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 * Kernel Memory Controller
14 * Copyright (C) 2012 Parallels Inc. and Google Inc.
15 * Authors: Glauber Costa and Suleiman Souhlal
18 * Charge lifetime sanitation
19 * Lockless page tracking & accounting
20 * Unified hierarchy configuration model
21 * Copyright (C) 2015 Red Hat, Inc., Johannes Weiner
23 * This program is free software; you can redistribute it and/or modify
24 * it under the terms of the GNU General Public License as published by
25 * the Free Software Foundation; either version 2 of the License, or
26 * (at your option) any later version.
28 * This program is distributed in the hope that it will be useful,
29 * but WITHOUT ANY WARRANTY; without even the implied warranty of
30 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
31 * GNU General Public License for more details.
34 #include <linux/page_counter.h>
35 #include <linux/memcontrol.h>
36 #include <linux/cgroup.h>
38 #include <linux/hugetlb.h>
39 #include <linux/pagemap.h>
40 #include <linux/smp.h>
41 #include <linux/page-flags.h>
42 #include <linux/backing-dev.h>
43 #include <linux/bit_spinlock.h>
44 #include <linux/rcupdate.h>
45 #include <linux/limits.h>
46 #include <linux/export.h>
47 #include <linux/mutex.h>
48 #include <linux/rbtree.h>
49 #include <linux/slab.h>
50 #include <linux/swap.h>
51 #include <linux/swapops.h>
52 #include <linux/spinlock.h>
53 #include <linux/eventfd.h>
54 #include <linux/poll.h>
55 #include <linux/sort.h>
57 #include <linux/seq_file.h>
58 #include <linux/vmpressure.h>
59 #include <linux/mm_inline.h>
60 #include <linux/swap_cgroup.h>
61 #include <linux/cpu.h>
62 #include <linux/oom.h>
63 #include <linux/lockdep.h>
64 #include <linux/file.h>
65 #include <linux/tracehook.h>
69 #include <net/tcp_memcontrol.h>
72 #include <asm/uaccess.h>
74 #include <trace/events/vmscan.h>
76 struct cgroup_subsys memory_cgrp_subsys __read_mostly
;
77 EXPORT_SYMBOL(memory_cgrp_subsys
);
79 #define MEM_CGROUP_RECLAIM_RETRIES 5
80 static struct mem_cgroup
*root_mem_cgroup __read_mostly
;
81 struct cgroup_subsys_state
*mem_cgroup_root_css __read_mostly
;
83 /* Whether the swap controller is active */
84 #ifdef CONFIG_MEMCG_SWAP
85 int do_swap_account __read_mostly
;
87 #define do_swap_account 0
90 static const char * const mem_cgroup_stat_names
[] = {
100 static const char * const mem_cgroup_events_names
[] = {
107 static const char * const mem_cgroup_lru_names
[] = {
115 #define THRESHOLDS_EVENTS_TARGET 128
116 #define SOFTLIMIT_EVENTS_TARGET 1024
117 #define NUMAINFO_EVENTS_TARGET 1024
120 * Cgroups above their limits are maintained in a RB-Tree, independent of
121 * their hierarchy representation
124 struct mem_cgroup_tree_per_zone
{
125 struct rb_root rb_root
;
129 struct mem_cgroup_tree_per_node
{
130 struct mem_cgroup_tree_per_zone rb_tree_per_zone
[MAX_NR_ZONES
];
133 struct mem_cgroup_tree
{
134 struct mem_cgroup_tree_per_node
*rb_tree_per_node
[MAX_NUMNODES
];
137 static struct mem_cgroup_tree soft_limit_tree __read_mostly
;
140 struct mem_cgroup_eventfd_list
{
141 struct list_head list
;
142 struct eventfd_ctx
*eventfd
;
146 * cgroup_event represents events which userspace want to receive.
148 struct mem_cgroup_event
{
150 * memcg which the event belongs to.
152 struct mem_cgroup
*memcg
;
154 * eventfd to signal userspace about the event.
156 struct eventfd_ctx
*eventfd
;
158 * Each of these stored in a list by the cgroup.
160 struct list_head list
;
162 * register_event() callback will be used to add new userspace
163 * waiter for changes related to this event. Use eventfd_signal()
164 * on eventfd to send notification to userspace.
166 int (*register_event
)(struct mem_cgroup
*memcg
,
167 struct eventfd_ctx
*eventfd
, const char *args
);
169 * unregister_event() callback will be called when userspace closes
170 * the eventfd or on cgroup removing. This callback must be set,
171 * if you want provide notification functionality.
173 void (*unregister_event
)(struct mem_cgroup
*memcg
,
174 struct eventfd_ctx
*eventfd
);
176 * All fields below needed to unregister event when
177 * userspace closes eventfd.
180 wait_queue_head_t
*wqh
;
182 struct work_struct remove
;
185 static void mem_cgroup_threshold(struct mem_cgroup
*memcg
);
186 static void mem_cgroup_oom_notify(struct mem_cgroup
*memcg
);
188 /* Stuffs for move charges at task migration. */
190 * Types of charges to be moved.
192 #define MOVE_ANON 0x1U
193 #define MOVE_FILE 0x2U
194 #define MOVE_MASK (MOVE_ANON | MOVE_FILE)
196 /* "mc" and its members are protected by cgroup_mutex */
197 static struct move_charge_struct
{
198 spinlock_t lock
; /* for from, to */
199 struct mem_cgroup
*from
;
200 struct mem_cgroup
*to
;
202 unsigned long precharge
;
203 unsigned long moved_charge
;
204 unsigned long moved_swap
;
205 struct task_struct
*moving_task
; /* a task moving charges */
206 wait_queue_head_t waitq
; /* a waitq for other context */
208 .lock
= __SPIN_LOCK_UNLOCKED(mc
.lock
),
209 .waitq
= __WAIT_QUEUE_HEAD_INITIALIZER(mc
.waitq
),
213 * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft
214 * limit reclaim to prevent infinite loops, if they ever occur.
216 #define MEM_CGROUP_MAX_RECLAIM_LOOPS 100
217 #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2
220 MEM_CGROUP_CHARGE_TYPE_CACHE
= 0,
221 MEM_CGROUP_CHARGE_TYPE_ANON
,
222 MEM_CGROUP_CHARGE_TYPE_SWAPOUT
, /* for accounting swapcache */
223 MEM_CGROUP_CHARGE_TYPE_DROP
, /* a page was unused swap cache */
227 /* for encoding cft->private value on file */
235 #define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val))
236 #define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff)
237 #define MEMFILE_ATTR(val) ((val) & 0xffff)
238 /* Used for OOM nofiier */
239 #define OOM_CONTROL (0)
242 * The memcg_create_mutex will be held whenever a new cgroup is created.
243 * As a consequence, any change that needs to protect against new child cgroups
244 * appearing has to hold it as well.
246 static DEFINE_MUTEX(memcg_create_mutex
);
248 /* Some nice accessors for the vmpressure. */
249 struct vmpressure
*memcg_to_vmpressure(struct mem_cgroup
*memcg
)
252 memcg
= root_mem_cgroup
;
253 return &memcg
->vmpressure
;
256 struct cgroup_subsys_state
*vmpressure_to_css(struct vmpressure
*vmpr
)
258 return &container_of(vmpr
, struct mem_cgroup
, vmpressure
)->css
;
261 static inline bool mem_cgroup_is_root(struct mem_cgroup
*memcg
)
263 return (memcg
== root_mem_cgroup
);
267 * We restrict the id in the range of [1, 65535], so it can fit into
270 #define MEM_CGROUP_ID_MAX USHRT_MAX
272 static inline unsigned short mem_cgroup_id(struct mem_cgroup
*memcg
)
274 return memcg
->css
.id
;
278 * A helper function to get mem_cgroup from ID. must be called under
279 * rcu_read_lock(). The caller is responsible for calling
280 * css_tryget_online() if the mem_cgroup is used for charging. (dropping
281 * refcnt from swap can be called against removed memcg.)
283 static inline struct mem_cgroup
*mem_cgroup_from_id(unsigned short id
)
285 struct cgroup_subsys_state
*css
;
287 css
= css_from_id(id
, &memory_cgrp_subsys
);
288 return mem_cgroup_from_css(css
);
291 /* Writing them here to avoid exposing memcg's inner layout */
292 #if defined(CONFIG_INET) && defined(CONFIG_MEMCG_KMEM)
294 void sock_update_memcg(struct sock
*sk
)
296 if (mem_cgroup_sockets_enabled
) {
297 struct mem_cgroup
*memcg
;
298 struct cg_proto
*cg_proto
;
300 BUG_ON(!sk
->sk_prot
->proto_cgroup
);
302 /* Socket cloning can throw us here with sk_cgrp already
303 * filled. It won't however, necessarily happen from
304 * process context. So the test for root memcg given
305 * the current task's memcg won't help us in this case.
307 * Respecting the original socket's memcg is a better
308 * decision in this case.
311 BUG_ON(mem_cgroup_is_root(sk
->sk_cgrp
->memcg
));
312 css_get(&sk
->sk_cgrp
->memcg
->css
);
317 memcg
= mem_cgroup_from_task(current
);
318 cg_proto
= sk
->sk_prot
->proto_cgroup(memcg
);
319 if (cg_proto
&& test_bit(MEMCG_SOCK_ACTIVE
, &cg_proto
->flags
) &&
320 css_tryget_online(&memcg
->css
)) {
321 sk
->sk_cgrp
= cg_proto
;
326 EXPORT_SYMBOL(sock_update_memcg
);
328 void sock_release_memcg(struct sock
*sk
)
330 if (mem_cgroup_sockets_enabled
&& sk
->sk_cgrp
) {
331 struct mem_cgroup
*memcg
;
332 WARN_ON(!sk
->sk_cgrp
->memcg
);
333 memcg
= sk
->sk_cgrp
->memcg
;
334 css_put(&sk
->sk_cgrp
->memcg
->css
);
338 struct cg_proto
*tcp_proto_cgroup(struct mem_cgroup
*memcg
)
340 if (!memcg
|| mem_cgroup_is_root(memcg
))
343 return &memcg
->tcp_mem
;
345 EXPORT_SYMBOL(tcp_proto_cgroup
);
349 #ifdef CONFIG_MEMCG_KMEM
351 * This will be the memcg's index in each cache's ->memcg_params.memcg_caches.
352 * The main reason for not using cgroup id for this:
353 * this works better in sparse environments, where we have a lot of memcgs,
354 * but only a few kmem-limited. Or also, if we have, for instance, 200
355 * memcgs, and none but the 200th is kmem-limited, we'd have to have a
356 * 200 entry array for that.
358 * The current size of the caches array is stored in memcg_nr_cache_ids. It
359 * will double each time we have to increase it.
361 static DEFINE_IDA(memcg_cache_ida
);
362 int memcg_nr_cache_ids
;
364 /* Protects memcg_nr_cache_ids */
365 static DECLARE_RWSEM(memcg_cache_ids_sem
);
367 void memcg_get_cache_ids(void)
369 down_read(&memcg_cache_ids_sem
);
372 void memcg_put_cache_ids(void)
374 up_read(&memcg_cache_ids_sem
);
378 * MIN_SIZE is different than 1, because we would like to avoid going through
379 * the alloc/free process all the time. In a small machine, 4 kmem-limited
380 * cgroups is a reasonable guess. In the future, it could be a parameter or
381 * tunable, but that is strictly not necessary.
383 * MAX_SIZE should be as large as the number of cgrp_ids. Ideally, we could get
384 * this constant directly from cgroup, but it is understandable that this is
385 * better kept as an internal representation in cgroup.c. In any case, the
386 * cgrp_id space is not getting any smaller, and we don't have to necessarily
387 * increase ours as well if it increases.
389 #define MEMCG_CACHES_MIN_SIZE 4
390 #define MEMCG_CACHES_MAX_SIZE MEM_CGROUP_ID_MAX
393 * A lot of the calls to the cache allocation functions are expected to be
394 * inlined by the compiler. Since the calls to memcg_kmem_get_cache are
395 * conditional to this static branch, we'll have to allow modules that does
396 * kmem_cache_alloc and the such to see this symbol as well
398 struct static_key memcg_kmem_enabled_key
;
399 EXPORT_SYMBOL(memcg_kmem_enabled_key
);
401 #endif /* CONFIG_MEMCG_KMEM */
403 static struct mem_cgroup_per_zone
*
404 mem_cgroup_zone_zoneinfo(struct mem_cgroup
*memcg
, struct zone
*zone
)
406 int nid
= zone_to_nid(zone
);
407 int zid
= zone_idx(zone
);
409 return &memcg
->nodeinfo
[nid
]->zoneinfo
[zid
];
413 * mem_cgroup_css_from_page - css of the memcg associated with a page
414 * @page: page of interest
416 * If memcg is bound to the default hierarchy, css of the memcg associated
417 * with @page is returned. The returned css remains associated with @page
418 * until it is released.
420 * If memcg is bound to a traditional hierarchy, the css of root_mem_cgroup
423 * XXX: The above description of behavior on the default hierarchy isn't
424 * strictly true yet as replace_page_cache_page() can modify the
425 * association before @page is released even on the default hierarchy;
426 * however, the current and planned usages don't mix the the two functions
427 * and replace_page_cache_page() will soon be updated to make the invariant
430 struct cgroup_subsys_state
*mem_cgroup_css_from_page(struct page
*page
)
432 struct mem_cgroup
*memcg
;
436 memcg
= page
->mem_cgroup
;
438 if (!memcg
|| !cgroup_on_dfl(memcg
->css
.cgroup
))
439 memcg
= root_mem_cgroup
;
446 * page_cgroup_ino - return inode number of the memcg a page is charged to
449 * Look up the closest online ancestor of the memory cgroup @page is charged to
450 * and return its inode number or 0 if @page is not charged to any cgroup. It
451 * is safe to call this function without holding a reference to @page.
453 * Note, this function is inherently racy, because there is nothing to prevent
454 * the cgroup inode from getting torn down and potentially reallocated a moment
455 * after page_cgroup_ino() returns, so it only should be used by callers that
456 * do not care (such as procfs interfaces).
458 ino_t
page_cgroup_ino(struct page
*page
)
460 struct mem_cgroup
*memcg
;
461 unsigned long ino
= 0;
464 memcg
= READ_ONCE(page
->mem_cgroup
);
465 while (memcg
&& !(memcg
->css
.flags
& CSS_ONLINE
))
466 memcg
= parent_mem_cgroup(memcg
);
468 ino
= cgroup_ino(memcg
->css
.cgroup
);
473 static struct mem_cgroup_per_zone
*
474 mem_cgroup_page_zoneinfo(struct mem_cgroup
*memcg
, struct page
*page
)
476 int nid
= page_to_nid(page
);
477 int zid
= page_zonenum(page
);
479 return &memcg
->nodeinfo
[nid
]->zoneinfo
[zid
];
482 static struct mem_cgroup_tree_per_zone
*
483 soft_limit_tree_node_zone(int nid
, int zid
)
485 return &soft_limit_tree
.rb_tree_per_node
[nid
]->rb_tree_per_zone
[zid
];
488 static struct mem_cgroup_tree_per_zone
*
489 soft_limit_tree_from_page(struct page
*page
)
491 int nid
= page_to_nid(page
);
492 int zid
= page_zonenum(page
);
494 return &soft_limit_tree
.rb_tree_per_node
[nid
]->rb_tree_per_zone
[zid
];
497 static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_zone
*mz
,
498 struct mem_cgroup_tree_per_zone
*mctz
,
499 unsigned long new_usage_in_excess
)
501 struct rb_node
**p
= &mctz
->rb_root
.rb_node
;
502 struct rb_node
*parent
= NULL
;
503 struct mem_cgroup_per_zone
*mz_node
;
508 mz
->usage_in_excess
= new_usage_in_excess
;
509 if (!mz
->usage_in_excess
)
513 mz_node
= rb_entry(parent
, struct mem_cgroup_per_zone
,
515 if (mz
->usage_in_excess
< mz_node
->usage_in_excess
)
518 * We can't avoid mem cgroups that are over their soft
519 * limit by the same amount
521 else if (mz
->usage_in_excess
>= mz_node
->usage_in_excess
)
524 rb_link_node(&mz
->tree_node
, parent
, p
);
525 rb_insert_color(&mz
->tree_node
, &mctz
->rb_root
);
529 static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_zone
*mz
,
530 struct mem_cgroup_tree_per_zone
*mctz
)
534 rb_erase(&mz
->tree_node
, &mctz
->rb_root
);
538 static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_zone
*mz
,
539 struct mem_cgroup_tree_per_zone
*mctz
)
543 spin_lock_irqsave(&mctz
->lock
, flags
);
544 __mem_cgroup_remove_exceeded(mz
, mctz
);
545 spin_unlock_irqrestore(&mctz
->lock
, flags
);
548 static unsigned long soft_limit_excess(struct mem_cgroup
*memcg
)
550 unsigned long nr_pages
= page_counter_read(&memcg
->memory
);
551 unsigned long soft_limit
= READ_ONCE(memcg
->soft_limit
);
552 unsigned long excess
= 0;
554 if (nr_pages
> soft_limit
)
555 excess
= nr_pages
- soft_limit
;
560 static void mem_cgroup_update_tree(struct mem_cgroup
*memcg
, struct page
*page
)
562 unsigned long excess
;
563 struct mem_cgroup_per_zone
*mz
;
564 struct mem_cgroup_tree_per_zone
*mctz
;
566 mctz
= soft_limit_tree_from_page(page
);
568 * Necessary to update all ancestors when hierarchy is used.
569 * because their event counter is not touched.
571 for (; memcg
; memcg
= parent_mem_cgroup(memcg
)) {
572 mz
= mem_cgroup_page_zoneinfo(memcg
, page
);
573 excess
= soft_limit_excess(memcg
);
575 * We have to update the tree if mz is on RB-tree or
576 * mem is over its softlimit.
578 if (excess
|| mz
->on_tree
) {
581 spin_lock_irqsave(&mctz
->lock
, flags
);
582 /* if on-tree, remove it */
584 __mem_cgroup_remove_exceeded(mz
, mctz
);
586 * Insert again. mz->usage_in_excess will be updated.
587 * If excess is 0, no tree ops.
589 __mem_cgroup_insert_exceeded(mz
, mctz
, excess
);
590 spin_unlock_irqrestore(&mctz
->lock
, flags
);
595 static void mem_cgroup_remove_from_trees(struct mem_cgroup
*memcg
)
597 struct mem_cgroup_tree_per_zone
*mctz
;
598 struct mem_cgroup_per_zone
*mz
;
602 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++) {
603 mz
= &memcg
->nodeinfo
[nid
]->zoneinfo
[zid
];
604 mctz
= soft_limit_tree_node_zone(nid
, zid
);
605 mem_cgroup_remove_exceeded(mz
, mctz
);
610 static struct mem_cgroup_per_zone
*
611 __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone
*mctz
)
613 struct rb_node
*rightmost
= NULL
;
614 struct mem_cgroup_per_zone
*mz
;
618 rightmost
= rb_last(&mctz
->rb_root
);
620 goto done
; /* Nothing to reclaim from */
622 mz
= rb_entry(rightmost
, struct mem_cgroup_per_zone
, tree_node
);
624 * Remove the node now but someone else can add it back,
625 * we will to add it back at the end of reclaim to its correct
626 * position in the tree.
628 __mem_cgroup_remove_exceeded(mz
, mctz
);
629 if (!soft_limit_excess(mz
->memcg
) ||
630 !css_tryget_online(&mz
->memcg
->css
))
636 static struct mem_cgroup_per_zone
*
637 mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_zone
*mctz
)
639 struct mem_cgroup_per_zone
*mz
;
641 spin_lock_irq(&mctz
->lock
);
642 mz
= __mem_cgroup_largest_soft_limit_node(mctz
);
643 spin_unlock_irq(&mctz
->lock
);
648 * Return page count for single (non recursive) @memcg.
650 * Implementation Note: reading percpu statistics for memcg.
652 * Both of vmstat[] and percpu_counter has threshold and do periodic
653 * synchronization to implement "quick" read. There are trade-off between
654 * reading cost and precision of value. Then, we may have a chance to implement
655 * a periodic synchronization of counter in memcg's counter.
657 * But this _read() function is used for user interface now. The user accounts
658 * memory usage by memory cgroup and he _always_ requires exact value because
659 * he accounts memory. Even if we provide quick-and-fuzzy read, we always
660 * have to visit all online cpus and make sum. So, for now, unnecessary
661 * synchronization is not implemented. (just implemented for cpu hotplug)
663 * If there are kernel internal actions which can make use of some not-exact
664 * value, and reading all cpu value can be performance bottleneck in some
665 * common workload, threshold and synchronization as vmstat[] should be
669 mem_cgroup_read_stat(struct mem_cgroup
*memcg
, enum mem_cgroup_stat_index idx
)
674 /* Per-cpu values can be negative, use a signed accumulator */
675 for_each_possible_cpu(cpu
)
676 val
+= per_cpu(memcg
->stat
->count
[idx
], cpu
);
678 * Summing races with updates, so val may be negative. Avoid exposing
679 * transient negative values.
686 static unsigned long mem_cgroup_read_events(struct mem_cgroup
*memcg
,
687 enum mem_cgroup_events_index idx
)
689 unsigned long val
= 0;
692 for_each_possible_cpu(cpu
)
693 val
+= per_cpu(memcg
->stat
->events
[idx
], cpu
);
697 static void mem_cgroup_charge_statistics(struct mem_cgroup
*memcg
,
702 * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is
703 * counted as CACHE even if it's on ANON LRU.
