1 /* memcontrol.c - Memory Controller
3 * Copyright IBM Corporation, 2007
4 * Author Balbir Singh <balbir@linux.vnet.ibm.com>
6 * Copyright 2007 OpenVZ SWsoft Inc
7 * Author: Pavel Emelianov <xemul@openvz.org>
10 * Copyright (C) 2009 Nokia Corporation
11 * Author: Kirill A. Shutemov
13 * This program is free software; you can redistribute it and/or modify
14 * it under the terms of the GNU General Public License as published by
15 * the Free Software Foundation; either version 2 of the License, or
16 * (at your option) any later version.
18 * This program is distributed in the hope that it will be useful,
19 * but WITHOUT ANY WARRANTY; without even the implied warranty of
20 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
21 * GNU General Public License for more details.
24 #include <linux/res_counter.h>
25 #include <linux/memcontrol.h>
26 #include <linux/cgroup.h>
28 #include <linux/hugetlb.h>
29 #include <linux/pagemap.h>
30 #include <linux/smp.h>
31 #include <linux/page-flags.h>
32 #include <linux/backing-dev.h>
33 #include <linux/bit_spinlock.h>
34 #include <linux/rcupdate.h>
35 #include <linux/limits.h>
36 #include <linux/export.h>
37 #include <linux/mutex.h>
38 #include <linux/rbtree.h>
39 #include <linux/slab.h>
40 #include <linux/swap.h>
41 #include <linux/swapops.h>
42 #include <linux/spinlock.h>
43 #include <linux/eventfd.h>
44 #include <linux/sort.h>
46 #include <linux/seq_file.h>
47 #include <linux/vmalloc.h>
48 #include <linux/mm_inline.h>
49 #include <linux/page_cgroup.h>
50 #include <linux/cpu.h>
51 #include <linux/oom.h>
54 #include <net/tcp_memcontrol.h>
56 #include <asm/uaccess.h>
58 #include <trace/events/vmscan.h>
60 struct cgroup_subsys mem_cgroup_subsys __read_mostly
;
61 #define MEM_CGROUP_RECLAIM_RETRIES 5
62 struct mem_cgroup
*root_mem_cgroup __read_mostly
;
64 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
65 /* Turned on only when memory cgroup is enabled && really_do_swap_account = 1 */
66 int do_swap_account __read_mostly
;
68 /* for remember boot option*/
69 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP_ENABLED
70 static int really_do_swap_account __initdata
= 1;
72 static int really_do_swap_account __initdata
= 0;
76 #define do_swap_account (0)
81 * Statistics for memory cgroup.
83 enum mem_cgroup_stat_index
{
85 * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss.
87 MEM_CGROUP_STAT_CACHE
, /* # of pages charged as cache */
88 MEM_CGROUP_STAT_RSS
, /* # of pages charged as anon rss */
89 MEM_CGROUP_STAT_FILE_MAPPED
, /* # of pages charged as file rss */
90 MEM_CGROUP_STAT_SWAPOUT
, /* # of pages, swapped out */
91 MEM_CGROUP_STAT_DATA
, /* end of data requires synchronization */
92 MEM_CGROUP_ON_MOVE
, /* someone is moving account between groups */
93 MEM_CGROUP_STAT_NSTATS
,
96 enum mem_cgroup_events_index
{
97 MEM_CGROUP_EVENTS_PGPGIN
, /* # of pages paged in */
98 MEM_CGROUP_EVENTS_PGPGOUT
, /* # of pages paged out */
99 MEM_CGROUP_EVENTS_COUNT
, /* # of pages paged in/out */
100 MEM_CGROUP_EVENTS_PGFAULT
, /* # of page-faults */
101 MEM_CGROUP_EVENTS_PGMAJFAULT
, /* # of major page-faults */
102 MEM_CGROUP_EVENTS_NSTATS
,
105 * Per memcg event counter is incremented at every pagein/pageout. With THP,
106 * it will be incremated by the number of pages. This counter is used for
107 * for trigger some periodic events. This is straightforward and better
108 * than using jiffies etc. to handle periodic memcg event.
110 enum mem_cgroup_events_target
{
111 MEM_CGROUP_TARGET_THRESH
,
112 MEM_CGROUP_TARGET_SOFTLIMIT
,
113 MEM_CGROUP_TARGET_NUMAINFO
,
116 #define THRESHOLDS_EVENTS_TARGET (128)
117 #define SOFTLIMIT_EVENTS_TARGET (1024)
118 #define NUMAINFO_EVENTS_TARGET (1024)
120 struct mem_cgroup_stat_cpu
{
121 long count
[MEM_CGROUP_STAT_NSTATS
];
122 unsigned long events
[MEM_CGROUP_EVENTS_NSTATS
];
123 unsigned long targets
[MEM_CGROUP_NTARGETS
];
126 struct mem_cgroup_reclaim_iter
{
127 /* css_id of the last scanned hierarchy member */
129 /* scan generation, increased every round-trip */
130 unsigned int generation
;
134 * per-zone information in memory controller.
136 struct mem_cgroup_per_zone
{
137 struct lruvec lruvec
;
138 unsigned long count
[NR_LRU_LISTS
];
140 struct mem_cgroup_reclaim_iter reclaim_iter
[DEF_PRIORITY
+ 1];
142 struct zone_reclaim_stat reclaim_stat
;
143 struct rb_node tree_node
; /* RB tree node */
144 unsigned long long usage_in_excess
;/* Set to the value by which */
145 /* the soft limit is exceeded*/
147 struct mem_cgroup
*mem
; /* Back pointer, we cannot */
148 /* use container_of */
150 /* Macro for accessing counter */
151 #define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)])
153 struct mem_cgroup_per_node
{
154 struct mem_cgroup_per_zone zoneinfo
[MAX_NR_ZONES
];
157 struct mem_cgroup_lru_info
{
158 struct mem_cgroup_per_node
*nodeinfo
[MAX_NUMNODES
];
162 * Cgroups above their limits are maintained in a RB-Tree, independent of
163 * their hierarchy representation
166 struct mem_cgroup_tree_per_zone
{
167 struct rb_root rb_root
;
171 struct mem_cgroup_tree_per_node
{
172 struct mem_cgroup_tree_per_zone rb_tree_per_zone
[MAX_NR_ZONES
];
175 struct mem_cgroup_tree
{
176 struct mem_cgroup_tree_per_node
*rb_tree_per_node
[MAX_NUMNODES
];
179 static struct mem_cgroup_tree soft_limit_tree __read_mostly
;
181 struct mem_cgroup_threshold
{
182 struct eventfd_ctx
*eventfd
;
187 struct mem_cgroup_threshold_ary
{
188 /* An array index points to threshold just below usage. */
189 int current_threshold
;
190 /* Size of entries[] */
192 /* Array of thresholds */
193 struct mem_cgroup_threshold entries
[0];
196 struct mem_cgroup_thresholds
{
197 /* Primary thresholds array */
198 struct mem_cgroup_threshold_ary
*primary
;
200 * Spare threshold array.
201 * This is needed to make mem_cgroup_unregister_event() "never fail".
202 * It must be able to store at least primary->size - 1 entries.
204 struct mem_cgroup_threshold_ary
*spare
;
208 struct mem_cgroup_eventfd_list
{
209 struct list_head list
;
210 struct eventfd_ctx
*eventfd
;
213 static void mem_cgroup_threshold(struct mem_cgroup
*memcg
);
214 static void mem_cgroup_oom_notify(struct mem_cgroup
*memcg
);
217 * The memory controller data structure. The memory controller controls both
218 * page cache and RSS per cgroup. We would eventually like to provide
219 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
220 * to help the administrator determine what knobs to tune.
222 * TODO: Add a water mark for the memory controller. Reclaim will begin when
223 * we hit the water mark. May be even add a low water mark, such that
224 * no reclaim occurs from a cgroup at it's low water mark, this is
225 * a feature that will be implemented much later in the future.
228 struct cgroup_subsys_state css
;
230 * the counter to account for memory usage
232 struct res_counter res
;
234 * the counter to account for mem+swap usage.
236 struct res_counter memsw
;
238 * Per cgroup active and inactive list, similar to the
239 * per zone LRU lists.
241 struct mem_cgroup_lru_info info
;
242 int last_scanned_node
;
244 nodemask_t scan_nodes
;
245 atomic_t numainfo_events
;
246 atomic_t numainfo_updating
;
249 * Should the accounting and control be hierarchical, per subtree?
259 /* OOM-Killer disable */
260 int oom_kill_disable
;
262 /* set when res.limit == memsw.limit */
263 bool memsw_is_minimum
;
265 /* protect arrays of thresholds */
266 struct mutex thresholds_lock
;
268 /* thresholds for memory usage. RCU-protected */
269 struct mem_cgroup_thresholds thresholds
;
271 /* thresholds for mem+swap usage. RCU-protected */
272 struct mem_cgroup_thresholds memsw_thresholds
;
274 /* For oom notifier event fd */
275 struct list_head oom_notify
;
278 * Should we move charges of a task when a task is moved into this
279 * mem_cgroup ? And what type of charges should we move ?
281 unsigned long move_charge_at_immigrate
;
285 struct mem_cgroup_stat_cpu
*stat
;
287 * used when a cpu is offlined or other synchronizations
288 * See mem_cgroup_read_stat().
290 struct mem_cgroup_stat_cpu nocpu_base
;
291 spinlock_t pcp_counter_lock
;
294 struct tcp_memcontrol tcp_mem
;
298 /* Stuffs for move charges at task migration. */
300 * Types of charges to be moved. "move_charge_at_immitgrate" is treated as a
301 * left-shifted bitmap of these types.
304 MOVE_CHARGE_TYPE_ANON
, /* private anonymous page and swap of it */
305 MOVE_CHARGE_TYPE_FILE
, /* file page(including tmpfs) and swap of it */
309 /* "mc" and its members are protected by cgroup_mutex */
310 static struct move_charge_struct
{
311 spinlock_t lock
; /* for from, to */
312 struct mem_cgroup
*from
;
313 struct mem_cgroup
*to
;
314 unsigned long precharge
;
315 unsigned long moved_charge
;
316 unsigned long moved_swap
;
317 struct task_struct
*moving_task
; /* a task moving charges */
318 wait_queue_head_t waitq
; /* a waitq for other context */
320 .lock
= __SPIN_LOCK_UNLOCKED(mc
.lock
),
321 .waitq
= __WAIT_QUEUE_HEAD_INITIALIZER(mc
.waitq
),
324 static bool move_anon(void)
326 return test_bit(MOVE_CHARGE_TYPE_ANON
,
327 &mc
.to
->move_charge_at_immigrate
);
330 static bool move_file(void)
332 return test_bit(MOVE_CHARGE_TYPE_FILE
,
333 &mc
.to
->move_charge_at_immigrate
);
337 * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft
338 * limit reclaim to prevent infinite loops, if they ever occur.
340 #define MEM_CGROUP_MAX_RECLAIM_LOOPS (100)
341 #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS (2)
344 MEM_CGROUP_CHARGE_TYPE_CACHE
= 0,
345 MEM_CGROUP_CHARGE_TYPE_MAPPED
,
346 MEM_CGROUP_CHARGE_TYPE_SHMEM
, /* used by page migration of shmem */
347 MEM_CGROUP_CHARGE_TYPE_FORCE
, /* used by force_empty */
348 MEM_CGROUP_CHARGE_TYPE_SWAPOUT
, /* for accounting swapcache */
349 MEM_CGROUP_CHARGE_TYPE_DROP
, /* a page was unused swap cache */
353 /* for encoding cft->private value on file */
356 #define _OOM_TYPE (2)
357 #define MEMFILE_PRIVATE(x, val) (((x) << 16) | (val))
358 #define MEMFILE_TYPE(val) (((val) >> 16) & 0xffff)
359 #define MEMFILE_ATTR(val) ((val) & 0xffff)
360 /* Used for OOM nofiier */
361 #define OOM_CONTROL (0)
364 * Reclaim flags for mem_cgroup_hierarchical_reclaim
366 #define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0
367 #define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT)
368 #define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1
369 #define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT)
371 static void mem_cgroup_get(struct mem_cgroup
*memcg
);
372 static void mem_cgroup_put(struct mem_cgroup
*memcg
);
374 /* Writing them here to avoid exposing memcg's inner layout */
375 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_KMEM
376 #include <net/sock.h>
379 static bool mem_cgroup_is_root(struct mem_cgroup
*memcg
);
380 void sock_update_memcg(struct sock
*sk
)
382 if (mem_cgroup_sockets_enabled
) {
383 struct mem_cgroup
*memcg
;
385 BUG_ON(!sk
->sk_prot
->proto_cgroup
);
387 /* Socket cloning can throw us here with sk_cgrp already
388 * filled. It won't however, necessarily happen from
389 * process context. So the test for root memcg given
390 * the current task's memcg won't help us in this case.
392 * Respecting the original socket's memcg is a better
393 * decision in this case.
396 BUG_ON(mem_cgroup_is_root(sk
->sk_cgrp
->memcg
));
397 mem_cgroup_get(sk
->sk_cgrp
->memcg
);
402 memcg
= mem_cgroup_from_task(current
);
403 if (!mem_cgroup_is_root(memcg
)) {
404 mem_cgroup_get(memcg
);
405 sk
->sk_cgrp
= sk
->sk_prot
->proto_cgroup(memcg
);
410 EXPORT_SYMBOL(sock_update_memcg
);
412 void sock_release_memcg(struct sock
*sk
)
414 if (mem_cgroup_sockets_enabled
&& sk
->sk_cgrp
) {
415 struct mem_cgroup
*memcg
;
416 WARN_ON(!sk
->sk_cgrp
->memcg
);
417 memcg
= sk
->sk_cgrp
->memcg
;
418 mem_cgroup_put(memcg
);
423 struct cg_proto
*tcp_proto_cgroup(struct mem_cgroup
*memcg
)
425 if (!memcg
|| mem_cgroup_is_root(memcg
))
428 return &memcg
->tcp_mem
.cg_proto
;
430 EXPORT_SYMBOL(tcp_proto_cgroup
);
431 #endif /* CONFIG_INET */
432 #endif /* CONFIG_CGROUP_MEM_RES_CTLR_KMEM */
434 static void drain_all_stock_async(struct mem_cgroup
*memcg
);
436 static struct mem_cgroup_per_zone
*
437 mem_cgroup_zoneinfo(struct mem_cgroup
*memcg
, int nid
, int zid
)
439 return &memcg
->info
.nodeinfo
[nid
]->zoneinfo
[zid
];
442 struct cgroup_subsys_state
*mem_cgroup_css(struct mem_cgroup
*memcg
)
447 static struct mem_cgroup_per_zone
*
448 page_cgroup_zoneinfo(struct mem_cgroup
*memcg
, struct page
*page
)
450 int nid
= page_to_nid(page
);
451 int zid
= page_zonenum(page
);
453 return mem_cgroup_zoneinfo(memcg
, nid
, zid
);
456 static struct mem_cgroup_tree_per_zone
*
457 soft_limit_tree_node_zone(int nid
, int zid
)
459 return &soft_limit_tree
.rb_tree_per_node
[nid
]->rb_tree_per_zone
[zid
];
462 static struct mem_cgroup_tree_per_zone
*
463 soft_limit_tree_from_page(struct page
*page
)
465 int nid
= page_to_nid(page
);
466 int zid
= page_zonenum(page
);
468 return &soft_limit_tree
.rb_tree_per_node
[nid
]->rb_tree_per_zone
[zid
];
472 __mem_cgroup_insert_exceeded(struct mem_cgroup
*memcg
,
473 struct mem_cgroup_per_zone
*mz
,
474 struct mem_cgroup_tree_per_zone
*mctz
,
475 unsigned long long new_usage_in_excess
)
477 struct rb_node
**p
= &mctz
->rb_root
.rb_node
;
478 struct rb_node
*parent
= NULL
;
479 struct mem_cgroup_per_zone
*mz_node
;
484 mz
->usage_in_excess
= new_usage_in_excess
;
485 if (!mz
->usage_in_excess
)
489 mz_node
= rb_entry(parent
, struct mem_cgroup_per_zone
,
491 if (mz
->usage_in_excess
< mz_node
->usage_in_excess
)
494 * We can't avoid mem cgroups that are over their soft
495 * limit by the same amount
497 else if (mz
->usage_in_excess
>= mz_node
->usage_in_excess
)
500 rb_link_node(&mz
->tree_node
, parent
, p
);
501 rb_insert_color(&mz
->tree_node
, &mctz
->rb_root
);
506 __mem_cgroup_remove_exceeded(struct mem_cgroup
*memcg
,
507 struct mem_cgroup_per_zone
*mz
,
508 struct mem_cgroup_tree_per_zone
*mctz
)
512 rb_erase(&mz
->tree_node
, &mctz
->rb_root
);
517 mem_cgroup_remove_exceeded(struct mem_cgroup
*memcg
,
518 struct mem_cgroup_per_zone
*mz
,
519 struct mem_cgroup_tree_per_zone
*mctz
)
521 spin_lock(&mctz
->lock
);
522 __mem_cgroup_remove_exceeded(memcg
, mz
, mctz
);
523 spin_unlock(&mctz
->lock
);
527 static void mem_cgroup_update_tree(struct mem_cgroup
*memcg
, struct page
*page
)
529 unsigned long long excess
;
530 struct mem_cgroup_per_zone
*mz
;
531 struct mem_cgroup_tree_per_zone
*mctz
;
532 int nid
= page_to_nid(page
);
533 int zid
= page_zonenum(page
);
534 mctz
= soft_limit_tree_from_page(page
);
537 * Necessary to update all ancestors when hierarchy is used.
538 * because their event counter is not touched.
540 for (; memcg
; memcg
= parent_mem_cgroup(memcg
)) {
541 mz
= mem_cgroup_zoneinfo(memcg
, nid
, zid
);
542 excess
= res_counter_soft_limit_excess(&memcg
->res
);
544 * We have to update the tree if mz is on RB-tree or
545 * mem is over its softlimit.
547 if (excess
|| mz
->on_tree
) {
548 spin_lock(&mctz
->lock
);
549 /* if on-tree, remove it */
551 __mem_cgroup_remove_exceeded(memcg
, mz
, mctz
);
553 * Insert again. mz->usage_in_excess will be updated.
554 * If excess is 0, no tree ops.
556 __mem_cgroup_insert_exceeded(memcg
, mz
, mctz
, excess
);
557 spin_unlock(&mctz
->lock
);
562 static void mem_cgroup_remove_from_trees(struct mem_cgroup
*memcg
)
565 struct mem_cgroup_per_zone
*mz
;
566 struct mem_cgroup_tree_per_zone
*mctz
;
568 for_each_node(node
) {
569 for (zone
= 0; zone
< MAX_NR_ZONES
; zone
++) {
570 mz
= mem_cgroup_zoneinfo(memcg
, node
, zone
);
571 mctz
= soft_limit_tree_node_zone(node
, zone
);
572 mem_cgroup_remove_exceeded(memcg
, mz
, mctz
);
577 static struct mem_cgroup_per_zone
*
578 __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone
*mctz
)
580 struct rb_node
*rightmost
= NULL
;
581 struct mem_cgroup_per_zone
*mz
;
585 rightmost
= rb_last(&mctz
->rb_root
);
587 goto done
; /* Nothing to reclaim from */
589 mz
= rb_entry(rightmost
, struct mem_cgroup_per_zone
, tree_node
);
591 * Remove the node now but someone else can add it back,
592 * we will to add it back at the end of reclaim to its correct
593 * position in the tree.
595 __mem_cgroup_remove_exceeded(mz
->mem
, mz
, mctz
);
596 if (!res_counter_soft_limit_excess(&mz
->mem
->res
) ||
597 !css_tryget(&mz
->mem
->css
))
603 static struct mem_cgroup_per_zone
*
604 mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone
*mctz
)
606 struct mem_cgroup_per_zone
*mz
;
608 spin_lock(&mctz
->lock
);
609 mz
= __mem_cgroup_largest_soft_limit_node(mctz
);
610 spin_unlock(&mctz
->lock
);
615 * Implementation Note: reading percpu statistics for memcg.
617 * Both of vmstat[] and percpu_counter has threshold and do periodic
618 * synchronization to implement "quick" read. There are trade-off between
619 * reading cost and precision of value. Then, we may have a chance to implement
620 * a periodic synchronizion of counter in memcg's counter.
