aboutsummaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorMandeep Singh Baines <msb@google.com>2009-02-06 18:37:47 -0500
committerIngo Molnar <mingo@elte.hu>2009-02-09 05:03:49 -0500
commit17406b82d621930cca8ccc1272cdac9a7dae8e40 (patch)
tree224c8b0b7c78a5467175fe10e116bdc6b73495ea
parent94be52dc075a32af4aa73d7e10f68734d62d6af2 (diff)
softlockup: remove timestamp checking from hung_task
Impact: saves sizeof(long) bytes per task_struct By guaranteeing that sysctl_hung_task_timeout_secs have elapsed between tasklist scans we can avoid using timestamps. Signed-off-by: Mandeep Singh Baines <msb@google.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
-rw-r--r--include/linux/sched.h1
-rw-r--r--kernel/fork.c8
-rw-r--r--kernel/hung_task.c48
3 files changed, 12 insertions, 45 deletions
diff --git a/include/linux/sched.h b/include/linux/sched.h
index 2a2811c6239d..e0d723fea9f5 100644
--- a/include/linux/sched.h
+++ b/include/linux/sched.h
@@ -1241,7 +1241,6 @@ struct task_struct {
1241#endif 1241#endif
1242#ifdef CONFIG_DETECT_HUNG_TASK 1242#ifdef CONFIG_DETECT_HUNG_TASK
1243/* hung task detection */ 1243/* hung task detection */
1244 unsigned long last_switch_timestamp;
1245 unsigned long last_switch_count; 1244 unsigned long last_switch_count;
1246#endif 1245#endif
1247/* CPU-specific state of this task */ 1246/* CPU-specific state of this task */
diff --git a/kernel/fork.c b/kernel/fork.c
index fb9444282836..bf582f75014b 100644
--- a/kernel/fork.c
+++ b/kernel/fork.c
@@ -639,6 +639,9 @@ static int copy_mm(unsigned long clone_flags, struct task_struct * tsk)
639 639
640 tsk->min_flt = tsk->maj_flt = 0; 640 tsk->min_flt = tsk->maj_flt = 0;
641 tsk->nvcsw = tsk->nivcsw = 0; 641 tsk->nvcsw = tsk->nivcsw = 0;
642#ifdef CONFIG_DETECT_HUNG_TASK
643 tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
644#endif
642 645
643 tsk->mm = NULL; 646 tsk->mm = NULL;
644 tsk->active_mm = NULL; 647 tsk->active_mm = NULL;
@@ -1041,11 +1044,6 @@ static struct task_struct *copy_process(unsigned long clone_flags,
1041 1044
1042 p->default_timer_slack_ns = current->timer_slack_ns; 1045 p->default_timer_slack_ns = current->timer_slack_ns;
1043 1046
1044#ifdef CONFIG_DETECT_HUNG_TASK
1045 p->last_switch_count = 0;
1046 p->last_switch_timestamp = 0;
1047#endif
1048
1049 task_io_accounting_init(&p->ioac); 1047 task_io_accounting_init(&p->ioac);
1050 acct_clear_integrals(p); 1048 acct_clear_integrals(p);
1051 1049
diff --git a/kernel/hung_task.c b/kernel/hung_task.c
index 3951a80e7cbe..0c924de58cb2 100644
--- a/kernel/hung_task.c
+++ b/kernel/hung_task.c
@@ -34,7 +34,6 @@ unsigned long __read_mostly sysctl_hung_task_check_count = PID_MAX_LIMIT;
34 * Zero means infinite timeout - no checking done: 34 * Zero means infinite timeout - no checking done:
35 */ 35 */
36unsigned long __read_mostly sysctl_hung_task_timeout_secs = 120; 36unsigned long __read_mostly sysctl_hung_task_timeout_secs = 120;
37static unsigned long __read_mostly hung_task_poll_jiffies;
38 37
39unsigned long __read_mostly sysctl_hung_task_warnings = 10; 38unsigned long __read_mostly sysctl_hung_task_warnings = 10;
40 39
@@ -69,33 +68,17 @@ static struct notifier_block panic_block = {
69 .notifier_call = hung_task_panic, 68 .notifier_call = hung_task_panic,
70}; 69};
71 70
72/* 71static void check_hung_task(struct task_struct *t, unsigned long timeout)
73 * Returns seconds, approximately. We don't need nanosecond
74 * resolution, and we don't need to waste time with a big divide when
75 * 2^30ns == 1.074s.
