blob: 069b64e521fccf2725199550653bfdb7055d6f81 [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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070036
KAMEZAWA Hiroyukifadd8fb2006-06-23 02:03:13 -070037int sysctl_panic_on_oom;
David Rientjesfe071d72007-10-16 23:25:56 -070038int sysctl_oom_kill_allocating_task;
David Rientjesad915c42010-08-09 17:18:53 -070039int sysctl_oom_dump_tasks = 1;
David Rientjesc7d4cae2009-01-06 14:39:00 -080040static DEFINE_SPINLOCK(zone_scan_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070041
David Rientjes43362a42011-10-31 17:07:18 -070042/*
43 * compare_swap_oom_score_adj() - compare and swap current's oom_score_adj
44 * @old_val: old oom_score_adj for compare
45 * @new_val: new oom_score_adj for swap
46 *
47 * Sets the oom_score_adj value for current to @new_val iff its present value is
48 * @old_val. Usually used to reinstate a previous value to prevent racing with
49 * userspacing tuning the value in the interim.
50 */
51void compare_swap_oom_score_adj(int old_val, int new_val)
52{
53 struct sighand_struct *sighand = current->sighand;
54
55 spin_lock_irq(&sighand->siglock);
56 if (current->signal->oom_score_adj == old_val)
57 current->signal->oom_score_adj = new_val;
58 spin_unlock_irq(&sighand->siglock);
59}
60
David Rientjes72788c32011-05-24 17:11:40 -070061/**
62 * test_set_oom_score_adj() - set current's oom_score_adj and return old value
63 * @new_val: new oom_score_adj value
64 *
65 * Sets the oom_score_adj value for current to @new_val with proper
66 * synchronization and returns the old value. Usually used to temporarily
67 * set a value, save the old value in the caller, and then reinstate it later.
68 */
69int test_set_oom_score_adj(int new_val)
70{
71 struct sighand_struct *sighand = current->sighand;
72 int old_val;
73
74 spin_lock_irq(&sighand->siglock);
75 old_val = current->signal->oom_score_adj;
David Rientjesc9f01242011-10-31 17:07:15 -070076 current->signal->oom_score_adj = new_val;
David Rientjes72788c32011-05-24 17:11:40 -070077 spin_unlock_irq(&sighand->siglock);
78
79 return old_val;
80}
81
David Rientjes6f48d0eb2010-08-09 17:18:52 -070082#ifdef CONFIG_NUMA
83/**
84 * has_intersects_mems_allowed() - check task eligiblity for kill
85 * @tsk: task struct of which task to consider
86 * @mask: nodemask passed to page allocator for mempolicy ooms
87 *
88 * Task eligibility is determined by whether or not a candidate task, @tsk,
89 * shares the same mempolicy nodes as current if it is bound by such a policy
90 * and whether or not it has the same set of allowed cpuset nodes.
KOSAKI Motohiro495789a2009-09-21 17:03:14 -070091 */
David Rientjes6f48d0eb2010-08-09 17:18:52 -070092static bool has_intersects_mems_allowed(struct task_struct *tsk,
93 const nodemask_t *mask)
KOSAKI Motohiro495789a2009-09-21 17:03:14 -070094{
David Rientjes6f48d0eb2010-08-09 17:18:52 -070095 struct task_struct *start = tsk;
KOSAKI Motohiro495789a2009-09-21 17:03:14 -070096
KOSAKI Motohiro495789a2009-09-21 17:03:14 -070097 do {
David Rientjes6f48d0eb2010-08-09 17:18:52 -070098 if (mask) {
99 /*
100 * If this is a mempolicy constrained oom, tsk's
101 * cpuset is irrelevant. Only return true if its
102 * mempolicy intersects current, otherwise it may be
103 * needlessly killed.
104 */
105 if (mempolicy_nodemask_intersects(tsk, mask))
106 return true;
107 } else {
108 /*
109 * This is not a mempolicy constrained oom, so only
110 * check the mems of tsk's cpuset.
111 */
112 if (cpuset_mems_allowed_intersects(current, tsk))
113 return true;
114 }
KOSAKI Motohirodf1090a2010-08-09 17:19:39 -0700115 } while_each_thread(start, tsk);
116
David Rientjes6f48d0eb2010-08-09 17:18:52 -0700117 return false;
KOSAKI Motohiro495789a2009-09-21 17:03:14 -0700118}
David Rientjes6f48d0eb2010-08-09 17:18:52 -0700119#else
120static bool has_intersects_mems_allowed(struct task_struct *tsk,
121 const nodemask_t *mask)
122{
123 return true;
124}
125#endif /* CONFIG_NUMA */
KOSAKI Motohiro495789a2009-09-21 17:03:14 -0700126
David Rientjes6f48d0eb2010-08-09 17:18:52 -0700127/*
128 * The process p may have detached its own ->mm while exiting or through
129 * use_mm(), but one or more of its subthreads may still have a valid
130 * pointer. Return p, or any of its subthreads with a valid ->mm, with
131 * task_lock() held.
