, tmp
);
3967 *mem_limit
= min_limit
;
3968 *memsw_limit
= min_memsw_limit
;
3971 static int mem_cgroup_reset(struct cgroup
*cont
, unsigned int event
)
3973 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
3976 type
= MEMFILE_TYPE(event
);
3977 name
= MEMFILE_ATTR(event
);
3979 if (!do_swap_account
&& type
== _MEMSWAP
)
3985 res_counter_reset_max(&memcg
->res
);
3987 res_counter_reset_max(&memcg
->memsw
);
3991 res_counter_reset_failcnt(&memcg
->res
);
3993 res_counter_reset_failcnt(&memcg
->memsw
);
4000 static u64
mem_cgroup_move_charge_read(struct cgroup
*cgrp
,
4003 return mem_cgroup_from_cont(cgrp
)->move_charge_at_immigrate
;
4007 static int mem_cgroup_move_charge_write(struct cgroup
*cgrp
,
4008 struct cftype
*cft
, u64 val
)
4010 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4012 if (val
>= (1 << NR_MOVE_TYPE
))
4015 * We check this value several times in both in can_attach() and
4016 * attach(), so we need cgroup lock to prevent this value from being
4020 memcg
->move_charge_at_immigrate
= val
;
4026 static int mem_cgroup_move_charge_write(struct cgroup
*cgrp
,
4027 struct cftype
*cft
, u64 val
)
4034 /* For read statistics */
4052 struct mcs_total_stat
{
4053 s64 stat
[NR_MCS_STAT
];
4059 } memcg_stat_strings
[NR_MCS_STAT
] = {
4060 {"cache", "total_cache"},
4061 {"rss", "total_rss"},
4062 {"mapped_file", "total_mapped_file"},
4063 {"pgpgin", "total_pgpgin"},
4064 {"pgpgout", "total_pgpgout"},
4065 {"swap", "total_swap"},
4066 {"pgfault", "total_pgfault"},
4067 {"pgmajfault", "total_pgmajfault"},
4068 {"inactive_anon", "total_inactive_anon"},
4069 {"active_anon", "total_active_anon"},
4070 {"inactive_file", "total_inactive_file"},
4071 {"active_file", "total_active_file"},
4072 {"unevictable", "total_unevictable"}
4077 mem_cgroup_get_local_stat(struct mem_cgroup
*memcg
, struct mcs_total_stat
*s
)
4082 val
= mem_cgroup_read_stat(memcg
, MEM_CGROUP_STAT_CACHE
);
4083 s
->stat
[MCS_CACHE
] += val
* PAGE_SIZE
;
4084 val
= mem_cgroup_read_stat(memcg
, MEM_CGROUP_STAT_RSS
);
4085 s
->stat
[MCS_RSS
] += val
* PAGE_SIZE
;
4086 val
= mem_cgroup_read_stat(memcg
, MEM_CGROUP_STAT_FILE_MAPPED
);
4087 s
->stat
[MCS_FILE_MAPPED
] += val
* PAGE_SIZE
;
4088 val
= mem_cgroup_read_events(memcg
, MEM_CGROUP_EVENTS_PGPGIN
);
4089 s
->stat
[MCS_PGPGIN
] += val
;
4090 val
= mem_cgroup_read_events(memcg
, MEM_CGROUP_EVENTS_PGPGOUT
);
4091 s
->stat
[MCS_PGPGOUT
] += val
;
4092 if (do_swap_account
) {
4093 val
= mem_cgroup_read_stat(memcg
, MEM_CGROUP_STAT_SWAPOUT
);
4094 s
->stat
[MCS_SWAP
] += val
* PAGE_SIZE
;
4096 val
= mem_cgroup_read_events(memcg
, MEM_CGROUP_EVENTS_PGFAULT
);
4097 s
->stat
[MCS_PGFAULT
] += val
;
4098 val
= mem_cgroup_read_events(memcg
, MEM_CGROUP_EVENTS_PGMAJFAULT
);
4099 s
->stat
[MCS_PGMAJFAULT
] += val
;
4102 val
= mem_cgroup_nr_lru_pages(memcg
, BIT(LRU_INACTIVE_ANON
));
4103 s
->stat
[MCS_INACTIVE_ANON
] += val
* PAGE_SIZE
;
4104 val
= mem_cgroup_nr_lru_pages(memcg
, BIT(LRU_ACTIVE_ANON
));
4105 s
->stat
[MCS_ACTIVE_ANON
] += val
* PAGE_SIZE
;
4106 val
= mem_cgroup_nr_lru_pages(memcg
, BIT(LRU_INACTIVE_FILE
));
4107 s
->stat
[MCS_INACTIVE_FILE
] += val
* PAGE_SIZE
;
4108 val
= mem_cgroup_nr_lru_pages(memcg
, BIT(LRU_ACTIVE_FILE
));
4109 s
->stat
[MCS_ACTIVE_FILE
] += val
* PAGE_SIZE
;
4110 val
= mem_cgroup_nr_lru_pages(memcg
, BIT(LRU_UNEVICTABLE
));
4111 s
->stat
[MCS_UNEVICTABLE
] += val
* PAGE_SIZE
;
4115 mem_cgroup_get_total_stat(struct mem_cgroup
*memcg
, struct mcs_total_stat
*s
)
4117 struct mem_cgroup
*iter
;
4119 for_each_mem_cgroup_tree(iter
, memcg
)
4120 mem_cgroup_get_local_stat(iter
, s
);
4124 static int mem_control_numa_stat_show(struct seq_file
*m
, void *arg
)
4127 unsigned long total_nr
, file_nr
, anon_nr
, unevictable_nr
;
4128 unsigned long node_nr
;
4129 struct cgroup
*cont
= m
->private;
4130 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
4132 total_nr
= mem_cgroup_nr_lru_pages(memcg
, LRU_ALL
);
4133 seq_printf(m
, "total=%lu", total_nr
);
4134 for_each_node_state(nid
, N_HIGH_MEMORY
) {
4135 node_nr
= mem_cgroup_node_nr_lru_pages(memcg
, nid
, LRU_ALL
);
4136 seq_printf(m
, " N%d=%lu", nid
, node_nr
);
4140 file_nr
= mem_cgroup_nr_lru_pages(memcg
, LRU_ALL_FILE
);
4141 seq_printf(m
, "file=%lu", file_nr
);
4142 for_each_node_state(nid
, N_HIGH_MEMORY
) {
4143 node_nr
= mem_cgroup_node_nr_lru_pages(memcg
