汇 编语言的优点是速度快，可以直接对硬件进行操作，这对诸如图形处理等关键应用是非常重要的。Linux 是一个用 C 语言开发的操作系统，这使得很多程序员开始忘记在 Linux 中还可以直接使用汇编这一底层语言来优化程序的性能。本文为那些在Linux 平台上编写汇编代码的程序员提供指南，介绍 Linux 汇编语言的语法格式和开发工具，并辅以具体的例子讲述如何开发实用的Linux 汇编程序。

一、简介

作 为最基本的编程语言之一，汇编语言虽然应用的范围不算很广，但重要性却勿庸置疑，因为它能够完成许多其它语言所无法完成的功能。就拿 Linux 内核来讲，虽然绝大部分代码是用 C 语言编写的，但仍然不可避免地在某些关键地方使用了汇编代码，其中主要是在 Linux 的启动部分。由于这部分代码与硬件的关系非常密切，即使是 C 语言也会有些力不从心，而汇编语言则能够很好扬长避短，最大限度地发挥硬件的性能。

大 多数情况下 Linux 程序员不需要使用汇编语言，因为即便是硬件驱动这样的底层程序在 Linux 操作系统中也可以用完全用 C 语言来实现，再加上 GCC 这一优秀的编译器目前已经能够对最终生成的代码进行很好的优化，的确有足够的理由让我们可以暂时将汇编语言抛在一边了。但实现情况是 Linux 程序员有时还是需要使用汇编，或者不得不使用汇编，理由很简单：精简、高效和 libc 无关性。假设要移植 Linux 到某一特定的嵌入式硬件环境下，首先必然面临如何减少系统大小、提高执行效率等问题，此时或许只有汇编语言能帮上忙了。

汇编语言直接同计算机的底层软件甚至硬件进行交互，它具有如下一些优点：

• 能够直接访问与硬件相关的存储器或 I/O 端口；
• 能够不受编译器的限制，对生成的二进制代码进行完全的控制；
• 能够对关键代码进行更准确的控制，避免因线程共同访问或者硬件设备共享引起的死锁；
• 能够根据特定的应用对代码做最佳的优化，提高运行速度；
• 能够最大限度地发挥硬件的功能。

同时还应该认识到，汇编语言是一种层次非常低的语言，它仅仅高于直接手工编写二进制的机器指令码，因此不可避免地存在一些缺点：

• 编写的代码非常难懂，不好维护；
• 很容易产生 bug，难于调试；
• 只能针对特定的体系结构和处理器进行优化；
• 开发效率很低，时间长且单调。

Linux 下用汇编语言编写的代码具有两种不同的形式。第一种是完全的汇编代码，指的是整个程序全部用汇编语言编写。尽管是完全的汇编代码，Linux 平台下的汇编工具也吸收了 C 语言的长处，使得程序员可以使用 #include、#ifdef 等预处理指令，并能够通过宏定义来简化代码。第二种是内嵌的汇编代码，指的是可以嵌入到C语言程序中的汇编代码片段。虽然 ANSI 的 C 语言标准中没有关于内嵌汇编代码的相应规定，但各种实际使用的 C 编译器都做了这方面的扩充，这其中当然就包括 Linux 平台下的 GCC。

二、Linux 汇编语法格式

绝大多数 Linux 程序员以前只接触过DOS/Windows 下的汇编语言，这些汇编代码都是 Intel 风格的。但在 Unix 和 Linux 系统中，更多采用的还是 AT&T 格式，两者在语法格式上有着很大的不同：

1. 在 AT&T 汇编格式中，寄存器名要加上 '%' 作为前缀；而在 Intel 汇编格式中，寄存器名不需要加前缀。例如：

   AT&T 格式	Intel 格式
   pushl %eax	push eax

2. 在 AT&T 汇编格式中，用 '$' 前缀表示一个立即操作数；而在 Intel 汇编格式中，立即数的表示不用带任何前缀。例如：

   AT&T 格式	Intel 格式
   pushl $1	push 1

3. AT&T 和 Intel 格式中的源操作数和目标操作数的位置正好相反。在 Intel 汇编格式中，目标操作数在源操作数的左边；而在 AT&T 汇编格式中，目标操作数在源操作数的右边。例如：

   AT&T 格式	Intel 格式
   addl $1, %eax	add eax, 1

4. 在 AT&T 汇编格式中，操作数的字长由操作符的最后一个字母决定，后缀'b'、'w'、'l'分别表示操作数为字节（byte，8 比特）、字（word，16 比特）和长字（long，32比特）；而在 Intel 汇编格式中，操作数的字长是用 "byte ptr" 和 "word ptr" 等前缀来表示的。例如：

