feat: tasks

src/arch/mod.rs

@@ -5,4 +5,7 @@ mod x86_64;
 pub use x86_64::*;
 
 #[cfg(target_arch = "x86_64")]
-pub(crate) use x86_64::{PageTableCreateError, PageTableMapError, PageTableUnmapError};
+pub(crate) use x86_64::{
+    PageTableCreateError, PageTableMapError, PageTableUnmapError, ThreadContext, set_kernel_stack,
+    switch_context,
+};

src/arch/x86_64/apic.rs

@@ -13,7 +13,6 @@ use crate::{
     memory::{
         AddressSpace, CachePolicy, FrameAllocator, PagePermissions, PhysicalAddr, VirtualAddr,
     },
-    println,
 };
 
 const APIC_ID: u32 = 0x20;

src/arch/x86_64/cpu.rs

@@ -1,8 +1,104 @@
-use core::arch::asm;
+use core::arch::{asm, naked_asm};
+
+use crate::memory::VirtualAddr;
+
+#[repr(C, align(64))]
+pub struct CpuLocal {
+    pub kernel_stack_top: usize,
+    pub user_rsp_scratch: usize,
+    pub cpu_id: u32,
+}
+
+pub static mut BOOT_CPU: CpuLocal = CpuLocal {
+    kernel_stack_top: 0,
+    user_rsp_scratch: 0,
+    cpu_id: 0,
+};
+
+#[derive(Debug)]
+pub struct ThreadContext {
+    rsp: usize,
+}
+
+impl ThreadContext {
+    pub fn new(
+        user_entry: VirtualAddr,
+        user_stack: VirtualAddr,
+        kernel_stack_top: VirtualAddr,
+    ) -> Self {
+        // Stack layout (grows downwards from kernel_stack_top):
+        // [top - 8]  = user_thread_entry (popped by `ret`)
+        // [top - 16] = rbp (0)
+        // [top - 24] = rbx (user_stack)
+        // [top - 32] = r12 (user_entry)
+        // [top - 40] = r13 (0)
+        // [top - 48] = r14 (0)
+        // [top - 56] = r15 (0) <- initial rsp
+
+        let stack_ptr = (kernel_stack_top.as_usize() - 56) as *mut usize;
+
+        unsafe {
+            stack_ptr.add(0).write(0); // r15
+            stack_ptr.add(1).write(0); // r14
+            stack_ptr.add(2).write(0); // r13
+            stack_ptr.add(3).write(user_entry.as_usize()); // r12 (user_entry)
+            stack_ptr.add(4).write(user_stack.as_usize()); // rbx (user_stack)
+            stack_ptr.add(5).write(0); // rbp (0)
+            stack_ptr
+                .add(6)
+                .write(user_thread_entry as *const () as usize); // return address
+        }
+
+        Self {
+            rsp: kernel_stack_top.as_usize() - 56,
+        }
+    }
+
+    pub fn empty() -> Self {
+        Self { rsp: 0 }
+    }
+}
+
+#[unsafe(naked)]
+unsafe extern "C" fn user_thread_entry() -> ! {
+    naked_asm!(
+        // r12 = user_entry, rbx = user_stack
+        "mov rdi, r12",
+        "mov rsi, rbx",
+        "call {enter_user}",
+        enter_user = sym crate::arch::enter_user,
+    );
+}
+
+#[unsafe(naked)]
+pub unsafe extern "C" fn switch_context(prev: *mut ThreadContext, next: *const ThreadContext) {
+    naked_asm!(
+        "push rbp",
+        "push rbx",
+        "push r12",
+        "push r13",
+        "push r14",
+        "push r15",
+        "",
+        // save current rsp into prev.rsp
+        "mov [rdi], rsp",
+        // load next rsp into rsp
+        "mov rsp, [rsi]",
+        "",
+        "pop r15",
+        "pop r14",
+        "pop r13",
+        "pop r12",
+        "pop rbx",
+        "pop rbp",
+        "ret",
+    )
+}
 
