Crash in gdbserver on ppc405 - what vintage of libthread_db is recommended?

Crash in gdbserver on ppc405 - what vintage of libthread_db is recommended?

[Dan Kegel]

I tried gdbserver on the ppc405 on a large multithreaded C++ program
compiled with g++3.0.2 or so.  All I did was 'break main'
and 'cont', and gdbserver crashed.  So I updated to last week's
snapshot (ftp://sources.redhat.com/pub/gdb/snapshots/current/gdb+dejagnu-weekly-CVS-20020709.tar.bz2),
built gdbserver for debugging, and reproduced the problem while debugging
gdbserver.  Here's what popped up.  Looks like the call to
td_ta_thr_iter is bombing.  I'm using the libthread_db from Hard Hat 2.0,
dated about a year ago; perhaps I need to use a newer version?
Or is gdbserver itself misbehaving?

Program received signal SIGSEGV, Segmentation fault.
0x10008844 in linux_read_memory (memaddr=1275058749, myaddr=0x7fefeda8 "",
     len=2094805880) at ../../../src/gdb/gdbserver/linux-low.c:1153
1153        = (PTRACE_XFER_TYPE *) alloca (count * sizeof (PTRACE_XFER_TYPE));
(gdb) bt
#0  0x10008844 in linux_read_memory (memaddr=1275058749, myaddr=0x7fefeda8 "",
     len=2094805880) at ../../../src/gdb/gdbserver/linux-low.c:1153
#1  0x10005494 in read_inferior_memory (memaddr=1275058749, myaddr=0x7fefeda8 "",
     len=2094805880) at ../../../src/gdb/gdbserver/target.c:63
#2  0x10009740 in ps_pdread (ph=0x1001c850, addr=1275058749, buf=0x7fefeda8,
     size=2094805880) at ../../../src/gdb/gdbserver/proc-service.c:112
#3  0x0ffdd000 in td_thr_setgregs () from /lib/libthread_db.so.1
#4  0x0ffdd1c4 in td_thr_tsd () from /lib/libthread_db.so.1
#5  0x1000958c in thread_db_find_new_threads ()
     at ../../../src/gdb/gdbserver/thread-db.c:308
#6  0x10009650 in thread_db_init () at ../../../src/gdb/gdbserver/thread-db.c:334
#7  0x10008bb8 in linux_look_up_symbols () at ../../../src/gdb/gdbserver/linux-low.c:1226
#8  0x10003ef4 in handle_query (own_buf=0x1001eee8 "qSymbol::")
     at ../../../src/gdb/gdbserver/server.c:71
#9  0x100042e8 in main (argc=3, argv=0x7ffffe84)
     at ../../../src/gdb/gdbserver/server.c:192
#10 0x0feb0dbc in ngettext () from /lib/libc.so.6
(gdb)

Re: Crash in gdbserver on ppc405 - what vintage of libthread_db is recommended?

[Daniel Jacobowitz]

I'll reiterate - I have about 99% confidence that your crash comes from
mismatched host and target libraries.  This is far and away the most
frequent cause of problems, since libthread_db is not at all capable of
handling errors in mismatched symbol lookup.

I'm still trying to figure out ways to detect this.  I'm almost leaning
towards using libelf or a simple clone to take checksums of on-disk
.text sections...

On Sat, Jul 13, 2002 at 04:00:43PM -0700, Dan Kegel wrote:
> I tried gdbserver on the ppc405 on a large multithreaded C++ program
> compiled with g++3.0.2 or so.  All I did was 'break main'
> and 'cont', and gdbserver crashed.  So I updated to last week's
> snapshot (ftp://sources.redhat.com/pub/gdb/snapshots/current/gdb+dejagnu-weekly-CVS-20020709.tar.bz2),
> built gdbserver for debugging, and reproduced the problem while debugging
> gdbserver.  Here's what popped up.  Looks like the call to
> td_ta_thr_iter is bombing.  I'm using the libthread_db from Hard Hat 2.0,
> dated about a year ago; perhaps I need to use a newer version?
> Or is gdbserver itself misbehaving?
> 
> Program received signal SIGSEGV, Segmentation fault.
> 0x10008844 in linux_read_memory (memaddr=1275058749, myaddr=0x7fefeda8 "",
>      len=2094805880) at ../../../src/gdb/gdbserver/linux-low.c:1153
> 1153        = (PTRACE_XFER_TYPE *) alloca (count * sizeof (PTRACE_XFER_TYPE));
> (gdb) bt
> #0  0x10008844 in linux_read_memory (memaddr=1275058749, myaddr=0x7fefeda8 "",
>      len=2094805880) at ../../../src/gdb/gdbserver/linux-low.c:1153
> #1  0x10005494 in read_inferior_memory (memaddr=1275058749, myaddr=0x7fefeda8 "",
>      len=2094805880) at ../../../src/gdb/gdbserver/target.c:63
> #2  0x10009740 in ps_pdread (ph=0x1001c850, addr=1275058749, buf=0x7fefeda8,
>      size=2094805880) at ../../../src/gdb/gdbserver/proc-service.c:112
> #3  0x0ffdd000 in td_thr_setgregs () from /lib/libthread_db.so.1
> #4  0x0ffdd1c4 in td_thr_tsd () from /lib/libthread_db.so.1
> #5  0x1000958c in thread_db_find_new_threads ()
>      at ../../../src/gdb/gdbserver/thread-db.c:308
> #6  0x10009650 in thread_db_init () at ../../../src/gdb/gdbserver/thread-db.c:334
> #7  0x10008bb8 in linux_look_up_symbols () at ../../../src/gdb/gdbserver/linux-low.c:1226
> #8  0x10003ef4 in handle_query (own_buf=0x1001eee8 "qSymbol::")
>      at ../../../src/gdb/gdbserver/server.c:71
> #9  0x100042e8 in main (argc=3, argv=0x7ffffe84)
>      at ../../../src/gdb/gdbserver/server.c:192
> #10 0x0feb0dbc in ngettext () from /lib/libc.so.6
> (gdb)
> 

