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
;
64 static int really_do_swap_account __initdata
= 1; /* for remember boot option*/
66 #define do_swap_account (0)
70 * Per memcg event counter is incremented at every pagein/pageout. This counter
71 * is used for trigger some periodic events. This is straightforward and better
72 * than using jiffies etc. to handle periodic memcg event.
74 * These values will be used as !((event) & ((1 <<(thresh)) - 1))
76 #define THRESHOLDS_EVENTS_THRESH (7) /* once in 128 */
77 #define SOFTLIMIT_EVENTS_THRESH (10) /* once in 1024 */
80 * Statistics for memory cgroup.
82 enum mem_cgroup_stat_index
{
84 * For MEM_CONTAINER_TYPE_ALL, usage = pagecache + rss.
86 MEM_CGROUP_STAT_CACHE
, /* # of pages charged as cache */
87 MEM_CGROUP_STAT_RSS
, /* # of pages charged as anon rss */
88 MEM_CGROUP_STAT_FILE_MAPPED
, /* # of pages charged as file rss */
89 MEM_CGROUP_STAT_PGPGIN_COUNT
, /* # of pages paged in */
90 MEM_CGROUP_STAT_PGPGOUT_COUNT
, /* # of pages paged out */
91 MEM_CGROUP_STAT_SWAPOUT
, /* # of pages, swapped out */
92 MEM_CGROUP_STAT_DATA
, /* end of data requires synchronization */
93 /* incremented at every pagein/pageout */
94 MEM_CGROUP_EVENTS
= MEM_CGROUP_STAT_DATA
,
95 MEM_CGROUP_ON_MOVE
, /* someone is moving account between groups */
97 MEM_CGROUP_STAT_NSTATS
,
100 struct mem_cgroup_stat_cpu
{
101 s64 count
[MEM_CGROUP_STAT_NSTATS
];
105 * per-zone information in memory controller.
107 struct mem_cgroup_per_zone
{
109 * spin_lock to protect the per cgroup LRU
111 struct list_head lists
[NR_LRU_LISTS
];
112 unsigned long count
[NR_LRU_LISTS
];
114 struct zone_reclaim_stat reclaim_stat
;
115 struct rb_node tree_node
; /* RB tree node */
116 unsigned long long usage_in_excess
;/* Set to the value by which */
117 /* the soft limit is exceeded*/
119 struct mem_cgroup
*mem
; /* Back pointer, we cannot */
120 /* use container_of */
122 /* Macro for accessing counter */
123 #define MEM_CGROUP_ZSTAT(mz, idx) ((mz)->count[(idx)])
125 struct mem_cgroup_per_node
{
126 struct mem_cgroup_per_zone zoneinfo
[MAX_NR_ZONES
];
129 struct mem_cgroup_lru_info
{
130 struct mem_cgroup_per_node
*nodeinfo
[MAX_NUMNODES
];
134 * Cgroups above their limits are maintained in a RB-Tree, independent of
135 * their hierarchy representation
138 struct mem_cgroup_tree_per_zone
{
139 struct rb_root rb_root
;
143 struct mem_cgroup_tree_per_node
{
144 struct mem_cgroup_tree_per_zone rb_tree_per_zone
[MAX_NR_ZONES
];
147 struct mem_cgroup_tree
{
148 struct mem_cgroup_tree_per_node
*rb_tree_per_node
[MAX_NUMNODES
];
151 static struct mem_cgroup_tree soft_limit_tree __read_mostly
;
153 struct mem_cgroup_threshold
{
154 struct eventfd_ctx
*eventfd
;
159 struct mem_cgroup_threshold_ary
{
160 /* An array index points to threshold just below usage. */
161 int current_threshold
;
162 /* Size of entries[] */
164 /* Array of thresholds */
165 struct mem_cgroup_threshold entries
[0];
168 struct mem_cgroup_thresholds
{
169 /* Primary thresholds array */
170 struct mem_cgroup_threshold_ary
*primary
;
172 * Spare threshold array.
173 * This is needed to make mem_cgroup_unregister_event() "never fail".
174 * It must be able to store at least primary->size - 1 entries.
176 struct mem_cgroup_threshold_ary
*spare
;
180 struct mem_cgroup_eventfd_list
{
181 struct list_head list
;
182 struct eventfd_ctx
*eventfd
;
185 static void mem_cgroup_threshold(struct mem_cgroup
*mem
);
186 static void mem_cgroup_oom_notify(struct mem_cgroup
*mem
);
189 * The memory controller data structure. The memory controller controls both
190 * page cache and RSS per cgroup. We would eventually like to provide
191 * statistics based on the statistics developed by Rik Van Riel for clock-pro,
192 * to help the administrator determine what knobs to tune.
194 * TODO: Add a water mark for the memory controller. Reclaim will begin when
195 * we hit the water mark. May be even add a low water mark, such that
196 * no reclaim occurs from a cgroup at it's low water mark, this is
197 * a feature that will be implemented much later in the future.
200 struct cgroup_subsys_state css
;
202 * the counter to account for memory usage
204 struct res_counter res
;
206 * the counter to account for mem+swap usage.
208 struct res_counter memsw
;
210 * Per cgroup active and inactive list, similar to the
211 * per zone LRU lists.
213 struct mem_cgroup_lru_info info
;
216 protect against reclaim related member.
218 spinlock_t reclaim_param_lock
;
221 * While reclaiming in a hierarchy, we cache the last child we
224 int last_scanned_child
;
226 * Should the accounting and control be hierarchical, per subtree?
232 unsigned int swappiness
;
233 /* OOM-Killer disable */
234 int oom_kill_disable
;
236 /* set when res.limit == memsw.limit */
237 bool memsw_is_minimum
;
239 /* protect arrays of thresholds */
240 struct mutex thresholds_lock
;
242 /* thresholds for memory usage. RCU-protected */
243 struct mem_cgroup_thresholds thresholds
;
245 /* thresholds for mem+swap usage. RCU-protected */
246 struct mem_cgroup_thresholds memsw_thresholds
;
248 /* For oom notifier event fd */
249 struct list_head oom_notify
;
252 * Should we move charges of a task when a task is moved into this
253 * mem_cgroup ? And what type of charges should we move ?
255 unsigned long move_charge_at_immigrate
;
259 struct mem_cgroup_stat_cpu
*stat
;
261 * used when a cpu is offlined or other synchronizations
262 * See mem_cgroup_read_stat().
264 struct mem_cgroup_stat_cpu nocpu_base
;
265 spinlock_t pcp_counter_lock
;
268 /* Stuffs for move charges at task migration. */
270 * Types of charges to be moved. "move_charge_at_immitgrate" is treated as a
271 * left-shifted bitmap of these types.
274 MOVE_CHARGE_TYPE_ANON
, /* private anonymous page and swap of it */
275 MOVE_CHARGE_TYPE_FILE
, /* file page(including tmpfs) and swap of it */
279 /* "mc" and its members are protected by cgroup_mutex */
280 static struct move_charge_struct
{
281 spinlock_t lock
; /* for from, to, moving_task */
282 struct mem_cgroup
*from
;
283 struct mem_cgroup
*to
;
284 unsigned long precharge
;
285 unsigned long moved_charge
;
286 unsigned long moved_swap
;
287 struct task_struct
*moving_task
; /* a task moving charges */
288 wait_queue_head_t waitq
; /* a waitq for other context */
290 .lock
= __SPIN_LOCK_UNLOCKED(mc
.lock
),
291 .waitq
= __WAIT_QUEUE_HEAD_INITIALIZER(mc
.waitq
),
294 static bool move_anon(void)
296 return test_bit(MOVE_CHARGE_TYPE_ANON
,
297 &mc
.to
->move_charge_at_immigrate
);
300 static bool move_file(void)
302 return test_bit(MOVE_CHARGE_TYPE_FILE
,
303 &mc
.to
->move_charge_at_immigrate
);
307 * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft
308 * limit reclaim to prevent infinite loops, if they ever occur.
310 #define MEM_CGROUP_MAX_RECLAIM_LOOPS (100)
311 #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS (2)
314 MEM_CGROUP_CHARGE_TYPE_CACHE
= 0,
315 MEM_CGROUP_CHARGE_TYPE_MAPPED
,
316 MEM_CGROUP_CHARGE_TYPE_SHMEM
, /* used by page migration of shmem */
317 MEM_CGROUP_CHARGE_TYPE_FORCE
, /* used by force_empty */
318 MEM_CGROUP_CHARGE_TYPE_SWAPOUT
, /* for accounting swapcache */
319 MEM_CGROUP_CHARGE_TYPE_DROP
, /* a page was unused swap cache */
323 /* only for here (for easy reading.) */
324 #define PCGF_CACHE (1UL << PCG_CACHE)
325 #define PCGF_USED (1UL << PCG_USED)
326 #define PCGF_LOCK (1UL << PCG_LOCK)
327 /* Not used, but added here for completeness */
328 #define PCGF_ACCT (1UL << PCG_ACCT)
330 /* for encoding cft->private value on file */
333 #define _OOM_TYPE (2)
334 #define MEMFILE_PRIVATE(x, val) (((x) << 16) | (val))
335 #define MEMFILE_TYPE(val) (((val) >> 16) & 0xffff)
336 #define MEMFILE_ATTR(val) ((val) & 0xffff)
337 /* Used for OOM nofiier */
338 #define OOM_CONTROL (0)
341 * Reclaim flags for mem_cgroup_hierarchical_reclaim
343 #define MEM_CGROUP_RECLAIM_NOSWAP_BIT 0x0
344 #define MEM_CGROUP_RECLAIM_NOSWAP (1 << MEM_CGROUP_RECLAIM_NOSWAP_BIT)
345 #define MEM_CGROUP_RECLAIM_SHRINK_BIT 0x1
346 #define MEM_CGROUP_RECLAIM_SHRINK (1 << MEM_CGROUP_RECLAIM_SHRINK_BIT)
347 #define MEM_CGROUP_RECLAIM_SOFT_BIT 0x2
348 #define MEM_CGROUP_RECLAIM_SOFT (1 << MEM_CGROUP_RECLAIM_SOFT_BIT)
350 static void mem_cgroup_get(struct mem_cgroup
*mem
);
351 static void mem_cgroup_put(struct mem_cgroup
*mem
);
352 static struct mem_cgroup
*parent_mem_cgroup(struct mem_cgroup
*mem
);
353 static void drain_all_stock_async(void);
355 static struct mem_cgroup_per_zone
*
356 mem_cgroup_zoneinfo(struct mem_cgroup
*mem
, int nid
, int zid
)
358 return &mem
->info
.nodeinfo
[nid
]->zoneinfo
[zid
];
361 struct cgroup_subsys_state
*mem_cgroup_css(struct mem_cgroup
*mem
)
366 static struct mem_cgroup_per_zone
*
367 page_cgroup_zoneinfo(struct page_cgroup
*pc
)
369 struct mem_cgroup
*mem
= pc
->mem_cgroup
;
370 int nid
= page_cgroup_nid(pc
);
371 int zid
= page_cgroup_zid(pc
);
376 return mem_cgroup_zoneinfo(mem
, nid
, zid
);
379 static struct mem_cgroup_tree_per_zone
*
380 soft_limit_tree_node_zone(int nid
, int zid
)
382 return &soft_limit_tree
.rb_tree_per_node
[nid
]->rb_tree_per_zone
[zid
];
385 static struct mem_cgroup_tree_per_zone
*
386 soft_limit_tree_from_page(struct page
*page
)
388 int nid
= page_to_nid(page
);
389 int zid
= page_zonenum(page
);
391 return &soft_limit_tree
.rb_tree_per_node
[nid
]->rb_tree_per_zone
[zid
];
395 __mem_cgroup_insert_exceeded(struct mem_cgroup
*mem
,
396 struct mem_cgroup_per_zone
*mz
,
397 struct mem_cgroup_tree_per_zone
*mctz
,
398 unsigned long long new_usage_in_excess
)
400 struct rb_node
**p
= &mctz
->rb_root
.rb_node
;
401 struct rb_node
*parent
= NULL
;
402 struct mem_cgroup_per_zone
*mz_node
;
407 mz
->usage_in_excess
= new_usage_in_excess
;
408 if (!mz
->usage_in_excess
)
412 mz_node
= rb_entry(parent
, struct mem_cgroup_per_zone
,
414 if (mz
->usage_in_excess
< mz_node
->usage_in_excess
)
417 * We can't avoid mem cgroups that are over their soft
418 * limit by the same amount
420 else if (mz
->usage_in_excess
>= mz_node
->usage_in_excess
)
423 rb_link_node(&mz
->tree_node
, parent
, p
);
424 rb_insert_color(&mz
->tree_node
, &mctz
->rb_root
);
429 __mem_cgroup_remove_exceeded(struct mem_cgroup
*mem
,
430 struct mem_cgroup_per_zone
*mz
,
431 struct mem_cgroup_tree_per_zone
*mctz
)
435 rb_erase(&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
)
444 spin_lock(&mctz
->lock
);
445 __mem_cgroup_remove_exceeded(mem
, mz
, mctz
);
446 spin_unlock(&mctz
->lock
);
450 static void mem_cgroup_update_tree(struct mem_cgroup
*mem
, struct page
*page
)
452 unsigned long long excess
;
453 struct mem_cgroup_per_zone
*mz
;
454 struct mem_cgroup_tree_per_zone
*mctz
;
455 int nid
= page_to_nid(page
);
456 int zid
= page_zonenum(page
);
457 mctz
= soft_limit_tree_from_page(page
);
460 * Necessary to update all ancestors when hierarchy is used.
461 * because their event counter is not touched.
463 for (; mem
; mem
= parent_mem_cgroup(mem
)) {
464 mz
= mem_cgroup_zoneinfo(mem
, nid
, zid
);
465 excess
= res_counter_soft_limit_excess(&mem
->res
);
467 * We have to update the tree if mz is on RB-tree or
468 * mem is over its softlimit.
470 if (excess
|| mz
->on_tree
) {
471 spin_lock(&mctz
->lock
);
472 /* if on-tree, remove it */
474 __mem_cgroup_remove_exceeded(mem
, mz
, mctz
);
476 * Insert again. mz->usage_in_excess will be updated.
477 * If excess is 0, no tree ops.
479 __mem_cgroup_insert_exceeded(mem
, mz
, mctz
, excess
);
480 spin_unlock(&mctz
->lock
);
485 static void mem_cgroup_remove_from_trees(struct mem_cgroup
*mem
)
488 struct mem_cgroup_per_zone
*mz
;
489 struct mem_cgroup_tree_per_zone
*mctz
;
491 for_each_node_state(node
, N_POSSIBLE
) {
492 for (zone
= 0; zone
< MAX_NR_ZONES
; zone
++) {
493 mz
= mem_cgroup_zoneinfo(mem
, node
, zone
);
494 mctz
= soft_limit_tree_node_zone(node
, zone
);
495 mem_cgroup_remove_exceeded(mem
, mz
, mctz
);
500 static inline unsigned long mem_cgroup_get_excess(struct mem_cgroup
*mem
)
502 return res_counter_soft_limit_excess(&mem
->res
) >> PAGE_SHIFT
;
505 static struct mem_cgroup_per_zone
*
506 __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone
*mctz
)
508 struct rb_node
*rightmost
= NULL
;
509 struct mem_cgroup_per_zone
*mz
;
513 rightmost
= rb_last(&mctz
->rb_root
);
515 goto done
; /* Nothing to reclaim from */
517 mz
= rb_entry(rightmost
, struct mem_cgroup_per_zone
, tree_node
);
519 * Remove the node now but someone else can add it back,
520 * we will to add it back at the end of reclaim to its correct
521 * position in the tree.
523 __mem_cgroup_remove_exceeded(mz
->mem
, mz
, mctz
);
524 if (!res_counter_soft_limit_excess(&mz
->mem
->res
) ||
525 !css_tryget(&mz
->mem
->css
))
531 static struct mem_cgroup_per_zone
*
532 mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone
*mctz
)
534 struct mem_cgroup_per_zone
*mz
;
536 spin_lock(&mctz
->lock
);
537 mz
= __mem_cgroup_largest_soft_limit_node(mctz
);
538 spin_unlock(&mctz
->lock
);
543 * Implementation Note: reading percpu statistics for memcg.
545 * Both of vmstat[] and percpu_counter has threshold and do periodic
546 * synchronization to implement "quick" read. There are trade-off between
547 * reading cost and precision of value. Then, we may have a chance to implement
548 * a periodic synchronizion of counter in memcg's counter.
550 * But this _read() function is used for user interface now. The user accounts
551 * memory usage by memory cgroup and he _always_ requires exact value because
552 * he accounts memory. Even if we provide quick-and-fuzzy read, we always
553 * have to visit all online cpus and make sum. So, for now, unnecessary
554 * synchronization is not implemented. (just implemented for cpu hotplug)
556 * If there are kernel internal actions which can make use of some not-exact
557 * value, and reading all cpu value can be performance bottleneck in some
558 * common workload, threashold and synchonization as vmstat[] should be
561 static s64
mem_cgroup_read_stat(struct mem_cgroup
*mem
,
562 enum mem_cgroup_stat_index idx
)
568 for_each_online_cpu(cpu
)
569 val
+= per_cpu(mem
->stat
->count
[idx
], cpu
);
570 #ifdef CONFIG_HOTPLUG_CPU
571 spin_lock(&mem
->pcp_counter_lock
);
572 val
+= mem
->nocpu_base
.count
[idx
];
573 spin_unlock(&mem
->pcp_counter_lock
);
579 static s64
mem_cgroup_local_usage(struct mem_cgroup
*mem
)
583 ret
= mem_cgroup_read_stat(mem
, MEM_CGROUP_STAT_RSS
);
584 ret
+= mem_cgroup_read_stat(mem
, MEM_CGROUP_STAT_CACHE
);
588 static void mem_cgroup_swap_statistics(struct mem_cgroup
*mem
,
591 int val
= (charge
) ? 1 : -1;
592 this_cpu_add(mem
->stat
->count
[MEM_CGROUP_STAT_SWAPOUT
], val
);
595 static void mem_cgroup_charge_statistics(struct mem_cgroup
*mem
,
596 struct page_cgroup
*pc
,
599 int val
= (charge
) ? 1 : -1;
603 if (PageCgroupCache(pc
))
604 __this_cpu_add(mem
->stat
->count
[MEM_CGROUP_STAT_CACHE
], val
);
606 __this_cpu_add(mem
->stat
->count
[MEM_CGROUP_STAT_RSS
], val
);
609 __this_cpu_inc(mem
->stat
->count
[MEM_CGROUP_STAT_PGPGIN_COUNT
]);
611 __this_cpu_inc(mem
->stat
->count
[MEM_CGROUP_STAT_PGPGOUT_COUNT
]);
612 __this_cpu_inc(mem
->stat
->count
[MEM_CGROUP_EVENTS
]);
617 static unsigned long mem_cgroup_get_local_zonestat(struct mem_cgroup
*mem
,
621 struct mem_cgroup_per_zone
*mz
;
624 for_each_online_node(nid
)
625 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++) {
626 mz
= mem_cgroup_zoneinfo(mem
, nid
, zid
);
627 total
+= MEM_CGROUP_ZSTAT(mz
, idx
);
632 static bool __memcg_event_check(struct mem_cgroup
*mem
, int event_mask_shift
)
636 val
= this_cpu_read(mem
->stat
->count
[MEM_CGROUP_EVENTS
]);
638 return !(val
& ((1 << event_mask_shift
) - 1));
642 * Check events in order.
