[PATCH v15 3/6] locking/qspinlock: Introduce CNA into the slow path of qspinlock

[Barry Song]

> +/*
> + * Implement a NUMA-aware version of MCS (aka CNA, or compact NUMA-aware lock).
> + *
> + * In CNA, spinning threads are organized in two queues, a primary queue for
> + * threads running on the same NUMA node as the current lock holder, and a
> + * secondary queue for threads running on other nodes. Schematically, it
> + * looks like this:
> + *
> + *    cna_node
> + *   +----------+     +--------+         +--------+
> + *   |mcs:next  | --> |mcs:next| --> ... |mcs:next| --> NULL  [Primary queue]
> + *   |mcs:locked| -.  +--------+         +--------+
> + *   +----------+  |
> + *                 `----------------------.
> + *                                        v
> + *                 +--------+         +--------+
> + *                 |mcs:next| --> ... |mcs:next|            [Secondary queue]
> + *                 +--------+         +--------+
> + *                     ^                    |
> + *                     `--------------------'
> + *

probably not only related with NUMA, it might be also related with cache topology.
For example, one NUMA might has a couple of sub domains, each domain shares some
last level cache. ZEN, Power and some ARM servers all have this kind of topology.

lock synchronization within this smaller range should be much faster. anyway, it
looks like a good start to be aware of numa only for this moment.

Thanks
barry