core: Add documentation to all scheduler functions + global thread list

This commit is contained in:
Gabriel 2025-02-21 19:37:13 +01:00
parent 8156ff57fe
commit 52cb29dbca
4 changed files with 186 additions and 91 deletions

View File

@ -1,7 +1,7 @@
const std = @import("std"); const std = @import("std");
const interrupts = @import("interrupts.zig"); const interrupts = @import("interrupts.zig");
pub inline fn enterTask(regs: *interrupts.InterruptStackFrame, base: u64, directory: u64) noreturn { pub inline fn enterThread(regs: *interrupts.InterruptStackFrame, base: u64, directory: u64) noreturn {
asm volatile ( asm volatile (
\\ addq %[base], %rsp \\ addq %[base], %rsp
\\ push %[ss] \\ push %[ss]

View File

@ -105,7 +105,7 @@ export fn _start(magic: u32, info: MultibootInfo) callconv(.C) noreturn {
platform.platformEndInit(); platform.platformEndInit();
thread.enterTask(init); thread.enterThread(init);
} }
pub fn panic(message: []const u8, _: ?*std.builtin.StackTrace, _: ?usize) noreturn { pub fn panic(message: []const u8, _: ?*std.builtin.StackTrace, _: ?usize) noreturn {

View File

@ -5,8 +5,8 @@ const cpu = @import("../arch/cpu.zig");
pub fn yield(regs: *interrupts.InterruptStackFrame, _: *sys.Arguments, _: *isize) anyerror!void { pub fn yield(regs: *interrupts.InterruptStackFrame, _: *sys.Arguments, _: *isize) anyerror!void {
const core = cpu.thisCore(); const core = cpu.thisCore();
const new_thread = thread.fetchNewTask(core, false) orelse return; const new_thread = thread.fetchNewThread(core, false) orelse return;
const current_thread = thread.scheduleNewTask(core, regs, new_thread); const current_thread = thread.scheduleNewThread(core, regs, new_thread);
thread.addThreadToPriorityQueue(core, current_thread); thread.addThreadToPriorityQueue(core, current_thread);
} }

