Table

of Contents
Introduction

1.1

Introduction about the x86 architecture and about our OS

1.2

Setup the development environment

1.3

First boot with GRUB

1.4

Backbone of the OS and C++ runtime

1.5

Base classes for managing x86 architecture

1.6

GDT

1.7

IDT and interrupts

1.8

Theory: physical and virtual memory

1.9

Memory management: physical and virtual

1.10

Process management and multitasking

1.11

External program execution: ELF files

1.12

Userland and syscalls

1.13

Modular drivers

1.14

Some basics modules: console, keyboard

1.15

IDE Hard disks

1.16

DOS Partitions

1.17

EXT2 read-only filesystems

1.18

Standard C library (libC)

1.19

UNIX basic tools: sh, cat

1.20

Lua interpreter

1.21

Introduction

How to Make a Computer Operating

System
Online book about how to write a computer operating system in C/C++ from scratch.
Caution: This repository is a remake of my old course. It was written several years ago as
one of my first projects when I was in High School, I'm still refactoring some parts. The
original course was in French and I'm not an English native. I'm going to continue and
improve this course in my free-time.
Book: An online version is available at http://samypesse.gitbooks.io/how-to-create-anoperating-system/ (PDF, Mobi and ePub). It was generated using GitBook.
Source Code: All the system source code will be stored in the src directory. Each step will
contain links to the different related files.
Contributions: This course is open to contributions, feel free to signal errors with issues or
directly correct the errors with pull-requests.
Questions: Feel free to ask any questions by adding issues or commenting sections.
You can follow me on Twitter @SamyPesse or GitHub.

What kind of OS are we building?

The goal is to build a very simple UNIX-based operating system in C++, not just a "proof-ofconcept". The OS should be able to boot, start a userland shell, and be extensible.

Introduction

Introduction about the x86 architecture and about our OS

Chapter 1: Introduction to the x86 architecture

and about our OS
What is the x86 architecture?
The term x86 denotes a family of backward compatible instruction set architectures
based on the Intel 8086 CPU.
The x86 architecture is the most common instruction set architecture since its introduction in
1981 for the IBM PC. A large amount of software, including operating systems (OS's) such
as DOS, Windows, Linux, BSD, Solaris and Mac OS X, function with x86-based hardware.
In this course we are not going to design an operating system for the x86-64 architecture but
for x86-32, thanks to backward compatibility, our OS will be compatible with our newer PCs
(but take caution if you want to test it on your real machine).

Our Operating System

The goal is to build a very simple UNIX-based operating system in C++, but the goal is not to
just build a "proof-of-concept". The OS should be able to boot, start a userland shell and be
extensible.
The OS will be built for the x86 architecture, running on 32 bits, and compatible with IBM
PCs.
Specifications:
Code in C++
x86, 32 bit architecture
Boot with Grub
Kind of modular system for drivers
Kind of UNIX style
Multitasking
ELF executable in userland
Modules (accessible in userland using /dev/...) :
IDE disks
DOS partitions
Clock
EXT2 (read only)
Boch VBE
Userland :

Introduction about the x86 architecture and about our OS

API Posix
LibC
"Can" run a shell or some executables (e.g., lua)

Setup the development environment

Chapter 2: Setup the development environment

The first step is to setup a good and viable development environment. Using Vagrant and
Virtualbox, you'll be able to compile and test your OS from all the OSs (Linux, Windows or
Mac).

Install Vagrant
Vagrant is free and open-source software for creating and configuring virtual
development environments. It can be considered a wrapper around VirtualBox.
Vagrant will help us create a clean virtual development environment on whatever system you
are using. The first step is to download and install Vagrant for your system at
http://www.vagrantup.com/.

Install Virtualbox
Oracle VM VirtualBox is a virtualization software package for x86 and AMD64/Intel64based computers.
Vagrant needs Virtualbox to work, Download and install for your system at
https://www.virtualbox.org/wiki/Downloads.

