While we may not be able to fix this problem in existing commercial operating systems, it shouldn't be too difficult for a few people in the free OS communities to put their heads together and solve this problem for the popular free operating systems. That's what this standard aims for. Basically, it specifies an interface between a boot loader and a operating system, such that any complying boot loader should be able to load any complying operating system. This standard does NOT specify how boot loaders should work - only how they must interface with the OS being loaded.
The term "OS image" is used to refer to the initial binary image that the boot loader loads into memory and transfers control to to start the OS. The OS image is typically an executable containing the OS kernel.
The term "boot module" refers to other auxiliary files that the boot loader loads into memory along with the OS image, but does not interpret in any way other than passing their locations to the OS when it is invoked.
Disk-based boot loaders may use a variety of techniques to find the relevant OS image and boot module data on disk, such as by interpretation of specific file systems (e.g. the BSD/Mach boot loader), using precalculated "block lists" (e.g. LILO), loading from a special "boot partition" (e.g. OS/2), or even loading from within another operating system (e.g. the VSTa boot code, which loads from DOS). Similarly, network-based boot loaders could use a variety of network hardware and protocols.
It is hoped that boot loaders will be created that support multiple loading mechanisms, increasing their portability, robustness, and user-friendliness.
Unfortunately, there is a horrendous variety of executable file formats even among free Unix-like PC-based OS's - generally a different format for each OS. Most of the relevant free OS's use some variant of a.out format, but some are moving to ELF. It is highly desirable for boot loaders not to have to be able to interpret all the different types of executable file formats in existence in order to load the OS image - otherwise the boot loader effectively becomes OS-specific again.
This standard adopts a compromise solution to this problem. MultiBoot compliant boot images always either (a) are in ELF format, or (b) contain a "magic MultiBoot header", described below, which allows the boot loader to load the image without having to understand numerous a.out variants or other executable formats. This magic header does not need to be at the very beginning of the executable file, so kernel images can still conform to the local a.out format variant in addition to being MultiBoot compliant.
Thus, this standard should provide a standard method for a boot loader to indicate to the OS what auxiliary boot modules were loaded, and where they can be found. Boot loaders don't have to support multiple boot modules, but they are strongly encouraged to, because some OS's will be unable to boot without them.
Unfortunately, the exact meaning of the text, data, bss, and entry fields of a.out headers tends to vary widely between different executable flavors, and it is sometimes very difficult to distinguish one flavor from another (e.g. Linux ZMAGIC executables and Mach ZMAGIC executables). Furthermore, there is no simple, reliable way of determining at what address in memory the text segment is supposed to start. Therefore, this standard requires that an additional header, known as a 'multiboot_header', appear somewhere near the beginning of the executable file. In general it should come "as early as possible", and is typically embedded in the beginning of the text segment after the "real" executable header. It _must_ be contained completely within the first 8192 bytes of the executable file, and must be longword (32-bit) aligned. These rules allow the boot loader to find and synchronize with the text segment in the a.out file without knowing beforehand the details of the a.out variant. The layout of the header is as follows:
+-------------------+ 0 | magic: 0x1BADB002 | (required) 4 | flags | (required) 8 | checksum | (required) +-------------------+ 8 | header_addr | (present if flags is set) 12 | load_addr | (present if flags is set) 16 | load_end_addr | (present if flags is set) 20 | bss_end_addr | (present if flags is set) 24 | entry_addr | (present if flags is set) +-------------------+All fields are in little-endian byte order, of course. The first field is the magic number identifying the header, which must be the hex value 0x1BADB002.
The flags field specifies features that the OS image requests or requires of the boot loader. Bits 0-15 indicate requirements; if the boot loader sees any of these bits set but doesn't understand the flag or can't fulfill the requirements it indicates for some reason, it must notify the user and fail to load the OS image. Bits 16-31 indicate optional features; if any bits in this range are set but the boot loader doesn't understand them, it can simply ignore them and proceed as usual. Naturally, all as-yet-undefined bits in the flags word must be set to zero in OS images. This way, the flags fields serves for version control as well as simple feature selection.
If bit 0 in the flags word is set, then all boot modules loaded along with the OS must be aligned on page (4KB) boundaries. Some OS's expect to be able to map the pages containing boot modules directly into a paged address space during startup, and thus need the boot modules to be page-aligned.
If bit 1 in the flags word is set, then information on available memory via at least the 'mem_*' fields of the multiboot_info structure defined below must be included. If the bootloader is capable of passing a memory map (the 'mmap_*' fields) and one exists, then it must be included as well.
If bit 16 in the flags word is set, then the fields at offsets 8-24 in the multiboot_header are valid, and the boot loader should use them instead of the fields in the actual executable header to calculate where to load the OS image. This information does not need to be provided if the kernel image is in ELF format, but it should be provided if the images is in a.out format or in some other format. Compliant boot loaders must be able to load images that either are in ELF format or contain the load address information embedded in the multiboot_header; they may also directly support other executable formats, such as particular a.out variants, but are not required to.
