X86 Assembly/Bootloaders

When a computer is turned on, there is some beeping, and some flashing lights, and then a loading screen appears. And then magically, the operating system loads into memory. The question is then raised, how does the operating system load up? What gets the ball rolling? The answer is bootloaders.

What is a Bootloader?
Bootloaders are small pieces of software that play a role in getting an operating system loaded and ready for execution when a computer is turned on. The way this happens varies between different computer designs (early computers required a person to manually set the computer up whenever it was turned on), and often there are several stages in the process of boot loading.

It's crucial to understand that the term "bootloader" is simply a classification of software (and sometimes a blurry one). To the processor, a bootloader is just another piece of code that it blindly executes. There are many different kinds of boot loaders. Some are small, others are large; some follow very simple rules while others show fancy screens and give the user a selection to choose from.

On IBM PC compatibles, the first program to load is the Basic Input/Output System (BIOS). The BIOS performs many tests and initialisations, and if everything is OK, the BIOS's boot loader begins. Its purpose is to load another boot loader! It selects a disk (or some other storage media) from which it loads a secondary boot loader.

In some cases, this boot loader loads enough of an operating system to start running it. In other cases, it loads yet another boot loader from somewhere else. This often happens when multiple operating systems are installed on a single computer; each OS may have its own specific bootloader, with a "central" bootloader that loads one of the specific ones according to the user's selection.

Most bootloaders are written exclusively in assembly language (or even machine code), because they need to be compact, they don't have access to OS routines (such as memory allocation) that other languages might require, they need to follow some unusual requirements, and they make frequent use of low-level features. However some bootloaders, particularly those that have many features and allow user input, are quite heavyweight. These are often written in a combination of assembly and C. The GRand Unified Bootloader (GRUB) is an example of such.

Some boot loaders are highly OS-specific, while others are less so - certainly the BIOS boot loader is not OS-specific. The MS-DOS boot loader (which was placed on all MS-DOS formatted floppy disks) simply checks if the files IO.SYS and MSDOS.SYS exist; if they are not present it displays the error "Non-System disk or disk error" otherwise it loads and begins execution of IO.SYS.

The final stage boot loader may be expected (by the OS) to prepare the computer in some way, for instance by placing the processor in protected mode and programming the interrupt controller. While it would be possible to do these things inside the OS's initialisation procedure, moving them into the bootloader can simplify the OS design. Some operating systems require their bootloader to set up a small, basic GDT (Global Descriptor Table) and enter protected mode, in order to remove the need for the OS to have any 16-bit code. However, the OS might replace this with its own sophisticated GDT soon after.

The Bootsector
The first 512 bytes of a disk are known as the bootsector or Master Boot Record. The boot sector is an area of the disk reserved for booting purposes. If the bootsector of a disk contains a valid boot sector (the last word of the sector must contain the signature 0xAA55), then the disk is treated by the BIOS as bootable.

The Boot Process
When switched on or reset, an x86 processor begins executing the instructions it finds at address FFFF:0000 (at this stage it is operating in Real Mode) (Intel Software Developer's Manual Volume 3 Chapter 9 contradicts this information: Execution starts at the physical address 0xFFFFFFF0, among other things). In IBM PC compatible processors, this address is mapped to a ROM chip that contains the computer's Basic Input/Output System (BIOS) code. The BIOS is responsible for many tests and initialisations; for instance the BIOS may perform a memory test, initialise the interrupt controller and system timer, and test that these devices are working.

Eventually the actual boot loading begins. First the BIOS searches for and initialises available storage media (such as floppy drives, hard disks, CD drives), then it decides which of these it will attempt to boot from. It checks each device for availability (e.g. ensuring a floppy drive contains a disk), then the 0xAA55 signature, in some predefined order (often the order is configurable using the BIOS setup tool). It loads the first sector of the first bootable device it comes across into RAM, and initiates execution.

Ideally, this will be another boot loader, and it will continue the job, making a few preparations, then passing control to something else.

