$Id$ COMBOOT and COM32 files SYSLINUX supports simple standalone programs, using a file format similar to DOS ".com" files. A 32-bit version, called COM32, is also provided. A simple API provides access to a limited set of filesystem and console functions. ++++ COMBOOT file format ++++ A COMBOOT file is a raw binary file containing 16-bit code. It should be linked to run at offset 0x100, and contain no absolute segment references. It is run in 16-bit real mode. A COMBOOT image can be written to be compatible with MS-DOS. Such a file will usually have extension ".com". A COMBOOT file which is not compatible with MS-DOS will usually have extension ".cbt". Before running the program, SYSLINUX sets up the following fields in the Program Segment Prefix (PSP), a structure at offset 0 in the program segment: Offset Size Meaning 0 word Contains an INT 20h instruction 2 word Contains the paragraph (16-byte "segment" address) at the end of memory available to the program. 128 byte Length of the command line arguments, including the leading space but not including the final CR character. 129 127b Command line arguments, starting with a space and ending with a CR character (ASCII 13). The program is allowed to use memory between the PSP paragraph (which all the CS, DS, ES and SS registers point to at program start) and the paragraph value given at offset 2. On startup, SP is set up to point to the end of the 64K segment, at 0xfffe. Under DOS it is possible for SP to contain a smaller value if memory is very tight; this is never the case under SYSLINUX. The program should make no assumptions about what segment address it will be loaded at; instead it should look at the segment registers on program startup. Both DOS and SYSLINUX will guarantee CS == DS == ES == SS on program start; the program should not assume anything about the values of FS or GS. To exit, a program can either execute a near RET (which will jump to offset 0 which contains an INT 20h instruction, terminating the program), or execute INT 20h or INT 21h AH=00h or INT 21h AH=4Ch. If compatiblity with SYSLINUX 1.xx is desired, use INT 20h. ++++ COM32 file format ++++ A COM32 file is a raw binary file containing 32-bit code. It should be linked to run at address 0x101000, and should not contain any segment references. It will be run in flat-memory 32-bit protected mode. Under SYSLINUX, it will be run in CPL 0, however, since it may be possible to create a COM32 execution engine that would run under something like Linux DOSEMU, it is recommended that the code does not assume CPL 0 unless absolutely necessary. It is highly recommended that every COM32 program begins with the byte sequence B8 FF 4C CD 21 (mov eax,21cd4cffh) as a magic number. A COM32 file should have extension ".c32". On startup, CS will be set up as a flat 32-bit code segment, and DS == ES == SS will be set up as the equivalent flat 32-bit data segment. FS and GS are reserved for future use and are currently initialized to zero. A COM32 image should not assume any particular values of segment selectors. ESP is set up at the end of available memory and also serves as notification to the program how much memory is available. The following arguments are passed to the program on the stack: Address Size Meaning [ESP] dword Return (termination) address [ESP+4] dword Number of additional arguments (currently 5) [ESP+8] dword Pointer to the command line arguments (null-terminated string) [ESP+12] dword Pointer to INT call helper function [ESP+16] dword Pointer to low memory bounce buffer [ESP+20] dword Size of low memory bounce buffer [ESP+24] dword Pointer to FAR call helper function (new in 2.05) This corresponds to the following C prototype, available in the file com32/include/com32.h: /* The standard prototype for _start() */ int _start(unsigned int __nargs, char *__cmdline, void (*__intcall)(uint8_t, com32sys_t *, com32sys_t *), void *__bounce_ptr, unsigned int __bounce_len, void (*__farcall)(uint32_t, uint16_t, com32sys_t *, com32sys_t *)); The intcall helper function can be used to issue BIOS or SYSLINUX API calls, and takes the interrupt number as first argument. The second