Briefly, a boot loader is the first software program that runs when a computer starts. It is responsible for loading and transferring control to an operating system kernel software (such as the Linux or GNU Hurd kernel). The kernel, in turn, initializes the rest of the operating system (e.g. a GNU system).
GNU GRUB is a very powerful boot loader, which can load a wide variety of free operating systems, as well as proprietary operating systems with chain-loading(1). GRUB is designed to address the complexity of booting a personal computer; both the program and this manual are tightly bound to that computer platform, although porting to other platforms may be addressed in the future.
One of the important features in GRUB is flexibility; GRUB understands filesystems and kernel executable formats, so you can load an arbitrary operating system the way you like, without recording the physical position of your kernel on the disk.
Thus you can load the kernel just by specifying its filename and the drive (and the partition) where the kernel resides. To let GRUB know the drive and the filename, you can either type in them manually via the command-line interface (see section The flexible command line interface), or use the nice menu interface (see section The simple menu interface) through which you can easily select which OS it boots. To allow you to customize the menu, GRUB will load a preexisting configuration file (see section Configuration). Note that you can not only enter the command-line interface whenever you like, but also you can edit specific menu entries prior to using them.
In the following chapters, you will learn how to specify a drive or a partition, and a file name (see section Naming convention) to GRUB, how to install GRUB on your drive (see section Installation), and how to boot your OSes (see section Booting), step by step.
The device syntax used in GRUB is a wee bit different from what you may have seen before in your operating system(s), and you need to know it so that you can specify a drive/partition.
Look at the following examples and explanations:
(fd0)
First of all, GRUB requires that the device name is enclosed with `(' and `)'. The `fd' part means that it is a floppy disk. The number `0' is the drive number, which is counted from zero. This expression means that GRUB will use the whole floppy disk.
(hd0,1)
Here, `hd' means it is a hard disk drive. The first integer `0' indicates the drive number, that is, the first hard disk, while the second integer, `1', indicates the partition number (or the PC slice number in the BSD terminology). Once again, please note that the partition numbers are counted from zero, not from one. This expression means the second partition of the first hard disk drive. In this case, GRUB uses one partition of the disk, instead of the whole disk.
(hd0,4)
This specifies the first extended partition of the first hard disk drive. Note that the partition numbers for extended partitions are counted from `4', regardless of the actual number of primary partitions on your hard disk.
(hd1,a)
This means the BSD `a' partition of the second hard disk. If you need to specify which PC slice number should be used, use something like this: `(hd1,0,a)'. If the PC slice number is omitted, GRUB searches for the first PC slice which has a BSD `a' partition.
Of course, to actually access the disks or partitions with GRUB, you need to use the device specification in a command, like `root (fd0)' or `unhide (hd0,2)'. To help you find out which number is a partition you want, the GRUB command-line (see section The flexible command line interface) options have argument completion. That means that, for example, you only need to type `root (', followed by a TAB, and GRUB will display the list of drives, partitions, or filenames, so it should be quite easy to determine the name of your target partition, even with minimal knowledge of the syntax.
Note that GRUB does not distinguish IDE from SCSI - it simply counts the drive numbers from zero, regardless of their type. Normally, any IDE drive number is less than any SCSI drive number, although that is not true if you change the boot sequence by swapping IDE and SCSI drives in your BIOS.
Now the question is, how to specify a file? Again, see this example:
(hd0,0)/vmlinuz
This specifies the file named `vmlinuz', found on the first partition of the first hard disk drive. Note that the argument completion works with file names, too.
That was easy, admit it. Do read the next chapter, to find out how to actually install GRUB on your drive.
First, you need to have GRUB itself properly installed on your system, (see section How to obtain and build GRUB) either from the source tarball, or as a package for your OS.
To use GRUB, you need to install it on your drive. There are two ways of doing that - either using the utility @command{grub-install} (see section Invoking grub-install) on a UNIX-like OS, or by using the native Stage 2. These are quite similar, however, the utility might probe a wrong BIOS drive, so better be careful.
Also, if you install GRUB on a UNIX-like OS, please make sure that you have an emergency boot disk ready, so that you can rescue your computer if, by any chance, your hard drive becomes unusable (unbootable).
GRUB comes with boot images, which are normally installed in the `/usr/share/grub/i386-pc' directory. You need to copy the files `stage1', `stage2', and `*stage1_5' to the directory `/boot/grub'.
To create a GRUB boot floppy, you need to take the files `stage1' and `stage2' from `/boot/grub' directory, and write them to the first and the second block of the floppy disk, respectively.
Caution: This procedure will destroy any data currently stored on the floppy.
On a UNIX-like operating system, that is done with the following commands:
# cd /boot/grub # dd if=stage1 of=/dev/fd0 bs=512 count=1 1+0 records in 1+0 records out # dd if=stage2 of=/dev/fd0 bs=512 seek=1 153+1 records in 153+1 records out #
The device filename may be different. Consult the manual for your OS.
Caution: Installing GRUB's stage1 in this manner will erase the normal boot-sector used by an OS.
GRUB can currently boot GNU Mach, Linux, FreeBSD, NetBSD, and OpenBSD directly, so using it on a boot sector should be okay. But generally, it would be a good idea to back up the first sector of the partition on which you are installing GRUB's stage1. This isn't as important if you are installing GRUB on the first sector of a hard disk, since it's easy to reinitialize it (e.g. by running `FDISK /MBR' from DOS).
If you decide to install GRUB in the native environment, which is definitely desirable, you'll need to create the GRUB boot disk, and reboot your computer with it. Otherwise, see section Installing GRUB using grub-install, for more details.
Once started, GRUB will show the command-line interface (see section The flexible command line interface). First, set the GRUB's root device(2).} to the partition which has your GRUB images, like this:
grub> root (hd0,0)
If you are not sure which partition actually holds these files, use the @command{find} command, like this:
grub> find /boot/grub/stage1
This will search for the filename `/boot/grub/stage1' and show the devices which contain the file.
Once you've set the root device correctly, run the command @command{setup}:
grub> setup (hd0)
This command will install GRUB on the MBR (see section The structure of Master Boot Record) in the first drive. If you want to install GRUB into the boot sector of a partition instead of the MBR, specify a partition into which you want to install GRUB:
grub> setup (hd0,0)
If you install GRUB into a partition or a drive other than the first one, you must chain-load GRUB from another boot loader. Refer to the manual for the boot loader to know how to chain-load GRUB.
Now you can boot GRUB without a GRUB floppy. See the chapter section Booting to find out how to boot your operating systems from GRUB.
Unfortunately, if you do want to install GRUB under a UNIX-like OS (such as GNU), invoke the program @command{grub-install} as the superuser (root).
The usage is basically very easy. You only need to specify one argument to the program, namely, where to install GRUB. The argument can be either of a device file or a GRUB's drive/partition. So, this will install GRUB into the MBR of the first IDE disk under Linux:
# grub-install /dev/hda
Likewise, under Hurd, this has the same effect:
# grub-install /dev/hd0
If it is the first BIOS drive, this is the same as well:
# grub-install '(hd0)'
But all the above examples assume that you use GRUB images under the root directory. If you want GRUB to use images under a directory other than the root directory, you need to specify the option @option{--root-directory}. The typical usage is that you create a GRUB boot floppy with a filesystem. Here is an example:
# mke2fs /dev/fd0 # mount -t ext2 /dev/fd0 /mnt # grub-install --root-directory=/mnt '(fd0)' # umount /mnt
Another example is in case that you have a separate boot partition which is mounted at `/boot'. Since GRUB is a boot loader, it doesn't know anything about mountpoints at all. Thus, you need to run @command{grub-install} like this:
# grub-install --root-directory=/boot /dev/hda
By the way, as you may already know, it is quite difficult to guess BIOS drives correctly under a UNIX-like OS. Thus, @command{grub-install} will prompt you to check if it could really guess the correct mappings, after the installation. The format is defined in section The map between BIOS drives and OS devices. Please be careful enough. If the output is wrong, it is unlikely that your computer can boot with no problem.
Note that @command{grub-install} is actually just a shell script and the real task is done by the grub shell @command{grub} (see section Invoking the grub shell). Therefore, you may run @command{grub} directly to install GRUB, without using @command{grub-install}. Don't do that, however, unless you are very familiar with the internals of GRUB. Installing a boot loader on a running OS may be dangerous.
For Multiboot-compliant kernels, GRUB can load them in a consistent way, but, for some free operating systems, you need to use some OS-specific magic.
