[task] support percpu run-queue and cpu affinity for axtask (arceos-o…

…rg#176)

* [task] support percpu run-queue and cpu affinity for axtask

* refactor: add preempt guard for AxRunQueue through AxRunQueueRef

* refactor: use kernel_guard hold by RQ in wait_queue.rs

* feat: introduce cpumask

* [feat] add on_cpu and prev_task_on_cpu_ptr field in AxTaskInner

* fix: delete redundant kernel guard in WaitQueue wait()

* Add percpu scheduler doc

* [fix] unit test error related to doc comment

* [feat] delete cancel_alarm from timers

* Update percpu_rq

* [fix] use timer ticket id to prevent potential bug introduced by timer event confusion

* [feat] use type CpuSet to wrap cpumask::CpuMask

* [fix] use allow(clippy::modulo_one) to bypass modulo 1 error in select_run_queue_index

* [fix] miss doc for type alias CpuSet

* [fix] little modification in axtask api.rs

* [fix] some review opinions

* [fix] delete if self.on_cpu() block in unblock_locked

* [refactor] remove in_timer_list flag in axtask

* [fix] doc error in unblock_task

* [refactor] timer ticket id

* [fix] use put_prev_task in unblock_task

* [fix] pontential bug in unblock_locked, set on_cpu as true for init tasks

* [feat] enrich code comments and scheduling-related docs

* [refactor] move percpu related docs to discussions

* [fix] first round fix of review suggestions

* [refactor] add block_current in run_queue.rs

* [refactor] pass SpinNoIrqGuard to blocked_resched

* [feat] add WaitQueueGuard

* [refactor] merge unblock_task, add TaskUnblockGuard

* [fix] second round fix of review suggestions

* [fix] third round fix of review suggestions

* [refactor] use weak reference for prev task to manipulate on_cpu flag

* [feat] enable on_cpu flag only with smp feature enabled

* [fix] enable on_cpu flag only with smp feature enabled

* [fix] change the pos of NoPreemptIrqSave guard hold by rq in wait_queue.rs

* [refactor] delete unblock_locked, use atomic compare_exchange in unblock_task

* [fix] bug in wait_timeout, modify assert in blocked_resched and switch_to

* [fix] modify assert about irqs_enabled in switch_to

* [refactor] keep the guard in the loop in wait_timeout_until

* [fix] some compile warnings

* Update the commit hash for arceos-apps

* Update the commit hash for arceos-apps again

* Update the commit hash for arceos-apps again and again

* [fix] bug in wait_timeout and wait_timeout_until

* [CI] update dependencies for qemu build

* [fix] fourth round fix of review suggestions

* [fix] problems related to WAIT_FOR_EXIT in gc_entry

* [refactor] use current_ref_mut_raw to get TIMER_LIST in check_events()

* [refactor] improve notify_all in wait_queue.rs

* [refactor] delete redundant assertion in block_resched()

* [fix] some compile warnings

.github/workflows/actions/setup-musl/action.yml

@@ -11,7 +11,7 @@ runs:
   steps:
   - name: Cache musl
     id: cache-musl
-    uses: actions/cache/restore@v3
+    uses: actions/cache/restore@v4
     with:
       path: ${{ inputs.arch }}-linux-musl-cross
       key: ${{ inputs.arch }}-linux-musl-cross
@@ -22,7 +22,7 @@ runs:
       MUSL_PATH=${{ inputs.arch }}-linux-musl-cross
       wget https://musl.cc/${MUSL_PATH}.tgz
       tar -xf ${MUSL_PATH}.tgz
-  - uses: actions/cache/save@v3
+  - uses: actions/cache/save@v4
     if: steps.cache-musl.outputs.cache-hit != 'true'
     with:
       path: ${{ inputs.arch }}-linux-musl-cross

.github/workflows/actions/setup-qemu/action.yml

@@ -11,7 +11,7 @@ runs:
   steps:
   - name: Cache QEMU
     id: cache-qemu
-    uses: actions/cache/restore@v3
+    uses: actions/cache/restore@v4
     with:
       path: qemu_build
       key: qemu-${{ inputs.qemu-version }}-slirp-1
@@ -22,13 +22,13 @@ runs:
       PREFIX: ${{ github.workspace }}/qemu_build
     shell: bash
     run: |
-      sudo apt-get update && sudo apt-get install -y ninja-build libslirp-dev
+      sudo apt-get update && sudo apt-get install -y ninja-build libslirp-dev libglib2.0-dev
       wget https://download.qemu.org/$QEMU_PATH.tar.xz && tar -xJf $QEMU_PATH.tar.xz
       cd $QEMU_PATH \
         && ./configure --prefix=$PREFIX --target-list=x86_64-softmmu,riscv64-softmmu,aarch64-softmmu --enable-slirp \
         && make -j > /dev/null 2>&1 \
         && make install
-  - uses: actions/cache/save@v3
+  - uses: actions/cache/save@v4
     if: steps.cache-qemu.outputs.cache-hit != 'true'
     with:
       path: qemu_build

