Re: [PATCH]Remove stack symbol definition from crt files for SH

[Jonathan Larmour <jlarmour at redhat dot com>]

"J. Johnston" wrote:
> 
> Jonathan Larmour wrote:
> >
> > "J. Johnston" wrote:
> > >
> > > anitak wrote:
> > > >
> > > > Hi,
> > > > When the stack region is specified at the top of the memory in linker
> > > > script,
> > > > the following definition in newlib/libc/sys/sh/crt0.S
> > > >  _stack: .long   0xdeaddead causes it to be put out of ram and gives error
> > > > for COFF.
> > > > Thanks for any reply.
> > > > Regards,
> > > > Anita Kulkarni
> > > >
> > >
> > > Patch accepted, thanks.
> >
> > Sorry to time travel backwards :-), but I think this patch has an
> > unintended effect. The linker scripts in libgloss (at least) assume that
> > _stack is defined. Without this the linker scripts are not usable - in fact
> > this makes it difficult to provide a stack without defining the stack in
> > the application or some other object which can't be the intention, as
> > opposed to simply its location being in the linker script.
> >
> > Anita's problem was that she didn't define a better location for _stack in
> > the linker script, that's all; but that was the way it was intended to be
> > used.
> 
> IIRC, the default sh linker script does not define it properly.  Is eCos using
> the libgloss scripts and depending on the 0xdeaddead value?

Not eCos - this is just something I had noticed for a wholely unrelated
reason. But after some offline chatting with Alexandre Oliva, I think I
understand the reason for it now. Sorry for the false alarm!

Jifl
-- 
Red Hat, Rustat House, Clifton Road, Cambridge, UK. Tel: +44 (1223) 271062
Maybe this world is another planet's Hell -Aldous Huxley || Opinions==mine