[PATCH] i386: Change offset of __private_ss to 0x30

[H.J. Lu]

On Mon, Jun 4, 2018 at 6:30 AM, Florian Weimer <fweimer@redhat.com> wrote:
> On 05/30/2018 06:47 PM, H.J. Lu wrote:
>>
>> On Wed, May 30, 2018 at 9:43 AM, Florian Weimer <fweimer@redhat.com>
>> wrote:
>>>
>>> On 05/30/2018 04:49 PM, H.J. Lu wrote:
>>>>
>>>>
>>>> diff --git a/sysdeps/i386/nptl/tls.h b/sysdeps/i386/nptl/tls.h
>>>> index 6c36e58e9c..afb71ce431 100644
>>>> --- a/sysdeps/i386/nptl/tls.h
>>>> +++ b/sysdeps/i386/nptl/tls.h
>>>> @@ -43,11 +43,16 @@ typedef struct
>>>>      int gscope_flag;
>>>>      int __glibc_reserved1;
>>>>      /* Reservation of some values for the TM ABI.  */
>>>> -  void *__private_tm[4];
>>>> +  void *__private_tm[3];
>>>>      /* GCC split stack support.  */
>>>>      void *__private_ss;
>>>> +  void *__glibc_reserved2;
>>>>    } tcbhead_t;
>>>
>>>
>>>
>>> What's the state of TM in GCC?  Has it already been removed?
>>
>>
>> libitm has its own issue:
>>
>> https://gcc.gnu.org/bugzilla/show_bug.cgi?id=85988
>>
>> Fortunately, only the first 2 elements in __private_tm are used by
>> libitm.
>
>
> Looks like the reuse of the TM fields was actually deliberate:

True.

>   <https://sourceware.org/bugzilla/show_bug.cgi?id=10686>
>
> But the offsets are still wrong.

Yes, we are stuck with the wrong offset.

> The question still remains whether we still need libitm.
>

libitm is enabled by default in GCC.


-- 
H.J.