blob: 597ecac5731a8b5f725efcac34a247186f2877e8 [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>
David Rientjesdc3f21e2012-03-21 16:33:47 -070037#include <linux/ratelimit.h>
KAMEZAWA Hiroyuki43d2b112012-01-10 15:08:09 -080038
39#define CREATE_TRACE_POINTS
40#include <trace/events/oom.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070041
KAMEZAWA Hiroyukifadd8fb2006-06-23 02:03:13 -070042int sysctl_panic_on_oom;
David Rientjesfe071d72007-10-16 23:25:56 -070043int sysctl_oom_kill_allocating_task;
David Rientjesad915c42010-08-09 17:18:53 -070044int sysctl_oom_dump_tasks = 1;
David Rientjesc7d4cae2009-01-06 14:39:00 -080045static DEFINE_SPINLOCK(zone_scan_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -070046
David Rientjes43362a42011-10-31 17:07:18 -070047/*
48 * compare_swap_oom_score_adj() - compare and swap current's oom_score_adj
49 * @old_val: old oom_score_adj for compare
50 * @new_val: new oom_score_adj for swap
51 *
52 * Sets the oom_score_adj value for current to @new_val iff its present value is
53 * @old_val. Usually used to reinstate a previous value to prevent racing with
54 * userspacing tuning the value in the interim.
55 */
56void compare_swap_oom_score_adj(int old_val, int new_val)
57{
58 struct sighand_struct *sighand = current->sighand;
59
60 spin_lock_irq(&sighand->siglock);
61 if (current->signal->oom_score_adj == old_val)
62 current->signal->oom_score_adj = new_val;
KAMEZAWA Hiroyuki43d2b112012-01-10 15:08:09 -080063 trace_oom_score_adj_update(current);
David Rientjes43362a42011-10-31 17:07:18 -070064 spin_unlock_irq(&sighand->siglock);
65}
66
David Rientjes72788c32011-05-24 17:11:40 -070067/**
68 * test_set_oom_score_adj() - set current's oom_score_adj and return old value
69 * @new_val: new oom_score_adj value
70 *
71 * Sets the oom_score_adj value for current to @new_val with proper
72 * synchronization and returns the old value. Usually used to temporarily
73 * set a value, save the old value in the caller, and then reinstate it later.
74 */
75int test_set_oom_score_adj(int new_val)
76{
77 struct sighand_struct *sighand = current->sighand;
78 int old_val;
79
80 spin_lock_irq(&sighand->siglock);
81 old_val = current->signal->oom_score_adj;
David Rientjesc9f01242011-10-31 17:07:15 -070082 current->signal->oom_score_adj = new_val;
KAMEZAWA Hiroyuki43d2b112012-01-10 15:08:09 -080083 trace_oom_score_adj_update(current);
David Rientjes72788c32011-05-24 17:11:40 -070084 spin_unlock_irq(&sighand->siglock);
85
86 return old_val;
87}
88
David Rientjes6f48d0eb2010-08-09 17:18:52 -070089#ifdef CONFIG_NUMA
90/**
91 * has_intersects_mems_allowed() - check task eligiblity for kill
92 * @tsk: task struct of which task to consider
93 * @mask: nodemask passed to page allocator for mempolicy ooms
94 *
95 * Task eligibility is determined by whether or not a candidate task, @tsk,
96 * shares the same mempolicy nodes as current if it is bound by such a policy
97 * and whether or not it has the same set of allowed cpuset nodes.
KOSAKI Motohiro495789a2009-09-21 17:03:14 -070098 */
David Rientjes6f48d0eb2010-08-09 17:18:52 -070099static bool has_intersects_mems_allowed(struct task_struct *tsk,
100 const nodemask_t *mask)
KOSAKI Motohiro495789a2009-09-21 17:03:14 -0700101{
David Rientjes6f48d0eb2010-08-09 17:18:52 -0700102 struct task_struct *start = tsk;
KOSAKI Motohiro495789a2009-09-21 17:03:14 -0700103
KOSAKI Motohiro495789a2009-09-21 17:03:14 -0700104 do {
David Rientjes6f48d0eb2010-08-09 17:18:52 -0700105 if (mask) {
106 /*
107 * If this is a mempolicy constrained oom, tsk's
108 * cpuset is irrelevant. Only return true if its
109 * mempolicy intersects current, otherwise it may be
110 * needlessly killed.
111 */
112 if (mempolicy_nodemask_intersects(tsk, mask))
113 return true;
114 } else {
115 /*
116 * This is not a mempolicy constrained oom, so only
117 * check the mems of tsk's cpuset.
118 */
119 if (cpuset_mems_allowed_intersects(current, tsk))
120 return true;
121 }
KOSAKI Motohirodf1090a2010-08-09 17:19:39 -0700122 } while_each_thread(start, tsk);
123
David Rientjes6f48d0eb2010-08-09 17:18:52 -0700124 return false;
KOSAKI Motohiro495789a2009-09-21 17:03:14 -0700125}
David Rientjes6f48d0eb2010-08-09 17:18:52 -0700126#else
127static bool has_intersects_mems_allowed(struct task_struct *tsk,
128 const nodemask_t *mask)
129{
130 return true;
131}
132#endif /* CONFIG_NUMA */
KOSAKI Motohiro495789a2009-09-21 17:03:14 -0700133
David Rientjes6f48d0eb2010-08-09 17:18:52 -0700134/*
135 * The process p may have detached its own ->mm while exiting or through
136 * use_mm(), but one or more of its subthreads may still have a valid
137 * pointer. Return p, or any of its subthreads with a valid ->mm, with
138 * task_lock() held.
