Re: tulip (pnic) errors in 2.6.5-rc1

["Randy.Dunlap" <rddunlap@osdl.org>]

On Thu, 18 Mar 2004 18:39:54 +0100 Mikael Pettersson wrote:

| Randy.Dunlap writes:
|  > On Thu, 18 Mar 2004 17:29:39 +0100 Mikael Pettersson wrote:
|  > 
|  > | Jeff Garzik writes:
|  > |  > Mikael Pettersson wrote:
|  > |  > > Jeff Garzik writes:
|  > |  > >  > er, oops... lemme find the right patch...
|  > |  > > 
|  > |  > > No change, still a flood of those tulip_rx() interrupt messages.
|  > |  > 
|  > |  > hmmm.  Well, it is something unrelated to tulip driver, then.
|  > | 
|  > | Testing older -bk versions I've found that 2.6.4-bk2
|  > | is Ok but 2.6.4-bk3 has this message flood problem.
|  > 
|  > Other than the netdev_priv() changes, I see removal of
|  > KERNEL_SYSCALLS and the addition of CONFIG_NET_POLL_CONTROLLER.
|  > Are you enabling CONFIG_NET_POLL_CONTROLLER?
|  > If so, can you test with it disabled?
| 
| No it wasn't NET_POLL_CONTROLLER (I didn't enable it).
| 
| I split the bk2->bk3 patch in pieces, and traced the bug to the
| netdev_priv() change in in loopback.c. The bug is that loopback's
| dev is static and its ->priv actually points to kmalloc:d space.
| The netdev_priv() transformation is invalid for anything not
| allocated with alloc_etherdev or alloc_netdev, so the patch
| to loopback caused it to access and clobber static memory just
| after its own dev.
| 
| It was my bad luck that tulip was the victim of that clobber.
| 
| Reverting the patch below solves the problem.
| 
| /Mikael
| 
| diff -ruN linux-2.6.4-bk2/drivers/net/loopback.c linux-2.6.4-bk3/drivers/net/loopback.c
| --- linux-2.6.4-bk2/drivers/net/loopback.c	2004-03-11 14:01:28.000000000 +0100
| +++ linux-2.6.4-bk3/drivers/net/loopback.c	2004-03-18 16:12:28.000000000 +0100
| @@ -123,7 +123,7 @@
|   */
|  static int loopback_xmit(struct sk_buff *skb, struct net_device *dev)
|  {
| -	struct net_device_stats *stats = (struct net_device_stats *)dev->priv;
| +	struct net_device_stats *stats = netdev_priv(dev);
|  
|  	skb_orphan(skb);
|  

Thanks for checking, and sorry about the trouble.

BTW, the loopback patch has already been reverted.

--
~Randy