Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/mm/oom_kill.c | |
3 | * | |
4 | * Copyright (C) 1998,2000 Rik van Riel | |
5 | * Thanks go out to Claus Fischer for some serious inspiration and | |
6 | * for goading me into coding this file... | |
7 | * | |
8 | * The routines in this file are used to kill a process when | |
a49335cc PJ |
9 | * we're seriously out of memory. This gets called from __alloc_pages() |
10 | * in mm/page_alloc.c when we really run out of memory. | |
1da177e4 LT |
11 | * |
12 | * Since we won't call these routines often (on a well-configured | |
13 | * machine) this file will double as a 'coding guide' and a signpost | |
14 | * for newbie kernel hackers. It features several pointers to major | |
15 | * kernel subsystems and hints as to where to find out what things do. | |
16 | */ | |
17 | ||
8ac773b4 | 18 | #include <linux/oom.h> |
1da177e4 | 19 | #include <linux/mm.h> |
4e950f6f | 20 | #include <linux/err.h> |
1da177e4 LT |
21 | #include <linux/sched.h> |
22 | #include <linux/swap.h> | |
23 | #include <linux/timex.h> | |
24 | #include <linux/jiffies.h> | |
ef08e3b4 | 25 | #include <linux/cpuset.h> |
8bc719d3 MS |
26 | #include <linux/module.h> |
27 | #include <linux/notifier.h> | |
c7ba5c9e | 28 | #include <linux/memcontrol.h> |
5cd9c58f | 29 | #include <linux/security.h> |
1da177e4 | 30 | |
fadd8fbd | 31 | int sysctl_panic_on_oom; |
fe071d7e | 32 | int sysctl_oom_kill_allocating_task; |
fef1bdd6 | 33 | int sysctl_oom_dump_tasks; |
c7d4caeb | 34 | static DEFINE_SPINLOCK(zone_scan_lock); |
1da177e4 LT |
35 | /* #define DEBUG */ |
36 | ||
37 | /** | |
6937a25c | 38 | * badness - calculate a numeric value for how bad this task has been |
1da177e4 | 39 | * @p: task struct of which task we should calculate |
a49335cc | 40 | * @uptime: current uptime in seconds |
1da177e4 LT |
41 | * |
42 | * The formula used is relatively simple and documented inline in the | |
43 | * function. The main rationale is that we want to select a good task | |
44 | * to kill when we run out of memory. | |
45 | * | |
46 | * Good in this context means that: | |
47 | * 1) we lose the minimum amount of work done | |
48 | * 2) we recover a large amount of memory | |
49 | * 3) we don't kill anything innocent of eating tons of memory | |
50 | * 4) we want to kill the minimum amount of processes (one) | |
51 | * 5) we try to kill the process the user expects us to kill, this | |
52 | * algorithm has been meticulously tuned to meet the principle | |
53 | * of least surprise ... (be careful when you change it) | |
54 | */ | |
55 | ||
97d87c97 | 56 | unsigned long badness(struct task_struct *p, unsigned long uptime) |
1da177e4 | 57 | { |
a12888f7 | 58 | unsigned long points, cpu_time, run_time; |
97c2c9b8 AM |
59 | struct mm_struct *mm; |
60 | struct task_struct *child; | |
2ff05b2b | 61 | int oom_adj; |
1da177e4 | 62 | |
97c2c9b8 AM |
63 | task_lock(p); |
64 | mm = p->mm; | |
65 | if (!mm) { | |
66 | task_unlock(p); | |
1da177e4 | 67 | return 0; |
97c2c9b8 | 68 | } |
2ff05b2b | 69 | oom_adj = mm->oom_adj; |
