Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 1 | /* memcontrol.c - Memory Controller |
| 2 | * |
| 3 | * Copyright IBM Corporation, 2007 |
| 4 | * Author Balbir Singh <balbir@linux.vnet.ibm.com> |
| 5 | * |
Pavel Emelianov | 78fb746 | 2008-02-07 00:13:51 -0800 | [diff] [blame] | 6 | * Copyright 2007 OpenVZ SWsoft Inc |
| 7 | * Author: Pavel Emelianov <xemul@openvz.org> |
| 8 | * |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 9 | * Memory thresholds |
| 10 | * Copyright (C) 2009 Nokia Corporation |
| 11 | * Author: Kirill A. Shutemov |
| 12 | * |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 13 | * Kernel Memory Controller |
| 14 | * Copyright (C) 2012 Parallels Inc. and Google Inc. |
| 15 | * Authors: Glauber Costa and Suleiman Souhlal |
| 16 | * |
Johannes Weiner | 1575e68 | 2015-04-14 15:44:51 -0700 | [diff] [blame] | 17 | * Native page reclaim |
| 18 | * Charge lifetime sanitation |
| 19 | * Lockless page tracking & accounting |
| 20 | * Unified hierarchy configuration model |
| 21 | * Copyright (C) 2015 Red Hat, Inc., Johannes Weiner |
| 22 | * |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 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. |
| 27 | * |
| 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. |
| 32 | */ |
| 33 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 34 | #include <linux/page_counter.h> |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 35 | #include <linux/memcontrol.h> |
| 36 | #include <linux/cgroup.h> |
Pavel Emelianov | 78fb746 | 2008-02-07 00:13:51 -0800 | [diff] [blame] | 37 | #include <linux/mm.h> |
Ingo Molnar | 6e84f31 | 2017-02-08 18:51:29 +0100 | [diff] [blame] | 38 | #include <linux/sched/mm.h> |
Hugh Dickins | 3a4f8a0 | 2017-02-24 14:59:36 -0800 | [diff] [blame] | 39 | #include <linux/shmem_fs.h> |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 40 | #include <linux/hugetlb.h> |
KAMEZAWA Hiroyuki | d13d144 | 2009-01-07 18:07:56 -0800 | [diff] [blame] | 41 | #include <linux/pagemap.h> |
KAMEZAWA Hiroyuki | d52aa41 | 2008-02-07 00:14:24 -0800 | [diff] [blame] | 42 | #include <linux/smp.h> |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 43 | #include <linux/page-flags.h> |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 44 | #include <linux/backing-dev.h> |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 45 | #include <linux/bit_spinlock.h> |
| 46 | #include <linux/rcupdate.h> |
Balbir Singh | e222432 | 2009-04-02 16:57:39 -0700 | [diff] [blame] | 47 | #include <linux/limits.h> |
Paul Gortmaker | b9e15ba | 2011-05-26 16:00:52 -0400 | [diff] [blame] | 48 | #include <linux/export.h> |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 49 | #include <linux/mutex.h> |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 50 | #include <linux/rbtree.h> |
Balbir Singh | b6ac57d | 2008-04-29 01:00:19 -0700 | [diff] [blame] | 51 | #include <linux/slab.h> |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 52 | #include <linux/swap.h> |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 53 | #include <linux/swapops.h> |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 54 | #include <linux/spinlock.h> |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 55 | #include <linux/eventfd.h> |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 56 | #include <linux/poll.h> |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 57 | #include <linux/sort.h> |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 58 | #include <linux/fs.h> |
KAMEZAWA Hiroyuki | d2ceb9b | 2008-02-07 00:14:25 -0800 | [diff] [blame] | 59 | #include <linux/seq_file.h> |
Anton Vorontsov | 70ddf63 | 2013-04-29 15:08:31 -0700 | [diff] [blame] | 60 | #include <linux/vmpressure.h> |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 61 | #include <linux/mm_inline.h> |
Johannes Weiner | 5d1ea48 | 2014-12-10 15:44:55 -0800 | [diff] [blame] | 62 | #include <linux/swap_cgroup.h> |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 63 | #include <linux/cpu.h> |
KAMEZAWA Hiroyuki | 158e0a2 | 2010-08-10 18:03:00 -0700 | [diff] [blame] | 64 | #include <linux/oom.h> |
Johannes Weiner | 0056f4e | 2013-10-31 16:34:14 -0700 | [diff] [blame] | 65 | #include <linux/lockdep.h> |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 66 | #include <linux/file.h> |
Tejun Heo | b23afb9 | 2015-11-05 18:46:11 -0800 | [diff] [blame] | 67 | #include <linux/tracehook.h> |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 68 | #include "internal.h" |
Glauber Costa | d1a4c0b | 2011-12-11 21:47:04 +0000 | [diff] [blame] | 69 | #include <net/sock.h> |
Michal Hocko | 4bd2c1e | 2012-10-08 16:33:10 -0700 | [diff] [blame] | 70 | #include <net/ip.h> |
Qiang Huang | f35c3a8 | 2013-11-12 15:08:22 -0800 | [diff] [blame] | 71 | #include "slab.h" |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 72 | |
Linus Torvalds | 7c0f6ba | 2016-12-24 11:46:01 -0800 | [diff] [blame] | 73 | #include <linux/uaccess.h> |
Balbir Singh | 8697d33 | 2008-02-07 00:13:59 -0800 | [diff] [blame] | 74 | |
KOSAKI Motohiro | cc8e970 | 2010-08-09 17:19:57 -0700 | [diff] [blame] | 75 | #include <trace/events/vmscan.h> |
| 76 | |
Tejun Heo | 073219e | 2014-02-08 10:36:58 -0500 | [diff] [blame] | 77 | struct cgroup_subsys memory_cgrp_subsys __read_mostly; |
| 78 | EXPORT_SYMBOL(memory_cgrp_subsys); |
David Rientjes | 68ae564 | 2012-12-12 13:51:57 -0800 | [diff] [blame] | 79 | |
Johannes Weiner | 7d82860 | 2016-01-14 15:20:56 -0800 | [diff] [blame] | 80 | struct mem_cgroup *root_mem_cgroup __read_mostly; |
| 81 | |
KAMEZAWA Hiroyuki | a181b0e | 2008-07-25 01:47:08 -0700 | [diff] [blame] | 82 | #define MEM_CGROUP_RECLAIM_RETRIES 5 |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 83 | |
Johannes Weiner | f7e1cb6 | 2016-01-14 15:21:29 -0800 | [diff] [blame] | 84 | /* Socket memory accounting disabled? */ |
| 85 | static bool cgroup_memory_nosocket; |
| 86 | |
Vladimir Davydov | 04823c8 | 2016-01-20 15:02:38 -0800 | [diff] [blame] | 87 | /* Kernel memory accounting disabled? */ |
| 88 | static bool cgroup_memory_nokmem; |
| 89 | |
Johannes Weiner | 21afa38 | 2015-02-11 15:26:36 -0800 | [diff] [blame] | 90 | /* Whether the swap controller is active */ |
Andrew Morton | c255a45 | 2012-07-31 16:43:02 -0700 | [diff] [blame] | 91 | #ifdef CONFIG_MEMCG_SWAP |
KAMEZAWA Hiroyuki | c077719 | 2009-01-07 18:07:57 -0800 | [diff] [blame] | 92 | int do_swap_account __read_mostly; |
KAMEZAWA Hiroyuki | c077719 | 2009-01-07 18:07:57 -0800 | [diff] [blame] | 93 | #else |
Kirill A. Shutemov | a0db00f | 2012-05-29 15:06:56 -0700 | [diff] [blame] | 94 | #define do_swap_account 0 |
KAMEZAWA Hiroyuki | c077719 | 2009-01-07 18:07:57 -0800 | [diff] [blame] | 95 | #endif |
| 96 | |
Johannes Weiner | 7941d21 | 2016-01-14 15:21:23 -0800 | [diff] [blame] | 97 | /* Whether legacy memory+swap accounting is active */ |
| 98 | static bool do_memsw_account(void) |
| 99 | { |
| 100 | return !cgroup_subsys_on_dfl(memory_cgrp_subsys) && do_swap_account; |
| 101 | } |
| 102 | |
Johannes Weiner | 71cd311 | 2017-05-03 14:55:13 -0700 | [diff] [blame] | 103 | static const char *const mem_cgroup_lru_names[] = { |
Sha Zhengju | 58cf188 | 2013-02-22 16:32:05 -0800 | [diff] [blame] | 104 | "inactive_anon", |
| 105 | "active_anon", |
| 106 | "inactive_file", |
| 107 | "active_file", |
| 108 | "unevictable", |
| 109 | }; |
| 110 | |
Kirill A. Shutemov | a0db00f | 2012-05-29 15:06:56 -0700 | [diff] [blame] | 111 | #define THRESHOLDS_EVENTS_TARGET 128 |
| 112 | #define SOFTLIMIT_EVENTS_TARGET 1024 |
| 113 | #define NUMAINFO_EVENTS_TARGET 1024 |
Johannes Weiner | e9f8974 | 2011-03-23 16:42:37 -0700 | [diff] [blame] | 114 | |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 115 | /* |
| 116 | * Cgroups above their limits are maintained in a RB-Tree, independent of |
| 117 | * their hierarchy representation |
| 118 | */ |
| 119 | |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 120 | struct mem_cgroup_tree_per_node { |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 121 | struct rb_root rb_root; |
| 122 | spinlock_t lock; |
| 123 | }; |
| 124 | |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 125 | struct mem_cgroup_tree { |
| 126 | struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES]; |
| 127 | }; |
| 128 | |
| 129 | static struct mem_cgroup_tree soft_limit_tree __read_mostly; |
| 130 | |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 131 | /* for OOM */ |
| 132 | struct mem_cgroup_eventfd_list { |
| 133 | struct list_head list; |
| 134 | struct eventfd_ctx *eventfd; |
| 135 | }; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 136 | |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 137 | /* |
| 138 | * cgroup_event represents events which userspace want to receive. |
| 139 | */ |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 140 | struct mem_cgroup_event { |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 141 | /* |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 142 | * memcg which the event belongs to. |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 143 | */ |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 144 | struct mem_cgroup *memcg; |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 145 | /* |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 146 | * eventfd to signal userspace about the event. |
| 147 | */ |
| 148 | struct eventfd_ctx *eventfd; |
| 149 | /* |
| 150 | * Each of these stored in a list by the cgroup. |
| 151 | */ |
| 152 | struct list_head list; |
| 153 | /* |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 154 | * register_event() callback will be used to add new userspace |
| 155 | * waiter for changes related to this event. Use eventfd_signal() |
| 156 | * on eventfd to send notification to userspace. |
| 157 | */ |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 158 | int (*register_event)(struct mem_cgroup *memcg, |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 159 | struct eventfd_ctx *eventfd, const char *args); |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 160 | /* |
| 161 | * unregister_event() callback will be called when userspace closes |
| 162 | * the eventfd or on cgroup removing. This callback must be set, |
| 163 | * if you want provide notification functionality. |
| 164 | */ |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 165 | void (*unregister_event)(struct mem_cgroup *memcg, |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 166 | struct eventfd_ctx *eventfd); |
| 167 | /* |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 168 | * All fields below needed to unregister event when |
| 169 | * userspace closes eventfd. |
| 170 | */ |
| 171 | poll_table pt; |
| 172 | wait_queue_head_t *wqh; |
Ingo Molnar | ac6424b | 2017-06-20 12:06:13 +0200 | [diff] [blame] | 173 | wait_queue_entry_t wait; |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 174 | struct work_struct remove; |
| 175 | }; |
| 176 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 177 | static void mem_cgroup_threshold(struct mem_cgroup *memcg); |
| 178 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg); |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 179 | |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 180 | /* Stuffs for move charges at task migration. */ |
| 181 | /* |
Johannes Weiner | 1dfab5a | 2015-02-11 15:26:09 -0800 | [diff] [blame] | 182 | * Types of charges to be moved. |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 183 | */ |
Johannes Weiner | 1dfab5a | 2015-02-11 15:26:09 -0800 | [diff] [blame] | 184 | #define MOVE_ANON 0x1U |
| 185 | #define MOVE_FILE 0x2U |
| 186 | #define MOVE_MASK (MOVE_ANON | MOVE_FILE) |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 187 | |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 188 | /* "mc" and its members are protected by cgroup_mutex */ |
| 189 | static struct move_charge_struct { |
Daisuke Nishimura | b1dd693 | 2010-11-24 12:57:06 -0800 | [diff] [blame] | 190 | spinlock_t lock; /* for from, to */ |
Tejun Heo | 264a0ae | 2016-04-21 19:09:02 -0400 | [diff] [blame] | 191 | struct mm_struct *mm; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 192 | struct mem_cgroup *from; |
| 193 | struct mem_cgroup *to; |
Johannes Weiner | 1dfab5a | 2015-02-11 15:26:09 -0800 | [diff] [blame] | 194 | unsigned long flags; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 195 | unsigned long precharge; |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 196 | unsigned long moved_charge; |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 197 | unsigned long moved_swap; |
Daisuke Nishimura | 8033b97 | 2010-03-10 15:22:16 -0800 | [diff] [blame] | 198 | struct task_struct *moving_task; /* a task moving charges */ |
| 199 | wait_queue_head_t waitq; /* a waitq for other context */ |
| 200 | } mc = { |
KAMEZAWA Hiroyuki | 2bd9bb2 | 2010-08-10 18:02:58 -0700 | [diff] [blame] | 201 | .lock = __SPIN_LOCK_UNLOCKED(mc.lock), |
Daisuke Nishimura | 8033b97 | 2010-03-10 15:22:16 -0800 | [diff] [blame] | 202 | .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq), |
| 203 | }; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 204 | |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 205 | /* |
| 206 | * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft |
| 207 | * limit reclaim to prevent infinite loops, if they ever occur. |
| 208 | */ |
Kirill A. Shutemov | a0db00f | 2012-05-29 15:06:56 -0700 | [diff] [blame] | 209 | #define MEM_CGROUP_MAX_RECLAIM_LOOPS 100 |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 210 | #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2 |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 211 | |
KAMEZAWA Hiroyuki | 217bc31 | 2008-02-07 00:14:17 -0800 | [diff] [blame] | 212 | enum charge_type { |
| 213 | MEM_CGROUP_CHARGE_TYPE_CACHE = 0, |
Kamezawa Hiroyuki | 41326c1 | 2012-07-31 16:41:40 -0700 | [diff] [blame] | 214 | MEM_CGROUP_CHARGE_TYPE_ANON, |
KAMEZAWA Hiroyuki | d13d144 | 2009-01-07 18:07:56 -0800 | [diff] [blame] | 215 | MEM_CGROUP_CHARGE_TYPE_SWAPOUT, /* for accounting swapcache */ |
KAMEZAWA Hiroyuki | 8a9478c | 2009-06-17 16:27:17 -0700 | [diff] [blame] | 216 | MEM_CGROUP_CHARGE_TYPE_DROP, /* a page was unused swap cache */ |
KAMEZAWA Hiroyuki | c05555b | 2008-10-18 20:28:11 -0700 | [diff] [blame] | 217 | NR_CHARGE_TYPE, |
| 218 | }; |
| 219 | |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 220 | /* for encoding cft->private value on file */ |
Glauber Costa | 86ae53e | 2012-12-18 14:21:45 -0800 | [diff] [blame] | 221 | enum res_type { |
| 222 | _MEM, |
| 223 | _MEMSWAP, |
| 224 | _OOM_TYPE, |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 225 | _KMEM, |
Vladimir Davydov | d55f90b | 2016-01-20 15:02:44 -0800 | [diff] [blame] | 226 | _TCP, |
Glauber Costa | 86ae53e | 2012-12-18 14:21:45 -0800 | [diff] [blame] | 227 | }; |
| 228 | |
Kirill A. Shutemov | a0db00f | 2012-05-29 15:06:56 -0700 | [diff] [blame] | 229 | #define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val)) |
| 230 | #define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff) |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 231 | #define MEMFILE_ATTR(val) ((val) & 0xffff) |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 232 | /* Used for OOM nofiier */ |
| 233 | #define OOM_CONTROL (0) |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 234 | |
Anton Vorontsov | 70ddf63 | 2013-04-29 15:08:31 -0700 | [diff] [blame] | 235 | /* Some nice accessors for the vmpressure. */ |
| 236 | struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg) |
| 237 | { |
| 238 | if (!memcg) |
| 239 | memcg = root_mem_cgroup; |
| 240 | return &memcg->vmpressure; |
| 241 | } |
| 242 | |
| 243 | struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr) |
| 244 | { |
| 245 | return &container_of(vmpr, struct mem_cgroup, vmpressure)->css; |
| 246 | } |
| 247 | |
Michal Hocko | 7ffc0ed | 2012-10-08 16:33:13 -0700 | [diff] [blame] | 248 | static inline bool mem_cgroup_is_root(struct mem_cgroup *memcg) |
| 249 | { |
| 250 | return (memcg == root_mem_cgroup); |
| 251 | } |
| 252 | |
Johannes Weiner | 127424c | 2016-01-20 15:02:32 -0800 | [diff] [blame] | 253 | #ifndef CONFIG_SLOB |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 254 | /* |
Vladimir Davydov | f7ce319 | 2015-02-12 14:59:20 -0800 | [diff] [blame] | 255 | * This will be the memcg's index in each cache's ->memcg_params.memcg_caches. |
Li Zefan | b862783 | 2013-09-23 16:56:47 +0800 | [diff] [blame] | 256 | * The main reason for not using cgroup id for this: |
| 257 | * this works better in sparse environments, where we have a lot of memcgs, |
| 258 | * but only a few kmem-limited. Or also, if we have, for instance, 200 |
| 259 | * memcgs, and none but the 200th is kmem-limited, we'd have to have a |
| 260 | * 200 entry array for that. |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 261 | * |
Vladimir Davydov | dbcf73e | 2015-02-12 14:58:57 -0800 | [diff] [blame] | 262 | * The current size of the caches array is stored in memcg_nr_cache_ids. It |
| 263 | * will double each time we have to increase it. |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 264 | */ |
Vladimir Davydov | dbcf73e | 2015-02-12 14:58:57 -0800 | [diff] [blame] | 265 | static DEFINE_IDA(memcg_cache_ida); |
| 266 | int memcg_nr_cache_ids; |
Glauber Costa | 749c541 | 2012-12-18 14:23:01 -0800 | [diff] [blame] | 267 | |
Vladimir Davydov | 05257a1 | 2015-02-12 14:59:01 -0800 | [diff] [blame] | 268 | /* Protects memcg_nr_cache_ids */ |
| 269 | static DECLARE_RWSEM(memcg_cache_ids_sem); |
| 270 | |
| 271 | void memcg_get_cache_ids(void) |
| 272 | { |
| 273 | down_read(&memcg_cache_ids_sem); |
| 274 | } |
| 275 | |
| 276 | void memcg_put_cache_ids(void) |
| 277 | { |
| 278 | up_read(&memcg_cache_ids_sem); |
| 279 | } |
| 280 | |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 281 | /* |
| 282 | * MIN_SIZE is different than 1, because we would like to avoid going through |
| 283 | * the alloc/free process all the time. In a small machine, 4 kmem-limited |
| 284 | * cgroups is a reasonable guess. In the future, it could be a parameter or |
| 285 | * tunable, but that is strictly not necessary. |
| 286 | * |
Li Zefan | b862783 | 2013-09-23 16:56:47 +0800 | [diff] [blame] | 287 | * MAX_SIZE should be as large as the number of cgrp_ids. Ideally, we could get |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 288 | * this constant directly from cgroup, but it is understandable that this is |
| 289 | * better kept as an internal representation in cgroup.c. In any case, the |
Li Zefan | b862783 | 2013-09-23 16:56:47 +0800 | [diff] [blame] | 290 | * cgrp_id space is not getting any smaller, and we don't have to necessarily |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 291 | * increase ours as well if it increases. |
| 292 | */ |
| 293 | #define MEMCG_CACHES_MIN_SIZE 4 |
Li Zefan | b862783 | 2013-09-23 16:56:47 +0800 | [diff] [blame] | 294 | #define MEMCG_CACHES_MAX_SIZE MEM_CGROUP_ID_MAX |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 295 | |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 296 | /* |
| 297 | * A lot of the calls to the cache allocation functions are expected to be |
| 298 | * inlined by the compiler. Since the calls to memcg_kmem_get_cache are |
| 299 | * conditional to this static branch, we'll have to allow modules that does |
| 300 | * kmem_cache_alloc and the such to see this symbol as well |
| 301 | */ |
Johannes Weiner | ef12947 | 2016-01-14 15:21:34 -0800 | [diff] [blame] | 302 | DEFINE_STATIC_KEY_FALSE(memcg_kmem_enabled_key); |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 303 | EXPORT_SYMBOL(memcg_kmem_enabled_key); |
Glauber Costa | a8964b9 | 2012-12-18 14:22:09 -0800 | [diff] [blame] | 304 | |
Tejun Heo | 17cc4df | 2017-02-22 15:41:36 -0800 | [diff] [blame] | 305 | struct workqueue_struct *memcg_kmem_cache_wq; |
| 306 | |
Johannes Weiner | 127424c | 2016-01-20 15:02:32 -0800 | [diff] [blame] | 307 | #endif /* !CONFIG_SLOB */ |
Glauber Costa | a8964b9 | 2012-12-18 14:22:09 -0800 | [diff] [blame] | 308 | |
Tejun Heo | ad7fa85 | 2015-05-27 20:00:02 -0400 | [diff] [blame] | 309 | /** |
| 310 | * mem_cgroup_css_from_page - css of the memcg associated with a page |
| 311 | * @page: page of interest |
| 312 | * |
| 313 | * If memcg is bound to the default hierarchy, css of the memcg associated |
| 314 | * with @page is returned. The returned css remains associated with @page |
| 315 | * until it is released. |
| 316 | * |
| 317 | * If memcg is bound to a traditional hierarchy, the css of root_mem_cgroup |
| 318 | * is returned. |
Tejun Heo | ad7fa85 | 2015-05-27 20:00:02 -0400 | [diff] [blame] | 319 | */ |
| 320 | struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page) |
| 321 | { |
| 322 | struct mem_cgroup *memcg; |
| 323 | |
Tejun Heo | ad7fa85 | 2015-05-27 20:00:02 -0400 | [diff] [blame] | 324 | memcg = page->mem_cgroup; |
| 325 | |
Tejun Heo | 9e10a13 | 2015-09-18 11:56:28 -0400 | [diff] [blame] | 326 | if (!memcg || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
Tejun Heo | ad7fa85 | 2015-05-27 20:00:02 -0400 | [diff] [blame] | 327 | memcg = root_mem_cgroup; |
| 328 | |
Tejun Heo | ad7fa85 | 2015-05-27 20:00:02 -0400 | [diff] [blame] | 329 | return &memcg->css; |
| 330 | } |
| 331 | |
Vladimir Davydov | 2fc0452 | 2015-09-09 15:35:28 -0700 | [diff] [blame] | 332 | /** |
| 333 | * page_cgroup_ino - return inode number of the memcg a page is charged to |
| 334 | * @page: the page |
| 335 | * |
| 336 | * Look up the closest online ancestor of the memory cgroup @page is charged to |
| 337 | * and return its inode number or 0 if @page is not charged to any cgroup. It |
| 338 | * is safe to call this function without holding a reference to @page. |
| 339 | * |
| 340 | * Note, this function is inherently racy, because there is nothing to prevent |
| 341 | * the cgroup inode from getting torn down and potentially reallocated a moment |
| 342 | * after page_cgroup_ino() returns, so it only should be used by callers that |
| 343 | * do not care (such as procfs interfaces). |
| 344 | */ |
| 345 | ino_t page_cgroup_ino(struct page *page) |
| 346 | { |
| 347 | struct mem_cgroup *memcg; |
| 348 | unsigned long ino = 0; |
| 349 | |
| 350 | rcu_read_lock(); |
| 351 | memcg = READ_ONCE(page->mem_cgroup); |
| 352 | while (memcg && !(memcg->css.flags & CSS_ONLINE)) |
| 353 | memcg = parent_mem_cgroup(memcg); |
| 354 | if (memcg) |
| 355 | ino = cgroup_ino(memcg->css.cgroup); |
| 356 | rcu_read_unlock(); |
| 357 | return ino; |
| 358 | } |
| 359 | |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 360 | static struct mem_cgroup_per_node * |
| 361 | mem_cgroup_page_nodeinfo(struct mem_cgroup *memcg, struct page *page) |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 362 | { |
Johannes Weiner | 97a6c37 | 2011-03-23 16:42:27 -0700 | [diff] [blame] | 363 | int nid = page_to_nid(page); |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 364 | |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 365 | return memcg->nodeinfo[nid]; |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 366 | } |
| 367 | |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 368 | static struct mem_cgroup_tree_per_node * |
| 369 | soft_limit_tree_node(int nid) |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 370 | { |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 371 | return soft_limit_tree.rb_tree_per_node[nid]; |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 372 | } |
| 373 | |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 374 | static struct mem_cgroup_tree_per_node * |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 375 | soft_limit_tree_from_page(struct page *page) |
| 376 | { |
| 377 | int nid = page_to_nid(page); |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 378 | |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 379 | return soft_limit_tree.rb_tree_per_node[nid]; |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 380 | } |
| 381 | |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 382 | static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_node *mz, |
| 383 | struct mem_cgroup_tree_per_node *mctz, |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 384 | unsigned long new_usage_in_excess) |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 385 | { |
| 386 | struct rb_node **p = &mctz->rb_root.rb_node; |
| 387 | struct rb_node *parent = NULL; |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 388 | struct mem_cgroup_per_node *mz_node; |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 389 | |
| 390 | if (mz->on_tree) |
| 391 | return; |
| 392 | |
| 393 | mz->usage_in_excess = new_usage_in_excess; |
| 394 | if (!mz->usage_in_excess) |
| 395 | return; |
| 396 | while (*p) { |
| 397 | parent = *p; |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 398 | mz_node = rb_entry(parent, struct mem_cgroup_per_node, |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 399 | tree_node); |
| 400 | if (mz->usage_in_excess < mz_node->usage_in_excess) |
| 401 | p = &(*p)->rb_left; |
| 402 | /* |
| 403 | * We can't avoid mem cgroups that are over their soft |
| 404 | * limit by the same amount |
| 405 | */ |
| 406 | else if (mz->usage_in_excess >= mz_node->usage_in_excess) |
| 407 | p = &(*p)->rb_right; |
| 408 | } |
| 409 | rb_link_node(&mz->tree_node, parent, p); |
| 410 | rb_insert_color(&mz->tree_node, &mctz->rb_root); |
| 411 | mz->on_tree = true; |
| 412 | } |
| 413 | |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 414 | static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz, |
| 415 | struct mem_cgroup_tree_per_node *mctz) |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 416 | { |
| 417 | if (!mz->on_tree) |
| 418 | return; |
| 419 | rb_erase(&mz->tree_node, &mctz->rb_root); |
| 420 | mz->on_tree = false; |
| 421 | } |
| 422 | |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 423 | static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz, |
| 424 | struct mem_cgroup_tree_per_node *mctz) |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 425 | { |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 426 | unsigned long flags; |
| 427 | |
| 428 | spin_lock_irqsave(&mctz->lock, flags); |
Johannes Weiner | cf2c812 | 2014-06-06 14:38:21 -0700 | [diff] [blame] | 429 | __mem_cgroup_remove_exceeded(mz, mctz); |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 430 | spin_unlock_irqrestore(&mctz->lock, flags); |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 431 | } |
| 432 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 433 | static unsigned long soft_limit_excess(struct mem_cgroup *memcg) |
| 434 | { |
| 435 | unsigned long nr_pages = page_counter_read(&memcg->memory); |
Jason Low | 4db0c3c | 2015-04-15 16:14:08 -0700 | [diff] [blame] | 436 | unsigned long soft_limit = READ_ONCE(memcg->soft_limit); |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 437 | unsigned long excess = 0; |
| 438 | |
| 439 | if (nr_pages > soft_limit) |
| 440 | excess = nr_pages - soft_limit; |
| 441 | |
| 442 | return excess; |
| 443 | } |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 444 | |
| 445 | static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page) |
| 446 | { |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 447 | unsigned long excess; |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 448 | struct mem_cgroup_per_node *mz; |
| 449 | struct mem_cgroup_tree_per_node *mctz; |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 450 | |
Jianyu Zhan | e231875 | 2014-06-06 14:38:20 -0700 | [diff] [blame] | 451 | mctz = soft_limit_tree_from_page(page); |
Laurent Dufour | bfc7228 | 2017-03-09 16:17:06 -0800 | [diff] [blame] | 452 | if (!mctz) |
| 453 | return; |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 454 | /* |
| 455 | * Necessary to update all ancestors when hierarchy is used. |
| 456 | * because their event counter is not touched. |
| 457 | */ |
| 458 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 459 | mz = mem_cgroup_page_nodeinfo(memcg, page); |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 460 | excess = soft_limit_excess(memcg); |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 461 | /* |
| 462 | * We have to update the tree if mz is on RB-tree or |
| 463 | * mem is over its softlimit. |
| 464 | */ |
| 465 | if (excess || mz->on_tree) { |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 466 | unsigned long flags; |
| 467 | |
| 468 | spin_lock_irqsave(&mctz->lock, flags); |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 469 | /* if on-tree, remove it */ |
| 470 | if (mz->on_tree) |
Johannes Weiner | cf2c812 | 2014-06-06 14:38:21 -0700 | [diff] [blame] | 471 | __mem_cgroup_remove_exceeded(mz, mctz); |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 472 | /* |
| 473 | * Insert again. mz->usage_in_excess will be updated. |
| 474 | * If excess is 0, no tree ops. |
| 475 | */ |
Johannes Weiner | cf2c812 | 2014-06-06 14:38:21 -0700 | [diff] [blame] | 476 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 477 | spin_unlock_irqrestore(&mctz->lock, flags); |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 478 | } |
| 479 | } |
| 480 | } |
| 481 | |
| 482 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg) |
| 483 | { |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 484 | struct mem_cgroup_tree_per_node *mctz; |
| 485 | struct mem_cgroup_per_node *mz; |
| 486 | int nid; |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 487 | |
Jianyu Zhan | e231875 | 2014-06-06 14:38:20 -0700 | [diff] [blame] | 488 | for_each_node(nid) { |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 489 | mz = mem_cgroup_nodeinfo(memcg, nid); |
| 490 | mctz = soft_limit_tree_node(nid); |
Laurent Dufour | bfc7228 | 2017-03-09 16:17:06 -0800 | [diff] [blame] | 491 | if (mctz) |
| 492 | mem_cgroup_remove_exceeded(mz, mctz); |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 493 | } |
| 494 | } |
| 495 | |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 496 | static struct mem_cgroup_per_node * |
| 497 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz) |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 498 | { |
| 499 | struct rb_node *rightmost = NULL; |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 500 | struct mem_cgroup_per_node *mz; |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 501 | |
| 502 | retry: |
| 503 | mz = NULL; |
| 504 | rightmost = rb_last(&mctz->rb_root); |
| 505 | if (!rightmost) |
| 506 | goto done; /* Nothing to reclaim from */ |
| 507 | |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 508 | mz = rb_entry(rightmost, struct mem_cgroup_per_node, tree_node); |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 509 | /* |
| 510 | * Remove the node now but someone else can add it back, |
| 511 | * we will to add it back at the end of reclaim to its correct |
| 512 | * position in the tree. |
| 513 | */ |
Johannes Weiner | cf2c812 | 2014-06-06 14:38:21 -0700 | [diff] [blame] | 514 | __mem_cgroup_remove_exceeded(mz, mctz); |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 515 | if (!soft_limit_excess(mz->memcg) || |
Tejun Heo | ec903c0 | 2014-05-13 12:11:01 -0400 | [diff] [blame] | 516 | !css_tryget_online(&mz->memcg->css)) |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 517 | goto retry; |
| 518 | done: |
| 519 | return mz; |
| 520 | } |
| 521 | |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 522 | static struct mem_cgroup_per_node * |
| 523 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz) |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 524 | { |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 525 | struct mem_cgroup_per_node *mz; |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 526 | |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 527 | spin_lock_irq(&mctz->lock); |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 528 | mz = __mem_cgroup_largest_soft_limit_node(mctz); |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 529 | spin_unlock_irq(&mctz->lock); |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 530 | return mz; |
| 531 | } |
| 532 | |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 533 | /* |
Greg Thelen | 484ebb3 | 2015-10-01 15:37:05 -0700 | [diff] [blame] | 534 | * Return page count for single (non recursive) @memcg. |
| 535 | * |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 536 | * Implementation Note: reading percpu statistics for memcg. |
| 537 | * |
| 538 | * Both of vmstat[] and percpu_counter has threshold and do periodic |
| 539 | * synchronization to implement "quick" read. There are trade-off between |
| 540 | * reading cost and precision of value. Then, we may have a chance to implement |
Greg Thelen | 484ebb3 | 2015-10-01 15:37:05 -0700 | [diff] [blame] | 541 | * a periodic synchronization of counter in memcg's counter. |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 542 | * |
| 543 | * But this _read() function is used for user interface now. The user accounts |
| 544 | * memory usage by memory cgroup and he _always_ requires exact value because |
| 545 | * he accounts memory. Even if we provide quick-and-fuzzy read, we always |
| 546 | * have to visit all online cpus and make sum. So, for now, unnecessary |
| 547 | * synchronization is not implemented. (just implemented for cpu hotplug) |
| 548 | * |
| 549 | * If there are kernel internal actions which can make use of some not-exact |
| 550 | * value, and reading all cpu value can be performance bottleneck in some |
Greg Thelen | 484ebb3 | 2015-10-01 15:37:05 -0700 | [diff] [blame] | 551 | * common workload, threshold and synchronization as vmstat[] should be |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 552 | * implemented. |
| 553 | */ |
KAMEZAWA Hiroyuki | c62b1a3 | 2010-03-10 15:22:29 -0800 | [diff] [blame] | 554 | |
Johannes Weiner | ccda7f4 | 2017-05-03 14:55:16 -0700 | [diff] [blame] | 555 | static unsigned long memcg_sum_events(struct mem_cgroup *memcg, |
| 556 | enum memcg_event_item event) |
Johannes Weiner | e9f8974 | 2011-03-23 16:42:37 -0700 | [diff] [blame] | 557 | { |
| 558 | unsigned long val = 0; |
| 559 | int cpu; |
| 560 | |
Tejun Heo | 733a572 | 2015-05-22 18:23:18 -0400 | [diff] [blame] | 561 | for_each_possible_cpu(cpu) |
Johannes Weiner | df0e53d | 2017-05-03 14:55:10 -0700 | [diff] [blame] | 562 | val += per_cpu(memcg->stat->events[event], cpu); |
Johannes Weiner | e9f8974 | 2011-03-23 16:42:37 -0700 | [diff] [blame] | 563 | return val; |
| 564 | } |
| 565 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 566 | static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg, |
David Rientjes | b070e65 | 2013-05-07 16:18:09 -0700 | [diff] [blame] | 567 | struct page *page, |
Kirill A. Shutemov | f627c2f | 2016-01-15 16:52:20 -0800 | [diff] [blame] | 568 | bool compound, int nr_pages) |
KAMEZAWA Hiroyuki | d52aa41 | 2008-02-07 00:14:24 -0800 | [diff] [blame] | 569 | { |
KAMEZAWA Hiroyuki | b2402857 | 2012-03-21 16:34:22 -0700 | [diff] [blame] | 570 | /* |
| 571 | * Here, RSS means 'mapped anon' and anon's SwapCache. Shmem/tmpfs is |
| 572 | * counted as CACHE even if it's on ANON LRU. |
| 573 | */ |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 574 | if (PageAnon(page)) |
Johannes Weiner | 71cd311 | 2017-05-03 14:55:13 -0700 | [diff] [blame] | 575 | __this_cpu_add(memcg->stat->count[MEMCG_RSS], nr_pages); |
Johannes Weiner | 9a4caf1 | 2017-05-03 14:52:45 -0700 | [diff] [blame] | 576 | else { |
Johannes Weiner | 71cd311 | 2017-05-03 14:55:13 -0700 | [diff] [blame] | 577 | __this_cpu_add(memcg->stat->count[MEMCG_CACHE], nr_pages); |
Johannes Weiner | 9a4caf1 | 2017-05-03 14:52:45 -0700 | [diff] [blame] | 578 | if (PageSwapBacked(page)) |
Johannes Weiner | 71cd311 | 2017-05-03 14:55:13 -0700 | [diff] [blame] | 579 | __this_cpu_add(memcg->stat->count[NR_SHMEM], nr_pages); |
Johannes Weiner | 9a4caf1 | 2017-05-03 14:52:45 -0700 | [diff] [blame] | 580 | } |
Balaji Rao | 55e462b | 2008-05-01 04:35:12 -0700 | [diff] [blame] | 581 | |
Kirill A. Shutemov | f627c2f | 2016-01-15 16:52:20 -0800 | [diff] [blame] | 582 | if (compound) { |
| 583 | VM_BUG_ON_PAGE(!PageTransHuge(page), page); |
Johannes Weiner | 71cd311 | 2017-05-03 14:55:13 -0700 | [diff] [blame] | 584 | __this_cpu_add(memcg->stat->count[MEMCG_RSS_HUGE], nr_pages); |
Kirill A. Shutemov | f627c2f | 2016-01-15 16:52:20 -0800 | [diff] [blame] | 585 | } |
David Rientjes | b070e65 | 2013-05-07 16:18:09 -0700 | [diff] [blame] | 586 | |
KAMEZAWA Hiroyuki | e401f176 | 2011-01-20 14:44:23 -0800 | [diff] [blame] | 587 | /* pagein of a big page is an event. So, ignore page size */ |
| 588 | if (nr_pages > 0) |
Johannes Weiner | df0e53d | 2017-05-03 14:55:10 -0700 | [diff] [blame] | 589 | __this_cpu_inc(memcg->stat->events[PGPGIN]); |
KAMEZAWA Hiroyuki | 3751d60 | 2011-02-01 15:52:45 -0800 | [diff] [blame] | 590 | else { |
Johannes Weiner | df0e53d | 2017-05-03 14:55:10 -0700 | [diff] [blame] | 591 | __this_cpu_inc(memcg->stat->events[PGPGOUT]); |
KAMEZAWA Hiroyuki | 3751d60 | 2011-02-01 15:52:45 -0800 | [diff] [blame] | 592 | nr_pages = -nr_pages; /* for event */ |
| 593 | } |
KAMEZAWA Hiroyuki | e401f176 | 2011-01-20 14:44:23 -0800 | [diff] [blame] | 594 | |
Johannes Weiner | 1311471 | 2012-05-29 15:07:07 -0700 | [diff] [blame] | 595 | __this_cpu_add(memcg->stat->nr_page_events, nr_pages); |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 596 | } |
KAMEZAWA Hiroyuki | d52aa41 | 2008-02-07 00:14:24 -0800 | [diff] [blame] | 597 | |
Vladimir Davydov | 0a6b76d | 2016-03-17 14:18:42 -0700 | [diff] [blame] | 598 | unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, |
| 599 | int nid, unsigned int lru_mask) |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 600 | { |
Michal Hocko | b4536f0c8 | 2017-01-10 16:58:04 -0800 | [diff] [blame] | 601 | struct lruvec *lruvec = mem_cgroup_lruvec(NODE_DATA(nid), memcg); |
Jianyu Zhan | e231875 | 2014-06-06 14:38:20 -0700 | [diff] [blame] | 602 | unsigned long nr = 0; |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 603 | enum lru_list lru; |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 604 | |
Jianyu Zhan | e231875 | 2014-06-06 14:38:20 -0700 | [diff] [blame] | 605 | VM_BUG_ON((unsigned)nid >= nr_node_ids); |
KAMEZAWA Hiroyuki | bb2a0de | 2011-07-26 16:08:22 -0700 | [diff] [blame] | 606 | |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 607 | for_each_lru(lru) { |
| 608 | if (!(BIT(lru) & lru_mask)) |
| 609 | continue; |
Michal Hocko | b4536f0c8 | 2017-01-10 16:58:04 -0800 | [diff] [blame] | 610 | nr += mem_cgroup_get_lru_size(lruvec, lru); |
Jianyu Zhan | e231875 | 2014-06-06 14:38:20 -0700 | [diff] [blame] | 611 | } |
| 612 | return nr; |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 613 | } |
KAMEZAWA Hiroyuki | bb2a0de | 2011-07-26 16:08:22 -0700 | [diff] [blame] | 614 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 615 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, |
KAMEZAWA Hiroyuki | bb2a0de | 2011-07-26 16:08:22 -0700 | [diff] [blame] | 616 | unsigned int lru_mask) |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 617 | { |
Jianyu Zhan | e231875 | 2014-06-06 14:38:20 -0700 | [diff] [blame] | 618 | unsigned long nr = 0; |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 619 | int nid; |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 620 | |
Lai Jiangshan | 31aaea4 | 2012-12-12 13:51:27 -0800 | [diff] [blame] | 621 | for_each_node_state(nid, N_MEMORY) |
Jianyu Zhan | e231875 | 2014-06-06 14:38:20 -0700 | [diff] [blame] | 622 | nr += mem_cgroup_node_nr_lru_pages(memcg, nid, lru_mask); |
| 623 | return nr; |
KAMEZAWA Hiroyuki | d52aa41 | 2008-02-07 00:14:24 -0800 | [diff] [blame] | 624 | } |
| 625 | |
Johannes Weiner | f53d7ce | 2012-01-12 17:18:23 -0800 | [diff] [blame] | 626 | static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg, |
| 627 | enum mem_cgroup_events_target target) |
KAMEZAWA Hiroyuki | d2265e6 | 2010-03-10 15:22:31 -0800 | [diff] [blame] | 628 | { |
Johannes Weiner | 7a159cc | 2011-03-23 16:42:38 -0700 | [diff] [blame] | 629 | unsigned long val, next; |
| 630 | |
Johannes Weiner | 1311471 | 2012-05-29 15:07:07 -0700 | [diff] [blame] | 631 | val = __this_cpu_read(memcg->stat->nr_page_events); |
Steven Rostedt | 4799401 | 2011-11-02 13:38:33 -0700 | [diff] [blame] | 632 | next = __this_cpu_read(memcg->stat->targets[target]); |
Johannes Weiner | 7a159cc | 2011-03-23 16:42:38 -0700 | [diff] [blame] | 633 | /* from time_after() in jiffies.h */ |
Johannes Weiner | f53d7ce | 2012-01-12 17:18:23 -0800 | [diff] [blame] | 634 | if ((long)next - (long)val < 0) { |
| 635 | switch (target) { |
| 636 | case MEM_CGROUP_TARGET_THRESH: |
| 637 | next = val + THRESHOLDS_EVENTS_TARGET; |
| 638 | break; |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 639 | case MEM_CGROUP_TARGET_SOFTLIMIT: |
| 640 | next = val + SOFTLIMIT_EVENTS_TARGET; |
| 641 | break; |
Johannes Weiner | f53d7ce | 2012-01-12 17:18:23 -0800 | [diff] [blame] | 642 | case MEM_CGROUP_TARGET_NUMAINFO: |
| 643 | next = val + NUMAINFO_EVENTS_TARGET; |
| 644 | break; |
| 645 | default: |
| 646 | break; |
| 647 | } |
| 648 | __this_cpu_write(memcg->stat->targets[target], next); |
| 649 | return true; |
Johannes Weiner | 7a159cc | 2011-03-23 16:42:38 -0700 | [diff] [blame] | 650 | } |
Johannes Weiner | f53d7ce | 2012-01-12 17:18:23 -0800 | [diff] [blame] | 651 | return false; |
KAMEZAWA Hiroyuki | d2265e6 | 2010-03-10 15:22:31 -0800 | [diff] [blame] | 652 | } |
| 653 | |
| 654 | /* |
| 655 | * Check events in order. |
| 656 | * |
| 657 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 658 | static void memcg_check_events(struct mem_cgroup *memcg, struct page *page) |
KAMEZAWA Hiroyuki | d2265e6 | 2010-03-10 15:22:31 -0800 | [diff] [blame] | 659 | { |
| 660 | /* threshold event is triggered in finer grain than soft limit */ |
Johannes Weiner | f53d7ce | 2012-01-12 17:18:23 -0800 | [diff] [blame] | 661 | if (unlikely(mem_cgroup_event_ratelimit(memcg, |
| 662 | MEM_CGROUP_TARGET_THRESH))) { |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 663 | bool do_softlimit; |
Andrew Morton | 82b3f2a | 2012-02-03 15:37:14 -0800 | [diff] [blame] | 664 | bool do_numainfo __maybe_unused; |
Johannes Weiner | f53d7ce | 2012-01-12 17:18:23 -0800 | [diff] [blame] | 665 | |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 666 | do_softlimit = mem_cgroup_event_ratelimit(memcg, |
| 667 | MEM_CGROUP_TARGET_SOFTLIMIT); |
KAMEZAWA Hiroyuki | 453a9bf | 2011-07-08 15:39:43 -0700 | [diff] [blame] | 668 | #if MAX_NUMNODES > 1 |
Johannes Weiner | f53d7ce | 2012-01-12 17:18:23 -0800 | [diff] [blame] | 669 | do_numainfo = mem_cgroup_event_ratelimit(memcg, |
| 670 | MEM_CGROUP_TARGET_NUMAINFO); |
KAMEZAWA Hiroyuki | 453a9bf | 2011-07-08 15:39:43 -0700 | [diff] [blame] | 671 | #endif |
Johannes Weiner | f53d7ce | 2012-01-12 17:18:23 -0800 | [diff] [blame] | 672 | mem_cgroup_threshold(memcg); |
Andrew Morton | bb4cc1a | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 673 | if (unlikely(do_softlimit)) |
| 674 | mem_cgroup_update_tree(memcg, page); |
Johannes Weiner | f53d7ce | 2012-01-12 17:18:23 -0800 | [diff] [blame] | 675 | #if MAX_NUMNODES > 1 |
| 676 | if (unlikely(do_numainfo)) |
| 677 | atomic_inc(&memcg->numainfo_events); |
| 678 | #endif |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 679 | } |
KAMEZAWA Hiroyuki | d2265e6 | 2010-03-10 15:22:31 -0800 | [diff] [blame] | 680 | } |
| 681 | |
Balbir Singh | cf475ad | 2008-04-29 01:00:16 -0700 | [diff] [blame] | 682 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
Pavel Emelianov | 78fb746 | 2008-02-07 00:13:51 -0800 | [diff] [blame] | 683 | { |
Balbir Singh | 31a78f2 | 2008-09-28 23:09:31 +0100 | [diff] [blame] | 684 | /* |
| 685 | * mm_update_next_owner() may clear mm->owner to NULL |
| 686 | * if it races with swapoff, page migration, etc. |
| 687 | * So this can be called with p == NULL. |
| 688 | */ |
| 689 | if (unlikely(!p)) |
| 690 | return NULL; |
| 691 | |
Tejun Heo | 073219e | 2014-02-08 10:36:58 -0500 | [diff] [blame] | 692 | return mem_cgroup_from_css(task_css(p, memory_cgrp_id)); |
Pavel Emelianov | 78fb746 | 2008-02-07 00:13:51 -0800 | [diff] [blame] | 693 | } |
Michal Hocko | 33398cf | 2015-09-08 15:01:02 -0700 | [diff] [blame] | 694 | EXPORT_SYMBOL(mem_cgroup_from_task); |
Pavel Emelianov | 78fb746 | 2008-02-07 00:13:51 -0800 | [diff] [blame] | 695 | |
Johannes Weiner | df38197 | 2014-04-07 15:37:43 -0700 | [diff] [blame] | 696 | static struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm) |
KAMEZAWA Hiroyuki | 54595fe | 2009-01-07 18:08:33 -0800 | [diff] [blame] | 697 | { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 698 | struct mem_cgroup *memcg = NULL; |
KAMEZAWA Hiroyuki | 0b7f569 | 2009-04-02 16:57:38 -0700 | [diff] [blame] | 699 | |
KAMEZAWA Hiroyuki | 54595fe | 2009-01-07 18:08:33 -0800 | [diff] [blame] | 700 | rcu_read_lock(); |
| 701 | do { |
Michal Hocko | 6f6acb0 | 2014-05-22 11:54:19 -0700 | [diff] [blame] | 702 | /* |
| 703 | * Page cache insertions can happen withou an |
| 704 | * actual mm context, e.g. during disk probing |
| 705 | * on boot, loopback IO, acct() writes etc. |
| 706 | */ |
| 707 | if (unlikely(!mm)) |
Johannes Weiner | df38197 | 2014-04-07 15:37:43 -0700 | [diff] [blame] | 708 | memcg = root_mem_cgroup; |
Michal Hocko | 6f6acb0 | 2014-05-22 11:54:19 -0700 | [diff] [blame] | 709 | else { |
| 710 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); |
| 711 | if (unlikely(!memcg)) |
| 712 | memcg = root_mem_cgroup; |
| 713 | } |
Tejun Heo | ec903c0 | 2014-05-13 12:11:01 -0400 | [diff] [blame] | 714 | } while (!css_tryget_online(&memcg->css)); |
KAMEZAWA Hiroyuki | 54595fe | 2009-01-07 18:08:33 -0800 | [diff] [blame] | 715 | rcu_read_unlock(); |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 716 | return memcg; |
KAMEZAWA Hiroyuki | 54595fe | 2009-01-07 18:08:33 -0800 | [diff] [blame] | 717 | } |
| 718 | |
Johannes Weiner | 5660048 | 2012-01-12 17:17:59 -0800 | [diff] [blame] | 719 | /** |
| 720 | * mem_cgroup_iter - iterate over memory cgroup hierarchy |
| 721 | * @root: hierarchy root |
| 722 | * @prev: previously returned memcg, NULL on first invocation |
| 723 | * @reclaim: cookie for shared reclaim walks, NULL for full walks |
| 724 | * |
| 725 | * Returns references to children of the hierarchy below @root, or |
| 726 | * @root itself, or %NULL after a full round-trip. |
| 727 | * |
| 728 | * Caller must pass the return value in @prev on subsequent |
| 729 | * invocations for reference counting, or use mem_cgroup_iter_break() |
| 730 | * to cancel a hierarchy walk before the round-trip is complete. |
| 731 | * |
| 732 | * Reclaimers can specify a zone and a priority level in @reclaim to |
| 733 | * divide up the memcgs in the hierarchy among all concurrent |
| 734 | * reclaimers operating on the same zone and priority. |
| 735 | */ |
Andrew Morton | 694fbc0 | 2013-09-24 15:27:37 -0700 | [diff] [blame] | 736 | struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root, |
Johannes Weiner | 5660048 | 2012-01-12 17:17:59 -0800 | [diff] [blame] | 737 | struct mem_cgroup *prev, |
Andrew Morton | 694fbc0 | 2013-09-24 15:27:37 -0700 | [diff] [blame] | 738 | struct mem_cgroup_reclaim_cookie *reclaim) |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 739 | { |
Michal Hocko | 33398cf | 2015-09-08 15:01:02 -0700 | [diff] [blame] | 740 | struct mem_cgroup_reclaim_iter *uninitialized_var(iter); |
Johannes Weiner | 5ac8fb3 | 2014-12-10 15:42:39 -0800 | [diff] [blame] | 741 | struct cgroup_subsys_state *css = NULL; |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 742 | struct mem_cgroup *memcg = NULL; |