139 */
KAMEZAWA Hiroyuki158e0a22010-08-10 18:03:00 -0700140struct task_struct *find_lock_task_mm(struct task_struct *p)
Oleg Nesterovdd8e8f42010-08-09 17:18:45 -0700141{
142 struct task_struct *t = p;
143
144 do {
145 task_lock(t);
146 if (likely(t->mm))
147 return t;
148 task_unlock(t);
149 } while_each_thread(p, t);
150
151 return NULL;
152}
153
KOSAKI Motohiroab290ad2010-08-09 17:19:35 -0700154/* return true if the task is not adequate as candidate victim task. */
David Rientjese85bfd32010-09-22 13:05:10 -0700155static bool oom_unkillable_task(struct task_struct *p,
Johannes Weiner72835c82012-01-12 17:18:32 -0800156 const struct mem_cgroup *memcg, const nodemask_t *nodemask)
KOSAKI Motohiroab290ad2010-08-09 17:19:35 -0700157{
158 if (is_global_init(p))
159 return true;
160 if (p->flags & PF_KTHREAD)
161 return true;
162
163 /* When mem_cgroup_out_of_memory() and p is not member of the group */
Johannes Weiner72835c82012-01-12 17:18:32 -0800164 if (memcg && !task_in_mem_cgroup(p, memcg))
KOSAKI Motohiroab290ad2010-08-09 17:19:35 -0700165 return true;
166
167 /* p may not have freeable memory in nodemask */
168 if (!has_intersects_mems_allowed(p, nodemask))
169 return true;
170
171 return false;
172}
173
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174/**
David Rientjesa63d83f2010-08-09 17:19:46 -0700175 * oom_badness - heuristic function to determine which candidate task to kill
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176 * @p: task struct of which task we should calculate
David Rientjesa63d83f2010-08-09 17:19:46 -0700177 * @totalpages: total present RAM allowed for page allocation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178 *
David Rientjesa63d83f2010-08-09 17:19:46 -0700179 * The heuristic for determining which task to kill is made to be as simple and
180 * predictable as possible. The goal is to return the highest value for the
181 * task consuming the most memory to avoid subsequent oom failures.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182 */
Johannes Weiner72835c82012-01-12 17:18:32 -0800183unsigned int oom_badness(struct task_struct *p, struct mem_cgroup *memcg,
David Rientjesa63d83f2010-08-09 17:19:46 -0700184 const nodemask_t *nodemask, unsigned long totalpages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185{
Frantisek Hrbataff05b6f2011-12-19 17:11:59 -0800186 long points;
KOSAKI Motohiro28b83c52009-09-21 17:03:13 -0700187
Johannes Weiner72835c82012-01-12 17:18:32 -0800188 if (oom_unkillable_task(p, memcg, nodemask))
KOSAKI Motohiro26ebc982010-08-09 17:19:37 -0700189 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190
Oleg Nesterovdd8e8f42010-08-09 17:18:45 -0700191 p = find_lock_task_mm(p);
192 if (!p)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193 return 0;
194
Michal Hocko5aecc852011-11-15 14:36:07 -0800195 if (p->signal->oom_score_adj == OOM_SCORE_ADJ_MIN) {
196 task_unlock(p);
197 return 0;
198 }
199
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200 /*
David Rientjesa63d83f2010-08-09 17:19:46 -0700201 * The memory controller may have a limit of 0 bytes, so avoid a divide
202 * by zero, if necessary.
203 */
204 if (!totalpages)
205 totalpages = 1;
206
207 /*
208 * The baseline for the badness score is the proportion of RAM that each
KOSAKI Motohirof755a042011-04-27 15:26:50 -0700209 * task's rss, pagetable and swap space use.
David Rientjesa63d83f2010-08-09 17:19:46 -0700210 */
KOSAKI Motohirof755a042011-04-27 15:26:50 -0700211 points = get_mm_rss(p->mm) + p->mm->nr_ptes;
212 points += get_mm_counter(p->mm, MM_SWAPENTS);
213
214 points *= 1000;
215 points /= totalpages;
Andrew Morton97c2c9b2006-04-18 22:20:38 -0700216 task_unlock(p);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217
218 /*
David Rientjesa63d83f2010-08-09 17:19:46 -0700219 * Root processes get 3% bonus, just like the __vm_enough_memory()
220 * implementation used by LSMs.
Hugh Dickins7ba34852007-01-05 16:37:03 -0800221 */
David Rientjesa63d83f2010-08-09 17:19:46 -0700222 if (has_capability_noaudit(p, CAP_SYS_ADMIN))
223 points -= 30;
Hugh Dickins7ba34852007-01-05 16:37:03 -0800224
225 /*
David Rientjesa63d83f2010-08-09 17:19:46 -0700226 * /proc/pid/oom_score_adj ranges from -1000 to +1000 such that it may
227 * either completely disable oom killing or always prefer a certain
228 * task.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229 */
David Rientjesa63d83f2010-08-09 17:19:46 -0700230 points += p->signal->oom_score_adj;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231
David Rientjesf19e8aa2010-09-22 13:04:52 -0700232 /*
233 * Never return 0 for an eligible task that may be killed since it's
234 * possible that no single user task uses more than 0.1% of memory and
235 * no single admin tasks uses more than 3.0%.
236 */
237 if (points <= 0)
238 return 1;
David Rientjesa63d83f2010-08-09 17:19:46 -0700239 return (points < 1000) ? points : 1000;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240}
241
242/*
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800243 * Determine the type of allocation constraint.