622 * But this _read() function is used for user interface now. The user accounts
623 * memory usage by memory cgroup and he _always_ requires exact value because
624 * he accounts memory. Even if we provide quick-and-fuzzy read, we always
625 * have to visit all online cpus and make sum. So, for now, unnecessary
626 * synchronization is not implemented. (just implemented for cpu hotplug)
628 * If there are kernel internal actions which can make use of some not-exact
629 * value, and reading all cpu value can be performance bottleneck in some
630 * common workload, threashold and synchonization as vmstat[] should be
633 static long mem_cgroup_read_stat(struct mem_cgroup
*memcg
,
634 enum mem_cgroup_stat_index idx
)
640 for_each_online_cpu(cpu
)
641 val
+= per_cpu(memcg
->stat
->count
[idx
], cpu
);
642 #ifdef CONFIG_HOTPLUG_CPU
643 spin_lock(&memcg
->pcp_counter_lock
);
644 val
+= memcg
->nocpu_base
.count
[idx
];
645 spin_unlock(&memcg
->pcp_counter_lock
);
651 static void mem_cgroup_swap_statistics(struct mem_cgroup
*memcg
,
654 int val
= (charge
) ? 1 : -1;
655 this_cpu_add(memcg
->stat
->count
[MEM_CGROUP_STAT_SWAPOUT
], val
);
658 static unsigned long mem_cgroup_read_events(struct mem_cgroup
*memcg
,
659 enum mem_cgroup_events_index idx
)
661 unsigned long val
= 0;
664 for_each_online_cpu(cpu
)
665 val
+= per_cpu(memcg
->stat
->events
[idx
], cpu
);
666 #ifdef CONFIG_HOTPLUG_CPU
667 spin_lock(&memcg
->pcp_counter_lock
);
668 val
+= memcg
->nocpu_base
.events
[idx
];
669 spin_unlock(&memcg
->pcp_counter_lock
);
674 static void mem_cgroup_charge_statistics(struct mem_cgroup
*memcg
,
675 bool file
, int nr_pages
)
680 __this_cpu_add(memcg
->stat
->count
[MEM_CGROUP_STAT_CACHE
],
683 __this_cpu_add(memcg
->stat
->count
[MEM_CGROUP_STAT_RSS
],
686 /* pagein of a big page is an event. So, ignore page size */
688 __this_cpu_inc(memcg
->stat
->events
[MEM_CGROUP_EVENTS_PGPGIN
]);
690 __this_cpu_inc(memcg
->stat
->events
[MEM_CGROUP_EVENTS_PGPGOUT
]);
691 nr_pages
= -nr_pages
; /* for event */
694 __this_cpu_add(memcg
->stat
->events
[MEM_CGROUP_EVENTS_COUNT
], nr_pages
);
700 mem_cgroup_zone_nr_lru_pages(struct mem_cgroup
*memcg
, int nid
, int zid
,
701 unsigned int lru_mask
)
703 struct mem_cgroup_per_zone
*mz
;
705 unsigned long ret
= 0;
707 mz
= mem_cgroup_zoneinfo(memcg
, nid
, zid
);
710 if (BIT(l
) & lru_mask
)
711 ret
+= MEM_CGROUP_ZSTAT(mz
, l
);
717 mem_cgroup_node_nr_lru_pages(struct mem_cgroup
*memcg
,
718 int nid
, unsigned int lru_mask
)
723 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++)
724 total
+= mem_cgroup_zone_nr_lru_pages(memcg
,
730 static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup
*memcg
,
731 unsigned int lru_mask
)
736 for_each_node_state(nid
, N_HIGH_MEMORY
)
737 total
+= mem_cgroup_node_nr_lru_pages(memcg
, nid
, lru_mask
);
741 static bool mem_cgroup_event_ratelimit(struct mem_cgroup
*memcg
,
742 enum mem_cgroup_events_target target
)
744 unsigned long val
, next
;
746 val
= __this_cpu_read(memcg
->stat
->events
[MEM_CGROUP_EVENTS_COUNT
]);
747 next
= __this_cpu_read(memcg
->stat
->targets
[target
]);
748 /* from time_after() in jiffies.h */
749 if ((long)next
- (long)val
< 0) {
751 case MEM_CGROUP_TARGET_THRESH
:
752 next
= val
+ THRESHOLDS_EVENTS_TARGET
;
754 case MEM_CGROUP_TARGET_SOFTLIMIT
:
755 next
= val
+ SOFTLIMIT_EVENTS_TARGET
;
757 case MEM_CGROUP_TARGET_NUMAINFO
:
758 next
= val
+ NUMAINFO_EVENTS_TARGET
;
763 __this_cpu_write(memcg
->stat
->targets
[target
], next
);
770 * Check events in order.
773 static void memcg_check_events(struct mem_cgroup
*memcg
, struct page
*page
)
776 /* threshold event is triggered in finer grain than soft limit */
777 if (unlikely(mem_cgroup_event_ratelimit(memcg
,
778 MEM_CGROUP_TARGET_THRESH
))) {
780 bool do_numainfo __maybe_unused
;
782 do_softlimit
= mem_cgroup_event_ratelimit(memcg
,
783 MEM_CGROUP_TARGET_SOFTLIMIT
);
785 do_numainfo
= mem_cgroup_event_ratelimit(memcg
,
786 MEM_CGROUP_TARGET_NUMAINFO
);
790 mem_cgroup_threshold(memcg
);
791 if (unlikely(do_softlimit
))
792 mem_cgroup_update_tree(memcg
, page
);
794 if (unlikely(do_numainfo
))
795 atomic_inc(&memcg
->numainfo_events
);
801 struct mem_cgroup
*mem_cgroup_from_cont(struct cgroup
*cont
)
803 return container_of(cgroup_subsys_state(cont
,
804 mem_cgroup_subsys_id
), struct mem_cgroup
,
808 struct mem_cgroup
*mem_cgroup_from_task(struct task_struct
*p
)
811 * mm_update_next_owner() may clear mm->owner to NULL
812 * if it races with swapoff, page migration, etc.
813 * So this can be called with p == NULL.
818 return container_of(task_subsys_state(p
, mem_cgroup_subsys_id
),
819 struct mem_cgroup
, css
);
822 struct mem_cgroup
*try_get_mem_cgroup_from_mm(struct mm_struct
*mm
)
824 struct mem_cgroup
*memcg
= NULL
;
829 * Because we have no locks, mm->owner's may be being moved to other
830 * cgroup. We use css_tryget() here even if this looks
831 * pessimistic (rather than adding locks here).
835 memcg
= mem_cgroup_from_task(rcu_dereference(mm
->owner
));
836 if (unlikely(!memcg
))
838 } while (!css_tryget(&memcg
->css
));
844 * mem_cgroup_iter - iterate over memory cgroup hierarchy
845 * @root: hierarchy root
846 * @prev: previously returned memcg, NULL on first invocation
847 * @reclaim: cookie for shared reclaim walks, NULL for full walks
849 * Returns references to children of the hierarchy below @root, or
850 * @root itself, or %NULL after a full round-trip.
852 * Caller must pass the return value in @prev on subsequent
853 * invocations for reference counting, or use mem_cgroup_iter_break()
854 * to cancel a hierarchy walk before the round-trip is complete.
856 * Reclaimers can specify a zone and a priority level in @reclaim to
857 * divide up the memcgs in the hierarchy among all concurrent
858 * reclaimers operating on the same zone and priority.
860 struct mem_cgroup
*mem_cgroup_iter(struct mem_cgroup
*root
,
861 struct mem_cgroup
*prev
,
862 struct mem_cgroup_reclaim_cookie
*reclaim
)
864 struct mem_cgroup
*memcg
= NULL
;
867 if (mem_cgroup_disabled())
871 root
= root_mem_cgroup
;
873 if (prev
&& !reclaim
)
874 id
= css_id(&prev
->css
);
876 if (prev
&& prev
!= root
)
879 if (!root
->use_hierarchy
&& root
!= root_mem_cgroup
) {
886 struct mem_cgroup_reclaim_iter
*uninitialized_var(iter
);
887 struct cgroup_subsys_state
*css
;
890 int nid
= zone_to_nid(reclaim
->zone
);
891 int zid
= zone_idx(reclaim
->zone
);
892 struct mem_cgroup_per_zone
*mz
;
894 mz
= mem_cgroup_zoneinfo(root
, nid
, zid
);
895 iter
= &mz
->reclaim_iter
[reclaim
->priority
];
896 if (prev
&& reclaim
->generation
!= iter
->generation
)
902 css
= css_get_next(&mem_cgroup_subsys
, id
+ 1, &root
->css
, &id
);
904 if (css
== &root
->css
|| css_tryget(css
))
905 memcg
= container_of(css
,
906 struct mem_cgroup
, css
);
915 else if (!prev
&& memcg
)
916 reclaim
->generation
= iter
->generation
;
926 * mem_cgroup_iter_break - abort a hierarchy walk prematurely
927 * @root: hierarchy root
928 * @prev: last visited hierarchy member as returned by mem_cgroup_iter()
930 void mem_cgroup_iter_break(struct mem_cgroup
*root
,
931 struct mem_cgroup
*prev
)
934 root
= root_mem_cgroup
;
935 if (prev
&& prev
!= root
)
940 * Iteration constructs for visiting all cgroups (under a tree). If
941 * loops are exited prematurely (break), mem_cgroup_iter_break() must
942 * be used for reference counting.
944 #define for_each_mem_cgroup_tree(iter, root) \
945 for (iter = mem_cgroup_iter(root, NULL, NULL); \
947 iter = mem_cgroup_iter(root, iter, NULL))
949 #define for_each_mem_cgroup(iter) \
950 for (iter = mem_cgroup_iter(NULL, NULL, NULL); \
952 iter = mem_cgroup_iter(NULL, iter, NULL))
954 static inline bool mem_cgroup_is_root(struct mem_cgroup
*memcg
)
956 return (memcg
== root_mem_cgroup
);
959 void mem_cgroup_count_vm_event(struct mm_struct
*mm
, enum vm_event_item idx
)
961 struct mem_cgroup
*memcg
;
967 memcg
= mem_cgroup_from_task(rcu_dereference(mm
->owner
));
968 if (unlikely(!memcg
))
973 this_cpu_inc(memcg
->stat
->events
[MEM_CGROUP_EVENTS_PGFAULT
]);
976 this_cpu_inc(memcg
->stat
->events
[MEM_CGROUP_EVENTS_PGMAJFAULT
]);
984 EXPORT_SYMBOL(mem_cgroup_count_vm_event
);
987 * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg
988 * @zone: zone of the wanted lruvec
989 * @mem: memcg of the wanted lruvec
991 * Returns the lru list vector holding pages for the given @zone and
992 * @mem. This can be the global zone lruvec, if the memory controller
995 struct lruvec
*mem_cgroup_zone_lruvec(struct zone
*zone
,
996 struct mem_cgroup
*memcg
)
998 struct mem_cgroup_per_zone
*mz
;
1000 if (mem_cgroup_disabled())
1001 return &zone
->lruvec
;
1003 mz
= mem_cgroup_zoneinfo(memcg
, zone_to_nid(zone
), zone_idx(zone
));
1008 * Following LRU functions are allowed to be used without PCG_LOCK.
1009 * Operations are called by routine of global LRU independently from memcg.
1010 * What we have to take care of here is validness of pc->mem_cgroup.
1012 * Changes to pc->mem_cgroup happens when
1015 * In typical case, "charge" is done before add-to-lru. Exception is SwapCache.
1016 * It is added to LRU before charge.
1017 * If PCG_USED bit is not set, page_cgroup is not added to this private LRU.
1018 * When moving account, the page is not on LRU. It's isolated.
1022 * mem_cgroup_lru_add_list - account for adding an lru page and return lruvec
1023 * @zone: zone of the page
1027 * This function accounts for @page being added to @lru, and returns
1028 * the lruvec for the given @zone and the memcg @page is charged to.
1030 * The callsite is then responsible for physically linking the page to
1031 * the returned lruvec->lists[@lru].
1033 struct lruvec
*mem_cgroup_lru_add_list(struct zone
*zone
, struct page
*page
,
1036 struct mem_cgroup_per_zone
*mz
;
1037 struct mem_cgroup
*memcg
;
1038 struct page_cgroup
*pc
;
1040 if (mem_cgroup_disabled())
1041 return &zone
->lruvec
;
1043 pc
= lookup_page_cgroup(page
);
1044 memcg
= pc
->mem_cgroup
;
1045 mz
= page_cgroup_zoneinfo(memcg
, page
);
1046 /* compound_order() is stabilized through lru_lock */
1047 MEM_CGROUP_ZSTAT(mz
, lru
) += 1 << compound_order(page
);
1052 * mem_cgroup_lru_del_list - account for removing an lru page
1056 * This function accounts for @page being removed from @lru.
1058 * The callsite is then responsible for physically unlinking
1061 void mem_cgroup_lru_del_list(struct page
*page
, enum lru_list lru
)
1063 struct mem_cgroup_per_zone
*mz
;
1064 struct mem_cgroup
*memcg
;
1065 struct page_cgroup
*pc
;
1067 if (mem_cgroup_disabled())
1070 pc
= lookup_page_cgroup(page
);
1071 memcg
= pc
->mem_cgroup
;
1073 mz
= page_cgroup_zoneinfo(memcg
, page
);
1074 /* huge page split is done under lru_lock. so, we have no races. */
1075 VM_BUG_ON(MEM_CGROUP_ZSTAT(mz
, lru
) < (1 << compound_order(page
)));
1076 MEM_CGROUP_ZSTAT(mz
, lru
) -= 1 << compound_order(page
);
1079 void mem_cgroup_lru_del(struct page
*page
)
1081 mem_cgroup_lru_del_list(page
, page_lru(page
));
1085 * mem_cgroup_lru_move_lists - account for moving a page between lrus
1086 * @zone: zone of the page
1088 * @from: current lru
1091 * This function accounts for @page being moved between the lrus @from
1092 * and @to, and returns the lruvec for the given @zone and the memcg
1093 * @page is charged to.
1095 * The callsite is then responsible for physically relinking
1096 * @page->lru to the returned lruvec->lists[@to].
1098 struct lruvec
*mem_cgroup_lru_move_lists(struct zone
*zone
,
1103 /* XXX: Optimize this, especially for @from == @to */
1104 mem_cgroup_lru_del_list(page
, from
);
1105 return mem_cgroup_lru_add_list(zone
, page
, to
);
1109 * Checks whether given mem is same or in the root_mem_cgroup's
1112 static bool mem_cgroup_same_or_subtree(const struct mem_cgroup
*root_memcg
,
1113 struct mem_cgroup
*memcg
)
1115 if (root_memcg
!= memcg
) {
1116 return (root_memcg
->use_hierarchy
&&
1117 css_is_ancestor(&memcg
->css
, &root_memcg
->css
));
1123 int task_in_mem_cgroup(struct task_struct
*task
, const struct mem_cgroup
*memcg
)
1126 struct mem_cgroup
*curr
= NULL
;
1127 struct task_struct
*p
;
1129 p
= find_lock_task_mm(task
);
1131 curr
= try_get_mem_cgroup_from_mm(p
->mm
);
1135 * All threads may have already detached their mm's, but the oom
1136 * killer still needs to detect if they have already been oom
1137 * killed to prevent needlessly killing additional tasks.
1140 curr
= mem_cgroup_from_task(task
);
1142 css_get(&curr
->css
);
1148 * We should check use_hierarchy of "memcg" not "curr". Because checking
1149 * use_hierarchy of "curr" here make this function true if hierarchy is
1150 * enabled in "curr" and "curr" is a child of "memcg" in *cgroup*
1151 * hierarchy(even if use_hierarchy is disabled in "memcg").
1153 ret
= mem_cgroup_same_or_subtree(memcg
, curr
);
1154 css_put(&curr
->css
);
1158 int mem_cgroup_inactive_anon_is_low(struct mem_cgroup
*memcg
, struct zone
*zone
)
1160 unsigned long inactive_ratio
;
1161 int nid
= zone_to_nid(zone
);
1162 int zid
= zone_idx(zone
);
1163 unsigned long inactive
;
1164 unsigned long active
;
1167 inactive
= mem_cgroup_zone_nr_lru_pages(memcg
, nid
, zid
,
1168 BIT(LRU_INACTIVE_ANON
));
1169 active
= mem_cgroup_zone_nr_lru_pages(memcg
, nid
, zid
,
1170 BIT(LRU_ACTIVE_ANON
));
1172 gb
= (inactive
+ active
) >> (30 - PAGE_SHIFT
);
1174 inactive_ratio
= int_sqrt(10 * gb
);
1178 return inactive
* inactive_ratio
< active
;
1181 int mem_cgroup_inactive_file_is_low(struct mem_cgroup
*memcg
, struct zone
*zone
)
1183 unsigned long active
;
1184 unsigned long inactive
;
1185 int zid
= zone_idx(zone
);
1186 int nid
= zone_to_nid(zone
);
1188 inactive
= mem_cgroup_zone_nr_lru_pages(memcg
, nid
, zid
,
1189 BIT(LRU_INACTIVE_FILE
));
1190 active
= mem_cgroup_zone_nr_lru_pages(memcg
, nid
, zid
,
1191 BIT(LRU_ACTIVE_FILE
));
1193 return (active
> inactive
);
1196 struct zone_reclaim_stat
*mem_cgroup_get_reclaim_stat(struct mem_cgroup
*memcg
,
1199 int nid
= zone_to_nid(zone
);
1200 int zid
= zone_idx(zone
);
1201 struct mem_cgroup_per_zone
*mz
= mem_cgroup_zoneinfo(memcg
, nid
, zid
);
1203 return &mz
->reclaim_stat
;
1206 struct zone_reclaim_stat
*
1207 mem_cgroup_get_reclaim_stat_from_page(struct page
*page
)
1209 struct page_cgroup
*pc
;
1210 struct mem_cgroup_per_zone
*mz
;
1212 if (mem_cgroup_disabled())
1215 pc
= lookup_page_cgroup(page
);
1216 if (!PageCgroupUsed(pc
))
1218 /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */
1220 mz
= page_cgroup_zoneinfo(pc
->mem_cgroup
, page
);
1221 return &mz
->reclaim_stat
;
1224 #define mem_cgroup_from_res_counter(counter, member) \
1225 container_of(counter, struct mem_cgroup, member)
1228 * mem_cgroup_margin - calculate chargeable space of a memory cgroup
1229 * @mem: the memory cgroup
1231 * Returns the maximum amount of memory @mem can be charged with, in
1234 static unsigned long mem_cgroup_margin(struct mem_cgroup
*memcg
)
1236 unsigned long long margin
;
1238 margin
= res_counter_margin(&memcg
->res
);
1239 if (do_swap_account
)
1240 margin
= min(margin
, res_counter_margin(&memcg
->memsw
));
1241 return margin
>> PAGE_SHIFT
;
1244 int mem_cgroup_swappiness(struct mem_cgroup
*memcg
)
1246 struct cgroup
*cgrp
= memcg
->css
.cgroup
;
1249 if (cgrp
->parent
== NULL
)
1250 return vm_swappiness
;
1252 return memcg
->swappiness
;
1255 static void mem_cgroup_start_move(struct mem_cgroup
*memcg
)
1260 spin_lock(&memcg
->pcp_counter_lock
);
1261 for_each_online_cpu(cpu
)
1262 per_cpu(memcg
->stat
->count
[MEM_CGROUP_ON_MOVE
], cpu
) += 1;
1263 memcg
->nocpu_base
.count
[MEM_CGROUP_ON_MOVE
] += 1;
1264 spin_unlock(&memcg
->pcp_counter_lock
);
1270 static void mem_cgroup_end_move(struct mem_cgroup
*memcg
)
1277 spin_lock(&memcg
->pcp_counter_lock
);
1278 for_each_online_cpu(cpu
)
1279 per_cpu(memcg
->stat
->count
[MEM_CGROUP_ON_MOVE
], cpu
) -= 1;
1280 memcg
->nocpu_base
.count
[MEM_CGROUP_ON_MOVE
] -= 1;
1281 spin_unlock(&memcg
->pcp_counter_lock
);
1285 * 2 routines for checking "mem" is under move_account() or not.
1287 * mem_cgroup_stealed() - checking a cgroup is mc.from or not. This is used
1288 * for avoiding race in accounting. If true,
1289 * pc->mem_cgroup may be overwritten.
1291 * mem_cgroup_under_move() - checking a cgroup is mc.from or mc.to or
1292 * under hierarchy of moving cgroups. This is for
1293 * waiting at hith-memory prressure caused by "move".
1296 static bool mem_cgroup_stealed(struct mem_cgroup
*memcg
)
1298 VM_BUG_ON(!rcu_read_lock_held());
1299 return this_cpu_read(memcg
->stat
->count
[MEM_CGROUP_ON_MOVE
]) > 0;
1302 static bool mem_cgroup_under_move(struct mem_cgroup
*memcg
)
1304 struct mem_cgroup
*from
;
1305 struct mem_cgroup
*to
;
1308 * Unlike task_move routines, we access mc.to, mc.from not under
1309 * mutual exclusion by cgroup_mutex. Here, we take spinlock instead.
1311 spin_lock(&mc
.lock
);
1317 ret
= mem_cgroup_same_or_subtree(memcg
, from
)
1318 || mem_cgroup_same_or_subtree(memcg
, to
);
1320 spin_unlock(&mc
.lock
);
1324 static bool mem_cgroup_wait_acct_move(struct mem_cgroup
*memcg
)
1326 if (mc
.moving_task
&& current
!= mc
.moving_task
) {
1327 if (mem_cgroup_under_move(memcg
)) {
1329 prepare_to_wait(&mc
.waitq
, &wait
, TASK_INTERRUPTIBLE
);
1330 /* moving charge context might have finished. */
1333 finish_wait(&mc
.waitq
, &wait
);
1341 * mem_cgroup_print_oom_info: Called from OOM with tasklist_lock held in read mode.
1342 * @memcg: The memory cgroup that went over limit
1343 * @p: Task that is going to be killed
1345 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1348 void mem_cgroup_print_oom_info(struct mem_cgroup
*memcg
, struct task_struct
*p
)
1350 struct cgroup
*task_cgrp
;
1351 struct cgroup
*mem_cgrp
;
1353 * Need a buffer in BSS, can't rely on allocations. The code relies
1354 * on the assumption that OOM is serialized for memory controller.
1355 * If this assumption is broken, revisit this code.
