Re: step out of range

[Fernando Nasser <fnasser at cygnus dot com>]

"J.T. Conklin" wrote:
> 
> I've encounter monitors and ICEs that have a "step out of range"
> command where execution continues until a instruction outside of a
> range of instructions is about to be executed or a breakpoint is
> triggered.  If GDB is able to take advantage of this feature, it won't
> have to do multiple single-steps for step, next, until, etc. commands.
> 
> This could siginficantly improve interactive debugging performance, as
> it eliminates most of the latency from the "step an instruction/ query
> registers to check whether we're within the address range/yep we
> are/repeat" sequence.
> 
> GDB already computes a range of instructions (step_range_start,
> step_range_end) that it uses internally when stepping through a range
> of instructions.  So it seems that all we need to do is to expose/use
> these variables in the target layer.  In fact, the vxWorks RDB target
> (remote-vx.c) does exactly this.  But it seems that these values
> should be passed through to target_remote() instead of being globals.
> 
> Any thoughts?  I'd like to add a step out of range command to the
> remote protocol.
> 

This idea is really neat.  I had a sort of stub some years ago 
(actually a "Debugger Helper" loaded on a multiprocessor node) that
was capable of taking care of the single step without having to
get back to the debugger.  The implementation was a hack, but the
result was effective.

You have my vote (as people are now in the voting mood :-).

-- 
Fernando Nasser
Red Hat - Toronto                       E-Mail:  fnasser@redhat.com
2323 Yonge Street, Suite #300
Toronto, Ontario   M4P 2C9