Johannes Weiner | 5ac8fb3 | 2014-12-10 15:42:39 -0800 | [diff] [blame] | 743 | struct mem_cgroup *pos = NULL; |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 744 | |
Andrew Morton | 694fbc0 | 2013-09-24 15:27:37 -0700 | [diff] [blame] | 745 | if (mem_cgroup_disabled()) |
| 746 | return NULL; |
Johannes Weiner | 5660048 | 2012-01-12 17:17:59 -0800 | [diff] [blame] | 747 | |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 748 | if (!root) |
| 749 | root = root_mem_cgroup; |
| 750 | |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 751 | if (prev && !reclaim) |
Johannes Weiner | 5ac8fb3 | 2014-12-10 15:42:39 -0800 | [diff] [blame] | 752 | pos = prev; |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 753 | |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 754 | if (!root->use_hierarchy && root != root_mem_cgroup) { |
| 755 | if (prev) |
Johannes Weiner | 5ac8fb3 | 2014-12-10 15:42:39 -0800 | [diff] [blame] | 756 | goto out; |
Andrew Morton | 694fbc0 | 2013-09-24 15:27:37 -0700 | [diff] [blame] | 757 | return root; |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 758 | } |
| 759 | |
Michal Hocko | 542f85f | 2013-04-29 15:07:15 -0700 | [diff] [blame] | 760 | rcu_read_lock(); |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 761 | |
Johannes Weiner | 5ac8fb3 | 2014-12-10 15:42:39 -0800 | [diff] [blame] | 762 | if (reclaim) { |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 763 | struct mem_cgroup_per_node *mz; |
Johannes Weiner | 527a5ec | 2012-01-12 17:17:55 -0800 | [diff] [blame] | 764 | |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 765 | mz = mem_cgroup_nodeinfo(root, reclaim->pgdat->node_id); |
Johannes Weiner | 5ac8fb3 | 2014-12-10 15:42:39 -0800 | [diff] [blame] | 766 | iter = &mz->iter[reclaim->priority]; |
Michal Hocko | 5f57816 | 2013-04-29 15:07:17 -0700 | [diff] [blame] | 767 | |
Johannes Weiner | 5ac8fb3 | 2014-12-10 15:42:39 -0800 | [diff] [blame] | 768 | if (prev && reclaim->generation != iter->generation) |
Michal Hocko | 542f85f | 2013-04-29 15:07:15 -0700 | [diff] [blame] | 769 | goto out_unlock; |
Johannes Weiner | 5ac8fb3 | 2014-12-10 15:42:39 -0800 | [diff] [blame] | 770 | |
Vladimir Davydov | 6df3868 | 2015-12-29 14:54:10 -0800 | [diff] [blame] | 771 | while (1) { |
Jason Low | 4db0c3c | 2015-04-15 16:14:08 -0700 | [diff] [blame] | 772 | pos = READ_ONCE(iter->position); |
Vladimir Davydov | 6df3868 | 2015-12-29 14:54:10 -0800 | [diff] [blame] | 773 | if (!pos || css_tryget(&pos->css)) |
| 774 | break; |
Johannes Weiner | 5ac8fb3 | 2014-12-10 15:42:39 -0800 | [diff] [blame] | 775 | /* |
Vladimir Davydov | 6df3868 | 2015-12-29 14:54:10 -0800 | [diff] [blame] | 776 | * css reference reached zero, so iter->position will |
| 777 | * be cleared by ->css_released. However, we should not |
| 778 | * rely on this happening soon, because ->css_released |
| 779 | * is called from a work queue, and by busy-waiting we |
| 780 | * might block it. So we clear iter->position right |
| 781 | * away. |
Johannes Weiner | 5ac8fb3 | 2014-12-10 15:42:39 -0800 | [diff] [blame] | 782 | */ |
Vladimir Davydov | 6df3868 | 2015-12-29 14:54:10 -0800 | [diff] [blame] | 783 | (void)cmpxchg(&iter->position, pos, NULL); |
| 784 | } |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 785 | } |
Johannes Weiner | 5ac8fb3 | 2014-12-10 15:42:39 -0800 | [diff] [blame] | 786 | |
| 787 | if (pos) |
| 788 | css = &pos->css; |
| 789 | |
| 790 | for (;;) { |
| 791 | css = css_next_descendant_pre(css, &root->css); |
| 792 | if (!css) { |
| 793 | /* |
| 794 | * Reclaimers share the hierarchy walk, and a |
| 795 | * new one might jump in right at the end of |
| 796 | * the hierarchy - make sure they see at least |
| 797 | * one group and restart from the beginning. |
| 798 | */ |
| 799 | if (!prev) |
| 800 | continue; |
| 801 | break; |
| 802 | } |
| 803 | |
| 804 | /* |
| 805 | * Verify the css and acquire a reference. The root |
| 806 | * is provided by the caller, so we know it's alive |
| 807 | * and kicking, and don't take an extra reference. |
| 808 | */ |
| 809 | memcg = mem_cgroup_from_css(css); |
| 810 | |
| 811 | if (css == &root->css) |
| 812 | break; |
| 813 | |
Johannes Weiner | 0b8f73e | 2016-01-20 15:02:53 -0800 | [diff] [blame] | 814 | if (css_tryget(css)) |
| 815 | break; |
Johannes Weiner | 5ac8fb3 | 2014-12-10 15:42:39 -0800 | [diff] [blame] | 816 | |
| 817 | memcg = NULL; |
| 818 | } |
| 819 | |
| 820 | if (reclaim) { |
Johannes Weiner | 5ac8fb3 | 2014-12-10 15:42:39 -0800 | [diff] [blame] | 821 | /* |
Vladimir Davydov | 6df3868 | 2015-12-29 14:54:10 -0800 | [diff] [blame] | 822 | * The position could have already been updated by a competing |
| 823 | * thread, so check that the value hasn't changed since we read |
| 824 | * it to avoid reclaiming from the same cgroup twice. |
Johannes Weiner | 5ac8fb3 | 2014-12-10 15:42:39 -0800 | [diff] [blame] | 825 | */ |
Vladimir Davydov | 6df3868 | 2015-12-29 14:54:10 -0800 | [diff] [blame] | 826 | (void)cmpxchg(&iter->position, pos, memcg); |
| 827 | |
Johannes Weiner | 5ac8fb3 | 2014-12-10 15:42:39 -0800 | [diff] [blame] | 828 | if (pos) |
| 829 | css_put(&pos->css); |
| 830 | |
| 831 | if (!memcg) |
| 832 | iter->generation++; |
| 833 | else if (!prev) |
| 834 | reclaim->generation = iter->generation; |
| 835 | } |
| 836 | |
Michal Hocko | 542f85f | 2013-04-29 15:07:15 -0700 | [diff] [blame] | 837 | out_unlock: |
| 838 | rcu_read_unlock(); |
Johannes Weiner | 5ac8fb3 | 2014-12-10 15:42:39 -0800 | [diff] [blame] | 839 | out: |
Michal Hocko | c40046f | 2013-04-29 15:07:14 -0700 | [diff] [blame] | 840 | if (prev && prev != root) |
| 841 | css_put(&prev->css); |
| 842 | |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 843 | return memcg; |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 844 | } |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 845 | |
Johannes Weiner | 5660048 | 2012-01-12 17:17:59 -0800 | [diff] [blame] | 846 | /** |
| 847 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely |
| 848 | * @root: hierarchy root |
| 849 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() |
| 850 | */ |
| 851 | void mem_cgroup_iter_break(struct mem_cgroup *root, |
| 852 | struct mem_cgroup *prev) |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 853 | { |
| 854 | if (!root) |
| 855 | root = root_mem_cgroup; |
| 856 | if (prev && prev != root) |
| 857 | css_put(&prev->css); |
| 858 | } |
| 859 | |
Vladimir Davydov | 6df3868 | 2015-12-29 14:54:10 -0800 | [diff] [blame] | 860 | static void invalidate_reclaim_iterators(struct mem_cgroup *dead_memcg) |
| 861 | { |
| 862 | struct mem_cgroup *memcg = dead_memcg; |
| 863 | struct mem_cgroup_reclaim_iter *iter; |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 864 | struct mem_cgroup_per_node *mz; |
| 865 | int nid; |
Vladimir Davydov | 6df3868 | 2015-12-29 14:54:10 -0800 | [diff] [blame] | 866 | int i; |
| 867 | |
| 868 | while ((memcg = parent_mem_cgroup(memcg))) { |
| 869 | for_each_node(nid) { |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 870 | mz = mem_cgroup_nodeinfo(memcg, nid); |
| 871 | for (i = 0; i <= DEF_PRIORITY; i++) { |
| 872 | iter = &mz->iter[i]; |
| 873 | cmpxchg(&iter->position, |
| 874 | dead_memcg, NULL); |
Vladimir Davydov | 6df3868 | 2015-12-29 14:54:10 -0800 | [diff] [blame] | 875 | } |
| 876 | } |
| 877 | } |
| 878 | } |
| 879 | |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 880 | /* |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 881 | * Iteration constructs for visiting all cgroups (under a tree). If |
| 882 | * loops are exited prematurely (break), mem_cgroup_iter_break() must |
| 883 | * be used for reference counting. |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 884 | */ |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 885 | #define for_each_mem_cgroup_tree(iter, root) \ |
Johannes Weiner | 527a5ec | 2012-01-12 17:17:55 -0800 | [diff] [blame] | 886 | for (iter = mem_cgroup_iter(root, NULL, NULL); \ |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 887 | iter != NULL; \ |
Johannes Weiner | 527a5ec | 2012-01-12 17:17:55 -0800 | [diff] [blame] | 888 | iter = mem_cgroup_iter(root, iter, NULL)) |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 889 | |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 890 | #define for_each_mem_cgroup(iter) \ |
Johannes Weiner | 527a5ec | 2012-01-12 17:17:55 -0800 | [diff] [blame] | 891 | for (iter = mem_cgroup_iter(NULL, NULL, NULL); \ |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 892 | iter != NULL; \ |
Johannes Weiner | 527a5ec | 2012-01-12 17:17:55 -0800 | [diff] [blame] | 893 | iter = mem_cgroup_iter(NULL, iter, NULL)) |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 894 | |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 895 | /** |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 896 | * mem_cgroup_scan_tasks - iterate over tasks of a memory cgroup hierarchy |
| 897 | * @memcg: hierarchy root |
| 898 | * @fn: function to call for each task |
| 899 | * @arg: argument passed to @fn |
| 900 | * |
| 901 | * This function iterates over tasks attached to @memcg or to any of its |
| 902 | * descendants and calls @fn for each task. If @fn returns a non-zero |
| 903 | * value, the function breaks the iteration loop and returns the value. |
| 904 | * Otherwise, it will iterate over all tasks and return 0. |
| 905 | * |
| 906 | * This function must not be called for the root memory cgroup. |
| 907 | */ |
| 908 | int mem_cgroup_scan_tasks(struct mem_cgroup *memcg, |
| 909 | int (*fn)(struct task_struct *, void *), void *arg) |
| 910 | { |
| 911 | struct mem_cgroup *iter; |
| 912 | int ret = 0; |
| 913 | |
| 914 | BUG_ON(memcg == root_mem_cgroup); |
| 915 | |
| 916 | for_each_mem_cgroup_tree(iter, memcg) { |
| 917 | struct css_task_iter it; |
| 918 | struct task_struct *task; |
| 919 | |
| 920 | css_task_iter_start(&iter->css, &it); |
| 921 | while (!ret && (task = css_task_iter_next(&it))) |
| 922 | ret = fn(task, arg); |
| 923 | css_task_iter_end(&it); |
| 924 | if (ret) { |
| 925 | mem_cgroup_iter_break(memcg, iter); |
| 926 | break; |
| 927 | } |
| 928 | } |
| 929 | return ret; |
| 930 | } |
| 931 | |
| 932 | /** |
Johannes Weiner | dfe0e77 | 2014-12-10 15:43:43 -0800 | [diff] [blame] | 933 | * mem_cgroup_page_lruvec - return lruvec for isolating/putting an LRU page |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 934 | * @page: the page |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 935 | * @zone: zone of the page |
Johannes Weiner | dfe0e77 | 2014-12-10 15:43:43 -0800 | [diff] [blame] | 936 | * |
| 937 | * This function is only safe when following the LRU page isolation |
| 938 | * and putback protocol: the LRU lock must be held, and the page must |
| 939 | * either be PageLRU() or the caller must have isolated/allocated it. |
Minchan Kim | 3f58a82 | 2011-03-22 16:32:53 -0700 | [diff] [blame] | 940 | */ |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 941 | struct lruvec *mem_cgroup_page_lruvec(struct page *page, struct pglist_data *pgdat) |
Minchan Kim | 3f58a82 | 2011-03-22 16:32:53 -0700 | [diff] [blame] | 942 | { |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 943 | struct mem_cgroup_per_node *mz; |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 944 | struct mem_cgroup *memcg; |
Hugh Dickins | bea8c15 | 2012-11-16 14:14:54 -0800 | [diff] [blame] | 945 | struct lruvec *lruvec; |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 946 | |
Hugh Dickins | bea8c15 | 2012-11-16 14:14:54 -0800 | [diff] [blame] | 947 | if (mem_cgroup_disabled()) { |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 948 | lruvec = &pgdat->lruvec; |
Hugh Dickins | bea8c15 | 2012-11-16 14:14:54 -0800 | [diff] [blame] | 949 | goto out; |
| 950 | } |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 951 | |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 952 | memcg = page->mem_cgroup; |
Hugh Dickins | 7512102 | 2012-03-05 14:59:18 -0800 | [diff] [blame] | 953 | /* |
Johannes Weiner | dfe0e77 | 2014-12-10 15:43:43 -0800 | [diff] [blame] | 954 | * Swapcache readahead pages are added to the LRU - and |
Johannes Weiner | 2983331 | 2014-12-10 15:44:02 -0800 | [diff] [blame] | 955 | * possibly migrated - before they are charged. |
Hugh Dickins | 7512102 | 2012-03-05 14:59:18 -0800 | [diff] [blame] | 956 | */ |
Johannes Weiner | 2983331 | 2014-12-10 15:44:02 -0800 | [diff] [blame] | 957 | if (!memcg) |
| 958 | memcg = root_mem_cgroup; |
Hugh Dickins | 7512102 | 2012-03-05 14:59:18 -0800 | [diff] [blame] | 959 | |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 960 | mz = mem_cgroup_page_nodeinfo(memcg, page); |
Hugh Dickins | bea8c15 | 2012-11-16 14:14:54 -0800 | [diff] [blame] | 961 | lruvec = &mz->lruvec; |
| 962 | out: |
| 963 | /* |
| 964 | * Since a node can be onlined after the mem_cgroup was created, |
| 965 | * we have to be prepared to initialize lruvec->zone here; |
| 966 | * and if offlined then reonlined, we need to reinitialize it. |
| 967 | */ |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 968 | if (unlikely(lruvec->pgdat != pgdat)) |
| 969 | lruvec->pgdat = pgdat; |
Hugh Dickins | bea8c15 | 2012-11-16 14:14:54 -0800 | [diff] [blame] | 970 | return lruvec; |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 971 | } |
| 972 | |
| 973 | /** |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 974 | * mem_cgroup_update_lru_size - account for adding or removing an lru page |
| 975 | * @lruvec: mem_cgroup per zone lru vector |
| 976 | * @lru: index of lru list the page is sitting on |
Michal Hocko | b4536f0c8 | 2017-01-10 16:58:04 -0800 | [diff] [blame] | 977 | * @zid: zone id of the accounted pages |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 978 | * @nr_pages: positive when adding or negative when removing |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 979 | * |
Hugh Dickins | ca70723 | 2016-05-19 17:12:35 -0700 | [diff] [blame] | 980 | * This function must be called under lru_lock, just before a page is added |
| 981 | * to or just after a page is removed from an lru list (that ordering being |
| 982 | * so as to allow it to check that lru_size 0 is consistent with list_empty). |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 983 | */ |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 984 | void mem_cgroup_update_lru_size(struct lruvec *lruvec, enum lru_list lru, |
Michal Hocko | b4536f0c8 | 2017-01-10 16:58:04 -0800 | [diff] [blame] | 985 | int zid, int nr_pages) |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 986 | { |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 987 | struct mem_cgroup_per_node *mz; |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 988 | unsigned long *lru_size; |
Hugh Dickins | ca70723 | 2016-05-19 17:12:35 -0700 | [diff] [blame] | 989 | long size; |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 990 | |
| 991 | if (mem_cgroup_disabled()) |
| 992 | return; |
| 993 | |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 994 | mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec); |
Michal Hocko | b4536f0c8 | 2017-01-10 16:58:04 -0800 | [diff] [blame] | 995 | lru_size = &mz->lru_zone_size[zid][lru]; |
Hugh Dickins | ca70723 | 2016-05-19 17:12:35 -0700 | [diff] [blame] | 996 | |
| 997 | if (nr_pages < 0) |
| 998 | *lru_size += nr_pages; |
| 999 | |
| 1000 | size = *lru_size; |
Michal Hocko | b4536f0c8 | 2017-01-10 16:58:04 -0800 | [diff] [blame] | 1001 | if (WARN_ONCE(size < 0, |
| 1002 | "%s(%p, %d, %d): lru_size %ld\n", |
| 1003 | __func__, lruvec, lru, nr_pages, size)) { |
Hugh Dickins | ca70723 | 2016-05-19 17:12:35 -0700 | [diff] [blame] | 1004 | VM_BUG_ON(1); |
| 1005 | *lru_size = 0; |
| 1006 | } |
| 1007 | |
| 1008 | if (nr_pages > 0) |
| 1009 | *lru_size += nr_pages; |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 1010 | } |
KAMEZAWA Hiroyuki | 544122e | 2009-01-07 18:08:34 -0800 | [diff] [blame] | 1011 | |
Johannes Weiner | 2314b42 | 2014-12-10 15:44:33 -0800 | [diff] [blame] | 1012 | bool task_in_mem_cgroup(struct task_struct *task, struct mem_cgroup *memcg) |
Johannes Weiner | c3ac9a8 | 2012-05-29 15:06:25 -0700 | [diff] [blame] | 1013 | { |
Johannes Weiner | 2314b42 | 2014-12-10 15:44:33 -0800 | [diff] [blame] | 1014 | struct mem_cgroup *task_memcg; |
KAMEZAWA Hiroyuki | 158e0a2 | 2010-08-10 18:03:00 -0700 | [diff] [blame] | 1015 | struct task_struct *p; |
David Rientjes | ffbdccf | 2013-07-03 15:01:23 -0700 | [diff] [blame] | 1016 | bool ret; |
David Rientjes | 4c4a221 | 2008-02-07 00:14:06 -0800 | [diff] [blame] | 1017 | |
KAMEZAWA Hiroyuki | 158e0a2 | 2010-08-10 18:03:00 -0700 | [diff] [blame] | 1018 | p = find_lock_task_mm(task); |
David Rientjes | de077d2 | 2012-01-12 17:18:52 -0800 | [diff] [blame] | 1019 | if (p) { |
Johannes Weiner | 2314b42 | 2014-12-10 15:44:33 -0800 | [diff] [blame] | 1020 | task_memcg = get_mem_cgroup_from_mm(p->mm); |
David Rientjes | de077d2 | 2012-01-12 17:18:52 -0800 | [diff] [blame] | 1021 | task_unlock(p); |
| 1022 | } else { |
| 1023 | /* |
| 1024 | * All threads may have already detached their mm's, but the oom |
| 1025 | * killer still needs to detect if they have already been oom |
| 1026 | * killed to prevent needlessly killing additional tasks. |
| 1027 | */ |
David Rientjes | ffbdccf | 2013-07-03 15:01:23 -0700 | [diff] [blame] | 1028 | rcu_read_lock(); |
Johannes Weiner | 2314b42 | 2014-12-10 15:44:33 -0800 | [diff] [blame] | 1029 | task_memcg = mem_cgroup_from_task(task); |
| 1030 | css_get(&task_memcg->css); |
David Rientjes | ffbdccf | 2013-07-03 15:01:23 -0700 | [diff] [blame] | 1031 | rcu_read_unlock(); |
David Rientjes | de077d2 | 2012-01-12 17:18:52 -0800 | [diff] [blame] | 1032 | } |
Johannes Weiner | 2314b42 | 2014-12-10 15:44:33 -0800 | [diff] [blame] | 1033 | ret = mem_cgroup_is_descendant(task_memcg, memcg); |
| 1034 | css_put(&task_memcg->css); |
David Rientjes | 4c4a221 | 2008-02-07 00:14:06 -0800 | [diff] [blame] | 1035 | return ret; |
| 1036 | } |
| 1037 | |
Johannes Weiner | 1994282 | 2011-02-01 15:52:43 -0800 | [diff] [blame] | 1038 | /** |
Johannes Weiner | 9d11ea9 | 2011-03-23 16:42:21 -0700 | [diff] [blame] | 1039 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
Wanpeng Li | dad7557 | 2012-06-20 12:53:01 -0700 | [diff] [blame] | 1040 | * @memcg: the memory cgroup |
Johannes Weiner | 1994282 | 2011-02-01 15:52:43 -0800 | [diff] [blame] | 1041 | * |
Johannes Weiner | 9d11ea9 | 2011-03-23 16:42:21 -0700 | [diff] [blame] | 1042 | * Returns the maximum amount of memory @mem can be charged with, in |
Johannes Weiner | 7ec99d6 | 2011-03-23 16:42:36 -0700 | [diff] [blame] | 1043 | * pages. |
Johannes Weiner | 1994282 | 2011-02-01 15:52:43 -0800 | [diff] [blame] | 1044 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1045 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
Johannes Weiner | 1994282 | 2011-02-01 15:52:43 -0800 | [diff] [blame] | 1046 | { |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1047 | unsigned long margin = 0; |
| 1048 | unsigned long count; |
| 1049 | unsigned long limit; |
Johannes Weiner | 9d11ea9 | 2011-03-23 16:42:21 -0700 | [diff] [blame] | 1050 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1051 | count = page_counter_read(&memcg->memory); |
Jason Low | 4db0c3c | 2015-04-15 16:14:08 -0700 | [diff] [blame] | 1052 | limit = READ_ONCE(memcg->memory.limit); |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1053 | if (count < limit) |
| 1054 | margin = limit - count; |
| 1055 | |
Johannes Weiner | 7941d21 | 2016-01-14 15:21:23 -0800 | [diff] [blame] | 1056 | if (do_memsw_account()) { |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1057 | count = page_counter_read(&memcg->memsw); |
Jason Low | 4db0c3c | 2015-04-15 16:14:08 -0700 | [diff] [blame] | 1058 | limit = READ_ONCE(memcg->memsw.limit); |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1059 | if (count <= limit) |
| 1060 | margin = min(margin, limit - count); |
Li RongQing | cbedbac | 2016-05-27 14:27:43 -0700 | [diff] [blame] | 1061 | else |
| 1062 | margin = 0; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1063 | } |
| 1064 | |
| 1065 | return margin; |
Johannes Weiner | 1994282 | 2011-02-01 15:52:43 -0800 | [diff] [blame] | 1066 | } |
| 1067 | |
KAMEZAWA Hiroyuki | 619d094 | 2012-03-21 16:34:23 -0700 | [diff] [blame] | 1068 | /* |
Qiang Huang | bdcbb65 | 2014-06-04 16:08:21 -0700 | [diff] [blame] | 1069 | * A routine for checking "mem" is under move_account() or not. |
KAMEZAWA Hiroyuki | 32047e2 | 2010-10-27 15:33:40 -0700 | [diff] [blame] | 1070 | * |
Qiang Huang | bdcbb65 | 2014-06-04 16:08:21 -0700 | [diff] [blame] | 1071 | * Checking a cgroup is mc.from or mc.to or under hierarchy of |
| 1072 | * moving cgroups. This is for waiting at high-memory pressure |
| 1073 | * caused by "move". |
KAMEZAWA Hiroyuki | 32047e2 | 2010-10-27 15:33:40 -0700 | [diff] [blame] | 1074 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1075 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 1076 | { |
KAMEZAWA Hiroyuki | 2bd9bb2 | 2010-08-10 18:02:58 -0700 | [diff] [blame] | 1077 | struct mem_cgroup *from; |
| 1078 | struct mem_cgroup *to; |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 1079 | bool ret = false; |
KAMEZAWA Hiroyuki | 2bd9bb2 | 2010-08-10 18:02:58 -0700 | [diff] [blame] | 1080 | /* |
| 1081 | * Unlike task_move routines, we access mc.to, mc.from not under |
| 1082 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. |
| 1083 | */ |
| 1084 | spin_lock(&mc.lock); |
| 1085 | from = mc.from; |
| 1086 | to = mc.to; |
| 1087 | if (!from) |
| 1088 | goto unlock; |
Michal Hocko | 3e92041 | 2011-07-26 16:08:29 -0700 | [diff] [blame] | 1089 | |
Johannes Weiner | 2314b42 | 2014-12-10 15:44:33 -0800 | [diff] [blame] | 1090 | ret = mem_cgroup_is_descendant(from, memcg) || |
| 1091 | mem_cgroup_is_descendant(to, memcg); |
KAMEZAWA Hiroyuki | 2bd9bb2 | 2010-08-10 18:02:58 -0700 | [diff] [blame] | 1092 | unlock: |
| 1093 | spin_unlock(&mc.lock); |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 1094 | return ret; |
| 1095 | } |
| 1096 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1097 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 1098 | { |
| 1099 | if (mc.moving_task && current != mc.moving_task) { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1100 | if (mem_cgroup_under_move(memcg)) { |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 1101 | DEFINE_WAIT(wait); |
| 1102 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); |
| 1103 | /* moving charge context might have finished. */ |
| 1104 | if (mc.moving_task) |
| 1105 | schedule(); |
| 1106 | finish_wait(&mc.waitq, &wait); |
| 1107 | return true; |
| 1108 | } |
| 1109 | } |
| 1110 | return false; |
| 1111 | } |
| 1112 | |
Johannes Weiner | 71cd311 | 2017-05-03 14:55:13 -0700 | [diff] [blame] | 1113 | unsigned int memcg1_stats[] = { |
| 1114 | MEMCG_CACHE, |
| 1115 | MEMCG_RSS, |
| 1116 | MEMCG_RSS_HUGE, |
| 1117 | NR_SHMEM, |
| 1118 | NR_FILE_MAPPED, |
| 1119 | NR_FILE_DIRTY, |
| 1120 | NR_WRITEBACK, |
| 1121 | MEMCG_SWAP, |
| 1122 | }; |
| 1123 | |
| 1124 | static const char *const memcg1_stat_names[] = { |
| 1125 | "cache", |
| 1126 | "rss", |
| 1127 | "rss_huge", |
| 1128 | "shmem", |
| 1129 | "mapped_file", |
| 1130 | "dirty", |
| 1131 | "writeback", |
| 1132 | "swap", |
| 1133 | }; |
| 1134 | |
Sha Zhengju | 58cf188 | 2013-02-22 16:32:05 -0800 | [diff] [blame] | 1135 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
Balbir Singh | e222432 | 2009-04-02 16:57:39 -0700 | [diff] [blame] | 1136 | /** |
Sha Zhengju | 58cf188 | 2013-02-22 16:32:05 -0800 | [diff] [blame] | 1137 | * mem_cgroup_print_oom_info: Print OOM information relevant to memory controller. |
Balbir Singh | e222432 | 2009-04-02 16:57:39 -0700 | [diff] [blame] | 1138 | * @memcg: The memory cgroup that went over limit |
| 1139 | * @p: Task that is going to be killed |
| 1140 | * |
| 1141 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is |
| 1142 | * enabled |
| 1143 | */ |
| 1144 | void mem_cgroup_print_oom_info(struct mem_cgroup *memcg, struct task_struct *p) |
| 1145 | { |
Sha Zhengju | 58cf188 | 2013-02-22 16:32:05 -0800 | [diff] [blame] | 1146 | struct mem_cgroup *iter; |
| 1147 | unsigned int i; |
Balbir Singh | e222432 | 2009-04-02 16:57:39 -0700 | [diff] [blame] | 1148 | |
Balbir Singh | e222432 | 2009-04-02 16:57:39 -0700 | [diff] [blame] | 1149 | rcu_read_lock(); |
| 1150 | |
Balasubramani Vivekanandan | 2415b9f | 2015-04-14 15:48:18 -0700 | [diff] [blame] | 1151 | if (p) { |
| 1152 | pr_info("Task in "); |
| 1153 | pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id)); |
| 1154 | pr_cont(" killed as a result of limit of "); |
| 1155 | } else { |
| 1156 | pr_info("Memory limit reached of cgroup "); |
| 1157 | } |
| 1158 | |
Tejun Heo | e61734c | 2014-02-12 09:29:50 -0500 | [diff] [blame] | 1159 | pr_cont_cgroup_path(memcg->css.cgroup); |
Greg Thelen | 0346dad | 2015-01-26 12:58:38 -0800 | [diff] [blame] | 1160 | pr_cont("\n"); |
Balbir Singh | e222432 | 2009-04-02 16:57:39 -0700 | [diff] [blame] | 1161 | |
Balbir Singh | e222432 | 2009-04-02 16:57:39 -0700 | [diff] [blame] | 1162 | rcu_read_unlock(); |
| 1163 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1164 | pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n", |
| 1165 | K((u64)page_counter_read(&memcg->memory)), |
| 1166 | K((u64)memcg->memory.limit), memcg->memory.failcnt); |
| 1167 | pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n", |
| 1168 | K((u64)page_counter_read(&memcg->memsw)), |
| 1169 | K((u64)memcg->memsw.limit), memcg->memsw.failcnt); |
| 1170 | pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n", |
| 1171 | K((u64)page_counter_read(&memcg->kmem)), |
| 1172 | K((u64)memcg->kmem.limit), memcg->kmem.failcnt); |
Sha Zhengju | 58cf188 | 2013-02-22 16:32:05 -0800 | [diff] [blame] | 1173 | |
| 1174 | for_each_mem_cgroup_tree(iter, memcg) { |
Tejun Heo | e61734c | 2014-02-12 09:29:50 -0500 | [diff] [blame] | 1175 | pr_info("Memory cgroup stats for "); |
| 1176 | pr_cont_cgroup_path(iter->css.cgroup); |
Sha Zhengju | 58cf188 | 2013-02-22 16:32:05 -0800 | [diff] [blame] | 1177 | pr_cont(":"); |
| 1178 | |
Johannes Weiner | 71cd311 | 2017-05-03 14:55:13 -0700 | [diff] [blame] | 1179 | for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) { |
| 1180 | if (memcg1_stats[i] == MEMCG_SWAP && !do_swap_account) |
Sha Zhengju | 58cf188 | 2013-02-22 16:32:05 -0800 | [diff] [blame] | 1181 | continue; |
Johannes Weiner | 71cd311 | 2017-05-03 14:55:13 -0700 | [diff] [blame] | 1182 | pr_cont(" %s:%luKB", memcg1_stat_names[i], |
Johannes Weiner | ccda7f4 | 2017-05-03 14:55:16 -0700 | [diff] [blame] | 1183 | K(memcg_page_state(iter, memcg1_stats[i]))); |
Sha Zhengju | 58cf188 | 2013-02-22 16:32:05 -0800 | [diff] [blame] | 1184 | } |
| 1185 | |
| 1186 | for (i = 0; i < NR_LRU_LISTS; i++) |
| 1187 | pr_cont(" %s:%luKB", mem_cgroup_lru_names[i], |
| 1188 | K(mem_cgroup_nr_lru_pages(iter, BIT(i)))); |
| 1189 | |
| 1190 | pr_cont("\n"); |
| 1191 | } |
Balbir Singh | e222432 | 2009-04-02 16:57:39 -0700 | [diff] [blame] | 1192 | } |
| 1193 | |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 1194 | /* |
| 1195 | * This function returns the number of memcg under hierarchy tree. Returns |
| 1196 | * 1(self count) if no children. |
| 1197 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1198 | static int mem_cgroup_count_children(struct mem_cgroup *memcg) |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 1199 | { |
| 1200 | int num = 0; |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 1201 | struct mem_cgroup *iter; |
| 1202 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1203 | for_each_mem_cgroup_tree(iter, memcg) |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 1204 | num++; |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 1205 | return num; |
| 1206 | } |
| 1207 | |
Balbir Singh | 6d61ef4 | 2009-01-07 18:08:06 -0800 | [diff] [blame] | 1208 | /* |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 1209 | * Return the memory (and swap, if configured) limit for a memcg. |
| 1210 | */ |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 1211 | unsigned long mem_cgroup_get_limit(struct mem_cgroup *memcg) |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 1212 | { |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1213 | unsigned long limit; |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 1214 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1215 | limit = memcg->memory.limit; |
Michal Hocko | 9a5a8f1 | 2012-11-16 14:14:49 -0800 | [diff] [blame] | 1216 | if (mem_cgroup_swappiness(memcg)) { |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1217 | unsigned long memsw_limit; |
Vladimir Davydov | 37e8435 | 2016-01-20 15:02:56 -0800 | [diff] [blame] | 1218 | unsigned long swap_limit; |
Michal Hocko | 9a5a8f1 | 2012-11-16 14:14:49 -0800 | [diff] [blame] | 1219 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1220 | memsw_limit = memcg->memsw.limit; |
Vladimir Davydov | 37e8435 | 2016-01-20 15:02:56 -0800 | [diff] [blame] | 1221 | swap_limit = memcg->swap.limit; |
| 1222 | swap_limit = min(swap_limit, (unsigned long)total_swap_pages); |
| 1223 | limit = min(limit + swap_limit, memsw_limit); |
Michal Hocko | 9a5a8f1 | 2012-11-16 14:14:49 -0800 | [diff] [blame] | 1224 | } |
Michal Hocko | 9a5a8f1 | 2012-11-16 14:14:49 -0800 | [diff] [blame] | 1225 | return limit; |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 1226 | } |
| 1227 | |
Johannes Weiner | b6e6edc | 2016-03-17 14:20:28 -0700 | [diff] [blame] | 1228 | static bool mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask, |
David Rientjes | 1996546 | 2012-12-11 16:00:26 -0800 | [diff] [blame] | 1229 | int order) |
David Rientjes | 9cbb78b | 2012-07-31 16:43:44 -0700 | [diff] [blame] | 1230 | { |
David Rientjes | 6e0fc46 | 2015-09-08 15:00:36 -0700 | [diff] [blame] | 1231 | struct oom_control oc = { |
| 1232 | .zonelist = NULL, |
| 1233 | .nodemask = NULL, |
Vladimir Davydov | 2a966b7 | 2016-07-26 15:22:33 -0700 | [diff] [blame] | 1234 | .memcg = memcg, |
David Rientjes | 6e0fc46 | 2015-09-08 15:00:36 -0700 | [diff] [blame] | 1235 | .gfp_mask = gfp_mask, |
| 1236 | .order = order, |
David Rientjes | 6e0fc46 | 2015-09-08 15:00:36 -0700 | [diff] [blame] | 1237 | }; |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 1238 | bool ret; |
David Rientjes | 9cbb78b | 2012-07-31 16:43:44 -0700 | [diff] [blame] | 1239 | |
Johannes Weiner | dc56401 | 2015-06-24 16:57:19 -0700 | [diff] [blame] | 1240 | mutex_lock(&oom_lock); |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 1241 | ret = out_of_memory(&oc); |
Johannes Weiner | dc56401 | 2015-06-24 16:57:19 -0700 | [diff] [blame] | 1242 | mutex_unlock(&oom_lock); |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 1243 | return ret; |
David Rientjes | 9cbb78b | 2012-07-31 16:43:44 -0700 | [diff] [blame] | 1244 | } |
| 1245 | |
Michele Curti | ae6e71d | 2014-12-12 16:56:35 -0800 | [diff] [blame] | 1246 | #if MAX_NUMNODES > 1 |
| 1247 | |
KAMEZAWA Hiroyuki | 4d0c066 | 2011-07-08 15:39:42 -0700 | [diff] [blame] | 1248 | /** |
| 1249 | * test_mem_cgroup_node_reclaimable |
Wanpeng Li | dad7557 | 2012-06-20 12:53:01 -0700 | [diff] [blame] | 1250 | * @memcg: the target memcg |
KAMEZAWA Hiroyuki | 4d0c066 | 2011-07-08 15:39:42 -0700 | [diff] [blame] | 1251 | * @nid: the node ID to be checked. |
| 1252 | * @noswap : specify true here if the user wants flle only information. |
| 1253 | * |
| 1254 | * This function returns whether the specified memcg contains any |
| 1255 | * reclaimable pages on a node. Returns true if there are any reclaimable |
| 1256 | * pages in the node. |
| 1257 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1258 | static bool test_mem_cgroup_node_reclaimable(struct mem_cgroup *memcg, |
KAMEZAWA Hiroyuki | 4d0c066 | 2011-07-08 15:39:42 -0700 | [diff] [blame] | 1259 | int nid, bool noswap) |
| 1260 | { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1261 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_FILE)) |
KAMEZAWA Hiroyuki | 4d0c066 | 2011-07-08 15:39:42 -0700 | [diff] [blame] | 1262 | return true; |
| 1263 | if (noswap || !total_swap_pages) |
| 1264 | return false; |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1265 | if (mem_cgroup_node_nr_lru_pages(memcg, nid, LRU_ALL_ANON)) |
KAMEZAWA Hiroyuki | 4d0c066 | 2011-07-08 15:39:42 -0700 | [diff] [blame] | 1266 | return true; |
| 1267 | return false; |
| 1268 | |
| 1269 | } |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1270 | |
| 1271 | /* |
| 1272 | * Always updating the nodemask is not very good - even if we have an empty |
| 1273 | * list or the wrong list here, we can start from some node and traverse all |
| 1274 | * nodes based on the zonelist. So update the list loosely once per 10 secs. |
| 1275 | * |
| 1276 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1277 | static void mem_cgroup_may_update_nodemask(struct mem_cgroup *memcg) |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1278 | { |
| 1279 | int nid; |
KAMEZAWA Hiroyuki | 453a9bf | 2011-07-08 15:39:43 -0700 | [diff] [blame] | 1280 | /* |
| 1281 | * numainfo_events > 0 means there was at least NUMAINFO_EVENTS_TARGET |
| 1282 | * pagein/pageout changes since the last update. |
| 1283 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1284 | if (!atomic_read(&memcg->numainfo_events)) |
KAMEZAWA Hiroyuki | 453a9bf | 2011-07-08 15:39:43 -0700 | [diff] [blame] | 1285 | return; |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1286 | if (atomic_inc_return(&memcg->numainfo_updating) > 1) |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1287 | return; |
| 1288 | |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1289 | /* make a nodemask where this memcg uses memory from */ |
Lai Jiangshan | 31aaea4 | 2012-12-12 13:51:27 -0800 | [diff] [blame] | 1290 | memcg->scan_nodes = node_states[N_MEMORY]; |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1291 | |
Lai Jiangshan | 31aaea4 | 2012-12-12 13:51:27 -0800 | [diff] [blame] | 1292 | for_each_node_mask(nid, node_states[N_MEMORY]) { |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1293 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1294 | if (!test_mem_cgroup_node_reclaimable(memcg, nid, false)) |
| 1295 | node_clear(nid, memcg->scan_nodes); |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1296 | } |
KAMEZAWA Hiroyuki | 453a9bf | 2011-07-08 15:39:43 -0700 | [diff] [blame] | 1297 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1298 | atomic_set(&memcg->numainfo_events, 0); |
| 1299 | atomic_set(&memcg->numainfo_updating, 0); |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1300 | } |
| 1301 | |
| 1302 | /* |
| 1303 | * Selecting a node where we start reclaim from. Because what we need is just |
| 1304 | * reducing usage counter, start from anywhere is O,K. Considering |
| 1305 | * memory reclaim from current node, there are pros. and cons. |
| 1306 | * |
| 1307 | * Freeing memory from current node means freeing memory from a node which |
| 1308 | * we'll use or we've used. So, it may make LRU bad. And if several threads |
| 1309 | * hit limits, it will see a contention on a node. But freeing from remote |
| 1310 | * node means more costs for memory reclaim because of memory latency. |
| 1311 | * |
| 1312 | * Now, we use round-robin. Better algorithm is welcomed. |
| 1313 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1314 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1315 | { |
| 1316 | int node; |
| 1317 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1318 | mem_cgroup_may_update_nodemask(memcg); |
| 1319 | node = memcg->last_scanned_node; |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1320 | |
Andrew Morton | 0edaf86 | 2016-05-19 17:10:58 -0700 | [diff] [blame] | 1321 | node = next_node_in(node, memcg->scan_nodes); |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1322 | /* |
Michal Hocko | fda3d69 | 2016-05-19 17:11:34 -0700 | [diff] [blame] | 1323 | * mem_cgroup_may_update_nodemask might have seen no reclaimmable pages |
| 1324 | * last time it really checked all the LRUs due to rate limiting. |
| 1325 | * Fallback to the current node in that case for simplicity. |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1326 | */ |
| 1327 | if (unlikely(node == MAX_NUMNODES)) |
| 1328 | node = numa_node_id(); |
| 1329 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1330 | memcg->last_scanned_node = node; |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1331 | return node; |
| 1332 | } |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1333 | #else |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1334 | int mem_cgroup_select_victim_node(struct mem_cgroup *memcg) |
Ying Han | 889976d | 2011-05-26 16:25:33 -0700 | [diff] [blame] | 1335 | { |
| 1336 | return 0; |
| 1337 | } |
| 1338 | #endif |
| 1339 | |
Andrew Morton | 0608f43 | 2013-09-24 15:27:41 -0700 | [diff] [blame] | 1340 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 1341 | pg_data_t *pgdat, |
Andrew Morton | 0608f43 | 2013-09-24 15:27:41 -0700 | [diff] [blame] | 1342 | gfp_t gfp_mask, |
| 1343 | unsigned long *total_scanned) |
Balbir Singh | 6d61ef4 | 2009-01-07 18:08:06 -0800 | [diff] [blame] | 1344 | { |
Andrew Morton | 0608f43 | 2013-09-24 15:27:41 -0700 | [diff] [blame] | 1345 | struct mem_cgroup *victim = NULL; |
| 1346 | int total = 0; |
| 1347 | int loop = 0; |
| 1348 | unsigned long excess; |
| 1349 | unsigned long nr_scanned; |
| 1350 | struct mem_cgroup_reclaim_cookie reclaim = { |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 1351 | .pgdat = pgdat, |
Andrew Morton | 0608f43 | 2013-09-24 15:27:41 -0700 | [diff] [blame] | 1352 | .priority = 0, |
| 1353 | }; |
Johannes Weiner | 9d11ea9 | 2011-03-23 16:42:21 -0700 | [diff] [blame] | 1354 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1355 | excess = soft_limit_excess(root_memcg); |
Balbir Singh | 6d61ef4 | 2009-01-07 18:08:06 -0800 | [diff] [blame] | 1356 | |
Andrew Morton | 0608f43 | 2013-09-24 15:27:41 -0700 | [diff] [blame] | 1357 | while (1) { |
| 1358 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); |
| 1359 | if (!victim) { |
| 1360 | loop++; |
| 1361 | if (loop >= 2) { |
| 1362 | /* |
| 1363 | * If we have not been able to reclaim |
| 1364 | * anything, it might because there are |
| 1365 | * no reclaimable pages under this hierarchy |
| 1366 | */ |
| 1367 | if (!total) |
| 1368 | break; |
| 1369 | /* |
| 1370 | * We want to do more targeted reclaim. |
| 1371 | * excess >> 2 is not to excessive so as to |
| 1372 | * reclaim too much, nor too less that we keep |
| 1373 | * coming back to reclaim from this cgroup |
| 1374 | */ |
| 1375 | if (total >= (excess >> 2) || |
| 1376 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) |
| 1377 | break; |
| 1378 | } |
| 1379 | continue; |
| 1380 | } |
Mel Gorman | a9dd0a8 | 2016-07-28 15:46:02 -0700 | [diff] [blame] | 1381 | total += mem_cgroup_shrink_node(victim, gfp_mask, false, |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 1382 | pgdat, &nr_scanned); |
Andrew Morton | 0608f43 | 2013-09-24 15:27:41 -0700 | [diff] [blame] | 1383 | *total_scanned += nr_scanned; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1384 | if (!soft_limit_excess(root_memcg)) |
Andrew Morton | 0608f43 | 2013-09-24 15:27:41 -0700 | [diff] [blame] | 1385 | break; |
Balbir Singh | 6d61ef4 | 2009-01-07 18:08:06 -0800 | [diff] [blame] | 1386 | } |
Andrew Morton | 0608f43 | 2013-09-24 15:27:41 -0700 | [diff] [blame] | 1387 | mem_cgroup_iter_break(root_memcg, victim); |
| 1388 | return total; |
Balbir Singh | 6d61ef4 | 2009-01-07 18:08:06 -0800 | [diff] [blame] | 1389 | } |
| 1390 | |
Johannes Weiner | 0056f4e | 2013-10-31 16:34:14 -0700 | [diff] [blame] | 1391 | #ifdef CONFIG_LOCKDEP |
| 1392 | static struct lockdep_map memcg_oom_lock_dep_map = { |
| 1393 | .name = "memcg_oom_lock", |
| 1394 | }; |
| 1395 | #endif |
| 1396 | |
Johannes Weiner | fb2a6fc | 2013-09-12 15:13:43 -0700 | [diff] [blame] | 1397 | static DEFINE_SPINLOCK(memcg_oom_lock); |
| 1398 | |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1399 | /* |
| 1400 | * Check OOM-Killer is already running under our hierarchy. |
| 1401 | * If someone is running, return false. |
| 1402 | */ |
Johannes Weiner | fb2a6fc | 2013-09-12 15:13:43 -0700 | [diff] [blame] | 1403 | static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg) |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1404 | { |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 1405 | struct mem_cgroup *iter, *failed = NULL; |
KAMEZAWA Hiroyuki | a636b32 | 2009-01-07 18:08:08 -0800 | [diff] [blame] | 1406 | |
Johannes Weiner | fb2a6fc | 2013-09-12 15:13:43 -0700 | [diff] [blame] | 1407 | spin_lock(&memcg_oom_lock); |
| 1408 | |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 1409 | for_each_mem_cgroup_tree(iter, memcg) { |
Johannes Weiner | 23751be | 2011-08-25 15:59:16 -0700 | [diff] [blame] | 1410 | if (iter->oom_lock) { |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 1411 | /* |
| 1412 | * this subtree of our hierarchy is already locked |
| 1413 | * so we cannot give a lock. |
| 1414 | */ |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 1415 | failed = iter; |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 1416 | mem_cgroup_iter_break(memcg, iter); |
| 1417 | break; |
Johannes Weiner | 23751be | 2011-08-25 15:59:16 -0700 | [diff] [blame] | 1418 | } else |
| 1419 | iter->oom_lock = true; |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 1420 | } |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1421 | |
Johannes Weiner | fb2a6fc | 2013-09-12 15:13:43 -0700 | [diff] [blame] | 1422 | if (failed) { |
| 1423 | /* |
| 1424 | * OK, we failed to lock the whole subtree so we have |
| 1425 | * to clean up what we set up to the failing subtree |
| 1426 | */ |
| 1427 | for_each_mem_cgroup_tree(iter, memcg) { |
| 1428 | if (iter == failed) { |
| 1429 | mem_cgroup_iter_break(memcg, iter); |
| 1430 | break; |
| 1431 | } |
| 1432 | iter->oom_lock = false; |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 1433 | } |
Johannes Weiner | 0056f4e | 2013-10-31 16:34:14 -0700 | [diff] [blame] | 1434 | } else |
| 1435 | mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_); |
Johannes Weiner | fb2a6fc | 2013-09-12 15:13:43 -0700 | [diff] [blame] | 1436 | |
| 1437 | spin_unlock(&memcg_oom_lock); |
| 1438 | |
| 1439 | return !failed; |
KAMEZAWA Hiroyuki | a636b32 | 2009-01-07 18:08:08 -0800 | [diff] [blame] | 1440 | } |
KAMEZAWA Hiroyuki | 0b7f569 | 2009-04-02 16:57:38 -0700 | [diff] [blame] | 1441 | |
Johannes Weiner | fb2a6fc | 2013-09-12 15:13:43 -0700 | [diff] [blame] | 1442 | static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
KAMEZAWA Hiroyuki | 0b7f569 | 2009-04-02 16:57:38 -0700 | [diff] [blame] | 1443 | { |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 1444 | struct mem_cgroup *iter; |
| 1445 | |
Johannes Weiner | fb2a6fc | 2013-09-12 15:13:43 -0700 | [diff] [blame] | 1446 | spin_lock(&memcg_oom_lock); |
Johannes Weiner | 0056f4e | 2013-10-31 16:34:14 -0700 | [diff] [blame] | 1447 | mutex_release(&memcg_oom_lock_dep_map, 1, _RET_IP_); |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1448 | for_each_mem_cgroup_tree(iter, memcg) |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 1449 | iter->oom_lock = false; |
Johannes Weiner | fb2a6fc | 2013-09-12 15:13:43 -0700 | [diff] [blame] | 1450 | spin_unlock(&memcg_oom_lock); |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 1451 | } |
| 1452 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1453 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 1454 | { |
| 1455 | struct mem_cgroup *iter; |
| 1456 | |
Tejun Heo | c2b42d3 | 2015-06-24 16:58:23 -0700 | [diff] [blame] | 1457 | spin_lock(&memcg_oom_lock); |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1458 | for_each_mem_cgroup_tree(iter, memcg) |
Tejun Heo | c2b42d3 | 2015-06-24 16:58:23 -0700 | [diff] [blame] | 1459 | iter->under_oom++; |
| 1460 | spin_unlock(&memcg_oom_lock); |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 1461 | } |
| 1462 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1463 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 1464 | { |
| 1465 | struct mem_cgroup *iter; |
| 1466 | |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1467 | /* |
| 1468 | * When a new child is created while the hierarchy is under oom, |
Tejun Heo | c2b42d3 | 2015-06-24 16:58:23 -0700 | [diff] [blame] | 1469 | * mem_cgroup_oom_lock() may not be called. Watch for underflow. |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1470 | */ |
Tejun Heo | c2b42d3 | 2015-06-24 16:58:23 -0700 | [diff] [blame] | 1471 | spin_lock(&memcg_oom_lock); |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1472 | for_each_mem_cgroup_tree(iter, memcg) |
Tejun Heo | c2b42d3 | 2015-06-24 16:58:23 -0700 | [diff] [blame] | 1473 | if (iter->under_oom > 0) |
| 1474 | iter->under_oom--; |
| 1475 | spin_unlock(&memcg_oom_lock); |
KAMEZAWA Hiroyuki | 0b7f569 | 2009-04-02 16:57:38 -0700 | [diff] [blame] | 1476 | } |
| 1477 | |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1478 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
| 1479 | |
KAMEZAWA Hiroyuki | dc98df5 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 1480 | struct oom_wait_info { |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 1481 | struct mem_cgroup *memcg; |
Ingo Molnar | ac6424b | 2017-06-20 12:06:13 +0200 | [diff] [blame] | 1482 | wait_queue_entry_t wait; |
KAMEZAWA Hiroyuki | dc98df5 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 1483 | }; |
| 1484 | |
Ingo Molnar | ac6424b | 2017-06-20 12:06:13 +0200 | [diff] [blame] | 1485 | static int memcg_oom_wake_function(wait_queue_entry_t *wait, |
KAMEZAWA Hiroyuki | dc98df5 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 1486 | unsigned mode, int sync, void *arg) |
| 1487 | { |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 1488 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
| 1489 | struct mem_cgroup *oom_wait_memcg; |
KAMEZAWA Hiroyuki | dc98df5 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 1490 | struct oom_wait_info *oom_wait_info; |
| 1491 | |
| 1492 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 1493 | oom_wait_memcg = oom_wait_info->memcg; |
KAMEZAWA Hiroyuki | dc98df5 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 1494 | |
Johannes Weiner | 2314b42 | 2014-12-10 15:44:33 -0800 | [diff] [blame] | 1495 | if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) && |
| 1496 | !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg)) |
KAMEZAWA Hiroyuki | dc98df5 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 1497 | return 0; |
KAMEZAWA Hiroyuki | dc98df5 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 1498 | return autoremove_wake_function(wait, mode, sync, arg); |
| 1499 | } |
| 1500 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1501 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 1502 | { |
Tejun Heo | c2b42d3 | 2015-06-24 16:58:23 -0700 | [diff] [blame] | 1503 | /* |
| 1504 | * For the following lockless ->under_oom test, the only required |
| 1505 | * guarantee is that it must see the state asserted by an OOM when |
| 1506 | * this function is called as a result of userland actions |
| 1507 | * triggered by the notification of the OOM. This is trivially |
| 1508 | * achieved by invoking mem_cgroup_mark_under_oom() before |
| 1509 | * triggering notification. |
| 1510 | */ |
| 1511 | if (memcg && memcg->under_oom) |
Tejun Heo | f4b90b70 | 2015-06-24 16:58:21 -0700 | [diff] [blame] | 1512 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 1513 | } |
| 1514 | |
Johannes Weiner | 3812c8c | 2013-09-12 15:13:44 -0700 | [diff] [blame] | 1515 | static void mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order) |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1516 | { |