1357 static char memcg_name
[PATH_MAX
];
1366 mem_cgrp
= memcg
->css
.cgroup
;
1367 task_cgrp
= task_cgroup(p
, mem_cgroup_subsys_id
);
1369 ret
= cgroup_path(task_cgrp
, memcg_name
, PATH_MAX
);
1372 * Unfortunately, we are unable to convert to a useful name
1373 * But we'll still print out the usage information
1380 printk(KERN_INFO
"Task in %s killed", memcg_name
);
1383 ret
= cgroup_path(mem_cgrp
, memcg_name
, PATH_MAX
);
1391 * Continues from above, so we don't need an KERN_ level
1393 printk(KERN_CONT
" as a result of limit of %s\n", memcg_name
);
1396 printk(KERN_INFO
"memory: usage %llukB, limit %llukB, failcnt %llu\n",
1397 res_counter_read_u64(&memcg
->res
, RES_USAGE
) >> 10,
1398 res_counter_read_u64(&memcg
->res
, RES_LIMIT
) >> 10,
1399 res_counter_read_u64(&memcg
->res
, RES_FAILCNT
));
1400 printk(KERN_INFO
"memory+swap: usage %llukB, limit %llukB, "
1402 res_counter_read_u64(&memcg
->memsw
, RES_USAGE
) >> 10,
1403 res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
) >> 10,
1404 res_counter_read_u64(&memcg
->memsw
, RES_FAILCNT
));
1408 * This function returns the number of memcg under hierarchy tree. Returns
1409 * 1(self count) if no children.
1411 static int mem_cgroup_count_children(struct mem_cgroup
*memcg
)
1414 struct mem_cgroup
*iter
;
1416 for_each_mem_cgroup_tree(iter
, memcg
)
1422 * Return the memory (and swap, if configured) limit for a memcg.
1424 u64
mem_cgroup_get_limit(struct mem_cgroup
*memcg
)
1429 limit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
1430 limit
+= total_swap_pages
<< PAGE_SHIFT
;
1432 memsw
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
1434 * If memsw is finite and limits the amount of swap space available
1435 * to this memcg, return that limit.
1437 return min(limit
, memsw
);
1440 static unsigned long mem_cgroup_reclaim(struct mem_cgroup
*memcg
,
1442 unsigned long flags
)
1444 unsigned long total
= 0;
1445 bool noswap
= false;
1448 if (flags
& MEM_CGROUP_RECLAIM_NOSWAP
)
1450 if (!(flags
& MEM_CGROUP_RECLAIM_SHRINK
) && memcg
->memsw_is_minimum
)
1453 for (loop
= 0; loop
< MEM_CGROUP_MAX_RECLAIM_LOOPS
; loop
++) {
1455 drain_all_stock_async(memcg
);
1456 total
+= try_to_free_mem_cgroup_pages(memcg
, gfp_mask
, noswap
);
1458 * Allow limit shrinkers, which are triggered directly
1459 * by userspace, to catch signals and stop reclaim
1460 * after minimal progress, regardless of the margin.
1462 if (total
&& (flags
& MEM_CGROUP_RECLAIM_SHRINK
))
1464 if (mem_cgroup_margin(memcg
))
1467 * If nothing was reclaimed after two attempts, there
1468 * may be no reclaimable pages in this hierarchy.
1477 * test_mem_cgroup_node_reclaimable
1478 * @mem: the target memcg
1479 * @nid: the node ID to be checked.
1480 * @noswap : specify true here if the user wants flle only information.
1482 * This function returns whether the specified memcg contains any
1483 * reclaimable pages on a node. Returns true if there are any reclaimable
1484 * pages in the node.
1486 static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup
*memcg
,
1487 int nid
, bool noswap
)
1489 if (mem_cgroup_node_nr_lru_pages(memcg
, nid
, LRU_ALL_FILE
))
1491 if (noswap
|| !total_swap_pages
)
1493 if (mem_cgroup_node_nr_lru_pages(memcg
, nid
, LRU_ALL_ANON
))
1498 #if MAX_NUMNODES > 1
1501 * Always updating the nodemask is not very good - even if we have an empty
1502 * list or the wrong list here, we can start from some node and traverse all
1503 * nodes based on the zonelist. So update the list loosely once per 10 secs.
1506 static void mem_cgroup_may_update_nodemask(struct mem_cgroup
*memcg
)
1510 * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET
1511 * pagein/pageout changes since the last update.
1513 if (!atomic_read(&memcg
->numainfo_events
))
1515 if (atomic_inc_return(&memcg
->numainfo_updating
) > 1)
1518 /* make a nodemask where this memcg uses memory from */
1519 memcg
->scan_nodes
= node_states
[N_HIGH_MEMORY
];
1521 for_each_node_mask(nid
, node_states
[N_HIGH_MEMORY
]) {
1523 if (!test_mem_cgroup_node_reclaimable(memcg
, nid
, false))
1524 node_clear(nid
, memcg
->scan_nodes
);
1527 atomic_set(&memcg
->numainfo_events
, 0);
1528 atomic_set(&memcg
->numainfo_updating
, 0);
1532 * Selecting a node where we start reclaim from. Because what we need is just
1533 * reducing usage counter, start from anywhere is O,K. Considering
1534 * memory reclaim from current node, there are pros. and cons.
1536 * Freeing memory from current node means freeing memory from a node which
1537 * we'll use or we've used. So, it may make LRU bad. And if several threads
1538 * hit limits, it will see a contention on a node. But freeing from remote
1539 * node means more costs for memory reclaim because of memory latency.
1541 * Now, we use round-robin. Better algorithm is welcomed.
1543 int mem_cgroup_select_victim_node(struct mem_cgroup
*memcg
)
1547 mem_cgroup_may_update_nodemask(memcg
);
1548 node
= memcg
->last_scanned_node
;
1550 node
= next_node(node
, memcg
->scan_nodes
);
1551 if (node
== MAX_NUMNODES
)
1552 node
= first_node(memcg
->scan_nodes
);
1554 * We call this when we hit limit, not when pages are added to LRU.
1555 * No LRU may hold pages because all pages are UNEVICTABLE or
1556 * memcg is too small and all pages are not on LRU. In that case,
1557 * we use curret node.
1559 if (unlikely(node
== MAX_NUMNODES
))
1560 node
= numa_node_id();
1562 memcg
->last_scanned_node
= node
;
1567 * Check all nodes whether it contains reclaimable pages or not.
1568 * For quick scan, we make use of scan_nodes. This will allow us to skip
1569 * unused nodes. But scan_nodes is lazily updated and may not cotain
1570 * enough new information. We need to do double check.
1572 bool mem_cgroup_reclaimable(struct mem_cgroup
*memcg
, bool noswap
)
1577 * quick check...making use of scan_node.
1578 * We can skip unused nodes.
1580 if (!nodes_empty(memcg
->scan_nodes
)) {
1581 for (nid
= first_node(memcg
->scan_nodes
);
1583 nid
= next_node(nid
, memcg
->scan_nodes
)) {
1585 if (test_mem_cgroup_node_reclaimable(memcg
, nid
, noswap
))
1590 * Check rest of nodes.
1592 for_each_node_state(nid
, N_HIGH_MEMORY
) {
1593 if (node_isset(nid
, memcg
->scan_nodes
))
1595 if (test_mem_cgroup_node_reclaimable(memcg
, nid
, noswap
))
1602 int mem_cgroup_select_victim_node(struct mem_cgroup
*memcg
)
1607 bool mem_cgroup_reclaimable(struct mem_cgroup
*memcg
, bool noswap
)
1609 return test_mem_cgroup_node_reclaimable(memcg
, 0, noswap
);
1613 static int mem_cgroup_soft_reclaim(struct mem_cgroup
*root_memcg
,
1616 unsigned long *total_scanned
)
1618 struct mem_cgroup
*victim
= NULL
;
1621 unsigned long excess
;
1622 unsigned long nr_scanned
;
1623 struct mem_cgroup_reclaim_cookie reclaim
= {
1628 excess
= res_counter_soft_limit_excess(&root_memcg
->res
) >> PAGE_SHIFT
;
1631 victim
= mem_cgroup_iter(root_memcg
, victim
, &reclaim
);
1636 * If we have not been able to reclaim
1637 * anything, it might because there are
1638 * no reclaimable pages under this hierarchy
1643 * We want to do more targeted reclaim.
1644 * excess >> 2 is not to excessive so as to
1645 * reclaim too much, nor too less that we keep
1646 * coming back to reclaim from this cgroup
1648 if (total
>= (excess
>> 2) ||
1649 (loop
> MEM_CGROUP_MAX_RECLAIM_LOOPS
))
1654 if (!mem_cgroup_reclaimable(victim
, false))
1656 total
+= mem_cgroup_shrink_node_zone(victim
, gfp_mask
, false,
1658 *total_scanned
+= nr_scanned
;
1659 if (!res_counter_soft_limit_excess(&root_memcg
->res
))
1662 mem_cgroup_iter_break(root_memcg
, victim
);
1667 * Check OOM-Killer is already running under our hierarchy.
1668 * If someone is running, return false.
1669 * Has to be called with memcg_oom_lock
1671 static bool mem_cgroup_oom_lock(struct mem_cgroup
*memcg
)
1673 struct mem_cgroup
*iter
, *failed
= NULL
;
1675 for_each_mem_cgroup_tree(iter
, memcg
) {
1676 if (iter
->oom_lock
) {
1678 * this subtree of our hierarchy is already locked
1679 * so we cannot give a lock.
1682 mem_cgroup_iter_break(memcg
, iter
);
1685 iter
->oom_lock
= true;
1692 * OK, we failed to lock the whole subtree so we have to clean up
1693 * what we set up to the failing subtree
1695 for_each_mem_cgroup_tree(iter
, memcg
) {
1696 if (iter
== failed
) {
1697 mem_cgroup_iter_break(memcg
, iter
);
1700 iter
->oom_lock
= false;
1706 * Has to be called with memcg_oom_lock
1708 static int mem_cgroup_oom_unlock(struct mem_cgroup
*memcg
)
1710 struct mem_cgroup
*iter
;
1712 for_each_mem_cgroup_tree(iter
, memcg
)
1713 iter
->oom_lock
= false;
1717 static void mem_cgroup_mark_under_oom(struct mem_cgroup
*memcg
)
1719 struct mem_cgroup
*iter
;
1721 for_each_mem_cgroup_tree(iter
, memcg
)
1722 atomic_inc(&iter
->under_oom
);
1725 static void mem_cgroup_unmark_under_oom(struct mem_cgroup
*memcg
)
1727 struct mem_cgroup
*iter
;
1730 * When a new child is created while the hierarchy is under oom,
1731 * mem_cgroup_oom_lock() may not be called. We have to use
1732 * atomic_add_unless() here.
1734 for_each_mem_cgroup_tree(iter
, memcg
)
1735 atomic_add_unless(&iter
->under_oom
, -1, 0);
1738 static DEFINE_SPINLOCK(memcg_oom_lock
);
1739 static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq
);
1741 struct oom_wait_info
{
1742 struct mem_cgroup
*mem
;
1746 static int memcg_oom_wake_function(wait_queue_t
*wait
,
1747 unsigned mode
, int sync
, void *arg
)
1749 struct mem_cgroup
*wake_memcg
= (struct mem_cgroup
*)arg
,
1751 struct oom_wait_info
*oom_wait_info
;
1753 oom_wait_info
= container_of(wait
, struct oom_wait_info
, wait
);
1754 oom_wait_memcg
= oom_wait_info
->mem
;
1757 * Both of oom_wait_info->mem and wake_mem are stable under us.
1758 * Then we can use css_is_ancestor without taking care of RCU.
1760 if (!mem_cgroup_same_or_subtree(oom_wait_memcg
, wake_memcg
)
1761 && !mem_cgroup_same_or_subtree(wake_memcg
, oom_wait_memcg
))
1763 return autoremove_wake_function(wait
, mode
, sync
, arg
);
1766 static void memcg_wakeup_oom(struct mem_cgroup
*memcg
)
1768 /* for filtering, pass "memcg" as argument. */
1769 __wake_up(&memcg_oom_waitq
, TASK_NORMAL
, 0, memcg
);
1772 static void memcg_oom_recover(struct mem_cgroup
*memcg
)
1774 if (memcg
&& atomic_read(&memcg
->under_oom
))
1775 memcg_wakeup_oom(memcg
);
1779 * try to call OOM killer. returns false if we should exit memory-reclaim loop.
1781 bool mem_cgroup_handle_oom(struct mem_cgroup
*memcg
, gfp_t mask
)
1783 struct oom_wait_info owait
;
1784 bool locked
, need_to_kill
;
1787 owait
.wait
.flags
= 0;
1788 owait
.wait
.func
= memcg_oom_wake_function
;
1789 owait
.wait
.private = current
;
1790 INIT_LIST_HEAD(&owait
.wait
.task_list
);
1791 need_to_kill
= true;
1792 mem_cgroup_mark_under_oom(memcg
);
1794 /* At first, try to OOM lock hierarchy under memcg.*/
1795 spin_lock(&memcg_oom_lock
);
1796 locked
= mem_cgroup_oom_lock(memcg
);
1798 * Even if signal_pending(), we can't quit charge() loop without
1799 * accounting. So, UNINTERRUPTIBLE is appropriate. But SIGKILL
1800 * under OOM is always welcomed, use TASK_KILLABLE here.
1802 prepare_to_wait(&memcg_oom_waitq
, &owait
.wait
, TASK_KILLABLE
);
1803 if (!locked
|| memcg
->oom_kill_disable
)
1804 need_to_kill
= false;
1806 mem_cgroup_oom_notify(memcg
);
1807 spin_unlock(&memcg_oom_lock
);
1810 finish_wait(&memcg_oom_waitq
, &owait
.wait
);
1811 mem_cgroup_out_of_memory(memcg
, mask
);
1814 finish_wait(&memcg_oom_waitq
, &owait
.wait
);
1816 spin_lock(&memcg_oom_lock
);
1818 mem_cgroup_oom_unlock(memcg
);
1819 memcg_wakeup_oom(memcg
);
1820 spin_unlock(&memcg_oom_lock
);
1822 mem_cgroup_unmark_under_oom(memcg
);
1824 if (test_thread_flag(TIF_MEMDIE
) || fatal_signal_pending(current
))
1826 /* Give chance to dying process */
1827 schedule_timeout_uninterruptible(1);
1832 * Currently used to update mapped file statistics, but the routine can be
1833 * generalized to update other statistics as well.
1835 * Notes: Race condition
1837 * We usually use page_cgroup_lock() for accessing page_cgroup member but
1838 * it tends to be costly. But considering some conditions, we doesn't need
1839 * to do so _always_.
1841 * Considering "charge", lock_page_cgroup() is not required because all
1842 * file-stat operations happen after a page is attached to radix-tree. There
1843 * are no race with "charge".
1845 * Considering "uncharge", we know that memcg doesn't clear pc->mem_cgroup
1846 * at "uncharge" intentionally. So, we always see valid pc->mem_cgroup even
1847 * if there are race with "uncharge". Statistics itself is properly handled
1850 * Considering "move", this is an only case we see a race. To make the race
1851 * small, we check MEM_CGROUP_ON_MOVE percpu value and detect there are
1852 * possibility of race condition. If there is, we take a lock.
1855 void mem_cgroup_update_page_stat(struct page
*page
,
1856 enum mem_cgroup_page_stat_item idx
, int val
)
1858 struct mem_cgroup
*memcg
;
1859 struct page_cgroup
*pc
= lookup_page_cgroup(page
);
1860 bool need_unlock
= false;
1861 unsigned long uninitialized_var(flags
);
1863 if (mem_cgroup_disabled())
1867 memcg
= pc
->mem_cgroup
;
1868 if (unlikely(!memcg
|| !PageCgroupUsed(pc
)))
1870 /* pc->mem_cgroup is unstable ? */
1871 if (unlikely(mem_cgroup_stealed(memcg
)) || PageTransHuge(page
)) {
1872 /* take a lock against to access pc->mem_cgroup */
1873 move_lock_page_cgroup(pc
, &flags
);
1875 memcg
= pc
->mem_cgroup
;
1876 if (!memcg
|| !PageCgroupUsed(pc
))
1881 case MEMCG_NR_FILE_MAPPED
:
1883 SetPageCgroupFileMapped(pc
);
1884 else if (!page_mapped(page
))
1885 ClearPageCgroupFileMapped(pc
);
1886 idx
= MEM_CGROUP_STAT_FILE_MAPPED
;
1892 this_cpu_add(memcg
->stat
->count
[idx
], val
);
1895 if (unlikely(need_unlock
))
1896 move_unlock_page_cgroup(pc
, &flags
);
1900 EXPORT_SYMBOL(mem_cgroup_update_page_stat
);
1903 * size of first charge trial. "32" comes from vmscan.c's magic value.
1904 * TODO: maybe necessary to use big numbers in big irons.
1906 #define CHARGE_BATCH 32U
1907 struct memcg_stock_pcp
{
1908 struct mem_cgroup
*cached
; /* this never be root cgroup */
1909 unsigned int nr_pages
;
1910 struct work_struct work
;
1911 unsigned long flags
;
1912 #define FLUSHING_CACHED_CHARGE (0)
1914 static DEFINE_PER_CPU(struct memcg_stock_pcp
, memcg_stock
);
1915 static DEFINE_MUTEX(percpu_charge_mutex
);
1918 * Try to consume stocked charge on this cpu. If success, one page is consumed
1919 * from local stock and true is returned. If the stock is 0 or charges from a
1920 * cgroup which is not current target, returns false. This stock will be
1923 static bool consume_stock(struct mem_cgroup
*memcg
)
1925 struct memcg_stock_pcp
*stock
;
1928 stock
= &get_cpu_var(memcg_stock
);
1929 if (memcg
== stock
->cached
&& stock
->nr_pages
)
1931 else /* need to call res_counter_charge */
1933 put_cpu_var(memcg_stock
);
1938 * Returns stocks cached in percpu to res_counter and reset cached information.
1940 static void drain_stock(struct memcg_stock_pcp
*stock
)
1942 struct mem_cgroup
*old
= stock
->cached
;
1944 if (stock
->nr_pages
) {
1945 unsigned long bytes
= stock
->nr_pages
* PAGE_SIZE
;
1947 res_counter_uncharge(&old
->res
, bytes
);
1948 if (do_swap_account
)
1949 res_counter_uncharge(&old
->memsw
, bytes
);
1950 stock
->nr_pages
= 0;
1952 stock
->cached
= NULL
;
1956 * This must be called under preempt disabled or must be called by
1957 * a thread which is pinned to local cpu.
1959 static void drain_local_stock(struct work_struct
*dummy
)
1961 struct memcg_stock_pcp
*stock
= &__get_cpu_var(memcg_stock
);
1963 clear_bit(FLUSHING_CACHED_CHARGE
, &stock
->flags
);
1967 * Cache charges(val) which is from res_counter, to local per_cpu area.
1968 * This will be consumed by consume_stock() function, later.
1970 static void refill_stock(struct mem_cgroup
*memcg
, unsigned int nr_pages
)
1972 struct memcg_stock_pcp
*stock
= &get_cpu_var(memcg_stock
);
1974 if (stock
->cached
!= memcg
) { /* reset if necessary */
1976 stock
->cached
= memcg
;
1978 stock
->nr_pages
+= nr_pages
;
1979 put_cpu_var(memcg_stock
);
1983 * Drains all per-CPU charge caches for given root_memcg resp. subtree
1984 * of the hierarchy under it. sync flag says whether we should block
1985 * until the work is done.
1987 static void drain_all_stock(struct mem_cgroup
*root_memcg
, bool sync
)
1991 /* Notify other cpus that system-wide "drain" is running */
1994 for_each_online_cpu(cpu
) {
1995 struct memcg_stock_pcp
*stock
= &per_cpu(memcg_stock
, cpu
);
1996 struct mem_cgroup
*memcg
;
1998 memcg
= stock
->cached
;
1999 if (!memcg
|| !stock
->nr_pages
)
2001 if (!mem_cgroup_same_or_subtree(root_memcg
, memcg
))
2003 if (!test_and_set_bit(FLUSHING_CACHED_CHARGE
, &stock
->flags
)) {
2005 drain_local_stock(&stock
->work
);
2007 schedule_work_on(cpu
, &stock
->work
);
2015 for_each_online_cpu(cpu
) {
2016 struct memcg_stock_pcp
*stock
= &per_cpu(memcg_stock
, cpu
);
2017 if (test_bit(FLUSHING_CACHED_CHARGE
, &stock
->flags
))
2018 flush_work(&stock
->work
);
2025 * Tries to drain stocked charges in other cpus. This function is asynchronous
2026 * and just put a work per cpu for draining localy on each cpu. Caller can
2027 * expects some charges will be back to res_counter later but cannot wait for
2030 static void drain_all_stock_async(struct mem_cgroup
*root_memcg
)
2033 * If someone calls draining, avoid adding more kworker runs.
2035 if (!mutex_trylock(&percpu_charge_mutex
))
2037 drain_all_stock(root_memcg
, false);
2038 mutex_unlock(&percpu_charge_mutex
);
2041 /* This is a synchronous drain interface. */
2042 static void drain_all_stock_sync(struct mem_cgroup
*root_memcg
)
2044 /* called when force_empty is called */
2045 mutex_lock(&percpu_charge_mutex
);
2046 drain_all_stock(root_memcg
, true);
2047 mutex_unlock(&percpu_charge_mutex
);
2051 * This function drains percpu counter value from DEAD cpu and
2052 * move it to local cpu. Note that this function can be preempted.
