On 30.12.2003, at 19:57, Dirk-Willem van Gulik wrote:

> First I hear of that - and would be interested in digging into it. Any
> pointers to a description ?

Basically, it seems that when pushing data at maximum speed (Ethernet 
-> Ethernet) through a net45xx that runs FreeBSD (m0n0wall actually), 
the CPU gets so fed up with handling all the interrupts (and packet 
filtering/NAT too) that the watchdogd process doesn't get any CPU time 
to tickle the watchdog anymore. Raising the priority of the process 
doesn't help either.

> I've used it both in wirelessleiden.nl and in commercial project and 
> not
> observed this; nor with 100Mbit ether or WiFi. Are we sure this is not 
> an
> artifact of polling mode and a HZ setting introduced at the same time ?

Au contraire - polling mode actually "solved" the problem by leaving 
watchdogd enough CPU time even at max. throughput.

> No trouble. Whichever is easiest; though given the way this code works 
> and
> given the way the HW of the Elan works - I seriously do dount that the
> watchdog is the cause - perhaps more a sympthom; a few cycles userland

Well, the watchdog and your watchdog daemon work as expected - 
resetting the box when it is "hung" (no more userland activity ;).

Greets,

Manuel