Herbert Xu wrote:

On Thu, May 31, 2007 at 08:23:24AM -0700, Kok, Auke wrote:

quoted

quoted

e1000: 0000:01:02.0: e1000_probe: (PCI-X:66MHz:64-bit) 00:15:60:04:d7:f8
e1000: eth0: e1000_probe: Intel(R) PRO/1000 Network Connection
e1000: 0000:01:02.1: e1000_probe: (PCI-X:66MHz:64-bit) 00:15:60:04:d7:f9
e1000: eth1: e1000_probe: Intel(R) PRO/1000 Network Connection
e1000: 0000:15:02.0: e1000_probe: (PCI-X:66MHz:64-bit) 00:12:79:9e:b7:c4
e1000: eth2: e1000_probe: Intel(R) PRO/1000 Network Connection
e1000: 0000:15:02.1: e1000_probe: (PCI-X:66MHz:64-bit) 00:12:79:9e:b7:c5
e1000: eth3: e1000_probe: Intel(R) PRO/1000 Network Connection
e1000: Unexpected interrupt icr=0x4

that's indeed the link status change we fire manually to trigger the 
watchdog run:

#define E1000_ICR_LSC           0x00000004 /* Link Status Change */

This still makes no sense.  The only triggers I can find for this occur
after e1000_irq_enable.  So unless we've got a problem with memory
barriers we shouldn't get the above printk.

Is there another trigger that happens earlier?

none that I know....

Auke