Introduce CPU clock and CPU timer

framework/aster-frame/src/arch/x86/trap.rs

@@ -1,5 +1,7 @@
 // SPDX-License-Identifier: MPL-2.0
 
+use core::sync::atomic::{AtomicBool, Ordering};
+
 use align_ext::AlignExt;
 use log::debug;
 #[cfg(feature = "intel_tdx")]
@@ -14,6 +16,7 @@ use crate::arch::{
 };
 use crate::{
     cpu::{CpuException, PageFaultErrorCode, PAGE_FAULT},
+    cpu_local,
     trap::call_irq_callback_functions,
     vm::{
         kspace::{KERNEL_PAGE_TABLE, LINEAR_MAPPING_BASE_VADDR, LINEAR_MAPPING_VADDR_RANGE},
@@ -22,6 +25,17 @@ use crate::{
     },
 };
 
+cpu_local! {
+    static IS_KERNEL_INTERRUPTED: AtomicBool = AtomicBool::new(false);
+}
+
+/// Returns true if this function is called within the context of an IRQ handler
+/// and the IRQ occurs while the CPU is executing in the kernel mode.
+/// Otherwise, it returns false.
+pub fn is_kernel_interrupted() -> bool {
+    IS_KERNEL_INTERRUPTED.load(Ordering::Acquire)
+}
+
 /// Only from kernel
 #[no_mangle]
 extern "sysv64" fn trap_handler(f: &mut TrapFrame) {
@@ -43,7 +57,9 @@ extern "sysv64" fn trap_handler(f: &mut TrapFrame) {
             }
         }
     } else {
+        IS_KERNEL_INTERRUPTED.store(true, Ordering::Release);
         call_irq_callback_functions(f);
+        IS_KERNEL_INTERRUPTED.store(false, Ordering::Release);
     }
 }
 

kernel/aster-nix/src/lib.rs

@@ -80,6 +80,7 @@ pub fn init() {
     device::init().unwrap();
     vdso::init();
     taskless::init();
+    process::init();
 }
 
 fn init_thread() {

kernel/aster-nix/src/process/mod.rs

@@ -32,3 +32,7 @@ pub use program_loader::{check_executable_file, load_program_to_vm};
 pub use rlimit::ResourceType;
 pub use term_status::TermStatus;
 pub use wait::{wait_child_exit, WaitOptions};
+
+pub(super) fn init() {
+    process::init();
+}

kernel/aster-nix/src/process/posix_thread/builder.rs

@@ -11,6 +11,7 @@ use crate::{
         Credentials, Process,
     },
     thread::{status::ThreadStatus, task, thread_table, Thread, Tid},
+    time::{clocks::ProfClock, TimerManager},
 };
 
 /// The builder to build a posix thread
@@ -94,6 +95,11 @@ impl PosixThreadBuilder {
         let thread = Arc::new_cyclic(|thread_ref| {
             let task = task::create_new_user_task(user_space, thread_ref.clone());
             let status = ThreadStatus::Init;
+
+            let prof_clock = ProfClock::new();
+            let virtual_timer_manager = TimerManager::new(prof_clock.user_clock().clone());
+            let prof_timer_manager = TimerManager::new(prof_clock.clone());
+
             let posix_thread = PosixThread {
                 process,
                 is_main_thread,
@@ -106,6 +112,9 @@ impl PosixThreadBuilder {
                 sig_context: Mutex::new(None),
                 sig_stack: Mutex::new(None),
                 robust_list: Mutex::new(None),
+                prof_clock,
+                virtual_timer_manager,
+                prof_timer_manager,
             };
 
             Thread::new(tid, task, posix_thread, status)

kernel/aster-nix/src/process/posix_thread/mod.rs

@@ -21,6 +21,7 @@ use crate::{
     prelude::*,
     process::signal::constants::SIGCONT,
     thread::{thread_table, Tid},
+    time::{clocks::ProfClock, Timer, TimerManager},
     util::write_val_to_user,
 };
 
