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/sched/mm.h>
39 #include <linux/shmem_fs.h>
40 #include <linux/hugetlb.h>
41 #include <linux/pagemap.h>
42 #include <linux/smp.h>
43 #include <linux/page-flags.h>
44 #include <linux/backing-dev.h>
45 #include <linux/bit_spinlock.h>
46 #include <linux/rcupdate.h>
47 #include <linux/limits.h>
48 #include <linux/export.h>
49 #include <linux/mutex.h>
50 #include <linux/rbtree.h>
51 #include <linux/slab.h>
52 #include <linux/swap.h>
53 #include <linux/swapops.h>
54 #include <linux/spinlock.h>
55 #include <linux/eventfd.h>
56 #include <linux/poll.h>
57 #include <linux/sort.h>
59 #include <linux/seq_file.h>
60 #include <linux/vmpressure.h>
61 #include <linux/mm_inline.h>
62 #include <linux/swap_cgroup.h>
63 #include <linux/cpu.h>
64 #include <linux/oom.h>
65 #include <linux/lockdep.h>
66 #include <linux/file.h>
67 #include <linux/tracehook.h>
73 #include <linux/uaccess.h>
75 #include <trace/events/vmscan.h>
77 struct cgroup_subsys memory_cgrp_subsys __read_mostly
;
78 EXPORT_SYMBOL(memory_cgrp_subsys
);
80 struct mem_cgroup
*root_mem_cgroup __read_mostly
;
82 #define MEM_CGROUP_RECLAIM_RETRIES 5
84 /* Socket memory accounting disabled? */
85 static bool cgroup_memory_nosocket
;
87 /* Kernel memory accounting disabled? */
88 static bool cgroup_memory_nokmem
;
90 /* Whether the swap controller is active */
91 #ifdef CONFIG_MEMCG_SWAP
92 int do_swap_account __read_mostly
;
94 #define do_swap_account 0
97 /* Whether legacy memory+swap accounting is active */
98 static bool do_memsw_account(void)
100 return !cgroup_subsys_on_dfl(memory_cgrp_subsys
) && do_swap_account
;
103 static const char *const mem_cgroup_lru_names
[] = {
111 #define THRESHOLDS_EVENTS_TARGET 128
112 #define SOFTLIMIT_EVENTS_TARGET 1024
113 #define NUMAINFO_EVENTS_TARGET 1024
116 * Cgroups above their limits are maintained in a RB-Tree, independent of
117 * their hierarchy representation
120 struct mem_cgroup_tree_per_node
{
121 struct rb_root rb_root
;
122 struct rb_node
*rb_rightmost
;
126 struct mem_cgroup_tree
{
127 struct mem_cgroup_tree_per_node
*rb_tree_per_node
[MAX_NUMNODES
];
130 static struct mem_cgroup_tree soft_limit_tree __read_mostly
;
133 struct mem_cgroup_eventfd_list
{
134 struct list_head list
;
135 struct eventfd_ctx
*eventfd
;
139 * cgroup_event represents events which userspace want to receive.
141 struct mem_cgroup_event
{
143 * memcg which the event belongs to.
145 struct mem_cgroup
*memcg
;
147 * eventfd to signal userspace about the event.
149 struct eventfd_ctx
*eventfd
;
151 * Each of these stored in a list by the cgroup.
153 struct list_head list
;
155 * register_event() callback will be used to add new userspace
156 * waiter for changes related to this event. Use eventfd_signal()
157 * on eventfd to send notification to userspace.
159 int (*register_event
)(struct mem_cgroup
*memcg
,
160 struct eventfd_ctx
*eventfd
, const char *args
);
162 * unregister_event() callback will be called when userspace closes
163 * the eventfd or on cgroup removing. This callback must be set,
164 * if you want provide notification functionality.
166 void (*unregister_event
)(struct mem_cgroup
*memcg
,
167 struct eventfd_ctx
*eventfd
);
169 * All fields below needed to unregister event when
170 * userspace closes eventfd.
173 wait_queue_head_t
*wqh
;
174 wait_queue_entry_t wait
;
175 struct work_struct remove
;
178 static void mem_cgroup_threshold(struct mem_cgroup
*memcg
);
179 static void mem_cgroup_oom_notify(struct mem_cgroup
*memcg
);
181 /* Stuffs for move charges at task migration. */
183 * Types of charges to be moved.
185 #define MOVE_ANON 0x1U
186 #define MOVE_FILE 0x2U
187 #define MOVE_MASK (MOVE_ANON | MOVE_FILE)
189 /* "mc" and its members are protected by cgroup_mutex */
190 static struct move_charge_struct
{
191 spinlock_t lock
; /* for from, to */
192 struct mm_struct
*mm
;
193 struct mem_cgroup
*from
;
194 struct mem_cgroup
*to
;
196 unsigned long precharge
;
197 unsigned long moved_charge
;
198 unsigned long moved_swap
;
199 struct task_struct
*moving_task
; /* a task moving charges */
200 wait_queue_head_t waitq
; /* a waitq for other context */
202 .lock
= __SPIN_LOCK_UNLOCKED(mc
.lock
),
203 .waitq
= __WAIT_QUEUE_HEAD_INITIALIZER(mc
.waitq
),
207 * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft
208 * limit reclaim to prevent infinite loops, if they ever occur.
210 #define MEM_CGROUP_MAX_RECLAIM_LOOPS 100
211 #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2
214 MEM_CGROUP_CHARGE_TYPE_CACHE
= 0,
215 MEM_CGROUP_CHARGE_TYPE_ANON
,
216 MEM_CGROUP_CHARGE_TYPE_SWAPOUT
, /* for accounting swapcache */
217 MEM_CGROUP_CHARGE_TYPE_DROP
, /* a page was unused swap cache */
221 /* for encoding cft->private value on file */
230 #define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val))
231 #define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff)
232 #define MEMFILE_ATTR(val) ((val) & 0xffff)
233 /* Used for OOM nofiier */
234 #define OOM_CONTROL (0)
236 static inline bool should_force_charge(void)
238 return tsk_is_oom_victim(current
) || fatal_signal_pending(current
) ||
239 (current
->flags
& PF_EXITING
);
242 /* Some nice accessors for the vmpressure. */
243 struct vmpressure
*memcg_to_vmpressure(struct mem_cgroup
*memcg
)
246 memcg
= root_mem_cgroup
;
247 return &memcg
->vmpressure
;
250 struct cgroup_subsys_state
*vmpressure_to_css(struct vmpressure
*vmpr
)
252 return &container_of(vmpr
, struct mem_cgroup
, vmpressure
)->css
;
255 static inline bool mem_cgroup_is_root(struct mem_cgroup
*memcg
)
257 return (memcg
== root_mem_cgroup
);
262 * This will be the memcg's index in each cache's ->memcg_params.memcg_caches.
263 * The main reason for not using cgroup id for this:
264 * this works better in sparse environments, where we have a lot of memcgs,
265 * but only a few kmem-limited. Or also, if we have, for instance, 200
266 * memcgs, and none but the 200th is kmem-limited, we'd have to have a
267 * 200 entry array for that.
269 * The current size of the caches array is stored in memcg_nr_cache_ids. It
270 * will double each time we have to increase it.
272 static DEFINE_IDA(memcg_cache_ida
);
273 int memcg_nr_cache_ids
;
275 /* Protects memcg_nr_cache_ids */
276 static DECLARE_RWSEM(memcg_cache_ids_sem
);
278 void memcg_get_cache_ids(void)
280 down_read(&memcg_cache_ids_sem
);
283 void memcg_put_cache_ids(void)
285 up_read(&memcg_cache_ids_sem
);
289 * MIN_SIZE is different than 1, because we would like to avoid going through
290 * the alloc/free process all the time. In a small machine, 4 kmem-limited
291 * cgroups is a reasonable guess. In the future, it could be a parameter or
292 * tunable, but that is strictly not necessary.
294 * MAX_SIZE should be as large as the number of cgrp_ids. Ideally, we could get
295 * this constant directly from cgroup, but it is understandable that this is
296 * better kept as an internal representation in cgroup.c. In any case, the
297 * cgrp_id space is not getting any smaller, and we don't have to necessarily
298 * increase ours as well if it increases.
300 #define MEMCG_CACHES_MIN_SIZE 4
301 #define MEMCG_CACHES_MAX_SIZE MEM_CGROUP_ID_MAX
304 * A lot of the calls to the cache allocation functions are expected to be
305 * inlined by the compiler. Since the calls to memcg_kmem_get_cache are
306 * conditional to this static branch, we'll have to allow modules that does
307 * kmem_cache_alloc and the such to see this symbol as well
309 DEFINE_STATIC_KEY_FALSE(memcg_kmem_enabled_key
);
310 EXPORT_SYMBOL(memcg_kmem_enabled_key
);
312 struct workqueue_struct
*memcg_kmem_cache_wq
;
314 #endif /* !CONFIG_SLOB */
317 * mem_cgroup_css_from_page - css of the memcg associated with a page
318 * @page: page of interest
320 * If memcg is bound to the default hierarchy, css of the memcg associated
321 * with @page is returned. The returned css remains associated with @page
322 * until it is released.
324 * If memcg is bound to a traditional hierarchy, the css of root_mem_cgroup
327 struct cgroup_subsys_state
*mem_cgroup_css_from_page(struct page
*page
)
329 struct mem_cgroup
*memcg
;
331 memcg
= page
->mem_cgroup
;
333 if (!memcg
|| !cgroup_subsys_on_dfl(memory_cgrp_subsys
))
334 memcg
= root_mem_cgroup
;
340 * page_cgroup_ino - return inode number of the memcg a page is charged to
343 * Look up the closest online ancestor of the memory cgroup @page is charged to
344 * and return its inode number or 0 if @page is not charged to any cgroup. It
345 * is safe to call this function without holding a reference to @page.
347 * Note, this function is inherently racy, because there is nothing to prevent
348 * the cgroup inode from getting torn down and potentially reallocated a moment
349 * after page_cgroup_ino() returns, so it only should be used by callers that
350 * do not care (such as procfs interfaces).
352 ino_t
page_cgroup_ino(struct page
*page
)
354 struct mem_cgroup
*memcg
;
355 unsigned long ino
= 0;
358 memcg
= READ_ONCE(page
->mem_cgroup
);
359 while (memcg
&& !(memcg
->css
.flags
& CSS_ONLINE
))
360 memcg
= parent_mem_cgroup(memcg
);
362 ino
= cgroup_ino(memcg
->css
.cgroup
);
367 static struct mem_cgroup_per_node
*
368 mem_cgroup_page_nodeinfo(struct mem_cgroup
*memcg
, struct page
*page
)
370 int nid
= page_to_nid(page
);
372 return memcg
->nodeinfo
[nid
];
375 static struct mem_cgroup_tree_per_node
*
376 soft_limit_tree_node(int nid
)
378 return soft_limit_tree
.rb_tree_per_node
[nid
];
381 static struct mem_cgroup_tree_per_node
*
382 soft_limit_tree_from_page(struct page
*page
)
384 int nid
= page_to_nid(page
);
386 return soft_limit_tree
.rb_tree_per_node
[nid
];
389 static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_node
*mz
,
390 struct mem_cgroup_tree_per_node
*mctz
,
391 unsigned long new_usage_in_excess
)
393 struct rb_node
**p
= &mctz
->rb_root
.rb_node
;
394 struct rb_node
*parent
= NULL
;
395 struct mem_cgroup_per_node
*mz_node
;
396 bool rightmost
= true;
401 mz
->usage_in_excess
= new_usage_in_excess
;
402 if (!mz
->usage_in_excess
)
406 mz_node
= rb_entry(parent
, struct mem_cgroup_per_node
,
408 if (mz
->usage_in_excess
< mz_node
->usage_in_excess
) {
414 * We can't avoid mem cgroups that are over their soft
415 * limit by the same amount
417 else if (mz
->usage_in_excess
>= mz_node
->usage_in_excess
)
422 mctz
->rb_rightmost
= &mz
->tree_node
;
424 rb_link_node(&mz
->tree_node
, parent
, p
);
425 rb_insert_color(&mz
->tree_node
, &mctz
->rb_root
);
429 static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_node
*mz
,
430 struct mem_cgroup_tree_per_node
*mctz
)
435 if (&mz
->tree_node
== mctz
->rb_rightmost
)
436 mctz
->rb_rightmost
= rb_prev(&mz
->tree_node
);
438 rb_erase(&mz
->tree_node
, &mctz
->rb_root
);
442 static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_node
*mz
,
443 struct mem_cgroup_tree_per_node
*mctz
)
447 spin_lock_irqsave(&mctz
->lock
, flags
);
448 __mem_cgroup_remove_exceeded(mz
, mctz
);
449 spin_unlock_irqrestore(&mctz
->lock
, flags
);
452 static unsigned long soft_limit_excess(struct mem_cgroup
*memcg
)
454 unsigned long nr_pages
= page_counter_read(&memcg
->memory
);
455 unsigned long soft_limit
= READ_ONCE(memcg
->soft_limit
);
456 unsigned long excess
= 0;
458 if (nr_pages
> soft_limit
)
459 excess
= nr_pages
- soft_limit
;
464 static void mem_cgroup_update_tree(struct mem_cgroup
*memcg
, struct page
*page
)
466 unsigned long excess
;
467 struct mem_cgroup_per_node
*mz
;
468 struct mem_cgroup_tree_per_node
*mctz
;
470 mctz
= soft_limit_tree_from_page(page
);
474 * Necessary to update all ancestors when hierarchy is used.
475 * because their event counter is not touched.
477 for (; memcg
; memcg
= parent_mem_cgroup(memcg
)) {
478 mz
= mem_cgroup_page_nodeinfo(memcg
, page
);
479 excess
= soft_limit_excess(memcg
);
481 * We have to update the tree if mz is on RB-tree or
482 * mem is over its softlimit.
484 if (excess
|| mz
->on_tree
) {
487 spin_lock_irqsave(&mctz
->lock
, flags
);
488 /* if on-tree, remove it */
490 __mem_cgroup_remove_exceeded(mz
, mctz
);
492 * Insert again. mz->usage_in_excess will be updated.
493 * If excess is 0, no tree ops.
495 __mem_cgroup_insert_exceeded(mz
, mctz
, excess
);
496 spin_unlock_irqrestore(&mctz
->lock
, flags
);
501 static void mem_cgroup_remove_from_trees(struct mem_cgroup
*memcg
)
503 struct mem_cgroup_tree_per_node
*mctz
;
504 struct mem_cgroup_per_node
*mz
;
508 mz
= mem_cgroup_nodeinfo(memcg
, nid
);
509 mctz
= soft_limit_tree_node(nid
);
511 mem_cgroup_remove_exceeded(mz
, mctz
);
515 static struct mem_cgroup_per_node
*
516 __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node
*mctz
)
518 struct mem_cgroup_per_node
*mz
;
522 if (!mctz
->rb_rightmost
)
523 goto done
; /* Nothing to reclaim from */
525 mz
= rb_entry(mctz
->rb_rightmost
,
526 struct mem_cgroup_per_node
, tree_node
);
528 * Remove the node now but someone else can add it back,
529 * we will to add it back at the end of reclaim to its correct
530 * position in the tree.
532 __mem_cgroup_remove_exceeded(mz
, mctz
);
533 if (!soft_limit_excess(mz
->memcg
) ||
534 !css_tryget_online(&mz
->memcg
->css
))
540 static struct mem_cgroup_per_node
*
541 mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node
*mctz
)
543 struct mem_cgroup_per_node
*mz
;
545 spin_lock_irq(&mctz
->lock
);
546 mz
= __mem_cgroup_largest_soft_limit_node(mctz
);
547 spin_unlock_irq(&mctz
->lock
);
552 * Return page count for single (non recursive) @memcg.
554 * Implementation Note: reading percpu statistics for memcg.
556 * Both of vmstat[] and percpu_counter has threshold and do periodic
557 * synchronization to implement "quick" read. There are trade-off between
558 * reading cost and precision of value. Then, we may have a chance to implement
559 * a periodic synchronization of counter in memcg's counter.
561 * But this _read() function is used for user interface now. The user accounts
562 * memory usage by memory cgroup and he _always_ requires exact value because
563 * he accounts memory. Even if we provide quick-and-fuzzy read, we always
564 * have to visit all online cpus and make sum. So, for now, unnecessary
565 * synchronization is not implemented. (just implemented for cpu hotplug)
567 * If there are kernel internal actions which can make use of some not-exact
568 * value, and reading all cpu value can be performance bottleneck in some
569 * common workload, threshold and synchronization as vmstat[] should be
572 * The parameter idx can be of type enum memcg_event_item or vm_event_item.
575 static unsigned long memcg_sum_events(struct mem_cgroup
*memcg
,
578 unsigned long val
= 0;
581 for_each_possible_cpu(cpu
)
582 val
+= per_cpu(memcg
->stat
->events
[event
], cpu
);
586 static void mem_cgroup_charge_statistics(struct mem_cgroup
*memcg
,
588 bool compound
, int nr_pages
)
591 * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is
592 * counted as CACHE even if it's on ANON LRU.
595 __this_cpu_add(memcg
->stat
->count
[MEMCG_RSS
], nr_pages
);
597 __this_cpu_add(memcg
->stat
->count
[MEMCG_CACHE
], nr_pages
);
598 if (PageSwapBacked(page
))
599 __this_cpu_add(memcg
->stat
->count
[NR_SHMEM
], nr_pages
);
603 VM_BUG_ON_PAGE(!PageTransHuge(page
), page
);
604 __this_cpu_add(memcg
->stat
->count
[MEMCG_RSS_HUGE
], nr_pages
);
607 /* pagein of a big page is an event. So, ignore page size */
609 __this_cpu_inc(memcg
->stat
->events
[PGPGIN
]);
611 __this_cpu_inc(memcg
->stat
->events
[PGPGOUT
]);
612 nr_pages
= -nr_pages
; /* for event */
615 __this_cpu_add(memcg
->stat
->nr_page_events
, nr_pages
);
618 unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup
*memcg
,
619 int nid
, unsigned int lru_mask
)
621 struct lruvec
*lruvec
= mem_cgroup_lruvec(NODE_DATA(nid
), memcg
);
622 unsigned long nr
= 0;
625 VM_BUG_ON((unsigned)nid
>= nr_node_ids
);
628 if (!(BIT(lru
) & lru_mask
))
630 nr
+= mem_cgroup_get_lru_size(lruvec
, lru
);
635 static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup
*memcg
,
636 unsigned int lru_mask
)
638 unsigned long nr
= 0;
641 for_each_node_state(nid
, N_MEMORY
)
642 nr
+= mem_cgroup_node_nr_lru_pages(memcg
, nid
, lru_mask
);
646 static bool mem_cgroup_event_ratelimit(struct mem_cgroup
*memcg
,
647 enum mem_cgroup_events_target target
)
649 unsigned long val
, next
;
651 val
= __this_cpu_read(memcg
->stat
->nr_page_events
);
652 next
= __this_cpu_read(memcg
->stat
->targets
[target
]);
653 /* from time_after() in jiffies.h */
654 if ((long)(next
- val
) < 0) {
656 case MEM_CGROUP_TARGET_THRESH
:
657 next
= val
+ THRESHOLDS_EVENTS_TARGET
;
659 case MEM_CGROUP_TARGET_SOFTLIMIT
:
660 next
= val
+ SOFTLIMIT_EVENTS_TARGET
;
662 case MEM_CGROUP_TARGET_NUMAINFO
:
663 next
= val
+ NUMAINFO_EVENTS_TARGET
;
668 __this_cpu_write(memcg
->stat
->targets
[target
], next
);
675 * Check events in order.
678 static void memcg_check_events(struct mem_cgroup
*memcg
, struct page
*page
)
680 /* threshold event is triggered in finer grain than soft limit */
681 if (unlikely(mem_cgroup_event_ratelimit(memcg
,
682 MEM_CGROUP_TARGET_THRESH
))) {
684 bool do_numainfo __maybe_unused
;
686 do_softlimit
= mem_cgroup_event_ratelimit(memcg
,
687 MEM_CGROUP_TARGET_SOFTLIMIT
);
689 do_numainfo
= mem_cgroup_event_ratelimit(memcg
,
690 MEM_CGROUP_TARGET_NUMAINFO
);
692 mem_cgroup_threshold(memcg
);
693 if (unlikely(do_softlimit
))
694 mem_cgroup_update_tree(memcg
, page
);
696 if (unlikely(do_numainfo
))
697 atomic_inc(&memcg
->numainfo_events
);
702 struct mem_cgroup
*mem_cgroup_from_task(struct task_struct
*p
)
705 * mm_update_next_owner() may clear mm->owner to NULL
706 * if it races with swapoff, page migration, etc.
707 * So this can be called with p == NULL.
712 return mem_cgroup_from_css(task_css(p
, memory_cgrp_id
));
714 EXPORT_SYMBOL(mem_cgroup_from_task
);
716 static struct mem_cgroup
*get_mem_cgroup_from_mm(struct mm_struct
*mm
)
718 struct mem_cgroup
*memcg
= NULL
;
723 * Page cache insertions can happen withou an
724 * actual mm context, e.g. during disk probing
725 * on boot, loopback IO, acct() writes etc.
728 memcg
= root_mem_cgroup
;
730 memcg
= mem_cgroup_from_task(rcu_dereference(mm
->owner
));
731 if (unlikely(!memcg
))
732 memcg
= root_mem_cgroup
;
734 } while (!css_tryget(&memcg
->css
));
740 * mem_cgroup_iter - iterate over memory cgroup hierarchy
741 * @root: hierarchy root
742 * @prev: previously returned memcg, NULL on first invocation
743 * @reclaim: cookie for shared reclaim walks, NULL for full walks
745 * Returns references to children of the hierarchy below @root, or
746 * @root itself, or %NULL after a full round-trip.
748 * Caller must pass the return value in @prev on subsequent
749 * invocations for reference counting, or use mem_cgroup_iter_break()
750 * to cancel a hierarchy walk before the round-trip is complete.
752 * Reclaimers can specify a zone and a priority level in @reclaim to
753 * divide up the memcgs in the hierarchy among all concurrent
754 * reclaimers operating on the same zone and priority.
756 struct mem_cgroup
*mem_cgroup_iter(struct mem_cgroup
*root
,
757 struct mem_cgroup
*prev
,
758 struct mem_cgroup_reclaim_cookie
*reclaim
)
760 struct mem_cgroup_reclaim_iter
*uninitialized_var(iter
);
761 struct cgroup_subsys_state
*css
= NULL
;
762 struct mem_cgroup
*memcg
= NULL
;
763 struct mem_cgroup
*pos
= NULL
;
765 if (mem_cgroup_disabled())
769 root
= root_mem_cgroup
;
771 if (prev
&& !reclaim
)
774 if (!root
->use_hierarchy
&& root
!= root_mem_cgroup
) {
783 struct mem_cgroup_per_node
*mz
;
785 mz
= mem_cgroup_nodeinfo(root
, reclaim
->pgdat
->node_id
);
786 iter
= &mz
->iter
[reclaim
->priority
];
788 if (prev
&& reclaim
->generation
!= iter
->generation
)
792 pos
= READ_ONCE(iter
->position
);
793 if (!pos
|| css_tryget(&pos
->css
))
796 * css reference reached zero, so iter->position will
797 * be cleared by ->css_released. However, we should not
798 * rely on this happening soon, because ->css_released
799 * is called from a work queue, and by busy-waiting we
800 * might block it. So we clear iter->position right
803 (void)cmpxchg(&iter
->position
, pos
, NULL
);
811 css
= css_next_descendant_pre(css
, &root
->css
);
814 * Reclaimers share the hierarchy walk, and a
815 * new one might jump in right at the end of
816 * the hierarchy - make sure they see at least
817 * one group and restart from the beginning.
825 * Verify the css and acquire a reference. The root
826 * is provided by the caller, so we know it's alive
827 * and kicking, and don't take an extra reference.
829 memcg
= mem_cgroup_from_css(css
);
831 if (css
== &root
->css
)
842 * The position could have already been updated by a competing
843 * thread, so check that the value hasn't changed since we read
844 * it to avoid reclaiming from the same cgroup twice.
