[PATCH] HVC: do not request the irq twice.

[PATCH] HVC: do not request the irq twice.

[Ian Campbell]

Otherwise:
  root@bogga:~# tty
  /dev/hvc0
  root@bogga:~# reboot


  Broadcast message from root@bogga (hvc0) (Thu Jan 22 17:51:20 2009):

  The system is going down for reboot NOW!
  INIT: Switching to runlevel: 6
  INIT: Sending processes the TERM signal
  hvc_open: request_irq failed with rc -16.

Signed-off-by: Ian Campbell <ian.campbell@citrix.com>
Cc: Milton Miller <miltonm@bga.com>
Cc: Christian Borntraeger <borntraeger@de.ibm.com>
Cc: Hendrik Brueckner <brueckner@linux.vnet.ibm.com>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Rusty Russell <rusty@rustcorp.com.au>
---
 drivers/char/hvc_irq.c |    4 ++++
 1 files changed, 4 insertions(+), 0 deletions(-)

diff --git a/drivers/char/hvc_irq.c b/drivers/char/hvc_irq.c
index 2623e17..8152a13 100644
--- a/drivers/char/hvc_irq.c
+++ b/drivers/char/hvc_irq.c
@@ -28,6 +28,10 @@ int notifier_add_irq(struct hvc_struct *hp, int irq)
 		hp->irq_requested = 0;
 		return 0;
 	}
+
+	if (hp->irq_requested)
+		return 0;
+
 	rc = request_irq(irq, hvc_handle_interrupt, IRQF_DISABLED,
 			   "hvc_console", hp);
 	if (!rc)
-- 
1.5.6.5

Re: [PATCH] HVC: do not request the irq twice.

[Christian Borntraeger]

Am Thursday 22 January 2009 18:56:08 schrieb Ian Campbell:
> --- a/drivers/char/hvc_irq.c
> +++ b/drivers/char/hvc_irq.c
> @@ -28,6 +28,10 @@ int notifier_add_irq(struct hvc_struct *hp, int irq)
>  		hp->irq_requested = 0;
>  		return 0;
>  	}
> +
> +	if (hp->irq_requested)
> +		return 0;
> +
>  	rc = request_irq(irq, hvc_handle_interrupt, IRQF_DISABLED,
>  			   "hvc_console", hp);
>  	if (!rc)

Unfortunately I have no system with an irq based hvc console - I cannot test 
this patch. Anyway, the patch looks sane:

Reviewed-by: Christian Borntraeger <borntraeger@de.ibm.com>

If this really fixes a problem, I am asking myself, if notifier_del can be 
called several times as well. If yes, it might be better to call free_irq at 
the last notifier_del - which means we need proper refcounting. Otherwise the 
console switches into polling mode....

Christian, blaming Rusty for using hvc_console in virtio_console which made me 
somehow responsible for this code...  ;-)

Re: [PATCH] HVC: do not request the irq twice.

[Ian Campbell]

On Thu, 2009-01-22 at 20:09 +0100, Christian Borntraeger wrote:
> Am Thursday 22 January 2009 18:56:08 schrieb Ian Campbell:
> > --- a/drivers/char/hvc_irq.c
> > +++ b/drivers/char/hvc_irq.c
> > @@ -28,6 +28,10 @@ int notifier_add_irq(struct hvc_struct *hp, int irq)
> >  		hp->irq_requested = 0;
> >  		return 0;
> >  	}
> > +
> > +	if (hp->irq_requested)
> > +		return 0;
> > +
> >  	rc = request_irq(irq, hvc_handle_interrupt, IRQF_DISABLED,
> >  			   "hvc_console", hp);
> >  	if (!rc)
> 
> Unfortunately I have no system with an irq based hvc console - I cannot test 
> this patch. Anyway, the patch looks sane:
> 
> Reviewed-by: Christian Borntraeger <borntraeger@de.ibm.com>
> 
> If this really fixes a problem, I am asking myself, if notifier_del can be 
> called several times as well. If yes, it might be better to call free_irq at 
> the last notifier_del - which means we need proper refcounting. Otherwise the 
> console switches into polling mode....

Hmm, interesting question. It looks like the count field in struct
hvc_struct is being used to count opens and closes which should
reference count the IRQ usage too.

Oh, I see, hvc_hangup resets count but the Xen dom0 hv_ops was missing
the hangup hook to release the IRQ -- I'll fix this.

I think this patch can be ignored, real bug is on the Xen side.

Ian.