706 __this_cpu_add(memcg
->stat
->count
[MEM_CGROUP_STAT_RSS
],
709 __this_cpu_add(memcg
->stat
->count
[MEM_CGROUP_STAT_CACHE
],
712 if (PageTransHuge(page
))
713 __this_cpu_add(memcg
->stat
->count
[MEM_CGROUP_STAT_RSS_HUGE
],
716 /* pagein of a big page is an event. So, ignore page size */
718 __this_cpu_inc(memcg
->stat
->events
[MEM_CGROUP_EVENTS_PGPGIN
]);
720 __this_cpu_inc(memcg
->stat
->events
[MEM_CGROUP_EVENTS_PGPGOUT
]);
721 nr_pages
= -nr_pages
; /* for event */
724 __this_cpu_add(memcg
->stat
->nr_page_events
, nr_pages
);
727 static unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup
*memcg
,
729 unsigned int lru_mask
)
731 unsigned long nr
= 0;
734 VM_BUG_ON((unsigned)nid
>= nr_node_ids
);
736 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++) {
737 struct mem_cgroup_per_zone
*mz
;
741 if (!(BIT(lru
) & lru_mask
))
743 mz
= &memcg
->nodeinfo
[nid
]->zoneinfo
[zid
];
744 nr
+= mz
->lru_size
[lru
];
750 static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup
*memcg
,
751 unsigned int lru_mask
)
753 unsigned long nr
= 0;
756 for_each_node_state(nid
, N_MEMORY
)
757 nr
+= mem_cgroup_node_nr_lru_pages(memcg
, nid
, lru_mask
);
761 static bool mem_cgroup_event_ratelimit(struct mem_cgroup
*memcg
,
762 enum mem_cgroup_events_target target
)
764 unsigned long val
, next
;
766 val
= __this_cpu_read(memcg
->stat
->nr_page_events
);
767 next
= __this_cpu_read(memcg
->stat
->targets
[target
]);
768 /* from time_after() in jiffies.h */
769 if ((long)next
- (long)val
< 0) {
771 case MEM_CGROUP_TARGET_THRESH
:
772 next
= val
+ THRESHOLDS_EVENTS_TARGET
;
774 case MEM_CGROUP_TARGET_SOFTLIMIT
:
775 next
= val
+ SOFTLIMIT_EVENTS_TARGET
;
777 case MEM_CGROUP_TARGET_NUMAINFO
:
778 next
= val
+ NUMAINFO_EVENTS_TARGET
;
783 __this_cpu_write(memcg
->stat
->targets
[target
], next
);
790 * Check events in order.
793 static void memcg_check_events(struct mem_cgroup
*memcg
, struct page
*page
)
795 /* threshold event is triggered in finer grain than soft limit */
796 if (unlikely(mem_cgroup_event_ratelimit(memcg
,
797 MEM_CGROUP_TARGET_THRESH
))) {
799 bool do_numainfo __maybe_unused
;
801 do_softlimit
= mem_cgroup_event_ratelimit(memcg
,
802 MEM_CGROUP_TARGET_SOFTLIMIT
);
804 do_numainfo
= mem_cgroup_event_ratelimit(memcg
,
805 MEM_CGROUP_TARGET_NUMAINFO
);
807 mem_cgroup_threshold(memcg
);
808 if (unlikely(do_softlimit
))
809 mem_cgroup_update_tree(memcg
, page
);
811 if (unlikely(do_numainfo
))
812 atomic_inc(&memcg
->numainfo_events
);
817 struct mem_cgroup
*mem_cgroup_from_task(struct task_struct
*p
)
820 * mm_update_next_owner() may clear mm->owner to NULL
821 * if it races with swapoff, page migration, etc.
822 * So this can be called with p == NULL.
827 return mem_cgroup_from_css(task_css(p
, memory_cgrp_id
));
829 EXPORT_SYMBOL(mem_cgroup_from_task
);
831 static struct mem_cgroup
*get_mem_cgroup_from_mm(struct mm_struct
*mm
)
833 struct mem_cgroup
*memcg
= NULL
;
838 * Page cache insertions can happen withou an
839 * actual mm context, e.g. during disk probing
840 * on boot, loopback IO, acct() writes etc.
843 memcg
= root_mem_cgroup
;
845 memcg
= mem_cgroup_from_task(rcu_dereference(mm
->owner
));
846 if (unlikely(!memcg
))
847 memcg
= root_mem_cgroup
;
849 } while (!css_tryget_online(&memcg
->css
));
855 * mem_cgroup_iter - iterate over memory cgroup hierarchy
856 * @root: hierarchy root
857 * @prev: previously returned memcg, NULL on first invocation
858 * @reclaim: cookie for shared reclaim walks, NULL for full walks
860 * Returns references to children of the hierarchy below @root, or
861 * @root itself, or %NULL after a full round-trip.
863 * Caller must pass the return value in @prev on subsequent
864 * invocations for reference counting, or use mem_cgroup_iter_break()
865 * to cancel a hierarchy walk before the round-trip is complete.
867 * Reclaimers can specify a zone and a priority level in @reclaim to
868 * divide up the memcgs in the hierarchy among all concurrent
869 * reclaimers operating on the same zone and priority.
871 struct mem_cgroup
*mem_cgroup_iter(struct mem_cgroup
*root
,
872 struct mem_cgroup
*prev
,
873 struct mem_cgroup_reclaim_cookie
*reclaim
)
875 struct mem_cgroup_reclaim_iter
*uninitialized_var(iter
);
876 struct cgroup_subsys_state
*css
= NULL
;
877 struct mem_cgroup
*memcg
= NULL
;
878 struct mem_cgroup
*pos
= NULL
;
880 if (mem_cgroup_disabled())
884 root
= root_mem_cgroup
;
886 if (prev
&& !reclaim
)
889 if (!root
->use_hierarchy
&& root
!= root_mem_cgroup
) {
898 struct mem_cgroup_per_zone
*mz
;
900 mz
= mem_cgroup_zone_zoneinfo(root
, reclaim
->zone
);
901 iter
= &mz
->iter
[reclaim
->priority
];
903 if (prev
&& reclaim
->generation
!= iter
->generation
)
907 pos
= READ_ONCE(iter
->position
);
909 * A racing update may change the position and
910 * put the last reference, hence css_tryget(),
911 * or retry to see the updated position.
913 } while (pos
&& !css_tryget(&pos
->css
));
920 css
= css_next_descendant_pre(css
, &root
->css
);
923 * Reclaimers share the hierarchy walk, and a
924 * new one might jump in right at the end of
925 * the hierarchy - make sure they see at least
926 * one group and restart from the beginning.
934 * Verify the css and acquire a reference. The root
935 * is provided by the caller, so we know it's alive
936 * and kicking, and don't take an extra reference.
938 memcg
= mem_cgroup_from_css(css
);
940 if (css
== &root
->css
)
943 if (css_tryget(css
)) {
945 * Make sure the memcg is initialized:
946 * mem_cgroup_css_online() orders the the
947 * initialization against setting the flag.
949 if (smp_load_acquire(&memcg
->initialized
))
959 if (cmpxchg(&iter
->position
, pos
, memcg
) == pos
) {
961 css_get(&memcg
->css
);
967 * pairs with css_tryget when dereferencing iter->position
976 reclaim
->generation
= iter
->generation
;
982 if (prev
&& prev
!= root
)
989 * mem_cgroup_iter_break - abort a hierarchy walk prematurely
990 * @root: hierarchy root
991 * @prev: last visited hierarchy member as returned by mem_cgroup_iter()
993 void mem_cgroup_iter_break(struct mem_cgroup
*root
,
994 struct mem_cgroup
*prev
)
997 root
= root_mem_cgroup
;
998 if (prev
&& prev
!= root
)
1003 * Iteration constructs for visiting all cgroups (under a tree). If
1004 * loops are exited prematurely (break), mem_cgroup_iter_break() must
1005 * be used for reference counting.
1007 #define for_each_mem_cgroup_tree(iter, root) \
1008 for (iter = mem_cgroup_iter(root, NULL, NULL); \
1010 iter = mem_cgroup_iter(root, iter, NULL))
1012 #define for_each_mem_cgroup(iter) \
1013 for (iter = mem_cgroup_iter(NULL, NULL, NULL); \
1015 iter = mem_cgroup_iter(NULL, iter, NULL))
1018 * mem_cgroup_zone_lruvec - get the lru list vector for a zone and memcg
1019 * @zone: zone of the wanted lruvec
1020 * @memcg: memcg of the wanted lruvec
1022 * Returns the lru list vector holding pages for the given @zone and
1023 * @mem. This can be the global zone lruvec, if the memory controller
1026 struct lruvec
*mem_cgroup_zone_lruvec(struct zone
*zone
,
1027 struct mem_cgroup
*memcg
)
1029 struct mem_cgroup_per_zone
*mz
;
1030 struct lruvec
*lruvec
;
1032 if (mem_cgroup_disabled()) {
1033 lruvec
= &zone
->lruvec
;
1037 mz
= mem_cgroup_zone_zoneinfo(memcg
, zone
);
1038 lruvec
= &mz
->lruvec
;
1041 * Since a node can be onlined after the mem_cgroup was created,
1042 * we have to be prepared to initialize lruvec->zone here;
1043 * and if offlined then reonlined, we need to reinitialize it.
1045 if (unlikely(lruvec
->zone
!= zone
))
1046 lruvec
->zone
= zone
;
1051 * mem_cgroup_page_lruvec - return lruvec for isolating/putting an LRU page
1053 * @zone: zone of the page
1055 * This function is only safe when following the LRU page isolation
1056 * and putback protocol: the LRU lock must be held, and the page must
1057 * either be PageLRU() or the caller must have isolated/allocated it.
1059 struct lruvec
*mem_cgroup_page_lruvec(struct page
*page
, struct zone
*zone
)
1061 struct mem_cgroup_per_zone
*mz
;
1062 struct mem_cgroup
*memcg
;
1063 struct lruvec
*lruvec
;
1065 if (mem_cgroup_disabled()) {
1066 lruvec
= &zone
->lruvec
;
1070 memcg
= page
->mem_cgroup
;
1072 * Swapcache readahead pages are added to the LRU - and
1073 * possibly migrated - before they are charged.
1076 memcg
= root_mem_cgroup
;
1078 mz
= mem_cgroup_page_zoneinfo(memcg
, page
);
1079 lruvec
= &mz
->lruvec
;
1082 * Since a node can be onlined after the mem_cgroup was created,
1083 * we have to be prepared to initialize lruvec->zone here;
1084 * and if offlined then reonlined, we need to reinitialize it.
1086 if (unlikely(lruvec
->zone
!= zone
))
1087 lruvec
->zone
= zone
;
1092 * mem_cgroup_update_lru_size - account for adding or removing an lru page
1093 * @lruvec: mem_cgroup per zone lru vector
1094 * @lru: index of lru list the page is sitting on
1095 * @nr_pages: positive when adding or negative when removing
1097 * This function must be called when a page is added to or removed from an
1100 void mem_cgroup_update_lru_size(struct lruvec
*lruvec
, enum lru_list lru
,
1103 struct mem_cgroup_per_zone
*mz
;
1104 unsigned long *lru_size
;
1106 if (mem_cgroup_disabled())
1109 mz
= container_of(lruvec
, struct mem_cgroup_per_zone
, lruvec
);
1110 lru_size
= mz
->lru_size
+ lru
;
1111 *lru_size
+= nr_pages
;
1112 VM_BUG_ON((long)(*lru_size
) < 0);
1115 bool task_in_mem_cgroup(struct task_struct
*task
, struct mem_cgroup
*memcg
)
1117 struct mem_cgroup
*task_memcg
;
1118 struct task_struct
*p
;
1121 p
= find_lock_task_mm(task
);
1123 task_memcg
= get_mem_cgroup_from_mm(p
->mm
);
1127 * All threads may have already detached their mm's, but the oom
1128 * killer still needs to detect if they have already been oom
1129 * killed to prevent needlessly killing additional tasks.
1132 task_memcg
= mem_cgroup_from_task(task
);
1133 css_get(&task_memcg
->css
);
1136 ret
= mem_cgroup_is_descendant(task_memcg
, memcg
);
1137 css_put(&task_memcg
->css
);
1141 #define mem_cgroup_from_counter(counter, member) \
1142 container_of(counter, struct mem_cgroup, member)
1145 * mem_cgroup_margin - calculate chargeable space of a memory cgroup
1146 * @memcg: the memory cgroup
1148 * Returns the maximum amount of memory @mem can be charged with, in
1151 static unsigned long mem_cgroup_margin(struct mem_cgroup
*memcg
)
1153 unsigned long margin
= 0;
1154 unsigned long count
;
1155 unsigned long limit
;
1157 count
= page_counter_read(&memcg
->memory
);
1158 limit
= READ_ONCE(memcg
->memory
.limit
);
1160 margin
= limit
- count
;
1162 if (do_swap_account
) {
1163 count
= page_counter_read(&memcg
->memsw
);
1164 limit
= READ_ONCE(memcg
->memsw
.limit
);
1166 margin
= min(margin
, limit
- count
);
1173 * A routine for checking "mem" is under move_account() or not.
1175 * Checking a cgroup is mc.from or mc.to or under hierarchy of
1176 * moving cgroups. This is for waiting at high-memory pressure
1179 static bool mem_cgroup_under_move(struct mem_cgroup
*memcg
)
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
);
1194 ret
= mem_cgroup_is_descendant(from
, memcg
) ||
1195 mem_cgroup_is_descendant(to
, memcg
);
1197 spin_unlock(&mc
.lock
);
1201 static bool mem_cgroup_wait_acct_move(struct mem_cgroup
*memcg
)
1203 if (mc
.moving_task
&& current
!= mc
.moving_task
) {
1204 if (mem_cgroup_under_move(memcg
)) {
1206 prepare_to_wait(&mc
.waitq
, &wait
, TASK_INTERRUPTIBLE
);
1207 /* moving charge context might have finished. */
1210 finish_wait(&mc
.waitq
, &wait
);
1217 #define K(x) ((x) << (PAGE_SHIFT-10))
1219 * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller.
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 /* oom_info_lock ensures that parallel ooms do not interleave */
1229 static DEFINE_MUTEX(oom_info_lock
);
1230 struct mem_cgroup
*iter
;
1233 mutex_lock(&oom_info_lock
);
1237 pr_info("Task in ");
1238 pr_cont_cgroup_path(task_cgroup(p
, memory_cgrp_id
));
1239 pr_cont(" killed as a result of limit of ");
1241 pr_info("Memory limit reached of cgroup ");
1244 pr_cont_cgroup_path(memcg
->css
.cgroup
);
1249 pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n",
1250 K((u64
)page_counter_read(&memcg
->memory
)),
1251 K((u64
)memcg
->memory
.limit
), memcg
->memory
.failcnt
);
1252 pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n",
1253 K((u64
)page_counter_read(&memcg
->memsw
)),
1254 K((u64
)memcg
->memsw
.limit
), memcg
->memsw
.failcnt
);
1255 pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n",
1256 K((u64
)page_counter_read(&memcg
->kmem
)),
1257 K((u64
)memcg
->kmem
.limit
), memcg
->kmem
.failcnt
);
1259 for_each_mem_cgroup_tree(iter
, memcg
) {
1260 pr_info("Memory cgroup stats for ");
1261 pr_cont_cgroup_path(iter
->css
.cgroup
);
1264 for (i
= 0; i
< MEM_CGROUP_STAT_NSTATS
; i
++) {
1265 if (i
== MEM_CGROUP_STAT_SWAP
&& !do_swap_account
)
1267 pr_cont(" %s:%luKB", mem_cgroup_stat_names
[i
],
1268 K(mem_cgroup_read_stat(iter
, i
)));
1271 for (i
= 0; i
< NR_LRU_LISTS
; i
++)
1272 pr_cont(" %s:%luKB", mem_cgroup_lru_names
[i
],
1273 K(mem_cgroup_nr_lru_pages(iter
, BIT(i
))));
1277 mutex_unlock(&oom_info_lock
);
1281 * This function returns the number of memcg under hierarchy tree. Returns
1282 * 1(self count) if no children.
1284 static int mem_cgroup_count_children(struct mem_cgroup
*memcg
)
1287 struct mem_cgroup
*iter
;
1289 for_each_mem_cgroup_tree(iter
, memcg
)
1295 * Return the memory (and swap, if configured) limit for a memcg.
1297 static unsigned long mem_cgroup_get_limit(struct mem_cgroup
*memcg
)
1299 unsigned long limit
;
1301 limit
= memcg
->memory
.limit
;
1302 if (mem_cgroup_swappiness(memcg
)) {
1303 unsigned long memsw_limit
;
1305 memsw_limit
= memcg
->memsw
.limit
;
1306 limit
= min(limit
+ total_swap_pages
, memsw_limit
);
1311 static void mem_cgroup_out_of_memory(struct mem_cgroup
*memcg
, gfp_t gfp_mask
,
1314 struct oom_control oc
= {
1317 .gfp_mask
= gfp_mask
,
1320 struct mem_cgroup
*iter
;
1321 unsigned long chosen_points
= 0;
1322 unsigned long totalpages
;
1323 unsigned int points
= 0;
1324 struct task_struct
*chosen
= NULL
;
1326 mutex_lock(&oom_lock
);
1329 * If current has a pending SIGKILL or is exiting, then automatically
1330 * select it. The goal is to allow it to allocate so that it may
1331 * quickly exit and free its memory.
1333 if (fatal_signal_pending(current
) || task_will_free_mem(current
)) {
1334 mark_oom_victim(current
);
1338 check_panic_on_oom(&oc
, CONSTRAINT_MEMCG
, memcg
);
1339 totalpages
= mem_cgroup_get_limit(memcg
) ? : 1;
1340 for_each_mem_cgroup_tree(iter
, memcg
) {
1341 struct css_task_iter it
;
1342 struct task_struct
*task
;
1344 css_task_iter_start(&iter
->css
, &it
);
1345 while ((task
= css_task_iter_next(&it
))) {
1346 switch (oom_scan_process_thread(&oc
, task
, totalpages
)) {
1347 case OOM_SCAN_SELECT
:
1349 put_task_struct(chosen
);
1351 chosen_points
= ULONG_MAX
;
1352 get_task_struct(chosen
);
1354 case OOM_SCAN_CONTINUE
:
1356 case OOM_SCAN_ABORT
:
1357 css_task_iter_end(&it
);
1358 mem_cgroup_iter_break(memcg
, iter
);
1360 put_task_struct(chosen
);
1365 points
= oom_badness(task
, memcg
, NULL
, totalpages
);
1366 if (!points
|| points
< chosen_points
)
1368 /* Prefer thread group leaders for display purposes */
1369 if (points
== chosen_points
&&
1370 thread_group_leader(chosen
))
1374 put_task_struct(chosen
);
1376 chosen_points
= points
;
1377 get_task_struct(chosen
);
1379 css_task_iter_end(&it
);
1383 points
= chosen_points
* 1000 / totalpages
;
1384 oom_kill_process(&oc
, chosen
, points
, totalpages
, memcg
,
1385 "Memory cgroup out of memory");
1388 mutex_unlock(&oom_lock
);
1391 #if MAX_NUMNODES > 1
1394 * test_mem_cgroup_node_reclaimable
1395 * @memcg: the target memcg
1396 * @nid: the node ID to be checked.
1397 * @noswap : specify true here if the user wants flle only information.
1399 * This function returns whether the specified memcg contains any
1400 * reclaimable pages on a node. Returns true if there are any reclaimable
1401 * pages in the node.
1403 static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup
*memcg
,
1404 int nid
, bool noswap
)
1406 if (mem_cgroup_node_nr_lru_pages(memcg
, nid
, LRU_ALL_FILE
))
1408 if (noswap
|| !total_swap_pages
)
1410 if (mem_cgroup_node_nr_lru_pages(memcg
, nid
, LRU_ALL_ANON
))
1417 * Always updating the nodemask is not very good - even if we have an empty
1418 * list or the wrong list here, we can start from some node and traverse all
1419 * nodes based on the zonelist. So update the list loosely once per 10 secs.
1422 static void mem_cgroup_may_update_nodemask(struct mem_cgroup
*memcg
)
1426 * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET
1427 * pagein/pageout changes since the last update.
1429 if (!atomic_read(&memcg
->numainfo_events
))
1431 if (atomic_inc_return(&memcg
->numainfo_updating
) > 1)
1434 /* make a nodemask where this memcg uses memory from */
1435 memcg
->scan_nodes
= node_states
[N_MEMORY
];
1437 for_each_node_mask(nid
, node_states
[N_MEMORY
]) {
1439 if (!test_mem_cgroup_node_reclaimable(memcg
, nid
, false))
1440 node_clear(nid
, memcg
->scan_nodes
);
1443 atomic_set(&memcg
->numainfo_events
, 0);
1444 atomic_set(&memcg
->numainfo_updating
, 0);
1448 * Selecting a node where we start reclaim from. Because what we need is just
1449 * reducing usage counter, start from anywhere is O,K. Considering
1450 * memory reclaim from current node, there are pros. and cons.
1452 * Freeing memory from current node means freeing memory from a node which
1453 * we'll use or we've used. So, it may make LRU bad. And if several threads
1454 * hit limits, it will see a contention on a node. But freeing from remote
1455 * node means more costs for memory reclaim because of memory latency.
1457 * Now, we use round-robin. Better algorithm is welcomed.
1459 int mem_cgroup_select_victim_node(struct mem_cgroup
*memcg
)
1463 mem_cgroup_may_update_nodemask(memcg
);
1464 node
= memcg
->last_scanned_node
;
1466 node
= next_node(node
, memcg
->scan_nodes
);
1467 if (node
== MAX_NUMNODES
)
1468 node
= first_node(memcg
->scan_nodes
);
1470 * We call this when we hit limit, not when pages are added to LRU.
1471 * No LRU may hold pages because all pages are UNEVICTABLE or
1472 * memcg is too small and all pages are not on LRU. In that case,
1473 * we use curret node.