argument is a pointer to the input register definition, an instance of the following structure (also available in com32.h): typedef union { uint32_t l; uint16_t w[2]; uint8_t b[4]; } reg32_t; typedef struct { uint16_t gs; /* Offset 0 */ uint16_t fs; /* Offset 2 */ uint16_t es; /* Offset 4 */ uint16_t ds; /* Offset 6 */ reg32_t edi; /* Offset 8 */ reg32_t esi; /* Offset 12 */ reg32_t ebp; /* Offset 16 */ reg32_t _unused; /* Offset 20 */ reg32_t ebx; /* Offset 24 */ reg32_t edx; /* Offset 28 */ reg32_t ecx; /* Offset 32 */ reg32_t eax; /* Offset 36 */ reg32_t eflags; /* Offset 40 */ } com32sys_t; The third argument is a pointer to the output register definition, an instance of the same structure. The third argument can also be zero (NULL). Since BIOS or SYSLINUX API calls can generally only manipulate data below address 0x100000, a "bounce buffer" in low memory, at least 64K in size, is available, to copy data in and out. The farcall helper function behaves similarly, but takes as its first argument the CS:IP (in the form (CS << 16) + IP) of procedure to be invoked via a FAR CALL. ++++ SYSLINUX API CALLS +++ SYSLINUX provides the following API calls. SYSLINUX 1.xx only supported INT 20h - terminate program. [] indicates the first version of SYSLINUX which supported this feature (correctly.) NOTE: Most of the API functionality is still experimental. Expect to find bugs. ++++ DOS-COMPATIBLE API CALLS ++++ INT 20h [1.48] Terminate program INT 21h AH=00h [2.00] Terminate program INT 21h AH=4Ch [2.00] Terminate program All of these terminate the program. INT 21h AH=01h [2.01] Get Key with Echo Reads a key from the console input, with echo to the console output. The read character is returned in AL. Extended characters received from the keyboard are returned as NUL (00h) + the extended character code. INT 21h AH=02h [2.01] Write Character Writes a character in DL to the console (video and serial) output. INT 21h AH=04h [2.01] Write Character to Serial Port Writes a character in DL to the serial console output (if enabled.) If no serial port is configured, this routine does nothing. INT 21h AH=08h [2.09] Get Key without Echo Reads a key fron the console input, without echoing it to the console output. The read character is returned in AL. INT 21h AH=09h [2.01] Write DOS String to Console Writes a DOS $-terminated string in DS:DX to the console. INT 21h AH=0Bh [2.00] Check Keyboard Returns AL=FFh if there is a keystroke waiting (which can then be read with INT 21h, AH=01h or AH=08h), otherwise AL=00h. INT 21h AH=30h [2.00] Check DOS Version This function returns AX=BX=CX=DX=0, corresponding to a hypothetical "DOS 0.0", but the high parts of EAX-EBX-ECX-EDX spell "SYSLINUX": EAX=59530000h EBX=4C530000h ECX=4E490000h EDX=58550000h This function can thus be used to distinguish running on SYSLINUX from running on DOS. ++++ SYSLINUX-SPECIFIC API CALLS ++++ SYSLINUX-specific API calls are executed using INT 22h, with a function number in AX. INT 22h is used by DOS for internal purposes; do not execute INT 22h under DOS. DOS-compatible function INT 21h, AH=30h can be used to detect if the SYSLINUX API calls are available. Any register not specifically listed as modified is preserved; however, future versions of SYSLINUX may add additional output registers to existing calls. All calls return CF=0 on success, CF=1 on failure. The noted outputs apply if CF=0 only unless otherwise noted. All calls clobber the arithmetric flags (CF, PF, AF, ZF, SF and OF) but leave all other flags unchanged unless otherwise noted. AX=0001h [2.00] Get Version Input: AX 0001h Output: AX number of INT 22h API functions available CH SYSLINUX major version number CL SYSLINUX minor version number DL SYSLINUX derivative ID (e.g. 32h = PXELINUX) ES:SI SYSLINUX version string ES:DI SYSLINUX copyright string This API call returns the SYSLINUX version and API information. AX=0002h [2.01] Write String Input: AX 0002h ES:BX null-terminated string Output: None Writes a null-terminated string on the console. AX=0003h [2.01] Run command Input: AX 0003h ES:BX null-terminated command string Output: Does not return This API call terminates the program