aboutsummaryrefslogtreecommitdiff
path: root/src/runtime/scheduler/multi_thread_alt/queue.rs
blob: e41e9fdb3a6ec0acac11972e6a515644d642a76f (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
//! Run-queue structures to support a work-stealing scheduler

use crate::loom::cell::UnsafeCell;
use crate::loom::sync::Arc;
use crate::runtime::scheduler::multi_thread_alt::{Overflow, Stats};
use crate::runtime::task;

use std::mem::{self, MaybeUninit};
use std::ptr;
use std::sync::atomic::Ordering::{AcqRel, Acquire, Relaxed, Release};

// Use wider integers when possible to increase ABA resilience.
//
// See issue #5041: <https://github.com/tokio-rs/tokio/issues/5041>.
cfg_has_atomic_u64! {
    type UnsignedShort = u32;
    type UnsignedLong = u64;
    type AtomicUnsignedShort = crate::loom::sync::atomic::AtomicU32;
    type AtomicUnsignedLong = crate::loom::sync::atomic::AtomicU64;
}
cfg_not_has_atomic_u64! {
    type UnsignedShort = u16;
    type UnsignedLong = u32;
    type AtomicUnsignedShort = crate::loom::sync::atomic::AtomicU16;
    type AtomicUnsignedLong = crate::loom::sync::atomic::AtomicU32;
}

/// Producer handle. May only be used from a single thread.
pub(crate) struct Local<T: 'static> {
    inner: Arc<Inner<T>>,
}

/// Consumer handle. May be used from many threads.
pub(crate) struct Steal<T: 'static>(Arc<Inner<T>>);

#[repr(align(128))]
pub(crate) struct Inner<T: 'static> {
    /// Concurrently updated by many threads.
    ///
    /// Contains two `UnsignedShort` values. The LSB byte is the "real" head of
    /// the queue. The `UnsignedShort` in the MSB is set by a stealer in process
    /// of stealing values. It represents the first value being stolen in the
    /// batch. The `UnsignedShort` indices are intentionally wider than strictly
    /// required for buffer indexing in order to provide ABA mitigation and make
    /// it possible to distinguish between full and empty buffers.
    ///
    /// When both `UnsignedShort` values are the same, there is no active
    /// stealer.
    ///
    /// Tracking an in-progress stealer prevents a wrapping scenario.
    head: AtomicUnsignedLong,

    /// Only updated by producer thread but read by many threads.
    tail: AtomicUnsignedShort,

    /// Elements
    buffer: Box<[UnsafeCell<MaybeUninit<task::Notified<T>>>]>,

    mask: usize,
}

unsafe impl<T> Send for Inner<T> {}
unsafe impl<T> Sync for Inner<T> {}

/// Create a new local run-queue
pub(crate) fn local<T: 'static>(capacity: usize) -> (Steal<T>, Local<T>) {
    assert!(capacity <= 4096);
    assert!(capacity >= 1);

    let mut buffer = Vec::with_capacity(capacity);

    for _ in 0..capacity {
        buffer.push(UnsafeCell::new(MaybeUninit::uninit()));
    }

    let inner = Arc::new(Inner {
        head: AtomicUnsignedLong::new(0),
        tail: AtomicUnsignedShort::new(0),
        buffer: buffer.into_boxed_slice(),
        mask: capacity - 1,
    });

    let local = Local {
        inner: inner.clone(),
    };

    let remote = Steal(inner);

    (remote, local)
}

impl<T> Local<T> {
    /// How many tasks can be pushed into the queue
    pub(crate) fn remaining_slots(&self) -> usize {
        self.inner.remaining_slots()
    }

    pub(crate) fn max_capacity(&self) -> usize {
        self.inner.buffer.len()
    }

    /// Returns `true` if there are no entries in the queue
    pub(crate) fn is_empty(&self) -> bool {
        self.inner.is_empty()
    }

    pub(crate) fn can_steal(&self) -> bool {
        self.remaining_slots() >= self.max_capacity() - self.max_capacity() / 2
    }

    /// Pushes a batch of tasks to the back of the queue. All tasks must fit in
    /// the local queue.
    ///
    /// # Panics
    ///
    /// The method panics if there is not enough capacity to fit in the queue.
    pub(crate) fn push_back(&mut self, tasks: impl ExactSizeIterator<Item = task::Notified<T>>) {
        let len = tasks.len();
        assert!(len <= self.inner.buffer.len());

        if len == 0 {
            // Nothing to do
            return;
        }

        let head = self.inner.head.load(Acquire);
        let (steal, real) = unpack(head);

