blob: 41b4e362221d943ba1759d9de3d740222f067c98 [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;
Linus Torvalds1da177e2005-04-16 15:20:36 -070030/* #define DEBUG */
31
32/**
Dave Peterson6937a252006-06-23 02:03:13 -070033 * badness - calculate a numeric value for how bad this task has been
Linus Torvalds1da177e2005-04-16 15:20:36 -070034 * @p: task struct of which task we should calculate
Paul Jacksona49335c2005-09-06 15:18:09 -070035 * @uptime: current uptime in seconds
Linus Torvalds1da177e2005-04-16 15:20:36 -070036 *
37 * The formula used is relatively simple and documented inline in the
38 * function. The main rationale is that we want to select a good task
39 * to kill when we run out of memory.
40 *
41 * Good in this context means that:
42 * 1) we lose the minimum amount of work done
43 * 2) we recover a large amount of memory
44 * 3) we don't kill anything innocent of eating tons of memory
45 * 4) we want to kill the minimum amount of processes (one)
46 * 5) we try to kill the process the user expects us to kill, this
47 * algorithm has been meticulously tuned to meet the principle
48 * of least surprise ... (be careful when you change it)
49 */
50
51unsigned long badness(struct task_struct *p, unsigned long uptime)
52{
53 unsigned long points, cpu_time, run_time, s;
Andrew Morton97c2c9b82006-04-18 22:20:38 -070054 struct mm_struct *mm;
55 struct task_struct *child;
Linus Torvalds1da177e2005-04-16 15:20:36 -070056
Andrew Morton97c2c9b82006-04-18 22:20:38 -070057 task_lock(p);
58 mm = p->mm;
59 if (!mm) {
60 task_unlock(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -070061 return 0;
Andrew Morton97c2c9b82006-04-18 22:20:38 -070062 }
Linus Torvalds1da177e2005-04-16 15:20:36 -070063
64 /*
65 * The memory size of the process is the basis for the badness.
66 */
Andrew Morton97c2c9b82006-04-18 22:20:38 -070067 points = mm->total_vm;
68
69 /*
70 * After this unlock we can no longer dereference local variable `mm'
71 */
72 task_unlock(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -070073
74 /*
Hugh Dickins7ba34852007-01-05 16:37:03 -080075 * swapoff can easily use up all memory, so kill those first.
76 */
77 if (p->flags & PF_SWAPOFF)
78 return ULONG_MAX;
79
80 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -070081 * Processes which fork a lot of child processes are likely
Kurt Garloff9827b782006-02-20 18:27:51 -080082 * a good choice. We add half the vmsize of the children if they
Linus Torvalds1da177e2005-04-16 15:20:36 -070083 * have an own mm. This prevents forking servers to flood the
Kurt Garloff9827b782006-02-20 18:27:51 -080084 * machine with an endless amount of children. In case a single
85 * child is eating the vast majority of memory, adding only half
86 * to the parents will make the child our kill candidate of choice.
Linus Torvalds1da177e2005-04-16 15:20:36 -070087 */
Andrew Morton97c2c9b82006-04-18 22:20:38 -070088 list_for_each_entry(child, &p->children, sibling) {
89 task_lock(child);
90 if (child->mm != mm && child->mm)
91 points += child->mm->total_vm/2 + 1;
92 task_unlock(child);
Linus Torvalds1da177e2005-04-16 15:20:36 -070093 }
94
95 /*
96 * CPU time is in tens of seconds and run time is in thousands
97 * of seconds. There is no particular reason for this other than
98 * that it turned out to work very well in practice.
99 */
100 cpu_time = (cputime_to_jiffies(p->utime) + cputime_to_jiffies(p->stime))
101 >> (SHIFT_HZ + 3);
102
103 if (uptime >= p->start_time.tv_sec)
104 run_time = (uptime - p->start_time.tv_sec) >> 10;
105 else
106 run_time = 0;
107
108 s = int_sqrt(cpu_time);
109 if (s)
110 points /= s;
111 s = int_sqrt(int_sqrt(run_time));
112 if (s)
113 points /= s;
114
115 /*
116 * Niced processes are most likely less important, so double
117 * their badness points.
