is CARP still impossible on SMP in 6.2?

[dima]

> On Wed, Feb 28, 2007 at 06:08:21PM +0300, dima wrote:
> d> > On Wed, Feb 28, 2007 at 02:28:13PM +0300, dima wrote:
> d> > d> I'm trying to make CARP working on a 2-Xeon hardware. It worked OK on 5.5, but not on 6.2.
> d> > d> The result of compiling both
> d> > d> options  SMP
> d> > d> device   carp
> d> > d> into the kernel makes it crash in 'swi1: net' kernel thread on a random (usually first) use of the CARP facility.
> d> > 
> d> > Can you please provide a backtrace?
> d> 
> d> It's not that easy. The server becomes extremely unstable under these conditions.
> d> I tried to build the kernel with '-g' option and included 'dumpdev' in /etc/rc.conf. The server can't even make a memory dump: it either gets one more page fault or just stops responding to anything but Reset key in the beginning of the dump.
> d> It's strange since it rebooted well while the kernel was built without debugging symbols and a dumpdev. Any ideas how I can make a memory dump?
> 
> Why do you think the CARP is guilty in this unstableness? Are you triggering
> MASTER<->BACKUP events and this causes instability?

CARP was my only 3rd subject for investigation. The previous two were polling (which is why I want to upgrade from 5.x to 6.x) and ACPI. I tried to build different kernels but as I included CARP even in GENERIC+SMP I saw the same effect.
The server can sometimes hang on 'ifconfig' commands using CARP, but the easier way to reproduce the bug is to connect() to a socket on a CARP address. bind() can have the same effect -- I haven't tested that, but bind to INADDR_ANY works ok.