Framebuffer驱动程序框架 skeletonfb.c 分析

最新推荐文章于 2024-05-10 19:14:42 发布

King Seal

最新推荐文章于 2024-05-10 19:14:42 发布

阅读量3.3k

点赞数 1

分类专栏： Linux LCD driver 文章标签： Linux 驱动开发内核

本文链接：https://blog.csdn.net/xinyuwuxian/article/details/9873279

版权

Linux LCD driver 专栏收录该内容

11 篇文章 0 订阅

订阅专栏

Framebuffer驱动程序框架 skeletonfb.c 分析

最近想好好研究一下lcd驱动开发过程，lcd驱动开发主要就是framebuffer的编写了，这里只要想做framebuffer驱动的开发可能这里是必经之路，因为这里这个skeletnfb.c是framebuffer驱动程序开发的骨架，他没有具体去实现任何功能，没有针对任何设备，但是，他的作用却十分惊人，他就是使用说明文档一样，教你怎样一步一步进行framebuffer驱动编写

写贴出这个架构的实现过程，下面再慢慢分析：

具體在以下路徑：kernel/driver/video/skeletonfb.c

这里就开始慢慢啃一啃这个硬骨架吧

 
  /*
 
  * linux/drivers/video/skeletonfb.c -- Skeleton for a frame buffer device
 
  *
 
  * Modified to new api Jan 2001 by James Simmons (jsimmons@transvirtual.com)
 
  *
 
  * Created 28 Dec 1997 by Geert Uytterhoeven
 
  *
 
  *
 
  * I have started rewriting this driver as a example of the upcoming new API
 
  * The primary goal is to remove the console code from fbdev and place it
 
  * into fbcon.c. This reduces the code and makes writing a new fbdev driver
 
  * easy since the author doesn't need to worry about console internals. It
 
  * also allows the ability to run fbdev without a console/tty system on top 
 
  * of it. 
 
  *
 
  * First the roles of struct fb_info and struct display have changed. Struct
 
  * display will go away. The way the new framebuffer console code will
 
  * work is that it will act to translate data about the tty/console in 
 
  * struct vc_data to data in a device independent way in struct fb_info. Then
 
  * various functions in struct fb_ops will be called to store the device 
 
  * dependent state in the par field in struct fb_info and to change the 
 
  * hardware to that state. This allows a very clean separation of the fbdev
 
  * layer from the console layer. It also allows one to use fbdev on its own
 
  * which is a bounus for embedded devices. The reason this approach works is 
 
  * for each framebuffer device when used as a tty/console device is allocated
 
  * a set of virtual terminals to it. Only one virtual terminal can be active 
 
  * per framebuffer device. We already have all the data we need in struct 
 
  * vc_data so why store a bunch of colormaps and other fbdev specific data
 
  * per virtual terminal. 
 
  *
 
  * As you can see doing this makes the con parameter pretty much useless
 
  * for struct fb_ops functions, as it should be. Also having struct 
 
  * fb_var_screeninfo and other data in fb_info pretty much eliminates the 
 
  * need for get_fix and get_var. Once all drivers use the fix, var, and cmap
 
  * fbcon can be written around these fields. This will also eliminate the
 
  * need to regenerate struct fb_var_screeninfo, struct fb_fix_screeninfo
 
  * struct fb_cmap every time get_var, get_fix, get_cmap functions are called
 
  * as many drivers do now. 
 
  *
 
  * This file is subject to the terms and conditions of the GNU General Public
 
  * License. See the file COPYING in the main directory of this archive for
 
  * more details.
 
  */
 
 #include <linux/module.h>
 
 #include <linux/kernel.h>
 
 #include <linux/errno.h>
 
 #include <linux/string.h>
 
 #include <linux/mm.h>
 
 #include <linux/slab.h>
 
 #include <linux/delay.h>
 
 #include <linux/fb.h>
 
 #include <linux/init.h>
 
 #include <linux/pci.h>
 
 /*
 
  * This is just simple sample code.
 
  * No warranty that it actually compiles.
 
  * Even less warranty that it actually works :-)
 
  */
 
 /*
 
  * Driver data
 
  */
 
 static char *mode_option __devinitdata;
 
 /*
 
  * If your driver supports multiple boards, you should make the 
 
  * below data types arrays, or allocate them dynamically (using kmalloc()). 
 
  */ 
 
 /* 
 
  * This structure defines the hardware state of the graphics card. Normally
 
  * you place this in a header file in linux/include/video. This file usually
 
  * also includes register information. That allows other driver subsystems
 
  * and userland applications the ability to use the same header file to 
 
  * avoid duplicate work and easy porting of software. 
 
  */
 
 struct xxx_par;
 
 /*
 
  * Here we define the default structs fb_fix_screeninfo and fb_var_screeninfo
 
  * if we don't use modedb. If we do use modedb see xxxfb_init how to use it
 
  * to get a fb_var_screeninfo. Otherwise define a default var as well. 
 
  */
 
 static struct fb_fix_screeninfo xxxfb_fix __devinitdata = {
 
     .id =        "FB's name", 
 
     .type =        FB_TYPE_PACKED_PIXELS,
 
     .visual =    FB_VISUAL_PSEUDOCOLOR,
 
     .xpanstep =    1,
 
     .ypanstep =    1,
 
     .ywrapstep = 1, 
 
     .accel =    FB_ACCEL_NONE,
 
 };
 
 /*
 
  * Modern graphical hardware not only supports pipelines but some 
 
  * also support multiple monitors where each display can have its 
 
  * its own unique data. In this case each display could be 
 
  * represented by a separate framebuffer device thus a separate 
 
  * struct fb_info. Now the struct xxx_par represents the graphics
 
  * hardware state thus only one exist per card. In this case the 
 
  * struct xxx_par for each graphics card would be shared between 
 
  * every struct fb_info that represents a framebuffer on that card. 
 
