Re: [PATCH 1/5] Linux: Implement a useful version of _startup_fatal

[Florian Weimer <fweimer@redhat.com>]

* Adhemerval Zanella via Libc-alpha:

> On 03/05/2022 13:57, Florian Weimer via Libc-alpha wrote:
>> * H. J. Lu:
>> 
>>> On Tue, May 3, 2022 at 9:44 AM Florian Weimer <fweimer@redhat.com> wrote:
>>>>
>>>> * H. J. Lu:
>>>>
>>>>> On Tue, May 3, 2022 at 9:01 AM Florian Weimer <fweimer@redhat.com> wrote:
>>>>>>
>>>>>> * H. J. Lu:
>>>>>>
>>>>>>> The startup codes can't use cancellation before it is ready.  Do we need to
>>>>>>> check SHARED in this case?
>>>>>>
>>>>>> I think __libc_fatal shouldn't be a cancellation point.  That's not
>>>>>> really related to SHARED.
>>>>>
>>>>> Should we define _startup_fatal unconditionally and always define
>>>>> I386_USE_SYSENTER as 0 for i386 so that _startup_fatal can be
>>>>> used safely during startup?
>>>>
>>>> Hmm.  Isn't that what my patch does?
>>>>
>>>
>>> Your patch still checks BUILD_PIE_DEFAULT and SHARED.
>> 
>> Ah, I see, thanks.  You are right, the i386 version should define
>> I386_USE_SYSENTER unconditionally.  Will fix.
>
> I think you might want to check ia64 as well, since it uses a TCB variable
> to call the syscall in some cases.

“might want to check“ isn't a phrase I would use in the context with
ia64, but yes, I can add a startup.h file for it, too.  Thanks for the
suggestion.

Florian