4d8b9135 DR |
70 | if (oom_adj == OOM_DISABLE) { |
71 | task_unlock(p); | |
72 | return 0; | |
73 | } | |
1da177e4 LT |
74 | |
75 | /* | |
76 | * The memory size of the process is the basis for the badness. | |
77 | */ | |
97c2c9b8 AM |
78 | points = mm->total_vm; |
79 | ||
80 | /* | |
81 | * After this unlock we can no longer dereference local variable `mm' | |
82 | */ | |
83 | task_unlock(p); | |
1da177e4 | 84 | |
7ba34859 HD |
85 | /* |
86 | * swapoff can easily use up all memory, so kill those first. | |
87 | */ | |
88 | if (p->flags & PF_SWAPOFF) | |
89 | return ULONG_MAX; | |
90 | ||
1da177e4 LT |
91 | /* |
92 | * Processes which fork a lot of child processes are likely | |
9827b781 | 93 | * a good choice. We add half the vmsize of the children if they |
1da177e4 | 94 | * have an own mm. This prevents forking servers to flood the |
9827b781 KG |
95 | * machine with an endless amount of children. In case a single |
96 | * child is eating the vast majority of memory, adding only half | |
97 | * to the parents will make the child our kill candidate of choice. | |
1da177e4 | 98 | */ |
97c2c9b8 AM |
99 | list_for_each_entry(child, &p->children, sibling) { |
100 | task_lock(child); | |
101 | if (child->mm != mm && child->mm) | |
102 | points += child->mm->total_vm/2 + 1; | |
103 | task_unlock(child); | |
1da177e4 LT |
104 | } |
105 | ||
106 | /* | |
107 | * CPU time is in tens of seconds and run time is in thousands | |
108 | * of seconds. There is no particular reason for this other than | |
109 | * that it turned out to work very well in practice. | |
110 | */ | |
111 | cpu_time = (cputime_to_jiffies(p->utime) + cputime_to_jiffies(p->stime)) | |
112 | >> (SHIFT_HZ + 3); | |
113 | ||
114 | if (uptime >= p->start_time.tv_sec) | |
115 | run_time = (uptime - p->start_time.tv_sec) >> 10; | |
116 | else | |
117 | run_time = 0; | |
118 | ||
a12888f7 CG |
119 | if (cpu_time) |
120 | points /= int_sqrt(cpu_time); | |
121 | if (run_time) | |
122 | points /= int_sqrt(int_sqrt(run_time)); | |
1da177e4 LT |
123 | |
124 | /* | |
125 | * Niced processes are most likely less important, so double | |
126 | * their badness points. | |
127 | */ | |
128 | if (task_nice(p) > 0) | |
129 | points *= 2; | |
130 | ||
131 | /* | |
132 | * Superuser processes are usually more important, so we make it | |
133 | * less likely that we kill those. | |
134 | */ | |
a2f2945a EP |
135 | if (has_capability_noaudit(p, CAP_SYS_ADMIN) || |
136 | has_capability_noaudit(p, CAP_SYS_RESOURCE)) | |
1da177e4 LT |
137 | points /= 4; |
138 | ||
139 | /* | |
140 | * We don't want to kill a process with direct hardware access. | |
141 | * Not only could that mess up the hardware, but usually users | |
142 | * tend to only have this flag set on applications they think | |
143 | * of as important. | |
144 | */ | |
a2f2945a | 145 | if (has_capability_noaudit(p, CAP_SYS_RAWIO)) |
1da177e4 LT |
146 | points /= 4; |
147 | ||
7887a3da NP |
148 | /* |
149 | * If p's nodes don't overlap ours, it may still help to kill p | |