  * This allows when one display changes it video resolution (info->var) 
 
  * the other displays know instantly. Each display can always be
 
  * aware of the entire hardware state that affects it because they share
 
  * the same xxx_par struct. The other side of the coin is multiple
 
  * graphics cards that pass data around until it is finally displayed
 
  * on one monitor. Such examples are the voodoo 1 cards and high end
 
  * NUMA graphics servers. For this case we have a bunch of pars, each
 
  * one that represents a graphics state, that belong to one struct 
 
  * fb_info. Their you would want to have *par point to a array of device
 
  * states and have each struct fb_ops function deal with all those 
 
  * states. I hope this covers every possible hardware design. If not
 
  * feel free to send your ideas at jsimmons@users.sf.net 
 
  */
 
 /*
 
  * If your driver supports multiple boards or it supports multiple 
 
  * framebuffers, you should make these arrays, or allocate them 
 
  * dynamically using framebuffer_alloc() and free them with
 
  * framebuffer_release().
 
  */ 
 
 static struct fb_info info;
 
 /* 
 
  * Each one represents the state of the hardware. Most hardware have
 
  * just one hardware state. These here represent the default state(s). 
 
  */
 
 static struct xxx_par __initdata current_par;
 
 int xxxfb_init(void);
 
 /**
 
  *    xxxfb_open - Optional function. Called when the framebuffer is first accessed.
 
  *    @info: frame buffer structure that represents a single frame buffer
 
  *    @user: tell us if the userland (value=1) or the console is accessing the framebuffer. 
 
  *
 
  *    This function is the first function called in the framebuffer api.
 
  *    Usually you don't need to provide this function. The case where it 
 
  *    is used is to change from a text mode hardware state to a graphics
 
  *     mode state. 
 
  *
 
  *    Returns negative errno on error, or zero on success.
 
  */
 
 static int xxxfb_open(struct fb_info *info, int user)
 
 {
 
     return 0;
 
 }
 
 /**
 
  *    xxxfb_release - Optional function. Called when the framebuffer device is closed. 
 
  *    @info: frame buffer structure that represents a single frame buffer
 
  *    @user: tell us if the userland (value=1) or the console is accessing the framebuffer. 
 
  *    
 
  *    Thus function is called when we close /dev/fb or the framebuffer 
 
  *    console system is released. Usually you don't need this function.
 
  *    The case where it is usually used is to go from a graphics state
 
  *    to a text mode state.
 
  *
 
  *    Returns negative errno on error, or zero on success.
 
  */
 
 static int xxxfb_release(struct fb_info *info, int user)
 
 {
 
     return 0;
 
 }
 
 /**
 
  * xxxfb_check_var - Optional function. Validates a var passed in. 
 
  * @var: frame buffer variable screen structure
 
  * @info: frame buffer structure that represents a single frame buffer 
 
  *    检查我们传入的var是否是硬件支持的属性，
       这个方法不改变硬件，也就是fb_info中保存的data不会改变
 
       这个方法用在我们只是想测试一下硬件，但并不是真正设置硬件属性
 
  *    Checks to see if the hardware supports the state requested by
 
  *    var passed in. This function does not alter the hardware 
 
  *    This means the data stored in struct fb_info and struct xxx_par do 
 
  *    not change. This includes the var inside of struct fb_info. 
 
  *    Do NOT change these. This function can be called on its own if we
 
  *    intent to only test a mode and not actually set it. The stuff in 
 
  *    modedb.c is a example of this. If the var passed in is slightly 
 
  *    off by what the hardware can support then we alter the var PASSED in
 
  *    to what we can do.
 
  *
 
  * For values that are off, this function must round them _up_ to the
 
  * next value that is supported by the hardware. If the value is
 
  * greater than the highest value supported by the hardware, then this
 
  * function must return -EINVAL.
 
  *
 
  * Exception to the above rule: Some drivers have a fixed mode, ie,
 
  * the hardware is already set at boot up, and cannot be changed. In
 
  * this case, it is more acceptable that this function just return
 
  * a copy of the currently working var (info->var). Better is to not
 
  * implement this function, as the upper layer will do the copying
 
  * of the current var for you.
 
  *
 
  * Note: This is the only function where the contents of var can be
 
  * freely adjusted after the driver has been registered. If you find
 
  * that you have code outside of this function that alters the content
 
  * of var, then you are doing something wrong. Note also that the
 
  * contents of info->var must be left untouched at all times after
 
  * driver registration.
 
  *
 
  *    Returns negative errno on error, or zero on success.
 
  */
 
 static int xxxfb_check_var(struct fb_var_screeninfo *var, struct fb_info *info)
 
 {
 
     /* ... */
 
     return 0;         
 
 }
 
 /**
 
  * xxxfb_set_par - Optional function. Alters the hardware state.
 
  * @info: frame buffer structure that represents a single frame buffer
 
  *    使用这个方法是用来设置framebuffer属性的，
       会改变fb_info中的par和fb_fix_screeinfo数据，
 
       但是不改变var的数据
 
  *    Using the fb_var_screeninfo in fb_info we set the resolution of the
 
  *    this particular framebuffer. This function alters the par AND the
 
  *    fb_fix_screeninfo stored in fb_info. It doesn't not alter var in 
 
  *    fb_info since we are using that data. This means we depend on the
 
  *    data in var inside fb_info to be supported by the hardware. 
 
  *
 
  * This function is also used to recover/restore the hardware to a
 
  * known working state.
 
  *
 
  *    xxxfb_check_var is always called before xxxfb_set_par to ensure that
 
  * the contents of var is always valid.
 
  *
 
  *    Again if you can't change the resolution you don't need this function.
 
