blob: 2958fd8e7c9abcfcf6cc7b38e2dac34133905b22 [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...
David Rientjesa63d83f2010-08-09 17:19:46 -07007 * Copyright (C) 2010 Google, Inc.
8 * Rewritten by David Rientjes
Linus Torvalds1da177e2005-04-16 15:20:36 -07009 *
10 * The routines in this file are used to kill a process when
Paul Jacksona49335c2005-09-06 15:18:09 -070011 * we're seriously out of memory. This gets called from __alloc_pages()
12 * in mm/page_alloc.c when we really run out of memory.
Linus Torvalds1da177e2005-04-16 15:20:36 -070013 *
14 * Since we won't call these routines often (on a well-configured
15 * machine) this file will double as a 'coding guide' and a signpost
16 * for newbie kernel hackers. It features several pointers to major
17 * kernel subsystems and hints as to where to find out what things do.
18 */
19
Alexey Dobriyan8ac773b2006-10-19 23:28:32 -070020#include <linux/oom.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070021#include <linux/mm.h>
Alexey Dobriyan4e950f62007-07-30 02:36:13 +040022#include <linux/err.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090023#include <linux/gfp.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include <linux/sched.h>
25#include <linux/swap.h>
26#include <linux/timex.h>
27#include <linux/jiffies.h>
Paul Jacksonef08e3b2005-09-06 15:18:13 -070028#include <linux/cpuset.h>
Paul Gortmakerb95f1b312011-10-16 02:01:52 -040029#include <linux/export.h>
Martin Schwidefsky8bc719d2006-09-25 23:31:20 -070030#include <linux/notifier.h>
Pavel Emelianovc7ba5c92008-02-07 00:13:58 -080031#include <linux/memcontrol.h>
David Rientjes6f48d0eb2010-08-09 17:18:52 -070032#include <linux/mempolicy.h>
David Howells5cd9c582008-08-14 11:37:28 +010033#include <linux/security.h>
David Rientjesedd45542011-03-22 16:30:12 -070034#include <linux/ptrace.h>
David Rientjesf660daa2011-10-31 17:07:07 -070035#include <linux/freezer.h>
KAMEZAWA Hiroyuki43d2b112012-01-10 15:08:09 -080036#include <linux/ftrace.h>
37
38#define CREATE_TRACE_POINTS
39#include <trace/events/oom.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070040
KAMEZAWA Hiroyukifadd8fb2006-06-23 02:03:13 -070041int sysctl_panic_on_oom;
David Rientjesfe071d72007-10-16 23:25:56 -070042int sysctl_oom_kill_allocating_task;
David Rientjesad915c42010-08-09 17:18:53 -070043int sysctl_oom_dump_tasks = 1;
David Rientjesc7d4cae2009-01-06 14:39:00 -080044static DEFINE_SPINLOCK(zone_scan_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070045
David Rientjes43362a42011-10-31 17:07:18 -070046/*
47 * compare_swap_oom_score_adj() - compare and swap current's oom_score_adj
48 * @old_val: old oom_score_adj for compare
49 * @new_val: new oom_score_adj for swap
50 *
51 * Sets the oom_score_adj value for current to @new_val iff its present value is
52 * @old_val. Usually used to reinstate a previous value to prevent racing with
53 * userspacing tuning the value in the interim.
54 */
55void compare_swap_oom_score_adj(int old_val, int new_val)
56{
57 struct sighand_struct *sighand = current->sighand;
58
59 spin_lock_irq(&sighand->siglock);
60 if (current->signal->oom_score_adj == old_val)
61 current->signal->oom_score_adj = new_val;
KAMEZAWA Hiroyuki43d2b112012-01-10 15:08:09 -080062 trace_oom_score_adj_update(current);
David Rientjes43362a42011-10-31 17:07:18 -070063 spin_unlock_irq(&sighand->siglock);
64}
65
David Rientjes72788c32011-05-24 17:11:40 -070066/**
67 * test_set_oom_score_adj() - set current's oom_score_adj and return old value
68 * @new_val: new oom_score_adj value
69 *
70 * Sets the oom_score_adj value for current to @new_val with proper
71 * synchronization and returns the old value. Usually used to temporarily
72 * set a value, save the old value in the caller, and then reinstate it later.
73 */
74int test_set_oom_score_adj(int new_val)
75{
76 struct sighand_struct *sighand = current->sighand;
77 int old_val;
78
79 spin_lock_irq(&sighand->siglock);
80 old_val = current->signal->oom_score_adj;
David Rientjesc9f01242011-10-31 17:07:15 -070081 current->signal->oom_score_adj = new_val;
KAMEZAWA Hiroyuki43d2b112012-01-10 15:08:09 -080082 trace_oom_score_adj_update(current);
David Rientjes72788c32011-05-24 17:11:40 -070083 spin_unlock_irq(&sighand->siglock);
84
85 return old_val;
86}
87
David Rientjes6f48d0eb2010-08-09 17:18:52 -070088#ifdef CONFIG_NUMA
89/**
90 * has_intersects_mems_allowed() - check task eligiblity for kill
91 * @tsk: task struct of which task to consider
92 * @mask: nodemask passed to page allocator for mempolicy ooms
93 *
94 * Task eligibility is determined by whether or not a candidate task, @tsk,
95 * shares the same mempolicy nodes as current if it is bound by such a policy
96 * and whether or not it has the same set of allowed cpuset nodes.
KOSAKI Motohiro495789a2009-09-21 17:03:14 -070097 */
David Rientjes6f48d0eb2010-08-09 17:18:52 -070098static bool has_intersects_mems_allowed(struct task_struct *tsk,
99 const nodemask_t *mask)
KOSAKI Motohiro495789a2009-09-21 17:03:14 -0700100{
David Rientjes6f48d0eb2010-08-09 17:18:52 -0700101 struct task_struct *start = tsk;
KOSAKI Motohiro495789a2009-09-21 17:03:14 -0700102
KOSAKI Motohiro495789a2009-09-21 17:03:14 -0700103 do {
David Rientjes6f48d0eb2010-08-09 17:18:52 -0700104 if (mask) {
105 /*
106 * If this is a mempolicy constrained oom, tsk's
107 * cpuset is irrelevant. Only return true if its
108 * mempolicy intersects current, otherwise it may be
109 * needlessly killed.
