Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 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 |
Paul Jackson | a49335c | 2005-09-06 15:18:09 -0700 | [diff] [blame] | 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. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 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 | |
Alexey Dobriyan | 8ac773b | 2006-10-19 23:28:32 -0700 | [diff] [blame] | 18 | #include <linux/oom.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 19 | #include <linux/mm.h> |
Alexey Dobriyan | 4e950f6 | 2007-07-30 02:36:13 +0400 | [diff] [blame] | 20 | #include <linux/err.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 21 | #include <linux/sched.h> |
| 22 | #include <linux/swap.h> |
| 23 | #include <linux/timex.h> |
| 24 | #include <linux/jiffies.h> |
Paul Jackson | ef08e3b | 2005-09-06 15:18:13 -0700 | [diff] [blame] | 25 | #include <linux/cpuset.h> |
Martin Schwidefsky | 8bc719d | 2006-09-25 23:31:20 -0700 | [diff] [blame] | 26 | #include <linux/module.h> |
| 27 | #include <linux/notifier.h> |
Pavel Emelianov | c7ba5c9 | 2008-02-07 00:13:58 -0800 | [diff] [blame] | 28 | #include <linux/memcontrol.h> |
David Howells | 5cd9c58 | 2008-08-14 11:37:28 +0100 | [diff] [blame] | 29 | #include <linux/security.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 30 | |
KAMEZAWA Hiroyuki | fadd8fb | 2006-06-23 02:03:13 -0700 | [diff] [blame] | 31 | int sysctl_panic_on_oom; |
David Rientjes | fe071d7 | 2007-10-16 23:25:56 -0700 | [diff] [blame] | 32 | int sysctl_oom_kill_allocating_task; |
David Rientjes | fef1bdd | 2008-02-07 00:14:07 -0800 | [diff] [blame] | 33 | int sysctl_oom_dump_tasks; |
David Rientjes | c7d4cae | 2009-01-06 14:39:00 -0800 | [diff] [blame] | 34 | static DEFINE_SPINLOCK(zone_scan_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 35 | /* #define DEBUG */ |
| 36 | |
KOSAKI Motohiro | 495789a | 2009-09-21 17:03:14 -0700 | [diff] [blame] | 37 | /* |
| 38 | * Is all threads of the target process nodes overlap ours? |
| 39 | */ |
| 40 | static int has_intersects_mems_allowed(struct task_struct *tsk) |
| 41 | { |
| 42 | struct task_struct *t; |
| 43 | |
| 44 | t = tsk; |
| 45 | do { |
| 46 | if (cpuset_mems_allowed_intersects(current, t)) |
| 47 | return 1; |
| 48 | t = next_thread(t); |
| 49 | } while (t != tsk); |
| 50 | |
| 51 | return 0; |
| 52 | } |
| 53 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 54 | /** |
Dave Peterson | 6937a25 | 2006-06-23 02:03:13 -0700 | [diff] [blame] | 55 | * badness - calculate a numeric value for how bad this task has been |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 56 | * @p: task struct of which task we should calculate |
Paul Jackson | a49335c | 2005-09-06 15:18:09 -0700 | [diff] [blame] | 57 | * @uptime: current uptime in seconds |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 58 | * |
| 59 | * The formula used is relatively simple and documented inline in the |
| 60 | * function. The main rationale is that we want to select a good task |
| 61 | * to kill when we run out of memory. |
| 62 | * |
| 63 | * Good in this context means that: |
| 64 | * 1) we lose the minimum amount of work done |
| 65 | * 2) we recover a large amount of memory |
| 66 | * 3) we don't kill anything innocent of eating tons of memory |
| 67 | * 4) we want to kill the minimum amount of processes (one) |
| 68 | * 5) we try to kill the process the user expects us to kill, this |
| 69 | * algorithm has been meticulously tuned to meet the principle |
| 70 | * of least surprise ... (be careful when you change it) |
| 71 | */ |
| 72 | |
Li Zefan | 97d87c9 | 2008-04-28 02:13:35 -0700 | [diff] [blame] | 73 | unsigned long badness(struct task_struct *p, unsigned long uptime) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 74 | { |
Cyrill Gorcunov | a12888f | 2009-03-31 15:19:27 -0700 | [diff] [blame] | 75 | unsigned long points, cpu_time, run_time; |
Andrew Morton | 97c2c9b | 2006-04-18 22:20:38 -0700 | [diff] [blame] | 76 | struct mm_struct *mm; |
| 77 | struct task_struct *child; |
KOSAKI Motohiro | 28b83c5 | 2009-09-21 17:03:13 -0700 | [diff] [blame] | 78 | int oom_adj = p->signal->oom_adj; |
KOSAKI Motohiro | 495789a | 2009-09-21 17:03:14 -0700 | [diff] [blame] | 79 | struct task_cputime task_time; |
| 80 | unsigned long utime; |
| 81 | unsigned long stime; |
KOSAKI Motohiro | 28b83c5 | 2009-09-21 17:03:13 -0700 | [diff] [blame] | 82 | |
| 83 | if (oom_adj == OOM_DISABLE) |
| 84 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 85 | |
Andrew Morton | 97c2c9b | 2006-04-18 22:20:38 -0700 | [diff] [blame] | 86 | task_lock(p); |
| 87 | mm = p->mm; |
| 88 | if (!mm) { |
| 89 | task_unlock(p); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 90 | return 0; |
