linux-source debugging with kgdb-patch

Hello,

BACKGROUND:

I was keen to see kgdb running for purely academic reasons. Thus, I made a setup of 2 machines for source-level debugging of the linux-kernel. The procedure mentioned on the web-site
[ kgdb.sourceforge.net] has been adhered to. I was able to successfully configure the setup. Also, I decided to use "ddd" front-end on gdb [local m/c] for debugging the kgdb-patched kernel on the remote machine, which is the usual setup for such debugging-efforts.
The m/c to be debugged stops with the message "Waiting for
connection from remote gdb..." until the "target remote" command is run from the "gdb" prompt of "ddd", upon which the m/c to be debugged continues it's bootup till it shows the command-prompt.

PROBLEM:

I was interested in setting a break-point in start_kernel thru' "ddd" such that the boot-up of the m/c to be debugged could be analysed step-by-step remotely. Though, I am able to set the breakpoint in start_kernel(),
the commands "run" or "continue" on the "gdb" prompt, only throw up the following errors :

(gdb) info break
Num Type Disp Enb Address What
7 breakpoint keep y 0xc027e7f0 in start_kernel at
init/main.c:614
(gdb) run
warning: shared library handler failed to enable breakpoint
warning: Cannot insert breakpoint 7:
Cannot access memory at address 0xc027e7f0

QUESTION:

I very strongly suspect that this exercise follows a particular sequence of steps to get it right. Either I am missing some step or I am not following the "order". In either case, I would be glad to receive some help/comments on my academic endeavour to be able to remotely debug the kernel.

- Kindly let me know a solution

- TIA

- Shobhit Mathur

The "run" command should not be used. Kgdb/gdb debugging of the kernel is more like attaching to a running program than running it. When you use the run command you are telling gdb to, among other things, transfer conrol to the beginning of the program, which in the case of the kernel, is where the boot loader jumps on boot. In the kernel this code in is memory that is released to free memory on the way up.

If you make this mistake often I recommend the following gdb macro:

define hook-run
echo "run is not available for kernel debugging \n" p .
end

The "p ." will error out the command and snach you from the jaws of reboot/ fsck :)

This macro should be in .gdbinit file in the same directory as the vmlinux file.

George Anzinger   [email protected]
High-res-timers:  http://sourceforge.net/projects/high-res-timers/
Preemption patch: http://www.kernel.org/pub/linux/kernel/people/rml