132 */
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -0700133struct task_struct *find_lock_task_mm(struct task_struct *p)
Oleg Nesterovdd8e8f42010-08-09 17:18:45 -0700134{
135 struct task_struct *t = p;
136
137 do {
138 task_lock(t);
139 if (likely(t->mm))
140 return t;
141 task_unlock(t);
142 } while_each_thread(p, t);
143
144 return NULL;
145}
146
KOSAKI Motohiroab290ad2010-08-09 17:19:35 -0700147/* return true if the task is not adequate as candidate victim task. */
David Rientjese85bfd32010-09-22 13:05:10 -0700148static bool oom_unkillable_task(struct task_struct *p,
149 const struct mem_cgroup *mem, const nodemask_t *nodemask)
KOSAKI Motohiroab290ad2010-08-09 17:19:35 -0700150{
151 if (is_global_init(p))
152 return true;
153 if (p->flags & PF_KTHREAD)
154 return true;
155
156 /* When mem_cgroup_out_of_memory() and p is not member of the group */
157 if (mem && !task_in_mem_cgroup(p, mem))
158 return true;
159
160 /* p may not have freeable memory in nodemask */
161 if (!has_intersects_mems_allowed(p, nodemask))
162 return true;
163
164 return false;
165}
166
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167/**
David Rientjesa63d83f2010-08-09 17:19:46 -0700168 * oom_badness - heuristic function to determine which candidate task to kill
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169 * @p: task struct of which task we should calculate
David Rientjesa63d83f2010-08-09 17:19:46 -0700170 * @totalpages: total present RAM allowed for page allocation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171 *
David Rientjesa63d83f2010-08-09 17:19:46 -0700172 * The heuristic for determining which task to kill is made to be as simple and
173 * predictable as possible. The goal is to return the highest value for the
174 * task consuming the most memory to avoid subsequent oom failures.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175 */
David Rientjesa63d83f2010-08-09 17:19:46 -0700176unsigned int oom_badness(struct task_struct *p, struct mem_cgroup *mem,
177 const nodemask_t *nodemask, unsigned long totalpages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178{
Frantisek Hrbataff05b6f2011-12-19 17:11:59 -0800179 long points;
KOSAKI Motohiro28b83c52009-09-21 17:03:13 -0700180
KOSAKI Motohiro26ebc982010-08-09 17:19:37 -0700181 if (oom_unkillable_task(p, mem, nodemask))
182 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183
Oleg Nesterovdd8e8f42010-08-09 17:18:45 -0700184 p = find_lock_task_mm(p);
185 if (!p)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186 return 0;
187
Michal Hocko5aecc852011-11-15 14:36:07 -0800188 if (p->signal->oom_score_adj == OOM_SCORE_ADJ_MIN) {
189 task_unlock(p);
190 return 0;
191 }
192
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193 /*
David Rientjesa63d83f2010-08-09 17:19:46 -0700194 * The memory controller may have a limit of 0 bytes, so avoid a divide
195 * by zero, if necessary.
196 */
197 if (!totalpages)
198 totalpages = 1;
199
200 /*
201 * The baseline for the badness score is the proportion of RAM that each
KOSAKI Motohirof755a042011-04-27 15:26:50 -0700202 * task's rss, pagetable and swap space use.
David Rientjesa63d83f2010-08-09 17:19:46 -0700203 */
KOSAKI Motohirof755a042011-04-27 15:26:50 -0700204 points = get_mm_rss(p->mm) + p->mm->nr_ptes;
205 points += get_mm_counter(p->mm, MM_SWAPENTS);
206
207 points *= 1000;
208 points /= totalpages;
Andrew Morton97c2c9b2006-04-18 22:20:38 -0700209 task_unlock(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700210
211 /*
David Rientjesa63d83f2010-08-09 17:19:46 -0700212 * Root processes get 3% bonus, just like the __vm_enough_memory()
213 * implementation used by LSMs.
Hugh Dickins7ba34852007-01-05 16:37:03 -0800214 */
David Rientjesa63d83f2010-08-09 17:19:46 -0700215 if (has_capability_noaudit(p, CAP_SYS_ADMIN))
216 points -= 30;
Hugh Dickins7ba34852007-01-05 16:37:03 -0800217
218 /*
David Rientjesa63d83f2010-08-09 17:19:46 -0700219 * /proc/pid/oom_score_adj ranges from -1000 to +1000 such that it may
220 * either completely disable oom killing or always prefer a certain
221 * task.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222 */
David Rientjesa63d83f2010-08-09 17:19:46 -0700223 points += p->signal->oom_score_adj;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224
David Rientjesf19e8aa2010-09-22 13:04:52 -0700225 /*
226 * Never return 0 for an eligible task that may be killed since it's
227 * possible that no single user task uses more than 0.1% of memory and
228 * no single admin tasks uses more than 3.0%.
229 */
230 if (points <= 0)
231 return 1;
David Rientjesa63d83f2010-08-09 17:19:46 -0700232 return (points < 1000) ? points : 1000;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700233}
234
235/*
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800236 * Determine the type of allocation constraint.
237 */
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800238#ifdef CONFIG_NUMA
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800239static enum oom_constraint constrained_alloc(struct zonelist *zonelist,
David Rientjesa63d83f2010-08-09 17:19:46 -0700240 gfp_t gfp_mask, nodemask_t *nodemask,
241 unsigned long *totalpages)
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800242{
Mel Gorman54a6eb52008-04-28 02:12:16 -0700243 struct zone *zone;
Mel Gormandd1a2392008-04-28 02:12:17 -0700244 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -0700245 enum zone_type high_zoneidx = gfp_zone(gfp_mask);
David Rientjesa63d83f2010-08-09 17:19:46 -0700246 bool cpuset_limited = false;
247 int nid;
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800248
David Rientjesa63d83f2010-08-09 17:19:46 -0700249 /* Default to all available memory */
250 *totalpages = totalram_pages + total_swap_pages;
251
252 if (!zonelist)
253 return CONSTRAINT_NONE;
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800254 /*
255 * Reach here only when __GFP_NOFAIL is used. So, we should avoid
256 * to kill current.We have to random task kill in this case.
257 * Hopefully, CONSTRAINT_THISNODE...but no way to handle it, now.