645 static void memcg_check_events(struct mem_cgroup
*mem
, struct page
*page
)
647 /* threshold event is triggered in finer grain than soft limit */
648 if (unlikely(__memcg_event_check(mem
, THRESHOLDS_EVENTS_THRESH
))) {
649 mem_cgroup_threshold(mem
);
650 if (unlikely(__memcg_event_check(mem
, SOFTLIMIT_EVENTS_THRESH
)))
651 mem_cgroup_update_tree(mem
, page
);
655 static struct mem_cgroup
*mem_cgroup_from_cont(struct cgroup
*cont
)
657 return container_of(cgroup_subsys_state(cont
,
658 mem_cgroup_subsys_id
), struct mem_cgroup
,
662 struct mem_cgroup
*mem_cgroup_from_task(struct task_struct
*p
)
665 * mm_update_next_owner() may clear mm->owner to NULL
666 * if it races with swapoff, page migration, etc.
667 * So this can be called with p == NULL.
672 return container_of(task_subsys_state(p
, mem_cgroup_subsys_id
),
673 struct mem_cgroup
, css
);
676 static struct mem_cgroup
*try_get_mem_cgroup_from_mm(struct mm_struct
*mm
)
678 struct mem_cgroup
*mem
= NULL
;
683 * Because we have no locks, mm->owner's may be being moved to other
684 * cgroup. We use css_tryget() here even if this looks
685 * pessimistic (rather than adding locks here).
689 mem
= mem_cgroup_from_task(rcu_dereference(mm
->owner
));
692 } while (!css_tryget(&mem
->css
));
697 /* The caller has to guarantee "mem" exists before calling this */
698 static struct mem_cgroup
*mem_cgroup_start_loop(struct mem_cgroup
*mem
)
700 struct cgroup_subsys_state
*css
;
703 if (!mem
) /* ROOT cgroup has the smallest ID */
704 return root_mem_cgroup
; /*css_put/get against root is ignored*/
705 if (!mem
->use_hierarchy
) {
706 if (css_tryget(&mem
->css
))
712 * searching a memory cgroup which has the smallest ID under given
713 * ROOT cgroup. (ID >= 1)
715 css
= css_get_next(&mem_cgroup_subsys
, 1, &mem
->css
, &found
);
716 if (css
&& css_tryget(css
))
717 mem
= container_of(css
, struct mem_cgroup
, css
);
724 static struct mem_cgroup
*mem_cgroup_get_next(struct mem_cgroup
*iter
,
725 struct mem_cgroup
*root
,
728 int nextid
= css_id(&iter
->css
) + 1;
731 struct cgroup_subsys_state
*css
;
733 hierarchy_used
= iter
->use_hierarchy
;
736 /* If no ROOT, walk all, ignore hierarchy */
737 if (!cond
|| (root
&& !hierarchy_used
))
741 root
= root_mem_cgroup
;
747 css
= css_get_next(&mem_cgroup_subsys
, nextid
,
749 if (css
&& css_tryget(css
))
750 iter
= container_of(css
, struct mem_cgroup
, css
);
752 /* If css is NULL, no more cgroups will be found */
754 } while (css
&& !iter
);
759 * for_eacn_mem_cgroup_tree() for visiting all cgroup under tree. Please
760 * be careful that "break" loop is not allowed. We have reference count.
761 * Instead of that modify "cond" to be false and "continue" to exit the loop.
763 #define for_each_mem_cgroup_tree_cond(iter, root, cond) \
764 for (iter = mem_cgroup_start_loop(root);\
766 iter = mem_cgroup_get_next(iter, root, cond))
768 #define for_each_mem_cgroup_tree(iter, root) \
769 for_each_mem_cgroup_tree_cond(iter, root, true)
771 #define for_each_mem_cgroup_all(iter) \
772 for_each_mem_cgroup_tree_cond(iter, NULL, true)
775 static inline bool mem_cgroup_is_root(struct mem_cgroup
*mem
)
777 return (mem
== root_mem_cgroup
);
781 * Following LRU functions are allowed to be used without PCG_LOCK.
782 * Operations are called by routine of global LRU independently from memcg.
783 * What we have to take care of here is validness of pc->mem_cgroup.
785 * Changes to pc->mem_cgroup happens when
788 * In typical case, "charge" is done before add-to-lru. Exception is SwapCache.
789 * It is added to LRU before charge.
790 * If PCG_USED bit is not set, page_cgroup is not added to this private LRU.
791 * When moving account, the page is not on LRU. It's isolated.
794 void mem_cgroup_del_lru_list(struct page
*page
, enum lru_list lru
)
796 struct page_cgroup
*pc
;
797 struct mem_cgroup_per_zone
*mz
;
799 if (mem_cgroup_disabled())
801 pc
= lookup_page_cgroup(page
);
802 /* can happen while we handle swapcache. */
803 if (!TestClearPageCgroupAcctLRU(pc
))
805 VM_BUG_ON(!pc
->mem_cgroup
);
807 * We don't check PCG_USED bit. It's cleared when the "page" is finally
808 * removed from global LRU.
810 mz
= page_cgroup_zoneinfo(pc
);
811 MEM_CGROUP_ZSTAT(mz
, lru
) -= 1;
812 if (mem_cgroup_is_root(pc
->mem_cgroup
))
814 VM_BUG_ON(list_empty(&pc
->lru
));
815 list_del_init(&pc
->lru
);
819 void mem_cgroup_del_lru(struct page
*page
)
821 mem_cgroup_del_lru_list(page
, page_lru(page
));
824 void mem_cgroup_rotate_lru_list(struct page
*page
, enum lru_list lru
)
826 struct mem_cgroup_per_zone
*mz
;
827 struct page_cgroup
*pc
;
829 if (mem_cgroup_disabled())
832 pc
= lookup_page_cgroup(page
);
834 * Used bit is set without atomic ops but after smp_wmb().
835 * For making pc->mem_cgroup visible, insert smp_rmb() here.
838 /* unused or root page is not rotated. */
839 if (!PageCgroupUsed(pc
) || mem_cgroup_is_root(pc
->mem_cgroup
))
841 mz
= page_cgroup_zoneinfo(pc
);
842 list_move(&pc
->lru
, &mz
->lists
[lru
]);
845 void mem_cgroup_add_lru_list(struct page
*page
, enum lru_list lru
)
847 struct page_cgroup
*pc
;
848 struct mem_cgroup_per_zone
*mz
;
850 if (mem_cgroup_disabled())
852 pc
= lookup_page_cgroup(page
);
853 VM_BUG_ON(PageCgroupAcctLRU(pc
));
855 * Used bit is set without atomic ops but after smp_wmb().
856 * For making pc->mem_cgroup visible, insert smp_rmb() here.
859 if (!PageCgroupUsed(pc
))
862 mz
= page_cgroup_zoneinfo(pc
);
863 MEM_CGROUP_ZSTAT(mz
, lru
) += 1;
864 SetPageCgroupAcctLRU(pc
);
865 if (mem_cgroup_is_root(pc
->mem_cgroup
))
867 list_add(&pc
->lru
, &mz
->lists
[lru
]);
871 * At handling SwapCache, pc->mem_cgroup may be changed while it's linked to
872 * lru because the page may.be reused after it's fully uncharged (because of
873 * SwapCache behavior).To handle that, unlink page_cgroup from LRU when charge
874 * it again. This function is only used to charge SwapCache. It's done under
875 * lock_page and expected that zone->lru_lock is never held.
877 static void mem_cgroup_lru_del_before_commit_swapcache(struct page
*page
)
880 struct zone
*zone
= page_zone(page
);
881 struct page_cgroup
*pc
= lookup_page_cgroup(page
);
883 spin_lock_irqsave(&zone
->lru_lock
, flags
);
885 * Forget old LRU when this page_cgroup is *not* used. This Used bit
886 * is guarded by lock_page() because the page is SwapCache.
888 if (!PageCgroupUsed(pc
))
889 mem_cgroup_del_lru_list(page
, page_lru(page
));
890 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
893 static void mem_cgroup_lru_add_after_commit_swapcache(struct page
*page
)
896 struct zone
*zone
= page_zone(page
);
897 struct page_cgroup
*pc
= lookup_page_cgroup(page
);
899 spin_lock_irqsave(&zone
->lru_lock
, flags
);
900 /* link when the page is linked to LRU but page_cgroup isn't */
901 if (PageLRU(page
) && !PageCgroupAcctLRU(pc
))
902 mem_cgroup_add_lru_list(page
, page_lru(page
));
903 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
907 void mem_cgroup_move_lists(struct page
*page
,
908 enum lru_list from
, enum lru_list to
)
910 if (mem_cgroup_disabled())
912 mem_cgroup_del_lru_list(page
, from
);
913 mem_cgroup_add_lru_list(page
, to
);
916 int task_in_mem_cgroup(struct task_struct
*task
, const struct mem_cgroup
*mem
)
919 struct mem_cgroup
*curr
= NULL
;
920 struct task_struct
*p
;
922 p
= find_lock_task_mm(task
);
925 curr
= try_get_mem_cgroup_from_mm(p
->mm
);
930 * We should check use_hierarchy of "mem" not "curr". Because checking
931 * use_hierarchy of "curr" here make this function true if hierarchy is
932 * enabled in "curr" and "curr" is a child of "mem" in *cgroup*
933 * hierarchy(even if use_hierarchy is disabled in "mem").
935 if (mem
->use_hierarchy
)
936 ret
= css_is_ancestor(&curr
->css
, &mem
->css
);
943 static int calc_inactive_ratio(struct mem_cgroup
*memcg
, unsigned long *present_pages
)
945 unsigned long active
;
946 unsigned long inactive
;
948 unsigned long inactive_ratio
;
950 inactive
= mem_cgroup_get_local_zonestat(memcg
, LRU_INACTIVE_ANON
);
951 active
= mem_cgroup_get_local_zonestat(memcg
, LRU_ACTIVE_ANON
);
953 gb
= (inactive
+ active
) >> (30 - PAGE_SHIFT
);
955 inactive_ratio
= int_sqrt(10 * gb
);
960 present_pages
[0] = inactive
;
961 present_pages
[1] = active
;
964 return inactive_ratio
;
967 int mem_cgroup_inactive_anon_is_low(struct mem_cgroup
*memcg
)
969 unsigned long active
;
970 unsigned long inactive
;
971 unsigned long present_pages
[2];
972 unsigned long inactive_ratio
;
974 inactive_ratio
= calc_inactive_ratio(memcg
, present_pages
);
976 inactive
= present_pages
[0];
977 active
= present_pages
[1];
979 if (inactive
* inactive_ratio
< active
)
985 int mem_cgroup_inactive_file_is_low(struct mem_cgroup
*memcg
)
987 unsigned long active
;
988 unsigned long inactive
;
990 inactive
= mem_cgroup_get_local_zonestat(memcg
, LRU_INACTIVE_FILE
);
991 active
= mem_cgroup_get_local_zonestat(memcg
, LRU_ACTIVE_FILE
);
993 return (active
> inactive
);
996 unsigned long mem_cgroup_zone_nr_pages(struct mem_cgroup
*memcg
,
1000 int nid
= zone_to_nid(zone
);
1001 int zid
= zone_idx(zone
);
1002 struct mem_cgroup_per_zone
*mz
= mem_cgroup_zoneinfo(memcg
, nid
, zid
);
1004 return MEM_CGROUP_ZSTAT(mz
, lru
);
1007 struct zone_reclaim_stat
*mem_cgroup_get_reclaim_stat(struct mem_cgroup
*memcg
,
1010 int nid
= zone_to_nid(zone
);
1011 int zid
= zone_idx(zone
);
1012 struct mem_cgroup_per_zone
*mz
= mem_cgroup_zoneinfo(memcg
, nid
, zid
);
1014 return &mz
->reclaim_stat
;
1017 struct zone_reclaim_stat
*
1018 mem_cgroup_get_reclaim_stat_from_page(struct page
*page
)
1020 struct page_cgroup
*pc
;
1021 struct mem_cgroup_per_zone
*mz
;
1023 if (mem_cgroup_disabled())
1026 pc
= lookup_page_cgroup(page
);
1028 * Used bit is set without atomic ops but after smp_wmb().
1029 * For making pc->mem_cgroup visible, insert smp_rmb() here.
1032 if (!PageCgroupUsed(pc
))
1035 mz
= page_cgroup_zoneinfo(pc
);
1039 return &mz
->reclaim_stat
;
1042 unsigned long mem_cgroup_isolate_pages(unsigned long nr_to_scan
,
1043 struct list_head
*dst
,
1044 unsigned long *scanned
, int order
,
1045 int mode
, struct zone
*z
,
1046 struct mem_cgroup
*mem_cont
,
1047 int active
, int file
)
1049 unsigned long nr_taken
= 0;
1053 struct list_head
*src
;
1054 struct page_cgroup
*pc
, *tmp
;
1055 int nid
= zone_to_nid(z
);
1056 int zid
= zone_idx(z
);
1057 struct mem_cgroup_per_zone
*mz
;
1058 int lru
= LRU_FILE
* file
+ active
;
1062 mz
= mem_cgroup_zoneinfo(mem_cont
, nid
, zid
);
1063 src
= &mz
->lists
[lru
];
1066 list_for_each_entry_safe_reverse(pc
, tmp
, src
, lru
) {
1067 if (scan
>= nr_to_scan
)
1071 if (unlikely(!PageCgroupUsed(pc
)))
1073 if (unlikely(!PageLRU(page
)))
1077 ret
= __isolate_lru_page(page
, mode
, file
);
1080 list_move(&page
->lru
, dst
);
1081 mem_cgroup_del_lru(page
);
1085 /* we don't affect global LRU but rotate in our LRU */
1086 mem_cgroup_rotate_lru_list(page
, page_lru(page
));
1095 trace_mm_vmscan_memcg_isolate(0, nr_to_scan
, scan
, nr_taken
,
1101 #define mem_cgroup_from_res_counter(counter, member) \
1102 container_of(counter, struct mem_cgroup, member)
1104 static bool mem_cgroup_check_under_limit(struct mem_cgroup
*mem
)
1106 if (do_swap_account
) {
1107 if (res_counter_check_under_limit(&mem
->res
) &&
1108 res_counter_check_under_limit(&mem
->memsw
))
1111 if (res_counter_check_under_limit(&mem
->res
))
1116 static unsigned int get_swappiness(struct mem_cgroup
*memcg
)
1118 struct cgroup
*cgrp
= memcg
->css
.cgroup
;
1119 unsigned int swappiness
;
1122 if (cgrp
->parent
== NULL
)
1123 return vm_swappiness
;
1125 spin_lock(&memcg
->reclaim_param_lock
);
1126 swappiness
= memcg
->swappiness
;
1127 spin_unlock(&memcg
->reclaim_param_lock
);
1132 static void mem_cgroup_start_move(struct mem_cgroup
*mem
)
1137 spin_lock(&mem
->pcp_counter_lock
);
1138 for_each_online_cpu(cpu
)
1139 per_cpu(mem
->stat
->count
[MEM_CGROUP_ON_MOVE
], cpu
) += 1;
1140 mem
->nocpu_base
.count
[MEM_CGROUP_ON_MOVE
] += 1;
1141 spin_unlock(&mem
->pcp_counter_lock
);
1147 static void mem_cgroup_end_move(struct mem_cgroup
*mem
)
1154 spin_lock(&mem
->pcp_counter_lock
);
1155 for_each_online_cpu(cpu
)
1156 per_cpu(mem
->stat
->count
[MEM_CGROUP_ON_MOVE
], cpu
) -= 1;
1157 mem
->nocpu_base
.count
[MEM_CGROUP_ON_MOVE
] -= 1;
1158 spin_unlock(&mem
->pcp_counter_lock
);
1162 * 2 routines for checking "mem" is under move_account() or not.
1164 * mem_cgroup_stealed() - checking a cgroup is mc.from or not. This is used
1165 * for avoiding race in accounting. If true,
1166 * pc->mem_cgroup may be overwritten.
1168 * mem_cgroup_under_move() - checking a cgroup is mc.from or mc.to or
1169 * under hierarchy of moving cgroups. This is for
1170 * waiting at hith-memory prressure caused by "move".
1173 static bool mem_cgroup_stealed(struct mem_cgroup
*mem
)
1175 VM_BUG_ON(!rcu_read_lock_held());
1176 return this_cpu_read(mem
->stat
->count
[MEM_CGROUP_ON_MOVE
]) > 0;
1179 static bool mem_cgroup_under_move(struct mem_cgroup
*mem
)
1181 struct mem_cgroup
*from
;
1182 struct mem_cgroup
*to
;
1185 * Unlike task_move routines, we access mc.to, mc.from not under
1186 * mutual exclusion by cgroup_mutex. Here, we take spinlock instead.
1188 spin_lock(&mc
.lock
);
1193 if (from
== mem
|| to
== mem
1194 || (mem
->use_hierarchy
&& css_is_ancestor(&from
->css
, &mem
->css
))
1195 || (mem
->use_hierarchy
&& css_is_ancestor(&to
->css
, &mem
->css
)))
1198 spin_unlock(&mc
.lock
);
1202 static bool mem_cgroup_wait_acct_move(struct mem_cgroup
*mem
)
1204 if (mc
.moving_task
&& current
!= mc
.moving_task
) {
1205 if (mem_cgroup_under_move(mem
)) {
1207 prepare_to_wait(&mc
.waitq
, &wait
, TASK_INTERRUPTIBLE
);
1208 /* moving charge context might have finished. */
1211 finish_wait(&mc
.waitq
, &wait
);
1219 * mem_cgroup_print_oom_info: Called from OOM with tasklist_lock held in read mode.
1220 * @memcg: The memory cgroup that went over limit
1221 * @p: Task that is going to be killed
1223 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1226 void mem_cgroup_print_oom_info(struct mem_cgroup
*memcg
, struct task_struct
*p
)
1228 struct cgroup
*task_cgrp
;
1229 struct cgroup
*mem_cgrp
;
1231 * Need a buffer in BSS, can't rely on allocations. The code relies
1232 * on the assumption that OOM is serialized for memory controller.