110 */
111 if (mempolicy_nodemask_intersects(tsk, mask))
112 return true;
113 } else {
114 /*
115 * This is not a mempolicy constrained oom, so only
116 * check the mems of tsk's cpuset.
117 */
118 if (cpuset_mems_allowed_intersects(current, tsk))
119 return true;
120 }
KOSAKI Motohirodf1090a2010-08-09 17:19:39 -0700121 } while_each_thread(start, tsk);
122
David Rientjes6f48d0eb2010-08-09 17:18:52 -0700123 return false;
KOSAKI Motohiro495789a2009-09-21 17:03:14 -0700124}
David Rientjes6f48d0eb2010-08-09 17:18:52 -0700125#else
126static bool has_intersects_mems_allowed(struct task_struct *tsk,
127 const nodemask_t *mask)
128{
129 return true;
130}
131#endif /* CONFIG_NUMA */
KOSAKI Motohiro495789a2009-09-21 17:03:14 -0700132
David Rientjes6f48d0eb2010-08-09 17:18:52 -0700133/*
134 * The process p may have detached its own ->mm while exiting or through
135 * use_mm(), but one or more of its subthreads may still have a valid
136 * pointer. Return p, or any of its subthreads with a valid ->mm, with
137 * task_lock() held.
138 */
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -0700139struct task_struct *find_lock_task_mm(struct task_struct *p)
Oleg Nesterovdd8e8f42010-08-09 17:18:45 -0700140{
141 struct task_struct *t = p;
142
143 do {
144 task_lock(t);
145 if (likely(t->mm))
146 return t;
147 task_unlock(t);
148 } while_each_thread(p, t);
149
150 return NULL;
151}
152
KOSAKI Motohiroab290ad2010-08-09 17:19:35 -0700153/* return true if the task is not adequate as candidate victim task. */
David Rientjese85bfd32010-09-22 13:05:10 -0700154static bool oom_unkillable_task(struct task_struct *p,
Johannes Weiner72835c82012-01-12 17:18:32 -0800155 const struct mem_cgroup *memcg, const nodemask_t *nodemask)
KOSAKI Motohiroab290ad2010-08-09 17:19:35 -0700156{
157 if (is_global_init(p))
158 return true;
159 if (p->flags & PF_KTHREAD)
160 return true;
161
162 /* When mem_cgroup_out_of_memory() and p is not member of the group */
Johannes Weiner72835c82012-01-12 17:18:32 -0800163 if (memcg && !task_in_mem_cgroup(p, memcg))
KOSAKI Motohiroab290ad2010-08-09 17:19:35 -0700164 return true;
165
166 /* p may not have freeable memory in nodemask */
167 if (!has_intersects_mems_allowed(p, nodemask))
168 return true;
169
170 return false;
171}
172
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173/**
David Rientjesa63d83f2010-08-09 17:19:46 -0700174 * oom_badness - heuristic function to determine which candidate task to kill
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175 * @p: task struct of which task we should calculate
David Rientjesa63d83f2010-08-09 17:19:46 -0700176 * @totalpages: total present RAM allowed for page allocation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177 *
David Rientjesa63d83f2010-08-09 17:19:46 -0700178 * The heuristic for determining which task to kill is made to be as simple and
179 * predictable as possible. The goal is to return the highest value for the
180 * task consuming the most memory to avoid subsequent oom failures.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181 */
Johannes Weiner72835c82012-01-12 17:18:32 -0800182unsigned int oom_badness(struct task_struct *p, struct mem_cgroup *memcg,
David Rientjesa63d83f2010-08-09 17:19:46 -0700183 const nodemask_t *nodemask, unsigned long totalpages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184{
Frantisek Hrbataff05b6f2011-12-19 17:11:59 -0800185 long points;
KOSAKI Motohiro28b83c52009-09-21 17:03:13 -0700186
Johannes Weiner72835c82012-01-12 17:18:32 -0800187 if (oom_unkillable_task(p, memcg, nodemask))
KOSAKI Motohiro26ebc982010-08-09 17:19:37 -0700188 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189
Oleg Nesterovdd8e8f42010-08-09 17:18:45 -0700190 p = find_lock_task_mm(p);
191 if (!p)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700192 return 0;
193
Michal Hocko5aecc852011-11-15 14:36:07 -0800194 if (p->signal->oom_score_adj == OOM_SCORE_ADJ_MIN) {
195 task_unlock(p);
196 return 0;
197 }
198
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199 /*
David Rientjesa63d83f2010-08-09 17:19:46 -0700200 * The memory controller may have a limit of 0 bytes, so avoid a divide
201 * by zero, if necessary.
202 */
203 if (!totalpages)
204 totalpages = 1;
205
206 /*
207 * The baseline for the badness score is the proportion of RAM that each
KOSAKI Motohirof755a042011-04-27 15:26:50 -0700208 * task's rss, pagetable and swap space use.
David Rientjesa63d83f2010-08-09 17:19:46 -0700209 */
KOSAKI Motohirof755a042011-04-27 15:26:50 -0700210 points = get_mm_rss(p->mm) + p->mm->nr_ptes;
211 points += get_mm_counter(p->mm, MM_SWAPENTS);
212
213 points *= 1000;
214 points /= totalpages;
Andrew Morton97c2c9b2006-04-18 22:20:38 -0700215 task_unlock(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216
217 /*
David Rientjesa63d83f2010-08-09 17:19:46 -0700218 * Root processes get 3% bonus, just like the __vm_enough_memory()
219 * implementation used by LSMs.