1475 if (unlikely(node
== MAX_NUMNODES
))
1476 node
= numa_node_id();
1478 memcg
->last_scanned_node
= node
;
1482 int mem_cgroup_select_victim_node(struct mem_cgroup
*memcg
)
1488 static int mem_cgroup_soft_reclaim(struct mem_cgroup
*root_memcg
,
1491 unsigned long *total_scanned
)
1493 struct mem_cgroup
*victim
= NULL
;
1496 unsigned long excess
;
1497 unsigned long nr_scanned
;
1498 struct mem_cgroup_reclaim_cookie reclaim
= {
1503 excess
= soft_limit_excess(root_memcg
);
1506 victim
= mem_cgroup_iter(root_memcg
, victim
, &reclaim
);
1511 * If we have not been able to reclaim
1512 * anything, it might because there are
1513 * no reclaimable pages under this hierarchy
1518 * We want to do more targeted reclaim.
1519 * excess >> 2 is not to excessive so as to
1520 * reclaim too much, nor too less that we keep
1521 * coming back to reclaim from this cgroup
1523 if (total
>= (excess
>> 2) ||
1524 (loop
> MEM_CGROUP_MAX_RECLAIM_LOOPS
))
1529 total
+= mem_cgroup_shrink_node_zone(victim
, gfp_mask
, false,
1531 *total_scanned
+= nr_scanned
;
1532 if (!soft_limit_excess(root_memcg
))
1535 mem_cgroup_iter_break(root_memcg
, victim
);
1539 #ifdef CONFIG_LOCKDEP
1540 static struct lockdep_map memcg_oom_lock_dep_map
= {
1541 .name
= "memcg_oom_lock",
1545 static DEFINE_SPINLOCK(memcg_oom_lock
);
1548 * Check OOM-Killer is already running under our hierarchy.
1549 * If someone is running, return false.
1551 static bool mem_cgroup_oom_trylock(struct mem_cgroup
*memcg
)
1553 struct mem_cgroup
*iter
, *failed
= NULL
;
1555 spin_lock(&memcg_oom_lock
);
1557 for_each_mem_cgroup_tree(iter
, memcg
) {
1558 if (iter
->oom_lock
) {
1560 * this subtree of our hierarchy is already locked
1561 * so we cannot give a lock.
1564 mem_cgroup_iter_break(memcg
, iter
);
1567 iter
->oom_lock
= true;
1572 * OK, we failed to lock the whole subtree so we have
1573 * to clean up what we set up to the failing subtree
1575 for_each_mem_cgroup_tree(iter
, memcg
) {
1576 if (iter
== failed
) {
1577 mem_cgroup_iter_break(memcg
, iter
);
1580 iter
->oom_lock
= false;
1583 mutex_acquire(&memcg_oom_lock_dep_map
, 0, 1, _RET_IP_
);
1585 spin_unlock(&memcg_oom_lock
);
1590 static void mem_cgroup_oom_unlock(struct mem_cgroup
*memcg
)
1592 struct mem_cgroup
*iter
;
1594 spin_lock(&memcg_oom_lock
);
1595 mutex_release(&memcg_oom_lock_dep_map
, 1, _RET_IP_
);
1596 for_each_mem_cgroup_tree(iter
, memcg
)
1597 iter
->oom_lock
= false;
1598 spin_unlock(&memcg_oom_lock
);
1601 static void mem_cgroup_mark_under_oom(struct mem_cgroup
*memcg
)
1603 struct mem_cgroup
*iter
;
1605 spin_lock(&memcg_oom_lock
);
1606 for_each_mem_cgroup_tree(iter
, memcg
)
1608 spin_unlock(&memcg_oom_lock
);
1611 static void mem_cgroup_unmark_under_oom(struct mem_cgroup
*memcg
)
1613 struct mem_cgroup
*iter
;
1616 * When a new child is created while the hierarchy is under oom,
1617 * mem_cgroup_oom_lock() may not be called. Watch for underflow.
1619 spin_lock(&memcg_oom_lock
);
1620 for_each_mem_cgroup_tree(iter
, memcg
)
1621 if (iter
->under_oom
> 0)
1623 spin_unlock(&memcg_oom_lock
);
1626 static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq
);
1628 struct oom_wait_info
{
1629 struct mem_cgroup
*memcg
;
1633 static int memcg_oom_wake_function(wait_queue_t
*wait
,
1634 unsigned mode
, int sync
, void *arg
)
1636 struct mem_cgroup
*wake_memcg
= (struct mem_cgroup
*)arg
;
1637 struct mem_cgroup
*oom_wait_memcg
;
1638 struct oom_wait_info
*oom_wait_info
;
1640 oom_wait_info
= container_of(wait
, struct oom_wait_info
, wait
);
1641 oom_wait_memcg
= oom_wait_info
->memcg
;
1643 if (!mem_cgroup_is_descendant(wake_memcg
, oom_wait_memcg
) &&
1644 !mem_cgroup_is_descendant(oom_wait_memcg
, wake_memcg
))
1646 return autoremove_wake_function(wait
, mode
, sync
, arg
);
1649 static void memcg_oom_recover(struct mem_cgroup
*memcg
)
1652 * For the following lockless ->under_oom test, the only required
1653 * guarantee is that it must see the state asserted by an OOM when
1654 * this function is called as a result of userland actions
1655 * triggered by the notification of the OOM. This is trivially
1656 * achieved by invoking mem_cgroup_mark_under_oom() before
1657 * triggering notification.
1659 if (memcg
&& memcg
->under_oom
)
1660 __wake_up(&memcg_oom_waitq
, TASK_NORMAL
, 0, memcg
);
1663 static void mem_cgroup_oom(struct mem_cgroup
*memcg
, gfp_t mask
, int order
)
1665 if (!current
->memcg_may_oom
)
1668 * We are in the middle of the charge context here, so we
1669 * don't want to block when potentially sitting on a callstack
1670 * that holds all kinds of filesystem and mm locks.
1672 * Also, the caller may handle a failed allocation gracefully
1673 * (like optional page cache readahead) and so an OOM killer
1674 * invocation might not even be necessary.
1676 * That's why we don't do anything here except remember the
1677 * OOM context and then deal with it at the end of the page
1678 * fault when the stack is unwound, the locks are released,
1679 * and when we know whether the fault was overall successful.
1681 css_get(&memcg
->css
);
1682 current
->memcg_in_oom
= memcg
;
1683 current
->memcg_oom_gfp_mask
= mask
;
1684 current
->memcg_oom_order
= order
;
1688 * mem_cgroup_oom_synchronize - complete memcg OOM handling
1689 * @handle: actually kill/wait or just clean up the OOM state
1691 * This has to be called at the end of a page fault if the memcg OOM
1692 * handler was enabled.
1694 * Memcg supports userspace OOM handling where failed allocations must
1695 * sleep on a waitqueue until the userspace task resolves the
1696 * situation. Sleeping directly in the charge context with all kinds
1697 * of locks held is not a good idea, instead we remember an OOM state
1698 * in the task and mem_cgroup_oom_synchronize() has to be called at
1699 * the end of the page fault to complete the OOM handling.
1701 * Returns %true if an ongoing memcg OOM situation was detected and
1702 * completed, %false otherwise.
1704 bool mem_cgroup_oom_synchronize(bool handle
)
1706 struct mem_cgroup
*memcg
= current
->memcg_in_oom
;
1707 struct oom_wait_info owait
;
1710 /* OOM is global, do not handle */
1714 if (!handle
|| oom_killer_disabled
)
1717 owait
.memcg
= memcg
;
1718 owait
.wait
.flags
= 0;
1719 owait
.wait
.func
= memcg_oom_wake_function
;
1720 owait
.wait
.private = current
;
1721 INIT_LIST_HEAD(&owait
.wait
.task_list
);
1723 prepare_to_wait(&memcg_oom_waitq
, &owait
.wait
, TASK_KILLABLE
);
1724 mem_cgroup_mark_under_oom(memcg
);
1726 locked
= mem_cgroup_oom_trylock(memcg
);
1729 mem_cgroup_oom_notify(memcg
);
1731 if (locked
&& !memcg
->oom_kill_disable
) {
1732 mem_cgroup_unmark_under_oom(memcg
);
1733 finish_wait(&memcg_oom_waitq
, &owait
.wait
);
1734 mem_cgroup_out_of_memory(memcg
, current
->memcg_oom_gfp_mask
,
1735 current
->memcg_oom_order
);
1738 mem_cgroup_unmark_under_oom(memcg
);
1739 finish_wait(&memcg_oom_waitq
, &owait
.wait
);
1743 mem_cgroup_oom_unlock(memcg
);
1745 * There is no guarantee that an OOM-lock contender
1746 * sees the wakeups triggered by the OOM kill
1747 * uncharges. Wake any sleepers explicitely.
1749 memcg_oom_recover(memcg
);
1752 current
->memcg_in_oom
= NULL
;
1753 css_put(&memcg
->css
);
1758 * mem_cgroup_begin_page_stat - begin a page state statistics transaction
1759 * @page: page that is going to change accounted state
1761 * This function must mark the beginning of an accounted page state
1762 * change to prevent double accounting when the page is concurrently
1763 * being moved to another memcg:
1765 * memcg = mem_cgroup_begin_page_stat(page);
1766 * if (TestClearPageState(page))
1767 * mem_cgroup_update_page_stat(memcg, state, -1);
1768 * mem_cgroup_end_page_stat(memcg);
1770 struct mem_cgroup
*mem_cgroup_begin_page_stat(struct page
*page
)
1772 struct mem_cgroup
*memcg
;
1773 unsigned long flags
;
1776 * The RCU lock is held throughout the transaction. The fast
1777 * path can get away without acquiring the memcg->move_lock
1778 * because page moving starts with an RCU grace period.
1780 * The RCU lock also protects the memcg from being freed when
1781 * the page state that is going to change is the only thing
1782 * preventing the page from being uncharged.
1783 * E.g. end-writeback clearing PageWriteback(), which allows
1784 * migration to go ahead and uncharge the page before the
1785 * account transaction might be complete.
1789 if (mem_cgroup_disabled())
1792 memcg
= page
->mem_cgroup
;
1793 if (unlikely(!memcg
))
1796 if (atomic_read(&memcg
->moving_account
) <= 0)
1799 spin_lock_irqsave(&memcg
->move_lock
, flags
);
1800 if (memcg
!= page
->mem_cgroup
) {
1801 spin_unlock_irqrestore(&memcg
->move_lock
, flags
);
1806 * When charge migration first begins, we can have locked and
1807 * unlocked page stat updates happening concurrently. Track
1808 * the task who has the lock for mem_cgroup_end_page_stat().
1810 memcg
->move_lock_task
= current
;
1811 memcg
->move_lock_flags
= flags
;
1815 EXPORT_SYMBOL(mem_cgroup_begin_page_stat
);
1818 * mem_cgroup_end_page_stat - finish a page state statistics transaction
1819 * @memcg: the memcg that was accounted against
1821 void mem_cgroup_end_page_stat(struct mem_cgroup
*memcg
)
1823 if (memcg
&& memcg
->move_lock_task
== current
) {
1824 unsigned long flags
= memcg
->move_lock_flags
;
1826 memcg
->move_lock_task
= NULL
;
1827 memcg
->move_lock_flags
= 0;
1829 spin_unlock_irqrestore(&memcg
->move_lock
, flags
);
1834 EXPORT_SYMBOL(mem_cgroup_end_page_stat
);
1837 * size of first charge trial. "32" comes from vmscan.c's magic value.
1838 * TODO: maybe necessary to use big numbers in big irons.
1840 #define CHARGE_BATCH 32U
1841 struct memcg_stock_pcp
{
1842 struct mem_cgroup
*cached
; /* this never be root cgroup */
1843 unsigned int nr_pages
;
1844 struct work_struct work
;
1845 unsigned long flags
;
1846 #define FLUSHING_CACHED_CHARGE 0
1848 static DEFINE_PER_CPU(struct memcg_stock_pcp
, memcg_stock
);
1849 static DEFINE_MUTEX(percpu_charge_mutex
);
1852 * consume_stock: Try to consume stocked charge on this cpu.
1853 * @memcg: memcg to consume from.
1854 * @nr_pages: how many pages to charge.
1856 * The charges will only happen if @memcg matches the current cpu's memcg
1857 * stock, and at least @nr_pages are available in that stock. Failure to
1858 * service an allocation will refill the stock.
1860 * returns true if successful, false otherwise.
1862 static bool consume_stock(struct mem_cgroup
*memcg
, unsigned int nr_pages
)
1864 struct memcg_stock_pcp
*stock
;
1867 if (nr_pages
> CHARGE_BATCH
)
1870 stock
= &get_cpu_var(memcg_stock
);
1871 if (memcg
== stock
->cached
&& stock
->nr_pages
>= nr_pages
) {
1872 stock
->nr_pages
-= nr_pages
;
1875 put_cpu_var(memcg_stock
);
1880 * Returns stocks cached in percpu and reset cached information.
1882 static void drain_stock(struct memcg_stock_pcp
*stock
)
1884 struct mem_cgroup
*old
= stock
->cached
;
1886 if (stock
->nr_pages
) {
1887 page_counter_uncharge(&old
->memory
, stock
->nr_pages
);
1888 if (do_swap_account
)
1889 page_counter_uncharge(&old
->memsw
, stock
->nr_pages
);
1890 css_put_many(&old
->css
, stock
->nr_pages
);
1891 stock
->nr_pages
= 0;
1893 stock
->cached
= NULL
;
1897 * This must be called under preempt disabled or must be called by
1898 * a thread which is pinned to local cpu.
1900 static void drain_local_stock(struct work_struct
*dummy
)
1902 struct memcg_stock_pcp
*stock
= this_cpu_ptr(&memcg_stock
);
1904 clear_bit(FLUSHING_CACHED_CHARGE
, &stock
->flags
);
1908 * Cache charges(val) to local per_cpu area.
1909 * This will be consumed by consume_stock() function, later.
1911 static void refill_stock(struct mem_cgroup
*memcg
, unsigned int nr_pages
)
1913 struct memcg_stock_pcp
*stock
= &get_cpu_var(memcg_stock
);
1915 if (stock
->cached
!= memcg
) { /* reset if necessary */
1917 stock
->cached
= memcg
;
1919 stock
->nr_pages
+= nr_pages
;
1920 put_cpu_var(memcg_stock
);
1924 * Drains all per-CPU charge caches for given root_memcg resp. subtree
1925 * of the hierarchy under it.
1927 static void drain_all_stock(struct mem_cgroup
*root_memcg
)
1931 /* If someone's already draining, avoid adding running more workers. */
1932 if (!mutex_trylock(&percpu_charge_mutex
))
1934 /* Notify other cpus that system-wide "drain" is running */
1937 for_each_online_cpu(cpu
) {
1938 struct memcg_stock_pcp
*stock
= &per_cpu(memcg_stock
, cpu
);
1939 struct mem_cgroup
*memcg
;
1941 memcg
= stock
->cached
;
1942 if (!memcg
|| !stock
->nr_pages
)
1944 if (!mem_cgroup_is_descendant(memcg
, root_memcg
))
1946 if (!test_and_set_bit(FLUSHING_CACHED_CHARGE
, &stock
->flags
)) {
1948 drain_local_stock(&stock
->work
);
1950 schedule_work_on(cpu
, &stock
->work
);
1955 mutex_unlock(&percpu_charge_mutex
);
1958 static int memcg_cpu_hotplug_callback(struct notifier_block
*nb
,
1959 unsigned long action
,
1962 int cpu
= (unsigned long)hcpu
;
1963 struct memcg_stock_pcp
*stock
;
1965 if (action
== CPU_ONLINE
)
1968 if (action
!= CPU_DEAD
&& action
!= CPU_DEAD_FROZEN
)
1971 stock
= &per_cpu(memcg_stock
, cpu
);
1977 * Scheduled by try_charge() to be executed from the userland return path
1978 * and reclaims memory over the high limit.
1980 void mem_cgroup_handle_over_high(void)
1982 unsigned int nr_pages
= current
->memcg_nr_pages_over_high
;
1983 struct mem_cgroup
*memcg
, *pos
;
1985 if (likely(!nr_pages
))
1988 pos
= memcg
= get_mem_cgroup_from_mm(current
->mm
);
1991 if (page_counter_read(&pos
->memory
) <= pos
->high
)
1993 mem_cgroup_events(pos
, MEMCG_HIGH
, 1);
1994 try_to_free_mem_cgroup_pages(pos
, nr_pages
, GFP_KERNEL
, true);
1995 } while ((pos
= parent_mem_cgroup(pos
)));
1997 css_put(&memcg
->css
);
1998 current
->memcg_nr_pages_over_high
= 0;
2001 static int try_charge(struct mem_cgroup
*memcg
, gfp_t gfp_mask
,
2002 unsigned int nr_pages
)
2004 unsigned int batch
= max(CHARGE_BATCH
, nr_pages
);
2005 int nr_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
2006 struct mem_cgroup
*mem_over_limit
;
2007 struct page_counter
*counter
;
2008 unsigned long nr_reclaimed
;
2009 bool may_swap
= true;
2010 bool drained
= false;
2012 if (mem_cgroup_is_root(memcg
))
2015 if (consume_stock(memcg
, nr_pages
))
2018 if (!do_swap_account
||
2019 !page_counter_try_charge(&memcg
->memsw
, batch
, &counter
)) {
2020 if (!page_counter_try_charge(&memcg
->memory
, batch
, &counter
))
2022 if (do_swap_account
)
2023 page_counter_uncharge(&memcg
->memsw
, batch
);
2024 mem_over_limit
= mem_cgroup_from_counter(counter
, memory
);
2026 mem_over_limit
= mem_cgroup_from_counter(counter
, memsw
);
2030 if (batch
> nr_pages
) {
2036 * Unlike in global OOM situations, memcg is not in a physical
2037 * memory shortage. Allow dying and OOM-killed tasks to
2038 * bypass the last charges so that they can exit quickly and
2039 * free their memory.
2041 if (unlikely(test_thread_flag(TIF_MEMDIE
) ||
2042 fatal_signal_pending(current
) ||
2043 current
->flags
& PF_EXITING
))
2046 if (unlikely(task_in_memcg_oom(current
)))
2049 if (!(gfp_mask
& __GFP_WAIT
))
2052 mem_cgroup_events(mem_over_limit
, MEMCG_MAX
, 1);
2054 nr_reclaimed
= try_to_free_mem_cgroup_pages(mem_over_limit
, nr_pages
,
2055 gfp_mask
, may_swap
);
2057 if (mem_cgroup_margin(mem_over_limit
) >= nr_pages
)
2061 drain_all_stock(mem_over_limit
);
2066 if (gfp_mask
& __GFP_NORETRY
)
2069 * Even though the limit is exceeded at this point, reclaim
2070 * may have been able to free some pages. Retry the charge
2071 * before killing the task.
2073 * Only for regular pages, though: huge pages are rather
2074 * unlikely to succeed so close to the limit, and we fall back
2075 * to regular pages anyway in case of failure.
2077 if (nr_reclaimed
&& nr_pages
<= (1 << PAGE_ALLOC_COSTLY_ORDER
))
2080 * At task move, charge accounts can be doubly counted. So, it's
2081 * better to wait until the end of task_move if something is going on.
2083 if (mem_cgroup_wait_acct_move(mem_over_limit
))
2089 if (gfp_mask
& __GFP_NOFAIL
)
2092 if (fatal_signal_pending(current
))
2095 mem_cgroup_events(mem_over_limit
, MEMCG_OOM
, 1);
2097 mem_cgroup_oom(mem_over_limit
, gfp_mask
,
2098 get_order(nr_pages
* PAGE_SIZE
));
2100 if (!(gfp_mask
& __GFP_NOFAIL
))
2104 * The allocation either can't fail or will lead to more memory
2105 * being freed very soon. Allow memory usage go over the limit
2106 * temporarily by force charging it.
2108 page_counter_charge(&memcg
->memory
, nr_pages
);
2109 if (do_swap_account
)
2110 page_counter_charge(&memcg
->memsw
, nr_pages
);
2111 css_get_many(&memcg
->css
, nr_pages
);
2116 css_get_many(&memcg
->css
, batch
);
2117 if (batch
> nr_pages
)
2118 refill_stock(memcg
, batch
- nr_pages
);
2121 * If the hierarchy is above the normal consumption range, schedule
2122 * reclaim on returning to userland. We can perform reclaim here
2123 * if __GFP_WAIT but let's always punt for simplicity and so that
2124 * GFP_KERNEL can consistently be used during reclaim. @memcg is
2125 * not recorded as it most likely matches current's and won't
2126 * change in the meantime. As high limit is checked again before
2127 * reclaim, the cost of mismatch is negligible.
2130 if (page_counter_read(&memcg
->memory
) > memcg
->high
) {
2131 current
->memcg_nr_pages_over_high
+= nr_pages
;
2132 set_notify_resume(current
);
2135 } while ((memcg
= parent_mem_cgroup(memcg
)));
2140 static void cancel_charge(struct mem_cgroup
*memcg
, unsigned int nr_pages
)
2142 if (mem_cgroup_is_root(memcg
))
2145 page_counter_uncharge(&memcg
->memory
, nr_pages
);
2146 if (do_swap_account
)
2147 page_counter_uncharge(&memcg
->memsw
, nr_pages
);
2149 css_put_many(&memcg
->css
, nr_pages
);
2152 static void lock_page_lru(struct page
*page
, int *isolated
)
2154 struct zone
*zone
= page_zone(page
);
2156 spin_lock_irq(&zone
->lru_lock
);
2157 if (PageLRU(page
)) {
2158 struct lruvec
*lruvec
;
2160 lruvec
= mem_cgroup_page_lruvec(page
, zone
);
2162 del_page_from_lru_list(page
, lruvec
, page_lru(page
));
2168 static void unlock_page_lru(struct page
*page
, int isolated
)
2170 struct zone
*zone
= page_zone(page
);
2173 struct lruvec
*lruvec
;
2175 lruvec
= mem_cgroup_page_lruvec(page
, zone
);
2176 VM_BUG_ON_PAGE(PageLRU(page
), page
);
2178 add_page_to_lru_list(page
, lruvec
, page_lru(page
));
2180 spin_unlock_irq(&zone
->lru_lock
);
2183 static void commit_charge(struct page
*page
, struct mem_cgroup
*memcg
,
2188 VM_BUG_ON_PAGE(page
->mem_cgroup
, page
);
2191 * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page
2192 * may already be on some other mem_cgroup's LRU. Take care of it.
