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/mutex.h>
37 #include <linux/rbtree.h>
38 #include <linux/slab.h>
39 #include <linux/swap.h>
40 #include <linux/swapops.h>
41 #include <linux/spinlock.h>
42 #include <linux/eventfd.h>
43 #include <linux/sort.h>
45 #include <linux/seq_file.h>
46 #include <linux/vmalloc.h>
47 #include <linux/mm_inline.h>
48 #include <linux/page_cgroup.h>
49 #include <linux/cpu.h>
50 #include <linux/oom.h>
53 #include <asm/uaccess.h>
55 #include <trace/events/vmscan.h>
57 struct cgroup_subsys mem_cgroup_subsys __read_mostly
;
58 #define MEM_CGROUP_RECLAIM_RETRIES 5
59 struct mem_cgroup
*root_mem_cgroup __read_mostly
;
61 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
62 /* Turned on only when memory cgroup is enabled && really_do_swap_account = 1 */
63 int do_swap_account __read_mostly
;
65 /* for remember boot option*/
66 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP_ENABLED
67 static int really_do_swap_account __initdata
= 1;
69 static int really_do_swap_account __initdata
= 0;
73 #define do_swap_account (0)
78 * Statistics for memory cgroup.
80 enum mem_cgroup_stat_index
{
82 * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss.
84 MEM_CGROUP_STAT_CACHE
, /* # of pages charged as cache */
85 MEM_CGROUP_STAT_RSS
, /* # of pages charged as anon rss */
86 MEM_CGROUP_STAT_FILE_MAPPED
, /* # of pages charged as file rss */
87 MEM_CGROUP_STAT_SWAPOUT
, /* # of pages, swapped out */
88 MEM_CGROUP_STAT_DATA
, /* end of data requires synchronization */
89 MEM_CGROUP_ON_MOVE
, /* someone is moving account between groups */
90 MEM_CGROUP_STAT_NSTATS
,
93 enum mem_cgroup_events_index
{
94 MEM_CGROUP_EVENTS_PGPGIN
, /* # of pages paged in */
95 MEM_CGROUP_EVENTS_PGPGOUT
, /* # of pages paged out */
96 MEM_CGROUP_EVENTS_COUNT
, /* # of pages paged in/out */
97 MEM_CGROUP_EVENTS_PGFAULT
, /* # of page-faults */
98 MEM_CGROUP_EVENTS_PGMAJFAULT
, /* # of major page-faults */
99 MEM_CGROUP_EVENTS_NSTATS
,
102 * Per memcg event counter is incremented at every pagein/pageout. With THP,
103 * it will be incremated by the number of pages. This counter is used for
104 * for trigger some periodic events. This is straightforward and better
105 * than using jiffies etc. to handle periodic memcg event.
107 enum mem_cgroup_events_target
{
108 MEM_CGROUP_TARGET_THRESH
,
109 MEM_CGROUP_TARGET_SOFTLIMIT
,
110 MEM_CGROUP_TARGET_NUMAINFO
,
113 #define THRESHOLDS_EVENTS_TARGET (128)
114 #define SOFTLIMIT_EVENTS_TARGET (1024)
115 #define NUMAINFO_EVENTS_TARGET (1024)
117 struct mem_cgroup_stat_cpu
{
118 long count
[MEM_CGROUP_STAT_NSTATS
];
119 unsigned long events
[MEM_CGROUP_EVENTS_NSTATS
];
120 unsigned long targets
[MEM_CGROUP_NTARGETS
];
124 * per-zone information in memory controller.
126 struct mem_cgroup_per_zone
{
128 * spin_lock to protect the per cgroup LRU
130 struct list_head lists
[NR_LRU_LISTS
];
131 unsigned long count
[NR_LRU_LISTS
];
133 struct zone_reclaim_stat reclaim_stat
;
134 struct rb_node tree_node
; /* RB tree node */
135 unsigned long long usage_in_excess
;/* Set to the value by which */
136 /* the soft limit is exceeded*/
138 struct mem_cgroup
*mem
; /* Back pointer, we cannot */
139 /* use container_of */
141 /* Macro for accessing counter */
142 #define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)])
144 struct mem_cgroup_per_node
{
145 struct mem_cgroup_per_zone zoneinfo
[MAX_NR_ZONES
];
148 struct mem_cgroup_lru_info
{
149 struct mem_cgroup_per_node
*nodeinfo
[MAX_NUMNODES
];
153 * Cgroups above their limits are maintained in a RB-Tree, independent of
154 * their hierarchy representation
157 struct mem_cgroup_tree_per_zone
{
158 struct rb_root rb_root
;
162 struct mem_cgroup_tree_per_node
{
163 struct mem_cgroup_tree_per_zone rb_tree_per_zone
[MAX_NR_ZONES
];
166 struct mem_cgroup_tree
{
167 struct mem_cgroup_tree_per_node
*rb_tree_per_node
[MAX_NUMNODES
];
170 static struct mem_cgroup_tree soft_limit_tree __read_mostly
;
172 struct mem_cgroup_threshold
{
173 struct eventfd_ctx
*eventfd
;
178 struct mem_cgroup_threshold_ary
{
179 /* An array index points to threshold just below usage. */
180 int current_threshold
;
181 /* Size of entries[] */
183 /* Array of thresholds */
184 struct mem_cgroup_threshold entries
[0];
187 struct mem_cgroup_thresholds
{
188 /* Primary thresholds array */
189 struct mem_cgroup_threshold_ary
*primary
;
191 * Spare threshold array.
192 * This is needed to make mem_cgroup_unregister_event() "never fail".
193 * It must be able to store at least primary->size - 1 entries.
195 struct mem_cgroup_threshold_ary
*spare
;
199 struct mem_cgroup_eventfd_list
{
200 struct list_head list
;
201 struct eventfd_ctx
*eventfd
;
204 static void mem_cgroup_threshold(struct mem_cgroup
*mem
);
205 static void mem_cgroup_oom_notify(struct mem_cgroup
*mem
);
208 * The memory controller data structure. The memory controller controls both
209 * page cache and RSS per cgroup. We would eventually like to provide
210 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
211 * to help the administrator determine what knobs to tune.
213 * TODO: Add a water mark for the memory controller. Reclaim will begin when
214 * we hit the water mark. May be even add a low water mark, such that
215 * no reclaim occurs from a cgroup at it's low water mark, this is
216 * a feature that will be implemented much later in the future.
219 struct cgroup_subsys_state css
;
221 * the counter to account for memory usage
223 struct res_counter res
;
225 * the counter to account for mem+swap usage.
227 struct res_counter memsw
;
229 * Per cgroup active and inactive list, similar to the
230 * per zone LRU lists.
232 struct mem_cgroup_lru_info info
;
234 * While reclaiming in a hierarchy, we cache the last child we
237 int last_scanned_child
;
238 int last_scanned_node
;
240 nodemask_t scan_nodes
;
241 atomic_t numainfo_events
;
242 atomic_t numainfo_updating
;
245 * Should the accounting and control be hierarchical, per subtree?
255 /* OOM-Killer disable */
256 int oom_kill_disable
;
258 /* set when res.limit == memsw.limit */
259 bool memsw_is_minimum
;
261 /* protect arrays of thresholds */
262 struct mutex thresholds_lock
;
264 /* thresholds for memory usage. RCU-protected */
265 struct mem_cgroup_thresholds thresholds
;
267 /* thresholds for mem+swap usage. RCU-protected */
268 struct mem_cgroup_thresholds memsw_thresholds
;
270 /* For oom notifier event fd */
271 struct list_head oom_notify
;
274 * Should we move charges of a task when a task is moved into this
275 * mem_cgroup ? And what type of charges should we move ?
277 unsigned long move_charge_at_immigrate
;
281 struct mem_cgroup_stat_cpu
*stat
;
283 * used when a cpu is offlined or other synchronizations
284 * See mem_cgroup_read_stat().
286 struct mem_cgroup_stat_cpu nocpu_base
;
287 spinlock_t pcp_counter_lock
;
290 /* Stuffs for move charges at task migration. */
292 * Types of charges to be moved. "move_charge_at_immitgrate" is treated as a
293 * left-shifted bitmap of these types.
296 MOVE_CHARGE_TYPE_ANON
, /* private anonymous page and swap of it */
297 MOVE_CHARGE_TYPE_FILE
, /* file page(including tmpfs) and swap of it */
301 /* "mc" and its members are protected by cgroup_mutex */
302 static struct move_charge_struct
{
303 spinlock_t lock
; /* for from, to */
304 struct mem_cgroup
*from
;
305 struct mem_cgroup
*to
;
306 unsigned long precharge
;
307 unsigned long moved_charge
;
308 unsigned long moved_swap
;
309 struct task_struct
*moving_task
; /* a task moving charges */
310 wait_queue_head_t waitq
; /* a waitq for other context */
312 .lock
= __SPIN_LOCK_UNLOCKED(mc
.lock
),
313 .waitq
= __WAIT_QUEUE_HEAD_INITIALIZER(mc
.waitq
),
316 static bool move_anon(void)
318 return test_bit(MOVE_CHARGE_TYPE_ANON
,
319 &mc
.to
->move_charge_at_immigrate
);
322 static bool move_file(void)
324 return test_bit(MOVE_CHARGE_TYPE_FILE
,
325 &mc
.to
->move_charge_at_immigrate
);
329 * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft
330 * limit reclaim to prevent infinite loops, if they ever occur.
332 #define MEM_CGROUP_MAX_RECLAIM_LOOPS (100)
333 #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS (2)
336 MEM_CGROUP_CHARGE_TYPE_CACHE
= 0,
337 MEM_CGROUP_CHARGE_TYPE_MAPPED
,
338 MEM_CGROUP_CHARGE_TYPE_SHMEM
, /* used by page migration of shmem */
339 MEM_CGROUP_CHARGE_TYPE_FORCE
, /* used by force_empty */
340 MEM_CGROUP_CHARGE_TYPE_SWAPOUT
, /* for accounting swapcache */
341 MEM_CGROUP_CHARGE_TYPE_DROP
, /* a page was unused swap cache */
345 /* for encoding cft->private value on file */
348 #define _OOM_TYPE (2)
349 #define MEMFILE_PRIVATE(x, val) (((x) << 16) | (val))
350 #define MEMFILE_TYPE(val) (((val) >> 16) & 0xffff)
351 #define MEMFILE_ATTR(val) ((val) & 0xffff)
352 /* Used for OOM nofiier */
353 #define OOM_CONTROL (0)
356 * Reclaim flags for mem_cgroup_hierarchical_reclaim
358 #define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0
359 #define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT)
360 #define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1
361 #define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT)
362 #define MEM_CGROUP_RECLAIM_SOFT_BIT 0x2
363 #define MEM_CGROUP_RECLAIM_SOFT (1 << MEM_CGROUP_RECLAIM_SOFT_BIT)
365 static void mem_cgroup_get(struct mem_cgroup
*mem
);
366 static void mem_cgroup_put(struct mem_cgroup
*mem
);
367 static struct mem_cgroup
*parent_mem_cgroup(struct mem_cgroup
*mem
);
368 static void drain_all_stock_async(struct mem_cgroup
*mem
);
370 static struct mem_cgroup_per_zone
*
371 mem_cgroup_zoneinfo(struct mem_cgroup
*mem
, int nid
, int zid
)
373 return &mem
->info
.nodeinfo
[nid
]->zoneinfo
[zid
];
376 struct cgroup_subsys_state
*mem_cgroup_css(struct mem_cgroup
*mem
)
381 static struct mem_cgroup_per_zone
*
382 page_cgroup_zoneinfo(struct mem_cgroup
*mem
, struct page
*page
)
384 int nid
= page_to_nid(page
);
385 int zid
= page_zonenum(page
);
387 return mem_cgroup_zoneinfo(mem
, nid
, zid
);
390 static struct mem_cgroup_tree_per_zone
*
391 soft_limit_tree_node_zone(int nid
, int zid
)
393 return &soft_limit_tree
.rb_tree_per_node
[nid
]->rb_tree_per_zone
[zid
];
396 static struct mem_cgroup_tree_per_zone
*
397 soft_limit_tree_from_page(struct page
*page
)
399 int nid
= page_to_nid(page
);
400 int zid
= page_zonenum(page
);
402 return &soft_limit_tree
.rb_tree_per_node
[nid
]->rb_tree_per_zone
[zid
];
406 __mem_cgroup_insert_exceeded(struct mem_cgroup
*mem
,
407 struct mem_cgroup_per_zone
*mz
,
408 struct mem_cgroup_tree_per_zone
*mctz
,
409 unsigned long long new_usage_in_excess
)
411 struct rb_node
**p
= &mctz
->rb_root
.rb_node
;
412 struct rb_node
*parent
= NULL
;
413 struct mem_cgroup_per_zone
*mz_node
;
418 mz
->usage_in_excess
= new_usage_in_excess
;
419 if (!mz
->usage_in_excess
)
423 mz_node
= rb_entry(parent
, struct mem_cgroup_per_zone
,
425 if (mz
->usage_in_excess
< mz_node
->usage_in_excess
)
428 * We can't avoid mem cgroups that are over their soft
429 * limit by the same amount
431 else if (mz
->usage_in_excess
>= mz_node
->usage_in_excess
)
434 rb_link_node(&mz
->tree_node
, parent
, p
);
435 rb_insert_color(&mz
->tree_node
, &mctz
->rb_root
);
440 __mem_cgroup_remove_exceeded(struct mem_cgroup
*mem
,
441 struct mem_cgroup_per_zone
*mz
,
442 struct mem_cgroup_tree_per_zone
*mctz
)
446 rb_erase(&mz
->tree_node
, &mctz
->rb_root
);
451 mem_cgroup_remove_exceeded(struct mem_cgroup
*mem
,
452 struct mem_cgroup_per_zone
*mz
,
453 struct mem_cgroup_tree_per_zone
*mctz
)
455 spin_lock(&mctz
->lock
);
456 __mem_cgroup_remove_exceeded(mem
, mz
, mctz
);
457 spin_unlock(&mctz
->lock
);
461 static void mem_cgroup_update_tree(struct mem_cgroup
*mem
, struct page
*page
)
463 unsigned long long excess
;
464 struct mem_cgroup_per_zone
*mz
;
465 struct mem_cgroup_tree_per_zone
*mctz
;
466 int nid
= page_to_nid(page
);
467 int zid
= page_zonenum(page
);
468 mctz
= soft_limit_tree_from_page(page
);
471 * Necessary to update all ancestors when hierarchy is used.
472 * because their event counter is not touched.
474 for (; mem
; mem
= parent_mem_cgroup(mem
)) {
475 mz
= mem_cgroup_zoneinfo(mem
, nid
, zid
);
476 excess
= res_counter_soft_limit_excess(&mem
->res
);
478 * We have to update the tree if mz is on RB-tree or
479 * mem is over its softlimit.
481 if (excess
|| mz
->on_tree
) {
482 spin_lock(&mctz
->lock
);
483 /* if on-tree, remove it */
485 __mem_cgroup_remove_exceeded(mem
, mz
, mctz
);
487 * Insert again. mz->usage_in_excess will be updated.
488 * If excess is 0, no tree ops.
490 __mem_cgroup_insert_exceeded(mem
, mz
, mctz
, excess
);
491 spin_unlock(&mctz
->lock
);
496 static void mem_cgroup_remove_from_trees(struct mem_cgroup
*mem
)
499 struct mem_cgroup_per_zone
*mz
;
500 struct mem_cgroup_tree_per_zone
*mctz
;
502 for_each_node_state(node
, N_POSSIBLE
) {
503 for (zone
= 0; zone
< MAX_NR_ZONES
; zone
++) {
504 mz
= mem_cgroup_zoneinfo(mem
, node
, zone
);
505 mctz
= soft_limit_tree_node_zone(node
, zone
);
506 mem_cgroup_remove_exceeded(mem
, mz
, mctz
);
511 static struct mem_cgroup_per_zone
*
512 __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone
*mctz
)
514 struct rb_node
*rightmost
= NULL
;
515 struct mem_cgroup_per_zone
*mz
;
519 rightmost
= rb_last(&mctz
->rb_root
);
521 goto done
; /* Nothing to reclaim from */
523 mz
= rb_entry(rightmost
, struct mem_cgroup_per_zone
, tree_node
);
525 * Remove the node now but someone else can add it back,
526 * we will to add it back at the end of reclaim to its correct
527 * position in the tree.
529 __mem_cgroup_remove_exceeded(mz
->mem
, mz
, mctz
);
530 if (!res_counter_soft_limit_excess(&mz
->mem
->res
) ||
531 !css_tryget(&mz
->mem
->css
))
537 static struct mem_cgroup_per_zone
*
538 mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone
*mctz
)
540 struct mem_cgroup_per_zone
*mz
;
542 spin_lock(&mctz
->lock
);
543 mz
= __mem_cgroup_largest_soft_limit_node(mctz
);
544 spin_unlock(&mctz
->lock
);
549 * Implementation Note: reading percpu statistics for memcg.
551 * Both of vmstat[] and percpu_counter has threshold and do periodic
552 * synchronization to implement "quick" read. There are trade-off between
553 * reading cost and precision of value. Then, we may have a chance to implement
554 * a periodic synchronizion of counter in memcg's counter.
556 * But this _read() function is used for user interface now. The user accounts
557 * memory usage by memory cgroup and he _always_ requires exact value because
558 * he accounts memory. Even if we provide quick-and-fuzzy read, we always
559 * have to visit all online cpus and make sum. So, for now, unnecessary
560 * synchronization is not implemented. (just implemented for cpu hotplug)
562 * If there are kernel internal actions which can make use of some not-exact
563 * value, and reading all cpu value can be performance bottleneck in some
564 * common workload, threashold and synchonization as vmstat[] should be
567 static long mem_cgroup_read_stat(struct mem_cgroup
*mem
,
568 enum mem_cgroup_stat_index idx
)
574 for_each_online_cpu(cpu
)
575 val
+= per_cpu(mem
->stat
->count
[idx
], cpu
);
576 #ifdef CONFIG_HOTPLUG_CPU
577 spin_lock(&mem
->pcp_counter_lock
);
578 val
+= mem
->nocpu_base
.count
[idx
];
579 spin_unlock(&mem
->pcp_counter_lock
);
585 static void mem_cgroup_swap_statistics(struct mem_cgroup
*mem
,
588 int val
= (charge
) ? 1 : -1;
589 this_cpu_add(mem
->stat
->count
[MEM_CGROUP_STAT_SWAPOUT
], val
);
592 void mem_cgroup_pgfault(struct mem_cgroup
*mem
, int val
)
594 this_cpu_add(mem
->stat
->events
[MEM_CGROUP_EVENTS_PGFAULT
], val
);
597 void mem_cgroup_pgmajfault(struct mem_cgroup
*mem
, int val
)
599 this_cpu_add(mem
->stat
->events
[MEM_CGROUP_EVENTS_PGMAJFAULT
], val
);
602 static unsigned long mem_cgroup_read_events(struct mem_cgroup
*mem
,
603 enum mem_cgroup_events_index idx
)
605 unsigned long val
= 0;
608 for_each_online_cpu(cpu
)
609 val
+= per_cpu(mem
->stat
->events
[idx
], cpu
);
610 #ifdef CONFIG_HOTPLUG_CPU
611 spin_lock(&mem
->pcp_counter_lock
);
612 val
+= mem
->nocpu_base
.events
[idx
];
613 spin_unlock(&mem
->pcp_counter_lock
);
618 static void mem_cgroup_charge_statistics(struct mem_cgroup
*mem
,
619 bool file
, int nr_pages
)
624 __this_cpu_add(mem
->stat
->count
[MEM_CGROUP_STAT_CACHE
], nr_pages
);
626 __this_cpu_add(mem
->stat
->count
[MEM_CGROUP_STAT_RSS
], nr_pages
);
628 /* pagein of a big page is an event. So, ignore page size */
630 __this_cpu_inc(mem
->stat
->events
[MEM_CGROUP_EVENTS_PGPGIN
]);
632 __this_cpu_inc(mem
->stat
->events
[MEM_CGROUP_EVENTS_PGPGOUT
]);
633 nr_pages
= -nr_pages
; /* for event */
636 __this_cpu_add(mem
->stat
->events
[MEM_CGROUP_EVENTS_COUNT
], nr_pages
);
642 mem_cgroup_zone_nr_lru_pages(struct mem_cgroup
*mem
, int nid
, int zid
,
643 unsigned int lru_mask
)
645 struct mem_cgroup_per_zone
*mz
;
647 unsigned long ret
= 0;
649 mz
= mem_cgroup_zoneinfo(mem
, nid
, zid
);
652 if (BIT(l
) & lru_mask
)
653 ret
+= MEM_CGROUP_ZSTAT(mz
, l
);
659 mem_cgroup_node_nr_lru_pages(struct mem_cgroup
*mem
,
660 int nid
, unsigned int lru_mask
)
665 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++)
666 total
+= mem_cgroup_zone_nr_lru_pages(mem
, nid
, zid
, lru_mask
);
671 static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup
*mem
,
672 unsigned int lru_mask
)
677 for_each_node_state(nid
, N_HIGH_MEMORY
)
678 total
+= mem_cgroup_node_nr_lru_pages(mem
, nid
, lru_mask
);
682 static bool __memcg_event_check(struct mem_cgroup
*mem
, int target
)
684 unsigned long val
, next
;
686 val
= this_cpu_read(mem
->stat
->events
[MEM_CGROUP_EVENTS_COUNT
]);
687 next
= this_cpu_read(mem
->stat
->targets
[target
]);
688 /* from time_after() in jiffies.h */
689 return ((long)next
- (long)val
< 0);
692 static void __mem_cgroup_target_update(struct mem_cgroup
*mem
, int target
)
694 unsigned long val
, next
;
696 val
= this_cpu_read(mem
->stat
->events
[MEM_CGROUP_EVENTS_COUNT
]);
699 case MEM_CGROUP_TARGET_THRESH
:
700 next
= val
+ THRESHOLDS_EVENTS_TARGET
;
702 case MEM_CGROUP_TARGET_SOFTLIMIT
:
703 next
= val
+ SOFTLIMIT_EVENTS_TARGET
;
705 case MEM_CGROUP_TARGET_NUMAINFO
:
706 next
= val
+ NUMAINFO_EVENTS_TARGET
;
712 this_cpu_write(mem
->stat
->targets
[target
], next
);
716 * Check events in order.
