From ca2362f858527c2d08b108cb629f9bc13ad35d8a Mon Sep 17 00:00:00 2001
From: "Justin C. Miller" <justin@devjustinian.com>
Date: Sun, 31 Mar 2019 22:49:24 -0700
Subject: [PATCH] Simplify task switches

No longer using the rsp from the entry to the kernel, but instead
switching rsp at task-switching time in assembly.

This currently breaks fork()
---
 modules.yaml                 |   1 +
 src/drivers/nulldrv/main.cpp |   2 +
 src/drivers/nulldrv/main.s   |  12 ++++
 src/kernel/cpu.h             |   5 ++
 src/kernel/debug.cpp         |   7 +-
 src/kernel/debug.h           |   1 +
 src/kernel/interrupts.cpp    |  45 +++++--------
 src/kernel/interrupts.s      |   2 -
 src/kernel/loader.s          |  11 ++--
 src/kernel/log.cpp           |  13 ++--
 src/kernel/log_areas.inc     |   8 +--
 src/kernel/page_manager.cpp  |   5 +-
 src/kernel/process.cpp       |  50 +++++++++------
 src/kernel/process.h         |  31 +++++----
 src/kernel/scheduler.cpp     | 121 ++++++++++++++++-------------------
 src/kernel/scheduler.h       |  14 ++--
 src/kernel/syscall.cpp       |  41 ++++++------
 src/kernel/syscall.h         |   2 +-
 src/kernel/syscall.s         |   1 -
 src/kernel/task.s            |  85 ++++++++++++++++++++++++
 src/kernel/tasking.inc       |  32 +++++++++
 21 files changed, 311 insertions(+), 178 deletions(-)
 create mode 100644 src/kernel/task.s
 create mode 100644 src/kernel/tasking.inc

diff --git a/modules.yaml b/modules.yaml
index cfbc445..cdae6ac 100644
--- a/modules.yaml
+++ b/modules.yaml
@@ -41,6 +41,7 @@ modules:
             - src/kernel/serial.cpp
             - src/kernel/syscall.cpp
             - src/kernel/syscall.s
+            - src/kernel/task.s
             - src/kernel/crtn.s
 
     boot:
diff --git a/src/drivers/nulldrv/main.cpp b/src/drivers/nulldrv/main.cpp
index 54100d6..4740524 100644
--- a/src/drivers/nulldrv/main.cpp
+++ b/src/drivers/nulldrv/main.cpp
@@ -3,6 +3,7 @@
 
 extern "C" {
 	int32_t getpid();
+	int32_t fork();
 	void sleep(uint64_t til);
 	void debug();
 
@@ -14,6 +15,7 @@ int
 main(int argc, const char **argv)
 {
 	int32_t pid = getpid();
+	//int32_t child = fork();
 	debug();
 	for (int i = 1; i < 5; ++i)
 		sleep(i*10);
diff --git a/src/drivers/nulldrv/main.s b/src/drivers/nulldrv/main.s
index 37785c8..03e01ea 100644
--- a/src/drivers/nulldrv/main.s
+++ b/src/drivers/nulldrv/main.s
@@ -39,6 +39,18 @@ sleep:
 	pop rbp
 	ret
 
+global fork
+fork:
+	push rbp
+	mov rbp, rsp
+
+	mov rax, 8
+	syscall    ; pid left in rax
+
+	pop rbp
+	ret
+	
+
 global _start
 _start:
 	xor rbp, rbp	; Sentinel rbp
diff --git a/src/kernel/cpu.h b/src/kernel/cpu.h
index 9c1d4a0..dd48549 100644
--- a/src/kernel/cpu.h
+++ b/src/kernel/cpu.h
@@ -2,6 +2,8 @@
 
 #include <stdint.h>
 
+struct process;
+
 struct cpu_state
 {
 	uint64_t r15, r14, r13, r12, r11, r10, r9,  r8;
@@ -10,10 +12,13 @@ struct cpu_state
 	uint64_t rip, cs, rflags, user_rsp, ss;
 };
 
+/// Per-cpu state data. If you change this, remember to update the assembly
+/// version in 'tasking.inc'
 struct cpu_data
 {
 	uintptr_t rsp0;
 	uintptr_t rsp3;
+	process *tcb;
 };
 
 extern cpu_data bsp_cpu_data;
diff --git a/src/kernel/debug.cpp b/src/kernel/debug.cpp
index 29fd217..1c3ba25 100644
--- a/src/kernel/debug.cpp
+++ b/src/kernel/debug.cpp
@@ -12,6 +12,9 @@ print_regs(const cpu_state &regs)
 {
 	console *cons = console::get();
 