GRUB has two distinct boot methods. One of the two is to load an operating system directly, and the other is to chain-load another boot loader which then will load an operating system actually. Generally speaking, the former is desirable, because you don't need to install or maintain other boot loaders and GRUB is flexible enough to load an operating system from an arbitrary disk/partition. However, the latter is sometimes required, since GRUB doesn't support all the existing operating systems natively.
Multiboot (see section `Motivation' in The Multiboot Specification) is the native format supported by GRUB. For the sake of convenience, there are also support for Linux, FreeBSD, NetBSD and OpenBSD. If you want to boot other operating systems, you will have to chain-load them (see section Load another boot loader to boot unsupported operating systems).
Generally, GRUB can boot any Multiboot-compliant OS in the following steps:
Linux, FreeBSD, NetBSD and OpenBSD can be booted in a similar manner. You can load a kernel image by the command @command{kernel} and then run the command @command{boot}. If the kernel requires some parameters, just append the parameters to @command{kernel}, after the filename of the kernel. Also, please refer to section Some caveats on OS-specific issues, for the information on your OS-specific issues.
If you want to boot an unsupported operating system (e.g. Windows 95), chain-load a boot loader for the operating system. Normally, the boot loader is embedded in the boot sector of the partition on which the operating system is installed.
First, set GRUB's root device to the partition by the command @command{rootnoverify}:
grub> rootnoverify (hd0,0)
Second, set the active flag in the partition by the command @command{makeactive}(3):
grub> makeactive
Third, load the boot loader by the command @command{chainloader}:
grub> chainloader +1
`+1' indicates that GRUB should read one sector from the start of the partition. The complete description about this syntax can be found in section Filesystem syntax and semantics. If this succeeds, run the command @command{boot}.
However, DOS and Windows have a deficiency, so you might have to use more complicated instructions. See section Some caveats on OS-specific issues, for more information.
Here, we describe some caveats on several operating systems.
Since GNU/Hurd is Multiboot-compliant, it is easy to boot it; there is nothing special about it. But do not forget that you have to specify a root partition to the kernel.
find /boot/gnumach
or similar can help you.
grub> kernel /boot/gnumach root=hd0s1 grub> module /boot/serverboot
It is relatively easy to boot GNU/Linux from GRUB, because it somewhat resembles to boot a Multiboot-compliant OS.
find /vmlinuz
or similar can help you.
grub> kernel /vmlinuz root=/dev/hda1If you need to specify some kernel parameters, just append them to the command. For example, to set @option{vga} to `ext', do this:
grub> kernel /vmlinuz root=/dev/hda1 vga=extSee the documentation in the Linux source tree for the complete information on the available options.
grub> initrd /initrd
Caution: If you use an initrd and specify the `mem=' option to the kernel, to let it use less than actual memory size, you will also have to specify the same memory size to GRUB. To let GRUB know the size, run the command @command{uppermem} before loading the kernel. See section The list of command-line and menu entry commands, for more information.
GRUB can load the kernel directly, either in ELF or a.out format. But this is not recommended, since FreeBSD's bootstrap interface sometimes changes heavily, so GRUB can't guarantee to pass kernel parameters correctly.
Thus, we'd recommend loading the very flexible loader `/boot/loader' instead. See this example:
grub> root (hd0,a) grub> kernel /boot/loader grub> boot
GRUB can load NetBSD a.out and ELF directly, follow these steps:
grub> kernel --type=netbsd /netbsd-elf
For now, however, GRUB doesn't allow you to pass kernel parameters, so it may be better to chain-load it instead, for more information please see section Load another boot loader to boot unsupported operating systems.
The booting instruction is exactly the same as for NetBSD (see section NetBSD).
GRUB cannot boot DOS or Windows directly, so you must chain-load them (see section Load another boot loader to boot unsupported operating systems). However, their boot loaders have some critical deficiencies, so it may not work to just chain-load them. To overcome the problems, GRUB provides you with two helper functions.
If you have installed DOS (or Windows) on a non-first hard disk, you have to use the disk swapping technique, because that OS cannot boot from any disks but the first one. The workaround used in GRUB is the command @command{map} (see section The list of available commands), like this:
grub> map (hd0) (hd1) grub> map (hd1) (hd0)
This performs a virtual swap between your first and second hard drive.
Caution: This is effective only if DOS (or Windows) uses BIOS to access the swapped disks. If that OS uses a special driver for the disks, this probably won't work.
Another problem arises if you installed more than one set of DOS/Windows onto one disk, because they could be confused if there are more than one primary partitions for DOS/Windows. Certainly you should avoid doing this, but there is a solution if you do want to do so. Use the partition hiding/unhiding technique.
If GRUB hides a DOS (or Windows) partition, it ignores the partition. If GRUB unhides a DOS (or Windows) partition, it detects the partition. Thus, if you have installed DOS (or Windows) on the first partition and the second of the first hard disk, and boot the first copy, do the following:
grub> unhide (hd0,0) grub> hide (hd0,1) grub> rootnoverify (hd0,0) grub> chainloader +1 grub> makeactive grub> boot
It is known that the signature in the boot loader for SCO UnixWare is wrong, so you will have to specify the option @option{--force} to @command{chainloader}, like this:
grub> rootnoverify (hd1,0) grub> chainloader --force +1 grub> makeactive grub> boot
Although GRUB is a disk-based boot loader, it does provide some network support. To use the network support, you need to enable at least one network driver in the GRUB build process. For more information please see `netboot/README.netboot' in the source distribution.
GRUB requires a server that will assign an IP address to the machine on which GRUB is running, and a file server. The former is either BOOTP, DHCP or a RARP server (RARP is deprecated, since it cannot serve much information). For the latter, only TFTP is supported at the moment. It is not necessary to run both the servers on one computer. How to configure these servers is beyond the scope of this document, so please refer to the manuals specific to those protocols/servers.
Once you have set up the servers, run @command{bootp}, @command{dhcp} or @command{rarp} for BOOTP, DHCP or RARP, respectively. Each command will show an assigned IP address, a netmask, an IP address for your TFTP server and a gateway. If any of the addresses is wrong or it causes an error, probably the configuration of your servers isn't set up properly.
Finally, download your OS images from your network. The network can be accessed using the network drive `(nd)'. Everything else is very similar to the normal instructions (see section Booting).
Here is an example:
grub> bootp Probing... [NE*000] NE2000 base ... Address: 192.168.110.23 Netmask: 255.255.255.0 Server: 192.168.110.14 Gateway: 192.168.110.1 grub> root (nd) grub> kernel /tftproot/gnumach.gz root=sd0s1 grub> module /tftproot/serverboot.gz grub> boot
It is sometimes very useful to boot from a network, especially, when you use a machine which has no local disk. In this case, you need to obtain a kind of Net Boot ROM, such as a PXE ROM or a free software package like Etherboot. Such a Boot ROM first boots the machine, sets up the network card installed into the machine, and downloads a second stage boot image from the network. Then, the second image will try to boot an operating system from the network actually.
GRUB provides two second stage images, `nbgrub' and `pxegrub'. Those images are the same as the normal Stage 2, except that they set up a network automatically, and try to load a configuration file from the network, if specified. The usage is very simple: If the machine has a PXE ROM, use `pxegrub'. If the machine has a NBI loader such as Etherboot, use `nbgrub'. There is no difference between them but their formats. As how to load a second stage image you want to use should be described in the manual on your Net Boot ROM, please refer to the manual, for more details. The topic is beyond the scope of this documentation.
However, there is one thing specific to GRUB. Namely, how to specify a configuration file in a BOOTP/DHCP server. For now, GRUB uses the tag `150', to get the name of a configuration file. This below is an example about a BOOTP configuration:
allhost:hd=/tmp:bf=null:\ :ds=145.71.35.1 145.71.32.1:\ :sm=255.255.254.0:\ :gw=145.71.35.1:\ :sa=145.71.35.5: foo:ht=1:ha=63655d0334a7:ip=145.71.35.127:\ :bf=/nbgrub:\ :tc=allhost:\ :T150="/tftpboot/menu.lst.foo":
See the manual about your BOOTP/DHCP server, for more information. The exact syntax should differ from the example, more or less.
This chapter describes how to use the serial terminal support in GRUB.
If you have many computers or computers with no display/keyboard, it would be very useful to control the computers with serial communications. To connect a computer with another via a serial line, you need to prepare a null-modem (cross) serial cable, and you may need to have multiport serial boards, if your computer doesn't have extra serial ports. In addition, a terminal emulator is also required, such as minicom. Refer to a manual of your operating system, for more information.
As for GRUB, the instruction to set up a serial terminal is quite simple. First of all, make sure that you haven't specified the option @option{--disable-serial} to the configure script when you built your GRUB images. If you get them in binary form, probably they have serial terminal support already.
