segmentation faults on RELENG_6

[walt]

Andreas Killaitis wrote:
> Hi list,
>
> I have some problems with one of my FreeBSD installations. I am not
> shure if this the correct list, but I hope so.
>
> The scenario:
>
> I am running three virtual machines (VMware Fusion 1.1 on a Mac Pro).
> All of them use exactly the same hardware playground. Each of the
> machines is running with a different FreeBSD installation: RELENG_5,
> RELENG_6 and RELENG_7. There is another virtual machine with 8-CURRENT
> with a slightly different setup, but all of the VMs are using two
> virtual processors, 512 MB RAM and a 100 GB harddisk. The RELENG_5,
> RELENG_7 and the CURRENT machine are working like a charm, but the
> RELENG_6 maching is causing trouble. Sometimes some sort of processes
> abort with a segmentation fault:
>
> Mar 15 23:49:30 beastie6 kernel: pid 77039 (sed), uid 0: exited on
> signal 11 (core dumped)
> Mar 16 00:12:47 beastie6 kernel: pid 35211 (conftest), uid 0: exited on
> signal 12 (core dumped)
> Mar 16 00:15:50 beastie6 kernel: pid 61400 (tr), uid 0: exited on signal
> 11 (core dumped)
> Mar 16 01:37:13 beastie6 kernel: pid 62336 (cc1), uid 0: exited on
> signal 11 (core dumped)
> Mar 16 01:51:06 beastie6 kernel: pid 41313 (sed), uid 0: exited on
> signal 11 (core dumped)
> Mar 16 02:03:10 beastie6 kernel: pid 3826 (cc1plus), uid 0: exited on
> signal 11 (core dumped)
> Mar 16 09:40:33 beastie6 kernel: pid 6094 (sed), uid 0: exited on signal
> 11 (core dumped)
> Mar 16 11:19:56 beastie6 kernel: pid 94667 (xargs), uid 0: exited on
> signal 11 (core dumped)
> Mar 16 11:20:12 beastie6 kernel: pid 97386 (sed), uid 0: exited on
> signal 11 (core dumped)
> Mar 16 11:27:49 beastie6 kernel: pid 22796 (pkg_info), uid 0: exited on
> signal 11 (core dumped)
> Mar 16 11:48:09 beastie6 kernel: pid 97969 (cat), uid 0: exited on
> signal 11 (core dumped)
> Mar 16 12:05:44 beastie6 kernel: pid 16624 (bdftopcf), uid 0: exited on
> signal 11 (core dumped)
> Mar 16 13:15:04 beastie6 kernel: pid 49750 (sed), uid 0: exited on
> signal 11 (core dumped)
>
> All those segmentatiopn faults occurred building some ports.
>
>
> None of my FreeBSD boxes uses special CFLAG settings for buildworld or
> ports, they just use the default settings, with the exception of
>
> CPUTYPE?=pentiumpro
>
> Initially I tried CPUTYPE?=prescott, reverting it to pentium4, pentium3
> and now even to pentiumpro, but even pentiumpro doesn't solve the
> problem. The whone world is built using those very conservative
> settings. My last idea is to remove even this last "tuning" parameter
> CPUTYPE, but I dislike this, for it would mean to run a plain 386 box. :-(
>
>
> Any ideas?
>
> btw: the kernel is build using the SMP flag as the only modification.
> The SCHED_4BSD is used. The source tree is RELENG_6 from yesterday.

I'm having some problems with sh segfaulting on one machine, but the
hardware is so different from yours that this idea may be irrelevant
to your situation:

Starting on 2006/08/27 RELENG_6 updated gcc from 3.4.4 to 3.4.6, and
I have a hunch that my problem may be related to that.  Unfortunately,
my old machine is so slow it takes 3 days to build world/kernel -- and
it's my firewall, too, so I haven't tried both compilers yet.

Sounds like your machine is a wee bit faster, so maybe you could try
checking out sources for RELENG_6 -D2006/08/26 and try a couple of
world/kernel cycles and see if it helps?  I'd be interested to see.