+	uint64_t cr2 = 0;
+	__asm__ __volatile__ ("mov %%cr2, %0" : "=r"(cr2));
+
 	print_regL("rax", regs.rax);
 	print_regM("rbx", regs.rbx);
 	print_regR("rcx", regs.rcx);
@@ -36,7 +39,9 @@ print_regs(const cpu_state &regs)
 
 	print_regL("rip", regs.rip);
 	print_regM("cr3", page_manager::get()->get_pml4());
-	cons->puts("\n\n");
+	print_regR("cr2", cr2);
+
+	cons->puts("\n");
 }
 
 struct frame
diff --git a/src/kernel/debug.h b/src/kernel/debug.h
index 9484b82..de08ffe 100644
--- a/src/kernel/debug.h
+++ b/src/kernel/debug.h
@@ -9,6 +9,7 @@ extern "C" {
 	uintptr_t get_rip();
 	uintptr_t get_frame(int frame);
 	uintptr_t get_gsbase();
+	void _halt();
 }
 
 extern size_t __counter_syscall_enter;
diff --git a/src/kernel/interrupts.cpp b/src/kernel/interrupts.cpp
index 2ffafcc..5db7a03 100644
--- a/src/kernel/interrupts.cpp
+++ b/src/kernel/interrupts.cpp
@@ -18,9 +18,9 @@ static const uint16_t PIC2 = 0xa0;
 extern "C" {
 	void _halt();
 
-	uintptr_t isr_handler(uintptr_t, cpu_state*);
-	uintptr_t irq_handler(uintptr_t, cpu_state*);
-	uintptr_t syscall_handler(uintptr_t, cpu_state);
+	void isr_handler(cpu_state*);
+	void irq_handler(cpu_state*);
+	void syscall_handler(cpu_state*);
 
 #define ISR(i, name)     extern void name ();
 #define EISR(i, name)    extern void name ();
@@ -104,8 +104,8 @@ interrupts_init()
 	log::info(logs::boot, "Interrupts enabled.");
 }
 
-uintptr_t
-isr_handler(uintptr_t return_rsp, cpu_state *regs)
+void
+isr_handler(cpu_state *regs)
 {
 	console *cons = console::get();
 
@@ -193,21 +193,14 @@ isr_handler(uintptr_t return_rsp, cpu_state *regs)
 			if (regs->errorcode & 0x08) cons->puts(" reserved");
 			if (regs->errorcode & 0x10) cons->puts(" ip");
 			cons->puts("\n");
-
-			uint64_t cr2 = 0;
-			__asm__ __volatile__ ("mov %%cr2, %0" : "=r"(cr2));
-			print_regL("cr2", cr2);
-			print_regM("rsp", regs->user_rsp);
-			print_regR("rip", regs->rip);
-			//print_stacktrace(2);
+			print_regs(*regs);
+			print_stacktrace(2);
 		}
 		_halt();
 		break;
 
-	case isr::isrTimer: {
-			scheduler &s = scheduler::get();
-			return_rsp = s.tick(return_rsp);
-		}
+	case isr::isrTimer:
+		scheduler::get().tick();
 		break;
 
 	case isr::isrLINT0:
@@ -226,14 +219,13 @@ isr_handler(uintptr_t return_rsp, cpu_state *regs)
 		_halt();
 		break;
 
-	case isr::isrSyscall: {
-			return_rsp = syscall_dispatch(return_rsp, *regs);
-		}
+	case isr::isrSyscall:
+		syscall_dispatch(regs);
 		break;
 
 	case isr::isrSpurious:
 		// No EOI for the spurious interrupt
-		return return_rsp;
+		return;
 
 	case isr::isrIgnore0:
 	case isr::isrIgnore1:
@@ -274,12 +266,10 @@ isr_handler(uintptr_t return_rsp, cpu_state *regs)
 		_halt();
 	}
 	*reinterpret_cast<uint32_t *>(0xffffff80fee000b0) = 0;
-
-	return return_rsp;
 }
 
-uintptr_t
-irq_handler(uintptr_t return_rsp, cpu_state *regs)
+void
+irq_handler(cpu_state *regs)
 {
 	console *cons = console::get();
 	uint8_t irq = get_irq(regs->interrupt);
@@ -293,11 +283,10 @@ irq_handler(uintptr_t return_rsp, cpu_state *regs)
 	}
 
 	*reinterpret_cast<uint32_t *>(0xffffff80fee000b0) = 0;
-	return return_rsp;
 }
 
-uintptr_t
-syscall_handler(uintptr_t return_rsp, cpu_state regs)
+void
+syscall_handler(cpu_state *regs)
 {
-	return syscall_dispatch(return_rsp, regs);
+	syscall_dispatch(regs);
 }
diff --git a/src/kernel/interrupts.s b/src/kernel/interrupts.s
index 8bbcba0..cbb9e7b 100644
--- a/src/kernel/interrupts.s
+++ b/src/kernel/interrupts.s
@@ -9,7 +9,6 @@ isr_handler_prelude:
 	mov rdi, rsp
 	mov rsi, rsp
 	call isr_handler
-	mov rsp, rax
 	jmp isr_handler_return
 
 extern irq_handler
@@ -21,7 +20,6 @@ irq_handler_prelude:
 	mov rdi, rsp
 	mov rsi, rsp
 	call irq_handler
-	mov rsp, rax
 	; fall through to isr_handler_return
 