, nid
,
4145 seq_printf(m
, " N%d=%lu", nid
, node_nr
);
4149 anon_nr
= mem_cgroup_nr_lru_pages(memcg
, LRU_ALL_ANON
);
4150 seq_printf(m
, "anon=%lu", anon_nr
);
4151 for_each_node_state(nid
, N_HIGH_MEMORY
) {
4152 node_nr
= mem_cgroup_node_nr_lru_pages(memcg
, nid
,
4154 seq_printf(m
, " N%d=%lu", nid
, node_nr
);
4158 unevictable_nr
= mem_cgroup_nr_lru_pages(memcg
, BIT(LRU_UNEVICTABLE
));
4159 seq_printf(m
, "unevictable=%lu", unevictable_nr
);
4160 for_each_node_state(nid
, N_HIGH_MEMORY
) {
4161 node_nr
= mem_cgroup_node_nr_lru_pages(memcg
, nid
,
4162 BIT(LRU_UNEVICTABLE
));
4163 seq_printf(m
, " N%d=%lu", nid
, node_nr
);
4168 #endif /* CONFIG_NUMA */
4170 static int mem_control_stat_show(struct cgroup
*cont
, struct cftype
*cft
,
4171 struct cgroup_map_cb
*cb
)
4173 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
4174 struct mcs_total_stat mystat
;
4177 memset(&mystat
, 0, sizeof(mystat
));
4178 mem_cgroup_get_local_stat(memcg
, &mystat
);
4181 for (i
= 0; i
< NR_MCS_STAT
; i
++) {
4182 if (i
== MCS_SWAP
&& !do_swap_account
)
4184 cb
->fill(cb
, memcg_stat_strings
[i
].local_name
, mystat
.stat
[i
]);
4187 /* Hierarchical information */
4189 unsigned long long limit
, memsw_limit
;
4190 memcg_get_hierarchical_limit(memcg
, &limit
, &memsw_limit
);
4191 cb
->fill(cb
, "hierarchical_memory_limit", limit
);
4192 if (do_swap_account
)
4193 cb
->fill(cb
, "hierarchical_memsw_limit", memsw_limit
);
4196 memset(&mystat
, 0, sizeof(mystat
));
4197 mem_cgroup_get_total_stat(memcg
, &mystat
);
4198 for (i
= 0; i
< NR_MCS_STAT
; i
++) {
4199 if (i
== MCS_SWAP
&& !do_swap_account
)
4201 cb
->fill(cb
, memcg_stat_strings
[i
].total_name
, mystat
.stat
[i
]);
4204 #ifdef CONFIG_DEBUG_VM
4207 struct mem_cgroup_per_zone
*mz
;
4208 struct zone_reclaim_stat
*rstat
;
4209 unsigned long recent_rotated
[2] = {0, 0};
4210 unsigned long recent_scanned
[2] = {0, 0};
4212 for_each_online_node(nid
)
4213 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++) {
4214 mz
= mem_cgroup_zoneinfo(memcg
, nid
, zid
);
4215 rstat
= &mz
->lruvec
.reclaim_stat
;
4217 recent_rotated
[0] += rstat
->recent_rotated
[0];
4218 recent_rotated
[1] += rstat
->recent_rotated
[1];
4219 recent_scanned
[0] += rstat
->recent_scanned
[0];
4220 recent_scanned
[1] += rstat
->recent_scanned
[1];
4222 cb
->fill(cb
, "recent_rotated_anon", recent_rotated
[0]);
4223 cb
->fill(cb
, "recent_rotated_file", recent_rotated
[1]);
4224 cb
->fill(cb
, "recent_scanned_anon", recent_scanned
[0]);
4225 cb
->fill(cb
, "recent_scanned_file", recent_scanned
[1]);
4232 static u64
mem_cgroup_swappiness_read(struct cgroup
*cgrp
, struct cftype
*cft
)
4234 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4236 return mem_cgroup_swappiness(memcg
);
4239 static int mem_cgroup_swappiness_write(struct cgroup
*cgrp
, struct cftype
*cft
,
4242 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4243 struct mem_cgroup
*parent
;
4248 if (cgrp
->parent
== NULL
)
4251 parent
= mem_cgroup_from_cont(cgrp
->parent
);
4255 /* If under hierarchy, only empty-root can set this value */
4256 if ((parent
->use_hierarchy
) ||
4257 (memcg
->use_hierarchy
&& !list_empty(&cgrp
->children
))) {
4262 memcg
->swappiness
= val
;
4269 static void __mem_cgroup_threshold(struct mem_cgroup
*memcg
, bool swap
)
4271 struct mem_cgroup_threshold_ary
*t
;
4277 t
= rcu_dereference(memcg
->thresholds
.primary
);
4279 t
= rcu_dereference(memcg
->memsw_thresholds
.primary
);
4284 usage
= mem_cgroup_usage(memcg
, swap
);
4287 * current_threshold points to threshold just below usage.
4288 * If it's not true, a threshold was crossed after last
4289 * call of __mem_cgroup_threshold().
4291 i
= t
->current_threshold
;
4294 * Iterate backward over array of thresholds starting from
4295 * current_threshold and check if a threshold is crossed.
4296 * If none of thresholds below usage is crossed, we read
4297 * only one element of the array here.
4299 for (; i
>= 0 && unlikely(t
->entries
[i
].threshold
> usage
); i
--)
4300 eventfd_signal(t
->entries
[i
].eventfd
, 1);
4302 /* i = current_threshold + 1 */
4306 * Iterate forward over array of thresholds starting from
4307 * current_threshold+1 and check if a threshold is crossed.
4308 * If none of thresholds above usage is crossed, we read
4309 * only one element of the array here.
