PATCH: Add x32 arch_prctl support

[H.J. Lu]

On Mon, May 21, 2012 at 4:26 PM, Roland McGrath <roland@hack.frob.com> wrote:
>> Change it to unsigned long long int makes it impossible
>> to get address as pointer since it will override memory on stack.
>
> There is some misunderstanding.  What I suggested would not make any
> difference to the compiled code of arch_prctl.

The goes back to user uint32_t vs kernel uint64_t:

http://sourceware.org/ml/libc-alpha/2012-05/msg01519.html

The difference is movq vs movl.  There is no need for movq
since x32 will only pass unsigned 32-bit value to kernel.

>> I am trying to limit my x32 GDB change as small as possible:
>>
>> hjl@gnu-6 gdb]$ grep PTRACE_ARCH_PRCTL  amd64-linux-nat.c
>> #ifndef PTRACE_ARCH_PRCTL
>> #define PTRACE_ARCH_PRCTL      30
>>         if (ptrace (PTRACE_ARCH_PRCTL, lwpid, base, ARCH_GET_FS) == 0)
>>         if (ptrace (PTRACE_ARCH_PRCTL, lwpid, base, ARCH_GET_GS) == 0)
>> [hjl@gnu-6 gdb]$
>>
>> Without PTRACE_ARCH_PRCTL means I have to write a whole
>> new patch to use the new interface.   It will make it harder to enable
>> x32 in GDB.
>
> Sometimes the least change everywhere is not the cleanest answer anywhere.
>
> diff --git a/gdb/amd64-linux-nat.c b/gdb/amd64-linux-nat.c
> index 5ebba3a..aaa1e62 100644
> --- a/gdb/amd64-linux-nat.c
> +++ b/gdb/amd64-linux-nat.c
> @@ -479,11 +479,41 @@ ps_get_thread_area (const struct ps_prochandle *ph,
>       switch (idx)
>        {
>        case FS:
> -         if (ptrace (PTRACE_ARCH_PRCTL, lwpid, base, ARCH_GET_FS) == 0)
> +         if (sizeof(unsigned long) == 4)
> +           {
> +             /* PTRACE_ARCH_PRCTL is obsolete since 2.6.25, where the
> +                fs_base and gs_base fields of user_regs_struct can be
> +                used directly.  We use it only to support older
> +                kernels.  But x32 kernels do not support it, so we
> +                don't try to use it.  There were no x32 kernels so old
> +                that fs_base wasn't usable.  */
> +             unsigned long fs;
> +             errno = 0;
> +             fs = ptrace (PTRACE_PEEKUSR, lwpid,
> +                          offsetof (struct user_regs_struct, fs_base), 0);
> +             if (errno == 0)
> +               {
> +                 *base = fs;
> +                 return PS_OK;
> +               }
> +           }
> +         else if (ptrace (PTRACE_ARCH_PRCTL, lwpid, base, ARCH_GET_FS) == 0)
>            return PS_OK;
>          break;
>        case GS:
> -         if (ptrace (PTRACE_ARCH_PRCTL, lwpid, base, ARCH_GET_GS) == 0)
> +         if (sizeof(unsigned long) == 4)
> +           {
> +             unsigned long gs;
> +             errno = 0;
> +             gs = ptrace (PTRACE_PEEKUSR, lwpid,
> +                          offsetof (struct user_regs_struct, gs_base), 0);
> +             if (errno == 0)
> +               {
> +                 *base = gs;
> +                 return PS_OK;
> +               }
> +           }
> +         else if (ptrace (PTRACE_ARCH_PRCTL, lwpid, base, ARCH_GET_GS) == 0)
>            return PS_OK;
>          break;
>        default:                   /* Should not happen.  */

Thanks.  But since x32 kernel takes PTRACE_ARCH_PRCTL, I
still need x32 ptrace to either handle it properly or set EINVAL
when PTRACE_ARCH_PRCTL is passed.

-- 
H.J.