150 | * because p may have allocated or otherwise mapped memory on | |
151 | * this node before. However it will be less likely. | |
152 | */ | |
bbe373f2 | 153 | if (!cpuset_mems_allowed_intersects(current, p)) |
7887a3da NP |
154 | points /= 8; |
155 | ||
1da177e4 | 156 | /* |
2ff05b2b | 157 | * Adjust the score by oom_adj. |
1da177e4 | 158 | */ |
2ff05b2b DR |
159 | if (oom_adj) { |
160 | if (oom_adj > 0) { | |
9a82782f JP |
161 | if (!points) |
162 | points = 1; | |
2ff05b2b | 163 | points <<= oom_adj; |
9a82782f | 164 | } else |
2ff05b2b | 165 | points >>= -(oom_adj); |
1da177e4 LT |
166 | } |
167 | ||
168 | #ifdef DEBUG | |
a5e58a61 | 169 | printk(KERN_DEBUG "OOMkill: task %d (%s) got %lu points\n", |
1da177e4 LT |
170 | p->pid, p->comm, points); |
171 | #endif | |
172 | return points; | |
173 | } | |
174 | ||
9b0f8b04 CL |
175 | /* |
176 | * Determine the type of allocation constraint. | |
177 | */ | |
70e24bdf DR |
178 | static inline enum oom_constraint constrained_alloc(struct zonelist *zonelist, |
179 | gfp_t gfp_mask) | |
9b0f8b04 CL |
180 | { |
181 | #ifdef CONFIG_NUMA | |
54a6eb5c | 182 | struct zone *zone; |
dd1a239f | 183 | struct zoneref *z; |
54a6eb5c | 184 | enum zone_type high_zoneidx = gfp_zone(gfp_mask); |
ee31af5d | 185 | nodemask_t nodes = node_states[N_HIGH_MEMORY]; |
9b0f8b04 | 186 | |
54a6eb5c MG |
187 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) |
188 | if (cpuset_zone_allowed_softwall(zone, gfp_mask)) | |
189 | node_clear(zone_to_nid(zone), nodes); | |
9b0f8b04 CL |
190 | else |
191 | return CONSTRAINT_CPUSET; | |
192 | ||
193 | if (!nodes_empty(nodes)) | |
194 | return CONSTRAINT_MEMORY_POLICY; | |
195 | #endif | |
196 | ||
197 | return CONSTRAINT_NONE; | |
198 | } | |
199 | ||
1da177e4 LT |
200 | /* |
201 | * Simple selection loop. We chose the process with the highest | |
202 | * number of 'points'. We expect the caller will lock the tasklist. | |
203 | * | |
204 | * (not docbooked, we don't want this one cluttering up the manual) | |
205 | */ | |
c7ba5c9e PE |
206 | static struct task_struct *select_bad_process(unsigned long *ppoints, |
207 | struct mem_cgroup *mem) | |
1da177e4 | 208 | { |
1da177e4 LT |
209 | struct task_struct *g, *p; |
210 | struct task_struct *chosen = NULL; | |
211 | struct timespec uptime; | |
9827b781 | 212 | *ppoints = 0; |
1da177e4 LT |
213 | |
214 | do_posix_clock_monotonic_gettime(&uptime); | |
a49335cc PJ |
215 | do_each_thread(g, p) { |
216 | unsigned long points; | |
a49335cc | 217 | |
28324d1d ON |
218 | /* |
219 | * skip kernel threads and tasks which have already released | |
220 | * their mm. | |
221 | */ | |
5081dde3 NP |
222 | if (!p->mm) |
223 | continue; | |
28324d1d | 224 | /* skip the init task */ |
b460cbc5 | 225 | if (is_global_init(p)) |
a49335cc | 226 | continue; |
4c4a2214 DR |
227 | if (mem && !task_in_mem_cgroup(p, mem)) |
228 | continue; | |
ef08e3b4 | 229 | |
b78483a4 NP |
230 | /* |
231 | * This task already has access to memory reserves and is | |
232 | * being killed. Don't allow any other task access to the | |
