4 * Copyright (C) 1998,2000 Rik van Riel
5 * Thanks go out to Claus Fischer for some serious inspiration and
6 * for goading me into coding this file...
8 * The routines in this file are used to kill a process when
9 * we're seriously out of memory. This gets called from __alloc_pages()
10 * in mm/page_alloc.c when we really run out of memory.
12 * Since we won't call these routines often (on a well-configured
13 * machine) this file will double as a 'coding guide' and a signpost
14 * for newbie kernel hackers. It features several pointers to major
15 * kernel subsystems and hints as to where to find out what things do.
19 #include <linux/sched.h>
20 #include <linux/swap.h>
21 #include <linux/timex.h>
22 #include <linux/jiffies.h>
23 #include <linux/cpuset.h>
28 * oom_badness - calculate a numeric value for how bad this task has been
29 * @p: task struct of which task we should calculate
30 * @uptime: current uptime in seconds
32 * The formula used is relatively simple and documented inline in the
33 * function. The main rationale is that we want to select a good task
34 * to kill when we run out of memory.
36 * Good in this context means that:
37 * 1) we lose the minimum amount of work done
38 * 2) we recover a large amount of memory
39 * 3) we don't kill anything innocent of eating tons of memory
40 * 4) we want to kill the minimum amount of processes (one)
41 * 5) we try to kill the process the user expects us to kill, this
42 * algorithm has been meticulously tuned to meet the principle
43 * of least surprise ... (be careful when you change it)
46 unsigned long badness(struct task_struct *p, unsigned long uptime)
48 unsigned long points, cpu_time, run_time, s;
49 struct list_head *tsk;
55 * The memory size of the process is the basis for the badness.
57 points = p->mm->total_vm;
60 * Processes which fork a lot of child processes are likely
61 * a good choice. We add the vmsize of the children if they
62 * have an own mm. This prevents forking servers to flood the
63 * machine with an endless amount of children
65 list_for_each(tsk, &p->children) {
66 struct task_struct *chld;
67 chld = list_entry(tsk, struct task_struct, sibling);
68 if (chld->mm != p->mm && chld->mm)
69 points += chld->mm->total_vm;
73 * CPU time is in tens of seconds and run time is in thousands
74 * of seconds. There is no particular reason for this other than
75 * that it turned out to work very well in practice.
77 cpu_time = (cputime_to_jiffies(p->utime) + cputime_to_jiffies(p->stime))
80 if (uptime >= p->start_time.tv_sec)
81 run_time = (uptime - p->start_time.tv_sec) >> 10;
85 s = int_sqrt(cpu_time);
88 s = int_sqrt(int_sqrt(run_time));
93 * Niced processes are most likely less important, so double
94 * their badness points.
100 * Superuser processes are usually more important, so we make it
101 * less likely that we kill those.
103 if (cap_t(p->cap_effective) & CAP_TO_MASK(CAP_SYS_ADMIN) ||
104 p->uid == 0 || p->euid == 0)
108 * We don't want to kill a process with direct hardware access.
109 * Not only could that mess up the hardware, but usually users
110 * tend to only have this flag set on applications they think
113 if (cap_t(p->cap_effective) & CAP_TO_MASK(CAP_SYS_RAWIO))
117 * Adjust the score by oomkilladj.
120 if (p->oomkilladj > 0)
121 points <<= p->oomkilladj;
123 points >>= -(p->oomkilladj);
127 printk(KERN_DEBUG "OOMkill: task %d (%s) got %d points\n",
128 p->pid, p->comm, points);
134 * Simple selection loop. We chose the process with the highest
135 * number of 'points'. We expect the caller will lock the tasklist.
137 * (not docbooked, we don't want this one cluttering up the manual)
139 static struct task_struct * select_bad_process(void)
141 unsigned long maxpoints = 0;
142 struct task_struct *g, *p;
143 struct task_struct *chosen = NULL;
144 struct timespec uptime;
146 do_posix_clock_monotonic_gettime(&uptime);
147 do_each_thread(g, p) {
148 unsigned long points;
151 /* skip the init task with pid == 1 */
154 if (p->oomkilladj == OOM_DISABLE)
156 /* If p's nodes don't overlap ours, it won't help to kill p. */
157 if (!cpuset_excl_nodes_overlap(p))
161 * This is in the process of releasing memory so for wait it
162 * to finish before killing some other task by mistake.
164 releasing = test_tsk_thread_flag(p, TIF_MEMDIE) ||
165 p->flags & PF_EXITING;
166 if (releasing && !(p->flags & PF_DEAD))
167 return ERR_PTR(-1UL);
168 if (p->flags & PF_SWAPOFF)
171 points = badness(p, uptime.tv_sec);
172 if (points > maxpoints || !chosen) {
176 } while_each_thread(g, p);
181 * We must be careful though to never send SIGKILL a process with
182 * CAP_SYS_RAW_IO set, send SIGTERM instead (but it's unlikely that
183 * we select a process with CAP_SYS_RAW_IO set).
185 static void __oom_kill_task(task_t *p)
189 printk(KERN_WARNING "tried to kill init!\n");
194 if (!p->mm || p->mm == &init_mm) {
196 printk(KERN_WARNING "tried to kill an mm-less task!\n");
201 printk(KERN_ERR "Out of Memory: Killed process %d (%s).\n",
205 * We give our sacrificial lamb high priority and access to
206 * all the memory it needs. That way it should be able to
207 * exit() and clear out its resources quickly...
210 set_tsk_thread_flag(p, TIF_MEMDIE);
212 force_sig(SIGKILL, p);
215 static struct mm_struct *oom_kill_task(task_t *p)
217 struct mm_struct *mm = get_task_mm(p);
222 if (mm == &init_mm) {
229 * kill all processes that share the ->mm (i.e. all threads),
230 * but are in a different thread group
233 if (q->mm == mm && q->tgid != p->tgid)
235 while_each_thread(g, q);
240 static struct mm_struct *oom_kill_process(struct task_struct *p)
242 struct mm_struct *mm;
243 struct task_struct *c;
244 struct list_head *tsk;
246 /* Try to kill a child first */
247 list_for_each(tsk, &p->children) {
248 c = list_entry(tsk, struct task_struct, sibling);
251 mm = oom_kill_task(c);
255 return oom_kill_task(p);
259 * oom_kill - kill the "best" process when we run out of memory
261 * If we run out of memory, we have the choice between either
262 * killing a random task (bad), letting the system crash (worse)
263 * OR try to be smart about which process to kill. Note that we
264 * don't have to be perfect here, we just have to be good.
266 void out_of_memory(gfp_t gfp_mask, int order)
268 struct mm_struct *mm = NULL;
271 if (printk_ratelimit()) {
272 printk("oom-killer: gfp_mask=0x%x, order=%d\n",
277 read_lock(&tasklist_lock);
279 p = select_bad_process();
281 if (PTR_ERR(p) == -1UL)
284 /* Found nothing?!?! Either we hang forever, or we panic. */
286 read_unlock(&tasklist_lock);
287 panic("Out of memory and no killable processes...\n");
290 mm = oom_kill_process(p);
295 read_unlock(&tasklist_lock);
300 * Give "p" a good chance of killing itself before we
301 * retry to allocate memory.
303 schedule_timeout_interruptible(1);