Thomas Gleixner | c942fdd | 2019-05-27 08:55:06 +0200 | [diff] [blame] | 1 | // SPDX-License-Identifier: GPL-2.0-or-later |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 2 | /* memcontrol.c - Memory Controller |
| 3 | * |
| 4 | * Copyright IBM Corporation, 2007 |
| 5 | * Author Balbir Singh <balbir@linux.vnet.ibm.com> |
| 6 | * |
Pavel Emelianov | 78fb746 | 2008-02-07 00:13:51 -0800 | [diff] [blame] | 7 | * Copyright 2007 OpenVZ SWsoft Inc |
| 8 | * Author: Pavel Emelianov <xemul@openvz.org> |
| 9 | * |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 10 | * Memory thresholds |
| 11 | * Copyright (C) 2009 Nokia Corporation |
| 12 | * Author: Kirill A. Shutemov |
| 13 | * |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 14 | * Kernel Memory Controller |
| 15 | * Copyright (C) 2012 Parallels Inc. and Google Inc. |
| 16 | * Authors: Glauber Costa and Suleiman Souhlal |
| 17 | * |
Johannes Weiner | 1575e68 | 2015-04-14 15:44:51 -0700 | [diff] [blame] | 18 | * Native page reclaim |
| 19 | * Charge lifetime sanitation |
| 20 | * Lockless page tracking & accounting |
| 21 | * Unified hierarchy configuration model |
| 22 | * Copyright (C) 2015 Red Hat, Inc., Johannes Weiner |
Alex Shi | 6168d0d | 2020-12-15 12:34:29 -0800 | [diff] [blame] | 23 | * |
| 24 | * Per memcg lru locking |
| 25 | * Copyright (C) 2020 Alibaba, Inc, Alex Shi |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 26 | */ |
| 27 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 28 | #include <linux/page_counter.h> |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 29 | #include <linux/memcontrol.h> |
| 30 | #include <linux/cgroup.h> |
Christoph Hellwig | a520110 | 2019-08-28 16:19:53 +0200 | [diff] [blame] | 31 | #include <linux/pagewalk.h> |
Ingo Molnar | 6e84f31 | 2017-02-08 18:51:29 +0100 | [diff] [blame] | 32 | #include <linux/sched/mm.h> |
Hugh Dickins | 3a4f8a0 | 2017-02-24 14:59:36 -0800 | [diff] [blame] | 33 | #include <linux/shmem_fs.h> |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 34 | #include <linux/hugetlb.h> |
KAMEZAWA Hiroyuki | d13d144 | 2009-01-07 18:07:56 -0800 | [diff] [blame] | 35 | #include <linux/pagemap.h> |
Chris Down | 1ff9e6e | 2019-03-05 15:48:09 -0800 | [diff] [blame] | 36 | #include <linux/vm_event_item.h> |
KAMEZAWA Hiroyuki | d52aa41 | 2008-02-07 00:14:24 -0800 | [diff] [blame] | 37 | #include <linux/smp.h> |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 38 | #include <linux/page-flags.h> |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 39 | #include <linux/backing-dev.h> |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 40 | #include <linux/bit_spinlock.h> |
| 41 | #include <linux/rcupdate.h> |
Balbir Singh | e222432 | 2009-04-02 16:57:39 -0700 | [diff] [blame] | 42 | #include <linux/limits.h> |
Paul Gortmaker | b9e15ba | 2011-05-26 16:00:52 -0400 | [diff] [blame] | 43 | #include <linux/export.h> |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 44 | #include <linux/mutex.h> |
Andrew Morton | bb4cc1a8 | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 45 | #include <linux/rbtree.h> |
Balbir Singh | b6ac57d | 2008-04-29 01:00:19 -0700 | [diff] [blame] | 46 | #include <linux/slab.h> |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 47 | #include <linux/swap.h> |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 48 | #include <linux/swapops.h> |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 49 | #include <linux/spinlock.h> |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 50 | #include <linux/eventfd.h> |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 51 | #include <linux/poll.h> |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 52 | #include <linux/sort.h> |
Balbir Singh | 66e1707 | 2008-02-07 00:13:56 -0800 | [diff] [blame] | 53 | #include <linux/fs.h> |
KAMEZAWA Hiroyuki | d2ceb9b | 2008-02-07 00:14:25 -0800 | [diff] [blame] | 54 | #include <linux/seq_file.h> |
Anton Vorontsov | 70ddf63 | 2013-04-29 15:08:31 -0700 | [diff] [blame] | 55 | #include <linux/vmpressure.h> |
Christoph Lameter | b69408e | 2008-10-18 20:26:14 -0700 | [diff] [blame] | 56 | #include <linux/mm_inline.h> |
Johannes Weiner | 5d1ea48 | 2014-12-10 15:44:55 -0800 | [diff] [blame] | 57 | #include <linux/swap_cgroup.h> |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 58 | #include <linux/cpu.h> |
KAMEZAWA Hiroyuki | 158e0a2 | 2010-08-10 18:03:00 -0700 | [diff] [blame] | 59 | #include <linux/oom.h> |
Johannes Weiner | 0056f4e | 2013-10-31 16:34:14 -0700 | [diff] [blame] | 60 | #include <linux/lockdep.h> |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 61 | #include <linux/file.h> |
Tejun Heo | b23afb9 | 2015-11-05 18:46:11 -0800 | [diff] [blame] | 62 | #include <linux/tracehook.h> |
Chris Down | 0e4b01d | 2019-09-23 15:34:55 -0700 | [diff] [blame] | 63 | #include <linux/psi.h> |
Johannes Weiner | c8713d0 | 2019-07-11 20:55:59 -0700 | [diff] [blame] | 64 | #include <linux/seq_buf.h> |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 65 | #include "internal.h" |
Glauber Costa | d1a4c0b | 2011-12-11 21:47:04 +0000 | [diff] [blame] | 66 | #include <net/sock.h> |
Michal Hocko | 4bd2c1e | 2012-10-08 16:33:10 -0700 | [diff] [blame] | 67 | #include <net/ip.h> |
Qiang Huang | f35c3a8 | 2013-11-12 15:08:22 -0800 | [diff] [blame] | 68 | #include "slab.h" |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 69 | |
Linus Torvalds | 7c0f6ba | 2016-12-24 11:46:01 -0800 | [diff] [blame] | 70 | #include <linux/uaccess.h> |
Balbir Singh | 8697d33 | 2008-02-07 00:13:59 -0800 | [diff] [blame] | 71 | |
KOSAKI Motohiro | cc8e970 | 2010-08-09 17:19:57 -0700 | [diff] [blame] | 72 | #include <trace/events/vmscan.h> |
| 73 | |
Tejun Heo | 073219e | 2014-02-08 10:36:58 -0500 | [diff] [blame] | 74 | struct cgroup_subsys memory_cgrp_subsys __read_mostly; |
| 75 | EXPORT_SYMBOL(memory_cgrp_subsys); |
David Rientjes | 68ae564 | 2012-12-12 13:51:57 -0800 | [diff] [blame] | 76 | |
Johannes Weiner | 7d82860 | 2016-01-14 15:20:56 -0800 | [diff] [blame] | 77 | struct mem_cgroup *root_mem_cgroup __read_mostly; |
| 78 | |
Roman Gushchin | 37d5985 | 2020-10-17 16:13:50 -0700 | [diff] [blame] | 79 | /* Active memory cgroup to use from an interrupt context */ |
| 80 | DEFINE_PER_CPU(struct mem_cgroup *, int_active_memcg); |
| 81 | |
Johannes Weiner | f7e1cb6 | 2016-01-14 15:21:29 -0800 | [diff] [blame] | 82 | /* Socket memory accounting disabled? */ |
| 83 | static bool cgroup_memory_nosocket; |
| 84 | |
Vladimir Davydov | 04823c8 | 2016-01-20 15:02:38 -0800 | [diff] [blame] | 85 | /* Kernel memory accounting disabled? */ |
| 86 | static bool cgroup_memory_nokmem; |
| 87 | |
Johannes Weiner | 21afa38 | 2015-02-11 15:26:36 -0800 | [diff] [blame] | 88 | /* Whether the swap controller is active */ |
Andrew Morton | c255a45 | 2012-07-31 16:43:02 -0700 | [diff] [blame] | 89 | #ifdef CONFIG_MEMCG_SWAP |
Johannes Weiner | eccb52e | 2020-06-03 16:02:11 -0700 | [diff] [blame] | 90 | bool cgroup_memory_noswap __read_mostly; |
KAMEZAWA Hiroyuki | c077719 | 2009-01-07 18:07:57 -0800 | [diff] [blame] | 91 | #else |
Johannes Weiner | eccb52e | 2020-06-03 16:02:11 -0700 | [diff] [blame] | 92 | #define cgroup_memory_noswap 1 |
Johannes Weiner | 2d1c498 | 2020-06-03 16:02:14 -0700 | [diff] [blame] | 93 | #endif |
KAMEZAWA Hiroyuki | c077719 | 2009-01-07 18:07:57 -0800 | [diff] [blame] | 94 | |
Tejun Heo | 97b2782 | 2019-08-26 09:06:56 -0700 | [diff] [blame] | 95 | #ifdef CONFIG_CGROUP_WRITEBACK |
| 96 | static DECLARE_WAIT_QUEUE_HEAD(memcg_cgwb_frn_waitq); |
| 97 | #endif |
| 98 | |
Johannes Weiner | 7941d21 | 2016-01-14 15:21:23 -0800 | [diff] [blame] | 99 | /* Whether legacy memory+swap accounting is active */ |
| 100 | static bool do_memsw_account(void) |
| 101 | { |
Johannes Weiner | eccb52e | 2020-06-03 16:02:11 -0700 | [diff] [blame] | 102 | return !cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_noswap; |
Johannes Weiner | 7941d21 | 2016-01-14 15:21:23 -0800 | [diff] [blame] | 103 | } |
| 104 | |
Kirill A. Shutemov | a0db00f | 2012-05-29 15:06:56 -0700 | [diff] [blame] | 105 | #define THRESHOLDS_EVENTS_TARGET 128 |
| 106 | #define SOFTLIMIT_EVENTS_TARGET 1024 |
Johannes Weiner | e9f8974 | 2011-03-23 16:42:37 -0700 | [diff] [blame] | 107 | |
Andrew Morton | bb4cc1a8 | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 108 | /* |
| 109 | * Cgroups above their limits are maintained in a RB-Tree, independent of |
| 110 | * their hierarchy representation |
| 111 | */ |
| 112 | |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 113 | struct mem_cgroup_tree_per_node { |
Andrew Morton | bb4cc1a8 | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 114 | struct rb_root rb_root; |
Davidlohr Bueso | fa90b2f | 2017-09-08 16:15:21 -0700 | [diff] [blame] | 115 | struct rb_node *rb_rightmost; |
Andrew Morton | bb4cc1a8 | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 116 | spinlock_t lock; |
| 117 | }; |
| 118 | |
Andrew Morton | bb4cc1a8 | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 119 | struct mem_cgroup_tree { |
| 120 | struct mem_cgroup_tree_per_node *rb_tree_per_node[MAX_NUMNODES]; |
| 121 | }; |
| 122 | |
| 123 | static struct mem_cgroup_tree soft_limit_tree __read_mostly; |
| 124 | |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 125 | /* for OOM */ |
| 126 | struct mem_cgroup_eventfd_list { |
| 127 | struct list_head list; |
| 128 | struct eventfd_ctx *eventfd; |
| 129 | }; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 130 | |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 131 | /* |
| 132 | * cgroup_event represents events which userspace want to receive. |
| 133 | */ |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 134 | struct mem_cgroup_event { |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 135 | /* |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 136 | * memcg which the event belongs to. |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 137 | */ |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 138 | struct mem_cgroup *memcg; |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 139 | /* |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 140 | * eventfd to signal userspace about the event. |
| 141 | */ |
| 142 | struct eventfd_ctx *eventfd; |
| 143 | /* |
| 144 | * Each of these stored in a list by the cgroup. |
| 145 | */ |
| 146 | struct list_head list; |
| 147 | /* |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 148 | * register_event() callback will be used to add new userspace |
| 149 | * waiter for changes related to this event. Use eventfd_signal() |
| 150 | * on eventfd to send notification to userspace. |
| 151 | */ |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 152 | int (*register_event)(struct mem_cgroup *memcg, |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 153 | struct eventfd_ctx *eventfd, const char *args); |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 154 | /* |
| 155 | * unregister_event() callback will be called when userspace closes |
| 156 | * the eventfd or on cgroup removing. This callback must be set, |
| 157 | * if you want provide notification functionality. |
| 158 | */ |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 159 | void (*unregister_event)(struct mem_cgroup *memcg, |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 160 | struct eventfd_ctx *eventfd); |
| 161 | /* |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 162 | * All fields below needed to unregister event when |
| 163 | * userspace closes eventfd. |
| 164 | */ |
| 165 | poll_table pt; |
| 166 | wait_queue_head_t *wqh; |
Ingo Molnar | ac6424b | 2017-06-20 12:06:13 +0200 | [diff] [blame] | 167 | wait_queue_entry_t wait; |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 168 | struct work_struct remove; |
| 169 | }; |
| 170 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 171 | static void mem_cgroup_threshold(struct mem_cgroup *memcg); |
| 172 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg); |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 173 | |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 174 | /* Stuffs for move charges at task migration. */ |
| 175 | /* |
Johannes Weiner | 1dfab5a | 2015-02-11 15:26:09 -0800 | [diff] [blame] | 176 | * Types of charges to be moved. |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 177 | */ |
Johannes Weiner | 1dfab5a | 2015-02-11 15:26:09 -0800 | [diff] [blame] | 178 | #define MOVE_ANON 0x1U |
| 179 | #define MOVE_FILE 0x2U |
| 180 | #define MOVE_MASK (MOVE_ANON | MOVE_FILE) |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 181 | |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 182 | /* "mc" and its members are protected by cgroup_mutex */ |
| 183 | static struct move_charge_struct { |
Daisuke Nishimura | b1dd693 | 2010-11-24 12:57:06 -0800 | [diff] [blame] | 184 | spinlock_t lock; /* for from, to */ |
Tejun Heo | 264a0ae | 2016-04-21 19:09:02 -0400 | [diff] [blame] | 185 | struct mm_struct *mm; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 186 | struct mem_cgroup *from; |
| 187 | struct mem_cgroup *to; |
Johannes Weiner | 1dfab5a | 2015-02-11 15:26:09 -0800 | [diff] [blame] | 188 | unsigned long flags; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 189 | unsigned long precharge; |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 190 | unsigned long moved_charge; |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 191 | unsigned long moved_swap; |
Daisuke Nishimura | 8033b97 | 2010-03-10 15:22:16 -0800 | [diff] [blame] | 192 | struct task_struct *moving_task; /* a task moving charges */ |
| 193 | wait_queue_head_t waitq; /* a waitq for other context */ |
| 194 | } mc = { |
KAMEZAWA Hiroyuki | 2bd9bb2 | 2010-08-10 18:02:58 -0700 | [diff] [blame] | 195 | .lock = __SPIN_LOCK_UNLOCKED(mc.lock), |
Daisuke Nishimura | 8033b97 | 2010-03-10 15:22:16 -0800 | [diff] [blame] | 196 | .waitq = __WAIT_QUEUE_HEAD_INITIALIZER(mc.waitq), |
| 197 | }; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 198 | |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 199 | /* |
| 200 | * Maximum loops in mem_cgroup_hierarchical_reclaim(), used for soft |
| 201 | * limit reclaim to prevent infinite loops, if they ever occur. |
| 202 | */ |
Kirill A. Shutemov | a0db00f | 2012-05-29 15:06:56 -0700 | [diff] [blame] | 203 | #define MEM_CGROUP_MAX_RECLAIM_LOOPS 100 |
Andrew Morton | bb4cc1a8 | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 204 | #define MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS 2 |
Balbir Singh | 4e41695 | 2009-09-23 15:56:39 -0700 | [diff] [blame] | 205 | |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 206 | /* for encoding cft->private value on file */ |
Glauber Costa | 86ae53e | 2012-12-18 14:21:45 -0800 | [diff] [blame] | 207 | enum res_type { |
| 208 | _MEM, |
| 209 | _MEMSWAP, |
| 210 | _OOM_TYPE, |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 211 | _KMEM, |
Vladimir Davydov | d55f90b | 2016-01-20 15:02:44 -0800 | [diff] [blame] | 212 | _TCP, |
Glauber Costa | 86ae53e | 2012-12-18 14:21:45 -0800 | [diff] [blame] | 213 | }; |
| 214 | |
Kirill A. Shutemov | a0db00f | 2012-05-29 15:06:56 -0700 | [diff] [blame] | 215 | #define MEMFILE_PRIVATE(x, val) ((x) << 16 | (val)) |
| 216 | #define MEMFILE_TYPE(val) ((val) >> 16 & 0xffff) |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 217 | #define MEMFILE_ATTR(val) ((val) & 0xffff) |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 218 | /* Used for OOM nofiier */ |
| 219 | #define OOM_CONTROL (0) |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 220 | |
Kirill Tkhai | b05706f | 2018-08-17 15:47:33 -0700 | [diff] [blame] | 221 | /* |
| 222 | * Iteration constructs for visiting all cgroups (under a tree). If |
| 223 | * loops are exited prematurely (break), mem_cgroup_iter_break() must |
| 224 | * be used for reference counting. |
| 225 | */ |
| 226 | #define for_each_mem_cgroup_tree(iter, root) \ |
| 227 | for (iter = mem_cgroup_iter(root, NULL, NULL); \ |
| 228 | iter != NULL; \ |
| 229 | iter = mem_cgroup_iter(root, iter, NULL)) |
| 230 | |
| 231 | #define for_each_mem_cgroup(iter) \ |
| 232 | for (iter = mem_cgroup_iter(NULL, NULL, NULL); \ |
| 233 | iter != NULL; \ |
| 234 | iter = mem_cgroup_iter(NULL, iter, NULL)) |
| 235 | |
Tetsuo Handa | 7775fac | 2019-03-05 15:46:47 -0800 | [diff] [blame] | 236 | static inline bool should_force_charge(void) |
| 237 | { |
| 238 | return tsk_is_oom_victim(current) || fatal_signal_pending(current) || |
| 239 | (current->flags & PF_EXITING); |
| 240 | } |
| 241 | |
Anton Vorontsov | 70ddf63 | 2013-04-29 15:08:31 -0700 | [diff] [blame] | 242 | /* Some nice accessors for the vmpressure. */ |
| 243 | struct vmpressure *memcg_to_vmpressure(struct mem_cgroup *memcg) |
| 244 | { |
| 245 | if (!memcg) |
| 246 | memcg = root_mem_cgroup; |
| 247 | return &memcg->vmpressure; |
| 248 | } |
| 249 | |
| 250 | struct cgroup_subsys_state *vmpressure_to_css(struct vmpressure *vmpr) |
| 251 | { |
| 252 | return &container_of(vmpr, struct mem_cgroup, vmpressure)->css; |
| 253 | } |
| 254 | |
Kirill Tkhai | 84c07d1 | 2018-08-17 15:47:25 -0700 | [diff] [blame] | 255 | #ifdef CONFIG_MEMCG_KMEM |
Roman Gushchin | bf4f059 | 2020-08-06 23:20:49 -0700 | [diff] [blame] | 256 | extern spinlock_t css_set_lock; |
| 257 | |
Roman Gushchin | c1a660d | 2021-02-24 12:03:58 -0800 | [diff] [blame] | 258 | static int __memcg_kmem_charge(struct mem_cgroup *memcg, gfp_t gfp, |
| 259 | unsigned int nr_pages); |
| 260 | static void __memcg_kmem_uncharge(struct mem_cgroup *memcg, |
| 261 | unsigned int nr_pages); |
| 262 | |
Roman Gushchin | bf4f059 | 2020-08-06 23:20:49 -0700 | [diff] [blame] | 263 | static void obj_cgroup_release(struct percpu_ref *ref) |
| 264 | { |
| 265 | struct obj_cgroup *objcg = container_of(ref, struct obj_cgroup, refcnt); |
| 266 | struct mem_cgroup *memcg; |
| 267 | unsigned int nr_bytes; |
| 268 | unsigned int nr_pages; |
| 269 | unsigned long flags; |
| 270 | |
| 271 | /* |
| 272 | * At this point all allocated objects are freed, and |
| 273 | * objcg->nr_charged_bytes can't have an arbitrary byte value. |
| 274 | * However, it can be PAGE_SIZE or (x * PAGE_SIZE). |
| 275 | * |
| 276 | * The following sequence can lead to it: |
| 277 | * 1) CPU0: objcg == stock->cached_objcg |
| 278 | * 2) CPU1: we do a small allocation (e.g. 92 bytes), |
| 279 | * PAGE_SIZE bytes are charged |
| 280 | * 3) CPU1: a process from another memcg is allocating something, |
| 281 | * the stock if flushed, |
| 282 | * objcg->nr_charged_bytes = PAGE_SIZE - 92 |
| 283 | * 5) CPU0: we do release this object, |
| 284 | * 92 bytes are added to stock->nr_bytes |
| 285 | * 6) CPU0: stock is flushed, |
| 286 | * 92 bytes are added to objcg->nr_charged_bytes |
| 287 | * |
| 288 | * In the result, nr_charged_bytes == PAGE_SIZE. |
| 289 | * This page will be uncharged in obj_cgroup_release(). |
| 290 | */ |
| 291 | nr_bytes = atomic_read(&objcg->nr_charged_bytes); |
| 292 | WARN_ON_ONCE(nr_bytes & (PAGE_SIZE - 1)); |
| 293 | nr_pages = nr_bytes >> PAGE_SHIFT; |
| 294 | |
| 295 | spin_lock_irqsave(&css_set_lock, flags); |
| 296 | memcg = obj_cgroup_memcg(objcg); |
| 297 | if (nr_pages) |
| 298 | __memcg_kmem_uncharge(memcg, nr_pages); |
| 299 | list_del(&objcg->list); |
| 300 | mem_cgroup_put(memcg); |
| 301 | spin_unlock_irqrestore(&css_set_lock, flags); |
| 302 | |
| 303 | percpu_ref_exit(ref); |
| 304 | kfree_rcu(objcg, rcu); |
| 305 | } |
| 306 | |
| 307 | static struct obj_cgroup *obj_cgroup_alloc(void) |
| 308 | { |
| 309 | struct obj_cgroup *objcg; |
| 310 | int ret; |
| 311 | |
| 312 | objcg = kzalloc(sizeof(struct obj_cgroup), GFP_KERNEL); |
| 313 | if (!objcg) |
| 314 | return NULL; |
| 315 | |
| 316 | ret = percpu_ref_init(&objcg->refcnt, obj_cgroup_release, 0, |
| 317 | GFP_KERNEL); |
| 318 | if (ret) { |
| 319 | kfree(objcg); |
| 320 | return NULL; |
| 321 | } |
| 322 | INIT_LIST_HEAD(&objcg->list); |
| 323 | return objcg; |
| 324 | } |
| 325 | |
| 326 | static void memcg_reparent_objcgs(struct mem_cgroup *memcg, |
| 327 | struct mem_cgroup *parent) |
| 328 | { |
| 329 | struct obj_cgroup *objcg, *iter; |
| 330 | |
| 331 | objcg = rcu_replace_pointer(memcg->objcg, NULL, true); |
| 332 | |
| 333 | spin_lock_irq(&css_set_lock); |
| 334 | |
| 335 | /* Move active objcg to the parent's list */ |
| 336 | xchg(&objcg->memcg, parent); |
| 337 | css_get(&parent->css); |
| 338 | list_add(&objcg->list, &parent->objcg_list); |
| 339 | |
| 340 | /* Move already reparented objcgs to the parent's list */ |
| 341 | list_for_each_entry(iter, &memcg->objcg_list, list) { |
| 342 | css_get(&parent->css); |
| 343 | xchg(&iter->memcg, parent); |
| 344 | css_put(&memcg->css); |
| 345 | } |
| 346 | list_splice(&memcg->objcg_list, &parent->objcg_list); |
| 347 | |
| 348 | spin_unlock_irq(&css_set_lock); |
| 349 | |
| 350 | percpu_ref_kill(&objcg->refcnt); |
| 351 | } |
| 352 | |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 353 | /* |
Roman Gushchin | 9855609 | 2020-08-06 23:21:10 -0700 | [diff] [blame] | 354 | * This will be used as a shrinker list's index. |
Li Zefan | b862783 | 2013-09-23 16:56:47 +0800 | [diff] [blame] | 355 | * The main reason for not using cgroup id for this: |
| 356 | * this works better in sparse environments, where we have a lot of memcgs, |
| 357 | * but only a few kmem-limited. Or also, if we have, for instance, 200 |
| 358 | * memcgs, and none but the 200th is kmem-limited, we'd have to have a |
| 359 | * 200 entry array for that. |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 360 | * |
Vladimir Davydov | dbcf73e | 2015-02-12 14:58:57 -0800 | [diff] [blame] | 361 | * The current size of the caches array is stored in memcg_nr_cache_ids. It |
| 362 | * will double each time we have to increase it. |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 363 | */ |
Vladimir Davydov | dbcf73e | 2015-02-12 14:58:57 -0800 | [diff] [blame] | 364 | static DEFINE_IDA(memcg_cache_ida); |
| 365 | int memcg_nr_cache_ids; |
Glauber Costa | 749c541 | 2012-12-18 14:23:01 -0800 | [diff] [blame] | 366 | |
Vladimir Davydov | 05257a1 | 2015-02-12 14:59:01 -0800 | [diff] [blame] | 367 | /* Protects memcg_nr_cache_ids */ |
| 368 | static DECLARE_RWSEM(memcg_cache_ids_sem); |
| 369 | |
| 370 | void memcg_get_cache_ids(void) |
| 371 | { |
| 372 | down_read(&memcg_cache_ids_sem); |
| 373 | } |
| 374 | |
| 375 | void memcg_put_cache_ids(void) |
| 376 | { |
| 377 | up_read(&memcg_cache_ids_sem); |
| 378 | } |
| 379 | |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 380 | /* |
| 381 | * MIN_SIZE is different than 1, because we would like to avoid going through |
| 382 | * the alloc/free process all the time. In a small machine, 4 kmem-limited |
| 383 | * cgroups is a reasonable guess. In the future, it could be a parameter or |
| 384 | * tunable, but that is strictly not necessary. |
| 385 | * |
Li Zefan | b862783 | 2013-09-23 16:56:47 +0800 | [diff] [blame] | 386 | * 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] | 387 | * this constant directly from cgroup, but it is understandable that this is |
| 388 | * 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] | 389 | * 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] | 390 | * increase ours as well if it increases. |
| 391 | */ |
| 392 | #define MEMCG_CACHES_MIN_SIZE 4 |
Li Zefan | b862783 | 2013-09-23 16:56:47 +0800 | [diff] [blame] | 393 | #define MEMCG_CACHES_MAX_SIZE MEM_CGROUP_ID_MAX |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 394 | |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 395 | /* |
| 396 | * A lot of the calls to the cache allocation functions are expected to be |
Roman Gushchin | 272911a | 2020-08-06 23:21:17 -0700 | [diff] [blame] | 397 | * inlined by the compiler. Since the calls to memcg_slab_pre_alloc_hook() are |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 398 | * conditional to this static branch, we'll have to allow modules that does |
| 399 | * kmem_cache_alloc and the such to see this symbol as well |
| 400 | */ |
Johannes Weiner | ef12947 | 2016-01-14 15:21:34 -0800 | [diff] [blame] | 401 | DEFINE_STATIC_KEY_FALSE(memcg_kmem_enabled_key); |
Glauber Costa | d7f25f8 | 2012-12-18 14:22:40 -0800 | [diff] [blame] | 402 | EXPORT_SYMBOL(memcg_kmem_enabled_key); |
Yang Shi | 0a432dc | 2019-09-23 15:38:12 -0700 | [diff] [blame] | 403 | #endif |
Tejun Heo | 17cc4df | 2017-02-22 15:41:36 -0800 | [diff] [blame] | 404 | |
Kirill Tkhai | 0a4465d | 2018-08-17 15:47:37 -0700 | [diff] [blame] | 405 | static int memcg_shrinker_map_size; |
| 406 | static DEFINE_MUTEX(memcg_shrinker_map_mutex); |
| 407 | |
| 408 | static void memcg_free_shrinker_map_rcu(struct rcu_head *head) |
| 409 | { |
| 410 | kvfree(container_of(head, struct memcg_shrinker_map, rcu)); |
| 411 | } |
| 412 | |
| 413 | static int memcg_expand_one_shrinker_map(struct mem_cgroup *memcg, |
| 414 | int size, int old_size) |
| 415 | { |
| 416 | struct memcg_shrinker_map *new, *old; |
Johannes Weiner | a3747b5 | 2021-04-29 22:56:14 -0700 | [diff] [blame] | 417 | struct mem_cgroup_per_node *pn; |
Kirill Tkhai | 0a4465d | 2018-08-17 15:47:37 -0700 | [diff] [blame] | 418 | int nid; |
| 419 | |
| 420 | lockdep_assert_held(&memcg_shrinker_map_mutex); |
| 421 | |
| 422 | for_each_node(nid) { |
Johannes Weiner | a3747b5 | 2021-04-29 22:56:14 -0700 | [diff] [blame] | 423 | pn = memcg->nodeinfo[nid]; |
| 424 | old = rcu_dereference_protected(pn->shrinker_map, true); |
Kirill Tkhai | 0a4465d | 2018-08-17 15:47:37 -0700 | [diff] [blame] | 425 | /* Not yet online memcg */ |
| 426 | if (!old) |
| 427 | return 0; |
| 428 | |
Kirill Tkhai | 86daf94 | 2020-04-01 21:06:33 -0700 | [diff] [blame] | 429 | new = kvmalloc_node(sizeof(*new) + size, GFP_KERNEL, nid); |
Kirill Tkhai | 0a4465d | 2018-08-17 15:47:37 -0700 | [diff] [blame] | 430 | if (!new) |
| 431 | return -ENOMEM; |
| 432 | |
| 433 | /* Set all old bits, clear all new bits */ |
| 434 | memset(new->map, (int)0xff, old_size); |
| 435 | memset((void *)new->map + old_size, 0, size - old_size); |
| 436 | |
Johannes Weiner | a3747b5 | 2021-04-29 22:56:14 -0700 | [diff] [blame] | 437 | rcu_assign_pointer(pn->shrinker_map, new); |
Kirill Tkhai | 0a4465d | 2018-08-17 15:47:37 -0700 | [diff] [blame] | 438 | call_rcu(&old->rcu, memcg_free_shrinker_map_rcu); |
| 439 | } |
| 440 | |
| 441 | return 0; |
| 442 | } |
| 443 | |
| 444 | static void memcg_free_shrinker_maps(struct mem_cgroup *memcg) |
| 445 | { |
| 446 | struct mem_cgroup_per_node *pn; |
| 447 | struct memcg_shrinker_map *map; |
| 448 | int nid; |
| 449 | |
| 450 | if (mem_cgroup_is_root(memcg)) |
| 451 | return; |
| 452 | |
| 453 | for_each_node(nid) { |
Johannes Weiner | a3747b5 | 2021-04-29 22:56:14 -0700 | [diff] [blame] | 454 | pn = memcg->nodeinfo[nid]; |
Kirill Tkhai | 0a4465d | 2018-08-17 15:47:37 -0700 | [diff] [blame] | 455 | map = rcu_dereference_protected(pn->shrinker_map, true); |
Yang Li | 8a26016 | 2021-02-24 12:04:05 -0800 | [diff] [blame] | 456 | kvfree(map); |
Kirill Tkhai | 0a4465d | 2018-08-17 15:47:37 -0700 | [diff] [blame] | 457 | rcu_assign_pointer(pn->shrinker_map, NULL); |
| 458 | } |
| 459 | } |
| 460 | |
| 461 | static int memcg_alloc_shrinker_maps(struct mem_cgroup *memcg) |
| 462 | { |
| 463 | struct memcg_shrinker_map *map; |
| 464 | int nid, size, ret = 0; |
| 465 | |
| 466 | if (mem_cgroup_is_root(memcg)) |
| 467 | return 0; |
| 468 | |
| 469 | mutex_lock(&memcg_shrinker_map_mutex); |
| 470 | size = memcg_shrinker_map_size; |
| 471 | for_each_node(nid) { |
Kirill Tkhai | 86daf94 | 2020-04-01 21:06:33 -0700 | [diff] [blame] | 472 | map = kvzalloc_node(sizeof(*map) + size, GFP_KERNEL, nid); |
Kirill Tkhai | 0a4465d | 2018-08-17 15:47:37 -0700 | [diff] [blame] | 473 | if (!map) { |
| 474 | memcg_free_shrinker_maps(memcg); |
| 475 | ret = -ENOMEM; |
| 476 | break; |
| 477 | } |
| 478 | rcu_assign_pointer(memcg->nodeinfo[nid]->shrinker_map, map); |
| 479 | } |
| 480 | mutex_unlock(&memcg_shrinker_map_mutex); |
| 481 | |
| 482 | return ret; |
| 483 | } |
| 484 | |
| 485 | int memcg_expand_shrinker_maps(int new_id) |
| 486 | { |
| 487 | int size, old_size, ret = 0; |
| 488 | struct mem_cgroup *memcg; |
| 489 | |
| 490 | size = DIV_ROUND_UP(new_id + 1, BITS_PER_LONG) * sizeof(unsigned long); |
| 491 | old_size = memcg_shrinker_map_size; |
| 492 | if (size <= old_size) |
| 493 | return 0; |
| 494 | |
| 495 | mutex_lock(&memcg_shrinker_map_mutex); |
| 496 | if (!root_mem_cgroup) |
| 497 | goto unlock; |
| 498 | |
| 499 | for_each_mem_cgroup(memcg) { |
| 500 | if (mem_cgroup_is_root(memcg)) |
| 501 | continue; |
| 502 | ret = memcg_expand_one_shrinker_map(memcg, size, old_size); |
Vasily Averin | 75866af | 2020-02-20 20:04:18 -0800 | [diff] [blame] | 503 | if (ret) { |
| 504 | mem_cgroup_iter_break(NULL, memcg); |
Kirill Tkhai | 0a4465d | 2018-08-17 15:47:37 -0700 | [diff] [blame] | 505 | goto unlock; |
Vasily Averin | 75866af | 2020-02-20 20:04:18 -0800 | [diff] [blame] | 506 | } |
Kirill Tkhai | 0a4465d | 2018-08-17 15:47:37 -0700 | [diff] [blame] | 507 | } |
| 508 | unlock: |
| 509 | if (!ret) |
| 510 | memcg_shrinker_map_size = size; |
| 511 | mutex_unlock(&memcg_shrinker_map_mutex); |
| 512 | return ret; |
| 513 | } |
Kirill Tkhai | fae91d6 | 2018-08-17 15:48:10 -0700 | [diff] [blame] | 514 | |
| 515 | void memcg_set_shrinker_bit(struct mem_cgroup *memcg, int nid, int shrinker_id) |
| 516 | { |
| 517 | if (shrinker_id >= 0 && memcg && !mem_cgroup_is_root(memcg)) { |
| 518 | struct memcg_shrinker_map *map; |
| 519 | |
| 520 | rcu_read_lock(); |
| 521 | map = rcu_dereference(memcg->nodeinfo[nid]->shrinker_map); |
Kirill Tkhai | f90280d | 2018-08-17 15:48:25 -0700 | [diff] [blame] | 522 | /* Pairs with smp mb in shrink_slab() */ |
| 523 | smp_mb__before_atomic(); |
Kirill Tkhai | fae91d6 | 2018-08-17 15:48:10 -0700 | [diff] [blame] | 524 | set_bit(shrinker_id, map->map); |
| 525 | rcu_read_unlock(); |
| 526 | } |
| 527 | } |
| 528 | |
Tejun Heo | ad7fa85 | 2015-05-27 20:00:02 -0400 | [diff] [blame] | 529 | /** |
| 530 | * mem_cgroup_css_from_page - css of the memcg associated with a page |
| 531 | * @page: page of interest |
| 532 | * |
| 533 | * If memcg is bound to the default hierarchy, css of the memcg associated |
| 534 | * with @page is returned. The returned css remains associated with @page |
| 535 | * until it is released. |
| 536 | * |
| 537 | * If memcg is bound to a traditional hierarchy, the css of root_mem_cgroup |
| 538 | * is returned. |
Tejun Heo | ad7fa85 | 2015-05-27 20:00:02 -0400 | [diff] [blame] | 539 | */ |
| 540 | struct cgroup_subsys_state *mem_cgroup_css_from_page(struct page *page) |
| 541 | { |
| 542 | struct mem_cgroup *memcg; |
| 543 | |
Roman Gushchin | bcfe06b | 2020-12-01 13:58:27 -0800 | [diff] [blame] | 544 | memcg = page_memcg(page); |
Tejun Heo | ad7fa85 | 2015-05-27 20:00:02 -0400 | [diff] [blame] | 545 | |
Tejun Heo | 9e10a13 | 2015-09-18 11:56:28 -0400 | [diff] [blame] | 546 | if (!memcg || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
Tejun Heo | ad7fa85 | 2015-05-27 20:00:02 -0400 | [diff] [blame] | 547 | memcg = root_mem_cgroup; |
| 548 | |
Tejun Heo | ad7fa85 | 2015-05-27 20:00:02 -0400 | [diff] [blame] | 549 | return &memcg->css; |
| 550 | } |
| 551 | |
Vladimir Davydov | 2fc0452 | 2015-09-09 15:35:28 -0700 | [diff] [blame] | 552 | /** |
| 553 | * page_cgroup_ino - return inode number of the memcg a page is charged to |
| 554 | * @page: the page |
| 555 | * |
| 556 | * Look up the closest online ancestor of the memory cgroup @page is charged to |
| 557 | * and return its inode number or 0 if @page is not charged to any cgroup. It |
| 558 | * is safe to call this function without holding a reference to @page. |
| 559 | * |
| 560 | * Note, this function is inherently racy, because there is nothing to prevent |
| 561 | * the cgroup inode from getting torn down and potentially reallocated a moment |
| 562 | * after page_cgroup_ino() returns, so it only should be used by callers that |
| 563 | * do not care (such as procfs interfaces). |
| 564 | */ |
| 565 | ino_t page_cgroup_ino(struct page *page) |
| 566 | { |
| 567 | struct mem_cgroup *memcg; |
| 568 | unsigned long ino = 0; |
| 569 | |
| 570 | rcu_read_lock(); |
Roman Gushchin | bcfe06b | 2020-12-01 13:58:27 -0800 | [diff] [blame] | 571 | memcg = page_memcg_check(page); |
Roman Gushchin | 286e04b | 2020-08-06 23:20:52 -0700 | [diff] [blame] | 572 | |
Vladimir Davydov | 2fc0452 | 2015-09-09 15:35:28 -0700 | [diff] [blame] | 573 | while (memcg && !(memcg->css.flags & CSS_ONLINE)) |
| 574 | memcg = parent_mem_cgroup(memcg); |
| 575 | if (memcg) |
| 576 | ino = cgroup_ino(memcg->css.cgroup); |
| 577 | rcu_read_unlock(); |
| 578 | return ino; |
| 579 | } |
| 580 | |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 581 | static struct mem_cgroup_per_node * |
| 582 | mem_cgroup_page_nodeinfo(struct mem_cgroup *memcg, struct page *page) |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 583 | { |
Johannes Weiner | 97a6c37 | 2011-03-23 16:42:27 -0700 | [diff] [blame] | 584 | int nid = page_to_nid(page); |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 585 | |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 586 | return memcg->nodeinfo[nid]; |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 587 | } |
| 588 | |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 589 | static struct mem_cgroup_tree_per_node * |
| 590 | soft_limit_tree_node(int nid) |
Andrew Morton | bb4cc1a8 | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 591 | { |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 592 | return soft_limit_tree.rb_tree_per_node[nid]; |
Andrew Morton | bb4cc1a8 | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 593 | } |
| 594 | |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 595 | static struct mem_cgroup_tree_per_node * |
Andrew Morton | bb4cc1a8 | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 596 | soft_limit_tree_from_page(struct page *page) |
| 597 | { |
| 598 | int nid = page_to_nid(page); |
Andrew Morton | bb4cc1a8 | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 599 | |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 600 | return soft_limit_tree.rb_tree_per_node[nid]; |
Andrew Morton | bb4cc1a8 | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 601 | } |
| 602 | |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 603 | static void __mem_cgroup_insert_exceeded(struct mem_cgroup_per_node *mz, |
| 604 | struct mem_cgroup_tree_per_node *mctz, |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 605 | unsigned long new_usage_in_excess) |
Andrew Morton | bb4cc1a8 | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 606 | { |
| 607 | struct rb_node **p = &mctz->rb_root.rb_node; |
| 608 | struct rb_node *parent = NULL; |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 609 | struct mem_cgroup_per_node *mz_node; |
Davidlohr Bueso | fa90b2f | 2017-09-08 16:15:21 -0700 | [diff] [blame] | 610 | bool rightmost = true; |
Andrew Morton | bb4cc1a8 | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 611 | |
| 612 | if (mz->on_tree) |
| 613 | return; |
| 614 | |
| 615 | mz->usage_in_excess = new_usage_in_excess; |
| 616 | if (!mz->usage_in_excess) |
| 617 | return; |
| 618 | while (*p) { |
| 619 | parent = *p; |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 620 | mz_node = rb_entry(parent, struct mem_cgroup_per_node, |
Andrew Morton | bb4cc1a8 | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 621 | tree_node); |
Davidlohr Bueso | fa90b2f | 2017-09-08 16:15:21 -0700 | [diff] [blame] | 622 | if (mz->usage_in_excess < mz_node->usage_in_excess) { |
Andrew Morton | bb4cc1a8 | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 623 | p = &(*p)->rb_left; |
Davidlohr Bueso | fa90b2f | 2017-09-08 16:15:21 -0700 | [diff] [blame] | 624 | rightmost = false; |
Miaohe Lin | 378876b | 2020-12-14 19:06:28 -0800 | [diff] [blame] | 625 | } else { |
Andrew Morton | bb4cc1a8 | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 626 | p = &(*p)->rb_right; |
Miaohe Lin | 378876b | 2020-12-14 19:06:28 -0800 | [diff] [blame] | 627 | } |
Andrew Morton | bb4cc1a8 | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 628 | } |
Davidlohr Bueso | fa90b2f | 2017-09-08 16:15:21 -0700 | [diff] [blame] | 629 | |
| 630 | if (rightmost) |
| 631 | mctz->rb_rightmost = &mz->tree_node; |
| 632 | |
Andrew Morton | bb4cc1a8 | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 633 | rb_link_node(&mz->tree_node, parent, p); |
| 634 | rb_insert_color(&mz->tree_node, &mctz->rb_root); |
| 635 | mz->on_tree = true; |
| 636 | } |
| 637 | |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 638 | static void __mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz, |
| 639 | struct mem_cgroup_tree_per_node *mctz) |
Andrew Morton | bb4cc1a8 | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 640 | { |
| 641 | if (!mz->on_tree) |
| 642 | return; |
Davidlohr Bueso | fa90b2f | 2017-09-08 16:15:21 -0700 | [diff] [blame] | 643 | |
| 644 | if (&mz->tree_node == mctz->rb_rightmost) |
| 645 | mctz->rb_rightmost = rb_prev(&mz->tree_node); |
| 646 | |
Andrew Morton | bb4cc1a8 | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 647 | rb_erase(&mz->tree_node, &mctz->rb_root); |
| 648 | mz->on_tree = false; |
| 649 | } |
| 650 | |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 651 | static void mem_cgroup_remove_exceeded(struct mem_cgroup_per_node *mz, |
| 652 | struct mem_cgroup_tree_per_node *mctz) |
Andrew Morton | bb4cc1a8 | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 653 | { |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 654 | unsigned long flags; |
| 655 | |
| 656 | spin_lock_irqsave(&mctz->lock, flags); |
Johannes Weiner | cf2c812 | 2014-06-06 14:38:21 -0700 | [diff] [blame] | 657 | __mem_cgroup_remove_exceeded(mz, mctz); |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 658 | spin_unlock_irqrestore(&mctz->lock, flags); |
Andrew Morton | bb4cc1a8 | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 659 | } |
| 660 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 661 | static unsigned long soft_limit_excess(struct mem_cgroup *memcg) |
| 662 | { |
| 663 | unsigned long nr_pages = page_counter_read(&memcg->memory); |
Jason Low | 4db0c3c | 2015-04-15 16:14:08 -0700 | [diff] [blame] | 664 | unsigned long soft_limit = READ_ONCE(memcg->soft_limit); |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 665 | unsigned long excess = 0; |
| 666 | |
| 667 | if (nr_pages > soft_limit) |
| 668 | excess = nr_pages - soft_limit; |
| 669 | |
| 670 | return excess; |
| 671 | } |
Andrew Morton | bb4cc1a8 | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 672 | |
| 673 | static void mem_cgroup_update_tree(struct mem_cgroup *memcg, struct page *page) |
| 674 | { |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 675 | unsigned long excess; |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 676 | struct mem_cgroup_per_node *mz; |
| 677 | struct mem_cgroup_tree_per_node *mctz; |
Andrew Morton | bb4cc1a8 | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 678 | |
Jianyu Zhan | e231875 | 2014-06-06 14:38:20 -0700 | [diff] [blame] | 679 | mctz = soft_limit_tree_from_page(page); |
Laurent Dufour | bfc7228 | 2017-03-09 16:17:06 -0800 | [diff] [blame] | 680 | if (!mctz) |
| 681 | return; |
Andrew Morton | bb4cc1a8 | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 682 | /* |
| 683 | * Necessary to update all ancestors when hierarchy is used. |
| 684 | * because their event counter is not touched. |
| 685 | */ |
| 686 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 687 | mz = mem_cgroup_page_nodeinfo(memcg, page); |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 688 | excess = soft_limit_excess(memcg); |
Andrew Morton | bb4cc1a8 | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 689 | /* |
| 690 | * We have to update the tree if mz is on RB-tree or |
| 691 | * mem is over its softlimit. |
| 692 | */ |
| 693 | if (excess || mz->on_tree) { |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 694 | unsigned long flags; |
| 695 | |
| 696 | spin_lock_irqsave(&mctz->lock, flags); |
Andrew Morton | bb4cc1a8 | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 697 | /* if on-tree, remove it */ |
| 698 | if (mz->on_tree) |
Johannes Weiner | cf2c812 | 2014-06-06 14:38:21 -0700 | [diff] [blame] | 699 | __mem_cgroup_remove_exceeded(mz, mctz); |
Andrew Morton | bb4cc1a8 | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 700 | /* |
| 701 | * Insert again. mz->usage_in_excess will be updated. |
| 702 | * If excess is 0, no tree ops. |
| 703 | */ |
Johannes Weiner | cf2c812 | 2014-06-06 14:38:21 -0700 | [diff] [blame] | 704 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 705 | spin_unlock_irqrestore(&mctz->lock, flags); |
Andrew Morton | bb4cc1a8 | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 706 | } |
| 707 | } |
| 708 | } |
| 709 | |
| 710 | static void mem_cgroup_remove_from_trees(struct mem_cgroup *memcg) |
| 711 | { |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 712 | struct mem_cgroup_tree_per_node *mctz; |
| 713 | struct mem_cgroup_per_node *mz; |
| 714 | int nid; |
Andrew Morton | bb4cc1a8 | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 715 | |
Jianyu Zhan | e231875 | 2014-06-06 14:38:20 -0700 | [diff] [blame] | 716 | for_each_node(nid) { |
Johannes Weiner | a3747b5 | 2021-04-29 22:56:14 -0700 | [diff] [blame] | 717 | mz = memcg->nodeinfo[nid]; |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 718 | mctz = soft_limit_tree_node(nid); |
Laurent Dufour | bfc7228 | 2017-03-09 16:17:06 -0800 | [diff] [blame] | 719 | if (mctz) |
| 720 | mem_cgroup_remove_exceeded(mz, mctz); |
Andrew Morton | bb4cc1a8 | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 721 | } |
| 722 | } |
| 723 | |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 724 | static struct mem_cgroup_per_node * |
| 725 | __mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz) |
Andrew Morton | bb4cc1a8 | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 726 | { |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 727 | struct mem_cgroup_per_node *mz; |
Andrew Morton | bb4cc1a8 | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 728 | |
| 729 | retry: |
| 730 | mz = NULL; |
Davidlohr Bueso | fa90b2f | 2017-09-08 16:15:21 -0700 | [diff] [blame] | 731 | if (!mctz->rb_rightmost) |
Andrew Morton | bb4cc1a8 | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 732 | goto done; /* Nothing to reclaim from */ |
| 733 | |
Davidlohr Bueso | fa90b2f | 2017-09-08 16:15:21 -0700 | [diff] [blame] | 734 | mz = rb_entry(mctz->rb_rightmost, |
| 735 | struct mem_cgroup_per_node, tree_node); |
Andrew Morton | bb4cc1a8 | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 736 | /* |
| 737 | * Remove the node now but someone else can add it back, |
| 738 | * we will to add it back at the end of reclaim to its correct |
| 739 | * position in the tree. |
| 740 | */ |
Johannes Weiner | cf2c812 | 2014-06-06 14:38:21 -0700 | [diff] [blame] | 741 | __mem_cgroup_remove_exceeded(mz, mctz); |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 742 | if (!soft_limit_excess(mz->memcg) || |
Shakeel Butt | 8965aa2 | 2020-04-01 21:07:10 -0700 | [diff] [blame] | 743 | !css_tryget(&mz->memcg->css)) |
Andrew Morton | bb4cc1a8 | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 744 | goto retry; |
| 745 | done: |
| 746 | return mz; |
| 747 | } |
| 748 | |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 749 | static struct mem_cgroup_per_node * |
| 750 | mem_cgroup_largest_soft_limit_node(struct mem_cgroup_tree_per_node *mctz) |
Andrew Morton | bb4cc1a8 | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 751 | { |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 752 | struct mem_cgroup_per_node *mz; |
Andrew Morton | bb4cc1a8 | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 753 | |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 754 | spin_lock_irq(&mctz->lock); |
Andrew Morton | bb4cc1a8 | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 755 | mz = __mem_cgroup_largest_soft_limit_node(mctz); |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 756 | spin_unlock_irq(&mctz->lock); |
Andrew Morton | bb4cc1a8 | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 757 | return mz; |
| 758 | } |
| 759 | |
Johannes Weiner | db9adbc | 2019-05-14 15:47:09 -0700 | [diff] [blame] | 760 | /** |
| 761 | * __mod_memcg_state - update cgroup memory statistics |
| 762 | * @memcg: the memory cgroup |
| 763 | * @idx: the stat item - can be enum memcg_stat_item or enum node_stat_item |
| 764 | * @val: delta to add to the counter, can be negative |
| 765 | */ |
| 766 | void __mod_memcg_state(struct mem_cgroup *memcg, int idx, int val) |
| 767 | { |
Johannes Weiner | db9adbc | 2019-05-14 15:47:09 -0700 | [diff] [blame] | 768 | if (mem_cgroup_disabled()) |
| 769 | return; |
| 770 | |
Johannes Weiner | 2d146aa | 2021-04-29 22:56:26 -0700 | [diff] [blame] | 771 | __this_cpu_add(memcg->vmstats_percpu->state[idx], val); |
| 772 | cgroup_rstat_updated(memcg->css.cgroup, smp_processor_id()); |
Johannes Weiner | db9adbc | 2019-05-14 15:47:09 -0700 | [diff] [blame] | 773 | } |
| 774 | |
Johannes Weiner | 2d146aa | 2021-04-29 22:56:26 -0700 | [diff] [blame] | 775 | /* idx can be of type enum memcg_stat_item or node_stat_item. */ |
Johannes Weiner | a18e6e6 | 2021-04-29 22:56:17 -0700 | [diff] [blame] | 776 | static unsigned long memcg_page_state(struct mem_cgroup *memcg, int idx) |
| 777 | { |
Johannes Weiner | 2d146aa | 2021-04-29 22:56:26 -0700 | [diff] [blame] | 778 | long x = READ_ONCE(memcg->vmstats.state[idx]); |
Johannes Weiner | a18e6e6 | 2021-04-29 22:56:17 -0700 | [diff] [blame] | 779 | #ifdef CONFIG_SMP |
| 780 | if (x < 0) |
| 781 | x = 0; |
| 782 | #endif |
| 783 | return x; |
| 784 | } |
| 785 | |
Johannes Weiner | 2d146aa | 2021-04-29 22:56:26 -0700 | [diff] [blame] | 786 | /* idx can be of type enum memcg_stat_item or node_stat_item. */ |
Johannes Weiner | a18e6e6 | 2021-04-29 22:56:17 -0700 | [diff] [blame] | 787 | static unsigned long memcg_page_state_local(struct mem_cgroup *memcg, int idx) |
| 788 | { |
| 789 | long x = 0; |
| 790 | int cpu; |
| 791 | |
| 792 | for_each_possible_cpu(cpu) |
Johannes Weiner | 2d146aa | 2021-04-29 22:56:26 -0700 | [diff] [blame] | 793 | x += per_cpu(memcg->vmstats_percpu->state[idx], cpu); |
Johannes Weiner | a18e6e6 | 2021-04-29 22:56:17 -0700 | [diff] [blame] | 794 | #ifdef CONFIG_SMP |
| 795 | if (x < 0) |
| 796 | x = 0; |
| 797 | #endif |
| 798 | return x; |
| 799 | } |
| 800 | |
Johannes Weiner | 42a3003 | 2019-05-14 15:47:12 -0700 | [diff] [blame] | 801 | static struct mem_cgroup_per_node * |
| 802 | parent_nodeinfo(struct mem_cgroup_per_node *pn, int nid) |
| 803 | { |
| 804 | struct mem_cgroup *parent; |
| 805 | |
| 806 | parent = parent_mem_cgroup(pn->memcg); |
| 807 | if (!parent) |
| 808 | return NULL; |
Johannes Weiner | a3747b5 | 2021-04-29 22:56:14 -0700 | [diff] [blame] | 809 | return parent->nodeinfo[nid]; |
Johannes Weiner | 42a3003 | 2019-05-14 15:47:12 -0700 | [diff] [blame] | 810 | } |
| 811 | |
Roman Gushchin | eedc4e5 | 2020-08-06 23:20:32 -0700 | [diff] [blame] | 812 | void __mod_memcg_lruvec_state(struct lruvec *lruvec, enum node_stat_item idx, |
| 813 | int val) |
| 814 | { |
| 815 | struct mem_cgroup_per_node *pn; |
| 816 | struct mem_cgroup *memcg; |
Roman Gushchin | ea426c2 | 2020-08-06 23:20:35 -0700 | [diff] [blame] | 817 | long x, threshold = MEMCG_CHARGE_BATCH; |
Roman Gushchin | eedc4e5 | 2020-08-06 23:20:32 -0700 | [diff] [blame] | 818 | |
| 819 | pn = container_of(lruvec, struct mem_cgroup_per_node, lruvec); |
| 820 | memcg = pn->memcg; |
| 821 | |
| 822 | /* Update memcg */ |
| 823 | __mod_memcg_state(memcg, idx, val); |
| 824 | |
| 825 | /* Update lruvec */ |
| 826 | __this_cpu_add(pn->lruvec_stat_local->count[idx], val); |
| 827 | |
Roman Gushchin | ea426c2 | 2020-08-06 23:20:35 -0700 | [diff] [blame] | 828 | if (vmstat_item_in_bytes(idx)) |
| 829 | threshold <<= PAGE_SHIFT; |
| 830 | |
Roman Gushchin | eedc4e5 | 2020-08-06 23:20:32 -0700 | [diff] [blame] | 831 | x = val + __this_cpu_read(pn->lruvec_stat_cpu->count[idx]); |
Roman Gushchin | ea426c2 | 2020-08-06 23:20:35 -0700 | [diff] [blame] | 832 | if (unlikely(abs(x) > threshold)) { |
Roman Gushchin | eedc4e5 | 2020-08-06 23:20:32 -0700 | [diff] [blame] | 833 | pg_data_t *pgdat = lruvec_pgdat(lruvec); |
| 834 | struct mem_cgroup_per_node *pi; |
| 835 | |
| 836 | for (pi = pn; pi; pi = parent_nodeinfo(pi, pgdat->node_id)) |
| 837 | atomic_long_add(x, &pi->lruvec_stat[idx]); |
| 838 | x = 0; |
| 839 | } |
| 840 | __this_cpu_write(pn->lruvec_stat_cpu->count[idx], x); |
| 841 | } |
| 842 | |
Johannes Weiner | db9adbc | 2019-05-14 15:47:09 -0700 | [diff] [blame] | 843 | /** |
| 844 | * __mod_lruvec_state - update lruvec memory statistics |
| 845 | * @lruvec: the lruvec |
| 846 | * @idx: the stat item |
| 847 | * @val: delta to add to the counter, can be negative |
| 848 | * |
| 849 | * The lruvec is the intersection of the NUMA node and a cgroup. This |
| 850 | * function updates the all three counters that are affected by a |
| 851 | * change of state at this level: per-node, per-cgroup, per-lruvec. |
| 852 | */ |
| 853 | void __mod_lruvec_state(struct lruvec *lruvec, enum node_stat_item idx, |
| 854 | int val) |
| 855 | { |
Johannes Weiner | db9adbc | 2019-05-14 15:47:09 -0700 | [diff] [blame] | 856 | /* Update node */ |
Roman Gushchin | eedc4e5 | 2020-08-06 23:20:32 -0700 | [diff] [blame] | 857 | __mod_node_page_state(lruvec_pgdat(lruvec), idx, val); |
Johannes Weiner | db9adbc | 2019-05-14 15:47:09 -0700 | [diff] [blame] | 858 | |
Roman Gushchin | eedc4e5 | 2020-08-06 23:20:32 -0700 | [diff] [blame] | 859 | /* Update memcg and lruvec */ |
| 860 | if (!mem_cgroup_disabled()) |
| 861 | __mod_memcg_lruvec_state(lruvec, idx, val); |
Johannes Weiner | db9adbc | 2019-05-14 15:47:09 -0700 | [diff] [blame] | 862 | } |
| 863 | |
Shakeel Butt | c47d503 | 2020-12-14 19:07:14 -0800 | [diff] [blame] | 864 | void __mod_lruvec_page_state(struct page *page, enum node_stat_item idx, |
| 865 | int val) |
| 866 | { |
| 867 | struct page *head = compound_head(page); /* rmap on tail pages */ |
Linus Torvalds | d635a69 | 2020-12-15 13:22:29 -0800 | [diff] [blame] | 868 | struct mem_cgroup *memcg = page_memcg(head); |
Shakeel Butt | c47d503 | 2020-12-14 19:07:14 -0800 | [diff] [blame] | 869 | pg_data_t *pgdat = page_pgdat(page); |
| 870 | struct lruvec *lruvec; |
| 871 | |
| 872 | /* Untracked pages have no memcg, no lruvec. Update only the node */ |
Linus Torvalds | d635a69 | 2020-12-15 13:22:29 -0800 | [diff] [blame] | 873 | if (!memcg) { |
Shakeel Butt | c47d503 | 2020-12-14 19:07:14 -0800 | [diff] [blame] | 874 | __mod_node_page_state(pgdat, idx, val); |
| 875 | return; |
| 876 | } |
| 877 | |
Linus Torvalds | d635a69 | 2020-12-15 13:22:29 -0800 | [diff] [blame] | 878 | lruvec = mem_cgroup_lruvec(memcg, pgdat); |
Shakeel Butt | c47d503 | 2020-12-14 19:07:14 -0800 | [diff] [blame] | 879 | __mod_lruvec_state(lruvec, idx, val); |
| 880 | } |
Shakeel Butt | f0c0c11 | 2020-12-14 19:07:17 -0800 | [diff] [blame] | 881 | EXPORT_SYMBOL(__mod_lruvec_page_state); |
Shakeel Butt | c47d503 | 2020-12-14 19:07:14 -0800 | [diff] [blame] | 882 | |
Muchun Song | da3ceef | 2020-12-14 19:07:04 -0800 | [diff] [blame] | 883 | void __mod_lruvec_kmem_state(void *p, enum node_stat_item idx, int val) |
Roman Gushchin | ec9f023 | 2019-08-13 15:37:41 -0700 | [diff] [blame] | 884 | { |
Roman Gushchin | 4f103c6 | 2020-04-01 21:06:36 -0700 | [diff] [blame] | 885 | pg_data_t *pgdat = page_pgdat(virt_to_page(p)); |
Roman Gushchin | ec9f023 | 2019-08-13 15:37:41 -0700 | [diff] [blame] | 886 | struct mem_cgroup *memcg; |
| 887 | struct lruvec *lruvec; |
| 888 | |
| 889 | rcu_read_lock(); |
Roman Gushchin | 4f103c6 | 2020-04-01 21:06:36 -0700 | [diff] [blame] | 890 | memcg = mem_cgroup_from_obj(p); |
Roman Gushchin | ec9f023 | 2019-08-13 15:37:41 -0700 | [diff] [blame] | 891 | |
Muchun Song | 8faeb1f | 2020-11-21 22:17:12 -0800 | [diff] [blame] | 892 | /* |
| 893 | * Untracked pages have no memcg, no lruvec. Update only the |
| 894 | * node. If we reparent the slab objects to the root memcg, |
| 895 | * when we free the slab object, we need to update the per-memcg |
| 896 | * vmstats to keep it correct for the root memcg. |
| 897 | */ |
| 898 | if (!memcg) { |
Roman Gushchin | ec9f023 | 2019-08-13 15:37:41 -0700 | [diff] [blame] | 899 | __mod_node_page_state(pgdat, idx, val); |
| 900 | } else { |
Johannes Weiner | 867e5e1 | 2019-11-30 17:55:34 -0800 | [diff] [blame] | 901 | lruvec = mem_cgroup_lruvec(memcg, pgdat); |
Roman Gushchin | ec9f023 | 2019-08-13 15:37:41 -0700 | [diff] [blame] | 902 | __mod_lruvec_state(lruvec, idx, val); |
| 903 | } |
| 904 | rcu_read_unlock(); |
| 905 | } |
| 906 | |
Johannes Weiner | db9adbc | 2019-05-14 15:47:09 -0700 | [diff] [blame] | 907 | /** |
| 908 | * __count_memcg_events - account VM events in a cgroup |
| 909 | * @memcg: the memory cgroup |
| 910 | * @idx: the event item |
| 911 | * @count: the number of events that occured |
| 912 | */ |
| 913 | void __count_memcg_events(struct mem_cgroup *memcg, enum vm_event_item idx, |
| 914 | unsigned long count) |
| 915 | { |
Johannes Weiner | db9adbc | 2019-05-14 15:47:09 -0700 | [diff] [blame] | 916 | if (mem_cgroup_disabled()) |
| 917 | return; |
| 918 | |
Johannes Weiner | 2d146aa | 2021-04-29 22:56:26 -0700 | [diff] [blame] | 919 | __this_cpu_add(memcg->vmstats_percpu->events[idx], count); |
| 920 | cgroup_rstat_updated(memcg->css.cgroup, smp_processor_id()); |
Johannes Weiner | db9adbc | 2019-05-14 15:47:09 -0700 | [diff] [blame] | 921 | } |
| 922 | |
Johannes Weiner | 42a3003 | 2019-05-14 15:47:12 -0700 | [diff] [blame] | 923 | static unsigned long memcg_events(struct mem_cgroup *memcg, int event) |
Johannes Weiner | e9f8974 | 2011-03-23 16:42:37 -0700 | [diff] [blame] | 924 | { |
Johannes Weiner | 2d146aa | 2021-04-29 22:56:26 -0700 | [diff] [blame] | 925 | return READ_ONCE(memcg->vmstats.events[event]); |
Johannes Weiner | e9f8974 | 2011-03-23 16:42:37 -0700 | [diff] [blame] | 926 | } |
| 927 | |
Johannes Weiner | 42a3003 | 2019-05-14 15:47:12 -0700 | [diff] [blame] | 928 | static unsigned long memcg_events_local(struct mem_cgroup *memcg, int event) |
| 929 | { |
Johannes Weiner | 815744d | 2019-06-13 15:55:46 -0700 | [diff] [blame] | 930 | long x = 0; |
| 931 | int cpu; |
| 932 | |
| 933 | for_each_possible_cpu(cpu) |
Johannes Weiner | 2d146aa | 2021-04-29 22:56:26 -0700 | [diff] [blame] | 934 | x += per_cpu(memcg->vmstats_percpu->events[event], cpu); |
Johannes Weiner | 815744d | 2019-06-13 15:55:46 -0700 | [diff] [blame] | 935 | return x; |
Johannes Weiner | 42a3003 | 2019-05-14 15:47:12 -0700 | [diff] [blame] | 936 | } |
| 937 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 938 | static void mem_cgroup_charge_statistics(struct mem_cgroup *memcg, |
David Rientjes | b070e65 | 2013-05-07 16:18:09 -0700 | [diff] [blame] | 939 | struct page *page, |
Johannes Weiner | 3fba69a | 2020-06-03 16:01:31 -0700 | [diff] [blame] | 940 | int nr_pages) |
KAMEZAWA Hiroyuki | d52aa41 | 2008-02-07 00:14:24 -0800 | [diff] [blame] | 941 | { |
KAMEZAWA Hiroyuki | e401f17 | 2011-01-20 14:44:23 -0800 | [diff] [blame] | 942 | /* pagein of a big page is an event. So, ignore page size */ |
| 943 | if (nr_pages > 0) |
Johannes Weiner | c9019e9 | 2018-01-31 16:16:37 -0800 | [diff] [blame] | 944 | __count_memcg_events(memcg, PGPGIN, 1); |
KAMEZAWA Hiroyuki | 3751d60 | 2011-02-01 15:52:45 -0800 | [diff] [blame] | 945 | else { |
Johannes Weiner | c9019e9 | 2018-01-31 16:16:37 -0800 | [diff] [blame] | 946 | __count_memcg_events(memcg, PGPGOUT, 1); |
KAMEZAWA Hiroyuki | 3751d60 | 2011-02-01 15:52:45 -0800 | [diff] [blame] | 947 | nr_pages = -nr_pages; /* for event */ |
| 948 | } |
KAMEZAWA Hiroyuki | e401f17 | 2011-01-20 14:44:23 -0800 | [diff] [blame] | 949 | |
Chris Down | 871789d | 2019-05-14 15:46:57 -0700 | [diff] [blame] | 950 | __this_cpu_add(memcg->vmstats_percpu->nr_page_events, nr_pages); |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 951 | } |
KAMEZAWA Hiroyuki | d52aa41 | 2008-02-07 00:14:24 -0800 | [diff] [blame] | 952 | |
Johannes Weiner | f53d7ce | 2012-01-12 17:18:23 -0800 | [diff] [blame] | 953 | static bool mem_cgroup_event_ratelimit(struct mem_cgroup *memcg, |
| 954 | enum mem_cgroup_events_target target) |
KAMEZAWA Hiroyuki | d2265e6 | 2010-03-10 15:22:31 -0800 | [diff] [blame] | 955 | { |
Johannes Weiner | 7a159cc | 2011-03-23 16:42:38 -0700 | [diff] [blame] | 956 | unsigned long val, next; |
| 957 | |
Chris Down | 871789d | 2019-05-14 15:46:57 -0700 | [diff] [blame] | 958 | val = __this_cpu_read(memcg->vmstats_percpu->nr_page_events); |
