Skip to contentSkip to conversation

scheduler fixes

GIT PULL Ingo Molnar 3 participants 5 messages Dec 19, 2010 – Dec 20, 2010

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

Linus,

Please pull the latest sched-fixes-for-linus git tree from:

git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip.git sched-fixes-for-linus

It's a bit larger than usual - partly because it accumulated over two weeks and partly because two complex fixes arrived late which needed more testing.

Thanks,

Ingo

------------------>
Heiko Carstens (1):
nohz: Fix get_next_timer_interrupt() vs cpu hotplug

Mike Galbraith (1):
Sched: fix skip_clock_update optimization

Peter Zijlstra (3):
sched: Cure more NO_HZ load average woes
sched: Fix the irqtime code to deal with u64 wraps
sched: Fix the irqtime code for 32bit

 include/linux/sched.h |    2 +-
 kernel/fork.c         |    1 +
 kernel/sched.c        |  288 ++++++++++++++++++++++++++++++++++++++++---------
 kernel/timer.c        |    8 ++-
 4 files changed, 246 insertions(+), 53 deletions(-)
include/linux/sched.h +1 -1
index 2c79e92..2238745 100644
@@ -143,7 +143,7 @@ extern unsigned long nr_iowait_cpu(int cpu);
 extern unsigned long this_cpu_load(void);
 
 
-extern void calc_global_load(void);
+extern void calc_global_load(unsigned long ticks);
 
 extern unsigned long get_parent_ip(unsigned long addr);
 
kernel/fork.c +1 -0
index 3b159c5..5447dc7 100644
@@ -273,6 +273,7 @@ static struct task_struct *dup_task_struct(struct task_struct *orig)
 
 	setup_thread_stack(tsk, orig);
 	clear_user_return_notifier(tsk);
+	clear_tsk_need_resched(tsk);
 	stackend = end_of_stack(tsk);
 	*stackend = STACK_END_MAGIC;	/* for overflow detection */
 
kernel/sched.c +237 -51

405 lines not on this page · raw message

kernel/timer.c +7 -1
index 68a9ae7..353b922 100644
@@ -1252,6 +1252,12 @@ unsigned long get_next_timer_interrupt(unsigned long now)
 	struct tvec_base *base = __get_cpu_var(tvec_bases);
 	unsigned long expires;
 
+	/*
+	 * Pretend that there is no timer pending if the cpu is offline.
+	 * Possible pending timers will be migrated later to an active cpu.
+	 */
+	if (cpu_is_offline(smp_processor_id()))
+		return now + NEXT_TIMER_MAX_DELTA;
 	spin_lock(&base->lock);
 	if (time_before_eq(base->next_timer, base->timer_jiffies))
 		base->next_timer = __next_timer_interrupt(base);
@@ -1319,7 +1325,7 @@ void do_timer(unsigned long ticks)
 {
 	jiffies_64 += ticks;
 	update_wall_time();
-	calc_global_load();
+	calc_global_load(ticks);
 }
 
 #ifdef __ARCH_WANT_SYS_ALARM
#1 lore raw

kern

Quoted 9 lines · Ingo Molnar
kernel/sched.c +1 -0
@@ -3943,6 +4128,7 @@ need_resched_nonpreemptible:
                rq->nr_switches++;
                rq->curr = next;
                ++*switch_count;
+               WARN_ON_ONCE(test_tsk_need_resched(next));
 

               context_switch(rq, prev, next); /* unlocks the rq */ /*

kern renamed from kernel/timer.c +0 -0

Please don't add like this in an -rc. It looks like it's some debugging aid, I don't think it should have gone in now.

And yes, it triggers for me.

