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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029
use crate::job::{JobFifo, JobRef, StackJob};
use crate::latch::{AsCoreLatch, CoreLatch, CountLatch, Latch, LatchRef, LockLatch, SpinLatch};
use crate::log::Event::*;
use crate::log::Logger;
use crate::sleep::Sleep;
use crate::unwind;
use crate::{
ErrorKind, ExitHandler, PanicHandler, StartHandler, ThreadPoolBuildError, ThreadPoolBuilder,
Yield,
};
use crossbeam_deque::{Injector, Steal, Stealer, Worker};
use std::cell::Cell;
use std::collections::hash_map::DefaultHasher;
use std::fmt;
use std::hash::Hasher;
use std::io;
use std::mem;
use std::ptr;
use std::sync::atomic::{AtomicUsize, Ordering};
use std::sync::{Arc, Mutex, Once};
use std::thread;
use std::usize;
/// Thread builder used for customization via
/// [`ThreadPoolBuilder::spawn_handler`](struct.ThreadPoolBuilder.html#method.spawn_handler).
pub struct ThreadBuilder {
name: Option<String>,
stack_size: Option<usize>,
worker: Worker<JobRef>,
stealer: Stealer<JobRef>,
registry: Arc<Registry>,
index: usize,
}
impl ThreadBuilder {
/// Gets the index of this thread in the pool, within `0..num_threads`.
pub fn index(&self) -> usize {
self.index
}
/// Gets the string that was specified by `ThreadPoolBuilder::name()`.
pub fn name(&self) -> Option<&str> {
self.name.as_deref()
}
/// Gets the value that was specified by `ThreadPoolBuilder::stack_size()`.
pub fn stack_size(&self) -> Option<usize> {
self.stack_size
}
/// Executes the main loop for this thread. This will not return until the
/// thread pool is dropped.
pub fn run(self) {
unsafe { main_loop(self) }
}
}
impl fmt::Debug for ThreadBuilder {
fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
f.debug_struct("ThreadBuilder")
.field("pool", &self.registry.id())
.field("index", &self.index)
.field("name", &self.name)
.field("stack_size", &self.stack_size)
.finish()
}
}
/// Generalized trait for spawning a thread in the `Registry`.
///
/// This trait is pub-in-private -- E0445 forces us to make it public,
/// but we don't actually want to expose these details in the API.
pub trait ThreadSpawn {
private_decl! {}
/// Spawn a thread with the `ThreadBuilder` parameters, and then
/// call `ThreadBuilder::run()`.
fn spawn(&mut self, thread: ThreadBuilder) -> io::Result<()>;
}
/// Spawns a thread in the "normal" way with `std::thread::Builder`.
///
/// This type is pub-in-private -- E0445 forces us to make it public,
/// but we don't actually want to expose these details in the API.
#[derive(Debug, Default)]
pub struct DefaultSpawn;
impl ThreadSpawn for DefaultSpawn {
private_impl! {}
fn spawn(&mut self, thread: ThreadBuilder) -> io::Result<()> {
let mut b = thread::Builder::new();
if let Some(name) = thread.name() {
b = b.name(name.to_owned());
}
if let Some(stack_size) = thread.stack_size() {
b = b.stack_size(stack_size);
}
b.spawn(|| thread.run())?;
Ok(())
}
}
/// Spawns a thread with a user's custom callback.
///
/// This type is pub-in-private -- E0445 forces us to make it public,
/// but we don't actually want to expose these details in the API.