| 959 | next = __this_cpu_read(memcg->vmstats_percpu->targets[target]); |
Johannes Weiner | 7a159cc | 2011-03-23 16:42:38 -0700 | [diff] [blame] | 960 | /* from time_after() in jiffies.h */ |
Michal Hocko | 6a1a8b8 | 2017-07-10 15:48:53 -0700 | [diff] [blame] | 961 | if ((long)(next - val) < 0) { |
Johannes Weiner | f53d7ce | 2012-01-12 17:18:23 -0800 | [diff] [blame] | 962 | switch (target) { |
| 963 | case MEM_CGROUP_TARGET_THRESH: |
| 964 | next = val + THRESHOLDS_EVENTS_TARGET; |
| 965 | break; |
Andrew Morton | bb4cc1a8 | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 966 | case MEM_CGROUP_TARGET_SOFTLIMIT: |
| 967 | next = val + SOFTLIMIT_EVENTS_TARGET; |
| 968 | break; |
Johannes Weiner | f53d7ce | 2012-01-12 17:18:23 -0800 | [diff] [blame] | 969 | default: |
| 970 | break; |
| 971 | } |
Chris Down | 871789d | 2019-05-14 15:46:57 -0700 | [diff] [blame] | 972 | __this_cpu_write(memcg->vmstats_percpu->targets[target], next); |
Johannes Weiner | f53d7ce | 2012-01-12 17:18:23 -0800 | [diff] [blame] | 973 | return true; |
Johannes Weiner | 7a159cc | 2011-03-23 16:42:38 -0700 | [diff] [blame] | 974 | } |
Johannes Weiner | f53d7ce | 2012-01-12 17:18:23 -0800 | [diff] [blame] | 975 | return false; |
KAMEZAWA Hiroyuki | d2265e6 | 2010-03-10 15:22:31 -0800 | [diff] [blame] | 976 | } |
| 977 | |
| 978 | /* |
| 979 | * Check events in order. |
| 980 | * |
| 981 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 982 | static void memcg_check_events(struct mem_cgroup *memcg, struct page *page) |
KAMEZAWA Hiroyuki | d2265e6 | 2010-03-10 15:22:31 -0800 | [diff] [blame] | 983 | { |
| 984 | /* threshold event is triggered in finer grain than soft limit */ |
Johannes Weiner | f53d7ce | 2012-01-12 17:18:23 -0800 | [diff] [blame] | 985 | if (unlikely(mem_cgroup_event_ratelimit(memcg, |
| 986 | MEM_CGROUP_TARGET_THRESH))) { |
Andrew Morton | bb4cc1a8 | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 987 | bool do_softlimit; |
Johannes Weiner | f53d7ce | 2012-01-12 17:18:23 -0800 | [diff] [blame] | 988 | |
Andrew Morton | bb4cc1a8 | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 989 | do_softlimit = mem_cgroup_event_ratelimit(memcg, |
| 990 | MEM_CGROUP_TARGET_SOFTLIMIT); |
Johannes Weiner | f53d7ce | 2012-01-12 17:18:23 -0800 | [diff] [blame] | 991 | mem_cgroup_threshold(memcg); |
Andrew Morton | bb4cc1a8 | 2013-09-24 15:27:40 -0700 | [diff] [blame] | 992 | if (unlikely(do_softlimit)) |
| 993 | mem_cgroup_update_tree(memcg, page); |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 994 | } |
KAMEZAWA Hiroyuki | d2265e6 | 2010-03-10 15:22:31 -0800 | [diff] [blame] | 995 | } |
| 996 | |
Balbir Singh | cf475ad | 2008-04-29 01:00:16 -0700 | [diff] [blame] | 997 | struct mem_cgroup *mem_cgroup_from_task(struct task_struct *p) |
Pavel Emelianov | 78fb746 | 2008-02-07 00:13:51 -0800 | [diff] [blame] | 998 | { |
Balbir Singh | 31a78f2 | 2008-09-28 23:09:31 +0100 | [diff] [blame] | 999 | /* |
| 1000 | * mm_update_next_owner() may clear mm->owner to NULL |
| 1001 | * if it races with swapoff, page migration, etc. |
| 1002 | * So this can be called with p == NULL. |
| 1003 | */ |
| 1004 | if (unlikely(!p)) |
| 1005 | return NULL; |
| 1006 | |
Tejun Heo | 073219e | 2014-02-08 10:36:58 -0500 | [diff] [blame] | 1007 | return mem_cgroup_from_css(task_css(p, memory_cgrp_id)); |
Pavel Emelianov | 78fb746 | 2008-02-07 00:13:51 -0800 | [diff] [blame] | 1008 | } |
Michal Hocko | 33398cf | 2015-09-08 15:01:02 -0700 | [diff] [blame] | 1009 | EXPORT_SYMBOL(mem_cgroup_from_task); |
Pavel Emelianov | 78fb746 | 2008-02-07 00:13:51 -0800 | [diff] [blame] | 1010 | |
Shakeel Butt | d46eb14b | 2018-08-17 15:46:39 -0700 | [diff] [blame] | 1011 | /** |
| 1012 | * get_mem_cgroup_from_mm: Obtain a reference on given mm_struct's memcg. |
| 1013 | * @mm: mm from which memcg should be extracted. It can be NULL. |
| 1014 | * |
| 1015 | * Obtain a reference on mm->memcg and returns it if successful. Otherwise |
| 1016 | * root_mem_cgroup is returned. However if mem_cgroup is disabled, NULL is |
| 1017 | * returned. |
| 1018 | */ |
| 1019 | struct mem_cgroup *get_mem_cgroup_from_mm(struct mm_struct *mm) |
KAMEZAWA Hiroyuki | 54595fe | 2009-01-07 18:08:33 -0800 | [diff] [blame] | 1020 | { |
Shakeel Butt | d46eb14b | 2018-08-17 15:46:39 -0700 | [diff] [blame] | 1021 | struct mem_cgroup *memcg; |
| 1022 | |
| 1023 | if (mem_cgroup_disabled()) |
| 1024 | return NULL; |
KAMEZAWA Hiroyuki | 0b7f569 | 2009-04-02 16:57:38 -0700 | [diff] [blame] | 1025 | |
KAMEZAWA Hiroyuki | 54595fe | 2009-01-07 18:08:33 -0800 | [diff] [blame] | 1026 | rcu_read_lock(); |
| 1027 | do { |
Michal Hocko | 6f6acb0 | 2014-05-22 11:54:19 -0700 | [diff] [blame] | 1028 | /* |
| 1029 | * Page cache insertions can happen withou an |
| 1030 | * actual mm context, e.g. during disk probing |
| 1031 | * on boot, loopback IO, acct() writes etc. |
| 1032 | */ |
| 1033 | if (unlikely(!mm)) |
Johannes Weiner | df38197 | 2014-04-07 15:37:43 -0700 | [diff] [blame] | 1034 | memcg = root_mem_cgroup; |
Michal Hocko | 6f6acb0 | 2014-05-22 11:54:19 -0700 | [diff] [blame] | 1035 | else { |
| 1036 | memcg = mem_cgroup_from_task(rcu_dereference(mm->owner)); |
| 1037 | if (unlikely(!memcg)) |
| 1038 | memcg = root_mem_cgroup; |
| 1039 | } |
Roman Gushchin | 00d484f | 2019-11-15 17:34:43 -0800 | [diff] [blame] | 1040 | } while (!css_tryget(&memcg->css)); |
KAMEZAWA Hiroyuki | 54595fe | 2009-01-07 18:08:33 -0800 | [diff] [blame] | 1041 | rcu_read_unlock(); |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1042 | return memcg; |
KAMEZAWA Hiroyuki | 54595fe | 2009-01-07 18:08:33 -0800 | [diff] [blame] | 1043 | } |
Shakeel Butt | d46eb14b | 2018-08-17 15:46:39 -0700 | [diff] [blame] | 1044 | EXPORT_SYMBOL(get_mem_cgroup_from_mm); |
| 1045 | |
Roman Gushchin | 37d5985 | 2020-10-17 16:13:50 -0700 | [diff] [blame] | 1046 | static __always_inline struct mem_cgroup *active_memcg(void) |
| 1047 | { |
| 1048 | if (in_interrupt()) |
| 1049 | return this_cpu_read(int_active_memcg); |
| 1050 | else |
| 1051 | return current->active_memcg; |
| 1052 | } |
| 1053 | |
| 1054 | static __always_inline struct mem_cgroup *get_active_memcg(void) |
| 1055 | { |
| 1056 | struct mem_cgroup *memcg; |
| 1057 | |
| 1058 | rcu_read_lock(); |
| 1059 | memcg = active_memcg(); |
Muchun Song | 1685bde | 2021-02-24 12:04:22 -0800 | [diff] [blame] | 1060 | /* remote memcg must hold a ref. */ |
| 1061 | if (memcg && WARN_ON_ONCE(!css_tryget(&memcg->css))) |
| 1062 | memcg = root_mem_cgroup; |
Roman Gushchin | 37d5985 | 2020-10-17 16:13:50 -0700 | [diff] [blame] | 1063 | rcu_read_unlock(); |
| 1064 | |
| 1065 | return memcg; |
| 1066 | } |
| 1067 | |
Roman Gushchin | 4127c65 | 2020-10-17 16:13:53 -0700 | [diff] [blame] | 1068 | static __always_inline bool memcg_kmem_bypass(void) |
| 1069 | { |
| 1070 | /* Allow remote memcg charging from any context. */ |
| 1071 | if (unlikely(active_memcg())) |
| 1072 | return false; |
| 1073 | |
| 1074 | /* Memcg to charge can't be determined. */ |
| 1075 | if (in_interrupt() || !current->mm || (current->flags & PF_KTHREAD)) |
| 1076 | return true; |
| 1077 | |
| 1078 | return false; |
| 1079 | } |
| 1080 | |
Shakeel Butt | f745c6f | 2018-08-17 15:46:44 -0700 | [diff] [blame] | 1081 | /** |
Roman Gushchin | 37d5985 | 2020-10-17 16:13:50 -0700 | [diff] [blame] | 1082 | * If active memcg is set, do not fallback to current->mm->memcg. |
Shakeel Butt | d46eb14b | 2018-08-17 15:46:39 -0700 | [diff] [blame] | 1083 | */ |
| 1084 | static __always_inline struct mem_cgroup *get_mem_cgroup_from_current(void) |
| 1085 | { |
Roman Gushchin | 279c339 | 2020-10-17 16:13:44 -0700 | [diff] [blame] | 1086 | if (memcg_kmem_bypass()) |
| 1087 | return NULL; |
| 1088 | |
Roman Gushchin | 37d5985 | 2020-10-17 16:13:50 -0700 | [diff] [blame] | 1089 | if (unlikely(active_memcg())) |
| 1090 | return get_active_memcg(); |
Shakeel Butt | d46eb14b | 2018-08-17 15:46:39 -0700 | [diff] [blame] | 1091 | |
Shakeel Butt | d46eb14b | 2018-08-17 15:46:39 -0700 | [diff] [blame] | 1092 | return get_mem_cgroup_from_mm(current->mm); |
| 1093 | } |
KAMEZAWA Hiroyuki | 54595fe | 2009-01-07 18:08:33 -0800 | [diff] [blame] | 1094 | |
Johannes Weiner | 5660048 | 2012-01-12 17:17:59 -0800 | [diff] [blame] | 1095 | /** |
| 1096 | * mem_cgroup_iter - iterate over memory cgroup hierarchy |
| 1097 | * @root: hierarchy root |
| 1098 | * @prev: previously returned memcg, NULL on first invocation |
| 1099 | * @reclaim: cookie for shared reclaim walks, NULL for full walks |
| 1100 | * |
| 1101 | * Returns references to children of the hierarchy below @root, or |
| 1102 | * @root itself, or %NULL after a full round-trip. |
| 1103 | * |
| 1104 | * Caller must pass the return value in @prev on subsequent |
| 1105 | * invocations for reference counting, or use mem_cgroup_iter_break() |
| 1106 | * to cancel a hierarchy walk before the round-trip is complete. |
| 1107 | * |
Miaohe Lin | 05bdc52 | 2020-10-13 16:52:45 -0700 | [diff] [blame] | 1108 | * Reclaimers can specify a node in @reclaim to divide up the memcgs |
| 1109 | * in the hierarchy among all concurrent reclaimers operating on the |
| 1110 | * same node. |
Johannes Weiner | 5660048 | 2012-01-12 17:17:59 -0800 | [diff] [blame] | 1111 | */ |
Andrew Morton | 694fbc0 | 2013-09-24 15:27:37 -0700 | [diff] [blame] | 1112 | struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root, |
Johannes Weiner | 5660048 | 2012-01-12 17:17:59 -0800 | [diff] [blame] | 1113 | struct mem_cgroup *prev, |
Andrew Morton | 694fbc0 | 2013-09-24 15:27:37 -0700 | [diff] [blame] | 1114 | struct mem_cgroup_reclaim_cookie *reclaim) |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 1115 | { |
Kees Cook | 3f649ab | 2020-06-03 13:09:38 -0700 | [diff] [blame] | 1116 | struct mem_cgroup_reclaim_iter *iter; |
Johannes Weiner | 5ac8fb3 | 2014-12-10 15:42:39 -0800 | [diff] [blame] | 1117 | struct cgroup_subsys_state *css = NULL; |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 1118 | struct mem_cgroup *memcg = NULL; |
Johannes Weiner | 5ac8fb3 | 2014-12-10 15:42:39 -0800 | [diff] [blame] | 1119 | struct mem_cgroup *pos = NULL; |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 1120 | |
Andrew Morton | 694fbc0 | 2013-09-24 15:27:37 -0700 | [diff] [blame] | 1121 | if (mem_cgroup_disabled()) |
| 1122 | return NULL; |
Johannes Weiner | 5660048 | 2012-01-12 17:17:59 -0800 | [diff] [blame] | 1123 | |
KAMEZAWA Hiroyuki | 711d3d2 | 2010-10-27 15:33:42 -0700 | [diff] [blame] | 1124 | if (!root) |
| 1125 | root = root_mem_cgroup; |
| 1126 | |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 1127 | if (prev && !reclaim) |
Johannes Weiner | 5ac8fb3 | 2014-12-10 15:42:39 -0800 | [diff] [blame] | 1128 | pos = prev; |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 1129 | |
Michal Hocko | 542f85f | 2013-04-29 15:07:15 -0700 | [diff] [blame] | 1130 | rcu_read_lock(); |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 1131 | |
Johannes Weiner | 5ac8fb3 | 2014-12-10 15:42:39 -0800 | [diff] [blame] | 1132 | if (reclaim) { |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 1133 | struct mem_cgroup_per_node *mz; |
Johannes Weiner | 527a5ec | 2012-01-12 17:17:55 -0800 | [diff] [blame] | 1134 | |
Johannes Weiner | a3747b5 | 2021-04-29 22:56:14 -0700 | [diff] [blame] | 1135 | mz = root->nodeinfo[reclaim->pgdat->node_id]; |
Yafang Shao | 9da83f3 | 2019-11-30 17:50:03 -0800 | [diff] [blame] | 1136 | iter = &mz->iter; |
Michal Hocko | 5f57816 | 2013-04-29 15:07:17 -0700 | [diff] [blame] | 1137 | |
Johannes Weiner | 5ac8fb3 | 2014-12-10 15:42:39 -0800 | [diff] [blame] | 1138 | if (prev && reclaim->generation != iter->generation) |
Michal Hocko | 542f85f | 2013-04-29 15:07:15 -0700 | [diff] [blame] | 1139 | goto out_unlock; |
Johannes Weiner | 5ac8fb3 | 2014-12-10 15:42:39 -0800 | [diff] [blame] | 1140 | |
Vladimir Davydov | 6df3868 | 2015-12-29 14:54:10 -0800 | [diff] [blame] | 1141 | while (1) { |
Jason Low | 4db0c3c | 2015-04-15 16:14:08 -0700 | [diff] [blame] | 1142 | pos = READ_ONCE(iter->position); |
Vladimir Davydov | 6df3868 | 2015-12-29 14:54:10 -0800 | [diff] [blame] | 1143 | if (!pos || css_tryget(&pos->css)) |
| 1144 | break; |
Johannes Weiner | 5ac8fb3 | 2014-12-10 15:42:39 -0800 | [diff] [blame] | 1145 | /* |
Vladimir Davydov | 6df3868 | 2015-12-29 14:54:10 -0800 | [diff] [blame] | 1146 | * css reference reached zero, so iter->position will |
| 1147 | * be cleared by ->css_released. However, we should not |
| 1148 | * rely on this happening soon, because ->css_released |
| 1149 | * is called from a work queue, and by busy-waiting we |
| 1150 | * might block it. So we clear iter->position right |
| 1151 | * away. |
Johannes Weiner | 5ac8fb3 | 2014-12-10 15:42:39 -0800 | [diff] [blame] | 1152 | */ |
Vladimir Davydov | 6df3868 | 2015-12-29 14:54:10 -0800 | [diff] [blame] | 1153 | (void)cmpxchg(&iter->position, pos, NULL); |
| 1154 | } |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 1155 | } |
Johannes Weiner | 5ac8fb3 | 2014-12-10 15:42:39 -0800 | [diff] [blame] | 1156 | |
| 1157 | if (pos) |
| 1158 | css = &pos->css; |
| 1159 | |
| 1160 | for (;;) { |
| 1161 | css = css_next_descendant_pre(css, &root->css); |
| 1162 | if (!css) { |
| 1163 | /* |
| 1164 | * Reclaimers share the hierarchy walk, and a |
| 1165 | * new one might jump in right at the end of |
| 1166 | * the hierarchy - make sure they see at least |
| 1167 | * one group and restart from the beginning. |
| 1168 | */ |
| 1169 | if (!prev) |
| 1170 | continue; |
| 1171 | break; |
| 1172 | } |
| 1173 | |
| 1174 | /* |
| 1175 | * Verify the css and acquire a reference. The root |
| 1176 | * is provided by the caller, so we know it's alive |
| 1177 | * and kicking, and don't take an extra reference. |
| 1178 | */ |
| 1179 | memcg = mem_cgroup_from_css(css); |
| 1180 | |
| 1181 | if (css == &root->css) |
| 1182 | break; |
| 1183 | |
Johannes Weiner | 0b8f73e | 2016-01-20 15:02:53 -0800 | [diff] [blame] | 1184 | if (css_tryget(css)) |
| 1185 | break; |
Johannes Weiner | 5ac8fb3 | 2014-12-10 15:42:39 -0800 | [diff] [blame] | 1186 | |
| 1187 | memcg = NULL; |
| 1188 | } |
| 1189 | |
| 1190 | if (reclaim) { |
Johannes Weiner | 5ac8fb3 | 2014-12-10 15:42:39 -0800 | [diff] [blame] | 1191 | /* |
Vladimir Davydov | 6df3868 | 2015-12-29 14:54:10 -0800 | [diff] [blame] | 1192 | * The position could have already been updated by a competing |
| 1193 | * thread, so check that the value hasn't changed since we read |
| 1194 | * it to avoid reclaiming from the same cgroup twice. |
Johannes Weiner | 5ac8fb3 | 2014-12-10 15:42:39 -0800 | [diff] [blame] | 1195 | */ |
Vladimir Davydov | 6df3868 | 2015-12-29 14:54:10 -0800 | [diff] [blame] | 1196 | (void)cmpxchg(&iter->position, pos, memcg); |
| 1197 | |
Johannes Weiner | 5ac8fb3 | 2014-12-10 15:42:39 -0800 | [diff] [blame] | 1198 | if (pos) |
| 1199 | css_put(&pos->css); |
| 1200 | |
| 1201 | if (!memcg) |
| 1202 | iter->generation++; |
| 1203 | else if (!prev) |
| 1204 | reclaim->generation = iter->generation; |
| 1205 | } |
| 1206 | |
Michal Hocko | 542f85f | 2013-04-29 15:07:15 -0700 | [diff] [blame] | 1207 | out_unlock: |
| 1208 | rcu_read_unlock(); |
Michal Hocko | c40046f | 2013-04-29 15:07:14 -0700 | [diff] [blame] | 1209 | if (prev && prev != root) |
| 1210 | css_put(&prev->css); |
| 1211 | |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 1212 | return memcg; |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 1213 | } |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 1214 | |
Johannes Weiner | 5660048 | 2012-01-12 17:17:59 -0800 | [diff] [blame] | 1215 | /** |
| 1216 | * mem_cgroup_iter_break - abort a hierarchy walk prematurely |
| 1217 | * @root: hierarchy root |
| 1218 | * @prev: last visited hierarchy member as returned by mem_cgroup_iter() |
| 1219 | */ |
| 1220 | void mem_cgroup_iter_break(struct mem_cgroup *root, |
| 1221 | struct mem_cgroup *prev) |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 1222 | { |
| 1223 | if (!root) |
| 1224 | root = root_mem_cgroup; |
| 1225 | if (prev && prev != root) |
| 1226 | css_put(&prev->css); |
| 1227 | } |
| 1228 | |
Miles Chen | 54a83d6 | 2019-08-13 15:37:28 -0700 | [diff] [blame] | 1229 | static void __invalidate_reclaim_iterators(struct mem_cgroup *from, |
| 1230 | struct mem_cgroup *dead_memcg) |
Vladimir Davydov | 6df3868 | 2015-12-29 14:54:10 -0800 | [diff] [blame] | 1231 | { |
Vladimir Davydov | 6df3868 | 2015-12-29 14:54:10 -0800 | [diff] [blame] | 1232 | struct mem_cgroup_reclaim_iter *iter; |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 1233 | struct mem_cgroup_per_node *mz; |
| 1234 | int nid; |
Vladimir Davydov | 6df3868 | 2015-12-29 14:54:10 -0800 | [diff] [blame] | 1235 | |
Miles Chen | 54a83d6 | 2019-08-13 15:37:28 -0700 | [diff] [blame] | 1236 | for_each_node(nid) { |
Johannes Weiner | a3747b5 | 2021-04-29 22:56:14 -0700 | [diff] [blame] | 1237 | mz = from->nodeinfo[nid]; |
Yafang Shao | 9da83f3 | 2019-11-30 17:50:03 -0800 | [diff] [blame] | 1238 | iter = &mz->iter; |
| 1239 | cmpxchg(&iter->position, dead_memcg, NULL); |
Vladimir Davydov | 6df3868 | 2015-12-29 14:54:10 -0800 | [diff] [blame] | 1240 | } |
| 1241 | } |
| 1242 | |
Miles Chen | 54a83d6 | 2019-08-13 15:37:28 -0700 | [diff] [blame] | 1243 | static void invalidate_reclaim_iterators(struct mem_cgroup *dead_memcg) |
| 1244 | { |
| 1245 | struct mem_cgroup *memcg = dead_memcg; |
| 1246 | struct mem_cgroup *last; |
| 1247 | |
| 1248 | do { |
| 1249 | __invalidate_reclaim_iterators(memcg, dead_memcg); |
| 1250 | last = memcg; |
| 1251 | } while ((memcg = parent_mem_cgroup(memcg))); |
| 1252 | |
| 1253 | /* |
| 1254 | * When cgruop1 non-hierarchy mode is used, |
| 1255 | * parent_mem_cgroup() does not walk all the way up to the |
| 1256 | * cgroup root (root_mem_cgroup). So we have to handle |
| 1257 | * dead_memcg from cgroup root separately. |
| 1258 | */ |
| 1259 | if (last != root_mem_cgroup) |
| 1260 | __invalidate_reclaim_iterators(root_mem_cgroup, |
| 1261 | dead_memcg); |
| 1262 | } |
| 1263 | |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 1264 | /** |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 1265 | * mem_cgroup_scan_tasks - iterate over tasks of a memory cgroup hierarchy |
| 1266 | * @memcg: hierarchy root |
| 1267 | * @fn: function to call for each task |
| 1268 | * @arg: argument passed to @fn |
| 1269 | * |
| 1270 | * This function iterates over tasks attached to @memcg or to any of its |
| 1271 | * descendants and calls @fn for each task. If @fn returns a non-zero |
| 1272 | * value, the function breaks the iteration loop and returns the value. |
| 1273 | * Otherwise, it will iterate over all tasks and return 0. |
| 1274 | * |
| 1275 | * This function must not be called for the root memory cgroup. |
| 1276 | */ |
| 1277 | int mem_cgroup_scan_tasks(struct mem_cgroup *memcg, |
| 1278 | int (*fn)(struct task_struct *, void *), void *arg) |
| 1279 | { |
| 1280 | struct mem_cgroup *iter; |
| 1281 | int ret = 0; |
| 1282 | |
| 1283 | BUG_ON(memcg == root_mem_cgroup); |
| 1284 | |
| 1285 | for_each_mem_cgroup_tree(iter, memcg) { |
| 1286 | struct css_task_iter it; |
| 1287 | struct task_struct *task; |
| 1288 | |
Tetsuo Handa | f168a9a | 2019-07-11 21:00:20 -0700 | [diff] [blame] | 1289 | css_task_iter_start(&iter->css, CSS_TASK_ITER_PROCS, &it); |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 1290 | while (!ret && (task = css_task_iter_next(&it))) |
| 1291 | ret = fn(task, arg); |
| 1292 | css_task_iter_end(&it); |
| 1293 | if (ret) { |
| 1294 | mem_cgroup_iter_break(memcg, iter); |
| 1295 | break; |
| 1296 | } |
| 1297 | } |
| 1298 | return ret; |
| 1299 | } |
| 1300 | |
Alex Shi | 6168d0d | 2020-12-15 12:34:29 -0800 | [diff] [blame] | 1301 | #ifdef CONFIG_DEBUG_VM |
| 1302 | void lruvec_memcg_debug(struct lruvec *lruvec, struct page *page) |
| 1303 | { |
| 1304 | struct mem_cgroup *memcg; |
| 1305 | |
| 1306 | if (mem_cgroup_disabled()) |
| 1307 | return; |
| 1308 | |
| 1309 | memcg = page_memcg(page); |
| 1310 | |
| 1311 | if (!memcg) |
| 1312 | VM_BUG_ON_PAGE(lruvec_memcg(lruvec) != root_mem_cgroup, page); |
| 1313 | else |
| 1314 | VM_BUG_ON_PAGE(lruvec_memcg(lruvec) != memcg, page); |
| 1315 | } |
| 1316 | #endif |
| 1317 | |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 1318 | /** |
Alex Shi | 6168d0d | 2020-12-15 12:34:29 -0800 | [diff] [blame] | 1319 | * lock_page_lruvec - lock and return lruvec for a given page. |
| 1320 | * @page: the page |
| 1321 | * |
Alex Shi | d7e3aba | 2021-02-24 12:03:47 -0800 | [diff] [blame] | 1322 | * These functions are safe to use under any of the following conditions: |
| 1323 | * - page locked |
| 1324 | * - PageLRU cleared |
| 1325 | * - lock_page_memcg() |
| 1326 | * - page->_refcount is zero |
Alex Shi | 6168d0d | 2020-12-15 12:34:29 -0800 | [diff] [blame] | 1327 | */ |
| 1328 | struct lruvec *lock_page_lruvec(struct page *page) |
| 1329 | { |
| 1330 | struct lruvec *lruvec; |
| 1331 | struct pglist_data *pgdat = page_pgdat(page); |
| 1332 | |
Alex Shi | 6168d0d | 2020-12-15 12:34:29 -0800 | [diff] [blame] | 1333 | lruvec = mem_cgroup_page_lruvec(page, pgdat); |
| 1334 | spin_lock(&lruvec->lru_lock); |
Alex Shi | 6168d0d | 2020-12-15 12:34:29 -0800 | [diff] [blame] | 1335 | |
| 1336 | lruvec_memcg_debug(lruvec, page); |
| 1337 | |
| 1338 | return lruvec; |
| 1339 | } |
| 1340 | |
| 1341 | struct lruvec *lock_page_lruvec_irq(struct page *page) |
| 1342 | { |
| 1343 | struct lruvec *lruvec; |
| 1344 | struct pglist_data *pgdat = page_pgdat(page); |
| 1345 | |
Alex Shi | 6168d0d | 2020-12-15 12:34:29 -0800 | [diff] [blame] | 1346 | lruvec = mem_cgroup_page_lruvec(page, pgdat); |
| 1347 | spin_lock_irq(&lruvec->lru_lock); |
Alex Shi | 6168d0d | 2020-12-15 12:34:29 -0800 | [diff] [blame] | 1348 | |
| 1349 | lruvec_memcg_debug(lruvec, page); |
| 1350 | |
| 1351 | return lruvec; |
| 1352 | } |
| 1353 | |
| 1354 | struct lruvec *lock_page_lruvec_irqsave(struct page *page, unsigned long *flags) |
| 1355 | { |
| 1356 | struct lruvec *lruvec; |
| 1357 | struct pglist_data *pgdat = page_pgdat(page); |
| 1358 | |
Alex Shi | 6168d0d | 2020-12-15 12:34:29 -0800 | [diff] [blame] | 1359 | lruvec = mem_cgroup_page_lruvec(page, pgdat); |
| 1360 | spin_lock_irqsave(&lruvec->lru_lock, *flags); |
Alex Shi | 6168d0d | 2020-12-15 12:34:29 -0800 | [diff] [blame] | 1361 | |
| 1362 | lruvec_memcg_debug(lruvec, page); |
| 1363 | |
| 1364 | return lruvec; |
| 1365 | } |
| 1366 | |
| 1367 | /** |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 1368 | * mem_cgroup_update_lru_size - account for adding or removing an lru page |
| 1369 | * @lruvec: mem_cgroup per zone lru vector |
| 1370 | * @lru: index of lru list the page is sitting on |
Michal Hocko | b4536f0c8 | 2017-01-10 16:58:04 -0800 | [diff] [blame] | 1371 | * @zid: zone id of the accounted pages |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 1372 | * @nr_pages: positive when adding or negative when removing |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 1373 | * |
Hugh Dickins | ca70723 | 2016-05-19 17:12:35 -0700 | [diff] [blame] | 1374 | * This function must be called under lru_lock, just before a page is added |
| 1375 | * to or just after a page is removed from an lru list (that ordering being |
| 1376 | * 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] | 1377 | */ |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 1378 | 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] | 1379 | int zid, int nr_pages) |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 1380 | { |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 1381 | struct mem_cgroup_per_node *mz; |
Hugh Dickins | fa9add6 | 2012-05-29 15:07:09 -0700 | [diff] [blame] | 1382 | unsigned long *lru_size; |
Hugh Dickins | ca70723 | 2016-05-19 17:12:35 -0700 | [diff] [blame] | 1383 | long size; |
Johannes Weiner | 925b767 | 2012-01-12 17:18:15 -0800 | [diff] [blame] | 1384 | |
| 1385 | if (mem_cgroup_disabled()) |
| 1386 | return; |
| 1387 | |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 1388 | mz = container_of(lruvec, struct mem_cgroup_per_node, lruvec); |
Michal Hocko | b4536f0c8 | 2017-01-10 16:58:04 -0800 | [diff] [blame] | 1389 | lru_size = &mz->lru_zone_size[zid][lru]; |
Hugh Dickins | ca70723 | 2016-05-19 17:12:35 -0700 | [diff] [blame] | 1390 | |
| 1391 | if (nr_pages < 0) |
| 1392 | *lru_size += nr_pages; |
| 1393 | |
| 1394 | size = *lru_size; |
Michal Hocko | b4536f0c8 | 2017-01-10 16:58:04 -0800 | [diff] [blame] | 1395 | if (WARN_ONCE(size < 0, |
| 1396 | "%s(%p, %d, %d): lru_size %ld\n", |
| 1397 | __func__, lruvec, lru, nr_pages, size)) { |
Hugh Dickins | ca70723 | 2016-05-19 17:12:35 -0700 | [diff] [blame] | 1398 | VM_BUG_ON(1); |
| 1399 | *lru_size = 0; |
| 1400 | } |
| 1401 | |
| 1402 | if (nr_pages > 0) |
| 1403 | *lru_size += nr_pages; |
KAMEZAWA Hiroyuki | 08e552c | 2009-01-07 18:08:01 -0800 | [diff] [blame] | 1404 | } |
KAMEZAWA Hiroyuki | 544122e | 2009-01-07 18:08:34 -0800 | [diff] [blame] | 1405 | |
Johannes Weiner | 1994282 | 2011-02-01 15:52:43 -0800 | [diff] [blame] | 1406 | /** |
Johannes Weiner | 9d11ea9 | 2011-03-23 16:42:21 -0700 | [diff] [blame] | 1407 | * mem_cgroup_margin - calculate chargeable space of a memory cgroup |
Wanpeng Li | dad7557 | 2012-06-20 12:53:01 -0700 | [diff] [blame] | 1408 | * @memcg: the memory cgroup |
Johannes Weiner | 1994282 | 2011-02-01 15:52:43 -0800 | [diff] [blame] | 1409 | * |
Johannes Weiner | 9d11ea9 | 2011-03-23 16:42:21 -0700 | [diff] [blame] | 1410 | * Returns the maximum amount of memory @mem can be charged with, in |
Johannes Weiner | 7ec99d6 | 2011-03-23 16:42:36 -0700 | [diff] [blame] | 1411 | * pages. |
Johannes Weiner | 1994282 | 2011-02-01 15:52:43 -0800 | [diff] [blame] | 1412 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1413 | static unsigned long mem_cgroup_margin(struct mem_cgroup *memcg) |
Johannes Weiner | 1994282 | 2011-02-01 15:52:43 -0800 | [diff] [blame] | 1414 | { |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1415 | unsigned long margin = 0; |
| 1416 | unsigned long count; |
| 1417 | unsigned long limit; |
Johannes Weiner | 9d11ea9 | 2011-03-23 16:42:21 -0700 | [diff] [blame] | 1418 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1419 | count = page_counter_read(&memcg->memory); |
Roman Gushchin | bbec2e1 | 2018-06-07 17:06:18 -0700 | [diff] [blame] | 1420 | limit = READ_ONCE(memcg->memory.max); |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1421 | if (count < limit) |
| 1422 | margin = limit - count; |
| 1423 | |
Johannes Weiner | 7941d21 | 2016-01-14 15:21:23 -0800 | [diff] [blame] | 1424 | if (do_memsw_account()) { |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1425 | count = page_counter_read(&memcg->memsw); |
Roman Gushchin | bbec2e1 | 2018-06-07 17:06:18 -0700 | [diff] [blame] | 1426 | limit = READ_ONCE(memcg->memsw.max); |
Kaixu Xia | 1c4448e | 2020-06-01 21:49:36 -0700 | [diff] [blame] | 1427 | if (count < limit) |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1428 | margin = min(margin, limit - count); |
Li RongQing | cbedbac | 2016-05-27 14:27:43 -0700 | [diff] [blame] | 1429 | else |
| 1430 | margin = 0; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1431 | } |
| 1432 | |
| 1433 | return margin; |
Johannes Weiner | 1994282 | 2011-02-01 15:52:43 -0800 | [diff] [blame] | 1434 | } |
| 1435 | |
KAMEZAWA Hiroyuki | 619d094 | 2012-03-21 16:34:23 -0700 | [diff] [blame] | 1436 | /* |
Qiang Huang | bdcbb65 | 2014-06-04 16:08:21 -0700 | [diff] [blame] | 1437 | * A routine for checking "mem" is under move_account() or not. |
KAMEZAWA Hiroyuki | 32047e2 | 2010-10-27 15:33:40 -0700 | [diff] [blame] | 1438 | * |
Qiang Huang | bdcbb65 | 2014-06-04 16:08:21 -0700 | [diff] [blame] | 1439 | * Checking a cgroup is mc.from or mc.to or under hierarchy of |
| 1440 | * moving cgroups. This is for waiting at high-memory pressure |
| 1441 | * caused by "move". |
KAMEZAWA Hiroyuki | 32047e2 | 2010-10-27 15:33:40 -0700 | [diff] [blame] | 1442 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1443 | static bool mem_cgroup_under_move(struct mem_cgroup *memcg) |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 1444 | { |
KAMEZAWA Hiroyuki | 2bd9bb2 | 2010-08-10 18:02:58 -0700 | [diff] [blame] | 1445 | struct mem_cgroup *from; |
| 1446 | struct mem_cgroup *to; |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 1447 | bool ret = false; |
KAMEZAWA Hiroyuki | 2bd9bb2 | 2010-08-10 18:02:58 -0700 | [diff] [blame] | 1448 | /* |
| 1449 | * Unlike task_move routines, we access mc.to, mc.from not under |
| 1450 | * mutual exclusion by cgroup_mutex. Here, we take spinlock instead. |
| 1451 | */ |
| 1452 | spin_lock(&mc.lock); |
| 1453 | from = mc.from; |
| 1454 | to = mc.to; |
| 1455 | if (!from) |
| 1456 | goto unlock; |
Michal Hocko | 3e92041 | 2011-07-26 16:08:29 -0700 | [diff] [blame] | 1457 | |
Johannes Weiner | 2314b42 | 2014-12-10 15:44:33 -0800 | [diff] [blame] | 1458 | ret = mem_cgroup_is_descendant(from, memcg) || |
| 1459 | mem_cgroup_is_descendant(to, memcg); |
KAMEZAWA Hiroyuki | 2bd9bb2 | 2010-08-10 18:02:58 -0700 | [diff] [blame] | 1460 | unlock: |
| 1461 | spin_unlock(&mc.lock); |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 1462 | return ret; |
| 1463 | } |
| 1464 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1465 | static bool mem_cgroup_wait_acct_move(struct mem_cgroup *memcg) |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 1466 | { |
| 1467 | if (mc.moving_task && current != mc.moving_task) { |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1468 | if (mem_cgroup_under_move(memcg)) { |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 1469 | DEFINE_WAIT(wait); |
| 1470 | prepare_to_wait(&mc.waitq, &wait, TASK_INTERRUPTIBLE); |
| 1471 | /* moving charge context might have finished. */ |
| 1472 | if (mc.moving_task) |
| 1473 | schedule(); |
| 1474 | finish_wait(&mc.waitq, &wait); |
| 1475 | return true; |
| 1476 | } |
| 1477 | } |
| 1478 | return false; |
| 1479 | } |
| 1480 | |
Muchun Song | 5f9a4f4 | 2020-10-13 16:52:59 -0700 | [diff] [blame] | 1481 | struct memory_stat { |
| 1482 | const char *name; |
Muchun Song | 5f9a4f4 | 2020-10-13 16:52:59 -0700 | [diff] [blame] | 1483 | unsigned int idx; |
| 1484 | }; |
| 1485 | |
Muchun Song | 57b2847 | 2021-02-24 12:03:31 -0800 | [diff] [blame] | 1486 | static const struct memory_stat memory_stats[] = { |
Muchun Song | fff66b7 | 2021-02-24 12:03:43 -0800 | [diff] [blame] | 1487 | { "anon", NR_ANON_MAPPED }, |
| 1488 | { "file", NR_FILE_PAGES }, |
| 1489 | { "kernel_stack", NR_KERNEL_STACK_KB }, |
| 1490 | { "pagetables", NR_PAGETABLE }, |
| 1491 | { "percpu", MEMCG_PERCPU_B }, |
| 1492 | { "sock", MEMCG_SOCK }, |
| 1493 | { "shmem", NR_SHMEM }, |
| 1494 | { "file_mapped", NR_FILE_MAPPED }, |
| 1495 | { "file_dirty", NR_FILE_DIRTY }, |
| 1496 | { "file_writeback", NR_WRITEBACK }, |
Shakeel Butt | b603894 | 2021-02-24 12:03:55 -0800 | [diff] [blame] | 1497 | #ifdef CONFIG_SWAP |
| 1498 | { "swapcached", NR_SWAPCACHE }, |
| 1499 | #endif |
Muchun Song | 5f9a4f4 | 2020-10-13 16:52:59 -0700 | [diff] [blame] | 1500 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
Muchun Song | fff66b7 | 2021-02-24 12:03:43 -0800 | [diff] [blame] | 1501 | { "anon_thp", NR_ANON_THPS }, |
| 1502 | { "file_thp", NR_FILE_THPS }, |
| 1503 | { "shmem_thp", NR_SHMEM_THPS }, |
Muchun Song | 5f9a4f4 | 2020-10-13 16:52:59 -0700 | [diff] [blame] | 1504 | #endif |
Muchun Song | fff66b7 | 2021-02-24 12:03:43 -0800 | [diff] [blame] | 1505 | { "inactive_anon", NR_INACTIVE_ANON }, |
| 1506 | { "active_anon", NR_ACTIVE_ANON }, |
| 1507 | { "inactive_file", NR_INACTIVE_FILE }, |
| 1508 | { "active_file", NR_ACTIVE_FILE }, |
| 1509 | { "unevictable", NR_UNEVICTABLE }, |
| 1510 | { "slab_reclaimable", NR_SLAB_RECLAIMABLE_B }, |
| 1511 | { "slab_unreclaimable", NR_SLAB_UNRECLAIMABLE_B }, |
Muchun Song | 5f9a4f4 | 2020-10-13 16:52:59 -0700 | [diff] [blame] | 1512 | |
| 1513 | /* The memory events */ |
Muchun Song | fff66b7 | 2021-02-24 12:03:43 -0800 | [diff] [blame] | 1514 | { "workingset_refault_anon", WORKINGSET_REFAULT_ANON }, |
| 1515 | { "workingset_refault_file", WORKINGSET_REFAULT_FILE }, |
| 1516 | { "workingset_activate_anon", WORKINGSET_ACTIVATE_ANON }, |
| 1517 | { "workingset_activate_file", WORKINGSET_ACTIVATE_FILE }, |
| 1518 | { "workingset_restore_anon", WORKINGSET_RESTORE_ANON }, |
| 1519 | { "workingset_restore_file", WORKINGSET_RESTORE_FILE }, |
| 1520 | { "workingset_nodereclaim", WORKINGSET_NODERECLAIM }, |
Muchun Song | 5f9a4f4 | 2020-10-13 16:52:59 -0700 | [diff] [blame] | 1521 | }; |
| 1522 | |
Muchun Song | fff66b7 | 2021-02-24 12:03:43 -0800 | [diff] [blame] | 1523 | /* Translate stat items to the correct unit for memory.stat output */ |
| 1524 | static int memcg_page_state_unit(int item) |
| 1525 | { |
| 1526 | switch (item) { |
| 1527 | case MEMCG_PERCPU_B: |
| 1528 | case NR_SLAB_RECLAIMABLE_B: |
| 1529 | case NR_SLAB_UNRECLAIMABLE_B: |
| 1530 | case WORKINGSET_REFAULT_ANON: |
| 1531 | case WORKINGSET_REFAULT_FILE: |
| 1532 | case WORKINGSET_ACTIVATE_ANON: |
| 1533 | case WORKINGSET_ACTIVATE_FILE: |
| 1534 | case WORKINGSET_RESTORE_ANON: |
| 1535 | case WORKINGSET_RESTORE_FILE: |
| 1536 | case WORKINGSET_NODERECLAIM: |
| 1537 | return 1; |
| 1538 | case NR_KERNEL_STACK_KB: |
| 1539 | return SZ_1K; |
| 1540 | default: |
| 1541 | return PAGE_SIZE; |
| 1542 | } |
| 1543 | } |
| 1544 | |
| 1545 | static inline unsigned long memcg_page_state_output(struct mem_cgroup *memcg, |
| 1546 | int item) |
| 1547 | { |
| 1548 | return memcg_page_state(memcg, item) * memcg_page_state_unit(item); |
| 1549 | } |
| 1550 | |
Johannes Weiner | c8713d0 | 2019-07-11 20:55:59 -0700 | [diff] [blame] | 1551 | static char *memory_stat_format(struct mem_cgroup *memcg) |
| 1552 | { |
| 1553 | struct seq_buf s; |
| 1554 | int i; |
Johannes Weiner | 71cd311 | 2017-05-03 14:55:13 -0700 | [diff] [blame] | 1555 | |
Johannes Weiner | c8713d0 | 2019-07-11 20:55:59 -0700 | [diff] [blame] | 1556 | seq_buf_init(&s, kmalloc(PAGE_SIZE, GFP_KERNEL), PAGE_SIZE); |
| 1557 | if (!s.buffer) |
| 1558 | return NULL; |
| 1559 | |
| 1560 | /* |
| 1561 | * Provide statistics on the state of the memory subsystem as |
| 1562 | * well as cumulative event counters that show past behavior. |
| 1563 | * |
| 1564 | * This list is ordered following a combination of these gradients: |
| 1565 | * 1) generic big picture -> specifics and details |
| 1566 | * 2) reflecting userspace activity -> reflecting kernel heuristics |
| 1567 | * |
| 1568 | * Current memory state: |
| 1569 | */ |
Johannes Weiner | 2d146aa | 2021-04-29 22:56:26 -0700 | [diff] [blame] | 1570 | cgroup_rstat_flush(memcg->css.cgroup); |
Johannes Weiner | c8713d0 | 2019-07-11 20:55:59 -0700 | [diff] [blame] | 1571 | |
Muchun Song | 5f9a4f4 | 2020-10-13 16:52:59 -0700 | [diff] [blame] | 1572 | for (i = 0; i < ARRAY_SIZE(memory_stats); i++) { |
| 1573 | u64 size; |
Johannes Weiner | c8713d0 | 2019-07-11 20:55:59 -0700 | [diff] [blame] | 1574 | |
Muchun Song | fff66b7 | 2021-02-24 12:03:43 -0800 | [diff] [blame] | 1575 | size = memcg_page_state_output(memcg, memory_stats[i].idx); |
Muchun Song | 5f9a4f4 | 2020-10-13 16:52:59 -0700 | [diff] [blame] | 1576 | seq_buf_printf(&s, "%s %llu\n", memory_stats[i].name, size); |
Johannes Weiner | c8713d0 | 2019-07-11 20:55:59 -0700 | [diff] [blame] | 1577 | |
Muchun Song | 5f9a4f4 | 2020-10-13 16:52:59 -0700 | [diff] [blame] | 1578 | if (unlikely(memory_stats[i].idx == NR_SLAB_UNRECLAIMABLE_B)) { |
Muchun Song | fff66b7 | 2021-02-24 12:03:43 -0800 | [diff] [blame] | 1579 | size += memcg_page_state_output(memcg, |
| 1580 | NR_SLAB_RECLAIMABLE_B); |
Muchun Song | 5f9a4f4 | 2020-10-13 16:52:59 -0700 | [diff] [blame] | 1581 | seq_buf_printf(&s, "slab %llu\n", size); |
| 1582 | } |
| 1583 | } |
Johannes Weiner | c8713d0 | 2019-07-11 20:55:59 -0700 | [diff] [blame] | 1584 | |
| 1585 | /* Accumulated memory events */ |
| 1586 | |
Konstantin Khlebnikov | ebc5d83d | 2019-12-04 16:49:53 -0800 | [diff] [blame] | 1587 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGFAULT), |
| 1588 | memcg_events(memcg, PGFAULT)); |
| 1589 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGMAJFAULT), |
| 1590 | memcg_events(memcg, PGMAJFAULT)); |
Konstantin Khlebnikov | ebc5d83d | 2019-12-04 16:49:53 -0800 | [diff] [blame] | 1591 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGREFILL), |
| 1592 | memcg_events(memcg, PGREFILL)); |
Johannes Weiner | c8713d0 | 2019-07-11 20:55:59 -0700 | [diff] [blame] | 1593 | seq_buf_printf(&s, "pgscan %lu\n", |
| 1594 | memcg_events(memcg, PGSCAN_KSWAPD) + |
| 1595 | memcg_events(memcg, PGSCAN_DIRECT)); |
| 1596 | seq_buf_printf(&s, "pgsteal %lu\n", |
| 1597 | memcg_events(memcg, PGSTEAL_KSWAPD) + |
| 1598 | memcg_events(memcg, PGSTEAL_DIRECT)); |
Konstantin Khlebnikov | ebc5d83d | 2019-12-04 16:49:53 -0800 | [diff] [blame] | 1599 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGACTIVATE), |
| 1600 | memcg_events(memcg, PGACTIVATE)); |
| 1601 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGDEACTIVATE), |
| 1602 | memcg_events(memcg, PGDEACTIVATE)); |
| 1603 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGLAZYFREE), |
| 1604 | memcg_events(memcg, PGLAZYFREE)); |
| 1605 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(PGLAZYFREED), |
| 1606 | memcg_events(memcg, PGLAZYFREED)); |
Johannes Weiner | c8713d0 | 2019-07-11 20:55:59 -0700 | [diff] [blame] | 1607 | |
| 1608 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
Konstantin Khlebnikov | ebc5d83d | 2019-12-04 16:49:53 -0800 | [diff] [blame] | 1609 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(THP_FAULT_ALLOC), |
Johannes Weiner | c8713d0 | 2019-07-11 20:55:59 -0700 | [diff] [blame] | 1610 | memcg_events(memcg, THP_FAULT_ALLOC)); |
Konstantin Khlebnikov | ebc5d83d | 2019-12-04 16:49:53 -0800 | [diff] [blame] | 1611 | seq_buf_printf(&s, "%s %lu\n", vm_event_name(THP_COLLAPSE_ALLOC), |
Johannes Weiner | c8713d0 | 2019-07-11 20:55:59 -0700 | [diff] [blame] | 1612 | memcg_events(memcg, THP_COLLAPSE_ALLOC)); |
| 1613 | #endif /* CONFIG_TRANSPARENT_HUGEPAGE */ |
| 1614 | |
| 1615 | /* The above should easily fit into one page */ |
| 1616 | WARN_ON_ONCE(seq_buf_has_overflowed(&s)); |
| 1617 | |
| 1618 | return s.buffer; |
| 1619 | } |
Johannes Weiner | 71cd311 | 2017-05-03 14:55:13 -0700 | [diff] [blame] | 1620 | |
Sha Zhengju | 58cf188 | 2013-02-22 16:32:05 -0800 | [diff] [blame] | 1621 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
Balbir Singh | e222432 | 2009-04-02 16:57:39 -0700 | [diff] [blame] | 1622 | /** |
yuzhoujian | f0c867d | 2018-12-28 00:36:10 -0800 | [diff] [blame] | 1623 | * mem_cgroup_print_oom_context: Print OOM information relevant to |
| 1624 | * memory controller. |
Balbir Singh | e222432 | 2009-04-02 16:57:39 -0700 | [diff] [blame] | 1625 | * @memcg: The memory cgroup that went over limit |
| 1626 | * @p: Task that is going to be killed |
| 1627 | * |
| 1628 | * NOTE: @memcg and @p's mem_cgroup can be different when hierarchy is |
| 1629 | * enabled |
| 1630 | */ |
yuzhoujian | f0c867d | 2018-12-28 00:36:10 -0800 | [diff] [blame] | 1631 | void mem_cgroup_print_oom_context(struct mem_cgroup *memcg, struct task_struct *p) |
| 1632 | { |
| 1633 | rcu_read_lock(); |
| 1634 | |
| 1635 | if (memcg) { |
| 1636 | pr_cont(",oom_memcg="); |
| 1637 | pr_cont_cgroup_path(memcg->css.cgroup); |
| 1638 | } else |
| 1639 | pr_cont(",global_oom"); |
| 1640 | if (p) { |
| 1641 | pr_cont(",task_memcg="); |
| 1642 | pr_cont_cgroup_path(task_cgroup(p, memory_cgrp_id)); |
| 1643 | } |
| 1644 | rcu_read_unlock(); |
| 1645 | } |
| 1646 | |
| 1647 | /** |
| 1648 | * mem_cgroup_print_oom_meminfo: Print OOM memory information relevant to |
| 1649 | * memory controller. |
| 1650 | * @memcg: The memory cgroup that went over limit |
| 1651 | */ |
| 1652 | void mem_cgroup_print_oom_meminfo(struct mem_cgroup *memcg) |
Balbir Singh | e222432 | 2009-04-02 16:57:39 -0700 | [diff] [blame] | 1653 | { |
Johannes Weiner | c8713d0 | 2019-07-11 20:55:59 -0700 | [diff] [blame] | 1654 | char *buf; |
Balbir Singh | e222432 | 2009-04-02 16:57:39 -0700 | [diff] [blame] | 1655 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1656 | pr_info("memory: usage %llukB, limit %llukB, failcnt %lu\n", |
| 1657 | K((u64)page_counter_read(&memcg->memory)), |
Chris Down | 15b4256 | 2020-04-01 21:07:20 -0700 | [diff] [blame] | 1658 | K((u64)READ_ONCE(memcg->memory.max)), memcg->memory.failcnt); |
Johannes Weiner | c8713d0 | 2019-07-11 20:55:59 -0700 | [diff] [blame] | 1659 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
| 1660 | pr_info("swap: usage %llukB, limit %llukB, failcnt %lu\n", |
| 1661 | K((u64)page_counter_read(&memcg->swap)), |
Chris Down | 32d087c | 2020-04-01 21:07:30 -0700 | [diff] [blame] | 1662 | K((u64)READ_ONCE(memcg->swap.max)), memcg->swap.failcnt); |
Johannes Weiner | c8713d0 | 2019-07-11 20:55:59 -0700 | [diff] [blame] | 1663 | else { |
| 1664 | pr_info("memory+swap: usage %llukB, limit %llukB, failcnt %lu\n", |
| 1665 | K((u64)page_counter_read(&memcg->memsw)), |
| 1666 | K((u64)memcg->memsw.max), memcg->memsw.failcnt); |
| 1667 | pr_info("kmem: usage %llukB, limit %llukB, failcnt %lu\n", |
| 1668 | K((u64)page_counter_read(&memcg->kmem)), |
| 1669 | K((u64)memcg->kmem.max), memcg->kmem.failcnt); |
Sha Zhengju | 58cf188 | 2013-02-22 16:32:05 -0800 | [diff] [blame] | 1670 | } |
Johannes Weiner | c8713d0 | 2019-07-11 20:55:59 -0700 | [diff] [blame] | 1671 | |
| 1672 | pr_info("Memory cgroup stats for "); |
| 1673 | pr_cont_cgroup_path(memcg->css.cgroup); |
| 1674 | pr_cont(":"); |
| 1675 | buf = memory_stat_format(memcg); |
| 1676 | if (!buf) |
| 1677 | return; |
| 1678 | pr_info("%s", buf); |
| 1679 | kfree(buf); |
Balbir Singh | e222432 | 2009-04-02 16:57:39 -0700 | [diff] [blame] | 1680 | } |
| 1681 | |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 1682 | /* |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 1683 | * Return the memory (and swap, if configured) limit for a memcg. |
| 1684 | */ |
Roman Gushchin | bbec2e1 | 2018-06-07 17:06:18 -0700 | [diff] [blame] | 1685 | unsigned long mem_cgroup_get_max(struct mem_cgroup *memcg) |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 1686 | { |
Waiman Long | 8d387a5 | 2020-10-13 16:52:52 -0700 | [diff] [blame] | 1687 | unsigned long max = READ_ONCE(memcg->memory.max); |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 1688 | |
Waiman Long | 8d387a5 | 2020-10-13 16:52:52 -0700 | [diff] [blame] | 1689 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
| 1690 | if (mem_cgroup_swappiness(memcg)) |
| 1691 | max += min(READ_ONCE(memcg->swap.max), |
| 1692 | (unsigned long)total_swap_pages); |
| 1693 | } else { /* v1 */ |
| 1694 | if (mem_cgroup_swappiness(memcg)) { |
| 1695 | /* Calculate swap excess capacity from memsw limit */ |
| 1696 | unsigned long swap = READ_ONCE(memcg->memsw.max) - max; |
Michal Hocko | 9a5a8f1 | 2012-11-16 14:14:49 -0800 | [diff] [blame] | 1697 | |
Waiman Long | 8d387a5 | 2020-10-13 16:52:52 -0700 | [diff] [blame] | 1698 | max += min(swap, (unsigned long)total_swap_pages); |
| 1699 | } |
Michal Hocko | 9a5a8f1 | 2012-11-16 14:14:49 -0800 | [diff] [blame] | 1700 | } |
Roman Gushchin | bbec2e1 | 2018-06-07 17:06:18 -0700 | [diff] [blame] | 1701 | return max; |
David Rientjes | a63d83f | 2010-08-09 17:19:46 -0700 | [diff] [blame] | 1702 | } |
| 1703 | |
Chris Down | 9783aa9 | 2019-10-06 17:58:32 -0700 | [diff] [blame] | 1704 | unsigned long mem_cgroup_size(struct mem_cgroup *memcg) |
| 1705 | { |
| 1706 | return page_counter_read(&memcg->memory); |
| 1707 | } |
| 1708 | |
Johannes Weiner | b6e6edc | 2016-03-17 14:20:28 -0700 | [diff] [blame] | 1709 | 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] | 1710 | int order) |
David Rientjes | 9cbb78b | 2012-07-31 16:43:44 -0700 | [diff] [blame] | 1711 | { |
David Rientjes | 6e0fc46 | 2015-09-08 15:00:36 -0700 | [diff] [blame] | 1712 | struct oom_control oc = { |
| 1713 | .zonelist = NULL, |
| 1714 | .nodemask = NULL, |
Vladimir Davydov | 2a966b7 | 2016-07-26 15:22:33 -0700 | [diff] [blame] | 1715 | .memcg = memcg, |
David Rientjes | 6e0fc46 | 2015-09-08 15:00:36 -0700 | [diff] [blame] | 1716 | .gfp_mask = gfp_mask, |
| 1717 | .order = order, |
David Rientjes | 6e0fc46 | 2015-09-08 15:00:36 -0700 | [diff] [blame] | 1718 | }; |
Yafang Shao | 1378b37 | 2020-08-06 23:22:08 -0700 | [diff] [blame] | 1719 | bool ret = true; |
David Rientjes | 9cbb78b | 2012-07-31 16:43:44 -0700 | [diff] [blame] | 1720 | |
Tetsuo Handa | 7775fac | 2019-03-05 15:46:47 -0800 | [diff] [blame] | 1721 | if (mutex_lock_killable(&oom_lock)) |
| 1722 | return true; |
Yafang Shao | 1378b37 | 2020-08-06 23:22:08 -0700 | [diff] [blame] | 1723 | |
| 1724 | if (mem_cgroup_margin(memcg) >= (1 << order)) |
| 1725 | goto unlock; |
| 1726 | |
Tetsuo Handa | 7775fac | 2019-03-05 15:46:47 -0800 | [diff] [blame] | 1727 | /* |
| 1728 | * A few threads which were not waiting at mutex_lock_killable() can |
| 1729 | * fail to bail out. Therefore, check again after holding oom_lock. |
| 1730 | */ |
| 1731 | ret = should_force_charge() || out_of_memory(&oc); |
Yafang Shao | 1378b37 | 2020-08-06 23:22:08 -0700 | [diff] [blame] | 1732 | |
| 1733 | unlock: |
Johannes Weiner | dc56401 | 2015-06-24 16:57:19 -0700 | [diff] [blame] | 1734 | mutex_unlock(&oom_lock); |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 1735 | return ret; |
David Rientjes | 9cbb78b | 2012-07-31 16:43:44 -0700 | [diff] [blame] | 1736 | } |
| 1737 | |
Andrew Morton | 0608f43 | 2013-09-24 15:27:41 -0700 | [diff] [blame] | 1738 | static int mem_cgroup_soft_reclaim(struct mem_cgroup *root_memcg, |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 1739 | pg_data_t *pgdat, |
Andrew Morton | 0608f43 | 2013-09-24 15:27:41 -0700 | [diff] [blame] | 1740 | gfp_t gfp_mask, |
| 1741 | unsigned long *total_scanned) |
Balbir Singh | 6d61ef4 | 2009-01-07 18:08:06 -0800 | [diff] [blame] | 1742 | { |
Andrew Morton | 0608f43 | 2013-09-24 15:27:41 -0700 | [diff] [blame] | 1743 | struct mem_cgroup *victim = NULL; |
| 1744 | int total = 0; |
| 1745 | int loop = 0; |
| 1746 | unsigned long excess; |
| 1747 | unsigned long nr_scanned; |
| 1748 | struct mem_cgroup_reclaim_cookie reclaim = { |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 1749 | .pgdat = pgdat, |
Andrew Morton | 0608f43 | 2013-09-24 15:27:41 -0700 | [diff] [blame] | 1750 | }; |
Johannes Weiner | 9d11ea9 | 2011-03-23 16:42:21 -0700 | [diff] [blame] | 1751 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1752 | excess = soft_limit_excess(root_memcg); |
Balbir Singh | 6d61ef4 | 2009-01-07 18:08:06 -0800 | [diff] [blame] | 1753 | |
Andrew Morton | 0608f43 | 2013-09-24 15:27:41 -0700 | [diff] [blame] | 1754 | while (1) { |
| 1755 | victim = mem_cgroup_iter(root_memcg, victim, &reclaim); |
| 1756 | if (!victim) { |
| 1757 | loop++; |
| 1758 | if (loop >= 2) { |
| 1759 | /* |
| 1760 | * If we have not been able to reclaim |
| 1761 | * anything, it might because there are |
| 1762 | * no reclaimable pages under this hierarchy |
| 1763 | */ |
| 1764 | if (!total) |
| 1765 | break; |
| 1766 | /* |
| 1767 | * We want to do more targeted reclaim. |
| 1768 | * excess >> 2 is not to excessive so as to |
| 1769 | * reclaim too much, nor too less that we keep |
| 1770 | * coming back to reclaim from this cgroup |
| 1771 | */ |
| 1772 | if (total >= (excess >> 2) || |
| 1773 | (loop > MEM_CGROUP_MAX_RECLAIM_LOOPS)) |
| 1774 | break; |
| 1775 | } |
| 1776 | continue; |
| 1777 | } |
Mel Gorman | a9dd0a8 | 2016-07-28 15:46:02 -0700 | [diff] [blame] | 1778 | total += mem_cgroup_shrink_node(victim, gfp_mask, false, |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 1779 | pgdat, &nr_scanned); |
Andrew Morton | 0608f43 | 2013-09-24 15:27:41 -0700 | [diff] [blame] | 1780 | *total_scanned += nr_scanned; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 1781 | if (!soft_limit_excess(root_memcg)) |
Andrew Morton | 0608f43 | 2013-09-24 15:27:41 -0700 | [diff] [blame] | 1782 | break; |
Balbir Singh | 6d61ef4 | 2009-01-07 18:08:06 -0800 | [diff] [blame] | 1783 | } |
Andrew Morton | 0608f43 | 2013-09-24 15:27:41 -0700 | [diff] [blame] | 1784 | mem_cgroup_iter_break(root_memcg, victim); |
| 1785 | return total; |
Balbir Singh | 6d61ef4 | 2009-01-07 18:08:06 -0800 | [diff] [blame] | 1786 | } |
| 1787 | |
Johannes Weiner | 0056f4e | 2013-10-31 16:34:14 -0700 | [diff] [blame] | 1788 | #ifdef CONFIG_LOCKDEP |
| 1789 | static struct lockdep_map memcg_oom_lock_dep_map = { |
| 1790 | .name = "memcg_oom_lock", |
| 1791 | }; |
| 1792 | #endif |
| 1793 | |
Johannes Weiner | fb2a6fc | 2013-09-12 15:13:43 -0700 | [diff] [blame] | 1794 | static DEFINE_SPINLOCK(memcg_oom_lock); |
| 1795 | |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1796 | /* |
| 1797 | * Check OOM-Killer is already running under our hierarchy. |
| 1798 | * If someone is running, return false. |
| 1799 | */ |
Johannes Weiner | fb2a6fc | 2013-09-12 15:13:43 -0700 | [diff] [blame] | 1800 | static bool mem_cgroup_oom_trylock(struct mem_cgroup *memcg) |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1801 | { |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 1802 | struct mem_cgroup *iter, *failed = NULL; |
KAMEZAWA Hiroyuki | a636b32 | 2009-01-07 18:08:08 -0800 | [diff] [blame] | 1803 | |
Johannes Weiner | fb2a6fc | 2013-09-12 15:13:43 -0700 | [diff] [blame] | 1804 | spin_lock(&memcg_oom_lock); |
| 1805 | |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 1806 | for_each_mem_cgroup_tree(iter, memcg) { |
Johannes Weiner | 23751be | 2011-08-25 15:59:16 -0700 | [diff] [blame] | 1807 | if (iter->oom_lock) { |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 1808 | /* |
| 1809 | * this subtree of our hierarchy is already locked |
| 1810 | * so we cannot give a lock. |
| 1811 | */ |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 1812 | failed = iter; |
Johannes Weiner | 9f3a0d0 | 2012-01-12 17:17:48 -0800 | [diff] [blame] | 1813 | mem_cgroup_iter_break(memcg, iter); |
| 1814 | break; |
Johannes Weiner | 23751be | 2011-08-25 15:59:16 -0700 | [diff] [blame] | 1815 | } else |
| 1816 | iter->oom_lock = true; |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 1817 | } |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1818 | |
Johannes Weiner | fb2a6fc | 2013-09-12 15:13:43 -0700 | [diff] [blame] | 1819 | if (failed) { |
| 1820 | /* |
| 1821 | * OK, we failed to lock the whole subtree so we have |
| 1822 | * to clean up what we set up to the failing subtree |
| 1823 | */ |
| 1824 | for_each_mem_cgroup_tree(iter, memcg) { |
| 1825 | if (iter == failed) { |
| 1826 | mem_cgroup_iter_break(memcg, iter); |
| 1827 | break; |
| 1828 | } |
| 1829 | iter->oom_lock = false; |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 1830 | } |
Johannes Weiner | 0056f4e | 2013-10-31 16:34:14 -0700 | [diff] [blame] | 1831 | } else |
| 1832 | mutex_acquire(&memcg_oom_lock_dep_map, 0, 1, _RET_IP_); |
Johannes Weiner | fb2a6fc | 2013-09-12 15:13:43 -0700 | [diff] [blame] | 1833 | |
| 1834 | spin_unlock(&memcg_oom_lock); |
| 1835 | |
| 1836 | return !failed; |
KAMEZAWA Hiroyuki | a636b32 | 2009-01-07 18:08:08 -0800 | [diff] [blame] | 1837 | } |
KAMEZAWA Hiroyuki | 0b7f569 | 2009-04-02 16:57:38 -0700 | [diff] [blame] | 1838 | |
Johannes Weiner | fb2a6fc | 2013-09-12 15:13:43 -0700 | [diff] [blame] | 1839 | static void mem_cgroup_oom_unlock(struct mem_cgroup *memcg) |
KAMEZAWA Hiroyuki | 0b7f569 | 2009-04-02 16:57:38 -0700 | [diff] [blame] | 1840 | { |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 1841 | struct mem_cgroup *iter; |
| 1842 | |
Johannes Weiner | fb2a6fc | 2013-09-12 15:13:43 -0700 | [diff] [blame] | 1843 | spin_lock(&memcg_oom_lock); |
Qian Cai | 5facae4 | 2019-09-19 12:09:40 -0400 | [diff] [blame] | 1844 | mutex_release(&memcg_oom_lock_dep_map, _RET_IP_); |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1845 | for_each_mem_cgroup_tree(iter, memcg) |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 1846 | iter->oom_lock = false; |
Johannes Weiner | fb2a6fc | 2013-09-12 15:13:43 -0700 | [diff] [blame] | 1847 | spin_unlock(&memcg_oom_lock); |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 1848 | } |
| 1849 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1850 | static void mem_cgroup_mark_under_oom(struct mem_cgroup *memcg) |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 1851 | { |
| 1852 | struct mem_cgroup *iter; |
| 1853 | |
Tejun Heo | c2b42d3 | 2015-06-24 16:58:23 -0700 | [diff] [blame] | 1854 | spin_lock(&memcg_oom_lock); |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1855 | for_each_mem_cgroup_tree(iter, memcg) |
Tejun Heo | c2b42d3 | 2015-06-24 16:58:23 -0700 | [diff] [blame] | 1856 | iter->under_oom++; |
| 1857 | spin_unlock(&memcg_oom_lock); |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 1858 | } |
| 1859 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1860 | static void mem_cgroup_unmark_under_oom(struct mem_cgroup *memcg) |
Michal Hocko | 79dfdac | 2011-07-26 16:08:23 -0700 | [diff] [blame] | 1861 | { |
| 1862 | struct mem_cgroup *iter; |
| 1863 | |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1864 | /* |
Miaohe Lin | 7a52d4d8 | 2020-10-13 16:53:05 -0700 | [diff] [blame] | 1865 | * Be careful about under_oom underflows becase a child memcg |
| 1866 | * could have been added after mem_cgroup_mark_under_oom. |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1867 | */ |
Tejun Heo | c2b42d3 | 2015-06-24 16:58:23 -0700 | [diff] [blame] | 1868 | spin_lock(&memcg_oom_lock); |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1869 | for_each_mem_cgroup_tree(iter, memcg) |
Tejun Heo | c2b42d3 | 2015-06-24 16:58:23 -0700 | [diff] [blame] | 1870 | if (iter->under_oom > 0) |
| 1871 | iter->under_oom--; |
| 1872 | spin_unlock(&memcg_oom_lock); |
KAMEZAWA Hiroyuki | 0b7f569 | 2009-04-02 16:57:38 -0700 | [diff] [blame] | 1873 | } |
| 1874 | |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1875 | static DECLARE_WAIT_QUEUE_HEAD(memcg_oom_waitq); |
| 1876 | |
KAMEZAWA Hiroyuki | dc98df5 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 1877 | struct oom_wait_info { |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 1878 | struct mem_cgroup *memcg; |
Ingo Molnar | ac6424b | 2017-06-20 12:06:13 +0200 | [diff] [blame] | 1879 | wait_queue_entry_t wait; |
KAMEZAWA Hiroyuki | dc98df5 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 1880 | }; |
| 1881 | |
Ingo Molnar | ac6424b | 2017-06-20 12:06:13 +0200 | [diff] [blame] | 1882 | static int memcg_oom_wake_function(wait_queue_entry_t *wait, |
KAMEZAWA Hiroyuki | dc98df5 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 1883 | unsigned mode, int sync, void *arg) |
| 1884 | { |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 1885 | struct mem_cgroup *wake_memcg = (struct mem_cgroup *)arg; |
| 1886 | struct mem_cgroup *oom_wait_memcg; |
KAMEZAWA Hiroyuki | dc98df5 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 1887 | struct oom_wait_info *oom_wait_info; |
| 1888 | |
| 1889 | oom_wait_info = container_of(wait, struct oom_wait_info, wait); |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 1890 | oom_wait_memcg = oom_wait_info->memcg; |
KAMEZAWA Hiroyuki | dc98df5 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 1891 | |
Johannes Weiner | 2314b42 | 2014-12-10 15:44:33 -0800 | [diff] [blame] | 1892 | if (!mem_cgroup_is_descendant(wake_memcg, oom_wait_memcg) && |
| 1893 | !mem_cgroup_is_descendant(oom_wait_memcg, wake_memcg)) |
KAMEZAWA Hiroyuki | dc98df5 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 1894 | return 0; |
KAMEZAWA Hiroyuki | dc98df5 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 1895 | return autoremove_wake_function(wait, mode, sync, arg); |
| 1896 | } |
| 1897 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 1898 | static void memcg_oom_recover(struct mem_cgroup *memcg) |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 1899 | { |
Tejun Heo | c2b42d3 | 2015-06-24 16:58:23 -0700 | [diff] [blame] | 1900 | /* |
| 1901 | * For the following lockless ->under_oom test, the only required |
| 1902 | * guarantee is that it must see the state asserted by an OOM when |
| 1903 | * this function is called as a result of userland actions |
| 1904 | * triggered by the notification of the OOM. This is trivially |
| 1905 | * achieved by invoking mem_cgroup_mark_under_oom() before |
| 1906 | * triggering notification. |
| 1907 | */ |
| 1908 | if (memcg && memcg->under_oom) |
Tejun Heo | f4b90b70 | 2015-06-24 16:58:21 -0700 | [diff] [blame] | 1909 | __wake_up(&memcg_oom_waitq, TASK_NORMAL, 0, memcg); |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 1910 | } |
| 1911 | |
Michal Hocko | 29ef680 | 2018-08-17 15:47:11 -0700 | [diff] [blame] | 1912 | enum oom_status { |
| 1913 | OOM_SUCCESS, |
| 1914 | OOM_FAILED, |