 global isr_handler_return
diff --git a/src/kernel/loader.s b/src/kernel/loader.s
index e9310eb..5e9031f 100644
--- a/src/kernel/loader.s
+++ b/src/kernel/loader.s
@@ -7,12 +7,11 @@ ramdisk_process_loader:
 	; create_process already pushed a cpu_state onto the stack for us, this
 	; acts both as the cpu_state parameter to load_process, and the saved
 	; state for the following iretq
-	;
-	; Additional parameters:
-	;  rdi - the address of the program image
-	;  rsi - the size of the program image
-	;  rdx - the address of this process' process structure
-	;  rcx - the stack pointer, which points at the cpu_state
+
+	pop rdi ; the address of the program image
+	pop rsi ; the size of the program image
+	pop rdx ; the address of this process' process structure
+	pop rcx ; the cpu_state
 	call load_process
 
 	swapgs
diff --git a/src/kernel/log.cpp b/src/kernel/log.cpp
index 666d30c..7f6543c 100644
--- a/src/kernel/log.cpp
+++ b/src/kernel/log.cpp
@@ -2,6 +2,7 @@
 #include "kutil/memory.h"
 #include "console.h"
 #include "log.h"
+#include "scheduler.h"
 
 namespace logs {
 
@@ -21,7 +22,7 @@ output_log(log::area_t area, log::level severity, const char *message)
 {
 	auto *cons = console::get();
 	cons->set_color(level_colors[static_cast<int>(severity)]);
-	cons->printf("%9s %8s: %s\n",
+	cons->printf("%7s %5s: %s\n",
 			g_logger.area_name(area),
 			g_logger.level_name(severity),
 			message);
@@ -35,21 +36,23 @@ logger_task()
 	auto *ent = reinterpret_cast<log::logger::entry *>(buffer);
 	auto *cons = console::get();
 
-	g_logger.set_immediate(nullptr);
+	//g_logger.set_immediate(nullptr);
 	log::info(logs::task, "Starting kernel logger task");
 
+	scheduler &s = scheduler::get();
+
 	while (true) {
-		__asm__ ("cli");
 		if(g_logger.get_entry(buffer, sizeof(buffer))) {
 			buffer[ent->bytes] = 0;
 			cons->set_color(level_colors[static_cast<int>(ent->severity)]);
-			cons->printf("%9s %8s: %s\n",
+			cons->printf("%7s %5s: %s\n",
 				g_logger.area_name(ent->area),
 				g_logger.level_name(ent->severity),
 				ent->message);
 			cons->set_color();
+		} else {
+			s.schedule();
 		}
-		__asm__ ("sti");
 	}
 }
 
diff --git a/src/kernel/log_areas.inc b/src/kernel/log_areas.inc
index ec3b215..dd49215 100644
--- a/src/kernel/log_areas.inc
+++ b/src/kernel/log_areas.inc
@@ -1,8 +1,8 @@
 LOG(apic,   info);
-LOG(device, debug);
-LOG(paging, info);
-LOG(driver, debug);
+LOG(device, info);
+LOG(paging, debug);
+LOG(driver, info);
 LOG(memory, debug);
-LOG(fs,     debug);
+LOG(fs,     info);
 LOG(task,   debug);
 LOG(boot,   debug);
diff --git a/src/kernel/page_manager.cpp b/src/kernel/page_manager.cpp
index c8e8786..6701c81 100644
--- a/src/kernel/page_manager.cpp
+++ b/src/kernel/page_manager.cpp
@@ -150,7 +150,7 @@ page_manager::delete_process_map(page_table *pml4)
 void
 page_manager::map_offset_pointer(void **pointer, size_t length)
 {
-	log::info(logs::paging, "Mapping offset pointer region at %016lx size 0x%lx", *pointer, length);
+	log::debug(logs::paging, "Mapping offset pointer region at %016lx size 0x%lx", *pointer, length);
 	*pointer = kutil::offset_pointer(*pointer, page_offset);
 }
 
@@ -185,6 +185,7 @@ page_manager::get_table_page()
 
 	free_page_header *page = m_page_cache;
 	m_page_cache = page->next;
+
 	return reinterpret_cast<page_table *>(page);
 }
 
@@ -294,8 +295,10 @@ page_manager::check_needs_page(page_table *table, unsigned index, bool user)
 void
 page_manager::page_in(page_table *pml4, uintptr_t phys_addr, uintptr_t virt_addr, size_t count, bool user, bool large)
 {
+	/*
 	log::debug(logs::paging, "page_in for table %016lx p:%016lx v:%016lx c:%4d u:%d l:%d",
 			pml4, phys_addr, virt_addr, count, user, large);
+	*/
 
 	page_table_indices idx{virt_addr};
 	page_table *tables[4] = {pml4, nullptr, nullptr, nullptr};
diff --git a/src/kernel/process.cpp b/src/kernel/process.cpp
index cc91a38..5db1981 100644
--- a/src/kernel/process.cpp
+++ b/src/kernel/process.cpp
@@ -1,4 +1,5 @@
 #include "cpu.h"
+#include "debug.h"
 #include "log.h"
 #include "process.h"
 #include "scheduler.h"
@@ -12,7 +13,7 @@ process::exit(uint32_t code)
 }
 
 pid_t
-process::fork(uintptr_t in_rsp)
+process::fork(cpu_state *regs)
 {
 	auto &sched = scheduler::get();
 	auto *child = sched.create_process();
@@ -25,40 +26,49 @@ process::fork(uintptr_t in_rsp)
 
 	sched.m_runlists[child->priority].push_back(child);
 