1233 * If this assumption is broken, revisit this code.
1235 static char memcg_name
[PATH_MAX
];
1244 mem_cgrp
= memcg
->css
.cgroup
;
1245 task_cgrp
= task_cgroup(p
, mem_cgroup_subsys_id
);
1247 ret
= cgroup_path(task_cgrp
, memcg_name
, PATH_MAX
);
1250 * Unfortunately, we are unable to convert to a useful name
1251 * But we'll still print out the usage information
1258 printk(KERN_INFO
"Task in %s killed", memcg_name
);
1261 ret
= cgroup_path(mem_cgrp
, memcg_name
, PATH_MAX
);
1269 * Continues from above, so we don't need an KERN_ level
1271 printk(KERN_CONT
" as a result of limit of %s\n", memcg_name
);
1274 printk(KERN_INFO
"memory: usage %llukB, limit %llukB, failcnt %llu\n",
1275 res_counter_read_u64(&memcg
->res
, RES_USAGE
) >> 10,
1276 res_counter_read_u64(&memcg
->res
, RES_LIMIT
) >> 10,
1277 res_counter_read_u64(&memcg
->res
, RES_FAILCNT
));
1278 printk(KERN_INFO
"memory+swap: usage %llukB, limit %llukB, "
1280 res_counter_read_u64(&memcg
->memsw
, RES_USAGE
) >> 10,
1281 res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
) >> 10,
1282 res_counter_read_u64(&memcg
->memsw
, RES_FAILCNT
));
1286 * This function returns the number of memcg under hierarchy tree. Returns
1287 * 1(self count) if no children.
1289 static int mem_cgroup_count_children(struct mem_cgroup
*mem
)
1292 struct mem_cgroup
*iter
;
1294 for_each_mem_cgroup_tree(iter
, mem
)
1300 * Return the memory (and swap, if configured) limit for a memcg.
1302 u64
mem_cgroup_get_limit(struct mem_cgroup
*memcg
)
1307 limit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
) +
1309 memsw
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
1311 * If memsw is finite and limits the amount of swap space available
1312 * to this memcg, return that limit.
1314 return min(limit
, memsw
);
1318 * Visit the first child (need not be the first child as per the ordering
1319 * of the cgroup list, since we track last_scanned_child) of @mem and use
1320 * that to reclaim free pages from.
1322 static struct mem_cgroup
*
1323 mem_cgroup_select_victim(struct mem_cgroup
*root_mem
)
1325 struct mem_cgroup
*ret
= NULL
;
1326 struct cgroup_subsys_state
*css
;
1329 if (!root_mem
->use_hierarchy
) {
1330 css_get(&root_mem
->css
);
1336 nextid
= root_mem
->last_scanned_child
+ 1;
1337 css
= css_get_next(&mem_cgroup_subsys
, nextid
, &root_mem
->css
,
1339 if (css
&& css_tryget(css
))
1340 ret
= container_of(css
, struct mem_cgroup
, css
);
1343 /* Updates scanning parameter */
1344 spin_lock(&root_mem
->reclaim_param_lock
);
1346 /* this means start scan from ID:1 */
1347 root_mem
->last_scanned_child
= 0;
1349 root_mem
->last_scanned_child
= found
;
1350 spin_unlock(&root_mem
->reclaim_param_lock
);
1357 * Scan the hierarchy if needed to reclaim memory. We remember the last child
1358 * we reclaimed from, so that we don't end up penalizing one child extensively
1359 * based on its position in the children list.
1361 * root_mem is the original ancestor that we've been reclaim from.
1363 * We give up and return to the caller when we visit root_mem twice.
1364 * (other groups can be removed while we're walking....)
1366 * If shrink==true, for avoiding to free too much, this returns immedieately.
1368 static int mem_cgroup_hierarchical_reclaim(struct mem_cgroup
*root_mem
,
1371 unsigned long reclaim_options
)
1373 struct mem_cgroup
*victim
;
1376 bool noswap
= reclaim_options
& MEM_CGROUP_RECLAIM_NOSWAP
;
1377 bool shrink
= reclaim_options
& MEM_CGROUP_RECLAIM_SHRINK
;
1378 bool check_soft
= reclaim_options
& MEM_CGROUP_RECLAIM_SOFT
;
1379 unsigned long excess
= mem_cgroup_get_excess(root_mem
);
1381 /* If memsw_is_minimum==1, swap-out is of-no-use. */
1382 if (root_mem
->memsw_is_minimum
)
1386 victim
= mem_cgroup_select_victim(root_mem
);
1387 if (victim
== root_mem
) {
1390 drain_all_stock_async();
1393 * If we have not been able to reclaim
1394 * anything, it might because there are
1395 * no reclaimable pages under this hierarchy
1397 if (!check_soft
|| !total
) {
1398 css_put(&victim
->css
);
1402 * We want to do more targetted reclaim.
1403 * excess >> 2 is not to excessive so as to
1404 * reclaim too much, nor too less that we keep
1405 * coming back to reclaim from this cgroup
1407 if (total
>= (excess
>> 2) ||
1408 (loop
> MEM_CGROUP_MAX_RECLAIM_LOOPS
)) {
1409 css_put(&victim
->css
);
1414 if (!mem_cgroup_local_usage(victim
)) {
1415 /* this cgroup's local usage == 0 */
1416 css_put(&victim
->css
);
1419 /* we use swappiness of local cgroup */
1421 ret
= mem_cgroup_shrink_node_zone(victim
, gfp_mask
,
1422 noswap
, get_swappiness(victim
), zone
);
1424 ret
= try_to_free_mem_cgroup_pages(victim
, gfp_mask
,
1425 noswap
, get_swappiness(victim
));
1426 css_put(&victim
->css
);
1428 * At shrinking usage, we can't check we should stop here or
1429 * reclaim more. It's depends on callers. last_scanned_child
1430 * will work enough for keeping fairness under tree.
1436 if (res_counter_check_under_soft_limit(&root_mem
->res
))
1438 } else if (mem_cgroup_check_under_limit(root_mem
))
1445 * Check OOM-Killer is already running under our hierarchy.
1446 * If someone is running, return false.
1448 static bool mem_cgroup_oom_lock(struct mem_cgroup
*mem
)
1450 int x
, lock_count
= 0;
1451 struct mem_cgroup
*iter
;
1453 for_each_mem_cgroup_tree(iter
, mem
) {
1454 x
= atomic_inc_return(&iter
->oom_lock
);
1455 lock_count
= max(x
, lock_count
);
1458 if (lock_count
== 1)
1463 static int mem_cgroup_oom_unlock(struct mem_cgroup
*mem
)
1465 struct mem_cgroup
*iter
;
1468 * When a new child is created while the hierarchy is under oom,
1469 * mem_cgroup_oom_lock() may not be called. We have to use
1470 * atomic_add_unless() here.
1472 for_each_mem_cgroup_tree(iter
, mem
)
1473 atomic_add_unless(&iter
->oom_lock
, -1, 0);
1478 static DEFINE_MUTEX(memcg_oom_mutex
);
1479 static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq
);
1481 struct oom_wait_info
{
1482 struct mem_cgroup
*mem
;
1486 static int memcg_oom_wake_function(wait_queue_t
*wait
,
1487 unsigned mode
, int sync
, void *arg
)
1489 struct mem_cgroup
*wake_mem
= (struct mem_cgroup
*)arg
;
1490 struct oom_wait_info
*oom_wait_info
;
1492 oom_wait_info
= container_of(wait
, struct oom_wait_info
, wait
);
1494 if (oom_wait_info
->mem
== wake_mem
)
1496 /* if no hierarchy, no match */
1497 if (!oom_wait_info
->mem
->use_hierarchy
|| !wake_mem
->use_hierarchy
)
1500 * Both of oom_wait_info->mem and wake_mem are stable under us.
1501 * Then we can use css_is_ancestor without taking care of RCU.
1503 if (!css_is_ancestor(&oom_wait_info
->mem
->css
, &wake_mem
->css
) &&
1504 !css_is_ancestor(&wake_mem
->css
, &oom_wait_info
->mem
->css
))
1508 return autoremove_wake_function(wait
, mode
, sync
, arg
);
1511 static void memcg_wakeup_oom(struct mem_cgroup
*mem
)
1513 /* for filtering, pass "mem" as argument. */
1514 __wake_up(&memcg_oom_waitq
, TASK_NORMAL
, 0, mem
);
1517 static void memcg_oom_recover(struct mem_cgroup
*mem
)
1519 if (mem
&& atomic_read(&mem
->oom_lock
))
1520 memcg_wakeup_oom(mem
);
1524 * try to call OOM killer. returns false if we should exit memory-reclaim loop.
1526 bool mem_cgroup_handle_oom(struct mem_cgroup
*mem
, gfp_t mask
)
1528 struct oom_wait_info owait
;
1529 bool locked
, need_to_kill
;
1532 owait
.wait
.flags
= 0;
1533 owait
.wait
.func
= memcg_oom_wake_function
;
1534 owait
.wait
.private = current
;
1535 INIT_LIST_HEAD(&owait
.wait
.task_list
);
1536 need_to_kill
= true;
1537 /* At first, try to OOM lock hierarchy under mem.*/
1538 mutex_lock(&memcg_oom_mutex
);
1539 locked
= mem_cgroup_oom_lock(mem
);
1541 * Even if signal_pending(), we can't quit charge() loop without
1542 * accounting. So, UNINTERRUPTIBLE is appropriate. But SIGKILL
1543 * under OOM is always welcomed, use TASK_KILLABLE here.
1545 prepare_to_wait(&memcg_oom_waitq
, &owait
.wait
, TASK_KILLABLE
);
1546 if (!locked
|| mem
->oom_kill_disable
)
1547 need_to_kill
= false;
1549 mem_cgroup_oom_notify(mem
);
1550 mutex_unlock(&memcg_oom_mutex
);
1553 finish_wait(&memcg_oom_waitq
, &owait
.wait
);
1554 mem_cgroup_out_of_memory(mem
, mask
);
1557 finish_wait(&memcg_oom_waitq
, &owait
.wait
);
1559 mutex_lock(&memcg_oom_mutex
);
1560 mem_cgroup_oom_unlock(mem
);
1561 memcg_wakeup_oom(mem
);
1562 mutex_unlock(&memcg_oom_mutex
);
1564 if (test_thread_flag(TIF_MEMDIE
) || fatal_signal_pending(current
))
1566 /* Give chance to dying process */
1567 schedule_timeout(1);
1572 * Currently used to update mapped file statistics, but the routine can be
1573 * generalized to update other statistics as well.
1575 * Notes: Race condition
1577 * We usually use page_cgroup_lock() for accessing page_cgroup member but
1578 * it tends to be costly. But considering some conditions, we doesn't need
1579 * to do so _always_.
1581 * Considering "charge", lock_page_cgroup() is not required because all
1582 * file-stat operations happen after a page is attached to radix-tree. There
1583 * are no race with "charge".
1585 * Considering "uncharge", we know that memcg doesn't clear pc->mem_cgroup
1586 * at "uncharge" intentionally. So, we always see valid pc->mem_cgroup even
1587 * if there are race with "uncharge". Statistics itself is properly handled
1590 * Considering "move", this is an only case we see a race. To make the race
1591 * small, we check MEM_CGROUP_ON_MOVE percpu value and detect there are
1592 * possibility of race condition. If there is, we take a lock.
1594 void mem_cgroup_update_file_mapped(struct page
*page
, int val
)
1596 struct mem_cgroup
*mem
;
1597 struct page_cgroup
*pc
= lookup_page_cgroup(page
);
1598 bool need_unlock
= false;
1604 mem
= pc
->mem_cgroup
;
1605 if (unlikely(!mem
|| !PageCgroupUsed(pc
)))
1607 /* pc->mem_cgroup is unstable ? */
1608 if (unlikely(mem_cgroup_stealed(mem
))) {
1609 /* take a lock against to access pc->mem_cgroup */
1610 lock_page_cgroup(pc
);
1612 mem
= pc
->mem_cgroup
;
1613 if (!mem
|| !PageCgroupUsed(pc
))
1617 this_cpu_inc(mem
->stat
->count
[MEM_CGROUP_STAT_FILE_MAPPED
]);
1618 SetPageCgroupFileMapped(pc
);
1620 this_cpu_dec(mem
->stat
->count
[MEM_CGROUP_STAT_FILE_MAPPED
]);
1621 if (!page_mapped(page
)) /* for race between dec->inc counter */
1622 ClearPageCgroupFileMapped(pc
);
1626 if (unlikely(need_unlock
))
1627 unlock_page_cgroup(pc
);
1633 * size of first charge trial. "32" comes from vmscan.c's magic value.
1634 * TODO: maybe necessary to use big numbers in big irons.
1636 #define CHARGE_SIZE (32 * PAGE_SIZE)
1637 struct memcg_stock_pcp
{
1638 struct mem_cgroup
*cached
; /* this never be root cgroup */
1640 struct work_struct work
;
1642 static DEFINE_PER_CPU(struct memcg_stock_pcp
, memcg_stock
);
1643 static atomic_t memcg_drain_count
;
1646 * Try to consume stocked charge on this cpu. If success, PAGE_SIZE is consumed
1647 * from local stock and true is returned. If the stock is 0 or charges from a
1648 * cgroup which is not current target, returns false. This stock will be
1651 static bool consume_stock(struct mem_cgroup
*mem
)
1653 struct memcg_stock_pcp
*stock
;
1656 stock
= &get_cpu_var(memcg_stock
);
1657 if (mem
== stock
->cached
&& stock
->charge
)
1658 stock
->charge
-= PAGE_SIZE
;
1659 else /* need to call res_counter_charge */
1661 put_cpu_var(memcg_stock
);
1666 * Returns stocks cached in percpu to res_counter and reset cached information.
1668 static void drain_stock(struct memcg_stock_pcp
*stock
)
1670 struct mem_cgroup
*old
= stock
->cached
;
1672 if (stock
->charge
) {
1673 res_counter_uncharge(&old
->res
, stock
->charge
);
1674 if (do_swap_account
)
1675 res_counter_uncharge(&old
->memsw
, stock
->charge
);
1677 stock
->cached
= NULL
;
1682 * This must be called under preempt disabled or must be called by
1683 * a thread which is pinned to local cpu.
1685 static void drain_local_stock(struct work_struct
*dummy
)
1687 struct memcg_stock_pcp
*stock
= &__get_cpu_var(memcg_stock
);
1692 * Cache charges(val) which is from res_counter, to local per_cpu area.
1693 * This will be consumed by consume_stock() function, later.
1695 static void refill_stock(struct mem_cgroup
*mem
, int val
)
1697 struct memcg_stock_pcp
*stock
= &get_cpu_var(memcg_stock
);
1699 if (stock
->cached
!= mem
) { /* reset if necessary */
1701 stock
->cached
= mem
;
1703 stock
->charge
+= val
;
1704 put_cpu_var(memcg_stock
);
1708 * Tries to drain stocked charges in other cpus. This function is asynchronous
1709 * and just put a work per cpu for draining localy on each cpu. Caller can
1710 * expects some charges will be back to res_counter later but cannot wait for
1713 static void drain_all_stock_async(void)
1716 /* This function is for scheduling "drain" in asynchronous way.
1717 * The result of "drain" is not directly handled by callers. Then,
1718 * if someone is calling drain, we don't have to call drain more.
1719 * Anyway, WORK_STRUCT_PENDING check in queue_work_on() will catch if
1720 * there is a race. We just do loose check here.
1722 if (atomic_read(&memcg_drain_count
))
1724 /* Notify other cpus that system-wide "drain" is running */
1725 atomic_inc(&memcg_drain_count
);
1727 for_each_online_cpu(cpu
) {
1728 struct memcg_stock_pcp
*stock
= &per_cpu(memcg_stock
, cpu
);
1729 schedule_work_on(cpu
, &stock
->work
);
1732 atomic_dec(&memcg_drain_count
);
1733 /* We don't wait for flush_work */
1736 /* This is a synchronous drain interface. */
1737 static void drain_all_stock_sync(void)
1739 /* called when force_empty is called */
1740 atomic_inc(&memcg_drain_count
);
1741 schedule_on_each_cpu(drain_local_stock
);
1742 atomic_dec(&memcg_drain_count
);
1746 * This function drains percpu counter value from DEAD cpu and
1747 * move it to local cpu. Note that this function can be preempted.
1749 static void mem_cgroup_drain_pcp_counter(struct mem_cgroup
*mem
, int cpu
)
1753 spin_lock(&mem
->pcp_counter_lock
);
1754 for (i
= 0; i
< MEM_CGROUP_STAT_DATA
; i
++) {
1755 s64 x
= per_cpu(mem
->stat
->count
[i
], cpu
);
1757 per_cpu(mem
->stat
->count
[i
], cpu
) = 0;
1758 mem
->nocpu_base
.count
[i
] += x
;
1760 /* need to clear ON_MOVE value, works as a kind of lock. */
1761 per_cpu(mem
->stat
->count
[MEM_CGROUP_ON_MOVE
], cpu
) = 0;
1762 spin_unlock(&mem
->pcp_counter_lock
);
1765 static void synchronize_mem_cgroup_on_move(struct mem_cgroup
*mem
, int cpu
)
1767 int idx
= MEM_CGROUP_ON_MOVE
;
1769 spin_lock(&mem
->pcp_counter_lock
);
1770 per_cpu(mem
->stat
->count
[idx
], cpu
) = mem
->nocpu_base
.count
[idx
];
1771 spin_unlock(&mem
->pcp_counter_lock
);
1774 static int __cpuinit
memcg_cpu_hotplug_callback(struct notifier_block
*nb
,
1775 unsigned long action
,
1778 int cpu
= (unsigned long)hcpu
;
1779 struct memcg_stock_pcp
*stock
;
1780 struct mem_cgroup
*iter
;
1782 if ((action
== CPU_ONLINE
)) {
1783 for_each_mem_cgroup_all(iter
)
1784 synchronize_mem_cgroup_on_move(iter
, cpu
);
1788 if ((action
!= CPU_DEAD
) || action
!= CPU_DEAD_FROZEN
)
1791 for_each_mem_cgroup_all(iter
)
1792 mem_cgroup_drain_pcp_counter(iter
, cpu
);
1794 stock
= &per_cpu(memcg_stock
, cpu
);
1800 /* See __mem_cgroup_try_charge() for details */
1802 CHARGE_OK
, /* success */
1803 CHARGE_RETRY
, /* need to retry but retry is not bad */
1804 CHARGE_NOMEM
, /* we can't do more. return -ENOMEM */
1805 CHARGE_WOULDBLOCK
, /* GFP_WAIT wasn't set and no enough res. */
1806 CHARGE_OOM_DIE
, /* the current is killed because of OOM */
1809 static int __mem_cgroup_do_charge(struct mem_cgroup
*mem
, gfp_t gfp_mask
,
1810 int csize
, bool oom_check
)
1812 struct mem_cgroup
*mem_over_limit
;
1813 struct res_counter
*fail_res
;
1814 unsigned long flags
= 0;
1817 ret
= res_counter_charge(&mem
->res
, csize
, &fail_res
);
1820 if (!do_swap_account
)
1822 ret
= res_counter_charge(&mem
->memsw
, csize
, &fail_res
);
1826 mem_over_limit
= mem_cgroup_from_res_counter(fail_res
, memsw
);
1827 flags
|= MEM_CGROUP_RECLAIM_NOSWAP
;
1829 mem_over_limit
= mem_cgroup_from_res_counter(fail_res
, res
);
1831 if (csize
> PAGE_SIZE
) /* change csize and retry */
1832 return CHARGE_RETRY
;
1834 if (!(gfp_mask
& __GFP_WAIT
))
1835 return CHARGE_WOULDBLOCK
;
1837 ret
= mem_cgroup_hierarchical_reclaim(mem_over_limit
, NULL
,
1840 * try_to_free_mem_cgroup_pages() might not give us a full
1841 * picture of reclaim. Some pages are reclaimed and might be
1842 * moved to swap cache or just unmapped from the cgroup.
