Re: [Xen-ia64-devel] Mini-OS registers and libgcc

[Dietmar Hahn <dietmar.hahn@xxxxxxxxxxxxxxxxxxx>]

Hi Samuel,

comments are within the text!

Am Mittwoch, 13. Februar 2008 schrieb Samuel Thibault:
> Hello,
>
> Dietmar Hahn, le Wed 13 Feb 2008 11:53:59 +0100, a écrit :
> > Am Dienstag, 12. Februar 2008 schrieb Samuel Thibault:
> > > Is there a reason for using -mfixed-range=f2-f5,f12-f15,f32-f127
> > > when compiling Mini-OS?  Is it really needed since it doesn't
> > > have a user-land? (though I guess some support would be needed in
> > > switch_context for instance)
> >
> > To handle this in a clean way the trap_frame has to be expanded and some
> > fp registers have to stored/restored in trap handling.
>
> Ok. Would this be needed in order to be able to use libgcc.a?

No, both things are independent from each other.

>
> > > Then, are there other reasons for using __*div/di*.S instead of just
> > > linking with libgcc?
> >
> > It's possible to use libgcc.a but the flag -mconstant-gp has to be
> > removed from the mini-os makefiles and a clean way to add libgcc.a is
> > needed.
>
> Something like the attached patch?
>
> Actually I'm asking because for other purpose, I would need __divdi on
> x86_32, and using libgcc.a is a really simple solution.

Ok I understand - the attached patch runs fine.

> > By the way current mini-os crashed with Unaligned Reference.
> > A small change is needed:
> >  static void maybe_split(struct xmalloc_hdr *hdr, size_t size, size_t
> > block) {
> >      struct xmalloc_hdr *extra;
> > -    size_t leftover = block - size;
> > +    size_t leftover;
> > +#if defined(__ia64__)
> > +    size = (size +7 ) & ~7;  /* alignment needed on 8 byte boundary. */
> > +#endif
> > +    leftover = block - size;
>
> Does the following patch work?
>
> diff -r 3cc84775d725 extras/mini-os/lib/xmalloc.c
> --- a/extras/mini-os/lib/xmalloc.c    Tue Feb 12 17:04:32 2008 +0000
> +++ b/extras/mini-os/lib/xmalloc.c    Wed Feb 13 12:03:52 2008 +0000
> @@ -62,10 +62,19 @@ struct xmalloc_pad
>      size_t hdr_size;
>  };
>
> +/* Return size, increased to alignment with align. */
> +static inline size_t align_up(size_t size, size_t align)
> +{
> +    return (size + align - 1) & ~(align - 1);
> +}
> +
>  static void maybe_split(struct xmalloc_hdr *hdr, size_t size, size_t
> block) {
>      struct xmalloc_hdr *extra;
> -    size_t leftover = block - size;
> +    size_t leftover;
> +    size = align_up(size, __alignof__(struct xmalloc_hdr));
> +    size = align_up(size, __alignof__(struct xmalloc_pad));
> +    leftover = block - size;
>
>      /* If enough is left to make a block, put it on free list. */
>      if ( leftover >= (2 * (sizeof(struct xmalloc_hdr) + sizeof(struct
> xmalloc_pad))) ) @@ -98,12 +107,6 @@ static struct xmalloc_hdr
> *xmalloc_new_p
>      maybe_split(hdr, size, PAGE_SIZE);
>
>      return hdr;
> -}
> -
> -/* Return size, increased to alignment with align. */
> -static inline size_t align_up(size_t size, size_t align)
> -{
> -    return (size + align - 1) & ~(align - 1);
>  }
>
>  /* Big object?  Just use the page allocator. */
>
> Samuel

This patch works fine too!

Dietmar.

_______________________________________________
Xen-ia64-devel mailing list
Xen-ia64-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-ia64-devel