Recent CPUTYPE changes breaking kernel on Centrino

[Jon Noack]

Bartosz Fabianowski wrote:
> I posted this question a few days ago, but it got lost in a thread I fear.
>
> With the recent changes to bsd.cpu.mk, the setting "CPUTYPE=pentium-m"
> in make.conf now gets picked up and leads to GCC flags being set
> accordingly. Unfortunately, something gets enabled that the Pentium M
> (actually a P3 with some additional features) does not support. What
> happens then is that a newly built world works just fine, while the
> kernel does not even boot. Both the boot loader and the kernel itself
> cause instant reboots at startup.
>
> I'm wondering if somebody knowledgeable with the GCC flags for a P3
> would have a clue which flag or feature it is that could be triggering
> this.

I reported this same issue in October with my Athlon-XP.  Here was the
last message of that thread (includes -save-temps data):
http://lists.freebsd.org/pipermail/freebsd-current/2004-November/042127.html

It appears like setting the CPUTYPE results in gcc producing bad code on
my Athlon-XP.  If I build _just_ the loader with a good gcc and
CPUTYPE?=athlon-xp, everything works.  However, if I build the loader as
part of a buildworld with CPUTYPE?=athlon-xp, the loader is bad.  Thus,
the issue appears to be that setting CPUTYPE might produce a broken gcc
(perhaps only when using -march).

Jon