blob: a7001410ab15964449f2b7188be3ecf05a9876eb [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>
20#include <linux/sched.h>
21#include <linux/swap.h>
22#include <linux/timex.h>
23#include <linux/jiffies.h>
Paul Jacksonef08e3b2005-09-06 15:18:13 -070024#include <linux/cpuset.h>
Martin Schwidefsky8bc719d2006-09-25 23:31:20 -070025#include <linux/module.h>
26#include <linux/notifier.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070027
KAMEZAWA Hiroyukifadd8fb2006-06-23 02:03:13 -070028int sysctl_panic_on_oom;
Linus Torvalds1da177e2005-04-16 15:20:36 -070029/* #define DEBUG */
30
31/**
Dave Peterson6937a252006-06-23 02:03:13 -070032 * badness - calculate a numeric value for how bad this task has been
Linus Torvalds1da177e2005-04-16 15:20:36 -070033 * @p: task struct of which task we should calculate
Paul Jacksona49335c2005-09-06 15:18:09 -070034 * @uptime: current uptime in seconds
Linus Torvalds1da177e2005-04-16 15:20:36 -070035 *
36 * The formula used is relatively simple and documented inline in the
37 * function. The main rationale is that we want to select a good task
38 * to kill when we run out of memory.
39 *
40 * Good in this context means that:
41 * 1) we lose the minimum amount of work done
42 * 2) we recover a large amount of memory
43 * 3) we don't kill anything innocent of eating tons of memory
44 * 4) we want to kill the minimum amount of processes (one)
45 * 5) we try to kill the process the user expects us to kill, this
46 * algorithm has been meticulously tuned to meet the principle
47 * of least surprise ... (be careful when you change it)
48 */
49
50unsigned long badness(struct task_struct *p, unsigned long uptime)
51{
52 unsigned long points, cpu_time, run_time, s;
Andrew Morton97c2c9b2006-04-18 22:20:38 -070053 struct mm_struct *mm;
54 struct task_struct *child;
Linus Torvalds1da177e2005-04-16 15:20:36 -070055
Andrew Morton97c2c9b2006-04-18 22:20:38 -070056 task_lock(p);
57 mm = p->mm;
58 if (!mm) {
59 task_unlock(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -070060 return 0;
Andrew Morton97c2c9b2006-04-18 22:20:38 -070061 }
Linus Torvalds1da177e2005-04-16 15:20:36 -070062
63 /*
64 * The memory size of the process is the basis for the badness.
65 */
Andrew Morton97c2c9b2006-04-18 22:20:38 -070066 points = mm->total_vm;
67
68 /*
69 * After this unlock we can no longer dereference local variable `mm'
70 */
71 task_unlock(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -070072
73 /*
Hugh Dickins7ba34852007-01-05 16:37:03 -080074 * swapoff can easily use up all memory, so kill those first.
75 */
76 if (p->flags & PF_SWAPOFF)
77 return ULONG_MAX;
78
79 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -070080 * Processes which fork a lot of child processes are likely
Kurt Garloff9827b782006-02-20 18:27:51 -080081 * a good choice. We add half the vmsize of the children if they
Linus Torvalds1da177e2005-04-16 15:20:36 -070082 * have an own mm. This prevents forking servers to flood the
Kurt Garloff9827b782006-02-20 18:27:51 -080083 * machine with an endless amount of children. In case a single
84 * child is eating the vast majority of memory, adding only half
85 * to the parents will make the child our kill candidate of choice.
Linus Torvalds1da177e2005-04-16 15:20:36 -070086 */
Andrew Morton97c2c9b2006-04-18 22:20:38 -070087 list_for_each_entry(child, &p->children, sibling) {
88 task_lock(child);
89 if (child->mm != mm && child->mm)
90 points += child->mm->total_vm/2 + 1;
91 task_unlock(child);
Linus Torvalds1da177e2005-04-16 15:20:36 -070092 }
93
94 /*
95 * CPU time is in tens of seconds and run time is in thousands
96 * of seconds. There is no particular reason for this other than
97 * that it turned out to work very well in practice.
98 */
99 cpu_time = (cputime_to_jiffies(p->utime) + cputime_to_jiffies(p->stime))
100 >> (SHIFT_HZ + 3);
101
102 if (uptime >= p->start_time.tv_sec)
103 run_time = (uptime - p->start_time.tv_sec) >> 10;
104 else
105 run_time = 0;
106
107 s = int_sqrt(cpu_time);
108 if (s)
109 points /= s;
110 s = int_sqrt(int_sqrt(run_time));
111 if (s)
112 points /= s;
113
114 /*
115 * Niced processes are most likely less important, so double
116 * their badness points.