#[derive(Debug)]
pub struct CustomSpawn<F>(F);
impl<F> CustomSpawn<F>
where
F: FnMut(ThreadBuilder) -> io::Result<()>,
{
pub(super) fn new(spawn: F) -> Self {
CustomSpawn(spawn)
}
}
impl<F> ThreadSpawn for CustomSpawn<F>
where
F: FnMut(ThreadBuilder) -> io::Result<()>,
{
private_impl! {}
#[inline]
fn spawn(&mut self, thread: ThreadBuilder) -> io::Result<()> {
(self.0)(thread)
}
}
pub(super) struct Registry {
logger: Logger,
thread_infos: Vec<ThreadInfo>,
sleep: Sleep,
injected_jobs: Injector<JobRef>,
broadcasts: Mutex<Vec<Worker<JobRef>>>,
panic_handler: Option<Box<PanicHandler>>,
start_handler: Option<Box<StartHandler>>,
exit_handler: Option<Box<ExitHandler>>,
// When this latch reaches 0, it means that all work on this
// registry must be complete. This is ensured in the following ways:
//
// - if this is the global registry, there is a ref-count that never
// gets released.
// - if this is a user-created thread-pool, then so long as the thread-pool
// exists, it holds a reference.
// - when we inject a "blocking job" into the registry with `ThreadPool::install()`,
// no adjustment is needed; the `ThreadPool` holds the reference, and since we won't
// return until the blocking job is complete, that ref will continue to be held.
// - when `join()` or `scope()` is invoked, similarly, no adjustments are needed.
// These are always owned by some other job (e.g., one injected by `ThreadPool::install()`)
// and that job will keep the pool alive.
terminate_count: AtomicUsize,
}
/// ////////////////////////////////////////////////////////////////////////
/// Initialization
static mut THE_REGISTRY: Option<Arc<Registry>> = None;
static THE_REGISTRY_SET: Once = Once::new();
/// Starts the worker threads (if that has not already happened). If
/// initialization has not already occurred, use the default
/// configuration.
pub(super) fn global_registry() -> &'static Arc<Registry> {
set_global_registry(default_global_registry)
.or_else(|err| unsafe { THE_REGISTRY.as_ref().ok_or(err) })
.expect("The global thread pool has not been initialized.")
}
/// Starts the worker threads (if that has not already happened) with
/// the given builder.
pub(super) fn init_global_registry<S>(
builder: ThreadPoolBuilder<S>,
) -> Result<&'static Arc<Registry>, ThreadPoolBuildError>
where
S: ThreadSpawn,
{
set_global_registry(|| Registry::new(builder))
}
/// Starts the worker threads (if that has not already happened)
/// by creating a registry with the given callback.
fn set_global_registry<F>(registry: F) -> Result<&'static Arc<Registry>, ThreadPoolBuildError>
where
F: FnOnce() -> Result<Arc<Registry>, ThreadPoolBuildError>,
{
let mut result = Err(ThreadPoolBuildError::new(
ErrorKind::GlobalPoolAlreadyInitialized,
));
THE_REGISTRY_SET.call_once(|| {
result = registry()
.map(|registry: Arc<Registry>| unsafe { &*THE_REGISTRY.get_or_insert(registry) })
});
result
}
fn default_global_registry() -> Result<Arc<Registry>, ThreadPoolBuildError> {
let result = Registry::new(ThreadPoolBuilder::new());
// If we're running in an environment that doesn't support threads at all, we can fall back to
// using the current thread alone. This is crude, and probably won't work for non-blocking
// calls like `spawn` or `broadcast_spawn`, but a lot of stuff does work fine.
//
// Notably, this allows current WebAssembly targets to work even though their threading support
// is stubbed out, and we won't have to change anything if they do add real threading.
let unsupported = matches!(&result, Err(e) if e.is_unsupported());
if unsupported && WorkerThread::current().is_null() {
let builder = ThreadPoolBuilder::new()
.num_threads(1)
.spawn_handler(|thread| {
// Rather than starting a new thread, we're just taking over the current thread
// *without* running the main loop, so we can still return from here.