118 */
119 if (task_nice(p) > 0)
120 points *= 2;
121
122 /*
123 * Superuser processes are usually more important, so we make it
124 * less likely that we kill those.
125 */
126 if (cap_t(p->cap_effective) & CAP_TO_MASK(CAP_SYS_ADMIN) ||
127 p->uid == 0 || p->euid == 0)
128 points /= 4;
129
130 /*
131 * We don't want to kill a process with direct hardware access.
132 * Not only could that mess up the hardware, but usually users
133 * tend to only have this flag set on applications they think
134 * of as important.
135 */
136 if (cap_t(p->cap_effective) & CAP_TO_MASK(CAP_SYS_RAWIO))
137 points /= 4;
138
139 /*
Nick Piggin7887a3d2006-09-25 23:31:29 -0700140 * If p's nodes don't overlap ours, it may still help to kill p
141 * because p may have allocated or otherwise mapped memory on
142 * this node before. However it will be less likely.
143 */
144 if (!cpuset_excl_nodes_overlap(p))
145 points /= 8;
146
147 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148 * Adjust the score by oomkilladj.
149 */
150 if (p->oomkilladj) {
Joshua N Pritikin9a827822007-05-06 14:49:07 -0700151 if (p->oomkilladj > 0) {
152 if (!points)
153 points = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154 points <<= p->oomkilladj;
Joshua N Pritikin9a827822007-05-06 14:49:07 -0700155 } else
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156 points >>= -(p->oomkilladj);
157 }
158
159#ifdef DEBUG
David Rientjesa5e58a62007-07-31 00:37:32 -0700160 printk(KERN_DEBUG "OOMkill: task %d (%s) got %lu points\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161 p->pid, p->comm, points);
162#endif
163 return points;
164}
165
166/*
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800167 * Types of limitations to the nodes from which allocations may occur
168 */
169#define CONSTRAINT_NONE 1
170#define CONSTRAINT_MEMORY_POLICY 2
171#define CONSTRAINT_CPUSET 3
172
173/*
174 * Determine the type of allocation constraint.
175 */
176static inline int constrained_alloc(struct zonelist *zonelist, gfp_t gfp_mask)
177{
178#ifdef CONFIG_NUMA
179 struct zone **z;
Christoph Lameteree31af52007-10-16 01:25:30 -0700180 nodemask_t nodes = node_states[N_HIGH_MEMORY];
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800181
182 for (z = zonelist->zones; *z; z++)
Paul Jackson02a0e532006-12-13 00:34:25 -0800183 if (cpuset_zone_allowed_softwall(*z, gfp_mask))
Christoph Lameter89fa3022006-09-25 23:31:55 -0700184 node_clear(zone_to_nid(*z), nodes);
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800185 else
186 return CONSTRAINT_CPUSET;
187
188 if (!nodes_empty(nodes))
189 return CONSTRAINT_MEMORY_POLICY;
190#endif
191
192 return CONSTRAINT_NONE;
193}
194
195/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196 * Simple selection loop. We chose the process with the highest
197 * number of 'points'. We expect the caller will lock the tasklist.
198 *
199 * (not docbooked, we don't want this one cluttering up the manual)
200 */
Kurt Garloff9827b782006-02-20 18:27:51 -0800201static struct task_struct *select_bad_process(unsigned long *ppoints)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203 struct task_struct *g, *p;
204 struct task_struct *chosen = NULL;
205 struct timespec uptime;
Kurt Garloff9827b782006-02-20 18:27:51 -0800206 *ppoints = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207
208 do_posix_clock_monotonic_gettime(&uptime);
Paul Jacksona49335c2005-09-06 15:18:09 -0700209 do_each_thread(g, p) {
210 unsigned long points;
Paul Jacksona49335c2005-09-06 15:18:09 -0700211
Oleg Nesterov28324d12006-09-29 02:01:12 -0700212 /*
213 * skip kernel threads and tasks which have already released
214 * their mm.
215 */
Nick Piggin5081dde2006-09-25 23:31:32 -0700216 if (!p->mm)
217 continue;
Oleg Nesterov28324d12006-09-29 02:01:12 -0700218 /* skip the init task */
219 if (is_init(p))
Paul Jacksona49335c2005-09-06 15:18:09 -0700220 continue;
Paul Jacksonef08e3b2005-09-06 15:18:13 -0700221
Paul Jacksona49335c2005-09-06 15:18:09 -0700222 /*
Nick Pigginb78483a2006-09-29 02:01:14 -0700223 * This task already has access to memory reserves and is
224 * being killed. Don't allow any other task access to the
225 * memory reserve.
