Modules and CONFIG_ALTIVEC

Modules and CONFIG_ALTIVEC

[Samuel Rydh]

On ppc, the size of thread_struct depends on the CONFIG_ALTIVEC setting.
Thus, the offset of various task_struct fields are dependent upon this
config option.

This makes it difficult to distribute precompiled module binaries
(in a safe manner). It really shouldn't be necessary to bundle:

	module-2.4.X
	module-2.4.X-smp
	module-2.4.X-altivec
	module-2.4.X-altivec-smp

To avoid this problem, one could do something like:

        unsigned long   fpscr;          /* Floating point status */
-#ifdef CONFIG_ALTIVEC
-       vector128       vr[32];         /* Complete AltiVec set */
-       vector128       vscr;           /* AltiVec status */
-       unsigned long   vrsave;
-#endif /* CONFIG_ALTIVEC */
+	altivec_t	*vregs;

I'm not sure the affected task_struct fields are
ever referenced by a typical kernel module. But this is
difficult to check and besides, new fields might be added
to the end of the struct.

Thoughts?

/Samuel


----------------------------------------------------------
 E-mail <samuel@ibrium.se>  WWW: <http://www.ibrium.se>
  Phone/fax: (home) +46 8 4418431, (work) +46 8 790nnnn
----------------------------------------------------------

** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/

Re: Modules and CONFIG_ALTIVEC

[Hollis]

On Sunday 05 August 2001 04:49, Samuel Rydh wrote:
> On ppc, the size of thread_struct depends on the CONFIG_ALTIVEC setting.
> Thus, the offset of various task_struct fields are dependent upon this
> config option.
[snip]
>         unsigned long   fpscr;          /* Floating point status */
> -#ifdef CONFIG_ALTIVEC
> -       vector128       vr[32];         /* Complete AltiVec set */
> -       vector128       vscr;           /* AltiVec status */
> -       unsigned long   vrsave;
> -#endif /* CONFIG_ALTIVEC */
> +	altivec_t	*vregs;

This also affects gdb. When gdb reads a core file, it gets upset if the size
of a regset isn't what it thinks it should be. Currently it's a bit too easy
to change that size...

-Hollis

** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/

Re: Modules and CONFIG_ALTIVEC

[Daniel Jacobowitz]

On Sun, Aug 05, 2001 at 06:08:21PM -0500, Hollis wrote:
>
> On Sunday 05 August 2001 04:49, Samuel Rydh wrote:
> > On ppc, the size of thread_struct depends on the CONFIG_ALTIVEC setting.
> > Thus, the offset of various task_struct fields are dependent upon this
> > config option.
> [snip]
> >         unsigned long   fpscr;          /* Floating point status */
> > -#ifdef CONFIG_ALTIVEC
> > -       vector128       vr[32];         /* Complete AltiVec set */
> > -       vector128       vscr;           /* AltiVec status */
> > -       unsigned long   vrsave;
> > -#endif /* CONFIG_ALTIVEC */
> > +	altivec_t	*vregs;
>
> This also affects gdb. When gdb reads a core file, it gets upset if the size
> of a regset isn't what it thinks it should be. Currently it's a bit too easy
> to change that size...

Yup.  Expect GDB patches to deal with that (and with altivec) from me
in the next month, if no one beats me to it.  It's approaching the top
of my TODO list.

--
Daniel Jacobowitz                           Carnegie Mellon University
MontaVista Software                         Debian GNU/Linux Developer

** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/

Re: Modules and CONFIG_ALTIVEC

[Franz Sirl]

At 08:38 06.08.2001, Daniel Jacobowitz wrote:

>On Sun, Aug 05, 2001 at 06:08:21PM -0500, Hollis wrote:
> >
> > On Sunday 05 August 2001 04:49, Samuel Rydh wrote:
> > > On ppc, the size of thread_struct depends on the CONFIG_ALTIVEC setting.
> > > Thus, the offset of various task_struct fields are dependent upon this
> > > config option.
> > [snip]
> > >         unsigned long   fpscr;          /* Floating point status */
> > > -#ifdef CONFIG_ALTIVEC
> > > -       vector128       vr[32];         /* Complete AltiVec set */
> > > -       vector128       vscr;           /* AltiVec status */
> > > -       unsigned long   vrsave;
> > > -#endif /* CONFIG_ALTIVEC */
> > > +   altivec_t       *vregs;
> >
> > This also affects gdb. When gdb reads a core file, it gets upset if the
> size
> > of a regset isn't what it thinks it should be. Currently it's a bit too
> easy
> > to change that size...
>
>Yup.  Expect GDB patches to deal with that (and with altivec) from me
>in the next month, if no one beats me to it.  It's approaching the top
>of my TODO list.

Uhm, I sent a patch to Daniel Berlin a while ago for commenting, but
haven't heard back yet. I can send it to you if you want to take a look.

Franz.


** Sent via the linuxppc-dev mail list. See http://lists.linuxppc.org/