and executes the command string as if the user had entered it at the SYSLINUX command line. This API call does not return. AX=0004h [2.01] Run default command Input: AX 0004h Output: Does not return This API call terminates the program and executes the default command string as if the user had pressed Enter alone on the SYSLINUX command line. This API call does not return. AX=0005h [2.00] Force text mode Input: AX 0005h Output: None If the screen was in graphics mode (due to displaying a splash screen using the command in a message file, or similar), return to text mode. AX=0006h [2.08] Open file Input: AX 0006h ES:SI null-terminated filename Output: SI file handle EAX length of file in bytes CX file block size Open a file for reading. The exact syntax of the filenames allowed depends on the particular SYSLINUX derivative. The SYSLINUX file system is block-oriented. The size of a block will always be a power of two and no greater than 16K. Note: SYSLINUX considers a zero-length file to be nonexistent. AX=0007h [2.08] Read file Input: AX 0007h SI file handle ES:BX buffer CX number of blocks to read Output: SI file handle, or 0 if EOF was reached Read blocks from a file. Note that the file handle that is returned in SI may not be the same value that was passed in. If end of file was reached (SI=0), the file was automatically closed. The address of the buffer (ES:BX) should be at least 512-byte aligned. SYSLINUX guarantees at least this alignment for the COMBOOT load segment or the COM32 bounce buffer. Keep in mind that a "file" may be a TFTP connection, and that leaving a file open for an extended period of time may result in a timeout. WARNING: Calling this function with an invalid file handle will probably crash the system. AX=0008h [2.08] Close file Input: AX 0008h SI file handle Output: None Close a file before reaching the end of file. WARNING: Calling this function with an invalid file handle will probably crash the system. AX=0009h [2.00] Call PXE Stack [PXELINUX ONLY] Input: AX 0009h BX PXE function number ES:DI PXE data buffer Output: AX PXE return status code Invoke an arbitrary PXE stack function. On SYSLINUX/ISOLINUX, this function returns with an error (CF=1) and no action is taken. On PXELINUX, this function always returns with CF=0 indicating that the PXE stack was successfully invoked; check the status code in AX and in the first word of the data buffer to determine if the PXE call succeeded or not. The PXE stack will have the UDP stack OPEN; if you change that you cannot call any of the file-related API functions, and must restore UDP OPEN before returning to PXELINUX. PXELINUX reserves UDP port numbers from 49152 to 65535 for its own use; port numbers below that range is available. AX=000Ah [2.00] Get Derivative-Specific Information [SYSLINUX, EXTLINUX] Input: AX 000Ah Output: AL 31h (SYSLINUX), 34h (EXTLINUX) DL drive number ES:BX pointer to partition table entry (if DL >= 80h) Note: This function was broken in EXTLINUX 3.00-3.02. [PXELINUX] Input: AX 000Ah Output: AL 32h (PXELINUX) DX PXE API version detected (DH=major, DL=minor) ES:BX pointer to PXENV+ or !PXE structure FS:SI pointer to original stack with invocation record Note: DX notes the API version detected by PXELINUX, which may be more conservative than the actual version available. For exact information examine the API version entry in the PXENV+ structure, or the API version entries in the ROMID structures pointed from the !PXE structure. PXELINUX will use, and provide, the !PXE structure over the PXENV+ structure. Examine the structure signature to determine which particular structure was provided. The FS:SI pointer points to the top of the original stack provided by the PXE stack, with the following values pushed at the time PXELINUX is started: [fs:si+0] GS <- top of stack [fs:si+2] FS [fs:si+4] ES [fs:si+6] DS [fs:si+8] EDI [fs:si+12] ESI [fs:si+16] EBP [fs:si+20] - [fs:si+24] EBX [fs:si+28] EDX [fs:si+32] ECX [fs:si+36] EAX [fs:si+40] EFLAGS [fs:si+44] PXE return IP <- t.o.s. when PXELINUX invoked [fs:si+46] PXE return CS [ISOLINUX] Input: AX 000Ah Output: AL 33h (ISOLINUX) DL drive number ES:BX pointer to El