Re: [PATCH v6 1/4] Add LLL_MUTEX_READ_LOCK [BZ #28537]

[Szabolcs Nagy <szabolcs.nagy@arm.com>]

The 11/11/2021 08:24, H.J. Lu via Libc-alpha wrote:
> CAS instruction is expensive.  From the x86 CPU's point of view, getting
> a cache line for writing is more expensive than reading.  See Appendix
> A.2 Spinlock in:
> 
> https://www.intel.com/content/dam/www/public/us/en/documents/white-papers/xeon-lock-scaling-analysis-paper.pdf
> 
> The full compare and swap will grab the cache line exclusive and cause
> excessive cache line bouncing.
> 
> Add LLL_MUTEX_READ_LOCK to do an atomic load and skip CAS in spinlock
> loop if compare may fail to reduce cache line bouncing on contended locks.

essentially the contended loop goes from

  if (cas(lock, 0, 1))
    break;

to

  if (read(lock) == 0)
    if (cas(lock, 0, 1))
      break;

i believe this helps on aarch64 too if it is only
done after contention is detected.

since the first cas access to lock is kept as is,
the common, uncontended case should not be affected.
this looks good to me.

> ---
>  nptl/pthread_mutex_lock.c | 7 +++++++
>  1 file changed, 7 insertions(+)
> 
> diff --git a/nptl/pthread_mutex_lock.c b/nptl/pthread_mutex_lock.c
> index 2bd41767e0..72058c719c 100644
> --- a/nptl/pthread_mutex_lock.c
> +++ b/nptl/pthread_mutex_lock.c
> @@ -64,6 +64,11 @@ lll_mutex_lock_optimized (pthread_mutex_t *mutex)
>  # define PTHREAD_MUTEX_VERSIONS 1
>  #endif
>  
> +#ifndef LLL_MUTEX_READ_LOCK
> +# define LLL_MUTEX_READ_LOCK(mutex) \
> +  atomic_load_relaxed (&(mutex)->__data.__lock)
> +#endif
> +
>  static int __pthread_mutex_lock_full (pthread_mutex_t *mutex)
>       __attribute_noinline__;
>  
> @@ -141,6 +146,8 @@ PTHREAD_MUTEX_LOCK (pthread_mutex_t *mutex)
>  		  break;
>  		}
>  	      atomic_spin_nop ();
> +	      if (LLL_MUTEX_READ_LOCK (mutex) != 0)
> +		continue;
>  	    }
>  	  while (LLL_MUTEX_TRYLOCK (mutex) != 0);
>  
> -- 
> 2.33.1
>