Then, initialize your serial terminal after GRUB starts up. Here is an example:
grub> serial --unit=0 --speed=9600 grub> terminal serial
The command @command{serial} initializes the serial unit 0 with the speed 9600bps. The serial unit 0 is usually called `COM1', so, if you want to use COM2, you must specify `--unit=1' instead. This command accepts many other options, so please refer to section The list of available commands, for more details.
The command @command{terminal} chooses which type of terminal you want
to use. In that case above, the terminal will be a serial terminal, but
you can also pass console
to the command, like `terminal
serial console'. In this case, a terminal in which you press any key
will be selected as a GRUB terminal.
However, note that GRUB assumes that your terminal emulator is compatible with VT100 by default. This is true for most terminal emulators nowadays, but you should pass the option @option{--dumb} to the command, if your terminal emulator is not VT100-compatible or implement few VT100 escape sequences. If you specify the option, then GRUB doesn't provide you with the menu interface, because the menu requires several fancy features for your terminal. Instead, GRUB only gives you the hidden menu interface and the command-line interface.
You probably noticed that you need to type several commands to boot your OS. There's a solution to that - GRUB provides a menu interface (see section The simple menu interface) from which you can select an item (using arrow keys) that will do everything to boot an OS.
To enable the menu, you need a configuration file, `/boot/grub/menu.lst'. We'll analyze an example file.
The file first contains some general settings, the menu interface related options. You can put these commands (see section The list of commands for the menu only) before any of the items (starting with @command{title}).
# # Sample boot menu configuration file #
As you may have guessed, these lines are comments. Lines starting with a hash character (`#'), and blank lines, are ignored by GRUB.
# By default, boot the first entry. default 0
The first entry (here, counting starts with number zero, not one!) will be the default choice.
# Boot automatically after 30 secs. timeout 30
As the comment says, GRUB will boot automatically in 30 seconds, unless interrupted with a keypress.
# Fallback to the second entry. fallback 1
If, for any reason, the default entry doesn't work, fall back to the second one (this is rarely used, for obvious reasons).
Note that the complete descriptions of these commands, which are menu interface specific, can be found in see section The list of commands for the menu only. Other descriptions can be found in section The list of available commands.
Now, on to the actual OS definitions. You will see that each entry begins with a special command, @command{title}, and the action is described after it. Note that there is no command @command{boot} at the end of each item. That is because GRUB automatically executes @command{boot} if it loads other commands successfully.
The argument for the command @command{title} is used to display a short title/description of the entry in the menu. Since @command{title} displays the argument as is, you can write basically anything in there.
# For booting the GNU Hurd title GNU/Hurd root (hd0,0) kernel /boot/gnumach.gz root=hd0s1 module /boot/serverboot.gz
This boots GNU/Hurd from the first hard disk.
# For booting Linux title GNU/Linux kernel (hd1,0)/vmlinuz root=/dev/hdb1
This boots GNU/Linux, but from the second hard disk.
# For booting Mach (getting kernel from floppy) title Utah Mach4 multiboot root (hd0,2) pause Insert the diskette now^G!! kernel (fd0)/boot/kernel root=hd0s3 module (fd0)/boot/bootstrap
This boots Mach with a kernel on a floppy, but the root filesystem at hd0s3. It also contains a @command{pause} line, which will cause GRUB to display a prompt and delay, before actually executing the rest of the commands and booting.
# For booting FreeBSD title FreeBSD root (hd0,2,a) kernel /boot/loader
This item will boot FreeBSD kernel loaded from the `a' partition of the third PC slice of the first hard disk.
# For booting OS/2 title OS/2 root (hd0,1) makeactive # chainload OS/2 bootloader from the first sector chainloader +1 # This is similar to "chainload", but loads a specific file #chainloader /boot/chain.os2
This will boot OS/2, using a chain-loader.
# For booting Windows NT or Windows95 title Windows NT / Windows 95 boot menu root (hd0,0) makeactive chainloader +1 # For loading DOS if Windows NT is installed # chainload /bootsect.dos
The same as the above, but for Windows.
# For installing GRUB into the hard disk title Install GRUB into the hard disk root (hd0,0) setup (hd0)
This will just (re)install GRUB onto the hard disk.
# Change the colors. title Change the colors color light-green/brown blink-red/blue
In the last entry, the command @command{color} is used, to change the menu colors (try it!). This command is somewhat special, because it can be used both in the command-line and in the menu (see section The list of command-line and menu commands). GRUB has several such commands, see the user reference.
We hope that you now understand how to use the basic features of GRUB. To learn more about GRUB, see the User Reference Manual (see section Introduction).
Have fun!
This part documents the user-visible aspect of GRUB. If you are looking for the information on the internals, see the Programmer Reference Manual (see section Implementation details).
GRUB originated in 1995 when Erich Boleyn was trying to boot the GNU Hurd with the University of Utah's Mach 4 microkernel (now known as GNU Mach). Erich and Brian Ford designed the Multiboot Specification (see section `Motivation' in The Multiboot Specification), because they were determined not to add to the large number of mutually-incompatible PC boot methods.
Erich then began modifying the FreeBSD boot loader so that it would understand Multiboot. He soon realized that it would be a lot easier to write his own boot loader from scratch than to keep working on the FreeBSD boot loader, and so GRUB was born.
Erich added many features to GRUB, but other priorities prevented him from keeping up with the demands of its quickly-expanding user base. In 1999, Gordon Matzigkeit and OKUJI Yoshinori adopted GRUB as an official GNU package, and opened its development by making the latest sources available via anonymous CVS. See section How to obtain and build GRUB, for more information.
The primary requirement for GRUB is that it be compliant with the Multiboot Specification, which is described in section `Motivation' in The Multiboot Specification.
The other goals, listed in approximate order of importance, are:
Except for specific compatibility modes (chain-loading and the Linux piggyback format), all kernels will be started in much the same state as in the Multiboot Specification. Only kernels loaded at 1 megabyte or above are presently supported. Any attempt to load below that boundary will simply result in immediate failure and an error message reporting the problem.
In addition to the requirements above, GRUB has the following features (note that the Multiboot Specification doesn't require all the features that GRUB supports):
Future directions might include an internal programming language for supporting richer sets of boot options with control statements (which would make GRUB its own kind of kernel). Support for non-PC hardware architectures is also planned.(5)
The following is a quotation from Gordon Matzigkeit, a GRUB fanatic:
Some people like to acknowledge both the operating system and kernel when they talk about their computers, so they might say they use "GNU/Linux" or "GNU/Hurd". Other people seem to think that the kernel is the most important part of the system, so they like to call their GNU operating systems "Linux systems."
I, personally, believe that this is a grave injustice, because the boot loader is the most important software of all. I used to refer to the above systems as either "LILO"(6) or "GRUB" systems.
Unfortunately, nobody ever understood what I was talking about; now I just use the word "GNU" as a pseudonym for GRUB.
So, if you ever hear people talking about their alleged "GNU" systems, remember that they are actually paying homage to the best boot loader around... GRUB!
We, the GRUB maintainers, do not (usually) encourage Gordon's level of fanaticism, but it helps to remember that boot loaders deserve recognition. We hope that you enjoy using GNU GRUB as much as we did writing it.
GRUB uses a special syntax for specifying disk drives which can be accessed by BIOS. Because of BIOS limitations, GRUB cannot distinguish between IDE, ESDI, SCSI, or others. You must know yourself which BIOS device is equivalent to which OS device. Normally, that will be clear if you see the files in a device or use the command @command{find} (see section The list of available commands).
The device syntax is like this:
(bios-device[,part-num][,bsd-subpart-letter])
`[]' means the parameter is optional. bios-device should be either `fd' or `hd' followed by a digit, like `fd0'. But you can also set bios-device to a hexadecimal or a decimal, which is a BIOS drive number, so the following are equivalent:
(hd0) (0x80) (128)
part-num represents the partition number of bios-device, starting from zero for primary partitions and from four for extended partitions, and bsd-subpart-letter represents the BSD disklabel subpartition, such as `a' or `e'.
A shortcut for specifying BSD subpartitions is
(bios-device,bsd-subpart-letter)
, in this case, GRUB
searches for the first PC partition containing a BSD disklabel, then
finds the subpartition bsd-subpart-letter. Here is an example:
(hd0,a)
The syntax like `(hd0)' represents using the entire disk (or the MBR when installing GRUB), while the syntax like `(hd0,0)' represents using the partition of the disk (or the boot sector of the partition when installing GRUB).