233 | * memory reserve. | |
234 | * | |
235 | * Note: this may have a chance of deadlock if it gets | |
236 | * blocked waiting for another task which itself is waiting | |
237 | * for memory. Is there a better alternative? | |
238 | */ | |
239 | if (test_tsk_thread_flag(p, TIF_MEMDIE)) | |
240 | return ERR_PTR(-1UL); | |
241 | ||
a49335cc | 242 | /* |
6937a25c | 243 | * This is in the process of releasing memory so wait for it |
a49335cc | 244 | * to finish before killing some other task by mistake. |
50ec3bbf NP |
245 | * |
246 | * However, if p is the current task, we allow the 'kill' to | |
247 | * go ahead if it is exiting: this will simply set TIF_MEMDIE, | |
248 | * which will allow it to gain access to memory reserves in | |
249 | * the process of exiting and releasing its resources. | |
b78483a4 | 250 | * Otherwise we could get an easy OOM deadlock. |
a49335cc | 251 | */ |
b78483a4 NP |
252 | if (p->flags & PF_EXITING) { |
253 | if (p != current) | |
254 | return ERR_PTR(-1UL); | |
255 | ||
972c4ea5 ON |
256 | chosen = p; |
257 | *ppoints = ULONG_MAX; | |
50ec3bbf | 258 | } |
972c4ea5 | 259 | |
97d87c97 | 260 | points = badness(p, uptime.tv_sec); |
4d8b9135 | 261 | if (points > *ppoints) { |
a49335cc | 262 | chosen = p; |
9827b781 | 263 | *ppoints = points; |
1da177e4 | 264 | } |
a49335cc | 265 | } while_each_thread(g, p); |
972c4ea5 | 266 | |
1da177e4 LT |
267 | return chosen; |
268 | } | |
269 | ||
fef1bdd6 | 270 | /** |
1b578df0 RD |
271 | * dump_tasks - dump current memory state of all system tasks |
272 | * @mem: target memory controller | |
273 | * | |
fef1bdd6 DR |
274 | * Dumps the current memory state of all system tasks, excluding kernel threads. |
275 | * State information includes task's pid, uid, tgid, vm size, rss, cpu, oom_adj | |
276 | * score, and name. | |
277 | * | |
278 | * If the actual is non-NULL, only tasks that are a member of the mem_cgroup are | |
279 | * shown. | |
280 | * | |
281 | * Call with tasklist_lock read-locked. | |
282 | */ | |
283 | static void dump_tasks(const struct mem_cgroup *mem) | |
284 | { | |
285 | struct task_struct *g, *p; | |
286 | ||
287 | printk(KERN_INFO "[ pid ] uid tgid total_vm rss cpu oom_adj " | |
288 | "name\n"); | |
289 | do_each_thread(g, p) { | |
6d2661ed DR |
290 | struct mm_struct *mm; |
291 | ||
fef1bdd6 DR |
292 | if (mem && !task_in_mem_cgroup(p, mem)) |
293 | continue; | |
b4416d2b DR |
294 | if (!thread_group_leader(p)) |
295 | continue; | |
fef1bdd6 DR |
296 | |
297 | task_lock(p); | |
6d2661ed DR |
298 | mm = p->mm; |
299 | if (!mm) { | |
300 | /* | |
301 | * total_vm and rss sizes do not exist for tasks with no | |
302 | * mm so there's no need to report them; they can't be | |
303 | * oom killed anyway. | |
304 | */ | |
305 | task_unlock(p); | |
306 | continue; | |
307 | } | |
fef1bdd6 | 308 | printk(KERN_INFO "[%5d] %5d %5d %8lu %8lu %3d %3d %s\n", |
6d2661ed | 309 | p->pid, __task_cred(p)->uid, p->tgid, mm->total_vm, |
2ff05b2b | 310 | get_mm_rss(mm), (int)task_cpu(p), mm->oom_adj, p->comm); |
fef1bdd6 DR |