719 static void memcg_check_events(struct mem_cgroup
*mem
, struct page
*page
)
721 /* threshold event is triggered in finer grain than soft limit */
722 if (unlikely(__memcg_event_check(mem
, MEM_CGROUP_TARGET_THRESH
))) {
723 mem_cgroup_threshold(mem
);
724 __mem_cgroup_target_update(mem
, MEM_CGROUP_TARGET_THRESH
);
725 if (unlikely(__memcg_event_check(mem
,
726 MEM_CGROUP_TARGET_SOFTLIMIT
))) {
727 mem_cgroup_update_tree(mem
, page
);
728 __mem_cgroup_target_update(mem
,
729 MEM_CGROUP_TARGET_SOFTLIMIT
);
732 if (unlikely(__memcg_event_check(mem
,
733 MEM_CGROUP_TARGET_NUMAINFO
))) {
734 atomic_inc(&mem
->numainfo_events
);
735 __mem_cgroup_target_update(mem
,
736 MEM_CGROUP_TARGET_NUMAINFO
);
742 static struct mem_cgroup
*mem_cgroup_from_cont(struct cgroup
*cont
)
744 return container_of(cgroup_subsys_state(cont
,
745 mem_cgroup_subsys_id
), struct mem_cgroup
,
749 struct mem_cgroup
*mem_cgroup_from_task(struct task_struct
*p
)
752 * mm_update_next_owner() may clear mm->owner to NULL
753 * if it races with swapoff, page migration, etc.
754 * So this can be called with p == NULL.
759 return container_of(task_subsys_state(p
, mem_cgroup_subsys_id
),
760 struct mem_cgroup
, css
);
763 struct mem_cgroup
*try_get_mem_cgroup_from_mm(struct mm_struct
*mm
)
765 struct mem_cgroup
*mem
= NULL
;
770 * Because we have no locks, mm->owner's may be being moved to other
771 * cgroup. We use css_tryget() here even if this looks
772 * pessimistic (rather than adding locks here).
776 mem
= mem_cgroup_from_task(rcu_dereference(mm
->owner
));
779 } while (!css_tryget(&mem
->css
));
784 /* The caller has to guarantee "mem" exists before calling this */
785 static struct mem_cgroup
*mem_cgroup_start_loop(struct mem_cgroup
*mem
)
787 struct cgroup_subsys_state
*css
;
790 if (!mem
) /* ROOT cgroup has the smallest ID */
791 return root_mem_cgroup
; /*css_put/get against root is ignored*/
792 if (!mem
->use_hierarchy
) {
793 if (css_tryget(&mem
->css
))
799 * searching a memory cgroup which has the smallest ID under given
800 * ROOT cgroup. (ID >= 1)
802 css
= css_get_next(&mem_cgroup_subsys
, 1, &mem
->css
, &found
);
803 if (css
&& css_tryget(css
))
804 mem
= container_of(css
, struct mem_cgroup
, css
);
811 static struct mem_cgroup
*mem_cgroup_get_next(struct mem_cgroup
*iter
,
812 struct mem_cgroup
*root
,
815 int nextid
= css_id(&iter
->css
) + 1;
818 struct cgroup_subsys_state
*css
;
820 hierarchy_used
= iter
->use_hierarchy
;
823 /* If no ROOT, walk all, ignore hierarchy */
824 if (!cond
|| (root
&& !hierarchy_used
))
828 root
= root_mem_cgroup
;
834 css
= css_get_next(&mem_cgroup_subsys
, nextid
,
836 if (css
&& css_tryget(css
))
837 iter
= container_of(css
, struct mem_cgroup
, css
);
839 /* If css is NULL, no more cgroups will be found */
841 } while (css
&& !iter
);
846 * for_eacn_mem_cgroup_tree() for visiting all cgroup under tree. Please
847 * be careful that "break" loop is not allowed. We have reference count.
848 * Instead of that modify "cond" to be false and "continue" to exit the loop.
850 #define for_each_mem_cgroup_tree_cond(iter, root, cond) \
851 for (iter = mem_cgroup_start_loop(root);\
853 iter = mem_cgroup_get_next(iter, root, cond))
855 #define for_each_mem_cgroup_tree(iter, root) \
856 for_each_mem_cgroup_tree_cond(iter, root, true)
858 #define for_each_mem_cgroup_all(iter) \
859 for_each_mem_cgroup_tree_cond(iter, NULL, true)
862 static inline bool mem_cgroup_is_root(struct mem_cgroup
*mem
)
864 return (mem
== root_mem_cgroup
);
867 void mem_cgroup_count_vm_event(struct mm_struct
*mm
, enum vm_event_item idx
)
869 struct mem_cgroup
*mem
;
875 mem
= mem_cgroup_from_task(rcu_dereference(mm
->owner
));
881 mem_cgroup_pgmajfault(mem
, 1);
884 mem_cgroup_pgfault(mem
, 1);
892 EXPORT_SYMBOL(mem_cgroup_count_vm_event
);
895 * Following LRU functions are allowed to be used without PCG_LOCK.
896 * Operations are called by routine of global LRU independently from memcg.
897 * What we have to take care of here is validness of pc->mem_cgroup.
899 * Changes to pc->mem_cgroup happens when
902 * In typical case, "charge" is done before add-to-lru. Exception is SwapCache.
903 * It is added to LRU before charge.
904 * If PCG_USED bit is not set, page_cgroup is not added to this private LRU.
905 * When moving account, the page is not on LRU. It's isolated.
908 void mem_cgroup_del_lru_list(struct page
*page
, enum lru_list lru
)
910 struct page_cgroup
*pc
;
911 struct mem_cgroup_per_zone
*mz
;
913 if (mem_cgroup_disabled())
915 pc
= lookup_page_cgroup(page
);
916 /* can happen while we handle swapcache. */
917 if (!TestClearPageCgroupAcctLRU(pc
))
919 VM_BUG_ON(!pc
->mem_cgroup
);
921 * We don't check PCG_USED bit. It's cleared when the "page" is finally
922 * removed from global LRU.
924 mz
= page_cgroup_zoneinfo(pc
->mem_cgroup
, page
);
925 /* huge page split is done under lru_lock. so, we have no races. */
926 MEM_CGROUP_ZSTAT(mz
, lru
) -= 1 << compound_order(page
);
927 if (mem_cgroup_is_root(pc
->mem_cgroup
))
929 VM_BUG_ON(list_empty(&pc
->lru
));
930 list_del_init(&pc
->lru
);
933 void mem_cgroup_del_lru(struct page
*page
)
935 mem_cgroup_del_lru_list(page
, page_lru(page
));
939 * Writeback is about to end against a page which has been marked for immediate
940 * reclaim. If it still appears to be reclaimable, move it to the tail of the
943 void mem_cgroup_rotate_reclaimable_page(struct page
*page
)
945 struct mem_cgroup_per_zone
*mz
;
946 struct page_cgroup
*pc
;
947 enum lru_list lru
= page_lru(page
);
949 if (mem_cgroup_disabled())
952 pc
= lookup_page_cgroup(page
);
953 /* unused or root page is not rotated. */
954 if (!PageCgroupUsed(pc
))
956 /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */
958 if (mem_cgroup_is_root(pc
->mem_cgroup
))
960 mz
= page_cgroup_zoneinfo(pc
->mem_cgroup
, page
);
961 list_move_tail(&pc
->lru
, &mz
->lists
[lru
]);
964 void mem_cgroup_rotate_lru_list(struct page
*page
, enum lru_list lru
)
966 struct mem_cgroup_per_zone
*mz
;
967 struct page_cgroup
*pc
;
969 if (mem_cgroup_disabled())
972 pc
= lookup_page_cgroup(page
);
973 /* unused or root page is not rotated. */
974 if (!PageCgroupUsed(pc
))
976 /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */
978 if (mem_cgroup_is_root(pc
->mem_cgroup
))
980 mz
= page_cgroup_zoneinfo(pc
->mem_cgroup
, page
);
981 list_move(&pc
->lru
, &mz
->lists
[lru
]);
984 void mem_cgroup_add_lru_list(struct page
*page
, enum lru_list lru
)
986 struct page_cgroup
*pc
;
987 struct mem_cgroup_per_zone
*mz
;
989 if (mem_cgroup_disabled())
991 pc
= lookup_page_cgroup(page
);
992 VM_BUG_ON(PageCgroupAcctLRU(pc
));
993 if (!PageCgroupUsed(pc
))
995 /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */
997 mz
= page_cgroup_zoneinfo(pc
->mem_cgroup
, page
);
998 /* huge page split is done under lru_lock. so, we have no races. */
999 MEM_CGROUP_ZSTAT(mz
, lru
) += 1 << compound_order(page
);
1000 SetPageCgroupAcctLRU(pc
);
1001 if (mem_cgroup_is_root(pc
->mem_cgroup
))
1003 list_add(&pc
->lru
, &mz
->lists
[lru
]);
1007 * At handling SwapCache and other FUSE stuff, pc->mem_cgroup may be changed
1008 * while it's linked to lru because the page may be reused after it's fully
1009 * uncharged. To handle that, unlink page_cgroup from LRU when charge it again.
1010 * It's done under lock_page and expected that zone->lru_lock isnever held.
1012 static void mem_cgroup_lru_del_before_commit(struct page
*page
)
1014 unsigned long flags
;
1015 struct zone
*zone
= page_zone(page
);
1016 struct page_cgroup
*pc
= lookup_page_cgroup(page
);
1019 * Doing this check without taking ->lru_lock seems wrong but this
1020 * is safe. Because if page_cgroup's USED bit is unset, the page
1021 * will not be added to any memcg's LRU. If page_cgroup's USED bit is
1022 * set, the commit after this will fail, anyway.
1023 * This all charge/uncharge is done under some mutual execustion.
1024 * So, we don't need to taking care of changes in USED bit.
1026 if (likely(!PageLRU(page
)))
1029 spin_lock_irqsave(&zone
->lru_lock
, flags
);
1031 * Forget old LRU when this page_cgroup is *not* used. This Used bit
1032 * is guarded by lock_page() because the page is SwapCache.
1034 if (!PageCgroupUsed(pc
))
1035 mem_cgroup_del_lru_list(page
, page_lru(page
));
1036 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
1039 static void mem_cgroup_lru_add_after_commit(struct page
*page
)
1041 unsigned long flags
;
1042 struct zone
*zone
= page_zone(page
);
1043 struct page_cgroup
*pc
= lookup_page_cgroup(page
);
1045 /* taking care of that the page is added to LRU while we commit it */
1046 if (likely(!PageLRU(page
)))
1048 spin_lock_irqsave(&zone
->lru_lock
, flags
);
1049 /* link when the page is linked to LRU but page_cgroup isn't */
1050 if (PageLRU(page
) && !PageCgroupAcctLRU(pc
))
1051 mem_cgroup_add_lru_list(page
, page_lru(page
));
1052 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
1056 void mem_cgroup_move_lists(struct page
*page
,
1057 enum lru_list from
, enum lru_list to
)
1059 if (mem_cgroup_disabled())
1061 mem_cgroup_del_lru_list(page
, from
);
1062 mem_cgroup_add_lru_list(page
, to
);
1066 * Checks whether given mem is same or in the root_mem's
1069 static bool mem_cgroup_same_or_subtree(const struct mem_cgroup
*root_mem
,
1070 struct mem_cgroup
*mem
)
1072 if (root_mem
!= mem
) {
1073 return (root_mem
->use_hierarchy
&&
1074 css_is_ancestor(&mem
->css
, &root_mem
->css
));
1080 int task_in_mem_cgroup(struct task_struct
*task
, const struct mem_cgroup
*mem
)
1083 struct mem_cgroup
*curr
= NULL
;
1084 struct task_struct
*p
;
1086 p
= find_lock_task_mm(task
);
1089 curr
= try_get_mem_cgroup_from_mm(p
->mm
);
1094 * We should check use_hierarchy of "mem" not "curr". Because checking
1095 * use_hierarchy of "curr" here make this function true if hierarchy is
1096 * enabled in "curr" and "curr" is a child of "mem" in *cgroup*
1097 * hierarchy(even if use_hierarchy is disabled in "mem").
1099 ret
= mem_cgroup_same_or_subtree(mem
, curr
);
1100 css_put(&curr
->css
);
1104 static int calc_inactive_ratio(struct mem_cgroup
*memcg
, unsigned long *present_pages
)
1106 unsigned long active
;
1107 unsigned long inactive
;
1109 unsigned long inactive_ratio
;
1111 inactive
= mem_cgroup_nr_lru_pages(memcg
, BIT(LRU_INACTIVE_ANON
));
1112 active
= mem_cgroup_nr_lru_pages(memcg
, BIT(LRU_ACTIVE_ANON
));
1114 gb
= (inactive
+ active
) >> (30 - PAGE_SHIFT
);
1116 inactive_ratio
= int_sqrt(10 * gb
);
1120 if (present_pages
) {
1121 present_pages
[0] = inactive
;
1122 present_pages
[1] = active
;
1125 return inactive_ratio
;
1128 int mem_cgroup_inactive_anon_is_low(struct mem_cgroup
*memcg
)
1130 unsigned long active
;
1131 unsigned long inactive
;
1132 unsigned long present_pages
[2];
1133 unsigned long inactive_ratio
;
1135 inactive_ratio
= calc_inactive_ratio(memcg
, present_pages
);
1137 inactive
= present_pages
[0];
1138 active
= present_pages
[1];
1140 if (inactive
* inactive_ratio
< active
)
1146 int mem_cgroup_inactive_file_is_low(struct mem_cgroup
*memcg
)
1148 unsigned long active
;
1149 unsigned long inactive
;
1151 inactive
= mem_cgroup_nr_lru_pages(memcg
, BIT(LRU_INACTIVE_FILE
));
1152 active
= mem_cgroup_nr_lru_pages(memcg
, BIT(LRU_ACTIVE_FILE
));
1154 return (active
> inactive
);
1157 struct zone_reclaim_stat
*mem_cgroup_get_reclaim_stat(struct mem_cgroup
*memcg
,
1160 int nid
= zone_to_nid(zone
);
1161 int zid
= zone_idx(zone
);
1162 struct mem_cgroup_per_zone
*mz
= mem_cgroup_zoneinfo(memcg
, nid
, zid
);
1164 return &mz
->reclaim_stat
;
1167 struct zone_reclaim_stat
*
1168 mem_cgroup_get_reclaim_stat_from_page(struct page
*page
)
1170 struct page_cgroup
*pc
;
1171 struct mem_cgroup_per_zone
*mz
;
1173 if (mem_cgroup_disabled())
1176 pc
= lookup_page_cgroup(page
);
1177 if (!PageCgroupUsed(pc
))
1179 /* Ensure pc->mem_cgroup is visible after reading PCG_USED. */
1181 mz
= page_cgroup_zoneinfo(pc
->mem_cgroup
, page
);
1182 return &mz
->reclaim_stat
;
1185 unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan
,
1186 struct list_head
*dst
,
1187 unsigned long *scanned
, int order
,
1188 isolate_mode_t mode
,
1190 struct mem_cgroup
*mem_cont
,
1191 int active
, int file
)
1193 unsigned long nr_taken
= 0;
1197 struct list_head
*src
;
1198 struct page_cgroup
*pc
, *tmp
;
1199 int nid
= zone_to_nid(z
);
1200 int zid
= zone_idx(z
);
1201 struct mem_cgroup_per_zone
*mz
;
1202 int lru
= LRU_FILE
* file
+ active
;
1206 mz
= mem_cgroup_zoneinfo(mem_cont
, nid
, zid
);
1207 src
= &mz
->lists
[lru
];
1210 list_for_each_entry_safe_reverse(pc
, tmp
, src
, lru
) {
1211 if (scan
>= nr_to_scan
)
1214 if (unlikely(!PageCgroupUsed(pc
)))
1217 page
= lookup_cgroup_page(pc
);
1219 if (unlikely(!PageLRU(page
)))
1223 ret
= __isolate_lru_page(page
, mode
, file
);
1226 list_move(&page
->lru
, dst
);
1227 mem_cgroup_del_lru(page
);
1228 nr_taken
+= hpage_nr_pages(page
);
1231 /* we don't affect global LRU but rotate in our LRU */
1232 mem_cgroup_rotate_lru_list(page
, page_lru(page
));
1241 trace_mm_vmscan_memcg_isolate(0, nr_to_scan
, scan
, nr_taken
,
1247 #define mem_cgroup_from_res_counter(counter, member) \
1248 container_of(counter, struct mem_cgroup, member)
1251 * mem_cgroup_margin - calculate chargeable space of a memory cgroup
1252 * @mem: the memory cgroup
1254 * Returns the maximum amount of memory @mem can be charged with, in
1257 static unsigned long mem_cgroup_margin(struct mem_cgroup
*mem
)
1259 unsigned long long margin
;
1261 margin
= res_counter_margin(&mem
->res
);
1262 if (do_swap_account
)
1263 margin
= min(margin
, res_counter_margin(&mem
->memsw
));
1264 return margin
>> PAGE_SHIFT
;
1267 int mem_cgroup_swappiness(struct mem_cgroup
*memcg
)
1269 struct cgroup
*cgrp
= memcg
->css
.cgroup
;
1272 if (cgrp
->parent
== NULL
)
1273 return vm_swappiness
;
1275 return memcg
->swappiness
;
1278 static void mem_cgroup_start_move(struct mem_cgroup
*mem
)
1283 spin_lock(&mem
->pcp_counter_lock
);
1284 for_each_online_cpu(cpu
)
1285 per_cpu(mem
->stat
->count
[MEM_CGROUP_ON_MOVE
], cpu
) += 1;
1286 mem
->nocpu_base
.count
[MEM_CGROUP_ON_MOVE
] += 1;
1287 spin_unlock(&mem
->pcp_counter_lock
);
1293 static void mem_cgroup_end_move(struct mem_cgroup
*mem
)
1300 spin_lock(&mem
->pcp_counter_lock
);
1301 for_each_online_cpu(cpu
)
1302 per_cpu(mem
->stat
->count
[MEM_CGROUP_ON_MOVE
], cpu
) -= 1;
1303 mem
->nocpu_base
.count
[MEM_CGROUP_ON_MOVE
] -= 1;
1304 spin_unlock(&mem
->pcp_counter_lock
);
1308 * 2 routines for checking "mem" is under move_account() or not.
1310 * mem_cgroup_stealed() - checking a cgroup is mc.from or not. This is used
1311 * for avoiding race in accounting. If true,
1312 * pc->mem_cgroup may be overwritten.
1314 * mem_cgroup_under_move() - checking a cgroup is mc.from or mc.to or
1315 * under hierarchy of moving cgroups. This is for
1316 * waiting at hith-memory prressure caused by "move".
1319 static bool mem_cgroup_stealed(struct mem_cgroup
*mem
)
1321 VM_BUG_ON(!rcu_read_lock_held());
1322 return this_cpu_read(mem
->stat
->count
[MEM_CGROUP_ON_MOVE
]) > 0;
1325 static bool mem_cgroup_under_move(struct mem_cgroup
*mem
)
1327 struct mem_cgroup
*from
;
1328 struct mem_cgroup
*to
;
1331 * Unlike task_move routines, we access mc.to, mc.from not under
1332 * mutual exclusion by cgroup_mutex. Here, we take spinlock instead.