-	child->rsp = in_rsp;
-
 	child->pml4 = page_manager::get()->copy_table(pml4);
 	kassert(child->pml4, "process::fork() got null pml4");
 
-	child->setup_kernel_stack(kernel_stack_size, kernel_stack);
-	child->rsp = child->kernel_stack + (in_rsp - kernel_stack);
-
 	log::debug(logs::task, "Copied process %d to %d, new PML4 %016lx.",
 			pid, child->pid, child->pml4);
-	log::debug(logs::task, "  copied stack %016lx to %016lx, rsp %016lx to %016lx.",
-			kernel_stack, child->kernel_stack, in_rsp, child->rsp);
+	log::debug(logs::task, "  copied stack %016lx to %016lx, rsp %016lx.",
+			kernel_stack, child->kernel_stack, child->rsp);
 
-	// Add in the faked fork return value
-	cpu_state *regs = reinterpret_cast<cpu_state *>(child->rsp);
-	regs->rax = 0;
+	child->setup_kernel_stack();
+	task_fork(child); // Both parent and child will return from this
+
+	if (bsp_cpu_data.tcb->pid == child->pid) {
+		return 0;
+	}
 
 	return child->pid;
 }
 
 void *
-process::setup_kernel_stack(size_t size, uintptr_t orig)
+process::setup_kernel_stack()
 {
-	void *stack0 = kutil::malloc(size);
+	constexpr unsigned null_frame_entries = 2;
+	constexpr size_t null_frame_size = null_frame_entries * sizeof(uint64_t);
 
-	if (orig)
-		kutil::memcpy(stack0, reinterpret_cast<void*>(orig), size);
-	else
-		kutil::memset(stack0, 0, size);
+	void *stack_bottom = kutil::malloc(initial_stack_size);
+	kutil::memset(stack_bottom, 0, initial_stack_size);
 
-	kernel_stack_size = size;
-	kernel_stack = reinterpret_cast<uintptr_t>(stack0);
+	log::debug(logs::memory, "Created kernel stack at %016lx size 0x%lx",
+			stack_bottom, initial_stack_size);
 
-	return stack0;
+	void *stack_top =
+		kutil::offset_pointer(stack_bottom,
+				initial_stack_size - null_frame_size);
+
+	uint64_t *null_frame = reinterpret_cast<uint64_t*>(stack_top);
+	for (unsigned i = 0; i < null_frame_entries; ++i)
+		null_frame[i] = 0;
+
+	kernel_stack_size = initial_stack_size;
+	kernel_stack = reinterpret_cast<uintptr_t>(stack_bottom);
+	rsp0 = reinterpret_cast<uintptr_t>(stack_top);
+
+	return stack_top;
 }
 
 bool
diff --git a/src/kernel/process.h b/src/kernel/process.h
index 6911da2..7c5e1ef 100644
--- a/src/kernel/process.h
+++ b/src/kernel/process.h
@@ -8,6 +8,7 @@
 #include "page_manager.h"
 
 typedef int32_t pid_t;
+struct cpu_state;
 
 
 enum class process_flags : uint32_t
@@ -32,9 +33,19 @@ enum class process_wait : uint8_t
 	receive
 };
 
-/// A process
+/// A process.
+/// 
 struct process
 {
+	static const size_t initial_stack_size = 0x1000;
+
+	// Fields used by assembly routines go first.  If you change any of these,
+	// be sure to change the assembly definitions in 'tasking.inc'
+	uintptr_t rsp;
+	uintptr_t rsp0;
+	page_table *pml4;
+	// End of assembly fields
+
 	pid_t pid;
 	pid_t ppid;
 
@@ -51,9 +62,6 @@ struct process
 
 	uint32_t reserved1;
 
-	uintptr_t rsp;
-	page_table *pml4;
-
 	uintptr_t kernel_stack;
 	size_t kernel_stack_size;
 
@@ -62,10 +70,10 @@ struct process
 	void exit(unsigned code);
 
 	/// Copy this process.
-	/// \arg in_rsp The RSP of the calling process
+	/// \arg regs   The saved state from the fork syscall
 	/// \returns    Returns the child's pid to the parent, and
 	///             0 to the child.
-	pid_t fork(uint64_t in_rsp);
+	pid_t fork(cpu_state *regs);
 
 	/// Unready this process until it gets a signal
 	/// \arg sigmask  A bitfield of signals to wake on
@@ -122,13 +130,10 @@ struct process
 private:
 	friend class scheduler;
 
-	/// Set up a new kernel stack for this process, optionally copying the
-	/// given stack. Sets the kernel stack on the process object, but also
-	/// returns it.
-	/// \arg size  Size of the stack to allocate
-	/// \arg orig  Address of a stack to copy, or 0 for no copying.
-	/// \returns   The address of the new stack as a pointer
-	void * setup_kernel_stack(size_t size, uintptr_t orig);
+	/// Set up a new empty kernel stack for this process. Sets rsp0 on this
+	/// process object, but also returns it.
+	/// \returns   The new rsp0 as a pointer
+	void * setup_kernel_stack();
 };
 
 using process_list = kutil::linked_list<process>;
diff --git a/src/kernel/scheduler.cpp b/src/kernel/scheduler.cpp
index a494984..3341b60 100644
--- a/src/kernel/scheduler.cpp
+++ b/src/kernel/scheduler.cpp
@@ -18,7 +18,6 @@ using memory::initial_stack;
 
 scheduler scheduler::s_instance(nullptr);
 
