Excerpts from Waiman Long's message of July 24, 2020 12:29 am:

On 7/23/20 9:30 AM, Nicholas Piggin wrote:

quoted

quoted

I would prefer to extract out the pending bit handling code out into a
separate helper function which can be overridden by the arch code
instead of breaking the slowpath into 2 pieces.

You mean have the arch provide a queued_spin_lock_slowpath_pending
function that the slow path calls?

I would actually prefer the pending handling can be made inline in
the queued_spin_lock function, especially with out-of-line locks it
makes sense to put it there.

We could ifdef out queued_spin_lock_slowpath_queue if it's not used,
then __queued_spin_lock_slowpath_queue would be inlined into the
caller so there would be no split?

The pending code is an optimization for lightly contended locks. That is 
why I think it is appropriate to extract it into a helper function and 
mark it as such.

You can certainly put the code in the arch's spin_lock code, you just 
has to override the generic pending code by a null function.

I see what you mean. I guess that would work fine.

Thanks,
Nick