os/kernel/boot.s

# Declare constants for the multiboot header.
.set ALIGN,    1<<0             # align loaded modules on page boundaries
.set MEMINFO,  1<<1             # provide memory map
.set FLAGS,    ALIGN | MEMINFO  # this is the Multiboot 'flag' field
.set MAGIC,    0x1BADB002       # 'magic number' lets bootloader find the header
.set CHECKSUM, -(MAGIC + FLAGS) # checksum of above, to prove we are multiboot

# Declare a multiboot header that marks the program as a kernel.
.section .multiboot
.align 4
.long MAGIC
.long FLAGS
.long CHECKSUM

# Allocate the initial stack.
.section .bootstrap_stack, "aw", @nobits
stack_bottom:
.skip 16384 # 16 KiB
stack_top:

.global int_stack_top

.section .interrupt_stack, "aw", @nobits
int_stack_bottom:
.skip 16384 # 16 KiB
int_stack_top:

# Preallocate pages used for paging. Don't hard-code addresses and assume they
# are available, as the bootloader might have loaded its multiboot structures or
# modules there. This lets the bootloader know it must avoid the addresses.
.section .bss, "aw", @nobits
	.align 4096
boot_page_directory:
	.skip 4096
boot_page_tables:
boot_page_table1:
	.skip 4096
boot_page_table2:
	.skip 4096
boot_page_table3:
	.skip 4096
boot_page_table4:
	.skip 4096
# Further page tables may be required if the kernel grows beyond 3 MiB.

# The kernel entry point.
.section .text
.global _start
.type _start, @function
_start:
	cmp $0x2BADB002, %eax
	jnz no_multiboot
	# Physical address of boot_page_tables.
	# TODO: I recall seeing some assembly that used a macro to do the
	#       conversions to and from physical. Maybe this should be done in this
	#       code as well?
	movl $(boot_page_tables - 0xC0000000), %edi
	# First address to map is address 0.
	# TODO: Start at the first kernel page instead. Alternatively map the first
	#       1 MiB as it can be generally useful, and there's no need to
	#       specially map the VGA buffer.
	movl $0, %esi
	# Map 4096 pages.
	movl $4096, %ecx

1:
	# Map physical address as "present, writable". Note that this maps
	# .text and .rodata as writable. Mind security and map them as non-writable.
	movl %esi, %edx
	orl $0x007, %edx
	movl %edx, (%edi)

	# Size of page is 4096 bytes.
	addl $4096, %esi
	# Size of entries in boot_page_tables is 4 bytes.
	addl $4, %edi
	# Loop to the next entry if we haven't finished.
	loop 1b

	# The page table is used at both page directory entry 0 (virtually from 0x0
	# to 0x3FFFFF) (thus identity mapping the kernel) and page directory entry
	# 768 (virtually from 0xC0000000 to 0xC03FFFFF) (thus mapping it in the
	# higher half). The kernel is identity mapped because enabling paging does
	# not change the next instruction, which continues to be physical. The CPU
	# would instead page fault if there was no identity mapping.

	# Map the page table to both virtual addresses 0x00000000 and 0xC0000000.
	movl $(boot_page_table1 - 0xC0000000 + 0x007), boot_page_directory - 0xC0000000 + 0
	movl $(boot_page_table1 - 0xC0000000 + 0x007), boot_page_directory - 0xC0000000 + 768 * 4
	movl $(boot_page_table2 - 0xC0000000 + 0x007), boot_page_directory - 0xC0000000 + 769 * 4
	movl $(boot_page_table3 - 0xC0000000 + 0x007), boot_page_directory - 0xC0000000 + 770 * 4
	movl $(boot_page_table4 - 0xC0000000 + 0x007), boot_page_directory - 0xC0000000 + 771 * 4
	# Set cr3 to the address of the boot_page_directory.
	movl $(boot_page_directory - 0xC0000000), %ecx
	movl %ecx, %cr3

	# Enable paging and the write-protect bit.
	movl %cr0, %ecx
	orl $0x80010000, %ecx
	movl %ecx, %cr0

	# Jump to higher half with an absolute jump.
	lea 4f, %ecx
	jmp *%ecx

4:
	# At this point, paging is fully set up and enabled.

	# Unmap the identity mapping as it is now unnecessary.
	movl $0, boot_page_directory + 0

	# Reload crc3 to force a TLB flush so the changes to take effect.
	movl %cr3, %ecx
	movl %ecx, %cr3

	# Set up the stack.
	mov $stack_top, %esp

	# Enter the high-level kernel.
	add $0xC0000000, %ebx
	push %ebx
	call kmain

	# Infinite loop if the system has nothing more to do.
	no_multiboot:
	loop: jmp loop