Andrew Morton | d0db7af | 2016-06-08 15:33:47 -0700 | [diff] [blame] | 1517 | if (!current->memcg_may_oom) |
Johannes Weiner | 3812c8c | 2013-09-12 15:13:44 -0700 | [diff] [blame] | 1518 | return; |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1519 | /* |
Johannes Weiner | 4942642 | 2013-10-16 13:46:59 -0700 | [diff] [blame] | 1520 | * We are in the middle of the charge context here, so we |
| 1521 | * don't want to block when potentially sitting on a callstack |
| 1522 | * that holds all kinds of filesystem and mm locks. |
| 1523 | * |
| 1524 | * Also, the caller may handle a failed allocation gracefully |
| 1525 | * (like optional page cache readahead) and so an OOM killer |
| 1526 | * invocation might not even be necessary. |
| 1527 | * |
| 1528 | * That's why we don't do anything here except remember the |
| 1529 | * OOM context and then deal with it at the end of the page |
| 1530 | * fault when the stack is unwound, the locks are released, |
| 1531 | * and when we know whether the fault was overall successful. |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1532 | */ |
Johannes Weiner | 4942642 | 2013-10-16 13:46:59 -0700 | [diff] [blame] | 1533 | css_get(&memcg->css); |
Tejun Heo | 626ebc4 | 2015-11-05 18:46:09 -0800 | [diff] [blame] | 1534 | current->memcg_in_oom = memcg; |
| 1535 | current->memcg_oom_gfp_mask = mask; |
| 1536 | current->memcg_oom_order = order; |
Johannes Weiner | 4942642 | 2013-10-16 13:46:59 -0700 | [diff] [blame] | 1537 | } |
| 1538 | |
| 1539 | /** |
| 1540 | * mem_cgroup_oom_synchronize - complete memcg OOM handling |
| 1541 | * @handle: actually kill/wait or just clean up the OOM state |
| 1542 | * |
| 1543 | * This has to be called at the end of a page fault if the memcg OOM |
| 1544 | * handler was enabled. |
| 1545 | * |
| 1546 | * Memcg supports userspace OOM handling where failed allocations must |
| 1547 | * sleep on a waitqueue until the userspace task resolves the |
| 1548 | * situation. Sleeping directly in the charge context with all kinds |
| 1549 | * of locks held is not a good idea, instead we remember an OOM state |
| 1550 | * in the task and mem_cgroup_oom_synchronize() has to be called at |
| 1551 | * the end of the page fault to complete the OOM handling. |
| 1552 | * |
| 1553 | * Returns %true if an ongoing memcg OOM situation was detected and |
| 1554 | * completed, %false otherwise. |
| 1555 | */ |
| 1556 | bool mem_cgroup_oom_synchronize(bool handle) |
| 1557 | { |
Tejun Heo | 626ebc4 | 2015-11-05 18:46:09 -0800 | [diff] [blame] | 1558 | struct mem_cgroup *memcg = current->memcg_in_oom; |
Johannes Weiner | 4942642 | 2013-10-16 13:46:59 -0700 | [diff] [blame] | 1559 | struct oom_wait_info owait; |
| 1560 | bool locked; |
| 1561 | |
| 1562 | /* OOM is global, do not handle */ |
| 1563 | if (!memcg) |
| 1564 | return false; |
| 1565 | |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 1566 | if (!handle) |
Johannes Weiner | 4942642 | 2013-10-16 13:46:59 -0700 | [diff] [blame] | 1567 | goto cleanup; |
| 1568 | |
| 1569 | owait.memcg = memcg; |
| 1570 | owait.wait.flags = 0; |
| 1571 | owait.wait.func = memcg_oom_wake_function; |
| 1572 | owait.wait.private = current; |
Ingo Molnar | 2055da9 | 2017-06-20 12:06:46 +0200 | [diff] [blame] | 1573 | INIT_LIST_HEAD(&owait.wait.entry); |
Johannes Weiner | 4942642 | 2013-10-16 13:46:59 -0700 | [diff] [blame] | 1574 | |
| 1575 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
Johannes Weiner | fb2a6fc | 2013-09-12 15:13:43 -0700 | [diff] [blame] | 1576 | mem_cgroup_mark_under_oom(memcg); |
| 1577 | |
| 1578 | locked = mem_cgroup_oom_trylock(memcg); |
| 1579 | |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 1580 | if (locked) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1581 | mem_cgroup_oom_notify(memcg); |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1582 | |
Johannes Weiner | fb2a6fc | 2013-09-12 15:13:43 -0700 | [diff] [blame] | 1583 | if (locked && !memcg->oom_kill_disable) { |
| 1584 | mem_cgroup_unmark_under_oom(memcg); |
Johannes Weiner | 4942642 | 2013-10-16 13:46:59 -0700 | [diff] [blame] | 1585 | finish_wait(&memcg_oom_waitq, &owait.wait); |
Tejun Heo | 626ebc4 | 2015-11-05 18:46:09 -0800 | [diff] [blame] | 1586 | mem_cgroup_out_of_memory(memcg, current->memcg_oom_gfp_mask, |
| 1587 | current->memcg_oom_order); |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 1588 | } else { |
Johannes Weiner | 3812c8c | 2013-09-12 15:13:44 -0700 | [diff] [blame] | 1589 | schedule(); |
Johannes Weiner | 4942642 | 2013-10-16 13:46:59 -0700 | [diff] [blame] | 1590 | mem_cgroup_unmark_under_oom(memcg); |
| 1591 | finish_wait(&memcg_oom_waitq, &owait.wait); |
| 1592 | } |
| 1593 | |
| 1594 | if (locked) { |
Johannes Weiner | fb2a6fc | 2013-09-12 15:13:43 -0700 | [diff] [blame] | 1595 | mem_cgroup_oom_unlock(memcg); |
| 1596 | /* |
| 1597 | * There is no guarantee that an OOM-lock contender |
| 1598 | * sees the wakeups triggered by the OOM kill |
| 1599 | * uncharges. Wake any sleepers explicitely. |
| 1600 | */ |
| 1601 | memcg_oom_recover(memcg); |
| 1602 | } |
Johannes Weiner | 4942642 | 2013-10-16 13:46:59 -0700 | [diff] [blame] | 1603 | cleanup: |
Tejun Heo | 626ebc4 | 2015-11-05 18:46:09 -0800 | [diff] [blame] | 1604 | current->memcg_in_oom = NULL; |
Johannes Weiner | 3812c8c | 2013-09-12 15:13:44 -0700 | [diff] [blame] | 1605 | css_put(&memcg->css); |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1606 | return true; |
KAMEZAWA Hiroyuki | 0b7f569 | 2009-04-02 16:57:38 -0700 | [diff] [blame] | 1607 | } |
| 1608 | |
Johannes Weiner | d7365e7 | 2014-10-29 14:50:48 -0700 | [diff] [blame] | 1609 | /** |
Johannes Weiner | 81f8c3a | 2016-03-15 14:57:04 -0700 | [diff] [blame] | 1610 | * lock_page_memcg - lock a page->mem_cgroup binding |
| 1611 | * @page: the page |
KAMEZAWA Hiroyuki | 32047e2 | 2010-10-27 15:33:40 -0700 | [diff] [blame] | 1612 | * |
Johannes Weiner | 81f8c3a | 2016-03-15 14:57:04 -0700 | [diff] [blame] | 1613 | * This function protects unlocked LRU pages from being moved to |
| 1614 | * another cgroup and stabilizes their page->mem_cgroup binding. |
Balbir Singh | d69b042 | 2009-06-17 16:26:34 -0700 | [diff] [blame] | 1615 | */ |
Johannes Weiner | 62cccb8 | 2016-03-15 14:57:22 -0700 | [diff] [blame] | 1616 | void lock_page_memcg(struct page *page) |
KAMEZAWA Hiroyuki | 89c06bd | 2012-03-21 16:34:25 -0700 | [diff] [blame] | 1617 | { |
| 1618 | struct mem_cgroup *memcg; |
Johannes Weiner | 6de2261 | 2015-02-11 15:25:01 -0800 | [diff] [blame] | 1619 | unsigned long flags; |
KAMEZAWA Hiroyuki | 89c06bd | 2012-03-21 16:34:25 -0700 | [diff] [blame] | 1620 | |
Johannes Weiner | 6de2261 | 2015-02-11 15:25:01 -0800 | [diff] [blame] | 1621 | /* |
| 1622 | * The RCU lock is held throughout the transaction. The fast |
| 1623 | * path can get away without acquiring the memcg->move_lock |
| 1624 | * because page moving starts with an RCU grace period. |
Johannes Weiner | 6de2261 | 2015-02-11 15:25:01 -0800 | [diff] [blame] | 1625 | */ |
Johannes Weiner | d7365e7 | 2014-10-29 14:50:48 -0700 | [diff] [blame] | 1626 | rcu_read_lock(); |
| 1627 | |
| 1628 | if (mem_cgroup_disabled()) |
Johannes Weiner | 62cccb8 | 2016-03-15 14:57:22 -0700 | [diff] [blame] | 1629 | return; |
KAMEZAWA Hiroyuki | 89c06bd | 2012-03-21 16:34:25 -0700 | [diff] [blame] | 1630 | again: |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 1631 | memcg = page->mem_cgroup; |
Johannes Weiner | 2983331 | 2014-12-10 15:44:02 -0800 | [diff] [blame] | 1632 | if (unlikely(!memcg)) |
Johannes Weiner | 62cccb8 | 2016-03-15 14:57:22 -0700 | [diff] [blame] | 1633 | return; |
Johannes Weiner | d7365e7 | 2014-10-29 14:50:48 -0700 | [diff] [blame] | 1634 | |
Qiang Huang | bdcbb65 | 2014-06-04 16:08:21 -0700 | [diff] [blame] | 1635 | if (atomic_read(&memcg->moving_account) <= 0) |
Johannes Weiner | 62cccb8 | 2016-03-15 14:57:22 -0700 | [diff] [blame] | 1636 | return; |
KAMEZAWA Hiroyuki | 89c06bd | 2012-03-21 16:34:25 -0700 | [diff] [blame] | 1637 | |
Johannes Weiner | 6de2261 | 2015-02-11 15:25:01 -0800 | [diff] [blame] | 1638 | spin_lock_irqsave(&memcg->move_lock, flags); |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 1639 | if (memcg != page->mem_cgroup) { |
Johannes Weiner | 6de2261 | 2015-02-11 15:25:01 -0800 | [diff] [blame] | 1640 | spin_unlock_irqrestore(&memcg->move_lock, flags); |
KAMEZAWA Hiroyuki | 89c06bd | 2012-03-21 16:34:25 -0700 | [diff] [blame] | 1641 | goto again; |
| 1642 | } |
Johannes Weiner | 6de2261 | 2015-02-11 15:25:01 -0800 | [diff] [blame] | 1643 | |
| 1644 | /* |
| 1645 | * When charge migration first begins, we can have locked and |
| 1646 | * unlocked page stat updates happening concurrently. Track |
Johannes Weiner | 81f8c3a | 2016-03-15 14:57:04 -0700 | [diff] [blame] | 1647 | * the task who has the lock for unlock_page_memcg(). |
Johannes Weiner | 6de2261 | 2015-02-11 15:25:01 -0800 | [diff] [blame] | 1648 | */ |
| 1649 | memcg->move_lock_task = current; |
| 1650 | memcg->move_lock_flags = flags; |
Johannes Weiner | d7365e7 | 2014-10-29 14:50:48 -0700 | [diff] [blame] | 1651 | |
Johannes Weiner | 62cccb8 | 2016-03-15 14:57:22 -0700 | [diff] [blame] | 1652 | return; |
KAMEZAWA Hiroyuki | 89c06bd | 2012-03-21 16:34:25 -0700 | [diff] [blame] | 1653 | } |
Johannes Weiner | 81f8c3a | 2016-03-15 14:57:04 -0700 | [diff] [blame] | 1654 | EXPORT_SYMBOL(lock_page_memcg); |
KAMEZAWA Hiroyuki | 89c06bd | 2012-03-21 16:34:25 -0700 | [diff] [blame] | 1655 | |
Johannes Weiner | d7365e7 | 2014-10-29 14:50:48 -0700 | [diff] [blame] | 1656 | /** |
Johannes Weiner | 81f8c3a | 2016-03-15 14:57:04 -0700 | [diff] [blame] | 1657 | * unlock_page_memcg - unlock a page->mem_cgroup binding |
Johannes Weiner | 62cccb8 | 2016-03-15 14:57:22 -0700 | [diff] [blame] | 1658 | * @page: the page |
Johannes Weiner | d7365e7 | 2014-10-29 14:50:48 -0700 | [diff] [blame] | 1659 | */ |
Johannes Weiner | 62cccb8 | 2016-03-15 14:57:22 -0700 | [diff] [blame] | 1660 | void unlock_page_memcg(struct page *page) |
KAMEZAWA Hiroyuki | 89c06bd | 2012-03-21 16:34:25 -0700 | [diff] [blame] | 1661 | { |
Johannes Weiner | 62cccb8 | 2016-03-15 14:57:22 -0700 | [diff] [blame] | 1662 | struct mem_cgroup *memcg = page->mem_cgroup; |
| 1663 | |
Johannes Weiner | 6de2261 | 2015-02-11 15:25:01 -0800 | [diff] [blame] | 1664 | if (memcg && memcg->move_lock_task == current) { |
| 1665 | unsigned long flags = memcg->move_lock_flags; |
| 1666 | |
| 1667 | memcg->move_lock_task = NULL; |
| 1668 | memcg->move_lock_flags = 0; |
| 1669 | |
| 1670 | spin_unlock_irqrestore(&memcg->move_lock, flags); |
| 1671 | } |
KAMEZAWA Hiroyuki | 89c06bd | 2012-03-21 16:34:25 -0700 | [diff] [blame] | 1672 | |
Johannes Weiner | d7365e7 | 2014-10-29 14:50:48 -0700 | [diff] [blame] | 1673 | rcu_read_unlock(); |
KAMEZAWA Hiroyuki | 89c06bd | 2012-03-21 16:34:25 -0700 | [diff] [blame] | 1674 | } |
Johannes Weiner | 81f8c3a | 2016-03-15 14:57:04 -0700 | [diff] [blame] | 1675 | EXPORT_SYMBOL(unlock_page_memcg); |
KAMEZAWA Hiroyuki | 89c06bd | 2012-03-21 16:34:25 -0700 | [diff] [blame] | 1676 | |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 1677 | /* |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1678 | * size of first charge trial. "32" comes from vmscan.c's magic value. |
| 1679 | * TODO: maybe necessary to use big numbers in big irons. |
| 1680 | */ |
Johannes Weiner | 7ec99d6 | 2011-03-23 16:42:36 -0700 | [diff] [blame] | 1681 | #define CHARGE_BATCH 32U |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1682 | struct memcg_stock_pcp { |
| 1683 | struct mem_cgroup *cached; /* this never be root cgroup */ |
Johannes Weiner | 11c9ea4 | 2011-03-23 16:42:34 -0700 | [diff] [blame] | 1684 | unsigned int nr_pages; |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1685 | struct work_struct work; |
KAMEZAWA Hiroyuki | 26fe616 | 2011-06-15 15:08:45 -0700 | [diff] [blame] | 1686 | unsigned long flags; |
Kirill A. Shutemov | a0db00f | 2012-05-29 15:06:56 -0700 | [diff] [blame] | 1687 | #define FLUSHING_CACHED_CHARGE 0 |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1688 | }; |
| 1689 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); |
Michal Hocko | 9f50fad | 2011-08-09 11:56:26 +0200 | [diff] [blame] | 1690 | static DEFINE_MUTEX(percpu_charge_mutex); |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1691 | |
Suleiman Souhlal | a0956d5 | 2012-12-18 14:21:36 -0800 | [diff] [blame] | 1692 | /** |
| 1693 | * consume_stock: Try to consume stocked charge on this cpu. |
| 1694 | * @memcg: memcg to consume from. |
| 1695 | * @nr_pages: how many pages to charge. |
| 1696 | * |
| 1697 | * The charges will only happen if @memcg matches the current cpu's memcg |
| 1698 | * stock, and at least @nr_pages are available in that stock. Failure to |
| 1699 | * service an allocation will refill the stock. |
| 1700 | * |
| 1701 | * returns true if successful, false otherwise. |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1702 | */ |
Suleiman Souhlal | a0956d5 | 2012-12-18 14:21:36 -0800 | [diff] [blame] | 1703 | static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1704 | { |
| 1705 | struct memcg_stock_pcp *stock; |
Johannes Weiner | db2ba40c | 2016-09-19 14:44:36 -0700 | [diff] [blame] | 1706 | unsigned long flags; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1707 | bool ret = false; |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1708 | |
Suleiman Souhlal | a0956d5 | 2012-12-18 14:21:36 -0800 | [diff] [blame] | 1709 | if (nr_pages > CHARGE_BATCH) |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1710 | return ret; |
Suleiman Souhlal | a0956d5 | 2012-12-18 14:21:36 -0800 | [diff] [blame] | 1711 | |
Johannes Weiner | db2ba40c | 2016-09-19 14:44:36 -0700 | [diff] [blame] | 1712 | local_irq_save(flags); |
| 1713 | |
| 1714 | stock = this_cpu_ptr(&memcg_stock); |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1715 | if (memcg == stock->cached && stock->nr_pages >= nr_pages) { |
Suleiman Souhlal | a0956d5 | 2012-12-18 14:21:36 -0800 | [diff] [blame] | 1716 | stock->nr_pages -= nr_pages; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1717 | ret = true; |
| 1718 | } |
Johannes Weiner | db2ba40c | 2016-09-19 14:44:36 -0700 | [diff] [blame] | 1719 | |
| 1720 | local_irq_restore(flags); |
| 1721 | |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1722 | return ret; |
| 1723 | } |
| 1724 | |
| 1725 | /* |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1726 | * Returns stocks cached in percpu and reset cached information. |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1727 | */ |
| 1728 | static void drain_stock(struct memcg_stock_pcp *stock) |
| 1729 | { |
| 1730 | struct mem_cgroup *old = stock->cached; |
| 1731 | |
Johannes Weiner | 11c9ea4 | 2011-03-23 16:42:34 -0700 | [diff] [blame] | 1732 | if (stock->nr_pages) { |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1733 | page_counter_uncharge(&old->memory, stock->nr_pages); |
Johannes Weiner | 7941d21 | 2016-01-14 15:21:23 -0800 | [diff] [blame] | 1734 | if (do_memsw_account()) |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1735 | page_counter_uncharge(&old->memsw, stock->nr_pages); |
Johannes Weiner | e8ea14c | 2014-12-10 15:42:42 -0800 | [diff] [blame] | 1736 | css_put_many(&old->css, stock->nr_pages); |
Johannes Weiner | 11c9ea4 | 2011-03-23 16:42:34 -0700 | [diff] [blame] | 1737 | stock->nr_pages = 0; |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1738 | } |
| 1739 | stock->cached = NULL; |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1740 | } |
| 1741 | |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1742 | static void drain_local_stock(struct work_struct *dummy) |
| 1743 | { |
Johannes Weiner | db2ba40c | 2016-09-19 14:44:36 -0700 | [diff] [blame] | 1744 | struct memcg_stock_pcp *stock; |
| 1745 | unsigned long flags; |
| 1746 | |
| 1747 | local_irq_save(flags); |
| 1748 | |
| 1749 | stock = this_cpu_ptr(&memcg_stock); |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1750 | drain_stock(stock); |
KAMEZAWA Hiroyuki | 26fe616 | 2011-06-15 15:08:45 -0700 | [diff] [blame] | 1751 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
Johannes Weiner | db2ba40c | 2016-09-19 14:44:36 -0700 | [diff] [blame] | 1752 | |
| 1753 | local_irq_restore(flags); |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1754 | } |
| 1755 | |
| 1756 | /* |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1757 | * Cache charges(val) to local per_cpu area. |
Greg Thelen | 320cc51 | 2010-03-15 15:27:28 +0100 | [diff] [blame] | 1758 | * This will be consumed by consume_stock() function, later. |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1759 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1760 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1761 | { |
Johannes Weiner | db2ba40c | 2016-09-19 14:44:36 -0700 | [diff] [blame] | 1762 | struct memcg_stock_pcp *stock; |
| 1763 | unsigned long flags; |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1764 | |
Johannes Weiner | db2ba40c | 2016-09-19 14:44:36 -0700 | [diff] [blame] | 1765 | local_irq_save(flags); |
| 1766 | |
| 1767 | stock = this_cpu_ptr(&memcg_stock); |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1768 | if (stock->cached != memcg) { /* reset if necessary */ |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1769 | drain_stock(stock); |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1770 | stock->cached = memcg; |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1771 | } |
Johannes Weiner | 11c9ea4 | 2011-03-23 16:42:34 -0700 | [diff] [blame] | 1772 | stock->nr_pages += nr_pages; |
Johannes Weiner | db2ba40c | 2016-09-19 14:44:36 -0700 | [diff] [blame] | 1773 | |
| 1774 | local_irq_restore(flags); |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1775 | } |
| 1776 | |
| 1777 | /* |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1778 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
Johannes Weiner | 6d3d6aa | 2014-12-10 15:42:50 -0800 | [diff] [blame] | 1779 | * of the hierarchy under it. |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1780 | */ |
Johannes Weiner | 6d3d6aa | 2014-12-10 15:42:50 -0800 | [diff] [blame] | 1781 | static void drain_all_stock(struct mem_cgroup *root_memcg) |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1782 | { |
KAMEZAWA Hiroyuki | 26fe616 | 2011-06-15 15:08:45 -0700 | [diff] [blame] | 1783 | int cpu, curcpu; |
Michal Hocko | d38144b | 2011-07-26 16:08:28 -0700 | [diff] [blame] | 1784 | |
Johannes Weiner | 6d3d6aa | 2014-12-10 15:42:50 -0800 | [diff] [blame] | 1785 | /* If someone's already draining, avoid adding running more workers. */ |
| 1786 | if (!mutex_trylock(&percpu_charge_mutex)) |
| 1787 | return; |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1788 | /* Notify other cpus that system-wide "drain" is running */ |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1789 | get_online_cpus(); |
Johannes Weiner | 5af12d0 | 2011-08-25 15:59:07 -0700 | [diff] [blame] | 1790 | curcpu = get_cpu(); |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1791 | for_each_online_cpu(cpu) { |
| 1792 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1793 | struct mem_cgroup *memcg; |
KAMEZAWA Hiroyuki | 26fe616 | 2011-06-15 15:08:45 -0700 | [diff] [blame] | 1794 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1795 | memcg = stock->cached; |
| 1796 | if (!memcg || !stock->nr_pages) |
KAMEZAWA Hiroyuki | 26fe616 | 2011-06-15 15:08:45 -0700 | [diff] [blame] | 1797 | continue; |
Johannes Weiner | 2314b42 | 2014-12-10 15:44:33 -0800 | [diff] [blame] | 1798 | if (!mem_cgroup_is_descendant(memcg, root_memcg)) |
Michal Hocko | 3e92041 | 2011-07-26 16:08:29 -0700 | [diff] [blame] | 1799 | continue; |
Michal Hocko | d1a05b6 | 2011-07-26 16:08:27 -0700 | [diff] [blame] | 1800 | if (!test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { |
| 1801 | if (cpu == curcpu) |
| 1802 | drain_local_stock(&stock->work); |
| 1803 | else |
| 1804 | schedule_work_on(cpu, &stock->work); |
| 1805 | } |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1806 | } |
Johannes Weiner | 5af12d0 | 2011-08-25 15:59:07 -0700 | [diff] [blame] | 1807 | put_cpu(); |
Andrew Morton | f894ffa | 2013-09-12 15:13:35 -0700 | [diff] [blame] | 1808 | put_online_cpus(); |
Michal Hocko | 9f50fad | 2011-08-09 11:56:26 +0200 | [diff] [blame] | 1809 | mutex_unlock(&percpu_charge_mutex); |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1810 | } |
| 1811 | |
Sebastian Andrzej Siewior | 308167f | 2016-11-03 15:49:59 +0100 | [diff] [blame] | 1812 | static int memcg_hotplug_cpu_dead(unsigned int cpu) |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1813 | { |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1814 | struct memcg_stock_pcp *stock; |
| 1815 | |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1816 | stock = &per_cpu(memcg_stock, cpu); |
| 1817 | drain_stock(stock); |
Sebastian Andrzej Siewior | 308167f | 2016-11-03 15:49:59 +0100 | [diff] [blame] | 1818 | return 0; |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1819 | } |
| 1820 | |
Johannes Weiner | f7e1cb6 | 2016-01-14 15:21:29 -0800 | [diff] [blame] | 1821 | static void reclaim_high(struct mem_cgroup *memcg, |
| 1822 | unsigned int nr_pages, |
| 1823 | gfp_t gfp_mask) |
| 1824 | { |
| 1825 | do { |
| 1826 | if (page_counter_read(&memcg->memory) <= memcg->high) |
| 1827 | continue; |
Johannes Weiner | 31176c7 | 2017-05-03 14:55:07 -0700 | [diff] [blame] | 1828 | mem_cgroup_event(memcg, MEMCG_HIGH); |
Johannes Weiner | f7e1cb6 | 2016-01-14 15:21:29 -0800 | [diff] [blame] | 1829 | try_to_free_mem_cgroup_pages(memcg, nr_pages, gfp_mask, true); |
| 1830 | } while ((memcg = parent_mem_cgroup(memcg))); |
| 1831 | } |
| 1832 | |
| 1833 | static void high_work_func(struct work_struct *work) |
| 1834 | { |
| 1835 | struct mem_cgroup *memcg; |
| 1836 | |
| 1837 | memcg = container_of(work, struct mem_cgroup, high_work); |
| 1838 | reclaim_high(memcg, CHARGE_BATCH, GFP_KERNEL); |
| 1839 | } |
| 1840 | |
Tejun Heo | b23afb9 | 2015-11-05 18:46:11 -0800 | [diff] [blame] | 1841 | /* |
| 1842 | * Scheduled by try_charge() to be executed from the userland return path |
| 1843 | * and reclaims memory over the high limit. |
| 1844 | */ |
| 1845 | void mem_cgroup_handle_over_high(void) |
| 1846 | { |
| 1847 | unsigned int nr_pages = current->memcg_nr_pages_over_high; |
Johannes Weiner | f7e1cb6 | 2016-01-14 15:21:29 -0800 | [diff] [blame] | 1848 | struct mem_cgroup *memcg; |
Tejun Heo | b23afb9 | 2015-11-05 18:46:11 -0800 | [diff] [blame] | 1849 | |
| 1850 | if (likely(!nr_pages)) |
| 1851 | return; |
| 1852 | |
Johannes Weiner | f7e1cb6 | 2016-01-14 15:21:29 -0800 | [diff] [blame] | 1853 | memcg = get_mem_cgroup_from_mm(current->mm); |
| 1854 | reclaim_high(memcg, nr_pages, GFP_KERNEL); |
Tejun Heo | b23afb9 | 2015-11-05 18:46:11 -0800 | [diff] [blame] | 1855 | css_put(&memcg->css); |
| 1856 | current->memcg_nr_pages_over_high = 0; |
| 1857 | } |
| 1858 | |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 1859 | static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
| 1860 | unsigned int nr_pages) |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 1861 | { |
Johannes Weiner | 7ec99d6 | 2011-03-23 16:42:36 -0700 | [diff] [blame] | 1862 | unsigned int batch = max(CHARGE_BATCH, nr_pages); |
Johannes Weiner | 9b13061 | 2014-08-06 16:05:51 -0700 | [diff] [blame] | 1863 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; |
Johannes Weiner | 6539cc0 | 2014-08-06 16:05:42 -0700 | [diff] [blame] | 1864 | struct mem_cgroup *mem_over_limit; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1865 | struct page_counter *counter; |
Johannes Weiner | 6539cc0 | 2014-08-06 16:05:42 -0700 | [diff] [blame] | 1866 | unsigned long nr_reclaimed; |
Johannes Weiner | b70a2a2 | 2014-10-09 15:28:56 -0700 | [diff] [blame] | 1867 | bool may_swap = true; |
| 1868 | bool drained = false; |
KAMEZAWA Hiroyuki | a636b32 | 2009-01-07 18:08:08 -0800 | [diff] [blame] | 1869 | |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 1870 | if (mem_cgroup_is_root(memcg)) |
Tejun Heo | 10d53c7 | 2015-11-05 18:46:17 -0800 | [diff] [blame] | 1871 | return 0; |
Johannes Weiner | 6539cc0 | 2014-08-06 16:05:42 -0700 | [diff] [blame] | 1872 | retry: |
Michal Hocko | b6b6cc7 | 2014-04-07 15:37:44 -0700 | [diff] [blame] | 1873 | if (consume_stock(memcg, nr_pages)) |
Tejun Heo | 10d53c7 | 2015-11-05 18:46:17 -0800 | [diff] [blame] | 1874 | return 0; |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 1875 | |
Johannes Weiner | 7941d21 | 2016-01-14 15:21:23 -0800 | [diff] [blame] | 1876 | if (!do_memsw_account() || |
Johannes Weiner | 6071ca5 | 2015-11-05 18:50:26 -0800 | [diff] [blame] | 1877 | page_counter_try_charge(&memcg->memsw, batch, &counter)) { |
| 1878 | if (page_counter_try_charge(&memcg->memory, batch, &counter)) |
Johannes Weiner | 6539cc0 | 2014-08-06 16:05:42 -0700 | [diff] [blame] | 1879 | goto done_restock; |
Johannes Weiner | 7941d21 | 2016-01-14 15:21:23 -0800 | [diff] [blame] | 1880 | if (do_memsw_account()) |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1881 | page_counter_uncharge(&memcg->memsw, batch); |
| 1882 | mem_over_limit = mem_cgroup_from_counter(counter, memory); |
Johannes Weiner | 3fbe724 | 2014-10-09 15:28:54 -0700 | [diff] [blame] | 1883 | } else { |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1884 | mem_over_limit = mem_cgroup_from_counter(counter, memsw); |
Johannes Weiner | b70a2a2 | 2014-10-09 15:28:56 -0700 | [diff] [blame] | 1885 | may_swap = false; |
Johannes Weiner | 3fbe724 | 2014-10-09 15:28:54 -0700 | [diff] [blame] | 1886 | } |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 1887 | |
Johannes Weiner | 6539cc0 | 2014-08-06 16:05:42 -0700 | [diff] [blame] | 1888 | if (batch > nr_pages) { |
| 1889 | batch = nr_pages; |
| 1890 | goto retry; |
| 1891 | } |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 1892 | |
Johannes Weiner | 06b078f | 2014-08-06 16:05:44 -0700 | [diff] [blame] | 1893 | /* |
| 1894 | * Unlike in global OOM situations, memcg is not in a physical |
| 1895 | * memory shortage. Allow dying and OOM-killed tasks to |
| 1896 | * bypass the last charges so that they can exit quickly and |
| 1897 | * free their memory. |
| 1898 | */ |
| 1899 | if (unlikely(test_thread_flag(TIF_MEMDIE) || |
| 1900 | fatal_signal_pending(current) || |
| 1901 | current->flags & PF_EXITING)) |
Tejun Heo | 10d53c7 | 2015-11-05 18:46:17 -0800 | [diff] [blame] | 1902 | goto force; |
Johannes Weiner | 06b078f | 2014-08-06 16:05:44 -0700 | [diff] [blame] | 1903 | |
Johannes Weiner | 89a2848 | 2016-10-27 17:46:56 -0700 | [diff] [blame] | 1904 | /* |
| 1905 | * Prevent unbounded recursion when reclaim operations need to |
| 1906 | * allocate memory. This might exceed the limits temporarily, |
| 1907 | * but we prefer facilitating memory reclaim and getting back |
| 1908 | * under the limit over triggering OOM kills in these cases. |
| 1909 | */ |
| 1910 | if (unlikely(current->flags & PF_MEMALLOC)) |
| 1911 | goto force; |
| 1912 | |
Johannes Weiner | 06b078f | 2014-08-06 16:05:44 -0700 | [diff] [blame] | 1913 | if (unlikely(task_in_memcg_oom(current))) |
| 1914 | goto nomem; |
| 1915 | |
Mel Gorman | d0164ad | 2015-11-06 16:28:21 -0800 | [diff] [blame] | 1916 | if (!gfpflags_allow_blocking(gfp_mask)) |
Johannes Weiner | 6539cc0 | 2014-08-06 16:05:42 -0700 | [diff] [blame] | 1917 | goto nomem; |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 1918 | |
Johannes Weiner | 31176c7 | 2017-05-03 14:55:07 -0700 | [diff] [blame] | 1919 | mem_cgroup_event(mem_over_limit, MEMCG_MAX); |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 1920 | |
Johannes Weiner | b70a2a2 | 2014-10-09 15:28:56 -0700 | [diff] [blame] | 1921 | nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages, |
| 1922 | gfp_mask, may_swap); |
Johannes Weiner | 6539cc0 | 2014-08-06 16:05:42 -0700 | [diff] [blame] | 1923 | |
Johannes Weiner | 61e02c7 | 2014-08-06 16:08:16 -0700 | [diff] [blame] | 1924 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
Johannes Weiner | 6539cc0 | 2014-08-06 16:05:42 -0700 | [diff] [blame] | 1925 | goto retry; |
Johannes Weiner | 28c34c2 | 2014-08-06 16:05:47 -0700 | [diff] [blame] | 1926 | |
Johannes Weiner | b70a2a2 | 2014-10-09 15:28:56 -0700 | [diff] [blame] | 1927 | if (!drained) { |
Johannes Weiner | 6d3d6aa | 2014-12-10 15:42:50 -0800 | [diff] [blame] | 1928 | drain_all_stock(mem_over_limit); |
Johannes Weiner | b70a2a2 | 2014-10-09 15:28:56 -0700 | [diff] [blame] | 1929 | drained = true; |
| 1930 | goto retry; |
| 1931 | } |
| 1932 | |
Johannes Weiner | 28c34c2 | 2014-08-06 16:05:47 -0700 | [diff] [blame] | 1933 | if (gfp_mask & __GFP_NORETRY) |
| 1934 | goto nomem; |
Johannes Weiner | 6539cc0 | 2014-08-06 16:05:42 -0700 | [diff] [blame] | 1935 | /* |
| 1936 | * Even though the limit is exceeded at this point, reclaim |
| 1937 | * may have been able to free some pages. Retry the charge |
| 1938 | * before killing the task. |
| 1939 | * |
| 1940 | * Only for regular pages, though: huge pages are rather |
| 1941 | * unlikely to succeed so close to the limit, and we fall back |
| 1942 | * to regular pages anyway in case of failure. |
| 1943 | */ |
Johannes Weiner | 61e02c7 | 2014-08-06 16:08:16 -0700 | [diff] [blame] | 1944 | if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER)) |
Johannes Weiner | 6539cc0 | 2014-08-06 16:05:42 -0700 | [diff] [blame] | 1945 | goto retry; |
| 1946 | /* |
| 1947 | * At task move, charge accounts can be doubly counted. So, it's |
| 1948 | * better to wait until the end of task_move if something is going on. |
| 1949 | */ |
| 1950 | if (mem_cgroup_wait_acct_move(mem_over_limit)) |
| 1951 | goto retry; |
| 1952 | |
Johannes Weiner | 9b13061 | 2014-08-06 16:05:51 -0700 | [diff] [blame] | 1953 | if (nr_retries--) |
| 1954 | goto retry; |
| 1955 | |
Johannes Weiner | 06b078f | 2014-08-06 16:05:44 -0700 | [diff] [blame] | 1956 | if (gfp_mask & __GFP_NOFAIL) |
Tejun Heo | 10d53c7 | 2015-11-05 18:46:17 -0800 | [diff] [blame] | 1957 | goto force; |
Johannes Weiner | 06b078f | 2014-08-06 16:05:44 -0700 | [diff] [blame] | 1958 | |
Johannes Weiner | 6539cc0 | 2014-08-06 16:05:42 -0700 | [diff] [blame] | 1959 | if (fatal_signal_pending(current)) |
Tejun Heo | 10d53c7 | 2015-11-05 18:46:17 -0800 | [diff] [blame] | 1960 | goto force; |
Johannes Weiner | 6539cc0 | 2014-08-06 16:05:42 -0700 | [diff] [blame] | 1961 | |
Johannes Weiner | 31176c7 | 2017-05-03 14:55:07 -0700 | [diff] [blame] | 1962 | mem_cgroup_event(mem_over_limit, MEMCG_OOM); |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 1963 | |
Jerome Marchand | 3608de0 | 2015-11-05 18:47:29 -0800 | [diff] [blame] | 1964 | mem_cgroup_oom(mem_over_limit, gfp_mask, |
| 1965 | get_order(nr_pages * PAGE_SIZE)); |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 1966 | nomem: |
Johannes Weiner | 6d1fdc4 | 2014-04-07 15:37:45 -0700 | [diff] [blame] | 1967 | if (!(gfp_mask & __GFP_NOFAIL)) |
Johannes Weiner | 3168ecb | 2013-10-31 16:34:13 -0700 | [diff] [blame] | 1968 | return -ENOMEM; |
Tejun Heo | 10d53c7 | 2015-11-05 18:46:17 -0800 | [diff] [blame] | 1969 | force: |
| 1970 | /* |
| 1971 | * The allocation either can't fail or will lead to more memory |
| 1972 | * being freed very soon. Allow memory usage go over the limit |
| 1973 | * temporarily by force charging it. |
| 1974 | */ |
| 1975 | page_counter_charge(&memcg->memory, nr_pages); |
Johannes Weiner | 7941d21 | 2016-01-14 15:21:23 -0800 | [diff] [blame] | 1976 | if (do_memsw_account()) |
Tejun Heo | 10d53c7 | 2015-11-05 18:46:17 -0800 | [diff] [blame] | 1977 | page_counter_charge(&memcg->memsw, nr_pages); |
| 1978 | css_get_many(&memcg->css, nr_pages); |
| 1979 | |
| 1980 | return 0; |
Johannes Weiner | 6539cc0 | 2014-08-06 16:05:42 -0700 | [diff] [blame] | 1981 | |
| 1982 | done_restock: |
Johannes Weiner | e8ea14c | 2014-12-10 15:42:42 -0800 | [diff] [blame] | 1983 | css_get_many(&memcg->css, batch); |
Johannes Weiner | 6539cc0 | 2014-08-06 16:05:42 -0700 | [diff] [blame] | 1984 | if (batch > nr_pages) |
| 1985 | refill_stock(memcg, batch - nr_pages); |
Tejun Heo | b23afb9 | 2015-11-05 18:46:11 -0800 | [diff] [blame] | 1986 | |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 1987 | /* |
Tejun Heo | b23afb9 | 2015-11-05 18:46:11 -0800 | [diff] [blame] | 1988 | * If the hierarchy is above the normal consumption range, schedule |
| 1989 | * reclaim on returning to userland. We can perform reclaim here |
Mel Gorman | 71baba4 | 2015-11-06 16:28:28 -0800 | [diff] [blame] | 1990 | * if __GFP_RECLAIM but let's always punt for simplicity and so that |
Tejun Heo | b23afb9 | 2015-11-05 18:46:11 -0800 | [diff] [blame] | 1991 | * GFP_KERNEL can consistently be used during reclaim. @memcg is |
| 1992 | * not recorded as it most likely matches current's and won't |
| 1993 | * change in the meantime. As high limit is checked again before |
| 1994 | * reclaim, the cost of mismatch is negligible. |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 1995 | */ |
| 1996 | do { |
Tejun Heo | b23afb9 | 2015-11-05 18:46:11 -0800 | [diff] [blame] | 1997 | if (page_counter_read(&memcg->memory) > memcg->high) { |
Johannes Weiner | f7e1cb6 | 2016-01-14 15:21:29 -0800 | [diff] [blame] | 1998 | /* Don't bother a random interrupted task */ |
| 1999 | if (in_interrupt()) { |
| 2000 | schedule_work(&memcg->high_work); |
| 2001 | break; |
| 2002 | } |
Vladimir Davydov | 9516a18 | 2015-12-11 13:40:24 -0800 | [diff] [blame] | 2003 | current->memcg_nr_pages_over_high += batch; |
Tejun Heo | b23afb9 | 2015-11-05 18:46:11 -0800 | [diff] [blame] | 2004 | set_notify_resume(current); |
| 2005 | break; |
| 2006 | } |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 2007 | } while ((memcg = parent_mem_cgroup(memcg))); |
Tejun Heo | 10d53c7 | 2015-11-05 18:46:17 -0800 | [diff] [blame] | 2008 | |
| 2009 | return 0; |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2010 | } |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 2011 | |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 2012 | static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages) |
Daisuke Nishimura | a3032a2 | 2009-12-15 16:47:10 -0800 | [diff] [blame] | 2013 | { |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 2014 | if (mem_cgroup_is_root(memcg)) |
| 2015 | return; |
| 2016 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2017 | page_counter_uncharge(&memcg->memory, nr_pages); |
Johannes Weiner | 7941d21 | 2016-01-14 15:21:23 -0800 | [diff] [blame] | 2018 | if (do_memsw_account()) |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2019 | page_counter_uncharge(&memcg->memsw, nr_pages); |
Daisuke Nishimura | a3032a2 | 2009-12-15 16:47:10 -0800 | [diff] [blame] | 2020 | |
Johannes Weiner | e8ea14c | 2014-12-10 15:42:42 -0800 | [diff] [blame] | 2021 | css_put_many(&memcg->css, nr_pages); |
KAMEZAWA Hiroyuki | d01dd17 | 2012-05-29 15:07:03 -0700 | [diff] [blame] | 2022 | } |
| 2023 | |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 2024 | static void lock_page_lru(struct page *page, int *isolated) |
| 2025 | { |
| 2026 | struct zone *zone = page_zone(page); |
| 2027 | |
Mel Gorman | a52633d | 2016-07-28 15:45:28 -0700 | [diff] [blame] | 2028 | spin_lock_irq(zone_lru_lock(zone)); |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 2029 | if (PageLRU(page)) { |
| 2030 | struct lruvec *lruvec; |
| 2031 | |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 2032 | lruvec = mem_cgroup_page_lruvec(page, zone->zone_pgdat); |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 2033 | ClearPageLRU(page); |
| 2034 | del_page_from_lru_list(page, lruvec, page_lru(page)); |
| 2035 | *isolated = 1; |
| 2036 | } else |
| 2037 | *isolated = 0; |
| 2038 | } |
| 2039 | |
| 2040 | static void unlock_page_lru(struct page *page, int isolated) |
| 2041 | { |
| 2042 | struct zone *zone = page_zone(page); |
| 2043 | |
| 2044 | if (isolated) { |
| 2045 | struct lruvec *lruvec; |
| 2046 | |
Mel Gorman | 599d0c9 | 2016-07-28 15:45:31 -0700 | [diff] [blame] | 2047 | lruvec = mem_cgroup_page_lruvec(page, zone->zone_pgdat); |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 2048 | VM_BUG_ON_PAGE(PageLRU(page), page); |
| 2049 | SetPageLRU(page); |
| 2050 | add_page_to_lru_list(page, lruvec, page_lru(page)); |
| 2051 | } |
Mel Gorman | a52633d | 2016-07-28 15:45:28 -0700 | [diff] [blame] | 2052 | spin_unlock_irq(zone_lru_lock(zone)); |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 2053 | } |
| 2054 | |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 2055 | static void commit_charge(struct page *page, struct mem_cgroup *memcg, |
Johannes Weiner | 6abb5a8 | 2014-08-08 14:19:33 -0700 | [diff] [blame] | 2056 | bool lrucare) |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2057 | { |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 2058 | int isolated; |
Hugh Dickins | 9ce70c0 | 2012-03-05 14:59:16 -0800 | [diff] [blame] | 2059 | |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 2060 | VM_BUG_ON_PAGE(page->mem_cgroup, page); |
Hugh Dickins | 9ce70c0 | 2012-03-05 14:59:16 -0800 | [diff] [blame] | 2061 | |
| 2062 | /* |
| 2063 | * In some cases, SwapCache and FUSE(splice_buf->radixtree), the page |
| 2064 | * may already be on some other mem_cgroup's LRU. Take care of it. |
| 2065 | */ |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 2066 | if (lrucare) |
| 2067 | lock_page_lru(page, &isolated); |
Hugh Dickins | 9ce70c0 | 2012-03-05 14:59:16 -0800 | [diff] [blame] | 2068 | |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 2069 | /* |
| 2070 | * Nobody should be changing or seriously looking at |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 2071 | * page->mem_cgroup at this point: |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 2072 | * |
| 2073 | * - the page is uncharged |
| 2074 | * |
| 2075 | * - the page is off-LRU |
| 2076 | * |
| 2077 | * - an anonymous fault has exclusive page access, except for |
| 2078 | * a locked page table |
| 2079 | * |
| 2080 | * - a page cache insertion, a swapin fault, or a migration |
| 2081 | * have the page locked |
| 2082 | */ |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 2083 | page->mem_cgroup = memcg; |
Hugh Dickins | 3be9127 | 2008-02-07 00:14:19 -0800 | [diff] [blame] | 2084 | |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 2085 | if (lrucare) |
| 2086 | unlock_page_lru(page, isolated); |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 2087 | } |
| 2088 | |
Johannes Weiner | 127424c | 2016-01-20 15:02:32 -0800 | [diff] [blame] | 2089 | #ifndef CONFIG_SLOB |
Vladimir Davydov | f3bb304 | 2014-10-09 15:28:45 -0700 | [diff] [blame] | 2090 | static int memcg_alloc_cache_id(void) |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 2091 | { |
Vladimir Davydov | f3bb304 | 2014-10-09 15:28:45 -0700 | [diff] [blame] | 2092 | int id, size; |
| 2093 | int err; |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 2094 | |
Vladimir Davydov | dbcf73e | 2015-02-12 14:58:57 -0800 | [diff] [blame] | 2095 | id = ida_simple_get(&memcg_cache_ida, |
Vladimir Davydov | f3bb304 | 2014-10-09 15:28:45 -0700 | [diff] [blame] | 2096 | 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL); |
| 2097 | if (id < 0) |
| 2098 | return id; |
| 2099 | |
Vladimir Davydov | dbcf73e | 2015-02-12 14:58:57 -0800 | [diff] [blame] | 2100 | if (id < memcg_nr_cache_ids) |
Vladimir Davydov | f3bb304 | 2014-10-09 15:28:45 -0700 | [diff] [blame] | 2101 | return id; |
| 2102 | |
| 2103 | /* |
| 2104 | * There's no space for the new id in memcg_caches arrays, |
| 2105 | * so we have to grow them. |
| 2106 | */ |
Vladimir Davydov | 05257a1 | 2015-02-12 14:59:01 -0800 | [diff] [blame] | 2107 | down_write(&memcg_cache_ids_sem); |
Vladimir Davydov | f3bb304 | 2014-10-09 15:28:45 -0700 | [diff] [blame] | 2108 | |
| 2109 | size = 2 * (id + 1); |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 2110 | if (size < MEMCG_CACHES_MIN_SIZE) |
| 2111 | size = MEMCG_CACHES_MIN_SIZE; |
| 2112 | else if (size > MEMCG_CACHES_MAX_SIZE) |
| 2113 | size = MEMCG_CACHES_MAX_SIZE; |
| 2114 | |
Vladimir Davydov | f3bb304 | 2014-10-09 15:28:45 -0700 | [diff] [blame] | 2115 | err = memcg_update_all_caches(size); |
Vladimir Davydov | 05257a1 | 2015-02-12 14:59:01 -0800 | [diff] [blame] | 2116 | if (!err) |
Vladimir Davydov | 60d3fd3 | 2015-02-12 14:59:10 -0800 | [diff] [blame] | 2117 | err = memcg_update_all_list_lrus(size); |
| 2118 | if (!err) |
Vladimir Davydov | 05257a1 | 2015-02-12 14:59:01 -0800 | [diff] [blame] | 2119 | memcg_nr_cache_ids = size; |
| 2120 | |
| 2121 | up_write(&memcg_cache_ids_sem); |
| 2122 | |
Vladimir Davydov | f3bb304 | 2014-10-09 15:28:45 -0700 | [diff] [blame] | 2123 | if (err) { |
Vladimir Davydov | dbcf73e | 2015-02-12 14:58:57 -0800 | [diff] [blame] | 2124 | ida_simple_remove(&memcg_cache_ida, id); |
Vladimir Davydov | f3bb304 | 2014-10-09 15:28:45 -0700 | [diff] [blame] | 2125 | return err; |
| 2126 | } |
| 2127 | return id; |
| 2128 | } |
| 2129 | |
| 2130 | static void memcg_free_cache_id(int id) |
| 2131 | { |
Vladimir Davydov | dbcf73e | 2015-02-12 14:58:57 -0800 | [diff] [blame] | 2132 | ida_simple_remove(&memcg_cache_ida, id); |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 2133 | } |
| 2134 | |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 2135 | struct memcg_kmem_cache_create_work { |
Vladimir Davydov | 5722d09 | 2014-04-07 15:39:24 -0700 | [diff] [blame] | 2136 | struct mem_cgroup *memcg; |
| 2137 | struct kmem_cache *cachep; |
| 2138 | struct work_struct work; |
| 2139 | }; |
| 2140 | |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 2141 | static void memcg_kmem_cache_create_func(struct work_struct *w) |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 2142 | { |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 2143 | struct memcg_kmem_cache_create_work *cw = |
| 2144 | container_of(w, struct memcg_kmem_cache_create_work, work); |
Vladimir Davydov | 5722d09 | 2014-04-07 15:39:24 -0700 | [diff] [blame] | 2145 | struct mem_cgroup *memcg = cw->memcg; |
| 2146 | struct kmem_cache *cachep = cw->cachep; |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 2147 | |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 2148 | memcg_create_kmem_cache(memcg, cachep); |
Vladimir Davydov | bd67314 | 2014-06-04 16:07:40 -0700 | [diff] [blame] | 2149 | |
Vladimir Davydov | 5722d09 | 2014-04-07 15:39:24 -0700 | [diff] [blame] | 2150 | css_put(&memcg->css); |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 2151 | kfree(cw); |
| 2152 | } |
| 2153 | |
| 2154 | /* |
| 2155 | * Enqueue the creation of a per-memcg kmem_cache. |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 2156 | */ |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 2157 | static void __memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg, |
| 2158 | struct kmem_cache *cachep) |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 2159 | { |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 2160 | struct memcg_kmem_cache_create_work *cw; |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 2161 | |
Vladimir Davydov | 776ed0f | 2014-06-04 16:10:02 -0700 | [diff] [blame] | 2162 | cw = kmalloc(sizeof(*cw), GFP_NOWAIT); |
Vladimir Davydov | 8135be5 | 2014-12-12 16:56:38 -0800 | [diff] [blame] | 2163 | if (!cw) |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 2164 | return; |
Vladimir Davydov | 8135be5 | 2014-12-12 16:56:38 -0800 | [diff] [blame] | 2165 | |
| 2166 | css_get(&memcg->css); |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 2167 | |
| 2168 | cw->memcg = memcg; |
| 2169 | cw->cachep = cachep; |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 2170 | INIT_WORK(&cw->work, memcg_kmem_cache_create_func); |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 2171 | |
Tejun Heo | 17cc4df | 2017-02-22 15:41:36 -0800 | [diff] [blame] | 2172 | queue_work(memcg_kmem_cache_wq, &cw->work); |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 2173 | } |
| 2174 | |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 2175 | static void memcg_schedule_kmem_cache_create(struct mem_cgroup *memcg, |
| 2176 | struct kmem_cache *cachep) |
Glauber Costa | 0e9d92f | 2012-12-18 14:22:42 -0800 | [diff] [blame] | 2177 | { |
| 2178 | /* |
| 2179 | * We need to stop accounting when we kmalloc, because if the |
| 2180 | * corresponding kmalloc cache is not yet created, the first allocation |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 2181 | * in __memcg_schedule_kmem_cache_create will recurse. |
Glauber Costa | 0e9d92f | 2012-12-18 14:22:42 -0800 | [diff] [blame] | 2182 | * |
| 2183 | * However, it is better to enclose the whole function. Depending on |
| 2184 | * the debugging options enabled, INIT_WORK(), for instance, can |
| 2185 | * trigger an allocation. This too, will make us recurse. Because at |
| 2186 | * this point we can't allow ourselves back into memcg_kmem_get_cache, |
| 2187 | * the safest choice is to do it like this, wrapping the whole function. |
| 2188 | */ |
Vladimir Davydov | 6f185c2 | 2014-12-12 16:55:15 -0800 | [diff] [blame] | 2189 | current->memcg_kmem_skip_account = 1; |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 2190 | __memcg_schedule_kmem_cache_create(memcg, cachep); |
Vladimir Davydov | 6f185c2 | 2014-12-12 16:55:15 -0800 | [diff] [blame] | 2191 | current->memcg_kmem_skip_account = 0; |
Glauber Costa | 0e9d92f | 2012-12-18 14:22:42 -0800 | [diff] [blame] | 2192 | } |
Vladimir Davydov | c67a8a6 | 2014-06-04 16:07:39 -0700 | [diff] [blame] | 2193 | |
Vladimir Davydov | 4526477 | 2016-07-26 15:24:21 -0700 | [diff] [blame] | 2194 | static inline bool memcg_kmem_bypass(void) |
| 2195 | { |
| 2196 | if (in_interrupt() || !current->mm || (current->flags & PF_KTHREAD)) |
| 2197 | return true; |
| 2198 | return false; |
| 2199 | } |
| 2200 | |
| 2201 | /** |
| 2202 | * memcg_kmem_get_cache: select the correct per-memcg cache for allocation |
| 2203 | * @cachep: the original global kmem cache |
| 2204 | * |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 2205 | * Return the kmem_cache we're supposed to use for a slab allocation. |
| 2206 | * We try to use the current memcg's version of the cache. |
| 2207 | * |
Vladimir Davydov | 4526477 | 2016-07-26 15:24:21 -0700 | [diff] [blame] | 2208 | * If the cache does not exist yet, if we are the first user of it, we |
| 2209 | * create it asynchronously in a workqueue and let the current allocation |
| 2210 | * go through with the original cache. |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 2211 | * |
Vladimir Davydov | 4526477 | 2016-07-26 15:24:21 -0700 | [diff] [blame] | 2212 | * This function takes a reference to the cache it returns to assure it |
| 2213 | * won't get destroyed while we are working with it. Once the caller is |
| 2214 | * done with it, memcg_kmem_put_cache() must be called to release the |
| 2215 | * reference. |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 2216 | */ |
Vladimir Davydov | 4526477 | 2016-07-26 15:24:21 -0700 | [diff] [blame] | 2217 | struct kmem_cache *memcg_kmem_get_cache(struct kmem_cache *cachep) |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 2218 | { |
| 2219 | struct mem_cgroup *memcg; |
Vladimir Davydov | 959c896 | 2014-01-23 15:52:59 -0800 | [diff] [blame] | 2220 | struct kmem_cache *memcg_cachep; |
Vladimir Davydov | 2a4db7e | 2015-02-12 14:59:32 -0800 | [diff] [blame] | 2221 | int kmemcg_id; |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 2222 | |
Vladimir Davydov | f7ce319 | 2015-02-12 14:59:20 -0800 | [diff] [blame] | 2223 | VM_BUG_ON(!is_root_cache(cachep)); |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 2224 | |
