[PATCH] 8250 PCI Serial driver not always disabling PCI device

[PATCH] 8250 PCI Serial driver not always disabling PCI device

[Rolf Eike Beer]

Hi,

I found that pciserial_remove_one() only disables the device if it finds 
private data. This looks suspicious to me. Patch attached.

Eike

Signed-off-by: Rolf Eike Beer <eike-kernel@sf-tec.de>

--- linux-2.6.13-rc3/drivers/serial/8250_pci.c	2005-07-13 06:46:46.000000000 
+0200
+++ linux-2.6.13-rc3-eike/drivers/serial/8250_pci.c	2005-07-20 
09:16:28.000000000 +0200
@@ -1778,10 +1778,10 @@ static void __devexit pciserial_remove_o
 		if (quirk->exit)
 			quirk->exit(dev);
 
-		pci_disable_device(dev);
-
 		kfree(priv);
 	}
+
+	pci_disable_device(dev);
 }
 
 static int pciserial_suspend_one(struct pci_dev *dev, pm_message_t state)

Re: [PATCH] 8250 PCI Serial driver not always disabling PCI device

[Russell King]

On Wed, Jul 20, 2005 at 09:27:05AM +0200, Rolf Eike Beer wrote:
> I found that pciserial_remove_one() only disables the device if it finds 
> private data. This looks suspicious to me. Patch attached.

Makes no difference.  If we don't have the private data set, then
we have errored out during the initialisation, and we disabled the
device on the way out.  Moreover, the exit function won't even be
called.

So we actually have the guarantee that there will always be private
data here.

-- 
Russell King
 Linux kernel    2.6 ARM Linux   - http://www.arm.linux.org.uk/
 maintainer of:  2.6 Serial core

Re: [PATCH] 8250 PCI Serial driver not always disabling PCI device

[Rolf Eike Beer]

[-- Attachment #1: Type: text/plain, Size: 767 bytes --]

Am Mittwoch, 20. Juli 2005 10:03 schrieb Russell King:
>On Wed, Jul 20, 2005 at 09:27:05AM +0200, Rolf Eike Beer wrote:
>> I found that pciserial_remove_one() only disables the device if it finds
>> private data. This looks suspicious to me. Patch attached.
>
>Makes no difference.  If we don't have the private data set, then
>we have errored out during the initialisation, and we disabled the
>device on the way out.  Moreover, the exit function won't even be
>called.
>
>So we actually have the guarantee that there will always be private
>data here.

In this case we should just kill the test if the private data is present. If 
something is seriously wrong this will oops and not hide the bug. This is 
done in other places the same way.

Eike

[-- Attachment #2: Type: application/pgp-signature, Size: 189 bytes --]