If you enabled the network support, the special drive, `(nd)', is also available. Before using the network drive, you must initialize the network. See section Downloading OS images from a network, for more information.
There are two ways to specify files, by absolute filename and by blocklist.
An absolute filename resembles a Unix absolute filename, using `/'
for the directory separator (not `\' as in DOS). One example is
`(hd0,0)/boot/grub/menu.lst'. This means the file
`/boot/grub/menu.lst' in the first partition of the first hard
disk. If you omit the device name in an absolute filename, GRUB uses
GRUB's root device implicitly. So if you set the root device to,
say, `(hd1,0)' by the command @command{root}, then
/boot/kernel
is the same as (hd1,0)/boot/kernel
.
A blocklist is used for specifying a file that doesn't appear in the
filesystem, like a chainloader. The syntax is
[offset]+length[,[offset]+length]...
.
Here is an example:
0+100,200+1,300+300
This represents that GRUB should read blocks 0 through 99, block 200, and blocks 300 through 599. If you omit an offset, then GRUB assumes the offset is zero.
Like the filename syntax (see section How to specify files), if a blocklist does
not contain a device name, then GRUB uses GRUB's root device. So
(hd0,1)+1
is the same as +1
when the root device is
`(hd0,1)'.
GRUB has both a simple menu interface for choosing preset entries from a configuration file, and a highly flexible command line for performing any desired combination of boot commands.
GRUB looks for its configuration file as soon as it is loaded. If one is found, then the full menu interface is activated using whatever entries were found in the file. If you choose the command line menu option, or if the configuration file was not found, then GRUB drops to the command line interface.
The command line interface provides a prompt and after it an editable text area much like a command line in Unix or DOS. Each command is immediately executed after it is entered (7). The commands (see section The list of available commands) are a subset of those available in the configuration file, used with exactly the same syntax.
Cursor movement and editing of the text on the line can be done via a subset of the functions available in the Bash shell:
When typing commands interactively, if the cursor is within or before the first word in the command-line, pressing the TAB key (or C-i) will display a listing of the available commands, and if the cursor is after the first word, the TAB will provide a completion listing of disks, partitions, and filenames depending on the context.
Note that you cannot use the completion functionality in the TFTP filesystem. This is because TFTP doesn't support filename listing for the security.
The menu interface is quite easy to use. Its commands are both reasonably intuitive and described on screen.
Basically, the menu interface provides a list of boot entries to the user to choose from. Use the arrow keys to select the entry of choice, then press RET to run it. An optional timeout is available to boot the default entry (the first one if not set), which is aborted by pressing any key.
Commands are available to enter a bare command line by pressing c (which operates exactly like the non-config-file version of GRUB, but allows one to return to the menu if desired by pressing ESC) or to edit any of the boot entries by pressing e.
The menu entry editor looks much like the main menu interface, but the lines in the menu are individual commands in the selected entry instead of entry names.
If an ESC is pressed in the editor, it aborts all the changes made to the configuration entry and returns to the main menu interface.
When a particular line is selected, the editor places the user at a special version of the GRUB command line to edit that line. When the user hits RET, GRUB replaces the line in question in the boot entry with the changes (unless it was aborted via ESC, in which case the changes are thrown away).
If you want to add a new line to the menu entry, press o if adding a line after the current line or press O if before the current line.
To delete a line, hit the key d. Although GRUB does not support undo unfortunately, you can do almost the same thing by just returning to the main menu.
In this chapter, we list all commands that are available in GRUB.
Commands belong to different groups. A few can only be used in the global section of the configuration file (or "menu"); most of them can be entered on the command line and can be either used in the menu or in the menu entries.
The semantics used in parsing the configuration file are the following:
These commands can only be used in the menu:
default
command). This obviously won't help if the machine was rebooted by a
kernel that GRUB loaded.
Commands usable both in the menu and in the command line.
foreground/background
. foreground and
background are symbolic color names. A symbolic color name must be
one of these:
But only the first eight names can be used for background. You can
prefix blink-
to foreground if you want a blinking
foreground color.
This command can be used in the configuration file and on the command line, so you may write something like this in your configuration file:
# Set default colors. color light-gray/blue black/light-gray # Change the colors. title OS-BS like color magenta/blue black/magenta
grub> device (fd0) /floppy-image grub> device (hd0) /dev/sd0
This command can be used only in the grub shell (see section Invoking the grub shell).
0-0xff
; from and
to are the starting and ending sectors, expressed as an absolute
sector number.
The serial port is not used as a communication channel unless the terminal command is used.
This command is only available if GRUB is compiled with serial support.
grub> setkey capslock control grub> setkey control capslock
A key must be an alphabet, a digit, or one of these symbols: `escape', `exclam', `at', `numbersign', `dollar', `percent', `caret', `ampersand', `asterisk', `parenleft', `parenright', `minus', `underscore', `equal', `plus', `backspace', `tab', `bracketleft', `braceleft', `bracketright', `braceright', `enter', `control', `semicolon', `colon', `quote', `doublequote', `backquote', `tilde', `shift', `backslash', `bar', `comma', `less', `period', `greater', `slash', `question', `alt', `space', `capslock', `FX' (`X' is a digit), and `delete'. This table describes to which character each of the symbols corresponds:
These commands are usable in the command line and in menu entries. If you forget a command, you can run the command @command{help}.
grub> cat /etc/fstab
Differ in size: 0x1234 [foo], 0x4321 [bar]
If the sizes are equal but the bytes at an offset differ, then print the bytes like this:
Differ at the offset 777: 0xbe [foo], 0xef [bar]
If they are completely identical, nothing will be printed.
/boot/grub/stage1
.
In slightly more detail, it will load stage1_file, validate that it is a GRUB Stage 1 of the right version number, install a blocklist for loading stage2_file as a Stage 2. If the option @option{d} is present, the Stage 1 will always look for the actual disk stage2_file was installed on, rather than using the booting drive. The Stage 2 will be loaded at address addr, which must be `0x8000' for a true Stage 2, and `0x2000' for a Stage 1.5. If addr is not present, GRUB will determine the address automatically. It then writes the completed Stage 1 to the first block of the device dest_dev. If the options @option{p} or config_file are present, then it reads the first block of stage2, modifies it with the values of the partition stage2_file was found on (for @option{p}) or places the string config_file into the area telling the stage2 where to look for a configuration file at boot time. Likewise, if real_config_file is present and stage2_file is a Stage 1.5, then the Stage 2 config_file is patched with the configuration filename real_config_file. This command preserves the DOS BPB (and for hard disks, the partition table) of the sector the Stage 1 is to be installed into.
Caution: Several buggy BIOSes don't pass a booting drive properly when booting from a hard disk drive. Therefore, you will have to specify the option @option{d}, whether your Stage2 resides at the booting drive or not, if you have such a BIOS unfortunately. We know these are defective in that:
Caution2: A number of BIOSes don't return a correct LBA support bitmap even if they do have the support. So GRUB provides a solution to ignore the wrong bitmap, that is, the option @option{--force-lba}. Don't use this option if you know that your BIOS doesn't have LBA support.
This command also accepts the option @option{--type} so that you can specify the kernel type of file explicitly. The argument type must be one of these: `netbsd', `freebsd', `openbsd', `linux', `biglinux', and `multiboot'. However, you need to specify it only if you want to load a NetBSD ELF kernel, because GRUB can automatically determine a kernel type in the other cases, quite safely.
This command is used in a menu, as shown in this this example:
title This entry is too dangerous to be executed by normal users lock root (hd0,a) kernel /no-security-os
grub> map (hd0) (hd1) grub> map (hd1) (hd0)
The example exchanges the order between the first hard disk and the second hard disk.
ES:ESI
, used by some chain-loaded boot loaders), the
BSD drive-type (for booting BSD kernels using their native boot format),
and correctly determine the PC partition where a BSD sub-partition is
located. The optional hdbias parameter is a number to tell a BSD
kernel how many BIOS drive numbers are on controllers before the current
one. For example, if there is an IDE disk and a SCSI disk, and your
FreeBSD root partition is on the SCSI disk, then use a `1' for
hdbias.
The option @option{--force-lba} is just passed to @command{install} if specified. See the description on @command{install}, for more information.
Caution: This should be used with great caution, and should only be necessary on some old machines. GRUB's BIOS probe can pick up all RAM on all new machines the author has ever heard of. It can also be used for debugging purposes to lie to an OS.
This chapter describes error messages reported by GRUB when you encounter trouble. See section Invoking the grub shell, if your problem is specific to the grub shell.
The general way that the Stage 1 handles errors is to print an error string and then halt. Pressing CTRL-ALT-DEL will reboot.