 #[derive(Debug)]
 pub enum CpuFeaturesError {
     CpuidFeaturesNotSupported,
+    SyscallNotSupported,
     InvalidPhysicalAddressWidth,
     InvalidVirtualAddressWidth,
 }
@@ -40,6 +136,18 @@ pub fn detect_features_and_enable() -> Result<CpuFeatures, CpuFeaturesError> {
 
     features.nx_supported = cpuid_result.edx & (1 << 20) != 0;
 
+    // TODO: on AMD K6 *only*, this bit is bit 10, should we consider that edge case?
+    let syscall_supported = cpuid_result.edx & (1 << 11) != 0;
+    if !syscall_supported {
+        return Err(CpuFeaturesError::SyscallNotSupported);
+    }
+
+    unsafe {
+        let efer = read_msr(IA32_EFER);
+        let value = efer | 1;
+        write_msr(IA32_EFER, value);
+    };
+
     let cpuid_result = core::arch::x86_64::__cpuid(0x80000008);
 
     features.physical_address_bits = (cpuid_result.eax & 0xFF) as u8;

src/arch/x86_64/gdt.rs

@@ -3,8 +3,8 @@ use core::arch::asm;
 use crate::memory::VirtualAddr;
 
 #[repr(C, align(8))]
-struct Gdt {
+pub(super) struct Gdt {
-    entries: [u64; 7],
+    pub entries: [u64; 7],
 }
 
 impl Gdt {
@@ -49,12 +49,12 @@ const DOUBLE_FAULT_STACK_SIZE: usize = 16 * 1024;
 
 pub(super) const KERNEL_CODE_SELECTOR: u16 = 1 * 8;
 pub(super) const KERNEL_DATA_SELECTOR: u16 = 2 * 8;
-pub(super) const USER_CODE_SELECTOR: u16 = (3 * 8) | 3;
+pub(super) const USER_DATA_SELECTOR: u16 = (3 * 8) | 3;
-pub(super) const USER_DATA_SELECTOR: u16 = (4 * 8) | 3;
+pub(super) const USER_CODE_SELECTOR: u16 = (4 * 8) | 3;
 pub(super) const TSS_SELECTOR: u16 = 5 * 8;
 
 #[repr(align(16))]
-#[allow(dead_code)] // field 0 is read, rust just cant tell
+#[allow(unused)] // field 0 is read, rust just cant tell
 struct ExceptionStack([u8; DOUBLE_FAULT_STACK_SIZE]);
 
 static mut GDT: Gdt = Gdt::new();
@@ -78,8 +78,10 @@ pub fn init() {
                 0,
                 kernel_code_descriptor(),
                 kernel_data_descriptor(),
-                user_code_descriptor(),
+                // In Long Mode, userland CS will be loaded from STAR 63:48 + 16
+                // and userland SS from STAR 63:48 + 8 on SYSRET.
                 user_data_descriptor(),
+                user_code_descriptor(),
                 tss_low,
                 tss_high,
             ],

src/arch/x86_64/interrupts/mod.rs

@@ -6,6 +6,33 @@ mod idt;
 
 pub use idt::idt_init as init;
 
+#[inline(always)]
+pub fn disable_interrupts_and_save() -> u64 {
+    let flags: u64;
+
+    unsafe {
+        asm!("
+            pushfq",
+            "pop {flags}",
+            "cli",
+            flags = out(reg) flags,
+        );
+    }
+
+    flags
+}
+
+#[inline(always)]
+pub fn restore_interrupts(flags: u64) {
+    unsafe {
+        asm!(
+            "push {flags}",
+            "popfq",
+            flags = in(reg) flags,
+        );
+    }
+}
+
 #[inline(always)]
 pub fn disable_interrupts() {
     unsafe {

src/arch/x86_64/io_apic.rs

@@ -3,7 +3,6 @@ use crate::{
         AddressSpace, CachePolicy, FrameAllocator, PagePermissions, PhysicalAddr, VirtualAddr,
     },
     platform::acpi::{InterruptPolarity, TriggerMode},
-    println,
 };
 
 const IOWIN: usize = 0x10;

src/arch/x86_64/mod.rs

@@ -1,16 +1,18 @@
-pub mod apic;
+pub(super) mod apic;
 mod cpu;
 mod gdt;
 mod interrupts;
-pub mod io_apic;
+pub(super) mod io_apic;
 mod paging;
 mod pit;
 pub mod port;
+mod syscall;
 pub mod timer;
 
 use core::arch::asm;
 
