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> |
Tejun Heo | 5a0e3ad | 2010-03-24 17:04:11 +0900 | [diff] [blame] | 21 | #include <linux/gfp.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 22 | #include <linux/sched.h> |
| 23 | #include <linux/swap.h> |
| 24 | #include <linux/timex.h> |
| 25 | #include <linux/jiffies.h> |
Paul Jackson | ef08e3b | 2005-09-06 15:18:13 -0700 | [diff] [blame] | 26 | #include <linux/cpuset.h> |
Martin Schwidefsky | 8bc719d | 2006-09-25 23:31:20 -0700 | [diff] [blame] | 27 | #include <linux/module.h> |
| 28 | #include <linux/notifier.h> |
Pavel Emelianov | c7ba5c9 | 2008-02-07 00:13:58 -0800 | [diff] [blame] | 29 | #include <linux/memcontrol.h> |
David Rientjes | 6f48d0eb | 2010-08-09 17:18:52 -0700 | [diff] [blame] | 30 | #include <linux/mempolicy.h> |
David Howells | 5cd9c58 | 2008-08-14 11:37:28 +0100 | [diff] [blame] | 31 | #include <linux/security.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 32 | |
KAMEZAWA Hiroyuki | fadd8fb | 2006-06-23 02:03:13 -0700 | [diff] [blame] | 33 | int sysctl_panic_on_oom; |
David Rientjes | fe071d7 | 2007-10-16 23:25:56 -0700 | [diff] [blame] | 34 | int sysctl_oom_kill_allocating_task; |
David Rientjes | ad915c4 | 2010-08-09 17:18:53 -0700 | [diff] [blame] | 35 | int sysctl_oom_dump_tasks = 1; |
David Rientjes | c7d4cae | 2009-01-06 14:39:00 -0800 | [diff] [blame] | 36 | static DEFINE_SPINLOCK(zone_scan_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 37 | /* #define DEBUG */ |
| 38 | |
David Rientjes | 6f48d0eb | 2010-08-09 17:18:52 -0700 | [diff] [blame] | 39 | #ifdef CONFIG_NUMA |
| 40 | /** |
| 41 | * has_intersects_mems_allowed() - check task eligiblity for kill |
| 42 | * @tsk: task struct of which task to consider |
| 43 | * @mask: nodemask passed to page allocator for mempolicy ooms |
| 44 | * |
| 45 | * Task eligibility is determined by whether or not a candidate task, @tsk, |
| 46 | * shares the same mempolicy nodes as current if it is bound by such a policy |
| 47 | * and whether or not it has the same set of allowed cpuset nodes. |
KOSAKI Motohiro | 495789a | 2009-09-21 17:03:14 -0700 | [diff] [blame] | 48 | */ |
David Rientjes | 6f48d0eb | 2010-08-09 17:18:52 -0700 | [diff] [blame] | 49 | static bool has_intersects_mems_allowed(struct task_struct *tsk, |
| 50 | const nodemask_t *mask) |
KOSAKI Motohiro | 495789a | 2009-09-21 17:03:14 -0700 | [diff] [blame] | 51 | { |
David Rientjes | 6f48d0eb | 2010-08-09 17:18:52 -0700 | [diff] [blame] | 52 | struct task_struct *start = tsk; |
KOSAKI Motohiro | 495789a | 2009-09-21 17:03:14 -0700 | [diff] [blame] | 53 | |
KOSAKI Motohiro | 495789a | 2009-09-21 17:03:14 -0700 | [diff] [blame] | 54 | do { |
David Rientjes | 6f48d0eb | 2010-08-09 17:18:52 -0700 | [diff] [blame] | 55 | if (mask) { |
| 56 | /* |
| 57 | * If this is a mempolicy constrained oom, tsk's |
| 58 | * cpuset is irrelevant. Only return true if its |
| 59 | * mempolicy intersects current, otherwise it may be |
| 60 | * needlessly killed. |
| 61 | */ |
| 62 | if (mempolicy_nodemask_intersects(tsk, mask)) |
| 63 | return true; |
| 64 | } else { |
| 65 | /* |
| 66 | * This is not a mempolicy constrained oom, so only |
| 67 | * check the mems of tsk's cpuset. |
| 68 | */ |
| 69 | if (cpuset_mems_allowed_intersects(current, tsk)) |
| 70 | return true; |
| 71 | } |
| 72 | tsk = next_thread(tsk); |
| 73 | } while (tsk != start); |
| 74 | return false; |
KOSAKI Motohiro | 495789a | 2009-09-21 17:03:14 -0700 | [diff] [blame] | 75 | } |
David Rientjes | 6f48d0eb | 2010-08-09 17:18:52 -0700 | [diff] [blame] | 76 | #else |
| 77 | static bool has_intersects_mems_allowed(struct task_struct *tsk, |
| 78 | const nodemask_t *mask) |
| 79 | { |
| 80 | return true; |
| 81 | } |
| 82 | #endif /* CONFIG_NUMA */ |
KOSAKI Motohiro | 495789a | 2009-09-21 17:03:14 -0700 | [diff] [blame] | 83 | |
David Rientjes | 6f48d0eb | 2010-08-09 17:18:52 -0700 | [diff] [blame] | 84 | /* |
| 85 | * The process p may have detached its own ->mm while exiting or through |
| 86 | * use_mm(), but one or more of its subthreads may still have a valid |
| 87 | * pointer. Return p, or any of its subthreads with a valid ->mm, with |
| 88 | * task_lock() held. |
| 89 | */ |
Oleg Nesterov | dd8e8f4 | 2010-08-09 17:18:45 -0700 | [diff] [blame] | 90 | static struct task_struct *find_lock_task_mm(struct task_struct *p) |
| 91 | { |
| 92 | struct task_struct *t = p; |
| 93 | |
| 94 | do { |
| 95 | task_lock(t); |
| 96 | if (likely(t->mm)) |
| 97 | return t; |
| 98 | task_unlock(t); |
| 99 | } while_each_thread(p, t); |
| 100 | |
| 101 | return NULL; |
| 102 | } |
| 103 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 104 | /** |
Dave Peterson | 6937a25 | 2006-06-23 02:03:13 -0700 | [diff] [blame] | 105 | * badness - calculate a numeric value for how bad this task has been |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 106 | * @p: task struct of which task we should calculate |
Paul Jackson | a49335c | 2005-09-06 15:18:09 -0700 | [diff] [blame] | 107 | * @uptime: current uptime in seconds |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 108 | * |
| 109 | * The formula used is relatively simple and documented inline in the |
| 110 | * function. The main rationale is that we want to select a good task |
| 111 | * to kill when we run out of memory. |
| 112 | * |
| 113 | * Good in this context means that: |
| 114 | * 1) we lose the minimum amount of work done |
