1 // Copyright 2014 The Rust Project Developers. See the COPYRIGHT
2 // file at the top-level directory of this distribution and at
3 // http://rust-lang.org/COPYRIGHT.
5 // Licensed under the Apache License, Version 2.0 <LICENSE-APACHE or
6 // http://www.apache.org/licenses/LICENSE-2.0> or the MIT license
7 // <LICENSE-MIT or http://opensource.org/licenses/MIT>, at your
8 // option. This file may not be copied, modified, or distributed
9 // except according to those terms.
13 /// This is the flavor of channels which are not necessarily optimized for any
14 /// particular use case, but are the most general in how they are used. Shared
15 /// channels are cloneable allowing for multiple senders.
17 /// High level implementation details can be found in the comment of the parent
18 /// module. You'll also note that the implementation of the shared and stream
19 /// channels are quite similar, and this is no coincidence!
21 pub use self::Failure::*;
25 use alloc::boxed::Box;
28 use rustrt::local::Local;
29 use rustrt::mutex::NativeMutex;
30 use rustrt::task::{Task, BlockedTask};
31 use rustrt::thread::Thread;
34 use sync::mpsc_queue as mpsc;
36 const DISCONNECTED: int = int::MIN;
37 const FUDGE: int = 1024;
39 const MAX_STEALS: int = 5;
41 const MAX_STEALS: int = 1 << 20;
43 pub struct Packet<T> {
44 queue: mpsc::Queue<T>,
45 cnt: atomic::AtomicInt, // How many items are on this channel
46 steals: int, // How many times has a port received without blocking?
47 to_wake: atomic::AtomicUint, // Task to wake up
49 // The number of channels which are currently using this packet.
50 channels: atomic::AtomicInt,
52 // See the discussion in Port::drop and the channel send methods for what
54 port_dropped: atomic::AtomicBool,
55 sender_drain: atomic::AtomicInt,
57 // this lock protects various portions of this implementation during
59 select_lock: NativeMutex,
67 impl<T: Send> Packet<T> {
68 // Creation of a packet *must* be followed by a call to postinit_lock
69 // and later by inherit_blocker
70 pub fn new() -> Packet<T> {
72 queue: mpsc::Queue::new(),
73 cnt: atomic::AtomicInt::new(0),
75 to_wake: atomic::AtomicUint::new(0),
76 channels: atomic::AtomicInt::new(2),
77 port_dropped: atomic::AtomicBool::new(false),
78 sender_drain: atomic::AtomicInt::new(0),
79 select_lock: unsafe { NativeMutex::new() },
84 // This function should be used after newly created Packet
85 // was wrapped with an Arc
86 // In other case mutex data will be duplicated while cloning
87 // and that could cause problems on platforms where it is
88 // represented by opaque data structure
89 pub fn postinit_lock(&mut self) {
90 unsafe { self.select_lock.lock_noguard() }
93 // This function is used at the creation of a shared packet to inherit a
94 // previously blocked task. This is done to prevent spurious wakeups of
97 // This can only be called at channel-creation time
98 pub fn inherit_blocker(&mut self, task: Option<BlockedTask>) {
101 assert_eq!(self.cnt.load(atomic::SeqCst), 0);
102 assert_eq!(self.to_wake.load(atomic::SeqCst), 0);
103 self.to_wake.store(unsafe { task.cast_to_uint() },
105 self.cnt.store(-1, atomic::SeqCst);
107 // This store is a little sketchy. What's happening here is
108 // that we're transferring a blocker from a oneshot or stream
109 // channel to this shared channel. In doing so, we never
110 // spuriously wake them up and rather only wake them up at the
111 // appropriate time. This implementation of shared channels
112 // assumes that any blocking recv() will undo the increment of
113 // steals performed in try_recv() once the recv is complete.
114 // This thread that we're inheriting, however, is not in the
115 // middle of recv. Hence, the first time we wake them up,
116 // they're going to wake up from their old port, move on to the
117 // upgraded port, and then call the block recv() function.