-pub use interrupts::disable_interrupts;
+pub use cpu::{ThreadContext, switch_context};
+pub use interrupts::{disable_interrupts, disable_interrupts_and_save, restore_interrupts};
 pub(crate) use paging::{
     MapError as PageTableMapError, PageTableCreateError, UnmapError as PageTableUnmapError,
 };
@@ -22,11 +24,7 @@ pub struct ArchState {
 
 use crate::{
     KernelHandoff,
-    arch::{
+    arch::x86_64::cpu::BOOT_CPU,
-        apic::LocalApic,
-        io_apic::{IOAPIC_VIRTUAL_ADDRESS, IoApic},
-        x86_64::interrupts::apic_vectors::PIT_CALIBRATION_VECTOR,
-    },
     memory::{AddressSpace, FrameAllocator, VirtualAddr},
     platform::acpi::Madt,
     println,
@@ -45,6 +43,14 @@ pub fn init() -> ArchState {
     ArchState { paging }
 }
 
+pub fn set_kernel_stack(stack_top: VirtualAddr) {
+    gdt::set_kernel_stack(stack_top);
+
+    unsafe {
+        BOOT_CPU.kernel_stack_top = stack_top.as_usize();
+    }
+}
+
 #[derive(Debug)]
 #[allow(unused)]
 pub enum InterruptInitError {
@@ -60,8 +66,8 @@ pub enum InterruptInitError {
 
 #[derive(Debug)]
 pub struct InterruptController {
-    local_apic: LocalApic,
+    local_apic: apic::LocalApic,
-    io_apic: IoApic,
+    io_apic: io_apic::IoApic,
     local_timer_frequency: u64,
 }
 
@@ -89,7 +95,7 @@ pub fn init_interrupt_controller(
         io_apic_info.id,
         io_apic_info.apic_address,
         io_apic_info.global_system_interrupt_base,
-        IOAPIC_VIRTUAL_ADDRESS,
+        io_apic::IOAPIC_VIRTUAL_ADDRESS,
         allocator,
         address_space,
     )
@@ -107,7 +113,7 @@ pub fn init_interrupt_controller(
         .configure_masked(
             pit_route.gsi,
             io_apic::RedirectionConfig {
-                vector: PIT_CALIBRATION_VECTOR,
+                vector: interrupts::apic_vectors::PIT_CALIBRATION_VECTOR,
                 destination,
                 polarity: pit_route.polarity,
                 trigger: pit_route.trigger,
@@ -165,6 +171,9 @@ pub unsafe fn enter_kernel(stack_top: VirtualAddr, handoff: *mut KernelHandoff)
     unsafe {
         gdt::set_kernel_stack(stack_top);
 
+        BOOT_CPU.kernel_stack_top = stack_top.as_usize();
+        syscall::init(&raw const BOOT_CPU);
+
         asm!(
             "mov rsp, {stack_top}",
             "xor rbp, rbp",
@@ -193,13 +202,31 @@ pub unsafe fn enter_user(
             "mov ds, {user_data_selector:x}",
             "mov es, {user_data_selector:x}",
             "mov fs, {user_data_selector:x}",
-            "mov gs, {user_data_selector:x}", // ss is handled by iretq
+            "mov gs, {user_data_selector:x}",
 
