Re: Only 23 days left of active development before master freezes.

["Carlos O'Donell" <carlos at redhat dot com>]

On 12/18/2014 03:50 PM, David Miller wrote:
> From: "Carlos O'Donell" <carlos@redhat.com>
> Date: Thu, 18 Dec 2014 15:37:58 -0500
> 
>> Do you need help fixing up sparc?
> 
> I'm running into problems that look like compiler bugs, for example while
> building the dynamic linker code we have, from a CPP pre-processed dl-load.c:

If you think it's a compiler bug and that it will take time to fix
then push a patch that uses a pragma to disable the warning and
add a comment about the sparc issue.
 
> inline void __attribute__ ((unused, always_inline))
> elf_get_dynamic_info (struct link_map *l, Elf32_Dyn *temp)
> {
>   Elf32_Dyn *dyn = l->l_ld;
>   Elf32_Dyn **info;
>  ...
>   info = l->l_info;
> 
>   while (dyn->d_tag != 0)
>     {
>       if ((d_tag_utype) dyn->d_tag < 34)
>  info[dyn->d_tag] = dyn;
>       else if (dyn->d_tag >= 0x70000000 &&
>         dyn->d_tag < 0x70000000 + 2)
>  info[dyn->d_tag - 0x70000000 + 34] = dyn;
>  ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
> 
> GCC warns about that final expression:
> 
> get-dynamic-info.h:50:40: error: array subscript is above array bounds [-Werror=array-bounds]
> 
> However, looking at l->l_info's definition:
> 
>     Elf32_Dyn *l_info[34 + 2 + 16
>         + 3 + 12 + 11];
> 
> The expression is certainly protected in a way such that it will be
> in-bounds.  The conditional only accepts d_tag values which are either
> 0x70000000 or 0x70000001.
> 
> This means the info[] array index evaluates to either 34 or 35, which
> is definitely in the range of the defined array size.
> 
> This is with Debian's gcc-4.6.3

Definitely looks like a compiler bug.

See 0cd8552045aeb0537b4f1ab5d1b30ada8ae50076 to fix it like Siddhesh
fixed similar code in nscd.

Cheers,
Carlos.