Feat: implement smp initially

.cargo/config.toml

@@ -1,5 +1,5 @@
-[source.crates-io]
-replace-with = "vendored-sources"
+# [source.crates-io]
+# replace-with = "vendored-sources"
 
-[source.vendored-sources]
-directory = "vendor"
+# [source.vendored-sources]
+# directory = "vendor"

AGENTS.md

@@ -31,13 +31,13 @@
 ### RISC-V 64
 
 ```bash
-timeout 360 qemu-system-riscv64 -machine virt -kernel kernel-rv -m 1G -nographic -smp 1 -bios default -drive file=sdcard-rv.img,if=none,format=raw,id=x0 -device virtio-blk-device,drive=x0,bus=virtio-mmio-bus.0 -no-reboot -device virtio-net-device,netdev=net -netdev user,id=net -rtc base=utc -drive file=disk.img,if=none,format=raw,id=x1 -device virtio-blk-device,drive=x1,bus=virtio-mmio-bus.1
+timeout 360 qemu-system-riscv64 -machine virt -kernel kernel-rv -m 8G -nographic -smp 8 -bios default -drive file=sdcard-rv.img,if=none,format=raw,id=x0 -device virtio-blk-device,drive=x0,bus=virtio-mmio-bus.0 -no-reboot -device virtio-net-device,netdev=net -netdev user,id=net -rtc base=utc -drive file=disk.img,if=none,format=raw,id=x1 -device virtio-blk-device,drive=x1,bus=virtio-mmio-bus.1
 ```
 
 ### LoongArch64
 
 ```bash
-timeout 360 qemu-system-loongarch64 -machine virt -kernel kernel-la -m 1G -nographic -smp 1 -drive file=sdcard-la.img,if=none,format=raw,id=x0 -device virtio-blk-pci,drive=x0 -no-reboot -device virtio-net-pci,netdev=net0 -netdev user,id=net0 -rtc base=utc -drive file=disk-la.img,if=none,format=raw,id=x1 -device virtio-blk-pci,drive=x1
+timeout 360 qemu-system-loongarch64 -machine virt -kernel kernel-la -m 8G -nographic -smp 8 -drive file=sdcard-la.img,if=none,format=raw,id=x0 -device virtio-blk-pci,drive=x0 -no-reboot -device virtio-net-pci,netdev=net0 -netdev user,id=net0 -rtc base=utc -drive file=disk-la.img,if=none,format=raw,id=x1 -device virtio-blk-pci,drive=x1
 ```
 
 ## 任务执行约定
@@ -46,4 +46,4 @@ timeout 360 qemu-system-loongarch64 -machine virt -kernel kernel-la -m 1G -nogra
 - 保持改动最小且可审查，不要顺手做无关重构。
 - 如果遇到权限问题，必须立即停止任务，并提示用户先修复权限后再继续。
 - 一旦出现权限不足、只读文件系统、无法写入产物目录等情况，不要尝试绕过限制或改用破坏性手段。
-- 制订计划时必须先考虑制订功能的测试
+- 制订计划时必须先考虑制订功能的测试

Cargo.toml

@@ -17,7 +17,7 @@ pulse_syscalls = { path = "pulse_syscalls" }
 
 axfeat = { path = "arceos/api/axfeat", features = [
     "alloc-tlsf",
-    "page-alloc-4g",
+    "page-alloc-64g",
     "multitask",
     "sched-rr",
     "fs",
@@ -109,6 +109,7 @@ repository.workspace = true
 [features]
 default = []
 qemu = []
+smp = ["axfeat/smp", "axfeat/ipi"]
 testcode = []
 
 [dependencies]

Makefile

@@ -6,7 +6,7 @@ export AX_LIB := axfeat
 export APP_FEATURES := qemu
 export BLK := y
 
-export MEM := 1G
+export MEM := 8G
 export ARCH ?= riscv64
 export LOG ?= info
 

arceos/api/axfeat/Cargo.toml

@@ -20,7 +20,7 @@ fp-simd = ["axhal/fp-simd"]
 
 # Interrupts
 irq = ["axhal/irq", "axruntime/irq", "axtask?/irq"]
-ipi = ["irq", "dep:axipi", "axhal/ipi", "axruntime/ipi"]
+ipi = ["irq", "dep:axipi", "axhal/ipi", "axruntime/ipi", "axtask?/ipi"]
 
