Re: gcc 4.5 and TLS

[Richard Zidlicky <rz@linux-m68k.org>]

On Fri, Jun 04, 2010 at 10:57:50PM +0200, Geert Uytterhoeven wrote:
> On Fri, Jun 4, 2010 at 21:15, Thorsten Glaser <tg@mirbsd.de> wrote:
> > Andreas Schwab dixit:
> >
> >>One cost is ABI breakage.
> >
> > Hrm, that is true. But then: is that syscall/structure used already?
> > I think I get an idea of the trouble though…
> >
> > On the other hand, changing gcc’s default alignment may break
> > the ABI as well, no?
> >
> >>Also, I'm not yet convinced this works in all
> >>situations, like for locks allocated on stack.
> >
> > That may not work, yes, since gcc is brain-dead and assumes that
> > the stack is already aligned “properly”. What is the default stack
> > alignment on m68k (in real life, not by some ABI)? If it’s 4, I
> > think it shouldn’t be a problem.
> 
> It's 2. Even if you push a byte to the stack, the SP will be aligned to
> an even address.

this is the hardware behaviour but the compiler is free to use more alignment
and it does that. At least for function calls I am pretty sure the minumum is 
guaranteed to be 4 bytes.
This is because for function calls the compiler must work work with the
assumption that the function may be invoked without a proper prototype
declaration from an external module (which is hackish but perfectly legal 
in K&R) and for this reason K&R mandates a storage size that is large enough
to accomodate both long or pointer.

However I am not sure if automatic variable allocation on the stack is 4 bytes
aligned. 
Even if it was not I am pretty sure it would not require an incompatible change 
of the ABI to guarantee 4 bytes stack alignment. It would also have close to zero 
effect on runtime performance and memory usage.

The big ABI change that was discussed some time ago would have involved lot 
more - increasing the alignment of elements inside structs and alignment of 
array elements. Especially the struct alignment change would break plenty
of stuff.

I am not sure what happens if the futex is inside a "misaligned struct" - would
that be handled with an attribute of the futex?

Richard