on acquire-release memory-model semantics

[Jason E. Aten]

https://davekilian.com/acquire-release.html

is a great introduction to acquire-release semantics, or

"how your ARM CPU really thinks about memory fences".

(hint: there are two half-fences rather than one full fence).

I had to learn about them to keep the race detector happy while

hacking on channels. I'm not sure why the race detector

does not see the mutex locks in the runtime, but for some

reason it needs to explicitly be told about the effects the

channel locks in runtime/chan.go have, or else it will raise false alarms.