[PATCH v5] Define PTHREAD_STACK_MIN to sysconf(_SC_THREAD_STACK_MIN)

[Carlos O'Donell]

On 7/16/21 6:17 PM, Khem Raj via Libc-alpha wrote:
> On Mon, Jul 12, 2021 at 12:51 PM Florian Weimer via Libc-alpha
> <libc-alpha@sourceware.org> wrote:
>>
>> * Maxim Kuvyrkov via Libc-alpha:
>>
>>> Hi H.J.,
>>>
>>> This breaks GCC’s libgomp build on at least aarch64-linux-gnu:
>>>
>>> /builds/destdir/x86_64-pc-linux-gnu/aarch64-linux-gnu/bin/ld: .libs/env.o: in function `initialize_env':
>>> /snapshots/gcc.git~master/libgomp/env.c:1536: undefined reference to `sysconf'
>>> /snapshots/gcc.git~master/libgomp/env.c:1538: undefined reference to `sysconf'
>>> /snapshots/gcc.git~master/libgomp/env.c:1539: undefined reference to `sysconf'
>>> /builds/destdir/x86_64-pc-linux-gnu/aarch64-linux-gnu/bin/ld: .libs/proc.o: in function `gomp_init_num_threads':
>>> /snapshots/gcc.git~master/libgomp/config/linux/proc.c:85: undefined reference to `sysconf'
>>> /snapshots/gcc.git~master/libgomp/config/linux/proc.c:143: undefined reference to `sysconf'
>>> /snapshots/gcc.git~master/libgomp/config/linux/proc.c:174: more undefined references to `sysconf' follow
>>> collect2: error: ld returned 1 exit status
>>> make[4]: *** [Makefile:728: libgomp.la] Error 1
>>>
>>> Both GCC and Glibc are tip of trunk.
>>>
>>> Would you please take a look?  Please let me know if it doesn’t easily
>>> reproduce and I’ll take a look.
>>
>> I pushed a patch today that should fix this on the glibc side:
>>
>>   Reduce <limits.h> pollution due to dynamic PTHREAD_STACK_MIN
>>   <https://sourceware.org/git/?p=glibc.git;a=commitdiff;h=aaacde11f2e814814f>
>>
>> It contains some magic that happens to fix GCC for now.  Jakub is
>> working on a patch that fixes the libgomp issue as well (the standard
>> header #include within the hidden-visibility area is invalid):
>>
>>   <https://gcc.gnu.org/pipermail/gcc-patches/2021-July/574960.html>
>>
> 
> I am seeing few packages failing since they use PTHREAD_STACK_MIN as constant
> see https://errors.yoctoproject.org/Errors/Build/124520/

Thanks for those reports, we'll need to keep an eye on these, but they will
all need to be fixed for non-constant pthread stack size. Finding and fixing
these issues will allow Intel and Arm to move forward with dynamic stack
sizes for new hardware features.

-- 
Cheers,
Carlos.