Linux设备模型之input子系统详解(second)

 
七:evdev概述
 Evdev对应的设备节点一般位于/dev/input/event0 ~ /dev/input/event4.理论上可以对应32个设备节点.分别代表被handler匹配的32个input device.
可以用cat /dev/input/event0.然后移动鼠标或者键盘按键就会有数据输出(两者之间只能选一.因为一个设备文件只能关能一个输入设备).还可以往这个文件里写数据,使其产生特定的事件.这个过程我们之后再详细分析.
为了分析这一过程,必须从input子系统的初始化说起.
 
八:input子系统的初始化
Input子系统的初始化函数为input_init().代码如下:
static int __init input_init(void)
{
         int err;
 
         err = class_register(&input_class);
         if (err) {
                   printk(KERN_ERR "input: unable to register input_dev class\n");
                   return err;
         }
 
         err = input_proc_init();
         if (err)
                   goto fail1;
 
         err = register_chrdev(INPUT_MAJOR, "input", &input_fops);
         if (err) {
                   printk(KERN_ERR "input: unable to register char major %d", INPUT_MAJOR);
                   goto fail2;
         }
 
         return 0;
 
 fail2:        input_proc_exit();
 fail1:        class_unregister(&input_class);
         return err;
}
在这个初始化函数里,先注册了一个名为”input”的类.所有input device都属于这个类.在sysfs中表现就是.所有input device所代表的目录都位于/dev/class/input下面.
然后调用input_proc_init()在/proc下面建立相关的交互文件.
再后调用register_chrdev()注册了主设备号为INPUT_MAJOR(13).次设备号为0~255的字符设备.它的操作指针为input_fops.
在这里,我们看到.所有主设备号13的字符设备的操作最终都会转入到input_fops中.在前面分析的/dev/input/event0~/dev/input/event4的主设备号为13.操作也不例外的落在了input_fops中.
Input_fops定义如下:
static const struct file_operations input_fops = {
         .owner = THIS_MODULE,
         .open = input_open_file,
};
打开文件所对应的操作函数为input_open_file.代码如下示:
static int input_open_file(struct inode *inode, struct file *file)
{
         struct input_handler *handler = input_table[iminor(inode) >> 5];
         const struct file_operations *old_fops, *new_fops = NULL;
         int err;
 
         /* No load-on-demand here? */
         if (!handler || !(new_fops = fops_get(handler->fops)))
                   return -ENODEV;
 
iminor(inode)为打开文件所对应的次设备号.input_table是一个struct input_handler全局数组.在这里.它先设备结点的次设备号右移5位做为索引值到input_table中取对应项.从这里我们也可以看到.一 个handle代表1<<5个设备节点(因为在input_table中取值是以次备号右移5位为索引的.即低5位相同的次备号对应的是同一 个索引).在这里,终于看到了input_talbe大显身手的地方了.input_talbe[ ]中取值和input_talbe[ ]的赋值,这两个过程是相对应的.
 
在input_table中找到对应的handler之后,就会检验这个handle是否存,是否带有fops文件操作集.如果没有.则返回一个设备不存在的错误.
         /*
          * That's _really_ odd. Usually NULL ->open means "nothing special",
          * not "no device". Oh, well...
          */
         if (!new_fops->open) {
                   fops_put(new_fops);
                   return -ENODEV;
         }
         old_fops = file->f_op;
         file->f_op = new_fops;
 
         err = new_fops->open(inode, file);
 
         if (err) {
                   fops_put(file->f_op);
                   file->f_op = fops_get(old_fops);
         }
         fops_put(old_fops);
         return err;
}
然后将handler中的fops替换掉当前的fops.如果新的fops中有open()函数,则调用它.
 
九:evdev的初始化
Evdev的模块初始化函数为evdev_init().代码如下:
static int __init evdev_init(void)
{
         return input_register_handler(&evdev_handler);
}
它调用了input_register_handler注册了一个handler.
注意到,在这里evdev_handler中定义的minor为EVDEV_MINOR_BASE(64).也就是说evdev_handler所表示的设备文件范围为(13,64)à(13,64+32).
从之前的分析我们知道.匹配成功的关键在于handler中的blacklist和id_talbe. Evdev_handler的id_table定义如下:
static const struct input_device_id evdev_ids[] = {
         { .driver_info = 1 },     /* Matches all devices */
         { },                       /* Terminating zero entry */
};
它没有定义flags.也没有定义匹配属性值.这个handler是匹配所有input device的.从前面的分析我们知道.匹配成功之后会调用handler->connect函数.
在Evdev_handler中,该成员函数如下所示:
 
