Re: Spinlock recursion in hvc_poll

[Christian Borntraeger <borntraeger@de.ibm.com>]

Am Dienstag, 5. August 2008 schrieb Jeremy Fitzhardinge:
> Alex Nixon wrote:
> >>> Well I say fixed - it just means I can proceed to a spinlock recursion
> >>> BUG() 2 secs into the boot process, but it should be easier to track 
down
> >>> with printks and a coherent stack dump at my disposal.
> >>>   
> >>>       
> >> What's the backtrace on this?
> >>     
> >
> > I just turned DEBUG_SPINLOCKS back on to try catch this bug again, and it
> > seems to occur (very roughly) 1/10 of the time, with nothing changing 
between
> > runs.

Ok, this is a guess, I dont fully understand the backtrace.

request_irq tries to call the handler if the IRQ is shared. The irq handler 
calls hvc_poll and hvc_kill which might take the same spinlock. 
Can you test it this patch fixes the problem?

Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
---
 drivers/char/hvc_console.c |    3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

Index: linux-2.6/drivers/char/hvc_console.c
===================================================================
--- linux-2.6.orig/drivers/char/hvc_console.c
+++ linux-2.6/drivers/char/hvc_console.c
@@ -322,10 +322,11 @@ static int hvc_open(struct tty_struct *t
 
 	hp->tty = tty;
 
+	spin_unlock_irqrestore(&hp->lock, flags);
+
 	if (hp->ops->notifier_add)
 		rc = hp->ops->notifier_add(hp, hp->data);
 
-	spin_unlock_irqrestore(&hp->lock, flags);
 
 
 	/*