846 (void)cmpxchg(&iter
->position
, pos
, memcg
);
854 reclaim
->generation
= iter
->generation
;
860 if (prev
&& prev
!= root
)
867 * mem_cgroup_iter_break - abort a hierarchy walk prematurely
868 * @root: hierarchy root
869 * @prev: last visited hierarchy member as returned by mem_cgroup_iter()
871 void mem_cgroup_iter_break(struct mem_cgroup
*root
,
872 struct mem_cgroup
*prev
)
875 root
= root_mem_cgroup
;
876 if (prev
&& prev
!= root
)
880 static void __invalidate_reclaim_iterators(struct mem_cgroup
*from
,
881 struct mem_cgroup
*dead_memcg
)
883 struct mem_cgroup_reclaim_iter
*iter
;
884 struct mem_cgroup_per_node
*mz
;
889 mz
= mem_cgroup_nodeinfo(from
, nid
);
890 for (i
= 0; i
<= DEF_PRIORITY
; i
++) {
892 cmpxchg(&iter
->position
,
898 static void invalidate_reclaim_iterators(struct mem_cgroup
*dead_memcg
)
900 struct mem_cgroup
*memcg
= dead_memcg
;
901 struct mem_cgroup
*last
;
904 __invalidate_reclaim_iterators(memcg
, dead_memcg
);
906 } while ((memcg
= parent_mem_cgroup(memcg
)));
909 * When cgruop1 non-hierarchy mode is used,
910 * parent_mem_cgroup() does not walk all the way up to the
911 * cgroup root (root_mem_cgroup). So we have to handle
912 * dead_memcg from cgroup root separately.
914 if (last
!= root_mem_cgroup
)
915 __invalidate_reclaim_iterators(root_mem_cgroup
,
920 * Iteration constructs for visiting all cgroups (under a tree). If
921 * loops are exited prematurely (break), mem_cgroup_iter_break() must
922 * be used for reference counting.
924 #define for_each_mem_cgroup_tree(iter, root) \
925 for (iter = mem_cgroup_iter(root, NULL, NULL); \
927 iter = mem_cgroup_iter(root, iter, NULL))
929 #define for_each_mem_cgroup(iter) \
930 for (iter = mem_cgroup_iter(NULL, NULL, NULL); \
932 iter = mem_cgroup_iter(NULL, iter, NULL))
935 * mem_cgroup_scan_tasks - iterate over tasks of a memory cgroup hierarchy
936 * @memcg: hierarchy root
937 * @fn: function to call for each task
938 * @arg: argument passed to @fn
940 * This function iterates over tasks attached to @memcg or to any of its
941 * descendants and calls @fn for each task. If @fn returns a non-zero
942 * value, the function breaks the iteration loop and returns the value.
943 * Otherwise, it will iterate over all tasks and return 0.
945 * This function must not be called for the root memory cgroup.
947 int mem_cgroup_scan_tasks(struct mem_cgroup
*memcg
,
948 int (*fn
)(struct task_struct
*, void *), void *arg
)
950 struct mem_cgroup
*iter
;
953 BUG_ON(memcg
== root_mem_cgroup
);
955 for_each_mem_cgroup_tree(iter
, memcg
) {
956 struct css_task_iter it
;
957 struct task_struct
*task
;
959 css_task_iter_start(&iter
->css
, 0, &it
);
960 while (!ret
&& (task
= css_task_iter_next(&it
)))
962 css_task_iter_end(&it
);
964 mem_cgroup_iter_break(memcg
, iter
);
972 * mem_cgroup_page_lruvec - return lruvec for isolating/putting an LRU page
974 * @zone: zone of the page
976 * This function is only safe when following the LRU page isolation
977 * and putback protocol: the LRU lock must be held, and the page must
978 * either be PageLRU() or the caller must have isolated/allocated it.
980 struct lruvec
*mem_cgroup_page_lruvec(struct page
*page
, struct pglist_data
*pgdat
)
982 struct mem_cgroup_per_node
*mz
;
983 struct mem_cgroup
*memcg
;
984 struct lruvec
*lruvec
;
986 if (mem_cgroup_disabled()) {
987 lruvec
= &pgdat
->lruvec
;
991 memcg
= page
->mem_cgroup
;
993 * Swapcache readahead pages are added to the LRU - and
994 * possibly migrated - before they are charged.
997 memcg
= root_mem_cgroup
;
999 mz
= mem_cgroup_page_nodeinfo(memcg
, page
);
1000 lruvec
= &mz
->lruvec
;
1003 * Since a node can be onlined after the mem_cgroup was created,
1004 * we have to be prepared to initialize lruvec->zone here;
1005 * and if offlined then reonlined, we need to reinitialize it.
1007 if (unlikely(lruvec
->pgdat
!= pgdat
))
1008 lruvec
->pgdat
= pgdat
;
1013 * mem_cgroup_update_lru_size - account for adding or removing an lru page
1014 * @lruvec: mem_cgroup per zone lru vector
1015 * @lru: index of lru list the page is sitting on
1016 * @zid: zone id of the accounted pages
1017 * @nr_pages: positive when adding or negative when removing
1019 * This function must be called under lru_lock, just before a page is added
1020 * to or just after a page is removed from an lru list (that ordering being
1021 * so as to allow it to check that lru_size 0 is consistent with list_empty).
1023 void mem_cgroup_update_lru_size(struct lruvec
*lruvec
, enum lru_list lru
,
1024 int zid
, int nr_pages
)
1026 struct mem_cgroup_per_node
*mz
;
1027 unsigned long *lru_size
;
1030 if (mem_cgroup_disabled())
1033 mz
= container_of(lruvec
, struct mem_cgroup_per_node
, lruvec
);
1034 lru_size
= &mz
->lru_zone_size
[zid
][lru
];
1037 *lru_size
+= nr_pages
;
1040 if (WARN_ONCE(size
< 0,
1041 "%s(%p, %d, %d): lru_size %ld\n",
1042 __func__
, lruvec
, lru
, nr_pages
, size
)) {
1048 *lru_size
+= nr_pages
;
1051 bool task_in_mem_cgroup(struct task_struct
*task
, struct mem_cgroup
*memcg
)
1053 struct mem_cgroup
*task_memcg
;
1054 struct task_struct
*p
;
1057 p
= find_lock_task_mm(task
);
1059 task_memcg
= get_mem_cgroup_from_mm(p
->mm
);
1063 * All threads may have already detached their mm's, but the oom
1064 * killer still needs to detect if they have already been oom
1065 * killed to prevent needlessly killing additional tasks.
1068 task_memcg
= mem_cgroup_from_task(task
);
1069 css_get(&task_memcg
->css
);
1072 ret
= mem_cgroup_is_descendant(task_memcg
, memcg
);
1073 css_put(&task_memcg
->css
);
1078 * mem_cgroup_margin - calculate chargeable space of a memory cgroup
1079 * @memcg: the memory cgroup
1081 * Returns the maximum amount of memory @mem can be charged with, in
1084 static unsigned long mem_cgroup_margin(struct mem_cgroup
*memcg
)
1086 unsigned long margin
= 0;
1087 unsigned long count
;
1088 unsigned long limit
;
1090 count
= page_counter_read(&memcg
->memory
);
1091 limit
= READ_ONCE(memcg
->memory
.limit
);
1093 margin
= limit
- count
;
1095 if (do_memsw_account()) {
1096 count
= page_counter_read(&memcg
->memsw
);
1097 limit
= READ_ONCE(memcg
->memsw
.limit
);
1099 margin
= min(margin
, limit
- count
);
1108 * A routine for checking "mem" is under move_account() or not.
1110 * Checking a cgroup is mc.from or mc.to or under hierarchy of
1111 * moving cgroups. This is for waiting at high-memory pressure
1114 static bool mem_cgroup_under_move(struct mem_cgroup
*memcg
)
1116 struct mem_cgroup
*from
;
1117 struct mem_cgroup
*to
;
1120 * Unlike task_move routines, we access mc.to, mc.from not under
1121 * mutual exclusion by cgroup_mutex. Here, we take spinlock instead.
1123 spin_lock(&mc
.lock
);
1129 ret
= mem_cgroup_is_descendant(from
, memcg
) ||
1130 mem_cgroup_is_descendant(to
, memcg
);
1132 spin_unlock(&mc
.lock
);
1136 static bool mem_cgroup_wait_acct_move(struct mem_cgroup
*memcg
)
1138 if (mc
.moving_task
&& current
!= mc
.moving_task
) {
1139 if (mem_cgroup_under_move(memcg
)) {
1141 prepare_to_wait(&mc
.waitq
, &wait
, TASK_INTERRUPTIBLE
);
1142 /* moving charge context might have finished. */
1145 finish_wait(&mc
.waitq
, &wait
);
1152 unsigned int memcg1_stats
[] = {
1163 static const char *const memcg1_stat_names
[] = {
1174 #define K(x) ((x) << (PAGE_SHIFT-10))
1176 * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller.
1177 * @memcg: The memory cgroup that went over limit
1178 * @p: Task that is going to be killed
1180 * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is
1183 void mem_cgroup_print_oom_info(struct mem_cgroup
*memcg
, struct task_struct
*p
)
1185 struct mem_cgroup
*iter
;
1191 pr_info("Task in ");
1192 pr_cont_cgroup_path(task_cgroup(p
, memory_cgrp_id
));
1193 pr_cont(" killed as a result of limit of ");
1195 pr_info("Memory limit reached of cgroup ");
1198 pr_cont_cgroup_path(memcg
->css
.cgroup
);
1203 pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n",
1204 K((u64
)page_counter_read(&memcg
->memory
)),
1205 K((u64
)memcg
->memory
.limit
), memcg
->memory
.failcnt
);
1206 pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n",
1207 K((u64
)page_counter_read(&memcg
->memsw
)),
1208 K((u64
)memcg
->memsw
.limit
), memcg
->memsw
.failcnt
);
1209 pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n",
1210 K((u64
)page_counter_read(&memcg
->kmem
)),
1211 K((u64
)memcg
->kmem
.limit
), memcg
->kmem
.failcnt
);
1213 for_each_mem_cgroup_tree(iter
, memcg
) {
1214 pr_info("Memory cgroup stats for ");
1215 pr_cont_cgroup_path(iter
->css
.cgroup
);
1218 for (i
= 0; i
< ARRAY_SIZE(memcg1_stats
); i
++) {
1219 if (memcg1_stats
[i
] == MEMCG_SWAP
&& !do_swap_account
)
1221 pr_cont(" %s:%luKB", memcg1_stat_names
[i
],
1222 K(memcg_page_state(iter
, memcg1_stats
[i
])));
1225 for (i
= 0; i
< NR_LRU_LISTS
; i
++)
1226 pr_cont(" %s:%luKB", mem_cgroup_lru_names
[i
],
1227 K(mem_cgroup_nr_lru_pages(iter
, BIT(i
))));
1234 * This function returns the number of memcg under hierarchy tree. Returns
1235 * 1(self count) if no children.
1237 static int mem_cgroup_count_children(struct mem_cgroup
*memcg
)
1240 struct mem_cgroup
*iter
;
1242 for_each_mem_cgroup_tree(iter
, memcg
)
1248 * Return the memory (and swap, if configured) limit for a memcg.
1250 unsigned long mem_cgroup_get_limit(struct mem_cgroup
*memcg
)
1252 unsigned long limit
;
1254 limit
= memcg
->memory
.limit
;
1255 if (mem_cgroup_swappiness(memcg
)) {
1256 unsigned long memsw_limit
;
1257 unsigned long swap_limit
;
1259 memsw_limit
= memcg
->memsw
.limit
;
1260 swap_limit
= memcg
->swap
.limit
;
1261 swap_limit
= min(swap_limit
, (unsigned long)total_swap_pages
);
1262 limit
= min(limit
+ swap_limit
, memsw_limit
);
1267 static bool mem_cgroup_out_of_memory(struct mem_cgroup
*memcg
, gfp_t gfp_mask
,
1270 struct oom_control oc
= {
1274 .gfp_mask
= gfp_mask
,
1279 if (mutex_lock_killable(&oom_lock
))
1282 * A few threads which were not waiting at mutex_lock_killable() can
1283 * fail to bail out. Therefore, check again after holding oom_lock.
1285 ret
= should_force_charge() || out_of_memory(&oc
);
1286 mutex_unlock(&oom_lock
);
1290 #if MAX_NUMNODES > 1
1293 * test_mem_cgroup_node_reclaimable
1294 * @memcg: the target memcg
1295 * @nid: the node ID to be checked.
1296 * @noswap : specify true here if the user wants flle only information.
1298 * This function returns whether the specified memcg contains any
1299 * reclaimable pages on a node. Returns true if there are any reclaimable
1300 * pages in the node.
1302 static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup
*memcg
,
1303 int nid
, bool noswap
)
1305 if (mem_cgroup_node_nr_lru_pages(memcg
, nid
, LRU_ALL_FILE
))
1307 if (noswap
|| !total_swap_pages
)
1309 if (mem_cgroup_node_nr_lru_pages(memcg
, nid
, LRU_ALL_ANON
))
1316 * Always updating the nodemask is not very good - even if we have an empty
1317 * list or the wrong list here, we can start from some node and traverse all
1318 * nodes based on the zonelist. So update the list loosely once per 10 secs.
1321 static void mem_cgroup_may_update_nodemask(struct mem_cgroup
*memcg
)
1325 * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET
1326 * pagein/pageout changes since the last update.
1328 if (!atomic_read(&memcg
->numainfo_events
))
1330 if (atomic_inc_return(&memcg
->numainfo_updating
) > 1)
1333 /* make a nodemask where this memcg uses memory from */
1334 memcg
->scan_nodes
= node_states
[N_MEMORY
];
1336 for_each_node_mask(nid
, node_states
[N_MEMORY
]) {
1338 if (!test_mem_cgroup_node_reclaimable(memcg
, nid
, false))
1339 node_clear(nid
, memcg
->scan_nodes
);
1342 atomic_set(&memcg
->numainfo_events
, 0);
1343 atomic_set(&memcg
->numainfo_updating
, 0);
1347 * Selecting a node where we start reclaim from. Because what we need is just
1348 * reducing usage counter, start from anywhere is O,K. Considering
1349 * memory reclaim from current node, there are pros. and cons.
1351 * Freeing memory from current node means freeing memory from a node which
1352 * we'll use or we've used. So, it may make LRU bad. And if several threads
1353 * hit limits, it will see a contention on a node. But freeing from remote
1354 * node means more costs for memory reclaim because of memory latency.
1356 * Now, we use round-robin. Better algorithm is welcomed.
1358 int mem_cgroup_select_victim_node(struct mem_cgroup
*memcg
)
1362 mem_cgroup_may_update_nodemask(memcg
);
1363 node
= memcg
->last_scanned_node
;
1365 node
= next_node_in(node
, memcg
->scan_nodes
);
1367 * mem_cgroup_may_update_nodemask might have seen no reclaimmable pages
1368 * last time it really checked all the LRUs due to rate limiting.
1369 * Fallback to the current node in that case for simplicity.
1371 if (unlikely(node
== MAX_NUMNODES
))
1372 node
= numa_node_id();
1374 memcg
->last_scanned_node
= node
;
1378 int mem_cgroup_select_victim_node(struct mem_cgroup
*memcg
)
1384 static int mem_cgroup_soft_reclaim(struct mem_cgroup
*root_memcg
,
1387 unsigned long *total_scanned
)
1389 struct mem_cgroup
*victim
= NULL
;
1392 unsigned long excess
;
1393 unsigned long nr_scanned
;
1394 struct mem_cgroup_reclaim_cookie reclaim
= {
1399 excess
= soft_limit_excess(root_memcg
);
1402 victim
= mem_cgroup_iter(root_memcg
, victim
, &reclaim
);
1407 * If we have not been able to reclaim
1408 * anything, it might because there are
1409 * no reclaimable pages under this hierarchy
1414 * We want to do more targeted reclaim.
1415 * excess >> 2 is not to excessive so as to
1416 * reclaim too much, nor too less that we keep
1417 * coming back to reclaim from this cgroup
1419 if (total
>= (excess
>> 2) ||
1420 (loop
> MEM_CGROUP_MAX_RECLAIM_LOOPS
))
1425 total
+= mem_cgroup_shrink_node(victim
, gfp_mask
, false,
1426 pgdat
, &nr_scanned
);
1427 *total_scanned
+= nr_scanned
;
1428 if (!soft_limit_excess(root_memcg
))
1431 mem_cgroup_iter_break(root_memcg
, victim
);
1435 #ifdef CONFIG_LOCKDEP
1436 static struct lockdep_map memcg_oom_lock_dep_map
= {
1437 .name
= "memcg_oom_lock",
1441 static DEFINE_SPINLOCK(memcg_oom_lock
);
1444 * Check OOM-Killer is already running under our hierarchy.
1445 * If someone is running, return false.
1447 static bool mem_cgroup_oom_trylock(struct mem_cgroup
*memcg
)
1449 struct mem_cgroup
*iter
, *failed
= NULL
;
1451 spin_lock(&memcg_oom_lock
);
1453 for_each_mem_cgroup_tree(iter
, memcg
) {
1454 if (iter
->oom_lock
) {
1456 * this subtree of our hierarchy is already locked
1457 * so we cannot give a lock.
1460 mem_cgroup_iter_break(memcg
, iter
);
1463 iter
->oom_lock
= true;
1468 * OK, we failed to lock the whole subtree so we have
1469 * to clean up what we set up to the failing subtree
1471 for_each_mem_cgroup_tree(iter
, memcg
) {
1472 if (iter
== failed
) {
1473 mem_cgroup_iter_break(memcg
, iter
);
1476 iter
->oom_lock
= false;
1479 mutex_acquire(&memcg_oom_lock_dep_map
, 0, 1, _RET_IP_
);
1481 spin_unlock(&memcg_oom_lock
);
1486 static void mem_cgroup_oom_unlock(struct mem_cgroup
*memcg
)
1488 struct mem_cgroup
*iter
;
1490 spin_lock(&memcg_oom_lock
);
1491 mutex_release(&memcg_oom_lock_dep_map
, 1, _RET_IP_
);
1492 for_each_mem_cgroup_tree(iter
, memcg
)
1493 iter
->oom_lock
= false;
1494 spin_unlock(&memcg_oom_lock
);
1497 static void mem_cgroup_mark_under_oom(struct mem_cgroup
*memcg
)
1499 struct mem_cgroup
*iter
;
1501 spin_lock(&memcg_oom_lock
);
1502 for_each_mem_cgroup_tree(iter
, memcg
)
1504 spin_unlock(&memcg_oom_lock
);
1507 static void mem_cgroup_unmark_under_oom(struct mem_cgroup
*memcg
)
1509 struct mem_cgroup
*iter
;
1512 * When a new child is created while the hierarchy is under oom,
1513 * mem_cgroup_oom_lock() may not be called. Watch for underflow.
1515 spin_lock(&memcg_oom_lock
);
1516 for_each_mem_cgroup_tree(iter
, memcg
)
1517 if (iter
->under_oom
> 0)
1519 spin_unlock(&memcg_oom_lock
);
1522 static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq
);
1524 struct oom_wait_info
{
1525 struct mem_cgroup
*memcg
;
1526 wait_queue_entry_t wait
;
1529 static int memcg_oom_wake_function(wait_queue_entry_t
*wait
,
1530 unsigned mode
, int sync
, void *arg
)
1532 struct mem_cgroup
*wake_memcg
= (struct mem_cgroup
*)arg
;
1533 struct mem_cgroup
*oom_wait_memcg
;
1534 struct oom_wait_info
*oom_wait_info
;
1536 oom_wait_info
= container_of(wait
, struct oom_wait_info
, wait
);
1537 oom_wait_memcg
= oom_wait_info
->memcg
;
1539 if (!mem_cgroup_is_descendant(wake_memcg
, oom_wait_memcg
) &&
1540 !mem_cgroup_is_descendant(oom_wait_memcg
, wake_memcg
))
1542 return autoremove_wake_function(wait
, mode
, sync
, arg
);
1545 static void memcg_oom_recover(struct mem_cgroup
*memcg
)
1548 * For the following lockless ->under_oom test, the only required
1549 * guarantee is that it must see the state asserted by an OOM when
1550 * this function is called as a result of userland actions
1551 * triggered by the notification of the OOM. This is trivially
1552 * achieved by invoking mem_cgroup_mark_under_oom() before
1553 * triggering notification.
1555 if (memcg
&& memcg
->under_oom
)
1556 __wake_up(&memcg_oom_waitq
, TASK_NORMAL
, 0, memcg
);
1559 static void mem_cgroup_oom(struct mem_cgroup
*memcg
, gfp_t mask
, int order
)
1561 if (!current
->memcg_may_oom
)
1564 * We are in the middle of the charge context here, so we
1565 * don't want to block when potentially sitting on a callstack
1566 * that holds all kinds of filesystem and mm locks.
1568 * Also, the caller may handle a failed allocation gracefully
1569 * (like optional page cache readahead) and so an OOM killer
1570 * invocation might not even be necessary.
1572 * That's why we don't do anything here except remember the
1573 * OOM context and then deal with it at the end of the page
1574 * fault when the stack is unwound, the locks are released,
1575 * and when we know whether the fault was overall successful.
1577 css_get(&memcg
->css
);
1578 current
->memcg_in_oom
= memcg
;
1579 current
->memcg_oom_gfp_mask
= mask
;
1580 current
->memcg_oom_order
= order
;
1584 * mem_cgroup_oom_synchronize - complete memcg OOM handling
1585 * @handle: actually kill/wait or just clean up the OOM state
1587 * This has to be called at the end of a page fault if the memcg OOM
1588 * handler was enabled.
1590 * Memcg supports userspace OOM handling where failed allocations must
1591 * sleep on a waitqueue until the userspace task resolves the
1592 * situation. Sleeping directly in the charge context with all kinds
1593 * of locks held is not a good idea, instead we remember an OOM state
1594 * in the task and mem_cgroup_oom_synchronize() has to be called at
1595 * the end of the page fault to complete the OOM handling.
1597 * Returns %true if an ongoing memcg OOM situation was detected and
1598 * completed, %false otherwise.
1600 bool mem_cgroup_oom_synchronize(bool handle
)
1602 struct mem_cgroup
*memcg
= current
->memcg_in_oom
;
1603 struct oom_wait_info owait
;
1606 /* OOM is global, do not handle */
1613 owait
.memcg
= memcg
;
1614 owait
.wait
.flags
= 0;
1615 owait
.wait
.func
= memcg_oom_wake_function
;
1616 owait
.wait
.private = current
;
1617 INIT_LIST_HEAD(&owait
.wait
.entry
);
1619 prepare_to_wait(&memcg_oom_waitq
, &owait
.wait
, TASK_KILLABLE
);
1620 mem_cgroup_mark_under_oom(memcg
);
1622 locked
= mem_cgroup_oom_trylock(memcg
);
1625 mem_cgroup_oom_notify(memcg
);
1627 if (locked
&& !memcg
->oom_kill_disable
) {
1628 mem_cgroup_unmark_under_oom(memcg
);
1629 finish_wait(&memcg_oom_waitq
, &owait
.wait
);
1630 mem_cgroup_out_of_memory(memcg
, current
->memcg_oom_gfp_mask
,
1631 current
->memcg_oom_order
);
1634 mem_cgroup_unmark_under_oom(memcg
);
1635 finish_wait(&memcg_oom_waitq
, &owait
.wait
);
1639 mem_cgroup_oom_unlock(memcg
);
1641 * There is no guarantee that an OOM-lock contender
1642 * sees the wakeups triggered by the OOM kill
1643 * uncharges. Wake any sleepers explicitely.
1645 memcg_oom_recover(memcg
);
1648 current
->memcg_in_oom
= NULL
;
1649 css_put(&memcg
->css
);
1654 * lock_page_memcg - lock a page->mem_cgroup binding
1657 * This function protects unlocked LRU pages from being moved to
1660 * It ensures lifetime of the returned memcg. Caller is responsible
1661 * for the lifetime of the page; __unlock_page_memcg() is available
1662 * when @page might get freed inside the locked section.
1664 struct mem_cgroup
*lock_page_memcg(struct page
*page
)
1666 struct mem_cgroup
*memcg
;
1667 unsigned long flags
;
1670 * The RCU lock is held throughout the transaction. The fast
1671 * path can get away without acquiring the memcg->move_lock
1672 * because page moving starts with an RCU grace period.
1674 * The RCU lock also protects the memcg from being freed when
1675 * the page state that is going to change is the only thing
1676 * preventing the page itself from being freed. E.g. writeback
1677 * doesn't hold a page reference and relies on PG_writeback to
1678 * keep off truncation, migration and so forth.