// The WorkerThread is leaked, but we never shutdown the global pool anyway.
let worker_thread = Box::leak(Box::new(WorkerThread::from(thread)));
let registry = &*worker_thread.registry;
let index = worker_thread.index;
unsafe {
WorkerThread::set_current(worker_thread);
// let registry know we are ready to do work
Latch::set(®istry.thread_infos[index].primed);
}
Ok(())
});
let fallback_result = Registry::new(builder);
if fallback_result.is_ok() {
return fallback_result;
}
}
result
}
struct Terminator<'a>(&'a Arc<Registry>);
impl<'a> Drop for Terminator<'a> {
fn drop(&mut self) {
self.0.terminate()
}
}
impl Registry {
pub(super) fn new<S>(
mut builder: ThreadPoolBuilder<S>,
) -> Result<Arc<Self>, ThreadPoolBuildError>
where
S: ThreadSpawn,
{
// Soft-limit the number of threads that we can actually support.
let n_threads = Ord::min(builder.get_num_threads(), crate::max_num_threads());
let breadth_first = builder.get_breadth_first();
let (workers, stealers): (Vec<_>, Vec<_>) = (0..n_threads)
.map(|_| {
let worker = if breadth_first {
Worker::new_fifo()
} else {
Worker::new_lifo()
};
let stealer = worker.stealer();
(worker, stealer)
})
.unzip();
let (broadcasts, broadcast_stealers): (Vec<_>, Vec<_>) = (0..n_threads)
.map(|_| {
let worker = Worker::new_fifo();
let stealer = worker.stealer();
(worker, stealer)
})
.unzip();
let logger = Logger::new(n_threads);
let registry = Arc::new(Registry {
logger: logger.clone(),
thread_infos: stealers.into_iter().map(ThreadInfo::new).collect(),
sleep: Sleep::new(logger, n_threads),
injected_jobs: Injector::new(),
broadcasts: Mutex::new(broadcasts),
terminate_count: AtomicUsize::new(1),
panic_handler: builder.take_panic_handler(),
start_handler: builder.take_start_handler(),
exit_handler: builder.take_exit_handler(),
});
// If we return early or panic, make sure to terminate existing threads.
let t1000 = Terminator(®istry);
for (index, (worker, stealer)) in workers.into_iter().zip(broadcast_stealers).enumerate() {
let thread = ThreadBuilder {
name: builder.get_thread_name(index),
stack_size: builder.get_stack_size(),
registry: Arc::clone(®istry),
worker,
stealer,
index,
};
if let Err(e) = builder.get_spawn_handler().spawn(thread) {
return Err(ThreadPoolBuildError::new(ErrorKind::IOError(e)));
}
}
// Returning normally now, without termination.
mem::forget(t1000);
Ok(registry)
}
pub(super) fn current() -> Arc<Registry> {
unsafe {
let worker_thread = WorkerThread::current();
let registry = if worker_thread.is_null() {
global_registry()
} else {
&(*worker_thread).registry
};
Arc::clone(registry)
}
}
/// Returns the number of threads in the current registry. This
/// is better than `Registry::current().num_threads()` because it
/// avoids incrementing the `Arc`.
pub(super) fn current_num_threads() -> usize {
unsafe {
let worker_thread = WorkerThread::current();
if worker_thread.is_null() {
global_registry().num_threads()
} else {
(*worker_thread).registry.num_threads()
}
}
}
/// Returns the current `WorkerThread` if it's part of this `Registry`.
pub(super) fn current_thread(&self) -> Option<&WorkerThread> {
unsafe {
let worker = WorkerThread::current().as_ref()?;
if worker.registry().id() == self.id() {
Some(worker)
} else {
None
}
}
}
/// Returns an opaque identifier for this registry.
pub(super) fn id(&self) -> RegistryId {
// We can rely on `self` not to change since we only ever create
// registries that are boxed up in an `Arc` (see `new()` above).