1334 spin_lock(&mc
.lock
);
1340 ret
= mem_cgroup_same_or_subtree(mem
, from
)
1341 || mem_cgroup_same_or_subtree(mem
, to
);
1343 spin_unlock(&mc
.lock
);
1347 static bool mem_cgroup_wait_acct_move(struct mem_cgroup
*mem
)
1349 if (mc
.moving_task
&& current
!= mc
.moving_task
) {
1350 if (mem_cgroup_under_move(mem
)) {
1352 prepare_to_wait(&mc
.waitq
, &wait
, TASK_INTERRUPTIBLE
);
1353 /* moving charge context might have finished. */
1356 finish_wait(&mc
.waitq
, &wait
);
1364 * mem_cgroup_print_oom_info: Called from OOM with tasklist_lock held in read mode.
1365 * @memcg: The memory cgroup that went over limit
1366 * @p: Task that is going to be killed
1368 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1371 void mem_cgroup_print_oom_info(struct mem_cgroup
*memcg
, struct task_struct
*p
)
1373 struct cgroup
*task_cgrp
;
1374 struct cgroup
*mem_cgrp
;
1376 * Need a buffer in BSS, can't rely on allocations. The code relies
1377 * on the assumption that OOM is serialized for memory controller.
1378 * If this assumption is broken, revisit this code.
1380 static char memcg_name
[PATH_MAX
];
1389 mem_cgrp
= memcg
->css
.cgroup
;
1390 task_cgrp
= task_cgroup(p
, mem_cgroup_subsys_id
);
1392 ret
= cgroup_path(task_cgrp
, memcg_name
, PATH_MAX
);
1395 * Unfortunately, we are unable to convert to a useful name
1396 * But we'll still print out the usage information
1403 printk(KERN_INFO
"Task in %s killed", memcg_name
);
1406 ret
= cgroup_path(mem_cgrp
, memcg_name
, PATH_MAX
);
1414 * Continues from above, so we don't need an KERN_ level
1416 printk(KERN_CONT
" as a result of limit of %s\n", memcg_name
);
1419 printk(KERN_INFO
"memory: usage %llukB, limit %llukB, failcnt %llu\n",
1420 res_counter_read_u64(&memcg
->res
, RES_USAGE
) >> 10,
1421 res_counter_read_u64(&memcg
->res
, RES_LIMIT
) >> 10,
1422 res_counter_read_u64(&memcg
->res
, RES_FAILCNT
));
1423 printk(KERN_INFO
"memory+swap: usage %llukB, limit %llukB, "
1425 res_counter_read_u64(&memcg
->memsw
, RES_USAGE
) >> 10,
1426 res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
) >> 10,
1427 res_counter_read_u64(&memcg
->memsw
, RES_FAILCNT
));
1431 * This function returns the number of memcg under hierarchy tree. Returns
1432 * 1(self count) if no children.
1434 static int mem_cgroup_count_children(struct mem_cgroup
*mem
)
1437 struct mem_cgroup
*iter
;
1439 for_each_mem_cgroup_tree(iter
, mem
)
1445 * Return the memory (and swap, if configured) limit for a memcg.
1447 u64
mem_cgroup_get_limit(struct mem_cgroup
*memcg
)
1452 limit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
1453 limit
+= total_swap_pages
<< PAGE_SHIFT
;
1455 memsw
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
1457 * If memsw is finite and limits the amount of swap space available
1458 * to this memcg, return that limit.
1460 return min(limit
, memsw
);
1464 * Visit the first child (need not be the first child as per the ordering
1465 * of the cgroup list, since we track last_scanned_child) of @mem and use
1466 * that to reclaim free pages from.
1468 static struct mem_cgroup
*
1469 mem_cgroup_select_victim(struct mem_cgroup
*root_mem
)
1471 struct mem_cgroup
*ret
= NULL
;
1472 struct cgroup_subsys_state
*css
;
1475 if (!root_mem
->use_hierarchy
) {
1476 css_get(&root_mem
->css
);
1482 nextid
= root_mem
->last_scanned_child
+ 1;
1483 css
= css_get_next(&mem_cgroup_subsys
, nextid
, &root_mem
->css
,
1485 if (css
&& css_tryget(css
))
1486 ret
= container_of(css
, struct mem_cgroup
, css
);
1489 /* Updates scanning parameter */
1491 /* this means start scan from ID:1 */
1492 root_mem
->last_scanned_child
= 0;
1494 root_mem
->last_scanned_child
= found
;
1501 * test_mem_cgroup_node_reclaimable
1502 * @mem: the target memcg
1503 * @nid: the node ID to be checked.
1504 * @noswap : specify true here if the user wants flle only information.
1506 * This function returns whether the specified memcg contains any
1507 * reclaimable pages on a node. Returns true if there are any reclaimable
1508 * pages in the node.
1510 static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup
*mem
,
1511 int nid
, bool noswap
)
1513 if (mem_cgroup_node_nr_lru_pages(mem
, nid
, LRU_ALL_FILE
))
1515 if (noswap
|| !total_swap_pages
)
1517 if (mem_cgroup_node_nr_lru_pages(mem
, nid
, LRU_ALL_ANON
))
1522 #if MAX_NUMNODES > 1
1525 * Always updating the nodemask is not very good - even if we have an empty
1526 * list or the wrong list here, we can start from some node and traverse all
1527 * nodes based on the zonelist. So update the list loosely once per 10 secs.
1530 static void mem_cgroup_may_update_nodemask(struct mem_cgroup
*mem
)
1534 * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET
1535 * pagein/pageout changes since the last update.
1537 if (!atomic_read(&mem
->numainfo_events
))
1539 if (atomic_inc_return(&mem
->numainfo_updating
) > 1)
1542 /* make a nodemask where this memcg uses memory from */
1543 mem
->scan_nodes
= node_states
[N_HIGH_MEMORY
];
1545 for_each_node_mask(nid
, node_states
[N_HIGH_MEMORY
]) {
1547 if (!test_mem_cgroup_node_reclaimable(mem
, nid
, false))
1548 node_clear(nid
, mem
->scan_nodes
);
1551 atomic_set(&mem
->numainfo_events
, 0);
1552 atomic_set(&mem
->numainfo_updating
, 0);
1556 * Selecting a node where we start reclaim from. Because what we need is just
1557 * reducing usage counter, start from anywhere is O,K. Considering
1558 * memory reclaim from current node, there are pros. and cons.
1560 * Freeing memory from current node means freeing memory from a node which
1561 * we'll use or we've used. So, it may make LRU bad. And if several threads
1562 * hit limits, it will see a contention on a node. But freeing from remote
1563 * node means more costs for memory reclaim because of memory latency.
1565 * Now, we use round-robin. Better algorithm is welcomed.
1567 int mem_cgroup_select_victim_node(struct mem_cgroup
*mem
)
1571 mem_cgroup_may_update_nodemask(mem
);
1572 node
= mem
->last_scanned_node
;
1574 node
= next_node(node
, mem
->scan_nodes
);
1575 if (node
== MAX_NUMNODES
)
1576 node
= first_node(mem
->scan_nodes
);
1578 * We call this when we hit limit, not when pages are added to LRU.
1579 * No LRU may hold pages because all pages are UNEVICTABLE or
1580 * memcg is too small and all pages are not on LRU. In that case,
1581 * we use curret node.
1583 if (unlikely(node
== MAX_NUMNODES
))
1584 node
= numa_node_id();
1586 mem
->last_scanned_node
= node
;
1591 * Check all nodes whether it contains reclaimable pages or not.
1592 * For quick scan, we make use of scan_nodes. This will allow us to skip
1593 * unused nodes. But scan_nodes is lazily updated and may not cotain
1594 * enough new information. We need to do double check.
1596 bool mem_cgroup_reclaimable(struct mem_cgroup
*mem
, bool noswap
)
1601 * quick check...making use of scan_node.
1602 * We can skip unused nodes.
1604 if (!nodes_empty(mem
->scan_nodes
)) {
1605 for (nid
= first_node(mem
->scan_nodes
);
1607 nid
= next_node(nid
, mem
->scan_nodes
)) {
1609 if (test_mem_cgroup_node_reclaimable(mem
, nid
, noswap
))
1614 * Check rest of nodes.
1616 for_each_node_state(nid
, N_HIGH_MEMORY
) {
1617 if (node_isset(nid
, mem
->scan_nodes
))
1619 if (test_mem_cgroup_node_reclaimable(mem
, nid
, noswap
))
1626 int mem_cgroup_select_victim_node(struct mem_cgroup
*mem
)
1631 bool mem_cgroup_reclaimable(struct mem_cgroup
*mem
, bool noswap
)
1633 return test_mem_cgroup_node_reclaimable(mem
, 0, noswap
);
1638 * Scan the hierarchy if needed to reclaim memory. We remember the last child
1639 * we reclaimed from, so that we don't end up penalizing one child extensively
1640 * based on its position in the children list.
1642 * root_mem is the original ancestor that we've been reclaim from.
1644 * We give up and return to the caller when we visit root_mem twice.
1645 * (other groups can be removed while we're walking....)
1647 * If shrink==true, for avoiding to free too much, this returns immedieately.
1649 static int mem_cgroup_hierarchical_reclaim(struct mem_cgroup
*root_mem
,
1652 unsigned long reclaim_options
,
1653 unsigned long *total_scanned
)
1655 struct mem_cgroup
*victim
;
1658 bool noswap
= reclaim_options
& MEM_CGROUP_RECLAIM_NOSWAP
;
1659 bool shrink
= reclaim_options
& MEM_CGROUP_RECLAIM_SHRINK
;
1660 bool check_soft
= reclaim_options
& MEM_CGROUP_RECLAIM_SOFT
;
1661 unsigned long excess
;
1662 unsigned long nr_scanned
;
1664 excess
= res_counter_soft_limit_excess(&root_mem
->res
) >> PAGE_SHIFT
;
1666 /* If memsw_is_minimum==1, swap-out is of-no-use. */
1667 if (!check_soft
&& !shrink
&& root_mem
->memsw_is_minimum
)
1671 victim
= mem_cgroup_select_victim(root_mem
);
1672 if (victim
== root_mem
) {
1675 * We are not draining per cpu cached charges during
1676 * soft limit reclaim because global reclaim doesn't
1677 * care about charges. It tries to free some memory and
1678 * charges will not give any.
1680 if (!check_soft
&& loop
>= 1)
1681 drain_all_stock_async(root_mem
);
1684 * If we have not been able to reclaim
1685 * anything, it might because there are
1686 * no reclaimable pages under this hierarchy
1688 if (!check_soft
|| !total
) {
1689 css_put(&victim
->css
);
1693 * We want to do more targeted reclaim.
1694 * excess >> 2 is not to excessive so as to
1695 * reclaim too much, nor too less that we keep
1696 * coming back to reclaim from this cgroup
1698 if (total
>= (excess
>> 2) ||
1699 (loop
> MEM_CGROUP_MAX_RECLAIM_LOOPS
)) {
1700 css_put(&victim
->css
);
1705 if (!mem_cgroup_reclaimable(victim
, noswap
)) {
1706 /* this cgroup's local usage == 0 */
1707 css_put(&victim
->css
);
1710 /* we use swappiness of local cgroup */
1712 ret
= mem_cgroup_shrink_node_zone(victim
, gfp_mask
,
1713 noswap
, zone
, &nr_scanned
);
1714 *total_scanned
+= nr_scanned
;
1716 ret
= try_to_free_mem_cgroup_pages(victim
, gfp_mask
,
1718 css_put(&victim
->css
);
1720 * At shrinking usage, we can't check we should stop here or
1721 * reclaim more. It's depends on callers. last_scanned_child
1722 * will work enough for keeping fairness under tree.
1728 if (!res_counter_soft_limit_excess(&root_mem
->res
))
1730 } else if (mem_cgroup_margin(root_mem
))
1737 * Check OOM-Killer is already running under our hierarchy.
1738 * If someone is running, return false.
1739 * Has to be called with memcg_oom_lock
1741 static bool mem_cgroup_oom_lock(struct mem_cgroup
*mem
)
1743 struct mem_cgroup
*iter
, *failed
= NULL
;
1746 for_each_mem_cgroup_tree_cond(iter
, mem
, cond
) {
1747 if (iter
->oom_lock
) {
1749 * this subtree of our hierarchy is already locked
1750 * so we cannot give a lock.
1755 iter
->oom_lock
= true;
1762 * OK, we failed to lock the whole subtree so we have to clean up
1763 * what we set up to the failing subtree
1766 for_each_mem_cgroup_tree_cond(iter
, mem
, cond
) {
1767 if (iter
== failed
) {
1771 iter
->oom_lock
= false;
1777 * Has to be called with memcg_oom_lock
1779 static int mem_cgroup_oom_unlock(struct mem_cgroup
*mem
)
1781 struct mem_cgroup
*iter
;
1783 for_each_mem_cgroup_tree(iter
, mem
)
1784 iter
->oom_lock
= false;
1788 static void mem_cgroup_mark_under_oom(struct mem_cgroup
*mem
)
1790 struct mem_cgroup
*iter
;
1792 for_each_mem_cgroup_tree(iter
, mem
)
1793 atomic_inc(&iter
->under_oom
);
1796 static void mem_cgroup_unmark_under_oom(struct mem_cgroup
*mem
)
1798 struct mem_cgroup
*iter
;
1801 * When a new child is created while the hierarchy is under oom,
1802 * mem_cgroup_oom_lock() may not be called. We have to use
1803 * atomic_add_unless() here.
1805 for_each_mem_cgroup_tree(iter
, mem
)
1806 atomic_add_unless(&iter
->under_oom
, -1, 0);
1809 static DEFINE_SPINLOCK(memcg_oom_lock
);
1810 static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq
);
1812 struct oom_wait_info
{
1813 struct mem_cgroup
*mem
;
1817 static int memcg_oom_wake_function(wait_queue_t
*wait
,
1818 unsigned mode
, int sync
, void *arg
)
1820 struct mem_cgroup
*wake_mem
= (struct mem_cgroup
*)arg
,
1822 struct oom_wait_info
*oom_wait_info
;
1824 oom_wait_info
= container_of(wait
, struct oom_wait_info
, wait
);
1825 oom_wait_mem
= oom_wait_info
->mem
;
1828 * Both of oom_wait_info->mem and wake_mem are stable under us.
1829 * Then we can use css_is_ancestor without taking care of RCU.
1831 if (!mem_cgroup_same_or_subtree(oom_wait_mem
, wake_mem
)
1832 && !mem_cgroup_same_or_subtree(wake_mem
, oom_wait_mem
))
1834 return autoremove_wake_function(wait
, mode
, sync
, arg
);
1837 static void memcg_wakeup_oom(struct mem_cgroup
*mem
)
1839 /* for filtering, pass "mem" as argument. */
1840 __wake_up(&memcg_oom_waitq
, TASK_NORMAL
, 0, mem
);
1843 static void memcg_oom_recover(struct mem_cgroup
*mem
)
1845 if (mem
&& atomic_read(&mem
->under_oom
))
1846 memcg_wakeup_oom(mem
);
1850 * try to call OOM killer. returns false if we should exit memory-reclaim loop.
1852 bool mem_cgroup_handle_oom(struct mem_cgroup
*mem
, gfp_t mask
)
1854 struct oom_wait_info owait
;
1855 bool locked
, need_to_kill
;
1858 owait
.wait
.flags
= 0;
1859 owait
.wait
.func
= memcg_oom_wake_function
;
1860 owait
.wait
.private = current
;
1861 INIT_LIST_HEAD(&owait
.wait
.task_list
);
1862 need_to_kill
= true;
1863 mem_cgroup_mark_under_oom(mem
);
1865 /* At first, try to OOM lock hierarchy under mem.*/
1866 spin_lock(&memcg_oom_lock
);
1867 locked
= mem_cgroup_oom_lock(mem
);
1869 * Even if signal_pending(), we can't quit charge() loop without
1870 * accounting. So, UNINTERRUPTIBLE is appropriate. But SIGKILL
1871 * under OOM is always welcomed, use TASK_KILLABLE here.
1873 prepare_to_wait(&memcg_oom_waitq
, &owait
.wait
, TASK_KILLABLE
);
1874 if (!locked
|| mem
->oom_kill_disable
)
1875 need_to_kill
= false;
1877 mem_cgroup_oom_notify(mem
);
1878 spin_unlock(&memcg_oom_lock
);
1881 finish_wait(&memcg_oom_waitq
, &owait
.wait
);
1882 mem_cgroup_out_of_memory(mem
, mask
);
1885 finish_wait(&memcg_oom_waitq
, &owait
.wait
);
1887 spin_lock(&memcg_oom_lock
);
1889 mem_cgroup_oom_unlock(mem
);
1890 memcg_wakeup_oom(mem
);
1891 spin_unlock(&memcg_oom_lock
);
1893 mem_cgroup_unmark_under_oom(mem
);
1895 if (test_thread_flag(TIF_MEMDIE
) || fatal_signal_pending(current
))
1897 /* Give chance to dying process */
1898 schedule_timeout(1);
1903 * Currently used to update mapped file statistics, but the routine can be
1904 * generalized to update other statistics as well.
1906 * Notes: Race condition
1908 * We usually use page_cgroup_lock() for accessing page_cgroup member but
1909 * it tends to be costly. But considering some conditions, we doesn't need
1910 * to do so _always_.
1912 * Considering "charge", lock_page_cgroup() is not required because all
1913 * file-stat operations happen after a page is attached to radix-tree. There
1914 * are no race with "charge".
1916 * Considering "uncharge", we know that memcg doesn't clear pc->mem_cgroup
1917 * at "uncharge" intentionally. So, we always see valid pc->mem_cgroup even
1918 * if there are race with "uncharge". Statistics itself is properly handled
1921 * Considering "move", this is an only case we see a race. To make the race
1922 * small, we check MEM_CGROUP_ON_MOVE percpu value and detect there are
1923 * possibility of race condition. If there is, we take a lock.
1926 void mem_cgroup_update_page_stat(struct page
*page
,
1927 enum mem_cgroup_page_stat_item idx
, int val
)
1929 struct mem_cgroup
*mem
;
1930 struct page_cgroup
*pc
= lookup_page_cgroup(page
);
1931 bool need_unlock
= false;
1932 unsigned long uninitialized_var(flags
);
1938 mem
= pc
->mem_cgroup
;
1939 if (unlikely(!mem
|| !PageCgroupUsed(pc
)))
1941 /* pc->mem_cgroup is unstable ? */
1942 if (unlikely(mem_cgroup_stealed(mem
)) || PageTransHuge(page
)) {
1943 /* take a lock against to access pc->mem_cgroup */
1944 move_lock_page_cgroup(pc
, &flags
);
1946 mem
= pc
->mem_cgroup
;
1947 if (!mem
|| !PageCgroupUsed(pc
))
1952 case MEMCG_NR_FILE_MAPPED
:
1954 SetPageCgroupFileMapped(pc
);
1955 else if (!page_mapped(page
))
1956 ClearPageCgroupFileMapped(pc
);
1957 idx
= MEM_CGROUP_STAT_FILE_MAPPED
;
1963 this_cpu_add(mem
->stat
->count
[idx
], val
);
1966 if (unlikely(need_unlock
))
1967 move_unlock_page_cgroup(pc
, &flags
);
1971 EXPORT_SYMBOL(mem_cgroup_update_page_stat
);
1974 * size of first charge trial. "32" comes from vmscan.c's magic value.
1975 * TODO: maybe necessary to use big numbers in big irons.
1977 #define CHARGE_BATCH 32U
1978 struct memcg_stock_pcp
{
1979 struct mem_cgroup
*cached
; /* this never be root cgroup */
1980 unsigned int nr_pages
;
1981 struct work_struct work
;
1982 unsigned long flags
;
1983 #define FLUSHING_CACHED_CHARGE (0)
1985 static DEFINE_PER_CPU(struct memcg_stock_pcp
, memcg_stock
);
1986 static DEFINE_MUTEX(percpu_charge_mutex
);
1989 * Try to consume stocked charge on this cpu. If success, one page is consumed
1990 * from local stock and true is returned. If the stock is 0 or charges from a
1991 * cgroup which is not current target, returns false. This stock will be
1994 static bool consume_stock(struct mem_cgroup
*mem
)
1996 struct memcg_stock_pcp
*stock
;
1999 stock
= &get_cpu_var(memcg_stock
);
2000 if (mem
== stock
->cached
&& stock
->nr_pages
)
2002 else /* need to call res_counter_charge */
2004 put_cpu_var(memcg_stock
);
2009 * Returns stocks cached in percpu to res_counter and reset cached information.
2011 static void drain_stock(struct memcg_stock_pcp
*stock
)
2013 struct mem_cgroup
*old
= stock
->cached
;
2015 if (stock
->nr_pages
) {
2016 unsigned long bytes
= stock
->nr_pages
* PAGE_SIZE
;
2018 res_counter_uncharge(&old
->res
, bytes
);
2019 if (do_swap_account
)
2020 res_counter_uncharge(&old
->memsw
, bytes
);
2021 stock
->nr_pages
= 0;
2023 stock
->cached
= NULL
;
2027 * This must be called under preempt disabled or must be called by
2028 * a thread which is pinned to local cpu.