226 *
227 * Note: this may have a chance of deadlock if it gets
228 * blocked waiting for another task which itself is waiting
229 * for memory. Is there a better alternative?
230 */
231 if (test_tsk_thread_flag(p, TIF_MEMDIE))
232 return ERR_PTR(-1UL);
233
234 /*
Dave Peterson6937a252006-06-23 02:03:13 -0700235 * This is in the process of releasing memory so wait for it
Paul Jacksona49335c2005-09-06 15:18:09 -0700236 * to finish before killing some other task by mistake.
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700237 *
238 * However, if p is the current task, we allow the 'kill' to
239 * go ahead if it is exiting: this will simply set TIF_MEMDIE,
240 * which will allow it to gain access to memory reserves in
241 * the process of exiting and releasing its resources.
Nick Pigginb78483a2006-09-29 02:01:14 -0700242 * Otherwise we could get an easy OOM deadlock.
Paul Jacksona49335c2005-09-06 15:18:09 -0700243 */
Nick Pigginb78483a2006-09-29 02:01:14 -0700244 if (p->flags & PF_EXITING) {
245 if (p != current)
246 return ERR_PTR(-1UL);
247
Oleg Nesterov972c4ea2006-09-29 02:01:12 -0700248 chosen = p;
249 *ppoints = ULONG_MAX;
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700250 }
Oleg Nesterov972c4ea2006-09-29 02:01:12 -0700251
Nick Piggin4a3ede12006-09-25 23:31:30 -0700252 if (p->oomkilladj == OOM_DISABLE)
253 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254
Paul Jacksona49335c2005-09-06 15:18:09 -0700255 points = badness(p, uptime.tv_sec);
Kurt Garloff9827b782006-02-20 18:27:51 -0800256 if (points > *ppoints || !chosen) {
Paul Jacksona49335c2005-09-06 15:18:09 -0700257 chosen = p;
Kurt Garloff9827b782006-02-20 18:27:51 -0800258 *ppoints = points;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259 }
Paul Jacksona49335c2005-09-06 15:18:09 -0700260 } while_each_thread(g, p);
Oleg Nesterov972c4ea2006-09-29 02:01:12 -0700261
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262 return chosen;
263}
264
265/**
Ram Gupta5a291b982006-09-25 23:31:54 -0700266 * Send SIGKILL to the selected process irrespective of CAP_SYS_RAW_IO
267 * flag though it's unlikely that we select a process with CAP_SYS_RAW_IO
268 * set.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269 */
Nick Pigginf3af38d2006-12-06 20:31:51 -0800270static void __oom_kill_task(struct task_struct *p, int verbose)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271{
Sukadev Bhattiproluf400e192006-09-29 02:00:07 -0700272 if (is_init(p)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273 WARN_ON(1);
274 printk(KERN_WARNING "tried to kill init!\n");
275 return;
276 }
277
Oleg Nesterov01017a22006-09-29 02:01:13 -0700278 if (!p->mm) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279 WARN_ON(1);
280 printk(KERN_WARNING "tried to kill an mm-less task!\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281 return;
282 }
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700283
Nick Pigginf3af38d2006-12-06 20:31:51 -0800284 if (verbose)
285 printk(KERN_ERR "Killed process %d (%s)\n", p->pid, p->comm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700286
287 /*
288 * We give our sacrificial lamb high priority and access to
289 * all the memory it needs. That way it should be able to
290 * exit() and clear out its resources quickly...
291 */
292 p->time_slice = HZ;
293 set_tsk_thread_flag(p, TIF_MEMDIE);
294
295 force_sig(SIGKILL, p);
296}
297
Nick Pigginf3af38d2006-12-06 20:31:51 -0800298static int oom_kill_task(struct task_struct *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299{
Dave Peterson01315922006-04-18 22:20:44 -0700300 struct mm_struct *mm;
Ingo Molnar36c8b582006-07-03 00:25:41 -0700301 struct task_struct *g, *q;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302
Dave Peterson01315922006-04-18 22:20:44 -0700303 mm = p->mm;
304
305 /* WARNING: mm may not be dereferenced since we did not obtain its
306 * value from get_task_mm(p). This is OK since all we need to do is
307 * compare mm to q->mm below.
