Re: NetBSD xen dom0 kernel panic

[Manuel Bouyer <bouyer%antioche.eu.org@localhost>]

On Tue, Mar 17, 2009 at 08:43:43PM -0400, Thor Lancelot Simon wrote:
> On Tue, Mar 17, 2009 at 11:44:22PM +0000, YAMAMOTO Takashi wrote:
> > 
> > inet_pfil_hook and inet6_pfil_hook are not expected to be called
> > from h/w interrupt context.  ie. BRIDGE_IPF is broken.
> > it should be fixed to use softint.
> 
> That's a general problem with the bridge code, no?  Because it runs from
> the hardware interrupt path, all kinds of nasty priority inversions and
> livelock conditions are possible, like if you bridge to a software
> interface like any of our network-in-network tunnels.

Couldn't we have a queue processed from software interrupt context like
it's done for upper protocols ? At a first look this could easily be done
in bridge_forward(). AFAIK pfil_run_hooks() is called from bridge_forward()
or bridge_enqueue() (itself called from bridge_output(), bridge_forward()
though bridge_broadcast(), or the stp code).

-- 
Manuel Bouyer <bouyer%antioche.eu.org@localhost>
     NetBSD: 26 ans d'experience feront toujours la difference
--