On Fri, Oct 17, 2008 at 12:18 PM, Duane Ellis <[EMAIL PROTECTED]> wrote:
> Øyvind Harboe wrote:
>>
>> Currently each target gets it's own GDB server.
>>
>> However, I'm thinking that it would be better in most cases to
>> have each CPU represented as a thread in GDB. That way only
>> one debug session needs to be opened.
>>
>> This would be for CPUs of the same type that share memory.
>>
>> Thoughts?
>>
>
> 1) What if I have two different ARCH, like ARM7tdmi + ARM-Cortex?   Or worse
> -  ARM + DSP on the same target?

That I think two GDB servers are more approperiate.

> 2) What if GDB already knows about my threads?

OpenOCD does not currently have any threads support. If it grows
threads support,
i don't see that as a conflict with representing each CPU as thread.

>
> 3) GDB assumes a single address space for all threads - *MAYBE* true for
> *SOME* multi-core cpu. But - not all multi-cores have common address space.
>  ie: Blackfin BF561 - has 2 cores, shares external SDRAM but - each core has
> its own private on chip SRAM that is not accessible by the other core, also
> the two cores are non-cache coherent.

At some point it's better to break it apart to multiple GDB servers, but for
the simplest SMP shared memory model case, I'd prefer a single gdb server
w/multiple threads.
>
>
> -Duane.
>
>
>
>



-- 
Øyvind Harboe
http://www.zylin.com/zy1000.html
ARM7 ARM9 XScale Cortex
JTAG debugger and flash programmer

Free eCos workshop in Oslo October 21!
http://www.zylin.com/workshop.html
_______________________________________________
Openocd-development mailing list
[email protected]
https://lists.berlios.de/mailman/listinfo/openocd-development

Reply via email to