XEN3_DOM0 crashes hard - pckbc_attach code

[Jean-Yves Migeon <jean-yves.migeon%espci.fr@localhost>]

Hi list,

This mail is the continuation of a previous thread on port-xen:

http://mail-index.netbsd.org/port-xen/2008/06/18/msg003906.html

I am bringing it to current, in order to get some advices on how to 
handle this case.

Latest modification to pckbc, for XEN3_DOM0 kernel:

http://cvsweb.netbsd.org/bsdweb.cgi/src/sys/dev/ic/pckbc.c.diff?r1=1.44&r2=1.45

leads to some kind of interrupt handling exhaustion, and/or causes a 
triple fault (?), which results in a violent shutdown of the computer.

Using ddb to step through autoconf confirms that the issue appear just 
after calling bus_space_read() in pckbc_poll_data1(): it loops in ddb in 
ilevel 8, and eventually, machine shuts down after a few seconds and 
lots of outputs to console.

Any advice on how I could handle this issue? Setting higher value for 
<s> (and lowering delay() accordingly) for the loop does fix it; 
however, it looks like an ugly hack to me. Note that it only happens 
with XEN3_DOM0, not GENERIC. Adding option PCKBCDEBUG does not help me much.

Comments/advices/hints... are appreciated.

Cheers,

--
Jean-Yves Migeon
jean-yves.migeon%espci.fr@localhost