4311 for (; i
< t
->size
&& unlikely(t
->entries
[i
].threshold
<= usage
); i
++)
4312 eventfd_signal(t
->entries
[i
].eventfd
, 1);
4314 /* Update current_threshold */
4315 t
->current_threshold
= i
- 1;
4320 static void mem_cgroup_threshold(struct mem_cgroup
*memcg
)
4323 __mem_cgroup_threshold(memcg
, false);
4324 if (do_swap_account
)
4325 __mem_cgroup_threshold(memcg
, true);
4327 memcg
= parent_mem_cgroup(memcg
);
4331 static int compare_thresholds(const void *a
, const void *b
)
4333 const struct mem_cgroup_threshold
*_a
= a
;
4334 const struct mem_cgroup_threshold
*_b
= b
;
4336 return _a
->threshold
- _b
->threshold
;
4339 static int mem_cgroup_oom_notify_cb(struct mem_cgroup
*memcg
)
4341 struct mem_cgroup_eventfd_list
*ev
;
4343 list_for_each_entry(ev
, &memcg
->oom_notify
, list
)
4344 eventfd_signal(ev
->eventfd
, 1);
4348 static void mem_cgroup_oom_notify(struct mem_cgroup
*memcg
)
4350 struct mem_cgroup
*iter
;
4352 for_each_mem_cgroup_tree(iter
, memcg
)
4353 mem_cgroup_oom_notify_cb(iter
);
4356 static int mem_cgroup_usage_register_event(struct cgroup
*cgrp
,
4357 struct cftype
*cft
, struct eventfd_ctx
*eventfd
, const char *args
)
4359 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4360 struct mem_cgroup_thresholds
*thresholds
;
4361 struct mem_cgroup_threshold_ary
*new;
4362 int type
= MEMFILE_TYPE(cft
->private);
4363 u64 threshold
, usage
;
4366 ret
= res_counter_memparse_write_strategy(args
, &threshold
);
4370 mutex_lock(&memcg
->thresholds_lock
);
4373 thresholds
= &memcg
->thresholds
;
4374 else if (type
== _MEMSWAP
)
4375 thresholds
= &memcg
->memsw_thresholds
;
4379 usage
= mem_cgroup_usage(memcg
, type
== _MEMSWAP
);
4381 /* Check if a threshold crossed before adding a new one */
4382 if (thresholds
->primary
)
4383 __mem_cgroup_threshold(memcg
, type
== _MEMSWAP
);
4385 size
= thresholds
->primary
? thresholds
->primary
->size
+ 1 : 1;
4387 /* Allocate memory for new array of thresholds */
4388 new = kmalloc(sizeof(*new) + size
* sizeof(struct mem_cgroup_threshold
),
4396 /* Copy thresholds (if any) to new array */
4397 if (thresholds
->primary
) {
4398 memcpy(new->entries
, thresholds
->primary
->entries
, (size
- 1) *
4399 sizeof(struct mem_cgroup_threshold
));
4402 /* Add new threshold */
4403 new->entries
[size
- 1].eventfd
= eventfd
;
4404 new->entries
[size
- 1].threshold
= threshold
;
4406 /* Sort thresholds. Registering of new threshold isn't time-critical */
4407 sort(new->entries
, size
, sizeof(struct mem_cgroup_threshold
),
4408 compare_thresholds
, NULL
);
4410 /* Find current threshold */
4411 new->current_threshold
= -1;
4412 for (i
= 0; i
< size
; i
++) {
4413 if (new->entries
[i
].threshold
< usage
) {
4415 * new->current_threshold will not be used until
4416 * rcu_assign_pointer(), so it's safe to increment
4419 ++new->current_threshold
;
4423 /* Free old spare buffer and save old primary buffer as spare */
4424 kfree(thresholds
->spare
);
4425 thresholds
->spare
= thresholds
->primary
;
4427 rcu_assign_pointer(thresholds
->primary
, new);
4429 /* To be sure that nobody uses thresholds */
4433 mutex_unlock(&memcg
->thresholds_lock
);
4438 static void mem_cgroup_usage_unregister_event(struct cgroup
*cgrp
,
4439 struct cftype
*cft
, struct eventfd_ctx
*eventfd
)
4441 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4442 struct mem_cgroup_thresholds
*thresholds
;
4443 struct mem_cgroup_threshold_ary
*new;
4444 int type
= MEMFILE_TYPE(cft
->private);
4448 mutex_lock(&memcg
->thresholds_lock
);
4450 thresholds
= &memcg
->thresholds
;
4451 else if (type
== _MEMSWAP
)
4452 thresholds
= &memcg
->memsw_thresholds
;
4456 if (!thresholds
->primary
)
4459 usage
= mem_cgroup_usage(memcg
, type
== _MEMSWAP
);
4461 /* Check if a threshold crossed before removing */
4462 __mem_cgroup_threshold(memcg
, type
== _MEMSWAP
);
4464 /* Calculate new number of threshold */
4466 for (i
= 0; i
< thresholds
->primary
->size
; i
++) {
4467 if (thresholds
->primary
->entries
[i
].eventfd
!= eventfd
)
4471 new = thresholds
->spare
;
4473 /* Set thresholds array to NULL if we don't have thresholds */
4482 /* Copy thresholds and find current threshold */
4483 new->current_threshold
= -1;
4484 for (i
= 0, j
= 0; i
< thresholds
->primary
->size
; i
++) {
4485 if (thresholds
->primary
->entries
[i
].eventfd
== eventfd
)
4488 new->entries
[j
] = thresholds
->primary
->entries
[i
];
4489 if (new->entries
[j
].threshold
< usage
) {
4491 * new->current_threshold will not be used
4492 * until rcu_assign_pointer(), so it's safe to increment
4495 ++new->current_threshold
;
4501 /* Swap primary and spare array */
4502 thresholds
->spare
= thresholds
->primary
;
4503 /* If all events are unregistered, free the spare array */
4505 kfree(thresholds
->spare
);
4506 thresholds
->spare
= NULL
;
4509 rcu_assign_pointer(thresholds
->primary
, new);
4511 /* To be sure that nobody uses thresholds */
4514 mutex_unlock(&memcg
->thresholds_lock
);
4517 static int mem_cgroup_oom_register_event(struct cgroup
*cgrp
,
4518 struct cftype
*cft
, struct eventfd_ctx
*eventfd
, const char *args
)
4520 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4521 struct mem_cgroup_eventfd_list
*event
;
4522 int type
= MEMFILE_TYPE(cft
->private);
4524 BUG_ON(type
!= _OOM_TYPE
);
4525 event
= kmalloc(sizeof(*event
), GFP_KERNEL
);
4529 spin_lock(&memcg_oom_lock
);
4531 event
->eventfd
= eventfd
;
4532 list_add(&event
->list
, &memcg
->oom_notify
