Segmentation violations in 'make check'

[Patrick Smith]

I'm trying to build glibc 2.3.1, and getting weird segmentation 
violations when running 'make check'.  So far, the following have 
produced segfaults:

        shm_open(...);  - when there is no tmpfs filesystem mounted
        fprintf(stderr, "sys_siglist[0] %p\n", sys_siglist[0]);
        dlerror();

My environment:

binutils  2.13.90.0.10
gcc       3.2
linux     2.4.19
glibc     2.3.1, with linuxthreads, and configured with

CC=gcc .../configure \
   --prefix=... \
   --libexecdir=... \
   --enable-add-ons \
   --with-headers=... \
   --without-gd \
   --enable-kernel=2.4.8 \
   --enable-static-nss

Running on a PowerPC (G4) (Mac Cube)


Details of the saga so far:

Initially, I was using binutils 2.13.

After getting around a couple of other problems (for which I've 
previously posted patches), 'make check' died in rt/tst-shm.  Upon 
investigation, there seemed to be a segmentation violation in the first 
call to shm_open in do_test.  In fact, if I inserted the code

   fprintf(stderr, "one\n");
   shm_open ("/shm-test", O_RDWR | O_CREAT | O_TRUNC | O_EXCL, 0600);
   fprintf(stderr, "two\n");

at the beginning of main(), the output would be just 'one' - followed by 
a segmentation violation.

Furthermore, the parent process (code in test-skeleton.c) was dying 
trying to report the problem.  This turned out to be a segmentation 
violation in strsignal().  And when I inserted this code

   fprintf(stderr, "sys_siglist %p\n", sys_siglist);
   fprintf(stderr, "sys_siglist[0] %p\n", sys_siglist[0]);

at the beginning of main, the output was

   sys_siglist 0x10011d8i

followed by a segmentation violation.

After a while, I realized that shm_open needs a tmpfs filesystem 
somewhere.  So I mounted one on /dev/shm.  Now tst-shm passed. 
(However, from my reading of the shm_open code, it should report an 
error if it can't find a suitable filesystem, not segfault.  So I'd 
think these problems have just been sidestepped, not solved.)

Then 'make check' died running elf/resolvfail.

At this point, I upgraded binutils to 2.13.90.0.10 (which required going 
from bison 1.50 to 1.75), rebuilt glibc, and tried 'make check' again. 
Segmentation violation in resolvfail, yet again.  This one seems to 
happen in the call to dlerror().

Does anyone have any ideas as to what might be wrong here?
--
address@hidden