| 115 | * 2) we recover a large amount of memory |
| 116 | * 3) we don't kill anything innocent of eating tons of memory |
| 117 | * 4) we want to kill the minimum amount of processes (one) |
| 118 | * 5) we try to kill the process the user expects us to kill, this |
| 119 | * algorithm has been meticulously tuned to meet the principle |
| 120 | * of least surprise ... (be careful when you change it) |
| 121 | */ |
| 122 | |
Li Zefan | 97d87c9 | 2008-04-28 02:13:35 -0700 | [diff] [blame] | 123 | unsigned long badness(struct task_struct *p, unsigned long uptime) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 124 | { |
Cyrill Gorcunov | a12888f | 2009-03-31 15:19:27 -0700 | [diff] [blame] | 125 | unsigned long points, cpu_time, run_time; |
Andrew Morton | 97c2c9b | 2006-04-18 22:20:38 -0700 | [diff] [blame] | 126 | struct task_struct *child; |
Oleg Nesterov | dd8e8f4 | 2010-08-09 17:18:45 -0700 | [diff] [blame] | 127 | struct task_struct *c, *t; |
KOSAKI Motohiro | 28b83c5 | 2009-09-21 17:03:13 -0700 | [diff] [blame] | 128 | int oom_adj = p->signal->oom_adj; |
KOSAKI Motohiro | 495789a | 2009-09-21 17:03:14 -0700 | [diff] [blame] | 129 | struct task_cputime task_time; |
| 130 | unsigned long utime; |
| 131 | unsigned long stime; |
KOSAKI Motohiro | 28b83c5 | 2009-09-21 17:03:13 -0700 | [diff] [blame] | 132 | |
| 133 | if (oom_adj == OOM_DISABLE) |
| 134 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 135 | |
Oleg Nesterov | dd8e8f4 | 2010-08-09 17:18:45 -0700 | [diff] [blame] | 136 | p = find_lock_task_mm(p); |
| 137 | if (!p) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 138 | return 0; |
| 139 | |
| 140 | /* |
| 141 | * The memory size of the process is the basis for the badness. |
| 142 | */ |
Oleg Nesterov | dd8e8f4 | 2010-08-09 17:18:45 -0700 | [diff] [blame] | 143 | points = p->mm->total_vm; |
Andrew Morton | 97c2c9b | 2006-04-18 22:20:38 -0700 | [diff] [blame] | 144 | task_unlock(p); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 145 | |
| 146 | /* |
Hugh Dickins | 7ba3485 | 2007-01-05 16:37:03 -0800 | [diff] [blame] | 147 | * swapoff can easily use up all memory, so kill those first. |
| 148 | */ |
Hugh Dickins | 35451be | 2009-09-21 17:02:27 -0700 | [diff] [blame] | 149 | if (p->flags & PF_OOM_ORIGIN) |
Hugh Dickins | 7ba3485 | 2007-01-05 16:37:03 -0800 | [diff] [blame] | 150 | return ULONG_MAX; |
| 151 | |
| 152 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 153 | * Processes which fork a lot of child processes are likely |
Kurt Garloff | 9827b78 | 2006-02-20 18:27:51 -0800 | [diff] [blame] | 154 | * 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] | 155 | * have an own mm. This prevents forking servers to flood the |
Kurt Garloff | 9827b78 | 2006-02-20 18:27:51 -0800 | [diff] [blame] | 156 | * machine with an endless amount of children. In case a single |
| 157 | * child is eating the vast majority of memory, adding only half |
| 158 | * to the parents will make the child our kill candidate of choice. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 159 | */ |
Oleg Nesterov | dd8e8f4 | 2010-08-09 17:18:45 -0700 | [diff] [blame] | 160 | t = p; |
| 161 | do { |
| 162 | list_for_each_entry(c, &t->children, sibling) { |
| 163 | child = find_lock_task_mm(c); |
| 164 | if (child) { |
| 165 | if (child->mm != p->mm) |
| 166 | points += child->mm->total_vm/2 + 1; |
| 167 | task_unlock(child); |
| 168 | } |
| 169 | } |
| 170 | } while_each_thread(p, t); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 171 | |
| 172 | /* |
| 173 | * CPU time is in tens of seconds and run time is in thousands |
| 174 | * of seconds. There is no particular reason for this other than |
| 175 | * that it turned out to work very well in practice. |
| 176 | */ |
KOSAKI Motohiro | 495789a | 2009-09-21 17:03:14 -0700 | [diff] [blame] | 177 | thread_group_cputime(p, &task_time); |
| 178 | utime = cputime_to_jiffies(task_time.utime); |
| 179 | stime = cputime_to_jiffies(task_time.stime); |
| 180 | cpu_time = (utime + stime) >> (SHIFT_HZ + 3); |
| 181 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 182 | |
| 183 | if (uptime >= p->start_time.tv_sec) |
| 184 | run_time = (uptime - p->start_time.tv_sec) >> 10; |
| 185 | else |
| 186 | run_time = 0; |
| 187 | |
Cyrill Gorcunov | a12888f | 2009-03-31 15:19:27 -0700 | [diff] [blame] | 188 | if (cpu_time) |
| 189 | points /= int_sqrt(cpu_time); |
| 190 | if (run_time) |
| 191 | points /= int_sqrt(int_sqrt(run_time)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 192 | |
| 193 | /* |
| 194 | * Niced processes are most likely less important, so double |
| 195 | * their badness points. |
| 196 | */ |
| 197 | if (task_nice(p) > 0) |
| 198 | points *= 2; |
| 199 | |
| 200 | /* |
| 201 | * Superuser processes are usually more important, so we make it |
| 202 | * less likely that we kill those. |
| 203 | */ |
Eric Paris | a2f2945 | 2008-11-11 22:02:54 +1100 | [diff] [blame] | 204 | if (has_capability_noaudit(p, CAP_SYS_ADMIN) || |
| 205 | has_capability_noaudit(p, CAP_SYS_RESOURCE)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 206 | points /= 4; |
| 207 | |
| 208 | /* |
| 209 | * We don't want to kill a process with direct hardware access. |
| 210 | * Not only could that mess up the hardware, but usually users |
| 211 | * tend to only have this flag set on applications they think |
| 212 | * of as important. |
| 213 | */ |
