Re: [PATCH 2/2] add powerpc64 port

[Bobby Bingham <koorogi@koorogi.info>]

On Sun, Apr 03, 2016 at 01:26:31PM -0400, Rich Felker wrote:
> On Sun, Apr 03, 2016 at 12:10:44PM -0500, Bobby Bingham wrote:
> > > > r2 is call-saved when calling to the local entry point, so setjmp needs
> > > > to save it.
> > >
> > > OK, I see how this works for local calls to setjmp. But how does the
> > > linker PLT magic work for setjmp?
> > >
> > > After the first return, the caller's stack slot where r2 was saved
> > > belongs to the caller, and the compiler can clobber it. Upon the
> > 
> > The ABI is very prescriptive about the layout of a stack frame.  Each
> > stack frame has several slots where callees are allowed to use part of
> > their caller's frame.  For example, the link register is saved to the
> > caller's frame, not the callee's.
> > 
> > For several of these slots, the ABI explicitly documents that they may
> > be used as temporary storage which should be considered call-clobbered.
> > For the slot used for saving the toc pointer (r2), the ABI makes no
> > mention of it being available for temporary storage.  It would be nice
> > if it were more explicit here, but I believe the intent is that the
> > compiler may not use this slot for any other purpose.
> 
> My concern was not that the function itself could clobber it (although
> I think it would be entitled to if it's no longer live, i.e. if no
> code paths remain that reference its value) but that future function
> calls might clobber it. However I think they all necessarily either
> don't write to this slot at all, or write the same value that was
> already there, so it's probably safe without longjmp having to restore
> it.

Yes, that's right.  All function calls from a single caller will either
not clobber that stack slot, or will write the same value to it.

> 
> Rich