Re: [PATCH] Add support for fpscr for Power / PowerPC targets

[Kevin Buettner <kevinb@redhat.com>]

On Apr 11,  3:25pm, Daniel Jacobowitz wrote:

> I know this is not adequately documented right now, but every change to
> the format of the register cache is a change to the remote protocol.
> The closest thing we've got to documentation are the files in
> regformats/; they don't document a fraction of the possibilities,
> though.

Thanks for reminding me of this.

> I've updated the PowerPC definition to match this change, as
> well as updating gdbserver.

Thanks.

> Before I commit either, though:
> 
> > @@ -376,6 +380,14 @@ store_register (int tid, int regno)
> >        ptrace (PT_WRITE_U, tid, (PTRACE_ARG3_TYPE) regaddr,
> >  	      *(PTRACE_XFER_TYPE *) & buf[i]);
> >        regaddr += sizeof (PTRACE_XFER_TYPE);
> > +
> > +      if (errno == EIO 
> > +          && regno == gdbarch_tdep (current_gdbarch)->ppc_fpscr_regnum)
> > +	{
> > +	  /* Some older kernel versions don't allow fpscr to be written.  */
> > +	  continue;
> > +	}
> > +
> >        if (errno != 0)
> >  	{
> >  	  sprintf (mess, "writing register %s (#%d)", 
> 
> What versions are we talking here?

2.2.X, for X <= 15.  Probably for X > 15 too, but I haven't checked.
I don't think the 2.4 kernels have this problem though.

> Is it worth silencing the gdbserver warning in this case?

If it's only a warning, it probably doesn't hurt to leave it in,
though it'd be nice to structure it so that the warning is only
printed once.

I would have preferred to not to have to add the above code to
ppc-linux-nat.c, but I was seeing too many regressions without it and
that made it difficult to evaluate the patch.  It was worth it to add
the above code though because I ended up catching some regressions
that I would've missed otherwise.  (I probably ought to just upgrade
my Linux/PPC box...)

Kevin