Re: [Linuxptp-users] phc offset explodes after a while

[Richard C. <ric...@gm...>]

On Thu, Jan 30, 2014 at 08:53:17AM +0000, Koehrer Mathias (ETAS/ESW5) wrote:

> Kernel 3.2.48-rt69 (RT_PREEMPT patch).
> I use the Intel Gigabit CT Desktop adapter (82574L).
> For this is use the latest e1000e driver (Version 2.5.4) and build it outside the kernel to have the PTP support.

...

> This works really fine for a while. However after a while the value of phc explodes dramatically.
> The value will be -35192325800 which looks like a kind of value overrun! 
> In hex this is 0xFFFFFFF7CE5FB958.

This sounds like a RT_PREEMPT issue. That patch makes spinlocks
preemptable, and the 82574 does need real locking, IIRC.

Can you reproduce the problem with RT_PREEMPT disabled?

If not, then you will probably have to find the guilty spinlock in the
driver and convert it to a raw spin lock.

Thanks,
Richard