The following is a comprehensive list of error messages for the Stage 1:
The general way that the Stage 1.5 handles errors is to print an error
number in the form Error num
and then halt. Pressing
CTRL-ALT-DEL will reboot.
The error numbers correspond to the errors reported by Stage 2. See section Errors reported by the Stage 2.
The general way that the Stage 2 handles errors is to abort the operation in question, print an error string, then (if possible) either continue based on the fact that an error occurred or wait for the user to deal with the error.
The following is a comprehensive list of error messages for the Stage 2 (error numbers for the Stage 1.5 are listed before the colon in each description):
This chapter documents the grub shell @command{grub}. Note that the grub shell is an emulator; it doesn't run under the native environment, so it sometimes does something wrong. Therefore, you shouldn't trust it too much. If there is anything wrong with it, don't hesitate to try the native GRUB environment, especially when it guesses a wrong map between BIOS drives and OS devices.
You can use the command @command{grub} for installing GRUB under your operating systems and for a testbed when you add a new feature into GRUB or when fix a bug. @command{grub} is almost the same as the Stage 2, and, in fact, it shares the source code with the Stage 2 and you can use the same commands (see section The list of available commands) in @command{grub}. It is emulated by replacing BIOS calls with UNIX system calls and libc functions.
The command @command{grub} accepts the following options:
The installation procedure is the same as under the native Stage 2. See section Installation, for more information. The command @command{grub}-specific information is described here.
What you should be careful about is buffer cache. @command{grub} makes use of raw devices instead of filesystems that your operating systems serve, so there exists a potential problem that some cache inconsistency may corrupt your filesystems. What we recommend is:
In addition, enter the command @command{quit} when you finish the installation. That is very important because @command{quit} makes the buffer cache consistent. Do not push C-c.
If you want to install GRUB non-interactively, specify `--batch' option in the command line. This is a simple example:
#!/bin/sh # Use /usr/sbin/grub if you are on an older system. /sbin/grub --batch <<EOT 1>/dev/null 2>/dev/null root (hd0,0) setup (hd0) quit EOT
When you specify the option @option{--device-map} (see section Introduction into the grub shell), the grub shell creates the device map file automatically unless it already exists. The filename `/boot/grub/device.map' is preferred.
If the device map file exists, the grub shell reads it to map BIOS drives to OS devices. This file consists of lines like this:
device file
device is a drive, which syntax is the same as the one in GRUB (see section How to specify devices), and file is an OS's file, which is normally a device file.
The reason why the grub shell gives you the device map file is that it cannot guess the map between BIOS drives and OS devices correctly in some environments. For example, if you exchange the boot sequence between IDE and SCSI in your BIOS, it mistakes the order.
Thus, edit the file if the grub shell makes a mistake. You can put any comments in the file if needed, as the grub shell assumes that a line is just a comment if the first character is `#'.
The program @command{grub-install} installs GRUB on your drive by the grub shell (see section Invoking the grub shell). You must specify the device name on which you want to install GRUB, like this:
grub-install install_device
The device name install_device is an OS device name or a GRUB device name.
@command{grub-install} accepts the following options:
grub-install --root-directory=/boot '(hd0)'
grub-install --grub-shell="grub --read-only" /dev/fd0
The program @command{mbchk} checks for the format of a Multiboot kernel. We recommend using this program before booting your own kernel by GRUB.
@command{mbchk} accepts the following options:
This part describes the GRUB internals so that developers can understand the implementation and start to hack GRUB. Of course, the source code has the complete information, so refer to it when you are not satisfied with this documentation.
GRUB is broken into 2 distinct components, or stages, which are loaded at different times in the boot process. The Stage 1 has to know where to find Stage 2, and the Stage 2 has to know where to find its configuration file (if Stage 2 doesn't have a configuration file, it drops into the command line interface and waits for a user command).
Here is the memory map of the various components (10):
See the file `stage2/shared.h', for more information.
GRUB's stage1 and stage2 have embedded variables whose locations are well-defined, so that the installation can patch the binary file directly without recompilation of the modules.
In stage1, these are defined (The number in the parenthesis of each entry is an offset number):
jmp
command to the starting address of
the component loaded by the stage1.
A stage1.5 should be loaded at address 0x2000, and a stage2
should be loaded at address 0x8000. Both use a CS of 0.
In the first sector of stage1.5 and stage2, the blocklists are recorded between firstlist (0x200) and lastlist (determined when assembling the file `stage2/start.S').
The trick here is that it is actually read backward, and the first 8-byte blocklist is not read here, but after the pointer is decremented 8 bytes, then after reading it, it decrements again, reads, decrements, reads, etc. until it is finished. The terminating condition is when the number of sectors to be read in the next blocklist is 0.
The format of a blocklist can be seen from the example in the code just
before the firstlist
label. Note that it is always from the
beginning of the disk, and not relative to the partition
boundaries.
In stage1.5 and stage2 (these are all defined at the beginning of `shared_src/asm.S'):
device:filename
(`:' is not present actually).
device is an unsigned long like install_partition, and
filename is an absolute filename or a blocklist. If device
is disabled, that is, the drive number is 0xff, then stage1.5 uses
the boot drive and the install partition instead.
For any particular partition, it is presumed that only one of the normal filesystems such as FAT, FFS, or ext2fs can be used, so there is a switch table managed by the functions in `disk_io.c'. The notation is that you can only mount one at a time.
The blocklist filesystem has a special place in the system. In addition to the normal filesystem (or even without one mounted), you can access disk blocks directly (in the indicated partition) via the blocklist notation. Using the blocklist filesystem doesn't effect any other filesystem mounts.
The variables which can be read by the filesystem backend are:
current_drive
current_partition
current_slice
saved_drive
saved_partition
part_start
part_length
print_possibilities
dir
function should print the possible completions
of a file, and false when it should try to actually open a file of that
name.
FSYS_BUF
The variables which need to be written by a filesystem backend are:
filepos
filemax
disk_read_func
NULL
at all other times (it will be
NULL
by default). If this isn't done correctly, then the
@command{testload} and @command{install} commands won't work
correctly.
The functions expected to be used by the filesystem backend are:
devread
grub_read
grub_read
can be used, after setting block_file
to 1.
The functions expected to be defined by the filesystem backend are described at least moderately in the file `filesys.h'. Their usage is fairly evident from their use in the functions in `disk_io.c', look for the use of the fsys_table array.
Caution: The semantics are such that then `mount'ing the filesystem, presume the filesystem buffer FSYS_BUF is corrupted, and (re-)load all important contents. When opening and reading a file, presume that the data from the `mount' is available, and doesn't get corrupted by the open/read (i.e. multiple opens and/or reads will be done with only one mount if in the same filesystem).
The disk space can be used in a boot loader is very restricted because a MBR (see section The structure of Master Boot Record) is only 512 bytes but it also contains a partition table (see section The format of partition table) and a BPB. So the question is how to make a boot loader code enough small to be fit in a MBR.
However, GRUB is a very large program, so we break GRUB into 2 (or 3) distinct components, stage1 and stage2 (and optionally stage1.5). See section The memory map of various components, for more information.
We embed stage1 in a MBR or in the boot sector of a partition , and place stage2 in a filesystem. The optional stage1.5 can be installed in a filesystem, in the boot loader area in a FFS, and in the sectors right after a MBR, because stage1.5 is enough small and the sectors right after a MBR is normally an unused region. The size of this region is the number of sectors per head minus 1.
Thus, all the stage1 must do is just load a stage2 or stage1.5. But even if stage1 needs not to support the user interface or the filesystem interface, it is impossible to make stage1 less than 400 bytes, because GRUB should support both the CHS mode and the LBA mode (see section INT 13H disk I/O interrupts).
The solution used by GRUB is that stage1 loads only the first sector of a stage2 (or a stage1.5) and stage2 itself loads the rest. The flow of stage1 is:
The flow of stage2 (and stage1.5) is:
Note that stage2 (or stage1.5) does not probe the geometry or the accessing mode of the loading drive, since stage1 has already probed them.
In the PC world, BIOS cannot detect if a hard disk drive is SCSI or IDE, generally speaking. Thus, it is not trivial to know which BIOS drive corresponds to an OS device. So the Multiboot Specification describes some techniques on how to guess mappings (see section `BIOS device mapping techniques' in The Multiboot Specification).
However, the techniques described are unreliable or difficult to be implemented, so we use a different technique from them in GRUB. Our technique is INT 13H tracking technique. More precisely, it runs the INT 13 call (see section INT 13H disk I/O interrupts) in single-step mode just like a debugger and parses the instructions.
