Scheduler multithreading

[brson]

r? @nikomatsakis

This implements very basic multithreading in the new scheduler. This is only using a trivial scheduling strategy with a shared task queue. It adds three new concurrent types to the scheduler: WorkQueue, MessageQueue, and SleeperList. WorkQueue is temporarily just a basic queue; MessageQueue is a multiple-producer, singe-consumer queue for sending messages to schedulers; and SleeperList is a shared stack used for waking up sleeping schedulers. All three are intended to be lock free but for now are implemented naively with locks.

The main scheduling routine, which uses all three of these data structures is here: https://github.com/brson/rust/blob/io-upstream/src/libstd/rt/sched.rs#L156

An example of how the multithreaded scheduler is constructed: https://github.com/brson/rust/blob/io-upstream/src/libstd/rt/test.rs#L62
#4419

[nikomatsakis]

/me reads

[nikomatsakis]

I made a branch where I inserted comments and made some minor changes:

https://github.com/nikomatsakis/rust/tree/brson-io-upstream

This commit in particular:

nikomatsakis@6889ddb

@brson maybe we can chat with vidyo on Monday to clear up some of the things I am confused about?

[nikomatsakis]

By the way the code is pretty nice! I'd like to see a file doc.rs with a high-level overview of the strategy. I was a bit confused about which threads could be running where at various points, and I still don't quite understand the logic behind cleanup jobs.

[brson]

@nikomatsakis sure we can talk about it monday.

I also threw in some new commits that implement SharedChan and SharedPort. When combined they amount to an unbounded, multiple producer/multiple consumer queue.

[brson]

Removed SharedChan/Port commits because they are buggy.

[brson]

I've cleaned this up so it only contains the parts that I believe are pretty stable. I intend to go apply some of feedback in your commit still, but may not write the docs for a while.

[brson]

Also, still waiting for incoming to be green. I've got that bug in my sights, though.

[nikomatsakis]

r+. I didn't add tihs to the commit since it's not mergeable. plus, I didn't expect you to write the docs yet :)