static int evdev_connect(struct input_handler *handler, struct input_dev *dev,
                             const struct input_device_id *id)
{
         struct evdev *evdev;
         int minor;
         int error;
 
         for (minor = 0; minor < EVDEV_MINORS; minor++)
                   if (!evdev_table[minor])
                            break;
 
         if (minor == EVDEV_MINORS) {
                   printk(KERN_ERR "evdev: no more free evdev devices\n");
                   return -ENFILE;
         }
EVDEV_MINORS定义为32.表示evdev_handler所表示的32个设备文件.evdev_talbe是一个struct evdev类型的数组.struct evdev是模块使用的封装结构.在接下来的代码中我们可以看到这个结构的使用.
这一段代码的在evdev_talbe找到为空的那一项.minor就是数组中第一项为空的序号.
 
         evdev = kzalloc(sizeof(struct evdev), GFP_KERNEL);
         if (!evdev)
                   return -ENOMEM;
 
         INIT_LIST_HEAD(&evdev->client_list);
         spin_lock_init(&evdev->client_lock);
         mutex_init(&evdev->mutex);
         init_waitqueue_head(&evdev->wait);
 
         snprintf(evdev->name, sizeof(evdev->name), "event%d", minor);
         evdev->exist = 1;
         evdev->minor = minor;
 
         evdev->handle.dev = input_get_device(dev);
         evdev->handle.name = evdev->name;
         evdev->handle.handler = handler;
         evdev->handle.private = evdev;
接下来,分配了一个evdev结构,并对这个结构进行初始化.在这里我们可以看到,这个结构封装了一个handle结构,这结构与我们之前所讨 论的handler是不相同的.注意有一个字母的差别哦.我们可以把handle看成是handler和input device的信息集合体.在这个结构里集合了匹配成功的handler和input device
 
         strlcpy(evdev->dev.bus_id, evdev->name, sizeof(evdev->dev.bus_id));
         evdev->dev.devt = MKDEV(INPUT_MAJOR, EVDEV_MINOR_BASE + minor);
         evdev->dev.class = &input_class;
         evdev->dev.parent = &dev->dev;
         evdev->dev.release = evdev_free;
         device_initialize(&evdev->dev);
在这段代码里主要完成evdev封装的device的初始化.注意在这里,使它所属的类指向input_class.这样在/sysfs中创建的设备目录就会在/sys/class/input/下面显示.
 
         error = input_register_handle(&evdev->handle);
         if (error)
                   goto err_free_evdev;
         error = evdev_install_chrdev(evdev);
         if (error)
                   goto err_unregister_handle;
 
         error = device_add(&evdev->dev);
         if (error)
                   goto err_cleanup_evdev;
 
         return 0;
 
 err_cleanup_evdev:
         evdev_cleanup(evdev);
 err_unregister_handle:
         input_unregister_handle(&evdev->handle);
 err_free_evdev:
         put_device(&evdev->dev);
         return error;
}
注册handle,如果是成功的,那么调用evdev_install_chrdev将evdev_table的minor项指向evdev. 然后将evdev->device注册到sysfs.如果失败,将进行相关的错误处理.
万事俱备了,但是要接收事件,还得要等”东风”.这个”东风”就是要打开相应的handle.这个打开过程是在文件的open()中完成的.
 
十:evdev设备结点的open()操作
我们知道.对主设备号为INPUT_MAJOR的设备节点进行操作,会将操作集转换成handler的操作集.在evdev中,这个操作集就是evdev_fops.对应的open函数如下示:
static int evdev_open(struct inode *inode, struct file *file)
{
         struct evdev *evdev;
         struct evdev_client *client;
         int i = iminor(inode) - EVDEV_MINOR_BASE;
         int error;
 
         if (i >= EVDEV_MINORS)
                   return -ENODEV;
 
         error = mutex_lock_interruptible(&evdev_table_mutex);
         if (error)
                   return error;
         evdev = evdev_table[i];
         if (evdev)
                   get_device(&evdev->dev);
         mutex_unlock(&evdev_table_mutex);
 
         if (!evdev)
                   return -ENODEV;
 
