<!--{
	"Title": "A Quick Guide to Go's Assembler",
	"Path":  "/doc/asm"
}-->

<h2 id="introduction">A Quick Guide to Go's Assembler</h2>

<p>
This document is a quick outline of the unusual form of assembly language used by the <code>gc</code>
suite of Go compilers (<code>6g</code>, <code>8g</code>, etc.).
The document is not comprehensive.
</p>

<p>
The assembler is based on the input to the Plan 9 assemblers, which is documented in detail
<a href="http://plan9.bell-labs.com/sys/doc/asm.html">on the Plan 9 site</a>.
If you plan to write assembly language, you should read that document although much of it is Plan 9-specific.
This document provides a summary of the syntax and
describes the peculiarities that apply when writing assembly code to interact with Go.
</p>

<p>
The most important thing to know about Go's assembler is that it is not a direct representation of the underlying machine.
Some of the details map precisely to the machine, but some do not.
This is because the compiler suite (see
<a href="http://plan9.bell-labs.com/sys/doc/compiler.html">this description</a>)
needs no assembler pass in the usual pipeline.
Instead, the compiler emits a kind of incompletely defined instruction set, in binary form, which the linker
then completes.
In particular, the linker does instruction selection, so when you see an instruction like <code>MOV</code>
what the linker actually generates for that operation might not be a move instruction at all, perhaps a clear or load.
Or it might correspond exactly to the machine instruction with that name.
In general, machine-specific operations tend to appear as themselves, while more general concepts like
memory move and subroutine call and return are more abstract.
The details vary with architecture, and we apologize for the imprecision; the situation is not well-defined.
</p>

<p>
The assembler program is a way to generate that intermediate, incompletely defined instruction sequence
as input for the linker.
If you want to see what the instructions look like in assembly for a given architecture, say amd64, there
are many examples in the sources of the standard library, in packages such as
<a href="/pkg/runtime/"><code>runtime</code></a> and
<a href="/pkg/math/big/"><code>math/big</code></a>.
You can also examine what the compiler emits as assembly code:
</p>

<pre>
$ cat x.go
package main

func main() {
	println(3)
}
$ go tool 6g -S x.go        # or: go build -gcflags -S x.go

--- prog list "main" ---
0000 (x.go:3) TEXT    main+0(SB),$8-0
0001 (x.go:3) FUNCDATA $0,gcargs·0+0(SB)
0002 (x.go:3) FUNCDATA $1,gclocals·0+0(SB)
0003 (x.go:4) MOVQ    $3,(SP)
0004 (x.go:4) PCDATA  $0,$8
0005 (x.go:4) CALL    ,runtime.printint+0(SB)
0006 (x.go:4) PCDATA  $0,$-1
0007 (x.go:4) PCDATA  $0,$0
0008 (x.go:4) CALL    ,runtime.printnl+0(SB)
0009 (x.go:4) PCDATA  $0,$-1
0010 (x.go:5) RET     ,
...
</pre>

<p>
The <code>FUNCDATA</code> and <code>PCDATA</code> directives contain information
for use by the garbage collector; they are introduced by the compiler.
</p> 

<!-- Commenting out because the feature is gone but it's popular and may come back.

<p>
To see what gets put in the binary after linking, add the <code>-a</code> flag to the linker:
</p>

<pre>
$ go tool 6l -a x.6        # or: go build -ldflags -a x.go
codeblk [0x2000,0x1d059) at offset 0x1000
002000	main.main            | (3)	TEXT	main.main+0(SB),$8
002000	65488b0c25a0080000   | (3)	MOVQ	2208(GS),CX
002009	483b21               | (3)	CMPQ	SP,(CX)
00200c	7707                 | (3)	JHI	,2015
00200e	e83da20100           | (3)	CALL	,1c250+runtime.morestack00
002013	ebeb                 | (3)	JMP	,2000
002015	4883ec08             | (3)	SUBQ	$8,SP
002019	                     | (3)	FUNCDATA	$0,main.gcargs·0+0(SB)
002019	                     | (3)	FUNCDATA	$1,main.gclocals·0+0(SB)
002019	48c7042403000000     | (4)	MOVQ	$3,(SP)
002021	                     | (4)	PCDATA	$0,$8
002021	e8aad20000           | (4)	CALL	,f2d0+runtime.printint
002026	                     | (4)	PCDATA	$0,$-1
002026	                     | (4)	PCDATA	$0,$0
002026	e865d40000           | (4)	CALL	,f490+runtime.printnl
00202b	                     | (4)	PCDATA	$0,$-1
00202b	4883c408             | (5)	ADDQ	$8,SP
00202f	c3                   | (5)	RET	,
...
</pre>

