[PR19826] fix non-LE TLS in static programs

[Florian Weimer]

On 12/02/2016 05:25 PM, Carlos O'Donell wrote:

>> This patch broke non-optimized global-dynamic TLS on aarch64 (for
>> shared builds), as shown by my new tst-tls-manydynamic test case.
>
> Does reverting Alex's patch fix tst-tls-manydynamic?

It's sufficient to revert the patch to nptl/allocatestack.c.  I don't 
know if it is really needed, I and I have no idea how to make 
initialization at this point safe because the threads themselves could 
be using the DTV concurrently, so we cannot enlarge it.

I filed <https://sourceware.org/bugzilla/show_bug.cgi?id=20915> so that 
we keep track of this regression and fix it before the release.

Thanks,
Florian