Hugh Dickins7ba34852007-01-05 16:37:03 -0800220 */
David Rientjesa63d83f2010-08-09 17:19:46 -0700221 if (has_capability_noaudit(p, CAP_SYS_ADMIN))
222 points -= 30;
Hugh Dickins7ba34852007-01-05 16:37:03 -0800223
224 /*
David Rientjesa63d83f2010-08-09 17:19:46 -0700225 * /proc/pid/oom_score_adj ranges from -1000 to +1000 such that it may
226 * either completely disable oom killing or always prefer a certain
227 * task.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228 */
David Rientjesa63d83f2010-08-09 17:19:46 -0700229 points += p->signal->oom_score_adj;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230
David Rientjesf19e8aa2010-09-22 13:04:52 -0700231 /*
232 * Never return 0 for an eligible task that may be killed since it's
233 * possible that no single user task uses more than 0.1% of memory and
234 * no single admin tasks uses more than 3.0%.
235 */
236 if (points <= 0)
237 return 1;
David Rientjesa63d83f2010-08-09 17:19:46 -0700238 return (points < 1000) ? points : 1000;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239}
240
241/*
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800242 * Determine the type of allocation constraint.
243 */
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800244#ifdef CONFIG_NUMA
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800245static enum oom_constraint constrained_alloc(struct zonelist *zonelist,
David Rientjesa63d83f2010-08-09 17:19:46 -0700246 gfp_t gfp_mask, nodemask_t *nodemask,
247 unsigned long *totalpages)
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800248{
Mel Gorman54a6eb52008-04-28 02:12:16 -0700249 struct zone *zone;
Mel Gormandd1a2392008-04-28 02:12:17 -0700250 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -0700251 enum zone_type high_zoneidx = gfp_zone(gfp_mask);
David Rientjesa63d83f2010-08-09 17:19:46 -0700252 bool cpuset_limited = false;
253 int nid;
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800254
David Rientjesa63d83f2010-08-09 17:19:46 -0700255 /* Default to all available memory */
256 *totalpages = totalram_pages + total_swap_pages;
257
258 if (!zonelist)
259 return CONSTRAINT_NONE;
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800260 /*
261 * Reach here only when __GFP_NOFAIL is used. So, we should avoid
262 * to kill current.We have to random task kill in this case.
263 * Hopefully, CONSTRAINT_THISNODE...but no way to handle it, now.
264 */
265 if (gfp_mask & __GFP_THISNODE)
266 return CONSTRAINT_NONE;
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800267
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800268 /*
David Rientjesa63d83f2010-08-09 17:19:46 -0700269 * This is not a __GFP_THISNODE allocation, so a truncated nodemask in
270 * the page allocator means a mempolicy is in effect. Cpuset policy
271 * is enforced in get_page_from_freelist().
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800272 */
David Rientjesa63d83f2010-08-09 17:19:46 -0700273 if (nodemask && !nodes_subset(node_states[N_HIGH_MEMORY], *nodemask)) {
274 *totalpages = total_swap_pages;
275 for_each_node_mask(nid, *nodemask)
276 *totalpages += node_spanned_pages(nid);
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800277 return CONSTRAINT_MEMORY_POLICY;
David Rientjesa63d83f2010-08-09 17:19:46 -0700278 }
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800279
280 /* Check this allocation failure is caused by cpuset's wall function */
281 for_each_zone_zonelist_nodemask(zone, z, zonelist,
282 high_zoneidx, nodemask)
283 if (!cpuset_zone_allowed_softwall(zone, gfp_mask))
David Rientjesa63d83f2010-08-09 17:19:46 -0700284 cpuset_limited = true;
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800285
David Rientjesa63d83f2010-08-09 17:19:46 -0700286 if (cpuset_limited) {
287 *totalpages = total_swap_pages;
288 for_each_node_mask(nid, cpuset_current_mems_allowed)
289 *totalpages += node_spanned_pages(nid);
290 return CONSTRAINT_CPUSET;
291 }
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800292 return CONSTRAINT_NONE;
293}
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800294#else
295static enum oom_constraint constrained_alloc(struct zonelist *zonelist,
David Rientjesa63d83f2010-08-09 17:19:46 -0700296 gfp_t gfp_mask, nodemask_t *nodemask,
297 unsigned long *totalpages)
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800298{
David Rientjesa63d83f2010-08-09 17:19:46 -0700299 *totalpages = totalram_pages + total_swap_pages;
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800300 return CONSTRAINT_NONE;
301}
302#endif
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800303
304/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700305 * Simple selection loop. We chose the process with the highest
306 * number of 'points'. We expect the caller will lock the tasklist.
307 *
308 * (not docbooked, we don't want this one cluttering up the manual)
309 */
David Rientjesa63d83f2010-08-09 17:19:46 -0700310static struct task_struct *select_bad_process(unsigned int *ppoints,
Johannes Weiner72835c82012-01-12 17:18:32 -0800311 unsigned long totalpages, struct mem_cgroup *memcg,
David Rientjesa63d83f2010-08-09 17:19:46 -0700312 const nodemask_t *nodemask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700313{
David Rientjes3a5dda72011-03-22 16:30:09 -0700314 struct task_struct *g, *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700315 struct task_struct *chosen = NULL;
Kurt Garloff9827b782006-02-20 18:27:51 -0800316 *ppoints = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317
David Rientjes3a5dda72011-03-22 16:30:09 -0700318 do_each_thread(g, p) {
David Rientjesa63d83f2010-08-09 17:19:46 -0700319 unsigned int points;
Paul Jacksona49335c2005-09-06 15:18:09 -0700320
Oleg Nesterovc027a472011-07-30 16:35:02 +0200321 if (p->exit_state)
Andrey Vagin30e2b412011-03-22 16:30:11 -0700322 continue;
Johannes Weiner72835c82012-01-12 17:18:32 -0800323 if (oom_unkillable_task(p, memcg, nodemask))
David Rientjes6cf86ac2010-08-09 17:18:50 -0700324 continue;
Paul Jacksonef08e3b2005-09-06 15:18:13 -0700325
Paul Jacksona49335c2005-09-06 15:18:09 -0700326 /*
Nick Pigginb78483a2006-09-29 02:01:14 -0700327 * This task already has access to memory reserves and is
328 * being killed. Don't allow any other task access to the
329 * memory reserve.