         client = kzalloc(sizeof(struct evdev_client), GFP_KERNEL);
         if (!client) {
                   error = -ENOMEM;
                   goto err_put_evdev;
         }
         spin_lock_init(&client->buffer_lock);
         client->evdev = evdev;
         evdev_attach_client(evdev, client);
 
         error = evdev_open_device(evdev);
         if (error)
                   goto err_free_client;
 
         file->private_data = client;
         return 0;
 
 err_free_client:
         evdev_detach_client(evdev, client);
         kfree(client);
 err_put_evdev:
         put_device(&evdev->dev);
         return error;
}
iminor(inode) - EVDEV_MINOR_BASE就得到了在evdev_table[ ]中的序号.然后将数组中对应的evdev取出.递增devdev中device的引用计数.
分配并初始化一个client.并将它和evdev关联起来: client->evdev指向它所表示的evdev. 将client挂到evdev->client_list上. 将client赋为file的私有区.
对应handle的打开是在此evdev_open_device()中完成的.代码如下:
static int evdev_open_device(struct evdev *evdev)
{
         int retval;
 
         retval = mutex_lock_interruptible(&evdev->mutex);
         if (retval)
                   return retval;
 
         if (!evdev->exist)
                   retval = -ENODEV;
         else if (!evdev->open++) {
                   retval = input_open_device(&evdev->handle);
                   if (retval)
                            evdev->open--;
         }
 
         mutex_unlock(&evdev->mutex);
         return retval;
}
如果evdev是第一次打开,就会调用input_open_device()打开evdev对应的handle.跟踪一下这个函数:
int input_open_device(struct input_handle *handle)
{
         struct input_dev *dev = handle->dev;
         int retval;
 
         retval = mutex_lock_interruptible(&dev->mutex);
         if (retval)
                   return retval;
 
         if (dev->going_away) {
                   retval = -ENODEV;
                   goto out;
         }
 
         handle->open++;
 
         if (!dev->users++ && dev->open)
                   retval = dev->open(dev);
 
         if (retval) {
                   dev->users--;
                   if (!--handle->open) {
                            /*
                             * Make sure we are not delivering any more events
                             * through this handle
                             */
                            synchronize_rcu();
                   }
         }
 
 out:
         mutex_unlock(&dev->mutex);
         return retval;
}
在这个函数中,我们看到.递增handle的打开计数.如果是第一次打开.则调用input device的open()函数.
 
十一:evdev的事件处理
经过上面的分析.每当input device上报一个事件时,会将其交给和它匹配的handler的event函数处理.在evdev中.这个event函数对应的代码为:
static void evdev_event(struct input_handle *handle,
                            unsigned int type, unsigned int code, int value)
{
         struct evdev *evdev = handle->private;
         struct evdev_client *client;
         struct input_event event;
 
         do_gettimeofday(&event.time);
         event.type = type;
         event.code = code;
         event.value = value;
 
         rcu_read_lock();
 
         client = rcu_dereference(evdev->grab);
         if (client)
                   evdev_pass_event(client, &event);
         else
                   list_for_each_entry_rcu(client, &evdev->client_list, node)
                            evdev_pass_event(client, &event);
 
         rcu_read_unlock();
 
         wake_up_interruptible(&evdev->wait);
}
首先构造一个struct input_event结构.并设备它的type.code,value为处理事件的相关属性.如果该设备被强制设置了handle.则调用如之对应的client.
我们在open的时候分析到.会初始化clinet并将其链入到evdev->client_list. 这样,就可以通过evdev->client_list找到这个client了.
对于找到的第一个client都会调用evdev_pass_event( ).代码如下:
static void evdev_pass_event(struct evdev_client *client,
                                 struct input_event *event)
{
         /*
          * Interrupts are disabled, just acquire the lock
          */
         spin_lock(&client->buffer_lock);
         client->buffer[client->head++] = *event;
         client->head &= EVDEV_BUFFER_SIZE - 1;
         spin_unlock(&client->buffer_lock);
 
         kill_fasync(&client->fasync, SIGIO, POLL_IN);
}
这里的操作很简单.就是将event保存到client->buffer中.而client->head就是当前的数据位置.注意这里是一个环形缓存区.写数据是从client->head写.而读数据则是从client->tail中读.
 
