blob: c1850bf991cda9794f6985a7f6ebf595c6adac87 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * linux/mm/oom_kill.c
3 *
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...
7 *
8 * The routines in this file are used to kill a process when
Paul Jacksona49335c2005-09-06 15:18:09 -07009 * 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.
Linus Torvalds1da177e2005-04-16 15:20:36 -070011 *
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.
16 */
17
Alexey Dobriyan8ac773b2006-10-19 23:28:32 -070018#include <linux/oom.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070019#include <linux/mm.h>
Alexey Dobriyan4e950f62007-07-30 02:36:13 +040020#include <linux/err.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070021#include <linux/sched.h>
22#include <linux/swap.h>
23#include <linux/timex.h>
24#include <linux/jiffies.h>
Paul Jacksonef08e3b2005-09-06 15:18:13 -070025#include <linux/cpuset.h>
Martin Schwidefsky8bc719d2006-09-25 23:31:20 -070026#include <linux/module.h>
27#include <linux/notifier.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070028
KAMEZAWA Hiroyukifadd8fb2006-06-23 02:03:13 -070029int sysctl_panic_on_oom;
David Rientjesfe071d72007-10-16 23:25:56 -070030int sysctl_oom_kill_allocating_task;
David Rientjesae741382007-10-16 23:26:00 -070031static DEFINE_SPINLOCK(zone_scan_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -070032/* #define DEBUG */
33
34/**
Dave Peterson6937a252006-06-23 02:03:13 -070035 * badness - calculate a numeric value for how bad this task has been
Linus Torvalds1da177e2005-04-16 15:20:36 -070036 * @p: task struct of which task we should calculate
Paul Jacksona49335c2005-09-06 15:18:09 -070037 * @uptime: current uptime in seconds
Linus Torvalds1da177e2005-04-16 15:20:36 -070038 *
39 * The formula used is relatively simple and documented inline in the
40 * function. The main rationale is that we want to select a good task
41 * to kill when we run out of memory.
42 *
43 * Good in this context means that:
44 * 1) we lose the minimum amount of work done
45 * 2) we recover a large amount of memory
46 * 3) we don't kill anything innocent of eating tons of memory
47 * 4) we want to kill the minimum amount of processes (one)
48 * 5) we try to kill the process the user expects us to kill, this
49 * algorithm has been meticulously tuned to meet the principle
50 * of least surprise ... (be careful when you change it)
51 */
52
53unsigned long badness(struct task_struct *p, unsigned long uptime)
54{
55 unsigned long points, cpu_time, run_time, s;
Andrew Morton97c2c9b2006-04-18 22:20:38 -070056 struct mm_struct *mm;
57 struct task_struct *child;
Linus Torvalds1da177e2005-04-16 15:20:36 -070058
Andrew Morton97c2c9b2006-04-18 22:20:38 -070059 task_lock(p);
60 mm = p->mm;
61 if (!mm) {
62 task_unlock(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -070063 return 0;
Andrew Morton97c2c9b2006-04-18 22:20:38 -070064 }
Linus Torvalds1da177e2005-04-16 15:20:36 -070065
66 /*
67 * The memory size of the process is the basis for the badness.
68 */
Andrew Morton97c2c9b2006-04-18 22:20:38 -070069 points = mm->total_vm;
70
71 /*
72 * After this unlock we can no longer dereference local variable `mm'
73 */
74 task_unlock(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -070075
76 /*
Hugh Dickins7ba34852007-01-05 16:37:03 -080077 * swapoff can easily use up all memory, so kill those first.
78 */
79 if (p->flags & PF_SWAPOFF)
80 return ULONG_MAX;
81
82 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -070083 * Processes which fork a lot of child processes are likely
Kurt Garloff9827b782006-02-20 18:27:51 -080084 * a good choice. We add half the vmsize of the children if they
Linus Torvalds1da177e2005-04-16 15:20:36 -070085 * have an own mm. This prevents forking servers to flood the
Kurt Garloff9827b782006-02-20 18:27:51 -080086 * machine with an endless amount of children. In case a single
87 * child is eating the vast majority of memory, adding only half
88 * to the parents will make the child our kill candidate of choice.
