Re: [OpenOCD-user] Core files on embedded ARM?

[Paul F. <fer...@gm...>]

Hi!

On Tue, May 23, 2017 at 08:33:04AM +0200, Ladislav Laska wrote:
> I could dump memory, save registers and be done with it, but I doubt I
> can load all of that back into gdb to be as useful as before, ispecting
> stack, objects and current variables.

Something I was puzzled about myself few weeks ago, and I came to the
conclusion there's no working solution yet.

You can probably write a core file generator based on the memory dump
plus register contents, I haven't investigated this option but it's
likely not prohibitively hard.

Alternatively, you can later connect to the same target (live),
replace all of its memory and registers and this way I think the state
will be replicated enough for inspection.

> I didn't find them in the gdb repository, but they seem to have been
> merged, or at least not rejected.

Looks like they were merged indeed, but what they actually do is
quering remote's gdbserver process to read /proc/<pid>/maps via fileio
extension. So basically they assume gdbserver running on Linux to
work.

> I wonder if there is some support necessary from openocd? Or is there
> some way that openocd could generate a core file by itself?

You can probably imitate reading of the /proc/<pid>/maps file,
heh. There already exists some FileIO code so you should be able to
repurpose it. 

Sorry about not having a ready-made solution for you.

In case you're fighting some faults, this might come useful for
Cortex-M3 targets, but with some modifications it should be applicable
to Cortex-M0 too:

https://github.com/karlp/zypsnips/blob/master/armv7m-vecstate-zippe.gdb

HTH
-- 
Be free, use free (http://www.gnu.org/philosophy/free-sw.html) software!
mailto:fer...@gm...