在Linux原始碼中，以.S為副檔名的檔案是“純”組合語言的檔案。這裡，我們結合具體的例子再介紹一些AT&T組合語言的相關知識。

  1．GNU彙編程式GAS（GNU Assembly和連線程式

當你編寫了一個程式後，就需要對其進行彙編（assembly）和連線。在Linux下有兩種方式，一種是使用匯程式設計序GAS和連線程式ld，一種是使用gcc。我們先來看一下GAS和ld:

GAS把組合語言原始檔（.o）轉換為目標檔案（.o），其基本語法如下：

as filename.s -o filename.o

一旦建立了一個目標檔案，就需要把它連線並執行，連線一個目標檔案的基本語法為：

ld filename.o -o filename

這裡 filename.o是目標檔名，而filename 是輸出(可執行) 檔案。

GAS使用的是AT&T的語法而不是Intel的語法，這就再次說明了AT&T語法是Unix世界的標準，你必須熟悉它。

如果要使用GNC的C編譯器gcc，就可以一步完成彙編和連線，例如：

gcc -o example example.S

     這裡，example.S是你的彙編程式，輸出檔案（可執行檔案）名為example。其中，副檔名必須為大寫的S，這是因為，大寫的S可以使gcc自動識別彙編程式中的C預處理命令，像#include、#define、#ifdef、 #endif等，也就是說，使用gcc進行編譯，你可以在彙編程式中使用C的預處理命令。

重點是：我們的彙編程式的字尾必須.S 注意是大寫S，不是小寫s

參考：http://oss.org.cn/kernel-book/ch02/2.6.2.htm

http://www.kerneltravel.net/chenlj/lecture6.pdf

http://blog.csdn.net/kevinshq/article/details/8228810  GNU彙編風格

---------------------------------

組合語言中出現#include, #define的問題

U_boot原始檔中，一些 .S檔案中出現了#include, #define，如board\smdk2410目錄下的lowlevel_init.S檔案：

#include <config.h> #include <version.h>

/* some parameters for the board */

/*  *  * Taken from linux/arch/arm/boot/compressed/head-s3c2410.S  *  * Copyright (C) 2002 Samsung Electronics SW.LEE  <[email protected]>  *  */

#define BWSCON 0x48000000

/* BWSCON */ #define DW8 (0x0) #define DW16 (0x1) #define DW32 (0x2) #define WAIT (0x1<<2) #define UBLB (0x1<<3)

#define B1_BWSCON   (DW32) #define B2_BWSCON   (DW16) #define B3_BWSCON   (DW16 + WAIT + UBLB) #define B4_BWSCON   (DW16) #define B5_BWSCON   (DW16) #define B6_BWSCON   (DW32) #define B7_BWSCON   (DW32)

/* BANK0CON */ #define B0_Tacs 0x0 /*  0clk */ #define B0_Tcos 0x0 /*  0clk */ #define B0_Tacc 0x7 /* 14clk */ #define B0_Tcoh 0x0 /*  0clk */ #define B0_Tah 0x0 /*  0clk */ #define B0_Tacp 0x0 #define B0_PMC 0x0 /* normal */

/* BANK1CON */ #define B1_Tacs 0x0 /*  0clk */ #define B1_Tcos 0x0 /*  0clk */ #define B1_Tacc 0x7 /* 14clk */ #define B1_Tcoh 0x0 /*  0clk */ #define B1_Tah 0x0 /*  0clk */ #define B1_Tacp 0x0 #define B1_PMC 0x0

#define B2_Tacs 0x0 #define B2_Tcos 0x0 #define B2_Tacc 0x7 #define B2_Tcoh 0x0 #define B2_Tah 0x0 #define B2_Tacp 0x0 #define B2_PMC 0x0

#define B3_Tacs 0x0 /*  0clk */ #define B3_Tcos 0x3 /*  4clk */ #define B3_Tacc 0x7 /* 14clk */ #define B3_Tcoh 0x1 /*  1clk */ #define B3_Tah 0x0 /*  0clk */ #define B3_Tacp 0x3     /*  6clk */ #define B3_PMC 0x0 /* normal */

#define B4_Tacs 0x0 /*  0clk */ #define B4_Tcos 0x0 /*  0clk */ #define B4_Tacc 0x7 /* 14clk */ #define B4_Tcoh 0x0 /*  0clk */ #define B4_Tah 0x0 /*  0clk */ #define B4_Tacp 0x0 #define B4_PMC 0x0 /* normal */ #define B5_Tacs 0x0 /*  0clk */ #define B5_Tcos 0x0 /*  0clk */ #define B5_Tacc 0x7 /* 14clk */ #define B5_Tcoh 0x0 /*  0clk */ #define B5_Tah 0x0 /*  0clk */ #define B5_Tacp 0x0 #define B5_PMC 0x0 /* normal */

#define B6_MT 0x3 /* SDRAM */ #define B6_Trcd 0x1 #define B6_SCAN 0x1 /* 9bit */

#define B7_MT 0x3 /* SDRAM */ #define B7_Trcd 0x1 /* 3clk */ #define B7_SCAN 0x1 /* 9bit */

/* REFRESH parameter */ #define REFEN 0x1 /* Refresh enable */ #define TREFMD 0x0 /* CBR(CAS before RAS)/Auto refresh */ #define Trp 0x0 /* 2clk */ #define Trc 0x3 /* 7clk */ #define Tchr 0x2 /* 3clk */ #define REFCNT 1113 /* period=15.6us, HCLK=60Mhz, (2048+1-15.6*60) */ /**************************************/