.github/workflows/test.yml

@@ -5,7 +5,7 @@ on: [push, pull_request]
 env:
   qemu-version: 8.2.0
   rust-toolchain: nightly-2024-05-02
-  arceos-apps: '68054e8'
+  arceos-apps: '57d495a'
 
 jobs:
   unit-test:

README.md

@@ -19,7 +19,7 @@ ArceOS was inspired a lot by [Unikraft](https://github.com/unikraft/unikraft).
 * [x] VirtIO net/blk/gpu drivers
 * [x] TCP/UDP net stack using [smoltcp](https://github.com/smoltcp-rs/smoltcp)
 * [x] Synchronization/Mutex
-* [x] SMP scheduling with single run queue
+* [x] SMP scheduling with [per-cpu run queue](https://github.com/arceos-org/arceos/discussions/181)
 * [x] File system
 * [ ] Compatible with Linux apps
 * [ ] Interrupt driven device I/O

api/axfeat/Cargo.toml

@@ -13,7 +13,7 @@ documentation = "https://arceos-org.github.io/arceos/axfeat/index.html"
 default = []
 
 # Multicore
-smp = ["axhal/smp", "axruntime/smp", "kspin/smp"]
+smp = ["axhal/smp", "axruntime/smp", "axtask?/smp", "kspin/smp"]
 
 # Floating point/SIMD
 fp_simd = ["axhal/fp_simd"]

doc/README.md

@@ -38,3 +38,10 @@ See [arceos-apps](https://github.com/arceos-org/arceos-apps) for example applica
 Documentation of ArceOS [modules](../modules), [api](../api), and [ulib](../ulib) are generated by [`rustdoc`](https://doc.rust-lang.org/rustdoc/what-is-rustdoc.html) and hosted on GitHub Pages:
 
 * https://arceos-org.github.io/arceos/
+
+## Discussions
+
+* [Rust std support](https://github.com/arceos-org/arceos/discussions/92)
+* [ArceOS for ARM64](https://github.com/arceos-org/arceos/discussions/101)
+* [ArceOS for RISCV Hardware](https://github.com/arceos-org/arceos/discussions/120)
+* [Per-CPU scheduling](https://github.com/arceos-org/arceos/discussions/181)

modules/axhal/src/arch/x86_64/mod.rs

@@ -19,13 +19,21 @@ pub use x86_64::structures::tss::TaskStateSegment;
 /// Allows the current CPU to respond to interrupts.
 #[inline]
 pub fn enable_irqs() {
+    #[cfg(not(target_os = "none"))]
+    {
+        warn!("enable_irqs: not implemented");
+    }
     #[cfg(target_os = "none")]
     interrupts::enable()
 }
 
 /// Makes the current CPU to ignore interrupts.
 #[inline]
 pub fn disable_irqs() {
+    #[cfg(not(target_os = "none"))]
+    {
+        warn!("disable_irqs: not implemented");
+    }
     #[cfg(target_os = "none")]
     interrupts::disable()
 }

modules/axruntime/Cargo.toml

@@ -12,7 +12,7 @@ documentation = "https://arceos-org.github.io/arceos/axruntime/index.html"
 [features]
 default = []
 
-smp = ["axhal/smp"]
+smp = ["axhal/smp", "axtask?/smp"]
 irq = ["axhal/irq", "axtask?/irq", "percpu", "kernel_guard"]
 tls = ["axhal/tls", "axtask?/tls"]
 alloc = ["axalloc"]

modules/axtask/Cargo.toml

@@ -13,12 +13,21 @@ documentation = "https://arceos-org.github.io/arceos/axtask/index.html"
 default = []
 
 multitask = [
-    "dep:axconfig", "dep:percpu", "dep:kspin", "dep:lazyinit", "dep:memory_addr",
-    "dep:scheduler", "dep:timer_list", "kernel_guard", "dep:crate_interface",
+    "dep:axconfig",
+    "dep:percpu",
+    "dep:kspin",
+    "dep:lazyinit",
+    "dep:memory_addr",
+    "dep:scheduler",
+    "dep:timer_list",
+    "kernel_guard",
+    "dep:crate_interface",
+    "dep:cpumask",
 ]
 irq = []
 tls = ["axhal/tls"]
 preempt = ["irq", "percpu?/preempt", "kernel_guard/preempt"]
+smp = ["kspin/smp"]
 
 sched_fifo = ["multitask"]
 sched_rr = ["multitask", "preempt"]
@@ -39,6 +48,7 @@ timer_list = { version = "0.1", optional = true }
 kernel_guard = { version = "0.1", optional = true }
 crate_interface = { version = "0.1", optional = true }
 scheduler = { git = "https://github.com/arceos-org/scheduler.git", tag = "v0.1.0", optional = true }
+cpumask = { git = "https://github.com/arceos-org/cpumask.git", optional = true }
 
