NASM is a portable assembler, designed to be able to compile on any ANSI
C-supporting platform and produce output to run on a variety of Intel x86
operating systems. For this reason, it has a large number of available
output formats, selected using the -f
option on the NASM
command line. Each of these formats, along with its extensions to the base
NASM syntax, is detailed in this chapter.
As stated in section 2.1.1,
NASM chooses a default name for your output file based on the input file
name and the chosen output format. This will be generated by removing the
extension (.asm
, .s
, or whatever you like to use)
from the input file name, and substituting an extension defined by the
output format. The extensions are given with each format below.
bin
: Flat-Form Binary OutputThe bin
format does not produce object files: it generates
nothing in the output file except the code you wrote. Such `pure binary'
files are used by MS-DOS: .COM
executables and
.SYS
device drivers are pure binary files. Pure binary output
is also useful for operating system and boot loader development.
The bin
format supports multiple section names. For details
of how NASM handles sections in the bin
format, see
section 7.1.3.
Using the bin
format puts NASM by default into 16-bit mode
(see section 6.1). In order to use
bin
to write 32-bit or 64-bit code, such as an OS kernel, you
need to explicitly issue the BITS 32
or BITS 64
directive.
bin
has no default output file name extension: instead, it
leaves your file name as it is once the original extension has been
removed. Thus, the default is for NASM to assemble binprog.asm
into a binary file called binprog
.
ORG
: Binary File Program OriginThe bin
format provides an additional directive to the list
given in chapter 6: ORG
. The
function of the ORG
directive is to specify the origin address
which NASM will assume the program begins at when it is loaded into memory.
For example, the following code will generate the longword
0x00000104
:
org 0x100 dd label label:
Unlike the ORG
directive provided by MASM-compatible
assemblers, which allows you to jump around in the object file and
overwrite code you have already generated, NASM's ORG
does
exactly what the directive says: origin. Its sole function is to
specify one offset which is added to all internal address references within
the section; it does not permit any of the trickery that MASM's version
does. See section 12.1.3 for
further comments.
bin
Extensions to the SECTION
DirectiveThe bin
output format extends the SECTION
(or
SEGMENT
) directive to allow you to specify the alignment
requirements of segments. This is done by appending the ALIGN
qualifier to the end of the section-definition line. For example,
section .data align=16
switches to the section .data
and also specifies that it
must be aligned on a 16-byte boundary.
The parameter to ALIGN
specifies how many low bits of the
section start address must be forced to zero. The alignment value given may
be any power of two.
bin
FormatThe bin
format allows the use of multiple sections, of
arbitrary names, besides the "known" .text
,
.data
, and .bss
names.
Sections may be designated progbits
or nobits
.
Default is progbits
(except .bss
, which defaults
to nobits
, of course).
Sections can be aligned at a specified boundary following the previous
section with align=
, or at an arbitrary byte-granular position
with start=
.
Sections can be given a virtual start address, which will be used for
the calculation of all memory references within that section with
vstart=
.
Sections can be ordered using
follows=
<section>
or
vfollows=
<section>
as an alternative to
specifying an explicit start address.
Arguments to org
, start
, vstart
,
and align=
are critical expressions. See
section 3.8. E.g.
align=(1 << ALIGN_SHIFT)
–
ALIGN_SHIFT
must be defined before it is used here.
Any code which comes before an explicit SECTION
directive
is directed by default into the .text
section.
If an ORG
statement is not given, ORG 0
is
used by default.
The .bss
section will be placed after the last
progbits
section, unless start=
,
vstart=
, follows=
, or vfollows=
has
been specified.
All sections are aligned on dword boundaries, unless a different alignment has been specified.
Sections may not overlap.
NASM creates the section.<secname>.start
for each
section, which may be used in your code.
Map files can be generated in -f bin
format by means of the
[map]
option. Map types of all
(default),
brief
, sections
, segments
, or
symbols
may be specified. Output may be directed to
stdout
(default), stderr
, or a specified file.
E.g. [map symbols myfile.map]
. No "user form" exists, the
square brackets must be used.
ith
: Intel Hex OutputThe ith
file format produces Intel hex-format files. Just
as the bin
format, this is a flat memory image format with no
support for relocation or linking. It is usually used with ROM programmers
and similar utilities.
All extensions supported by the bin
file format is also
supported by the ith
file format.
ith
provides a default output file-name extension of
.ith
.
srec
: Motorola S-Records OutputThe srec
file format produces Motorola S-records files.
Just as the bin
format, this is a flat memory image format
with no support for relocation or linking. It is usually used with ROM
programmers and similar utilities.
All extensions supported by the bin
file format is also
supported by the srec
file format.
srec
provides a default output file-name extension of
.srec
.
obj
: Microsoft OMF Object FilesThe obj
file format (NASM calls it obj
rather
than omf
for historical reasons) is the one produced by MASM
and TASM, which is typically fed to 16-bit DOS linkers to produce
.EXE
files. It is also the format used by OS/2.
obj
provides a default output file-name extension of
.obj
.
obj
is not exclusively a 16-bit format, though: NASM has
full support for the 32-bit extensions to the format. In particular, 32-bit
obj
format files are used by Borland's Win32 compilers,
instead of using Microsoft's newer win32
object file format.
The obj
format does not define any special segment names:
you can call your segments anything you like. Typical names for segments in
obj
format files are CODE
, DATA
and
BSS
.
If your source file contains code before specifying an explicit
SEGMENT
directive, then NASM will invent its own segment
called __NASMDEFSEG
for you.