2195 lock_page_lru(page
, &isolated
);
2198 * Nobody should be changing or seriously looking at
2199 * page->mem_cgroup at this point:
2201 * - the page is uncharged
2203 * - the page is off-LRU
2205 * - an anonymous fault has exclusive page access, except for
2206 * a locked page table
2208 * - a page cache insertion, a swapin fault, or a migration
2209 * have the page locked
2211 page
->mem_cgroup
= memcg
;
2214 unlock_page_lru(page
, isolated
);
2217 #ifdef CONFIG_MEMCG_KMEM
2218 static int memcg_alloc_cache_id(void)
2223 id
= ida_simple_get(&memcg_cache_ida
,
2224 0, MEMCG_CACHES_MAX_SIZE
, GFP_KERNEL
);
2228 if (id
< memcg_nr_cache_ids
)
2232 * There's no space for the new id in memcg_caches arrays,
2233 * so we have to grow them.
2235 down_write(&memcg_cache_ids_sem
);
2237 size
= 2 * (id
+ 1);
2238 if (size
< MEMCG_CACHES_MIN_SIZE
)
2239 size
= MEMCG_CACHES_MIN_SIZE
;
2240 else if (size
> MEMCG_CACHES_MAX_SIZE
)
2241 size
= MEMCG_CACHES_MAX_SIZE
;
2243 err
= memcg_update_all_caches(size
);
2245 err
= memcg_update_all_list_lrus(size
);
2247 memcg_nr_cache_ids
= size
;
2249 up_write(&memcg_cache_ids_sem
);
2252 ida_simple_remove(&memcg_cache_ida
, id
);
2258 static void memcg_free_cache_id(int id
)
2260 ida_simple_remove(&memcg_cache_ida
, id
);
2263 struct memcg_kmem_cache_create_work
{
2264 struct mem_cgroup
*memcg
;
2265 struct kmem_cache
*cachep
;
2266 struct work_struct work
;
2269 static void memcg_kmem_cache_create_func(struct work_struct
*w
)
2271 struct memcg_kmem_cache_create_work
*cw
=
2272 container_of(w
, struct memcg_kmem_cache_create_work
, work
);
2273 struct mem_cgroup
*memcg
= cw
->memcg
;
2274 struct kmem_cache
*cachep
= cw
->cachep
;
2276 memcg_create_kmem_cache(memcg
, cachep
);
2278 css_put(&memcg
->css
);
2283 * Enqueue the creation of a per-memcg kmem_cache.
2285 static void __memcg_schedule_kmem_cache_create(struct mem_cgroup
*memcg
,
2286 struct kmem_cache
*cachep
)
2288 struct memcg_kmem_cache_create_work
*cw
;
2290 cw
= kmalloc(sizeof(*cw
), GFP_NOWAIT
);
2294 css_get(&memcg
->css
);
2297 cw
->cachep
= cachep
;
2298 INIT_WORK(&cw
->work
, memcg_kmem_cache_create_func
);
2300 schedule_work(&cw
->work
);
2303 static void memcg_schedule_kmem_cache_create(struct mem_cgroup
*memcg
,
2304 struct kmem_cache
*cachep
)
2307 * We need to stop accounting when we kmalloc, because if the
2308 * corresponding kmalloc cache is not yet created, the first allocation
2309 * in __memcg_schedule_kmem_cache_create will recurse.
2311 * However, it is better to enclose the whole function. Depending on
2312 * the debugging options enabled, INIT_WORK(), for instance, can
2313 * trigger an allocation. This too, will make us recurse. Because at
2314 * this point we can't allow ourselves back into memcg_kmem_get_cache,
2315 * the safest choice is to do it like this, wrapping the whole function.
2317 current
->memcg_kmem_skip_account
= 1;
2318 __memcg_schedule_kmem_cache_create(memcg
, cachep
);
2319 current
->memcg_kmem_skip_account
= 0;
2323 * Return the kmem_cache we're supposed to use for a slab allocation.
2324 * We try to use the current memcg's version of the cache.
2326 * If the cache does not exist yet, if we are the first user of it,
2327 * we either create it immediately, if possible, or create it asynchronously
2329 * In the latter case, we will let the current allocation go through with
2330 * the original cache.
2332 * Can't be called in interrupt context or from kernel threads.
2333 * This function needs to be called with rcu_read_lock() held.
2335 struct kmem_cache
*__memcg_kmem_get_cache(struct kmem_cache
*cachep
)
2337 struct mem_cgroup
*memcg
;
2338 struct kmem_cache
*memcg_cachep
;
2341 VM_BUG_ON(!is_root_cache(cachep
));
2343 if (current
->memcg_kmem_skip_account
)
2346 memcg
= get_mem_cgroup_from_mm(current
->mm
);
2347 kmemcg_id
= READ_ONCE(memcg
->kmemcg_id
);
2351 memcg_cachep
= cache_from_memcg_idx(cachep
, kmemcg_id
);
2352 if (likely(memcg_cachep
))
2353 return memcg_cachep
;
2356 * If we are in a safe context (can wait, and not in interrupt
2357 * context), we could be be predictable and return right away.
2358 * This would guarantee that the allocation being performed
2359 * already belongs in the new cache.
2361 * However, there are some clashes that can arrive from locking.
2362 * For instance, because we acquire the slab_mutex while doing
2363 * memcg_create_kmem_cache, this means no further allocation
2364 * could happen with the slab_mutex held. So it's better to
2367 memcg_schedule_kmem_cache_create(memcg
, cachep
);
2369 css_put(&memcg
->css
);
2373 void __memcg_kmem_put_cache(struct kmem_cache
*cachep
)
2375 if (!is_root_cache(cachep
))
2376 css_put(&cachep
->memcg_params
.memcg
->css
);
2379 int __memcg_kmem_charge_memcg(struct page
*page
, gfp_t gfp
, int order
,
2380 struct mem_cgroup
*memcg
)
2382 unsigned int nr_pages
= 1 << order
;
2383 struct page_counter
*counter
;
2386 if (!memcg_kmem_is_active(memcg
))
2389 ret
= page_counter_try_charge(&memcg
->kmem
, nr_pages
, &counter
);
2393 ret
= try_charge(memcg
, gfp
, nr_pages
);
2395 page_counter_uncharge(&memcg
->kmem
, nr_pages
);
2399 page
->mem_cgroup
= memcg
;
2404 int __memcg_kmem_charge(struct page
*page
, gfp_t gfp
, int order
)
2406 struct mem_cgroup
*memcg
;
2409 memcg
= get_mem_cgroup_from_mm(current
->mm
);
2410 ret
= __memcg_kmem_charge_memcg(page
, gfp
, order
, memcg
);
2411 css_put(&memcg
->css
);
2415 void __memcg_kmem_uncharge(struct page
*page
, int order
)
2417 struct mem_cgroup
*memcg
= page
->mem_cgroup
;
2418 unsigned int nr_pages
= 1 << order
;
2423 VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg
), page
);
2425 page_counter_uncharge(&memcg
->kmem
, nr_pages
);
2426 page_counter_uncharge(&memcg
->memory
, nr_pages
);
2427 if (do_swap_account
)
2428 page_counter_uncharge(&memcg
->memsw
, nr_pages
);
2430 page
->mem_cgroup
= NULL
;
2431 css_put_many(&memcg
->css
, nr_pages
);
2433 #endif /* CONFIG_MEMCG_KMEM */
2435 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
2438 * Because tail pages are not marked as "used", set it. We're under
2439 * zone->lru_lock, 'splitting on pmd' and compound_lock.
2440 * charge/uncharge will be never happen and move_account() is done under
2441 * compound_lock(), so we don't have to take care of races.
2443 void mem_cgroup_split_huge_fixup(struct page
*head
)
2447 if (mem_cgroup_disabled())
2450 for (i
= 1; i
< HPAGE_PMD_NR
; i
++)
2451 head
[i
].mem_cgroup
= head
->mem_cgroup
;
2453 __this_cpu_sub(head
->mem_cgroup
->stat
->count
[MEM_CGROUP_STAT_RSS_HUGE
],
2456 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
2458 #ifdef CONFIG_MEMCG_SWAP
2459 static void mem_cgroup_swap_statistics(struct mem_cgroup
*memcg
,
2462 int val
= (charge
) ? 1 : -1;
2463 this_cpu_add(memcg
->stat
->count
[MEM_CGROUP_STAT_SWAP
], val
);
2467 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
2468 * @entry: swap entry to be moved
2469 * @from: mem_cgroup which the entry is moved from
2470 * @to: mem_cgroup which the entry is moved to
2472 * It succeeds only when the swap_cgroup's record for this entry is the same
2473 * as the mem_cgroup's id of @from.
2475 * Returns 0 on success, -EINVAL on failure.
2477 * The caller must have charged to @to, IOW, called page_counter_charge() about
2478 * both res and memsw, and called css_get().
2480 static int mem_cgroup_move_swap_account(swp_entry_t entry
,
2481 struct mem_cgroup
*from
, struct mem_cgroup
*to
)
2483 unsigned short old_id
, new_id
;
2485 old_id
= mem_cgroup_id(from
);
2486 new_id
= mem_cgroup_id(to
);
2488 if (swap_cgroup_cmpxchg(entry
, old_id
, new_id
) == old_id
) {
2489 mem_cgroup_swap_statistics(from
, false);
2490 mem_cgroup_swap_statistics(to
, true);
2496 static inline int mem_cgroup_move_swap_account(swp_entry_t entry
,
2497 struct mem_cgroup
*from
, struct mem_cgroup
*to
)
2503 static DEFINE_MUTEX(memcg_limit_mutex
);
2505 static int mem_cgroup_resize_limit(struct mem_cgroup
*memcg
,
2506 unsigned long limit
)
2508 unsigned long curusage
;
2509 unsigned long oldusage
;
2510 bool enlarge
= false;
2515 * For keeping hierarchical_reclaim simple, how long we should retry
2516 * is depends on callers. We set our retry-count to be function
2517 * of # of children which we should visit in this loop.
2519 retry_count
= MEM_CGROUP_RECLAIM_RETRIES
*
2520 mem_cgroup_count_children(memcg
);
2522 oldusage
= page_counter_read(&memcg
->memory
);
2525 if (signal_pending(current
)) {
2530 mutex_lock(&memcg_limit_mutex
);
2531 if (limit
> memcg
->memsw
.limit
) {
2532 mutex_unlock(&memcg_limit_mutex
);
2536 if (limit
> memcg
->memory
.limit
)
2538 ret
= page_counter_limit(&memcg
->memory
, limit
);
2539 mutex_unlock(&memcg_limit_mutex
);
2544 try_to_free_mem_cgroup_pages(memcg
, 1, GFP_KERNEL
, true);
2546 curusage
= page_counter_read(&memcg
->memory
);
2547 /* Usage is reduced ? */
2548 if (curusage
>= oldusage
)
2551 oldusage
= curusage
;
2552 } while (retry_count
);
2554 if (!ret
&& enlarge
)
2555 memcg_oom_recover(memcg
);
2560 static int mem_cgroup_resize_memsw_limit(struct mem_cgroup
*memcg
,
2561 unsigned long limit
)
2563 unsigned long curusage
;
2564 unsigned long oldusage
;
2565 bool enlarge
= false;
2569 /* see mem_cgroup_resize_res_limit */
2570 retry_count
= MEM_CGROUP_RECLAIM_RETRIES
*
2571 mem_cgroup_count_children(memcg
);
2573 oldusage
= page_counter_read(&memcg
->memsw
);
2576 if (signal_pending(current
)) {
2581 mutex_lock(&memcg_limit_mutex
);
2582 if (limit
< memcg
->memory
.limit
) {
2583 mutex_unlock(&memcg_limit_mutex
);
2587 if (limit
> memcg
->memsw
.limit
)
2589 ret
= page_counter_limit(&memcg
->memsw
, limit
);
2590 mutex_unlock(&memcg_limit_mutex
);
2595 try_to_free_mem_cgroup_pages(memcg
, 1, GFP_KERNEL
, false);
2597 curusage
= page_counter_read(&memcg
->memsw
);
2598 /* Usage is reduced ? */
2599 if (curusage
>= oldusage
)
2602 oldusage
= curusage
;
2603 } while (retry_count
);
2605 if (!ret
&& enlarge
)
2606 memcg_oom_recover(memcg
);
2611 unsigned long mem_cgroup_soft_limit_reclaim(struct zone
*zone
, int order
,
2613 unsigned long *total_scanned
)
2615 unsigned long nr_reclaimed
= 0;
2616 struct mem_cgroup_per_zone
*mz
, *next_mz
= NULL
;
2617 unsigned long reclaimed
;
2619 struct mem_cgroup_tree_per_zone
*mctz
;
2620 unsigned long excess
;
2621 unsigned long nr_scanned
;
2626 mctz
= soft_limit_tree_node_zone(zone_to_nid(zone
), zone_idx(zone
));
2628 * This loop can run a while, specially if mem_cgroup's continuously
2629 * keep exceeding their soft limit and putting the system under
2636 mz
= mem_cgroup_largest_soft_limit_node(mctz
);
2641 reclaimed
= mem_cgroup_soft_reclaim(mz
->memcg
, zone
,
2642 gfp_mask
, &nr_scanned
);
2643 nr_reclaimed
+= reclaimed
;
2644 *total_scanned
+= nr_scanned
;
2645 spin_lock_irq(&mctz
->lock
);
2646 __mem_cgroup_remove_exceeded(mz
, mctz
);
2649 * If we failed to reclaim anything from this memory cgroup
2650 * it is time to move on to the next cgroup
2654 next_mz
= __mem_cgroup_largest_soft_limit_node(mctz
);
2656 excess
= soft_limit_excess(mz
->memcg
);
2658 * One school of thought says that we should not add
2659 * back the node to the tree if reclaim returns 0.
2660 * But our reclaim could return 0, simply because due
2661 * to priority we are exposing a smaller subset of
2662 * memory to reclaim from. Consider this as a longer
2665 /* If excess == 0, no tree ops */
2666 __mem_cgroup_insert_exceeded(mz
, mctz
, excess
);
2667 spin_unlock_irq(&mctz
->lock
);
2668 css_put(&mz
->memcg
->css
);
2671 * Could not reclaim anything and there are no more
2672 * mem cgroups to try or we seem to be looping without
2673 * reclaiming anything.
2675 if (!nr_reclaimed
&&
2677 loop
> MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS
))
2679 } while (!nr_reclaimed
);
2681 css_put(&next_mz
->memcg
->css
);
2682 return nr_reclaimed
;
2686 * Test whether @memcg has children, dead or alive. Note that this
2687 * function doesn't care whether @memcg has use_hierarchy enabled and
2688 * returns %true if there are child csses according to the cgroup
2689 * hierarchy. Testing use_hierarchy is the caller's responsiblity.
2691 static inline bool memcg_has_children(struct mem_cgroup
*memcg
)
2696 * The lock does not prevent addition or deletion of children, but
2697 * it prevents a new child from being initialized based on this
2698 * parent in css_online(), so it's enough to decide whether
2699 * hierarchically inherited attributes can still be changed or not.
2701 lockdep_assert_held(&memcg_create_mutex
);
2704 ret
= css_next_child(NULL
, &memcg
->css
);
2710 * Reclaims as many pages from the given memcg as possible and moves
2711 * the rest to the parent.
2713 * Caller is responsible for holding css reference for memcg.
2715 static int mem_cgroup_force_empty(struct mem_cgroup
*memcg
)
2717 int nr_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
2719 /* we call try-to-free pages for make this cgroup empty */
2720 lru_add_drain_all();
2721 /* try to free all pages in this cgroup */
2722 while (nr_retries
&& page_counter_read(&memcg
->memory
)) {
2725 if (signal_pending(current
))
2728 progress
= try_to_free_mem_cgroup_pages(memcg
, 1,
2732 /* maybe some writeback is necessary */
2733 congestion_wait(BLK_RW_ASYNC
, HZ
/10);
2741 static ssize_t
mem_cgroup_force_empty_write(struct kernfs_open_file
*of
,
2742 char *buf
, size_t nbytes
,
2745 struct mem_cgroup
*memcg
= mem_cgroup_from_css(of_css(of
));
2747 if (mem_cgroup_is_root(memcg
))
2749 return mem_cgroup_force_empty(memcg
) ?: nbytes
;
2752 static u64
mem_cgroup_hierarchy_read(struct cgroup_subsys_state
*css
,
2755 return mem_cgroup_from_css(css
)->use_hierarchy
;
2758 static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state
*css
,
2759 struct cftype
*cft
, u64 val
)
2762 struct mem_cgroup
*memcg
= mem_cgroup_from_css(css
);
2763 struct mem_cgroup
*parent_memcg
= mem_cgroup_from_css(memcg
->css
.parent
);
2765 mutex_lock(&memcg_create_mutex
);
2767 if (memcg
->use_hierarchy
== val
)
2771 * If parent's use_hierarchy is set, we can't make any modifications
2772 * in the child subtrees. If it is unset, then the change can
2773 * occur, provided the current cgroup has no children.
2775 * For the root cgroup, parent_mem is NULL, we allow value to be
2776 * set if there are no children.
2778 if ((!parent_memcg
|| !parent_memcg
->use_hierarchy
) &&
2779 (val
== 1 || val
== 0)) {
2780 if (!memcg_has_children(memcg
))
2781 memcg
->use_hierarchy
= val
;
2788 mutex_unlock(&memcg_create_mutex
);
2793 static unsigned long tree_stat(struct mem_cgroup
*memcg
,
2794 enum mem_cgroup_stat_index idx
)
2796 struct mem_cgroup
*iter
;
2797 unsigned long val
= 0;
2799 for_each_mem_cgroup_tree(iter
, memcg
)
2800 val
+= mem_cgroup_read_stat(iter
, idx
);
2805 static inline u64
mem_cgroup_usage(struct mem_cgroup
*memcg
, bool swap
)
2809 if (mem_cgroup_is_root(memcg
)) {
2810 val
= tree_stat(memcg
, MEM_CGROUP_STAT_CACHE
);
2811 val
+= tree_stat(memcg
, MEM_CGROUP_STAT_RSS
);
2813 val
+= tree_stat(memcg
, MEM_CGROUP_STAT_SWAP
);
2816 val
= page_counter_read(&memcg
->memory
);
2818 val
= page_counter_read(&memcg
->memsw
);
2820 return val
<< PAGE_SHIFT
;
2831 static u64
mem_cgroup_read_u64(struct cgroup_subsys_state
*css
,
2834 struct mem_cgroup
*memcg
= mem_cgroup_from_css(css
);
2835 struct page_counter
*counter
;
2837 switch (MEMFILE_TYPE(cft
->private)) {
2839 counter
= &memcg
->memory
;
2842 counter
= &memcg
->memsw
;
2845 counter
= &memcg
->kmem
;
2851 switch (MEMFILE_ATTR(cft
->private)) {
2853 if (counter
== &memcg
->memory
)
2854 return mem_cgroup_usage(memcg
, false);
2855 if (counter
== &memcg
->memsw
)
2856 return mem_cgroup_usage(memcg
, true);
2857 return (u64
)page_counter_read(counter
) * PAGE_SIZE
;
2859 return (u64
)counter
->limit
* PAGE_SIZE
;
2861 return (u64
)counter
->watermark
* PAGE_SIZE
;
2863 return counter
->failcnt
;
2864 case RES_SOFT_LIMIT
:
2865 return (u64
)memcg
->soft_limit
* PAGE_SIZE
;
2871 #ifdef CONFIG_MEMCG_KMEM
2872 static int memcg_activate_kmem(struct mem_cgroup
*memcg
,
2873 unsigned long nr_pages
)
2878 BUG_ON(memcg
->kmemcg_id
>= 0);
2879 BUG_ON(memcg
->kmem_acct_activated
);
2880 BUG_ON(memcg
->kmem_acct_active
);
2883 * For simplicity, we won't allow this to be disabled. It also can't
2884 * be changed if the cgroup has children already, or if tasks had
2887 * If tasks join before we set the limit, a person looking at
2888 * kmem.usage_in_bytes will have no way to determine when it took
2889 * place, which makes the value quite meaningless.
2891 * After it first became limited, changes in the value of the limit are
2892 * of course permitted.
2894 mutex_lock(&memcg_create_mutex
);
2895 if (cgroup_has_tasks(memcg
->css
.cgroup
) ||
2896 (memcg
->use_hierarchy
&& memcg_has_children(memcg
)))
2898 mutex_unlock(&memcg_create_mutex
);
2902 memcg_id
= memcg_alloc_cache_id();
2909 * We couldn't have accounted to this cgroup, because it hasn't got
2910 * activated yet, so this should succeed.