1682 if (mem_cgroup_disabled())
1685 memcg
= page
->mem_cgroup
;
1686 if (unlikely(!memcg
))
1689 if (atomic_read(&memcg
->moving_account
) <= 0)
1692 spin_lock_irqsave(&memcg
->move_lock
, flags
);
1693 if (memcg
!= page
->mem_cgroup
) {
1694 spin_unlock_irqrestore(&memcg
->move_lock
, flags
);
1699 * When charge migration first begins, we can have locked and
1700 * unlocked page stat updates happening concurrently. Track
1701 * the task who has the lock for unlock_page_memcg().
1703 memcg
->move_lock_task
= current
;
1704 memcg
->move_lock_flags
= flags
;
1708 EXPORT_SYMBOL(lock_page_memcg
);
1711 * __unlock_page_memcg - unlock and unpin a memcg
1714 * Unlock and unpin a memcg returned by lock_page_memcg().
1716 void __unlock_page_memcg(struct mem_cgroup
*memcg
)
1718 if (memcg
&& memcg
->move_lock_task
== current
) {
1719 unsigned long flags
= memcg
->move_lock_flags
;
1721 memcg
->move_lock_task
= NULL
;
1722 memcg
->move_lock_flags
= 0;
1724 spin_unlock_irqrestore(&memcg
->move_lock
, flags
);
1731 * unlock_page_memcg - unlock a page->mem_cgroup binding
1734 void unlock_page_memcg(struct page
*page
)
1736 __unlock_page_memcg(page
->mem_cgroup
);
1738 EXPORT_SYMBOL(unlock_page_memcg
);
1741 * size of first charge trial. "32" comes from vmscan.c's magic value.
1742 * TODO: maybe necessary to use big numbers in big irons.
1744 #define CHARGE_BATCH 32U
1745 struct memcg_stock_pcp
{
1746 struct mem_cgroup
*cached
; /* this never be root cgroup */
1747 unsigned int nr_pages
;
1748 struct work_struct work
;
1749 unsigned long flags
;
1750 #define FLUSHING_CACHED_CHARGE 0
1752 static DEFINE_PER_CPU(struct memcg_stock_pcp
, memcg_stock
);
1753 static DEFINE_MUTEX(percpu_charge_mutex
);
1756 * consume_stock: Try to consume stocked charge on this cpu.
1757 * @memcg: memcg to consume from.
1758 * @nr_pages: how many pages to charge.
1760 * The charges will only happen if @memcg matches the current cpu's memcg
1761 * stock, and at least @nr_pages are available in that stock. Failure to
1762 * service an allocation will refill the stock.
1764 * returns true if successful, false otherwise.
1766 static bool consume_stock(struct mem_cgroup
*memcg
, unsigned int nr_pages
)
1768 struct memcg_stock_pcp
*stock
;
1769 unsigned long flags
;
1772 if (nr_pages
> CHARGE_BATCH
)
1775 local_irq_save(flags
);
1777 stock
= this_cpu_ptr(&memcg_stock
);
1778 if (memcg
== stock
->cached
&& stock
->nr_pages
>= nr_pages
) {
1779 stock
->nr_pages
-= nr_pages
;
1783 local_irq_restore(flags
);
1789 * Returns stocks cached in percpu and reset cached information.
1791 static void drain_stock(struct memcg_stock_pcp
*stock
)
1793 struct mem_cgroup
*old
= stock
->cached
;
1795 if (stock
->nr_pages
) {
1796 page_counter_uncharge(&old
->memory
, stock
->nr_pages
);
1797 if (do_memsw_account())
1798 page_counter_uncharge(&old
->memsw
, stock
->nr_pages
);
1799 css_put_many(&old
->css
, stock
->nr_pages
);
1800 stock
->nr_pages
= 0;
1802 stock
->cached
= NULL
;
1805 static void drain_local_stock(struct work_struct
*dummy
)
1807 struct memcg_stock_pcp
*stock
;
1808 unsigned long flags
;
1811 * The only protection from memory hotplug vs. drain_stock races is
1812 * that we always operate on local CPU stock here with IRQ disabled
1814 local_irq_save(flags
);
1816 stock
= this_cpu_ptr(&memcg_stock
);
1818 clear_bit(FLUSHING_CACHED_CHARGE
, &stock
->flags
);
1820 local_irq_restore(flags
);
1824 * Cache charges(val) to local per_cpu area.
1825 * This will be consumed by consume_stock() function, later.
1827 static void refill_stock(struct mem_cgroup
*memcg
, unsigned int nr_pages
)
1829 struct memcg_stock_pcp
*stock
;
1830 unsigned long flags
;
1832 local_irq_save(flags
);
1834 stock
= this_cpu_ptr(&memcg_stock
);
1835 if (stock
->cached
!= memcg
) { /* reset if necessary */
1837 stock
->cached
= memcg
;
1839 stock
->nr_pages
+= nr_pages
;
1841 if (stock
->nr_pages
> CHARGE_BATCH
)
1844 local_irq_restore(flags
);
1848 * Drains all per-CPU charge caches for given root_memcg resp. subtree
1849 * of the hierarchy under it.
1851 static void drain_all_stock(struct mem_cgroup
*root_memcg
)
1855 /* If someone's already draining, avoid adding running more workers. */
1856 if (!mutex_trylock(&percpu_charge_mutex
))
1859 * Notify other cpus that system-wide "drain" is running
1860 * We do not care about races with the cpu hotplug because cpu down
1861 * as well as workers from this path always operate on the local
1862 * per-cpu data. CPU up doesn't touch memcg_stock at all.
1865 for_each_online_cpu(cpu
) {
1866 struct memcg_stock_pcp
*stock
= &per_cpu(memcg_stock
, cpu
);
1867 struct mem_cgroup
*memcg
;
1869 memcg
= stock
->cached
;
1870 if (!memcg
|| !stock
->nr_pages
|| !css_tryget(&memcg
->css
))
1872 if (!mem_cgroup_is_descendant(memcg
, root_memcg
)) {
1873 css_put(&memcg
->css
);
1876 if (!test_and_set_bit(FLUSHING_CACHED_CHARGE
, &stock
->flags
)) {
1878 drain_local_stock(&stock
->work
);
1880 schedule_work_on(cpu
, &stock
->work
);
1882 css_put(&memcg
->css
);
1885 mutex_unlock(&percpu_charge_mutex
);
1888 static int memcg_hotplug_cpu_dead(unsigned int cpu
)
1890 struct memcg_stock_pcp
*stock
;
1892 stock
= &per_cpu(memcg_stock
, cpu
);
1897 static void reclaim_high(struct mem_cgroup
*memcg
,
1898 unsigned int nr_pages
,
1902 if (page_counter_read(&memcg
->memory
) <= memcg
->high
)
1904 mem_cgroup_event(memcg
, MEMCG_HIGH
);
1905 try_to_free_mem_cgroup_pages(memcg
, nr_pages
, gfp_mask
, true);
1906 } while ((memcg
= parent_mem_cgroup(memcg
)));
1909 static void high_work_func(struct work_struct
*work
)
1911 struct mem_cgroup
*memcg
;
1913 memcg
= container_of(work
, struct mem_cgroup
, high_work
);
1914 reclaim_high(memcg
, CHARGE_BATCH
, GFP_KERNEL
);
1918 * Scheduled by try_charge() to be executed from the userland return path
1919 * and reclaims memory over the high limit.
1921 void mem_cgroup_handle_over_high(void)
1923 unsigned int nr_pages
= current
->memcg_nr_pages_over_high
;
1924 struct mem_cgroup
*memcg
;
1926 if (likely(!nr_pages
))
1929 memcg
= get_mem_cgroup_from_mm(current
->mm
);
1930 reclaim_high(memcg
, nr_pages
, GFP_KERNEL
);
1931 css_put(&memcg
->css
);
1932 current
->memcg_nr_pages_over_high
= 0;
1935 static int try_charge(struct mem_cgroup
*memcg
, gfp_t gfp_mask
,
1936 unsigned int nr_pages
)
1938 unsigned int batch
= max(CHARGE_BATCH
, nr_pages
);
1939 int nr_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
1940 struct mem_cgroup
*mem_over_limit
;
1941 struct page_counter
*counter
;
1942 unsigned long nr_reclaimed
;
1943 bool may_swap
= true;
1944 bool drained
= false;
1946 if (mem_cgroup_is_root(memcg
))
1949 if (consume_stock(memcg
, nr_pages
))
1952 if (!do_memsw_account() ||
1953 page_counter_try_charge(&memcg
->memsw
, batch
, &counter
)) {
1954 if (page_counter_try_charge(&memcg
->memory
, batch
, &counter
))
1956 if (do_memsw_account())
1957 page_counter_uncharge(&memcg
->memsw
, batch
);
1958 mem_over_limit
= mem_cgroup_from_counter(counter
, memory
);
1960 mem_over_limit
= mem_cgroup_from_counter(counter
, memsw
);
1964 if (batch
> nr_pages
) {
1970 * Memcg doesn't have a dedicated reserve for atomic
1971 * allocations. But like the global atomic pool, we need to
1972 * put the burden of reclaim on regular allocation requests
1973 * and let these go through as privileged allocations.
1975 if (gfp_mask
& __GFP_ATOMIC
)
1979 * Unlike in global OOM situations, memcg is not in a physical
1980 * memory shortage. Allow dying and OOM-killed tasks to
1981 * bypass the last charges so that they can exit quickly and
1982 * free their memory.
1984 if (unlikely(should_force_charge()))
1988 * Prevent unbounded recursion when reclaim operations need to
1989 * allocate memory. This might exceed the limits temporarily,
1990 * but we prefer facilitating memory reclaim and getting back
1991 * under the limit over triggering OOM kills in these cases.
1993 if (unlikely(current
->flags
& PF_MEMALLOC
))
1996 if (unlikely(task_in_memcg_oom(current
)))
1999 if (!gfpflags_allow_blocking(gfp_mask
))
2002 mem_cgroup_event(mem_over_limit
, MEMCG_MAX
);
2004 nr_reclaimed
= try_to_free_mem_cgroup_pages(mem_over_limit
, nr_pages
,
2005 gfp_mask
, may_swap
);
2007 if (mem_cgroup_margin(mem_over_limit
) >= nr_pages
)
2011 drain_all_stock(mem_over_limit
);
2016 if (gfp_mask
& __GFP_NORETRY
)
2019 * Even though the limit is exceeded at this point, reclaim
2020 * may have been able to free some pages. Retry the charge
2021 * before killing the task.
2023 * Only for regular pages, though: huge pages are rather
2024 * unlikely to succeed so close to the limit, and we fall back
2025 * to regular pages anyway in case of failure.
2027 if (nr_reclaimed
&& nr_pages
<= (1 << PAGE_ALLOC_COSTLY_ORDER
))
2030 * At task move, charge accounts can be doubly counted. So, it's
2031 * better to wait until the end of task_move if something is going on.
2033 if (mem_cgroup_wait_acct_move(mem_over_limit
))
2039 if (gfp_mask
& __GFP_NOFAIL
)
2042 if (fatal_signal_pending(current
))
2045 mem_cgroup_event(mem_over_limit
, MEMCG_OOM
);
2047 mem_cgroup_oom(mem_over_limit
, gfp_mask
,
2048 get_order(nr_pages
* PAGE_SIZE
));
2050 if (!(gfp_mask
& __GFP_NOFAIL
))
2054 * The allocation either can't fail or will lead to more memory
2055 * being freed very soon. Allow memory usage go over the limit
2056 * temporarily by force charging it.
2058 page_counter_charge(&memcg
->memory
, nr_pages
);
2059 if (do_memsw_account())
2060 page_counter_charge(&memcg
->memsw
, nr_pages
);
2061 css_get_many(&memcg
->css
, nr_pages
);
2066 css_get_many(&memcg
->css
, batch
);
2067 if (batch
> nr_pages
)
2068 refill_stock(memcg
, batch
- nr_pages
);
2071 * If the hierarchy is above the normal consumption range, schedule
2072 * reclaim on returning to userland. We can perform reclaim here
2073 * if __GFP_RECLAIM but let's always punt for simplicity and so that
2074 * GFP_KERNEL can consistently be used during reclaim. @memcg is
2075 * not recorded as it most likely matches current's and won't
2076 * change in the meantime. As high limit is checked again before
2077 * reclaim, the cost of mismatch is negligible.
2080 if (page_counter_read(&memcg
->memory
) > memcg
->high
) {
2081 /* Don't bother a random interrupted task */
2082 if (in_interrupt()) {
2083 schedule_work(&memcg
->high_work
);
2086 current
->memcg_nr_pages_over_high
+= batch
;
2087 set_notify_resume(current
);
2090 } while ((memcg
= parent_mem_cgroup(memcg
)));
2095 static void cancel_charge(struct mem_cgroup
*memcg
, unsigned int nr_pages
)
2097 if (mem_cgroup_is_root(memcg
))
2100 page_counter_uncharge(&memcg
->memory
, nr_pages
);
2101 if (do_memsw_account())
2102 page_counter_uncharge(&memcg
->memsw
, nr_pages
);
2104 css_put_many(&memcg
->css
, nr_pages
);
2107 static void lock_page_lru(struct page
*page
, int *isolated
)
2109 struct zone
*zone
= page_zone(page
);
2111 spin_lock_irq(zone_lru_lock(zone
));
2112 if (PageLRU(page
)) {
2113 struct lruvec
*lruvec
;
2115 lruvec
= mem_cgroup_page_lruvec(page
, zone
->zone_pgdat
);
2117 del_page_from_lru_list(page
, lruvec
, page_lru(page
));
2123 static void unlock_page_lru(struct page
*page
, int isolated
)
2125 struct zone
*zone
= page_zone(page
);
2128 struct lruvec
*lruvec
;
2130 lruvec
= mem_cgroup_page_lruvec(page
, zone
->zone_pgdat
);
2131 VM_BUG_ON_PAGE(PageLRU(page
), page
);
2133 add_page_to_lru_list(page
, lruvec
, page_lru(page
));
2135 spin_unlock_irq(zone_lru_lock(zone
));
2138 static void commit_charge(struct page
*page
, struct mem_cgroup
*memcg
,
2143 VM_BUG_ON_PAGE(page
->mem_cgroup
, page
);
2146 * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page
2147 * may already be on some other mem_cgroup's LRU. Take care of it.
2150 lock_page_lru(page
, &isolated
);
2153 * Nobody should be changing or seriously looking at
2154 * page->mem_cgroup at this point:
2156 * - the page is uncharged
2158 * - the page is off-LRU
2160 * - an anonymous fault has exclusive page access, except for
2161 * a locked page table
2163 * - a page cache insertion, a swapin fault, or a migration
2164 * have the page locked
2166 page
->mem_cgroup
= memcg
;
2169 unlock_page_lru(page
, isolated
);
2173 static int memcg_alloc_cache_id(void)
2178 id
= ida_simple_get(&memcg_cache_ida
,
2179 0, MEMCG_CACHES_MAX_SIZE
, GFP_KERNEL
);
2183 if (id
< memcg_nr_cache_ids
)
2187 * There's no space for the new id in memcg_caches arrays,
2188 * so we have to grow them.
2190 down_write(&memcg_cache_ids_sem
);
2192 size
= 2 * (id
+ 1);
2193 if (size
< MEMCG_CACHES_MIN_SIZE
)
2194 size
= MEMCG_CACHES_MIN_SIZE
;
2195 else if (size
> MEMCG_CACHES_MAX_SIZE
)
2196 size
= MEMCG_CACHES_MAX_SIZE
;
2198 err
= memcg_update_all_caches(size
);
2200 err
= memcg_update_all_list_lrus(size
);
2202 memcg_nr_cache_ids
= size
;
2204 up_write(&memcg_cache_ids_sem
);
2207 ida_simple_remove(&memcg_cache_ida
, id
);
2213 static void memcg_free_cache_id(int id
)
2215 ida_simple_remove(&memcg_cache_ida
, id
);
2218 struct memcg_kmem_cache_create_work
{
2219 struct mem_cgroup
*memcg
;
2220 struct kmem_cache
*cachep
;
2221 struct work_struct work
;
2224 static void memcg_kmem_cache_create_func(struct work_struct
*w
)
2226 struct memcg_kmem_cache_create_work
*cw
=
2227 container_of(w
, struct memcg_kmem_cache_create_work
, work
);
2228 struct mem_cgroup
*memcg
= cw
->memcg
;
2229 struct kmem_cache
*cachep
= cw
->cachep
;
2231 memcg_create_kmem_cache(memcg
, cachep
);
2233 css_put(&memcg
->css
);
2238 * Enqueue the creation of a per-memcg kmem_cache.
2240 static void __memcg_schedule_kmem_cache_create(struct mem_cgroup
*memcg
,
2241 struct kmem_cache
*cachep
)
2243 struct memcg_kmem_cache_create_work
*cw
;
2245 cw
= kmalloc(sizeof(*cw
), GFP_NOWAIT
| __GFP_NOWARN
);
2249 css_get(&memcg
->css
);
2252 cw
->cachep
= cachep
;
2253 INIT_WORK(&cw
->work
, memcg_kmem_cache_create_func
);
2255 queue_work(memcg_kmem_cache_wq
, &cw
->work
);
2258 static void memcg_schedule_kmem_cache_create(struct mem_cgroup
*memcg
,
2259 struct kmem_cache
*cachep
)
2262 * We need to stop accounting when we kmalloc, because if the
2263 * corresponding kmalloc cache is not yet created, the first allocation
2264 * in __memcg_schedule_kmem_cache_create will recurse.
2266 * However, it is better to enclose the whole function. Depending on
2267 * the debugging options enabled, INIT_WORK(), for instance, can
2268 * trigger an allocation. This too, will make us recurse. Because at
2269 * this point we can't allow ourselves back into memcg_kmem_get_cache,
2270 * the safest choice is to do it like this, wrapping the whole function.
2272 current
->memcg_kmem_skip_account
= 1;
2273 __memcg_schedule_kmem_cache_create(memcg
, cachep
);
2274 current
->memcg_kmem_skip_account
= 0;
2277 static inline bool memcg_kmem_bypass(void)
2279 if (in_interrupt() || !current
->mm
|| (current
->flags
& PF_KTHREAD
))
2285 * memcg_kmem_get_cache: select the correct per-memcg cache for allocation
2286 * @cachep: the original global kmem cache
2288 * Return the kmem_cache we're supposed to use for a slab allocation.
2289 * We try to use the current memcg's version of the cache.
2291 * If the cache does not exist yet, if we are the first user of it, we
2292 * create it asynchronously in a workqueue and let the current allocation
2293 * go through with the original cache.
2295 * This function takes a reference to the cache it returns to assure it
2296 * won't get destroyed while we are working with it. Once the caller is
2297 * done with it, memcg_kmem_put_cache() must be called to release the
2300 struct kmem_cache
*memcg_kmem_get_cache(struct kmem_cache
*cachep
)
2302 struct mem_cgroup
*memcg
;
2303 struct kmem_cache
*memcg_cachep
;
2306 VM_BUG_ON(!is_root_cache(cachep
));
2308 if (memcg_kmem_bypass())
2311 if (current
->memcg_kmem_skip_account
)
2314 memcg
= get_mem_cgroup_from_mm(current
->mm
);
2315 kmemcg_id
= READ_ONCE(memcg
->kmemcg_id
);
2319 memcg_cachep
= cache_from_memcg_idx(cachep
, kmemcg_id
);
2320 if (likely(memcg_cachep
))
2321 return memcg_cachep
;
2324 * If we are in a safe context (can wait, and not in interrupt
2325 * context), we could be be predictable and return right away.
2326 * This would guarantee that the allocation being performed
2327 * already belongs in the new cache.
2329 * However, there are some clashes that can arrive from locking.
2330 * For instance, because we acquire the slab_mutex while doing
2331 * memcg_create_kmem_cache, this means no further allocation
2332 * could happen with the slab_mutex held. So it's better to
2335 memcg_schedule_kmem_cache_create(memcg
, cachep
);
2337 css_put(&memcg
->css
);
2342 * memcg_kmem_put_cache: drop reference taken by memcg_kmem_get_cache
2343 * @cachep: the cache returned by memcg_kmem_get_cache
2345 void memcg_kmem_put_cache(struct kmem_cache
*cachep
)
2347 if (!is_root_cache(cachep
))
2348 css_put(&cachep
->memcg_params
.memcg
->css
);
2352 * memcg_kmem_charge: charge a kmem page
2353 * @page: page to charge
2354 * @gfp: reclaim mode
2355 * @order: allocation order
2356 * @memcg: memory cgroup to charge
2358 * Returns 0 on success, an error code on failure.
2360 int memcg_kmem_charge_memcg(struct page
*page
, gfp_t gfp
, int order
,
2361 struct mem_cgroup
*memcg
)
2363 unsigned int nr_pages
= 1 << order
;
2364 struct page_counter
*counter
;
2367 ret
= try_charge(memcg
, gfp
, nr_pages
);
2371 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys
) &&
2372 !page_counter_try_charge(&memcg
->kmem
, nr_pages
, &counter
)) {
2375 * Enforce __GFP_NOFAIL allocation because callers are not
2376 * prepared to see failures and likely do not have any failure
2379 if (gfp
& __GFP_NOFAIL
) {
2380 page_counter_charge(&memcg
->kmem
, nr_pages
);
2383 cancel_charge(memcg
, nr_pages
);
2387 page
->mem_cgroup
= memcg
;
2393 * memcg_kmem_charge: charge a kmem page to the current memory cgroup
2394 * @page: page to charge
2395 * @gfp: reclaim mode
2396 * @order: allocation order
2398 * Returns 0 on success, an error code on failure.
2400 int memcg_kmem_charge(struct page
*page
, gfp_t gfp
, int order
)
2402 struct mem_cgroup
*memcg
;
2405 if (mem_cgroup_disabled() || memcg_kmem_bypass())
2408 memcg
= get_mem_cgroup_from_mm(current
->mm
);
2409 if (!mem_cgroup_is_root(memcg
)) {
2410 ret
= memcg_kmem_charge_memcg(page
, gfp
, order
, memcg
);
2412 __SetPageKmemcg(page
);
2414 css_put(&memcg
->css
);
2418 * memcg_kmem_uncharge: uncharge a kmem page
2419 * @page: page to uncharge
2420 * @order: allocation order
2422 void memcg_kmem_uncharge(struct page
*page
, int order
)
2424 struct mem_cgroup
*memcg
= page
->mem_cgroup
;
2425 unsigned int nr_pages
= 1 << order
;
2430 VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg
), page
);
2432 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys
))
2433 page_counter_uncharge(&memcg
->kmem
, nr_pages
);
2435 page_counter_uncharge(&memcg
->memory
, nr_pages
);
2436 if (do_memsw_account())
2437 page_counter_uncharge(&memcg
->memsw
, nr_pages
);
2439 page
->mem_cgroup
= NULL
;
2441 /* slab pages do not have PageKmemcg flag set */
2442 if (PageKmemcg(page
))
2443 __ClearPageKmemcg(page
);
2445 css_put_many(&memcg
->css
, nr_pages
);
2447 #endif /* !CONFIG_SLOB */
2449 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
2452 * Because tail pages are not marked as "used", set it. We're under
2453 * zone_lru_lock and migration entries setup in all page mappings.
2455 void mem_cgroup_split_huge_fixup(struct page
*head
)
2459 if (mem_cgroup_disabled())
2462 for (i
= 1; i
< HPAGE_PMD_NR
; i
++)
2463 head
[i
].mem_cgroup
= head
->mem_cgroup
;
2465 __this_cpu_sub(head
->mem_cgroup
->stat
->count
[MEMCG_RSS_HUGE
],
2468 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
2470 #ifdef CONFIG_MEMCG_SWAP
2471 static void mem_cgroup_swap_statistics(struct mem_cgroup
*memcg
,
2474 this_cpu_add(memcg
->stat
->count
[MEMCG_SWAP
], nr_entries
);
2478 * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record.
2479 * @entry: swap entry to be moved
2480 * @from: mem_cgroup which the entry is moved from
2481 * @to: mem_cgroup which the entry is moved to
2483 * It succeeds only when the swap_cgroup's record for this entry is the same
2484 * as the mem_cgroup's id of @from.
2486 * Returns 0 on success, -EINVAL on failure.
2488 * The caller must have charged to @to, IOW, called page_counter_charge() about
2489 * both res and memsw, and called css_get().
