MIPS64 N64 SIGBUS error

Mon Nov 25 12:11:21 UTC 2013

On 24 November 2013 10:51, Waldemar Brodkorb <wbx at openadk.org> wrote:
> Hi Developers,
>
> I am trying to find the reason for a SIGBUS error, when

which binutils? ld.gold or bfd?
thanks,

> using uClibc (git from yesterday) with Qemu 1.6.1
> (qemu-system-mips64).
>
> My uClibc config is attached. I am cross-compiling from x86_64
> Debian. To successfully startup the Linux system, I use a
> eglibc based toolchain and system and then using chroot to trigger
> the SIGBUS. Using O32 ABI with uClibc works fine. Using N64 ABI
> results in a SIGBUS error, but only when using a binary using shared
> libraries. A static binary works fine.
>
> GDB shows following backtrace:
> # gdb /hello
> GNU gdb (GDB) 7.6
> Copyright (C) 2013 Free Software Foundation, Inc.
> License GPLv3+: GNU GPL version 3 or later
> <http://gnu.org/licenses/gpl.html>
> This is free software: you are free to change and redistribute it.
> There is NO WARRANTY, to the extent permitted by law.  Type "show
> copying"
> and "show warranty" for details.
> This GDB was configured as "mips64-openadk-linux".
> For bug reporting instructions, please see:
> <http://www.gnu.org/software/gdb/bugs/>...
> Reading symbols from /hello...done.
> (gdb) run
> Starting program: /hello
> _dl_get_ready_to_run:450: Cool, ldso survived making function calls
> _dl_malloc:240: mmapping more memory
> _dl_ldsopath_init:156: Lib Loader: (0xfffffffff7fe4000)
> /lib/ld64-uClibc.so.0: using path: /lib
> _dl_get_ready_to_run:661: calling mprotect on the application
> program
> _dl_load_elf_shared_library:772: Found TLS header for /lib/libc.so.0
> _dl_load_elf_shared_library:799: Relocated TLS initial image from
> 0xe7ad0 to 0xfff7fdbad0 (size = 0x8)
> _dl_get_ready_to_run:1052: Loading: (0xfff7ef4000) /lib/libc.so.0
> _dl_get_ready_to_run:1052: Loading: (0xfff7fe4000)
> /lib/ld64-uClibc.so.0
> _dl_get_ready_to_run:1193: Calling init_tls()!
> _dl_malloc:240: mmapping more memory
> _dl_get_ready_to_run:1295: Beginning relocation fixups
> TLS_TPREL  : , 0x8, 0xffffffffffff9000
> TLS_TPREL  : , 0xc, 0xffffffffffff9000
> TLS_TPREL  : , 0x0, 0xffffffffffff9000
> TLS_TPREL  : , 0x10, 0xffffffffffff9000
> TLS_TPREL  : __libc_tsd_RPC_VARS, 0x0, 0xffffffffffff9018
> _dl_get_ready_to_run:1325: Calling _dl_allocate_tls_init()!
>
> Program received signal SIGBUS, Bus error.
> 0x000000fff7f63cfc in _stdio_init () at libc/stdio/_stdio.c:257
> 257     libc/stdio/_stdio.c: No such file or directory.
> (gdb) bt
> #0  0x000000fff7f63cfc in _stdio_init () at libc/stdio/_stdio.c:257
> #1  0x000000fff7fba700 in __uClibc_init ()
>     at libc/misc/internals/__uClibc_main.c:276
> #2  0x000000fff7fec9cc in _dl_get_ready_to_run (tpnt=0xfff7ffd2b0,
>     load_addr=<optimized out>, auxvt=0xffffffec58,
> envp=0xffffffedb8,
>     argv=<optimized out>) at ldso/ldso/ldso.c:1398
> #3  0x000000fff7fed598 in _dl_start (args=1099511623072)