Start and test your development environment

Once Vagrant and Virtualbox are installed, you need to download the ubuntu lucid32 image
for Vagrant:
vagrant box add lucid32 http://files.vagrantup.com/lucid32.box

Once the lucid32 image is ready, we need to define our development environment using a
Vagrantfile, create a file named Vagrantfile. This file defines what prerequisites our
environment needs: nasm, make, build-essential, grub and qemu.
Start your box using:
vagrant up

You can now access your box by using ssh to connect to the virtual box using:

Setup the development environment

vagrant ssh

The directory containing the Vagrantfile will be mounted by default in the /vagrant directory
of the guest VM (in this case, Ubuntu Lucid32):
cd /vagrant

Build and test our operating system

The file Makefile defines some basics rules for building the kernel, the user libc and some
userland programs.
Build:
make all

Test our operating system with qemu:

make run

The documentation for qemu is available at QEMU Emulator Documentation.

You can exit the emulator using: Ctrl-a.

First boot with GRUB

Chapter 3: First boot with GRUB

How the boot works?
When an x86-based computer is turned on, it begins a complex path to get to the stage
where control is transferred to our kernel's "main" routine ( kmain() ). For this course, we are
only going to consider the BIOS boot method and not it's successor (UEFI).
The BIOS boot sequence is: RAM detection -> Hardware detection/Initialization -> Boot
sequence.
The most important step for us is the "Boot sequence", where the BIOS is done with its
initialization and tries to transfer control to the next stage of the bootloader process.
During the "Boot sequence", the BIOS will try to determine a "boot device" (e.g. floppy disk,
hard-disk, CD, USB flash memory device or network). Our Operating System will initially
boot from the hard-disk (but it will be possible to boot it from a CD or a USB flash memory
device in future). A device is considered bootable if the bootsector contains the valid
signature bytes 0x55 and 0xAA at offsets 511 and 512 respectively (called the magic bytes
of the Master Boot Record, also known as the MBR). This signature is represented (in
binary) as 0b1010101001010101. The alternating bit pattern was thought to be a protection
against certain failures (drive or controller). If this pattern is garbled or 0x00, the device is
not considered bootable.
BIOS physically searches for a boot device by loading the first 512 bytes from the bootsector
of each device into physical memory, starting at the address 0x7C00 (1 KiB below the 32
KiB mark). When the valid signature bytes are detected, BIOS transfers control to the
0x7C00 memory address (via a jump instruction) in order to execute the bootsector code.

Throughout this process the CPU has been running in 16-bit Real Mode, which is the default
state for x86 CPUs in order to maintain backwards compatibility. To execute the 32-bit
instructions within our kernel, a bootloader is required to switch the CPU into Protected
Mode.

What is GRUB?
GNU GRUB (short for GNU GRand Unified Bootloader) is a boot loader package from
the GNU Project. GRUB is the reference implementation of the Free Software
Foundation's Multiboot Specification, which provides a user the choice to boot one of
multiple operating systems installed on a computer or select a specific kernel
configuration available on a particular operating system's partitions.

First boot with GRUB

To make it simple, GRUB is the first thing booted by the machine (a boot-loader) and will
simplify the loading of our kernel stored on the hard-disk.

Why are we using GRUB?

GRUB is very simple to use
Make it very simple to load 32bits kernels without needs of 16bits code
Multiboot with Linux, Windows and others
Make it easy to load external modules in memory

How to use GRUB?

GRUB uses the Multiboot specification, the executable binary should be 32bits and must
contain a special header (multiboot header) in its 8192 first bytes. Our kernel will be a ELF
executable file ("Executable and Linkable Format", a common standard file format for
executables in most UNIX system).
The first boot sequence of our kernel is written in Assembly: start.asm and we use a linker
file to define our executable structure: linker.ld.
This boot process also initializes some of our C++ runtime, it will be described in the next
chapter.
Multiboot header structure:

10

First boot with GRUB

struct multiboot_info {
u32 flags;
u32 low_mem;
u32 high_mem;
u32 boot_device;
u32 cmdline;
u32 mods_count;
u32 mods_addr;
struct {
u32 num;
u32 size;
u32 addr;
u32 shndx;
} elf_sec;
unsigned long mmap_length;
unsigned long mmap_addr;
unsigned long drives_length;
unsigned long drives_addr;
unsigned long config_table;
unsigned long boot_loader_name;
unsigned long apm_table;
unsigned long vbe_control_info;
unsigned long vbe_mode_info;
unsigned long vbe_mode;
unsigned long vbe_interface_seg;
unsigned long vbe_interface_off;
unsigned long vbe_interface_len;
};

You can use the command mbchk kernel.elf to validate your kernel.elf file against the
multiboot standard. You can also use the command nm -n kernel.elf to validate the offset
of the different objects in the ELF binary.