While BIOSes remain compatible with 20-year-old software, they have also become more sophisticated over time. Early BIOSes could not boot from CD drives, but now CD and even DVD booting are standard BIOS features. Booting from USB storage devices is also possible, and some systems can boot from over the network. To achieve such advanced functioning, BIOSes sometimes enter protected mode and the like, but then return to real mode in order to be compatible with legacy boot loaders. This creates a chicken-and-egg problem: bootloaders are written to work with the ubiquitous BIOS, and BIOSes are written to support all those bootloaders, preventing much in the way of new boot loading features.

However, a new bootstrap technology, the UEFI, is beginning to gain momentum. It is much more sophisticated and will not be discussed in this article.

Note also that other computer systems - even some that use x86 processors - may boot in different ways. Indeed, some embedded systems whose software is compact enough to be stored on ROM chips may not need bootloaders at all.

Technical Details
A bootloader runs under certain conditions that the programmer must appreciate in order to make a successful bootloader. The following pertains to bootloaders initiated by the PC BIOS:


 * 1) The first sector of a drive contains its boot loader.
 * 2) One sector is 512 bytes — the last two bytes of which must be 0xAA55 (i.e. 0x55 followed by 0xAA), or else the BIOS will treat the drive as unbootable.
 * 3) If everything is in order, said first sector will be placed at RAM address 0000:7C00, and the BIOS's role is over as it transfers control to 0000:7C00 (that is, it JMPs to that address).
 * 4) The DL register will contain the drive number that is being booted from, useful if you want to read more data from elsewhere on the drive.
 * 5) The BIOS leaves behind a lot of code, both to handle hardware interrupts (such as a keypress) and to provide services to the bootloader and OS (such as keyboard input, disk read, and writing to the screen). You must understand the purpose of the Interrupt Vector Table (IVT), and be careful not to interfere with the parts of the BIOS that you depend on. Most operating systems replace the BIOS code with their own code, but the boot loader can't use anything but its own code and what the BIOS provides. Useful BIOS services include  (for displaying text/graphics),   (disk functions) and   (keyboard input).
 * 6) This means that any code or data that the boot loader needs must either be included in the first sector (be careful not to accidentally execute data) or manually loaded from another sector of the disk to somewhere in RAM. Because the OS is not running yet, most of the RAM will be unused. However, you must take care not to interfere with the RAM that is required by the BIOS interrupt handlers and services mentioned above.
 * 7) The OS code itself (or the next bootloader) will need to be loaded into RAM as well.
 * 8) The BIOS places the stack pointer 512 bytes beyond the end of the boot sector, meaning that the stack cannot exceed 512 bytes. It may be necessary to move the stack to a larger area.
 * 9) There are some conventions that need to be respected if the disk is to be readable under mainstream operating systems. For instance you may wish to include a BIOS Parameter Block on a floppy disk to render the disk readable under most PC operating systems.

Most assemblers will have a command or directive similar to  that informs the assembler that the code will be loaded starting at offset 7C00h. The assembler will take this into account when calculating instruction and data addresses. If you leave this out, the assembler assumes the code is loaded at address 0 and this must be compensated for manually in the code.

Usually, the bootloader will load the kernel into memory, and then jump to the kernel. The kernel will then be able to reclaim the memory used by the bootloader (because it has already performed its job). However it is possible to include OS code within the boot sector and keep it resident after the OS begins.

Here is a simple bootloader demo designed for NASM:

To compile the above file, suppose it is called 'floppy.asm', you can use following command: nasm -f bin -o floppy.img floppy.asm While strictly speaking this is not a bootloader, it is bootable, and demonstrates several things:
 * How to include and access data in the boot sector
 * How to skip over included data (this is required for a BIOS Parameter Block)
 * How to place the 0xAA55 signature at the end of the sector (NASM will issue an error if there is too much code to fit in a sector)
 * The use of BIOS interrupts

On Linux, you can issue a command like cat floppy.img > /dev/fd0 to write the image to the floppy disk (the image may be smaller than the size of the disk in which case only as much information as is in the image will be written to the disk). A more sophisticated option is to use the dd utility: dd if=floppy.img of=/dev/fd0 Under Windows you can use software such as RAWRITE.

