On Thu, 15 Nov 2001, Jason R Thorpe wrote:

> On Mon, Oct 29, 2001 at 09:03:49AM +0000, Chris Gilbert wrote:
>
>  > > Needed before the changeover:
>
>  > > * (gcc) Ensure 8-byte stack alignment on function entry [rearnsha]
>
> Any guesses on how hard this will be to implement in the compiler?

Non-trivial, at least.  The obvious #defines (STACK_BOUNDARY and
PREFERRED_STACK_BOUNDARY) don't seem to do anything useful, anyway.

>  > > * (gcc) packed enums (should be trivial to implement in the compiler, but
>  > >   I've heard rumours it might break openssh)
>
> What does the ARM ELF ABI document actually say about this?  If it's
> sufficiently vague (which I've heard that it is), why not just ignore
> the packed enum issue (i.e. stick w/ fixed-size enums like we have now)?

This is Richard's field, so we should wait for him to get back before
deciding anything.

>  > > * (asm sources) Ensure these preserve 8-byte stack alignment if not leaf
>  > >   functions [chris]
>
> I see Chris committed this.  What about the kernel... do we have to care
> about the kernel, in this case?

No, except where it sets up a user stack (and I think I've dealt with them
already).  We can assume users recompile the whole kernel at once, so we
only need to fix things when GCC starts generating code which requires an
aligned stack.

-- 
Ben Harris                                                   <bjh21@netbsd.org>
Portmaster, NetBSD/arm26               <URL:http://www.netbsd.org/Ports/arm26/>