Linus

[ 1321.625764] ------------[ cut here ]------------
[ 1321.625773] WARNING: at kernel/sched.c:4131 schedule+0x5d1/0x978()
[ 1321.625774] Hardware name: System Product Name
[ 1321.625776] Modules linked in: [last unloaded: scsi_wait_scan]
[ 1321.625779] Pid: 3630, comm: chrome Not tainted 2.6.37-rc6-00132-g55ec86f #56
[ 1321.625781] Call Trace:
[ 1321.625787]  [<ffffffff810385c5>] warn_slowpath_common+0x80/0x98
[ 1321.625790]  [<ffffffff810385f2>] warn_slowpath_null+0x15/0x17
[ 1321.625792]  [<ffffffff8153b2ed>] schedule+0x5d1/0x978
[ 1321.625795]  [<ffffffff810324d0>] ? sub_preempt_count+0x92/0xa5
[ 1321.625798]  [<ffffffff8153e0aa>] ? _raw_spin_unlock_irqrestore+0x46/0x64
[ 1321.625802]  [<ffffffff810536fe>] ? __hrtimer_start_range_ns+0x328/0x33a
[ 1321.625805]  [<ffffffff8153cb6c>] schedule_hrtimeout_range_clock+0xc8/0xfa
[ 1321.625808]  [<ffffffff81052f0a>] ? hrtimer_wakeup+0x0/0x21
[ 1321.625810]  [<ffffffff81050497>] ? add_wait_queue+0x3f/0x44
[ 1321.625812]  [<ffffffff8153cbac>] schedule_hrtimeout_range+0xe/0x10
[ 1321.625816]  [<ffffffff810cc3b0>] poll_schedule_timeout+0x43/0x5f
[ 1321.625818]  [<ffffffff810cd65f>] do_sys_poll+0x351/0x3e8
[ 1321.625820]  [<ffffffff810cc45b>] ? __pollwait+0x0/0xc7
[ 1321.625822]  [<ffffffff810cc522>] ? pollwake+0x0/0x4f
[ 1321.625824]  [<ffffffff810cc522>] ? pollwake+0x0/0x4f
[ 1321.625827]  [<ffffffff810cc522>] ? pollwake+0x0/0x4f
[ 1321.625829]  [<ffffffff8103240e>] ? get_parent_ip+0x11/0x41
[ 1321.625832]  [<ffffffff811e7000>] ? sys_semtimedop+0x8bc/0x972
[ 1321.625834]  [<ffffffff811e704c>] ? sys_semtimedop+0x908/0x972
[ 1321.625838]  [<ffffffff81462299>] ? sock_aio_read+0x123/0x13b
[ 1321.625842]  [<ffffffff8103d001>] ? timespec_add_safe+0x32/0x61
[ 1321.625844]  [<ffffffff810cc6c2>] ? poll_select_set_timeout+0x61/0x75
[ 1321.625847]  [<ffffffff810cd793>] sys_poll+0x50/0xba
[ 1321.625851]  [<ffffffff81001ffb>] system_call_fastpath+0x16/0x1b
[ 1321.625853] ---[ end trace 3ba46a028c3b4324 ]---
#2 lore raw
Quoted 7 lines · Linus Torvalds
Quoted 6 lines · Ingo Molnar
(quoted diff) +1 -0
@@ -3943,6 +4128,7 @@ need_resched_nonpreemptible:
                rq->nr_switches++;
                rq->curr = next;
                ++*switch_count;
+               WARN_ON_ONCE(test_tsk_need_resched(next));
 
Quoted 6 lines

kern

Quoted 3 lines

               context_switch(rq, prev, next); /* unlocks the rq */ /*

kern renamed from kernel/timer.c +0 -0

Please don't add like this in an -rc. It looks like it's some debugging aid, I don't think it should have gone in now.

Yeah - sorry about that. The warning is bogus - please pull the updated tree below. (Can also be pulled on top of the previous one.)

Thanks,

Ingo

git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip.git sched-fixes-for-linus

------------------>
Heiko Carstens (1):
nohz: Fix get_next_timer_interrupt() vs cpu hotplug

Ingo Molnar (1):
sched: Remove debugging check

Mike Galbraith (1):
Sched: fix skip_clock_update optimization

Peter Zijlstra (3):
sched: Cure more NO_HZ load average woes
sched: Fix the irqtime code to deal with u64 wraps
sched: Fix the irqtime code for 32bit

 include/linux/sched.h |    2 +-
 kernel/fork.c         |    1 +
 kernel/sched.c        |  287 ++++++++++++++++++++++++++++++++++++++++---------
 kernel/timer.c        |    8 ++-
 4 files changed, 245 insertions(+), 53 deletions(-)
include/linux/sched.h +1 -1
index 2c79e92..2238745 100644
@@ -143,7 +143,7 @@ extern unsigned long nr_iowait_cpu(int cpu);
 extern unsigned long this_cpu_load(void);
 