);
4534 /* already in OOM ? */
4535 if (atomic_read(&memcg
->under_oom
))
4536 eventfd_signal(eventfd
, 1);
4537 spin_unlock(&memcg_oom_lock
);
4542 static void mem_cgroup_oom_unregister_event(struct cgroup
*cgrp
,
4543 struct cftype
*cft
, struct eventfd_ctx
*eventfd
)
4545 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4546 struct mem_cgroup_eventfd_list
*ev
, *tmp
;
4547 int type
= MEMFILE_TYPE(cft
->private);
4549 BUG_ON(type
!= _OOM_TYPE
);
4551 spin_lock(&memcg_oom_lock
);
4553 list_for_each_entry_safe(ev
, tmp
, &memcg
->oom_notify
, list
) {
4554 if (ev
->eventfd
== eventfd
) {
4555 list_del(&ev
->list
);
4560 spin_unlock(&memcg_oom_lock
);
4563 static int mem_cgroup_oom_control_read(struct cgroup
*cgrp
,
4564 struct cftype
*cft
, struct cgroup_map_cb
*cb
)
4566 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4568 cb
->fill(cb
, "oom_kill_disable", memcg
->oom_kill_disable
);
4570 if (atomic_read(&memcg
->under_oom
))
4571 cb
->fill(cb
, "under_oom", 1);
4573 cb
->fill(cb
, "under_oom", 0);
4577 static int mem_cgroup_oom_control_write(struct cgroup
*cgrp
,
4578 struct cftype
*cft
, u64 val
)
4580 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4581 struct mem_cgroup
*parent
;
4583 /* cannot set to root cgroup and only 0 and 1 are allowed */
4584 if (!cgrp
->parent
|| !((val
== 0) || (val
== 1)))
4587 parent
= mem_cgroup_from_cont(cgrp
->parent
);
4590 /* oom-kill-disable is a flag for subhierarchy. */
4591 if ((parent
->use_hierarchy
) ||
4592 (memcg
->use_hierarchy
&& !list_empty(&cgrp
->children
))) {
4596 memcg
->oom_kill_disable
= val
;
4598 memcg_oom_recover(memcg
);
4604 static const struct file_operations mem_control_numa_stat_file_operations
= {
4606 .llseek
= seq_lseek
,
4607 .release
= single_release
,
4610 static int mem_control_numa_stat_open(struct inode
*unused
, struct file
*file
)
4612 struct cgroup
*cont
= file
->f_dentry
->d_parent
->d_fsdata
;
4614 file
->f_op
= &mem_control_numa_stat_file_operations
;
4615 return single_open(file
, mem_control_numa_stat_show
, cont
);
4617 #endif /* CONFIG_NUMA */
4619 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_KMEM
4620 static int memcg_init_kmem(struct mem_cgroup
*memcg
, struct cgroup_subsys
*ss
)
4622 return mem_cgroup_sockets_init(memcg
, ss
);
4625 static void kmem_cgroup_destroy(struct mem_cgroup
*memcg
)
4627 mem_cgroup_sockets_destroy(memcg
);
4630 static int memcg_init_kmem(struct mem_cgroup
*memcg
, struct cgroup_subsys
*ss
)
4635 static void kmem_cgroup_destroy(struct mem_cgroup
*memcg
)
4640 static struct cftype mem_cgroup_files
[] = {
4642 .name
= "usage_in_bytes",
4643 .private = MEMFILE_PRIVATE(_MEM
, RES_USAGE
),
4644 .read
= mem_cgroup_read
,
4645 .register_event
= mem_cgroup_usage_register_event
,
4646 .unregister_event
= mem_cgroup_usage_unregister_event
,
4649 .name
= "max_usage_in_bytes",
4650 .private = MEMFILE_PRIVATE(_MEM
, RES_MAX_USAGE
),
4651 .trigger
= mem_cgroup_reset
,
4652 .read
= mem_cgroup_read
,
4655 .name
= "limit_in_bytes",
4656 .private = MEMFILE_PRIVATE(_MEM
, RES_LIMIT
),
4657 .write_string
= mem_cgroup_write
,
4658 .read
= mem_cgroup_read
,
4661 .name
= "soft_limit_in_bytes",
4662 .private = MEMFILE_PRIVATE(_MEM
, RES_SOFT_LIMIT
),
4663 .write_string
= mem_cgroup_write
,
4664 .read
= mem_cgroup_read
,
4668 .private = MEMFILE_PRIVATE(_MEM
, RES_FAILCNT
),
4669 .trigger
= mem_cgroup_reset
,
4670 .read
= mem_cgroup_read
,
4674 .read_map
= mem_control_stat_show
,
4677 .name
= "force_empty",
4678 .trigger
= mem_cgroup_force_empty_write
,
4681 .name
= "use_hierarchy",
4682 .write_u64
= mem_cgroup_hierarchy_write
,
4683 .read_u64
= mem_cgroup_hierarchy_read
,
4686 .name
= "swappiness",
4687 .read_u64
= mem_cgroup_swappiness_read
,
4688 .write_u64
= mem_cgroup_swappiness_write
,
4691 .name
= "move_charge_at_immigrate",
4692 .read_u64
= mem_cgroup_move_charge_read
,
4693 .write_u64
= mem_cgroup_move_charge_write
,
4696 .name
= "oom_control",
4697 .read_map
= mem_cgroup_oom_control_read
,
4698 .write_u64
= mem_cgroup_oom_control_write
,
4699 .register_event
= mem_cgroup_oom_register_event
,
4700 .unregister_event
= mem_cgroup_oom_unregister_event
,
4701 .private = MEMFILE_PRIVATE(_OOM_TYPE
, OOM_CONTROL
),
4705 .name
= "numa_stat",
4706 .open
= mem_control_numa_stat_open
,
4710 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
4712 .name
= "memsw.usage_in_bytes",
4713 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_USAGE
),
4714 .read
= mem_cgroup_read
,
4715 .register_event
= mem_cgroup_usage_register_event
,
4716 .unregister_event
= mem_cgroup_usage_unregister_event
,
4719 .name
= "memsw.max_usage_in_bytes",
4720 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_MAX_USAGE
),
4721 .trigger
= mem_cgroup_reset
,
4722 .read
= mem_cgroup_read
,
4725 .name
= "memsw.limit_in_bytes",
4726 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_LIMIT
),
4727 .write_string
= mem_cgroup_write
,
4728 .read
= mem_cgroup_read
,
4731 .name
= "memsw.failcnt",
4732 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_FAILCNT
),
4733 .trigger
= mem_cgroup_reset
,
4734 .read
= mem_cgroup_read
,
4737 { }, /* terminate */
4740 static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup
*memcg
, int node
)
4742 struct mem_cgroup_per_node
*pn
;
4743 struct mem_cgroup_per_zone
*mz
;
4745 int zone
, tmp
= node
;
4747 * This routine is called against possible nodes.
4748 * But it's BUG to call kmalloc() against offline node.