2030 static void drain_local_stock(struct work_struct
*dummy
)
2032 struct memcg_stock_pcp
*stock
= &__get_cpu_var(memcg_stock
);
2034 clear_bit(FLUSHING_CACHED_CHARGE
, &stock
->flags
);
2038 * Cache charges(val) which is from res_counter, to local per_cpu area.
2039 * This will be consumed by consume_stock() function, later.
2041 static void refill_stock(struct mem_cgroup
*mem
, unsigned int nr_pages
)
2043 struct memcg_stock_pcp
*stock
= &get_cpu_var(memcg_stock
);
2045 if (stock
->cached
!= mem
) { /* reset if necessary */
2047 stock
->cached
= mem
;
2049 stock
->nr_pages
+= nr_pages
;
2050 put_cpu_var(memcg_stock
);
2054 * Drains all per-CPU charge caches for given root_mem resp. subtree
2055 * of the hierarchy under it. sync flag says whether we should block
2056 * until the work is done.
2058 static void drain_all_stock(struct mem_cgroup
*root_mem
, bool sync
)
2062 /* Notify other cpus that system-wide "drain" is running */
2065 for_each_online_cpu(cpu
) {
2066 struct memcg_stock_pcp
*stock
= &per_cpu(memcg_stock
, cpu
);
2067 struct mem_cgroup
*mem
;
2069 mem
= stock
->cached
;
2070 if (!mem
|| !stock
->nr_pages
)
2072 if (!mem_cgroup_same_or_subtree(root_mem
, mem
))
2074 if (!test_and_set_bit(FLUSHING_CACHED_CHARGE
, &stock
->flags
)) {
2076 drain_local_stock(&stock
->work
);
2078 schedule_work_on(cpu
, &stock
->work
);
2086 for_each_online_cpu(cpu
) {
2087 struct memcg_stock_pcp
*stock
= &per_cpu(memcg_stock
, cpu
);
2088 if (test_bit(FLUSHING_CACHED_CHARGE
, &stock
->flags
))
2089 flush_work(&stock
->work
);
2096 * Tries to drain stocked charges in other cpus. This function is asynchronous
2097 * and just put a work per cpu for draining localy on each cpu. Caller can
2098 * expects some charges will be back to res_counter later but cannot wait for
2101 static void drain_all_stock_async(struct mem_cgroup
*root_mem
)
2104 * If someone calls draining, avoid adding more kworker runs.
2106 if (!mutex_trylock(&percpu_charge_mutex
))
2108 drain_all_stock(root_mem
, false);
2109 mutex_unlock(&percpu_charge_mutex
);
2112 /* This is a synchronous drain interface. */
2113 static void drain_all_stock_sync(struct mem_cgroup
*root_mem
)
2115 /* called when force_empty is called */
2116 mutex_lock(&percpu_charge_mutex
);
2117 drain_all_stock(root_mem
, true);
2118 mutex_unlock(&percpu_charge_mutex
);
2122 * This function drains percpu counter value from DEAD cpu and
2123 * move it to local cpu. Note that this function can be preempted.
2125 static void mem_cgroup_drain_pcp_counter(struct mem_cgroup
*mem
, int cpu
)
2129 spin_lock(&mem
->pcp_counter_lock
);
2130 for (i
= 0; i
< MEM_CGROUP_STAT_DATA
; i
++) {
2131 long x
= per_cpu(mem
->stat
->count
[i
], cpu
);
2133 per_cpu(mem
->stat
->count
[i
], cpu
) = 0;
2134 mem
->nocpu_base
.count
[i
] += x
;
2136 for (i
= 0; i
< MEM_CGROUP_EVENTS_NSTATS
; i
++) {
2137 unsigned long x
= per_cpu(mem
->stat
->events
[i
], cpu
);
2139 per_cpu(mem
->stat
->events
[i
], cpu
) = 0;
2140 mem
->nocpu_base
.events
[i
] += x
;
2142 /* need to clear ON_MOVE value, works as a kind of lock. */
2143 per_cpu(mem
->stat
->count
[MEM_CGROUP_ON_MOVE
], cpu
) = 0;
2144 spin_unlock(&mem
->pcp_counter_lock
);
2147 static void synchronize_mem_cgroup_on_move(struct mem_cgroup
*mem
, int cpu
)
2149 int idx
= MEM_CGROUP_ON_MOVE
;
2151 spin_lock(&mem
->pcp_counter_lock
);
2152 per_cpu(mem
->stat
->count
[idx
], cpu
) = mem
->nocpu_base
.count
[idx
];
2153 spin_unlock(&mem
->pcp_counter_lock
);
2156 static int __cpuinit
memcg_cpu_hotplug_callback(struct notifier_block
*nb
,
2157 unsigned long action
,
2160 int cpu
= (unsigned long)hcpu
;
2161 struct memcg_stock_pcp
*stock
;
2162 struct mem_cgroup
*iter
;
2164 if ((action
== CPU_ONLINE
)) {
2165 for_each_mem_cgroup_all(iter
)
2166 synchronize_mem_cgroup_on_move(iter
, cpu
);
2170 if ((action
!= CPU_DEAD
) || action
!= CPU_DEAD_FROZEN
)
2173 for_each_mem_cgroup_all(iter
)
2174 mem_cgroup_drain_pcp_counter(iter
, cpu
);
2176 stock
= &per_cpu(memcg_stock
, cpu
);
2182 /* See __mem_cgroup_try_charge() for details */
2184 CHARGE_OK
, /* success */
2185 CHARGE_RETRY
, /* need to retry but retry is not bad */
2186 CHARGE_NOMEM
, /* we can't do more. return -ENOMEM */
2187 CHARGE_WOULDBLOCK
, /* GFP_WAIT wasn't set and no enough res. */
2188 CHARGE_OOM_DIE
, /* the current is killed because of OOM */
2191 static int mem_cgroup_do_charge(struct mem_cgroup
*mem
, gfp_t gfp_mask
,
2192 unsigned int nr_pages
, bool oom_check
)
2194 unsigned long csize
= nr_pages
* PAGE_SIZE
;
2195 struct mem_cgroup
*mem_over_limit
;
2196 struct res_counter
*fail_res
;
2197 unsigned long flags
= 0;
2200 ret
= res_counter_charge(&mem
->res
, csize
, &fail_res
);
2203 if (!do_swap_account
)
2205 ret
= res_counter_charge(&mem
->memsw
, csize
, &fail_res
);
2209 res_counter_uncharge(&mem
->res
, csize
);
2210 mem_over_limit
= mem_cgroup_from_res_counter(fail_res
, memsw
);
2211 flags
|= MEM_CGROUP_RECLAIM_NOSWAP
;
2213 mem_over_limit
= mem_cgroup_from_res_counter(fail_res
, res
);
2215 * nr_pages can be either a huge page (HPAGE_PMD_NR), a batch
2216 * of regular pages (CHARGE_BATCH), or a single regular page (1).
2218 * Never reclaim on behalf of optional batching, retry with a
2219 * single page instead.
2221 if (nr_pages
== CHARGE_BATCH
)
2222 return CHARGE_RETRY
;
2224 if (!(gfp_mask
& __GFP_WAIT
))
2225 return CHARGE_WOULDBLOCK
;
2227 ret
= mem_cgroup_hierarchical_reclaim(mem_over_limit
, NULL
,
2228 gfp_mask
, flags
, NULL
);
2229 if (mem_cgroup_margin(mem_over_limit
) >= nr_pages
)
2230 return CHARGE_RETRY
;
2232 * Even though the limit is exceeded at this point, reclaim
2233 * may have been able to free some pages. Retry the charge
2234 * before killing the task.
2236 * Only for regular pages, though: huge pages are rather
2237 * unlikely to succeed so close to the limit, and we fall back
2238 * to regular pages anyway in case of failure.
2240 if (nr_pages
== 1 && ret
)
2241 return CHARGE_RETRY
;
2244 * At task move, charge accounts can be doubly counted. So, it's
2245 * better to wait until the end of task_move if something is going on.
2247 if (mem_cgroup_wait_acct_move(mem_over_limit
))
2248 return CHARGE_RETRY
;
2250 /* If we don't need to call oom-killer at el, return immediately */
2252 return CHARGE_NOMEM
;
2254 if (!mem_cgroup_handle_oom(mem_over_limit
, gfp_mask
))
2255 return CHARGE_OOM_DIE
;
2257 return CHARGE_RETRY
;
2261 * Unlike exported interface, "oom" parameter is added. if oom==true,
2262 * oom-killer can be invoked.
2264 static int __mem_cgroup_try_charge(struct mm_struct
*mm
,
2266 unsigned int nr_pages
,
2267 struct mem_cgroup
**memcg
,
2270 unsigned int batch
= max(CHARGE_BATCH
, nr_pages
);
2271 int nr_oom_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
2272 struct mem_cgroup
*mem
= NULL
;
2276 * Unlike gloval-vm's OOM-kill, we're not in memory shortage
2277 * in system level. So, allow to go ahead dying process in addition to
2280 if (unlikely(test_thread_flag(TIF_MEMDIE
)
2281 || fatal_signal_pending(current
)))
2285 * We always charge the cgroup the mm_struct belongs to.
2286 * The mm_struct's mem_cgroup changes on task migration if the
2287 * thread group leader migrates. It's possible that mm is not
2288 * set, if so charge the init_mm (happens for pagecache usage).
2293 if (*memcg
) { /* css should be a valid one */
2295 VM_BUG_ON(css_is_removed(&mem
->css
));
2296 if (mem_cgroup_is_root(mem
))
2298 if (nr_pages
== 1 && consume_stock(mem
))
2302 struct task_struct
*p
;
2305 p
= rcu_dereference(mm
->owner
);
2307 * Because we don't have task_lock(), "p" can exit.
2308 * In that case, "mem" can point to root or p can be NULL with
2309 * race with swapoff. Then, we have small risk of mis-accouning.
2310 * But such kind of mis-account by race always happens because
2311 * we don't have cgroup_mutex(). It's overkill and we allo that
2313 * (*) swapoff at el will charge against mm-struct not against
2314 * task-struct. So, mm->owner can be NULL.
2316 mem
= mem_cgroup_from_task(p
);
2317 if (!mem
|| mem_cgroup_is_root(mem
)) {
2321 if (nr_pages
== 1 && consume_stock(mem
)) {
2323 * It seems dagerous to access memcg without css_get().
2324 * But considering how consume_stok works, it's not
2325 * necessary. If consume_stock success, some charges
2326 * from this memcg are cached on this cpu. So, we
2327 * don't need to call css_get()/css_tryget() before
2328 * calling consume_stock().
2333 /* after here, we may be blocked. we need to get refcnt */
2334 if (!css_tryget(&mem
->css
)) {
2344 /* If killed, bypass charge */
2345 if (fatal_signal_pending(current
)) {
2351 if (oom
&& !nr_oom_retries
) {
2353 nr_oom_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
2356 ret
= mem_cgroup_do_charge(mem
, gfp_mask
, batch
, oom_check
);
2360 case CHARGE_RETRY
: /* not in OOM situation but retry */
2365 case CHARGE_WOULDBLOCK
: /* !__GFP_WAIT */
2368 case CHARGE_NOMEM
: /* OOM routine works */
2373 /* If oom, we never return -ENOMEM */
2376 case CHARGE_OOM_DIE
: /* Killed by OOM Killer */
2380 } while (ret
!= CHARGE_OK
);
2382 if (batch
> nr_pages
)
2383 refill_stock(mem
, batch
- nr_pages
);
2397 * Somemtimes we have to undo a charge we got by try_charge().
2398 * This function is for that and do uncharge, put css's refcnt.
2399 * gotten by try_charge().
2401 static void __mem_cgroup_cancel_charge(struct mem_cgroup
*mem
,
2402 unsigned int nr_pages
)
2404 if (!mem_cgroup_is_root(mem
)) {
2405 unsigned long bytes
= nr_pages
* PAGE_SIZE
;
2407 res_counter_uncharge(&mem
->res
, bytes
);
2408 if (do_swap_account
)
2409 res_counter_uncharge(&mem
->memsw
, bytes
);
2414 * A helper function to get mem_cgroup from ID. must be called under
2415 * rcu_read_lock(). The caller must check css_is_removed() or some if
2416 * it's concern. (dropping refcnt from swap can be called against removed
2419 static struct mem_cgroup
*mem_cgroup_lookup(unsigned short id
)
2421 struct cgroup_subsys_state
*css
;
2423 /* ID 0 is unused ID */
2426 css
= css_lookup(&mem_cgroup_subsys
, id
);
2429 return container_of(css
, struct mem_cgroup
, css
);
2432 struct mem_cgroup
*try_get_mem_cgroup_from_page(struct page
*page
)
2434 struct mem_cgroup
*mem
= NULL
;
2435 struct page_cgroup
*pc
;
2439 VM_BUG_ON(!PageLocked(page
));
2441 pc
= lookup_page_cgroup(page
);
2442 lock_page_cgroup(pc
);
2443 if (PageCgroupUsed(pc
)) {
2444 mem
= pc
->mem_cgroup
;
2445 if (mem
&& !css_tryget(&mem
->css
))
2447 } else if (PageSwapCache(page
)) {
2448 ent
.val
= page_private(page
);
2449 id
= lookup_swap_cgroup(ent
);
2451 mem
= mem_cgroup_lookup(id
);
2452 if (mem
&& !css_tryget(&mem
->css
))
2456 unlock_page_cgroup(pc
);
2460 static void __mem_cgroup_commit_charge(struct mem_cgroup
*mem
,
2462 unsigned int nr_pages
,
2463 struct page_cgroup
*pc
,
2464 enum charge_type ctype
)
2466 lock_page_cgroup(pc
);
2467 if (unlikely(PageCgroupUsed(pc
))) {
2468 unlock_page_cgroup(pc
);
2469 __mem_cgroup_cancel_charge(mem
, nr_pages
);
2473 * we don't need page_cgroup_lock about tail pages, becase they are not
2474 * accessed by any other context at this point.
2476 pc
->mem_cgroup
= mem
;
2478 * We access a page_cgroup asynchronously without lock_page_cgroup().
2479 * Especially when a page_cgroup is taken from a page, pc->mem_cgroup
2480 * is accessed after testing USED bit. To make pc->mem_cgroup visible
2481 * before USED bit, we need memory barrier here.
2482 * See mem_cgroup_add_lru_list(), etc.
2486 case MEM_CGROUP_CHARGE_TYPE_CACHE
:
2487 case MEM_CGROUP_CHARGE_TYPE_SHMEM
:
2488 SetPageCgroupCache(pc
);
2489 SetPageCgroupUsed(pc
);
2491 case MEM_CGROUP_CHARGE_TYPE_MAPPED
:
2492 ClearPageCgroupCache(pc
);
2493 SetPageCgroupUsed(pc
);
2499 mem_cgroup_charge_statistics(mem
, PageCgroupCache(pc
), nr_pages
);
2500 unlock_page_cgroup(pc
);
2502 * "charge_statistics" updated event counter. Then, check it.
2503 * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree.
2504 * if they exceeds softlimit.
2506 memcg_check_events(mem
, page
);
2509 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
2511 #define PCGF_NOCOPY_AT_SPLIT ((1 << PCG_LOCK) | (1 << PCG_MOVE_LOCK) |\
2512 (1 << PCG_ACCT_LRU) | (1 << PCG_MIGRATION))
2514 * Because tail pages are not marked as "used", set it. We're under
2515 * zone->lru_lock, 'splitting on pmd' and compund_lock.
2517 void mem_cgroup_split_huge_fixup(struct page
*head
, struct page
*tail
)
2519 struct page_cgroup
*head_pc
= lookup_page_cgroup(head
);
2520 struct page_cgroup
*tail_pc
= lookup_page_cgroup(tail
);
2521 unsigned long flags
;
2523 if (mem_cgroup_disabled())
2526 * We have no races with charge/uncharge but will have races with
2527 * page state accounting.
2529 move_lock_page_cgroup(head_pc
, &flags
);
2531 tail_pc
->mem_cgroup
= head_pc
->mem_cgroup
;
2532 smp_wmb(); /* see __commit_charge() */
2533 if (PageCgroupAcctLRU(head_pc
)) {
2535 struct mem_cgroup_per_zone
*mz
;
2538 * LRU flags cannot be copied because we need to add tail
2539 *.page to LRU by generic call and our hook will be called.
2540 * We hold lru_lock, then, reduce counter directly.
2542 lru
= page_lru(head
);
2543 mz
= page_cgroup_zoneinfo(head_pc
->mem_cgroup
, head
);
2544 MEM_CGROUP_ZSTAT(mz
, lru
) -= 1;
2546 tail_pc
->flags
= head_pc
->flags
& ~PCGF_NOCOPY_AT_SPLIT
;
2547 move_unlock_page_cgroup(head_pc
, &flags
);
2552 * mem_cgroup_move_account - move account of the page
2554 * @nr_pages: number of regular pages (>1 for huge pages)
2555 * @pc: page_cgroup of the page.
2556 * @from: mem_cgroup which the page is moved from.
2557 * @to: mem_cgroup which the page is moved to. @from != @to.
2558 * @uncharge: whether we should call uncharge and css_put against @from.
2560 * The caller must confirm following.
2561 * - page is not on LRU (isolate_page() is useful.)
2562 * - compound_lock is held when nr_pages > 1
2564 * This function doesn't do "charge" nor css_get to new cgroup. It should be
2565 * done by a caller(__mem_cgroup_try_charge would be useful). If @uncharge is
2566 * true, this function does "uncharge" from old cgroup, but it doesn't if
2567 * @uncharge is false, so a caller should do "uncharge".
2569 static int mem_cgroup_move_account(struct page
*page
,
2570 unsigned int nr_pages
,
2571 struct page_cgroup
*pc
,
2572 struct mem_cgroup
*from
,
2573 struct mem_cgroup
*to
,
2576 unsigned long flags
;
2579 VM_BUG_ON(from
== to
);
2580 VM_BUG_ON(PageLRU(page
));
2582 * The page is isolated from LRU. So, collapse function
2583 * will not handle this page. But page splitting can happen.
2584 * Do this check under compound_page_lock(). The caller should
2588 if (nr_pages
> 1 && !PageTransHuge(page
))
2591 lock_page_cgroup(pc
);
2594 if (!PageCgroupUsed(pc
) || pc
->mem_cgroup
!= from
)
2597 move_lock_page_cgroup(pc
, &flags
);
2599 if (PageCgroupFileMapped(pc
)) {
2600 /* Update mapped_file data for mem_cgroup */
2602 __this_cpu_dec(from
->stat
->count
[MEM_CGROUP_STAT_FILE_MAPPED
]);
2603 __this_cpu_inc(to
->stat
->count
[MEM_CGROUP_STAT_FILE_MAPPED
]);
2606 mem_cgroup_charge_statistics(from
, PageCgroupCache(pc
), -nr_pages
);
2608 /* This is not "cancel", but cancel_charge does all we need. */
2609 __mem_cgroup_cancel_charge(from
, nr_pages
);
2611 /* caller should have done css_get */
2612 pc
->mem_cgroup
= to
;
2613 mem_cgroup_charge_statistics(to
, PageCgroupCache(pc
), nr_pages
);
2615 * We charges against "to" which may not have any tasks. Then, "to"
2616 * can be under rmdir(). But in current implementation, caller of
2617 * this function is just force_empty() and move charge, so it's
2618 * guaranteed that "to" is never removed. So, we don't check rmdir
2621 move_unlock_page_cgroup(pc
, &flags
);
2624 unlock_page_cgroup(pc
);
2628 memcg_check_events(to
, page
);
2629 memcg_check_events(from
, page
);
2635 * move charges to its parent.
2638 static int mem_cgroup_move_parent(struct page
*page
,
2639 struct page_cgroup
*pc
,
2640 struct mem_cgroup
*child
,
2643 struct cgroup
*cg
= child
->css
.cgroup
;
2644 struct cgroup
*pcg
= cg
->parent
;
2645 struct mem_cgroup
*parent
;
2646 unsigned int nr_pages
;
2647 unsigned long uninitialized_var(flags
);
2655 if (!get_page_unless_zero(page
))
2657 if (isolate_lru_page(page
))
2660 nr_pages
= hpage_nr_pages(page
);
2662 parent
= mem_cgroup_from_cont(pcg
);
2663 ret
= __mem_cgroup_try_charge(NULL
, gfp_mask
, nr_pages
, &parent
, false);
2668 flags
= compound_lock_irqsave(page
);
2670 ret
= mem_cgroup_move_account(page
, nr_pages
, pc
, child
, parent
, true);
2672 __mem_cgroup_cancel_charge(parent
, nr_pages
);
2675 compound_unlock_irqrestore(page
, flags
);
2677 putback_lru_page(page
);
2685 * Charge the memory controller for page usage.