When you define a segment in an obj
file, NASM defines the
segment name as a symbol as well, so that you can access the segment
address of the segment. So, for example:
segment data dvar: dw 1234 segment code function: mov ax,data ; get segment address of data mov ds,ax ; and move it into DS inc word [dvar] ; now this reference will work ret
The obj
format also enables the use of the SEG
and WRT
operators, so that you can write code which does
things like
extern foo mov ax,seg foo ; get preferred segment of foo mov ds,ax mov ax,data ; a different segment mov es,ax mov ax,[ds:foo] ; this accesses `foo' mov [es:foo wrt data],bx ; so does this
obj
Extensions to the SEGMENT
DirectiveThe obj
output format extends the SEGMENT
(or
SECTION
) directive to allow you to specify various properties
of the segment you are defining. This is done by appending extra qualifiers
to the end of the segment-definition line. For example,
segment code private align=16
defines the segment code
, but also declares it to be a
private segment, and requires that the portion of it described in this code
module must be aligned on a 16-byte boundary.
The available qualifiers are:
PRIVATE
, PUBLIC
, COMMON
and
STACK
specify the combination characteristics of the segment.
PRIVATE
segments do not get combined with any others by the
linker; PUBLIC
and STACK
segments get
concatenated together at link time; and COMMON
segments all
get overlaid on top of each other rather than stuck end-to-end.
ALIGN
is used, as shown above, to specify how many low bits
of the segment start address must be forced to zero. The alignment value
given may be any power of two from 1 to 4096; in reality, the only values
supported are 1, 2, 4, 16, 256 and 4096, so if 8 is specified it will be
rounded up to 16, and 32, 64 and 128 will all be rounded up to 256, and so
on. Note that alignment to 4096-byte boundaries is a PharLap extension to
the format and may not be supported by all linkers.
CLASS
can be used to specify the segment class; this
feature indicates to the linker that segments of the same class should be
placed near each other in the output file. The class name can be any word,
e.g. CLASS=CODE
.
OVERLAY
, like CLASS
, is specified with an
arbitrary word as an argument, and provides overlay information to an
overlay-capable linker.
Segments can be declared as USE16
or USE32
,
which has the effect of recording the choice in the object file and also
ensuring that NASM's default assembly mode when assembling in that segment
is 16-bit or 32-bit respectively.
When writing OS/2 object files, you should declare 32-bit segments as
FLAT
, which causes the default segment base for anything in
the segment to be the special group FLAT
, and also defines the
group if it is not already defined.
The obj
file format also allows segments to be declared as
having a pre-defined absolute segment address, although no linkers are
currently known to make sensible use of this feature; nevertheless, NASM
allows you to declare a segment such as
SEGMENT SCREEN ABSOLUTE=0xB800
if you need to. The
ABSOLUTE
and ALIGN
keywords are mutually
exclusive.
NASM's default segment attributes are PUBLIC
,
ALIGN=1
, no class, no overlay, and USE16
.
GROUP
: Defining Groups of SegmentsThe obj
format also allows segments to be grouped, so that
a single segment register can be used to refer to all the segments in a
group. NASM therefore supplies the GROUP
directive, whereby
you can code
segment data ; some data segment bss ; some uninitialized data group dgroup data bss
which will define a group called dgroup
to contain the
segments data
and bss
. Like SEGMENT
,
GROUP
causes the group name to be defined as a symbol, so that
you can refer to a variable var
in the data
segment as var wrt data
or as var wrt dgroup
,
depending on which segment value is currently in your segment register.
If you just refer to var
, however, and var
is
declared in a segment which is part of a group, then NASM will default to
giving you the offset of var
from the beginning of the
group, not the segment. Therefore SEG var
,
also, will return the group base rather than the segment base.
NASM will allow a segment to be part of more than one group, but will generate a warning if you do this. Variables declared in a segment which is part of more than one group will default to being relative to the first group that was defined to contain the segment.
A group does not have to contain any segments; you can still make
WRT
references to a group which does not contain the variable
you are referring to. OS/2, for example, defines the special group
FLAT
with no segments in it.
UPPERCASE
: Disabling Case Sensitivity in OutputAlthough NASM itself is case sensitive, some OMF linkers are not;
therefore it can be useful for NASM to output single-case object files. The
UPPERCASE
format-specific directive causes all segment, group
and symbol names that are written to the object file to be forced to upper
case just before being written. Within a source file, NASM is still
case-sensitive; but the object file can be written entirely in upper case
if desired.
UPPERCASE
is used alone on a line; it requires no
parameters.
IMPORT
: Importing DLL SymbolsThe IMPORT
format-specific directive defines a symbol to be
imported from a DLL, for use if you are writing a DLL's import library in
NASM. You still need to declare the symbol as EXTERN
as well
as using the IMPORT
directive.
The IMPORT
directive takes two required parameters,
separated by white space, which are (respectively) the name of the symbol
you wish to import and the name of the library you wish to import it from.
For example:
import WSAStartup wsock32.dll
A third optional parameter gives the name by which the symbol is known in the library you are importing it from, in case this is not the same as the name you wish the symbol to be known by to your code once you have imported it. For example:
import asyncsel wsock32.dll WSAAsyncSelect
EXPORT
: Exporting DLL SymbolsThe EXPORT
format-specific directive defines a global
symbol to be exported as a DLL symbol, for use if you are writing a DLL in
NASM. You still need to declare the symbol as GLOBAL
as well
as using the EXPORT
directive.
EXPORT
takes one required parameter, which is the name of
the symbol you wish to export, as it was defined in your source file. An
optional second parameter (separated by white space from the first) gives
the external name of the symbol: the name by which you wish the
symbol to be known to programs using the DLL. If this name is the same as
the internal name, you may leave the second parameter off.
Further parameters can be given to define attributes of the exported symbol. These parameters, like the second, are separated by white space. If further parameters are given, the external name must also be specified, even if it is the same as the internal name. The available attributes are:
resident
indicates that the exported name is to be kept
resident by the system loader. This is an optimisation for frequently used
symbols imported by name.
nodata
indicates that the exported symbol is a function
which does not make use of any initialized data.
parm=NNN
, where NNN
is an integer, sets the
number of parameter words for the case in which the symbol is a call gate
between 32-bit and 16-bit segments.