Eric Paris | a2f2945 | 2008-11-11 22:02:54 +1100 | [diff] [blame] | 214 | if (has_capability_noaudit(p, CAP_SYS_RAWIO)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 215 | points /= 4; |
| 216 | |
| 217 | /* |
KOSAKI Motohiro | 28b83c5 | 2009-09-21 17:03:13 -0700 | [diff] [blame] | 218 | * Adjust the score by oom_adj. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 219 | */ |
KOSAKI Motohiro | 28b83c5 | 2009-09-21 17:03:13 -0700 | [diff] [blame] | 220 | if (oom_adj) { |
| 221 | if (oom_adj > 0) { |
Joshua N Pritikin | 9a82782 | 2007-05-06 14:49:07 -0700 | [diff] [blame] | 222 | if (!points) |
| 223 | points = 1; |
KOSAKI Motohiro | 28b83c5 | 2009-09-21 17:03:13 -0700 | [diff] [blame] | 224 | points <<= oom_adj; |
Joshua N Pritikin | 9a82782 | 2007-05-06 14:49:07 -0700 | [diff] [blame] | 225 | } else |
KOSAKI Motohiro | 28b83c5 | 2009-09-21 17:03:13 -0700 | [diff] [blame] | 226 | points >>= -(oom_adj); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 227 | } |
| 228 | |
| 229 | #ifdef DEBUG |
David Rientjes | a5e58a6 | 2007-07-31 00:37:32 -0700 | [diff] [blame] | 230 | printk(KERN_DEBUG "OOMkill: task %d (%s) got %lu points\n", |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 231 | p->pid, p->comm, points); |
| 232 | #endif |
| 233 | return points; |
| 234 | } |
| 235 | |
| 236 | /* |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 237 | * Determine the type of allocation constraint. |
| 238 | */ |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 239 | #ifdef CONFIG_NUMA |
KAMEZAWA Hiroyuki | 4365a56 | 2009-12-15 16:45:33 -0800 | [diff] [blame] | 240 | static enum oom_constraint constrained_alloc(struct zonelist *zonelist, |
| 241 | gfp_t gfp_mask, nodemask_t *nodemask) |
| 242 | { |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 243 | struct zone *zone; |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 244 | struct zoneref *z; |
Mel Gorman | 54a6eb5 | 2008-04-28 02:12:16 -0700 | [diff] [blame] | 245 | enum zone_type high_zoneidx = gfp_zone(gfp_mask); |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 246 | |
KAMEZAWA Hiroyuki | 4365a56 | 2009-12-15 16:45:33 -0800 | [diff] [blame] | 247 | /* |
| 248 | * Reach here only when __GFP_NOFAIL is used. So, we should avoid |
| 249 | * to kill current.We have to random task kill in this case. |
| 250 | * Hopefully, CONSTRAINT_THISNODE...but no way to handle it, now. |
| 251 | */ |
| 252 | if (gfp_mask & __GFP_THISNODE) |
| 253 | return CONSTRAINT_NONE; |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 254 | |
KAMEZAWA Hiroyuki | 4365a56 | 2009-12-15 16:45:33 -0800 | [diff] [blame] | 255 | /* |
| 256 | * The nodemask here is a nodemask passed to alloc_pages(). Now, |
| 257 | * cpuset doesn't use this nodemask for its hardwall/softwall/hierarchy |
| 258 | * feature. mempolicy is an only user of nodemask here. |
| 259 | * check mempolicy's nodemask contains all N_HIGH_MEMORY |
| 260 | */ |
| 261 | if (nodemask && !nodes_subset(node_states[N_HIGH_MEMORY], *nodemask)) |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 262 | return CONSTRAINT_MEMORY_POLICY; |
KAMEZAWA Hiroyuki | 4365a56 | 2009-12-15 16:45:33 -0800 | [diff] [blame] | 263 | |
| 264 | /* Check this allocation failure is caused by cpuset's wall function */ |
| 265 | for_each_zone_zonelist_nodemask(zone, z, zonelist, |
| 266 | high_zoneidx, nodemask) |
| 267 | if (!cpuset_zone_allowed_softwall(zone, gfp_mask)) |
| 268 | return CONSTRAINT_CPUSET; |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 269 | |
| 270 | return CONSTRAINT_NONE; |
| 271 | } |
KAMEZAWA Hiroyuki | 4365a56 | 2009-12-15 16:45:33 -0800 | [diff] [blame] | 272 | #else |
| 273 | static enum oom_constraint constrained_alloc(struct zonelist *zonelist, |
| 274 | gfp_t gfp_mask, nodemask_t *nodemask) |
| 275 | { |
| 276 | return CONSTRAINT_NONE; |
| 277 | } |
| 278 | #endif |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 279 | |
| 280 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 281 | * Simple selection loop. We chose the process with the highest |
| 282 | * number of 'points'. We expect the caller will lock the tasklist. |
| 283 | * |
| 284 | * (not docbooked, we don't want this one cluttering up the manual) |
| 285 | */ |
Pavel Emelianov | c7ba5c9 | 2008-02-07 00:13:58 -0800 | [diff] [blame] | 286 | static struct task_struct *select_bad_process(unsigned long *ppoints, |
David Rientjes | f442003 | 2010-08-09 17:18:58 -0700 | [diff] [blame^] | 287 | struct mem_cgroup *mem, const nodemask_t *nodemask) |
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 | struct task_struct *p; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 290 | struct task_struct *chosen = NULL; |
| 291 | struct timespec uptime; |
Kurt Garloff | 9827b78 | 2006-02-20 18:27:51 -0800 | [diff] [blame] | 292 | *ppoints = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 293 | |
| 294 | do_posix_clock_monotonic_gettime(&uptime); |
KOSAKI Motohiro | 495789a | 2009-09-21 17:03:14 -0700 | [diff] [blame] | 295 | for_each_process(p) { |
Paul Jackson | a49335c | 2005-09-06 15:18:09 -0700 | [diff] [blame] | 296 | unsigned long points; |
Paul Jackson | a49335c | 2005-09-06 15:18:09 -0700 | [diff] [blame] | 297 | |
Oleg Nesterov | 455c0e5 | 2010-08-09 17:18:43 -0700 | [diff] [blame] | 298 | /* skip the init task and kthreads */ |
| 299 | if (is_global_init(p) || (p->flags & PF_KTHREAD)) |
Paul Jackson | a49335c | 2005-09-06 15:18:09 -0700 | [diff] [blame] | 300 | continue; |