RegistryId {
addr: self as *const Self as usize,
}
}
#[inline]
pub(super) fn log(&self, event: impl FnOnce() -> crate::log::Event) {
self.logger.log(event)
}
pub(super) fn num_threads(&self) -> usize {
self.thread_infos.len()
}
pub(super) fn catch_unwind(&self, f: impl FnOnce()) {
if let Err(err) = unwind::halt_unwinding(f) {
// If there is no handler, or if that handler itself panics, then we abort.
let abort_guard = unwind::AbortIfPanic;
if let Some(ref handler) = self.panic_handler {
handler(err);
mem::forget(abort_guard);
}
}
}
/// Waits for the worker threads to get up and running. This is
/// meant to be used for benchmarking purposes, primarily, so that
/// you can get more consistent numbers by having everything
/// "ready to go".
pub(super) fn wait_until_primed(&self) {
for info in &self.thread_infos {
info.primed.wait();
}
}
/// Waits for the worker threads to stop. This is used for testing
/// -- so we can check that termination actually works.
#[cfg(test)]
pub(super) fn wait_until_stopped(&self) {
for info in &self.thread_infos {
info.stopped.wait();
}
}
/// ////////////////////////////////////////////////////////////////////////
/// MAIN LOOP
///
/// So long as all of the worker threads are hanging out in their
/// top-level loop, there is no work to be done.
/// Push a job into the given `registry`. If we are running on a
/// worker thread for the registry, this will push onto the
/// deque. Else, it will inject from the outside (which is slower).
pub(super) fn inject_or_push(&self, job_ref: JobRef) {
let worker_thread = WorkerThread::current();
unsafe {
if !worker_thread.is_null() && (*worker_thread).registry().id() == self.id() {
(*worker_thread).push(job_ref);
} else {
self.inject(job_ref);
}
}
}
/// Push a job into the "external jobs" queue; it will be taken by
/// whatever worker has nothing to do. Use this if you know that
/// you are not on a worker of this registry.
pub(super) fn inject(&self, injected_job: JobRef) {
self.log(|| JobsInjected { count: 1 });
// It should not be possible for `state.terminate` to be true
// here. It is only set to true when the user creates (and
// drops) a `ThreadPool`; and, in that case, they cannot be
// calling `inject()` later, since they dropped their
// `ThreadPool`.
debug_assert_ne!(
self.terminate_count.load(Ordering::Acquire),
0,
"inject() sees state.terminate as true"
);
let queue_was_empty = self.injected_jobs.is_empty();
self.injected_jobs.push(injected_job);
self.sleep.new_injected_jobs(usize::MAX, 1, queue_was_empty);
}
fn has_injected_job(&self) -> bool {
!self.injected_jobs.is_empty()
}
fn pop_injected_job(&self, worker_index: usize) -> Option<JobRef> {
loop {
match self.injected_jobs.steal() {
Steal::Success(job) => {
self.log(|| JobUninjected {
worker: worker_index,
});
return Some(job);
}
Steal::Empty => return None,
Steal::Retry => {}
}
}
}
/// Push a job into each thread's own "external jobs" queue; it will be
/// executed only on that thread, when it has nothing else to do locally,
/// before it tries to steal other work.
///
/// **Panics** if not given exactly as many jobs as there are threads.
pub(super) fn inject_broadcast(&self, injected_jobs: impl ExactSizeIterator<Item = JobRef>) {
assert_eq!(self.num_threads(), injected_jobs.len());
self.log(|| JobBroadcast {
count: self.num_threads(),
});
{
let broadcasts = self.broadcasts.lock().unwrap();
// It should not be possible for `state.terminate` to be true
// here. It is only set to true when the user creates (and
// drops) a `ThreadPool`; and, in that case, they cannot be
// calling `inject_broadcast()` later, since they dropped their
// `ThreadPool`.
debug_assert_ne!(
self.terminate_count.load(Ordering::Acquire),
0,
"inject_broadcast() sees state.terminate as true"
);
assert_eq!(broadcasts.len(), injected_jobs.len());
for (worker, job_ref) in broadcasts.iter().zip(injected_jobs) {
worker.push(job_ref);
}
}
for i in 0..self.num_threads() {
self.sleep.notify_worker_latch_is_set(i);
}
}
/// If already in a worker-thread of this registry, just execute `op`.