An attribute which is just a number indicates that the symbol should be exported with an identifying number (ordinal), and gives the desired number.
For example:
export myfunc export myfunc TheRealMoreFormalLookingFunctionName export myfunc myfunc 1234 ; export by ordinal export myfunc myfunc resident parm=23 nodata
..start
: Defining the Program Entry PointOMF
linkers require exactly one of the object files being
linked to define the program entry point, where execution will begin when
the program is run. If the object file that defines the entry point is
assembled using NASM, you specify the entry point by declaring the special
symbol ..start
at the point where you wish execution to begin.
obj
Extensions to the EXTERN
DirectiveIf you declare an external symbol with the directive
extern foo
then references such as mov ax,foo
will give you the offset
of foo
from its preferred segment base (as specified in
whichever module foo
is actually defined in). So to access the
contents of foo
you will usually need to do something like
mov ax,seg foo ; get preferred segment base mov es,ax ; move it into ES mov ax,[es:foo] ; and use offset `foo' from it
This is a little unwieldy, particularly if you know that an external is
going to be accessible from a given segment or group, say
dgroup
. So if DS
already contained
dgroup
, you could simply code
mov ax,[foo wrt dgroup]
However, having to type this every time you want to access
foo
can be a pain; so NASM allows you to declare
foo
in the alternative form
extern foo:wrt dgroup
This form causes NASM to pretend that the preferred segment base of
foo
is in fact dgroup
; so the expression
seg foo
will now return dgroup
, and the
expression foo
is equivalent to foo wrt dgroup
.
This default-WRT
mechanism can be used to make externals
appear to be relative to any group or segment in your program. It can also
be applied to common variables: see section
7.4.8.
obj
Extensions to the COMMON
DirectiveThe obj
format allows common variables to be either near or
far; NASM allows you to specify which your variables should be by the use
of the syntax
common nearvar 2:near ; `nearvar' is a near common common farvar 10:far ; and `farvar' is far
Far common variables may be greater in size than 64Kb, and so the OMF specification says that they are declared as a number of elements of a given size. So a 10-byte far common variable could be declared as ten one-byte elements, five two-byte elements, two five-byte elements or one ten-byte element.
Some OMF
linkers require the element size, as well as the
variable size, to match when resolving common variables declared in more
than one module. Therefore NASM must allow you to specify the element size
on your far common variables. This is done by the following syntax:
common c_5by2 10:far 5 ; two five-byte elements common c_2by5 10:far 2 ; five two-byte elements
If no element size is specified, the default is 1. Also, the
FAR
keyword is not required when an element size is specified,
since only far commons may have element sizes at all. So the above
declarations could equivalently be
common c_5by2 10:5 ; two five-byte elements common c_2by5 10:2 ; five two-byte elements
In addition to these extensions, the COMMON
directive in
obj
also supports default-WRT
specification like
EXTERN
does (explained in section
7.4.7). So you can also declare things like
common foo 10:wrt dgroup common bar 16:far 2:wrt data common baz 24:wrt data:6
Since NASM 2.13.02, obj
files contain embedded dependency
file information. To suppress the generation of dependencies, use
%pragma obj nodepend
win32
: Microsoft Win32 Object FilesThe win32
output format generates Microsoft Win32 object
files, suitable for passing to Microsoft linkers such as Visual C++. Note
that Borland Win32 compilers do not use this format, but use
obj
instead (see section 7.4).
win32
provides a default output file-name extension of
.obj
.
Note that although Microsoft say that Win32 object files follow the
COFF
(Common Object File Format) standard, the object files
produced by Microsoft Win32 compilers are not compatible with COFF linkers
such as DJGPP's, and vice versa. This is due to a difference of opinion
over the precise semantics of PC-relative relocations. To produce COFF
files suitable for DJGPP, use NASM's coff
output format;
conversely, the coff
format does not produce object files that
Win32 linkers can generate correct output from.
win32
Extensions to the SECTION
DirectiveLike the obj
format, win32
allows you to
specify additional information on the SECTION
directive line,
to control the type and properties of sections you declare. Section types
and properties are generated automatically by NASM for the standard section
names .text
, .data
and .bss
, but may
still be overridden by these qualifiers.
The available qualifiers are:
code
, or equivalently text
, defines the
section to be a code section. This marks the section as readable and
executable, but not writable, and also indicates to the linker that the
type of the section is code.
data
and bss
define the section to be a data
section, analogously to code
. Data sections are marked as
readable and writable, but not executable. data
declares an
initialized data section, whereas bss
declares an
uninitialized data section.
rdata
declares an initialized data section that is readable
but not writable. Microsoft compilers use this section to place constants
in it.
info
defines the section to be an informational section,
which is not included in the executable file by the linker, but may (for
example) pass information to the linker. For example, declaring an
info
–type section called .drectve
causes
the linker to interpret the contents of the section as command-line
options.
align=
, used with a trailing number as in obj
,
gives the alignment requirements of the section. The maximum you may
specify is 64: the Win32 object file format contains no means to request a
greater section alignment than this. If alignment is not explicitly
specified, the defaults are 16-byte alignment for code sections, 8-byte
alignment for rdata sections and 4-byte alignment for data (and BSS)
sections. Informational sections get a default alignment of 1 byte (no
alignment), though the value does not matter.
The defaults assumed by NASM if you do not specify the above qualifiers are:
section .text code align=16 section .data data align=4 section .rdata rdata align=8 section .bss bss align=4
Any other section name is treated by default like .text
.
win32
: Safe Structured Exception HandlingAmong other improvements in Windows XP SP2 and Windows Server 2003
Microsoft has introduced concept of "safe structured exception handling."
