1/* drivers/misc/lowmemorykiller.c
2 *
3 * The lowmemorykiller driver lets user-space specify a set of memory thresholds
4 * where processes with a range of oom_score_adj values will get killed. Specify
5 * the minimum oom_score_adj values in
6 * /sys/module/lowmemorykiller/parameters/adj and the number of free pages in
7 * /sys/module/lowmemorykiller/parameters/minfree. Both files take a comma
8 * separated list of numbers in ascending order.
9 *
10 * For example, write "0,8" to /sys/module/lowmemorykiller/parameters/adj and
11 * "1024,4096" to /sys/module/lowmemorykiller/parameters/minfree to kill
12 * processes with a oom_score_adj value of 8 or higher when the free memory
13 * drops below 4096 pages and kill processes with a oom_score_adj value of 0 or
14 * higher when the free memory drops below 1024 pages.
15 *
16 * The driver considers memory used for caches to be free, but if a large
17 * percentage of the cached memory is locked this can be very inaccurate
18 * and processes may not get killed until the normal oom killer is triggered.
19 *
20 * Copyright (C) 2007-2008 Google, Inc.
21 *
22 * This software is licensed under the terms of the GNU General Public
23 * License version 2, as published by the Free Software Foundation, and
24 * may be copied, distributed, and modified under those terms.
25 *
26 * This program is distributed in the hope that it will be useful,
27 * but WITHOUT ANY WARRANTY; without even the implied warranty of
28 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
29 * GNU General Public License for more details.
30 *
31 */
32
33#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
34
35#include <linux/init.h>
36#include <linux/moduleparam.h>
37#include <linux/kernel.h>
38#include <linux/mm.h>
39#include <linux/oom.h>
40#include <linux/sched.h>
41#include <linux/swap.h>
42#include <linux/rcupdate.h>
43#include <linux/profile.h>
44#include <linux/notifier.h>
45
46static uint32_t lowmem_debug_level = 1;
47static short lowmem_adj[6] = {
48	0,
49	1,
50	6,
51	12,
52};
53static int lowmem_adj_size = 4;
54static int lowmem_minfree[6] = {
55	3 * 512,	/* 6MB */
56	2 * 1024,	/* 8MB */
57	4 * 1024,	/* 16MB */
58	16 * 1024,	/* 64MB */
59};
60static int lowmem_minfree_size = 4;
61
62static unsigned long lowmem_deathpending_timeout;
63
64#define lowmem_print(level, x...)			\
65	do {						\
66		if (lowmem_debug_level >= (level))	\
67			pr_info(x);			\
68	} while (0)
69
70static unsigned long lowmem_count(struct shrinker *s,
71				  struct shrink_control *sc)
72{
73	return global_page_state(NR_ACTIVE_ANON) +
74		global_page_state(NR_ACTIVE_FILE) +
75		global_page_state(NR_INACTIVE_ANON) +
76		global_page_state(NR_INACTIVE_FILE);
77}
78
79static unsigned long lowmem_scan(struct shrinker *s, struct shrink_control *sc)
80{
81	struct task_struct *tsk;
82	struct task_struct *selected = NULL;
83	unsigned long rem = 0;
84	int tasksize;
85	int i;
86	short min_score_adj = OOM_SCORE_ADJ_MAX + 1;
87	int selected_tasksize = 0;
88	short selected_oom_score_adj;
89	int array_size = ARRAY_SIZE(lowmem_adj);
90	int other_free = global_page_state(NR_FREE_PAGES) - totalreserve_pages;
91	int other_file = global_page_state(NR_FILE_PAGES) -
92						global_page_state(NR_SHMEM) -
93						total_swapcache_pages();
94
95	if (lowmem_adj_size < array_size)
96		array_size = lowmem_adj_size;