/// Otherwise, inject `op` in this thread-pool. Either way, block until `op`
/// completes and return its return value. If `op` panics, that panic will
/// be propagated as well. The second argument indicates `true` if injection
/// was performed, `false` if executed directly.
pub(super) fn in_worker<OP, R>(&self, op: OP) -> R
where
OP: FnOnce(&WorkerThread, bool) -> R + Send,
R: Send,
{
unsafe {
let worker_thread = WorkerThread::current();
if worker_thread.is_null() {
self.in_worker_cold(op)
} else if (*worker_thread).registry().id() != self.id() {
self.in_worker_cross(&*worker_thread, op)
} else {
// Perfectly valid to give them a `&T`: this is the
// current thread, so we know the data structure won't be
// invalidated until we return.
op(&*worker_thread, false)
}
}
}
#[cold]
unsafe fn in_worker_cold<OP, R>(&self, op: OP) -> R
where
OP: FnOnce(&WorkerThread, bool) -> R + Send,
R: Send,
{
thread_local!(static LOCK_LATCH: LockLatch = LockLatch::new());
LOCK_LATCH.with(|l| {
// This thread isn't a member of *any* thread pool, so just block.
debug_assert!(WorkerThread::current().is_null());
let job = StackJob::new(
|injected| {
let worker_thread = WorkerThread::current();
assert!(injected && !worker_thread.is_null());
op(&*worker_thread, true)
},
LatchRef::new(l),
);
self.inject(job.as_job_ref());
job.latch.wait_and_reset(); // Make sure we can use the same latch again next time.
// flush accumulated logs as we exit the thread
self.logger.log(|| Flush);
job.into_result()
})
}
#[cold]
unsafe fn in_worker_cross<OP, R>(&self, current_thread: &WorkerThread, op: OP) -> R
where
OP: FnOnce(&WorkerThread, bool) -> R + Send,
R: Send,
{
// This thread is a member of a different pool, so let it process
// other work while waiting for this `op` to complete.
debug_assert!(current_thread.registry().id() != self.id());
let latch = SpinLatch::cross(current_thread);
let job = StackJob::new(
|injected| {
let worker_thread = WorkerThread::current();
assert!(injected && !worker_thread.is_null());
op(&*worker_thread, true)
},
latch,
);
self.inject(job.as_job_ref());
current_thread.wait_until(&job.latch);
job.into_result()
}
/// Increments the terminate counter. This increment should be
/// balanced by a call to `terminate`, which will decrement. This
/// is used when spawning asynchronous work, which needs to
/// prevent the registry from terminating so long as it is active.
///
/// Note that blocking functions such as `join` and `scope` do not
/// need to concern themselves with this fn; their context is
/// responsible for ensuring the current thread-pool will not
/// terminate until they return.
///
/// The global thread-pool always has an outstanding reference
/// (the initial one). Custom thread-pools have one outstanding
/// reference that is dropped when the `ThreadPool` is dropped:
/// since installing the thread-pool blocks until any joins/scopes
/// complete, this ensures that joins/scopes are covered.
///
/// The exception is `::spawn()`, which can create a job outside
/// of any blocking scope. In that case, the job itself holds a
/// terminate count and is responsible for invoking `terminate()`
/// when finished.
pub(super) fn increment_terminate_count(&self) {
let previous = self.terminate_count.fetch_add(1, Ordering::AcqRel);
debug_assert!(previous != 0, "registry ref count incremented from zero");
assert!(
previous != std::usize::MAX,
"overflow in registry ref count"
);
}
/// Signals that the thread-pool which owns this registry has been
/// dropped. The worker threads will gradually terminate, once any
/// extant work is completed.
pub(super) fn terminate(&self) {
if self.terminate_count.fetch_sub(1, Ordering::AcqRel) == 1 {
for (i, thread_info) in self.thread_infos.iter().enumerate() {
unsafe { CountLatch::set_and_tickle_one(&thread_info.terminate, self, i) };
}
}
}
/// Notify the worker that the latch they are sleeping on has been "set".