244 */
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800245#ifdef CONFIG_NUMA
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800246static enum oom_constraint constrained_alloc(struct zonelist *zonelist,
David Rientjesa63d83f2010-08-09 17:19:46 -0700247 gfp_t gfp_mask, nodemask_t *nodemask,
248 unsigned long *totalpages)
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800249{
Mel Gorman54a6eb52008-04-28 02:12:16 -0700250 struct zone *zone;
Mel Gormandd1a2392008-04-28 02:12:17 -0700251 struct zoneref *z;
Mel Gorman54a6eb52008-04-28 02:12:16 -0700252 enum zone_type high_zoneidx = gfp_zone(gfp_mask);
David Rientjesa63d83f2010-08-09 17:19:46 -0700253 bool cpuset_limited = false;
254 int nid;
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800255
David Rientjesa63d83f2010-08-09 17:19:46 -0700256 /* Default to all available memory */
257 *totalpages = totalram_pages + total_swap_pages;
258
259 if (!zonelist)
260 return CONSTRAINT_NONE;
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800261 /*
262 * Reach here only when __GFP_NOFAIL is used. So, we should avoid
263 * to kill current.We have to random task kill in this case.
264 * Hopefully, CONSTRAINT_THISNODE...but no way to handle it, now.
265 */
266 if (gfp_mask & __GFP_THISNODE)
267 return CONSTRAINT_NONE;
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800268
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800269 /*
David Rientjesa63d83f2010-08-09 17:19:46 -0700270 * This is not a __GFP_THISNODE allocation, so a truncated nodemask in
271 * the page allocator means a mempolicy is in effect. Cpuset policy
272 * is enforced in get_page_from_freelist().
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800273 */
David Rientjesa63d83f2010-08-09 17:19:46 -0700274 if (nodemask && !nodes_subset(node_states[N_HIGH_MEMORY], *nodemask)) {
275 *totalpages = total_swap_pages;
276 for_each_node_mask(nid, *nodemask)
277 *totalpages += node_spanned_pages(nid);
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800278 return CONSTRAINT_MEMORY_POLICY;
David Rientjesa63d83f2010-08-09 17:19:46 -0700279 }
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800280
281 /* Check this allocation failure is caused by cpuset's wall function */
282 for_each_zone_zonelist_nodemask(zone, z, zonelist,
283 high_zoneidx, nodemask)
284 if (!cpuset_zone_allowed_softwall(zone, gfp_mask))
David Rientjesa63d83f2010-08-09 17:19:46 -0700285 cpuset_limited = true;
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800286
David Rientjesa63d83f2010-08-09 17:19:46 -0700287 if (cpuset_limited) {
288 *totalpages = total_swap_pages;
289 for_each_node_mask(nid, cpuset_current_mems_allowed)
290 *totalpages += node_spanned_pages(nid);
291 return CONSTRAINT_CPUSET;
292 }
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800293 return CONSTRAINT_NONE;
294}
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800295#else
296static enum oom_constraint constrained_alloc(struct zonelist *zonelist,
David Rientjesa63d83f2010-08-09 17:19:46 -0700297 gfp_t gfp_mask, nodemask_t *nodemask,
298 unsigned long *totalpages)
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800299{
David Rientjesa63d83f2010-08-09 17:19:46 -0700300 *totalpages = totalram_pages + total_swap_pages;
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800301 return CONSTRAINT_NONE;
302}
303#endif
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800304
305/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306 * Simple selection loop. We chose the process with the highest
307 * number of 'points'. We expect the caller will lock the tasklist.
308 *
309 * (not docbooked, we don't want this one cluttering up the manual)
310 */
David Rientjesa63d83f2010-08-09 17:19:46 -0700311static struct task_struct *select_bad_process(unsigned int *ppoints,
Johannes Weiner72835c82012-01-12 17:18:32 -0800312 unsigned long totalpages, struct mem_cgroup *memcg,
David Rientjes08ab9b12012-03-21 16:34:04 -0700313 const nodemask_t *nodemask, bool force_kill)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314{
David Rientjes3a5dda72011-03-22 16:30:09 -0700315 struct task_struct *g, *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316 struct task_struct *chosen = NULL;
Kurt Garloff9827b782006-02-20 18:27:51 -0800317 *ppoints = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700318
David Rientjes3a5dda72011-03-22 16:30:09 -0700319 do_each_thread(g, p) {
David Rientjesa63d83f2010-08-09 17:19:46 -0700320 unsigned int points;
Paul Jacksona49335c2005-09-06 15:18:09 -0700321
Oleg Nesterovc027a472011-07-30 16:35:02 +0200322 if (p->exit_state)
Andrey Vagin30e2b412011-03-22 16:30:11 -0700323 continue;
Johannes Weiner72835c82012-01-12 17:18:32 -0800324 if (oom_unkillable_task(p, memcg, nodemask))
David Rientjes6cf86ac2010-08-09 17:18:50 -0700325 continue;
Paul Jacksonef08e3b2005-09-06 15:18:13 -0700326
Paul Jacksona49335c2005-09-06 15:18:09 -0700327 /*
Nick Pigginb78483a2006-09-29 02:01:14 -0700328 * This task already has access to memory reserves and is
329 * being killed. Don't allow any other task access to the
330 * memory reserve.
331 *
332 * Note: this may have a chance of deadlock if it gets
333 * blocked waiting for another task which itself is waiting
334 * for memory. Is there a better alternative?
335 */
David Rientjesf660daa2011-10-31 17:07:07 -0700336 if (test_tsk_thread_flag(p, TIF_MEMDIE)) {
337 if (unlikely(frozen(p)))
Tejun Heoa5be2d02011-11-21 12:32:23 -0800338 __thaw_task(p);
David Rientjes08ab9b12012-03-21 16:34:04 -0700339 if (!force_kill)
340 return ERR_PTR(-1UL);
David Rientjesf660daa2011-10-31 17:07:07 -0700341 }
Oleg Nesterovc027a472011-07-30 16:35:02 +0200342 if (!p->mm)
343 continue;
Nick Pigginb78483a2006-09-29 02:01:14 -0700344
Andrey Vagin30e2b412011-03-22 16:30:11 -0700345 if (p->flags & PF_EXITING) {
David Rientjesedd45542011-03-22 16:30:12 -0700346 /*
347 * If p is the current task and is in the process of
348 * releasing memory, we allow the "kill" to set
349 * TIF_MEMDIE, which will allow it to gain access to
350 * memory reserves. Otherwise, it may stall forever.
