基于V4L2的视频驱动开发（2）

三、 V4L2 API及数据结构

V4L2是V4L的升级版本，为linux下视频设备程序提供了一套接口规范。包括一套数据结构和底层V4L2驱动接口。

1、常用的结构体在内核目录include/linux/videodev2.h中定义

struct v4l2_requestbuffers        //申请帧缓冲，对应命令VIDIOC_REQBUFS

        struct v4l2_capability        //视频设备的功能，对应命令VIDIOC_QUERYCAP

        struct v4l2_input        //视频输入信息，对应命令VIDIOC_ENUMINPUT

        struct v4l2_standard        //视频的制式，比如PAL，NTSC，对应命令VIDIOC_ENUMSTD

        struct v4l2_format        //帧的格式，对应命令VIDIOC_G_FMT、VIDIOC_S_FMT等

        struct v4l2_buffer        //驱动中的一帧图像缓存，对应命令VIDIOC_QUERYBUF

        struct v4l2_crop        //视频信号矩形边框

        v4l2_std_id        //视频制式

2、常用的IOCTL接口命令也在include/linux/videodev2.h中定义

VIDIOC_REQBUFS //分配内存

        VIDIOC_QUERYBUF         //把VIDIOC_REQBUFS中分配的数据缓存转换成物理地址

        VIDIOC_QUERYCAP        //查询驱动功能

        VIDIOC_ENUM_FMT        //获取当前驱动支持的视频格式

        VIDIOC_S_FMT        //设置当前驱动的频捕获格式

        VIDIOC_G_FMT        //读取当前驱动的频捕获格式

        VIDIOC_TRY_FMT        //验证当前驱动的显示格式

        VIDIOC_CROPCAP        //查询驱动的修剪能力

        VIDIOC_S_CROP        //设置视频信号的矩形边框

        VIDIOC_G_CROP        //读取视频信号的矩形边框

        VIDIOC_QBUF        //把数据从缓存中读取出来

        VIDIOC_DQBUF        //把数据放回缓存队列

        VIDIOC_STREAMON        //开始视频显示函数

        VIDIOC_STREAMOFF        //结束视频显示函数

        VIDIOC_QUERYSTD         //检查当前视频设备支持的标准，例如PAL或NTSC。

3、操作流程

V4L2提供了很多访问接口，你可以根据具体需要选择操作方法。需要注意的是，很少有驱动完全实现了所有的接口功能。所以在使用时需要参考驱动源码，或仔细阅读驱动提供者的使用说明。

下面列举出一种操作的流程，供参考。

（1）打开设备文件

                  int fd = open(Devicename,mode);

                  Devicename：/dev/video0、/dev/video1 ……

                  Mode：O_RDWR [| O_NONBLOCK]

如果使用非阻塞模式调用视频设备，则当没有可用的视频数据时，不会阻塞，而立刻返回。

（2）取得设备的capability

          struct v4l2_capability capability；

                  int ret = ioctl(fd, VIDIOC_QUERYCAP, &capability);

看看设备具有什么功能，比如是否具有视频输入特性。

（3）选择视频输入

          struct v4l2_input input；

                  ……初始化input

                  int ret = ioctl(fd, VIDIOC_QUERYCAP, &input);

一个视频设备可以有多个视频输入。如果只有一路输入，这个功能可以没有。

（4）检测视频支持的制式

          v4l2_std_id std;

                  do {

                                ret = ioctl(fd, VIDIOC_QUERYSTD, &std);

                   } while (ret == -1 && errno == EAGAIN);

                switch (std) {

                case V4L2_STD_NTSC:

                                //……

                case V4L2_STD_PAL:

                                //……

                }

（5）设置视频捕获格式

          struct v4l2_format fmt;

                  fmt.type = V4L2_BUF_TYPE_VIDEO_OUTPUT;

                  fmt.fmt.pix.pixelformat = V4L2_PIX_FMT_UYVY;

                  fmt.fmt.pix.height = height;

                  fmt.fmt.pix.width = width;

                  fmt.fmt.pix.field = V4L2_FIELD_INTERLACED;

                  ret = ioctl(fd, VIDIOC_S_FMT, &fmt);

                  if(ret) {

                          perror("VIDIOC_S_FMT\n");

                          close(fd);

                          return -1;

                  }

（6）向驱动申请帧缓存

          struct v4l2_requestbuffers req;

                   if (ioctl(fd, VIDIOC_REQBUFS, &req) == -1) {

                          return -1;

                   }

v4l2_requestbuffers结构中定义了缓存的数量，驱动会据此申请对应数量的视频缓存。多个缓存可以用于建立FIFO，来提高视频采集的效率。

（7）获取每个缓存的信息，并mmap到用户空间

          typedef struct VideoBuffer {

                          void *start;

                          size_t length;

                  } VideoBuffer;

          VideoBuffer* buffers = calloc( req.count, sizeof(*buffers) );

                  struct v4l2_buffer buf;

          for (numBufs = 0; numBufs < req.count; numBufs++) {//映射所有的缓存

                          memset( &buf, 0, sizeof(buf) );

