How to write a simple operating system

How to write a simple operating system

PC primer
If you're writing an OS for x86 PCs (the best choice, due to the huge amount of documentation available), you'll need to understand the basics of how a PC starts up. Fortunately, you don't need to dwell on complicated subjects such as graphics drivers and network protocols, as you'll be focusing on the essential parts first.

When a PC is powered-up, it starts executing the BIOS (Basic Input/Output System), which is essentially a mini-OS built into the system. It performs a few hardware tests (eg memory checks) and typically spurts out a graphic (eg Dell logo) or diagnostic text to the screen. Then, when it's done, it starts to load your operating system from any media it can find. Many PCs jump to the hard drive and start executing code they find in the Master Boot Record (MBR), a 512-byte section at the start of the hard drive; some try to find executable code on a floppy disk (boot sector) or CD-ROM.

This all depends on the boot order - you can normally specify it in the BIOS options screen. The BIOS loads 512 bytes from the chosen media into its memory, and begins executing it. This is the bootloader, the small program that then loads the main OS kernel or a larger boot program (eg GRUB/LILO for Linux systems). This 512 byte bootloader has two special numbers at the end to tell the OS that it's a boot sector - we'll cover that later.

Note that PCs have an interesting feature for booting. Historically, most PCs had a floppy drive, so the BIOS was configured to boot from that device. Today, however, many PCs don't have a floppy drive - only a CD-ROM - so a hack was developed to cater for this. When you're booting from a CD-ROM, it can emulate a floppy disk; the BIOS reads the CD-ROM drive, loads in a chunk of data, and executes it as if it was a floppy disk. This is incredibly useful for us OS developers, as we can make floppy disk versions of our OS, but still boot it on CD-only machines. (Floppy disks are really easy to work with, whereas CD-ROM filesystems are much more complicated.)

Assembly language primer
Like most programming languages, assembly is a list of instructions followed in order. You can jump around between various places and set up subroutines/functions, but it's much more minimal than C# and friends. You can't just print "Hello world" to the screen - the CPU has no concept of what a screen is! Instead, you work with memory, manipulating chunks of RAM, performing arithmetic on them and putting the results in the right place. Sounds scary? It's a bit alien at first, but it's not hard to grasp.

At the assembly language level, there is no such thing as variables in the high-level language sense. What you do have, however, is a set of registers, which are on-CPU memory stores. You can put numbers into these registers and perform calculations on them. In 16-bit mode, these registers can hold numbers between 0 and 65535. Here's a list of the fundamental registers on a typical x86 CPU:

AX, BX, CX, DX	General-purpose registers for storing numbers that you're using. For instance, you may use AX to store the character that has been pressed on the keyboard, while using CX to act as a counter in a loop. (Note: these 16-bit registers can be split into 8-bit registers such as AH/AL, BH/BL etc.)
SI, DI	Source and destination data index registers. These point to places in memory for retrieving and storing data.
SP	The Stack Pointer (explained in a moment).
IP (sometimes CP)	The Instruction/Code Pointer. This contains the location in memory of the instruction being executed. When an instruction has finished, it is incremented and moves on to the next instruction. You can change the contents of this register to move around in your code.

So you can use these registers to store numbers as you work - a bit like variables, but they're much more fixed in size and purpose. There are a few others, notably segment registers. Due to limitations in old PCs, memory was handled in 64K chunks called segments. This is a really messy subject, but thankfully you don't have to worry about it - for the time being, your OS will be less than a kilobyte anyway! In MikeOS, we limit ourselves to a single 64K segment so that we don't have to mess around with segment registers.

The stack is an area of your main RAM used for storing temporary information. It's called a stack because numbers are stacked one-on-top of another. Imagine a Pringles tube: if you put in a playing card, an iPod Shuffle and a beermat, you'll pull them out in the reverse order (beermat, then iPod, and finally playing card). It's the same with numbers: if you push the numbers 5, 7 and 15 onto the stack, you will pop them out as 15 first, then 7, and lastly 5. In assembly, you can push registers onto the stack and pop them out later - it's useful when you want to store temporarily the value of a register while you use that register for something else.

PC memory can be viewed as a linear line of pigeon-holes ranging from byte 0 to whatever you have installed (millions of bytes on modern machines). At byte number 53,634,246 in your RAM, for instance, you may have your web browser code to view this document. But whereas we humans count in powers of 10 (10, 100, 1000 etc. - decimal), computers are better off with powers of two (because they're based on binary). So we use hexadecimal, which is base 16, as a way of representing numbers. See this chart to understand:
Decimal 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20
Hexadecimal 0 1 2 3 4 5 6 7 8 9 A B C D E F 10 11 12 13 14

As you can see, whereas our normal decimal system uses 0 - 9, hexadecimal uses 0 - F in counting. It's a bit weird at first, but you'll get the hang of it. In assembly programming, we identify hexadecimal (hex) numbers by tagging a 'h' onto the end - so 0Ah is hex for the number 10. (You can also denote hexadecimal in assembly by prefixing the number with 0x - for instance, 0x0A.)

Let's finish off with a few common assembly instructions. These move memory around, compare them and perform calculations. They're the building blocks of your OS - there are hundreds of instructions, but you don't have to memorise them all, because the most important handful are used 90% of the time.
mov Copies memory from one location or register to another. For instance, mov ax, 30 places the number 30 into the AX register. Using square brackets, you can get the number at the memory location pointed to by the register. For instance, if BX contains 80, then mov ax, [bx] means "get the number in memory location 80, and put it into AX". You can move numbers between registers too: mov bx, cx.
add / sub Adds a number to a register. add ax, FFh adds FF in hexadecimal (255 in our normal decimal) to the AX register. You can use sub in the same way: sub dx, 50.
cmp Compares a register with a number. cmp cx, 12 compares the CX register with the number 12. It then updates a special register on the CPU called FLAGS - a special register that contains information about the last operation. In this case, if the number 12 is bigger than the value in CX, it generates a negative result, and notes that negative in the FLAGS register. We can use this in the following instructions...
jmp / jg / jl... Jump to a different part of the code. jmp label jumps (GOTOs) to the part of our source code where we have label: written. But there's more - you can jump conditionally, based on the CPU flags set in the previous command. For instance, if a cmp instruction determined that a register held a smaller value than the one with which it was compared, you can act on that with jl label (jump if less-than to label). Similarly, jge label jumps to 'label' in the code if the value in the cmp was greater-than or equal to its compared number.
int Interrupt the program and jump to a specified place in memory. Operating systems set up interrupts which are analogous to subroutines in high-level languages. For instance, in MS-DOS, the 21h interrupt provides DOS services (eg as opening a file). Typically, you put a value in the AX register, then call an interrupt and wait for a result (passed back in a register too). When you're writing an OS from scratch, you can call the BIOS with int 10h, int 13h, int 14h or int 16h to perform tasks like printing strings, reading sectors from a floppy disk etc.