Create a disk image for our kernel and grub

The script diskimage.sh will generate a hard disk image that can be used by QEMU.
The first step is to create a hard-disk image (c.img) using qemu-img:
qemu-img create c.img 2M

We need now to partition the disk using fdisk:

11

First boot with GRUB

fdisk ./c.img
# Switch to Expert commands
> x
# Change number of cylinders (1-1048576)
> c
> 4
# Change number of heads (1-256, default 16):
> h
> 16
# Change number of sectors/track (1-63, default 63)
> s
> 63
# Return to main menu
> r
# Add a new partition
> n
# Choose primary partition
> p
# Choose partition number
> 1
# Choose first sector (1-4, default 1)
> 1
# Choose last sector, +cylinders or +size{K,M,G} (1-4, default 4)
> 4
# Toggle bootable flag
> a
# Choose first partition for bootable flag
> 1
# Write table to disk and exit
> w

We need now to attach the created partition to the loop-device using losetup. This allows a
file to be access like a block device. The offset of the partition is passed as an argument and
calculated using: offset= start_sector * bytes_by_sector.
Using fdisk -l -u c.img , you get: 63 * 512 = 32256.

12

First boot with GRUB

losetup -o 32256 /dev/loop1 ./c.img

We create a EXT2 filesystem on this new device using:

mke2fs /dev/loop1

We copy our files on a mounted disk:

mount /dev/loop1 /mnt/
cp -R bootdisk/* /mnt/
umount /mnt/

Install GRUB on the disk:

grub --device-map=/dev/null << EOF
device (hd0) ./c.img
geometry (hd0) 4 16 63
root (hd0,0)
setup (hd0)
quit
EOF

And finally we detach the loop device:

losetup -d /dev/loop1

See Also
GNU GRUB on Wikipedia
Multiboot specification

13

Backbone of the OS and C++ runtime

Chapter 4: Backbone of the OS and C++

runtime
C++ kernel run-time
A kernel can be written in C++ just as it can be in C, with the exception of a few pitfalls that
come with using C++ (runtime support, constructors, etc).
The compiler will assume that all the necessary C++ runtime support is available by default,
but as we are not linking libsupc++ into your C++ kernel, we need to add some basic
functions that can be found in the cxx.cc file.
Caution: The operators new and delete cannot be used before virtual memory and
pagination have been initialized.

Basic C/C++ functions

The kernel code can't use functions from the standard libraries so we need to add some
basic functions for managing memory and strings:
void itoa(char *buf, unsigned long int n, int base);
void * memset(char *dst,char src, int n);
void * memcpy(char *dst, char *src, int n);
int strlen(char *s);
int strcmp(const char *dst, char *src);
int strcpy(char *dst,const char *src);
void strcat(void *dest,const void *src);
char * strncpy(char *destString, const char *sourceString,int maxLength);
int strncmp( const char* s1, const char* s2, int c );

These functions are defined in string.cc, memory.cc, itoa.cc

C types
In the next step, we're going to define different types we're going to use in our code. Most of
our variable types are going to be unsigned. This means that all the bits are used to store
the integer. Signed variables use their first bit to indicate their sign.

14

Backbone of the OS and C++ runtime

typedef unsigned char u8;

typedef unsigned short u16;
typedef unsigned int u32;
typedef unsigned long long u64;
typedef signed char s8;
typedef signed short s16;
typedef signed int s32;
typedef signed long long s64;

Compile our kernel

Compiling a kernel is not the same thing as compiling a linux executable, we can't use a
standard library and should have no dependencies to the system.
Our Makefile will define the process to compile and link our kernel.
For x86 architecture, the followings arguments will be used for gcc/g++/ld:
# Linker
LD=ld
LDFLAG= -melf_i386 -static -L ./ -T ./arch/$(ARCH)/linker.ld
# C++ compiler
SC=g++
FLAG= $(INCDIR) -g -O2 -w -trigraphs -fno-builtin -fno-exceptions -fno-stack-protecto
r -O0 -m32 -fno-rtti -nostdlib -nodefaultlibs
# Assembly compiler
ASM=nasm
ASMFLAG=-f elf -o

15

Base classes for managing x86 architecture

Chapter 5: Base classes for managing x86

architecture
Now that we know how to compile our C++ kernel and boot the binary using GRUB, we can
start to do some cool things in C/C++.