General idea is to collect handlers' entry points in designated read-only
table and have alleged entry point verified against this table prior
exception control is passed to the handler. In order for an executable
module to be equipped with such "safe exception handler table," all object
modules on linker command line has to comply with certain criteria. If one
single module among them does not, then the table in question is omitted
and above mentioned run-time checks will not be performed for application
in question. Table omission is by default silent and therefore can be
easily overlooked. One can instruct linker to refuse to produce binary
without such table by passing /safeseh
command line option.
Without regard to this run-time check merits it's natural to expect NASM
to be capable of generating modules suitable for /safeseh
linking. From developer's viewpoint the problem is two-fold:
how to adapt modules not deploying exception handlers of their own;
how to adapt/develop modules utilizing custom exception handling;
Former can be easily achieved with any NASM version by adding following line to source code:
$@feat.00 equ 1
As of version 2.03 NASM adds this absolute symbol automatically. If it's not already present to be precise. I.e. if for whatever reason developer would choose to assign another value in source file, it would still be perfectly possible.
Registering custom exception handler on the other hand requires certain
"magic." As of version 2.03 additional directive is implemented,
safeseh
, which instructs the assembler to produce
appropriately formatted input data for above mentioned "safe exception
handler table." Its typical use would be:
section .text extern _MessageBoxA@16 %if __NASM_VERSION_ID__ >= 0x02030000 safeseh handler ; register handler as "safe handler" %endif handler: push DWORD 1 ; MB_OKCANCEL push DWORD caption push DWORD text push DWORD 0 call _MessageBoxA@16 sub eax,1 ; incidentally suits as return value ; for exception handler ret global _main _main: push DWORD handler push DWORD [fs:0] mov DWORD [fs:0],esp ; engage exception handler xor eax,eax mov eax,DWORD[eax] ; cause exception pop DWORD [fs:0] ; disengage exception handler add esp,4 ret text: db 'OK to rethrow, CANCEL to generate core dump',0 caption:db 'SEGV',0 section .drectve info db '/defaultlib:user32.lib /defaultlib:msvcrt.lib '
As you might imagine, it's perfectly possible to produce .exe binary
with "safe exception handler table" and yet engage unregistered exception
handler. Indeed, handler is engaged by simply manipulating
[fs:0]
location at run-time, something linker has no power
over, run-time that is. It should be explicitly mentioned that such failure
to register handler's entry point with safeseh
directive has
undesired side effect at run-time. If exception is raised and unregistered
handler is to be executed, the application is abruptly terminated without
any notification whatsoever. One can argue that system could at least have
logged some kind "non-safe exception handler in x.exe at address n" message
in event log, but no, literally no notification is provided and user is
left with no clue on what caused application failure.
Finally, all mentions of linker in this paragraph refer to Microsoft
linker version 7.x and later. Presence of @feat.00
symbol and
input data for "safe exception handler table" causes no backward
incompatibilities and "safeseh" modules generated by NASM 2.03 and later
can still be linked by earlier versions or non-Microsoft linkers.
The win32
and win64
formats support the
Microsoft CodeView debugging format. Currently CodeView version 8 format is
supported (cv8
), but newer versions of the CodeView debugger
should be able to handle this format as well.
win64
: Microsoft Win64 Object FilesThe win64
output format generates Microsoft Win64 object
files, which is nearly 100% identical to the win32
object
format (section 7.5) with the exception that it
is meant to target 64-bit code and the x86-64 platform altogether. This
object file is used exactly the same as the win32
object
format (section 7.5), in NASM, with regard to
this exception.
win64
: Writing Position-Independent CodeWhile REL
takes good care of RIP-relative addressing, there
is one aspect that is easy to overlook for a Win64 programmer: indirect
references. Consider a switch dispatch table:
jmp qword [dsptch+rax*8] ... dsptch: dq case0 dq case1 ...
Even a novice Win64 assembler programmer will soon realize that the code is not 64-bit savvy. Most notably linker will refuse to link it with
'ADDR32' relocation to '.text' invalid without /LARGEADDRESSAWARE:NO
So [s]he will have to split jmp instruction as following:
lea rbx,[rel dsptch] jmp qword [rbx+rax*8]
What happens behind the scene is that effective address in
lea
is encoded relative to instruction pointer, or in
perfectly position-independent manner. But this is only part of the
problem! Trouble is that in .dll context caseN
relocations
will make their way to the final module and might have to be adjusted at
.dll load time. To be specific when it can't be loaded at preferred
address. And when this occurs, pages with such relocations will be rendered
private to current process, which kind of undermines the idea of sharing
.dll. But no worry, it's trivial to fix:
lea rbx,[rel dsptch] add rbx,[rbx+rax*8] jmp rbx ... dsptch: dq case0-dsptch dq case1-dsptch ...
NASM version 2.03 and later provides another alternative,
wrt ..imagebase
operator, which returns offset from base
address of the current image, be it .exe or .dll module, therefore the
name. For those acquainted with PE-COFF format base address denotes start
of IMAGE_DOS_HEADER
structure. Here is how to implement switch
with these image-relative references:
lea rbx,[rel dsptch] mov eax,[rbx+rax*4] sub rbx,dsptch wrt ..imagebase add rbx,rax jmp rbx ... dsptch: dd case0 wrt ..imagebase dd case1 wrt ..imagebase
One can argue that the operator is redundant. Indeed, snippet before
last works just fine with any NASM version and is not even Windows
specific... The real reason for implementing wrt ..imagebase
will become apparent in next paragraph.