-const int stack_size = 0x1000;
 const uint64_t rflags_noint = 0x002;
 const uint64_t rflags_int = 0x202;
 
@@ -32,6 +31,7 @@ scheduler::scheduler(lapic *apic) :
 	m_next_pid(1)
 {
 	auto *idle = m_process_allocator.pop();
+	idle->setup_kernel_stack();
 
 	uint8_t last_pri = num_priorities - 1;
 
@@ -49,6 +49,9 @@ scheduler::scheduler(lapic *apic) :
 
 	m_runlists[last_pri].push_back(idle);
 	m_current = idle;
+
+	bsp_cpu_data.rsp0 = idle->rsp0;
+	bsp_cpu_data.tcb = idle;
 }
 
 void
@@ -120,6 +123,24 @@ scheduler::create_process(pid_t pid)
 	return proc;
 }
 
+static uintptr_t
+add_fake_stack_return(uintptr_t rsp, uintptr_t rbp, uintptr_t rip)
+{
+	// Initialize a new empty stack with a fake return segment
+	// for returning out of task_switch
+	rsp -= sizeof(uintptr_t) * 7;
+	uintptr_t *stack = reinterpret_cast<uintptr_t*>(rsp);
+
+	stack[0] = rbp;        // rbp
+	stack[1] = 0xbbbbbbbb; // rbx
+	stack[2] = 0x12121212; // r12
+	stack[3] = 0x13131313; // r13
+	stack[4] = 0x14141414; // r14
+	stack[5] = 0x15151515; // r15
+	stack[6] = rip;        // return rip
+	return rsp;
+}
+
 void
 scheduler::load_process(const char *name, const void *data, size_t size)
 {
@@ -132,15 +153,10 @@ scheduler::load_process(const char *name, const void *data, size_t size)
 	uint16_t ss = (4 << 3) | 3;  // User SS is GDT entry 4, ring 3
 
 	// Set up the page tables - this also allocates an initial user stack
-	page_table *pml4 = page_manager::get()->create_process_map();
-
-	// Create a one-page kernel stack space
-	void *stack0 = proc->setup_kernel_stack(stack_size, 0);
-
-	// Stack grows down, point to the end, resere space for initial null frame
-	static const size_t null_frame = sizeof(uint64_t);
-	void *sp0 = kutil::offset_pointer(stack0, stack_size - null_frame);
+	proc->pml4 = page_manager::get()->create_process_map();
 
+	// Create an initial kernel stack space
+	void *sp0 = proc->setup_kernel_stack();
 	cpu_state *state = reinterpret_cast<cpu_state *>(sp0) - 1;
 
 	// Highest state in the stack is the process' kernel stack for the loader
@@ -151,24 +167,18 @@ scheduler::load_process(const char *name, const void *data, size_t size)
 	state->rip = 0; // to be filled by the loader
 	state->user_rsp = initial_stack;
 
-	// Next state in the stack is the loader's kernel stack. The scheduler will
-	// iret to this which will kick off the loading:
-	cpu_state *loader_state = reinterpret_cast<cpu_state *>(sp0) - 2;
+	// Pass args to ramdisk_process_loader on the stack
+	uintptr_t *stack = reinterpret_cast<uintptr_t *>(state) - 4;
+	stack[0] = reinterpret_cast<uintptr_t>(data);
+	stack[1] = reinterpret_cast<uintptr_t>(size);
+	stack[2] = reinterpret_cast<uintptr_t>(proc);
+	stack[3] = reinterpret_cast<uintptr_t>(state);
 
-	loader_state->ss = kss;
-	loader_state->cs = kcs;
-	loader_state->rflags = rflags_noint;
-	loader_state->rip = reinterpret_cast<uint64_t>(ramdisk_process_loader);
-	loader_state->user_rsp = reinterpret_cast<uint64_t>(state);
+	proc->rsp = add_fake_stack_return(
+			reinterpret_cast<uintptr_t>(stack),
+			proc->rsp0,
+			reinterpret_cast<uintptr_t>(ramdisk_process_loader));
 
-	// Set up the registers to have the arguments to the load_process call
-	loader_state->rdi = reinterpret_cast<uint64_t>(data); // arg 1
-	loader_state->rsi = size; // arg 2
-	loader_state->rdx = reinterpret_cast<uint64_t>(proc); // arg 3
-	loader_state->rcx = loader_state->user_rsp; // arg 4
-
-	proc->rsp = reinterpret_cast<uintptr_t>(loader_state);
-	proc->pml4 = pml4;
 	proc->quanta = process_quanta;
 	proc->flags =
 		process_flags::running |
@@ -177,10 +187,9 @@ scheduler::load_process(const char *name, const void *data, size_t size)
 
 	m_runlists[default_priority].push_back(proc);
 