Printing to the screen console

We are going to use VGA default mode (03h) to display some text to the user. The screen
can be directly accessed using the video memory at 0xB8000. The screen resolution is
80x25 and each character on the screen is defined by 2 bytes: one for the character code,
and one for the style flag. This means that the total size of the video memory is 4000B
(80B25B2B).
In the IO class (io.cc),:
x,y: define the cursor position on the screen
real_screen: define the video memory pointer
putc(char c): print a unique character on the screen and manage cursor position
printf(char* s, ...): print a string
We add a method putc to the IO Class to put a character on the screen and update the (x,y)
position.

16

Base classes for managing x86 architecture

/* put a byte on screen */

void Io::putc(char c){
kattr = 0x07;
unsigned char *video;
video = (unsigned char *) (real_screen+ 2 * x + 160 * y);
// newline
if (c == '\n') {
x = 0;
y++;
// back space
} else if (c == '\b') {
if (x) {
*(video + 1) = 0x0;
x--;
}
// horizontal tab
} else if (c == '\t') {
x = x + 8 - (x % 8);
// carriage return
} else if (c == '\r') {
x = 0;
} else {
*video = c;
*(video + 1) = kattr;
x++;
if (x > 79) {
x = 0;
y++;
}
}
if (y > 24)
scrollup(y - 24);
}

We also add a useful and very known method: printf

/* put a string in screen */
void Io::print(const char *s, ...){
va_list ap;
char buf[16];
int i, j, size, buflen, neg;
unsigned char c;
int ival;
unsigned int uival;
va_start(ap, s);
while ((c = *s++)) {

17

Base classes for managing x86 architecture

size = 0;
neg = 0;
if (c == 0)
break;
else if (c == '%') {
c = *s++;
if (c >= '0' && c <= '9') {
size = c - '0';
c = *s++;
}
if (c == 'd') {
ival = va_arg(ap, int);
if (ival < 0) {
uival = 0 - ival;
neg++;
} else
uival = ival;
itoa(buf, uival, 10);
buflen = strlen(buf);
if (buflen < size)
for (i = size, j = buflen; i >= 0;
i--, j--)
buf[i] =
(j >=
0) ? buf[j] : '0';
if (neg)
print("-%s", buf);
else
print(buf);
}
else if (c == 'u') {
uival = va_arg(ap, int);
itoa(buf, uival, 10);
buflen = strlen(buf);
if (buflen < size)
for (i = size, j = buflen; i >= 0;
i--, j--)
buf[i] =
(j >=
0) ? buf[j] : '0';
print(buf);
} else if (c == 'x' || c == 'X') {
uival = va_arg(ap, int);
itoa(buf, uival, 16);
buflen = strlen(buf);
if (buflen < size)

18

Base classes for managing x86 architecture

for (i = size, j = buflen; i >= 0;

i--, j--)
buf[i] =
(j >=
0) ? buf[j] : '0';
print("0x%s", buf);
} else if (c == 'p') {
uival = va_arg(ap, int);
itoa(buf, uival, 16);
size = 8;
buflen = strlen(buf);
if (buflen < size)
for (i = size, j = buflen; i >= 0;
i--, j--)
buf[i] =
(j >=
0) ? buf[j] : '0';
print("0x%s", buf);
} else if (c == 's') {
print((char *) va_arg(ap, int));
}
} else
putc(c);
}
return;
}

Assembly interface
A large number of instructions are available in Assembly but there is not equivalent in C (like
cli, sti, in and out), so we need an interface to these instructions.
In C, we can include Assembly using the directive "asm()", gcc use gas to compile the
assembly.
Caution: gas uses the AT&T syntax.