It should be noted that wrt ..imagebase
is defined as
32-bit operand only:
dd label wrt ..imagebase ; ok dq label wrt ..imagebase ; bad mov eax,label wrt ..imagebase ; ok mov rax,label wrt ..imagebase ; bad
win64
: Structured Exception HandlingStructured exception handing in Win64 is completely different matter
from Win32. Upon exception program counter value is noted, and
linker-generated table comprising start and end addresses of all the
functions [in given executable module] is traversed and compared to the
saved program counter. Thus so called UNWIND_INFO
structure is
identified. If it's not found, then offending subroutine is assumed to be
"leaf" and just mentioned lookup procedure is attempted for its caller. In
Win64 leaf function is such function that does not call any other function
nor modifies any Win64 non-volatile registers, including stack
pointer. The latter ensures that it's possible to identify leaf function's
caller by simply pulling the value from the top of the stack.
While majority of subroutines written in assembler are not calling any
other function, requirement for non-volatile registers' immutability leaves
developer with not more than 7 registers and no stack frame, which is not
necessarily what [s]he counted with. Customarily one would meet the
requirement by saving non-volatile registers on stack and restoring them
upon return, so what can go wrong? If [and only if] an exception is raised
at run-time and no UNWIND_INFO
structure is associated with
such "leaf" function, the stack unwind procedure will expect to find
caller's return address on the top of stack immediately followed by its
frame. Given that developer pushed caller's non-volatile registers on
stack, would the value on top point at some code segment or even
addressable space? Well, developer can attempt copying caller's return
address to the top of stack and this would actually work in some very
specific circumstances. But unless developer can guarantee that these
circumstances are always met, it's more appropriate to assume worst case
scenario, i.e. stack unwind procedure going berserk. Relevant question is
what happens then? Application is abruptly terminated without any
notification whatsoever. Just like in Win32 case, one can argue that system
could at least have logged "unwind procedure went berserk in x.exe at
address n" in event log, but no, no trace of failure is left.
Now, when we understand significance of the UNWIND_INFO
structure, let's discuss what's in it and/or how it's processed. First of
all it is checked for presence of reference to custom language-specific
exception handler. If there is one, then it's invoked. Depending on the
return value, execution flow is resumed (exception is said to be
"handled"), or rest of UNWIND_INFO
structure is
processed as following. Beside optional reference to custom handler, it
carries information about current callee's stack frame and where
non-volatile registers are saved. Information is detailed enough to be able
to reconstruct contents of caller's non-volatile registers upon call to
current callee. And so caller's context is reconstructed, and then unwind
procedure is repeated, i.e. another UNWIND_INFO
structure is
associated, this time, with caller's instruction pointer, which is then
checked for presence of reference to language-specific handler, etc. The
procedure is recursively repeated till exception is handled. As last resort
system "handles" it by generating memory core dump and terminating the
application.
As for the moment of this writing NASM unfortunately does not facilitate generation of above mentioned detailed information about stack frame layout. But as of version 2.03 it implements building blocks for generating structures involved in stack unwinding. As simplest example, here is how to deploy custom exception handler for leaf function:
default rel section .text extern MessageBoxA handler: sub rsp,40 mov rcx,0 lea rdx,[text] lea r8,[caption] mov r9,1 ; MB_OKCANCEL call MessageBoxA sub eax,1 ; incidentally suits as return value ; for exception handler add rsp,40 ret global main main: xor rax,rax mov rax,QWORD[rax] ; cause exception ret main_end: text: db 'OK to rethrow, CANCEL to generate core dump',0 caption:db 'SEGV',0 section .pdata rdata align=4 dd main wrt ..imagebase dd main_end wrt ..imagebase dd xmain wrt ..imagebase section .xdata rdata align=8 xmain: db 9,0,0,0 dd handler wrt ..imagebase section .drectve info db '/defaultlib:user32.lib /defaultlib:msvcrt.lib '
What you see in .pdata
section is element of the "table
comprising start and end addresses of function" along with reference to
associated UNWIND_INFO
structure. And what you see in
.xdata
section is UNWIND_INFO
structure
describing function with no frame, but with designated exception handler.
References are required to be image-relative (which is the real
reason for implementing wrt ..imagebase
operator). It should
be noted that rdata align=n
, as well as
wrt ..imagebase
, are optional in these two segments' contexts,
i.e. can be omitted. Latter means that all 32-bit references, not
only above listed required ones, placed into these two segments turn out
image-relative. Why is it important to understand? Developer is allowed to
append handler-specific data to UNWIND_INFO
structure, and if
[s]he adds a 32-bit reference, then [s]he will have to remember to adjust
its value to obtain the real pointer.
As already mentioned, in Win64 terms leaf function is one that does not
call any other function nor modifies any non-volatile register,
including stack pointer. But it's not uncommon that assembler programmer
plans to utilize every single register and sometimes even have variable
stack frame. Is there anything one can do with bare building blocks? I.e.
besides manually composing fully-fledged UNWIND_INFO
structure, which would surely be considered error-prone? Yes, there is.
