ASM bootloader that sets up long mode paging with a self-referential PML4E

boot.S

[bits 16]

load:
	mov ah, 0x02 ; read disk sectors
	mov al, 0x01 ; 1 sector (probably should statically calculate this somehow, or use linker script)
	mov ch, 0x00 ; track
	mov cl, 0x02 ; sector
	mov dh, 0x00; head
;	mov dl, 0x00
	mov bx, 0
	mov es, bx
	mov bx, 0
.readsector:
	int 0x13
	jc .readsector ; try again if it fails
	jmp _start

; -- real mode --
_start:
	lea esp, [__stack_top]

	mov eax, cr0
	or eax, 1 << 0
	mov cr0, eax
	lgdt [GDT32.Pointer]
	jmp GDT32.Code:protected_mode_entry

; -- protected mode w/ segmentation --
[bits 32]
protected_mode_entry:
	mov edi, 0x3000
	mov cr3, edi
	xor eax, eax
	mov ecx, 4096
	rep stosd
	mov eax, cr3
	mov dword [eax], 0x4000 | 3
	add eax, 4096
	mov dword [eax], 0x5000 | 3
	add eax, 4096
	mov dword [eax], 0x6000 | 3
	add eax, 4096
	mov ebx, 0x00000003

	mov ecx, 512

	set_page_entry:
	mov dword [eax], ebx
	add eax, 8
	add ebx, 0x1000
	loop set_page_entry


	mov ecx, 0xC0000080
	rdmsr
	or eax, 1 << 8 ; efer.lme
	wrmsr

	mov edi, cr4
	or edi, 1 << 5 ; PAE
	mov cr4, edi

	mov edi, cr0
	or edi, 1 << 31 ; paging on
	mov cr0, edi

	lgdt [GDT64.Pointer]
	jmp GDT64.Code:long_mode_entry


[bits 64]
long_mode_entry:
	mov rbp, rsp
	xor r8, r8

	rdtsc
	and eax, 0xff
	mov r8, rax
	nop
	shl r8, 9

	mov qword [0x3000 + 8*eax], 0x8000 | 0x3 ; pml4e

	rdtsc
	and eax, 0x1ff
	or r8, rax
	shl r8, 9

	mov qword [0x8000 + 8*eax], 0x8000 | 0x3 ; pdpte

	rdtsc
	and eax, 0x1ff
	or r8, rax
	shl r8, 9

	mov qword [0x8000 + 8*eax], 0x9000 | 0x3 ; pde

	mov qword [0x9000], 0x0 | 0x3

	shl r8, 12
	lea r14, [r8 + add_self_refential_mapping]
	jmp r14

add_self_refential_mapping:
	rdtsc
	and eax, 0x1ff ; meagre 9 bits of entropy
	cmp eax, 0
	je add_self_refential_mapping
	mov qword [0x3000 + 8*eax], 0x3000 | 0x3 ; pml4e

	mov qword [0x3000], 0x0
	cli
.done:
	   hlt

GDT64:
	dq 0
.Code: equ $ - GDT64
	dq (1<<44) | (1<<47) | (1<<41) | (1<<43) | (1<<53)
.Data: equ $ - GDT64
	dq (1<<44) | (1<<47) | (1<<41)
.Pointer:
	dw .Pointer - GDT64 - 1
	dq GDT64

GDT32:
	.Null: equ $ - GDT32
	dd 0x0, 0x0

	.Code: equ $ - GDT32
	dw 0xffff, 0x0
	db 0x0, 10011010b, 11001111b, 0x0

	.Pointer:
	dw $ - GDT32 - 1
	dd GDT32