19

Base classes for managing x86 architecture

/* output byte */
void Io::outb(u32 ad, u8 v){
asmv("outb %%al, %%dx" :: "d" (ad), "a" (v));;
}
/* output word */
void Io::outw(u32 ad, u16 v){
asmv("outw %%ax, %%dx" :: "d" (ad), "a" (v));
}
/* output word */
void Io::outl(u32 ad, u32 v){
asmv("outl %%eax, %%dx" : : "d" (ad), "a" (v));
}
/* input byte */
u8 Io::inb(u32 ad){
u8 _v; \
asmv("inb %%dx, %%al" : "=a" (_v) : "d" (ad)); \
return _v;
}
/* input word */
u16 Io::inw(u32 ad){
u16 _v; \
asmv("inw %%dx, %%ax" : "=a" (_v) : "d" (ad)); \
return _v;
}
/* input word */
u32 Io::inl(u32 ad){
u32 _v; \
asmv("inl %%dx, %%eax" : "=a" (_v) : "d" (ad)); \
return _v;
}

20

GDT

Chapter 6: GDT
Thanks to GRUB, your kernel is no longer in real-mode, but already in protected mode, this
mode allows us to use all the possibilities of the microprocessor such as virtual memory
management, paging and safe multi-tasking.

What is the GDT?

The GDT ("Global Descriptor Table") is a data structure used to define the different memory
areas: the base address, the size and access privileges like execute and write. These
memory areas are called "segments".
We are going to use the GDT to define different memory segments:
"code": kernel code, used to stored the executable binary code
"data": kernel data
"stack": kernel stack, used to stored the call stack during kernel execution
"ucode": user code, used to stored the executable binary code for user program
"udata": user program data
"ustack": user stack, used to stored the call stack during execution in userland

How to load our GDT?

GRUB initializes a GDT but this GDT is does not correspond to our kernel. The GDT is
loaded using the LGDT assembly instruction. It expects the location of a GDT description
structure:

And the C structure:

21

GDT

struct gdtr {
u16 limite;
u32 base;
} __attribute__ ((packed));

Caution: the directive __attribute__ ((packed)) signal to gcc that the structure should use
as little memory as possible. Without this directive, gcc include some bytes to optimize the
memory alignment and the access during execution.
Now we need to define our GDT table and then load it using LGDT. The GDT table can be
stored wherever we want in memory, its address should just be signaled to the process
using the GDTR registry.
The GDT table is composed of segments with the following structure:

And the C structure:

struct gdtdesc {
u16 lim0_15;
u16 base0_15;
u8 base16_23;
u8 acces;
u8 lim16_19:4;
u8 other:4;
u8 base24_31;
} __attribute__ ((packed));

How to define our GDT table?

We need now to define our GDT in memory and finally load it using the GDTR registry.
We are going to store our GDT at the address:
#define GDTBASE 0x00000800

22

GDT

The function init_gdt_desc in x86.cc initialize a gdt segment descriptor.

void init_gdt_desc(u32 base, u32 limite, u8 acces, u8 other, struct gdtdesc *desc)
{
desc->lim0_15 = (limite & 0xffff);
desc->base0_15 = (base & 0xffff);
desc->base16_23 = (base & 0xff0000) >> 16;
desc->acces = acces;
desc->lim16_19 = (limite & 0xf0000) >> 16;
desc->other = (other & 0xf);
desc->base24_31 = (base & 0xff000000) >> 24;
return;
}

And the function init_gdt initialize the GDT, some parts of the below function will be
explained later and are used for multitasking.

23

GDT

void init_gdt(void)
{
default_tss.debug_flag = 0x00;
default_tss.io_map = 0x00;
default_tss.esp0 = 0x1FFF0;
default_tss.ss0 = 0x18;
/* initialize gdt segments */
init_gdt_desc(0x0, 0x0, 0x0, 0x0, &kgdt[0]);
init_gdt_desc(0x0, 0xFFFFF, 0x9B, 0x0D, &kgdt[1]); /* code */
init_gdt_desc(0x0, 0xFFFFF, 0x93, 0x0D, &kgdt[2]); /* data */
init_gdt_desc(0x0, 0x0, 0x97, 0x0D, &kgdt[3]); /* stack */
init_gdt_desc(0x0, 0xFFFFF, 0xFF, 0x0D, &kgdt[4]); /* ucode */
init_gdt_desc(0x0, 0xFFFFF, 0xF3, 0x0D, &kgdt[5]); /* udata */
init_gdt_desc(0x0, 0x0, 0xF7, 0x0D, &kgdt[6]); /* ustack */
init_gdt_desc((u32) & default_tss, 0x67, 0xE9, 0x00, &kgdt[7]); /* descripteur
de tss */
/* initialize the gdtr structure */
kgdtr.limite = GDTSIZE * 8;
kgdtr.base = GDTBASE;
/* copy the gdtr to its memory area */
memcpy((char *) kgdtr.base, (char *) kgdt, kgdtr.limite);
/* load the gdtr registry */
asm("lgdtl (kgdtr)");
/* initiliaz the segments */
asm(" movw $0x10, %ax \n \
movw %ax, %ds \n \
movw %ax, %es \n \
movw %ax, %fs \n \
movw %ax, %gs \n \
ljmp $0x08, $next \n \
next: \n");
}