2054 static void mem_cgroup_drain_pcp_counter(struct mem_cgroup
*memcg
, int cpu
)
2058 spin_lock(&memcg
->pcp_counter_lock
);
2059 for (i
= 0; i
< MEM_CGROUP_STAT_DATA
; i
++) {
2060 long x
= per_cpu(memcg
->stat
->count
[i
], cpu
);
2062 per_cpu(memcg
->stat
->count
[i
], cpu
) = 0;
2063 memcg
->nocpu_base
.count
[i
] += x
;
2065 for (i
= 0; i
< MEM_CGROUP_EVENTS_NSTATS
; i
++) {
2066 unsigned long x
= per_cpu(memcg
->stat
->events
[i
], cpu
);
2068 per_cpu(memcg
->stat
->events
[i
], cpu
) = 0;
2069 memcg
->nocpu_base
.events
[i
] += x
;
2071 /* need to clear ON_MOVE value, works as a kind of lock. */
2072 per_cpu(memcg
->stat
->count
[MEM_CGROUP_ON_MOVE
], cpu
) = 0;
2073 spin_unlock(&memcg
->pcp_counter_lock
);
2076 static void synchronize_mem_cgroup_on_move(struct mem_cgroup
*memcg
, int cpu
)
2078 int idx
= MEM_CGROUP_ON_MOVE
;
2080 spin_lock(&memcg
->pcp_counter_lock
);
2081 per_cpu(memcg
->stat
->count
[idx
], cpu
) = memcg
->nocpu_base
.count
[idx
];
2082 spin_unlock(&memcg
->pcp_counter_lock
);
2085 static int __cpuinit
memcg_cpu_hotplug_callback(struct notifier_block
*nb
,
2086 unsigned long action
,
2089 int cpu
= (unsigned long)hcpu
;
2090 struct memcg_stock_pcp
*stock
;
2091 struct mem_cgroup
*iter
;
2093 if ((action
== CPU_ONLINE
)) {
2094 for_each_mem_cgroup(iter
)
2095 synchronize_mem_cgroup_on_move(iter
, cpu
);
2099 if ((action
!= CPU_DEAD
) || action
!= CPU_DEAD_FROZEN
)
2102 for_each_mem_cgroup(iter
)
2103 mem_cgroup_drain_pcp_counter(iter
, cpu
);
2105 stock
= &per_cpu(memcg_stock
, cpu
);
2111 /* See __mem_cgroup_try_charge() for details */
2113 CHARGE_OK
, /* success */
2114 CHARGE_RETRY
, /* need to retry but retry is not bad */
2115 CHARGE_NOMEM
, /* we can't do more. return -ENOMEM */
2116 CHARGE_WOULDBLOCK
, /* GFP_WAIT wasn't set and no enough res. */
2117 CHARGE_OOM_DIE
, /* the current is killed because of OOM */
2120 static int mem_cgroup_do_charge(struct mem_cgroup
*memcg
, gfp_t gfp_mask
,
2121 unsigned int nr_pages
, bool oom_check
)
2123 unsigned long csize
= nr_pages
* PAGE_SIZE
;
2124 struct mem_cgroup
*mem_over_limit
;
2125 struct res_counter
*fail_res
;
2126 unsigned long flags
= 0;
2129 ret
= res_counter_charge(&memcg
->res
, csize
, &fail_res
);
2132 if (!do_swap_account
)
2134 ret
= res_counter_charge(&memcg
->memsw
, csize
, &fail_res
);
2138 res_counter_uncharge(&memcg
->res
, csize
);
2139 mem_over_limit
= mem_cgroup_from_res_counter(fail_res
, memsw
);
2140 flags
|= MEM_CGROUP_RECLAIM_NOSWAP
;
2142 mem_over_limit
= mem_cgroup_from_res_counter(fail_res
, res
);
2144 * nr_pages can be either a huge page (HPAGE_PMD_NR), a batch
2145 * of regular pages (CHARGE_BATCH), or a single regular page (1).
2147 * Never reclaim on behalf of optional batching, retry with a
2148 * single page instead.
2150 if (nr_pages
== CHARGE_BATCH
)
2151 return CHARGE_RETRY
;
2153 if (!(gfp_mask
& __GFP_WAIT
))
2154 return CHARGE_WOULDBLOCK
;
2156 ret
= mem_cgroup_reclaim(mem_over_limit
, gfp_mask
, flags
);
2157 if (mem_cgroup_margin(mem_over_limit
) >= nr_pages
)
2158 return CHARGE_RETRY
;
2160 * Even though the limit is exceeded at this point, reclaim
2161 * may have been able to free some pages. Retry the charge
2162 * before killing the task.
2164 * Only for regular pages, though: huge pages are rather
2165 * unlikely to succeed so close to the limit, and we fall back
2166 * to regular pages anyway in case of failure.
2168 if (nr_pages
== 1 && ret
)
2169 return CHARGE_RETRY
;
2172 * At task move, charge accounts can be doubly counted. So, it's
2173 * better to wait until the end of task_move if something is going on.
2175 if (mem_cgroup_wait_acct_move(mem_over_limit
))
2176 return CHARGE_RETRY
;
2178 /* If we don't need to call oom-killer at el, return immediately */
2180 return CHARGE_NOMEM
;
2182 if (!mem_cgroup_handle_oom(mem_over_limit
, gfp_mask
))
2183 return CHARGE_OOM_DIE
;
2185 return CHARGE_RETRY
;
2189 * __mem_cgroup_try_charge() does
2190 * 1. detect memcg to be charged against from passed *mm and *ptr,
2191 * 2. update res_counter
2192 * 3. call memory reclaim if necessary.
2194 * In some special case, if the task is fatal, fatal_signal_pending() or
2195 * has TIF_MEMDIE, this function returns -EINTR while writing root_mem_cgroup
2196 * to *ptr. There are two reasons for this. 1: fatal threads should quit as soon
2197 * as possible without any hazards. 2: all pages should have a valid
2198 * pc->mem_cgroup. If mm is NULL and the caller doesn't pass a valid memcg
2199 * pointer, that is treated as a charge to root_mem_cgroup.
2201 * So __mem_cgroup_try_charge() will return
2202 * 0 ... on success, filling *ptr with a valid memcg pointer.
2203 * -ENOMEM ... charge failure because of resource limits.
2204 * -EINTR ... if thread is fatal. *ptr is filled with root_mem_cgroup.
2206 * Unlike the exported interface, an "oom" parameter is added. if oom==true,
2207 * the oom-killer can be invoked.
2209 static int __mem_cgroup_try_charge(struct mm_struct
*mm
,
2211 unsigned int nr_pages
,
2212 struct mem_cgroup
**ptr
,
2215 unsigned int batch
= max(CHARGE_BATCH
, nr_pages
);
2216 int nr_oom_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
2217 struct mem_cgroup
*memcg
= NULL
;
2221 * Unlike gloval-vm's OOM-kill, we're not in memory shortage
2222 * in system level. So, allow to go ahead dying process in addition to
2225 if (unlikely(test_thread_flag(TIF_MEMDIE
)
2226 || fatal_signal_pending(current
)))
2230 * We always charge the cgroup the mm_struct belongs to.
2231 * The mm_struct's mem_cgroup changes on task migration if the
2232 * thread group leader migrates. It's possible that mm is not
2233 * set, if so charge the init_mm (happens for pagecache usage).
2236 *ptr
= root_mem_cgroup
;
2238 if (*ptr
) { /* css should be a valid one */
2240 VM_BUG_ON(css_is_removed(&memcg
->css
));
2241 if (mem_cgroup_is_root(memcg
))
2243 if (nr_pages
== 1 && consume_stock(memcg
))
2245 css_get(&memcg
->css
);
2247 struct task_struct
*p
;
2250 p
= rcu_dereference(mm
->owner
);
2252 * Because we don't have task_lock(), "p" can exit.
2253 * In that case, "memcg" can point to root or p can be NULL with
2254 * race with swapoff. Then, we have small risk of mis-accouning.
2255 * But such kind of mis-account by race always happens because
2256 * we don't have cgroup_mutex(). It's overkill and we allo that
2258 * (*) swapoff at el will charge against mm-struct not against
2259 * task-struct. So, mm->owner can be NULL.
2261 memcg
= mem_cgroup_from_task(p
);
2263 memcg
= root_mem_cgroup
;
2264 if (mem_cgroup_is_root(memcg
)) {
2268 if (nr_pages
== 1 && consume_stock(memcg
)) {
2270 * It seems dagerous to access memcg without css_get().
2271 * But considering how consume_stok works, it's not
2272 * necessary. If consume_stock success, some charges
2273 * from this memcg are cached on this cpu. So, we
2274 * don't need to call css_get()/css_tryget() before
2275 * calling consume_stock().
2280 /* after here, we may be blocked. we need to get refcnt */
2281 if (!css_tryget(&memcg
->css
)) {
2291 /* If killed, bypass charge */
2292 if (fatal_signal_pending(current
)) {
2293 css_put(&memcg
->css
);
2298 if (oom
&& !nr_oom_retries
) {
2300 nr_oom_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
2303 ret
= mem_cgroup_do_charge(memcg
, gfp_mask
, batch
, oom_check
);
2307 case CHARGE_RETRY
: /* not in OOM situation but retry */
2309 css_put(&memcg
->css
);
2312 case CHARGE_WOULDBLOCK
: /* !__GFP_WAIT */
2313 css_put(&memcg
->css
);
2315 case CHARGE_NOMEM
: /* OOM routine works */
2317 css_put(&memcg
->css
);
2320 /* If oom, we never return -ENOMEM */
2323 case CHARGE_OOM_DIE
: /* Killed by OOM Killer */
2324 css_put(&memcg
->css
);
2327 } while (ret
!= CHARGE_OK
);
2329 if (batch
> nr_pages
)
2330 refill_stock(memcg
, batch
- nr_pages
);
2331 css_put(&memcg
->css
);
2339 *ptr
= root_mem_cgroup
;
2344 * Somemtimes we have to undo a charge we got by try_charge().
2345 * This function is for that and do uncharge, put css's refcnt.
2346 * gotten by try_charge().
2348 static void __mem_cgroup_cancel_charge(struct mem_cgroup
*memcg
,
2349 unsigned int nr_pages
)
2351 if (!mem_cgroup_is_root(memcg
)) {
2352 unsigned long bytes
= nr_pages
* PAGE_SIZE
;
2354 res_counter_uncharge(&memcg
->res
, bytes
);
2355 if (do_swap_account
)
2356 res_counter_uncharge(&memcg
->memsw
, bytes
);
2361 * A helper function to get mem_cgroup from ID. must be called under
2362 * rcu_read_lock(). The caller must check css_is_removed() or some if
2363 * it's concern. (dropping refcnt from swap can be called against removed
2366 static struct mem_cgroup
*mem_cgroup_lookup(unsigned short id
)
2368 struct cgroup_subsys_state
*css
;
2370 /* ID 0 is unused ID */
2373 css
= css_lookup(&mem_cgroup_subsys
, id
);
2376 return container_of(css
, struct mem_cgroup
, css
);
2379 struct mem_cgroup
*try_get_mem_cgroup_from_page(struct page
*page
)
2381 struct mem_cgroup
*memcg
= NULL
;
2382 struct page_cgroup
*pc
;
2386 VM_BUG_ON(!PageLocked(page
));
2388 pc
= lookup_page_cgroup(page
);
2389 lock_page_cgroup(pc
);
2390 if (PageCgroupUsed(pc
)) {
2391 memcg
= pc
->mem_cgroup
;
2392 if (memcg
&& !css_tryget(&memcg
->css
))
2394 } else if (PageSwapCache(page
)) {
2395 ent
.val
= page_private(page
);
2396 id
= lookup_swap_cgroup_id(ent
);
2398 memcg
= mem_cgroup_lookup(id
);
2399 if (memcg
&& !css_tryget(&memcg
->css
))
2403 unlock_page_cgroup(pc
);
2407 static void __mem_cgroup_commit_charge(struct mem_cgroup
*memcg
,
2409 unsigned int nr_pages
,
2410 struct page_cgroup
*pc
,
2411 enum charge_type ctype
,
2414 struct zone
*uninitialized_var(zone
);
2415 bool was_on_lru
= false;
2417 lock_page_cgroup(pc
);
2418 if (unlikely(PageCgroupUsed(pc
))) {
2419 unlock_page_cgroup(pc
);
2420 __mem_cgroup_cancel_charge(memcg
, nr_pages
);
2424 * we don't need page_cgroup_lock about tail pages, becase they are not
2425 * accessed by any other context at this point.
2429 * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page
2430 * may already be on some other mem_cgroup's LRU. Take care of it.
2433 zone
= page_zone(page
);
2434 spin_lock_irq(&zone
->lru_lock
);
2435 if (PageLRU(page
)) {
2437 del_page_from_lru_list(zone
, page
, page_lru(page
));
2442 pc
->mem_cgroup
= memcg
;
2444 * We access a page_cgroup asynchronously without lock_page_cgroup().
2445 * Especially when a page_cgroup is taken from a page, pc->mem_cgroup
2446 * is accessed after testing USED bit. To make pc->mem_cgroup visible
2447 * before USED bit, we need memory barrier here.
2448 * See mem_cgroup_add_lru_list(), etc.
2452 case MEM_CGROUP_CHARGE_TYPE_CACHE
:
2453 case MEM_CGROUP_CHARGE_TYPE_SHMEM
:
2454 SetPageCgroupCache(pc
);
2455 SetPageCgroupUsed(pc
);
2457 case MEM_CGROUP_CHARGE_TYPE_MAPPED
:
2458 ClearPageCgroupCache(pc
);
2459 SetPageCgroupUsed(pc
);
2467 VM_BUG_ON(PageLRU(page
));
2469 add_page_to_lru_list(zone
, page
, page_lru(page
));
2471 spin_unlock_irq(&zone
->lru_lock
);
2474 mem_cgroup_charge_statistics(memcg
, PageCgroupCache(pc
), nr_pages
);
2475 unlock_page_cgroup(pc
);
2478 * "charge_statistics" updated event counter. Then, check it.
2479 * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree.
2480 * if they exceeds softlimit.
2482 memcg_check_events(memcg
, page
);
2485 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
2487 #define PCGF_NOCOPY_AT_SPLIT ((1 << PCG_LOCK) | (1 << PCG_MOVE_LOCK) |\
2488 (1 << PCG_MIGRATION))
2490 * Because tail pages are not marked as "used", set it. We're under
2491 * zone->lru_lock, 'splitting on pmd' and compound_lock.
2492 * charge/uncharge will be never happen and move_account() is done under
2493 * compound_lock(), so we don't have to take care of races.
2495 void mem_cgroup_split_huge_fixup(struct page
*head
)
2497 struct page_cgroup
*head_pc
= lookup_page_cgroup(head
);
2498 struct page_cgroup
*pc
;
2501 if (mem_cgroup_disabled())
2503 for (i
= 1; i
< HPAGE_PMD_NR
; i
++) {
2505 pc
->mem_cgroup
= head_pc
->mem_cgroup
;
2506 smp_wmb();/* see __commit_charge() */
2507 pc
->flags
= head_pc
->flags
& ~PCGF_NOCOPY_AT_SPLIT
;
2510 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
2513 * mem_cgroup_move_account - move account of the page
2515 * @nr_pages: number of regular pages (>1 for huge pages)
2516 * @pc: page_cgroup of the page.
2517 * @from: mem_cgroup which the page is moved from.
2518 * @to: mem_cgroup which the page is moved to. @from != @to.
2519 * @uncharge: whether we should call uncharge and css_put against @from.
2521 * The caller must confirm following.
2522 * - page is not on LRU (isolate_page() is useful.)
2523 * - compound_lock is held when nr_pages > 1
2525 * This function doesn't do "charge" nor css_get to new cgroup. It should be
2526 * done by a caller(__mem_cgroup_try_charge would be useful). If @uncharge is
2527 * true, this function does "uncharge" from old cgroup, but it doesn't if
2528 * @uncharge is false, so a caller should do "uncharge".
2530 static int mem_cgroup_move_account(struct page
*page
,
2531 unsigned int nr_pages
,
2532 struct page_cgroup
*pc
,
2533 struct mem_cgroup
*from
,
2534 struct mem_cgroup
*to
,
2537 unsigned long flags
;
2540 VM_BUG_ON(from
== to
);
2541 VM_BUG_ON(PageLRU(page
));
2543 * The page is isolated from LRU. So, collapse function
2544 * will not handle this page. But page splitting can happen.
2545 * Do this check under compound_page_lock(). The caller should
2549 if (nr_pages
> 1 && !PageTransHuge(page
))
2552 lock_page_cgroup(pc
);
2555 if (!PageCgroupUsed(pc
) || pc
->mem_cgroup
!= from
)
2558 move_lock_page_cgroup(pc
, &flags
);
2560 if (PageCgroupFileMapped(pc
)) {
2561 /* Update mapped_file data for mem_cgroup */
2563 __this_cpu_dec(from
->stat
->count
[MEM_CGROUP_STAT_FILE_MAPPED
]);
2564 __this_cpu_inc(to
->stat
->count
[MEM_CGROUP_STAT_FILE_MAPPED
]);
2567 mem_cgroup_charge_statistics(from
, PageCgroupCache(pc
), -nr_pages
);
2569 /* This is not "cancel", but cancel_charge does all we need. */
2570 __mem_cgroup_cancel_charge(from
, nr_pages
);
2572 /* caller should have done css_get */
2573 pc
->mem_cgroup
= to
;
2574 mem_cgroup_charge_statistics(to
, PageCgroupCache(pc
), nr_pages
);
2576 * We charges against "to" which may not have any tasks. Then, "to"
2577 * can be under rmdir(). But in current implementation, caller of
2578 * this function is just force_empty() and move charge, so it's
2579 * guaranteed that "to" is never removed. So, we don't check rmdir
2582 move_unlock_page_cgroup(pc
, &flags
);
2585 unlock_page_cgroup(pc
);
2589 memcg_check_events(to
, page
);
2590 memcg_check_events(from
, page
);
2596 * move charges to its parent.
2599 static int mem_cgroup_move_parent(struct page
*page
,
2600 struct page_cgroup
*pc
,
2601 struct mem_cgroup
*child
,
2604 struct cgroup
*cg
= child
->css
.cgroup
;
2605 struct cgroup
*pcg
= cg
->parent
;
2606 struct mem_cgroup
*parent
;
2607 unsigned int nr_pages
;
2608 unsigned long uninitialized_var(flags
);
2616 if (!get_page_unless_zero(page
))
2618 if (isolate_lru_page(page
))
2621 nr_pages
= hpage_nr_pages(page
);
2623 parent
= mem_cgroup_from_cont(pcg
);
2624 ret
= __mem_cgroup_try_charge(NULL
, gfp_mask
, nr_pages
, &parent
, false);
2629 flags
= compound_lock_irqsave(page
);
2631 ret
= mem_cgroup_move_account(page
, nr_pages
, pc
, child
, parent
, true);
2633 __mem_cgroup_cancel_charge(parent
, nr_pages
);
2636 compound_unlock_irqrestore(page
, flags
);
2638 putback_lru_page(page
);
2646 * Charge the memory controller for page usage.
2648 * 0 if the charge was successful
2649 * < 0 if the cgroup is over its limit
2651 static int mem_cgroup_charge_common(struct page
*page
, struct mm_struct
*mm
,
2652 gfp_t gfp_mask
, enum charge_type ctype
)
2654 struct mem_cgroup
*memcg
= NULL
;
2655 unsigned int nr_pages
= 1;
2656 struct page_cgroup
*pc
;
2660 if (PageTransHuge(page
)) {
2661 nr_pages
<<= compound_order(page
);
2662 VM_BUG_ON(!PageTransHuge(page
));
2664 * Never OOM-kill a process for a huge page. The
2665 * fault handler will fall back to regular pages.
2670 pc
= lookup_page_cgroup(page
);
2671 ret
= __mem_cgroup_try_charge(mm
, gfp_mask
, nr_pages
, &memcg
, oom
);
2674 __mem_cgroup_commit_charge(memcg
, page
, nr_pages
, pc
, ctype
, false);
2678 int mem_cgroup_newpage_charge(struct page
*page
,
2679 struct mm_struct
*mm
, gfp_t gfp_mask
)
2681 if (mem_cgroup_disabled())
2683 VM_BUG_ON(page_mapped(page
));
2684 VM_BUG_ON(page
->mapping
&& !PageAnon(page
));
2686 return mem_cgroup_charge_common(page
, mm
, gfp_mask
,
2687 MEM_CGROUP_CHARGE_TYPE_MAPPED
);
2691 __mem_cgroup_commit_charge_swapin(struct page
*page
, struct mem_cgroup
*ptr
,
2692 enum charge_type ctype
);
2694 int mem_cgroup_cache_charge(struct page
*page
, struct mm_struct
*mm
,
2697 struct mem_cgroup
*memcg
= NULL
;
2698 enum charge_type type
= MEM_CGROUP_CHARGE_TYPE_CACHE
;
2701 if (mem_cgroup_disabled())
2703 if (PageCompound(page
))
2708 if (!page_is_file_cache(page
))
2709 type
= MEM_CGROUP_CHARGE_TYPE_SHMEM
;
2711 if (!PageSwapCache(page
))
2712 ret
= mem_cgroup_charge_common(page
, mm
, gfp_mask
, type
);
2713 else { /* page is swapcache/shmem */
2714 ret
= mem_cgroup_try_charge_swapin(mm
, page
, gfp_mask
, &memcg
);
2716 __mem_cgroup_commit_charge_swapin(page
, memcg
, type
);
2722 * While swap-in, try_charge -> commit or cancel, the page is locked.