2912 err
= page_counter_limit(&memcg
->kmem
, nr_pages
);
2915 static_key_slow_inc(&memcg_kmem_enabled_key
);
2917 * A memory cgroup is considered kmem-active as soon as it gets
2918 * kmemcg_id. Setting the id after enabling static branching will
2919 * guarantee no one starts accounting before all call sites are
2922 memcg
->kmemcg_id
= memcg_id
;
2923 memcg
->kmem_acct_activated
= true;
2924 memcg
->kmem_acct_active
= true;
2929 static int memcg_update_kmem_limit(struct mem_cgroup
*memcg
,
2930 unsigned long limit
)
2934 mutex_lock(&memcg_limit_mutex
);
2935 if (!memcg_kmem_is_active(memcg
))
2936 ret
= memcg_activate_kmem(memcg
, limit
);
2938 ret
= page_counter_limit(&memcg
->kmem
, limit
);
2939 mutex_unlock(&memcg_limit_mutex
);
2943 static int memcg_propagate_kmem(struct mem_cgroup
*memcg
)
2946 struct mem_cgroup
*parent
= parent_mem_cgroup(memcg
);
2951 mutex_lock(&memcg_limit_mutex
);
2953 * If the parent cgroup is not kmem-active now, it cannot be activated
2954 * after this point, because it has at least one child already.
2956 if (memcg_kmem_is_active(parent
))
2957 ret
= memcg_activate_kmem(memcg
, PAGE_COUNTER_MAX
);
2958 mutex_unlock(&memcg_limit_mutex
);
2962 static int memcg_update_kmem_limit(struct mem_cgroup
*memcg
,
2963 unsigned long limit
)
2967 #endif /* CONFIG_MEMCG_KMEM */
2970 * The user of this function is...
2973 static ssize_t
mem_cgroup_write(struct kernfs_open_file
*of
,
2974 char *buf
, size_t nbytes
, loff_t off
)
2976 struct mem_cgroup
*memcg
= mem_cgroup_from_css(of_css(of
));
2977 unsigned long nr_pages
;
2980 buf
= strstrip(buf
);
2981 ret
= page_counter_memparse(buf
, "-1", &nr_pages
);
2985 switch (MEMFILE_ATTR(of_cft(of
)->private)) {
2987 if (mem_cgroup_is_root(memcg
)) { /* Can't set limit on root */
2991 switch (MEMFILE_TYPE(of_cft(of
)->private)) {
2993 ret
= mem_cgroup_resize_limit(memcg
, nr_pages
);
2996 ret
= mem_cgroup_resize_memsw_limit(memcg
, nr_pages
);
2999 ret
= memcg_update_kmem_limit(memcg
, nr_pages
);
3003 case RES_SOFT_LIMIT
:
3004 memcg
->soft_limit
= nr_pages
;
3008 return ret
?: nbytes
;
3011 static ssize_t
mem_cgroup_reset(struct kernfs_open_file
*of
, char *buf
,
3012 size_t nbytes
, loff_t off
)
3014 struct mem_cgroup
*memcg
= mem_cgroup_from_css(of_css(of
));
3015 struct page_counter
*counter
;
3017 switch (MEMFILE_TYPE(of_cft(of
)->private)) {
3019 counter
= &memcg
->memory
;
3022 counter
= &memcg
->memsw
;
3025 counter
= &memcg
->kmem
;
3031 switch (MEMFILE_ATTR(of_cft(of
)->private)) {
3033 page_counter_reset_watermark(counter
);
3036 counter
->failcnt
= 0;
3045 static u64
mem_cgroup_move_charge_read(struct cgroup_subsys_state
*css
,
3048 return mem_cgroup_from_css(css
)->move_charge_at_immigrate
;
3052 static int mem_cgroup_move_charge_write(struct cgroup_subsys_state
*css
,
3053 struct cftype
*cft
, u64 val
)
3055 struct mem_cgroup
*memcg
= mem_cgroup_from_css(css
);
3057 if (val
& ~MOVE_MASK
)
3061 * No kind of locking is needed in here, because ->can_attach() will
3062 * check this value once in the beginning of the process, and then carry
3063 * on with stale data. This means that changes to this value will only
3064 * affect task migrations starting after the change.
3066 memcg
->move_charge_at_immigrate
= val
;
3070 static int mem_cgroup_move_charge_write(struct cgroup_subsys_state
*css
,
3071 struct cftype
*cft
, u64 val
)
3078 static int memcg_numa_stat_show(struct seq_file
*m
, void *v
)
3082 unsigned int lru_mask
;
3085 static const struct numa_stat stats
[] = {
3086 { "total", LRU_ALL
},
3087 { "file", LRU_ALL_FILE
},
3088 { "anon", LRU_ALL_ANON
},
3089 { "unevictable", BIT(LRU_UNEVICTABLE
) },
3091 const struct numa_stat
*stat
;
3094 struct mem_cgroup
*memcg
= mem_cgroup_from_css(seq_css(m
));
3096 for (stat
= stats
; stat
< stats
+ ARRAY_SIZE(stats
); stat
++) {
3097 nr
= mem_cgroup_nr_lru_pages(memcg
, stat
->lru_mask
);
3098 seq_printf(m
, "%s=%lu", stat
->name
, nr
);
3099 for_each_node_state(nid
, N_MEMORY
) {
3100 nr
= mem_cgroup_node_nr_lru_pages(memcg
, nid
,
3102 seq_printf(m
, " N%d=%lu", nid
, nr
);
3107 for (stat
= stats
; stat
< stats
+ ARRAY_SIZE(stats
); stat
++) {
3108 struct mem_cgroup
*iter
;
3111 for_each_mem_cgroup_tree(iter
, memcg
)
3112 nr
+= mem_cgroup_nr_lru_pages(iter
, stat
->lru_mask
);
3113 seq_printf(m
, "hierarchical_%s=%lu", stat
->name
, nr
);
3114 for_each_node_state(nid
, N_MEMORY
) {
3116 for_each_mem_cgroup_tree(iter
, memcg
)
3117 nr
+= mem_cgroup_node_nr_lru_pages(
3118 iter
, nid
, stat
->lru_mask
);
3119 seq_printf(m
, " N%d=%lu", nid
, nr
);
3126 #endif /* CONFIG_NUMA */
3128 static int memcg_stat_show(struct seq_file
*m
, void *v
)
3130 struct mem_cgroup
*memcg
= mem_cgroup_from_css(seq_css(m
));
3131 unsigned long memory
, memsw
;
3132 struct mem_cgroup
*mi
;
3135 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_stat_names
) !=
3136 MEM_CGROUP_STAT_NSTATS
);
3137 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_events_names
) !=
3138 MEM_CGROUP_EVENTS_NSTATS
);
3139 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names
) != NR_LRU_LISTS
);
3141 for (i
= 0; i
< MEM_CGROUP_STAT_NSTATS
; i
++) {
3142 if (i
== MEM_CGROUP_STAT_SWAP
&& !do_swap_account
)
3144 seq_printf(m
, "%s %lu\n", mem_cgroup_stat_names
[i
],
3145 mem_cgroup_read_stat(memcg
, i
) * PAGE_SIZE
);
3148 for (i
= 0; i
< MEM_CGROUP_EVENTS_NSTATS
; i
++)
3149 seq_printf(m
, "%s %lu\n", mem_cgroup_events_names
[i
],
3150 mem_cgroup_read_events(memcg
, i
));
3152 for (i
= 0; i
< NR_LRU_LISTS
; i
++)
3153 seq_printf(m
, "%s %lu\n", mem_cgroup_lru_names
[i
],
3154 mem_cgroup_nr_lru_pages(memcg
, BIT(i
)) * PAGE_SIZE
);
3156 /* Hierarchical information */
3157 memory
= memsw
= PAGE_COUNTER_MAX
;
3158 for (mi
= memcg
; mi
; mi
= parent_mem_cgroup(mi
)) {
3159 memory
= min(memory
, mi
->memory
.limit
);
3160 memsw
= min(memsw
, mi
->memsw
.limit
);
3162 seq_printf(m
, "hierarchical_memory_limit %llu\n",
3163 (u64
)memory
* PAGE_SIZE
);
3164 if (do_swap_account
)
3165 seq_printf(m
, "hierarchical_memsw_limit %llu\n",
3166 (u64
)memsw
* PAGE_SIZE
);
3168 for (i
= 0; i
< MEM_CGROUP_STAT_NSTATS
; i
++) {
3169 unsigned long long val
= 0;
3171 if (i
== MEM_CGROUP_STAT_SWAP
&& !do_swap_account
)
3173 for_each_mem_cgroup_tree(mi
, memcg
)
3174 val
+= mem_cgroup_read_stat(mi
, i
) * PAGE_SIZE
;
3175 seq_printf(m
, "total_%s %llu\n", mem_cgroup_stat_names
[i
], val
);
3178 for (i
= 0; i
< MEM_CGROUP_EVENTS_NSTATS
; i
++) {
3179 unsigned long long val
= 0;
3181 for_each_mem_cgroup_tree(mi
, memcg
)
3182 val
+= mem_cgroup_read_events(mi
, i
);
3183 seq_printf(m
, "total_%s %llu\n",
3184 mem_cgroup_events_names
[i
], val
);
3187 for (i
= 0; i
< NR_LRU_LISTS
; i
++) {
3188 unsigned long long val
= 0;
3190 for_each_mem_cgroup_tree(mi
, memcg
)
3191 val
+= mem_cgroup_nr_lru_pages(mi
, BIT(i
)) * PAGE_SIZE
;
3192 seq_printf(m
, "total_%s %llu\n", mem_cgroup_lru_names
[i
], val
);
3195 #ifdef CONFIG_DEBUG_VM
3198 struct mem_cgroup_per_zone
*mz
;
3199 struct zone_reclaim_stat
*rstat
;
3200 unsigned long recent_rotated
[2] = {0, 0};
3201 unsigned long recent_scanned
[2] = {0, 0};
3203 for_each_online_node(nid
)
3204 for (zid
= 0; zid
< MAX_NR_ZONES
; zid
++) {
3205 mz
= &memcg
->nodeinfo
[nid
]->zoneinfo
[zid
];
3206 rstat
= &mz
->lruvec
.reclaim_stat
;
3208 recent_rotated
[0] += rstat
->recent_rotated
[0];
3209 recent_rotated
[1] += rstat
->recent_rotated
[1];
3210 recent_scanned
[0] += rstat
->recent_scanned
[0];
3211 recent_scanned
[1] += rstat
->recent_scanned
[1];
3213 seq_printf(m
, "recent_rotated_anon %lu\n", recent_rotated
[0]);
3214 seq_printf(m
, "recent_rotated_file %lu\n", recent_rotated
[1]);
3215 seq_printf(m
, "recent_scanned_anon %lu\n", recent_scanned
[0]);
3216 seq_printf(m
, "recent_scanned_file %lu\n", recent_scanned
[1]);
3223 static u64
mem_cgroup_swappiness_read(struct cgroup_subsys_state
*css
,
3226 struct mem_cgroup
*memcg
= mem_cgroup_from_css(css
);
3228 return mem_cgroup_swappiness(memcg
);
3231 static int mem_cgroup_swappiness_write(struct cgroup_subsys_state
*css
,
3232 struct cftype
*cft
, u64 val
)
3234 struct mem_cgroup
*memcg
= mem_cgroup_from_css(css
);
3240 memcg
->swappiness
= val
;
3242 vm_swappiness
= val
;
3247 static void __mem_cgroup_threshold(struct mem_cgroup
*memcg
, bool swap
)
3249 struct mem_cgroup_threshold_ary
*t
;
3250 unsigned long usage
;
3255 t
= rcu_dereference(memcg
->thresholds
.primary
);
3257 t
= rcu_dereference(memcg
->memsw_thresholds
.primary
);
3262 usage
= mem_cgroup_usage(memcg
, swap
);
3265 * current_threshold points to threshold just below or equal to usage.
3266 * If it's not true, a threshold was crossed after last
3267 * call of __mem_cgroup_threshold().
3269 i
= t
->current_threshold
;
3272 * Iterate backward over array of thresholds starting from
3273 * current_threshold and check if a threshold is crossed.
3274 * If none of thresholds below usage is crossed, we read
3275 * only one element of the array here.
3277 for (; i
>= 0 && unlikely(t
->entries
[i
].threshold
> usage
); i
--)
3278 eventfd_signal(t
->entries
[i
].eventfd
, 1);
3280 /* i = current_threshold + 1 */
3284 * Iterate forward over array of thresholds starting from
3285 * current_threshold+1 and check if a threshold is crossed.
3286 * If none of thresholds above usage is crossed, we read
3287 * only one element of the array here.
3289 for (; i
< t
->size
&& unlikely(t
->entries
[i
].threshold
<= usage
); i
++)
3290 eventfd_signal(t
->entries
[i
].eventfd
, 1);
3292 /* Update current_threshold */
3293 t
->current_threshold
= i
- 1;
3298 static void mem_cgroup_threshold(struct mem_cgroup
*memcg
)
3301 __mem_cgroup_threshold(memcg
, false);
3302 if (do_swap_account
)
3303 __mem_cgroup_threshold(memcg
, true);
3305 memcg
= parent_mem_cgroup(memcg
);
3309 static int compare_thresholds(const void *a
, const void *b
)
3311 const struct mem_cgroup_threshold
*_a
= a
;
3312 const struct mem_cgroup_threshold
*_b
= b
;
3314 if (_a
->threshold
> _b
->threshold
)
3317 if (_a
->threshold
< _b
->threshold
)
3323 static int mem_cgroup_oom_notify_cb(struct mem_cgroup
*memcg
)
3325 struct mem_cgroup_eventfd_list
*ev
;
3327 spin_lock(&memcg_oom_lock
);
3329 list_for_each_entry(ev
, &memcg
->oom_notify
, list
)
3330 eventfd_signal(ev
->eventfd
, 1);
3332 spin_unlock(&memcg_oom_lock
);
3336 static void mem_cgroup_oom_notify(struct mem_cgroup
*memcg
)
3338 struct mem_cgroup
*iter
;
3340 for_each_mem_cgroup_tree(iter
, memcg
)
3341 mem_cgroup_oom_notify_cb(iter
);
3344 static int __mem_cgroup_usage_register_event(struct mem_cgroup
*memcg
,
3345 struct eventfd_ctx
*eventfd
, const char *args
, enum res_type type
)
3347 struct mem_cgroup_thresholds
*thresholds
;
3348 struct mem_cgroup_threshold_ary
*new;
3349 unsigned long threshold
;
3350 unsigned long usage
;
3353 ret
= page_counter_memparse(args
, "-1", &threshold
);
3356 threshold
<<= PAGE_SHIFT
;
3358 mutex_lock(&memcg
->thresholds_lock
);
3361 thresholds
= &memcg
->thresholds
;
3362 usage
= mem_cgroup_usage(memcg
, false);
3363 } else if (type
== _MEMSWAP
) {
3364 thresholds
= &memcg
->memsw_thresholds
;
3365 usage
= mem_cgroup_usage(memcg
, true);
3369 /* Check if a threshold crossed before adding a new one */
3370 if (thresholds
->primary
)
3371 __mem_cgroup_threshold(memcg
, type
== _MEMSWAP
);
3373 size
= thresholds
->primary
? thresholds
->primary
->size
+ 1 : 1;
3375 /* Allocate memory for new array of thresholds */
3376 new = kmalloc(sizeof(*new) + size
* sizeof(struct mem_cgroup_threshold
),
3384 /* Copy thresholds (if any) to new array */
3385 if (thresholds
->primary
) {
3386 memcpy(new->entries
, thresholds
->primary
->entries
, (size
- 1) *
3387 sizeof(struct mem_cgroup_threshold
));
3390 /* Add new threshold */
3391 new->entries
[size
- 1].eventfd
= eventfd
;
3392 new->entries
[size
- 1].threshold
= threshold
;
3394 /* Sort thresholds. Registering of new threshold isn't time-critical */
3395 sort(new->entries
, size
, sizeof(struct mem_cgroup_threshold
),
3396 compare_thresholds
, NULL
);
3398 /* Find current threshold */
3399 new->current_threshold
= -1;
3400 for (i
= 0; i
< size
; i
++) {
3401 if (new->entries
[i
].threshold
<= usage
) {
3403 * new->current_threshold will not be used until
3404 * rcu_assign_pointer(), so it's safe to increment
3407 ++new->current_threshold
;
3412 /* Free old spare buffer and save old primary buffer as spare */
3413 kfree(thresholds
->spare
);
3414 thresholds
->spare
= thresholds
->primary
;
3416 rcu_assign_pointer(thresholds
->primary
, new);
3418 /* To be sure that nobody uses thresholds */
3422 mutex_unlock(&memcg
->thresholds_lock
);
3427 static int mem_cgroup_usage_register_event(struct mem_cgroup
*memcg
,
3428 struct eventfd_ctx
*eventfd
, const char *args
)
3430 return __mem_cgroup_usage_register_event(memcg
, eventfd
, args
, _MEM
);
3433 static int memsw_cgroup_usage_register_event(struct mem_cgroup
*memcg
,
3434 struct eventfd_ctx
*eventfd
, const char *args
)
3436 return __mem_cgroup_usage_register_event(memcg
, eventfd
, args
, _MEMSWAP
);
3439 static void __mem_cgroup_usage_unregister_event(struct mem_cgroup
*memcg
,
3440 struct eventfd_ctx
*eventfd
, enum res_type type
)
3442 struct mem_cgroup_thresholds
*thresholds
;
3443 struct mem_cgroup_threshold_ary
*new;
3444 unsigned long usage
;
3447 mutex_lock(&memcg
->thresholds_lock
);
3450 thresholds
= &memcg
->thresholds
;
3451 usage
= mem_cgroup_usage(memcg
, false);
3452 } else if (type
== _MEMSWAP
) {
3453 thresholds
= &memcg
->memsw_thresholds
;
3454 usage
= mem_cgroup_usage(memcg
, true);
3458 if (!thresholds
->primary
)
3461 /* Check if a threshold crossed before removing */
3462 __mem_cgroup_threshold(memcg
, type
== _MEMSWAP
);
3464 /* Calculate new number of threshold */
3466 for (i
= 0; i
< thresholds
->primary
->size
; i
++) {
3467 if (thresholds
->primary
->entries
[i
].eventfd
!= eventfd
)
3471 new = thresholds
->spare
;
3473 /* Set thresholds array to NULL if we don't have thresholds */
3482 /* Copy thresholds and find current threshold */
3483 new->current_threshold
= -1;
3484 for (i
= 0, j
= 0; i
< thresholds
->primary
->size
; i
++) {
3485 if (thresholds
->primary
->entries
[i
].eventfd
== eventfd
)
3488 new->entries
[j
] = thresholds
->primary
->entries
[i
];
3489 if (new->entries
[j
].threshold
<= usage
) {
3491 * new->current_threshold will not be used
3492 * until rcu_assign_pointer(), so it's safe to increment
3495 ++new->current_threshold
;
3501 /* Swap primary and spare array */
3502 thresholds
->spare
= thresholds
->primary
;
3503 /* If all events are unregistered, free the spare array */
3505 kfree(thresholds
->spare
);
3506 thresholds
->spare
= NULL
;
3509 rcu_assign_pointer(thresholds
->primary
, new);
3511 /* To be sure that nobody uses thresholds */
3514 mutex_unlock(&memcg
->thresholds_lock
);
3517 static void mem_cgroup_usage_unregister_event(struct mem_cgroup
*memcg
,
3518 struct eventfd_ctx
*eventfd
)
3520 return __mem_cgroup_usage_unregister_event(memcg
, eventfd
, _MEM
);
3523 static void memsw_cgroup_usage_unregister_event(struct mem_cgroup
*memcg
,
3524 struct eventfd_ctx
*eventfd
)
3526 return __mem_cgroup_usage_unregister_event(memcg
, eventfd
, _MEMSWAP
);
3529 static int mem_cgroup_oom_register_event(struct mem_cgroup
*memcg
,
3530 struct eventfd_ctx
*eventfd
, const char *args
)
3532 struct mem_cgroup_eventfd_list
*event
;
3534 event
= kmalloc(sizeof(*event
), GFP_KERNEL
);
3538 spin_lock(&memcg_oom_lock
);
3540 event
->eventfd
= eventfd
;
3541 list_add(&event
->list
, &memcg
->oom_notify
);
3543 /* already in OOM ? */
3544 if (memcg
->under_oom
)
3545 eventfd_signal(eventfd
, 1);
3546 spin_unlock(&memcg_oom_lock
);
3551 static void mem_cgroup_oom_unregister_event(struct mem_cgroup
*memcg
,
3552 struct eventfd_ctx
*eventfd
)
3554 struct mem_cgroup_eventfd_list
*ev
, *tmp
;
3556 spin_lock(&memcg_oom_lock
);
3558 list_for_each_entry_safe(ev
, tmp
, &memcg
->oom_notify
, list
) {
3559 if (ev
->eventfd
== eventfd
) {
3560 list_del(&ev
->list
);
3565 spin_unlock(&memcg_oom_lock
);
3568 static int mem_cgroup_oom_control_read(struct seq_file
*sf
, void *v
)
3570 struct mem_cgroup
*memcg
= mem_cgroup_from_css(seq_css(sf
));
3572 seq_printf(sf
, "oom_kill_disable %d\n", memcg
->oom_kill_disable
);
3573 seq_printf(sf
, "under_oom %d\n", (bool)memcg
->under_oom
);
3577 static int mem_cgroup_oom_control_write(struct cgroup_subsys_state
*css
,
3578 struct cftype
*cft
, u64 val
)
3580 struct mem_cgroup
*memcg
= mem_cgroup_from_css(css
);
3582 /* cannot set to root cgroup and only 0 and 1 are allowed */
3583 if (!css
->parent
|| !((val
== 0) || (val
== 1)))
3586 memcg
->oom_kill_disable
= val
;
3588 memcg_oom_recover(memcg
);
3593 #ifdef CONFIG_MEMCG_KMEM
3594 static int memcg_init_kmem(struct mem_cgroup
*memcg
, struct cgroup_subsys
*ss
)
3598 ret
= memcg_propagate_kmem(memcg
);
3602 return mem_cgroup_sockets_init(memcg
, ss
);
3605 static void memcg_deactivate_kmem(struct mem_cgroup
*memcg
)
3607 struct cgroup_subsys_state
*css
;
3608 struct mem_cgroup
*parent
, *child
;
3611 if (!memcg
->kmem_acct_active
)
3615 * Clear the 'active' flag before clearing memcg_caches arrays entries.