 # Custom or default platforms
 myplat = ["axhal/myplat"]

arceos/modules/axfs/src/fs/ext4/fs.rs

@@ -27,6 +27,7 @@ pub struct Ext4Filesystem {
     deletion_generation: AtomicU64,
     deletion_worker_running: AtomicBool,
     deletion_processing: async_lock::Mutex<()>,
+    pub(super) directory_mutation: async_lock::Mutex<()>,
 }
 
 impl Ext4Filesystem {
@@ -82,6 +83,7 @@ impl Ext4Filesystem {
             deletion_generation: AtomicU64::new(0),
             deletion_worker_running: AtomicBool::new(false),
             deletion_processing: async_lock::Mutex::new(()),
+            directory_mutation: async_lock::Mutex::new(()),
         });
         let root_dir = DirEntry::new_dir(
             |this| DirNode::new(Inode::new(fs.clone(), ROOT_INODE, Some(this))),

arceos/modules/axfs/src/fs/ext4/inode.rs

@@ -541,6 +541,7 @@ impl DirNodeOps for Inode {
     ) -> VfsResult<DirEntry> {
         let fs = &self.fs.inner;
         self.validate_inode_num(&fs, self.ino)?;
+        let _directory_guard = self.fs.directory_mutation.lock().await;
 
         let exists = if let Some(snapshot) = self.dir_cache.get() {
             self.cached_entry(&snapshot, name).is_some()
@@ -608,6 +609,7 @@ impl DirNodeOps for Inode {
     async fn link(&self, name: &str, node: &DirEntry) -> VfsResult<DirEntry> {
         let fs = &self.fs.inner;
         self.validate_inode_num(&fs, self.ino)?;
+        let _directory_guard = self.fs.directory_mutation.lock().await;
         let dir_idx = core::num::NonZeroU32::new(self.ino).ok_or(VfsError::InvalidData)?;
         let dir_inode = ext4plus::inode::Inode::read(&fs, dir_idx).await.map_err(into_vfs_err)?;
         let mut dir = ext4plus::dir::Dir::open_inode(&fs, dir_inode).map_err(into_vfs_err)?;
@@ -634,6 +636,7 @@ impl DirNodeOps for Inode {
     async fn unlink(&self, name: &str) -> VfsResult<()> {
         let fs = &self.fs.inner;
         self.validate_inode_num(&fs, self.ino)?;
+        let _directory_guard = self.fs.directory_mutation.lock().await;
 
         let dir_idx = core::num::NonZeroU32::new(self.ino).ok_or(VfsError::InvalidData)?;
         let dir_inode = ext4plus::inode::Inode::read(&fs, dir_idx).await.map_err(into_vfs_err)?;
@@ -687,6 +690,7 @@ impl DirNodeOps for Inode {
         let fs = &self.fs.inner;
         self.validate_inode_num(&fs, self.ino)?;
         self.validate_inode_num(&fs, dst_dir.ino)?;
+        let _directory_guard = self.fs.directory_mutation.lock().await;
 
         let src_dir_idx = core::num::NonZeroU32::new(self.ino).ok_or(VfsError::InvalidData)?;
         let src_dir_inode = ext4plus::inode::Inode::read(&fs, src_dir_idx).await.map_err(into_vfs_err)?;

arceos/modules/axhal/src/dummy.rs

@@ -1,11 +1,12 @@
 //! Dummy implementation of platform-related interfaces defined in [`axplat`].
 
+#[cfg(feature = "irq")]
 use axplat::impl_plat_interface;
 
 use axplat::console::ConsoleIf;
 use axplat::init::InitIf;
 #[cfg(feature = "irq")]
-use axplat::irq::{IpiTarget, IrqHandler, IrqIf};
+use axplat::irq::{IpiError, IpiTarget, IrqHandler, IrqIf};
 use axplat::mem::{MemIf, RawRange};
 use axplat::power::PowerIf;
 use axplat::time::TimeIf;
@@ -90,7 +91,12 @@ impl TimeIf for DummyTime {
 #[impl_plat_interface]
 impl PowerIf for DummyPower {
     #[cfg(feature = "smp")]
-    fn cpu_boot(_cpu_id: usize, _stack_top_paddr: usize) {}
+    fn cpu_boot(
+        _cpu_id: usize,
+        _stack_top_paddr: usize,
+    ) -> Result<(), axplat::power::CpuBootError> {
+        Ok(())
+    }
 
     fn system_off() -> ! {
         unimplemented!()
@@ -114,7 +120,11 @@ impl IrqIf for DummyIrq {
         None
     }
 
-    fn handle(_irq: usize) {}
+    fn handle(_irq: usize, _cpu_id: usize) {}
 
-    fn send_ipi(_irq: usize, _target: IpiTarget) {}
+    fn cpu_online(_cpu_id: usize) {}
+
+    fn send_ipi(_irq: usize, _target: IpiTarget) -> Result<(), IpiError> {
+        Err(IpiError::NotSupported)
+    }
 }

arceos/modules/axhal/src/irq.rs

@@ -5,7 +5,7 @@ use axcpu::trap::{IRQ, register_trap_handler};
 pub use axplat::irq::{handle, register, set_enable, unregister};
 