2723 * And when try_charge() successfully returns, one refcnt to memcg without
2724 * struct page_cgroup is acquired. This refcnt will be consumed by
2725 * "commit()" or removed by "cancel()"
2727 int mem_cgroup_try_charge_swapin(struct mm_struct
*mm
,
2729 gfp_t mask
, struct mem_cgroup
**memcgp
)
2731 struct mem_cgroup
*memcg
;
2736 if (mem_cgroup_disabled())
2739 if (!do_swap_account
)
2742 * A racing thread's fault, or swapoff, may have already updated
2743 * the pte, and even removed page from swap cache: in those cases
2744 * do_swap_page()'s pte_same() test will fail; but there's also a
2745 * KSM case which does need to charge the page.
2747 if (!PageSwapCache(page
))
2749 memcg
= try_get_mem_cgroup_from_page(page
);
2753 ret
= __mem_cgroup_try_charge(NULL
, mask
, 1, memcgp
, true);
2754 css_put(&memcg
->css
);
2761 ret
= __mem_cgroup_try_charge(mm
, mask
, 1, memcgp
, true);
2768 __mem_cgroup_commit_charge_swapin(struct page
*page
, struct mem_cgroup
*memcg
,
2769 enum charge_type ctype
)
2771 struct page_cgroup
*pc
;
2773 if (mem_cgroup_disabled())
2777 cgroup_exclude_rmdir(&memcg
->css
);
2779 pc
= lookup_page_cgroup(page
);
2780 __mem_cgroup_commit_charge(memcg
, page
, 1, pc
, ctype
, true);
2782 * Now swap is on-memory. This means this page may be
2783 * counted both as mem and swap....double count.
2784 * Fix it by uncharging from memsw. Basically, this SwapCache is stable
2785 * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page()
2786 * may call delete_from_swap_cache() before reach here.
2788 if (do_swap_account
&& PageSwapCache(page
)) {
2789 swp_entry_t ent
= {.val
= page_private(page
)};
2790 struct mem_cgroup
*swap_memcg
;
2793 id
= swap_cgroup_record(ent
, 0);
2795 swap_memcg
= mem_cgroup_lookup(id
);
2798 * This recorded memcg can be obsolete one. So, avoid
2799 * calling css_tryget
2801 if (!mem_cgroup_is_root(swap_memcg
))
2802 res_counter_uncharge(&swap_memcg
->memsw
,
2804 mem_cgroup_swap_statistics(swap_memcg
, false);
2805 mem_cgroup_put(swap_memcg
);
2810 * At swapin, we may charge account against cgroup which has no tasks.
2811 * So, rmdir()->pre_destroy() can be called while we do this charge.
2812 * In that case, we need to call pre_destroy() again. check it here.
2814 cgroup_release_and_wakeup_rmdir(&memcg
->css
);
2817 void mem_cgroup_commit_charge_swapin(struct page
*page
,
2818 struct mem_cgroup
*memcg
)
2820 __mem_cgroup_commit_charge_swapin(page
, memcg
,
2821 MEM_CGROUP_CHARGE_TYPE_MAPPED
);
2824 void mem_cgroup_cancel_charge_swapin(struct mem_cgroup
*memcg
)
2826 if (mem_cgroup_disabled())
2830 __mem_cgroup_cancel_charge(memcg
, 1);
2833 static void mem_cgroup_do_uncharge(struct mem_cgroup
*memcg
,
2834 unsigned int nr_pages
,
2835 const enum charge_type ctype
)
2837 struct memcg_batch_info
*batch
= NULL
;
2838 bool uncharge_memsw
= true;
2840 /* If swapout, usage of swap doesn't decrease */
2841 if (!do_swap_account
|| ctype
== MEM_CGROUP_CHARGE_TYPE_SWAPOUT
)
2842 uncharge_memsw
= false;
2844 batch
= ¤t
->memcg_batch
;
2846 * In usual, we do css_get() when we remember memcg pointer.
2847 * But in this case, we keep res->usage until end of a series of
2848 * uncharges. Then, it's ok to ignore memcg's refcnt.
2851 batch
->memcg
= memcg
;
2853 * do_batch > 0 when unmapping pages or inode invalidate/truncate.
2854 * In those cases, all pages freed continuously can be expected to be in
2855 * the same cgroup and we have chance to coalesce uncharges.
2856 * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE)
2857 * because we want to do uncharge as soon as possible.
2860 if (!batch
->do_batch
|| test_thread_flag(TIF_MEMDIE
))
2861 goto direct_uncharge
;
2864 goto direct_uncharge
;
2867 * In typical case, batch->memcg == mem. This means we can
2868 * merge a series of uncharges to an uncharge of res_counter.
2869 * If not, we uncharge res_counter ony by one.
2871 if (batch
->memcg
!= memcg
)
2872 goto direct_uncharge
;
2873 /* remember freed charge and uncharge it later */
2876 batch
->memsw_nr_pages
++;
2879 res_counter_uncharge(&memcg
->res
, nr_pages
* PAGE_SIZE
);
2881 res_counter_uncharge(&memcg
->memsw
, nr_pages
* PAGE_SIZE
);
2882 if (unlikely(batch
->memcg
!= memcg
))
2883 memcg_oom_recover(memcg
);
2888 * uncharge if !page_mapped(page)
2890 static struct mem_cgroup
*
2891 __mem_cgroup_uncharge_common(struct page
*page
, enum charge_type ctype
)
2893 struct mem_cgroup
*memcg
= NULL
;
2894 unsigned int nr_pages
= 1;
2895 struct page_cgroup
*pc
;
2897 if (mem_cgroup_disabled())
2900 if (PageSwapCache(page
))
2903 if (PageTransHuge(page
)) {
2904 nr_pages
<<= compound_order(page
);
2905 VM_BUG_ON(!PageTransHuge(page
));
2908 * Check if our page_cgroup is valid
2910 pc
= lookup_page_cgroup(page
);
2911 if (unlikely(!PageCgroupUsed(pc
)))
2914 lock_page_cgroup(pc
);
2916 memcg
= pc
->mem_cgroup
;
2918 if (!PageCgroupUsed(pc
))
2922 case MEM_CGROUP_CHARGE_TYPE_MAPPED
:
2923 case MEM_CGROUP_CHARGE_TYPE_DROP
:
2924 /* See mem_cgroup_prepare_migration() */
2925 if (page_mapped(page
) || PageCgroupMigration(pc
))
2928 case MEM_CGROUP_CHARGE_TYPE_SWAPOUT
:
2929 if (!PageAnon(page
)) { /* Shared memory */
2930 if (page
->mapping
&& !page_is_file_cache(page
))
2932 } else if (page_mapped(page
)) /* Anon */
2939 mem_cgroup_charge_statistics(memcg
, PageCgroupCache(pc
), -nr_pages
);
2941 ClearPageCgroupUsed(pc
);
2943 * pc->mem_cgroup is not cleared here. It will be accessed when it's
2944 * freed from LRU. This is safe because uncharged page is expected not
2945 * to be reused (freed soon). Exception is SwapCache, it's handled by
2946 * special functions.
2949 unlock_page_cgroup(pc
);
2951 * even after unlock, we have memcg->res.usage here and this memcg
2952 * will never be freed.
2954 memcg_check_events(memcg
, page
);
2955 if (do_swap_account
&& ctype
== MEM_CGROUP_CHARGE_TYPE_SWAPOUT
) {
2956 mem_cgroup_swap_statistics(memcg
, true);
2957 mem_cgroup_get(memcg
);
2959 if (!mem_cgroup_is_root(memcg
))
2960 mem_cgroup_do_uncharge(memcg
, nr_pages
, ctype
);
2965 unlock_page_cgroup(pc
);
2969 void mem_cgroup_uncharge_page(struct page
*page
)
2972 if (page_mapped(page
))
2974 VM_BUG_ON(page
->mapping
&& !PageAnon(page
));
2975 __mem_cgroup_uncharge_common(page
, MEM_CGROUP_CHARGE_TYPE_MAPPED
);
2978 void mem_cgroup_uncharge_cache_page(struct page
*page
)
2980 VM_BUG_ON(page_mapped(page
));
2981 VM_BUG_ON(page
->mapping
);
2982 __mem_cgroup_uncharge_common(page
, MEM_CGROUP_CHARGE_TYPE_CACHE
);
2986 * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate.
2987 * In that cases, pages are freed continuously and we can expect pages
2988 * are in the same memcg. All these calls itself limits the number of
2989 * pages freed at once, then uncharge_start/end() is called properly.
2990 * This may be called prural(2) times in a context,
2993 void mem_cgroup_uncharge_start(void)
2995 current
->memcg_batch
.do_batch
++;
2996 /* We can do nest. */
2997 if (current
->memcg_batch
.do_batch
== 1) {
2998 current
->memcg_batch
.memcg
= NULL
;
2999 current
->memcg_batch
.nr_pages
= 0;
3000 current
->memcg_batch
.memsw_nr_pages
= 0;
3004 void mem_cgroup_uncharge_end(void)
3006 struct memcg_batch_info
*batch
= ¤t
->memcg_batch
;
3008 if (!batch
->do_batch
)
3012 if (batch
->do_batch
) /* If stacked, do nothing. */
3018 * This "batch->memcg" is valid without any css_get/put etc...
3019 * bacause we hide charges behind us.
3021 if (batch
->nr_pages
)
3022 res_counter_uncharge(&batch
->memcg
->res
,
3023 batch
->nr_pages
* PAGE_SIZE
);
3024 if (batch
->memsw_nr_pages
)
3025 res_counter_uncharge(&batch
->memcg
->memsw
,
3026 batch
->memsw_nr_pages
* PAGE_SIZE
);
3027 memcg_oom_recover(batch
->memcg
);
3028 /* forget this pointer (for sanity check) */
3029 batch
->memcg
= NULL
;
3033 * A function for resetting pc->mem_cgroup for newly allocated pages.
3034 * This function should be called if the newpage will be added to LRU
3035 * before start accounting.
3037 void mem_cgroup_reset_owner(struct page
*newpage
)
3039 struct page_cgroup
*pc
;
3041 if (mem_cgroup_disabled())
3044 pc
= lookup_page_cgroup(newpage
);
3045 VM_BUG_ON(PageCgroupUsed(pc
));
3046 pc
->mem_cgroup
= root_mem_cgroup
;
3051 * called after __delete_from_swap_cache() and drop "page" account.
3052 * memcg information is recorded to swap_cgroup of "ent"
3055 mem_cgroup_uncharge_swapcache(struct page
*page
, swp_entry_t ent
, bool swapout
)
3057 struct mem_cgroup
*memcg
;
3058 int ctype
= MEM_CGROUP_CHARGE_TYPE_SWAPOUT
;
3060 if (!swapout
) /* this was a swap cache but the swap is unused ! */
3061 ctype
= MEM_CGROUP_CHARGE_TYPE_DROP
;
3063 memcg
= __mem_cgroup_uncharge_common(page
, ctype
);
3066 * record memcg information, if swapout && memcg != NULL,
3067 * mem_cgroup_get() was called in uncharge().
3069 if (do_swap_account
&& swapout
&& memcg
)
3070 swap_cgroup_record(ent
, css_id(&memcg
->css
));
3074 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
3076 * called from swap_entry_free(). remove record in swap_cgroup and
3077 * uncharge "memsw" account.
3079 void mem_cgroup_uncharge_swap(swp_entry_t ent
)
3081 struct mem_cgroup
*memcg
;
3084 if (!do_swap_account
)
3087 id
= swap_cgroup_record(ent
, 0);
3089 memcg
= mem_cgroup_lookup(id
);
3092 * We uncharge this because swap is freed.
3093 * This memcg can be obsolete one. We avoid calling css_tryget
3095 if (!mem_cgroup_is_root(memcg
))
3096 res_counter_uncharge(&memcg
->memsw
, PAGE_SIZE
);
3097 mem_cgroup_swap_statistics(memcg
, false);
3098 mem_cgroup_put(memcg
);
3104 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
3105 * @entry: swap entry to be moved
3106 * @from: mem_cgroup which the entry is moved from
3107 * @to: mem_cgroup which the entry is moved to
3108 * @need_fixup: whether we should fixup res_counters and refcounts.
3110 * It succeeds only when the swap_cgroup's record for this entry is the same
3111 * as the mem_cgroup's id of @from.
3113 * Returns 0 on success, -EINVAL on failure.
3115 * The caller must have charged to @to, IOW, called res_counter_charge() about
3116 * both res and memsw, and called css_get().
3118 static int mem_cgroup_move_swap_account(swp_entry_t entry
,
3119 struct mem_cgroup
*from
, struct mem_cgroup
*to
, bool need_fixup
)
3121 unsigned short old_id
, new_id
;
3123 old_id
= css_id(&from
->css
);
3124 new_id
= css_id(&to
->css
);
3126 if (swap_cgroup_cmpxchg(entry
, old_id
, new_id
) == old_id
) {
3127 mem_cgroup_swap_statistics(from
, false);
3128 mem_cgroup_swap_statistics(to
, true);
3130 * This function is only called from task migration context now.
3131 * It postpones res_counter and refcount handling till the end
3132 * of task migration(mem_cgroup_clear_mc()) for performance
3133 * improvement. But we cannot postpone mem_cgroup_get(to)
3134 * because if the process that has been moved to @to does
3135 * swap-in, the refcount of @to might be decreased to 0.
3139 if (!mem_cgroup_is_root(from
))
3140 res_counter_uncharge(&from
->memsw
, PAGE_SIZE
);
3141 mem_cgroup_put(from
);
3143 * we charged both to->res and to->memsw, so we should
3146 if (!mem_cgroup_is_root(to
))
3147 res_counter_uncharge(&to
->res
, PAGE_SIZE
);
3154 static inline int mem_cgroup_move_swap_account(swp_entry_t entry
,
3155 struct mem_cgroup
*from
, struct mem_cgroup
*to
, bool need_fixup
)
3162 * Before starting migration, account PAGE_SIZE to mem_cgroup that the old
3165 int mem_cgroup_prepare_migration(struct page
*page
,
3166 struct page
*newpage
, struct mem_cgroup
**memcgp
, gfp_t gfp_mask
)
3168 struct mem_cgroup
*memcg
= NULL
;
3169 struct page_cgroup
*pc
;
3170 enum charge_type ctype
;
3175 VM_BUG_ON(PageTransHuge(page
));
3176 if (mem_cgroup_disabled())
3179 pc
= lookup_page_cgroup(page
);
3180 lock_page_cgroup(pc
);
3181 if (PageCgroupUsed(pc
)) {
3182 memcg
= pc
->mem_cgroup
;
3183 css_get(&memcg
->css
);
3185 * At migrating an anonymous page, its mapcount goes down
3186 * to 0 and uncharge() will be called. But, even if it's fully
3187 * unmapped, migration may fail and this page has to be
3188 * charged again. We set MIGRATION flag here and delay uncharge
3189 * until end_migration() is called
3191 * Corner Case Thinking
3193 * When the old page was mapped as Anon and it's unmap-and-freed
3194 * while migration was ongoing.
3195 * If unmap finds the old page, uncharge() of it will be delayed
3196 * until end_migration(). If unmap finds a new page, it's
3197 * uncharged when it make mapcount to be 1->0. If unmap code
3198 * finds swap_migration_entry, the new page will not be mapped
3199 * and end_migration() will find it(mapcount==0).
3202 * When the old page was mapped but migraion fails, the kernel
3203 * remaps it. A charge for it is kept by MIGRATION flag even
3204 * if mapcount goes down to 0. We can do remap successfully
3205 * without charging it again.
3208 * The "old" page is under lock_page() until the end of
3209 * migration, so, the old page itself will not be swapped-out.
3210 * If the new page is swapped out before end_migraton, our
3211 * hook to usual swap-out path will catch the event.
3214 SetPageCgroupMigration(pc
);
3216 unlock_page_cgroup(pc
);
3218 * If the page is not charged at this point,
3225 ret
= __mem_cgroup_try_charge(NULL
, gfp_mask
, 1, memcgp
, false);
3226 css_put(&memcg
->css
);/* drop extra refcnt */
3228 if (PageAnon(page
)) {
3229 lock_page_cgroup(pc
);
3230 ClearPageCgroupMigration(pc
);
3231 unlock_page_cgroup(pc
);
3233 * The old page may be fully unmapped while we kept it.
3235 mem_cgroup_uncharge_page(page
);
3237 /* we'll need to revisit this error code (we have -EINTR) */
3241 * We charge new page before it's used/mapped. So, even if unlock_page()
3242 * is called before end_migration, we can catch all events on this new
3243 * page. In the case new page is migrated but not remapped, new page's
3244 * mapcount will be finally 0 and we call uncharge in end_migration().
3246 pc
= lookup_page_cgroup(newpage
);
3248 ctype
= MEM_CGROUP_CHARGE_TYPE_MAPPED
;
3249 else if (page_is_file_cache(page
))
3250 ctype
= MEM_CGROUP_CHARGE_TYPE_CACHE
;
3252 ctype
= MEM_CGROUP_CHARGE_TYPE_SHMEM
;
3253 __mem_cgroup_commit_charge(memcg
, newpage
, 1, pc
, ctype
, false);
3257 /* remove redundant charge if migration failed*/
3258 void mem_cgroup_end_migration(struct mem_cgroup
*memcg
,
3259 struct page
*oldpage
, struct page
*newpage
, bool migration_ok
)
3261 struct page
*used
, *unused
;
3262 struct page_cgroup
*pc
;
3266 /* blocks rmdir() */
3267 cgroup_exclude_rmdir(&memcg
->css
);
3268 if (!migration_ok
) {
3276 * We disallowed uncharge of pages under migration because mapcount
3277 * of the page goes down to zero, temporarly.
3278 * Clear the flag and check the page should be charged.
3280 pc
= lookup_page_cgroup(oldpage
);
3281 lock_page_cgroup(pc
);
3282 ClearPageCgroupMigration(pc
);
3283 unlock_page_cgroup(pc
);
3285 __mem_cgroup_uncharge_common(unused
, MEM_CGROUP_CHARGE_TYPE_FORCE
);
3288 * If a page is a file cache, radix-tree replacement is very atomic
3289 * and we can skip this check. When it was an Anon page, its mapcount
3290 * goes down to 0. But because we added MIGRATION flage, it's not
3291 * uncharged yet. There are several case but page->mapcount check
3292 * and USED bit check in mem_cgroup_uncharge_page() will do enough
3293 * check. (see prepare_charge() also)
3296 mem_cgroup_uncharge_page(used
);
3298 * At migration, we may charge account against cgroup which has no
3300 * So, rmdir()->pre_destroy() can be called while we do this charge.
3301 * In that case, we need to call pre_destroy() again. check it here.
3303 cgroup_release_and_wakeup_rmdir(&memcg
->css
);
3307 * At replace page cache, newpage is not under any memcg but it's on
3308 * LRU. So, this function doesn't touch res_counter but handles LRU
3309 * in correct way. Both pages are locked so we cannot race with uncharge.
3311 void mem_cgroup_replace_page_cache(struct page
*oldpage
,
3312 struct page
*newpage
)
3314 struct mem_cgroup
*memcg
;
3315 struct page_cgroup
*pc
;
3316 enum charge_type type
= MEM_CGROUP_CHARGE_TYPE_CACHE
;
3318 if (mem_cgroup_disabled())
3321 pc
= lookup_page_cgroup(oldpage
);
3322 /* fix accounting on old pages */
3323 lock_page_cgroup(pc
);
3324 memcg
= pc
->mem_cgroup
;
3325 mem_cgroup_charge_statistics(memcg
, PageCgroupCache(pc
), -1);
3326 ClearPageCgroupUsed(pc
);
3327 unlock_page_cgroup(pc
);
3329 if (PageSwapBacked(oldpage
))
3330 type
= MEM_CGROUP_CHARGE_TYPE_SHMEM
;
3333 * Even if newpage->mapping was NULL before starting replacement,
3334 * the newpage may be on LRU(or pagevec for LRU) already. We lock
3335 * LRU while we overwrite pc->mem_cgroup.
3337 __mem_cgroup_commit_charge(memcg
, newpage
, 1, pc
, type
, true);
3340 #ifdef CONFIG_DEBUG_VM
3341 static struct page_cgroup
*lookup_page_cgroup_used(struct page
*page
)
3343 struct page_cgroup
*pc
;
3345 pc
= lookup_page_cgroup(page
);
3347 * Can be NULL while feeding pages into the page allocator for
3348 * the first time, i.e. during boot or memory hotplug;
3349 * or when mem_cgroup_disabled().