  *
 
  * However, even if your hardware does not support mode changing,
 
  * a set_par might be needed to at least initialize the hardware to
 
  * a known working state, especially if it came back from another
 
  * process that also modifies the same hardware, such as X.
 
  *
 
  * If this is the case, a combination such as the following should work:
 
  *
 
  * static int xxxfb_check_var(struct fb_var_screeninfo *var,
 
  * struct fb_info *info)
 
  * {
 
  * *var = info->var;
 
  * return 0;
 
  * }
 
  *
 
  * static int xxxfb_set_par(struct fb_info *info)
 
  * {
 
  * init your hardware here
 
  * }
 
  *
 
  *    Returns negative errno on error, or zero on success.
 
  */
 
 static int xxxfb_set_par(struct fb_info *info)
 
 {
 
     struct xxx_par *par = info->par;
 
     /* ... */
 
     return 0;    
 
 }
 
 /**
 
  *     xxxfb_setcolreg - Optional function. Sets a color register.
 
  * @regno: Which register in the CLUT we are programming 
 
  * @red: The red value which can be up to 16 bits wide 
 
  *    @green: The green value which can be up to 16 bits wide 
 
  *    @blue: The blue value which can be up to 16 bits wide.
 
  *    @transp: If supported, the alpha value which can be up to 16 bits wide.
 
  * @info: frame buffer info structure
 
  * 
 
  *     Set a single color register. The values supplied have a 16 bit
 
  *     magnitude which needs to be scaled in this function for the hardware. 
 
  *    Things to take into consideration are how many color registers, if
 
  *    any, are supported with the current color visual. With truecolor mode
 
  *    no color palettes are supported. Here a pseudo palette is created
 
  *    which we store the value in pseudo_palette in struct fb_info. For
 
  *    pseudocolor mode we have a limited color palette. To deal with this
 
  *    we can program what color is displayed for a particular pixel value.
 
  *    DirectColor is similar in that we can program each color field. If
 
  *    we have a static colormap we don't need to implement this function. 
 
  * 
 
  *    Returns negative errno on error, or zero on success.
 
  */
 
 static int xxxfb_setcolreg(unsigned regno, unsigned red, unsigned green,
 
              unsigned blue, unsigned transp,
 
              struct fb_info *info)
 
 {
 
     if (regno >= 256) /* no. of hw registers */
 
        return -EINVAL;
 
     /*
 
      * Program hardware... do anything you want with transp
 
      */
 
     /* grayscale works only partially under directcolor */
 
     if (info->var.grayscale) {
 
        /* grayscale = 0.30*R + 0.59*G + 0.11*B */
 
        red = green = blue = (red * 77 + green * 151 + blue * 28) >> 8;
 
     }
 
     /* Directcolor:
 
      * var->{color}.offset contains start of bitfield
 
      * var->{color}.length contains length of bitfield
 
      * {hardwarespecific} contains width of DAC
 
      * pseudo_palette[X] is programmed to (X << red.offset) |
 
      * (X << green.offset) |
 
      * (X << blue.offset)
 
      * RAMDAC[X] is programmed to (red, green, blue)
 
      * color depth = SUM(var->{color}.length)
 
      *
 
      *     Pseudocolor:
 
      * var->{color}.offset is 0 unless the palette index takes less than
 
      * bits_per_pixel bits and is stored in the upper
 
      * bits of the pixel value
 
      * var->{color}.length is set so that 1 << length is the number of
 
      * available palette entries
 
      * pseudo_palette is not used
 
      * RAMDAC[X] is programmed to (red, green, blue)
 
      * color depth = var->{color}.length
 
      *
 
      *     Static pseudocolor:
 
      * same as Pseudocolor, but the RAMDAC is not programmed (read-only)
 
      *
 
      *     Mono01/Mono10:
 
      * Has only 2 values, black on white or white on black (fg on bg),
 
      * var->{color}.offset is 0
 
      * white = (1 << var->{color}.length) - 1, black = 0
 
      * pseudo_palette is not used
 
      * RAMDAC does not exist
 
      * color depth is always 2
 
      *
 
      *     Truecolor:
 
      * does not use RAMDAC (usually has 3 of them).
 
      * var->{color}.offset contains start of bitfield
 
      * var->{color}.length contains length of bitfield
 
      * pseudo_palette is programmed to (red << red.offset) |
 
      * (green << green.offset) |
 
      * (blue << blue.offset) |
 
      * (transp << transp.offset)
 
      * RAMDAC does not exist
 
      * color depth = SUM(var->{color}.length})
 
      *
 
      *     The color depth is used by fbcon for choosing the logo and also
 
      *     for color palette transformation if color depth < 4
 
      *
 
      *    As can be seen from the above, the field bits_per_pixel is _NOT_
 
      *    a criteria for describing the color visual.
 
      *
 
      *    A common mistake is assuming that bits_per_pixel <= 8 is pseudocolor,
 
      *    and higher than that, true/directcolor. This is incorrect, one needs
 
      *    to look at the fix->visual.
 
      *
 
      *     Another common mistake is using bits_per_pixel to calculate the color
 
      *     depth. The bits_per_pixel field does not directly translate to color
 
      *     depth. You have to compute for the color depth (using the color
 
      *     bitfields) and fix->visual as seen above.
 
      */
 
     /*
 
      *     This is the point where the color is converted to something that
 
      *     is acceptable by the hardware.
 
      */
 
 #define CNVT_TOHW(val,width) ((((val)<<(width))+0x7FFF-(val))>>16)
 
     red = CNVT_TOHW(red, info->var.red.length);
 
     green = CNVT_TOHW(green, info->var.green.length);
 
     blue = CNVT_TOHW(blue, info->var.blue.length);
 
     transp = CNVT_TOHW(transp, info->var.transp.length);
 
 #undef CNVT_TOHW
 
     /*
 
      * This is the point where the function feeds the color to the hardware
 
      * palette after converting the colors to something acceptable by
 
      * the hardware. Note, only FB_VISUAL_DIRECTCOLOR and
 
      * FB_VISUAL_PSEUDOCOLOR visuals need to write to the hardware palette.
 