119 // When calling this function, they'll find there's data
120 // immediately available, counting it as a steal. This in fact
121 // wasn't a steal because we appropriately blocked them waiting
124 // To offset this bad increment, we initially set the steal
125 // count to -1. You'll find some special code in
126 // abort_selection() as well to ensure that this -1 steal count
127 // doesn't escape too far.
133 // When the shared packet is constructed, we grabbed this lock. The
134 // purpose of this lock is to ensure that abort_selection() doesn't
135 // interfere with this method. After we unlock this lock, we're
136 // signifying that we're done modifying self.cnt and self.to_wake and
137 // the port is ready for the world to continue using it.
138 unsafe { self.select_lock.unlock_noguard() }
141 pub fn send(&mut self, t: T) -> Result<(), T> {
142 // See Port::drop for what's going on
143 if self.port_dropped.load(atomic::SeqCst) { return Err(t) }
145 // Note that the multiple sender case is a little trickier
146 // semantically than the single sender case. The logic for
147 // incrementing is "add and if disconnected store disconnected".
148 // This could end up leading some senders to believe that there
149 // wasn't a disconnect if in fact there was a disconnect. This means
150 // that while one thread is attempting to re-store the disconnected
151 // states, other threads could walk through merrily incrementing
152 // this very-negative disconnected count. To prevent senders from
153 // spuriously attempting to send when the channels is actually
154 // disconnected, the count has a ranged check here.
156 // This is also done for another reason. Remember that the return
157 // value of this function is:
159 // `true` == the data *may* be received, this essentially has no
161 // `false` == the data will *never* be received, this has a lot of
164 // In the SPSC case, we have a check of 'queue.is_empty()' to see
165 // whether the data was actually received, but this same condition
166 // means nothing in a multi-producer context. As a result, this
167 // preflight check serves as the definitive "this will never be
168 // received". Once we get beyond this check, we have permanently
169 // entered the realm of "this may be received"
170 if self.cnt.load(atomic::SeqCst) < DISCONNECTED + FUDGE {
175 match self.cnt.fetch_add(1, atomic::SeqCst) {
177 self.take_to_wake().wake().map(|t| t.reawaken());
180 // In this case, we have possibly failed to send our data, and
181 // we need to consider re-popping the data in order to fully
182 // destroy it. We must arbitrate among the multiple senders,
183 // however, because the queues that we're using are
184 // single-consumer queues. In order to do this, all exiting
185 // pushers will use an atomic count in order to count those
186 // flowing through. Pushers who see 0 are required to drain as
187 // much as possible, and then can only exit when they are the
188 // only pusher (otherwise they must try again).
189 n if n < DISCONNECTED + FUDGE => {
190 // see the comment in 'try' for a shared channel for why this
191 // window of "not disconnected" is ok.
192 self.cnt.store(DISCONNECTED, atomic::SeqCst);
194 if self.sender_drain.fetch_add(1, atomic::SeqCst) == 0 {
196 // drain the queue, for info on the thread yield see the
197 // discussion in try_recv
199 match self.queue.pop() {
201 mpsc::Empty => break,
202 mpsc::Inconsistent => Thread::yield_now(),
205 // maybe we're done, if we're not the last ones
206 // here, then we need to go try again.
207 if self.sender_drain.fetch_sub(1, atomic::SeqCst) == 1 {
212 // At this point, there may still be data on the queue,
213 // but only if the count hasn't been incremented and
214 // some other sender hasn't finished pushing data just
215 // yet. That sender in question will drain its own data.
219 // Can't make any assumptions about this case like in the SPSC case.
226 pub fn recv(&mut self) -> Result<T, Failure> {
227 // This code is essentially the exact same as that found in the stream
228 // case (see stream.rs)
229 match self.try_recv() {
234 let task: Box<Task> = Local::take();
235 task.deschedule(1, |task| {
239 match self.try_recv() {
240 data @ Ok(..) => { self.steals -= 1; data }
245 // Essentially the exact same thing as the stream decrement function.