Andrew Morton | 97c2c9b | 2006-04-18 22:20:38 -0700 | [diff] [blame] | 91 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 92 | |
| 93 | /* |
| 94 | * The memory size of the process is the basis for the badness. |
| 95 | */ |
Andrew Morton | 97c2c9b | 2006-04-18 22:20:38 -0700 | [diff] [blame] | 96 | points = mm->total_vm; |
| 97 | |
| 98 | /* |
| 99 | * After this unlock we can no longer dereference local variable `mm' |
| 100 | */ |
| 101 | task_unlock(p); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 102 | |
| 103 | /* |
Hugh Dickins | 7ba3485 | 2007-01-05 16:37:03 -0800 | [diff] [blame] | 104 | * swapoff can easily use up all memory, so kill those first. |
| 105 | */ |
Hugh Dickins | 35451be | 2009-09-21 17:02:27 -0700 | [diff] [blame] | 106 | if (p->flags & PF_OOM_ORIGIN) |
Hugh Dickins | 7ba3485 | 2007-01-05 16:37:03 -0800 | [diff] [blame] | 107 | return ULONG_MAX; |
| 108 | |
| 109 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 110 | * Processes which fork a lot of child processes are likely |
Kurt Garloff | 9827b78 | 2006-02-20 18:27:51 -0800 | [diff] [blame] | 111 | * a good choice. We add half the vmsize of the children if they |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 112 | * have an own mm. This prevents forking servers to flood the |
Kurt Garloff | 9827b78 | 2006-02-20 18:27:51 -0800 | [diff] [blame] | 113 | * machine with an endless amount of children. In case a single |
| 114 | * child is eating the vast majority of memory, adding only half |
| 115 | * to the parents will make the child our kill candidate of choice. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 116 | */ |
Andrew Morton | 97c2c9b | 2006-04-18 22:20:38 -0700 | [diff] [blame] | 117 | list_for_each_entry(child, &p->children, sibling) { |
| 118 | task_lock(child); |
| 119 | if (child->mm != mm && child->mm) |
| 120 | points += child->mm->total_vm/2 + 1; |
| 121 | task_unlock(child); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 122 | } |
| 123 | |
| 124 | /* |
| 125 | * CPU time is in tens of seconds and run time is in thousands |
| 126 | * of seconds. There is no particular reason for this other than |
| 127 | * that it turned out to work very well in practice. |
| 128 | */ |
KOSAKI Motohiro | 495789a | 2009-09-21 17:03:14 -0700 | [diff] [blame] | 129 | thread_group_cputime(p, &task_time); |
| 130 | utime = cputime_to_jiffies(task_time.utime); |
| 131 | stime = cputime_to_jiffies(task_time.stime); |
| 132 | cpu_time = (utime + stime) >> (SHIFT_HZ + 3); |
| 133 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 134 | |
| 135 | if (uptime >= p->start_time.tv_sec) |
| 136 | run_time = (uptime - p->start_time.tv_sec) >> 10; |
| 137 | else |
| 138 | run_time = 0; |
| 139 | |
Cyrill Gorcunov | a12888f | 2009-03-31 15:19:27 -0700 | [diff] [blame] | 140 | if (cpu_time) |
| 141 | points /= int_sqrt(cpu_time); |
| 142 | if (run_time) |
| 143 | points /= int_sqrt(int_sqrt(run_time)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 144 | |
| 145 | /* |
| 146 | * Niced processes are most likely less important, so double |
| 147 | * their badness points. |
| 148 | */ |
| 149 | if (task_nice(p) > 0) |
| 150 | points *= 2; |
| 151 | |
| 152 | /* |
| 153 | * Superuser processes are usually more important, so we make it |
| 154 | * less likely that we kill those. |
| 155 | */ |
Eric Paris | a2f2945 | 2008-11-11 22:02:54 +1100 | [diff] [blame] | 156 | if (has_capability_noaudit(p, CAP_SYS_ADMIN) || |
| 157 | has_capability_noaudit(p, CAP_SYS_RESOURCE)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 158 | points /= 4; |
| 159 | |
| 160 | /* |
| 161 | * We don't want to kill a process with direct hardware access. |
| 162 | * Not only could that mess up the hardware, but usually users |
| 163 | * tend to only have this flag set on applications they think |
| 164 | * of as important. |
| 165 | */ |
Eric Paris | a2f2945 | 2008-11-11 22:02:54 +1100 | [diff] [blame] | 166 | if (has_capability_noaudit(p, CAP_SYS_RAWIO)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 167 | points /= 4; |
| 168 | |
| 169 | /* |
Nick Piggin | 7887a3d | 2006-09-25 23:31:29 -0700 | [diff] [blame] | 170 | * If p's nodes don't overlap ours, it may still help to kill p |
| 171 | * because p may have allocated or otherwise mapped memory on |
| 172 | * this node before. However it will be less likely. |
| 173 | */ |
KOSAKI Motohiro | 495789a | 2009-09-21 17:03:14 -0700 | [diff] [blame] | 174 | if (!has_intersects_mems_allowed(p)) |
Nick Piggin | 7887a3d | 2006-09-25 23:31:29 -0700 | [diff] [blame] | 175 | points /= 8; |
| 176 | |
| 177 | /* |
KOSAKI Motohiro | 28b83c5 | 2009-09-21 17:03:13 -0700 | [diff] [blame] | 178 | * Adjust the score by oom_adj. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 179 | */ |