        // safety: this is the **only** thread that updates this cell.
        let mut tail = unsafe { self.inner.tail.unsync_load() };

        if tail.wrapping_sub(steal) <= (self.inner.buffer.len() - len) as UnsignedShort {
            // Yes, this if condition is structured a bit weird (first block
            // does nothing, second returns an error). It is this way to match
            // `push_back_or_overflow`.
        } else {
            panic!(
                "not enough capacity; len={}; tail={}; steal={}; real={}",
                len, tail, steal, real
            );
        }

        for task in tasks {
            let idx = tail as usize & self.inner.mask;

            self.inner.buffer[idx].with_mut(|ptr| {
                // Write the task to the slot
                //
                // Safety: There is only one producer and the above `if`
                // condition ensures we don't touch a cell if there is a
                // value, thus no consumer.
                unsafe {
                    ptr::write((*ptr).as_mut_ptr(), task);
                }
            });

            tail = tail.wrapping_add(1);
        }

        self.inner.tail.store(tail, Release);
    }

    /// Pushes a task to the back of the local queue, if there is not enough
    /// capacity in the queue, this triggers the overflow operation.
    ///
    /// When the queue overflows, half of the curent contents of the queue is
    /// moved to the given Injection queue. This frees up capacity for more
    /// tasks to be pushed into the local queue.
    pub(crate) fn push_back_or_overflow<O: Overflow<T>>(
        &mut self,
        mut task: task::Notified<T>,
        overflow: &O,
        stats: &mut Stats,
    ) {
        let tail = loop {
            let head = self.inner.head.load(Acquire);
            let (steal, real) = unpack(head);

            // safety: this is the **only** thread that updates this cell.
            let tail = unsafe { self.inner.tail.unsync_load() };

            if tail.wrapping_sub(steal) < self.inner.buffer.len() as UnsignedShort {
                // There is capacity for the task
                break tail;
            } else if steal != real {
                super::counters::inc_num_overflows();
                // Concurrently stealing, this will free up capacity, so only
                // push the task onto the inject queue
                overflow.push(task);
                return;
            } else {
                super::counters::inc_num_overflows();
                // Push the current task and half of the queue into the
                // inject queue.
                match self.push_overflow(task, real, tail, overflow, stats) {
                    Ok(_) => return,
                    // Lost the race, try again
                    Err(v) => {
                        task = v;
                    }
                }
            }
        };

        self.push_back_finish(task, tail);
    }

    // Second half of `push_back`
    fn push_back_finish(&self, task: task::Notified<T>, tail: UnsignedShort) {
        // Map the position to a slot index.
        let idx = tail as usize & self.inner.mask;

        self.inner.buffer[idx].with_mut(|ptr| {
            // Write the task to the slot
            //
            // Safety: There is only one producer and the above `if`
            // condition ensures we don't touch a cell if there is a
            // value, thus no consumer.
            unsafe {
                ptr::write((*ptr).as_mut_ptr(), task);
            }
        });

        // Make the task available. Synchronizes with a load in
        // `steal_into2`.
        self.inner.tail.store(tail.wrapping_add(1), Release);
    }

    /// Moves a batch of tasks into the inject queue.
    ///
    /// This will temporarily make some of the tasks unavailable to stealers.
    /// Once `push_overflow` is done, a notification is sent out, so if other
    /// workers "missed" some of the tasks during a steal, they will get
    /// another opportunity.
    #[inline(never)]
    fn push_overflow<O: Overflow<T>>(
        &mut self,
        task: task::Notified<T>,
        head: UnsignedShort,
        tail: UnsignedShort,
        overflow: &O,
        stats: &mut Stats,
    ) -> Result<(), task::Notified<T>> {
        // How many elements are we taking from the local queue.
        //
        // This is one less than the number of tasks pushed to the inject
        // queue as we are also inserting the `task` argument.
        let num_tasks_taken: UnsignedShort = (self.inner.buffer.len() / 2) as UnsignedShort;

        assert_eq!(
            tail.wrapping_sub(head) as usize,
            self.inner.buffer.len(),
            "queue is not full; tail = {}; head = {}",
            tail,
            head
        );

        let prev = pack(head, head);

        // Claim a bunch of tasks
        //
        // We are claiming the tasks **before** reading them out of the buffer.
        // This is safe because only the **current** thread is able to push new
        // tasks.
        //
        // There isn't really any need for memory ordering... Relaxed would
        // work. This is because all tasks are pushed into the queue from the
        // current thread (or memory has been acquired if the local queue handle
        // moved).
        if self
            .inner
            .head
            .compare_exchange(
                prev,
                pack(
                    head.wrapping_add(num_tasks_taken),
                    head.wrapping_add(num_tasks_taken),
                ),
                Release,
                Relaxed,
            )
            .is_err()
        {
            // We failed to claim the tasks, losing the race. Return out of
            // this function and try the full `push` routine again. The queue
            // may not be full anymore.
            return Err(task);
        }