@@ -62,6 +63,15 @@ pub struct PosixThread {
     /// FIXME: This field may be removed. For glibc applications with RESTORER flag set, the sig_context is always equals with rsp.
     sig_context: Mutex<Option<Vaddr>>,
     sig_stack: Mutex<Option<SigStack>>,
+
+    /// A profiling clock measures the user CPU time and kernel CPU time in the thread.
+    prof_clock: Arc<ProfClock>,
+
+    /// A manager that manages timers based on the user CPU time of the current thread.
+    virtual_timer_manager: Arc<TimerManager>,
+
+    /// A manager that manages timers based on the profiling clock of the current thread.
+    prof_timer_manager: Arc<TimerManager>,
 }
 
 impl PosixThread {
@@ -148,10 +158,39 @@ impl PosixThread {
         return_errno_with_message!(Errno::EPERM, "sending signal to the thread is not allowed.");
     }
 
-    pub(in crate::process) fn enqueue_signal(&self, signal: Box<dyn Signal>) {
+    /// Enqueues a thread-directed signal. This method should only be used for enqueue kernel
+    /// signal and fault signal.
+    pub fn enqueue_signal(&self, signal: Box<dyn Signal>) {
         self.sig_queues.enqueue(signal);
     }
 
+    /// Returns a reference to the profiling clock of the current thread.
+    pub fn prof_clock(&self) -> &Arc<ProfClock> {
+        &self.prof_clock
+    }
+
+    /// Creates a timer based on the profiling CPU clock of the current thread.
+    pub fn create_prof_timer<F>(&self, func: F) -> Arc<Timer>
+    where
+        F: Fn() + Send + Sync + 'static,
+    {
+        self.prof_timer_manager.create_timer(func)
+    }
+
+    /// Creates a timer based on the user CPU clock of the current thread.
+    pub fn create_virtual_timer<F>(&self, func: F) -> Arc<Timer>
+    where
+        F: Fn() + Send + Sync + 'static,
+    {
+        self.virtual_timer_manager.create_timer(func)
+    }
+
+    /// Checks the `TimerCallback`s that are managed by the `prof_timer_manager`.
+    /// If any have timed out, call the corresponding callback functions.
+    pub fn process_expired_timers(&self) {
+        self.prof_timer_manager.process_expired_timers();
+    }
+
     pub fn dequeue_signal(&self, mask: &SigMask) -> Option<Box<dyn Signal>> {
         self.sig_queues.dequeue(mask)
     }

kernel/aster-nix/src/process/process/mod.rs

@@ -1,15 +1,13 @@
 // SPDX-License-Identifier: MPL-2.0
 
+use self::timer_manager::PosixTimerManager;
 use super::{
     posix_thread::PosixThreadExt,
     process_table,
     process_vm::{Heap, InitStackReader, ProcessVm},
     rlimit::ResourceLimits,
     signal::{
-        constants::{SIGALRM, SIGCHLD},
-        sig_disposition::SigDispositions,
-        sig_mask::SigMask,
-        signals::{kernel::KernelSignal, Signal},
+        constants::SIGCHLD, sig_disposition::SigDispositions, sig_mask::SigMask, signals::Signal,
         Pauser,
     },
     status::ProcessStatus,
@@ -20,12 +18,8 @@ use crate::{
     fs::{file_table::FileTable, fs_resolver::FsResolver, utils::FileCreationMask},
     prelude::*,
     sched::nice::Nice,
-    thread::{
-        allocate_tid,
-        work_queue::{submit_work_item, work_item::WorkItem},
-        Thread,
-    },
-    time::{clocks::RealTimeClock, Timer},
+    thread::{allocate_tid, Thread},
+    time::clocks::ProfClock,
     vm::vmar::Vmar,
 };
 
@@ -34,6 +28,7 @@ mod job_control;
 mod process_group;
 mod session;
 mod terminal;
+mod timer_manager;
 
 use aster_rights::Full;
 use atomic::Atomic;
@@ -52,6 +47,10 @@ pub type Sid = u32;
 
 pub type ExitCode = i32;
 