24

IDT and interrupts

Chapter 7: IDT and interrupts

An interrupt is a signal to the processor emitted by hardware or software indicating an event
that needs immediate attention.
There are 3 types of interrupts:
Hardware interrupts: are sent to the processor from an external device (keyboard,
mouse, hard disk, ...). Hardware interrupts were introduced as a way to reduce wasting
the processor's valuable time in polling loops, waiting for external events.
Software interrupts: are initiated voluntarily by the software. It's used to manage
system calls.
Exceptions: are used for errors or events occurring during program execution that are
exceptional enough that they cannot be handled within the program itself (division by
zero, page fault, ...)

The keyboard example:

When the user pressed a key on the keyboard, the keyboard controller will signal an
interrupt to the Interrupt Controller. If the interrupt is not masked, the controller will signal the
interrupt to the processor, the processor will execute a routine to manage the interrupt (key
pressed or key released), this routine could, for example, get the pressed key from the
keyboard controller and print the key to the screen. Once the character processing routine is
completed, the interrupted job can be resumed.

What is the PIC?

The PIC (Programmable interrupt controller)is a device that is used to combine several
sources of interrupt onto one or more CPU lines, while allowing priority levels to be assigned
to its interrupt outputs. When the device has multiple interrupt outputs to assert, it asserts
them in the order of their relative priority.
The best known PIC is the 8259A, each 8259A can handle 8 devices but most computers
have two controllers: one master and one slave, this allows the computer to manage
interrupts from 14 devices.
In this chapter, we will need to program this controller to initialize and mask interrupts.

What is the IDT?

25

IDT and interrupts

The Interrupt Descriptor Table (IDT) is a data structure used by the x86 architecture to
implement an interrupt vector table. The IDT is used by the processor to determine the
correct response to interrupts and exceptions.
Our kernel is going to use the IDT to define the different functions to be executed when an
interrupt occurred.
Like the GDT, the IDT is loaded using the LIDTL assembly instruction. It expects the location
of a IDT description structure:
struct idtr {
u16 limite;
u32 base;
} __attribute__ ((packed));

The IDT table is composed of IDT segments with the following structure:
struct idtdesc {
u16 offset0_15;
u16 select;
u16 type;
u16 offset16_31;
} __attribute__ ((packed));

Caution: the directive __attribute__ ((packed)) signal to gcc that the structure should use
as little memory as possible. Without this directive, gcc includes some bytes to optimize the
memory alignment and the access during execution.
Now we need to define our IDT table and then load it using LIDTL. The IDT table can be
stored wherever we want in memory, its address should just be signaled to the process
using the IDTR registry.
Here is a table of common interrupts (Maskable hardware interrupt are called IRQ):

26

IDT and interrupts

IRQ

Description

Programmable Interrupt Timer Interrupt

Keyboard Interrupt

Cascade (used internally by the two PICs. never raised)

COM2 (if enabled)

COM1 (if enabled)

LPT2 (if enabled)

Floppy Disk

LPT1

CMOS real-time clock (if enabled)

Free for peripherals / legacy SCSI / NIC

10

Free for peripherals / SCSI / NIC

11

Free for peripherals / SCSI / NIC

12

PS2 Mouse

13

FPU / Coprocessor / Inter-processor

14

Primary ATA Hard Disk

15

Secondary ATA Hard Disk

How to initialize the interrupts?