3616 * Since we take the slab_mutex in memcg_deactivate_kmem_caches(), it
3617 * guarantees no cache will be created for this cgroup after we are
3618 * done (see memcg_create_kmem_cache()).
3620 memcg
->kmem_acct_active
= false;
3622 memcg_deactivate_kmem_caches(memcg
);
3624 kmemcg_id
= memcg
->kmemcg_id
;
3625 BUG_ON(kmemcg_id
< 0);
3627 parent
= parent_mem_cgroup(memcg
);
3629 parent
= root_mem_cgroup
;
3632 * Change kmemcg_id of this cgroup and all its descendants to the
3633 * parent's id, and then move all entries from this cgroup's list_lrus
3634 * to ones of the parent. After we have finished, all list_lrus
3635 * corresponding to this cgroup are guaranteed to remain empty. The
3636 * ordering is imposed by list_lru_node->lock taken by
3637 * memcg_drain_all_list_lrus().
3639 css_for_each_descendant_pre(css
, &memcg
->css
) {
3640 child
= mem_cgroup_from_css(css
);
3641 BUG_ON(child
->kmemcg_id
!= kmemcg_id
);
3642 child
->kmemcg_id
= parent
->kmemcg_id
;
3643 if (!memcg
->use_hierarchy
)
3646 memcg_drain_all_list_lrus(kmemcg_id
, parent
->kmemcg_id
);
3648 memcg_free_cache_id(kmemcg_id
);
3651 static void memcg_destroy_kmem(struct mem_cgroup
*memcg
)
3653 if (memcg
->kmem_acct_activated
) {
3654 memcg_destroy_kmem_caches(memcg
);
3655 static_key_slow_dec(&memcg_kmem_enabled_key
);
3656 WARN_ON(page_counter_read(&memcg
->kmem
));
3658 mem_cgroup_sockets_destroy(memcg
);
3661 static int memcg_init_kmem(struct mem_cgroup
*memcg
, struct cgroup_subsys
*ss
)
3666 static void memcg_deactivate_kmem(struct mem_cgroup
*memcg
)
3670 static void memcg_destroy_kmem(struct mem_cgroup
*memcg
)
3675 #ifdef CONFIG_CGROUP_WRITEBACK
3677 struct list_head
*mem_cgroup_cgwb_list(struct mem_cgroup
*memcg
)
3679 return &memcg
->cgwb_list
;
3682 static int memcg_wb_domain_init(struct mem_cgroup
*memcg
, gfp_t gfp
)
3684 return wb_domain_init(&memcg
->cgwb_domain
, gfp
);
3687 static void memcg_wb_domain_exit(struct mem_cgroup
*memcg
)
3689 wb_domain_exit(&memcg
->cgwb_domain
);
3692 static void memcg_wb_domain_size_changed(struct mem_cgroup
*memcg
)
3694 wb_domain_size_changed(&memcg
->cgwb_domain
);
3697 struct wb_domain
*mem_cgroup_wb_domain(struct bdi_writeback
*wb
)
3699 struct mem_cgroup
*memcg
= mem_cgroup_from_css(wb
->memcg_css
);
3701 if (!memcg
->css
.parent
)
3704 return &memcg
->cgwb_domain
;
3708 * mem_cgroup_wb_stats - retrieve writeback related stats from its memcg
3709 * @wb: bdi_writeback in question
3710 * @pfilepages: out parameter for number of file pages
3711 * @pheadroom: out parameter for number of allocatable pages according to memcg
3712 * @pdirty: out parameter for number of dirty pages
3713 * @pwriteback: out parameter for number of pages under writeback
3715 * Determine the numbers of file, headroom, dirty, and writeback pages in
3716 * @wb's memcg. File, dirty and writeback are self-explanatory. Headroom
3717 * is a bit more involved.
3719 * A memcg's headroom is "min(max, high) - used". In the hierarchy, the
3720 * headroom is calculated as the lowest headroom of itself and the
3721 * ancestors. Note that this doesn't consider the actual amount of
3722 * available memory in the system. The caller should further cap
3723 * *@pheadroom accordingly.
3725 void mem_cgroup_wb_stats(struct bdi_writeback
*wb
, unsigned long *pfilepages
,
3726 unsigned long *pheadroom
, unsigned long *pdirty
,
3727 unsigned long *pwriteback
)
3729 struct mem_cgroup
*memcg
= mem_cgroup_from_css(wb
->memcg_css
);
3730 struct mem_cgroup
*parent
;
3732 *pdirty
= mem_cgroup_read_stat(memcg
, MEM_CGROUP_STAT_DIRTY
);
3734 /* this should eventually include NR_UNSTABLE_NFS */
3735 *pwriteback
= mem_cgroup_read_stat(memcg
, MEM_CGROUP_STAT_WRITEBACK
);
3736 *pfilepages
= mem_cgroup_nr_lru_pages(memcg
, (1 << LRU_INACTIVE_FILE
) |
3737 (1 << LRU_ACTIVE_FILE
));
3738 *pheadroom
= PAGE_COUNTER_MAX
;
3740 while ((parent
= parent_mem_cgroup(memcg
))) {
3741 unsigned long ceiling
= min(memcg
->memory
.limit
, memcg
->high
);
3742 unsigned long used
= page_counter_read(&memcg
->memory
);
3744 *pheadroom
= min(*pheadroom
, ceiling
- min(ceiling
, used
));
3749 #else /* CONFIG_CGROUP_WRITEBACK */
3751 static int memcg_wb_domain_init(struct mem_cgroup
*memcg
, gfp_t gfp
)
3756 static void memcg_wb_domain_exit(struct mem_cgroup
*memcg
)
3760 static void memcg_wb_domain_size_changed(struct mem_cgroup
*memcg
)
3764 #endif /* CONFIG_CGROUP_WRITEBACK */
3767 * DO NOT USE IN NEW FILES.
3769 * "cgroup.event_control" implementation.
3771 * This is way over-engineered. It tries to support fully configurable
3772 * events for each user. Such level of flexibility is completely
3773 * unnecessary especially in the light of the planned unified hierarchy.
3775 * Please deprecate this and replace with something simpler if at all
3780 * Unregister event and free resources.
3782 * Gets called from workqueue.
3784 static void memcg_event_remove(struct work_struct
*work
)
3786 struct mem_cgroup_event
*event
=
3787 container_of(work
, struct mem_cgroup_event
, remove
);
3788 struct mem_cgroup
*memcg
= event
->memcg
;
3790 remove_wait_queue(event
->wqh
, &event
->wait
);
3792 event
->unregister_event(memcg
, event
->eventfd
);
3794 /* Notify userspace the event is going away. */
3795 eventfd_signal(event
->eventfd
, 1);
3797 eventfd_ctx_put(event
->eventfd
);
3799 css_put(&memcg
->css
);
3803 * Gets called on POLLHUP on eventfd when user closes it.
3805 * Called with wqh->lock held and interrupts disabled.
3807 static int memcg_event_wake(wait_queue_t
*wait
, unsigned mode
,
3808 int sync
, void *key
)
3810 struct mem_cgroup_event
*event
=
3811 container_of(wait
, struct mem_cgroup_event
, wait
);
3812 struct mem_cgroup
*memcg
= event
->memcg
;
3813 unsigned long flags
= (unsigned long)key
;
3815 if (flags
& POLLHUP
) {
3817 * If the event has been detached at cgroup removal, we
3818 * can simply return knowing the other side will cleanup
3821 * We can't race against event freeing since the other
3822 * side will require wqh->lock via remove_wait_queue(),
3825 spin_lock(&memcg
->event_list_lock
);
3826 if (!list_empty(&event
->list
)) {
3827 list_del_init(&event
->list
);
3829 * We are in atomic context, but cgroup_event_remove()
3830 * may sleep, so we have to call it in workqueue.
3832 schedule_work(&event
->remove
);
3834 spin_unlock(&memcg
->event_list_lock
);
3840 static void memcg_event_ptable_queue_proc(struct file
*file
,
3841 wait_queue_head_t
*wqh
, poll_table
*pt
)
3843 struct mem_cgroup_event
*event
=
3844 container_of(pt
, struct mem_cgroup_event
, pt
);
3847 add_wait_queue(wqh
, &event
->wait
);
3851 * DO NOT USE IN NEW FILES.
3853 * Parse input and register new cgroup event handler.
3855 * Input must be in format '<event_fd> <control_fd> <args>'.
3856 * Interpretation of args is defined by control file implementation.
3858 static ssize_t
memcg_write_event_control(struct kernfs_open_file
*of
,
3859 char *buf
, size_t nbytes
, loff_t off
)
3861 struct cgroup_subsys_state
*css
= of_css(of
);
3862 struct mem_cgroup
*memcg
= mem_cgroup_from_css(css
);
3863 struct mem_cgroup_event
*event
;
3864 struct cgroup_subsys_state
*cfile_css
;
3865 unsigned int efd
, cfd
;
3872 buf
= strstrip(buf
);
3874 efd
= simple_strtoul(buf
, &endp
, 10);
3879 cfd
= simple_strtoul(buf
, &endp
, 10);
3880 if ((*endp
!= ' ') && (*endp
!= '\0'))
3884 event
= kzalloc(sizeof(*event
), GFP_KERNEL
);
3888 event
->memcg
= memcg
;
3889 INIT_LIST_HEAD(&event
->list
);
3890 init_poll_funcptr(&event
->pt
, memcg_event_ptable_queue_proc
);
3891 init_waitqueue_func_entry(&event
->wait
, memcg_event_wake
);
3892 INIT_WORK(&event
->remove
, memcg_event_remove
);
3900 event
->eventfd
= eventfd_ctx_fileget(efile
.file
);
3901 if (IS_ERR(event
->eventfd
)) {
3902 ret
= PTR_ERR(event
->eventfd
);
3909 goto out_put_eventfd
;
3912 /* the process need read permission on control file */
3913 /* AV: shouldn't we check that it's been opened for read instead? */
3914 ret
= inode_permission(file_inode(cfile
.file
), MAY_READ
);
3919 * Determine the event callbacks and set them in @event. This used
3920 * to be done via struct cftype but cgroup core no longer knows
3921 * about these events. The following is crude but the whole thing
3922 * is for compatibility anyway.
3924 * DO NOT ADD NEW FILES.
3926 name
= cfile
.file
->f_path
.dentry
->d_name
.name
;
3928 if (!strcmp(name
, "memory.usage_in_bytes")) {
3929 event
->register_event
= mem_cgroup_usage_register_event
;
3930 event
->unregister_event
= mem_cgroup_usage_unregister_event
;
3931 } else if (!strcmp(name
, "memory.oom_control")) {
3932 event
->register_event
= mem_cgroup_oom_register_event
;
3933 event
->unregister_event
= mem_cgroup_oom_unregister_event
;
3934 } else if (!strcmp(name
, "memory.pressure_level")) {
3935 event
->register_event
= vmpressure_register_event
;
3936 event
->unregister_event
= vmpressure_unregister_event
;
3937 } else if (!strcmp(name
, "memory.memsw.usage_in_bytes")) {
3938 event
->register_event
= memsw_cgroup_usage_register_event
;
3939 event
->unregister_event
= memsw_cgroup_usage_unregister_event
;
3946 * Verify @cfile should belong to @css. Also, remaining events are
3947 * automatically removed on cgroup destruction but the removal is
3948 * asynchronous, so take an extra ref on @css.
3950 cfile_css
= css_tryget_online_from_dir(cfile
.file
->f_path
.dentry
->d_parent
,
3951 &memory_cgrp_subsys
);
3953 if (IS_ERR(cfile_css
))
3955 if (cfile_css
!= css
) {
3960 ret
= event
->register_event(memcg
, event
->eventfd
, buf
);
3964 efile
.file
->f_op
->poll(efile
.file
, &event
->pt
);
3966 spin_lock(&memcg
->event_list_lock
);
3967 list_add(&event
->list
, &memcg
->event_list
);
3968 spin_unlock(&memcg
->event_list_lock
);
3980 eventfd_ctx_put(event
->eventfd
);
3989 static struct cftype mem_cgroup_legacy_files
[] = {
3991 .name
= "usage_in_bytes",
3992 .private = MEMFILE_PRIVATE(_MEM
, RES_USAGE
),
3993 .read_u64
= mem_cgroup_read_u64
,
3996 .name
= "max_usage_in_bytes",
3997 .private = MEMFILE_PRIVATE(_MEM
, RES_MAX_USAGE
),
3998 .write
= mem_cgroup_reset
,
3999 .read_u64
= mem_cgroup_read_u64
,
4002 .name
= "limit_in_bytes",
4003 .private = MEMFILE_PRIVATE(_MEM
, RES_LIMIT
),
4004 .write
= mem_cgroup_write
,
4005 .read_u64
= mem_cgroup_read_u64
,
4008 .name
= "soft_limit_in_bytes",
4009 .private = MEMFILE_PRIVATE(_MEM
, RES_SOFT_LIMIT
),
4010 .write
= mem_cgroup_write
,
4011 .read_u64
= mem_cgroup_read_u64
,
4015 .private = MEMFILE_PRIVATE(_MEM
, RES_FAILCNT
),
4016 .write
= mem_cgroup_reset
,
4017 .read_u64
= mem_cgroup_read_u64
,
4021 .seq_show
= memcg_stat_show
,
4024 .name
= "force_empty",
4025 .write
= mem_cgroup_force_empty_write
,
4028 .name
= "use_hierarchy",
4029 .write_u64
= mem_cgroup_hierarchy_write
,
4030 .read_u64
= mem_cgroup_hierarchy_read
,
4033 .name
= "cgroup.event_control", /* XXX: for compat */
4034 .write
= memcg_write_event_control
,
4035 .flags
= CFTYPE_NO_PREFIX
,
4039 .name
= "swappiness",
4040 .read_u64
= mem_cgroup_swappiness_read
,
4041 .write_u64
= mem_cgroup_swappiness_write
,
4044 .name
= "move_charge_at_immigrate",
4045 .read_u64
= mem_cgroup_move_charge_read
,
4046 .write_u64
= mem_cgroup_move_charge_write
,
4049 .name
= "oom_control",
4050 .seq_show
= mem_cgroup_oom_control_read
,
4051 .write_u64
= mem_cgroup_oom_control_write
,
4052 .private = MEMFILE_PRIVATE(_OOM_TYPE
, OOM_CONTROL
),
4055 .name
= "pressure_level",
4059 .name
= "numa_stat",
4060 .seq_show
= memcg_numa_stat_show
,
4063 #ifdef CONFIG_MEMCG_KMEM
4065 .name
= "kmem.limit_in_bytes",
4066 .private = MEMFILE_PRIVATE(_KMEM
, RES_LIMIT
),
4067 .write
= mem_cgroup_write
,
4068 .read_u64
= mem_cgroup_read_u64
,
4071 .name
= "kmem.usage_in_bytes",
4072 .private = MEMFILE_PRIVATE(_KMEM
, RES_USAGE
),
4073 .read_u64
= mem_cgroup_read_u64
,
4076 .name
= "kmem.failcnt",
4077 .private = MEMFILE_PRIVATE(_KMEM
, RES_FAILCNT
),
4078 .write
= mem_cgroup_reset
,
4079 .read_u64
= mem_cgroup_read_u64
,
4082 .name
= "kmem.max_usage_in_bytes",
4083 .private = MEMFILE_PRIVATE(_KMEM
, RES_MAX_USAGE
),
4084 .write
= mem_cgroup_reset
,
4085 .read_u64
= mem_cgroup_read_u64
,
4087 #ifdef CONFIG_SLABINFO
4089 .name
= "kmem.slabinfo",
4090 .seq_start
= slab_start
,
4091 .seq_next
= slab_next
,
4092 .seq_stop
= slab_stop
,
4093 .seq_show
= memcg_slab_show
,
4097 { }, /* terminate */
4100 static int alloc_mem_cgroup_per_zone_info(struct mem_cgroup
*memcg
, int node
)
4102 struct mem_cgroup_per_node
*pn
;
4103 struct mem_cgroup_per_zone
*mz
;
4104 int zone
, tmp
= node
;
4106 * This routine is called against possible nodes.
4107 * But it's BUG to call kmalloc() against offline node.
4109 * TODO: this routine can waste much memory for nodes which will
4110 * never be onlined. It's better to use memory hotplug callback
4113 if (!node_state(node
, N_NORMAL_MEMORY
))
4115 pn
= kzalloc_node(sizeof(*pn
), GFP_KERNEL
, tmp
);
4119 for (zone
= 0; zone
< MAX_NR_ZONES
; zone
++) {
4120 mz
= &pn
->zoneinfo
[zone
];
4121 lruvec_init(&mz
->lruvec
);
4122 mz
->usage_in_excess
= 0;
4123 mz
->on_tree
= false;
4126 memcg
->nodeinfo
[node
] = pn
;
4130 static void free_mem_cgroup_per_zone_info(struct mem_cgroup
*memcg
, int node
)
4132 kfree(memcg
->nodeinfo
[node
]);
4135 static struct mem_cgroup
*mem_cgroup_alloc(void)
4137 struct mem_cgroup
*memcg
;
4140 size
= sizeof(struct mem_cgroup
);
4141 size
+= nr_node_ids
* sizeof(struct mem_cgroup_per_node
*);
4143 memcg
= kzalloc(size
, GFP_KERNEL
);
4147 memcg
->stat
= alloc_percpu(struct mem_cgroup_stat_cpu
);
4151 if (memcg_wb_domain_init(memcg
, GFP_KERNEL
))
4157 free_percpu(memcg
->stat
);
4164 * At destroying mem_cgroup, references from swap_cgroup can remain.
4165 * (scanning all at force_empty is too costly...)
4167 * Instead of clearing all references at force_empty, we remember
4168 * the number of reference from swap_cgroup and free mem_cgroup when
4169 * it goes down to 0.
4171 * Removal of cgroup itself succeeds regardless of refs from swap.
4174 static void __mem_cgroup_free(struct mem_cgroup
*memcg
)
4178 mem_cgroup_remove_from_trees(memcg
);
4181 free_mem_cgroup_per_zone_info(memcg
, node
);
4183 free_percpu(memcg
->stat
);
4184 memcg_wb_domain_exit(memcg
);
4189 * Returns the parent mem_cgroup in memcgroup hierarchy with hierarchy enabled.
4191 struct mem_cgroup
*parent_mem_cgroup(struct mem_cgroup
*memcg
)
4193 if (!memcg
->memory
.parent
)
4195 return mem_cgroup_from_counter(memcg
->memory
.parent
, memory
);
4197 EXPORT_SYMBOL(parent_mem_cgroup
);
4199 static struct cgroup_subsys_state
* __ref
4200 mem_cgroup_css_alloc(struct cgroup_subsys_state
*parent_css
)
4202 struct mem_cgroup
*memcg
;
4203 long error
= -ENOMEM
;
4206 memcg
= mem_cgroup_alloc();
4208 return ERR_PTR(error
);
4211 if (alloc_mem_cgroup_per_zone_info(memcg
, node
))
4215 if (parent_css
== NULL
) {
4216 root_mem_cgroup
= memcg
;
4217 mem_cgroup_root_css
= &memcg
->css
;
4218 page_counter_init(&memcg
->memory
, NULL
);
4219 memcg
->high
= PAGE_COUNTER_MAX
;
4220 memcg
->soft_limit
= PAGE_COUNTER_MAX
;
4221 page_counter_init(&memcg
->memsw
, NULL
);
4222 page_counter_init(&memcg
->kmem
, NULL
);
4225 memcg
->last_scanned_node
= MAX_NUMNODES
;
4226 INIT_LIST_HEAD(&memcg
->oom_notify
);
4227 memcg
->move_charge_at_immigrate
= 0;
4228 mutex_init(&memcg
->thresholds_lock
);
4229 spin_lock_init(&memcg
->move_lock
);
4230 vmpressure_init(&memcg
->vmpressure
);
4231 INIT_LIST_HEAD(&memcg
->event_list
);
4232 spin_lock_init(&memcg
->event_list_lock
);
4233 #ifdef CONFIG_MEMCG_KMEM
4234 memcg
->kmemcg_id
= -1;
4236 #ifdef CONFIG_CGROUP_WRITEBACK
4237 INIT_LIST_HEAD(&memcg
->cgwb_list
);
4242 __mem_cgroup_free(memcg
);
4243 return ERR_PTR(error
);
4247 mem_cgroup_css_online(struct cgroup_subsys_state
*css
)
4249 struct mem_cgroup
*memcg
= mem_cgroup_from_css(css
);
4250 struct mem_cgroup
*parent
= mem_cgroup_from_css(css
->parent
);
4253 if (css
->id
> MEM_CGROUP_ID_MAX
)
4259 mutex_lock(&memcg_create_mutex
);
4261 memcg
->use_hierarchy
= parent
->use_hierarchy
;
4262 memcg
->oom_kill_disable
= parent
->oom_kill_disable
;
4263 memcg
->swappiness
= mem_cgroup_swappiness(parent
);
4265 if (parent
->use_hierarchy
) {
4266 page_counter_init(&memcg
->memory
, &parent
->memory
);
4267 memcg
->high
= PAGE_COUNTER_MAX
;
4268 memcg
->soft_limit
= PAGE_COUNTER_MAX
;
4269 page_counter_init(&memcg
->memsw
, &parent
->memsw
);
4270 page_counter_init(&memcg
->kmem
, &parent
->kmem
);
4273 * No need to take a reference to the parent because cgroup
4274 * core guarantees its existence.