                          buf.type = V4L2_BUF_TYPE_VIDEO_CAPTURE;

                          buf.memory = V4L2_MEMORY_MMAP;

                          buf.index = numBufs;

                          if (ioctl(fd, VIDIOC_QUERYBUF, &buf) == -1) {//获取到对应index的缓存信息，此处主要利用length信息及offset信息来完成后面的mmap操作。

                                  return -1;

                          }

                  buffers[numBufs].length = buf.length;

                          // 转换成相对地址

                          buffers[numBufs].start = mmap(NULL, buf.length,

                                  PROT_READ | PROT_WRITE,

                                  MAP_SHARED,

                                  fd, buf.m.offset);

                  if (buffers[numBufs].start == MAP_FAILED) {

                                  return -1;

                          }

（8）开始采集视频

          int buf_type= V4L2_BUF_TYPE_VIDEO_CAPTURE；

                  int ret = ioctl(fd, VIDIOC_STREAMON, &buf_type);

（9）取出FIFO缓存中已经采样的帧缓存

          struct v4l2_buffer buf;

                  memset(&buf,0,sizeof(buf));

                  buf.type=V4L2_BUF_TYPE_VIDEO_CAPTURE;

                  buf.memory=V4L2_MEMORY_MMAP;

                  buf.index=0;//此值由下面的ioctl返回

                  if (ioctl(fd, VIDIOC_DQBUF, &buf) == -1)

                  {

                          return -1;

                  }

根据返回的buf.index找到对应的mmap映射好的缓存，取出视频数据。

（10）将刚刚处理完的缓冲重新入队列尾，这样可以循环采集

          if (ioctl(fd, VIDIOC_QBUF, &buf) == -1) {

                          return -1;

                  }

（11）停止视频的采集

          int ret = ioctl(fd, VIDIOC_STREAMOFF, &buf_type);

（12）关闭视频设备

          close(fd);

四、 V4L2驱动框架

上述流程的各个操作都需要有底层V4L2驱动的支持。内核中有一些非常完善的例子。

比如：linux-2.6.26内核目录/drivers/media/video//zc301/zc301_core.c 中的ZC301视频驱动代码。上面的V4L2操作流程涉及的功能在其中都有实现。

1、V4L2驱动注册、注销函数

Video核心层（drivers/media/video/videodev.c）提供了注册函数

              int video_register_device(struct video_device *vfd, int type, int nr)

                    video_device: 要构建的核心数据结构

                    Type: 表示设备类型，此设备号的基地址受此变量的影响

                    Nr: 如果end-base>nr>0 ：次设备号=base（基准值，受type影响）+nr；

                    否则：系统自动分配合适的次设备号

具体驱动只需要构建video_device结构，然后调用注册函数既可。

如：zc301_core.c中的

                    err = video_register_device(cam->v4ldev, VFL_TYPE_GRABBER,

                              video_nr[dev_nr]);

          Video核心层（drivers/media/video/videodev.c）提供了注销函数

                    void video_unregister_device(struct video_device *vfd)

2、struct video_device 的构建

      video_device结构包含了视频设备的属性和操作方法。参见zc301_core.c

      strcpy(cam->v4ldev->name, "ZC0301[P] PC Camera");

              cam->v4ldev->owner = THIS_MODULE;

              cam->v4ldev->type = VID_TYPE_CAPTURE | VID_TYPE_SCALES;

              cam->v4ldev->fops = &zc0301_fops;

              cam->v4ldev->minor = video_nr[dev_nr];

              cam->v4ldev->release = video_device_release;

              video_set_drvdata(cam->v4ldev, cam);

大家发现在这个zc301的驱动中并没有实现struct video_device中的很多操作函数，如：vidioc_querycap、vidioc_g_fmt_cap等。主要原因是struct file_operations zc0301_fops中的zc0301_ioctl实现了前面的所有ioctl操作。所以就不需要在struct video_device再实现struct video_device中的那些操作了。

另一种实现方法如下：

static struct video_device camif_dev =

        {

                .name = "s3c2440 camif",

                .type = VID_TYPE_CAPTURE|VID_TYPE_SCALES|VID_TYPE_SUBCAPTURE,

                .fops = &camif_fops,

                .minor = -1,

                .release = camif_dev_release,

                .vidioc_querycap = vidioc_querycap,

                .vidioc_enum_fmt_cap = vidioc_enum_fmt_cap,

                .vidioc_g_fmt_cap = vidioc_g_fmt_cap,

                .vidioc_s_fmt_cap = vidioc_s_fmt_cap,

                .vidioc_queryctrl = vidioc_queryctrl,

                .vidioc_g_ctrl = vidioc_g_ctrl,

                .vidioc_s_ctrl = vidioc_s_ctrl,

        };

        static struct file_operations camif_fops =

        {

                .owner = THIS_MODULE,

                .open = camif_open,

                .release = camif_release,

                .read = camif_read,

                .poll = camif_poll,

                .ioctl = video_ioctl2, /* V4L2 ioctl handler */

                .mmap = camif_mmap,

                .llseek = no_llseek,

        };

注意：video_ioctl2是videodev.c中是实现的。video_ioctl2中会根据ioctl不同的cmd来调用video_device中的操作方法。