+pub(super) fn init() {
+    timer_manager::init();
+}
+
 /// Process stands for a set of threads that shares the same userspace.
 pub struct Process {
     // Immutable Part
@@ -60,8 +59,7 @@ pub struct Process {
     process_vm: ProcessVm,
     /// Wait for child status changed
     children_pauser: Arc<Pauser>,
-    /// The timer counts down in real (i.e., wall clock) time
-    alarm_timer: Arc<Timer>,
+
     // Mutable Part
     /// The executable path.
     executable_path: RwLock<String>,
@@ -91,26 +89,12 @@ pub struct Process {
     // Signal
     /// Sig dispositions
     sig_dispositions: Arc<Mutex<SigDispositions>>,
-}
 
-fn create_process_timer_callback(process_ref: &Weak<Process>) -> impl Fn() {
-    let current_process = process_ref.clone();
-    let sent_signal = move || {
-        let signal = KernelSignal::new(SIGALRM);
-        if let Some(process) = current_process.upgrade() {
-            process.enqueue_signal(signal);
-        }
-    };
+    /// A profiling clock measures the user CPU time and kernel CPU time of the current process.
+    prof_clock: Arc<ProfClock>,
 
-    let work_func = Box::new(sent_signal);
-    let work_item = Arc::new(WorkItem::new(work_func));
-
-    move || {
-        submit_work_item(
-            work_item.clone(),
-            crate::thread::work_queue::WorkPriority::High,
-        );
-    }
+    /// A manager that manages timer resources and utilities of the process.
+    timer_manager: PosixTimerManager,
 }
 
 impl Process {
@@ -135,28 +119,26 @@ impl Process {
             Pauser::new_with_mask(sigmask)
         };
 
-        Arc::new_cyclic(|process_ref: &Weak<Process>| {
-            let callback = create_process_timer_callback(process_ref);
-            let alarm_timer = RealTimeClock::timer_manager().create_timer(callback);
+        let prof_clock = ProfClock::new();
 
-            Self {
-                pid,
-                threads: Mutex::new(threads),
-                executable_path: RwLock::new(executable_path),
-                process_vm,
-                children_pauser,
-                alarm_timer,
-                status: Mutex::new(ProcessStatus::Uninit),
-                parent: Mutex::new(parent),
-                children: Mutex::new(BTreeMap::new()),
-                process_group: Mutex::new(Weak::new()),
-                file_table,
-                fs,
-                umask,
-                sig_dispositions,
-                resource_limits: Mutex::new(resource_limits),
-                nice: Atomic::new(nice),
-            }
+        Arc::new_cyclic(|process_ref: &Weak<Process>| Self {
+            pid,
+            threads: Mutex::new(threads),
+            executable_path: RwLock::new(executable_path),
+            process_vm,
+            children_pauser,
+            status: Mutex::new(ProcessStatus::Uninit),
+            parent: Mutex::new(parent),
+            children: Mutex::new(BTreeMap::new()),
+            process_group: Mutex::new(Weak::new()),
+            file_table,
+            fs,
+            umask,
+            sig_dispositions,
+            resource_limits: Mutex::new(resource_limits),
+            nice: Atomic::new(nice),
+            timer_manager: PosixTimerManager::new(&prof_clock, process_ref),
+            prof_clock,
         })
     }
 
@@ -233,8 +215,14 @@ impl Process {
         self.pid
     }
 
-    pub fn alarm_timer(&self) -> &Arc<Timer> {
-        &self.alarm_timer
+    /// Gets the profiling clock of the process.
+    pub fn prof_clock(&self) -> &Arc<ProfClock> {
+        &self.prof_clock
+    }
+
+    /// Gets the timer resources and utilities of the process.
+    pub fn timer_manager(&self) -> &PosixTimerManager {
+        &self.timer_manager
     }
 
     pub fn threads(&self) -> &Mutex<Vec<Arc<Thread>>> {
@@ -288,7 +276,7 @@ impl Process {
 