330 *
331 * Note: this may have a chance of deadlock if it gets
332 * blocked waiting for another task which itself is waiting
333 * for memory. Is there a better alternative?
334 */
David Rientjesf660daa2011-10-31 17:07:07 -0700335 if (test_tsk_thread_flag(p, TIF_MEMDIE)) {
336 if (unlikely(frozen(p)))
Tejun Heoa5be2d02011-11-21 12:32:23 -0800337 __thaw_task(p);
Nick Pigginb78483a2006-09-29 02:01:14 -0700338 return ERR_PTR(-1UL);
David Rientjesf660daa2011-10-31 17:07:07 -0700339 }
Oleg Nesterovc027a472011-07-30 16:35:02 +0200340 if (!p->mm)
341 continue;
Nick Pigginb78483a2006-09-29 02:01:14 -0700342
Andrey Vagin30e2b412011-03-22 16:30:11 -0700343 if (p->flags & PF_EXITING) {
David Rientjesedd45542011-03-22 16:30:12 -0700344 /*
345 * If p is the current task and is in the process of
346 * releasing memory, we allow the "kill" to set
347 * TIF_MEMDIE, which will allow it to gain access to
348 * memory reserves. Otherwise, it may stall forever.
349 *
350 * The loop isn't broken here, however, in case other
351 * threads are found to have already been oom killed.
352 */
353 if (p == current) {
354 chosen = p;
355 *ppoints = 1000;
356 } else {
357 /*
358 * If this task is not being ptraced on exit,
359 * then wait for it to finish before killing
360 * some other task unnecessarily.
361 */
Tejun Heod21142e2011-06-17 16:50:34 +0200362 if (!(p->group_leader->ptrace & PT_TRACE_EXIT))
David Rientjesedd45542011-03-22 16:30:12 -0700363 return ERR_PTR(-1UL);
364 }
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700365 }
Oleg Nesterov972c4ea2006-09-29 02:01:12 -0700366
Johannes Weiner72835c82012-01-12 17:18:32 -0800367 points = oom_badness(p, memcg, nodemask, totalpages);
David Rientjesa63d83f2010-08-09 17:19:46 -0700368 if (points > *ppoints) {
Paul Jacksona49335c2005-09-06 15:18:09 -0700369 chosen = p;
Kurt Garloff9827b782006-02-20 18:27:51 -0800370 *ppoints = points;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700371 }
David Rientjes3a5dda72011-03-22 16:30:09 -0700372 } while_each_thread(g, p);
Oleg Nesterov972c4ea2006-09-29 02:01:12 -0700373
Linus Torvalds1da177e2005-04-16 15:20:36 -0700374 return chosen;
375}
376
377/**
Randy Dunlap1b578df2008-03-19 17:00:42 -0700378 * dump_tasks - dump current memory state of all system tasks
David Rientjes74ab7f12010-08-09 17:18:46 -0700379 * @mem: current's memory controller, if constrained
David Rientjese85bfd32010-09-22 13:05:10 -0700380 * @nodemask: nodemask passed to page allocator for mempolicy ooms
Randy Dunlap1b578df2008-03-19 17:00:42 -0700381 *
David Rientjese85bfd32010-09-22 13:05:10 -0700382 * Dumps the current memory state of all eligible tasks. Tasks not in the same
383 * memcg, not in the same cpuset, or bound to a disjoint set of mempolicy nodes
384 * are not shown.
David Rientjesfef1bdd2008-02-07 00:14:07 -0800385 * State information includes task's pid, uid, tgid, vm size, rss, cpu, oom_adj
David Rientjesa63d83f2010-08-09 17:19:46 -0700386 * value, oom_score_adj value, and name.
David Rientjesfef1bdd2008-02-07 00:14:07 -0800387 *
David Rientjesfef1bdd2008-02-07 00:14:07 -0800388 * Call with tasklist_lock read-locked.
389 */
Johannes Weiner72835c82012-01-12 17:18:32 -0800390static void dump_tasks(const struct mem_cgroup *memcg, const nodemask_t *nodemask)
David Rientjesfef1bdd2008-02-07 00:14:07 -0800391{
KOSAKI Motohiroc55db952010-08-09 17:18:46 -0700392 struct task_struct *p;
393 struct task_struct *task;
David Rientjesfef1bdd2008-02-07 00:14:07 -0800394
David Rientjesa63d83f2010-08-09 17:19:46 -0700395 pr_info("[ pid ] uid tgid total_vm rss cpu oom_adj oom_score_adj name\n");
KOSAKI Motohiroc55db952010-08-09 17:18:46 -0700396 for_each_process(p) {
Johannes Weiner72835c82012-01-12 17:18:32 -0800397 if (oom_unkillable_task(p, memcg, nodemask))
David Rientjesfef1bdd2008-02-07 00:14:07 -0800398 continue;
399
KOSAKI Motohiroc55db952010-08-09 17:18:46 -0700400 task = find_lock_task_mm(p);
401 if (!task) {
David Rientjes6d2661e2009-05-28 14:34:19 -0700402 /*
David Rientjes74ab7f12010-08-09 17:18:46 -0700403 * This is a kthread or all of p's threads have already
404 * detached their mm's. There's no need to report
KOSAKI Motohiroc55db952010-08-09 17:18:46 -0700405 * them; they can't be oom killed anyway.