KOSAKI Motohiro | 28b83c5 | 2009-09-21 17:03:13 -0700 | [diff] [blame] | 180 | if (oom_adj) { |
| 181 | if (oom_adj > 0) { |
Joshua N Pritikin | 9a82782 | 2007-05-06 14:49:07 -0700 | [diff] [blame] | 182 | if (!points) |
| 183 | points = 1; |
KOSAKI Motohiro | 28b83c5 | 2009-09-21 17:03:13 -0700 | [diff] [blame] | 184 | points <<= oom_adj; |
Joshua N Pritikin | 9a82782 | 2007-05-06 14:49:07 -0700 | [diff] [blame] | 185 | } else |
KOSAKI Motohiro | 28b83c5 | 2009-09-21 17:03:13 -0700 | [diff] [blame] | 186 | points >>= -(oom_adj); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 187 | } |
| 188 | |
| 189 | #ifdef DEBUG |
David Rientjes | a5e58a6 | 2007-07-31 00:37:32 -0700 | [diff] [blame] | 190 | printk(KERN_DEBUG "OOMkill: task %d (%s) got %lu points\n", |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 191 | p->pid, p->comm, points); |
| 192 | #endif |
| 193 | return points; |
| 194 | } |
| 195 | |
| 196 | /* |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 197 | * Determine the type of allocation constraint. |
| 198 | */ |
David Rientjes | 70e24bd | 2007-10-16 23:25:53 -0700 | [diff] [blame] | 199 | static inline enum oom_constraint constrained_alloc(struct zonelist *zonelist, |
| 200 | gfp_t gfp_mask) |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 201 | { |
| 202 | #ifdef CONFIG_NUMA |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 203 | struct zone *zone; |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 204 | struct zoneref *z; |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 205 | enum zone_type high_zoneidx = gfp_zone(gfp_mask); |
Christoph Lameter | ee31af5 | 2007-10-16 01:25:30 -0700 | [diff] [blame] | 206 | nodemask_t nodes = node_states[N_HIGH_MEMORY]; |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 207 | |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 208 | for_each_zone_zonelist(zone, z, zonelist, high_zoneidx) |
| 209 | if (cpuset_zone_allowed_softwall(zone, gfp_mask)) |
| 210 | node_clear(zone_to_nid(zone), nodes); |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 211 | else |
| 212 | return CONSTRAINT_CPUSET; |
| 213 | |
| 214 | if (!nodes_empty(nodes)) |
| 215 | return CONSTRAINT_MEMORY_POLICY; |
| 216 | #endif |
| 217 | |
| 218 | return CONSTRAINT_NONE; |
| 219 | } |
| 220 | |
| 221 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 222 | * Simple selection loop. We chose the process with the highest |
| 223 | * number of 'points'. We expect the caller will lock the tasklist. |
| 224 | * |
| 225 | * (not docbooked, we don't want this one cluttering up the manual) |
| 226 | */ |
Pavel Emelianov | c7ba5c9 | 2008-02-07 00:13:58 -0800 | [diff] [blame] | 227 | static struct task_struct *select_bad_process(unsigned long *ppoints, |
| 228 | struct mem_cgroup *mem) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 229 | { |
KOSAKI Motohiro | 495789a | 2009-09-21 17:03:14 -0700 | [diff] [blame] | 230 | struct task_struct *p; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 231 | struct task_struct *chosen = NULL; |
| 232 | struct timespec uptime; |
Kurt Garloff | 9827b78 | 2006-02-20 18:27:51 -0800 | [diff] [blame] | 233 | *ppoints = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 234 | |
| 235 | do_posix_clock_monotonic_gettime(&uptime); |
KOSAKI Motohiro | 495789a | 2009-09-21 17:03:14 -0700 | [diff] [blame] | 236 | for_each_process(p) { |
Paul Jackson | a49335c | 2005-09-06 15:18:09 -0700 | [diff] [blame] | 237 | unsigned long points; |
Paul Jackson | a49335c | 2005-09-06 15:18:09 -0700 | [diff] [blame] | 238 | |
Oleg Nesterov | 28324d1 | 2006-09-29 02:01:12 -0700 | [diff] [blame] | 239 | /* |
| 240 | * skip kernel threads and tasks which have already released |
| 241 | * their mm. |
| 242 | */ |
Nick Piggin | 5081dde | 2006-09-25 23:31:32 -0700 | [diff] [blame] | 243 | if (!p->mm) |
| 244 | continue; |
Oleg Nesterov | 28324d1 | 2006-09-29 02:01:12 -0700 | [diff] [blame] | 245 | /* skip the init task */ |
Serge E. Hallyn | b460cbc | 2007-10-18 23:39:52 -0700 | [diff] [blame] | 246 | if (is_global_init(p)) |
Paul Jackson | a49335c | 2005-09-06 15:18:09 -0700 | [diff] [blame] | 247 | continue; |
David Rientjes | 4c4a221 | 2008-02-07 00:14:06 -0800 | [diff] [blame] | 248 | if (mem && !task_in_mem_cgroup(p, mem)) |
| 249 | continue; |
Paul Jackson | ef08e3b | 2005-09-06 15:18:13 -0700 | [diff] [blame] | 250 | |
Paul Jackson | a49335c | 2005-09-06 15:18:09 -0700 | [diff] [blame] | 251 | /* |
Nick Piggin | b78483a | 2006-09-29 02:01:14 -0700 | [diff] [blame] | 252 | * This task already has access to memory reserves and is |
| 253 | * being killed. Don't allow any other task access to the |
| 254 | * memory reserve. |
| 255 | * |
| 256 | * Note: this may have a chance of deadlock if it gets |
| 257 | * blocked waiting for another task which itself is waiting |