     // *********** Process group & Session***********
 
-    /// Returns the process group id of the process.
+    /// Returns the process group ID of the process.
     pub fn pgid(&self) -> Pgid {
         if let Some(process_group) = self.process_group.lock().upgrade() {
             process_group.pgid()
@@ -341,7 +329,7 @@ impl Process {
     ///
     /// This method may return the following errors:
     /// * `EPERM`, if the process is a process group leader, or some existing session
-    /// or process group has the same id as the process.
+    /// or process group has the same ID as the process.
     pub fn to_new_session(self: &Arc<Self>) -> Result<Arc<Session>> {
         if self.is_session_leader() {
             return Ok(self.session().unwrap());
@@ -439,7 +427,7 @@ impl Process {
             if pgid != self.pid() {
                 return_errno_with_message!(
                     Errno::EPERM,
-                    "the new process group should have the same id as the process."
+                    "the new process group should have the same ID as the process."
                 );
             }
 
@@ -637,8 +625,6 @@ pub fn current() -> Arc<Process> {
 #[cfg(ktest)]
 mod test {
 
-    use spin::Once;
-
     use super::*;
 
     fn new_process(parent: Option<Arc<Process>>) -> Arc<Process> {

kernel/aster-nix/src/process/process/timer_manager.rs

@@ -0,0 +1,200 @@
+// SPDX-License-Identifier: MPL-2.0
+
+use alloc::{
+    boxed::Box,
+    sync::{Arc, Weak},
+    vec::Vec,
+};
+use core::time::Duration;
+
+use aster_frame::{
+    arch::{
+        timer::{self, TIMER_FREQ},
+        x86::trap::is_kernel_interrupted,
+    },
+    sync::Mutex,
+};
+use id_alloc::IdAlloc;
+
+use super::Process;
+use crate::{
+    process::{
+        posix_thread::PosixThreadExt,
+        signal::{constants::SIGALRM, signals::kernel::KernelSignal},
+    },
+    thread::{
+        work_queue::{submit_work_item, work_item::WorkItem},
+        Thread,
+    },
+    time::{
+        clocks::{ProfClock, RealTimeClock},
+        Timer, TimerManager,
+    },
+};
+
+/// Updates the CPU time recorded in the CPU clocks of current Process.
+///
+/// This function will be invoked at the system timer interrupt, and
+/// invoke the callbacks of expired timers which are based on the updated
+/// CPU clock.
+fn update_cpu_time() {
+    let current_thread = Thread::current();
+    if let Some(posix_thread) = current_thread.as_posix_thread() {
+        let process = posix_thread.process();
+        let timer_manager = process.timer_manager();
+        let jiffies_interval = Duration::from_millis(1000 / TIMER_FREQ);
+        // Based on whether the timer interrupt occurs in kernel mode or user mode,
+        // the function will add the duration of one timer interrupt interval to the
+        // corresponding CPU clocks.
+        if is_kernel_interrupted() {
+            posix_thread
+                .prof_clock()
+                .kernel_clock()
+                .add_time(jiffies_interval);
+            process
+                .prof_clock()
+                .kernel_clock()
+                .add_time(jiffies_interval);
+        } else {
+            posix_thread
+                .prof_clock()
+                .user_clock()
+                .add_time(jiffies_interval);
+            process.prof_clock().user_clock().add_time(jiffies_interval);
+            timer_manager
+                .virtual_timer()
+                .timer_manager()
+                .process_expired_timers();
+        }
+        timer_manager
+            .prof_timer()
+            .timer_manager()
+            .process_expired_timers();
+        posix_thread.process_expired_timers();
+    }
+}
+
+/// Registers a function to update the CPU clock in processes and
+/// threads during the system timer interrupt.
+pub(super) fn init() {
+    timer::register_callback(update_cpu_time);
+}
+
+/// Represents timer resources and utilities for a POSIX process.
+pub struct PosixTimerManager {
+    /// A real-time countdown timer, measuring in wall clock time.
+    alarm_timer: Arc<Timer>,
+    /// A timer based on user CPU clock.