311 | task_unlock(p); |
312 | } while_each_thread(g, p); | |
313 | } | |
314 | ||
1b578df0 | 315 | /* |
5a291b98 RG |
316 | * Send SIGKILL to the selected process irrespective of CAP_SYS_RAW_IO |
317 | * flag though it's unlikely that we select a process with CAP_SYS_RAW_IO | |
318 | * set. | |
1da177e4 | 319 | */ |
f3af38d3 | 320 | static void __oom_kill_task(struct task_struct *p, int verbose) |
1da177e4 | 321 | { |
b460cbc5 | 322 | if (is_global_init(p)) { |
1da177e4 LT |
323 | WARN_ON(1); |
324 | printk(KERN_WARNING "tried to kill init!\n"); | |
325 | return; | |
326 | } | |
327 | ||
01017a22 | 328 | if (!p->mm) { |
1da177e4 LT |
329 | WARN_ON(1); |
330 | printk(KERN_WARNING "tried to kill an mm-less task!\n"); | |
1da177e4 LT |
331 | return; |
332 | } | |
50ec3bbf | 333 | |
f3af38d3 | 334 | if (verbose) |
ba25f9dc PE |
335 | printk(KERN_ERR "Killed process %d (%s)\n", |
336 | task_pid_nr(p), p->comm); | |
1da177e4 LT |
337 | |
338 | /* | |
339 | * We give our sacrificial lamb high priority and access to | |
340 | * all the memory it needs. That way it should be able to | |
341 | * exit() and clear out its resources quickly... | |
342 | */ | |
fa717060 | 343 | p->rt.time_slice = HZ; |
1da177e4 LT |
344 | set_tsk_thread_flag(p, TIF_MEMDIE); |
345 | ||
346 | force_sig(SIGKILL, p); | |
347 | } | |
348 | ||
f3af38d3 | 349 | static int oom_kill_task(struct task_struct *p) |
1da177e4 | 350 | { |
01315922 | 351 | struct mm_struct *mm; |
36c8b586 | 352 | struct task_struct *g, *q; |
1da177e4 | 353 | |
4d8b9135 | 354 | task_lock(p); |
01315922 | 355 | mm = p->mm; |
4d8b9135 DR |
356 | if (!mm || mm->oom_adj == OOM_DISABLE) { |
357 | task_unlock(p); | |
01315922 | 358 | return 1; |
4d8b9135 DR |
359 | } |
360 | task_unlock(p); | |
f3af38d3 | 361 | __oom_kill_task(p, 1); |
c33e0fca | 362 | |
1da177e4 LT |
363 | /* |
364 | * kill all processes that share the ->mm (i.e. all threads), | |
f2a2a710 NP |
365 | * but are in a different thread group. Don't let them have access |
366 | * to memory reserves though, otherwise we might deplete all memory. | |
1da177e4 | 367 | */ |
c33e0fca | 368 | do_each_thread(g, q) { |
bac0abd6 | 369 | if (q->mm == mm && !same_thread_group(q, p)) |
650a7c97 | 370 | force_sig(SIGKILL, q); |
c33e0fca | 371 | } while_each_thread(g, q); |
1da177e4 | 372 | |
01315922 | 373 | return 0; |
1da177e4 LT |
374 | } |
375 | ||
7213f506 | 376 | static int oom_kill_process(struct task_struct *p, gfp_t gfp_mask, int order, |
fef1bdd6 DR |
377 | unsigned long points, struct mem_cgroup *mem, |
378 | const char *message) | |
1da177e4 | 379 | { |
1da177e4 | 380 | struct task_struct *c; |
1da177e4 | 381 | |
7213f506 | 382 | if (printk_ratelimit()) { |
75aa1994 | 383 | task_lock(current); |
2ff05b2b DR |
384 | printk(KERN_WARNING "%s invoked oom-killer: " |
385 | "gfp_mask=0x%x, order=%d, oom_adj=%d\n", | |
386 | current->comm, gfp_mask, order, | |
387 | current->mm ? current->mm->oom_adj : OOM_DISABLE); | |
75aa1994 DR |
388 | cpuset_print_task_mems_allowed(current); |
389 | task_unlock(current); | |