| 1915 | OOM_ASYNC, |
| 1916 | OOM_SKIPPED |
| 1917 | }; |
| 1918 | |
| 1919 | static enum oom_status mem_cgroup_oom(struct mem_cgroup *memcg, gfp_t mask, int order) |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1920 | { |
Michal Hocko | 7056d3a | 2018-12-28 00:39:57 -0800 | [diff] [blame] | 1921 | enum oom_status ret; |
| 1922 | bool locked; |
| 1923 | |
Michal Hocko | 29ef680 | 2018-08-17 15:47:11 -0700 | [diff] [blame] | 1924 | if (order > PAGE_ALLOC_COSTLY_ORDER) |
| 1925 | return OOM_SKIPPED; |
| 1926 | |
Roman Gushchin | 7a1adfd | 2018-10-26 15:09:48 -0700 | [diff] [blame] | 1927 | memcg_memory_event(memcg, MEMCG_OOM); |
| 1928 | |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1929 | /* |
Johannes Weiner | 4942642 | 2013-10-16 13:46:59 -0700 | [diff] [blame] | 1930 | * We are in the middle of the charge context here, so we |
| 1931 | * don't want to block when potentially sitting on a callstack |
| 1932 | * that holds all kinds of filesystem and mm locks. |
| 1933 | * |
Michal Hocko | 29ef680 | 2018-08-17 15:47:11 -0700 | [diff] [blame] | 1934 | * cgroup1 allows disabling the OOM killer and waiting for outside |
| 1935 | * handling until the charge can succeed; remember the context and put |
| 1936 | * the task to sleep at the end of the page fault when all locks are |
| 1937 | * released. |
Johannes Weiner | 4942642 | 2013-10-16 13:46:59 -0700 | [diff] [blame] | 1938 | * |
Michal Hocko | 29ef680 | 2018-08-17 15:47:11 -0700 | [diff] [blame] | 1939 | * On the other hand, in-kernel OOM killer allows for an async victim |
| 1940 | * memory reclaim (oom_reaper) and that means that we are not solely |
| 1941 | * relying on the oom victim to make a forward progress and we can |
| 1942 | * invoke the oom killer here. |
| 1943 | * |
| 1944 | * Please note that mem_cgroup_out_of_memory might fail to find a |
| 1945 | * victim and then we have to bail out from the charge path. |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 1946 | */ |
Michal Hocko | 29ef680 | 2018-08-17 15:47:11 -0700 | [diff] [blame] | 1947 | if (memcg->oom_kill_disable) { |
| 1948 | if (!current->in_user_fault) |
| 1949 | return OOM_SKIPPED; |
| 1950 | css_get(&memcg->css); |
| 1951 | current->memcg_in_oom = memcg; |
| 1952 | current->memcg_oom_gfp_mask = mask; |
| 1953 | current->memcg_oom_order = order; |
| 1954 | |
| 1955 | return OOM_ASYNC; |
| 1956 | } |
| 1957 | |
Michal Hocko | 7056d3a | 2018-12-28 00:39:57 -0800 | [diff] [blame] | 1958 | mem_cgroup_mark_under_oom(memcg); |
Michal Hocko | 29ef680 | 2018-08-17 15:47:11 -0700 | [diff] [blame] | 1959 | |
Michal Hocko | 7056d3a | 2018-12-28 00:39:57 -0800 | [diff] [blame] | 1960 | locked = mem_cgroup_oom_trylock(memcg); |
| 1961 | |
| 1962 | if (locked) |
| 1963 | mem_cgroup_oom_notify(memcg); |
| 1964 | |
| 1965 | mem_cgroup_unmark_under_oom(memcg); |
| 1966 | if (mem_cgroup_out_of_memory(memcg, mask, order)) |
| 1967 | ret = OOM_SUCCESS; |
| 1968 | else |
| 1969 | ret = OOM_FAILED; |
| 1970 | |
| 1971 | if (locked) |
| 1972 | mem_cgroup_oom_unlock(memcg); |
| 1973 | |
| 1974 | return ret; |
Johannes Weiner | 4942642 | 2013-10-16 13:46:59 -0700 | [diff] [blame] | 1975 | } |
| 1976 | |
| 1977 | /** |
| 1978 | * mem_cgroup_oom_synchronize - complete memcg OOM handling |
| 1979 | * @handle: actually kill/wait or just clean up the OOM state |
| 1980 | * |
| 1981 | * This has to be called at the end of a page fault if the memcg OOM |
| 1982 | * handler was enabled. |
| 1983 | * |
| 1984 | * Memcg supports userspace OOM handling where failed allocations must |
| 1985 | * sleep on a waitqueue until the userspace task resolves the |
| 1986 | * situation. Sleeping directly in the charge context with all kinds |
| 1987 | * of locks held is not a good idea, instead we remember an OOM state |
| 1988 | * in the task and mem_cgroup_oom_synchronize() has to be called at |
| 1989 | * the end of the page fault to complete the OOM handling. |
| 1990 | * |
| 1991 | * Returns %true if an ongoing memcg OOM situation was detected and |
| 1992 | * completed, %false otherwise. |
| 1993 | */ |
| 1994 | bool mem_cgroup_oom_synchronize(bool handle) |
| 1995 | { |
Tejun Heo | 626ebc4 | 2015-11-05 18:46:09 -0800 | [diff] [blame] | 1996 | struct mem_cgroup *memcg = current->memcg_in_oom; |
Johannes Weiner | 4942642 | 2013-10-16 13:46:59 -0700 | [diff] [blame] | 1997 | struct oom_wait_info owait; |
| 1998 | bool locked; |
| 1999 | |
| 2000 | /* OOM is global, do not handle */ |
| 2001 | if (!memcg) |
| 2002 | return false; |
| 2003 | |
Vladimir Davydov | 7c5f64f | 2016-10-07 16:57:23 -0700 | [diff] [blame] | 2004 | if (!handle) |
Johannes Weiner | 4942642 | 2013-10-16 13:46:59 -0700 | [diff] [blame] | 2005 | goto cleanup; |
| 2006 | |
| 2007 | owait.memcg = memcg; |
| 2008 | owait.wait.flags = 0; |
| 2009 | owait.wait.func = memcg_oom_wake_function; |
| 2010 | owait.wait.private = current; |
Ingo Molnar | 2055da9 | 2017-06-20 12:06:46 +0200 | [diff] [blame] | 2011 | INIT_LIST_HEAD(&owait.wait.entry); |
Johannes Weiner | 4942642 | 2013-10-16 13:46:59 -0700 | [diff] [blame] | 2012 | |
| 2013 | prepare_to_wait(&memcg_oom_waitq, &owait.wait, TASK_KILLABLE); |
Johannes Weiner | fb2a6fc | 2013-09-12 15:13:43 -0700 | [diff] [blame] | 2014 | mem_cgroup_mark_under_oom(memcg); |
| 2015 | |
| 2016 | locked = mem_cgroup_oom_trylock(memcg); |
| 2017 | |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 2018 | if (locked) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2019 | mem_cgroup_oom_notify(memcg); |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 2020 | |
Johannes Weiner | fb2a6fc | 2013-09-12 15:13:43 -0700 | [diff] [blame] | 2021 | if (locked && !memcg->oom_kill_disable) { |
| 2022 | mem_cgroup_unmark_under_oom(memcg); |
Johannes Weiner | 4942642 | 2013-10-16 13:46:59 -0700 | [diff] [blame] | 2023 | finish_wait(&memcg_oom_waitq, &owait.wait); |
Tejun Heo | 626ebc4 | 2015-11-05 18:46:09 -0800 | [diff] [blame] | 2024 | mem_cgroup_out_of_memory(memcg, current->memcg_oom_gfp_mask, |
| 2025 | current->memcg_oom_order); |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 2026 | } else { |
Johannes Weiner | 3812c8c | 2013-09-12 15:13:44 -0700 | [diff] [blame] | 2027 | schedule(); |
Johannes Weiner | 4942642 | 2013-10-16 13:46:59 -0700 | [diff] [blame] | 2028 | mem_cgroup_unmark_under_oom(memcg); |
| 2029 | finish_wait(&memcg_oom_waitq, &owait.wait); |
| 2030 | } |
| 2031 | |
| 2032 | if (locked) { |
Johannes Weiner | fb2a6fc | 2013-09-12 15:13:43 -0700 | [diff] [blame] | 2033 | mem_cgroup_oom_unlock(memcg); |
| 2034 | /* |
| 2035 | * There is no guarantee that an OOM-lock contender |
| 2036 | * sees the wakeups triggered by the OOM kill |
| 2037 | * uncharges. Wake any sleepers explicitely. |
| 2038 | */ |
| 2039 | memcg_oom_recover(memcg); |
| 2040 | } |
Johannes Weiner | 4942642 | 2013-10-16 13:46:59 -0700 | [diff] [blame] | 2041 | cleanup: |
Tejun Heo | 626ebc4 | 2015-11-05 18:46:09 -0800 | [diff] [blame] | 2042 | current->memcg_in_oom = NULL; |
Johannes Weiner | 3812c8c | 2013-09-12 15:13:44 -0700 | [diff] [blame] | 2043 | css_put(&memcg->css); |
KAMEZAWA Hiroyuki | 867578c | 2010-03-10 15:22:39 -0800 | [diff] [blame] | 2044 | return true; |
KAMEZAWA Hiroyuki | 0b7f569 | 2009-04-02 16:57:38 -0700 | [diff] [blame] | 2045 | } |
| 2046 | |
Johannes Weiner | d7365e7 | 2014-10-29 14:50:48 -0700 | [diff] [blame] | 2047 | /** |
Roman Gushchin | 3d8b38e | 2018-08-21 21:53:54 -0700 | [diff] [blame] | 2048 | * mem_cgroup_get_oom_group - get a memory cgroup to clean up after OOM |
| 2049 | * @victim: task to be killed by the OOM killer |
| 2050 | * @oom_domain: memcg in case of memcg OOM, NULL in case of system-wide OOM |
| 2051 | * |
| 2052 | * Returns a pointer to a memory cgroup, which has to be cleaned up |
| 2053 | * by killing all belonging OOM-killable tasks. |
| 2054 | * |
| 2055 | * Caller has to call mem_cgroup_put() on the returned non-NULL memcg. |
| 2056 | */ |
| 2057 | struct mem_cgroup *mem_cgroup_get_oom_group(struct task_struct *victim, |
| 2058 | struct mem_cgroup *oom_domain) |
| 2059 | { |
| 2060 | struct mem_cgroup *oom_group = NULL; |
| 2061 | struct mem_cgroup *memcg; |
| 2062 | |
| 2063 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
| 2064 | return NULL; |
| 2065 | |
| 2066 | if (!oom_domain) |
| 2067 | oom_domain = root_mem_cgroup; |
| 2068 | |
| 2069 | rcu_read_lock(); |
| 2070 | |
| 2071 | memcg = mem_cgroup_from_task(victim); |
| 2072 | if (memcg == root_mem_cgroup) |
| 2073 | goto out; |
| 2074 | |
| 2075 | /* |
Roman Gushchin | 48fe267 | 2020-04-01 21:07:39 -0700 | [diff] [blame] | 2076 | * If the victim task has been asynchronously moved to a different |
| 2077 | * memory cgroup, we might end up killing tasks outside oom_domain. |
| 2078 | * In this case it's better to ignore memory.group.oom. |
| 2079 | */ |
| 2080 | if (unlikely(!mem_cgroup_is_descendant(memcg, oom_domain))) |
| 2081 | goto out; |
| 2082 | |
| 2083 | /* |
Roman Gushchin | 3d8b38e | 2018-08-21 21:53:54 -0700 | [diff] [blame] | 2084 | * Traverse the memory cgroup hierarchy from the victim task's |
| 2085 | * cgroup up to the OOMing cgroup (or root) to find the |
| 2086 | * highest-level memory cgroup with oom.group set. |
| 2087 | */ |
| 2088 | for (; memcg; memcg = parent_mem_cgroup(memcg)) { |
| 2089 | if (memcg->oom_group) |
| 2090 | oom_group = memcg; |
| 2091 | |
| 2092 | if (memcg == oom_domain) |
| 2093 | break; |
| 2094 | } |
| 2095 | |
| 2096 | if (oom_group) |
| 2097 | css_get(&oom_group->css); |
| 2098 | out: |
| 2099 | rcu_read_unlock(); |
| 2100 | |
| 2101 | return oom_group; |
| 2102 | } |
| 2103 | |
| 2104 | void mem_cgroup_print_oom_group(struct mem_cgroup *memcg) |
| 2105 | { |
| 2106 | pr_info("Tasks in "); |
| 2107 | pr_cont_cgroup_path(memcg->css.cgroup); |
| 2108 | pr_cont(" are going to be killed due to memory.oom.group set\n"); |
| 2109 | } |
| 2110 | |
| 2111 | /** |
Roman Gushchin | bcfe06b | 2020-12-01 13:58:27 -0800 | [diff] [blame] | 2112 | * lock_page_memcg - lock a page and memcg binding |
Johannes Weiner | 81f8c3a | 2016-03-15 14:57:04 -0700 | [diff] [blame] | 2113 | * @page: the page |
KAMEZAWA Hiroyuki | 32047e2 | 2010-10-27 15:33:40 -0700 | [diff] [blame] | 2114 | * |
Johannes Weiner | 81f8c3a | 2016-03-15 14:57:04 -0700 | [diff] [blame] | 2115 | * This function protects unlocked LRU pages from being moved to |
Johannes Weiner | 739f79f | 2017-08-18 15:15:48 -0700 | [diff] [blame] | 2116 | * another cgroup. |
| 2117 | * |
Johannes Weiner | 1c824a6 | 2021-04-29 22:55:32 -0700 | [diff] [blame] | 2118 | * It ensures lifetime of the locked memcg. Caller is responsible |
| 2119 | * for the lifetime of the page. |
Balbir Singh | d69b042 | 2009-06-17 16:26:34 -0700 | [diff] [blame] | 2120 | */ |
Johannes Weiner | 1c824a6 | 2021-04-29 22:55:32 -0700 | [diff] [blame] | 2121 | void lock_page_memcg(struct page *page) |
KAMEZAWA Hiroyuki | 89c06bd | 2012-03-21 16:34:25 -0700 | [diff] [blame] | 2122 | { |
Johannes Weiner | 9da7b52 | 2020-06-03 16:01:51 -0700 | [diff] [blame] | 2123 | struct page *head = compound_head(page); /* rmap on tail pages */ |
KAMEZAWA Hiroyuki | 89c06bd | 2012-03-21 16:34:25 -0700 | [diff] [blame] | 2124 | struct mem_cgroup *memcg; |
Johannes Weiner | 6de2261 | 2015-02-11 15:25:01 -0800 | [diff] [blame] | 2125 | unsigned long flags; |
KAMEZAWA Hiroyuki | 89c06bd | 2012-03-21 16:34:25 -0700 | [diff] [blame] | 2126 | |
Johannes Weiner | 6de2261 | 2015-02-11 15:25:01 -0800 | [diff] [blame] | 2127 | /* |
| 2128 | * The RCU lock is held throughout the transaction. The fast |
| 2129 | * path can get away without acquiring the memcg->move_lock |
| 2130 | * because page moving starts with an RCU grace period. |
Johannes Weiner | 739f79f | 2017-08-18 15:15:48 -0700 | [diff] [blame] | 2131 | */ |
Johannes Weiner | d7365e7 | 2014-10-29 14:50:48 -0700 | [diff] [blame] | 2132 | rcu_read_lock(); |
| 2133 | |
| 2134 | if (mem_cgroup_disabled()) |
Johannes Weiner | 1c824a6 | 2021-04-29 22:55:32 -0700 | [diff] [blame] | 2135 | return; |
KAMEZAWA Hiroyuki | 89c06bd | 2012-03-21 16:34:25 -0700 | [diff] [blame] | 2136 | again: |
Roman Gushchin | bcfe06b | 2020-12-01 13:58:27 -0800 | [diff] [blame] | 2137 | memcg = page_memcg(head); |
Johannes Weiner | 2983331 | 2014-12-10 15:44:02 -0800 | [diff] [blame] | 2138 | if (unlikely(!memcg)) |
Johannes Weiner | 1c824a6 | 2021-04-29 22:55:32 -0700 | [diff] [blame] | 2139 | return; |
Johannes Weiner | d7365e7 | 2014-10-29 14:50:48 -0700 | [diff] [blame] | 2140 | |
Alex Shi | 20ad50d | 2020-12-15 12:33:51 -0800 | [diff] [blame] | 2141 | #ifdef CONFIG_PROVE_LOCKING |
| 2142 | local_irq_save(flags); |
| 2143 | might_lock(&memcg->move_lock); |
| 2144 | local_irq_restore(flags); |
| 2145 | #endif |
| 2146 | |
Qiang Huang | bdcbb65 | 2014-06-04 16:08:21 -0700 | [diff] [blame] | 2147 | if (atomic_read(&memcg->moving_account) <= 0) |
Johannes Weiner | 1c824a6 | 2021-04-29 22:55:32 -0700 | [diff] [blame] | 2148 | return; |
KAMEZAWA Hiroyuki | 89c06bd | 2012-03-21 16:34:25 -0700 | [diff] [blame] | 2149 | |
Johannes Weiner | 6de2261 | 2015-02-11 15:25:01 -0800 | [diff] [blame] | 2150 | spin_lock_irqsave(&memcg->move_lock, flags); |
Roman Gushchin | bcfe06b | 2020-12-01 13:58:27 -0800 | [diff] [blame] | 2151 | if (memcg != page_memcg(head)) { |
Johannes Weiner | 6de2261 | 2015-02-11 15:25:01 -0800 | [diff] [blame] | 2152 | spin_unlock_irqrestore(&memcg->move_lock, flags); |
KAMEZAWA Hiroyuki | 89c06bd | 2012-03-21 16:34:25 -0700 | [diff] [blame] | 2153 | goto again; |
| 2154 | } |
Johannes Weiner | 6de2261 | 2015-02-11 15:25:01 -0800 | [diff] [blame] | 2155 | |
| 2156 | /* |
Johannes Weiner | 1c824a6 | 2021-04-29 22:55:32 -0700 | [diff] [blame] | 2157 | * When charge migration first begins, we can have multiple |
| 2158 | * critical sections holding the fast-path RCU lock and one |
| 2159 | * holding the slowpath move_lock. Track the task who has the |
| 2160 | * move_lock for unlock_page_memcg(). |
Johannes Weiner | 6de2261 | 2015-02-11 15:25:01 -0800 | [diff] [blame] | 2161 | */ |
| 2162 | memcg->move_lock_task = current; |
| 2163 | memcg->move_lock_flags = flags; |
KAMEZAWA Hiroyuki | 89c06bd | 2012-03-21 16:34:25 -0700 | [diff] [blame] | 2164 | } |
Johannes Weiner | 81f8c3a | 2016-03-15 14:57:04 -0700 | [diff] [blame] | 2165 | EXPORT_SYMBOL(lock_page_memcg); |
KAMEZAWA Hiroyuki | 89c06bd | 2012-03-21 16:34:25 -0700 | [diff] [blame] | 2166 | |
Johannes Weiner | 1c824a6 | 2021-04-29 22:55:32 -0700 | [diff] [blame] | 2167 | static void __unlock_page_memcg(struct mem_cgroup *memcg) |
KAMEZAWA Hiroyuki | 89c06bd | 2012-03-21 16:34:25 -0700 | [diff] [blame] | 2168 | { |
Johannes Weiner | 6de2261 | 2015-02-11 15:25:01 -0800 | [diff] [blame] | 2169 | if (memcg && memcg->move_lock_task == current) { |
| 2170 | unsigned long flags = memcg->move_lock_flags; |
| 2171 | |
| 2172 | memcg->move_lock_task = NULL; |
| 2173 | memcg->move_lock_flags = 0; |
| 2174 | |
| 2175 | spin_unlock_irqrestore(&memcg->move_lock, flags); |
| 2176 | } |
KAMEZAWA Hiroyuki | 89c06bd | 2012-03-21 16:34:25 -0700 | [diff] [blame] | 2177 | |
Johannes Weiner | d7365e7 | 2014-10-29 14:50:48 -0700 | [diff] [blame] | 2178 | rcu_read_unlock(); |
KAMEZAWA Hiroyuki | 89c06bd | 2012-03-21 16:34:25 -0700 | [diff] [blame] | 2179 | } |
Johannes Weiner | 739f79f | 2017-08-18 15:15:48 -0700 | [diff] [blame] | 2180 | |
| 2181 | /** |
Roman Gushchin | bcfe06b | 2020-12-01 13:58:27 -0800 | [diff] [blame] | 2182 | * unlock_page_memcg - unlock a page and memcg binding |
Johannes Weiner | 739f79f | 2017-08-18 15:15:48 -0700 | [diff] [blame] | 2183 | * @page: the page |
| 2184 | */ |
| 2185 | void unlock_page_memcg(struct page *page) |
| 2186 | { |
Johannes Weiner | 9da7b52 | 2020-06-03 16:01:51 -0700 | [diff] [blame] | 2187 | struct page *head = compound_head(page); |
| 2188 | |
Roman Gushchin | bcfe06b | 2020-12-01 13:58:27 -0800 | [diff] [blame] | 2189 | __unlock_page_memcg(page_memcg(head)); |
Johannes Weiner | 739f79f | 2017-08-18 15:15:48 -0700 | [diff] [blame] | 2190 | } |
Johannes Weiner | 81f8c3a | 2016-03-15 14:57:04 -0700 | [diff] [blame] | 2191 | EXPORT_SYMBOL(unlock_page_memcg); |
KAMEZAWA Hiroyuki | 89c06bd | 2012-03-21 16:34:25 -0700 | [diff] [blame] | 2192 | |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2193 | struct memcg_stock_pcp { |
| 2194 | struct mem_cgroup *cached; /* this never be root cgroup */ |
Johannes Weiner | 11c9ea4 | 2011-03-23 16:42:34 -0700 | [diff] [blame] | 2195 | unsigned int nr_pages; |
Roman Gushchin | bf4f059 | 2020-08-06 23:20:49 -0700 | [diff] [blame] | 2196 | |
| 2197 | #ifdef CONFIG_MEMCG_KMEM |
| 2198 | struct obj_cgroup *cached_objcg; |
| 2199 | unsigned int nr_bytes; |
| 2200 | #endif |
| 2201 | |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2202 | struct work_struct work; |
KAMEZAWA Hiroyuki | 26fe616 | 2011-06-15 15:08:45 -0700 | [diff] [blame] | 2203 | unsigned long flags; |
Kirill A. Shutemov | a0db00f | 2012-05-29 15:06:56 -0700 | [diff] [blame] | 2204 | #define FLUSHING_CACHED_CHARGE 0 |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2205 | }; |
| 2206 | static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock); |
Michal Hocko | 9f50fad | 2011-08-09 11:56:26 +0200 | [diff] [blame] | 2207 | static DEFINE_MUTEX(percpu_charge_mutex); |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2208 | |
Roman Gushchin | bf4f059 | 2020-08-06 23:20:49 -0700 | [diff] [blame] | 2209 | #ifdef CONFIG_MEMCG_KMEM |
| 2210 | static void drain_obj_stock(struct memcg_stock_pcp *stock); |
| 2211 | static bool obj_stock_flush_required(struct memcg_stock_pcp *stock, |
| 2212 | struct mem_cgroup *root_memcg); |
| 2213 | |
| 2214 | #else |
| 2215 | static inline void drain_obj_stock(struct memcg_stock_pcp *stock) |
| 2216 | { |
| 2217 | } |
| 2218 | static bool obj_stock_flush_required(struct memcg_stock_pcp *stock, |
| 2219 | struct mem_cgroup *root_memcg) |
| 2220 | { |
| 2221 | return false; |
| 2222 | } |
| 2223 | #endif |
| 2224 | |
Suleiman Souhlal | a0956d5 | 2012-12-18 14:21:36 -0800 | [diff] [blame] | 2225 | /** |
| 2226 | * consume_stock: Try to consume stocked charge on this cpu. |
| 2227 | * @memcg: memcg to consume from. |
| 2228 | * @nr_pages: how many pages to charge. |
| 2229 | * |
| 2230 | * The charges will only happen if @memcg matches the current cpu's memcg |
| 2231 | * stock, and at least @nr_pages are available in that stock. Failure to |
| 2232 | * service an allocation will refill the stock. |
| 2233 | * |
| 2234 | * returns true if successful, false otherwise. |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2235 | */ |
Suleiman Souhlal | a0956d5 | 2012-12-18 14:21:36 -0800 | [diff] [blame] | 2236 | static bool consume_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2237 | { |
| 2238 | struct memcg_stock_pcp *stock; |
Johannes Weiner | db2ba40c | 2016-09-19 14:44:36 -0700 | [diff] [blame] | 2239 | unsigned long flags; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2240 | bool ret = false; |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2241 | |
Johannes Weiner | a983b5e | 2018-01-31 16:16:45 -0800 | [diff] [blame] | 2242 | if (nr_pages > MEMCG_CHARGE_BATCH) |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2243 | return ret; |
Suleiman Souhlal | a0956d5 | 2012-12-18 14:21:36 -0800 | [diff] [blame] | 2244 | |
Johannes Weiner | db2ba40c | 2016-09-19 14:44:36 -0700 | [diff] [blame] | 2245 | local_irq_save(flags); |
| 2246 | |
| 2247 | stock = this_cpu_ptr(&memcg_stock); |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2248 | if (memcg == stock->cached && stock->nr_pages >= nr_pages) { |
Suleiman Souhlal | a0956d5 | 2012-12-18 14:21:36 -0800 | [diff] [blame] | 2249 | stock->nr_pages -= nr_pages; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2250 | ret = true; |
| 2251 | } |
Johannes Weiner | db2ba40c | 2016-09-19 14:44:36 -0700 | [diff] [blame] | 2252 | |
| 2253 | local_irq_restore(flags); |
| 2254 | |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2255 | return ret; |
| 2256 | } |
| 2257 | |
| 2258 | /* |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2259 | * Returns stocks cached in percpu and reset cached information. |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2260 | */ |
| 2261 | static void drain_stock(struct memcg_stock_pcp *stock) |
| 2262 | { |
| 2263 | struct mem_cgroup *old = stock->cached; |
| 2264 | |
Johannes Weiner | 1a3e1f4 | 2020-08-06 23:20:45 -0700 | [diff] [blame] | 2265 | if (!old) |
| 2266 | return; |
| 2267 | |
Johannes Weiner | 11c9ea4 | 2011-03-23 16:42:34 -0700 | [diff] [blame] | 2268 | if (stock->nr_pages) { |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2269 | page_counter_uncharge(&old->memory, stock->nr_pages); |
Johannes Weiner | 7941d21 | 2016-01-14 15:21:23 -0800 | [diff] [blame] | 2270 | if (do_memsw_account()) |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2271 | page_counter_uncharge(&old->memsw, stock->nr_pages); |
Johannes Weiner | 11c9ea4 | 2011-03-23 16:42:34 -0700 | [diff] [blame] | 2272 | stock->nr_pages = 0; |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2273 | } |
Johannes Weiner | 1a3e1f4 | 2020-08-06 23:20:45 -0700 | [diff] [blame] | 2274 | |
| 2275 | css_put(&old->css); |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2276 | stock->cached = NULL; |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2277 | } |
| 2278 | |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2279 | static void drain_local_stock(struct work_struct *dummy) |
| 2280 | { |
Johannes Weiner | db2ba40c | 2016-09-19 14:44:36 -0700 | [diff] [blame] | 2281 | struct memcg_stock_pcp *stock; |
| 2282 | unsigned long flags; |
| 2283 | |
Michal Hocko | 72f0184 | 2017-10-03 16:14:53 -0700 | [diff] [blame] | 2284 | /* |
| 2285 | * The only protection from memory hotplug vs. drain_stock races is |
| 2286 | * that we always operate on local CPU stock here with IRQ disabled |
| 2287 | */ |
Johannes Weiner | db2ba40c | 2016-09-19 14:44:36 -0700 | [diff] [blame] | 2288 | local_irq_save(flags); |
| 2289 | |
| 2290 | stock = this_cpu_ptr(&memcg_stock); |
Roman Gushchin | bf4f059 | 2020-08-06 23:20:49 -0700 | [diff] [blame] | 2291 | drain_obj_stock(stock); |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2292 | drain_stock(stock); |
KAMEZAWA Hiroyuki | 26fe616 | 2011-06-15 15:08:45 -0700 | [diff] [blame] | 2293 | clear_bit(FLUSHING_CACHED_CHARGE, &stock->flags); |
Johannes Weiner | db2ba40c | 2016-09-19 14:44:36 -0700 | [diff] [blame] | 2294 | |
| 2295 | local_irq_restore(flags); |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2296 | } |
| 2297 | |
| 2298 | /* |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2299 | * Cache charges(val) to local per_cpu area. |
Greg Thelen | 320cc51 | 2010-03-15 15:27:28 +0100 | [diff] [blame] | 2300 | * This will be consumed by consume_stock() function, later. |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2301 | */ |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2302 | static void refill_stock(struct mem_cgroup *memcg, unsigned int nr_pages) |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2303 | { |
Johannes Weiner | db2ba40c | 2016-09-19 14:44:36 -0700 | [diff] [blame] | 2304 | struct memcg_stock_pcp *stock; |
| 2305 | unsigned long flags; |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2306 | |
Johannes Weiner | db2ba40c | 2016-09-19 14:44:36 -0700 | [diff] [blame] | 2307 | local_irq_save(flags); |
| 2308 | |
| 2309 | stock = this_cpu_ptr(&memcg_stock); |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2310 | if (stock->cached != memcg) { /* reset if necessary */ |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2311 | drain_stock(stock); |
Johannes Weiner | 1a3e1f4 | 2020-08-06 23:20:45 -0700 | [diff] [blame] | 2312 | css_get(&memcg->css); |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2313 | stock->cached = memcg; |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2314 | } |
Johannes Weiner | 11c9ea4 | 2011-03-23 16:42:34 -0700 | [diff] [blame] | 2315 | stock->nr_pages += nr_pages; |
Johannes Weiner | db2ba40c | 2016-09-19 14:44:36 -0700 | [diff] [blame] | 2316 | |
Johannes Weiner | a983b5e | 2018-01-31 16:16:45 -0800 | [diff] [blame] | 2317 | if (stock->nr_pages > MEMCG_CHARGE_BATCH) |
Roman Gushchin | 475d048 | 2017-09-08 16:13:09 -0700 | [diff] [blame] | 2318 | drain_stock(stock); |
| 2319 | |
Johannes Weiner | db2ba40c | 2016-09-19 14:44:36 -0700 | [diff] [blame] | 2320 | local_irq_restore(flags); |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2321 | } |
| 2322 | |
| 2323 | /* |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2324 | * Drains all per-CPU charge caches for given root_memcg resp. subtree |
Johannes Weiner | 6d3d6aa | 2014-12-10 15:42:50 -0800 | [diff] [blame] | 2325 | * of the hierarchy under it. |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2326 | */ |
Johannes Weiner | 6d3d6aa | 2014-12-10 15:42:50 -0800 | [diff] [blame] | 2327 | static void drain_all_stock(struct mem_cgroup *root_memcg) |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2328 | { |
KAMEZAWA Hiroyuki | 26fe616 | 2011-06-15 15:08:45 -0700 | [diff] [blame] | 2329 | int cpu, curcpu; |
Michal Hocko | d38144b | 2011-07-26 16:08:28 -0700 | [diff] [blame] | 2330 | |
Johannes Weiner | 6d3d6aa | 2014-12-10 15:42:50 -0800 | [diff] [blame] | 2331 | /* If someone's already draining, avoid adding running more workers. */ |
| 2332 | if (!mutex_trylock(&percpu_charge_mutex)) |
| 2333 | return; |
Michal Hocko | 72f0184 | 2017-10-03 16:14:53 -0700 | [diff] [blame] | 2334 | /* |
| 2335 | * Notify other cpus that system-wide "drain" is running |
| 2336 | * We do not care about races with the cpu hotplug because cpu down |
| 2337 | * as well as workers from this path always operate on the local |
| 2338 | * per-cpu data. CPU up doesn't touch memcg_stock at all. |
| 2339 | */ |
Johannes Weiner | 5af12d0 | 2011-08-25 15:59:07 -0700 | [diff] [blame] | 2340 | curcpu = get_cpu(); |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2341 | for_each_online_cpu(cpu) { |
| 2342 | struct memcg_stock_pcp *stock = &per_cpu(memcg_stock, cpu); |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2343 | struct mem_cgroup *memcg; |
Roman Gushchin | e1a366b | 2019-09-23 15:34:58 -0700 | [diff] [blame] | 2344 | bool flush = false; |
KAMEZAWA Hiroyuki | 26fe616 | 2011-06-15 15:08:45 -0700 | [diff] [blame] | 2345 | |
Roman Gushchin | e1a366b | 2019-09-23 15:34:58 -0700 | [diff] [blame] | 2346 | rcu_read_lock(); |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 2347 | memcg = stock->cached; |
Roman Gushchin | e1a366b | 2019-09-23 15:34:58 -0700 | [diff] [blame] | 2348 | if (memcg && stock->nr_pages && |
| 2349 | mem_cgroup_is_descendant(memcg, root_memcg)) |
| 2350 | flush = true; |
Roman Gushchin | bf4f059 | 2020-08-06 23:20:49 -0700 | [diff] [blame] | 2351 | if (obj_stock_flush_required(stock, root_memcg)) |
| 2352 | flush = true; |
Roman Gushchin | e1a366b | 2019-09-23 15:34:58 -0700 | [diff] [blame] | 2353 | rcu_read_unlock(); |
| 2354 | |
| 2355 | if (flush && |
| 2356 | !test_and_set_bit(FLUSHING_CACHED_CHARGE, &stock->flags)) { |
Michal Hocko | d1a05b6 | 2011-07-26 16:08:27 -0700 | [diff] [blame] | 2357 | if (cpu == curcpu) |
| 2358 | drain_local_stock(&stock->work); |
| 2359 | else |
| 2360 | schedule_work_on(cpu, &stock->work); |
| 2361 | } |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2362 | } |
Johannes Weiner | 5af12d0 | 2011-08-25 15:59:07 -0700 | [diff] [blame] | 2363 | put_cpu(); |
Michal Hocko | 9f50fad | 2011-08-09 11:56:26 +0200 | [diff] [blame] | 2364 | mutex_unlock(&percpu_charge_mutex); |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2365 | } |
| 2366 | |
Johannes Weiner | 2cd21c8 | 2021-04-29 22:56:29 -0700 | [diff] [blame] | 2367 | static void memcg_flush_lruvec_page_state(struct mem_cgroup *memcg, int cpu) |
| 2368 | { |
| 2369 | int nid; |
| 2370 | |
| 2371 | for_each_node(nid) { |
| 2372 | struct mem_cgroup_per_node *pn = memcg->nodeinfo[nid]; |
| 2373 | unsigned long stat[NR_VM_NODE_STAT_ITEMS]; |
| 2374 | struct batched_lruvec_stat *lstatc; |
| 2375 | int i; |
| 2376 | |
| 2377 | lstatc = per_cpu_ptr(pn->lruvec_stat_cpu, cpu); |
| 2378 | for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) { |
| 2379 | stat[i] = lstatc->count[i]; |
| 2380 | lstatc->count[i] = 0; |
| 2381 | } |
| 2382 | |
| 2383 | do { |
| 2384 | for (i = 0; i < NR_VM_NODE_STAT_ITEMS; i++) |
| 2385 | atomic_long_add(stat[i], &pn->lruvec_stat[i]); |
| 2386 | } while ((pn = parent_nodeinfo(pn, nid))); |
| 2387 | } |
| 2388 | } |
| 2389 | |
Sebastian Andrzej Siewior | 308167f | 2016-11-03 15:49:59 +0100 | [diff] [blame] | 2390 | static int memcg_hotplug_cpu_dead(unsigned int cpu) |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2391 | { |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2392 | struct memcg_stock_pcp *stock; |
Johannes Weiner | a3d4c05 | 2021-04-29 22:56:11 -0700 | [diff] [blame] | 2393 | struct mem_cgroup *memcg; |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2394 | |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2395 | stock = &per_cpu(memcg_stock, cpu); |
| 2396 | drain_stock(stock); |
Johannes Weiner | a983b5e | 2018-01-31 16:16:45 -0800 | [diff] [blame] | 2397 | |
Johannes Weiner | 2cd21c8 | 2021-04-29 22:56:29 -0700 | [diff] [blame] | 2398 | for_each_mem_cgroup(memcg) |
| 2399 | memcg_flush_lruvec_page_state(memcg, cpu); |
Johannes Weiner | a983b5e | 2018-01-31 16:16:45 -0800 | [diff] [blame] | 2400 | |
Sebastian Andrzej Siewior | 308167f | 2016-11-03 15:49:59 +0100 | [diff] [blame] | 2401 | return 0; |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2402 | } |
| 2403 | |
Chris Down | b3ff929 | 2020-08-06 23:21:54 -0700 | [diff] [blame] | 2404 | static unsigned long reclaim_high(struct mem_cgroup *memcg, |
| 2405 | unsigned int nr_pages, |
| 2406 | gfp_t gfp_mask) |
Johannes Weiner | f7e1cb6 | 2016-01-14 15:21:29 -0800 | [diff] [blame] | 2407 | { |
Chris Down | b3ff929 | 2020-08-06 23:21:54 -0700 | [diff] [blame] | 2408 | unsigned long nr_reclaimed = 0; |
| 2409 | |
Johannes Weiner | f7e1cb6 | 2016-01-14 15:21:29 -0800 | [diff] [blame] | 2410 | do { |
Johannes Weiner | e22c6ed | 2020-08-06 23:22:15 -0700 | [diff] [blame] | 2411 | unsigned long pflags; |
| 2412 | |
Jakub Kicinski | d1663a9 | 2020-06-01 21:49:49 -0700 | [diff] [blame] | 2413 | if (page_counter_read(&memcg->memory) <= |
| 2414 | READ_ONCE(memcg->memory.high)) |
Johannes Weiner | f7e1cb6 | 2016-01-14 15:21:29 -0800 | [diff] [blame] | 2415 | continue; |
Johannes Weiner | e22c6ed | 2020-08-06 23:22:15 -0700 | [diff] [blame] | 2416 | |
Johannes Weiner | e27be24 | 2018-04-10 16:29:45 -0700 | [diff] [blame] | 2417 | memcg_memory_event(memcg, MEMCG_HIGH); |
Johannes Weiner | e22c6ed | 2020-08-06 23:22:15 -0700 | [diff] [blame] | 2418 | |
| 2419 | psi_memstall_enter(&pflags); |
Chris Down | b3ff929 | 2020-08-06 23:21:54 -0700 | [diff] [blame] | 2420 | nr_reclaimed += try_to_free_mem_cgroup_pages(memcg, nr_pages, |
| 2421 | gfp_mask, true); |
Johannes Weiner | e22c6ed | 2020-08-06 23:22:15 -0700 | [diff] [blame] | 2422 | psi_memstall_leave(&pflags); |
Chris Down | 4bf1730 | 2020-04-06 20:03:30 -0700 | [diff] [blame] | 2423 | } while ((memcg = parent_mem_cgroup(memcg)) && |
| 2424 | !mem_cgroup_is_root(memcg)); |
Chris Down | b3ff929 | 2020-08-06 23:21:54 -0700 | [diff] [blame] | 2425 | |
| 2426 | return nr_reclaimed; |
Johannes Weiner | f7e1cb6 | 2016-01-14 15:21:29 -0800 | [diff] [blame] | 2427 | } |
| 2428 | |
| 2429 | static void high_work_func(struct work_struct *work) |
| 2430 | { |
| 2431 | struct mem_cgroup *memcg; |
| 2432 | |
| 2433 | memcg = container_of(work, struct mem_cgroup, high_work); |
Johannes Weiner | a983b5e | 2018-01-31 16:16:45 -0800 | [diff] [blame] | 2434 | reclaim_high(memcg, MEMCG_CHARGE_BATCH, GFP_KERNEL); |
Johannes Weiner | f7e1cb6 | 2016-01-14 15:21:29 -0800 | [diff] [blame] | 2435 | } |
| 2436 | |
Tejun Heo | b23afb9 | 2015-11-05 18:46:11 -0800 | [diff] [blame] | 2437 | /* |
Chris Down | 0e4b01d | 2019-09-23 15:34:55 -0700 | [diff] [blame] | 2438 | * Clamp the maximum sleep time per allocation batch to 2 seconds. This is |
| 2439 | * enough to still cause a significant slowdown in most cases, while still |
| 2440 | * allowing diagnostics and tracing to proceed without becoming stuck. |
| 2441 | */ |
| 2442 | #define MEMCG_MAX_HIGH_DELAY_JIFFIES (2UL*HZ) |
| 2443 | |
| 2444 | /* |
| 2445 | * When calculating the delay, we use these either side of the exponentiation to |
| 2446 | * maintain precision and scale to a reasonable number of jiffies (see the table |
| 2447 | * below. |
| 2448 | * |
| 2449 | * - MEMCG_DELAY_PRECISION_SHIFT: Extra precision bits while translating the |
| 2450 | * overage ratio to a delay. |
Randy Dunlap | ac5ddd0 | 2020-08-11 18:33:02 -0700 | [diff] [blame] | 2451 | * - MEMCG_DELAY_SCALING_SHIFT: The number of bits to scale down the |
Chris Down | 0e4b01d | 2019-09-23 15:34:55 -0700 | [diff] [blame] | 2452 | * proposed penalty in order to reduce to a reasonable number of jiffies, and |
| 2453 | * to produce a reasonable delay curve. |
| 2454 | * |
| 2455 | * MEMCG_DELAY_SCALING_SHIFT just happens to be a number that produces a |
| 2456 | * reasonable delay curve compared to precision-adjusted overage, not |
| 2457 | * penalising heavily at first, but still making sure that growth beyond the |
| 2458 | * limit penalises misbehaviour cgroups by slowing them down exponentially. For |
| 2459 | * example, with a high of 100 megabytes: |
| 2460 | * |
| 2461 | * +-------+------------------------+ |
| 2462 | * | usage | time to allocate in ms | |
| 2463 | * +-------+------------------------+ |
| 2464 | * | 100M | 0 | |
| 2465 | * | 101M | 6 | |
| 2466 | * | 102M | 25 | |
| 2467 | * | 103M | 57 | |
| 2468 | * | 104M | 102 | |
| 2469 | * | 105M | 159 | |
| 2470 | * | 106M | 230 | |
| 2471 | * | 107M | 313 | |
| 2472 | * | 108M | 409 | |
| 2473 | * | 109M | 518 | |
| 2474 | * | 110M | 639 | |
| 2475 | * | 111M | 774 | |
| 2476 | * | 112M | 921 | |
| 2477 | * | 113M | 1081 | |
| 2478 | * | 114M | 1254 | |
| 2479 | * | 115M | 1439 | |
| 2480 | * | 116M | 1638 | |
| 2481 | * | 117M | 1849 | |
| 2482 | * | 118M | 2000 | |
| 2483 | * | 119M | 2000 | |
| 2484 | * | 120M | 2000 | |
| 2485 | * +-------+------------------------+ |
| 2486 | */ |
| 2487 | #define MEMCG_DELAY_PRECISION_SHIFT 20 |
| 2488 | #define MEMCG_DELAY_SCALING_SHIFT 14 |
| 2489 | |
Jakub Kicinski | 8a5dbc6 | 2020-06-01 21:49:42 -0700 | [diff] [blame] | 2490 | static u64 calculate_overage(unsigned long usage, unsigned long high) |
| 2491 | { |
| 2492 | u64 overage; |
| 2493 | |
| 2494 | if (usage <= high) |
| 2495 | return 0; |
| 2496 | |
| 2497 | /* |
| 2498 | * Prevent division by 0 in overage calculation by acting as if |
| 2499 | * it was a threshold of 1 page |
| 2500 | */ |
| 2501 | high = max(high, 1UL); |
| 2502 | |
| 2503 | overage = usage - high; |
| 2504 | overage <<= MEMCG_DELAY_PRECISION_SHIFT; |
| 2505 | return div64_u64(overage, high); |
| 2506 | } |
| 2507 | |
| 2508 | static u64 mem_find_max_overage(struct mem_cgroup *memcg) |
| 2509 | { |
| 2510 | u64 overage, max_overage = 0; |
| 2511 | |
| 2512 | do { |
| 2513 | overage = calculate_overage(page_counter_read(&memcg->memory), |
Jakub Kicinski | d1663a9 | 2020-06-01 21:49:49 -0700 | [diff] [blame] | 2514 | READ_ONCE(memcg->memory.high)); |
Jakub Kicinski | 8a5dbc6 | 2020-06-01 21:49:42 -0700 | [diff] [blame] | 2515 | max_overage = max(overage, max_overage); |
| 2516 | } while ((memcg = parent_mem_cgroup(memcg)) && |
| 2517 | !mem_cgroup_is_root(memcg)); |
| 2518 | |
| 2519 | return max_overage; |
| 2520 | } |
| 2521 | |
Jakub Kicinski | 4b82ab4 | 2020-06-01 21:49:52 -0700 | [diff] [blame] | 2522 | static u64 swap_find_max_overage(struct mem_cgroup *memcg) |
| 2523 | { |
| 2524 | u64 overage, max_overage = 0; |
| 2525 | |
| 2526 | do { |
| 2527 | overage = calculate_overage(page_counter_read(&memcg->swap), |
| 2528 | READ_ONCE(memcg->swap.high)); |
| 2529 | if (overage) |
| 2530 | memcg_memory_event(memcg, MEMCG_SWAP_HIGH); |
| 2531 | max_overage = max(overage, max_overage); |
| 2532 | } while ((memcg = parent_mem_cgroup(memcg)) && |
| 2533 | !mem_cgroup_is_root(memcg)); |
| 2534 | |
| 2535 | return max_overage; |
| 2536 | } |
| 2537 | |
Chris Down | 0e4b01d | 2019-09-23 15:34:55 -0700 | [diff] [blame] | 2538 | /* |
Chris Down | e26733e | 2020-03-21 18:22:23 -0700 | [diff] [blame] | 2539 | * Get the number of jiffies that we should penalise a mischievous cgroup which |
| 2540 | * is exceeding its memory.high by checking both it and its ancestors. |
Tejun Heo | b23afb9 | 2015-11-05 18:46:11 -0800 | [diff] [blame] | 2541 | */ |
Chris Down | e26733e | 2020-03-21 18:22:23 -0700 | [diff] [blame] | 2542 | static unsigned long calculate_high_delay(struct mem_cgroup *memcg, |
Jakub Kicinski | 8a5dbc6 | 2020-06-01 21:49:42 -0700 | [diff] [blame] | 2543 | unsigned int nr_pages, |
| 2544 | u64 max_overage) |
Tejun Heo | b23afb9 | 2015-11-05 18:46:11 -0800 | [diff] [blame] | 2545 | { |
Chris Down | e26733e | 2020-03-21 18:22:23 -0700 | [diff] [blame] | 2546 | unsigned long penalty_jiffies; |
Chris Down | e26733e | 2020-03-21 18:22:23 -0700 | [diff] [blame] | 2547 | |
| 2548 | if (!max_overage) |
| 2549 | return 0; |
Chris Down | 0e4b01d | 2019-09-23 15:34:55 -0700 | [diff] [blame] | 2550 | |
| 2551 | /* |
Chris Down | 0e4b01d | 2019-09-23 15:34:55 -0700 | [diff] [blame] | 2552 | * We use overage compared to memory.high to calculate the number of |
| 2553 | * jiffies to sleep (penalty_jiffies). Ideally this value should be |
| 2554 | * fairly lenient on small overages, and increasingly harsh when the |
| 2555 | * memcg in question makes it clear that it has no intention of stopping |
| 2556 | * its crazy behaviour, so we exponentially increase the delay based on |
| 2557 | * overage amount. |
| 2558 | */ |
Chris Down | e26733e | 2020-03-21 18:22:23 -0700 | [diff] [blame] | 2559 | penalty_jiffies = max_overage * max_overage * HZ; |
| 2560 | penalty_jiffies >>= MEMCG_DELAY_PRECISION_SHIFT; |
| 2561 | penalty_jiffies >>= MEMCG_DELAY_SCALING_SHIFT; |
Chris Down | 0e4b01d | 2019-09-23 15:34:55 -0700 | [diff] [blame] | 2562 | |
| 2563 | /* |
| 2564 | * Factor in the task's own contribution to the overage, such that four |
| 2565 | * N-sized allocations are throttled approximately the same as one |
| 2566 | * 4N-sized allocation. |
| 2567 | * |
| 2568 | * MEMCG_CHARGE_BATCH pages is nominal, so work out how much smaller or |
| 2569 | * larger the current charge patch is than that. |
| 2570 | */ |
Jakub Kicinski | ff144e6 | 2020-06-01 21:49:45 -0700 | [diff] [blame] | 2571 | return penalty_jiffies * nr_pages / MEMCG_CHARGE_BATCH; |
Chris Down | e26733e | 2020-03-21 18:22:23 -0700 | [diff] [blame] | 2572 | } |
| 2573 | |
| 2574 | /* |
| 2575 | * Scheduled by try_charge() to be executed from the userland return path |
| 2576 | * and reclaims memory over the high limit. |
| 2577 | */ |
| 2578 | void mem_cgroup_handle_over_high(void) |
| 2579 | { |
| 2580 | unsigned long penalty_jiffies; |
| 2581 | unsigned long pflags; |
Chris Down | b3ff929 | 2020-08-06 23:21:54 -0700 | [diff] [blame] | 2582 | unsigned long nr_reclaimed; |
Chris Down | e26733e | 2020-03-21 18:22:23 -0700 | [diff] [blame] | 2583 | unsigned int nr_pages = current->memcg_nr_pages_over_high; |
Chris Down | d977aa9 | 2020-08-06 23:21:58 -0700 | [diff] [blame] | 2584 | int nr_retries = MAX_RECLAIM_RETRIES; |
Chris Down | e26733e | 2020-03-21 18:22:23 -0700 | [diff] [blame] | 2585 | struct mem_cgroup *memcg; |
Chris Down | b3ff929 | 2020-08-06 23:21:54 -0700 | [diff] [blame] | 2586 | bool in_retry = false; |
Chris Down | e26733e | 2020-03-21 18:22:23 -0700 | [diff] [blame] | 2587 | |
| 2588 | if (likely(!nr_pages)) |
| 2589 | return; |
| 2590 | |
| 2591 | memcg = get_mem_cgroup_from_mm(current->mm); |
Chris Down | e26733e | 2020-03-21 18:22:23 -0700 | [diff] [blame] | 2592 | current->memcg_nr_pages_over_high = 0; |
| 2593 | |
Chris Down | b3ff929 | 2020-08-06 23:21:54 -0700 | [diff] [blame] | 2594 | retry_reclaim: |
| 2595 | /* |
| 2596 | * The allocating task should reclaim at least the batch size, but for |
| 2597 | * subsequent retries we only want to do what's necessary to prevent oom |
| 2598 | * or breaching resource isolation. |
| 2599 | * |
| 2600 | * This is distinct from memory.max or page allocator behaviour because |
| 2601 | * memory.high is currently batched, whereas memory.max and the page |
| 2602 | * allocator run every time an allocation is made. |
| 2603 | */ |
| 2604 | nr_reclaimed = reclaim_high(memcg, |
| 2605 | in_retry ? SWAP_CLUSTER_MAX : nr_pages, |
| 2606 | GFP_KERNEL); |
| 2607 | |
Chris Down | e26733e | 2020-03-21 18:22:23 -0700 | [diff] [blame] | 2608 | /* |
| 2609 | * memory.high is breached and reclaim is unable to keep up. Throttle |
| 2610 | * allocators proactively to slow down excessive growth. |
| 2611 | */ |
Jakub Kicinski | 8a5dbc6 | 2020-06-01 21:49:42 -0700 | [diff] [blame] | 2612 | penalty_jiffies = calculate_high_delay(memcg, nr_pages, |
| 2613 | mem_find_max_overage(memcg)); |
Chris Down | 0e4b01d | 2019-09-23 15:34:55 -0700 | [diff] [blame] | 2614 | |
Jakub Kicinski | 4b82ab4 | 2020-06-01 21:49:52 -0700 | [diff] [blame] | 2615 | penalty_jiffies += calculate_high_delay(memcg, nr_pages, |
| 2616 | swap_find_max_overage(memcg)); |
| 2617 | |
Chris Down | 0e4b01d | 2019-09-23 15:34:55 -0700 | [diff] [blame] | 2618 | /* |
Jakub Kicinski | ff144e6 | 2020-06-01 21:49:45 -0700 | [diff] [blame] | 2619 | * Clamp the max delay per usermode return so as to still keep the |
| 2620 | * application moving forwards and also permit diagnostics, albeit |
| 2621 | * extremely slowly. |
| 2622 | */ |
| 2623 | penalty_jiffies = min(penalty_jiffies, MEMCG_MAX_HIGH_DELAY_JIFFIES); |
| 2624 | |
| 2625 | /* |
Chris Down | 0e4b01d | 2019-09-23 15:34:55 -0700 | [diff] [blame] | 2626 | * Don't sleep if the amount of jiffies this memcg owes us is so low |
| 2627 | * that it's not even worth doing, in an attempt to be nice to those who |
| 2628 | * go only a small amount over their memory.high value and maybe haven't |
| 2629 | * been aggressively reclaimed enough yet. |
| 2630 | */ |
| 2631 | if (penalty_jiffies <= HZ / 100) |
| 2632 | goto out; |
| 2633 | |
| 2634 | /* |
Chris Down | b3ff929 | 2020-08-06 23:21:54 -0700 | [diff] [blame] | 2635 | * If reclaim is making forward progress but we're still over |
| 2636 | * memory.high, we want to encourage that rather than doing allocator |
| 2637 | * throttling. |
| 2638 | */ |
| 2639 | if (nr_reclaimed || nr_retries--) { |
| 2640 | in_retry = true; |
| 2641 | goto retry_reclaim; |
| 2642 | } |
| 2643 | |
| 2644 | /* |
Chris Down | 0e4b01d | 2019-09-23 15:34:55 -0700 | [diff] [blame] | 2645 | * If we exit early, we're guaranteed to die (since |
| 2646 | * schedule_timeout_killable sets TASK_KILLABLE). This means we don't |
| 2647 | * need to account for any ill-begotten jiffies to pay them off later. |
| 2648 | */ |
| 2649 | psi_memstall_enter(&pflags); |
| 2650 | schedule_timeout_killable(penalty_jiffies); |
| 2651 | psi_memstall_leave(&pflags); |
| 2652 | |
| 2653 | out: |
| 2654 | css_put(&memcg->css); |
Tejun Heo | b23afb9 | 2015-11-05 18:46:11 -0800 | [diff] [blame] | 2655 | } |
| 2656 | |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 2657 | static int try_charge(struct mem_cgroup *memcg, gfp_t gfp_mask, |
| 2658 | unsigned int nr_pages) |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 2659 | { |
Johannes Weiner | a983b5e | 2018-01-31 16:16:45 -0800 | [diff] [blame] | 2660 | unsigned int batch = max(MEMCG_CHARGE_BATCH, nr_pages); |
Chris Down | d977aa9 | 2020-08-06 23:21:58 -0700 | [diff] [blame] | 2661 | int nr_retries = MAX_RECLAIM_RETRIES; |
Johannes Weiner | 6539cc0 | 2014-08-06 16:05:42 -0700 | [diff] [blame] | 2662 | struct mem_cgroup *mem_over_limit; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2663 | struct page_counter *counter; |
Johannes Weiner | e22c6ed | 2020-08-06 23:22:15 -0700 | [diff] [blame] | 2664 | enum oom_status oom_status; |
Johannes Weiner | 6539cc0 | 2014-08-06 16:05:42 -0700 | [diff] [blame] | 2665 | unsigned long nr_reclaimed; |
Johannes Weiner | b70a2a2 | 2014-10-09 15:28:56 -0700 | [diff] [blame] | 2666 | bool may_swap = true; |
| 2667 | bool drained = false; |
Johannes Weiner | e22c6ed | 2020-08-06 23:22:15 -0700 | [diff] [blame] | 2668 | unsigned long pflags; |
KAMEZAWA Hiroyuki | a636b32 | 2009-01-07 18:08:08 -0800 | [diff] [blame] | 2669 | |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 2670 | if (mem_cgroup_is_root(memcg)) |
Tejun Heo | 10d53c7 | 2015-11-05 18:46:17 -0800 | [diff] [blame] | 2671 | return 0; |
Johannes Weiner | 6539cc0 | 2014-08-06 16:05:42 -0700 | [diff] [blame] | 2672 | retry: |
Michal Hocko | b6b6cc7 | 2014-04-07 15:37:44 -0700 | [diff] [blame] | 2673 | if (consume_stock(memcg, nr_pages)) |
Tejun Heo | 10d53c7 | 2015-11-05 18:46:17 -0800 | [diff] [blame] | 2674 | return 0; |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 2675 | |
Johannes Weiner | 7941d21 | 2016-01-14 15:21:23 -0800 | [diff] [blame] | 2676 | if (!do_memsw_account() || |
Johannes Weiner | 6071ca5 | 2015-11-05 18:50:26 -0800 | [diff] [blame] | 2677 | page_counter_try_charge(&memcg->memsw, batch, &counter)) { |
| 2678 | if (page_counter_try_charge(&memcg->memory, batch, &counter)) |
Johannes Weiner | 6539cc0 | 2014-08-06 16:05:42 -0700 | [diff] [blame] | 2679 | goto done_restock; |
Johannes Weiner | 7941d21 | 2016-01-14 15:21:23 -0800 | [diff] [blame] | 2680 | if (do_memsw_account()) |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2681 | page_counter_uncharge(&memcg->memsw, batch); |
| 2682 | mem_over_limit = mem_cgroup_from_counter(counter, memory); |
Johannes Weiner | 3fbe724 | 2014-10-09 15:28:54 -0700 | [diff] [blame] | 2683 | } else { |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2684 | mem_over_limit = mem_cgroup_from_counter(counter, memsw); |
Johannes Weiner | b70a2a2 | 2014-10-09 15:28:56 -0700 | [diff] [blame] | 2685 | may_swap = false; |
Johannes Weiner | 3fbe724 | 2014-10-09 15:28:54 -0700 | [diff] [blame] | 2686 | } |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2687 | |
Johannes Weiner | 6539cc0 | 2014-08-06 16:05:42 -0700 | [diff] [blame] | 2688 | if (batch > nr_pages) { |
| 2689 | batch = nr_pages; |
| 2690 | goto retry; |
| 2691 | } |
KAMEZAWA Hiroyuki | cdec2e4 | 2009-12-15 16:47:08 -0800 | [diff] [blame] | 2692 | |
Johannes Weiner | 06b078f | 2014-08-06 16:05:44 -0700 | [diff] [blame] | 2693 | /* |
Johannes Weiner | 869712f | 2019-11-05 21:17:13 -0800 | [diff] [blame] | 2694 | * Memcg doesn't have a dedicated reserve for atomic |
| 2695 | * allocations. But like the global atomic pool, we need to |
| 2696 | * put the burden of reclaim on regular allocation requests |
| 2697 | * and let these go through as privileged allocations. |
| 2698 | */ |
| 2699 | if (gfp_mask & __GFP_ATOMIC) |
| 2700 | goto force; |
| 2701 | |
| 2702 | /* |
Johannes Weiner | 06b078f | 2014-08-06 16:05:44 -0700 | [diff] [blame] | 2703 | * Unlike in global OOM situations, memcg is not in a physical |
| 2704 | * memory shortage. Allow dying and OOM-killed tasks to |
| 2705 | * bypass the last charges so that they can exit quickly and |
| 2706 | * free their memory. |
| 2707 | */ |
Tetsuo Handa | 7775fac | 2019-03-05 15:46:47 -0800 | [diff] [blame] | 2708 | if (unlikely(should_force_charge())) |
Tejun Heo | 10d53c7 | 2015-11-05 18:46:17 -0800 | [diff] [blame] | 2709 | goto force; |
Johannes Weiner | 06b078f | 2014-08-06 16:05:44 -0700 | [diff] [blame] | 2710 | |
Johannes Weiner | 89a2848 | 2016-10-27 17:46:56 -0700 | [diff] [blame] | 2711 | /* |
| 2712 | * Prevent unbounded recursion when reclaim operations need to |
| 2713 | * allocate memory. This might exceed the limits temporarily, |
| 2714 | * but we prefer facilitating memory reclaim and getting back |
| 2715 | * under the limit over triggering OOM kills in these cases. |
| 2716 | */ |
| 2717 | if (unlikely(current->flags & PF_MEMALLOC)) |
| 2718 | goto force; |
| 2719 | |
Johannes Weiner | 06b078f | 2014-08-06 16:05:44 -0700 | [diff] [blame] | 2720 | if (unlikely(task_in_memcg_oom(current))) |
| 2721 | goto nomem; |
| 2722 | |
Mel Gorman | d0164ad | 2015-11-06 16:28:21 -0800 | [diff] [blame] | 2723 | if (!gfpflags_allow_blocking(gfp_mask)) |
Johannes Weiner | 6539cc0 | 2014-08-06 16:05:42 -0700 | [diff] [blame] | 2724 | goto nomem; |
KAMEZAWA Hiroyuki | 4b53433 | 2010-08-10 18:02:57 -0700 | [diff] [blame] | 2725 | |
Johannes Weiner | e27be24 | 2018-04-10 16:29:45 -0700 | [diff] [blame] | 2726 | memcg_memory_event(mem_over_limit, MEMCG_MAX); |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 2727 | |
Johannes Weiner | e22c6ed | 2020-08-06 23:22:15 -0700 | [diff] [blame] | 2728 | psi_memstall_enter(&pflags); |
Johannes Weiner | b70a2a2 | 2014-10-09 15:28:56 -0700 | [diff] [blame] | 2729 | nr_reclaimed = try_to_free_mem_cgroup_pages(mem_over_limit, nr_pages, |
| 2730 | gfp_mask, may_swap); |
Johannes Weiner | e22c6ed | 2020-08-06 23:22:15 -0700 | [diff] [blame] | 2731 | psi_memstall_leave(&pflags); |
Johannes Weiner | 6539cc0 | 2014-08-06 16:05:42 -0700 | [diff] [blame] | 2732 | |
Johannes Weiner | 61e02c7 | 2014-08-06 16:08:16 -0700 | [diff] [blame] | 2733 | if (mem_cgroup_margin(mem_over_limit) >= nr_pages) |
Johannes Weiner | 6539cc0 | 2014-08-06 16:05:42 -0700 | [diff] [blame] | 2734 | goto retry; |
Johannes Weiner | 28c34c2 | 2014-08-06 16:05:47 -0700 | [diff] [blame] | 2735 | |
Johannes Weiner | b70a2a2 | 2014-10-09 15:28:56 -0700 | [diff] [blame] | 2736 | if (!drained) { |
Johannes Weiner | 6d3d6aa | 2014-12-10 15:42:50 -0800 | [diff] [blame] | 2737 | drain_all_stock(mem_over_limit); |
Johannes Weiner | b70a2a2 | 2014-10-09 15:28:56 -0700 | [diff] [blame] | 2738 | drained = true; |
| 2739 | goto retry; |
| 2740 | } |
| 2741 | |
Johannes Weiner | 28c34c2 | 2014-08-06 16:05:47 -0700 | [diff] [blame] | 2742 | if (gfp_mask & __GFP_NORETRY) |
| 2743 | goto nomem; |
Johannes Weiner | 6539cc0 | 2014-08-06 16:05:42 -0700 | [diff] [blame] | 2744 | /* |
| 2745 | * Even though the limit is exceeded at this point, reclaim |
| 2746 | * may have been able to free some pages. Retry the charge |
| 2747 | * before killing the task. |
| 2748 | * |
| 2749 | * Only for regular pages, though: huge pages are rather |
| 2750 | * unlikely to succeed so close to the limit, and we fall back |
| 2751 | * to regular pages anyway in case of failure. |
| 2752 | */ |
Johannes Weiner | 61e02c7 | 2014-08-06 16:08:16 -0700 | [diff] [blame] | 2753 | if (nr_reclaimed && nr_pages <= (1 << PAGE_ALLOC_COSTLY_ORDER)) |
Johannes Weiner | 6539cc0 | 2014-08-06 16:05:42 -0700 | [diff] [blame] | 2754 | goto retry; |
| 2755 | /* |
| 2756 | * At task move, charge accounts can be doubly counted. So, it's |
| 2757 | * better to wait until the end of task_move if something is going on. |
| 2758 | */ |
| 2759 | if (mem_cgroup_wait_acct_move(mem_over_limit)) |
| 2760 | goto retry; |
| 2761 | |
Johannes Weiner | 9b13061 | 2014-08-06 16:05:51 -0700 | [diff] [blame] | 2762 | if (nr_retries--) |
| 2763 | goto retry; |
| 2764 | |
Shakeel Butt | 38d3849 | 2019-07-11 20:55:48 -0700 | [diff] [blame] | 2765 | if (gfp_mask & __GFP_RETRY_MAYFAIL) |
Michal Hocko | 29ef680 | 2018-08-17 15:47:11 -0700 | [diff] [blame] | 2766 | goto nomem; |
| 2767 | |
Johannes Weiner | 6539cc0 | 2014-08-06 16:05:42 -0700 | [diff] [blame] | 2768 | if (fatal_signal_pending(current)) |
Tejun Heo | 10d53c7 | 2015-11-05 18:46:17 -0800 | [diff] [blame] | 2769 | goto force; |
Johannes Weiner | 6539cc0 | 2014-08-06 16:05:42 -0700 | [diff] [blame] | 2770 | |
Michal Hocko | 29ef680 | 2018-08-17 15:47:11 -0700 | [diff] [blame] | 2771 | /* |
| 2772 | * keep retrying as long as the memcg oom killer is able to make |
| 2773 | * a forward progress or bypass the charge if the oom killer |
| 2774 | * couldn't make any progress. |
| 2775 | */ |
| 2776 | oom_status = mem_cgroup_oom(mem_over_limit, gfp_mask, |
Jerome Marchand | 3608de0 | 2015-11-05 18:47:29 -0800 | [diff] [blame] | 2777 | get_order(nr_pages * PAGE_SIZE)); |
Michal Hocko | 29ef680 | 2018-08-17 15:47:11 -0700 | [diff] [blame] | 2778 | switch (oom_status) { |
| 2779 | case OOM_SUCCESS: |
Chris Down | d977aa9 | 2020-08-06 23:21:58 -0700 | [diff] [blame] | 2780 | nr_retries = MAX_RECLAIM_RETRIES; |
Michal Hocko | 29ef680 | 2018-08-17 15:47:11 -0700 | [diff] [blame] | 2781 | goto retry; |
| 2782 | case OOM_FAILED: |
| 2783 | goto force; |
| 2784 | default: |
| 2785 | goto nomem; |
| 2786 | } |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2787 | nomem: |
Johannes Weiner | 6d1fdc4 | 2014-04-07 15:37:45 -0700 | [diff] [blame] | 2788 | if (!(gfp_mask & __GFP_NOFAIL)) |
Johannes Weiner | 3168ecb | 2013-10-31 16:34:13 -0700 | [diff] [blame] | 2789 | return -ENOMEM; |
Tejun Heo | 10d53c7 | 2015-11-05 18:46:17 -0800 | [diff] [blame] | 2790 | force: |
| 2791 | /* |
| 2792 | * The allocation either can't fail or will lead to more memory |
| 2793 | * being freed very soon. Allow memory usage go over the limit |
| 2794 | * temporarily by force charging it. |
| 2795 | */ |
| 2796 | page_counter_charge(&memcg->memory, nr_pages); |
Johannes Weiner | 7941d21 | 2016-01-14 15:21:23 -0800 | [diff] [blame] | 2797 | if (do_memsw_account()) |
Tejun Heo | 10d53c7 | 2015-11-05 18:46:17 -0800 | [diff] [blame] | 2798 | page_counter_charge(&memcg->memsw, nr_pages); |
Tejun Heo | 10d53c7 | 2015-11-05 18:46:17 -0800 | [diff] [blame] | 2799 | |
| 2800 | return 0; |
Johannes Weiner | 6539cc0 | 2014-08-06 16:05:42 -0700 | [diff] [blame] | 2801 | |
| 2802 | done_restock: |
| 2803 | if (batch > nr_pages) |
| 2804 | refill_stock(memcg, batch - nr_pages); |
Tejun Heo | b23afb9 | 2015-11-05 18:46:11 -0800 | [diff] [blame] | 2805 | |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 2806 | /* |
Tejun Heo | b23afb9 | 2015-11-05 18:46:11 -0800 | [diff] [blame] | 2807 | * If the hierarchy is above the normal consumption range, schedule |
| 2808 | * reclaim on returning to userland. We can perform reclaim here |
Mel Gorman | 71baba4 | 2015-11-06 16:28:28 -0800 | [diff] [blame] | 2809 | * 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] | 2810 | * GFP_KERNEL can consistently be used during reclaim. @memcg is |
| 2811 | * not recorded as it most likely matches current's and won't |
| 2812 | * change in the meantime. As high limit is checked again before |
| 2813 | * reclaim, the cost of mismatch is negligible. |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 2814 | */ |
| 2815 | do { |
Jakub Kicinski | 4b82ab4 | 2020-06-01 21:49:52 -0700 | [diff] [blame] | 2816 | bool mem_high, swap_high; |
| 2817 | |
| 2818 | mem_high = page_counter_read(&memcg->memory) > |
| 2819 | READ_ONCE(memcg->memory.high); |
| 2820 | swap_high = page_counter_read(&memcg->swap) > |
| 2821 | READ_ONCE(memcg->swap.high); |
| 2822 | |
| 2823 | /* Don't bother a random interrupted task */ |
| 2824 | if (in_interrupt()) { |
| 2825 | if (mem_high) { |
Johannes Weiner | f7e1cb6 | 2016-01-14 15:21:29 -0800 | [diff] [blame] | 2826 | schedule_work(&memcg->high_work); |
| 2827 | break; |
| 2828 | } |
Jakub Kicinski | 4b82ab4 | 2020-06-01 21:49:52 -0700 | [diff] [blame] | 2829 | continue; |
| 2830 | } |
| 2831 | |
| 2832 | if (mem_high || swap_high) { |
| 2833 | /* |
| 2834 | * The allocating tasks in this cgroup will need to do |
| 2835 | * reclaim or be throttled to prevent further growth |
| 2836 | * of the memory or swap footprints. |
| 2837 | * |
| 2838 | * Target some best-effort fairness between the tasks, |
| 2839 | * and distribute reclaim work and delay penalties |
| 2840 | * based on how much each task is actually allocating. |
| 2841 | */ |
Vladimir Davydov | 9516a18 | 2015-12-11 13:40:24 -0800 | [diff] [blame] | 2842 | current->memcg_nr_pages_over_high += batch; |
Tejun Heo | b23afb9 | 2015-11-05 18:46:11 -0800 | [diff] [blame] | 2843 | set_notify_resume(current); |
| 2844 | break; |
| 2845 | } |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 2846 | } while ((memcg = parent_mem_cgroup(memcg))); |
Tejun Heo | 10d53c7 | 2015-11-05 18:46:17 -0800 | [diff] [blame] | 2847 | |
| 2848 | return 0; |
KAMEZAWA Hiroyuki | 7a81b88 | 2009-01-07 18:07:48 -0800 | [diff] [blame] | 2849 | } |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 2850 | |
Johannes Weiner | f0e45fb | 2020-06-03 16:02:07 -0700 | [diff] [blame] | 2851 | #if defined(CONFIG_MEMCG_KMEM) || defined(CONFIG_MMU) |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 2852 | static void cancel_charge(struct mem_cgroup *memcg, unsigned int nr_pages) |
Daisuke Nishimura | a3032a2 | 2009-12-15 16:47:10 -0800 | [diff] [blame] | 2853 | { |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 2854 | if (mem_cgroup_is_root(memcg)) |
| 2855 | return; |