Vladimir Davydov | 4526477 | 2016-07-26 15:24:21 -0700 | [diff] [blame] | 2225 | if (memcg_kmem_bypass()) |
Vladimir Davydov | 230e9fc | 2016-01-14 15:18:15 -0800 | [diff] [blame] | 2226 | return cachep; |
| 2227 | |
Vladimir Davydov | 9d100c5 | 2014-12-12 16:54:53 -0800 | [diff] [blame] | 2228 | if (current->memcg_kmem_skip_account) |
Glauber Costa | 0e9d92f | 2012-12-18 14:22:42 -0800 | [diff] [blame] | 2229 | return cachep; |
| 2230 | |
Vladimir Davydov | 8135be5 | 2014-12-12 16:56:38 -0800 | [diff] [blame] | 2231 | memcg = get_mem_cgroup_from_mm(current->mm); |
Jason Low | 4db0c3c | 2015-04-15 16:14:08 -0700 | [diff] [blame] | 2232 | kmemcg_id = READ_ONCE(memcg->kmemcg_id); |
Vladimir Davydov | 2a4db7e | 2015-02-12 14:59:32 -0800 | [diff] [blame] | 2233 | if (kmemcg_id < 0) |
Li Zefan | ca0dde9 | 2013-04-29 15:08:57 -0700 | [diff] [blame] | 2234 | goto out; |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 2235 | |
Vladimir Davydov | 2a4db7e | 2015-02-12 14:59:32 -0800 | [diff] [blame] | 2236 | memcg_cachep = cache_from_memcg_idx(cachep, kmemcg_id); |
Vladimir Davydov | 8135be5 | 2014-12-12 16:56:38 -0800 | [diff] [blame] | 2237 | if (likely(memcg_cachep)) |
| 2238 | return memcg_cachep; |
Li Zefan | ca0dde9 | 2013-04-29 15:08:57 -0700 | [diff] [blame] | 2239 | |
| 2240 | /* |
| 2241 | * If we are in a safe context (can wait, and not in interrupt |
| 2242 | * context), we could be be predictable and return right away. |
| 2243 | * This would guarantee that the allocation being performed |
| 2244 | * already belongs in the new cache. |
| 2245 | * |
| 2246 | * However, there are some clashes that can arrive from locking. |
| 2247 | * For instance, because we acquire the slab_mutex while doing |
Vladimir Davydov | 776ed0f | 2014-06-04 16:10:02 -0700 | [diff] [blame] | 2248 | * memcg_create_kmem_cache, this means no further allocation |
| 2249 | * could happen with the slab_mutex held. So it's better to |
| 2250 | * defer everything. |
Li Zefan | ca0dde9 | 2013-04-29 15:08:57 -0700 | [diff] [blame] | 2251 | */ |
Vladimir Davydov | d5b3cf7 | 2015-02-10 14:11:47 -0800 | [diff] [blame] | 2252 | memcg_schedule_kmem_cache_create(memcg, cachep); |
Li Zefan | ca0dde9 | 2013-04-29 15:08:57 -0700 | [diff] [blame] | 2253 | out: |
Vladimir Davydov | 8135be5 | 2014-12-12 16:56:38 -0800 | [diff] [blame] | 2254 | css_put(&memcg->css); |
Li Zefan | ca0dde9 | 2013-04-29 15:08:57 -0700 | [diff] [blame] | 2255 | return cachep; |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 2256 | } |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 2257 | |
Vladimir Davydov | 4526477 | 2016-07-26 15:24:21 -0700 | [diff] [blame] | 2258 | /** |
| 2259 | * memcg_kmem_put_cache: drop reference taken by memcg_kmem_get_cache |
| 2260 | * @cachep: the cache returned by memcg_kmem_get_cache |
| 2261 | */ |
| 2262 | void memcg_kmem_put_cache(struct kmem_cache *cachep) |
Vladimir Davydov | 8135be5 | 2014-12-12 16:56:38 -0800 | [diff] [blame] | 2263 | { |
| 2264 | if (!is_root_cache(cachep)) |
Vladimir Davydov | f7ce319 | 2015-02-12 14:59:20 -0800 | [diff] [blame] | 2265 | css_put(&cachep->memcg_params.memcg->css); |
Vladimir Davydov | 8135be5 | 2014-12-12 16:56:38 -0800 | [diff] [blame] | 2266 | } |
| 2267 | |
Vladimir Davydov | 4526477 | 2016-07-26 15:24:21 -0700 | [diff] [blame] | 2268 | /** |
| 2269 | * memcg_kmem_charge: charge a kmem page |
| 2270 | * @page: page to charge |
| 2271 | * @gfp: reclaim mode |
| 2272 | * @order: allocation order |
| 2273 | * @memcg: memory cgroup to charge |
| 2274 | * |
| 2275 | * Returns 0 on success, an error code on failure. |
| 2276 | */ |
| 2277 | int memcg_kmem_charge_memcg(struct page *page, gfp_t gfp, int order, |
| 2278 | struct mem_cgroup *memcg) |
Vladimir Davydov | f3ccb2c4 | 2015-11-05 18:49:01 -0800 | [diff] [blame] | 2279 | { |
| 2280 | unsigned int nr_pages = 1 << order; |
| 2281 | struct page_counter *counter; |
Johannes Weiner | 6071ca5 | 2015-11-05 18:50:26 -0800 | [diff] [blame] | 2282 | int ret; |
Vladimir Davydov | f3ccb2c4 | 2015-11-05 18:49:01 -0800 | [diff] [blame] | 2283 | |
Vladimir Davydov | f3ccb2c4 | 2015-11-05 18:49:01 -0800 | [diff] [blame] | 2284 | ret = try_charge(memcg, gfp, nr_pages); |
Johannes Weiner | 52c29b0 | 2016-01-20 15:02:35 -0800 | [diff] [blame] | 2285 | if (ret) |
Vladimir Davydov | f3ccb2c4 | 2015-11-05 18:49:01 -0800 | [diff] [blame] | 2286 | return ret; |
Johannes Weiner | 52c29b0 | 2016-01-20 15:02:35 -0800 | [diff] [blame] | 2287 | |
| 2288 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && |
| 2289 | !page_counter_try_charge(&memcg->kmem, nr_pages, &counter)) { |
| 2290 | cancel_charge(memcg, nr_pages); |
| 2291 | return -ENOMEM; |
Vladimir Davydov | f3ccb2c4 | 2015-11-05 18:49:01 -0800 | [diff] [blame] | 2292 | } |
| 2293 | |
| 2294 | page->mem_cgroup = memcg; |
| 2295 | |
| 2296 | return 0; |
| 2297 | } |
| 2298 | |
Vladimir Davydov | 4526477 | 2016-07-26 15:24:21 -0700 | [diff] [blame] | 2299 | /** |
| 2300 | * memcg_kmem_charge: charge a kmem page to the current memory cgroup |
| 2301 | * @page: page to charge |
| 2302 | * @gfp: reclaim mode |
| 2303 | * @order: allocation order |
| 2304 | * |
| 2305 | * Returns 0 on success, an error code on failure. |
| 2306 | */ |
| 2307 | int memcg_kmem_charge(struct page *page, gfp_t gfp, int order) |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 2308 | { |
| 2309 | struct mem_cgroup *memcg; |
Vladimir Davydov | fcff7d7 | 2016-03-17 14:17:29 -0700 | [diff] [blame] | 2310 | int ret = 0; |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 2311 | |
Vladimir Davydov | 4526477 | 2016-07-26 15:24:21 -0700 | [diff] [blame] | 2312 | if (memcg_kmem_bypass()) |
| 2313 | return 0; |
| 2314 | |
Johannes Weiner | df38197 | 2014-04-07 15:37:43 -0700 | [diff] [blame] | 2315 | memcg = get_mem_cgroup_from_mm(current->mm); |
Vladimir Davydov | c4159a7 | 2016-08-08 23:03:12 +0300 | [diff] [blame] | 2316 | if (!mem_cgroup_is_root(memcg)) { |
Vladimir Davydov | 4526477 | 2016-07-26 15:24:21 -0700 | [diff] [blame] | 2317 | ret = memcg_kmem_charge_memcg(page, gfp, order, memcg); |
Vladimir Davydov | c4159a7 | 2016-08-08 23:03:12 +0300 | [diff] [blame] | 2318 | if (!ret) |
| 2319 | __SetPageKmemcg(page); |
| 2320 | } |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 2321 | css_put(&memcg->css); |
Vladimir Davydov | d05e83a | 2015-11-05 18:48:59 -0800 | [diff] [blame] | 2322 | return ret; |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 2323 | } |
Vladimir Davydov | 4526477 | 2016-07-26 15:24:21 -0700 | [diff] [blame] | 2324 | /** |
| 2325 | * memcg_kmem_uncharge: uncharge a kmem page |
| 2326 | * @page: page to uncharge |
| 2327 | * @order: allocation order |
| 2328 | */ |
| 2329 | void memcg_kmem_uncharge(struct page *page, int order) |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 2330 | { |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 2331 | struct mem_cgroup *memcg = page->mem_cgroup; |
Vladimir Davydov | f3ccb2c4 | 2015-11-05 18:49:01 -0800 | [diff] [blame] | 2332 | unsigned int nr_pages = 1 << order; |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 2333 | |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 2334 | if (!memcg) |
| 2335 | return; |
| 2336 | |
Sasha Levin | 309381fea | 2014-01-23 15:52:54 -0800 | [diff] [blame] | 2337 | VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page); |
Johannes Weiner | 2983331 | 2014-12-10 15:44:02 -0800 | [diff] [blame] | 2338 | |
Johannes Weiner | 52c29b0 | 2016-01-20 15:02:35 -0800 | [diff] [blame] | 2339 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
| 2340 | page_counter_uncharge(&memcg->kmem, nr_pages); |
| 2341 | |
Vladimir Davydov | f3ccb2c4 | 2015-11-05 18:49:01 -0800 | [diff] [blame] | 2342 | page_counter_uncharge(&memcg->memory, nr_pages); |
Johannes Weiner | 7941d21 | 2016-01-14 15:21:23 -0800 | [diff] [blame] | 2343 | if (do_memsw_account()) |
Vladimir Davydov | f3ccb2c4 | 2015-11-05 18:49:01 -0800 | [diff] [blame] | 2344 | page_counter_uncharge(&memcg->memsw, nr_pages); |
| 2345 | |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 2346 | page->mem_cgroup = NULL; |
Vladimir Davydov | c4159a7 | 2016-08-08 23:03:12 +0300 | [diff] [blame] | 2347 | |
| 2348 | /* slab pages do not have PageKmemcg flag set */ |
| 2349 | if (PageKmemcg(page)) |
| 2350 | __ClearPageKmemcg(page); |
| 2351 | |
Vladimir Davydov | f3ccb2c4 | 2015-11-05 18:49:01 -0800 | [diff] [blame] | 2352 | css_put_many(&memcg->css, nr_pages); |
Vladimir Davydov | 60d3fd3 | 2015-02-12 14:59:10 -0800 | [diff] [blame] | 2353 | } |
Johannes Weiner | 127424c | 2016-01-20 15:02:32 -0800 | [diff] [blame] | 2354 | #endif /* !CONFIG_SLOB */ |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 2355 | |
KAMEZAWA Hiroyuki | ca3e021 | 2011-01-20 14:44:24 -0800 | [diff] [blame] | 2356 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
| 2357 | |
KAMEZAWA Hiroyuki | ca3e021 | 2011-01-20 14:44:24 -0800 | [diff] [blame] | 2358 | /* |
| 2359 | * Because tail pages are not marked as "used", set it. We're under |
Mel Gorman | a52633d | 2016-07-28 15:45:28 -0700 | [diff] [blame] | 2360 | * zone_lru_lock and migration entries setup in all page mappings. |
KAMEZAWA Hiroyuki | ca3e021 | 2011-01-20 14:44:24 -0800 | [diff] [blame] | 2361 | */ |
KAMEZAWA Hiroyuki | e94c8a9 | 2012-01-12 17:18:20 -0800 | [diff] [blame] | 2362 | void mem_cgroup_split_huge_fixup(struct page *head) |
KAMEZAWA Hiroyuki | ca3e021 | 2011-01-20 14:44:24 -0800 | [diff] [blame] | 2363 | { |
KAMEZAWA Hiroyuki | e94c8a9 | 2012-01-12 17:18:20 -0800 | [diff] [blame] | 2364 | int i; |
KAMEZAWA Hiroyuki | ca3e021 | 2011-01-20 14:44:24 -0800 | [diff] [blame] | 2365 | |
KAMEZAWA Hiroyuki | 3d37c4a | 2011-01-25 15:07:28 -0800 | [diff] [blame] | 2366 | if (mem_cgroup_disabled()) |
| 2367 | return; |
David Rientjes | b070e65 | 2013-05-07 16:18:09 -0700 | [diff] [blame] | 2368 | |
Johannes Weiner | 2983331 | 2014-12-10 15:44:02 -0800 | [diff] [blame] | 2369 | for (i = 1; i < HPAGE_PMD_NR; i++) |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 2370 | head[i].mem_cgroup = head->mem_cgroup; |
Michal Hocko | b9982f8 | 2014-12-10 15:43:51 -0800 | [diff] [blame] | 2371 | |
Johannes Weiner | 71cd311 | 2017-05-03 14:55:13 -0700 | [diff] [blame] | 2372 | __this_cpu_sub(head->mem_cgroup->stat->count[MEMCG_RSS_HUGE], |
David Rientjes | b070e65 | 2013-05-07 16:18:09 -0700 | [diff] [blame] | 2373 | HPAGE_PMD_NR); |
KAMEZAWA Hiroyuki | ca3e021 | 2011-01-20 14:44:24 -0800 | [diff] [blame] | 2374 | } |
Hugh Dickins | 12d2710 | 2012-01-12 17:19:52 -0800 | [diff] [blame] | 2375 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
KAMEZAWA Hiroyuki | ca3e021 | 2011-01-20 14:44:24 -0800 | [diff] [blame] | 2376 | |
Andrew Morton | c255a45 | 2012-07-31 16:43:02 -0700 | [diff] [blame] | 2377 | #ifdef CONFIG_MEMCG_SWAP |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 2378 | static void mem_cgroup_swap_statistics(struct mem_cgroup *memcg, |
Huang Ying | 38d8b4e | 2017-07-06 15:37:18 -0700 | [diff] [blame] | 2379 | int nr_entries) |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2380 | { |
Huang Ying | 38d8b4e | 2017-07-06 15:37:18 -0700 | [diff] [blame] | 2381 | this_cpu_add(memcg->stat->count[MEMCG_SWAP], nr_entries); |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2382 | } |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 2383 | |
| 2384 | /** |
| 2385 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. |
| 2386 | * @entry: swap entry to be moved |
| 2387 | * @from: mem_cgroup which the entry is moved from |
| 2388 | * @to: mem_cgroup which the entry is moved to |
| 2389 | * |
| 2390 | * It succeeds only when the swap_cgroup's record for this entry is the same |
| 2391 | * as the mem_cgroup's id of @from. |
| 2392 | * |
| 2393 | * Returns 0 on success, -EINVAL on failure. |
| 2394 | * |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2395 | * The caller must have charged to @to, IOW, called page_counter_charge() about |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 2396 | * both res and memsw, and called css_get(). |
| 2397 | */ |
| 2398 | static int mem_cgroup_move_swap_account(swp_entry_t entry, |
Hugh Dickins | e91cbb4 | 2012-05-29 15:06:51 -0700 | [diff] [blame] | 2399 | struct mem_cgroup *from, struct mem_cgroup *to) |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 2400 | { |
| 2401 | unsigned short old_id, new_id; |
| 2402 | |
Li Zefan | 34c00c3 | 2013-09-23 16:56:01 +0800 | [diff] [blame] | 2403 | old_id = mem_cgroup_id(from); |
| 2404 | new_id = mem_cgroup_id(to); |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 2405 | |
| 2406 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { |
Huang Ying | 38d8b4e | 2017-07-06 15:37:18 -0700 | [diff] [blame] | 2407 | mem_cgroup_swap_statistics(from, -1); |
| 2408 | mem_cgroup_swap_statistics(to, 1); |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 2409 | return 0; |
| 2410 | } |
| 2411 | return -EINVAL; |
| 2412 | } |
| 2413 | #else |
| 2414 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, |
Hugh Dickins | e91cbb4 | 2012-05-29 15:06:51 -0700 | [diff] [blame] | 2415 | struct mem_cgroup *from, struct mem_cgroup *to) |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 2416 | { |
| 2417 | return -EINVAL; |
| 2418 | } |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2419 | #endif |
| 2420 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2421 | static DEFINE_MUTEX(memcg_limit_mutex); |
Daisuke Nishimura | f212ad7 | 2011-03-23 16:42:25 -0700 | [diff] [blame] | 2422 | |
KOSAKI Motohiro | d38d2a7 | 2009-01-06 14:39:44 -0800 | [diff] [blame] | 2423 | static int mem_cgroup_resize_limit(struct mem_cgroup *memcg, |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2424 | unsigned long limit) |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 2425 | { |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2426 | unsigned long curusage; |
| 2427 | unsigned long oldusage; |
| 2428 | bool enlarge = false; |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 2429 | int retry_count; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2430 | int ret; |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 2431 | |
| 2432 | /* |
| 2433 | * For keeping hierarchical_reclaim simple, how long we should retry |
| 2434 | * is depends on callers. We set our retry-count to be function |
| 2435 | * of # of children which we should visit in this loop. |
| 2436 | */ |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2437 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * |
| 2438 | mem_cgroup_count_children(memcg); |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 2439 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2440 | oldusage = page_counter_read(&memcg->memory); |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 2441 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2442 | do { |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 2443 | if (signal_pending(current)) { |
| 2444 | ret = -EINTR; |
| 2445 | break; |
| 2446 | } |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2447 | |
| 2448 | mutex_lock(&memcg_limit_mutex); |
| 2449 | if (limit > memcg->memsw.limit) { |
| 2450 | mutex_unlock(&memcg_limit_mutex); |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2451 | ret = -EINVAL; |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 2452 | break; |
| 2453 | } |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2454 | if (limit > memcg->memory.limit) |
| 2455 | enlarge = true; |
| 2456 | ret = page_counter_limit(&memcg->memory, limit); |
| 2457 | mutex_unlock(&memcg_limit_mutex); |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2458 | |
| 2459 | if (!ret) |
| 2460 | break; |
| 2461 | |
Johannes Weiner | b70a2a2 | 2014-10-09 15:28:56 -0700 | [diff] [blame] | 2462 | try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, true); |
| 2463 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2464 | curusage = page_counter_read(&memcg->memory); |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 2465 | /* Usage is reduced ? */ |
Andrew Morton | f894ffa | 2013-09-12 15:13:35 -0700 | [diff] [blame] | 2466 | if (curusage >= oldusage) |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 2467 | retry_count--; |
| 2468 | else |
| 2469 | oldusage = curusage; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2470 | } while (retry_count); |
| 2471 | |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 2472 | if (!ret && enlarge) |
| 2473 | memcg_oom_recover(memcg); |
KOSAKI Motohiro | 14797e2 | 2009-01-07 18:08:18 -0800 | [diff] [blame] | 2474 | |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2475 | return ret; |
| 2476 | } |
| 2477 | |
Li Zefan | 338c843 | 2009-06-17 16:27:15 -0700 | [diff] [blame] | 2478 | static int mem_cgroup_resize_memsw_limit(struct mem_cgroup *memcg, |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2479 | unsigned long limit) |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2480 | { |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2481 | unsigned long curusage; |
| 2482 | unsigned long oldusage; |
| 2483 | bool enlarge = false; |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 2484 | int retry_count; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2485 | int ret; |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2486 | |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 2487 | /* see mem_cgroup_resize_res_limit */ |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2488 | retry_count = MEM_CGROUP_RECLAIM_RETRIES * |
| 2489 | mem_cgroup_count_children(memcg); |
| 2490 | |
| 2491 | oldusage = page_counter_read(&memcg->memsw); |
| 2492 | |
| 2493 | do { |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2494 | if (signal_pending(current)) { |
| 2495 | ret = -EINTR; |
| 2496 | break; |
| 2497 | } |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2498 | |
| 2499 | mutex_lock(&memcg_limit_mutex); |
| 2500 | if (limit < memcg->memory.limit) { |
| 2501 | mutex_unlock(&memcg_limit_mutex); |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2502 | ret = -EINVAL; |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2503 | break; |
| 2504 | } |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2505 | if (limit > memcg->memsw.limit) |
| 2506 | enlarge = true; |
| 2507 | ret = page_counter_limit(&memcg->memsw, limit); |
| 2508 | mutex_unlock(&memcg_limit_mutex); |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2509 | |
| 2510 | if (!ret) |
| 2511 | break; |
| 2512 | |
Johannes Weiner | b70a2a2 | 2014-10-09 15:28:56 -0700 | [diff] [blame] | 2513 | try_to_free_mem_cgroup_pages(memcg, 1, GFP_KERNEL, false); |
| 2514 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2515 | curusage = page_counter_read(&memcg->memsw); |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 2516 | /* Usage is reduced ? */ |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2517 | if (curusage >= oldusage) |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 2518 | retry_count--; |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 2519 | else |
| 2520 | oldusage = curusage; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2521 | } while (retry_count); |
| 2522 | |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 2523 | if (!ret && enlarge) |
| 2524 | memcg_oom_recover(memcg); |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2525 | |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 2526 | return ret; |
| 2527 | } |
| 2528 | |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 2529 | unsigned long mem_cgroup_soft_limit_reclaim(pg_data_t *pgdat, int order, |
Andrew Morton | 0608f43 | 2013-09-24 15:27:41 -0700 | [diff] [blame] | 2530 | gfp_t gfp_mask, |
| 2531 | unsigned long *total_scanned) |
| 2532 | { |
| 2533 | unsigned long nr_reclaimed = 0; |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 2534 | struct mem_cgroup_per_node *mz, *next_mz = NULL; |
Andrew Morton | 0608f43 | 2013-09-24 15:27:41 -0700 | [diff] [blame] | 2535 | unsigned long reclaimed; |
| 2536 | int loop = 0; |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 2537 | struct mem_cgroup_tree_per_node *mctz; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2538 | unsigned long excess; |
Andrew Morton | 0608f43 | 2013-09-24 15:27:41 -0700 | [diff] [blame] | 2539 | unsigned long nr_scanned; |
| 2540 | |
| 2541 | if (order > 0) |
| 2542 | return 0; |
| 2543 | |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 2544 | mctz = soft_limit_tree_node(pgdat->node_id); |
Michal Hocko | d6507ff | 2016-08-02 14:02:37 -0700 | [diff] [blame] | 2545 | |
| 2546 | /* |
| 2547 | * Do not even bother to check the largest node if the root |
| 2548 | * is empty. Do it lockless to prevent lock bouncing. Races |
| 2549 | * are acceptable as soft limit is best effort anyway. |
| 2550 | */ |
Laurent Dufour | bfc7228 | 2017-03-09 16:17:06 -0800 | [diff] [blame] | 2551 | if (!mctz || RB_EMPTY_ROOT(&mctz->rb_root)) |
Michal Hocko | d6507ff | 2016-08-02 14:02:37 -0700 | [diff] [blame] | 2552 | return 0; |
| 2553 | |
Andrew Morton | 0608f43 | 2013-09-24 15:27:41 -0700 | [diff] [blame] | 2554 | /* |
| 2555 | * This loop can run a while, specially if mem_cgroup's continuously |
| 2556 | * keep exceeding their soft limit and putting the system under |
| 2557 | * pressure |
| 2558 | */ |
| 2559 | do { |
| 2560 | if (next_mz) |
| 2561 | mz = next_mz; |
| 2562 | else |
| 2563 | mz = mem_cgroup_largest_soft_limit_node(mctz); |
| 2564 | if (!mz) |
| 2565 | break; |
| 2566 | |
| 2567 | nr_scanned = 0; |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 2568 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, pgdat, |
Andrew Morton | 0608f43 | 2013-09-24 15:27:41 -0700 | [diff] [blame] | 2569 | gfp_mask, &nr_scanned); |
| 2570 | nr_reclaimed += reclaimed; |
| 2571 | *total_scanned += nr_scanned; |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 2572 | spin_lock_irq(&mctz->lock); |
Vladimir Davydov | bc2f2e7 | 2014-12-10 15:43:40 -0800 | [diff] [blame] | 2573 | __mem_cgroup_remove_exceeded(mz, mctz); |
Andrew Morton | 0608f43 | 2013-09-24 15:27:41 -0700 | [diff] [blame] | 2574 | |
| 2575 | /* |
| 2576 | * If we failed to reclaim anything from this memory cgroup |
| 2577 | * it is time to move on to the next cgroup |
| 2578 | */ |
| 2579 | next_mz = NULL; |
Vladimir Davydov | bc2f2e7 | 2014-12-10 15:43:40 -0800 | [diff] [blame] | 2580 | if (!reclaimed) |
| 2581 | next_mz = __mem_cgroup_largest_soft_limit_node(mctz); |
| 2582 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2583 | excess = soft_limit_excess(mz->memcg); |
Andrew Morton | 0608f43 | 2013-09-24 15:27:41 -0700 | [diff] [blame] | 2584 | /* |
| 2585 | * One school of thought says that we should not add |
| 2586 | * back the node to the tree if reclaim returns 0. |
| 2587 | * But our reclaim could return 0, simply because due |
| 2588 | * to priority we are exposing a smaller subset of |
| 2589 | * memory to reclaim from. Consider this as a longer |
| 2590 | * term TODO. |
| 2591 | */ |
| 2592 | /* If excess == 0, no tree ops */ |
Johannes Weiner | cf2c812 | 2014-06-06 14:38:21 -0700 | [diff] [blame] | 2593 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 2594 | spin_unlock_irq(&mctz->lock); |
Andrew Morton | 0608f43 | 2013-09-24 15:27:41 -0700 | [diff] [blame] | 2595 | css_put(&mz->memcg->css); |
| 2596 | loop++; |
| 2597 | /* |
| 2598 | * Could not reclaim anything and there are no more |
| 2599 | * mem cgroups to try or we seem to be looping without |
| 2600 | * reclaiming anything. |
| 2601 | */ |
| 2602 | if (!nr_reclaimed && |
| 2603 | (next_mz == NULL || |
| 2604 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) |
| 2605 | break; |
| 2606 | } while (!nr_reclaimed); |
| 2607 | if (next_mz) |
| 2608 | css_put(&next_mz->memcg->css); |
| 2609 | return nr_reclaimed; |
| 2610 | } |
| 2611 | |
Tejun Heo | ea280e7 | 2014-05-16 13:22:48 -0400 | [diff] [blame] | 2612 | /* |
| 2613 | * Test whether @memcg has children, dead or alive. Note that this |
| 2614 | * function doesn't care whether @memcg has use_hierarchy enabled and |
| 2615 | * returns %true if there are child csses according to the cgroup |
| 2616 | * hierarchy. Testing use_hierarchy is the caller's responsiblity. |
| 2617 | */ |
Glauber Costa | b5f99b5 | 2013-02-22 16:34:53 -0800 | [diff] [blame] | 2618 | static inline bool memcg_has_children(struct mem_cgroup *memcg) |
| 2619 | { |
Tejun Heo | ea280e7 | 2014-05-16 13:22:48 -0400 | [diff] [blame] | 2620 | bool ret; |
| 2621 | |
Tejun Heo | ea280e7 | 2014-05-16 13:22:48 -0400 | [diff] [blame] | 2622 | rcu_read_lock(); |
| 2623 | ret = css_next_child(NULL, &memcg->css); |
| 2624 | rcu_read_unlock(); |
| 2625 | return ret; |
Glauber Costa | b5f99b5 | 2013-02-22 16:34:53 -0800 | [diff] [blame] | 2626 | } |
| 2627 | |
| 2628 | /* |
Greg Thelen | 5103817 | 2016-05-20 16:58:18 -0700 | [diff] [blame] | 2629 | * Reclaims as many pages from the given memcg as possible. |
Michal Hocko | c26251f | 2012-10-26 13:37:28 +0200 | [diff] [blame] | 2630 | * |
| 2631 | * Caller is responsible for holding css reference for memcg. |
| 2632 | */ |
| 2633 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg) |
| 2634 | { |
| 2635 | int nr_retries = MEM_CGROUP_RECLAIM_RETRIES; |
Michal Hocko | c26251f | 2012-10-26 13:37:28 +0200 | [diff] [blame] | 2636 | |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 2637 | /* we call try-to-free pages for make this cgroup empty */ |
| 2638 | lru_add_drain_all(); |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 2639 | /* try to free all pages in this cgroup */ |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2640 | while (nr_retries && page_counter_read(&memcg->memory)) { |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 2641 | int progress; |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 2642 | |
Michal Hocko | c26251f | 2012-10-26 13:37:28 +0200 | [diff] [blame] | 2643 | if (signal_pending(current)) |
| 2644 | return -EINTR; |
| 2645 | |
Johannes Weiner | b70a2a2 | 2014-10-09 15:28:56 -0700 | [diff] [blame] | 2646 | progress = try_to_free_mem_cgroup_pages(memcg, 1, |
| 2647 | GFP_KERNEL, true); |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 2648 | if (!progress) { |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 2649 | nr_retries--; |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 2650 | /* maybe some writeback is necessary */ |
Jens Axboe | 8aa7e84 | 2009-07-09 14:52:32 +0200 | [diff] [blame] | 2651 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 2652 | } |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 2653 | |
| 2654 | } |
Michal Hocko | ab5196c | 2012-10-26 13:37:32 +0200 | [diff] [blame] | 2655 | |
| 2656 | return 0; |
KAMEZAWA Hiroyuki | cc84758 | 2008-02-07 00:14:16 -0800 | [diff] [blame] | 2657 | } |
| 2658 | |
Tejun Heo | 6770c64 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 2659 | static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of, |
| 2660 | char *buf, size_t nbytes, |
| 2661 | loff_t off) |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 2662 | { |
Tejun Heo | 6770c64 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 2663 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
Michal Hocko | c26251f | 2012-10-26 13:37:28 +0200 | [diff] [blame] | 2664 | |
Michal Hocko | d842301 | 2012-10-26 13:37:29 +0200 | [diff] [blame] | 2665 | if (mem_cgroup_is_root(memcg)) |
| 2666 | return -EINVAL; |
Tejun Heo | 6770c64 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 2667 | return mem_cgroup_force_empty(memcg) ?: nbytes; |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 2668 | } |
| 2669 | |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 2670 | static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css, |
| 2671 | struct cftype *cft) |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 2672 | { |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 2673 | return mem_cgroup_from_css(css)->use_hierarchy; |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 2674 | } |
| 2675 | |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 2676 | static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css, |
| 2677 | struct cftype *cft, u64 val) |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 2678 | { |
| 2679 | int retval = 0; |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 2680 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
Tejun Heo | 5c9d535 | 2014-05-16 13:22:48 -0400 | [diff] [blame] | 2681 | struct mem_cgroup *parent_memcg = mem_cgroup_from_css(memcg->css.parent); |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 2682 | |
Glauber Costa | 567fb43 | 2012-07-31 16:43:07 -0700 | [diff] [blame] | 2683 | if (memcg->use_hierarchy == val) |
Johannes Weiner | 0b8f73e | 2016-01-20 15:02:53 -0800 | [diff] [blame] | 2684 | return 0; |
Glauber Costa | 567fb43 | 2012-07-31 16:43:07 -0700 | [diff] [blame] | 2685 | |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 2686 | /* |
André Goddard Rosa | af901ca | 2009-11-14 13:09:05 -0200 | [diff] [blame] | 2687 | * If parent's use_hierarchy is set, we can't make any modifications |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 2688 | * in the child subtrees. If it is unset, then the change can |
| 2689 | * occur, provided the current cgroup has no children. |
| 2690 | * |
| 2691 | * For the root cgroup, parent_mem is NULL, we allow value to be |
| 2692 | * set if there are no children. |
| 2693 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2694 | if ((!parent_memcg || !parent_memcg->use_hierarchy) && |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 2695 | (val == 1 || val == 0)) { |
Tejun Heo | ea280e7 | 2014-05-16 13:22:48 -0400 | [diff] [blame] | 2696 | if (!memcg_has_children(memcg)) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2697 | memcg->use_hierarchy = val; |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 2698 | else |
| 2699 | retval = -EBUSY; |
| 2700 | } else |
| 2701 | retval = -EINVAL; |
Glauber Costa | 567fb43 | 2012-07-31 16:43:07 -0700 | [diff] [blame] | 2702 | |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 2703 | return retval; |
| 2704 | } |
| 2705 | |
Vladimir Davydov | 72b54e7 | 2016-03-17 14:17:32 -0700 | [diff] [blame] | 2706 | static void tree_stat(struct mem_cgroup *memcg, unsigned long *stat) |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 2707 | { |
| 2708 | struct mem_cgroup *iter; |
Vladimir Davydov | 72b54e7 | 2016-03-17 14:17:32 -0700 | [diff] [blame] | 2709 | int i; |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 2710 | |
Vladimir Davydov | 72b54e7 | 2016-03-17 14:17:32 -0700 | [diff] [blame] | 2711 | memset(stat, 0, sizeof(*stat) * MEMCG_NR_STAT); |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 2712 | |
Vladimir Davydov | 72b54e7 | 2016-03-17 14:17:32 -0700 | [diff] [blame] | 2713 | for_each_mem_cgroup_tree(iter, memcg) { |
| 2714 | for (i = 0; i < MEMCG_NR_STAT; i++) |
Johannes Weiner | ccda7f4 | 2017-05-03 14:55:16 -0700 | [diff] [blame] | 2715 | stat[i] += memcg_page_state(iter, i); |
Vladimir Davydov | 72b54e7 | 2016-03-17 14:17:32 -0700 | [diff] [blame] | 2716 | } |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 2717 | } |
| 2718 | |
Vladimir Davydov | 72b54e7 | 2016-03-17 14:17:32 -0700 | [diff] [blame] | 2719 | static void tree_events(struct mem_cgroup *memcg, unsigned long *events) |
Johannes Weiner | 587d9f7 | 2016-01-20 15:03:19 -0800 | [diff] [blame] | 2720 | { |
| 2721 | struct mem_cgroup *iter; |
Vladimir Davydov | 72b54e7 | 2016-03-17 14:17:32 -0700 | [diff] [blame] | 2722 | int i; |
Johannes Weiner | 587d9f7 | 2016-01-20 15:03:19 -0800 | [diff] [blame] | 2723 | |
Vladimir Davydov | 72b54e7 | 2016-03-17 14:17:32 -0700 | [diff] [blame] | 2724 | memset(events, 0, sizeof(*events) * MEMCG_NR_EVENTS); |
Johannes Weiner | 587d9f7 | 2016-01-20 15:03:19 -0800 | [diff] [blame] | 2725 | |
Vladimir Davydov | 72b54e7 | 2016-03-17 14:17:32 -0700 | [diff] [blame] | 2726 | for_each_mem_cgroup_tree(iter, memcg) { |
| 2727 | for (i = 0; i < MEMCG_NR_EVENTS; i++) |
Johannes Weiner | ccda7f4 | 2017-05-03 14:55:16 -0700 | [diff] [blame] | 2728 | events[i] += memcg_sum_events(iter, i); |
Vladimir Davydov | 72b54e7 | 2016-03-17 14:17:32 -0700 | [diff] [blame] | 2729 | } |
Johannes Weiner | 587d9f7 | 2016-01-20 15:03:19 -0800 | [diff] [blame] | 2730 | } |
| 2731 | |
Andrew Morton | 6f64615 | 2015-11-06 16:28:58 -0800 | [diff] [blame] | 2732 | static unsigned long mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 2733 | { |
Vladimir Davydov | 72b54e7 | 2016-03-17 14:17:32 -0700 | [diff] [blame] | 2734 | unsigned long val = 0; |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 2735 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2736 | if (mem_cgroup_is_root(memcg)) { |
Vladimir Davydov | 72b54e7 | 2016-03-17 14:17:32 -0700 | [diff] [blame] | 2737 | struct mem_cgroup *iter; |
| 2738 | |
| 2739 | for_each_mem_cgroup_tree(iter, memcg) { |
Johannes Weiner | ccda7f4 | 2017-05-03 14:55:16 -0700 | [diff] [blame] | 2740 | val += memcg_page_state(iter, MEMCG_CACHE); |
| 2741 | val += memcg_page_state(iter, MEMCG_RSS); |
Vladimir Davydov | 72b54e7 | 2016-03-17 14:17:32 -0700 | [diff] [blame] | 2742 | if (swap) |
Johannes Weiner | ccda7f4 | 2017-05-03 14:55:16 -0700 | [diff] [blame] | 2743 | val += memcg_page_state(iter, MEMCG_SWAP); |
Vladimir Davydov | 72b54e7 | 2016-03-17 14:17:32 -0700 | [diff] [blame] | 2744 | } |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2745 | } else { |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 2746 | if (!swap) |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2747 | val = page_counter_read(&memcg->memory); |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 2748 | else |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2749 | val = page_counter_read(&memcg->memsw); |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 2750 | } |
Michal Hocko | c12176d | 2015-11-05 18:50:29 -0800 | [diff] [blame] | 2751 | return val; |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 2752 | } |
| 2753 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2754 | enum { |
| 2755 | RES_USAGE, |
| 2756 | RES_LIMIT, |
| 2757 | RES_MAX_USAGE, |
| 2758 | RES_FAILCNT, |
| 2759 | RES_SOFT_LIMIT, |
| 2760 | }; |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 2761 | |
Tejun Heo | 791badb | 2013-12-05 12:28:02 -0500 | [diff] [blame] | 2762 | static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css, |
Johannes Weiner | 05b8430 | 2014-08-06 16:05:59 -0700 | [diff] [blame] | 2763 | struct cftype *cft) |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 2764 | { |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 2765 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2766 | struct page_counter *counter; |
Tejun Heo | af36f90 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 2767 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2768 | switch (MEMFILE_TYPE(cft->private)) { |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2769 | case _MEM: |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2770 | counter = &memcg->memory; |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 2771 | break; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2772 | case _MEMSWAP: |
| 2773 | counter = &memcg->memsw; |
| 2774 | break; |
| 2775 | case _KMEM: |
| 2776 | counter = &memcg->kmem; |
| 2777 | break; |
Vladimir Davydov | d55f90b | 2016-01-20 15:02:44 -0800 | [diff] [blame] | 2778 | case _TCP: |
Johannes Weiner | 0db1529 | 2016-01-20 15:02:50 -0800 | [diff] [blame] | 2779 | counter = &memcg->tcpmem; |
Vladimir Davydov | d55f90b | 2016-01-20 15:02:44 -0800 | [diff] [blame] | 2780 | break; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2781 | default: |
| 2782 | BUG(); |
| 2783 | } |
| 2784 | |
| 2785 | switch (MEMFILE_ATTR(cft->private)) { |
| 2786 | case RES_USAGE: |
| 2787 | if (counter == &memcg->memory) |
Michal Hocko | c12176d | 2015-11-05 18:50:29 -0800 | [diff] [blame] | 2788 | return (u64)mem_cgroup_usage(memcg, false) * PAGE_SIZE; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2789 | if (counter == &memcg->memsw) |
Michal Hocko | c12176d | 2015-11-05 18:50:29 -0800 | [diff] [blame] | 2790 | return (u64)mem_cgroup_usage(memcg, true) * PAGE_SIZE; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2791 | return (u64)page_counter_read(counter) * PAGE_SIZE; |
| 2792 | case RES_LIMIT: |
| 2793 | return (u64)counter->limit * PAGE_SIZE; |
| 2794 | case RES_MAX_USAGE: |
| 2795 | return (u64)counter->watermark * PAGE_SIZE; |
| 2796 | case RES_FAILCNT: |
| 2797 | return counter->failcnt; |
| 2798 | case RES_SOFT_LIMIT: |
| 2799 | return (u64)memcg->soft_limit * PAGE_SIZE; |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2800 | default: |
| 2801 | BUG(); |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2802 | } |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 2803 | } |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 2804 | |
Johannes Weiner | 127424c | 2016-01-20 15:02:32 -0800 | [diff] [blame] | 2805 | #ifndef CONFIG_SLOB |
Johannes Weiner | 567e9ab | 2016-01-20 15:02:24 -0800 | [diff] [blame] | 2806 | static int memcg_online_kmem(struct mem_cgroup *memcg) |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 2807 | { |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 2808 | int memcg_id; |
| 2809 | |
Vladimir Davydov | b313aee | 2016-03-17 14:18:27 -0700 | [diff] [blame] | 2810 | if (cgroup_memory_nokmem) |
| 2811 | return 0; |
| 2812 | |
Vladimir Davydov | 2a4db7e | 2015-02-12 14:59:32 -0800 | [diff] [blame] | 2813 | BUG_ON(memcg->kmemcg_id >= 0); |
Johannes Weiner | 567e9ab | 2016-01-20 15:02:24 -0800 | [diff] [blame] | 2814 | BUG_ON(memcg->kmem_state); |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 2815 | |
Vladimir Davydov | f3bb304 | 2014-10-09 15:28:45 -0700 | [diff] [blame] | 2816 | memcg_id = memcg_alloc_cache_id(); |
Johannes Weiner | 0b8f73e | 2016-01-20 15:02:53 -0800 | [diff] [blame] | 2817 | if (memcg_id < 0) |
| 2818 | return memcg_id; |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 2819 | |
Johannes Weiner | ef12947 | 2016-01-14 15:21:34 -0800 | [diff] [blame] | 2820 | static_branch_inc(&memcg_kmem_enabled_key); |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 2821 | /* |
Johannes Weiner | 567e9ab | 2016-01-20 15:02:24 -0800 | [diff] [blame] | 2822 | * A memory cgroup is considered kmem-online as soon as it gets |
Vladimir Davydov | 900a38f | 2014-12-12 16:55:10 -0800 | [diff] [blame] | 2823 | * kmemcg_id. Setting the id after enabling static branching will |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 2824 | * guarantee no one starts accounting before all call sites are |
| 2825 | * patched. |
| 2826 | */ |
Vladimir Davydov | 900a38f | 2014-12-12 16:55:10 -0800 | [diff] [blame] | 2827 | memcg->kmemcg_id = memcg_id; |
Johannes Weiner | 567e9ab | 2016-01-20 15:02:24 -0800 | [diff] [blame] | 2828 | memcg->kmem_state = KMEM_ONLINE; |
Tejun Heo | bc2791f | 2017-02-22 15:41:21 -0800 | [diff] [blame] | 2829 | INIT_LIST_HEAD(&memcg->kmem_caches); |
Johannes Weiner | 0b8f73e | 2016-01-20 15:02:53 -0800 | [diff] [blame] | 2830 | |
| 2831 | return 0; |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 2832 | } |
| 2833 | |
Johannes Weiner | 8e0a891 | 2016-01-20 15:02:26 -0800 | [diff] [blame] | 2834 | static void memcg_offline_kmem(struct mem_cgroup *memcg) |
| 2835 | { |
| 2836 | struct cgroup_subsys_state *css; |
| 2837 | struct mem_cgroup *parent, *child; |
| 2838 | int kmemcg_id; |
| 2839 | |
| 2840 | if (memcg->kmem_state != KMEM_ONLINE) |
| 2841 | return; |
| 2842 | /* |
| 2843 | * Clear the online state before clearing memcg_caches array |
| 2844 | * entries. The slab_mutex in memcg_deactivate_kmem_caches() |
| 2845 | * guarantees that no cache will be created for this cgroup |
| 2846 | * after we are done (see memcg_create_kmem_cache()). |
| 2847 | */ |
| 2848 | memcg->kmem_state = KMEM_ALLOCATED; |
| 2849 | |
| 2850 | memcg_deactivate_kmem_caches(memcg); |
| 2851 | |
| 2852 | kmemcg_id = memcg->kmemcg_id; |
| 2853 | BUG_ON(kmemcg_id < 0); |
| 2854 | |
| 2855 | parent = parent_mem_cgroup(memcg); |
| 2856 | if (!parent) |
| 2857 | parent = root_mem_cgroup; |
| 2858 | |
| 2859 | /* |
| 2860 | * Change kmemcg_id of this cgroup and all its descendants to the |
| 2861 | * parent's id, and then move all entries from this cgroup's list_lrus |
| 2862 | * to ones of the parent. After we have finished, all list_lrus |
| 2863 | * corresponding to this cgroup are guaranteed to remain empty. The |
| 2864 | * ordering is imposed by list_lru_node->lock taken by |
| 2865 | * memcg_drain_all_list_lrus(). |
| 2866 | */ |
Tejun Heo | 3a06bb7 | 2016-06-03 14:55:44 -0700 | [diff] [blame] | 2867 | rcu_read_lock(); /* can be called from css_free w/o cgroup_mutex */ |
Johannes Weiner | 8e0a891 | 2016-01-20 15:02:26 -0800 | [diff] [blame] | 2868 | css_for_each_descendant_pre(css, &memcg->css) { |
| 2869 | child = mem_cgroup_from_css(css); |
| 2870 | BUG_ON(child->kmemcg_id != kmemcg_id); |
| 2871 | child->kmemcg_id = parent->kmemcg_id; |
| 2872 | if (!memcg->use_hierarchy) |
| 2873 | break; |
| 2874 | } |
Tejun Heo | 3a06bb7 | 2016-06-03 14:55:44 -0700 | [diff] [blame] | 2875 | rcu_read_unlock(); |
| 2876 | |
Johannes Weiner | 8e0a891 | 2016-01-20 15:02:26 -0800 | [diff] [blame] | 2877 | memcg_drain_all_list_lrus(kmemcg_id, parent->kmemcg_id); |
| 2878 | |
| 2879 | memcg_free_cache_id(kmemcg_id); |
| 2880 | } |
| 2881 | |
| 2882 | static void memcg_free_kmem(struct mem_cgroup *memcg) |
| 2883 | { |
Johannes Weiner | 0b8f73e | 2016-01-20 15:02:53 -0800 | [diff] [blame] | 2884 | /* css_alloc() failed, offlining didn't happen */ |
| 2885 | if (unlikely(memcg->kmem_state == KMEM_ONLINE)) |
| 2886 | memcg_offline_kmem(memcg); |
| 2887 | |
Johannes Weiner | 8e0a891 | 2016-01-20 15:02:26 -0800 | [diff] [blame] | 2888 | if (memcg->kmem_state == KMEM_ALLOCATED) { |
| 2889 | memcg_destroy_kmem_caches(memcg); |
| 2890 | static_branch_dec(&memcg_kmem_enabled_key); |
| 2891 | WARN_ON(page_counter_read(&memcg->kmem)); |
| 2892 | } |
Johannes Weiner | 8e0a891 | 2016-01-20 15:02:26 -0800 | [diff] [blame] | 2893 | } |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 2894 | #else |
Johannes Weiner | 0b8f73e | 2016-01-20 15:02:53 -0800 | [diff] [blame] | 2895 | static int memcg_online_kmem(struct mem_cgroup *memcg) |
Johannes Weiner | 127424c | 2016-01-20 15:02:32 -0800 | [diff] [blame] | 2896 | { |
| 2897 | return 0; |
| 2898 | } |
| 2899 | static void memcg_offline_kmem(struct mem_cgroup *memcg) |
| 2900 | { |
| 2901 | } |
| 2902 | static void memcg_free_kmem(struct mem_cgroup *memcg) |
| 2903 | { |
| 2904 | } |
| 2905 | #endif /* !CONFIG_SLOB */ |
| 2906 | |
Johannes Weiner | 127424c | 2016-01-20 15:02:32 -0800 | [diff] [blame] | 2907 | static int memcg_update_kmem_limit(struct mem_cgroup *memcg, |
| 2908 | unsigned long limit) |
| 2909 | { |
Vladimir Davydov | b313aee | 2016-03-17 14:18:27 -0700 | [diff] [blame] | 2910 | int ret; |
Johannes Weiner | 127424c | 2016-01-20 15:02:32 -0800 | [diff] [blame] | 2911 | |
| 2912 | mutex_lock(&memcg_limit_mutex); |
Johannes Weiner | 127424c | 2016-01-20 15:02:32 -0800 | [diff] [blame] | 2913 | ret = page_counter_limit(&memcg->kmem, limit); |
Johannes Weiner | 127424c | 2016-01-20 15:02:32 -0800 | [diff] [blame] | 2914 | mutex_unlock(&memcg_limit_mutex); |
| 2915 | return ret; |
| 2916 | } |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 2917 | |
Vladimir Davydov | d55f90b | 2016-01-20 15:02:44 -0800 | [diff] [blame] | 2918 | static int memcg_update_tcp_limit(struct mem_cgroup *memcg, unsigned long limit) |
| 2919 | { |
| 2920 | int ret; |
| 2921 | |
| 2922 | mutex_lock(&memcg_limit_mutex); |
| 2923 | |
Johannes Weiner | 0db1529 | 2016-01-20 15:02:50 -0800 | [diff] [blame] | 2924 | ret = page_counter_limit(&memcg->tcpmem, limit); |
Vladimir Davydov | d55f90b | 2016-01-20 15:02:44 -0800 | [diff] [blame] | 2925 | if (ret) |
| 2926 | goto out; |
| 2927 | |
Johannes Weiner | 0db1529 | 2016-01-20 15:02:50 -0800 | [diff] [blame] | 2928 | if (!memcg->tcpmem_active) { |
Vladimir Davydov | d55f90b | 2016-01-20 15:02:44 -0800 | [diff] [blame] | 2929 | /* |
| 2930 | * The active flag needs to be written after the static_key |
| 2931 | * update. This is what guarantees that the socket activation |
Johannes Weiner | 2d75807 | 2016-10-07 17:00:58 -0700 | [diff] [blame] | 2932 | * function is the last one to run. See mem_cgroup_sk_alloc() |
| 2933 | * for details, and note that we don't mark any socket as |
| 2934 | * belonging to this memcg until that flag is up. |
Vladimir Davydov | d55f90b | 2016-01-20 15:02:44 -0800 | [diff] [blame] | 2935 | * |
| 2936 | * We need to do this, because static_keys will span multiple |
| 2937 | * sites, but we can't control their order. If we mark a socket |
| 2938 | * as accounted, but the accounting functions are not patched in |
| 2939 | * yet, we'll lose accounting. |
| 2940 | * |
Johannes Weiner | 2d75807 | 2016-10-07 17:00:58 -0700 | [diff] [blame] | 2941 | * We never race with the readers in mem_cgroup_sk_alloc(), |
Vladimir Davydov | d55f90b | 2016-01-20 15:02:44 -0800 | [diff] [blame] | 2942 | * because when this value change, the code to process it is not |
| 2943 | * patched in yet. |
| 2944 | */ |
| 2945 | static_branch_inc(&memcg_sockets_enabled_key); |
Johannes Weiner | 0db1529 | 2016-01-20 15:02:50 -0800 | [diff] [blame] | 2946 | memcg->tcpmem_active = true; |
Vladimir Davydov | d55f90b | 2016-01-20 15:02:44 -0800 | [diff] [blame] | 2947 | } |
| 2948 | out: |
| 2949 | mutex_unlock(&memcg_limit_mutex); |
| 2950 | return ret; |
| 2951 | } |
Vladimir Davydov | d55f90b | 2016-01-20 15:02:44 -0800 | [diff] [blame] | 2952 | |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 2953 | /* |
| 2954 | * The user of this function is... |
| 2955 | * RES_LIMIT. |
| 2956 | */ |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 2957 | static ssize_t mem_cgroup_write(struct kernfs_open_file *of, |
| 2958 | char *buf, size_t nbytes, loff_t off) |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 2959 | { |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 2960 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2961 | unsigned long nr_pages; |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 2962 | int ret; |
| 2963 | |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 2964 | buf = strstrip(buf); |
Johannes Weiner | 650c5e5 | 2015-02-11 15:26:03 -0800 | [diff] [blame] | 2965 | ret = page_counter_memparse(buf, "-1", &nr_pages); |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2966 | if (ret) |
| 2967 | return ret; |
Tejun Heo | af36f90 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 2968 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2969 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 2970 | case RES_LIMIT: |
Balbir Singh | 4b3bde4 | 2009-09-23 15:56:32 -0700 | [diff] [blame] | 2971 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
| 2972 | ret = -EINVAL; |
| 2973 | break; |
| 2974 | } |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2975 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
| 2976 | case _MEM: |
| 2977 | ret = mem_cgroup_resize_limit(memcg, nr_pages); |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 2978 | break; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2979 | case _MEMSWAP: |
| 2980 | ret = mem_cgroup_resize_memsw_limit(memcg, nr_pages); |
| 2981 | break; |
| 2982 | case _KMEM: |
| 2983 | ret = memcg_update_kmem_limit(memcg, nr_pages); |
| 2984 | break; |
Vladimir Davydov | d55f90b | 2016-01-20 15:02:44 -0800 | [diff] [blame] | 2985 | case _TCP: |
| 2986 | ret = memcg_update_tcp_limit(memcg, nr_pages); |
| 2987 | break; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2988 | } |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 2989 | break; |