Linus Torvalds1da177e2005-04-16 15:20:36 -070089 */
Andrew Morton97c2c9b2006-04-18 22:20:38 -070090 list_for_each_entry(child, &p->children, sibling) {
91 task_lock(child);
92 if (child->mm != mm && child->mm)
93 points += child->mm->total_vm/2 + 1;
94 task_unlock(child);
Linus Torvalds1da177e2005-04-16 15:20:36 -070095 }
96
97 /*
98 * CPU time is in tens of seconds and run time is in thousands
99 * of seconds. There is no particular reason for this other than
100 * that it turned out to work very well in practice.
101 */
102 cpu_time = (cputime_to_jiffies(p->utime) + cputime_to_jiffies(p->stime))
103 >> (SHIFT_HZ + 3);
104
105 if (uptime >= p->start_time.tv_sec)
106 run_time = (uptime - p->start_time.tv_sec) >> 10;
107 else
108 run_time = 0;
109
110 s = int_sqrt(cpu_time);
111 if (s)
112 points /= s;
113 s = int_sqrt(int_sqrt(run_time));
114 if (s)
115 points /= s;
116
117 /*
118 * Niced processes are most likely less important, so double
119 * their badness points.
120 */
121 if (task_nice(p) > 0)
122 points *= 2;
123
124 /*
125 * Superuser processes are usually more important, so we make it
126 * less likely that we kill those.
127 */
Serge E. Hallyn97829952008-02-04 22:29:47 -0800128 if (__capable(p, CAP_SYS_ADMIN) || __capable(p, CAP_SYS_RESOURCE))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129 points /= 4;
130
131 /*
132 * We don't want to kill a process with direct hardware access.
133 * Not only could that mess up the hardware, but usually users
134 * tend to only have this flag set on applications they think
135 * of as important.
136 */
Andrew Morgane338d262008-02-04 22:29:42 -0800137 if (__capable(p, CAP_SYS_RAWIO))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138 points /= 4;
139
140 /*
Nick Piggin7887a3d2006-09-25 23:31:29 -0700141 * If p's nodes don't overlap ours, it may still help to kill p
142 * because p may have allocated or otherwise mapped memory on
143 * this node before. However it will be less likely.
144 */
David Rientjesbbe373f2007-10-16 23:25:58 -0700145 if (!cpuset_mems_allowed_intersects(current, p))
Nick Piggin7887a3d2006-09-25 23:31:29 -0700146 points /= 8;
147
148 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149 * Adjust the score by oomkilladj.
150 */
151 if (p->oomkilladj) {
Joshua N Pritikin9a827822007-05-06 14:49:07 -0700152 if (p->oomkilladj > 0) {
153 if (!points)
154 points = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155 points <<= p->oomkilladj;
Joshua N Pritikin9a827822007-05-06 14:49:07 -0700156 } else
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157 points >>= -(p->oomkilladj);
158 }
159
160#ifdef DEBUG
David Rientjesa5e58a62007-07-31 00:37:32 -0700161 printk(KERN_DEBUG "OOMkill: task %d (%s) got %lu points\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162 p->pid, p->comm, points);
163#endif
164 return points;
165}
166
167/*
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800168 * Determine the type of allocation constraint.
169 */
David Rientjes70e24bd2007-10-16 23:25:53 -0700170static inline enum oom_constraint constrained_alloc(struct zonelist *zonelist,
171 gfp_t gfp_mask)
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800172{
173#ifdef CONFIG_NUMA
174 struct zone **z;
Christoph Lameteree31af52007-10-16 01:25:30 -0700175 nodemask_t nodes = node_states[N_HIGH_MEMORY];
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800176
177 for (z = zonelist->zones; *z; z++)
Paul Jackson02a0e532006-12-13 00:34:25 -0800178 if (cpuset_zone_allowed_softwall(*z, gfp_mask))
Christoph Lameter89fa3022006-09-25 23:31:55 -0700179 node_clear(zone_to_nid(*z), nodes);
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800180 else
181 return CONSTRAINT_CPUSET;
182
183 if (!nodes_empty(nodes))
184 return CONSTRAINT_MEMORY_POLICY;
185#endif
186
187 return CONSTRAINT_NONE;
188}
189
190/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191 * Simple selection loop. We chose the process with the highest
192 * number of 'points'. We expect the caller will lock the tasklist.