David Rientjes | 4c4a221 | 2008-02-07 00:14:06 -0800 | [diff] [blame] | 301 | if (mem && !task_in_mem_cgroup(p, mem)) |
| 302 | continue; |
David Rientjes | f442003 | 2010-08-09 17:18:58 -0700 | [diff] [blame^] | 303 | if (!has_intersects_mems_allowed(p, nodemask)) |
David Rientjes | 6cf86ac | 2010-08-09 17:18:50 -0700 | [diff] [blame] | 304 | continue; |
Paul Jackson | ef08e3b | 2005-09-06 15:18:13 -0700 | [diff] [blame] | 305 | |
Paul Jackson | a49335c | 2005-09-06 15:18:09 -0700 | [diff] [blame] | 306 | /* |
Nick Piggin | b78483a | 2006-09-29 02:01:14 -0700 | [diff] [blame] | 307 | * This task already has access to memory reserves and is |
| 308 | * being killed. Don't allow any other task access to the |
| 309 | * memory reserve. |
| 310 | * |
| 311 | * Note: this may have a chance of deadlock if it gets |
| 312 | * blocked waiting for another task which itself is waiting |
| 313 | * for memory. Is there a better alternative? |
| 314 | */ |
| 315 | if (test_tsk_thread_flag(p, TIF_MEMDIE)) |
| 316 | return ERR_PTR(-1UL); |
| 317 | |
| 318 | /* |
Dave Peterson | 6937a25 | 2006-06-23 02:03:13 -0700 | [diff] [blame] | 319 | * This is in the process of releasing memory so wait for it |
Paul Jackson | a49335c | 2005-09-06 15:18:09 -0700 | [diff] [blame] | 320 | * to finish before killing some other task by mistake. |
Nick Piggin | 50ec3bb | 2006-09-25 23:31:29 -0700 | [diff] [blame] | 321 | * |
| 322 | * However, if p is the current task, we allow the 'kill' to |
| 323 | * go ahead if it is exiting: this will simply set TIF_MEMDIE, |
| 324 | * which will allow it to gain access to memory reserves in |
| 325 | * the process of exiting and releasing its resources. |
Nick Piggin | b78483a | 2006-09-29 02:01:14 -0700 | [diff] [blame] | 326 | * Otherwise we could get an easy OOM deadlock. |
Paul Jackson | a49335c | 2005-09-06 15:18:09 -0700 | [diff] [blame] | 327 | */ |
Oleg Nesterov | b522794 | 2010-08-09 17:18:44 -0700 | [diff] [blame] | 328 | if ((p->flags & PF_EXITING) && p->mm) { |
Nick Piggin | b78483a | 2006-09-29 02:01:14 -0700 | [diff] [blame] | 329 | if (p != current) |
| 330 | return ERR_PTR(-1UL); |
| 331 | |
Oleg Nesterov | 972c4ea | 2006-09-29 02:01:12 -0700 | [diff] [blame] | 332 | chosen = p; |
| 333 | *ppoints = ULONG_MAX; |
Nick Piggin | 50ec3bb | 2006-09-25 23:31:29 -0700 | [diff] [blame] | 334 | } |
Oleg Nesterov | 972c4ea | 2006-09-29 02:01:12 -0700 | [diff] [blame] | 335 | |
KOSAKI Motohiro | 28b83c5 | 2009-09-21 17:03:13 -0700 | [diff] [blame] | 336 | if (p->signal->oom_adj == OOM_DISABLE) |
KOSAKI Motohiro | 0753ba0 | 2009-08-18 14:11:10 -0700 | [diff] [blame] | 337 | continue; |
| 338 | |
Li Zefan | 97d87c9 | 2008-04-28 02:13:35 -0700 | [diff] [blame] | 339 | points = badness(p, uptime.tv_sec); |
KOSAKI Motohiro | 0753ba0 | 2009-08-18 14:11:10 -0700 | [diff] [blame] | 340 | if (points > *ppoints || !chosen) { |
Paul Jackson | a49335c | 2005-09-06 15:18:09 -0700 | [diff] [blame] | 341 | chosen = p; |
Kurt Garloff | 9827b78 | 2006-02-20 18:27:51 -0800 | [diff] [blame] | 342 | *ppoints = points; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 343 | } |
KOSAKI Motohiro | 495789a | 2009-09-21 17:03:14 -0700 | [diff] [blame] | 344 | } |
Oleg Nesterov | 972c4ea | 2006-09-29 02:01:12 -0700 | [diff] [blame] | 345 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 346 | return chosen; |
| 347 | } |
| 348 | |
| 349 | /** |
Randy Dunlap | 1b578df | 2008-03-19 17:00:42 -0700 | [diff] [blame] | 350 | * dump_tasks - dump current memory state of all system tasks |
David Rientjes | 74ab7f1 | 2010-08-09 17:18:46 -0700 | [diff] [blame] | 351 | * @mem: current's memory controller, if constrained |
Randy Dunlap | 1b578df | 2008-03-19 17:00:42 -0700 | [diff] [blame] | 352 | * |
David Rientjes | fef1bdd | 2008-02-07 00:14:07 -0800 | [diff] [blame] | 353 | * Dumps the current memory state of all system tasks, excluding kernel threads. |
| 354 | * State information includes task's pid, uid, tgid, vm size, rss, cpu, oom_adj |
| 355 | * score, and name. |
| 356 | * |
| 357 | * If the actual is non-NULL, only tasks that are a member of the mem_cgroup are |
| 358 | * shown. |
| 359 | * |
| 360 | * Call with tasklist_lock read-locked. |
| 361 | */ |
| 362 | static void dump_tasks(const struct mem_cgroup *mem) |
| 363 | { |
KOSAKI Motohiro | c55db95 | 2010-08-09 17:18:46 -0700 | [diff] [blame] | 364 | struct task_struct *p; |
| 365 | struct task_struct *task; |
David Rientjes | fef1bdd | 2008-02-07 00:14:07 -0800 | [diff] [blame] | 366 | |
| 367 | printk(KERN_INFO "[ pid ] uid tgid total_vm rss cpu oom_adj " |
| 368 | "name\n"); |
KOSAKI Motohiro | c55db95 | 2010-08-09 17:18:46 -0700 | [diff] [blame] | 369 | for_each_process(p) { |
KOSAKI Motohiro | c55db95 | 2010-08-09 17:18:46 -0700 | [diff] [blame] | 370 | if (p->flags & PF_KTHREAD) |
| 371 | continue; |
David Rientjes | fef1bdd | 2008-02-07 00:14:07 -0800 | [diff] [blame] | 372 | if (mem && !task_in_mem_cgroup(p, mem)) |
| 373 | continue; |
| 374 | |
KOSAKI Motohiro | c55db95 | 2010-08-09 17:18:46 -0700 | [diff] [blame] | 375 | task = find_lock_task_mm(p); |
| 376 | if (!task) { |
David Rientjes | 6d2661e | 2009-05-28 14:34:19 -0700 | [diff] [blame] | 377 | /* |
David Rientjes | 74ab7f1 | 2010-08-09 17:18:46 -0700 | [diff] [blame] | 378 | * This is a kthread or all of p's threads have already |
| 379 | * detached their mm's. There's no need to report |