-
 	log::debug(logs::task, "Creating process %s: pid %d  pri %d", name, proc->pid, proc->priority);
 	log::debug(logs::task, "     RSP0 %016lx", state);
-	log::debug(logs::task, "     PML4 %016lx", pml4);
+	log::debug(logs::task, "     PML4 %016lx", proc->pml4);
 }
 
 void
@@ -191,24 +200,12 @@ scheduler::create_kernel_task(pid_t pid, void (*task)())
 	uint16_t kcs = (1 << 3) | 0; // Kernel CS is GDT entry 1, ring 0
 	uint16_t kss = (2 << 3) | 0; // Kernel SS is GDT entry 2, ring 0
 
-	// Create a one-page kernel stack space
-	void *stack0 = proc->setup_kernel_stack(stack_size, 0);
+	// Create an initial kernel stack space
+	proc->setup_kernel_stack();
+	proc->rsp = add_fake_stack_return(
+			proc->rsp0, proc->rsp0,
+			reinterpret_cast<uintptr_t>(task));
 
-	// Stack grows down, point to the end, resere space for initial null frame
-	static const size_t null_frame = sizeof(uint64_t);
-	void *sp0 = kutil::offset_pointer(stack0, stack_size - null_frame);
-
-	cpu_state *state = reinterpret_cast<cpu_state *>(sp0) - 1;
-
-	// Highest state in the stack is the process' kernel stack for the loader
-	// to iret to:
-	state->ss = kss;
-	state->cs = kcs;
-	state->rflags = rflags_int;
-	state->rip = reinterpret_cast<uint64_t>(task);
-	state->user_rsp = reinterpret_cast<uint64_t>(state);
-
-	proc->rsp = reinterpret_cast<uintptr_t>(state);
 	proc->pml4 = page_manager::get()->get_kernel_pml4();
 	proc->quanta = process_quanta;
 	proc->flags =
@@ -218,7 +215,8 @@ scheduler::create_kernel_task(pid_t pid, void (*task)())
 	m_runlists[default_priority].push_back(proc);
 
 	log::debug(logs::task, "Creating kernel task: pid %d  pri %d", proc->pid, proc->priority);
-	log::debug(logs::task, "     RSP0 %016lx", state);
+	log::debug(logs::task, "    RSP0 %016lx", proc->rsp0);
+	log::debug(logs::task, "     RSP %016lx", proc->rsp);
 }
 
 void
@@ -278,17 +276,15 @@ void scheduler::prune(uint64_t now)
 	}
 }
 
-uintptr_t
-scheduler::schedule(uintptr_t rsp0)
+void
+scheduler::schedule()
 {
 	// TODO: lol a real clock
 	static uint64_t now = 0;
 
 	pid_t lastpid = m_current->pid;
 
-	m_current->rsp = rsp0;
 	m_runlists[m_current->priority].remove(m_current);
-
 	if (m_current->flags && process_flags::ready) {
 		m_runlists[m_current->priority].push_back(m_current);
 	} else {
@@ -304,34 +300,25 @@ scheduler::schedule(uintptr_t rsp0)
 	}
 
 	m_current = m_runlists[pri].pop_front();
-	rsp0 = m_current->rsp;
-
-	// Set rsp0 to after the end of the about-to-be-popped cpu state
-	tss_set_stack(0, rsp0 + sizeof(cpu_state));
-	bsp_cpu_data.rsp0 = rsp0;
-
-	// Swap page tables
-	page_table *pml4 = m_current->pml4;
-	page_manager::set_pml4(pml4);
 
 	if (lastpid != m_current->pid) {
-		bool loading = m_current->flags && process_flags::loading;
-		log::debug(logs::task, "Scheduler switched to process %d, priority %d%s.",
-				m_current->pid, m_current->priority, loading ? " (loading)" : "");
-	}
 
-	return rsp0;
+		bool loading = m_current->flags && process_flags::loading;
+		log::debug(logs::task, "Scheduler switching to process %d, priority %d%s.",
+				m_current->pid, m_current->priority, loading ? " (loading)" : "");
+
+		task_switch(m_current);
+	}
 }
 
-uintptr_t
-scheduler::tick(uintptr_t rsp0)
+void
+scheduler::tick()
 {
 	if (--m_current->quanta == 0) {
 		m_current->quanta = process_quanta;
-		rsp0 = schedule(rsp0);
+		schedule();
 	}
 	m_apic->reset_timer(m_tick_count);
-	return rsp0;
 }
 
 process_node *
diff --git a/src/kernel/scheduler.h b/src/kernel/scheduler.h
index 55d2b09..9a137aa 100644
--- a/src/kernel/scheduler.h
+++ b/src/kernel/scheduler.h
@@ -10,7 +10,9 @@ class lapic;
 struct page_table;
 struct cpu_state;
 
-extern "C" uintptr_t isr_handler(uintptr_t, cpu_state*);
+extern "C" void isr_handler(cpu_state*);
+extern "C" void task_switch(process *next);
+extern "C" void task_fork(process *child);
 
 
 /// The task scheduler
@@ -46,9 +48,7 @@ public:
 	void start();
 
 	/// Run the scheduler, possibly switching to a new task
-	/// \arg rsp0  The stack pointer of the current interrupt handler
-	/// \returns   The stack pointer to switch to
-	uintptr_t schedule(uintptr_t rsp0);
+	void schedule();
 