Hard disks
Hard disks usually add an extra layer to this process, since they may be partitioned. The first sector of a hard disk is known as the Master Boot Record (MBR). Conventionally, the partition information for a hard disk is included at the end of the MBR, just before the 0xAA55 signature.

The role of the BIOS is no different to before: to read the first sector of the disk (that is, the MBR) into RAM, and transfer execution to the first byte of this sector. The BIOS is oblivious to partitioning schemes - all it checks for is the presence of the 0xAA55 signature.

While this means that one can use the MBR in any way one would like (for instance, omit or extend the partition table) this is seldom done. Despite the fact that the partition table design is very old and limited - it is limited to four partitions - virtually all operating systems for IBM PC compatibles assume that the MBR will be formatted like this. Therefore to break with convention is to render your disk inoperable except to operating systems specifically designed to use it.

In practice, the MBR usually contains a boot loader whose purpose is to load another boot loader - to be found at the start of one of the partitions. This is often a very simple program which finds the first partition marked Active, loads its first sector into RAM, and commences its execution. Since by convention the new boot loader is also loaded to address 7C00h, the old loader may need to relocate all or part of itself to a different location before doing this. Also, ES:SI is expected to contain the address in RAM of the partition table, and DL the boot drive number. Breaking such conventions may render a bootloader incompatible with other bootloaders.

However, many boot managers (software that enables the user to select a partition, and sometimes even kernel, to boot from) use custom MBR code which loads the remainder of the boot manager code from somewhere on disk, then provides the user with options on how to continue the bootstrap process. It is also possible for the boot manager to reside within a partition, in which case it must first be loaded by another boot loader.

Most boot managers support chain loading (that is, starting another boot loader via the usual first-sector-of-partition-to-address-7C00 process) and this is often used for systems such as DOS and Windows. However, some boot managers (notably GRUB) support the loading of a user-selected kernel image. This can be used with systems such as GNU/Linux and Solaris, allowing more flexibility in starting the system. The mechanism may differ somewhat from that of chain loading.

Clearly, the partition table presents a chicken-and-egg problem that is placing unreasonable limitations on partitioning schemes. One solution gaining momentum is the GUID Partition Table; it uses a dummy MBR partition table so that legacy operating systems will not interfere with the GPT, while newer operating systems can take advantage of the many improvements offered by the system.

GNU GRUB
The GRand Unified Bootloader supports the flexible multiboot boot protocol. This protocol aims to simplify the boot process by providing a single, flexible protocol for booting a variety of operating systems. Many free operating systems can be booted using multiboot.

GRUB is extremely powerful and is practically a small operating system. It can read various file systems and thus lets you specify a kernel image by filename as well as separate module files that the kernel may make use of. Command-line arguments can be passed to the kernel as well - this is a nice way of starting an OS in maintenance mode, or "safe mode", or with VGA graphics, and so on. GRUB can provide a menu for the user to select from as well as allowing custom loading parameters to be entered.

Obviously this functionality cannot possibly be provided in 512 bytes of code. This is why GRUB is split into two or three "stages":
 * Stage 1 - this is a 512-byte block that has the location of stage 1.5 or stage 2 hardcoded into it. It loads the next stage.
 * Stage 1.5 - an optional stage which understands the filesystem (e.g. FAT32 or ext3) where stage 2 resides. It will find out where stage 2 is located and load it. This stage is quite small and is located in a fixed area, often just after Stage 1.
 * Stage 2 - this is a much larger image that has all the GRUB functionality.

Note that Stage 1 may be installed to the Master Boot Record of a hard disk, or may be installed in one of the partitions and chainloaded by another boot loader.

Windows can not be loaded using multiboot, but the Windows bootloader (like those of other non-multiboot operating systems) can be chainloaded from GRUB, which isn't quite as good, but does let you boot such systems.

Testing the Bootloader
Perhaps the easiest way to test a bootloader is inside a virtual machine, like VirtualBox or VMware.

Sometimes it is useful if the bootloader supports the GDB remote debug protocol.