arm-none-eabi-gdb and openocd: Malformed response to offset query, qOffsets?

I am attempting to use GDB to debug a Stellaris LM3S8962 Evaluation board using OpenOCD and the GNU ARM toolchain (installed with Ma开发者_StackOverflow中文版cPorts), whenever I set the remote target in GDB it always returns "Malfomred response to offset query, qOffsets". Any ideas on what could be going wrong? Is there anything that I am missing?

[bcochran@narada arm]$ arm-none-eabi-gdb
GNU gdb (GDB) 7.3
Copyright (C) 2011 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 "--host=x86_64-apple-darwin10.7.0 --target=arm-none-eabi".
For bug reporting instructions, please see:
<http://www.gnu.org/software/gdb/bugs/>.
(gdb) set remotebaud 115200
(gdb) set debug remote 1
(gdb) file ~/dev/eclipse_workspace/hello_world_arm/bin/main.axf
Reading symbols from /Users/bcochran/dev/eclipse_workspace/hello_world_arm/bin/main.axf...(no debugging symbols found)...done.
(gdb) target remote localhost:4444
Remote debugging using localhost:4444
Sending packet: $qSupported:qRelocInsn+#9a...putpkt: Junk: {{}~Open On
Nak
Sending packet: $qSupported:qRelocInsn+#9a...putpkt: Junk: Chip Debugger
> 
Ack
Packet received: qSupported:qRelocInsn+
Packet qSupported (supported-packets) is supported
...
Packet qAttached (query-attached) is supported
Sending packet: $qOffsets#4b...Ack
Packet received: qOffsets
Malformed response to offset query, qOffsets

Here is the openocd output... as soon as the malformed response comes across openocd drops the telnet connection...

[bcochran@narada bin]$ openocd -f ../openocd/luminary.cfg -f ../openocd/stellaris.cfg
Open On-Chip Debugger 0.6.0-dev-00014-g827057f (2011-08-09-22:02)
Licensed under GNU GPL v2
For bug reports, read
    http://openocd.berlios.de/doc/doxygen/bugs.html
Info : only one transport option; autoselect 'jtag'
500 kHz
Info : clock speed 500 kHz
Info : JTAG tap: lm3s.cpu tap/device found: 0x3ba00477 (mfg: 0x23b, part: 0xba00, ver: 0x3)
Info : lm3s.cpu: hardware has 6 breakpoints, 4 watchpoints
Info : accepting 'telnet' connection from 4444
Error: error during read: Connection reset by peer
Info : dropped 'telnet' connection

Here are the version outputs of my arm-none-eabi-* toolchain...

[bcochran@narada tcl]$ arm-none-eabi-gcc -v
Using built-in specs.
COLLECT_GCC=arm-none-eabi-gcc
COLLECT_LTO_WRAPPER=/opt/local/libexec/gcc/arm-none-eabi/4.6.1/lto-wrapper
Target: arm-none-eabi
Configured with: ../gcc-4.6.1/configure --prefix=/opt/local --target=arm-none-eabi --enable-languages=c,objc,c++,obj-c++ --infodir=/opt/local/share/info --mandir=/opt/local/share/man --datarootdir=/opt/local/share/arm-none-eabi-gcc --with-system-zlib --disable-nls --with-gmp=/opt/local --with-mpfr=/opt/local --with-mpc=/opt/local --enable-stage1-checking --enable-multilib --with-newlib --enable-interwork
Thread model: single
gcc version 4.6.1 (GCC)

[bcochran@narada tcl]$ arm-none-eabi-gdb -v
GNU gdb (GDB) 7.3
Copyright (C) 2011 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 "--host=x86_64-apple-darwin10.7.0 --target=arm-none-eabi".
For bug reporting instructions, please see:
<http://www.gnu.org/software/gdb/bugs/>.

I am able to compile using the toolchain, and flash the resultant .bin file using OpenOCD. I have been unable to find a solution to the "malformed response" issue just by searching the web.

Thanks in advance for any advice or assistance!

UPDATES

Thanks to answers from @turbo-j and @guy-sirton, I was able to progress a bit further... The most helpful thus far was that I was indeed using the wrong port (4444 instead of 3333) but now I am getting the following whether I add -c "init" -c "halt" -c "reset halt" to my openocd command string or not:

(gdb) target remote localhost:3333
Remote debugging using localhost:3333
Remote 'g' packet reply is too long:     0080004000000000040000220f0000002833405451332abc009600a4d2b86092c0c118c03040d6f0284dbb93204b40c2000000000c010020ffffffff550400000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000000020000001

(this is right after the qOffsets req/resp, which now passes)

On the OpenOCD side:

Info : accepting 'gdb' connection from 3333
Warn : acknowledgment received, but no packet pending
Info : dropped 'gdb' connection

With sometimes a undefined debug reason 6 - target needs reset on the OpenOCD console...not sure what is going on now but it seems closer to functioning

UPDATES 2

It seems if I do not load the file 'main.axf' or 'main.o' I do not run into the Remote 'g' packet reply is too long but I get no symbols... I've noticed other websites deal primarily with the .elf extension. What is the difference? I'm using the "Hello World" example from StellarisWare which it generates main.axf, main.bin (flash writes and runs fine), main.d, main.o from the make command. Something odd about my Makefile?

Here is a patch like what @athquad mentioned. See his answer for more information. Thanks to him for efficiently pointing me at the right spot, and great shame indeed for offering a patch without providing it. :-P

http://pastebin.com/rdFF2eZd

Patch was made against openocd commit 631b80fd0835055bb385314f569f589b99d7441d

To use: (./configure options depend on jtag hardware)

git clone git://git.code.sf.net/p/openocd/code openocd
cd openocd
patch -p1 < /path/to/patch/file
./bootstrap
./configure --enable-maintainer-mode --enable-ft2232_libftdi
nice make && sudo make install

You used the wrong port. Use target remote localhost 3333 for the GDB-to-OpenOCD connection. The port 4444 is intended for human interaction via a terminal and can be used in addition to a GDB connection.

If you don't want to compile OpenOCD, as tacos' & atquad's answer, you can use an older release of the GNU ARM tool chain (7.2.5 is fine for example). (0.4.0 and 0.5.0 windows precompiled are both giving this problem).

I have just encountered the same "'g' packet reply is too long" bug and traced it to its cause.

It seems that for a very long time GNU ARM tools have used an ARM model with 9 obsolete floating point registers. OpenOCD is wise to this and in response to GDB's register request supplies dummy values. However, recent eabi toolchains including GDB have come up to date and so your GDB no longer expects these registers - hence the message.

In the OpenOCD source I have patched armv7m_get_gdb_reg_list() in armv7m.c to make this work, but my patch isn't smart enough to know which GDB version is being used - it simply uses an #ifdef switch - so it would need a bit more work to integrate with the OpenOCD source code.

A patch could be available to anyone who wants.

I do a lot of remote debugging on ARM but with a different processor, OS, and toolchain :-) Nevertheless, here are some thoughts.

This suggestion:

openocd -f openocd.cfg -c "init" -c "halt" -c "reset halt"

Comes from here: http://michaldemin.wordpress.com/2010/02/22/part-2-debugging-with-gdb-and-openocd/

Otherwise:

• A mismatch in the remote gdb protocol between the remote and your gdb. You can try going to a different version of the toolchain. Google and find what works for others.

• A communication issue. Try lower baud rate? Make sure the serial port parameters are sound.