258 */
259 if (gfp_mask & __GFP_THISNODE)
260 return CONSTRAINT_NONE;
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800261
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800262 /*
David Rientjesa63d83f2010-08-09 17:19:46 -0700263 * This is not a __GFP_THISNODE allocation, so a truncated nodemask in
264 * the page allocator means a mempolicy is in effect. Cpuset policy
265 * is enforced in get_page_from_freelist().
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800266 */
David Rientjesa63d83f2010-08-09 17:19:46 -0700267 if (nodemask && !nodes_subset(node_states[N_HIGH_MEMORY], *nodemask)) {
268 *totalpages = total_swap_pages;
269 for_each_node_mask(nid, *nodemask)
270 *totalpages += node_spanned_pages(nid);
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800271 return CONSTRAINT_MEMORY_POLICY;
David Rientjesa63d83f2010-08-09 17:19:46 -0700272 }
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800273
274 /* Check this allocation failure is caused by cpuset's wall function */
275 for_each_zone_zonelist_nodemask(zone, z, zonelist,
276 high_zoneidx, nodemask)
277 if (!cpuset_zone_allowed_softwall(zone, gfp_mask))
David Rientjesa63d83f2010-08-09 17:19:46 -0700278 cpuset_limited = true;
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800279
David Rientjesa63d83f2010-08-09 17:19:46 -0700280 if (cpuset_limited) {
281 *totalpages = total_swap_pages;
282 for_each_node_mask(nid, cpuset_current_mems_allowed)
283 *totalpages += node_spanned_pages(nid);
284 return CONSTRAINT_CPUSET;
285 }
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800286 return CONSTRAINT_NONE;
287}
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800288#else
289static enum oom_constraint constrained_alloc(struct zonelist *zonelist,
David Rientjesa63d83f2010-08-09 17:19:46 -0700290 gfp_t gfp_mask, nodemask_t *nodemask,
291 unsigned long *totalpages)
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800292{
David Rientjesa63d83f2010-08-09 17:19:46 -0700293 *totalpages = totalram_pages + total_swap_pages;
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800294 return CONSTRAINT_NONE;
295}
296#endif
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800297
298/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299 * Simple selection loop. We chose the process with the highest
300 * number of 'points'. We expect the caller will lock the tasklist.
301 *
302 * (not docbooked, we don't want this one cluttering up the manual)
303 */
David Rientjesa63d83f2010-08-09 17:19:46 -0700304static struct task_struct *select_bad_process(unsigned int *ppoints,
305 unsigned long totalpages, struct mem_cgroup *mem,
306 const nodemask_t *nodemask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700307{
David Rientjes3a5dda72011-03-22 16:30:09 -0700308 struct task_struct *g, *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309 struct task_struct *chosen = NULL;
Kurt Garloff9827b782006-02-20 18:27:51 -0800310 *ppoints = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700311
David Rientjes3a5dda72011-03-22 16:30:09 -0700312 do_each_thread(g, p) {
David Rientjesa63d83f2010-08-09 17:19:46 -0700313 unsigned int points;
Paul Jacksona49335c2005-09-06 15:18:09 -0700314
Oleg Nesterovc027a472011-07-30 16:35:02 +0200315 if (p->exit_state)
Andrey Vagin30e2b412011-03-22 16:30:11 -0700316 continue;
KOSAKI Motohiroab290ad2010-08-09 17:19:35 -0700317 if (oom_unkillable_task(p, mem, nodemask))
David Rientjes6cf86ac2010-08-09 17:18:50 -0700318 continue;
Paul Jacksonef08e3b2005-09-06 15:18:13 -0700319
Paul Jacksona49335c2005-09-06 15:18:09 -0700320 /*
Nick Pigginb78483a2006-09-29 02:01:14 -0700321 * This task already has access to memory reserves and is
322 * being killed. Don't allow any other task access to the
323 * memory reserve.
324 *
325 * Note: this may have a chance of deadlock if it gets
326 * blocked waiting for another task which itself is waiting
327 * for memory. Is there a better alternative?
328 */
David Rientjesf660daa2011-10-31 17:07:07 -0700329 if (test_tsk_thread_flag(p, TIF_MEMDIE)) {
330 if (unlikely(frozen(p)))
331 thaw_process(p);
Nick Pigginb78483a2006-09-29 02:01:14 -0700332 return ERR_PTR(-1UL);
David Rientjesf660daa2011-10-31 17:07:07 -0700333 }
Oleg Nesterovc027a472011-07-30 16:35:02 +0200334 if (!p->mm)
335 continue;
Nick Pigginb78483a2006-09-29 02:01:14 -0700336
Andrey Vagin30e2b412011-03-22 16:30:11 -0700337 if (p->flags & PF_EXITING) {
David Rientjesedd45542011-03-22 16:30:12 -0700338 /*
339 * If p is the current task and is in the process of
340 * releasing memory, we allow the "kill" to set
341 * TIF_MEMDIE, which will allow it to gain access to
342 * memory reserves. Otherwise, it may stall forever.
343 *
344 * The loop isn't broken here, however, in case other
345 * threads are found to have already been oom killed.
346 */
347 if (p == current) {
348 chosen = p;
349 *ppoints = 1000;
350 } else {
351 /*
352 * If this task is not being ptraced on exit,
353 * then wait for it to finish before killing
354 * some other task unnecessarily.