View File

@ -4,6 +4,7 @@ const interrupts = @import("arch/interrupts.zig").arch;
pub const arch = @import("arch/thread.zig").arch; pub const arch = @import("arch/thread.zig").arch;
const pmm = @import("pmm.zig"); const pmm = @import("pmm.zig");
const cpu = @import("arch/cpu.zig"); const cpu = @import("arch/cpu.zig");
const locking = @import("lib/spinlock.zig");
pub const ThreadState = enum { pub const ThreadState = enum {
Inactive, Inactive,
@ -19,7 +20,9 @@ pub const ThreadControlBlock = struct {
state: ThreadState, state: ThreadState,
user_priority: u8, user_priority: u8,
// Managed by scheduleNewTask(), no need to set manually. // Managed by addThreadToGlobalList(), no need to set manually.
tag: GlobalThreadList.Node,
// Managed by scheduleNewThread(), no need to set manually.
ticks: u64, ticks: u64,
// Managed by addThreadToPriorityQueue(), no need to set manually. // Managed by addThreadToPriorityQueue(), no need to set manually.
current_priority: u32, current_priority: u32,
@ -29,20 +32,21 @@ pub const ThreadControlBlock = struct {
pub const ThreadList = std.DoublyLinkedList(ThreadControlBlock); pub const ThreadList = std.DoublyLinkedList(ThreadControlBlock);
const ALLOCATED_TICKS_PER_TASK = 20; const ALLOCATED_TICKS_PER_THREAD = 20;
pub fn enterTask(task: *ThreadControlBlock) noreturn { /// Starts the scheduler by running a thread. This function never returns.
cpu.thisCore().current_thread = task; pub fn enterThread(thread: *ThreadControlBlock) noreturn {
cpu.thisCore().current_thread = thread;
task.ticks = ALLOCATED_TICKS_PER_TASK; thread.ticks = ALLOCATED_TICKS_PER_THREAD;
var table = vmm.readPageTable(); var table = vmm.readPageTable();
if (task.address_space) |space| { if (thread.address_space) |space| {
table = space.phys; table = space.phys;
} }
task.state = .Running; thread.state = .Running;
// If the stack is in user memory, then we need a pointer to its higher-half version. If it's already in kernel memory, no need to do anything. // If the stack is in user memory, then we need a pointer to its higher-half version. If it's already in kernel memory, no need to do anything.
var base: usize = 0; var base: usize = 0;
@ -50,50 +54,42 @@ pub fn enterTask(task: *ThreadControlBlock) noreturn {
base += vmm.PHYSICAL_MAPPING_BASE; base += vmm.PHYSICAL_MAPPING_BASE;
} }
arch.enterTask(&task.regs, base, table.address); arch.enterThread(&thread.regs, base, table.address);
} }
fn switchTask(regs: *interrupts.InterruptStackFrame, new_task: *ThreadControlBlock) void { /// Updates the processor state to run a new thread.
fn switchThread(regs: *interrupts.InterruptStackFrame, new_thread: *ThreadControlBlock) void {
const core = cpu.thisCore(); const core = cpu.thisCore();
core.current_thread.regs = regs.*; core.current_thread.regs = regs.*;
regs.* = new_task.regs; regs.* = new_thread.regs;
if (new_task.address_space) |space| { if (new_thread.address_space) |space| {
if (vmm.readPageTable().address != space.phys.address) vmm.setPageTable(space.phys); if (vmm.readPageTable().address != space.phys.address) vmm.setPageTable(space.phys);
} }
new_task.ticks = ALLOCATED_TICKS_PER_TASK; new_thread.ticks = ALLOCATED_TICKS_PER_THREAD;
core.current_thread = new_task; core.current_thread = new_thread;
} }
pub fn fetchNewTask(core: *cpu.arch.Core, should_idle_if_not_found: bool) ?*ThreadControlBlock { /// Changes the running thread to a new one and returns the previous one.
const last = core.active_thread_list.last orelse { pub fn scheduleNewThread(core: *cpu.arch.Core, regs: *interrupts.InterruptStackFrame, new_thread: *ThreadControlBlock) *ThreadControlBlock {
if (should_idle_if_not_found) {
return &core.idle_thread.data;
} else return null;
};
const new_task = &last.data;
removeThreadFromPriorityQueue(core, new_task);
return new_task;
}
pub fn scheduleNewTask(core: *cpu.arch.Core, regs: *interrupts.InterruptStackFrame, new_thread: *ThreadControlBlock) *ThreadControlBlock {
if (core.active_thread_list.first) |first| { if (core.active_thread_list.first) |first| {
first.data.current_priority +|= 4; first.data.current_priority +|= 4;
} }
const current_thread = core.current_thread; const current_thread = core.current_thread;