-->

<h3 id="symbols">Symbols</h3>

<p>
Some symbols, such as <code>PC</code>, <code>R0</code> and <code>SP</code>, are predeclared and refer to registers.
There are two other predeclared symbols, <code>SB</code> (static base) and <code>FP</code> (frame pointer).
All user-defined symbols other than jump labels are written as offsets to these pseudo-registers.
</p>

<p>
The <code>SB</code> pseudo-register can be thought of as the origin of memory, so the symbol <code>foo(SB)</code>
is the name <code>foo</code> as an address in memory.
</p>

<p>
The <code>FP</code> pseudo-register is a virtual frame pointer
used to refer to function arguments.
The compilers maintain a virtual frame pointer and refer to the arguments on the stack as offsets from that pseudo-register.
Thus <code>0(FP)</code> is the first argument to the function,
<code>8(FP)</code> is the second (on a 64-bit machine), and so on.
When referring to a function argument this way, it is conventional to place the name
at the beginning, as in <code>first_arg+0(FP)</code> and <code>second_arg+8(FP)</code>.
Some of the assemblers enforce this convention, rejecting plain <code>0(FP)</code> and <code>8(FP)</code>.
For assembly functions with Go prototypes, <code>go vet</code> will check that the argument names
and offsets match.
</p>

<p>
The <code>SP</code> pseudo-register is a virtual stack pointer
used to refer to frame-local variables and the arguments being
prepared for function calls.
It points to the top of the local stack frame, so references should use negative offsets
in the range [−framesize, 0):
<code>x-8(SP)</code>, <code>y-4(SP)</code>, and so on.
On architectures with a real register named <code>SP</code>, the name prefix distinguishes
references to the virtual stack pointer from references to the architectural <code>SP</code> register.
That is, <code>x-8(SP)</code> and <code>-8(SP)</code> are different memory locations:
the first refers to the virtual stack pointer pseudo-register, while the second refers to the
hardware's <code>SP</code> register.
</p>

<p>
Instructions, registers, and assembler directives are always in UPPER CASE to remind you
that assembly programming is a fraught endeavor.
(Exceptions: the <code>m</code> and <code>g</code> register renamings on ARM.)
</p>

<p>
In Go object files and binaries, the full name of a symbol is the 
package path followed by a period and the symbol name:
<code>fmt.Printf</code> or <code>math/rand.Int</code>.
Because the assembler's parser treats period and slash as punctuation,
those strings cannot be used directly as identifier names.
Instead, the assembler allows the middle dot character U+00B7
and the division slash U+2215 in identifiers and rewrites them to
plain period and slash.
Within an assembler source file, the symbols above are written as
<code>fmt·Printf</code> and <code>math∕rand·Int</code>.
The assembly listings generated by the compilers when using the <code>-S</code> flag
show the period and slash directly instead of the Unicode replacements
required by the assemblers.
</p>

<p>
Most hand-written assembly files do not include the full package path
in symbol names, because the linker inserts the package path of the current
object file at the beginning of any name starting with a period:
in an assembly source file within the math/rand package implementation,
the package's Int function can be referred to as <code>·Int</code>.
This convention avoids the need to hard-code a package's import path in its
own source code, making it easier to move the code from one location to another.
</p>

<h3 id="directives">Directives</h3>

<p>
The assembler uses various directives to bind text and data to symbol names.
For example, here is a simple complete function definition. The <code>TEXT</code>
directive declares the symbol <code>runtime·profileloop</code> and the instructions
that follow form the body of the function.
The last instruction in a <code>TEXT</code> block must be some sort of jump, usually a <code>RET</code> (pseudo-)instruction.
(If it's not, the linker will append a jump-to-itself instruction; there is no fallthrough in <code>TEXTs</code>.)
After the symbol, the arguments are flags (see below)
and the frame size, a constant (but see below):
</p>

