Native preemption is the culprit [was Re: today's CURRENT lockups]

[Robert Watson]

On Mon, 12 Jul 2004, Johan Pettersson wrote:

> > Per my earlier post, I've experienced similar unreliability.  The work
> > around I'm using is to build with "#define PREEMPTION" from param.h
> > disabled.  This results in a quite usable kernel, although quite
> > probably more poor interrupt handling latency, etc.
> 
> Do you or John or any one else know whats can cause this? Is there any
> ongoing work to fix this? 

It's not currently clear what the cause is -- reports suggest that it's an
existing bug that's getting triggered by preemption, although I wouldn't
rule out a bug in the preemption implementation itself.  I'm currently
setting up a box with NMI to see if I can get some more useful
information, as serial break isn't cutting it (suggesting a leaked spin
lock or critical section).  I don't have an ETA, but I'm guessing that
we'll have some sort of resolution in the next day or two.  If not, I
guess we back out the change, or disable PREEMPTION by default.

Robert N M Watson             FreeBSD Core Team, TrustedBSD Projects
robert at fledge.watson.org      Principal Research Scientist, McAfee Research