Add directly to global pool and reduce global lock acquiring

This commit is contained in:
Zhang Junyang 2025-03-20 17:24:08 +08:00 committed by Tate, Hongliang Tian
parent d8e076f58a
commit abc74151fb
2 changed files with 65 additions and 47 deletions

View File

@ -2,13 +2,9 @@
//! Controlling the balancing between CPU-local free pools and the global free pool.
use core::sync::atomic::Ordering;
use ostd::cpu::num_cpus;
use super::{
lesser_order_of, BuddyOrder, BuddySet, GLOBAL_POOL, GLOBAL_POOL_SIZE, MAX_LOCAL_BUDDY_ORDER,
};
use super::{lesser_order_of, BuddyOrder, BuddySet, OnDemandGlobalLock, MAX_LOCAL_BUDDY_ORDER};
use crate::chunk::split_to_order;
@ -50,8 +46,8 @@ fn cache_maximal_size(global_size: usize) -> usize {
}
/// Balances a local cache and the global free pool.
pub fn balance(local: &mut BuddySet<MAX_LOCAL_BUDDY_ORDER>) {
let global_size = GLOBAL_POOL_SIZE.load(Ordering::Relaxed);
pub fn balance(local: &mut BuddySet<MAX_LOCAL_BUDDY_ORDER>, global: &mut OnDemandGlobalLock) {
let global_size = global.get_global_size();
let minimal_local_size = cache_minimal_size(global_size);
let expected_local_size = cache_expected_size(global_size);
@ -67,11 +63,8 @@ pub fn balance(local: &mut BuddySet<MAX_LOCAL_BUDDY_ORDER>) {
let expected_removal = local_size - expected_local_size;
let lesser_order = lesser_order_of(expected_removal);
let mut global_pool_lock = GLOBAL_POOL.lock();
balance_to(local, &mut *global_pool_lock, lesser_order);
GLOBAL_POOL_SIZE.store(global_pool_lock.total_size(), Ordering::Relaxed);
balance_to(local, &mut *global.get(), lesser_order);
} else if local_size < minimal_local_size {
// Move global frames to the local pool.
if global_size == 0 {
@ -80,11 +73,8 @@ pub fn balance(local: &mut BuddySet<MAX_LOCAL_BUDDY_ORDER>) {
let expected_allocation = expected_local_size - local_size;
let lesser_order = lesser_order_of(expected_allocation);
let mut global_pool_lock = GLOBAL_POOL.lock();
balance_to(&mut *global_pool_lock, local, lesser_order);
GLOBAL_POOL_SIZE.store(global_pool_lock.total_size(), Ordering::Relaxed);
balance_to(&mut *global.get(), local, lesser_order);
}
}