switchTask(regs, new_thread); switchThread(regs, new_thread);
return current_thread; return current_thread;
} }
/// Called on every timer interrupt.
///
/// Updates the core's sleep queue, checks if the running thread's time
/// is up, and if it is, schedules a new one.
pub fn preempt(regs: *interrupts.InterruptStackFrame) void { pub fn preempt(regs: *interrupts.InterruptStackFrame) void {
const core = cpu.thisCore(); const core = cpu.thisCore();
@ -104,87 +100,41 @@ pub fn preempt(regs: *interrupts.InterruptStackFrame) void {
core.current_thread.ticks -|= 1; core.current_thread.ticks -|= 1;
if (core.current_thread.ticks == 0) { if (core.current_thread.ticks == 0) {
const new_thread = fetchNewTask(core, false) orelse return; const new_thread = fetchNewThread(core, false) orelse return;
const current_thread = scheduleNewTask(core, regs, new_thread); const current_thread = scheduleNewThread(core, regs, new_thread);
addThreadToPriorityQueue(core, current_thread); addThreadToPriorityQueue(core, current_thread);
} }
} }
/// Sets the current thread's state to "Blocked" and schedules a new one to replace it.
pub fn block(regs: *interrupts.InterruptStackFrame) *ThreadControlBlock { pub fn block(regs: *interrupts.InterruptStackFrame) *ThreadControlBlock {
const core = cpu.thisCore(); const core = cpu.thisCore();
// fetchNewTask() always returns a thread if should_idle_if_not_found is set to true. // fetchNewThread() always returns a thread if should_idle_if_not_found is set to true.
const new_thread = fetchNewTask(core, true) orelse unreachable; const new_thread = fetchNewThread(core, true) orelse unreachable;
const current_thread = scheduleNewTask(core, regs, new_thread); const current_thread = scheduleNewThread(core, regs, new_thread);
current_thread.state = .Blocked; current_thread.state = .Blocked;
return current_thread; return current_thread;
} }
/// Puts the current thread to sleep, adding it to the sleep queue, and schedules a new one to replace it.
pub fn startSleep(regs: *interrupts.InterruptStackFrame, ticks: u64) *ThreadControlBlock { pub fn startSleep(regs: *interrupts.InterruptStackFrame, ticks: u64) *ThreadControlBlock {
const core = cpu.thisCore(); const core = cpu.thisCore();
// fetchNewTask() always returns a thread if should_idle_if_not_found is set to true. // fetchNewThread() always returns a thread if should_idle_if_not_found is set to true.
const new_thread = fetchNewTask(core, true) orelse unreachable; const new_thread = fetchNewThread(core, true) orelse unreachable;
const current_thread = scheduleNewTask(core, regs, new_thread); const current_thread = scheduleNewThread(core, regs, new_thread);
current_thread.state = .Sleeping; current_thread.state = .Sleeping;
addThreadToSleepQueue(core, current_thread, ticks); addThreadToSleepQueue(core, current_thread, ticks);
return current_thread; return current_thread;
} }
fn addThreadToSleepQueue(core: *cpu.arch.Core, thread: *ThreadControlBlock, ticks: u64) void {
thread.sleep_ticks = ticks;
var it: ?*ThreadList.Node = core.sleeping_thread_list.first;
while (it) |n| : (it = n.next) {
if (thread.sleep_ticks <= n.data.sleep_ticks) {
n.data.sleep_ticks -|= thread.sleep_ticks;
core.sleeping_thread_list.insertBefore(n, @fieldParentPtr("data", thread));
return;
}
thread.sleep_ticks -|= n.data.sleep_ticks;
}
core.sleeping_thread_list.append(@fieldParentPtr("data", thread));
}
pub fn removeThreadFromSleepQueue(core: *cpu.arch.Core, thread: *ThreadControlBlock) void {
const node: *ThreadList.Node = @fieldParentPtr("data", thread);
if (node.next) |n| {
n.data.sleep_ticks +|= thread.sleep_ticks;
}
core.sleeping_thread_list.remove(node);
reviveThread(core, thread);
}
fn updateSleepQueue(core: *cpu.arch.Core) void {
const first = core.sleeping_thread_list.first orelse return;
first.data.sleep_ticks -|= 1;
}
fn popSleepQueue(core: *cpu.arch.Core) ?*ThreadControlBlock {
const first = core.sleeping_thread_list.first orelse return null;
if (first.data.sleep_ticks == 0) {
core.sleeping_thread_list.remove(first);
return &first.data;
}
return null;
}
pub fn reviveThread(core: *cpu.arch.Core, thread: *ThreadControlBlock) void {