        /// An iterator that takes elements out of the run queue.
        struct BatchTaskIter<'a, T: 'static> {
            buffer: &'a [UnsafeCell<MaybeUninit<task::Notified<T>>>],
            mask: usize,
            head: UnsignedLong,
            i: UnsignedLong,
            num: UnsignedShort,
        }
        impl<'a, T: 'static> Iterator for BatchTaskIter<'a, T> {
            type Item = task::Notified<T>;

            #[inline]
            fn next(&mut self) -> Option<task::Notified<T>> {
                if self.i == UnsignedLong::from(self.num) {
                    None
                } else {
                    let i_idx = self.i.wrapping_add(self.head) as usize & self.mask;
                    let slot = &self.buffer[i_idx];

                    // safety: Our CAS from before has assumed exclusive ownership
                    // of the task pointers in this range.
                    let task = slot.with(|ptr| unsafe { ptr::read((*ptr).as_ptr()) });

                    self.i += 1;
                    Some(task)
                }
            }
        }

        // safety: The CAS above ensures that no consumer will look at these
        // values again, and we are the only producer.
        let batch_iter = BatchTaskIter {
            buffer: &self.inner.buffer,
            mask: self.inner.mask,
            head: head as UnsignedLong,
            i: 0,
            num: num_tasks_taken,
        };
        overflow.push_batch(batch_iter.chain(std::iter::once(task)));

        // Add 1 to factor in the task currently being scheduled.
        stats.incr_overflow_count();

        Ok(())
    }

    /// Pops a task from the local queue.
    pub(crate) fn pop(&mut self) -> Option<task::Notified<T>> {
        let mut head = self.inner.head.load(Acquire);

        let idx = loop {
            let (steal, real) = unpack(head);

            // safety: this is the **only** thread that updates this cell.
            let tail = unsafe { self.inner.tail.unsync_load() };

            if real == tail {
                // queue is empty
                return None;
            }

            let next_real = real.wrapping_add(1);

            // If `steal == real` there are no concurrent stealers. Both `steal`
            // and `real` are updated.
            let next = if steal == real {
                pack(next_real, next_real)
            } else {
                assert_ne!(steal, next_real);
                pack(steal, next_real)
            };

            // Attempt to claim a task.
            let res = self
                .inner
                .head
                .compare_exchange(head, next, AcqRel, Acquire);

            match res {
                Ok(_) => break real as usize & self.inner.mask,
                Err(actual) => head = actual,
            }
        };

        Some(self.inner.buffer[idx].with(|ptr| unsafe { ptr::read(ptr).assume_init() }))
    }
}

impl<T> Steal<T> {
    pub(crate) fn is_empty(&self) -> bool {
        self.0.is_empty()
    }

    /// Steals half the tasks from self and place them into `dst`.
    pub(crate) fn steal_into(
        &self,
        dst: &mut Local<T>,
        dst_stats: &mut Stats,
    ) -> Option<task::Notified<T>> {
        // Safety: the caller is the only thread that mutates `dst.tail` and
        // holds a mutable reference.
        let dst_tail = unsafe { dst.inner.tail.unsync_load() };

        // To the caller, `dst` may **look** empty but still have values
        // contained in the buffer. If another thread is concurrently stealing
        // from `dst` there may not be enough capacity to steal.
        let (steal, _) = unpack(dst.inner.head.load(Acquire));

        if dst_tail.wrapping_sub(steal) > self.0.buffer.len() as UnsignedShort / 2 {
            // we *could* try to steal less here, but for simplicity, we're just
            // going to abort.
            return None;
        }

        // Steal the tasks into `dst`'s buffer. This does not yet expose the
        // tasks in `dst`.
        let mut n = self.steal_into2(dst, dst_tail);

        if n == 0 {
            // No tasks were stolen
            return None;
        }

        super::counters::inc_num_steals();

        dst_stats.incr_steal_count(n as u16);
        dst_stats.incr_steal_operations();

        // We are returning a task here
        n -= 1;

        let ret_pos = dst_tail.wrapping_add(n);
        let ret_idx = ret_pos as usize & dst.inner.mask;

        // safety: the value was written as part of `steal_into2` and not
        // exposed to stealers, so no other thread can access it.
        let ret = dst.inner.buffer[ret_idx].with(|ptr| unsafe { ptr::read((*ptr).as_ptr()) });

        if n == 0 {
            // The `dst` queue is empty, but a single task was stolen
            return Some(ret);
        }

        // Make the stolen items available to consumers
        dst.inner.tail.store(dst_tail.wrapping_add(n), Release);