4277 page_counter_init(&memcg
->memory
, NULL
);
4278 memcg
->high
= PAGE_COUNTER_MAX
;
4279 memcg
->soft_limit
= PAGE_COUNTER_MAX
;
4280 page_counter_init(&memcg
->memsw
, NULL
);
4281 page_counter_init(&memcg
->kmem
, NULL
);
4283 * Deeper hierachy with use_hierarchy == false doesn't make
4284 * much sense so let cgroup subsystem know about this
4285 * unfortunate state in our controller.
4287 if (parent
!= root_mem_cgroup
)
4288 memory_cgrp_subsys
.broken_hierarchy
= true;
4290 mutex_unlock(&memcg_create_mutex
);
4292 ret
= memcg_init_kmem(memcg
, &memory_cgrp_subsys
);
4297 * Make sure the memcg is initialized: mem_cgroup_iter()
4298 * orders reading memcg->initialized against its callers
4299 * reading the memcg members.
4301 smp_store_release(&memcg
->initialized
, 1);
4306 static void mem_cgroup_css_offline(struct cgroup_subsys_state
*css
)
4308 struct mem_cgroup
*memcg
= mem_cgroup_from_css(css
);
4309 struct mem_cgroup_event
*event
, *tmp
;
4312 * Unregister events and notify userspace.
4313 * Notify userspace about cgroup removing only after rmdir of cgroup
4314 * directory to avoid race between userspace and kernelspace.
4316 spin_lock(&memcg
->event_list_lock
);
4317 list_for_each_entry_safe(event
, tmp
, &memcg
->event_list
, list
) {
4318 list_del_init(&event
->list
);
4319 schedule_work(&event
->remove
);
4321 spin_unlock(&memcg
->event_list_lock
);
4323 vmpressure_cleanup(&memcg
->vmpressure
);
4325 memcg_deactivate_kmem(memcg
);
4327 wb_memcg_offline(memcg
);
4330 static void mem_cgroup_css_free(struct cgroup_subsys_state
*css
)
4332 struct mem_cgroup
*memcg
= mem_cgroup_from_css(css
);
4334 memcg_destroy_kmem(memcg
);
4335 __mem_cgroup_free(memcg
);
4339 * mem_cgroup_css_reset - reset the states of a mem_cgroup
4340 * @css: the target css
4342 * Reset the states of the mem_cgroup associated with @css. This is
4343 * invoked when the userland requests disabling on the default hierarchy
4344 * but the memcg is pinned through dependency. The memcg should stop
4345 * applying policies and should revert to the vanilla state as it may be
4346 * made visible again.
4348 * The current implementation only resets the essential configurations.
4349 * This needs to be expanded to cover all the visible parts.
4351 static void mem_cgroup_css_reset(struct cgroup_subsys_state
*css
)
4353 struct mem_cgroup
*memcg
= mem_cgroup_from_css(css
);
4355 mem_cgroup_resize_limit(memcg
, PAGE_COUNTER_MAX
);
4356 mem_cgroup_resize_memsw_limit(memcg
, PAGE_COUNTER_MAX
);
4357 memcg_update_kmem_limit(memcg
, PAGE_COUNTER_MAX
);
4359 memcg
->high
= PAGE_COUNTER_MAX
;
4360 memcg
->soft_limit
= PAGE_COUNTER_MAX
;
4361 memcg_wb_domain_size_changed(memcg
);
4365 /* Handlers for move charge at task migration. */
4366 static int mem_cgroup_do_precharge(unsigned long count
)
4370 /* Try a single bulk charge without reclaim first */
4371 ret
= try_charge(mc
.to
, GFP_KERNEL
& ~__GFP_WAIT
, count
);
4373 mc
.precharge
+= count
;
4377 /* Try charges one by one with reclaim */
4379 ret
= try_charge(mc
.to
, GFP_KERNEL
& ~__GFP_NORETRY
, 1);
4389 * get_mctgt_type - get target type of moving charge
4390 * @vma: the vma the pte to be checked belongs
4391 * @addr: the address corresponding to the pte to be checked
4392 * @ptent: the pte to be checked
4393 * @target: the pointer the target page or swap ent will be stored(can be NULL)
4396 * 0(MC_TARGET_NONE): if the pte is not a target for move charge.
4397 * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
4398 * move charge. if @target is not NULL, the page is stored in target->page
4399 * with extra refcnt got(Callers should handle it).
4400 * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
4401 * target for charge migration. if @target is not NULL, the entry is stored
4404 * Called with pte lock held.
4411 enum mc_target_type
{
4417 static struct page
*mc_handle_present_pte(struct vm_area_struct
*vma
,
4418 unsigned long addr
, pte_t ptent
)
4420 struct page
*page
= vm_normal_page(vma
, addr
, ptent
);
4422 if (!page
|| !page_mapped(page
))
4424 if (PageAnon(page
)) {
4425 if (!(mc
.flags
& MOVE_ANON
))
4428 if (!(mc
.flags
& MOVE_FILE
))
4431 if (!get_page_unless_zero(page
))
4438 static struct page
*mc_handle_swap_pte(struct vm_area_struct
*vma
,
4439 unsigned long addr
, pte_t ptent
, swp_entry_t
*entry
)
4441 struct page
*page
= NULL
;
4442 swp_entry_t ent
= pte_to_swp_entry(ptent
);
4444 if (!(mc
.flags
& MOVE_ANON
) || non_swap_entry(ent
))
4447 * Because lookup_swap_cache() updates some statistics counter,
4448 * we call find_get_page() with swapper_space directly.
4450 page
= find_get_page(swap_address_space(ent
), ent
.val
);
4451 if (do_swap_account
)
4452 entry
->val
= ent
.val
;
4457 static struct page
*mc_handle_swap_pte(struct vm_area_struct
*vma
,
4458 unsigned long addr
, pte_t ptent
, swp_entry_t
*entry
)
4464 static struct page
*mc_handle_file_pte(struct vm_area_struct
*vma
,
4465 unsigned long addr
, pte_t ptent
, swp_entry_t
*entry
)
4467 struct page
*page
= NULL
;
4468 struct address_space
*mapping
;
4471 if (!vma
->vm_file
) /* anonymous vma */
4473 if (!(mc
.flags
& MOVE_FILE
))
4476 mapping
= vma
->vm_file
->f_mapping
;
4477 pgoff
= linear_page_index(vma
, addr
);
4479 /* page is moved even if it's not RSS of this task(page-faulted). */
4481 /* shmem/tmpfs may report page out on swap: account for that too. */
4482 if (shmem_mapping(mapping
)) {
4483 page
= find_get_entry(mapping
, pgoff
);
4484 if (radix_tree_exceptional_entry(page
)) {
4485 swp_entry_t swp
= radix_to_swp_entry(page
);
4486 if (do_swap_account
)
4488 page
= find_get_page(swap_address_space(swp
), swp
.val
);
4491 page
= find_get_page(mapping
, pgoff
);
4493 page
= find_get_page(mapping
, pgoff
);
4499 * mem_cgroup_move_account - move account of the page
4501 * @nr_pages: number of regular pages (>1 for huge pages)
4502 * @from: mem_cgroup which the page is moved from.
4503 * @to: mem_cgroup which the page is moved to. @from != @to.
4505 * The caller must confirm following.
4506 * - page is not on LRU (isolate_page() is useful.)
4507 * - compound_lock is held when nr_pages > 1
4509 * This function doesn't do "charge" to new cgroup and doesn't do "uncharge"
4512 static int mem_cgroup_move_account(struct page
*page
,
4513 unsigned int nr_pages
,
4514 struct mem_cgroup
*from
,
4515 struct mem_cgroup
*to
)
4517 unsigned long flags
;
4521 VM_BUG_ON(from
== to
);
4522 VM_BUG_ON_PAGE(PageLRU(page
), page
);
4524 * The page is isolated from LRU. So, collapse function
4525 * will not handle this page. But page splitting can happen.
4526 * Do this check under compound_page_lock(). The caller should
4530 if (nr_pages
> 1 && !PageTransHuge(page
))
4534 * Prevent mem_cgroup_replace_page() from looking at
4535 * page->mem_cgroup of its source page while we change it.
4537 if (!trylock_page(page
))
4541 if (page
->mem_cgroup
!= from
)
4544 anon
= PageAnon(page
);
4546 spin_lock_irqsave(&from
->move_lock
, flags
);
4548 if (!anon
&& page_mapped(page
)) {
4549 __this_cpu_sub(from
->stat
->count
[MEM_CGROUP_STAT_FILE_MAPPED
],
4551 __this_cpu_add(to
->stat
->count
[MEM_CGROUP_STAT_FILE_MAPPED
],
4556 * move_lock grabbed above and caller set from->moving_account, so
4557 * mem_cgroup_update_page_stat() will serialize updates to PageDirty.
4558 * So mapping should be stable for dirty pages.
4560 if (!anon
&& PageDirty(page
)) {
4561 struct address_space
*mapping
= page_mapping(page
);
4563 if (mapping_cap_account_dirty(mapping
)) {
4564 __this_cpu_sub(from
->stat
->count
[MEM_CGROUP_STAT_DIRTY
],
4566 __this_cpu_add(to
->stat
->count
[MEM_CGROUP_STAT_DIRTY
],
4571 if (PageWriteback(page
)) {
4572 __this_cpu_sub(from
->stat
->count
[MEM_CGROUP_STAT_WRITEBACK
],
4574 __this_cpu_add(to
->stat
->count
[MEM_CGROUP_STAT_WRITEBACK
],
4579 * It is safe to change page->mem_cgroup here because the page
4580 * is referenced, charged, and isolated - we can't race with
4581 * uncharging, charging, migration, or LRU putback.
4584 /* caller should have done css_get */
4585 page
->mem_cgroup
= to
;
4586 spin_unlock_irqrestore(&from
->move_lock
, flags
);
4590 local_irq_disable();
4591 mem_cgroup_charge_statistics(to
, page
, nr_pages
);
4592 memcg_check_events(to
, page
);
4593 mem_cgroup_charge_statistics(from
, page
, -nr_pages
);
4594 memcg_check_events(from
, page
);
4602 static enum mc_target_type
get_mctgt_type(struct vm_area_struct
*vma
,
4603 unsigned long addr
, pte_t ptent
, union mc_target
*target
)
4605 struct page
*page
= NULL
;
4606 enum mc_target_type ret
= MC_TARGET_NONE
;
4607 swp_entry_t ent
= { .val
= 0 };
4609 if (pte_present(ptent
))
4610 page
= mc_handle_present_pte(vma
, addr
, ptent
);
4611 else if (is_swap_pte(ptent
))
4612 page
= mc_handle_swap_pte(vma
, addr
, ptent
, &ent
);
4613 else if (pte_none(ptent
))
4614 page
= mc_handle_file_pte(vma
, addr
, ptent
, &ent
);
4616 if (!page
&& !ent
.val
)
4620 * Do only loose check w/o serialization.
4621 * mem_cgroup_move_account() checks the page is valid or
4622 * not under LRU exclusion.
4624 if (page
->mem_cgroup
== mc
.from
) {
4625 ret
= MC_TARGET_PAGE
;
4627 target
->page
= page
;
4629 if (!ret
|| !target
)
4632 /* There is a swap entry and a page doesn't exist or isn't charged */
4633 if (ent
.val
&& !ret
&&
4634 mem_cgroup_id(mc
.from
) == lookup_swap_cgroup_id(ent
)) {
4635 ret
= MC_TARGET_SWAP
;
4642 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
4644 * We don't consider swapping or file mapped pages because THP does not
4645 * support them for now.
4646 * Caller should make sure that pmd_trans_huge(pmd) is true.
4648 static enum mc_target_type
get_mctgt_type_thp(struct vm_area_struct
*vma
,
4649 unsigned long addr
, pmd_t pmd
, union mc_target
*target
)
4651 struct page
*page
= NULL
;
4652 enum mc_target_type ret
= MC_TARGET_NONE
;
4654 page
= pmd_page(pmd
);
4655 VM_BUG_ON_PAGE(!page
|| !PageHead(page
), page
);
4656 if (!(mc
.flags
& MOVE_ANON
))
4658 if (page
->mem_cgroup
== mc
.from
) {
4659 ret
= MC_TARGET_PAGE
;
4662 target
->page
= page
;
4668 static inline enum mc_target_type
get_mctgt_type_thp(struct vm_area_struct
*vma
,
4669 unsigned long addr
, pmd_t pmd
, union mc_target
*target
)
4671 return MC_TARGET_NONE
;
4675 static int mem_cgroup_count_precharge_pte_range(pmd_t
*pmd
,
4676 unsigned long addr
, unsigned long end
,
4677 struct mm_walk
*walk
)
4679 struct vm_area_struct
*vma
= walk
->vma
;
4683 if (pmd_trans_huge_lock(pmd
, vma
, &ptl
) == 1) {
4684 if (get_mctgt_type_thp(vma
, addr
, *pmd
, NULL
) == MC_TARGET_PAGE
)
4685 mc
.precharge
+= HPAGE_PMD_NR
;
4690 if (pmd_trans_unstable(pmd
))
4692 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
4693 for (; addr
!= end
; pte
++, addr
+= PAGE_SIZE
)
4694 if (get_mctgt_type(vma
, addr
, *pte
, NULL
))
4695 mc
.precharge
++; /* increment precharge temporarily */
4696 pte_unmap_unlock(pte
- 1, ptl
);
4702 static unsigned long mem_cgroup_count_precharge(struct mm_struct
*mm
)
4704 unsigned long precharge
;
4706 struct mm_walk mem_cgroup_count_precharge_walk
= {
4707 .pmd_entry
= mem_cgroup_count_precharge_pte_range
,
4710 down_read(&mm
->mmap_sem
);
4711 walk_page_range(0, ~0UL, &mem_cgroup_count_precharge_walk
);
4712 up_read(&mm
->mmap_sem
);
4714 precharge
= mc
.precharge
;
4720 static int mem_cgroup_precharge_mc(struct mm_struct
*mm
)
4722 unsigned long precharge
= mem_cgroup_count_precharge(mm
);
4724 VM_BUG_ON(mc
.moving_task
);
4725 mc
.moving_task
= current
;
4726 return mem_cgroup_do_precharge(precharge
);
4729 /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */
4730 static void __mem_cgroup_clear_mc(void)
4732 struct mem_cgroup
*from
= mc
.from
;
4733 struct mem_cgroup
*to
= mc
.to
;
4735 /* we must uncharge all the leftover precharges from mc.to */
4737 cancel_charge(mc
.to
, mc
.precharge
);
4741 * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
4742 * we must uncharge here.
4744 if (mc
.moved_charge
) {
4745 cancel_charge(mc
.from
, mc
.moved_charge
);
4746 mc
.moved_charge
= 0;
4748 /* we must fixup refcnts and charges */
4749 if (mc
.moved_swap
) {
4750 /* uncharge swap account from the old cgroup */
4751 if (!mem_cgroup_is_root(mc
.from
))
4752 page_counter_uncharge(&mc
.from
->memsw
, mc
.moved_swap
);
4755 * we charged both to->memory and to->memsw, so we
4756 * should uncharge to->memory.
4758 if (!mem_cgroup_is_root(mc
.to
))
4759 page_counter_uncharge(&mc
.to
->memory
, mc
.moved_swap
);
4761 css_put_many(&mc
.from
->css
, mc
.moved_swap
);
4763 /* we've already done css_get(mc.to) */
4766 memcg_oom_recover(from
);
4767 memcg_oom_recover(to
);
4768 wake_up_all(&mc
.waitq
);
4771 static void mem_cgroup_clear_mc(void)
4774 * we must clear moving_task before waking up waiters at the end of
4777 mc
.moving_task
= NULL
;
4778 __mem_cgroup_clear_mc();
4779 spin_lock(&mc
.lock
);
4782 spin_unlock(&mc
.lock
);
4785 static int mem_cgroup_can_attach(struct cgroup_subsys_state
*css
,
4786 struct cgroup_taskset
*tset
)
4788 struct mem_cgroup
*memcg
= mem_cgroup_from_css(css
);
4789 struct mem_cgroup
*from
;
4790 struct task_struct
*p
;
4791 struct mm_struct
*mm
;
4792 unsigned long move_flags
;
4796 * We are now commited to this value whatever it is. Changes in this
4797 * tunable will only affect upcoming migrations, not the current one.
4798 * So we need to save it, and keep it going.
4800 move_flags
= READ_ONCE(memcg
->move_charge_at_immigrate
);
4804 p
= cgroup_taskset_first(tset
);
4805 from
= mem_cgroup_from_task(p
);
4807 VM_BUG_ON(from
== memcg
);
4809 mm
= get_task_mm(p
);
4812 /* We move charges only when we move a owner of the mm */
4813 if (mm
->owner
== p
) {
4816 VM_BUG_ON(mc
.precharge
);
4817 VM_BUG_ON(mc
.moved_charge
);
4818 VM_BUG_ON(mc
.moved_swap
);
4820 spin_lock(&mc
.lock
);
4823 mc
.flags
= move_flags
;
4824 spin_unlock(&mc
.lock
);
4825 /* We set mc.moving_task later */
4827 ret
= mem_cgroup_precharge_mc(mm
);
4829 mem_cgroup_clear_mc();
4835 static void mem_cgroup_cancel_attach(struct cgroup_subsys_state
*css
,
4836 struct cgroup_taskset
*tset
)
4839 mem_cgroup_clear_mc();
4842 static int mem_cgroup_move_charge_pte_range(pmd_t
*pmd
,
4843 unsigned long addr
, unsigned long end
,
4844 struct mm_walk
*walk
)
4847 struct vm_area_struct
*vma
= walk
->vma
;
4850 enum mc_target_type target_type
;
4851 union mc_target target
;
4855 * We don't take compound_lock() here but no race with splitting thp
4857 * - if pmd_trans_huge_lock() returns 1, the relevant thp is not
4858 * under splitting, which means there's no concurrent thp split,
4859 * - if another thread runs into split_huge_page() just after we
4860 * entered this if-block, the thread must wait for page table lock
4861 * to be unlocked in __split_huge_page_splitting(), where the main
4862 * part of thp split is not executed yet.
4864 if (pmd_trans_huge_lock(pmd
, vma
, &ptl
) == 1) {
4865 if (mc
.precharge
< HPAGE_PMD_NR
) {
4869 target_type
= get_mctgt_type_thp(vma
, addr
, *pmd
, &target
);
4870 if (target_type
== MC_TARGET_PAGE
) {
4872 if (!isolate_lru_page(page
)) {
4873 if (!mem_cgroup_move_account(page
, HPAGE_PMD_NR
,
4875 mc
.precharge
-= HPAGE_PMD_NR
;
4876 mc
.moved_charge
+= HPAGE_PMD_NR
;
4878 putback_lru_page(page
);
4886 if (pmd_trans_unstable(pmd
))
4889 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
4890 for (; addr
!= end
; addr
+= PAGE_SIZE
) {
4891 pte_t ptent
= *(pte
++);
4897 switch (get_mctgt_type(vma
, addr
, ptent
, &target
)) {
4898 case MC_TARGET_PAGE
:
4900 if (isolate_lru_page(page
))
4902 if (!mem_cgroup_move_account(page
, 1, mc
.from
, mc
.to
)) {
4904 /* we uncharge from mc.from later. */
4907 putback_lru_page(page
);
4908 put
: /* get_mctgt_type() gets the page */
4911 case MC_TARGET_SWAP
:
4913 if (!mem_cgroup_move_swap_account(ent
, mc
.from
, mc
.to
)) {
4915 /* we fixup refcnts and charges later. */
4923 pte_unmap_unlock(pte
- 1, ptl
);
4928 * We have consumed all precharges we got in can_attach().
4929 * We try charge one by one, but don't do any additional
4930 * charges to mc.to if we have failed in charge once in attach()
4933 ret
= mem_cgroup_do_precharge(1);
4941 static void mem_cgroup_move_charge(struct mm_struct
*mm
)
4943 struct mm_walk mem_cgroup_move_charge_walk
= {
4944 .pmd_entry
= mem_cgroup_move_charge_pte_range
,
4948 lru_add_drain_all();
4950 * Signal mem_cgroup_begin_page_stat() to take the memcg's
4951 * move_lock while we're moving its pages to another memcg.
4952 * Then wait for already started RCU-only updates to finish.
4954 atomic_inc(&mc
.from
->moving_account
);
4957 if (unlikely(!down_read_trylock(&mm
->mmap_sem
))) {
4959 * Someone who are holding the mmap_sem might be waiting in
4960 * waitq. So we cancel all extra charges, wake up all waiters,
4961 * and retry. Because we cancel precharges, we might not be able
4962 * to move enough charges, but moving charge is a best-effort
4963 * feature anyway, so it wouldn't be a big problem.
4965 __mem_cgroup_clear_mc();
4970 * When we have consumed all precharges and failed in doing
4971 * additional charge, the page walk just aborts.
4973 walk_page_range(0, ~0UL, &mem_cgroup_move_charge_walk
);
4974 up_read(&mm
->mmap_sem
);
4975 atomic_dec(&mc
.from
->moving_account
);
4978 static void mem_cgroup_move_task(struct cgroup_subsys_state
*css
,
4979 struct cgroup_taskset
*tset
)
4981 struct task_struct
*p
= cgroup_taskset_first(tset
);
4982 struct mm_struct
*mm
= get_task_mm(p
);
4986 mem_cgroup_move_charge(mm
);
4990 mem_cgroup_clear_mc();
4992 #else /* !CONFIG_MMU */
4993 static int mem_cgroup_can_attach(struct cgroup_subsys_state
*css
,
4994 struct cgroup_taskset
*tset
)
4998 static void mem_cgroup_cancel_attach(struct cgroup_subsys_state
*css
,
4999 struct cgroup_taskset
*tset
)
5002 static void mem_cgroup_move_task(struct cgroup_subsys_state
*css
,
5003 struct cgroup_taskset
*tset
)
5009 * Cgroup retains root cgroups across [un]mount cycles making it necessary
5010 * to verify whether we're attached to the default hierarchy on each mount
5013 static void mem_cgroup_bind(struct cgroup_subsys_state
*root_css
)
5016 * use_hierarchy is forced on the default hierarchy. cgroup core
5017 * guarantees that @root doesn't have any children, so turning it
5018 * on for the root memcg is enough.