             "push {user_data_selector}",
             "push {user_stack_pointer}",
-            "pushfq",
+            "push 0x202", // RFLAGS (IF=1, bit 1 reserved=1)
             "push {user_code_selector}",
             "push {user_instruction_pointer}",
+
+            // clear GPRs
+            "xor rax, rax",
+            "xor rbx, rbx",
+            "xor rcx, rcx",
+            "xor rdx, rdx",
+            "xor rsi, rsi",
+            "xor rdi, rdi",
+            "xor rbp, rbp",
+            "xor r8, r8",
+            "xor r9, r9",
+            "xor r10, r10",
+            "xor r11, r11",
+            "xor r12, r12",
+            "xor r13, r13",
+            "xor r14, r14",
+            "xor r15, r15",
+
             "iretq",
             user_data_selector = in(reg) gdt::USER_DATA_SELECTOR as usize,
             user_code_selector = in(reg) gdt::USER_CODE_SELECTOR as usize,

src/arch/x86_64/paging.rs

@@ -264,12 +264,21 @@ pub(crate) enum PageTableCreateError {
     OutOfFrames,
 }
 
+#[derive(Debug)]
 pub struct PageTable {
     pub direct_map: DirectMap,
     config: PagingConfig,
     frame: OwnedFrame,
 }
 
+impl PartialEq for PageTable {
+    fn eq(&self, other: &Self) -> bool {
+        self.frame.frame_address() == other.frame.frame_address()
+    }
+}
+
+impl Eq for PageTable {}
+
 impl PageTable {
     pub fn new(
         direct_map: DirectMap,

src/arch/x86_64/syscall.rs

@@ -0,0 +1,111 @@
+use core::arch::{asm, naked_asm};
+
+use crate::arch::x86_64::{
+    cpu::{CpuLocal, write_msr},
+    gdt::{KERNEL_CODE_SELECTOR, KERNEL_DATA_SELECTOR},
+};
+
+const IA32_STAR: u32 = 0xC000_0081;
+const IA32_LSTAR: u32 = 0xC000_0082;
+const IA32_CSTAR: u32 = 0xC000_0083;
+const IA32_FMASK: u32 = 0xC000_0084;
+const IA32_GS_BASE: u32 = 0xC000_0101;
+const IA32_KERNEL_GS_BASE: u32 = 0xC000_0102;
+
+const RFLAGS_MASK: u64 = 0x257FD5; // Clear IF, TF, DF, IOPL, NT, AC
+
+#[repr(C)]
+#[derive(Debug)]
+struct SyscallFrame {
+    pub r15: u64,
+    pub r14: u64,
+    pub r13: u64,
+    pub r12: u64,
+    pub rbp: u64,
+    pub rbx: u64,
+    pub r9: u64,          // arg5
+    pub r8: u64,          // arg4
+    pub r10: u64,         // arg3
+    pub rdx: u64,         // arg2
+    pub rsi: u64,         // arg1
+    pub rdi: u64,         // arg0
+    pub rax: u64,         // syscall number on entry / return value on exit
+    pub user_rip: u64,    // rcx
+    pub user_rflags: u64, // r11
+    pub user_rsp: u64,
+}
+
+pub fn init(cpu_local: *const CpuLocal) {
+    unsafe {
+        let star = ((KERNEL_DATA_SELECTOR as u64) << 48) | ((KERNEL_CODE_SELECTOR as u64) << 32);
+        write_msr(IA32_STAR, star);
+        write_msr(IA32_LSTAR, syscall_entry as *const () as u64);
+        write_msr(IA32_CSTAR, 0);
+        write_msr(IA32_FMASK, RFLAGS_MASK);
+        write_msr(IA32_GS_BASE, 0);
+        write_msr(IA32_KERNEL_GS_BASE, cpu_local as u64);
+    }
+}
+
+#[unsafe(naked)]
+unsafe extern "C" fn syscall_entry() {
+    naked_asm!(
+        "swapgs",
+        "mov gs:[8], rsp", // user_rsp_scratch
+        "mov rsp, gs:[0]", // kernel_stack_top
+        "",
+        // build the syscall frame
+        "push qword ptr gs:[8]", // user_rsp
+        "push r11",              // user_rflags
+        "push rcx",              // user_rip
+        "push rax",
+        "push rdi",
+        "push rsi",
+        "push rdx",
+        "push r10",
+        "push r8",
+        "push r9",
+        "push rbx",
+        "push rbp",
+        "push r12",
+        "push r13",
+        "push r14",
+        "push r15",
+        "",
+        // Syscall calling convention:
+        // Syscall number in rax, args in rdi, rsi, rdx, r10, r8, r9
+        "mov rdi, rsp",
+        "call {dispatch}",
+        "",
+        // restore the syscall frame
+        "pop r15",
+        "pop r14",
+        "pop r13",
+        "pop r12",
+        "pop rbp",
+        "pop rbx",
+        "pop r9",
+        "pop r8",
+        "pop r10",
+        "pop rdx",
+        "pop rsi",
+        "pop rdi",
+        "pop rax",              // return value
+        "pop rcx",              // user_rip for sysret
+        "pop r11",              // user_rflags for sysret
+        "pop qword ptr gs:[8]", // user_rsp
+        "",
+        "mov rsp, gs:[8]", // switch to user stack
+        "swapgs",
+        "sysretq",
+        dispatch = sym syscall_dispatch,
+    );
+}
+
+extern "C" fn syscall_dispatch(frame: &mut SyscallFrame) {
+    let ret = crate::task::syscall::handle(
+        frame.rax, frame.rdi, frame.rsi, frame.rdx, frame.r10, frame.r8, frame.r9,
+    );
+
+    frame.rax = ret;
+}