3351 if (likely(pc
) && PageCgroupUsed(pc
))
3356 bool mem_cgroup_bad_page_check(struct page
*page
)
3358 if (mem_cgroup_disabled())
3361 return lookup_page_cgroup_used(page
) != NULL
;
3364 void mem_cgroup_print_bad_page(struct page
*page
)
3366 struct page_cgroup
*pc
;
3368 pc
= lookup_page_cgroup_used(page
);
3370 printk(KERN_ALERT
"pc:%p pc->flags:%lx pc->mem_cgroup:%p\n",
3371 pc
, pc
->flags
, pc
->mem_cgroup
);
3376 static DEFINE_MUTEX(set_limit_mutex
);
3378 static int mem_cgroup_resize_limit(struct mem_cgroup
*memcg
,
3379 unsigned long long val
)
3382 u64 memswlimit
, memlimit
;
3384 int children
= mem_cgroup_count_children(memcg
);
3385 u64 curusage
, oldusage
;
3389 * For keeping hierarchical_reclaim simple, how long we should retry
3390 * is depends on callers. We set our retry-count to be function
3391 * of # of children which we should visit in this loop.
3393 retry_count
= MEM_CGROUP_RECLAIM_RETRIES
* children
;
3395 oldusage
= res_counter_read_u64(&memcg
->res
, RES_USAGE
);
3398 while (retry_count
) {
3399 if (signal_pending(current
)) {
3404 * Rather than hide all in some function, I do this in
3405 * open coded manner. You see what this really does.
3406 * We have to guarantee memcg->res.limit < memcg->memsw.limit.
3408 mutex_lock(&set_limit_mutex
);
3409 memswlimit
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
3410 if (memswlimit
< val
) {
3412 mutex_unlock(&set_limit_mutex
);
3416 memlimit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
3420 ret
= res_counter_set_limit(&memcg
->res
, val
);
3422 if (memswlimit
== val
)
3423 memcg
->memsw_is_minimum
= true;
3425 memcg
->memsw_is_minimum
= false;
3427 mutex_unlock(&set_limit_mutex
);
3432 mem_cgroup_reclaim(memcg
, GFP_KERNEL
,
3433 MEM_CGROUP_RECLAIM_SHRINK
);
3434 curusage
= res_counter_read_u64(&memcg
->res
, RES_USAGE
);
3435 /* Usage is reduced ? */
3436 if (curusage
>= oldusage
)
3439 oldusage
= curusage
;
3441 if (!ret
&& enlarge
)
3442 memcg_oom_recover(memcg
);
3447 static int mem_cgroup_resize_memsw_limit(struct mem_cgroup
*memcg
,
3448 unsigned long long val
)
3451 u64 memlimit
, memswlimit
, oldusage
, curusage
;
3452 int children
= mem_cgroup_count_children(memcg
);
3456 /* see mem_cgroup_resize_res_limit */
3457 retry_count
= children
* MEM_CGROUP_RECLAIM_RETRIES
;
3458 oldusage
= res_counter_read_u64(&memcg
->memsw
, RES_USAGE
);
3459 while (retry_count
) {
3460 if (signal_pending(current
)) {
3465 * Rather than hide all in some function, I do this in
3466 * open coded manner. You see what this really does.
3467 * We have to guarantee memcg->res.limit < memcg->memsw.limit.
3469 mutex_lock(&set_limit_mutex
);
3470 memlimit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
3471 if (memlimit
> val
) {
3473 mutex_unlock(&set_limit_mutex
);
3476 memswlimit
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
3477 if (memswlimit
< val
)
3479 ret
= res_counter_set_limit(&memcg
->memsw
, val
);
3481 if (memlimit
== val
)
3482 memcg
->memsw_is_minimum
= true;
3484 memcg
->memsw_is_minimum
= false;
3486 mutex_unlock(&set_limit_mutex
);
3491 mem_cgroup_reclaim(memcg
, GFP_KERNEL
,
3492 MEM_CGROUP_RECLAIM_NOSWAP
|
3493 MEM_CGROUP_RECLAIM_SHRINK
);
3494 curusage
= res_counter_read_u64(&memcg
->memsw
, RES_USAGE
);
3495 /* Usage is reduced ? */
3496 if (curusage
>= oldusage
)
3499 oldusage
= curusage
;
3501 if (!ret
&& enlarge
)
3502 memcg_oom_recover(memcg
);
3506 unsigned long mem_cgroup_soft_limit_reclaim(struct zone
*zone
, int order
,
3508 unsigned long *total_scanned
)
3510 unsigned long nr_reclaimed
= 0;
3511 struct mem_cgroup_per_zone
*mz
, *next_mz
= NULL
;
3512 unsigned long reclaimed
;
3514 struct mem_cgroup_tree_per_zone
*mctz
;
3515 unsigned long long excess
;
3516 unsigned long nr_scanned
;
3521 mctz
= soft_limit_tree_node_zone(zone_to_nid(zone
), zone_idx(zone
));
3523 * This loop can run a while, specially if mem_cgroup's continuously
3524 * keep exceeding their soft limit and putting the system under
3531 mz
= mem_cgroup_largest_soft_limit_node(mctz
);
3536 reclaimed
= mem_cgroup_soft_reclaim(mz
->mem
, zone
,
3537 gfp_mask
, &nr_scanned
);
3538 nr_reclaimed
+= reclaimed
;
3539 *total_scanned
+= nr_scanned
;
3540 spin_lock(&mctz
->lock
);
3543 * If we failed to reclaim anything from this memory cgroup
3544 * it is time to move on to the next cgroup
3550 * Loop until we find yet another one.
3552 * By the time we get the soft_limit lock
3553 * again, someone might have aded the
3554 * group back on the RB tree. Iterate to
3555 * make sure we get a different mem.
3556 * mem_cgroup_largest_soft_limit_node returns
3557 * NULL if no other cgroup is present on
3561 __mem_cgroup_largest_soft_limit_node(mctz
);
3563 css_put(&next_mz
->mem
->css
);
3564 else /* next_mz == NULL or other memcg */
3568 __mem_cgroup_remove_exceeded(mz
->mem
, mz
, mctz
);
3569 excess
= res_counter_soft_limit_excess(&mz
->mem
->res
);
3571 * One school of thought says that we should not add
3572 * back the node to the tree if reclaim returns 0.
3573 * But our reclaim could return 0, simply because due
3574 * to priority we are exposing a smaller subset of
3575 * memory to reclaim from. Consider this as a longer
3578 /* If excess == 0, no tree ops */
3579 __mem_cgroup_insert_exceeded(mz
->mem
, mz
, mctz
, excess
);
3580 spin_unlock(&mctz
->lock
);
3581 css_put(&mz
->mem
->css
);
3584 * Could not reclaim anything and there are no more
3585 * mem cgroups to try or we seem to be looping without
3586 * reclaiming anything.
3588 if (!nr_reclaimed
&&
3590 loop
> MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS
))
3592 } while (!nr_reclaimed
);
3594 css_put(&next_mz
->mem
->css
);
3595 return nr_reclaimed
;
3599 * This routine traverse page_cgroup in given list and drop them all.
3600 * *And* this routine doesn't reclaim page itself, just removes page_cgroup.
3602 static int mem_cgroup_force_empty_list(struct mem_cgroup
*memcg
,
3603 int node
, int zid
, enum lru_list lru
)
3605 struct mem_cgroup_per_zone
*mz
;
3606 unsigned long flags
, loop
;
3607 struct list_head
*list
;
3612 zone
= &NODE_DATA(node
)->node_zones
[zid
];
3613 mz
= mem_cgroup_zoneinfo(memcg
, node
, zid
);
3614 list
= &mz
->lruvec
.lists
[lru
];
3616 loop
= MEM_CGROUP_ZSTAT(mz
, lru
);
3617 /* give some margin against EBUSY etc...*/
3621 struct page_cgroup
*pc
;
3625 spin_lock_irqsave(&zone
->lru_lock
, flags
);
3626 if (list_empty(list
)) {
3627 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
3630 page
= list_entry(list
->prev
, struct page
, lru
);
3632 list_move(&page
->lru
, list
);
3634 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
3637 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
3639 pc
= lookup_page_cgroup(page
);
3641 ret
= mem_cgroup_move_parent(page
, pc
, memcg
, GFP_KERNEL
);
3642 if (ret
== -ENOMEM
|| ret
== -EINTR
)
3645 if (ret
== -EBUSY
|| ret
== -EINVAL
) {
3646 /* found lock contention or "pc" is obsolete. */
3653 if (!ret
&& !list_empty(list
))
3659 * make mem_cgroup's charge to be 0 if there is no task.
3660 * This enables deleting this mem_cgroup.
3662 static int mem_cgroup_force_empty(struct mem_cgroup
*memcg
, bool free_all
)
3665 int node
, zid
, shrink
;
3666 int nr_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
3667 struct cgroup
*cgrp
= memcg
->css
.cgroup
;
3669 css_get(&memcg
->css
);
3672 /* should free all ? */
3678 if (cgroup_task_count(cgrp
) || !list_empty(&cgrp
->children
))
3681 if (signal_pending(current
))
3683 /* This is for making all *used* pages to be on LRU. */
3684 lru_add_drain_all();
3685 drain_all_stock_sync(memcg
);
3687 mem_cgroup_start_move(memcg
);
3688 for_each_node_state(node
, N_HIGH_MEMORY
) {
3689 for (zid
= 0; !ret
&& zid
< MAX_NR_ZONES
; zid
++) {
3692 ret
= mem_cgroup_force_empty_list(memcg
,
3701 mem_cgroup_end_move(memcg
);
3702 memcg_oom_recover(memcg
);
3703 /* it seems parent cgroup doesn't have enough mem */
3707 /* "ret" should also be checked to ensure all lists are empty. */
3708 } while (memcg
->res
.usage
> 0 || ret
);
3710 css_put(&memcg
->css
);
3714 /* returns EBUSY if there is a task or if we come here twice. */
3715 if (cgroup_task_count(cgrp
) || !list_empty(&cgrp
->children
) || shrink
) {
3719 /* we call try-to-free pages for make this cgroup empty */
3720 lru_add_drain_all();
3721 /* try to free all pages in this cgroup */
3723 while (nr_retries
&& memcg
->res
.usage
> 0) {
3726 if (signal_pending(current
)) {
3730 progress
= try_to_free_mem_cgroup_pages(memcg
, GFP_KERNEL
,
3734 /* maybe some writeback is necessary */
3735 congestion_wait(BLK_RW_ASYNC
, HZ
/10);
3740 /* try move_account...there may be some *locked* pages. */
3744 int mem_cgroup_force_empty_write(struct cgroup
*cont
, unsigned int event
)
3746 return mem_cgroup_force_empty(mem_cgroup_from_cont(cont
), true);
3750 static u64
mem_cgroup_hierarchy_read(struct cgroup
*cont
, struct cftype
*cft
)
3752 return mem_cgroup_from_cont(cont
)->use_hierarchy
;
3755 static int mem_cgroup_hierarchy_write(struct cgroup
*cont
, struct cftype
*cft
,
3759 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
3760 struct cgroup
*parent
= cont
->parent
;
3761 struct mem_cgroup
*parent_memcg
= NULL
;
3764 parent_memcg
= mem_cgroup_from_cont(parent
);
3768 * If parent's use_hierarchy is set, we can't make any modifications
3769 * in the child subtrees. If it is unset, then the change can
3770 * occur, provided the current cgroup has no children.
3772 * For the root cgroup, parent_mem is NULL, we allow value to be
3773 * set if there are no children.
3775 if ((!parent_memcg
|| !parent_memcg
->use_hierarchy
) &&
3776 (val
== 1 || val
== 0)) {
3777 if (list_empty(&cont
->children
))
3778 memcg
->use_hierarchy
= val
;
3789 static unsigned long mem_cgroup_recursive_stat(struct mem_cgroup
*memcg
,
3790 enum mem_cgroup_stat_index idx
)
3792 struct mem_cgroup
*iter
;
3795 /* Per-cpu values can be negative, use a signed accumulator */
3796 for_each_mem_cgroup_tree(iter
, memcg
)
3797 val
+= mem_cgroup_read_stat(iter
, idx
);
3799 if (val
< 0) /* race ? */
3804 static inline u64
mem_cgroup_usage(struct mem_cgroup
*memcg
, bool swap
)
3808 if (!mem_cgroup_is_root(memcg
)) {
3810 return res_counter_read_u64(&memcg
->res
, RES_USAGE
);
3812 return res_counter_read_u64(&memcg
->memsw
, RES_USAGE
);
3815 val
= mem_cgroup_recursive_stat(memcg
, MEM_CGROUP_STAT_CACHE
);
3816 val
+= mem_cgroup_recursive_stat(memcg
, MEM_CGROUP_STAT_RSS
);
3819 val
+= mem_cgroup_recursive_stat(memcg
, MEM_CGROUP_STAT_SWAPOUT
);
3821 return val
<< PAGE_SHIFT
;
3824 static u64
mem_cgroup_read(struct cgroup
*cont
, struct cftype
*cft
)
3826 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
3830 type
= MEMFILE_TYPE(cft
->private);
3831 name
= MEMFILE_ATTR(cft
->private);
3834 if (name
== RES_USAGE
)
3835 val
= mem_cgroup_usage(memcg
, false);
3837 val
= res_counter_read_u64(&memcg
->res
, name
);
3840 if (name
== RES_USAGE
)
3841 val
= mem_cgroup_usage(memcg
, true);
3843 val
= res_counter_read_u64(&memcg
->memsw
, name
);
3852 * The user of this function is...
3855 static int mem_cgroup_write(struct cgroup
*cont
, struct cftype
*cft
,
3858 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
3860 unsigned long long val
;
3863 type
= MEMFILE_TYPE(cft
->private);
3864 name
= MEMFILE_ATTR(cft
->private);
3867 if (mem_cgroup_is_root(memcg
)) { /* Can't set limit on root */
3871 /* This function does all necessary parse...reuse it */
3872 ret
= res_counter_memparse_write_strategy(buffer
, &val
);
3876 ret
= mem_cgroup_resize_limit(memcg
, val
);
3878 ret
= mem_cgroup_resize_memsw_limit(memcg
, val
);
3880 case RES_SOFT_LIMIT
:
3881 ret
= res_counter_memparse_write_strategy(buffer
, &val
);
3885 * For memsw, soft limits are hard to implement in terms
3886 * of semantics, for now, we support soft limits for
3887 * control without swap
3890 ret
= res_counter_set_soft_limit(&memcg
->res
, val
);
3895 ret
= -EINVAL
; /* should be BUG() ? */
3901 static void memcg_get_hierarchical_limit(struct mem_cgroup
*memcg
,
3902 unsigned long long *mem_limit
, unsigned long long *memsw_limit
)
3904 struct cgroup
*cgroup
;
3905 unsigned long long min_limit
, min_memsw_limit
, tmp
;
3907 min_limit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
3908 min_memsw_limit
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
3909 cgroup
= memcg
->css
.cgroup
;
3910 if (!memcg
->use_hierarchy
)
3913 while (cgroup
->parent
) {
3914 cgroup
= cgroup
->parent
;
3915 memcg
= mem_cgroup_from_cont(cgroup
);
3916 if (!memcg
->use_hierarchy
)
3918 tmp
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
3919 min_limit
= min(min_limit
, tmp
);
3920 tmp
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
3921 min_memsw_limit
= min(min_memsw_limit
, tmp
);
3924 *mem_limit
= min_limit
;
3925 *memsw_limit
= min_memsw_limit
;
3929 static int mem_cgroup_reset(struct cgroup
*cont
, unsigned int event
)
3931 struct mem_cgroup
*memcg
;
3934 memcg
= mem_cgroup_from_cont(cont
);
3935 type
= MEMFILE_TYPE(event
);
3936 name
= MEMFILE_ATTR(event
);
3940 res_counter_reset_max(&memcg
->res
);
3942 res_counter_reset_max(&memcg
->memsw
);
3946 res_counter_reset_failcnt(&memcg
->res
);
3948 res_counter_reset_failcnt(&memcg
->memsw
);
3955 static u64
mem_cgroup_move_charge_read(struct cgroup
*cgrp
,
3958 return mem_cgroup_from_cont(cgrp
)->move_charge_at_immigrate
;
3962 static int mem_cgroup_move_charge_write(struct cgroup
*cgrp
,
3963 struct cftype
*cft
, u64 val
)
3965 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
3967 if (val
>= (1 << NR_MOVE_TYPE
))
3970 * We check this value several times in both in can_attach() and
3971 * attach(), so we need cgroup lock to prevent this value from being
3975 memcg
->move_charge_at_immigrate
= val
;
3981 static int mem_cgroup_move_charge_write(struct cgroup
*cgrp
,
3982 struct cftype
*cft
, u64 val
)
3989 /* For read statistics */
4007 struct mcs_total_stat
{
4008 s64 stat
[NR_MCS_STAT
];
4014 } memcg_stat_strings
[NR_MCS_STAT
] = {
4015 {"cache", "total_cache"},
4016 {"rss", "total_rss"},
4017 {"mapped_file", "total_mapped_file"},
4018 {"pgpgin", "total_pgpgin"},
4019 {"pgpgout", "total_pgpgout"},
4020 {"swap", "total_swap"},
4021 {"pgfault", "total_pgfault"},
4022 {"pgmajfault", "total_pgmajfault"},
4023 {"inactive_anon", "total_inactive_anon"},
4024 {"active_anon", "total_active_anon"},
4025 {"inactive_file", "total_inactive_file"},
4026 {"active_file", "total_active_file"},
4027 {"unevictable", "total_unevictable"}
4032 mem_cgroup_get_local_stat(struct mem_cgroup
*memcg
, struct mcs_total_stat
*s
)
4037 val
= mem_cgroup_read_stat(memcg
, MEM_CGROUP_STAT_CACHE
);
4038 s
->stat
[MCS_CACHE
] += val
* PAGE_SIZE
;
4039 val
= mem_cgroup_read_stat(memcg
, MEM_CGROUP_STAT_RSS
);
4040 s
->stat
[MCS_RSS
] += val
* PAGE_SIZE
;
4041 val
= mem_cgroup_read_stat(memcg
, MEM_CGROUP_STAT_FILE_MAPPED
);
4042 s
->stat
[MCS_FILE_MAPPED
] += val
* PAGE_SIZE
;
4043 val
= mem_cgroup_read_events(memcg
, MEM_CGROUP_EVENTS_PGPGIN
);
4044 s
->stat
[MCS_PGPGIN
] += val
;
4045 val
= mem_cgroup_read_events(memcg
, MEM_CGROUP_EVENTS_PGPGOUT
);
4046 s
->stat
[MCS_PGPGOUT
] += val
;
4047 if (do_swap_account
) {
4048 val
= mem_cgroup_read_stat(memcg
, MEM_CGROUP_STAT_SWAPOUT
);
4049 s
->stat
[MCS_SWAP
] += val
* PAGE_SIZE
;
4051 val
= mem_cgroup_read_events(memcg
, MEM_CGROUP_EVENTS_PGFAULT
);
4052 s
->stat
[MCS_PGFAULT
] += val
;
4053 val
= mem_cgroup_read_events(memcg
, MEM_CGROUP_EVENTS_PGMAJFAULT
);
4054 s
->stat
[MCS_PGMAJFAULT
] += val
;
4057 val
= mem_cgroup_nr_lru_pages(memcg
, BIT(LRU_INACTIVE_ANON
));
4058 s
->stat
[MCS_INACTIVE_ANON
] += val
* PAGE_SIZE
;
4059 val
= mem_cgroup_nr_lru_pages(memcg
, BIT(LRU_ACTIVE_ANON
));
4060 s
->stat
[MCS_ACTIVE_ANON
] += val
* PAGE_SIZE
;
4061 val
= mem_cgroup_nr_lru_pages(memcg
, BIT(LRU_INACTIVE_FILE
));
4062 s
->stat
[MCS_INACTIVE_FILE
] += val
* PAGE_SIZE
;
4063 val
= mem_cgroup_nr_lru_pages(memcg
, BIT(LRU_ACTIVE_FILE
));
4064 s
->stat
[MCS_ACTIVE_FILE
] += val
* PAGE_SIZE
;
4065 val
= mem_cgroup_nr_lru_pages(memcg
, BIT(LRU_UNEVICTABLE
));
4066 s
->stat
[MCS_UNEVICTABLE
] += val
* PAGE_SIZE
;
4070 mem_cgroup_get_total_stat(struct mem_cgroup
*memcg
, struct mcs_total_stat
*s
)
4072 struct mem_cgroup
*iter
;
4074 for_each_mem_cgroup_tree(iter
, memcg
)
4075 mem_cgroup_get_local_stat(iter
, s
);
4079 static int mem_control_numa_stat_show(struct seq_file
*m
, void *arg
)
4082 unsigned long total_nr
, file_nr
, anon_nr
, unevictable_nr
;
4083 unsigned long node_nr
;
4084 struct cgroup
*cont
= m
->private;
4085 struct mem_cgroup
*mem_cont
= mem_cgroup_from_cont(cont
);
4087 total_nr
= mem_cgroup_nr_lru_pages(mem_cont
, LRU_ALL
);
4088 seq_printf(m
, "total=%lu", total_nr
);
4089 for_each_node_state(nid
, N_HIGH_MEMORY
) {
4090 node_nr
= mem_cgroup_node_nr_lru_pages(mem_cont
, nid
, LRU_ALL
);
4091 seq_printf(m
, " N%d=%lu", nid
, node_nr
);
4095 file_nr
= mem_cgroup_nr_lru_pages(mem_cont
, LRU_ALL_FILE
);
4096 seq_printf(m
, "file=%lu", file_nr
);
4097 for_each_node_state(nid
, N_HIGH_MEMORY
) {
4098 node_nr
= mem_cgroup_node_nr_lru_pages(mem_cont
, nid
,
4100 seq_printf(m
, " N%d=%lu", nid
, node_nr
);
4104 anon_nr
= mem_cgroup_nr_lru_pages(mem_cont
, LRU_ALL_ANON
);
4105 seq_printf(m
, "anon=%lu", anon_nr
);
4106 for_each_node_state(nid
, N_HIGH_MEMORY
) {
4107 node_nr
= mem_cgroup_node_nr_lru_pages(mem_cont
, nid
,
4109 seq_printf(m
, " N%d=%lu", nid
, node_nr
);
4113 unevictable_nr
= mem_cgroup_nr_lru_pages(mem_cont
, BIT(LRU_UNEVICTABLE
));
4114 seq_printf(m
, "unevictable=%lu", unevictable_nr
);
4115 for_each_node_state(nid
, N_HIGH_MEMORY
) {
4116 node_nr
= mem_cgroup_node_nr_lru_pages(mem_cont
, nid
,
4117 BIT(LRU_UNEVICTABLE
));
4118 seq_printf(m
, " N%d=%lu", nid
, node_nr
);
4123 #endif /* CONFIG_NUMA */
4125 static int mem_control_stat_show(struct cgroup
*cont
, struct cftype
*cft
,
4126 struct cgroup_map_cb
*cb
)
4128 struct mem_cgroup
*mem_cont
= mem_cgroup_from_cont(cont
);
4129 struct mcs_total_stat mystat
;
4132 memset(&mystat
, 0, sizeof(mystat
));
4133 mem_cgroup_get_local_stat(mem_cont
, &mystat
);
4136 for (i
= 0; i
< NR_MCS_STAT
; i
++) {
4137 if (i
== MCS_SWAP
&& !do_swap_account
)
4139 cb
->fill(cb
, memcg_stat_strings
[i
].local_name
, mystat
.stat
[i
]);
4142 /* Hierarchical information */
4144 unsigned long long limit
, memsw_limit
;
4145 memcg_get_hierarchical_limit(mem_cont
, &limit
, &memsw_limit
);
4146 cb
->fill(cb
, "hierarchical_memory_limit", limit
);
4147 if (do_swap_account
)
4148 cb
->fill(cb
, "hierarchical_memsw_limit", memsw_limit
);
4151 memset(&mystat
, 0, sizeof(mystat
));
4152 mem_cgroup_get_total_stat(mem_cont
, &mystat
);
4153 for (i
= 0; i
< NR_MCS_STAT
; i
++) {
4154 if (i
== MCS_SWAP
&& !do_swap_account
)
4156 cb
->fill(cb
, memcg_stat_strings
[i
].total_name
, mystat
.stat
[i
]);
4159 #ifdef CONFIG_DEBUG_VM
4162 struct mem_cgroup_per_zone
*mz
;
4163 unsigned long recent_rotated
[2] = {0, 0};
4164 unsigned long recent_scanned
[2] = {0, 0};
4166 for_each_online_node(nid
)
4167 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++) {
4168 mz
= mem_cgroup_zoneinfo(mem_cont
, nid
, zid
);
4170 recent_rotated
[0] +=
4171 mz
->reclaim_stat
.recent_rotated
[0];
4172 recent_rotated
[1] +=
4173 mz
->reclaim_stat
.recent_rotated
[1];
4174 recent_scanned
[0] +=
4175 mz
->reclaim_stat
.recent_scanned
[0];
4176 recent_scanned
[1] +=
4177 mz
->reclaim_stat
.recent_scanned
[1];
4179 cb
->fill(cb
, "recent_rotated_anon", recent_rotated
[0]);
4180 cb
->fill(cb
, "recent_rotated_file", recent_rotated
[1]);
4181 cb
->fill(cb
, "recent_scanned_anon", recent_scanned
[0]);
4182 cb
->fill(cb
, "recent_scanned_file", recent_scanned
[1]);
4189 static u64
mem_cgroup_swappiness_read(struct cgroup
*cgrp
, struct cftype
*cft
)
4191 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4193 return mem_cgroup_swappiness(memcg
);
4196 static int mem_cgroup_swappiness_write(struct cgroup
*cgrp
, struct cftype
*cft
,
4199 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4200 struct mem_cgroup
*parent
;
4205 if (cgrp
->parent
== NULL
)
4208 parent
= mem_cgroup_from_cont(cgrp
->parent
);
4212 /* If under hierarchy, only empty-root can set this value */
4213 if ((parent
->use_hierarchy
) ||
4214 (memcg
->use_hierarchy
&& !list_empty(&cgrp
->children
))) {
4219 memcg
->swappiness
= val
;
4226 static void __mem_cgroup_threshold(struct mem_cgroup
*memcg
, bool swap
)
4228 struct mem_cgroup_threshold_ary
*t
;
4234 t
= rcu_dereference(memcg
->thresholds
.primary
);
4236 t
= rcu_dereference(memcg
->memsw_thresholds
.primary
);
4241 usage
= mem_cgroup_usage(memcg
, swap
);
4244 * current_threshold points to threshold just below usage.