2687 * 0 if the charge was successful
2688 * < 0 if the cgroup is over its limit
2690 static int mem_cgroup_charge_common(struct page
*page
, struct mm_struct
*mm
,
2691 gfp_t gfp_mask
, enum charge_type ctype
)
2693 struct mem_cgroup
*mem
= NULL
;
2694 unsigned int nr_pages
= 1;
2695 struct page_cgroup
*pc
;
2699 if (PageTransHuge(page
)) {
2700 nr_pages
<<= compound_order(page
);
2701 VM_BUG_ON(!PageTransHuge(page
));
2703 * Never OOM-kill a process for a huge page. The
2704 * fault handler will fall back to regular pages.
2709 pc
= lookup_page_cgroup(page
);
2710 BUG_ON(!pc
); /* XXX: remove this and move pc lookup into commit */
2712 ret
= __mem_cgroup_try_charge(mm
, gfp_mask
, nr_pages
, &mem
, oom
);
2716 __mem_cgroup_commit_charge(mem
, page
, nr_pages
, pc
, ctype
);
2720 int mem_cgroup_newpage_charge(struct page
*page
,
2721 struct mm_struct
*mm
, gfp_t gfp_mask
)
2723 if (mem_cgroup_disabled())
2726 * If already mapped, we don't have to account.
2727 * If page cache, page->mapping has address_space.
2728 * But page->mapping may have out-of-use anon_vma pointer,
2729 * detecit it by PageAnon() check. newly-mapped-anon's page->mapping
2732 if (page_mapped(page
) || (page
->mapping
&& !PageAnon(page
)))
2736 return mem_cgroup_charge_common(page
, mm
, gfp_mask
,
2737 MEM_CGROUP_CHARGE_TYPE_MAPPED
);
2741 __mem_cgroup_commit_charge_swapin(struct page
*page
, struct mem_cgroup
*ptr
,
2742 enum charge_type ctype
);
2745 __mem_cgroup_commit_charge_lrucare(struct page
*page
, struct mem_cgroup
*mem
,
2746 enum charge_type ctype
)
2748 struct page_cgroup
*pc
= lookup_page_cgroup(page
);
2750 * In some case, SwapCache, FUSE(splice_buf->radixtree), the page
2751 * is already on LRU. It means the page may on some other page_cgroup's
2752 * LRU. Take care of it.
2754 mem_cgroup_lru_del_before_commit(page
);
2755 __mem_cgroup_commit_charge(mem
, page
, 1, pc
, ctype
);
2756 mem_cgroup_lru_add_after_commit(page
);
2760 int mem_cgroup_cache_charge(struct page
*page
, struct mm_struct
*mm
,
2763 struct mem_cgroup
*mem
= NULL
;
2766 if (mem_cgroup_disabled())
2768 if (PageCompound(page
))
2774 if (page_is_file_cache(page
)) {
2775 ret
= __mem_cgroup_try_charge(mm
, gfp_mask
, 1, &mem
, true);
2780 * FUSE reuses pages without going through the final
2781 * put that would remove them from the LRU list, make
2782 * sure that they get relinked properly.
2784 __mem_cgroup_commit_charge_lrucare(page
, mem
,
2785 MEM_CGROUP_CHARGE_TYPE_CACHE
);
2789 if (PageSwapCache(page
)) {
2790 ret
= mem_cgroup_try_charge_swapin(mm
, page
, gfp_mask
, &mem
);
2792 __mem_cgroup_commit_charge_swapin(page
, mem
,
2793 MEM_CGROUP_CHARGE_TYPE_SHMEM
);
2795 ret
= mem_cgroup_charge_common(page
, mm
, gfp_mask
,
2796 MEM_CGROUP_CHARGE_TYPE_SHMEM
);
2802 * While swap-in, try_charge -> commit or cancel, the page is locked.
2803 * And when try_charge() successfully returns, one refcnt to memcg without
2804 * struct page_cgroup is acquired. This refcnt will be consumed by
2805 * "commit()" or removed by "cancel()"
2807 int mem_cgroup_try_charge_swapin(struct mm_struct
*mm
,
2809 gfp_t mask
, struct mem_cgroup
**ptr
)
2811 struct mem_cgroup
*mem
;
2816 if (mem_cgroup_disabled())
2819 if (!do_swap_account
)
2822 * A racing thread's fault, or swapoff, may have already updated
2823 * the pte, and even removed page from swap cache: in those cases
2824 * do_swap_page()'s pte_same() test will fail; but there's also a
2825 * KSM case which does need to charge the page.
2827 if (!PageSwapCache(page
))
2829 mem
= try_get_mem_cgroup_from_page(page
);
2833 ret
= __mem_cgroup_try_charge(NULL
, mask
, 1, ptr
, true);
2839 return __mem_cgroup_try_charge(mm
, mask
, 1, ptr
, true);
2843 __mem_cgroup_commit_charge_swapin(struct page
*page
, struct mem_cgroup
*ptr
,
2844 enum charge_type ctype
)
2846 if (mem_cgroup_disabled())
2850 cgroup_exclude_rmdir(&ptr
->css
);
2852 __mem_cgroup_commit_charge_lrucare(page
, ptr
, ctype
);
2854 * Now swap is on-memory. This means this page may be
2855 * counted both as mem and swap....double count.
2856 * Fix it by uncharging from memsw. Basically, this SwapCache is stable
2857 * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page()
2858 * may call delete_from_swap_cache() before reach here.
2860 if (do_swap_account
&& PageSwapCache(page
)) {
2861 swp_entry_t ent
= {.val
= page_private(page
)};
2863 struct mem_cgroup
*memcg
;
2865 id
= swap_cgroup_record(ent
, 0);
2867 memcg
= mem_cgroup_lookup(id
);
2870 * This recorded memcg can be obsolete one. So, avoid
2871 * calling css_tryget
2873 if (!mem_cgroup_is_root(memcg
))
2874 res_counter_uncharge(&memcg
->memsw
, PAGE_SIZE
);
2875 mem_cgroup_swap_statistics(memcg
, false);
2876 mem_cgroup_put(memcg
);
2881 * At swapin, we may charge account against cgroup which has no tasks.
2882 * So, rmdir()->pre_destroy() can be called while we do this charge.
2883 * In that case, we need to call pre_destroy() again. check it here.
2885 cgroup_release_and_wakeup_rmdir(&ptr
->css
);
2888 void mem_cgroup_commit_charge_swapin(struct page
*page
, struct mem_cgroup
*ptr
)
2890 __mem_cgroup_commit_charge_swapin(page
, ptr
,
2891 MEM_CGROUP_CHARGE_TYPE_MAPPED
);
2894 void mem_cgroup_cancel_charge_swapin(struct mem_cgroup
*mem
)
2896 if (mem_cgroup_disabled())
2900 __mem_cgroup_cancel_charge(mem
, 1);
2903 static void mem_cgroup_do_uncharge(struct mem_cgroup
*mem
,
2904 unsigned int nr_pages
,
2905 const enum charge_type ctype
)
2907 struct memcg_batch_info
*batch
= NULL
;
2908 bool uncharge_memsw
= true;
2910 /* If swapout, usage of swap doesn't decrease */
2911 if (!do_swap_account
|| ctype
== MEM_CGROUP_CHARGE_TYPE_SWAPOUT
)
2912 uncharge_memsw
= false;
2914 batch
= ¤t
->memcg_batch
;
2916 * In usual, we do css_get() when we remember memcg pointer.
2917 * But in this case, we keep res->usage until end of a series of
2918 * uncharges. Then, it's ok to ignore memcg's refcnt.
2923 * do_batch > 0 when unmapping pages or inode invalidate/truncate.
2924 * In those cases, all pages freed continuously can be expected to be in
2925 * the same cgroup and we have chance to coalesce uncharges.
2926 * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE)
2927 * because we want to do uncharge as soon as possible.
2930 if (!batch
->do_batch
|| test_thread_flag(TIF_MEMDIE
))
2931 goto direct_uncharge
;
2934 goto direct_uncharge
;
2937 * In typical case, batch->memcg == mem. This means we can
2938 * merge a series of uncharges to an uncharge of res_counter.
2939 * If not, we uncharge res_counter ony by one.
2941 if (batch
->memcg
!= mem
)
2942 goto direct_uncharge
;
2943 /* remember freed charge and uncharge it later */
2946 batch
->memsw_nr_pages
++;
2949 res_counter_uncharge(&mem
->res
, nr_pages
* PAGE_SIZE
);
2951 res_counter_uncharge(&mem
->memsw
, nr_pages
* PAGE_SIZE
);
2952 if (unlikely(batch
->memcg
!= mem
))
2953 memcg_oom_recover(mem
);
2958 * uncharge if !page_mapped(page)
2960 static struct mem_cgroup
*
2961 __mem_cgroup_uncharge_common(struct page
*page
, enum charge_type ctype
)
2963 struct mem_cgroup
*mem
= NULL
;
2964 unsigned int nr_pages
= 1;
2965 struct page_cgroup
*pc
;
2967 if (mem_cgroup_disabled())
2970 if (PageSwapCache(page
))
2973 if (PageTransHuge(page
)) {
2974 nr_pages
<<= compound_order(page
);
2975 VM_BUG_ON(!PageTransHuge(page
));
2978 * Check if our page_cgroup is valid
2980 pc
= lookup_page_cgroup(page
);
2981 if (unlikely(!pc
|| !PageCgroupUsed(pc
)))
2984 lock_page_cgroup(pc
);
2986 mem
= pc
->mem_cgroup
;
2988 if (!PageCgroupUsed(pc
))
2992 case MEM_CGROUP_CHARGE_TYPE_MAPPED
:
2993 case MEM_CGROUP_CHARGE_TYPE_DROP
:
2994 /* See mem_cgroup_prepare_migration() */
2995 if (page_mapped(page
) || PageCgroupMigration(pc
))
2998 case MEM_CGROUP_CHARGE_TYPE_SWAPOUT
:
2999 if (!PageAnon(page
)) { /* Shared memory */
3000 if (page
->mapping
&& !page_is_file_cache(page
))
3002 } else if (page_mapped(page
)) /* Anon */
3009 mem_cgroup_charge_statistics(mem
, PageCgroupCache(pc
), -nr_pages
);
3011 ClearPageCgroupUsed(pc
);
3013 * pc->mem_cgroup is not cleared here. It will be accessed when it's
3014 * freed from LRU. This is safe because uncharged page is expected not
3015 * to be reused (freed soon). Exception is SwapCache, it's handled by
3016 * special functions.
3019 unlock_page_cgroup(pc
);
3021 * even after unlock, we have mem->res.usage here and this memcg
3022 * will never be freed.
3024 memcg_check_events(mem
, page
);
3025 if (do_swap_account
&& ctype
== MEM_CGROUP_CHARGE_TYPE_SWAPOUT
) {
3026 mem_cgroup_swap_statistics(mem
, true);
3027 mem_cgroup_get(mem
);
3029 if (!mem_cgroup_is_root(mem
))
3030 mem_cgroup_do_uncharge(mem
, nr_pages
, ctype
);
3035 unlock_page_cgroup(pc
);
3039 void mem_cgroup_uncharge_page(struct page
*page
)
3042 if (page_mapped(page
))
3044 if (page
->mapping
&& !PageAnon(page
))
3046 __mem_cgroup_uncharge_common(page
, MEM_CGROUP_CHARGE_TYPE_MAPPED
);
3049 void mem_cgroup_uncharge_cache_page(struct page
*page
)
3051 VM_BUG_ON(page_mapped(page
));
3052 VM_BUG_ON(page
->mapping
);
3053 __mem_cgroup_uncharge_common(page
, MEM_CGROUP_CHARGE_TYPE_CACHE
);
3057 * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate.
3058 * In that cases, pages are freed continuously and we can expect pages
3059 * are in the same memcg. All these calls itself limits the number of
3060 * pages freed at once, then uncharge_start/end() is called properly.
3061 * This may be called prural(2) times in a context,
3064 void mem_cgroup_uncharge_start(void)
3066 current
->memcg_batch
.do_batch
++;
3067 /* We can do nest. */
3068 if (current
->memcg_batch
.do_batch
== 1) {
3069 current
->memcg_batch
.memcg
= NULL
;
3070 current
->memcg_batch
.nr_pages
= 0;
3071 current
->memcg_batch
.memsw_nr_pages
= 0;
3075 void mem_cgroup_uncharge_end(void)
3077 struct memcg_batch_info
*batch
= ¤t
->memcg_batch
;
3079 if (!batch
->do_batch
)
3083 if (batch
->do_batch
) /* If stacked, do nothing. */
3089 * This "batch->memcg" is valid without any css_get/put etc...
3090 * bacause we hide charges behind us.
3092 if (batch
->nr_pages
)
3093 res_counter_uncharge(&batch
->memcg
->res
,
3094 batch
->nr_pages
* PAGE_SIZE
);
3095 if (batch
->memsw_nr_pages
)
3096 res_counter_uncharge(&batch
->memcg
->memsw
,
3097 batch
->memsw_nr_pages
* PAGE_SIZE
);
3098 memcg_oom_recover(batch
->memcg
);
3099 /* forget this pointer (for sanity check) */
3100 batch
->memcg
= NULL
;
3105 * called after __delete_from_swap_cache() and drop "page" account.
3106 * memcg information is recorded to swap_cgroup of "ent"
3109 mem_cgroup_uncharge_swapcache(struct page
*page
, swp_entry_t ent
, bool swapout
)
3111 struct mem_cgroup
*memcg
;
3112 int ctype
= MEM_CGROUP_CHARGE_TYPE_SWAPOUT
;
3114 if (!swapout
) /* this was a swap cache but the swap is unused ! */
3115 ctype
= MEM_CGROUP_CHARGE_TYPE_DROP
;
3117 memcg
= __mem_cgroup_uncharge_common(page
, ctype
);
3120 * record memcg information, if swapout && memcg != NULL,
3121 * mem_cgroup_get() was called in uncharge().
3123 if (do_swap_account
&& swapout
&& memcg
)
3124 swap_cgroup_record(ent
, css_id(&memcg
->css
));
3128 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
3130 * called from swap_entry_free(). remove record in swap_cgroup and
3131 * uncharge "memsw" account.
3133 void mem_cgroup_uncharge_swap(swp_entry_t ent
)
3135 struct mem_cgroup
*memcg
;
3138 if (!do_swap_account
)
3141 id
= swap_cgroup_record(ent
, 0);
3143 memcg
= mem_cgroup_lookup(id
);
3146 * We uncharge this because swap is freed.
3147 * This memcg can be obsolete one. We avoid calling css_tryget
3149 if (!mem_cgroup_is_root(memcg
))
3150 res_counter_uncharge(&memcg
->memsw
, PAGE_SIZE
);
3151 mem_cgroup_swap_statistics(memcg
, false);
3152 mem_cgroup_put(memcg
);
3158 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
3159 * @entry: swap entry to be moved
3160 * @from: mem_cgroup which the entry is moved from
3161 * @to: mem_cgroup which the entry is moved to
3162 * @need_fixup: whether we should fixup res_counters and refcounts.
3164 * It succeeds only when the swap_cgroup's record for this entry is the same
3165 * as the mem_cgroup's id of @from.
3167 * Returns 0 on success, -EINVAL on failure.
3169 * The caller must have charged to @to, IOW, called res_counter_charge() about
3170 * both res and memsw, and called css_get().
3172 static int mem_cgroup_move_swap_account(swp_entry_t entry
,
3173 struct mem_cgroup
*from
, struct mem_cgroup
*to
, bool need_fixup
)
3175 unsigned short old_id
, new_id
;
3177 old_id
= css_id(&from
->css
);
3178 new_id
= css_id(&to
->css
);
3180 if (swap_cgroup_cmpxchg(entry
, old_id
, new_id
) == old_id
) {
3181 mem_cgroup_swap_statistics(from
, false);
3182 mem_cgroup_swap_statistics(to
, true);
3184 * This function is only called from task migration context now.
3185 * It postpones res_counter and refcount handling till the end
3186 * of task migration(mem_cgroup_clear_mc()) for performance
3187 * improvement. But we cannot postpone mem_cgroup_get(to)
3188 * because if the process that has been moved to @to does
3189 * swap-in, the refcount of @to might be decreased to 0.
3193 if (!mem_cgroup_is_root(from
))
3194 res_counter_uncharge(&from
->memsw
, PAGE_SIZE
);
3195 mem_cgroup_put(from
);
3197 * we charged both to->res and to->memsw, so we should
3200 if (!mem_cgroup_is_root(to
))
3201 res_counter_uncharge(&to
->res
, PAGE_SIZE
);
3208 static inline int mem_cgroup_move_swap_account(swp_entry_t entry
,
3209 struct mem_cgroup
*from
, struct mem_cgroup
*to
, bool need_fixup
)
3216 * Before starting migration, account PAGE_SIZE to mem_cgroup that the old
3219 int mem_cgroup_prepare_migration(struct page
*page
,
3220 struct page
*newpage
, struct mem_cgroup
**ptr
, gfp_t gfp_mask
)
3222 struct mem_cgroup
*mem
= NULL
;
3223 struct page_cgroup
*pc
;
3224 enum charge_type ctype
;
3229 VM_BUG_ON(PageTransHuge(page
));
3230 if (mem_cgroup_disabled())
3233 pc
= lookup_page_cgroup(page
);
3234 lock_page_cgroup(pc
);
3235 if (PageCgroupUsed(pc
)) {
3236 mem
= pc
->mem_cgroup
;
3239 * At migrating an anonymous page, its mapcount goes down
3240 * to 0 and uncharge() will be called. But, even if it's fully
3241 * unmapped, migration may fail and this page has to be
3242 * charged again. We set MIGRATION flag here and delay uncharge
3243 * until end_migration() is called
3245 * Corner Case Thinking
3247 * When the old page was mapped as Anon and it's unmap-and-freed
3248 * while migration was ongoing.
3249 * If unmap finds the old page, uncharge() of it will be delayed
3250 * until end_migration(). If unmap finds a new page, it's
3251 * uncharged when it make mapcount to be 1->0. If unmap code
3252 * finds swap_migration_entry, the new page will not be mapped
3253 * and end_migration() will find it(mapcount==0).
3256 * When the old page was mapped but migraion fails, the kernel
3257 * remaps it. A charge for it is kept by MIGRATION flag even
3258 * if mapcount goes down to 0. We can do remap successfully
3259 * without charging it again.
3262 * The "old" page is under lock_page() until the end of
3263 * migration, so, the old page itself will not be swapped-out.
3264 * If the new page is swapped out before end_migraton, our
3265 * hook to usual swap-out path will catch the event.
3268 SetPageCgroupMigration(pc
);
3270 unlock_page_cgroup(pc
);
3272 * If the page is not charged at this point,
3279 ret
= __mem_cgroup_try_charge(NULL
, gfp_mask
, 1, ptr
, false);
3280 css_put(&mem
->css
);/* drop extra refcnt */
3281 if (ret
|| *ptr
== NULL
) {
3282 if (PageAnon(page
)) {
3283 lock_page_cgroup(pc
);
3284 ClearPageCgroupMigration(pc
);
3285 unlock_page_cgroup(pc
);
3287 * The old page may be fully unmapped while we kept it.
3289 mem_cgroup_uncharge_page(page
);
3294 * We charge new page before it's used/mapped. So, even if unlock_page()
3295 * is called before end_migration, we can catch all events on this new
3296 * page. In the case new page is migrated but not remapped, new page's
3297 * mapcount will be finally 0 and we call uncharge in end_migration().
3299 pc
= lookup_page_cgroup(newpage
);
3301 ctype
= MEM_CGROUP_CHARGE_TYPE_MAPPED
;
3302 else if (page_is_file_cache(page
))
3303 ctype
= MEM_CGROUP_CHARGE_TYPE_CACHE
;
3305 ctype
= MEM_CGROUP_CHARGE_TYPE_SHMEM
;
3306 __mem_cgroup_commit_charge(mem
, page
, 1, pc
, ctype
);
3310 /* remove redundant charge if migration failed*/
3311 void mem_cgroup_end_migration(struct mem_cgroup
*mem
,
3312 struct page
*oldpage
, struct page
*newpage
, bool migration_ok
)
3314 struct page
*used
, *unused
;
3315 struct page_cgroup
*pc
;
3319 /* blocks rmdir() */
3320 cgroup_exclude_rmdir(&mem
->css
);
3321 if (!migration_ok
) {
3329 * We disallowed uncharge of pages under migration because mapcount
3330 * of the page goes down to zero, temporarly.
3331 * Clear the flag and check the page should be charged.
3333 pc
= lookup_page_cgroup(oldpage
);
3334 lock_page_cgroup(pc
);
3335 ClearPageCgroupMigration(pc
);
3336 unlock_page_cgroup(pc
);
3338 __mem_cgroup_uncharge_common(unused
, MEM_CGROUP_CHARGE_TYPE_FORCE
);
3341 * If a page is a file cache, radix-tree replacement is very atomic
3342 * and we can skip this check. When it was an Anon page, its mapcount
3343 * goes down to 0. But because we added MIGRATION flage, it's not
3344 * uncharged yet. There are several case but page->mapcount check
3345 * and USED bit check in mem_cgroup_uncharge_page() will do enough
3346 * check. (see prepare_charge() also)
3349 mem_cgroup_uncharge_page(used
);
3351 * At migration, we may charge account against cgroup which has no
3353 * So, rmdir()->pre_destroy() can be called while we do this charge.