_TEXT_BASE: .word TEXT_BASE

.globl lowlevel_init lowlevel_init: /* memory control configuration */ /* make r0 relative the current location so that it */ /* reads SMRDATA out of FLASH rather than memory ! */ ldr     r0, =SMRDATA ldr r1, _TEXT_BASE sub r0, r0, r1 ldr r1, =BWSCON /* Bus Width Status Controller */ add     r2, r0, #13*4 0: ldr     r3, [r0], #4 str     r3, [r1], #4 cmp     r2, r0 bne     0b

/* everything is fine now */ mov pc, lr

.ltorg /* the literal pools origin */

SMRDATA:     .word (0+(B1_BWSCON<<4)+(B2_BWSCON<<8)+(B3_BWSCON<<12)+(B4_BWSCON<<16)+(B5_BWSCON<<20)+(B6_BWSCON<<24)+(B7_BWSCON<<28))     .word ((B0_Tacs<<13)+(B0_Tcos<<11)+(B0_Tacc<<8)+(B0_Tcoh<<6)+(B0_Tah<<4)+(B0_Tacp<<2)+(B0_PMC))     .word ((B1_Tacs<<13)+(B1_Tcos<<11)+(B1_Tacc<<8)+(B1_Tcoh<<6)+(B1_Tah<<4)+(B1_Tacp<<2)+(B1_PMC))     .word ((B2_Tacs<<13)+(B2_Tcos<<11)+(B2_Tacc<<8)+(B2_Tcoh<<6)+(B2_Tah<<4)+(B2_Tacp<<2)+(B2_PMC))     .word ((B3_Tacs<<13)+(B3_Tcos<<11)+(B3_Tacc<<8)+(B3_Tcoh<<6)+(B3_Tah<<4)+(B3_Tacp<<2)+(B3_PMC))     .word ((B4_Tacs<<13)+(B4_Tcos<<11)+(B4_Tacc<<8)+(B4_Tcoh<<6)+(B4_Tah<<4)+(B4_Tacp<<2)+(B4_PMC))     .word ((B5_Tacs<<13)+(B5_Tcos<<11)+(B5_Tacc<<8)+(B5_Tcoh<<6)+(B5_Tah<<4)+(B5_Tacp<<2)+(B5_PMC))     .word ((B6_MT<<15)+(B6_Trcd<<2)+(B6_SCAN))     .word ((B7_MT<<15)+(B7_Trcd<<2)+(B7_SCAN))     .word ((REFEN<<23)+(TREFMD<<22)+(Trp<<20)+(Trc<<18)+(Tchr<<16)+REFCNT)     .word 0x32     .word 0x30     .word 0x30

在網上查找了一些資料，現將http://gcc.gnu.org/ml/gcc-help/2003-09/msg00111.html網站上的解釋貼上如下：

On Wed, Sep 10, 2003 at 01:07:25PM -0700, garret.spears wrote:

> Refernce: > gcc-2.95.3 -m5200 -x assembler led.S

Before you compile an assembly source file, you should decide whether you want to pass it through the C preprocessor (CPP) or not. If you decide to use the preprocessor then you should name your source-file "something.S" (capital "S"); if not, then you should name it "something.s" (lowcase "s"). Provided that you follow this convention you don't *have* to specify the source-language explicitly using the "-x <lang>" option; gcc can figure-it out from the filename extension. If you don't want to or can't use this naming convention, then "-x <lang>" is required:

   gcc-2.95.3 -m5200 asmcode.s -o asmcode.o    gcc-2.95.3 -m5200 -x assembler asmcode.asm -o asmcode.o              Compile without passing the source through CPP.

   gcc-2.95.3 -m5200 asmcode1.S -o asmcode1.o    gcc-2.95.3 -m5200 -x assembler-with-cpp asmcode1.asm -o asmcode1.o

     Pass the source through CPP and then compile.

If you decide to pass your assembly through CPP, then (and only then) you *can* use all the C preprocessor goodies, like "#define", "#include", "#ifdef", etc. But remember: these are handled by the preprocessor *not* the assmebler!

>  > #DEFINE INIT_SECTION_ASM_OP  // is this a requird line or should ther be > another? >

This is neither required nor allowed! First: you're not using "assembler-with-cpp" so #DEFINE has no sense! Second: Even if you used it, CPP is CaSe-SeNsItIvE, so it should be "#define" and not "#DEFINE"   > // Base addr of internal resources & SIM resources > MBAR    EQU    0x10000000                // alt I have seen ".set

What you should have said is (observe the dot!):

  MBAR    .EQU    0x10000000 

Which is the same as saying

  #define MBAR 0x10000000

>  > Should I be using a dot h file for some of this and a dot s file for my > actual assembly coding? >

You can use ".h" files (and the respective "#include" directives) only if you pass your assembly through CPP. You don't *have* to use them, though!

> Essentially when I did this years > ago I dedicated a section to defines or equates, a section to data space, > and a section to code - assembly language.

data section:

 .data

     ... contents of the data-section ...

code section:

 .text

     ... contents of the text-section ...

or if you use an object-format that can support arbitrarily-named sections:

  .section ".data"

     ... contents of the data-section ...

  .section ".text"

     ... contents of the text-section ...

You don't need a section for the "equ"s since they produce no output!

Hope this helps /npat

--  As for systems that are not like Unix, such as MSDOS, Windows, the Macintosh, VMS, and MVS, supporting them is usually so much work that it is better if you don't.   -- Richard Stallman "GNU Coding Standards"