[PATCH] uio: fix devm_request_irq usage

[PATCH] uio: fix devm_request_irq usage

[Aaro Koskinen]

Commit e6789cd3dfb553077606ccafeb05e0043f072481 (uio: Simplify uio error
path by using devres functions) converted uio to use devm_request_irq().
This introduced a change in behaviour since the IRQ is associated with
the parent device instead of the created UIO device. The IRQ will remain
active after uio_unregister_device() is called, and some drivers will
crash because of this. The patch fixes this.

Signed-off-by: Aaro Koskinen <aaro.koskinen@nsn.com>
---
 drivers/uio/uio.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/uio/uio.c b/drivers/uio/uio.c
index f7beb6e..a673e5b 100644
--- a/drivers/uio/uio.c
+++ b/drivers/uio/uio.c
@@ -847,7 +847,7 @@ int __uio_register_device(struct module *owner,
 	info->uio_dev = idev;
 
 	if (info->irq && (info->irq != UIO_IRQ_CUSTOM)) {
-		ret = devm_request_irq(parent, info->irq, uio_interrupt,
+		ret = devm_request_irq(idev->dev, info->irq, uio_interrupt,
 				  info->irq_flags, info->name, idev);
 		if (ret)
 			goto err_request_irq;
-- 
1.8.5.1

Re: [PATCH] uio: fix devm_request_irq usage

[Greg Kroah-Hartman]

On Fri, Dec 20, 2013 at 04:19:47PM +0200, Aaro Koskinen wrote:
> Commit e6789cd3dfb553077606ccafeb05e0043f072481 (uio: Simplify uio error
> path by using devres functions) converted uio to use devm_request_irq().
> This introduced a change in behaviour since the IRQ is associated with
> the parent device instead of the created UIO device. The IRQ will remain
> active after uio_unregister_device() is called, and some drivers will
> crash because of this. The patch fixes this.

What drivers crash because of this?  Any in-kernel drivers?

thanks,

greg k-h

Re: [PATCH] uio: fix devm_request_irq usage

[Aaro Koskinen]

Hi,

On Fri, Dec 20, 2013 at 08:49:13AM -0800, Greg Kroah-Hartman wrote:
> On Fri, Dec 20, 2013 at 04:19:47PM +0200, Aaro Koskinen wrote:
> > Commit e6789cd3dfb553077606ccafeb05e0043f072481 (uio: Simplify uio error
> > path by using devres functions) converted uio to use devm_request_irq().
> > This introduced a change in behaviour since the IRQ is associated with
> > the parent device instead of the created UIO device. The IRQ will remain
> > active after uio_unregister_device() is called, and some drivers will
> > crash because of this. The patch fixes this.
> 
> What drivers crash because of this?  Any in-kernel drivers?

I saw a crash with Intel DPDK (http://www.dpdk.org/) igb_uio
driver. Basically, they do:

	uio_unregister_device
	pci_disable_msix <-- this will BUG() if there is an active IRQ

I cannot test any of the in-tree UIO drivers, but at least some
of them seem to release resources the IRQ handler might use after
uio_unregister_device(). So if the IRQ fires results would not probably
be very good.

A.

Re: [PATCH] uio: fix devm_request_irq usage

[Greg Kroah-Hartman]

On Fri, Dec 20, 2013 at 08:01:13PM +0200, Aaro Koskinen wrote:
> Hi,
> 
> On Fri, Dec 20, 2013 at 08:49:13AM -0800, Greg Kroah-Hartman wrote:
> > On Fri, Dec 20, 2013 at 04:19:47PM +0200, Aaro Koskinen wrote:
> > > Commit e6789cd3dfb553077606ccafeb05e0043f072481 (uio: Simplify uio error
> > > path by using devres functions) converted uio to use devm_request_irq().
> > > This introduced a change in behaviour since the IRQ is associated with
> > > the parent device instead of the created UIO device. The IRQ will remain
> > > active after uio_unregister_device() is called, and some drivers will
> > > crash because of this. The patch fixes this.
> > 
> > What drivers crash because of this?  Any in-kernel drivers?
> 
> I saw a crash with Intel DPDK (http://www.dpdk.org/) igb_uio
> driver. Basically, they do:
> 
> 	uio_unregister_device
> 	pci_disable_msix <-- this will BUG() if there is an active IRQ
> 
> I cannot test any of the in-tree UIO drivers, but at least some
> of them seem to release resources the IRQ handler might use after
> uio_unregister_device(). So if the IRQ fires results would not probably
> be very good.

Ok, thanks, I'll queue this up for 3.14-rc1 and tag it for 3.13-stable
so that people who have external modules can get the fix then.

greg k-h