KOSAKI Motohiro | c55db95 | 2010-08-09 17:18:46 -0700 | [diff] [blame] | 380 | * them; they can't be oom killed anyway. |
David Rientjes | 6d2661e | 2009-05-28 14:34:19 -0700 | [diff] [blame] | 381 | */ |
David Rientjes | 6d2661e | 2009-05-28 14:34:19 -0700 | [diff] [blame] | 382 | continue; |
| 383 | } |
KOSAKI Motohiro | c55db95 | 2010-08-09 17:18:46 -0700 | [diff] [blame] | 384 | |
David Rientjes | c81fac5 | 2010-08-09 17:18:47 -0700 | [diff] [blame] | 385 | printk(KERN_INFO "[%5d] %5d %5d %8lu %8lu %3u %3d %s\n", |
KOSAKI Motohiro | c55db95 | 2010-08-09 17:18:46 -0700 | [diff] [blame] | 386 | task->pid, __task_cred(task)->uid, task->tgid, |
| 387 | task->mm->total_vm, get_mm_rss(task->mm), |
David Rientjes | c81fac5 | 2010-08-09 17:18:47 -0700 | [diff] [blame] | 388 | task_cpu(task), task->signal->oom_adj, task->comm); |
KOSAKI Motohiro | c55db95 | 2010-08-09 17:18:46 -0700 | [diff] [blame] | 389 | task_unlock(task); |
| 390 | } |
David Rientjes | fef1bdd | 2008-02-07 00:14:07 -0800 | [diff] [blame] | 391 | } |
| 392 | |
Daisuke Nishimura | d31f56d | 2009-12-15 16:47:12 -0800 | [diff] [blame] | 393 | static void dump_header(struct task_struct *p, gfp_t gfp_mask, int order, |
| 394 | struct mem_cgroup *mem) |
David Rientjes | 1b604d7 | 2009-12-14 17:57:47 -0800 | [diff] [blame] | 395 | { |
David Rientjes | 5e9d834 | 2010-08-09 17:18:51 -0700 | [diff] [blame] | 396 | task_lock(current); |
David Rientjes | 1b604d7 | 2009-12-14 17:57:47 -0800 | [diff] [blame] | 397 | pr_warning("%s invoked oom-killer: gfp_mask=0x%x, order=%d, " |
| 398 | "oom_adj=%d\n", |
| 399 | current->comm, gfp_mask, order, current->signal->oom_adj); |
David Rientjes | 1b604d7 | 2009-12-14 17:57:47 -0800 | [diff] [blame] | 400 | cpuset_print_task_mems_allowed(current); |
| 401 | task_unlock(current); |
| 402 | dump_stack(); |
Daisuke Nishimura | d31f56d | 2009-12-15 16:47:12 -0800 | [diff] [blame] | 403 | mem_cgroup_print_oom_info(mem, p); |
David Rientjes | 1b604d7 | 2009-12-14 17:57:47 -0800 | [diff] [blame] | 404 | show_mem(); |
| 405 | if (sysctl_oom_dump_tasks) |
| 406 | dump_tasks(mem); |
| 407 | } |
| 408 | |
KOSAKI Motohiro | 3b4798c | 2009-12-15 16:45:32 -0800 | [diff] [blame] | 409 | #define K(x) ((x) << (PAGE_SHIFT-10)) |
Nick Piggin | f3af38d | 2006-12-06 20:31:51 -0800 | [diff] [blame] | 410 | static int oom_kill_task(struct task_struct *p) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 411 | { |
David Rientjes | b940fd7 | 2010-08-09 17:18:57 -0700 | [diff] [blame] | 412 | p = find_lock_task_mm(p); |
| 413 | if (!p || p->signal->oom_adj == OOM_DISABLE) { |
| 414 | task_unlock(p); |
Dave Peterson | 0131592 | 2006-04-18 22:20:44 -0700 | [diff] [blame] | 415 | return 1; |
David Rientjes | b940fd7 | 2010-08-09 17:18:57 -0700 | [diff] [blame] | 416 | } |
| 417 | pr_err("Killed process %d (%s) total-vm:%lukB, anon-rss:%lukB, file-rss:%lukB\n", |
| 418 | task_pid_nr(p), p->comm, K(p->mm->total_vm), |
| 419 | K(get_mm_counter(p->mm, MM_ANONPAGES)), |
| 420 | K(get_mm_counter(p->mm, MM_FILEPAGES))); |
| 421 | task_unlock(p); |
KOSAKI Motohiro | 0753ba0 | 2009-08-18 14:11:10 -0700 | [diff] [blame] | 422 | |
David Rientjes | b940fd7 | 2010-08-09 17:18:57 -0700 | [diff] [blame] | 423 | p->rt.time_slice = HZ; |
| 424 | set_tsk_thread_flag(p, TIF_MEMDIE); |
| 425 | force_sig(SIGKILL, p); |
Dave Peterson | 0131592 | 2006-04-18 22:20:44 -0700 | [diff] [blame] | 426 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 427 | } |
David Rientjes | b940fd7 | 2010-08-09 17:18:57 -0700 | [diff] [blame] | 428 | #undef K |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 429 | |
David Rientjes | 7213f50 | 2007-10-16 23:25:57 -0700 | [diff] [blame] | 430 | 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] | 431 | unsigned long points, struct mem_cgroup *mem, |
| 432 | const char *message) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 433 | { |
David Rientjes | 5e9d834 | 2010-08-09 17:18:51 -0700 | [diff] [blame] | 434 | struct task_struct *victim = p; |
| 435 | struct task_struct *child; |
Oleg Nesterov | dd8e8f4 | 2010-08-09 17:18:45 -0700 | [diff] [blame] | 436 | struct task_struct *t = p; |
David Rientjes | 5e9d834 | 2010-08-09 17:18:51 -0700 | [diff] [blame] | 437 | unsigned long victim_points = 0; |
| 438 | struct timespec uptime; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 439 | |
David Rientjes | 1b604d7 | 2009-12-14 17:57:47 -0800 | [diff] [blame] | 440 | if (printk_ratelimit()) |
Daisuke Nishimura | d31f56d | 2009-12-15 16:47:12 -0800 | [diff] [blame] | 441 | dump_header(p, gfp_mask, order, mem); |
David Rientjes | 7213f50 | 2007-10-16 23:25:57 -0700 | [diff] [blame] | 442 | |
Nick Piggin | 50ec3bb | 2006-09-25 23:31:29 -0700 | [diff] [blame] | 443 | /* |
| 444 | * If the task is already exiting, don't alarm the sysadmin or kill |
| 445 | * its children or threads, just set TIF_MEMDIE so it can die quickly |
| 446 | */ |
KOSAKI Motohiro | 0753ba0 | 2009-08-18 14:11:10 -0700 | [diff] [blame] | 447 | if (p->flags & PF_EXITING) { |
David Rientjes | 4358997 | 2010-08-09 17:18:49 -0700 | [diff] [blame] | 448 | set_tsk_thread_flag(p, TIF_MEMDIE); |
Nick Piggin | 50ec3bb | 2006-09-25 23:31:29 -0700 | [diff] [blame] | 449 | return 0; |
| 450 | } |
| 451 | |
David Rientjes | 5e9d834 | 2010-08-09 17:18:51 -0700 | [diff] [blame] | 452 | task_lock(p); |
| 453 | pr_err("%s: Kill process %d (%s) score %lu or sacrifice child\n", |
| 454 | message, task_pid_nr(p), p->comm, points); |