src/main.rs

@@ -8,12 +8,14 @@ mod boot;
 mod debug;
 mod memory;
 mod platform;
+mod task;
 
 use crate::{
     debug::serial,
     memory::{
-        AddressSpace, KernelStack, MemoryRegionKind, PagePermissions, UserStack, VirtualAddr,
+        AddressSpace, KernelStackPool, MemoryRegionKind, PagePermissions, UserStack, VirtualAddr,
     },
+    task::tcb::Tcb,
 };
 
 pub struct KernelHandoff {
@@ -21,6 +23,7 @@ pub struct KernelHandoff {
     address_space: AddressSpace,
     direct_map: memory::DirectMap,
     boot_info: boot::BootInfo,
+    kernel_stack_pool: KernelStackPool,
     handoff_frame: memory::OwnedFrame,
 }
 
@@ -49,7 +52,10 @@ pub extern "C" fn _start() -> ! {
 
     println!("Entering kernel main...");
 
-    let bootstrap_stack = KernelStack::allocate(&mut address_space, &mut allocator)
+    let mut kernel_stack_pool = KernelStackPool::new();
+
+    let kernel_stack = kernel_stack_pool
+        .allocate(&mut address_space, &mut allocator)
         .expect("failed to allocate bootstrap stack");
 
     let handoff_frame = allocator
@@ -59,12 +65,13 @@ pub extern "C" fn _start() -> ! {
         .to_virtual(handoff_frame.frame_address().start_address())
         .expect("failed to map kernel handoff");
 
-    let bootstrap_stack_top = bootstrap_stack.top();
+    let bootstrap_stack_top = kernel_stack.top();
     let handoff = KernelHandoff {
         allocator,
         address_space,
         direct_map,
         boot_info,
+        kernel_stack_pool,
         handoff_frame,
     };
 
@@ -82,13 +89,21 @@ pub extern "C" fn _start() -> ! {
 }
 
 pub unsafe extern "C" fn kernel_main(handoff: *mut KernelHandoff) -> ! {
-    let (mut allocator, mut address_space, direct_map, boot_info, handoff_frame) = unsafe {
+    let (
+        mut allocator,
+        mut address_space,
+        direct_map,
+        boot_info,
+        mut kernel_stack_pool,
+        handoff_frame,
+    ) = unsafe {
         let handoff = handoff.read();
         (
             handoff.allocator,
             handoff.address_space,
             handoff.direct_map,
             handoff.boot_info,
+            handoff.kernel_stack_pool,
             handoff.handoff_frame,
         )
     };
@@ -106,19 +121,31 @@ pub unsafe extern "C" fn kernel_main(handoff: *mut KernelHandoff) -> ! {
 
     let acpi = platform::acpi::init(&boot_info, direct_map).expect("failed to initialize ACPI");
 
