This is the mail archive of the
binutils@sources.redhat.com
mailing list for the binutils project.
GNU as 2.14 on IRIX 6: crashes with shared libs
- From: Rainer Orth <ro at TechFak dot Uni-Bielefeld dot DE>
- To: binutils at sources dot redhat dot com
- Date: Fri, 18 Jul 2003 19:44:00 +0200 (MEST)
- Subject: GNU as 2.14 on IRIX 6: crashes with shared libs
Preparing for the integration of my (currently separate) IRIX 6 O32
configuration for GCC into the generic IRIX 6 code, I tried to use gas 2.14
(and 2.14.90 from current CVS mainline) to bootstrap and regtest mainline
GCC for mips-sgi-irix6.5 (i.e. N32 and N64 ABIs). While the C tests run
mostly ok (only a few regressions here and there, which may be due to not
testing exactly the same version of GCC sources with native as and gas),
all execution tests involving shared libraries created with gas
(i.e. libstdc++.so and libg2c.so) fail: the resulting binaries SEGV very
early in startup. GDB cannot debug this at all, and neither using
rld.debug or setting _RLD_ARGS=-trace or whatever runtime debugging
facilities rld provides gave any output. Using SGIs latest dbx 7.3.4
reveals the following stacktrace, though:
Process 1396451 (hello) stopped on signal SIGBUS: Bus error (default) at [resolve_relocations:2847 ,0xfb66368]
Source (of /j7/mtibuild/v74/workarea/v7.4/rld/rld.c) not available for Process 1396451
(dbx) where
> 0 resolve_relocations(0xfbdd438, 0x14, 0x3c, 0xa8, 0x1, 0x4002f30, 0x5ffb2f30, 0x5ffb2f30) ["/j7/mtibuild/v74/workarea/v7.4/rld/rld.c":2847, 0xfb66368]
1 fix_all_defineds(0xfbdd438, 0x24, 0x3c, 0xa8, 0x1, 0x4002f30, 0x5ffb2f30, 0x5ffb2f30) ["/j7/mtibuild/v74/workarea/v7.4/rld/rld.c":3654, 0xfb66f2c]
2 handle_undefineds(0x3, 0x24, 0x3c, 0xa8, 0x1, 0x4002f30, 0x5ffb2f30, 0x5ffb2f30) ["/j7/mtibuild/v74/workarea/v7.4/rld/rld.c":2655, 0xfb660d0]
3 foreach_obj(0x3, 0xfb66020, 0x0, 0xa8, 0x1, 0x4002f30, 0x5ffb2f30, 0x5ffb2f30) ["/j7/mtibuild/v74/workarea/v7.4/rld/list.c":225, 0xfb61454]
4 do_initial_resolutions_and_run_inits(0x3, 0x24, 0x3c, 0xa8, 0x1, 0x4002f30, 0x5ffb2f30, 0x5ffb2f30) ["/j7/mtibuild/v74/workarea/v7.4/rld/rld_startup.c":1123, 0xfb705a4]
5 sgi_main(0x0, 0x1, 0x7ffb7f04, 0xa8, 0x1, 0x4002f30, 0x5ffb2f30, 0x5ffb2f30) ["/j7/mtibuild/v74/workarea/v7.4/rld/rld_startup.c":1659, 0xfb70cb0]
6 newmain() ["/j7/mtibuild/v74/workarea/v7.4/rld/rld_newmain.s":68, 0xfb60184]
Obviously, the runtime linker (rld) crashes early during startup (which
explains no output from rld.debug ;-). While crashing rld is clearly a bug
there (which I may well report to SGI), it looks like relocations are
seriously busted. Has anyone else seen this, or been able to successfully
use this combination?
Any insights welcome.
Rainer
-----------------------------------------------------------------------------
Rainer Orth, Faculty of Technology, Bielefeld University