This is a simple method to define an IDT segment
void init_idt_desc(u16 select, u32 offset, u16 type, struct idtdesc *desc)
{
desc->offset0_15 = (offset & 0xffff);
desc->select = select;
desc->type = type;
desc->offset16_31 = (offset & 0xffff0000) >> 16;
return;
}

And we can now initialize the interupts:

#define IDTBASE 0x00000000
#define IDTSIZE 0xFF
idtr kidtr;

27

IDT and interrupts

void init_idt(void)
{
/* Init irq */
int i;
for (i = 0; i < IDTSIZE; i++)
init_idt_desc(0x08, (u32)_asm_schedule, INTGATE, &kidt[i]); //
/* Vectors 0 -> 31 are for exceptions */
init_idt_desc(0x08, (u32) _asm_exc_GP, INTGATE, &kidt[13]); /* #GP */
init_idt_desc(0x08, (u32) _asm_exc_PF, INTGATE, &kidt[14]); /* #PF */
init_idt_desc(0x08, (u32) _asm_schedule, INTGATE, &kidt[32]);
init_idt_desc(0x08, (u32) _asm_int_1, INTGATE, &kidt[33]);
init_idt_desc(0x08, (u32) _asm_syscalls, TRAPGATE, &kidt[48]);
init_idt_desc(0x08, (u32) _asm_syscalls, TRAPGATE, &kidt[128]); //48
kidtr.limite = IDTSIZE * 8;
kidtr.base = IDTBASE;

/* Copy the IDT to the memory */

memcpy((char *) kidtr.base, (char *) kidt, kidtr.limite);
/* Load the IDTR registry */
asm("lidtl (kidtr)");
}

After intialization of our IDT, we need to activate interrupts by configuring the PIC. The
following function will configure the two PICs by writting in their internal registries using the
output ports of the processor io.outb . We configure the PICs using the ports:
Master PIC: 0x20 and 0x21
Slave PIC: 0xA0 and 0xA1
For a PIC, there are 2 types of registries:
ICW (Initialization Command Word): reinit the controller
OCW (Operation Control Word): configure the controller once initialized (used to
mask/unmask the interrupts)

28

IDT and interrupts

void init_pic(void)
{
/* Initialization of ICW1 */
io.outb(0x20, 0x11);
io.outb(0xA0, 0x11);
/* Initialization of ICW2 */
io.outb(0x21, 0x20); /* start vector = 32 */
io.outb(0xA1, 0x70); /* start vector = 96 */
/* Initialization of ICW3 */
io.outb(0x21, 0x04);
io.outb(0xA1, 0x02);
/* Initialization of ICW4 */
io.outb(0x21, 0x01);
io.outb(0xA1, 0x01);
/* mask interrupts */
io.outb(0x21, 0x0);
io.outb(0xA1, 0x0);
}

PIC ICW configurations details

The registries have to be configured in order.
ICW1 (port 0x20 / port 0xA0)
|0|0|0|1|x|0|x|x|
| | +--- with ICW4 (1) or without (0)
| +----- one controller (1), or cascade (0)
+--------- triggering by level (level) (1) or by edge (edge) (0)

ICW2 (port 0x21 / port 0xA1)

|x|x|x|x|x|0|0|0|
| | | | |
+----------------- base address for interrupts vectors

ICW2 (port 0x21 / port 0xA1)

For the master:

29

IDT and interrupts

|x|x|x|x|x|x|x|x|
| | | | | | | |
+------------------ slave controller connected to the port yes (1), or no (0)

For the slave:

|0|0|0|0|0|x|x|x| pour l'esclave
| | |
+-------- Slave ID which is equal to the master port

ICW4 (port 0x21 / port 0xA1)

It is used to define in which mode the controller should work.
|0|0|0|x|x|x|x|1|
| | | +------ mode "automatic end of interrupt" AEOI (1)
| | +-------- mode buffered slave (0) or master (1)
| +---------- mode buffered (1)
+------------ mode "fully nested" (1)

Why do idt segments offset our ASM functions?