351 *
352 * The loop isn't broken here, however, in case other
353 * threads are found to have already been oom killed.
354 */
355 if (p == current) {
356 chosen = p;
357 *ppoints = 1000;
David Rientjes08ab9b12012-03-21 16:34:04 -0700358 } else if (!force_kill) {
David Rientjesedd45542011-03-22 16:30:12 -0700359 /*
360 * If this task is not being ptraced on exit,
361 * then wait for it to finish before killing
362 * some other task unnecessarily.
363 */
Tejun Heod21142e2011-06-17 16:50:34 +0200364 if (!(p->group_leader->ptrace & PT_TRACE_EXIT))
David Rientjesedd45542011-03-22 16:30:12 -0700365 return ERR_PTR(-1UL);
366 }
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700367 }
Oleg Nesterov972c4ea2006-09-29 02:01:12 -0700368
Johannes Weiner72835c82012-01-12 17:18:32 -0800369 points = oom_badness(p, memcg, nodemask, totalpages);
David Rientjesa63d83f2010-08-09 17:19:46 -0700370 if (points > *ppoints) {
Paul Jacksona49335c2005-09-06 15:18:09 -0700371 chosen = p;
Kurt Garloff9827b782006-02-20 18:27:51 -0800372 *ppoints = points;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373 }
David Rientjes3a5dda72011-03-22 16:30:09 -0700374 } while_each_thread(g, p);
Oleg Nesterov972c4ea2006-09-29 02:01:12 -0700375
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376 return chosen;
377}
378
379/**
Randy Dunlap1b578df2008-03-19 17:00:42 -0700380 * dump_tasks - dump current memory state of all system tasks
David Rientjes74ab7f12010-08-09 17:18:46 -0700381 * @mem: current's memory controller, if constrained
David Rientjese85bfd32010-09-22 13:05:10 -0700382 * @nodemask: nodemask passed to page allocator for mempolicy ooms
Randy Dunlap1b578df2008-03-19 17:00:42 -0700383 *
David Rientjese85bfd32010-09-22 13:05:10 -0700384 * Dumps the current memory state of all eligible tasks. Tasks not in the same
385 * memcg, not in the same cpuset, or bound to a disjoint set of mempolicy nodes
386 * are not shown.
David Rientjesfef1bdd2008-02-07 00:14:07 -0800387 * State information includes task's pid, uid, tgid, vm size, rss, cpu, oom_adj
David Rientjesa63d83f2010-08-09 17:19:46 -0700388 * value, oom_score_adj value, and name.
David Rientjesfef1bdd2008-02-07 00:14:07 -0800389 *
David Rientjesfef1bdd2008-02-07 00:14:07 -0800390 * Call with tasklist_lock read-locked.
391 */
Johannes Weiner72835c82012-01-12 17:18:32 -0800392static void dump_tasks(const struct mem_cgroup *memcg, const nodemask_t *nodemask)
David Rientjesfef1bdd2008-02-07 00:14:07 -0800393{
KOSAKI Motohiroc55db952010-08-09 17:18:46 -0700394 struct task_struct *p;
395 struct task_struct *task;
David Rientjesfef1bdd2008-02-07 00:14:07 -0800396
David Rientjesa63d83f2010-08-09 17:19:46 -0700397 pr_info("[ pid ] uid tgid total_vm rss cpu oom_adj oom_score_adj name\n");
KOSAKI Motohiroc55db952010-08-09 17:18:46 -0700398 for_each_process(p) {
Johannes Weiner72835c82012-01-12 17:18:32 -0800399 if (oom_unkillable_task(p, memcg, nodemask))
David Rientjesfef1bdd2008-02-07 00:14:07 -0800400 continue;
401
KOSAKI Motohiroc55db952010-08-09 17:18:46 -0700402 task = find_lock_task_mm(p);
403 if (!task) {
David Rientjes6d2661e2009-05-28 14:34:19 -0700404 /*
David Rientjes74ab7f12010-08-09 17:18:46 -0700405 * This is a kthread or all of p's threads have already
406 * detached their mm's. There's no need to report
KOSAKI Motohiroc55db952010-08-09 17:18:46 -0700407 * them; they can't be oom killed anyway.