117 */
118 if (task_nice(p) > 0)
119 points *= 2;
120
121 /*
122 * Superuser processes are usually more important, so we make it
123 * less likely that we kill those.
124 */
125 if (cap_t(p->cap_effective) & CAP_TO_MASK(CAP_SYS_ADMIN) ||
126 p->uid == 0 || p->euid == 0)
127 points /= 4;
128
129 /*
130 * We don't want to kill a process with direct hardware access.
131 * Not only could that mess up the hardware, but usually users
132 * tend to only have this flag set on applications they think
133 * of as important.
134 */
135 if (cap_t(p->cap_effective) & CAP_TO_MASK(CAP_SYS_RAWIO))
136 points /= 4;
137
138 /*
Nick Piggin7887a3d2006-09-25 23:31:29 -0700139 * If p's nodes don't overlap ours, it may still help to kill p
140 * because p may have allocated or otherwise mapped memory on
141 * this node before. However it will be less likely.
142 */
143 if (!cpuset_excl_nodes_overlap(p))
144 points /= 8;
145
146 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700147 * Adjust the score by oomkilladj.
148 */
149 if (p->oomkilladj) {
Joshua N Pritikin9a827822007-05-06 14:49:07 -0700150 if (p->oomkilladj > 0) {
151 if (!points)
152 points = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153 points <<= p->oomkilladj;
Joshua N Pritikin9a827822007-05-06 14:49:07 -0700154 } else
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155 points >>= -(p->oomkilladj);
156 }
157
158#ifdef DEBUG
159 printk(KERN_DEBUG "OOMkill: task %d (%s) got %d points\n",
160 p->pid, p->comm, points);
161#endif
162 return points;
163}
164
165/*
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800166 * Types of limitations to the nodes from which allocations may occur
167 */
168#define CONSTRAINT_NONE 1
169#define CONSTRAINT_MEMORY_POLICY 2
170#define CONSTRAINT_CPUSET 3
171
172/*
173 * Determine the type of allocation constraint.
174 */
175static inline int constrained_alloc(struct zonelist *zonelist, gfp_t gfp_mask)
176{
177#ifdef CONFIG_NUMA
178 struct zone **z;
KAMEZAWA Hiroyuki96ac5912006-12-29 16:47:20 -0800179 nodemask_t nodes;
180 int node;
Hugh Dickins3d124cb2007-04-23 14:41:02 -0700181
182 nodes_clear(nodes);
KAMEZAWA Hiroyuki96ac5912006-12-29 16:47:20 -0800183 /* node has memory ? */
184 for_each_online_node(node)
185 if (NODE_DATA(node)->node_present_pages)
186 node_set(node, nodes);
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800187
188 for (z = zonelist->zones; *z; z++)
Paul Jackson02a0e532006-12-13 00:34:25 -0800189 if (cpuset_zone_allowed_softwall(*z, gfp_mask))
Christoph Lameter89fa3022006-09-25 23:31:55 -0700190 node_clear(zone_to_nid(*z), nodes);
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800191 else
192 return CONSTRAINT_CPUSET;
193
194 if (!nodes_empty(nodes))
195 return CONSTRAINT_MEMORY_POLICY;
196#endif
197
198 return CONSTRAINT_NONE;
199}
200
201/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202 * Simple selection loop. We chose the process with the highest
203 * number of 'points'. We expect the caller will lock the tasklist.
204 *
205 * (not docbooked, we don't want this one cluttering up the manual)
206 */
Kurt Garloff9827b782006-02-20 18:27:51 -0800207static struct task_struct *select_bad_process(unsigned long *ppoints)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700209 struct task_struct *g, *p;
210 struct task_struct *chosen = NULL;
211 struct timespec uptime;
Kurt Garloff9827b782006-02-20 18:27:51 -0800212 *ppoints = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213
214 do_posix_clock_monotonic_gettime(&uptime);
Paul Jacksona49335c2005-09-06 15:18:09 -0700215 do_each_thread(g, p) {
216 unsigned long points;
Paul Jacksona49335c2005-09-06 15:18:09 -0700217
Oleg Nesterov28324d12006-09-29 02:01:12 -0700218 /*
219 * skip kernel threads and tasks which have already released
220 * their mm.