1843 * Check the limit again to see if the reclaim reduced the
1844 * current usage of the cgroup before giving up
1846 if (ret
|| mem_cgroup_check_under_limit(mem_over_limit
))
1847 return CHARGE_RETRY
;
1850 * At task move, charge accounts can be doubly counted. So, it's
1851 * better to wait until the end of task_move if something is going on.
1853 if (mem_cgroup_wait_acct_move(mem_over_limit
))
1854 return CHARGE_RETRY
;
1856 /* If we don't need to call oom-killer at el, return immediately */
1858 return CHARGE_NOMEM
;
1860 if (!mem_cgroup_handle_oom(mem_over_limit
, gfp_mask
))
1861 return CHARGE_OOM_DIE
;
1863 return CHARGE_RETRY
;
1867 * Unlike exported interface, "oom" parameter is added. if oom==true,
1868 * oom-killer can be invoked.
1870 static int __mem_cgroup_try_charge(struct mm_struct
*mm
,
1871 gfp_t gfp_mask
, struct mem_cgroup
**memcg
, bool oom
)
1873 int nr_oom_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
1874 struct mem_cgroup
*mem
= NULL
;
1876 int csize
= CHARGE_SIZE
;
1879 * Unlike gloval-vm's OOM-kill, we're not in memory shortage
1880 * in system level. So, allow to go ahead dying process in addition to
1883 if (unlikely(test_thread_flag(TIF_MEMDIE
)
1884 || fatal_signal_pending(current
)))
1888 * We always charge the cgroup the mm_struct belongs to.
1889 * The mm_struct's mem_cgroup changes on task migration if the
1890 * thread group leader migrates. It's possible that mm is not
1891 * set, if so charge the init_mm (happens for pagecache usage).
1896 if (*memcg
) { /* css should be a valid one */
1898 VM_BUG_ON(css_is_removed(&mem
->css
));
1899 if (mem_cgroup_is_root(mem
))
1901 if (consume_stock(mem
))
1905 struct task_struct
*p
;
1908 p
= rcu_dereference(mm
->owner
);
1911 * because we don't have task_lock(), "p" can exit while
1912 * we're here. In that case, "mem" can point to root
1913 * cgroup but never be NULL. (and task_struct itself is freed
1914 * by RCU, cgroup itself is RCU safe.) Then, we have small
1915 * risk here to get wrong cgroup. But such kind of mis-account
1916 * by race always happens because we don't have cgroup_mutex().
1917 * It's overkill and we allow that small race, here.
1919 mem
= mem_cgroup_from_task(p
);
1921 if (mem_cgroup_is_root(mem
)) {
1925 if (consume_stock(mem
)) {
1927 * It seems dagerous to access memcg without css_get().
1928 * But considering how consume_stok works, it's not
1929 * necessary. If consume_stock success, some charges
1930 * from this memcg are cached on this cpu. So, we
1931 * don't need to call css_get()/css_tryget() before
1932 * calling consume_stock().
1937 /* after here, we may be blocked. we need to get refcnt */
1938 if (!css_tryget(&mem
->css
)) {
1948 /* If killed, bypass charge */
1949 if (fatal_signal_pending(current
)) {
1955 if (oom
&& !nr_oom_retries
) {
1957 nr_oom_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
1960 ret
= __mem_cgroup_do_charge(mem
, gfp_mask
, csize
, oom_check
);
1965 case CHARGE_RETRY
: /* not in OOM situation but retry */
1970 case CHARGE_WOULDBLOCK
: /* !__GFP_WAIT */
1973 case CHARGE_NOMEM
: /* OOM routine works */
1978 /* If oom, we never return -ENOMEM */
1981 case CHARGE_OOM_DIE
: /* Killed by OOM Killer */
1985 } while (ret
!= CHARGE_OK
);
1987 if (csize
> PAGE_SIZE
)
1988 refill_stock(mem
, csize
- PAGE_SIZE
);
2002 * Somemtimes we have to undo a charge we got by try_charge().
2003 * This function is for that and do uncharge, put css's refcnt.
2004 * gotten by try_charge().
2006 static void __mem_cgroup_cancel_charge(struct mem_cgroup
*mem
,
2007 unsigned long count
)
2009 if (!mem_cgroup_is_root(mem
)) {
2010 res_counter_uncharge(&mem
->res
, PAGE_SIZE
* count
);
2011 if (do_swap_account
)
2012 res_counter_uncharge(&mem
->memsw
, PAGE_SIZE
* count
);
2016 static void mem_cgroup_cancel_charge(struct mem_cgroup
*mem
)
2018 __mem_cgroup_cancel_charge(mem
, 1);
2022 * A helper function to get mem_cgroup from ID. must be called under
2023 * rcu_read_lock(). The caller must check css_is_removed() or some if
2024 * it's concern. (dropping refcnt from swap can be called against removed
2027 static struct mem_cgroup
*mem_cgroup_lookup(unsigned short id
)
2029 struct cgroup_subsys_state
*css
;
2031 /* ID 0 is unused ID */
2034 css
= css_lookup(&mem_cgroup_subsys
, id
);
2037 return container_of(css
, struct mem_cgroup
, css
);
2040 struct mem_cgroup
*try_get_mem_cgroup_from_page(struct page
*page
)
2042 struct mem_cgroup
*mem
= NULL
;
2043 struct page_cgroup
*pc
;
2047 VM_BUG_ON(!PageLocked(page
));
2049 pc
= lookup_page_cgroup(page
);
2050 lock_page_cgroup(pc
);
2051 if (PageCgroupUsed(pc
)) {
2052 mem
= pc
->mem_cgroup
;
2053 if (mem
&& !css_tryget(&mem
->css
))
2055 } else if (PageSwapCache(page
)) {
2056 ent
.val
= page_private(page
);
2057 id
= lookup_swap_cgroup(ent
);
2059 mem
= mem_cgroup_lookup(id
);
2060 if (mem
&& !css_tryget(&mem
->css
))
2064 unlock_page_cgroup(pc
);
2069 * commit a charge got by __mem_cgroup_try_charge() and makes page_cgroup to be
2070 * USED state. If already USED, uncharge and return.
2073 static void __mem_cgroup_commit_charge(struct mem_cgroup
*mem
,
2074 struct page_cgroup
*pc
,
2075 enum charge_type ctype
)
2077 /* try_charge() can return NULL to *memcg, taking care of it. */
2081 lock_page_cgroup(pc
);
2082 if (unlikely(PageCgroupUsed(pc
))) {
2083 unlock_page_cgroup(pc
);
2084 mem_cgroup_cancel_charge(mem
);
2088 pc
->mem_cgroup
= mem
;
2090 * We access a page_cgroup asynchronously without lock_page_cgroup().
2091 * Especially when a page_cgroup is taken from a page, pc->mem_cgroup
2092 * is accessed after testing USED bit. To make pc->mem_cgroup visible
2093 * before USED bit, we need memory barrier here.
2094 * See mem_cgroup_add_lru_list(), etc.
2098 case MEM_CGROUP_CHARGE_TYPE_CACHE
:
2099 case MEM_CGROUP_CHARGE_TYPE_SHMEM
:
2100 SetPageCgroupCache(pc
);
2101 SetPageCgroupUsed(pc
);
2103 case MEM_CGROUP_CHARGE_TYPE_MAPPED
:
2104 ClearPageCgroupCache(pc
);
2105 SetPageCgroupUsed(pc
);
2111 mem_cgroup_charge_statistics(mem
, pc
, true);
2113 unlock_page_cgroup(pc
);
2115 * "charge_statistics" updated event counter. Then, check it.
2116 * Insert ancestor (and ancestor's ancestors), to softlimit RB-tree.
2117 * if they exceeds softlimit.
2119 memcg_check_events(mem
, pc
->page
);
2123 * __mem_cgroup_move_account - move account of the page
2124 * @pc: page_cgroup of the page.
2125 * @from: mem_cgroup which the page is moved from.
2126 * @to: mem_cgroup which the page is moved to. @from != @to.
2127 * @uncharge: whether we should call uncharge and css_put against @from.
2129 * The caller must confirm following.
2130 * - page is not on LRU (isolate_page() is useful.)
2131 * - the pc is locked, used, and ->mem_cgroup points to @from.
2133 * This function doesn't do "charge" nor css_get to new cgroup. It should be
2134 * done by a caller(__mem_cgroup_try_charge would be usefull). If @uncharge is
2135 * true, this function does "uncharge" from old cgroup, but it doesn't if
2136 * @uncharge is false, so a caller should do "uncharge".
2139 static void __mem_cgroup_move_account(struct page_cgroup
*pc
,
2140 struct mem_cgroup
*from
, struct mem_cgroup
*to
, bool uncharge
)
2142 VM_BUG_ON(from
== to
);
2143 VM_BUG_ON(PageLRU(pc
->page
));
2144 VM_BUG_ON(!PageCgroupLocked(pc
));
2145 VM_BUG_ON(!PageCgroupUsed(pc
));
2146 VM_BUG_ON(pc
->mem_cgroup
!= from
);
2148 if (PageCgroupFileMapped(pc
)) {
2149 /* Update mapped_file data for mem_cgroup */
2151 __this_cpu_dec(from
->stat
->count
[MEM_CGROUP_STAT_FILE_MAPPED
]);
2152 __this_cpu_inc(to
->stat
->count
[MEM_CGROUP_STAT_FILE_MAPPED
]);
2155 mem_cgroup_charge_statistics(from
, pc
, false);
2157 /* This is not "cancel", but cancel_charge does all we need. */
2158 mem_cgroup_cancel_charge(from
);
2160 /* caller should have done css_get */
2161 pc
->mem_cgroup
= to
;
2162 mem_cgroup_charge_statistics(to
, pc
, true);
2164 * We charges against "to" which may not have any tasks. Then, "to"
2165 * can be under rmdir(). But in current implementation, caller of
2166 * this function is just force_empty() and move charge, so it's
2167 * garanteed that "to" is never removed. So, we don't check rmdir
2173 * check whether the @pc is valid for moving account and call
2174 * __mem_cgroup_move_account()
2176 static int mem_cgroup_move_account(struct page_cgroup
*pc
,
2177 struct mem_cgroup
*from
, struct mem_cgroup
*to
, bool uncharge
)
2180 lock_page_cgroup(pc
);
2181 if (PageCgroupUsed(pc
) && pc
->mem_cgroup
== from
) {
2182 __mem_cgroup_move_account(pc
, from
, to
, uncharge
);
2185 unlock_page_cgroup(pc
);
2189 memcg_check_events(to
, pc
->page
);
2190 memcg_check_events(from
, pc
->page
);
2195 * move charges to its parent.
2198 static int mem_cgroup_move_parent(struct page_cgroup
*pc
,
2199 struct mem_cgroup
*child
,
2202 struct page
*page
= pc
->page
;
2203 struct cgroup
*cg
= child
->css
.cgroup
;
2204 struct cgroup
*pcg
= cg
->parent
;
2205 struct mem_cgroup
*parent
;
2213 if (!get_page_unless_zero(page
))
2215 if (isolate_lru_page(page
))
2218 parent
= mem_cgroup_from_cont(pcg
);
2219 ret
= __mem_cgroup_try_charge(NULL
, gfp_mask
, &parent
, false);
2223 ret
= mem_cgroup_move_account(pc
, child
, parent
, true);
2225 mem_cgroup_cancel_charge(parent
);
2227 putback_lru_page(page
);
2235 * Charge the memory controller for page usage.
2237 * 0 if the charge was successful
2238 * < 0 if the cgroup is over its limit
2240 static int mem_cgroup_charge_common(struct page
*page
, struct mm_struct
*mm
,
2241 gfp_t gfp_mask
, enum charge_type ctype
)
2243 struct mem_cgroup
*mem
= NULL
;
2244 struct page_cgroup
*pc
;
2247 pc
= lookup_page_cgroup(page
);
2248 /* can happen at boot */
2253 ret
= __mem_cgroup_try_charge(mm
, gfp_mask
, &mem
, true);
2257 __mem_cgroup_commit_charge(mem
, pc
, ctype
);
2261 int mem_cgroup_newpage_charge(struct page
*page
,
2262 struct mm_struct
*mm
, gfp_t gfp_mask
)
2264 if (mem_cgroup_disabled())
2266 if (PageCompound(page
))
2269 * If already mapped, we don't have to account.
2270 * If page cache, page->mapping has address_space.
2271 * But page->mapping may have out-of-use anon_vma pointer,
2272 * detecit it by PageAnon() check. newly-mapped-anon's page->mapping
2275 if (page_mapped(page
) || (page
->mapping
&& !PageAnon(page
)))
2279 return mem_cgroup_charge_common(page
, mm
, gfp_mask
,
2280 MEM_CGROUP_CHARGE_TYPE_MAPPED
);
2284 __mem_cgroup_commit_charge_swapin(struct page
*page
, struct mem_cgroup
*ptr
,
2285 enum charge_type ctype
);
2287 int mem_cgroup_cache_charge(struct page
*page
, struct mm_struct
*mm
,
2292 if (mem_cgroup_disabled())
2294 if (PageCompound(page
))
2297 * Corner case handling. This is called from add_to_page_cache()
2298 * in usual. But some FS (shmem) precharges this page before calling it
2299 * and call add_to_page_cache() with GFP_NOWAIT.
2301 * For GFP_NOWAIT case, the page may be pre-charged before calling
2302 * add_to_page_cache(). (See shmem.c) check it here and avoid to call
2303 * charge twice. (It works but has to pay a bit larger cost.)
2304 * And when the page is SwapCache, it should take swap information
2305 * into account. This is under lock_page() now.
2307 if (!(gfp_mask
& __GFP_WAIT
)) {
2308 struct page_cgroup
*pc
;
2310 pc
= lookup_page_cgroup(page
);
2313 lock_page_cgroup(pc
);
2314 if (PageCgroupUsed(pc
)) {
2315 unlock_page_cgroup(pc
);
2318 unlock_page_cgroup(pc
);
2324 if (page_is_file_cache(page
))
2325 return mem_cgroup_charge_common(page
, mm
, gfp_mask
,
2326 MEM_CGROUP_CHARGE_TYPE_CACHE
);
2329 if (PageSwapCache(page
)) {
2330 struct mem_cgroup
*mem
= NULL
;
2332 ret
= mem_cgroup_try_charge_swapin(mm
, page
, gfp_mask
, &mem
);
2334 __mem_cgroup_commit_charge_swapin(page
, mem
,
2335 MEM_CGROUP_CHARGE_TYPE_SHMEM
);
2337 ret
= mem_cgroup_charge_common(page
, mm
, gfp_mask
,
2338 MEM_CGROUP_CHARGE_TYPE_SHMEM
);
2344 * While swap-in, try_charge -> commit or cancel, the page is locked.
2345 * And when try_charge() successfully returns, one refcnt to memcg without
2346 * struct page_cgroup is acquired. This refcnt will be consumed by
2347 * "commit()" or removed by "cancel()"
2349 int mem_cgroup_try_charge_swapin(struct mm_struct
*mm
,
2351 gfp_t mask
, struct mem_cgroup
**ptr
)
2353 struct mem_cgroup
*mem
;
2356 if (mem_cgroup_disabled())
2359 if (!do_swap_account
)
2362 * A racing thread's fault, or swapoff, may have already updated
2363 * the pte, and even removed page from swap cache: in those cases
2364 * do_swap_page()'s pte_same() test will fail; but there's also a
2365 * KSM case which does need to charge the page.
2367 if (!PageSwapCache(page
))
2369 mem
= try_get_mem_cgroup_from_page(page
);
2373 ret
= __mem_cgroup_try_charge(NULL
, mask
, ptr
, true);
2379 return __mem_cgroup_try_charge(mm
, mask
, ptr
, true);
2383 __mem_cgroup_commit_charge_swapin(struct page
*page
, struct mem_cgroup
*ptr
,
2384 enum charge_type ctype
)
2386 struct page_cgroup
*pc
;
2388 if (mem_cgroup_disabled())
2392 cgroup_exclude_rmdir(&ptr
->css
);
2393 pc
= lookup_page_cgroup(page
);
2394 mem_cgroup_lru_del_before_commit_swapcache(page
);
2395 __mem_cgroup_commit_charge(ptr
, pc
, ctype
);
2396 mem_cgroup_lru_add_after_commit_swapcache(page
);
2398 * Now swap is on-memory. This means this page may be
2399 * counted both as mem and swap....double count.
2400 * Fix it by uncharging from memsw. Basically, this SwapCache is stable
2401 * under lock_page(). But in do_swap_page()::memory.c, reuse_swap_page()
2402 * may call delete_from_swap_cache() before reach here.
2404 if (do_swap_account
&& PageSwapCache(page
)) {
2405 swp_entry_t ent
= {.val
= page_private(page
)};
2407 struct mem_cgroup
*memcg
;
2409 id
= swap_cgroup_record(ent
, 0);
2411 memcg
= mem_cgroup_lookup(id
);
2414 * This recorded memcg can be obsolete one. So, avoid
2415 * calling css_tryget
2417 if (!mem_cgroup_is_root(memcg
))
2418 res_counter_uncharge(&memcg
->memsw
, PAGE_SIZE
);
2419 mem_cgroup_swap_statistics(memcg
, false);
2420 mem_cgroup_put(memcg
);
2425 * At swapin, we may charge account against cgroup which has no tasks.
2426 * So, rmdir()->pre_destroy() can be called while we do this charge.