| 2856 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2857 | page_counter_uncharge(&memcg->memory, nr_pages); |
Johannes Weiner | 7941d21 | 2016-01-14 15:21:23 -0800 | [diff] [blame] | 2858 | if (do_memsw_account()) |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 2859 | page_counter_uncharge(&memcg->memsw, nr_pages); |
KAMEZAWA Hiroyuki | d01dd17 | 2012-05-29 15:07:03 -0700 | [diff] [blame] | 2860 | } |
Johannes Weiner | f0e45fb | 2020-06-03 16:02:07 -0700 | [diff] [blame] | 2861 | #endif |
KAMEZAWA Hiroyuki | d01dd17 | 2012-05-29 15:07:03 -0700 | [diff] [blame] | 2862 | |
Johannes Weiner | d9eb1ea | 2020-06-03 16:02:24 -0700 | [diff] [blame] | 2863 | static void commit_charge(struct page *page, struct mem_cgroup *memcg) |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 2864 | { |
Roman Gushchin | bcfe06b | 2020-12-01 13:58:27 -0800 | [diff] [blame] | 2865 | VM_BUG_ON_PAGE(page_memcg(page), page); |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 2866 | /* |
Alex Shi | a5eb011 | 2020-12-14 19:06:42 -0800 | [diff] [blame] | 2867 | * Any of the following ensures page's memcg stability: |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 2868 | * |
Johannes Weiner | a0b5b41 | 2020-06-03 16:02:27 -0700 | [diff] [blame] | 2869 | * - the page lock |
| 2870 | * - LRU isolation |
| 2871 | * - lock_page_memcg() |
| 2872 | * - exclusive reference |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 2873 | */ |
Roman Gushchin | bcfe06b | 2020-12-01 13:58:27 -0800 | [diff] [blame] | 2874 | page->memcg_data = (unsigned long)memcg; |
Balbir Singh | 8a9f3cc | 2008-02-07 00:13:53 -0800 | [diff] [blame] | 2875 | } |
| 2876 | |
Kirill Tkhai | 84c07d1 | 2018-08-17 15:47:25 -0700 | [diff] [blame] | 2877 | #ifdef CONFIG_MEMCG_KMEM |
Roman Gushchin | 10befea | 2020-08-06 23:21:27 -0700 | [diff] [blame] | 2878 | int memcg_alloc_page_obj_cgroups(struct page *page, struct kmem_cache *s, |
Roman Gushchin | 2e9bd48 | 2021-02-24 12:03:11 -0800 | [diff] [blame] | 2879 | gfp_t gfp, bool new_page) |
Roman Gushchin | 10befea | 2020-08-06 23:21:27 -0700 | [diff] [blame] | 2880 | { |
| 2881 | unsigned int objects = objs_per_slab_page(s, page); |
Roman Gushchin | 2e9bd48 | 2021-02-24 12:03:11 -0800 | [diff] [blame] | 2882 | unsigned long memcg_data; |
Roman Gushchin | 10befea | 2020-08-06 23:21:27 -0700 | [diff] [blame] | 2883 | void *vec; |
| 2884 | |
| 2885 | vec = kcalloc_node(objects, sizeof(struct obj_cgroup *), gfp, |
| 2886 | page_to_nid(page)); |
| 2887 | if (!vec) |
| 2888 | return -ENOMEM; |
| 2889 | |
Roman Gushchin | 2e9bd48 | 2021-02-24 12:03:11 -0800 | [diff] [blame] | 2890 | memcg_data = (unsigned long) vec | MEMCG_DATA_OBJCGS; |
| 2891 | if (new_page) { |
| 2892 | /* |
| 2893 | * If the slab page is brand new and nobody can yet access |
| 2894 | * it's memcg_data, no synchronization is required and |
| 2895 | * memcg_data can be simply assigned. |
| 2896 | */ |
| 2897 | page->memcg_data = memcg_data; |
| 2898 | } else if (cmpxchg(&page->memcg_data, 0, memcg_data)) { |
| 2899 | /* |
| 2900 | * If the slab page is already in use, somebody can allocate |
| 2901 | * and assign obj_cgroups in parallel. In this case the existing |
| 2902 | * objcg vector should be reused. |
| 2903 | */ |
Roman Gushchin | 10befea | 2020-08-06 23:21:27 -0700 | [diff] [blame] | 2904 | kfree(vec); |
Roman Gushchin | 2e9bd48 | 2021-02-24 12:03:11 -0800 | [diff] [blame] | 2905 | return 0; |
| 2906 | } |
Roman Gushchin | 10befea | 2020-08-06 23:21:27 -0700 | [diff] [blame] | 2907 | |
Roman Gushchin | 2e9bd48 | 2021-02-24 12:03:11 -0800 | [diff] [blame] | 2908 | kmemleak_not_leak(vec); |
Roman Gushchin | 10befea | 2020-08-06 23:21:27 -0700 | [diff] [blame] | 2909 | return 0; |
| 2910 | } |
| 2911 | |
Roman Gushchin | 8380ce4 | 2020-03-28 19:17:25 -0700 | [diff] [blame] | 2912 | /* |
| 2913 | * Returns a pointer to the memory cgroup to which the kernel object is charged. |
| 2914 | * |
Roman Gushchin | bcfe06b | 2020-12-01 13:58:27 -0800 | [diff] [blame] | 2915 | * A passed kernel object can be a slab object or a generic kernel page, so |
| 2916 | * different mechanisms for getting the memory cgroup pointer should be used. |
| 2917 | * In certain cases (e.g. kernel stacks or large kmallocs with SLUB) the caller |
| 2918 | * can not know for sure how the kernel object is implemented. |
| 2919 | * mem_cgroup_from_obj() can be safely used in such cases. |
| 2920 | * |
Roman Gushchin | 8380ce4 | 2020-03-28 19:17:25 -0700 | [diff] [blame] | 2921 | * The caller must ensure the memcg lifetime, e.g. by taking rcu_read_lock(), |
| 2922 | * cgroup_mutex, etc. |
| 2923 | */ |
| 2924 | struct mem_cgroup *mem_cgroup_from_obj(void *p) |
| 2925 | { |
| 2926 | struct page *page; |
| 2927 | |
| 2928 | if (mem_cgroup_disabled()) |
| 2929 | return NULL; |
| 2930 | |
| 2931 | page = virt_to_head_page(p); |
| 2932 | |
| 2933 | /* |
Roman Gushchin | 9855609 | 2020-08-06 23:21:10 -0700 | [diff] [blame] | 2934 | * Slab objects are accounted individually, not per-page. |
| 2935 | * Memcg membership data for each individual object is saved in |
| 2936 | * the page->obj_cgroups. |
Roman Gushchin | 8380ce4 | 2020-03-28 19:17:25 -0700 | [diff] [blame] | 2937 | */ |
Roman Gushchin | 270c6a7 | 2020-12-01 13:58:28 -0800 | [diff] [blame] | 2938 | if (page_objcgs_check(page)) { |
Roman Gushchin | 9855609 | 2020-08-06 23:21:10 -0700 | [diff] [blame] | 2939 | struct obj_cgroup *objcg; |
| 2940 | unsigned int off; |
| 2941 | |
| 2942 | off = obj_to_index(page->slab_cache, page, p); |
Roman Gushchin | 270c6a7 | 2020-12-01 13:58:28 -0800 | [diff] [blame] | 2943 | objcg = page_objcgs(page)[off]; |
Roman Gushchin | 10befea | 2020-08-06 23:21:27 -0700 | [diff] [blame] | 2944 | if (objcg) |
| 2945 | return obj_cgroup_memcg(objcg); |
| 2946 | |
| 2947 | return NULL; |
Roman Gushchin | 9855609 | 2020-08-06 23:21:10 -0700 | [diff] [blame] | 2948 | } |
Roman Gushchin | 8380ce4 | 2020-03-28 19:17:25 -0700 | [diff] [blame] | 2949 | |
Roman Gushchin | bcfe06b | 2020-12-01 13:58:27 -0800 | [diff] [blame] | 2950 | /* |
| 2951 | * page_memcg_check() is used here, because page_has_obj_cgroups() |
| 2952 | * check above could fail because the object cgroups vector wasn't set |
| 2953 | * at that moment, but it can be set concurrently. |
| 2954 | * page_memcg_check(page) will guarantee that a proper memory |
| 2955 | * cgroup pointer or NULL will be returned. |
| 2956 | */ |
| 2957 | return page_memcg_check(page); |
Roman Gushchin | 8380ce4 | 2020-03-28 19:17:25 -0700 | [diff] [blame] | 2958 | } |
| 2959 | |
Roman Gushchin | bf4f059 | 2020-08-06 23:20:49 -0700 | [diff] [blame] | 2960 | __always_inline struct obj_cgroup *get_obj_cgroup_from_current(void) |
| 2961 | { |
| 2962 | struct obj_cgroup *objcg = NULL; |
| 2963 | struct mem_cgroup *memcg; |
| 2964 | |
Roman Gushchin | 279c339 | 2020-10-17 16:13:44 -0700 | [diff] [blame] | 2965 | if (memcg_kmem_bypass()) |
| 2966 | return NULL; |
| 2967 | |
Roman Gushchin | bf4f059 | 2020-08-06 23:20:49 -0700 | [diff] [blame] | 2968 | rcu_read_lock(); |
Roman Gushchin | 37d5985 | 2020-10-17 16:13:50 -0700 | [diff] [blame] | 2969 | if (unlikely(active_memcg())) |
| 2970 | memcg = active_memcg(); |
Roman Gushchin | bf4f059 | 2020-08-06 23:20:49 -0700 | [diff] [blame] | 2971 | else |
| 2972 | memcg = mem_cgroup_from_task(current); |
| 2973 | |
| 2974 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) { |
| 2975 | objcg = rcu_dereference(memcg->objcg); |
| 2976 | if (objcg && obj_cgroup_tryget(objcg)) |
| 2977 | break; |
Muchun Song | 2f7659a | 2020-12-14 19:06:31 -0800 | [diff] [blame] | 2978 | objcg = NULL; |
Roman Gushchin | bf4f059 | 2020-08-06 23:20:49 -0700 | [diff] [blame] | 2979 | } |
| 2980 | rcu_read_unlock(); |
| 2981 | |
| 2982 | return objcg; |
| 2983 | } |
| 2984 | |
Vladimir Davydov | f3bb304 | 2014-10-09 15:28:45 -0700 | [diff] [blame] | 2985 | static int memcg_alloc_cache_id(void) |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 2986 | { |
Vladimir Davydov | f3bb304 | 2014-10-09 15:28:45 -0700 | [diff] [blame] | 2987 | int id, size; |
| 2988 | int err; |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 2989 | |
Vladimir Davydov | dbcf73e | 2015-02-12 14:58:57 -0800 | [diff] [blame] | 2990 | id = ida_simple_get(&memcg_cache_ida, |
Vladimir Davydov | f3bb304 | 2014-10-09 15:28:45 -0700 | [diff] [blame] | 2991 | 0, MEMCG_CACHES_MAX_SIZE, GFP_KERNEL); |
| 2992 | if (id < 0) |
| 2993 | return id; |
| 2994 | |
Vladimir Davydov | dbcf73e | 2015-02-12 14:58:57 -0800 | [diff] [blame] | 2995 | if (id < memcg_nr_cache_ids) |
Vladimir Davydov | f3bb304 | 2014-10-09 15:28:45 -0700 | [diff] [blame] | 2996 | return id; |
| 2997 | |
| 2998 | /* |
| 2999 | * There's no space for the new id in memcg_caches arrays, |
| 3000 | * so we have to grow them. |
| 3001 | */ |
Vladimir Davydov | 05257a1 | 2015-02-12 14:59:01 -0800 | [diff] [blame] | 3002 | down_write(&memcg_cache_ids_sem); |
Vladimir Davydov | f3bb304 | 2014-10-09 15:28:45 -0700 | [diff] [blame] | 3003 | |
| 3004 | size = 2 * (id + 1); |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 3005 | if (size < MEMCG_CACHES_MIN_SIZE) |
| 3006 | size = MEMCG_CACHES_MIN_SIZE; |
| 3007 | else if (size > MEMCG_CACHES_MAX_SIZE) |
| 3008 | size = MEMCG_CACHES_MAX_SIZE; |
| 3009 | |
Roman Gushchin | 9855609 | 2020-08-06 23:21:10 -0700 | [diff] [blame] | 3010 | err = memcg_update_all_list_lrus(size); |
Vladimir Davydov | 60d3fd3 | 2015-02-12 14:59:10 -0800 | [diff] [blame] | 3011 | if (!err) |
Vladimir Davydov | 05257a1 | 2015-02-12 14:59:01 -0800 | [diff] [blame] | 3012 | memcg_nr_cache_ids = size; |
| 3013 | |
| 3014 | up_write(&memcg_cache_ids_sem); |
| 3015 | |
Vladimir Davydov | f3bb304 | 2014-10-09 15:28:45 -0700 | [diff] [blame] | 3016 | if (err) { |
Vladimir Davydov | dbcf73e | 2015-02-12 14:58:57 -0800 | [diff] [blame] | 3017 | ida_simple_remove(&memcg_cache_ida, id); |
Vladimir Davydov | f3bb304 | 2014-10-09 15:28:45 -0700 | [diff] [blame] | 3018 | return err; |
| 3019 | } |
| 3020 | return id; |
| 3021 | } |
| 3022 | |
| 3023 | static void memcg_free_cache_id(int id) |
| 3024 | { |
Vladimir Davydov | dbcf73e | 2015-02-12 14:58:57 -0800 | [diff] [blame] | 3025 | ida_simple_remove(&memcg_cache_ida, id); |
Glauber Costa | 55007d8 | 2012-12-18 14:22:38 -0800 | [diff] [blame] | 3026 | } |
| 3027 | |
Vladimir Davydov | 4526477 | 2016-07-26 15:24:21 -0700 | [diff] [blame] | 3028 | /** |
Roman Gushchin | 4b13f64 | 2020-04-01 21:06:56 -0700 | [diff] [blame] | 3029 | * __memcg_kmem_charge: charge a number of kernel pages to a memcg |
Roman Gushchin | 10eaec2 | 2020-04-01 21:06:39 -0700 | [diff] [blame] | 3030 | * @memcg: memory cgroup to charge |
Vladimir Davydov | 4526477 | 2016-07-26 15:24:21 -0700 | [diff] [blame] | 3031 | * @gfp: reclaim mode |
Roman Gushchin | 92d0510c | 2020-04-01 21:06:49 -0700 | [diff] [blame] | 3032 | * @nr_pages: number of pages to charge |
Vladimir Davydov | 4526477 | 2016-07-26 15:24:21 -0700 | [diff] [blame] | 3033 | * |
| 3034 | * Returns 0 on success, an error code on failure. |
| 3035 | */ |
Roman Gushchin | c1a660d | 2021-02-24 12:03:58 -0800 | [diff] [blame] | 3036 | static int __memcg_kmem_charge(struct mem_cgroup *memcg, gfp_t gfp, |
| 3037 | unsigned int nr_pages) |
Vladimir Davydov | f3ccb2c4 | 2015-11-05 18:49:01 -0800 | [diff] [blame] | 3038 | { |
Vladimir Davydov | f3ccb2c4 | 2015-11-05 18:49:01 -0800 | [diff] [blame] | 3039 | struct page_counter *counter; |
Johannes Weiner | 6071ca5 | 2015-11-05 18:50:26 -0800 | [diff] [blame] | 3040 | int ret; |
Vladimir Davydov | f3ccb2c4 | 2015-11-05 18:49:01 -0800 | [diff] [blame] | 3041 | |
Vladimir Davydov | f3ccb2c4 | 2015-11-05 18:49:01 -0800 | [diff] [blame] | 3042 | ret = try_charge(memcg, gfp, nr_pages); |
Johannes Weiner | 52c29b0 | 2016-01-20 15:02:35 -0800 | [diff] [blame] | 3043 | if (ret) |
Vladimir Davydov | f3ccb2c4 | 2015-11-05 18:49:01 -0800 | [diff] [blame] | 3044 | return ret; |
Johannes Weiner | 52c29b0 | 2016-01-20 15:02:35 -0800 | [diff] [blame] | 3045 | |
| 3046 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && |
| 3047 | !page_counter_try_charge(&memcg->kmem, nr_pages, &counter)) { |
Michal Hocko | e55d9d9 | 2019-09-25 16:45:53 -0700 | [diff] [blame] | 3048 | |
| 3049 | /* |
| 3050 | * Enforce __GFP_NOFAIL allocation because callers are not |
| 3051 | * prepared to see failures and likely do not have any failure |
| 3052 | * handling code. |
| 3053 | */ |
| 3054 | if (gfp & __GFP_NOFAIL) { |
| 3055 | page_counter_charge(&memcg->kmem, nr_pages); |
| 3056 | return 0; |
| 3057 | } |
Johannes Weiner | 52c29b0 | 2016-01-20 15:02:35 -0800 | [diff] [blame] | 3058 | cancel_charge(memcg, nr_pages); |
| 3059 | return -ENOMEM; |
Vladimir Davydov | f3ccb2c4 | 2015-11-05 18:49:01 -0800 | [diff] [blame] | 3060 | } |
Vladimir Davydov | f3ccb2c4 | 2015-11-05 18:49:01 -0800 | [diff] [blame] | 3061 | return 0; |
| 3062 | } |
| 3063 | |
Vladimir Davydov | 4526477 | 2016-07-26 15:24:21 -0700 | [diff] [blame] | 3064 | /** |
Roman Gushchin | 4b13f64 | 2020-04-01 21:06:56 -0700 | [diff] [blame] | 3065 | * __memcg_kmem_uncharge: uncharge a number of kernel pages from a memcg |
| 3066 | * @memcg: memcg to uncharge |
| 3067 | * @nr_pages: number of pages to uncharge |
| 3068 | */ |
Roman Gushchin | c1a660d | 2021-02-24 12:03:58 -0800 | [diff] [blame] | 3069 | static void __memcg_kmem_uncharge(struct mem_cgroup *memcg, unsigned int nr_pages) |
Roman Gushchin | 4b13f64 | 2020-04-01 21:06:56 -0700 | [diff] [blame] | 3070 | { |
| 3071 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
| 3072 | page_counter_uncharge(&memcg->kmem, nr_pages); |
| 3073 | |
Roman Gushchin | 3de7d4f | 2021-01-23 21:01:07 -0800 | [diff] [blame] | 3074 | refill_stock(memcg, nr_pages); |
Roman Gushchin | 4b13f64 | 2020-04-01 21:06:56 -0700 | [diff] [blame] | 3075 | } |
| 3076 | |
| 3077 | /** |
Roman Gushchin | f4b00ea | 2020-04-01 21:06:46 -0700 | [diff] [blame] | 3078 | * __memcg_kmem_charge_page: charge a kmem page to the current memory cgroup |
Vladimir Davydov | 4526477 | 2016-07-26 15:24:21 -0700 | [diff] [blame] | 3079 | * @page: page to charge |
| 3080 | * @gfp: reclaim mode |
| 3081 | * @order: allocation order |
| 3082 | * |
| 3083 | * Returns 0 on success, an error code on failure. |
| 3084 | */ |
Roman Gushchin | f4b00ea | 2020-04-01 21:06:46 -0700 | [diff] [blame] | 3085 | int __memcg_kmem_charge_page(struct page *page, gfp_t gfp, int order) |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 3086 | { |
| 3087 | struct mem_cgroup *memcg; |
Vladimir Davydov | fcff7d7 | 2016-03-17 14:17:29 -0700 | [diff] [blame] | 3088 | int ret = 0; |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 3089 | |
Shakeel Butt | d46eb14b | 2018-08-17 15:46:39 -0700 | [diff] [blame] | 3090 | memcg = get_mem_cgroup_from_current(); |
Roman Gushchin | 279c339 | 2020-10-17 16:13:44 -0700 | [diff] [blame] | 3091 | if (memcg && !mem_cgroup_is_root(memcg)) { |
Roman Gushchin | 4b13f64 | 2020-04-01 21:06:56 -0700 | [diff] [blame] | 3092 | ret = __memcg_kmem_charge(memcg, gfp, 1 << order); |
Roman Gushchin | 4d96ba3 | 2019-07-11 20:56:31 -0700 | [diff] [blame] | 3093 | if (!ret) { |
Roman Gushchin | 18b2db3 | 2020-12-01 13:58:30 -0800 | [diff] [blame] | 3094 | page->memcg_data = (unsigned long)memcg | |
| 3095 | MEMCG_DATA_KMEM; |
Johannes Weiner | 1a3e1f4 | 2020-08-06 23:20:45 -0700 | [diff] [blame] | 3096 | return 0; |
Roman Gushchin | 4d96ba3 | 2019-07-11 20:56:31 -0700 | [diff] [blame] | 3097 | } |
Roman Gushchin | 279c339 | 2020-10-17 16:13:44 -0700 | [diff] [blame] | 3098 | css_put(&memcg->css); |
Vladimir Davydov | c4159a7 | 2016-08-08 23:03:12 +0300 | [diff] [blame] | 3099 | } |
Vladimir Davydov | d05e83a | 2015-11-05 18:48:59 -0800 | [diff] [blame] | 3100 | return ret; |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 3101 | } |
Roman Gushchin | 49a18ea | 2019-07-11 20:56:13 -0700 | [diff] [blame] | 3102 | |
| 3103 | /** |
Roman Gushchin | f4b00ea | 2020-04-01 21:06:46 -0700 | [diff] [blame] | 3104 | * __memcg_kmem_uncharge_page: uncharge a kmem page |
Vladimir Davydov | 4526477 | 2016-07-26 15:24:21 -0700 | [diff] [blame] | 3105 | * @page: page to uncharge |
| 3106 | * @order: allocation order |
| 3107 | */ |
Roman Gushchin | f4b00ea | 2020-04-01 21:06:46 -0700 | [diff] [blame] | 3108 | void __memcg_kmem_uncharge_page(struct page *page, int order) |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 3109 | { |
Roman Gushchin | bcfe06b | 2020-12-01 13:58:27 -0800 | [diff] [blame] | 3110 | struct mem_cgroup *memcg = page_memcg(page); |
Vladimir Davydov | f3ccb2c4 | 2015-11-05 18:49:01 -0800 | [diff] [blame] | 3111 | unsigned int nr_pages = 1 << order; |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 3112 | |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 3113 | if (!memcg) |
| 3114 | return; |
| 3115 | |
Sasha Levin | 309381fea | 2014-01-23 15:52:54 -0800 | [diff] [blame] | 3116 | VM_BUG_ON_PAGE(mem_cgroup_is_root(memcg), page); |
Roman Gushchin | 4b13f64 | 2020-04-01 21:06:56 -0700 | [diff] [blame] | 3117 | __memcg_kmem_uncharge(memcg, nr_pages); |
Roman Gushchin | bcfe06b | 2020-12-01 13:58:27 -0800 | [diff] [blame] | 3118 | page->memcg_data = 0; |
Johannes Weiner | 1a3e1f4 | 2020-08-06 23:20:45 -0700 | [diff] [blame] | 3119 | css_put(&memcg->css); |
Vladimir Davydov | 60d3fd3 | 2015-02-12 14:59:10 -0800 | [diff] [blame] | 3120 | } |
Roman Gushchin | bf4f059 | 2020-08-06 23:20:49 -0700 | [diff] [blame] | 3121 | |
| 3122 | static bool consume_obj_stock(struct obj_cgroup *objcg, unsigned int nr_bytes) |
| 3123 | { |
| 3124 | struct memcg_stock_pcp *stock; |
| 3125 | unsigned long flags; |
| 3126 | bool ret = false; |
| 3127 | |
| 3128 | local_irq_save(flags); |
| 3129 | |
| 3130 | stock = this_cpu_ptr(&memcg_stock); |
| 3131 | if (objcg == stock->cached_objcg && stock->nr_bytes >= nr_bytes) { |
| 3132 | stock->nr_bytes -= nr_bytes; |
| 3133 | ret = true; |
| 3134 | } |
| 3135 | |
| 3136 | local_irq_restore(flags); |
| 3137 | |
| 3138 | return ret; |
| 3139 | } |
| 3140 | |
| 3141 | static void drain_obj_stock(struct memcg_stock_pcp *stock) |
| 3142 | { |
| 3143 | struct obj_cgroup *old = stock->cached_objcg; |
| 3144 | |
| 3145 | if (!old) |
| 3146 | return; |
| 3147 | |
| 3148 | if (stock->nr_bytes) { |
| 3149 | unsigned int nr_pages = stock->nr_bytes >> PAGE_SHIFT; |
| 3150 | unsigned int nr_bytes = stock->nr_bytes & (PAGE_SIZE - 1); |
| 3151 | |
| 3152 | if (nr_pages) { |
| 3153 | rcu_read_lock(); |
| 3154 | __memcg_kmem_uncharge(obj_cgroup_memcg(old), nr_pages); |
| 3155 | rcu_read_unlock(); |
| 3156 | } |
| 3157 | |
| 3158 | /* |
| 3159 | * The leftover is flushed to the centralized per-memcg value. |
| 3160 | * On the next attempt to refill obj stock it will be moved |
| 3161 | * to a per-cpu stock (probably, on an other CPU), see |
| 3162 | * refill_obj_stock(). |
| 3163 | * |
| 3164 | * How often it's flushed is a trade-off between the memory |
| 3165 | * limit enforcement accuracy and potential CPU contention, |
| 3166 | * so it might be changed in the future. |
| 3167 | */ |
| 3168 | atomic_add(nr_bytes, &old->nr_charged_bytes); |
| 3169 | stock->nr_bytes = 0; |
| 3170 | } |
| 3171 | |
| 3172 | obj_cgroup_put(old); |
| 3173 | stock->cached_objcg = NULL; |
| 3174 | } |
| 3175 | |
| 3176 | static bool obj_stock_flush_required(struct memcg_stock_pcp *stock, |
| 3177 | struct mem_cgroup *root_memcg) |
| 3178 | { |
| 3179 | struct mem_cgroup *memcg; |
| 3180 | |
| 3181 | if (stock->cached_objcg) { |
| 3182 | memcg = obj_cgroup_memcg(stock->cached_objcg); |
| 3183 | if (memcg && mem_cgroup_is_descendant(memcg, root_memcg)) |
| 3184 | return true; |
| 3185 | } |
| 3186 | |
| 3187 | return false; |
| 3188 | } |
| 3189 | |
| 3190 | static void refill_obj_stock(struct obj_cgroup *objcg, unsigned int nr_bytes) |
| 3191 | { |
| 3192 | struct memcg_stock_pcp *stock; |
| 3193 | unsigned long flags; |
| 3194 | |
| 3195 | local_irq_save(flags); |
| 3196 | |
| 3197 | stock = this_cpu_ptr(&memcg_stock); |
| 3198 | if (stock->cached_objcg != objcg) { /* reset if necessary */ |
| 3199 | drain_obj_stock(stock); |
| 3200 | obj_cgroup_get(objcg); |
| 3201 | stock->cached_objcg = objcg; |
| 3202 | stock->nr_bytes = atomic_xchg(&objcg->nr_charged_bytes, 0); |
| 3203 | } |
| 3204 | stock->nr_bytes += nr_bytes; |
| 3205 | |
| 3206 | if (stock->nr_bytes > PAGE_SIZE) |
| 3207 | drain_obj_stock(stock); |
| 3208 | |
| 3209 | local_irq_restore(flags); |
| 3210 | } |
| 3211 | |
| 3212 | int obj_cgroup_charge(struct obj_cgroup *objcg, gfp_t gfp, size_t size) |
| 3213 | { |
| 3214 | struct mem_cgroup *memcg; |
| 3215 | unsigned int nr_pages, nr_bytes; |
| 3216 | int ret; |
| 3217 | |
| 3218 | if (consume_obj_stock(objcg, size)) |
| 3219 | return 0; |
| 3220 | |
| 3221 | /* |
| 3222 | * In theory, memcg->nr_charged_bytes can have enough |
| 3223 | * pre-charged bytes to satisfy the allocation. However, |
| 3224 | * flushing memcg->nr_charged_bytes requires two atomic |
| 3225 | * operations, and memcg->nr_charged_bytes can't be big, |
| 3226 | * so it's better to ignore it and try grab some new pages. |
| 3227 | * memcg->nr_charged_bytes will be flushed in |
| 3228 | * refill_obj_stock(), called from this function or |
| 3229 | * independently later. |
| 3230 | */ |
| 3231 | rcu_read_lock(); |
Muchun Song | eefbfa7 | 2020-12-14 19:06:35 -0800 | [diff] [blame] | 3232 | retry: |
Roman Gushchin | bf4f059 | 2020-08-06 23:20:49 -0700 | [diff] [blame] | 3233 | memcg = obj_cgroup_memcg(objcg); |
Muchun Song | eefbfa7 | 2020-12-14 19:06:35 -0800 | [diff] [blame] | 3234 | if (unlikely(!css_tryget(&memcg->css))) |
| 3235 | goto retry; |
Roman Gushchin | bf4f059 | 2020-08-06 23:20:49 -0700 | [diff] [blame] | 3236 | rcu_read_unlock(); |
| 3237 | |
| 3238 | nr_pages = size >> PAGE_SHIFT; |
| 3239 | nr_bytes = size & (PAGE_SIZE - 1); |
| 3240 | |
| 3241 | if (nr_bytes) |
| 3242 | nr_pages += 1; |
| 3243 | |
| 3244 | ret = __memcg_kmem_charge(memcg, gfp, nr_pages); |
| 3245 | if (!ret && nr_bytes) |
| 3246 | refill_obj_stock(objcg, PAGE_SIZE - nr_bytes); |
| 3247 | |
| 3248 | css_put(&memcg->css); |
| 3249 | return ret; |
| 3250 | } |
| 3251 | |
| 3252 | void obj_cgroup_uncharge(struct obj_cgroup *objcg, size_t size) |
| 3253 | { |
| 3254 | refill_obj_stock(objcg, size); |
| 3255 | } |
| 3256 | |
Kirill Tkhai | 84c07d1 | 2018-08-17 15:47:25 -0700 | [diff] [blame] | 3257 | #endif /* CONFIG_MEMCG_KMEM */ |
Glauber Costa | 7ae1e1d | 2012-12-18 14:21:56 -0800 | [diff] [blame] | 3258 | |
KAMEZAWA Hiroyuki | ca3e021 | 2011-01-20 14:44:24 -0800 | [diff] [blame] | 3259 | /* |
Zhou Guanghui | be6c898 | 2021-03-12 21:08:30 -0800 | [diff] [blame] | 3260 | * Because page_memcg(head) is not set on tails, set it now. |
KAMEZAWA Hiroyuki | ca3e021 | 2011-01-20 14:44:24 -0800 | [diff] [blame] | 3261 | */ |
Zhou Guanghui | be6c898 | 2021-03-12 21:08:30 -0800 | [diff] [blame] | 3262 | void split_page_memcg(struct page *head, unsigned int nr) |
KAMEZAWA Hiroyuki | ca3e021 | 2011-01-20 14:44:24 -0800 | [diff] [blame] | 3263 | { |
Roman Gushchin | bcfe06b | 2020-12-01 13:58:27 -0800 | [diff] [blame] | 3264 | struct mem_cgroup *memcg = page_memcg(head); |
KAMEZAWA Hiroyuki | e94c8a9 | 2012-01-12 17:18:20 -0800 | [diff] [blame] | 3265 | int i; |
KAMEZAWA Hiroyuki | ca3e021 | 2011-01-20 14:44:24 -0800 | [diff] [blame] | 3266 | |
Zhou Guanghui | be6c898 | 2021-03-12 21:08:30 -0800 | [diff] [blame] | 3267 | if (mem_cgroup_disabled() || !memcg) |
KAMEZAWA Hiroyuki | 3d37c4a | 2011-01-25 15:07:28 -0800 | [diff] [blame] | 3268 | return; |
David Rientjes | b070e65 | 2013-05-07 16:18:09 -0700 | [diff] [blame] | 3269 | |
Zhou Guanghui | be6c898 | 2021-03-12 21:08:30 -0800 | [diff] [blame] | 3270 | for (i = 1; i < nr; i++) |
| 3271 | head[i].memcg_data = head->memcg_data; |
| 3272 | css_get_many(&memcg->css, nr - 1); |
KAMEZAWA Hiroyuki | ca3e021 | 2011-01-20 14:44:24 -0800 | [diff] [blame] | 3273 | } |
KAMEZAWA Hiroyuki | ca3e021 | 2011-01-20 14:44:24 -0800 | [diff] [blame] | 3274 | |
Andrew Morton | c255a45 | 2012-07-31 16:43:02 -0700 | [diff] [blame] | 3275 | #ifdef CONFIG_MEMCG_SWAP |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 3276 | /** |
| 3277 | * mem_cgroup_move_swap_account - move swap charge and swap_cgroup's record. |
| 3278 | * @entry: swap entry to be moved |
| 3279 | * @from: mem_cgroup which the entry is moved from |
| 3280 | * @to: mem_cgroup which the entry is moved to |
| 3281 | * |
| 3282 | * It succeeds only when the swap_cgroup's record for this entry is the same |
| 3283 | * as the mem_cgroup's id of @from. |
| 3284 | * |
| 3285 | * Returns 0 on success, -EINVAL on failure. |
| 3286 | * |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3287 | * 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] | 3288 | * both res and memsw, and called css_get(). |
| 3289 | */ |
| 3290 | static int mem_cgroup_move_swap_account(swp_entry_t entry, |
Hugh Dickins | e91cbb4 | 2012-05-29 15:06:51 -0700 | [diff] [blame] | 3291 | struct mem_cgroup *from, struct mem_cgroup *to) |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 3292 | { |
| 3293 | unsigned short old_id, new_id; |
| 3294 | |
Li Zefan | 34c00c3 | 2013-09-23 16:56:01 +0800 | [diff] [blame] | 3295 | old_id = mem_cgroup_id(from); |
| 3296 | new_id = mem_cgroup_id(to); |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 3297 | |
| 3298 | if (swap_cgroup_cmpxchg(entry, old_id, new_id) == old_id) { |
Johannes Weiner | c9019e9 | 2018-01-31 16:16:37 -0800 | [diff] [blame] | 3299 | mod_memcg_state(from, MEMCG_SWAP, -1); |
| 3300 | mod_memcg_state(to, MEMCG_SWAP, 1); |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 3301 | return 0; |
| 3302 | } |
| 3303 | return -EINVAL; |
| 3304 | } |
| 3305 | #else |
| 3306 | static inline int mem_cgroup_move_swap_account(swp_entry_t entry, |
Hugh Dickins | e91cbb4 | 2012-05-29 15:06:51 -0700 | [diff] [blame] | 3307 | struct mem_cgroup *from, struct mem_cgroup *to) |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 3308 | { |
| 3309 | return -EINVAL; |
| 3310 | } |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3311 | #endif |
| 3312 | |
Roman Gushchin | bbec2e1 | 2018-06-07 17:06:18 -0700 | [diff] [blame] | 3313 | static DEFINE_MUTEX(memcg_max_mutex); |
Daisuke Nishimura | f212ad7 | 2011-03-23 16:42:25 -0700 | [diff] [blame] | 3314 | |
Roman Gushchin | bbec2e1 | 2018-06-07 17:06:18 -0700 | [diff] [blame] | 3315 | static int mem_cgroup_resize_max(struct mem_cgroup *memcg, |
| 3316 | unsigned long max, bool memsw) |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 3317 | { |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3318 | bool enlarge = false; |
Shakeel Butt | bb4a7ea | 2018-06-07 17:07:27 -0700 | [diff] [blame] | 3319 | bool drained = false; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3320 | int ret; |
Yu Zhao | c054a78 | 2018-01-31 16:20:02 -0800 | [diff] [blame] | 3321 | bool limits_invariant; |
| 3322 | struct page_counter *counter = memsw ? &memcg->memsw : &memcg->memory; |
KAMEZAWA Hiroyuki | 81d39c2 | 2009-04-02 16:57:36 -0700 | [diff] [blame] | 3323 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3324 | do { |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 3325 | if (signal_pending(current)) { |
| 3326 | ret = -EINTR; |
| 3327 | break; |
| 3328 | } |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3329 | |
Roman Gushchin | bbec2e1 | 2018-06-07 17:06:18 -0700 | [diff] [blame] | 3330 | mutex_lock(&memcg_max_mutex); |
Yu Zhao | c054a78 | 2018-01-31 16:20:02 -0800 | [diff] [blame] | 3331 | /* |
| 3332 | * Make sure that the new limit (memsw or memory limit) doesn't |
Roman Gushchin | bbec2e1 | 2018-06-07 17:06:18 -0700 | [diff] [blame] | 3333 | * break our basic invariant rule memory.max <= memsw.max. |
Yu Zhao | c054a78 | 2018-01-31 16:20:02 -0800 | [diff] [blame] | 3334 | */ |
Chris Down | 15b4256 | 2020-04-01 21:07:20 -0700 | [diff] [blame] | 3335 | limits_invariant = memsw ? max >= READ_ONCE(memcg->memory.max) : |
Roman Gushchin | bbec2e1 | 2018-06-07 17:06:18 -0700 | [diff] [blame] | 3336 | max <= memcg->memsw.max; |
Yu Zhao | c054a78 | 2018-01-31 16:20:02 -0800 | [diff] [blame] | 3337 | if (!limits_invariant) { |
Roman Gushchin | bbec2e1 | 2018-06-07 17:06:18 -0700 | [diff] [blame] | 3338 | mutex_unlock(&memcg_max_mutex); |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3339 | ret = -EINVAL; |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 3340 | break; |
| 3341 | } |
Roman Gushchin | bbec2e1 | 2018-06-07 17:06:18 -0700 | [diff] [blame] | 3342 | if (max > counter->max) |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3343 | enlarge = true; |
Roman Gushchin | bbec2e1 | 2018-06-07 17:06:18 -0700 | [diff] [blame] | 3344 | ret = page_counter_set_max(counter, max); |
| 3345 | mutex_unlock(&memcg_max_mutex); |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3346 | |
| 3347 | if (!ret) |
| 3348 | break; |
| 3349 | |
Shakeel Butt | bb4a7ea | 2018-06-07 17:07:27 -0700 | [diff] [blame] | 3350 | if (!drained) { |
| 3351 | drain_all_stock(memcg); |
| 3352 | drained = true; |
| 3353 | continue; |
| 3354 | } |
| 3355 | |
Andrey Ryabinin | 1ab5c05 | 2018-01-31 16:20:37 -0800 | [diff] [blame] | 3356 | if (!try_to_free_mem_cgroup_pages(memcg, 1, |
| 3357 | GFP_KERNEL, !memsw)) { |
| 3358 | ret = -EBUSY; |
| 3359 | break; |
| 3360 | } |
| 3361 | } while (true); |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3362 | |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 3363 | if (!ret && enlarge) |
| 3364 | memcg_oom_recover(memcg); |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3365 | |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 3366 | return ret; |
| 3367 | } |
| 3368 | |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 3369 | 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] | 3370 | gfp_t gfp_mask, |
| 3371 | unsigned long *total_scanned) |
| 3372 | { |
| 3373 | unsigned long nr_reclaimed = 0; |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 3374 | struct mem_cgroup_per_node *mz, *next_mz = NULL; |
Andrew Morton | 0608f43 | 2013-09-24 15:27:41 -0700 | [diff] [blame] | 3375 | unsigned long reclaimed; |
| 3376 | int loop = 0; |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 3377 | struct mem_cgroup_tree_per_node *mctz; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3378 | unsigned long excess; |
Andrew Morton | 0608f43 | 2013-09-24 15:27:41 -0700 | [diff] [blame] | 3379 | unsigned long nr_scanned; |
| 3380 | |
| 3381 | if (order > 0) |
| 3382 | return 0; |
| 3383 | |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 3384 | mctz = soft_limit_tree_node(pgdat->node_id); |
Michal Hocko | d6507ff | 2016-08-02 14:02:37 -0700 | [diff] [blame] | 3385 | |
| 3386 | /* |
| 3387 | * Do not even bother to check the largest node if the root |
| 3388 | * is empty. Do it lockless to prevent lock bouncing. Races |
| 3389 | * are acceptable as soft limit is best effort anyway. |
| 3390 | */ |
Laurent Dufour | bfc7228 | 2017-03-09 16:17:06 -0800 | [diff] [blame] | 3391 | if (!mctz || RB_EMPTY_ROOT(&mctz->rb_root)) |
Michal Hocko | d6507ff | 2016-08-02 14:02:37 -0700 | [diff] [blame] | 3392 | return 0; |
| 3393 | |
Andrew Morton | 0608f43 | 2013-09-24 15:27:41 -0700 | [diff] [blame] | 3394 | /* |
| 3395 | * This loop can run a while, specially if mem_cgroup's continuously |
| 3396 | * keep exceeding their soft limit and putting the system under |
| 3397 | * pressure |
| 3398 | */ |
| 3399 | do { |
| 3400 | if (next_mz) |
| 3401 | mz = next_mz; |
| 3402 | else |
| 3403 | mz = mem_cgroup_largest_soft_limit_node(mctz); |
| 3404 | if (!mz) |
| 3405 | break; |
| 3406 | |
| 3407 | nr_scanned = 0; |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 3408 | reclaimed = mem_cgroup_soft_reclaim(mz->memcg, pgdat, |
Andrew Morton | 0608f43 | 2013-09-24 15:27:41 -0700 | [diff] [blame] | 3409 | gfp_mask, &nr_scanned); |
| 3410 | nr_reclaimed += reclaimed; |
| 3411 | *total_scanned += nr_scanned; |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 3412 | spin_lock_irq(&mctz->lock); |
Vladimir Davydov | bc2f2e7 | 2014-12-10 15:43:40 -0800 | [diff] [blame] | 3413 | __mem_cgroup_remove_exceeded(mz, mctz); |
Andrew Morton | 0608f43 | 2013-09-24 15:27:41 -0700 | [diff] [blame] | 3414 | |
| 3415 | /* |
| 3416 | * If we failed to reclaim anything from this memory cgroup |
| 3417 | * it is time to move on to the next cgroup |
| 3418 | */ |
| 3419 | next_mz = NULL; |
Vladimir Davydov | bc2f2e7 | 2014-12-10 15:43:40 -0800 | [diff] [blame] | 3420 | if (!reclaimed) |
| 3421 | next_mz = __mem_cgroup_largest_soft_limit_node(mctz); |
| 3422 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3423 | excess = soft_limit_excess(mz->memcg); |
Andrew Morton | 0608f43 | 2013-09-24 15:27:41 -0700 | [diff] [blame] | 3424 | /* |
| 3425 | * One school of thought says that we should not add |
| 3426 | * back the node to the tree if reclaim returns 0. |
| 3427 | * But our reclaim could return 0, simply because due |
| 3428 | * to priority we are exposing a smaller subset of |
| 3429 | * memory to reclaim from. Consider this as a longer |
| 3430 | * term TODO. |
| 3431 | */ |
| 3432 | /* If excess == 0, no tree ops */ |
Johannes Weiner | cf2c812 | 2014-06-06 14:38:21 -0700 | [diff] [blame] | 3433 | __mem_cgroup_insert_exceeded(mz, mctz, excess); |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 3434 | spin_unlock_irq(&mctz->lock); |
Andrew Morton | 0608f43 | 2013-09-24 15:27:41 -0700 | [diff] [blame] | 3435 | css_put(&mz->memcg->css); |
| 3436 | loop++; |
| 3437 | /* |
| 3438 | * Could not reclaim anything and there are no more |
| 3439 | * mem cgroups to try or we seem to be looping without |
| 3440 | * reclaiming anything. |
| 3441 | */ |
| 3442 | if (!nr_reclaimed && |
| 3443 | (next_mz == NULL || |
| 3444 | loop > MEM_CGROUP_MAX_SOFT_LIMIT_RECLAIM_LOOPS)) |
| 3445 | break; |
| 3446 | } while (!nr_reclaimed); |
| 3447 | if (next_mz) |
| 3448 | css_put(&next_mz->memcg->css); |
| 3449 | return nr_reclaimed; |
| 3450 | } |
| 3451 | |
Tejun Heo | ea280e7 | 2014-05-16 13:22:48 -0400 | [diff] [blame] | 3452 | /* |
Greg Thelen | 5103817 | 2016-05-20 16:58:18 -0700 | [diff] [blame] | 3453 | * Reclaims as many pages from the given memcg as possible. |
Michal Hocko | c26251f | 2012-10-26 13:37:28 +0200 | [diff] [blame] | 3454 | * |
| 3455 | * Caller is responsible for holding css reference for memcg. |
| 3456 | */ |
| 3457 | static int mem_cgroup_force_empty(struct mem_cgroup *memcg) |
| 3458 | { |
Chris Down | d977aa9 | 2020-08-06 23:21:58 -0700 | [diff] [blame] | 3459 | int nr_retries = MAX_RECLAIM_RETRIES; |
Michal Hocko | c26251f | 2012-10-26 13:37:28 +0200 | [diff] [blame] | 3460 | |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 3461 | /* we call try-to-free pages for make this cgroup empty */ |
| 3462 | lru_add_drain_all(); |
Junaid Shahid | d12c60f | 2018-06-07 17:07:31 -0700 | [diff] [blame] | 3463 | |
| 3464 | drain_all_stock(memcg); |
| 3465 | |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3466 | /* try to free all pages in this cgroup */ |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3467 | while (nr_retries && page_counter_read(&memcg->memory)) { |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3468 | int progress; |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 3469 | |
Michal Hocko | c26251f | 2012-10-26 13:37:28 +0200 | [diff] [blame] | 3470 | if (signal_pending(current)) |
| 3471 | return -EINTR; |
| 3472 | |
Johannes Weiner | b70a2a2 | 2014-10-09 15:28:56 -0700 | [diff] [blame] | 3473 | progress = try_to_free_mem_cgroup_pages(memcg, 1, |
| 3474 | GFP_KERNEL, true); |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 3475 | if (!progress) { |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3476 | nr_retries--; |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 3477 | /* maybe some writeback is necessary */ |
Jens Axboe | 8aa7e84 | 2009-07-09 14:52:32 +0200 | [diff] [blame] | 3478 | congestion_wait(BLK_RW_ASYNC, HZ/10); |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 3479 | } |
KAMEZAWA Hiroyuki | f817ed4 | 2009-01-07 18:07:53 -0800 | [diff] [blame] | 3480 | |
| 3481 | } |
Michal Hocko | ab5196c | 2012-10-26 13:37:32 +0200 | [diff] [blame] | 3482 | |
| 3483 | return 0; |
KAMEZAWA Hiroyuki | cc84758 | 2008-02-07 00:14:16 -0800 | [diff] [blame] | 3484 | } |
| 3485 | |
Tejun Heo | 6770c64 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 3486 | static ssize_t mem_cgroup_force_empty_write(struct kernfs_open_file *of, |
| 3487 | char *buf, size_t nbytes, |
| 3488 | loff_t off) |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 3489 | { |
Tejun Heo | 6770c64 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 3490 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
Michal Hocko | c26251f | 2012-10-26 13:37:28 +0200 | [diff] [blame] | 3491 | |
Michal Hocko | d842301 | 2012-10-26 13:37:29 +0200 | [diff] [blame] | 3492 | if (mem_cgroup_is_root(memcg)) |
| 3493 | return -EINVAL; |
Tejun Heo | 6770c64 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 3494 | return mem_cgroup_force_empty(memcg) ?: nbytes; |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 3495 | } |
| 3496 | |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 3497 | static u64 mem_cgroup_hierarchy_read(struct cgroup_subsys_state *css, |
| 3498 | struct cftype *cft) |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 3499 | { |
Roman Gushchin | bef8620 | 2020-12-14 19:06:49 -0800 | [diff] [blame] | 3500 | return 1; |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 3501 | } |
| 3502 | |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 3503 | static int mem_cgroup_hierarchy_write(struct cgroup_subsys_state *css, |
| 3504 | struct cftype *cft, u64 val) |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 3505 | { |
Roman Gushchin | bef8620 | 2020-12-14 19:06:49 -0800 | [diff] [blame] | 3506 | if (val == 1) |
Johannes Weiner | 0b8f73e | 2016-01-20 15:02:53 -0800 | [diff] [blame] | 3507 | return 0; |
Glauber Costa | 567fb43 | 2012-07-31 16:43:07 -0700 | [diff] [blame] | 3508 | |
Roman Gushchin | bef8620 | 2020-12-14 19:06:49 -0800 | [diff] [blame] | 3509 | pr_warn_once("Non-hierarchical mode is deprecated. " |
| 3510 | "Please report your usecase to linux-mm@kvack.org if you " |
| 3511 | "depend on this functionality.\n"); |
Glauber Costa | 567fb43 | 2012-07-31 16:43:07 -0700 | [diff] [blame] | 3512 | |
Roman Gushchin | bef8620 | 2020-12-14 19:06:49 -0800 | [diff] [blame] | 3513 | return -EINVAL; |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 3514 | } |
| 3515 | |
Andrew Morton | 6f64615 | 2015-11-06 16:28:58 -0800 | [diff] [blame] | 3516 | static unsigned long mem_cgroup_usage(struct mem_cgroup *memcg, bool swap) |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 3517 | { |
Johannes Weiner | 42a3003 | 2019-05-14 15:47:12 -0700 | [diff] [blame] | 3518 | unsigned long val; |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 3519 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3520 | if (mem_cgroup_is_root(memcg)) { |
Johannes Weiner | 2d146aa | 2021-04-29 22:56:26 -0700 | [diff] [blame] | 3521 | cgroup_rstat_flush(memcg->css.cgroup); |
Johannes Weiner | 0d1c207 | 2020-06-03 16:01:54 -0700 | [diff] [blame] | 3522 | val = memcg_page_state(memcg, NR_FILE_PAGES) + |
Johannes Weiner | be5d0a7 | 2020-06-03 16:01:57 -0700 | [diff] [blame] | 3523 | memcg_page_state(memcg, NR_ANON_MAPPED); |
Johannes Weiner | 42a3003 | 2019-05-14 15:47:12 -0700 | [diff] [blame] | 3524 | if (swap) |
| 3525 | val += memcg_page_state(memcg, MEMCG_SWAP); |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3526 | } else { |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 3527 | if (!swap) |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3528 | val = page_counter_read(&memcg->memory); |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 3529 | else |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3530 | val = page_counter_read(&memcg->memsw); |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 3531 | } |
Michal Hocko | c12176d | 2015-11-05 18:50:29 -0800 | [diff] [blame] | 3532 | return val; |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 3533 | } |
| 3534 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3535 | enum { |
| 3536 | RES_USAGE, |
| 3537 | RES_LIMIT, |
| 3538 | RES_MAX_USAGE, |
| 3539 | RES_FAILCNT, |
| 3540 | RES_SOFT_LIMIT, |
| 3541 | }; |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 3542 | |
Tejun Heo | 791badb | 2013-12-05 12:28:02 -0500 | [diff] [blame] | 3543 | static u64 mem_cgroup_read_u64(struct cgroup_subsys_state *css, |
Johannes Weiner | 05b8430 | 2014-08-06 16:05:59 -0700 | [diff] [blame] | 3544 | struct cftype *cft) |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 3545 | { |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 3546 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3547 | struct page_counter *counter; |
Tejun Heo | af36f90 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 3548 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3549 | switch (MEMFILE_TYPE(cft->private)) { |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3550 | case _MEM: |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3551 | counter = &memcg->memory; |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 3552 | break; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3553 | case _MEMSWAP: |
| 3554 | counter = &memcg->memsw; |
| 3555 | break; |
| 3556 | case _KMEM: |
| 3557 | counter = &memcg->kmem; |
| 3558 | break; |
Vladimir Davydov | d55f90b | 2016-01-20 15:02:44 -0800 | [diff] [blame] | 3559 | case _TCP: |
Johannes Weiner | 0db1529 | 2016-01-20 15:02:50 -0800 | [diff] [blame] | 3560 | counter = &memcg->tcpmem; |
Vladimir Davydov | d55f90b | 2016-01-20 15:02:44 -0800 | [diff] [blame] | 3561 | break; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3562 | default: |
| 3563 | BUG(); |
| 3564 | } |
| 3565 | |
| 3566 | switch (MEMFILE_ATTR(cft->private)) { |
| 3567 | case RES_USAGE: |
| 3568 | if (counter == &memcg->memory) |
Michal Hocko | c12176d | 2015-11-05 18:50:29 -0800 | [diff] [blame] | 3569 | return (u64)mem_cgroup_usage(memcg, false) * PAGE_SIZE; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3570 | if (counter == &memcg->memsw) |
Michal Hocko | c12176d | 2015-11-05 18:50:29 -0800 | [diff] [blame] | 3571 | return (u64)mem_cgroup_usage(memcg, true) * PAGE_SIZE; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3572 | return (u64)page_counter_read(counter) * PAGE_SIZE; |
| 3573 | case RES_LIMIT: |
Roman Gushchin | bbec2e1 | 2018-06-07 17:06:18 -0700 | [diff] [blame] | 3574 | return (u64)counter->max * PAGE_SIZE; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3575 | case RES_MAX_USAGE: |
| 3576 | return (u64)counter->watermark * PAGE_SIZE; |
| 3577 | case RES_FAILCNT: |
| 3578 | return counter->failcnt; |
| 3579 | case RES_SOFT_LIMIT: |
| 3580 | return (u64)memcg->soft_limit * PAGE_SIZE; |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3581 | default: |
| 3582 | BUG(); |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3583 | } |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 3584 | } |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 3585 | |
Kirill Tkhai | 84c07d1 | 2018-08-17 15:47:25 -0700 | [diff] [blame] | 3586 | #ifdef CONFIG_MEMCG_KMEM |
Johannes Weiner | 567e9ab | 2016-01-20 15:02:24 -0800 | [diff] [blame] | 3587 | static int memcg_online_kmem(struct mem_cgroup *memcg) |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 3588 | { |
Roman Gushchin | bf4f059 | 2020-08-06 23:20:49 -0700 | [diff] [blame] | 3589 | struct obj_cgroup *objcg; |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 3590 | int memcg_id; |
| 3591 | |
Vladimir Davydov | b313aee | 2016-03-17 14:18:27 -0700 | [diff] [blame] | 3592 | if (cgroup_memory_nokmem) |
| 3593 | return 0; |
| 3594 | |
Vladimir Davydov | 2a4db7e | 2015-02-12 14:59:32 -0800 | [diff] [blame] | 3595 | BUG_ON(memcg->kmemcg_id >= 0); |
Johannes Weiner | 567e9ab | 2016-01-20 15:02:24 -0800 | [diff] [blame] | 3596 | BUG_ON(memcg->kmem_state); |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 3597 | |
Vladimir Davydov | f3bb304 | 2014-10-09 15:28:45 -0700 | [diff] [blame] | 3598 | memcg_id = memcg_alloc_cache_id(); |
Johannes Weiner | 0b8f73e | 2016-01-20 15:02:53 -0800 | [diff] [blame] | 3599 | if (memcg_id < 0) |
| 3600 | return memcg_id; |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 3601 | |
Roman Gushchin | bf4f059 | 2020-08-06 23:20:49 -0700 | [diff] [blame] | 3602 | objcg = obj_cgroup_alloc(); |
| 3603 | if (!objcg) { |
| 3604 | memcg_free_cache_id(memcg_id); |
| 3605 | return -ENOMEM; |
| 3606 | } |
| 3607 | objcg->memcg = memcg; |
| 3608 | rcu_assign_pointer(memcg->objcg, objcg); |
| 3609 | |
Roman Gushchin | d648bcc | 2020-08-06 23:20:28 -0700 | [diff] [blame] | 3610 | static_branch_enable(&memcg_kmem_enabled_key); |
| 3611 | |
Vladimir Davydov | 900a38f | 2014-12-12 16:55:10 -0800 | [diff] [blame] | 3612 | memcg->kmemcg_id = memcg_id; |
Johannes Weiner | 567e9ab | 2016-01-20 15:02:24 -0800 | [diff] [blame] | 3613 | memcg->kmem_state = KMEM_ONLINE; |
Johannes Weiner | 0b8f73e | 2016-01-20 15:02:53 -0800 | [diff] [blame] | 3614 | |
| 3615 | return 0; |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 3616 | } |
| 3617 | |
Johannes Weiner | 8e0a891 | 2016-01-20 15:02:26 -0800 | [diff] [blame] | 3618 | static void memcg_offline_kmem(struct mem_cgroup *memcg) |
| 3619 | { |
| 3620 | struct cgroup_subsys_state *css; |
| 3621 | struct mem_cgroup *parent, *child; |
| 3622 | int kmemcg_id; |
| 3623 | |
| 3624 | if (memcg->kmem_state != KMEM_ONLINE) |
| 3625 | return; |
Roman Gushchin | 9855609 | 2020-08-06 23:21:10 -0700 | [diff] [blame] | 3626 | |
Johannes Weiner | 8e0a891 | 2016-01-20 15:02:26 -0800 | [diff] [blame] | 3627 | memcg->kmem_state = KMEM_ALLOCATED; |
| 3628 | |
Johannes Weiner | 8e0a891 | 2016-01-20 15:02:26 -0800 | [diff] [blame] | 3629 | parent = parent_mem_cgroup(memcg); |
| 3630 | if (!parent) |
| 3631 | parent = root_mem_cgroup; |
| 3632 | |
Roman Gushchin | bf4f059 | 2020-08-06 23:20:49 -0700 | [diff] [blame] | 3633 | memcg_reparent_objcgs(memcg, parent); |
Roman Gushchin | fb2f2b0 | 2019-07-11 20:56:34 -0700 | [diff] [blame] | 3634 | |
| 3635 | kmemcg_id = memcg->kmemcg_id; |
| 3636 | BUG_ON(kmemcg_id < 0); |
| 3637 | |
Johannes Weiner | 8e0a891 | 2016-01-20 15:02:26 -0800 | [diff] [blame] | 3638 | /* |
| 3639 | * Change kmemcg_id of this cgroup and all its descendants to the |
| 3640 | * parent's id, and then move all entries from this cgroup's list_lrus |
| 3641 | * to ones of the parent. After we have finished, all list_lrus |
| 3642 | * corresponding to this cgroup are guaranteed to remain empty. The |
| 3643 | * ordering is imposed by list_lru_node->lock taken by |
| 3644 | * memcg_drain_all_list_lrus(). |
| 3645 | */ |
Tejun Heo | 3a06bb7 | 2016-06-03 14:55:44 -0700 | [diff] [blame] | 3646 | 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] | 3647 | css_for_each_descendant_pre(css, &memcg->css) { |
| 3648 | child = mem_cgroup_from_css(css); |
| 3649 | BUG_ON(child->kmemcg_id != kmemcg_id); |
| 3650 | child->kmemcg_id = parent->kmemcg_id; |
Johannes Weiner | 8e0a891 | 2016-01-20 15:02:26 -0800 | [diff] [blame] | 3651 | } |
Tejun Heo | 3a06bb7 | 2016-06-03 14:55:44 -0700 | [diff] [blame] | 3652 | rcu_read_unlock(); |
| 3653 | |
Kirill Tkhai | 9bec5c3 | 2018-08-17 15:47:58 -0700 | [diff] [blame] | 3654 | memcg_drain_all_list_lrus(kmemcg_id, parent); |
Johannes Weiner | 8e0a891 | 2016-01-20 15:02:26 -0800 | [diff] [blame] | 3655 | |
| 3656 | memcg_free_cache_id(kmemcg_id); |
| 3657 | } |
| 3658 | |
| 3659 | static void memcg_free_kmem(struct mem_cgroup *memcg) |
| 3660 | { |
Johannes Weiner | 0b8f73e | 2016-01-20 15:02:53 -0800 | [diff] [blame] | 3661 | /* css_alloc() failed, offlining didn't happen */ |
| 3662 | if (unlikely(memcg->kmem_state == KMEM_ONLINE)) |
| 3663 | memcg_offline_kmem(memcg); |
Johannes Weiner | 8e0a891 | 2016-01-20 15:02:26 -0800 | [diff] [blame] | 3664 | } |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 3665 | #else |
Johannes Weiner | 0b8f73e | 2016-01-20 15:02:53 -0800 | [diff] [blame] | 3666 | static int memcg_online_kmem(struct mem_cgroup *memcg) |
Johannes Weiner | 127424c | 2016-01-20 15:02:32 -0800 | [diff] [blame] | 3667 | { |
| 3668 | return 0; |
| 3669 | } |
| 3670 | static void memcg_offline_kmem(struct mem_cgroup *memcg) |
| 3671 | { |
| 3672 | } |
| 3673 | static void memcg_free_kmem(struct mem_cgroup *memcg) |
| 3674 | { |
| 3675 | } |
Kirill Tkhai | 84c07d1 | 2018-08-17 15:47:25 -0700 | [diff] [blame] | 3676 | #endif /* CONFIG_MEMCG_KMEM */ |
Johannes Weiner | 127424c | 2016-01-20 15:02:32 -0800 | [diff] [blame] | 3677 | |
Roman Gushchin | bbec2e1 | 2018-06-07 17:06:18 -0700 | [diff] [blame] | 3678 | static int memcg_update_kmem_max(struct mem_cgroup *memcg, |
| 3679 | unsigned long max) |
Johannes Weiner | 127424c | 2016-01-20 15:02:32 -0800 | [diff] [blame] | 3680 | { |
Vladimir Davydov | b313aee | 2016-03-17 14:18:27 -0700 | [diff] [blame] | 3681 | int ret; |
Johannes Weiner | 127424c | 2016-01-20 15:02:32 -0800 | [diff] [blame] | 3682 | |
Roman Gushchin | bbec2e1 | 2018-06-07 17:06:18 -0700 | [diff] [blame] | 3683 | mutex_lock(&memcg_max_mutex); |
| 3684 | ret = page_counter_set_max(&memcg->kmem, max); |
| 3685 | mutex_unlock(&memcg_max_mutex); |
Johannes Weiner | 127424c | 2016-01-20 15:02:32 -0800 | [diff] [blame] | 3686 | return ret; |
| 3687 | } |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 3688 | |
Roman Gushchin | bbec2e1 | 2018-06-07 17:06:18 -0700 | [diff] [blame] | 3689 | static int memcg_update_tcp_max(struct mem_cgroup *memcg, unsigned long max) |
Vladimir Davydov | d55f90b | 2016-01-20 15:02:44 -0800 | [diff] [blame] | 3690 | { |
| 3691 | int ret; |
| 3692 | |
Roman Gushchin | bbec2e1 | 2018-06-07 17:06:18 -0700 | [diff] [blame] | 3693 | mutex_lock(&memcg_max_mutex); |
Vladimir Davydov | d55f90b | 2016-01-20 15:02:44 -0800 | [diff] [blame] | 3694 | |
Roman Gushchin | bbec2e1 | 2018-06-07 17:06:18 -0700 | [diff] [blame] | 3695 | ret = page_counter_set_max(&memcg->tcpmem, max); |
Vladimir Davydov | d55f90b | 2016-01-20 15:02:44 -0800 | [diff] [blame] | 3696 | if (ret) |
| 3697 | goto out; |
| 3698 | |
Johannes Weiner | 0db1529 | 2016-01-20 15:02:50 -0800 | [diff] [blame] | 3699 | if (!memcg->tcpmem_active) { |
Vladimir Davydov | d55f90b | 2016-01-20 15:02:44 -0800 | [diff] [blame] | 3700 | /* |
| 3701 | * The active flag needs to be written after the static_key |
| 3702 | * update. This is what guarantees that the socket activation |
Johannes Weiner | 2d75807 | 2016-10-07 17:00:58 -0700 | [diff] [blame] | 3703 | * function is the last one to run. See mem_cgroup_sk_alloc() |
| 3704 | * for details, and note that we don't mark any socket as |
| 3705 | * belonging to this memcg until that flag is up. |
Vladimir Davydov | d55f90b | 2016-01-20 15:02:44 -0800 | [diff] [blame] | 3706 | * |
| 3707 | * We need to do this, because static_keys will span multiple |
| 3708 | * sites, but we can't control their order. If we mark a socket |
| 3709 | * as accounted, but the accounting functions are not patched in |
| 3710 | * yet, we'll lose accounting. |
| 3711 | * |
Johannes Weiner | 2d75807 | 2016-10-07 17:00:58 -0700 | [diff] [blame] | 3712 | * We never race with the readers in mem_cgroup_sk_alloc(), |
Vladimir Davydov | d55f90b | 2016-01-20 15:02:44 -0800 | [diff] [blame] | 3713 | * because when this value change, the code to process it is not |
| 3714 | * patched in yet. |
| 3715 | */ |
| 3716 | static_branch_inc(&memcg_sockets_enabled_key); |
Johannes Weiner | 0db1529 | 2016-01-20 15:02:50 -0800 | [diff] [blame] | 3717 | memcg->tcpmem_active = true; |
Vladimir Davydov | d55f90b | 2016-01-20 15:02:44 -0800 | [diff] [blame] | 3718 | } |
| 3719 | out: |
Roman Gushchin | bbec2e1 | 2018-06-07 17:06:18 -0700 | [diff] [blame] | 3720 | mutex_unlock(&memcg_max_mutex); |
Vladimir Davydov | d55f90b | 2016-01-20 15:02:44 -0800 | [diff] [blame] | 3721 | return ret; |
| 3722 | } |
Vladimir Davydov | d55f90b | 2016-01-20 15:02:44 -0800 | [diff] [blame] | 3723 | |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 3724 | /* |
| 3725 | * The user of this function is... |
| 3726 | * RES_LIMIT. |
| 3727 | */ |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 3728 | static ssize_t mem_cgroup_write(struct kernfs_open_file *of, |
| 3729 | char *buf, size_t nbytes, loff_t off) |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 3730 | { |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 3731 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3732 | unsigned long nr_pages; |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 3733 | int ret; |
| 3734 | |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 3735 | buf = strstrip(buf); |
Johannes Weiner | 650c5e5 | 2015-02-11 15:26:03 -0800 | [diff] [blame] | 3736 | ret = page_counter_memparse(buf, "-1", &nr_pages); |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3737 | if (ret) |
| 3738 | return ret; |
Tejun Heo | af36f90 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 3739 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3740 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 3741 | case RES_LIMIT: |
Balbir Singh | 4b3bde4 | 2009-09-23 15:56:32 -0700 | [diff] [blame] | 3742 | if (mem_cgroup_is_root(memcg)) { /* Can't set limit on root */ |
| 3743 | ret = -EINVAL; |
| 3744 | break; |
| 3745 | } |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3746 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
| 3747 | case _MEM: |
Roman Gushchin | bbec2e1 | 2018-06-07 17:06:18 -0700 | [diff] [blame] | 3748 | ret = mem_cgroup_resize_max(memcg, nr_pages, false); |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 3749 | break; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3750 | case _MEMSWAP: |
Roman Gushchin | bbec2e1 | 2018-06-07 17:06:18 -0700 | [diff] [blame] | 3751 | ret = mem_cgroup_resize_max(memcg, nr_pages, true); |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3752 | break; |
| 3753 | case _KMEM: |
Michal Hocko | 0158115 | 2019-09-23 15:37:22 -0700 | [diff] [blame] | 3754 | pr_warn_once("kmem.limit_in_bytes is deprecated and will be removed. " |
| 3755 | "Please report your usecase to linux-mm@kvack.org if you " |
| 3756 | "depend on this functionality.\n"); |
Roman Gushchin | bbec2e1 | 2018-06-07 17:06:18 -0700 | [diff] [blame] | 3757 | ret = memcg_update_kmem_max(memcg, nr_pages); |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3758 | break; |
Vladimir Davydov | d55f90b | 2016-01-20 15:02:44 -0800 | [diff] [blame] | 3759 | case _TCP: |
Roman Gushchin | bbec2e1 | 2018-06-07 17:06:18 -0700 | [diff] [blame] | 3760 | ret = memcg_update_tcp_max(memcg, nr_pages); |
Vladimir Davydov | d55f90b | 2016-01-20 15:02:44 -0800 | [diff] [blame] | 3761 | break; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3762 | } |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 3763 | break; |
Balbir Singh | 296c81d | 2009-09-23 15:56:36 -0700 | [diff] [blame] | 3764 | case RES_SOFT_LIMIT: |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3765 | memcg->soft_limit = nr_pages; |
| 3766 | ret = 0; |
KAMEZAWA Hiroyuki | 628f423 | 2008-07-25 01:47:20 -0700 | [diff] [blame] | 3767 | break; |
| 3768 | } |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 3769 | return ret ?: nbytes; |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 3770 | } |
| 3771 | |
Tejun Heo | 6770c64 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 3772 | static ssize_t mem_cgroup_reset(struct kernfs_open_file *of, char *buf, |
| 3773 | size_t nbytes, loff_t off) |
Pavel Emelyanov | c84872e | 2008-04-29 01:00:17 -0700 | [diff] [blame] | 3774 | { |
Tejun Heo | 6770c64 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 3775 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3776 | struct page_counter *counter; |
Pavel Emelyanov | c84872e | 2008-04-29 01:00:17 -0700 | [diff] [blame] | 3777 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3778 | switch (MEMFILE_TYPE(of_cft(of)->private)) { |
| 3779 | case _MEM: |
| 3780 | counter = &memcg->memory; |
| 3781 | break; |
| 3782 | case _MEMSWAP: |
| 3783 | counter = &memcg->memsw; |
| 3784 | break; |
| 3785 | case _KMEM: |
| 3786 | counter = &memcg->kmem; |
| 3787 | break; |
Vladimir Davydov | d55f90b | 2016-01-20 15:02:44 -0800 | [diff] [blame] | 3788 | case _TCP: |
Johannes Weiner | 0db1529 | 2016-01-20 15:02:50 -0800 | [diff] [blame] | 3789 | counter = &memcg->tcpmem; |
Vladimir Davydov | d55f90b | 2016-01-20 15:02:44 -0800 | [diff] [blame] | 3790 | break; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3791 | default: |
| 3792 | BUG(); |
| 3793 | } |
