Re: [PATCH v2 5/5] locking/rwsem: Remove reader optimistic spinning

[Peter Zijlstra]

On Fri, Nov 20, 2020 at 11:14:16PM -0500, Waiman Long wrote:


> @@ -1032,40 +901,16 @@ rwsem_down_read_slowpath(struct rw_semaphore *sem, int state, long count)
>  	 *
>  	 * We can take the read lock directly without doing
>  	 * rwsem_optimistic_spin() if the conditions are right.

This comment no longer makes sense..

> -	 * Also wake up other readers if it is the first reader.
>  	 */
> -	if (!(count & (RWSEM_WRITER_LOCKED | RWSEM_FLAG_HANDOFF)) &&
> -	    rwsem_no_spinners(sem)) {
> +	if (!(count & (RWSEM_WRITER_LOCKED | RWSEM_FLAG_HANDOFF))) {
>  		rwsem_set_reader_owned(sem);
>  		lockevent_inc(rwsem_rlock_steal);
> -		if (rcnt == 1)
> -			goto wake_readers;
> -		return sem;
> -	}
>  
> -	/*
> -	 * Save the current read-owner of rwsem, if available, and the
> -	 * reader nonspinnable bit.
> -	 */
> -	waiter.last_rowner = owner;
> -	if (!(waiter.last_rowner & RWSEM_READER_OWNED))
> -		waiter.last_rowner &= RWSEM_RD_NONSPINNABLE;
> -
> -	if (!rwsem_can_spin_on_owner(sem, RWSEM_RD_NONSPINNABLE))
> -		goto queue;
> -
> -	/*
> -	 * Undo read bias from down_read() and do optimistic spinning.
> -	 */
> -	atomic_long_add(-RWSEM_READER_BIAS, &sem->count);
> -	adjustment = 0;
> -	if (rwsem_optimistic_spin(sem, false)) {

since we're removing the optimistic spinning entirely on the read side.

Also, I was looking at skipping patch #4, which mucks with the reader
wakeup logic, and afaict this removal doesn't really depend on it.

Or am I missing something?