Balbir Singh | 296c81d | 2009-09-23 15:56:36 -0700 | [diff] [blame] | 2990 | case RES_SOFT_LIMIT: |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2991 | memcg->soft_limit = nr_pages; |
| 2992 | ret = 0; |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 2993 | break; |
| 2994 | } |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 2995 | return ret ?: nbytes; |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 2996 | } |
| 2997 | |
Tejun Heo | 6770c64 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 2998 | static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf, |
| 2999 | size_t nbytes, loff_t off) |
Pavel Emelyanov | c84872e | 2008-04-29 01:00:17 -0700 | [diff] [blame] | 3000 | { |
Tejun Heo | 6770c64 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 3001 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3002 | struct page_counter *counter; |
Pavel Emelyanov | c84872e | 2008-04-29 01:00:17 -0700 | [diff] [blame] | 3003 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3004 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
| 3005 | case _MEM: |
| 3006 | counter = &memcg->memory; |
| 3007 | break; |
| 3008 | case _MEMSWAP: |
| 3009 | counter = &memcg->memsw; |
| 3010 | break; |
| 3011 | case _KMEM: |
| 3012 | counter = &memcg->kmem; |
| 3013 | break; |
Vladimir Davydov | d55f90b | 2016-01-20 15:02:44 -0800 | [diff] [blame] | 3014 | case _TCP: |
Johannes Weiner | 0db1529 | 2016-01-20 15:02:50 -0800 | [diff] [blame] | 3015 | counter = &memcg->tcpmem; |
Vladimir Davydov | d55f90b | 2016-01-20 15:02:44 -0800 | [diff] [blame] | 3016 | break; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3017 | default: |
| 3018 | BUG(); |
| 3019 | } |
Tejun Heo | af36f90 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 3020 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3021 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
Pavel Emelyanov | 29f2a4d | 2008-04-29 01:00:21 -0700 | [diff] [blame] | 3022 | case RES_MAX_USAGE: |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3023 | page_counter_reset_watermark(counter); |
Pavel Emelyanov | 29f2a4d | 2008-04-29 01:00:21 -0700 | [diff] [blame] | 3024 | break; |
| 3025 | case RES_FAILCNT: |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3026 | counter->failcnt = 0; |
Pavel Emelyanov | 29f2a4d | 2008-04-29 01:00:21 -0700 | [diff] [blame] | 3027 | break; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3028 | default: |
| 3029 | BUG(); |
Pavel Emelyanov | 29f2a4d | 2008-04-29 01:00:21 -0700 | [diff] [blame] | 3030 | } |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 3031 | |
Tejun Heo | 6770c64 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 3032 | return nbytes; |
Pavel Emelyanov | c84872e | 2008-04-29 01:00:17 -0700 | [diff] [blame] | 3033 | } |
| 3034 | |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 3035 | static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css, |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 3036 | struct cftype *cft) |
| 3037 | { |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 3038 | return mem_cgroup_from_css(css)->move_charge_at_immigrate; |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 3039 | } |
| 3040 | |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 3041 | #ifdef CONFIG_MMU |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 3042 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 3043 | struct cftype *cft, u64 val) |
| 3044 | { |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 3045 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 3046 | |
Johannes Weiner | 1dfab5a | 2015-02-11 15:26:09 -0800 | [diff] [blame] | 3047 | if (val & ~MOVE_MASK) |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 3048 | return -EINVAL; |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 3049 | |
Glauber Costa | ee5e847 | 2013-02-22 16:34:50 -0800 | [diff] [blame] | 3050 | /* |
| 3051 | * No kind of locking is needed in here, because ->can_attach() will |
| 3052 | * check this value once in the beginning of the process, and then carry |
| 3053 | * on with stale data. This means that changes to this value will only |
| 3054 | * affect task migrations starting after the change. |
| 3055 | */ |
| 3056 | memcg->move_charge_at_immigrate = val; |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 3057 | return 0; |
| 3058 | } |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 3059 | #else |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 3060 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 3061 | struct cftype *cft, u64 val) |
| 3062 | { |
| 3063 | return -ENOSYS; |
| 3064 | } |
| 3065 | #endif |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 3066 | |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 3067 | #ifdef CONFIG_NUMA |
Tejun Heo | 2da8ca8 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 3068 | static int memcg_numa_stat_show(struct seq_file *m, void *v) |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 3069 | { |
Greg Thelen | 25485de | 2013-11-12 15:07:40 -0800 | [diff] [blame] | 3070 | struct numa_stat { |
| 3071 | const char *name; |
| 3072 | unsigned int lru_mask; |
| 3073 | }; |
| 3074 | |
| 3075 | static const struct numa_stat stats[] = { |
| 3076 | { "total", LRU_ALL }, |
| 3077 | { "file", LRU_ALL_FILE }, |
| 3078 | { "anon", LRU_ALL_ANON }, |
| 3079 | { "unevictable", BIT(LRU_UNEVICTABLE) }, |
| 3080 | }; |
| 3081 | const struct numa_stat *stat; |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 3082 | int nid; |
Greg Thelen | 25485de | 2013-11-12 15:07:40 -0800 | [diff] [blame] | 3083 | unsigned long nr; |
Tejun Heo | 2da8ca8 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 3084 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 3085 | |
Greg Thelen | 25485de | 2013-11-12 15:07:40 -0800 | [diff] [blame] | 3086 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
| 3087 | nr = mem_cgroup_nr_lru_pages(memcg, stat->lru_mask); |
| 3088 | seq_printf(m, "%s=%lu", stat->name, nr); |
| 3089 | for_each_node_state(nid, N_MEMORY) { |
| 3090 | nr = mem_cgroup_node_nr_lru_pages(memcg, nid, |
| 3091 | stat->lru_mask); |
| 3092 | seq_printf(m, " N%d=%lu", nid, nr); |
| 3093 | } |
| 3094 | seq_putc(m, '\n'); |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 3095 | } |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 3096 | |
Ying Han | 071aee1 | 2013-11-12 15:07:41 -0800 | [diff] [blame] | 3097 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
| 3098 | struct mem_cgroup *iter; |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 3099 | |
Ying Han | 071aee1 | 2013-11-12 15:07:41 -0800 | [diff] [blame] | 3100 | nr = 0; |
| 3101 | for_each_mem_cgroup_tree(iter, memcg) |
| 3102 | nr += mem_cgroup_nr_lru_pages(iter, stat->lru_mask); |
| 3103 | seq_printf(m, "hierarchical_%s=%lu", stat->name, nr); |
| 3104 | for_each_node_state(nid, N_MEMORY) { |
| 3105 | nr = 0; |
| 3106 | for_each_mem_cgroup_tree(iter, memcg) |
| 3107 | nr += mem_cgroup_node_nr_lru_pages( |
| 3108 | iter, nid, stat->lru_mask); |
| 3109 | seq_printf(m, " N%d=%lu", nid, nr); |
| 3110 | } |
| 3111 | seq_putc(m, '\n'); |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 3112 | } |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 3113 | |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 3114 | return 0; |
| 3115 | } |
| 3116 | #endif /* CONFIG_NUMA */ |
| 3117 | |
Johannes Weiner | df0e53d | 2017-05-03 14:55:10 -0700 | [diff] [blame] | 3118 | /* Universal VM events cgroup1 shows, original sort order */ |
| 3119 | unsigned int memcg1_events[] = { |
| 3120 | PGPGIN, |
| 3121 | PGPGOUT, |
| 3122 | PGFAULT, |
| 3123 | PGMAJFAULT, |
| 3124 | }; |
| 3125 | |
| 3126 | static const char *const memcg1_event_names[] = { |
| 3127 | "pgpgin", |
| 3128 | "pgpgout", |
| 3129 | "pgfault", |
| 3130 | "pgmajfault", |
| 3131 | }; |
| 3132 | |
Tejun Heo | 2da8ca8 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 3133 | static int memcg_stat_show(struct seq_file *m, void *v) |
KAMEZAWA Hiroyuki | d2ceb9b | 2008-02-07 00:14:25 -0800 | [diff] [blame] | 3134 | { |
Tejun Heo | 2da8ca8 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 3135 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3136 | unsigned long memory, memsw; |
Johannes Weiner | af7c4b0 | 2012-05-29 15:07:08 -0700 | [diff] [blame] | 3137 | struct mem_cgroup *mi; |
| 3138 | unsigned int i; |
KAMEZAWA Hiroyuki | d2ceb9b | 2008-02-07 00:14:25 -0800 | [diff] [blame] | 3139 | |
Johannes Weiner | 71cd311 | 2017-05-03 14:55:13 -0700 | [diff] [blame] | 3140 | BUILD_BUG_ON(ARRAY_SIZE(memcg1_stat_names) != ARRAY_SIZE(memcg1_stats)); |
Rickard Strandqvist | 70bc068 | 2014-12-12 16:56:41 -0800 | [diff] [blame] | 3141 | BUILD_BUG_ON(ARRAY_SIZE(mem_cgroup_lru_names) != NR_LRU_LISTS); |
| 3142 | |
Johannes Weiner | 71cd311 | 2017-05-03 14:55:13 -0700 | [diff] [blame] | 3143 | for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) { |
| 3144 | if (memcg1_stats[i] == MEMCG_SWAP && !do_memsw_account()) |
Daisuke Nishimura | 1dd3a27 | 2009-09-23 15:56:43 -0700 | [diff] [blame] | 3145 | continue; |
Johannes Weiner | 71cd311 | 2017-05-03 14:55:13 -0700 | [diff] [blame] | 3146 | seq_printf(m, "%s %lu\n", memcg1_stat_names[i], |
Johannes Weiner | ccda7f4 | 2017-05-03 14:55:16 -0700 | [diff] [blame] | 3147 | memcg_page_state(memcg, memcg1_stats[i]) * |
Johannes Weiner | 71cd311 | 2017-05-03 14:55:13 -0700 | [diff] [blame] | 3148 | PAGE_SIZE); |
Daisuke Nishimura | 1dd3a27 | 2009-09-23 15:56:43 -0700 | [diff] [blame] | 3149 | } |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 3150 | |
Johannes Weiner | df0e53d | 2017-05-03 14:55:10 -0700 | [diff] [blame] | 3151 | for (i = 0; i < ARRAY_SIZE(memcg1_events); i++) |
| 3152 | seq_printf(m, "%s %lu\n", memcg1_event_names[i], |
Johannes Weiner | ccda7f4 | 2017-05-03 14:55:16 -0700 | [diff] [blame] | 3153 | memcg_sum_events(memcg, memcg1_events[i])); |
Johannes Weiner | af7c4b0 | 2012-05-29 15:07:08 -0700 | [diff] [blame] | 3154 | |
| 3155 | for (i = 0; i < NR_LRU_LISTS; i++) |
| 3156 | seq_printf(m, "%s %lu\n", mem_cgroup_lru_names[i], |
| 3157 | mem_cgroup_nr_lru_pages(memcg, BIT(i)) * PAGE_SIZE); |
| 3158 | |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 3159 | /* Hierarchical information */ |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3160 | memory = memsw = PAGE_COUNTER_MAX; |
| 3161 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) { |
| 3162 | memory = min(memory, mi->memory.limit); |
| 3163 | memsw = min(memsw, mi->memsw.limit); |
KAMEZAWA Hiroyuki | fee7b54 | 2009-01-07 18:08:26 -0800 | [diff] [blame] | 3164 | } |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3165 | seq_printf(m, "hierarchical_memory_limit %llu\n", |
| 3166 | (u64)memory * PAGE_SIZE); |
Johannes Weiner | 7941d21 | 2016-01-14 15:21:23 -0800 | [diff] [blame] | 3167 | if (do_memsw_account()) |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3168 | seq_printf(m, "hierarchical_memsw_limit %llu\n", |
| 3169 | (u64)memsw * PAGE_SIZE); |
KOSAKI Motohiro | 7f016ee | 2009-01-07 18:08:22 -0800 | [diff] [blame] | 3170 | |
Johannes Weiner | 71cd311 | 2017-05-03 14:55:13 -0700 | [diff] [blame] | 3171 | for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) { |
Greg Thelen | 484ebb3 | 2015-10-01 15:37:05 -0700 | [diff] [blame] | 3172 | unsigned long long val = 0; |
Johannes Weiner | af7c4b0 | 2012-05-29 15:07:08 -0700 | [diff] [blame] | 3173 | |
Johannes Weiner | 71cd311 | 2017-05-03 14:55:13 -0700 | [diff] [blame] | 3174 | if (memcg1_stats[i] == MEMCG_SWAP && !do_memsw_account()) |
Daisuke Nishimura | 1dd3a27 | 2009-09-23 15:56:43 -0700 | [diff] [blame] | 3175 | continue; |
Johannes Weiner | af7c4b0 | 2012-05-29 15:07:08 -0700 | [diff] [blame] | 3176 | for_each_mem_cgroup_tree(mi, memcg) |
Johannes Weiner | ccda7f4 | 2017-05-03 14:55:16 -0700 | [diff] [blame] | 3177 | val += memcg_page_state(mi, memcg1_stats[i]) * |
Johannes Weiner | 71cd311 | 2017-05-03 14:55:13 -0700 | [diff] [blame] | 3178 | PAGE_SIZE; |
| 3179 | seq_printf(m, "total_%s %llu\n", memcg1_stat_names[i], val); |
Johannes Weiner | af7c4b0 | 2012-05-29 15:07:08 -0700 | [diff] [blame] | 3180 | } |
| 3181 | |
Johannes Weiner | df0e53d | 2017-05-03 14:55:10 -0700 | [diff] [blame] | 3182 | for (i = 0; i < ARRAY_SIZE(memcg1_events); i++) { |
Johannes Weiner | af7c4b0 | 2012-05-29 15:07:08 -0700 | [diff] [blame] | 3183 | unsigned long long val = 0; |
| 3184 | |
| 3185 | for_each_mem_cgroup_tree(mi, memcg) |
Johannes Weiner | ccda7f4 | 2017-05-03 14:55:16 -0700 | [diff] [blame] | 3186 | val += memcg_sum_events(mi, memcg1_events[i]); |
Johannes Weiner | df0e53d | 2017-05-03 14:55:10 -0700 | [diff] [blame] | 3187 | seq_printf(m, "total_%s %llu\n", memcg1_event_names[i], val); |
Johannes Weiner | af7c4b0 | 2012-05-29 15:07:08 -0700 | [diff] [blame] | 3188 | } |
| 3189 | |
| 3190 | for (i = 0; i < NR_LRU_LISTS; i++) { |
| 3191 | unsigned long long val = 0; |
| 3192 | |
| 3193 | for_each_mem_cgroup_tree(mi, memcg) |
| 3194 | val += mem_cgroup_nr_lru_pages(mi, BIT(i)) * PAGE_SIZE; |
| 3195 | seq_printf(m, "total_%s %llu\n", mem_cgroup_lru_names[i], val); |
Daisuke Nishimura | 1dd3a27 | 2009-09-23 15:56:43 -0700 | [diff] [blame] | 3196 | } |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 3197 | |
KOSAKI Motohiro | 7f016ee | 2009-01-07 18:08:22 -0800 | [diff] [blame] | 3198 | #ifdef CONFIG_DEBUG_VM |
KOSAKI Motohiro | 7f016ee | 2009-01-07 18:08:22 -0800 | [diff] [blame] | 3199 | { |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 3200 | pg_data_t *pgdat; |
| 3201 | struct mem_cgroup_per_node *mz; |
Hugh Dickins | 89abfab | 2012-05-29 15:06:53 -0700 | [diff] [blame] | 3202 | struct zone_reclaim_stat *rstat; |
KOSAKI Motohiro | 7f016ee | 2009-01-07 18:08:22 -0800 | [diff] [blame] | 3203 | unsigned long recent_rotated[2] = {0, 0}; |
| 3204 | unsigned long recent_scanned[2] = {0, 0}; |
| 3205 | |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 3206 | for_each_online_pgdat(pgdat) { |
| 3207 | mz = mem_cgroup_nodeinfo(memcg, pgdat->node_id); |
| 3208 | rstat = &mz->lruvec.reclaim_stat; |
KOSAKI Motohiro | 7f016ee | 2009-01-07 18:08:22 -0800 | [diff] [blame] | 3209 | |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 3210 | recent_rotated[0] += rstat->recent_rotated[0]; |
| 3211 | recent_rotated[1] += rstat->recent_rotated[1]; |
| 3212 | recent_scanned[0] += rstat->recent_scanned[0]; |
| 3213 | recent_scanned[1] += rstat->recent_scanned[1]; |
| 3214 | } |
Johannes Weiner | 78ccf5b | 2012-05-29 15:07:06 -0700 | [diff] [blame] | 3215 | seq_printf(m, "recent_rotated_anon %lu\n", recent_rotated[0]); |
| 3216 | seq_printf(m, "recent_rotated_file %lu\n", recent_rotated[1]); |
| 3217 | seq_printf(m, "recent_scanned_anon %lu\n", recent_scanned[0]); |
| 3218 | seq_printf(m, "recent_scanned_file %lu\n", recent_scanned[1]); |
KOSAKI Motohiro | 7f016ee | 2009-01-07 18:08:22 -0800 | [diff] [blame] | 3219 | } |
| 3220 | #endif |
| 3221 | |
KAMEZAWA Hiroyuki | d2ceb9b | 2008-02-07 00:14:25 -0800 | [diff] [blame] | 3222 | return 0; |
| 3223 | } |
| 3224 | |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 3225 | static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css, |
| 3226 | struct cftype *cft) |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 3227 | { |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 3228 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 3229 | |
KAMEZAWA Hiroyuki | 1f4c025 | 2011-07-26 16:08:21 -0700 | [diff] [blame] | 3230 | return mem_cgroup_swappiness(memcg); |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 3231 | } |
| 3232 | |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 3233 | static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css, |
| 3234 | struct cftype *cft, u64 val) |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 3235 | { |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 3236 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
Li Zefan | 068b38c | 2009-01-15 13:51:26 -0800 | [diff] [blame] | 3237 | |
Johannes Weiner | 3dae7fe | 2014-06-04 16:07:01 -0700 | [diff] [blame] | 3238 | if (val > 100) |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 3239 | return -EINVAL; |
| 3240 | |
Linus Torvalds | 14208b0 | 2014-06-09 15:03:33 -0700 | [diff] [blame] | 3241 | if (css->parent) |
Johannes Weiner | 3dae7fe | 2014-06-04 16:07:01 -0700 | [diff] [blame] | 3242 | memcg->swappiness = val; |
| 3243 | else |
| 3244 | vm_swappiness = val; |
Li Zefan | 068b38c | 2009-01-15 13:51:26 -0800 | [diff] [blame] | 3245 | |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 3246 | return 0; |
| 3247 | } |
| 3248 | |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3249 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
| 3250 | { |
| 3251 | struct mem_cgroup_threshold_ary *t; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3252 | unsigned long usage; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3253 | int i; |
| 3254 | |
| 3255 | rcu_read_lock(); |
| 3256 | if (!swap) |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3257 | t = rcu_dereference(memcg->thresholds.primary); |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3258 | else |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3259 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3260 | |
| 3261 | if (!t) |
| 3262 | goto unlock; |
| 3263 | |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 3264 | usage = mem_cgroup_usage(memcg, swap); |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3265 | |
| 3266 | /* |
Sha Zhengju | 748dad3 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 3267 | * current_threshold points to threshold just below or equal to usage. |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3268 | * If it's not true, a threshold was crossed after last |
| 3269 | * call of __mem_cgroup_threshold(). |
| 3270 | */ |
Phil Carmody | 5407a56 | 2010-05-26 14:42:42 -0700 | [diff] [blame] | 3271 | i = t->current_threshold; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3272 | |
| 3273 | /* |
| 3274 | * Iterate backward over array of thresholds starting from |
| 3275 | * current_threshold and check if a threshold is crossed. |
| 3276 | * If none of thresholds below usage is crossed, we read |
| 3277 | * only one element of the array here. |
| 3278 | */ |
| 3279 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) |
| 3280 | eventfd_signal(t->entries[i].eventfd, 1); |
| 3281 | |
| 3282 | /* i = current_threshold + 1 */ |
| 3283 | i++; |
| 3284 | |
| 3285 | /* |
| 3286 | * Iterate forward over array of thresholds starting from |
| 3287 | * current_threshold+1 and check if a threshold is crossed. |
| 3288 | * If none of thresholds above usage is crossed, we read |
| 3289 | * only one element of the array here. |
| 3290 | */ |
| 3291 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) |
| 3292 | eventfd_signal(t->entries[i].eventfd, 1); |
| 3293 | |
| 3294 | /* Update current_threshold */ |
Phil Carmody | 5407a56 | 2010-05-26 14:42:42 -0700 | [diff] [blame] | 3295 | t->current_threshold = i - 1; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3296 | unlock: |
| 3297 | rcu_read_unlock(); |
| 3298 | } |
| 3299 | |
| 3300 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) |
| 3301 | { |
Kirill A. Shutemov | ad4ca5f | 2010-10-07 12:59:27 -0700 | [diff] [blame] | 3302 | while (memcg) { |
| 3303 | __mem_cgroup_threshold(memcg, false); |
Johannes Weiner | 7941d21 | 2016-01-14 15:21:23 -0800 | [diff] [blame] | 3304 | if (do_memsw_account()) |
Kirill A. Shutemov | ad4ca5f | 2010-10-07 12:59:27 -0700 | [diff] [blame] | 3305 | __mem_cgroup_threshold(memcg, true); |
| 3306 | |
| 3307 | memcg = parent_mem_cgroup(memcg); |
| 3308 | } |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3309 | } |
| 3310 | |
| 3311 | static int compare_thresholds(const void *a, const void *b) |
| 3312 | { |
| 3313 | const struct mem_cgroup_threshold *_a = a; |
| 3314 | const struct mem_cgroup_threshold *_b = b; |
| 3315 | |
Greg Thelen | 2bff24a | 2013-09-11 14:23:08 -0700 | [diff] [blame] | 3316 | if (_a->threshold > _b->threshold) |
| 3317 | return 1; |
| 3318 | |
| 3319 | if (_a->threshold < _b->threshold) |
| 3320 | return -1; |
| 3321 | |
| 3322 | return 0; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3323 | } |
| 3324 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3325 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3326 | { |
| 3327 | struct mem_cgroup_eventfd_list *ev; |
| 3328 | |
Michal Hocko | 2bcf2e9 | 2014-07-30 16:08:33 -0700 | [diff] [blame] | 3329 | spin_lock(&memcg_oom_lock); |
| 3330 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3331 | list_for_each_entry(ev, &memcg->oom_notify, list) |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3332 | eventfd_signal(ev->eventfd, 1); |
Michal Hocko | 2bcf2e9 | 2014-07-30 16:08:33 -0700 | [diff] [blame] | 3333 | |
| 3334 | spin_unlock(&memcg_oom_lock); |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3335 | return 0; |
| 3336 | } |
| 3337 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3338 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3339 | { |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 3340 | struct mem_cgroup *iter; |
| 3341 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3342 | for_each_mem_cgroup_tree(iter, memcg) |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 3343 | mem_cgroup_oom_notify_cb(iter); |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3344 | } |
| 3345 | |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3346 | static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3347 | struct eventfd_ctx *eventfd, const char *args, enum res_type type) |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3348 | { |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3349 | struct mem_cgroup_thresholds *thresholds; |
| 3350 | struct mem_cgroup_threshold_ary *new; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3351 | unsigned long threshold; |
| 3352 | unsigned long usage; |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3353 | int i, size, ret; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3354 | |
Johannes Weiner | 650c5e5 | 2015-02-11 15:26:03 -0800 | [diff] [blame] | 3355 | ret = page_counter_memparse(args, "-1", &threshold); |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3356 | if (ret) |
| 3357 | return ret; |
| 3358 | |
| 3359 | mutex_lock(&memcg->thresholds_lock); |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3360 | |
Johannes Weiner | 05b8430 | 2014-08-06 16:05:59 -0700 | [diff] [blame] | 3361 | if (type == _MEM) { |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3362 | thresholds = &memcg->thresholds; |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 3363 | usage = mem_cgroup_usage(memcg, false); |
Johannes Weiner | 05b8430 | 2014-08-06 16:05:59 -0700 | [diff] [blame] | 3364 | } else if (type == _MEMSWAP) { |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3365 | thresholds = &memcg->memsw_thresholds; |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 3366 | usage = mem_cgroup_usage(memcg, true); |
Johannes Weiner | 05b8430 | 2014-08-06 16:05:59 -0700 | [diff] [blame] | 3367 | } else |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3368 | BUG(); |
| 3369 | |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3370 | /* Check if a threshold crossed before adding a new one */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3371 | if (thresholds->primary) |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3372 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
| 3373 | |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3374 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3375 | |
| 3376 | /* Allocate memory for new array of thresholds */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3377 | new = kmalloc(sizeof(*new) + size * sizeof(struct mem_cgroup_threshold), |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3378 | GFP_KERNEL); |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3379 | if (!new) { |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3380 | ret = -ENOMEM; |
| 3381 | goto unlock; |
| 3382 | } |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3383 | new->size = size; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3384 | |
| 3385 | /* Copy thresholds (if any) to new array */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3386 | if (thresholds->primary) { |
| 3387 | memcpy(new->entries, thresholds->primary->entries, (size - 1) * |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3388 | sizeof(struct mem_cgroup_threshold)); |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3389 | } |
| 3390 | |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3391 | /* Add new threshold */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3392 | new->entries[size - 1].eventfd = eventfd; |
| 3393 | new->entries[size - 1].threshold = threshold; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3394 | |
| 3395 | /* Sort thresholds. Registering of new threshold isn't time-critical */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3396 | sort(new->entries, size, sizeof(struct mem_cgroup_threshold), |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3397 | compare_thresholds, NULL); |
| 3398 | |
| 3399 | /* Find current threshold */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3400 | new->current_threshold = -1; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3401 | for (i = 0; i < size; i++) { |
Sha Zhengju | 748dad3 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 3402 | if (new->entries[i].threshold <= usage) { |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3403 | /* |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3404 | * new->current_threshold will not be used until |
| 3405 | * rcu_assign_pointer(), so it's safe to increment |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3406 | * it here. |
| 3407 | */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3408 | ++new->current_threshold; |
Sha Zhengju | 748dad3 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 3409 | } else |
| 3410 | break; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3411 | } |
| 3412 | |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3413 | /* Free old spare buffer and save old primary buffer as spare */ |
| 3414 | kfree(thresholds->spare); |
| 3415 | thresholds->spare = thresholds->primary; |
| 3416 | |
| 3417 | rcu_assign_pointer(thresholds->primary, new); |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3418 | |
Kirill A. Shutemov | 907860e | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 3419 | /* To be sure that nobody uses thresholds */ |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3420 | synchronize_rcu(); |
| 3421 | |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3422 | unlock: |
| 3423 | mutex_unlock(&memcg->thresholds_lock); |
| 3424 | |
| 3425 | return ret; |
| 3426 | } |
| 3427 | |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3428 | static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3429 | struct eventfd_ctx *eventfd, const char *args) |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3430 | { |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3431 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM); |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3432 | } |
| 3433 | |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3434 | static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg, |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3435 | struct eventfd_ctx *eventfd, const char *args) |
| 3436 | { |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3437 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP); |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3438 | } |
| 3439 | |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3440 | static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3441 | struct eventfd_ctx *eventfd, enum res_type type) |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3442 | { |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3443 | struct mem_cgroup_thresholds *thresholds; |
| 3444 | struct mem_cgroup_threshold_ary *new; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3445 | unsigned long usage; |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3446 | int i, j, size; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3447 | |
| 3448 | mutex_lock(&memcg->thresholds_lock); |
Johannes Weiner | 05b8430 | 2014-08-06 16:05:59 -0700 | [diff] [blame] | 3449 | |
| 3450 | if (type == _MEM) { |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3451 | thresholds = &memcg->thresholds; |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 3452 | usage = mem_cgroup_usage(memcg, false); |
Johannes Weiner | 05b8430 | 2014-08-06 16:05:59 -0700 | [diff] [blame] | 3453 | } else if (type == _MEMSWAP) { |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3454 | thresholds = &memcg->memsw_thresholds; |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 3455 | usage = mem_cgroup_usage(memcg, true); |
Johannes Weiner | 05b8430 | 2014-08-06 16:05:59 -0700 | [diff] [blame] | 3456 | } else |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3457 | BUG(); |
| 3458 | |
Anton Vorontsov | 371528c | 2012-02-24 05:14:46 +0400 | [diff] [blame] | 3459 | if (!thresholds->primary) |
| 3460 | goto unlock; |
| 3461 | |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3462 | /* Check if a threshold crossed before removing */ |
| 3463 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
| 3464 | |
| 3465 | /* Calculate new number of threshold */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3466 | size = 0; |
| 3467 | for (i = 0; i < thresholds->primary->size; i++) { |
| 3468 | if (thresholds->primary->entries[i].eventfd != eventfd) |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3469 | size++; |
| 3470 | } |
| 3471 | |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3472 | new = thresholds->spare; |
Kirill A. Shutemov | 907860e | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 3473 | |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3474 | /* Set thresholds array to NULL if we don't have thresholds */ |
| 3475 | if (!size) { |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3476 | kfree(new); |
| 3477 | new = NULL; |
Kirill A. Shutemov | 907860e | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 3478 | goto swap_buffers; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3479 | } |
| 3480 | |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3481 | new->size = size; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3482 | |
| 3483 | /* Copy thresholds and find current threshold */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3484 | new->current_threshold = -1; |
| 3485 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { |
| 3486 | if (thresholds->primary->entries[i].eventfd == eventfd) |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3487 | continue; |
| 3488 | |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3489 | new->entries[j] = thresholds->primary->entries[i]; |
Sha Zhengju | 748dad3 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 3490 | if (new->entries[j].threshold <= usage) { |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3491 | /* |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3492 | * new->current_threshold will not be used |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3493 | * until rcu_assign_pointer(), so it's safe to increment |
| 3494 | * it here. |
| 3495 | */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3496 | ++new->current_threshold; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3497 | } |
| 3498 | j++; |
| 3499 | } |
| 3500 | |
Kirill A. Shutemov | 907860e | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 3501 | swap_buffers: |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3502 | /* Swap primary and spare array */ |
| 3503 | thresholds->spare = thresholds->primary; |
Sha Zhengju | 8c75776 | 2012-05-10 13:01:45 -0700 | [diff] [blame] | 3504 | |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 3505 | rcu_assign_pointer(thresholds->primary, new); |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3506 | |
Kirill A. Shutemov | 907860e | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 3507 | /* To be sure that nobody uses thresholds */ |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3508 | synchronize_rcu(); |
Martijn Coenen | 6611d8d | 2016-01-15 16:57:49 -0800 | [diff] [blame] | 3509 | |
| 3510 | /* If all events are unregistered, free the spare array */ |
| 3511 | if (!new) { |
| 3512 | kfree(thresholds->spare); |
| 3513 | thresholds->spare = NULL; |
| 3514 | } |
Anton Vorontsov | 371528c | 2012-02-24 05:14:46 +0400 | [diff] [blame] | 3515 | unlock: |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3516 | mutex_unlock(&memcg->thresholds_lock); |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 3517 | } |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 3518 | |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3519 | static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3520 | struct eventfd_ctx *eventfd) |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3521 | { |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3522 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM); |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3523 | } |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3524 | |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3525 | static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3526 | struct eventfd_ctx *eventfd) |
| 3527 | { |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3528 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP); |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3529 | } |
| 3530 | |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3531 | static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg, |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3532 | struct eventfd_ctx *eventfd, const char *args) |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3533 | { |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3534 | struct mem_cgroup_eventfd_list *event; |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3535 | |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3536 | event = kmalloc(sizeof(*event), GFP_KERNEL); |
| 3537 | if (!event) |
| 3538 | return -ENOMEM; |
| 3539 | |
Michal Hocko | 1af8efe | 2011-07-26 16:08:24 -0700 | [diff] [blame] | 3540 | spin_lock(&memcg_oom_lock); |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3541 | |
| 3542 | event->eventfd = eventfd; |
| 3543 | list_add(&event->list, &memcg->oom_notify); |
| 3544 | |
| 3545 | /* already in OOM ? */ |
Tejun Heo | c2b42d3 | 2015-06-24 16:58:23 -0700 | [diff] [blame] | 3546 | if (memcg->under_oom) |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3547 | eventfd_signal(eventfd, 1); |
Michal Hocko | 1af8efe | 2011-07-26 16:08:24 -0700 | [diff] [blame] | 3548 | spin_unlock(&memcg_oom_lock); |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3549 | |
| 3550 | return 0; |
| 3551 | } |
| 3552 | |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3553 | static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg, |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3554 | struct eventfd_ctx *eventfd) |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3555 | { |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3556 | struct mem_cgroup_eventfd_list *ev, *tmp; |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3557 | |
Michal Hocko | 1af8efe | 2011-07-26 16:08:24 -0700 | [diff] [blame] | 3558 | spin_lock(&memcg_oom_lock); |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3559 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3560 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3561 | if (ev->eventfd == eventfd) { |
| 3562 | list_del(&ev->list); |
| 3563 | kfree(ev); |
| 3564 | } |
| 3565 | } |
| 3566 | |
Michal Hocko | 1af8efe | 2011-07-26 16:08:24 -0700 | [diff] [blame] | 3567 | spin_unlock(&memcg_oom_lock); |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3568 | } |
| 3569 | |
Tejun Heo | 2da8ca8 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 3570 | static int mem_cgroup_oom_control_read(struct seq_file *sf, void *v) |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 3571 | { |
Tejun Heo | 2da8ca8 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 3572 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(sf)); |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 3573 | |
Tejun Heo | 791badb | 2013-12-05 12:28:02 -0500 | [diff] [blame] | 3574 | seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable); |
Tejun Heo | c2b42d3 | 2015-06-24 16:58:23 -0700 | [diff] [blame] | 3575 | seq_printf(sf, "under_oom %d\n", (bool)memcg->under_oom); |
Konstantin Khlebnikov | 8e675f7 | 2017-07-06 15:40:28 -0700 | [diff] [blame] | 3576 | seq_printf(sf, "oom_kill %lu\n", memcg_sum_events(memcg, OOM_KILL)); |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 3577 | return 0; |
| 3578 | } |
| 3579 | |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 3580 | static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css, |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 3581 | struct cftype *cft, u64 val) |
| 3582 | { |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 3583 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 3584 | |
| 3585 | /* cannot set to root cgroup and only 0 and 1 are allowed */ |
Linus Torvalds | 14208b0 | 2014-06-09 15:03:33 -0700 | [diff] [blame] | 3586 | if (!css->parent || !((val == 0) || (val == 1))) |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 3587 | return -EINVAL; |
| 3588 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3589 | memcg->oom_kill_disable = val; |
KAMEZAWA Hiroyuki | 4d845eb | 2010-06-29 15:05:18 -0700 | [diff] [blame] | 3590 | if (!val) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 3591 | memcg_oom_recover(memcg); |
Johannes Weiner | 3dae7fe | 2014-06-04 16:07:01 -0700 | [diff] [blame] | 3592 | |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 3593 | return 0; |
| 3594 | } |
| 3595 | |
Tejun Heo | 52ebea7 | 2015-05-22 17:13:37 -0400 | [diff] [blame] | 3596 | #ifdef CONFIG_CGROUP_WRITEBACK |
| 3597 | |
| 3598 | struct list_head *mem_cgroup_cgwb_list(struct mem_cgroup *memcg) |
| 3599 | { |
| 3600 | return &memcg->cgwb_list; |
| 3601 | } |
| 3602 | |
Tejun Heo | 841710a | 2015-05-22 18:23:33 -0400 | [diff] [blame] | 3603 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) |
| 3604 | { |
| 3605 | return wb_domain_init(&memcg->cgwb_domain, gfp); |
| 3606 | } |
| 3607 | |
| 3608 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) |
| 3609 | { |
| 3610 | wb_domain_exit(&memcg->cgwb_domain); |
| 3611 | } |
| 3612 | |
Tejun Heo | 2529bb3 | 2015-05-22 18:23:34 -0400 | [diff] [blame] | 3613 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
| 3614 | { |
| 3615 | wb_domain_size_changed(&memcg->cgwb_domain); |
| 3616 | } |
| 3617 | |
Tejun Heo | 841710a | 2015-05-22 18:23:33 -0400 | [diff] [blame] | 3618 | struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb) |
| 3619 | { |
| 3620 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); |
| 3621 | |
| 3622 | if (!memcg->css.parent) |
| 3623 | return NULL; |
| 3624 | |
| 3625 | return &memcg->cgwb_domain; |
| 3626 | } |
| 3627 | |
Tejun Heo | c2aa723 | 2015-05-22 18:23:35 -0400 | [diff] [blame] | 3628 | /** |
| 3629 | * mem_cgroup_wb_stats - retrieve writeback related stats from its memcg |
| 3630 | * @wb: bdi_writeback in question |
Tejun Heo | c5edf9c | 2015-09-29 13:04:26 -0400 | [diff] [blame] | 3631 | * @pfilepages: out parameter for number of file pages |
| 3632 | * @pheadroom: out parameter for number of allocatable pages according to memcg |
Tejun Heo | c2aa723 | 2015-05-22 18:23:35 -0400 | [diff] [blame] | 3633 | * @pdirty: out parameter for number of dirty pages |
| 3634 | * @pwriteback: out parameter for number of pages under writeback |
| 3635 | * |
Tejun Heo | c5edf9c | 2015-09-29 13:04:26 -0400 | [diff] [blame] | 3636 | * Determine the numbers of file, headroom, dirty, and writeback pages in |
| 3637 | * @wb's memcg. File, dirty and writeback are self-explanatory. Headroom |
| 3638 | * is a bit more involved. |
Tejun Heo | c2aa723 | 2015-05-22 18:23:35 -0400 | [diff] [blame] | 3639 | * |
Tejun Heo | c5edf9c | 2015-09-29 13:04:26 -0400 | [diff] [blame] | 3640 | * A memcg's headroom is "min(max, high) - used". In the hierarchy, the |
| 3641 | * headroom is calculated as the lowest headroom of itself and the |
| 3642 | * ancestors. Note that this doesn't consider the actual amount of |
| 3643 | * available memory in the system. The caller should further cap |
| 3644 | * *@pheadroom accordingly. |
Tejun Heo | c2aa723 | 2015-05-22 18:23:35 -0400 | [diff] [blame] | 3645 | */ |
Tejun Heo | c5edf9c | 2015-09-29 13:04:26 -0400 | [diff] [blame] | 3646 | void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages, |
| 3647 | unsigned long *pheadroom, unsigned long *pdirty, |
| 3648 | unsigned long *pwriteback) |
Tejun Heo | c2aa723 | 2015-05-22 18:23:35 -0400 | [diff] [blame] | 3649 | { |
| 3650 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); |
| 3651 | struct mem_cgroup *parent; |
Tejun Heo | c2aa723 | 2015-05-22 18:23:35 -0400 | [diff] [blame] | 3652 | |
Johannes Weiner | ccda7f4 | 2017-05-03 14:55:16 -0700 | [diff] [blame] | 3653 | *pdirty = memcg_page_state(memcg, NR_FILE_DIRTY); |
Tejun Heo | c2aa723 | 2015-05-22 18:23:35 -0400 | [diff] [blame] | 3654 | |
| 3655 | /* this should eventually include NR_UNSTABLE_NFS */ |
Johannes Weiner | ccda7f4 | 2017-05-03 14:55:16 -0700 | [diff] [blame] | 3656 | *pwriteback = memcg_page_state(memcg, NR_WRITEBACK); |
Tejun Heo | c5edf9c | 2015-09-29 13:04:26 -0400 | [diff] [blame] | 3657 | *pfilepages = mem_cgroup_nr_lru_pages(memcg, (1 << LRU_INACTIVE_FILE) | |
| 3658 | (1 << LRU_ACTIVE_FILE)); |
| 3659 | *pheadroom = PAGE_COUNTER_MAX; |
Tejun Heo | c2aa723 | 2015-05-22 18:23:35 -0400 | [diff] [blame] | 3660 | |
Tejun Heo | c2aa723 | 2015-05-22 18:23:35 -0400 | [diff] [blame] | 3661 | while ((parent = parent_mem_cgroup(memcg))) { |
| 3662 | unsigned long ceiling = min(memcg->memory.limit, memcg->high); |
| 3663 | unsigned long used = page_counter_read(&memcg->memory); |
| 3664 | |
Tejun Heo | c5edf9c | 2015-09-29 13:04:26 -0400 | [diff] [blame] | 3665 | *pheadroom = min(*pheadroom, ceiling - min(ceiling, used)); |
Tejun Heo | c2aa723 | 2015-05-22 18:23:35 -0400 | [diff] [blame] | 3666 | memcg = parent; |
| 3667 | } |
Tejun Heo | c2aa723 | 2015-05-22 18:23:35 -0400 | [diff] [blame] | 3668 | } |
| 3669 | |
Tejun Heo | 841710a | 2015-05-22 18:23:33 -0400 | [diff] [blame] | 3670 | #else /* CONFIG_CGROUP_WRITEBACK */ |
| 3671 | |
| 3672 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) |
| 3673 | { |
| 3674 | return 0; |
| 3675 | } |
| 3676 | |
| 3677 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) |
| 3678 | { |
| 3679 | } |
| 3680 | |
Tejun Heo | 2529bb3 | 2015-05-22 18:23:34 -0400 | [diff] [blame] | 3681 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
| 3682 | { |
| 3683 | } |
| 3684 | |
Tejun Heo | 52ebea7 | 2015-05-22 17:13:37 -0400 | [diff] [blame] | 3685 | #endif /* CONFIG_CGROUP_WRITEBACK */ |
| 3686 | |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 3687 | /* |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 3688 | * DO NOT USE IN NEW FILES. |
| 3689 | * |
| 3690 | * "cgroup.event_control" implementation. |
| 3691 | * |
| 3692 | * This is way over-engineered. It tries to support fully configurable |
| 3693 | * events for each user. Such level of flexibility is completely |
| 3694 | * unnecessary especially in the light of the planned unified hierarchy. |
| 3695 | * |
| 3696 | * Please deprecate this and replace with something simpler if at all |
| 3697 | * possible. |
| 3698 | */ |
| 3699 | |
| 3700 | /* |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 3701 | * Unregister event and free resources. |
| 3702 | * |
| 3703 | * Gets called from workqueue. |
| 3704 | */ |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 3705 | static void memcg_event_remove(struct work_struct *work) |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 3706 | { |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 3707 | struct mem_cgroup_event *event = |
| 3708 | container_of(work, struct mem_cgroup_event, remove); |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3709 | struct mem_cgroup *memcg = event->memcg; |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 3710 | |
| 3711 | remove_wait_queue(event->wqh, &event->wait); |
| 3712 | |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3713 | event->unregister_event(memcg, event->eventfd); |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 3714 | |
| 3715 | /* Notify userspace the event is going away. */ |
| 3716 | eventfd_signal(event->eventfd, 1); |
| 3717 | |
| 3718 | eventfd_ctx_put(event->eventfd); |
| 3719 | kfree(event); |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3720 | css_put(&memcg->css); |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 3721 | } |
| 3722 | |
| 3723 | /* |
| 3724 | * Gets called on POLLHUP on eventfd when user closes it. |
| 3725 | * |
| 3726 | * Called with wqh->lock held and interrupts disabled. |
| 3727 | */ |
Ingo Molnar | ac6424b | 2017-06-20 12:06:13 +0200 | [diff] [blame] | 3728 | static int memcg_event_wake(wait_queue_entry_t *wait, unsigned mode, |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 3729 | int sync, void *key) |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 3730 | { |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 3731 | struct mem_cgroup_event *event = |
| 3732 | container_of(wait, struct mem_cgroup_event, wait); |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3733 | struct mem_cgroup *memcg = event->memcg; |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 3734 | unsigned long flags = (unsigned long)key; |
| 3735 | |
| 3736 | if (flags & POLLHUP) { |
| 3737 | /* |
| 3738 | * If the event has been detached at cgroup removal, we |
| 3739 | * can simply return knowing the other side will cleanup |
| 3740 | * for us. |
| 3741 | * |
| 3742 | * We can't race against event freeing since the other |
| 3743 | * side will require wqh->lock via remove_wait_queue(), |
| 3744 | * which we hold. |
| 3745 | */ |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3746 | spin_lock(&memcg->event_list_lock); |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 3747 | if (!list_empty(&event->list)) { |
| 3748 | list_del_init(&event->list); |
| 3749 | /* |
| 3750 | * We are in atomic context, but cgroup_event_remove() |
| 3751 | * may sleep, so we have to call it in workqueue. |
| 3752 | */ |
| 3753 | schedule_work(&event->remove); |
| 3754 | } |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3755 | spin_unlock(&memcg->event_list_lock); |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 3756 | } |
| 3757 | |
| 3758 | return 0; |
| 3759 | } |
| 3760 | |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 3761 | static void memcg_event_ptable_queue_proc(struct file *file, |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 3762 | wait_queue_head_t *wqh, poll_table *pt) |
| 3763 | { |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 3764 | struct mem_cgroup_event *event = |
| 3765 | container_of(pt, struct mem_cgroup_event, pt); |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 3766 | |
| 3767 | event->wqh = wqh; |