To execute the call one instruction at a time, set the TF (trap flag) flag in the register FLAGS. By this, your CPU generates Break Point Trap after each instruction is executed and call INT 1. In the stack in the interrupt handler, callee's FLAGS and the far pointer which points to the next instruction to be executed are pushed, so we can know what instruction will be executed in the next time and the current contents of all the registers. If the next instruction is an I/O operation, the interrupt handler adds the I/O port into the I/O map.
If the INT 13 handler returns, the TF flag is cleared automatically by
the instruction iret
, and then output the I/O map on the screen.
See the source code for the command @command{ioprobe}
(see section The list of command-line and menu entry commands), for more information.
There are three BIOS calls which return the information of installed RAM. GRUB uses these calls to detect all installed RAM and which address range should be treated by operating systems.
Real mode only.
This call returns a memory map of all the installed RAM, and of physical memory ranges reserved by the BIOS. The address map is returned by making successive calls to this API, each returning one "run" of physical address information. Each run has a type which dictates how this run of physical address range should be treated by the operating system.
If the information returned from INT 15h, AX=E820h in some way differs from INT 15h, AX=E801h (see section INT 15H, AX=E801h interrupt call) or INT 15h AH=88h (see section INT 15H, AX=88h interrupt call), then the information returned from E820h supersedes what is returned from these older interfaces. This allows the BIOS to return whatever information it wishes to for compatibility reasons.
Input:
EAX | Function Code | E820h |
EBX | Continuation | Contains the continuation | value to get the next run of physical memory. This is the value returned by a previous call to this routine. If this is the first call,
ES:DI | Buffer Pointer | Pointer to an Address Range | Descriptor structure which the BIOS is to fill in.
ECX | Buffer Size | The length in bytes of the | structure passed to the BIOS. The BIOS will fill in at most
EDX | Signature | `SMAP' - Used by the BIOS to | verify the caller is requesting the system map information to be returned in
Output:
CF | Carry Flag | Non-Carry - indicates no error |
EAX | Signature | `SMAP' - Signature to verify | correct BIOS revision.
ES:DI | Buffer Pointer | Returned Address Range | Descriptor pointer. Same value as on input.
ECX | Buffer Size | Number of bytes returned by the | BIOS in the address range descriptor. The minimum size structure returned by the BIOS is 20 bytes.
EBX | Continuation | Contains the continuation value | to get the next address descriptor. The actual significance of the continuation value is up to the discretion of the BIOS. The caller must pass the continuation value unchanged as input to the next iteration of the E820h call in order to get the next Address Range Descriptor. A return value of zero means that this is the last descriptor. Note that the BIOS indicates that the last valid descriptor has been returned by either returning a zero as the continuation value, or by returning carry.
The Address Range Descriptor Structure is:
Offset in Bytes | Name | Description |
0 | BaseAddrLow | Low 32 Bits of Base Address |
4 | BaseAddrHigh | High 32 Bits of Base Address |
8 | LengthLow | Low 32 Bits of Length in Bytes |
12 | LengthHigh | High 32 Bits of Length in Bytes |
16 | Type | Address type of this range |
The BaseAddrLow and BaseAddrHigh together are the 64 bit BaseAddress of this range. The BaseAddress is the physical address of the start of the range being specified.
The LengthLow and LengthHigh together are the 64 bit Length of this range. The Length is the physical contiguous length in bytes of a range being specified.
The Type field describes the usage of the described address range as defined in the table below:
Value | Mnemonic | Description |
1 | AddressRangeMemory | This run is available | RAM usable by the operating system.
2 | AddressRangeReserved | This run of addresses is in | use or reserved by the system, and must not be used by the operating system.
Other | Undefined | Undefined - Reserved for future | use. Any range of this type must be treated by the OS as if the type returned was AddressRangeReserved.
The BIOS can use the AddressRangeReserved address range type to block out various addresses as not suitable for use by a programmable device.
Some of the reasons a BIOS would do this are:
Here is the list of assumptions and limitations:
Here we explain an example address map. This sample address map describes a machine which has 128 MB RAM, 640K of base memory and 127 MB extended. The base memory has 639K available for the user and 1K for an extended BIOS data area. There is a 4 MB Linear Frame Buffer (LFB) based at 12 MB. The memory hole created by the chipset is from 8 M to 16 M. There are memory mapped APIC devices in the system. The IO Unit is at FEC00000 and the Local Unit is at FEE00000. The system BIOS is remapped to 4G - 64K.
Note that the 639K endpoint of the first memory range is also the base memory size reported in the BIOS data segment at 40:13.
Key to types: ARM is AddressRangeMemory, ARR is AddressRangeReserved.
Base (Hex) | Length | Type | Description |
0000 0000 | 639K | ARM | Available Base memory - | typically the same value as is returned via the INT 12 function.
0009 FC00 | 1K | ARR | Memory reserved for use by the | BIOS(s). This area typically includes the Extended BIOS data area.
000F 0000 | 64K | ARR | System BIOS. |
0010 0000 | 7M | ARM | Extended memory, this is not | limited to the 64MB address range.
0080 0000 | 8M | ARR | Chipset memory hole required to | support the LFB mapping at 12 MB.
0100 0000 | 120M | ARM | Base board RAM relocated | above a chipset memory hole.
FEC0 0000 | 4K | ARR | IO APIC memory mapped I/O at | FEC00000. Note the range of addresses required for an APIC device may vary from one motherboard manufacturer to another
FEE0 0000 | 4K | ARR | Local APIC memory mapped I/O at | FEE00000.
FFFF 0000 | 64K | ARR | Remapped System BIOS at end of | address space.
The following code segment is intended to describe the algorithm needed when calling the Query System Address Map function. It is an implementation example and uses non standard mechanisms.
E820Present = FALSE; Regs.ebx = 0; do { Regs.eax = 0xE820; Regs.es = SEGMENT (&Descriptor); Regs.di = OFFSET (&Descriptor); Regs.ecx = sizeof (Descriptor); Regs.edx = 'SMAP'; _int (0x15, Regs); if ((Regs.eflags & EFLAGS_CARRY) || Regs.eax != 'SMAP') { break; } if (Regs.ecx < 20 || Regs.ecx > sizeof (Descriptor)) { /* bug in bios - all returned descriptors must be at least 20 bytes long, and can not be larger than the input buffer. */ break; } E820Present = TRUE; . . . Add address range Descriptor.BaseAddress through Descriptor.BaseAddress + Descriptor.Length as type Descriptor.Type . . . } while (Regs.ebx != 0); if (! E820Present) { . . . call INT 15H, AX E801h and/or INT 15H, AH=88h to obtain old style memory information . . . }
Real mode only.
Originally defined for EISA servers, this interface is capable of reporting up to 4 GB of RAM. While not nearly as flexible as E820h, it is present in many more systems.
Input:
AX | Function Code | E801h. |
Output:
CF | Carry Flag | Non-Carry - indicates no error. |
AX | Extended 1 | Number of contiguous KB between 1 | and 16 MB, maximum 0x3C00 = 15 MB.
BX | Extended 2 | Number of contiguous 64KB blocks | between 16 MB and 4GB.
CX | Configured 1 | Number of contiguous KB between 1 | and 16 MB, maximum 0x3c00 = 15 MB.
DX | Configured 2 | Number of contiguous 64KB blocks | between 16 MB and 4 GB.
Not sure what this difference between the Extended and Configured numbers are, but they appear to be identical, as reported from the BIOS.
It is possible for a machine using this interface to report a memory hole just under 16 MB (Count 1 is less than 15 MB, but Count 2 is non-zero).
Real mode only.
This interface is quite primitive. It returns a single value for contiguous memory above 1 MB. The biggest limitation is that the value returned is a 16-bit value, in KB, so it has a maximum saturation of just under 64 MB even presuming it returns as much as it can. On some systems, it won't return anything above the 16 MB boundary.
The one useful point is that it works on every PC available.
Input:
AH | Function Code | 88h |
Output:
CF | Carry Flag | Non-Carry - indicates no error. |
AX | Memory Count | Number of contiguous KB above 1 | MB.
In the PC world, living with the BIOS disk interface is definitely a nightmare. This section documents how awful the chaos is and how GRUB deals with the BIOS disks.
CHS -- Cylinder/Head/Sector -- is the traditional way to address sectors on a disk. There are at least two types of CHS addressing; the CHS that is used at the INT 13H interface and the CHS that is used at the ATA device interface. In the MFM/RLL/ESDI and early ATA days the CHS used at the INT 13H interface was the same as the CHS used at the device interface.
