Re: memory alignment problems with -current on amd64? [Found Cause]

Mark Atkinson wrote:
> Mark Atkinson wrote:
>> With recent kernels on HAMMER/amd64 I cannot complete a buildworld.  The
>> compilation keeps failing with problems like:
[...]
>> /usr/src/gnu/usr.bin/binutils/as/../../../../contrib/binutils/gas/subsegs.c:205:
>> internal compiler error: Bus error: 10 Please submit a full bug report,
>> with preprocessed source if appropriate. See
>> <URL:http://gcc.gnu.org/bugs.html> for instructions. *** Error code 1 1
[...]
>> If I boot back to a kernel from sources Oct 15th 2008, I can complete a
>> buildworld on this machine no problem.
>> 
>> * This is a HP DL385G5 with 1 quad core AMD 2100 and 10G of memory.
>> * This the amd64 GENERIC kernel
>> * I've tried reducing hw.physmem to 2G, but that didn't make any
>> difference. * I will recieve bus errors when running buildworld w/ -j1
>> * If I run buildworld with a larger number the machine will reset w/ no
>> panic.
[...]
> Since I can go back to the Oct 15th kernel and do a complete buildworld, I
> can only assume -current has changed since then for amd64 sizing to maybe
> access something other than what the e820 method would return?

Well, taking the information I knew -- OCT 15th == good, Mid DEC == BAD,
I trolled every commit logged between.   Eventually I found this one:

http://svn.freebsd.org/viewvc/base?view=revision&revision=185715

http://docs.freebsd.org/cgi/mid.cgi?200812061937.mB6JbqAI003273

I set vm.pmap.pg_ps_enabled="0" in /boot/loader.conf, and
was able to complete buildworld and -j16 buildworld and -j8 buildkernel
no problem.

It appears superpage mapping causes alignment problems on this box.  
-- 
Mark Atkinson
atkin901_at_yahoo.com
(!wired)?(coffee++):(wired);