        Some(ret)
    }

    // Steal tasks from `self`, placing them into `dst`. Returns the number of
    // tasks that were stolen.
    fn steal_into2(&self, dst: &mut Local<T>, dst_tail: UnsignedShort) -> UnsignedShort {
        let mut prev_packed = self.0.head.load(Acquire);
        let mut next_packed;

        let n = loop {
            let (src_head_steal, src_head_real) = unpack(prev_packed);
            let src_tail = self.0.tail.load(Acquire);

            // If these two do not match, another thread is concurrently
            // stealing from the queue.
            if src_head_steal != src_head_real {
                return 0;
            }

            // Number of available tasks to steal
            let n = src_tail.wrapping_sub(src_head_real);
            let n = n - n / 2;

            if n == 0 {
                // No tasks available to steal
                return 0;
            }

            // Update the real head index to acquire the tasks.
            let steal_to = src_head_real.wrapping_add(n);
            assert_ne!(src_head_steal, steal_to);
            next_packed = pack(src_head_steal, steal_to);

            // Claim all those tasks. This is done by incrementing the "real"
            // head but not the steal. By doing this, no other thread is able to
            // steal from this queue until the current thread completes.
            let res = self
                .0
                .head
                .compare_exchange(prev_packed, next_packed, AcqRel, Acquire);

            match res {
                Ok(_) => break n,
                Err(actual) => prev_packed = actual,
            }
        };

        debug_assert!(
            n <= (self.0.buffer.len() - self.0.buffer.len() / 2) as UnsignedShort,
            "actual = {}",
            n
        );

        let (first, _) = unpack(next_packed);

        // Take all the tasks
        for i in 0..n {
            // Compute the positions
            let src_pos = first.wrapping_add(i);
            let dst_pos = dst_tail.wrapping_add(i);

            // Map to slots
            let src_idx = src_pos as usize & self.0.mask;
            let dst_idx = dst_pos as usize & self.0.mask;

            // Read the task
            //
            // safety: We acquired the task with the atomic exchange above.
            let task = self.0.buffer[src_idx].with(|ptr| unsafe { ptr::read((*ptr).as_ptr()) });

            // Write the task to the new slot
            //
            // safety: `dst` queue is empty and we are the only producer to
            // this queue.
            dst.inner.buffer[dst_idx]
                .with_mut(|ptr| unsafe { ptr::write((*ptr).as_mut_ptr(), task) });
        }

        let mut prev_packed = next_packed;

        // Update `src_head_steal` to match `src_head_real` signalling that the
        // stealing routine is complete.
        loop {
            let head = unpack(prev_packed).1;
            next_packed = pack(head, head);

            let res = self
                .0
                .head
                .compare_exchange(prev_packed, next_packed, AcqRel, Acquire);

            match res {
                Ok(_) => return n,
                Err(actual) => {
                    let (actual_steal, actual_real) = unpack(actual);

                    assert_ne!(actual_steal, actual_real);

                    prev_packed = actual;
                }
            }
        }
    }
}

cfg_metrics! {
    impl<T> Steal<T> {
        pub(crate) fn len(&self) -> usize {
            self.0.len() as _
        }
    }
}

impl<T> Clone for Steal<T> {
    fn clone(&self) -> Steal<T> {
        Steal(self.0.clone())
    }
}

impl<T> Drop for Local<T> {
    fn drop(&mut self) {
        if !std::thread::panicking() {
            assert!(self.pop().is_none(), "queue not empty");
        }
    }
}

impl<T> Inner<T> {
    fn remaining_slots(&self) -> usize {
        let (steal, _) = unpack(self.head.load(Acquire));
        let tail = self.tail.load(Acquire);

        self.buffer.len() - (tail.wrapping_sub(steal) as usize)
    }

    fn len(&self) -> UnsignedShort {
        let (_, head) = unpack(self.head.load(Acquire));
        let tail = self.tail.load(Acquire);

        tail.wrapping_sub(head)
    }

    fn is_empty(&self) -> bool {
        self.len() == 0
    }
}

/// Split the head value into the real head and the index a stealer is working
/// on.
fn unpack(n: UnsignedLong) -> (UnsignedShort, UnsignedShort) {
    let real = n & UnsignedShort::MAX as UnsignedLong;
    let steal = n >> (mem::size_of::<UnsignedShort>() * 8);

    (steal as UnsignedShort, real as UnsignedShort)
}

/// Join the two head values
fn pack(steal: UnsignedShort, real: UnsignedShort) -> UnsignedLong {
    (real as UnsignedLong) | ((steal as UnsignedLong) << (mem::size_of::<UnsignedShort>() * 8))
}