aboutsummaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
-rw-r--r--Documentation/filesystems/proc.txt2
-rw-r--r--fs/proc/base.c2
-rw-r--r--mm/oom_kill.c35
3 files changed, 31 insertions, 8 deletions
diff --git a/Documentation/filesystems/proc.txt b/Documentation/filesystems/proc.txt
index ae7f8bb1b7bc..75988ba26a51 100644
--- a/Documentation/filesystems/proc.txt
+++ b/Documentation/filesystems/proc.txt
@@ -1205,7 +1205,7 @@ The following heuristics are then applied:
1205 * if the task was reniced, its score doubles 1205 * if the task was reniced, its score doubles
1206 * superuser or direct hardware access tasks (CAP_SYS_ADMIN, CAP_SYS_RESOURCE 1206 * superuser or direct hardware access tasks (CAP_SYS_ADMIN, CAP_SYS_RESOURCE
1207 or CAP_SYS_RAWIO) have their score divided by 4 1207 or CAP_SYS_RAWIO) have their score divided by 4
1208 * if oom condition happened in one cpuset and checked task does not belong 1208 * if oom condition happened in one cpuset and checked process does not belong
1209 to it, its score is divided by 8 1209 to it, its score is divided by 8
1210 * the resulting score is multiplied by two to the power of oom_adj, i.e. 1210 * the resulting score is multiplied by two to the power of oom_adj, i.e.
1211 points <<= oom_adj when it is positive and 1211 points <<= oom_adj when it is positive and
diff --git a/fs/proc/base.c b/fs/proc/base.c
index 81cfff82875b..71a34253dcbb 100644
--- a/fs/proc/base.c
+++ b/fs/proc/base.c
@@ -447,7 +447,7 @@ static int proc_oom_score(struct task_struct *task, char *buffer)
447 447
448 do_posix_clock_monotonic_gettime(&uptime); 448 do_posix_clock_monotonic_gettime(&uptime);
449 read_lock(&tasklist_lock); 449 read_lock(&tasklist_lock);
450 points = badness(task, uptime.tv_sec); 450 points = badness(task->group_leader, uptime.tv_sec);
451 read_unlock(&tasklist_lock); 451 read_unlock(&tasklist_lock);
452 return sprintf(buffer, "%lu\n", points); 452 return sprintf(buffer, "%lu\n", points);
453} 453}
diff --git a/mm/oom_kill.c b/mm/oom_kill.c
index 630b77fe862f..372692294844 100644
--- a/mm/oom_kill.c
+++ b/mm/oom_kill.c
@@ -34,6 +34,23 @@ int sysctl_oom_dump_tasks;
34static DEFINE_SPINLOCK(zone_scan_lock); 34static DEFINE_SPINLOCK(zone_scan_lock);
35/* #define DEBUG */ 35/* #define DEBUG */
36 36
37/*
38 * Is all threads of the target process nodes overlap ours?
39 */
40static 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
37/** 54/**
38 * badness - calculate a numeric value for how bad this task has been 55 * badness - calculate a numeric value for how bad this task has been
39 * @p: task struct of which task we should calculate 56 * @p: task struct of which task we should calculate
@@ -59,6 +76,9 @@ unsigned long badness(struct task_struct *p, unsigned long uptime)
59 struct mm_struct *mm; 76 struct mm_struct *mm;
60 struct task_struct *child; 77 struct task_struct *child;
61 int oom_adj = p->signal->oom_adj; 78 int oom_adj = p->signal->oom_adj;
79 struct task_cputime task_time;
80 unsigned long utime;
81 unsigned long stime;
62 82
63 if (oom_adj == OOM_DISABLE) 83 if (oom_adj == OOM_DISABLE)
64 return 0; 84 return 0;
@@ -106,8 +126,11 @@ unsigned long badness(struct task_struct *p, unsigned long uptime)
106 * of seconds. There is no particular reason for this other than 126 * of seconds. There is no particular reason for this other than
107 * that it turned out to work very well in practice. 127 * that it turned out to work very well in practice.
108 */ 128 */
109 cpu_time = (cputime_to_jiffies(p->utime) + cputime_to_jiffies(p->stime)) 129 thread_group_cputime(p, &task_time);
110 >> (SHIFT_HZ + 3); 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
111 134
112 if (uptime >= p->start_time.tv_sec) 135 if (uptime >= p->start_time.tv_sec)
113 run_time = (uptime - p->start_time.tv_sec) >> 10; 136 run_time = (uptime - p->start_time.tv_sec) >> 10;
@@ -148,7 +171,7 @@ unsigned long badness(struct task_struct *p, unsigned long uptime)
148 * because p may have allocated or otherwise mapped memory on 171 * because p may have allocated or otherwise mapped memory on
149 * this node before. However it will be less likely. 172 * this node before. However it will be less likely.
150 */ 173 */
151 if (!cpuset_mems_allowed_intersects(current, p)) 174 if (!has_intersects_mems_allowed(p))
152 points /= 8; 175 points /= 8;
153 176
154 /* 177 /*
@@ -204,13 +227,13 @@ static inline enum oom_constraint constrained_alloc(struct zonelist *zonelist,
204static struct task_struct *select_bad_process(unsigned long *ppoints, 227static struct task_struct *select_bad_process(unsigned long *ppoints,
205 struct mem_cgroup *mem) 228 struct mem_cgroup *mem)
206{ 229{
207 struct task_struct *g, *p; 230 struct task_struct *p;
208 struct task_struct *chosen = NULL; 231 struct task_struct *chosen = NULL;
209 struct timespec uptime; 232 struct timespec uptime;
210 *ppoints = 0; 233 *ppoints = 0;
211 234
212 do_posix_clock_monotonic_gettime(&uptime); 235 do_posix_clock_monotonic_gettime(&uptime);
213 do_each_thread(g, p) { 236 for_each_process(p) {
214 unsigned long points; 237 unsigned long points;
215 238
216 /* 239 /*
@@ -263,7 +286,7 @@ static struct task_struct *select_bad_process(unsigned long *ppoints,
263 chosen = p; 286 chosen = p;
264 *ppoints = points; 287 *ppoints = points;
265 } 288 }
266 } while_each_thread(g, p); 289 }
267 290
268 return chosen; 291 return chosen;
269} 292}