[PATCH][ARM] Fix _dl_tlsdesc_resolve_hold to save r0

[Szabolcs Nagy]

On 14/10/15 16:09, Joseph Myers wrote:
> On Wed, 14 Oct 2015, Szabolcs Nagy wrote:
>> On 14/10/15 13:17, Joseph Myers wrote:
>>> On Wed, 14 Oct 2015, Szabolcs Nagy wrote:
>> i have a test that triggers this or the other tlsdesc race
>> i was about to fix, but it is a bit expensive.
>>
>> (666 tls objects accessed from 66 threads, 2 versions to
>> get tlsdesc entries 8byte-unaligned, running 10x triggers
>> the failure, a smaller test triggers less frequently.)
>>
>> i'm not sure what's the glibc policy for expensive tests.
>
> Well, at least it should be attached to the bug in Bugzilla so it's
> available to other people.  What are the actual memory / CPU time
> requirements of this test?
>

i can run the test 10 times under a second,
the memory usage is small (but the address
space usage of the thread stacks may cause
problems with memory overcommit disabled).

native compilation of the 666 sections takes some
time (i put each tls object into different section
so each gets its own tlsdesc relocations).

(the code is partly generated i see if i can
cut it down a bit more for inclusion into glibc).