十二:设备节点的read处理
对于evdev设备节点的read操作都会由evdev_read()完成.它的代码如下:
static ssize_t evdev_read(struct file *file, char __user *buffer,
                              size_t count, loff_t *ppos)
{
         struct evdev_client *client = file->private_data;
         struct evdev *evdev = client->evdev;
         struct input_event event;
         int retval;
 
         if (count < evdev_event_size())
                   return -EINVAL;
 
         if (client->head == client->tail && evdev->exist &&
             (file->f_flags & O_NONBLOCK))
                   return -EAGAIN;
 
         retval = wait_event_interruptible(evdev->wait,
                   client->head != client->tail || !evdev->exist);
         if (retval)
                   return retval;
 
         if (!evdev->exist)
                   return -ENODEV;
 
         while (retval + evdev_event_size() <= count &&
                evdev_fetch_next_event(client, &event)) {
 
                   if (evdev_event_to_user(buffer + retval, &event))
                            return -EFAULT;
 
                   retval += evdev_event_size();
         }
 
         return retval;
}
首先,它判断缓存区大小是否足够.在读取数据的情况下,可能当前缓存区内没有数据可读.在这里先睡眠等待缓存区中有数据.如果在睡眠的时候,.条件满足.是不会进行睡眠状态而直接返回的.
然后根据read()提够的缓存区大小.将client中的数据写入到用户空间的缓存区中.
十三:设备节点的写操作
同样.对设备节点的写操作是由evdev_write()完成的.代码如下:
 
static ssize_t evdev_write(struct file *file, const char __user *buffer,
                               size_t count, loff_t *ppos)
{
         struct evdev_client *client = file->private_data;
         struct evdev *evdev = client->evdev;
         struct input_event event;
         int retval;
 
         retval = mutex_lock_interruptible(&evdev->mutex);
         if (retval)
                   return retval;
 
         if (!evdev->exist) {
                   retval = -ENODEV;
                   goto out;
         }
 
         while (retval < count) {
 
                   if (evdev_event_from_user(buffer + retval, &event)) {
                            retval = -EFAULT;
                            goto out;
                   }
 
                   input_inject_event(&evdev->handle,
                                        event.type, event.code, event.value);
                   retval += evdev_event_size();
         }
 
 out:
         mutex_unlock(&evdev->mutex);
         return retval;
}
首先取得操作设备文件所对应的evdev.
实际上,这里写入设备文件的是一个event结构的数组.我们在之前分析过,这个结构里包含了事件的type.code和event.
将写入设备的event数组取出.然后对每一项调用event_inject_event().
这个函数的操作和input_event()差不多.就是将第一个参数handle转换为输入设备结构.然后这个设备再产生一个事件.
代码如下:
void input_inject_event(struct input_handle *handle,
                            unsigned int type, unsigned int code, int value)
{
         struct input_dev *dev = handle->dev;
         struct input_handle *grab;
         unsigned long flags;
 
         if (is_event_supported(type, dev->evbit, EV_MAX)) {
                   spin_lock_irqsave(&dev->event_lock, flags);
 
                   rcu_read_lock();
                   grab = rcu_dereference(dev->grab);
                   if (!grab || grab == handle)
                            input_handle_event(dev, type, code, value);
                   rcu_read_unlock();
 
                   spin_unlock_irqrestore(&dev->event_lock, flags);
         }
}
我们在这里也可以跟input_event()对比一下,这里设备可以产生任意事件,而不需要和设备所支持的事件类型相匹配.
由此可见.对于写操作而言.就是让与设备文件相关的输入设备产生一个特定的事件.
将上述设备文件的操作过程以图的方式表示如下:
 
Linux设备模型之input子系统详解(xu) - liaowb1234 - liaowb1234的博客
 
十四:小结
在这一节点,分析了整个input子系统的架构,各个环节的流程.最后还以evdev为例.将各个流程贯穿在一起.以加深对input子系统的 理解.由此也可以看出.linux设备驱动采用了分层的模式.从最下层的设备模型到设备,驱动,总线再到input子系统最后到input device.这样的分层结构使得最上层的驱动不必关心下层是怎么实现的.而下层驱动又为多种型号同样功能的驱动提供了一个统一的接口.

 原文地址 http://blog.chinaunix.net/u1/51562/showart_1090628.html
阅读更多
个人分类: linux驱动
想对作者说点什么? 我来说一句

没有更多推荐了,返回首页

关闭
关闭
关闭