 #[cfg(feature = "ipi")]
-pub use axplat::irq::{IpiTarget, send_ipi};
+pub use axplat::irq::{IpiError, IpiTarget, send_ipi};
 
 #[cfg(feature = "ipi")]
 pub use axconfig::devices::IPI_IRQ;
@@ -18,7 +18,7 @@ pub use axconfig::devices::IPI_IRQ;
 #[register_trap_handler(IRQ)]
 pub fn irq_handler(vector: usize) -> bool {
     let guard = kernel_guard::NoPreempt::new();
-    handle(vector);
+    handle(vector, crate::percpu::this_cpu_id());
     drop(guard); // rescheduling may occur when preemption is re-enabled.
     true
 }

arceos/modules/axhal/src/lib.rs

@@ -81,7 +81,7 @@ pub mod console {
 pub mod power {
     #[cfg(feature = "smp")]
     pub use axplat::power::cpu_boot;
-    pub use axplat::power::system_off;
+    pub use axplat::power::{CpuBootError, system_off};
 }
 
 /// Trap handling.
@@ -106,7 +106,6 @@ pub mod context {
 }
 
 pub use axcpu::asm;
-
 #[cfg(feature = "smp")]
 pub use axplat::init::{init_early_secondary, init_later_secondary};
 #[cfg(feature = "smp")]
@@ -152,12 +151,16 @@ pub fn get_bootarg() -> usize {
     *BOOT_ARG
 }
 
-/// The number of CPUs in the system. Based on the number declared by the
-/// platform crate and limited by the configured maximum CPU number.
+/// The number of possible CPUs in the system. Based on the number declared by
+/// the platform crate and limited by the configured maximum CPU number.
 #[cfg(feature = "smp")]
 static CPU_NUM: AtomicUsize = AtomicUsize::new(1);
 
-/// Gets the number of CPUs running in the system.
+/// Bit mask of CPUs that completed per-CPU initialization and can receive work.
+#[cfg(feature = "smp")]
+static ONLINE_CPU_MASK: AtomicUsize = AtomicUsize::new(0);
+
+/// Gets the number of possible CPUs in the system.
 ///
 /// When SMP is disabled, this function always returns 1.
 ///
@@ -186,6 +189,43 @@ pub fn cpu_num() -> usize {
     }
 }
 
+/// Returns a bit mask containing every online logical CPU.
+pub fn online_cpu_mask() -> usize {
+    #[cfg(feature = "smp")]
+    {
+        ONLINE_CPU_MASK.load(Ordering::Acquire)
+    }
+    #[cfg(not(feature = "smp"))]
+    {
+        1
+    }
+}
+
+/// Returns the number of CPUs that are ready to receive tasks and interrupts.
+pub fn online_cpu_count() -> usize {
+    online_cpu_mask().count_ones() as usize
+}
+
+/// Returns whether a logical CPU is online.
+pub fn is_cpu_online(cpu_id: usize) -> bool {
+    cpu_id < usize::BITS as usize && online_cpu_mask() & (1usize << cpu_id) != 0
+}
+
+/// Publishes a logical CPU after all of its per-CPU runtime state is initialized.
+pub fn mark_cpu_online(cpu_id: usize) {
+    assert!(cpu_id < cpu_num(), "invalid online CPU ID {cpu_id}");
+    assert!(
+        cpu_id < usize::BITS as usize,
+        "online CPU mask is too small"
+    );
+    #[cfg(feature = "irq")]
+    axplat::irq::cpu_online(cpu_id);
+    #[cfg(feature = "smp")]
+    ONLINE_CPU_MASK.fetch_or(1usize << cpu_id, Ordering::Release);
+    #[cfg(not(feature = "smp"))]
+    let _ = cpu_id;
+}
+
 /// Initializes the CPU number information.
 fn init_cpu_num() {
     #[cfg(feature = "smp")]
@@ -194,6 +234,11 @@ fn init_cpu_num() {
         let max_cpu_num = axconfig::plat::MAX_CPU_NUM;
         let cpu_num = plat_cpu_num.min(max_cpu_num);
 
+        assert!(
+            cpu_num <= usize::BITS as usize,
+            "configured CPU count exceeds the online CPU mask width"
+        );
+
         info!("CPU number: max = {max_cpu_num}, platform = {plat_cpu_num}, use = {cpu_num}",);
         ax_println!("smp = {}", cpu_num); // for test purposes