+    println!("Parsing MADT...");
+
     let madt = acpi
         .madt()
         .expect("failed to parse ACPI")
         .expect("MADT not found");
 
+    println!("Initializing interrupt controller...");
+
     let interrupt_controller =
         arch::init_interrupt_controller(&madt, &mut allocator, &mut address_space)
             .expect("failed to initialize interrupt controller");
 
+    println!("Creating user address space...");
+
+    let task_kernel_stack = kernel_stack_pool
+        .allocate(&mut address_space, &mut allocator)
+        .expect("failed to allocate user task stack");
+
     let mut user_addr_space = address_space
         .new_user(&mut allocator)
         .expect("failed to create user address space");
 
+    println!("Allocating user stack...");
+
     let user_stack = UserStack::allocate(&mut user_addr_space, &mut allocator)
         .expect("failed to allocate user stack");
 
@@ -138,21 +165,37 @@ pub unsafe extern "C" fn kernel_main(handoff: *mut KernelHandoff) -> ! {
         .expect("failed to map user code page");
 
     unsafe {
-        user_addr_space.activate();
+        let user_code: &[u8] = &[
-
-        let user_code: [u8; 5] = [
             0xCC, // INT3
-            0xCD, 0x80, // INT 0x80
+            0x0F, 0x05, // SYSCALL
             0xEB, 0xFE, // JMP -2 (loop forever if exit returns)
         ];
 
+        let user_code_virtual = direct_map
+            .translate(user_code_page.start_address())
+            .unwrap();
+
+        println!("Copying user code to {:#X}", user_code_virtual.as_usize());
+
         core::ptr::copy_nonoverlapping(
             user_code.as_ptr(),
-            user_instruction_pointer.as_mut_ptr::<u8>(),
+            user_code_virtual.as_mut_ptr::<u8>(),
             user_code.len(),
         );
 
-        arch::enter_user(user_instruction_pointer, user_stack.top());
+        println!("Activating user address space...");
+
+        let user_tcb = Tcb::new_user(
+            0,
+            user_addr_space,
+            task_kernel_stack,
+            user_instruction_pointer,
+            user_stack.top(),
+        );
+
+        println!("Running first user task {user_tcb:?}...");
+
+        task::tcb::run_first_user(&user_tcb);
     }
 
     hcf();

src/memory/address_space.rs

@@ -88,6 +88,7 @@ enum AddressSpaceKind {
     User,
 }
 
+#[derive(Debug, PartialEq, Eq)]
 pub struct AddressSpace {
     root: PageTable,
     kind: AddressSpaceKind,

src/memory/mod.rs

@@ -1,11 +1,12 @@
 mod address_space;
 mod frame;
-pub mod stack;
+mod stack;
 
 #[allow(unused)]
 pub use address_space::{AddressSpace, AddressSpaceCreateError, MapError, UnmapError};
 pub use frame::{FRAME_SIZE, FrameAddr, FrameAllocator, OwnedFrame};
-pub use stack::{KernelStack, StackCreateError, UserStack};
+#[allow(unused)]
+pub use stack::{KernelStack, KernelStackPool, StackCreateError, UserStack};
 
 pub struct KernelSegment {
     pub physical_base: PhysicalAddr,

src/memory/stack.rs

@@ -3,10 +3,16 @@ use crate::memory::{
     VirtualAddr,
 };
 
-const STACK_PAGES: usize = 16;
+const GUARD_PAGES: usize = 1; // 4KiB
-const STACK_SIZE: usize = STACK_PAGES * FRAME_SIZE;
+const KERNEL_STACK_PAGES: usize = 8; // 32KiB
+const USER_STACK_PAGES: usize = 16; // 64KiB
+const KERNEL_STACK_SIZE: usize = KERNEL_STACK_PAGES * FRAME_SIZE;
+const USER_STACK_SIZE: usize = USER_STACK_PAGES * FRAME_SIZE;
+
+const KERNEL_SLOT_SIZE: usize = (KERNEL_STACK_PAGES + GUARD_PAGES) * FRAME_SIZE;
+const MAX_KERNEL_STACKS: usize = 64;
+const KERNEL_STACK_BASE: usize = 0xFFFF_FFFE_0000_0000;
 
