PATCH] net: b44.c fix sleeping-with-spinlock-helt during resume

PATCH] net: b44.c fix sleeping-with-spinlock-helt during resume

[Arjan van de Ven]

From: Arjan van de Ven <arjan@linux.intel.com>
Subject: [PATCH] net: b44.c fix sleeping-with-spinlock-helt during resume
CC: Michael Buesch <mb@bu3sch.de>

The b44.c driver calls b44_chip_reset() from the resume path,
with the device spinlock held.
Unfortunately, b44_chip_reset() calls ssb_pcicore_dev_irqvecs_enable()
which is a sleeping function (and calls might_sleep), if and only if
the device hasn't been enabled yet.

Not having this hardware, the safest solution seems to be to enable
the irqvec before taking the spinlock...

http://www.kerneloops.org/search.php?search=ssb_pcicore_dev_irqvecs_enable

Reported-by: www.kerneloops.org
Signed-off-by: Arjan van de Ven
---
 drivers/net/b44.c |   12 ++++++++++--
 1 files changed, 10 insertions(+), 2 deletions(-)

diff --git a/drivers/net/b44.c b/drivers/net/b44.c
index 59dce6a..6028129 100644
--- a/drivers/net/b44.c
+++ b/drivers/net/b44.c
@@ -1278,6 +1278,7 @@ static void b44_chip_reset(struct b44 *bp, int reset_kind)
 		bw32(bp, B44_DMARX_CTRL, 0);
 		bp->rx_prod = bp->rx_cons = 0;
 	} else
+		/* this function sleeps! */
 		ssb_pcicore_dev_irqvecs_enable(&sdev->bus->pcicore, sdev);
 
 	ssb_device_enable(bp->sdev, 0);
@@ -1374,7 +1375,7 @@ static int b44_set_mac_addr(struct net_device *dev, void *p)
 }
 
 /* Called at device open time to get the chip ready for
- * packet processing.  Invoked with bp->lock held.
+ * packet processing.  Sometimes invoked with bp->lock held.
  */
 static void __b44_set_rx_mode(struct net_device *);
 static void b44_init_hw(struct b44 *bp, int reset_kind)
@@ -2288,6 +2289,13 @@ static int b44_resume(struct ssb_device *sdev)
 		return rc;
 	}
 
+	/*
+	 * If the device isn't enabled yet, we need to do so before taking
+	 * the spinlock, since the enablinging function sleeps.
+	 */
+	if (!ssb_device_is_enabled(bp->sdev))
+		ssb_pcicore_dev_irqvecs_enable(&sdev->bus->pcicore, sdev);
+
 	spin_lock_irq(&bp->lock);
 
 	b44_init_rings(bp);
-- 
1.5.5.1



-- 
If you want to reach me at my work email, use arjan@linux.intel.com
For development, discussion and tips for power savings, 
visit http://www.lesswatts.org

Re: PATCH] net: b44.c fix sleeping-with-spinlock-helt during resume

[Michael Buesch]

On Sunday 01 June 2008 05:11:30 Arjan van de Ven wrote:
> From: Arjan van de Ven <arjan@linux.intel.com>
> Subject: [PATCH] net: b44.c fix sleeping-with-spinlock-helt during resume
> CC: Michael Buesch <mb@bu3sch.de>
> 
> The b44.c driver calls b44_chip_reset() from the resume path,
> with the device spinlock held.
> Unfortunately, b44_chip_reset() calls ssb_pcicore_dev_irqvecs_enable()
> which is a sleeping function (and calls might_sleep), if and only if
> the device hasn't been enabled yet.
> 
> Not having this hardware, the safest solution seems to be to enable
> the irqvec before taking the spinlock...
> 
> http://www.kerneloops.org/search.php?search=ssb_pcicore_dev_irqvecs_enable
> 
> Reported-by: www.kerneloops.org
> Signed-off-by: Arjan van de Ven
> ---
>  drivers/net/b44.c |   12 ++++++++++--
>  1 files changed, 10 insertions(+), 2 deletions(-)
> 
> diff --git a/drivers/net/b44.c b/drivers/net/b44.c
> index 59dce6a..6028129 100644
> --- a/drivers/net/b44.c
> +++ b/drivers/net/b44.c
> @@ -1278,6 +1278,7 @@ static void b44_chip_reset(struct b44 *bp, int reset_kind)
>  		bw32(bp, B44_DMARX_CTRL, 0);
>  		bp->rx_prod = bp->rx_cons = 0;
>  	} else
> +		/* this function sleeps! */
>  		ssb_pcicore_dev_irqvecs_enable(&sdev->bus->pcicore, sdev);
>  
>  	ssb_device_enable(bp->sdev, 0);
> @@ -1374,7 +1375,7 @@ static int b44_set_mac_addr(struct net_device *dev, void *p)
>  }
>  
>  /* Called at device open time to get the chip ready for
> - * packet processing.  Invoked with bp->lock held.
> + * packet processing.  Sometimes invoked with bp->lock held.
>   */
>  static void __b44_set_rx_mode(struct net_device *);
>  static void b44_init_hw(struct b44 *bp, int reset_kind)
> @@ -2288,6 +2289,13 @@ static int b44_resume(struct ssb_device *sdev)
>  		return rc;
>  	}
>  
> +	/*
> +	 * If the device isn't enabled yet, we need to do so before taking
> +	 * the spinlock, since the enablinging function sleeps.
> +	 */
> +	if (!ssb_device_is_enabled(bp->sdev))
> +		ssb_pcicore_dev_irqvecs_enable(&sdev->bus->pcicore, sdev);
> +
>  	spin_lock_irq(&bp->lock);
>  
>  	b44_init_rings(bp);

I'm not sure how this is supposed to fix the bug.
I think you will call ssb_pcicore_dev_irqvecs_enable() twice this way.
The call in b44_chip_reset() will still trigger.