Recall that exception handler is called first, before stack layout is
analyzed. As it turned out, it's perfectly possible to manipulate current
callee's context in custom handler in manner that permits further stack
unwinding. General idea is that handler would not actually "handle" the
exception, but instead restore callee's context, as it was at its entry
point and thus mimic leaf function. In other words, handler would simply
undertake part of unwinding procedure. Consider following example:
function: mov rax,rsp ; copy rsp to volatile register push r15 ; save non-volatile registers push rbx push rbp mov r11,rsp ; prepare variable stack frame sub r11,rcx and r11,-64 mov QWORD[r11],rax ; check for exceptions mov rsp,r11 ; allocate stack frame mov QWORD[rsp],rax ; save original rsp value magic_point: ... mov r11,QWORD[rsp] ; pull original rsp value mov rbp,QWORD[r11-24] mov rbx,QWORD[r11-16] mov r15,QWORD[r11-8] mov rsp,r11 ; destroy frame ret
The keyword is that up to magic_point
original
rsp
value remains in chosen volatile register and no
non-volatile register, except for rsp
, is modified. While past
magic_point
rsp
remains constant till the very
end of the function
. In this case custom language-specific
exception handler would look like this:
EXCEPTION_DISPOSITION handler (EXCEPTION_RECORD *rec,ULONG64 frame, CONTEXT *context,DISPATCHER_CONTEXT *disp) { ULONG64 *rsp; if (context->Rip<(ULONG64)magic_point) rsp = (ULONG64 *)context->Rax; else { rsp = ((ULONG64 **)context->Rsp)[0]; context->Rbp = rsp[-3]; context->Rbx = rsp[-2]; context->R15 = rsp[-1]; } context->Rsp = (ULONG64)rsp; memcpy (disp->ContextRecord,context,sizeof(CONTEXT)); RtlVirtualUnwind(UNW_FLAG_NHANDLER,disp->ImageBase, dips->ControlPc,disp->FunctionEntry,disp->ContextRecord, &disp->HandlerData,&disp->EstablisherFrame,NULL); return ExceptionContinueSearch; }
As custom handler mimics leaf function, corresponding
UNWIND_INFO
structure does not have to contain any information
about stack frame and its layout.
coff
: Common Object File FormatThe coff
output type produces COFF
object
files suitable for linking with the DJGPP linker.
coff
provides a default output file-name extension of
.o
.
The coff
format supports the same extensions to the
SECTION
directive as win32
does, except that the
align
qualifier and the info
section type are not
supported.
macho32
and macho64
: Mach Object File FormatThe macho32
and macho64
output formts produces
Mach-O object files suitable for linking with the MacOS X linker.
macho
is a synonym for macho32
.
macho
provides a default output file-name extension of
.o
.
macho
extensions to the SECTION
Directive The macho
output format specifies section names in the
format "segment,
section". No spaces are
allowed around the comma. The following flags can also be specified:
data
– this section contains initialized data items
text
– this section contains code exclusively
mixed
– this section contains both code and data
bss
– this section is uninitialized and filled with
zero
zerofill
– same as bss
no_dead_strip
– inhibit dead code stripping for this
section
live_support
– set the live support flag for this
section
strip_static_syms
– strip static symbols for this
section
debug
– this section contains debugging information
align=
alignment – specify section alignment
The default is data
, unless the section name is
__text
or __bss
in which case the default is
text
or bss
, respectively.
For compatibility with other Unix platforms, the following standard names are also supported:
.text = __TEXT,__text text .rodata = __DATA,__const data .data = __DATA,__data data .bss = __DATA,__bss bss
If the .rodata
section contains no relocations, it is
instead put into the __TEXT,__const
section unless this
section has already been specified explicitly. However, it is probably
better to specify __TEXT,__const
and
__DATA,__const
explicitly as appropriate.
macho
special symbols and WRT
Mach-O defines the following special symbols that can be used on the
right-hand side of the WRT
operator:
..tlvp
is used to specify access to thread-local storage.
..gotpcrel
is used to specify references to the Global
Offset Table. The GOT is supported in the macho64
format only.
macho
specfic directive subsections_via_symbols
The directive subsections_via_symbols
sets the
MH_SUBSECTIONS_VIA_SYMBOLS
flag in the Mach-O header, which
tells the linker that the symbols in the file matches the conventions
required to allow for link-time dead code elimination.
This directive takes no arguments.
This is a macro implemented as a %pragma
. It can also be
specified in its %pragma
form, in which case it will not
affect non-Mach-O builds of the same source code:
%pragma macho subsections_via_symbols
macho
specfic directive no_dead_strip
The directive no_dead_strip
sets the Mach-O
SH_NO_DEAD_STRIP
section flag on the section containing a a
specific symbol. This directive takes a list of symbols as its arguments.
This is a macro implemented as a %pragma
. It can also be
specified in its %pragma
form, in which case it will not
affect non-Mach-O builds of the same source code:
%pragma macho no_dead_strip symbol...
elf32
, elf64
, elfx32
: Executable and Linkable Format Object FilesThe elf32
, elf64
and elfx32
output formats generate ELF32 and ELF64
(Executable and
Linkable Format) object files, as used by Linux as well as Unix System V,
including Solaris x86, UnixWare and SCO Unix. elf
provides a
default output file-name extension of .o
. elf
is
a synonym for elf32
.
The elfx32
format is used for the x32 ABI, which is a
32-bit ABI with the CPU in 64-bit mode.
osabi
The ELF header specifies the application binary interface for the target
operating system (OSABI). This field can be set by using the
osabi
directive with the numeric value (0-255) of the target
system. If this directive is not used, the default value will be "UNIX
System V ABI" (0) which will work on most systems which support ELF.
elf
extensions to the SECTION
Directive Like the obj
format, elf
allows you to specify
additional information on the SECTION
directive line, to
control the type and properties of sections you declare. Section types and
properties are generated automatically by NASM for the standard section
names, but may still be overridden by these qualifiers.
The available qualifiers are:
alloc
defines the section to be one which is loaded into
memory when the program is run. noalloc
defines it to be one
which is not, such as an informational or comment section.
exec
defines the section to be one which should have
execute permission when the program is run. noexec
defines it
as one which should not.
write
defines the section to be one which should be
writable when the program is run. nowrite
defines it as one
which should not.
progbits
defines the section to be one with explicit
contents stored in the object file: an ordinary code or data section, for
example, nobits
defines the section to be one with no explicit
contents given, such as a BSS section.
align=
, used with a trailing number as in obj
,
gives the alignment requirements of the section.
tls
defines the section to be one which contains thread
local variables.