221 */
Nick Piggin5081dde2006-09-25 23:31:32 -0700222 if (!p->mm)
223 continue;
Oleg Nesterov28324d12006-09-29 02:01:12 -0700224 /* skip the init task */
225 if (is_init(p))
Paul Jacksona49335c2005-09-06 15:18:09 -0700226 continue;
Paul Jacksonef08e3b2005-09-06 15:18:13 -0700227
Paul Jacksona49335c2005-09-06 15:18:09 -0700228 /*
Nick Pigginb78483a2006-09-29 02:01:14 -0700229 * This task already has access to memory reserves and is
230 * being killed. Don't allow any other task access to the
231 * memory reserve.
232 *
233 * Note: this may have a chance of deadlock if it gets
234 * blocked waiting for another task which itself is waiting
235 * for memory. Is there a better alternative?
236 */
237 if (test_tsk_thread_flag(p, TIF_MEMDIE))
238 return ERR_PTR(-1UL);
239
240 /*
Dave Peterson6937a252006-06-23 02:03:13 -0700241 * This is in the process of releasing memory so wait for it
Paul Jacksona49335c2005-09-06 15:18:09 -0700242 * to finish before killing some other task by mistake.
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700243 *
244 * However, if p is the current task, we allow the 'kill' to
245 * go ahead if it is exiting: this will simply set TIF_MEMDIE,
246 * which will allow it to gain access to memory reserves in
247 * the process of exiting and releasing its resources.
Nick Pigginb78483a2006-09-29 02:01:14 -0700248 * Otherwise we could get an easy OOM deadlock.
Paul Jacksona49335c2005-09-06 15:18:09 -0700249 */
Nick Pigginb78483a2006-09-29 02:01:14 -0700250 if (p->flags & PF_EXITING) {
251 if (p != current)
252 return ERR_PTR(-1UL);
253
Oleg Nesterov972c4ea2006-09-29 02:01:12 -0700254 chosen = p;
255 *ppoints = ULONG_MAX;
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700256 }
Oleg Nesterov972c4ea2006-09-29 02:01:12 -0700257
Nick Piggin4a3ede12006-09-25 23:31:30 -0700258 if (p->oomkilladj == OOM_DISABLE)
259 continue;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260
Paul Jacksona49335c2005-09-06 15:18:09 -0700261 points = badness(p, uptime.tv_sec);
Kurt Garloff9827b782006-02-20 18:27:51 -0800262 if (points > *ppoints || !chosen) {
Paul Jacksona49335c2005-09-06 15:18:09 -0700263 chosen = p;
Kurt Garloff9827b782006-02-20 18:27:51 -0800264 *ppoints = points;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265 }
Paul Jacksona49335c2005-09-06 15:18:09 -0700266 } while_each_thread(g, p);
Oleg Nesterov972c4ea2006-09-29 02:01:12 -0700267
Linus Torvalds1da177e2005-04-16 15:20:36 -0700268 return chosen;
269}
270
271/**
Ram Gupta5a291b982006-09-25 23:31:54 -0700272 * Send SIGKILL to the selected process irrespective of CAP_SYS_RAW_IO
273 * flag though it's unlikely that we select a process with CAP_SYS_RAW_IO
274 * set.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700275 */
Nick Pigginf3af38d2006-12-06 20:31:51 -0800276static void __oom_kill_task(struct task_struct *p, int verbose)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700277{
Sukadev Bhattiproluf400e192006-09-29 02:00:07 -0700278 if (is_init(p)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279 WARN_ON(1);
280 printk(KERN_WARNING "tried to kill init!\n");
281 return;
282 }
283
Oleg Nesterov01017a22006-09-29 02:01:13 -0700284 if (!p->mm) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285 WARN_ON(1);
286 printk(KERN_WARNING "tried to kill an mm-less task!\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287 return;
288 }
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700289
Nick Pigginf3af38d2006-12-06 20:31:51 -0800290 if (verbose)
291 printk(KERN_ERR "Killed process %d (%s)\n", p->pid, p->comm);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700292
293 /*
294 * We give our sacrificial lamb high priority and access to
295 * all the memory it needs. That way it should be able to
296 * exit() and clear out its resources quickly...