193 *
194 * (not docbooked, we don't want this one cluttering up the manual)
195 */
Kurt Garloff9827b782006-02-20 18:27:51 -0800196static struct task_struct *select_bad_process(unsigned long *ppoints)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700198 struct task_struct *g, *p;
199 struct task_struct *chosen = NULL;
200 struct timespec uptime;
Kurt Garloff9827b782006-02-20 18:27:51 -0800201 *ppoints = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202
203 do_posix_clock_monotonic_gettime(&uptime);
Paul Jacksona49335c2005-09-06 15:18:09 -0700204 do_each_thread(g, p) {
205 unsigned long points;
Paul Jacksona49335c2005-09-06 15:18:09 -0700206
Oleg Nesterov28324d12006-09-29 02:01:12 -0700207 /*
208 * skip kernel threads and tasks which have already released
209 * their mm.
210 */
Nick Piggin5081dde2006-09-25 23:31:32 -0700211 if (!p->mm)
212 continue;
Oleg Nesterov28324d12006-09-29 02:01:12 -0700213 /* skip the init task */
Serge E. Hallynb460cbc2007-10-18 23:39:52 -0700214 if (is_global_init(p))
Paul Jacksona49335c2005-09-06 15:18:09 -0700215 continue;
Paul Jacksonef08e3b2005-09-06 15:18:13 -0700216
Paul Jacksona49335c2005-09-06 15:18:09 -0700217 /*
Nick Pigginb78483a2006-09-29 02:01:14 -0700218 * This task already has access to memory reserves and is
219 * being killed. Don't allow any other task access to the
220 * memory reserve.
221 *
222 * Note: this may have a chance of deadlock if it gets
223 * blocked waiting for another task which itself is waiting
224 * for memory. Is there a better alternative?
225 */
226 if (test_tsk_thread_flag(p, TIF_MEMDIE))
227 return ERR_PTR(-1UL);
228
229 /*
Dave Peterson6937a252006-06-23 02:03:13 -0700230 * This is in the process of releasing memory so wait for it
Paul Jacksona49335c2005-09-06 15:18:09 -0700231 * to finish before killing some other task by mistake.
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700232 *
233 * However, if p is the current task, we allow the 'kill' to
234 * go ahead if it is exiting: this will simply set TIF_MEMDIE,
235 * which will allow it to gain access to memory reserves in
236 * the process of exiting and releasing its resources.
Nick Pigginb78483a2006-09-29 02:01:14 -0700237 * Otherwise we could get an easy OOM deadlock.
Paul Jacksona49335c2005-09-06 15:18:09 -0700238 */
Nick Pigginb78483a2006-09-29 02:01:14 -0700239 if (p->flags & PF_EXITING) {
240 if (p != current)
241 return ERR_PTR(-1UL);
242
Oleg Nesterov972c4ea2006-09-29 02:01:12 -0700243 chosen = p;
244 *ppoints = ULONG_MAX;
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700245 }
Oleg Nesterov972c4ea2006-09-29 02:01:12 -0700246
Nick Piggin4a3ede12006-09-25 23:31:30 -0700247 if (p->oomkilladj == OOM_DISABLE)
248 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249
Paul Jacksona49335c2005-09-06 15:18:09 -0700250 points = badness(p, uptime.tv_sec);
Kurt Garloff9827b782006-02-20 18:27:51 -0800251 if (points > *ppoints || !chosen) {
Paul Jacksona49335c2005-09-06 15:18:09 -0700252 chosen = p;
Kurt Garloff9827b782006-02-20 18:27:51 -0800253 *ppoints = points;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254 }
Paul Jacksona49335c2005-09-06 15:18:09 -0700255 } while_each_thread(g, p);
Oleg Nesterov972c4ea2006-09-29 02:01:12 -0700256
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257 return chosen;
258}
259
260/**
Ram Gupta5a291b982006-09-25 23:31:54 -0700261 * Send SIGKILL to the selected process irrespective of CAP_SYS_RAW_IO
262 * flag though it's unlikely that we select a process with CAP_SYS_RAW_IO
263 * set.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264 */
Nick Pigginf3af38d2006-12-06 20:31:51 -0800265static void __oom_kill_task(struct task_struct *p, int verbose)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700266{
Serge E. Hallynb460cbc2007-10-18 23:39:52 -0700267 if (is_global_init(p)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268 WARN_ON(1);
269 printk(KERN_WARNING "tried to kill init!\n");
270 return;
271 }
272
Oleg Nesterov01017a22006-09-29 02:01:13 -0700273 if (!p->mm) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700274 WARN_ON(1);
275 printk(KERN_WARNING "tried to kill an mm-less task!\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276 return;
277 }
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700278
Nick Pigginf3af38d2006-12-06 20:31:51 -0800279 if (verbose)
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -0700280 printk(KERN_ERR "Killed process %d (%s)\n",
281 task_pid_nr(p), p->comm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282
283 /*
284 * We give our sacrificial lamb high priority and access to
285 * all the memory it needs. That way it should be able to
286 * exit() and clear out its resources quickly...