| 258 | * for memory. Is there a better alternative? |
| 259 | */ |
| 260 | if (test_tsk_thread_flag(p, TIF_MEMDIE)) |
| 261 | return ERR_PTR(-1UL); |
| 262 | |
| 263 | /* |
Dave Peterson | 6937a25 | 2006-06-23 02:03:13 -0700 | [diff] [blame] | 264 | * This is in the process of releasing memory so wait for it |
Paul Jackson | a49335c | 2005-09-06 15:18:09 -0700 | [diff] [blame] | 265 | * to finish before killing some other task by mistake. |
Nick Piggin | 50ec3bb | 2006-09-25 23:31:29 -0700 | [diff] [blame] | 266 | * |
| 267 | * However, if p is the current task, we allow the 'kill' to |
| 268 | * go ahead if it is exiting: this will simply set TIF_MEMDIE, |
| 269 | * which will allow it to gain access to memory reserves in |
| 270 | * the process of exiting and releasing its resources. |
Nick Piggin | b78483a | 2006-09-29 02:01:14 -0700 | [diff] [blame] | 271 | * Otherwise we could get an easy OOM deadlock. |
Paul Jackson | a49335c | 2005-09-06 15:18:09 -0700 | [diff] [blame] | 272 | */ |
Nick Piggin | b78483a | 2006-09-29 02:01:14 -0700 | [diff] [blame] | 273 | if (p->flags & PF_EXITING) { |
| 274 | if (p != current) |
| 275 | return ERR_PTR(-1UL); |
| 276 | |
Oleg Nesterov | 972c4ea | 2006-09-29 02:01:12 -0700 | [diff] [blame] | 277 | chosen = p; |
| 278 | *ppoints = ULONG_MAX; |
Nick Piggin | 50ec3bb | 2006-09-25 23:31:29 -0700 | [diff] [blame] | 279 | } |
Oleg Nesterov | 972c4ea | 2006-09-29 02:01:12 -0700 | [diff] [blame] | 280 | |
KOSAKI Motohiro | 28b83c5 | 2009-09-21 17:03:13 -0700 | [diff] [blame] | 281 | if (p->signal->oom_adj == OOM_DISABLE) |
KOSAKI Motohiro | 0753ba0 | 2009-08-18 14:11:10 -0700 | [diff] [blame] | 282 | continue; |
| 283 | |
Li Zefan | 97d87c9 | 2008-04-28 02:13:35 -0700 | [diff] [blame] | 284 | points = badness(p, uptime.tv_sec); |
KOSAKI Motohiro | 0753ba0 | 2009-08-18 14:11:10 -0700 | [diff] [blame] | 285 | if (points > *ppoints || !chosen) { |
Paul Jackson | a49335c | 2005-09-06 15:18:09 -0700 | [diff] [blame] | 286 | chosen = p; |
Kurt Garloff | 9827b78 | 2006-02-20 18:27:51 -0800 | [diff] [blame] | 287 | *ppoints = points; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 288 | } |
KOSAKI Motohiro | 495789a | 2009-09-21 17:03:14 -0700 | [diff] [blame] | 289 | } |
Oleg Nesterov | 972c4ea | 2006-09-29 02:01:12 -0700 | [diff] [blame] | 290 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 291 | return chosen; |
| 292 | } |
| 293 | |
| 294 | /** |
Randy Dunlap | 1b578df | 2008-03-19 17:00:42 -0700 | [diff] [blame] | 295 | * dump_tasks - dump current memory state of all system tasks |
| 296 | * @mem: target memory controller |
| 297 | * |
David Rientjes | fef1bdd | 2008-02-07 00:14:07 -0800 | [diff] [blame] | 298 | * Dumps the current memory state of all system tasks, excluding kernel threads. |
| 299 | * State information includes task's pid, uid, tgid, vm size, rss, cpu, oom_adj |
| 300 | * score, and name. |
| 301 | * |
| 302 | * If the actual is non-NULL, only tasks that are a member of the mem_cgroup are |
| 303 | * shown. |
| 304 | * |
| 305 | * Call with tasklist_lock read-locked. |
| 306 | */ |
| 307 | static void dump_tasks(const struct mem_cgroup *mem) |
| 308 | { |
| 309 | struct task_struct *g, *p; |
| 310 | |
| 311 | printk(KERN_INFO "[ pid ] uid tgid total_vm rss cpu oom_adj " |
| 312 | "name\n"); |
| 313 | do_each_thread(g, p) { |
David Rientjes | 6d2661e | 2009-05-28 14:34:19 -0700 | [diff] [blame] | 314 | struct mm_struct *mm; |
| 315 | |
David Rientjes | fef1bdd | 2008-02-07 00:14:07 -0800 | [diff] [blame] | 316 | if (mem && !task_in_mem_cgroup(p, mem)) |
| 317 | continue; |
David Rientjes | b4416d2 | 2008-11-06 12:53:29 -0800 | [diff] [blame] | 318 | if (!thread_group_leader(p)) |
| 319 | continue; |
David Rientjes | fef1bdd | 2008-02-07 00:14:07 -0800 | [diff] [blame] | 320 | |
| 321 | task_lock(p); |
David Rientjes | 6d2661e | 2009-05-28 14:34:19 -0700 | [diff] [blame] | 322 | mm = p->mm; |
| 323 | if (!mm) { |
| 324 | /* |
| 325 | * total_vm and rss sizes do not exist for tasks with no |
| 326 | * mm so there's no need to report them; they can't be |
| 327 | * oom killed anyway. |
| 328 | */ |
| 329 | task_unlock(p); |
| 330 | continue; |
| 331 | } |
David Rientjes | fef1bdd | 2008-02-07 00:14:07 -0800 | [diff] [blame] | 332 | printk(KERN_INFO "[%5d] %5d %5d %8lu %8lu %3d %3d %s\n", |
David Rientjes | 6d2661e | 2009-05-28 14:34:19 -0700 | [diff] [blame] | 333 | p->pid, __task_cred(p)->uid, p->tgid, mm->total_vm, |
KOSAKI Motohiro | 28b83c5 | 2009-09-21 17:03:13 -0700 | [diff] [blame] | 334 | get_mm_rss(mm), (int)task_cpu(p), p->signal->oom_adj, |
KOSAKI Motohiro | 0753ba0 | 2009-08-18 14:11:10 -0700 | [diff] [blame] | 335 | p->comm); |
David Rientjes | fef1bdd | 2008-02-07 00:14:07 -0800 | [diff] [blame] | 336 | task_unlock(p); |
| 337 | } while_each_thread(g, p); |