<pre>
TEXT runtime·profileloop(SB),NOSPLIT,$8
	MOVQ	$runtime·profileloop1(SB), CX
	MOVQ	CX, 0(SP)
	CALL	runtime·externalthreadhandler(SB)
	RET
</pre>

<p>
In the general case, the frame size is followed by an argument size, separated by a minus sign.
(It's not an subtraction, just idiosyncratic syntax.)
The frame size <code>$24-8</code> states that the function has a 24-byte frame
and is called with 8 bytes of argument, which live on the caller's frame.
If <code>NOSPLIT</code> is not specified for the <code>TEXT</code>,
the argument size must be provided.
</p>

<p>
Note that the symbol name uses a middle dot to separate the components and is specified as an offset from the
static base pseudo-register <code>SB</code>.
This function would be called from Go source for package <code>runtime</code> using the
simple name <code>profileloop</code>.
</p>

<p>
For <code>DATA</code> directives, the symbol is followed by a slash and the number
of bytes the memory associated with the symbol occupies.
The arguments are optional flags and the data itself.
For instance,
</p>

<pre>
DATA  runtime·isplan9(SB)/4, $1
</pre>

<p>
declares the local symbol <code>runtime·isplan9</code> of size 4 and value 1.
Again the symbol has the middle dot and is offset from <code>SB</code>.
</p>

<p>
The <code>GLOBL</code> directive declares a symbol to be global.
The arguments are optional flags and the size of the data being declared as a global,
which will have initial value all zeros unless a <code>DATA</code> directive
has initialized it.
The <code>GLOBL</code> directive must follow any corresponding <code>DATA</code> directives.
This example
</p>

<pre>
GLOBL runtime·tlsoffset(SB),$4
</pre>

<p>
declares <code>runtime·tlsoffset</code> to have size 4.
</p>

<p>
There may be one or two arguments to the directives.
If there are two, the first is a bit mask of flags,
which can be written as numeric expressions, added or or-ed together,
or can be set symbolically for easier absorption by a human.
Their values, defined in the file <code>src/cmd/ld/textflag.h</code>, are:
</p>

<ul>
<li>
<code>NOPROF</code> = 1
<br>
(For <code>TEXT</code> items.)
Don't profile the marked function.  This flag is deprecated.
</li>
<li>
<code>DUPOK</code> = 2
<br>
It is legal to have multiple instances of this symbol in a single binary.
The linker will choose one of the duplicates to use.
</li>
<li>
<code>NOSPLIT</code> = 4
<br>
(For <code>TEXT</code> items.)
Don't insert the preamble to check if the stack must be split.
The frame for the routine, plus anything it calls, must fit in the
spare space at the top of the stack segment.
Used to protect routines such as the stack splitting code itself.
</li>
<li>
<code>RODATA</code> = 8
<br>
(For <code>DATA</code> and <code>GLOBL</code> items.)
Put this data in a read-only section.
</li>
<li>
<code>NOPTR</code> = 16
<br>
(For <code>DATA</code> and <code>GLOBL</code> items.)
This data contains no pointers and therefore does not need to be
scanned by the garbage collector.
</li>
<li>
<code>WRAPPER</code>  = 32
<br>
(For <code>TEXT</code> items.)
This is a wrapper function and should not count as disabling <code>recover</code>.
</li>
</ul>

<h2 id="architectures">Architecture-specific details</h2>

<p>
It is impractical to list all the instructions and other details for each machine.
To see what instructions are defined for a given machine, say 32-bit Intel x86,
look in the top-level header file for the corresponding linker, in this case <code>8l</code>.
That is, the file <code>$GOROOT/src/cmd/8l/8.out.h</code> contains a C enumeration, called <code>as</code>,
of the instructions and their spellings as known to the assembler and linker for that architecture.
In that file you'll find a declaration that begins
</p>

<pre>
enum	as
{
	AXXX,
	AAAA,
	AAAD,
	AAAM,
	AAAS,
	AADCB,
	...
</pre>

<p>
Each instruction begins with a  initial capital <code>A</code> in this list, so <code>AADCB</code>
represents the <code>ADCB</code> (add carry byte) instruction.
The enumeration is in alphabetical order, plus some late additions (<code>AXXX</code> occupies
the zero slot as an invalid instruction).
The sequence has nothing to do with the actual encoding of the machine instructions.
Again, the linker takes care of that detail.
</p>

