Linux设备模型之tty驱动架构分析(转载)

------------------------------------------ 本文系本站原创,欢迎转载! 转载请注明出处:http://ericxiao.cublog.cn/ ------------------------------------------ 一：前言 Tty这个名称源于电传打字节的简称。在linux表示各种终端。终端通常都跟硬件相对应。比如对应于输入设备键盘鼠标。输出设备显示器的控制终端和串口终端.也有对应于不存在设备的pty驱动。在如此众多的终端模型之中，linux是怎么将它们统一建模的呢?这就是我们今天要讨论的问题.   二：tty驱动概貌 Tty架构如下所示： 如上图所示,用户空间主要是通过设备文件同tty_core交互.tty_core根据用空间操作的类型再选择跟line discipline和tty_driver交互.例如设置硬件的ioctl指令就直接交给tty_driver处理。Read和write操作就会交给line discipline处理. Line discipline是线路规程的意思。正如它的名字一样，它表示的是这条终端 ”线程 ”的输入与输出规范设置.主要用来进行输入/输出数据的预处理。处理之后。就会将数据交给tty_driver Tty_driver就是终端对应的驱动了。它将字符转换成终端可以理解的字串.将其传给终端设备。 值得注意的是,这个架构没有为tty_drivero提供read操作。也就是说tty_core  和line discipline都没有办法从tty_driver里直接读终端信息。这是因为tty_driver对就的hardware并不一定是输入数据和输出数据的共同负载者。例如控制终端，输出设备是显示器。输入设备是键盘。基于这样的原理。在line discipline中有一个输入缓存区。并提供了一个名叫receive_buf()的接口函数。对应的终端设备只要调用line discipine的receiver_buf函数，将数据写入到输入缓存区就可以了。 如果一个设备同时是输入设备又是输出设备。那在设备的中断处理中调用receive_buf()将数据写入即可.   三：tty驱动接口分析 具体的tty驱动设计可以参考LDD3。这里只对它的接口实现做一个分析.tty driver的所有操作都包含在tty_driver中。内核即供了一个名叫alloc_tty_driver()来分配这个tty_driver。当然我们也可以在自己的驱动中将它定义成一个静态的结构。对tty_driver进行一些必要的初始化之后，调用tty_register_driver()将其注册. alloc_tty_driver()接口代码如下所示： struct tty_driver *alloc_tty_driver(int lines) {          struct tty_driver *driver;            driver = kzalloc(sizeof(struct tty_driver), GFP_KERNEL);          if (driver) {                    driver->magic = TTY_DRIVER_MAGIC;                    driver->num = lines;                    /* later we'll move allocation of tables here */          }          return driver; } 这个函数只有一个参数。这个参数的含义为line的个数。也即次设备号的个数。注意每个设备文件都会对应一个line. 在这个接口里为tty_driver分配内存，然后将driver->mage.driver->num初始化之后就返回了.   tty_register_driver()用来注册一个tty_driver。代码如下： int tty_register_driver(struct tty_driver *driver) {          int error;          int i;          dev_t dev;          void **p = NULL;            //TTY_DRIVER_INSTALLED:已安装的          if (driver->flags & TTY_DRIVER_INSTALLED)                    return 0;            //TTY_DRIVER_DEVPTS_MEM:使用devpts进行动态内存映射          if (!(driver->flags & TTY_DRIVER_DEVPTS_MEM) && driver->num) {                    p = kzalloc(driver->num * 3 * sizeof(void *), GFP_KERNEL);                    if (!p)                             return -ENOMEM;          }            //注册字符设备号          //如果没有指定driver->major          if (!driver->major) {                    error = alloc_chrdev_region(&dev, driver->minor_start,                                                         driver->num, driver->name);                    if (!error) {                             driver->major = MAJOR(dev);                             driver->minor_start = MINOR(dev);                    }          } else {                    dev = MKDEV(driver->major, driver->minor_start);                    error = register_chrdev_region(dev, driver->num, driver->name);          }          if (error < 0) {                    kfree(p);                   return error;          }            if (p) {                    driver->ttys = (struct tty_struct **)p;                    driver->termios = (struct ktermios **)(p + driver->num);                    driver->termios_locked = (struct ktermios **)                                                                  (p + driver->num * 2);          } else {                    driver->ttys = NULL;                    driver->termios = NULL;                    driver->termios_locked = NULL;          }            //注册字符设备          cdev_init(&driver->cdev, &tty_fops);          driver->cdev.owner = driver->owner;          error = cdev_add(&driver->cdev, dev, driver->num);          if (error) {                    unregister_chrdev_region(dev, driver->num);                    driver->ttys = NULL;                    driver->termios = driver->termios_locked = NULL;                    kfree(p);                    return error;          }            //指定默认的put_char          if (!driver->put_char)                    driver->put_char = tty_default_put_char;            mutex_lock(&tty_mutex);          list_add(&driver->tty_drivers, &tty_drivers);          mutex_unlock(&tty_mutex);            //如果没有指定TTY_DRIVER_DYNAMIC_DEV.即动态设备管理          if (!(driver->flags & TTY_DRIVER_DYNAMIC_DEV)) {                    for (i = 0; i < driver->num; i++)                        tty_register_device(driver, i, NULL);          }          proc_tty_register_driver(driver);          return 0; } 这个函数操作比较简单。就是为tty_driver创建字符设备。然后将字符设备的操作集指定为tty_fops.并且将tty_driver挂载到tty_drivers链表中.其实这个链表的作用跟我们之前分析的input子系统中的input_dev[ ]数组类似。都是以设备号为关键字找到对应的driver. 特别的。如果没有定义TTY_DRIVER_DYNAMIC_DEV.还会在sysfs中创建一个类设备.这样主要是为了udev管理设备. 以流程图的方式将上述操作表示如下：     四：设备文件的操作 设备文件的操作是本节分析的重点。它的主要操作是将各项操作对应到ldsic或者是tty_driver.   4．1：打开tty设备的操作 从注册的过程可以看到，所有的操作都会对应到tty_fops中。Open操作对应的操作接口是tty_open()。