lowmemorykiller.c revision 1eda5166c7640092f512138be6899d050c3d62ed 
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#include <linux/module.h>
34#include <linux/kernel.h>
35#include <linux/mm.h>
36#include <linux/oom.h>
37#include <linux/sched.h>
38#include <linux/rcupdate.h>
39#include <linux/profile.h>
40#include <linux/notifier.h>
41
42static uint32_t lowmem_debug_level = 2;
43static int lowmem_adj[6] = {
44	0,
45	1,
46	6,
47	12,
48};
49static int lowmem_adj_size = 4;
50static size_t lowmem_minfree[6] = {
51	3 * 512,	/* 6MB */
52	2 * 1024,	/* 8MB */
53	4 * 1024,	/* 16MB */
54	16 * 1024,	/* 64MB */
55};
56static int lowmem_minfree_size = 4;
57
58static struct task_struct *lowmem_deathpending;
59static unsigned long lowmem_deathpending_timeout;
60
61#define lowmem_print(level, x...)			\
62	do {						\
63		if (lowmem_debug_level >= (level))	\
64			printk(x);			\
65	} while (0)
66
67static int
68task_notify_func(struct notifier_block *self, unsigned long val, void *data);
69
70static struct notifier_block task_nb = {
71	.notifier_call	= task_notify_func,
72};
73
74static int
75task_notify_func(struct notifier_block *self, unsigned long val, void *data)
76{
77	struct task_struct *task = data;
78
79	if (task == lowmem_deathpending)
80		lowmem_deathpending = NULL;
81
82	return NOTIFY_OK;
83}
84
85static int lowmem_shrink(struct shrinker *s, struct shrink_control *sc)
86{
87	struct task_struct *tsk;
88	struct task_struct *selected = NULL;
89	int rem = 0;
90	int tasksize;
91	int i;
92	int min_score_adj = OOM_SCORE_ADJ_MAX + 1;
93	int selected_tasksize = 0;
94	int selected_oom_score_adj;
95	int array_size = ARRAY_SIZE(lowmem_adj);
96	int other_free = global_page_state(NR_FREE_PAGES);
97	int other_file = global_page_state(NR_FILE_PAGES) -
98						global_page_state(NR_SHMEM);
99
100	/*
101	 * If we already have a death outstanding, then
102	 * bail out right away; indicating to vmscan
103	 * that we have nothing further to offer on
104	 * this pass.
105	 *
106	 * Note: Currently you need CONFIG_PROFILING
107	 * for this to work correctly.
108	 */
109	if (lowmem_deathpending &&
110	    time_before_eq(jiffies, lowmem_deathpending_timeout))
111		return 0;
112
113	if (lowmem_adj_size < array_size)
114		array_size = lowmem_adj_size;
115	if (lowmem_minfree_size < array_size)
116		array_size = lowmem_minfree_size;
117	for (i = 0; i < array_size; i++) {
118		if (other_free < lowmem_minfree[i] &&
119		    other_file < lowmem_minfree[i]) {
120			min_score_adj = lowmem_adj[i];
121			break;
122		}
123	}
124	if (sc->nr_to_scan > 0)
125		lowmem_print(3, "lowmem_shrink %lu, %x, ofree %d %d, ma %d\n",
126				sc->nr_to_scan, sc->gfp_mask, other_free,
127				other_file, min_score_adj);
128	rem = global_page_state(NR_ACTIVE_ANON) +
129		global_page_state(NR_ACTIVE_FILE) +
130		global_page_state(NR_INACTIVE_ANON) +
131		global_page_state(NR_INACTIVE_FILE);
132	if (sc->nr_to_scan <= 0 || min_score_adj == OOM_SCORE_ADJ_MAX + 1) {
133		lowmem_print(5, "lowmem_shrink %lu, %x, return %d\n",
134			     sc->nr_to_scan, sc->gfp_mask, rem);
135		return rem;
136	}
137	selected_oom_score_adj = min_score_adj;
138
139	rcu_read_lock();
140	for_each_process(tsk) {
141		struct task_struct *p;
142		int oom_score_adj;
143
144		if (tsk->flags & PF_KTHREAD)
145			continue;
146
147		p = find_lock_task_mm(tsk);
148		if (!p)
149			continue;
150
151		oom_score_adj = p->signal->oom_score_adj;
152		if (oom_score_adj < min_score_adj) {
153			task_unlock(p);
154			continue;
155		}
156		tasksize = get_mm_rss(p->mm);
157		task_unlock(p);
158		if (tasksize <= 0)
159			continue;
160		if (selected) {
161			if (oom_score_adj < selected_oom_score_adj)
162				continue;
163			if (oom_score_adj == selected_oom_score_adj &&
164			    tasksize <= selected_tasksize)
165				continue;
166		}
167		selected = p;
168		selected_tasksize = tasksize;
169		selected_oom_score_adj = oom_score_adj;
170		lowmem_print(2, "select %d (%s), adj %d, size %d, to kill\n",
171			     p->pid, p->comm, oom_score_adj, tasksize);
172	}
173	if (selected) {
174		lowmem_print(1, "send sigkill to %d (%s), adj %d, size %d\n",
175			     selected->pid, selected->comm,
176			     selected_oom_score_adj, selected_tasksize);
177		/*
178		 * If CONFIG_PROFILING is off, then we don't want to stall
179		 * the killer by setting lowmem_deathpending.
180		 */
181#ifdef CONFIG_PROFILING
182		lowmem_deathpending = selected;
183		lowmem_deathpending_timeout = jiffies + HZ;
184#endif
185		send_sig(SIGKILL, selected, 0);
186		rem -= selected_tasksize;
187	}
188	lowmem_print(4, "lowmem_shrink %lu, %x, return %d\n",
189		     sc->nr_to_scan, sc->gfp_mask, rem);
190	rcu_read_unlock();
191	return rem;
192}
193
194static struct shrinker lowmem_shrinker = {
195	.shrink = lowmem_shrink,
196	.seeks = DEFAULT_SEEKS * 16
197};
198
199static int __init lowmem_init(void)
200{
201	task_handoff_register(&task_nb);
202	register_shrinker(&lowmem_shrinker);
203	return 0;
204}
205
206static void __exit lowmem_exit(void)
207{
208	unregister_shrinker(&lowmem_shrinker);
209	task_handoff_unregister(&task_nb);
210}
211
212module_param_named(cost, lowmem_shrinker.seeks, int, S_IRUGO | S_IWUSR);
213module_param_array_named(adj, lowmem_adj, int, &lowmem_adj_size,
214			 S_IRUGO | S_IWUSR);
215module_param_array_named(minfree, lowmem_minfree, uint, &lowmem_minfree_size,
216			 S_IRUGO | S_IWUSR);
217module_param_named(debug_level, lowmem_debug_level, uint, S_IRUGO | S_IWUSR);
218
219module_init(lowmem_init);
220module_exit(lowmem_exit);
221
222MODULE_LICENSE("GPL");
223
224