-- 
Daniel Jacobowitz                           Debian GNU/Linux Developer
MontaVista Software                         Carnegie Mellon University

Re: Crash in gdbserver on ppc405 - what vintage of libthread_db is recommended?

[Dan Kegel]

Right again.  Once I switched to
   set solib-absolute-prefix /opt/hardhat/devkit/ppc/405/target
as you suggested, everything was fine.

Good luck figuring out how to detect this kind of misconfiguration,
it would be helpful.
- Dan

Daniel Jacobowitz wrote:
> I'll reiterate - I have about 99% confidence that your crash comes from
> mismatched host and target libraries.  This is far and away the most
> frequent cause of problems, since libthread_db is not at all capable of
> handling errors in mismatched symbol lookup.
> 
> I'm still trying to figure out ways to detect this.  I'm almost leaning
> towards using libelf or a simple clone to take checksums of on-disk
> .text sections...
> 
> On Sat, Jul 13, 2002 at 04:00:43PM -0700, Dan Kegel wrote:
> 
>>I tried gdbserver on the ppc405 on a large multithreaded C++ program
>>compiled with g++3.0.2 or so.  All I did was 'break main'
>>and 'cont', and gdbserver crashed.  So I updated to last week's
>>snapshot (ftp://sources.redhat.com/pub/gdb/snapshots/current/gdb+dejagnu-weekly-CVS-20020709.tar.bz2),
>>built gdbserver for debugging, and reproduced the problem while debugging
>>gdbserver.  Here's what popped up.  Looks like the call to
>>td_ta_thr_iter is bombing.  I'm using the libthread_db from Hard Hat 2.0,
>>dated about a year ago; perhaps I need to use a newer version?
>>Or is gdbserver itself misbehaving?
>>
>>Program received signal SIGSEGV, Segmentation fault.
>>0x10008844 in linux_read_memory (memaddr=1275058749, myaddr=0x7fefeda8 "",
>>     len=2094805880) at ../../../src/gdb/gdbserver/linux-low.c:1153
>>1153        = (PTRACE_XFER_TYPE *) alloca (count * sizeof (PTRACE_XFER_TYPE));
>>(gdb) bt
>>#0  0x10008844 in linux_read_memory (memaddr=1275058749, myaddr=0x7fefeda8 "",
>>     len=2094805880) at ../../../src/gdb/gdbserver/linux-low.c:1153
>>#1  0x10005494 in read_inferior_memory (memaddr=1275058749, myaddr=0x7fefeda8 "",
>>     len=2094805880) at ../../../src/gdb/gdbserver/target.c:63
>>#2  0x10009740 in ps_pdread (ph=0x1001c850, addr=1275058749, buf=0x7fefeda8,
>>     size=2094805880) at ../../../src/gdb/gdbserver/proc-service.c:112
>>#3  0x0ffdd000 in td_thr_setgregs () from /lib/libthread_db.so.1
>>#4  0x0ffdd1c4 in td_thr_tsd () from /lib/libthread_db.so.1
>>#5  0x1000958c in thread_db_find_new_threads ()
>>     at ../../../src/gdb/gdbserver/thread-db.c:308
>>#6  0x10009650 in thread_db_init () at ../../../src/gdb/gdbserver/thread-db.c:334
>>#7  0x10008bb8 in linux_look_up_symbols () at ../../../src/gdb/gdbserver/linux-low.c:1226
>>#8  0x10003ef4 in handle_query (own_buf=0x1001eee8 "qSymbol::")
>>     at ../../../src/gdb/gdbserver/server.c:71
>>#9  0x100042e8 in main (argc=3, argv=0x7ffffe84)
>>     at ../../../src/gdb/gdbserver/server.c:192
>>#10 0x0feb0dbc in ngettext () from /lib/libc.so.6
>>(gdb)
>>
> 
>