355 */
Tejun Heod21142e2011-06-17 16:50:34 +0200356 if (!(p->group_leader->ptrace & PT_TRACE_EXIT))
David Rientjesedd45542011-03-22 16:30:12 -0700357 return ERR_PTR(-1UL);
358 }
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700359 }
Oleg Nesterov972c4ea2006-09-29 02:01:12 -0700360
David Rientjesa63d83f2010-08-09 17:19:46 -0700361 points = oom_badness(p, mem, nodemask, totalpages);
362 if (points > *ppoints) {
Paul Jacksona49335c2005-09-06 15:18:09 -0700363 chosen = p;
Kurt Garloff9827b782006-02-20 18:27:51 -0800364 *ppoints = points;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365 }
David Rientjes3a5dda72011-03-22 16:30:09 -0700366 } while_each_thread(g, p);
Oleg Nesterov972c4ea2006-09-29 02:01:12 -0700367
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368 return chosen;
369}
370
371/**
Randy Dunlap1b578df2008-03-19 17:00:42 -0700372 * dump_tasks - dump current memory state of all system tasks
David Rientjes74ab7f12010-08-09 17:18:46 -0700373 * @mem: current's memory controller, if constrained
David Rientjese85bfd32010-09-22 13:05:10 -0700374 * @nodemask: nodemask passed to page allocator for mempolicy ooms
Randy Dunlap1b578df2008-03-19 17:00:42 -0700375 *
David Rientjese85bfd32010-09-22 13:05:10 -0700376 * Dumps the current memory state of all eligible tasks. Tasks not in the same
377 * memcg, not in the same cpuset, or bound to a disjoint set of mempolicy nodes
378 * are not shown.
David Rientjesfef1bdd2008-02-07 00:14:07 -0800379 * State information includes task's pid, uid, tgid, vm size, rss, cpu, oom_adj
David Rientjesa63d83f2010-08-09 17:19:46 -0700380 * value, oom_score_adj value, and name.
David Rientjesfef1bdd2008-02-07 00:14:07 -0800381 *
David Rientjesfef1bdd2008-02-07 00:14:07 -0800382 * Call with tasklist_lock read-locked.
383 */
David Rientjese85bfd32010-09-22 13:05:10 -0700384static void dump_tasks(const struct mem_cgroup *mem, const nodemask_t *nodemask)
David Rientjesfef1bdd2008-02-07 00:14:07 -0800385{
KOSAKI Motohiroc55db952010-08-09 17:18:46 -0700386 struct task_struct *p;
387 struct task_struct *task;
David Rientjesfef1bdd2008-02-07 00:14:07 -0800388
David Rientjesa63d83f2010-08-09 17:19:46 -0700389 pr_info("[ pid ] uid tgid total_vm rss cpu oom_adj oom_score_adj name\n");
KOSAKI Motohiroc55db952010-08-09 17:18:46 -0700390 for_each_process(p) {
David Rientjese85bfd32010-09-22 13:05:10 -0700391 if (oom_unkillable_task(p, mem, nodemask))
David Rientjesfef1bdd2008-02-07 00:14:07 -0800392 continue;
393
KOSAKI Motohiroc55db952010-08-09 17:18:46 -0700394 task = find_lock_task_mm(p);
395 if (!task) {
David Rientjes6d2661e2009-05-28 14:34:19 -0700396 /*
David Rientjes74ab7f12010-08-09 17:18:46 -0700397 * This is a kthread or all of p's threads have already
398 * detached their mm's. There's no need to report
KOSAKI Motohiroc55db952010-08-09 17:18:46 -0700399 * them; they can't be oom killed anyway.
David Rientjes6d2661e2009-05-28 14:34:19 -0700400 */
David Rientjes6d2661e2009-05-28 14:34:19 -0700401 continue;
402 }
KOSAKI Motohiroc55db952010-08-09 17:18:46 -0700403
David Rientjesa63d83f2010-08-09 17:19:46 -0700404 pr_info("[%5d] %5d %5d %8lu %8lu %3u %3d %5d %s\n",
KOSAKI Motohiro8d6c83f2010-08-19 14:13:39 -0700405 task->pid, task_uid(task), task->tgid,
David Rientjesa63d83f2010-08-09 17:19:46 -0700406 task->mm->total_vm, get_mm_rss(task->mm),
407 task_cpu(task), task->signal->oom_adj,
408 task->signal->oom_score_adj, task->comm);
KOSAKI Motohiroc55db952010-08-09 17:18:46 -0700409 task_unlock(task);
410 }
David Rientjesfef1bdd2008-02-07 00:14:07 -0800411}
412
Daisuke Nishimurad31f56d2009-12-15 16:47:12 -0800413static void dump_header(struct task_struct *p, gfp_t gfp_mask, int order,
David Rientjese85bfd32010-09-22 13:05:10 -0700414 struct mem_cgroup *mem, const nodemask_t *nodemask)
David Rientjes1b604d72009-12-14 17:57:47 -0800415{
David Rientjes5e9d8342010-08-09 17:18:51 -0700416 task_lock(current);
David Rientjes1b604d72009-12-14 17:57:47 -0800417 pr_warning("%s invoked oom-killer: gfp_mask=0x%x, order=%d, "
David Rientjesa63d83f2010-08-09 17:19:46 -0700418 "oom_adj=%d, oom_score_adj=%d\n",
419 current->comm, gfp_mask, order, current->signal->oom_adj,
420 current->signal->oom_score_adj);
David Rientjes1b604d72009-12-14 17:57:47 -0800421 cpuset_print_task_mems_allowed(current);
422 task_unlock(current);
423 dump_stack();
Daisuke Nishimurad31f56d2009-12-15 16:47:12 -0800424 mem_cgroup_print_oom_info(mem, p);
David Rientjesb2b755b2011-03-24 15:18:15 -0700425 show_mem(SHOW_MEM_FILTER_NODES);
David Rientjes1b604d72009-12-14 17:57:47 -0800426 if (sysctl_oom_dump_tasks)
David Rientjese85bfd32010-09-22 13:05:10 -0700427 dump_tasks(mem, nodemask);
David Rientjes1b604d72009-12-14 17:57:47 -0800428}
429
KOSAKI Motohiro3b4798c2009-12-15 16:45:32 -0800430#define K(x) ((x) << (PAGE_SHIFT-10))