308 *
309 * Furthermore, even if mm contains a non-NULL value, p->mm may
310 * change to NULL at any time since we do not hold task_lock(p).
311 * However, this is of no concern to us.
312 */
313
Oleg Nesterov01017a22006-09-29 02:01:13 -0700314 if (mm == NULL)
Dave Peterson01315922006-04-18 22:20:44 -0700315 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316
Nick Pigginc33e0fc2006-12-06 20:31:50 -0800317 /*
318 * Don't kill the process if any threads are set to OOM_DISABLE
319 */
320 do_each_thread(g, q) {
Ankita Garg35ae8342007-03-16 13:38:33 -0800321 if (q->mm == mm && q->oomkilladj == OOM_DISABLE)
Nick Pigginc33e0fc2006-12-06 20:31:50 -0800322 return 1;
323 } while_each_thread(g, q);
324
Nick Pigginf3af38d2006-12-06 20:31:51 -0800325 __oom_kill_task(p, 1);
Nick Pigginc33e0fc2006-12-06 20:31:50 -0800326
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327 /*
328 * kill all processes that share the ->mm (i.e. all threads),
Nick Pigginf2a2a712006-12-06 20:31:52 -0800329 * but are in a different thread group. Don't let them have access
330 * to memory reserves though, otherwise we might deplete all memory.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700331 */
Nick Pigginc33e0fc2006-12-06 20:31:50 -0800332 do_each_thread(g, q) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333 if (q->mm == mm && q->tgid != p->tgid)
David Rientjes650a7c92007-04-23 21:36:13 -0700334 force_sig(SIGKILL, q);
Nick Pigginc33e0fc2006-12-06 20:31:50 -0800335 } while_each_thread(g, q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700336
Dave Peterson01315922006-04-18 22:20:44 -0700337 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338}
339
Dave Peterson01315922006-04-18 22:20:44 -0700340static int oom_kill_process(struct task_struct *p, unsigned long points,
341 const char *message)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343 struct task_struct *c;
344 struct list_head *tsk;
345
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700346 /*
347 * If the task is already exiting, don't alarm the sysadmin or kill
348 * its children or threads, just set TIF_MEMDIE so it can die quickly
349 */
350 if (p->flags & PF_EXITING) {
Nick Pigginf3af38d2006-12-06 20:31:51 -0800351 __oom_kill_task(p, 0);
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700352 return 0;
353 }
354
Nick Pigginf3af38d2006-12-06 20:31:51 -0800355 printk(KERN_ERR "%s: kill process %d (%s) score %li or a child\n",
356 message, p->pid, p->comm, points);
357
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358 /* Try to kill a child first */
359 list_for_each(tsk, &p->children) {
360 c = list_entry(tsk, struct task_struct, sibling);
361 if (c->mm == p->mm)
362 continue;
Nick Pigginf3af38d2006-12-06 20:31:51 -0800363 if (!oom_kill_task(c))
Dave Peterson01315922006-04-18 22:20:44 -0700364 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365 }
Nick Pigginf3af38d2006-12-06 20:31:51 -0800366 return oom_kill_task(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367}
368
Martin Schwidefsky8bc719d2006-09-25 23:31:20 -0700369static BLOCKING_NOTIFIER_HEAD(oom_notify_list);
370
371int register_oom_notifier(struct notifier_block *nb)
372{
373 return blocking_notifier_chain_register(&oom_notify_list, nb);
374}
375EXPORT_SYMBOL_GPL(register_oom_notifier);
376
377int unregister_oom_notifier(struct notifier_block *nb)
378{
379 return blocking_notifier_chain_unregister(&oom_notify_list, nb);
380}
381EXPORT_SYMBOL_GPL(unregister_oom_notifier);
382
Linus Torvalds1da177e2005-04-16 15:20:36 -0700383/**
Dave Peterson6937a252006-06-23 02:03:13 -0700384 * out_of_memory - kill the "best" process when we run out of memory
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385 *
386 * If we run out of memory, we have the choice between either
387 * killing a random task (bad), letting the system crash (worse)
388 * OR try to be smart about which process to kill. Note that we
389 * don't have to be perfect here, we just have to be good.