pub(super) fn notify_worker_latch_is_set(&self, target_worker_index: usize) {
self.sleep.notify_worker_latch_is_set(target_worker_index);
}
}
#[derive(Copy, Clone, Debug, PartialEq, Eq, PartialOrd, Ord)]
pub(super) struct RegistryId {
addr: usize,
}
struct ThreadInfo {
/// Latch set once thread has started and we are entering into the
/// main loop. Used to wait for worker threads to become primed,
/// primarily of interest for benchmarking.
primed: LockLatch,
/// Latch is set once worker thread has completed. Used to wait
/// until workers have stopped; only used for tests.
stopped: LockLatch,
/// The latch used to signal that terminated has been requested.
/// This latch is *set* by the `terminate` method on the
/// `Registry`, once the registry's main "terminate" counter
/// reaches zero.
///
/// NB. We use a `CountLatch` here because it has no lifetimes and is
/// meant for async use, but the count never gets higher than one.
terminate: CountLatch,
/// the "stealer" half of the worker's deque
stealer: Stealer<JobRef>,
}
impl ThreadInfo {
fn new(stealer: Stealer<JobRef>) -> ThreadInfo {
ThreadInfo {
primed: LockLatch::new(),
stopped: LockLatch::new(),
terminate: CountLatch::new(),
stealer,
}
}
}
/// ////////////////////////////////////////////////////////////////////////
/// WorkerThread identifiers
pub(super) struct WorkerThread {
/// the "worker" half of our local deque
worker: Worker<JobRef>,
/// the "stealer" half of the worker's broadcast deque
stealer: Stealer<JobRef>,
/// local queue used for `spawn_fifo` indirection
fifo: JobFifo,
index: usize,
/// A weak random number generator.
rng: XorShift64Star,
registry: Arc<Registry>,
}
// This is a bit sketchy, but basically: the WorkerThread is
// allocated on the stack of the worker on entry and stored into this
// thread local variable. So it will remain valid at least until the
// worker is fully unwound. Using an unsafe pointer avoids the need
// for a RefCell<T> etc.
thread_local! {
static WORKER_THREAD_STATE: Cell<*const WorkerThread> = const { Cell::new(ptr::null()) };
}
impl From<ThreadBuilder> for WorkerThread {
fn from(thread: ThreadBuilder) -> Self {
Self {
worker: thread.worker,
stealer: thread.stealer,
fifo: JobFifo::new(),
index: thread.index,
rng: XorShift64Star::new(),
registry: thread.registry,
}
}
}
impl Drop for WorkerThread {
fn drop(&mut self) {
// Undo `set_current`
WORKER_THREAD_STATE.with(|t| {
assert!(t.get().eq(&(self as *const _)));
t.set(ptr::null());
});
}
}
impl WorkerThread {
/// Gets the `WorkerThread` index for the current thread; returns
/// NULL if this is not a worker thread. This pointer is valid
/// anywhere on the current thread.
#[inline]
pub(super) fn current() -> *const WorkerThread {
WORKER_THREAD_STATE.with(Cell::get)
}
/// Sets `self` as the worker thread index for the current thread.
/// This is done during worker thread startup.
unsafe fn set_current(thread: *const WorkerThread) {
WORKER_THREAD_STATE.with(|t| {
assert!(t.get().is_null());
t.set(thread);
});
}
/// Returns the registry that owns this worker thread.
#[inline]
pub(super) fn registry(&self) -> &Arc<Registry> {
&self.registry
}
#[inline]
pub(super) fn log(&self, event: impl FnOnce() -> crate::log::Event) {
self.registry.logger.log(event)
}
/// Our index amongst the worker threads (ranges from `0..self.num_threads()`).