4245 * If it's not true, a threshold was crossed after last
4246 * call of __mem_cgroup_threshold().
4248 i
= t
->current_threshold
;
4251 * Iterate backward over array of thresholds starting from
4252 * current_threshold and check if a threshold is crossed.
4253 * If none of thresholds below usage is crossed, we read
4254 * only one element of the array here.
4256 for (; i
>= 0 && unlikely(t
->entries
[i
].threshold
> usage
); i
--)
4257 eventfd_signal(t
->entries
[i
].eventfd
, 1);
4259 /* i = current_threshold + 1 */
4263 * Iterate forward over array of thresholds starting from
4264 * current_threshold+1 and check if a threshold is crossed.
4265 * If none of thresholds above usage is crossed, we read
4266 * only one element of the array here.
4268 for (; i
< t
->size
&& unlikely(t
->entries
[i
].threshold
<= usage
); i
++)
4269 eventfd_signal(t
->entries
[i
].eventfd
, 1);
4271 /* Update current_threshold */
4272 t
->current_threshold
= i
- 1;
4277 static void mem_cgroup_threshold(struct mem_cgroup
*memcg
)
4280 __mem_cgroup_threshold(memcg
, false);
4281 if (do_swap_account
)
4282 __mem_cgroup_threshold(memcg
, true);
4284 memcg
= parent_mem_cgroup(memcg
);
4288 static int compare_thresholds(const void *a
, const void *b
)
4290 const struct mem_cgroup_threshold
*_a
= a
;
4291 const struct mem_cgroup_threshold
*_b
= b
;
4293 return _a
->threshold
- _b
->threshold
;
4296 static int mem_cgroup_oom_notify_cb(struct mem_cgroup
*memcg
)
4298 struct mem_cgroup_eventfd_list
*ev
;
4300 list_for_each_entry(ev
, &memcg
->oom_notify
, list
)
4301 eventfd_signal(ev
->eventfd
, 1);
4305 static void mem_cgroup_oom_notify(struct mem_cgroup
*memcg
)
4307 struct mem_cgroup
*iter
;
4309 for_each_mem_cgroup_tree(iter
, memcg
)
4310 mem_cgroup_oom_notify_cb(iter
);
4313 static int mem_cgroup_usage_register_event(struct cgroup
*cgrp
,
4314 struct cftype
*cft
, struct eventfd_ctx
*eventfd
, const char *args
)
4316 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4317 struct mem_cgroup_thresholds
*thresholds
;
4318 struct mem_cgroup_threshold_ary
*new;
4319 int type
= MEMFILE_TYPE(cft
->private);
4320 u64 threshold
, usage
;
4323 ret
= res_counter_memparse_write_strategy(args
, &threshold
);
4327 mutex_lock(&memcg
->thresholds_lock
);
4330 thresholds
= &memcg
->thresholds
;
4331 else if (type
== _MEMSWAP
)
4332 thresholds
= &memcg
->memsw_thresholds
;
4336 usage
= mem_cgroup_usage(memcg
, type
== _MEMSWAP
);
4338 /* Check if a threshold crossed before adding a new one */
4339 if (thresholds
->primary
)
4340 __mem_cgroup_threshold(memcg
, type
== _MEMSWAP
);
4342 size
= thresholds
->primary
? thresholds
->primary
->size
+ 1 : 1;
4344 /* Allocate memory for new array of thresholds */
4345 new = kmalloc(sizeof(*new) + size
* sizeof(struct mem_cgroup_threshold
),
4353 /* Copy thresholds (if any) to new array */
4354 if (thresholds
->primary
) {
4355 memcpy(new->entries
, thresholds
->primary
->entries
, (size
- 1) *
4356 sizeof(struct mem_cgroup_threshold
));
4359 /* Add new threshold */
4360 new->entries
[size
- 1].eventfd
= eventfd
;
4361 new->entries
[size
- 1].threshold
= threshold
;
4363 /* Sort thresholds. Registering of new threshold isn't time-critical */
4364 sort(new->entries
, size
, sizeof(struct mem_cgroup_threshold
),
4365 compare_thresholds
, NULL
);
4367 /* Find current threshold */
4368 new->current_threshold
= -1;
4369 for (i
= 0; i
< size
; i
++) {
4370 if (new->entries
[i
].threshold
< usage
) {
4372 * new->current_threshold will not be used until
4373 * rcu_assign_pointer(), so it's safe to increment
4376 ++new->current_threshold
;
4380 /* Free old spare buffer and save old primary buffer as spare */
4381 kfree(thresholds
->spare
);
4382 thresholds
->spare
= thresholds
->primary
;
4384 rcu_assign_pointer(thresholds
->primary
, new);
4386 /* To be sure that nobody uses thresholds */
4390 mutex_unlock(&memcg
->thresholds_lock
);
4395 static void mem_cgroup_usage_unregister_event(struct cgroup
*cgrp
,
4396 struct cftype
*cft
, struct eventfd_ctx
*eventfd
)
4398 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4399 struct mem_cgroup_thresholds
*thresholds
;
4400 struct mem_cgroup_threshold_ary
*new;
4401 int type
= MEMFILE_TYPE(cft
->private);
4405 mutex_lock(&memcg
->thresholds_lock
);
4407 thresholds
= &memcg
->thresholds
;
4408 else if (type
== _MEMSWAP
)
4409 thresholds
= &memcg
->memsw_thresholds
;
4414 * Something went wrong if we trying to unregister a threshold
4415 * if we don't have thresholds
4417 BUG_ON(!thresholds
);
4419 if (!thresholds
->primary
)
4422 usage
= mem_cgroup_usage(memcg
, type
== _MEMSWAP
);
4424 /* Check if a threshold crossed before removing */
4425 __mem_cgroup_threshold(memcg
, type
== _MEMSWAP
);
4427 /* Calculate new number of threshold */
4429 for (i
= 0; i
< thresholds
->primary
->size
; i
++) {
4430 if (thresholds
->primary
->entries
[i
].eventfd
!= eventfd
)
4434 new = thresholds
->spare
;
4436 /* Set thresholds array to NULL if we don't have thresholds */
4445 /* Copy thresholds and find current threshold */
4446 new->current_threshold
= -1;
4447 for (i
= 0, j
= 0; i
< thresholds
->primary
->size
; i
++) {
4448 if (thresholds
->primary
->entries
[i
].eventfd
== eventfd
)
4451 new->entries
[j
] = thresholds
->primary
->entries
[i
];
4452 if (new->entries
[j
].threshold
< usage
) {
4454 * new->current_threshold will not be used
4455 * until rcu_assign_pointer(), so it's safe to increment
4458 ++new->current_threshold
;
4464 /* Swap primary and spare array */
4465 thresholds
->spare
= thresholds
->primary
;
4466 rcu_assign_pointer(thresholds
->primary
, new);
4468 /* To be sure that nobody uses thresholds */
4471 mutex_unlock(&memcg
->thresholds_lock
);
4474 static int mem_cgroup_oom_register_event(struct cgroup
*cgrp
,
4475 struct cftype
*cft
, struct eventfd_ctx
*eventfd
, const char *args
)
4477 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4478 struct mem_cgroup_eventfd_list
*event
;
4479 int type
= MEMFILE_TYPE(cft
->private);
4481 BUG_ON(type
!= _OOM_TYPE
);
4482 event
= kmalloc(sizeof(*event
), GFP_KERNEL
);
4486 spin_lock(&memcg_oom_lock
);
4488 event
->eventfd
= eventfd
;
4489 list_add(&event
->list
, &memcg
->oom_notify
);
4491 /* already in OOM ? */
4492 if (atomic_read(&memcg
->under_oom
))
4493 eventfd_signal(eventfd
, 1);
4494 spin_unlock(&memcg_oom_lock
);
4499 static void mem_cgroup_oom_unregister_event(struct cgroup
*cgrp
,
4500 struct cftype
*cft
, struct eventfd_ctx
*eventfd
)
4502 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4503 struct mem_cgroup_eventfd_list
*ev
, *tmp
;
4504 int type
= MEMFILE_TYPE(cft
->private);
4506 BUG_ON(type
!= _OOM_TYPE
);
4508 spin_lock(&memcg_oom_lock
);
4510 list_for_each_entry_safe(ev
, tmp
, &memcg
->oom_notify
, list
) {
4511 if (ev
->eventfd
== eventfd
) {
4512 list_del(&ev
->list
);
4517 spin_unlock(&memcg_oom_lock
);
4520 static int mem_cgroup_oom_control_read(struct cgroup
*cgrp
,
4521 struct cftype
*cft
, struct cgroup_map_cb
*cb
)
4523 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4525 cb
->fill(cb
, "oom_kill_disable", memcg
->oom_kill_disable
);
4527 if (atomic_read(&memcg
->under_oom
))
4528 cb
->fill(cb
, "under_oom", 1);
4530 cb
->fill(cb
, "under_oom", 0);
4534 static int mem_cgroup_oom_control_write(struct cgroup
*cgrp
,
4535 struct cftype
*cft
, u64 val
)
4537 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4538 struct mem_cgroup
*parent
;
4540 /* cannot set to root cgroup and only 0 and 1 are allowed */
4541 if (!cgrp
->parent
|| !((val
== 0) || (val
== 1)))
4544 parent
= mem_cgroup_from_cont(cgrp
->parent
);
4547 /* oom-kill-disable is a flag for subhierarchy. */
4548 if ((parent
->use_hierarchy
) ||
4549 (memcg
->use_hierarchy
&& !list_empty(&cgrp
->children
))) {
4553 memcg
->oom_kill_disable
= val
;
4555 memcg_oom_recover(memcg
);
4561 static const struct file_operations mem_control_numa_stat_file_operations
= {
4563 .llseek
= seq_lseek
,
4564 .release
= single_release
,
4567 static int mem_control_numa_stat_open(struct inode
*unused
, struct file
*file
)
4569 struct cgroup
*cont
= file
->f_dentry
->d_parent
->d_fsdata
;
4571 file
->f_op
= &mem_control_numa_stat_file_operations
;
4572 return single_open(file
, mem_control_numa_stat_show
, cont
);
4574 #endif /* CONFIG_NUMA */
4576 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_KMEM
4577 static int register_kmem_files(struct cgroup
*cont
, struct cgroup_subsys
*ss
)
4580 * Part of this would be better living in a separate allocation
4581 * function, leaving us with just the cgroup tree population work.
4582 * We, however, depend on state such as network's proto_list that
4583 * is only initialized after cgroup creation. I found the less
4584 * cumbersome way to deal with it to defer it all to populate time
4586 return mem_cgroup_sockets_init(cont
, ss
);
4589 static void kmem_cgroup_destroy(struct cgroup_subsys
*ss
,
4590 struct cgroup
*cont
)
4592 mem_cgroup_sockets_destroy(cont
, ss
);
4595 static int register_kmem_files(struct cgroup
*cont
, struct cgroup_subsys
*ss
)
4600 static void kmem_cgroup_destroy(struct cgroup_subsys
*ss
,
4601 struct cgroup
*cont
)
4606 static struct cftype mem_cgroup_files
[] = {
4608 .name
= "usage_in_bytes",
4609 .private = MEMFILE_PRIVATE(_MEM
, RES_USAGE
),
4610 .read_u64
= mem_cgroup_read
,
4611 .register_event
= mem_cgroup_usage_register_event
,
4612 .unregister_event
= mem_cgroup_usage_unregister_event
,
4615 .name
= "max_usage_in_bytes",
4616 .private = MEMFILE_PRIVATE(_MEM
, RES_MAX_USAGE
),
4617 .trigger
= mem_cgroup_reset
,
4618 .read_u64
= mem_cgroup_read
,
4621 .name
= "limit_in_bytes",
4622 .private = MEMFILE_PRIVATE(_MEM
, RES_LIMIT
),
4623 .write_string
= mem_cgroup_write
,
4624 .read_u64
= mem_cgroup_read
,
4627 .name
= "soft_limit_in_bytes",
4628 .private = MEMFILE_PRIVATE(_MEM
, RES_SOFT_LIMIT
),
4629 .write_string
= mem_cgroup_write
,
4630 .read_u64
= mem_cgroup_read
,
4634 .private = MEMFILE_PRIVATE(_MEM
, RES_FAILCNT
),
4635 .trigger
= mem_cgroup_reset
,
4636 .read_u64
= mem_cgroup_read
,
4640 .read_map
= mem_control_stat_show
,
4643 .name
= "force_empty",
4644 .trigger
= mem_cgroup_force_empty_write
,
4647 .name
= "use_hierarchy",
4648 .write_u64
= mem_cgroup_hierarchy_write
,
4649 .read_u64
= mem_cgroup_hierarchy_read
,
4652 .name
= "swappiness",
4653 .read_u64
= mem_cgroup_swappiness_read
,
4654 .write_u64
= mem_cgroup_swappiness_write
,
4657 .name
= "move_charge_at_immigrate",
4658 .read_u64
= mem_cgroup_move_charge_read
,
4659 .write_u64
= mem_cgroup_move_charge_write
,
4662 .name
= "oom_control",
4663 .read_map
= mem_cgroup_oom_control_read
,
4664 .write_u64
= mem_cgroup_oom_control_write
,
4665 .register_event
= mem_cgroup_oom_register_event
,
4666 .unregister_event
= mem_cgroup_oom_unregister_event
,
4667 .private = MEMFILE_PRIVATE(_OOM_TYPE
, OOM_CONTROL
),
4671 .name
= "numa_stat",
4672 .open
= mem_control_numa_stat_open
,
4678 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
4679 static struct cftype memsw_cgroup_files
[] = {
4681 .name
= "memsw.usage_in_bytes",
4682 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_USAGE
),
4683 .read_u64
= mem_cgroup_read
,
4684 .register_event
= mem_cgroup_usage_register_event
,
4685 .unregister_event
= mem_cgroup_usage_unregister_event
,
4688 .name
= "memsw.max_usage_in_bytes",
4689 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_MAX_USAGE
),
4690 .trigger
= mem_cgroup_reset
,
4691 .read_u64
= mem_cgroup_read
,
4694 .name
= "memsw.limit_in_bytes",
4695 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_LIMIT
),
4696 .write_string
= mem_cgroup_write
,
4697 .read_u64
= mem_cgroup_read
,
4700 .name
= "memsw.failcnt",
4701 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_FAILCNT
),
4702 .trigger
= mem_cgroup_reset
,
4703 .read_u64
= mem_cgroup_read
,
4707 static int register_memsw_files(struct cgroup
*cont
, struct cgroup_subsys
*ss
)
4709 if (!do_swap_account
)
4711 return cgroup_add_files(cont
, ss
, memsw_cgroup_files
,
4712 ARRAY_SIZE(memsw_cgroup_files
));
4715 static int register_memsw_files(struct cgroup
*cont
, struct cgroup_subsys
*ss
)
4721 static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup
*memcg
, int node
)
4723 struct mem_cgroup_per_node
*pn
;
4724 struct mem_cgroup_per_zone
*mz
;
4726 int zone
, tmp
= node
;
4728 * This routine is called against possible nodes.
4729 * But it's BUG to call kmalloc() against offline node.
4731 * TODO: this routine can waste much memory for nodes which will
4732 * never be onlined. It's better to use memory hotplug callback
4735 if (!node_state(node
, N_NORMAL_MEMORY
))
4737 pn
= kzalloc_node(sizeof(*pn
), GFP_KERNEL
, tmp
);
4741 for (zone
= 0; zone
< MAX_NR_ZONES
; zone
++) {
4742 mz
= &pn
->zoneinfo
[zone
];
4744 INIT_LIST_HEAD(&mz
->lruvec
.lists
[l
]);
4745 mz
->usage_in_excess
= 0;
4746 mz
->on_tree
= false;
4749 memcg
->info
.nodeinfo
[node
] = pn
;
4753 static void free_mem_cgroup_per_zone_info(struct mem_cgroup
*memcg
, int node
)
4755 kfree(memcg
->info
.nodeinfo
[node
]);
4758 static struct mem_cgroup
*mem_cgroup_alloc(void)
4760 struct mem_cgroup
*mem
;
4761 int size
= sizeof(struct mem_cgroup
);
4763 /* Can be very big if MAX_NUMNODES is very big */
4764 if (size
< PAGE_SIZE
)
4765 mem
= kzalloc(size
, GFP_KERNEL
);
4767 mem
= vzalloc(size
);
4772 mem
->stat
= alloc_percpu(struct mem_cgroup_stat_cpu
);
4775 spin_lock_init(&mem
->pcp_counter_lock
);
4779 if (size
< PAGE_SIZE
)
4787 * At destroying mem_cgroup, references from swap_cgroup can remain.