   AT&T 格式	Intel 格式
   movb val, %al	mov al, byte ptr val

5. 在 AT&T 汇编格式中，绝对转移和调用指令（jump/call）的操作数前要加上'*'作为前缀，而在 Intel 格式中则不需要。

6. 远程转移指令和远程子调用指令的操作码，在 AT&T 汇编格式中为 "ljump" 和 "lcall"，而在 Intel 汇编格式中则为 "jmp far" 和 "call far"，即：

   AT&T 格式	Intel 格式
   ljump $section, $offset	jmp far section:offset
   lcall $section, $offset	call far section:offset

   与之相应的远程返回指令则为：

   AT&T 格式	Intel 格式
   lret $stack_adjust	ret far stack_adjust

7. 在 AT&T 汇编格式中，内存操作数的寻址方式是

   section:disp(base, index, scale)

   而在 Intel 汇编格式中，内存操作数的寻址方式为：

   section:[base + index*scale + disp]

   由于 Linux 工作在保护模式下，用的是 32 位线性地址，所以在计算地址时不用考虑段基址和偏移量，而是采用如下的地址计算方法：

   disp + base + index * scale

   下面是一些内存操作数的例子：

   AT&T 格式	Intel 格式
   movl -4(%ebp), %eax	mov eax, [ebp - 4]
   movl array(, %eax, 4), %eax	mov eax, [eax*4 + array]
   movw array(%ebx, %eax, 4), %cx	mov cx, [ebx + 4*eax + array]
   movb $4, %fs:(%eax)	mov fs:eax, 4

fblogo(0.5.2版本)将启动小企鹅logo改为自己的logo.png。现在遇到以下几个问

题，请求高手指点：
(1)linux_logo.h跟fblogo是什么编译器（gcc）得到的有关吗？也就是说我用PC

机上的gcc对fblogo源文件编译链接生成可执行文件fblogo，而后用./fblogo

mylogo.png生成的linux_logo.h是要编译到ARM-LINUX内核中的，行吗？
(2)为什么对于大多数png图片执行./fblogo_pc mylogo.png后有错误：
fblogo error: only palette PNGs supported
难道有什么其他的格式要求？
(3)如何对任意BMP图片生成符合要求的PNG图片，有这方面的工具吗？

非常谢谢！dengyueming@163.net

1) 首先下载linux progres
1) 首先下载linux progress project软件包lpp-0.4.2.tar.gz,解压，在contrib的boot_logo目录中运行make,生成boot_logo工具；
2) 将要在启动时显示的图片(例如:leo.bmp)用GIMP打开，另存为leo2.pcx格式文件;
3) 运行#convert -colors 214 -dither -resize 640x480 leo2.pcx logo.pcx生成214色，640x480的pcx文件；
4) 运行./boot_logo logo.pcx linux_logo.h生成linux_logo文件，将它拷贝到/include/linux/目录下，
5) 修改drivers/video/fbcon.c为
#define LOGO_H              480
#define LOGO_W              640

重新编译内核即可.

我用fblogo生成的是linux_logo.h，出现的还是小企鹅
我用fblogo生成的是linux_logo.h文件里的linux_logo[]，
而我的小企鹅是linux_logo16[],而不是linux_logo[].
我的液晶屏是16位的,从fbcon_show_logo()可以看出,当depth=16时,src = linux_logo16;而不是linux_logo.
所以我用fblogo生成的是linux_logo.h取代原来的文件后，出现的还是小企鹅，而不是我所预想的图片。
只有当depth>   =24时,才会显示linux_logo[]里的数据.
我想现在好像很少有24位以上的液晶屏吧。
请问各位仁兄：你们的情况是怎样的呢？谢谢!

Basis

This HOW-TO is written for those that want to hack the linux bootup logo. The main files in the kernel for the vesa frame buffer bootup logo are: /usr/src/linux/drivers/video/* & /usr/src/linux/include/linux/linux_logo.h Those files are responsible for the bootup logo when the kernel starts in debug or quiet mode. If we want to change this logo we have to modify the files, by using the program that is appended at the end of this text.The image file that we have to use is a PCX image file with no more than 256 colors and not bigger than 786432 pixels(1024x768).