David Rientjes6d2661e2009-05-28 14:34:19 -0700408 */
David Rientjes6d2661e2009-05-28 14:34:19 -0700409 continue;
410 }
KOSAKI Motohiroc55db952010-08-09 17:18:46 -0700411
David Rientjesa63d83f2010-08-09 17:19:46 -0700412 pr_info("[%5d] %5d %5d %8lu %8lu %3u %3d %5d %s\n",
KOSAKI Motohiro8d6c83f2010-08-19 14:13:39 -0700413 task->pid, task_uid(task), task->tgid,
David Rientjesa63d83f2010-08-09 17:19:46 -0700414 task->mm->total_vm, get_mm_rss(task->mm),
415 task_cpu(task), task->signal->oom_adj,
416 task->signal->oom_score_adj, task->comm);
KOSAKI Motohiroc55db952010-08-09 17:18:46 -0700417 task_unlock(task);
418 }
David Rientjesfef1bdd2008-02-07 00:14:07 -0800419}
420
Daisuke Nishimurad31f56d2009-12-15 16:47:12 -0800421static void dump_header(struct task_struct *p, gfp_t gfp_mask, int order,
Johannes Weiner72835c82012-01-12 17:18:32 -0800422 struct mem_cgroup *memcg, const nodemask_t *nodemask)
David Rientjes1b604d72009-12-14 17:57:47 -0800423{
David Rientjes5e9d8342010-08-09 17:18:51 -0700424 task_lock(current);
David Rientjes1b604d72009-12-14 17:57:47 -0800425 pr_warning("%s invoked oom-killer: gfp_mask=0x%x, order=%d, "
David Rientjesa63d83f2010-08-09 17:19:46 -0700426 "oom_adj=%d, oom_score_adj=%d\n",
427 current->comm, gfp_mask, order, current->signal->oom_adj,
428 current->signal->oom_score_adj);
David Rientjes1b604d72009-12-14 17:57:47 -0800429 cpuset_print_task_mems_allowed(current);
430 task_unlock(current);
431 dump_stack();
Johannes Weiner72835c82012-01-12 17:18:32 -0800432 mem_cgroup_print_oom_info(memcg, p);
David Rientjesb2b755b2011-03-24 15:18:15 -0700433 show_mem(SHOW_MEM_FILTER_NODES);
David Rientjes1b604d72009-12-14 17:57:47 -0800434 if (sysctl_oom_dump_tasks)
Johannes Weiner72835c82012-01-12 17:18:32 -0800435 dump_tasks(memcg, nodemask);
David Rientjes1b604d72009-12-14 17:57:47 -0800436}
437
KOSAKI Motohiro3b4798c2009-12-15 16:45:32 -0800438#define K(x) ((x) << (PAGE_SHIFT-10))
David Rientjes2a1c9b12012-03-21 16:33:46 -0700439static void oom_kill_process(struct task_struct *p, gfp_t gfp_mask, int order,
440 unsigned int points, unsigned long totalpages,
441 struct mem_cgroup *memcg, nodemask_t *nodemask,
442 const char *message)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700443{
Linus Torvalds52d3c032011-03-14 15:17:07 -0700444 struct task_struct *victim = p;
David Rientjes5e9d8342010-08-09 17:18:51 -0700445 struct task_struct *child;
Linus Torvalds52d3c032011-03-14 15:17:07 -0700446 struct task_struct *t = p;
David Rientjes647f2bd2012-03-21 16:33:46 -0700447 struct mm_struct *mm;
Linus Torvalds52d3c032011-03-14 15:17:07 -0700448 unsigned int victim_points = 0;
David Rientjesdc3f21e2012-03-21 16:33:47 -0700449 static DEFINE_RATELIMIT_STATE(oom_rs, DEFAULT_RATELIMIT_INTERVAL,
450 DEFAULT_RATELIMIT_BURST);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700452 /*
453 * If the task is already exiting, don't alarm the sysadmin or kill
454 * its children or threads, just set TIF_MEMDIE so it can die quickly
455 */
KOSAKI Motohiro0753ba02009-08-18 14:11:10 -0700456 if (p->flags & PF_EXITING) {
David Rientjes43589972010-08-09 17:18:49 -0700457 set_tsk_thread_flag(p, TIF_MEMDIE);
David Rientjes2a1c9b12012-03-21 16:33:46 -0700458 return;
Nick Piggin50ec3bb2006-09-25 23:31:29 -0700459 }
460
David Rientjesdc3f21e2012-03-21 16:33:47 -0700461 if (__ratelimit(&oom_rs))
David Rientjes8447d952012-03-21 16:33:47 -0700462 dump_header(p, gfp_mask, order, memcg, nodemask);
463
David Rientjes5e9d8342010-08-09 17:18:51 -0700464 task_lock(p);
David Rientjesa63d83f2010-08-09 17:19:46 -0700465 pr_err("%s: Kill process %d (%s) score %d or sacrifice child\n",
David Rientjes5e9d8342010-08-09 17:18:51 -0700466 message, task_pid_nr(p), p->comm, points);
467 task_unlock(p);
Nick Pigginf3af38d2006-12-06 20:31:51 -0800468
David Rientjes5e9d8342010-08-09 17:18:51 -0700469 /*
470 * If any of p's children has a different mm and is eligible for kill,
David Rientjes11239832011-07-25 17:12:17 -0700471 * the one with the highest oom_badness() score is sacrificed for its
David Rientjes5e9d8342010-08-09 17:18:51 -0700472 * parent. This attempts to lose the minimal amount of work done while
473 * still freeing memory.
474 */
Oleg Nesterovdd8e8f42010-08-09 17:18:45 -0700475 do {
David Rientjes5e9d8342010-08-09 17:18:51 -0700476 list_for_each_entry(child, &t->children, sibling) {
David Rientjesa63d83f2010-08-09 17:19:46 -0700477 unsigned int child_points;
David Rientjes5e9d8342010-08-09 17:18:51 -0700478
David Rientjesedd45542011-03-22 16:30:12 -0700479 if (child->mm == p->mm)
480 continue;
David Rientjesa63d83f2010-08-09 17:19:46 -0700481 /*
482 * oom_badness() returns 0 if the thread is unkillable
483 */
Johannes Weiner72835c82012-01-12 17:18:32 -0800484 child_points = oom_badness(child, memcg, nodemask,
David Rientjesa63d83f2010-08-09 17:19:46 -0700485 totalpages);
David Rientjes5e9d8342010-08-09 17:18:51 -0700486 if (child_points > victim_points) {
487 victim = child;
488 victim_points = child_points;
489 }
Oleg Nesterovdd8e8f42010-08-09 17:18:45 -0700490 }
491 } while_each_thread(p, t);
492
David Rientjes647f2bd2012-03-21 16:33:46 -0700493 victim = find_lock_task_mm(victim);
494 if (!victim)
495 return;
496
497 /* mm cannot safely be dereferenced after task_unlock(victim) */
498 mm = victim->mm;
499 pr_err("Killed process %d (%s) total-vm:%lukB, anon-rss:%lukB, file-rss:%lukB\n",
500 task_pid_nr(victim), victim->comm, K(victim->mm->total_vm),
501 K(get_mm_counter(victim->mm, MM_ANONPAGES)),
502 K(get_mm_counter(victim->mm, MM_FILEPAGES)));
503 task_unlock(victim);
504
505 /*
506 * Kill all user processes sharing victim->mm in other thread groups, if
507 * any. They don't get access to memory reserves, though, to avoid
508 * depletion of all memory. This prevents mm->mmap_sem livelock when an
509 * oom killed thread cannot exit because it requires the semaphore and
510 * its contended by another thread trying to allocate memory itself.
