Revision control
Copy as Markdown
Other Tools
//! The global data and participant for garbage collection.
//!
//! # Registration
//!
//! In order to track all participants in one place, we need some form of participant
//! registration. When a participant is created, it is registered to a global lock-free
//! singly-linked list of registries; and when a participant is leaving, it is unregistered from the
//! list.
//!
//! # Pinning
//!
//! Every participant contains an integer that tells whether the participant is pinned and if so,
//! what was the global epoch at the time it was pinned. Participants also hold a pin counter that
//! aids in periodic global epoch advancement.
//!
//! When a participant is pinned, a `Guard` is returned as a witness that the participant is pinned.
//! Guards are necessary for performing atomic operations, and for freeing/dropping locations.
//!
//! # Thread-local bag
//!
//! Objects that get unlinked from concurrent data structures must be stashed away until the global
//! epoch sufficiently advances so that they become safe for destruction. Pointers to such objects
//! are pushed into a thread-local bag, and when it becomes full, the bag is marked with the current
//! global epoch and pushed into the global queue of bags. We store objects in thread-local storages
//! for amortizing the synchronization cost of pushing the garbages to a global queue.
//!
//! # Global queue
//!
//! Whenever a bag is pushed into a queue, the objects in some bags in the queue are collected and
//! destroyed along the way. This design reduces contention on data structures. The global queue
//! cannot be explicitly accessed: the only way to interact with it is by calling functions
//! `defer()` that adds an object to the thread-local bag, or `collect()` that manually triggers
//! garbage collection.
//!
//! Ideally each instance of concurrent data structure may have its own queue that gets fully
//! destroyed as soon as the data structure gets dropped.
use crate::primitive::cell::UnsafeCell;
use crate::primitive::sync::atomic;
use core::cell::Cell;
use core::mem::{self, ManuallyDrop};
use core::num::Wrapping;
use core::sync::atomic::Ordering;
use core::{fmt, ptr};
use crossbeam_utils::CachePadded;
use memoffset::offset_of;
use crate::atomic::{Owned, Shared};
use crate::collector::{Collector, LocalHandle};
use crate::deferred::Deferred;
use crate::epoch::{AtomicEpoch, Epoch};
use crate::guard::{unprotected, Guard};
use crate::sync::list::{Entry, IsElement, IterError, List};
use crate::sync::queue::Queue;
/// Maximum number of objects a bag can contain.
#[cfg(not(any(crossbeam_sanitize, miri)))]
const MAX_OBJECTS: usize = 64;
// Makes it more likely to trigger any potential data races.
#[cfg(any(crossbeam_sanitize, miri))]
const MAX_OBJECTS: usize = 4;
/// A bag of deferred functions.
pub(crate) struct Bag {
/// Stashed objects.
deferreds: [Deferred; MAX_OBJECTS],
len: usize,
}
/// `Bag::try_push()` requires that it is safe for another thread to execute the given functions.
unsafe impl Send for Bag {}
impl Bag {
/// Returns a new, empty bag.
pub(crate) fn new() -> Self {
Self::default()
}
/// Returns `true` if the bag is empty.
pub(crate) fn is_empty(&self) -> bool {
self.len == 0
}
/// Attempts to insert a deferred function into the bag.
///
/// Returns `Ok(())` if successful, and `Err(deferred)` for the given `deferred` if the bag is
/// full.
///
/// # Safety
///
/// It should be safe for another thread to execute the given function.
pub(crate) unsafe fn try_push(&mut self, deferred: Deferred) -> Result<(), Deferred> {
if self.len < MAX_OBJECTS {
self.deferreds[self.len] = deferred;
self.len += 1;
Ok(())
} else {
Err(deferred)
}
}
/// Seals the bag with the given epoch.
fn seal(self, epoch: Epoch) -> SealedBag {
SealedBag { epoch, _bag: self }
}
}
impl Default for Bag {
fn default() -> Self {
Bag {
len: 0,
deferreds: [Deferred::NO_OP; MAX_OBJECTS],
}
}
}
impl Drop for Bag {
fn drop(&mut self) {
// Call all deferred functions.
for deferred in &mut self.deferreds[..self.len] {
let no_op = Deferred::NO_OP;
let owned_deferred = mem::replace(deferred, no_op);
owned_deferred.call();
}
}
}
// can't #[derive(Debug)] because Debug is not implemented for arrays 64 items long
impl fmt::Debug for Bag {
fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
f.debug_struct("Bag")
.field("deferreds", &&self.deferreds[..self.len])
.finish()
}
}
/// A pair of an epoch and a bag.