76 */
77static unsigned long get_timestamp(void)
78{
79 int this_cpu = raw_smp_processor_id();
80
81 return cpu_clock(this_cpu) >> 30LL; /* 2^30 ~= 10^9 */
82}
83
84static void check_hung_task(struct task_struct *t, unsigned long now,
85 unsigned long timeout)
86{ 72{
87 unsigned long switch_count = t->nvcsw + t->nivcsw; 73 unsigned long switch_count = t->nvcsw + t->nivcsw;
88 74
89 if (t->flags & PF_FROZEN) 75 if (t->flags & PF_FROZEN)
90 return; 76 return;
91 77
92 if (switch_count != t->last_switch_count || !t->last_switch_timestamp) { 78 if (switch_count != t->last_switch_count) {
93 t->last_switch_count = switch_count; 79 t->last_switch_count = switch_count;
94 t->last_switch_timestamp = now;
95 return; 80 return;
96 } 81 }
97 if ((long)(now - t->last_switch_timestamp) < timeout)
98 return;
99 if (!sysctl_hung_task_warnings) 82 if (!sysctl_hung_task_warnings)
100 return; 83 return;
101 sysctl_hung_task_warnings--; 84 sysctl_hung_task_warnings--;
@@ -111,7 +94,6 @@ static void check_hung_task(struct task_struct *t, unsigned long now,
111 sched_show_task(t); 94 sched_show_task(t);
112 __debug_show_held_locks(t); 95 __debug_show_held_locks(t);
113 96
114 t->last_switch_timestamp = now;
115 touch_nmi_watchdog(); 97 touch_nmi_watchdog();
116 98
117 if (sysctl_hung_task_panic) 99 if (sysctl_hung_task_panic)
@@ -145,7 +127,6 @@ static void check_hung_uninterruptible_tasks(unsigned long timeout)
145{ 127{
146 int max_count = sysctl_hung_task_check_count; 128 int max_count = sysctl_hung_task_check_count;
147 int batch_count = HUNG_TASK_BATCHING; 129 int batch_count = HUNG_TASK_BATCHING;
148 unsigned long now = get_timestamp();
149 struct task_struct *g, *t; 130 struct task_struct *g, *t;
150 131
151 /* 132 /*
@@ -168,19 +149,16 @@ static void check_hung_uninterruptible_tasks(unsigned long timeout)
168 } 149 }
169 /* use "==" to skip the TASK_KILLABLE tasks waiting on NFS */ 150 /* use "==" to skip the TASK_KILLABLE tasks waiting on NFS */
170 if (t->state == TASK_UNINTERRUPTIBLE) 151 if (t->state == TASK_UNINTERRUPTIBLE)
171 check_hung_task(t, now, timeout); 152 check_hung_task(t, timeout);
172 } while_each_thread(g, t); 153 } while_each_thread(g, t);
173 unlock: 154 unlock:
174 rcu_read_unlock(); 155 rcu_read_unlock();
175} 156}
176 157
177static void update_poll_jiffies(void) 158static unsigned long timeout_jiffies(unsigned long timeout)
178{ 159{
179 /* timeout of 0 will disable the watchdog */ 160 /* timeout of 0 will disable the watchdog */
180 if (sysctl_hung_task_timeout_secs == 0) 161 return timeout ? timeout * HZ : MAX_SCHEDULE_TIMEOUT;
181 hung_task_poll_jiffies = MAX_SCHEDULE_TIMEOUT;
182 else
183 hung_task_poll_jiffies = sysctl_hung_task_timeout_secs * HZ / 2;
184} 162}
185 163
186/* 164/*
@@ -197,8 +175,6 @@ int proc_dohung_task_timeout_secs(struct ctl_table *table, int write,
197 if (ret || !write) 175 if (ret || !write)
198 goto out; 176 goto out;
199 177
200 update_poll_jiffies();
201
202 wake_up_process(watchdog_task); 178 wake_up_process(watchdog_task);
203 179
204 out: 180 out:
@@ -211,20 +187,14 @@ int proc_dohung_task_timeout_secs(struct ctl_table *table, int write,
211static int watchdog(void *dummy) 187static int watchdog(void *dummy)
212{ 188{
213 set_user_nice(current, 0); 189 set_user_nice(current, 0);
214 update_poll_jiffies();
215 190
216 for ( ; ; ) { 191 for ( ; ; ) {
217 unsigned long timeout; 192 unsigned long timeout = sysctl_hung_task_timeout_secs;
218 193
219 while (schedule_timeout_interruptible(hung_task_poll_jiffies)); 194 while (schedule_timeout_interruptible(timeout_jiffies(timeout)))
195 timeout = sysctl_hung_task_timeout_secs;
220 196
221 /* 197 check_hung_uninterruptible_tasks(timeout);
222 * Need to cache timeout here to avoid timeout being set
223 * to 0 via sysctl while inside check_hung_*_tasks().
224 */
225 timeout = sysctl_hung_task_timeout_secs;
226 if (timeout)
227 check_hung_uninterruptible_tasks(timeout);
228 } 198 }
229 199
230 return 0; 200 return 0;