| 455 | task_unlock(p); |
Nick Piggin | f3af38d | 2006-12-06 20:31:51 -0800 | [diff] [blame] | 456 | |
David Rientjes | 5e9d834 | 2010-08-09 17:18:51 -0700 | [diff] [blame] | 457 | /* |
| 458 | * If any of p's children has a different mm and is eligible for kill, |
| 459 | * the one with the highest badness() score is sacrificed for its |
| 460 | * parent. This attempts to lose the minimal amount of work done while |
| 461 | * still freeing memory. |
| 462 | */ |
| 463 | do_posix_clock_monotonic_gettime(&uptime); |
Oleg Nesterov | dd8e8f4 | 2010-08-09 17:18:45 -0700 | [diff] [blame] | 464 | do { |
David Rientjes | 5e9d834 | 2010-08-09 17:18:51 -0700 | [diff] [blame] | 465 | list_for_each_entry(child, &t->children, sibling) { |
| 466 | unsigned long child_points; |
| 467 | |
| 468 | if (child->mm == p->mm) |
Oleg Nesterov | dd8e8f4 | 2010-08-09 17:18:45 -0700 | [diff] [blame] | 469 | continue; |
David Rientjes | 5e9d834 | 2010-08-09 17:18:51 -0700 | [diff] [blame] | 470 | if (mem && !task_in_mem_cgroup(child, mem)) |
Oleg Nesterov | dd8e8f4 | 2010-08-09 17:18:45 -0700 | [diff] [blame] | 471 | continue; |
David Rientjes | 5e9d834 | 2010-08-09 17:18:51 -0700 | [diff] [blame] | 472 | |
| 473 | /* badness() returns 0 if the thread is unkillable */ |
| 474 | child_points = badness(child, uptime.tv_sec); |
| 475 | if (child_points > victim_points) { |
| 476 | victim = child; |
| 477 | victim_points = child_points; |
| 478 | } |
Oleg Nesterov | dd8e8f4 | 2010-08-09 17:18:45 -0700 | [diff] [blame] | 479 | } |
| 480 | } while_each_thread(p, t); |
| 481 | |
David Rientjes | 5e9d834 | 2010-08-09 17:18:51 -0700 | [diff] [blame] | 482 | return oom_kill_task(victim); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 483 | } |
| 484 | |
David Rientjes | 309ed88 | 2010-08-09 17:18:54 -0700 | [diff] [blame] | 485 | /* |
| 486 | * Determines whether the kernel must panic because of the panic_on_oom sysctl. |
| 487 | */ |
| 488 | static void check_panic_on_oom(enum oom_constraint constraint, gfp_t gfp_mask, |
| 489 | int order) |
| 490 | { |
| 491 | if (likely(!sysctl_panic_on_oom)) |
| 492 | return; |
| 493 | if (sysctl_panic_on_oom != 2) { |
| 494 | /* |
| 495 | * panic_on_oom == 1 only affects CONSTRAINT_NONE, the kernel |
| 496 | * does not panic for cpuset, mempolicy, or memcg allocation |
| 497 | * failures. |
| 498 | */ |
| 499 | if (constraint != CONSTRAINT_NONE) |
| 500 | return; |
| 501 | } |
| 502 | read_lock(&tasklist_lock); |
| 503 | dump_header(NULL, gfp_mask, order, NULL); |
| 504 | read_unlock(&tasklist_lock); |
| 505 | panic("Out of memory: %s panic_on_oom is enabled\n", |
| 506 | sysctl_panic_on_oom == 2 ? "compulsory" : "system-wide"); |
| 507 | } |
| 508 | |
Balbir Singh | 00f0b82 | 2008-03-04 14:28:39 -0800 | [diff] [blame] | 509 | #ifdef CONFIG_CGROUP_MEM_RES_CTLR |
Pavel Emelianov | c7ba5c9 | 2008-02-07 00:13:58 -0800 | [diff] [blame] | 510 | void mem_cgroup_out_of_memory(struct mem_cgroup *mem, gfp_t gfp_mask) |
| 511 | { |
| 512 | unsigned long points = 0; |
| 513 | struct task_struct *p; |
| 514 | |
David Rientjes | 309ed88 | 2010-08-09 17:18:54 -0700 | [diff] [blame] | 515 | check_panic_on_oom(CONSTRAINT_MEMCG, gfp_mask, 0); |
Li Zefan | e115f2d | 2008-04-15 14:34:37 -0700 | [diff] [blame] | 516 | read_lock(&tasklist_lock); |
Pavel Emelianov | c7ba5c9 | 2008-02-07 00:13:58 -0800 | [diff] [blame] | 517 | retry: |
David Rientjes | f442003 | 2010-08-09 17:18:58 -0700 | [diff] [blame^] | 518 | p = select_bad_process(&points, mem, NULL); |
David Rientjes | df64f81 | 2010-05-26 14:42:41 -0700 | [diff] [blame] | 519 | if (!p || PTR_ERR(p) == -1UL) |
Pavel Emelianov | c7ba5c9 | 2008-02-07 00:13:58 -0800 | [diff] [blame] | 520 | goto out; |
| 521 | |
David Rientjes | fef1bdd | 2008-02-07 00:14:07 -0800 | [diff] [blame] | 522 | if (oom_kill_process(p, gfp_mask, 0, points, mem, |
Pavel Emelianov | c7ba5c9 | 2008-02-07 00:13:58 -0800 | [diff] [blame] | 523 | "Memory cgroup out of memory")) |
| 524 | goto retry; |
| 525 | out: |
Li Zefan | e115f2d | 2008-04-15 14:34:37 -0700 | [diff] [blame] | 526 | read_unlock(&tasklist_lock); |
Pavel Emelianov | c7ba5c9 | 2008-02-07 00:13:58 -0800 | [diff] [blame] | 527 | } |
| 528 | #endif |
| 529 | |
Martin Schwidefsky | 8bc719d | 2006-09-25 23:31:20 -0700 | [diff] [blame] | 530 | static BLOCKING_NOTIFIER_HEAD(oom_notify_list); |
| 531 | |
| 532 | int register_oom_notifier(struct notifier_block *nb) |
| 533 | { |
| 534 | return blocking_notifier_chain_register(&oom_notify_list, nb); |
| 535 | } |
| 536 | EXPORT_SYMBOL_GPL(register_oom_notifier); |
| 537 | |
| 538 | int unregister_oom_notifier(struct notifier_block *nb) |
| 539 | { |
| 540 | return blocking_notifier_chain_unregister(&oom_notify_list, nb); |
| 541 | } |
| 542 | EXPORT_SYMBOL_GPL(unregister_oom_notifier); |
| 543 | |
David Rientjes | 098d7f1 | 2007-10-16 23:25:55 -0700 | [diff] [blame] | 544 | /* |
| 545 | * Try to acquire the OOM killer lock for the zones in zonelist. Returns zero |
| 546 | * if a parallel OOM killing is already taking place that includes a zone in |
| 547 | * the zonelist. Otherwise, locks all zones in the zonelist and returns 1. |
| 548 | */ |
Minchan Kim | ff321fe | 2010-08-09 17:18:57 -0700 | [diff] [blame] | 549 | int try_set_zonelist_oom(struct zonelist *zonelist, gfp_t gfp_mask) |
David Rientjes | 098d7f1 | 2007-10-16 23:25:55 -0700 | [diff] [blame] | 550 | { |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 551 | struct zoneref *z; |