You should have noticed that when I'm initializing our IDT segments, I'm using offsets to
segment the code in Assembly. The different functions are defined in x86int.asm and are of
the following scheme:

30

IDT and interrupts

%macro SAVE_REGS 0
pushad
push ds
push es
push fs
push gs
push ebx
mov bx,0x10
mov ds,bx
pop ebx
%endmacro
%macro RESTORE_REGS 0
pop gs
pop fs
pop es
pop ds
popad
%endmacro
%macro INTERRUPT 1
global _asm_int_%1
_asm_int_%1:
SAVE_REGS
push %1
call isr_default_int
pop eax ;;a enlever sinon
mov al,0x20
out 0x20,al
RESTORE_REGS
iret
%endmacro

These macros will be used to define the interrupt segment that will prevent corruption of the
different registries, it will be very useful for multitasking.

31

Theory: physical and virtual memory

Chapter 8: Theory: physical and virtual

memory
In the chapter related to the GDT, we saw that using segmentation a physical memory
address is calculated using a segment selector and an offset.
In this chapter, we are going to implement paging, paging will translate a linear address from
segmentation into a physical address.

Why do we need paging?

Paging will allow our kernel to:
use the hard-drive as a memory and not be limited by the machine ram memory limit
to have a unique memory space for each process
to allow and unallow memory space in a dynamic way
In a paged system, each process may execute in its own 4gb area of memory, without any
chance of effecting any other process's memory, or the kernel's. It simplifies multitasking.

How does it work?

The translation of a linear address to a physical address is done in multiple steps:
1. The processor use the registry CR3 to know the physical address of the pages
directory.
2. The first 10 bits of the linear address represent an offset (between 0 and 1023), pointing
to an entry in the pages directory. This entry contains the physical address of a pages
table.
3. the next 10 bits of the linear address represent an offset, pointing to an entry in the
pages table. This entry is pointing to a 4ko page.
4. The last 12 bits of the linear address represent an offset (between 0 and 4095), which
indicates the position in the 4ko page.

32

Theory: physical and virtual memory

Format for pages table and directory

The two types of entries (table and directory) look like the same. Only the field in gray will be
used in our OS.

P : indicate if the page or table is in physical memory

R/W : indicate if the page or table is accessible in writting (equals 1)
U/S : equals 1 to allow access to non-preferred tasks
A : indicate if the page or table was accessed
D : (only for pages table) indicate if the page was written
PS (only for pages directory) indicate the size of pages:

0 = 4kb
1 = 4mb
Note: Physical addresses in the pages diretcory or pages table are written using 20 bits
because these addresses are aligned on 4kb, so the last 12bits should be equal to 0.
A pages directory or pages table used 1024*4 = 4096 bytes = 4k

33

Theory: physical and virtual memory

A pages table can address 1024 * 4k = 4 Mb

A pages directory can address 1024 (1024 4k) = 4 Gb

How to enable pagination?

To enable pagination, we just need to set bit 31 of the CR0 registry to 1:
asm(" mov %%cr0, %%eax; \
or %1, %%eax; \
mov %%eax, %%cr0" \
:: "i"(0x80000000));

But before, we need to initialize our pages directory with at least one pages table.

Identity Mapping
With the identity mapping model, the page will apply only to the kernel as the first 4 MB of
virtual memory coincide with the first 4 MB of physical memory:

This model is simple: the first virtual memory page coincide to the first page in physical
memory, the second page coincide to the second page on physical memory and so on ...

34

Memory management: physical and virtual

Memory management: physical and virtual

The kernel knows the size of the physical memory available thanks to GRUB.
In our implementation, the first 8 megabytes of physical memory will be reserved for use by
the kernel and will contain:
The kernel
GDT, IDT et TSS
Kernel Stack
Some space reserved to hardware (video memory, ...)
Page directory and pages table for the kernel
The rest of the physical memory is freely available to the kernel and applications.

Virtual Memory Mapping

The address space between the beginning of memory and 0x40000000 address is the kernel
space, while the space between the address 0x40000000 and the end of the memory
corresponds to user space:

35

Memory management: physical and virtual

The kernel space in virtual memory, which is using 1Gb of virtual memory, is common to all
tasks (kernel and user).
This is implemented by pointing the first 256 entries of the task page directory to the kernel
page directory (In vmm.cc):
/*
* Kernel Space. v_addr < USER_OFFSET are addressed by the kernel pages table
*/
for (i=0; i<256; i++)
pdir[i] = pd0[i];

36