[PATCH] Add PTRACE_{GET|SET}VRREGS

[Steve Munroe]

Changing you internal structures is OK with me. But don't change the 
size/alignment for PTRACE_GETVRREGS and PTRACE_SETVRREGS

Steven J. Munroe
Linux on Power Toolchain Architect
IBM Corporation, Linux Technology Center




Kumar Gala <kumar.gala at freescale.com> 
09/09/2005 09:09 AM

To
Steve Munroe/Rochester/IBM at IBMUS
cc
Robert Jennings <rcjenn at austin.ibm.com>, Anton Blanchard 
<anton at samba.org>, linuxppc64-dev <linuxppc64-dev at ozlabs.org>, Paul 
Mackerras <paulus at samba.org>
Subject
Re: [PATCH] Add PTRACE_{GET|SET}VRREGS






Steve,

Can you provide any insight into Anton's comment.

What I would like to see is on ppc64:

struct thread_struct {
...
#ifdef CONFIG_ALTIVEC
         /* Complete AltiVec register set */
         vector128       vr[32] __attribute((aligned(16)));
         /* AltiVec status */
         vector128       vscr __attribute((aligned(16)));
         unsigned long   vrsave;
         int             used_vr;        /* set if process has used 
altivec */
#endif /* CONFIG_ALTIVEC */
};

turn into:

struct thread_struct {
...
#ifdef CONFIG_ALTIVEC
         /* Complete AltiVec register set */
         vector128       vr[32] __attribute((aligned(16)));
         /* AltiVec status */
         vector128       vscr __attribute((aligned(16)));
         unsigned int   vrsave;
         int             used_vr;        /* set if process has used 
altivec */
#endif /* CONFIG_ALTIVEC */
};

(Change of unsigned long to unsigned int for vrsave).  This is 
because vrsave is truly only 32-bits in all implementations.

Is there some reason this breaks the ppc64 ABI?

- kumar


On Sep 9, 2005, at 1:47 AM, Anton Blanchard wrote:

>
> Hi Kumar,
>
>
>> I repeat my request that we make vrsave an unsigned int instead of an
>> unsigned long in thread_struct since we are touch this code anyways.
>>
>
> Apparently the 3 word pad is there for a reason in the 64bit ABI, 
> Steve
> Munroe would know the details.
>
> Anton
>