Tejun Heo | af36f90 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 3794 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3795 | switch (MEMFILE_ATTR(of_cft(of)->private)) { |
Pavel Emelyanov | 29f2a4d | 2008-04-29 01:00:21 -0700 | [diff] [blame] | 3796 | case RES_MAX_USAGE: |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3797 | page_counter_reset_watermark(counter); |
Pavel Emelyanov | 29f2a4d | 2008-04-29 01:00:21 -0700 | [diff] [blame] | 3798 | break; |
| 3799 | case RES_FAILCNT: |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3800 | counter->failcnt = 0; |
Pavel Emelyanov | 29f2a4d | 2008-04-29 01:00:21 -0700 | [diff] [blame] | 3801 | break; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3802 | default: |
| 3803 | BUG(); |
Pavel Emelyanov | 29f2a4d | 2008-04-29 01:00:21 -0700 | [diff] [blame] | 3804 | } |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 3805 | |
Tejun Heo | 6770c64 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 3806 | return nbytes; |
Pavel Emelyanov | c84872e | 2008-04-29 01:00:17 -0700 | [diff] [blame] | 3807 | } |
| 3808 | |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 3809 | static u64 mem_cgroup_move_charge_read(struct cgroup_subsys_state *css, |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 3810 | struct cftype *cft) |
| 3811 | { |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 3812 | return mem_cgroup_from_css(css)->move_charge_at_immigrate; |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 3813 | } |
| 3814 | |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 3815 | #ifdef CONFIG_MMU |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 3816 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 3817 | struct cftype *cft, u64 val) |
| 3818 | { |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 3819 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 3820 | |
Johannes Weiner | 1dfab5a | 2015-02-11 15:26:09 -0800 | [diff] [blame] | 3821 | if (val & ~MOVE_MASK) |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 3822 | return -EINVAL; |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 3823 | |
Glauber Costa | ee5e847 | 2013-02-22 16:34:50 -0800 | [diff] [blame] | 3824 | /* |
| 3825 | * No kind of locking is needed in here, because ->can_attach() will |
| 3826 | * check this value once in the beginning of the process, and then carry |
| 3827 | * on with stale data. This means that changes to this value will only |
| 3828 | * affect task migrations starting after the change. |
| 3829 | */ |
| 3830 | memcg->move_charge_at_immigrate = val; |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 3831 | return 0; |
| 3832 | } |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 3833 | #else |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 3834 | static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css, |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 3835 | struct cftype *cft, u64 val) |
| 3836 | { |
| 3837 | return -ENOSYS; |
| 3838 | } |
| 3839 | #endif |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 3840 | |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 3841 | #ifdef CONFIG_NUMA |
Johannes Weiner | 113b7df | 2019-05-13 17:18:11 -0700 | [diff] [blame] | 3842 | |
| 3843 | #define LRU_ALL_FILE (BIT(LRU_INACTIVE_FILE) | BIT(LRU_ACTIVE_FILE)) |
| 3844 | #define LRU_ALL_ANON (BIT(LRU_INACTIVE_ANON) | BIT(LRU_ACTIVE_ANON)) |
| 3845 | #define LRU_ALL ((1 << NR_LRU_LISTS) - 1) |
| 3846 | |
| 3847 | static unsigned long mem_cgroup_node_nr_lru_pages(struct mem_cgroup *memcg, |
Shakeel Butt | dd8657b | 2020-06-03 15:56:24 -0700 | [diff] [blame] | 3848 | int nid, unsigned int lru_mask, bool tree) |
Johannes Weiner | 113b7df | 2019-05-13 17:18:11 -0700 | [diff] [blame] | 3849 | { |
Johannes Weiner | 867e5e1 | 2019-11-30 17:55:34 -0800 | [diff] [blame] | 3850 | struct lruvec *lruvec = mem_cgroup_lruvec(memcg, NODE_DATA(nid)); |
Johannes Weiner | 113b7df | 2019-05-13 17:18:11 -0700 | [diff] [blame] | 3851 | unsigned long nr = 0; |
| 3852 | enum lru_list lru; |
| 3853 | |
| 3854 | VM_BUG_ON((unsigned)nid >= nr_node_ids); |
| 3855 | |
| 3856 | for_each_lru(lru) { |
| 3857 | if (!(BIT(lru) & lru_mask)) |
| 3858 | continue; |
Shakeel Butt | dd8657b | 2020-06-03 15:56:24 -0700 | [diff] [blame] | 3859 | if (tree) |
| 3860 | nr += lruvec_page_state(lruvec, NR_LRU_BASE + lru); |
| 3861 | else |
| 3862 | nr += lruvec_page_state_local(lruvec, NR_LRU_BASE + lru); |
Johannes Weiner | 113b7df | 2019-05-13 17:18:11 -0700 | [diff] [blame] | 3863 | } |
| 3864 | return nr; |
| 3865 | } |
| 3866 | |
| 3867 | static unsigned long mem_cgroup_nr_lru_pages(struct mem_cgroup *memcg, |
Shakeel Butt | dd8657b | 2020-06-03 15:56:24 -0700 | [diff] [blame] | 3868 | unsigned int lru_mask, |
| 3869 | bool tree) |
Johannes Weiner | 113b7df | 2019-05-13 17:18:11 -0700 | [diff] [blame] | 3870 | { |
| 3871 | unsigned long nr = 0; |
| 3872 | enum lru_list lru; |
| 3873 | |
| 3874 | for_each_lru(lru) { |
| 3875 | if (!(BIT(lru) & lru_mask)) |
| 3876 | continue; |
Shakeel Butt | dd8657b | 2020-06-03 15:56:24 -0700 | [diff] [blame] | 3877 | if (tree) |
| 3878 | nr += memcg_page_state(memcg, NR_LRU_BASE + lru); |
| 3879 | else |
| 3880 | nr += memcg_page_state_local(memcg, NR_LRU_BASE + lru); |
Johannes Weiner | 113b7df | 2019-05-13 17:18:11 -0700 | [diff] [blame] | 3881 | } |
| 3882 | return nr; |
| 3883 | } |
| 3884 | |
Tejun Heo | 2da8ca8 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 3885 | static int memcg_numa_stat_show(struct seq_file *m, void *v) |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 3886 | { |
Greg Thelen | 25485de | 2013-11-12 15:07:40 -0800 | [diff] [blame] | 3887 | struct numa_stat { |
| 3888 | const char *name; |
| 3889 | unsigned int lru_mask; |
| 3890 | }; |
| 3891 | |
| 3892 | static const struct numa_stat stats[] = { |
| 3893 | { "total", LRU_ALL }, |
| 3894 | { "file", LRU_ALL_FILE }, |
| 3895 | { "anon", LRU_ALL_ANON }, |
| 3896 | { "unevictable", BIT(LRU_UNEVICTABLE) }, |
| 3897 | }; |
| 3898 | const struct numa_stat *stat; |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 3899 | int nid; |
Chris Down | aa9694b | 2019-03-05 15:45:52 -0800 | [diff] [blame] | 3900 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 3901 | |
Johannes Weiner | 2d146aa | 2021-04-29 22:56:26 -0700 | [diff] [blame] | 3902 | cgroup_rstat_flush(memcg->css.cgroup); |
| 3903 | |
Greg Thelen | 25485de | 2013-11-12 15:07:40 -0800 | [diff] [blame] | 3904 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
Shakeel Butt | dd8657b | 2020-06-03 15:56:24 -0700 | [diff] [blame] | 3905 | seq_printf(m, "%s=%lu", stat->name, |
| 3906 | mem_cgroup_nr_lru_pages(memcg, stat->lru_mask, |
| 3907 | false)); |
| 3908 | for_each_node_state(nid, N_MEMORY) |
| 3909 | seq_printf(m, " N%d=%lu", nid, |
| 3910 | mem_cgroup_node_nr_lru_pages(memcg, nid, |
| 3911 | stat->lru_mask, false)); |
Greg Thelen | 25485de | 2013-11-12 15:07:40 -0800 | [diff] [blame] | 3912 | seq_putc(m, '\n'); |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 3913 | } |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 3914 | |
Ying Han | 071aee1 | 2013-11-12 15:07:41 -0800 | [diff] [blame] | 3915 | for (stat = stats; stat < stats + ARRAY_SIZE(stats); stat++) { |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 3916 | |
Shakeel Butt | dd8657b | 2020-06-03 15:56:24 -0700 | [diff] [blame] | 3917 | seq_printf(m, "hierarchical_%s=%lu", stat->name, |
| 3918 | mem_cgroup_nr_lru_pages(memcg, stat->lru_mask, |
| 3919 | true)); |
| 3920 | for_each_node_state(nid, N_MEMORY) |
| 3921 | seq_printf(m, " N%d=%lu", nid, |
| 3922 | mem_cgroup_node_nr_lru_pages(memcg, nid, |
| 3923 | stat->lru_mask, true)); |
Ying Han | 071aee1 | 2013-11-12 15:07:41 -0800 | [diff] [blame] | 3924 | seq_putc(m, '\n'); |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 3925 | } |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 3926 | |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 3927 | return 0; |
| 3928 | } |
| 3929 | #endif /* CONFIG_NUMA */ |
| 3930 | |
Johannes Weiner | c8713d0 | 2019-07-11 20:55:59 -0700 | [diff] [blame] | 3931 | static const unsigned int memcg1_stats[] = { |
Johannes Weiner | 0d1c207 | 2020-06-03 16:01:54 -0700 | [diff] [blame] | 3932 | NR_FILE_PAGES, |
Johannes Weiner | be5d0a7 | 2020-06-03 16:01:57 -0700 | [diff] [blame] | 3933 | NR_ANON_MAPPED, |
Johannes Weiner | 468c398 | 2020-06-03 16:02:01 -0700 | [diff] [blame] | 3934 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
| 3935 | NR_ANON_THPS, |
| 3936 | #endif |
Johannes Weiner | c8713d0 | 2019-07-11 20:55:59 -0700 | [diff] [blame] | 3937 | NR_SHMEM, |
| 3938 | NR_FILE_MAPPED, |
| 3939 | NR_FILE_DIRTY, |
| 3940 | NR_WRITEBACK, |
| 3941 | MEMCG_SWAP, |
| 3942 | }; |
| 3943 | |
| 3944 | static const char *const memcg1_stat_names[] = { |
| 3945 | "cache", |
| 3946 | "rss", |
Johannes Weiner | 468c398 | 2020-06-03 16:02:01 -0700 | [diff] [blame] | 3947 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
Johannes Weiner | c8713d0 | 2019-07-11 20:55:59 -0700 | [diff] [blame] | 3948 | "rss_huge", |
Johannes Weiner | 468c398 | 2020-06-03 16:02:01 -0700 | [diff] [blame] | 3949 | #endif |
Johannes Weiner | c8713d0 | 2019-07-11 20:55:59 -0700 | [diff] [blame] | 3950 | "shmem", |
| 3951 | "mapped_file", |
| 3952 | "dirty", |
| 3953 | "writeback", |
| 3954 | "swap", |
| 3955 | }; |
| 3956 | |
Johannes Weiner | df0e53d | 2017-05-03 14:55:10 -0700 | [diff] [blame] | 3957 | /* Universal VM events cgroup1 shows, original sort order */ |
Greg Thelen | 8dd53fd | 2018-06-07 17:07:23 -0700 | [diff] [blame] | 3958 | static const unsigned int memcg1_events[] = { |
Johannes Weiner | df0e53d | 2017-05-03 14:55:10 -0700 | [diff] [blame] | 3959 | PGPGIN, |
| 3960 | PGPGOUT, |
| 3961 | PGFAULT, |
| 3962 | PGMAJFAULT, |
| 3963 | }; |
| 3964 | |
Tejun Heo | 2da8ca8 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 3965 | static int memcg_stat_show(struct seq_file *m, void *v) |
KAMEZAWA Hiroyuki | d2ceb9b | 2008-02-07 00:14:25 -0800 | [diff] [blame] | 3966 | { |
Chris Down | aa9694b | 2019-03-05 15:45:52 -0800 | [diff] [blame] | 3967 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3968 | unsigned long memory, memsw; |
Johannes Weiner | af7c4b0 | 2012-05-29 15:07:08 -0700 | [diff] [blame] | 3969 | struct mem_cgroup *mi; |
| 3970 | unsigned int i; |
KAMEZAWA Hiroyuki | d2ceb9b | 2008-02-07 00:14:25 -0800 | [diff] [blame] | 3971 | |
Johannes Weiner | 71cd311 | 2017-05-03 14:55:13 -0700 | [diff] [blame] | 3972 | BUILD_BUG_ON(ARRAY_SIZE(memcg1_stat_names) != ARRAY_SIZE(memcg1_stats)); |
Rickard Strandqvist | 70bc068 | 2014-12-12 16:56:41 -0800 | [diff] [blame] | 3973 | |
Johannes Weiner | 2d146aa | 2021-04-29 22:56:26 -0700 | [diff] [blame] | 3974 | cgroup_rstat_flush(memcg->css.cgroup); |
| 3975 | |
Johannes Weiner | 71cd311 | 2017-05-03 14:55:13 -0700 | [diff] [blame] | 3976 | for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) { |
Johannes Weiner | 468c398 | 2020-06-03 16:02:01 -0700 | [diff] [blame] | 3977 | unsigned long nr; |
| 3978 | |
Johannes Weiner | 71cd311 | 2017-05-03 14:55:13 -0700 | [diff] [blame] | 3979 | if (memcg1_stats[i] == MEMCG_SWAP && !do_memsw_account()) |
Daisuke Nishimura | 1dd3a27 | 2009-09-23 15:56:43 -0700 | [diff] [blame] | 3980 | continue; |
Johannes Weiner | 468c398 | 2020-06-03 16:02:01 -0700 | [diff] [blame] | 3981 | nr = memcg_page_state_local(memcg, memcg1_stats[i]); |
Johannes Weiner | 468c398 | 2020-06-03 16:02:01 -0700 | [diff] [blame] | 3982 | seq_printf(m, "%s %lu\n", memcg1_stat_names[i], nr * PAGE_SIZE); |
Daisuke Nishimura | 1dd3a27 | 2009-09-23 15:56:43 -0700 | [diff] [blame] | 3983 | } |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 3984 | |
Johannes Weiner | df0e53d | 2017-05-03 14:55:10 -0700 | [diff] [blame] | 3985 | for (i = 0; i < ARRAY_SIZE(memcg1_events); i++) |
Konstantin Khlebnikov | ebc5d83d | 2019-12-04 16:49:53 -0800 | [diff] [blame] | 3986 | seq_printf(m, "%s %lu\n", vm_event_name(memcg1_events[i]), |
Johannes Weiner | 205b20c | 2019-05-14 15:47:06 -0700 | [diff] [blame] | 3987 | memcg_events_local(memcg, memcg1_events[i])); |
Johannes Weiner | af7c4b0 | 2012-05-29 15:07:08 -0700 | [diff] [blame] | 3988 | |
| 3989 | for (i = 0; i < NR_LRU_LISTS; i++) |
Konstantin Khlebnikov | ebc5d83d | 2019-12-04 16:49:53 -0800 | [diff] [blame] | 3990 | seq_printf(m, "%s %lu\n", lru_list_name(i), |
Johannes Weiner | 205b20c | 2019-05-14 15:47:06 -0700 | [diff] [blame] | 3991 | memcg_page_state_local(memcg, NR_LRU_BASE + i) * |
Johannes Weiner | 21d89d1 | 2019-05-13 17:18:08 -0700 | [diff] [blame] | 3992 | PAGE_SIZE); |
Johannes Weiner | af7c4b0 | 2012-05-29 15:07:08 -0700 | [diff] [blame] | 3993 | |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 3994 | /* Hierarchical information */ |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 3995 | memory = memsw = PAGE_COUNTER_MAX; |
| 3996 | for (mi = memcg; mi; mi = parent_mem_cgroup(mi)) { |
Chris Down | 15b4256 | 2020-04-01 21:07:20 -0700 | [diff] [blame] | 3997 | memory = min(memory, READ_ONCE(mi->memory.max)); |
| 3998 | memsw = min(memsw, READ_ONCE(mi->memsw.max)); |
KAMEZAWA Hiroyuki | fee7b54 | 2009-01-07 18:08:26 -0800 | [diff] [blame] | 3999 | } |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 4000 | seq_printf(m, "hierarchical_memory_limit %llu\n", |
| 4001 | (u64)memory * PAGE_SIZE); |
Johannes Weiner | 7941d21 | 2016-01-14 15:21:23 -0800 | [diff] [blame] | 4002 | if (do_memsw_account()) |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 4003 | seq_printf(m, "hierarchical_memsw_limit %llu\n", |
| 4004 | (u64)memsw * PAGE_SIZE); |
KOSAKI Motohiro | 7f016ee | 2009-01-07 18:08:22 -0800 | [diff] [blame] | 4005 | |
Shakeel Butt | 8de7ecc6 | 2018-08-21 21:53:17 -0700 | [diff] [blame] | 4006 | for (i = 0; i < ARRAY_SIZE(memcg1_stats); i++) { |
zhongjiang-ali | 7de2e9f | 2020-11-01 17:07:30 -0800 | [diff] [blame] | 4007 | unsigned long nr; |
| 4008 | |
Johannes Weiner | 71cd311 | 2017-05-03 14:55:13 -0700 | [diff] [blame] | 4009 | if (memcg1_stats[i] == MEMCG_SWAP && !do_memsw_account()) |
Daisuke Nishimura | 1dd3a27 | 2009-09-23 15:56:43 -0700 | [diff] [blame] | 4010 | continue; |
zhongjiang-ali | 7de2e9f | 2020-11-01 17:07:30 -0800 | [diff] [blame] | 4011 | nr = memcg_page_state(memcg, memcg1_stats[i]); |
Shakeel Butt | 8de7ecc6 | 2018-08-21 21:53:17 -0700 | [diff] [blame] | 4012 | seq_printf(m, "total_%s %llu\n", memcg1_stat_names[i], |
zhongjiang-ali | 7de2e9f | 2020-11-01 17:07:30 -0800 | [diff] [blame] | 4013 | (u64)nr * PAGE_SIZE); |
Johannes Weiner | af7c4b0 | 2012-05-29 15:07:08 -0700 | [diff] [blame] | 4014 | } |
| 4015 | |
Shakeel Butt | 8de7ecc6 | 2018-08-21 21:53:17 -0700 | [diff] [blame] | 4016 | for (i = 0; i < ARRAY_SIZE(memcg1_events); i++) |
Konstantin Khlebnikov | ebc5d83d | 2019-12-04 16:49:53 -0800 | [diff] [blame] | 4017 | seq_printf(m, "total_%s %llu\n", |
| 4018 | vm_event_name(memcg1_events[i]), |
Yafang Shao | dd92399 | 2019-07-11 20:52:11 -0700 | [diff] [blame] | 4019 | (u64)memcg_events(memcg, memcg1_events[i])); |
Johannes Weiner | af7c4b0 | 2012-05-29 15:07:08 -0700 | [diff] [blame] | 4020 | |
Shakeel Butt | 8de7ecc6 | 2018-08-21 21:53:17 -0700 | [diff] [blame] | 4021 | for (i = 0; i < NR_LRU_LISTS; i++) |
Konstantin Khlebnikov | ebc5d83d | 2019-12-04 16:49:53 -0800 | [diff] [blame] | 4022 | seq_printf(m, "total_%s %llu\n", lru_list_name(i), |
Johannes Weiner | 42a3003 | 2019-05-14 15:47:12 -0700 | [diff] [blame] | 4023 | (u64)memcg_page_state(memcg, NR_LRU_BASE + i) * |
| 4024 | PAGE_SIZE); |
KAMEZAWA Hiroyuki | 14067bb | 2009-04-02 16:57:35 -0700 | [diff] [blame] | 4025 | |
KOSAKI Motohiro | 7f016ee | 2009-01-07 18:08:22 -0800 | [diff] [blame] | 4026 | #ifdef CONFIG_DEBUG_VM |
KOSAKI Motohiro | 7f016ee | 2009-01-07 18:08:22 -0800 | [diff] [blame] | 4027 | { |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 4028 | pg_data_t *pgdat; |
| 4029 | struct mem_cgroup_per_node *mz; |
Johannes Weiner | 1431d4d | 2020-06-03 16:02:53 -0700 | [diff] [blame] | 4030 | unsigned long anon_cost = 0; |
| 4031 | unsigned long file_cost = 0; |
KOSAKI Motohiro | 7f016ee | 2009-01-07 18:08:22 -0800 | [diff] [blame] | 4032 | |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 4033 | for_each_online_pgdat(pgdat) { |
Johannes Weiner | a3747b5 | 2021-04-29 22:56:14 -0700 | [diff] [blame] | 4034 | mz = memcg->nodeinfo[pgdat->node_id]; |
KOSAKI Motohiro | 7f016ee | 2009-01-07 18:08:22 -0800 | [diff] [blame] | 4035 | |
Johannes Weiner | 1431d4d | 2020-06-03 16:02:53 -0700 | [diff] [blame] | 4036 | anon_cost += mz->lruvec.anon_cost; |
| 4037 | file_cost += mz->lruvec.file_cost; |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 4038 | } |
Johannes Weiner | 1431d4d | 2020-06-03 16:02:53 -0700 | [diff] [blame] | 4039 | seq_printf(m, "anon_cost %lu\n", anon_cost); |
| 4040 | seq_printf(m, "file_cost %lu\n", file_cost); |
KOSAKI Motohiro | 7f016ee | 2009-01-07 18:08:22 -0800 | [diff] [blame] | 4041 | } |
| 4042 | #endif |
| 4043 | |
KAMEZAWA Hiroyuki | d2ceb9b | 2008-02-07 00:14:25 -0800 | [diff] [blame] | 4044 | return 0; |
| 4045 | } |
| 4046 | |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 4047 | static u64 mem_cgroup_swappiness_read(struct cgroup_subsys_state *css, |
| 4048 | struct cftype *cft) |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 4049 | { |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 4050 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 4051 | |
KAMEZAWA Hiroyuki | 1f4c025 | 2011-07-26 16:08:21 -0700 | [diff] [blame] | 4052 | return mem_cgroup_swappiness(memcg); |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 4053 | } |
| 4054 | |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 4055 | static int mem_cgroup_swappiness_write(struct cgroup_subsys_state *css, |
| 4056 | struct cftype *cft, u64 val) |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 4057 | { |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 4058 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
Li Zefan | 068b38c | 2009-01-15 13:51:26 -0800 | [diff] [blame] | 4059 | |
Johannes Weiner | 3dae7fe | 2014-06-04 16:07:01 -0700 | [diff] [blame] | 4060 | if (val > 100) |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 4061 | return -EINVAL; |
| 4062 | |
Shakeel Butt | a479203 | 2021-04-29 22:56:05 -0700 | [diff] [blame] | 4063 | if (!mem_cgroup_is_root(memcg)) |
Johannes Weiner | 3dae7fe | 2014-06-04 16:07:01 -0700 | [diff] [blame] | 4064 | memcg->swappiness = val; |
| 4065 | else |
| 4066 | vm_swappiness = val; |
Li Zefan | 068b38c | 2009-01-15 13:51:26 -0800 | [diff] [blame] | 4067 | |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 4068 | return 0; |
| 4069 | } |
| 4070 | |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4071 | static void __mem_cgroup_threshold(struct mem_cgroup *memcg, bool swap) |
| 4072 | { |
| 4073 | struct mem_cgroup_threshold_ary *t; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 4074 | unsigned long usage; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4075 | int i; |
| 4076 | |
| 4077 | rcu_read_lock(); |
| 4078 | if (!swap) |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4079 | t = rcu_dereference(memcg->thresholds.primary); |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4080 | else |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4081 | t = rcu_dereference(memcg->memsw_thresholds.primary); |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4082 | |
| 4083 | if (!t) |
| 4084 | goto unlock; |
| 4085 | |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 4086 | usage = mem_cgroup_usage(memcg, swap); |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4087 | |
| 4088 | /* |
Sha Zhengju | 748dad3 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 4089 | * current_threshold points to threshold just below or equal to usage. |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4090 | * If it's not true, a threshold was crossed after last |
| 4091 | * call of __mem_cgroup_threshold(). |
| 4092 | */ |
Phil Carmody | 5407a56 | 2010-05-26 14:42:42 -0700 | [diff] [blame] | 4093 | i = t->current_threshold; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4094 | |
| 4095 | /* |
| 4096 | * Iterate backward over array of thresholds starting from |
| 4097 | * current_threshold and check if a threshold is crossed. |
| 4098 | * If none of thresholds below usage is crossed, we read |
| 4099 | * only one element of the array here. |
| 4100 | */ |
| 4101 | for (; i >= 0 && unlikely(t->entries[i].threshold > usage); i--) |
| 4102 | eventfd_signal(t->entries[i].eventfd, 1); |
| 4103 | |
| 4104 | /* i = current_threshold + 1 */ |
| 4105 | i++; |
| 4106 | |
| 4107 | /* |
| 4108 | * Iterate forward over array of thresholds starting from |
| 4109 | * current_threshold+1 and check if a threshold is crossed. |
| 4110 | * If none of thresholds above usage is crossed, we read |
| 4111 | * only one element of the array here. |
| 4112 | */ |
| 4113 | for (; i < t->size && unlikely(t->entries[i].threshold <= usage); i++) |
| 4114 | eventfd_signal(t->entries[i].eventfd, 1); |
| 4115 | |
| 4116 | /* Update current_threshold */ |
Phil Carmody | 5407a56 | 2010-05-26 14:42:42 -0700 | [diff] [blame] | 4117 | t->current_threshold = i - 1; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4118 | unlock: |
| 4119 | rcu_read_unlock(); |
| 4120 | } |
| 4121 | |
| 4122 | static void mem_cgroup_threshold(struct mem_cgroup *memcg) |
| 4123 | { |
Kirill A. Shutemov | ad4ca5f | 2010-10-07 12:59:27 -0700 | [diff] [blame] | 4124 | while (memcg) { |
| 4125 | __mem_cgroup_threshold(memcg, false); |
Johannes Weiner | 7941d21 | 2016-01-14 15:21:23 -0800 | [diff] [blame] | 4126 | if (do_memsw_account()) |
Kirill A. Shutemov | ad4ca5f | 2010-10-07 12:59:27 -0700 | [diff] [blame] | 4127 | __mem_cgroup_threshold(memcg, true); |
| 4128 | |
| 4129 | memcg = parent_mem_cgroup(memcg); |
| 4130 | } |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4131 | } |
| 4132 | |
| 4133 | static int compare_thresholds(const void *a, const void *b) |
| 4134 | { |
| 4135 | const struct mem_cgroup_threshold *_a = a; |
| 4136 | const struct mem_cgroup_threshold *_b = b; |
| 4137 | |
Greg Thelen | 2bff24a | 2013-09-11 14:23:08 -0700 | [diff] [blame] | 4138 | if (_a->threshold > _b->threshold) |
| 4139 | return 1; |
| 4140 | |
| 4141 | if (_a->threshold < _b->threshold) |
| 4142 | return -1; |
| 4143 | |
| 4144 | return 0; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4145 | } |
| 4146 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4147 | static int mem_cgroup_oom_notify_cb(struct mem_cgroup *memcg) |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 4148 | { |
| 4149 | struct mem_cgroup_eventfd_list *ev; |
| 4150 | |
Michal Hocko | 2bcf2e9 | 2014-07-30 16:08:33 -0700 | [diff] [blame] | 4151 | spin_lock(&memcg_oom_lock); |
| 4152 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4153 | list_for_each_entry(ev, &memcg->oom_notify, list) |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 4154 | eventfd_signal(ev->eventfd, 1); |
Michal Hocko | 2bcf2e9 | 2014-07-30 16:08:33 -0700 | [diff] [blame] | 4155 | |
| 4156 | spin_unlock(&memcg_oom_lock); |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 4157 | return 0; |
| 4158 | } |
| 4159 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4160 | static void mem_cgroup_oom_notify(struct mem_cgroup *memcg) |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 4161 | { |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 4162 | struct mem_cgroup *iter; |
| 4163 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4164 | for_each_mem_cgroup_tree(iter, memcg) |
KAMEZAWA Hiroyuki | 7d74b06 | 2010-10-27 15:33:41 -0700 | [diff] [blame] | 4165 | mem_cgroup_oom_notify_cb(iter); |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 4166 | } |
| 4167 | |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4168 | static int __mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4169 | struct eventfd_ctx *eventfd, const char *args, enum res_type type) |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4170 | { |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4171 | struct mem_cgroup_thresholds *thresholds; |
| 4172 | struct mem_cgroup_threshold_ary *new; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 4173 | unsigned long threshold; |
| 4174 | unsigned long usage; |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4175 | int i, size, ret; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4176 | |
Johannes Weiner | 650c5e5 | 2015-02-11 15:26:03 -0800 | [diff] [blame] | 4177 | ret = page_counter_memparse(args, "-1", &threshold); |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4178 | if (ret) |
| 4179 | return ret; |
| 4180 | |
| 4181 | mutex_lock(&memcg->thresholds_lock); |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4182 | |
Johannes Weiner | 05b8430 | 2014-08-06 16:05:59 -0700 | [diff] [blame] | 4183 | if (type == _MEM) { |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4184 | thresholds = &memcg->thresholds; |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 4185 | usage = mem_cgroup_usage(memcg, false); |
Johannes Weiner | 05b8430 | 2014-08-06 16:05:59 -0700 | [diff] [blame] | 4186 | } else if (type == _MEMSWAP) { |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4187 | thresholds = &memcg->memsw_thresholds; |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 4188 | usage = mem_cgroup_usage(memcg, true); |
Johannes Weiner | 05b8430 | 2014-08-06 16:05:59 -0700 | [diff] [blame] | 4189 | } else |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4190 | BUG(); |
| 4191 | |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4192 | /* Check if a threshold crossed before adding a new one */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4193 | if (thresholds->primary) |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4194 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
| 4195 | |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4196 | size = thresholds->primary ? thresholds->primary->size + 1 : 1; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4197 | |
| 4198 | /* Allocate memory for new array of thresholds */ |
Gustavo A. R. Silva | 67b8046 | 2019-03-05 15:44:05 -0800 | [diff] [blame] | 4199 | new = kmalloc(struct_size(new, entries, size), GFP_KERNEL); |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4200 | if (!new) { |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4201 | ret = -ENOMEM; |
| 4202 | goto unlock; |
| 4203 | } |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4204 | new->size = size; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4205 | |
| 4206 | /* Copy thresholds (if any) to new array */ |
Gustavo A. R. Silva | e90342e | 2020-10-13 16:52:36 -0700 | [diff] [blame] | 4207 | if (thresholds->primary) |
| 4208 | memcpy(new->entries, thresholds->primary->entries, |
| 4209 | flex_array_size(new, entries, size - 1)); |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4210 | |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4211 | /* Add new threshold */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4212 | new->entries[size - 1].eventfd = eventfd; |
| 4213 | new->entries[size - 1].threshold = threshold; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4214 | |
| 4215 | /* Sort thresholds. Registering of new threshold isn't time-critical */ |
Gustavo A. R. Silva | 61e604e6 | 2020-10-13 16:52:39 -0700 | [diff] [blame] | 4216 | sort(new->entries, size, sizeof(*new->entries), |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4217 | compare_thresholds, NULL); |
| 4218 | |
| 4219 | /* Find current threshold */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4220 | new->current_threshold = -1; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4221 | for (i = 0; i < size; i++) { |
Sha Zhengju | 748dad3 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 4222 | if (new->entries[i].threshold <= usage) { |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4223 | /* |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4224 | * new->current_threshold will not be used until |
| 4225 | * rcu_assign_pointer(), so it's safe to increment |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4226 | * it here. |
| 4227 | */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4228 | ++new->current_threshold; |
Sha Zhengju | 748dad3 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 4229 | } else |
| 4230 | break; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4231 | } |
| 4232 | |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4233 | /* Free old spare buffer and save old primary buffer as spare */ |
| 4234 | kfree(thresholds->spare); |
| 4235 | thresholds->spare = thresholds->primary; |
| 4236 | |
| 4237 | rcu_assign_pointer(thresholds->primary, new); |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4238 | |
Kirill A. Shutemov | 907860e | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 4239 | /* To be sure that nobody uses thresholds */ |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4240 | synchronize_rcu(); |
| 4241 | |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4242 | unlock: |
| 4243 | mutex_unlock(&memcg->thresholds_lock); |
| 4244 | |
| 4245 | return ret; |
| 4246 | } |
| 4247 | |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4248 | static int mem_cgroup_usage_register_event(struct mem_cgroup *memcg, |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4249 | struct eventfd_ctx *eventfd, const char *args) |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4250 | { |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4251 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEM); |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4252 | } |
| 4253 | |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4254 | static int memsw_cgroup_usage_register_event(struct mem_cgroup *memcg, |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4255 | struct eventfd_ctx *eventfd, const char *args) |
| 4256 | { |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4257 | return __mem_cgroup_usage_register_event(memcg, eventfd, args, _MEMSWAP); |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4258 | } |
| 4259 | |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4260 | static void __mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4261 | struct eventfd_ctx *eventfd, enum res_type type) |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4262 | { |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4263 | struct mem_cgroup_thresholds *thresholds; |
| 4264 | struct mem_cgroup_threshold_ary *new; |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 4265 | unsigned long usage; |
Chunguang Xu | 7d36665 | 2020-03-21 18:22:10 -0700 | [diff] [blame] | 4266 | int i, j, size, entries; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4267 | |
| 4268 | mutex_lock(&memcg->thresholds_lock); |
Johannes Weiner | 05b8430 | 2014-08-06 16:05:59 -0700 | [diff] [blame] | 4269 | |
| 4270 | if (type == _MEM) { |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4271 | thresholds = &memcg->thresholds; |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 4272 | usage = mem_cgroup_usage(memcg, false); |
Johannes Weiner | 05b8430 | 2014-08-06 16:05:59 -0700 | [diff] [blame] | 4273 | } else if (type == _MEMSWAP) { |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4274 | thresholds = &memcg->memsw_thresholds; |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 4275 | usage = mem_cgroup_usage(memcg, true); |
Johannes Weiner | 05b8430 | 2014-08-06 16:05:59 -0700 | [diff] [blame] | 4276 | } else |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4277 | BUG(); |
| 4278 | |
Anton Vorontsov | 371528c | 2012-02-24 05:14:46 +0400 | [diff] [blame] | 4279 | if (!thresholds->primary) |
| 4280 | goto unlock; |
| 4281 | |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4282 | /* Check if a threshold crossed before removing */ |
| 4283 | __mem_cgroup_threshold(memcg, type == _MEMSWAP); |
| 4284 | |
| 4285 | /* Calculate new number of threshold */ |
Chunguang Xu | 7d36665 | 2020-03-21 18:22:10 -0700 | [diff] [blame] | 4286 | size = entries = 0; |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4287 | for (i = 0; i < thresholds->primary->size; i++) { |
| 4288 | if (thresholds->primary->entries[i].eventfd != eventfd) |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4289 | size++; |
Chunguang Xu | 7d36665 | 2020-03-21 18:22:10 -0700 | [diff] [blame] | 4290 | else |
| 4291 | entries++; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4292 | } |
| 4293 | |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4294 | new = thresholds->spare; |
Kirill A. Shutemov | 907860e | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 4295 | |
Chunguang Xu | 7d36665 | 2020-03-21 18:22:10 -0700 | [diff] [blame] | 4296 | /* If no items related to eventfd have been cleared, nothing to do */ |
| 4297 | if (!entries) |
| 4298 | goto unlock; |
| 4299 | |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4300 | /* Set thresholds array to NULL if we don't have thresholds */ |
| 4301 | if (!size) { |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4302 | kfree(new); |
| 4303 | new = NULL; |
Kirill A. Shutemov | 907860e | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 4304 | goto swap_buffers; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4305 | } |
| 4306 | |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4307 | new->size = size; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4308 | |
| 4309 | /* Copy thresholds and find current threshold */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4310 | new->current_threshold = -1; |
| 4311 | for (i = 0, j = 0; i < thresholds->primary->size; i++) { |
| 4312 | if (thresholds->primary->entries[i].eventfd == eventfd) |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4313 | continue; |
| 4314 | |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4315 | new->entries[j] = thresholds->primary->entries[i]; |
Sha Zhengju | 748dad3 | 2012-05-29 15:06:57 -0700 | [diff] [blame] | 4316 | if (new->entries[j].threshold <= usage) { |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4317 | /* |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4318 | * new->current_threshold will not be used |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4319 | * until rcu_assign_pointer(), so it's safe to increment |
| 4320 | * it here. |
| 4321 | */ |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4322 | ++new->current_threshold; |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4323 | } |
| 4324 | j++; |
| 4325 | } |
| 4326 | |
Kirill A. Shutemov | 907860e | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 4327 | swap_buffers: |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4328 | /* Swap primary and spare array */ |
| 4329 | thresholds->spare = thresholds->primary; |
Sha Zhengju | 8c75776 | 2012-05-10 13:01:45 -0700 | [diff] [blame] | 4330 | |
Kirill A. Shutemov | 2c488db | 2010-05-26 14:42:47 -0700 | [diff] [blame] | 4331 | rcu_assign_pointer(thresholds->primary, new); |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4332 | |
Kirill A. Shutemov | 907860e | 2010-05-26 14:42:46 -0700 | [diff] [blame] | 4333 | /* To be sure that nobody uses thresholds */ |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4334 | synchronize_rcu(); |
Martijn Coenen | 6611d8d | 2016-01-15 16:57:49 -0800 | [diff] [blame] | 4335 | |
| 4336 | /* If all events are unregistered, free the spare array */ |
| 4337 | if (!new) { |
| 4338 | kfree(thresholds->spare); |
| 4339 | thresholds->spare = NULL; |
| 4340 | } |
Anton Vorontsov | 371528c | 2012-02-24 05:14:46 +0400 | [diff] [blame] | 4341 | unlock: |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4342 | mutex_unlock(&memcg->thresholds_lock); |
Kirill A. Shutemov | 2e72b63 | 2010-03-10 15:22:24 -0800 | [diff] [blame] | 4343 | } |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 4344 | |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4345 | static void mem_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4346 | struct eventfd_ctx *eventfd) |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 4347 | { |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4348 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEM); |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4349 | } |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 4350 | |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4351 | static void memsw_cgroup_usage_unregister_event(struct mem_cgroup *memcg, |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4352 | struct eventfd_ctx *eventfd) |
| 4353 | { |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4354 | return __mem_cgroup_usage_unregister_event(memcg, eventfd, _MEMSWAP); |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4355 | } |
| 4356 | |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4357 | static int mem_cgroup_oom_register_event(struct mem_cgroup *memcg, |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4358 | struct eventfd_ctx *eventfd, const char *args) |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 4359 | { |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 4360 | struct mem_cgroup_eventfd_list *event; |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 4361 | |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 4362 | event = kmalloc(sizeof(*event), GFP_KERNEL); |
| 4363 | if (!event) |
| 4364 | return -ENOMEM; |
| 4365 | |
Michal Hocko | 1af8efe | 2011-07-26 16:08:24 -0700 | [diff] [blame] | 4366 | spin_lock(&memcg_oom_lock); |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 4367 | |
| 4368 | event->eventfd = eventfd; |
| 4369 | list_add(&event->list, &memcg->oom_notify); |
| 4370 | |
| 4371 | /* already in OOM ? */ |
Tejun Heo | c2b42d3 | 2015-06-24 16:58:23 -0700 | [diff] [blame] | 4372 | if (memcg->under_oom) |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 4373 | eventfd_signal(eventfd, 1); |
Michal Hocko | 1af8efe | 2011-07-26 16:08:24 -0700 | [diff] [blame] | 4374 | spin_unlock(&memcg_oom_lock); |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 4375 | |
| 4376 | return 0; |
| 4377 | } |
| 4378 | |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4379 | static void mem_cgroup_oom_unregister_event(struct mem_cgroup *memcg, |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4380 | struct eventfd_ctx *eventfd) |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 4381 | { |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 4382 | struct mem_cgroup_eventfd_list *ev, *tmp; |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 4383 | |
Michal Hocko | 1af8efe | 2011-07-26 16:08:24 -0700 | [diff] [blame] | 4384 | spin_lock(&memcg_oom_lock); |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 4385 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4386 | list_for_each_entry_safe(ev, tmp, &memcg->oom_notify, list) { |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 4387 | if (ev->eventfd == eventfd) { |
| 4388 | list_del(&ev->list); |
| 4389 | kfree(ev); |
| 4390 | } |
| 4391 | } |
| 4392 | |
Michal Hocko | 1af8efe | 2011-07-26 16:08:24 -0700 | [diff] [blame] | 4393 | spin_unlock(&memcg_oom_lock); |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 4394 | } |
| 4395 | |
Tejun Heo | 2da8ca8 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 4396 | 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] | 4397 | { |
Chris Down | aa9694b | 2019-03-05 15:45:52 -0800 | [diff] [blame] | 4398 | struct mem_cgroup *memcg = mem_cgroup_from_seq(sf); |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 4399 | |
Tejun Heo | 791badb | 2013-12-05 12:28:02 -0500 | [diff] [blame] | 4400 | seq_printf(sf, "oom_kill_disable %d\n", memcg->oom_kill_disable); |
Tejun Heo | c2b42d3 | 2015-06-24 16:58:23 -0700 | [diff] [blame] | 4401 | seq_printf(sf, "under_oom %d\n", (bool)memcg->under_oom); |
Roman Gushchin | fe6bdfc | 2018-06-14 15:28:05 -0700 | [diff] [blame] | 4402 | seq_printf(sf, "oom_kill %lu\n", |
| 4403 | atomic_long_read(&memcg->memory_events[MEMCG_OOM_KILL])); |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 4404 | return 0; |
| 4405 | } |
| 4406 | |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 4407 | static int mem_cgroup_oom_control_write(struct cgroup_subsys_state *css, |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 4408 | struct cftype *cft, u64 val) |
| 4409 | { |
Tejun Heo | 182446d | 2013-08-08 20:11:24 -0400 | [diff] [blame] | 4410 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 4411 | |
| 4412 | /* cannot set to root cgroup and only 0 and 1 are allowed */ |
Shakeel Butt | a479203 | 2021-04-29 22:56:05 -0700 | [diff] [blame] | 4413 | if (mem_cgroup_is_root(memcg) || !((val == 0) || (val == 1))) |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 4414 | return -EINVAL; |
| 4415 | |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4416 | memcg->oom_kill_disable = val; |
KAMEZAWA Hiroyuki | 4d845eb | 2010-06-29 15:05:18 -0700 | [diff] [blame] | 4417 | if (!val) |
Raghavendra K T | c0ff4b8 | 2011-11-02 13:38:15 -0700 | [diff] [blame] | 4418 | memcg_oom_recover(memcg); |
Johannes Weiner | 3dae7fe | 2014-06-04 16:07:01 -0700 | [diff] [blame] | 4419 | |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 4420 | return 0; |
| 4421 | } |
| 4422 | |
Tejun Heo | 52ebea7 | 2015-05-22 17:13:37 -0400 | [diff] [blame] | 4423 | #ifdef CONFIG_CGROUP_WRITEBACK |
| 4424 | |
Tejun Heo | 3a8e9ac | 2019-08-29 15:47:19 -0700 | [diff] [blame] | 4425 | #include <trace/events/writeback.h> |
| 4426 | |
Tejun Heo | 841710a | 2015-05-22 18:23:33 -0400 | [diff] [blame] | 4427 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) |
| 4428 | { |
| 4429 | return wb_domain_init(&memcg->cgwb_domain, gfp); |
| 4430 | } |
| 4431 | |
| 4432 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) |
| 4433 | { |
| 4434 | wb_domain_exit(&memcg->cgwb_domain); |
| 4435 | } |
| 4436 | |
Tejun Heo | 2529bb3 | 2015-05-22 18:23:34 -0400 | [diff] [blame] | 4437 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
| 4438 | { |
| 4439 | wb_domain_size_changed(&memcg->cgwb_domain); |
| 4440 | } |
| 4441 | |
Tejun Heo | 841710a | 2015-05-22 18:23:33 -0400 | [diff] [blame] | 4442 | struct wb_domain *mem_cgroup_wb_domain(struct bdi_writeback *wb) |
| 4443 | { |
| 4444 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); |
| 4445 | |
| 4446 | if (!memcg->css.parent) |
| 4447 | return NULL; |
| 4448 | |
| 4449 | return &memcg->cgwb_domain; |
| 4450 | } |
| 4451 | |
Tejun Heo | c2aa723 | 2015-05-22 18:23:35 -0400 | [diff] [blame] | 4452 | /** |
| 4453 | * mem_cgroup_wb_stats - retrieve writeback related stats from its memcg |
| 4454 | * @wb: bdi_writeback in question |
Tejun Heo | c5edf9c | 2015-09-29 13:04:26 -0400 | [diff] [blame] | 4455 | * @pfilepages: out parameter for number of file pages |
| 4456 | * @pheadroom: out parameter for number of allocatable pages according to memcg |
Tejun Heo | c2aa723 | 2015-05-22 18:23:35 -0400 | [diff] [blame] | 4457 | * @pdirty: out parameter for number of dirty pages |
| 4458 | * @pwriteback: out parameter for number of pages under writeback |
| 4459 | * |
Tejun Heo | c5edf9c | 2015-09-29 13:04:26 -0400 | [diff] [blame] | 4460 | * Determine the numbers of file, headroom, dirty, and writeback pages in |
| 4461 | * @wb's memcg. File, dirty and writeback are self-explanatory. Headroom |
| 4462 | * is a bit more involved. |
Tejun Heo | c2aa723 | 2015-05-22 18:23:35 -0400 | [diff] [blame] | 4463 | * |
Tejun Heo | c5edf9c | 2015-09-29 13:04:26 -0400 | [diff] [blame] | 4464 | * A memcg's headroom is "min(max, high) - used". In the hierarchy, the |
| 4465 | * headroom is calculated as the lowest headroom of itself and the |
| 4466 | * ancestors. Note that this doesn't consider the actual amount of |
| 4467 | * available memory in the system. The caller should further cap |
| 4468 | * *@pheadroom accordingly. |
Tejun Heo | c2aa723 | 2015-05-22 18:23:35 -0400 | [diff] [blame] | 4469 | */ |
Tejun Heo | c5edf9c | 2015-09-29 13:04:26 -0400 | [diff] [blame] | 4470 | void mem_cgroup_wb_stats(struct bdi_writeback *wb, unsigned long *pfilepages, |
| 4471 | unsigned long *pheadroom, unsigned long *pdirty, |
| 4472 | unsigned long *pwriteback) |
Tejun Heo | c2aa723 | 2015-05-22 18:23:35 -0400 | [diff] [blame] | 4473 | { |
| 4474 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); |
| 4475 | struct mem_cgroup *parent; |
Tejun Heo | c2aa723 | 2015-05-22 18:23:35 -0400 | [diff] [blame] | 4476 | |
Johannes Weiner | 2d146aa | 2021-04-29 22:56:26 -0700 | [diff] [blame] | 4477 | cgroup_rstat_flush_irqsafe(memcg->css.cgroup); |
Tejun Heo | c2aa723 | 2015-05-22 18:23:35 -0400 | [diff] [blame] | 4478 | |
Johannes Weiner | 2d146aa | 2021-04-29 22:56:26 -0700 | [diff] [blame] | 4479 | *pdirty = memcg_page_state(memcg, NR_FILE_DIRTY); |
| 4480 | *pwriteback = memcg_page_state(memcg, NR_WRITEBACK); |
| 4481 | *pfilepages = memcg_page_state(memcg, NR_INACTIVE_FILE) + |
| 4482 | memcg_page_state(memcg, NR_ACTIVE_FILE); |
| 4483 | |
Tejun Heo | c5edf9c | 2015-09-29 13:04:26 -0400 | [diff] [blame] | 4484 | *pheadroom = PAGE_COUNTER_MAX; |
Tejun Heo | c2aa723 | 2015-05-22 18:23:35 -0400 | [diff] [blame] | 4485 | while ((parent = parent_mem_cgroup(memcg))) { |
Chris Down | 15b4256 | 2020-04-01 21:07:20 -0700 | [diff] [blame] | 4486 | unsigned long ceiling = min(READ_ONCE(memcg->memory.max), |
Jakub Kicinski | d1663a9 | 2020-06-01 21:49:49 -0700 | [diff] [blame] | 4487 | READ_ONCE(memcg->memory.high)); |
Tejun Heo | c2aa723 | 2015-05-22 18:23:35 -0400 | [diff] [blame] | 4488 | unsigned long used = page_counter_read(&memcg->memory); |
| 4489 | |
Tejun Heo | c5edf9c | 2015-09-29 13:04:26 -0400 | [diff] [blame] | 4490 | *pheadroom = min(*pheadroom, ceiling - min(ceiling, used)); |
Tejun Heo | c2aa723 | 2015-05-22 18:23:35 -0400 | [diff] [blame] | 4491 | memcg = parent; |
| 4492 | } |
Tejun Heo | c2aa723 | 2015-05-22 18:23:35 -0400 | [diff] [blame] | 4493 | } |
| 4494 | |
Tejun Heo | 97b2782 | 2019-08-26 09:06:56 -0700 | [diff] [blame] | 4495 | /* |
| 4496 | * Foreign dirty flushing |
| 4497 | * |
| 4498 | * There's an inherent mismatch between memcg and writeback. The former |
| 4499 | * trackes ownership per-page while the latter per-inode. This was a |
| 4500 | * deliberate design decision because honoring per-page ownership in the |
| 4501 | * writeback path is complicated, may lead to higher CPU and IO overheads |
| 4502 | * and deemed unnecessary given that write-sharing an inode across |
| 4503 | * different cgroups isn't a common use-case. |
| 4504 | * |
| 4505 | * Combined with inode majority-writer ownership switching, this works well |
| 4506 | * enough in most cases but there are some pathological cases. For |
| 4507 | * example, let's say there are two cgroups A and B which keep writing to |
| 4508 | * different but confined parts of the same inode. B owns the inode and |
| 4509 | * A's memory is limited far below B's. A's dirty ratio can rise enough to |
| 4510 | * trigger balance_dirty_pages() sleeps but B's can be low enough to avoid |
| 4511 | * triggering background writeback. A will be slowed down without a way to |
| 4512 | * make writeback of the dirty pages happen. |
| 4513 | * |
| 4514 | * Conditions like the above can lead to a cgroup getting repatedly and |
| 4515 | * severely throttled after making some progress after each |
| 4516 | * dirty_expire_interval while the underyling IO device is almost |
| 4517 | * completely idle. |
| 4518 | * |
| 4519 | * Solving this problem completely requires matching the ownership tracking |
| 4520 | * granularities between memcg and writeback in either direction. However, |
| 4521 | * the more egregious behaviors can be avoided by simply remembering the |
| 4522 | * most recent foreign dirtying events and initiating remote flushes on |
| 4523 | * them when local writeback isn't enough to keep the memory clean enough. |
| 4524 | * |
| 4525 | * The following two functions implement such mechanism. When a foreign |
| 4526 | * page - a page whose memcg and writeback ownerships don't match - is |
| 4527 | * dirtied, mem_cgroup_track_foreign_dirty() records the inode owning |
| 4528 | * bdi_writeback on the page owning memcg. When balance_dirty_pages() |
| 4529 | * decides that the memcg needs to sleep due to high dirty ratio, it calls |
| 4530 | * mem_cgroup_flush_foreign() which queues writeback on the recorded |
| 4531 | * foreign bdi_writebacks which haven't expired. Both the numbers of |
| 4532 | * recorded bdi_writebacks and concurrent in-flight foreign writebacks are |
| 4533 | * limited to MEMCG_CGWB_FRN_CNT. |
| 4534 | * |
| 4535 | * The mechanism only remembers IDs and doesn't hold any object references. |
| 4536 | * As being wrong occasionally doesn't matter, updates and accesses to the |
| 4537 | * records are lockless and racy. |
| 4538 | */ |
| 4539 | void mem_cgroup_track_foreign_dirty_slowpath(struct page *page, |
| 4540 | struct bdi_writeback *wb) |
| 4541 | { |
Roman Gushchin | bcfe06b | 2020-12-01 13:58:27 -0800 | [diff] [blame] | 4542 | struct mem_cgroup *memcg = page_memcg(page); |
Tejun Heo | 97b2782 | 2019-08-26 09:06:56 -0700 | [diff] [blame] | 4543 | struct memcg_cgwb_frn *frn; |
| 4544 | u64 now = get_jiffies_64(); |
| 4545 | u64 oldest_at = now; |
| 4546 | int oldest = -1; |
| 4547 | int i; |
| 4548 | |
Tejun Heo | 3a8e9ac | 2019-08-29 15:47:19 -0700 | [diff] [blame] | 4549 | trace_track_foreign_dirty(page, wb); |
| 4550 | |
Tejun Heo | 97b2782 | 2019-08-26 09:06:56 -0700 | [diff] [blame] | 4551 | /* |
| 4552 | * Pick the slot to use. If there is already a slot for @wb, keep |
| 4553 | * using it. If not replace the oldest one which isn't being |
| 4554 | * written out. |
| 4555 | */ |
| 4556 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) { |
| 4557 | frn = &memcg->cgwb_frn[i]; |
| 4558 | if (frn->bdi_id == wb->bdi->id && |
| 4559 | frn->memcg_id == wb->memcg_css->id) |
| 4560 | break; |
| 4561 | if (time_before64(frn->at, oldest_at) && |
| 4562 | atomic_read(&frn->done.cnt) == 1) { |
| 4563 | oldest = i; |
| 4564 | oldest_at = frn->at; |
| 4565 | } |
| 4566 | } |
| 4567 | |
| 4568 | if (i < MEMCG_CGWB_FRN_CNT) { |
| 4569 | /* |
| 4570 | * Re-using an existing one. Update timestamp lazily to |
| 4571 | * avoid making the cacheline hot. We want them to be |
| 4572 | * reasonably up-to-date and significantly shorter than |
| 4573 | * dirty_expire_interval as that's what expires the record. |
| 4574 | * Use the shorter of 1s and dirty_expire_interval / 8. |
| 4575 | */ |
| 4576 | unsigned long update_intv = |
| 4577 | min_t(unsigned long, HZ, |
| 4578 | msecs_to_jiffies(dirty_expire_interval * 10) / 8); |
| 4579 | |
| 4580 | if (time_before64(frn->at, now - update_intv)) |
| 4581 | frn->at = now; |
| 4582 | } else if (oldest >= 0) { |
| 4583 | /* replace the oldest free one */ |
| 4584 | frn = &memcg->cgwb_frn[oldest]; |
| 4585 | frn->bdi_id = wb->bdi->id; |
| 4586 | frn->memcg_id = wb->memcg_css->id; |
| 4587 | frn->at = now; |
| 4588 | } |
| 4589 | } |
| 4590 | |
| 4591 | /* issue foreign writeback flushes for recorded foreign dirtying events */ |
| 4592 | void mem_cgroup_flush_foreign(struct bdi_writeback *wb) |
| 4593 | { |
| 4594 | struct mem_cgroup *memcg = mem_cgroup_from_css(wb->memcg_css); |
| 4595 | unsigned long intv = msecs_to_jiffies(dirty_expire_interval * 10); |
| 4596 | u64 now = jiffies_64; |
| 4597 | int i; |
| 4598 | |
| 4599 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) { |
| 4600 | struct memcg_cgwb_frn *frn = &memcg->cgwb_frn[i]; |
| 4601 | |
| 4602 | /* |
| 4603 | * If the record is older than dirty_expire_interval, |
| 4604 | * writeback on it has already started. No need to kick it |
| 4605 | * off again. Also, don't start a new one if there's |
| 4606 | * already one in flight. |
| 4607 | */ |
| 4608 | if (time_after64(frn->at, now - intv) && |
| 4609 | atomic_read(&frn->done.cnt) == 1) { |
| 4610 | frn->at = 0; |
Tejun Heo | 3a8e9ac | 2019-08-29 15:47:19 -0700 | [diff] [blame] | 4611 | trace_flush_foreign(wb, frn->bdi_id, frn->memcg_id); |
Tejun Heo | 97b2782 | 2019-08-26 09:06:56 -0700 | [diff] [blame] | 4612 | cgroup_writeback_by_id(frn->bdi_id, frn->memcg_id, 0, |
| 4613 | WB_REASON_FOREIGN_FLUSH, |
| 4614 | &frn->done); |
| 4615 | } |
| 4616 | } |
| 4617 | } |
| 4618 | |
Tejun Heo | 841710a | 2015-05-22 18:23:33 -0400 | [diff] [blame] | 4619 | #else /* CONFIG_CGROUP_WRITEBACK */ |
| 4620 | |
| 4621 | static int memcg_wb_domain_init(struct mem_cgroup *memcg, gfp_t gfp) |
| 4622 | { |
| 4623 | return 0; |
| 4624 | } |
| 4625 | |
| 4626 | static void memcg_wb_domain_exit(struct mem_cgroup *memcg) |
| 4627 | { |
| 4628 | } |
| 4629 | |
Tejun Heo | 2529bb3 | 2015-05-22 18:23:34 -0400 | [diff] [blame] | 4630 | static void memcg_wb_domain_size_changed(struct mem_cgroup *memcg) |
| 4631 | { |
| 4632 | } |
| 4633 | |
Tejun Heo | 52ebea7 | 2015-05-22 17:13:37 -0400 | [diff] [blame] | 4634 | #endif /* CONFIG_CGROUP_WRITEBACK */ |
| 4635 | |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4636 | /* |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 4637 | * DO NOT USE IN NEW FILES. |
| 4638 | * |
| 4639 | * "cgroup.event_control" implementation. |
| 4640 | * |
| 4641 | * This is way over-engineered. It tries to support fully configurable |
| 4642 | * events for each user. Such level of flexibility is completely |
| 4643 | * unnecessary especially in the light of the planned unified hierarchy. |
| 4644 | * |
| 4645 | * Please deprecate this and replace with something simpler if at all |
| 4646 | * possible. |
| 4647 | */ |
| 4648 | |
| 4649 | /* |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4650 | * Unregister event and free resources. |
| 4651 | * |
| 4652 | * Gets called from workqueue. |
| 4653 | */ |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 4654 | static void memcg_event_remove(struct work_struct *work) |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4655 | { |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 4656 | struct mem_cgroup_event *event = |
| 4657 | container_of(work, struct mem_cgroup_event, remove); |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4658 | struct mem_cgroup *memcg = event->memcg; |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4659 | |
| 4660 | remove_wait_queue(event->wqh, &event->wait); |
| 4661 | |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4662 | event->unregister_event(memcg, event->eventfd); |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4663 | |
| 4664 | /* Notify userspace the event is going away. */ |
| 4665 | eventfd_signal(event->eventfd, 1); |
| 4666 | |
| 4667 | eventfd_ctx_put(event->eventfd); |
| 4668 | kfree(event); |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4669 | css_put(&memcg->css); |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4670 | } |
| 4671 | |
| 4672 | /* |
Linus Torvalds | a9a0884 | 2018-02-11 14:34:03 -0800 | [diff] [blame] | 4673 | * Gets called on EPOLLHUP on eventfd when user closes it. |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4674 | * |
| 4675 | * Called with wqh->lock held and interrupts disabled. |
| 4676 | */ |
Ingo Molnar | ac6424b | 2017-06-20 12:06:13 +0200 | [diff] [blame] | 4677 | static int memcg_event_wake(wait_queue_entry_t *wait, unsigned mode, |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 4678 | int sync, void *key) |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4679 | { |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 4680 | struct mem_cgroup_event *event = |
| 4681 | container_of(wait, struct mem_cgroup_event, wait); |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4682 | struct mem_cgroup *memcg = event->memcg; |
Al Viro | 3ad6f93 | 2017-07-03 20:14:56 -0400 | [diff] [blame] | 4683 | __poll_t flags = key_to_poll(key); |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4684 | |
Linus Torvalds | a9a0884 | 2018-02-11 14:34:03 -0800 | [diff] [blame] | 4685 | if (flags & EPOLLHUP) { |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4686 | /* |
| 4687 | * If the event has been detached at cgroup removal, we |
| 4688 | * can simply return knowing the other side will cleanup |
| 4689 | * for us. |
| 4690 | * |
| 4691 | * We can't race against event freeing since the other |
| 4692 | * side will require wqh->lock via remove_wait_queue(), |
| 4693 | * which we hold. |
| 4694 | */ |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4695 | spin_lock(&memcg->event_list_lock); |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4696 | if (!list_empty(&event->list)) { |
| 4697 | list_del_init(&event->list); |
| 4698 | /* |
| 4699 | * We are in atomic context, but cgroup_event_remove() |