4750 * TODO: this routine can waste much memory for nodes which will
4751 * never be onlined. It's better to use memory hotplug callback
4754 if (!node_state(node
, N_NORMAL_MEMORY
))
4756 pn
= kzalloc_node(sizeof(*pn
), GFP_KERNEL
, tmp
);
4760 for (zone
= 0; zone
< MAX_NR_ZONES
; zone
++) {
4761 mz
= &pn
->zoneinfo
[zone
];
4763 INIT_LIST_HEAD(&mz
->lruvec
.lists
[lru
]);
4764 mz
->usage_in_excess
= 0;
4765 mz
->on_tree
= false;
4768 memcg
->info
.nodeinfo
[node
] = pn
;
4772 static void free_mem_cgroup_per_zone_info(struct mem_cgroup
*memcg
, int node
)
4774 kfree(memcg
->info
.nodeinfo
[node
]);
4777 static struct mem_cgroup
*mem_cgroup_alloc(void)
4779 struct mem_cgroup
*memcg
;
4780 int size
= sizeof(struct mem_cgroup
);
4782 /* Can be very big if MAX_NUMNODES is very big */
4783 if (size
< PAGE_SIZE
)
4784 memcg
= kzalloc(size
, GFP_KERNEL
);
4786 memcg
= vzalloc(size
);
4791 memcg
->stat
= alloc_percpu(struct mem_cgroup_stat_cpu
);
4794 spin_lock_init(&memcg
->pcp_counter_lock
);
4798 if (size
< PAGE_SIZE
)
4806 * Helpers for freeing a vzalloc()ed mem_cgroup by RCU,
4807 * but in process context. The work_freeing structure is overlaid
4808 * on the rcu_freeing structure, which itself is overlaid on memsw.
4810 static void vfree_work(struct work_struct
*work
)
4812 struct mem_cgroup
*memcg
;
4814 memcg
= container_of(work
, struct mem_cgroup
, work_freeing
);
4817 static void vfree_rcu(struct rcu_head
*rcu_head
)
4819 struct mem_cgroup
*memcg
;
4821 memcg
= container_of(rcu_head
, struct mem_cgroup
, rcu_freeing
);
4822 INIT_WORK(&memcg
->work_freeing
, vfree_work
);
4823 schedule_work(&memcg
->work_freeing
);
4827 * At destroying mem_cgroup, references from swap_cgroup can remain.
4828 * (scanning all at force_empty is too costly...)
4830 * Instead of clearing all references at force_empty, we remember
4831 * the number of reference from swap_cgroup and free mem_cgroup when
4832 * it goes down to 0.
4834 * Removal of cgroup itself succeeds regardless of refs from swap.
4837 static void __mem_cgroup_free(struct mem_cgroup
*memcg
)
4841 mem_cgroup_remove_from_trees(memcg
);
4842 free_css_id(&mem_cgroup_subsys
, &memcg
->css
);
4845 free_mem_cgroup_per_zone_info(memcg
, node
);
4847 free_percpu(memcg
->stat
);
4848 if (sizeof(struct mem_cgroup
) < PAGE_SIZE
)
4849 kfree_rcu(memcg
, rcu_freeing
);
4851 call_rcu(&memcg
->rcu_freeing
, vfree_rcu
);
4854 static void mem_cgroup_get(struct mem_cgroup
*memcg
)
4856 atomic_inc(&memcg
->refcnt
);
4859 static void __mem_cgroup_put(struct mem_cgroup
*memcg
, int count
)
4861 if (atomic_sub_and_test(count
, &memcg
->refcnt
)) {
4862 struct mem_cgroup
*parent
= parent_mem_cgroup(memcg
);
4863 __mem_cgroup_free(memcg
);
4865 mem_cgroup_put(parent
);
4869 static void mem_cgroup_put(struct mem_cgroup
*memcg
)
4871 __mem_cgroup_put(memcg
, 1);
4875 * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled.
4877 struct mem_cgroup
*parent_mem_cgroup(struct mem_cgroup
*memcg
)
4879 if (!memcg
->res
.parent
)
4881 return mem_cgroup_from_res_counter(memcg
->res
.parent
, res
);
4883 EXPORT_SYMBOL(parent_mem_cgroup
);
4885 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
4886 static void __init
enable_swap_cgroup(void)
4888 if (!mem_cgroup_disabled() && really_do_swap_account
)
4889 do_swap_account
= 1;
4892 static void __init
enable_swap_cgroup(void)
4897 static int mem_cgroup_soft_limit_tree_init(void)
4899 struct mem_cgroup_tree_per_node
*rtpn
;
4900 struct mem_cgroup_tree_per_zone
*rtpz
;
4901 int tmp
, node
, zone
;
4903 for_each_node(node
) {
4905 if (!node_state(node
, N_NORMAL_MEMORY
))
4907 rtpn
= kzalloc_node(sizeof(*rtpn
), GFP_KERNEL
, tmp
);
4911 soft_limit_tree
.rb_tree_per_node
[node
] = rtpn
;
4913 for (zone
= 0; zone
< MAX_NR_ZONES
; zone
++) {
4914 rtpz
= &rtpn
->rb_tree_per_zone
[zone
];
4915 rtpz
->rb_root
= RB_ROOT
;
4916 spin_lock_init(&rtpz
->lock
);
4922 for_each_node(node
) {
4923 if (!soft_limit_tree
.rb_tree_per_node
[node
])
4925 kfree(soft_limit_tree
.rb_tree_per_node
[node
]);
4926 soft_limit_tree
.rb_tree_per_node
[node
] = NULL
;
4932 static struct cgroup_subsys_state
* __ref
4933 mem_cgroup_create(struct cgroup
*cont
)
4935 struct mem_cgroup
*memcg
, *parent
;
4936 long error
= -ENOMEM
;
4939 memcg
= mem_cgroup_alloc();
4941 return ERR_PTR(error
);
4944 if (alloc_mem_cgroup_per_zone_info(memcg
, node
))
4948 if (cont
->parent
== NULL
) {
4950 enable_swap_cgroup();
4952 if (mem_cgroup_soft_limit_tree_init())
4954 root_mem_cgroup
= memcg
;
4955 for_each_possible_cpu(cpu
) {
4956 struct memcg_stock_pcp
*stock
=
4957 &per_cpu(memcg_stock
, cpu
);
4958 INIT_WORK(&stock
->work
, drain_local_stock
);
4960 hotcpu_notifier(memcg_cpu_hotplug_callback
, 0);
4962 parent
= mem_cgroup_from_cont(cont
->parent
);
4963 memcg
->use_hierarchy
= parent
->use_hierarchy
;
4964 memcg
->oom_kill_disable
= parent
->oom_kill_disable
;
4967 if (parent
&& parent
->use_hierarchy
) {
4968 res_counter_init(&memcg
->res
, &parent
->res
);
4969 res_counter_init(&memcg
->memsw
, &parent
->memsw
);
4971 * We increment refcnt of the parent to ensure that we can
4972 * safely access it on res_counter_charge/uncharge.
