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/module.h>
36#include <linux/kernel.h>
37#include <linux/mm.h>
38#include <linux/oom.h>
39#include <linux/sched.h>
40#include <linux/swap.h>
41#include <linux/rcupdate.h>
42#include <linux/profile.h>
43#include <linux/notifier.h>
44
45static uint32_t lowmem_debug_level = 1;
46static short lowmem_adj[6] = {
47	0,
48	1,
49	6,
50	12,
51};
52static int lowmem_adj_size = 4;
53static int lowmem_minfree[6] = {
54	3 * 512,	/* 6MB */
55	2 * 1024,	/* 8MB */
56	4 * 1024,	/* 16MB */
57	16 * 1024,	/* 64MB */
58};
59static int lowmem_minfree_size = 4;
60
61static unsigned long lowmem_deathpending_timeout;
62
63#define lowmem_print(level, x...)			\
64	do {						\
65		if (lowmem_debug_level >= (level))	\
66			pr_info(x);			\
67	} while (0)
68
69static unsigned long lowmem_count(struct shrinker *s,
70				  struct shrink_control *sc)
71{
72	return global_page_state(NR_ACTIVE_ANON) +
73		global_page_state(NR_ACTIVE_FILE) +
74		global_page_state(NR_INACTIVE_ANON) +
75		global_page_state(NR_INACTIVE_FILE);
76}
77
78static unsigned long lowmem_scan(struct shrinker *s, struct shrink_control *sc)
79{
80	struct task_struct *tsk;
81	struct task_struct *selected = NULL;
82	unsigned long rem = 0;
83	int tasksize;
84	int i;
85	short min_score_adj = OOM_SCORE_ADJ_MAX + 1;
86	int selected_tasksize = 0;
87	short selected_oom_score_adj;
88	int array_size = ARRAY_SIZE(lowmem_adj);
89	int other_free = global_page_state(NR_FREE_PAGES) - totalreserve_pages;
90	int other_file = global_page_state(NR_FILE_PAGES) -
91						global_page_state(NR_SHMEM) -
92						total_swapcache_pages();
93
94	if (lowmem_adj_size < array_size)
95		array_size = lowmem_adj_size;
96	if (lowmem_minfree_size < array_size)
97		array_size = lowmem_minfree_size;
98	for (i = 0; i < array_size; i++) {
99		if (other_free < lowmem_minfree[i] &&
100		    other_file < lowmem_minfree[i]) {
101			min_score_adj = lowmem_adj[i];
102			break;
103		}
104	}
105
106	lowmem_print(3, "lowmem_scan %lu, %x, ofree %d %d, ma %hd\n",
107			sc->nr_to_scan, sc->gfp_mask, other_free,
108			other_file, min_score_adj);
109
110	if (min_score_adj == OOM_SCORE_ADJ_MAX + 1) {
111		lowmem_print(5, "lowmem_scan %lu, %x, return 0\n",
112			     sc->nr_to_scan, sc->gfp_mask);
113		return 0;
114	}
115
116	selected_oom_score_adj = min_score_adj;
117
118	rcu_read_lock();
119	for_each_process(tsk) {
120		struct task_struct *p;
121		short oom_score_adj;
122
123		if (tsk->flags & PF_KTHREAD)
124			continue;
125
126		p = find_lock_task_mm(tsk);
127		if (!p)
128			continue;
129
130		if (test_tsk_thread_flag(p, TIF_MEMDIE) &&
131		    time_before_eq(jiffies, lowmem_deathpending_timeout)) {
132			task_unlock(p);
133			rcu_read_unlock();
134			return 0;
135		}
136		oom_score_adj = p->signal->oom_score_adj;
137		if (oom_score_adj < min_score_adj) {
138			task_unlock(p);
139			continue;
140		}
141		tasksize = get_mm_rss(p->mm);
142		task_unlock(p);
143		if (tasksize <= 0)
144			continue;
145		if (selected) {
146			if (oom_score_adj < selected_oom_score_adj)
147				continue;
148			if (oom_score_adj == selected_oom_score_adj &&
149			    tasksize <= selected_tasksize)
150				continue;
151		}
152		selected = p;
153		selected_tasksize = tasksize;
154		selected_oom_score_adj = oom_score_adj;
155		lowmem_print(2, "select %d (%s), adj %hd, size %d, to kill\n",
156			     p->pid, p->comm, oom_score_adj, tasksize);
157	}
158	if (selected) {
159		lowmem_print(1, "send sigkill to %d (%s), adj %hd, size %d\n",
160			     selected->pid, selected->comm,
161			     selected_oom_score_adj, selected_tasksize);
162		lowmem_deathpending_timeout = jiffies + HZ;
163		/*
164		 * FIXME: lowmemorykiller shouldn't abuse global OOM killer
165		 * infrastructure. There is no real reason why the selected
166		 * task should have access to the memory reserves.
167		 */
168		mark_tsk_oom_victim(selected);
169		send_sig(SIGKILL, selected, 0);
170		rem += selected_tasksize;
171	}
172
173	lowmem_print(4, "lowmem_scan %lu, %x, return %lu\n",
174		     sc->nr_to_scan, sc->gfp_mask, rem);
175	rcu_read_unlock();
176	return rem;
177}
178
179static struct shrinker lowmem_shrinker = {
180	.scan_objects = lowmem_scan,
181	.count_objects = lowmem_count,
182	.seeks = DEFAULT_SEEKS * 16
183};
184
185static int __init lowmem_init(void)
186{
187	register_shrinker(&lowmem_shrinker);
188	return 0;
189}
190
191static void __exit lowmem_exit(void)
192{
193	unregister_shrinker(&lowmem_shrinker);
194}
195
196module_param_named(cost, lowmem_shrinker.seeks, int, S_IRUGO | S_IWUSR);
197module_param_array_named(adj, lowmem_adj, short, &lowmem_adj_size,
198			 S_IRUGO | S_IWUSR);
199module_param_array_named(minfree, lowmem_minfree, uint, &lowmem_minfree_size,
200			 S_IRUGO | S_IWUSR);
201module_param_named(debug_level, lowmem_debug_level, uint, S_IRUGO | S_IWUSR);
202
203module_init(lowmem_init);
204module_exit(lowmem_exit);
205
206MODULE_LICENSE("GPL");
207
208