3354 * In that case, we need to call pre_destroy() again. check it here.
3356 cgroup_release_and_wakeup_rmdir(&mem
->css
);
3359 #ifdef CONFIG_DEBUG_VM
3360 static struct page_cgroup
*lookup_page_cgroup_used(struct page
*page
)
3362 struct page_cgroup
*pc
;
3364 pc
= lookup_page_cgroup(page
);
3365 if (likely(pc
) && PageCgroupUsed(pc
))
3370 bool mem_cgroup_bad_page_check(struct page
*page
)
3372 if (mem_cgroup_disabled())
3375 return lookup_page_cgroup_used(page
) != NULL
;
3378 void mem_cgroup_print_bad_page(struct page
*page
)
3380 struct page_cgroup
*pc
;
3382 pc
= lookup_page_cgroup_used(page
);
3387 printk(KERN_ALERT
"pc:%p pc->flags:%lx pc->mem_cgroup:%p",
3388 pc
, pc
->flags
, pc
->mem_cgroup
);
3390 path
= kmalloc(PATH_MAX
, GFP_KERNEL
);
3393 ret
= cgroup_path(pc
->mem_cgroup
->css
.cgroup
,
3398 printk(KERN_CONT
"(%s)\n",
3399 (ret
< 0) ? "cannot get the path" : path
);
3405 static DEFINE_MUTEX(set_limit_mutex
);
3407 static int mem_cgroup_resize_limit(struct mem_cgroup
*memcg
,
3408 unsigned long long val
)
3411 u64 memswlimit
, memlimit
;
3413 int children
= mem_cgroup_count_children(memcg
);
3414 u64 curusage
, oldusage
;
3418 * For keeping hierarchical_reclaim simple, how long we should retry
3419 * is depends on callers. We set our retry-count to be function
3420 * of # of children which we should visit in this loop.
3422 retry_count
= MEM_CGROUP_RECLAIM_RETRIES
* children
;
3424 oldusage
= res_counter_read_u64(&memcg
->res
, RES_USAGE
);
3427 while (retry_count
) {
3428 if (signal_pending(current
)) {
3433 * Rather than hide all in some function, I do this in
3434 * open coded manner. You see what this really does.
3435 * We have to guarantee mem->res.limit < mem->memsw.limit.
3437 mutex_lock(&set_limit_mutex
);
3438 memswlimit
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
3439 if (memswlimit
< val
) {
3441 mutex_unlock(&set_limit_mutex
);
3445 memlimit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
3449 ret
= res_counter_set_limit(&memcg
->res
, val
);
3451 if (memswlimit
== val
)
3452 memcg
->memsw_is_minimum
= true;
3454 memcg
->memsw_is_minimum
= false;
3456 mutex_unlock(&set_limit_mutex
);
3461 mem_cgroup_hierarchical_reclaim(memcg
, NULL
, GFP_KERNEL
,
3462 MEM_CGROUP_RECLAIM_SHRINK
,
3464 curusage
= res_counter_read_u64(&memcg
->res
, RES_USAGE
);
3465 /* Usage is reduced ? */
3466 if (curusage
>= oldusage
)
3469 oldusage
= curusage
;
3471 if (!ret
&& enlarge
)
3472 memcg_oom_recover(memcg
);
3477 static int mem_cgroup_resize_memsw_limit(struct mem_cgroup
*memcg
,
3478 unsigned long long val
)
3481 u64 memlimit
, memswlimit
, oldusage
, curusage
;
3482 int children
= mem_cgroup_count_children(memcg
);
3486 /* see mem_cgroup_resize_res_limit */
3487 retry_count
= children
* MEM_CGROUP_RECLAIM_RETRIES
;
3488 oldusage
= res_counter_read_u64(&memcg
->memsw
, RES_USAGE
);
3489 while (retry_count
) {
3490 if (signal_pending(current
)) {
3495 * Rather than hide all in some function, I do this in
3496 * open coded manner. You see what this really does.
3497 * We have to guarantee mem->res.limit < mem->memsw.limit.
3499 mutex_lock(&set_limit_mutex
);
3500 memlimit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
3501 if (memlimit
> val
) {
3503 mutex_unlock(&set_limit_mutex
);
3506 memswlimit
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
3507 if (memswlimit
< val
)
3509 ret
= res_counter_set_limit(&memcg
->memsw
, val
);
3511 if (memlimit
== val
)
3512 memcg
->memsw_is_minimum
= true;
3514 memcg
->memsw_is_minimum
= false;
3516 mutex_unlock(&set_limit_mutex
);
3521 mem_cgroup_hierarchical_reclaim(memcg
, NULL
, GFP_KERNEL
,
3522 MEM_CGROUP_RECLAIM_NOSWAP
|
3523 MEM_CGROUP_RECLAIM_SHRINK
,
3525 curusage
= res_counter_read_u64(&memcg
->memsw
, RES_USAGE
);
3526 /* Usage is reduced ? */
3527 if (curusage
>= oldusage
)
3530 oldusage
= curusage
;
3532 if (!ret
&& enlarge
)
3533 memcg_oom_recover(memcg
);
3537 unsigned long mem_cgroup_soft_limit_reclaim(struct zone
*zone
, int order
,
3539 unsigned long *total_scanned
)
3541 unsigned long nr_reclaimed
= 0;
3542 struct mem_cgroup_per_zone
*mz
, *next_mz
= NULL
;
3543 unsigned long reclaimed
;
3545 struct mem_cgroup_tree_per_zone
*mctz
;
3546 unsigned long long excess
;
3547 unsigned long nr_scanned
;
3552 mctz
= soft_limit_tree_node_zone(zone_to_nid(zone
), zone_idx(zone
));
3554 * This loop can run a while, specially if mem_cgroup's continuously
3555 * keep exceeding their soft limit and putting the system under
3562 mz
= mem_cgroup_largest_soft_limit_node(mctz
);
3567 reclaimed
= mem_cgroup_hierarchical_reclaim(mz
->mem
, zone
,
3569 MEM_CGROUP_RECLAIM_SOFT
,
3571 nr_reclaimed
+= reclaimed
;
3572 *total_scanned
+= nr_scanned
;
3573 spin_lock(&mctz
->lock
);
3576 * If we failed to reclaim anything from this memory cgroup
3577 * it is time to move on to the next cgroup
3583 * Loop until we find yet another one.
3585 * By the time we get the soft_limit lock
3586 * again, someone might have aded the
3587 * group back on the RB tree. Iterate to
3588 * make sure we get a different mem.
3589 * mem_cgroup_largest_soft_limit_node returns
3590 * NULL if no other cgroup is present on
3594 __mem_cgroup_largest_soft_limit_node(mctz
);
3596 css_put(&next_mz
->mem
->css
);
3597 else /* next_mz == NULL or other memcg */
3601 __mem_cgroup_remove_exceeded(mz
->mem
, mz
, mctz
);
3602 excess
= res_counter_soft_limit_excess(&mz
->mem
->res
);
3604 * One school of thought says that we should not add
3605 * back the node to the tree if reclaim returns 0.
3606 * But our reclaim could return 0, simply because due
3607 * to priority we are exposing a smaller subset of
3608 * memory to reclaim from. Consider this as a longer
3611 /* If excess == 0, no tree ops */
3612 __mem_cgroup_insert_exceeded(mz
->mem
, mz
, mctz
, excess
);
3613 spin_unlock(&mctz
->lock
);
3614 css_put(&mz
->mem
->css
);
3617 * Could not reclaim anything and there are no more
3618 * mem cgroups to try or we seem to be looping without
3619 * reclaiming anything.
3621 if (!nr_reclaimed
&&
3623 loop
> MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS
))
3625 } while (!nr_reclaimed
);
3627 css_put(&next_mz
->mem
->css
);
3628 return nr_reclaimed
;
3632 * This routine traverse page_cgroup in given list and drop them all.
3633 * *And* this routine doesn't reclaim page itself, just removes page_cgroup.
3635 static int mem_cgroup_force_empty_list(struct mem_cgroup
*mem
,
3636 int node
, int zid
, enum lru_list lru
)
3639 struct mem_cgroup_per_zone
*mz
;
3640 struct page_cgroup
*pc
, *busy
;
3641 unsigned long flags
, loop
;
3642 struct list_head
*list
;
3645 zone
= &NODE_DATA(node
)->node_zones
[zid
];
3646 mz
= mem_cgroup_zoneinfo(mem
, node
, zid
);
3647 list
= &mz
->lists
[lru
];
3649 loop
= MEM_CGROUP_ZSTAT(mz
, lru
);
3650 /* give some margin against EBUSY etc...*/
3657 spin_lock_irqsave(&zone
->lru_lock
, flags
);
3658 if (list_empty(list
)) {
3659 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
3662 pc
= list_entry(list
->prev
, struct page_cgroup
, lru
);
3664 list_move(&pc
->lru
, list
);
3666 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
3669 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
3671 page
= lookup_cgroup_page(pc
);
3673 ret
= mem_cgroup_move_parent(page
, pc
, mem
, GFP_KERNEL
);
3677 if (ret
== -EBUSY
|| ret
== -EINVAL
) {
3678 /* found lock contention or "pc" is obsolete. */
3685 if (!ret
&& !list_empty(list
))
3691 * make mem_cgroup's charge to be 0 if there is no task.
3692 * This enables deleting this mem_cgroup.
3694 static int mem_cgroup_force_empty(struct mem_cgroup
*mem
, bool free_all
)
3697 int node
, zid
, shrink
;
3698 int nr_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
3699 struct cgroup
*cgrp
= mem
->css
.cgroup
;
3704 /* should free all ? */
3710 if (cgroup_task_count(cgrp
) || !list_empty(&cgrp
->children
))
3713 if (signal_pending(current
))
3715 /* This is for making all *used* pages to be on LRU. */
3716 lru_add_drain_all();
3717 drain_all_stock_sync(mem
);
3719 mem_cgroup_start_move(mem
);
3720 for_each_node_state(node
, N_HIGH_MEMORY
) {
3721 for (zid
= 0; !ret
&& zid
< MAX_NR_ZONES
; zid
++) {
3724 ret
= mem_cgroup_force_empty_list(mem
,
3733 mem_cgroup_end_move(mem
);
3734 memcg_oom_recover(mem
);
3735 /* it seems parent cgroup doesn't have enough mem */
3739 /* "ret" should also be checked to ensure all lists are empty. */
3740 } while (mem
->res
.usage
> 0 || ret
);
3746 /* returns EBUSY if there is a task or if we come here twice. */
3747 if (cgroup_task_count(cgrp
) || !list_empty(&cgrp
->children
) || shrink
) {
3751 /* we call try-to-free pages for make this cgroup empty */
3752 lru_add_drain_all();
3753 /* try to free all pages in this cgroup */
3755 while (nr_retries
&& mem
->res
.usage
> 0) {
3758 if (signal_pending(current
)) {
3762 progress
= try_to_free_mem_cgroup_pages(mem
, GFP_KERNEL
,
3766 /* maybe some writeback is necessary */
3767 congestion_wait(BLK_RW_ASYNC
, HZ
/10);
3772 /* try move_account...there may be some *locked* pages. */
3776 int mem_cgroup_force_empty_write(struct cgroup
*cont
, unsigned int event
)
3778 return mem_cgroup_force_empty(mem_cgroup_from_cont(cont
), true);
3782 static u64
mem_cgroup_hierarchy_read(struct cgroup
*cont
, struct cftype
*cft
)
3784 return mem_cgroup_from_cont(cont
)->use_hierarchy
;
3787 static int mem_cgroup_hierarchy_write(struct cgroup
*cont
, struct cftype
*cft
,
3791 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cont
);
3792 struct cgroup
*parent
= cont
->parent
;
3793 struct mem_cgroup
*parent_mem
= NULL
;
3796 parent_mem
= mem_cgroup_from_cont(parent
);
3800 * If parent's use_hierarchy is set, we can't make any modifications
3801 * in the child subtrees. If it is unset, then the change can
3802 * occur, provided the current cgroup has no children.
3804 * For the root cgroup, parent_mem is NULL, we allow value to be
3805 * set if there are no children.
3807 if ((!parent_mem
|| !parent_mem
->use_hierarchy
) &&
3808 (val
== 1 || val
== 0)) {
3809 if (list_empty(&cont
->children
))
3810 mem
->use_hierarchy
= val
;
3821 static unsigned long mem_cgroup_recursive_stat(struct mem_cgroup
*mem
,
3822 enum mem_cgroup_stat_index idx
)
3824 struct mem_cgroup
*iter
;
3827 /* Per-cpu values can be negative, use a signed accumulator */
3828 for_each_mem_cgroup_tree(iter
, mem
)
3829 val
+= mem_cgroup_read_stat(iter
, idx
);
3831 if (val
< 0) /* race ? */
3836 static inline u64
mem_cgroup_usage(struct mem_cgroup
*mem
, bool swap
)
3840 if (!mem_cgroup_is_root(mem
)) {
3842 return res_counter_read_u64(&mem
->res
, RES_USAGE
);
3844 return res_counter_read_u64(&mem
->memsw
, RES_USAGE
);
3847 val
= mem_cgroup_recursive_stat(mem
, MEM_CGROUP_STAT_CACHE
);
3848 val
+= mem_cgroup_recursive_stat(mem
, MEM_CGROUP_STAT_RSS
);
3851 val
+= mem_cgroup_recursive_stat(mem
, MEM_CGROUP_STAT_SWAPOUT
);
3853 return val
<< PAGE_SHIFT
;
3856 static u64
mem_cgroup_read(struct cgroup
*cont
, struct cftype
*cft
)
3858 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cont
);
3862 type
= MEMFILE_TYPE(cft
->private);
3863 name
= MEMFILE_ATTR(cft
->private);
3866 if (name
== RES_USAGE
)
3867 val
= mem_cgroup_usage(mem
, false);
3869 val
= res_counter_read_u64(&mem
->res
, name
);
3872 if (name
== RES_USAGE
)
3873 val
= mem_cgroup_usage(mem
, true);
3875 val
= res_counter_read_u64(&mem
->memsw
, name
);
3884 * The user of this function is...
3887 static int mem_cgroup_write(struct cgroup
*cont
, struct cftype
*cft
,
3890 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
3892 unsigned long long val
;
3895 type
= MEMFILE_TYPE(cft
->private);
3896 name
= MEMFILE_ATTR(cft
->private);
3899 if (mem_cgroup_is_root(memcg
)) { /* Can't set limit on root */
3903 /* This function does all necessary parse...reuse it */
3904 ret
= res_counter_memparse_write_strategy(buffer
, &val
);
3908 ret
= mem_cgroup_resize_limit(memcg
, val
);
3910 ret
= mem_cgroup_resize_memsw_limit(memcg
, val
);
3912 case RES_SOFT_LIMIT
:
3913 ret
= res_counter_memparse_write_strategy(buffer
, &val
);
3917 * For memsw, soft limits are hard to implement in terms
3918 * of semantics, for now, we support soft limits for
3919 * control without swap
3922 ret
= res_counter_set_soft_limit(&memcg
->res
, val
);
3927 ret
= -EINVAL
; /* should be BUG() ? */
3933 static void memcg_get_hierarchical_limit(struct mem_cgroup
*memcg
,
3934 unsigned long long *mem_limit
, unsigned long long *memsw_limit
)
3936 struct cgroup
*cgroup
;
3937 unsigned long long min_limit
, min_memsw_limit
, tmp
;
3939 min_limit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
3940 min_memsw_limit
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
3941 cgroup
= memcg
->css
.cgroup
;
3942 if (!memcg
->use_hierarchy
)
3945 while (cgroup
->parent
) {
3946 cgroup
= cgroup
->parent
;
3947 memcg
= mem_cgroup_from_cont(cgroup
);
3948 if (!memcg
->use_hierarchy
)
3950 tmp
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
3951 min_limit
= min(min_limit
, tmp
);
3952 tmp
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
3953 min_memsw_limit
= min(min_memsw_limit
, tmp
);
3956 *mem_limit
= min_limit
;
3957 *memsw_limit
= min_memsw_limit
;
3961 static int mem_cgroup_reset(struct cgroup
*cont
, unsigned int event
)
3963 struct mem_cgroup
*mem
;
3966 mem
= mem_cgroup_from_cont(cont
);
3967 type
= MEMFILE_TYPE(event
);
3968 name
= MEMFILE_ATTR(event
);
3972 res_counter_reset_max(&mem
->res
);
3974 res_counter_reset_max(&mem
->memsw
);
3978 res_counter_reset_failcnt(&mem
->res
);
3980 res_counter_reset_failcnt(&mem
->memsw
);
3987 static u64
mem_cgroup_move_charge_read(struct cgroup
*cgrp
,
3990 return mem_cgroup_from_cont(cgrp
)->move_charge_at_immigrate
;
3994 static int mem_cgroup_move_charge_write(struct cgroup
*cgrp
,
3995 struct cftype
*cft
, u64 val
)
3997 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cgrp
);
3999 if (val
>= (1 << NR_MOVE_TYPE
))
4002 * We check this value several times in both in can_attach() and
4003 * attach(), so we need cgroup lock to prevent this value from being
4007 mem
->move_charge_at_immigrate
= val
;
4013 static int mem_cgroup_move_charge_write(struct cgroup
*cgrp
,
4014 struct cftype
*cft
, u64 val
)
4021 /* For read statistics */
4039 struct mcs_total_stat
{
4040 s64 stat
[NR_MCS_STAT
];
4046 } memcg_stat_strings
[NR_MCS_STAT
] = {
4047 {"cache", "total_cache"},
4048 {"rss", "total_rss"},
4049 {"mapped_file", "total_mapped_file"},
4050 {"pgpgin", "total_pgpgin"},
4051 {"pgpgout", "total_pgpgout"},
4052 {"swap", "total_swap"},
4053 {"pgfault", "total_pgfault"},
4054 {"pgmajfault", "total_pgmajfault"},
4055 {"inactive_anon", "total_inactive_anon"},
4056 {"active_anon", "total_active_anon"},
4057 {"inactive_file", "total_inactive_file"},
4058 {"active_file", "total_active_file"},
4059 {"unevictable", "total_unevictable"}
4064 mem_cgroup_get_local_stat(struct mem_cgroup
*mem
, struct mcs_total_stat
*s
)
4069 val
= mem_cgroup_read_stat(mem
, MEM_CGROUP_STAT_CACHE
);
4070 s
->stat
[MCS_CACHE
] += val
* PAGE_SIZE
;
4071 val
= mem_cgroup_read_stat(mem
, MEM_CGROUP_STAT_RSS
);
4072 s
->stat
[MCS_RSS
] += val
* PAGE_SIZE
;
4073 val
= mem_cgroup_read_stat(mem
, MEM_CGROUP_STAT_FILE_MAPPED
);
4074 s
->stat
[MCS_FILE_MAPPED
] += val
* PAGE_SIZE
;
4075 val
= mem_cgroup_read_events(mem
, MEM_CGROUP_EVENTS_PGPGIN
);
4076 s
->stat
[MCS_PGPGIN
] += val
;
4077 val
= mem_cgroup_read_events(mem
, MEM_CGROUP_EVENTS_PGPGOUT
);
4078 s
->stat
[MCS_PGPGOUT
] += val
;
4079 if (do_swap_account
) {
4080 val
= mem_cgroup_read_stat(mem
, MEM_CGROUP_STAT_SWAPOUT
);
4081 s
->stat
[MCS_SWAP
] += val
* PAGE_SIZE
;
4083 val
= mem_cgroup_read_events(mem
, MEM_CGROUP_EVENTS_PGFAULT
);
4084 s
->stat
[MCS_PGFAULT
] += val
;
4085 val
= mem_cgroup_read_events(mem
, MEM_CGROUP_EVENTS_PGMAJFAULT
);
4086 s
->stat
[MCS_PGMAJFAULT
] += val
;
4089 val
= mem_cgroup_nr_lru_pages(mem
, BIT(LRU_INACTIVE_ANON
));
4090 s
->stat
[MCS_INACTIVE_ANON
] += val
* PAGE_SIZE
;
4091 val
= mem_cgroup_nr_lru_pages(mem
, BIT(LRU_ACTIVE_ANON
));
4092 s
->stat
[MCS_ACTIVE_ANON
] += val
* PAGE_SIZE
;
4093 val
= mem_cgroup_nr_lru_pages(mem
, BIT(LRU_INACTIVE_FILE
));
4094 s
->stat
[MCS_INACTIVE_FILE
] += val
* PAGE_SIZE
;
4095 val
= mem_cgroup_nr_lru_pages(mem
, BIT(LRU_ACTIVE_FILE
));
4096 s
->stat
[MCS_ACTIVE_FILE