2427 * In that case, we need to call pre_destroy() again. check it here.
2429 cgroup_release_and_wakeup_rmdir(&ptr
->css
);
2432 void mem_cgroup_commit_charge_swapin(struct page
*page
, struct mem_cgroup
*ptr
)
2434 __mem_cgroup_commit_charge_swapin(page
, ptr
,
2435 MEM_CGROUP_CHARGE_TYPE_MAPPED
);
2438 void mem_cgroup_cancel_charge_swapin(struct mem_cgroup
*mem
)
2440 if (mem_cgroup_disabled())
2444 mem_cgroup_cancel_charge(mem
);
2448 __do_uncharge(struct mem_cgroup
*mem
, const enum charge_type ctype
)
2450 struct memcg_batch_info
*batch
= NULL
;
2451 bool uncharge_memsw
= true;
2452 /* If swapout, usage of swap doesn't decrease */
2453 if (!do_swap_account
|| ctype
== MEM_CGROUP_CHARGE_TYPE_SWAPOUT
)
2454 uncharge_memsw
= false;
2456 batch
= ¤t
->memcg_batch
;
2458 * In usual, we do css_get() when we remember memcg pointer.
2459 * But in this case, we keep res->usage until end of a series of
2460 * uncharges. Then, it's ok to ignore memcg's refcnt.
2465 * do_batch > 0 when unmapping pages or inode invalidate/truncate.
2466 * In those cases, all pages freed continously can be expected to be in
2467 * the same cgroup and we have chance to coalesce uncharges.
2468 * But we do uncharge one by one if this is killed by OOM(TIF_MEMDIE)
2469 * because we want to do uncharge as soon as possible.
2472 if (!batch
->do_batch
|| test_thread_flag(TIF_MEMDIE
))
2473 goto direct_uncharge
;
2476 * In typical case, batch->memcg == mem. This means we can
2477 * merge a series of uncharges to an uncharge of res_counter.
2478 * If not, we uncharge res_counter ony by one.
2480 if (batch
->memcg
!= mem
)
2481 goto direct_uncharge
;
2482 /* remember freed charge and uncharge it later */
2483 batch
->bytes
+= PAGE_SIZE
;
2485 batch
->memsw_bytes
+= PAGE_SIZE
;
2488 res_counter_uncharge(&mem
->res
, PAGE_SIZE
);
2490 res_counter_uncharge(&mem
->memsw
, PAGE_SIZE
);
2491 if (unlikely(batch
->memcg
!= mem
))
2492 memcg_oom_recover(mem
);
2497 * uncharge if !page_mapped(page)
2499 static struct mem_cgroup
*
2500 __mem_cgroup_uncharge_common(struct page
*page
, enum charge_type ctype
)
2502 struct page_cgroup
*pc
;
2503 struct mem_cgroup
*mem
= NULL
;
2505 if (mem_cgroup_disabled())
2508 if (PageSwapCache(page
))
2512 * Check if our page_cgroup is valid
2514 pc
= lookup_page_cgroup(page
);
2515 if (unlikely(!pc
|| !PageCgroupUsed(pc
)))
2518 lock_page_cgroup(pc
);
2520 mem
= pc
->mem_cgroup
;
2522 if (!PageCgroupUsed(pc
))
2526 case MEM_CGROUP_CHARGE_TYPE_MAPPED
:
2527 case MEM_CGROUP_CHARGE_TYPE_DROP
:
2528 /* See mem_cgroup_prepare_migration() */
2529 if (page_mapped(page
) || PageCgroupMigration(pc
))
2532 case MEM_CGROUP_CHARGE_TYPE_SWAPOUT
:
2533 if (!PageAnon(page
)) { /* Shared memory */
2534 if (page
->mapping
&& !page_is_file_cache(page
))
2536 } else if (page_mapped(page
)) /* Anon */
2543 mem_cgroup_charge_statistics(mem
, pc
, false);
2545 ClearPageCgroupUsed(pc
);
2547 * pc->mem_cgroup is not cleared here. It will be accessed when it's
2548 * freed from LRU. This is safe because uncharged page is expected not
2549 * to be reused (freed soon). Exception is SwapCache, it's handled by
2550 * special functions.
2553 unlock_page_cgroup(pc
);
2555 * even after unlock, we have mem->res.usage here and this memcg
2556 * will never be freed.
2558 memcg_check_events(mem
, page
);
2559 if (do_swap_account
&& ctype
== MEM_CGROUP_CHARGE_TYPE_SWAPOUT
) {
2560 mem_cgroup_swap_statistics(mem
, true);
2561 mem_cgroup_get(mem
);
2563 if (!mem_cgroup_is_root(mem
))
2564 __do_uncharge(mem
, ctype
);
2569 unlock_page_cgroup(pc
);
2573 void mem_cgroup_uncharge_page(struct page
*page
)
2576 if (page_mapped(page
))
2578 if (page
->mapping
&& !PageAnon(page
))
2580 __mem_cgroup_uncharge_common(page
, MEM_CGROUP_CHARGE_TYPE_MAPPED
);
2583 void mem_cgroup_uncharge_cache_page(struct page
*page
)
2585 VM_BUG_ON(page_mapped(page
));
2586 VM_BUG_ON(page
->mapping
);
2587 __mem_cgroup_uncharge_common(page
, MEM_CGROUP_CHARGE_TYPE_CACHE
);
2591 * Batch_start/batch_end is called in unmap_page_range/invlidate/trucate.
2592 * In that cases, pages are freed continuously and we can expect pages
2593 * are in the same memcg. All these calls itself limits the number of
2594 * pages freed at once, then uncharge_start/end() is called properly.
2595 * This may be called prural(2) times in a context,
2598 void mem_cgroup_uncharge_start(void)
2600 current
->memcg_batch
.do_batch
++;
2601 /* We can do nest. */
2602 if (current
->memcg_batch
.do_batch
== 1) {
2603 current
->memcg_batch
.memcg
= NULL
;
2604 current
->memcg_batch
.bytes
= 0;
2605 current
->memcg_batch
.memsw_bytes
= 0;
2609 void mem_cgroup_uncharge_end(void)
2611 struct memcg_batch_info
*batch
= ¤t
->memcg_batch
;
2613 if (!batch
->do_batch
)
2617 if (batch
->do_batch
) /* If stacked, do nothing. */
2623 * This "batch->memcg" is valid without any css_get/put etc...
2624 * bacause we hide charges behind us.
2627 res_counter_uncharge(&batch
->memcg
->res
, batch
->bytes
);
2628 if (batch
->memsw_bytes
)
2629 res_counter_uncharge(&batch
->memcg
->memsw
, batch
->memsw_bytes
);
2630 memcg_oom_recover(batch
->memcg
);
2631 /* forget this pointer (for sanity check) */
2632 batch
->memcg
= NULL
;
2637 * called after __delete_from_swap_cache() and drop "page" account.
2638 * memcg information is recorded to swap_cgroup of "ent"
2641 mem_cgroup_uncharge_swapcache(struct page
*page
, swp_entry_t ent
, bool swapout
)
2643 struct mem_cgroup
*memcg
;
2644 int ctype
= MEM_CGROUP_CHARGE_TYPE_SWAPOUT
;
2646 if (!swapout
) /* this was a swap cache but the swap is unused ! */
2647 ctype
= MEM_CGROUP_CHARGE_TYPE_DROP
;
2649 memcg
= __mem_cgroup_uncharge_common(page
, ctype
);
2652 * record memcg information, if swapout && memcg != NULL,
2653 * mem_cgroup_get() was called in uncharge().
2655 if (do_swap_account
&& swapout
&& memcg
)
2656 swap_cgroup_record(ent
, css_id(&memcg
->css
));
2660 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
2662 * called from swap_entry_free(). remove record in swap_cgroup and
2663 * uncharge "memsw" account.
2665 void mem_cgroup_uncharge_swap(swp_entry_t ent
)
2667 struct mem_cgroup
*memcg
;
2670 if (!do_swap_account
)
2673 id
= swap_cgroup_record(ent
, 0);
2675 memcg
= mem_cgroup_lookup(id
);
2678 * We uncharge this because swap is freed.
2679 * This memcg can be obsolete one. We avoid calling css_tryget
2681 if (!mem_cgroup_is_root(memcg
))
2682 res_counter_uncharge(&memcg
->memsw
, PAGE_SIZE
);
2683 mem_cgroup_swap_statistics(memcg
, false);
2684 mem_cgroup_put(memcg
);
2690 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
2691 * @entry: swap entry to be moved
2692 * @from: mem_cgroup which the entry is moved from
2693 * @to: mem_cgroup which the entry is moved to
2694 * @need_fixup: whether we should fixup res_counters and refcounts.
2696 * It succeeds only when the swap_cgroup's record for this entry is the same
2697 * as the mem_cgroup's id of @from.
2699 * Returns 0 on success, -EINVAL on failure.
2701 * The caller must have charged to @to, IOW, called res_counter_charge() about
2702 * both res and memsw, and called css_get().
2704 static int mem_cgroup_move_swap_account(swp_entry_t entry
,
2705 struct mem_cgroup
*from
, struct mem_cgroup
*to
, bool need_fixup
)
2707 unsigned short old_id
, new_id
;
2709 old_id
= css_id(&from
->css
);
2710 new_id
= css_id(&to
->css
);
2712 if (swap_cgroup_cmpxchg(entry
, old_id
, new_id
) == old_id
) {
2713 mem_cgroup_swap_statistics(from
, false);
2714 mem_cgroup_swap_statistics(to
, true);
2716 * This function is only called from task migration context now.
2717 * It postpones res_counter and refcount handling till the end
2718 * of task migration(mem_cgroup_clear_mc()) for performance
2719 * improvement. But we cannot postpone mem_cgroup_get(to)
2720 * because if the process that has been moved to @to does
2721 * swap-in, the refcount of @to might be decreased to 0.
2725 if (!mem_cgroup_is_root(from
))
2726 res_counter_uncharge(&from
->memsw
, PAGE_SIZE
);
2727 mem_cgroup_put(from
);
2729 * we charged both to->res and to->memsw, so we should
2732 if (!mem_cgroup_is_root(to
))
2733 res_counter_uncharge(&to
->res
, PAGE_SIZE
);
2740 static inline int mem_cgroup_move_swap_account(swp_entry_t entry
,
2741 struct mem_cgroup
*from
, struct mem_cgroup
*to
, bool need_fixup
)
2748 * Before starting migration, account PAGE_SIZE to mem_cgroup that the old
2751 int mem_cgroup_prepare_migration(struct page
*page
,
2752 struct page
*newpage
, struct mem_cgroup
**ptr
)
2754 struct page_cgroup
*pc
;
2755 struct mem_cgroup
*mem
= NULL
;
2756 enum charge_type ctype
;
2759 if (mem_cgroup_disabled())
2762 pc
= lookup_page_cgroup(page
);
2763 lock_page_cgroup(pc
);
2764 if (PageCgroupUsed(pc
)) {
2765 mem
= pc
->mem_cgroup
;
2768 * At migrating an anonymous page, its mapcount goes down
2769 * to 0 and uncharge() will be called. But, even if it's fully
2770 * unmapped, migration may fail and this page has to be
2771 * charged again. We set MIGRATION flag here and delay uncharge
2772 * until end_migration() is called
2774 * Corner Case Thinking
2776 * When the old page was mapped as Anon and it's unmap-and-freed
2777 * while migration was ongoing.
2778 * If unmap finds the old page, uncharge() of it will be delayed
2779 * until end_migration(). If unmap finds a new page, it's
2780 * uncharged when it make mapcount to be 1->0. If unmap code
2781 * finds swap_migration_entry, the new page will not be mapped
2782 * and end_migration() will find it(mapcount==0).
2785 * When the old page was mapped but migraion fails, the kernel
2786 * remaps it. A charge for it is kept by MIGRATION flag even
2787 * if mapcount goes down to 0. We can do remap successfully
2788 * without charging it again.
2791 * The "old" page is under lock_page() until the end of
2792 * migration, so, the old page itself will not be swapped-out.
2793 * If the new page is swapped out before end_migraton, our
2794 * hook to usual swap-out path will catch the event.
2797 SetPageCgroupMigration(pc
);
2799 unlock_page_cgroup(pc
);
2801 * If the page is not charged at this point,
2808 ret
= __mem_cgroup_try_charge(NULL
, GFP_KERNEL
, ptr
, false);
2809 css_put(&mem
->css
);/* drop extra refcnt */
2810 if (ret
|| *ptr
== NULL
) {
2811 if (PageAnon(page
)) {
2812 lock_page_cgroup(pc
);
2813 ClearPageCgroupMigration(pc
);
2814 unlock_page_cgroup(pc
);
2816 * The old page may be fully unmapped while we kept it.
2818 mem_cgroup_uncharge_page(page
);
2823 * We charge new page before it's used/mapped. So, even if unlock_page()
2824 * is called before end_migration, we can catch all events on this new
2825 * page. In the case new page is migrated but not remapped, new page's
2826 * mapcount will be finally 0 and we call uncharge in end_migration().
2828 pc
= lookup_page_cgroup(newpage
);
2830 ctype
= MEM_CGROUP_CHARGE_TYPE_MAPPED
;
2831 else if (page_is_file_cache(page
))
2832 ctype
= MEM_CGROUP_CHARGE_TYPE_CACHE
;
2834 ctype
= MEM_CGROUP_CHARGE_TYPE_SHMEM
;
2835 __mem_cgroup_commit_charge(mem
, pc
, ctype
);
2839 /* remove redundant charge if migration failed*/
2840 void mem_cgroup_end_migration(struct mem_cgroup
*mem
,
2841 struct page
*oldpage
, struct page
*newpage
)
2843 struct page
*used
, *unused
;
2844 struct page_cgroup
*pc
;
2848 /* blocks rmdir() */
2849 cgroup_exclude_rmdir(&mem
->css
);
2850 /* at migration success, oldpage->mapping is NULL. */
2851 if (oldpage
->mapping
) {
2859 * We disallowed uncharge of pages under migration because mapcount
2860 * of the page goes down to zero, temporarly.
2861 * Clear the flag and check the page should be charged.
2863 pc
= lookup_page_cgroup(oldpage
);
2864 lock_page_cgroup(pc
);
2865 ClearPageCgroupMigration(pc
);
2866 unlock_page_cgroup(pc
);
2868 __mem_cgroup_uncharge_common(unused
, MEM_CGROUP_CHARGE_TYPE_FORCE
);
2871 * If a page is a file cache, radix-tree replacement is very atomic
2872 * and we can skip this check. When it was an Anon page, its mapcount
2873 * goes down to 0. But because we added MIGRATION flage, it's not
2874 * uncharged yet. There are several case but page->mapcount check
2875 * and USED bit check in mem_cgroup_uncharge_page() will do enough
2876 * check. (see prepare_charge() also)
2879 mem_cgroup_uncharge_page(used
);
2881 * At migration, we may charge account against cgroup which has no
2883 * So, rmdir()->pre_destroy() can be called while we do this charge.
2884 * In that case, we need to call pre_destroy() again. check it here.
2886 cgroup_release_and_wakeup_rmdir(&mem
->css
);
2890 * A call to try to shrink memory usage on charge failure at shmem's swapin.
2891 * Calling hierarchical_reclaim is not enough because we should update
2892 * last_oom_jiffies to prevent pagefault_out_of_memory from invoking global OOM.
2893 * Moreover considering hierarchy, we should reclaim from the mem_over_limit,
2894 * not from the memcg which this page would be charged to.
2895 * try_charge_swapin does all of these works properly.
2897 int mem_cgroup_shmem_charge_fallback(struct page
*page
,
2898 struct mm_struct
*mm
,
2901 struct mem_cgroup
*mem
= NULL
;
2904 if (mem_cgroup_disabled())
2907 ret
= mem_cgroup_try_charge_swapin(mm
, page
, gfp_mask
, &mem
);
2909 mem_cgroup_cancel_charge_swapin(mem
); /* it does !mem check */
2914 static DEFINE_MUTEX(set_limit_mutex
);
2916 static int mem_cgroup_resize_limit(struct mem_cgroup
*memcg
,
2917 unsigned long long val
)
2920 u64 memswlimit
, memlimit
;
2922 int children
= mem_cgroup_count_children(memcg
);
2923 u64 curusage
, oldusage
;
2927 * For keeping hierarchical_reclaim simple, how long we should retry
2928 * is depends on callers. We set our retry-count to be function
2929 * of # of children which we should visit in this loop.
2931 retry_count
= MEM_CGROUP_RECLAIM_RETRIES
* children
;
2933 oldusage
= res_counter_read_u64(&memcg
->res
, RES_USAGE
);
2936 while (retry_count
) {
2937 if (signal_pending(current
)) {
2942 * Rather than hide all in some function, I do this in
2943 * open coded manner. You see what this really does.
2944 * We have to guarantee mem->res.limit < mem->memsw.limit.
2946 mutex_lock(&set_limit_mutex
);
2947 memswlimit
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
2948 if (memswlimit
< val
) {
2950 mutex_unlock(&set_limit_mutex
);
2954 memlimit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
2958 ret
= res_counter_set_limit(&memcg
->res
, val
);
2960 if (memswlimit
== val
)
2961 memcg
->memsw_is_minimum
= true;
2963 memcg
->memsw_is_minimum
= false;
2965 mutex_unlock(&set_limit_mutex
);
2970 mem_cgroup_hierarchical_reclaim(memcg
, NULL
, GFP_KERNEL
,
2971 MEM_CGROUP_RECLAIM_SHRINK
);
2972 curusage
= res_counter_read_u64(&memcg
->res
, RES_USAGE
);
2973 /* Usage is reduced ? */
2974 if (curusage
>= oldusage
)
2977 oldusage
= curusage
;
2979 if (!ret
&& enlarge
)
2980 memcg_oom_recover(memcg
);
2985 static int mem_cgroup_resize_memsw_limit(struct mem_cgroup
*memcg
,
2986 unsigned long long val
)
2989 u64 memlimit
, memswlimit
, oldusage
, curusage
;
2990 int children
= mem_cgroup_count_children(memcg
);
2994 /* see mem_cgroup_resize_res_limit */
2995 retry_count
= children
* MEM_CGROUP_RECLAIM_RETRIES
;
2996 oldusage
= res_counter_read_u64(&memcg
->memsw
, RES_USAGE
);
2997 while (retry_count
) {
2998 if (signal_pending(current
)) {
3003 * Rather than hide all in some function, I do this in
3004 * open coded manner. You see what this really does.
3005 * We have to guarantee mem->res.limit < mem->memsw.limit.