7213f506 | 390 | dump_stack(); |
e222432b | 391 | mem_cgroup_print_oom_info(mem, current); |
7213f506 | 392 | show_mem(); |
fef1bdd6 DR |
393 | if (sysctl_oom_dump_tasks) |
394 | dump_tasks(mem); | |
7213f506 DR |
395 | } |
396 | ||
50ec3bbf NP |
397 | /* |
398 | * If the task is already exiting, don't alarm the sysadmin or kill | |
399 | * its children or threads, just set TIF_MEMDIE so it can die quickly | |
400 | */ | |
401 | if (p->flags & PF_EXITING) { | |
f3af38d3 | 402 | __oom_kill_task(p, 0); |
50ec3bbf NP |
403 | return 0; |
404 | } | |
405 | ||
f3af38d3 | 406 | printk(KERN_ERR "%s: kill process %d (%s) score %li or a child\n", |
ba25f9dc | 407 | message, task_pid_nr(p), p->comm, points); |
f3af38d3 | 408 | |
1da177e4 | 409 | /* Try to kill a child first */ |
7b1915a9 | 410 | list_for_each_entry(c, &p->children, sibling) { |
1da177e4 LT |
411 | if (c->mm == p->mm) |
412 | continue; | |
f3af38d3 | 413 | if (!oom_kill_task(c)) |
01315922 | 414 | return 0; |
1da177e4 | 415 | } |
f3af38d3 | 416 | return oom_kill_task(p); |
1da177e4 LT |
417 | } |
418 | ||
00f0b825 | 419 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR |
c7ba5c9e PE |
420 | void mem_cgroup_out_of_memory(struct mem_cgroup *mem, gfp_t gfp_mask) |
421 | { | |
422 | unsigned long points = 0; | |
423 | struct task_struct *p; | |
424 | ||
e115f2d8 | 425 | read_lock(&tasklist_lock); |
c7ba5c9e PE |
426 | retry: |
427 | p = select_bad_process(&points, mem); | |
428 | if (PTR_ERR(p) == -1UL) | |
429 | goto out; | |
430 | ||
431 | if (!p) | |
432 | p = current; | |
433 | ||
fef1bdd6 | 434 | if (oom_kill_process(p, gfp_mask, 0, points, mem, |
c7ba5c9e PE |
435 | "Memory cgroup out of memory")) |
436 | goto retry; | |
437 | out: | |
e115f2d8 | 438 | read_unlock(&tasklist_lock); |
c7ba5c9e PE |
439 | } |
440 | #endif | |
441 | ||
8bc719d3 MS |
442 | static BLOCKING_NOTIFIER_HEAD(oom_notify_list); |
443 | ||
444 | int register_oom_notifier(struct notifier_block *nb) | |
445 | { | |
446 | return blocking_notifier_chain_register(&oom_notify_list, nb); | |
447 | } | |
448 | EXPORT_SYMBOL_GPL(register_oom_notifier); | |
449 | ||
450 | int unregister_oom_notifier(struct notifier_block *nb) | |
451 | { | |
452 | return blocking_notifier_chain_unregister(&oom_notify_list, nb); | |
453 | } | |
454 | EXPORT_SYMBOL_GPL(unregister_oom_notifier); | |
455 | ||
098d7f12 DR |
456 | /* |
457 | * Try to acquire the OOM killer lock for the zones in zonelist. Returns zero | |
458 | * if a parallel OOM killing is already taking place that includes a zone in | |
459 | * the zonelist. Otherwise, locks all zones in the zonelist and returns 1. | |
460 | */ | |
dd1a239f | 461 | int try_set_zone_oom(struct zonelist *zonelist, gfp_t gfp_mask) |
098d7f12 | 462 | { |
dd1a239f MG |
463 | struct zoneref *z; |
464 | struct zone *zone; | |
098d7f12 DR |
465 | int ret = 1; |
466 | ||
c7d4caeb | 467 | spin_lock(&zone_scan_lock); |
dd1a239f MG |
468 | for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) { |
469 | if (zone_is_oom_locked(zone)) { | |
098d7f12 DR |