All of the address fields enabled by flag bit 16 are physical addresses. The meaning of each is as follows:
The multiboot_info structure and its related substructures may be placed anywhere in memory by the boot loader (with the exception of the memory reserved for the kernel and boot modules, of course). It is the OS's responsibility to avoid overwriting this memory until it is done using it.
The format of the multiboot_info structure (as defined so far) follows:
+-------------------+ 0 | flags | (required) +-------------------+ 4 | mem_lower | (present if flags is set) 8 | mem_upper | (present if flags is set) +-------------------+ 12 | boot_device | (present if flags is set) +-------------------+ 16 | cmdline | (present if flags is set) +-------------------+ 20 | mods_count | (present if flags is set) 24 | mods_addr | (present if flags is set) +-------------------+ 28 - 40 | syms | (present if flags or flags is set) +-------------------+ 44 | mmap_length | (present if flags is set) 48 | mmap_addr | (present if flags is set) +-------------------+The first longword indicates the presence and validity of other fields in the multiboot_info structure. All as-yet-undefined bits must be set to zero by the boot loader. Any set bits that the OS does not understand should be ignored. Thus, the flags field also functions as a version indicator, allowing the multiboot_info structure to be expanded in the future without breaking anything.
If bit 0 in the multiboot_info.flags word is set, then the 'mem_*' fields are valid. 'mem_lower' and 'mem_upper' indicate the amount of lower and upper memory, respectively, in kilobytes. Lower memory starts at address 0, and upper memory starts at address 1 megabyte. The maximum possible value for lower memory is 640 kilobytes. The value returned for upper memory is maximally the address of the first upper memory hole minus 1 megabyte. It is not guaranteed to be this value.
If bit 1 in the multiboot_info.flags word is set, then the 'boot_device' field is valid, and indicates which BIOS disk device the boot loader loaded the OS from. If the OS was not loaded from a BIOS disk, then this field must not be present (bit 3 must be clear). The OS may use this field as a hint for determining its own "root" device, but is not required to. The boot_device field is layed out in four one-byte subfields as follows:
+-------+-------+-------+-------+ | drive | part1 | part2 | part3 | +-------+-------+-------+-------+The first byte contains the BIOS drive number as understood by the BIOS INT 0x13 low-level disk interface: e.g. 0x00 for the first floppy disk or 0x80 for the first hard disk.
The three remaining bytes specify the boot partition. 'part1' specifies the "top-level" partition number, 'part2' specifies a "sub-partition" in the top-level partition, etc. Partition numbers always start from zero. Unused partition bytes must be set to 0xFF. For example, if the disk is partitioned using a simple one-level DOS partitioning scheme, then 'part1' contains the DOS partition number, and 'part2' and 'part3' are both zero. As another example, if a disk is partitioned first into DOS partitions, and then one of those DOS partitions is subdivided into several BSD partitions using BSD's "disklabel" strategy, then 'part1' contains the DOS partition number, 'part2' contains the BSD sub-partition within that DOS partition, and 'part3' is 0xFF.
DOS extended partitions are indicated as partition numbers starting from 4 and increasing, rather than as nested sub-partitions, even though the underlying disk layout of extended partitions is hierarchical in nature. For example, if the boot loader boots from the second extended partition on a disk partitioned in conventional DOS style, then 'part1' will be 5, and 'part2' and 'part3' will both be 0xFF.
If bit 2 of the flags longword is set, the 'cmdline' field is valid, and contains the physical address of the the command line to be passed to the kernel. The command line is a normal C-style null-terminated string.
If bit 3 of the flags is set, then the 'mods' fields indicate to the kernel what boot modules were loaded along with the kernel image, and where they can be found. 'mods_count' contains the number of modules loaded; 'mods_addr' contains the physical address of the first module structure. 'mods_count' may be zero, indicating no boot modules were loaded, even if bit 1 of 'flags' is set. Each module structure is formatted as follows:
+-------------------+ 0 | mod_start | 4 | mod_end | +-------------------+ 8 | string | +-------------------+ 12 | reserved (0) | +-------------------+The first two fields contain the start and end addresses of the boot module itself. The 'string' field provides an arbitrary string to be associated with that particular boot module; it is a null-terminated ASCII string, just like the kernel command line. The 'string' field may be 0 if there is no string associated with the module. Typically the string might be a command line (e.g. if the OS treats boot modules as executable programs), or a pathname (e.g. if the OS treats boot modules as files in a file system), but its exact use is specific to the OS. The 'reserved' field must be set to 0 by the boot loader and ignored by the OS.
NOTE: Bits 4 & 5 are mutually exclusive.
If bit 4 in the multiboot_info.flags word is set, then the following fields in the multiboot_info structure starting at byte 28 are valid:
+-------------------+ 28 | tabsize | 32 | strsize | 36 | addr | 40 | reserved (0) | +-------------------+These indicate where the symbol table from an a.out kernel image can be found. 'addr' is the physical address of the size (4-byte unsigned long) of an array of a.out-format 'nlist' structures, followed immediately by the array itself, then the size (4-byte unsigned long) of a set of null-terminated ASCII strings (plus sizeof(unsigned long) in this case), and finally the set of strings itself. 'tabsize' is equal to it's size parameter (found at the beginning of the symbol section), and 'strsize' is equal to it's size parameter (found at the beginning of the string section) of the following string table to which the symbol table refers. Note that 'tabsize' may be 0, indicating no symbols, even if bit 4 in the flags word is set.
