[PATCH] scc fix for 2.6.0

[PATCH] scc fix for 2.6.0

[Jeroen Vreeken]

[-- Attachment #1: Type: text/plain, Size: 217 bytes --]

Hi,

This is a fix for the scc driver in 2.6.0
It was calling register_netdev from within a ioctl and thus holding
rtnl_lock()
It now uses register_netdevice and calls rtnl_lock() itself if outside the
ioctl.

Jeroen

[-- Attachment #2: scc-linux-2.6.0.rxq.diff --]
[-- Type: application/octet-stream, Size: 918 bytes --]

--- linux-2.6.0/drivers/net/hamradio/scc.c	2003-12-18 03:59:27.000000000 +0100
+++ linux-2.6.0.rxq/drivers/net/hamradio/scc.c	2003-12-25 17:57:58.000000000 +0100
@@ -163,6 +163,7 @@
 #include <linux/delay.h>
 #include <linux/skbuff.h>
 #include <linux/netdevice.h>
+#include <linux/rtnetlink.h>
 #include <linux/if_ether.h>
 #include <linux/if_arp.h>
 #include <linux/socket.h>
@@ -1526,7 +1527,7 @@
 	scc->dev = dev;
 	spin_lock_init(&scc->lock);
 
-	err = register_netdev(dev);
+	err = register_netdevice(dev);
 	if (err) {
 		printk(KERN_ERR "%s: can't register network device (%d)\n", 
 		       name, err);
@@ -2119,10 +2120,13 @@
 	
 	sprintf(devname,"%s0", SCC_DriverName);
 	
+	rtnl_lock();
 	if (scc_net_alloc(devname, SCC_Info)) {
+		rtnl_unlock();
 		printk(KERN_ERR "z8530drv: cannot initialize module\n");
 		return -EIO;
 	}
+	rtnl_unlock();
 
 	proc_net_fops_create("z8530drv", 0, &scc_net_seq_fops);
 

Re : [PATCH] scc fix for 2.6.0

[Jean-Luc Coulon (f5ibh)]

[-- Attachment #1: Type: text/plain, Size: 409 bytes --]

Hi,

I've tested this patch with 2.6.0 and a drsi card.
It works fine.

--
Regards and 73
			- f5ibh - Jean-Luc

Le 25.12.2003 19:52, Jeroen Vreeken a écrit :
|Hi,
|
|This is a fix for the scc driver in 2.6.0
|It was calling register_netdev from within a ioctl and thus holding
|rtnl_lock()
|It now uses register_netdevice and calls rtnl_lock() itself if
|outside the
|ioctl.
|
|Jeroen
|

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

Re: [PATCH] scc fix for 2.6.0

[Ralf Baechle DO1GRB]

On Thu, Dec 25, 2003 at 07:52:33PM +0100, Jeroen Vreeken wrote:

> This is a fix for the scc driver in 2.6.0
> It was calling register_netdev from within a ioctl and thus holding
> rtnl_lock()
> It now uses register_netdevice and calls rtnl_lock() itself if outside the
> ioctl.

Somehow the whole device registration in that driver looks suspect.
Registering from the driver's module_init() method is fine - but doing so
from ioctl looks more it's some kind of ancient workaround or so?

73 de DO1GRB op Ralf

--
Loc. JN47BS / CQ 14 / ITU 28 / DOK A21

Re: [PATCH] scc fix for 2.6.0

[Jeroen Vreeken]

On 2003.12.27 05:05:23 +0100 Ralf Baechle DO1GRB wrote:
> On Thu, Dec 25, 2003 at 07:52:33PM +0100, Jeroen Vreeken wrote:
> 
> > This is a fix for the scc driver in 2.6.0
> > It was calling register_netdev from within a ioctl and thus holding
> > rtnl_lock()
> > It now uses register_netdevice and calls rtnl_lock() itself if outside
> the
> > ioctl.
> 
> Somehow the whole device registration in that driver looks suspect.
> Registering from the driver's module_init() method is fine - but doing so
> from ioctl looks more it's some kind of ancient workaround or so?

At module init it only registers scc0.
sccinit later uses ioctls on this device to tell the driver how many chips
there really are, what kind of boards they are on, what crystal they have
and what addresses they use for the various registers.
At that time the ioctl will generate additional scc network devices if
necessary.
To my knowledge the only way to do this otherwise is to generate some kind
of scc specific /proc device... Which would be even more ugly.

The way scc works now has been like that since altleast the 2.0 kernels and
I never heard anybody come up with a better way.

Jeroen