#[derive(Default, Debug)]
struct SealedBag {
epoch: Epoch,
_bag: Bag,
}
/// It is safe to share `SealedBag` because `is_expired` only inspects the epoch.
unsafe impl Sync for SealedBag {}
impl SealedBag {
/// Checks if it is safe to drop the bag w.r.t. the given global epoch.
fn is_expired(&self, global_epoch: Epoch) -> bool {
// A pinned participant can witness at most one epoch advancement. Therefore, any bag that
// is within one epoch of the current one cannot be destroyed yet.
global_epoch.wrapping_sub(self.epoch) >= 2
}
}
/// The global data for a garbage collector.
pub(crate) struct Global {
/// The intrusive linked list of `Local`s.
locals: List<Local>,
/// The global queue of bags of deferred functions.
queue: Queue<SealedBag>,
/// The global epoch.
pub(crate) epoch: CachePadded<AtomicEpoch>,
}
impl Global {
/// Number of bags to destroy.
const COLLECT_STEPS: usize = 8;
/// Creates a new global data for garbage collection.
#[inline]
pub(crate) fn new() -> Self {
Self {
locals: List::new(),
queue: Queue::new(),
epoch: CachePadded::new(AtomicEpoch::new(Epoch::starting())),
}
}
/// Pushes the bag into the global queue and replaces the bag with a new empty bag.
pub(crate) fn push_bag(&self, bag: &mut Bag, guard: &Guard) {
let bag = mem::replace(bag, Bag::new());
atomic::fence(Ordering::SeqCst);
let epoch = self.epoch.load(Ordering::Relaxed);
self.queue.push(bag.seal(epoch), guard);
}
/// Collects several bags from the global queue and executes deferred functions in them.
///
/// Note: This may itself produce garbage and in turn allocate new bags.
///
/// `pin()` rarely calls `collect()`, so we want the compiler to place that call on a cold
/// path. In other words, we want the compiler to optimize branching for the case when
/// `collect()` is not called.
#[cold]
pub(crate) fn collect(&self, guard: &Guard) {
let global_epoch = self.try_advance(guard);
let steps = if cfg!(crossbeam_sanitize) {
usize::max_value()
} else {
Self::COLLECT_STEPS
};
for _ in 0..steps {
match self.queue.try_pop_if(
&|sealed_bag: &SealedBag| sealed_bag.is_expired(global_epoch),
guard,
) {
None => break,
Some(sealed_bag) => drop(sealed_bag),
}
}
}
/// Attempts to advance the global epoch.
///
/// The global epoch can advance only if all currently pinned participants have been pinned in
/// the current epoch.
///
/// Returns the current global epoch.
///
/// `try_advance()` is annotated `#[cold]` because it is rarely called.
#[cold]
pub(crate) fn try_advance(&self, guard: &Guard) -> Epoch {
let global_epoch = self.epoch.load(Ordering::Relaxed);
atomic::fence(Ordering::SeqCst);
// TODO(stjepang): `Local`s are stored in a linked list because linked lists are fairly
// easy to implement in a lock-free manner. However, traversal can be slow due to cache
// misses and data dependencies. We should experiment with other data structures as well.
for local in self.locals.iter(guard) {
match local {
Err(IterError::Stalled) => {
// A concurrent thread stalled this iteration. That thread might also try to
// advance the epoch, in which case we leave the job to it. Otherwise, the
// epoch will not be advanced.
return global_epoch;
}
Ok(local) => {
let local_epoch = local.epoch.load(Ordering::Relaxed);
// If the participant was pinned in a different epoch, we cannot advance the
// global epoch just yet.
if local_epoch.is_pinned() && local_epoch.unpinned() != global_epoch {
return global_epoch;
}
}
}
}
atomic::fence(Ordering::Acquire);
// All pinned participants were pinned in the current global epoch.
// Now let's advance the global epoch...
//
// Note that if another thread already advanced it before us, this store will simply
// overwrite the global epoch with the same value. This is true because `try_advance` was
// called from a thread that was pinned in `global_epoch`, and the global epoch cannot be
// advanced two steps ahead of it.
let new_epoch = global_epoch.successor();
self.epoch.store(new_epoch, Ordering::Release);
new_epoch
}
}
/// Participant for garbage collection.
pub(crate) struct Local {
/// A node in the intrusive linked list of `Local`s.
entry: Entry,
/// The local epoch.
epoch: AtomicEpoch,
/// A reference to the global data.
///
/// When all guards and handles get dropped, this reference is destroyed.
collector: UnsafeCell<ManuallyDrop<Collector>>,
/// The local bag of deferred functions.
pub(crate) bag: UnsafeCell<Bag>,
/// The number of guards keeping this participant pinned.
guard_count: Cell<usize>,
/// The number of active handles.
handle_count: Cell<usize>,
/// Total number of pinnings performed.
///
/// This is just an auxiliary counter that sometimes kicks off collection.
pin_count: Cell<Wrapping<usize>>,
}
// Make sure `Local` is less than or equal to 2048 bytes.
