error : preempt_count 1

error : preempt_count 1

[Oliver Kowalke]

Hello,

after shutdown (kernel 2.5.15) I've got :

erro: halt[8635] exited with preempt_count 1

What does it mean?
so long,
Oliver

Re: error : preempt_count 1

[Robert Love]

On Sun, 2002-05-12 at 23:25, Oliver Kowalke wrote:

> after shutdown (kernel 2.5.15) I've got :
> 
> erro: halt[8635] exited with preempt_count 1
> 
> What does it mean?

Absolutely nothing bad.  It is a debugging check to catch bad code that
does funny things with locks.  Ideally, every program should call unlock
for each instance it called lock - balancing everything out and giving a
preempt_count of zero.

Some code in the kernel, knowing it is shutting down, does not bother to
drop any held locks and subsequently you see that message.

Since it is triggering false positives, I will remove it eventually. 
For now it is incredibly useful for catching real problems.  And the
above, while harmless, could be fixed for "cleanliness" concerns.

	Robert Love

Re: error : preempt_count 1

[Denis Vlasenko]

On 13 May 2002 14:18, Robert Love wrote:
> > erro: halt[8635] exited with preempt_count 1
> >
> > What does it mean?
>
> Absolutely nothing bad.  It is a debugging check to catch bad code that
> does funny things with locks.  Ideally, every program should call unlock
> for each instance it called lock - balancing everything out and giving a
> preempt_count of zero.

> Some code in the kernel, knowing it is shutting down, does not bother to
> drop any held locks and subsequently you see that message.

> Since it is triggering false positives, I will remove it eventually.

I'd say don't remove it, just omit the 'error:' part - this will
reduce panic mails on the subject.

> For now it is incredibly useful for catching real problems.  And the
> above, while harmless, could be fixed for "cleanliness" concerns.
--
vda

Re: error : preempt_count 1

[Robert Love]

On Tue, 2002-05-14 at 09:45, Denis Vlasenko wrote:

> On 13 May 2002 14:18, Robert Love wrote:
>
> > Absolutely nothing bad.  It is a debugging check to catch bad code that
> > does funny things with locks.  Ideally, every program should call unlock
> > for each instance it called lock - balancing everything out and giving a
> > preempt_count of zero.
> 
> > Some code in the kernel, knowing it is shutting down, does not bother to
> > drop any held locks and subsequently you see that message.
> 
> > Since it is triggering false positives, I will remove it eventually.
> 
> I'd say don't remove it, just omit the 'error:' part - this will
> reduce panic mails on the subject.
> 
> > For now it is incredibly useful for catching real problems.  And the
> > above, while harmless, could be fixed for "cleanliness" concerns.

Not a bad idea ;-)

For now this will hopefully curb the inquiries - I can still remove it
later when I am confident that the core code is sane and if people still
whine.

Thanks,

	Robert Love

--- linux-2.5.15/kernel/exit.c	Sun May  5 20:37:59 2002
+++ linux/kernel/exit.c	Tue May 14 09:22:52 2002
@@ -526,7 +526,7 @@
 	del_timer_sync(&tsk->real_timer);
 
 	if (unlikely(preempt_get_count()))
-		printk(KERN_ERR "error: %s[%d] exited with preempt_count %d\n",
+		printk(KERN_ERR "%s[%d] exited with preempt_count %d\n",
 				current->comm, current->pid,
 				preempt_get_count());