287 */
Peter Zijlstrafa717062008-01-25 21:08:27 +0100288 p->rt.time_slice = HZ;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700289 set_tsk_thread_flag(p, TIF_MEMDIE);
290
291 force_sig(SIGKILL, p);
292}
293
Nick Pigginf3af38d2006-12-06 20:31:51 -0800294static int oom_kill_task(struct task_struct *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295{
Dave Peterson01315922006-04-18 22:20:44 -0700296 struct mm_struct *mm;
Ingo Molnar36c8b582006-07-03 00:25:41 -0700297 struct task_struct *g, *q;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700298
Dave Peterson01315922006-04-18 22:20:44 -0700299 mm = p->mm;
300
301 /* WARNING: mm may not be dereferenced since we did not obtain its
302 * value from get_task_mm(p). This is OK since all we need to do is
303 * compare mm to q->mm below.
304 *
305 * Furthermore, even if mm contains a non-NULL value, p->mm may
306 * change to NULL at any time since we do not hold task_lock(p).
307 * However, this is of no concern to us.
308 */
309
Oleg Nesterov01017a22006-09-29 02:01:13 -0700310 if (mm == NULL)
Dave Peterson01315922006-04-18 22:20:44 -0700311 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312
Nick Pigginc33e0fc2006-12-06 20:31:50 -0800313 /*
314 * Don't kill the process if any threads are set to OOM_DISABLE
315 */
316 do_each_thread(g, q) {
Ankita Garg35ae8342007-03-16 13:38:33 -0800317 if (q->mm == mm && q->oomkilladj == OOM_DISABLE)
Nick Pigginc33e0fc2006-12-06 20:31:50 -0800318 return 1;
319 } while_each_thread(g, q);
320
Nick Pigginf3af38d2006-12-06 20:31:51 -0800321 __oom_kill_task(p, 1);
Nick Pigginc33e0fc2006-12-06 20:31:50 -0800322
Linus Torvalds1da177e2005-04-16 15:20:36 -0700323 /*
324 * kill all processes that share the ->mm (i.e. all threads),
Nick Pigginf2a2a712006-12-06 20:31:52 -0800325 * but are in a different thread group. Don't let them have access
326 * to memory reserves though, otherwise we might deplete all memory.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327 */
Nick Pigginc33e0fc2006-12-06 20:31:50 -0800328 do_each_thread(g, q) {
Pavel Emelyanovbac0abd2007-10-18 23:40:18 -0700329 if (q->mm == mm && !same_thread_group(q, p))
David Rientjes650a7c92007-04-23 21:36:13 -0700330 force_sig(SIGKILL, q);
Nick Pigginc33e0fc2006-12-06 20:31:50 -0800331 } while_each_thread(g, q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332
Dave Peterson01315922006-04-18 22:20:44 -0700333 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334}
335
David Rientjes7213f502007-10-16 23:25:57 -0700336static int oom_kill_process(struct task_struct *p, gfp_t gfp_mask, int order,
337 unsigned long points, const char *message)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339 struct task_struct *c;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340
David Rientjes7213f502007-10-16 23:25:57 -0700341 if (printk_ratelimit()) {
342 printk(KERN_WARNING "%s invoked oom-killer: "
343 "gfp_mask=0x%x, order=%d, oomkilladj=%d\n",
344 current->comm, gfp_mask, order, current->oomkilladj);
345 dump_stack();
