Re: 2.4.29, e100 and a WOL packet causes keventd going mad

[Marcelo Tosatti <marcelo.tosatti@cyclades.com>]

On Fri, Jan 28, 2005 at 05:48:11PM +0100, Michael Gernoth wrote:
> Hi,
> 
> we have about 70 P4 uniprocessor machines (some with Hyperthreading
> capable CPUs) running linux 2.4.29, which are woken up on the weekdays
> by sending a WOL packet to them. The machines all have a E100 nic with
> WOL enabled in the bios. The E100 driver is compiled into the kernel
> and not loaded as a module.
> 
> If the machine which should be woken up is already running (because
> someone switched it on by hand), the WOL packet causes keventd to go
> mad and "use" 100% CPU:
> 
>   PID USER      PR  NI  VIRT  RES  SHR S %CPU %MEM    TIME+  COMMAND
>     2 root      15   0     0    0    0 R 99.9  0.0 140:50.94 keventd

Probably a task event is rescheduling itself repeatedly? e100 does not seem 
to schedule_task() events directly, so I wonder what is going on.

Can you boot a machine with profile=2, then send the WOL packet causing
keventd to go mad and run:

readprofile | sort -nr +2 | head -20

After a few minutes.

Ganesh, Scott, Jeff, any ideas?

> This can be reproduced on any of the 70 machines by simply sending a WOL
> packet to it, when it's already running... No entry is made in the
> kernel log.
> 
> The dmesg of an affected machine can be found at:
> http://wwwcip.informatik.uni-erlangen.de/~simigern/cip-dmesg
> Our kernel-config is at:
> http://wwwcip.informatik.uni-erlangen.de/~simigern/cip-generic-config
> lspci -vvv is at:
> http://wwwcip.informatik.uni-erlangen.de/~simigern/cip-lspci
> 
> We are using a kernel.org linux 2.4.29 kernel patched with the current
> autofs patch and ACL support.