      * If you have code that writes to the hardware CLUT, and it's not
 
      * any of the above visuals, then you are doing something wrong.
 
      */
 
     if (info->fix.visual == FB_VISUAL_DIRECTCOLOR ||
 
     info->fix.visual == FB_VISUAL_TRUECOLOR)
 
      write_{red|green|blue|transp}_to_clut();
 
     /*This is the point were you need to fill up the contents of
 
      * info->pseudo_palette. This structure is used _only_ by fbcon, thus
 
      * it only contains 16 entries to match the number of colors supported
 
      * by the console. The pseudo_palette is used only if the visual is
 
      * in directcolor or truecolor mode. With other visuals, the
 
      * pseudo_palette is not used. (This might change in the future.)
 
      *
 
      * The contents of the pseudo_palette is in raw pixel format. Ie, each
 
      * entry can be written directly to the framebuffer without any conversion.
 
      * The pseudo_palette is (void *). However, if using the generic
 
      * drawing functions (cfb_imageblit, cfb_fillrect), the pseudo_palette
 
      * must be casted to (u32 *) _regardless_ of the bits per pixel. If the
 
      * driver is using its own drawing functions, then it can use whatever
 
      * size it wants.
 
      */
 
     if (info->fix.visual == FB_VISUAL_TRUECOLOR ||
 
     info->fix.visual == FB_VISUAL_DIRECTCOLOR) {
 
      u32 v;
 
      if (regno >= 16)
 
          return -EINVAL;
 
      v = (red << info->var.red.offset) |
 
          (green << info->var.green.offset) |
 
          (blue << info->var.blue.offset) |
 
          (transp << info->var.transp.offset);
 
      ((u32*)(info->pseudo_palette))[regno] = v;
 
     }
 
     /* ... */
 
     return 0;
 
 }
 
 /**
 
  * xxxfb_pan_display - NOT a required function. Pans the display.
 
  * @var: frame buffer variable screen structure
 
  * @info: frame buffer structure that represents a single frame buffer
 
  *
 
  * Pan (or wrap, depending on the `vmode' field) the display using the
 
  * 'xoffset' and `yoffset' fields of the `var' structure.
 
  * If the values don't fit, return -EINVAL.
 
  *
 
  * Returns negative errno on error, or zero on success.
 
  */
 
 static int xxxfb_pan_display(struct fb_var_screeninfo *var,
 
              struct fb_info *info)
 
 {
 
     /*
 
      * If your hardware does not support panning, _do_ _not_ implement this
 
      * function. Creating a dummy function will just confuse user apps.
 
      */
 
     /*
 
      * Note that even if this function is fully functional, a setting of
 
      * 0 in both xpanstep and ypanstep means that this function will never
 
      * get called.
 
      */
 
     /* ... */
 
     return 0;
 
 }
 
 /**
 
  * xxxfb_blank - NOT a required function. Blanks the display.
 
  * @blank_mode: the blank mode we want. 
 
  * @info: frame buffer structure that represents a single frame buffer
 
  *
 
  * Blank the screen if blank_mode != FB_BLANK_UNBLANK, else unblank.
 
  * Return 0 if blanking succeeded, != 0 if un-/blanking failed due to
 
  * e.g. a video mode which doesn't support it.
 
  *
 
  * Implements VESA suspend and powerdown modes on hardware that supports
 
  * disabling hsync/vsync:
 
  *
 
  * FB_BLANK_NORMAL = display is blanked, syncs are on.
 
  * FB_BLANK_HSYNC_SUSPEND = hsync off
 
  * FB_BLANK_VSYNC_SUSPEND = vsync off
 
  * FB_BLANK_POWERDOWN = hsync and vsync off
 
  *
 
  * If implementing this function, at least support FB_BLANK_UNBLANK.
 
  * Return !0 for any modes that are unimplemented.
 
  *
 
  */
 
 static int xxxfb_blank(int blank_mode, struct fb_info *info)
 
 {
 
     /* ... */
 
     return 0;
 
 }
 
 /* ------------ Accelerated Functions --------------------- */
 
 /*
 
  * We provide our own functions if we have hardware acceleration
 
  * or non packed pixel format layouts. If we have no hardware 
 
  * acceleration, we can use a generic unaccelerated function. If using
 
  * a pack pixel format just use the functions in cfb_*.c. Each file 
 
  * has one of the three different accel functions we support.
 
  */
 
 /**
 
  * xxxfb_fillrect - REQUIRED function. Can use generic routines if 
 
  *              non acclerated hardware and packed pixel based.
 
  *             Draws a rectangle on the screen.        
 
  *
 
  * @info: frame buffer structure that represents a single frame buffer
 
  *    @region: The structure representing the rectangular region we 
 
  *         wish to draw to.
 
  *
 
  *    This drawing operation places/removes a retangle on the screen 
 
  *    depending on the rastering operation with the value of color which
 
  *    is in the current color depth format.
 
  */
 
 void xxxfb_fillrect(struct fb_info *p, const struct fb_fillrect *region)
 
 {
 
 /*    Meaning of struct fb_fillrect
 
  *
 
  *    @dx: The x and y corrdinates of the upper left hand corner of the 
 
  *    @dy: area we want to draw to. 
 
  *    @width: How wide the rectangle is we want to draw.
 
  *    @height: How tall the rectangle is we want to draw.
 
  *    @color:    The color to fill in the rectangle with. 
 
  *    @rop: The raster operation. We can draw the rectangle with a COPY
 
  *     of XOR which provides erasing effect. 
 