| 3768 | add_wait_queue(wqh, &event->wait); |
| 3769 | } |
| 3770 | |
| 3771 | /* |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 3772 | * DO NOT USE IN NEW FILES. |
| 3773 | * |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 3774 | * Parse input and register new cgroup event handler. |
| 3775 | * |
| 3776 | * Input must be in format '<event_fd> <control_fd> <args>'. |
| 3777 | * Interpretation of args is defined by control file implementation. |
| 3778 | */ |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 3779 | static ssize_t memcg_write_event_control(struct kernfs_open_file *of, |
| 3780 | char *buf, size_t nbytes, loff_t off) |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 3781 | { |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 3782 | struct cgroup_subsys_state *css = of_css(of); |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3783 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 3784 | struct mem_cgroup_event *event; |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 3785 | struct cgroup_subsys_state *cfile_css; |
| 3786 | unsigned int efd, cfd; |
| 3787 | struct fd efile; |
| 3788 | struct fd cfile; |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3789 | const char *name; |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 3790 | char *endp; |
| 3791 | int ret; |
| 3792 | |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 3793 | buf = strstrip(buf); |
| 3794 | |
| 3795 | efd = simple_strtoul(buf, &endp, 10); |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 3796 | if (*endp != ' ') |
| 3797 | return -EINVAL; |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 3798 | buf = endp + 1; |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 3799 | |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 3800 | cfd = simple_strtoul(buf, &endp, 10); |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 3801 | if ((*endp != ' ') && (*endp != '\0')) |
| 3802 | return -EINVAL; |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 3803 | buf = endp + 1; |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 3804 | |
| 3805 | event = kzalloc(sizeof(*event), GFP_KERNEL); |
| 3806 | if (!event) |
| 3807 | return -ENOMEM; |
| 3808 | |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3809 | event->memcg = memcg; |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 3810 | INIT_LIST_HEAD(&event->list); |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 3811 | init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc); |
| 3812 | init_waitqueue_func_entry(&event->wait, memcg_event_wake); |
| 3813 | INIT_WORK(&event->remove, memcg_event_remove); |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 3814 | |
| 3815 | efile = fdget(efd); |
| 3816 | if (!efile.file) { |
| 3817 | ret = -EBADF; |
| 3818 | goto out_kfree; |
| 3819 | } |
| 3820 | |
| 3821 | event->eventfd = eventfd_ctx_fileget(efile.file); |
| 3822 | if (IS_ERR(event->eventfd)) { |
| 3823 | ret = PTR_ERR(event->eventfd); |
| 3824 | goto out_put_efile; |
| 3825 | } |
| 3826 | |
| 3827 | cfile = fdget(cfd); |
| 3828 | if (!cfile.file) { |
| 3829 | ret = -EBADF; |
| 3830 | goto out_put_eventfd; |
| 3831 | } |
| 3832 | |
| 3833 | /* the process need read permission on control file */ |
| 3834 | /* AV: shouldn't we check that it's been opened for read instead? */ |
| 3835 | ret = inode_permission(file_inode(cfile.file), MAY_READ); |
| 3836 | if (ret < 0) |
| 3837 | goto out_put_cfile; |
| 3838 | |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 3839 | /* |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3840 | * Determine the event callbacks and set them in @event. This used |
| 3841 | * to be done via struct cftype but cgroup core no longer knows |
| 3842 | * about these events. The following is crude but the whole thing |
| 3843 | * is for compatibility anyway. |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 3844 | * |
| 3845 | * DO NOT ADD NEW FILES. |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3846 | */ |
Al Viro | b583043 | 2014-10-31 01:22:04 -0400 | [diff] [blame] | 3847 | name = cfile.file->f_path.dentry->d_name.name; |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3848 | |
| 3849 | if (!strcmp(name, "memory.usage_in_bytes")) { |
| 3850 | event->register_event = mem_cgroup_usage_register_event; |
| 3851 | event->unregister_event = mem_cgroup_usage_unregister_event; |
| 3852 | } else if (!strcmp(name, "memory.oom_control")) { |
| 3853 | event->register_event = mem_cgroup_oom_register_event; |
| 3854 | event->unregister_event = mem_cgroup_oom_unregister_event; |
| 3855 | } else if (!strcmp(name, "memory.pressure_level")) { |
| 3856 | event->register_event = vmpressure_register_event; |
| 3857 | event->unregister_event = vmpressure_unregister_event; |
| 3858 | } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) { |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3859 | event->register_event = memsw_cgroup_usage_register_event; |
| 3860 | event->unregister_event = memsw_cgroup_usage_unregister_event; |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3861 | } else { |
| 3862 | ret = -EINVAL; |
| 3863 | goto out_put_cfile; |
| 3864 | } |
| 3865 | |
| 3866 | /* |
Tejun Heo | b5557c4 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 3867 | * Verify @cfile should belong to @css. Also, remaining events are |
| 3868 | * automatically removed on cgroup destruction but the removal is |
| 3869 | * asynchronous, so take an extra ref on @css. |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 3870 | */ |
Al Viro | b583043 | 2014-10-31 01:22:04 -0400 | [diff] [blame] | 3871 | cfile_css = css_tryget_online_from_dir(cfile.file->f_path.dentry->d_parent, |
Tejun Heo | ec903c0 | 2014-05-13 12:11:01 -0400 | [diff] [blame] | 3872 | &memory_cgrp_subsys); |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 3873 | ret = -EINVAL; |
Tejun Heo | 5a17f54 | 2014-02-11 11:52:47 -0500 | [diff] [blame] | 3874 | if (IS_ERR(cfile_css)) |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 3875 | goto out_put_cfile; |
Tejun Heo | 5a17f54 | 2014-02-11 11:52:47 -0500 | [diff] [blame] | 3876 | if (cfile_css != css) { |
| 3877 | css_put(cfile_css); |
| 3878 | goto out_put_cfile; |
| 3879 | } |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 3880 | |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 3881 | ret = event->register_event(memcg, event->eventfd, buf); |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 3882 | if (ret) |
| 3883 | goto out_put_css; |
| 3884 | |
| 3885 | efile.file->f_op->poll(efile.file, &event->pt); |
| 3886 | |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 3887 | spin_lock(&memcg->event_list_lock); |
| 3888 | list_add(&event->list, &memcg->event_list); |
| 3889 | spin_unlock(&memcg->event_list_lock); |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 3890 | |
| 3891 | fdput(cfile); |
| 3892 | fdput(efile); |
| 3893 | |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 3894 | return nbytes; |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 3895 | |
| 3896 | out_put_css: |
Tejun Heo | b5557c4 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 3897 | css_put(css); |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 3898 | out_put_cfile: |
| 3899 | fdput(cfile); |
| 3900 | out_put_eventfd: |
| 3901 | eventfd_ctx_put(event->eventfd); |
| 3902 | out_put_efile: |
| 3903 | fdput(efile); |
| 3904 | out_kfree: |
| 3905 | kfree(event); |
| 3906 | |
| 3907 | return ret; |
| 3908 | } |
| 3909 | |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 3910 | static struct cftype mem_cgroup_legacy_files[] = { |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 3911 | { |
Balbir Singh | 0eea103 | 2008-02-07 00:13:57 -0800 | [diff] [blame] | 3912 | .name = "usage_in_bytes", |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3913 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
Tejun Heo | 791badb | 2013-12-05 12:28:02 -0500 | [diff] [blame] | 3914 | .read_u64 = mem_cgroup_read_u64, |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 3915 | }, |
| 3916 | { |
Pavel Emelyanov | c84872e | 2008-04-29 01:00:17 -0700 | [diff] [blame] | 3917 | .name = "max_usage_in_bytes", |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3918 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
Tejun Heo | 6770c64 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 3919 | .write = mem_cgroup_reset, |
Tejun Heo | 791badb | 2013-12-05 12:28:02 -0500 | [diff] [blame] | 3920 | .read_u64 = mem_cgroup_read_u64, |
Pavel Emelyanov | c84872e | 2008-04-29 01:00:17 -0700 | [diff] [blame] | 3921 | }, |
| 3922 | { |
Balbir Singh | 0eea103 | 2008-02-07 00:13:57 -0800 | [diff] [blame] | 3923 | .name = "limit_in_bytes", |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3924 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 3925 | .write = mem_cgroup_write, |
Tejun Heo | 791badb | 2013-12-05 12:28:02 -0500 | [diff] [blame] | 3926 | .read_u64 = mem_cgroup_read_u64, |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 3927 | }, |
| 3928 | { |
Balbir Singh | 296c81d | 2009-09-23 15:56:36 -0700 | [diff] [blame] | 3929 | .name = "soft_limit_in_bytes", |
| 3930 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 3931 | .write = mem_cgroup_write, |
Tejun Heo | 791badb | 2013-12-05 12:28:02 -0500 | [diff] [blame] | 3932 | .read_u64 = mem_cgroup_read_u64, |
Balbir Singh | 296c81d | 2009-09-23 15:56:36 -0700 | [diff] [blame] | 3933 | }, |
| 3934 | { |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 3935 | .name = "failcnt", |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3936 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
Tejun Heo | 6770c64 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 3937 | .write = mem_cgroup_reset, |
Tejun Heo | 791badb | 2013-12-05 12:28:02 -0500 | [diff] [blame] | 3938 | .read_u64 = mem_cgroup_read_u64, |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 3939 | }, |
Balbir Singh | 8697d33 | 2008-02-07 00:13:59 -0800 | [diff] [blame] | 3940 | { |
KAMEZAWA Hiroyuki | d2ceb9b | 2008-02-07 00:14:25 -0800 | [diff] [blame] | 3941 | .name = "stat", |
Tejun Heo | 2da8ca8 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 3942 | .seq_show = memcg_stat_show, |
KAMEZAWA Hiroyuki | d2ceb9b | 2008-02-07 00:14:25 -0800 | [diff] [blame] | 3943 | }, |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 3944 | { |
| 3945 | .name = "force_empty", |
Tejun Heo | 6770c64 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 3946 | .write = mem_cgroup_force_empty_write, |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 3947 | }, |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 3948 | { |
| 3949 | .name = "use_hierarchy", |
| 3950 | .write_u64 = mem_cgroup_hierarchy_write, |
| 3951 | .read_u64 = mem_cgroup_hierarchy_read, |
| 3952 | }, |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 3953 | { |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 3954 | .name = "cgroup.event_control", /* XXX: for compat */ |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 3955 | .write = memcg_write_event_control, |
Tejun Heo | 7dbdb19 | 2015-09-18 17:54:23 -0400 | [diff] [blame] | 3956 | .flags = CFTYPE_NO_PREFIX | CFTYPE_WORLD_WRITABLE, |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 3957 | }, |
| 3958 | { |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 3959 | .name = "swappiness", |
| 3960 | .read_u64 = mem_cgroup_swappiness_read, |
| 3961 | .write_u64 = mem_cgroup_swappiness_write, |
| 3962 | }, |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 3963 | { |
| 3964 | .name = "move_charge_at_immigrate", |
| 3965 | .read_u64 = mem_cgroup_move_charge_read, |
| 3966 | .write_u64 = mem_cgroup_move_charge_write, |
| 3967 | }, |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3968 | { |
| 3969 | .name = "oom_control", |
Tejun Heo | 2da8ca8 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 3970 | .seq_show = mem_cgroup_oom_control_read, |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 3971 | .write_u64 = mem_cgroup_oom_control_write, |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 3972 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), |
| 3973 | }, |
Anton Vorontsov | 70ddf63 | 2013-04-29 15:08:31 -0700 | [diff] [blame] | 3974 | { |
| 3975 | .name = "pressure_level", |
Anton Vorontsov | 70ddf63 | 2013-04-29 15:08:31 -0700 | [diff] [blame] | 3976 | }, |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 3977 | #ifdef CONFIG_NUMA |
| 3978 | { |
| 3979 | .name = "numa_stat", |
Tejun Heo | 2da8ca8 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 3980 | .seq_show = memcg_numa_stat_show, |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 3981 | }, |
| 3982 | #endif |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 3983 | { |
| 3984 | .name = "kmem.limit_in_bytes", |
| 3985 | .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT), |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 3986 | .write = mem_cgroup_write, |
Tejun Heo | 791badb | 2013-12-05 12:28:02 -0500 | [diff] [blame] | 3987 | .read_u64 = mem_cgroup_read_u64, |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 3988 | }, |
| 3989 | { |
| 3990 | .name = "kmem.usage_in_bytes", |
| 3991 | .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE), |
Tejun Heo | 791badb | 2013-12-05 12:28:02 -0500 | [diff] [blame] | 3992 | .read_u64 = mem_cgroup_read_u64, |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 3993 | }, |
| 3994 | { |
| 3995 | .name = "kmem.failcnt", |
| 3996 | .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT), |
Tejun Heo | 6770c64 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 3997 | .write = mem_cgroup_reset, |
Tejun Heo | 791badb | 2013-12-05 12:28:02 -0500 | [diff] [blame] | 3998 | .read_u64 = mem_cgroup_read_u64, |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 3999 | }, |
| 4000 | { |
| 4001 | .name = "kmem.max_usage_in_bytes", |
| 4002 | .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE), |
Tejun Heo | 6770c64 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 4003 | .write = mem_cgroup_reset, |
Tejun Heo | 791badb | 2013-12-05 12:28:02 -0500 | [diff] [blame] | 4004 | .read_u64 = mem_cgroup_read_u64, |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 4005 | }, |
Glauber Costa | 749c541 | 2012-12-18 14:23:01 -0800 | [diff] [blame] | 4006 | #ifdef CONFIG_SLABINFO |
| 4007 | { |
| 4008 | .name = "kmem.slabinfo", |
Tejun Heo | bc2791f | 2017-02-22 15:41:21 -0800 | [diff] [blame] | 4009 | .seq_start = memcg_slab_start, |
| 4010 | .seq_next = memcg_slab_next, |
| 4011 | .seq_stop = memcg_slab_stop, |
Vladimir Davydov | b047501 | 2014-12-10 15:44:19 -0800 | [diff] [blame] | 4012 | .seq_show = memcg_slab_show, |
Glauber Costa | 749c541 | 2012-12-18 14:23:01 -0800 | [diff] [blame] | 4013 | }, |
| 4014 | #endif |
Vladimir Davydov | d55f90b | 2016-01-20 15:02:44 -0800 | [diff] [blame] | 4015 | { |
| 4016 | .name = "kmem.tcp.limit_in_bytes", |
| 4017 | .private = MEMFILE_PRIVATE(_TCP, RES_LIMIT), |
| 4018 | .write = mem_cgroup_write, |
| 4019 | .read_u64 = mem_cgroup_read_u64, |
| 4020 | }, |
| 4021 | { |
| 4022 | .name = "kmem.tcp.usage_in_bytes", |
| 4023 | .private = MEMFILE_PRIVATE(_TCP, RES_USAGE), |
| 4024 | .read_u64 = mem_cgroup_read_u64, |
| 4025 | }, |
| 4026 | { |
| 4027 | .name = "kmem.tcp.failcnt", |
| 4028 | .private = MEMFILE_PRIVATE(_TCP, RES_FAILCNT), |
| 4029 | .write = mem_cgroup_reset, |
| 4030 | .read_u64 = mem_cgroup_read_u64, |
| 4031 | }, |
| 4032 | { |
| 4033 | .name = "kmem.tcp.max_usage_in_bytes", |
| 4034 | .private = MEMFILE_PRIVATE(_TCP, RES_MAX_USAGE), |
| 4035 | .write = mem_cgroup_reset, |
| 4036 | .read_u64 = mem_cgroup_read_u64, |
| 4037 | }, |
Tejun Heo | 6bc1034 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 4038 | { }, /* terminate */ |
Tejun Heo | af36f90 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 4039 | }; |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4040 | |
Johannes Weiner | 73f576c | 2016-07-20 15:44:57 -0700 | [diff] [blame] | 4041 | /* |
| 4042 | * Private memory cgroup IDR |
| 4043 | * |
| 4044 | * Swap-out records and page cache shadow entries need to store memcg |
| 4045 | * references in constrained space, so we maintain an ID space that is |
| 4046 | * limited to 16 bit (MEM_CGROUP_ID_MAX), limiting the total number of |
| 4047 | * memory-controlled cgroups to 64k. |
| 4048 | * |
| 4049 | * However, there usually are many references to the oflline CSS after |
| 4050 | * the cgroup has been destroyed, such as page cache or reclaimable |
| 4051 | * slab objects, that don't need to hang on to the ID. We want to keep |
| 4052 | * those dead CSS from occupying IDs, or we might quickly exhaust the |
| 4053 | * relatively small ID space and prevent the creation of new cgroups |
| 4054 | * even when there are much fewer than 64k cgroups - possibly none. |
| 4055 | * |
| 4056 | * Maintain a private 16-bit ID space for memcg, and allow the ID to |
| 4057 | * be freed and recycled when it's no longer needed, which is usually |
| 4058 | * when the CSS is offlined. |
| 4059 | * |
| 4060 | * The only exception to that are records of swapped out tmpfs/shmem |
| 4061 | * pages that need to be attributed to live ancestors on swapin. But |
| 4062 | * those references are manageable from userspace. |
| 4063 | */ |
| 4064 | |
| 4065 | static DEFINE_IDR(mem_cgroup_idr); |
| 4066 | |
Vladimir Davydov | 615d66c | 2016-08-11 15:33:03 -0700 | [diff] [blame] | 4067 | static void mem_cgroup_id_get_many(struct mem_cgroup *memcg, unsigned int n) |
Johannes Weiner | 73f576c | 2016-07-20 15:44:57 -0700 | [diff] [blame] | 4068 | { |
Vladimir Davydov | 58fa2a5 | 2016-10-07 16:57:29 -0700 | [diff] [blame] | 4069 | VM_BUG_ON(atomic_read(&memcg->id.ref) <= 0); |
Vladimir Davydov | 615d66c | 2016-08-11 15:33:03 -0700 | [diff] [blame] | 4070 | atomic_add(n, &memcg->id.ref); |
Johannes Weiner | 73f576c | 2016-07-20 15:44:57 -0700 | [diff] [blame] | 4071 | } |
| 4072 | |
Vladimir Davydov | 615d66c | 2016-08-11 15:33:03 -0700 | [diff] [blame] | 4073 | static void mem_cgroup_id_put_many(struct mem_cgroup *memcg, unsigned int n) |
Johannes Weiner | 73f576c | 2016-07-20 15:44:57 -0700 | [diff] [blame] | 4074 | { |
Vladimir Davydov | 58fa2a5 | 2016-10-07 16:57:29 -0700 | [diff] [blame] | 4075 | VM_BUG_ON(atomic_read(&memcg->id.ref) < n); |
Vladimir Davydov | 615d66c | 2016-08-11 15:33:03 -0700 | [diff] [blame] | 4076 | if (atomic_sub_and_test(n, &memcg->id.ref)) { |
Johannes Weiner | 73f576c | 2016-07-20 15:44:57 -0700 | [diff] [blame] | 4077 | idr_remove(&mem_cgroup_idr, memcg->id.id); |
| 4078 | memcg->id.id = 0; |
| 4079 | |
| 4080 | /* Memcg ID pins CSS */ |
| 4081 | css_put(&memcg->css); |
| 4082 | } |
| 4083 | } |
| 4084 | |
Vladimir Davydov | 615d66c | 2016-08-11 15:33:03 -0700 | [diff] [blame] | 4085 | static inline void mem_cgroup_id_get(struct mem_cgroup *memcg) |
| 4086 | { |
| 4087 | mem_cgroup_id_get_many(memcg, 1); |
| 4088 | } |
| 4089 | |
| 4090 | static inline void mem_cgroup_id_put(struct mem_cgroup *memcg) |
| 4091 | { |
| 4092 | mem_cgroup_id_put_many(memcg, 1); |
| 4093 | } |
| 4094 | |
Johannes Weiner | 73f576c | 2016-07-20 15:44:57 -0700 | [diff] [blame] | 4095 | /** |
| 4096 | * mem_cgroup_from_id - look up a memcg from a memcg id |
| 4097 | * @id: the memcg id to look up |
| 4098 | * |
| 4099 | * Caller must hold rcu_read_lock(). |
| 4100 | */ |
| 4101 | struct mem_cgroup *mem_cgroup_from_id(unsigned short id) |
| 4102 | { |
| 4103 | WARN_ON_ONCE(!rcu_read_lock_held()); |
| 4104 | return idr_find(&mem_cgroup_idr, id); |
| 4105 | } |
| 4106 | |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 4107 | static int alloc_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node) |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 4108 | { |
| 4109 | struct mem_cgroup_per_node *pn; |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 4110 | int tmp = node; |
KAMEZAWA Hiroyuki | 1ecaab2 | 2008-02-07 00:14:38 -0800 | [diff] [blame] | 4111 | /* |
| 4112 | * This routine is called against possible nodes. |
| 4113 | * But it's BUG to call kmalloc() against offline node. |
| 4114 | * |
| 4115 | * TODO: this routine can waste much memory for nodes which will |
| 4116 | * never be onlined. It's better to use memory hotplug callback |
| 4117 | * function. |
| 4118 | */ |
KAMEZAWA Hiroyuki | 41e3355 | 2008-04-08 17:41:54 -0700 | [diff] [blame] | 4119 | if (!node_state(node, N_NORMAL_MEMORY)) |
| 4120 | tmp = -1; |
Jesper Juhl | 17295c8 | 2011-01-13 15:47:42 -0800 | [diff] [blame] | 4121 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 4122 | if (!pn) |
| 4123 | return 1; |
KAMEZAWA Hiroyuki | 1ecaab2 | 2008-02-07 00:14:38 -0800 | [diff] [blame] | 4124 | |
Johannes Weiner | 00f3ca2 | 2017-07-06 15:40:52 -0700 | [diff] [blame] | 4125 | pn->lruvec_stat = alloc_percpu(struct lruvec_stat); |
| 4126 | if (!pn->lruvec_stat) { |
| 4127 | kfree(pn); |
| 4128 | return 1; |
| 4129 | } |
| 4130 | |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 4131 | lruvec_init(&pn->lruvec); |
| 4132 | pn->usage_in_excess = 0; |
| 4133 | pn->on_tree = false; |
| 4134 | pn->memcg = memcg; |
| 4135 | |
Johannes Weiner | 54f72fe | 2013-07-08 15:59:49 -0700 | [diff] [blame] | 4136 | memcg->nodeinfo[node] = pn; |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 4137 | return 0; |
| 4138 | } |
| 4139 | |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 4140 | static void free_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node) |
KAMEZAWA Hiroyuki | 1ecaab2 | 2008-02-07 00:14:38 -0800 | [diff] [blame] | 4141 | { |
Johannes Weiner | 00f3ca2 | 2017-07-06 15:40:52 -0700 | [diff] [blame] | 4142 | struct mem_cgroup_per_node *pn = memcg->nodeinfo[node]; |
| 4143 | |
| 4144 | free_percpu(pn->lruvec_stat); |
| 4145 | kfree(pn); |
KAMEZAWA Hiroyuki | 1ecaab2 | 2008-02-07 00:14:38 -0800 | [diff] [blame] | 4146 | } |
| 4147 | |
Tahsin Erdogan | 40e952f | 2017-03-09 16:17:26 -0800 | [diff] [blame] | 4148 | static void __mem_cgroup_free(struct mem_cgroup *memcg) |
Johannes Weiner | 0b8f73e | 2016-01-20 15:02:53 -0800 | [diff] [blame] | 4149 | { |
| 4150 | int node; |
| 4151 | |
Johannes Weiner | 0b8f73e | 2016-01-20 15:02:53 -0800 | [diff] [blame] | 4152 | for_each_node(node) |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 4153 | free_mem_cgroup_per_node_info(memcg, node); |
Johannes Weiner | 0b8f73e | 2016-01-20 15:02:53 -0800 | [diff] [blame] | 4154 | free_percpu(memcg->stat); |
| 4155 | kfree(memcg); |
| 4156 | } |
| 4157 | |
Tahsin Erdogan | 40e952f | 2017-03-09 16:17:26 -0800 | [diff] [blame] | 4158 | static void mem_cgroup_free(struct mem_cgroup *memcg) |
| 4159 | { |
| 4160 | memcg_wb_domain_exit(memcg); |
| 4161 | __mem_cgroup_free(memcg); |
| 4162 | } |
| 4163 | |
KAMEZAWA Hiroyuki | 3332794 | 2008-04-29 01:00:24 -0700 | [diff] [blame] | 4164 | static struct mem_cgroup *mem_cgroup_alloc(void) |
| 4165 | { |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 4166 | struct mem_cgroup *memcg; |
Vladimir Davydov | 8ff69e2 | 2014-01-23 15:52:52 -0800 | [diff] [blame] | 4167 | size_t size; |
Johannes Weiner | 0b8f73e | 2016-01-20 15:02:53 -0800 | [diff] [blame] | 4168 | int node; |
KAMEZAWA Hiroyuki | 3332794 | 2008-04-29 01:00:24 -0700 | [diff] [blame] | 4169 | |
Vladimir Davydov | 8ff69e2 | 2014-01-23 15:52:52 -0800 | [diff] [blame] | 4170 | size = sizeof(struct mem_cgroup); |
| 4171 | size += nr_node_ids * sizeof(struct mem_cgroup_per_node *); |
KAMEZAWA Hiroyuki | 3332794 | 2008-04-29 01:00:24 -0700 | [diff] [blame] | 4172 | |
Vladimir Davydov | 8ff69e2 | 2014-01-23 15:52:52 -0800 | [diff] [blame] | 4173 | memcg = kzalloc(size, GFP_KERNEL); |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 4174 | if (!memcg) |
Dan Carpenter | e7bbcdf | 2010-03-23 13:35:12 -0700 | [diff] [blame] | 4175 | return NULL; |
| 4176 | |
Johannes Weiner | 73f576c | 2016-07-20 15:44:57 -0700 | [diff] [blame] | 4177 | memcg->id.id = idr_alloc(&mem_cgroup_idr, NULL, |
| 4178 | 1, MEM_CGROUP_ID_MAX, |
| 4179 | GFP_KERNEL); |
| 4180 | if (memcg->id.id < 0) |
| 4181 | goto fail; |
| 4182 | |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 4183 | memcg->stat = alloc_percpu(struct mem_cgroup_stat_cpu); |
| 4184 | if (!memcg->stat) |
Johannes Weiner | 0b8f73e | 2016-01-20 15:02:53 -0800 | [diff] [blame] | 4185 | goto fail; |
Pavel Emelianov | 78fb746 | 2008-02-07 00:13:51 -0800 | [diff] [blame] | 4186 | |
Bob Liu | 3ed28fa | 2012-01-12 17:19:04 -0800 | [diff] [blame] | 4187 | for_each_node(node) |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 4188 | if (alloc_mem_cgroup_per_node_info(memcg, node)) |
Johannes Weiner | 0b8f73e | 2016-01-20 15:02:53 -0800 | [diff] [blame] | 4189 | goto fail; |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 4190 | |
Johannes Weiner | 0b8f73e | 2016-01-20 15:02:53 -0800 | [diff] [blame] | 4191 | if (memcg_wb_domain_init(memcg, GFP_KERNEL)) |
| 4192 | goto fail; |
Balbir Singh | 28dbc4b | 2009-01-07 18:08:05 -0800 | [diff] [blame] | 4193 | |
Johannes Weiner | f7e1cb6 | 2016-01-14 15:21:29 -0800 | [diff] [blame] | 4194 | INIT_WORK(&memcg->high_work, high_work_func); |
Glauber Costa | d142e3e | 2013-02-22 16:34:52 -0800 | [diff] [blame] | 4195 | memcg->last_scanned_node = MAX_NUMNODES; |
| 4196 | INIT_LIST_HEAD(&memcg->oom_notify); |
Glauber Costa | d142e3e | 2013-02-22 16:34:52 -0800 | [diff] [blame] | 4197 | mutex_init(&memcg->thresholds_lock); |
| 4198 | spin_lock_init(&memcg->move_lock); |
Anton Vorontsov | 70ddf63 | 2013-04-29 15:08:31 -0700 | [diff] [blame] | 4199 | vmpressure_init(&memcg->vmpressure); |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4200 | INIT_LIST_HEAD(&memcg->event_list); |
| 4201 | spin_lock_init(&memcg->event_list_lock); |
Johannes Weiner | d886f4e | 2016-01-20 15:02:47 -0800 | [diff] [blame] | 4202 | memcg->socket_pressure = jiffies; |
Johannes Weiner | 127424c | 2016-01-20 15:02:32 -0800 | [diff] [blame] | 4203 | #ifndef CONFIG_SLOB |
Vladimir Davydov | 900a38f | 2014-12-12 16:55:10 -0800 | [diff] [blame] | 4204 | memcg->kmemcg_id = -1; |
Vladimir Davydov | 900a38f | 2014-12-12 16:55:10 -0800 | [diff] [blame] | 4205 | #endif |
Tejun Heo | 52ebea7 | 2015-05-22 17:13:37 -0400 | [diff] [blame] | 4206 | #ifdef CONFIG_CGROUP_WRITEBACK |
| 4207 | INIT_LIST_HEAD(&memcg->cgwb_list); |
| 4208 | #endif |
Johannes Weiner | 73f576c | 2016-07-20 15:44:57 -0700 | [diff] [blame] | 4209 | idr_replace(&mem_cgroup_idr, memcg, memcg->id.id); |
Johannes Weiner | 0b8f73e | 2016-01-20 15:02:53 -0800 | [diff] [blame] | 4210 | return memcg; |
| 4211 | fail: |
Johannes Weiner | 73f576c | 2016-07-20 15:44:57 -0700 | [diff] [blame] | 4212 | if (memcg->id.id > 0) |
| 4213 | idr_remove(&mem_cgroup_idr, memcg->id.id); |
Tahsin Erdogan | 40e952f | 2017-03-09 16:17:26 -0800 | [diff] [blame] | 4214 | __mem_cgroup_free(memcg); |
Johannes Weiner | 0b8f73e | 2016-01-20 15:02:53 -0800 | [diff] [blame] | 4215 | return NULL; |
Glauber Costa | d142e3e | 2013-02-22 16:34:52 -0800 | [diff] [blame] | 4216 | } |
| 4217 | |
Johannes Weiner | 0b8f73e | 2016-01-20 15:02:53 -0800 | [diff] [blame] | 4218 | static struct cgroup_subsys_state * __ref |
| 4219 | mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) |
Glauber Costa | d142e3e | 2013-02-22 16:34:52 -0800 | [diff] [blame] | 4220 | { |
Johannes Weiner | 0b8f73e | 2016-01-20 15:02:53 -0800 | [diff] [blame] | 4221 | struct mem_cgroup *parent = mem_cgroup_from_css(parent_css); |
| 4222 | struct mem_cgroup *memcg; |
| 4223 | long error = -ENOMEM; |
Glauber Costa | d142e3e | 2013-02-22 16:34:52 -0800 | [diff] [blame] | 4224 | |
Johannes Weiner | 0b8f73e | 2016-01-20 15:02:53 -0800 | [diff] [blame] | 4225 | memcg = mem_cgroup_alloc(); |
| 4226 | if (!memcg) |
| 4227 | return ERR_PTR(error); |
Li Zefan | 4219b2d | 2013-09-23 16:56:29 +0800 | [diff] [blame] | 4228 | |
Johannes Weiner | 0b8f73e | 2016-01-20 15:02:53 -0800 | [diff] [blame] | 4229 | memcg->high = PAGE_COUNTER_MAX; |
| 4230 | memcg->soft_limit = PAGE_COUNTER_MAX; |
| 4231 | if (parent) { |
| 4232 | memcg->swappiness = mem_cgroup_swappiness(parent); |
| 4233 | memcg->oom_kill_disable = parent->oom_kill_disable; |
| 4234 | } |
| 4235 | if (parent && parent->use_hierarchy) { |
| 4236 | memcg->use_hierarchy = true; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 4237 | page_counter_init(&memcg->memory, &parent->memory); |
Vladimir Davydov | 37e8435 | 2016-01-20 15:02:56 -0800 | [diff] [blame] | 4238 | page_counter_init(&memcg->swap, &parent->swap); |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 4239 | page_counter_init(&memcg->memsw, &parent->memsw); |
| 4240 | page_counter_init(&memcg->kmem, &parent->kmem); |
Johannes Weiner | 0db1529 | 2016-01-20 15:02:50 -0800 | [diff] [blame] | 4241 | page_counter_init(&memcg->tcpmem, &parent->tcpmem); |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 4242 | } else { |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 4243 | page_counter_init(&memcg->memory, NULL); |
Vladimir Davydov | 37e8435 | 2016-01-20 15:02:56 -0800 | [diff] [blame] | 4244 | page_counter_init(&memcg->swap, NULL); |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 4245 | page_counter_init(&memcg->memsw, NULL); |
| 4246 | page_counter_init(&memcg->kmem, NULL); |
Johannes Weiner | 0db1529 | 2016-01-20 15:02:50 -0800 | [diff] [blame] | 4247 | page_counter_init(&memcg->tcpmem, NULL); |
Tejun Heo | 8c7f6ed | 2012-09-13 12:20:58 -0700 | [diff] [blame] | 4248 | /* |
| 4249 | * Deeper hierachy with use_hierarchy == false doesn't make |
| 4250 | * much sense so let cgroup subsystem know about this |
| 4251 | * unfortunate state in our controller. |
| 4252 | */ |
Glauber Costa | d142e3e | 2013-02-22 16:34:52 -0800 | [diff] [blame] | 4253 | if (parent != root_mem_cgroup) |
Tejun Heo | 073219e | 2014-02-08 10:36:58 -0500 | [diff] [blame] | 4254 | memory_cgrp_subsys.broken_hierarchy = true; |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 4255 | } |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 4256 | |
Johannes Weiner | 0b8f73e | 2016-01-20 15:02:53 -0800 | [diff] [blame] | 4257 | /* The following stuff does not apply to the root */ |
| 4258 | if (!parent) { |
| 4259 | root_mem_cgroup = memcg; |
| 4260 | return &memcg->css; |
| 4261 | } |
| 4262 | |
Vladimir Davydov | b313aee | 2016-03-17 14:18:27 -0700 | [diff] [blame] | 4263 | error = memcg_online_kmem(memcg); |
Johannes Weiner | 0b8f73e | 2016-01-20 15:02:53 -0800 | [diff] [blame] | 4264 | if (error) |
| 4265 | goto fail; |
Johannes Weiner | 127424c | 2016-01-20 15:02:32 -0800 | [diff] [blame] | 4266 | |
Johannes Weiner | f7e1cb6 | 2016-01-14 15:21:29 -0800 | [diff] [blame] | 4267 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket) |
Johannes Weiner | ef12947 | 2016-01-14 15:21:34 -0800 | [diff] [blame] | 4268 | static_branch_inc(&memcg_sockets_enabled_key); |
Johannes Weiner | f7e1cb6 | 2016-01-14 15:21:29 -0800 | [diff] [blame] | 4269 | |
Johannes Weiner | 0b8f73e | 2016-01-20 15:02:53 -0800 | [diff] [blame] | 4270 | return &memcg->css; |
| 4271 | fail: |
| 4272 | mem_cgroup_free(memcg); |
Tejun Heo | ea3a964 | 2016-06-24 14:49:58 -0700 | [diff] [blame] | 4273 | return ERR_PTR(-ENOMEM); |
Johannes Weiner | 0b8f73e | 2016-01-20 15:02:53 -0800 | [diff] [blame] | 4274 | } |
| 4275 | |
Johannes Weiner | 73f576c | 2016-07-20 15:44:57 -0700 | [diff] [blame] | 4276 | static int mem_cgroup_css_online(struct cgroup_subsys_state *css) |
Johannes Weiner | 0b8f73e | 2016-01-20 15:02:53 -0800 | [diff] [blame] | 4277 | { |
Vladimir Davydov | 58fa2a5 | 2016-10-07 16:57:29 -0700 | [diff] [blame] | 4278 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
| 4279 | |
Johannes Weiner | 73f576c | 2016-07-20 15:44:57 -0700 | [diff] [blame] | 4280 | /* Online state pins memcg ID, memcg ID pins CSS */ |
Vladimir Davydov | 58fa2a5 | 2016-10-07 16:57:29 -0700 | [diff] [blame] | 4281 | atomic_set(&memcg->id.ref, 1); |
Johannes Weiner | 73f576c | 2016-07-20 15:44:57 -0700 | [diff] [blame] | 4282 | css_get(css); |
Johannes Weiner | 2f7dd7a | 2014-10-02 16:16:57 -0700 | [diff] [blame] | 4283 | return 0; |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 4284 | } |
| 4285 | |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 4286 | static void mem_cgroup_css_offline(struct cgroup_subsys_state *css) |
KAMEZAWA Hiroyuki | df878fb | 2008-02-07 00:14:28 -0800 | [diff] [blame] | 4287 | { |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 4288 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 4289 | struct mem_cgroup_event *event, *tmp; |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4290 | |
| 4291 | /* |
| 4292 | * Unregister events and notify userspace. |
| 4293 | * Notify userspace about cgroup removing only after rmdir of cgroup |
| 4294 | * directory to avoid race between userspace and kernelspace. |
| 4295 | */ |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4296 | spin_lock(&memcg->event_list_lock); |
| 4297 | list_for_each_entry_safe(event, tmp, &memcg->event_list, list) { |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4298 | list_del_init(&event->list); |
| 4299 | schedule_work(&event->remove); |
| 4300 | } |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4301 | spin_unlock(&memcg->event_list_lock); |
KAMEZAWA Hiroyuki | ec64f51 | 2009-04-02 16:57:26 -0700 | [diff] [blame] | 4302 | |
Johannes Weiner | 567e9ab | 2016-01-20 15:02:24 -0800 | [diff] [blame] | 4303 | memcg_offline_kmem(memcg); |
Tejun Heo | 52ebea7 | 2015-05-22 17:13:37 -0400 | [diff] [blame] | 4304 | wb_memcg_offline(memcg); |
Johannes Weiner | 73f576c | 2016-07-20 15:44:57 -0700 | [diff] [blame] | 4305 | |
| 4306 | mem_cgroup_id_put(memcg); |
KAMEZAWA Hiroyuki | df878fb | 2008-02-07 00:14:28 -0800 | [diff] [blame] | 4307 | } |
| 4308 | |
Vladimir Davydov | 6df3868 | 2015-12-29 14:54:10 -0800 | [diff] [blame] | 4309 | static void mem_cgroup_css_released(struct cgroup_subsys_state *css) |
| 4310 | { |
| 4311 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
| 4312 | |
| 4313 | invalidate_reclaim_iterators(memcg); |
| 4314 | } |
| 4315 | |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 4316 | static void mem_cgroup_css_free(struct cgroup_subsys_state *css) |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 4317 | { |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 4318 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
Daisuke Nishimura | c268e99 | 2009-01-15 13:51:13 -0800 | [diff] [blame] | 4319 | |
Johannes Weiner | f7e1cb6 | 2016-01-14 15:21:29 -0800 | [diff] [blame] | 4320 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket) |
Johannes Weiner | ef12947 | 2016-01-14 15:21:34 -0800 | [diff] [blame] | 4321 | static_branch_dec(&memcg_sockets_enabled_key); |
Johannes Weiner | 3893e30 | 2016-01-20 15:02:29 -0800 | [diff] [blame] | 4322 | |
Johannes Weiner | 0db1529 | 2016-01-20 15:02:50 -0800 | [diff] [blame] | 4323 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg->tcpmem_active) |
Vladimir Davydov | d55f90b | 2016-01-20 15:02:44 -0800 | [diff] [blame] | 4324 | static_branch_dec(&memcg_sockets_enabled_key); |
Johannes Weiner | 3893e30 | 2016-01-20 15:02:29 -0800 | [diff] [blame] | 4325 | |
Johannes Weiner | 0b8f73e | 2016-01-20 15:02:53 -0800 | [diff] [blame] | 4326 | vmpressure_cleanup(&memcg->vmpressure); |
| 4327 | cancel_work_sync(&memcg->high_work); |
| 4328 | mem_cgroup_remove_from_trees(memcg); |
Johannes Weiner | d886f4e | 2016-01-20 15:02:47 -0800 | [diff] [blame] | 4329 | memcg_free_kmem(memcg); |
Johannes Weiner | 0b8f73e | 2016-01-20 15:02:53 -0800 | [diff] [blame] | 4330 | mem_cgroup_free(memcg); |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 4331 | } |
| 4332 | |
Tejun Heo | 1ced953 | 2014-07-08 18:02:57 -0400 | [diff] [blame] | 4333 | /** |
| 4334 | * mem_cgroup_css_reset - reset the states of a mem_cgroup |
| 4335 | * @css: the target css |
| 4336 | * |
| 4337 | * Reset the states of the mem_cgroup associated with @css. This is |
| 4338 | * invoked when the userland requests disabling on the default hierarchy |
| 4339 | * but the memcg is pinned through dependency. The memcg should stop |
| 4340 | * applying policies and should revert to the vanilla state as it may be |
| 4341 | * made visible again. |
| 4342 | * |
| 4343 | * The current implementation only resets the essential configurations. |
| 4344 | * This needs to be expanded to cover all the visible parts. |
| 4345 | */ |
| 4346 | static void mem_cgroup_css_reset(struct cgroup_subsys_state *css) |
| 4347 | { |
| 4348 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
| 4349 | |
Vladimir Davydov | d334c9bc | 2016-03-17 14:19:38 -0700 | [diff] [blame] | 4350 | page_counter_limit(&memcg->memory, PAGE_COUNTER_MAX); |
| 4351 | page_counter_limit(&memcg->swap, PAGE_COUNTER_MAX); |
| 4352 | page_counter_limit(&memcg->memsw, PAGE_COUNTER_MAX); |
| 4353 | page_counter_limit(&memcg->kmem, PAGE_COUNTER_MAX); |
| 4354 | page_counter_limit(&memcg->tcpmem, PAGE_COUNTER_MAX); |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 4355 | memcg->low = 0; |
| 4356 | memcg->high = PAGE_COUNTER_MAX; |
Johannes Weiner | 24d404d | 2015-01-08 14:32:35 -0800 | [diff] [blame] | 4357 | memcg->soft_limit = PAGE_COUNTER_MAX; |
Tejun Heo | 2529bb3 | 2015-05-22 18:23:34 -0400 | [diff] [blame] | 4358 | memcg_wb_domain_size_changed(memcg); |
Tejun Heo | 1ced953 | 2014-07-08 18:02:57 -0400 | [diff] [blame] | 4359 | } |
| 4360 | |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4361 | #ifdef CONFIG_MMU |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 4362 | /* Handlers for move charge at task migration. */ |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 4363 | static int mem_cgroup_do_precharge(unsigned long count) |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 4364 | { |
Johannes Weiner | 05b8430 | 2014-08-06 16:05:59 -0700 | [diff] [blame] | 4365 | int ret; |
Johannes Weiner | 9476db9 | 2014-08-06 16:05:55 -0700 | [diff] [blame] | 4366 | |
Mel Gorman | d0164ad | 2015-11-06 16:28:21 -0800 | [diff] [blame] | 4367 | /* Try a single bulk charge without reclaim first, kswapd may wake */ |
| 4368 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_DIRECT_RECLAIM, count); |
Johannes Weiner | 9476db9 | 2014-08-06 16:05:55 -0700 | [diff] [blame] | 4369 | if (!ret) { |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 4370 | mc.precharge += count; |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 4371 | return ret; |
| 4372 | } |
Johannes Weiner | 9476db9 | 2014-08-06 16:05:55 -0700 | [diff] [blame] | 4373 | |
David Rientjes | 3674534 | 2017-01-24 15:18:10 -0800 | [diff] [blame] | 4374 | /* Try charges one by one with reclaim, but do not retry */ |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 4375 | while (count--) { |
David Rientjes | 3674534 | 2017-01-24 15:18:10 -0800 | [diff] [blame] | 4376 | ret = try_charge(mc.to, GFP_KERNEL | __GFP_NORETRY, 1); |
KAMEZAWA Hiroyuki | 38c5d72 | 2012-01-12 17:19:01 -0800 | [diff] [blame] | 4377 | if (ret) |
KAMEZAWA Hiroyuki | 38c5d72 | 2012-01-12 17:19:01 -0800 | [diff] [blame] | 4378 | return ret; |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 4379 | mc.precharge++; |
Johannes Weiner | 9476db9 | 2014-08-06 16:05:55 -0700 | [diff] [blame] | 4380 | cond_resched(); |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 4381 | } |
Johannes Weiner | 9476db9 | 2014-08-06 16:05:55 -0700 | [diff] [blame] | 4382 | return 0; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4383 | } |
| 4384 | |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4385 | union mc_target { |
| 4386 | struct page *page; |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4387 | swp_entry_t ent; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4388 | }; |
| 4389 | |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4390 | enum mc_target_type { |
Naoya Horiguchi | 8d32ff8 | 2012-03-21 16:34:27 -0700 | [diff] [blame] | 4391 | MC_TARGET_NONE = 0, |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4392 | MC_TARGET_PAGE, |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4393 | MC_TARGET_SWAP, |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4394 | }; |
| 4395 | |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 4396 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
| 4397 | unsigned long addr, pte_t ptent) |
| 4398 | { |
| 4399 | struct page *page = vm_normal_page(vma, addr, ptent); |
| 4400 | |
| 4401 | if (!page || !page_mapped(page)) |
| 4402 | return NULL; |
| 4403 | if (PageAnon(page)) { |
Johannes Weiner | 1dfab5a | 2015-02-11 15:26:09 -0800 | [diff] [blame] | 4404 | if (!(mc.flags & MOVE_ANON)) |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 4405 | return NULL; |
Johannes Weiner | 1dfab5a | 2015-02-11 15:26:09 -0800 | [diff] [blame] | 4406 | } else { |
| 4407 | if (!(mc.flags & MOVE_FILE)) |
| 4408 | return NULL; |
| 4409 | } |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 4410 | if (!get_page_unless_zero(page)) |
| 4411 | return NULL; |
| 4412 | |
| 4413 | return page; |
| 4414 | } |
| 4415 | |
KAMEZAWA Hiroyuki | 4b91355 | 2012-05-29 15:06:51 -0700 | [diff] [blame] | 4416 | #ifdef CONFIG_SWAP |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 4417 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
Li RongQing | 48406ef | 2016-07-26 15:22:14 -0700 | [diff] [blame] | 4418 | pte_t ptent, swp_entry_t *entry) |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 4419 | { |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 4420 | struct page *page = NULL; |
| 4421 | swp_entry_t ent = pte_to_swp_entry(ptent); |
| 4422 | |
Johannes Weiner | 1dfab5a | 2015-02-11 15:26:09 -0800 | [diff] [blame] | 4423 | if (!(mc.flags & MOVE_ANON) || non_swap_entry(ent)) |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 4424 | return NULL; |
KAMEZAWA Hiroyuki | 4b91355 | 2012-05-29 15:06:51 -0700 | [diff] [blame] | 4425 | /* |
| 4426 | * Because lookup_swap_cache() updates some statistics counter, |
| 4427 | * we call find_get_page() with swapper_space directly. |
| 4428 | */ |
Huang Ying | f6ab1f7 | 2016-10-07 17:00:21 -0700 | [diff] [blame] | 4429 | page = find_get_page(swap_address_space(ent), swp_offset(ent)); |
Johannes Weiner | 7941d21 | 2016-01-14 15:21:23 -0800 | [diff] [blame] | 4430 | if (do_memsw_account()) |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 4431 | entry->val = ent.val; |
| 4432 | |
| 4433 | return page; |
| 4434 | } |
KAMEZAWA Hiroyuki | 4b91355 | 2012-05-29 15:06:51 -0700 | [diff] [blame] | 4435 | #else |
| 4436 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
Li RongQing | 48406ef | 2016-07-26 15:22:14 -0700 | [diff] [blame] | 4437 | pte_t ptent, swp_entry_t *entry) |
KAMEZAWA Hiroyuki | 4b91355 | 2012-05-29 15:06:51 -0700 | [diff] [blame] | 4438 | { |
| 4439 | return NULL; |
| 4440 | } |
| 4441 | #endif |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 4442 | |
Daisuke Nishimura | 87946a7 | 2010-05-26 14:42:39 -0700 | [diff] [blame] | 4443 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
| 4444 | unsigned long addr, pte_t ptent, swp_entry_t *entry) |
| 4445 | { |
| 4446 | struct page *page = NULL; |
Daisuke Nishimura | 87946a7 | 2010-05-26 14:42:39 -0700 | [diff] [blame] | 4447 | struct address_space *mapping; |
| 4448 | pgoff_t pgoff; |
| 4449 | |
| 4450 | if (!vma->vm_file) /* anonymous vma */ |
| 4451 | return NULL; |
Johannes Weiner | 1dfab5a | 2015-02-11 15:26:09 -0800 | [diff] [blame] | 4452 | if (!(mc.flags & MOVE_FILE)) |
Daisuke Nishimura | 87946a7 | 2010-05-26 14:42:39 -0700 | [diff] [blame] | 4453 | return NULL; |
| 4454 | |
Daisuke Nishimura | 87946a7 | 2010-05-26 14:42:39 -0700 | [diff] [blame] | 4455 | mapping = vma->vm_file->f_mapping; |
Kirill A. Shutemov | 0661a33 | 2015-02-10 14:10:04 -0800 | [diff] [blame] | 4456 | pgoff = linear_page_index(vma, addr); |
Daisuke Nishimura | 87946a7 | 2010-05-26 14:42:39 -0700 | [diff] [blame] | 4457 | |
| 4458 | /* page is moved even if it's not RSS of this task(page-faulted). */ |
Hugh Dickins | aa3b189 | 2011-08-03 16:21:24 -0700 | [diff] [blame] | 4459 | #ifdef CONFIG_SWAP |
| 4460 | /* shmem/tmpfs may report page out on swap: account for that too. */ |
Johannes Weiner | 139b6a6 | 2014-05-06 12:50:05 -0700 | [diff] [blame] | 4461 | if (shmem_mapping(mapping)) { |
| 4462 | page = find_get_entry(mapping, pgoff); |
| 4463 | if (radix_tree_exceptional_entry(page)) { |
| 4464 | swp_entry_t swp = radix_to_swp_entry(page); |
Johannes Weiner | 7941d21 | 2016-01-14 15:21:23 -0800 | [diff] [blame] | 4465 | if (do_memsw_account()) |
Johannes Weiner | 139b6a6 | 2014-05-06 12:50:05 -0700 | [diff] [blame] | 4466 | *entry = swp; |
Huang Ying | f6ab1f7 | 2016-10-07 17:00:21 -0700 | [diff] [blame] | 4467 | page = find_get_page(swap_address_space(swp), |
| 4468 | swp_offset(swp)); |
Johannes Weiner | 139b6a6 | 2014-05-06 12:50:05 -0700 | [diff] [blame] | 4469 | } |
| 4470 | } else |
| 4471 | page = find_get_page(mapping, pgoff); |
| 4472 | #else |
| 4473 | page = find_get_page(mapping, pgoff); |
Hugh Dickins | aa3b189 | 2011-08-03 16:21:24 -0700 | [diff] [blame] | 4474 | #endif |
Daisuke Nishimura | 87946a7 | 2010-05-26 14:42:39 -0700 | [diff] [blame] | 4475 | return page; |
| 4476 | } |
| 4477 | |
Chen Gang | b1b0dea | 2015-04-14 15:47:35 -0700 | [diff] [blame] | 4478 | /** |
| 4479 | * mem_cgroup_move_account - move account of the page |
| 4480 | * @page: the page |
Li RongQing | 25843c2 | 2016-07-26 15:26:56 -0700 | [diff] [blame] | 4481 | * @compound: charge the page as compound or small page |
Chen Gang | b1b0dea | 2015-04-14 15:47:35 -0700 | [diff] [blame] | 4482 | * @from: mem_cgroup which the page is moved from. |
| 4483 | * @to: mem_cgroup which the page is moved to. @from != @to. |
| 4484 | * |
Kirill A. Shutemov | 3ac808f | 2016-01-15 16:53:07 -0800 | [diff] [blame] | 4485 | * The caller must make sure the page is not on LRU (isolate_page() is useful.) |
Chen Gang | b1b0dea | 2015-04-14 15:47:35 -0700 | [diff] [blame] | 4486 | * |
| 4487 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" |
| 4488 | * from old cgroup. |
| 4489 | */ |
| 4490 | static int mem_cgroup_move_account(struct page *page, |
Kirill A. Shutemov | f627c2f | 2016-01-15 16:52:20 -0800 | [diff] [blame] | 4491 | bool compound, |
Chen Gang | b1b0dea | 2015-04-14 15:47:35 -0700 | [diff] [blame] | 4492 | struct mem_cgroup *from, |
| 4493 | struct mem_cgroup *to) |
| 4494 | { |
| 4495 | unsigned long flags; |
Kirill A. Shutemov | f627c2f | 2016-01-15 16:52:20 -0800 | [diff] [blame] | 4496 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
Chen Gang | b1b0dea | 2015-04-14 15:47:35 -0700 | [diff] [blame] | 4497 | int ret; |
Greg Thelen | c4843a7 | 2015-05-22 17:13:16 -0400 | [diff] [blame] | 4498 | bool anon; |
Chen Gang | b1b0dea | 2015-04-14 15:47:35 -0700 | [diff] [blame] | 4499 | |
| 4500 | VM_BUG_ON(from == to); |
| 4501 | VM_BUG_ON_PAGE(PageLRU(page), page); |
Kirill A. Shutemov | f627c2f | 2016-01-15 16:52:20 -0800 | [diff] [blame] | 4502 | VM_BUG_ON(compound && !PageTransHuge(page)); |
Chen Gang | b1b0dea | 2015-04-14 15:47:35 -0700 | [diff] [blame] | 4503 | |
| 4504 | /* |
Johannes Weiner | 6a93ca8 | 2016-03-15 14:57:19 -0700 | [diff] [blame] | 4505 | * Prevent mem_cgroup_migrate() from looking at |
Hugh Dickins | 45637ba | 2015-11-05 18:49:40 -0800 | [diff] [blame] | 4506 | * page->mem_cgroup of its source page while we change it. |
Chen Gang | b1b0dea | 2015-04-14 15:47:35 -0700 | [diff] [blame] | 4507 | */ |
Kirill A. Shutemov | f627c2f | 2016-01-15 16:52:20 -0800 | [diff] [blame] | 4508 | ret = -EBUSY; |
Chen Gang | b1b0dea | 2015-04-14 15:47:35 -0700 | [diff] [blame] | 4509 | if (!trylock_page(page)) |
| 4510 | goto out; |
| 4511 | |
| 4512 | ret = -EINVAL; |
| 4513 | if (page->mem_cgroup != from) |
| 4514 | goto out_unlock; |
| 4515 | |