 [dev-dependencies]
 rand = "0.8"

modules/axtask/src/api.rs

@@ -2,7 +2,9 @@
 
 use alloc::{string::String, sync::Arc};
 
-pub(crate) use crate::run_queue::{AxRunQueue, RUN_QUEUE};
+use kernel_guard::NoPreemptIrqSave;
+
+pub(crate) use crate::run_queue::{current_run_queue, select_run_queue};
 
 #[doc(cfg(feature = "multitask"))]
 pub use crate::task::{CurrentTask, TaskId, TaskInner};
@@ -14,6 +16,9 @@ pub use crate::wait_queue::WaitQueue;
 /// The reference type of a task.
 pub type AxTaskRef = Arc<AxTask>;
 
+/// The wrapper type for cpumask::CpuMask with SMP configuration.
+pub type CpuMask = cpumask::CpuMask<{ axconfig::SMP }>;
+
 cfg_if::cfg_if! {
     if #[cfg(feature = "sched_rr")] {
         const MAX_TIME_SLICE: usize = 5;
@@ -77,6 +82,8 @@ pub fn init_scheduler() {
 /// Initializes the task scheduler for secondary CPUs.
 pub fn init_scheduler_secondary() {
     crate::run_queue::init_secondary();
+    #[cfg(feature = "irq")]
+    crate::timers::init();
 }
 
 /// Handles periodic timer ticks for the task manager.
@@ -85,14 +92,17 @@ pub fn init_scheduler_secondary() {
 #[cfg(feature = "irq")]
 #[doc(cfg(feature = "irq"))]
 pub fn on_timer_tick() {
+    use kernel_guard::NoOp;
     crate::timers::check_events();
-    RUN_QUEUE.lock().scheduler_timer_tick();
+    // Since irq and preemption are both disabled here,
+    // we can get current run queue with the default `kernel_guard::NoOp`.
+    current_run_queue::<NoOp>().scheduler_timer_tick();
 }
 
 /// Adds the given task to the run queue, returns the task reference.
 pub fn spawn_task(task: TaskInner) -> AxTaskRef {
     let task_ref = task.into_arc();
-    RUN_QUEUE.lock().add_task(task_ref.clone());
+    select_run_queue::<NoPreemptIrqSave>(&task_ref).add_task(task_ref.clone());
     task_ref
 }
 
@@ -129,13 +139,13 @@ where
 ///
 /// [CFS]: https://en.wikipedia.org/wiki/Completely_Fair_Scheduler
 pub fn set_priority(prio: isize) -> bool {
-    RUN_QUEUE.lock().set_current_priority(prio)
+    current_run_queue::<NoPreemptIrqSave>().set_current_priority(prio)
 }
 
 /// Current task gives up the CPU time voluntarily, and switches to another
 /// ready task.
 pub fn yield_now() {
-    RUN_QUEUE.lock().yield_current();
+    current_run_queue::<NoPreemptIrqSave>().yield_current()
 }
 
 /// Current task is going to sleep for the given duration.
@@ -150,14 +160,14 @@ pub fn sleep(dur: core::time::Duration) {
 /// If the feature `irq` is not enabled, it uses busy-wait instead.
 pub fn sleep_until(deadline: axhal::time::TimeValue) {
     #[cfg(feature = "irq")]
-    RUN_QUEUE.lock().sleep_until(deadline);
+    current_run_queue::<NoPreemptIrqSave>().sleep_until(deadline);
     #[cfg(not(feature = "irq"))]
     axhal::time::busy_wait_until(deadline);
 }
 
 /// Exits the current task.
 pub fn exit(exit_code: i32) -> ! {
-    RUN_QUEUE.lock().exit_current(exit_code)
+    current_run_queue::<NoPreemptIrqSave>().exit_current(exit_code)
 }
 
 /// The idle task routine.

modules/axtask/src/lib.rs

@@ -42,6 +42,7 @@ cfg_if::cfg_if! {
         extern crate log;
         extern crate alloc;
 
+        #[macro_use]
         mod run_queue;
         mod task;
         mod task_ext;