470 | ret = 0; |
471 | goto out; | |
472 | } | |
dd1a239f MG |
473 | } |
474 | ||
475 | for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) { | |
476 | /* | |
c7d4caeb | 477 | * Lock each zone in the zonelist under zone_scan_lock so a |
dd1a239f MG |
478 | * parallel invocation of try_set_zone_oom() doesn't succeed |
479 | * when it shouldn't. | |
480 | */ | |
481 | zone_set_flag(zone, ZONE_OOM_LOCKED); | |
482 | } | |
098d7f12 | 483 | |
098d7f12 | 484 | out: |
c7d4caeb | 485 | spin_unlock(&zone_scan_lock); |
098d7f12 DR |
486 | return ret; |
487 | } | |
488 | ||
489 | /* | |
490 | * Clears the ZONE_OOM_LOCKED flag for all zones in the zonelist so that failed | |
491 | * allocation attempts with zonelists containing them may now recall the OOM | |
492 | * killer, if necessary. | |
493 | */ | |
dd1a239f | 494 | void clear_zonelist_oom(struct zonelist *zonelist, gfp_t gfp_mask) |
098d7f12 | 495 | { |
dd1a239f MG |
496 | struct zoneref *z; |
497 | struct zone *zone; | |
098d7f12 | 498 | |
c7d4caeb | 499 | spin_lock(&zone_scan_lock); |
dd1a239f MG |
500 | for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) { |
501 | zone_clear_flag(zone, ZONE_OOM_LOCKED); | |
502 | } | |
c7d4caeb | 503 | spin_unlock(&zone_scan_lock); |
098d7f12 DR |
504 | } |
505 | ||
1c0fe6e3 NP |
506 | /* |
507 | * Must be called with tasklist_lock held for read. | |
508 | */ | |
509 | static void __out_of_memory(gfp_t gfp_mask, int order) | |
510 | { | |
184101bf DR |
511 | struct task_struct *p; |
512 | unsigned long points; | |
1c0fe6e3 | 513 | |
184101bf DR |
514 | if (sysctl_oom_kill_allocating_task) |
515 | if (!oom_kill_process(current, gfp_mask, order, 0, NULL, | |
516 | "Out of memory (oom_kill_allocating_task)")) | |
1c0fe6e3 | 517 | return; |
184101bf DR |
518 | retry: |
519 | /* | |
520 | * Rambo mode: Shoot down a process and hope it solves whatever | |
521 | * issues we may have. | |
522 | */ | |
523 | p = select_bad_process(&points, NULL); | |
1c0fe6e3 | 524 | |
184101bf DR |
525 | if (PTR_ERR(p) == -1UL) |
526 | return; | |
1c0fe6e3 | 527 | |
184101bf DR |
528 | /* Found nothing?!?! Either we hang forever, or we panic. */ |
529 | if (!p) { | |
530 | read_unlock(&tasklist_lock); | |
531 | panic("Out of memory and no killable processes...\n"); | |
1c0fe6e3 | 532 | } |
184101bf DR |
533 | |
534 | if (oom_kill_process(p, gfp_mask, order, points, NULL, | |
535 | "Out of memory")) | |
536 | goto retry; | |
1c0fe6e3 NP |
537 | } |
538 | ||
539 | /* | |
540 | * pagefault handler calls into here because it is out of memory but | |
541 | * doesn't know exactly how or why. | |
542 | */ | |
543 | void pagefault_out_of_memory(void) | |
544 | { | |
545 | unsigned long freed = 0; | |
546 | ||
547 | blocking_notifier_call_chain(&oom_notify_list, 0, &freed); | |
548 | if (freed > 0) | |
549 | /* Got some memory back in the last second. */ | |
550 | return; | |
551 | ||
a636b327 KH |
552 | /* |
553 | * If this is from memcg, oom-killer is already invoked. | |
554 | * and not worth to go system-wide-oom. | |
555 | */ | |
