lowmemorykiller.c revision 9823ec9ded836ee7ca4fe5ab7964b9cdc8af010c 
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_adj values will get killed. Specify the
5 * minimum oom_adj values in /sys/module/lowmemorykiller/parameters/adj and the
6 * number of free pages in /sys/module/lowmemorykiller/parameters/minfree. Both
7 * files take a comma separated list of numbers in ascending order.
8 *
9 * For example, write "0,8" to /sys/module/lowmemorykiller/parameters/adj and
10 * "1024,4096" to /sys/module/lowmemorykiller/parameters/minfree to kill
11 * processes with a oom_adj value of 8 or higher when the free memory drops
12 * below 4096 pages and kill processes with a oom_adj value of 0 or higher
13 * when the free memory drops below 1024 pages.
14 *
15 * The driver considers memory used for caches to be free, but if a large
16 * percentage of the cached memory is locked this can be very inaccurate
17 * and processes may not get killed until the normal oom killer is triggered.
18 *
19 * Copyright (C) 2007-2008 Google, Inc.
20 *
21 * This software is licensed under the terms of the GNU General Public
22 * License version 2, as published by the Free Software Foundation, and
23 * may be copied, distributed, and modified under those terms.
24 *
25 * This program is distributed in the hope that it will be useful,
26 * but WITHOUT ANY WARRANTY; without even the implied warranty of
27 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
28 * GNU General Public License for more details.
29 *
30 */
31
32#include <linux/module.h>
33#include <linux/kernel.h>
34#include <linux/mm.h>
35#include <linux/oom.h>
36#include <linux/sched.h>
37#include <linux/rcupdate.h>
38#include <linux/profile.h>
39#include <linux/notifier.h>
40
41static uint32_t lowmem_debug_level = 2;
42static int lowmem_adj[6] = {
43	0,
44	1,
45	6,
46	12,
47};
48static int lowmem_adj_size = 4;
49static size_t lowmem_minfree[6] = {
50	3 * 512,	/* 6MB */
51	2 * 1024,	/* 8MB */
52	4 * 1024,	/* 16MB */
53	16 * 1024,	/* 64MB */
54};
55static int lowmem_minfree_size = 4;
56
57static struct task_struct *lowmem_deathpending;
58
59#define lowmem_print(level, x...)			\
60	do {						\
61		if (lowmem_debug_level >= (level))	\
62			printk(x);			\
63	} while (0)
64
65static int
66task_notify_func(struct notifier_block *self, unsigned long val, void *data);
67
68static struct notifier_block task_nb = {
69	.notifier_call	= task_notify_func,
70};
71
72static int
73task_notify_func(struct notifier_block *self, unsigned long val, void *data)
74{
75	struct task_struct *task = data;
76	if (task == lowmem_deathpending) {
77		lowmem_deathpending = NULL;
78		task_handoff_unregister(&task_nb);
79	}
80	return NOTIFY_OK;
81}
82
83static int lowmem_shrink(struct shrinker *s, struct shrink_control *sc)
84{
85	struct task_struct *tsk;
86	struct task_struct *selected = NULL;
87	int rem = 0;
88	int tasksize;
89	int i;
90	int min_adj = OOM_ADJUST_MAX + 1;
91	int selected_tasksize = 0;
92	int selected_oom_adj;
93	int array_size = ARRAY_SIZE(lowmem_adj);
94	int other_free = global_page_state(NR_FREE_PAGES);
95	int other_file = global_page_state(NR_FILE_PAGES) -
96						global_page_state(NR_SHMEM);
97
98	/*
99	 * If we already have a death outstanding, then
100	 * bail out right away; indicating to vmscan
101	 * that we have nothing further to offer on
102	 * this pass.
103	 *
104	 * Note: Currently you need CONFIG_PROFILING
105	 * for this to work correctly.