#[cfg(not(any(crossbeam_sanitize, miri)))] // `crossbeam_sanitize` and `miri` reduce the size of `Local`
#[test]
fn local_size() {
// assert!(
// core::mem::size_of::<Local>() <= 2048,
// "An allocation of `Local` should be <= 2048 bytes."
// );
}
impl Local {
/// Number of pinnings after which a participant will execute some deferred functions from the
/// global queue.
const PINNINGS_BETWEEN_COLLECT: usize = 128;
/// Registers a new `Local` in the provided `Global`.
pub(crate) fn register(collector: &Collector) -> LocalHandle {
unsafe {
// Since we dereference no pointers in this block, it is safe to use `unprotected`.
let local = Owned::new(Local {
entry: Entry::default(),
epoch: AtomicEpoch::new(Epoch::starting()),
collector: UnsafeCell::new(ManuallyDrop::new(collector.clone())),
bag: UnsafeCell::new(Bag::new()),
guard_count: Cell::new(0),
handle_count: Cell::new(1),
pin_count: Cell::new(Wrapping(0)),
})
.into_shared(unprotected());
collector.global.locals.insert(local, unprotected());
LocalHandle {
local: local.as_raw(),
}
}
}
/// Returns a reference to the `Global` in which this `Local` resides.
#[inline]
pub(crate) fn global(&self) -> &Global {
&self.collector().global
}
/// Returns a reference to the `Collector` in which this `Local` resides.
#[inline]
pub(crate) fn collector(&self) -> &Collector {
self.collector.with(|c| unsafe { &**c })
}
/// Returns `true` if the current participant is pinned.
#[inline]
pub(crate) fn is_pinned(&self) -> bool {
self.guard_count.get() > 0
}
/// Adds `deferred` to the thread-local bag.
///
/// # Safety
///
/// It should be safe for another thread to execute the given function.
pub(crate) unsafe fn defer(&self, mut deferred: Deferred, guard: &Guard) {
let bag = self.bag.with_mut(|b| &mut *b);
while let Err(d) = bag.try_push(deferred) {
self.global().push_bag(bag, guard);
deferred = d;
}
}
pub(crate) fn flush(&self, guard: &Guard) {
let bag = self.bag.with_mut(|b| unsafe { &mut *b });
if !bag.is_empty() {
self.global().push_bag(bag, guard);
}
self.global().collect(guard);
}
/// Pins the `Local`.
#[inline]
pub(crate) fn pin(&self) -> Guard {
let guard = Guard { local: self };
let guard_count = self.guard_count.get();
self.guard_count.set(guard_count.checked_add(1).unwrap());
if guard_count == 0 {
let global_epoch = self.global().epoch.load(Ordering::Relaxed);
let new_epoch = global_epoch.pinned();
// Now we must store `new_epoch` into `self.epoch` and execute a `SeqCst` fence.
// The fence makes sure that any future loads from `Atomic`s will not happen before
// this store.
if cfg!(all(
any(target_arch = "x86", target_arch = "x86_64"),
not(miri)
)) {
// HACK(stjepang): On x86 architectures there are two different ways of executing
// a `SeqCst` fence.
//
// 1. `atomic::fence(SeqCst)`, which compiles into a `mfence` instruction.
// 2. `_.compare_exchange(_, _, SeqCst, SeqCst)`, which compiles into a `lock cmpxchg`
// instruction.
//
// Both instructions have the effect of a full barrier, but benchmarks have shown
// that the second one makes pinning faster in this particular case. It is not
// clear that this is permitted by the C++ memory model (SC fences work very
// differently from SC accesses), but experimental evidence suggests that this
// works fine. Using inline assembly would be a viable (and correct) alternative,
// but alas, that is not possible on stable Rust.
let current = Epoch::starting();
let res = self.epoch.compare_exchange(
current,
new_epoch,
Ordering::SeqCst,
Ordering::SeqCst,
);
debug_assert!(res.is_ok(), "participant was expected to be unpinned");
// We add a compiler fence to make it less likely for LLVM to do something wrong
// here. Formally, this is not enough to get rid of data races; practically,
// it should go a long way.
atomic::compiler_fence(Ordering::SeqCst);
} else {
self.epoch.store(new_epoch, Ordering::Relaxed);
atomic::fence(Ordering::SeqCst);
}
// Increment the pin counter.
let count = self.pin_count.get();
self.pin_count.set(count + Wrapping(1));
// After every `PINNINGS_BETWEEN_COLLECT` try advancing the epoch and collecting
// some garbage.
if count.0 % Self::PINNINGS_BETWEEN_COLLECT == 0 {
self.global().collect(&guard);
}
}
guard
}
/// Unpins the `Local`.