Torito spec packet Note: Some very broken El Torito implementations do not provide the spec packet information. If so, ES:BX may point to all zeroes or to garbage. Call INT 13h, AX=4B01h to obtain the spec packet directly from the BIOS if necessary. This call gives information specific to a particular SYSLINUX derivative. The value returned in AL is the same as is returned in DL by INT 22h AX=0001h. AX=000Bh [2.00] Get Serial Console Configuration Input: AX 000Bh Output: DX serial port I/O base (e.g. 3F8h = COM1...) CX baud rate divisor (1 = 115200 bps, 2 = 57600 bps...) BX flow control configuration bits (see syslinux.doc) -> bit 15 is set if the video console is disabled If no serial port is configured, DX will be set to 0 and the other registers are undefined. AX=000Ch [2.00] Perform final cleanup Input: AX 000Ch DX derivative-specific flags (0000h = clean up all) Output: Does not return This routine performs any "final cleanup" the boot loader would normally perform before loading a kernel, such as unloading the PXE stack in the case of PXELINUX. AFTER INVOKING THIS CALL, NO OTHER API CALLS MAY BE INVOKED, NOR MAY THE PROGRAM TERMINATE AND RETURN TO THE BOOT LOADER. This call basically tells the boot loader "get out of the way, I'll handle it from here." For COM32 images, the boot loader will continue to provide interrupt and BIOS call thunking services as long its memory areas (0x0800-0xffff, 0x100000-0x100fff) are not overwritten. MAKE SURE TO DISABLE INTERRUPTS, AND INSTALL NEW GDT AND IDTS BEFORE OVERWRITING THESE MEMORY AREAS. The permissible values for DX is an OR of these values: SYSLINUX: 0000h Normal cleanup PXELINUX: 0000h Normal cleanup 0003h Keep UNDI and PXE stacks loaded ISOLINUX: 0000h Normal cleanup EXTLINUX: 0000h Normal cleanup All other values are undefined, and may have different meanings in future versions of SYSLINUX. AX=000Dh [2.08] Cleanup and replace bootstrap code Input: AX 000Dh DX derivative-specific flags (see previous function) EDI bootstrap code (linear address, can be in high memory) ECX bootstrap code length in bytes (must fit in low mem) EBX(!) initial value of EDX after bootstrap ESI initial value of ESI after bootstrap DS initial value of DS after bootstrap Output: Does not return This routine performs final cleanup, then takes a piece of code, copies it over the primary bootstrap at address 7C00h, and jumps to it. This can be used to chainload boot sectors, MBRs, bootstraps, etc. Normal boot sectors expect DL to contain the drive number, and, for hard drives (DL >= 80h) DS:SI to contain a pointer to the 16-byte partition table entry. The memory between 600h-7FFh is available to put the partition table entry in. For PXELINUX, if the PXE stack is not unloaded, all registers (except DS, ESI and EDX) and the stack will be set up as they were set up by the PXE ROM. AX=000Eh [2.11] Get configuration file name Input: AX 0000Eh Output: ES:BX null-terminated file name string Returns the name of the configuration file. Note that it is possible that the configuration file doesn't actually exist. AX=000Fh [3.00] Get IPAPPEND strings [PXELINUX] Input: AX 000Fh Output: CX number of strings (currently 2) ES:BX pointer to an array of NEAR pointers in the same segment, one for each of the above strings Returns the same strings that the "ipappend" option would have added to the command line, one for each bit of the "ipappend" flag value, so entry 0 is the "ip=" string and entry 1 is the "BOOTIF=" string. AX=0010h [3.00] Resolve hostname [PXELINUX] Input: ES:BX pointer to null-terminated hostname Output: EAX IP address of hostname (zero if not found) Queries the DNS server(s) for a specific hostname. If the hostname does not contain a dot (.), the local domain name is automatically appended. This function only return CF=1 if the function is not supported. If the function is supported, but the hostname did not resolve, it returns with CF=0, EAX=0. The IP address is returned in network byte order, i.e. if the IP address is 1.2.3.4, EAX will contain 0x04030201. Note that all uses of IP addresses in PXE are also in network byte order. AX=0011h [3.05] Maximum number of shuffle