For more information you can also read the /usr/src/linux/Documentation/fb/vesafb.txt

Do it Right

As far as I know, this is the least "hacky" way you can get a custom logo. It preserves all the previous available logos, and lets you choose between them in the Kernel config.

It requires you to edit three files in the /usr/src/linux/drivers/video/logo directory, and (obviously) create the image(s) you want to appear. It only covers 224-colour logos - if you want 16-colour or black&white ones, you'll have to figure that out on your own. I did all this yesterday, and the kernel booted fine this morning with my new logo. I changed Tux to Larry, so that's the example I'll be using throughout this tutorial. Of course you can change this to whatever you want - just make sure you're consistant with your names (eg. don't accidentally switch from "LARRY" to "BARRY" for one file).

OK, let's get down to it.

• Create your image, (I used the GIMP), and save it as a PNG (mine's called larry.png). It might be a good idea to convert it to indexed (223 colours) first - this is going to happen later anyway, so you may as well get it how you like it now (whether you want it dithered, any custom palettes, etc). I should also point out that the default images are all 80x80 pixels, but you don't have to abide by that. As far as I know, you can go as big as you want (within reason).

• Change to root, and put yourself in the directory we're working in:

su -
cd /usr/src/linux/drivers/video/logo/

• Convert the image with the netpbm tools (media-libs/netpbm):

pngtopnm /path/to/larry.png | ppmquant -fs 223 | pnmtoplainpnm > logo_larry_clut224.ppm

Don't worry if the existing images seem to have more than one file each - the *.c and *.o ones get created automatically when you compile the kernel.

• Open Kconfig in your favourite editor, and insert the following section (editing the first two lines as appropriate):

File: /usr/src/linux/drivers/video/logo/Kconfig

config LOGO_LARRY_CLUT224 bool "Gentoo-ised 224-colour logo" depends on LOGO default y

I put it underneath the entry for LOGO_LINUX_CLUT224, so it'll appear below the default logos in the kernel setup.

• Open logo.c in your favourite editor, and insert the following sections. Everything's a bit more cosily packed in this file, so make sure you get them in the right place. Again, change everything that says "larry" to your chosen name. The bit between /* these parentheses */ is a comment, and doesn't make any difference. I changed it anyway.

File: /usr/src/linux/drivers/video/logo/logo.c

Add this to the block of similar definitions at the top of the file: extern const struct linux_logo logo_larry_clut224; Add this to the section headed by "if (depth >= 8) {": #ifdef CONFIG_LOGO_LARRY_CLUT224 /* Gentoo-ised logo */ logo = &logo_larry_clut224; #endif

• Open Makefile ...blahblahblah... and add the following line to that big block of definitions at the top. You know the drill by now.

File: /usr/src/linux/drivers/video/logo/Makefile

obj-$(CONFIG_LOGO_LARRY_CLUT224) += logo_larry_clut224.o

• You should be able to repeat those steps however many times you like if you want to add more than one image.

• Save all the files, cd down to /usr/src/linux, and follow your normal kernel-rolling procedure - making sure that you select your new image in the configuration:

Linux Kernel Configuration: Kernel 2.6

Device Drivers ->      Graphics Support ->          [*] Support for frame buffer devices          [*] VESA VGA graphics support              VESA driver type ->          Console display driver support ->              [*] Video mode selection support              <*> Framebuffer Console support          Logo configuration->              [*] Bootup logo              [ ] <More logos>              [*] Your Custom Logo ...and run make. (There is no need to run make modules_install.)

• Finish up.

Finishing Up

This section is common to both methods.

• Stick your newly customised kernel in its usual place under /boot

• Make sure you've got a decent framebuffer by adding vga=0x318 or similar to your kernel's command-line:

NOTE: You can ignore this step if you selected vesafb-tng under "VESA driver type".

File: /boot/grub/menu.lst or /boot/grub/grub.conf

kernel (hd0,0)/vmlinuz root=/dev/sda3 vga=0x318

• When you boot into the new kernel, you should see the fruit of your labours!