556 | if (mem_cgroup_oom_called(current)) | |
557 | goto rest_and_return; | |
558 | ||
1c0fe6e3 NP |
559 | if (sysctl_panic_on_oom) |
560 | panic("out of memory from page fault. panic_on_oom is selected.\n"); | |
561 | ||
562 | read_lock(&tasklist_lock); | |
563 | __out_of_memory(0, 0); /* unknown gfp_mask and order */ | |
564 | read_unlock(&tasklist_lock); | |
565 | ||
566 | /* | |
567 | * Give "p" a good chance of killing itself before we | |
568 | * retry to allocate memory. | |
569 | */ | |
a636b327 | 570 | rest_and_return: |
1c0fe6e3 NP |
571 | if (!test_thread_flag(TIF_MEMDIE)) |
572 | schedule_timeout_uninterruptible(1); | |
573 | } | |
574 | ||
1da177e4 | 575 | /** |
6937a25c | 576 | * out_of_memory - kill the "best" process when we run out of memory |
1b578df0 RD |
577 | * @zonelist: zonelist pointer |
578 | * @gfp_mask: memory allocation flags | |
579 | * @order: amount of memory being requested as a power of 2 | |
1da177e4 LT |
580 | * |
581 | * If we run out of memory, we have the choice between either | |
582 | * killing a random task (bad), letting the system crash (worse) | |
583 | * OR try to be smart about which process to kill. Note that we | |
584 | * don't have to be perfect here, we just have to be good. | |
585 | */ | |
9b0f8b04 | 586 | void out_of_memory(struct zonelist *zonelist, gfp_t gfp_mask, int order) |
1da177e4 | 587 | { |
8bc719d3 | 588 | unsigned long freed = 0; |
70e24bdf | 589 | enum oom_constraint constraint; |
8bc719d3 MS |
590 | |
591 | blocking_notifier_call_chain(&oom_notify_list, 0, &freed); | |
592 | if (freed > 0) | |
593 | /* Got some memory back in the last second. */ | |
594 | return; | |
1da177e4 | 595 | |
2b744c01 YG |
596 | if (sysctl_panic_on_oom == 2) |
597 | panic("out of memory. Compulsory panic_on_oom is selected.\n"); | |
598 | ||
9b0f8b04 CL |
599 | /* |
600 | * Check if there were limitations on the allocation (only relevant for | |
601 | * NUMA) that may require different handling. | |
602 | */ | |
2b45ab33 | 603 | constraint = constrained_alloc(zonelist, gfp_mask); |
2b45ab33 DR |
604 | read_lock(&tasklist_lock); |
605 | ||
606 | switch (constraint) { | |
9b0f8b04 | 607 | case CONSTRAINT_MEMORY_POLICY: |
1c0fe6e3 | 608 | oom_kill_process(current, gfp_mask, order, 0, NULL, |
9b0f8b04 CL |
609 | "No available memory (MPOL_BIND)"); |
610 | break; | |
611 | ||
9b0f8b04 | 612 | case CONSTRAINT_NONE: |
fadd8fbd KH |
613 | if (sysctl_panic_on_oom) |
614 | panic("out of memory. panic_on_oom is selected\n"); | |
fe071d7e DR |
615 | /* Fall-through */ |
616 | case CONSTRAINT_CPUSET: | |
1c0fe6e3 | 617 | __out_of_memory(gfp_mask, order); |
9b0f8b04 CL |
618 | break; |
619 | } | |
1da177e4 | 620 | |
140ffcec | 621 | read_unlock(&tasklist_lock); |
1da177e4 LT |
622 | |
623 | /* | |
624 | * Give "p" a good chance of killing itself before we | |
2f659f46 | 625 | * retry to allocate memory unless "p" is current |
1da177e4 | 626 | */ |
2f659f46 | 627 | if (!test_thread_flag(TIF_MEMDIE)) |
140ffcec | 628 | schedule_timeout_uninterruptible(1); |
1da177e4 | 629 | } |