297 */
298 p->time_slice = HZ;
299 set_tsk_thread_flag(p, TIF_MEMDIE);
300
301 force_sig(SIGKILL, p);
302}
303
Nick Pigginf3af38d2006-12-06 20:31:51 -0800304static int oom_kill_task(struct task_struct *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700305{
Dave Peterson01315922006-04-18 22:20:44 -0700306 struct mm_struct *mm;
Ingo Molnar36c8b582006-07-03 00:25:41 -0700307 struct task_struct *g, *q;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308
Dave Peterson01315922006-04-18 22:20:44 -0700309 mm = p->mm;
310
311 /* WARNING: mm may not be dereferenced since we did not obtain its
312 * value from get_task_mm(p). This is OK since all we need to do is
313 * compare mm to q->mm below.
314 *
315 * Furthermore, even if mm contains a non-NULL value, p->mm may
316 * change to NULL at any time since we do not hold task_lock(p).
317 * However, this is of no concern to us.
318 */
319
Oleg Nesterov01017a22006-09-29 02:01:13 -0700320 if (mm == NULL)
Dave Peterson01315922006-04-18 22:20:44 -0700321 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322
Nick Pigginc33e0fc2006-12-06 20:31:50 -0800323 /*
324 * Don't kill the process if any threads are set to OOM_DISABLE
325 */
326 do_each_thread(g, q) {
Ankita Garg35ae8342007-03-16 13:38:33 -0800327 if (q->mm == mm && q->oomkilladj == OOM_DISABLE)
Nick Pigginc33e0fc2006-12-06 20:31:50 -0800328 return 1;
329 } while_each_thread(g, q);
330
Nick Pigginf3af38d2006-12-06 20:31:51 -0800331 __oom_kill_task(p, 1);
Nick Pigginc33e0fc2006-12-06 20:31:50 -0800332
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333 /*
334 * kill all processes that share the ->mm (i.e. all threads),
Nick Pigginf2a2a712006-12-06 20:31:52 -0800335 * but are in a different thread group. Don't let them have access
336 * to memory reserves though, otherwise we might deplete all memory.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337 */
Nick Pigginc33e0fc2006-12-06 20:31:50 -0800338 do_each_thread(g, q) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700339 if (q->mm == mm && q->tgid != p->tgid)
David Rientjes650a7c92007-04-23 21:36:13 -0700340 force_sig(SIGKILL, q);
Nick Pigginc33e0fc2006-12-06 20:31:50 -0800341 } while_each_thread(g, q);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342
Dave Peterson01315922006-04-18 22:20:44 -0700343 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344}
345
Dave Peterson01315922006-04-18 22:20:44 -0700346static int oom_kill_process(struct task_struct *p, unsigned long points,
347 const char *message)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349 struct task_struct *c;
350 struct list_head *tsk;
351
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700352 /*
353 * If the task is already exiting, don't alarm the sysadmin or kill
354 * its children or threads, just set TIF_MEMDIE so it can die quickly
355 */
356 if (p->flags & PF_EXITING) {
Nick Pigginf3af38d2006-12-06 20:31:51 -0800357 __oom_kill_task(p, 0);
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700358 return 0;
359 }
360
Nick Pigginf3af38d2006-12-06 20:31:51 -0800361 printk(KERN_ERR "%s: kill process %d (%s) score %li or a child\n",
362 message, p->pid, p->comm, points);
363
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364 /* Try to kill a child first */
365 list_for_each(tsk, &p->children) {
366 c = list_entry(tsk, struct task_struct, sibling);
367 if (c->mm == p->mm)
368 continue;
Nick Pigginf3af38d2006-12-06 20:31:51 -0800369 if (!oom_kill_task(c))
Dave Peterson01315922006-04-18 22:20:44 -0700370 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371 }
Nick Pigginf3af38d2006-12-06 20:31:51 -0800372 return oom_kill_task(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373}
374
Martin Schwidefsky8bc719d2006-09-25 23:31:20 -0700375static BLOCKING_NOTIFIER_HEAD(oom_notify_list);
376
377int register_oom_notifier(struct notifier_block *nb)
378{
379 return blocking_notifier_chain_register(&oom_notify_list, nb);
380}
381EXPORT_SYMBOL_GPL(register_oom_notifier);
382
383int unregister_oom_notifier(struct notifier_block *nb)
384{
385 return blocking_notifier_chain_unregister(&oom_notify_list, nb);
386}
387EXPORT_SYMBOL_GPL(unregister_oom_notifier);
388
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389/**
Dave Peterson6937a252006-06-23 02:03:13 -0700390 * out_of_memory - kill the "best" process when we run out of memory
Linus Torvalds1da177e2005-04-16 15:20:36 -0700391 *
392 * If we run out of memory, we have the choice between either
393 * killing a random task (bad), letting the system crash (worse)
394 * OR try to be smart about which process to kill. Note that we
395 * don't have to be perfect here, we just have to be good.