511 * That thread will now get access to memory reserves since it has a
512 * pending fatal signal.
513 */
514 for_each_process(p)
515 if (p->mm == mm && !same_thread_group(p, victim) &&
516 !(p->flags & PF_KTHREAD)) {
517 if (p->signal->oom_score_adj == OOM_SCORE_ADJ_MIN)
518 continue;
519
520 task_lock(p); /* Protect ->comm from prctl() */
521 pr_err("Kill process %d (%s) sharing same memory\n",
522 task_pid_nr(p), p->comm);
523 task_unlock(p);
Oleg Nesterovd2d39302012-03-23 15:02:45 -0700524 do_send_sig_info(SIGKILL, SEND_SIG_FORCED, p, true);
David Rientjes647f2bd2012-03-21 16:33:46 -0700525 }
526
527 set_tsk_thread_flag(victim, TIF_MEMDIE);
Oleg Nesterovd2d39302012-03-23 15:02:45 -0700528 do_send_sig_info(SIGKILL, SEND_SIG_FORCED, victim, true);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529}
David Rientjes647f2bd2012-03-21 16:33:46 -0700530#undef K
Linus Torvalds1da177e2005-04-16 15:20:36 -0700531
David Rientjes309ed882010-08-09 17:18:54 -0700532/*
533 * Determines whether the kernel must panic because of the panic_on_oom sysctl.
534 */
535static void check_panic_on_oom(enum oom_constraint constraint, gfp_t gfp_mask,
David Rientjese85bfd32010-09-22 13:05:10 -0700536 int order, const nodemask_t *nodemask)
David Rientjes309ed882010-08-09 17:18:54 -0700537{
538 if (likely(!sysctl_panic_on_oom))
539 return;
540 if (sysctl_panic_on_oom != 2) {
541 /*
542 * panic_on_oom == 1 only affects CONSTRAINT_NONE, the kernel
543 * does not panic for cpuset, mempolicy, or memcg allocation
544 * failures.
545 */
546 if (constraint != CONSTRAINT_NONE)
547 return;
548 }
549 read_lock(&tasklist_lock);
David Rientjese85bfd32010-09-22 13:05:10 -0700550 dump_header(NULL, gfp_mask, order, NULL, nodemask);
David Rientjes309ed882010-08-09 17:18:54 -0700551 read_unlock(&tasklist_lock);
552 panic("Out of memory: %s panic_on_oom is enabled\n",
553 sysctl_panic_on_oom == 2 ? "compulsory" : "system-wide");
554}
555
Balbir Singh00f0b822008-03-04 14:28:39 -0800556#ifdef CONFIG_CGROUP_MEM_RES_CTLR
David Rientjese845e192012-03-21 16:34:10 -0700557void mem_cgroup_out_of_memory(struct mem_cgroup *memcg, gfp_t gfp_mask,
558 int order)
Pavel Emelianovc7ba5c92008-02-07 00:13:58 -0800559{
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 /*
David Rientjes83040f32013-09-27 17:08:49 +0800565 * If current has a pending SIGKILL or is exiting, then automatically
566 * select it. The goal is to allow it to allocate so that it may
567 * quickly exit and free its memory.
David Rientjesf9434ad2011-03-23 16:42:44 -0700568 */
David Rientjes83040f32013-09-27 17:08:49 +0800569 if (fatal_signal_pending(current) || current->flags & PF_EXITING) {
David Rientjesf9434ad2011-03-23 16:42:44 -0700570 set_thread_flag(TIF_MEMDIE);
David Rientjesf9434ad2011-03-23 16:42:44 -0700571 return;
572 }
573
David Rientjese845e192012-03-21 16:34:10 -0700574 check_panic_on_oom(CONSTRAINT_MEMCG, gfp_mask, order, NULL);
Johannes Weiner72835c82012-01-12 17:18:32 -0800575 limit = mem_cgroup_get_limit(memcg) >> PAGE_SHIFT;
Li Zefane115f2d2008-04-15 14:34:37 -0700576 read_lock(&tasklist_lock);
David Rientjes08ab9b12012-03-21 16:34:04 -0700577 p = select_bad_process(&points, limit, memcg, NULL, false);
David Rientjes2a1c9b12012-03-21 16:33:46 -0700578 if (p && PTR_ERR(p) != -1UL)
David Rientjese845e192012-03-21 16:34:10 -0700579 oom_kill_process(p, gfp_mask, order, points, limit, memcg, NULL,
David Rientjes2a1c9b12012-03-21 16:33:46 -0700580 "Memory cgroup out of memory");
Li Zefane115f2d2008-04-15 14:34:37 -0700581 read_unlock(&tasklist_lock);
Pavel Emelianovc7ba5c92008-02-07 00:13:58 -0800582}
583#endif
584
Martin Schwidefsky8bc719d2006-09-25 23:31:20 -0700585static BLOCKING_NOTIFIER_HEAD(oom_notify_list);
586
587int register_oom_notifier(struct notifier_block *nb)
588{
589 return blocking_notifier_chain_register(&oom_notify_list, nb);
590}
591EXPORT_SYMBOL_GPL(register_oom_notifier);
592
593int unregister_oom_notifier(struct notifier_block *nb)
594{
595 return blocking_notifier_chain_unregister(&oom_notify_list, nb);
596}
597EXPORT_SYMBOL_GPL(unregister_oom_notifier);
598
David Rientjes098d7f12007-10-16 23:25:55 -0700599/*
600 * Try to acquire the OOM killer lock for the zones in zonelist. Returns zero
601 * if a parallel OOM killing is already taking place that includes a zone in
602 * the zonelist. Otherwise, locks all zones in the zonelist and returns 1.
