[uClibc] uclibc gdb and threads

[Manuel Novoa III]

On Fri, Nov 05, 2004 at 02:07:02PM +0000, Andrew de Quincey wrote:
> Hi - I'm having slightly more progress here. I'm getting symbols now. Although 
> objdump still claims there aren't any - my guess is a crosscompiled objdump 
> has a bug.
> 
> I'm remote debugging using gdb/gdbserver 6.2.1, compiled with roughly these 
> instructions: http://www.kegel.com/linux/gdbserver.txt. 
> 
> Note I'm not using hardhat - as I said previously its gcc 3.3.4 from the 
> uclibc "toolchain" scripts. GDB is compiled against this version of uclibc.
> 
> First issue is that breakpoints dont seem to be working - I can set them, but 
> the execution never stops. However I don't need those right now, although 
> they would be *useful*.
> 
> The main concern for me is now threaded debugging. I keep getting "Program 
> received signal SIG32, Real-time event 32." messages in GDB.
> 
> I've seen these before, and generally its caused by problems with 
> libthread_db.so.1 - but that library is definitely resident on the target 
> machine. 
> 
> If I run crosscompile gdb and run it on the target machine, I can see it 
> claims to load libthread_db.so.1 ok:
> 
> This GDB was configured as "ppc-405-linux"...Using host libthread_db library 
> "/mnt/core/lib/libthread_db.so.1".
> 
> However when I actually run the program, I still get "Program received signal 
> SIG32, Real-time event 32.". In fact, I get about 3 of those, followed by:
> 
> warning: Cannot initialize thread debugging library: generic error
> warning: Cannot initialize thread debugging library: generic error
> warning: Cannot initialize thread debugging library: generic error
> warning: Cannot initialize thread debugging library: generic error
> warning: Cannot initialize thread debugging library: generic error
> warning: Cannot initialize thread debugging library: generic error
> This GDB was configured as "ppc-405-linux"...Using host libthread_db library 
> "/mnt/core/lib/libthread_db.so.1".
> 
> I don't see how the libthread_db can be returning an error - I looked at the 
> code, and the init function just returns "OK" always.

There are patches in buildroot for several gdb versions (including 6.2.1)
to configure for <arch>-linux-uclibc.  Remote debugging of threaded apps
works on mips.  Haven't tried powerpc though.

Manuel