2491 static int mem_cgroup_move_swap_account(swp_entry_t entry
,
2492 struct mem_cgroup
*from
, struct mem_cgroup
*to
)
2494 unsigned short old_id
, new_id
;
2496 old_id
= mem_cgroup_id(from
);
2497 new_id
= mem_cgroup_id(to
);
2499 if (swap_cgroup_cmpxchg(entry
, old_id
, new_id
) == old_id
) {
2500 mem_cgroup_swap_statistics(from
, -1);
2501 mem_cgroup_swap_statistics(to
, 1);
2507 static inline int mem_cgroup_move_swap_account(swp_entry_t entry
,
2508 struct mem_cgroup
*from
, struct mem_cgroup
*to
)
2514 static DEFINE_MUTEX(memcg_limit_mutex
);
2516 static int mem_cgroup_resize_limit(struct mem_cgroup
*memcg
,
2517 unsigned long limit
)
2519 unsigned long curusage
;
2520 unsigned long oldusage
;
2521 bool enlarge
= false;
2526 * For keeping hierarchical_reclaim simple, how long we should retry
2527 * is depends on callers. We set our retry-count to be function
2528 * of # of children which we should visit in this loop.
2530 retry_count
= MEM_CGROUP_RECLAIM_RETRIES
*
2531 mem_cgroup_count_children(memcg
);
2533 oldusage
= page_counter_read(&memcg
->memory
);
2536 if (signal_pending(current
)) {
2541 mutex_lock(&memcg_limit_mutex
);
2542 if (limit
> memcg
->memsw
.limit
) {
2543 mutex_unlock(&memcg_limit_mutex
);
2547 if (limit
> memcg
->memory
.limit
)
2549 ret
= page_counter_limit(&memcg
->memory
, limit
);
2550 mutex_unlock(&memcg_limit_mutex
);
2555 try_to_free_mem_cgroup_pages(memcg
, 1, GFP_KERNEL
, true);
2557 curusage
= page_counter_read(&memcg
->memory
);
2558 /* Usage is reduced ? */
2559 if (curusage
>= oldusage
)
2562 oldusage
= curusage
;
2563 } while (retry_count
);
2565 if (!ret
&& enlarge
)
2566 memcg_oom_recover(memcg
);
2571 static int mem_cgroup_resize_memsw_limit(struct mem_cgroup
*memcg
,
2572 unsigned long limit
)
2574 unsigned long curusage
;
2575 unsigned long oldusage
;
2576 bool enlarge
= false;
2580 /* see mem_cgroup_resize_res_limit */
2581 retry_count
= MEM_CGROUP_RECLAIM_RETRIES
*
2582 mem_cgroup_count_children(memcg
);
2584 oldusage
= page_counter_read(&memcg
->memsw
);
2587 if (signal_pending(current
)) {
2592 mutex_lock(&memcg_limit_mutex
);
2593 if (limit
< memcg
->memory
.limit
) {
2594 mutex_unlock(&memcg_limit_mutex
);
2598 if (limit
> memcg
->memsw
.limit
)
2600 ret
= page_counter_limit(&memcg
->memsw
, limit
);
2601 mutex_unlock(&memcg_limit_mutex
);
2606 try_to_free_mem_cgroup_pages(memcg
, 1, GFP_KERNEL
, false);
2608 curusage
= page_counter_read(&memcg
->memsw
);
2609 /* Usage is reduced ? */
2610 if (curusage
>= oldusage
)
2613 oldusage
= curusage
;
2614 } while (retry_count
);
2616 if (!ret
&& enlarge
)
2617 memcg_oom_recover(memcg
);
2622 unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t
*pgdat
, int order
,
2624 unsigned long *total_scanned
)
2626 unsigned long nr_reclaimed
= 0;
2627 struct mem_cgroup_per_node
*mz
, *next_mz
= NULL
;
2628 unsigned long reclaimed
;
2630 struct mem_cgroup_tree_per_node
*mctz
;
2631 unsigned long excess
;
2632 unsigned long nr_scanned
;
2637 mctz
= soft_limit_tree_node(pgdat
->node_id
);
2640 * Do not even bother to check the largest node if the root
2641 * is empty. Do it lockless to prevent lock bouncing. Races
2642 * are acceptable as soft limit is best effort anyway.
2644 if (!mctz
|| RB_EMPTY_ROOT(&mctz
->rb_root
))
2648 * This loop can run a while, specially if mem_cgroup's continuously
2649 * keep exceeding their soft limit and putting the system under
2656 mz
= mem_cgroup_largest_soft_limit_node(mctz
);
2661 reclaimed
= mem_cgroup_soft_reclaim(mz
->memcg
, pgdat
,
2662 gfp_mask
, &nr_scanned
);
2663 nr_reclaimed
+= reclaimed
;
2664 *total_scanned
+= nr_scanned
;
2665 spin_lock_irq(&mctz
->lock
);
2666 __mem_cgroup_remove_exceeded(mz
, mctz
);
2669 * If we failed to reclaim anything from this memory cgroup
2670 * it is time to move on to the next cgroup
2674 next_mz
= __mem_cgroup_largest_soft_limit_node(mctz
);
2676 excess
= soft_limit_excess(mz
->memcg
);
2678 * One school of thought says that we should not add
2679 * back the node to the tree if reclaim returns 0.
2680 * But our reclaim could return 0, simply because due
2681 * to priority we are exposing a smaller subset of
2682 * memory to reclaim from. Consider this as a longer
2685 /* If excess == 0, no tree ops */
2686 __mem_cgroup_insert_exceeded(mz
, mctz
, excess
);
2687 spin_unlock_irq(&mctz
->lock
);
2688 css_put(&mz
->memcg
->css
);
2691 * Could not reclaim anything and there are no more
2692 * mem cgroups to try or we seem to be looping without
2693 * reclaiming anything.
2695 if (!nr_reclaimed
&&
2697 loop
> MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS
))
2699 } while (!nr_reclaimed
);
2701 css_put(&next_mz
->memcg
->css
);
2702 return nr_reclaimed
;
2706 * Test whether @memcg has children, dead or alive. Note that this
2707 * function doesn't care whether @memcg has use_hierarchy enabled and
2708 * returns %true if there are child csses according to the cgroup
2709 * hierarchy. Testing use_hierarchy is the caller's responsiblity.
2711 static inline bool memcg_has_children(struct mem_cgroup
*memcg
)
2716 ret
= css_next_child(NULL
, &memcg
->css
);
2722 * Reclaims as many pages from the given memcg as possible.
2724 * Caller is responsible for holding css reference for memcg.
2726 static int mem_cgroup_force_empty(struct mem_cgroup
*memcg
)
2728 int nr_retries
= MEM_CGROUP_RECLAIM_RETRIES
;
2730 /* we call try-to-free pages for make this cgroup empty */
2731 lru_add_drain_all();
2732 /* try to free all pages in this cgroup */
2733 while (nr_retries
&& page_counter_read(&memcg
->memory
)) {
2736 if (signal_pending(current
))
2739 progress
= try_to_free_mem_cgroup_pages(memcg
, 1,
2743 /* maybe some writeback is necessary */
2744 congestion_wait(BLK_RW_ASYNC
, HZ
/10);
2752 static ssize_t
mem_cgroup_force_empty_write(struct kernfs_open_file
*of
,
2753 char *buf
, size_t nbytes
,
2756 struct mem_cgroup
*memcg
= mem_cgroup_from_css(of_css(of
));
2758 if (mem_cgroup_is_root(memcg
))
2760 return mem_cgroup_force_empty(memcg
) ?: nbytes
;
2763 static u64
mem_cgroup_hierarchy_read(struct cgroup_subsys_state
*css
,
2766 return mem_cgroup_from_css(css
)->use_hierarchy
;
2769 static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state
*css
,
2770 struct cftype
*cft
, u64 val
)
2773 struct mem_cgroup
*memcg
= mem_cgroup_from_css(css
);
2774 struct mem_cgroup
*parent_memcg
= mem_cgroup_from_css(memcg
->css
.parent
);
2776 if (memcg
->use_hierarchy
== val
)
2780 * If parent's use_hierarchy is set, we can't make any modifications
2781 * in the child subtrees. If it is unset, then the change can
2782 * occur, provided the current cgroup has no children.
2784 * For the root cgroup, parent_mem is NULL, we allow value to be
2785 * set if there are no children.
2787 if ((!parent_memcg
|| !parent_memcg
->use_hierarchy
) &&
2788 (val
== 1 || val
== 0)) {
2789 if (!memcg_has_children(memcg
))
2790 memcg
->use_hierarchy
= val
;
2799 static void tree_stat(struct mem_cgroup
*memcg
, unsigned long *stat
)
2801 struct mem_cgroup
*iter
;
2804 memset(stat
, 0, sizeof(*stat
) * MEMCG_NR_STAT
);
2806 for_each_mem_cgroup_tree(iter
, memcg
) {
2807 for (i
= 0; i
< MEMCG_NR_STAT
; i
++)
2808 stat
[i
] += memcg_page_state(iter
, i
);
2812 static void tree_events(struct mem_cgroup
*memcg
, unsigned long *events
)
2814 struct mem_cgroup
*iter
;
2817 memset(events
, 0, sizeof(*events
) * MEMCG_NR_EVENTS
);
2819 for_each_mem_cgroup_tree(iter
, memcg
) {
2820 for (i
= 0; i
< MEMCG_NR_EVENTS
; i
++)
2821 events
[i
] += memcg_sum_events(iter
, i
);
2825 static unsigned long mem_cgroup_usage(struct mem_cgroup
*memcg
, bool swap
)
2827 unsigned long val
= 0;
2829 if (mem_cgroup_is_root(memcg
)) {
2830 struct mem_cgroup
*iter
;
2832 for_each_mem_cgroup_tree(iter
, memcg
) {
2833 val
+= memcg_page_state(iter
, MEMCG_CACHE
);
2834 val
+= memcg_page_state(iter
, MEMCG_RSS
);
2836 val
+= memcg_page_state(iter
, MEMCG_SWAP
);
2840 val
= page_counter_read(&memcg
->memory
);
2842 val
= page_counter_read(&memcg
->memsw
);
2855 static u64
mem_cgroup_read_u64(struct cgroup_subsys_state
*css
,
2858 struct mem_cgroup
*memcg
= mem_cgroup_from_css(css
);
2859 struct page_counter
*counter
;
2861 switch (MEMFILE_TYPE(cft
->private)) {
2863 counter
= &memcg
->memory
;
2866 counter
= &memcg
->memsw
;
2869 counter
= &memcg
->kmem
;
2872 counter
= &memcg
->tcpmem
;
2878 switch (MEMFILE_ATTR(cft
->private)) {
2880 if (counter
== &memcg
->memory
)
2881 return (u64
)mem_cgroup_usage(memcg
, false) * PAGE_SIZE
;
2882 if (counter
== &memcg
->memsw
)
2883 return (u64
)mem_cgroup_usage(memcg
, true) * PAGE_SIZE
;
2884 return (u64
)page_counter_read(counter
) * PAGE_SIZE
;
2886 return (u64
)counter
->limit
* PAGE_SIZE
;
2888 return (u64
)counter
->watermark
* PAGE_SIZE
;
2890 return counter
->failcnt
;
2891 case RES_SOFT_LIMIT
:
2892 return (u64
)memcg
->soft_limit
* PAGE_SIZE
;
2899 static int memcg_online_kmem(struct mem_cgroup
*memcg
)
2903 if (cgroup_memory_nokmem
)
2906 BUG_ON(memcg
->kmemcg_id
>= 0);
2907 BUG_ON(memcg
->kmem_state
);
2909 memcg_id
= memcg_alloc_cache_id();
2913 static_branch_inc(&memcg_kmem_enabled_key
);
2915 * A memory cgroup is considered kmem-online as soon as it gets
2916 * kmemcg_id. Setting the id after enabling static branching will
2917 * guarantee no one starts accounting before all call sites are
2920 memcg
->kmemcg_id
= memcg_id
;
2921 memcg
->kmem_state
= KMEM_ONLINE
;
2922 INIT_LIST_HEAD(&memcg
->kmem_caches
);
2927 static void memcg_offline_kmem(struct mem_cgroup
*memcg
)
2929 struct cgroup_subsys_state
*css
;
2930 struct mem_cgroup
*parent
, *child
;
2933 if (memcg
->kmem_state
!= KMEM_ONLINE
)
2936 * Clear the online state before clearing memcg_caches array
2937 * entries. The slab_mutex in memcg_deactivate_kmem_caches()
2938 * guarantees that no cache will be created for this cgroup
2939 * after we are done (see memcg_create_kmem_cache()).
2941 memcg
->kmem_state
= KMEM_ALLOCATED
;
2943 memcg_deactivate_kmem_caches(memcg
);
2945 kmemcg_id
= memcg
->kmemcg_id
;
2946 BUG_ON(kmemcg_id
< 0);
2948 parent
= parent_mem_cgroup(memcg
);
2950 parent
= root_mem_cgroup
;
2953 * Change kmemcg_id of this cgroup and all its descendants to the
2954 * parent's id, and then move all entries from this cgroup's list_lrus
2955 * to ones of the parent. After we have finished, all list_lrus
2956 * corresponding to this cgroup are guaranteed to remain empty. The
2957 * ordering is imposed by list_lru_node->lock taken by
2958 * memcg_drain_all_list_lrus().
2960 rcu_read_lock(); /* can be called from css_free w/o cgroup_mutex */
2961 css_for_each_descendant_pre(css
, &memcg
->css
) {
2962 child
= mem_cgroup_from_css(css
);
2963 BUG_ON(child
->kmemcg_id
!= kmemcg_id
);
2964 child
->kmemcg_id
= parent
->kmemcg_id
;
2965 if (!memcg
->use_hierarchy
)
2970 memcg_drain_all_list_lrus(kmemcg_id
, parent
->kmemcg_id
);
2972 memcg_free_cache_id(kmemcg_id
);
2975 static void memcg_free_kmem(struct mem_cgroup
*memcg
)
2977 /* css_alloc() failed, offlining didn't happen */
2978 if (unlikely(memcg
->kmem_state
== KMEM_ONLINE
))
2979 memcg_offline_kmem(memcg
);
2981 if (memcg
->kmem_state
== KMEM_ALLOCATED
) {
2982 memcg_destroy_kmem_caches(memcg
);
2983 static_branch_dec(&memcg_kmem_enabled_key
);
2984 WARN_ON(page_counter_read(&memcg
->kmem
));
2988 static int memcg_online_kmem(struct mem_cgroup
*memcg
)
2992 static void memcg_offline_kmem(struct mem_cgroup
*memcg
)
2995 static void memcg_free_kmem(struct mem_cgroup
*memcg
)
2998 #endif /* !CONFIG_SLOB */
3000 static int memcg_update_kmem_limit(struct mem_cgroup
*memcg
,
3001 unsigned long limit
)
3005 mutex_lock(&memcg_limit_mutex
);
3006 ret
= page_counter_limit(&memcg
->kmem
, limit
);
3007 mutex_unlock(&memcg_limit_mutex
);
3011 static int memcg_update_tcp_limit(struct mem_cgroup
*memcg
, unsigned long limit
)
3015 mutex_lock(&memcg_limit_mutex
);
3017 ret
= page_counter_limit(&memcg
->tcpmem
, limit
);
3021 if (!memcg
->tcpmem_active
) {
3023 * The active flag needs to be written after the static_key
3024 * update. This is what guarantees that the socket activation
3025 * function is the last one to run. See mem_cgroup_sk_alloc()
3026 * for details, and note that we don't mark any socket as
3027 * belonging to this memcg until that flag is up.
3029 * We need to do this, because static_keys will span multiple
3030 * sites, but we can't control their order. If we mark a socket
3031 * as accounted, but the accounting functions are not patched in
3032 * yet, we'll lose accounting.
3034 * We never race with the readers in mem_cgroup_sk_alloc(),
3035 * because when this value change, the code to process it is not
3038 static_branch_inc(&memcg_sockets_enabled_key
);
3039 memcg
->tcpmem_active
= true;
3042 mutex_unlock(&memcg_limit_mutex
);
3047 * The user of this function is...
3050 static ssize_t
mem_cgroup_write(struct kernfs_open_file
*of
,
3051 char *buf
, size_t nbytes
, loff_t off
)
3053 struct mem_cgroup
*memcg
= mem_cgroup_from_css(of_css(of
));
3054 unsigned long nr_pages
;
3057 buf
= strstrip(buf
);
3058 ret
= page_counter_memparse(buf
, "-1", &nr_pages
);
3062 switch (MEMFILE_ATTR(of_cft(of
)->private)) {
3064 if (mem_cgroup_is_root(memcg
)) { /* Can't set limit on root */
3068 switch (MEMFILE_TYPE(of_cft(of
)->private)) {
3070 ret
= mem_cgroup_resize_limit(memcg
, nr_pages
);
3073 ret
= mem_cgroup_resize_memsw_limit(memcg
, nr_pages
);
3076 ret
= memcg_update_kmem_limit(memcg
, nr_pages
);
3079 ret
= memcg_update_tcp_limit(memcg
, nr_pages
);
3083 case RES_SOFT_LIMIT
:
3084 memcg
->soft_limit
= nr_pages
;
3088 return ret
?: nbytes
;
3091 static ssize_t
mem_cgroup_reset(struct kernfs_open_file
*of
, char *buf
,
3092 size_t nbytes
, loff_t off
)
3094 struct mem_cgroup
*memcg
= mem_cgroup_from_css(of_css(of
));
3095 struct page_counter
*counter
;
3097 switch (MEMFILE_TYPE(of_cft(of
)->private)) {
3099 counter
= &memcg
->memory
;
3102 counter
= &memcg
->memsw
;
3105 counter
= &memcg
->kmem
;
3108 counter
= &memcg
->tcpmem
;
3114 switch (MEMFILE_ATTR(of_cft(of
)->private)) {
3116 page_counter_reset_watermark(counter
);
3119 counter
->failcnt
= 0;
3128 static u64
mem_cgroup_move_charge_read(struct cgroup_subsys_state
*css
,
3131 return mem_cgroup_from_css(css
)->move_charge_at_immigrate
;
3135 static int mem_cgroup_move_charge_write(struct cgroup_subsys_state
*css
,
3136 struct cftype
*cft
, u64 val
)
3138 struct mem_cgroup
*memcg
= mem_cgroup_from_css(css
);
3140 if (val
& ~MOVE_MASK
)
3144 * No kind of locking is needed in here, because ->can_attach() will
3145 * check this value once in the beginning of the process, and then carry
3146 * on with stale data. This means that changes to this value will only
3147 * affect task migrations starting after the change.
3149 memcg
->move_charge_at_immigrate
= val
;
3153 static int mem_cgroup_move_charge_write(struct cgroup_subsys_state
*css
,
3154 struct cftype
*cft
, u64 val
)
3161 static int memcg_numa_stat_show(struct seq_file
*m
, void *v
)
3165 unsigned int lru_mask
;
3168 static const struct numa_stat stats
[] = {
3169 { "total", LRU_ALL
},
3170 { "file", LRU_ALL_FILE
},
3171 { "anon", LRU_ALL_ANON
},
3172 { "unevictable", BIT(LRU_UNEVICTABLE
) },
3174 const struct numa_stat
*stat
;
3177 struct mem_cgroup
*memcg
= mem_cgroup_from_css(seq_css(m
));
3179 for (stat
= stats
; stat
< stats
+ ARRAY_SIZE(stats
); stat
++) {
3180 nr
= mem_cgroup_nr_lru_pages(memcg
, stat
->lru_mask
);
3181 seq_printf(m
, "%s=%lu", stat
->name
, nr
);
3182 for_each_node_state(nid
, N_MEMORY
) {
3183 nr
= mem_cgroup_node_nr_lru_pages(memcg
, nid
,
3185 seq_printf(m
, " N%d=%lu", nid
, nr
);
3190 for (stat
= stats
; stat
< stats
+ ARRAY_SIZE(stats
); stat
++) {
3191 struct mem_cgroup
*iter
;
3194 for_each_mem_cgroup_tree(iter
, memcg
)
3195 nr
+= mem_cgroup_nr_lru_pages(iter
, stat
->lru_mask
);
3196 seq_printf(m
, "hierarchical_%s=%lu", stat
->name
, nr
);
3197 for_each_node_state(nid
, N_MEMORY
) {
3199 for_each_mem_cgroup_tree(iter
, memcg
)
3200 nr
+= mem_cgroup_node_nr_lru_pages(
3201 iter
, nid
, stat
->lru_mask
);
3202 seq_printf(m
, " N%d=%lu", nid
, nr
);
3209 #endif /* CONFIG_NUMA */
3211 /* Universal VM events cgroup1 shows, original sort order */
3212 unsigned int memcg1_events
[] = {
3219 static const char *const memcg1_event_names
[] = {
3226 static int memcg_stat_show(struct seq_file
*m
, void *v
)
3228 struct mem_cgroup
*memcg
= mem_cgroup_from_css(seq_css(m
));
3229 unsigned long memory
, memsw
;
3230 struct mem_cgroup
*mi
;
3233 BUILD_BUG_ON(ARRAY_SIZE(memcg1_stat_names
) != ARRAY_SIZE(memcg1_stats
));
3234 BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names
) != NR_LRU_LISTS
);
3236 for (i
= 0; i
< ARRAY_SIZE(memcg1_stats
); i
++) {
3237 if (memcg1_stats
[i
] == MEMCG_SWAP
&& !do_memsw_account())
3239 seq_printf(m
, "%s %lu\n", memcg1_stat_names
[i
],
3240 memcg_page_state(memcg
, memcg1_stats
[i
]) *
3244 for (i
= 0; i
< ARRAY_SIZE(memcg1_events
); i
++)
3245 seq_printf(m
, "%s %lu\n", memcg1_event_names
[i
],
3246 memcg_sum_events(memcg
, memcg1_events
[i
]));
3248 for (i
= 0; i
< NR_LRU_LISTS
; i
++)
3249 seq_printf(m
, "%s %lu\n", mem_cgroup_lru_names
[i
],
3250 mem_cgroup_nr_lru_pages(memcg
, BIT(i
)) * PAGE_SIZE
);
3252 /* Hierarchical information */
3253 memory
= memsw
= PAGE_COUNTER_MAX
;
3254 for (mi
= memcg
; mi
; mi
= parent_mem_cgroup(mi
)) {
3255 memory
= min(memory
, mi
->memory
.limit
);
3256 memsw
= min(memsw
, mi
->memsw
.limit
);
3258 seq_printf(m
, "hierarchical_memory_limit %llu\n",
3259 (u64
)memory
* PAGE_SIZE
);
3260 if (do_memsw_account())
3261 seq_printf(m
, "hierarchical_memsw_limit %llu\n",
3262 (u64
)memsw
* PAGE_SIZE
);
3264 for (i
= 0; i
< ARRAY_SIZE(memcg1_stats
); i
++) {
3265 unsigned long long val
= 0;
3267 if (memcg1_stats
[i
] == MEMCG_SWAP
&& !do_memsw_account())
3269 for_each_mem_cgroup_tree(mi
, memcg
)
3270 val
+= memcg_page_state(mi
, memcg1_stats
[i
]) *
3272 seq_printf(m
, "total_%s %llu\n", memcg1_stat_names
[i
], val
);
3275 for (i
= 0; i
< ARRAY_SIZE(memcg1_events
); i
++) {
3276 unsigned long long val
= 0;
3278 for_each_mem_cgroup_tree(mi
, memcg
)
3279 val
+= memcg_sum_events(mi
, memcg1_events
[i
]);
3280 seq_printf(m
, "total_%s %llu\n", memcg1_event_names
[i
], val
);
3283 for (i
= 0; i
< NR_LRU_LISTS
; i
++) {
3284 unsigned long long val
= 0;
3286 for_each_mem_cgroup_tree(mi
, memcg
)
3287 val
+= mem_cgroup_nr_lru_pages(mi
, BIT(i
)) * PAGE_SIZE
;
3288 seq_printf(m
, "total_%s %llu\n", mem_cgroup_lru_names
[i
], val
);
3291 #ifdef CONFIG_DEBUG_VM
3294 struct mem_cgroup_per_node
*mz
;
3295 struct zone_reclaim_stat
*rstat
;
3296 unsigned long recent_rotated
[2] = {0, 0};
3297 unsigned long recent_scanned
[2] = {0, 0};
3299 for_each_online_pgdat(pgdat
) {
3300 mz
= mem_cgroup_nodeinfo(memcg
, pgdat
->node_id
);
3301 rstat
= &mz
->lruvec
.reclaim_stat
;
3303 recent_rotated
[0] += rstat
->recent_rotated
[0];
3304 recent_rotated
[1] += rstat
->recent_rotated
[1];
3305 recent_scanned
[0] += rstat
->recent_scanned
[0];
3306 recent_scanned
[1] += rstat
->recent_scanned
[1];
3308 seq_printf(m
, "recent_rotated_anon %lu\n", recent_rotated
[0]);
3309 seq_printf(m
, "recent_rotated_file %lu\n", recent_rotated
[1]);
3310 seq_printf(m
, "recent_scanned_anon %lu\n", recent_scanned
[0]);
3311 seq_printf(m
, "recent_scanned_file %lu\n", recent_scanned
[1]);
3318 static u64
mem_cgroup_swappiness_read(struct cgroup_subsys_state
*css
,
3321 struct mem_cgroup
*memcg
= mem_cgroup_from_css(css
);
3323 return mem_cgroup_swappiness(memcg
);
3326 static int mem_cgroup_swappiness_write(struct cgroup_subsys_state
*css
,
3327 struct cftype
*cft
, u64 val
)
3329 struct mem_cgroup
*memcg
= mem_cgroup_from_css(css
);
3335 memcg
->swappiness
= val
;
3337 vm_swappiness
= val
;
3342 static void __mem_cgroup_threshold(struct mem_cgroup
*memcg
, bool swap
)
3344 struct mem_cgroup_threshold_ary
*t
;
3345 unsigned long usage
;
3350 t
= rcu_dereference(memcg
->thresholds
.primary
);
3352 t
= rcu_dereference(memcg
->memsw_thresholds
.primary
);
3357 usage
= mem_cgroup_usage(memcg
, swap
);
3360 * current_threshold points to threshold just below or equal to usage.