thread.state = .Running;
addThreadToPriorityQueue(core, thread);
}
var next_id: std.atomic.Value(u64) = std.atomic.Value(u64).init(1); var next_id: std.atomic.Value(u64) = std.atomic.Value(u64).init(1);
/// Allocates a physical frame and create a new thread control block inside it, adding
/// it to the global thread list.
pub fn createThreadControlBlock(allocator: *pmm.FrameAllocator) !*ThreadControlBlock { pub fn createThreadControlBlock(allocator: *pmm.FrameAllocator) !*ThreadControlBlock {
const frame = try pmm.allocFrame(allocator); const frame = try pmm.allocFrame(allocator);
@ -196,9 +146,21 @@ pub fn createThreadControlBlock(allocator: *pmm.FrameAllocator) !*ThreadControlB
thread.state = .Inactive; thread.state = .Inactive;
thread.user_priority = 127; thread.user_priority = 127;
addThreadToGlobalList(thread);
return thread; return thread;
} }
// The "priority queue" is the main scheduling queue for each core. In the code, it is referred to as "core.active_thread_list".
// In the priority queue, we store all threads currently waiting to run, sorted by priority.
// Threads are added to the list with an initial priority equal to their own "user_priority", but every time a new thread
// is scheduled, every other thread waiting to run that didn't get to run that time gets their priority incremented by 4.
//
// In the priority queue, threads are not stored with their absolute priority value, but their relative priority instead (that is,
// how much more priority they have than the previous thread). That way, all threads' priorities can be incremented by adding a number
// to the lowest priority thread.
/// Adds a thread to the specified core's priority queue.
pub fn addThreadToPriorityQueue(core: *cpu.arch.Core, thread: *ThreadControlBlock) void { pub fn addThreadToPriorityQueue(core: *cpu.arch.Core, thread: *ThreadControlBlock) void {
thread.current_priority = thread.user_priority; thread.current_priority = thread.user_priority;
@ -215,6 +177,10 @@ pub fn addThreadToPriorityQueue(core: *cpu.arch.Core, thread: *ThreadControlBloc
core.active_thread_list.append(@fieldParentPtr("data", thread)); core.active_thread_list.append(@fieldParentPtr("data", thread));
} }
/// Removes a thread from the specified core's priority queue.
///
/// This function is private because threads are automatically removed
/// when scheduled, and that's the only instance when they should be removed.
fn removeThreadFromPriorityQueue(core: *cpu.arch.Core, thread: *ThreadControlBlock) void { fn removeThreadFromPriorityQueue(core: *cpu.arch.Core, thread: *ThreadControlBlock) void {
const node: *ThreadList.Node = @fieldParentPtr("data", thread); const node: *ThreadList.Node = @fieldParentPtr("data", thread);
@ -224,3 +190,132 @@ fn removeThreadFromPriorityQueue(core: *cpu.arch.Core, thread: *ThreadControlBlo
core.active_thread_list.remove(node); core.active_thread_list.remove(node);
} }
/// Finds the thread with the highest priority, removes it from the specified core's
/// priority queue, and returns it so that it can be scheduled.
///
/// This function's behaviour when there are no threads waiting to be run depends on the value
/// passed to "should_idle_if_not_found". If this value is true (the currently running thread cannot continue to run,
/// because for example it wants to block), we return the core's idle thread.
/// Otherwise, we return null, signalling that the thread currently running should continue to run
/// until a new thread is available.
pub fn fetchNewThread(core: *cpu.arch.Core, should_idle_if_not_found: bool) ?*ThreadControlBlock {
const last = core.active_thread_list.last orelse {
if (should_idle_if_not_found) {
return &core.idle_thread.data;
} else return null;
};
const new_thread = &last.data;
removeThreadFromPriorityQueue(core, new_thread);
return new_thread;
}
/// Adds a previously blocked or sleeping thread back to the specified core's priority queue.
pub fn reviveThread(core: *cpu.arch.Core, thread: *ThreadControlBlock) void {
thread.state = .Running;
addThreadToPriorityQueue(core, thread);
}
// The "sleep queue" is the secondary scheduling queue for each core. In the code, it is referred to as "core.sleeping_thread_list".