Besides that, this only is a theoretical issue, as the function does
only sleep on a PCI-E device. But as b44 PCI-E devices don't exist,
it will never sleep. So I'd rather replace the might_sleep() with a
might_sleep_if() call.
Something like the following:

Index: wireless-testing/drivers/ssb/driver_pcicore.c
===================================================================
--- wireless-testing.orig/drivers/ssb/driver_pcicore.c	2008-04-23 16:06:56.000000000 +0200
+++ wireless-testing/drivers/ssb/driver_pcicore.c	2008-06-01 12:02:33.000000000 +0200
@@ -537,12 +537,12 @@ int ssb_pcicore_dev_irqvecs_enable(struc
 	int err = 0;
 	u32 tmp;
 
-	might_sleep();
-
 	if (!pdev)
 		goto out;
 	bus = pdev->bus;
 
+	might_sleep_if(pdev->id.coreid != SSB_DEV_PCI);
+
 	/* Enable interrupts for this device. */
 	if (bus->host_pci &&
 	    ((pdev->id.revision >= 6) || (pdev->id.coreid == SSB_DEV_PCIE))) {



-- 
Greetings Michael.

Re: PATCH] net: b44.c fix sleeping-with-spinlock-helt during resume

[Arjan van de Ven]

On Sun, 1 Jun 2008 12:03:40 +0200
Michael Buesch <mb@bu3sch.de> wrote:

> Index: wireless-testing/drivers/ssb/driver_pcicore.c
> ===================================================================
> --- wireless-testing.orig/drivers/ssb/driver_pcicore.c
> 2008-04-23 16:06:56.000000000 +0200 +++
> wireless-testing/drivers/ssb/driver_pcicore.c	2008-06-01
> 12:02:33.000000000 +0200 @@ -537,12 +537,12 @@ int
> ssb_pcicore_dev_irqvecs_enable(struc int err = 0; u32 tmp;
>  
> -	might_sleep();
> -
>  	if (!pdev)
>  		goto out;
>  	bus = pdev->bus;
>  
> +	might_sleep_if(pdev->id.coreid != SSB_DEV_PCI);
> +
>  	/* Enable interrupts for this device. */
>  	if (bus->host_pci &&
>  	    ((pdev->id.revision >= 6) || (pdev->id.coreid ==
> SSB_DEV_PCIE))) {
> 
> 
> 

fair enough.... can we get something like this into 2.6.26 still? Quite
a few people are hitting this (even though the warning is harmless, it
still is better to silence it)

-- 
If you want to reach me at my work email, use arjan@linux.intel.com
For development, discussion and tips for power savings, 
visit http://www.lesswatts.org

Re: PATCH] net: b44.c fix sleeping-with-spinlock-helt during resume

[Willy Tarreau]

Hi Arjan, Michael,

On Sun, Jun 01, 2008 at 09:01:07AM -0700, Arjan van de Ven wrote:
> On Sun, 1 Jun 2008 12:03:40 +0200
> Michael Buesch <mb@bu3sch.de> wrote:
> 
> > Index: wireless-testing/drivers/ssb/driver_pcicore.c
> > ===================================================================
> > --- wireless-testing.orig/drivers/ssb/driver_pcicore.c
> > 2008-04-23 16:06:56.000000000 +0200 +++
> > wireless-testing/drivers/ssb/driver_pcicore.c	2008-06-01
> > 12:02:33.000000000 +0200 @@ -537,12 +537,12 @@ int
> > ssb_pcicore_dev_irqvecs_enable(struc int err = 0; u32 tmp;
> >  
> > -	might_sleep();
> > -
> >  	if (!pdev)
> >  		goto out;
> >  	bus = pdev->bus;
> >  
> > +	might_sleep_if(pdev->id.coreid != SSB_DEV_PCI);
> > +
> >  	/* Enable interrupts for this device. */
> >  	if (bus->host_pci &&
> >  	    ((pdev->id.revision >= 6) || (pdev->id.coreid ==
> > SSB_DEV_PCIE))) {
> > 
> > 
> > 
> 
> fair enough.... can we get something like this into 2.6.26 still? Quite
> a few people are hitting this (even though the warning is harmless, it
> still is better to silence it)

Care to CC stable when people are already hitting the problem ?

FWIW, the fix was merged in mainline as commit
  a3bafeedfff2ac5fa0a316bea4570e27900b6fcc

thanks!
willy