| 4700 | * may sleep, so we have to call it in workqueue. |
| 4701 | */ |
| 4702 | schedule_work(&event->remove); |
| 4703 | } |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4704 | spin_unlock(&memcg->event_list_lock); |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4705 | } |
| 4706 | |
| 4707 | return 0; |
| 4708 | } |
| 4709 | |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 4710 | static void memcg_event_ptable_queue_proc(struct file *file, |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4711 | wait_queue_head_t *wqh, poll_table *pt) |
| 4712 | { |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 4713 | struct mem_cgroup_event *event = |
| 4714 | container_of(pt, struct mem_cgroup_event, pt); |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4715 | |
| 4716 | event->wqh = wqh; |
| 4717 | add_wait_queue(wqh, &event->wait); |
| 4718 | } |
| 4719 | |
| 4720 | /* |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 4721 | * DO NOT USE IN NEW FILES. |
| 4722 | * |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4723 | * Parse input and register new cgroup event handler. |
| 4724 | * |
| 4725 | * Input must be in format '<event_fd> <control_fd> <args>'. |
| 4726 | * Interpretation of args is defined by control file implementation. |
| 4727 | */ |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 4728 | static ssize_t memcg_write_event_control(struct kernfs_open_file *of, |
| 4729 | char *buf, size_t nbytes, loff_t off) |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4730 | { |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 4731 | struct cgroup_subsys_state *css = of_css(of); |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4732 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 4733 | struct mem_cgroup_event *event; |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4734 | struct cgroup_subsys_state *cfile_css; |
| 4735 | unsigned int efd, cfd; |
| 4736 | struct fd efile; |
| 4737 | struct fd cfile; |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4738 | const char *name; |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4739 | char *endp; |
| 4740 | int ret; |
| 4741 | |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 4742 | buf = strstrip(buf); |
| 4743 | |
| 4744 | efd = simple_strtoul(buf, &endp, 10); |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4745 | if (*endp != ' ') |
| 4746 | return -EINVAL; |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 4747 | buf = endp + 1; |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4748 | |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 4749 | cfd = simple_strtoul(buf, &endp, 10); |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4750 | if ((*endp != ' ') && (*endp != '\0')) |
| 4751 | return -EINVAL; |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 4752 | buf = endp + 1; |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4753 | |
| 4754 | event = kzalloc(sizeof(*event), GFP_KERNEL); |
| 4755 | if (!event) |
| 4756 | return -ENOMEM; |
| 4757 | |
Tejun Heo | 59b6f87 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4758 | event->memcg = memcg; |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4759 | INIT_LIST_HEAD(&event->list); |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 4760 | init_poll_funcptr(&event->pt, memcg_event_ptable_queue_proc); |
| 4761 | init_waitqueue_func_entry(&event->wait, memcg_event_wake); |
| 4762 | INIT_WORK(&event->remove, memcg_event_remove); |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4763 | |
| 4764 | efile = fdget(efd); |
| 4765 | if (!efile.file) { |
| 4766 | ret = -EBADF; |
| 4767 | goto out_kfree; |
| 4768 | } |
| 4769 | |
| 4770 | event->eventfd = eventfd_ctx_fileget(efile.file); |
| 4771 | if (IS_ERR(event->eventfd)) { |
| 4772 | ret = PTR_ERR(event->eventfd); |
| 4773 | goto out_put_efile; |
| 4774 | } |
| 4775 | |
| 4776 | cfile = fdget(cfd); |
| 4777 | if (!cfile.file) { |
| 4778 | ret = -EBADF; |
| 4779 | goto out_put_eventfd; |
| 4780 | } |
| 4781 | |
| 4782 | /* the process need read permission on control file */ |
| 4783 | /* AV: shouldn't we check that it's been opened for read instead? */ |
Christian Brauner | 02f92b3 | 2021-01-21 14:19:22 +0100 | [diff] [blame] | 4784 | ret = file_permission(cfile.file, MAY_READ); |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4785 | if (ret < 0) |
| 4786 | goto out_put_cfile; |
| 4787 | |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4788 | /* |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4789 | * Determine the event callbacks and set them in @event. This used |
| 4790 | * to be done via struct cftype but cgroup core no longer knows |
| 4791 | * about these events. The following is crude but the whole thing |
| 4792 | * is for compatibility anyway. |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 4793 | * |
| 4794 | * DO NOT ADD NEW FILES. |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4795 | */ |
Al Viro | b583043 | 2014-10-31 01:22:04 -0400 | [diff] [blame] | 4796 | name = cfile.file->f_path.dentry->d_name.name; |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4797 | |
| 4798 | if (!strcmp(name, "memory.usage_in_bytes")) { |
| 4799 | event->register_event = mem_cgroup_usage_register_event; |
| 4800 | event->unregister_event = mem_cgroup_usage_unregister_event; |
| 4801 | } else if (!strcmp(name, "memory.oom_control")) { |
| 4802 | event->register_event = mem_cgroup_oom_register_event; |
| 4803 | event->unregister_event = mem_cgroup_oom_unregister_event; |
| 4804 | } else if (!strcmp(name, "memory.pressure_level")) { |
| 4805 | event->register_event = vmpressure_register_event; |
| 4806 | event->unregister_event = vmpressure_unregister_event; |
| 4807 | } else if (!strcmp(name, "memory.memsw.usage_in_bytes")) { |
Tejun Heo | 347c4a8 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4808 | event->register_event = memsw_cgroup_usage_register_event; |
| 4809 | event->unregister_event = memsw_cgroup_usage_unregister_event; |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4810 | } else { |
| 4811 | ret = -EINVAL; |
| 4812 | goto out_put_cfile; |
| 4813 | } |
| 4814 | |
| 4815 | /* |
Tejun Heo | b5557c4 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4816 | * Verify @cfile should belong to @css. Also, remaining events are |
| 4817 | * automatically removed on cgroup destruction but the removal is |
| 4818 | * asynchronous, so take an extra ref on @css. |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4819 | */ |
Al Viro | b583043 | 2014-10-31 01:22:04 -0400 | [diff] [blame] | 4820 | 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] | 4821 | &memory_cgrp_subsys); |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4822 | ret = -EINVAL; |
Tejun Heo | 5a17f54 | 2014-02-11 11:52:47 -0500 | [diff] [blame] | 4823 | if (IS_ERR(cfile_css)) |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4824 | goto out_put_cfile; |
Tejun Heo | 5a17f54 | 2014-02-11 11:52:47 -0500 | [diff] [blame] | 4825 | if (cfile_css != css) { |
| 4826 | css_put(cfile_css); |
| 4827 | goto out_put_cfile; |
| 4828 | } |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4829 | |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 4830 | ret = event->register_event(memcg, event->eventfd, buf); |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4831 | if (ret) |
| 4832 | goto out_put_css; |
| 4833 | |
Christoph Hellwig | 9965ed17 | 2018-03-05 07:26:05 -0800 | [diff] [blame] | 4834 | vfs_poll(efile.file, &event->pt); |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4835 | |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 4836 | spin_lock(&memcg->event_list_lock); |
| 4837 | list_add(&event->list, &memcg->event_list); |
| 4838 | spin_unlock(&memcg->event_list_lock); |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4839 | |
| 4840 | fdput(cfile); |
| 4841 | fdput(efile); |
| 4842 | |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 4843 | return nbytes; |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4844 | |
| 4845 | out_put_css: |
Tejun Heo | b5557c4 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4846 | css_put(css); |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4847 | out_put_cfile: |
| 4848 | fdput(cfile); |
| 4849 | out_put_eventfd: |
| 4850 | eventfd_ctx_put(event->eventfd); |
| 4851 | out_put_efile: |
| 4852 | fdput(efile); |
| 4853 | out_kfree: |
| 4854 | kfree(event); |
| 4855 | |
| 4856 | return ret; |
| 4857 | } |
| 4858 | |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 4859 | static struct cftype mem_cgroup_legacy_files[] = { |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 4860 | { |
Balbir Singh | 0eea103 | 2008-02-07 00:13:57 -0800 | [diff] [blame] | 4861 | .name = "usage_in_bytes", |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4862 | .private = MEMFILE_PRIVATE(_MEM, RES_USAGE), |
Tejun Heo | 791badb | 2013-12-05 12:28:02 -0500 | [diff] [blame] | 4863 | .read_u64 = mem_cgroup_read_u64, |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 4864 | }, |
| 4865 | { |
Pavel Emelyanov | c84872e | 2008-04-29 01:00:17 -0700 | [diff] [blame] | 4866 | .name = "max_usage_in_bytes", |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4867 | .private = MEMFILE_PRIVATE(_MEM, RES_MAX_USAGE), |
Tejun Heo | 6770c64 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 4868 | .write = mem_cgroup_reset, |
Tejun Heo | 791badb | 2013-12-05 12:28:02 -0500 | [diff] [blame] | 4869 | .read_u64 = mem_cgroup_read_u64, |
Pavel Emelyanov | c84872e | 2008-04-29 01:00:17 -0700 | [diff] [blame] | 4870 | }, |
| 4871 | { |
Balbir Singh | 0eea103 | 2008-02-07 00:13:57 -0800 | [diff] [blame] | 4872 | .name = "limit_in_bytes", |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4873 | .private = MEMFILE_PRIVATE(_MEM, RES_LIMIT), |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 4874 | .write = mem_cgroup_write, |
Tejun Heo | 791badb | 2013-12-05 12:28:02 -0500 | [diff] [blame] | 4875 | .read_u64 = mem_cgroup_read_u64, |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 4876 | }, |
| 4877 | { |
Balbir Singh | 296c81d | 2009-09-23 15:56:36 -0700 | [diff] [blame] | 4878 | .name = "soft_limit_in_bytes", |
| 4879 | .private = MEMFILE_PRIVATE(_MEM, RES_SOFT_LIMIT), |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 4880 | .write = mem_cgroup_write, |
Tejun Heo | 791badb | 2013-12-05 12:28:02 -0500 | [diff] [blame] | 4881 | .read_u64 = mem_cgroup_read_u64, |
Balbir Singh | 296c81d | 2009-09-23 15:56:36 -0700 | [diff] [blame] | 4882 | }, |
| 4883 | { |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 4884 | .name = "failcnt", |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4885 | .private = MEMFILE_PRIVATE(_MEM, RES_FAILCNT), |
Tejun Heo | 6770c64 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 4886 | .write = mem_cgroup_reset, |
Tejun Heo | 791badb | 2013-12-05 12:28:02 -0500 | [diff] [blame] | 4887 | .read_u64 = mem_cgroup_read_u64, |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 4888 | }, |
Balbir Singh | 8697d33 | 2008-02-07 00:13:59 -0800 | [diff] [blame] | 4889 | { |
KAMEZAWA Hiroyuki | d2ceb9b | 2008-02-07 00:14:25 -0800 | [diff] [blame] | 4890 | .name = "stat", |
Tejun Heo | 2da8ca8 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 4891 | .seq_show = memcg_stat_show, |
KAMEZAWA Hiroyuki | d2ceb9b | 2008-02-07 00:14:25 -0800 | [diff] [blame] | 4892 | }, |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 4893 | { |
| 4894 | .name = "force_empty", |
Tejun Heo | 6770c64 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 4895 | .write = mem_cgroup_force_empty_write, |
KAMEZAWA Hiroyuki | c1e862c | 2009-01-07 18:07:55 -0800 | [diff] [blame] | 4896 | }, |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 4897 | { |
| 4898 | .name = "use_hierarchy", |
| 4899 | .write_u64 = mem_cgroup_hierarchy_write, |
| 4900 | .read_u64 = mem_cgroup_hierarchy_read, |
| 4901 | }, |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 4902 | { |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 4903 | .name = "cgroup.event_control", /* XXX: for compat */ |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 4904 | .write = memcg_write_event_control, |
Tejun Heo | 7dbdb19 | 2015-09-18 17:54:23 -0400 | [diff] [blame] | 4905 | .flags = CFTYPE_NO_PREFIX | CFTYPE_WORLD_WRITABLE, |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 4906 | }, |
| 4907 | { |
KOSAKI Motohiro | a7885eb | 2009-01-07 18:08:24 -0800 | [diff] [blame] | 4908 | .name = "swappiness", |
| 4909 | .read_u64 = mem_cgroup_swappiness_read, |
| 4910 | .write_u64 = mem_cgroup_swappiness_write, |
| 4911 | }, |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 4912 | { |
| 4913 | .name = "move_charge_at_immigrate", |
| 4914 | .read_u64 = mem_cgroup_move_charge_read, |
| 4915 | .write_u64 = mem_cgroup_move_charge_write, |
| 4916 | }, |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 4917 | { |
| 4918 | .name = "oom_control", |
Tejun Heo | 2da8ca8 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 4919 | .seq_show = mem_cgroup_oom_control_read, |
KAMEZAWA Hiroyuki | 3c11ecf | 2010-05-26 14:42:37 -0700 | [diff] [blame] | 4920 | .write_u64 = mem_cgroup_oom_control_write, |
KAMEZAWA Hiroyuki | 9490ff2 | 2010-05-26 14:42:36 -0700 | [diff] [blame] | 4921 | .private = MEMFILE_PRIVATE(_OOM_TYPE, OOM_CONTROL), |
| 4922 | }, |
Anton Vorontsov | 70ddf63 | 2013-04-29 15:08:31 -0700 | [diff] [blame] | 4923 | { |
| 4924 | .name = "pressure_level", |
Anton Vorontsov | 70ddf63 | 2013-04-29 15:08:31 -0700 | [diff] [blame] | 4925 | }, |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 4926 | #ifdef CONFIG_NUMA |
| 4927 | { |
| 4928 | .name = "numa_stat", |
Tejun Heo | 2da8ca8 | 2013-12-05 12:28:04 -0500 | [diff] [blame] | 4929 | .seq_show = memcg_numa_stat_show, |
Ying Han | 406eb0c | 2011-05-26 16:25:37 -0700 | [diff] [blame] | 4930 | }, |
| 4931 | #endif |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 4932 | { |
| 4933 | .name = "kmem.limit_in_bytes", |
| 4934 | .private = MEMFILE_PRIVATE(_KMEM, RES_LIMIT), |
Tejun Heo | 451af50 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 4935 | .write = mem_cgroup_write, |
Tejun Heo | 791badb | 2013-12-05 12:28:02 -0500 | [diff] [blame] | 4936 | .read_u64 = mem_cgroup_read_u64, |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 4937 | }, |
| 4938 | { |
| 4939 | .name = "kmem.usage_in_bytes", |
| 4940 | .private = MEMFILE_PRIVATE(_KMEM, RES_USAGE), |
Tejun Heo | 791badb | 2013-12-05 12:28:02 -0500 | [diff] [blame] | 4941 | .read_u64 = mem_cgroup_read_u64, |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 4942 | }, |
| 4943 | { |
| 4944 | .name = "kmem.failcnt", |
| 4945 | .private = MEMFILE_PRIVATE(_KMEM, RES_FAILCNT), |
Tejun Heo | 6770c64 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 4946 | .write = mem_cgroup_reset, |
Tejun Heo | 791badb | 2013-12-05 12:28:02 -0500 | [diff] [blame] | 4947 | .read_u64 = mem_cgroup_read_u64, |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 4948 | }, |
| 4949 | { |
| 4950 | .name = "kmem.max_usage_in_bytes", |
| 4951 | .private = MEMFILE_PRIVATE(_KMEM, RES_MAX_USAGE), |
Tejun Heo | 6770c64 | 2014-05-13 12:16:21 -0400 | [diff] [blame] | 4952 | .write = mem_cgroup_reset, |
Tejun Heo | 791badb | 2013-12-05 12:28:02 -0500 | [diff] [blame] | 4953 | .read_u64 = mem_cgroup_read_u64, |
Glauber Costa | 510fc4e | 2012-12-18 14:21:47 -0800 | [diff] [blame] | 4954 | }, |
Yafang Shao | a87425a | 2020-04-01 21:06:30 -0700 | [diff] [blame] | 4955 | #if defined(CONFIG_MEMCG_KMEM) && \ |
| 4956 | (defined(CONFIG_SLAB) || defined(CONFIG_SLUB_DEBUG)) |
Glauber Costa | 749c541 | 2012-12-18 14:23:01 -0800 | [diff] [blame] | 4957 | { |
| 4958 | .name = "kmem.slabinfo", |
Vladimir Davydov | b047501 | 2014-12-10 15:44:19 -0800 | [diff] [blame] | 4959 | .seq_show = memcg_slab_show, |
Glauber Costa | 749c541 | 2012-12-18 14:23:01 -0800 | [diff] [blame] | 4960 | }, |
| 4961 | #endif |
Vladimir Davydov | d55f90b | 2016-01-20 15:02:44 -0800 | [diff] [blame] | 4962 | { |
| 4963 | .name = "kmem.tcp.limit_in_bytes", |
| 4964 | .private = MEMFILE_PRIVATE(_TCP, RES_LIMIT), |
| 4965 | .write = mem_cgroup_write, |
| 4966 | .read_u64 = mem_cgroup_read_u64, |
| 4967 | }, |
| 4968 | { |
| 4969 | .name = "kmem.tcp.usage_in_bytes", |
| 4970 | .private = MEMFILE_PRIVATE(_TCP, RES_USAGE), |
| 4971 | .read_u64 = mem_cgroup_read_u64, |
| 4972 | }, |
| 4973 | { |
| 4974 | .name = "kmem.tcp.failcnt", |
| 4975 | .private = MEMFILE_PRIVATE(_TCP, RES_FAILCNT), |
| 4976 | .write = mem_cgroup_reset, |
| 4977 | .read_u64 = mem_cgroup_read_u64, |
| 4978 | }, |
| 4979 | { |
| 4980 | .name = "kmem.tcp.max_usage_in_bytes", |
| 4981 | .private = MEMFILE_PRIVATE(_TCP, RES_MAX_USAGE), |
| 4982 | .write = mem_cgroup_reset, |
| 4983 | .read_u64 = mem_cgroup_read_u64, |
| 4984 | }, |
Tejun Heo | 6bc1034 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 4985 | { }, /* terminate */ |
Tejun Heo | af36f90 | 2012-04-01 12:09:55 -0700 | [diff] [blame] | 4986 | }; |
KAMEZAWA Hiroyuki | 8c7c6e34 | 2009-01-07 18:08:00 -0800 | [diff] [blame] | 4987 | |
Johannes Weiner | 73f576c | 2016-07-20 15:44:57 -0700 | [diff] [blame] | 4988 | /* |
| 4989 | * Private memory cgroup IDR |
| 4990 | * |
| 4991 | * Swap-out records and page cache shadow entries need to store memcg |
| 4992 | * references in constrained space, so we maintain an ID space that is |
| 4993 | * limited to 16 bit (MEM_CGROUP_ID_MAX), limiting the total number of |
| 4994 | * memory-controlled cgroups to 64k. |
| 4995 | * |
Ethon Paul | b8f2935 | 2020-06-04 16:49:28 -0700 | [diff] [blame] | 4996 | * However, there usually are many references to the offline CSS after |
Johannes Weiner | 73f576c | 2016-07-20 15:44:57 -0700 | [diff] [blame] | 4997 | * the cgroup has been destroyed, such as page cache or reclaimable |
| 4998 | * slab objects, that don't need to hang on to the ID. We want to keep |
| 4999 | * those dead CSS from occupying IDs, or we might quickly exhaust the |
| 5000 | * relatively small ID space and prevent the creation of new cgroups |
| 5001 | * even when there are much fewer than 64k cgroups - possibly none. |
| 5002 | * |
| 5003 | * Maintain a private 16-bit ID space for memcg, and allow the ID to |
| 5004 | * be freed and recycled when it's no longer needed, which is usually |
| 5005 | * when the CSS is offlined. |
| 5006 | * |
| 5007 | * The only exception to that are records of swapped out tmpfs/shmem |
| 5008 | * pages that need to be attributed to live ancestors on swapin. But |
| 5009 | * those references are manageable from userspace. |
| 5010 | */ |
| 5011 | |
| 5012 | static DEFINE_IDR(mem_cgroup_idr); |
| 5013 | |
Kirill Tkhai | 7e97de0 | 2018-08-02 15:36:01 -0700 | [diff] [blame] | 5014 | static void mem_cgroup_id_remove(struct mem_cgroup *memcg) |
| 5015 | { |
| 5016 | if (memcg->id.id > 0) { |
| 5017 | idr_remove(&mem_cgroup_idr, memcg->id.id); |
| 5018 | memcg->id.id = 0; |
| 5019 | } |
| 5020 | } |
| 5021 | |
Vincenzo Frascino | c1514c0 | 2020-04-01 21:07:13 -0700 | [diff] [blame] | 5022 | static void __maybe_unused mem_cgroup_id_get_many(struct mem_cgroup *memcg, |
| 5023 | unsigned int n) |
Johannes Weiner | 73f576c | 2016-07-20 15:44:57 -0700 | [diff] [blame] | 5024 | { |
Kirill Tkhai | 1c2d479 | 2018-10-26 15:09:28 -0700 | [diff] [blame] | 5025 | refcount_add(n, &memcg->id.ref); |
Johannes Weiner | 73f576c | 2016-07-20 15:44:57 -0700 | [diff] [blame] | 5026 | } |
| 5027 | |
Vladimir Davydov | 615d66c | 2016-08-11 15:33:03 -0700 | [diff] [blame] | 5028 | 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] | 5029 | { |
Kirill Tkhai | 1c2d479 | 2018-10-26 15:09:28 -0700 | [diff] [blame] | 5030 | if (refcount_sub_and_test(n, &memcg->id.ref)) { |
Kirill Tkhai | 7e97de0 | 2018-08-02 15:36:01 -0700 | [diff] [blame] | 5031 | mem_cgroup_id_remove(memcg); |
Johannes Weiner | 73f576c | 2016-07-20 15:44:57 -0700 | [diff] [blame] | 5032 | |
| 5033 | /* Memcg ID pins CSS */ |
| 5034 | css_put(&memcg->css); |
| 5035 | } |
| 5036 | } |
| 5037 | |
Vladimir Davydov | 615d66c | 2016-08-11 15:33:03 -0700 | [diff] [blame] | 5038 | static inline void mem_cgroup_id_put(struct mem_cgroup *memcg) |
| 5039 | { |
| 5040 | mem_cgroup_id_put_many(memcg, 1); |
| 5041 | } |
| 5042 | |
Johannes Weiner | 73f576c | 2016-07-20 15:44:57 -0700 | [diff] [blame] | 5043 | /** |
| 5044 | * mem_cgroup_from_id - look up a memcg from a memcg id |
| 5045 | * @id: the memcg id to look up |
| 5046 | * |
| 5047 | * Caller must hold rcu_read_lock(). |
| 5048 | */ |
| 5049 | struct mem_cgroup *mem_cgroup_from_id(unsigned short id) |
| 5050 | { |
| 5051 | WARN_ON_ONCE(!rcu_read_lock_held()); |
| 5052 | return idr_find(&mem_cgroup_idr, id); |
| 5053 | } |
| 5054 | |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 5055 | 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] | 5056 | { |
| 5057 | struct mem_cgroup_per_node *pn; |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 5058 | int tmp = node; |
KAMEZAWA Hiroyuki | 1ecaab2 | 2008-02-07 00:14:38 -0800 | [diff] [blame] | 5059 | /* |
| 5060 | * This routine is called against possible nodes. |
| 5061 | * But it's BUG to call kmalloc() against offline node. |
| 5062 | * |
| 5063 | * TODO: this routine can waste much memory for nodes which will |
| 5064 | * never be onlined. It's better to use memory hotplug callback |
| 5065 | * function. |
| 5066 | */ |
KAMEZAWA Hiroyuki | 41e3355 | 2008-04-08 17:41:54 -0700 | [diff] [blame] | 5067 | if (!node_state(node, N_NORMAL_MEMORY)) |
| 5068 | tmp = -1; |
Jesper Juhl | 17295c8 | 2011-01-13 15:47:42 -0800 | [diff] [blame] | 5069 | pn = kzalloc_node(sizeof(*pn), GFP_KERNEL, tmp); |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 5070 | if (!pn) |
| 5071 | return 1; |
KAMEZAWA Hiroyuki | 1ecaab2 | 2008-02-07 00:14:38 -0800 | [diff] [blame] | 5072 | |
Roman Gushchin | 3e38e0a | 2020-08-11 18:30:25 -0700 | [diff] [blame] | 5073 | pn->lruvec_stat_local = alloc_percpu_gfp(struct lruvec_stat, |
| 5074 | GFP_KERNEL_ACCOUNT); |
Johannes Weiner | 815744d | 2019-06-13 15:55:46 -0700 | [diff] [blame] | 5075 | if (!pn->lruvec_stat_local) { |
| 5076 | kfree(pn); |
| 5077 | return 1; |
| 5078 | } |
| 5079 | |
Muchun Song | f3344ad | 2021-02-24 12:03:15 -0800 | [diff] [blame] | 5080 | pn->lruvec_stat_cpu = alloc_percpu_gfp(struct batched_lruvec_stat, |
Roman Gushchin | 3e38e0a | 2020-08-11 18:30:25 -0700 | [diff] [blame] | 5081 | GFP_KERNEL_ACCOUNT); |
Johannes Weiner | a983b5e | 2018-01-31 16:16:45 -0800 | [diff] [blame] | 5082 | if (!pn->lruvec_stat_cpu) { |
Johannes Weiner | 815744d | 2019-06-13 15:55:46 -0700 | [diff] [blame] | 5083 | free_percpu(pn->lruvec_stat_local); |
Johannes Weiner | 00f3ca2 | 2017-07-06 15:40:52 -0700 | [diff] [blame] | 5084 | kfree(pn); |
| 5085 | return 1; |
| 5086 | } |
| 5087 | |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 5088 | lruvec_init(&pn->lruvec); |
| 5089 | pn->usage_in_excess = 0; |
| 5090 | pn->on_tree = false; |
| 5091 | pn->memcg = memcg; |
| 5092 | |
Johannes Weiner | 54f72fe | 2013-07-08 15:59:49 -0700 | [diff] [blame] | 5093 | memcg->nodeinfo[node] = pn; |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 5094 | return 0; |
| 5095 | } |
| 5096 | |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 5097 | 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] | 5098 | { |
Johannes Weiner | 00f3ca2 | 2017-07-06 15:40:52 -0700 | [diff] [blame] | 5099 | struct mem_cgroup_per_node *pn = memcg->nodeinfo[node]; |
| 5100 | |
Michal Hocko | 4eaf431 | 2018-04-10 16:29:52 -0700 | [diff] [blame] | 5101 | if (!pn) |
| 5102 | return; |
| 5103 | |
Johannes Weiner | a983b5e | 2018-01-31 16:16:45 -0800 | [diff] [blame] | 5104 | free_percpu(pn->lruvec_stat_cpu); |
Johannes Weiner | 815744d | 2019-06-13 15:55:46 -0700 | [diff] [blame] | 5105 | free_percpu(pn->lruvec_stat_local); |
Johannes Weiner | 00f3ca2 | 2017-07-06 15:40:52 -0700 | [diff] [blame] | 5106 | kfree(pn); |
KAMEZAWA Hiroyuki | 1ecaab2 | 2008-02-07 00:14:38 -0800 | [diff] [blame] | 5107 | } |
| 5108 | |
Tahsin Erdogan | 40e952f | 2017-03-09 16:17:26 -0800 | [diff] [blame] | 5109 | static void __mem_cgroup_free(struct mem_cgroup *memcg) |
Johannes Weiner | 0b8f73e | 2016-01-20 15:02:53 -0800 | [diff] [blame] | 5110 | { |
| 5111 | int node; |
| 5112 | |
Johannes Weiner | 0b8f73e | 2016-01-20 15:02:53 -0800 | [diff] [blame] | 5113 | for_each_node(node) |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 5114 | free_mem_cgroup_per_node_info(memcg, node); |
Chris Down | 871789d | 2019-05-14 15:46:57 -0700 | [diff] [blame] | 5115 | free_percpu(memcg->vmstats_percpu); |
Johannes Weiner | 0b8f73e | 2016-01-20 15:02:53 -0800 | [diff] [blame] | 5116 | kfree(memcg); |
| 5117 | } |
| 5118 | |
Tahsin Erdogan | 40e952f | 2017-03-09 16:17:26 -0800 | [diff] [blame] | 5119 | static void mem_cgroup_free(struct mem_cgroup *memcg) |
| 5120 | { |
Johannes Weiner | 2cd21c8 | 2021-04-29 22:56:29 -0700 | [diff] [blame] | 5121 | int cpu; |
| 5122 | |
Tahsin Erdogan | 40e952f | 2017-03-09 16:17:26 -0800 | [diff] [blame] | 5123 | memcg_wb_domain_exit(memcg); |
Shakeel Butt | 7961eee | 2019-11-05 21:16:21 -0800 | [diff] [blame] | 5124 | /* |
Johannes Weiner | 2d146aa | 2021-04-29 22:56:26 -0700 | [diff] [blame] | 5125 | * Flush percpu lruvec stats to guarantee the value |
| 5126 | * correctness on parent's and all ancestor levels. |
Shakeel Butt | 7961eee | 2019-11-05 21:16:21 -0800 | [diff] [blame] | 5127 | */ |
Johannes Weiner | 2cd21c8 | 2021-04-29 22:56:29 -0700 | [diff] [blame] | 5128 | for_each_online_cpu(cpu) |
| 5129 | memcg_flush_lruvec_page_state(memcg, cpu); |
Tahsin Erdogan | 40e952f | 2017-03-09 16:17:26 -0800 | [diff] [blame] | 5130 | __mem_cgroup_free(memcg); |
| 5131 | } |
| 5132 | |
KAMEZAWA Hiroyuki | 3332794 | 2008-04-29 01:00:24 -0700 | [diff] [blame] | 5133 | static struct mem_cgroup *mem_cgroup_alloc(void) |
| 5134 | { |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 5135 | struct mem_cgroup *memcg; |
Alexey Dobriyan | b9726c2 | 2019-03-05 15:48:26 -0800 | [diff] [blame] | 5136 | unsigned int size; |
Johannes Weiner | 0b8f73e | 2016-01-20 15:02:53 -0800 | [diff] [blame] | 5137 | int node; |
Tejun Heo | 97b2782 | 2019-08-26 09:06:56 -0700 | [diff] [blame] | 5138 | int __maybe_unused i; |
Yafang Shao | 11d6761 | 2020-05-07 18:35:43 -0700 | [diff] [blame] | 5139 | long error = -ENOMEM; |
KAMEZAWA Hiroyuki | 3332794 | 2008-04-29 01:00:24 -0700 | [diff] [blame] | 5140 | |
Vladimir Davydov | 8ff69e2 | 2014-01-23 15:52:52 -0800 | [diff] [blame] | 5141 | size = sizeof(struct mem_cgroup); |
| 5142 | size += nr_node_ids * sizeof(struct mem_cgroup_per_node *); |
KAMEZAWA Hiroyuki | 3332794 | 2008-04-29 01:00:24 -0700 | [diff] [blame] | 5143 | |
Vladimir Davydov | 8ff69e2 | 2014-01-23 15:52:52 -0800 | [diff] [blame] | 5144 | memcg = kzalloc(size, GFP_KERNEL); |
Hugh Dickins | d79154b | 2012-03-21 16:34:18 -0700 | [diff] [blame] | 5145 | if (!memcg) |
Yafang Shao | 11d6761 | 2020-05-07 18:35:43 -0700 | [diff] [blame] | 5146 | return ERR_PTR(error); |
Dan Carpenter | e7bbcdf | 2010-03-23 13:35:12 -0700 | [diff] [blame] | 5147 | |
Johannes Weiner | 73f576c | 2016-07-20 15:44:57 -0700 | [diff] [blame] | 5148 | memcg->id.id = idr_alloc(&mem_cgroup_idr, NULL, |
| 5149 | 1, MEM_CGROUP_ID_MAX, |
| 5150 | GFP_KERNEL); |
Yafang Shao | 11d6761 | 2020-05-07 18:35:43 -0700 | [diff] [blame] | 5151 | if (memcg->id.id < 0) { |
| 5152 | error = memcg->id.id; |
Johannes Weiner | 73f576c | 2016-07-20 15:44:57 -0700 | [diff] [blame] | 5153 | goto fail; |
Yafang Shao | 11d6761 | 2020-05-07 18:35:43 -0700 | [diff] [blame] | 5154 | } |
Johannes Weiner | 73f576c | 2016-07-20 15:44:57 -0700 | [diff] [blame] | 5155 | |
Roman Gushchin | 3e38e0a | 2020-08-11 18:30:25 -0700 | [diff] [blame] | 5156 | memcg->vmstats_percpu = alloc_percpu_gfp(struct memcg_vmstats_percpu, |
| 5157 | GFP_KERNEL_ACCOUNT); |
Chris Down | 871789d | 2019-05-14 15:46:57 -0700 | [diff] [blame] | 5158 | if (!memcg->vmstats_percpu) |
Johannes Weiner | 0b8f73e | 2016-01-20 15:02:53 -0800 | [diff] [blame] | 5159 | goto fail; |
Pavel Emelianov | 78fb746 | 2008-02-07 00:13:51 -0800 | [diff] [blame] | 5160 | |
Bob Liu | 3ed28fa | 2012-01-12 17:19:04 -0800 | [diff] [blame] | 5161 | for_each_node(node) |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 5162 | if (alloc_mem_cgroup_per_node_info(memcg, node)) |
Johannes Weiner | 0b8f73e | 2016-01-20 15:02:53 -0800 | [diff] [blame] | 5163 | goto fail; |
Balbir Singh | f64c3f5 | 2009-09-23 15:56:37 -0700 | [diff] [blame] | 5164 | |
Johannes Weiner | 0b8f73e | 2016-01-20 15:02:53 -0800 | [diff] [blame] | 5165 | if (memcg_wb_domain_init(memcg, GFP_KERNEL)) |
| 5166 | goto fail; |
Balbir Singh | 28dbc4b | 2009-01-07 18:08:05 -0800 | [diff] [blame] | 5167 | |
Johannes Weiner | f7e1cb6 | 2016-01-14 15:21:29 -0800 | [diff] [blame] | 5168 | INIT_WORK(&memcg->high_work, high_work_func); |
Glauber Costa | d142e3e | 2013-02-22 16:34:52 -0800 | [diff] [blame] | 5169 | INIT_LIST_HEAD(&memcg->oom_notify); |
Glauber Costa | d142e3e | 2013-02-22 16:34:52 -0800 | [diff] [blame] | 5170 | mutex_init(&memcg->thresholds_lock); |
| 5171 | spin_lock_init(&memcg->move_lock); |
Anton Vorontsov | 70ddf63 | 2013-04-29 15:08:31 -0700 | [diff] [blame] | 5172 | vmpressure_init(&memcg->vmpressure); |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 5173 | INIT_LIST_HEAD(&memcg->event_list); |
| 5174 | spin_lock_init(&memcg->event_list_lock); |
Johannes Weiner | d886f4e | 2016-01-20 15:02:47 -0800 | [diff] [blame] | 5175 | memcg->socket_pressure = jiffies; |
Kirill Tkhai | 84c07d1 | 2018-08-17 15:47:25 -0700 | [diff] [blame] | 5176 | #ifdef CONFIG_MEMCG_KMEM |
Vladimir Davydov | 900a38f | 2014-12-12 16:55:10 -0800 | [diff] [blame] | 5177 | memcg->kmemcg_id = -1; |
Roman Gushchin | bf4f059 | 2020-08-06 23:20:49 -0700 | [diff] [blame] | 5178 | INIT_LIST_HEAD(&memcg->objcg_list); |
Vladimir Davydov | 900a38f | 2014-12-12 16:55:10 -0800 | [diff] [blame] | 5179 | #endif |
Tejun Heo | 52ebea7 | 2015-05-22 17:13:37 -0400 | [diff] [blame] | 5180 | #ifdef CONFIG_CGROUP_WRITEBACK |
| 5181 | INIT_LIST_HEAD(&memcg->cgwb_list); |
Tejun Heo | 97b2782 | 2019-08-26 09:06:56 -0700 | [diff] [blame] | 5182 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) |
| 5183 | memcg->cgwb_frn[i].done = |
| 5184 | __WB_COMPLETION_INIT(&memcg_cgwb_frn_waitq); |
Tejun Heo | 52ebea7 | 2015-05-22 17:13:37 -0400 | [diff] [blame] | 5185 | #endif |
Yang Shi | 87eaceb | 2019-09-23 15:38:15 -0700 | [diff] [blame] | 5186 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
| 5187 | spin_lock_init(&memcg->deferred_split_queue.split_queue_lock); |
| 5188 | INIT_LIST_HEAD(&memcg->deferred_split_queue.split_queue); |
| 5189 | memcg->deferred_split_queue.split_queue_len = 0; |
| 5190 | #endif |
Johannes Weiner | 73f576c | 2016-07-20 15:44:57 -0700 | [diff] [blame] | 5191 | idr_replace(&mem_cgroup_idr, memcg, memcg->id.id); |
Johannes Weiner | 0b8f73e | 2016-01-20 15:02:53 -0800 | [diff] [blame] | 5192 | return memcg; |
| 5193 | fail: |
Kirill Tkhai | 7e97de0 | 2018-08-02 15:36:01 -0700 | [diff] [blame] | 5194 | mem_cgroup_id_remove(memcg); |
Tahsin Erdogan | 40e952f | 2017-03-09 16:17:26 -0800 | [diff] [blame] | 5195 | __mem_cgroup_free(memcg); |
Yafang Shao | 11d6761 | 2020-05-07 18:35:43 -0700 | [diff] [blame] | 5196 | return ERR_PTR(error); |
Glauber Costa | d142e3e | 2013-02-22 16:34:52 -0800 | [diff] [blame] | 5197 | } |
| 5198 | |
Johannes Weiner | 0b8f73e | 2016-01-20 15:02:53 -0800 | [diff] [blame] | 5199 | static struct cgroup_subsys_state * __ref |
| 5200 | mem_cgroup_css_alloc(struct cgroup_subsys_state *parent_css) |
Glauber Costa | d142e3e | 2013-02-22 16:34:52 -0800 | [diff] [blame] | 5201 | { |
Johannes Weiner | 0b8f73e | 2016-01-20 15:02:53 -0800 | [diff] [blame] | 5202 | struct mem_cgroup *parent = mem_cgroup_from_css(parent_css); |
Roman Gushchin | b87d8ce | 2020-10-17 16:13:40 -0700 | [diff] [blame] | 5203 | struct mem_cgroup *memcg, *old_memcg; |
Johannes Weiner | 0b8f73e | 2016-01-20 15:02:53 -0800 | [diff] [blame] | 5204 | long error = -ENOMEM; |
Glauber Costa | d142e3e | 2013-02-22 16:34:52 -0800 | [diff] [blame] | 5205 | |
Roman Gushchin | b87d8ce | 2020-10-17 16:13:40 -0700 | [diff] [blame] | 5206 | old_memcg = set_active_memcg(parent); |
Johannes Weiner | 0b8f73e | 2016-01-20 15:02:53 -0800 | [diff] [blame] | 5207 | memcg = mem_cgroup_alloc(); |
Roman Gushchin | b87d8ce | 2020-10-17 16:13:40 -0700 | [diff] [blame] | 5208 | set_active_memcg(old_memcg); |
Yafang Shao | 11d6761 | 2020-05-07 18:35:43 -0700 | [diff] [blame] | 5209 | if (IS_ERR(memcg)) |
| 5210 | return ERR_CAST(memcg); |
Li Zefan | 4219b2d | 2013-09-23 16:56:29 +0800 | [diff] [blame] | 5211 | |
Jakub Kicinski | d1663a9 | 2020-06-01 21:49:49 -0700 | [diff] [blame] | 5212 | page_counter_set_high(&memcg->memory, PAGE_COUNTER_MAX); |
Johannes Weiner | 0b8f73e | 2016-01-20 15:02:53 -0800 | [diff] [blame] | 5213 | memcg->soft_limit = PAGE_COUNTER_MAX; |
Jakub Kicinski | 4b82ab4 | 2020-06-01 21:49:52 -0700 | [diff] [blame] | 5214 | page_counter_set_high(&memcg->swap, PAGE_COUNTER_MAX); |
Johannes Weiner | 0b8f73e | 2016-01-20 15:02:53 -0800 | [diff] [blame] | 5215 | if (parent) { |
| 5216 | memcg->swappiness = mem_cgroup_swappiness(parent); |
| 5217 | memcg->oom_kill_disable = parent->oom_kill_disable; |
Roman Gushchin | bef8620 | 2020-12-14 19:06:49 -0800 | [diff] [blame] | 5218 | |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 5219 | page_counter_init(&memcg->memory, &parent->memory); |
Vladimir Davydov | 37e8435 | 2016-01-20 15:02:56 -0800 | [diff] [blame] | 5220 | page_counter_init(&memcg->swap, &parent->swap); |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 5221 | page_counter_init(&memcg->kmem, &parent->kmem); |
Johannes Weiner | 0db1529 | 2016-01-20 15:02:50 -0800 | [diff] [blame] | 5222 | page_counter_init(&memcg->tcpmem, &parent->tcpmem); |
Balbir Singh | 18f59ea | 2009-01-07 18:08:07 -0800 | [diff] [blame] | 5223 | } else { |
Roman Gushchin | bef8620 | 2020-12-14 19:06:49 -0800 | [diff] [blame] | 5224 | page_counter_init(&memcg->memory, NULL); |
| 5225 | page_counter_init(&memcg->swap, NULL); |
| 5226 | page_counter_init(&memcg->kmem, NULL); |
| 5227 | page_counter_init(&memcg->tcpmem, NULL); |
Vladimir Davydov | d644163 | 2014-01-23 15:53:09 -0800 | [diff] [blame] | 5228 | |
Johannes Weiner | 0b8f73e | 2016-01-20 15:02:53 -0800 | [diff] [blame] | 5229 | root_mem_cgroup = memcg; |
| 5230 | return &memcg->css; |
| 5231 | } |
| 5232 | |
Roman Gushchin | bef8620 | 2020-12-14 19:06:49 -0800 | [diff] [blame] | 5233 | /* The following stuff does not apply to the root */ |
Vladimir Davydov | b313aee | 2016-03-17 14:18:27 -0700 | [diff] [blame] | 5234 | error = memcg_online_kmem(memcg); |
Johannes Weiner | 0b8f73e | 2016-01-20 15:02:53 -0800 | [diff] [blame] | 5235 | if (error) |
| 5236 | goto fail; |
Johannes Weiner | 127424c | 2016-01-20 15:02:32 -0800 | [diff] [blame] | 5237 | |
Johannes Weiner | f7e1cb6 | 2016-01-14 15:21:29 -0800 | [diff] [blame] | 5238 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket) |
Johannes Weiner | ef12947 | 2016-01-14 15:21:34 -0800 | [diff] [blame] | 5239 | static_branch_inc(&memcg_sockets_enabled_key); |
Johannes Weiner | f7e1cb6 | 2016-01-14 15:21:29 -0800 | [diff] [blame] | 5240 | |
Johannes Weiner | 0b8f73e | 2016-01-20 15:02:53 -0800 | [diff] [blame] | 5241 | return &memcg->css; |
| 5242 | fail: |
Kirill Tkhai | 7e97de0 | 2018-08-02 15:36:01 -0700 | [diff] [blame] | 5243 | mem_cgroup_id_remove(memcg); |
Johannes Weiner | 0b8f73e | 2016-01-20 15:02:53 -0800 | [diff] [blame] | 5244 | mem_cgroup_free(memcg); |
Yafang Shao | 11d6761 | 2020-05-07 18:35:43 -0700 | [diff] [blame] | 5245 | return ERR_PTR(error); |
Johannes Weiner | 0b8f73e | 2016-01-20 15:02:53 -0800 | [diff] [blame] | 5246 | } |
| 5247 | |
Johannes Weiner | 73f576c | 2016-07-20 15:44:57 -0700 | [diff] [blame] | 5248 | static int mem_cgroup_css_online(struct cgroup_subsys_state *css) |
Johannes Weiner | 0b8f73e | 2016-01-20 15:02:53 -0800 | [diff] [blame] | 5249 | { |
Vladimir Davydov | 58fa2a5 | 2016-10-07 16:57:29 -0700 | [diff] [blame] | 5250 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
| 5251 | |
Kirill Tkhai | 0a4465d | 2018-08-17 15:47:37 -0700 | [diff] [blame] | 5252 | /* |
| 5253 | * A memcg must be visible for memcg_expand_shrinker_maps() |
| 5254 | * by the time the maps are allocated. So, we allocate maps |
| 5255 | * here, when for_each_mem_cgroup() can't skip it. |
| 5256 | */ |
| 5257 | if (memcg_alloc_shrinker_maps(memcg)) { |
| 5258 | mem_cgroup_id_remove(memcg); |
| 5259 | return -ENOMEM; |
| 5260 | } |
| 5261 | |
Johannes Weiner | 73f576c | 2016-07-20 15:44:57 -0700 | [diff] [blame] | 5262 | /* Online state pins memcg ID, memcg ID pins CSS */ |
Kirill Tkhai | 1c2d479 | 2018-10-26 15:09:28 -0700 | [diff] [blame] | 5263 | refcount_set(&memcg->id.ref, 1); |
Johannes Weiner | 73f576c | 2016-07-20 15:44:57 -0700 | [diff] [blame] | 5264 | css_get(css); |
Johannes Weiner | 2f7dd7a | 2014-10-02 16:16:57 -0700 | [diff] [blame] | 5265 | return 0; |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 5266 | } |
| 5267 | |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 5268 | static void mem_cgroup_css_offline(struct cgroup_subsys_state *css) |
KAMEZAWA Hiroyuki | df878fb | 2008-02-07 00:14:28 -0800 | [diff] [blame] | 5269 | { |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 5270 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
Tejun Heo | 3bc942f | 2013-11-22 18:20:44 -0500 | [diff] [blame] | 5271 | struct mem_cgroup_event *event, *tmp; |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 5272 | |
| 5273 | /* |
| 5274 | * Unregister events and notify userspace. |
| 5275 | * Notify userspace about cgroup removing only after rmdir of cgroup |
| 5276 | * directory to avoid race between userspace and kernelspace. |
| 5277 | */ |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 5278 | spin_lock(&memcg->event_list_lock); |
| 5279 | list_for_each_entry_safe(event, tmp, &memcg->event_list, list) { |
Tejun Heo | 79bd981 | 2013-11-22 18:20:42 -0500 | [diff] [blame] | 5280 | list_del_init(&event->list); |
| 5281 | schedule_work(&event->remove); |
| 5282 | } |
Tejun Heo | fba9480 | 2013-11-22 18:20:43 -0500 | [diff] [blame] | 5283 | spin_unlock(&memcg->event_list_lock); |
KAMEZAWA Hiroyuki | ec64f51 | 2009-04-02 16:57:26 -0700 | [diff] [blame] | 5284 | |
Roman Gushchin | bf8d5d5 | 2018-06-07 17:07:46 -0700 | [diff] [blame] | 5285 | page_counter_set_min(&memcg->memory, 0); |
Roman Gushchin | 2306715 | 2018-06-07 17:06:22 -0700 | [diff] [blame] | 5286 | page_counter_set_low(&memcg->memory, 0); |
Roman Gushchin | 63677c74 | 2017-09-06 16:21:47 -0700 | [diff] [blame] | 5287 | |
Johannes Weiner | 567e9ab | 2016-01-20 15:02:24 -0800 | [diff] [blame] | 5288 | memcg_offline_kmem(memcg); |
Tejun Heo | 52ebea7 | 2015-05-22 17:13:37 -0400 | [diff] [blame] | 5289 | wb_memcg_offline(memcg); |
Johannes Weiner | 73f576c | 2016-07-20 15:44:57 -0700 | [diff] [blame] | 5290 | |
Roman Gushchin | 591edfb | 2018-10-26 15:03:23 -0700 | [diff] [blame] | 5291 | drain_all_stock(memcg); |
| 5292 | |
Johannes Weiner | 73f576c | 2016-07-20 15:44:57 -0700 | [diff] [blame] | 5293 | mem_cgroup_id_put(memcg); |
KAMEZAWA Hiroyuki | df878fb | 2008-02-07 00:14:28 -0800 | [diff] [blame] | 5294 | } |
| 5295 | |
Vladimir Davydov | 6df3868 | 2015-12-29 14:54:10 -0800 | [diff] [blame] | 5296 | static void mem_cgroup_css_released(struct cgroup_subsys_state *css) |
| 5297 | { |
| 5298 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
| 5299 | |
| 5300 | invalidate_reclaim_iterators(memcg); |
| 5301 | } |
| 5302 | |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 5303 | static void mem_cgroup_css_free(struct cgroup_subsys_state *css) |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 5304 | { |
Tejun Heo | eb95419 | 2013-08-08 20:11:23 -0400 | [diff] [blame] | 5305 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
Tejun Heo | 97b2782 | 2019-08-26 09:06:56 -0700 | [diff] [blame] | 5306 | int __maybe_unused i; |
Daisuke Nishimura | c268e99 | 2009-01-15 13:51:13 -0800 | [diff] [blame] | 5307 | |
Tejun Heo | 97b2782 | 2019-08-26 09:06:56 -0700 | [diff] [blame] | 5308 | #ifdef CONFIG_CGROUP_WRITEBACK |
| 5309 | for (i = 0; i < MEMCG_CGWB_FRN_CNT; i++) |
| 5310 | wb_wait_for_completion(&memcg->cgwb_frn[i].done); |
| 5311 | #endif |
Johannes Weiner | f7e1cb6 | 2016-01-14 15:21:29 -0800 | [diff] [blame] | 5312 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys) && !cgroup_memory_nosocket) |
Johannes Weiner | ef12947 | 2016-01-14 15:21:34 -0800 | [diff] [blame] | 5313 | static_branch_dec(&memcg_sockets_enabled_key); |
Johannes Weiner | 3893e30 | 2016-01-20 15:02:29 -0800 | [diff] [blame] | 5314 | |
Johannes Weiner | 0db1529 | 2016-01-20 15:02:50 -0800 | [diff] [blame] | 5315 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && memcg->tcpmem_active) |
Vladimir Davydov | d55f90b | 2016-01-20 15:02:44 -0800 | [diff] [blame] | 5316 | static_branch_dec(&memcg_sockets_enabled_key); |
Johannes Weiner | 3893e30 | 2016-01-20 15:02:29 -0800 | [diff] [blame] | 5317 | |
Johannes Weiner | 0b8f73e | 2016-01-20 15:02:53 -0800 | [diff] [blame] | 5318 | vmpressure_cleanup(&memcg->vmpressure); |
| 5319 | cancel_work_sync(&memcg->high_work); |
| 5320 | mem_cgroup_remove_from_trees(memcg); |
Kirill Tkhai | 0a4465d | 2018-08-17 15:47:37 -0700 | [diff] [blame] | 5321 | memcg_free_shrinker_maps(memcg); |
Johannes Weiner | d886f4e | 2016-01-20 15:02:47 -0800 | [diff] [blame] | 5322 | memcg_free_kmem(memcg); |
Johannes Weiner | 0b8f73e | 2016-01-20 15:02:53 -0800 | [diff] [blame] | 5323 | mem_cgroup_free(memcg); |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 5324 | } |
| 5325 | |
Tejun Heo | 1ced953 | 2014-07-08 18:02:57 -0400 | [diff] [blame] | 5326 | /** |
| 5327 | * mem_cgroup_css_reset - reset the states of a mem_cgroup |
| 5328 | * @css: the target css |
| 5329 | * |
| 5330 | * Reset the states of the mem_cgroup associated with @css. This is |
| 5331 | * invoked when the userland requests disabling on the default hierarchy |
| 5332 | * but the memcg is pinned through dependency. The memcg should stop |
| 5333 | * applying policies and should revert to the vanilla state as it may be |
| 5334 | * made visible again. |
| 5335 | * |
| 5336 | * The current implementation only resets the essential configurations. |
| 5337 | * This needs to be expanded to cover all the visible parts. |
| 5338 | */ |
| 5339 | static void mem_cgroup_css_reset(struct cgroup_subsys_state *css) |
| 5340 | { |
| 5341 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
| 5342 | |
Roman Gushchin | bbec2e1 | 2018-06-07 17:06:18 -0700 | [diff] [blame] | 5343 | page_counter_set_max(&memcg->memory, PAGE_COUNTER_MAX); |
| 5344 | page_counter_set_max(&memcg->swap, PAGE_COUNTER_MAX); |
Roman Gushchin | bbec2e1 | 2018-06-07 17:06:18 -0700 | [diff] [blame] | 5345 | page_counter_set_max(&memcg->kmem, PAGE_COUNTER_MAX); |
| 5346 | page_counter_set_max(&memcg->tcpmem, PAGE_COUNTER_MAX); |
Roman Gushchin | bf8d5d5 | 2018-06-07 17:07:46 -0700 | [diff] [blame] | 5347 | page_counter_set_min(&memcg->memory, 0); |
Roman Gushchin | 2306715 | 2018-06-07 17:06:22 -0700 | [diff] [blame] | 5348 | page_counter_set_low(&memcg->memory, 0); |
Jakub Kicinski | d1663a9 | 2020-06-01 21:49:49 -0700 | [diff] [blame] | 5349 | page_counter_set_high(&memcg->memory, PAGE_COUNTER_MAX); |
Johannes Weiner | 24d404d | 2015-01-08 14:32:35 -0800 | [diff] [blame] | 5350 | memcg->soft_limit = PAGE_COUNTER_MAX; |
Jakub Kicinski | 4b82ab4 | 2020-06-01 21:49:52 -0700 | [diff] [blame] | 5351 | page_counter_set_high(&memcg->swap, PAGE_COUNTER_MAX); |
Tejun Heo | 2529bb3 | 2015-05-22 18:23:34 -0400 | [diff] [blame] | 5352 | memcg_wb_domain_size_changed(memcg); |
Tejun Heo | 1ced953 | 2014-07-08 18:02:57 -0400 | [diff] [blame] | 5353 | } |
| 5354 | |
Johannes Weiner | 2d146aa | 2021-04-29 22:56:26 -0700 | [diff] [blame] | 5355 | static void mem_cgroup_css_rstat_flush(struct cgroup_subsys_state *css, int cpu) |
| 5356 | { |
| 5357 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
| 5358 | struct mem_cgroup *parent = parent_mem_cgroup(memcg); |
| 5359 | struct memcg_vmstats_percpu *statc; |
| 5360 | long delta, v; |
| 5361 | int i; |
| 5362 | |
| 5363 | statc = per_cpu_ptr(memcg->vmstats_percpu, cpu); |
| 5364 | |
| 5365 | for (i = 0; i < MEMCG_NR_STAT; i++) { |
| 5366 | /* |
| 5367 | * Collect the aggregated propagation counts of groups |
| 5368 | * below us. We're in a per-cpu loop here and this is |
| 5369 | * a global counter, so the first cycle will get them. |
| 5370 | */ |
| 5371 | delta = memcg->vmstats.state_pending[i]; |
| 5372 | if (delta) |
| 5373 | memcg->vmstats.state_pending[i] = 0; |
| 5374 | |
| 5375 | /* Add CPU changes on this level since the last flush */ |
| 5376 | v = READ_ONCE(statc->state[i]); |
| 5377 | if (v != statc->state_prev[i]) { |
| 5378 | delta += v - statc->state_prev[i]; |
| 5379 | statc->state_prev[i] = v; |
| 5380 | } |
| 5381 | |
| 5382 | if (!delta) |
| 5383 | continue; |
| 5384 | |
| 5385 | /* Aggregate counts on this level and propagate upwards */ |
| 5386 | memcg->vmstats.state[i] += delta; |
| 5387 | if (parent) |
| 5388 | parent->vmstats.state_pending[i] += delta; |
| 5389 | } |
| 5390 | |
| 5391 | for (i = 0; i < NR_VM_EVENT_ITEMS; i++) { |
| 5392 | delta = memcg->vmstats.events_pending[i]; |
| 5393 | if (delta) |
| 5394 | memcg->vmstats.events_pending[i] = 0; |
| 5395 | |
| 5396 | v = READ_ONCE(statc->events[i]); |
| 5397 | if (v != statc->events_prev[i]) { |
| 5398 | delta += v - statc->events_prev[i]; |
| 5399 | statc->events_prev[i] = v; |
| 5400 | } |
| 5401 | |
| 5402 | if (!delta) |
| 5403 | continue; |
| 5404 | |
| 5405 | memcg->vmstats.events[i] += delta; |
| 5406 | if (parent) |
| 5407 | parent->vmstats.events_pending[i] += delta; |
| 5408 | } |
| 5409 | } |
| 5410 | |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 5411 | #ifdef CONFIG_MMU |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 5412 | /* Handlers for move charge at task migration. */ |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 5413 | static int mem_cgroup_do_precharge(unsigned long count) |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 5414 | { |
Johannes Weiner | 05b8430 | 2014-08-06 16:05:59 -0700 | [diff] [blame] | 5415 | int ret; |
Johannes Weiner | 9476db9 | 2014-08-06 16:05:55 -0700 | [diff] [blame] | 5416 | |
Mel Gorman | d0164ad | 2015-11-06 16:28:21 -0800 | [diff] [blame] | 5417 | /* Try a single bulk charge without reclaim first, kswapd may wake */ |
| 5418 | ret = try_charge(mc.to, GFP_KERNEL & ~__GFP_DIRECT_RECLAIM, count); |
Johannes Weiner | 9476db9 | 2014-08-06 16:05:55 -0700 | [diff] [blame] | 5419 | if (!ret) { |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 5420 | mc.precharge += count; |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 5421 | return ret; |
| 5422 | } |
Johannes Weiner | 9476db9 | 2014-08-06 16:05:55 -0700 | [diff] [blame] | 5423 | |
David Rientjes | 3674534 | 2017-01-24 15:18:10 -0800 | [diff] [blame] | 5424 | /* Try charges one by one with reclaim, but do not retry */ |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 5425 | while (count--) { |
David Rientjes | 3674534 | 2017-01-24 15:18:10 -0800 | [diff] [blame] | 5426 | ret = try_charge(mc.to, GFP_KERNEL | __GFP_NORETRY, 1); |
KAMEZAWA Hiroyuki | 38c5d72 | 2012-01-12 17:19:01 -0800 | [diff] [blame] | 5427 | if (ret) |
KAMEZAWA Hiroyuki | 38c5d72 | 2012-01-12 17:19:01 -0800 | [diff] [blame] | 5428 | return ret; |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 5429 | mc.precharge++; |
Johannes Weiner | 9476db9 | 2014-08-06 16:05:55 -0700 | [diff] [blame] | 5430 | cond_resched(); |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 5431 | } |
Johannes Weiner | 9476db9 | 2014-08-06 16:05:55 -0700 | [diff] [blame] | 5432 | return 0; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5433 | } |
| 5434 | |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5435 | union mc_target { |
| 5436 | struct page *page; |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 5437 | swp_entry_t ent; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5438 | }; |
| 5439 | |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5440 | enum mc_target_type { |
Naoya Horiguchi | 8d32ff8 | 2012-03-21 16:34:27 -0700 | [diff] [blame] | 5441 | MC_TARGET_NONE = 0, |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5442 | MC_TARGET_PAGE, |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 5443 | MC_TARGET_SWAP, |
Jérôme Glisse | c733a82 | 2017-09-08 16:11:54 -0700 | [diff] [blame] | 5444 | MC_TARGET_DEVICE, |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5445 | }; |
| 5446 | |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 5447 | static struct page *mc_handle_present_pte(struct vm_area_struct *vma, |
| 5448 | unsigned long addr, pte_t ptent) |
| 5449 | { |
Christoph Hellwig | 25b2995 | 2019-06-13 22:50:49 +0200 | [diff] [blame] | 5450 | struct page *page = vm_normal_page(vma, addr, ptent); |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 5451 | |
| 5452 | if (!page || !page_mapped(page)) |
| 5453 | return NULL; |
| 5454 | if (PageAnon(page)) { |
Johannes Weiner | 1dfab5a | 2015-02-11 15:26:09 -0800 | [diff] [blame] | 5455 | if (!(mc.flags & MOVE_ANON)) |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 5456 | return NULL; |
Johannes Weiner | 1dfab5a | 2015-02-11 15:26:09 -0800 | [diff] [blame] | 5457 | } else { |
| 5458 | if (!(mc.flags & MOVE_FILE)) |
| 5459 | return NULL; |
| 5460 | } |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 5461 | if (!get_page_unless_zero(page)) |
| 5462 | return NULL; |
| 5463 | |
| 5464 | return page; |
| 5465 | } |
| 5466 | |
Jérôme Glisse | c733a82 | 2017-09-08 16:11:54 -0700 | [diff] [blame] | 5467 | #if defined(CONFIG_SWAP) || defined(CONFIG_DEVICE_PRIVATE) |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 5468 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
Li RongQing | 48406ef | 2016-07-26 15:22:14 -0700 | [diff] [blame] | 5469 | pte_t ptent, swp_entry_t *entry) |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 5470 | { |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 5471 | struct page *page = NULL; |
| 5472 | swp_entry_t ent = pte_to_swp_entry(ptent); |
| 5473 | |
Ralph Campbell | 9a13715 | 2020-10-13 16:53:13 -0700 | [diff] [blame] | 5474 | if (!(mc.flags & MOVE_ANON)) |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 5475 | return NULL; |
Jérôme Glisse | c733a82 | 2017-09-08 16:11:54 -0700 | [diff] [blame] | 5476 | |
| 5477 | /* |
| 5478 | * Handle MEMORY_DEVICE_PRIVATE which are ZONE_DEVICE page belonging to |
| 5479 | * a device and because they are not accessible by CPU they are store |
| 5480 | * as special swap entry in the CPU page table. |
| 5481 | */ |
| 5482 | if (is_device_private_entry(ent)) { |
| 5483 | page = device_private_entry_to_page(ent); |
| 5484 | /* |
| 5485 | * MEMORY_DEVICE_PRIVATE means ZONE_DEVICE page and which have |
| 5486 | * a refcount of 1 when free (unlike normal page) |
| 5487 | */ |
| 5488 | if (!page_ref_add_unless(page, 1, 1)) |
| 5489 | return NULL; |
| 5490 | return page; |
| 5491 | } |
| 5492 | |
Ralph Campbell | 9a13715 | 2020-10-13 16:53:13 -0700 | [diff] [blame] | 5493 | if (non_swap_entry(ent)) |
| 5494 | return NULL; |
| 5495 | |
KAMEZAWA Hiroyuki | 4b91355 | 2012-05-29 15:06:51 -0700 | [diff] [blame] | 5496 | /* |
| 5497 | * Because lookup_swap_cache() updates some statistics counter, |
| 5498 | * we call find_get_page() with swapper_space directly. |
| 5499 | */ |
Huang Ying | f6ab1f7 | 2016-10-07 17:00:21 -0700 | [diff] [blame] | 5500 | page = find_get_page(swap_address_space(ent), swp_offset(ent)); |
Johannes Weiner | 2d1c498 | 2020-06-03 16:02:14 -0700 | [diff] [blame] | 5501 | entry->val = ent.val; |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 5502 | |
| 5503 | return page; |
| 5504 | } |
KAMEZAWA Hiroyuki | 4b91355 | 2012-05-29 15:06:51 -0700 | [diff] [blame] | 5505 | #else |
| 5506 | static struct page *mc_handle_swap_pte(struct vm_area_struct *vma, |
Li RongQing | 48406ef | 2016-07-26 15:22:14 -0700 | [diff] [blame] | 5507 | pte_t ptent, swp_entry_t *entry) |
KAMEZAWA Hiroyuki | 4b91355 | 2012-05-29 15:06:51 -0700 | [diff] [blame] | 5508 | { |
| 5509 | return NULL; |
| 5510 | } |
| 5511 | #endif |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 5512 | |
Daisuke Nishimura | 87946a7 | 2010-05-26 14:42:39 -0700 | [diff] [blame] | 5513 | static struct page *mc_handle_file_pte(struct vm_area_struct *vma, |
| 5514 | unsigned long addr, pte_t ptent, swp_entry_t *entry) |
| 5515 | { |
Daisuke Nishimura | 87946a7 | 2010-05-26 14:42:39 -0700 | [diff] [blame] | 5516 | if (!vma->vm_file) /* anonymous vma */ |
| 5517 | return NULL; |
Johannes Weiner | 1dfab5a | 2015-02-11 15:26:09 -0800 | [diff] [blame] | 5518 | if (!(mc.flags & MOVE_FILE)) |
Daisuke Nishimura | 87946a7 | 2010-05-26 14:42:39 -0700 | [diff] [blame] | 5519 | return NULL; |
| 5520 | |
Daisuke Nishimura | 87946a7 | 2010-05-26 14:42:39 -0700 | [diff] [blame] | 5521 | /* 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] | 5522 | /* shmem/tmpfs may report page out on swap: account for that too. */ |
Matthew Wilcox (Oracle) | f5df863 | 2020-10-13 16:51:21 -0700 | [diff] [blame] | 5523 | return find_get_incore_page(vma->vm_file->f_mapping, |
| 5524 | linear_page_index(vma, addr)); |
Daisuke Nishimura | 87946a7 | 2010-05-26 14:42:39 -0700 | [diff] [blame] | 5525 | } |
| 5526 | |
Chen Gang | b1b0dea | 2015-04-14 15:47:35 -0700 | [diff] [blame] | 5527 | /** |
| 5528 | * mem_cgroup_move_account - move account of the page |
| 5529 | * @page: the page |
Li RongQing | 25843c2 | 2016-07-26 15:26:56 -0700 | [diff] [blame] | 5530 | * @compound: charge the page as compound or small page |
Chen Gang | b1b0dea | 2015-04-14 15:47:35 -0700 | [diff] [blame] | 5531 | * @from: mem_cgroup which the page is moved from. |
| 5532 | * @to: mem_cgroup which the page is moved to. @from != @to. |
| 5533 | * |
Kirill A. Shutemov | 3ac808f | 2016-01-15 16:53:07 -0800 | [diff] [blame] | 5534 | * 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] | 5535 | * |
| 5536 | * This function doesn't do "charge" to new cgroup and doesn't do "uncharge" |
| 5537 | * from old cgroup. |
| 5538 | */ |
| 5539 | static int mem_cgroup_move_account(struct page *page, |
Kirill A. Shutemov | f627c2f | 2016-01-15 16:52:20 -0800 | [diff] [blame] | 5540 | bool compound, |
Chen Gang | b1b0dea | 2015-04-14 15:47:35 -0700 | [diff] [blame] | 5541 | struct mem_cgroup *from, |
| 5542 | struct mem_cgroup *to) |
| 5543 | { |
Konstantin Khlebnikov | ae8af43 | 2019-10-18 20:20:11 -0700 | [diff] [blame] | 5544 | struct lruvec *from_vec, *to_vec; |
| 5545 | struct pglist_data *pgdat; |
Matthew Wilcox (Oracle) | 6c35784 | 2020-08-14 17:30:37 -0700 | [diff] [blame] | 5546 | unsigned int nr_pages = compound ? thp_nr_pages(page) : 1; |
Chen Gang | b1b0dea | 2015-04-14 15:47:35 -0700 | [diff] [blame] | 5547 | int ret; |
| 5548 | |
| 5549 | VM_BUG_ON(from == to); |
| 5550 | VM_BUG_ON_PAGE(PageLRU(page), page); |
Kirill A. Shutemov | f627c2f | 2016-01-15 16:52:20 -0800 | [diff] [blame] | 5551 | VM_BUG_ON(compound && !PageTransHuge(page)); |
Chen Gang | b1b0dea | 2015-04-14 15:47:35 -0700 | [diff] [blame] | 5552 | |
| 5553 | /* |
Johannes Weiner | 6a93ca8 | 2016-03-15 14:57:19 -0700 | [diff] [blame] | 5554 | * Prevent mem_cgroup_migrate() from looking at |
Roman Gushchin | bcfe06b | 2020-12-01 13:58:27 -0800 | [diff] [blame] | 5555 | * page's memory cgroup of its source page while we change it. |
Chen Gang | b1b0dea | 2015-04-14 15:47:35 -0700 | [diff] [blame] | 5556 | */ |
Kirill A. Shutemov | f627c2f | 2016-01-15 16:52:20 -0800 | [diff] [blame] | 5557 | ret = -EBUSY; |
Chen Gang | b1b0dea | 2015-04-14 15:47:35 -0700 | [diff] [blame] | 5558 | if (!trylock_page(page)) |
| 5559 | goto out; |
| 5560 | |
| 5561 | ret = -EINVAL; |
Roman Gushchin | bcfe06b | 2020-12-01 13:58:27 -0800 | [diff] [blame] | 5562 | if (page_memcg(page) != from) |
Chen Gang | b1b0dea | 2015-04-14 15:47:35 -0700 | [diff] [blame] | 5563 | goto out_unlock; |
| 5564 | |
Konstantin Khlebnikov | ae8af43 | 2019-10-18 20:20:11 -0700 | [diff] [blame] | 5565 | pgdat = page_pgdat(page); |
Johannes Weiner | 867e5e1 | 2019-11-30 17:55:34 -0800 | [diff] [blame] | 5566 | from_vec = mem_cgroup_lruvec(from, pgdat); |
| 5567 | to_vec = mem_cgroup_lruvec(to, pgdat); |
Konstantin Khlebnikov | ae8af43 | 2019-10-18 20:20:11 -0700 | [diff] [blame] | 5568 | |
Johannes Weiner | abb242f | 2020-06-03 16:01:28 -0700 | [diff] [blame] | 5569 | lock_page_memcg(page); |
Chen Gang | b1b0dea | 2015-04-14 15:47:35 -0700 | [diff] [blame] | 5570 | |