| 552 | struct zone *zone; |
David Rientjes | 098d7f1 | 2007-10-16 23:25:55 -0700 | [diff] [blame] | 553 | int ret = 1; |
| 554 | |
David Rientjes | c7d4cae | 2009-01-06 14:39:00 -0800 | [diff] [blame] | 555 | spin_lock(&zone_scan_lock); |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 556 | for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) { |
| 557 | if (zone_is_oom_locked(zone)) { |
David Rientjes | 098d7f1 | 2007-10-16 23:25:55 -0700 | [diff] [blame] | 558 | ret = 0; |
| 559 | goto out; |
| 560 | } |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 561 | } |
David Rientjes | 098d7f1 | 2007-10-16 23:25:55 -0700 | [diff] [blame] | 562 | |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 563 | for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) { |
| 564 | /* |
David Rientjes | c7d4cae | 2009-01-06 14:39:00 -0800 | [diff] [blame] | 565 | * Lock each zone in the zonelist under zone_scan_lock so a |
Minchan Kim | ff321fe | 2010-08-09 17:18:57 -0700 | [diff] [blame] | 566 | * parallel invocation of try_set_zonelist_oom() doesn't succeed |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 567 | * when it shouldn't. |
| 568 | */ |
| 569 | zone_set_flag(zone, ZONE_OOM_LOCKED); |
| 570 | } |
| 571 | |
David Rientjes | 098d7f1 | 2007-10-16 23:25:55 -0700 | [diff] [blame] | 572 | out: |
David Rientjes | c7d4cae | 2009-01-06 14:39:00 -0800 | [diff] [blame] | 573 | spin_unlock(&zone_scan_lock); |
David Rientjes | 098d7f1 | 2007-10-16 23:25:55 -0700 | [diff] [blame] | 574 | return ret; |
| 575 | } |
| 576 | |
| 577 | /* |
| 578 | * Clears the ZONE_OOM_LOCKED flag for all zones in the zonelist so that failed |
| 579 | * allocation attempts with zonelists containing them may now recall the OOM |
| 580 | * killer, if necessary. |
| 581 | */ |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 582 | void clear_zonelist_oom(struct zonelist *zonelist, gfp_t gfp_mask) |
David Rientjes | 098d7f1 | 2007-10-16 23:25:55 -0700 | [diff] [blame] | 583 | { |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 584 | struct zoneref *z; |
| 585 | struct zone *zone; |
David Rientjes | 098d7f1 | 2007-10-16 23:25:55 -0700 | [diff] [blame] | 586 | |
David Rientjes | c7d4cae | 2009-01-06 14:39:00 -0800 | [diff] [blame] | 587 | spin_lock(&zone_scan_lock); |
Mel Gorman | dd1a239 | 2008-04-28 02:12:17 -0700 | [diff] [blame] | 588 | for_each_zone_zonelist(zone, z, zonelist, gfp_zone(gfp_mask)) { |
| 589 | zone_clear_flag(zone, ZONE_OOM_LOCKED); |
| 590 | } |
David Rientjes | c7d4cae | 2009-01-06 14:39:00 -0800 | [diff] [blame] | 591 | spin_unlock(&zone_scan_lock); |
David Rientjes | 098d7f1 | 2007-10-16 23:25:55 -0700 | [diff] [blame] | 592 | } |
| 593 | |
Nick Piggin | 1c0fe6e | 2009-01-06 14:38:59 -0800 | [diff] [blame] | 594 | /* |
David Rientjes | e365893 | 2010-08-09 17:18:55 -0700 | [diff] [blame] | 595 | * Try to acquire the oom killer lock for all system zones. Returns zero if a |
| 596 | * parallel oom killing is taking place, otherwise locks all zones and returns |
| 597 | * non-zero. |
| 598 | */ |
| 599 | static int try_set_system_oom(void) |
| 600 | { |
| 601 | struct zone *zone; |
| 602 | int ret = 1; |
| 603 | |
| 604 | spin_lock(&zone_scan_lock); |
| 605 | for_each_populated_zone(zone) |
| 606 | if (zone_is_oom_locked(zone)) { |
| 607 | ret = 0; |
| 608 | goto out; |
| 609 | } |
| 610 | for_each_populated_zone(zone) |
| 611 | zone_set_flag(zone, ZONE_OOM_LOCKED); |
| 612 | out: |
| 613 | spin_unlock(&zone_scan_lock); |
| 614 | return ret; |
| 615 | } |
| 616 | |
| 617 | /* |
| 618 | * Clears ZONE_OOM_LOCKED for all system zones so that failed allocation |
| 619 | * attempts or page faults may now recall the oom killer, if necessary. |
| 620 | */ |
| 621 | static void clear_system_oom(void) |
| 622 | { |
| 623 | struct zone *zone; |
| 624 | |
| 625 | spin_lock(&zone_scan_lock); |
| 626 | for_each_populated_zone(zone) |
| 627 | zone_clear_flag(zone, ZONE_OOM_LOCKED); |
| 628 | spin_unlock(&zone_scan_lock); |
| 629 | } |
| 630 | |
| 631 | |
| 632 | /* |
Nick Piggin | 1c0fe6e | 2009-01-06 14:38:59 -0800 | [diff] [blame] | 633 | * Must be called with tasklist_lock held for read. |
| 634 | */ |
David Rientjes | f442003 | 2010-08-09 17:18:58 -0700 | [diff] [blame^] | 635 | static void __out_of_memory(gfp_t gfp_mask, int order, const nodemask_t *mask) |
Nick Piggin | 1c0fe6e | 2009-01-06 14:38:59 -0800 | [diff] [blame] | 636 | { |
David Rientjes | 184101b | 2009-05-06 16:02:55 -0700 | [diff] [blame] | 637 | struct task_struct *p; |
| 638 | unsigned long points; |
Nick Piggin | 1c0fe6e | 2009-01-06 14:38:59 -0800 | [diff] [blame] | 639 | |
David Rientjes | 184101b | 2009-05-06 16:02:55 -0700 | [diff] [blame] | 640 | if (sysctl_oom_kill_allocating_task) |
| 641 | if (!oom_kill_process(current, gfp_mask, order, 0, NULL, |
| 642 | "Out of memory (oom_kill_allocating_task)")) |
Nick Piggin | 1c0fe6e | 2009-01-06 14:38:59 -0800 | [diff] [blame] | 643 | return; |
David Rientjes | 184101b | 2009-05-06 16:02:55 -0700 | [diff] [blame] | 644 | retry: |
| 645 | /* |
| 646 | * Rambo mode: Shoot down a process and hope it solves whatever |
| 647 | * issues we may have. |
| 648 | */ |
David Rientjes | f442003 | 2010-08-09 17:18:58 -0700 | [diff] [blame^] | 649 | p = select_bad_process(&points, NULL, mask); |
Nick Piggin | 1c0fe6e | 2009-01-06 14:38:59 -0800 | [diff] [blame] | 650 | |