] += val
* PAGE_SIZE
;
4097 val
= mem_cgroup_nr_lru_pages(mem
, BIT(LRU_UNEVICTABLE
));
4098 s
->stat
[MCS_UNEVICTABLE
] += val
* PAGE_SIZE
;
4102 mem_cgroup_get_total_stat(struct mem_cgroup
*mem
, struct mcs_total_stat
*s
)
4104 struct mem_cgroup
*iter
;
4106 for_each_mem_cgroup_tree(iter
, mem
)
4107 mem_cgroup_get_local_stat(iter
, s
);
4111 static int mem_control_numa_stat_show(struct seq_file
*m
, void *arg
)
4114 unsigned long total_nr
, file_nr
, anon_nr
, unevictable_nr
;
4115 unsigned long node_nr
;
4116 struct cgroup
*cont
= m
->private;
4117 struct mem_cgroup
*mem_cont
= mem_cgroup_from_cont(cont
);
4119 total_nr
= mem_cgroup_nr_lru_pages(mem_cont
, LRU_ALL
);
4120 seq_printf(m
, "total=%lu", total_nr
);
4121 for_each_node_state(nid
, N_HIGH_MEMORY
) {
4122 node_nr
= mem_cgroup_node_nr_lru_pages(mem_cont
, nid
, LRU_ALL
);
4123 seq_printf(m
, " N%d=%lu", nid
, node_nr
);
4127 file_nr
= mem_cgroup_nr_lru_pages(mem_cont
, LRU_ALL_FILE
);
4128 seq_printf(m
, "file=%lu", file_nr
);
4129 for_each_node_state(nid
, N_HIGH_MEMORY
) {
4130 node_nr
= mem_cgroup_node_nr_lru_pages(mem_cont
, nid
,
4132 seq_printf(m
, " N%d=%lu", nid
, node_nr
);
4136 anon_nr
= mem_cgroup_nr_lru_pages(mem_cont
, LRU_ALL_ANON
);
4137 seq_printf(m
, "anon=%lu", anon_nr
);
4138 for_each_node_state(nid
, N_HIGH_MEMORY
) {
4139 node_nr
= mem_cgroup_node_nr_lru_pages(mem_cont
, nid
,
4141 seq_printf(m
, " N%d=%lu", nid
, node_nr
);
4145 unevictable_nr
= mem_cgroup_nr_lru_pages(mem_cont
, BIT(LRU_UNEVICTABLE
));
4146 seq_printf(m
, "unevictable=%lu", unevictable_nr
);
4147 for_each_node_state(nid
, N_HIGH_MEMORY
) {
4148 node_nr
= mem_cgroup_node_nr_lru_pages(mem_cont
, nid
,
4149 BIT(LRU_UNEVICTABLE
));
4150 seq_printf(m
, " N%d=%lu", nid
, node_nr
);
4155 #endif /* CONFIG_NUMA */
4157 static int mem_control_stat_show(struct cgroup
*cont
, struct cftype
*cft
,
4158 struct cgroup_map_cb
*cb
)
4160 struct mem_cgroup
*mem_cont
= mem_cgroup_from_cont(cont
);
4161 struct mcs_total_stat mystat
;
4164 memset(&mystat
, 0, sizeof(mystat
));
4165 mem_cgroup_get_local_stat(mem_cont
, &mystat
);
4168 for (i
= 0; i
< NR_MCS_STAT
; i
++) {
4169 if (i
== MCS_SWAP
&& !do_swap_account
)
4171 cb
->fill(cb
, memcg_stat_strings
[i
].local_name
, mystat
.stat
[i
]);
4174 /* Hierarchical information */
4176 unsigned long long limit
, memsw_limit
;
4177 memcg_get_hierarchical_limit(mem_cont
, &limit
, &memsw_limit
);
4178 cb
->fill(cb
, "hierarchical_memory_limit", limit
);
4179 if (do_swap_account
)
4180 cb
->fill(cb
, "hierarchical_memsw_limit", memsw_limit
);
4183 memset(&mystat
, 0, sizeof(mystat
));
4184 mem_cgroup_get_total_stat(mem_cont
, &mystat
);
4185 for (i
= 0; i
< NR_MCS_STAT
; i
++) {
4186 if (i
== MCS_SWAP
&& !do_swap_account
)
4188 cb
->fill(cb
, memcg_stat_strings
[i
].total_name
, mystat
.stat
[i
]);
4191 #ifdef CONFIG_DEBUG_VM
4192 cb
->fill(cb
, "inactive_ratio", calc_inactive_ratio(mem_cont
, NULL
));
4196 struct mem_cgroup_per_zone
*mz
;
4197 unsigned long recent_rotated
[2] = {0, 0};
4198 unsigned long recent_scanned
[2] = {0, 0};
4200 for_each_online_node(nid
)
4201 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++) {
4202 mz
= mem_cgroup_zoneinfo(mem_cont
, nid
, zid
);
4204 recent_rotated
[0] +=
4205 mz
->reclaim_stat
.recent_rotated
[0];
4206 recent_rotated
[1] +=
4207 mz
->reclaim_stat
.recent_rotated
[1];
4208 recent_scanned
[0] +=
4209 mz
->reclaim_stat
.recent_scanned
[0];
4210 recent_scanned
[1] +=
4211 mz
->reclaim_stat
.recent_scanned
[1];
4213 cb
->fill(cb
, "recent_rotated_anon", recent_rotated
[0]);
4214 cb
->fill(cb
, "recent_rotated_file", recent_rotated
[1]);
4215 cb
->fill(cb
, "recent_scanned_anon", recent_scanned
[0]);
4216 cb
->fill(cb
, "recent_scanned_file", recent_scanned
[1]);
4223 static u64
mem_cgroup_swappiness_read(struct cgroup
*cgrp
, struct cftype
*cft
)
4225 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4227 return mem_cgroup_swappiness(memcg
);
4230 static int mem_cgroup_swappiness_write(struct cgroup
*cgrp
, struct cftype
*cft
,
4233 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4234 struct mem_cgroup
*parent
;
4239 if (cgrp
->parent
== NULL
)
4242 parent
= mem_cgroup_from_cont(cgrp
->parent
);
4246 /* If under hierarchy, only empty-root can set this value */
4247 if ((parent
->use_hierarchy
) ||
4248 (memcg
->use_hierarchy
&& !list_empty(&cgrp
->children
))) {
4253 memcg
->swappiness
= val
;
4260 static void __mem_cgroup_threshold(struct mem_cgroup
*memcg
, bool swap
)
4262 struct mem_cgroup_threshold_ary
*t
;
4268 t
= rcu_dereference(memcg
->thresholds
.primary
);
4270 t
= rcu_dereference(memcg
->memsw_thresholds
.primary
);
4275 usage
= mem_cgroup_usage(memcg
, swap
);
4278 * current_threshold points to threshold just below usage.
4279 * If it's not true, a threshold was crossed after last
4280 * call of __mem_cgroup_threshold().
4282 i
= t
->current_threshold
;
4285 * Iterate backward over array of thresholds starting from
4286 * current_threshold and check if a threshold is crossed.
4287 * If none of thresholds below usage is crossed, we read
4288 * only one element of the array here.
4290 for (; i
>= 0 && unlikely(t
->entries
[i
].threshold
> usage
); i
--)
4291 eventfd_signal(t
->entries
[i
].eventfd
, 1);
4293 /* i = current_threshold + 1 */
4297 * Iterate forward over array of thresholds starting from
4298 * current_threshold+1 and check if a threshold is crossed.
4299 * If none of thresholds above usage is crossed, we read
4300 * only one element of the array here.
4302 for (; i
< t
->size
&& unlikely(t
->entries
[i
].threshold
<= usage
); i
++)
4303 eventfd_signal(t
->entries
[i
].eventfd
, 1);
4305 /* Update current_threshold */
4306 t
->current_threshold
= i
- 1;
4311 static void mem_cgroup_threshold(struct mem_cgroup
*memcg
)
4314 __mem_cgroup_threshold(memcg
, false);
4315 if (do_swap_account
)
4316 __mem_cgroup_threshold(memcg
, true);
4318 memcg
= parent_mem_cgroup(memcg
);
4322 static int compare_thresholds(const void *a
, const void *b
)
4324 const struct mem_cgroup_threshold
*_a
= a
;
4325 const struct mem_cgroup_threshold
*_b
= b
;
4327 return _a
->threshold
- _b
->threshold
;
4330 static int mem_cgroup_oom_notify_cb(struct mem_cgroup
*mem
)
4332 struct mem_cgroup_eventfd_list
*ev
;
4334 list_for_each_entry(ev
, &mem
->oom_notify
, list
)
4335 eventfd_signal(ev
->eventfd
, 1);
4339 static void mem_cgroup_oom_notify(struct mem_cgroup
*mem
)
4341 struct mem_cgroup
*iter
;
4343 for_each_mem_cgroup_tree(iter
, mem
)
4344 mem_cgroup_oom_notify_cb(iter
);
4347 static int mem_cgroup_usage_register_event(struct cgroup
*cgrp
,
4348 struct cftype
*cft
, struct eventfd_ctx
*eventfd
, const char *args
)
4350 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4351 struct mem_cgroup_thresholds
*thresholds
;
4352 struct mem_cgroup_threshold_ary
*new;
4353 int type
= MEMFILE_TYPE(cft
->private);
4354 u64 threshold
, usage
;
4357 ret
= res_counter_memparse_write_strategy(args
, &threshold
);
4361 mutex_lock(&memcg
->thresholds_lock
);
4364 thresholds
= &memcg
->thresholds
;
4365 else if (type
== _MEMSWAP
)
4366 thresholds
= &memcg
->memsw_thresholds
;
4370 usage
= mem_cgroup_usage(memcg
, type
== _MEMSWAP
);
4372 /* Check if a threshold crossed before adding a new one */
4373 if (thresholds
->primary
)
4374 __mem_cgroup_threshold(memcg
, type
== _MEMSWAP
);
4376 size
= thresholds
->primary
? thresholds
->primary
->size
+ 1 : 1;
4378 /* Allocate memory for new array of thresholds */
4379 new = kmalloc(sizeof(*new) + size
* sizeof(struct mem_cgroup_threshold
),
4387 /* Copy thresholds (if any) to new array */
4388 if (thresholds
->primary
) {
4389 memcpy(new->entries
, thresholds
->primary
->entries
, (size
- 1) *
4390 sizeof(struct mem_cgroup_threshold
));
4393 /* Add new threshold */
4394 new->entries
[size
- 1].eventfd
= eventfd
;
4395 new->entries
[size
- 1].threshold
= threshold
;
4397 /* Sort thresholds. Registering of new threshold isn't time-critical */
4398 sort(new->entries
, size
, sizeof(struct mem_cgroup_threshold
),
4399 compare_thresholds
, NULL
);
4401 /* Find current threshold */
4402 new->current_threshold
= -1;
4403 for (i
= 0; i
< size
; i
++) {
4404 if (new->entries
[i
].threshold
< usage
) {
4406 * new->current_threshold will not be used until
4407 * rcu_assign_pointer(), so it's safe to increment
4410 ++new->current_threshold
;
4414 /* Free old spare buffer and save old primary buffer as spare */
4415 kfree(thresholds
->spare
);
4416 thresholds
->spare
= thresholds
->primary
;
4418 rcu_assign_pointer(thresholds
->primary
, new);
4420 /* To be sure that nobody uses thresholds */
4424 mutex_unlock(&memcg
->thresholds_lock
);
4429 static void mem_cgroup_usage_unregister_event(struct cgroup
*cgrp
,
4430 struct cftype
*cft
, struct eventfd_ctx
*eventfd
)
4432 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4433 struct mem_cgroup_thresholds
*thresholds
;
4434 struct mem_cgroup_threshold_ary
*new;
4435 int type
= MEMFILE_TYPE(cft
->private);
4439 mutex_lock(&memcg
->thresholds_lock
);
4441 thresholds
= &memcg
->thresholds
;
4442 else if (type
== _MEMSWAP
)
4443 thresholds
= &memcg
->memsw_thresholds
;
4448 * Something went wrong if we trying to unregister a threshold
4449 * if we don't have thresholds
4451 BUG_ON(!thresholds
);
4453 usage
= mem_cgroup_usage(memcg
, type
== _MEMSWAP
);
4455 /* Check if a threshold crossed before removing */
4456 __mem_cgroup_threshold(memcg
, type
== _MEMSWAP
);
4458 /* Calculate new number of threshold */
4460 for (i
= 0; i
< thresholds
->primary
->size
; i
++) {
4461 if (thresholds
->primary
->entries
[i
].eventfd
!= eventfd
)
4465 new = thresholds
->spare
;
4467 /* Set thresholds array to NULL if we don't have thresholds */
4476 /* Copy thresholds and find current threshold */
4477 new->current_threshold
= -1;
4478 for (i
= 0, j
= 0; i
< thresholds
->primary
->size
; i
++) {
4479 if (thresholds
->primary
->entries
[i
].eventfd
== eventfd
)
4482 new->entries
[j
] = thresholds
->primary
->entries
[i
];
4483 if (new->entries
[j
].threshold
< usage
) {
4485 * new->current_threshold will not be used
4486 * until rcu_assign_pointer(), so it's safe to increment
4489 ++new->current_threshold
;
4495 /* Swap primary and spare array */
4496 thresholds
->spare
= thresholds
->primary
;
4497 rcu_assign_pointer(thresholds
->primary
, new);
4499 /* To be sure that nobody uses thresholds */
4502 mutex_unlock(&memcg
->thresholds_lock
);
4505 static int mem_cgroup_oom_register_event(struct cgroup
*cgrp
,
4506 struct cftype
*cft
, struct eventfd_ctx
*eventfd
, const char *args
)
4508 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
4509 struct mem_cgroup_eventfd_list
*event
;
4510 int type
= MEMFILE_TYPE(cft
->private);
4512 BUG_ON(type
!= _OOM_TYPE
);
4513 event
= kmalloc(sizeof(*event
), GFP_KERNEL
);
4517 spin_lock(&memcg_oom_lock
);
4519 event
->eventfd
= eventfd
;
4520 list_add(&event
->list
, &memcg
->oom_notify
);
4522 /* already in OOM ? */
4523 if (atomic_read(&memcg
->under_oom
))
4524 eventfd_signal(eventfd
, 1);
4525 spin_unlock(&memcg_oom_lock
);
4530 static void mem_cgroup_oom_unregister_event(struct cgroup
*cgrp
,
4531 struct cftype
*cft
, struct eventfd_ctx
*eventfd
)
4533 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cgrp
);
4534 struct mem_cgroup_eventfd_list
*ev
, *tmp
;
4535 int type
= MEMFILE_TYPE(cft
->private);
4537 BUG_ON(type
!= _OOM_TYPE
);
4539 spin_lock(&memcg_oom_lock
);
4541 list_for_each_entry_safe(ev
, tmp
, &mem
->oom_notify
, list
) {
4542 if (ev
->eventfd
== eventfd
) {
4543 list_del(&ev
->list
);
4548 spin_unlock(&memcg_oom_lock
);
4551 static int mem_cgroup_oom_control_read(struct cgroup
*cgrp
,
4552 struct cftype
*cft
, struct cgroup_map_cb
*cb
)
4554 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cgrp
);
4556 cb
->fill(cb
, "oom_kill_disable", mem
->oom_kill_disable
);
4558 if (atomic_read(&mem
->under_oom
))
4559 cb
->fill(cb
, "under_oom", 1);
4561 cb
->fill(cb
, "under_oom", 0);
4565 static int mem_cgroup_oom_control_write(struct cgroup
*cgrp
,
4566 struct cftype
*cft
, u64 val
)
4568 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cgrp
);
4569 struct mem_cgroup
*parent
;
4571 /* cannot set to root cgroup and only 0 and 1 are allowed */
4572 if (!cgrp
->parent
|| !((val
== 0) || (val
== 1)))
4575 parent
= mem_cgroup_from_cont(cgrp
->parent
);
4578 /* oom-kill-disable is a flag for subhierarchy. */
4579 if ((parent
->use_hierarchy
) ||
4580 (mem
->use_hierarchy
&& !list_empty(&cgrp
->children
))) {
4584 mem
->oom_kill_disable
= val
;
4586 memcg_oom_recover(mem
);
4592 static const struct file_operations mem_control_numa_stat_file_operations
= {
4594 .llseek
= seq_lseek
,
4595 .release
= single_release
,
4598 static int mem_control_numa_stat_open(struct inode
*unused
, struct file
*file
)
4600 struct cgroup
*cont
= file
->f_dentry
->d_parent
->d_fsdata
;
4602 file
->f_op
= &mem_control_numa_stat_file_operations
;
4603 return single_open(file
, mem_control_numa_stat_show
, cont
);
4605 #endif /* CONFIG_NUMA */
4607 static struct cftype mem_cgroup_files
[] = {
4609 .name
= "usage_in_bytes",
4610 .private = MEMFILE_PRIVATE(_MEM
, RES_USAGE
),
4611 .read_u64
= mem_cgroup_read
,
4612 .register_event
= mem_cgroup_usage_register_event
,
4613 .unregister_event
= mem_cgroup_usage_unregister_event
,
4616 .name
= "max_usage_in_bytes",
4617 .private = MEMFILE_PRIVATE(_MEM
, RES_MAX_USAGE
),
4618 .trigger
= mem_cgroup_reset
,
4619 .read_u64
= mem_cgroup_read
,
4622 .name
= "limit_in_bytes",
4623 .private = MEMFILE_PRIVATE(_MEM
, RES_LIMIT
),
4624 .write_string
= mem_cgroup_write
,
4625 .read_u64
= mem_cgroup_read
,
4628 .name
= "soft_limit_in_bytes",
4629 .private = MEMFILE_PRIVATE(_MEM
, RES_SOFT_LIMIT
),
4630 .write_string
= mem_cgroup_write
,
4631 .read_u64
= mem_cgroup_read
,
4635 .private = MEMFILE_PRIVATE(_MEM
, RES_FAILCNT
),
4636 .trigger
= mem_cgroup_reset
,
4637 .read_u64
= mem_cgroup_read
,
4641 .read_map
= mem_control_stat_show
,
4644 .name
= "force_empty",
4645 .trigger
= mem_cgroup_force_empty_write
,
4648 .name
= "use_hierarchy",
4649 .write_u64
= mem_cgroup_hierarchy_write
,
4650 .read_u64
= mem_cgroup_hierarchy_read
,
4653 .name
= "swappiness",
4654 .read_u64
= mem_cgroup_swappiness_read
,
4655 .write_u64
= mem_cgroup_swappiness_write
,
4658 .name
= "move_charge_at_immigrate",
4659 .read_u64
= mem_cgroup_move_charge_read
,
4660 .write_u64
= mem_cgroup_move_charge_write
,
4663 .name
= "oom_control",
4664 .read_map
= mem_cgroup_oom_control_read
,
4665 .write_u64
= mem_cgroup_oom_control_write
,
4666 .register_event
= mem_cgroup_oom_register_event
,
4667 .unregister_event
= mem_cgroup_oom_unregister_event
,
4668 .private = MEMFILE_PRIVATE(_OOM_TYPE
, OOM_CONTROL
),
4672 .name
= "numa_stat",
4673 .open
= mem_control_numa_stat_open
,
4679 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
4680 static struct cftype memsw_cgroup_files
[] = {
4682 .name
= "memsw.usage_in_bytes",
4683 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_USAGE
),
4684 .read_u64
= mem_cgroup_read
,
4685 .register_event
= mem_cgroup_usage_register_event
,
4686 .unregister_event
= mem_cgroup_usage_unregister_event
,
4689 .name
= "memsw.max_usage_in_bytes",
4690 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_MAX_USAGE
),
4691 .trigger
= mem_cgroup_reset
,
4692 .read_u64
= mem_cgroup_read
,
4695 .name
= "memsw.limit_in_bytes",
4696 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_LIMIT
),
4697 .write_string
= mem_cgroup_write
,
4698 .read_u64
= mem_cgroup_read
,
4701 .name
= "memsw.failcnt",
4702 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_FAILCNT
),
4703 .trigger
= mem_cgroup_reset
,
4704 .read_u64
= mem_cgroup_read
,
4708 static int register_memsw_files(struct cgroup
*cont
, struct cgroup_subsys
*ss
)
4710 if (!do_swap_account
)
4712 return cgroup_add_files(cont
, ss
, memsw_cgroup_files
,
4713 ARRAY_SIZE(memsw_cgroup_files
));
4716 static int register_memsw_files(struct cgroup
*cont
, struct cgroup_subsys
*ss
)
4722 static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup
*mem
, int node
)
4724 struct mem_cgroup_per_node
*pn
;
4725 struct mem_cgroup_per_zone
*mz
;
4727 int zone
, tmp
= node
;
4729 * This routine is called against possible nodes.
4730 * But it's BUG to call kmalloc() against offline node.
