Re: [patch/rfc] Revise REGISTER_SIM_REGNO()

[Michael Snyder <msnyder at redhat dot com>]

Richard Earnshaw wrote:
> 
> > > I'd like to see a target have a way to report that it is permanently
> > > unable to recover a register -- because there's nothing in the protocol to
> > > allow its recovery.
> > >
> > > For example, I've added the privileged mode registers to my ARM target
> > > code; when the target is using a ptrace() interface for debugging a user
> > > program, then these registers are never available and it's pointless
> > > having gdb report them.
> >
> > How about having target_fetch_register set them to -1 in the cache?
> > See remote.c:remote_fetch_registers:
> >
> >    set_register_cached (i, -1);
> >
> > This tells the rest of GDB that the value of the register is
> > "not available".  You could unconditionally mark certain regs
> > as unavailable whenever target_fetch_registers is called.
> 
> Already tried that idea.  It doesn't work.
> 
> set_register_cached (-1) means that the register is "temporarily
> unavailable" at this time (due to the way we gathered the registers).
> Each time registers_changed() is called the value is reset to zero.
> 
> I need a way the target vector to let REGISTER_NAME() know that the
> register is "unavailable this session", so that it can return an empty
> string for the register; so that gdb won't think it exists at all.


I see.  This is something that depends on both the arch and the target.