 
-extern void calc_global_load(void);
+extern void calc_global_load(unsigned long ticks);
 
 extern unsigned long get_parent_ip(unsigned long addr);
 
kernel/fork.c +1 -0
index 3b159c5..5447dc7 100644
@@ -273,6 +273,7 @@ static struct task_struct *dup_task_struct(struct task_struct *orig)
 
 	setup_thread_stack(tsk, orig);
 	clear_user_return_notifier(tsk);
+	clear_tsk_need_resched(tsk);
 	stackend = end_of_stack(tsk);
 	*stackend = STACK_END_MAGIC;	/* for overflow detection */
 
kernel/sched.c +236 -51

398 lines not on this page · raw message

kernel/timer.c +7 -1
index 68a9ae7..353b922 100644
@@ -1252,6 +1252,12 @@ unsigned long get_next_timer_interrupt(unsigned long now)
 	struct tvec_base *base = __get_cpu_var(tvec_bases);
 	unsigned long expires;
 
+	/*
+	 * Pretend that there is no timer pending if the cpu is offline.
+	 * Possible pending timers will be migrated later to an active cpu.
+	 */
+	if (cpu_is_offline(smp_processor_id()))
+		return now + NEXT_TIMER_MAX_DELTA;
 	spin_lock(&base->lock);
 	if (time_before_eq(base->next_timer, base->timer_jiffies))
 		base->next_timer = __next_timer_interrupt(base);
@@ -1319,7 +1325,7 @@ void do_timer(unsigned long ticks)
 {
 	jiffies_64 += ticks;
 	update_wall_time();
-	calc_global_load();
+	calc_global_load(ticks);
 }
 
 #ifdef __ARCH_WANT_SYS_ALARM
#3 lore raw

[tip:sched/urgent] sched: Remove debugging check

In reply to Linus Torvalds
Commit-ID:  050c6c9b896625d9fa498265be17b82c5fc65257
Gitweb:     http://git.kernel.org/tip/050c6c9b896625d9fa498265be17b82c5fc65257
Author:     Ingo Molnar <[email protected]>
AuthorDate: Sun, 19 Dec 2010 23:24:27 +0100
Committer:  Ingo Molnar <[email protected]>
CommitDate: Sun, 19 Dec 2010 23:24:27 +0100

sched: Remove debugging check

Linus reported that the new warning introduced by commit f26f9aff6aaf "Sched: fix skip_clock_update optimization" triggers. The need_resched flag can be set by other CPUs asynchronously so this debug check is bogus - remove it.

Cc Peter Zijlstra <[email protected]>
Cc Mike Galbraith <[email protected]>

LKML-Reference: <[email protected]>

---
 kernel/sched.c |    1 -
 1 files changed, 0 insertions(+), 1 deletions(-)
kernel/sched.c +0 -1
index 456c990..297d1a0 100644
@@ -4128,7 +4128,6 @@ need_resched_nonpreemptible:
 		rq->nr_switches++;
 		rq->curr = next;
 		++*switch_count;
-		WARN_ON_ONCE(test_tsk_need_resched(next));
 
 		context_switch(rq, prev, next); /* unlocks the rq */
 		/*
#4 lore raw
In reply to Linus Torvalds
Quoted 7 lines · Linus Torvalds
Quoted 6 lines · Ingo Molnar
kernel/sched.c +1 -0
@@ -3943,6 +4128,7 @@ need_resched_nonpreemptible:
               rq->nr_switches++;
               rq->curr = next;
               ++*switch_count;
+               WARN_ON_ONCE(test_tsk_need_resched(next));
 
Quoted 7 lines

kern

Quoted 3 lines
kernel/sched.c +0 -0
               context_switch(rq, prev, next); /* unlocks the rq */
               /*
kern renamed from kernel/timer.c +0 -0

Please don't add like this in an -rc. It looks like it's some debugging aid, I don't think it should have gone in now.

Correct, I did ask Ingo to remove that line before he pushed it out to you. Anyway, it seems to be sorted now. Sorry for the bother.