Johannes Weiner | be5d0a7 | 2020-06-03 16:01:57 -0700 | [diff] [blame] | 5571 | if (PageAnon(page)) { |
| 5572 | if (page_mapped(page)) { |
| 5573 | __mod_lruvec_state(from_vec, NR_ANON_MAPPED, -nr_pages); |
| 5574 | __mod_lruvec_state(to_vec, NR_ANON_MAPPED, nr_pages); |
Johannes Weiner | 468c398 | 2020-06-03 16:02:01 -0700 | [diff] [blame] | 5575 | if (PageTransHuge(page)) { |
Muchun Song | 69473e5 | 2021-02-24 12:03:23 -0800 | [diff] [blame] | 5576 | __mod_lruvec_state(from_vec, NR_ANON_THPS, |
| 5577 | -nr_pages); |
| 5578 | __mod_lruvec_state(to_vec, NR_ANON_THPS, |
| 5579 | nr_pages); |
Johannes Weiner | 468c398 | 2020-06-03 16:02:01 -0700 | [diff] [blame] | 5580 | } |
Johannes Weiner | be5d0a7 | 2020-06-03 16:01:57 -0700 | [diff] [blame] | 5581 | } |
| 5582 | } else { |
Johannes Weiner | 0d1c207 | 2020-06-03 16:01:54 -0700 | [diff] [blame] | 5583 | __mod_lruvec_state(from_vec, NR_FILE_PAGES, -nr_pages); |
| 5584 | __mod_lruvec_state(to_vec, NR_FILE_PAGES, nr_pages); |
| 5585 | |
| 5586 | if (PageSwapBacked(page)) { |
| 5587 | __mod_lruvec_state(from_vec, NR_SHMEM, -nr_pages); |
| 5588 | __mod_lruvec_state(to_vec, NR_SHMEM, nr_pages); |
| 5589 | } |
| 5590 | |
Johannes Weiner | 49e50d2 | 2020-06-03 16:01:47 -0700 | [diff] [blame] | 5591 | if (page_mapped(page)) { |
| 5592 | __mod_lruvec_state(from_vec, NR_FILE_MAPPED, -nr_pages); |
| 5593 | __mod_lruvec_state(to_vec, NR_FILE_MAPPED, nr_pages); |
| 5594 | } |
Chen Gang | b1b0dea | 2015-04-14 15:47:35 -0700 | [diff] [blame] | 5595 | |
Johannes Weiner | 49e50d2 | 2020-06-03 16:01:47 -0700 | [diff] [blame] | 5596 | if (PageDirty(page)) { |
| 5597 | struct address_space *mapping = page_mapping(page); |
Greg Thelen | c4843a7 | 2015-05-22 17:13:16 -0400 | [diff] [blame] | 5598 | |
Christoph Hellwig | f56753a | 2020-09-24 08:51:40 +0200 | [diff] [blame] | 5599 | if (mapping_can_writeback(mapping)) { |
Johannes Weiner | 49e50d2 | 2020-06-03 16:01:47 -0700 | [diff] [blame] | 5600 | __mod_lruvec_state(from_vec, NR_FILE_DIRTY, |
| 5601 | -nr_pages); |
| 5602 | __mod_lruvec_state(to_vec, NR_FILE_DIRTY, |
| 5603 | nr_pages); |
| 5604 | } |
Greg Thelen | c4843a7 | 2015-05-22 17:13:16 -0400 | [diff] [blame] | 5605 | } |
| 5606 | } |
| 5607 | |
Chen Gang | b1b0dea | 2015-04-14 15:47:35 -0700 | [diff] [blame] | 5608 | if (PageWriteback(page)) { |
Konstantin Khlebnikov | ae8af43 | 2019-10-18 20:20:11 -0700 | [diff] [blame] | 5609 | __mod_lruvec_state(from_vec, NR_WRITEBACK, -nr_pages); |
| 5610 | __mod_lruvec_state(to_vec, NR_WRITEBACK, nr_pages); |
Chen Gang | b1b0dea | 2015-04-14 15:47:35 -0700 | [diff] [blame] | 5611 | } |
| 5612 | |
| 5613 | /* |
Johannes Weiner | abb242f | 2020-06-03 16:01:28 -0700 | [diff] [blame] | 5614 | * All state has been migrated, let's switch to the new memcg. |
| 5615 | * |
Roman Gushchin | bcfe06b | 2020-12-01 13:58:27 -0800 | [diff] [blame] | 5616 | * It is safe to change page's memcg here because the page |
Johannes Weiner | abb242f | 2020-06-03 16:01:28 -0700 | [diff] [blame] | 5617 | * is referenced, charged, isolated, and locked: we can't race |
| 5618 | * with (un)charging, migration, LRU putback, or anything else |
Roman Gushchin | bcfe06b | 2020-12-01 13:58:27 -0800 | [diff] [blame] | 5619 | * that would rely on a stable page's memory cgroup. |
Johannes Weiner | abb242f | 2020-06-03 16:01:28 -0700 | [diff] [blame] | 5620 | * |
| 5621 | * Note that lock_page_memcg is a memcg lock, not a page lock, |
Roman Gushchin | bcfe06b | 2020-12-01 13:58:27 -0800 | [diff] [blame] | 5622 | * to save space. As soon as we switch page's memory cgroup to a |
Johannes Weiner | abb242f | 2020-06-03 16:01:28 -0700 | [diff] [blame] | 5623 | * new memcg that isn't locked, the above state can change |
| 5624 | * concurrently again. Make sure we're truly done with it. |
Chen Gang | b1b0dea | 2015-04-14 15:47:35 -0700 | [diff] [blame] | 5625 | */ |
Johannes Weiner | abb242f | 2020-06-03 16:01:28 -0700 | [diff] [blame] | 5626 | smp_mb(); |
Chen Gang | b1b0dea | 2015-04-14 15:47:35 -0700 | [diff] [blame] | 5627 | |
Johannes Weiner | 1a3e1f4 | 2020-08-06 23:20:45 -0700 | [diff] [blame] | 5628 | css_get(&to->css); |
| 5629 | css_put(&from->css); |
| 5630 | |
Roman Gushchin | bcfe06b | 2020-12-01 13:58:27 -0800 | [diff] [blame] | 5631 | page->memcg_data = (unsigned long)to; |
Yang Shi | 87eaceb | 2019-09-23 15:38:15 -0700 | [diff] [blame] | 5632 | |
Johannes Weiner | abb242f | 2020-06-03 16:01:28 -0700 | [diff] [blame] | 5633 | __unlock_page_memcg(from); |
Chen Gang | b1b0dea | 2015-04-14 15:47:35 -0700 | [diff] [blame] | 5634 | |
| 5635 | ret = 0; |
| 5636 | |
| 5637 | local_irq_disable(); |
Johannes Weiner | 3fba69a | 2020-06-03 16:01:31 -0700 | [diff] [blame] | 5638 | mem_cgroup_charge_statistics(to, page, nr_pages); |
Chen Gang | b1b0dea | 2015-04-14 15:47:35 -0700 | [diff] [blame] | 5639 | memcg_check_events(to, page); |
Johannes Weiner | 3fba69a | 2020-06-03 16:01:31 -0700 | [diff] [blame] | 5640 | mem_cgroup_charge_statistics(from, page, -nr_pages); |
Chen Gang | b1b0dea | 2015-04-14 15:47:35 -0700 | [diff] [blame] | 5641 | memcg_check_events(from, page); |
| 5642 | local_irq_enable(); |
| 5643 | out_unlock: |
| 5644 | unlock_page(page); |
| 5645 | out: |
| 5646 | return ret; |
| 5647 | } |
| 5648 | |
Li RongQing | 7cf7806 | 2016-05-27 14:27:46 -0700 | [diff] [blame] | 5649 | /** |
| 5650 | * get_mctgt_type - get target type of moving charge |
| 5651 | * @vma: the vma the pte to be checked belongs |
| 5652 | * @addr: the address corresponding to the pte to be checked |
| 5653 | * @ptent: the pte to be checked |
| 5654 | * @target: the pointer the target page or swap ent will be stored(can be NULL) |
| 5655 | * |
| 5656 | * Returns |
| 5657 | * 0(MC_TARGET_NONE): if the pte is not a target for move charge. |
| 5658 | * 1(MC_TARGET_PAGE): if the page corresponding to this pte is a target for |
| 5659 | * move charge. if @target is not NULL, the page is stored in target->page |
| 5660 | * with extra refcnt got(Callers should handle it). |
| 5661 | * 2(MC_TARGET_SWAP): if the swap entry corresponding to this pte is a |
| 5662 | * target for charge migration. if @target is not NULL, the entry is stored |
| 5663 | * in target->ent. |
Christoph Hellwig | 25b2995 | 2019-06-13 22:50:49 +0200 | [diff] [blame] | 5664 | * 3(MC_TARGET_DEVICE): like MC_TARGET_PAGE but page is MEMORY_DEVICE_PRIVATE |
| 5665 | * (so ZONE_DEVICE page and thus not on the lru). |
Jérôme Glisse | df6ad69 | 2017-09-08 16:12:24 -0700 | [diff] [blame] | 5666 | * For now we such page is charge like a regular page would be as for all |
| 5667 | * intent and purposes it is just special memory taking the place of a |
| 5668 | * regular page. |
Jérôme Glisse | c733a82 | 2017-09-08 16:11:54 -0700 | [diff] [blame] | 5669 | * |
| 5670 | * See Documentations/vm/hmm.txt and include/linux/hmm.h |
Li RongQing | 7cf7806 | 2016-05-27 14:27:46 -0700 | [diff] [blame] | 5671 | * |
| 5672 | * Called with pte lock held. |
| 5673 | */ |
| 5674 | |
Naoya Horiguchi | 8d32ff8 | 2012-03-21 16:34:27 -0700 | [diff] [blame] | 5675 | 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] | 5676 | unsigned long addr, pte_t ptent, union mc_target *target) |
| 5677 | { |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 5678 | struct page *page = NULL; |
Naoya Horiguchi | 8d32ff8 | 2012-03-21 16:34:27 -0700 | [diff] [blame] | 5679 | enum mc_target_type ret = MC_TARGET_NONE; |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 5680 | swp_entry_t ent = { .val = 0 }; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5681 | |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 5682 | if (pte_present(ptent)) |
| 5683 | page = mc_handle_present_pte(vma, addr, ptent); |
| 5684 | else if (is_swap_pte(ptent)) |
Li RongQing | 48406ef | 2016-07-26 15:22:14 -0700 | [diff] [blame] | 5685 | page = mc_handle_swap_pte(vma, ptent, &ent); |
Kirill A. Shutemov | 0661a33 | 2015-02-10 14:10:04 -0800 | [diff] [blame] | 5686 | else if (pte_none(ptent)) |
Daisuke Nishimura | 87946a7 | 2010-05-26 14:42:39 -0700 | [diff] [blame] | 5687 | page = mc_handle_file_pte(vma, addr, ptent, &ent); |
Daisuke Nishimura | 90254a6 | 2010-05-26 14:42:38 -0700 | [diff] [blame] | 5688 | |
| 5689 | if (!page && !ent.val) |
Naoya Horiguchi | 8d32ff8 | 2012-03-21 16:34:27 -0700 | [diff] [blame] | 5690 | return ret; |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 5691 | if (page) { |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 5692 | /* |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 5693 | * Do only loose check w/o serialization. |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 5694 | * mem_cgroup_move_account() checks the page is valid or |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 5695 | * not under LRU exclusion. |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 5696 | */ |
Roman Gushchin | bcfe06b | 2020-12-01 13:58:27 -0800 | [diff] [blame] | 5697 | if (page_memcg(page) == mc.from) { |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 5698 | ret = MC_TARGET_PAGE; |
Christoph Hellwig | 25b2995 | 2019-06-13 22:50:49 +0200 | [diff] [blame] | 5699 | if (is_device_private_page(page)) |
Jérôme Glisse | c733a82 | 2017-09-08 16:11:54 -0700 | [diff] [blame] | 5700 | ret = MC_TARGET_DEVICE; |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 5701 | if (target) |
| 5702 | target->page = page; |
| 5703 | } |
| 5704 | if (!ret || !target) |
| 5705 | put_page(page); |
| 5706 | } |
Huang Ying | 3e14a57 | 2017-09-06 16:22:37 -0700 | [diff] [blame] | 5707 | /* |
| 5708 | * There is a swap entry and a page doesn't exist or isn't charged. |
| 5709 | * But we cannot move a tail-page in a THP. |
| 5710 | */ |
| 5711 | if (ent.val && !ret && (!page || !PageTransCompound(page)) && |
Li Zefan | 34c00c3 | 2013-09-23 16:56:01 +0800 | [diff] [blame] | 5712 | mem_cgroup_id(mc.from) == lookup_swap_cgroup_id(ent)) { |
KAMEZAWA Hiroyuki | 7f0f154 | 2010-05-11 14:06:58 -0700 | [diff] [blame] | 5713 | ret = MC_TARGET_SWAP; |
| 5714 | if (target) |
| 5715 | target->ent = ent; |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 5716 | } |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5717 | return ret; |
| 5718 | } |
| 5719 | |
Naoya Horiguchi | 1272485 | 2012-03-21 16:34:28 -0700 | [diff] [blame] | 5720 | #ifdef CONFIG_TRANSPARENT_HUGEPAGE |
| 5721 | /* |
Huang Ying | d6810d7 | 2017-09-06 16:22:45 -0700 | [diff] [blame] | 5722 | * We don't consider PMD mapped swapping or file mapped pages because THP does |
| 5723 | * not support them for now. |
Naoya Horiguchi | 1272485 | 2012-03-21 16:34:28 -0700 | [diff] [blame] | 5724 | * Caller should make sure that pmd_trans_huge(pmd) is true. |
| 5725 | */ |
| 5726 | static enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, |
| 5727 | unsigned long addr, pmd_t pmd, union mc_target *target) |
| 5728 | { |
| 5729 | struct page *page = NULL; |
Naoya Horiguchi | 1272485 | 2012-03-21 16:34:28 -0700 | [diff] [blame] | 5730 | enum mc_target_type ret = MC_TARGET_NONE; |
| 5731 | |
Zi Yan | 84c3fc4 | 2017-09-08 16:11:01 -0700 | [diff] [blame] | 5732 | if (unlikely(is_swap_pmd(pmd))) { |
| 5733 | VM_BUG_ON(thp_migration_supported() && |
| 5734 | !is_pmd_migration_entry(pmd)); |
| 5735 | return ret; |
| 5736 | } |
Naoya Horiguchi | 1272485 | 2012-03-21 16:34:28 -0700 | [diff] [blame] | 5737 | page = pmd_page(pmd); |
Sasha Levin | 309381fea | 2014-01-23 15:52:54 -0800 | [diff] [blame] | 5738 | VM_BUG_ON_PAGE(!page || !PageHead(page), page); |
Johannes Weiner | 1dfab5a | 2015-02-11 15:26:09 -0800 | [diff] [blame] | 5739 | if (!(mc.flags & MOVE_ANON)) |
Naoya Horiguchi | 1272485 | 2012-03-21 16:34:28 -0700 | [diff] [blame] | 5740 | return ret; |
Roman Gushchin | bcfe06b | 2020-12-01 13:58:27 -0800 | [diff] [blame] | 5741 | if (page_memcg(page) == mc.from) { |
Naoya Horiguchi | 1272485 | 2012-03-21 16:34:28 -0700 | [diff] [blame] | 5742 | ret = MC_TARGET_PAGE; |
| 5743 | if (target) { |
| 5744 | get_page(page); |
| 5745 | target->page = page; |
| 5746 | } |
| 5747 | } |
| 5748 | return ret; |
| 5749 | } |
| 5750 | #else |
| 5751 | static inline enum mc_target_type get_mctgt_type_thp(struct vm_area_struct *vma, |
| 5752 | unsigned long addr, pmd_t pmd, union mc_target *target) |
| 5753 | { |
| 5754 | return MC_TARGET_NONE; |
| 5755 | } |
| 5756 | #endif |
| 5757 | |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5758 | static int mem_cgroup_count_precharge_pte_range(pmd_t *pmd, |
| 5759 | unsigned long addr, unsigned long end, |
| 5760 | struct mm_walk *walk) |
| 5761 | { |
Naoya Horiguchi | 26bcd64 | 2015-02-11 15:27:57 -0800 | [diff] [blame] | 5762 | struct vm_area_struct *vma = walk->vma; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5763 | pte_t *pte; |
| 5764 | spinlock_t *ptl; |
| 5765 | |
Kirill A. Shutemov | b6ec57f | 2016-01-21 16:40:25 -0800 | [diff] [blame] | 5766 | ptl = pmd_trans_huge_lock(pmd, vma); |
| 5767 | if (ptl) { |
Jérôme Glisse | c733a82 | 2017-09-08 16:11:54 -0700 | [diff] [blame] | 5768 | /* |
| 5769 | * Note their can not be MC_TARGET_DEVICE for now as we do not |
Christoph Hellwig | 25b2995 | 2019-06-13 22:50:49 +0200 | [diff] [blame] | 5770 | * support transparent huge page with MEMORY_DEVICE_PRIVATE but |
| 5771 | * this might change. |
Jérôme Glisse | c733a82 | 2017-09-08 16:11:54 -0700 | [diff] [blame] | 5772 | */ |
Naoya Horiguchi | 1272485 | 2012-03-21 16:34:28 -0700 | [diff] [blame] | 5773 | if (get_mctgt_type_thp(vma, addr, *pmd, NULL) == MC_TARGET_PAGE) |
| 5774 | mc.precharge += HPAGE_PMD_NR; |
Kirill A. Shutemov | bf92915 | 2013-11-14 14:30:54 -0800 | [diff] [blame] | 5775 | spin_unlock(ptl); |
Andrea Arcangeli | 1a5a990 | 2012-03-21 16:33:42 -0700 | [diff] [blame] | 5776 | return 0; |
Naoya Horiguchi | 1272485 | 2012-03-21 16:34:28 -0700 | [diff] [blame] | 5777 | } |
Dave Hansen | 0331932 | 2011-03-22 16:32:56 -0700 | [diff] [blame] | 5778 | |
Andrea Arcangeli | 45f83ce | 2012-03-28 14:42:40 -0700 | [diff] [blame] | 5779 | if (pmd_trans_unstable(pmd)) |
| 5780 | return 0; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5781 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
| 5782 | for (; addr != end; pte++, addr += PAGE_SIZE) |
Naoya Horiguchi | 8d32ff8 | 2012-03-21 16:34:27 -0700 | [diff] [blame] | 5783 | if (get_mctgt_type(vma, addr, *pte, NULL)) |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5784 | mc.precharge++; /* increment precharge temporarily */ |
| 5785 | pte_unmap_unlock(pte - 1, ptl); |
| 5786 | cond_resched(); |
| 5787 | |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 5788 | return 0; |
| 5789 | } |
| 5790 | |
Christoph Hellwig | 7b86ac3 | 2019-08-28 16:19:54 +0200 | [diff] [blame] | 5791 | static const struct mm_walk_ops precharge_walk_ops = { |
| 5792 | .pmd_entry = mem_cgroup_count_precharge_pte_range, |
| 5793 | }; |
| 5794 | |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5795 | static unsigned long mem_cgroup_count_precharge(struct mm_struct *mm) |
| 5796 | { |
| 5797 | unsigned long precharge; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5798 | |
Michel Lespinasse | d8ed45c | 2020-06-08 21:33:25 -0700 | [diff] [blame] | 5799 | mmap_read_lock(mm); |
Christoph Hellwig | 7b86ac3 | 2019-08-28 16:19:54 +0200 | [diff] [blame] | 5800 | walk_page_range(mm, 0, mm->highest_vm_end, &precharge_walk_ops, NULL); |
Michel Lespinasse | d8ed45c | 2020-06-08 21:33:25 -0700 | [diff] [blame] | 5801 | mmap_read_unlock(mm); |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5802 | |
| 5803 | precharge = mc.precharge; |
| 5804 | mc.precharge = 0; |
| 5805 | |
| 5806 | return precharge; |
| 5807 | } |
| 5808 | |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5809 | static int mem_cgroup_precharge_mc(struct mm_struct *mm) |
| 5810 | { |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 5811 | unsigned long precharge = mem_cgroup_count_precharge(mm); |
| 5812 | |
| 5813 | VM_BUG_ON(mc.moving_task); |
| 5814 | mc.moving_task = current; |
| 5815 | return mem_cgroup_do_precharge(precharge); |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5816 | } |
| 5817 | |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 5818 | /* cancels all extra charges on mc.from and mc.to, and wakes up all waiters. */ |
| 5819 | static void __mem_cgroup_clear_mc(void) |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5820 | { |
KAMEZAWA Hiroyuki | 2bd9bb2 | 2010-08-10 18:02:58 -0700 | [diff] [blame] | 5821 | struct mem_cgroup *from = mc.from; |
| 5822 | struct mem_cgroup *to = mc.to; |
| 5823 | |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5824 | /* we must uncharge all the leftover precharges from mc.to */ |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 5825 | if (mc.precharge) { |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 5826 | cancel_charge(mc.to, mc.precharge); |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 5827 | mc.precharge = 0; |
| 5828 | } |
| 5829 | /* |
| 5830 | * we didn't uncharge from mc.from at mem_cgroup_move_account(), so |
| 5831 | * we must uncharge here. |
| 5832 | */ |
| 5833 | if (mc.moved_charge) { |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 5834 | cancel_charge(mc.from, mc.moved_charge); |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 5835 | mc.moved_charge = 0; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5836 | } |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 5837 | /* we must fixup refcnts and charges */ |
| 5838 | if (mc.moved_swap) { |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 5839 | /* uncharge swap account from the old cgroup */ |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 5840 | if (!mem_cgroup_is_root(mc.from)) |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 5841 | page_counter_uncharge(&mc.from->memsw, mc.moved_swap); |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 5842 | |
Vladimir Davydov | 615d66c | 2016-08-11 15:33:03 -0700 | [diff] [blame] | 5843 | mem_cgroup_id_put_many(mc.from, mc.moved_swap); |
| 5844 | |
Johannes Weiner | 05b8430 | 2014-08-06 16:05:59 -0700 | [diff] [blame] | 5845 | /* |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 5846 | * we charged both to->memory and to->memsw, so we |
| 5847 | * should uncharge to->memory. |
Johannes Weiner | 05b8430 | 2014-08-06 16:05:59 -0700 | [diff] [blame] | 5848 | */ |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 5849 | if (!mem_cgroup_is_root(mc.to)) |
Johannes Weiner | 3e32cb2 | 2014-12-10 15:42:31 -0800 | [diff] [blame] | 5850 | page_counter_uncharge(&mc.to->memory, mc.moved_swap); |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 5851 | |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 5852 | mc.moved_swap = 0; |
| 5853 | } |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 5854 | memcg_oom_recover(from); |
| 5855 | memcg_oom_recover(to); |
| 5856 | wake_up_all(&mc.waitq); |
| 5857 | } |
| 5858 | |
| 5859 | static void mem_cgroup_clear_mc(void) |
| 5860 | { |
Tejun Heo | 264a0ae | 2016-04-21 19:09:02 -0400 | [diff] [blame] | 5861 | struct mm_struct *mm = mc.mm; |
| 5862 | |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 5863 | /* |
| 5864 | * we must clear moving_task before waking up waiters at the end of |
| 5865 | * task migration. |
| 5866 | */ |
| 5867 | mc.moving_task = NULL; |
| 5868 | __mem_cgroup_clear_mc(); |
KAMEZAWA Hiroyuki | 2bd9bb2 | 2010-08-10 18:02:58 -0700 | [diff] [blame] | 5869 | spin_lock(&mc.lock); |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5870 | mc.from = NULL; |
| 5871 | mc.to = NULL; |
Tejun Heo | 264a0ae | 2016-04-21 19:09:02 -0400 | [diff] [blame] | 5872 | mc.mm = NULL; |
KAMEZAWA Hiroyuki | 2bd9bb2 | 2010-08-10 18:02:58 -0700 | [diff] [blame] | 5873 | spin_unlock(&mc.lock); |
Tejun Heo | 264a0ae | 2016-04-21 19:09:02 -0400 | [diff] [blame] | 5874 | |
| 5875 | mmput(mm); |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5876 | } |
| 5877 | |
Tejun Heo | 1f7dd3e5 | 2015-12-03 10:18:21 -0500 | [diff] [blame] | 5878 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 5879 | { |
Tejun Heo | 1f7dd3e5 | 2015-12-03 10:18:21 -0500 | [diff] [blame] | 5880 | struct cgroup_subsys_state *css; |
Ross Zwisler | eed67d7 | 2015-12-23 14:53:27 -0700 | [diff] [blame] | 5881 | struct mem_cgroup *memcg = NULL; /* unneeded init to make gcc happy */ |
Tejun Heo | 9f2115f | 2015-09-08 15:01:10 -0700 | [diff] [blame] | 5882 | struct mem_cgroup *from; |
Tejun Heo | 4530edd | 2015-09-11 15:00:19 -0400 | [diff] [blame] | 5883 | struct task_struct *leader, *p; |
Tejun Heo | 9f2115f | 2015-09-08 15:01:10 -0700 | [diff] [blame] | 5884 | struct mm_struct *mm; |
Johannes Weiner | 1dfab5a | 2015-02-11 15:26:09 -0800 | [diff] [blame] | 5885 | unsigned long move_flags; |
Tejun Heo | 9f2115f | 2015-09-08 15:01:10 -0700 | [diff] [blame] | 5886 | int ret = 0; |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 5887 | |
Tejun Heo | 1f7dd3e5 | 2015-12-03 10:18:21 -0500 | [diff] [blame] | 5888 | /* charge immigration isn't supported on the default hierarchy */ |
| 5889 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
Tejun Heo | 9f2115f | 2015-09-08 15:01:10 -0700 | [diff] [blame] | 5890 | return 0; |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 5891 | |
Tejun Heo | 4530edd | 2015-09-11 15:00:19 -0400 | [diff] [blame] | 5892 | /* |
| 5893 | * Multi-process migrations only happen on the default hierarchy |
| 5894 | * where charge immigration is not used. Perform charge |
| 5895 | * immigration if @tset contains a leader and whine if there are |
| 5896 | * multiple. |
| 5897 | */ |
| 5898 | p = NULL; |
Tejun Heo | 1f7dd3e5 | 2015-12-03 10:18:21 -0500 | [diff] [blame] | 5899 | cgroup_taskset_for_each_leader(leader, css, tset) { |
Tejun Heo | 4530edd | 2015-09-11 15:00:19 -0400 | [diff] [blame] | 5900 | WARN_ON_ONCE(p); |
| 5901 | p = leader; |
Tejun Heo | 1f7dd3e5 | 2015-12-03 10:18:21 -0500 | [diff] [blame] | 5902 | memcg = mem_cgroup_from_css(css); |
Tejun Heo | 4530edd | 2015-09-11 15:00:19 -0400 | [diff] [blame] | 5903 | } |
| 5904 | if (!p) |
| 5905 | return 0; |
| 5906 | |
Tejun Heo | 1f7dd3e5 | 2015-12-03 10:18:21 -0500 | [diff] [blame] | 5907 | /* |
| 5908 | * We are now commited to this value whatever it is. Changes in this |
| 5909 | * tunable will only affect upcoming migrations, not the current one. |
| 5910 | * So we need to save it, and keep it going. |
| 5911 | */ |
| 5912 | move_flags = READ_ONCE(memcg->move_charge_at_immigrate); |
| 5913 | if (!move_flags) |
| 5914 | return 0; |
| 5915 | |
Tejun Heo | 9f2115f | 2015-09-08 15:01:10 -0700 | [diff] [blame] | 5916 | from = mem_cgroup_from_task(p); |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 5917 | |
Tejun Heo | 9f2115f | 2015-09-08 15:01:10 -0700 | [diff] [blame] | 5918 | VM_BUG_ON(from == memcg); |
Johannes Weiner | 247b144 | 2014-12-10 15:44:11 -0800 | [diff] [blame] | 5919 | |
Tejun Heo | 9f2115f | 2015-09-08 15:01:10 -0700 | [diff] [blame] | 5920 | mm = get_task_mm(p); |
| 5921 | if (!mm) |
| 5922 | return 0; |
| 5923 | /* We move charges only when we move a owner of the mm */ |
| 5924 | if (mm->owner == p) { |
| 5925 | VM_BUG_ON(mc.from); |
| 5926 | VM_BUG_ON(mc.to); |
| 5927 | VM_BUG_ON(mc.precharge); |
| 5928 | VM_BUG_ON(mc.moved_charge); |
| 5929 | VM_BUG_ON(mc.moved_swap); |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 5930 | |
Tejun Heo | 9f2115f | 2015-09-08 15:01:10 -0700 | [diff] [blame] | 5931 | spin_lock(&mc.lock); |
Tejun Heo | 264a0ae | 2016-04-21 19:09:02 -0400 | [diff] [blame] | 5932 | mc.mm = mm; |
Tejun Heo | 9f2115f | 2015-09-08 15:01:10 -0700 | [diff] [blame] | 5933 | mc.from = from; |
| 5934 | mc.to = memcg; |
| 5935 | mc.flags = move_flags; |
| 5936 | spin_unlock(&mc.lock); |
| 5937 | /* We set mc.moving_task later */ |
| 5938 | |
| 5939 | ret = mem_cgroup_precharge_mc(mm); |
| 5940 | if (ret) |
| 5941 | mem_cgroup_clear_mc(); |
Tejun Heo | 264a0ae | 2016-04-21 19:09:02 -0400 | [diff] [blame] | 5942 | } else { |
| 5943 | mmput(mm); |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 5944 | } |
| 5945 | return ret; |
| 5946 | } |
| 5947 | |
Tejun Heo | 1f7dd3e5 | 2015-12-03 10:18:21 -0500 | [diff] [blame] | 5948 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 5949 | { |
Johannes Weiner | 4e2f245 | 2014-12-10 15:44:08 -0800 | [diff] [blame] | 5950 | if (mc.to) |
| 5951 | mem_cgroup_clear_mc(); |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 5952 | } |
| 5953 | |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5954 | static int mem_cgroup_move_charge_pte_range(pmd_t *pmd, |
| 5955 | unsigned long addr, unsigned long end, |
| 5956 | struct mm_walk *walk) |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 5957 | { |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5958 | int ret = 0; |
Naoya Horiguchi | 26bcd64 | 2015-02-11 15:27:57 -0800 | [diff] [blame] | 5959 | struct vm_area_struct *vma = walk->vma; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5960 | pte_t *pte; |
| 5961 | spinlock_t *ptl; |
Naoya Horiguchi | 1272485 | 2012-03-21 16:34:28 -0700 | [diff] [blame] | 5962 | enum mc_target_type target_type; |
| 5963 | union mc_target target; |
| 5964 | struct page *page; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5965 | |
Kirill A. Shutemov | b6ec57f | 2016-01-21 16:40:25 -0800 | [diff] [blame] | 5966 | ptl = pmd_trans_huge_lock(pmd, vma); |
| 5967 | if (ptl) { |
Hugh Dickins | 62ade86 | 2012-05-18 11:28:34 -0700 | [diff] [blame] | 5968 | if (mc.precharge < HPAGE_PMD_NR) { |
Kirill A. Shutemov | bf92915 | 2013-11-14 14:30:54 -0800 | [diff] [blame] | 5969 | spin_unlock(ptl); |
Naoya Horiguchi | 1272485 | 2012-03-21 16:34:28 -0700 | [diff] [blame] | 5970 | return 0; |
| 5971 | } |
| 5972 | target_type = get_mctgt_type_thp(vma, addr, *pmd, &target); |
| 5973 | if (target_type == MC_TARGET_PAGE) { |
| 5974 | page = target.page; |
| 5975 | if (!isolate_lru_page(page)) { |
Kirill A. Shutemov | f627c2f | 2016-01-15 16:52:20 -0800 | [diff] [blame] | 5976 | if (!mem_cgroup_move_account(page, true, |
Johannes Weiner | 1306a85 | 2014-12-10 15:44:52 -0800 | [diff] [blame] | 5977 | mc.from, mc.to)) { |
Naoya Horiguchi | 1272485 | 2012-03-21 16:34:28 -0700 | [diff] [blame] | 5978 | mc.precharge -= HPAGE_PMD_NR; |
| 5979 | mc.moved_charge += HPAGE_PMD_NR; |
| 5980 | } |
| 5981 | putback_lru_page(page); |
| 5982 | } |
| 5983 | put_page(page); |
Jérôme Glisse | c733a82 | 2017-09-08 16:11:54 -0700 | [diff] [blame] | 5984 | } else if (target_type == MC_TARGET_DEVICE) { |
| 5985 | page = target.page; |
| 5986 | if (!mem_cgroup_move_account(page, true, |
| 5987 | mc.from, mc.to)) { |
| 5988 | mc.precharge -= HPAGE_PMD_NR; |
| 5989 | mc.moved_charge += HPAGE_PMD_NR; |
| 5990 | } |
| 5991 | put_page(page); |
Naoya Horiguchi | 1272485 | 2012-03-21 16:34:28 -0700 | [diff] [blame] | 5992 | } |
Kirill A. Shutemov | bf92915 | 2013-11-14 14:30:54 -0800 | [diff] [blame] | 5993 | spin_unlock(ptl); |
Andrea Arcangeli | 1a5a990 | 2012-03-21 16:33:42 -0700 | [diff] [blame] | 5994 | return 0; |
Naoya Horiguchi | 1272485 | 2012-03-21 16:34:28 -0700 | [diff] [blame] | 5995 | } |
| 5996 | |
Andrea Arcangeli | 45f83ce | 2012-03-28 14:42:40 -0700 | [diff] [blame] | 5997 | if (pmd_trans_unstable(pmd)) |
| 5998 | return 0; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 5999 | retry: |
| 6000 | pte = pte_offset_map_lock(vma->vm_mm, pmd, addr, &ptl); |
| 6001 | for (; addr != end; addr += PAGE_SIZE) { |
| 6002 | pte_t ptent = *(pte++); |
Jérôme Glisse | c733a82 | 2017-09-08 16:11:54 -0700 | [diff] [blame] | 6003 | bool device = false; |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 6004 | swp_entry_t ent; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 6005 | |
| 6006 | if (!mc.precharge) |
| 6007 | break; |
| 6008 | |
Naoya Horiguchi | 8d32ff8 | 2012-03-21 16:34:27 -0700 | [diff] [blame] | 6009 | switch (get_mctgt_type(vma, addr, ptent, &target)) { |
Jérôme Glisse | c733a82 | 2017-09-08 16:11:54 -0700 | [diff] [blame] | 6010 | case MC_TARGET_DEVICE: |
| 6011 | device = true; |
Joe Perches | e4a9bc5 | 2020-04-06 20:08:39 -0700 | [diff] [blame] | 6012 | fallthrough; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 6013 | case MC_TARGET_PAGE: |
| 6014 | page = target.page; |
Kirill A. Shutemov | 53f9263 | 2016-01-15 16:53:42 -0800 | [diff] [blame] | 6015 | /* |
| 6016 | * We can have a part of the split pmd here. Moving it |
| 6017 | * can be done but it would be too convoluted so simply |
| 6018 | * ignore such a partial THP and keep it in original |
| 6019 | * memcg. There should be somebody mapping the head. |
| 6020 | */ |
| 6021 | if (PageTransCompound(page)) |
| 6022 | goto put; |
Jérôme Glisse | c733a82 | 2017-09-08 16:11:54 -0700 | [diff] [blame] | 6023 | if (!device && isolate_lru_page(page)) |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 6024 | goto put; |
Kirill A. Shutemov | f627c2f | 2016-01-15 16:52:20 -0800 | [diff] [blame] | 6025 | if (!mem_cgroup_move_account(page, false, |
| 6026 | mc.from, mc.to)) { |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 6027 | mc.precharge--; |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 6028 | /* we uncharge from mc.from later. */ |
| 6029 | mc.moved_charge++; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 6030 | } |
Jérôme Glisse | c733a82 | 2017-09-08 16:11:54 -0700 | [diff] [blame] | 6031 | if (!device) |
| 6032 | putback_lru_page(page); |
Naoya Horiguchi | 8d32ff8 | 2012-03-21 16:34:27 -0700 | [diff] [blame] | 6033 | put: /* get_mctgt_type() gets the page */ |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 6034 | put_page(page); |
| 6035 | break; |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 6036 | case MC_TARGET_SWAP: |
| 6037 | ent = target.ent; |
Hugh Dickins | e91cbb4 | 2012-05-29 15:06:51 -0700 | [diff] [blame] | 6038 | if (!mem_cgroup_move_swap_account(ent, mc.from, mc.to)) { |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 6039 | mc.precharge--; |
Hugh Dickins | 8d22a93 | 2020-07-23 21:15:24 -0700 | [diff] [blame] | 6040 | mem_cgroup_id_get_many(mc.to, 1); |
| 6041 | /* we fixup other refcnts and charges later. */ |
Daisuke Nishimura | 483c30b | 2010-03-10 15:22:18 -0800 | [diff] [blame] | 6042 | mc.moved_swap++; |
| 6043 | } |
Daisuke Nishimura | 0249144 | 2010-03-10 15:22:17 -0800 | [diff] [blame] | 6044 | break; |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 6045 | default: |
| 6046 | break; |
| 6047 | } |
| 6048 | } |
| 6049 | pte_unmap_unlock(pte - 1, ptl); |
| 6050 | cond_resched(); |
| 6051 | |
| 6052 | if (addr != end) { |
| 6053 | /* |
| 6054 | * We have consumed all precharges we got in can_attach(). |
| 6055 | * We try charge one by one, but don't do any additional |
| 6056 | * charges to mc.to if we have failed in charge once in attach() |
| 6057 | * phase. |
| 6058 | */ |
Daisuke Nishimura | 854ffa8 | 2010-03-10 15:22:15 -0800 | [diff] [blame] | 6059 | ret = mem_cgroup_do_precharge(1); |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 6060 | if (!ret) |
| 6061 | goto retry; |
| 6062 | } |
| 6063 | |
| 6064 | return ret; |
| 6065 | } |
| 6066 | |
Christoph Hellwig | 7b86ac3 | 2019-08-28 16:19:54 +0200 | [diff] [blame] | 6067 | static const struct mm_walk_ops charge_walk_ops = { |
| 6068 | .pmd_entry = mem_cgroup_move_charge_pte_range, |
| 6069 | }; |
| 6070 | |
Tejun Heo | 264a0ae | 2016-04-21 19:09:02 -0400 | [diff] [blame] | 6071 | static void mem_cgroup_move_charge(void) |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 6072 | { |
Daisuke Nishimura | 4ffef5f | 2010-03-10 15:22:14 -0800 | [diff] [blame] | 6073 | lru_add_drain_all(); |
Johannes Weiner | 312722c | 2014-12-10 15:44:25 -0800 | [diff] [blame] | 6074 | /* |
Johannes Weiner | 81f8c3a | 2016-03-15 14:57:04 -0700 | [diff] [blame] | 6075 | * Signal lock_page_memcg() to take the memcg's move_lock |
| 6076 | * while we're moving its pages to another memcg. Then wait |
| 6077 | * for already started RCU-only updates to finish. |
Johannes Weiner | 312722c | 2014-12-10 15:44:25 -0800 | [diff] [blame] | 6078 | */ |
| 6079 | atomic_inc(&mc.from->moving_account); |
| 6080 | synchronize_rcu(); |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 6081 | retry: |
Michel Lespinasse | d8ed45c | 2020-06-08 21:33:25 -0700 | [diff] [blame] | 6082 | if (unlikely(!mmap_read_trylock(mc.mm))) { |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 6083 | /* |
Michel Lespinasse | c1e8d7c | 2020-06-08 21:33:54 -0700 | [diff] [blame] | 6084 | * Someone who are holding the mmap_lock might be waiting in |
Daisuke Nishimura | dfe076b | 2011-01-13 15:47:41 -0800 | [diff] [blame] | 6085 | * waitq. So we cancel all extra charges, wake up all waiters, |
| 6086 | * and retry. Because we cancel precharges, we might not be able |
| 6087 | * to move enough charges, but moving charge is a best-effort |
| 6088 | * feature anyway, so it wouldn't be a big problem. |
| 6089 | */ |
| 6090 | __mem_cgroup_clear_mc(); |
| 6091 | cond_resched(); |
| 6092 | goto retry; |
| 6093 | } |
Naoya Horiguchi | 26bcd64 | 2015-02-11 15:27:57 -0800 | [diff] [blame] | 6094 | /* |
| 6095 | * When we have consumed all precharges and failed in doing |
| 6096 | * additional charge, the page walk just aborts. |
| 6097 | */ |
Christoph Hellwig | 7b86ac3 | 2019-08-28 16:19:54 +0200 | [diff] [blame] | 6098 | walk_page_range(mc.mm, 0, mc.mm->highest_vm_end, &charge_walk_ops, |
| 6099 | NULL); |
James Morse | 0247f3f | 2016-10-07 17:00:12 -0700 | [diff] [blame] | 6100 | |
Michel Lespinasse | d8ed45c | 2020-06-08 21:33:25 -0700 | [diff] [blame] | 6101 | mmap_read_unlock(mc.mm); |
Johannes Weiner | 312722c | 2014-12-10 15:44:25 -0800 | [diff] [blame] | 6102 | atomic_dec(&mc.from->moving_account); |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 6103 | } |
| 6104 | |
Tejun Heo | 264a0ae | 2016-04-21 19:09:02 -0400 | [diff] [blame] | 6105 | static void mem_cgroup_move_task(void) |
Balbir Singh | 67e465a | 2008-02-07 00:13:54 -0800 | [diff] [blame] | 6106 | { |
Tejun Heo | 264a0ae | 2016-04-21 19:09:02 -0400 | [diff] [blame] | 6107 | if (mc.to) { |
| 6108 | mem_cgroup_move_charge(); |
KOSAKI Motohiro | a433658 | 2011-06-15 15:08:13 -0700 | [diff] [blame] | 6109 | mem_cgroup_clear_mc(); |
Tejun Heo | 264a0ae | 2016-04-21 19:09:02 -0400 | [diff] [blame] | 6110 | } |
Balbir Singh | 67e465a | 2008-02-07 00:13:54 -0800 | [diff] [blame] | 6111 | } |
Daisuke Nishimura | 5cfb80a | 2010-03-23 13:35:11 -0700 | [diff] [blame] | 6112 | #else /* !CONFIG_MMU */ |
Tejun Heo | 1f7dd3e5 | 2015-12-03 10:18:21 -0500 | [diff] [blame] | 6113 | static int mem_cgroup_can_attach(struct cgroup_taskset *tset) |
Daisuke Nishimura | 5cfb80a | 2010-03-23 13:35:11 -0700 | [diff] [blame] | 6114 | { |
| 6115 | return 0; |
| 6116 | } |
Tejun Heo | 1f7dd3e5 | 2015-12-03 10:18:21 -0500 | [diff] [blame] | 6117 | static void mem_cgroup_cancel_attach(struct cgroup_taskset *tset) |
Daisuke Nishimura | 5cfb80a | 2010-03-23 13:35:11 -0700 | [diff] [blame] | 6118 | { |
| 6119 | } |
Tejun Heo | 264a0ae | 2016-04-21 19:09:02 -0400 | [diff] [blame] | 6120 | static void mem_cgroup_move_task(void) |
Daisuke Nishimura | 5cfb80a | 2010-03-23 13:35:11 -0700 | [diff] [blame] | 6121 | { |
| 6122 | } |
| 6123 | #endif |
Balbir Singh | 67e465a | 2008-02-07 00:13:54 -0800 | [diff] [blame] | 6124 | |
Chris Down | 677dc97 | 2019-03-05 15:45:55 -0800 | [diff] [blame] | 6125 | static int seq_puts_memcg_tunable(struct seq_file *m, unsigned long value) |
| 6126 | { |
| 6127 | if (value == PAGE_COUNTER_MAX) |
| 6128 | seq_puts(m, "max\n"); |
| 6129 | else |
| 6130 | seq_printf(m, "%llu\n", (u64)value * PAGE_SIZE); |
| 6131 | |
| 6132 | return 0; |
| 6133 | } |
| 6134 | |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 6135 | static u64 memory_current_read(struct cgroup_subsys_state *css, |
| 6136 | struct cftype *cft) |
| 6137 | { |
Johannes Weiner | f5fc3c5d | 2015-11-05 18:50:23 -0800 | [diff] [blame] | 6138 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
| 6139 | |
| 6140 | return (u64)page_counter_read(&memcg->memory) * PAGE_SIZE; |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 6141 | } |
| 6142 | |
Roman Gushchin | bf8d5d5 | 2018-06-07 17:07:46 -0700 | [diff] [blame] | 6143 | static int memory_min_show(struct seq_file *m, void *v) |
| 6144 | { |
Chris Down | 677dc97 | 2019-03-05 15:45:55 -0800 | [diff] [blame] | 6145 | return seq_puts_memcg_tunable(m, |
| 6146 | READ_ONCE(mem_cgroup_from_seq(m)->memory.min)); |
Roman Gushchin | bf8d5d5 | 2018-06-07 17:07:46 -0700 | [diff] [blame] | 6147 | } |
| 6148 | |
| 6149 | static ssize_t memory_min_write(struct kernfs_open_file *of, |
| 6150 | char *buf, size_t nbytes, loff_t off) |
| 6151 | { |
| 6152 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
| 6153 | unsigned long min; |
| 6154 | int err; |
| 6155 | |
| 6156 | buf = strstrip(buf); |
| 6157 | err = page_counter_memparse(buf, "max", &min); |
| 6158 | if (err) |
| 6159 | return err; |
| 6160 | |
| 6161 | page_counter_set_min(&memcg->memory, min); |
| 6162 | |
| 6163 | return nbytes; |
| 6164 | } |
| 6165 | |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 6166 | static int memory_low_show(struct seq_file *m, void *v) |
| 6167 | { |
Chris Down | 677dc97 | 2019-03-05 15:45:55 -0800 | [diff] [blame] | 6168 | return seq_puts_memcg_tunable(m, |
| 6169 | READ_ONCE(mem_cgroup_from_seq(m)->memory.low)); |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 6170 | } |
| 6171 | |
| 6172 | static ssize_t memory_low_write(struct kernfs_open_file *of, |
| 6173 | char *buf, size_t nbytes, loff_t off) |
| 6174 | { |
| 6175 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
| 6176 | unsigned long low; |
| 6177 | int err; |
| 6178 | |
| 6179 | buf = strstrip(buf); |
Johannes Weiner | d297369 | 2015-02-27 15:52:04 -0800 | [diff] [blame] | 6180 | err = page_counter_memparse(buf, "max", &low); |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 6181 | if (err) |
| 6182 | return err; |
| 6183 | |
Roman Gushchin | 2306715 | 2018-06-07 17:06:22 -0700 | [diff] [blame] | 6184 | page_counter_set_low(&memcg->memory, low); |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 6185 | |
| 6186 | return nbytes; |
| 6187 | } |
| 6188 | |
| 6189 | static int memory_high_show(struct seq_file *m, void *v) |
| 6190 | { |
Jakub Kicinski | d1663a9 | 2020-06-01 21:49:49 -0700 | [diff] [blame] | 6191 | return seq_puts_memcg_tunable(m, |
| 6192 | READ_ONCE(mem_cgroup_from_seq(m)->memory.high)); |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 6193 | } |
| 6194 | |
| 6195 | static ssize_t memory_high_write(struct kernfs_open_file *of, |
| 6196 | char *buf, size_t nbytes, loff_t off) |
| 6197 | { |
| 6198 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
Chris Down | d977aa9 | 2020-08-06 23:21:58 -0700 | [diff] [blame] | 6199 | unsigned int nr_retries = MAX_RECLAIM_RETRIES; |
Johannes Weiner | 8c8c383 | 2019-11-30 17:50:09 -0800 | [diff] [blame] | 6200 | bool drained = false; |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 6201 | unsigned long high; |
| 6202 | int err; |
| 6203 | |
| 6204 | buf = strstrip(buf); |
Johannes Weiner | d297369 | 2015-02-27 15:52:04 -0800 | [diff] [blame] | 6205 | err = page_counter_memparse(buf, "max", &high); |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 6206 | if (err) |
| 6207 | return err; |
| 6208 | |
Johannes Weiner | e82553c | 2021-02-09 13:42:28 -0800 | [diff] [blame] | 6209 | page_counter_set_high(&memcg->memory, high); |
| 6210 | |
Johannes Weiner | 8c8c383 | 2019-11-30 17:50:09 -0800 | [diff] [blame] | 6211 | for (;;) { |
| 6212 | unsigned long nr_pages = page_counter_read(&memcg->memory); |
| 6213 | unsigned long reclaimed; |
Johannes Weiner | 588083b | 2016-03-17 14:20:25 -0700 | [diff] [blame] | 6214 | |
Johannes Weiner | 8c8c383 | 2019-11-30 17:50:09 -0800 | [diff] [blame] | 6215 | if (nr_pages <= high) |
| 6216 | break; |
| 6217 | |
| 6218 | if (signal_pending(current)) |
| 6219 | break; |
| 6220 | |
| 6221 | if (!drained) { |
| 6222 | drain_all_stock(memcg); |
| 6223 | drained = true; |
| 6224 | continue; |
| 6225 | } |
| 6226 | |
| 6227 | reclaimed = try_to_free_mem_cgroup_pages(memcg, nr_pages - high, |
| 6228 | GFP_KERNEL, true); |
| 6229 | |
| 6230 | if (!reclaimed && !nr_retries--) |
| 6231 | break; |
| 6232 | } |
| 6233 | |
Johannes Weiner | 19ce33a | 2020-08-06 23:22:12 -0700 | [diff] [blame] | 6234 | memcg_wb_domain_size_changed(memcg); |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 6235 | return nbytes; |
| 6236 | } |
| 6237 | |
| 6238 | static int memory_max_show(struct seq_file *m, void *v) |
| 6239 | { |
Chris Down | 677dc97 | 2019-03-05 15:45:55 -0800 | [diff] [blame] | 6240 | return seq_puts_memcg_tunable(m, |
| 6241 | READ_ONCE(mem_cgroup_from_seq(m)->memory.max)); |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 6242 | } |
| 6243 | |
| 6244 | static ssize_t memory_max_write(struct kernfs_open_file *of, |
| 6245 | char *buf, size_t nbytes, loff_t off) |
| 6246 | { |
| 6247 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
Chris Down | d977aa9 | 2020-08-06 23:21:58 -0700 | [diff] [blame] | 6248 | unsigned int nr_reclaims = MAX_RECLAIM_RETRIES; |
Johannes Weiner | b6e6edc | 2016-03-17 14:20:28 -0700 | [diff] [blame] | 6249 | bool drained = false; |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 6250 | unsigned long max; |
| 6251 | int err; |
| 6252 | |
| 6253 | buf = strstrip(buf); |
Johannes Weiner | d297369 | 2015-02-27 15:52:04 -0800 | [diff] [blame] | 6254 | err = page_counter_memparse(buf, "max", &max); |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 6255 | if (err) |
| 6256 | return err; |
| 6257 | |
Roman Gushchin | bbec2e1 | 2018-06-07 17:06:18 -0700 | [diff] [blame] | 6258 | xchg(&memcg->memory.max, max); |
Johannes Weiner | b6e6edc | 2016-03-17 14:20:28 -0700 | [diff] [blame] | 6259 | |
| 6260 | for (;;) { |
| 6261 | unsigned long nr_pages = page_counter_read(&memcg->memory); |
| 6262 | |
| 6263 | if (nr_pages <= max) |
| 6264 | break; |
| 6265 | |
Johannes Weiner | 7249c9f | 2019-11-30 17:50:06 -0800 | [diff] [blame] | 6266 | if (signal_pending(current)) |
Johannes Weiner | b6e6edc | 2016-03-17 14:20:28 -0700 | [diff] [blame] | 6267 | break; |
Johannes Weiner | b6e6edc | 2016-03-17 14:20:28 -0700 | [diff] [blame] | 6268 | |
| 6269 | if (!drained) { |
| 6270 | drain_all_stock(memcg); |
| 6271 | drained = true; |
| 6272 | continue; |
| 6273 | } |
| 6274 | |
| 6275 | if (nr_reclaims) { |
| 6276 | if (!try_to_free_mem_cgroup_pages(memcg, nr_pages - max, |
| 6277 | GFP_KERNEL, true)) |
| 6278 | nr_reclaims--; |
| 6279 | continue; |
| 6280 | } |
| 6281 | |
Johannes Weiner | e27be24 | 2018-04-10 16:29:45 -0700 | [diff] [blame] | 6282 | memcg_memory_event(memcg, MEMCG_OOM); |
Johannes Weiner | b6e6edc | 2016-03-17 14:20:28 -0700 | [diff] [blame] | 6283 | if (!mem_cgroup_out_of_memory(memcg, GFP_KERNEL, 0)) |
| 6284 | break; |
| 6285 | } |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 6286 | |
Tejun Heo | 2529bb3 | 2015-05-22 18:23:34 -0400 | [diff] [blame] | 6287 | memcg_wb_domain_size_changed(memcg); |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 6288 | return nbytes; |
| 6289 | } |
| 6290 | |
Shakeel Butt | 1e577f9 | 2019-07-11 20:55:55 -0700 | [diff] [blame] | 6291 | static void __memory_events_show(struct seq_file *m, atomic_long_t *events) |
| 6292 | { |
| 6293 | seq_printf(m, "low %lu\n", atomic_long_read(&events[MEMCG_LOW])); |
| 6294 | seq_printf(m, "high %lu\n", atomic_long_read(&events[MEMCG_HIGH])); |
| 6295 | seq_printf(m, "max %lu\n", atomic_long_read(&events[MEMCG_MAX])); |
| 6296 | seq_printf(m, "oom %lu\n", atomic_long_read(&events[MEMCG_OOM])); |
| 6297 | seq_printf(m, "oom_kill %lu\n", |
| 6298 | atomic_long_read(&events[MEMCG_OOM_KILL])); |
| 6299 | } |
| 6300 | |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 6301 | static int memory_events_show(struct seq_file *m, void *v) |
| 6302 | { |
Chris Down | aa9694b | 2019-03-05 15:45:52 -0800 | [diff] [blame] | 6303 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 6304 | |
Shakeel Butt | 1e577f9 | 2019-07-11 20:55:55 -0700 | [diff] [blame] | 6305 | __memory_events_show(m, memcg->memory_events); |
| 6306 | return 0; |
| 6307 | } |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 6308 | |
Shakeel Butt | 1e577f9 | 2019-07-11 20:55:55 -0700 | [diff] [blame] | 6309 | static int memory_events_local_show(struct seq_file *m, void *v) |
| 6310 | { |
| 6311 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
| 6312 | |
| 6313 | __memory_events_show(m, memcg->memory_events_local); |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 6314 | return 0; |
| 6315 | } |
| 6316 | |
Johannes Weiner | 587d9f7 | 2016-01-20 15:03:19 -0800 | [diff] [blame] | 6317 | static int memory_stat_show(struct seq_file *m, void *v) |
| 6318 | { |
Chris Down | aa9694b | 2019-03-05 15:45:52 -0800 | [diff] [blame] | 6319 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
Johannes Weiner | c8713d0 | 2019-07-11 20:55:59 -0700 | [diff] [blame] | 6320 | char *buf; |
Johannes Weiner | 587d9f7 | 2016-01-20 15:03:19 -0800 | [diff] [blame] | 6321 | |
Johannes Weiner | c8713d0 | 2019-07-11 20:55:59 -0700 | [diff] [blame] | 6322 | buf = memory_stat_format(memcg); |
| 6323 | if (!buf) |
| 6324 | return -ENOMEM; |
| 6325 | seq_puts(m, buf); |
| 6326 | kfree(buf); |
Johannes Weiner | 587d9f7 | 2016-01-20 15:03:19 -0800 | [diff] [blame] | 6327 | return 0; |
| 6328 | } |
| 6329 | |
Muchun Song | 5f9a4f4 | 2020-10-13 16:52:59 -0700 | [diff] [blame] | 6330 | #ifdef CONFIG_NUMA |
Muchun Song | fff66b7 | 2021-02-24 12:03:43 -0800 | [diff] [blame] | 6331 | static inline unsigned long lruvec_page_state_output(struct lruvec *lruvec, |
| 6332 | int item) |
| 6333 | { |
| 6334 | return lruvec_page_state(lruvec, item) * memcg_page_state_unit(item); |
| 6335 | } |
| 6336 | |
Muchun Song | 5f9a4f4 | 2020-10-13 16:52:59 -0700 | [diff] [blame] | 6337 | static int memory_numa_stat_show(struct seq_file *m, void *v) |
| 6338 | { |
| 6339 | int i; |
| 6340 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
| 6341 | |
| 6342 | for (i = 0; i < ARRAY_SIZE(memory_stats); i++) { |
| 6343 | int nid; |
| 6344 | |
| 6345 | if (memory_stats[i].idx >= NR_VM_NODE_STAT_ITEMS) |
| 6346 | continue; |
| 6347 | |
| 6348 | seq_printf(m, "%s", memory_stats[i].name); |
| 6349 | for_each_node_state(nid, N_MEMORY) { |
| 6350 | u64 size; |
| 6351 | struct lruvec *lruvec; |
| 6352 | |
| 6353 | lruvec = mem_cgroup_lruvec(memcg, NODE_DATA(nid)); |
Muchun Song | fff66b7 | 2021-02-24 12:03:43 -0800 | [diff] [blame] | 6354 | size = lruvec_page_state_output(lruvec, |
| 6355 | memory_stats[i].idx); |
Muchun Song | 5f9a4f4 | 2020-10-13 16:52:59 -0700 | [diff] [blame] | 6356 | seq_printf(m, " N%d=%llu", nid, size); |
| 6357 | } |
| 6358 | seq_putc(m, '\n'); |
| 6359 | } |
| 6360 | |
| 6361 | return 0; |
| 6362 | } |
| 6363 | #endif |
| 6364 | |
Roman Gushchin | 3d8b38e | 2018-08-21 21:53:54 -0700 | [diff] [blame] | 6365 | static int memory_oom_group_show(struct seq_file *m, void *v) |
| 6366 | { |
Chris Down | aa9694b | 2019-03-05 15:45:52 -0800 | [diff] [blame] | 6367 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
Roman Gushchin | 3d8b38e | 2018-08-21 21:53:54 -0700 | [diff] [blame] | 6368 | |
| 6369 | seq_printf(m, "%d\n", memcg->oom_group); |
| 6370 | |
| 6371 | return 0; |
| 6372 | } |
| 6373 | |
| 6374 | static ssize_t memory_oom_group_write(struct kernfs_open_file *of, |
| 6375 | char *buf, size_t nbytes, loff_t off) |
| 6376 | { |
| 6377 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
| 6378 | int ret, oom_group; |
| 6379 | |
| 6380 | buf = strstrip(buf); |
| 6381 | if (!buf) |
| 6382 | return -EINVAL; |
| 6383 | |
| 6384 | ret = kstrtoint(buf, 0, &oom_group); |
| 6385 | if (ret) |
| 6386 | return ret; |
| 6387 | |
| 6388 | if (oom_group != 0 && oom_group != 1) |
| 6389 | return -EINVAL; |
| 6390 | |
| 6391 | memcg->oom_group = oom_group; |
| 6392 | |
| 6393 | return nbytes; |
| 6394 | } |
| 6395 | |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 6396 | static struct cftype memory_files[] = { |
| 6397 | { |
| 6398 | .name = "current", |
Johannes Weiner | f5fc3c5d | 2015-11-05 18:50:23 -0800 | [diff] [blame] | 6399 | .flags = CFTYPE_NOT_ON_ROOT, |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 6400 | .read_u64 = memory_current_read, |
| 6401 | }, |
| 6402 | { |
Roman Gushchin | bf8d5d5 | 2018-06-07 17:07:46 -0700 | [diff] [blame] | 6403 | .name = "min", |
| 6404 | .flags = CFTYPE_NOT_ON_ROOT, |
| 6405 | .seq_show = memory_min_show, |
| 6406 | .write = memory_min_write, |
| 6407 | }, |
| 6408 | { |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 6409 | .name = "low", |
| 6410 | .flags = CFTYPE_NOT_ON_ROOT, |
| 6411 | .seq_show = memory_low_show, |
| 6412 | .write = memory_low_write, |
| 6413 | }, |
| 6414 | { |
| 6415 | .name = "high", |
| 6416 | .flags = CFTYPE_NOT_ON_ROOT, |
| 6417 | .seq_show = memory_high_show, |
| 6418 | .write = memory_high_write, |
| 6419 | }, |
| 6420 | { |
| 6421 | .name = "max", |
| 6422 | .flags = CFTYPE_NOT_ON_ROOT, |
| 6423 | .seq_show = memory_max_show, |
| 6424 | .write = memory_max_write, |
| 6425 | }, |
| 6426 | { |
| 6427 | .name = "events", |
| 6428 | .flags = CFTYPE_NOT_ON_ROOT, |
Tejun Heo | 472912a | 2015-09-18 18:01:59 -0400 | [diff] [blame] | 6429 | .file_offset = offsetof(struct mem_cgroup, events_file), |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 6430 | .seq_show = memory_events_show, |
| 6431 | }, |
Johannes Weiner | 587d9f7 | 2016-01-20 15:03:19 -0800 | [diff] [blame] | 6432 | { |
Shakeel Butt | 1e577f9 | 2019-07-11 20:55:55 -0700 | [diff] [blame] | 6433 | .name = "events.local", |
| 6434 | .flags = CFTYPE_NOT_ON_ROOT, |
| 6435 | .file_offset = offsetof(struct mem_cgroup, events_local_file), |
| 6436 | .seq_show = memory_events_local_show, |
| 6437 | }, |
| 6438 | { |
Johannes Weiner | 587d9f7 | 2016-01-20 15:03:19 -0800 | [diff] [blame] | 6439 | .name = "stat", |
Johannes Weiner | 587d9f7 | 2016-01-20 15:03:19 -0800 | [diff] [blame] | 6440 | .seq_show = memory_stat_show, |
| 6441 | }, |
Muchun Song | 5f9a4f4 | 2020-10-13 16:52:59 -0700 | [diff] [blame] | 6442 | #ifdef CONFIG_NUMA |
| 6443 | { |
| 6444 | .name = "numa_stat", |
| 6445 | .seq_show = memory_numa_stat_show, |
| 6446 | }, |
| 6447 | #endif |
Roman Gushchin | 3d8b38e | 2018-08-21 21:53:54 -0700 | [diff] [blame] | 6448 | { |
| 6449 | .name = "oom.group", |
| 6450 | .flags = CFTYPE_NOT_ON_ROOT | CFTYPE_NS_DELEGATABLE, |
| 6451 | .seq_show = memory_oom_group_show, |
| 6452 | .write = memory_oom_group_write, |
| 6453 | }, |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 6454 | { } /* terminate */ |
| 6455 | }; |
| 6456 | |
Tejun Heo | 073219e | 2014-02-08 10:36:58 -0500 | [diff] [blame] | 6457 | struct cgroup_subsys memory_cgrp_subsys = { |
Tejun Heo | 92fb974 | 2012-11-19 08:13:38 -0800 | [diff] [blame] | 6458 | .css_alloc = mem_cgroup_css_alloc, |
Glauber Costa | d142e3e | 2013-02-22 16:34:52 -0800 | [diff] [blame] | 6459 | .css_online = mem_cgroup_css_online, |
Tejun Heo | 92fb974 | 2012-11-19 08:13:38 -0800 | [diff] [blame] | 6460 | .css_offline = mem_cgroup_css_offline, |
Vladimir Davydov | 6df3868 | 2015-12-29 14:54:10 -0800 | [diff] [blame] | 6461 | .css_released = mem_cgroup_css_released, |
Tejun Heo | 92fb974 | 2012-11-19 08:13:38 -0800 | [diff] [blame] | 6462 | .css_free = mem_cgroup_css_free, |
Tejun Heo | 1ced953 | 2014-07-08 18:02:57 -0400 | [diff] [blame] | 6463 | .css_reset = mem_cgroup_css_reset, |
Johannes Weiner | 2d146aa | 2021-04-29 22:56:26 -0700 | [diff] [blame] | 6464 | .css_rstat_flush = mem_cgroup_css_rstat_flush, |
Daisuke Nishimura | 7dc74be | 2010-03-10 15:22:13 -0800 | [diff] [blame] | 6465 | .can_attach = mem_cgroup_can_attach, |
| 6466 | .cancel_attach = mem_cgroup_cancel_attach, |
Tejun Heo | 264a0ae | 2016-04-21 19:09:02 -0400 | [diff] [blame] | 6467 | .post_attach = mem_cgroup_move_task, |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 6468 | .dfl_cftypes = memory_files, |
| 6469 | .legacy_cftypes = mem_cgroup_legacy_files, |
KAMEZAWA Hiroyuki | 6d12e2d | 2008-02-07 00:14:31 -0800 | [diff] [blame] | 6470 | .early_init = 0, |
Balbir Singh | 8cdea7c | 2008-02-07 00:13:50 -0800 | [diff] [blame] | 6471 | }; |
KAMEZAWA Hiroyuki | c077719 | 2009-01-07 18:07:57 -0800 | [diff] [blame] | 6472 | |
Johannes Weiner | bc50bcc | 2020-04-01 21:07:03 -0700 | [diff] [blame] | 6473 | /* |
| 6474 | * This function calculates an individual cgroup's effective |
| 6475 | * protection which is derived from its own memory.min/low, its |
| 6476 | * parent's and siblings' settings, as well as the actual memory |
| 6477 | * distribution in the tree. |
| 6478 | * |
| 6479 | * The following rules apply to the effective protection values: |
| 6480 | * |
| 6481 | * 1. At the first level of reclaim, effective protection is equal to |
| 6482 | * the declared protection in memory.min and memory.low. |
| 6483 | * |
| 6484 | * 2. To enable safe delegation of the protection configuration, at |
| 6485 | * subsequent levels the effective protection is capped to the |
| 6486 | * parent's effective protection. |
| 6487 | * |
| 6488 | * 3. To make complex and dynamic subtrees easier to configure, the |
| 6489 | * user is allowed to overcommit the declared protection at a given |
| 6490 | * level. If that is the case, the parent's effective protection is |
| 6491 | * distributed to the children in proportion to how much protection |
| 6492 | * they have declared and how much of it they are utilizing. |
| 6493 | * |
| 6494 | * This makes distribution proportional, but also work-conserving: |
| 6495 | * if one cgroup claims much more protection than it uses memory, |
| 6496 | * the unused remainder is available to its siblings. |
| 6497 | * |
| 6498 | * 4. Conversely, when the declared protection is undercommitted at a |
| 6499 | * given level, the distribution of the larger parental protection |
| 6500 | * budget is NOT proportional. A cgroup's protection from a sibling |
| 6501 | * is capped to its own memory.min/low setting. |
| 6502 | * |
Johannes Weiner | 8a931f8 | 2020-04-01 21:07:07 -0700 | [diff] [blame] | 6503 | * 5. However, to allow protecting recursive subtrees from each other |
| 6504 | * without having to declare each individual cgroup's fixed share |
| 6505 | * of the ancestor's claim to protection, any unutilized - |
| 6506 | * "floating" - protection from up the tree is distributed in |
| 6507 | * proportion to each cgroup's *usage*. This makes the protection |
| 6508 | * neutral wrt sibling cgroups and lets them compete freely over |
| 6509 | * the shared parental protection budget, but it protects the |
| 6510 | * subtree as a whole from neighboring subtrees. |
| 6511 | * |
| 6512 | * Note that 4. and 5. are not in conflict: 4. is about protecting |
| 6513 | * against immediate siblings whereas 5. is about protecting against |
| 6514 | * neighboring subtrees. |
Johannes Weiner | bc50bcc | 2020-04-01 21:07:03 -0700 | [diff] [blame] | 6515 | */ |
| 6516 | static unsigned long effective_protection(unsigned long usage, |
Johannes Weiner | 8a931f8 | 2020-04-01 21:07:07 -0700 | [diff] [blame] | 6517 | unsigned long parent_usage, |
Johannes Weiner | bc50bcc | 2020-04-01 21:07:03 -0700 | [diff] [blame] | 6518 | unsigned long setting, |
| 6519 | unsigned long parent_effective, |
| 6520 | unsigned long siblings_protected) |
| 6521 | { |
| 6522 | unsigned long protected; |
Johannes Weiner | 8a931f8 | 2020-04-01 21:07:07 -0700 | [diff] [blame] | 6523 | unsigned long ep; |
Johannes Weiner | bc50bcc | 2020-04-01 21:07:03 -0700 | [diff] [blame] | 6524 | |
| 6525 | protected = min(usage, setting); |
| 6526 | /* |
| 6527 | * If all cgroups at this level combined claim and use more |
| 6528 | * protection then what the parent affords them, distribute |
| 6529 | * shares in proportion to utilization. |
| 6530 | * |
| 6531 | * We are using actual utilization rather than the statically |
| 6532 | * claimed protection in order to be work-conserving: claimed |
| 6533 | * but unused protection is available to siblings that would |
| 6534 | * otherwise get a smaller chunk than what they claimed. |
| 6535 | */ |
| 6536 | if (siblings_protected > parent_effective) |
| 6537 | return protected * parent_effective / siblings_protected; |
| 6538 | |
| 6539 | /* |
| 6540 | * Ok, utilized protection of all children is within what the |
| 6541 | * parent affords them, so we know whatever this child claims |
| 6542 | * and utilizes is effectively protected. |
| 6543 | * |
| 6544 | * If there is unprotected usage beyond this value, reclaim |
| 6545 | * will apply pressure in proportion to that amount. |
| 6546 | * |
| 6547 | * If there is unutilized protection, the cgroup will be fully |
| 6548 | * shielded from reclaim, but we do return a smaller value for |
| 6549 | * protection than what the group could enjoy in theory. This |