Luis Claudio R. Goncalves93b43fa2010-08-09 17:19:41 -0700431static int oom_kill_task(struct task_struct *p, struct mem_cgroup *mem)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432{
David Rientjes1e99bad2010-10-26 14:21:24 -0700433 struct task_struct *q;
434 struct mm_struct *mm;
435
David Rientjesb940fd72010-08-09 17:18:57 -0700436 p = find_lock_task_mm(p);
KOSAKI Motohirobe71cf22010-08-19 14:13:38 -0700437 if (!p)
Dave Peterson01315922006-04-18 22:20:44 -0700438 return 1;
KOSAKI Motohirobe71cf22010-08-19 14:13:38 -0700439
David Rientjes1e99bad2010-10-26 14:21:24 -0700440 /* mm cannot be safely dereferenced after task_unlock(p) */
441 mm = p->mm;
442
David Rientjesb940fd72010-08-09 17:18:57 -0700443 pr_err("Killed process %d (%s) total-vm:%lukB, anon-rss:%lukB, file-rss:%lukB\n",
444 task_pid_nr(p), p->comm, K(p->mm->total_vm),
445 K(get_mm_counter(p->mm, MM_ANONPAGES)),
446 K(get_mm_counter(p->mm, MM_FILEPAGES)));
447 task_unlock(p);
KOSAKI Motohiro0753ba02009-08-18 14:11:10 -0700448
David Rientjes1e99bad2010-10-26 14:21:24 -0700449 /*
David Rientjes7b0d44f2011-10-31 17:07:11 -0700450 * Kill all user processes sharing p->mm in other thread groups, if any.
David Rientjes1e99bad2010-10-26 14:21:24 -0700451 * They don't get access to memory reserves or a higher scheduler
452 * priority, though, to avoid depletion of all memory or task
453 * starvation. This prevents mm->mmap_sem livelock when an oom killed
454 * task cannot exit because it requires the semaphore and its contended
455 * by another thread trying to allocate memory itself. That thread will
456 * now get access to memory reserves since it has a pending fatal
457 * signal.
458 */
459 for_each_process(q)
David Rientjes7b0d44f2011-10-31 17:07:11 -0700460 if (q->mm == mm && !same_thread_group(q, p) &&
461 !(q->flags & PF_KTHREAD)) {
David Rientjesc9f01242011-10-31 17:07:15 -0700462 if (q->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
463 continue;
464
David Rientjes1e99bad2010-10-26 14:21:24 -0700465 task_lock(q); /* Protect ->comm from prctl() */
466 pr_err("Kill process %d (%s) sharing same memory\n",
467 task_pid_nr(q), q->comm);
468 task_unlock(q);
469 force_sig(SIGKILL, q);
470 }
Luis Claudio R. Goncalves93b43fa2010-08-09 17:19:41 -0700471
David Rientjesb940fd72010-08-09 17:18:57 -0700472 set_tsk_thread_flag(p, TIF_MEMDIE);
473 force_sig(SIGKILL, p);
Luis Claudio R. Goncalves93b43fa2010-08-09 17:19:41 -0700474
Dave Peterson01315922006-04-18 22:20:44 -0700475 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476}
David Rientjesb940fd72010-08-09 17:18:57 -0700477#undef K
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478
David Rientjes7213f502007-10-16 23:25:57 -0700479static int oom_kill_process(struct task_struct *p, gfp_t gfp_mask, int order,
David Rientjesa63d83f2010-08-09 17:19:46 -0700480 unsigned int points, unsigned long totalpages,
481 struct mem_cgroup *mem, nodemask_t *nodemask,
482 const char *message)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483{
Linus Torvalds52d3c032011-03-14 15:17:07 -0700484 struct task_struct *victim = p;
David Rientjes5e9d8342010-08-09 17:18:51 -0700485 struct task_struct *child;
Linus Torvalds52d3c032011-03-14 15:17:07 -0700486 struct task_struct *t = p;
487 unsigned int victim_points = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488
David Rientjes1b604d72009-12-14 17:57:47 -0800489 if (printk_ratelimit())
David Rientjese85bfd32010-09-22 13:05:10 -0700490 dump_header(p, gfp_mask, order, mem, nodemask);
David Rientjes7213f502007-10-16 23:25:57 -0700491
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700492 /*
493 * If the task is already exiting, don't alarm the sysadmin or kill
494 * its children or threads, just set TIF_MEMDIE so it can die quickly
495 */
KOSAKI Motohiro0753ba02009-08-18 14:11:10 -0700496 if (p->flags & PF_EXITING) {
David Rientjes43589972010-08-09 17:18:49 -0700497 set_tsk_thread_flag(p, TIF_MEMDIE);
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700498 return 0;
499 }
500
David Rientjes5e9d8342010-08-09 17:18:51 -0700501 task_lock(p);
David Rientjesa63d83f2010-08-09 17:19:46 -0700502 pr_err("%s: Kill process %d (%s) score %d or sacrifice child\n",
David Rientjes5e9d8342010-08-09 17:18:51 -0700503 message, task_pid_nr(p), p->comm, points);
504 task_unlock(p);
Nick Pigginf3af38d2006-12-06 20:31:51 -0800505
David Rientjes5e9d8342010-08-09 17:18:51 -0700506 /*
507 * If any of p's children has a different mm and is eligible for kill,
David Rientjes11239832011-07-25 17:12:17 -0700508 * the one with the highest oom_badness() score is sacrificed for its
David Rientjes5e9d8342010-08-09 17:18:51 -0700509 * parent. This attempts to lose the minimal amount of work done while
510 * still freeing memory.
