gdbserver and multithreaded programs

gdbserver and multithreaded programs

[Ankur Sheth]

We tried out today's cvs on a simple remote multithreaded program on x86,
ppc405 and sh4.  

Results :
	x86 - worked fine.
	ppc405 - worked on single threaded program, but on multithreaded
program output zillions of 

			Ignoring packet error, continuing...
			Ignoring packet error, continuing...
		as soon as we tried to step thru main.

	sh4 - worked fine on single threaded program, but on multithreaded
program it output the following message.

			Program received signal SIG32, Real-time event 32.


Has anybody else tried this stuff yet?  

Regards,

ankur

Re: gdbserver and multithreaded programs

[Dan Kegel]

Ankur Sheth wrote:
> We tried out today's cvs on a simple remote multithreaded program on
> x86, ppc405 and sh4.  
> 
> Results :
> 	x86 - worked fine.
> 	ppc405 - worked on single threaded program, but on multithreaded
>       program output zillions of 
> 		Ignoring packet error, continuing...
> 		Ignoring packet error, continuing...
> 	as soon as we tried to step thru main.
> 
> 	sh4 - worked fine on single threaded program, but on
>       multithreaded program it output the following message.
> 		Program received signal SIG32, Real-time event 32.

I should mention we are running Linux 2.4.16 or so,
using gcc3.0.2 (gcc2.9.7 from Montavista had same problem on sh4),
and we configured once for target powerpc-unknown-linux-gnu
and in a different copy of the tree for sh-unknown-linux-gnu.

Thanks very much to Daniel J. and Montavista for contributing
these patches.  We're really looking forward to using them.

- Dan

Re: gdbserver and multithreaded programs

[Daniel Jacobowitz]

On Fri, Jun 14, 2002 at 07:06:35PM -0700, Ankur Sheth wrote:
> We tried out today's cvs on a simple remote multithreaded program on x86,
> ppc405 and sh4.  

Let's see what we can do for you.

> Results :
> 	x86 - worked fine.

OK...

> 	ppc405 - worked on single threaded program, but on multithreaded
> program output zillions of 
> 
> 			Ignoring packet error, continuing...
> 			Ignoring packet error, continuing...
> 		as soon as we tried to step thru main.
> 
> 	sh4 - worked fine on single threaded program, but on multithreaded
> program it output the following message.
> 
> 			Program received signal SIG32, Real-time event 32.
> 
> 
> Has anybody else tried this stuff yet?  

The "received signal SIG32" means that thread_db did not load.  Most
likely this means GDB did not find your libraries. On ppc405 I'd guess
that it found an incorrect copy of the libraries.  Are you setting
solib-absolute-prefix correctly?

I'm trying to come up with better ways to diagnose this sort of
problem, as the remote multithread debug is exceedingly fragile at
present.

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

RE: gdbserver and multithreaded programs

[Dan Kegel]

Daniel Jacobowitz wrote:
> >     ppc405 - worked on single threaded program, but on multithreaded
> > program output zillions of
> >
> >         Ignoring packet error, continuing...
> >         Ignoring packet error, continuing...
> >     as soon as we tried to step thru main.
> >
> >     sh4 - worked fine on single threaded program, but on
> >       multithreaded program it output the following message.
> >
> >         Program received signal SIG32, Real-time event 32.
>
> The "received signal SIG32" means that thread_db did not load.  Most
> likely this means GDB did not find your libraries. On ppc405 I'd guess
> that it found an incorrect copy of the libraries.  Are you setting
> solib-absolute-prefix correctly?

No, we weren't setting it at all.  Thanks for the tip; works much
better now on ppc405.  I did see a strange message
"Program received signal SIGTRAP, Trace/breakpoint trap."
a few times, but in general I'm able to step through a program
that creates lots of threads on ppc405 now.  Huzzah!
(Haven't tried sh4 again yet.)

Still not trivial to build and use in a cross-compiled environment
(what is?), but a good FAQ ought to help that.

Thanks for the fast help!
- Dan

Re: gdbserver and multithreaded programs

[Daniel Jacobowitz]

On Fri, Jun 14, 2002 at 11:16:11PM -0700, Dan Kegel wrote:
> Daniel Jacobowitz wrote:
> > >     ppc405 - worked on single threaded program, but on multithreaded
> > > program output zillions of
> > >
> > >         Ignoring packet error, continuing...
> > >         Ignoring packet error, continuing...
> > >     as soon as we tried to step thru main.
> > >
> > >     sh4 - worked fine on single threaded program, but on
> > >       multithreaded program it output the following message.
> > >
> > >         Program received signal SIG32, Real-time event 32.
> >
> > The "received signal SIG32" means that thread_db did not load.  Most
> > likely this means GDB did not find your libraries. On ppc405 I'd guess
> > that it found an incorrect copy of the libraries.  Are you setting
> > solib-absolute-prefix correctly?
> 
> No, we weren't setting it at all.  Thanks for the tip; works much
> better now on ppc405.  I did see a strange message
> "Program received signal SIGTRAP, Trace/breakpoint trap."
> a few times, but in general I'm able to step through a program
> that creates lots of threads on ppc405 now.  Huzzah!
> (Haven't tried sh4 again yet.)

Hmm, if you can make a simple program that reproduces the SIGTRAPs I
would greatly appreciate it.  I'm pretty sure I committed the right
copy; I spent a long time trying to kill those things.

> Still not trivial to build and use in a cross-compiled environment
> (what is?), but a good FAQ ought to help that.

Documentation, documentation.  And some more sanity checks.  I'm
working on it.

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