3361 * If it's not true, a threshold was crossed after last
3362 * call of __mem_cgroup_threshold().
3364 i
= t
->current_threshold
;
3367 * Iterate backward over array of thresholds starting from
3368 * current_threshold and check if a threshold is crossed.
3369 * If none of thresholds below usage is crossed, we read
3370 * only one element of the array here.
3372 for (; i
>= 0 && unlikely(t
->entries
[i
].threshold
> usage
); i
--)
3373 eventfd_signal(t
->entries
[i
].eventfd
, 1);
3375 /* i = current_threshold + 1 */
3379 * Iterate forward over array of thresholds starting from
3380 * current_threshold+1 and check if a threshold is crossed.
3381 * If none of thresholds above usage is crossed, we read
3382 * only one element of the array here.
3384 for (; i
< t
->size
&& unlikely(t
->entries
[i
].threshold
<= usage
); i
++)
3385 eventfd_signal(t
->entries
[i
].eventfd
, 1);
3387 /* Update current_threshold */
3388 t
->current_threshold
= i
- 1;
3393 static void mem_cgroup_threshold(struct mem_cgroup
*memcg
)
3396 __mem_cgroup_threshold(memcg
, false);
3397 if (do_memsw_account())
3398 __mem_cgroup_threshold(memcg
, true);
3400 memcg
= parent_mem_cgroup(memcg
);
3404 static int compare_thresholds(const void *a
, const void *b
)
3406 const struct mem_cgroup_threshold
*_a
= a
;
3407 const struct mem_cgroup_threshold
*_b
= b
;
3409 if (_a
->threshold
> _b
->threshold
)
3412 if (_a
->threshold
< _b
->threshold
)
3418 static int mem_cgroup_oom_notify_cb(struct mem_cgroup
*memcg
)
3420 struct mem_cgroup_eventfd_list
*ev
;
3422 spin_lock(&memcg_oom_lock
);
3424 list_for_each_entry(ev
, &memcg
->oom_notify
, list
)
3425 eventfd_signal(ev
->eventfd
, 1);
3427 spin_unlock(&memcg_oom_lock
);
3431 static void mem_cgroup_oom_notify(struct mem_cgroup
*memcg
)
3433 struct mem_cgroup
*iter
;
3435 for_each_mem_cgroup_tree(iter
, memcg
)
3436 mem_cgroup_oom_notify_cb(iter
);
3439 static int __mem_cgroup_usage_register_event(struct mem_cgroup
*memcg
,
3440 struct eventfd_ctx
*eventfd
, const char *args
, enum res_type type
)
3442 struct mem_cgroup_thresholds
*thresholds
;
3443 struct mem_cgroup_threshold_ary
*new;
3444 unsigned long threshold
;
3445 unsigned long usage
;
3448 ret
= page_counter_memparse(args
, "-1", &threshold
);
3452 mutex_lock(&memcg
->thresholds_lock
);
3455 thresholds
= &memcg
->thresholds
;
3456 usage
= mem_cgroup_usage(memcg
, false);
3457 } else if (type
== _MEMSWAP
) {
3458 thresholds
= &memcg
->memsw_thresholds
;
3459 usage
= mem_cgroup_usage(memcg
, true);
3463 /* Check if a threshold crossed before adding a new one */
3464 if (thresholds
->primary
)
3465 __mem_cgroup_threshold(memcg
, type
== _MEMSWAP
);
3467 size
= thresholds
->primary
? thresholds
->primary
->size
+ 1 : 1;
3469 /* Allocate memory for new array of thresholds */
3470 new = kmalloc(sizeof(*new) + size
* sizeof(struct mem_cgroup_threshold
),
3478 /* Copy thresholds (if any) to new array */
3479 if (thresholds
->primary
) {
3480 memcpy(new->entries
, thresholds
->primary
->entries
, (size
- 1) *
3481 sizeof(struct mem_cgroup_threshold
));
3484 /* Add new threshold */
3485 new->entries
[size
- 1].eventfd
= eventfd
;
3486 new->entries
[size
- 1].threshold
= threshold
;
3488 /* Sort thresholds. Registering of new threshold isn't time-critical */
3489 sort(new->entries
, size
, sizeof(struct mem_cgroup_threshold
),
3490 compare_thresholds
, NULL
);
3492 /* Find current threshold */
3493 new->current_threshold
= -1;
3494 for (i
= 0; i
< size
; i
++) {
3495 if (new->entries
[i
].threshold
<= usage
) {
3497 * new->current_threshold will not be used until
3498 * rcu_assign_pointer(), so it's safe to increment
3501 ++new->current_threshold
;
3506 /* Free old spare buffer and save old primary buffer as spare */
3507 kfree(thresholds
->spare
);
3508 thresholds
->spare
= thresholds
->primary
;
3510 rcu_assign_pointer(thresholds
->primary
, new);
3512 /* To be sure that nobody uses thresholds */
3516 mutex_unlock(&memcg
->thresholds_lock
);
3521 static int mem_cgroup_usage_register_event(struct mem_cgroup
*memcg
,
3522 struct eventfd_ctx
*eventfd
, const char *args
)
3524 return __mem_cgroup_usage_register_event(memcg
, eventfd
, args
, _MEM
);
3527 static int memsw_cgroup_usage_register_event(struct mem_cgroup
*memcg
,
3528 struct eventfd_ctx
*eventfd
, const char *args
)
3530 return __mem_cgroup_usage_register_event(memcg
, eventfd
, args
, _MEMSWAP
);
3533 static void __mem_cgroup_usage_unregister_event(struct mem_cgroup
*memcg
,
3534 struct eventfd_ctx
*eventfd
, enum res_type type
)
3536 struct mem_cgroup_thresholds
*thresholds
;
3537 struct mem_cgroup_threshold_ary
*new;
3538 unsigned long usage
;
3541 mutex_lock(&memcg
->thresholds_lock
);
3544 thresholds
= &memcg
->thresholds
;
3545 usage
= mem_cgroup_usage(memcg
, false);
3546 } else if (type
== _MEMSWAP
) {
3547 thresholds
= &memcg
->memsw_thresholds
;
3548 usage
= mem_cgroup_usage(memcg
, true);
3552 if (!thresholds
->primary
)
3555 /* Check if a threshold crossed before removing */
3556 __mem_cgroup_threshold(memcg
, type
== _MEMSWAP
);
3558 /* Calculate new number of threshold */
3560 for (i
= 0; i
< thresholds
->primary
->size
; i
++) {
3561 if (thresholds
->primary
->entries
[i
].eventfd
!= eventfd
)
3565 new = thresholds
->spare
;
3567 /* Set thresholds array to NULL if we don't have thresholds */
3576 /* Copy thresholds and find current threshold */
3577 new->current_threshold
= -1;
3578 for (i
= 0, j
= 0; i
< thresholds
->primary
->size
; i
++) {
3579 if (thresholds
->primary
->entries
[i
].eventfd
== eventfd
)
3582 new->entries
[j
] = thresholds
->primary
->entries
[i
];
3583 if (new->entries
[j
].threshold
<= usage
) {
3585 * new->current_threshold will not be used
3586 * until rcu_assign_pointer(), so it's safe to increment
3589 ++new->current_threshold
;
3595 /* Swap primary and spare array */
3596 thresholds
->spare
= thresholds
->primary
;
3598 rcu_assign_pointer(thresholds
->primary
, new);
3600 /* To be sure that nobody uses thresholds */
3603 /* If all events are unregistered, free the spare array */
3605 kfree(thresholds
->spare
);
3606 thresholds
->spare
= NULL
;
3609 mutex_unlock(&memcg
->thresholds_lock
);
3612 static void mem_cgroup_usage_unregister_event(struct mem_cgroup
*memcg
,
3613 struct eventfd_ctx
*eventfd
)
3615 return __mem_cgroup_usage_unregister_event(memcg
, eventfd
, _MEM
);
3618 static void memsw_cgroup_usage_unregister_event(struct mem_cgroup
*memcg
,
3619 struct eventfd_ctx
*eventfd
)
3621 return __mem_cgroup_usage_unregister_event(memcg
, eventfd
, _MEMSWAP
);
3624 static int mem_cgroup_oom_register_event(struct mem_cgroup
*memcg
,
3625 struct eventfd_ctx
*eventfd
, const char *args
)
3627 struct mem_cgroup_eventfd_list
*event
;
3629 event
= kmalloc(sizeof(*event
), GFP_KERNEL
);
3633 spin_lock(&memcg_oom_lock
);
3635 event
->eventfd
= eventfd
;
3636 list_add(&event
->list
, &memcg
->oom_notify
);
3638 /* already in OOM ? */
3639 if (memcg
->under_oom
)
3640 eventfd_signal(eventfd
, 1);
3641 spin_unlock(&memcg_oom_lock
);
3646 static void mem_cgroup_oom_unregister_event(struct mem_cgroup
*memcg
,
3647 struct eventfd_ctx
*eventfd
)
3649 struct mem_cgroup_eventfd_list
*ev
, *tmp
;
3651 spin_lock(&memcg_oom_lock
);
3653 list_for_each_entry_safe(ev
, tmp
, &memcg
->oom_notify
, list
) {
3654 if (ev
->eventfd
== eventfd
) {
3655 list_del(&ev
->list
);
3660 spin_unlock(&memcg_oom_lock
);
3663 static int mem_cgroup_oom_control_read(struct seq_file
*sf
, void *v
)
3665 struct mem_cgroup
*memcg
= mem_cgroup_from_css(seq_css(sf
));
3667 seq_printf(sf
, "oom_kill_disable %d\n", memcg
->oom_kill_disable
);
3668 seq_printf(sf
, "under_oom %d\n", (bool)memcg
->under_oom
);
3669 seq_printf(sf
, "oom_kill %lu\n", memcg_sum_events(memcg
, OOM_KILL
));
3673 static int mem_cgroup_oom_control_write(struct cgroup_subsys_state
*css
,
3674 struct cftype
*cft
, u64 val
)
3676 struct mem_cgroup
*memcg
= mem_cgroup_from_css(css
);
3678 /* cannot set to root cgroup and only 0 and 1 are allowed */
3679 if (!css
->parent
|| !((val
== 0) || (val
== 1)))
3682 memcg
->oom_kill_disable
= val
;
3684 memcg_oom_recover(memcg
);
3689 #ifdef CONFIG_CGROUP_WRITEBACK
3691 struct list_head
*mem_cgroup_cgwb_list(struct mem_cgroup
*memcg
)
3693 return &memcg
->cgwb_list
;
3696 static int memcg_wb_domain_init(struct mem_cgroup
*memcg
, gfp_t gfp
)
3698 return wb_domain_init(&memcg
->cgwb_domain
, gfp
);
3701 static void memcg_wb_domain_exit(struct mem_cgroup
*memcg
)
3703 wb_domain_exit(&memcg
->cgwb_domain
);
3706 static void memcg_wb_domain_size_changed(struct mem_cgroup
*memcg
)
3708 wb_domain_size_changed(&memcg
->cgwb_domain
);
3711 struct wb_domain
*mem_cgroup_wb_domain(struct bdi_writeback
*wb
)
3713 struct mem_cgroup
*memcg
= mem_cgroup_from_css(wb
->memcg_css
);
3715 if (!memcg
->css
.parent
)
3718 return &memcg
->cgwb_domain
;
3722 * mem_cgroup_wb_stats - retrieve writeback related stats from its memcg
3723 * @wb: bdi_writeback in question
3724 * @pfilepages: out parameter for number of file pages
3725 * @pheadroom: out parameter for number of allocatable pages according to memcg
3726 * @pdirty: out parameter for number of dirty pages
3727 * @pwriteback: out parameter for number of pages under writeback
3729 * Determine the numbers of file, headroom, dirty, and writeback pages in
3730 * @wb's memcg. File, dirty and writeback are self-explanatory. Headroom
3731 * is a bit more involved.
3733 * A memcg's headroom is "min(max, high) - used". In the hierarchy, the
3734 * headroom is calculated as the lowest headroom of itself and the
3735 * ancestors. Note that this doesn't consider the actual amount of
3736 * available memory in the system. The caller should further cap
3737 * *@pheadroom accordingly.
3739 void mem_cgroup_wb_stats(struct bdi_writeback
*wb
, unsigned long *pfilepages
,
3740 unsigned long *pheadroom
, unsigned long *pdirty
,
3741 unsigned long *pwriteback
)
3743 struct mem_cgroup
*memcg
= mem_cgroup_from_css(wb
->memcg_css
);
3744 struct mem_cgroup
*parent
;
3746 *pdirty
= memcg_page_state(memcg
, NR_FILE_DIRTY
);
3748 /* this should eventually include NR_UNSTABLE_NFS */
3749 *pwriteback
= memcg_page_state(memcg
, NR_WRITEBACK
);
3750 *pfilepages
= mem_cgroup_nr_lru_pages(memcg
, (1 << LRU_INACTIVE_FILE
) |
3751 (1 << LRU_ACTIVE_FILE
));
3752 *pheadroom
= PAGE_COUNTER_MAX
;
3754 while ((parent
= parent_mem_cgroup(memcg
))) {
3755 unsigned long ceiling
= min(memcg
->memory
.limit
, memcg
->high
);
3756 unsigned long used
= page_counter_read(&memcg
->memory
);
3758 *pheadroom
= min(*pheadroom
, ceiling
- min(ceiling
, used
));
3763 #else /* CONFIG_CGROUP_WRITEBACK */
3765 static int memcg_wb_domain_init(struct mem_cgroup
*memcg
, gfp_t gfp
)
3770 static void memcg_wb_domain_exit(struct mem_cgroup
*memcg
)
3774 static void memcg_wb_domain_size_changed(struct mem_cgroup
*memcg
)
3778 #endif /* CONFIG_CGROUP_WRITEBACK */
3781 * DO NOT USE IN NEW FILES.
3783 * "cgroup.event_control" implementation.
3785 * This is way over-engineered. It tries to support fully configurable
3786 * events for each user. Such level of flexibility is completely
3787 * unnecessary especially in the light of the planned unified hierarchy.
3789 * Please deprecate this and replace with something simpler if at all
3794 * Unregister event and free resources.
3796 * Gets called from workqueue.
3798 static void memcg_event_remove(struct work_struct
*work
)
3800 struct mem_cgroup_event
*event
=
3801 container_of(work
, struct mem_cgroup_event
, remove
);
3802 struct mem_cgroup
*memcg
= event
->memcg
;
3804 remove_wait_queue(event
->wqh
, &event
->wait
);
3806 event
->unregister_event(memcg
, event
->eventfd
);
3808 /* Notify userspace the event is going away. */
3809 eventfd_signal(event
->eventfd
, 1);
3811 eventfd_ctx_put(event
->eventfd
);
3813 css_put(&memcg
->css
);
3817 * Gets called on POLLHUP on eventfd when user closes it.
3819 * Called with wqh->lock held and interrupts disabled.
3821 static int memcg_event_wake(wait_queue_entry_t
*wait
, unsigned mode
,
3822 int sync
, void *key
)
3824 struct mem_cgroup_event
*event
=
3825 container_of(wait
, struct mem_cgroup_event
, wait
);
3826 struct mem_cgroup
*memcg
= event
->memcg
;
3827 unsigned long flags
= (unsigned long)key
;
3829 if (flags
& POLLHUP
) {
3831 * If the event has been detached at cgroup removal, we
3832 * can simply return knowing the other side will cleanup
3835 * We can't race against event freeing since the other
3836 * side will require wqh->lock via remove_wait_queue(),
3839 spin_lock(&memcg
->event_list_lock
);
3840 if (!list_empty(&event
->list
)) {
3841 list_del_init(&event
->list
);
3843 * We are in atomic context, but cgroup_event_remove()
3844 * may sleep, so we have to call it in workqueue.
3846 schedule_work(&event
->remove
);
3848 spin_unlock(&memcg
->event_list_lock
);
3854 static void memcg_event_ptable_queue_proc(struct file
*file
,
3855 wait_queue_head_t
*wqh
, poll_table
*pt
)
3857 struct mem_cgroup_event
*event
=
3858 container_of(pt
, struct mem_cgroup_event
, pt
);
3861 add_wait_queue(wqh
, &event
->wait
);
3865 * DO NOT USE IN NEW FILES.
3867 * Parse input and register new cgroup event handler.
3869 * Input must be in format '<event_fd> <control_fd> <args>'.
3870 * Interpretation of args is defined by control file implementation.
3872 static ssize_t
memcg_write_event_control(struct kernfs_open_file
*of
,
3873 char *buf
, size_t nbytes
, loff_t off
)
3875 struct cgroup_subsys_state
*css
= of_css(of
);
3876 struct mem_cgroup
*memcg
= mem_cgroup_from_css(css
);
3877 struct mem_cgroup_event
*event
;
3878 struct cgroup_subsys_state
*cfile_css
;
3879 unsigned int efd
, cfd
;
3886 buf
= strstrip(buf
);
3888 efd
= simple_strtoul(buf
, &endp
, 10);
3893 cfd
= simple_strtoul(buf
, &endp
, 10);
3894 if ((*endp
!= ' ') && (*endp
!= '\0'))
3898 event
= kzalloc(sizeof(*event
), GFP_KERNEL
);
3902 event
->memcg
= memcg
;
3903 INIT_LIST_HEAD(&event
->list
);
3904 init_poll_funcptr(&event
->pt
, memcg_event_ptable_queue_proc
);
3905 init_waitqueue_func_entry(&event
->wait
, memcg_event_wake
);
3906 INIT_WORK(&event
->remove
, memcg_event_remove
);
3914 event
->eventfd
= eventfd_ctx_fileget(efile
.file
);
3915 if (IS_ERR(event
->eventfd
)) {
3916 ret
= PTR_ERR(event
->eventfd
);
3923 goto out_put_eventfd
;
3926 /* the process need read permission on control file */
3927 /* AV: shouldn't we check that it's been opened for read instead? */
3928 ret
= inode_permission(file_inode(cfile
.file
), MAY_READ
);
3933 * Determine the event callbacks and set them in @event. This used
3934 * to be done via struct cftype but cgroup core no longer knows
3935 * about these events. The following is crude but the whole thing
3936 * is for compatibility anyway.
3938 * DO NOT ADD NEW FILES.
3940 name
= cfile
.file
->f_path
.dentry
->d_name
.name
;
3942 if (!strcmp(name
, "memory.usage_in_bytes")) {
3943 event
->register_event
= mem_cgroup_usage_register_event
;
3944 event
->unregister_event
= mem_cgroup_usage_unregister_event
;
3945 } else if (!strcmp(name
, "memory.oom_control")) {
3946 event
->register_event
= mem_cgroup_oom_register_event
;
3947 event
->unregister_event
= mem_cgroup_oom_unregister_event
;
3948 } else if (!strcmp(name
, "memory.pressure_level")) {
3949 event
->register_event
= vmpressure_register_event
;
3950 event
->unregister_event
= vmpressure_unregister_event
;
3951 } else if (!strcmp(name
, "memory.memsw.usage_in_bytes")) {
3952 event
->register_event
= memsw_cgroup_usage_register_event
;
3953 event
->unregister_event
= memsw_cgroup_usage_unregister_event
;
3960 * Verify @cfile should belong to @css. Also, remaining events are
3961 * automatically removed on cgroup destruction but the removal is
3962 * asynchronous, so take an extra ref on @css.
3964 cfile_css
= css_tryget_online_from_dir(cfile
.file
->f_path
.dentry
->d_parent
,
3965 &memory_cgrp_subsys
);
3967 if (IS_ERR(cfile_css
))
3969 if (cfile_css
!= css
) {
3974 ret
= event
->register_event(memcg
, event
->eventfd
, buf
);
3978 efile
.file
->f_op
->poll(efile
.file
, &event
->pt
);
3980 spin_lock(&memcg
->event_list_lock
);
3981 list_add(&event
->list
, &memcg
->event_list
);
3982 spin_unlock(&memcg
->event_list_lock
);
3994 eventfd_ctx_put(event
->eventfd
);
4003 static struct cftype mem_cgroup_legacy_files
[] = {
4005 .name
= "usage_in_bytes",
4006 .private = MEMFILE_PRIVATE(_MEM
, RES_USAGE
),
4007 .read_u64
= mem_cgroup_read_u64
,
4010 .name
= "max_usage_in_bytes",
4011 .private = MEMFILE_PRIVATE(_MEM
, RES_MAX_USAGE
),
4012 .write
= mem_cgroup_reset
,
4013 .read_u64
= mem_cgroup_read_u64
,
4016 .name
= "limit_in_bytes",
4017 .private = MEMFILE_PRIVATE(_MEM
, RES_LIMIT
),
4018 .write
= mem_cgroup_write
,
4019 .read_u64
= mem_cgroup_read_u64
,
4022 .name
= "soft_limit_in_bytes",
4023 .private = MEMFILE_PRIVATE(_MEM
, RES_SOFT_LIMIT
),
4024 .write
= mem_cgroup_write
,
4025 .read_u64
= mem_cgroup_read_u64
,
4029 .private = MEMFILE_PRIVATE(_MEM
, RES_FAILCNT
),
4030 .write
= mem_cgroup_reset
,
4031 .read_u64
= mem_cgroup_read_u64
,
4035 .seq_show
= memcg_stat_show
,
4038 .name
= "force_empty",
4039 .write
= mem_cgroup_force_empty_write
,
4042 .name
= "use_hierarchy",
4043 .write_u64
= mem_cgroup_hierarchy_write
,
4044 .read_u64
= mem_cgroup_hierarchy_read
,
4047 .name
= "cgroup.event_control", /* XXX: for compat */
4048 .write
= memcg_write_event_control
,
4049 .flags
= CFTYPE_NO_PREFIX
| CFTYPE_WORLD_WRITABLE
,
4052 .name
= "swappiness",
4053 .read_u64
= mem_cgroup_swappiness_read
,
4054 .write_u64
= mem_cgroup_swappiness_write
,
4057 .name
= "move_charge_at_immigrate",
4058 .read_u64
= mem_cgroup_move_charge_read
,
4059 .write_u64
= mem_cgroup_move_charge_write
,
4062 .name
= "oom_control",
4063 .seq_show
= mem_cgroup_oom_control_read
,
4064 .write_u64
= mem_cgroup_oom_control_write
,
4065 .private = MEMFILE_PRIVATE(_OOM_TYPE
, OOM_CONTROL
),
4068 .name
= "pressure_level",
4072 .name
= "numa_stat",
4073 .seq_show
= memcg_numa_stat_show
,
4077 .name
= "kmem.limit_in_bytes",
4078 .private = MEMFILE_PRIVATE(_KMEM
, RES_LIMIT
),
4079 .write
= mem_cgroup_write
,
4080 .read_u64
= mem_cgroup_read_u64
,
4083 .name
= "kmem.usage_in_bytes",
4084 .private = MEMFILE_PRIVATE(_KMEM
, RES_USAGE
),
4085 .read_u64
= mem_cgroup_read_u64
,
4088 .name
= "kmem.failcnt",
4089 .private = MEMFILE_PRIVATE(_KMEM
, RES_FAILCNT
),
4090 .write
= mem_cgroup_reset
,
4091 .read_u64
= mem_cgroup_read_u64
,
4094 .name
= "kmem.max_usage_in_bytes",
4095 .private = MEMFILE_PRIVATE(_KMEM
, RES_MAX_USAGE
),
4096 .write
= mem_cgroup_reset
,
4097 .read_u64
= mem_cgroup_read_u64
,
4099 #if defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG)
4101 .name
= "kmem.slabinfo",
4102 .seq_start
= memcg_slab_start
,
4103 .seq_next
= memcg_slab_next
,
4104 .seq_stop
= memcg_slab_stop
,
4105 .seq_show
= memcg_slab_show
,
4109 .name
= "kmem.tcp.limit_in_bytes",
4110 .private = MEMFILE_PRIVATE(_TCP
, RES_LIMIT
),
4111 .write
= mem_cgroup_write
,
4112 .read_u64
= mem_cgroup_read_u64
,
4115 .name
= "kmem.tcp.usage_in_bytes",
4116 .private = MEMFILE_PRIVATE(_TCP
, RES_USAGE
),
4117 .read_u64
= mem_cgroup_read_u64
,
4120 .name
= "kmem.tcp.failcnt",
4121 .private = MEMFILE_PRIVATE(_TCP
, RES_FAILCNT
),
4122 .write
= mem_cgroup_reset
,
4123 .read_u64
= mem_cgroup_read_u64
,
4126 .name
= "kmem.tcp.max_usage_in_bytes",
4127 .private = MEMFILE_PRIVATE(_TCP
, RES_MAX_USAGE
),
4128 .write
= mem_cgroup_reset
,
4129 .read_u64
= mem_cgroup_read_u64
,
4131 { }, /* terminate */
4135 * Private memory cgroup IDR
4137 * Swap-out records and page cache shadow entries need to store memcg
4138 * references in constrained space, so we maintain an ID space that is
4139 * limited to 16 bit (MEM_CGROUP_ID_MAX), limiting the total number of
4140 * memory-controlled cgroups to 64k.