descriptors Input: AX 0011h Output: CX maximum number of descriptors This routine reports the maximum number of shuffle descriptors permitted in a call to function 0012h. Typical values are 682 and 1365. AX=0012h [3.05] Cleanup, shuffle and boot Input: AX 0012h DX derivative-specific flags (see previous function) ES:DI shuffle descriptor list (must be in low memory) CX number of shuffle descriptors EBX(!) initial value of EDX after bootstrap ESI initial value of ESI after bootstrap DS initial value of DS after bootstrap EBP CS:IP of routine to jump to Output: Does not return (if CX is too large the routine returns with CF=1) This routine performs final cleanup, then performs a sequence of copies, and jumps to a specified real mode entry point. This is a more general version of function 000Dh, which can also be used to load other types of programs. The copies must not touch memory below address 7C00h. ES:DI points to a list of CX descriptors each of the form: Offset Size Meaning 0 dword destination address 4 dword source address 8 dword length in bytes The copies are overlap-safe, like memmove(). Normal boot sectors expect DL to contain the drive number, and, for hard drives (DL >= 80h) DS:SI to contain a pointer to the 16-byte partition table entry. The memory between 600h-7FFh is available to put the partition table entry in. For PXELINUX, if the PXE stack is not unloaded, all registers (except DS, ESI and EDX) and the stack will be set up as they were set up by the PXE ROM. AX=0013h [3.08] Idle loop call Input: AX 0013h Output: None Call this routine while sitting in an idle loop. It performs any periodic activities required by the filesystem code. At the moment, this is a no-op on all derivatives except PXELINUX, where it executes PXE calls to answer ARP queries. Starting with version 3.10, this API call harmlessly returns failure (CF=1) if invoked on a platform which does not need idle calls. Additionally, it's safe to call this API call on previous SYSLINUX versions (2.00 or later); it will just harmlessly fail. Thus, if this call returns failure (CF=1), it means that there is no technical reason to call this function again, although doing so is of course safe. AX=0014h [3.10] Local boot [PXELINUX, ISOLINUX] Input: AX 0014h DX Local boot parameter Output: Does not return This function invokes the equivalent of the "localboot" configuration file option. The parameter in DX is the same parameter as would be entered after "localboot" in the configuration file; this parameter is derivative-specific -- see syslinux.doc for the definition. AX=0015h [3.10] Get feature flags Input: AX 0015h Output: ES:BX pointer to flags in memory CX number of flag bytes This function reports whether or not this SYSLINUX version and derivative supports specific features. Keep in mind that future versions might have more bits; remember to treat any bits beyond the end of the array (as defined by the value in CX) as zero. Currently the following feature flag is defined: Byte Bit Definition ---------------------------------------------------- 0 0 Local boot (AX=0014h) supported 1 Idle loop call (AX=0013h) is a no-op All other flags are reserved. AX=0016h [3.10] Run kernel image Input: AX 0016h DS:SI Filename of kernel image (zero-terminated string) ES:BX Command line (zero-terminated string) ECX IPAPPEND flags [PXELINUX] EDX Reserved - MUST BE ZERO Output: Does not return if successful; returns with CF=1 if the kernel image is not found. This function is similiar to AX=0003h Run command, except that the filename and command line are treated as if specified in a KERNEL and APPEND statement of a LABEL statement, which means: - The filename has to be exact; no variants are tried; - No global APPEND statement is applied; - ALLOWOPTIONS and IMPLICIT statements in the configuration file do not apply. It is therefore important that the COMBOOT module doesn't allow the end user to violate the intent of the administrator. Additionally, this function returns with a failure if the file doesn't exist, instead of returning to the command line. (It may still return to the command line if the image is somehow corrupt, however.)