gdbserver multithread support broken in 5.2.1?

gdbserver multithread support broken in 5.2.1?

[Dan Kegel]

With gdb+dejagnu-weekly-CVS-20020709, remote debugging works fine on ppc405.

With gdb-5.2.1 built the same way, it doesn't seem to work, and I
get the error message
   GDB bug: target.c (target_signal_from_host): unrecognized real-time signal
from gdbserver.

The problem seems to be in gdbserver, as the new gdb works fine with the old
gdbserver.  When I debug a trivial 9-threaded program:
the breakpoint on main works fine, but a breakpoint on the first created
thread's main routine fails.  Here's the gdb log:

(gdb) target remote 10.0.1.1:1234
Remote debugging using 10.0.1.1:1234
0x30013ae8 in ?? ()
(gdb) break main
Breakpoint 1 at 0x10000a00: file atomicity_test.c, line 46.
(gdb) cont
Continuing.

Breakpoint 1, main (argc=1, argv=0x7ffffe84) at atomicity_test.c:46
46              int switches = 0;
(gdb) break my_thread_main
Breakpoint 2 at 0x10000964: file atomicity_test.c, line 33.
(gdb) cont
Continuing.

And here's the log on the remote system:

# gdbserver foo:1234 ./a.out
Process ./a.out created; pid = 452
Remote debugging from host 192.168.6.130
nthreads 8, nloops 100000, nrepeats 100
GDB bug: target.c (target_signal_from_host): unrecognized real-time signal

There wasn't much time between those two releases, so it shouldn't be
too hard to track the problem down.  I haven't looked at the
diffs yet; I was kind of hoping somebody would know what was up...

(In both cases, I had applied a patch to give an initial value for
solib-absolute-prefix, but it's hard to imagine that was causing a problem...)

Thanks for any suggestions!
- Dan

Re: gdbserver multithread support broken in 5.2.1?

[Daniel Jacobowitz]

On Fri, Jul 26, 2002 at 05:53:59PM -0700, Dan Kegel wrote:
> With gdb+dejagnu-weekly-CVS-20020709, remote debugging works fine on ppc405.
> 
> With gdb-5.2.1 built the same way, it doesn't seem to work, and I
> get the error message
>   GDB bug: target.c (target_signal_from_host): unrecognized real-time signal
> from gdbserver.

CVS branches strike again.  5.2.1 is a maintenance release only; it
contains some important bugfixes.  CVS-20020709 is a development
snapshot, from the main trunk.  It contains new features, like the
multithreaded gdbserver.  If you want a released version of GDB with
this support, you have to wait for 5.3 (or 6.0, whatever we settle
on...).  September was the plan, if I remember right.


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

Re: gdbserver multithread support broken in 5.2.1?

[Andrew Cagney]

> 
> CVS branches strike again.  5.2.1 is a maintenance release only; it
> contains some important bugfixes.  CVS-20020709 is a development
> snapshot, from the main trunk.  It contains new features, like the
> multithreaded gdbserver.  If you want a released version of GDB with
> this support, you have to wait for 5.3 (or 6.0, whatever we settle
> on...).  September was the plan, if I remember right.

I think it will be called 5.3 (if we work on 6.0 being multi-arch only). 
  Mutter something about HP/UX :-)

Andrew

Re: gdbserver multithread support broken in 5.2.1?

[Dan Kegel]

Andrew Cagney wrote:
>> CVS branches strike again.  5.2.1 is a maintenance release only; it
>> contains some important bugfixes.  CVS-20020709 is a development
>> snapshot, from the main trunk.  It contains new features, like the
>> multithreaded gdbserver.  If you want a released version of GDB with
>> this support, you have to wait for 5.3 (or 6.0, whatever we settle
>> on...).  September was the plan, if I remember right.
> 
> 
> I think it will be called 5.3 (if we work on 6.0 being multi-arch only). 
>  Mutter something about HP/UX :-)

Thanks for the explanation.  Guess I'll stick with that old weekly snapshot
for another six weeks then.

- Dan