4142 * However, there usually are many references to the oflline CSS after
4143 * the cgroup has been destroyed, such as page cache or reclaimable
4144 * slab objects, that don't need to hang on to the ID. We want to keep
4145 * those dead CSS from occupying IDs, or we might quickly exhaust the
4146 * relatively small ID space and prevent the creation of new cgroups
4147 * even when there are much fewer than 64k cgroups - possibly none.
4149 * Maintain a private 16-bit ID space for memcg, and allow the ID to
4150 * be freed and recycled when it's no longer needed, which is usually
4151 * when the CSS is offlined.
4153 * The only exception to that are records of swapped out tmpfs/shmem
4154 * pages that need to be attributed to live ancestors on swapin. But
4155 * those references are manageable from userspace.
4158 static DEFINE_IDR(mem_cgroup_idr
);
4160 static void mem_cgroup_id_remove(struct mem_cgroup
*memcg
)
4162 if (memcg
->id
.id
> 0) {
4163 idr_remove(&mem_cgroup_idr
, memcg
->id
.id
);
4168 static void mem_cgroup_id_get_many(struct mem_cgroup
*memcg
, unsigned int n
)
4170 VM_BUG_ON(atomic_read(&memcg
->id
.ref
) <= 0);
4171 atomic_add(n
, &memcg
->id
.ref
);
4174 static void mem_cgroup_id_put_many(struct mem_cgroup
*memcg
, unsigned int n
)
4176 VM_BUG_ON(atomic_read(&memcg
->id
.ref
) < n
);
4177 if (atomic_sub_and_test(n
, &memcg
->id
.ref
)) {
4178 mem_cgroup_id_remove(memcg
);
4180 /* Memcg ID pins CSS */
4181 css_put(&memcg
->css
);
4185 static inline void mem_cgroup_id_get(struct mem_cgroup
*memcg
)
4187 mem_cgroup_id_get_many(memcg
, 1);
4190 static inline void mem_cgroup_id_put(struct mem_cgroup
*memcg
)
4192 mem_cgroup_id_put_many(memcg
, 1);
4196 * mem_cgroup_from_id - look up a memcg from a memcg id
4197 * @id: the memcg id to look up
4199 * Caller must hold rcu_read_lock().
4201 struct mem_cgroup
*mem_cgroup_from_id(unsigned short id
)
4203 WARN_ON_ONCE(!rcu_read_lock_held());
4204 return idr_find(&mem_cgroup_idr
, id
);
4207 static int alloc_mem_cgroup_per_node_info(struct mem_cgroup
*memcg
, int node
)
4209 struct mem_cgroup_per_node
*pn
;
4212 * This routine is called against possible nodes.
4213 * But it's BUG to call kmalloc() against offline node.
4215 * TODO: this routine can waste much memory for nodes which will
4216 * never be onlined. It's better to use memory hotplug callback
4219 if (!node_state(node
, N_NORMAL_MEMORY
))
4221 pn
= kzalloc_node(sizeof(*pn
), GFP_KERNEL
, tmp
);
4225 pn
->lruvec_stat
= alloc_percpu(struct lruvec_stat
);
4226 if (!pn
->lruvec_stat
) {
4231 lruvec_init(&pn
->lruvec
);
4232 pn
->usage_in_excess
= 0;
4233 pn
->on_tree
= false;
4236 memcg
->nodeinfo
[node
] = pn
;
4240 static void free_mem_cgroup_per_node_info(struct mem_cgroup
*memcg
, int node
)
4242 struct mem_cgroup_per_node
*pn
= memcg
->nodeinfo
[node
];
4247 free_percpu(pn
->lruvec_stat
);
4251 static void __mem_cgroup_free(struct mem_cgroup
*memcg
)
4256 free_mem_cgroup_per_node_info(memcg
, node
);
4257 free_percpu(memcg
->stat
);
4261 static void mem_cgroup_free(struct mem_cgroup
*memcg
)
4263 memcg_wb_domain_exit(memcg
);
4264 __mem_cgroup_free(memcg
);
4267 static struct mem_cgroup
*mem_cgroup_alloc(void)
4269 struct mem_cgroup
*memcg
;
4273 size
= sizeof(struct mem_cgroup
);
4274 size
+= nr_node_ids
* sizeof(struct mem_cgroup_per_node
*);
4276 memcg
= kzalloc(size
, GFP_KERNEL
);
4280 memcg
->id
.id
= idr_alloc(&mem_cgroup_idr
, NULL
,
4281 1, MEM_CGROUP_ID_MAX
,
4283 if (memcg
->id
.id
< 0)
4286 memcg
->stat
= alloc_percpu(struct mem_cgroup_stat_cpu
);
4291 if (alloc_mem_cgroup_per_node_info(memcg
, node
))
4294 if (memcg_wb_domain_init(memcg
, GFP_KERNEL
))
4297 INIT_WORK(&memcg
->high_work
, high_work_func
);
4298 memcg
->last_scanned_node
= MAX_NUMNODES
;
4299 INIT_LIST_HEAD(&memcg
->oom_notify
);
4300 mutex_init(&memcg
->thresholds_lock
);
4301 spin_lock_init(&memcg
->move_lock
);
4302 vmpressure_init(&memcg
->vmpressure
);
4303 INIT_LIST_HEAD(&memcg
->event_list
);
4304 spin_lock_init(&memcg
->event_list_lock
);
4305 memcg
->socket_pressure
= jiffies
;
4307 memcg
->kmemcg_id
= -1;
4309 #ifdef CONFIG_CGROUP_WRITEBACK
4310 INIT_LIST_HEAD(&memcg
->cgwb_list
);
4312 idr_replace(&mem_cgroup_idr
, memcg
, memcg
->id
.id
);
4315 mem_cgroup_id_remove(memcg
);
4316 __mem_cgroup_free(memcg
);
4320 static struct cgroup_subsys_state
* __ref
4321 mem_cgroup_css_alloc(struct cgroup_subsys_state
*parent_css
)
4323 struct mem_cgroup
*parent
= mem_cgroup_from_css(parent_css
);
4324 struct mem_cgroup
*memcg
;
4325 long error
= -ENOMEM
;
4327 memcg
= mem_cgroup_alloc();
4329 return ERR_PTR(error
);
4331 memcg
->high
= PAGE_COUNTER_MAX
;
4332 memcg
->soft_limit
= PAGE_COUNTER_MAX
;
4334 memcg
->swappiness
= mem_cgroup_swappiness(parent
);
4335 memcg
->oom_kill_disable
= parent
->oom_kill_disable
;
4337 if (parent
&& parent
->use_hierarchy
) {
4338 memcg
->use_hierarchy
= true;
4339 page_counter_init(&memcg
->memory
, &parent
->memory
);
4340 page_counter_init(&memcg
->swap
, &parent
->swap
);
4341 page_counter_init(&memcg
->memsw
, &parent
->memsw
);
4342 page_counter_init(&memcg
->kmem
, &parent
->kmem
);
4343 page_counter_init(&memcg
->tcpmem
, &parent
->tcpmem
);
4345 page_counter_init(&memcg
->memory
, NULL
);
4346 page_counter_init(&memcg
->swap
, NULL
);
4347 page_counter_init(&memcg
->memsw
, NULL
);
4348 page_counter_init(&memcg
->kmem
, NULL
);
4349 page_counter_init(&memcg
->tcpmem
, NULL
);
4351 * Deeper hierachy with use_hierarchy == false doesn't make
4352 * much sense so let cgroup subsystem know about this
4353 * unfortunate state in our controller.
4355 if (parent
!= root_mem_cgroup
)
4356 memory_cgrp_subsys
.broken_hierarchy
= true;
4359 /* The following stuff does not apply to the root */
4361 root_mem_cgroup
= memcg
;
4365 error
= memcg_online_kmem(memcg
);
4369 if (cgroup_subsys_on_dfl(memory_cgrp_subsys
) && !cgroup_memory_nosocket
)
4370 static_branch_inc(&memcg_sockets_enabled_key
);
4374 mem_cgroup_id_remove(memcg
);
4375 mem_cgroup_free(memcg
);
4376 return ERR_PTR(-ENOMEM
);
4379 static int mem_cgroup_css_online(struct cgroup_subsys_state
*css
)
4381 struct mem_cgroup
*memcg
= mem_cgroup_from_css(css
);
4383 /* Online state pins memcg ID, memcg ID pins CSS */
4384 atomic_set(&memcg
->id
.ref
, 1);
4389 static void mem_cgroup_css_offline(struct cgroup_subsys_state
*css
)
4391 struct mem_cgroup
*memcg
= mem_cgroup_from_css(css
);
4392 struct mem_cgroup_event
*event
, *tmp
;
4395 * Unregister events and notify userspace.
4396 * Notify userspace about cgroup removing only after rmdir of cgroup
4397 * directory to avoid race between userspace and kernelspace.
4399 spin_lock(&memcg
->event_list_lock
);
4400 list_for_each_entry_safe(event
, tmp
, &memcg
->event_list
, list
) {
4401 list_del_init(&event
->list
);
4402 schedule_work(&event
->remove
);
4404 spin_unlock(&memcg
->event_list_lock
);
4408 memcg_offline_kmem(memcg
);
4409 wb_memcg_offline(memcg
);
4411 mem_cgroup_id_put(memcg
);
4414 static void mem_cgroup_css_released(struct cgroup_subsys_state
*css
)
4416 struct mem_cgroup
*memcg
= mem_cgroup_from_css(css
);
4418 invalidate_reclaim_iterators(memcg
);
4421 static void mem_cgroup_css_free(struct cgroup_subsys_state
*css
)
4423 struct mem_cgroup
*memcg
= mem_cgroup_from_css(css
);
4425 if (cgroup_subsys_on_dfl(memory_cgrp_subsys
) && !cgroup_memory_nosocket
)
4426 static_branch_dec(&memcg_sockets_enabled_key
);
4428 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys
) && memcg
->tcpmem_active
)
4429 static_branch_dec(&memcg_sockets_enabled_key
);
4431 vmpressure_cleanup(&memcg
->vmpressure
);
4432 cancel_work_sync(&memcg
->high_work
);
4433 mem_cgroup_remove_from_trees(memcg
);
4434 memcg_free_kmem(memcg
);
4435 mem_cgroup_free(memcg
);
4439 * mem_cgroup_css_reset - reset the states of a mem_cgroup
4440 * @css: the target css
4442 * Reset the states of the mem_cgroup associated with @css. This is
4443 * invoked when the userland requests disabling on the default hierarchy
4444 * but the memcg is pinned through dependency. The memcg should stop
4445 * applying policies and should revert to the vanilla state as it may be
4446 * made visible again.
4448 * The current implementation only resets the essential configurations.
4449 * This needs to be expanded to cover all the visible parts.
4451 static void mem_cgroup_css_reset(struct cgroup_subsys_state
*css
)
4453 struct mem_cgroup
*memcg
= mem_cgroup_from_css(css
);
4455 page_counter_limit(&memcg
->memory
, PAGE_COUNTER_MAX
);
4456 page_counter_limit(&memcg
->swap
, PAGE_COUNTER_MAX
);
4457 page_counter_limit(&memcg
->memsw
, PAGE_COUNTER_MAX
);
4458 page_counter_limit(&memcg
->kmem
, PAGE_COUNTER_MAX
);
4459 page_counter_limit(&memcg
->tcpmem
, PAGE_COUNTER_MAX
);
4461 memcg
->high
= PAGE_COUNTER_MAX
;
4462 memcg
->soft_limit
= PAGE_COUNTER_MAX
;
4463 memcg_wb_domain_size_changed(memcg
);
4467 /* Handlers for move charge at task migration. */
4468 static int mem_cgroup_do_precharge(unsigned long count
)
4472 /* Try a single bulk charge without reclaim first, kswapd may wake */
4473 ret
= try_charge(mc
.to
, GFP_KERNEL
& ~__GFP_DIRECT_RECLAIM
, count
);
4475 mc
.precharge
+= count
;
4479 /* Try charges one by one with reclaim, but do not retry */
4481 ret
= try_charge(mc
.to
, GFP_KERNEL
| __GFP_NORETRY
, 1);
4495 enum mc_target_type
{
4502 static struct page
*mc_handle_present_pte(struct vm_area_struct
*vma
,
4503 unsigned long addr
, pte_t ptent
)
4505 struct page
*page
= _vm_normal_page(vma
, addr
, ptent
, true);
4507 if (!page
|| !page_mapped(page
))
4509 if (PageAnon(page
)) {
4510 if (!(mc
.flags
& MOVE_ANON
))
4513 if (!(mc
.flags
& MOVE_FILE
))
4516 if (!get_page_unless_zero(page
))
4522 #if defined(CONFIG_SWAP) || defined(CONFIG_DEVICE_PRIVATE)
4523 static struct page
*mc_handle_swap_pte(struct vm_area_struct
*vma
,
4524 pte_t ptent
, swp_entry_t
*entry
)
4526 struct page
*page
= NULL
;
4527 swp_entry_t ent
= pte_to_swp_entry(ptent
);
4529 if (!(mc
.flags
& MOVE_ANON
) || non_swap_entry(ent
))
4533 * Handle MEMORY_DEVICE_PRIVATE which are ZONE_DEVICE page belonging to
4534 * a device and because they are not accessible by CPU they are store
4535 * as special swap entry in the CPU page table.
4537 if (is_device_private_entry(ent
)) {
4538 page
= device_private_entry_to_page(ent
);
4540 * MEMORY_DEVICE_PRIVATE means ZONE_DEVICE page and which have
4541 * a refcount of 1 when free (unlike normal page)
4543 if (!page_ref_add_unless(page
, 1, 1))
4549 * Because lookup_swap_cache() updates some statistics counter,
4550 * we call find_get_page() with swapper_space directly.
4552 page
= find_get_page(swap_address_space(ent
), swp_offset(ent
));
4553 if (do_memsw_account())
4554 entry
->val
= ent
.val
;
4559 static struct page
*mc_handle_swap_pte(struct vm_area_struct
*vma
,
4560 pte_t ptent
, swp_entry_t
*entry
)
4566 static struct page
*mc_handle_file_pte(struct vm_area_struct
*vma
,
4567 unsigned long addr
, pte_t ptent
, swp_entry_t
*entry
)
4569 struct page
*page
= NULL
;
4570 struct address_space
*mapping
;
4573 if (!vma
->vm_file
) /* anonymous vma */
4575 if (!(mc
.flags
& MOVE_FILE
))
4578 mapping
= vma
->vm_file
->f_mapping
;
4579 pgoff
= linear_page_index(vma
, addr
);
4581 /* page is moved even if it's not RSS of this task(page-faulted). */
4583 /* shmem/tmpfs may report page out on swap: account for that too. */
4584 if (shmem_mapping(mapping
)) {
4585 page
= find_get_entry(mapping
, pgoff
);
4586 if (radix_tree_exceptional_entry(page
)) {
4587 swp_entry_t swp
= radix_to_swp_entry(page
);
4588 if (do_memsw_account())
4590 page
= find_get_page(swap_address_space(swp
),
4594 page
= find_get_page(mapping
, pgoff
);
4596 page
= find_get_page(mapping
, pgoff
);
4602 * mem_cgroup_move_account - move account of the page
4604 * @compound: charge the page as compound or small page
4605 * @from: mem_cgroup which the page is moved from.
4606 * @to: mem_cgroup which the page is moved to. @from != @to.
4608 * The caller must make sure the page is not on LRU (isolate_page() is useful.)
4610 * This function doesn't do "charge" to new cgroup and doesn't do "uncharge"
4613 static int mem_cgroup_move_account(struct page
*page
,
4615 struct mem_cgroup
*from
,
4616 struct mem_cgroup
*to
)
4618 unsigned long flags
;
4619 unsigned int nr_pages
= compound
? hpage_nr_pages(page
) : 1;
4623 VM_BUG_ON(from
== to
);
4624 VM_BUG_ON_PAGE(PageLRU(page
), page
);
4625 VM_BUG_ON(compound
&& !PageTransHuge(page
));
4628 * Prevent mem_cgroup_migrate() from looking at
4629 * page->mem_cgroup of its source page while we change it.
4632 if (!trylock_page(page
))
4636 if (page
->mem_cgroup
!= from
)
4639 anon
= PageAnon(page
);
4641 spin_lock_irqsave(&from
->move_lock
, flags
);
4643 if (!anon
&& page_mapped(page
)) {
4644 __this_cpu_sub(from
->stat
->count
[NR_FILE_MAPPED
], nr_pages
);
4645 __this_cpu_add(to
->stat
->count
[NR_FILE_MAPPED
], nr_pages
);
4649 * move_lock grabbed above and caller set from->moving_account, so
4650 * mod_memcg_page_state will serialize updates to PageDirty.
4651 * So mapping should be stable for dirty pages.
4653 if (!anon
&& PageDirty(page
)) {
4654 struct address_space
*mapping
= page_mapping(page
);
4656 if (mapping_cap_account_dirty(mapping
)) {
4657 __this_cpu_sub(from
->stat
->count
[NR_FILE_DIRTY
],
4659 __this_cpu_add(to
->stat
->count
[NR_FILE_DIRTY
],
4664 if (PageWriteback(page
)) {
4665 __this_cpu_sub(from
->stat
->count
[NR_WRITEBACK
], nr_pages
);
4666 __this_cpu_add(to
->stat
->count
[NR_WRITEBACK
], nr_pages
);
4670 * It is safe to change page->mem_cgroup here because the page
4671 * is referenced, charged, and isolated - we can't race with
4672 * uncharging, charging, migration, or LRU putback.
4675 /* caller should have done css_get */
4676 page
->mem_cgroup
= to
;
4677 spin_unlock_irqrestore(&from
->move_lock
, flags
);
4681 local_irq_disable();
4682 mem_cgroup_charge_statistics(to
, page
, compound
, nr_pages
);
4683 memcg_check_events(to
, page
);
4684 mem_cgroup_charge_statistics(from
, page
, compound
, -nr_pages
);
4685 memcg_check_events(from
, page
);
4694 * get_mctgt_type - get target type of moving charge
4695 * @vma: the vma the pte to be checked belongs
4696 * @addr: the address corresponding to the pte to be checked
4697 * @ptent: the pte to be checked
4698 * @target: the pointer the target page or swap ent will be stored(can be NULL)
4701 * 0(MC_TARGET_NONE): if the pte is not a target for move charge.
4702 * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for
4703 * move charge. if @target is not NULL, the page is stored in target->page
4704 * with extra refcnt got(Callers should handle it).
4705 * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a
4706 * target for charge migration. if @target is not NULL, the entry is stored
4708 * 3(MC_TARGET_DEVICE): like MC_TARGET_PAGE but page is MEMORY_DEVICE_PUBLIC
4709 * or MEMORY_DEVICE_PRIVATE (so ZONE_DEVICE page and thus not on the lru).
4710 * For now we such page is charge like a regular page would be as for all
4711 * intent and purposes it is just special memory taking the place of a
4714 * See Documentations/vm/hmm.txt and include/linux/hmm.h
4716 * Called with pte lock held.
4719 static enum mc_target_type
get_mctgt_type(struct vm_area_struct
*vma
,
4720 unsigned long addr
, pte_t ptent
, union mc_target
*target
)
4722 struct page
*page
= NULL
;
4723 enum mc_target_type ret
= MC_TARGET_NONE
;
4724 swp_entry_t ent
= { .val
= 0 };
4726 if (pte_present(ptent
))
4727 page
= mc_handle_present_pte(vma
, addr
, ptent
);
4728 else if (is_swap_pte(ptent
))
4729 page
= mc_handle_swap_pte(vma
, ptent
, &ent
);
4730 else if (pte_none(ptent
))
4731 page
= mc_handle_file_pte(vma
, addr
, ptent
, &ent
);
4733 if (!page
&& !ent
.val
)
4737 * Do only loose check w/o serialization.
4738 * mem_cgroup_move_account() checks the page is valid or
4739 * not under LRU exclusion.
4741 if (page
->mem_cgroup
== mc
.from
) {
4742 ret
= MC_TARGET_PAGE
;
4743 if (is_device_private_page(page
) ||
4744 is_device_public_page(page
))
4745 ret
= MC_TARGET_DEVICE
;
4747 target
->page
= page
;
4749 if (!ret
|| !target
)
4753 * There is a swap entry and a page doesn't exist or isn't charged.
4754 * But we cannot move a tail-page in a THP.
4756 if (ent
.val
&& !ret
&& (!page
|| !PageTransCompound(page
)) &&
4757 mem_cgroup_id(mc
.from
) == lookup_swap_cgroup_id(ent
)) {
4758 ret
= MC_TARGET_SWAP
;
4765 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
4767 * We don't consider PMD mapped swapping or file mapped pages because THP does
4768 * not support them for now.
4769 * Caller should make sure that pmd_trans_huge(pmd) is true.
4771 static enum mc_target_type
get_mctgt_type_thp(struct vm_area_struct
*vma
,
4772 unsigned long addr
, pmd_t pmd
, union mc_target
*target
)
4774 struct page
*page
= NULL
;
4775 enum mc_target_type ret
= MC_TARGET_NONE
;
4777 if (unlikely(is_swap_pmd(pmd
))) {
4778 VM_BUG_ON(thp_migration_supported() &&
4779 !is_pmd_migration_entry(pmd
));
4782 page
= pmd_page(pmd
);
4783 VM_BUG_ON_PAGE(!page
|| !PageHead(page
), page
);
4784 if (!(mc
.flags
& MOVE_ANON
))
4786 if (page
->mem_cgroup
== mc
.from
) {
4787 ret
= MC_TARGET_PAGE
;
4790 target
->page
= page
;
4796 static inline enum mc_target_type
get_mctgt_type_thp(struct vm_area_struct
*vma
,
4797 unsigned long addr
, pmd_t pmd
, union mc_target
*target
)
4799 return MC_TARGET_NONE
;
4803 static int mem_cgroup_count_precharge_pte_range(pmd_t
*pmd
,
4804 unsigned long addr
, unsigned long end
,
4805 struct mm_walk
*walk
)
4807 struct vm_area_struct
*vma
= walk
->vma
;
4811 ptl
= pmd_trans_huge_lock(pmd
, vma
);
4814 * Note their can not be MC_TARGET_DEVICE for now as we do not
4815 * support transparent huge page with MEMORY_DEVICE_PUBLIC or
4816 * MEMORY_DEVICE_PRIVATE but this might change.