  */
 
 }
 
 /**
 
  * xxxfb_copyarea - REQUIRED function. Can use generic routines if
 
  * non acclerated hardware and packed pixel based.
 
  * Copies one area of the screen to another area.
 
  *
 
  * @info: frame buffer structure that represents a single frame buffer
 
  * @area: Structure providing the data to copy the framebuffer contents
 
  *     from one region to another.
 
  *
 
  * This drawing operation copies a rectangular area from one area of the
 
  *    screen to another area.
 
  */
 
 void xxxfb_copyarea(struct fb_info *p, const struct fb_copyarea *area) 
 
 {
 
 /*
 
  * @dx: The x and y coordinates of the upper left hand corner of the
 
  *    @dy: destination area on the screen.
 
  * @width: How wide the rectangle is we want to copy.
 
  * @height: How tall the rectangle is we want to copy.
 
  * @sx: The x and y coordinates of the upper left hand corner of the
 
  * @sy: source area on the screen.
 
  */
 
 }
 
 /**
 
  * xxxfb_imageblit - REQUIRED function. Can use generic routines if
 
  * non acclerated hardware and packed pixel based.
 
  * Copies a image from system memory to the screen. 
 
  *
 
  * @info: frame buffer structure that represents a single frame buffer
 
  *    @image:    structure defining the image.
 
  *
 
  * This drawing operation draws a image on the screen. It can be a 
 
  *    mono image (needed for font handling) or a color image (needed for
 
  *    tux). 
 
  */
 
 void xxxfb_imageblit(struct fb_info *p, const struct fb_image *image) 
 
 {
 
 /*
 
  * @dx: The x and y coordinates of the upper left hand corner of the
 
  *    @dy: destination area to place the image on the screen.
 
  * @width: How wide the image is we want to copy.
 
  * @height: How tall the image is we want to copy.
 
  * @fg_color: For mono bitmap images this is color data for 
 
  * @bg_color: the foreground and background of the image to
 
  *         write directly to the frmaebuffer.
 
  *    @depth:    How many bits represent a single pixel for this image.
 
  *    @data: The actual data used to construct the image on the display.
 
  *    @cmap: The colormap used for color images. 
 
  */
 
 /*
 
  * The generic function, cfb_imageblit, expects that the bitmap scanlines are
 
  * padded to the next byte. Most hardware accelerators may require padding to
 
  * the next u16 or the next u32. If that is the case, the driver can specify
 
  * this by setting info->pixmap.scan_align = 2 or 4. See a more
 
  * comprehensive description of the pixmap below.
 
  */
 
 }
 
 /**
 
  *    xxxfb_cursor -     OPTIONAL. If your hardware lacks support
 
  *            for a cursor, leave this field NULL.
 
  *
 
  * @info: frame buffer structure that represents a single frame buffer
 
  *    @cursor: structure defining the cursor to draw.
 
  *
 
  * This operation is used to set or alter the properities of the cursor.
  *
 
  *    Returns negative errno on error, or zero on success.
 
  */
 
 int xxxfb_cursor(struct fb_info *info, struct fb_cursor *cursor)
 
 {
 
 /*
 
  * @set:     Which fields we are altering in struct fb_cursor 
 
  *    @enable: Disable or enable the cursor 
 
  * @rop:     The bit operation we want to do. 
 
  * @mask: This is the cursor mask bitmap. 
 
  * @dest: A image of the area we are going to display the cursor.
 
  *        Used internally by the driver.     
 
  * @hot:    The hot spot. 
 
  *    @image:    The actual data for the cursor image.
 
  *
 
  * NOTES ON FLAGS (cursor->set):
 
  *
 
  * FB_CUR_SETIMAGE - the cursor image has changed (cursor->image.data)
 
  * FB_CUR_SETPOS - the cursor position has changed (cursor->image.dx|dy)
 
  * FB_CUR_SETHOT - the cursor hot spot has changed (cursor->hot.dx|dy)
 
  * FB_CUR_SETCMAP - the cursor colors has changed (cursor->fg_color|bg_color)
 
  * FB_CUR_SETSHAPE - the cursor bitmask has changed (cursor->mask)
 
  * FB_CUR_SETSIZE - the cursor size has changed (cursor->width|height)
 
  * FB_CUR_SETALL - everything has changed
 
  *
 
  * NOTES ON ROPs (cursor->rop, Raster Operation)
 
  *
 
  * ROP_XOR - cursor->image.data XOR cursor->mask
 
  * ROP_COPY - curosr->image.data AND cursor->mask
 
  *
 
  * OTHER NOTES:
 
  *
 
  * - fbcon only supports a 2-color cursor (cursor->image.depth = 1)
 
  * - The fb_cursor structure, @cursor, _will_ always contain valid
 
  * fields, whether any particular bitfields in cursor->set is set
 
  * or not.
 
  */
 
 }
 
 /**
 
  *    xxxfb_rotate - NOT a required function. If your hardware
 
  *            supports rotation the whole screen then 
 
  *            you would provide a hook for this. 
 
  *
 
  * @info: frame buffer structure that represents a single frame buffer
 
  *    @angle: The angle we rotate the screen. 
 
  *
 
  * This operation is used to set or alter the properities of the
 
  *    cursor.
 
  */
 
 void xxxfb_rotate(struct fb_info *info, int angle)
 
 {
 
 /* Will be deprecated */
 
 }
 
 /**
 
  *    xxxfb_sync - NOT a required function. Normally the accel engine 
 
  *         for a graphics card take a specific amount of time.
 
  *         Often we have to wait for the accelerator to finish
 
  *         its operation before we can write to the framebuffer
 
  *         so we can have consistent display output. 
 
  *
 
  * @info: frame buffer structure that represents a single frame buffer
 
  *
 
  * If the driver has implemented its own hardware-based drawing function,
 
  * implementing this function is highly recommended.
 