396 */
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800397void out_of_memory(struct zonelist *zonelist, gfp_t gfp_mask, int order)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700398{
Ingo Molnar36c8b582006-07-03 00:25:41 -0700399 struct task_struct *p;
Andrew Mortond6713e02006-02-28 16:59:19 -0800400 unsigned long points = 0;
Martin Schwidefsky8bc719d2006-09-25 23:31:20 -0700401 unsigned long freed = 0;
David Rientjes2b45ab32007-05-06 14:50:00 -0700402 int constraint;
Martin Schwidefsky8bc719d2006-09-25 23:31:20 -0700403
404 blocking_notifier_call_chain(&oom_notify_list, 0, &freed);
405 if (freed > 0)
406 /* Got some memory back in the last second. */
407 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408
Anton Blanchard42639262005-07-07 17:56:06 -0700409 if (printk_ratelimit()) {
Nick Pigginb72f1602006-09-25 23:31:32 -0700410 printk(KERN_WARNING "%s invoked oom-killer: "
411 "gfp_mask=0x%x, order=%d, oomkilladj=%d\n",
412 current->comm, gfp_mask, order, current->oomkilladj);
Andrew Mortonb958f7d2006-02-01 03:05:51 -0800413 dump_stack();
Anton Blanchard42639262005-07-07 17:56:06 -0700414 show_mem();
415 }
Janet Morgan578c2fd2005-06-21 17:14:56 -0700416
Yasunori Goto2b744c02007-05-06 14:49:59 -0700417 if (sysctl_panic_on_oom == 2)
418 panic("out of memory. Compulsory panic_on_oom is selected.\n");
419
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800420 /*
421 * Check if there were limitations on the allocation (only relevant for
422 * NUMA) that may require different handling.
423 */
David Rientjes2b45ab32007-05-06 14:50:00 -0700424 constraint = constrained_alloc(zonelist, gfp_mask);
425 cpuset_lock();
426 read_lock(&tasklist_lock);
427
428 switch (constraint) {
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800429 case CONSTRAINT_MEMORY_POLICY:
Dave Peterson01315922006-04-18 22:20:44 -0700430 oom_kill_process(current, points,
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800431 "No available memory (MPOL_BIND)");
432 break;
433
434 case CONSTRAINT_CPUSET:
Dave Peterson01315922006-04-18 22:20:44 -0700435 oom_kill_process(current, points,
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800436 "No available memory in cpuset");
437 break;
438
439 case CONSTRAINT_NONE:
KAMEZAWA Hiroyukifadd8fb2006-06-23 02:03:13 -0700440 if (sysctl_panic_on_oom)
441 panic("out of memory. panic_on_oom is selected\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442retry:
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800443 /*
444 * Rambo mode: Shoot down a process and hope it solves whatever
445 * issues we may have.
446 */
447 p = select_bad_process(&points);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800449 if (PTR_ERR(p) == -1UL)
450 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800452 /* Found nothing?!?! Either we hang forever, or we panic. */
453 if (!p) {
454 read_unlock(&tasklist_lock);
455 cpuset_unlock();
456 panic("Out of memory and no killable processes...\n");
457 }
458
Dave Peterson01315922006-04-18 22:20:44 -0700459 if (oom_kill_process(p, points, "Out of memory"))
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800460 goto retry;
461
462 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463 }
464
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800465out:
Andrew Morton140ffce2006-03-02 02:54:28 -0800466 read_unlock(&tasklist_lock);
Paul Jackson505970b2006-01-14 13:21:06 -0800467 cpuset_unlock();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468
469 /*
470 * Give "p" a good chance of killing itself before we
Kirill Korotaev2f659f42006-01-08 01:01:05 -0800471 * retry to allocate memory unless "p" is current
Linus Torvalds1da177e2005-04-16 15:20:36 -0700472 */
Kirill Korotaev2f659f42006-01-08 01:01:05 -0800473 if (!test_thread_flag(TIF_MEMDIE))
Andrew Morton140ffce2006-03-02 02:54:28 -0800474 schedule_timeout_uninterruptible(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475}