If bit 5 in the multiboot_info.flags word is set, then the following fields in the multiboot_info structure starting at byte 28 are valid:
+-------------------+ 28 | num | 32 | size | 36 | addr | 40 | shndx | +-------------------+These indicate where the section header table from an ELF kernel is, the size of each entry, number of entries, and the string table used as the index of names. They correspond to the 'shdr_*' entries ('shdr_num', etc.) in the Executable and Linkable Format (ELF) specification in the program header. All sections are loaded, and the physical address fields of the elf section header then refer to where the sections are in memory (refer to the i386 ELF documentation for details as to how to read the section header(s)). Note that 'shdr_num' may be 0, indicating no symbols, even if bit 5 in the flags word is set.
If bit 6 in the multiboot_info.flags word is set, then the 'mmap_*' fields are valid, and indicate the address and length of a buffer containing a memory map of the machine provided by the BIOS. 'mmap_addr' is the address, and 'mmap_length' is the total size of the buffer. The buffer consists of one or more of the following size/structure pairs ('size' is really used for skipping to the next pair):
+-------------------+ -4 | size | +-------------------+ 0 | BaseAddrLow | 4 | BaseAddrHigh | 8 | LengthLow | 12 | LengthHigh | 16 | Type | +-------------------+where 'size' is the size of the associated structure in bytes, which can be greater than the minimum of 20 bytes. 'BaseAddrLow' is the lower 32 bits of the starting address, and 'BaseAddrHigh' is the upper 32 bits, for a total of a 64-bit starting address. 'LengthLow' is the lower 32 bits of the size of the memory region in bytes, and 'LengthHigh' is the upper 32 bits, for a total of a 64-bit length. 'Type' is the variety of address range represented, where a value of 1 indicates available RAM, and all other values currently indicated a reserved area.
The map provided is guaranteed to list all standard RAM that should be available for normal use.
Bryan Ford Computer Systems Laboratory University of Utah Salt Lake City, UT 84112 (801) 581-4280 firstname.lastname@example.org Erich Stefan Boleyn 924 S.W. 16th Ave, #202 Portland, OR, USA 97205 (503) 226-0741 email@example.comWe would also like to thank the many other people have provided comments, ideas, information, and other forms of support for our work.
Version 0.6 3/29/96 (a few wording changes, header checksum, and clarification of machine state passed to the OS) Version 0.5 2/23/96 (name change) Version 0.4 2/1/96 (major changes plus HTMLification) Version 0.3 12/23/95 Version 0.2 10/22/95 Version 0.1 6/26/95
In reference to bit 1 of the multiboot_info.flags parameter, it is recognized that determination of which BIOS drive maps to which OS-level device-driver is non-trivial, at best. Many kludges have been made to various OSes instead of solving this problem, most of them breaking under many conditions. To encourage the use of general-purpose solutions to this problem, here are 2 BIOS Device Mapping Techniques.
In reference to bit 6 of the multiboot_info.flags parameter, it is important to note that the data structure used there (starting with 'BaseAddrLow') is the data returned by the INT 15h, AX=E820h - Query System Address Map call. More information on reserved memory regions is defined on that web page. The interface here is meant to allow a bootloader to work unmodified with any reasonable extensions of the BIOS interface, passing along any extra data to be interpreted by the OS as desired.
The Mach 4 distribution, available by anonymous FTP from flux.cs.utah.edu:/flux, contains a C header file that defines the MultiBoot data structures described above; anyone is welcome to rip it out and use it for other boot loaders and OS's:
mach4-i386/include/mach/machine/multiboot.hThis distribution also contains code implementing a "Linux boot adaptor", which collects a MultiBoot-compliant OS image and an optional set of boot modules, compresses them, and packages them into a single traditional Linux boot image that can be loaded from LILO or other Linux boot loaders. There is also a corresponding "BSD boot adaptor" which can be used to wrap a MultiBoot kernel and set of modules and produce an image that can be loaded from the FreeBSD and NetBSD boot loaders. All of this code can be used as-is or as a basis for other boot loaders. These are the directories of primary relevance:
mach4-i386/boot mach4-i386/boot/bsd mach4-i386/boot/linuxThe Mach kernel itself in this distribution contains code that demonstrates how to create a compliant OS. The following files are of primary relevance:
mach4-i386/kernel/i386at/boothdr.S mach4-i386/kernel/i386at/model_dep.cFinally, I have created patches against the Linux 1.2.2 and FreeBSD 2.0 kernels, in order to make them compliant with this proposed standard. These patches are available in kahlua.cs.utah.edu:/private/boot.
A final release has not been made, but both the GRUB beta release (which is quite stable) and a patch for Multiboot version 0.6 for Mach4 UK22 are available in the GRUB public release area.