4973 * This refcnt will be decremented when freeing this
4974 * mem_cgroup(see mem_cgroup_put).
4976 mem_cgroup_get(parent
);
4978 res_counter_init(&memcg
->res
, NULL
);
4979 res_counter_init(&memcg
->memsw
, NULL
);
4981 memcg
->last_scanned_node
= MAX_NUMNODES
;
4982 INIT_LIST_HEAD(&memcg
->oom_notify
);
4985 memcg
->swappiness
= mem_cgroup_swappiness(parent
);
4986 atomic_set(&memcg
->refcnt
, 1);
4987 memcg
->move_charge_at_immigrate
= 0;
4988 mutex_init(&memcg
->thresholds_lock
);
4989 spin_lock_init(&memcg
->move_lock
);
4991 error
= memcg_init_kmem(memcg
, &mem_cgroup_subsys
);
4994 * We call put now because our (and parent's) refcnts
4995 * are already in place. mem_cgroup_put() will internally
4996 * call __mem_cgroup_free, so return directly
4998 mem_cgroup_put(memcg
);
4999 return ERR_PTR(error
);
5003 __mem_cgroup_free(memcg
);
5004 return ERR_PTR(error
);
5007 static int mem_cgroup_pre_destroy(struct cgroup
*cont
)
5009 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
5011 return mem_cgroup_force_empty(memcg
, false);
5014 static void mem_cgroup_destroy(struct cgroup
*cont
)
5016 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
5018 kmem_cgroup_destroy(memcg
);
5020 mem_cgroup_put(memcg
);
5024 /* Handlers for move charge at task migration. */
5025 #define PRECHARGE_COUNT_AT_ONCE 256
5026 static int mem_cgroup_do_precharge(unsigned long count
)
5029 int batch_count
= PRECHARGE_COUNT_AT_ONCE
;
5030 struct mem_cgroup
*memcg
= mc
.to
;
5032 if (mem_cgroup_is_root(memcg
)) {
5033 mc
.precharge
+= count
;
5034 /* we don't need css_get for root */
5037 /* try to charge at once */
5039 struct res_counter
*dummy
;
5041 * "memcg" cannot be under rmdir() because we've already checked
5042 * by cgroup_lock_live_cgroup() that it is not removed and we
5043 * are still under the same cgroup_mutex. So we can postpone
5046 if (res_counter_charge(&memcg
->res
, PAGE_SIZE
* count
, &dummy
))
5048 if (do_swap_account
&& res_counter_charge(&memcg
->memsw
,
5049 PAGE_SIZE
* count
, &dummy
)) {
5050 res_counter_uncharge(&memcg
->res
, PAGE_SIZE
* count
);
5053 mc
.precharge
+= count
;
5057 /* fall back to one by one charge */
5059 if (signal_pending(current
)) {
5063 if (!batch_count
--) {
5064 batch_count
= PRECHARGE_COUNT_AT_ONCE
;
5067 ret
= __mem_cgroup_try_charge(NULL
,
5068 GFP_KERNEL
, 1, &memcg
, false);
5070 /* mem_cgroup_clear_mc() will do uncharge later */
5078 * get_mctgt_type - get target type of moving charge
5079 * @vma: the vma the pte to be checked belongs
5080 * @addr: the address corresponding to the pte to be checked
5081 * @ptent: the pte to be checked
5082 * @target: the pointer the target page or swap ent will be stored(can be NULL)
5085 * 0(MC_TARGET_NONE): if the pte is not a target for move charge.
5086 * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
5087 * move charge. if @target is not NULL, the page is stored in target->page
5088 * with extra refcnt got(Callers should handle it).
5089 * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
5090 * target for charge migration. if @target is not NULL, the entry is stored
5093 * Called with pte lock held.
5100 enum mc_target_type
{
5106 static struct page
*mc_handle_present_pte(struct vm_area_struct
*vma
,
5107 unsigned long addr
, pte_t ptent
)
5109 struct page
*page
= vm_normal_page(vma
, addr
, ptent
);
5111 if (!page
|| !page_mapped(page
))
5113 if (PageAnon(page
)) {
5114 /* we don't move shared anon */
5117 } else if (!move_file())
5118 /* we ignore mapcount for file pages */
5120 if (!get_page_unless_zero(page
))
5127 static struct page
*mc_handle_swap_pte(struct vm_area_struct
*vma
,
5128 unsigned long addr
, pte_t ptent
, swp_entry_t
*entry
)
5130 struct page
*page
= NULL
;
5131 swp_entry_t ent
= pte_to_swp_entry(ptent
);
5133 if (!move_anon() || non_swap_entry(ent
))
5136 * Because lookup_swap_cache() updates some statistics counter,
5137 * we call find_get_page() with swapper_space directly.
5139 page
= find_get_page(&swapper_space
, ent
.val
);
5140 if (do_swap_account
)
5141 entry
->val
= ent
.val
;
5146 static struct page
*mc_handle_swap_pte(struct vm_area_struct
*vma
,
5147 unsigned long addr
, pte_t ptent
, swp_entry_t
*entry
)
5153 static struct page
*mc_handle_file_pte(struct vm_area_struct
*vma
,
5154 unsigned long addr
, pte_t ptent
, swp_entry_t
*entry
)
5156 struct page
*page
= NULL
;
5157 struct inode
*inode
;
5158 struct address_space
*mapping
;
5161 if (!vma
->vm_file
) /* anonymous vma */
5166 inode
= vma
->vm_file
->f_path
.dentry
->d_inode
;
5167 mapping
= vma
->vm_file
->f_mapping
;
5168 if (pte_none(ptent
))
5169 pgoff
= linear_page_index(vma
, addr
);
5170 else /* pte_file(ptent) is true */
5171 pgoff
= pte_to_pgoff(ptent
);
5173 /* page is moved even if it's not RSS of this task(page-faulted). */
5174 page
= find_get_page(mapping
, pgoff
);
5177 /* shmem/tmpfs may report page out on swap: account for that too. */
5178 if (radix_tree_exceptional_entry(page
)) {
5179 swp_entry_t swap
= radix_to_swp_entry(page
);
5180 if (do_swap_account
)
5182 page
= find_get_page(&swapper_space
, swap
.val
);
5188 static enum mc_target_type
get_mctgt_type(struct vm_area_struct
*vma
,
5189 unsigned long addr
, pte_t ptent
, union mc_target
*target
)
5191 struct page
*page
= NULL
;
5192 struct page_cgroup
*pc
;
5193 enum mc_target_type ret
= MC_TARGET_NONE
;
5194 swp_entry_t ent
= { .val
= 0 };
5196 if (pte_present(ptent
))
5197 page
= mc_handle_present_pte(vma
, addr
, ptent
);
5198 else if (is_swap_pte(ptent
))
5199 page
= mc_handle_swap_pte(vma
, addr
, ptent
, &ent
);
5200 else if (pte_none(ptent
) || pte_file(ptent
))
5201 page
= mc_handle_file_pte(vma
, addr
, ptent
, &ent
);
5203 if (!page
&& !ent
.val
)
5206 pc
= lookup_page_cgroup(page
);
5208 * Do only loose check w/o page_cgroup lock.