246 fn decrement(&mut self, task: BlockedTask) -> Result<(), BlockedTask> {
247 assert_eq!(self.to_wake.load(atomic::SeqCst), 0);
248 let n = unsafe { task.cast_to_uint() };
249 self.to_wake.store(n, atomic::SeqCst);
251 let steals = self.steals;
254 match self.cnt.fetch_sub(1 + steals, atomic::SeqCst) {
255 DISCONNECTED => { self.cnt.store(DISCONNECTED, atomic::SeqCst); }
256 // If we factor in our steals and notice that the channel has no
257 // data, we successfully sleep
260 if n - steals <= 0 { return Ok(()) }
264 self.to_wake.store(0, atomic::SeqCst);
265 Err(unsafe { BlockedTask::cast_from_uint(n) })
268 pub fn try_recv(&mut self) -> Result<T, Failure> {
269 let ret = match self.queue.pop() {
270 mpsc::Data(t) => Some(t),
273 // This is a bit of an interesting case. The channel is
274 // reported as having data available, but our pop() has
275 // failed due to the queue being in an inconsistent state.
276 // This means that there is some pusher somewhere which has
277 // yet to complete, but we are guaranteed that a pop will
278 // eventually succeed. In this case, we spin in a yield loop
279 // because the remote sender should finish their enqueue
280 // operation "very quickly".
282 // Avoiding this yield loop would require a different queue
283 // abstraction which provides the guarantee that after M
284 // pushes have succeeded, at least M pops will succeed. The
285 // current queues guarantee that if there are N active
286 // pushes, you can pop N times once all N have finished.
287 mpsc::Inconsistent => {
291 match self.queue.pop() {
292 mpsc::Data(t) => { data = t; break }
293 mpsc::Empty => panic!("inconsistent => empty"),
294 mpsc::Inconsistent => {}
301 // See the discussion in the stream implementation for why we
302 // might decrement steals.
304 if self.steals > MAX_STEALS {
305 match self.cnt.swap(0, atomic::SeqCst) {
307 self.cnt.store(DISCONNECTED, atomic::SeqCst);
310 let m = cmp::min(n, self.steals);
315 assert!(self.steals >= 0);
321 // See the discussion in the stream implementation for why we try
324 match self.cnt.load(atomic::SeqCst) {
325 n if n != DISCONNECTED => Err(Empty),
327 match self.queue.pop() {
328 mpsc::Data(t) => Ok(t),
329 mpsc::Empty => Err(Disconnected),
330 // with no senders, an inconsistency is impossible.
331 mpsc::Inconsistent => unreachable!(),
339 // Prepares this shared packet for a channel clone, essentially just bumping
341 pub fn clone_chan(&mut self) {
342 self.channels.fetch_add(1, atomic::SeqCst);
345 // Decrement the reference count on a channel. This is called whenever a
346 // Chan is dropped and may end up waking up a receiver. It's the receiver's
347 // responsibility on the other end to figure out that we've disconnected.
348 pub fn drop_chan(&mut self) {
349 match self.channels.fetch_sub(1, atomic::SeqCst) {
351 n if n > 1 => return,
352 n => panic!("bad number of channels left {}", n),
355 match self.cnt.swap(DISCONNECTED, atomic::SeqCst) {
356 -1 => { self.take_to_wake().wake().map(|t| t.reawaken()); }
358 n => { assert!(n >= 0); }
362 // See the long discussion inside of stream.rs for why the queue is drained,
363 // and why it is done in this fashion.
364 pub fn drop_port(&mut self) {
365 self.port_dropped.store(true, atomic::SeqCst);
366 let mut steals = self.steals;
368 let cnt = self.cnt.compare_and_swap(
369 steals, DISCONNECTED, atomic::SeqCst);
370 cnt != DISCONNECTED && cnt != steals
372 // See the discussion in 'try_recv' for why we yield
373 // control of this thread.