-const KERNEL_STACK_TOP: VirtualAddr = VirtualAddr::new(0xFFFF_FFFE_0000_0000);
 const USER_STACK_TOP: VirtualAddr = VirtualAddr::new(0x0000_7FFF_FFFF_F000);
 
 #[derive(Debug)]
@@ -14,13 +20,16 @@ pub enum StackCreateError {
     AddressOverflow,
     UnalignedStackTop,
     OutOfFrames,
+    OutOfStacks,
     GuardPageMapped,
     Map(MapError),
 }
 
+#[derive(Debug)]
 struct StackMapping {
     guard_page: VirtualAddr,
     mapped_start: VirtualAddr,
+    stack_size: usize,
     top: VirtualAddr,
 }
 
@@ -28,6 +37,7 @@ impl StackMapping {
     fn allocate(
         address_space: &mut AddressSpace,
         allocator: &mut FrameAllocator,
+        stack_size: usize,
         top: VirtualAddr,
         permissions: PagePermissions,
     ) -> Result<Self, StackCreateError> {
@@ -37,7 +47,7 @@ impl StackMapping {
 
         let mapped_start = VirtualAddr::new(
             top.as_usize()
-                .checked_sub(STACK_SIZE)
+                .checked_sub(stack_size)
                 .ok_or(StackCreateError::AddressOverflow)?,
         );
         let guard_page = VirtualAddr::new(
@@ -53,7 +63,7 @@ impl StackMapping {
 
         let mut mapped_pages = 0;
 
-        while mapped_pages < STACK_PAGES {
+        while mapped_pages < stack_size / FRAME_SIZE {
             let virtual_address = VirtualAddr::new(
                 mapped_start
                     .as_usize()
@@ -89,6 +99,7 @@ impl StackMapping {
         Ok(Self {
             guard_page,
             mapped_start,
+            stack_size,
             top,
         })
     }
@@ -115,7 +126,7 @@ impl StackMapping {
     /// The caller must ensure this stack is not active on any CPU and cannot be accessed by any
     /// kernel operation while it is being destroyed.
     unsafe fn destroy(self, address_space: &mut AddressSpace, allocator: &mut FrameAllocator) {
-        for page in (0..STACK_PAGES).rev() {
+        for page in (0..self.stack_size / FRAME_SIZE).rev() {
             let virtual_address =
                 VirtualAddr::new(self.mapped_start.as_usize() + page * FRAME_SIZE);
             let frame = unsafe {
@@ -130,6 +141,7 @@ impl StackMapping {
     }
 }
 
+#[derive(Debug)]
 pub struct KernelStack {
     mapping: StackMapping,
 }
@@ -137,12 +149,14 @@ pub struct KernelStack {
 impl KernelStack {
     pub fn allocate(
         address_space: &mut AddressSpace,
+        top: usize,
         allocator: &mut FrameAllocator,
     ) -> Result<Self, StackCreateError> {
         let mapping = StackMapping::allocate(
             address_space,
             allocator,
-            KERNEL_STACK_TOP,
+            KERNEL_STACK_SIZE,
+            VirtualAddr::new(top),
             PagePermissions::new(true, false, false),
         )?;
 