5209 * mem_cgroup_move_account() checks the pc is valid or not under
5212 if (PageCgroupUsed(pc
) && pc
->mem_cgroup
== mc
.from
) {
5213 ret
= MC_TARGET_PAGE
;
5215 target
->page
= page
;
5217 if (!ret
|| !target
)
5220 /* There is a swap entry and a page doesn't exist or isn't charged */
5221 if (ent
.val
&& !ret
&&
5222 css_id(&mc
.from
->css
) == lookup_swap_cgroup_id(ent
)) {
5223 ret
= MC_TARGET_SWAP
;
5230 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
5232 * We don't consider swapping or file mapped pages because THP does not
5233 * support them for now.
5234 * Caller should make sure that pmd_trans_huge(pmd) is true.
5236 static enum mc_target_type
get_mctgt_type_thp(struct vm_area_struct
*vma
,
5237 unsigned long addr
, pmd_t pmd
, union mc_target
*target
)
5239 struct page
*page
= NULL
;
5240 struct page_cgroup
*pc
;
5241 enum mc_target_type ret
= MC_TARGET_NONE
;
5243 page
= pmd_page(pmd
);
5244 VM_BUG_ON(!page
|| !PageHead(page
));
5247 pc
= lookup_page_cgroup(page
);
5248 if (PageCgroupUsed(pc
) && pc
->mem_cgroup
== mc
.from
) {
5249 ret
= MC_TARGET_PAGE
;
5252 target
->page
= page
;
5258 static inline enum mc_target_type
get_mctgt_type_thp(struct vm_area_struct
*vma
,
5259 unsigned long addr
, pmd_t pmd
, union mc_target
*target
)
5261 return MC_TARGET_NONE
;
5265 static int mem_cgroup_count_precharge_pte_range(pmd_t
*pmd
,
5266 unsigned long addr
, unsigned long end
,
5267 struct mm_walk
*walk
)
5269 struct vm_area_struct
*vma
= walk
->private;
5273 if (pmd_trans_huge_lock(pmd
, vma
) == 1) {
5274 if (get_mctgt_type_thp(vma
, addr
, *pmd
, NULL
) == MC_TARGET_PAGE
)
5275 mc
.precharge
+= HPAGE_PMD_NR
;
5276 spin_unlock(&vma
->vm_mm
->page_table_lock
);
5280 if (pmd_trans_unstable(pmd
))
5282 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
5283 for (; addr
!= end
; pte
++, addr
+= PAGE_SIZE
)
5284 if (get_mctgt_type(vma
, addr
, *pte
, NULL
))
5285 mc
.precharge
++; /* increment precharge temporarily */
5286 pte_unmap_unlock(pte
- 1, ptl
);
5292 static unsigned long mem_cgroup_count_precharge(struct mm_struct
*mm
)
5294 unsigned long precharge
;
5295 struct vm_area_struct
*vma
;
5297 down_read(&mm
->mmap_sem
);
5298 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
5299 struct mm_walk mem_cgroup_count_precharge_walk
= {
5300 .pmd_entry
= mem_cgroup_count_precharge_pte_range
,
5304 if (is_vm_hugetlb_page(vma
))
5306 walk_page_range(vma
->vm_start
, vma
->vm_end
,
5307 &mem_cgroup_count_precharge_walk
);
5309 up_read(&mm
->mmap_sem
);
5311 precharge
= mc
.precharge
;
5317 static int mem_cgroup_precharge_mc(struct mm_struct
*mm
)
5319 unsigned long precharge
= mem_cgroup_count_precharge(mm
);
5321 VM_BUG_ON(mc
.moving_task
);
5322 mc
.moving_task
= current
;
5323 return mem_cgroup_do_precharge(precharge
);
5326 /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */
5327 static void __mem_cgroup_clear_mc(void)
5329 struct mem_cgroup
*from
= mc
.from
;
5330 struct mem_cgroup
*to
= mc
.to
;
5332 /* we must uncharge all the leftover precharges from mc.to */
5334 __mem_cgroup_cancel_charge(mc
.to
, mc
.precharge
);
5338 * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
5339 * we must uncharge here.
5341 if (mc
.moved_charge
) {
5342 __mem_cgroup_cancel_charge(mc
.from
, mc
.moved_charge
);
5343 mc
.moved_charge
= 0;
5345 /* we must fixup refcnts and charges */
5346 if (mc
.moved_swap
) {
5347 /* uncharge swap account from the old cgroup */
5348 if (!mem_cgroup_is_root(mc
.from
))
5349 res_counter_uncharge(&mc
.from
->memsw
,
5350 PAGE_SIZE
* mc
.moved_swap
);
5351 __mem_cgroup_put(mc
.from
, mc
.moved_swap
);
5353 if (!mem_cgroup_is_root(mc
.to
)) {
5355 * we charged both to->res and to->memsw, so we should
5358 res_counter_uncharge(&mc
.to
->res
,
5359 PAGE_SIZE
* mc
.moved_swap
);
5361 /* we've already done mem_cgroup_get(mc.to) */
5364 memcg_oom_recover(from
);
5365 memcg_oom_recover(to
);
5366 wake_up_all(&mc
.waitq
);
5369 static void mem_cgroup_clear_mc(void)
5371 struct mem_cgroup
*from
= mc
.from
;
5374 * we must clear moving_task before waking up waiters at the end of
5377 mc
.moving_task
= NULL
;
5378 __mem_cgroup_clear_mc();
5379 spin_lock(&mc
.lock
);
5382 spin_unlock(&mc
.lock
);
5383 mem_cgroup_end_move(from
);
5386 static int mem_cgroup_can_attach(struct cgroup
*cgroup
,
5387 struct cgroup_taskset
*tset
)
5389 struct task_struct
*p
= cgroup_taskset_first(tset
);
5391 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgroup
);
5393 if (memcg
->move_charge_at_immigrate
) {
5394 struct mm_struct
*mm
;
5395 struct mem_cgroup
*from
= mem_cgroup_from_task(p
);
5397 VM_BUG_ON(from
== memcg
);
5399 mm
= get_task_mm(p
);
5402 /* We move charges only when we move a owner of the mm */
5403 if (mm
->owner
== p
) {
5406 VM_BUG_ON(mc
.precharge
);
5407 VM_BUG_ON(mc
.moved_charge
);
5408 VM_BUG_ON(mc
.moved_swap
);
5409 mem_cgroup_start_move(from
);
5410 spin_lock(&mc
.lock
);
5413 spin_unlock(&mc
.lock
);
5414 /* We set mc.moving_task later */
5416 ret
= mem_cgroup_precharge_mc(mm
);
5418 mem_cgroup_clear_mc();
5425 static void mem_cgroup_cancel_attach(struct cgroup
*cgroup
,
5426 struct cgroup_taskset
*tset
)
5428 mem_cgroup_clear_mc();
5431 static int mem_cgroup_move_charge_pte_range(pmd_t
*pmd
,
5432 unsigned long addr
, unsigned long end
,
5433 struct mm_walk
*walk
)
5436 struct vm_area_struct
*vma
= walk
->private;
5439 enum mc_target_type target_type
;
5440 union mc_target target
;
5442 struct page_cgroup
*pc
;
5445 * We don't take compound_lock() here but no race with splitting thp
5447 * - if pmd_trans_huge_lock() returns 1, the relevant thp is not
5448 * under splitting, which means there's no concurrent thp split,
5449 * - if another thread runs into split_huge_page() just after we
5450 * entered this if-block, the thread must wait for page table lock
5451 * to be unlocked in __split_huge_page_splitting(), where the main
5452 * part of thp split is not executed yet.
