Andrew Cagney wrote:
> 
> > I think you left out an "else" when you made this change in december.
> > We must not call target_[fetch/store]_register with a pseudo-register,
> > or they will choke.  By definition, a pseudo-register is one that
> > cannot be fetched from the target.
> 
> Oops, kind of.
> 
> > Index: regcache.c
> > ===================================================================
> > RCS file: /cvs/src/src/gdb/regcache.c,v
> > retrieving revision 1.31
> > diff -c -3 -p -r1.31 regcache.c
> > *** regcache.c        2002/01/29 20:24:27     1.31
> > --- regcache.c        2002/02/06 02:30:46
> > *************** fetch_register (int regnum)
> > *** 125,131 ****
> >     if (FETCH_PSEUDO_REGISTER_P ()
> >         && pseudo_register (regnum))
> >       FETCH_PSEUDO_REGISTER (regnum);
> > !   target_fetch_registers (regnum);
> >   }
> >
> >   /* Write register REGNUM cached value to the target.  */
> > --- 125,132 ----
> >     if (FETCH_PSEUDO_REGISTER_P ()
> >         && pseudo_register (regnum))
> >       FETCH_PSEUDO_REGISTER (regnum);
> > !   else if (real_register (regnum))
> > !     target_fetch_registers (regnum);
> >   }
> >
> >   /* Write register REGNUM cached value to the target.  */
> 
> There should be an else but not else if (real_register ()) ....
> 
> A target can handle registers two ways - using pseudo-registers or using
> register_{read,write}().  If a target uses the latter then it will
> likely end up fetching registers in the pseudo-register range from the
> target.
> 
> A simple example of this might be a modified x86 where not only the user
> registers but also the system registers (MSRs?) are accessable.  The
> MSR's wouldn't be returned via the G packet but instead fetched
> individually.  Unfortunatly, NUM_REGS determines which registers are
> included in a G packet (and many other obscure things).

Good.  Replaced with a simple "else" and checked in as attached below.