ref: 2af6b0896040ab24c2d411a36781d7588935c6cf
dir: /sys/man/6/a.out/
.TH A.OUT 6
.SH NAME
a.out \- object file format
.SH SYNOPSIS
.B #include <a.out.h>
.SH DESCRIPTION
An executable Plan 9 binary file has up to six sections:
a header, the program text, the data,
a symbol table, a PC/SP offset table (MC68020 only),
and finally a PC/line number table.
The header, given by a structure in
.BR <a.out.h> ,
contains 4-byte integers in big-endian order:
.PP
.EX
.ta \w'#define 'u +\w'_MAGIC(b) 'u +\w'_MAGIC(10) 'u +4n +4n +4n +4n
typedef struct Exec {
long magic; /* magic number */
long text; /* size of text segment */
long data; /* size of initialized data */
long bss; /* size of uninitialized data */
long syms; /* size of symbol table */
long entry; /* entry point */
long spsz; /* size of pc/sp offset table */
long pcsz; /* size of pc/line number table */
} Exec;
#define HDR_MAGIC 0x00008000
#define _MAGIC(f, b) ((f)|((((4*(b))+0)*(b))+7))
#define A_MAGIC _MAGIC(0, 8) /* 68020 */
#define I_MAGIC _MAGIC(0, 11) /* intel 386 */
#define J_MAGIC _MAGIC(0, 12) /* intel 960 (retired) */
#define K_MAGIC _MAGIC(0, 13) /* sparc */
#define V_MAGIC _MAGIC(0, 16) /* mips 3000 BE */
#define X_MAGIC _MAGIC(0, 17) /* att dsp 3210 (retired) */
#define M_MAGIC _MAGIC(0, 18) /* mips 4000 BE */
#define D_MAGIC _MAGIC(0, 19) /* amd 29000 (retired) */
#define E_MAGIC _MAGIC(0, 20) /* arm */
#define Q_MAGIC _MAGIC(0, 21) /* powerpc */
#define N_MAGIC _MAGIC(0, 22) /* mips 4000 LE */
#define L_MAGIC _MAGIC(0, 23) /* dec alpha (retired) */
#define P_MAGIC _MAGIC(0, 24) /* mips 3000 LE */
#define U_MAGIC _MAGIC(0, 25) /* sparc64 */
#define S_MAGIC _MAGIC(HDR_MAGIC, 26) /* amd64 */
#define T_MAGIC _MAGIC(HDR_MAGIC, 27) /* powerpc64 */
#define R_MAGIC _MAGIC(HDR_MAGIC, 28) /* arm64 */
.EE
.DT
.PP
Sizes are expressed in bytes.
The size of the header is not included in any of the other sizes.
.PP
When a Plan 9 binary file is executed,
a memory image of three segments is
set up: the text segment, the data segment, and the stack.
The text segment begins at a virtual address which is
a multiple of the machine-dependent page size.
The text segment consists of the header and the first
.B text
bytes of the binary file.
The
.B entry
field gives the virtual address of the entry point of the program.
The data segment starts at the first page-rounded virtual address
after the text segment.
It consists of the next
.B data
bytes of the binary file, followed by
.B bss
bytes initialized to zero.
The stack occupies the highest possible locations
in the core image, automatically growing downwards.
The bss segment may be extended by
.IR brk (2).
.PP
The next
.B syms
(possibly zero)
bytes of the file contain symbol table
entries, each laid out as:
.IP
.EX
uchar value[4];
char type;
char name[\f2n\fP]; /* NUL-terminated */
.EE
.PP
The
.B value
is in big-endian order and
the size of the
.B name
field is not pre-defined: it is a zero-terminated array of
variable length.