| 338 | } |
| 339 | |
Randy Dunlap | 1b578df | 2008-03-19 17:00:42 -0700 | [diff] [blame] | 340 | /* |
Ram Gupta | 5a291b98 | 2006-09-25 23:31:54 -0700 | [diff] [blame] | 341 | * Send SIGKILL to the selected process irrespective of CAP_SYS_RAW_IO |
| 342 | * flag though it's unlikely that we select a process with CAP_SYS_RAW_IO |
| 343 | * set. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 344 | */ |
Nick Piggin | f3af38d | 2006-12-06 20:31:51 -0800 | [diff] [blame] | 345 | static void __oom_kill_task(struct task_struct *p, int verbose) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 346 | { |
Serge E. Hallyn | b460cbc | 2007-10-18 23:39:52 -0700 | [diff] [blame] | 347 | if (is_global_init(p)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 348 | WARN_ON(1); |
| 349 | printk(KERN_WARNING "tried to kill init!\n"); |
| 350 | return; |
| 351 | } |
| 352 | |
KOSAKI Motohiro | 0753ba0 | 2009-08-18 14:11:10 -0700 | [diff] [blame] | 353 | if (!p->mm) { |
| 354 | WARN_ON(1); |
| 355 | printk(KERN_WARNING "tried to kill an mm-less task!\n"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 356 | return; |
KOSAKI Motohiro | 0753ba0 | 2009-08-18 14:11:10 -0700 | [diff] [blame] | 357 | } |
Nick Piggin | 50ec3bb | 2006-09-25 23:31:29 -0700 | [diff] [blame] | 358 | |
Nick Piggin | f3af38d | 2006-12-06 20:31:51 -0800 | [diff] [blame] | 359 | if (verbose) |
Pavel Emelyanov | ba25f9d | 2007-10-18 23:40:40 -0700 | [diff] [blame] | 360 | printk(KERN_ERR "Killed process %d (%s)\n", |
| 361 | task_pid_nr(p), p->comm); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 362 | |
| 363 | /* |
| 364 | * We give our sacrificial lamb high priority and access to |
| 365 | * all the memory it needs. That way it should be able to |
| 366 | * exit() and clear out its resources quickly... |
| 367 | */ |
Peter Zijlstra | fa71706 | 2008-01-25 21:08:27 +0100 | [diff] [blame] | 368 | p->rt.time_slice = HZ; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 369 | set_tsk_thread_flag(p, TIF_MEMDIE); |
| 370 | |
| 371 | force_sig(SIGKILL, p); |
| 372 | } |
| 373 | |
Nick Piggin | f3af38d | 2006-12-06 20:31:51 -0800 | [diff] [blame] | 374 | static int oom_kill_task(struct task_struct *p) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 375 | { |
KOSAKI Motohiro | 0753ba0 | 2009-08-18 14:11:10 -0700 | [diff] [blame] | 376 | /* WARNING: mm may not be dereferenced since we did not obtain its |
| 377 | * value from get_task_mm(p). This is OK since all we need to do is |
| 378 | * compare mm to q->mm below. |
| 379 | * |
| 380 | * Furthermore, even if mm contains a non-NULL value, p->mm may |
| 381 | * change to NULL at any time since we do not hold task_lock(p). |
| 382 | * However, this is of no concern to us. |
| 383 | */ |
KOSAKI Motohiro | 8c5cd6f | 2009-09-21 17:03:15 -0700 | [diff] [blame^] | 384 | if (!p->mm || p->signal->oom_adj == OOM_DISABLE) |
Dave Peterson | 0131592 | 2006-04-18 22:20:44 -0700 | [diff] [blame] | 385 | return 1; |
KOSAKI Motohiro | 0753ba0 | 2009-08-18 14:11:10 -0700 | [diff] [blame] | 386 | |
Nick Piggin | f3af38d | 2006-12-06 20:31:51 -0800 | [diff] [blame] | 387 | __oom_kill_task(p, 1); |
Nick Piggin | c33e0fc | 2006-12-06 20:31:50 -0800 | [diff] [blame] | 388 | |
Dave Peterson | 0131592 | 2006-04-18 22:20:44 -0700 | [diff] [blame] | 389 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 390 | } |
| 391 | |
David Rientjes | 7213f50 | 2007-10-16 23:25:57 -0700 | [diff] [blame] | 392 | static int oom_kill_process(struct task_struct *p, gfp_t gfp_mask, int order, |
David Rientjes | fef1bdd | 2008-02-07 00:14:07 -0800 | [diff] [blame] | 393 | unsigned long points, struct mem_cgroup *mem, |
| 394 | const char *message) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 395 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 396 | struct task_struct *c; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 397 | |
David Rientjes | 7213f50 | 2007-10-16 23:25:57 -0700 | [diff] [blame] | 398 | if (printk_ratelimit()) { |
David Rientjes | 2ff05b2 | 2009-06-16 15:32:56 -0700 | [diff] [blame] | 399 | printk(KERN_WARNING "%s invoked oom-killer: " |
KOSAKI Motohiro | 28b83c5 | 2009-09-21 17:03:13 -0700 | [diff] [blame] | 400 | "gfp_mask=0x%x, order=%d, oom_adj=%d\n", |
| 401 | current->comm, gfp_mask, order, |
| 402 | current->signal->oom_adj); |
KOSAKI Motohiro | 0753ba0 | 2009-08-18 14:11:10 -0700 | [diff] [blame] | 403 | task_lock(current); |
David Rientjes | 75aa199 | 2009-01-06 14:39:01 -0800 | [diff] [blame] | 404 | cpuset_print_task_mems_allowed(current); |
| 405 | task_unlock(current); |
David Rientjes | 7213f50 | 2007-10-16 23:25:57 -0700 | [diff] [blame] | 406 | dump_stack(); |
Balbir Singh | e222432 | 2009-04-02 16:57:39 -0700 | [diff] [blame] | 407 | mem_cgroup_print_oom_info(mem, current); |