511 */
Oleg Nesterovdd8e8f42010-08-09 17:18:45 -0700512 do {
David Rientjes5e9d8342010-08-09 17:18:51 -0700513 list_for_each_entry(child, &t->children, sibling) {
David Rientjesa63d83f2010-08-09 17:19:46 -0700514 unsigned int child_points;
David Rientjes5e9d8342010-08-09 17:18:51 -0700515
David Rientjesedd45542011-03-22 16:30:12 -0700516 if (child->mm == p->mm)
517 continue;
David Rientjesa63d83f2010-08-09 17:19:46 -0700518 /*
519 * oom_badness() returns 0 if the thread is unkillable
520 */
521 child_points = oom_badness(child, mem, nodemask,
522 totalpages);
David Rientjes5e9d8342010-08-09 17:18:51 -0700523 if (child_points > victim_points) {
524 victim = child;
525 victim_points = child_points;
526 }
Oleg Nesterovdd8e8f42010-08-09 17:18:45 -0700527 }
528 } while_each_thread(p, t);
529
Luis Claudio R. Goncalves93b43fa2010-08-09 17:19:41 -0700530 return oom_kill_task(victim, mem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531}
532
David Rientjes309ed882010-08-09 17:18:54 -0700533/*
534 * Determines whether the kernel must panic because of the panic_on_oom sysctl.
535 */
536static void check_panic_on_oom(enum oom_constraint constraint, gfp_t gfp_mask,
David Rientjese85bfd32010-09-22 13:05:10 -0700537 int order, const nodemask_t *nodemask)
David Rientjes309ed882010-08-09 17:18:54 -0700538{
539 if (likely(!sysctl_panic_on_oom))
540 return;
541 if (sysctl_panic_on_oom != 2) {
542 /*
543 * panic_on_oom == 1 only affects CONSTRAINT_NONE, the kernel
544 * does not panic for cpuset, mempolicy, or memcg allocation
545 * failures.
546 */
547 if (constraint != CONSTRAINT_NONE)
548 return;
549 }
550 read_lock(&tasklist_lock);
David Rientjese85bfd32010-09-22 13:05:10 -0700551 dump_header(NULL, gfp_mask, order, NULL, nodemask);
David Rientjes309ed882010-08-09 17:18:54 -0700552 read_unlock(&tasklist_lock);
553 panic("Out of memory: %s panic_on_oom is enabled\n",
554 sysctl_panic_on_oom == 2 ? "compulsory" : "system-wide");
555}
556
Balbir Singh00f0b822008-03-04 14:28:39 -0800557#ifdef CONFIG_CGROUP_MEM_RES_CTLR
Pavel Emelianovc7ba5c92008-02-07 00:13:58 -0800558void mem_cgroup_out_of_memory(struct mem_cgroup *mem, gfp_t gfp_mask)
559{
David Rientjesa63d83f2010-08-09 17:19:46 -0700560 unsigned long limit;
561 unsigned int points = 0;
Pavel Emelianovc7ba5c92008-02-07 00:13:58 -0800562 struct task_struct *p;
563
David Rientjesf9434ad2011-03-23 16:42:44 -0700564 /*
565 * If current has a pending SIGKILL, then automatically select it. The
566 * goal is to allow it to allocate so that it may quickly exit and free
567 * its memory.
568 */
569 if (fatal_signal_pending(current)) {
570 set_thread_flag(TIF_MEMDIE);
David Rientjesf9434ad2011-03-23 16:42:44 -0700571 return;
572 }
573
David Rientjese85bfd32010-09-22 13:05:10 -0700574 check_panic_on_oom(CONSTRAINT_MEMCG, gfp_mask, 0, NULL);
David Rientjesa63d83f2010-08-09 17:19:46 -0700575 limit = mem_cgroup_get_limit(mem) >> PAGE_SHIFT;
Li Zefane115f2d2008-04-15 14:34:37 -0700576 read_lock(&tasklist_lock);
Pavel Emelianovc7ba5c92008-02-07 00:13:58 -0800577retry:
David Rientjesa63d83f2010-08-09 17:19:46 -0700578 p = select_bad_process(&points, limit, mem, NULL);
David Rientjesdf64f812010-05-26 14:42:41 -0700579 if (!p || PTR_ERR(p) == -1UL)
Pavel Emelianovc7ba5c92008-02-07 00:13:58 -0800580 goto out;
581
David Rientjesa63d83f2010-08-09 17:19:46 -0700582 if (oom_kill_process(p, gfp_mask, 0, points, limit, mem, NULL,
Pavel Emelianovc7ba5c92008-02-07 00:13:58 -0800583 "Memory cgroup out of memory"))
584 goto retry;
585out:
Li Zefane115f2d2008-04-15 14:34:37 -0700586 read_unlock(&tasklist_lock);
Pavel Emelianovc7ba5c92008-02-07 00:13:58 -0800587}
588#endif
589
Martin Schwidefsky8bc719d2006-09-25 23:31:20 -0700590static BLOCKING_NOTIFIER_HEAD(oom_notify_list);
591
592int register_oom_notifier(struct notifier_block *nb)
593{
594 return blocking_notifier_chain_register(&oom_notify_list, nb);
595}
596EXPORT_SYMBOL_GPL(register_oom_notifier);
597
598int unregister_oom_notifier(struct notifier_block *nb)
599{
600 return blocking_notifier_chain_unregister(&oom_notify_list, nb);
601}
602EXPORT_SYMBOL_GPL(unregister_oom_notifier);
603
David Rientjes098d7f12007-10-16 23:25:55 -0700604/*
605 * Try to acquire the OOM killer lock for the zones in zonelist. Returns zero
606 * if a parallel OOM killing is already taking place that includes a zone in
607 * the zonelist. Otherwise, locks all zones in the zonelist and returns 1.
