Re: [PATCH] softlockup: remove timestamp checking from hung_task

[Frederic Weisbecker <fweisbec@gmail.com>]

On Sat, Feb 07, 2009 at 05:34:40PM +0100, Frederic Weisbecker wrote:
> On Sat, Feb 07, 2009 at 05:23:28PM +0100, Frederic Weisbecker wrote:
> > Hi Mandeep,
> > 
> > On Fri, Feb 06, 2009 at 03:37:47PM -0800, Mandeep Singh Baines wrote:
> > > Patch against tip/core/softlockup
> > > 
> > > ---
> > > 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>
> > 
> > 
> > Good idea.
> > BTW, why haven't you put your name on top of this file?
> > That would help those who will send patches knowing to whom they have to
> > route their mails.
> > 
> > I made some comments below about small things...
> > 
> > > ---
> > >  include/linux/sched.h |    1 -
> > >  kernel/fork.c         |    8 +++-----
> > >  kernel/hung_task.c    |   48 +++++++++---------------------------------------
> > >  3 files changed, 12 insertions(+), 45 deletions(-)
> > > 
> > > diff --git a/include/linux/sched.h b/include/linux/sched.h
> > > index 2a2811c..e0d723f 100644
> > > --- a/include/linux/sched.h
> > > +++ b/include/linux/sched.h
> > > @@ -1241,7 +1241,6 @@ struct task_struct {
> > >  #endif
> > >  #ifdef CONFIG_DETECT_HUNG_TASK
> > >  /* hung task detection */
> > > -	unsigned long last_switch_timestamp;
> > >  	unsigned long last_switch_count;
> > >  #endif
> > >  /* CPU-specific state of this task */
> > > diff --git a/kernel/fork.c b/kernel/fork.c
> > > index fb94442..bf582f7 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)
> > >  
> > >  	tsk->min_flt = tsk->maj_flt = 0;
> > >  	tsk->nvcsw = tsk->nivcsw = 0;
> > > +#ifdef CONFIG_DETECT_HUNG_TASK
> > > +	tsk->last_switch_count = tsk->nvcsw + tsk->nivcsw;
> > > +#endif
> > 
> > 
> > I think you can directly assign a zero here :-)
> > Or you want to let it as is to give some sense and explanation
> > about the role of this field?
> > Why not, I guess gcc will optimize it anyway.
> > 
> > 
> > >  	tsk->mm = NULL;
> > >  	tsk->active_mm = NULL;
> > > @@ -1041,11 +1044,6 @@ static struct task_struct *copy_process(unsigned long clone_flags,
> > >  
> > >  	p->default_timer_slack_ns = current->timer_slack_ns;
> > >  
> > > -#ifdef CONFIG_DETECT_HUNG_TASK
> > > -	p->last_switch_count = 0;
> > > -	p->last_switch_timestamp = 0;
> > > -#endif
> > > -
> > >  	task_io_accounting_init(&p->ioac);
> > >  	acct_clear_integrals(p);
> > >  
> > > diff --git a/kernel/hung_task.c b/kernel/hung_task.c
> > > index 3951a80..4a10756 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;
> > >   * Zero means infinite timeout - no checking done:
> > >   */
> > >  unsigned long __read_mostly sysctl_hung_task_timeout_secs = 120;
> > > -static unsigned long __read_mostly hung_task_poll_jiffies;
> > >  
> > >  unsigned long __read_mostly sysctl_hung_task_warnings = 10;
> > >  
> > > @@ -69,33 +68,17 @@ static struct notifier_block panic_block = {
> > >  	.notifier_call = hung_task_panic,
> > >  };
> > >  
> > > -/*
> > > - * Returns seconds, approximately.  We don't need nanosecond
> > > - * resolution, and we don't need to waste time with a big divide when
> > > - * 2^30ns == 1.074s.
> > > - */
> > > -static unsigned long get_timestamp(void)
> > > -{
> > > -	int this_cpu = raw_smp_processor_id();
> > > -
> > > -	return cpu_clock(this_cpu) >> 30LL;  /* 2^30 ~= 10^9 */
> > > -}
> > > -
> > > -static void check_hung_task(struct task_struct *t, unsigned long now,
> > > -			    unsigned long timeout)
> > > +static void check_hung_task(struct task_struct *t, unsigned long timeout)
> > >  {
> > >  	unsigned long switch_count = t->nvcsw + t->nivcsw;
> > >  
> > >  	if (t->flags & PF_FROZEN)
> > >  		return;
> > >  
> > > -	if (switch_count != t->last_switch_count || !t->last_switch_timestamp) {
> > > +	if (switch_count != t->last_switch_count) {
> > >  		t->last_switch_count = switch_count;
> > > -		t->last_switch_timestamp = now;
> > >  		return;
> > >  	}
> > 
> > 
> > 
> > What happens here if khungtaskd is scheduled in after tsk is inserted on the task_list
> > in copy_process() but before tsk has been scheduled once?
> > 
> > tsk->last_switch_count and  tsk->nvcsw + tsk->nivcsw will still be equal to zero right?
> > 
> > Perhaps you could add another check such as
> > 
> > if (!switch_count)
> > 	return;
> > 
> > 
> > > -	if ((long)(now - t->last_switch_timestamp) < timeout)
> > > -		return;
> > >  	if (!sysctl_hung_task_warnings)
> > >  		return;
> > >  	sysctl_hung_task_warnings--;
> > > @@ -111,7 +94,6 @@ static void check_hung_task(struct task_struct *t, unsigned long now,
> > >  	sched_show_task(t);
> > >  	__debug_show_held_locks(t);
> > >  
> > > -	t->last_switch_timestamp = now;
> > >  	touch_nmi_watchdog();
> > >  
> > >  	if (sysctl_hung_task_panic)
> > > @@ -145,7 +127,6 @@ static void check_hung_uninterruptible_tasks(unsigned long timeout)
> > >  {
> > >  	int max_count = sysctl_hung_task_check_count;
> > >  	int batch_count = HUNG_TASK_BATCHING;
> > > -	unsigned long now = get_timestamp();
> > >  	struct task_struct *g, *t;
> > >  
> > >  	/*
> > > @@ -168,19 +149,16 @@ static void check_hung_uninterruptible_tasks(unsigned long timeout)
> > >  		}
> > >  		/* use "==" to skip the TASK_KILLABLE tasks waiting on NFS */
> > >  		if (t->state == TASK_UNINTERRUPTIBLE)
> > > -			check_hung_task(t, now, timeout);
> > > +			check_hung_task(t, timeout);
> > >  	} while_each_thread(g, t);
> > >   unlock:
> > >  	rcu_read_unlock();
> > >  }
> > >  
> > > -static void update_poll_jiffies(void)
> > > +static unsigned long timeout_jiffies(unsigned long timeout)
> > >  {
> > >  	/* timeout of 0 will disable the watchdog */
> > > -	if (sysctl_hung_task_timeout_secs == 0)
> > > -		hung_task_poll_jiffies = MAX_SCHEDULE_TIMEOUT;
> > > -	else
> > > -		hung_task_poll_jiffies = sysctl_hung_task_timeout_secs * HZ / 2;
> > > +	return (timeout ? timeout * HZ : MAX_SCHEDULE_TIMEOUT);
> > >  }
> > >  
> > >  /*
> > > @@ -197,8 +175,6 @@ int proc_dohung_task_timeout_secs(struct ctl_table *table, int write,
> > >  	if (ret || !write)
> > >  		goto out;
> > >  
> > > -	update_poll_jiffies();
> > > -
> > >  	wake_up_process(watchdog_task);
> > 
> > 
> > I'm not sure what does this function now that you dropped update_poll_jiffies()
> > So if the user sets up a new timeout value, the only effect will be that khungtaskd will
> > be awakened?
> > 
> > But actually the /sys file doesn't seem to be set up.
> 
> 
> Oops, I should have grep on proc_dohung_task_timeout_secs which is set on kernel/sysctl.
> Sorry.
> But still, sysctl_hung_task_timeout_secs doesn't seem to be set :-)