346 show_mem();
347 }
348
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700349 /*
350 * If the task is already exiting, don't alarm the sysadmin or kill
351 * its children or threads, just set TIF_MEMDIE so it can die quickly
352 */
353 if (p->flags & PF_EXITING) {
Nick Pigginf3af38d2006-12-06 20:31:51 -0800354 __oom_kill_task(p, 0);
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700355 return 0;
356 }
357
Nick Pigginf3af38d2006-12-06 20:31:51 -0800358 printk(KERN_ERR "%s: kill process %d (%s) score %li or a child\n",
Pavel Emelyanovba25f9d2007-10-18 23:40:40 -0700359 message, task_pid_nr(p), p->comm, points);
Nick Pigginf3af38d2006-12-06 20:31:51 -0800360
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361 /* Try to kill a child first */
Matthias Kaehlcke7b1915a2007-10-18 23:39:58 -0700362 list_for_each_entry(c, &p->children, sibling) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363 if (c->mm == p->mm)
364 continue;
Nick Pigginf3af38d2006-12-06 20:31:51 -0800365 if (!oom_kill_task(c))
Dave Peterson01315922006-04-18 22:20:44 -0700366 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367 }
Nick Pigginf3af38d2006-12-06 20:31:51 -0800368 return oom_kill_task(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700369}
370
Martin Schwidefsky8bc719d2006-09-25 23:31:20 -0700371static BLOCKING_NOTIFIER_HEAD(oom_notify_list);
372
373int register_oom_notifier(struct notifier_block *nb)
374{
375 return blocking_notifier_chain_register(&oom_notify_list, nb);
376}
377EXPORT_SYMBOL_GPL(register_oom_notifier);
378
379int unregister_oom_notifier(struct notifier_block *nb)
380{
381 return blocking_notifier_chain_unregister(&oom_notify_list, nb);
382}
383EXPORT_SYMBOL_GPL(unregister_oom_notifier);
384
David Rientjes098d7f12007-10-16 23:25:55 -0700385/*
386 * Try to acquire the OOM killer lock for the zones in zonelist. Returns zero
387 * if a parallel OOM killing is already taking place that includes a zone in
388 * the zonelist. Otherwise, locks all zones in the zonelist and returns 1.
389 */
390int try_set_zone_oom(struct zonelist *zonelist)
391{
392 struct zone **z;
393 int ret = 1;
394
395 z = zonelist->zones;
396
David Rientjesae741382007-10-16 23:26:00 -0700397 spin_lock(&zone_scan_mutex);
David Rientjes098d7f12007-10-16 23:25:55 -0700398 do {
399 if (zone_is_oom_locked(*z)) {
400 ret = 0;
401 goto out;
402 }
403 } while (*(++z) != NULL);
404
405 /*
406 * Lock each zone in the zonelist under zone_scan_mutex so a parallel
407 * invocation of try_set_zone_oom() doesn't succeed when it shouldn't.
408 */
409 z = zonelist->zones;
410 do {
411 zone_set_flag(*z, ZONE_OOM_LOCKED);
412 } while (*(++z) != NULL);
413out:
David Rientjesae741382007-10-16 23:26:00 -0700414 spin_unlock(&zone_scan_mutex);
David Rientjes098d7f12007-10-16 23:25:55 -0700415 return ret;
416}
417
418/*
419 * Clears the ZONE_OOM_LOCKED flag for all zones in the zonelist so that failed
420 * allocation attempts with zonelists containing them may now recall the OOM
421 * killer, if necessary.