608 */
Minchan Kimff321fe2010-08-09 17:18:57 -0700609int try_set_zonelist_oom(struct zonelist *zonelist, gfp_t gfp_mask)
David Rientjes098d7f12007-10-16 23:25:55 -0700610{
Mel Gormandd1a2392008-04-28 02:12:17 -0700611 struct zoneref *z;
612 struct zone *zone;
David Rientjes098d7f12007-10-16 23:25:55 -0700613 int ret = 1;
614
David Rientjesc7d4cae2009-01-06 14:39:00 -0800615 spin_lock(&zone_scan_lock);
Mel Gormandd1a2392008-04-28 02:12:17 -0700616 for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) {
617 if (zone_is_oom_locked(zone)) {
David Rientjes098d7f12007-10-16 23:25:55 -0700618 ret = 0;
619 goto out;
620 }
Mel Gormandd1a2392008-04-28 02:12:17 -0700621 }
David Rientjes098d7f12007-10-16 23:25:55 -0700622
Mel Gormandd1a2392008-04-28 02:12:17 -0700623 for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) {
624 /*
David Rientjesc7d4cae2009-01-06 14:39:00 -0800625 * Lock each zone in the zonelist under zone_scan_lock so a
Minchan Kimff321fe2010-08-09 17:18:57 -0700626 * parallel invocation of try_set_zonelist_oom() doesn't succeed
Mel Gormandd1a2392008-04-28 02:12:17 -0700627 * when it shouldn't.
628 */
629 zone_set_flag(zone, ZONE_OOM_LOCKED);
630 }
631
David Rientjes098d7f12007-10-16 23:25:55 -0700632out:
David Rientjesc7d4cae2009-01-06 14:39:00 -0800633 spin_unlock(&zone_scan_lock);
David Rientjes098d7f12007-10-16 23:25:55 -0700634 return ret;
635}
636
637/*
638 * Clears the ZONE_OOM_LOCKED flag for all zones in the zonelist so that failed
639 * allocation attempts with zonelists containing them may now recall the OOM
640 * killer, if necessary.
641 */
Mel Gormandd1a2392008-04-28 02:12:17 -0700642void clear_zonelist_oom(struct zonelist *zonelist, gfp_t gfp_mask)
David Rientjes098d7f12007-10-16 23:25:55 -0700643{
Mel Gormandd1a2392008-04-28 02:12:17 -0700644 struct zoneref *z;
645 struct zone *zone;
David Rientjes098d7f12007-10-16 23:25:55 -0700646
David Rientjesc7d4cae2009-01-06 14:39:00 -0800647 spin_lock(&zone_scan_lock);
Mel Gormandd1a2392008-04-28 02:12:17 -0700648 for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) {
649 zone_clear_flag(zone, ZONE_OOM_LOCKED);
650 }
David Rientjesc7d4cae2009-01-06 14:39:00 -0800651 spin_unlock(&zone_scan_lock);
David Rientjes098d7f12007-10-16 23:25:55 -0700652}
653
Nick Piggin1c0fe6e2009-01-06 14:38:59 -0800654/*
David Rientjese3658932010-08-09 17:18:55 -0700655 * Try to acquire the oom killer lock for all system zones. Returns zero if a
656 * parallel oom killing is taking place, otherwise locks all zones and returns
657 * non-zero.
658 */
659static int try_set_system_oom(void)
660{
661 struct zone *zone;
662 int ret = 1;
663
664 spin_lock(&zone_scan_lock);
665 for_each_populated_zone(zone)
666 if (zone_is_oom_locked(zone)) {
667 ret = 0;
668 goto out;
669 }
670 for_each_populated_zone(zone)
671 zone_set_flag(zone, ZONE_OOM_LOCKED);
672out:
673 spin_unlock(&zone_scan_lock);
674 return ret;
675}
676
677/*
678 * Clears ZONE_OOM_LOCKED for all system zones so that failed allocation
679 * attempts or page faults may now recall the oom killer, if necessary.
680 */
681static void clear_system_oom(void)
682{
683 struct zone *zone;
684
685 spin_lock(&zone_scan_lock);
686 for_each_populated_zone(zone)
687 zone_clear_flag(zone, ZONE_OOM_LOCKED);
688 spin_unlock(&zone_scan_lock);
689}
690
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691/**
Dave Peterson6937a252006-06-23 02:03:13 -0700692 * out_of_memory - kill the "best" process when we run out of memory
Randy Dunlap1b578df2008-03-19 17:00:42 -0700693 * @zonelist: zonelist pointer
694 * @gfp_mask: memory allocation flags
695 * @order: amount of memory being requested as a power of 2
David Rientjes6f48d0eb2010-08-09 17:18:52 -0700696 * @nodemask: nodemask passed to page allocator
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697 *
698 * If we run out of memory, we have the choice between either
699 * killing a random task (bad), letting the system crash (worse)
700 * OR try to be smart about which process to kill. Note that we
701 * don't have to be perfect here, we just have to be good.