And once again I'm wrong, I shoud read sysctl.c or at least read the sysfs documentation.
Sorry for the noise.

BTW, here is a small fixlet on top of your patch about what I commented concerning
the tasks than weren't yet scheduled once:

--
>From e7120e424b031978e482b5fe311d90916ffb8b7e Mon Sep 17 00:00:00 2001
From: Frederic Weisbecker <fweisbec@gmail.com>
Date: Sat, 7 Feb 2009 17:45:12 +0100
Subject: [PATCH] softlockup: ensure the task has been scheduled once

When we check if the task has been scheduled since the last scan, we might
have a race condition if the task has been inserted on the task list but not
yet scheduled once. So we just add a small check to ensure it has been switched
in at least one time to avoid false positive.

Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
---
 kernel/hung_task.c |    6 +++++-
 1 files changed, 5 insertions(+), 1 deletions(-)

diff --git a/kernel/hung_task.c b/kernel/hung_task.c
index 4a10756..7f57a71 100644
--- a/kernel/hung_task.c
+++ b/kernel/hung_task.c
@@ -72,7 +72,11 @@ static void check_hung_task(struct task_struct *t, unsigned long timeout)
 {
 	unsigned long switch_count = t->nvcsw + t->nivcsw;
 
-	if (t->flags & PF_FROZEN)
+	/*
+	 * Ensure the task is not frozen and that it has been scheduled
+	 * at least once.
+	 */
+	if (t->flags & PF_FROZEN || !switch_count)
 		return;
 
 	if (switch_count != t->last_switch_count) {
-- 
1.6.1