3、Video核心层的实现

参见内核/drivers/media/videodev.c

（1）注册256个视频设备

static int __init videodev_init(void)

        {

                int ret;

                if (register_chrdev(VIDEO_MAJOR, VIDEO_NAME, &video_fops)) {

                        return -EIO;

                }

                ret = class_register(&video_class);

                ……

        }

上面的代码注册了256个视频设备，并注册了video_class类。video_fops为这256个设备共同的操作方法。

（2）V4L2驱动注册函数的实现

int video_register_device(struct video_device *vfd, int type, int nr)

        {

                int i=0;

                int base;

                int end;

                int ret;

                char *name_base;

        switch(type) //根据不同的type确定设备名称、次设备号

                {

                        case VFL_TYPE_GRABBER:

                                base=MINOR_VFL_TYPE_GRABBER_MIN;

                                end=MINOR_VFL_TYPE_GRABBER_MAX+1;

                                name_base = "video";

                                break;

                        case VFL_TYPE_VTX:

                                base=MINOR_VFL_TYPE_VTX_MIN;

                                end=MINOR_VFL_TYPE_VTX_MAX+1;

                                name_base = "vtx";

                                break;

                        case VFL_TYPE_VBI:

                                base=MINOR_VFL_TYPE_VBI_MIN;

                                end=MINOR_VFL_TYPE_VBI_MAX+1;

                                name_base = "vbi";

                                break;

                        case VFL_TYPE_RADIO:

                                base=MINOR_VFL_TYPE_RADIO_MIN;

                                end=MINOR_VFL_TYPE_RADIO_MAX+1;

                                name_base = "radio";

                                break;

                        default:

                                printk(KERN_ERR "%s called with unknown type: %d\n",

                                        __func__, type);

                                return -1;

                }

        /* 计算出次设备号 */

                mutex_lock(&videodev_lock);

                if (nr >= 0 && nr < end-base) {

                        /* use the one the driver asked for */

                        i = base+nr;

                        if (NULL != video_device[i]) {

                                mutex_unlock(&videodev_lock);

                                return -ENFILE;

                        }

                } else {

                        /* use first free */

                        for(i=base;i<end;i++)

                                if (NULL == video_device[i])

                                        break;

                        if (i == end) {

                                mutex_unlock(&videodev_lock);

                                return -ENFILE;

                        }

                }

                video_device[i]=vfd; //保存video_device结构指针到系统的结构数组中，最终的次设备号和i相关。

                vfd->minor=i;

                mutex_unlock(&videodev_lock);

                mutex_init(&vfd->lock);

        /* sysfs class */

                memset(&vfd->class_dev, 0x00, sizeof(vfd->class_dev));

                if (vfd->dev)

                        vfd->class_dev.parent = vfd->dev;

                vfd->class_dev.class = &video_class;

                vfd->class_dev.devt = MKDEV(VIDEO_MAJOR, vfd->minor);

                sprintf(vfd->class_dev.bus_id, "%s%d", name_base, i - base);//最后在/dev目录下的名称

                ret = device_register(&vfd->class_dev);//结合udev或mdev可以实现自动在/dev下创建设备节点

                ……

        }

从上面的注册函数中可以看出V4L2驱动的注册事实上只是完成了设备节点的创建，如：/dev/video0。和video_device结构指针的保存。

（3）视频驱动的打开过程

当用户空间调用open打开对应的视频文件时，如：

int fd = open(/dev/video0, O_RDWR);

对应/dev/video0的文件操作结构是/drivers/media/videodev.c中定义的video_fops。

static const struct file_operations video_fops=

        {

                .owner = THIS_MODULE,

                .llseek = no_llseek,

                .open = video_open,

        };

奇怪吧，这里只实现了open操作。那么后面的其它操作呢？还是先看看video_open吧。

static int video_open(struct inode *inode, struct file *file)

        {

                unsigned int minor = iminor(inode);

                int err = 0;

                struct video_device *vfl;

                const struct file_operations *old_fops;

        if(minor>=VIDEO_NUM_DEVICES)

                        return -ENODEV;

                mutex_lock(&videodev_lock);

                vfl=video_device[minor];

                if(vfl==NULL) {

                        mutex_unlock(&videodev_lock);

                        request_module("char-major-%d-%d", VIDEO_MAJOR, minor);

                        mutex_lock(&videodev_lock);

                        vfl=video_device[minor]; //根据次设备号取出video_device结构

                        if (vfl==NULL) {

                                mutex_unlock(&videodev_lock);

                                return -ENODEV;

                        }

                }

                old_fops = file->f_op;

                file->f_op = fops_get(vfl->fops);//替换此打开文件的file_operation结构。后面的其它针对此文件的操作都由新的结构来负责了。也就是由每个具体的video_device的fops负责。

                if(file->f_op->open)

                        err = file->f_op->open(inode,file);

                if (err) {

                        fops_put(file->f_op);

                        file->f_op = fops_get(old_fops);

                }

                ……

        }

以上是我对V4L2的一些理解，希望能对大家了解V4L2有一些帮助！