Re: Dead code in pthread_cond_wait() for spin-wait

[Mathieu Desnoyers <mathieu.desnoyers@efficios.com>]

On 2023-10-11 14:30, Olivier Dion wrote:
> Hi,
> 
> Commit ed19993b5b0d05d62cc883571519a67dae481a14 from 2016 introduce "a
> new implementation of condition variables".  This new implementation
> seems to introduce a spin-wait fast-path to avoid taking the underlying
> futex when doing a pthread_cond_wait().

Adding libc-alpha, Carlos and Florian in CC.

Thanks,

Mathieu

> 
> However, looking at the following snippet in nptl/pthread_cond_wait.c:
> 
>   static __always_inline int
>   __pthread_cond_wait_common (pthread_cond_t *cond, pthread_mutex_t *mutex,
>       clockid_t clockid, const struct __timespec64 *abstime)
>   {
>     const int maxspin = 0;
>     [...]
>             /* Spin-wait first. [...] */
> 	   unsigned int spin = maxspin;
> 	   while (signals == 0 && spin > 0) {
>               [...]
>             }
>     [...]
>   }
> 
> it seems to me that the try-spinning portion is dead code in
> pthread_cond_wait_common(). I was wondering why this was the case?
> 
> I also think that there would maybe some interest in using the Userspace
> RCU wait algorithm [0] for this.
> 
>   [0] https://github.com/urcu/userspace-rcu/blob/master/src/urcu-wait.h
> 
> Thanks,
> 
> Olivier
> 

-- 
Mathieu Desnoyers
EfficiOS Inc.
https://www.efficios.com