5020 if (cgroup_on_dfl(root_css
->cgroup
))
5021 root_mem_cgroup
->use_hierarchy
= true;
5023 root_mem_cgroup
->use_hierarchy
= false;
5026 static u64
memory_current_read(struct cgroup_subsys_state
*css
,
5029 return mem_cgroup_usage(mem_cgroup_from_css(css
), false);
5032 static int memory_low_show(struct seq_file
*m
, void *v
)
5034 struct mem_cgroup
*memcg
= mem_cgroup_from_css(seq_css(m
));
5035 unsigned long low
= READ_ONCE(memcg
->low
);
5037 if (low
== PAGE_COUNTER_MAX
)
5038 seq_puts(m
, "max\n");
5040 seq_printf(m
, "%llu\n", (u64
)low
* PAGE_SIZE
);
5045 static ssize_t
memory_low_write(struct kernfs_open_file
*of
,
5046 char *buf
, size_t nbytes
, loff_t off
)
5048 struct mem_cgroup
*memcg
= mem_cgroup_from_css(of_css(of
));
5052 buf
= strstrip(buf
);
5053 err
= page_counter_memparse(buf
, "max", &low
);
5062 static int memory_high_show(struct seq_file
*m
, void *v
)
5064 struct mem_cgroup
*memcg
= mem_cgroup_from_css(seq_css(m
));
5065 unsigned long high
= READ_ONCE(memcg
->high
);
5067 if (high
== PAGE_COUNTER_MAX
)
5068 seq_puts(m
, "max\n");
5070 seq_printf(m
, "%llu\n", (u64
)high
* PAGE_SIZE
);
5075 static ssize_t
memory_high_write(struct kernfs_open_file
*of
,
5076 char *buf
, size_t nbytes
, loff_t off
)
5078 struct mem_cgroup
*memcg
= mem_cgroup_from_css(of_css(of
));
5082 buf
= strstrip(buf
);
5083 err
= page_counter_memparse(buf
, "max", &high
);
5089 memcg_wb_domain_size_changed(memcg
);
5093 static int memory_max_show(struct seq_file
*m
, void *v
)
5095 struct mem_cgroup
*memcg
= mem_cgroup_from_css(seq_css(m
));
5096 unsigned long max
= READ_ONCE(memcg
->memory
.limit
);
5098 if (max
== PAGE_COUNTER_MAX
)
5099 seq_puts(m
, "max\n");
5101 seq_printf(m
, "%llu\n", (u64
)max
* PAGE_SIZE
);
5106 static ssize_t
memory_max_write(struct kernfs_open_file
*of
,
5107 char *buf
, size_t nbytes
, loff_t off
)
5109 struct mem_cgroup
*memcg
= mem_cgroup_from_css(of_css(of
));
5113 buf
= strstrip(buf
);
5114 err
= page_counter_memparse(buf
, "max", &max
);
5118 err
= mem_cgroup_resize_limit(memcg
, max
);
5122 memcg_wb_domain_size_changed(memcg
);
5126 static int memory_events_show(struct seq_file
*m
, void *v
)
5128 struct mem_cgroup
*memcg
= mem_cgroup_from_css(seq_css(m
));
5130 seq_printf(m
, "low %lu\n", mem_cgroup_read_events(memcg
, MEMCG_LOW
));
5131 seq_printf(m
, "high %lu\n", mem_cgroup_read_events(memcg
, MEMCG_HIGH
));
5132 seq_printf(m
, "max %lu\n", mem_cgroup_read_events(memcg
, MEMCG_MAX
));
5133 seq_printf(m
, "oom %lu\n", mem_cgroup_read_events(memcg
, MEMCG_OOM
));
5138 static struct cftype memory_files
[] = {
5141 .read_u64
= memory_current_read
,
5145 .flags
= CFTYPE_NOT_ON_ROOT
,
5146 .seq_show
= memory_low_show
,
5147 .write
= memory_low_write
,
5151 .flags
= CFTYPE_NOT_ON_ROOT
,
5152 .seq_show
= memory_high_show
,
5153 .write
= memory_high_write
,
5157 .flags
= CFTYPE_NOT_ON_ROOT
,
5158 .seq_show
= memory_max_show
,
5159 .write
= memory_max_write
,
5163 .flags
= CFTYPE_NOT_ON_ROOT
,
5164 .seq_show
= memory_events_show
,
5169 struct cgroup_subsys memory_cgrp_subsys
= {
5170 .css_alloc
= mem_cgroup_css_alloc
,
5171 .css_online
= mem_cgroup_css_online
,
5172 .css_offline
= mem_cgroup_css_offline
,
5173 .css_free
= mem_cgroup_css_free
,
5174 .css_reset
= mem_cgroup_css_reset
,
5175 .can_attach
= mem_cgroup_can_attach
,
5176 .cancel_attach
= mem_cgroup_cancel_attach
,
5177 .attach
= mem_cgroup_move_task
,
5178 .bind
= mem_cgroup_bind
,
5179 .dfl_cftypes
= memory_files
,
5180 .legacy_cftypes
= mem_cgroup_legacy_files
,
5185 * mem_cgroup_low - check if memory consumption is below the normal range
5186 * @root: the highest ancestor to consider
5187 * @memcg: the memory cgroup to check
5189 * Returns %true if memory consumption of @memcg, and that of all
5190 * configurable ancestors up to @root, is below the normal range.
5192 bool mem_cgroup_low(struct mem_cgroup
*root
, struct mem_cgroup
*memcg
)
5194 if (mem_cgroup_disabled())
5198 * The toplevel group doesn't have a configurable range, so
5199 * it's never low when looked at directly, and it is not
5200 * considered an ancestor when assessing the hierarchy.
5203 if (memcg
== root_mem_cgroup
)
5206 if (page_counter_read(&memcg
->memory
) >= memcg
->low
)
5209 while (memcg
!= root
) {
5210 memcg
= parent_mem_cgroup(memcg
);
5212 if (memcg
== root_mem_cgroup
)
5215 if (page_counter_read(&memcg
->memory
) >= memcg
->low
)
5222 * mem_cgroup_try_charge - try charging a page
5223 * @page: page to charge
5224 * @mm: mm context of the victim
5225 * @gfp_mask: reclaim mode
5226 * @memcgp: charged memcg return
5228 * Try to charge @page to the memcg that @mm belongs to, reclaiming
5229 * pages according to @gfp_mask if necessary.
5231 * Returns 0 on success, with *@memcgp pointing to the charged memcg.
5232 * Otherwise, an error code is returned.
5234 * After page->mapping has been set up, the caller must finalize the
5235 * charge with mem_cgroup_commit_charge(). Or abort the transaction
5236 * with mem_cgroup_cancel_charge() in case page instantiation fails.
5238 int mem_cgroup_try_charge(struct page
*page
, struct mm_struct
*mm
,
5239 gfp_t gfp_mask
, struct mem_cgroup
**memcgp
)
5241 struct mem_cgroup
*memcg
= NULL
;
5242 unsigned int nr_pages
= 1;
5245 if (mem_cgroup_disabled())
5248 if (PageSwapCache(page
)) {
5250 * Every swap fault against a single page tries to charge the
5251 * page, bail as early as possible. shmem_unuse() encounters
5252 * already charged pages, too. The USED bit is protected by
5253 * the page lock, which serializes swap cache removal, which
5254 * in turn serializes uncharging.
5256 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
5257 if (page
->mem_cgroup
)
5260 if (do_swap_account
) {
5261 swp_entry_t ent
= { .val
= page_private(page
), };
5262 unsigned short id
= lookup_swap_cgroup_id(ent
);
5265 memcg
= mem_cgroup_from_id(id
);
5266 if (memcg
&& !css_tryget_online(&memcg
->css
))
5272 if (PageTransHuge(page
)) {
5273 nr_pages
<<= compound_order(page
);
5274 VM_BUG_ON_PAGE(!PageTransHuge(page
), page
);
5278 memcg
= get_mem_cgroup_from_mm(mm
);
5280 ret
= try_charge(memcg
, gfp_mask
, nr_pages
);
5282 css_put(&memcg
->css
);
5289 * mem_cgroup_commit_charge - commit a page charge
5290 * @page: page to charge
5291 * @memcg: memcg to charge the page to
5292 * @lrucare: page might be on LRU already
5294 * Finalize a charge transaction started by mem_cgroup_try_charge(),
5295 * after page->mapping has been set up. This must happen atomically
5296 * as part of the page instantiation, i.e. under the page table lock
5297 * for anonymous pages, under the page lock for page and swap cache.
5299 * In addition, the page must not be on the LRU during the commit, to
5300 * prevent racing with task migration. If it might be, use @lrucare.
5302 * Use mem_cgroup_cancel_charge() to cancel the transaction instead.
5304 void mem_cgroup_commit_charge(struct page
*page
, struct mem_cgroup
*memcg
,
5307 unsigned int nr_pages
= 1;
5309 VM_BUG_ON_PAGE(!page
->mapping
, page
);
5310 VM_BUG_ON_PAGE(PageLRU(page
) && !lrucare
, page
);
5312 if (mem_cgroup_disabled())
5315 * Swap faults will attempt to charge the same page multiple
5316 * times. But reuse_swap_page() might have removed the page
5317 * from swapcache already, so we can't check PageSwapCache().
5322 commit_charge(page
, memcg
, lrucare
);
5324 if (PageTransHuge(page
)) {
5325 nr_pages
<<= compound_order(page
);
5326 VM_BUG_ON_PAGE(!PageTransHuge(page
), page
);
5329 local_irq_disable();
5330 mem_cgroup_charge_statistics(memcg
, page
, nr_pages
);
5331 memcg_check_events(memcg
, page
);
5334 if (do_swap_account
&& PageSwapCache(page
)) {
5335 swp_entry_t entry
= { .val
= page_private(page
) };
5337 * The swap entry might not get freed for a long time,
5338 * let's not wait for it. The page already received a
5339 * memory+swap charge, drop the swap entry duplicate.
5341 mem_cgroup_uncharge_swap(entry
);
5346 * mem_cgroup_cancel_charge - cancel a page charge
5347 * @page: page to charge
5348 * @memcg: memcg to charge the page to
5350 * Cancel a charge transaction started by mem_cgroup_try_charge().
5352 void mem_cgroup_cancel_charge(struct page
*page
, struct mem_cgroup
*memcg
)
5354 unsigned int nr_pages
= 1;
5356 if (mem_cgroup_disabled())
5359 * Swap faults will attempt to charge the same page multiple
5360 * times. But reuse_swap_page() might have removed the page
5361 * from swapcache already, so we can't check PageSwapCache().
5366 if (PageTransHuge(page
)) {
5367 nr_pages
<<= compound_order(page
);
5368 VM_BUG_ON_PAGE(!PageTransHuge(page
), page
);
5371 cancel_charge(memcg
, nr_pages
);
5374 static void uncharge_batch(struct mem_cgroup
*memcg
, unsigned long pgpgout
,
5375 unsigned long nr_anon
, unsigned long nr_file
,
5376 unsigned long nr_huge
, struct page
*dummy_page
)
5378 unsigned long nr_pages
= nr_anon
+ nr_file
;
5379 unsigned long flags
;
5381 if (!mem_cgroup_is_root(memcg
)) {
5382 page_counter_uncharge(&memcg
->memory
, nr_pages
);
5383 if (do_swap_account
)
5384 page_counter_uncharge(&memcg
->memsw
, nr_pages
);
5385 memcg_oom_recover(memcg
);
5388 local_irq_save(flags
);
5389 __this_cpu_sub(memcg
->stat
->count
[MEM_CGROUP_STAT_RSS
], nr_anon
);
5390 __this_cpu_sub(memcg
->stat
->count
[MEM_CGROUP_STAT_CACHE
], nr_file
);
5391 __this_cpu_sub(memcg
->stat
->count
[MEM_CGROUP_STAT_RSS_HUGE
], nr_huge
);
5392 __this_cpu_add(memcg
->stat
->events
[MEM_CGROUP_EVENTS_PGPGOUT
], pgpgout
);
5393 __this_cpu_add(memcg
->stat
->nr_page_events
, nr_pages
);
5394 memcg_check_events(memcg
, dummy_page
);
5395 local_irq_restore(flags
);
5397 if (!mem_cgroup_is_root(memcg
))
5398 css_put_many(&memcg
->css
, nr_pages
);
5401 static void uncharge_list(struct list_head
*page_list
)
5403 struct mem_cgroup
*memcg
= NULL
;
5404 unsigned long nr_anon
= 0;
5405 unsigned long nr_file
= 0;
5406 unsigned long nr_huge
= 0;
5407 unsigned long pgpgout
= 0;
5408 struct list_head
*next
;
5411 next
= page_list
->next
;
5413 unsigned int nr_pages
= 1;
5415 page
= list_entry(next
, struct page
, lru
);
5416 next
= page
->lru
.next
;
5418 VM_BUG_ON_PAGE(PageLRU(page
), page
);
5419 VM_BUG_ON_PAGE(page_count(page
), page
);
5421 if (!page
->mem_cgroup
)
5425 * Nobody should be changing or seriously looking at
5426 * page->mem_cgroup at this point, we have fully
5427 * exclusive access to the page.
5430 if (memcg
!= page
->mem_cgroup
) {
5432 uncharge_batch(memcg
, pgpgout
, nr_anon
, nr_file
,
5434 pgpgout
= nr_anon
= nr_file
= nr_huge
= 0;
5436 memcg
= page
->mem_cgroup
;
5439 if (PageTransHuge(page
)) {
5440 nr_pages
<<= compound_order(page
);
5441 VM_BUG_ON_PAGE(!PageTransHuge(page
), page
);
5442 nr_huge
+= nr_pages
;
5446 nr_anon
+= nr_pages
;
5448 nr_file
+= nr_pages
;
5450 page
->mem_cgroup
= NULL
;
5453 } while (next
!= page_list
);
5456 uncharge_batch(memcg
, pgpgout
, nr_anon
, nr_file
,
5461 * mem_cgroup_uncharge - uncharge a page
5462 * @page: page to uncharge
5464 * Uncharge a page previously charged with mem_cgroup_try_charge() and
5465 * mem_cgroup_commit_charge().
5467 void mem_cgroup_uncharge(struct page
*page
)
5469 if (mem_cgroup_disabled())
5472 /* Don't touch page->lru of any random page, pre-check: */
5473 if (!page
->mem_cgroup
)
5476 INIT_LIST_HEAD(&page
->lru
);
5477 uncharge_list(&page
->lru
);
5481 * mem_cgroup_uncharge_list - uncharge a list of page
5482 * @page_list: list of pages to uncharge
5484 * Uncharge a list of pages previously charged with
5485 * mem_cgroup_try_charge() and mem_cgroup_commit_charge().
5487 void mem_cgroup_uncharge_list(struct list_head
*page_list
)
5489 if (mem_cgroup_disabled())
5492 if (!list_empty(page_list
))
5493 uncharge_list(page_list
);
5497 * mem_cgroup_replace_page - migrate a charge to another page
5498 * @oldpage: currently charged page
5499 * @newpage: page to transfer the charge to
5500 * @lrucare: either or both pages might be on the LRU already
5502 * Migrate the charge from @oldpage to @newpage.
5504 * Both pages must be locked, @newpage->mapping must be set up.
5506 void mem_cgroup_replace_page(struct page
*oldpage
, struct page
*newpage
)
5508 struct mem_cgroup
*memcg
;
5511 VM_BUG_ON_PAGE(!PageLocked(oldpage
), oldpage
);
5512 VM_BUG_ON_PAGE(!PageLocked(newpage
), newpage
);
5513 VM_BUG_ON_PAGE(PageAnon(oldpage
) != PageAnon(newpage
), newpage
);
5514 VM_BUG_ON_PAGE(PageTransHuge(oldpage
) != PageTransHuge(newpage
),
5517 if (mem_cgroup_disabled())
5520 /* Page cache replacement: new page already charged? */
5521 if (newpage
->mem_cgroup
)
5524 /* Swapcache readahead pages can get replaced before being charged */
5525 memcg
= oldpage
->mem_cgroup
;
5529 lock_page_lru(oldpage
, &isolated
);
5530 oldpage
->mem_cgroup
= NULL
;
5531 unlock_page_lru(oldpage
, isolated
);
5533 commit_charge(newpage
, memcg
, true);
5537 * subsys_initcall() for memory controller.
5539 * Some parts like hotcpu_notifier() have to be initialized from this context
5540 * because of lock dependencies (cgroup_lock -> cpu hotplug) but basically
5541 * everything that doesn't depend on a specific mem_cgroup structure should
5542 * be initialized from here.
5544 static int __init
mem_cgroup_init(void)
5548 hotcpu_notifier(memcg_cpu_hotplug_callback
, 0);
5550 for_each_possible_cpu(cpu
)
5551 INIT_WORK(&per_cpu_ptr(&memcg_stock
, cpu
)->work
,
5554 for_each_node(node
) {
5555 struct mem_cgroup_tree_per_node
*rtpn
;
5558 rtpn
= kzalloc_node(sizeof(*rtpn
), GFP_KERNEL
,
5559 node_online(node
) ? node
: NUMA_NO_NODE
);
5561 for (zone
= 0; zone
< MAX_NR_ZONES
; zone
++) {
5562 struct mem_cgroup_tree_per_zone
*rtpz
;
5564 rtpz
= &rtpn
->rb_tree_per_zone
[zone
];
5565 rtpz
->rb_root
= RB_ROOT
;
5566 spin_lock_init(&rtpz
->lock
);
5568 soft_limit_tree
.rb_tree_per_node
[node
] = rtpn
;
5573 subsys_initcall(mem_cgroup_init
);
5575 #ifdef CONFIG_MEMCG_SWAP
5577 * mem_cgroup_swapout - transfer a memsw charge to swap
5578 * @page: page whose memsw charge to transfer
5579 * @entry: swap entry to move the charge to
5581 * Transfer the memsw charge of @page to @entry.
5583 void mem_cgroup_swapout(struct page
*page
, swp_entry_t entry
)
5585 struct mem_cgroup
*memcg
;
5586 unsigned short oldid
;
5588 VM_BUG_ON_PAGE(PageLRU(page
), page
);
5589 VM_BUG_ON_PAGE(page_count(page
), page
);
5591 if (!do_swap_account
)
5594 memcg
= page
->mem_cgroup
;
5596 /* Readahead page, never charged */
5600 oldid
= swap_cgroup_record(entry
, mem_cgroup_id(memcg
));
5601 VM_BUG_ON_PAGE(oldid
, page
);
5602 mem_cgroup_swap_statistics(memcg
, true);
5604 page
->mem_cgroup
= NULL
;
5606 if (!mem_cgroup_is_root(memcg
))
5607 page_counter_uncharge(&memcg
->memory
, 1);
5610 * Interrupts should be disabled here because the caller holds the
5611 * mapping->tree_lock lock which is taken with interrupts-off. It is
5612 * important here to have the interrupts disabled because it is the
5613 * only synchronisation we have for udpating the per-CPU variables.
5615 VM_BUG_ON(!irqs_disabled());
5616 mem_cgroup_charge_statistics(memcg
, page
, -1);
5617 memcg_check_events(memcg
, page
);
5621 * mem_cgroup_uncharge_swap - uncharge a swap entry
5622 * @entry: swap entry to uncharge
5624 * Drop the memsw charge associated with @entry.
5626 void mem_cgroup_uncharge_swap(swp_entry_t entry
)
5628 struct mem_cgroup
*memcg
;
5631 if (!do_swap_account
)
5634 id
= swap_cgroup_record(entry
, 0);
5636 memcg
= mem_cgroup_from_id(id
);
5638 if (!mem_cgroup_is_root(memcg
))
5639 page_counter_uncharge(&memcg
->memsw
, 1);
5640 mem_cgroup_swap_statistics(memcg
, false);
5641 css_put(&memcg
->css
);
5646 /* for remember boot option*/
5647 #ifdef CONFIG_MEMCG_SWAP_ENABLED
5648 static int really_do_swap_account __initdata
= 1;
5650 static int really_do_swap_account __initdata
;
5653 static int __init
enable_swap_account(char *s
)
5655 if (!strcmp(s
, "1"))
5656 really_do_swap_account
= 1;
5657 else if (!strcmp(s
, "0"))
5658 really_do_swap_account
= 0;
5661 __setup("swapaccount=", enable_swap_account
);
5663 static struct cftype memsw_cgroup_files
[] = {
5665 .name
= "memsw.usage_in_bytes",
5666 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_USAGE
),
5667 .read_u64
= mem_cgroup_read_u64
,
5670 .name
= "memsw.max_usage_in_bytes",
5671 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_MAX_USAGE
),
5672 .write
= mem_cgroup_reset
,
5673 .read_u64
= mem_cgroup_read_u64
,
5676 .name
= "memsw.limit_in_bytes",
5677 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_LIMIT
),
5678 .write
= mem_cgroup_write
,
5679 .read_u64
= mem_cgroup_read_u64
,
5682 .name
= "memsw.failcnt",
5683 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_FAILCNT
),
5684 .write
= mem_cgroup_reset
,
5685 .read_u64
= mem_cgroup_read_u64
,
5687 { }, /* terminate */
5690 static int __init
mem_cgroup_swap_init(void)
5692 if (!mem_cgroup_disabled() && really_do_swap_account
) {
5693 do_swap_account
= 1;
5694 WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys
,
5695 memsw_cgroup_files
));
5699 subsys_initcall(mem_cgroup_swap_init
);
5701 #endif /* CONFIG_MEMCG_SWAP */