The defaults assumed by NASM if you do not specify the above qualifiers are:
section .text progbits alloc exec nowrite align=16 section .rodata progbits alloc noexec nowrite align=4 section .lrodata progbits alloc noexec nowrite align=4 section .data progbits alloc noexec write align=4 section .ldata progbits alloc noexec write align=4 section .bss nobits alloc noexec write align=4 section .lbss nobits alloc noexec write align=4 section .tdata progbits alloc noexec write align=4 tls section .tbss nobits alloc noexec write align=4 tls section .comment progbits noalloc noexec nowrite align=1 section other progbits alloc noexec nowrite align=1
(Any section name other than those in the above table is treated by
default like other
in the above table. Please note that
section names are case sensitive.)
macho
Special Symbols and WRT
Since ELF
does not support segment-base references, the
WRT
operator is not used for its normal purpose; therefore
NASM's elf
output format makes use of WRT
for a
different purpose, namely the PIC-specific relocation types.
elf
defines five special symbols which you can use as the
right-hand side of the WRT
operator to obtain PIC relocation
types. They are ..gotpc
, ..gotoff
,
..got
, ..plt
and ..sym
. Their
functions are summarized here:
Referring to the symbol marking the global offset table base using
wrt ..gotpc
will end up giving the distance from the beginning
of the current section to the global offset table.
(_GLOBAL_OFFSET_TABLE_
is the standard symbol name used to
refer to the GOT.) So you would then need to add $$
to the
result to get the real address of the GOT.
Referring to a location in one of your own sections using
wrt ..gotoff
will give the distance from the beginning of the
GOT to the specified location, so that adding on the address of the GOT
would give the real address of the location you wanted.
Referring to an external or global symbol using wrt ..got
causes the linker to build an entry in the GOT containing the
address of the symbol, and the reference gives the distance from the
beginning of the GOT to the entry; so you can add on the address of the
GOT, load from the resulting address, and end up with the address of the
symbol.
Referring to a procedure name using wrt ..plt
causes the
linker to build a procedure linkage table entry for the symbol, and the
reference gives the address of the PLT entry. You can only use this in
contexts which would generate a PC-relative relocation normally (i.e. as
the destination for CALL
or JMP
), since ELF
contains no relocation type to refer to PLT entries absolutely.
Referring to a symbol name using wrt ..sym
causes NASM to
write an ordinary relocation, but instead of making the relocation relative
to the start of the section and then adding on the offset to the symbol, it
will write a relocation record aimed directly at the symbol in question.
The distinction is a necessary one due to a peculiarity of the dynamic
linker.
A fuller explanation of how to use these relocation types to write shared libraries entirely in NASM is given in section 9.2.
elf
Special Symbols and WRT
In ELF32 mode, referring to an external or global symbol using
wrt ..tlsie
causes the linker to build an entry in
the GOT containing the offset of the symbol within the TLS block, so you
can access the value of the symbol with code such as:
mov eax,[tid wrt ..tlsie] mov [gs:eax],ebx
In ELF64 or ELFx32 mode, referring to an external or global symbol using
wrt ..gottpoff
causes the linker to build an entry
in the GOT containing the offset of the symbol within the TLS
block, so you can access the value of the symbol with code such as:
mov rax,[rel tid wrt ..gottpoff] mov rcx,[fs:rax]
elf
Extensions to the GLOBAL
DirectiveELF
object files can contain more information about a
global symbol than just its address: they can contain the size of the
symbol and its type as well. These are not merely debugger conveniences,
but are actually necessary when the program being written is a shared
library. NASM therefore supports some extensions to the GLOBAL
directive, allowing you to specify these features.
You can specify whether a global variable is a function or a data object
by suffixing the name with a colon and the word function
or
data
. (object
is a synonym for
data
.) For example:
global hashlookup:function, hashtable:data
exports the global symbol hashlookup
as a function and
hashtable
as a data object.
Optionally, you can control the ELF visibility of the symbol. Just add
one of the visibility keywords: default
,
internal
, hidden
, or protected
. The
default is default
of course. For example, to make
hashlookup
hidden:
global hashlookup:function hidden
You can also specify the size of the data associated with the symbol, as a numeric expression (which may involve labels, and even forward references) after the type specifier. Like this:
global hashtable:data (hashtable.end - hashtable) hashtable: db this,that,theother ; some data here .end:
This makes NASM automatically calculate the length of the table and
place that information into the ELF
symbol table.
Declaring the type and size of global symbols is necessary when writing shared library code. For more information, see section 9.2.4.
elf
Extensions to the COMMON
Directive ELF
also allows you to specify alignment requirements on
common variables. This is done by putting a number (which must be a power
of two) after the name and size of the common variable, separated (as
usual) by a colon. For example, an array of doublewords would benefit from
4-byte alignment:
common dwordarray 128:4
This declares the total size of the array to be 128 bytes, and requires that it be aligned on a 4-byte boundary.
The ELF32
specification doesn't provide relocations for 8-
and 16-bit values, but the GNU ld
linker adds these as an
extension. NASM can generate GNU-compatible relocations, to allow 16-bit
code to be linked as ELF using GNU ld
. If NASM is used with
the -w+gnu-elf-extensions
option, a warning is issued when one
of these relocations is generated.