4732 * TODO: this routine can waste much memory for nodes which will
4733 * never be onlined. It's better to use memory hotplug callback
4736 if (!node_state(node
, N_NORMAL_MEMORY
))
4738 pn
= kzalloc_node(sizeof(*pn
), GFP_KERNEL
, tmp
);
4742 mem
->info
.nodeinfo
[node
] = pn
;
4743 for (zone
= 0; zone
< MAX_NR_ZONES
; zone
++) {
4744 mz
= &pn
->zoneinfo
[zone
];
4746 INIT_LIST_HEAD(&mz
->lists
[l
]);
4747 mz
->usage_in_excess
= 0;
4748 mz
->on_tree
= false;
4754 static void free_mem_cgroup_per_zone_info(struct mem_cgroup
*mem
, int node
)
4756 kfree(mem
->info
.nodeinfo
[node
]);
4759 static struct mem_cgroup
*mem_cgroup_alloc(void)
4761 struct mem_cgroup
*mem
;
4762 int size
= sizeof(struct mem_cgroup
);
4764 /* Can be very big if MAX_NUMNODES is very big */
4765 if (size
< PAGE_SIZE
)
4766 mem
= kzalloc(size
, GFP_KERNEL
);
4768 mem
= vzalloc(size
);
4773 mem
->stat
= alloc_percpu(struct mem_cgroup_stat_cpu
);
4776 spin_lock_init(&mem
->pcp_counter_lock
);
4780 if (size
< PAGE_SIZE
)
4788 * At destroying mem_cgroup, references from swap_cgroup can remain.
4789 * (scanning all at force_empty is too costly...)
4791 * Instead of clearing all references at force_empty, we remember
4792 * the number of reference from swap_cgroup and free mem_cgroup when
4793 * it goes down to 0.
4795 * Removal of cgroup itself succeeds regardless of refs from swap.
4798 static void __mem_cgroup_free(struct mem_cgroup
*mem
)
4802 mem_cgroup_remove_from_trees(mem
);
4803 free_css_id(&mem_cgroup_subsys
, &mem
->css
);
4805 for_each_node_state(node
, N_POSSIBLE
)
4806 free_mem_cgroup_per_zone_info(mem
, node
);
4808 free_percpu(mem
->stat
);
4809 if (sizeof(struct mem_cgroup
) < PAGE_SIZE
)
4815 static void mem_cgroup_get(struct mem_cgroup
*mem
)
4817 atomic_inc(&mem
->refcnt
);
4820 static void __mem_cgroup_put(struct mem_cgroup
*mem
, int count
)
4822 if (atomic_sub_and_test(count
, &mem
->refcnt
)) {
4823 struct mem_cgroup
*parent
= parent_mem_cgroup(mem
);
4824 __mem_cgroup_free(mem
);
4826 mem_cgroup_put(parent
);
4830 static void mem_cgroup_put(struct mem_cgroup
*mem
)
4832 __mem_cgroup_put(mem
, 1);
4836 * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled.
4838 static struct mem_cgroup
*parent_mem_cgroup(struct mem_cgroup
*mem
)
4840 if (!mem
->res
.parent
)
4842 return mem_cgroup_from_res_counter(mem
->res
.parent
, res
);
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_state(node
, N_POSSIBLE
) {
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 static struct cgroup_subsys_state
* __ref
4883 mem_cgroup_create(struct cgroup_subsys
*ss
, struct cgroup
*cont
)
4885 struct mem_cgroup
*mem
, *parent
;
4886 long error
= -ENOMEM
;
4889 mem
= mem_cgroup_alloc();
4891 return ERR_PTR(error
);
4893 for_each_node_state(node
, N_POSSIBLE
)
4894 if (alloc_mem_cgroup_per_zone_info(mem
, node
))
4898 if (cont
->parent
== NULL
) {
4900 enable_swap_cgroup();
4902 root_mem_cgroup
= mem
;
4903 if (mem_cgroup_soft_limit_tree_init())
4905 for_each_possible_cpu(cpu
) {
4906 struct memcg_stock_pcp
*stock
=
4907 &per_cpu(memcg_stock
, cpu
);
4908 INIT_WORK(&stock
->work
, drain_local_stock
);
4910 hotcpu_notifier(memcg_cpu_hotplug_callback
, 0);
4912 parent
= mem_cgroup_from_cont(cont
->parent
);
4913 mem
->use_hierarchy
= parent
->use_hierarchy
;
4914 mem
->oom_kill_disable
= parent
->oom_kill_disable
;
4917 if (parent
&& parent
->use_hierarchy
) {
4918 res_counter_init(&mem
->res
, &parent
->res
);
4919 res_counter_init(&mem
->memsw
, &parent
->memsw
);
4921 * We increment refcnt of the parent to ensure that we can
4922 * safely access it on res_counter_charge/uncharge.
4923 * This refcnt will be decremented when freeing this
4924 * mem_cgroup(see mem_cgroup_put).
4926 mem_cgroup_get(parent
);
4928 res_counter_init(&mem
->res
, NULL
);
4929 res_counter_init(&mem
->memsw
, NULL
);
4931 mem
->last_scanned_child
= 0;
4932 mem
->last_scanned_node
= MAX_NUMNODES
;
4933 INIT_LIST_HEAD(&mem
->oom_notify
);
4936 mem
->swappiness
= mem_cgroup_swappiness(parent
);
4937 atomic_set(&mem
->refcnt
, 1);
4938 mem
->move_charge_at_immigrate
= 0;
4939 mutex_init(&mem
->thresholds_lock
);
4942 __mem_cgroup_free(mem
);
4943 root_mem_cgroup
= NULL
;
4944 return ERR_PTR(error
);
4947 static int mem_cgroup_pre_destroy(struct cgroup_subsys
*ss
,
4948 struct cgroup
*cont
)
4950 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cont
);
4952 return mem_cgroup_force_empty(mem
, false);
4955 static void mem_cgroup_destroy(struct cgroup_subsys
*ss
,
4956 struct cgroup
*cont
)
4958 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cont
);
4960 mem_cgroup_put(mem
);
4963 static int mem_cgroup_populate(struct cgroup_subsys
*ss
,
4964 struct cgroup
*cont
)
4968 ret
= cgroup_add_files(cont
, ss
, mem_cgroup_files
,
4969 ARRAY_SIZE(mem_cgroup_files
));
4972 ret
= register_memsw_files(cont
, ss
);
4977 /* Handlers for move charge at task migration. */
4978 #define PRECHARGE_COUNT_AT_ONCE 256
4979 static int mem_cgroup_do_precharge(unsigned long count
)
4982 int batch_count
= PRECHARGE_COUNT_AT_ONCE
;
4983 struct mem_cgroup
*mem
= mc
.to
;
4985 if (mem_cgroup_is_root(mem
)) {
4986 mc
.precharge
+= count
;
4987 /* we don't need css_get for root */
4990 /* try to charge at once */
4992 struct res_counter
*dummy
;
4994 * "mem" cannot be under rmdir() because we've already checked
4995 * by cgroup_lock_live_cgroup() that it is not removed and we
4996 * are still under the same cgroup_mutex. So we can postpone
4999 if (res_counter_charge(&mem
->res
, PAGE_SIZE
* count
, &dummy
))
5001 if (do_swap_account
&& res_counter_charge(&mem
->memsw
,
5002 PAGE_SIZE
* count
, &dummy
)) {
5003 res_counter_uncharge(&mem
->res
, PAGE_SIZE
* count
);
5006 mc
.precharge
+= count
;
5010 /* fall back to one by one charge */
5012 if (signal_pending(current
)) {
5016 if (!batch_count
--) {
5017 batch_count
= PRECHARGE_COUNT_AT_ONCE
;
5020 ret
= __mem_cgroup_try_charge(NULL
, GFP_KERNEL
, 1, &mem
, false);
5022 /* mem_cgroup_clear_mc() will do uncharge later */
5030 * is_target_pte_for_mc - check a pte whether it is valid for move charge
5031 * @vma: the vma the pte to be checked belongs
5032 * @addr: the address corresponding to the pte to be checked
5033 * @ptent: the pte to be checked
5034 * @target: the pointer the target page or swap ent will be stored(can be NULL)
5037 * 0(MC_TARGET_NONE): if the pte is not a target for move charge.
5038 * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
5039 * move charge. if @target is not NULL, the page is stored in target->page
5040 * with extra refcnt got(Callers should handle it).
5041 * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
5042 * target for charge migration. if @target is not NULL, the entry is stored
5045 * Called with pte lock held.
5052 enum mc_target_type
{
5053 MC_TARGET_NONE
, /* not used */
5058 static struct page
*mc_handle_present_pte(struct vm_area_struct
*vma
,
5059 unsigned long addr
, pte_t ptent
)
5061 struct page
*page
= vm_normal_page(vma
, addr
, ptent
);
5063 if (!page
|| !page_mapped(page
))
5065 if (PageAnon(page
)) {
5066 /* we don't move shared anon */
5067 if (!move_anon() || page_mapcount(page
) > 2)
5069 } else if (!move_file())
5070 /* we ignore mapcount for file pages */
5072 if (!get_page_unless_zero(page
))
5078 static struct page
*mc_handle_swap_pte(struct vm_area_struct
*vma
,
5079 unsigned long addr
, pte_t ptent
, swp_entry_t
*entry
)
5082 struct page
*page
= NULL
;
5083 swp_entry_t ent
= pte_to_swp_entry(ptent
);
5085 if (!move_anon() || non_swap_entry(ent
))
5087 usage_count
= mem_cgroup_count_swap_user(ent
, &page
);
5088 if (usage_count
> 1) { /* we don't move shared anon */
5093 if (do_swap_account
)
5094 entry
->val
= ent
.val
;
5099 static struct page
*mc_handle_file_pte(struct vm_area_struct
*vma
,
5100 unsigned long addr
, pte_t ptent
, swp_entry_t
*entry
)
5102 struct page
*page
= NULL
;
5103 struct inode
*inode
;
5104 struct address_space
*mapping
;
5107 if (!vma
->vm_file
) /* anonymous vma */
5112 inode
= vma
->vm_file
->f_path
.dentry
->d_inode
;
5113 mapping
= vma
->vm_file
->f_mapping
;
5114 if (pte_none(ptent
))
5115 pgoff
= linear_page_index(vma
, addr
);
5116 else /* pte_file(ptent) is true */
5117 pgoff
= pte_to_pgoff(ptent
);
5119 /* page is moved even if it's not RSS of this task(page-faulted). */
5120 page
= find_get_page(mapping
, pgoff
);
5123 /* shmem/tmpfs may report page out on swap: account for that too. */
5124 if (radix_tree_exceptional_entry(page
)) {
5125 swp_entry_t swap
= radix_to_swp_entry(page
);
5126 if (do_swap_account
)
5128 page
= find_get_page(&swapper_space
, swap
.val
);
5134 static int is_target_pte_for_mc(struct vm_area_struct
*vma
,
5135 unsigned long addr
, pte_t ptent
, union mc_target
*target
)
5137 struct page
*page
= NULL
;
5138 struct page_cgroup
*pc
;
5140 swp_entry_t ent
= { .val
= 0 };
5142 if (pte_present(ptent
))
5143 page
= mc_handle_present_pte(vma
, addr
, ptent
);
5144 else if (is_swap_pte(ptent
))
5145 page
= mc_handle_swap_pte(vma
, addr
, ptent
, &ent
);
5146 else if (pte_none(ptent
) || pte_file(ptent
))
5147 page
= mc_handle_file_pte(vma
, addr
, ptent
, &ent
);
5149 if (!page
&& !ent
.val
)
5152 pc
= lookup_page_cgroup(page
);
5154 * Do only loose check w/o page_cgroup lock.
5155 * mem_cgroup_move_account() checks the pc is valid or not under
5158 if (PageCgroupUsed(pc
) && pc
->mem_cgroup
== mc
.from
) {
5159 ret
= MC_TARGET_PAGE
;
5161 target
->page
= page
;
5163 if (!ret
|| !target
)
5166 /* There is a swap entry and a page doesn't exist or isn't charged */
5167 if (ent
.val
&& !ret
&&
5168 css_id(&mc
.from
->css
) == lookup_swap_cgroup(ent
)) {
5169 ret
= MC_TARGET_SWAP
;
5176 static int mem_cgroup_count_precharge_pte_range(pmd_t
*pmd
,
5177 unsigned long addr
, unsigned long end
,
5178 struct mm_walk
*walk
)
5180 struct vm_area_struct
*vma
= walk
->private;
5184 split_huge_page_pmd(walk
->mm
, pmd
);
5186 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
5187 for (; addr
!= end
; pte
++, addr
+= PAGE_SIZE
)
5188 if (is_target_pte_for_mc(vma
, addr
, *pte
, NULL
))
5189 mc
.precharge
++; /* increment precharge temporarily */
5190 pte_unmap_unlock(pte
- 1, ptl
);
5196 static unsigned long mem_cgroup_count_precharge(struct mm_struct
*mm
)
5198 unsigned long precharge
;
5199 struct vm_area_struct
*vma
;
5201 down_read(&mm
->mmap_sem
);
5202 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
5203 struct mm_walk mem_cgroup_count_precharge_walk
= {
5204 .pmd_entry
= mem_cgroup_count_precharge_pte_range
,
5208 if (is_vm_hugetlb_page(vma
))
5210 walk_page_range(vma
->vm_start
, vma
->vm_end
,
5211 &mem_cgroup_count_precharge_walk
);
5213 up_read(&mm
->mmap_sem
);
5215 precharge
= mc
.precharge
;
5221 static int mem_cgroup_precharge_mc(struct mm_struct
*mm
)
5223 unsigned long precharge
= mem_cgroup_count_precharge(mm
);
5225 VM_BUG_ON(mc
.moving_task
);
5226 mc
.moving_task
= current
;
5227 return mem_cgroup_do_precharge(precharge
);
5230 /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */
5231 static void __mem_cgroup_clear_mc(void)
5233 struct mem_cgroup
*from
= mc
.from
;
5234 struct mem_cgroup
*to
= mc
.to
;
5236 /* we must uncharge all the leftover precharges from mc.to */
5238 __mem_cgroup_cancel_charge(mc
.to
, mc
.precharge
);
5242 * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
5243 * we must uncharge here.
5245 if (mc
.moved_charge
) {
5246 __mem_cgroup_cancel_charge(mc
.from
, mc
.moved_charge
);
5247 mc
.moved_charge
= 0;
5249 /* we must fixup refcnts and charges */
5250 if (mc
.moved_swap
) {
5251 /* uncharge swap account from the old cgroup */
5252 if (!mem_cgroup_is_root(mc
.from
))
5253 res_counter_uncharge(&mc
.from
->memsw
,
5254 PAGE_SIZE
* mc
.moved_swap
);
5255 __mem_cgroup_put(mc
.from
, mc
.moved_swap
);
5257 if (!mem_cgroup_is_root(mc
.to
)) {
5259 * we charged both to->res and to->memsw, so we should
5262 res_counter_uncharge(&mc
.to
->res
,
5263 PAGE_SIZE
* mc
.moved_swap
);
5265 /* we've already done mem_cgroup_get(mc.to) */
5268 memcg_oom_recover(from
);
5269 memcg_oom_recover(to
);
5270 wake_up_all(&mc
.waitq
);
5273 static void mem_cgroup_clear_mc(void)
5275 struct mem_cgroup
*from
= mc
.from
;
5278 * we must clear moving_task before waking up waiters at the end of
5281 mc
.moving_task
= NULL
;
5282 __mem_cgroup_clear_mc();
5283 spin_lock(&mc
.lock
);
5286 spin_unlock(&mc
.lock
);
5287 mem_cgroup_end_move(from
);
5290 static int mem_cgroup_can_attach(struct cgroup_subsys
*ss
,
5291 struct cgroup
*cgroup
,
5292 struct task_struct
*p
)
5295 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cgroup
);
5297 if (mem
->move_charge_at_immigrate
) {
5298 struct mm_struct
*mm
;
5299 struct mem_cgroup
*from
= mem_cgroup_from_task(p
);
5301 VM_BUG_ON(from
== mem
);
5303 mm
= get_task_mm(p
);
5306 /* We move charges only when we move a owner of the mm */
5307 if (mm
->owner
== p
) {
5310 VM_BUG_ON(mc
.precharge
);
5311 VM_BUG_ON(mc
.moved_charge
);
5312 VM_BUG_ON(mc
.moved_swap
);
5313 mem_cgroup_start_move(from
);
5314 spin_lock(&mc
.lock
);
5317 spin_unlock(&mc
.lock
);
5318 /* We set mc.moving_task later */
5320 ret
= mem_cgroup_precharge_mc(mm
);
5322 mem_cgroup_clear_mc();
5329 static void mem_cgroup_cancel_attach(struct cgroup_subsys
*ss
,
5330 struct cgroup
*cgroup
,
5331 struct task_struct
*p
)
5333 mem_cgroup_clear_mc();
5336 static int mem_cgroup_move_charge_pte_range(pmd_t
*pmd
,
5337 unsigned long addr
, unsigned long end
,
5338 struct mm_walk
*walk
)
5341 struct vm_area_struct
*vma
= walk
->private;
5345 split_huge_page_pmd(walk
->mm
, pmd
);
5347 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
5348 for (; addr
!= end
; addr
+= PAGE_SIZE
) {
5349 pte_t ptent
= *(pte
++);
5350 union mc_target target
;
5353 struct page_cgroup
*pc
;
5359 type
= is_target_pte_for_mc(vma
, addr
, ptent
, &target
);
5361 case MC_TARGET_PAGE
:
5363 if (isolate_lru_page(page
))
5365 pc
= lookup_page_cgroup(page
);
5366 if (!mem_cgroup_move_account(page
, 1, pc
,
5367 mc
.from
, mc
.to
, false)) {
5369 /* we uncharge from mc.from later. */
5372 putback_lru_page(page
);
5373 put
: /* is_target_pte_for_mc() gets the page */
5376 case MC_TARGET_SWAP
:
5378 if (!mem_cgroup_move_swap_account(ent
,
5379 mc
.from
, mc
.to
, false)) {
5381 /* we fixup refcnts and charges later. */
5389 pte_unmap_unlock(pte
- 1, ptl
);
5394 * We have consumed all precharges we got in can_attach().
5395 * We try charge one by one, but don't do any additional
5396 * charges to mc.to if we have failed in charge once in attach()
5399 ret
= mem_cgroup_do_precharge(1);
5407 static void mem_cgroup_move_charge(struct mm_struct
*mm
)
5409 struct vm_area_struct
*vma
;
5411 lru_add_drain_all();
5413 if (unlikely(!down_read_trylock(&mm
->mmap_sem
))) {
5415 * Someone who are holding the mmap_sem might be waiting in
5416 * waitq. So we cancel all extra charges, wake up all waiters,
5417 * and retry. Because we cancel precharges, we might not be able
5418 * to move enough charges, but moving charge is a best-effort
5419 * feature anyway, so it wouldn't be a big problem.
5421 __mem_cgroup_clear_mc();
5425 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
5427 struct mm_walk mem_cgroup_move_charge_walk
= {
5428 .pmd_entry
= mem_cgroup_move_charge_pte_range
,
5432 if (is_vm_hugetlb_page(vma
))
5434 ret
= walk_page_range(vma
->vm_start
, vma
->vm_end
,
5435 &mem_cgroup_move_charge_walk
);
5438 * means we have consumed all precharges and failed in
5439 * doing additional charge. Just abandon here.
5443 up_read(&mm
->mmap_sem
);
5446 static void mem_cgroup_move_task(struct cgroup_subsys
*ss
,
5447 struct cgroup
*cont
,
5448 struct cgroup
*old_cont
,
5449 struct task_struct
*p
)
5451 struct mm_struct
*mm
= get_task_mm(p
);
5455 mem_cgroup_move_charge(mm
);
5460 mem_cgroup_clear_mc();
5462 #else /* !CONFIG_MMU */
5463 static int mem_cgroup_can_attach(struct cgroup_subsys
*ss
,
5464 struct cgroup
*cgroup
,
5465 struct task_struct
*p
)
5469 static void mem_cgroup_cancel_attach(struct cgroup_subsys
*ss
,
5470 struct cgroup
*cgroup
,
5471 struct task_struct
*p
)
5474 static void mem_cgroup_move_task(struct cgroup_subsys
*ss
,
5475 struct cgroup
*cont
,
5476 struct cgroup
*old_cont
,
5477 struct task_struct
*p
)
5482 struct cgroup_subsys mem_cgroup_subsys
= {
5484 .subsys_id
= mem_cgroup_subsys_id
,
5485 .create
= mem_cgroup_create
,
5486 .pre_destroy
= mem_cgroup_pre_destroy
,
5487 .destroy
= mem_cgroup_destroy
,
5488 .populate
= mem_cgroup_populate
,
5489 .can_attach
= mem_cgroup_can_attach
,
5490 .cancel_attach
= mem_cgroup_cancel_attach
,
5491 .attach
= mem_cgroup_move_task
,
5496 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
5497 static int __init
enable_swap_account(char *s
)
5499 /* consider enabled if no parameter or 1 is given */
5500 if (!strcmp(s
, "1"))
5501 really_do_swap_account
= 1;
5502 else if (!strcmp(s
, "0"))
5503 really_do_swap_account
= 0;
5506 __setup("swapaccount=", enable_swap_account
);