David Rientjes6d2661e2009-05-28 14:34:19 -0700406 */
David Rientjes6d2661e2009-05-28 14:34:19 -0700407 continue;
408 }
KOSAKI Motohiroc55db952010-08-09 17:18:46 -0700409
David Rientjesa63d83f2010-08-09 17:19:46 -0700410 pr_info("[%5d] %5d %5d %8lu %8lu %3u %3d %5d %s\n",
KOSAKI Motohiro8d6c83f2010-08-19 14:13:39 -0700411 task->pid, task_uid(task), task->tgid,
David Rientjesa63d83f2010-08-09 17:19:46 -0700412 task->mm->total_vm, get_mm_rss(task->mm),
413 task_cpu(task), task->signal->oom_adj,
414 task->signal->oom_score_adj, task->comm);
KOSAKI Motohiroc55db952010-08-09 17:18:46 -0700415 task_unlock(task);
416 }
David Rientjesfef1bdd2008-02-07 00:14:07 -0800417}
418
Daisuke Nishimurad31f56d2009-12-15 16:47:12 -0800419static void dump_header(struct task_struct *p, gfp_t gfp_mask, int order,
Johannes Weiner72835c82012-01-12 17:18:32 -0800420 struct mem_cgroup *memcg, const nodemask_t *nodemask)
David Rientjes1b604d72009-12-14 17:57:47 -0800421{
David Rientjes5e9d8342010-08-09 17:18:51 -0700422 task_lock(current);
David Rientjes1b604d72009-12-14 17:57:47 -0800423 pr_warning("%s invoked oom-killer: gfp_mask=0x%x, order=%d, "
David Rientjesa63d83f2010-08-09 17:19:46 -0700424 "oom_adj=%d, oom_score_adj=%d\n",
425 current->comm, gfp_mask, order, current->signal->oom_adj,
426 current->signal->oom_score_adj);
David Rientjes1b604d72009-12-14 17:57:47 -0800427 cpuset_print_task_mems_allowed(current);
428 task_unlock(current);
429 dump_stack();
Johannes Weiner72835c82012-01-12 17:18:32 -0800430 mem_cgroup_print_oom_info(memcg, p);
David Rientjesb2b755b2011-03-24 15:18:15 -0700431 show_mem(SHOW_MEM_FILTER_NODES);
David Rientjes1b604d72009-12-14 17:57:47 -0800432 if (sysctl_oom_dump_tasks)
Johannes Weiner72835c82012-01-12 17:18:32 -0800433 dump_tasks(memcg, nodemask);
David Rientjes1b604d72009-12-14 17:57:47 -0800434}
435
KOSAKI Motohiro3b4798c2009-12-15 16:45:32 -0800436#define K(x) ((x) << (PAGE_SHIFT-10))
Johannes Weinerec0fffd2012-01-12 17:18:29 -0800437static int oom_kill_task(struct task_struct *p)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438{
David Rientjes1e99bad2010-10-26 14:21:24 -0700439 struct task_struct *q;
440 struct mm_struct *mm;
441
David Rientjesb940fd72010-08-09 17:18:57 -0700442 p = find_lock_task_mm(p);
KOSAKI Motohirobe71cf22010-08-19 14:13:38 -0700443 if (!p)
Dave Peterson01315922006-04-18 22:20:44 -0700444 return 1;
KOSAKI Motohirobe71cf22010-08-19 14:13:38 -0700445
David Rientjes1e99bad2010-10-26 14:21:24 -0700446 /* mm cannot be safely dereferenced after task_unlock(p) */
447 mm = p->mm;
448
David Rientjesb940fd72010-08-09 17:18:57 -0700449 pr_err("Killed process %d (%s) total-vm:%lukB, anon-rss:%lukB, file-rss:%lukB\n",
450 task_pid_nr(p), p->comm, K(p->mm->total_vm),
451 K(get_mm_counter(p->mm, MM_ANONPAGES)),
452 K(get_mm_counter(p->mm, MM_FILEPAGES)));
453 task_unlock(p);
KOSAKI Motohiro0753ba02009-08-18 14:11:10 -0700454
David Rientjes1e99bad2010-10-26 14:21:24 -0700455 /*
David Rientjes7b0d44f2011-10-31 17:07:11 -0700456 * Kill all user processes sharing p->mm in other thread groups, if any.
David Rientjes1e99bad2010-10-26 14:21:24 -0700457 * They don't get access to memory reserves or a higher scheduler
458 * priority, though, to avoid depletion of all memory or task
459 * starvation. This prevents mm->mmap_sem livelock when an oom killed
460 * task cannot exit because it requires the semaphore and its contended
461 * by another thread trying to allocate memory itself. That thread will
462 * now get access to memory reserves since it has a pending fatal
463 * signal.
464 */
465 for_each_process(q)
David Rientjes7b0d44f2011-10-31 17:07:11 -0700466 if (q->mm == mm && !same_thread_group(q, p) &&
467 !(q->flags & PF_KTHREAD)) {
David Rientjesc9f01242011-10-31 17:07:15 -0700468 if (q->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
469 continue;
470
David Rientjes1e99bad2010-10-26 14:21:24 -0700471 task_lock(q); /* Protect ->comm from prctl() */
472 pr_err("Kill process %d (%s) sharing same memory\n",
473 task_pid_nr(q), q->comm);
474 task_unlock(q);
475 force_sig(SIGKILL, q);
476 }
Luis Claudio R. Goncalves93b43fa2010-08-09 17:19:41 -0700477
David Rientjesb940fd72010-08-09 17:18:57 -0700478 set_tsk_thread_flag(p, TIF_MEMDIE);
479 force_sig(SIGKILL, p);
Luis Claudio R. Goncalves93b43fa2010-08-09 17:19:41 -0700480
Dave Peterson01315922006-04-18 22:20:44 -0700481 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482}
David Rientjesb940fd72010-08-09 17:18:57 -0700483#undef K
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484
David Rientjes7213f502007-10-16 23:25:57 -0700485static int oom_kill_process(struct task_struct *p, gfp_t gfp_mask, int order,
David Rientjesa63d83f2010-08-09 17:19:46 -0700486 unsigned int points, unsigned long totalpages,
Johannes Weiner72835c82012-01-12 17:18:32 -0800487 struct mem_cgroup *memcg, nodemask_t *nodemask,
David Rientjesa63d83f2010-08-09 17:19:46 -0700488 const char *message)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489{
Linus Torvalds52d3c032011-03-14 15:17:07 -0700490 struct task_struct *victim = p;
David Rientjes5e9d8342010-08-09 17:18:51 -0700491 struct task_struct *child;
Linus Torvalds52d3c032011-03-14 15:17:07 -0700492 struct task_struct *t = p;
493 unsigned int victim_points = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494
David Rientjes1b604d72009-12-14 17:57:47 -0800495 if (printk_ratelimit())
Johannes Weiner72835c82012-01-12 17:18:32 -0800496 dump_header(p, gfp_mask, order, memcg, nodemask);
David Rientjes7213f502007-10-16 23:25:57 -0700497
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700498 /*
499 * If the task is already exiting, don't alarm the sysadmin or kill
500 * its children or threads, just set TIF_MEMDIE so it can die quickly
501 */
KOSAKI Motohiro0753ba02009-08-18 14:11:10 -0700502 if (p->flags & PF_EXITING) {
David Rientjes43589972010-08-09 17:18:49 -0700503 set_tsk_thread_flag(p, TIF_MEMDIE);
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700504 return 0;
505 }
506
David Rientjes5e9d8342010-08-09 17:18:51 -0700507 task_lock(p);
David Rientjesa63d83f2010-08-09 17:19:46 -0700508 pr_err("%s: Kill process %d (%s) score %d or sacrifice child\n",
David Rientjes5e9d8342010-08-09 17:18:51 -0700509 message, task_pid_nr(p), p->comm, points);
510 task_unlock(p);
Nick Pigginf3af38d2006-12-06 20:31:51 -0800511
David Rientjes5e9d8342010-08-09 17:18:51 -0700512 /*
513 * If any of p's children has a different mm and is eligible for kill,
David Rientjes11239832011-07-25 17:12:17 -0700514 * the one with the highest oom_badness() score is sacrificed for its
David Rientjes5e9d8342010-08-09 17:18:51 -0700515 * parent. This attempts to lose the minimal amount of work done while
516 * still freeing memory.
