Random Illegal instructions and Segmentation faults

[jaycemil]

I'm out of ideas. I've installed a stage 3 Gentoo setup on a Sparcstation 10 (Sun4M), completely compiled the kernel, the system boots and runs just fine. I can loop on any random command, uname for example, and it runs with no errors for hours. However, if try to emerge anything, or for that matter, even run configure for anything, I get random Illegal instruction and Segmentation faults. They don't always happen in the same places in the configure which would point to a hardware problem or overheating but I can beat the crap out of the machine with stuff that is already compiled and it works just fine plus the kernel compiled just fine as well. I guess my next step would be to completely rebuild the machine with a gcc 2.95 install given that that is what the kernel compiles with but I'd really like to not have to do that without some idea if that's the problem Has anyone else encountered this on a 32-bit sparc system?

-------------------------------------------------------------------------

prompt> ./configure --prefix=/usr --host=sparc-unknown-linux-gnu --mandir=/usr/share/man --infodir=/usr/share/info --datadir=/usr/share --sysconfdir=/etc --localstatedir=/var/lib --with-posix-regex --without-readlib--disable-etags --enable-tmpdir=/tmp
configure: WARNING: If you wanted to set the --build type, don't use --host.
If a cross compiler is detected then cross compile mode will be used.
Exuberant Ctags, version 5.5.4
Linux 2.4.30-sparc #3 Sat Apr 30 09:46:50 MDT 2005 sparc
checking whether to install link to etags... no
checking whether to install readtags object file... no
checking for sparc-unknown-linux-gnu-gcc... sparc-unknown-linux-gnu-gcc
checking for C compiler default output... a.out
checking whether the C compiler works... yes
checking whether we are cross compiling... no
checking for suffix of executables... ./configure: line 1890: 6616 Illegal instruction rm -f conftest$ac_cv_exeext

checking for suffix of object files... o
checking whether we are using the GNU C compiler... ./configure: line 1994: 6634 Segmentation fault cat >conftest.$ac_ext <<_ACEOF
#line $LINENO "configure"
/* confdefs.h. */
_ACEOF

yes
checking whether sparc-unknown-linux-gnu-gcc accepts -g... yes
checking for sparc-unknown-linux-gnu-gcc option to accept ANSI C...

[labrador]

I've seen this sort of behaviour twice.

Once with hardware issues. emerge is very intensive on a system. I had run Redhat 7
on a x86 box for 3 years with no problems. I couldn't build from stage 1 on it. When I
backed off the memory settings in CMOS to normal from strong, then
the system could build, and it is still running strong now. Faulty memory
would also flake out the way you describe. Try the bootprom system
diagnostic for your Sun box. You'd need to refer to Sun docs for your
version of bootprom to see what commands are available. If you
do full diagnostic it will require a floppy disk.

If the errors always bomb out the same way each time, it could be another problem
I've seen. ReiserFS on a Sun 2.6 kernel is bad medicine. I saw errors
that way during emerge that no one else in the world was experiencing.
However in that case they were related to accessing the
file system, and so the errrors were seemingly random, but they
were reproducible for a particular release of ebuild.

[jaycemil]

I did run the BootProm full diagnostic to no avail. Everything came clean and, like I said, the whole bootstrap works fine too which is pretty intensive. When it dies trying to emerge, the failures are at random locations. I can rerun the same command again and it fails in different spots. It is the wierdest behavior. It still smells like a hardware problem but there is definitely tied to a software trigger as well.