View File

@ -10,10 +10,9 @@ use core::{
};
use ostd::{
cpu::PinCurrentCpu,
cpu_local,
mm::Paddr,
sync::{LocalIrqDisabled, SpinLock},
sync::{LocalIrqDisabled, SpinLock, SpinLockGuard},
trap::DisabledLocalIrqGuard,
};
@ -24,12 +23,12 @@ use super::set::BuddySet;
/// The global free buddies.
static GLOBAL_POOL: SpinLock<BuddySet<MAX_BUDDY_ORDER>, LocalIrqDisabled> =
SpinLock::new(BuddySet::new_empty());
/// A snapshot of the total size of the global free buddies, not precise.
static GLOBAL_POOL_SIZE: AtomicUsize = AtomicUsize::new(0);
// CPU-local free buddies.
cpu_local! {
static LOCAL_POOL: RefCell<BuddySet<MAX_LOCAL_BUDDY_ORDER>> = RefCell::new(BuddySet::new_empty());
static LOCAL_POOL_SIZE: AtomicUsize = AtomicUsize::new(0);
}
/// Maximum supported order of the buddy system.
@ -55,6 +54,7 @@ const MAX_LOCAL_BUDDY_ORDER: BuddyOrder = 18;
pub(super) fn alloc(guard: &DisabledLocalIrqGuard, layout: Layout) -> Option<Paddr> {
let local_pool_cell = LOCAL_POOL.get_with(guard);
let mut local_pool = local_pool_cell.borrow_mut();
let mut global_pool = OnDemandGlobalLock::new();
let size_order = greater_order_of(layout.size());
let align_order = greater_order_of(layout.align());
@ -68,7 +68,7 @@ pub(super) fn alloc(guard: &DisabledLocalIrqGuard, layout: Layout) -> Option<Pad
// Fall back to the global free lists if the local free lists are empty.
if chunk_addr.is_none() {
chunk_addr = alloc_from_global_pool(order);
chunk_addr = global_pool.get().alloc_chunk(order);
}
// TODO: On memory pressure the global pool may be not enough. We may need
// to merge all buddy chunks from the local pools to the global pool and
@ -79,20 +79,18 @@ pub(super) fn alloc(guard: &DisabledLocalIrqGuard, layout: Layout) -> Option<Pad
let allocated_size = size_of_order(order);
if allocated_size > layout.size() {
if let Some(chunk_addr) = chunk_addr {
add_free_memory_to(
do_dealloc(
&mut local_pool,
guard,
&mut global_pool,
chunk_addr + layout.size(),
allocated_size - layout.size(),
);
}
}
balancing::balance(local_pool.deref_mut());
balancing::balance(local_pool.deref_mut(), &mut global_pool);
LOCAL_POOL_SIZE
.get_on_cpu(guard.current_cpu())
.store(local_pool.total_size(), Ordering::Relaxed);
global_pool.update_global_size_if_locked();
chunk_addr
}
@ -100,46 +98,76 @@ pub(super) fn alloc(guard: &DisabledLocalIrqGuard, layout: Layout) -> Option<Pad
pub(super) fn dealloc(guard: &DisabledLocalIrqGuard, addr: Paddr, size: usize) {
let local_pool_cell = LOCAL_POOL.get_with(guard);
let mut local_pool = local_pool_cell.borrow_mut();
let mut global_pool = OnDemandGlobalLock::new();
add_free_memory_to(&mut local_pool, guard, addr, size);
do_dealloc(&mut local_pool, &mut global_pool, addr, size);
balancing::balance(local_pool.deref_mut(), &mut global_pool);
global_pool.update_global_size_if_locked();
}
pub(super) fn add_free_memory(guard: &DisabledLocalIrqGuard, addr: Paddr, size: usize) {
let local_pool_cell = LOCAL_POOL.get_with(guard);
let mut local_pool = local_pool_cell.borrow_mut();
pub(super) fn add_free_memory(_guard: &DisabledLocalIrqGuard, addr: Paddr, size: usize) {
let mut global_pool = OnDemandGlobalLock::new();
add_free_memory_to(&mut local_pool, guard, addr, size);
split_to_chunks(addr, size).for_each(|(addr, order)| {
global_pool.get().insert_chunk(addr, order);
});
global_pool.update_global_size_if_locked();
}
fn add_free_memory_to(
fn do_dealloc(
local_pool: &mut BuddySet<MAX_LOCAL_BUDDY_ORDER>,
guard: &DisabledLocalIrqGuard,
global_pool: &mut OnDemandGlobalLock,
addr: Paddr,
size: usize,
) {
split_to_chunks(addr, size).for_each(|(addr, order)| {
if order >= MAX_LOCAL_BUDDY_ORDER {
dealloc_to_global_pool(addr, order);
global_pool.get().insert_chunk(addr, order);
} else {
local_pool.insert_chunk(addr, order);
}
});
balancing::balance(local_pool);
LOCAL_POOL_SIZE
.get_on_cpu(guard.current_cpu())
.store(local_pool.total_size(), Ordering::Relaxed);
}
fn alloc_from_global_pool(order: BuddyOrder) -> Option<Paddr> {
let mut lock_guard = GLOBAL_POOL.lock();
let res = lock_guard.alloc_chunk(order);
GLOBAL_POOL_SIZE.store(lock_guard.total_size(), Ordering::Relaxed);
res
type GlobalLockGuard = SpinLockGuard<'static, BuddySet<MAX_BUDDY_ORDER>, LocalIrqDisabled>;
/// An on-demand guard that locks the global pool when needed.
///
/// It helps to avoid unnecessarily locking the global pool, and also avoids
/// repeatedly locking the global pool when it is needed multiple times.
struct OnDemandGlobalLock {
guard: Option<GlobalLockGuard>,
}
fn dealloc_to_global_pool(addr: Paddr, order: BuddyOrder) {
let mut lock_guard = GLOBAL_POOL.lock();
lock_guard.insert_chunk(addr, order);
GLOBAL_POOL_SIZE.store(lock_guard.total_size(), Ordering::Relaxed);
impl OnDemandGlobalLock {
fn new() -> Self {
Self { guard: None }
}
fn get(&mut self) -> &mut GlobalLockGuard {
self.guard.get_or_insert_with(|| GLOBAL_POOL.lock())
}
/// Updates [`GLOBAL_POOL_SIZE`] if the global pool is locked.
fn update_global_size_if_locked(&self) {
if let Some(guard) = self.guard.as_ref() {
GLOBAL_POOL_SIZE.store(guard.total_size(), Ordering::Relaxed);
}
}
/// Returns the size of the global pool.
///
/// If the global pool is locked, returns the actual size of the global pool.
/// Otherwise, returns the last snapshot of the global pool size by loading
/// [`GLOBAL_POOL_SIZE`].
fn get_global_size(&self) -> usize {
if let Some(guard) = self.guard.as_ref() {
guard.total_size()
} else {
GLOBAL_POOL_SIZE.load(Ordering::Relaxed)
}
}
}