Today we have CHS translating BIOS types that can use one CHS at the INT 13H interface and a different CHS at the device interface. These two types of CHS will be called the logical CHS or L-CHS and the physical CHS or P-CHS in this section. L-CHS is the CHS used at the INT 13H interface and P-CHS is the CHS used at the device interface.
The L-CHS used at the INT 13 interface allows up to 256 heads, up to 1024 cylinders and up to 63 sectors. This allows support of up to 8GB drives. This scheme started with either ESDI or SCSI adapters many years ago.
The P-CHS used at the device interface allows up to 16 heads up to 65535 cylinders, and up to 63 sectors. This allows access to about 2^26 sectors (32GB) on an ATA device. When a P-CHS is used at the INT 13H interface it is limited to 1024 cylinders, 16 heads and 63 sectors. This is where the old 528MB limit originated.
LBA -- Logical Block Address -- is another way of addressing sectors that uses a simple numbering scheme starting with zero as the address of the first sector on a device. The ATA standard requires that cylinder 0, head 0, sector 1 address the same sector as addressed by LBA 0. LBA addressing can be used at the ATA interface if the ATA device supports it. LBA addressing is also used at the INT 13H interface by the AH=4xH read/write calls.
ATA devices may also support LBA at the device interface. LBA allows access to approximately 2^28 sectors (137GB) on an ATA device.
A SCSI host adapter can convert a L-CHS directly to an LBA used in the SCSI read/write commands. On a PC today, SCSI is also limited to 8GB when CHS addressing is used at the INT 13H interface.
First, all OS's that want to be co-resident with another OS (and that is all of the PC based OS's that we know of) must use INT 13H to determine the capacity of a hard disk. And that capacity information must be determined in L-CHS mode. Why is this? Because:
During the boot processing, all of the disk read accesses are done in L-CHS mode via INT 13H and this includes loading the first of the OS's kernel code or boot manager's code.
Second, because there can be multiple BIOS types in any one system, each drive may be under the control of a different type of BIOS. For example, drive 80H (the first hard drive) could be controlled by the original system BIOS, drive 81H (the second drive) could be controlled by a option ROM BIOS and drive 82H (the third drive) could be controlled by a software driver. Also, be aware that each drive could be a different type, for example, drive 80H could be an MFM drive, drive 81H could be an ATA drive, drive 82H could be a SCSI drive.
Third, not all OS's understand or use BIOS drive numbers greater than 81H. Even if there is INT 13H support for drives 82H or greater, the OS may not use that support.
Fourth, the BIOS INT 13H configuration calls are:
An ATA disk must implement both CHS and LBA addressing and must at any given time support only one P-CHS at the device interface. And, the drive must maintain a strict relationship between the sector addressing in CHS mode and LBA mode. Quoting the ATA-2 document:
LBA = ( (cylinder * heads_per_cylinder + heads ) * sectors_per_track ) + sector - 1 where heads_per_cylinder and sectors_per_track are the current translation mode values.
This algorithm can also be used by a BIOS or an OS to convert a L-CHS to an LBA.
This algorithm can be reversed such that an LBA can be converted to a CHS:
cylinder = LBA / (heads_per_cylinder * sectors_per_track) temp = LBA % (heads_per_cylinder * sectors_per_track) head = temp / sectors_per_track sector = temp % sectors_per_track + 1
While most OS's compute disk addresses in an LBA scheme, an OS like DOS must convert that LBA to a CHS in order to call INT 13H.
The basic problem is that there is no requirement that a CHS translating BIOS followed these rules. There are many other algorithms that can be implemented to perform a similar function. Today, there are at least two popular implementations: the Phoenix implementation (described above) and the non-Phoenix implementations. Because a protected mode OS that does not want to use INT 13H must implement the same CHS translation algorithm. If it doesn't, your data gets scrambled.
In the perfect world of tomorrow, maybe only LBA will be used. But today we are faced with the following problems:
These are difficult problems to overcome in today's industry environment. The result: chaos.
Real mode only. These functions are the traditional CHS mode disk interface. GRUB calls them only if LBA mode is not available.
INT 13H, AH=02h reads sectors into memory.
Input:
AH | 02h |
AL | The number of sectors to read (must be non-zero). |
CH | Low 8 bits of cylinder number. |
CL | Sector number in bits 0-5, and high 2 bits of | cylinder number in bits 6-7.
DH | Head number. |
DL | Drive number (bit 7 set for hard disk). |
ES:BX | Data buffer. |
Output:
CF | Set on error. |
AH | Status. |
AL | The number of sectors transferred (only valid if CF | set for some BIOSes).
INT 13H, AH=03h writes disk sectors.
Input:
AH | 03h |
AL | The number of sectors to write (must be non-zero). |
CH | Low 8 bits of cylinder number. |
CL | Sector number in bits 0-5, and high 2 bits of | cylinder number in bits 6-7.
DH | Head number. |
DL | Drive number (bit 7 set for hard disk). |
ES:BX | Data buffer. |
Output:
CF | Set on error. |
AH | Status. |
AL | The number of sectors transferred (only valid if CF | set for some BIOSes).
INT 13H, AH=08h returns drive parameters. For systems predating the IBM PC/AT, this call is only valid for hard disks.
Input:
AH | 08h |
DL | Drive number (bit 7 set for hard disk). |
Output:
CF | Set on error. |
AH | 0. |
AL | 0 on at least some BIOSes. |
BL | Drive type (AT/PS2 floppies only). |
CH | Low 8 bits of maximum cylinder number. |
CL | Maximum sector number in bits 0-5, and high 2 bits | of maximum cylinder number in bits 6-7.
DH | Maximum head number. |
DL | The number of drives. |
ES:DI | Drive parameter table (floppies only). |
Real mode only. These functions are IBM/MS INT 13 Extensions to support LBA mode. GRUB uses them if available so that it can read/write over 8GB area.
INT 13, AH=41h checks if LBA is supported.
Input:
AH | 41h. |
BX | 55AAh. |
DL | Drive number. |
Output:
CF | Set on error. |
AH | Major version of extensions (10h for 1.x, 20h for | 2.0 / EDD-1.0, 21h for 2.1 / EDD-1.1 and 30h for EDD-3.0) if successful, otherwise 01h (the error code of invalid function).
BX | AA55h if installed. |
AL | Internal use. |
CX | API subset support bitmap (see below). |
DH | Extension version. |
The bitfields for the API subset support bitmap are(11):
Bit(s) | Description |
0 | Extended disk access functions (AH=42h-44h, 47h, 48h) | supported.
1 | Removable drive controller functions (AH=45h, 46h, 48h, | 49h, INT 15H, AH=52h) supported.
2 | Enhanced disk drive (EDD) functions (AH=48h, 4Eh) | supported.
3-15 | Reserved (0). |
INT 13, AH=42h reads sectors into memory.
Input:
AH | 42h. |
DL | Drive number. |
DS:SI | Disk Address Packet (see below). |
Output:
CF | Set on error. |
AH | 0 if successful, otherwise error code. |
The format of Disk Address Packet is:
Offset (hex) | Size (byte) | Description |
00 | 1 | 10h (The size of packet). |
01 | 1 | Reserved (0). |
02 | 2 | The number of blocks to transfer (max 007F for | Phoenix EDD).
04 | 4 | Transfer buffer (SEGMENT:OFFSET). |
08 | 8 | Starting absolute block number. |
INT 13, AH=43h writes disk sectors.
Input:
AH | 43h. |
AL | Write flags (In version 1.0 and 2.0, bit 0 is the | flag for verify write and other bits are reserved (0). In version 2.1, 00h and 01h indicates write without verify, and 02h indicates write with verify.
DL | Drive number. |
DS:SI | Disk Address Packet (see above). |
Output:
CF | Set on error. |
AH | 0 if successful, otherwise error code. |
INT 13, AH=48h returns drive parameters. GRUB only makes use of the total number of sectors, and ignore the CHS information, because only L-CHS makes sense. See section CHS addressing and LBA addressing, for more information.
Input:
AH | 48h. |
DL | Drive number. |
DS:SI | Buffer for drive parameters (see below). |
Output:
CF | Set on error. |
AH | 0 if successful, otherwise error code. |
The format of drive parameters is:
Offset (hex) | Size (byte) | Description |
00 | 2 | The size of buffer. Before calling this function, | set to the maximum buffer size, at least 1Ah. The size actually filled is returned (1Ah for version 1.0, 1Eh for 2.x and 42h for 3.0).
02 | 2 | Information flags (see below). |
04 | 4 | The number of physical cylinders. |
08 | 4 | The number of physical heads. |
0C | 4 | The number of physical sectors per track. |
10 | 8 | The total number of sectors. |
18 | 2 | The bytes per sector. |
v2.0 and later | ||
1A | 4 | EDD configuration parameters. |
v3.0 | ||
1E | 2 | Signature BEDD to indicate presence of Device Path | information.