517 */
Oleg Nesterovdd8e8f42010-08-09 17:18:45 -0700518 do {
David Rientjes5e9d8342010-08-09 17:18:51 -0700519 list_for_each_entry(child, &t->children, sibling) {
David Rientjesa63d83f2010-08-09 17:19:46 -0700520 unsigned int child_points;
David Rientjes5e9d8342010-08-09 17:18:51 -0700521
David Rientjesedd45542011-03-22 16:30:12 -0700522 if (child->mm == p->mm)
523 continue;
David Rientjesa63d83f2010-08-09 17:19:46 -0700524 /*
525 * oom_badness() returns 0 if the thread is unkillable
526 */
Johannes Weiner72835c82012-01-12 17:18:32 -0800527 child_points = oom_badness(child, memcg, nodemask,
David Rientjesa63d83f2010-08-09 17:19:46 -0700528 totalpages);
David Rientjes5e9d8342010-08-09 17:18:51 -0700529 if (child_points > victim_points) {
530 victim = child;
531 victim_points = child_points;
532 }
Oleg Nesterovdd8e8f42010-08-09 17:18:45 -0700533 }
534 } while_each_thread(p, t);
535
Johannes Weinerec0fffd2012-01-12 17:18:29 -0800536 return oom_kill_task(victim);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700537}
538
David Rientjes309ed882010-08-09 17:18:54 -0700539/*
540 * Determines whether the kernel must panic because of the panic_on_oom sysctl.
541 */
542static void check_panic_on_oom(enum oom_constraint constraint, gfp_t gfp_mask,
David Rientjese85bfd32010-09-22 13:05:10 -0700543 int order, const nodemask_t *nodemask)
David Rientjes309ed882010-08-09 17:18:54 -0700544{
545 if (likely(!sysctl_panic_on_oom))
546 return;
547 if (sysctl_panic_on_oom != 2) {
548 /*
549 * panic_on_oom == 1 only affects CONSTRAINT_NONE, the kernel
550 * does not panic for cpuset, mempolicy, or memcg allocation
551 * failures.
552 */
553 if (constraint != CONSTRAINT_NONE)
554 return;
555 }
556 read_lock(&tasklist_lock);
David Rientjese85bfd32010-09-22 13:05:10 -0700557 dump_header(NULL, gfp_mask, order, NULL, nodemask);
David Rientjes309ed882010-08-09 17:18:54 -0700558 read_unlock(&tasklist_lock);
559 panic("Out of memory: %s panic_on_oom is enabled\n",
560 sysctl_panic_on_oom == 2 ? "compulsory" : "system-wide");
561}
562
Balbir Singh00f0b822008-03-04 14:28:39 -0800563#ifdef CONFIG_CGROUP_MEM_RES_CTLR
Johannes Weiner72835c82012-01-12 17:18:32 -0800564void mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask)
Pavel Emelianovc7ba5c92008-02-07 00:13:58 -0800565{
David Rientjesa63d83f2010-08-09 17:19:46 -0700566 unsigned long limit;
567 unsigned int points = 0;
Pavel Emelianovc7ba5c92008-02-07 00:13:58 -0800568 struct task_struct *p;
569
David Rientjesf9434ad2011-03-23 16:42:44 -0700570 /*
571 * If current has a pending SIGKILL, then automatically select it. The
572 * goal is to allow it to allocate so that it may quickly exit and free
573 * its memory.