+    virtual_timer: Arc<Timer>,
+    /// A timer based on the profiling clock.
+    prof_timer: Arc<Timer>,
+    /// An ID allocator to allocate unique timer IDs.
+    id_allocator: Mutex<IdAlloc>,
+    /// A container managing all POSIX timers created by `timer_create()` syscall
+    /// within the process context.
+    posix_timers: Mutex<Vec<Option<Arc<Timer>>>>,
+}
+
+fn create_process_timer_callback(process_ref: &Weak<Process>) -> impl Fn() + Clone {
+    let current_process = process_ref.clone();
+    let sent_signal = move || {
+        let signal = KernelSignal::new(SIGALRM);
+        if let Some(process) = current_process.upgrade() {
+            process.enqueue_signal(signal);
+        }
+    };
+
+    let work_func = Box::new(sent_signal);
+    let work_item = Arc::new(WorkItem::new(work_func));
+
+    move || {
+        submit_work_item(
+            work_item.clone(),
+            crate::thread::work_queue::WorkPriority::High,
+        );
+    }
+}
+
+impl PosixTimerManager {
+    pub(super) fn new(prof_clock: &Arc<ProfClock>, process_ref: &Weak<Process>) -> Self {
+        const MAX_NUM_OF_POSIX_TIMERS: usize = 10000;
+
+        let callback = create_process_timer_callback(process_ref);
+
+        let alarm_timer = RealTimeClock::timer_manager().create_timer(callback.clone());
+
+        let virtual_timer =
+            TimerManager::new(prof_clock.user_clock().clone()).create_timer(callback.clone());
+        let prof_timer = TimerManager::new(prof_clock.clone()).create_timer(callback);
+
+        Self {
+            alarm_timer,
+            virtual_timer,
+            prof_timer,
+            id_allocator: Mutex::new(IdAlloc::with_capacity(MAX_NUM_OF_POSIX_TIMERS)),
+            posix_timers: Mutex::new(Vec::new()),
+        }
+    }
+
+    /// Gets the alarm timer of the corresponding process.
+    pub fn alarm_timer(&self) -> &Arc<Timer> {
+        &self.alarm_timer
+    }
+
+    /// Gets the virtual timer of the corresponding process.
+    pub fn virtual_timer(&self) -> &Arc<Timer> {
+        &self.virtual_timer
+    }
+
+    /// Gets the profiling timer of the corresponding process.
+    pub fn prof_timer(&self) -> &Arc<Timer> {
+        &self.prof_timer
+    }
+
+    /// Creates a timer based on the profiling CPU clock of the current process.
+    pub fn create_prof_timer<F>(&self, func: F) -> Arc<Timer>
+    where
+        F: Fn() + Send + Sync + 'static,
+    {
+        self.prof_timer.timer_manager().create_timer(func)
+    }
+
+    /// Creates a timer based on the user CPU clock of the current process.
+    pub fn create_virtual_timer<F>(&self, func: F) -> Arc<Timer>
+    where
+        F: Fn() + Send + Sync + 'static,
+    {
+        self.virtual_timer.timer_manager().create_timer(func)
+    }
+
+    /// Adds a POSIX timer to the managed `posix_timers`, and allocate a timer ID for this timer.
+    /// Return the timer ID.
+    pub fn add_posix_timer(&self, posix_timer: Arc<Timer>) -> usize {
+        let mut timers = self.posix_timers.lock();
+        // Holding the lock of `posix_timers` is required to operate the `id_allocator`.
+        let timer_id = self.id_allocator.lock().alloc().unwrap();
+        if timers.len() < timer_id + 1 {
+            timers.resize(timer_id + 1, None);
+        }
+        // The ID allocated is not used by any other timers so this index in `timers`
+        // must be `None`.
+        timers[timer_id] = Some(posix_timer);
+        timer_id
+    }
+
+    /// Finds a POSIX timer by the input `timer_id`.
+    pub fn find_posix_timer(&self, timer_id: usize) -> Option<Arc<Timer>> {
+        self.posix_timers.lock()[timer_id].clone()
+    }
+
+    /// Removes the POSIX timer with the ID `timer_id`.
+    pub fn remove_posix_timer(&self, timer_id: usize) -> Option<Arc<Timer>> {
+        let mut timers = self.posix_timers.lock();
+        let timer = timers[timer_id].take();
+        if timer.is_some() {
+            // Holding the lock of `posix_timers` is required to operate the `id_allocator`.
+            self.id_allocator.lock().free(timer_id);
+        }
+        timer
+    }
+}