.PP
The
.B type
field is one of the following characters with the high bit set:
.RS
.TP
.B T
text segment symbol
.PD0
.TP
.B t
static text segment symbol
.TP
.B L
leaf function text segment symbol
.TP
.B l
static leaf function text segment symbol
.TP
.B D
data segment symbol
.TP
.B d
static data segment symbol
.TP
.B B
bss segment symbol
.TP
.B b
static bss segment symbol
.TP
.B a
automatic (local) variable symbol
.TP
.B p
function parameter symbol
.RE
.PD
.PP
A few others are described below.
The symbols in the symbol table appear in the same order
as the program components they describe.
.PP
The Plan 9 compilers implement a virtual stack frame pointer rather
than dedicating a register;
moreover, on the MC680X0 architectures
there is a variable offset between the stack pointer and the
frame pointer.
Following the symbol table,
MC680X0 executable files contain a
.BR spsz -byte
table encoding the offset
of the stack frame pointer as a function of program location;
this section is not present for other architectures.
The PC/SP table is encoded as a byte stream.
By setting the PC to the base of the text segment
and the offset to zero and interpreting the stream,
the offset can be computed for any PC.
A byte value of 0 is followed by four bytes that hold, in big-endian order,
a constant to be added to the offset.
A byte value of 1 to 64 is multiplied by four and added, without sign
extension, to the offset.
A byte value of 65 to 128 is reduced by 64, multiplied by four, and
subtracted from the offset.
A byte value of 129 to 255 is reduced by 129, multiplied by the quantum
of instruction size
(e.g. two on the MC680X0),
and added to the current PC without changing the offset.
After any of these operations, the instruction quantum is added to the PC.
.PP
A similar table, occupying
.BR pcsz -bytes,
is the next section in an executable; it is present for all architectures.
The same algorithm may be run using this table to
recover the absolute source line number from a given program location.
The absolute line number (starting from zero) counts the newlines
in the C-preprocessed source seen by the compiler.
Three symbol types in the main symbol table facilitate conversion of the absolute
number to source file and line number:
.RS
.TP
.B f
source file name components
.TP
.B z
source file name
.TP
.B Z
source file line offset
.RE
.PP
The
.B f
symbol associates an integer (the
.B value
field of the `symbol') with
a unique file path name component (the
.B name
of the `symbol').
These path components are used by the
.B z
symbol to represent a file name: the
first byte of the name field is always 0; the remaining
bytes hold a zero-terminated array of 16-bit values (in big-endian order)
that represent file name components from
.B f
symbols.
These components, when separated by slashes, form a file name.
The initial slash of a file name is recorded in the symbol table by an
.B f
symbol; when forming file names from
.B z
symbols an initial slash is not to be assumed.
The
.B z
symbols are clustered, one set for each object file in the program,
before any text symbols from that object file.
The set of
.B z
symbols for an object file form a
.I history stack
of the included source files from which the object file was compiled.
The value associated with each
.B z
symbol is the absolute line number at which that file was included in the source;
if the name associated with the
.B z
symbol is null, the symbol represents the end of an included file, that is,
a pop of the history stack.
If the value of the
.B z
symbol is 1 (one),
it represents the start of a new history stack.
To recover the source file and line number for a program location,
find the text symbol containing the location
and then the first history stack preceding the text symbol in the symbol table.
Next, interpret the PC/line offset table to discover the absolute line number
for the program location.
Using the line number, scan the history stack to find the set of source
files open at that location.
The line number within the file can be found using the line numbers
in the history stack.
The
.B Z
symbols correspond to
.B #line
directives in the source; they specify an adjustment to the line number
to be printed by the above algorithm. The offset is associated with the
first previous
.B z
symbol in the symbol table.
.SH "SEE ALSO"
.IR db (1),
.IR acid (1),
.IR 2a (1),
.IR 2l (1),
.IR nm (1),
.IR strip (1),
.IR mach (2),
.IR symbol (2)
.SH BUGS
There is no type information in the symbol table; however, the
.B -a
flags on the compilers will produce symbols for
.IR acid (1).