>     at ldso/ldso/dl-startup.c:349
> #4  0x000000fff7fe592c in _start () from /lib/ld64-uClibc.so.0
> Backtrace stopped: frame did not save the PC
> (gdb)
>
> It is this line in the code:
> int old_errno = errno;
>
> And after preprocessing:
> int old_errno = __libc_errno;
>
> I have NPTL/TLS enabled in my config. So errno seems to be a per-thread
> value. When playing a little bit with gdb, breaking in _stdio_init
> I see following:
> (gdb) run
> Starting program: /hello
> _dl_get_ready_to_run:450: Cool, ldso survived making function calls
> _dl_malloc:240: mmapping more memory
> _dl_ldsopath_init:156: Lib Loader: (0xfffffffff7fe4000)
> /lib/ld64-uClibc.so.0: using path: /lib
> _dl_get_ready_to_run:661: calling mprotect on the application
> program
> _dl_load_elf_shared_library:772: Found TLS header for /lib/libc.so.0
> _dl_load_elf_shared_library:799: Relocated TLS initial image from
> 0xe7ad0 to 0xfff7fdbad0 (size = 0x8)
> _dl_get_ready_to_run:1053: Loading: (0xfff7ef4000) /lib/libc.so.0
> _dl_get_ready_to_run:1053: Loading: (0xfff7fe4000)
> /lib/ld64-uClibc.so.0
> _dl_get_ready_to_run:1194: Calling init_tls()!
> _dl_malloc:240: mmapping more memory
> _dl_get_ready_to_run:1296: Beginning relocation fixups
> TLS_TPREL  : , 0x8, 0xffffffffffff9000
> TLS_TPREL  : , 0xc, 0xffffffffffff9000
> TLS_TPREL  : , 0x0, 0xffffffffffff9000
> TLS_TPREL  : , 0x10, 0xffffffffffff9000
> TLS_TPREL  : __libc_tsd_RPC_VARS, 0x0, 0xffffffffffff9018
> _dl_get_ready_to_run:1326: Calling _dl_allocate_tls_init()!
>
> Breakpoint 1, _stdio_init () at libc/stdio/_stdio.c:257
> 257     libc/stdio/_stdio.c: No such file or directory.
> (gdb) si
> 0x000000fff7f63ce0      257     in libc/stdio/_stdio.c
> (gdb) disas 0x000000fff7f63ce0,+8
> Dump of assembler code from 0xfff7f63ce0 to 0xfff7f63ce8:
> => 0x000000fff7f63ce0 <_stdio_init+36>: move    v0,v1
>    0x000000fff7f63ce4 <_stdio_init+40>: ld      v1,-26904(gp)
> End of assembler dump.
> (gdb) p/x $gp-26904
> $1 = 0xfff7fddc58
> (gdb) x/g 0xfff7fddc58
> 0xfff7fddc58:   0xffff900000000008
> (gdb)
>
> Looks wrong or not?
> When I manually fix it, it survives:
>
> (gdb) set {unsigned long}0xfff7fddc58=0xffffffffffff9000
> (gdb) si
> 0x000000fff7f63ce4      257     in libc/stdio/_stdio.c
> (gdb) si
> 0x000000fff7f63ce8      257     in libc/stdio/_stdio.c
> (gdb) si
> 0x000000fff7f63cec      257     in libc/stdio/_stdio.c
> (gdb) si
> 0x000000fff7f63cf0      257     in libc/stdio/_stdio.c
> (gdb) si
> 259     in libc/stdio/_stdio.c
> (gdb)
>
> So who does set $gp-26904 to the wrong data?
> Can anybody help me to fix this? After that there seems a good
> chance to unbreak uClibc N64 support. There is another problem, when
> using clone systemcall, but I already have a fix for it.
>
> Thanks for any help,
>  Waldemar
>
>
> _______________________________________________
> uClibc mailing list
> uClibc at uclibc.org
> http://lists.busybox.net/mailman/listinfo/uclibc