603 */
Minchan Kimff321fe2010-08-09 17:18:57 -0700604int try_set_zonelist_oom(struct zonelist *zonelist, gfp_t gfp_mask)
David Rientjes098d7f12007-10-16 23:25:55 -0700605{
Mel Gormandd1a2392008-04-28 02:12:17 -0700606 struct zoneref *z;
607 struct zone *zone;
David Rientjes098d7f12007-10-16 23:25:55 -0700608 int ret = 1;
609
David Rientjesc7d4cae2009-01-06 14:39:00 -0800610 spin_lock(&zone_scan_lock);
Mel Gormandd1a2392008-04-28 02:12:17 -0700611 for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) {
612 if (zone_is_oom_locked(zone)) {
David Rientjes098d7f12007-10-16 23:25:55 -0700613 ret = 0;
614 goto out;
615 }
Mel Gormandd1a2392008-04-28 02:12:17 -0700616 }
David Rientjes098d7f12007-10-16 23:25:55 -0700617
Mel Gormandd1a2392008-04-28 02:12:17 -0700618 for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) {
619 /*
David Rientjesc7d4cae2009-01-06 14:39:00 -0800620 * Lock each zone in the zonelist under zone_scan_lock so a
Minchan Kimff321fe2010-08-09 17:18:57 -0700621 * parallel invocation of try_set_zonelist_oom() doesn't succeed
Mel Gormandd1a2392008-04-28 02:12:17 -0700622 * when it shouldn't.
623 */
624 zone_set_flag(zone, ZONE_OOM_LOCKED);
625 }
626
David Rientjes098d7f12007-10-16 23:25:55 -0700627out:
David Rientjesc7d4cae2009-01-06 14:39:00 -0800628 spin_unlock(&zone_scan_lock);
David Rientjes098d7f12007-10-16 23:25:55 -0700629 return ret;
630}
631
632/*
633 * Clears the ZONE_OOM_LOCKED flag for all zones in the zonelist so that failed
634 * allocation attempts with zonelists containing them may now recall the OOM
635 * killer, if necessary.
636 */
Mel Gormandd1a2392008-04-28 02:12:17 -0700637void clear_zonelist_oom(struct zonelist *zonelist, gfp_t gfp_mask)
David Rientjes098d7f12007-10-16 23:25:55 -0700638{
Mel Gormandd1a2392008-04-28 02:12:17 -0700639 struct zoneref *z;
640 struct zone *zone;
David Rientjes098d7f12007-10-16 23:25:55 -0700641
David Rientjesc7d4cae2009-01-06 14:39:00 -0800642 spin_lock(&zone_scan_lock);
Mel Gormandd1a2392008-04-28 02:12:17 -0700643 for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) {
644 zone_clear_flag(zone, ZONE_OOM_LOCKED);
645 }
David Rientjesc7d4cae2009-01-06 14:39:00 -0800646 spin_unlock(&zone_scan_lock);
David Rientjes098d7f12007-10-16 23:25:55 -0700647}
648
Nick Piggin1c0fe6e2009-01-06 14:38:59 -0800649/*
David Rientjese3658932010-08-09 17:18:55 -0700650 * Try to acquire the oom killer lock for all system zones. Returns zero if a
651 * parallel oom killing is taking place, otherwise locks all zones and returns
652 * non-zero.
653 */
654static int try_set_system_oom(void)
655{
656 struct zone *zone;
657 int ret = 1;
658
659 spin_lock(&zone_scan_lock);
660 for_each_populated_zone(zone)
661 if (zone_is_oom_locked(zone)) {
662 ret = 0;
663 goto out;
664 }
665 for_each_populated_zone(zone)
666 zone_set_flag(zone, ZONE_OOM_LOCKED);
667out:
668 spin_unlock(&zone_scan_lock);
669 return ret;
670}
671
672/*
673 * Clears ZONE_OOM_LOCKED for all system zones so that failed allocation
674 * attempts or page faults may now recall the oom killer, if necessary.
675 */
676static void clear_system_oom(void)
677{
678 struct zone *zone;
679
680 spin_lock(&zone_scan_lock);
681 for_each_populated_zone(zone)
682 zone_clear_flag(zone, ZONE_OOM_LOCKED);
683 spin_unlock(&zone_scan_lock);
684}
685
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686/**
Dave Peterson6937a252006-06-23 02:03:13 -0700687 * out_of_memory - kill the "best" process when we run out of memory
Randy Dunlap1b578df2008-03-19 17:00:42 -0700688 * @zonelist: zonelist pointer
689 * @gfp_mask: memory allocation flags
690 * @order: amount of memory being requested as a power of 2
David Rientjes6f48d0eb2010-08-09 17:18:52 -0700691 * @nodemask: nodemask passed to page allocator
David Rientjes08ab9b12012-03-21 16:34:04 -0700692 * @force_kill: true if a task must be killed, even if others are exiting
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693 *
694 * If we run out of memory, we have the choice between either
695 * killing a random task (bad), letting the system crash (worse)
696 * OR try to be smart about which process to kill. Note that we
697 * don't have to be perfect here, we just have to be good.
