[Backgroud: several of us see random core dumps on SS1+ or IPC hardware,
 and this is not reproductible on SS2 or IPX, this loosk like a pmap problem ]

Hi,
Chuck's post about NEW_PIPE and cache issues gave me an idea:
I recompiled a kernel with the call to cpuinfo.cache_enable() in cpu_setup().
Well, the box is now much, much slower, but it looks stable (has run a
make -j2 kernel compile for the whole day without a core dump, where a stock
kernel would get one in the first hour).
Not a real fix of course, just a data point, which may or may not be relevant.

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