#[inline]
pub(super) fn index(&self) -> usize {
self.index
}
#[inline]
pub(super) unsafe fn push(&self, job: JobRef) {
self.log(|| JobPushed { worker: self.index });
let queue_was_empty = self.worker.is_empty();
self.worker.push(job);
self.registry
.sleep
.new_internal_jobs(self.index, 1, queue_was_empty);
}
#[inline]
pub(super) unsafe fn push_fifo(&self, job: JobRef) {
self.push(self.fifo.push(job));
}
#[inline]
pub(super) fn local_deque_is_empty(&self) -> bool {
self.worker.is_empty()
}
/// Attempts to obtain a "local" job -- typically this means
/// popping from the top of the stack, though if we are configured
/// for breadth-first execution, it would mean dequeuing from the
/// bottom.
#[inline]
pub(super) fn take_local_job(&self) -> Option<JobRef> {
let popped_job = self.worker.pop();
if popped_job.is_some() {
self.log(|| JobPopped { worker: self.index });
return popped_job;
}
loop {
match self.stealer.steal() {
Steal::Success(job) => return Some(job),
Steal::Empty => return None,
Steal::Retry => {}
}
}
}
fn has_injected_job(&self) -> bool {
!self.stealer.is_empty() || self.registry.has_injected_job()
}
/// Wait until the latch is set. Try to keep busy by popping and
/// stealing tasks as necessary.
#[inline]
pub(super) unsafe fn wait_until<L: AsCoreLatch + ?Sized>(&self, latch: &L) {
let latch = latch.as_core_latch();
if !latch.probe() {
self.wait_until_cold(latch);
}
}
#[cold]
unsafe fn wait_until_cold(&self, latch: &CoreLatch) {
// the code below should swallow all panics and hence never
// unwind; but if something does wrong, we want to abort,
// because otherwise other code in rayon may assume that the
// latch has been signaled, and that can lead to random memory
// accesses, which would be *very bad*
let abort_guard = unwind::AbortIfPanic;
let mut idle_state = self.registry.sleep.start_looking(self.index, latch);
while !latch.probe() {
if let Some(job) = self.find_work() {
self.registry.sleep.work_found(idle_state);
self.execute(job);
idle_state = self.registry.sleep.start_looking(self.index, latch);
} else {
self.registry
.sleep
.no_work_found(&mut idle_state, latch, || self.has_injected_job())
}
}
// If we were sleepy, we are not anymore. We "found work" --
// whatever the surrounding thread was doing before it had to
// wait.
self.registry.sleep.work_found(idle_state);
self.log(|| ThreadSawLatchSet {
worker: self.index,
latch_addr: latch.addr(),
});
mem::forget(abort_guard); // successful execution, do not abort
}
fn find_work(&self) -> Option<JobRef> {
// Try to find some work to do. We give preference first
// to things in our local deque, then in other workers
// deques, and finally to injected jobs from the
// outside. The idea is to finish what we started before
// we take on something new.
self.take_local_job()
.or_else(|| self.steal())
.or_else(|| self.registry.pop_injected_job(self.index))
}
pub(super) fn yield_now(&self) -> Yield {
match self.find_work() {
Some(job) => unsafe {
self.execute(job);
Yield::Executed
},
None => Yield::Idle,
}
}
pub(super) fn yield_local(&self) -> Yield {
match self.take_local_job() {
Some(job) => unsafe {
self.execute(job);
Yield::Executed
},
None => Yield::Idle,
}
}
#[inline]
pub(super) unsafe fn execute(&self, job: JobRef) {
job.execute();
}
/// Try to steal a single job and return it.