698 */
KAMEZAWA Hiroyuki4365a562009-12-15 16:45:33 -0800699void out_of_memory(struct zonelist *zonelist, gfp_t gfp_mask,
David Rientjes08ab9b12012-03-21 16:34:04 -0700700 int order, nodemask_t *nodemask, bool force_kill)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701{
David Rientjese85bfd32010-09-22 13:05:10 -0700702 const nodemask_t *mpol_mask;
David Rientjes0aad4b32010-08-09 17:18:59 -0700703 struct task_struct *p;
David Rientjesa63d83f2010-08-09 17:19:46 -0700704 unsigned long totalpages;
Martin Schwidefsky8bc719d2006-09-25 23:31:20 -0700705 unsigned long freed = 0;
David Rientjesa63d83f2010-08-09 17:19:46 -0700706 unsigned int points;
David Rientjese3658932010-08-09 17:18:55 -0700707 enum oom_constraint constraint = CONSTRAINT_NONE;
KOSAKI Motohirob52723c2010-08-19 14:13:39 -0700708 int killed = 0;
Martin Schwidefsky8bc719d2006-09-25 23:31:20 -0700709
710 blocking_notifier_call_chain(&oom_notify_list, 0, &freed);
711 if (freed > 0)
712 /* Got some memory back in the last second. */
713 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700714
David Rientjes7b98c2e2010-08-09 17:18:48 -0700715 /*
716 * If current has a pending SIGKILL, then automatically select it. The
717 * goal is to allow it to allocate so that it may quickly exit and free
718 * its memory.
719 */
720 if (fatal_signal_pending(current)) {
721 set_thread_flag(TIF_MEMDIE);
722 return;
723 }
724
Christoph Lameter9b0f8b02006-02-20 18:27:52 -0800725 /*
726 * Check if there were limitations on the allocation (only relevant for
727 * NUMA) that may require different handling.
728 */
David Rientjesa63d83f2010-08-09 17:19:46 -0700729 constraint = constrained_alloc(zonelist, gfp_mask, nodemask,
730 &totalpages);
David Rientjese85bfd32010-09-22 13:05:10 -0700731 mpol_mask = (constraint == CONSTRAINT_MEMORY_POLICY) ? nodemask : NULL;
732 check_panic_on_oom(constraint, gfp_mask, order, mpol_mask);
David Rientjes0aad4b32010-08-09 17:18:59 -0700733
David Rientjes2b45ab32007-05-06 14:50:00 -0700734 read_lock(&tasklist_lock);
KOSAKI Motohirof88ccad2010-08-09 17:19:36 -0700735 if (sysctl_oom_kill_allocating_task &&
KOSAKI Motohiroa96cfd62010-08-09 17:19:38 -0700736 !oom_unkillable_task(current, NULL, nodemask) &&
David Rientjesc9f01242011-10-31 17:07:15 -0700737 current->mm) {
David Rientjes2a1c9b12012-03-21 16:33:46 -0700738 oom_kill_process(current, gfp_mask, order, 0, totalpages, NULL,
739 nodemask,
740 "Out of memory (oom_kill_allocating_task)");
741 goto out;
David Rientjes0aad4b32010-08-09 17:18:59 -0700742 }
743
David Rientjes08ab9b12012-03-21 16:34:04 -0700744 p = select_bad_process(&points, totalpages, NULL, mpol_mask,
745 force_kill);
David Rientjes0aad4b32010-08-09 17:18:59 -0700746 /* Found nothing?!?! Either we hang forever, or we panic. */
747 if (!p) {
David Rientjese85bfd32010-09-22 13:05:10 -0700748 dump_header(NULL, gfp_mask, order, NULL, mpol_mask);
David Rientjes0aad4b32010-08-09 17:18:59 -0700749 read_unlock(&tasklist_lock);
750 panic("Out of memory and no killable processes...\n");
751 }
David Rientjes2a1c9b12012-03-21 16:33:46 -0700752 if (PTR_ERR(p) != -1UL) {
753 oom_kill_process(p, gfp_mask, order, points, totalpages, NULL,
754 nodemask, "Out of memory");
755 killed = 1;
756 }
KOSAKI Motohirob52723c2010-08-19 14:13:39 -0700757out:
Andrew Morton140ffce2006-03-02 02:54:28 -0800758 read_unlock(&tasklist_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759
760 /*
761 * Give "p" a good chance of killing itself before we
Kirill Korotaev2f659f42006-01-08 01:01:05 -0800762 * retry to allocate memory unless "p" is current
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763 */
KOSAKI Motohirob52723c2010-08-19 14:13:39 -0700764 if (killed && !test_thread_flag(TIF_MEMDIE))
Andrew Morton140ffce2006-03-02 02:54:28 -0800765 schedule_timeout_uninterruptible(1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766}
David Rientjese3658932010-08-09 17:18:55 -0700767
768/*
769 * The pagefault handler calls here because it is out of memory, so kill a
770 * memory-hogging task. If a populated zone has ZONE_OOM_LOCKED set, a parallel
771 * oom killing is already in progress so do nothing. If a task is found with
772 * TIF_MEMDIE set, it has been killed so do nothing and allow it to exit.
773 */
774void pagefault_out_of_memory(void)
775{
776 if (try_set_system_oom()) {
David Rientjes08ab9b12012-03-21 16:34:04 -0700777 out_of_memory(NULL, 0, 0, NULL, false);
David Rientjese3658932010-08-09 17:18:55 -0700778 clear_system_oom();
779 }
780 if (!test_thread_flag(TIF_MEMDIE))
781 schedule_timeout_uninterruptible(1);
782}