#[inline]
pub(crate) fn unpin(&self) {
let guard_count = self.guard_count.get();
self.guard_count.set(guard_count - 1);
if guard_count == 1 {
self.epoch.store(Epoch::starting(), Ordering::Release);
if self.handle_count.get() == 0 {
self.finalize();
}
}
}
/// Unpins and then pins the `Local`.
#[inline]
pub(crate) fn repin(&self) {
let guard_count = self.guard_count.get();
// Update the local epoch only if there's only one guard.
if guard_count == 1 {
let epoch = self.epoch.load(Ordering::Relaxed);
let global_epoch = self.global().epoch.load(Ordering::Relaxed).pinned();
// Update the local epoch only if the global epoch is greater than the local epoch.
if epoch != global_epoch {
// We store the new epoch with `Release` because we need to ensure any memory
// accesses from the previous epoch do not leak into the new one.
self.epoch.store(global_epoch, Ordering::Release);
// However, we don't need a following `SeqCst` fence, because it is safe for memory
// accesses from the new epoch to be executed before updating the local epoch. At
// worse, other threads will see the new epoch late and delay GC slightly.
}
}
}
/// Increments the handle count.
#[inline]
pub(crate) fn acquire_handle(&self) {
let handle_count = self.handle_count.get();
debug_assert!(handle_count >= 1);
self.handle_count.set(handle_count + 1);
}
/// Decrements the handle count.
#[inline]
pub(crate) fn release_handle(&self) {
let guard_count = self.guard_count.get();
let handle_count = self.handle_count.get();
debug_assert!(handle_count >= 1);
self.handle_count.set(handle_count - 1);
if guard_count == 0 && handle_count == 1 {
self.finalize();
}
}
/// Removes the `Local` from the global linked list.
#[cold]
fn finalize(&self) {
debug_assert_eq!(self.guard_count.get(), 0);
debug_assert_eq!(self.handle_count.get(), 0);
// Temporarily increment handle count. This is required so that the following call to `pin`
// doesn't call `finalize` again.
self.handle_count.set(1);
unsafe {
// Pin and move the local bag into the global queue. It's important that `push_bag`
// doesn't defer destruction on any new garbage.
let guard = &self.pin();
self.global()
.push_bag(self.bag.with_mut(|b| &mut *b), guard);
}
// Revert the handle count back to zero.
self.handle_count.set(0);
unsafe {
// Take the reference to the `Global` out of this `Local`. Since we're not protected
// by a guard at this time, it's crucial that the reference is read before marking the
// `Local` as deleted.
let collector: Collector = ptr::read(self.collector.with(|c| &*(*c)));
// Mark this node in the linked list as deleted.
self.entry.delete(unprotected());
// Finally, drop the reference to the global. Note that this might be the last reference
// to the `Global`. If so, the global data will be destroyed and all deferred functions
// in its queue will be executed.
drop(collector);
}
}
}
impl IsElement<Local> for Local {
fn entry_of(local: &Local) -> &Entry {
let entry_ptr = (local as *const Local as usize + offset_of!(Local, entry)) as *const Entry;
unsafe { &*entry_ptr }
}
unsafe fn element_of(entry: &Entry) -> &Local {
// offset_of! macro uses unsafe, but it's unnecessary in this context.
#[allow(unused_unsafe)]
let local_ptr = (entry as *const Entry as usize - offset_of!(Local, entry)) as *const Local;
&*local_ptr
}
unsafe fn finalize(entry: &Entry, guard: &Guard) {
guard.defer_destroy(Shared::from(Self::element_of(entry) as *const _));
}
}
#[cfg(all(test, not(crossbeam_loom)))]
mod tests {
use std::sync::atomic::{AtomicUsize, Ordering};
use super::*;
#[test]
fn check_defer() {
static FLAG: AtomicUsize = AtomicUsize::new(0);
fn set() {
FLAG.store(42, Ordering::Relaxed);
}
let d = Deferred::new(set);
assert_eq!(FLAG.load(Ordering::Relaxed), 0);
d.call();
assert_eq!(FLAG.load(Ordering::Relaxed), 42);
}
#[test]
fn check_bag() {
static FLAG: AtomicUsize = AtomicUsize::new(0);
fn incr() {
FLAG.fetch_add(1, Ordering::Relaxed);
}
let mut bag = Bag::new();
assert!(bag.is_empty());
for _ in 0..MAX_OBJECTS {
assert!(unsafe { bag.try_push(Deferred::new(incr)).is_ok() });
assert!(!bag.is_empty());
assert_eq!(FLAG.load(Ordering::Relaxed), 0);
}
let result = unsafe { bag.try_push(Deferred::new(incr)) };
assert!(result.is_err());
assert!(!bag.is_empty());
assert_eq!(FLAG.load(Ordering::Relaxed), 0);
drop(bag);
assert_eq!(FLAG.load(Ordering::Relaxed), MAX_OBJECTS);
}
}