On Thu, 2007-10-11 at 08:04 -0500, Kumar Gala wrote:
quoted text
> On Oct 11, 2007, at 2:15 AM, markn@au1.ibm.com wrote:
>=20
> > Replace NT_PRXFPREG with ELF_CORE_XFPREG_TYPE in the coredump code =20
> > which
> > allows for more flexibility in the note type for the state of =20
> > 'extended
> > floating point' implementations in coredumps. New note types can =20
> > now be
> > added with an appropriate #define.
> >
> > #define ELF_CORE_XFPREG_TYPE to be NT_PRXFPREG in all current users so
> > there's are no change in behaviour.
>=20
> Can we make this ELF_CORE_VECREG_TYPE or something that is so coupled =20
> to the x86 specific name?
>=20
> >
> > Signed-off-by: Mark Nelson <markn@au1.ibm.com>
> > ---
> >  arch/ia64/ia32/elfcore32.h     |    1 +
> >  arch/x86_64/ia32/ia32_binfmt.c |    1 +
> >  fs/binfmt_elf.c                |    4 ++--
> >  include/asm-i386/elf.h         |    1 +
> >  4 files changed, 5 insertions(+), 2 deletions(-)
> >
> > Index: linux/arch/ia64/ia32/elfcore32.h
> > =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D
quoted text
> > --- linux.orig/arch/ia64/ia32/elfcore32.h
> > +++ linux/arch/ia64/ia32/elfcore32.h
> > @@ -117,6 +117,7 @@ elf_core_copy_task_fpregs(struct task_st
> >  }
> >
> >  #define ELF_CORE_COPY_XFPREGS 1
> > +#define ELF_CORE_XFPREG_TYPE NT_PRXFPREG
> >  static inline int
> >  elf_core_copy_task_xfpregs(struct task_struct *tsk, =20
> > elf_fpxregset_t *xfpu)
> >  {
> > Index: linux/arch/x86_64/ia32/ia32_binfmt.c
> > =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D
quoted text
> > --- linux.orig/arch/x86_64/ia32/ia32_binfmt.c
> > +++ linux/arch/x86_64/ia32/ia32_binfmt.c
> > @@ -188,6 +188,7 @@ elf_core_copy_task_fpregs(struct task_st
> >  }
> >
> >  #define ELF_CORE_COPY_XFPREGS 1
> > +#define ELF_CORE_XFPREG_TYPE NT_PRXFPREG
> >  static inline int
> >  elf_core_copy_task_xfpregs(struct task_struct *t, elf_fpxregset_t =20
> > *xfpu)
> >  {
> > Index: linux/fs/binfmt_elf.c
> > =3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=
=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D=3D
quoted text
> > --- linux.orig/fs/binfmt_elf.c
> > +++ linux/fs/binfmt_elf.c
> > @@ -1446,8 +1446,8 @@ static int elf_dump_thread_status(long s
> >
> >  #ifdef ELF_CORE_COPY_XFPREGS
> >  	if (elf_core_copy_task_xfpregs(p, &t->xfpu)) {
> > -		fill_note(&t->notes[2], "LINUX", NT_PRXFPREG, sizeof(t->xfpu),
> > -			  &t->xfpu);
> > +		fill_note(&t->notes[2], "LINUX", ELF_CORE_XFPREG_TYPE,
> > +			  sizeof(t->xfpu), &t->xfpu);
> >  		t->num_notes++;
> >  		sz +=3D notesize(&t->notes[2]);
> >  	}
>=20
> You've only fixed up one of 4 NT_PRXFPREG uses in the kernel.
>=20
> Also, I'm not a fan of your proposed mechanism to "overload" the =20
> elf_fpxregset_t.  I'd rather see us introduce a new elf_vecregset_t =20
> and have it typedef'd to be elf_fpxregset_t on i386, x86_64, ia64

He's not overloading it, there is no such thing as an elf_fpxregset_t,
unless it's typedef'd to something else by the arch code, eg:

include/asm-i386/elf.h:typedef struct user_fxsr_struct elf_fpxregset_t;
include/asm-ia64/elf.h:typedef unsigned long elf_fpxregset_t;

Introducing another typedef would just be another layer of obfuscation
IMO, and the elf code doesn't need any more of that.

cheers

--=20
Michael Ellerman
OzLabs, IBM Australia Development Lab

wwweb: http://michael.ellerman.id.au
phone: +61 2 6212 1183 (tie line 70 21183)

We do not inherit the earth from our ancestors,
we borrow it from our children. - S.M.A.R.T Person