David Rientjes | 7213f50 | 2007-10-16 23:25:57 -0700 | [diff] [blame] | 408 | show_mem(); |
David Rientjes | fef1bdd | 2008-02-07 00:14:07 -0800 | [diff] [blame] | 409 | if (sysctl_oom_dump_tasks) |
| 410 | dump_tasks(mem); |
David Rientjes | 7213f50 | 2007-10-16 23:25:57 -0700 | [diff] [blame] | 411 | } |
| 412 | |
Nick Piggin | 50ec3bb | 2006-09-25 23:31:29 -0700 | [diff] [blame] | 413 | /* |
| 414 | * If the task is already exiting, don't alarm the sysadmin or kill |
| 415 | * its children or threads, just set TIF_MEMDIE so it can die quickly |
| 416 | */ |
KOSAKI Motohiro | 0753ba0 | 2009-08-18 14:11:10 -0700 | [diff] [blame] | 417 | if (p->flags & PF_EXITING) { |
Nick Piggin | f3af38d | 2006-12-06 20:31:51 -0800 | [diff] [blame] | 418 | __oom_kill_task(p, 0); |
Nick Piggin | 50ec3bb | 2006-09-25 23:31:29 -0700 | [diff] [blame] | 419 | return 0; |
| 420 | } |
| 421 | |
Nick Piggin | f3af38d | 2006-12-06 20:31:51 -0800 | [diff] [blame] | 422 | printk(KERN_ERR "%s: kill process %d (%s) score %li or a child\n", |
Pavel Emelyanov | ba25f9d | 2007-10-18 23:40:40 -0700 | [diff] [blame] | 423 | message, task_pid_nr(p), p->comm, points); |
Nick Piggin | f3af38d | 2006-12-06 20:31:51 -0800 | [diff] [blame] | 424 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 425 | /* Try to kill a child first */ |
Matthias Kaehlcke | 7b1915a | 2007-10-18 23:39:58 -0700 | [diff] [blame] | 426 | list_for_each_entry(c, &p->children, sibling) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 427 | if (c->mm == p->mm) |
| 428 | continue; |
Nick Piggin | f3af38d | 2006-12-06 20:31:51 -0800 | [diff] [blame] | 429 | if (!oom_kill_task(c)) |
Dave Peterson | 0131592 | 2006-04-18 22:20:44 -0700 | [diff] [blame] | 430 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 431 | } |
Nick Piggin | f3af38d | 2006-12-06 20:31:51 -0800 | [diff] [blame] | 432 | return oom_kill_task(p); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 433 | } |
| 434 | |
Balbir Singh | 00f0b82 | 2008-03-04 14:28:39 -0800 | [diff] [blame] | 435 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR |
Pavel Emelianov | c7ba5c9 | 2008-02-07 00:13:58 -0800 | [diff] [blame] | 436 | void mem_cgroup_out_of_memory(struct mem_cgroup *mem, gfp_t gfp_mask) |
| 437 | { |
| 438 | unsigned long points = 0; |
| 439 | struct task_struct *p; |
| 440 | |
Li Zefan | e115f2d | 2008-04-15 14:34:37 -0700 | [diff] [blame] | 441 | read_lock(&tasklist_lock); |
Pavel Emelianov | c7ba5c9 | 2008-02-07 00:13:58 -0800 | [diff] [blame] | 442 | retry: |
| 443 | p = select_bad_process(&points, mem); |
| 444 | if (PTR_ERR(p) == -1UL) |
| 445 | goto out; |
| 446 | |
| 447 | if (!p) |
| 448 | p = current; |
| 449 | |
David Rientjes | fef1bdd | 2008-02-07 00:14:07 -0800 | [diff] [blame] | 450 | if (oom_kill_process(p, gfp_mask, 0, points, mem, |
Pavel Emelianov | c7ba5c9 | 2008-02-07 00:13:58 -0800 | [diff] [blame] | 451 | "Memory cgroup out of memory")) |
| 452 | goto retry; |
| 453 | out: |
Li Zefan | e115f2d | 2008-04-15 14:34:37 -0700 | [diff] [blame] | 454 | read_unlock(&tasklist_lock); |
Pavel Emelianov | c7ba5c9 | 2008-02-07 00:13:58 -0800 | [diff] [blame] | 455 | } |
| 456 | #endif |
| 457 | |
Martin Schwidefsky | 8bc719d | 2006-09-25 23:31:20 -0700 | [diff] [blame] | 458 | static BLOCKING_NOTIFIER_HEAD(oom_notify_list); |
| 459 | |
| 460 | int register_oom_notifier(struct notifier_block *nb) |
| 461 | { |
| 462 | return blocking_notifier_chain_register(&oom_notify_list, nb); |
| 463 | } |
| 464 | EXPORT_SYMBOL_GPL(register_oom_notifier); |
| 465 | |
| 466 | int unregister_oom_notifier(struct notifier_block *nb) |
| 467 | { |
| 468 | return blocking_notifier_chain_unregister(&oom_notify_list, nb); |
| 469 | } |
| 470 | EXPORT_SYMBOL_GPL(unregister_oom_notifier); |
| 471 | |
David Rientjes | 098d7f1 | 2007-10-16 23:25:55 -0700 | [diff] [blame] | 472 | /* |
| 473 | * Try to acquire the OOM killer lock for the zones in zonelist. Returns zero |
| 474 | * if a parallel OOM killing is already taking place that includes a zone in |
| 475 | * the zonelist. Otherwise, locks all zones in the zonelist and returns 1. |
| 476 | */ |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 477 | int try_set_zone_oom(struct zonelist *zonelist, gfp_t gfp_mask) |
David Rientjes | 098d7f1 | 2007-10-16 23:25:55 -0700 | [diff] [blame] | 478 | { |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 479 | struct zoneref *z; |
| 480 | struct zone *zone; |
David Rientjes | 098d7f1 | 2007-10-16 23:25:55 -0700 | [diff] [blame] | 481 | int ret = 1; |
| 482 | |
David Rientjes | c7d4cae | 2009-01-06 14:39:00 -0800 | [diff] [blame] | 483 | spin_lock(&zone_scan_lock); |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 484 | for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) { |