3007 mutex_lock(&set_limit_mutex
);
3008 memlimit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
3009 if (memlimit
> val
) {
3011 mutex_unlock(&set_limit_mutex
);
3014 memswlimit
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
3015 if (memswlimit
< val
)
3017 ret
= res_counter_set_limit(&memcg
->memsw
, val
);
3019 if (memlimit
== val
)
3020 memcg
->memsw_is_minimum
= true;
3022 memcg
->memsw_is_minimum
= false;
3024 mutex_unlock(&set_limit_mutex
);
3029 mem_cgroup_hierarchical_reclaim(memcg
, NULL
, GFP_KERNEL
,
3030 MEM_CGROUP_RECLAIM_NOSWAP
|
3031 MEM_CGROUP_RECLAIM_SHRINK
);
3032 curusage
= res_counter_read_u64(&memcg
->memsw
, RES_USAGE
);
3033 /* Usage is reduced ? */
3034 if (curusage
>= oldusage
)
3037 oldusage
= curusage
;
3039 if (!ret
&& enlarge
)
3040 memcg_oom_recover(memcg
);
3044 unsigned long mem_cgroup_soft_limit_reclaim(struct zone
*zone
, int order
,
3047 unsigned long nr_reclaimed
= 0;
3048 struct mem_cgroup_per_zone
*mz
, *next_mz
= NULL
;
3049 unsigned long reclaimed
;
3051 struct mem_cgroup_tree_per_zone
*mctz
;
3052 unsigned long long excess
;
3057 mctz
= soft_limit_tree_node_zone(zone_to_nid(zone
), zone_idx(zone
));
3059 * This loop can run a while, specially if mem_cgroup's continuously
3060 * keep exceeding their soft limit and putting the system under
3067 mz
= mem_cgroup_largest_soft_limit_node(mctz
);
3071 reclaimed
= mem_cgroup_hierarchical_reclaim(mz
->mem
, zone
,
3073 MEM_CGROUP_RECLAIM_SOFT
);
3074 nr_reclaimed
+= reclaimed
;
3075 spin_lock(&mctz
->lock
);
3078 * If we failed to reclaim anything from this memory cgroup
3079 * it is time to move on to the next cgroup
3085 * Loop until we find yet another one.
3087 * By the time we get the soft_limit lock
3088 * again, someone might have aded the
3089 * group back on the RB tree. Iterate to
3090 * make sure we get a different mem.
3091 * mem_cgroup_largest_soft_limit_node returns
3092 * NULL if no other cgroup is present on
3096 __mem_cgroup_largest_soft_limit_node(mctz
);
3097 if (next_mz
== mz
) {
3098 css_put(&next_mz
->mem
->css
);
3100 } else /* next_mz == NULL or other memcg */
3104 __mem_cgroup_remove_exceeded(mz
->mem
, mz
, mctz
);
3105 excess
= res_counter_soft_limit_excess(&mz
->mem
->res
);
3107 * One school of thought says that we should not add
3108 * back the node to the tree if reclaim returns 0.
3109 * But our reclaim could return 0, simply because due
3110 * to priority we are exposing a smaller subset of
3111 * memory to reclaim from. Consider this as a longer
3114 /* If excess == 0, no tree ops */
3115 __mem_cgroup_insert_exceeded(mz
->mem
, mz
, mctz
, excess
);
3116 spin_unlock(&mctz
->lock
);
3117 css_put(&mz
->mem
->css
);
3120 * Could not reclaim anything and there are no more
3121 * mem cgroups to try or we seem to be looping without
3122 * reclaiming anything.
3124 if (!nr_reclaimed
&&
3126 loop
> MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS
))
3128 } while (!nr_reclaimed
);
3130 css_put(&next_mz
->mem
->css
);
3131 return nr_reclaimed
;
3135 * This routine traverse page_cgroup in given list and drop them all.
3136 * *And* this routine doesn't reclaim page itself, just removes page_cgroup.
3138 static int mem_cgroup_force_empty_list(struct mem_cgroup
*mem
,
3139 int node
, int zid
, enum lru_list lru
)
3142 struct mem_cgroup_per_zone
*mz
;
3143 struct page_cgroup
*pc
, *busy
;
3144 unsigned long flags
, loop
;
3145 struct list_head
*list
;
3148 zone
= &NODE_DATA(node
)->node_zones
[zid
];
3149 mz
= mem_cgroup_zoneinfo(mem
, node
, zid
);
3150 list
= &mz
->lists
[lru
];
3152 loop
= MEM_CGROUP_ZSTAT(mz
, lru
);
3153 /* give some margin against EBUSY etc...*/
3158 spin_lock_irqsave(&zone
->lru_lock
, flags
);
3159 if (list_empty(list
)) {
3160 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
3163 pc
= list_entry(list
->prev
, struct page_cgroup
, lru
);
3165 list_move(&pc
->lru
, list
);
3167 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
3170 spin_unlock_irqrestore(&zone
->lru_lock
, flags
);
3172 ret
= mem_cgroup_move_parent(pc
, mem
, GFP_KERNEL
);
3176 if (ret
== -EBUSY
|| ret
== -EINVAL
) {
3177 /* found lock contention or "pc" is obsolete. */
3184 if (!ret
&& !list_empty(list
))
3190 * make mem_cgroup's charge to be 0 if there is no task.
3191 * This enables deleting this mem_cgroup.
3193 static int mem_cgroup_force_empty(struct mem_cgroup
*mem
, bool free_all
)
3196 int node
, zid
, shrink
;
3197 int nr_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
3198 struct cgroup
*cgrp
= mem
->css
.cgroup
;
3203 /* should free all ? */
3209 if (cgroup_task_count(cgrp
) || !list_empty(&cgrp
->children
))
3212 if (signal_pending(current
))
3214 /* This is for making all *used* pages to be on LRU. */
3215 lru_add_drain_all();
3216 drain_all_stock_sync();
3218 mem_cgroup_start_move(mem
);
3219 for_each_node_state(node
, N_HIGH_MEMORY
) {
3220 for (zid
= 0; !ret
&& zid
< MAX_NR_ZONES
; zid
++) {
3223 ret
= mem_cgroup_force_empty_list(mem
,
3232 mem_cgroup_end_move(mem
);
3233 memcg_oom_recover(mem
);
3234 /* it seems parent cgroup doesn't have enough mem */
3238 /* "ret" should also be checked to ensure all lists are empty. */
3239 } while (mem
->res
.usage
> 0 || ret
);
3245 /* returns EBUSY if there is a task or if we come here twice. */
3246 if (cgroup_task_count(cgrp
) || !list_empty(&cgrp
->children
) || shrink
) {
3250 /* we call try-to-free pages for make this cgroup empty */
3251 lru_add_drain_all();
3252 /* try to free all pages in this cgroup */
3254 while (nr_retries
&& mem
->res
.usage
> 0) {
3257 if (signal_pending(current
)) {
3261 progress
= try_to_free_mem_cgroup_pages(mem
, GFP_KERNEL
,
3262 false, get_swappiness(mem
));
3265 /* maybe some writeback is necessary */
3266 congestion_wait(BLK_RW_ASYNC
, HZ
/10);
3271 /* try move_account...there may be some *locked* pages. */
3275 int mem_cgroup_force_empty_write(struct cgroup
*cont
, unsigned int event
)
3277 return mem_cgroup_force_empty(mem_cgroup_from_cont(cont
), true);
3281 static u64
mem_cgroup_hierarchy_read(struct cgroup
*cont
, struct cftype
*cft
)
3283 return mem_cgroup_from_cont(cont
)->use_hierarchy
;
3286 static int mem_cgroup_hierarchy_write(struct cgroup
*cont
, struct cftype
*cft
,
3290 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cont
);
3291 struct cgroup
*parent
= cont
->parent
;
3292 struct mem_cgroup
*parent_mem
= NULL
;
3295 parent_mem
= mem_cgroup_from_cont(parent
);
3299 * If parent's use_hierarchy is set, we can't make any modifications
3300 * in the child subtrees. If it is unset, then the change can
3301 * occur, provided the current cgroup has no children.
3303 * For the root cgroup, parent_mem is NULL, we allow value to be
3304 * set if there are no children.
3306 if ((!parent_mem
|| !parent_mem
->use_hierarchy
) &&
3307 (val
== 1 || val
== 0)) {
3308 if (list_empty(&cont
->children
))
3309 mem
->use_hierarchy
= val
;
3320 static u64
mem_cgroup_get_recursive_idx_stat(struct mem_cgroup
*mem
,
3321 enum mem_cgroup_stat_index idx
)
3323 struct mem_cgroup
*iter
;
3326 /* each per cpu's value can be minus.Then, use s64 */
3327 for_each_mem_cgroup_tree(iter
, mem
)
3328 val
+= mem_cgroup_read_stat(iter
, idx
);
3330 if (val
< 0) /* race ? */
3335 static inline u64
mem_cgroup_usage(struct mem_cgroup
*mem
, bool swap
)
3339 if (!mem_cgroup_is_root(mem
)) {
3341 return res_counter_read_u64(&mem
->res
, RES_USAGE
);
3343 return res_counter_read_u64(&mem
->memsw
, RES_USAGE
);
3346 val
= mem_cgroup_get_recursive_idx_stat(mem
, MEM_CGROUP_STAT_CACHE
);
3347 val
+= mem_cgroup_get_recursive_idx_stat(mem
, MEM_CGROUP_STAT_RSS
);
3350 val
+= mem_cgroup_get_recursive_idx_stat(mem
,
3351 MEM_CGROUP_STAT_SWAPOUT
);
3353 return val
<< PAGE_SHIFT
;
3356 static u64
mem_cgroup_read(struct cgroup
*cont
, struct cftype
*cft
)
3358 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cont
);
3362 type
= MEMFILE_TYPE(cft
->private);
3363 name
= MEMFILE_ATTR(cft
->private);
3366 if (name
== RES_USAGE
)
3367 val
= mem_cgroup_usage(mem
, false);
3369 val
= res_counter_read_u64(&mem
->res
, name
);
3372 if (name
== RES_USAGE
)
3373 val
= mem_cgroup_usage(mem
, true);
3375 val
= res_counter_read_u64(&mem
->memsw
, name
);
3384 * The user of this function is...
3387 static int mem_cgroup_write(struct cgroup
*cont
, struct cftype
*cft
,
3390 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cont
);
3392 unsigned long long val
;
3395 type
= MEMFILE_TYPE(cft
->private);
3396 name
= MEMFILE_ATTR(cft
->private);
3399 if (mem_cgroup_is_root(memcg
)) { /* Can't set limit on root */
3403 /* This function does all necessary parse...reuse it */
3404 ret
= res_counter_memparse_write_strategy(buffer
, &val
);
3408 ret
= mem_cgroup_resize_limit(memcg
, val
);
3410 ret
= mem_cgroup_resize_memsw_limit(memcg
, val
);
3412 case RES_SOFT_LIMIT
:
3413 ret
= res_counter_memparse_write_strategy(buffer
, &val
);
3417 * For memsw, soft limits are hard to implement in terms
3418 * of semantics, for now, we support soft limits for
3419 * control without swap
3422 ret
= res_counter_set_soft_limit(&memcg
->res
, val
);
3427 ret
= -EINVAL
; /* should be BUG() ? */
3433 static void memcg_get_hierarchical_limit(struct mem_cgroup
*memcg
,
3434 unsigned long long *mem_limit
, unsigned long long *memsw_limit
)
3436 struct cgroup
*cgroup
;
3437 unsigned long long min_limit
, min_memsw_limit
, tmp
;
3439 min_limit
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
3440 min_memsw_limit
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
3441 cgroup
= memcg
->css
.cgroup
;
3442 if (!memcg
->use_hierarchy
)
3445 while (cgroup
->parent
) {
3446 cgroup
= cgroup
->parent
;
3447 memcg
= mem_cgroup_from_cont(cgroup
);
3448 if (!memcg
->use_hierarchy
)
3450 tmp
= res_counter_read_u64(&memcg
->res
, RES_LIMIT
);
3451 min_limit
= min(min_limit
, tmp
);
3452 tmp
= res_counter_read_u64(&memcg
->memsw
, RES_LIMIT
);
3453 min_memsw_limit
= min(min_memsw_limit
, tmp
);
3456 *mem_limit
= min_limit
;
3457 *memsw_limit
= min_memsw_limit
;
3461 static int mem_cgroup_reset(struct cgroup
*cont
, unsigned int event
)
3463 struct mem_cgroup
*mem
;
3466 mem
= mem_cgroup_from_cont(cont
);
3467 type
= MEMFILE_TYPE(event
);
3468 name
= MEMFILE_ATTR(event
);
3472 res_counter_reset_max(&mem
->res
);
3474 res_counter_reset_max(&mem
->memsw
);
3478 res_counter_reset_failcnt(&mem
->res
);
3480 res_counter_reset_failcnt(&mem
->memsw
);
3487 static u64
mem_cgroup_move_charge_read(struct cgroup
*cgrp
,
3490 return mem_cgroup_from_cont(cgrp
)->move_charge_at_immigrate
;
3494 static int mem_cgroup_move_charge_write(struct cgroup
*cgrp
,
3495 struct cftype
*cft
, u64 val
)
3497 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cgrp
);
3499 if (val
>= (1 << NR_MOVE_TYPE
))
3502 * We check this value several times in both in can_attach() and
3503 * attach(), so we need cgroup lock to prevent this value from being
3507 mem
->move_charge_at_immigrate
= val
;
3513 static int mem_cgroup_move_charge_write(struct cgroup
*cgrp
,
3514 struct cftype
*cft
, u64 val
)
3521 /* For read statistics */
3537 struct mcs_total_stat
{
3538 s64 stat
[NR_MCS_STAT
];
3544 } memcg_stat_strings
[NR_MCS_STAT
] = {
3545 {"cache", "total_cache"},
3546 {"rss", "total_rss"},
3547 {"mapped_file", "total_mapped_file"},
3548 {"pgpgin", "total_pgpgin"},
3549 {"pgpgout", "total_pgpgout"},
3550 {"swap", "total_swap"},
3551 {"inactive_anon", "total_inactive_anon"},
3552 {"active_anon", "total_active_anon"},
3553 {"inactive_file", "total_inactive_file"},
3554 {"active_file", "total_active_file"},
3555 {"unevictable", "total_unevictable"}
3560 mem_cgroup_get_local_stat(struct mem_cgroup
*mem
, struct mcs_total_stat
*s
)
3565 val
= mem_cgroup_read_stat(mem
, MEM_CGROUP_STAT_CACHE
);
3566 s
->stat
[MCS_CACHE
] += val
* PAGE_SIZE
;
3567 val
= mem_cgroup_read_stat(mem
, MEM_CGROUP_STAT_RSS
);
3568 s
->stat
[MCS_RSS
] += val
* PAGE_SIZE
;
3569 val
= mem_cgroup_read_stat(mem
, MEM_CGROUP_STAT_FILE_MAPPED
);
3570 s
->stat
[MCS_FILE_MAPPED
] += val
* PAGE_SIZE
;
3571 val
= mem_cgroup_read_stat(mem
, MEM_CGROUP_STAT_PGPGIN_COUNT
);
3572 s
->stat
[MCS_PGPGIN
] += val
;
3573 val
= mem_cgroup_read_stat(mem
, MEM_CGROUP_STAT_PGPGOUT_COUNT
);
3574 s
->stat
[MCS_PGPGOUT
] += val
;
3575 if (do_swap_account
) {
3576 val
= mem_cgroup_read_stat(mem
, MEM_CGROUP_STAT_SWAPOUT
);
3577 s
->stat
[MCS_SWAP
] += val
* PAGE_SIZE
;
3581 val
= mem_cgroup_get_local_zonestat(mem
, LRU_INACTIVE_ANON
);
3582 s
->stat
[MCS_INACTIVE_ANON
] += val
* PAGE_SIZE
;
3583 val
= mem_cgroup_get_local_zonestat(mem
, LRU_ACTIVE_ANON
);
3584 s
->stat
[MCS_ACTIVE_ANON
] += val
* PAGE_SIZE
;
3585 val
= mem_cgroup_get_local_zonestat(mem
, LRU_INACTIVE_FILE
);
3586 s
->stat
[MCS_INACTIVE_FILE
] += val
* PAGE_SIZE
;
3587 val
= mem_cgroup_get_local_zonestat(mem
, LRU_ACTIVE_FILE
);
3588 s
->stat
[MCS_ACTIVE_FILE
] += val
* PAGE_SIZE
;
3589 val
= mem_cgroup_get_local_zonestat(mem
, LRU_UNEVICTABLE
);
3590 s
->stat
[MCS_UNEVICTABLE
] += val
* PAGE_SIZE
;
3594 mem_cgroup_get_total_stat(struct mem_cgroup
*mem
, struct mcs_total_stat
*s
)
3596 struct mem_cgroup
*iter
;
3598 for_each_mem_cgroup_tree(iter
, mem
)
3599 mem_cgroup_get_local_stat(iter
, s
);
3602 static int mem_control_stat_show(struct cgroup
*cont
, struct cftype
*cft
,
3603 struct cgroup_map_cb
*cb
)
3605 struct mem_cgroup
*mem_cont
= mem_cgroup_from_cont(cont
);
3606 struct mcs_total_stat mystat
;
3609 memset(&mystat
, 0, sizeof(mystat
));
3610 mem_cgroup_get_local_stat(mem_cont
, &mystat
);
3612 for (i
= 0; i
< NR_MCS_STAT
; i
++) {
3613 if (i
== MCS_SWAP
&& !do_swap_account
)
3615 cb
->fill(cb
, memcg_stat_strings
[i
].local_name
, mystat
.stat
[i
]);
3618 /* Hierarchical information */
3620 unsigned long long limit
, memsw_limit
;
3621 memcg_get_hierarchical_limit(mem_cont
, &limit
, &memsw_limit
);
3622 cb
->fill(cb
, "hierarchical_memory_limit", limit
);
3623 if (do_swap_account
)
3624 cb
->fill(cb
, "hierarchical_memsw_limit", memsw_limit
);
3627 memset(&mystat
, 0, sizeof(mystat
));
3628 mem_cgroup_get_total_stat(mem_cont
, &mystat
);
3629 for (i
= 0; i
< NR_MCS_STAT
; i
++) {
3630 if (i
== MCS_SWAP
&& !do_swap_account
)
3632 cb
->fill(cb
, memcg_stat_strings
[i
].total_name
, mystat
.stat
[i
]);
3635 #ifdef CONFIG_DEBUG_VM
3636 cb
->fill(cb
, "inactive_ratio", calc_inactive_ratio(mem_cont
, NULL
));
3640 struct mem_cgroup_per_zone
*mz
;
3641 unsigned long recent_rotated
[2] = {0, 0};
3642 unsigned long recent_scanned
[2] = {0, 0};
3644 for_each_online_node(nid
)
3645 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++) {
3646 mz
= mem_cgroup_zoneinfo(mem_cont
, nid
, zid
);
3648 recent_rotated
[0] +=
3649 mz
->reclaim_stat
.recent_rotated
[0];
3650 recent_rotated
[1] +=
3651 mz
->reclaim_stat
.recent_rotated
[1];
3652 recent_scanned
[0] +=
3653 mz
->reclaim_stat
.recent_scanned
[0];
3654 recent_scanned
[1] +=
3655 mz
->reclaim_stat
.recent_scanned
[1];
3657 cb
->fill(cb
, "recent_rotated_anon", recent_rotated
[0]);
3658 cb
->fill(cb
, "recent_rotated_file", recent_rotated
[1]);
3659 cb
->fill(cb
, "recent_scanned_anon", recent_scanned
[0]);
3660 cb
->fill(cb
, "recent_scanned_file", recent_scanned
[1]);
3667 static u64
mem_cgroup_swappiness_read(struct cgroup
*cgrp
, struct cftype
*cft
)
3669 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
3671 return get_swappiness(memcg
);
3674 static int mem_cgroup_swappiness_write(struct cgroup
*cgrp
, struct cftype
*cft
,
3677 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
3678 struct mem_cgroup
*parent
;
3683 if (cgrp
->parent
== NULL
)
3686 parent
= mem_cgroup_from_cont(cgrp
->parent
);
3690 /* If under hierarchy, only empty-root can set this value */
3691 if ((parent
->use_hierarchy
) ||
3692 (memcg
->use_hierarchy
&& !list_empty(&cgrp
->children
))) {
3697 spin_lock(&memcg
->reclaim_param_lock
);
3698 memcg
->swappiness
= val
;
3699 spin_unlock(&memcg
->reclaim_param_lock
);
3706 static void __mem_cgroup_threshold(struct mem_cgroup
*memcg
, bool swap
)
3708 struct mem_cgroup_threshold_ary
*t
;
3714 t
= rcu_dereference(memcg
->thresholds
.primary
);
3716 t
= rcu_dereference(memcg
->memsw_thresholds
.primary
);
3721 usage
= mem_cgroup_usage(memcg
, swap
);
3724 * current_threshold points to threshold just below usage.