  */
 
 int xxxfb_sync(struct fb_info *info)
 
 {
 
     return 0;
 
 }
 
     /*
 
      * Frame buffer operations
 
      */
 
 static struct fb_ops xxxfb_ops = {
 
     .owner        = THIS_MODULE,
 
     .fb_open    = xxxfb_open,
 
     .fb_read    = xxxfb_read,
 
     .fb_write    = xxxfb_write,
 
     .fb_release    = xxxfb_release,
 
     .fb_check_var    = xxxfb_check_var,
 
     .fb_set_par    = xxxfb_set_par,
 
     .fb_setcolreg    = xxxfb_setcolreg,
 
     .fb_blank    = xxxfb_blank,
 
     .fb_pan_display    = xxxfb_pan_display,
 
     .fb_fillrect    = xxxfb_fillrect,     /* Needed !!! */
 
     .fb_copyarea    = xxxfb_copyarea,    /* Needed !!! */
 
     .fb_imageblit    = xxxfb_imageblit,    /* Needed !!! */
 
     .fb_cursor    = xxxfb_cursor,        /* Optional !!! */
 
     .fb_rotate    = xxxfb_rotate,
 
     .fb_sync    = xxxfb_sync,
 
     .fb_ioctl    = xxxfb_ioctl,
 
     .fb_mmap    = xxxfb_mmap,
 
 };
 
 /* ------------------------------------------------------------------------- */
 
 /*
 
  * Initialization
 
  */
 
 这里他们的定义是不一样的，唯一的不同
 /* static int __init xxfb_probe (struct platform_device *pdev) -- for platform devs */
 
 static int __devinit xxxfb_probe(struct pci_dev *dev, const struct pci_device_id *ent)
 {
 
     struct fb_info *info;
 
     struct xxx_par *par;
 
     struct device *device = &dev->dev; /* or &pdev->dev */
 
     int cmap_len, retval;    
 
     /*
 
      * Dynamically allocate info and par
 
      */
 
     info = framebuffer_alloc(sizeof(struct xxx_par), device);
 
     if (!info) {
 
      /* goto error path */
 
     }
 
     par = info->par;把我们驱动中的私有数据指向info->par，我们在上面已经通过framebuffer_alloc为他申请了内存
 
     /* 
 
      * Here we set the screen_base to the virtual memory address
 
      * for the framebuffer. Usually we obtain the resource address
 
      * from the bus layer and then translate it to virtual memory
 
      * space via ioremap. Consult ioport.h. 
 
      */
 
     这里注释也已经说的很清楚了，
     我们通常通过注册好的设备信息获取到memory address，
     然后把这个地址转换为virtual memory通过ioremap这个方法
 
     info->screen_base = framebuffer_virtual_memory;
 
     info->fbops = &xxxfb_ops; 
     这个定义好的xxxfb_ops结构中包含了用户空间访问driver的所有方法，
     用户访问/dev/fb0最终通过fbmem中注册的方法调用这里的方法
 
     info->fix = xxxfb_fix; /* this will be the only time xxxfb_fix will be used, so mark it as __devinitdata */
     info->pseudo_palette = pseudo_palette; /* The pseudopalette is an 16-member array*/
     /*
 
      * Set up flags to indicate what sort of acceleration your
 
      * driver can provide (pan/wrap/copyarea/etc.) and whether it
 
      * is a module -- see FBINFO_* in include/linux/fb.h
 
      *
 
      * If your hardware can support any of the hardware accelerated functions
 
      * fbcon performance will improve if info->flags is set properly.
 
      *
 
      * FBINFO_HWACCEL_COPYAREA - hardware moves
 
      * FBINFO_HWACCEL_FILLRECT - hardware fills
 
      * FBINFO_HWACCEL_IMAGEBLIT - hardware mono->color expansion
 
      * FBINFO_HWACCEL_YPAN - hardware can pan display in y-axis
 
      * FBINFO_HWACCEL_YWRAP - hardware can wrap display in y-axis
 
      * FBINFO_HWACCEL_DISABLED - supports hardware accels, but disabled
 
      * FBINFO_READS_FAST - if set, prefer moves over mono->color expansion
 
      * FBINFO_MISC_TILEBLITTING - hardware can do tile blits
 
      *
 
      * NOTE: These are for fbcon use only.
 
      */
 
     info->flags = FBINFO_DEFAULT;
 
 /********************* This stage is optional ******************************/
 
      /*
 
      * The struct pixmap is a scratch pad for the drawing functions. This
 
      * is where the monochrome bitmap is constructed by the higher layers
 
      * and then passed to the accelerator. For drivers that uses
 
      * cfb_imageblit, you can skip this part. For those that have a more
 
      * rigorous requirement, this stage is needed
 
      */
 
     /* PIXMAP_SIZE should be small enough to optimize drawing, but not
 
      * large enough that memory is wasted. A safe size is
 
      * (max_xres * max_font_height/8). max_xres is driver dependent,
 
      * max_font_height is 32.
 
      */
 
     info->pixmap.addr = kmalloc(PIXMAP_SIZE, GFP_KERNEL);
 
     if (!info->pixmap.addr) {
 
      /* goto error */
 
     }
 
     info->pixmap.size = PIXMAP_SIZE;
 
     /*
 
      * FB_PIXMAP_SYSTEM - memory is in system ram
 
      * FB_PIXMAP_IO - memory is iomapped
 
      * FB_PIXMAP_SYNC - if set, will call fb_sync() per access to pixmap,
 
      * usually if FB_PIXMAP_IO is set.
 
      *
 
      * Currently, FB_PIXMAP_IO is unimplemented.
 
      */
 
     info->pixmap.flags = FB_PIXMAP_SYSTEM;
 
     /*
 
      * scan_align is the number of padding for each scanline. It is in bytes.
 
      * Thus for accelerators that need padding to the next u32, put 4 here.
 