| 485 | if (zone_is_oom_locked(zone)) { |
David Rientjes | 098d7f1 | 2007-10-16 23:25:55 -0700 | [diff] [blame] | 486 | ret = 0; |
| 487 | goto out; |
| 488 | } |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 489 | } |
David Rientjes | 098d7f1 | 2007-10-16 23:25:55 -0700 | [diff] [blame] | 490 | |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 491 | for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) { |
| 492 | /* |
David Rientjes | c7d4cae | 2009-01-06 14:39:00 -0800 | [diff] [blame] | 493 | * Lock each zone in the zonelist under zone_scan_lock so a |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 494 | * parallel invocation of try_set_zone_oom() doesn't succeed |
| 495 | * when it shouldn't. |
| 496 | */ |
| 497 | zone_set_flag(zone, ZONE_OOM_LOCKED); |
| 498 | } |
| 499 | |
David Rientjes | 098d7f1 | 2007-10-16 23:25:55 -0700 | [diff] [blame] | 500 | out: |
David Rientjes | c7d4cae | 2009-01-06 14:39:00 -0800 | [diff] [blame] | 501 | spin_unlock(&zone_scan_lock); |
David Rientjes | 098d7f1 | 2007-10-16 23:25:55 -0700 | [diff] [blame] | 502 | return ret; |
| 503 | } |
| 504 | |
| 505 | /* |
| 506 | * Clears the ZONE_OOM_LOCKED flag for all zones in the zonelist so that failed |
| 507 | * allocation attempts with zonelists containing them may now recall the OOM |
| 508 | * killer, if necessary. |
| 509 | */ |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 510 | void clear_zonelist_oom(struct zonelist *zonelist, gfp_t gfp_mask) |
David Rientjes | 098d7f1 | 2007-10-16 23:25:55 -0700 | [diff] [blame] | 511 | { |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 512 | struct zoneref *z; |
| 513 | struct zone *zone; |
David Rientjes | 098d7f1 | 2007-10-16 23:25:55 -0700 | [diff] [blame] | 514 | |
David Rientjes | c7d4cae | 2009-01-06 14:39:00 -0800 | [diff] [blame] | 515 | spin_lock(&zone_scan_lock); |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 516 | for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) { |
| 517 | zone_clear_flag(zone, ZONE_OOM_LOCKED); |
| 518 | } |
David Rientjes | c7d4cae | 2009-01-06 14:39:00 -0800 | [diff] [blame] | 519 | spin_unlock(&zone_scan_lock); |
David Rientjes | 098d7f1 | 2007-10-16 23:25:55 -0700 | [diff] [blame] | 520 | } |
| 521 | |
Nick Piggin | 1c0fe6e | 2009-01-06 14:38:59 -0800 | [diff] [blame] | 522 | /* |
| 523 | * Must be called with tasklist_lock held for read. |
| 524 | */ |
| 525 | static void __out_of_memory(gfp_t gfp_mask, int order) |
| 526 | { |
David Rientjes | 184101b | 2009-05-06 16:02:55 -0700 | [diff] [blame] | 527 | struct task_struct *p; |
| 528 | unsigned long points; |
Nick Piggin | 1c0fe6e | 2009-01-06 14:38:59 -0800 | [diff] [blame] | 529 | |
David Rientjes | 184101b | 2009-05-06 16:02:55 -0700 | [diff] [blame] | 530 | if (sysctl_oom_kill_allocating_task) |
| 531 | if (!oom_kill_process(current, gfp_mask, order, 0, NULL, |
| 532 | "Out of memory (oom_kill_allocating_task)")) |
Nick Piggin | 1c0fe6e | 2009-01-06 14:38:59 -0800 | [diff] [blame] | 533 | return; |
David Rientjes | 184101b | 2009-05-06 16:02:55 -0700 | [diff] [blame] | 534 | retry: |
| 535 | /* |
| 536 | * Rambo mode: Shoot down a process and hope it solves whatever |
| 537 | * issues we may have. |
| 538 | */ |
| 539 | p = select_bad_process(&points, NULL); |
Nick Piggin | 1c0fe6e | 2009-01-06 14:38:59 -0800 | [diff] [blame] | 540 | |
David Rientjes | 184101b | 2009-05-06 16:02:55 -0700 | [diff] [blame] | 541 | if (PTR_ERR(p) == -1UL) |
| 542 | return; |
Nick Piggin | 1c0fe6e | 2009-01-06 14:38:59 -0800 | [diff] [blame] | 543 | |
David Rientjes | 184101b | 2009-05-06 16:02:55 -0700 | [diff] [blame] | 544 | /* Found nothing?!?! Either we hang forever, or we panic. */ |
| 545 | if (!p) { |
| 546 | read_unlock(&tasklist_lock); |
| 547 | panic("Out of memory and no killable processes...\n"); |
Nick Piggin | 1c0fe6e | 2009-01-06 14:38:59 -0800 | [diff] [blame] | 548 | } |
David Rientjes | 184101b | 2009-05-06 16:02:55 -0700 | [diff] [blame] | 549 | |
| 550 | if (oom_kill_process(p, gfp_mask, order, points, NULL, |
| 551 | "Out of memory")) |
| 552 | goto retry; |
Nick Piggin | 1c0fe6e | 2009-01-06 14:38:59 -0800 | [diff] [blame] | 553 | } |
| 554 | |
| 555 | /* |
| 556 | * pagefault handler calls into here because it is out of memory but |
| 557 | * doesn't know exactly how or why. |
| 558 | */ |
| 559 | void pagefault_out_of_memory(void) |
| 560 | { |
| 561 | unsigned long freed = 0; |
| 562 | |
| 563 | blocking_notifier_call_chain(&oom_notify_list, 0, &freed); |
| 564 | if (freed > 0) |
| 565 | /* Got some memory back in the last second. */ |
| 566 | return; |
| 567 | |
KAMEZAWA Hiroyuki | a636b32 | 2009-01-07 18:08:08 -0800 | [diff] [blame] | 568 | /* |
| 569 | * If this is from memcg, oom-killer is already invoked. |