///
/// This should only be done as a last resort, when there is no
/// local work to do.
fn steal(&self) -> Option<JobRef> {
// we only steal when we don't have any work to do locally
debug_assert!(self.local_deque_is_empty());
// otherwise, try to steal
let thread_infos = &self.registry.thread_infos.as_slice();
let num_threads = thread_infos.len();
if num_threads <= 1 {
return None;
}
loop {
let mut retry = false;
let start = self.rng.next_usize(num_threads);
let job = (start..num_threads)
.chain(0..start)
.filter(move |&i| i != self.index)
.find_map(|victim_index| {
let victim = &thread_infos[victim_index];
match victim.stealer.steal() {
Steal::Success(job) => {
self.log(|| JobStolen {
worker: self.index,
victim: victim_index,
});
Some(job)
}
Steal::Empty => None,
Steal::Retry => {
retry = true;
None
}
}
});
if job.is_some() || !retry {
return job;
}
}
}
}
/// ////////////////////////////////////////////////////////////////////////
unsafe fn main_loop(thread: ThreadBuilder) {
let worker_thread = &WorkerThread::from(thread);
WorkerThread::set_current(worker_thread);
let registry = &*worker_thread.registry;
let index = worker_thread.index;
// let registry know we are ready to do work
Latch::set(®istry.thread_infos[index].primed);
// Worker threads should not panic. If they do, just abort, as the
// internal state of the threadpool is corrupted. Note that if
// **user code** panics, we should catch that and redirect.
let abort_guard = unwind::AbortIfPanic;
// Inform a user callback that we started a thread.
if let Some(ref handler) = registry.start_handler {
registry.catch_unwind(|| handler(index));
}
let my_terminate_latch = ®istry.thread_infos[index].terminate;
worker_thread.log(|| ThreadStart {
worker: index,
terminate_addr: my_terminate_latch.as_core_latch().addr(),
});
worker_thread.wait_until(my_terminate_latch);
// Should not be any work left in our queue.
debug_assert!(worker_thread.take_local_job().is_none());
// let registry know we are done
Latch::set(®istry.thread_infos[index].stopped);
// Normal termination, do not abort.
mem::forget(abort_guard);
worker_thread.log(|| ThreadTerminate { worker: index });
// Inform a user callback that we exited a thread.
if let Some(ref handler) = registry.exit_handler {
registry.catch_unwind(|| handler(index));
// We're already exiting the thread, there's nothing else to do.
}
}
/// If already in a worker-thread, just execute `op`. Otherwise,
/// execute `op` in the default thread-pool. Either way, block until
/// `op` completes and return its return value. If `op` panics, that
/// panic will be propagated as well. The second argument indicates
/// `true` if injection was performed, `false` if executed directly.
pub(super) fn in_worker<OP, R>(op: OP) -> R
where
OP: FnOnce(&WorkerThread, bool) -> R + Send,
R: Send,
{
unsafe {
let owner_thread = WorkerThread::current();
if !owner_thread.is_null() {
// Perfectly valid to give them a `&T`: this is the
// current thread, so we know the data structure won't be
// invalidated until we return.
op(&*owner_thread, false)
} else {
global_registry().in_worker(op)
}
}
}
/// [xorshift*] is a fast pseudorandom number generator which will
/// even tolerate weak seeding, as long as it's not zero.
///
/// [xorshift*]: https://en.wikipedia.org/wiki/Xorshift#xorshift*
struct XorShift64Star {
state: Cell<u64>,
}
impl XorShift64Star {
fn new() -> Self {
// Any non-zero seed will do -- this uses the hash of a global counter.
let mut seed = 0;
while seed == 0 {
let mut hasher = DefaultHasher::new();
static COUNTER: AtomicUsize = AtomicUsize::new(0);
hasher.write_usize(COUNTER.fetch_add(1, Ordering::Relaxed));
seed = hasher.finish();
}
XorShift64Star {
state: Cell::new(seed),
}
}
fn next(&self) -> u64 {
let mut x = self.state.get();
debug_assert_ne!(x, 0);
x ^= x >> 12;
x ^= x << 25;
x ^= x >> 27;
self.state.set(x);
x.wrapping_mul(0x2545_f491_4f6c_dd1d)
}
/// Return a value from `0..n`.
fn next_usize(&self, n: usize) -> usize {
(self.next() % n as u64) as usize
}
}