[PATCH] Async signal safe TLS accesses

[Paul Pluzhnikov]

On Thu, Jan 2, 2014 at 11:04 AM, Andrew Hunter <ahh@google.com> wrote:

> This is version 6 of the patch, containing:

There were still some new build warnings generated, which I fixed ...

> @@ -114,8 +143,17 @@ void
>  internal_function __attribute_noinline__
>  _dl_allocate_static_tls (struct link_map *map)
>  {
> -  if (map->l_tls_offset == FORCED_DYNAMIC_TLS_OFFSET
> -      || _dl_try_allocate_static_tls (map))
> +  /* We wrap this in a signal mask because it has to iterate all threads
> +     (including this one) and update this map's TLS entry. A signal handler
> +     accessing TLS would try to do the same update and break.  */
> +  sigset_t old;
> +  _dl_mask_all_signals (&old);
> +  int err = -1;
> +  if (map->l_tls_offset != FORCED_DYNAMIC_TLS_OFFSET)
> +    err = _dl_try_allocate_static_tls (map);
> +
> +  _dl_unmask_signals (&old);
> +  if (err != 0)
>      {
>        _dl_signal_error (0, map->l_name, NULL, N_("\
>  cannot allocate memory in static TLS block"));

This one looks like a possible bug to me:
it will always error if map->l_tls_offset == FORCED_DYNAMIC_TLS_OFFSET.

Shouldn't it be 'int err = 0;' ?

Thanks,
-- 
Paul Pluzhnikov