David Rientjes | 184101b | 2009-05-06 16:02:55 -0700 | [diff] [blame] | 651 | if (PTR_ERR(p) == -1UL) |
| 652 | return; |
Nick Piggin | 1c0fe6e | 2009-01-06 14:38:59 -0800 | [diff] [blame] | 653 | |
David Rientjes | 184101b | 2009-05-06 16:02:55 -0700 | [diff] [blame] | 654 | /* Found nothing?!?! Either we hang forever, or we panic. */ |
| 655 | if (!p) { |
Daisuke Nishimura | d31f56d | 2009-12-15 16:47:12 -0800 | [diff] [blame] | 656 | dump_header(NULL, gfp_mask, order, NULL); |
David Rientjes | 309ed88 | 2010-08-09 17:18:54 -0700 | [diff] [blame] | 657 | read_unlock(&tasklist_lock); |
David Rientjes | 184101b | 2009-05-06 16:02:55 -0700 | [diff] [blame] | 658 | panic("Out of memory and no killable processes...\n"); |
Nick Piggin | 1c0fe6e | 2009-01-06 14:38:59 -0800 | [diff] [blame] | 659 | } |
David Rientjes | 184101b | 2009-05-06 16:02:55 -0700 | [diff] [blame] | 660 | |
| 661 | if (oom_kill_process(p, gfp_mask, order, points, NULL, |
| 662 | "Out of memory")) |
| 663 | goto retry; |
Nick Piggin | 1c0fe6e | 2009-01-06 14:38:59 -0800 | [diff] [blame] | 664 | } |
| 665 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 666 | /** |
Dave Peterson | 6937a25 | 2006-06-23 02:03:13 -0700 | [diff] [blame] | 667 | * 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] | 668 | * @zonelist: zonelist pointer |
| 669 | * @gfp_mask: memory allocation flags |
| 670 | * @order: amount of memory being requested as a power of 2 |
David Rientjes | 6f48d0eb | 2010-08-09 17:18:52 -0700 | [diff] [blame] | 671 | * @nodemask: nodemask passed to page allocator |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 672 | * |
| 673 | * If we run out of memory, we have the choice between either |
| 674 | * killing a random task (bad), letting the system crash (worse) |
| 675 | * OR try to be smart about which process to kill. Note that we |
| 676 | * don't have to be perfect here, we just have to be good. |
| 677 | */ |
KAMEZAWA Hiroyuki | 4365a56 | 2009-12-15 16:45:33 -0800 | [diff] [blame] | 678 | void out_of_memory(struct zonelist *zonelist, gfp_t gfp_mask, |
| 679 | int order, nodemask_t *nodemask) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 680 | { |
Martin Schwidefsky | 8bc719d | 2006-09-25 23:31:20 -0700 | [diff] [blame] | 681 | unsigned long freed = 0; |
David Rientjes | e365893 | 2010-08-09 17:18:55 -0700 | [diff] [blame] | 682 | enum oom_constraint constraint = CONSTRAINT_NONE; |
Martin Schwidefsky | 8bc719d | 2006-09-25 23:31:20 -0700 | [diff] [blame] | 683 | |
| 684 | blocking_notifier_call_chain(&oom_notify_list, 0, &freed); |
| 685 | if (freed > 0) |
| 686 | /* Got some memory back in the last second. */ |
| 687 | return; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 688 | |
David Rientjes | 7b98c2e | 2010-08-09 17:18:48 -0700 | [diff] [blame] | 689 | /* |
| 690 | * If current has a pending SIGKILL, then automatically select it. The |
| 691 | * goal is to allow it to allocate so that it may quickly exit and free |
| 692 | * its memory. |
| 693 | */ |
| 694 | if (fatal_signal_pending(current)) { |
| 695 | set_thread_flag(TIF_MEMDIE); |
| 696 | return; |
| 697 | } |
| 698 | |
Christoph Lameter | 9b0f8b0 | 2006-02-20 18:27:52 -0800 | [diff] [blame] | 699 | /* |
| 700 | * Check if there were limitations on the allocation (only relevant for |
| 701 | * NUMA) that may require different handling. |
| 702 | */ |
David Rientjes | e365893 | 2010-08-09 17:18:55 -0700 | [diff] [blame] | 703 | if (zonelist) |
| 704 | constraint = constrained_alloc(zonelist, gfp_mask, nodemask); |
David Rientjes | 309ed88 | 2010-08-09 17:18:54 -0700 | [diff] [blame] | 705 | check_panic_on_oom(constraint, gfp_mask, order); |
David Rientjes | 2b45ab3 | 2007-05-06 14:50:00 -0700 | [diff] [blame] | 706 | read_lock(&tasklist_lock); |
David Rientjes | f442003 | 2010-08-09 17:18:58 -0700 | [diff] [blame^] | 707 | __out_of_memory(gfp_mask, order, |
| 708 | constraint == CONSTRAINT_MEMORY_POLICY ? nodemask : |
| 709 | NULL); |
Andrew Morton | 140ffce | 2006-03-02 02:54:28 -0800 | [diff] [blame] | 710 | read_unlock(&tasklist_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 711 | |
| 712 | /* |
| 713 | * Give "p" a good chance of killing itself before we |
Kirill Korotaev | 2f659f4 | 2006-01-08 01:01:05 -0800 | [diff] [blame] | 714 | * retry to allocate memory unless "p" is current |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 715 | */ |
Kirill Korotaev | 2f659f4 | 2006-01-08 01:01:05 -0800 | [diff] [blame] | 716 | if (!test_thread_flag(TIF_MEMDIE)) |
Andrew Morton | 140ffce | 2006-03-02 02:54:28 -0800 | [diff] [blame] | 717 | schedule_timeout_uninterruptible(1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 718 | } |
David Rientjes | e365893 | 2010-08-09 17:18:55 -0700 | [diff] [blame] | 719 | |
| 720 | /* |
| 721 | * The pagefault handler calls here because it is out of memory, so kill a |
| 722 | * memory-hogging task. If a populated zone has ZONE_OOM_LOCKED set, a parallel |
| 723 | * oom killing is already in progress so do nothing. If a task is found with |
| 724 | * TIF_MEMDIE set, it has been killed so do nothing and allow it to exit. |
| 725 | */ |
| 726 | void pagefault_out_of_memory(void) |
| 727 | { |
| 728 | if (try_set_system_oom()) { |
| 729 | out_of_memory(NULL, 0, 0, NULL); |
| 730 | clear_system_oom(); |
| 731 | } |
| 732 | if (!test_thread_flag(TIF_MEMDIE)) |
| 733 | schedule_timeout_uninterruptible(1); |
| 734 | } |