| 6550 | * is okay. With the overcommit distribution above, effective |
| 6551 | * protection is always dependent on how memory is actually |
| 6552 | * consumed among the siblings anyway. |
| 6553 | */ |
Johannes Weiner | 8a931f8 | 2020-04-01 21:07:07 -0700 | [diff] [blame] | 6554 | ep = protected; |
| 6555 | |
| 6556 | /* |
| 6557 | * If the children aren't claiming (all of) the protection |
| 6558 | * afforded to them by the parent, distribute the remainder in |
| 6559 | * proportion to the (unprotected) memory of each cgroup. That |
| 6560 | * way, cgroups that aren't explicitly prioritized wrt each |
| 6561 | * other compete freely over the allowance, but they are |
| 6562 | * collectively protected from neighboring trees. |
| 6563 | * |
| 6564 | * We're using unprotected memory for the weight so that if |
| 6565 | * some cgroups DO claim explicit protection, we don't protect |
| 6566 | * the same bytes twice. |
Johannes Weiner | cd324ed | 2020-06-25 20:30:16 -0700 | [diff] [blame] | 6567 | * |
| 6568 | * Check both usage and parent_usage against the respective |
| 6569 | * protected values. One should imply the other, but they |
| 6570 | * aren't read atomically - make sure the division is sane. |
Johannes Weiner | 8a931f8 | 2020-04-01 21:07:07 -0700 | [diff] [blame] | 6571 | */ |
| 6572 | if (!(cgrp_dfl_root.flags & CGRP_ROOT_MEMORY_RECURSIVE_PROT)) |
| 6573 | return ep; |
Johannes Weiner | cd324ed | 2020-06-25 20:30:16 -0700 | [diff] [blame] | 6574 | if (parent_effective > siblings_protected && |
| 6575 | parent_usage > siblings_protected && |
| 6576 | usage > protected) { |
Johannes Weiner | 8a931f8 | 2020-04-01 21:07:07 -0700 | [diff] [blame] | 6577 | unsigned long unclaimed; |
| 6578 | |
| 6579 | unclaimed = parent_effective - siblings_protected; |
| 6580 | unclaimed *= usage - protected; |
| 6581 | unclaimed /= parent_usage - siblings_protected; |
| 6582 | |
| 6583 | ep += unclaimed; |
| 6584 | } |
| 6585 | |
| 6586 | return ep; |
Johannes Weiner | bc50bcc | 2020-04-01 21:07:03 -0700 | [diff] [blame] | 6587 | } |
| 6588 | |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 6589 | /** |
Roman Gushchin | bf8d5d5 | 2018-06-07 17:07:46 -0700 | [diff] [blame] | 6590 | * mem_cgroup_protected - check if memory consumption is in the normal range |
Sean Christopherson | 34c8105 | 2017-07-10 15:48:05 -0700 | [diff] [blame] | 6591 | * @root: the top ancestor of the sub-tree being checked |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 6592 | * @memcg: the memory cgroup to check |
| 6593 | * |
Roman Gushchin | 2306715 | 2018-06-07 17:06:22 -0700 | [diff] [blame] | 6594 | * WARNING: This function is not stateless! It can only be used as part |
| 6595 | * of a top-down tree iteration, not for isolated queries. |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 6596 | */ |
Chris Down | 45c7f7e | 2020-08-06 23:22:05 -0700 | [diff] [blame] | 6597 | void mem_cgroup_calculate_protection(struct mem_cgroup *root, |
| 6598 | struct mem_cgroup *memcg) |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 6599 | { |
Johannes Weiner | 8a931f8 | 2020-04-01 21:07:07 -0700 | [diff] [blame] | 6600 | unsigned long usage, parent_usage; |
Roman Gushchin | 2306715 | 2018-06-07 17:06:22 -0700 | [diff] [blame] | 6601 | struct mem_cgroup *parent; |
| 6602 | |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 6603 | if (mem_cgroup_disabled()) |
Chris Down | 45c7f7e | 2020-08-06 23:22:05 -0700 | [diff] [blame] | 6604 | return; |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 6605 | |
Sean Christopherson | 34c8105 | 2017-07-10 15:48:05 -0700 | [diff] [blame] | 6606 | if (!root) |
| 6607 | root = root_mem_cgroup; |
Yafang Shao | 22f7496 | 2020-08-06 23:22:01 -0700 | [diff] [blame] | 6608 | |
| 6609 | /* |
| 6610 | * Effective values of the reclaim targets are ignored so they |
| 6611 | * can be stale. Have a look at mem_cgroup_protection for more |
| 6612 | * details. |
| 6613 | * TODO: calculation should be more robust so that we do not need |
| 6614 | * that special casing. |
| 6615 | */ |
Sean Christopherson | 34c8105 | 2017-07-10 15:48:05 -0700 | [diff] [blame] | 6616 | if (memcg == root) |
Chris Down | 45c7f7e | 2020-08-06 23:22:05 -0700 | [diff] [blame] | 6617 | return; |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 6618 | |
Roman Gushchin | 2306715 | 2018-06-07 17:06:22 -0700 | [diff] [blame] | 6619 | usage = page_counter_read(&memcg->memory); |
Roman Gushchin | bf8d5d5 | 2018-06-07 17:07:46 -0700 | [diff] [blame] | 6620 | if (!usage) |
Chris Down | 45c7f7e | 2020-08-06 23:22:05 -0700 | [diff] [blame] | 6621 | return; |
Sean Christopherson | 34c8105 | 2017-07-10 15:48:05 -0700 | [diff] [blame] | 6622 | |
Roman Gushchin | bf8d5d5 | 2018-06-07 17:07:46 -0700 | [diff] [blame] | 6623 | parent = parent_mem_cgroup(memcg); |
Roman Gushchin | df2a419 | 2018-06-14 15:26:17 -0700 | [diff] [blame] | 6624 | /* No parent means a non-hierarchical mode on v1 memcg */ |
| 6625 | if (!parent) |
Chris Down | 45c7f7e | 2020-08-06 23:22:05 -0700 | [diff] [blame] | 6626 | return; |
Roman Gushchin | df2a419 | 2018-06-14 15:26:17 -0700 | [diff] [blame] | 6627 | |
Johannes Weiner | bc50bcc | 2020-04-01 21:07:03 -0700 | [diff] [blame] | 6628 | if (parent == root) { |
Chris Down | c3d5320 | 2020-04-01 21:07:27 -0700 | [diff] [blame] | 6629 | memcg->memory.emin = READ_ONCE(memcg->memory.min); |
Chris Down | 03960e3 | 2020-06-25 20:30:22 -0700 | [diff] [blame] | 6630 | memcg->memory.elow = READ_ONCE(memcg->memory.low); |
Chris Down | 45c7f7e | 2020-08-06 23:22:05 -0700 | [diff] [blame] | 6631 | return; |
Roman Gushchin | bf8d5d5 | 2018-06-07 17:07:46 -0700 | [diff] [blame] | 6632 | } |
| 6633 | |
Johannes Weiner | 8a931f8 | 2020-04-01 21:07:07 -0700 | [diff] [blame] | 6634 | parent_usage = page_counter_read(&parent->memory); |
| 6635 | |
Chris Down | b3a7822 | 2020-04-01 21:07:33 -0700 | [diff] [blame] | 6636 | WRITE_ONCE(memcg->memory.emin, effective_protection(usage, parent_usage, |
Chris Down | c3d5320 | 2020-04-01 21:07:27 -0700 | [diff] [blame] | 6637 | READ_ONCE(memcg->memory.min), |
| 6638 | READ_ONCE(parent->memory.emin), |
Chris Down | b3a7822 | 2020-04-01 21:07:33 -0700 | [diff] [blame] | 6639 | atomic_long_read(&parent->memory.children_min_usage))); |
Roman Gushchin | 2306715 | 2018-06-07 17:06:22 -0700 | [diff] [blame] | 6640 | |
Chris Down | b3a7822 | 2020-04-01 21:07:33 -0700 | [diff] [blame] | 6641 | WRITE_ONCE(memcg->memory.elow, effective_protection(usage, parent_usage, |
Chris Down | 03960e3 | 2020-06-25 20:30:22 -0700 | [diff] [blame] | 6642 | READ_ONCE(memcg->memory.low), |
| 6643 | READ_ONCE(parent->memory.elow), |
Chris Down | b3a7822 | 2020-04-01 21:07:33 -0700 | [diff] [blame] | 6644 | atomic_long_read(&parent->memory.children_low_usage))); |
Johannes Weiner | 241994ed | 2015-02-11 15:26:06 -0800 | [diff] [blame] | 6645 | } |
| 6646 | |
Shakeel Butt | 0add0c7 | 2021-04-29 22:56:36 -0700 | [diff] [blame^] | 6647 | static int __mem_cgroup_charge(struct page *page, struct mem_cgroup *memcg, |
| 6648 | gfp_t gfp) |
| 6649 | { |
| 6650 | unsigned int nr_pages = thp_nr_pages(page); |
| 6651 | int ret; |
| 6652 | |
| 6653 | ret = try_charge(memcg, gfp, nr_pages); |
| 6654 | if (ret) |
| 6655 | goto out; |
| 6656 | |
| 6657 | css_get(&memcg->css); |
| 6658 | commit_charge(page, memcg); |
| 6659 | |
| 6660 | local_irq_disable(); |
| 6661 | mem_cgroup_charge_statistics(memcg, page, nr_pages); |
| 6662 | memcg_check_events(memcg, page); |
| 6663 | local_irq_enable(); |
| 6664 | out: |
| 6665 | return ret; |
| 6666 | } |
| 6667 | |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 6668 | /** |
Johannes Weiner | f0e45fb | 2020-06-03 16:02:07 -0700 | [diff] [blame] | 6669 | * mem_cgroup_charge - charge a newly allocated page to a cgroup |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 6670 | * @page: page to charge |
| 6671 | * @mm: mm context of the victim |
| 6672 | * @gfp_mask: reclaim mode |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 6673 | * |
| 6674 | * Try to charge @page to the memcg that @mm belongs to, reclaiming |
| 6675 | * pages according to @gfp_mask if necessary. |
| 6676 | * |
Shakeel Butt | 0add0c7 | 2021-04-29 22:56:36 -0700 | [diff] [blame^] | 6677 | * Do not use this for pages allocated for swapin. |
| 6678 | * |
Johannes Weiner | f0e45fb | 2020-06-03 16:02:07 -0700 | [diff] [blame] | 6679 | * Returns 0 on success. Otherwise, an error code is returned. |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 6680 | */ |
Johannes Weiner | d9eb1ea | 2020-06-03 16:02:24 -0700 | [diff] [blame] | 6681 | int mem_cgroup_charge(struct page *page, struct mm_struct *mm, gfp_t gfp_mask) |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 6682 | { |
Shakeel Butt | 0add0c7 | 2021-04-29 22:56:36 -0700 | [diff] [blame^] | 6683 | struct mem_cgroup *memcg; |
| 6684 | int ret; |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 6685 | |
| 6686 | if (mem_cgroup_disabled()) |
Shakeel Butt | 0add0c7 | 2021-04-29 22:56:36 -0700 | [diff] [blame^] | 6687 | return 0; |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 6688 | |
Shakeel Butt | 0add0c7 | 2021-04-29 22:56:36 -0700 | [diff] [blame^] | 6689 | memcg = get_mem_cgroup_from_mm(mm); |
| 6690 | ret = __mem_cgroup_charge(page, memcg, gfp_mask); |
| 6691 | css_put(&memcg->css); |
Johannes Weiner | 2d1c498 | 2020-06-03 16:02:14 -0700 | [diff] [blame] | 6692 | |
Shakeel Butt | 0add0c7 | 2021-04-29 22:56:36 -0700 | [diff] [blame^] | 6693 | return ret; |
| 6694 | } |
Vladimir Davydov | e993d90 | 2015-09-09 15:35:35 -0700 | [diff] [blame] | 6695 | |
Shakeel Butt | 0add0c7 | 2021-04-29 22:56:36 -0700 | [diff] [blame^] | 6696 | /** |
| 6697 | * mem_cgroup_swapin_charge_page - charge a newly allocated page for swapin |
| 6698 | * @page: page to charge |
| 6699 | * @mm: mm context of the victim |
| 6700 | * @gfp: reclaim mode |
| 6701 | * @entry: swap entry for which the page is allocated |
| 6702 | * |
| 6703 | * This function charges a page allocated for swapin. Please call this before |
| 6704 | * adding the page to the swapcache. |
| 6705 | * |
| 6706 | * Returns 0 on success. Otherwise, an error code is returned. |
| 6707 | */ |
| 6708 | int mem_cgroup_swapin_charge_page(struct page *page, struct mm_struct *mm, |
| 6709 | gfp_t gfp, swp_entry_t entry) |
| 6710 | { |
| 6711 | struct mem_cgroup *memcg; |
| 6712 | unsigned short id; |
| 6713 | int ret; |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 6714 | |
Shakeel Butt | 0add0c7 | 2021-04-29 22:56:36 -0700 | [diff] [blame^] | 6715 | if (mem_cgroup_disabled()) |
| 6716 | return 0; |
| 6717 | |
| 6718 | id = lookup_swap_cgroup_id(entry); |
| 6719 | rcu_read_lock(); |
| 6720 | memcg = mem_cgroup_from_id(id); |
| 6721 | if (!memcg || !css_tryget_online(&memcg->css)) |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 6722 | memcg = get_mem_cgroup_from_mm(mm); |
Shakeel Butt | 0add0c7 | 2021-04-29 22:56:36 -0700 | [diff] [blame^] | 6723 | rcu_read_unlock(); |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 6724 | |
Shakeel Butt | 0add0c7 | 2021-04-29 22:56:36 -0700 | [diff] [blame^] | 6725 | ret = __mem_cgroup_charge(page, memcg, gfp); |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 6726 | |
Shakeel Butt | 0add0c7 | 2021-04-29 22:56:36 -0700 | [diff] [blame^] | 6727 | css_put(&memcg->css); |
| 6728 | return ret; |
| 6729 | } |
Johannes Weiner | 6abb5a8 | 2014-08-08 14:19:33 -0700 | [diff] [blame] | 6730 | |
Shakeel Butt | 0add0c7 | 2021-04-29 22:56:36 -0700 | [diff] [blame^] | 6731 | /* |
| 6732 | * mem_cgroup_swapin_uncharge_swap - uncharge swap slot |
| 6733 | * @entry: swap entry for which the page is charged |
| 6734 | * |
| 6735 | * Call this function after successfully adding the charged page to swapcache. |
| 6736 | * |
| 6737 | * Note: This function assumes the page for which swap slot is being uncharged |
| 6738 | * is order 0 page. |
| 6739 | */ |
| 6740 | void mem_cgroup_swapin_uncharge_swap(swp_entry_t entry) |
| 6741 | { |
Muchun Song | cae3af6 | 2021-02-24 12:04:19 -0800 | [diff] [blame] | 6742 | /* |
| 6743 | * Cgroup1's unified memory+swap counter has been charged with the |
| 6744 | * new swapcache page, finish the transfer by uncharging the swap |
| 6745 | * slot. The swap slot would also get uncharged when it dies, but |
| 6746 | * it can stick around indefinitely and we'd count the page twice |
| 6747 | * the entire time. |
| 6748 | * |
| 6749 | * Cgroup2 has separate resource counters for memory and swap, |
| 6750 | * so this is a non-issue here. Memory and swap charge lifetimes |
| 6751 | * correspond 1:1 to page and swap slot lifetimes: we charge the |
| 6752 | * page to memory here, and uncharge swap when the slot is freed. |
| 6753 | */ |
Shakeel Butt | 0add0c7 | 2021-04-29 22:56:36 -0700 | [diff] [blame^] | 6754 | if (!mem_cgroup_disabled() && do_memsw_account()) { |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 6755 | /* |
| 6756 | * The swap entry might not get freed for a long time, |
| 6757 | * let's not wait for it. The page already received a |
| 6758 | * memory+swap charge, drop the swap entry duplicate. |
| 6759 | */ |
Shakeel Butt | 0add0c7 | 2021-04-29 22:56:36 -0700 | [diff] [blame^] | 6760 | mem_cgroup_uncharge_swap(entry, 1); |
Johannes Weiner | 00501b5 | 2014-08-08 14:19:20 -0700 | [diff] [blame] | 6761 | } |
Johannes Weiner | 3fea5a4 | 2020-06-03 16:01:41 -0700 | [diff] [blame] | 6762 | } |
| 6763 | |
Jérôme Glisse | a9d5ade | 2017-09-08 16:11:50 -0700 | [diff] [blame] | 6764 | struct uncharge_gather { |
| 6765 | struct mem_cgroup *memcg; |
Johannes Weiner | 9f762db | 2020-06-03 16:01:44 -0700 | [diff] [blame] | 6766 | unsigned long nr_pages; |
Jérôme Glisse | a9d5ade | 2017-09-08 16:11:50 -0700 | [diff] [blame] | 6767 | unsigned long pgpgout; |
Jérôme Glisse | a9d5ade | 2017-09-08 16:11:50 -0700 | [diff] [blame] | 6768 | unsigned long nr_kmem; |
Jérôme Glisse | a9d5ade | 2017-09-08 16:11:50 -0700 | [diff] [blame] | 6769 | struct page *dummy_page; |
| 6770 | }; |
| 6771 | |
| 6772 | static inline void uncharge_gather_clear(struct uncharge_gather *ug) |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 6773 | { |
Jérôme Glisse | a9d5ade | 2017-09-08 16:11:50 -0700 | [diff] [blame] | 6774 | memset(ug, 0, sizeof(*ug)); |
| 6775 | } |
| 6776 | |
| 6777 | static void uncharge_batch(const struct uncharge_gather *ug) |
| 6778 | { |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 6779 | unsigned long flags; |
| 6780 | |
Jérôme Glisse | a9d5ade | 2017-09-08 16:11:50 -0700 | [diff] [blame] | 6781 | if (!mem_cgroup_is_root(ug->memcg)) { |
Johannes Weiner | 9f762db | 2020-06-03 16:01:44 -0700 | [diff] [blame] | 6782 | page_counter_uncharge(&ug->memcg->memory, ug->nr_pages); |
Johannes Weiner | 7941d21 | 2016-01-14 15:21:23 -0800 | [diff] [blame] | 6783 | if (do_memsw_account()) |
Johannes Weiner | 9f762db | 2020-06-03 16:01:44 -0700 | [diff] [blame] | 6784 | page_counter_uncharge(&ug->memcg->memsw, ug->nr_pages); |
Jérôme Glisse | a9d5ade | 2017-09-08 16:11:50 -0700 | [diff] [blame] | 6785 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && ug->nr_kmem) |
| 6786 | page_counter_uncharge(&ug->memcg->kmem, ug->nr_kmem); |
| 6787 | memcg_oom_recover(ug->memcg); |
Johannes Weiner | ce00a96 | 2014-09-05 08:43:57 -0400 | [diff] [blame] | 6788 | } |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 6789 | |
| 6790 | local_irq_save(flags); |
Johannes Weiner | c9019e9 | 2018-01-31 16:16:37 -0800 | [diff] [blame] | 6791 | __count_memcg_events(ug->memcg, PGPGOUT, ug->pgpgout); |
Johannes Weiner | 9f762db | 2020-06-03 16:01:44 -0700 | [diff] [blame] | 6792 | __this_cpu_add(ug->memcg->vmstats_percpu->nr_page_events, ug->nr_pages); |
Jérôme Glisse | a9d5ade | 2017-09-08 16:11:50 -0700 | [diff] [blame] | 6793 | memcg_check_events(ug->memcg, ug->dummy_page); |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 6794 | local_irq_restore(flags); |
Michal Hocko | f179654 | 2020-09-04 16:35:24 -0700 | [diff] [blame] | 6795 | |
| 6796 | /* drop reference from uncharge_page */ |
| 6797 | css_put(&ug->memcg->css); |
Jérôme Glisse | a9d5ade | 2017-09-08 16:11:50 -0700 | [diff] [blame] | 6798 | } |
| 6799 | |
| 6800 | static void uncharge_page(struct page *page, struct uncharge_gather *ug) |
| 6801 | { |
Johannes Weiner | 9f762db | 2020-06-03 16:01:44 -0700 | [diff] [blame] | 6802 | unsigned long nr_pages; |
| 6803 | |
Jérôme Glisse | a9d5ade | 2017-09-08 16:11:50 -0700 | [diff] [blame] | 6804 | VM_BUG_ON_PAGE(PageLRU(page), page); |
Jérôme Glisse | a9d5ade | 2017-09-08 16:11:50 -0700 | [diff] [blame] | 6805 | |
Roman Gushchin | bcfe06b | 2020-12-01 13:58:27 -0800 | [diff] [blame] | 6806 | if (!page_memcg(page)) |
Jérôme Glisse | a9d5ade | 2017-09-08 16:11:50 -0700 | [diff] [blame] | 6807 | return; |
| 6808 | |
| 6809 | /* |
| 6810 | * Nobody should be changing or seriously looking at |
Roman Gushchin | bcfe06b | 2020-12-01 13:58:27 -0800 | [diff] [blame] | 6811 | * page_memcg(page) at this point, we have fully |
Jérôme Glisse | a9d5ade | 2017-09-08 16:11:50 -0700 | [diff] [blame] | 6812 | * exclusive access to the page. |
| 6813 | */ |
| 6814 | |
Roman Gushchin | bcfe06b | 2020-12-01 13:58:27 -0800 | [diff] [blame] | 6815 | if (ug->memcg != page_memcg(page)) { |
Jérôme Glisse | a9d5ade | 2017-09-08 16:11:50 -0700 | [diff] [blame] | 6816 | if (ug->memcg) { |
| 6817 | uncharge_batch(ug); |
| 6818 | uncharge_gather_clear(ug); |
| 6819 | } |
Roman Gushchin | bcfe06b | 2020-12-01 13:58:27 -0800 | [diff] [blame] | 6820 | ug->memcg = page_memcg(page); |
Michal Hocko | f179654 | 2020-09-04 16:35:24 -0700 | [diff] [blame] | 6821 | |
| 6822 | /* pairs with css_put in uncharge_batch */ |
| 6823 | css_get(&ug->memcg->css); |
Jérôme Glisse | a9d5ade | 2017-09-08 16:11:50 -0700 | [diff] [blame] | 6824 | } |
| 6825 | |
Johannes Weiner | 9f762db | 2020-06-03 16:01:44 -0700 | [diff] [blame] | 6826 | nr_pages = compound_nr(page); |
| 6827 | ug->nr_pages += nr_pages; |
Jérôme Glisse | a9d5ade | 2017-09-08 16:11:50 -0700 | [diff] [blame] | 6828 | |
Roman Gushchin | 18b2db3 | 2020-12-01 13:58:30 -0800 | [diff] [blame] | 6829 | if (PageMemcgKmem(page)) |
Johannes Weiner | 9f762db | 2020-06-03 16:01:44 -0700 | [diff] [blame] | 6830 | ug->nr_kmem += nr_pages; |
Roman Gushchin | 18b2db3 | 2020-12-01 13:58:30 -0800 | [diff] [blame] | 6831 | else |
| 6832 | ug->pgpgout++; |
Jérôme Glisse | a9d5ade | 2017-09-08 16:11:50 -0700 | [diff] [blame] | 6833 | |
| 6834 | ug->dummy_page = page; |
Roman Gushchin | bcfe06b | 2020-12-01 13:58:27 -0800 | [diff] [blame] | 6835 | page->memcg_data = 0; |
Johannes Weiner | 1a3e1f4 | 2020-08-06 23:20:45 -0700 | [diff] [blame] | 6836 | css_put(&ug->memcg->css); |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 6837 | } |
| 6838 | |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 6839 | /** |
| 6840 | * mem_cgroup_uncharge - uncharge a page |
| 6841 | * @page: page to uncharge |
| 6842 | * |
Johannes Weiner | f0e45fb | 2020-06-03 16:02:07 -0700 | [diff] [blame] | 6843 | * Uncharge a page previously charged with mem_cgroup_charge(). |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 6844 | */ |
| 6845 | void mem_cgroup_uncharge(struct page *page) |
| 6846 | { |
Jérôme Glisse | a9d5ade | 2017-09-08 16:11:50 -0700 | [diff] [blame] | 6847 | struct uncharge_gather ug; |
| 6848 | |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 6849 | if (mem_cgroup_disabled()) |
| 6850 | return; |
| 6851 | |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 6852 | /* Don't touch page->lru of any random page, pre-check: */ |
Roman Gushchin | bcfe06b | 2020-12-01 13:58:27 -0800 | [diff] [blame] | 6853 | if (!page_memcg(page)) |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 6854 | return; |
| 6855 | |
Jérôme Glisse | a9d5ade | 2017-09-08 16:11:50 -0700 | [diff] [blame] | 6856 | uncharge_gather_clear(&ug); |
| 6857 | uncharge_page(page, &ug); |
| 6858 | uncharge_batch(&ug); |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 6859 | } |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 6860 | |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 6861 | /** |
| 6862 | * mem_cgroup_uncharge_list - uncharge a list of page |
| 6863 | * @page_list: list of pages to uncharge |
| 6864 | * |
| 6865 | * Uncharge a list of pages previously charged with |
Johannes Weiner | f0e45fb | 2020-06-03 16:02:07 -0700 | [diff] [blame] | 6866 | * mem_cgroup_charge(). |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 6867 | */ |
| 6868 | void mem_cgroup_uncharge_list(struct list_head *page_list) |
| 6869 | { |
Muchun Song | c41a40b | 2021-02-24 12:04:08 -0800 | [diff] [blame] | 6870 | struct uncharge_gather ug; |
| 6871 | struct page *page; |
| 6872 | |
Johannes Weiner | 747db95 | 2014-08-08 14:19:24 -0700 | [diff] [blame] | 6873 | if (mem_cgroup_disabled()) |
| 6874 | return; |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 6875 | |
Muchun Song | c41a40b | 2021-02-24 12:04:08 -0800 | [diff] [blame] | 6876 | uncharge_gather_clear(&ug); |
| 6877 | list_for_each_entry(page, page_list, lru) |
| 6878 | uncharge_page(page, &ug); |
| 6879 | if (ug.memcg) |
| 6880 | uncharge_batch(&ug); |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 6881 | } |
| 6882 | |
| 6883 | /** |
Johannes Weiner | 6a93ca8 | 2016-03-15 14:57:19 -0700 | [diff] [blame] | 6884 | * mem_cgroup_migrate - charge a page's replacement |
| 6885 | * @oldpage: currently circulating page |
| 6886 | * @newpage: replacement page |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 6887 | * |
Johannes Weiner | 6a93ca8 | 2016-03-15 14:57:19 -0700 | [diff] [blame] | 6888 | * Charge @newpage as a replacement page for @oldpage. @oldpage will |
| 6889 | * be uncharged upon free. |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 6890 | * |
| 6891 | * Both pages must be locked, @newpage->mapping must be set up. |
| 6892 | */ |
Johannes Weiner | 6a93ca8 | 2016-03-15 14:57:19 -0700 | [diff] [blame] | 6893 | void mem_cgroup_migrate(struct page *oldpage, struct page *newpage) |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 6894 | { |
Johannes Weiner | 2983331 | 2014-12-10 15:44:02 -0800 | [diff] [blame] | 6895 | struct mem_cgroup *memcg; |
Johannes Weiner | 44b7a8d | 2016-01-20 15:03:16 -0800 | [diff] [blame] | 6896 | unsigned int nr_pages; |
Tejun Heo | d93c413 | 2016-06-24 14:49:54 -0700 | [diff] [blame] | 6897 | unsigned long flags; |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 6898 | |
| 6899 | VM_BUG_ON_PAGE(!PageLocked(oldpage), oldpage); |
| 6900 | VM_BUG_ON_PAGE(!PageLocked(newpage), newpage); |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 6901 | VM_BUG_ON_PAGE(PageAnon(oldpage) != PageAnon(newpage), newpage); |
Johannes Weiner | 6abb5a8 | 2014-08-08 14:19:33 -0700 | [diff] [blame] | 6902 | VM_BUG_ON_PAGE(PageTransHuge(oldpage) != PageTransHuge(newpage), |
| 6903 | newpage); |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 6904 | |
| 6905 | if (mem_cgroup_disabled()) |
| 6906 | return; |
| 6907 | |
| 6908 | /* Page cache replacement: new page already charged? */ |
Roman Gushchin | bcfe06b | 2020-12-01 13:58:27 -0800 | [diff] [blame] | 6909 | if (page_memcg(newpage)) |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 6910 | return; |
| 6911 | |
Roman Gushchin | bcfe06b | 2020-12-01 13:58:27 -0800 | [diff] [blame] | 6912 | memcg = page_memcg(oldpage); |
Alex Shi | a405588 | 2020-12-18 14:01:31 -0800 | [diff] [blame] | 6913 | VM_WARN_ON_ONCE_PAGE(!memcg, oldpage); |
Johannes Weiner | 2983331 | 2014-12-10 15:44:02 -0800 | [diff] [blame] | 6914 | if (!memcg) |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 6915 | return; |
| 6916 | |
Johannes Weiner | 44b7a8d | 2016-01-20 15:03:16 -0800 | [diff] [blame] | 6917 | /* Force-charge the new page. The old one will be freed soon */ |
Matthew Wilcox (Oracle) | 6c35784 | 2020-08-14 17:30:37 -0700 | [diff] [blame] | 6918 | nr_pages = thp_nr_pages(newpage); |
Johannes Weiner | 44b7a8d | 2016-01-20 15:03:16 -0800 | [diff] [blame] | 6919 | |
| 6920 | page_counter_charge(&memcg->memory, nr_pages); |
| 6921 | if (do_memsw_account()) |
| 6922 | page_counter_charge(&memcg->memsw, nr_pages); |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 6923 | |
Johannes Weiner | 1a3e1f4 | 2020-08-06 23:20:45 -0700 | [diff] [blame] | 6924 | css_get(&memcg->css); |
Johannes Weiner | d9eb1ea | 2020-06-03 16:02:24 -0700 | [diff] [blame] | 6925 | commit_charge(newpage, memcg); |
Johannes Weiner | 44b7a8d | 2016-01-20 15:03:16 -0800 | [diff] [blame] | 6926 | |
Tejun Heo | d93c413 | 2016-06-24 14:49:54 -0700 | [diff] [blame] | 6927 | local_irq_save(flags); |
Johannes Weiner | 3fba69a | 2020-06-03 16:01:31 -0700 | [diff] [blame] | 6928 | mem_cgroup_charge_statistics(memcg, newpage, nr_pages); |
Johannes Weiner | 44b7a8d | 2016-01-20 15:03:16 -0800 | [diff] [blame] | 6929 | memcg_check_events(memcg, newpage); |
Tejun Heo | d93c413 | 2016-06-24 14:49:54 -0700 | [diff] [blame] | 6930 | local_irq_restore(flags); |
Johannes Weiner | 0a31bc9 | 2014-08-08 14:19:22 -0700 | [diff] [blame] | 6931 | } |
| 6932 | |
Johannes Weiner | ef12947 | 2016-01-14 15:21:34 -0800 | [diff] [blame] | 6933 | DEFINE_STATIC_KEY_FALSE(memcg_sockets_enabled_key); |
Johannes Weiner | 1109208 | 2016-01-14 15:21:26 -0800 | [diff] [blame] | 6934 | EXPORT_SYMBOL(memcg_sockets_enabled_key); |
| 6935 | |
Johannes Weiner | 2d75807 | 2016-10-07 17:00:58 -0700 | [diff] [blame] | 6936 | void mem_cgroup_sk_alloc(struct sock *sk) |
Johannes Weiner | 1109208 | 2016-01-14 15:21:26 -0800 | [diff] [blame] | 6937 | { |
| 6938 | struct mem_cgroup *memcg; |
| 6939 | |
Johannes Weiner | 2d75807 | 2016-10-07 17:00:58 -0700 | [diff] [blame] | 6940 | if (!mem_cgroup_sockets_enabled) |
| 6941 | return; |
| 6942 | |
Shakeel Butt | e876ecc | 2020-03-09 22:16:05 -0700 | [diff] [blame] | 6943 | /* Do not associate the sock with unrelated interrupted task's memcg. */ |
| 6944 | if (in_interrupt()) |
| 6945 | return; |
| 6946 | |
Johannes Weiner | 1109208 | 2016-01-14 15:21:26 -0800 | [diff] [blame] | 6947 | rcu_read_lock(); |
| 6948 | memcg = mem_cgroup_from_task(current); |
Johannes Weiner | f7e1cb6 | 2016-01-14 15:21:29 -0800 | [diff] [blame] | 6949 | if (memcg == root_mem_cgroup) |
| 6950 | goto out; |
Johannes Weiner | 0db1529 | 2016-01-20 15:02:50 -0800 | [diff] [blame] | 6951 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys) && !memcg->tcpmem_active) |
Johannes Weiner | f7e1cb6 | 2016-01-14 15:21:29 -0800 | [diff] [blame] | 6952 | goto out; |
Shakeel Butt | 8965aa2 | 2020-04-01 21:07:10 -0700 | [diff] [blame] | 6953 | if (css_tryget(&memcg->css)) |
Johannes Weiner | 1109208 | 2016-01-14 15:21:26 -0800 | [diff] [blame] | 6954 | sk->sk_memcg = memcg; |
Johannes Weiner | f7e1cb6 | 2016-01-14 15:21:29 -0800 | [diff] [blame] | 6955 | out: |
Johannes Weiner | 1109208 | 2016-01-14 15:21:26 -0800 | [diff] [blame] | 6956 | rcu_read_unlock(); |
| 6957 | } |
Johannes Weiner | 1109208 | 2016-01-14 15:21:26 -0800 | [diff] [blame] | 6958 | |
Johannes Weiner | 2d75807 | 2016-10-07 17:00:58 -0700 | [diff] [blame] | 6959 | void mem_cgroup_sk_free(struct sock *sk) |
Johannes Weiner | 1109208 | 2016-01-14 15:21:26 -0800 | [diff] [blame] | 6960 | { |
Johannes Weiner | 2d75807 | 2016-10-07 17:00:58 -0700 | [diff] [blame] | 6961 | if (sk->sk_memcg) |
| 6962 | css_put(&sk->sk_memcg->css); |
Johannes Weiner | 1109208 | 2016-01-14 15:21:26 -0800 | [diff] [blame] | 6963 | } |
| 6964 | |
| 6965 | /** |
| 6966 | * mem_cgroup_charge_skmem - charge socket memory |
| 6967 | * @memcg: memcg to charge |
| 6968 | * @nr_pages: number of pages to charge |
| 6969 | * |
| 6970 | * Charges @nr_pages to @memcg. Returns %true if the charge fit within |
| 6971 | * @memcg's configured limit, %false if the charge had to be forced. |
| 6972 | */ |
| 6973 | bool mem_cgroup_charge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages) |
| 6974 | { |
Johannes Weiner | f7e1cb6 | 2016-01-14 15:21:29 -0800 | [diff] [blame] | 6975 | gfp_t gfp_mask = GFP_KERNEL; |
Johannes Weiner | 1109208 | 2016-01-14 15:21:26 -0800 | [diff] [blame] | 6976 | |
Johannes Weiner | f7e1cb6 | 2016-01-14 15:21:29 -0800 | [diff] [blame] | 6977 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
Johannes Weiner | 0db1529 | 2016-01-20 15:02:50 -0800 | [diff] [blame] | 6978 | struct page_counter *fail; |
Johannes Weiner | f7e1cb6 | 2016-01-14 15:21:29 -0800 | [diff] [blame] | 6979 | |
Johannes Weiner | 0db1529 | 2016-01-20 15:02:50 -0800 | [diff] [blame] | 6980 | if (page_counter_try_charge(&memcg->tcpmem, nr_pages, &fail)) { |
| 6981 | memcg->tcpmem_pressure = 0; |
Johannes Weiner | f7e1cb6 | 2016-01-14 15:21:29 -0800 | [diff] [blame] | 6982 | return true; |
| 6983 | } |
Johannes Weiner | 0db1529 | 2016-01-20 15:02:50 -0800 | [diff] [blame] | 6984 | page_counter_charge(&memcg->tcpmem, nr_pages); |
| 6985 | memcg->tcpmem_pressure = 1; |
Johannes Weiner | f7e1cb6 | 2016-01-14 15:21:29 -0800 | [diff] [blame] | 6986 | return false; |
Johannes Weiner | 1109208 | 2016-01-14 15:21:26 -0800 | [diff] [blame] | 6987 | } |
Johannes Weiner | d886f4e | 2016-01-20 15:02:47 -0800 | [diff] [blame] | 6988 | |
Johannes Weiner | f7e1cb6 | 2016-01-14 15:21:29 -0800 | [diff] [blame] | 6989 | /* Don't block in the packet receive path */ |
| 6990 | if (in_softirq()) |
| 6991 | gfp_mask = GFP_NOWAIT; |
| 6992 | |
Johannes Weiner | c9019e9 | 2018-01-31 16:16:37 -0800 | [diff] [blame] | 6993 | mod_memcg_state(memcg, MEMCG_SOCK, nr_pages); |
Johannes Weiner | b2807f0 | 2016-01-20 15:03:22 -0800 | [diff] [blame] | 6994 | |
Johannes Weiner | f7e1cb6 | 2016-01-14 15:21:29 -0800 | [diff] [blame] | 6995 | if (try_charge(memcg, gfp_mask, nr_pages) == 0) |
| 6996 | return true; |
| 6997 | |
| 6998 | try_charge(memcg, gfp_mask|__GFP_NOFAIL, nr_pages); |
Johannes Weiner | 1109208 | 2016-01-14 15:21:26 -0800 | [diff] [blame] | 6999 | return false; |
| 7000 | } |
| 7001 | |
| 7002 | /** |
| 7003 | * mem_cgroup_uncharge_skmem - uncharge socket memory |
Mike Rapoport | b7701a5 | 2018-02-06 15:42:13 -0800 | [diff] [blame] | 7004 | * @memcg: memcg to uncharge |
| 7005 | * @nr_pages: number of pages to uncharge |
Johannes Weiner | 1109208 | 2016-01-14 15:21:26 -0800 | [diff] [blame] | 7006 | */ |
| 7007 | void mem_cgroup_uncharge_skmem(struct mem_cgroup *memcg, unsigned int nr_pages) |
| 7008 | { |
Johannes Weiner | f7e1cb6 | 2016-01-14 15:21:29 -0800 | [diff] [blame] | 7009 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) { |
Johannes Weiner | 0db1529 | 2016-01-20 15:02:50 -0800 | [diff] [blame] | 7010 | page_counter_uncharge(&memcg->tcpmem, nr_pages); |
Johannes Weiner | f7e1cb6 | 2016-01-14 15:21:29 -0800 | [diff] [blame] | 7011 | return; |
| 7012 | } |
Johannes Weiner | d886f4e | 2016-01-20 15:02:47 -0800 | [diff] [blame] | 7013 | |
Johannes Weiner | c9019e9 | 2018-01-31 16:16:37 -0800 | [diff] [blame] | 7014 | mod_memcg_state(memcg, MEMCG_SOCK, -nr_pages); |
Johannes Weiner | b2807f0 | 2016-01-20 15:03:22 -0800 | [diff] [blame] | 7015 | |
Roman Gushchin | 475d048 | 2017-09-08 16:13:09 -0700 | [diff] [blame] | 7016 | refill_stock(memcg, nr_pages); |
Johannes Weiner | 1109208 | 2016-01-14 15:21:26 -0800 | [diff] [blame] | 7017 | } |
| 7018 | |
Johannes Weiner | f7e1cb6 | 2016-01-14 15:21:29 -0800 | [diff] [blame] | 7019 | static int __init cgroup_memory(char *s) |
| 7020 | { |
| 7021 | char *token; |
| 7022 | |
| 7023 | while ((token = strsep(&s, ",")) != NULL) { |
| 7024 | if (!*token) |
| 7025 | continue; |
| 7026 | if (!strcmp(token, "nosocket")) |
| 7027 | cgroup_memory_nosocket = true; |
Vladimir Davydov | 04823c8 | 2016-01-20 15:02:38 -0800 | [diff] [blame] | 7028 | if (!strcmp(token, "nokmem")) |
| 7029 | cgroup_memory_nokmem = true; |
Johannes Weiner | f7e1cb6 | 2016-01-14 15:21:29 -0800 | [diff] [blame] | 7030 | } |
| 7031 | return 0; |
| 7032 | } |
| 7033 | __setup("cgroup.memory=", cgroup_memory); |
Johannes Weiner | 1109208 | 2016-01-14 15:21:26 -0800 | [diff] [blame] | 7034 | |
Michal Hocko | 2d11085 | 2013-02-22 16:34:43 -0800 | [diff] [blame] | 7035 | /* |
Michal Hocko | 1081312 | 2013-02-22 16:35:41 -0800 | [diff] [blame] | 7036 | * subsys_initcall() for memory controller. |
| 7037 | * |
Sebastian Andrzej Siewior | 308167f | 2016-11-03 15:49:59 +0100 | [diff] [blame] | 7038 | * Some parts like memcg_hotplug_cpu_dead() have to be initialized from this |
| 7039 | * context because of lock dependencies (cgroup_lock -> cpu hotplug) but |
| 7040 | * basically everything that doesn't depend on a specific mem_cgroup structure |
| 7041 | * should be initialized from here. |
Michal Hocko | 2d11085 | 2013-02-22 16:34:43 -0800 | [diff] [blame] | 7042 | */ |
| 7043 | static int __init mem_cgroup_init(void) |
| 7044 | { |
Johannes Weiner | 95a045f | 2015-02-11 15:26:33 -0800 | [diff] [blame] | 7045 | int cpu, node; |
| 7046 | |
Muchun Song | f3344ad | 2021-02-24 12:03:15 -0800 | [diff] [blame] | 7047 | /* |
| 7048 | * Currently s32 type (can refer to struct batched_lruvec_stat) is |
| 7049 | * used for per-memcg-per-cpu caching of per-node statistics. In order |
| 7050 | * to work fine, we should make sure that the overfill threshold can't |
| 7051 | * exceed S32_MAX / PAGE_SIZE. |
| 7052 | */ |
| 7053 | BUILD_BUG_ON(MEMCG_CHARGE_BATCH > S32_MAX / PAGE_SIZE); |
| 7054 | |
Sebastian Andrzej Siewior | 308167f | 2016-11-03 15:49:59 +0100 | [diff] [blame] | 7055 | cpuhp_setup_state_nocalls(CPUHP_MM_MEMCQ_DEAD, "mm/memctrl:dead", NULL, |
| 7056 | memcg_hotplug_cpu_dead); |
Johannes Weiner | 95a045f | 2015-02-11 15:26:33 -0800 | [diff] [blame] | 7057 | |
| 7058 | for_each_possible_cpu(cpu) |
| 7059 | INIT_WORK(&per_cpu_ptr(&memcg_stock, cpu)->work, |
| 7060 | drain_local_stock); |
| 7061 | |
| 7062 | for_each_node(node) { |
| 7063 | struct mem_cgroup_tree_per_node *rtpn; |
Johannes Weiner | 95a045f | 2015-02-11 15:26:33 -0800 | [diff] [blame] | 7064 | |
| 7065 | rtpn = kzalloc_node(sizeof(*rtpn), GFP_KERNEL, |
| 7066 | node_online(node) ? node : NUMA_NO_NODE); |
| 7067 | |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 7068 | rtpn->rb_root = RB_ROOT; |
Davidlohr Bueso | fa90b2f | 2017-09-08 16:15:21 -0700 | [diff] [blame] | 7069 | rtpn->rb_rightmost = NULL; |
Mel Gorman | ef8f232 | 2016-07-28 15:46:05 -0700 | [diff] [blame] | 7070 | spin_lock_init(&rtpn->lock); |
Johannes Weiner | 95a045f | 2015-02-11 15:26:33 -0800 | [diff] [blame] | 7071 | soft_limit_tree.rb_tree_per_node[node] = rtpn; |
| 7072 | } |
| 7073 | |
Michal Hocko | 2d11085 | 2013-02-22 16:34:43 -0800 | [diff] [blame] | 7074 | return 0; |
| 7075 | } |
| 7076 | subsys_initcall(mem_cgroup_init); |
Johannes Weiner | 21afa38 | 2015-02-11 15:26:36 -0800 | [diff] [blame] | 7077 | |
| 7078 | #ifdef CONFIG_MEMCG_SWAP |
Arnd Bergmann | 358c07f | 2016-08-25 15:17:08 -0700 | [diff] [blame] | 7079 | static struct mem_cgroup *mem_cgroup_id_get_online(struct mem_cgroup *memcg) |
| 7080 | { |
Kirill Tkhai | 1c2d479 | 2018-10-26 15:09:28 -0700 | [diff] [blame] | 7081 | while (!refcount_inc_not_zero(&memcg->id.ref)) { |
Arnd Bergmann | 358c07f | 2016-08-25 15:17:08 -0700 | [diff] [blame] | 7082 | /* |
| 7083 | * The root cgroup cannot be destroyed, so it's refcount must |
| 7084 | * always be >= 1. |
| 7085 | */ |
| 7086 | if (WARN_ON_ONCE(memcg == root_mem_cgroup)) { |
| 7087 | VM_BUG_ON(1); |
| 7088 | break; |
| 7089 | } |
| 7090 | memcg = parent_mem_cgroup(memcg); |
| 7091 | if (!memcg) |
| 7092 | memcg = root_mem_cgroup; |
| 7093 | } |
| 7094 | return memcg; |
| 7095 | } |
| 7096 | |
Johannes Weiner | 21afa38 | 2015-02-11 15:26:36 -0800 | [diff] [blame] | 7097 | /** |
| 7098 | * mem_cgroup_swapout - transfer a memsw charge to swap |
| 7099 | * @page: page whose memsw charge to transfer |
| 7100 | * @entry: swap entry to move the charge to |
| 7101 | * |
| 7102 | * Transfer the memsw charge of @page to @entry. |
| 7103 | */ |
| 7104 | void mem_cgroup_swapout(struct page *page, swp_entry_t entry) |
| 7105 | { |
Vladimir Davydov | 1f47b61 | 2016-08-11 15:33:00 -0700 | [diff] [blame] | 7106 | struct mem_cgroup *memcg, *swap_memcg; |
Huang Ying | d6810d7 | 2017-09-06 16:22:45 -0700 | [diff] [blame] | 7107 | unsigned int nr_entries; |
Johannes Weiner | 21afa38 | 2015-02-11 15:26:36 -0800 | [diff] [blame] | 7108 | unsigned short oldid; |
| 7109 | |
| 7110 | VM_BUG_ON_PAGE(PageLRU(page), page); |
| 7111 | VM_BUG_ON_PAGE(page_count(page), page); |
| 7112 | |
Alex Shi | 76358ab | 2020-12-18 14:01:28 -0800 | [diff] [blame] | 7113 | if (mem_cgroup_disabled()) |
| 7114 | return; |
| 7115 | |
Johannes Weiner | 2d1c498 | 2020-06-03 16:02:14 -0700 | [diff] [blame] | 7116 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
Johannes Weiner | 21afa38 | 2015-02-11 15:26:36 -0800 | [diff] [blame] | 7117 | return; |
| 7118 | |
Roman Gushchin | bcfe06b | 2020-12-01 13:58:27 -0800 | [diff] [blame] | 7119 | memcg = page_memcg(page); |
Johannes Weiner | 21afa38 | 2015-02-11 15:26:36 -0800 | [diff] [blame] | 7120 | |
Alex Shi | a405588 | 2020-12-18 14:01:31 -0800 | [diff] [blame] | 7121 | VM_WARN_ON_ONCE_PAGE(!memcg, page); |
Johannes Weiner | 21afa38 | 2015-02-11 15:26:36 -0800 | [diff] [blame] | 7122 | if (!memcg) |
| 7123 | return; |
| 7124 | |
Vladimir Davydov | 1f47b61 | 2016-08-11 15:33:00 -0700 | [diff] [blame] | 7125 | /* |
| 7126 | * In case the memcg owning these pages has been offlined and doesn't |
| 7127 | * have an ID allocated to it anymore, charge the closest online |
| 7128 | * ancestor for the swap instead and transfer the memory+swap charge. |
| 7129 | */ |
| 7130 | swap_memcg = mem_cgroup_id_get_online(memcg); |
Matthew Wilcox (Oracle) | 6c35784 | 2020-08-14 17:30:37 -0700 | [diff] [blame] | 7131 | nr_entries = thp_nr_pages(page); |
Huang Ying | d6810d7 | 2017-09-06 16:22:45 -0700 | [diff] [blame] | 7132 | /* Get references for the tail pages, too */ |
| 7133 | if (nr_entries > 1) |
| 7134 | mem_cgroup_id_get_many(swap_memcg, nr_entries - 1); |
| 7135 | oldid = swap_cgroup_record(entry, mem_cgroup_id(swap_memcg), |
| 7136 | nr_entries); |
Johannes Weiner | 21afa38 | 2015-02-11 15:26:36 -0800 | [diff] [blame] | 7137 | VM_BUG_ON_PAGE(oldid, page); |
Johannes Weiner | c9019e9 | 2018-01-31 16:16:37 -0800 | [diff] [blame] | 7138 | mod_memcg_state(swap_memcg, MEMCG_SWAP, nr_entries); |
Johannes Weiner | 21afa38 | 2015-02-11 15:26:36 -0800 | [diff] [blame] | 7139 | |
Roman Gushchin | bcfe06b | 2020-12-01 13:58:27 -0800 | [diff] [blame] | 7140 | page->memcg_data = 0; |
Johannes Weiner | 21afa38 | 2015-02-11 15:26:36 -0800 | [diff] [blame] | 7141 | |
| 7142 | if (!mem_cgroup_is_root(memcg)) |
Huang Ying | d6810d7 | 2017-09-06 16:22:45 -0700 | [diff] [blame] | 7143 | page_counter_uncharge(&memcg->memory, nr_entries); |
Johannes Weiner | 21afa38 | 2015-02-11 15:26:36 -0800 | [diff] [blame] | 7144 | |
Johannes Weiner | 2d1c498 | 2020-06-03 16:02:14 -0700 | [diff] [blame] | 7145 | if (!cgroup_memory_noswap && memcg != swap_memcg) { |
Vladimir Davydov | 1f47b61 | 2016-08-11 15:33:00 -0700 | [diff] [blame] | 7146 | if (!mem_cgroup_is_root(swap_memcg)) |
Huang Ying | d6810d7 | 2017-09-06 16:22:45 -0700 | [diff] [blame] | 7147 | page_counter_charge(&swap_memcg->memsw, nr_entries); |
| 7148 | page_counter_uncharge(&memcg->memsw, nr_entries); |
Vladimir Davydov | 1f47b61 | 2016-08-11 15:33:00 -0700 | [diff] [blame] | 7149 | } |
| 7150 | |
Sebastian Andrzej Siewior | ce9ce66 | 2015-09-04 15:47:50 -0700 | [diff] [blame] | 7151 | /* |
| 7152 | * Interrupts should be disabled here because the caller holds the |
Matthew Wilcox | b93b016 | 2018-04-10 16:36:56 -0700 | [diff] [blame] | 7153 | * i_pages lock which is taken with interrupts-off. It is |
Sebastian Andrzej Siewior | ce9ce66 | 2015-09-04 15:47:50 -0700 | [diff] [blame] | 7154 | * important here to have the interrupts disabled because it is the |
Matthew Wilcox | b93b016 | 2018-04-10 16:36:56 -0700 | [diff] [blame] | 7155 | * only synchronisation we have for updating the per-CPU variables. |
Sebastian Andrzej Siewior | ce9ce66 | 2015-09-04 15:47:50 -0700 | [diff] [blame] | 7156 | */ |
| 7157 | VM_BUG_ON(!irqs_disabled()); |
Johannes Weiner | 3fba69a | 2020-06-03 16:01:31 -0700 | [diff] [blame] | 7158 | mem_cgroup_charge_statistics(memcg, page, -nr_entries); |
Johannes Weiner | 21afa38 | 2015-02-11 15:26:36 -0800 | [diff] [blame] | 7159 | memcg_check_events(memcg, page); |
Johannes Weiner | 73f576c | 2016-07-20 15:44:57 -0700 | [diff] [blame] | 7160 | |
Johannes Weiner | 1a3e1f4 | 2020-08-06 23:20:45 -0700 | [diff] [blame] | 7161 | css_put(&memcg->css); |
Johannes Weiner | 21afa38 | 2015-02-11 15:26:36 -0800 | [diff] [blame] | 7162 | } |
| 7163 | |
Huang Ying | 38d8b4e | 2017-07-06 15:37:18 -0700 | [diff] [blame] | 7164 | /** |
| 7165 | * mem_cgroup_try_charge_swap - try charging swap space for a page |
Vladimir Davydov | 37e8435 | 2016-01-20 15:02:56 -0800 | [diff] [blame] | 7166 | * @page: page being added to swap |
| 7167 | * @entry: swap entry to charge |
| 7168 | * |
Huang Ying | 38d8b4e | 2017-07-06 15:37:18 -0700 | [diff] [blame] | 7169 | * Try to charge @page's memcg for the swap space at @entry. |
Vladimir Davydov | 37e8435 | 2016-01-20 15:02:56 -0800 | [diff] [blame] | 7170 | * |
| 7171 | * Returns 0 on success, -ENOMEM on failure. |
| 7172 | */ |
| 7173 | int mem_cgroup_try_charge_swap(struct page *page, swp_entry_t entry) |
| 7174 | { |
Matthew Wilcox (Oracle) | 6c35784 | 2020-08-14 17:30:37 -0700 | [diff] [blame] | 7175 | unsigned int nr_pages = thp_nr_pages(page); |
Vladimir Davydov | 37e8435 | 2016-01-20 15:02:56 -0800 | [diff] [blame] | 7176 | struct page_counter *counter; |
Huang Ying | 38d8b4e | 2017-07-06 15:37:18 -0700 | [diff] [blame] | 7177 | struct mem_cgroup *memcg; |
Vladimir Davydov | 37e8435 | 2016-01-20 15:02:56 -0800 | [diff] [blame] | 7178 | unsigned short oldid; |
| 7179 | |
Alex Shi | 76358ab | 2020-12-18 14:01:28 -0800 | [diff] [blame] | 7180 | if (mem_cgroup_disabled()) |
| 7181 | return 0; |
| 7182 | |
Johannes Weiner | 2d1c498 | 2020-06-03 16:02:14 -0700 | [diff] [blame] | 7183 | if (!cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
Vladimir Davydov | 37e8435 | 2016-01-20 15:02:56 -0800 | [diff] [blame] | 7184 | return 0; |
| 7185 | |
Roman Gushchin | bcfe06b | 2020-12-01 13:58:27 -0800 | [diff] [blame] | 7186 | memcg = page_memcg(page); |
Vladimir Davydov | 37e8435 | 2016-01-20 15:02:56 -0800 | [diff] [blame] | 7187 | |
Alex Shi | a405588 | 2020-12-18 14:01:31 -0800 | [diff] [blame] | 7188 | VM_WARN_ON_ONCE_PAGE(!memcg, page); |
Vladimir Davydov | 37e8435 | 2016-01-20 15:02:56 -0800 | [diff] [blame] | 7189 | if (!memcg) |
| 7190 | return 0; |
| 7191 | |
Tejun Heo | f3a53a3 | 2018-06-07 17:05:35 -0700 | [diff] [blame] | 7192 | if (!entry.val) { |
| 7193 | memcg_memory_event(memcg, MEMCG_SWAP_FAIL); |
Tejun Heo | bb98f2c | 2018-06-07 17:05:31 -0700 | [diff] [blame] | 7194 | return 0; |
Tejun Heo | f3a53a3 | 2018-06-07 17:05:35 -0700 | [diff] [blame] | 7195 | } |
Tejun Heo | bb98f2c | 2018-06-07 17:05:31 -0700 | [diff] [blame] | 7196 | |
Vladimir Davydov | 1f47b61 | 2016-08-11 15:33:00 -0700 | [diff] [blame] | 7197 | memcg = mem_cgroup_id_get_online(memcg); |
Vladimir Davydov | 37e8435 | 2016-01-20 15:02:56 -0800 | [diff] [blame] | 7198 | |
Johannes Weiner | 2d1c498 | 2020-06-03 16:02:14 -0700 | [diff] [blame] | 7199 | if (!cgroup_memory_noswap && !mem_cgroup_is_root(memcg) && |
Huang Ying | 38d8b4e | 2017-07-06 15:37:18 -0700 | [diff] [blame] | 7200 | !page_counter_try_charge(&memcg->swap, nr_pages, &counter)) { |
Tejun Heo | f3a53a3 | 2018-06-07 17:05:35 -0700 | [diff] [blame] | 7201 | memcg_memory_event(memcg, MEMCG_SWAP_MAX); |
| 7202 | memcg_memory_event(memcg, MEMCG_SWAP_FAIL); |
Vladimir Davydov | 1f47b61 | 2016-08-11 15:33:00 -0700 | [diff] [blame] | 7203 | mem_cgroup_id_put(memcg); |
| 7204 | return -ENOMEM; |
| 7205 | } |
| 7206 | |
Huang Ying | 38d8b4e | 2017-07-06 15:37:18 -0700 | [diff] [blame] | 7207 | /* Get references for the tail pages, too */ |
| 7208 | if (nr_pages > 1) |
| 7209 | mem_cgroup_id_get_many(memcg, nr_pages - 1); |
| 7210 | oldid = swap_cgroup_record(entry, mem_cgroup_id(memcg), nr_pages); |
Vladimir Davydov | 37e8435 | 2016-01-20 15:02:56 -0800 | [diff] [blame] | 7211 | VM_BUG_ON_PAGE(oldid, page); |
Johannes Weiner | c9019e9 | 2018-01-31 16:16:37 -0800 | [diff] [blame] | 7212 | mod_memcg_state(memcg, MEMCG_SWAP, nr_pages); |
Vladimir Davydov | 37e8435 | 2016-01-20 15:02:56 -0800 | [diff] [blame] | 7213 | |
Vladimir Davydov | 37e8435 | 2016-01-20 15:02:56 -0800 | [diff] [blame] | 7214 | return 0; |
| 7215 | } |
| 7216 | |
Johannes Weiner | 21afa38 | 2015-02-11 15:26:36 -0800 | [diff] [blame] | 7217 | /** |
Huang Ying | 38d8b4e | 2017-07-06 15:37:18 -0700 | [diff] [blame] | 7218 | * mem_cgroup_uncharge_swap - uncharge swap space |
Johannes Weiner | 21afa38 | 2015-02-11 15:26:36 -0800 | [diff] [blame] | 7219 | * @entry: swap entry to uncharge |
Huang Ying | 38d8b4e | 2017-07-06 15:37:18 -0700 | [diff] [blame] | 7220 | * @nr_pages: the amount of swap space to uncharge |
Johannes Weiner | 21afa38 | 2015-02-11 15:26:36 -0800 | [diff] [blame] | 7221 | */ |
Huang Ying | 38d8b4e | 2017-07-06 15:37:18 -0700 | [diff] [blame] | 7222 | 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] | 7223 | { |
| 7224 | struct mem_cgroup *memcg; |
| 7225 | unsigned short id; |
| 7226 | |
Huang Ying | 38d8b4e | 2017-07-06 15:37:18 -0700 | [diff] [blame] | 7227 | id = swap_cgroup_record(entry, 0, nr_pages); |
Johannes Weiner | 21afa38 | 2015-02-11 15:26:36 -0800 | [diff] [blame] | 7228 | rcu_read_lock(); |
Vladimir Davydov | adbe427 | 2015-04-15 16:13:00 -0700 | [diff] [blame] | 7229 | memcg = mem_cgroup_from_id(id); |
Johannes Weiner | 21afa38 | 2015-02-11 15:26:36 -0800 | [diff] [blame] | 7230 | if (memcg) { |
Johannes Weiner | 2d1c498 | 2020-06-03 16:02:14 -0700 | [diff] [blame] | 7231 | if (!cgroup_memory_noswap && !mem_cgroup_is_root(memcg)) { |
Vladimir Davydov | 37e8435 | 2016-01-20 15:02:56 -0800 | [diff] [blame] | 7232 | if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
Huang Ying | 38d8b4e | 2017-07-06 15:37:18 -0700 | [diff] [blame] | 7233 | page_counter_uncharge(&memcg->swap, nr_pages); |
Vladimir Davydov | 37e8435 | 2016-01-20 15:02:56 -0800 | [diff] [blame] | 7234 | else |
Huang Ying | 38d8b4e | 2017-07-06 15:37:18 -0700 | [diff] [blame] | 7235 | page_counter_uncharge(&memcg->memsw, nr_pages); |
Vladimir Davydov | 37e8435 | 2016-01-20 15:02:56 -0800 | [diff] [blame] | 7236 | } |
Johannes Weiner | c9019e9 | 2018-01-31 16:16:37 -0800 | [diff] [blame] | 7237 | mod_memcg_state(memcg, MEMCG_SWAP, -nr_pages); |
Huang Ying | 38d8b4e | 2017-07-06 15:37:18 -0700 | [diff] [blame] | 7238 | mem_cgroup_id_put_many(memcg, nr_pages); |
Johannes Weiner | 21afa38 | 2015-02-11 15:26:36 -0800 | [diff] [blame] | 7239 | } |
| 7240 | rcu_read_unlock(); |
| 7241 | } |
| 7242 | |
Vladimir Davydov | d8b3843 | 2016-01-20 15:03:07 -0800 | [diff] [blame] | 7243 | long mem_cgroup_get_nr_swap_pages(struct mem_cgroup *memcg) |
| 7244 | { |
| 7245 | long nr_swap_pages = get_nr_swap_pages(); |
| 7246 | |
Johannes Weiner | eccb52e | 2020-06-03 16:02:11 -0700 | [diff] [blame] | 7247 | if (cgroup_memory_noswap || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
Vladimir Davydov | d8b3843 | 2016-01-20 15:03:07 -0800 | [diff] [blame] | 7248 | return nr_swap_pages; |
| 7249 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) |
| 7250 | nr_swap_pages = min_t(long, nr_swap_pages, |
Roman Gushchin | bbec2e1 | 2018-06-07 17:06:18 -0700 | [diff] [blame] | 7251 | READ_ONCE(memcg->swap.max) - |
Vladimir Davydov | d8b3843 | 2016-01-20 15:03:07 -0800 | [diff] [blame] | 7252 | page_counter_read(&memcg->swap)); |
| 7253 | return nr_swap_pages; |
| 7254 | } |
| 7255 | |
Vladimir Davydov | 5ccc5ab | 2016-01-20 15:03:10 -0800 | [diff] [blame] | 7256 | bool mem_cgroup_swap_full(struct page *page) |
| 7257 | { |
| 7258 | struct mem_cgroup *memcg; |
| 7259 | |
| 7260 | VM_BUG_ON_PAGE(!PageLocked(page), page); |
| 7261 | |
| 7262 | if (vm_swap_full()) |
| 7263 | return true; |
Johannes Weiner | eccb52e | 2020-06-03 16:02:11 -0700 | [diff] [blame] | 7264 | if (cgroup_memory_noswap || !cgroup_subsys_on_dfl(memory_cgrp_subsys)) |
Vladimir Davydov | 5ccc5ab | 2016-01-20 15:03:10 -0800 | [diff] [blame] | 7265 | return false; |
| 7266 | |
Roman Gushchin | bcfe06b | 2020-12-01 13:58:27 -0800 | [diff] [blame] | 7267 | memcg = page_memcg(page); |
Vladimir Davydov | 5ccc5ab | 2016-01-20 15:03:10 -0800 | [diff] [blame] | 7268 | if (!memcg) |
| 7269 | return false; |
| 7270 | |
Jakub Kicinski | 4b82ab4 | 2020-06-01 21:49:52 -0700 | [diff] [blame] | 7271 | for (; memcg != root_mem_cgroup; memcg = parent_mem_cgroup(memcg)) { |
| 7272 | unsigned long usage = page_counter_read(&memcg->swap); |
| 7273 | |
| 7274 | if (usage * 2 >= READ_ONCE(memcg->swap.high) || |
| 7275 | usage * 2 >= READ_ONCE(memcg->swap.max)) |
Vladimir Davydov | 5ccc5ab | 2016-01-20 15:03:10 -0800 | [diff] [blame] | 7276 | return true; |
Jakub Kicinski | 4b82ab4 | 2020-06-01 21:49:52 -0700 | [diff] [blame] | 7277 | } |
Vladimir Davydov | 5ccc5ab | 2016-01-20 15:03:10 -0800 | [diff] [blame] | 7278 | |
| 7279 | return false; |
| 7280 | } |
| 7281 | |
Johannes Weiner | eccb52e | 2020-06-03 16:02:11 -0700 | [diff] [blame] | 7282 | static int __init setup_swap_account(char *s) |
Johannes Weiner | 21afa38 | 2015-02-11 15:26:36 -0800 | [diff] [blame] | 7283 | { |
| 7284 | if (!strcmp(s, "1")) |
Kaixu Xia | 5ab9290 | 2020-12-14 19:07:07 -0800 | [diff] [blame] | 7285 | cgroup_memory_noswap = false; |
Johannes Weiner | 21afa38 | 2015-02-11 15:26:36 -0800 | [diff] [blame] | 7286 | else if (!strcmp(s, "0")) |
Kaixu Xia | 5ab9290 | 2020-12-14 19:07:07 -0800 | [diff] [blame] | 7287 | cgroup_memory_noswap = true; |
Johannes Weiner | 21afa38 | 2015-02-11 15:26:36 -0800 | [diff] [blame] | 7288 | return 1; |
| 7289 | } |
Johannes Weiner | eccb52e | 2020-06-03 16:02:11 -0700 | [diff] [blame] | 7290 | __setup("swapaccount=", setup_swap_account); |
Johannes Weiner | 21afa38 | 2015-02-11 15:26:36 -0800 | [diff] [blame] | 7291 | |
Vladimir Davydov | 37e8435 | 2016-01-20 15:02:56 -0800 | [diff] [blame] | 7292 | static u64 swap_current_read(struct cgroup_subsys_state *css, |
| 7293 | struct cftype *cft) |
| 7294 | { |
| 7295 | struct mem_cgroup *memcg = mem_cgroup_from_css(css); |
| 7296 | |
| 7297 | return (u64)page_counter_read(&memcg->swap) * PAGE_SIZE; |
| 7298 | } |
| 7299 | |
Jakub Kicinski | 4b82ab4 | 2020-06-01 21:49:52 -0700 | [diff] [blame] | 7300 | static int swap_high_show(struct seq_file *m, void *v) |
| 7301 | { |
| 7302 | return seq_puts_memcg_tunable(m, |
| 7303 | READ_ONCE(mem_cgroup_from_seq(m)->swap.high)); |
| 7304 | } |
| 7305 | |
| 7306 | static ssize_t swap_high_write(struct kernfs_open_file *of, |
| 7307 | char *buf, size_t nbytes, loff_t off) |
| 7308 | { |
| 7309 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
| 7310 | unsigned long high; |
| 7311 | int err; |
| 7312 | |
| 7313 | buf = strstrip(buf); |
| 7314 | err = page_counter_memparse(buf, "max", &high); |
| 7315 | if (err) |
| 7316 | return err; |
| 7317 | |
| 7318 | page_counter_set_high(&memcg->swap, high); |
| 7319 | |
| 7320 | return nbytes; |
| 7321 | } |
| 7322 | |
Vladimir Davydov | 37e8435 | 2016-01-20 15:02:56 -0800 | [diff] [blame] | 7323 | static int swap_max_show(struct seq_file *m, void *v) |
| 7324 | { |
Chris Down | 677dc97 | 2019-03-05 15:45:55 -0800 | [diff] [blame] | 7325 | return seq_puts_memcg_tunable(m, |
| 7326 | READ_ONCE(mem_cgroup_from_seq(m)->swap.max)); |
Vladimir Davydov | 37e8435 | 2016-01-20 15:02:56 -0800 | [diff] [blame] | 7327 | } |
| 7328 | |
| 7329 | static ssize_t swap_max_write(struct kernfs_open_file *of, |
| 7330 | char *buf, size_t nbytes, loff_t off) |
| 7331 | { |
| 7332 | struct mem_cgroup *memcg = mem_cgroup_from_css(of_css(of)); |
| 7333 | unsigned long max; |
| 7334 | int err; |
| 7335 | |
| 7336 | buf = strstrip(buf); |
| 7337 | err = page_counter_memparse(buf, "max", &max); |
| 7338 | if (err) |
| 7339 | return err; |
| 7340 | |
Tejun Heo | be09102 | 2018-06-07 17:09:21 -0700 | [diff] [blame] | 7341 | xchg(&memcg->swap.max, max); |
Vladimir Davydov | 37e8435 | 2016-01-20 15:02:56 -0800 | [diff] [blame] | 7342 | |
| 7343 | return nbytes; |
| 7344 | } |
| 7345 | |
Tejun Heo | f3a53a3 | 2018-06-07 17:05:35 -0700 | [diff] [blame] | 7346 | static int swap_events_show(struct seq_file *m, void *v) |
| 7347 | { |
Chris Down | aa9694b | 2019-03-05 15:45:52 -0800 | [diff] [blame] | 7348 | struct mem_cgroup *memcg = mem_cgroup_from_seq(m); |
Tejun Heo | f3a53a3 | 2018-06-07 17:05:35 -0700 | [diff] [blame] | 7349 | |
Jakub Kicinski | 4b82ab4 | 2020-06-01 21:49:52 -0700 | [diff] [blame] | 7350 | seq_printf(m, "high %lu\n", |
| 7351 | atomic_long_read(&memcg->memory_events[MEMCG_SWAP_HIGH])); |
Tejun Heo | f3a53a3 | 2018-06-07 17:05:35 -0700 | [diff] [blame] | 7352 | seq_printf(m, "max %lu\n", |
| 7353 | atomic_long_read(&memcg->memory_events[MEMCG_SWAP_MAX])); |
| 7354 | seq_printf(m, "fail %lu\n", |
| 7355 | atomic_long_read(&memcg->memory_events[MEMCG_SWAP_FAIL])); |
| 7356 | |
| 7357 | return 0; |
| 7358 | } |
| 7359 | |
Vladimir Davydov | 37e8435 | 2016-01-20 15:02:56 -0800 | [diff] [blame] | 7360 | static struct cftype swap_files[] = { |
| 7361 | { |
| 7362 | .name = "swap.current", |
| 7363 | .flags = CFTYPE_NOT_ON_ROOT, |
| 7364 | .read_u64 = swap_current_read, |
| 7365 | }, |
| 7366 | { |
Jakub Kicinski | 4b82ab4 | 2020-06-01 21:49:52 -0700 | [diff] [blame] | 7367 | .name = "swap.high", |
| 7368 | .flags = CFTYPE_NOT_ON_ROOT, |
| 7369 | .seq_show = swap_high_show, |
| 7370 | .write = swap_high_write, |
| 7371 | }, |
| 7372 | { |
Vladimir Davydov | 37e8435 | 2016-01-20 15:02:56 -0800 | [diff] [blame] | 7373 | .name = "swap.max", |
| 7374 | .flags = CFTYPE_NOT_ON_ROOT, |
| 7375 | .seq_show = swap_max_show, |
| 7376 | .write = swap_max_write, |
| 7377 | }, |
Tejun Heo | f3a53a3 | 2018-06-07 17:05:35 -0700 | [diff] [blame] | 7378 | { |
| 7379 | .name = "swap.events", |
| 7380 | .flags = CFTYPE_NOT_ON_ROOT, |
| 7381 | .file_offset = offsetof(struct mem_cgroup, swap_events_file), |
| 7382 | .seq_show = swap_events_show, |
| 7383 | }, |
Vladimir Davydov | 37e8435 | 2016-01-20 15:02:56 -0800 | [diff] [blame] | 7384 | { } /* terminate */ |
| 7385 | }; |
| 7386 | |
Johannes Weiner | eccb52e | 2020-06-03 16:02:11 -0700 | [diff] [blame] | 7387 | static struct cftype memsw_files[] = { |
Johannes Weiner | 21afa38 | 2015-02-11 15:26:36 -0800 | [diff] [blame] | 7388 | { |
| 7389 | .name = "memsw.usage_in_bytes", |
| 7390 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_USAGE), |
| 7391 | .read_u64 = mem_cgroup_read_u64, |
| 7392 | }, |
| 7393 | { |
| 7394 | .name = "memsw.max_usage_in_bytes", |
| 7395 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_MAX_USAGE), |
| 7396 | .write = mem_cgroup_reset, |
| 7397 | .read_u64 = mem_cgroup_read_u64, |
| 7398 | }, |
| 7399 | { |
| 7400 | .name = "memsw.limit_in_bytes", |
| 7401 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_LIMIT), |
| 7402 | .write = mem_cgroup_write, |
| 7403 | .read_u64 = mem_cgroup_read_u64, |
| 7404 | }, |
| 7405 | { |
| 7406 | .name = "memsw.failcnt", |
| 7407 | .private = MEMFILE_PRIVATE(_MEMSWAP, RES_FAILCNT), |
| 7408 | .write = mem_cgroup_reset, |
| 7409 | .read_u64 = mem_cgroup_read_u64, |
| 7410 | }, |
| 7411 | { }, /* terminate */ |
| 7412 | }; |
| 7413 | |
Bhupesh Sharma | 82ff165 | 2020-07-23 21:15:21 -0700 | [diff] [blame] | 7414 | /* |
| 7415 | * If mem_cgroup_swap_init() is implemented as a subsys_initcall() |
| 7416 | * instead of a core_initcall(), this could mean cgroup_memory_noswap still |
| 7417 | * remains set to false even when memcg is disabled via "cgroup_disable=memory" |
| 7418 | * boot parameter. This may result in premature OOPS inside |
| 7419 | * mem_cgroup_get_nr_swap_pages() function in corner cases. |
| 7420 | */ |
Johannes Weiner | 21afa38 | 2015-02-11 15:26:36 -0800 | [diff] [blame] | 7421 | static int __init mem_cgroup_swap_init(void) |
| 7422 | { |
Johannes Weiner | 2d1c498 | 2020-06-03 16:02:14 -0700 | [diff] [blame] | 7423 | /* No memory control -> no swap control */ |
| 7424 | if (mem_cgroup_disabled()) |
| 7425 | cgroup_memory_noswap = true; |
| 7426 | |
| 7427 | if (cgroup_memory_noswap) |
Johannes Weiner | eccb52e | 2020-06-03 16:02:11 -0700 | [diff] [blame] | 7428 | return 0; |
| 7429 | |
| 7430 | WARN_ON(cgroup_add_dfl_cftypes(&memory_cgrp_subsys, swap_files)); |
| 7431 | WARN_ON(cgroup_add_legacy_cftypes(&memory_cgrp_subsys, memsw_files)); |
| 7432 | |
Johannes Weiner | 21afa38 | 2015-02-11 15:26:36 -0800 | [diff] [blame] | 7433 | return 0; |
| 7434 | } |
Bhupesh Sharma | 82ff165 | 2020-07-23 21:15:21 -0700 | [diff] [blame] | 7435 | core_initcall(mem_cgroup_swap_init); |
Johannes Weiner | 21afa38 | 2015-02-11 15:26:36 -0800 | [diff] [blame] | 7436 | |
| 7437 | #endif /* CONFIG_MEMCG_SWAP */ |