@@ -175,6 +189,7 @@ impl UserStack {
         let mapping = StackMapping::allocate(
             address_space,
             allocator,
+            USER_STACK_SIZE,
             top,
             PagePermissions::new(true, false, true),
         )?;
@@ -194,3 +209,36 @@ impl UserStack {
         unsafe { self.mapping.destroy(address_space, allocator) };
     }
 }
+
+pub struct KernelStackPool {
+    free_slots: u64, // bitmap
+}
+
+impl KernelStackPool {
+    pub fn new() -> Self {
+        Self { free_slots: 0 }
+    }
+
+    pub fn allocate(
+        &mut self,
+        address_space: &mut AddressSpace,
+        allocator: &mut FrameAllocator,
+    ) -> Result<KernelStack, StackCreateError> {
+        let mut slot = 0;
+        while slot < MAX_KERNEL_STACKS {
+            if self.free_slots & (1 << slot) == 0 {
+                self.free_slots |= 1 << slot;
+                let top = KERNEL_STACK_BASE + ((slot + 1) * KERNEL_SLOT_SIZE);
+                return KernelStack::allocate(address_space, top, allocator);
+            }
+            slot += 1;
+        }
+
+        Err(StackCreateError::OutOfStacks)
+    }
+
+    pub fn free(&mut self, stack: &KernelStack) {
+        let slot = (stack.top().as_usize() - KERNEL_STACK_BASE) / KERNEL_SLOT_SIZE - 1;
+        self.free_slots &= !(1 << slot);
+    }
+}

src/task/mod.rs

@@ -0,0 +1,2 @@
+pub mod syscall;
+pub mod tcb;

src/task/syscall.rs

@@ -0,0 +1,18 @@
+use crate::{hcf, println};
+
+pub fn handle(
+    syscall_num: u64,
+    arg0: u64,
+    arg1: u64,
+    arg2: u64,
+    arg3: u64,
+    arg4: u64,
+    arg5: u64,
+) -> u64 {
+    println!(
+        "Syscall nr={:#X} args=({:#X}, {:#X}, {:#X}, {:#X}, {:#X}, {:#X})",
+        syscall_num, arg0, arg1, arg2, arg3, arg4, arg5
+    );
+    hcf();
+    0
+}

src/task/tcb.rs

@@ -0,0 +1,72 @@
+use crate::{
+    arch::ThreadContext,
+    memory::{AddressSpace, KernelStack, VirtualAddr},
+    println,
+};
+
+// Thread Control Block
+#[derive(Clone, Copy, Debug, PartialEq, Eq)]
+pub enum ThreadState {
+    Ready,
+    Running,
+    Blocked,
+    Dead,
+}
+
+#[derive(Debug)]
+pub struct Tcb {
+    pub id: usize,
+    pub state: ThreadState,
+    pub kernel_stack: KernelStack,
+    pub context: ThreadContext,
+    pub address_space: AddressSpace,
+}
+
+impl Tcb {
+    pub fn new_user(
+        id: usize,
+        address_space: AddressSpace,
+        kernel_stack: KernelStack,
+        user_entry: VirtualAddr,
+        user_stack: VirtualAddr,
+    ) -> Self {
+        let context = ThreadContext::new(user_entry, user_stack, kernel_stack.top());
+
+        Self {
+            id,
+            state: ThreadState::Ready,
+            kernel_stack,
+            context,
+            address_space,
+        }
+    }
+}
+
+pub unsafe fn run_first_user(user_tcb: &Tcb) {
+    let previous_interrupts = crate::arch::disable_interrupts_and_save();
+
+    let mut boot_thread_ctx = ThreadContext::empty();
+    unsafe {
+        user_tcb.address_space.activate();
+        crate::arch::set_kernel_stack(user_tcb.kernel_stack.top());
+        crate::arch::switch_context(&mut boot_thread_ctx, &user_tcb.context);
+    }
+
+    crate::arch::restore_interrupts(previous_interrupts);
+}
+
+pub unsafe fn switch(prev: &mut Tcb, next: &Tcb) {
+    let previous_interrupts = crate::arch::disable_interrupts_and_save();
+
+    if prev.address_space != next.address_space {
+        unsafe { next.address_space.activate() };
+    }
+
+    crate::arch::set_kernel_stack(next.kernel_stack.top());
+
+    unsafe {
+        crate::arch::switch_context(&mut prev.context, &next.context);
+    }
+
+    crate::arch::restore_interrupts(previous_interrupts);
+}