422 */
423void clear_zonelist_oom(struct zonelist *zonelist)
424{
425 struct zone **z;
426
427 z = zonelist->zones;
428
David Rientjesae741382007-10-16 23:26:00 -0700429 spin_lock(&zone_scan_mutex);
David Rientjes098d7f12007-10-16 23:25:55 -0700430 do {
431 zone_clear_flag(*z, ZONE_OOM_LOCKED);
432 } while (*(++z) != NULL);
David Rientjesae741382007-10-16 23:26:00 -0700433 spin_unlock(&zone_scan_mutex);
David Rientjes098d7f12007-10-16 23:25:55 -0700434}
435
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436/**
Dave Peterson6937a252006-06-23 02:03:13 -0700437 * out_of_memory - kill the "best" process when we run out of memory
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438 *
439 * If we run out of memory, we have the choice between either
440 * killing a random task (bad), letting the system crash (worse)
441 * OR try to be smart about which process to kill. Note that we
442 * don't have to be perfect here, we just have to be good.
443 */
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800444void out_of_memory(struct zonelist *zonelist, gfp_t gfp_mask, int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445{
Ingo Molnar36c8b582006-07-03 00:25:41 -0700446 struct task_struct *p;
Andrew Mortond6713e02006-02-28 16:59:19 -0800447 unsigned long points = 0;
Martin Schwidefsky8bc719d2006-09-25 23:31:20 -0700448 unsigned long freed = 0;
David Rientjes70e24bd2007-10-16 23:25:53 -0700449 enum oom_constraint constraint;
Martin Schwidefsky8bc719d2006-09-25 23:31:20 -0700450
451 blocking_notifier_call_chain(&oom_notify_list, 0, &freed);
452 if (freed > 0)
453 /* Got some memory back in the last second. */
454 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455
Yasunori Goto2b744c02007-05-06 14:49:59 -0700456 if (sysctl_panic_on_oom == 2)
457 panic("out of memory. Compulsory panic_on_oom is selected.\n");
458
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800459 /*
460 * Check if there were limitations on the allocation (only relevant for
461 * NUMA) that may require different handling.
462 */
David Rientjes2b45ab32007-05-06 14:50:00 -0700463 constraint = constrained_alloc(zonelist, gfp_mask);
David Rientjes2b45ab32007-05-06 14:50:00 -0700464 read_lock(&tasklist_lock);
465
466 switch (constraint) {
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800467 case CONSTRAINT_MEMORY_POLICY:
David Rientjes7213f502007-10-16 23:25:57 -0700468 oom_kill_process(current, gfp_mask, order, points,
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800469 "No available memory (MPOL_BIND)");
470 break;
471
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800472 case CONSTRAINT_NONE:
KAMEZAWA Hiroyukifadd8fb2006-06-23 02:03:13 -0700473 if (sysctl_panic_on_oom)
474 panic("out of memory. panic_on_oom is selected\n");
David Rientjesfe071d72007-10-16 23:25:56 -0700475 /* Fall-through */
476 case CONSTRAINT_CPUSET:
477 if (sysctl_oom_kill_allocating_task) {
David Rientjes7213f502007-10-16 23:25:57 -0700478 oom_kill_process(current, gfp_mask, order, points,
David Rientjesfe071d72007-10-16 23:25:56 -0700479 "Out of memory (oom_kill_allocating_task)");
480 break;
481 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482retry:
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800483 /*
484 * Rambo mode: Shoot down a process and hope it solves whatever
485 * issues we may have.
486 */
487 p = select_bad_process(&points);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800489 if (PTR_ERR(p) == -1UL)
490 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800492 /* Found nothing?!?! Either we hang forever, or we panic. */
493 if (!p) {
494 read_unlock(&tasklist_lock);
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800495 panic("Out of memory and no killable processes...\n");
496 }
497
Al Viroe91a8102007-10-20 21:45:33 +0100498 if (oom_kill_process(p, gfp_mask, order, points,
David Rientjes7213f502007-10-16 23:25:57 -0700499 "Out of memory"))
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800500 goto retry;
501
502 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700503 }
504
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800505out:
Andrew Morton140ffce2006-03-02 02:54:28 -0800506 read_unlock(&tasklist_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507
508 /*
509 * Give "p" a good chance of killing itself before we
Kirill Korotaev2f659f42006-01-08 01:01:05 -0800510 * retry to allocate memory unless "p" is current
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511 */
Kirill Korotaev2f659f42006-01-08 01:01:05 -0800512 if (!test_thread_flag(TIF_MEMDIE))
Andrew Morton140ffce2006-03-02 02:54:28 -0800513 schedule_timeout_uninterruptible(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700514}