Greg Thelen | c4843a7 | 2015-05-22 17:13:16 -0400 | [diff] [blame] | 4516 | anon = PageAnon(page); |
| 4517 | |
Chen Gang | b1b0dea | 2015-04-14 15:47:35 -0700 | [diff] [blame] | 4518 | spin_lock_irqsave(&from->move_lock, flags); |
| 4519 | |
Greg Thelen | c4843a7 | 2015-05-22 17:13:16 -0400 | [diff] [blame] | 4520 | if (!anon && page_mapped(page)) { |
Johannes Weiner | 71cd311 | 2017-05-03 14:55:13 -0700 | [diff] [blame] | 4521 | __this_cpu_sub(from->stat->count[NR_FILE_MAPPED], nr_pages); |
| 4522 | __this_cpu_add(to->stat->count[NR_FILE_MAPPED], nr_pages); |
Chen Gang | b1b0dea | 2015-04-14 15:47:35 -0700 | [diff] [blame] | 4523 | } |
| 4524 | |
Greg Thelen | c4843a7 | 2015-05-22 17:13:16 -0400 | [diff] [blame] | 4525 | /* |
| 4526 | * move_lock grabbed above and caller set from->moving_account, so |
Johannes Weiner | ccda7f4 | 2017-05-03 14:55:16 -0700 | [diff] [blame] | 4527 | * mod_memcg_page_state will serialize updates to PageDirty. |
Greg Thelen | c4843a7 | 2015-05-22 17:13:16 -0400 | [diff] [blame] | 4528 | * So mapping should be stable for dirty pages. |
| 4529 | */ |
| 4530 | if (!anon && PageDirty(page)) { |
| 4531 | struct address_space *mapping = page_mapping(page); |
| 4532 | |
| 4533 | if (mapping_cap_account_dirty(mapping)) { |
Johannes Weiner | 71cd311 | 2017-05-03 14:55:13 -0700 | [diff] [blame] | 4534 | __this_cpu_sub(from->stat->count[NR_FILE_DIRTY], |
Greg Thelen | c4843a7 | 2015-05-22 17:13:16 -0400 | [diff] [blame] | 4535 | nr_pages); |
Johannes Weiner | 71cd311 | 2017-05-03 14:55:13 -0700 | [diff] [blame] | 4536 | __this_cpu_add(to->stat->count[NR_FILE_DIRTY], |
Greg Thelen | c4843a7 | 2015-05-22 17:13:16 -0400 | [diff] [blame] | 4537 | nr_pages); |
| 4538 | } |
| 4539 | } |
| 4540 | |
Chen Gang | b1b0dea | 2015-04-14 15:47:35 -0700 | [diff] [blame] | 4541 | if (PageWriteback(page)) { |
Johannes Weiner | 71cd311 | 2017-05-03 14:55:13 -0700 | [diff] [blame] | 4542 | __this_cpu_sub(from->stat->count[NR_WRITEBACK], nr_pages); |
| 4543 | __this_cpu_add(to->stat->count[NR_WRITEBACK], nr_pages); |
Chen Gang | b1b0dea | 2015-04-14 15:47:35 -0700 | [diff] [blame] | 4544 | } |
| 4545 | |
| 4546 | /* |
| 4547 | * It is safe to change page->mem_cgroup here because the page |
| 4548 | * is referenced, charged, and isolated - we can't race with |
| 4549 | * uncharging, charging, migration, or LRU putback. |
| 4550 | */ |
| 4551 | |
| 4552 | /* caller should have done css_get */ |
| 4553 | page->mem_cgroup = to; |
| 4554 | spin_unlock_irqrestore(&from->move_lock, flags); |
| 4555 | |
| 4556 | ret = 0; |
| 4557 | |
| 4558 | local_irq_disable(); |
Kirill A. Shutemov | f627c2f | 2016-01-15 16:52:20 -0800 | [diff] [blame] | 4559 | mem_cgroup_charge_statistics(to, page, compound, nr_pages); |
Chen Gang | b1b0dea | 2015-04-14 15:47:35 -0700 | [diff] [blame] | 4560 | memcg_check_events(to, page); |
Kirill A. Shutemov | f627c2f | 2016-01-15 16:52:20 -0800 | [diff] [blame] | 4561 | mem_cgroup_charge_statistics(from, page, compound, -nr_pages); |
Chen Gang | b1b0dea | 2015-04-14 15:47:35 -0700 | [diff] [blame] | 4562 | memcg_check_events(from, page); |
| 4563 | local_irq_enable(); |
| 4564 | out_unlock: |
| 4565 | unlock_page(page); |
| 4566 | out: |
| 4567 | return ret; |
| 4568 | } |
| 4569 | |
Li RongQing | 7cf7806 | 2016-05-27 14:27:46 -0700 | [diff] [blame] | 4570 | /** |
| 4571 | * get_mctgt_type - get target type of moving charge |
| 4572 | * @vma: the vma the pte to be checked belongs |
| 4573 | * @addr: the address corresponding to the pte to be checked |
| 4574 | * @ptent: the pte to be checked |
| 4575 | * @target: the pointer the target page or swap ent will be stored(can be NULL) |
| 4576 | * |
| 4577 | * Returns |
| 4578 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. |
| 4579 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for |
| 4580 | * move charge. if @target is not NULL, the page is stored in target->page |
| 4581 | * with extra refcnt got(Callers should handle it). |
| 4582 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a |
| 4583 | * target for charge migration. if @target is not NULL, the entry is stored |
| 4584 | * in target->ent. |
| 4585 | * |
| 4586 | * Called with pte lock held. |
| 4587 | */ |
| 4588 | |
Naoya Horiguchi | 8d32ff8 | 2012-03-21 16:34:27 -0700 | [diff] [blame] | 4589 | static enum mc_target_type get_mctgt_type(struct vm_area_struct *vma, |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4590 | unsigned long addr, pte_t ptent, union mc_target *target) |
| 4591 | { |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4592 | struct page *page = NULL; |
Naoya Horiguchi | 8d32ff8 | 2012-03-21 16:34:27 -0700 | [diff] [blame] | 4593 | enum mc_target_type ret = MC_TARGET_NONE; |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4594 | swp_entry_t ent = { .val = 0 }; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4595 | |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 4596 | if (pte_present(ptent)) |
| 4597 | page = mc_handle_present_pte(vma, addr, ptent); |
| 4598 | else if (is_swap_pte(ptent)) |
Li RongQing | 48406ef | 2016-07-26 15:22:14 -0700 | [diff] [blame] | 4599 | page = mc_handle_swap_pte(vma, ptent, &ent); |
Kirill A. Shutemov | 0661a33 | 2015-02-10 14:10:04 -0800 | [diff] [blame] | 4600 | else if (pte_none(ptent)) |
Daisuke Nishimura | 87946a7 | 2010-05-26 14:42:39 -0700 | [diff] [blame] | 4601 | page = mc_handle_file_pte(vma, addr, ptent, &ent); |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 4602 | |
| 4603 | if (!page && !ent.val) |
Naoya Horiguchi | 8d32ff8 | 2012-03-21 16:34:27 -0700 | [diff] [blame] | 4604 | return ret; |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4605 | if (page) { |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4606 | /* |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 4607 | * Do only loose check w/o serialization. |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 4608 | * mem_cgroup_move_account() checks the page is valid or |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 4609 | * not under LRU exclusion. |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4610 | */ |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 4611 | if (page->mem_cgroup == mc.from) { |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4612 | ret = MC_TARGET_PAGE; |
| 4613 | if (target) |
| 4614 | target->page = page; |
| 4615 | } |
| 4616 | if (!ret || !target) |
| 4617 | put_page(page); |
| 4618 | } |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 4619 | /* There is a swap entry and a page doesn't exist or isn't charged */ |
| 4620 | if (ent.val && !ret && |
Li Zefan | 34c00c3 | 2013-09-23 16:56:01 +0800 | [diff] [blame] | 4621 | mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) { |
KAMEZAWA Hiroyuki | 7f0f154 | 2010-05-11 14:06:58 -0700 | [diff] [blame] | 4622 | ret = MC_TARGET_SWAP; |
| 4623 | if (target) |
| 4624 | target->ent = ent; |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4625 | } |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4626 | return ret; |
| 4627 | } |
| 4628 | |
Naoya Horiguchi | 1272485 | 2012-03-21 16:34:28 -0700 | [diff] [blame] | 4629 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
| 4630 | /* |
| 4631 | * We don't consider swapping or file mapped pages because THP does not |
| 4632 | * support them for now. |
| 4633 | * Caller should make sure that pmd_trans_huge(pmd) is true. |
| 4634 | */ |
| 4635 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, |
| 4636 | unsigned long addr, pmd_t pmd, union mc_target *target) |
| 4637 | { |
| 4638 | struct page *page = NULL; |
Naoya Horiguchi | 1272485 | 2012-03-21 16:34:28 -0700 | [diff] [blame] | 4639 | enum mc_target_type ret = MC_TARGET_NONE; |
| 4640 | |
| 4641 | page = pmd_page(pmd); |
Sasha Levin | 309381fea | 2014-01-23 15:52:54 -0800 | [diff] [blame] | 4642 | VM_BUG_ON_PAGE(!page || !PageHead(page), page); |
Johannes Weiner | 1dfab5a | 2015-02-11 15:26:09 -0800 | [diff] [blame] | 4643 | if (!(mc.flags & MOVE_ANON)) |
Naoya Horiguchi | 1272485 | 2012-03-21 16:34:28 -0700 | [diff] [blame] | 4644 | return ret; |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 4645 | if (page->mem_cgroup == mc.from) { |
Naoya Horiguchi | 1272485 | 2012-03-21 16:34:28 -0700 | [diff] [blame] | 4646 | ret = MC_TARGET_PAGE; |
| 4647 | if (target) { |
| 4648 | get_page(page); |
| 4649 | target->page = page; |
| 4650 | } |
| 4651 | } |
| 4652 | return ret; |
| 4653 | } |
| 4654 | #else |
| 4655 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, |
| 4656 | unsigned long addr, pmd_t pmd, union mc_target *target) |
| 4657 | { |
| 4658 | return MC_TARGET_NONE; |
| 4659 | } |
| 4660 | #endif |
| 4661 | |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4662 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
| 4663 | unsigned long addr, unsigned long end, |
| 4664 | struct mm_walk *walk) |
| 4665 | { |
Naoya Horiguchi | 26bcd64 | 2015-02-11 15:27:57 -0800 | [diff] [blame] | 4666 | struct vm_area_struct *vma = walk->vma; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4667 | pte_t *pte; |
| 4668 | spinlock_t *ptl; |
| 4669 | |
Kirill A. Shutemov | b6ec57f | 2016-01-21 16:40:25 -0800 | [diff] [blame] | 4670 | ptl = pmd_trans_huge_lock(pmd, vma); |
| 4671 | if (ptl) { |
Naoya Horiguchi | 1272485 | 2012-03-21 16:34:28 -0700 | [diff] [blame] | 4672 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) |
| 4673 | mc.precharge += HPAGE_PMD_NR; |
Kirill A. Shutemov | bf92915 | 2013-11-14 14:30:54 -0800 | [diff] [blame] | 4674 | spin_unlock(ptl); |
Andrea Arcangeli | 1a5a990 | 2012-03-21 16:33:42 -0700 | [diff] [blame] | 4675 | return 0; |
Naoya Horiguchi | 1272485 | 2012-03-21 16:34:28 -0700 | [diff] [blame] | 4676 | } |
Dave Hansen | 0331932 | 2011-03-22 16:32:56 -0700 | [diff] [blame] | 4677 | |
Andrea Arcangeli | 45f83ce | 2012-03-28 14:42:40 -0700 | [diff] [blame] | 4678 | if (pmd_trans_unstable(pmd)) |
| 4679 | return 0; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4680 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
| 4681 | for (; addr != end; pte++, addr += PAGE_SIZE) |
Naoya Horiguchi | 8d32ff8 | 2012-03-21 16:34:27 -0700 | [diff] [blame] | 4682 | if (get_mctgt_type(vma, addr, *pte, NULL)) |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4683 | mc.precharge++; /* increment precharge temporarily */ |
| 4684 | pte_unmap_unlock(pte - 1, ptl); |
| 4685 | cond_resched(); |
| 4686 | |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 4687 | return 0; |
| 4688 | } |
| 4689 | |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4690 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
| 4691 | { |
| 4692 | unsigned long precharge; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4693 | |
Naoya Horiguchi | 26bcd64 | 2015-02-11 15:27:57 -0800 | [diff] [blame] | 4694 | struct mm_walk mem_cgroup_count_precharge_walk = { |
| 4695 | .pmd_entry = mem_cgroup_count_precharge_pte_range, |
| 4696 | .mm = mm, |
| 4697 | }; |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 4698 | down_read(&mm->mmap_sem); |
James Morse | 0247f3f | 2016-10-07 17:00:12 -0700 | [diff] [blame] | 4699 | walk_page_range(0, mm->highest_vm_end, |
| 4700 | &mem_cgroup_count_precharge_walk); |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 4701 | up_read(&mm->mmap_sem); |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4702 | |
| 4703 | precharge = mc.precharge; |
| 4704 | mc.precharge = 0; |
| 4705 | |
| 4706 | return precharge; |
| 4707 | } |
| 4708 | |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4709 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
| 4710 | { |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 4711 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
| 4712 | |
| 4713 | VM_BUG_ON(mc.moving_task); |
| 4714 | mc.moving_task = current; |
| 4715 | return mem_cgroup_do_precharge(precharge); |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4716 | } |
| 4717 | |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 4718 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
| 4719 | static void __mem_cgroup_clear_mc(void) |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4720 | { |
KAMEZAWA Hiroyuki | 2bd9bb2 | 2010-08-10 18:02:58 -0700 | [diff] [blame] | 4721 | struct mem_cgroup *from = mc.from; |
| 4722 | struct mem_cgroup *to = mc.to; |
| 4723 | |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4724 | /* we must uncharge all the leftover precharges from mc.to */ |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 4725 | if (mc.precharge) { |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 4726 | cancel_charge(mc.to, mc.precharge); |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 4727 | mc.precharge = 0; |
| 4728 | } |
| 4729 | /* |
| 4730 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so |
| 4731 | * we must uncharge here. |
| 4732 | */ |
| 4733 | if (mc.moved_charge) { |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 4734 | cancel_charge(mc.from, mc.moved_charge); |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 4735 | mc.moved_charge = 0; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4736 | } |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 4737 | /* we must fixup refcnts and charges */ |
| 4738 | if (mc.moved_swap) { |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 4739 | /* uncharge swap account from the old cgroup */ |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 4740 | if (!mem_cgroup_is_root(mc.from)) |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 4741 | page_counter_uncharge(&mc.from->memsw, mc.moved_swap); |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 4742 | |
Vladimir Davydov | 615d66c | 2016-08-11 15:33:03 -0700 | [diff] [blame] | 4743 | mem_cgroup_id_put_many(mc.from, mc.moved_swap); |
| 4744 | |
Johannes Weiner | 05b8430 | 2014-08-06 16:05:59 -0700 | [diff] [blame] | 4745 | /* |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 4746 | * we charged both to->memory and to->memsw, so we |
| 4747 | * should uncharge to->memory. |
Johannes Weiner | 05b8430 | 2014-08-06 16:05:59 -0700 | [diff] [blame] | 4748 | */ |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 4749 | if (!mem_cgroup_is_root(mc.to)) |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 4750 | page_counter_uncharge(&mc.to->memory, mc.moved_swap); |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 4751 | |
Vladimir Davydov | 615d66c | 2016-08-11 15:33:03 -0700 | [diff] [blame] | 4752 | mem_cgroup_id_get_many(mc.to, mc.moved_swap); |
| 4753 | css_put_many(&mc.to->css, mc.moved_swap); |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 4754 | |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 4755 | mc.moved_swap = 0; |
| 4756 | } |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 4757 | memcg_oom_recover(from); |
| 4758 | memcg_oom_recover(to); |
| 4759 | wake_up_all(&mc.waitq); |
| 4760 | } |
| 4761 | |
| 4762 | static void mem_cgroup_clear_mc(void) |
| 4763 | { |
Tejun Heo | 264a0ae | 2016-04-21 19:09:02 -0400 | [diff] [blame] | 4764 | struct mm_struct *mm = mc.mm; |
| 4765 | |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 4766 | /* |
| 4767 | * we must clear moving_task before waking up waiters at the end of |
| 4768 | * task migration. |
| 4769 | */ |
| 4770 | mc.moving_task = NULL; |
| 4771 | __mem_cgroup_clear_mc(); |
KAMEZAWA Hiroyuki | 2bd9bb2 | 2010-08-10 18:02:58 -0700 | [diff] [blame] | 4772 | spin_lock(&mc.lock); |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4773 | mc.from = NULL; |
| 4774 | mc.to = NULL; |
Tejun Heo | 264a0ae | 2016-04-21 19:09:02 -0400 | [diff] [blame] | 4775 | mc.mm = NULL; |
KAMEZAWA Hiroyuki | 2bd9bb2 | 2010-08-10 18:02:58 -0700 | [diff] [blame] | 4776 | spin_unlock(&mc.lock); |
Tejun Heo | 264a0ae | 2016-04-21 19:09:02 -0400 | [diff] [blame] | 4777 | |
| 4778 | mmput(mm); |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4779 | } |
| 4780 | |
Tejun Heo | 1f7dd3e5 | 2015-12-03 10:18:21 -0500 | [diff] [blame] | 4781 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 4782 | { |
Tejun Heo | 1f7dd3e5 | 2015-12-03 10:18:21 -0500 | [diff] [blame] | 4783 | struct cgroup_subsys_state *css; |
Ross Zwisler | eed67d7 | 2015-12-23 14:53:27 -0700 | [diff] [blame] | 4784 | struct mem_cgroup *memcg = NULL; /* unneeded init to make gcc happy */ |
Tejun Heo | 9f2115f | 2015-09-08 15:01:10 -0700 | [diff] [blame] | 4785 | struct mem_cgroup *from; |
Tejun Heo | 4530edd | 2015-09-11 15:00:19 -0400 | [diff] [blame] | 4786 | struct task_struct *leader, *p; |
Tejun Heo | 9f2115f | 2015-09-08 15:01:10 -0700 | [diff] [blame] | 4787 | struct mm_struct *mm; |
Johannes Weiner | 1dfab5a | 2015-02-11 15:26:09 -0800 | [diff] [blame] | 4788 | unsigned long move_flags; |
Tejun Heo | 9f2115f | 2015-09-08 15:01:10 -0700 | [diff] [blame] | 4789 | int ret = 0; |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 4790 | |
Tejun Heo | 1f7dd3e5 | 2015-12-03 10:18:21 -0500 | [diff] [blame] | 4791 | /* charge immigration isn't supported on the default hierarchy */ |
| 4792 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
Tejun Heo | 9f2115f | 2015-09-08 15:01:10 -0700 | [diff] [blame] | 4793 | return 0; |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 4794 | |
Tejun Heo | 4530edd | 2015-09-11 15:00:19 -0400 | [diff] [blame] | 4795 | /* |
| 4796 | * Multi-process migrations only happen on the default hierarchy |
| 4797 | * where charge immigration is not used. Perform charge |
| 4798 | * immigration if @tset contains a leader and whine if there are |
| 4799 | * multiple. |
| 4800 | */ |
| 4801 | p = NULL; |
Tejun Heo | 1f7dd3e5 | 2015-12-03 10:18:21 -0500 | [diff] [blame] | 4802 | cgroup_taskset_for_each_leader(leader, css, tset) { |
Tejun Heo | 4530edd | 2015-09-11 15:00:19 -0400 | [diff] [blame] | 4803 | WARN_ON_ONCE(p); |
| 4804 | p = leader; |
Tejun Heo | 1f7dd3e5 | 2015-12-03 10:18:21 -0500 | [diff] [blame] | 4805 | memcg = mem_cgroup_from_css(css); |
Tejun Heo | 4530edd | 2015-09-11 15:00:19 -0400 | [diff] [blame] | 4806 | } |
| 4807 | if (!p) |
| 4808 | return 0; |
| 4809 | |
Tejun Heo | 1f7dd3e5 | 2015-12-03 10:18:21 -0500 | [diff] [blame] | 4810 | /* |
| 4811 | * We are now commited to this value whatever it is. Changes in this |
| 4812 | * tunable will only affect upcoming migrations, not the current one. |
| 4813 | * So we need to save it, and keep it going. |
| 4814 | */ |
| 4815 | move_flags = READ_ONCE(memcg->move_charge_at_immigrate); |
| 4816 | if (!move_flags) |
| 4817 | return 0; |
| 4818 | |
Tejun Heo | 9f2115f | 2015-09-08 15:01:10 -0700 | [diff] [blame] | 4819 | from = mem_cgroup_from_task(p); |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 4820 | |
Tejun Heo | 9f2115f | 2015-09-08 15:01:10 -0700 | [diff] [blame] | 4821 | VM_BUG_ON(from == memcg); |
Johannes Weiner | 247b144 | 2014-12-10 15:44:11 -0800 | [diff] [blame] | 4822 | |
Tejun Heo | 9f2115f | 2015-09-08 15:01:10 -0700 | [diff] [blame] | 4823 | mm = get_task_mm(p); |
| 4824 | if (!mm) |
| 4825 | return 0; |
| 4826 | /* We move charges only when we move a owner of the mm */ |
| 4827 | if (mm->owner == p) { |
| 4828 | VM_BUG_ON(mc.from); |
| 4829 | VM_BUG_ON(mc.to); |
| 4830 | VM_BUG_ON(mc.precharge); |
| 4831 | VM_BUG_ON(mc.moved_charge); |
| 4832 | VM_BUG_ON(mc.moved_swap); |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 4833 | |
Tejun Heo | 9f2115f | 2015-09-08 15:01:10 -0700 | [diff] [blame] | 4834 | spin_lock(&mc.lock); |
Tejun Heo | 264a0ae | 2016-04-21 19:09:02 -0400 | [diff] [blame] | 4835 | mc.mm = mm; |
Tejun Heo | 9f2115f | 2015-09-08 15:01:10 -0700 | [diff] [blame] | 4836 | mc.from = from; |
| 4837 | mc.to = memcg; |
| 4838 | mc.flags = move_flags; |
| 4839 | spin_unlock(&mc.lock); |
| 4840 | /* We set mc.moving_task later */ |
| 4841 | |
| 4842 | ret = mem_cgroup_precharge_mc(mm); |
| 4843 | if (ret) |
| 4844 | mem_cgroup_clear_mc(); |
Tejun Heo | 264a0ae | 2016-04-21 19:09:02 -0400 | [diff] [blame] | 4845 | } else { |
| 4846 | mmput(mm); |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 4847 | } |
| 4848 | return ret; |
| 4849 | } |
| 4850 | |
Tejun Heo | 1f7dd3e5 | 2015-12-03 10:18:21 -0500 | [diff] [blame] | 4851 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 4852 | { |
Johannes Weiner | 4e2f245 | 2014-12-10 15:44:08 -0800 | [diff] [blame] | 4853 | if (mc.to) |
| 4854 | mem_cgroup_clear_mc(); |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 4855 | } |
| 4856 | |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4857 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
| 4858 | unsigned long addr, unsigned long end, |
| 4859 | struct mm_walk *walk) |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 4860 | { |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4861 | int ret = 0; |
Naoya Horiguchi | 26bcd64 | 2015-02-11 15:27:57 -0800 | [diff] [blame] | 4862 | struct vm_area_struct *vma = walk->vma; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4863 | pte_t *pte; |
| 4864 | spinlock_t *ptl; |
Naoya Horiguchi | 1272485 | 2012-03-21 16:34:28 -0700 | [diff] [blame] | 4865 | enum mc_target_type target_type; |
| 4866 | union mc_target target; |
| 4867 | struct page *page; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4868 | |
Kirill A. Shutemov | b6ec57f | 2016-01-21 16:40:25 -0800 | [diff] [blame] | 4869 | ptl = pmd_trans_huge_lock(pmd, vma); |
| 4870 | if (ptl) { |
Hugh Dickins | 62ade86 | 2012-05-18 11:28:34 -0700 | [diff] [blame] | 4871 | if (mc.precharge < HPAGE_PMD_NR) { |
Kirill A. Shutemov | bf92915 | 2013-11-14 14:30:54 -0800 | [diff] [blame] | 4872 | spin_unlock(ptl); |
Naoya Horiguchi | 1272485 | 2012-03-21 16:34:28 -0700 | [diff] [blame] | 4873 | return 0; |
| 4874 | } |
| 4875 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); |
| 4876 | if (target_type == MC_TARGET_PAGE) { |
| 4877 | page = target.page; |
| 4878 | if (!isolate_lru_page(page)) { |
Kirill A. Shutemov | f627c2f | 2016-01-15 16:52:20 -0800 | [diff] [blame] | 4879 | if (!mem_cgroup_move_account(page, true, |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 4880 | mc.from, mc.to)) { |
Naoya Horiguchi | 1272485 | 2012-03-21 16:34:28 -0700 | [diff] [blame] | 4881 | mc.precharge -= HPAGE_PMD_NR; |
| 4882 | mc.moved_charge += HPAGE_PMD_NR; |
| 4883 | } |
| 4884 | putback_lru_page(page); |
| 4885 | } |
| 4886 | put_page(page); |
| 4887 | } |
Kirill A. Shutemov | bf92915 | 2013-11-14 14:30:54 -0800 | [diff] [blame] | 4888 | spin_unlock(ptl); |
Andrea Arcangeli | 1a5a990 | 2012-03-21 16:33:42 -0700 | [diff] [blame] | 4889 | return 0; |
Naoya Horiguchi | 1272485 | 2012-03-21 16:34:28 -0700 | [diff] [blame] | 4890 | } |
| 4891 | |
Andrea Arcangeli | 45f83ce | 2012-03-28 14:42:40 -0700 | [diff] [blame] | 4892 | if (pmd_trans_unstable(pmd)) |
| 4893 | return 0; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4894 | retry: |
| 4895 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
| 4896 | for (; addr != end; addr += PAGE_SIZE) { |
| 4897 | pte_t ptent = *(pte++); |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4898 | swp_entry_t ent; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4899 | |
| 4900 | if (!mc.precharge) |
| 4901 | break; |
| 4902 | |
Naoya Horiguchi | 8d32ff8 | 2012-03-21 16:34:27 -0700 | [diff] [blame] | 4903 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4904 | case MC_TARGET_PAGE: |
| 4905 | page = target.page; |
Kirill A. Shutemov | 53f9263 | 2016-01-15 16:53:42 -0800 | [diff] [blame] | 4906 | /* |
| 4907 | * We can have a part of the split pmd here. Moving it |
| 4908 | * can be done but it would be too convoluted so simply |
| 4909 | * ignore such a partial THP and keep it in original |
| 4910 | * memcg. There should be somebody mapping the head. |
| 4911 | */ |
| 4912 | if (PageTransCompound(page)) |
| 4913 | goto put; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4914 | if (isolate_lru_page(page)) |
| 4915 | goto put; |
Kirill A. Shutemov | f627c2f | 2016-01-15 16:52:20 -0800 | [diff] [blame] | 4916 | if (!mem_cgroup_move_account(page, false, |
| 4917 | mc.from, mc.to)) { |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4918 | mc.precharge--; |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 4919 | /* we uncharge from mc.from later. */ |
| 4920 | mc.moved_charge++; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4921 | } |
| 4922 | putback_lru_page(page); |
Naoya Horiguchi | 8d32ff8 | 2012-03-21 16:34:27 -0700 | [diff] [blame] | 4923 | put: /* get_mctgt_type() gets the page */ |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4924 | put_page(page); |
| 4925 | break; |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4926 | case MC_TARGET_SWAP: |
| 4927 | ent = target.ent; |
Hugh Dickins | e91cbb4 | 2012-05-29 15:06:51 -0700 | [diff] [blame] | 4928 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4929 | mc.precharge--; |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 4930 | /* we fixup refcnts and charges later. */ |
| 4931 | mc.moved_swap++; |
| 4932 | } |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 4933 | break; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4934 | default: |
| 4935 | break; |
| 4936 | } |
| 4937 | } |
| 4938 | pte_unmap_unlock(pte - 1, ptl); |
| 4939 | cond_resched(); |
| 4940 | |
| 4941 | if (addr != end) { |
| 4942 | /* |
| 4943 | * We have consumed all precharges we got in can_attach(). |
| 4944 | * We try charge one by one, but don't do any additional |
| 4945 | * charges to mc.to if we have failed in charge once in attach() |
| 4946 | * phase. |
| 4947 | */ |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 4948 | ret = mem_cgroup_do_precharge(1); |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4949 | if (!ret) |
| 4950 | goto retry; |
| 4951 | } |
| 4952 | |
| 4953 | return ret; |
| 4954 | } |
| 4955 | |
Tejun Heo | 264a0ae | 2016-04-21 19:09:02 -0400 | [diff] [blame] | 4956 | static void mem_cgroup_move_charge(void) |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4957 | { |
Naoya Horiguchi | 26bcd64 | 2015-02-11 15:27:57 -0800 | [diff] [blame] | 4958 | struct mm_walk mem_cgroup_move_charge_walk = { |
| 4959 | .pmd_entry = mem_cgroup_move_charge_pte_range, |
Tejun Heo | 264a0ae | 2016-04-21 19:09:02 -0400 | [diff] [blame] | 4960 | .mm = mc.mm, |
Naoya Horiguchi | 26bcd64 | 2015-02-11 15:27:57 -0800 | [diff] [blame] | 4961 | }; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 4962 | |
| 4963 | lru_add_drain_all(); |
Johannes Weiner | 312722c | 2014-12-10 15:44:25 -0800 | [diff] [blame] | 4964 | /* |
Johannes Weiner | 81f8c3a | 2016-03-15 14:57:04 -0700 | [diff] [blame] | 4965 | * Signal lock_page_memcg() to take the memcg's move_lock |
| 4966 | * while we're moving its pages to another memcg. Then wait |
| 4967 | * for already started RCU-only updates to finish. |
Johannes Weiner | 312722c | 2014-12-10 15:44:25 -0800 | [diff] [blame] | 4968 | */ |
| 4969 | atomic_inc(&mc.from->moving_account); |
| 4970 | synchronize_rcu(); |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 4971 | retry: |
Tejun Heo | 264a0ae | 2016-04-21 19:09:02 -0400 | [diff] [blame] | 4972 | if (unlikely(!down_read_trylock(&mc.mm->mmap_sem))) { |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 4973 | /* |
| 4974 | * Someone who are holding the mmap_sem might be waiting in |
| 4975 | * waitq. So we cancel all extra charges, wake up all waiters, |
| 4976 | * and retry. Because we cancel precharges, we might not be able |
| 4977 | * to move enough charges, but moving charge is a best-effort |
| 4978 | * feature anyway, so it wouldn't be a big problem. |
| 4979 | */ |
| 4980 | __mem_cgroup_clear_mc(); |
| 4981 | cond_resched(); |
| 4982 | goto retry; |
| 4983 | } |
Naoya Horiguchi | 26bcd64 | 2015-02-11 15:27:57 -0800 | [diff] [blame] | 4984 | /* |
| 4985 | * When we have consumed all precharges and failed in doing |
| 4986 | * additional charge, the page walk just aborts. |
| 4987 | */ |
James Morse | 0247f3f | 2016-10-07 17:00:12 -0700 | [diff] [blame] | 4988 | walk_page_range(0, mc.mm->highest_vm_end, &mem_cgroup_move_charge_walk); |
| 4989 | |
Tejun Heo | 264a0ae | 2016-04-21 19:09:02 -0400 | [diff] [blame] | 4990 | up_read(&mc.mm->mmap_sem); |
Johannes Weiner | 312722c | 2014-12-10 15:44:25 -0800 | [diff] [blame] | 4991 | atomic_dec(&mc.from->moving_account); |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 4992 | } |
| 4993 | |
Tejun Heo | 264a0ae | 2016-04-21 19:09:02 -0400 | [diff] [blame] | 4994 | static void mem_cgroup_move_task(void) |
Balbir Singh | 67e465a | 2008-02-07 00:13:54 -0800 | [diff] [blame] | 4995 | { |
Tejun Heo | 264a0ae | 2016-04-21 19:09:02 -0400 | [diff] [blame] | 4996 | if (mc.to) { |
| 4997 | mem_cgroup_move_charge(); |
KOSAKI Motohiro | a433658 | 2011-06-15 15:08:13 -0700 | [diff] [blame] | 4998 | mem_cgroup_clear_mc(); |
Tejun Heo | 264a0ae | 2016-04-21 19:09:02 -0400 | [diff] [blame] | 4999 | } |
Balbir Singh | 67e465a | 2008-02-07 00:13:54 -0800 | [diff] [blame] | 5000 | } |
Daisuke Nishimura | 5cfb80a | 2010-03-23 13:35:11 -0700 | [diff] [blame] | 5001 | #else /* !CONFIG_MMU */ |
Tejun Heo | 1f7dd3e5 | 2015-12-03 10:18:21 -0500 | [diff] [blame] | 5002 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
Daisuke Nishimura | 5cfb80a | 2010-03-23 13:35:11 -0700 | [diff] [blame] | 5003 | { |
| 5004 | return 0; |
| 5005 | } |
Tejun Heo | 1f7dd3e5 | 2015-12-03 10:18:21 -0500 | [diff] [blame] | 5006 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
Daisuke Nishimura | 5cfb80a | 2010-03-23 13:35:11 -0700 | [diff] [blame] | 5007 | { |
| 5008 | } |
Tejun Heo | 264a0ae | 2016-04-21 19:09:02 -0400 | [diff] [blame] | 5009 | static void mem_cgroup_move_task(void) |
Daisuke Nishimura | 5cfb80a | 2010-03-23 13:35:11 -0700 | [diff] [blame] | 5010 | { |
| 5011 | } |
| 5012 | #endif |
Balbir Singh | 67e465a | 2008-02-07 00:13:54 -0800 | [diff] [blame] | 5013 | |
Tejun Heo | f00baae | 2013-04-15 13:41:15 -0700 | [diff] [blame] | 5014 | /* |
| 5015 | * Cgroup retains root cgroups across [un]mount cycles making it necessary |
Tejun Heo | aa6ec29 | 2014-07-09 10:08:08 -0400 | [diff] [blame] | 5016 | * to verify whether we're attached to the default hierarchy on each mount |
| 5017 | * attempt. |
Tejun Heo | f00baae | 2013-04-15 13:41:15 -0700 | [diff] [blame] | 5018 | */ |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 5019 | static void mem_cgroup_bind(struct cgroup_subsys_state *root_css) |
Tejun Heo | f00baae | 2013-04-15 13:41:15 -0700 | [diff] [blame] | 5020 | { |
| 5021 | /* |
Tejun Heo | aa6ec29 | 2014-07-09 10:08:08 -0400 | [diff] [blame] | 5022 | * use_hierarchy is forced on the default hierarchy. cgroup core |
Tejun Heo | f00baae | 2013-04-15 13:41:15 -0700 | [diff] [blame] | 5023 | * guarantees that @root doesn't have any children, so turning it |
| 5024 | * on for the root memcg is enough. |
| 5025 | */ |
Tejun Heo | 9e10a13 | 2015-09-18 11:56:28 -0400 | [diff] [blame] | 5026 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
Vladimir Davydov | 7feee59 | 2015-03-12 16:26:19 -0700 | [diff] [blame] | 5027 | root_mem_cgroup->use_hierarchy = true; |
| 5028 | else |
| 5029 | root_mem_cgroup->use_hierarchy = false; |
Tejun Heo | f00baae | 2013-04-15 13:41:15 -0700 | [diff] [blame] | 5030 | } |
| 5031 | |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5032 | static u64 memory_current_read(struct cgroup_subsys_state *css, |
| 5033 | struct cftype *cft) |
| 5034 | { |
Johannes Weiner | f5fc3c5d | 2015-11-05 18:50:23 -0800 | [diff] [blame] | 5035 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
| 5036 | |
| 5037 | return (u64)page_counter_read(&memcg->memory) * PAGE_SIZE; |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5038 | } |
| 5039 | |
| 5040 | static int memory_low_show(struct seq_file *m, void *v) |
| 5041 | { |
| 5042 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
Jason Low | 4db0c3c | 2015-04-15 16:14:08 -0700 | [diff] [blame] | 5043 | unsigned long low = READ_ONCE(memcg->low); |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5044 | |
| 5045 | if (low == PAGE_COUNTER_MAX) |
Johannes Weiner | d297369 | 2015-02-27 15:52:04 -0800 | [diff] [blame] | 5046 | seq_puts(m, "max\n"); |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5047 | else |
| 5048 | seq_printf(m, "%llu\n", (u64)low * PAGE_SIZE); |
| 5049 | |
| 5050 | return 0; |
| 5051 | } |
| 5052 | |
| 5053 | static ssize_t memory_low_write(struct kernfs_open_file *of, |
| 5054 | char *buf, size_t nbytes, loff_t off) |
| 5055 | { |
| 5056 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
| 5057 | unsigned long low; |
| 5058 | int err; |
| 5059 | |
| 5060 | buf = strstrip(buf); |
Johannes Weiner | d297369 | 2015-02-27 15:52:04 -0800 | [diff] [blame] | 5061 | err = page_counter_memparse(buf, "max", &low); |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5062 | if (err) |
| 5063 | return err; |
| 5064 | |
| 5065 | memcg->low = low; |
| 5066 | |
| 5067 | return nbytes; |
| 5068 | } |
| 5069 | |
| 5070 | static int memory_high_show(struct seq_file *m, void *v) |
| 5071 | { |
| 5072 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
Jason Low | 4db0c3c | 2015-04-15 16:14:08 -0700 | [diff] [blame] | 5073 | unsigned long high = READ_ONCE(memcg->high); |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5074 | |
| 5075 | if (high == PAGE_COUNTER_MAX) |
Johannes Weiner | d297369 | 2015-02-27 15:52:04 -0800 | [diff] [blame] | 5076 | seq_puts(m, "max\n"); |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5077 | else |
| 5078 | seq_printf(m, "%llu\n", (u64)high * PAGE_SIZE); |
| 5079 | |
| 5080 | return 0; |
| 5081 | } |
| 5082 | |
| 5083 | static ssize_t memory_high_write(struct kernfs_open_file *of, |
| 5084 | char *buf, size_t nbytes, loff_t off) |
| 5085 | { |
| 5086 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
Johannes Weiner | 588083b | 2016-03-17 14:20:25 -0700 | [diff] [blame] | 5087 | unsigned long nr_pages; |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5088 | unsigned long high; |
| 5089 | int err; |
| 5090 | |
| 5091 | buf = strstrip(buf); |
Johannes Weiner | d297369 | 2015-02-27 15:52:04 -0800 | [diff] [blame] | 5092 | err = page_counter_memparse(buf, "max", &high); |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5093 | if (err) |
| 5094 | return err; |
| 5095 | |
| 5096 | memcg->high = high; |
| 5097 | |
Johannes Weiner | 588083b | 2016-03-17 14:20:25 -0700 | [diff] [blame] | 5098 | nr_pages = page_counter_read(&memcg->memory); |
| 5099 | if (nr_pages > high) |
| 5100 | try_to_free_mem_cgroup_pages(memcg, nr_pages - high, |
| 5101 | GFP_KERNEL, true); |
| 5102 | |
Tejun Heo | 2529bb3 | 2015-05-22 18:23:34 -0400 | [diff] [blame] | 5103 | memcg_wb_domain_size_changed(memcg); |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5104 | return nbytes; |
| 5105 | } |
| 5106 | |
| 5107 | static int memory_max_show(struct seq_file *m, void *v) |
| 5108 | { |
| 5109 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
Jason Low | 4db0c3c | 2015-04-15 16:14:08 -0700 | [diff] [blame] | 5110 | unsigned long max = READ_ONCE(memcg->memory.limit); |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5111 | |
| 5112 | if (max == PAGE_COUNTER_MAX) |
Johannes Weiner | d297369 | 2015-02-27 15:52:04 -0800 | [diff] [blame] | 5113 | seq_puts(m, "max\n"); |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5114 | else |
| 5115 | seq_printf(m, "%llu\n", (u64)max * PAGE_SIZE); |
| 5116 | |
| 5117 | return 0; |
| 5118 | } |
| 5119 | |
| 5120 | static ssize_t memory_max_write(struct kernfs_open_file *of, |
| 5121 | char *buf, size_t nbytes, loff_t off) |
| 5122 | { |
| 5123 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
Johannes Weiner | b6e6edc | 2016-03-17 14:20:28 -0700 | [diff] [blame] | 5124 | unsigned int nr_reclaims = MEM_CGROUP_RECLAIM_RETRIES; |
| 5125 | bool drained = false; |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5126 | unsigned long max; |
| 5127 | int err; |
| 5128 | |
| 5129 | buf = strstrip(buf); |
Johannes Weiner | d297369 | 2015-02-27 15:52:04 -0800 | [diff] [blame] | 5130 | err = page_counter_memparse(buf, "max", &max); |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5131 | if (err) |
| 5132 | return err; |
| 5133 | |
Johannes Weiner | b6e6edc | 2016-03-17 14:20:28 -0700 | [diff] [blame] | 5134 | xchg(&memcg->memory.limit, max); |
| 5135 | |
| 5136 | for (;;) { |
| 5137 | unsigned long nr_pages = page_counter_read(&memcg->memory); |
| 5138 | |
| 5139 | if (nr_pages <= max) |
| 5140 | break; |
| 5141 | |
| 5142 | if (signal_pending(current)) { |
| 5143 | err = -EINTR; |
| 5144 | break; |
| 5145 | } |
| 5146 | |
| 5147 | if (!drained) { |
| 5148 | drain_all_stock(memcg); |
| 5149 | drained = true; |
| 5150 | continue; |
| 5151 | } |
| 5152 | |
| 5153 | if (nr_reclaims) { |
| 5154 | if (!try_to_free_mem_cgroup_pages(memcg, nr_pages - max, |
| 5155 | GFP_KERNEL, true)) |
| 5156 | nr_reclaims--; |
| 5157 | continue; |
| 5158 | } |
| 5159 | |
Johannes Weiner | 31176c7 | 2017-05-03 14:55:07 -0700 | [diff] [blame] | 5160 | mem_cgroup_event(memcg, MEMCG_OOM); |
Johannes Weiner | b6e6edc | 2016-03-17 14:20:28 -0700 | [diff] [blame] | 5161 | if (!mem_cgroup_out_of_memory(memcg, GFP_KERNEL, 0)) |
| 5162 | break; |
| 5163 | } |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5164 | |
Tejun Heo | 2529bb3 | 2015-05-22 18:23:34 -0400 | [diff] [blame] | 5165 | memcg_wb_domain_size_changed(memcg); |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5166 | return nbytes; |
| 5167 | } |
| 5168 | |
| 5169 | static int memory_events_show(struct seq_file *m, void *v) |
| 5170 | { |
| 5171 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
| 5172 | |
Johannes Weiner | ccda7f4 | 2017-05-03 14:55:16 -0700 | [diff] [blame] | 5173 | seq_printf(m, "low %lu\n", memcg_sum_events(memcg, MEMCG_LOW)); |
| 5174 | seq_printf(m, "high %lu\n", memcg_sum_events(memcg, MEMCG_HIGH)); |
| 5175 | seq_printf(m, "max %lu\n", memcg_sum_events(memcg, MEMCG_MAX)); |
| 5176 | seq_printf(m, "oom %lu\n", memcg_sum_events(memcg, MEMCG_OOM)); |
Konstantin Khlebnikov | 8e675f7 | 2017-07-06 15:40:28 -0700 | [diff] [blame] | 5177 | seq_printf(m, "oom_kill %lu\n", memcg_sum_events(memcg, OOM_KILL)); |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5178 | |
| 5179 | return 0; |
| 5180 | } |
| 5181 | |
Johannes Weiner | 587d9f7 | 2016-01-20 15:03:19 -0800 | [diff] [blame] | 5182 | static int memory_stat_show(struct seq_file *m, void *v) |
| 5183 | { |
| 5184 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
Vladimir Davydov | 72b54e7 | 2016-03-17 14:17:32 -0700 | [diff] [blame] | 5185 | unsigned long stat[MEMCG_NR_STAT]; |
| 5186 | unsigned long events[MEMCG_NR_EVENTS]; |
Johannes Weiner | 587d9f7 | 2016-01-20 15:03:19 -0800 | [diff] [blame] | 5187 | int i; |
| 5188 | |
| 5189 | /* |
| 5190 | * Provide statistics on the state of the memory subsystem as |
| 5191 | * well as cumulative event counters that show past behavior. |
| 5192 | * |
| 5193 | * This list is ordered following a combination of these gradients: |
| 5194 | * 1) generic big picture -> specifics and details |
| 5195 | * 2) reflecting userspace activity -> reflecting kernel heuristics |
| 5196 | * |
| 5197 | * Current memory state: |
| 5198 | */ |
| 5199 | |
Vladimir Davydov | 72b54e7 | 2016-03-17 14:17:32 -0700 | [diff] [blame] | 5200 | tree_stat(memcg, stat); |
| 5201 | tree_events(memcg, events); |
| 5202 | |
Johannes Weiner | 587d9f7 | 2016-01-20 15:03:19 -0800 | [diff] [blame] | 5203 | seq_printf(m, "anon %llu\n", |
Johannes Weiner | 71cd311 | 2017-05-03 14:55:13 -0700 | [diff] [blame] | 5204 | (u64)stat[MEMCG_RSS] * PAGE_SIZE); |
Johannes Weiner | 587d9f7 | 2016-01-20 15:03:19 -0800 | [diff] [blame] | 5205 | seq_printf(m, "file %llu\n", |
Johannes Weiner | 71cd311 | 2017-05-03 14:55:13 -0700 | [diff] [blame] | 5206 | (u64)stat[MEMCG_CACHE] * PAGE_SIZE); |
Vladimir Davydov | 12580e4 | 2016-03-17 14:17:38 -0700 | [diff] [blame] | 5207 | seq_printf(m, "kernel_stack %llu\n", |
Andy Lutomirski | efdc949 | 2016-07-28 15:48:17 -0700 | [diff] [blame] | 5208 | (u64)stat[MEMCG_KERNEL_STACK_KB] * 1024); |
Vladimir Davydov | 27ee57c | 2016-03-17 14:17:35 -0700 | [diff] [blame] | 5209 | seq_printf(m, "slab %llu\n", |
Johannes Weiner | 3204929 | 2017-07-06 15:40:46 -0700 | [diff] [blame] | 5210 | (u64)(stat[NR_SLAB_RECLAIMABLE] + |
| 5211 | stat[NR_SLAB_UNRECLAIMABLE]) * PAGE_SIZE); |
Johannes Weiner | b2807f0 | 2016-01-20 15:03:22 -0800 | [diff] [blame] | 5212 | seq_printf(m, "sock %llu\n", |
Vladimir Davydov | 72b54e7 | 2016-03-17 14:17:32 -0700 | [diff] [blame] | 5213 | (u64)stat[MEMCG_SOCK] * PAGE_SIZE); |
Johannes Weiner | 587d9f7 | 2016-01-20 15:03:19 -0800 | [diff] [blame] | 5214 | |
Johannes Weiner | 9a4caf1 | 2017-05-03 14:52:45 -0700 | [diff] [blame] | 5215 | seq_printf(m, "shmem %llu\n", |
Johannes Weiner | 71cd311 | 2017-05-03 14:55:13 -0700 | [diff] [blame] | 5216 | (u64)stat[NR_SHMEM] * PAGE_SIZE); |
Johannes Weiner | 587d9f7 | 2016-01-20 15:03:19 -0800 | [diff] [blame] | 5217 | seq_printf(m, "file_mapped %llu\n", |
Johannes Weiner | 71cd311 | 2017-05-03 14:55:13 -0700 | [diff] [blame] | 5218 | (u64)stat[NR_FILE_MAPPED] * PAGE_SIZE); |
Johannes Weiner | 587d9f7 | 2016-01-20 15:03:19 -0800 | [diff] [blame] | 5219 | seq_printf(m, "file_dirty %llu\n", |
Johannes Weiner | 71cd311 | 2017-05-03 14:55:13 -0700 | [diff] [blame] | 5220 | (u64)stat[NR_FILE_DIRTY] * PAGE_SIZE); |
Johannes Weiner | 587d9f7 | 2016-01-20 15:03:19 -0800 | [diff] [blame] | 5221 | seq_printf(m, "file_writeback %llu\n", |
Johannes Weiner | 71cd311 | 2017-05-03 14:55:13 -0700 | [diff] [blame] | 5222 | (u64)stat[NR_WRITEBACK] * PAGE_SIZE); |
Johannes Weiner | 587d9f7 | 2016-01-20 15:03:19 -0800 | [diff] [blame] | 5223 | |
| 5224 | for (i = 0; i < NR_LRU_LISTS; i++) { |
| 5225 | struct mem_cgroup *mi; |
| 5226 | unsigned long val = 0; |
| 5227 | |
| 5228 | for_each_mem_cgroup_tree(mi, memcg) |
| 5229 | val += mem_cgroup_nr_lru_pages(mi, BIT(i)); |
| 5230 | seq_printf(m, "%s %llu\n", |
| 5231 | mem_cgroup_lru_names[i], (u64)val * PAGE_SIZE); |
| 5232 | } |
| 5233 | |
Vladimir Davydov | 27ee57c | 2016-03-17 14:17:35 -0700 | [diff] [blame] | 5234 | seq_printf(m, "slab_reclaimable %llu\n", |
Johannes Weiner | 3204929 | 2017-07-06 15:40:46 -0700 | [diff] [blame] | 5235 | (u64)stat[NR_SLAB_RECLAIMABLE] * PAGE_SIZE); |
Vladimir Davydov | 27ee57c | 2016-03-17 14:17:35 -0700 | [diff] [blame] | 5236 | seq_printf(m, "slab_unreclaimable %llu\n", |
Johannes Weiner | 3204929 | 2017-07-06 15:40:46 -0700 | [diff] [blame] | 5237 | (u64)stat[NR_SLAB_UNRECLAIMABLE] * PAGE_SIZE); |
Vladimir Davydov | 27ee57c | 2016-03-17 14:17:35 -0700 | [diff] [blame] | 5238 | |
Johannes Weiner | 587d9f7 | 2016-01-20 15:03:19 -0800 | [diff] [blame] | 5239 | /* Accumulated memory events */ |
| 5240 | |
Johannes Weiner | df0e53d | 2017-05-03 14:55:10 -0700 | [diff] [blame] | 5241 | seq_printf(m, "pgfault %lu\n", events[PGFAULT]); |
| 5242 | seq_printf(m, "pgmajfault %lu\n", events[PGMAJFAULT]); |
Johannes Weiner | 587d9f7 | 2016-01-20 15:03:19 -0800 | [diff] [blame] | 5243 | |
Roman Gushchin | 2262185 | 2017-07-06 15:40:25 -0700 | [diff] [blame] | 5244 | seq_printf(m, "pgrefill %lu\n", events[PGREFILL]); |
| 5245 | seq_printf(m, "pgscan %lu\n", events[PGSCAN_KSWAPD] + |
| 5246 | events[PGSCAN_DIRECT]); |
| 5247 | seq_printf(m, "pgsteal %lu\n", events[PGSTEAL_KSWAPD] + |
| 5248 | events[PGSTEAL_DIRECT]); |
| 5249 | seq_printf(m, "pgactivate %lu\n", events[PGACTIVATE]); |
| 5250 | seq_printf(m, "pgdeactivate %lu\n", events[PGDEACTIVATE]); |
| 5251 | seq_printf(m, "pglazyfree %lu\n", events[PGLAZYFREE]); |
| 5252 | seq_printf(m, "pglazyfreed %lu\n", events[PGLAZYFREED]); |
| 5253 | |
Johannes Weiner | 2a2e488 | 2017-05-03 14:55:03 -0700 | [diff] [blame] | 5254 | seq_printf(m, "workingset_refault %lu\n", |
Johannes Weiner | 71cd311 | 2017-05-03 14:55:13 -0700 | [diff] [blame] | 5255 | stat[WORKINGSET_REFAULT]); |
Johannes Weiner | 2a2e488 | 2017-05-03 14:55:03 -0700 | [diff] [blame] | 5256 | seq_printf(m, "workingset_activate %lu\n", |
Johannes Weiner | 71cd311 | 2017-05-03 14:55:13 -0700 | [diff] [blame] | 5257 | stat[WORKINGSET_ACTIVATE]); |
Johannes Weiner | 2a2e488 | 2017-05-03 14:55:03 -0700 | [diff] [blame] | 5258 | seq_printf(m, "workingset_nodereclaim %lu\n", |
Johannes Weiner | 71cd311 | 2017-05-03 14:55:13 -0700 | [diff] [blame] | 5259 | stat[WORKINGSET_NODERECLAIM]); |
Johannes Weiner | 2a2e488 | 2017-05-03 14:55:03 -0700 | [diff] [blame] | 5260 | |
Johannes Weiner | 587d9f7 | 2016-01-20 15:03:19 -0800 | [diff] [blame] | 5261 | return 0; |
| 5262 | } |
| 5263 | |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5264 | static struct cftype memory_files[] = { |
| 5265 | { |
| 5266 | .name = "current", |
Johannes Weiner | f5fc3c5d | 2015-11-05 18:50:23 -0800 | [diff] [blame] | 5267 | .flags = CFTYPE_NOT_ON_ROOT, |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5268 | .read_u64 = memory_current_read, |
| 5269 | }, |
| 5270 | { |
| 5271 | .name = "low", |
| 5272 | .flags = CFTYPE_NOT_ON_ROOT, |
| 5273 | .seq_show = memory_low_show, |
| 5274 | .write = memory_low_write, |
| 5275 | }, |
| 5276 | { |
| 5277 | .name = "high", |
| 5278 | .flags = CFTYPE_NOT_ON_ROOT, |
| 5279 | .seq_show = memory_high_show, |
| 5280 | .write = memory_high_write, |
| 5281 | }, |
| 5282 | { |
| 5283 | .name = "max", |
| 5284 | .flags = CFTYPE_NOT_ON_ROOT, |
| 5285 | .seq_show = memory_max_show, |
| 5286 | .write = memory_max_write, |
| 5287 | }, |
| 5288 | { |
| 5289 | .name = "events", |
| 5290 | .flags = CFTYPE_NOT_ON_ROOT, |
Tejun Heo | 472912a | 2015-09-18 18:01:59 -0400 | [diff] [blame] | 5291 | .file_offset = offsetof(struct mem_cgroup, events_file), |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5292 | .seq_show = memory_events_show, |
| 5293 | }, |
Johannes Weiner | 587d9f7 | 2016-01-20 15:03:19 -0800 | [diff] [blame] | 5294 | { |
| 5295 | .name = "stat", |
| 5296 | .flags = CFTYPE_NOT_ON_ROOT, |
| 5297 | .seq_show = memory_stat_show, |
| 5298 | }, |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5299 | { } /* terminate */ |
| 5300 | }; |
| 5301 | |
Tejun Heo | 073219e | 2014-02-08 10:36:58 -0500 | [diff] [blame] | 5302 | struct cgroup_subsys memory_cgrp_subsys = { |
Tejun Heo | 92fb974 | 2012-11-19 08:13:38 -0800 | [diff] [blame] | 5303 | .css_alloc = mem_cgroup_css_alloc, |
Glauber Costa | d142e3e | 2013-02-22 16:34:52 -0800 | [diff] [blame] | 5304 | .css_online = mem_cgroup_css_online, |
Tejun Heo | 92fb974 | 2012-11-19 08:13:38 -0800 | [diff] [blame] | 5305 | .css_offline = mem_cgroup_css_offline, |
Vladimir Davydov | 6df3868 | 2015-12-29 14:54:10 -0800 | [diff] [blame] | 5306 | .css_released = mem_cgroup_css_released, |
Tejun Heo | 92fb974 | 2012-11-19 08:13:38 -0800 | [diff] [blame] | 5307 | .css_free = mem_cgroup_css_free, |
Tejun Heo | 1ced953 | 2014-07-08 18:02:57 -0400 | [diff] [blame] | 5308 | .css_reset = mem_cgroup_css_reset, |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 5309 | .can_attach = mem_cgroup_can_attach, |
| 5310 | .cancel_attach = mem_cgroup_cancel_attach, |
Tejun Heo | 264a0ae | 2016-04-21 19:09:02 -0400 | [diff] [blame] | 5311 | .post_attach = mem_cgroup_move_task, |
Tejun Heo | f00baae | 2013-04-15 13:41:15 -0700 | [diff] [blame] | 5312 | .bind = mem_cgroup_bind, |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5313 | .dfl_cftypes = memory_files, |
| 5314 | .legacy_cftypes = mem_cgroup_legacy_files, |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 5315 | .early_init = 0, |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 5316 | }; |
KAMEZAWA Hiroyuki | c077719 | 2009-01-07 18:07:57 -0800 | [diff] [blame] | 5317 | |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5318 | /** |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5319 | * mem_cgroup_low - check if memory consumption is below the normal range |
