Jamie Lokier wrote:
> Paul Brook wrote:
>> On most targets watchpoint traps occur after the instruction
>> completes, so you have to defer the DEBUG exception.  Normal MMU
>> faults occur before the instruction completes.
> 
> Ok.  It might be a useful architecture-independent debugging feature
> to _also_ have the ability to trap a watchpoint before the memory
> access though - especially writes.  Are there any tracing tools which
> use the gdbstub support and could make use of that?  Can GDB make use
> of that?

The spec of gdb's remote protocol does not mention where the instruction
pointer is supposed to point at when a watchpoint is reported. But
making this configurable is a good idea, will take it into account. I'm
currently reworking my queue, and I just added a watchpoint flag that
will decide about this behavior. Just needs to be implemented, but that
shouldn't be hard. Will likely take yet another monitor command, just
like for allow/block IRQs on singlestep.

Jan