390 */
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800391void out_of_memory(struct zonelist *zonelist, gfp_t gfp_mask, int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392{
Ingo Molnar36c8b582006-07-03 00:25:41 -0700393 struct task_struct *p;
Andrew Mortond6713e02006-02-28 16:59:19 -0800394 unsigned long points = 0;
Martin Schwidefsky8bc719d2006-09-25 23:31:20 -0700395 unsigned long freed = 0;
David Rientjes2b45ab32007-05-06 14:50:00 -0700396 int constraint;
Martin Schwidefsky8bc719d2006-09-25 23:31:20 -0700397
398 blocking_notifier_call_chain(&oom_notify_list, 0, &freed);
399 if (freed > 0)
400 /* Got some memory back in the last second. */
401 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700402
Anton Blanchard42639262005-07-07 17:56:06 -0700403 if (printk_ratelimit()) {
Nick Pigginb72f1602006-09-25 23:31:32 -0700404 printk(KERN_WARNING "%s invoked oom-killer: "
405 "gfp_mask=0x%x, order=%d, oomkilladj=%d\n",
406 current->comm, gfp_mask, order, current->oomkilladj);
Andrew Mortonb958f7d2006-02-01 03:05:51 -0800407 dump_stack();
Anton Blanchard42639262005-07-07 17:56:06 -0700408 show_mem();
409 }
Janet Morgan578c2fd2005-06-21 17:14:56 -0700410
Yasunori Goto2b744c02007-05-06 14:49:59 -0700411 if (sysctl_panic_on_oom == 2)
412 panic("out of memory. Compulsory panic_on_oom is selected.\n");
413
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800414 /*
415 * Check if there were limitations on the allocation (only relevant for
416 * NUMA) that may require different handling.
417 */
David Rientjes2b45ab32007-05-06 14:50:00 -0700418 constraint = constrained_alloc(zonelist, gfp_mask);
419 cpuset_lock();
420 read_lock(&tasklist_lock);
421
422 switch (constraint) {
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800423 case CONSTRAINT_MEMORY_POLICY:
Dave Peterson01315922006-04-18 22:20:44 -0700424 oom_kill_process(current, points,
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800425 "No available memory (MPOL_BIND)");
426 break;
427
428 case CONSTRAINT_CPUSET:
Dave Peterson01315922006-04-18 22:20:44 -0700429 oom_kill_process(current, points,
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800430 "No available memory in cpuset");
431 break;
432
433 case CONSTRAINT_NONE:
KAMEZAWA Hiroyukifadd8fb2006-06-23 02:03:13 -0700434 if (sysctl_panic_on_oom)
435 panic("out of memory. panic_on_oom is selected\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436retry:
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800437 /*
438 * Rambo mode: Shoot down a process and hope it solves whatever
439 * issues we may have.
440 */
441 p = select_bad_process(&points);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800443 if (PTR_ERR(p) == -1UL)
444 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700445
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800446 /* Found nothing?!?! Either we hang forever, or we panic. */
447 if (!p) {
448 read_unlock(&tasklist_lock);
449 cpuset_unlock();
450 panic("Out of memory and no killable processes...\n");
451 }
452
Dave Peterson01315922006-04-18 22:20:44 -0700453 if (oom_kill_process(p, points, "Out of memory"))
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800454 goto retry;
455
456 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700457 }
458
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800459out:
Andrew Morton140ffce2006-03-02 02:54:28 -0800460 read_unlock(&tasklist_lock);
Paul Jackson505970b2006-01-14 13:21:06 -0800461 cpuset_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462
463 /*
464 * Give "p" a good chance of killing itself before we
Kirill Korotaev2f659f42006-01-08 01:01:05 -0800465 * retry to allocate memory unless "p" is current
Linus Torvalds1da177e2005-04-16 15:20:36 -0700466 */
Kirill Korotaev2f659f42006-01-08 01:01:05 -0800467 if (!test_thread_flag(TIF_MEMDIE))
Andrew Morton140ffce2006-03-02 02:54:28 -0800468 schedule_timeout_uninterruptible(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700469}