5454 if (pmd_trans_huge_lock(pmd
, vma
) == 1) {
5455 if (mc
.precharge
< HPAGE_PMD_NR
) {
5456 spin_unlock(&vma
->vm_mm
->page_table_lock
);
5459 target_type
= get_mctgt_type_thp(vma
, addr
, *pmd
, &target
);
5460 if (target_type
== MC_TARGET_PAGE
) {
5462 if (!isolate_lru_page(page
)) {
5463 pc
= lookup_page_cgroup(page
);
5464 if (!mem_cgroup_move_account(page
, HPAGE_PMD_NR
,
5467 mc
.precharge
-= HPAGE_PMD_NR
;
5468 mc
.moved_charge
+= HPAGE_PMD_NR
;
5470 putback_lru_page(page
);
5474 spin_unlock(&vma
->vm_mm
->page_table_lock
);
5478 if (pmd_trans_unstable(pmd
))
5481 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
5482 for (; addr
!= end
; addr
+= PAGE_SIZE
) {
5483 pte_t ptent
= *(pte
++);
5489 switch (get_mctgt_type(vma
, addr
, ptent
, &target
)) {
5490 case MC_TARGET_PAGE
:
5492 if (isolate_lru_page(page
))
5494 pc
= lookup_page_cgroup(page
);
5495 if (!mem_cgroup_move_account(page
, 1, pc
,
5496 mc
.from
, mc
.to
, false)) {
5498 /* we uncharge from mc.from later. */
5501 putback_lru_page(page
);
5502 put
: /* get_mctgt_type() gets the page */
5505 case MC_TARGET_SWAP
:
5507 if (!mem_cgroup_move_swap_account(ent
, mc
.from
, mc
.to
)) {
5509 /* we fixup refcnts and charges later. */
5517 pte_unmap_unlock(pte
- 1, ptl
);
5522 * We have consumed all precharges we got in can_attach().
5523 * We try charge one by one, but don't do any additional
5524 * charges to mc.to if we have failed in charge once in attach()
5527 ret
= mem_cgroup_do_precharge(1);
5535 static void mem_cgroup_move_charge(struct mm_struct
*mm
)
5537 struct vm_area_struct
*vma
;
5539 lru_add_drain_all();
5541 if (unlikely(!down_read_trylock(&mm
->mmap_sem
))) {
5543 * Someone who are holding the mmap_sem might be waiting in
5544 * waitq. So we cancel all extra charges, wake up all waiters,
5545 * and retry. Because we cancel precharges, we might not be able
5546 * to move enough charges, but moving charge is a best-effort
5547 * feature anyway, so it wouldn't be a big problem.
5549 __mem_cgroup_clear_mc();
5553 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
5555 struct mm_walk mem_cgroup_move_charge_walk
= {
5556 .pmd_entry
= mem_cgroup_move_charge_pte_range
,
5560 if (is_vm_hugetlb_page(vma
))
5562 ret
= walk_page_range(vma
->vm_start
, vma
->vm_end
,
5563 &mem_cgroup_move_charge_walk
);
5566 * means we have consumed all precharges and failed in
5567 * doing additional charge. Just abandon here.
5571 up_read(&mm
->mmap_sem
);
5574 static void mem_cgroup_move_task(struct cgroup
*cont
,
5575 struct cgroup_taskset
*tset
)
5577 struct task_struct
*p
= cgroup_taskset_first(tset
);
5578 struct mm_struct
*mm
= get_task_mm(p
);
5582 mem_cgroup_move_charge(mm
);
5586 mem_cgroup_clear_mc();
5588 #else /* !CONFIG_MMU */
5589 static int mem_cgroup_can_attach(struct cgroup
*cgroup
,
5590 struct cgroup_taskset
*tset
)
5594 static void mem_cgroup_cancel_attach(struct cgroup
*cgroup
,
5595 struct cgroup_taskset
*tset
)
5598 static void mem_cgroup_move_task(struct cgroup
*cont
,
5599 struct cgroup_taskset
*tset
)
5604 struct cgroup_subsys mem_cgroup_subsys
= {
5606 .subsys_id
= mem_cgroup_subsys_id
,
5607 .create
= mem_cgroup_create
,
5608 .pre_destroy
= mem_cgroup_pre_destroy
,
5609 .destroy
= mem_cgroup_destroy
,
5610 .can_attach
= mem_cgroup_can_attach
,
5611 .cancel_attach
= mem_cgroup_cancel_attach
,
5612 .attach
= mem_cgroup_move_task
,
5613 .base_cftypes
= mem_cgroup_files
,
5616 .__DEPRECATED_clear_css_refs
= true,
5619 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
5620 static int __init
enable_swap_account(char *s
)
5622 /* consider enabled if no parameter or 1 is given */
5623 if (!strcmp(s
, "1"))
5624 really_do_swap_account
= 1;
5625 else if (!strcmp(s
, "0"))
5626 really_do_swap_account
= 0;
5629 __setup("swapaccount=", enable_swap_account
);