574 */
575 if (fatal_signal_pending(current)) {
576 set_thread_flag(TIF_MEMDIE);
David Rientjesf9434ad2011-03-23 16:42:44 -0700577 return;
578 }
579
David Rientjese85bfd32010-09-22 13:05:10 -0700580 check_panic_on_oom(CONSTRAINT_MEMCG, gfp_mask, 0, NULL);
Johannes Weiner72835c82012-01-12 17:18:32 -0800581 limit = mem_cgroup_get_limit(memcg) >> PAGE_SHIFT;
Li Zefane115f2d2008-04-15 14:34:37 -0700582 read_lock(&tasklist_lock);
Pavel Emelianovc7ba5c92008-02-07 00:13:58 -0800583retry:
Johannes Weiner72835c82012-01-12 17:18:32 -0800584 p = select_bad_process(&points, limit, memcg, NULL);
David Rientjesdf64f812010-05-26 14:42:41 -0700585 if (!p || PTR_ERR(p) == -1UL)
Pavel Emelianovc7ba5c92008-02-07 00:13:58 -0800586 goto out;
587
Johannes Weiner72835c82012-01-12 17:18:32 -0800588 if (oom_kill_process(p, gfp_mask, 0, points, limit, memcg, NULL,
Pavel Emelianovc7ba5c92008-02-07 00:13:58 -0800589 "Memory cgroup out of memory"))
590 goto retry;
591out:
Li Zefane115f2d2008-04-15 14:34:37 -0700592 read_unlock(&tasklist_lock);
Pavel Emelianovc7ba5c92008-02-07 00:13:58 -0800593}
594#endif
595
Martin Schwidefsky8bc719d2006-09-25 23:31:20 -0700596static BLOCKING_NOTIFIER_HEAD(oom_notify_list);
597
598int register_oom_notifier(struct notifier_block *nb)
599{
600 return blocking_notifier_chain_register(&oom_notify_list, nb);
601}
602EXPORT_SYMBOL_GPL(register_oom_notifier);
603
604int unregister_oom_notifier(struct notifier_block *nb)
605{
606 return blocking_notifier_chain_unregister(&oom_notify_list, nb);
607}
608EXPORT_SYMBOL_GPL(unregister_oom_notifier);
609
David Rientjes098d7f12007-10-16 23:25:55 -0700610/*
611 * Try to acquire the OOM killer lock for the zones in zonelist. Returns zero
612 * if a parallel OOM killing is already taking place that includes a zone in
613 * the zonelist. Otherwise, locks all zones in the zonelist and returns 1.
614 */
Minchan Kimff321fe2010-08-09 17:18:57 -0700615int try_set_zonelist_oom(struct zonelist *zonelist, gfp_t gfp_mask)
David Rientjes098d7f12007-10-16 23:25:55 -0700616{
Mel Gormandd1a2392008-04-28 02:12:17 -0700617 struct zoneref *z;
618 struct zone *zone;
David Rientjes098d7f12007-10-16 23:25:55 -0700619 int ret = 1;
620
David Rientjesc7d4cae2009-01-06 14:39:00 -0800621 spin_lock(&zone_scan_lock);
Mel Gormandd1a2392008-04-28 02:12:17 -0700622 for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) {
623 if (zone_is_oom_locked(zone)) {
David Rientjes098d7f12007-10-16 23:25:55 -0700624 ret = 0;
625 goto out;
626 }
Mel Gormandd1a2392008-04-28 02:12:17 -0700627 }
David Rientjes098d7f12007-10-16 23:25:55 -0700628
Mel Gormandd1a2392008-04-28 02:12:17 -0700629 for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) {
630 /*
David Rientjesc7d4cae2009-01-06 14:39:00 -0800631 * Lock each zone in the zonelist under zone_scan_lock so a
Minchan Kimff321fe2010-08-09 17:18:57 -0700632 * parallel invocation of try_set_zonelist_oom() doesn't succeed
Mel Gormandd1a2392008-04-28 02:12:17 -0700633 * when it shouldn't.
634 */
635 zone_set_flag(zone, ZONE_OOM_LOCKED);
636 }
637
David Rientjes098d7f12007-10-16 23:25:55 -0700638out:
David Rientjesc7d4cae2009-01-06 14:39:00 -0800639 spin_unlock(&zone_scan_lock);
David Rientjes098d7f12007-10-16 23:25:55 -0700640 return ret;
641}
642
643/*
644 * Clears the ZONE_OOM_LOCKED flag for all zones in the zonelist so that failed
645 * allocation attempts with zonelists containing them may now recall the OOM
646 * killer, if necessary.
647 */
Mel Gormandd1a2392008-04-28 02:12:17 -0700648void clear_zonelist_oom(struct zonelist *zonelist, gfp_t gfp_mask)
David Rientjes098d7f12007-10-16 23:25:55 -0700649{
Mel Gormandd1a2392008-04-28 02:12:17 -0700650 struct zoneref *z;
651 struct zone *zone;
David Rientjes098d7f12007-10-16 23:25:55 -0700652
David Rientjesc7d4cae2009-01-06 14:39:00 -0800653 spin_lock(&zone_scan_lock);
Mel Gormandd1a2392008-04-28 02:12:17 -0700654 for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) {
655 zone_clear_flag(zone, ZONE_OOM_LOCKED);
656 }
David Rientjesc7d4cae2009-01-06 14:39:00 -0800657 spin_unlock(&zone_scan_lock);
David Rientjes098d7f12007-10-16 23:25:55 -0700658}
659
Nick Piggin1c0fe6e2009-01-06 14:38:59 -0800660/*
David Rientjese3658932010-08-09 17:18:55 -0700661 * Try to acquire the oom killer lock for all system zones. Returns zero if a
662 * parallel oom killing is taking place, otherwise locks all zones and returns
663 * non-zero.
664 */
665static int try_set_system_oom(void)
666{
667 struct zone *zone;
668 int ret = 1;
669
670 spin_lock(&zone_scan_lock);
671 for_each_populated_zone(zone)
672 if (zone_is_oom_locked(zone)) {
673 ret = 0;
674 goto out;
675 }
676 for_each_populated_zone(zone)
677 zone_set_flag(zone, ZONE_OOM_LOCKED);
678out:
679 spin_unlock(&zone_scan_lock);
680 return ret;
681}
682
683/*
684 * Clears ZONE_OOM_LOCKED for all system zones so that failed allocation
685 * attempts or page faults may now recall the oom killer, if necessary.
686 */
687static void clear_system_oom(void)
688{
689 struct zone *zone;
690
691 spin_lock(&zone_scan_lock);
692 for_each_populated_zone(zone)
693 zone_clear_flag(zone, ZONE_OOM_LOCKED);
694 spin_unlock(&zone_scan_lock);
695}
696
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697/**
Dave Peterson6937a252006-06-23 02:03:13 -0700698 * out_of_memory - kill the "best" process when we run out of memory
Randy Dunlap1b578df2008-03-19 17:00:42 -0700699 * @zonelist: zonelist pointer
700 * @gfp_mask: memory allocation flags
701 * @order: amount of memory being requested as a power of 2
David Rientjes6f48d0eb2010-08-09 17:18:52 -0700702 * @nodemask: nodemask passed to page allocator
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703 *
704 * If we run out of memory, we have the choice between either
705 * killing a random task (bad), letting the system crash (worse)
706 * OR try to be smart about which process to kill. Note that we
707 * don't have to be perfect here, we just have to be good.
