Re: [PATCH v3] nptl: Deallocate the thread stack on setup failure (BZ #19511)

[Adhemerval Zanella <adhemerval.zanella@linaro.org>]

On 02/06/2021 15:30, Florian Weimer via Libc-alpha wrote:
> * Joseph Myers:
> 
>> On Wed, 2 Jun 2021, Andreas Schwab wrote:
>>
>>> On Jun 02 2021, Adhemerval Zanella wrote:
>>>
>>>> I don't have a strong preference here, maybe use a 'bool' then?
>>>
>>> bool has architecture dependent size.
>>
>> bool is one byte for all configurations supported by glibc (the only 
>> configuration supported by GCC for which it isn't is powerpc-darwin).  
>> Although I don't see any likely reason for glibc code to depend on it 
>> being one byte.
> 
> However, one reason not to use bool in internal code is that our own
> atomics do not support one-byte types on all architectures.
> 
> For example this definition in sysdeps/powerpc/atomic-machine.h:
> 
> #define atomic_exchange_acq(mem, value) \
>   ({                                                                          \
>     __typeof (*(mem)) __result;                                               \
>     if (sizeof (*mem) == 4)                                                   \
>       __result = __arch_atomic_exchange_32_acq (mem, value);                  \
>     else if (sizeof (*mem) == 8)                                              \
>       __result = __arch_atomic_exchange_64_acq (mem, value);                  \
>     else                                                                      \
>        abort ();                                                              \
>     __result;                                                                 \
>   })

I think it should be safe to use on code that won't be used for atomic
operations.  The this specific flag, for instance, is protected by
the PD lock.