106	 */
107	if (lowmem_deathpending)
108		return 0;
109
110	if (lowmem_adj_size < array_size)
111		array_size = lowmem_adj_size;
112	if (lowmem_minfree_size < array_size)
113		array_size = lowmem_minfree_size;
114	for (i = 0; i < array_size; i++) {
115		if (other_free < lowmem_minfree[i] &&
116		    other_file < lowmem_minfree[i]) {
117			min_adj = lowmem_adj[i];
118			break;
119		}
120	}
121	if (sc->nr_to_scan > 0)
122		lowmem_print(3, "lowmem_shrink %lu, %x, ofree %d %d, ma %d\n",
123				sc->nr_to_scan, sc->gfp_mask, other_free,
124				other_file, min_adj);
125	rem = global_page_state(NR_ACTIVE_ANON) +
126		global_page_state(NR_ACTIVE_FILE) +
127		global_page_state(NR_INACTIVE_ANON) +
128		global_page_state(NR_INACTIVE_FILE);
129	if (sc->nr_to_scan <= 0 || min_adj == OOM_ADJUST_MAX + 1) {
130		lowmem_print(5, "lowmem_shrink %lu, %x, return %d\n",
131			     sc->nr_to_scan, sc->gfp_mask, rem);
132		return rem;
133	}
134	selected_oom_adj = min_adj;
135
136	rcu_read_lock();
137	for_each_process(tsk) {
138		struct task_struct *p;
139		int oom_adj;
140
141		if (tsk->flags & PF_KTHREAD)
142			continue;
143
144		p = find_lock_task_mm(tsk);
145		if (!p)
146			continue;
147
148		oom_adj = p->signal->oom_adj;
149		if (oom_adj < min_adj) {
150			task_unlock(p);
151			continue;
152		}
153		tasksize = get_mm_rss(p->mm);
154		task_unlock(p);
155		if (tasksize <= 0)
156			continue;
157		if (selected) {
158			if (oom_adj < selected_oom_adj)
159				continue;
160			if (oom_adj == selected_oom_adj &&
161			    tasksize <= selected_tasksize)
162				continue;
163		}
164		selected = p;
165		selected_tasksize = tasksize;
166		selected_oom_adj = oom_adj;
167		lowmem_print(2, "select %d (%s), adj %d, size %d, to kill\n",
168			     p->pid, p->comm, oom_adj, tasksize);
169	}
170	if (selected) {
171		lowmem_print(1, "send sigkill to %d (%s), adj %d, size %d\n",
172			     selected->pid, selected->comm,
173			     selected_oom_adj, selected_tasksize);
174		/*
175		 * If CONFIG_PROFILING is off, then task_handoff_register()
176		 * is a nop. In that case we don't want to stall the killer
177		 * by setting lowmem_deathpending.
178		 */
179#ifdef CONFIG_PROFILING
180		lowmem_deathpending = selected;
181		task_handoff_register(&task_nb);
182#endif
183		send_sig(SIGKILL, selected, 0);
184		rem -= selected_tasksize;
185	}
186	lowmem_print(4, "lowmem_shrink %lu, %x, return %d\n",
187		     sc->nr_to_scan, sc->gfp_mask, rem);
188	rcu_read_unlock();
189	return rem;
190}
191
192static struct shrinker lowmem_shrinker = {
193	.shrink = lowmem_shrink,
194	.seeks = DEFAULT_SEEKS * 16
195};
196
197static int __init lowmem_init(void)
198{
199	register_shrinker(&lowmem_shrinker);
200	return 0;
201}
202
203static void __exit lowmem_exit(void)
204{
205	unregister_shrinker(&lowmem_shrinker);
206}
207
208module_param_named(cost, lowmem_shrinker.seeks, int, S_IRUGO | S_IWUSR);
209module_param_array_named(adj, lowmem_adj, int, &lowmem_adj_size,
210			 S_IRUGO | S_IWUSR);
211module_param_array_named(minfree, lowmem_minfree, uint, &lowmem_minfree_size,
212			 S_IRUGO | S_IWUSR);
213module_param_named(debug_level, lowmem_debug_level, uint, S_IRUGO | S_IWUSR);
214
215module_init(lowmem_init);
216module_exit(lowmem_exit);
217
218MODULE_LICENSE("GPL");
219
220