<p>
One detail evident in the examples from the previous sections is that data in the instructions flows from left to right:
<code>MOVQ</code> <code>$0,</code> <code>CX</code> clears <code>CX</code>.
This convention applies even on architectures where the usual mode is the opposite direction.
</p>

<p>
Here follows some descriptions of key Go-specific details for the supported architectures.
</p>

<h3 id="x86">32-bit Intel 386</h3>

<p>
The runtime pointers to the <code>m</code> and <code>g</code> structures are maintained
through the value of an otherwise unused (as far as Go is concerned) register in the MMU.
A OS-dependent macro <code>get_tls</code> is defined for the assembler if the source includes
an architecture-dependent header file, like this:
</p>

<pre>
#include "zasm_GOOS_GOARCH.h"
</pre>

<p>
Within the runtime, the <code>get_tls</code> macro loads its argument register
with a pointer to a pair of words representing the <code>g</code> and <code>m</code> pointers.
The sequence to load <code>g</code> and <code>m</code> using <code>CX</code> looks like this:
</p>

<pre>
get_tls(CX)
MOVL	g(CX), AX	// Move g into AX.
MOVL	m(CX), BX	// Move m into BX.
</pre>

<h3 id="amd64">64-bit Intel 386 (a.k.a. amd64)</h3>

<p>
The assembly code to access the <code>m</code> and <code>g</code>
pointers is the same as on the 386, except it uses <code>MOVQ</code> rather than
<code>MOVL</code>:
</p>

<pre>
get_tls(CX)
MOVQ	g(CX), AX	// Move g into AX.
MOVQ	m(CX), BX	// Move m into BX.
</pre>

<h3 id="arm">ARM</h3>

<p>
The registers <code>R9</code>, <code>R10</code>, and <code>R11</code>
are reserved by the compiler and linker.
</p>

<p>
<code>R9</code> and <code>R10</code> point to the <code>m</code> (machine) and <code>g</code>
(goroutine) structures, respectively.
Within assembler source code, these pointers must be referred to as <code>m</code> and <code>g</code>;
the names <code>R9</code> and <code>R10</code> are not recognized.
</p>

<p>
To make it easier for people and compilers to write assembly, the ARM linker
allows general addressing forms and pseudo-operations like <code>DIV</code> or <code>MOD</code>
that may not be expressible using a single hardware instruction.
It implements these forms as multiple instructions, often using the <code>R11</code> register
to hold temporary values.
Hand-written assembly can use <code>R11</code>, but doing so requires
being sure that the linker is not also using it to implement any of the other
instructions in the function.
</p>

<p>
When defining a <code>TEXT</code>, specifying frame size <code>$-4</code>
tells the linker that this is a leaf function that does not need to save <code>LR</code> on entry.
</p>

<p>
The name <code>SP</code> always refers to the virtual stack pointer described earlier.
For the hardware register, use <code>R13</code>.
</p>

<h3 id="unsupported_opcodes">Unsupported opcodes</h3>

<p>
The assemblers are designed to support the compiler so not all hardware instructions
are defined for all architectures: if the compiler doesn't generate it, it might not be there.
If you need to use a missing instruction, there are two ways to proceed.
One is to update the assembler to support that instruction, which is straightforward
but only worthwhile if it's likely the instruction will be used again.
Instead, for simple one-off cases, it's possible to use the <code>BYTE</code>
and <code>WORD</code> directives
to lay down explicit data into the instruction stream within a <code>TEXT</code>.
Here's how the 386 runtime defines the 64-bit atomic load function.
</p>

<pre>
// uint64 atomicload64(uint64 volatile* addr);
// so actually
// void atomicload64(uint64 *res, uint64 volatile *addr);
TEXT runtime·atomicload64(SB), NOSPLIT, $0-8
	MOVL	4(SP), BX
	MOVL	8(SP), AX
	// MOVQ (%EAX), %MM0
	BYTE $0x0f; BYTE $0x6f; BYTE $0x00
	// MOVQ %MM0, 0(%EBX)
	BYTE $0x0f; BYTE $0x7f; BYTE $0x03
	// EMMS
	BYTE $0x0F; BYTE $0x77
	RET
</pre>