 	/// Get the current process.
 	/// \returns  A pointer to the current process' process struct
@@ -65,7 +65,7 @@ public:
 
 private:
 	friend uintptr_t syscall_dispatch(uintptr_t, cpu_state &);
-	friend uintptr_t isr_handler(uintptr_t, cpu_state*);
+	friend void isr_handler(cpu_state*);
 	friend class process;
 
 	/// Create a new process object. This process will have its pid
@@ -75,9 +75,7 @@ private:
 	process_node * create_process(pid_t pid = 0);
 
 	/// Handle a timer tick
-	/// \arg rsp0  The stack pointer of the current interrupt handler
-	/// \returns   The stack pointer to switch to
-	uintptr_t tick(uintptr_t rsp0);
+	void tick();
 
 	void prune(uint64_t now);
 
diff --git a/src/kernel/syscall.cpp b/src/kernel/syscall.cpp
index bcee5b3..676228b 100644
--- a/src/kernel/syscall.cpp
+++ b/src/kernel/syscall.cpp
@@ -34,11 +34,11 @@ syscall_enable()
 	wrmsr(msr::ia32_fmask, 0x200);
 }
 
-uintptr_t
-syscall_dispatch(uintptr_t return_rsp, cpu_state &regs)
+void
+syscall_dispatch(cpu_state *regs)
 {
 	console *cons = console::get();
-	syscall call = static_cast<syscall>(regs.rax);
+	syscall call = static_cast<syscall>(regs->rax);
 
 	auto &s = scheduler::get();
 	auto *p = s.current();
@@ -51,7 +51,7 @@ syscall_dispatch(uintptr_t return_rsp, cpu_state &regs)
 		cons->set_color(11);
 		cons->printf("\nProcess %d: Received DEBUG syscall\n", p->pid);
 		cons->set_color();
-		print_regs(regs);
+		print_regs(*regs);
 		cons->printf("\n         Syscall enters: %8d\n", __counter_syscall_enter);
 		cons->printf("         Syscall sysret: %8d\n", __counter_syscall_sysret);
 		break;
@@ -71,7 +71,7 @@ syscall_dispatch(uintptr_t return_rsp, cpu_state &regs)
 			p->wait_on_signal(-1ull);
 			cons->printf("\nProcess %d: Received PAUSE syscall\n", p->pid);
 			cons->set_color();
-			return_rsp = s.schedule(return_rsp);
+			s.schedule();
 		}
 		break;
 
@@ -79,11 +79,11 @@ syscall_dispatch(uintptr_t return_rsp, cpu_state &regs)
 		{
 			cons->set_color(11);
 			cons->printf("\nProcess %d: Received SLEEP syscall\n", p->pid);
-			cons->printf("Sleeping until %lu\n", regs.rdi);
+			cons->printf("Sleeping until %lu\n", regs->rdi);
 			cons->set_color();
 
-			p->wait_on_time(regs.rdi);
-			return_rsp = s.schedule(return_rsp);
+			p->wait_on_time(regs->rdi);
+			s.schedule();
 		}
 		break;
 
@@ -91,34 +91,34 @@ syscall_dispatch(uintptr_t return_rsp, cpu_state &regs)
 		cons->set_color(11);
 		cons->printf("\nProcess %d: Received GETPID syscall\n", p->pid);
 		cons->set_color();
-		regs.rax = p->pid;
+		regs->rax = p->pid;
 		break;
 
 	case syscall::send:
 		{
-			pid_t target = regs.rdi;
-			uintptr_t data = regs.rsi;
+			pid_t target = regs->rdi;
+			uintptr_t data = regs->rsi;
 
 			cons->set_color(11);
 			cons->printf("\nProcess %d: Received SEND syscall, target %d, data %016lx\n", p->pid, target, data);
 			cons->set_color();
 
 			if (p->wait_on_send(target))
-				return_rsp = s.schedule(return_rsp);
+				s.schedule();
 		}
 		break;
 
 	case syscall::receive:
 		{
-			pid_t source = regs.rdi;
-			uintptr_t data = regs.rsi;
+			pid_t source = regs->rdi;
+			uintptr_t data = regs->rsi;
 
 			cons->set_color(11);
 			cons->printf("\nProcess %d: Received RECEIVE syscall, source %d, dat %016lx\n", p->pid, source, data);
 			cons->set_color();
 
 			if (p->wait_on_receive(source))
-				return_rsp = s.schedule(return_rsp);
+				s.schedule();
 		}
 		break;
 
@@ -128,8 +128,9 @@ syscall_dispatch(uintptr_t return_rsp, cpu_state &regs)
 			cons->printf("\nProcess %d: Received FORK syscall\n", p->pid);
 			cons->set_color();
 
-			pid_t pid = p->fork(return_rsp);
-			regs.rax = pid;
+			pid_t pid = p->fork(regs);
+			cons->printf("\n   fork returning %d\n", pid);
+			regs->rax = pid;
 		}
 		break;
 