      */
 
     info->pixmap.scan_align = 4;
 
     /*
 
      * buf_align is the amount to be padded for the buffer. For example,
 
      * the i810fb needs a scan_align of 2 but expects it to be fed with
 
      * dwords, so a buf_align = 4 is required.
 
      */
 
     info->pixmap.buf_align = 4;
 
     /* access_align is how many bits can be accessed from the framebuffer
 
      * ie. some epson cards allow 16-bit access only. Most drivers will
 
      * be safe with u32 here.
 
      *
 
      * NOTE: This field is currently unused.
 
      */
 
     info->pixmap.access_align = 32;
 
 /***************************** End optional stage ***************************/
 
     /*
 
      * This should give a reasonable default video mode. The following is
 
      * done when we can set a video mode. 
 
      */
 
     if (!mode_option)
 
     mode_option = "640x480@60";         
 
     retval = fb_find_mode(&info->var, info, mode_option, NULL, 0, NULL, 8);
 
     if (!retval || retval == 4)
 
     return -EINVAL;            
 
     /* This has to be */
 
     if (fb_alloc_cmap(&info->cmap, cmap_len, 0))
 
     return -ENOMEM;
 
     /* 
 
      * The following is done in the case of having hardware with a static 
 
      * mode. If we are setting the mode ourselves we don't call this. 
 
      */    
 
     info->var = xxxfb_var;
 
     /*
 
      * For drivers that can...
 
      */
 
     xxxfb_check_var(&info->var, info);
 
     /*
 
      * Does a call to fb_set_par() before register_framebuffer needed? This
 
      * will depend on you and the hardware. If you are sure that your driver
 
      * is the only device in the system, a call to fb_set_par() is safe.
 
      *
 
      * Hardware in x86 systems has a VGA core. Calling set_par() at this
 
      * point will corrupt the VGA console, so it might be safer to skip a
 
      * call to set_par here and just allow fbcon to do it for you.
 
      */
 
     /* xxxfb_set_par(info); */
 
     if (register_framebuffer(info) < 0) {
     注册设备，fbmem中负责注册了一个fb 字符设备，
     只有调用了这个方法，才真正创建了这个fb字符设备
 
     fb_dealloc_cmap(&info->cmap);
 
     return -EINVAL;
 
     }
 
     printk(KERN_INFO "fb%d: %s frame buffer device\n", info->node,
 
      info->fix.id);
 
     pci_set_drvdata(dev, info); /* or platform_set_drvdata(pdev, info) */
     方便后面使用info这个结构中的所有数据
 
     return 0;
 
 }
 
     /*
 
      * Cleanup
 
      */
 
 /* static void __devexit xxxfb_remove(struct platform_device *pdev) */
 
 static void __devexit xxxfb_remove(struct pci_dev *dev)
 
 {
 
     struct fb_info *info = pci_get_drvdata(dev);
 
     /* or platform_get_drvdata(pdev); */
 
     if (info) {
 
         unregister_framebuffer(info);
 
         fb_dealloc_cmap(&info->cmap);
 
         /* ... */
 
         framebuffer_release(info);
 
     }
 
 }
 
 #ifdef CONFIG_PCI
 
 #ifdef CONFIG_PM
 
 /**
 
  *    xxxfb_suspend - Optional but recommended function. Suspend the device.
 
  *    @dev: PCI device
 
  *    @msg: the suspend event code.
 
  *
 
  * See Documentation/power/devices.txt for more information
 
  */
 
 static int xxxfb_suspend(struct pci_dev *dev, pm_message_t msg)
 
 {
 
     struct fb_info *info = pci_get_drvdata(dev);
 
     struct xxxfb_par *par = info->par;
 
     /* suspend here */
 
     return 0;
 
 }
 
 /**
 
  *    xxxfb_resume - Optional but recommended function. Resume the device.
 
  *    @dev: PCI device
 
  *
 
  * See Documentation/power/devices.txt for more information
 
  */
 
 static int xxxfb_resume(struct pci_dev *dev)
 
 {
 
     struct fb_info *info = pci_get_drvdata(dev);
 
     struct xxxfb_par *par = info->par;
 
     /* resume here */
 
     return 0;
 
 }
 
 #else
 
 #define xxxfb_suspend NULL
 
 #define xxxfb_resume NULL
 
 #endif /* CONFIG_PM */
 
 static struct pci_device_id xxxfb_id_table[] = {
 
     { PCI_VENDOR_ID_XXX, PCI_DEVICE_ID_XXX,
 
      PCI_ANY_ID, PCI_ANY_ID, PCI_BASE_CLASS_DISPLAY << 16,
 
      PCI_CLASS_MASK, 0 },
 
     { 0, }
 
 };
 
 /* For PCI drivers */
 
 static struct pci_driver xxxfb_driver = {
 
     .name =        "xxxfb",
 
     .id_table =    xxxfb_id_table,
 
     .probe =    xxxfb_probe,
 
     .remove =    __devexit_p(xxxfb_remove),
 
     .suspend = xxxfb_suspend, /* optional but recommended */
 
     .resume = xxxfb_resume, /* optional but recommended */
 
 };
 
 MODULE_DEVICE_TABLE(pci, xxxfb_id_table);
 
 int __init xxxfb_init(void)
 
 {
 
     /*
 
      * For kernel boot options (in 'video=xxxfb:<options>' format)
 
      */
 
 #ifndef MODULE
 
     char *option = NULL;
 
     if (fb_get_options("xxxfb", &option))
 
         return -ENODEV;
 
     xxxfb_setup(option);
 
 #endif
 
     return pci_register_driver(&xxxfb_driver);
 
 }
 
 static void __exit xxxfb_exit(void)
 
 {
 
     pci_unregister_driver(&xxxfb_driver);
 
 }
 
 #else /* non PCI, platform drivers */
 
 #include <linux/platform_device.h>
 
 /* for platform devices */
 
 #ifdef CONFIG_PM
 
 /**
 
  *    xxxfb_suspend - Optional but recommended function. Suspend the device.
 