702 */
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800703void out_of_memory(struct zonelist *zonelist, gfp_t gfp_mask,
704 int order, nodemask_t *nodemask)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705{
David Rientjese85bfd32010-09-22 13:05:10 -0700706 const nodemask_t *mpol_mask;
David Rientjes0aad4b32010-08-09 17:18:59 -0700707 struct task_struct *p;
David Rientjesa63d83f2010-08-09 17:19:46 -0700708 unsigned long totalpages;
Martin Schwidefsky8bc719d2006-09-25 23:31:20 -0700709 unsigned long freed = 0;
David Rientjesa63d83f2010-08-09 17:19:46 -0700710 unsigned int points;
David Rientjese3658932010-08-09 17:18:55 -0700711 enum oom_constraint constraint = CONSTRAINT_NONE;
KOSAKI Motohirob52723c2010-08-19 14:13:39 -0700712 int killed = 0;
Martin Schwidefsky8bc719d2006-09-25 23:31:20 -0700713
714 blocking_notifier_call_chain(&oom_notify_list, 0, &freed);
715 if (freed > 0)
716 /* Got some memory back in the last second. */
717 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718
David Rientjes7b98c2e2010-08-09 17:18:48 -0700719 /*
720 * If current has a pending SIGKILL, then automatically select it. The
721 * goal is to allow it to allocate so that it may quickly exit and free
722 * its memory.
723 */
724 if (fatal_signal_pending(current)) {
725 set_thread_flag(TIF_MEMDIE);
726 return;
727 }
728
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800729 /*
730 * Check if there were limitations on the allocation (only relevant for
731 * NUMA) that may require different handling.
732 */
David Rientjesa63d83f2010-08-09 17:19:46 -0700733 constraint = constrained_alloc(zonelist, gfp_mask, nodemask,
734 &totalpages);
David Rientjese85bfd32010-09-22 13:05:10 -0700735 mpol_mask = (constraint == CONSTRAINT_MEMORY_POLICY) ? nodemask : NULL;
736 check_panic_on_oom(constraint, gfp_mask, order, mpol_mask);
David Rientjes0aad4b32010-08-09 17:18:59 -0700737
David Rientjes2b45ab32007-05-06 14:50:00 -0700738 read_lock(&tasklist_lock);
KOSAKI Motohirof88ccad2010-08-09 17:19:36 -0700739 if (sysctl_oom_kill_allocating_task &&
KOSAKI Motohiroa96cfd62010-08-09 17:19:38 -0700740 !oom_unkillable_task(current, NULL, nodemask) &&
David Rientjesc9f01242011-10-31 17:07:15 -0700741 current->mm) {
David Rientjes0aad4b32010-08-09 17:18:59 -0700742 /*
743 * oom_kill_process() needs tasklist_lock held. If it returns
744 * non-zero, current could not be killed so we must fallback to
745 * the tasklist scan.
746 */
David Rientjesa63d83f2010-08-09 17:19:46 -0700747 if (!oom_kill_process(current, gfp_mask, order, 0, totalpages,
748 NULL, nodemask,
David Rientjes0aad4b32010-08-09 17:18:59 -0700749 "Out of memory (oom_kill_allocating_task)"))
KOSAKI Motohirob52723c2010-08-19 14:13:39 -0700750 goto out;
David Rientjes0aad4b32010-08-09 17:18:59 -0700751 }
752
753retry:
David Rientjese85bfd32010-09-22 13:05:10 -0700754 p = select_bad_process(&points, totalpages, NULL, mpol_mask);
David Rientjes0aad4b32010-08-09 17:18:59 -0700755 if (PTR_ERR(p) == -1UL)
KOSAKI Motohirob52723c2010-08-19 14:13:39 -0700756 goto out;
David Rientjes0aad4b32010-08-09 17:18:59 -0700757
758 /* Found nothing?!?! Either we hang forever, or we panic. */
759 if (!p) {
David Rientjese85bfd32010-09-22 13:05:10 -0700760 dump_header(NULL, gfp_mask, order, NULL, mpol_mask);
David Rientjes0aad4b32010-08-09 17:18:59 -0700761 read_unlock(&tasklist_lock);
762 panic("Out of memory and no killable processes...\n");
763 }
764
David Rientjesa63d83f2010-08-09 17:19:46 -0700765 if (oom_kill_process(p, gfp_mask, order, points, totalpages, NULL,
766 nodemask, "Out of memory"))
David Rientjes0aad4b32010-08-09 17:18:59 -0700767 goto retry;
KOSAKI Motohirob52723c2010-08-19 14:13:39 -0700768 killed = 1;
769out:
Andrew Morton140ffce2006-03-02 02:54:28 -0800770 read_unlock(&tasklist_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700771
772 /*
773 * Give "p" a good chance of killing itself before we
Kirill Korotaev2f659f42006-01-08 01:01:05 -0800774 * retry to allocate memory unless "p" is current
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775 */
KOSAKI Motohirob52723c2010-08-19 14:13:39 -0700776 if (killed && !test_thread_flag(TIF_MEMDIE))
Andrew Morton140ffce2006-03-02 02:54:28 -0800777 schedule_timeout_uninterruptible(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778}
David Rientjese3658932010-08-09 17:18:55 -0700779
780/*
781 * The pagefault handler calls here because it is out of memory, so kill a
782 * memory-hogging task. If a populated zone has ZONE_OOM_LOCKED set, a parallel
783 * oom killing is already in progress so do nothing. If a task is found with
784 * TIF_MEMDIE set, it has been killed so do nothing and allow it to exit.
785 */
786void pagefault_out_of_memory(void)
787{
788 if (try_set_system_oom()) {
789 out_of_memory(NULL, 0, 0, NULL);
790 clear_system_oom();
791 }
792 if (!test_thread_flag(TIF_MEMDIE))
793 schedule_timeout_uninterruptible(1);
794}