@@ -137,8 +138,8 @@ syscall_dispatch(uintptr_t return_rsp, cpu_state &regs)
 		cons->set_color(11);
 		cons->printf("\nProcess %d: Received EXIT syscall\n", p->pid);
 		cons->set_color();
-		p->exit(regs.rdi);
-		return_rsp = s.schedule(return_rsp);
+		p->exit(regs->rdi);
+		s.schedule();
 		break;
 
 	default:
@@ -148,7 +149,5 @@ syscall_dispatch(uintptr_t return_rsp, cpu_state &regs)
 		_halt();
 		break;
 	}
-
-	return return_rsp;
 }
 
diff --git a/src/kernel/syscall.h b/src/kernel/syscall.h
index fb2f08e..27bed28 100644
--- a/src/kernel/syscall.h
+++ b/src/kernel/syscall.h
@@ -21,5 +21,5 @@ enum class syscall : uint64_t
 };
 
 void syscall_enable();
-uintptr_t syscall_dispatch(uintptr_t, cpu_state &);
+void syscall_dispatch(cpu_state *);
 
diff --git a/src/kernel/syscall.s b/src/kernel/syscall.s
index 02b9b3f..dfe6973 100644
--- a/src/kernel/syscall.s
+++ b/src/kernel/syscall.s
@@ -28,7 +28,6 @@ syscall_handler_prelude:
 
 	mov rdi, rsp
 	call syscall_handler
-	mov rsp, rax
 
 	mov rax, [rsp + 0x90]
 	and rax, 0x3
diff --git a/src/kernel/task.s b/src/kernel/task.s
new file mode 100644
index 0000000..3659578
--- /dev/null
+++ b/src/kernel/task.s
@@ -0,0 +1,85 @@
+%include "tasking.inc"
+
+extern g_tss
+
+global task_switch
+task_switch:
+	push rbp
+	mov rbp, rsp
+
+	; Save the rest of the callee-saved regs
+	push rbx
+	push r12
+	push r13
+	push r14
+	push r15
+
+	; Update previous task's TCB
+	mov rax, [gs:CPU_DATA.tcb]     ; rax: current task TCB
+	mov [rax + TCB.rsp], rsp
+
+	; Install next task's TCB
+	mov [gs:CPU_DATA.tcb], rdi     ; rdi: next TCB (function param)
+	mov rsp, [rdi + TCB.rsp]       ; next task's stack pointer
+	mov rax, 0x0000007fffffffff
+	and rax, [rdi + TCB.pml4]      ; rax: next task's pml4 (phys portion of address)
+
+	; Update syscall/interrupt rsp
+	mov rcx, [rdi + TCB.rsp0]      ; rcx: top of next task's kernel stack
+	mov [gs:CPU_DATA.rsp0], rcx
+
+	lea rdx, [rel g_tss]           ; rdx: address of TSS
+	mov [rdx + TSS.rsp0], rcx
+
+	; check if we need to update CR3
+	mov rdx, cr3                   ; rdx: old CR3
+	cmp rax, rdx
+	je .no_cr3
+	mov cr3, rax
+.no_cr3:
+
+	pop r15
+	pop r14
+	pop r13
+	pop r12
+	pop rbx
+
+	pop rbp
+	ret
+
+global task_fork
+task_fork:
+	push rbp
+	mov rbp, rsp
+
+	; Save the rest of the callee-saved regs
+	push rbx
+	push r12
+	push r13
+	push r14
+	push r15
+
+	mov r14, rdi                  ; r14: child task TCB (function argument)
+
+	mov rax, [gs:CPU_DATA.tcb]    ; rax: current task TCB
+	mov rax, [rax + TCB.rsp0]     ; rax: current task rsp0
+	sub rax, rsp                  ; rax: size of kernel stack in bytes
+
+	mov rcx, rax
+	shr rcx, 3                    ; rcx: size of kernel stack in qwords
+
+	mov rdi, [r14 + TCB.rsp0]     ; rdi: child task rsp0
+	sub rdi, rax                  ; rdi: child task rsp
+	mov rsi, rsp                  ; rsi: current rsp
+
+	rep movsq
+
+	pop r15
+	pop r14
+	pop r13
+	pop r12
+	pop rbx
+
+	pop rbp
+	ret
+
diff --git a/src/kernel/tasking.inc b/src/kernel/tasking.inc
new file mode 100644
index 0000000..ad1d0c0
--- /dev/null
+++ b/src/kernel/tasking.inc
@@ -0,0 +1,32 @@
+struc TCB
+.rsp:          resq 1
+.rsp0:		   resq 1
+.pml4:         resq 1
+endstruc
+
+struc CPU_DATA
+.rsp0:         resq 1
+.rsp3:         resq 1
+.tcb:          resq 1
+endstruc
+
+struc TSS
+.res0:         resd 1
+.rsp0:         resq 1
+.rsp1:         resq 1
+.rsp2:         resq 1
+.ist0:         resq 1
+.ist1:         resq 1
+.ist2:         resq 1
+.ist3:         resq 1
+.ist4:         resq 1
+.ist5:         resq 1
+.ist6:         resq 1
+.ist7:         resq 1
+.res1:         resq 1
+.res2:         resw 1
+.iomap:        resw 1
+endstruc
+
+
+; vim: ft=asm