// In the sleep queue, we store all threads currently sleeping on this core, sorted by the time remaining until they wake.
//
// In the sleep queue, threads are not stored with the absolute time remaining, but the relative time remaining instead (that is,
// how much more time they have left to sleep than the previous thread). That way, all threads' time remaining values can be decremented
// by one by doing this to the first thread in the list (see updateSleepQueue()).
/// Adds a new thread to the specified core's sleep queue, with the specified ticks left to wake.
///
/// This function is private because you should not call it directly, use startSleep() instead, which
/// also schedules a new thread to replace the one going to sleep.
fn addThreadToSleepQueue(core: *cpu.arch.Core, thread: *ThreadControlBlock, ticks: u64) void {
thread.sleep_ticks = ticks;
var it: ?*ThreadList.Node = core.sleeping_thread_list.first;
while (it) |n| : (it = n.next) {
if (thread.sleep_ticks <= n.data.sleep_ticks) {
n.data.sleep_ticks -|= thread.sleep_ticks;
core.sleeping_thread_list.insertBefore(n, @fieldParentPtr("data", thread));
return;
}
thread.sleep_ticks -|= n.data.sleep_ticks;
}
core.sleeping_thread_list.append(@fieldParentPtr("data", thread));
}
/// Removes a new thread from the specified core's sleep queue and adds it back to the priority queue.
pub fn removeThreadFromSleepQueue(core: *cpu.arch.Core, thread: *ThreadControlBlock) void {
const node: *ThreadList.Node = @fieldParentPtr("data", thread);
if (node.next) |n| {
n.data.sleep_ticks +|= thread.sleep_ticks;
}
core.sleeping_thread_list.remove(node);
reviveThread(core, thread);
}
/// Decrements the time left for all threads in a core's sleep queue to wake by one.
/// As mentioned earlier, since the time left is stored relative to the other threads,
/// decrementing the time left for the first thread is enough to change it for all threads.
fn updateSleepQueue(core: *cpu.arch.Core) void {
const first = core.sleeping_thread_list.first orelse return;
first.data.sleep_ticks -|= 1;
}
/// If a thread in the sleep queue has finished its sleeping time, removes it from the queue and returns it.
/// Otherwise, returns null.
///
/// This function should be called in a loop until it returns null,
/// since multiple threads could wake up at the same time.
///
/// This function also does not revive the threads; you must call reviveThread() yourself.
fn popSleepQueue(core: *cpu.arch.Core) ?*ThreadControlBlock {
const first = core.sleeping_thread_list.first orelse return null;
if (first.data.sleep_ticks == 0) {
core.sleeping_thread_list.remove(first);
return &first.data;
}
return null;
}
// The global thread list is only used to keep track of threads so they can be looked up when needed.
// It does not perform any scheduling functions, unlike the core-specific lists.
// While threads can move in and out of scheduling-related thread lists at will, and can even remain outside of them (example: blocked threads),
// threads should be added to this list on creation and never removed until destruction, with the exception of core-specific idle threads, which for now
// will not be added to this list, as they all have the same ID and there is little a user process can do with them.
// Since this list is shared by all cores, locking is needed.
pub const GlobalThreadList = std.DoublyLinkedList(u8);
var global_thread_list_lock: locking.SpinLock = .{};
var global_thread_list: GlobalThreadList = .{};
/// Adds a newly created thread to the global list.
pub fn addThreadToGlobalList(thread: *ThreadControlBlock) void {
global_thread_list_lock.lock();
defer global_thread_list_lock.unlock();
global_thread_list.append(&thread.tag);
}
/// Finds the thread with a matching thread ID by iterating through the global list.
pub fn lookupThreadById(id: u64) ?*ThreadControlBlock {
global_thread_list_lock.lock();
defer global_thread_list_lock.unlock();
var it: ?*GlobalThreadList.Node = global_thread_list.first;
while (it) |n| : (it = n.next) {
const thread: *ThreadControlBlock = @fieldParentPtr("tag", n);
if (thread.id == id) return thread;
}
return null;
}