"warning: many lost ticks." when restarting networking

[James Wilkinson]

Alex Greg wrote:
> When I run ifconfig or restart the networking on a couple of our
> machines, I sometimes get this error:
> 
> warning: many lost ticks.
> Your time source seems to be instable or some driver is hogging interupts

<snip>

> The machines are IBM x346 dual EM64T's with two Broadcom network ports
> on board and two Broadcom network cards - these are configured as two
> bonded pairs. They're running Fedora Core 3 x86_64.
> 
> What do these errors mean, and are they likely to cause a serious
> problem? Everything seems OK so far (I've pulled 20GB of data the
> network already, no issues).

Kernel-level problem. Do you have any kernel modules loaded that didn't
come from Fedora? If not (and you're up to date), you should bugzilla
it.

Given that this only seems to happen when you're playing with the
driver, I think the time source is unlikely to be the culprit.

What's happening is that computer hardware (like network cards)
sometimes need to get a CPU's attention. When this happens, they send
the CPU(s) an "interrupt". One of the CPUs will respond, and go to the
interrupt handler part of the kernel.

This might happen when the timer expires (I think it's currently at 1024
Hz), or when a network card receives a packet.

There I'm left handwaving a bit, not knowing too much about the drivers.
But when the driver has handled an interrupt, it's supposed to cancel
it. Sometimes (if the CPU has been handling another interrupt), there
may be a couple of interrupts to handle.

It sounds as though the driver has been cancelling the timer's
interrupts as well as its own.

James.

-- 
E-mail address: james | Just remember:
@westexe.demon.co.uk  | 1 virus         3 viriii
                      | 2 virii         4 viriv
                      |     -- Matt S Trout