3725 * If it's not true, a threshold was crossed after last
3726 * call of __mem_cgroup_threshold().
3728 i
= t
->current_threshold
;
3731 * Iterate backward over array of thresholds starting from
3732 * current_threshold and check if a threshold is crossed.
3733 * If none of thresholds below usage is crossed, we read
3734 * only one element of the array here.
3736 for (; i
>= 0 && unlikely(t
->entries
[i
].threshold
> usage
); i
--)
3737 eventfd_signal(t
->entries
[i
].eventfd
, 1);
3739 /* i = current_threshold + 1 */
3743 * Iterate forward over array of thresholds starting from
3744 * current_threshold+1 and check if a threshold is crossed.
3745 * If none of thresholds above usage is crossed, we read
3746 * only one element of the array here.
3748 for (; i
< t
->size
&& unlikely(t
->entries
[i
].threshold
<= usage
); i
++)
3749 eventfd_signal(t
->entries
[i
].eventfd
, 1);
3751 /* Update current_threshold */
3752 t
->current_threshold
= i
- 1;
3757 static void mem_cgroup_threshold(struct mem_cgroup
*memcg
)
3760 __mem_cgroup_threshold(memcg
, false);
3761 if (do_swap_account
)
3762 __mem_cgroup_threshold(memcg
, true);
3764 memcg
= parent_mem_cgroup(memcg
);
3768 static int compare_thresholds(const void *a
, const void *b
)
3770 const struct mem_cgroup_threshold
*_a
= a
;
3771 const struct mem_cgroup_threshold
*_b
= b
;
3773 return _a
->threshold
- _b
->threshold
;
3776 static int mem_cgroup_oom_notify_cb(struct mem_cgroup
*mem
)
3778 struct mem_cgroup_eventfd_list
*ev
;
3780 list_for_each_entry(ev
, &mem
->oom_notify
, list
)
3781 eventfd_signal(ev
->eventfd
, 1);
3785 static void mem_cgroup_oom_notify(struct mem_cgroup
*mem
)
3787 struct mem_cgroup
*iter
;
3789 for_each_mem_cgroup_tree(iter
, mem
)
3790 mem_cgroup_oom_notify_cb(iter
);
3793 static int mem_cgroup_usage_register_event(struct cgroup
*cgrp
,
3794 struct cftype
*cft
, struct eventfd_ctx
*eventfd
, const char *args
)
3796 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
3797 struct mem_cgroup_thresholds
*thresholds
;
3798 struct mem_cgroup_threshold_ary
*new;
3799 int type
= MEMFILE_TYPE(cft
->private);
3800 u64 threshold
, usage
;
3803 ret
= res_counter_memparse_write_strategy(args
, &threshold
);
3807 mutex_lock(&memcg
->thresholds_lock
);
3810 thresholds
= &memcg
->thresholds
;
3811 else if (type
== _MEMSWAP
)
3812 thresholds
= &memcg
->memsw_thresholds
;
3816 usage
= mem_cgroup_usage(memcg
, type
== _MEMSWAP
);
3818 /* Check if a threshold crossed before adding a new one */
3819 if (thresholds
->primary
)
3820 __mem_cgroup_threshold(memcg
, type
== _MEMSWAP
);
3822 size
= thresholds
->primary
? thresholds
->primary
->size
+ 1 : 1;
3824 /* Allocate memory for new array of thresholds */
3825 new = kmalloc(sizeof(*new) + size
* sizeof(struct mem_cgroup_threshold
),
3833 /* Copy thresholds (if any) to new array */
3834 if (thresholds
->primary
) {
3835 memcpy(new->entries
, thresholds
->primary
->entries
, (size
- 1) *
3836 sizeof(struct mem_cgroup_threshold
));
3839 /* Add new threshold */
3840 new->entries
[size
- 1].eventfd
= eventfd
;
3841 new->entries
[size
- 1].threshold
= threshold
;
3843 /* Sort thresholds. Registering of new threshold isn't time-critical */
3844 sort(new->entries
, size
, sizeof(struct mem_cgroup_threshold
),
3845 compare_thresholds
, NULL
);
3847 /* Find current threshold */
3848 new->current_threshold
= -1;
3849 for (i
= 0; i
< size
; i
++) {
3850 if (new->entries
[i
].threshold
< usage
) {
3852 * new->current_threshold will not be used until
3853 * rcu_assign_pointer(), so it's safe to increment
3856 ++new->current_threshold
;
3860 /* Free old spare buffer and save old primary buffer as spare */
3861 kfree(thresholds
->spare
);
3862 thresholds
->spare
= thresholds
->primary
;
3864 rcu_assign_pointer(thresholds
->primary
, new);
3866 /* To be sure that nobody uses thresholds */
3870 mutex_unlock(&memcg
->thresholds_lock
);
3875 static void mem_cgroup_usage_unregister_event(struct cgroup
*cgrp
,
3876 struct cftype
*cft
, struct eventfd_ctx
*eventfd
)
3878 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
3879 struct mem_cgroup_thresholds
*thresholds
;
3880 struct mem_cgroup_threshold_ary
*new;
3881 int type
= MEMFILE_TYPE(cft
->private);
3885 mutex_lock(&memcg
->thresholds_lock
);
3887 thresholds
= &memcg
->thresholds
;
3888 else if (type
== _MEMSWAP
)
3889 thresholds
= &memcg
->memsw_thresholds
;
3894 * Something went wrong if we trying to unregister a threshold
3895 * if we don't have thresholds
3897 BUG_ON(!thresholds
);
3899 usage
= mem_cgroup_usage(memcg
, type
== _MEMSWAP
);
3901 /* Check if a threshold crossed before removing */
3902 __mem_cgroup_threshold(memcg
, type
== _MEMSWAP
);
3904 /* Calculate new number of threshold */
3906 for (i
= 0; i
< thresholds
->primary
->size
; i
++) {
3907 if (thresholds
->primary
->entries
[i
].eventfd
!= eventfd
)
3911 new = thresholds
->spare
;
3913 /* Set thresholds array to NULL if we don't have thresholds */
3922 /* Copy thresholds and find current threshold */
3923 new->current_threshold
= -1;
3924 for (i
= 0, j
= 0; i
< thresholds
->primary
->size
; i
++) {
3925 if (thresholds
->primary
->entries
[i
].eventfd
== eventfd
)
3928 new->entries
[j
] = thresholds
->primary
->entries
[i
];
3929 if (new->entries
[j
].threshold
< usage
) {
3931 * new->current_threshold will not be used
3932 * until rcu_assign_pointer(), so it's safe to increment
3935 ++new->current_threshold
;
3941 /* Swap primary and spare array */
3942 thresholds
->spare
= thresholds
->primary
;
3943 rcu_assign_pointer(thresholds
->primary
, new);
3945 /* To be sure that nobody uses thresholds */
3948 mutex_unlock(&memcg
->thresholds_lock
);
3951 static int mem_cgroup_oom_register_event(struct cgroup
*cgrp
,
3952 struct cftype
*cft
, struct eventfd_ctx
*eventfd
, const char *args
)
3954 struct mem_cgroup
*memcg
= mem_cgroup_from_cont(cgrp
);
3955 struct mem_cgroup_eventfd_list
*event
;
3956 int type
= MEMFILE_TYPE(cft
->private);
3958 BUG_ON(type
!= _OOM_TYPE
);
3959 event
= kmalloc(sizeof(*event
), GFP_KERNEL
);
3963 mutex_lock(&memcg_oom_mutex
);
3965 event
->eventfd
= eventfd
;
3966 list_add(&event
->list
, &memcg
->oom_notify
);
3968 /* already in OOM ? */
3969 if (atomic_read(&memcg
->oom_lock
))
3970 eventfd_signal(eventfd
, 1);
3971 mutex_unlock(&memcg_oom_mutex
);
3976 static void mem_cgroup_oom_unregister_event(struct cgroup
*cgrp
,
3977 struct cftype
*cft
, struct eventfd_ctx
*eventfd
)
3979 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cgrp
);
3980 struct mem_cgroup_eventfd_list
*ev
, *tmp
;
3981 int type
= MEMFILE_TYPE(cft
->private);
3983 BUG_ON(type
!= _OOM_TYPE
);
3985 mutex_lock(&memcg_oom_mutex
);
3987 list_for_each_entry_safe(ev
, tmp
, &mem
->oom_notify
, list
) {
3988 if (ev
->eventfd
== eventfd
) {
3989 list_del(&ev
->list
);
3994 mutex_unlock(&memcg_oom_mutex
);
3997 static int mem_cgroup_oom_control_read(struct cgroup
*cgrp
,
3998 struct cftype
*cft
, struct cgroup_map_cb
*cb
)
4000 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cgrp
);
4002 cb
->fill(cb
, "oom_kill_disable", mem
->oom_kill_disable
);
4004 if (atomic_read(&mem
->oom_lock
))
4005 cb
->fill(cb
, "under_oom", 1);
4007 cb
->fill(cb
, "under_oom", 0);
4011 static int mem_cgroup_oom_control_write(struct cgroup
*cgrp
,
4012 struct cftype
*cft
, u64 val
)
4014 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cgrp
);
4015 struct mem_cgroup
*parent
;
4017 /* cannot set to root cgroup and only 0 and 1 are allowed */
4018 if (!cgrp
->parent
|| !((val
== 0) || (val
== 1)))
4021 parent
= mem_cgroup_from_cont(cgrp
->parent
);
4024 /* oom-kill-disable is a flag for subhierarchy. */
4025 if ((parent
->use_hierarchy
) ||
4026 (mem
->use_hierarchy
&& !list_empty(&cgrp
->children
))) {
4030 mem
->oom_kill_disable
= val
;
4032 memcg_oom_recover(mem
);
4037 static struct cftype mem_cgroup_files
[] = {
4039 .name
= "usage_in_bytes",
4040 .private = MEMFILE_PRIVATE(_MEM
, RES_USAGE
),
4041 .read_u64
= mem_cgroup_read
,
4042 .register_event
= mem_cgroup_usage_register_event
,
4043 .unregister_event
= mem_cgroup_usage_unregister_event
,
4046 .name
= "max_usage_in_bytes",
4047 .private = MEMFILE_PRIVATE(_MEM
, RES_MAX_USAGE
),
4048 .trigger
= mem_cgroup_reset
,
4049 .read_u64
= mem_cgroup_read
,
4052 .name
= "limit_in_bytes",
4053 .private = MEMFILE_PRIVATE(_MEM
, RES_LIMIT
),
4054 .write_string
= mem_cgroup_write
,
4055 .read_u64
= mem_cgroup_read
,
4058 .name
= "soft_limit_in_bytes",
4059 .private = MEMFILE_PRIVATE(_MEM
, RES_SOFT_LIMIT
),
4060 .write_string
= mem_cgroup_write
,
4061 .read_u64
= mem_cgroup_read
,
4065 .private = MEMFILE_PRIVATE(_MEM
, RES_FAILCNT
),
4066 .trigger
= mem_cgroup_reset
,
4067 .read_u64
= mem_cgroup_read
,
4071 .read_map
= mem_control_stat_show
,
4074 .name
= "force_empty",
4075 .trigger
= mem_cgroup_force_empty_write
,
4078 .name
= "use_hierarchy",
4079 .write_u64
= mem_cgroup_hierarchy_write
,
4080 .read_u64
= mem_cgroup_hierarchy_read
,
4083 .name
= "swappiness",
4084 .read_u64
= mem_cgroup_swappiness_read
,
4085 .write_u64
= mem_cgroup_swappiness_write
,
4088 .name
= "move_charge_at_immigrate",
4089 .read_u64
= mem_cgroup_move_charge_read
,
4090 .write_u64
= mem_cgroup_move_charge_write
,
4093 .name
= "oom_control",
4094 .read_map
= mem_cgroup_oom_control_read
,
4095 .write_u64
= mem_cgroup_oom_control_write
,
4096 .register_event
= mem_cgroup_oom_register_event
,
4097 .unregister_event
= mem_cgroup_oom_unregister_event
,
4098 .private = MEMFILE_PRIVATE(_OOM_TYPE
, OOM_CONTROL
),
4102 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
4103 static struct cftype memsw_cgroup_files
[] = {
4105 .name
= "memsw.usage_in_bytes",
4106 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_USAGE
),
4107 .read_u64
= mem_cgroup_read
,
4108 .register_event
= mem_cgroup_usage_register_event
,
4109 .unregister_event
= mem_cgroup_usage_unregister_event
,
4112 .name
= "memsw.max_usage_in_bytes",
4113 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_MAX_USAGE
),
4114 .trigger
= mem_cgroup_reset
,
4115 .read_u64
= mem_cgroup_read
,
4118 .name
= "memsw.limit_in_bytes",
4119 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_LIMIT
),
4120 .write_string
= mem_cgroup_write
,
4121 .read_u64
= mem_cgroup_read
,
4124 .name
= "memsw.failcnt",
4125 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_FAILCNT
),
4126 .trigger
= mem_cgroup_reset
,
4127 .read_u64
= mem_cgroup_read
,
4131 static int register_memsw_files(struct cgroup
*cont
, struct cgroup_subsys
*ss
)
4133 if (!do_swap_account
)
4135 return cgroup_add_files(cont
, ss
, memsw_cgroup_files
,
4136 ARRAY_SIZE(memsw_cgroup_files
));
4139 static int register_memsw_files(struct cgroup
*cont
, struct cgroup_subsys
*ss
)
4145 static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup
*mem
, int node
)
4147 struct mem_cgroup_per_node
*pn
;
4148 struct mem_cgroup_per_zone
*mz
;
4150 int zone
, tmp
= node
;
4152 * This routine is called against possible nodes.
4153 * But it's BUG to call kmalloc() against offline node.
4155 * TODO: this routine can waste much memory for nodes which will
4156 * never be onlined. It's better to use memory hotplug callback
4159 if (!node_state(node
, N_NORMAL_MEMORY
))
4161 pn
= kmalloc_node(sizeof(*pn
), GFP_KERNEL
, tmp
);
4165 mem
->info
.nodeinfo
[node
] = pn
;
4166 memset(pn
, 0, sizeof(*pn
));
4168 for (zone
= 0; zone
< MAX_NR_ZONES
; zone
++) {
4169 mz
= &pn
->zoneinfo
[zone
];
4171 INIT_LIST_HEAD(&mz
->lists
[l
]);
4172 mz
->usage_in_excess
= 0;
4173 mz
->on_tree
= false;
4179 static void free_mem_cgroup_per_zone_info(struct mem_cgroup
*mem
, int node
)
4181 kfree(mem
->info
.nodeinfo
[node
]);
4184 static struct mem_cgroup
*mem_cgroup_alloc(void)
4186 struct mem_cgroup
*mem
;
4187 int size
= sizeof(struct mem_cgroup
);
4189 /* Can be very big if MAX_NUMNODES is very big */
4190 if (size
< PAGE_SIZE
)
4191 mem
= kmalloc(size
, GFP_KERNEL
);
4193 mem
= vmalloc(size
);
4198 memset(mem
, 0, size
);
4199 mem
->stat
= alloc_percpu(struct mem_cgroup_stat_cpu
);
4201 if (size
< PAGE_SIZE
)
4207 spin_lock_init(&mem
->pcp_counter_lock
);
4212 * At destroying mem_cgroup, references from swap_cgroup can remain.
4213 * (scanning all at force_empty is too costly...)
4215 * Instead of clearing all references at force_empty, we remember
4216 * the number of reference from swap_cgroup and free mem_cgroup when
4217 * it goes down to 0.
4219 * Removal of cgroup itself succeeds regardless of refs from swap.
4222 static void __mem_cgroup_free(struct mem_cgroup
*mem
)
4226 mem_cgroup_remove_from_trees(mem
);
4227 free_css_id(&mem_cgroup_subsys
, &mem
->css
);
4229 for_each_node_state(node
, N_POSSIBLE
)
4230 free_mem_cgroup_per_zone_info(mem
, node
);
4232 free_percpu(mem
->stat
);
4233 if (sizeof(struct mem_cgroup
) < PAGE_SIZE
)
4239 static void mem_cgroup_get(struct mem_cgroup
*mem
)
4241 atomic_inc(&mem
->refcnt
);
4244 static void __mem_cgroup_put(struct mem_cgroup
*mem
, int count
)
4246 if (atomic_sub_and_test(count
, &mem
->refcnt
)) {
4247 struct mem_cgroup
*parent
= parent_mem_cgroup(mem
);
4248 __mem_cgroup_free(mem
);
4250 mem_cgroup_put(parent
);
4254 static void mem_cgroup_put(struct mem_cgroup
*mem
)
4256 __mem_cgroup_put(mem
, 1);
4260 * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled.