708 */
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800709void out_of_memory(struct zonelist *zonelist, gfp_t gfp_mask,
710 int order, nodemask_t *nodemask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711{
David Rientjese85bfd32010-09-22 13:05:10 -0700712 const nodemask_t *mpol_mask;
David Rientjes0aad4b32010-08-09 17:18:59 -0700713 struct task_struct *p;
David Rientjesa63d83f2010-08-09 17:19:46 -0700714 unsigned long totalpages;
Martin Schwidefsky8bc719d2006-09-25 23:31:20 -0700715 unsigned long freed = 0;
David Rientjesa63d83f2010-08-09 17:19:46 -0700716 unsigned int points;
David Rientjese3658932010-08-09 17:18:55 -0700717 enum oom_constraint constraint = CONSTRAINT_NONE;
KOSAKI Motohirob52723c2010-08-19 14:13:39 -0700718 int killed = 0;
Martin Schwidefsky8bc719d2006-09-25 23:31:20 -0700719
720 blocking_notifier_call_chain(&oom_notify_list, 0, &freed);
721 if (freed > 0)
722 /* Got some memory back in the last second. */
723 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700724
David Rientjes7b98c2e2010-08-09 17:18:48 -0700725 /*
726 * If current has a pending SIGKILL, then automatically select it. The
727 * goal is to allow it to allocate so that it may quickly exit and free
728 * its memory.
729 */
730 if (fatal_signal_pending(current)) {
731 set_thread_flag(TIF_MEMDIE);
732 return;
733 }
734
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800735 /*
736 * Check if there were limitations on the allocation (only relevant for
737 * NUMA) that may require different handling.
738 */
David Rientjesa63d83f2010-08-09 17:19:46 -0700739 constraint = constrained_alloc(zonelist, gfp_mask, nodemask,
740 &totalpages);
David Rientjese85bfd32010-09-22 13:05:10 -0700741 mpol_mask = (constraint == CONSTRAINT_MEMORY_POLICY) ? nodemask : NULL;
742 check_panic_on_oom(constraint, gfp_mask, order, mpol_mask);
David Rientjes0aad4b32010-08-09 17:18:59 -0700743
David Rientjes2b45ab32007-05-06 14:50:00 -0700744 read_lock(&tasklist_lock);
KOSAKI Motohirof88ccad2010-08-09 17:19:36 -0700745 if (sysctl_oom_kill_allocating_task &&
KOSAKI Motohiroa96cfd62010-08-09 17:19:38 -0700746 !oom_unkillable_task(current, NULL, nodemask) &&
David Rientjesc9f01242011-10-31 17:07:15 -0700747 current->mm) {
David Rientjes0aad4b32010-08-09 17:18:59 -0700748 /*
749 * oom_kill_process() needs tasklist_lock held. If it returns
750 * non-zero, current could not be killed so we must fallback to
751 * the tasklist scan.
752 */
David Rientjesa63d83f2010-08-09 17:19:46 -0700753 if (!oom_kill_process(current, gfp_mask, order, 0, totalpages,
754 NULL, nodemask,
David Rientjes0aad4b32010-08-09 17:18:59 -0700755 "Out of memory (oom_kill_allocating_task)"))
KOSAKI Motohirob52723c2010-08-19 14:13:39 -0700756 goto out;
David Rientjes0aad4b32010-08-09 17:18:59 -0700757 }
758
759retry:
David Rientjese85bfd32010-09-22 13:05:10 -0700760 p = select_bad_process(&points, totalpages, NULL, mpol_mask);
David Rientjes0aad4b32010-08-09 17:18:59 -0700761 if (PTR_ERR(p) == -1UL)
KOSAKI Motohirob52723c2010-08-19 14:13:39 -0700762 goto out;
David Rientjes0aad4b32010-08-09 17:18:59 -0700763
764 /* Found nothing?!?! Either we hang forever, or we panic. */
765 if (!p) {
David Rientjese85bfd32010-09-22 13:05:10 -0700766 dump_header(NULL, gfp_mask, order, NULL, mpol_mask);
David Rientjes0aad4b32010-08-09 17:18:59 -0700767 read_unlock(&tasklist_lock);
768 panic("Out of memory and no killable processes...\n");
769 }
770
David Rientjesa63d83f2010-08-09 17:19:46 -0700771 if (oom_kill_process(p, gfp_mask, order, points, totalpages, NULL,
772 nodemask, "Out of memory"))
David Rientjes0aad4b32010-08-09 17:18:59 -0700773 goto retry;
KOSAKI Motohirob52723c2010-08-19 14:13:39 -0700774 killed = 1;
775out:
Andrew Morton140ffce2006-03-02 02:54:28 -0800776 read_unlock(&tasklist_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777
778 /*
779 * Give "p" a good chance of killing itself before we
Kirill Korotaev2f659f42006-01-08 01:01:05 -0800780 * retry to allocate memory unless "p" is current
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781 */
KOSAKI Motohirob52723c2010-08-19 14:13:39 -0700782 if (killed && !test_thread_flag(TIF_MEMDIE))
Andrew Morton140ffce2006-03-02 02:54:28 -0800783 schedule_timeout_uninterruptible(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784}
David Rientjese3658932010-08-09 17:18:55 -0700785
786/*
787 * The pagefault handler calls here because it is out of memory, so kill a
788 * memory-hogging task. If a populated zone has ZONE_OOM_LOCKED set, a parallel
789 * oom killing is already in progress so do nothing. If a task is found with
790 * TIF_MEMDIE set, it has been killed so do nothing and allow it to exit.
791 */
792void pagefault_out_of_memory(void)
793{
794 if (try_set_system_oom()) {
795 out_of_memory(NULL, 0, 0, NULL);
796 clear_system_oom();
797 }
798 if (!test_thread_flag(TIF_MEMDIE))
799 schedule_timeout_uninterruptible(1);
800}