Hi,
an update on this problem:
thanks to Chuck which sent me a small cache_zap() function (which invalidates
the whole cache) I've been able to narrow down a bit the problem:
- if I remplace only sun4_vcache_flush_segment() with cache_zap() gives me
  a stable kernel.
- if I remplace all sun4_vcache* but sun4_vcache_flush_segment() with
  cache_zap() I get the core dumps.

So it looks like the problem is related to sun4_vcache_flush_segment().
Then I started remplacing calls to cache_flush_segment() by cache_zap(),
and it seems that the only place where it matters is in me_alloc().
My next idea is to remplace the cache_flush_segment() is me_alloc() with
something else. I'd try to remplace is with a cache_flush(),
but I don't know how to get (va,len) in me_alloc(). Any idea ?

-- 
Manuel Bouyer <bouyer@antioche.eu.org>
--