4262 static struct mem_cgroup
*parent_mem_cgroup(struct mem_cgroup
*mem
)
4264 if (!mem
->res
.parent
)
4266 return mem_cgroup_from_res_counter(mem
->res
.parent
, res
);
4269 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
4270 static void __init
enable_swap_cgroup(void)
4272 if (!mem_cgroup_disabled() && really_do_swap_account
)
4273 do_swap_account
= 1;
4276 static void __init
enable_swap_cgroup(void)
4281 static int mem_cgroup_soft_limit_tree_init(void)
4283 struct mem_cgroup_tree_per_node
*rtpn
;
4284 struct mem_cgroup_tree_per_zone
*rtpz
;
4285 int tmp
, node
, zone
;
4287 for_each_node_state(node
, N_POSSIBLE
) {
4289 if (!node_state(node
, N_NORMAL_MEMORY
))
4291 rtpn
= kzalloc_node(sizeof(*rtpn
), GFP_KERNEL
, tmp
);
4295 soft_limit_tree
.rb_tree_per_node
[node
] = rtpn
;
4297 for (zone
= 0; zone
< MAX_NR_ZONES
; zone
++) {
4298 rtpz
= &rtpn
->rb_tree_per_zone
[zone
];
4299 rtpz
->rb_root
= RB_ROOT
;
4300 spin_lock_init(&rtpz
->lock
);
4306 static struct cgroup_subsys_state
* __ref
4307 mem_cgroup_create(struct cgroup_subsys
*ss
, struct cgroup
*cont
)
4309 struct mem_cgroup
*mem
, *parent
;
4310 long error
= -ENOMEM
;
4313 mem
= mem_cgroup_alloc();
4315 return ERR_PTR(error
);
4317 for_each_node_state(node
, N_POSSIBLE
)
4318 if (alloc_mem_cgroup_per_zone_info(mem
, node
))
4322 if (cont
->parent
== NULL
) {
4324 enable_swap_cgroup();
4326 root_mem_cgroup
= mem
;
4327 if (mem_cgroup_soft_limit_tree_init())
4329 for_each_possible_cpu(cpu
) {
4330 struct memcg_stock_pcp
*stock
=
4331 &per_cpu(memcg_stock
, cpu
);
4332 INIT_WORK(&stock
->work
, drain_local_stock
);
4334 hotcpu_notifier(memcg_cpu_hotplug_callback
, 0);
4336 parent
= mem_cgroup_from_cont(cont
->parent
);
4337 mem
->use_hierarchy
= parent
->use_hierarchy
;
4338 mem
->oom_kill_disable
= parent
->oom_kill_disable
;
4341 if (parent
&& parent
->use_hierarchy
) {
4342 res_counter_init(&mem
->res
, &parent
->res
);
4343 res_counter_init(&mem
->memsw
, &parent
->memsw
);
4345 * We increment refcnt of the parent to ensure that we can
4346 * safely access it on res_counter_charge/uncharge.
4347 * This refcnt will be decremented when freeing this
4348 * mem_cgroup(see mem_cgroup_put).
4350 mem_cgroup_get(parent
);
4352 res_counter_init(&mem
->res
, NULL
);
4353 res_counter_init(&mem
->memsw
, NULL
);
4355 mem
->last_scanned_child
= 0;
4356 spin_lock_init(&mem
->reclaim_param_lock
);
4357 INIT_LIST_HEAD(&mem
->oom_notify
);
4360 mem
->swappiness
= get_swappiness(parent
);
4361 atomic_set(&mem
->refcnt
, 1);
4362 mem
->move_charge_at_immigrate
= 0;
4363 mutex_init(&mem
->thresholds_lock
);
4366 __mem_cgroup_free(mem
);
4367 root_mem_cgroup
= NULL
;
4368 return ERR_PTR(error
);
4371 static int mem_cgroup_pre_destroy(struct cgroup_subsys
*ss
,
4372 struct cgroup
*cont
)
4374 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cont
);
4376 return mem_cgroup_force_empty(mem
, false);
4379 static void mem_cgroup_destroy(struct cgroup_subsys
*ss
,
4380 struct cgroup
*cont
)
4382 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cont
);
4384 mem_cgroup_put(mem
);
4387 static int mem_cgroup_populate(struct cgroup_subsys
*ss
,
4388 struct cgroup
*cont
)
4392 ret
= cgroup_add_files(cont
, ss
, mem_cgroup_files
,
4393 ARRAY_SIZE(mem_cgroup_files
));
4396 ret
= register_memsw_files(cont
, ss
);
4401 /* Handlers for move charge at task migration. */
4402 #define PRECHARGE_COUNT_AT_ONCE 256
4403 static int mem_cgroup_do_precharge(unsigned long count
)
4406 int batch_count
= PRECHARGE_COUNT_AT_ONCE
;
4407 struct mem_cgroup
*mem
= mc
.to
;
4409 if (mem_cgroup_is_root(mem
)) {
4410 mc
.precharge
+= count
;
4411 /* we don't need css_get for root */
4414 /* try to charge at once */
4416 struct res_counter
*dummy
;
4418 * "mem" cannot be under rmdir() because we've already checked
4419 * by cgroup_lock_live_cgroup() that it is not removed and we
4420 * are still under the same cgroup_mutex. So we can postpone
4423 if (res_counter_charge(&mem
->res
, PAGE_SIZE
* count
, &dummy
))
4425 if (do_swap_account
&& res_counter_charge(&mem
->memsw
,
4426 PAGE_SIZE
* count
, &dummy
)) {
4427 res_counter_uncharge(&mem
->res
, PAGE_SIZE
* count
);
4430 mc
.precharge
+= count
;
4434 /* fall back to one by one charge */
4436 if (signal_pending(current
)) {
4440 if (!batch_count
--) {
4441 batch_count
= PRECHARGE_COUNT_AT_ONCE
;
4444 ret
= __mem_cgroup_try_charge(NULL
, GFP_KERNEL
, &mem
, false);
4446 /* mem_cgroup_clear_mc() will do uncharge later */
4454 * is_target_pte_for_mc - check a pte whether it is valid for move charge
4455 * @vma: the vma the pte to be checked belongs
4456 * @addr: the address corresponding to the pte to be checked
4457 * @ptent: the pte to be checked
4458 * @target: the pointer the target page or swap ent will be stored(can be NULL)
4461 * 0(MC_TARGET_NONE): if the pte is not a target for move charge.
4462 * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
4463 * move charge. if @target is not NULL, the page is stored in target->page
4464 * with extra refcnt got(Callers should handle it).
4465 * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
4466 * target for charge migration. if @target is not NULL, the entry is stored
4469 * Called with pte lock held.
4476 enum mc_target_type
{
4477 MC_TARGET_NONE
, /* not used */
4482 static struct page
*mc_handle_present_pte(struct vm_area_struct
*vma
,
4483 unsigned long addr
, pte_t ptent
)
4485 struct page
*page
= vm_normal_page(vma
, addr
, ptent
);
4487 if (!page
|| !page_mapped(page
))
4489 if (PageAnon(page
)) {
4490 /* we don't move shared anon */
4491 if (!move_anon() || page_mapcount(page
) > 2)
4493 } else if (!move_file())
4494 /* we ignore mapcount for file pages */
4496 if (!get_page_unless_zero(page
))
4502 static struct page
*mc_handle_swap_pte(struct vm_area_struct
*vma
,
4503 unsigned long addr
, pte_t ptent
, swp_entry_t
*entry
)
4506 struct page
*page
= NULL
;
4507 swp_entry_t ent
= pte_to_swp_entry(ptent
);
4509 if (!move_anon() || non_swap_entry(ent
))
4511 usage_count
= mem_cgroup_count_swap_user(ent
, &page
);
4512 if (usage_count
> 1) { /* we don't move shared anon */
4517 if (do_swap_account
)
4518 entry
->val
= ent
.val
;
4523 static struct page
*mc_handle_file_pte(struct vm_area_struct
*vma
,
4524 unsigned long addr
, pte_t ptent
, swp_entry_t
*entry
)
4526 struct page
*page
= NULL
;
4527 struct inode
*inode
;
4528 struct address_space
*mapping
;
4531 if (!vma
->vm_file
) /* anonymous vma */
4536 inode
= vma
->vm_file
->f_path
.dentry
->d_inode
;
4537 mapping
= vma
->vm_file
->f_mapping
;
4538 if (pte_none(ptent
))
4539 pgoff
= linear_page_index(vma
, addr
);
4540 else /* pte_file(ptent) is true */
4541 pgoff
= pte_to_pgoff(ptent
);
4543 /* page is moved even if it's not RSS of this task(page-faulted). */
4544 if (!mapping_cap_swap_backed(mapping
)) { /* normal file */
4545 page
= find_get_page(mapping
, pgoff
);
4546 } else { /* shmem/tmpfs file. we should take account of swap too. */
4548 mem_cgroup_get_shmem_target(inode
, pgoff
, &page
, &ent
);
4549 if (do_swap_account
)
4550 entry
->val
= ent
.val
;
4556 static int is_target_pte_for_mc(struct vm_area_struct
*vma
,
4557 unsigned long addr
, pte_t ptent
, union mc_target
*target
)
4559 struct page
*page
= NULL
;
4560 struct page_cgroup
*pc
;
4562 swp_entry_t ent
= { .val
= 0 };
4564 if (pte_present(ptent
))
4565 page
= mc_handle_present_pte(vma
, addr
, ptent
);
4566 else if (is_swap_pte(ptent
))
4567 page
= mc_handle_swap_pte(vma
, addr
, ptent
, &ent
);
4568 else if (pte_none(ptent
) || pte_file(ptent
))
4569 page
= mc_handle_file_pte(vma
, addr
, ptent
, &ent
);
4571 if (!page
&& !ent
.val
)
4574 pc
= lookup_page_cgroup(page
);
4576 * Do only loose check w/o page_cgroup lock.
4577 * mem_cgroup_move_account() checks the pc is valid or not under
4580 if (PageCgroupUsed(pc
) && pc
->mem_cgroup
== mc
.from
) {
4581 ret
= MC_TARGET_PAGE
;
4583 target
->page
= page
;
4585 if (!ret
|| !target
)
4588 /* There is a swap entry and a page doesn't exist or isn't charged */
4589 if (ent
.val
&& !ret
&&
4590 css_id(&mc
.from
->css
) == lookup_swap_cgroup(ent
)) {
4591 ret
= MC_TARGET_SWAP
;
4598 static int mem_cgroup_count_precharge_pte_range(pmd_t
*pmd
,
4599 unsigned long addr
, unsigned long end
,
4600 struct mm_walk
*walk
)
4602 struct vm_area_struct
*vma
= walk
->private;
4606 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
4607 for (; addr
!= end
; pte
++, addr
+= PAGE_SIZE
)
4608 if (is_target_pte_for_mc(vma
, addr
, *pte
, NULL
))
4609 mc
.precharge
++; /* increment precharge temporarily */
4610 pte_unmap_unlock(pte
- 1, ptl
);
4616 static unsigned long mem_cgroup_count_precharge(struct mm_struct
*mm
)
4618 unsigned long precharge
;
4619 struct vm_area_struct
*vma
;
4621 down_read(&mm
->mmap_sem
);
4622 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
4623 struct mm_walk mem_cgroup_count_precharge_walk
= {
4624 .pmd_entry
= mem_cgroup_count_precharge_pte_range
,
4628 if (is_vm_hugetlb_page(vma
))
4630 walk_page_range(vma
->vm_start
, vma
->vm_end
,
4631 &mem_cgroup_count_precharge_walk
);
4633 up_read(&mm
->mmap_sem
);
4635 precharge
= mc
.precharge
;
4641 static int mem_cgroup_precharge_mc(struct mm_struct
*mm
)
4643 return mem_cgroup_do_precharge(mem_cgroup_count_precharge(mm
));
4646 static void mem_cgroup_clear_mc(void)
4648 struct mem_cgroup
*from
= mc
.from
;
4649 struct mem_cgroup
*to
= mc
.to
;
4651 /* we must uncharge all the leftover precharges from mc.to */
4653 __mem_cgroup_cancel_charge(mc
.to
, mc
.precharge
);
4657 * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
4658 * we must uncharge here.
4660 if (mc
.moved_charge
) {
4661 __mem_cgroup_cancel_charge(mc
.from
, mc
.moved_charge
);
4662 mc
.moved_charge
= 0;
4664 /* we must fixup refcnts and charges */
4665 if (mc
.moved_swap
) {
4666 /* uncharge swap account from the old cgroup */
4667 if (!mem_cgroup_is_root(mc
.from
))
4668 res_counter_uncharge(&mc
.from
->memsw
,
4669 PAGE_SIZE
* mc
.moved_swap
);
4670 __mem_cgroup_put(mc
.from
, mc
.moved_swap
);
4672 if (!mem_cgroup_is_root(mc
.to
)) {
4674 * we charged both to->res and to->memsw, so we should
4677 res_counter_uncharge(&mc
.to
->res
,
4678 PAGE_SIZE
* mc
.moved_swap
);
4680 /* we've already done mem_cgroup_get(mc.to) */
4684 spin_lock(&mc
.lock
);
4687 mc
.moving_task
= NULL
;
4688 spin_unlock(&mc
.lock
);
4689 mem_cgroup_end_move(from
);
4690 memcg_oom_recover(from
);
4691 memcg_oom_recover(to
);
4692 wake_up_all(&mc
.waitq
);
4695 static int mem_cgroup_can_attach(struct cgroup_subsys
*ss
,
4696 struct cgroup
*cgroup
,
4697 struct task_struct
*p
,
4701 struct mem_cgroup
*mem
= mem_cgroup_from_cont(cgroup
);
4703 if (mem
->move_charge_at_immigrate
) {
4704 struct mm_struct
*mm
;
4705 struct mem_cgroup
*from
= mem_cgroup_from_task(p
);
4707 VM_BUG_ON(from
== mem
);
4709 mm
= get_task_mm(p
);
4712 /* We move charges only when we move a owner of the mm */
4713 if (mm
->owner
== p
) {
4716 VM_BUG_ON(mc
.precharge
);
4717 VM_BUG_ON(mc
.moved_charge
);
4718 VM_BUG_ON(mc
.moved_swap
);
4719 VM_BUG_ON(mc
.moving_task
);
4720 mem_cgroup_start_move(from
);
4721 spin_lock(&mc
.lock
);
4725 mc
.moved_charge
= 0;
4727 mc
.moving_task
= current
;
4728 spin_unlock(&mc
.lock
);
4730 ret
= mem_cgroup_precharge_mc(mm
);
4732 mem_cgroup_clear_mc();
4739 static void mem_cgroup_cancel_attach(struct cgroup_subsys
*ss
,
4740 struct cgroup
*cgroup
,
4741 struct task_struct
*p
,
4744 mem_cgroup_clear_mc();
4747 static int mem_cgroup_move_charge_pte_range(pmd_t
*pmd
,
4748 unsigned long addr
, unsigned long end
,
4749 struct mm_walk
*walk
)
4752 struct vm_area_struct
*vma
= walk
->private;
4757 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
4758 for (; addr
!= end
; addr
+= PAGE_SIZE
) {
4759 pte_t ptent
= *(pte
++);
4760 union mc_target target
;
4763 struct page_cgroup
*pc
;
4769 type
= is_target_pte_for_mc(vma
, addr
, ptent
, &target
);
4771 case MC_TARGET_PAGE
:
4773 if (isolate_lru_page(page
))
4775 pc
= lookup_page_cgroup(page
);
4776 if (!mem_cgroup_move_account(pc
,
4777 mc
.from
, mc
.to
, false)) {
4779 /* we uncharge from mc.from later. */
4782 putback_lru_page(page
);
4783 put
: /* is_target_pte_for_mc() gets the page */
4786 case MC_TARGET_SWAP
:
4788 if (!mem_cgroup_move_swap_account(ent
,
4789 mc
.from
, mc
.to
, false)) {
4791 /* we fixup refcnts and charges later. */
4799 pte_unmap_unlock(pte
- 1, ptl
);
4804 * We have consumed all precharges we got in can_attach().
4805 * We try charge one by one, but don't do any additional
4806 * charges to mc.to if we have failed in charge once in attach()
4809 ret
= mem_cgroup_do_precharge(1);
4817 static void mem_cgroup_move_charge(struct mm_struct
*mm
)
4819 struct vm_area_struct
*vma
;
4821 lru_add_drain_all();
4822 down_read(&mm
->mmap_sem
);
4823 for (vma
= mm
->mmap
; vma
; vma
= vma
->vm_next
) {
4825 struct mm_walk mem_cgroup_move_charge_walk
= {
4826 .pmd_entry
= mem_cgroup_move_charge_pte_range
,
4830 if (is_vm_hugetlb_page(vma
))
4832 ret
= walk_page_range(vma
->vm_start
, vma
->vm_end
,
4833 &mem_cgroup_move_charge_walk
);
4836 * means we have consumed all precharges and failed in
4837 * doing additional charge. Just abandon here.
4841 up_read(&mm
->mmap_sem
);
4844 static void mem_cgroup_move_task(struct cgroup_subsys
*ss
,
4845 struct cgroup
*cont
,
4846 struct cgroup
*old_cont
,
4847 struct task_struct
*p
,
4850 struct mm_struct
*mm
;
4853 /* no need to move charge */
4856 mm
= get_task_mm(p
);
4858 mem_cgroup_move_charge(mm
);
4861 mem_cgroup_clear_mc();
4863 #else /* !CONFIG_MMU */
4864 static int mem_cgroup_can_attach(struct cgroup_subsys
*ss
,
4865 struct cgroup
*cgroup
,
4866 struct task_struct
*p
,
4871 static void mem_cgroup_cancel_attach(struct cgroup_subsys
*ss
,
4872 struct cgroup
*cgroup
,
4873 struct task_struct
*p
,
4877 static void mem_cgroup_move_task(struct cgroup_subsys
*ss
,
4878 struct cgroup
*cont
,
4879 struct cgroup
*old_cont
,
4880 struct task_struct
*p
,
4886 struct cgroup_subsys mem_cgroup_subsys
= {
4888 .subsys_id
= mem_cgroup_subsys_id
,
4889 .create
= mem_cgroup_create
,
4890 .pre_destroy
= mem_cgroup_pre_destroy
,
4891 .destroy
= mem_cgroup_destroy
,
4892 .populate
= mem_cgroup_populate
,
4893 .can_attach
= mem_cgroup_can_attach
,
4894 .cancel_attach
= mem_cgroup_cancel_attach
,
4895 .attach
= mem_cgroup_move_task
,
4900 #ifdef CONFIG_CGROUP_MEM_RES_CTLR_SWAP
4902 static int __init
disable_swap_account(char *s
)
4904 really_do_swap_account
= 0;
4907 __setup("noswapaccount", disable_swap_account
);