375 match self.queue.pop() {
376 mpsc::Data(..) => { steals += 1; }
377 mpsc::Empty | mpsc::Inconsistent => break,
383 // Consumes ownership of the 'to_wake' field.
384 fn take_to_wake(&mut self) -> BlockedTask {
385 let task = self.to_wake.load(atomic::SeqCst);
386 self.to_wake.store(0, atomic::SeqCst);
388 unsafe { BlockedTask::cast_from_uint(task) }
391 ////////////////////////////////////////////////////////////////////////////
392 // select implementation
393 ////////////////////////////////////////////////////////////////////////////
395 // Helper function for select, tests whether this port can receive without
396 // blocking (obviously not an atomic decision).
398 // This is different than the stream version because there's no need to peek
399 // at the queue, we can just look at the local count.
400 pub fn can_recv(&mut self) -> bool {
401 let cnt = self.cnt.load(atomic::SeqCst);
402 cnt == DISCONNECTED || cnt - self.steals > 0
405 // increment the count on the channel (used for selection)
406 fn bump(&mut self, amt: int) -> int {
407 match self.cnt.fetch_add(amt, atomic::SeqCst) {
409 self.cnt.store(DISCONNECTED, atomic::SeqCst);
416 // Inserts the blocked task for selection on this port, returning it back if
417 // the port already has data on it.
419 // The code here is the same as in stream.rs, except that it doesn't need to
420 // peek at the channel to see if an upgrade is pending.
421 pub fn start_selection(&mut self,
422 task: BlockedTask) -> Result<(), BlockedTask> {
423 match self.decrement(task) {
426 let prev = self.bump(1);
427 assert!(prev == DISCONNECTED || prev >= 0);
433 // Cancels a previous task waiting on this port, returning whether there's
436 // This is similar to the stream implementation (hence fewer comments), but
437 // uses a different value for the "steals" variable.
438 pub fn abort_selection(&mut self, _was_upgrade: bool) -> bool {
439 // Before we do anything else, we bounce on this lock. The reason for
440 // doing this is to ensure that any upgrade-in-progress is gone and
441 // done with. Without this bounce, we can race with inherit_blocker
442 // about looking at and dealing with to_wake. Once we have acquired the
443 // lock, we are guaranteed that inherit_blocker is done.
445 let _guard = self.select_lock.lock();
448 // Like the stream implementation, we want to make sure that the count
449 // on the channel goes non-negative. We don't know how negative the
450 // stream currently is, so instead of using a steal value of 1, we load
451 // the channel count and figure out what we should do to make it
454 let cnt = self.cnt.load(atomic::SeqCst);
455 if cnt < 0 && cnt != DISCONNECTED {-cnt} else {0}
457 let prev = self.bump(steals + 1);
459 if prev == DISCONNECTED {
460 assert_eq!(self.to_wake.load(atomic::SeqCst), 0);
463 let cur = prev + steals + 1;
466 self.take_to_wake().trash();
468 while self.to_wake.load(atomic::SeqCst) != 0 {
472 // if the number of steals is -1, it was the pre-emptive -1 steal
473 // count from when we inherited a blocker. This is fine because
474 // we're just going to overwrite it with a real value.
475 assert!(self.steals == 0 || self.steals == -1);
476 self.steals = steals;
483 impl<T: Send> Drop for Packet<T> {
485 // Note that this load is not only an assert for correctness about
486 // disconnection, but also a proper fence before the read of
487 // `to_wake`, so this assert cannot be removed with also removing
488 // the `to_wake` assert.
489 assert_eq!(self.cnt.load(atomic::SeqCst), DISCONNECTED);
490 assert_eq!(self.to_wake.load(atomic::SeqCst), 0);
491 assert_eq!(self.channels.load(atomic::SeqCst), 0);