4818 if (get_mctgt_type_thp(vma
, addr
, *pmd
, NULL
) == MC_TARGET_PAGE
)
4819 mc
.precharge
+= HPAGE_PMD_NR
;
4824 if (pmd_trans_unstable(pmd
))
4826 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
4827 for (; addr
!= end
; pte
++, addr
+= PAGE_SIZE
)
4828 if (get_mctgt_type(vma
, addr
, *pte
, NULL
))
4829 mc
.precharge
++; /* increment precharge temporarily */
4830 pte_unmap_unlock(pte
- 1, ptl
);
4836 static unsigned long mem_cgroup_count_precharge(struct mm_struct
*mm
)
4838 unsigned long precharge
;
4840 struct mm_walk mem_cgroup_count_precharge_walk
= {
4841 .pmd_entry
= mem_cgroup_count_precharge_pte_range
,
4844 down_read(&mm
->mmap_sem
);
4845 walk_page_range(0, mm
->highest_vm_end
,
4846 &mem_cgroup_count_precharge_walk
);
4847 up_read(&mm
->mmap_sem
);
4849 precharge
= mc
.precharge
;
4855 static int mem_cgroup_precharge_mc(struct mm_struct
*mm
)
4857 unsigned long precharge
= mem_cgroup_count_precharge(mm
);
4859 VM_BUG_ON(mc
.moving_task
);
4860 mc
.moving_task
= current
;
4861 return mem_cgroup_do_precharge(precharge
);
4864 /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */
4865 static void __mem_cgroup_clear_mc(void)
4867 struct mem_cgroup
*from
= mc
.from
;
4868 struct mem_cgroup
*to
= mc
.to
;
4870 /* we must uncharge all the leftover precharges from mc.to */
4872 cancel_charge(mc
.to
, mc
.precharge
);
4876 * we didn't uncharge from mc.from at mem_cgroup_move_account(), so
4877 * we must uncharge here.
4879 if (mc
.moved_charge
) {
4880 cancel_charge(mc
.from
, mc
.moved_charge
);
4881 mc
.moved_charge
= 0;
4883 /* we must fixup refcnts and charges */
4884 if (mc
.moved_swap
) {
4885 /* uncharge swap account from the old cgroup */
4886 if (!mem_cgroup_is_root(mc
.from
))
4887 page_counter_uncharge(&mc
.from
->memsw
, mc
.moved_swap
);
4889 mem_cgroup_id_put_many(mc
.from
, mc
.moved_swap
);
4892 * we charged both to->memory and to->memsw, so we
4893 * should uncharge to->memory.
4895 if (!mem_cgroup_is_root(mc
.to
))
4896 page_counter_uncharge(&mc
.to
->memory
, mc
.moved_swap
);
4898 mem_cgroup_id_get_many(mc
.to
, mc
.moved_swap
);
4899 css_put_many(&mc
.to
->css
, mc
.moved_swap
);
4903 memcg_oom_recover(from
);
4904 memcg_oom_recover(to
);
4905 wake_up_all(&mc
.waitq
);
4908 static void mem_cgroup_clear_mc(void)
4910 struct mm_struct
*mm
= mc
.mm
;
4913 * we must clear moving_task before waking up waiters at the end of
4916 mc
.moving_task
= NULL
;
4917 __mem_cgroup_clear_mc();
4918 spin_lock(&mc
.lock
);
4922 spin_unlock(&mc
.lock
);
4927 static int mem_cgroup_can_attach(struct cgroup_taskset
*tset
)
4929 struct cgroup_subsys_state
*css
;
4930 struct mem_cgroup
*memcg
= NULL
; /* unneeded init to make gcc happy */
4931 struct mem_cgroup
*from
;
4932 struct task_struct
*leader
, *p
;
4933 struct mm_struct
*mm
;
4934 unsigned long move_flags
;
4937 /* charge immigration isn't supported on the default hierarchy */
4938 if (cgroup_subsys_on_dfl(memory_cgrp_subsys
))
4942 * Multi-process migrations only happen on the default hierarchy
4943 * where charge immigration is not used. Perform charge
4944 * immigration if @tset contains a leader and whine if there are
4948 cgroup_taskset_for_each_leader(leader
, css
, tset
) {
4951 memcg
= mem_cgroup_from_css(css
);
4957 * We are now commited to this value whatever it is. Changes in this
4958 * tunable will only affect upcoming migrations, not the current one.
4959 * So we need to save it, and keep it going.
4961 move_flags
= READ_ONCE(memcg
->move_charge_at_immigrate
);
4965 from
= mem_cgroup_from_task(p
);
4967 VM_BUG_ON(from
== memcg
);
4969 mm
= get_task_mm(p
);
4972 /* We move charges only when we move a owner of the mm */
4973 if (mm
->owner
== p
) {
4976 VM_BUG_ON(mc
.precharge
);
4977 VM_BUG_ON(mc
.moved_charge
);
4978 VM_BUG_ON(mc
.moved_swap
);
4980 spin_lock(&mc
.lock
);
4984 mc
.flags
= move_flags
;
4985 spin_unlock(&mc
.lock
);
4986 /* We set mc.moving_task later */
4988 ret
= mem_cgroup_precharge_mc(mm
);
4990 mem_cgroup_clear_mc();
4997 static void mem_cgroup_cancel_attach(struct cgroup_taskset
*tset
)
5000 mem_cgroup_clear_mc();
5003 static int mem_cgroup_move_charge_pte_range(pmd_t
*pmd
,
5004 unsigned long addr
, unsigned long end
,
5005 struct mm_walk
*walk
)
5008 struct vm_area_struct
*vma
= walk
->vma
;
5011 enum mc_target_type target_type
;
5012 union mc_target target
;
5015 ptl
= pmd_trans_huge_lock(pmd
, vma
);
5017 if (mc
.precharge
< HPAGE_PMD_NR
) {
5021 target_type
= get_mctgt_type_thp(vma
, addr
, *pmd
, &target
);
5022 if (target_type
== MC_TARGET_PAGE
) {
5024 if (!isolate_lru_page(page
)) {
5025 if (!mem_cgroup_move_account(page
, true,
5027 mc
.precharge
-= HPAGE_PMD_NR
;
5028 mc
.moved_charge
+= HPAGE_PMD_NR
;
5030 putback_lru_page(page
);
5033 } else if (target_type
== MC_TARGET_DEVICE
) {
5035 if (!mem_cgroup_move_account(page
, true,
5037 mc
.precharge
-= HPAGE_PMD_NR
;
5038 mc
.moved_charge
+= HPAGE_PMD_NR
;
5046 if (pmd_trans_unstable(pmd
))
5049 pte
= pte_offset_map_lock(vma
->vm_mm
, pmd
, addr
, &ptl
);
5050 for (; addr
!= end
; addr
+= PAGE_SIZE
) {
5051 pte_t ptent
= *(pte
++);
5052 bool device
= false;
5058 switch (get_mctgt_type(vma
, addr
, ptent
, &target
)) {
5059 case MC_TARGET_DEVICE
:
5062 case MC_TARGET_PAGE
:
5065 * We can have a part of the split pmd here. Moving it
5066 * can be done but it would be too convoluted so simply
5067 * ignore such a partial THP and keep it in original
5068 * memcg. There should be somebody mapping the head.
5070 if (PageTransCompound(page
))
5072 if (!device
&& isolate_lru_page(page
))
5074 if (!mem_cgroup_move_account(page
, false,
5077 /* we uncharge from mc.from later. */
5081 putback_lru_page(page
);
5082 put
: /* get_mctgt_type() gets the page */
5085 case MC_TARGET_SWAP
:
5087 if (!mem_cgroup_move_swap_account(ent
, mc
.from
, mc
.to
)) {
5089 /* we fixup refcnts and charges later. */
5097 pte_unmap_unlock(pte
- 1, ptl
);
5102 * We have consumed all precharges we got in can_attach().
5103 * We try charge one by one, but don't do any additional
5104 * charges to mc.to if we have failed in charge once in attach()
5107 ret
= mem_cgroup_do_precharge(1);
5115 static void mem_cgroup_move_charge(void)
5117 struct mm_walk mem_cgroup_move_charge_walk
= {
5118 .pmd_entry
= mem_cgroup_move_charge_pte_range
,
5122 lru_add_drain_all();
5124 * Signal lock_page_memcg() to take the memcg's move_lock
5125 * while we're moving its pages to another memcg. Then wait
5126 * for already started RCU-only updates to finish.
5128 atomic_inc(&mc
.from
->moving_account
);
5131 if (unlikely(!down_read_trylock(&mc
.mm
->mmap_sem
))) {
5133 * Someone who are holding the mmap_sem might be waiting in
5134 * waitq. So we cancel all extra charges, wake up all waiters,
5135 * and retry. Because we cancel precharges, we might not be able
5136 * to move enough charges, but moving charge is a best-effort
5137 * feature anyway, so it wouldn't be a big problem.
5139 __mem_cgroup_clear_mc();
5144 * When we have consumed all precharges and failed in doing
5145 * additional charge, the page walk just aborts.
5147 walk_page_range(0, mc
.mm
->highest_vm_end
, &mem_cgroup_move_charge_walk
);
5149 up_read(&mc
.mm
->mmap_sem
);
5150 atomic_dec(&mc
.from
->moving_account
);
5153 static void mem_cgroup_move_task(void)
5156 mem_cgroup_move_charge();
5157 mem_cgroup_clear_mc();
5160 #else /* !CONFIG_MMU */
5161 static int mem_cgroup_can_attach(struct cgroup_taskset
*tset
)
5165 static void mem_cgroup_cancel_attach(struct cgroup_taskset
*tset
)
5168 static void mem_cgroup_move_task(void)
5174 * Cgroup retains root cgroups across [un]mount cycles making it necessary
5175 * to verify whether we're attached to the default hierarchy on each mount
5178 static void mem_cgroup_bind(struct cgroup_subsys_state
*root_css
)
5181 * use_hierarchy is forced on the default hierarchy. cgroup core
5182 * guarantees that @root doesn't have any children, so turning it
5183 * on for the root memcg is enough.
5185 if (cgroup_subsys_on_dfl(memory_cgrp_subsys
))
5186 root_mem_cgroup
->use_hierarchy
= true;
5188 root_mem_cgroup
->use_hierarchy
= false;
5191 static u64
memory_current_read(struct cgroup_subsys_state
*css
,
5194 struct mem_cgroup
*memcg
= mem_cgroup_from_css(css
);
5196 return (u64
)page_counter_read(&memcg
->memory
) * PAGE_SIZE
;
5199 static int memory_low_show(struct seq_file
*m
, void *v
)
5201 struct mem_cgroup
*memcg
= mem_cgroup_from_css(seq_css(m
));
5202 unsigned long low
= READ_ONCE(memcg
->low
);
5204 if (low
== PAGE_COUNTER_MAX
)
5205 seq_puts(m
, "max\n");
5207 seq_printf(m
, "%llu\n", (u64
)low
* PAGE_SIZE
);
5212 static ssize_t
memory_low_write(struct kernfs_open_file
*of
,
5213 char *buf
, size_t nbytes
, loff_t off
)
5215 struct mem_cgroup
*memcg
= mem_cgroup_from_css(of_css(of
));
5219 buf
= strstrip(buf
);
5220 err
= page_counter_memparse(buf
, "max", &low
);
5229 static int memory_high_show(struct seq_file
*m
, void *v
)
5231 struct mem_cgroup
*memcg
= mem_cgroup_from_css(seq_css(m
));
5232 unsigned long high
= READ_ONCE(memcg
->high
);
5234 if (high
== PAGE_COUNTER_MAX
)
5235 seq_puts(m
, "max\n");
5237 seq_printf(m
, "%llu\n", (u64
)high
* PAGE_SIZE
);
5242 static ssize_t
memory_high_write(struct kernfs_open_file
*of
,
5243 char *buf
, size_t nbytes
, loff_t off
)
5245 struct mem_cgroup
*memcg
= mem_cgroup_from_css(of_css(of
));
5246 unsigned long nr_pages
;
5250 buf
= strstrip(buf
);
5251 err
= page_counter_memparse(buf
, "max", &high
);
5257 nr_pages
= page_counter_read(&memcg
->memory
);
5258 if (nr_pages
> high
)
5259 try_to_free_mem_cgroup_pages(memcg
, nr_pages
- high
,
5262 memcg_wb_domain_size_changed(memcg
);
5266 static int memory_max_show(struct seq_file
*m
, void *v
)
5268 struct mem_cgroup
*memcg
= mem_cgroup_from_css(seq_css(m
));
5269 unsigned long max
= READ_ONCE(memcg
->memory
.limit
);
5271 if (max
== PAGE_COUNTER_MAX
)
5272 seq_puts(m
, "max\n");
5274 seq_printf(m
, "%llu\n", (u64
)max
* PAGE_SIZE
);
5279 static ssize_t
memory_max_write(struct kernfs_open_file
*of
,
5280 char *buf
, size_t nbytes
, loff_t off
)
5282 struct mem_cgroup
*memcg
= mem_cgroup_from_css(of_css(of
));
5283 unsigned int nr_reclaims
= MEM_CGROUP_RECLAIM_RETRIES
;
5284 bool drained
= false;
5288 buf
= strstrip(buf
);
5289 err
= page_counter_memparse(buf
, "max", &max
);
5293 xchg(&memcg
->memory
.limit
, max
);
5296 unsigned long nr_pages
= page_counter_read(&memcg
->memory
);
5298 if (nr_pages
<= max
)
5301 if (signal_pending(current
)) {
5307 drain_all_stock(memcg
);
5313 if (!try_to_free_mem_cgroup_pages(memcg
, nr_pages
- max
,
5319 mem_cgroup_event(memcg
, MEMCG_OOM
);
5320 if (!mem_cgroup_out_of_memory(memcg
, GFP_KERNEL
, 0))
5324 memcg_wb_domain_size_changed(memcg
);
5328 static int memory_events_show(struct seq_file
*m
, void *v
)
5330 struct mem_cgroup
*memcg
= mem_cgroup_from_css(seq_css(m
));
5332 seq_printf(m
, "low %lu\n", memcg_sum_events(memcg
, MEMCG_LOW
));
5333 seq_printf(m
, "high %lu\n", memcg_sum_events(memcg
, MEMCG_HIGH
));
5334 seq_printf(m
, "max %lu\n", memcg_sum_events(memcg
, MEMCG_MAX
));
5335 seq_printf(m
, "oom %lu\n", memcg_sum_events(memcg
, MEMCG_OOM
));
5336 seq_printf(m
, "oom_kill %lu\n", memcg_sum_events(memcg
, OOM_KILL
));
5341 static int memory_stat_show(struct seq_file
*m
, void *v
)
5343 struct mem_cgroup
*memcg
= mem_cgroup_from_css(seq_css(m
));
5344 unsigned long stat
[MEMCG_NR_STAT
];
5345 unsigned long events
[MEMCG_NR_EVENTS
];
5349 * Provide statistics on the state of the memory subsystem as
5350 * well as cumulative event counters that show past behavior.
5352 * This list is ordered following a combination of these gradients:
5353 * 1) generic big picture -> specifics and details
5354 * 2) reflecting userspace activity -> reflecting kernel heuristics
5356 * Current memory state:
5359 tree_stat(memcg
, stat
);
5360 tree_events(memcg
, events
);
5362 seq_printf(m
, "anon %llu\n",
5363 (u64
)stat
[MEMCG_RSS
] * PAGE_SIZE
);
5364 seq_printf(m
, "file %llu\n",
5365 (u64
)stat
[MEMCG_CACHE
] * PAGE_SIZE
);
5366 seq_printf(m
, "kernel_stack %llu\n",
5367 (u64
)stat
[MEMCG_KERNEL_STACK_KB
] * 1024);
5368 seq_printf(m
, "slab %llu\n",
5369 (u64
)(stat
[NR_SLAB_RECLAIMABLE
] +
5370 stat
[NR_SLAB_UNRECLAIMABLE
]) * PAGE_SIZE
);
5371 seq_printf(m
, "sock %llu\n",
5372 (u64
)stat
[MEMCG_SOCK
] * PAGE_SIZE
);
5374 seq_printf(m
, "shmem %llu\n",
5375 (u64
)stat
[NR_SHMEM
] * PAGE_SIZE
);
5376 seq_printf(m
, "file_mapped %llu\n",
5377 (u64
)stat
[NR_FILE_MAPPED
] * PAGE_SIZE
);
5378 seq_printf(m
, "file_dirty %llu\n",
5379 (u64
)stat
[NR_FILE_DIRTY
] * PAGE_SIZE
);
5380 seq_printf(m
, "file_writeback %llu\n",
5381 (u64
)stat
[NR_WRITEBACK
] * PAGE_SIZE
);
5383 for (i
= 0; i
< NR_LRU_LISTS
; i
++) {
5384 struct mem_cgroup
*mi
;
5385 unsigned long val
= 0;
5387 for_each_mem_cgroup_tree(mi
, memcg
)
5388 val
+= mem_cgroup_nr_lru_pages(mi
, BIT(i
));
5389 seq_printf(m
, "%s %llu\n",
5390 mem_cgroup_lru_names
[i
], (u64
)val
* PAGE_SIZE
);
5393 seq_printf(m
, "slab_reclaimable %llu\n",
5394 (u64
)stat
[NR_SLAB_RECLAIMABLE
] * PAGE_SIZE
);
5395 seq_printf(m
, "slab_unreclaimable %llu\n",
5396 (u64
)stat
[NR_SLAB_UNRECLAIMABLE
] * PAGE_SIZE
);
5398 /* Accumulated memory events */
5400 seq_printf(m
, "pgfault %lu\n", events
[PGFAULT
]);
5401 seq_printf(m
, "pgmajfault %lu\n", events
[PGMAJFAULT
]);
5403 seq_printf(m
, "pgrefill %lu\n", events
[PGREFILL
]);
5404 seq_printf(m
, "pgscan %lu\n", events
[PGSCAN_KSWAPD
] +
5405 events
[PGSCAN_DIRECT
]);
5406 seq_printf(m
, "pgsteal %lu\n", events
[PGSTEAL_KSWAPD
] +
5407 events
[PGSTEAL_DIRECT
]);
5408 seq_printf(m
, "pgactivate %lu\n", events
[PGACTIVATE
]);
5409 seq_printf(m
, "pgdeactivate %lu\n", events
[PGDEACTIVATE
]);
5410 seq_printf(m
, "pglazyfree %lu\n", events
[PGLAZYFREE
]);
5411 seq_printf(m
, "pglazyfreed %lu\n", events
[PGLAZYFREED
]);
5413 seq_printf(m
, "workingset_refault %lu\n",
5414 stat
[WORKINGSET_REFAULT
]);
5415 seq_printf(m
, "workingset_activate %lu\n",
5416 stat
[WORKINGSET_ACTIVATE
]);
5417 seq_printf(m
, "workingset_nodereclaim %lu\n",
5418 stat
[WORKINGSET_NODERECLAIM
]);
5423 static struct cftype memory_files
[] = {
5426 .flags
= CFTYPE_NOT_ON_ROOT
,
5427 .read_u64
= memory_current_read
,
5431 .flags
= CFTYPE_NOT_ON_ROOT
,
5432 .seq_show
= memory_low_show
,
5433 .write
= memory_low_write
,
5437 .flags
= CFTYPE_NOT_ON_ROOT
,
5438 .seq_show
= memory_high_show
,
5439 .write
= memory_high_write
,
5443 .flags
= CFTYPE_NOT_ON_ROOT
,
5444 .seq_show
= memory_max_show
,
5445 .write
= memory_max_write
,
5449 .flags
= CFTYPE_NOT_ON_ROOT
,
5450 .file_offset
= offsetof(struct mem_cgroup
, events_file
),
5451 .seq_show
= memory_events_show
,
5455 .flags
= CFTYPE_NOT_ON_ROOT
,
5456 .seq_show
= memory_stat_show
,
5461 struct cgroup_subsys memory_cgrp_subsys
= {
5462 .css_alloc
= mem_cgroup_css_alloc
,
5463 .css_online
= mem_cgroup_css_online
,
5464 .css_offline
= mem_cgroup_css_offline
,
5465 .css_released
= mem_cgroup_css_released
,
5466 .css_free
= mem_cgroup_css_free
,
5467 .css_reset
= mem_cgroup_css_reset
,
5468 .can_attach
= mem_cgroup_can_attach
,
5469 .cancel_attach
= mem_cgroup_cancel_attach
,
5470 .post_attach
= mem_cgroup_move_task
,
5471 .bind
= mem_cgroup_bind
,
5472 .dfl_cftypes
= memory_files
,
5473 .legacy_cftypes
= mem_cgroup_legacy_files
,
5478 * mem_cgroup_low - check if memory consumption is below the normal range
5479 * @root: the top ancestor of the sub-tree being checked
5480 * @memcg: the memory cgroup to check
5482 * Returns %true if memory consumption of @memcg, and that of all
5483 * ancestors up to (but not including) @root, is below the normal range.
5485 * @root is exclusive; it is never low when looked at directly and isn't
5486 * checked when traversing the hierarchy.
5488 * Excluding @root enables using memory.low to prioritize memory usage
5489 * between cgroups within a subtree of the hierarchy that is limited by
5490 * memory.high or memory.max.
5492 * For example, given cgroup A with children B and C:
5500 * 1. A/memory.current > A/memory.high
5501 * 2. A/B/memory.current < A/B/memory.low
5502 * 3. A/C/memory.current >= A/C/memory.low
5504 * As 'A' is high, i.e. triggers reclaim from 'A', and 'B' is low, we
5505 * should reclaim from 'C' until 'A' is no longer high or until we can
5506 * no longer reclaim from 'C'. If 'A', i.e. @root, isn't excluded by
5507 * mem_cgroup_low when reclaming from 'A', then 'B' won't be considered
5508 * low and we will reclaim indiscriminately from both 'B' and 'C'.
5510 bool mem_cgroup_low(struct mem_cgroup
*root
, struct mem_cgroup
*memcg
)
5512 if (mem_cgroup_disabled())
5516 root
= root_mem_cgroup
;
5520 for (; memcg
!= root
; memcg
= parent_mem_cgroup(memcg
)) {
5521 if (page_counter_read(&memcg
->memory
) >= memcg
->low
)
5529 * mem_cgroup_try_charge - try charging a page
5530 * @page: page to charge
5531 * @mm: mm context of the victim
5532 * @gfp_mask: reclaim mode
5533 * @memcgp: charged memcg return
5534 * @compound: charge the page as compound or small page
5536 * Try to charge @page to the memcg that @mm belongs to, reclaiming
5537 * pages according to @gfp_mask if necessary.
5539 * Returns 0 on success, with *@memcgp pointing to the charged memcg.
5540 * Otherwise, an error code is returned.
5542 * After page->mapping has been set up, the caller must finalize the
5543 * charge with mem_cgroup_commit_charge(). Or abort the transaction
5544 * with mem_cgroup_cancel_charge() in case page instantiation fails.
5546 int mem_cgroup_try_charge(struct page
*page
, struct mm_struct
*mm
,
5547 gfp_t gfp_mask
, struct mem_cgroup
**memcgp
,
5550 struct mem_cgroup
*memcg
= NULL
;
5551 unsigned int nr_pages
= compound
? hpage_nr_pages(page
) : 1;
5554 if (mem_cgroup_disabled())
5557 if (PageSwapCache(page
)) {
5559 * Every swap fault against a single page tries to charge the
5560 * page, bail as early as possible. shmem_unuse() encounters
5561 * already charged pages, too. The USED bit is protected by
5562 * the page lock, which serializes swap cache removal, which
5563 * in turn serializes uncharging.
5565 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
5566 if (compound_head(page
)->mem_cgroup
)
5569 if (do_swap_account
) {
5570 swp_entry_t ent
= { .val
= page_private(page
), };
5571 unsigned short id
= lookup_swap_cgroup_id(ent
);
5574 memcg
= mem_cgroup_from_id(id
);
5575 if (memcg
&& !css_tryget_online(&memcg
->css
))
5582 memcg
= get_mem_cgroup_from_mm(mm
);
5584 ret
= try_charge(memcg
, gfp_mask
, nr_pages
);
5586 css_put(&memcg
->css
);
5593 * mem_cgroup_commit_charge - commit a page charge
5594 * @page: page to charge
5595 * @memcg: memcg to charge the page to
5596 * @lrucare: page might be on LRU already
5597 * @compound: charge the page as compound or small page
5599 * Finalize a charge transaction started by mem_cgroup_try_charge(),
5600 * after page->mapping has been set up. This must happen atomically
5601 * as part of the page instantiation, i.e. under the page table lock
5602 * for anonymous pages, under the page lock for page and swap cache.