20 | 1 | The length of Device Path information, including | signature and this byte (24h for version 3.0).
21 | 3 | Reserved (0). |
24 | 4 | ASCIZ name of host bus (`ISA' or `PCI'). |
28 | 8 | ASCIZ name of interface type (`ATA', | `ATAPI', `SCSI', `USB', `1394' or `FIBRE').
30 | 8 | Interface Path. |
38 | 8 | Device Path. |
40 | 1 | Reserved (0). |
41 | 1 | Checksum of bytes 1Eh-40h (2's complement of sum, | which makes the 8 bit sum of bytes 1Eh-41h equal to 00h).
The information flags are:
Bit(s) | Description |
0 | DMA boundary errors handles transparently. |
1 | CHS information is valid. |
2 | Removable drive. |
3 | Write with verify supported. |
4 | Drive has change-line support (required if drive is | removable).
5 | Drive can be locked (required if drive is removable). |
6 | CHS information set to maximum supported values, not | current media.
7-15 | Reserved (0). |
A Master Boot Record (MBR) is the sector at cylinder 0, head 0, sector 1 of a hard disk. A MBR-like structure must be created in each of partitions by the FDISK program.
At the completion of your system's Power On Self Test (POST), INT 19H is called. Usually INT 19 tries to read a boot sector from the first floppy drive(12). If a boot sector is found on the floppy disk, that boot sector is read into memory at location 0000:7C00 and INT 19H jumps to memory location 0000:7C00. However, if no boot sector is found on the first floppy drive, INT 19H tries to read the MBR from the first hard drive. If an MBR is found it is read into memory at location 0000:7C00 and INT 19H jumps to memory location 0000:7C00. The small program in the MBR will attempt to locate an active (bootable) partition in its partition table(13). The small program in the boot sector must locate the first part of the operating system's kernel loader program (or perhaps the kernel itself or perhaps a boot manager program) and read that into memory.
INT 19H is also called when the CTRL-ALT-DEL keys are used. On most systems, CTRL-ALT-DEL causes an short version of the POST to be executed before INT 19H is called.
The stuff is:
However, the first 62 bytes of a boot sector are known as the BIOS Parameter Block (BPB), so GRUB cannot use these bytes for its own purpose.
If an active partition is found, that partition's boot record is read
into 0000:7C00 and the MBR code jumps to 0000:7C00 with SI
pointing to the partition table entry that describes the partition being
booted. The boot record program uses this data to determine the drive
being booted from and the location of the partition on the disk.
The first byte of an active partition table entry is 80. This byte is
loaded into the DL
register before INT 13H is called to read the
boot sector. When INT 13H is called, DL
is the BIOS device
number. Because of this, the boot sector read by this MBR program can
only be read from BIOS device number 80 (the first hard disk). This is
one of the reasons why it is usually not possible to boot from any other
hard disk.
FDISK creates all partition records (sectors). The primary purpose of a partition record is to hold a partition table. The rules for how FDISK works are unwritten but so far most FDISK programs seem to follow the same basic idea.
First, all partition table records (sectors) have the same format. This includes the partition table record at cylinder 0, head 0, sector 1 -- what is known as the Master Boot Record (MBR). The last 66 bytes of a partition table record contain a partition table and a 2 byte signature. The first 446 bytes of these sectors usually contain a program but only the program in the MBR is ever executed (so extended partition table records could contain something other than a program in the first 466 bytes). For more information, see section The structure of Master Boot Record.
Second, extended partitions are nested inside one another and extended partition table records form a linked list. We will attempt to show this in a diagram at section The format of the table entry.
Each partition table entry is 16 bytes and contains things like the start and end location of a partition in CHS, the start in LBA, the size in sectors, the partition type and the active flag. Older versions of FDISK may compute incorrect LBA or size values. And when your computer boots itself, only the CHS fields of the partition table entries are used (another reason LBA doesn't solve the >528MB problem). The CHS fields in the partition tables are in L-CHS format, see section CHS addressing and LBA addressing.
There is no central clearing house to assign the codes used in the one byte type field. But codes are assigned (or used) to define most every type of file system that anyone has ever implemented on the x86 PC: 12-bit FAT, 16-bit FAT, HPFS, NTFS, etc. Plus, an extended partition also has a unique type code.
In the FDISK program `sfdisk', the following list is assumed:
The 16 bytes of a partition table entry are used as follows:
+--- Bit 7 is the active partition flag, bits 6-0 are zero. | | +--- Starting CHS in INT 13 call format. | | | | +--- Partition type byte. | | | | | | +--- Ending CHS in INT 13 call format. | | | | | | | | +-- Starting LBA. | | | | | | | | | | +-- Size in sectors. | | | | | | v <--+---> v <--+--> v v 0 1 2 3 4 5 6 7 8 9 A B C D E F DH DL CH CL TB DL CH CL LBA..... SIZE.... 80 01 01 00 06 0e be 94 3e000000 0c610900 1st entry 00 00 81 95 05 0e fe 7d 4a610900 724e0300 2nd entry 00 00 00 00 00 00 00 00 00000000 00000000 3rd entry 00 00 00 00 00 00 00 00 00000000 00000000 4th entry
Bytes 0-3 are used by the small program in the Master Boot Record to read the first sector of an active partition into memory. The DH, DL, CH and CL above show which x86 register is loaded when the MBR program calls INT 13H AH=02h to read the active partition's boot sector. For more information, see section The structure of Master Boot Record.
These entries define the following partitions:
Keep in mind that there are no written rules and no industry standards on how FDISK should work but here are some basic rules that seem to be followed by most versions of FDISK:
There are no written rules as to how an OS scans the partition table entries so each OS can have a different method. For DOS, this means that different versions could assign different drive letters to the same FAT file system partitions.
$ mke2fs /dev/fd0
$ /sbin/grub --batch <<EOT root (fd0) setup (fd0) quit EOT
$ ld -vThis will show two versions, but only the latter is important. If the version is identical with what you have installed, the installation was not bad. Well, please try:
$ gcc -Wl,-v 2>&1 | grep "GNU ld"If this is not identical with the result above, you should specify the directory where you have installed binutils for the script configure, like this:
$ ./configure --with-binutils=/usr/local/binIf you follow the instructions above but GRUB still crashes, probably there is a serious bug in GRUB. See section Reporting bugs.
grub> kernel /vmlinuz mem=128MYou may pass other options in the same way. See See section GNU/Linux, for more details.
grub> root (hd0,1) grub> install /grub/stage1 d (hd0) /grub/stage2 p /grub/menu.lst
Caution: GRUB requires binutils-2.9.1.0.23 or later because the GNU assembler has been changed so that it can produce real 16bits machine code between 2.9.1 and 2.9.1.0.x. See http://sourceware.cygnus.com/binutils/, to obtain information on how to get the latest version.
GRUB is available from the GNU alpha archive site alpha.gnu.org:/gnu/grub or any of its mirrors. The file will be named grub-version.tar.gz. The current version is 0.5.96, so the file you should grab is:
ftp://alpha.gnu.org/gnu/grub/grub-0.5.96.tar.gz
To unbundle GRUB use the instruction:
zcat grub-0.5.96.tar.gz | tar xvf -
which will create a directory called `grub-0.5.96' with all the sources. You can look at the file `INSTALL' for detailed instructions on how to build and install GRUB, but you should be able to just do:
$ cd grub-0.5.96 $ ./configure $ make install
This will install the grub shell `grub' (see section Invoking the grub shell), the Multiboot checker `mbchk' (see section Invoking mbchk), and the GRUB images It will also install the GRUB manual.
Also, the latest version is available from the CVS. The repository is:
:pserver:anoncvs@subversions.gnu.org:/home/cvs
and the module is:
grub
The password for anoncvs is empty. So the instruction is:
$ cvs -d :pserver:anoncvs@subversions.gnu.org:/home/cvs \ login Password: ENTER $ cvs -d :pserver:anoncvs@subversions.gnu.org:/home/cvs \ checkout grub
Get the recent version of GNU Automake from the CVS to regenerate `Makefile.in's. See http://sourceware.cygnus.com/automake/, for more information.
When you encounter any problem or bug, please submit it to bug-grub@gnu.org with information about your computer and what you did as much as possible. Take a look at this list before you send e-mail to the address:
Jump to: b - c - d - e - f - g - h - i - k - l - m - p - q - r - s - t - u
This document was generated on 25 February 2001 using texi2html 1.56k.