  *    @dev: platform device
 
  *    @msg: the suspend event code.
 
  *
 
  * See Documentation/power/devices.txt for more information
 
  */
 
 static int xxxfb_suspend(struct platform_device *dev, pm_message_t msg)
 
 {
 
     struct fb_info *info = platform_get_drvdata(dev);
 
     struct xxxfb_par *par = info->par;
 
     /* suspend here */
 
     return 0;
 
 }
 
 /**
 
  *    xxxfb_resume - Optional but recommended function. Resume the device.
 
  *    @dev: platform device
 
  *
 
  * See Documentation/power/devices.txt for more information
 
  */
 
 static int xxxfb_resume(struct platform_dev *dev)
 
 {
 
     struct fb_info *info = platform_get_drvdata(dev);
 
     struct xxxfb_par *par = info->par;
 
     /* resume here */
 
     return 0;
 
 }
 
 #else
 
 #define xxxfb_suspend NULL
 
 #define xxxfb_resume NULL
 
 #endif /* CONFIG_PM */
 
 static struct platform_device_driver xxxfb_driver = {
 
     .probe = xxxfb_probe,
 
     .remove = xxxfb_remove,
 
     .suspend = xxxfb_suspend, /* optional but recommended */
 
     .resume = xxxfb_resume, /* optional but recommended */
 
     .driver = {
 
         .name = "xxxfb",
 
     },
 
 };
 
 static struct platform_device *xxxfb_device;
 
 #ifndef MODULE
 
     /*
 
      * Setup
 
      */
 
 /*
 
  * Only necessary if your driver takes special options,
 
  * otherwise we fall back on the generic fb_setup().
 
  */
 
 int __init xxxfb_setup(char *options)
 
 {
 
     /* Parse user speficied options (`video=xxxfb:') */
 
 }
 
 #endif /* MODULE */
 
 static int __init xxxfb_init(void)
 
 {
 
     int ret;
 
     /*
 
      * For kernel boot options (in 'video=xxxfb:<options>' format)
 
      */
 
 #ifndef MODULE
 
     char *option = NULL;
 
     if (fb_get_options("xxxfb", &option))
 
         return -ENODEV;
 
     xxxfb_setup(option);
 
 #endif
 
     ret = platform_driver_register(&xxxfb_driver);
 
     if (!ret) {
 
         xxxfb_device = platform_device_register_simple("xxxfb", 0,
 
                                 NULL, 0);
 
         if (IS_ERR(xxxfb_device)) {
 
             platform_driver_unregister(&xxxfb_driver);
 
             ret = PTR_ERR(xxxfb_device);
 
         }
 
     }
 
     return ret;
 
 }
 
 static void __exit xxxfb_exit(void)
 
 {
 
     platform_device_unregister(xxxfb_device);
 
     platform_driver_unregister(&xxxfb_driver);
 
 }
 
 #endif /* CONFIG_PCI */
 
 /* ------------------------------------------------------------------------- */
 
     /*
 
      * Modularization
 
      */
 
 module_init(xxxfb_init);
 
 module_exit(xxxfb_remove);
 
 MODULE_LICENSE("GPL");

这里就开始慢慢啃一啃这个硬骨架吧

毫无疑问，先找到module_init(xxxfb_init)，xxxfb_init这个方式是这个驱动的入口，我们要找到这个方法，代码中我已经标注出来，原来在这个架构中采用了两种方法来实现framebuffer驱动，分别是pci子系统方式和platform子系统方式，这里他们的实现方法基本相似，只是在匹配device时有所不同，我们没有必要过于关注这里，大家也可以看到，他们的proble方法是一样的，这个proble才是我们要真正好好分析的

第一件事就是为fb_info动态申请内存空间，很有必要分析一下这个方法

/**
 * framebuffer_alloc - creates a new frame buffer info structure
 *
 * @size: size of driver private data, can be zero
 * @dev: pointer to the device for this fb, this can be NULL
 *
 * Creates a new frame buffer info structure. Also reserves @size bytes
 * for driver private data (info->par). info->par (if any) will be
 * aligned to sizeof(long).
 *
 * Returns the new structure, or NULL if an error occured.
 *
 */
struct fb_info *framebuffer_alloc(size_t size, struct device *dev)
{
#define BYTES_PER_LONG (BITS_PER_LONG/8)
#define PADDING (BYTES_PER_LONG - (sizeof(struct fb_info) % BYTES_PER_LONG))
    int fb_info_size = sizeof(struct fb_info);
    struct fb_info *info;
    char *p;

    if (size)
        fb_info_size += PADDING;

    p = kzalloc(fb_info_size + size, GFP_KERNEL);

    if (!p)
        return NULL;

    info = (struct fb_info *) p;

    if (size)
        info->par = p + fb_info_size;

    info->device = dev;

#ifdef CONFIG_FB_BACKLIGHT
    mutex_init(&info->bl_curve_mutex);
#endif

    return info;
#undef PADDING
#undef BYTES_PER_LONG
}

其实这里注释也已经说清楚了，这个方法不止为fb_info申请了内存空间，还为fb_info中的一个void型指针par分配了内存空间，这个par将会保存之后在这个驱动中的私有变量，是一个结构，保存了很多driver的相关信息，保存到par中，之后想要使用的时候就很容易得到。

这个架构中的其他方法都没有实现，他们就是访问driver时候具体的处理函数，就是要我们要去花大力气去做的事情，之后在真正的framebuffer驱动程序中再去好好分析这些方法吧。

附上不錯的一篇文章：http://www.linuxgraphics.cn/graphics/writing_lcd_driver.html