kernel/aster-nix/src/time/clocks/cpu_clock.rs

@@ -0,0 +1,67 @@
+// SPDX-License-Identifier: MPL-2.0
+
+use alloc::sync::Arc;
+use core::time::Duration;
+
+use aster_frame::sync::SpinLock;
+
+use crate::time::Clock;
+
+/// A clock used to record the CPU time for processes and threads.
+pub struct CpuClock {
+    time: SpinLock<Duration>,
+}
+
+/// A profiling clock that contains a user CPU clock and a kernel CPU clock.
+/// These two clocks record the CPU time in user mode and kernel mode respectively.
+/// Reading this clock directly returns the sum of both times.
+pub struct ProfClock {
+    user_clock: Arc<CpuClock>,
+    kernel_clock: Arc<CpuClock>,
+}
+
+impl CpuClock {
+    /// Creates a new `CpuClock`. The recorded time is initialized to 0.
+    pub fn new() -> Arc<Self> {
+        Arc::new(Self {
+            time: SpinLock::new(Duration::ZERO),
+        })
+    }
+
+    /// Adds `interval` to the original recorded time to update the `CpuClock`.
+    pub fn add_time(&self, interval: Duration) {
+        *self.time.lock_irq_disabled() += interval;
+    }
+}
+
+impl Clock for CpuClock {
+    fn read_time(&self) -> Duration {
+        *self.time.lock_irq_disabled()
+    }
+}
+
+impl ProfClock {
+    /// Creates a new `ProfClock`. The recorded time is initialized to 0.
+    pub fn new() -> Arc<Self> {
+        Arc::new(Self {
+            user_clock: CpuClock::new(),
+            kernel_clock: CpuClock::new(),
+        })
+    }
+
+    /// Returns a reference to the user CPU clock in this profiling clock.
+    pub fn user_clock(&self) -> &Arc<CpuClock> {
+        &self.user_clock
+    }
+
+    /// Returns a reference to the kernel CPU clock in this profiling clock.
+    pub fn kernel_clock(&self) -> &Arc<CpuClock> {
+        &self.kernel_clock
+    }
+}
+
+impl Clock for ProfClock {
+    fn read_time(&self) -> Duration {
+        self.user_clock.read_time() + self.kernel_clock.read_time()
+    }
+}

kernel/aster-nix/src/time/clocks/mod.rs

@@ -1,7 +1,9 @@
 // SPDX-License-Identifier: MPL-2.0
 
+pub use cpu_clock::*;
 pub use system_wide::*;
 
+mod cpu_clock;
 mod system_wide;
 
 pub(super) fn init() {

kernel/aster-nix/src/time/clocks/system_wide.rs

@@ -124,6 +124,11 @@ impl BootTimeClock {
     pub fn get() -> &'static Arc<BootTimeClock> {
         CLOCK_BOOTTIME_INSTANCE.get().unwrap()
     }
+
+    /// Get the cpu-local system-wide `TimerManager` singleton of this clock.
+    pub fn timer_manager() -> &'static Arc<TimerManager> {
+        CLOCK_BOOTTIME_MANAGER.get().unwrap()
+    }
 }
 
 impl Clock for JiffiesClock {
@@ -232,7 +237,7 @@ define_system_clocks! {
     CLOCK_BOOTTIME          => BootTimeClock,
 }
 
-define_timer_managers![CLOCK_REALTIME, CLOCK_MONOTONIC,];
+define_timer_managers![CLOCK_REALTIME, CLOCK_MONOTONIC, CLOCK_BOOTTIME,];
 
 /// Init the system-wide clocks.
 fn init_system_wide_clocks() {