4788 * (scanning all at force_empty is too costly...)
4790 * Instead of clearing all references at force_empty, we remember
4791 * the number of reference from swap_cgroup and free mem_cgroup when
4792 * it goes down to 0.
4794 * Removal of cgroup itself succeeds regardless of refs from swap.
4797 static void __mem_cgroup_free(struct mem_cgroup
*memcg
)
4801 mem_cgroup_remove_from_trees(memcg
);
4802 free_css_id(&mem_cgroup_subsys
, &memcg
->css
);
4805 free_mem_cgroup_per_zone_info(memcg
, node
);
4807 free_percpu(memcg
->stat
);
4808 if (sizeof(struct mem_cgroup
) < PAGE_SIZE
)
4814 static void mem_cgroup_get(struct mem_cgroup
*memcg
)
4816 atomic_inc(&memcg
->refcnt
);
4819 static void __mem_cgroup_put(struct mem_cgroup
*memcg
, int count
)
4821 if (atomic_sub_and_test(count
, &memcg
->refcnt
)) {
4822 struct mem_cgroup
*parent
= parent_mem_cgroup(memcg
);
4823 __mem_cgroup_free(memcg
);
4825 mem_cgroup_put(parent
);
4829 static void mem_cgroup_put(struct mem_cgroup
*memcg
)
4831 __mem_cgroup_put(memcg
, 1);
4835 * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled.
4837 struct mem_cgroup
*parent_mem_cgroup(struct mem_cgroup
*memcg
)
4839 if (!memcg
->res
.parent
)
4841 return mem_cgroup_from_res_counter(memcg
->res
.parent
, res
);
4843 EXPORT_SYMBOL(parent_mem_cgroup
);
4845 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
4846 static void __init
enable_swap_cgroup(void)
4848 if (!mem_cgroup_disabled() && really_do_swap_account
)
4849 do_swap_account
= 1;
4852 static void __init
enable_swap_cgroup(void)
4857 static int mem_cgroup_soft_limit_tree_init(void)
4859 struct mem_cgroup_tree_per_node
*rtpn
;
4860 struct mem_cgroup_tree_per_zone
*rtpz
;
4861 int tmp
, node
, zone
;
4863 for_each_node(node
) {
4865 if (!node_state(node
, N_NORMAL_MEMORY
))
4867 rtpn
= kzalloc_node(sizeof(*rtpn
), GFP_KERNEL
, tmp
);
4871 soft_limit_tree
.rb_tree_per_node
[node
] = rtpn
;
4873 for (zone
= 0; zone
< MAX_NR_ZONES
; zone
++) {
4874 rtpz
= &rtpn
->rb_tree_per_zone
[zone
];
4875 rtpz
->rb_root
= RB_ROOT
;
4876 spin_lock_init(&rtpz
->lock
);
4882 for_each_node(node
) {
4883 if (!soft_limit_tree
.rb_tree_per_node
[node
])
4885 kfree(soft_limit_tree
.rb_tree_per_node
[node
]);
4886 soft_limit_tree
.rb_tree_per_node
[node
] = NULL
;
4892 static struct cgroup_subsys_state
* __ref
4893 mem_cgroup_create(struct cgroup_subsys
*ss
, struct cgroup
*cont
)
4895 struct mem_cgroup
*memcg
, *parent
;
4896 long error
= -ENOMEM
;
4899 memcg
= mem_cgroup_alloc();
4901 return ERR_PTR(error
);
4904 if (alloc_mem_cgroup_per_zone_info(memcg
, node
))
4908 if (cont
->parent
== NULL
) {
4910 enable_swap_cgroup();
4912 if (mem_cgroup_soft_limit_tree_init())
4914 root_mem_cgroup
= memcg
;
4915 for_each_possible_cpu(cpu
) {
4916 struct memcg_stock_pcp
*stock
=
4917 &per_cpu(memcg_stock
, cpu
);
4918 INIT_WORK(&stock
->work
, drain_local_stock
);
4920 hotcpu_notifier(memcg_cpu_hotplug_callback
, 0);
4922 parent
= mem_cgroup_from_cont(cont
->parent
);
4923 memcg
->use_hierarchy
= parent
->use_hierarchy
;
4924 memcg
->oom_kill_disable
= parent
->oom_kill_disable
;
4927 if (parent
&& parent
->use_hierarchy
) {
4928 res_counter_init(&memcg
->res
, &parent
->res
);
4929 res_counter_init(&memcg
->memsw
, &parent
->memsw
);
4931 * We increment refcnt of the parent to ensure that we can
4932 * safely access it on res_counter_charge/uncharge.
4933 * This refcnt will be decremented when freeing this
4934 * mem_cgroup(see mem_cgroup_put).
4936 mem_cgroup_get(parent
);
4938 res_counter_init(&memcg
->res
, NULL
);
4939 res_counter_init(&memcg
->memsw
, NULL
);
4941 memcg
->last_scanned_node
= MAX_NUMNODES
;
4942 INIT_LIST_HEAD(&memcg
->oom_notify
);
4945 memcg
->swappiness
= mem_cgroup_swappiness(parent
);
4946 atomic_set(&memcg
->refcnt
, 1);
4947 memcg
->move_charge_at_immigrate
= 0;
4948 mutex_init(&memcg
->thresholds_lock
);
4951 __mem_cgroup_free(memcg
);
4952 return ERR_PTR(error
);
4955 static int mem_cgroup_pre_destroy(struct cgroup_subsys
*ss
,
4956 struct cgroup
*cont
)
4958 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
4960 return mem_cgroup_force_empty(memcg
, false);
4963 static void mem_cgroup_destroy(struct cgroup_subsys
*ss
,
4964 struct cgroup
*cont
)
4966 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
4968 kmem_cgroup_destroy(ss
, cont
);
4970 mem_cgroup_put(memcg
);
4973 static int mem_cgroup_populate(struct cgroup_subsys
*ss
,
4974 struct cgroup
*cont
)
4978 ret
= cgroup_add_files(cont
, ss
, mem_cgroup_files
,
4979 ARRAY_SIZE(mem_cgroup_files
));
4982 ret
= register_memsw_files(cont
, ss
);
4985 ret
= register_kmem_files(cont
, ss
);
4991 /* Handlers for move charge at task migration. */
4992 #define PRECHARGE_COUNT_AT_ONCE 256
4993 static int mem_cgroup_do_precharge(unsigned long count
)
4996 int batch_count
= PRECHARGE_COUNT_AT_ONCE
;
4997 struct mem_cgroup
*memcg
= mc
.to
;
4999 if (mem_cgroup_is_root(memcg
)) {
5000 mc
.precharge
+= count
;
5001 /* we don't need css_get for root */
5004 /* try to charge at once */
5006 struct res_counter
*dummy
;
5008 * "memcg" cannot be under rmdir() because we've already checked
5009 * by cgroup_lock_live_cgroup() that it is not removed and we
5010 * are still under the same cgroup_mutex. So we can postpone
5013 if (res_counter_charge(&memcg
->res
, PAGE_SIZE
* count
, &dummy
))
5015 if (do_swap_account
&& res_counter_charge(&memcg
->memsw
,
5016 PAGE_SIZE
* count
, &dummy
)) {
5017 res_counter_uncharge(&memcg
->res
, PAGE_SIZE
* count
);
5020 mc
.precharge
+= count
;
5024 /* fall back to one by one charge */
5026 if (signal_pending(current
)) {
5030 if (!batch_count
--) {
5031 batch_count
= PRECHARGE_COUNT_AT_ONCE
;
5034 ret
= __mem_cgroup_try_charge(NULL
,
5035 GFP_KERNEL
, 1, &memcg
, false);
5037 /* mem_cgroup_clear_mc() will do uncharge later */
5045 * is_target_pte_for_mc - check a pte whether it is valid for move charge
5046 * @vma: the vma the pte to be checked belongs
5047 * @addr: the address corresponding to the pte to be checked
5048 * @ptent: the pte to be checked
5049 * @target: the pointer the target page or swap ent will be stored(can be NULL)
5052 * 0(MC_TARGET_NONE): if the pte is not a target for move charge.
5053 * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
5054 * move charge. if @target is not NULL, the page is stored in target->page
5055 * with extra refcnt got(Callers should handle it).
5056 * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
5057 * target for charge migration. if @target is not NULL, the entry is stored
5060 * Called with pte lock held.
5067 enum mc_target_type
{
5068 MC_TARGET_NONE
, /* not used */
5073 static struct page
*mc_handle_present_pte(struct vm_area_struct
*vma
,
5074 unsigned long addr
, pte_t ptent
)
5076 struct page
*page
= vm_normal_page(vma
, addr
, ptent
);
5078 if (!page
|| !page_mapped(page
))
5080 if (PageAnon(page
)) {
5081 /* we don't move shared anon */
5082 if (!move_anon() || page_mapcount(page
) > 2)
5084 } else if (!move_file())
5085 /* we ignore mapcount for file pages */
5087 if (!get_page_unless_zero(page
))
5093 static struct page
*mc_handle_swap_pte(struct vm_area_struct
*vma
,
5094 unsigned long addr
, pte_t ptent
, swp_entry_t
*entry
)
5097 struct page
*page
= NULL
;
5098 swp_entry_t ent
= pte_to_swp_entry(ptent
);
5100 if (!move_anon() || non_swap_entry(ent
))
5102 usage_count
= mem_cgroup_count_swap_user(ent
, &page
);
5103 if (usage_count
> 1) { /* we don't move shared anon */
5108 if (do_swap_account
)
5109 entry
->val
= ent
.val
;
5114 static struct page
*mc_handle_file_pte(struct vm_area_struct
*vma
,
5115 unsigned long addr
, pte_t ptent
, swp_entry_t
*entry
)
5117 struct page
*page
= NULL
;
5118 struct inode
*inode
;
5119 struct address_space
*mapping
;
5122 if (!vma
->vm_file
) /* anonymous vma */
5127 inode
= vma
->vm_file
->f_path
.dentry
->d_inode
;
5128 mapping
= vma
->vm_file
->f_mapping
;
5129 if (pte_none(ptent
))
5130 pgoff
= linear_page_index(vma
, addr
);
5131 else /* pte_file(ptent) is true */
5132 pgoff
= pte_to_pgoff(ptent
);
5134 /* page is moved even if it's not RSS of this task(page-faulted). */
5135 page
= find_get_page(mapping
, pgoff
);
5138 /* shmem/tmpfs may report page out on swap: account for that too. */
5139 if (radix_tree_exceptional_entry(page
)) {
5140 swp_entry_t swap
= radix_to_swp_entry(page
);
5141 if (do_swap_account
)
5143 page
= find_get_page(&swapper_space
, swap
.val
);
5149 static int is_target_pte_for_mc(struct vm_area_struct
*vma
,
5150 unsigned long addr
, pte_t ptent
, union mc_target
*target
)
5152 struct page
*page
= NULL
;
5153 struct page_cgroup
*pc
;
5155 swp_entry_t ent
= { .val
= 0 };
5157 if (pte_present(ptent
))
5158 page
= mc_handle_present_pte(vma
, addr
, ptent
);
5159 else if (is_swap_pte(ptent
))
5160 page
= mc_handle_swap_pte(vma
, addr
, ptent
, &ent
);
5161 else if (pte_none(ptent
) || pte_file(ptent
))
5162 page
= mc_handle_file_pte(vma
, addr
, ptent
, &ent
);
5164 if (!page
&& !ent
.val
)
5167 pc
= lookup_page_cgroup(page
);
5169 * Do only loose check w/o page_cgroup lock.
5170 * mem_cgroup_move_account() checks the pc is valid or not under
5173 if (PageCgroupUsed(pc
) && pc
->mem_cgroup
== mc
.from
) {
5174 ret
= MC_TARGET_PAGE
;
5176 target
->page
= page
;
5178 if (!ret
|| !target
)
5181 /* There is a swap entry and a page doesn't exist or isn't charged */
5182 if (ent
.val
&& !ret
&&
5183 css_id(&mc
.from
->css
) == lookup_swap_cgroup_id(ent
)) {
5184 ret
= MC_TARGET_SWAP
;
5191 static int mem_cgroup_count_precharge_pte_range(pmd_t
*pmd
,
5192 unsigned long addr
, unsigned long end
,
5193 struct mm_walk
*walk
)
5195 struct vm_area_struct
*vma
= walk
->private;
5199 split_huge_page_pmd(walk
->mm
, pmd
);
5201 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
5202 for (; addr
!= end
; pte
++, addr
+= PAGE_SIZE
)
5203 if (is_target_pte_for_mc(vma
, addr
, *pte
, NULL
))
5204 mc
.precharge
++; /* increment precharge temporarily */
5205 pte_unmap_unlock(pte
- 1, ptl
);
5211 static unsigned long mem_cgroup_count_precharge(struct mm_struct
*mm
)
5213 unsigned long precharge
;
5214 struct vm_area_struct
*vma
;
5216 down_read(&mm
->mmap_sem
);
5217 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
5218 struct mm_walk mem_cgroup_count_precharge_walk
= {
5219 .pmd_entry
= mem_cgroup_count_precharge_pte_range
,
5223 if (is_vm_hugetlb_page(vma
))
5225 walk_page_range(vma
->vm_start
, vma
->vm_end
,
5226 &mem_cgroup_count_precharge_walk
);
5228 up_read(&mm
->mmap_sem
);
5230 precharge
= mc
.precharge
;
5236 static int mem_cgroup_precharge_mc(struct mm_struct
*mm
)
5238 unsigned long precharge
= mem_cgroup_count_precharge(mm
);
5240 VM_BUG_ON(mc
.moving_task
);
5241 mc
.moving_task
= current
;
5242 return mem_cgroup_do_precharge(precharge
);
5245 /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */
5246 static void __mem_cgroup_clear_mc(void)
5248 struct mem_cgroup
*from
= mc
.from
;
5249 struct mem_cgroup
*to
= mc
.to
;
5251 /* we must uncharge all the leftover precharges from mc.to */
5253 __mem_cgroup_cancel_charge(mc
.to
, mc
.precharge
);
5257 * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
5258 * we must uncharge here.
5260 if (mc
.moved_charge
) {
5261 __mem_cgroup_cancel_charge(mc
.from
, mc
.moved_charge
);
5262 mc
.moved_charge
= 0;
5264 /* we must fixup refcnts and charges */
5265 if (mc
.moved_swap
) {
5266 /* uncharge swap account from the old cgroup */
5267 if (!mem_cgroup_is_root(mc
.from
))
5268 res_counter_uncharge(&mc
.from
->memsw
,
5269 PAGE_SIZE
* mc
.moved_swap
);
5270 __mem_cgroup_put(mc
.from
, mc
.moved_swap
);
5272 if (!mem_cgroup_is_root(mc
.to
)) {
5274 * we charged both to->res and to->memsw, so we should
5277 res_counter_uncharge(&mc
.to
->res
,
5278 PAGE_SIZE
* mc
.moved_swap
);
5280 /* we've already done mem_cgroup_get(mc.to) */
5283 memcg_oom_recover(from
);
5284 memcg_oom_recover(to
);
5285 wake_up_all(&mc
.waitq
);
5288 static void mem_cgroup_clear_mc(void)
5290 struct mem_cgroup
*from
= mc
.from
;
5293 * we must clear moving_task before waking up waiters at the end of
5296 mc
.moving_task
= NULL
;
5297 __mem_cgroup_clear_mc();
5298 spin_lock(&mc
.lock
);
5301 spin_unlock(&mc
.lock
);
5302 mem_cgroup_end_move(from
);
5305 static int mem_cgroup_can_attach(struct cgroup_subsys
*ss
,
5306 struct cgroup
*cgroup
,
5307 struct cgroup_taskset
*tset
)
5309 struct task_struct
*p
= cgroup_taskset_first(tset
);
5311 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgroup
);
5313 if (memcg
->move_charge_at_immigrate
) {
5314 struct mm_struct
*mm
;
5315 struct mem_cgroup
*from
= mem_cgroup_from_task(p
);
5317 VM_BUG_ON(from
== memcg
);
5319 mm
= get_task_mm(p
);
5322 /* We move charges only when we move a owner of the mm */
5323 if (mm
->owner
== p
) {
5326 VM_BUG_ON(mc
.precharge
);
5327 VM_BUG_ON(mc
.moved_charge
);
5328 VM_BUG_ON(mc
.moved_swap
);
5329 mem_cgroup_start_move(from
);
5330 spin_lock(&mc
.lock
);
5333 spin_unlock(&mc
.lock
);
5334 /* We set mc.moving_task later */
5336 ret
= mem_cgroup_precharge_mc(mm
);
5338 mem_cgroup_clear_mc();
5345 static void mem_cgroup_cancel_attach(struct cgroup_subsys
*ss
,
5346 struct cgroup
*cgroup
,
5347 struct cgroup_taskset
*tset
)
5349 mem_cgroup_clear_mc();
5352 static int mem_cgroup_move_charge_pte_range(pmd_t
*pmd
,
5353 unsigned long addr
, unsigned long end
,
5354 struct mm_walk
*walk
)
5357 struct vm_area_struct
*vma
= walk
->private;
5361 split_huge_page_pmd(walk
->mm
, pmd
);
5363 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
5364 for (; addr
!= end
; addr
+= PAGE_SIZE
) {
5365 pte_t ptent
= *(pte
++);
5366 union mc_target target
;
5369 struct page_cgroup
*pc
;
5375 type
= is_target_pte_for_mc(vma
, addr
, ptent
, &target
);
5377 case MC_TARGET_PAGE
:
5379 if (isolate_lru_page(page
))
5381 pc
= lookup_page_cgroup(page
);
5382 if (!mem_cgroup_move_account(page
, 1, pc
,
5383 mc
.from
, mc
.to
, false)) {
5385 /* we uncharge from mc.from later. */
5388 putback_lru_page(page
);
5389 put
: /* is_target_pte_for_mc() gets the page */
5392 case MC_TARGET_SWAP
:
5394 if (!mem_cgroup_move_swap_account(ent
,
5395 mc
.from
, mc
.to
, false)) {
5397 /* we fixup refcnts and charges later. */
5405 pte_unmap_unlock(pte
- 1, ptl
);
5410 * We have consumed all precharges we got in can_attach().
5411 * We try charge one by one, but don't do any additional
5412 * charges to mc.to if we have failed in charge once in attach()
5415 ret
= mem_cgroup_do_precharge(1);
5423 static void mem_cgroup_move_charge(struct mm_struct
*mm
)
5425 struct vm_area_struct
*vma
;
5427 lru_add_drain_all();
5429 if (unlikely(!down_read_trylock(&mm
->mmap_sem
))) {
5431 * Someone who are holding the mmap_sem might be waiting in
5432 * waitq. So we cancel all extra charges, wake up all waiters,
5433 * and retry. Because we cancel precharges, we might not be able
5434 * to move enough charges, but moving charge is a best-effort
5435 * feature anyway, so it wouldn't be a big problem.
5437 __mem_cgroup_clear_mc();
5441 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
5443 struct mm_walk mem_cgroup_move_charge_walk
= {
5444 .pmd_entry
= mem_cgroup_move_charge_pte_range
,
5448 if (is_vm_hugetlb_page(vma
))
5450 ret
= walk_page_range(vma
->vm_start
, vma
->vm_end
,
5451 &mem_cgroup_move_charge_walk
);
5454 * means we have consumed all precharges and failed in
5455 * doing additional charge. Just abandon here.
5459 up_read(&mm
->mmap_sem
);
5462 static void mem_cgroup_move_task(struct cgroup_subsys
*ss
,
5463 struct cgroup
*cont
,
5464 struct cgroup_taskset
*tset
)
5466 struct task_struct
*p
= cgroup_taskset_first(tset
);
5467 struct mm_struct
*mm
= get_task_mm(p
);
5471 mem_cgroup_move_charge(mm
);
5476 mem_cgroup_clear_mc();
5478 #else /* !CONFIG_MMU */
5479 static int mem_cgroup_can_attach(struct cgroup_subsys
*ss
,
5480 struct cgroup
*cgroup
,
5481 struct cgroup_taskset
*tset
)
5485 static void mem_cgroup_cancel_attach(struct cgroup_subsys
*ss
,
5486 struct cgroup
*cgroup
,
5487 struct cgroup_taskset
*tset
)
5490 static void mem_cgroup_move_task(struct cgroup_subsys
*ss
,
5491 struct cgroup
*cont
,
5492 struct cgroup_taskset
*tset
)
5497 struct cgroup_subsys mem_cgroup_subsys
= {
5499 .subsys_id
= mem_cgroup_subsys_id
,
5500 .create
= mem_cgroup_create
,
5501 .pre_destroy
= mem_cgroup_pre_destroy
,
5502 .destroy
= mem_cgroup_destroy
,
5503 .populate
= mem_cgroup_populate
,
5504 .can_attach
= mem_cgroup_can_attach
,
5505 .cancel_attach
= mem_cgroup_cancel_attach
,
5506 .attach
= mem_cgroup_move_task
,
5511 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
5512 static int __init
enable_swap_account(char *s
)
5514 /* consider enabled if no parameter or 1 is given */
5515 if (!strcmp(s
, "1"))
5516 really_do_swap_account
= 1;
5517 else if (!strcmp(s
, "0"))
5518 really_do_swap_account
= 0;
5521 __setup("swapaccount=", enable_swap_account
);