ELF provides debug information in STABS
and
DWARF
formats. Line number information is generated for all
executable sections, but please note that only the ".text" section is
executable by default.
aout
: Linux a.out
Object FilesThe aout
format generates a.out
object files,
in the form used by early Linux systems (current Linux systems use ELF, see
section 7.9.) These differ from other
a.out
object files in that the magic number in the first four
bytes of the file is different; also, some implementations of
a.out
, for example NetBSD's, support position-independent
code, which Linux's implementation does not.
a.out
provides a default output file-name extension of
.o
.
a.out
is a very simple object format. It supports no
special directives, no special symbols, no use of SEG
or
WRT
, and no extensions to any standard directives. It supports
only the three standard section names .text
,
.data
and .bss
.
aoutb
: NetBSD/FreeBSD/OpenBSD a.out
Object FilesThe aoutb
format generates a.out
object files,
in the form used by the various free BSD Unix
clones,
NetBSD
, FreeBSD
and OpenBSD
. For
simple object files, this object format is exactly the same as
aout
except for the magic number in the first four bytes of
the file. However, the aoutb
format supports
position-independent code in the same way as the elf
format,
so you can use it to write BSD
shared libraries.
aoutb
provides a default output file-name extension of
.o
.
aoutb
supports no special directives, no special symbols,
and only the three standard section names .text
,
.data
and .bss
. However, it also supports the
same use of WRT
as elf
does, to provide
position-independent code relocation types. See
section 7.9.3 for full documentation of this
feature.
aoutb
also supports the same extensions to the
GLOBAL
directive as elf
does: see
section 7.9.5 for documentation of this.
as86
: Minix/Linux as86
Object FilesThe Minix/Linux 16-bit assembler as86
has its own
non-standard object file format. Although its companion linker
ld86
produces something close to ordinary a.out
binaries as output, the object file format used to communicate between
as86
and ld86
is not itself a.out
.
NASM supports this format, just in case it is useful, as
as86
. as86
provides a default output file-name
extension of .o
.
as86
is a very simple object format (from the NASM user's
point of view). It supports no special directives, no use of
SEG
or WRT
, and no extensions to any standard
directives. It supports only the three standard section names
.text
, .data
and .bss
. The only
special symbol supported is ..start
.
rdf
: Relocatable Dynamic Object File FormatThe rdf
output format produces RDOFF
object
files. RDOFF
(Relocatable Dynamic Object File Format) is a
home-grown object-file format, designed alongside NASM itself and
reflecting in its file format the internal structure of the assembler.
RDOFF
is not used by any well-known operating systems.
Those writing their own systems, however, may well wish to use
RDOFF
as their object format, on the grounds that it is
designed primarily for simplicity and contains very little file-header
bureaucracy.
The Unix NASM archive, and the DOS archive which includes sources, both
contain an rdoff
subdirectory holding a set of RDOFF
utilities: an RDF linker, an RDF
static-library manager, an
RDF file dump utility, and a program which will load and execute an RDF
executable under Linux.
rdf
supports only the standard section names
.text
, .data
and .bss
.
LIBRARY
DirectiveRDOFF
contains a mechanism for an object file to demand a
given library to be linked to the module, either at load time or run time.
This is done by the LIBRARY
directive, which takes one
argument which is the name of the module:
library mylib.rdl
MODULE
DirectiveSpecial RDOFF
header record is used to store the name of
the module. It can be used, for example, by run-time loader to perform
dynamic linking. MODULE
directive takes one argument which is
the name of current module:
module mymodname
Note that when you statically link modules and tell linker to strip the
symbols from output file, all module names will be stripped too. To avoid
it, you should start module names with $
, like:
module $kernel.core
rdf
Extensions to the GLOBAL
DirectiveRDOFF
global symbols can contain additional information
needed by the static linker. You can mark a global symbol as exported, thus
telling the linker do not strip it from target executable or library file.
Like in ELF
, you can also specify whether an exported symbol
is a procedure (function) or data object.
Suffixing the name with a colon and the word export
you
make the symbol exported:
global sys_open:export
To specify that exported symbol is a procedure (function), you add the
word proc
or function
after declaration:
global sys_open:export proc
Similarly, to specify exported data object, add the word
data
or object
to the directive:
global kernel_ticks:export data
rdf
Extensions to the EXTERN
DirectiveBy default the EXTERN
directive in RDOFF
declares a "pure external" symbol (i.e. the static linker will complain if
such a symbol is not resolved). To declare an "imported" symbol, which must
be resolved later during a dynamic linking phase, RDOFF
offers
an additional import
modifier. As in GLOBAL
, you
can also specify whether an imported symbol is a procedure (function) or
data object. For example:
library $libc extern _open:import extern _printf:import proc extern _errno:import data
Here the directive LIBRARY
is also included, which gives
the dynamic linker a hint as to where to find requested symbols.
dbg
: Debugging FormatThe dbg
format does not output an object file as such;
instead, it outputs a text file which contains a complete list of all the
transactions between the main body of NASM and the output-format back end
module. It is primarily intended to aid people who want to write their own
output drivers, so that they can get a clearer idea of the various requests
the main program makes of the output driver, and in what order they happen.
For simple files, one can easily use the dbg
format like
this:
nasm -f dbg filename.asm
which will generate a diagnostic file called filename.dbg
.
However, this will not work well on files which were designed for a
different object format, because each object format defines its own macros
(usually user-level forms of directives), and those macros will not be
defined in the dbg
format. Therefore it can be useful to run
NASM twice, in order to do the preprocessing with the native object format
selected:
nasm -e -f rdf -o rdfprog.i rdfprog.asm nasm -a -f dbg rdfprog.i
This preprocesses rdfprog.asm
into rdfprog.i
,
keeping the rdf
object format selected in order to make sure
RDF special directives are converted into primitive form correctly. Then
the preprocessed source is fed through the dbg
format to
generate the final diagnostic output.
This workaround will still typically not work for programs intended for
obj
format, because the obj
SEGMENT
and GROUP
directives have side effects of defining the segment
and group names as symbols; dbg
will not do this, so the
program will not assemble. You will have to work around that by defining
the symbols yourself (using EXTERN
, for example) if you really
need to get a dbg
trace of an obj
–specific
source file.
dbg
accepts any section name and any directives at all, and
logs them all to its output file.
dbg
accepts and logs any %pragma
, but the
specific %pragma
:
%pragma dbg maxdump <size>
where <size>
is either a number or
unlimited
, can be used to control the maximum size for dumping
the full contents of a rawdata
output object.