97	if (lowmem_minfree_size < array_size)
98		array_size = lowmem_minfree_size;
99	for (i = 0; i < array_size; i++) {
100		if (other_free < lowmem_minfree[i] &&
101		    other_file < lowmem_minfree[i]) {
102			min_score_adj = lowmem_adj[i];
103			break;
104		}
105	}
106
107	lowmem_print(3, "lowmem_scan %lu, %x, ofree %d %d, ma %hd\n",
108			sc->nr_to_scan, sc->gfp_mask, other_free,
109			other_file, min_score_adj);
110
111	if (min_score_adj == OOM_SCORE_ADJ_MAX + 1) {
112		lowmem_print(5, "lowmem_scan %lu, %x, return 0\n",
113			     sc->nr_to_scan, sc->gfp_mask);
114		return 0;
115	}
116
117	selected_oom_score_adj = min_score_adj;
118
119	rcu_read_lock();
120	for_each_process(tsk) {
121		struct task_struct *p;
122		short oom_score_adj;
123
124		if (tsk->flags & PF_KTHREAD)
125			continue;
126
127		p = find_lock_task_mm(tsk);
128		if (!p)
129			continue;
130
131		if (test_tsk_thread_flag(p, TIF_MEMDIE) &&
132		    time_before_eq(jiffies, lowmem_deathpending_timeout)) {
133			task_unlock(p);
134			rcu_read_unlock();
135			return 0;
136		}
137		oom_score_adj = p->signal->oom_score_adj;
138		if (oom_score_adj < min_score_adj) {
139			task_unlock(p);
140			continue;
141		}
142		tasksize = get_mm_rss(p->mm);
143		task_unlock(p);
144		if (tasksize <= 0)
145			continue;
146		if (selected) {
147			if (oom_score_adj < selected_oom_score_adj)
148				continue;
149			if (oom_score_adj == selected_oom_score_adj &&
150			    tasksize <= selected_tasksize)
151				continue;
152		}
153		selected = p;
154		selected_tasksize = tasksize;
155		selected_oom_score_adj = oom_score_adj;
156		lowmem_print(2, "select %d (%s), adj %hd, size %d, to kill\n",
157			     p->pid, p->comm, oom_score_adj, tasksize);
158	}
159	if (selected) {
160		task_lock(selected);
161		send_sig(SIGKILL, selected, 0);
162		/*
163		 * FIXME: lowmemorykiller shouldn't abuse global OOM killer
164		 * infrastructure. There is no real reason why the selected
165		 * task should have access to the memory reserves.
166		 */
167		if (selected->mm)
168			mark_oom_victim(selected);
169		task_unlock(selected);
170		lowmem_print(1, "send sigkill to %d (%s), adj %hd, size %d\n",
171			     selected->pid, selected->comm,
172			     selected_oom_score_adj, selected_tasksize);
173		lowmem_deathpending_timeout = jiffies + HZ;
174		rem += selected_tasksize;
175	}
176
177	lowmem_print(4, "lowmem_scan %lu, %x, return %lu\n",
178		     sc->nr_to_scan, sc->gfp_mask, rem);
179	rcu_read_unlock();
180	return rem;
181}
182
183static struct shrinker lowmem_shrinker = {
184	.scan_objects = lowmem_scan,
185	.count_objects = lowmem_count,
186	.seeks = DEFAULT_SEEKS * 16
187};
188
189static int __init lowmem_init(void)
190{
191	register_shrinker(&lowmem_shrinker);
192	return 0;
193}
194device_initcall(lowmem_init);
195
196/*
197 * not really modular, but the easiest way to keep compat with existing
198 * bootargs behaviour is to continue using module_param here.
199 */
200module_param_named(cost, lowmem_shrinker.seeks, int, S_IRUGO | S_IWUSR);
201module_param_array_named(adj, lowmem_adj, short, &lowmem_adj_size,
202			 S_IRUGO | S_IWUSR);
203module_param_array_named(minfree, lowmem_minfree, uint, &lowmem_minfree_size,
204			 S_IRUGO | S_IWUSR);
205module_param_named(debug_level, lowmem_debug_level, uint, S_IRUGO | S_IWUSR);
206
207