5604 * In addition, the page must not be on the LRU during the commit, to
5605 * prevent racing with task migration. If it might be, use @lrucare.
5607 * Use mem_cgroup_cancel_charge() to cancel the transaction instead.
5609 void mem_cgroup_commit_charge(struct page
*page
, struct mem_cgroup
*memcg
,
5610 bool lrucare
, bool compound
)
5612 unsigned int nr_pages
= compound
? hpage_nr_pages(page
) : 1;
5614 VM_BUG_ON_PAGE(!page
->mapping
, page
);
5615 VM_BUG_ON_PAGE(PageLRU(page
) && !lrucare
, page
);
5617 if (mem_cgroup_disabled())
5620 * Swap faults will attempt to charge the same page multiple
5621 * times. But reuse_swap_page() might have removed the page
5622 * from swapcache already, so we can't check PageSwapCache().
5627 commit_charge(page
, memcg
, lrucare
);
5629 local_irq_disable();
5630 mem_cgroup_charge_statistics(memcg
, page
, compound
, nr_pages
);
5631 memcg_check_events(memcg
, page
);
5634 if (do_memsw_account() && PageSwapCache(page
)) {
5635 swp_entry_t entry
= { .val
= page_private(page
) };
5637 * The swap entry might not get freed for a long time,
5638 * let's not wait for it. The page already received a
5639 * memory+swap charge, drop the swap entry duplicate.
5641 mem_cgroup_uncharge_swap(entry
, nr_pages
);
5646 * mem_cgroup_cancel_charge - cancel a page charge
5647 * @page: page to charge
5648 * @memcg: memcg to charge the page to
5649 * @compound: charge the page as compound or small page
5651 * Cancel a charge transaction started by mem_cgroup_try_charge().
5653 void mem_cgroup_cancel_charge(struct page
*page
, struct mem_cgroup
*memcg
,
5656 unsigned int nr_pages
= compound
? hpage_nr_pages(page
) : 1;
5658 if (mem_cgroup_disabled())
5661 * Swap faults will attempt to charge the same page multiple
5662 * times. But reuse_swap_page() might have removed the page
5663 * from swapcache already, so we can't check PageSwapCache().
5668 cancel_charge(memcg
, nr_pages
);
5671 struct uncharge_gather
{
5672 struct mem_cgroup
*memcg
;
5673 unsigned long pgpgout
;
5674 unsigned long nr_anon
;
5675 unsigned long nr_file
;
5676 unsigned long nr_kmem
;
5677 unsigned long nr_huge
;
5678 unsigned long nr_shmem
;
5679 struct page
*dummy_page
;
5682 static inline void uncharge_gather_clear(struct uncharge_gather
*ug
)
5684 memset(ug
, 0, sizeof(*ug
));
5687 static void uncharge_batch(const struct uncharge_gather
*ug
)
5689 unsigned long nr_pages
= ug
->nr_anon
+ ug
->nr_file
+ ug
->nr_kmem
;
5690 unsigned long flags
;
5692 if (!mem_cgroup_is_root(ug
->memcg
)) {
5693 page_counter_uncharge(&ug
->memcg
->memory
, nr_pages
);
5694 if (do_memsw_account())
5695 page_counter_uncharge(&ug
->memcg
->memsw
, nr_pages
);
5696 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys
) && ug
->nr_kmem
)
5697 page_counter_uncharge(&ug
->memcg
->kmem
, ug
->nr_kmem
);
5698 memcg_oom_recover(ug
->memcg
);
5701 local_irq_save(flags
);
5702 __this_cpu_sub(ug
->memcg
->stat
->count
[MEMCG_RSS
], ug
->nr_anon
);
5703 __this_cpu_sub(ug
->memcg
->stat
->count
[MEMCG_CACHE
], ug
->nr_file
);
5704 __this_cpu_sub(ug
->memcg
->stat
->count
[MEMCG_RSS_HUGE
], ug
->nr_huge
);
5705 __this_cpu_sub(ug
->memcg
->stat
->count
[NR_SHMEM
], ug
->nr_shmem
);
5706 __this_cpu_add(ug
->memcg
->stat
->events
[PGPGOUT
], ug
->pgpgout
);
5707 __this_cpu_add(ug
->memcg
->stat
->nr_page_events
, nr_pages
);
5708 memcg_check_events(ug
->memcg
, ug
->dummy_page
);
5709 local_irq_restore(flags
);
5711 if (!mem_cgroup_is_root(ug
->memcg
))
5712 css_put_many(&ug
->memcg
->css
, nr_pages
);
5715 static void uncharge_page(struct page
*page
, struct uncharge_gather
*ug
)
5717 VM_BUG_ON_PAGE(PageLRU(page
), page
);
5718 VM_BUG_ON_PAGE(page_count(page
) && !is_zone_device_page(page
) &&
5719 !PageHWPoison(page
) , page
);
5721 if (!page
->mem_cgroup
)
5725 * Nobody should be changing or seriously looking at
5726 * page->mem_cgroup at this point, we have fully
5727 * exclusive access to the page.
5730 if (ug
->memcg
!= page
->mem_cgroup
) {
5733 uncharge_gather_clear(ug
);
5735 ug
->memcg
= page
->mem_cgroup
;
5738 if (!PageKmemcg(page
)) {
5739 unsigned int nr_pages
= 1;
5741 if (PageTransHuge(page
)) {
5742 nr_pages
<<= compound_order(page
);
5743 ug
->nr_huge
+= nr_pages
;
5746 ug
->nr_anon
+= nr_pages
;
5748 ug
->nr_file
+= nr_pages
;
5749 if (PageSwapBacked(page
))
5750 ug
->nr_shmem
+= nr_pages
;
5754 ug
->nr_kmem
+= 1 << compound_order(page
);
5755 __ClearPageKmemcg(page
);
5758 ug
->dummy_page
= page
;
5759 page
->mem_cgroup
= NULL
;
5762 static void uncharge_list(struct list_head
*page_list
)
5764 struct uncharge_gather ug
;
5765 struct list_head
*next
;
5767 uncharge_gather_clear(&ug
);
5770 * Note that the list can be a single page->lru; hence the
5771 * do-while loop instead of a simple list_for_each_entry().
5773 next
= page_list
->next
;
5777 page
= list_entry(next
, struct page
, lru
);
5778 next
= page
->lru
.next
;
5780 uncharge_page(page
, &ug
);
5781 } while (next
!= page_list
);
5784 uncharge_batch(&ug
);
5788 * mem_cgroup_uncharge - uncharge a page
5789 * @page: page to uncharge
5791 * Uncharge a page previously charged with mem_cgroup_try_charge() and
5792 * mem_cgroup_commit_charge().
5794 void mem_cgroup_uncharge(struct page
*page
)
5796 struct uncharge_gather ug
;
5798 if (mem_cgroup_disabled())
5801 /* Don't touch page->lru of any random page, pre-check: */
5802 if (!page
->mem_cgroup
)
5805 uncharge_gather_clear(&ug
);
5806 uncharge_page(page
, &ug
);
5807 uncharge_batch(&ug
);
5811 * mem_cgroup_uncharge_list - uncharge a list of page
5812 * @page_list: list of pages to uncharge
5814 * Uncharge a list of pages previously charged with
5815 * mem_cgroup_try_charge() and mem_cgroup_commit_charge().
5817 void mem_cgroup_uncharge_list(struct list_head
*page_list
)
5819 if (mem_cgroup_disabled())
5822 if (!list_empty(page_list
))
5823 uncharge_list(page_list
);
5827 * mem_cgroup_migrate - charge a page's replacement
5828 * @oldpage: currently circulating page
5829 * @newpage: replacement page
5831 * Charge @newpage as a replacement page for @oldpage. @oldpage will
5832 * be uncharged upon free.
5834 * Both pages must be locked, @newpage->mapping must be set up.
5836 void mem_cgroup_migrate(struct page
*oldpage
, struct page
*newpage
)
5838 struct mem_cgroup
*memcg
;
5839 unsigned int nr_pages
;
5841 unsigned long flags
;
5843 VM_BUG_ON_PAGE(!PageLocked(oldpage
), oldpage
);
5844 VM_BUG_ON_PAGE(!PageLocked(newpage
), newpage
);
5845 VM_BUG_ON_PAGE(PageAnon(oldpage
) != PageAnon(newpage
), newpage
);
5846 VM_BUG_ON_PAGE(PageTransHuge(oldpage
) != PageTransHuge(newpage
),
5849 if (mem_cgroup_disabled())
5852 /* Page cache replacement: new page already charged? */
5853 if (newpage
->mem_cgroup
)
5856 /* Swapcache readahead pages can get replaced before being charged */
5857 memcg
= oldpage
->mem_cgroup
;
5861 /* Force-charge the new page. The old one will be freed soon */
5862 compound
= PageTransHuge(newpage
);
5863 nr_pages
= compound
? hpage_nr_pages(newpage
) : 1;
5865 page_counter_charge(&memcg
->memory
, nr_pages
);
5866 if (do_memsw_account())
5867 page_counter_charge(&memcg
->memsw
, nr_pages
);
5868 css_get_many(&memcg
->css
, nr_pages
);
5870 commit_charge(newpage
, memcg
, false);
5872 local_irq_save(flags
);
5873 mem_cgroup_charge_statistics(memcg
, newpage
, compound
, nr_pages
);
5874 memcg_check_events(memcg
, newpage
);
5875 local_irq_restore(flags
);
5878 DEFINE_STATIC_KEY_FALSE(memcg_sockets_enabled_key
);
5879 EXPORT_SYMBOL(memcg_sockets_enabled_key
);
5881 void mem_cgroup_sk_alloc(struct sock
*sk
)
5883 struct mem_cgroup
*memcg
;
5885 if (!mem_cgroup_sockets_enabled
)
5889 * Socket cloning can throw us here with sk_memcg already
5890 * filled. It won't however, necessarily happen from
5891 * process context. So the test for root memcg given
5892 * the current task's memcg won't help us in this case.
5894 * Respecting the original socket's memcg is a better
5895 * decision in this case.
5898 css_get(&sk
->sk_memcg
->css
);
5903 memcg
= mem_cgroup_from_task(current
);
5904 if (memcg
== root_mem_cgroup
)
5906 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys
) && !memcg
->tcpmem_active
)
5908 if (css_tryget_online(&memcg
->css
))
5909 sk
->sk_memcg
= memcg
;
5914 void mem_cgroup_sk_free(struct sock
*sk
)
5917 css_put(&sk
->sk_memcg
->css
);
5921 * mem_cgroup_charge_skmem - charge socket memory
5922 * @memcg: memcg to charge
5923 * @nr_pages: number of pages to charge
5925 * Charges @nr_pages to @memcg. Returns %true if the charge fit within
5926 * @memcg's configured limit, %false if the charge had to be forced.
5928 bool mem_cgroup_charge_skmem(struct mem_cgroup
*memcg
, unsigned int nr_pages
)
5930 gfp_t gfp_mask
= GFP_KERNEL
;
5932 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys
)) {
5933 struct page_counter
*fail
;
5935 if (page_counter_try_charge(&memcg
->tcpmem
, nr_pages
, &fail
)) {
5936 memcg
->tcpmem_pressure
= 0;
5939 page_counter_charge(&memcg
->tcpmem
, nr_pages
);
5940 memcg
->tcpmem_pressure
= 1;
5944 /* Don't block in the packet receive path */
5946 gfp_mask
= GFP_NOWAIT
;
5948 this_cpu_add(memcg
->stat
->count
[MEMCG_SOCK
], nr_pages
);
5950 if (try_charge(memcg
, gfp_mask
, nr_pages
) == 0)
5953 try_charge(memcg
, gfp_mask
|__GFP_NOFAIL
, nr_pages
);
5958 * mem_cgroup_uncharge_skmem - uncharge socket memory
5959 * @memcg - memcg to uncharge
5960 * @nr_pages - number of pages to uncharge
5962 void mem_cgroup_uncharge_skmem(struct mem_cgroup
*memcg
, unsigned int nr_pages
)
5964 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys
)) {
5965 page_counter_uncharge(&memcg
->tcpmem
, nr_pages
);
5969 this_cpu_sub(memcg
->stat
->count
[MEMCG_SOCK
], nr_pages
);
5971 refill_stock(memcg
, nr_pages
);
5974 static int __init
cgroup_memory(char *s
)
5978 while ((token
= strsep(&s
, ",")) != NULL
) {
5981 if (!strcmp(token
, "nosocket"))
5982 cgroup_memory_nosocket
= true;
5983 if (!strcmp(token
, "nokmem"))
5984 cgroup_memory_nokmem
= true;
5988 __setup("cgroup.memory=", cgroup_memory
);
5991 * subsys_initcall() for memory controller.
5993 * Some parts like memcg_hotplug_cpu_dead() have to be initialized from this
5994 * context because of lock dependencies (cgroup_lock -> cpu hotplug) but
5995 * basically everything that doesn't depend on a specific mem_cgroup structure
5996 * should be initialized from here.
5998 static int __init
mem_cgroup_init(void)
6004 * Kmem cache creation is mostly done with the slab_mutex held,
6005 * so use a workqueue with limited concurrency to avoid stalling
6006 * all worker threads in case lots of cgroups are created and
6007 * destroyed simultaneously.
6009 memcg_kmem_cache_wq
= alloc_workqueue("memcg_kmem_cache", 0, 1);
6010 BUG_ON(!memcg_kmem_cache_wq
);
6013 cpuhp_setup_state_nocalls(CPUHP_MM_MEMCQ_DEAD
, "mm/memctrl:dead", NULL
,
6014 memcg_hotplug_cpu_dead
);
6016 for_each_possible_cpu(cpu
)
6017 INIT_WORK(&per_cpu_ptr(&memcg_stock
, cpu
)->work
,
6020 for_each_node(node
) {
6021 struct mem_cgroup_tree_per_node
*rtpn
;
6023 rtpn
= kzalloc_node(sizeof(*rtpn
), GFP_KERNEL
,
6024 node_online(node
) ? node
: NUMA_NO_NODE
);
6026 rtpn
->rb_root
= RB_ROOT
;
6027 rtpn
->rb_rightmost
= NULL
;
6028 spin_lock_init(&rtpn
->lock
);
6029 soft_limit_tree
.rb_tree_per_node
[node
] = rtpn
;
6034 subsys_initcall(mem_cgroup_init
);
6036 #ifdef CONFIG_MEMCG_SWAP
6037 static struct mem_cgroup
*mem_cgroup_id_get_online(struct mem_cgroup
*memcg
)
6039 while (!atomic_inc_not_zero(&memcg
->id
.ref
)) {
6041 * The root cgroup cannot be destroyed, so it's refcount must
6044 if (WARN_ON_ONCE(memcg
== root_mem_cgroup
)) {
6048 memcg
= parent_mem_cgroup(memcg
);
6050 memcg
= root_mem_cgroup
;
6056 * mem_cgroup_swapout - transfer a memsw charge to swap
6057 * @page: page whose memsw charge to transfer
6058 * @entry: swap entry to move the charge to
6060 * Transfer the memsw charge of @page to @entry.
6062 void mem_cgroup_swapout(struct page
*page
, swp_entry_t entry
)
6064 struct mem_cgroup
*memcg
, *swap_memcg
;
6065 unsigned int nr_entries
;
6066 unsigned short oldid
;
6068 VM_BUG_ON_PAGE(PageLRU(page
), page
);
6069 VM_BUG_ON_PAGE(page_count(page
), page
);
6071 if (!do_memsw_account())
6074 memcg
= page
->mem_cgroup
;
6076 /* Readahead page, never charged */
6081 * In case the memcg owning these pages has been offlined and doesn't
6082 * have an ID allocated to it anymore, charge the closest online
6083 * ancestor for the swap instead and transfer the memory+swap charge.
6085 swap_memcg
= mem_cgroup_id_get_online(memcg
);
6086 nr_entries
= hpage_nr_pages(page
);
6087 /* Get references for the tail pages, too */
6089 mem_cgroup_id_get_many(swap_memcg
, nr_entries
- 1);
6090 oldid
= swap_cgroup_record(entry
, mem_cgroup_id(swap_memcg
),
6092 VM_BUG_ON_PAGE(oldid
, page
);
6093 mem_cgroup_swap_statistics(swap_memcg
, nr_entries
);
6095 page
->mem_cgroup
= NULL
;
6097 if (!mem_cgroup_is_root(memcg
))
6098 page_counter_uncharge(&memcg
->memory
, nr_entries
);
6100 if (memcg
!= swap_memcg
) {
6101 if (!mem_cgroup_is_root(swap_memcg
))
6102 page_counter_charge(&swap_memcg
->memsw
, nr_entries
);
6103 page_counter_uncharge(&memcg
->memsw
, nr_entries
);
6107 * Interrupts should be disabled here because the caller holds the
6108 * mapping->tree_lock lock which is taken with interrupts-off. It is
6109 * important here to have the interrupts disabled because it is the
6110 * only synchronisation we have for udpating the per-CPU variables.
6112 VM_BUG_ON(!irqs_disabled());
6113 mem_cgroup_charge_statistics(memcg
, page
, PageTransHuge(page
),
6115 memcg_check_events(memcg
, page
);
6117 if (!mem_cgroup_is_root(memcg
))
6118 css_put_many(&memcg
->css
, nr_entries
);
6122 * mem_cgroup_try_charge_swap - try charging swap space for a page
6123 * @page: page being added to swap
6124 * @entry: swap entry to charge
6126 * Try to charge @page's memcg for the swap space at @entry.
6128 * Returns 0 on success, -ENOMEM on failure.
6130 int mem_cgroup_try_charge_swap(struct page
*page
, swp_entry_t entry
)
6132 unsigned int nr_pages
= hpage_nr_pages(page
);
6133 struct page_counter
*counter
;
6134 struct mem_cgroup
*memcg
;
6135 unsigned short oldid
;
6137 if (!cgroup_subsys_on_dfl(memory_cgrp_subsys
) || !do_swap_account
)
6140 memcg
= page
->mem_cgroup
;
6142 /* Readahead page, never charged */
6146 memcg
= mem_cgroup_id_get_online(memcg
);
6148 if (!mem_cgroup_is_root(memcg
) &&
6149 !page_counter_try_charge(&memcg
->swap
, nr_pages
, &counter
)) {
6150 mem_cgroup_id_put(memcg
);
6154 /* Get references for the tail pages, too */
6156 mem_cgroup_id_get_many(memcg
, nr_pages
- 1);
6157 oldid
= swap_cgroup_record(entry
, mem_cgroup_id(memcg
), nr_pages
);
6158 VM_BUG_ON_PAGE(oldid
, page
);
6159 mem_cgroup_swap_statistics(memcg
, nr_pages
);
6165 * mem_cgroup_uncharge_swap - uncharge swap space
6166 * @entry: swap entry to uncharge
6167 * @nr_pages: the amount of swap space to uncharge
6169 void mem_cgroup_uncharge_swap(swp_entry_t entry
, unsigned int nr_pages
)
6171 struct mem_cgroup
*memcg
;
6174 if (!do_swap_account
)
6177 id
= swap_cgroup_record(entry
, 0, nr_pages
);
6179 memcg
= mem_cgroup_from_id(id
);
6181 if (!mem_cgroup_is_root(memcg
)) {
6182 if (cgroup_subsys_on_dfl(memory_cgrp_subsys
))
6183 page_counter_uncharge(&memcg
->swap
, nr_pages
);
6185 page_counter_uncharge(&memcg
->memsw
, nr_pages
);
6187 mem_cgroup_swap_statistics(memcg
, -nr_pages
);
6188 mem_cgroup_id_put_many(memcg
, nr_pages
);
6193 long mem_cgroup_get_nr_swap_pages(struct mem_cgroup
*memcg
)
6195 long nr_swap_pages
= get_nr_swap_pages();
6197 if (!do_swap_account
|| !cgroup_subsys_on_dfl(memory_cgrp_subsys
))
6198 return nr_swap_pages
;
6199 for (; memcg
!= root_mem_cgroup
; memcg
= parent_mem_cgroup(memcg
))
6200 nr_swap_pages
= min_t(long, nr_swap_pages
,
6201 READ_ONCE(memcg
->swap
.limit
) -
6202 page_counter_read(&memcg
->swap
));
6203 return nr_swap_pages
;
6206 bool mem_cgroup_swap_full(struct page
*page
)
6208 struct mem_cgroup
*memcg
;
6210 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
6214 if (!do_swap_account
|| !cgroup_subsys_on_dfl(memory_cgrp_subsys
))
6217 memcg
= page
->mem_cgroup
;
6221 for (; memcg
!= root_mem_cgroup
; memcg
= parent_mem_cgroup(memcg
))
6222 if (page_counter_read(&memcg
->swap
) * 2 >= memcg
->swap
.limit
)
6228 /* for remember boot option*/
6229 #ifdef CONFIG_MEMCG_SWAP_ENABLED
6230 static int really_do_swap_account __initdata
= 1;
6232 static int really_do_swap_account __initdata
;
6235 static int __init
enable_swap_account(char *s
)
6237 if (!strcmp(s
, "1"))
6238 really_do_swap_account
= 1;
6239 else if (!strcmp(s
, "0"))
6240 really_do_swap_account
= 0;
6243 __setup("swapaccount=", enable_swap_account
);
6245 static u64
swap_current_read(struct cgroup_subsys_state
*css
,
6248 struct mem_cgroup
*memcg
= mem_cgroup_from_css(css
);
6250 return (u64
)page_counter_read(&memcg
->swap
) * PAGE_SIZE
;
6253 static int swap_max_show(struct seq_file
*m
, void *v
)
6255 struct mem_cgroup
*memcg
= mem_cgroup_from_css(seq_css(m
));
6256 unsigned long max
= READ_ONCE(memcg
->swap
.limit
);
6258 if (max
== PAGE_COUNTER_MAX
)
6259 seq_puts(m
, "max\n");
6261 seq_printf(m
, "%llu\n", (u64
)max
* PAGE_SIZE
);
6266 static ssize_t
swap_max_write(struct kernfs_open_file
*of
,
6267 char *buf
, size_t nbytes
, loff_t off
)
6269 struct mem_cgroup
*memcg
= mem_cgroup_from_css(of_css(of
));
6273 buf
= strstrip(buf
);
6274 err
= page_counter_memparse(buf
, "max", &max
);
6278 mutex_lock(&memcg_limit_mutex
);
6279 err
= page_counter_limit(&memcg
->swap
, max
);
6280 mutex_unlock(&memcg_limit_mutex
);
6287 static struct cftype swap_files
[] = {
6289 .name
= "swap.current",
6290 .flags
= CFTYPE_NOT_ON_ROOT
,
6291 .read_u64
= swap_current_read
,
6295 .flags
= CFTYPE_NOT_ON_ROOT
,
6296 .seq_show
= swap_max_show
,
6297 .write
= swap_max_write
,
6302 static struct cftype memsw_cgroup_files
[] = {
6304 .name
= "memsw.usage_in_bytes",
6305 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_USAGE
),
6306 .read_u64
= mem_cgroup_read_u64
,
6309 .name
= "memsw.max_usage_in_bytes",
6310 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_MAX_USAGE
),
6311 .write
= mem_cgroup_reset
,
6312 .read_u64
= mem_cgroup_read_u64
,
6315 .name
= "memsw.limit_in_bytes",
6316 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_LIMIT
),
6317 .write
= mem_cgroup_write
,
6318 .read_u64
= mem_cgroup_read_u64
,
6321 .name
= "memsw.failcnt",
6322 .private = MEMFILE_PRIVATE(_MEMSWAP
, RES_FAILCNT
),
6323 .write
= mem_cgroup_reset
,
6324 .read_u64
= mem_cgroup_read_u64
,
6326 { }, /* terminate */
6329 static int __init
mem_cgroup_swap_init(void)
6331 if (!mem_cgroup_disabled() && really_do_swap_account
) {
6332 do_swap_account
= 1;
6333 WARN_ON(cgroup_add_dfl_cftypes(&memory_cgrp_subsys
,
6335 WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys
,
6336 memsw_cgroup_files
));
6340 subsys_initcall(mem_cgroup_swap_init
);
6342 #endif /* CONFIG_MEMCG_SWAP */