• fblogo - Converts images to framebuffer-logo header file
• SYNOPSIS
  
  fblogo [-2] [-h] [-d] [-v] [inputfile [outputfile]]
  
• DESCRIPTION
  
  
  
  fblogo takes a png image as input and generates an appropriate linux_logo.h which can simply be included in the kernel. It will be displayed when the framebuffer device is enabled. You can also use it to generate full screen splash screens for the Linux Progress Patch.
• OPTIONS
  

  -2, --linux-2.2

  Use this option to create a linux_logo.h suitable for 2.2.x kernels. By default, a 2.4.x/2.5.x - suitable linux_logo.h will be created

  -4, --linux-2.4

  Create a linux_logo.h suitable for 2.4.x kernels (default).

  -h, --help

  Displays a short help message

  -v, --version

  Display version string

  -d, --verbose

  Verbose mode, give some feedback.

  inputfile

  The input file, has to be a PNG with less than 224 colours. If omitted, stdin will be used.

  outputfile

  This will most likely be linux_logo.h, so you can simply replace the file in the linux source tree. If omitted, stdout will be used.

• REQUIREMENTS
  
  In order to actually see the resulted logo at boot time, framebuffer-support needs to be enabled in the kernel. CONFIG_EXPERIMENTAL=y

  CONFIG_FB=y

  CONFIG_FB_VESA=y

  - or whichever driver suits your graphics card.

  You also need to tell the kernel to boot with the correct resolution, e.g. if you use lilo, specify a correct value for the vga-Parameter:

  vga=791

  This will enable booting with a resolution of 1024x768 at 16bpp. See Documentation/fb/vesafb.txt in the linux source tree for more modes.

• USAGE
  
  Generate a linux_logo.h using fblogo. This is actually the most difficult part, it is tough to get the colours looking good. The best results can be achieved by using netpbm or ImageMagick to do the quantization like so:

  Using netpbm:

  pngtopnm imagefile.png | ppmquant -fs 223 | pnmtopng | fblogo > linux_logo.h

  Using ImageMagick:

  convert -colors 223 -dither <inputimage> <image>.png ; fblogo <image>.png linux_logo.h

  
  

  Place this file in include/linux/linux_logo.h

  Update drivers/video/fbcon.c to reflect the new size of the logo:

  #define LOGO_H                   80

  #define LOGO_W                   80

  If you are going to compile a Linux 2.2 kernel, you have to do one additional step:

  Edit the file linux/include/asm-<arch>/linux_logo.h and change to the correct LINUX_LOGO_COLORS to the right number, as pointed out when using -v or inside the linux_logo.h produced by fblogo.

  Recompile your new kernel.

• SEE ALSO
  
  fblogo_patch (1)
  
• AUTHOR
  
  fblogo was originally written by Daniel Ved?y <daniel@kongsberg.online.no>, then adopted by Gordon Fraser <gordon@debian.org>. It is licensed under the terms of the GNU   General Public License.

printk在src/kernel/Printk.c中实现：
int printk(const char *fmt, ...)
在该函数内申请了一块静态内存static char printk_buf[1024];作为输出缓冲区。也就是说，不管console是否存在，printk都可以成功返回。
if (!down_trylock(&console_sem)) {
   /*
    * We own the drivers.   We can drop the spinlock and let
    * release_console_sem() print the text
    */
   spin_unlock_irqrestore(&logbuf_lock, flags);
   console_may_schedule = 0;
   release_console_sem();
printk函数，时时刻刻希望得到当前系统的console，因此只要系统调用了console init函数，printk就可以觉察到。在下次调用printk时，printk会把printk_buf中存储的数据输出到console中。这个过程在release_console_sem();实现：
for ( ; ; ) {
   spin_lock_irqsave(&logbuf_lock, flags);
   must_wake_klogd |= log_start - log_end;
   if (con_start == log_end)
    break;    /* Nothing to print */
   _con_start = con_start;
   _log_end = log_end;
   con_start = log_end;   /* Flush */
   spin_unlock_irqrestore(&logbuf_lock, flags);
call_console_drivers(_con_start, _log_end);
}
通过调用call_console_drivers，printk输出数据到console!
call_console_drivers再次调用函数：
_call_console_drivers(start_print, cur_index, msg_level);
再次调用更为底层的__call_console_drivers（）实现。
正是在__call_console_drivers（）中完成了对console驱动中write的调用！
for (con = console_drivers; con; con = con->next) {
   if ((con->flags & CON_ENABLED) && con->write)
   con->write(con, &LOG_BUF(start), end - start);
}

因此，在系统没有调用console_init();之前，printk并没有真正把数据发送到console。更改printk，可以很方便的输出到任何需要的设备上。