| 570 | * and not worth to go system-wide-oom. |
| 571 | */ |
| 572 | if (mem_cgroup_oom_called(current)) |
| 573 | goto rest_and_return; |
| 574 | |
Nick Piggin | 1c0fe6e | 2009-01-06 14:38:59 -0800 | [diff] [blame] | 575 | if (sysctl_panic_on_oom) |
| 576 | panic("out of memory from page fault. panic_on_oom is selected.\n"); |
| 577 | |
| 578 | read_lock(&tasklist_lock); |
| 579 | __out_of_memory(0, 0); /* unknown gfp_mask and order */ |
| 580 | read_unlock(&tasklist_lock); |
| 581 | |
| 582 | /* |
| 583 | * Give "p" a good chance of killing itself before we |
| 584 | * retry to allocate memory. |
| 585 | */ |
KAMEZAWA Hiroyuki | a636b32 | 2009-01-07 18:08:08 -0800 | [diff] [blame] | 586 | rest_and_return: |
Nick Piggin | 1c0fe6e | 2009-01-06 14:38:59 -0800 | [diff] [blame] | 587 | if (!test_thread_flag(TIF_MEMDIE)) |
| 588 | schedule_timeout_uninterruptible(1); |
| 589 | } |
| 590 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 591 | /** |
Dave Peterson | 6937a25 | 2006-06-23 02:03:13 -0700 | [diff] [blame] | 592 | * out_of_memory - kill the "best" process when we run out of memory |
Randy Dunlap | 1b578df | 2008-03-19 17:00:42 -0700 | [diff] [blame] | 593 | * @zonelist: zonelist pointer |
| 594 | * @gfp_mask: memory allocation flags |
| 595 | * @order: amount of memory being requested as a power of 2 |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 596 | * |
| 597 | * If we run out of memory, we have the choice between either |
| 598 | * killing a random task (bad), letting the system crash (worse) |
| 599 | * OR try to be smart about which process to kill. Note that we |
| 600 | * don't have to be perfect here, we just have to be good. |
| 601 | */ |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 602 | void out_of_memory(struct zonelist *zonelist, gfp_t gfp_mask, int order) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 603 | { |
Martin Schwidefsky | 8bc719d | 2006-09-25 23:31:20 -0700 | [diff] [blame] | 604 | unsigned long freed = 0; |
David Rientjes | 70e24bd | 2007-10-16 23:25:53 -0700 | [diff] [blame] | 605 | enum oom_constraint constraint; |
Martin Schwidefsky | 8bc719d | 2006-09-25 23:31:20 -0700 | [diff] [blame] | 606 | |
| 607 | blocking_notifier_call_chain(&oom_notify_list, 0, &freed); |
| 608 | if (freed > 0) |
| 609 | /* Got some memory back in the last second. */ |
| 610 | return; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 611 | |
Yasunori Goto | 2b744c0 | 2007-05-06 14:49:59 -0700 | [diff] [blame] | 612 | if (sysctl_panic_on_oom == 2) |
| 613 | panic("out of memory. Compulsory panic_on_oom is selected.\n"); |
| 614 | |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 615 | /* |
| 616 | * Check if there were limitations on the allocation (only relevant for |
| 617 | * NUMA) that may require different handling. |
| 618 | */ |
David Rientjes | 2b45ab3 | 2007-05-06 14:50:00 -0700 | [diff] [blame] | 619 | constraint = constrained_alloc(zonelist, gfp_mask); |
David Rientjes | 2b45ab3 | 2007-05-06 14:50:00 -0700 | [diff] [blame] | 620 | read_lock(&tasklist_lock); |
| 621 | |
| 622 | switch (constraint) { |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 623 | case CONSTRAINT_MEMORY_POLICY: |
Nick Piggin | 1c0fe6e | 2009-01-06 14:38:59 -0800 | [diff] [blame] | 624 | oom_kill_process(current, gfp_mask, order, 0, NULL, |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 625 | "No available memory (MPOL_BIND)"); |
| 626 | break; |
| 627 | |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 628 | case CONSTRAINT_NONE: |
KAMEZAWA Hiroyuki | fadd8fb | 2006-06-23 02:03:13 -0700 | [diff] [blame] | 629 | if (sysctl_panic_on_oom) |
| 630 | panic("out of memory. panic_on_oom is selected\n"); |
David Rientjes | fe071d7 | 2007-10-16 23:25:56 -0700 | [diff] [blame] | 631 | /* Fall-through */ |
| 632 | case CONSTRAINT_CPUSET: |
Nick Piggin | 1c0fe6e | 2009-01-06 14:38:59 -0800 | [diff] [blame] | 633 | __out_of_memory(gfp_mask, order); |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 634 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 635 | } |
| 636 | |
Andrew Morton | 140ffce | 2006-03-02 02:54:28 -0800 | [diff] [blame] | 637 | read_unlock(&tasklist_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 638 | |
| 639 | /* |
| 640 | * Give "p" a good chance of killing itself before we |
Kirill Korotaev | 2f659f4 | 2006-01-08 01:01:05 -0800 | [diff] [blame] | 641 | * retry to allocate memory unless "p" is current |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 642 | */ |
Kirill Korotaev | 2f659f4 | 2006-01-08 01:01:05 -0800 | [diff] [blame] | 643 | if (!test_thread_flag(TIF_MEMDIE)) |
Andrew Morton | 140ffce | 2006-03-02 02:54:28 -0800 | [diff] [blame] | 644 | schedule_timeout_uninterruptible(1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 645 | } |