Sean Christopherson | 34c8105 | 2017-07-10 15:48:05 -0700 | [diff] [blame^] | 5320 | * @root: the top ancestor of the sub-tree being checked |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5321 | * @memcg: the memory cgroup to check |
| 5322 | * |
| 5323 | * Returns %true if memory consumption of @memcg, and that of all |
Sean Christopherson | 34c8105 | 2017-07-10 15:48:05 -0700 | [diff] [blame^] | 5324 | * ancestors up to (but not including) @root, is below the normal range. |
| 5325 | * |
| 5326 | * @root is exclusive; it is never low when looked at directly and isn't |
| 5327 | * checked when traversing the hierarchy. |
| 5328 | * |
| 5329 | * Excluding @root enables using memory.low to prioritize memory usage |
| 5330 | * between cgroups within a subtree of the hierarchy that is limited by |
| 5331 | * memory.high or memory.max. |
| 5332 | * |
| 5333 | * For example, given cgroup A with children B and C: |
| 5334 | * |
| 5335 | * A |
| 5336 | * / \ |
| 5337 | * B C |
| 5338 | * |
| 5339 | * and |
| 5340 | * |
| 5341 | * 1. A/memory.current > A/memory.high |
| 5342 | * 2. A/B/memory.current < A/B/memory.low |
| 5343 | * 3. A/C/memory.current >= A/C/memory.low |
| 5344 | * |
| 5345 | * As 'A' is high, i.e. triggers reclaim from 'A', and 'B' is low, we |
| 5346 | * should reclaim from 'C' until 'A' is no longer high or until we can |
| 5347 | * no longer reclaim from 'C'. If 'A', i.e. @root, isn't excluded by |
| 5348 | * mem_cgroup_low when reclaming from 'A', then 'B' won't be considered |
| 5349 | * low and we will reclaim indiscriminately from both 'B' and 'C'. |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5350 | */ |
| 5351 | bool mem_cgroup_low(struct mem_cgroup *root, struct mem_cgroup *memcg) |
| 5352 | { |
| 5353 | if (mem_cgroup_disabled()) |
| 5354 | return false; |
| 5355 | |
Sean Christopherson | 34c8105 | 2017-07-10 15:48:05 -0700 | [diff] [blame^] | 5356 | if (!root) |
| 5357 | root = root_mem_cgroup; |
| 5358 | if (memcg == root) |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5359 | return false; |
| 5360 | |
Sean Christopherson | 34c8105 | 2017-07-10 15:48:05 -0700 | [diff] [blame^] | 5361 | for (; memcg != root; memcg = parent_mem_cgroup(memcg)) { |
Michal Hocko | 4e54ded | 2015-02-27 15:51:46 -0800 | [diff] [blame] | 5362 | if (page_counter_read(&memcg->memory) >= memcg->low) |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5363 | return false; |
| 5364 | } |
Sean Christopherson | 34c8105 | 2017-07-10 15:48:05 -0700 | [diff] [blame^] | 5365 | |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 5366 | return true; |
| 5367 | } |
| 5368 | |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 5369 | /** |
| 5370 | * mem_cgroup_try_charge - try charging a page |
| 5371 | * @page: page to charge |
| 5372 | * @mm: mm context of the victim |
| 5373 | * @gfp_mask: reclaim mode |
| 5374 | * @memcgp: charged memcg return |
Li RongQing | 25843c2 | 2016-07-26 15:26:56 -0700 | [diff] [blame] | 5375 | * @compound: charge the page as compound or small page |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 5376 | * |
| 5377 | * Try to charge @page to the memcg that @mm belongs to, reclaiming |
| 5378 | * pages according to @gfp_mask if necessary. |
| 5379 | * |
| 5380 | * Returns 0 on success, with *@memcgp pointing to the charged memcg. |
| 5381 | * Otherwise, an error code is returned. |
| 5382 | * |
| 5383 | * After page->mapping has been set up, the caller must finalize the |
| 5384 | * charge with mem_cgroup_commit_charge(). Or abort the transaction |
| 5385 | * with mem_cgroup_cancel_charge() in case page instantiation fails. |
| 5386 | */ |
| 5387 | int mem_cgroup_try_charge(struct page *page, struct mm_struct *mm, |
Kirill A. Shutemov | f627c2f | 2016-01-15 16:52:20 -0800 | [diff] [blame] | 5388 | gfp_t gfp_mask, struct mem_cgroup **memcgp, |
| 5389 | bool compound) |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 5390 | { |
| 5391 | struct mem_cgroup *memcg = NULL; |
Kirill A. Shutemov | f627c2f | 2016-01-15 16:52:20 -0800 | [diff] [blame] | 5392 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 5393 | int ret = 0; |
| 5394 | |
| 5395 | if (mem_cgroup_disabled()) |
| 5396 | goto out; |
| 5397 | |
| 5398 | if (PageSwapCache(page)) { |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 5399 | /* |
| 5400 | * Every swap fault against a single page tries to charge the |
| 5401 | * page, bail as early as possible. shmem_unuse() encounters |
| 5402 | * already charged pages, too. The USED bit is protected by |
| 5403 | * the page lock, which serializes swap cache removal, which |
| 5404 | * in turn serializes uncharging. |
| 5405 | */ |
Vladimir Davydov | e993d90 | 2015-09-09 15:35:35 -0700 | [diff] [blame] | 5406 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 5407 | if (page->mem_cgroup) |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 5408 | goto out; |
Vladimir Davydov | e993d90 | 2015-09-09 15:35:35 -0700 | [diff] [blame] | 5409 | |
Vladimir Davydov | 37e8435 | 2016-01-20 15:02:56 -0800 | [diff] [blame] | 5410 | if (do_swap_account) { |
Vladimir Davydov | e993d90 | 2015-09-09 15:35:35 -0700 | [diff] [blame] | 5411 | swp_entry_t ent = { .val = page_private(page), }; |
| 5412 | unsigned short id = lookup_swap_cgroup_id(ent); |
| 5413 | |
| 5414 | rcu_read_lock(); |
| 5415 | memcg = mem_cgroup_from_id(id); |
| 5416 | if (memcg && !css_tryget_online(&memcg->css)) |
| 5417 | memcg = NULL; |
| 5418 | rcu_read_unlock(); |
| 5419 | } |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 5420 | } |
| 5421 | |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 5422 | if (!memcg) |
| 5423 | memcg = get_mem_cgroup_from_mm(mm); |
| 5424 | |
| 5425 | ret = try_charge(memcg, gfp_mask, nr_pages); |
| 5426 | |
| 5427 | css_put(&memcg->css); |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 5428 | out: |
| 5429 | *memcgp = memcg; |
| 5430 | return ret; |
| 5431 | } |
| 5432 | |
| 5433 | /** |
| 5434 | * mem_cgroup_commit_charge - commit a page charge |
| 5435 | * @page: page to charge |
| 5436 | * @memcg: memcg to charge the page to |
| 5437 | * @lrucare: page might be on LRU already |
Li RongQing | 25843c2 | 2016-07-26 15:26:56 -0700 | [diff] [blame] | 5438 | * @compound: charge the page as compound or small page |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 5439 | * |
| 5440 | * Finalize a charge transaction started by mem_cgroup_try_charge(), |
| 5441 | * after page->mapping has been set up. This must happen atomically |
| 5442 | * as part of the page instantiation, i.e. under the page table lock |
| 5443 | * for anonymous pages, under the page lock for page and swap cache. |
| 5444 | * |
| 5445 | * In addition, the page must not be on the LRU during the commit, to |
| 5446 | * prevent racing with task migration. If it might be, use @lrucare. |
| 5447 | * |
| 5448 | * Use mem_cgroup_cancel_charge() to cancel the transaction instead. |
| 5449 | */ |
| 5450 | void mem_cgroup_commit_charge(struct page *page, struct mem_cgroup *memcg, |
Kirill A. Shutemov | f627c2f | 2016-01-15 16:52:20 -0800 | [diff] [blame] | 5451 | bool lrucare, bool compound) |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 5452 | { |
Kirill A. Shutemov | f627c2f | 2016-01-15 16:52:20 -0800 | [diff] [blame] | 5453 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 5454 | |
| 5455 | VM_BUG_ON_PAGE(!page->mapping, page); |
| 5456 | VM_BUG_ON_PAGE(PageLRU(page) && !lrucare, page); |
| 5457 | |
| 5458 | if (mem_cgroup_disabled()) |
| 5459 | return; |
| 5460 | /* |
| 5461 | * Swap faults will attempt to charge the same page multiple |
| 5462 | * times. But reuse_swap_page() might have removed the page |
| 5463 | * from swapcache already, so we can't check PageSwapCache(). |
| 5464 | */ |
| 5465 | if (!memcg) |
| 5466 | return; |
| 5467 | |
Johannes Weiner | 6abb5a8 | 2014-08-08 14:19:33 -0700 | [diff] [blame] | 5468 | commit_charge(page, memcg, lrucare); |
| 5469 | |
Johannes Weiner | 6abb5a8 | 2014-08-08 14:19:33 -0700 | [diff] [blame] | 5470 | local_irq_disable(); |
Kirill A. Shutemov | f627c2f | 2016-01-15 16:52:20 -0800 | [diff] [blame] | 5471 | mem_cgroup_charge_statistics(memcg, page, compound, nr_pages); |
Johannes Weiner | 6abb5a8 | 2014-08-08 14:19:33 -0700 | [diff] [blame] | 5472 | memcg_check_events(memcg, page); |
| 5473 | local_irq_enable(); |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 5474 | |
Johannes Weiner | 7941d21 | 2016-01-14 15:21:23 -0800 | [diff] [blame] | 5475 | if (do_memsw_account() && PageSwapCache(page)) { |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 5476 | swp_entry_t entry = { .val = page_private(page) }; |
| 5477 | /* |
| 5478 | * The swap entry might not get freed for a long time, |
| 5479 | * let's not wait for it. The page already received a |
| 5480 | * memory+swap charge, drop the swap entry duplicate. |
| 5481 | */ |
Huang Ying | 38d8b4e | 2017-07-06 15:37:18 -0700 | [diff] [blame] | 5482 | mem_cgroup_uncharge_swap(entry, nr_pages); |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 5483 | } |
| 5484 | } |
| 5485 | |
| 5486 | /** |
| 5487 | * mem_cgroup_cancel_charge - cancel a page charge |
| 5488 | * @page: page to charge |
| 5489 | * @memcg: memcg to charge the page to |
Li RongQing | 25843c2 | 2016-07-26 15:26:56 -0700 | [diff] [blame] | 5490 | * @compound: charge the page as compound or small page |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 5491 | * |
| 5492 | * Cancel a charge transaction started by mem_cgroup_try_charge(). |
| 5493 | */ |
Kirill A. Shutemov | f627c2f | 2016-01-15 16:52:20 -0800 | [diff] [blame] | 5494 | void mem_cgroup_cancel_charge(struct page *page, struct mem_cgroup *memcg, |
| 5495 | bool compound) |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 5496 | { |
Kirill A. Shutemov | f627c2f | 2016-01-15 16:52:20 -0800 | [diff] [blame] | 5497 | unsigned int nr_pages = compound ? hpage_nr_pages(page) : 1; |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 5498 | |
| 5499 | if (mem_cgroup_disabled()) |
| 5500 | return; |
| 5501 | /* |
| 5502 | * Swap faults will attempt to charge the same page multiple |
| 5503 | * times. But reuse_swap_page() might have removed the page |
| 5504 | * from swapcache already, so we can't check PageSwapCache(). |
| 5505 | */ |
| 5506 | if (!memcg) |
| 5507 | return; |
| 5508 | |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 5509 | cancel_charge(memcg, nr_pages); |
| 5510 | } |
| 5511 | |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5512 | static void uncharge_batch(struct mem_cgroup *memcg, unsigned long pgpgout, |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5513 | unsigned long nr_anon, unsigned long nr_file, |
Johannes Weiner | 9a4caf1 | 2017-05-03 14:52:45 -0700 | [diff] [blame] | 5514 | unsigned long nr_kmem, unsigned long nr_huge, |
| 5515 | unsigned long nr_shmem, struct page *dummy_page) |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5516 | { |
Vladimir Davydov | 5e8d35f | 2016-07-26 15:24:27 -0700 | [diff] [blame] | 5517 | unsigned long nr_pages = nr_anon + nr_file + nr_kmem; |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5518 | unsigned long flags; |
| 5519 | |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 5520 | if (!mem_cgroup_is_root(memcg)) { |
Johannes Weiner | 18eca2e | 2014-12-10 15:43:57 -0800 | [diff] [blame] | 5521 | page_counter_uncharge(&memcg->memory, nr_pages); |
Johannes Weiner | 7941d21 | 2016-01-14 15:21:23 -0800 | [diff] [blame] | 5522 | if (do_memsw_account()) |
Johannes Weiner | 18eca2e | 2014-12-10 15:43:57 -0800 | [diff] [blame] | 5523 | page_counter_uncharge(&memcg->memsw, nr_pages); |
Vladimir Davydov | 5e8d35f | 2016-07-26 15:24:27 -0700 | [diff] [blame] | 5524 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && nr_kmem) |
| 5525 | page_counter_uncharge(&memcg->kmem, nr_kmem); |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 5526 | memcg_oom_recover(memcg); |
| 5527 | } |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5528 | |
| 5529 | local_irq_save(flags); |
Johannes Weiner | 71cd311 | 2017-05-03 14:55:13 -0700 | [diff] [blame] | 5530 | __this_cpu_sub(memcg->stat->count[MEMCG_RSS], nr_anon); |
| 5531 | __this_cpu_sub(memcg->stat->count[MEMCG_CACHE], nr_file); |
| 5532 | __this_cpu_sub(memcg->stat->count[MEMCG_RSS_HUGE], nr_huge); |
| 5533 | __this_cpu_sub(memcg->stat->count[NR_SHMEM], nr_shmem); |
Johannes Weiner | df0e53d | 2017-05-03 14:55:10 -0700 | [diff] [blame] | 5534 | __this_cpu_add(memcg->stat->events[PGPGOUT], pgpgout); |
Johannes Weiner | 18eca2e | 2014-12-10 15:43:57 -0800 | [diff] [blame] | 5535 | __this_cpu_add(memcg->stat->nr_page_events, nr_pages); |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5536 | memcg_check_events(memcg, dummy_page); |
| 5537 | local_irq_restore(flags); |
Johannes Weiner | e8ea14c | 2014-12-10 15:42:42 -0800 | [diff] [blame] | 5538 | |
| 5539 | if (!mem_cgroup_is_root(memcg)) |
Johannes Weiner | 18eca2e | 2014-12-10 15:43:57 -0800 | [diff] [blame] | 5540 | css_put_many(&memcg->css, nr_pages); |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5541 | } |
| 5542 | |
| 5543 | static void uncharge_list(struct list_head *page_list) |
| 5544 | { |
| 5545 | struct mem_cgroup *memcg = NULL; |
Johannes Weiner | 9a4caf1 | 2017-05-03 14:52:45 -0700 | [diff] [blame] | 5546 | unsigned long nr_shmem = 0; |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5547 | unsigned long nr_anon = 0; |
| 5548 | unsigned long nr_file = 0; |
| 5549 | unsigned long nr_huge = 0; |
Vladimir Davydov | 5e8d35f | 2016-07-26 15:24:27 -0700 | [diff] [blame] | 5550 | unsigned long nr_kmem = 0; |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5551 | unsigned long pgpgout = 0; |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5552 | struct list_head *next; |
| 5553 | struct page *page; |
| 5554 | |
Johannes Weiner | 8b59265 | 2016-03-17 14:20:31 -0700 | [diff] [blame] | 5555 | /* |
| 5556 | * Note that the list can be a single page->lru; hence the |
| 5557 | * do-while loop instead of a simple list_for_each_entry(). |
| 5558 | */ |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5559 | next = page_list->next; |
| 5560 | do { |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5561 | page = list_entry(next, struct page, lru); |
| 5562 | next = page->lru.next; |
| 5563 | |
| 5564 | VM_BUG_ON_PAGE(PageLRU(page), page); |
Michal Hocko | 1836522 | 2017-05-12 15:46:26 -0700 | [diff] [blame] | 5565 | VM_BUG_ON_PAGE(!PageHWPoison(page) && page_count(page), page); |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5566 | |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 5567 | if (!page->mem_cgroup) |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5568 | continue; |
| 5569 | |
| 5570 | /* |
| 5571 | * Nobody should be changing or seriously looking at |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 5572 | * page->mem_cgroup at this point, we have fully |
Johannes Weiner | 2983331 | 2014-12-10 15:44:02 -0800 | [diff] [blame] | 5573 | * exclusive access to the page. |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5574 | */ |
| 5575 | |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 5576 | if (memcg != page->mem_cgroup) { |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5577 | if (memcg) { |
Johannes Weiner | 18eca2e | 2014-12-10 15:43:57 -0800 | [diff] [blame] | 5578 | uncharge_batch(memcg, pgpgout, nr_anon, nr_file, |
Johannes Weiner | 9a4caf1 | 2017-05-03 14:52:45 -0700 | [diff] [blame] | 5579 | nr_kmem, nr_huge, nr_shmem, page); |
| 5580 | pgpgout = nr_anon = nr_file = nr_kmem = 0; |
| 5581 | nr_huge = nr_shmem = 0; |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5582 | } |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 5583 | memcg = page->mem_cgroup; |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5584 | } |
| 5585 | |
Vladimir Davydov | 5e8d35f | 2016-07-26 15:24:27 -0700 | [diff] [blame] | 5586 | if (!PageKmemcg(page)) { |
| 5587 | unsigned int nr_pages = 1; |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5588 | |
Vladimir Davydov | 5e8d35f | 2016-07-26 15:24:27 -0700 | [diff] [blame] | 5589 | if (PageTransHuge(page)) { |
| 5590 | nr_pages <<= compound_order(page); |
Vladimir Davydov | 5e8d35f | 2016-07-26 15:24:27 -0700 | [diff] [blame] | 5591 | nr_huge += nr_pages; |
| 5592 | } |
| 5593 | if (PageAnon(page)) |
| 5594 | nr_anon += nr_pages; |
Johannes Weiner | 9a4caf1 | 2017-05-03 14:52:45 -0700 | [diff] [blame] | 5595 | else { |
Vladimir Davydov | 5e8d35f | 2016-07-26 15:24:27 -0700 | [diff] [blame] | 5596 | nr_file += nr_pages; |
Johannes Weiner | 9a4caf1 | 2017-05-03 14:52:45 -0700 | [diff] [blame] | 5597 | if (PageSwapBacked(page)) |
| 5598 | nr_shmem += nr_pages; |
| 5599 | } |
Vladimir Davydov | 5e8d35f | 2016-07-26 15:24:27 -0700 | [diff] [blame] | 5600 | pgpgout++; |
Vladimir Davydov | c4159a7 | 2016-08-08 23:03:12 +0300 | [diff] [blame] | 5601 | } else { |
Vladimir Davydov | 5e8d35f | 2016-07-26 15:24:27 -0700 | [diff] [blame] | 5602 | nr_kmem += 1 << compound_order(page); |
Vladimir Davydov | c4159a7 | 2016-08-08 23:03:12 +0300 | [diff] [blame] | 5603 | __ClearPageKmemcg(page); |
| 5604 | } |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5605 | |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 5606 | page->mem_cgroup = NULL; |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5607 | } while (next != page_list); |
| 5608 | |
| 5609 | if (memcg) |
Johannes Weiner | 18eca2e | 2014-12-10 15:43:57 -0800 | [diff] [blame] | 5610 | uncharge_batch(memcg, pgpgout, nr_anon, nr_file, |
Johannes Weiner | 9a4caf1 | 2017-05-03 14:52:45 -0700 | [diff] [blame] | 5611 | nr_kmem, nr_huge, nr_shmem, page); |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5612 | } |
| 5613 | |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 5614 | /** |
| 5615 | * mem_cgroup_uncharge - uncharge a page |
| 5616 | * @page: page to uncharge |
| 5617 | * |
| 5618 | * Uncharge a page previously charged with mem_cgroup_try_charge() and |
| 5619 | * mem_cgroup_commit_charge(). |
| 5620 | */ |
| 5621 | void mem_cgroup_uncharge(struct page *page) |
| 5622 | { |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 5623 | if (mem_cgroup_disabled()) |
| 5624 | return; |
| 5625 | |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5626 | /* Don't touch page->lru of any random page, pre-check: */ |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 5627 | if (!page->mem_cgroup) |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 5628 | return; |
| 5629 | |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5630 | INIT_LIST_HEAD(&page->lru); |
| 5631 | uncharge_list(&page->lru); |
| 5632 | } |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 5633 | |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5634 | /** |
| 5635 | * mem_cgroup_uncharge_list - uncharge a list of page |
| 5636 | * @page_list: list of pages to uncharge |
| 5637 | * |
| 5638 | * Uncharge a list of pages previously charged with |
| 5639 | * mem_cgroup_try_charge() and mem_cgroup_commit_charge(). |
| 5640 | */ |
| 5641 | void mem_cgroup_uncharge_list(struct list_head *page_list) |
| 5642 | { |
| 5643 | if (mem_cgroup_disabled()) |
| 5644 | return; |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 5645 | |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 5646 | if (!list_empty(page_list)) |
| 5647 | uncharge_list(page_list); |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 5648 | } |
| 5649 | |
| 5650 | /** |
Johannes Weiner | 6a93ca8 | 2016-03-15 14:57:19 -0700 | [diff] [blame] | 5651 | * mem_cgroup_migrate - charge a page's replacement |
| 5652 | * @oldpage: currently circulating page |
| 5653 | * @newpage: replacement page |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 5654 | * |
Johannes Weiner | 6a93ca8 | 2016-03-15 14:57:19 -0700 | [diff] [blame] | 5655 | * Charge @newpage as a replacement page for @oldpage. @oldpage will |
| 5656 | * be uncharged upon free. |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 5657 | * |
| 5658 | * Both pages must be locked, @newpage->mapping must be set up. |
| 5659 | */ |
Johannes Weiner | 6a93ca8 | 2016-03-15 14:57:19 -0700 | [diff] [blame] | 5660 | void mem_cgroup_migrate(struct page *oldpage, struct page *newpage) |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 5661 | { |
Johannes Weiner | 2983331 | 2014-12-10 15:44:02 -0800 | [diff] [blame] | 5662 | struct mem_cgroup *memcg; |
Johannes Weiner | 44b7a8d | 2016-01-20 15:03:16 -0800 | [diff] [blame] | 5663 | unsigned int nr_pages; |
| 5664 | bool compound; |
Tejun Heo | d93c413 | 2016-06-24 14:49:54 -0700 | [diff] [blame] | 5665 | unsigned long flags; |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 5666 | |
| 5667 | VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage); |
| 5668 | VM_BUG_ON_PAGE(!PageLocked(newpage), newpage); |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 5669 | VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage); |
Johannes Weiner | 6abb5a8 | 2014-08-08 14:19:33 -0700 | [diff] [blame] | 5670 | VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage), |
| 5671 | newpage); |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 5672 | |
| 5673 | if (mem_cgroup_disabled()) |
| 5674 | return; |
| 5675 | |
| 5676 | /* Page cache replacement: new page already charged? */ |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 5677 | if (newpage->mem_cgroup) |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 5678 | return; |
| 5679 | |
Hugh Dickins | 45637ba | 2015-11-05 18:49:40 -0800 | [diff] [blame] | 5680 | /* Swapcache readahead pages can get replaced before being charged */ |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 5681 | memcg = oldpage->mem_cgroup; |
Johannes Weiner | 2983331 | 2014-12-10 15:44:02 -0800 | [diff] [blame] | 5682 | if (!memcg) |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 5683 | return; |
| 5684 | |
Johannes Weiner | 44b7a8d | 2016-01-20 15:03:16 -0800 | [diff] [blame] | 5685 | /* Force-charge the new page. The old one will be freed soon */ |
| 5686 | compound = PageTransHuge(newpage); |
| 5687 | nr_pages = compound ? hpage_nr_pages(newpage) : 1; |
| 5688 | |
| 5689 | page_counter_charge(&memcg->memory, nr_pages); |
| 5690 | if (do_memsw_account()) |
| 5691 | page_counter_charge(&memcg->memsw, nr_pages); |
| 5692 | css_get_many(&memcg->css, nr_pages); |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 5693 | |
Johannes Weiner | 9cf7666 | 2016-03-15 14:57:58 -0700 | [diff] [blame] | 5694 | commit_charge(newpage, memcg, false); |
Johannes Weiner | 44b7a8d | 2016-01-20 15:03:16 -0800 | [diff] [blame] | 5695 | |
Tejun Heo | d93c413 | 2016-06-24 14:49:54 -0700 | [diff] [blame] | 5696 | local_irq_save(flags); |
Johannes Weiner | 44b7a8d | 2016-01-20 15:03:16 -0800 | [diff] [blame] | 5697 | mem_cgroup_charge_statistics(memcg, newpage, compound, nr_pages); |
| 5698 | memcg_check_events(memcg, newpage); |
Tejun Heo | d93c413 | 2016-06-24 14:49:54 -0700 | [diff] [blame] | 5699 | local_irq_restore(flags); |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 5700 | } |
| 5701 | |
Johannes Weiner | ef12947 | 2016-01-14 15:21:34 -0800 | [diff] [blame] | 5702 | DEFINE_STATIC_KEY_FALSE(memcg_sockets_enabled_key); |
Johannes Weiner | 1109208 | 2016-01-14 15:21:26 -0800 | [diff] [blame] | 5703 | EXPORT_SYMBOL(memcg_sockets_enabled_key); |
| 5704 | |
Johannes Weiner | 2d75807 | 2016-10-07 17:00:58 -0700 | [diff] [blame] | 5705 | void mem_cgroup_sk_alloc(struct sock *sk) |
Johannes Weiner | 1109208 | 2016-01-14 15:21:26 -0800 | [diff] [blame] | 5706 | { |
| 5707 | struct mem_cgroup *memcg; |
| 5708 | |
Johannes Weiner | 2d75807 | 2016-10-07 17:00:58 -0700 | [diff] [blame] | 5709 | if (!mem_cgroup_sockets_enabled) |
| 5710 | return; |
| 5711 | |
| 5712 | /* |
| 5713 | * Socket cloning can throw us here with sk_memcg already |
Johannes Weiner | 1109208 | 2016-01-14 15:21:26 -0800 | [diff] [blame] | 5714 | * filled. It won't however, necessarily happen from |
| 5715 | * process context. So the test for root memcg given |
| 5716 | * the current task's memcg won't help us in this case. |
| 5717 | * |
| 5718 | * Respecting the original socket's memcg is a better |
| 5719 | * decision in this case. |
| 5720 | */ |
| 5721 | if (sk->sk_memcg) { |
| 5722 | BUG_ON(mem_cgroup_is_root(sk->sk_memcg)); |
| 5723 | css_get(&sk->sk_memcg->css); |
| 5724 | return; |
| 5725 | } |
| 5726 | |
| 5727 | rcu_read_lock(); |
| 5728 | memcg = mem_cgroup_from_task(current); |
Johannes Weiner | f7e1cb6 | 2016-01-14 15:21:29 -0800 | [diff] [blame] | 5729 | if (memcg == root_mem_cgroup) |
| 5730 | goto out; |
Johannes Weiner | 0db1529 | 2016-01-20 15:02:50 -0800 | [diff] [blame] | 5731 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && !memcg->tcpmem_active) |
Johannes Weiner | f7e1cb6 | 2016-01-14 15:21:29 -0800 | [diff] [blame] | 5732 | goto out; |
Johannes Weiner | f7e1cb6 | 2016-01-14 15:21:29 -0800 | [diff] [blame] | 5733 | if (css_tryget_online(&memcg->css)) |
Johannes Weiner | 1109208 | 2016-01-14 15:21:26 -0800 | [diff] [blame] | 5734 | sk->sk_memcg = memcg; |
Johannes Weiner | f7e1cb6 | 2016-01-14 15:21:29 -0800 | [diff] [blame] | 5735 | out: |
Johannes Weiner | 1109208 | 2016-01-14 15:21:26 -0800 | [diff] [blame] | 5736 | rcu_read_unlock(); |
| 5737 | } |
Johannes Weiner | 1109208 | 2016-01-14 15:21:26 -0800 | [diff] [blame] | 5738 | |
Johannes Weiner | 2d75807 | 2016-10-07 17:00:58 -0700 | [diff] [blame] | 5739 | void mem_cgroup_sk_free(struct sock *sk) |
Johannes Weiner | 1109208 | 2016-01-14 15:21:26 -0800 | [diff] [blame] | 5740 | { |
Johannes Weiner | 2d75807 | 2016-10-07 17:00:58 -0700 | [diff] [blame] | 5741 | if (sk->sk_memcg) |
| 5742 | css_put(&sk->sk_memcg->css); |
Johannes Weiner | 1109208 | 2016-01-14 15:21:26 -0800 | [diff] [blame] | 5743 | } |
| 5744 | |
| 5745 | /** |
| 5746 | * mem_cgroup_charge_skmem - charge socket memory |
| 5747 | * @memcg: memcg to charge |
| 5748 | * @nr_pages: number of pages to charge |
| 5749 | * |
| 5750 | * Charges @nr_pages to @memcg. Returns %true if the charge fit within |
| 5751 | * @memcg's configured limit, %false if the charge had to be forced. |
| 5752 | */ |
| 5753 | bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages) |
| 5754 | { |
Johannes Weiner | f7e1cb6 | 2016-01-14 15:21:29 -0800 | [diff] [blame] | 5755 | gfp_t gfp_mask = GFP_KERNEL; |
Johannes Weiner | 1109208 | 2016-01-14 15:21:26 -0800 | [diff] [blame] | 5756 | |
Johannes Weiner | f7e1cb6 | 2016-01-14 15:21:29 -0800 | [diff] [blame] | 5757 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
Johannes Weiner | 0db1529 | 2016-01-20 15:02:50 -0800 | [diff] [blame] | 5758 | struct page_counter *fail; |
Johannes Weiner | f7e1cb6 | 2016-01-14 15:21:29 -0800 | [diff] [blame] | 5759 | |
Johannes Weiner | 0db1529 | 2016-01-20 15:02:50 -0800 | [diff] [blame] | 5760 | if (page_counter_try_charge(&memcg->tcpmem, nr_pages, &fail)) { |
| 5761 | memcg->tcpmem_pressure = 0; |
Johannes Weiner | f7e1cb6 | 2016-01-14 15:21:29 -0800 | [diff] [blame] | 5762 | return true; |
| 5763 | } |
Johannes Weiner | 0db1529 | 2016-01-20 15:02:50 -0800 | [diff] [blame] | 5764 | page_counter_charge(&memcg->tcpmem, nr_pages); |
| 5765 | memcg->tcpmem_pressure = 1; |
Johannes Weiner | f7e1cb6 | 2016-01-14 15:21:29 -0800 | [diff] [blame] | 5766 | return false; |
Johannes Weiner | 1109208 | 2016-01-14 15:21:26 -0800 | [diff] [blame] | 5767 | } |
Johannes Weiner | d886f4e | 2016-01-20 15:02:47 -0800 | [diff] [blame] | 5768 | |
Johannes Weiner | f7e1cb6 | 2016-01-14 15:21:29 -0800 | [diff] [blame] | 5769 | /* Don't block in the packet receive path */ |
| 5770 | if (in_softirq()) |
| 5771 | gfp_mask = GFP_NOWAIT; |
| 5772 | |
Johannes Weiner | b2807f0 | 2016-01-20 15:03:22 -0800 | [diff] [blame] | 5773 | this_cpu_add(memcg->stat->count[MEMCG_SOCK], nr_pages); |
| 5774 | |
Johannes Weiner | f7e1cb6 | 2016-01-14 15:21:29 -0800 | [diff] [blame] | 5775 | if (try_charge(memcg, gfp_mask, nr_pages) == 0) |
| 5776 | return true; |
| 5777 | |
| 5778 | try_charge(memcg, gfp_mask|__GFP_NOFAIL, nr_pages); |
Johannes Weiner | 1109208 | 2016-01-14 15:21:26 -0800 | [diff] [blame] | 5779 | return false; |
| 5780 | } |
| 5781 | |
| 5782 | /** |
| 5783 | * mem_cgroup_uncharge_skmem - uncharge socket memory |
| 5784 | * @memcg - memcg to uncharge |
| 5785 | * @nr_pages - number of pages to uncharge |
| 5786 | */ |
| 5787 | void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages) |
| 5788 | { |
Johannes Weiner | f7e1cb6 | 2016-01-14 15:21:29 -0800 | [diff] [blame] | 5789 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
Johannes Weiner | 0db1529 | 2016-01-20 15:02:50 -0800 | [diff] [blame] | 5790 | page_counter_uncharge(&memcg->tcpmem, nr_pages); |
Johannes Weiner | f7e1cb6 | 2016-01-14 15:21:29 -0800 | [diff] [blame] | 5791 | return; |
| 5792 | } |
Johannes Weiner | d886f4e | 2016-01-20 15:02:47 -0800 | [diff] [blame] | 5793 | |
Johannes Weiner | b2807f0 | 2016-01-20 15:03:22 -0800 | [diff] [blame] | 5794 | this_cpu_sub(memcg->stat->count[MEMCG_SOCK], nr_pages); |
| 5795 | |
Johannes Weiner | f7e1cb6 | 2016-01-14 15:21:29 -0800 | [diff] [blame] | 5796 | page_counter_uncharge(&memcg->memory, nr_pages); |
| 5797 | css_put_many(&memcg->css, nr_pages); |
Johannes Weiner | 1109208 | 2016-01-14 15:21:26 -0800 | [diff] [blame] | 5798 | } |
| 5799 | |
Johannes Weiner | f7e1cb6 | 2016-01-14 15:21:29 -0800 | [diff] [blame] | 5800 | static int __init cgroup_memory(char *s) |
| 5801 | { |
| 5802 | char *token; |
| 5803 | |
| 5804 | while ((token = strsep(&s, ",")) != NULL) { |
| 5805 | if (!*token) |
| 5806 | continue; |
| 5807 | if (!strcmp(token, "nosocket")) |
| 5808 | cgroup_memory_nosocket = true; |
Vladimir Davydov | 04823c8 | 2016-01-20 15:02:38 -0800 | [diff] [blame] | 5809 | if (!strcmp(token, "nokmem")) |
| 5810 | cgroup_memory_nokmem = true; |
Johannes Weiner | f7e1cb6 | 2016-01-14 15:21:29 -0800 | [diff] [blame] | 5811 | } |
| 5812 | return 0; |
| 5813 | } |
| 5814 | __setup("cgroup.memory=", cgroup_memory); |
Johannes Weiner | 1109208 | 2016-01-14 15:21:26 -0800 | [diff] [blame] | 5815 | |
Michal Hocko | 2d11085 | 2013-02-22 16:34:43 -0800 | [diff] [blame] | 5816 | /* |
Michal Hocko | 1081312 | 2013-02-22 16:35:41 -0800 | [diff] [blame] | 5817 | * subsys_initcall() for memory controller. |
| 5818 | * |
Sebastian Andrzej Siewior | 308167f | 2016-11-03 15:49:59 +0100 | [diff] [blame] | 5819 | * Some parts like memcg_hotplug_cpu_dead() have to be initialized from this |
| 5820 | * context because of lock dependencies (cgroup_lock -> cpu hotplug) but |
| 5821 | * basically everything that doesn't depend on a specific mem_cgroup structure |
| 5822 | * should be initialized from here. |
Michal Hocko | 2d11085 | 2013-02-22 16:34:43 -0800 | [diff] [blame] | 5823 | */ |
| 5824 | static int __init mem_cgroup_init(void) |
| 5825 | { |
Johannes Weiner | 95a045f | 2015-02-11 15:26:33 -0800 | [diff] [blame] | 5826 | int cpu, node; |
| 5827 | |
Vladimir Davydov | 13583c3 | 2016-12-12 16:41:29 -0800 | [diff] [blame] | 5828 | #ifndef CONFIG_SLOB |
| 5829 | /* |
| 5830 | * Kmem cache creation is mostly done with the slab_mutex held, |
Tejun Heo | 17cc4df | 2017-02-22 15:41:36 -0800 | [diff] [blame] | 5831 | * so use a workqueue with limited concurrency to avoid stalling |
| 5832 | * all worker threads in case lots of cgroups are created and |
| 5833 | * destroyed simultaneously. |
Vladimir Davydov | 13583c3 | 2016-12-12 16:41:29 -0800 | [diff] [blame] | 5834 | */ |
Tejun Heo | 17cc4df | 2017-02-22 15:41:36 -0800 | [diff] [blame] | 5835 | memcg_kmem_cache_wq = alloc_workqueue("memcg_kmem_cache", 0, 1); |
| 5836 | BUG_ON(!memcg_kmem_cache_wq); |
Vladimir Davydov | 13583c3 | 2016-12-12 16:41:29 -0800 | [diff] [blame] | 5837 | #endif |
| 5838 | |
Sebastian Andrzej Siewior | 308167f | 2016-11-03 15:49:59 +0100 | [diff] [blame] | 5839 | cpuhp_setup_state_nocalls(CPUHP_MM_MEMCQ_DEAD, "mm/memctrl:dead", NULL, |
| 5840 | memcg_hotplug_cpu_dead); |
Johannes Weiner | 95a045f | 2015-02-11 15:26:33 -0800 | [diff] [blame] | 5841 | |
| 5842 | for_each_possible_cpu(cpu) |
| 5843 | INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work, |
| 5844 | drain_local_stock); |
| 5845 | |
| 5846 | for_each_node(node) { |
| 5847 | struct mem_cgroup_tree_per_node *rtpn; |
Johannes Weiner | 95a045f | 2015-02-11 15:26:33 -0800 | [diff] [blame] | 5848 | |
| 5849 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, |
| 5850 | node_online(node) ? node : NUMA_NO_NODE); |
| 5851 | |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 5852 | rtpn->rb_root = RB_ROOT; |
| 5853 | spin_lock_init(&rtpn->lock); |
Johannes Weiner | 95a045f | 2015-02-11 15:26:33 -0800 | [diff] [blame] | 5854 | soft_limit_tree.rb_tree_per_node[node] = rtpn; |
| 5855 | } |
| 5856 | |
Michal Hocko | 2d11085 | 2013-02-22 16:34:43 -0800 | [diff] [blame] | 5857 | return 0; |
| 5858 | } |
| 5859 | subsys_initcall(mem_cgroup_init); |
Johannes Weiner | 21afa38 | 2015-02-11 15:26:36 -0800 | [diff] [blame] | 5860 | |
| 5861 | #ifdef CONFIG_MEMCG_SWAP |
Arnd Bergmann | 358c07f | 2016-08-25 15:17:08 -0700 | [diff] [blame] | 5862 | static struct mem_cgroup *mem_cgroup_id_get_online(struct mem_cgroup *memcg) |
| 5863 | { |
| 5864 | while (!atomic_inc_not_zero(&memcg->id.ref)) { |
| 5865 | /* |
| 5866 | * The root cgroup cannot be destroyed, so it's refcount must |
| 5867 | * always be >= 1. |
| 5868 | */ |
| 5869 | if (WARN_ON_ONCE(memcg == root_mem_cgroup)) { |
| 5870 | VM_BUG_ON(1); |
| 5871 | break; |
| 5872 | } |
| 5873 | memcg = parent_mem_cgroup(memcg); |
| 5874 | if (!memcg) |
| 5875 | memcg = root_mem_cgroup; |
| 5876 | } |
| 5877 | return memcg; |
| 5878 | } |
| 5879 | |
Johannes Weiner | 21afa38 | 2015-02-11 15:26:36 -0800 | [diff] [blame] | 5880 | /** |
| 5881 | * mem_cgroup_swapout - transfer a memsw charge to swap |
| 5882 | * @page: page whose memsw charge to transfer |
| 5883 | * @entry: swap entry to move the charge to |
| 5884 | * |
| 5885 | * Transfer the memsw charge of @page to @entry. |
| 5886 | */ |
| 5887 | void mem_cgroup_swapout(struct page *page, swp_entry_t entry) |
| 5888 | { |
Vladimir Davydov | 1f47b61 | 2016-08-11 15:33:00 -0700 | [diff] [blame] | 5889 | struct mem_cgroup *memcg, *swap_memcg; |
Johannes Weiner | 21afa38 | 2015-02-11 15:26:36 -0800 | [diff] [blame] | 5890 | unsigned short oldid; |
| 5891 | |
| 5892 | VM_BUG_ON_PAGE(PageLRU(page), page); |
| 5893 | VM_BUG_ON_PAGE(page_count(page), page); |
| 5894 | |
Johannes Weiner | 7941d21 | 2016-01-14 15:21:23 -0800 | [diff] [blame] | 5895 | if (!do_memsw_account()) |
Johannes Weiner | 21afa38 | 2015-02-11 15:26:36 -0800 | [diff] [blame] | 5896 | return; |
| 5897 | |
| 5898 | memcg = page->mem_cgroup; |
| 5899 | |
| 5900 | /* Readahead page, never charged */ |
| 5901 | if (!memcg) |
| 5902 | return; |
| 5903 | |
Vladimir Davydov | 1f47b61 | 2016-08-11 15:33:00 -0700 | [diff] [blame] | 5904 | /* |
| 5905 | * In case the memcg owning these pages has been offlined and doesn't |
| 5906 | * have an ID allocated to it anymore, charge the closest online |
| 5907 | * ancestor for the swap instead and transfer the memory+swap charge. |
| 5908 | */ |
| 5909 | swap_memcg = mem_cgroup_id_get_online(memcg); |
Huang Ying | 38d8b4e | 2017-07-06 15:37:18 -0700 | [diff] [blame] | 5910 | oldid = swap_cgroup_record(entry, mem_cgroup_id(swap_memcg), 1); |
Johannes Weiner | 21afa38 | 2015-02-11 15:26:36 -0800 | [diff] [blame] | 5911 | VM_BUG_ON_PAGE(oldid, page); |
Huang Ying | 38d8b4e | 2017-07-06 15:37:18 -0700 | [diff] [blame] | 5912 | mem_cgroup_swap_statistics(swap_memcg, 1); |
Johannes Weiner | 21afa38 | 2015-02-11 15:26:36 -0800 | [diff] [blame] | 5913 | |
| 5914 | page->mem_cgroup = NULL; |
| 5915 | |
| 5916 | if (!mem_cgroup_is_root(memcg)) |
| 5917 | page_counter_uncharge(&memcg->memory, 1); |
| 5918 | |
Vladimir Davydov | 1f47b61 | 2016-08-11 15:33:00 -0700 | [diff] [blame] | 5919 | if (memcg != swap_memcg) { |
| 5920 | if (!mem_cgroup_is_root(swap_memcg)) |
| 5921 | page_counter_charge(&swap_memcg->memsw, 1); |
| 5922 | page_counter_uncharge(&memcg->memsw, 1); |
| 5923 | } |
| 5924 | |
Sebastian Andrzej Siewior | ce9ce66 | 2015-09-04 15:47:50 -0700 | [diff] [blame] | 5925 | /* |
| 5926 | * Interrupts should be disabled here because the caller holds the |
| 5927 | * mapping->tree_lock lock which is taken with interrupts-off. It is |
| 5928 | * important here to have the interrupts disabled because it is the |
| 5929 | * only synchronisation we have for udpating the per-CPU variables. |
| 5930 | */ |
| 5931 | VM_BUG_ON(!irqs_disabled()); |
Kirill A. Shutemov | f627c2f | 2016-01-15 16:52:20 -0800 | [diff] [blame] | 5932 | mem_cgroup_charge_statistics(memcg, page, false, -1); |
Johannes Weiner | 21afa38 | 2015-02-11 15:26:36 -0800 | [diff] [blame] | 5933 | memcg_check_events(memcg, page); |
Johannes Weiner | 73f576c | 2016-07-20 15:44:57 -0700 | [diff] [blame] | 5934 | |
| 5935 | if (!mem_cgroup_is_root(memcg)) |
| 5936 | css_put(&memcg->css); |
Johannes Weiner | 21afa38 | 2015-02-11 15:26:36 -0800 | [diff] [blame] | 5937 | } |
| 5938 | |
Huang Ying | 38d8b4e | 2017-07-06 15:37:18 -0700 | [diff] [blame] | 5939 | /** |
| 5940 | * mem_cgroup_try_charge_swap - try charging swap space for a page |
Vladimir Davydov | 37e8435 | 2016-01-20 15:02:56 -0800 | [diff] [blame] | 5941 | * @page: page being added to swap |
| 5942 | * @entry: swap entry to charge |
| 5943 | * |
Huang Ying | 38d8b4e | 2017-07-06 15:37:18 -0700 | [diff] [blame] | 5944 | * Try to charge @page's memcg for the swap space at @entry. |
Vladimir Davydov | 37e8435 | 2016-01-20 15:02:56 -0800 | [diff] [blame] | 5945 | * |
| 5946 | * Returns 0 on success, -ENOMEM on failure. |
| 5947 | */ |
| 5948 | int mem_cgroup_try_charge_swap(struct page *page, swp_entry_t entry) |
| 5949 | { |
Huang Ying | 38d8b4e | 2017-07-06 15:37:18 -0700 | [diff] [blame] | 5950 | unsigned int nr_pages = hpage_nr_pages(page); |
Vladimir Davydov | 37e8435 | 2016-01-20 15:02:56 -0800 | [diff] [blame] | 5951 | struct page_counter *counter; |
Huang Ying | 38d8b4e | 2017-07-06 15:37:18 -0700 | [diff] [blame] | 5952 | struct mem_cgroup *memcg; |
Vladimir Davydov | 37e8435 | 2016-01-20 15:02:56 -0800 | [diff] [blame] | 5953 | unsigned short oldid; |
| 5954 | |
| 5955 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) || !do_swap_account) |
| 5956 | return 0; |
| 5957 | |
| 5958 | memcg = page->mem_cgroup; |
| 5959 | |
| 5960 | /* Readahead page, never charged */ |
| 5961 | if (!memcg) |
| 5962 | return 0; |
| 5963 | |
Vladimir Davydov | 1f47b61 | 2016-08-11 15:33:00 -0700 | [diff] [blame] | 5964 | memcg = mem_cgroup_id_get_online(memcg); |
Vladimir Davydov | 37e8435 | 2016-01-20 15:02:56 -0800 | [diff] [blame] | 5965 | |
Vladimir Davydov | 1f47b61 | 2016-08-11 15:33:00 -0700 | [diff] [blame] | 5966 | if (!mem_cgroup_is_root(memcg) && |
Huang Ying | 38d8b4e | 2017-07-06 15:37:18 -0700 | [diff] [blame] | 5967 | !page_counter_try_charge(&memcg->swap, nr_pages, &counter)) { |
Vladimir Davydov | 1f47b61 | 2016-08-11 15:33:00 -0700 | [diff] [blame] | 5968 | mem_cgroup_id_put(memcg); |
| 5969 | return -ENOMEM; |
| 5970 | } |
| 5971 | |
Huang Ying | 38d8b4e | 2017-07-06 15:37:18 -0700 | [diff] [blame] | 5972 | /* Get references for the tail pages, too */ |
| 5973 | if (nr_pages > 1) |
| 5974 | mem_cgroup_id_get_many(memcg, nr_pages - 1); |
| 5975 | oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg), nr_pages); |
Vladimir Davydov | 37e8435 | 2016-01-20 15:02:56 -0800 | [diff] [blame] | 5976 | VM_BUG_ON_PAGE(oldid, page); |
Huang Ying | 38d8b4e | 2017-07-06 15:37:18 -0700 | [diff] [blame] | 5977 | mem_cgroup_swap_statistics(memcg, nr_pages); |
Vladimir Davydov | 37e8435 | 2016-01-20 15:02:56 -0800 | [diff] [blame] | 5978 | |
Vladimir Davydov | 37e8435 | 2016-01-20 15:02:56 -0800 | [diff] [blame] | 5979 | return 0; |
| 5980 | } |
| 5981 | |
Johannes Weiner | 21afa38 | 2015-02-11 15:26:36 -0800 | [diff] [blame] | 5982 | /** |
Huang Ying | 38d8b4e | 2017-07-06 15:37:18 -0700 | [diff] [blame] | 5983 | * mem_cgroup_uncharge_swap - uncharge swap space |
Johannes Weiner | 21afa38 | 2015-02-11 15:26:36 -0800 | [diff] [blame] | 5984 | * @entry: swap entry to uncharge |
Huang Ying | 38d8b4e | 2017-07-06 15:37:18 -0700 | [diff] [blame] | 5985 | * @nr_pages: the amount of swap space to uncharge |
Johannes Weiner | 21afa38 | 2015-02-11 15:26:36 -0800 | [diff] [blame] | 5986 | */ |
Huang Ying | 38d8b4e | 2017-07-06 15:37:18 -0700 | [diff] [blame] | 5987 | void mem_cgroup_uncharge_swap(swp_entry_t entry, unsigned int nr_pages) |
Johannes Weiner | 21afa38 | 2015-02-11 15:26:36 -0800 | [diff] [blame] | 5988 | { |
| 5989 | struct mem_cgroup *memcg; |
| 5990 | unsigned short id; |
| 5991 | |
Vladimir Davydov | 37e8435 | 2016-01-20 15:02:56 -0800 | [diff] [blame] | 5992 | if (!do_swap_account) |
Johannes Weiner | 21afa38 | 2015-02-11 15:26:36 -0800 | [diff] [blame] | 5993 | return; |
| 5994 | |
Huang Ying | 38d8b4e | 2017-07-06 15:37:18 -0700 | [diff] [blame] | 5995 | id = swap_cgroup_record(entry, 0, nr_pages); |
Johannes Weiner | 21afa38 | 2015-02-11 15:26:36 -0800 | [diff] [blame] | 5996 | rcu_read_lock(); |
Vladimir Davydov | adbe427 | 2015-04-15 16:13:00 -0700 | [diff] [blame] | 5997 | memcg = mem_cgroup_from_id(id); |
Johannes Weiner | 21afa38 | 2015-02-11 15:26:36 -0800 | [diff] [blame] | 5998 | if (memcg) { |
Vladimir Davydov | 37e8435 | 2016-01-20 15:02:56 -0800 | [diff] [blame] | 5999 | if (!mem_cgroup_is_root(memcg)) { |
| 6000 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
Huang Ying | 38d8b4e | 2017-07-06 15:37:18 -0700 | [diff] [blame] | 6001 | page_counter_uncharge(&memcg->swap, nr_pages); |
Vladimir Davydov | 37e8435 | 2016-01-20 15:02:56 -0800 | [diff] [blame] | 6002 | else |
Huang Ying | 38d8b4e | 2017-07-06 15:37:18 -0700 | [diff] [blame] | 6003 | page_counter_uncharge(&memcg->memsw, nr_pages); |
Vladimir Davydov | 37e8435 | 2016-01-20 15:02:56 -0800 | [diff] [blame] | 6004 | } |
Huang Ying | 38d8b4e | 2017-07-06 15:37:18 -0700 | [diff] [blame] | 6005 | mem_cgroup_swap_statistics(memcg, -nr_pages); |
| 6006 | mem_cgroup_id_put_many(memcg, nr_pages); |
Johannes Weiner | 21afa38 | 2015-02-11 15:26:36 -0800 | [diff] [blame] | 6007 | } |
| 6008 | rcu_read_unlock(); |
| 6009 | } |
| 6010 | |
Vladimir Davydov | d8b3843 | 2016-01-20 15:03:07 -0800 | [diff] [blame] | 6011 | long mem_cgroup_get_nr_swap_pages(struct mem_cgroup *memcg) |
| 6012 | { |
| 6013 | long nr_swap_pages = get_nr_swap_pages(); |
| 6014 | |
| 6015 | if (!do_swap_account || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
| 6016 | return nr_swap_pages; |
| 6017 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) |
| 6018 | nr_swap_pages = min_t(long, nr_swap_pages, |
| 6019 | READ_ONCE(memcg->swap.limit) - |
| 6020 | page_counter_read(&memcg->swap)); |
| 6021 | return nr_swap_pages; |
| 6022 | } |
| 6023 | |
Vladimir Davydov | 5ccc5ab | 2016-01-20 15:03:10 -0800 | [diff] [blame] | 6024 | bool mem_cgroup_swap_full(struct page *page) |
| 6025 | { |
| 6026 | struct mem_cgroup *memcg; |
| 6027 | |
| 6028 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
| 6029 | |
| 6030 | if (vm_swap_full()) |
| 6031 | return true; |
| 6032 | if (!do_swap_account || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
| 6033 | return false; |
| 6034 | |
| 6035 | memcg = page->mem_cgroup; |
| 6036 | if (!memcg) |
| 6037 | return false; |
| 6038 | |
| 6039 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) |
| 6040 | if (page_counter_read(&memcg->swap) * 2 >= memcg->swap.limit) |
| 6041 | return true; |
| 6042 | |
| 6043 | return false; |
| 6044 | } |
| 6045 | |
Johannes Weiner | 21afa38 | 2015-02-11 15:26:36 -0800 | [diff] [blame] | 6046 | /* for remember boot option*/ |
| 6047 | #ifdef CONFIG_MEMCG_SWAP_ENABLED |
| 6048 | static int really_do_swap_account __initdata = 1; |
| 6049 | #else |
| 6050 | static int really_do_swap_account __initdata; |
| 6051 | #endif |
| 6052 | |
| 6053 | static int __init enable_swap_account(char *s) |
| 6054 | { |
| 6055 | if (!strcmp(s, "1")) |
| 6056 | really_do_swap_account = 1; |
| 6057 | else if (!strcmp(s, "0")) |
| 6058 | really_do_swap_account = 0; |
| 6059 | return 1; |
| 6060 | } |
| 6061 | __setup("swapaccount=", enable_swap_account); |
| 6062 | |
Vladimir Davydov | 37e8435 | 2016-01-20 15:02:56 -0800 | [diff] [blame] | 6063 | static u64 swap_current_read(struct cgroup_subsys_state *css, |
| 6064 | struct cftype *cft) |
| 6065 | { |
| 6066 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
| 6067 | |
| 6068 | return (u64)page_counter_read(&memcg->swap) * PAGE_SIZE; |
| 6069 | } |
| 6070 | |
| 6071 | static int swap_max_show(struct seq_file *m, void *v) |
| 6072 | { |
| 6073 | struct mem_cgroup *memcg = mem_cgroup_from_css(seq_css(m)); |
| 6074 | unsigned long max = READ_ONCE(memcg->swap.limit); |
| 6075 | |
| 6076 | if (max == PAGE_COUNTER_MAX) |
| 6077 | seq_puts(m, "max\n"); |
| 6078 | else |
| 6079 | seq_printf(m, "%llu\n", (u64)max * PAGE_SIZE); |
| 6080 | |
| 6081 | return 0; |
| 6082 | } |
| 6083 | |
| 6084 | static ssize_t swap_max_write(struct kernfs_open_file *of, |
| 6085 | char *buf, size_t nbytes, loff_t off) |
| 6086 | { |
| 6087 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
| 6088 | unsigned long max; |
| 6089 | int err; |
| 6090 | |
| 6091 | buf = strstrip(buf); |
| 6092 | err = page_counter_memparse(buf, "max", &max); |
| 6093 | if (err) |
| 6094 | return err; |
| 6095 | |
| 6096 | mutex_lock(&memcg_limit_mutex); |
| 6097 | err = page_counter_limit(&memcg->swap, max); |
| 6098 | mutex_unlock(&memcg_limit_mutex); |
| 6099 | if (err) |
| 6100 | return err; |
| 6101 | |
| 6102 | return nbytes; |
| 6103 | } |
| 6104 | |
| 6105 | static struct cftype swap_files[] = { |
| 6106 | { |
| 6107 | .name = "swap.current", |
| 6108 | .flags = CFTYPE_NOT_ON_ROOT, |
| 6109 | .read_u64 = swap_current_read, |
| 6110 | }, |
| 6111 | { |
| 6112 | .name = "swap.max", |
| 6113 | .flags = CFTYPE_NOT_ON_ROOT, |
| 6114 | .seq_show = swap_max_show, |
| 6115 | .write = swap_max_write, |
| 6116 | }, |
| 6117 | { } /* terminate */ |
| 6118 | }; |
| 6119 | |
Johannes Weiner | 21afa38 | 2015-02-11 15:26:36 -0800 | [diff] [blame] | 6120 | static struct cftype memsw_cgroup_files[] = { |
| 6121 | { |
| 6122 | .name = "memsw.usage_in_bytes", |
| 6123 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), |
| 6124 | .read_u64 = mem_cgroup_read_u64, |
| 6125 | }, |
| 6126 | { |
| 6127 | .name = "memsw.max_usage_in_bytes", |
| 6128 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), |
| 6129 | .write = mem_cgroup_reset, |
| 6130 | .read_u64 = mem_cgroup_read_u64, |
| 6131 | }, |
| 6132 | { |
| 6133 | .name = "memsw.limit_in_bytes", |
| 6134 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), |
| 6135 | .write = mem_cgroup_write, |
| 6136 | .read_u64 = mem_cgroup_read_u64, |
| 6137 | }, |
| 6138 | { |
| 6139 | .name = "memsw.failcnt", |
| 6140 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), |
| 6141 | .write = mem_cgroup_reset, |
| 6142 | .read_u64 = mem_cgroup_read_u64, |
| 6143 | }, |
| 6144 | { }, /* terminate */ |
| 6145 | }; |
| 6146 | |
| 6147 | static int __init mem_cgroup_swap_init(void) |
| 6148 | { |
| 6149 | if (!mem_cgroup_disabled() && really_do_swap_account) { |
| 6150 | do_swap_account = 1; |
Vladimir Davydov | 37e8435 | 2016-01-20 15:02:56 -0800 | [diff] [blame] | 6151 | WARN_ON(cgroup_add_dfl_cftypes(&memory_cgrp_subsys, |
| 6152 | swap_files)); |
Johannes Weiner | 21afa38 | 2015-02-11 15:26:36 -0800 | [diff] [blame] | 6153 | WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys, |
| 6154 | memsw_cgroup_files)); |
| 6155 | } |
| 6156 | return 0; |
| 6157 | } |
| 6158 | subsys_initcall(mem_cgroup_swap_init); |
| 6159 | |
| 6160 | #endif /* CONFIG_MEMCG_SWAP */ |