代码如下： static int tty_open(struct inode *inode, struct file *filp) {          struct tty_struct *tty;          int noctty, retval;          struct tty_driver *driver;          int index;          dev_t device = inode->i_rdev;          unsigned short saved_flags = filp->f_flags;            nonseekable_open(inode, filp);   retry_open:          //O_NOCTTY 如果路径名指向终端设备，不要把这个设备用作控制终端            //noctty:需不需要更改当前进程的控制终端          noctty = filp->f_flags & O_NOCTTY;          index  = -1;          retval = 0;            mutex_lock(&tty_mutex);            //设备号(5,0) 即/dev/tty.表示当前进程的控制终端          if (device == MKDEV(TTYAUX_MAJOR, 0)) {                    tty = get_current_tty();                    //如果当前进程的控制终端不存在,退出                    if (!tty) {                             mutex_unlock(&tty_mutex);                             return -ENXIO;                    }                      //取得当前进程的tty_driver                    driver = tty->driver;                    index = tty->index;                    filp->f_flags |= O_NONBLOCK; /* Don't let /dev/tty block */                    /* noctty = 1; */                    goto got_driver;          } #ifdef CONFIG_VT          //设备号(4,0).即/dev/tty0:表示当前的控制台          if (device == MKDEV(TTY_MAJOR, 0)) {                    extern struct tty_driver *console_driver;                    driver = console_driver;                    //fg_console: 表示当前的控制台                    index = fg_console;                    noctty = 1;                    goto got_driver;          } #endif          //设备号(5,1).即/dev/console.表示外接的控制台. 通过regesit_console()          if (device == MKDEV(TTYAUX_MAJOR, 1)) {                    driver = console_device(&index);                    if (driver) {                             /* Don't let /dev/console block */                             filp->f_flags |= O_NONBLOCK;                             noctty = 1;                             goto got_driver;                    }                    mutex_unlock(&tty_mutex);                    return -ENODEV;          }            //以文件的设备号为关键字,到tty_drivers中搜索所注册的driver          driver = get_tty_driver(device, &index);          if (!driver) {                    mutex_unlock(&tty_mutex);                    return -ENODEV;          } got_driver:          //index表示它的次设备号          retval = init_dev(driver, index, &tty);          mutex_unlock(&tty_mutex);          if (retval)                    return retval;            filp->private_data = tty;          file_move(filp, &tty->tty_files);          check_tty_count(tty, "tty_open");          if (tty->driver->type == TTY_DRIVER_TYPE_PTY &&              tty->driver->subtype == PTY_TYPE_MASTER)                    noctty = 1; #ifdef TTY_DEBUG_HANGUP          printk(KERN_DEBUG "opening %s...", tty->name); #endif          if (!retval) {                    if (tty->driver->open)                             retval = tty->driver->open(tty, filp);                    else                             retval = -ENODEV;          }          filp->f_flags = saved_flags;            if (!retval && test_bit(TTY_EXCLUSIVE, &tty->flags) &&                                                         !capable(CAP_SYS_ADMIN))                    retval = -EBUSY;            if (retval) { #ifdef TTY_DEBUG_HANGUP                    printk(KERN_DEBUG "error %d in opening %s...", retval,                           tty->name); #endif                    release_dev(filp);                    if (retval != -ERESTARTSYS)                             return retval;                    if (signal_pending(current))                             return retval;                    schedule();                    /*                     * Need to reset f_op in case a hangup happened.                     */                    if (filp->f_op == &hung_up_tty_fops)                             filp->f_op = &tty_fops;                    goto retry_open;          }            mutex_lock(&tty_mutex);          spin_lock_irq(&current->sighand->siglock);            //设置当前进程的终端          if (!noctty &&              current->signal->leader &&              !current->signal->tty &&              tty->session == NULL)                    __proc_set_tty(current, tty);          spin_unlock_irq(&current->sighand->siglock);          mutex_unlock(&tty_mutex);          tty_audit_opening();          return 0; } 注意在这里有个容易忽略的操作：init_dev（）。 Init_dev() -à initialize_tty_struct() à tty_ldisc_assign(tty, tty_ldisc_get(N_TTY)); 看一下tty_ldisc_assign(tty, tty_ldisc_get(N_TTY))的操作: Tty_ldisc_get(): struct tty_ldisc *tty_ldisc_get(int disc) {          unsigned long flags;          struct tty_ldisc *ld;            if (disc < N_TTY || disc >= NR_LDISCS)                    return NULL;            spin_lock_irqsave(&tty_ldisc_lock, flags);            ld = &tty_ldiscs[disc];          /* Check the entry is defined */          if (ld->flags & LDISC_FLAG_DEFINED) {                    /* If the module is being unloaded we can't use it */                    if (!try_module_get(ld->owner))                             ld = NULL;                    else /* lock it */                             ld->refcount++;          } else                    ld = NULL;          spin_unlock_irqrestore(&tty_ldisc_lock, flags);          return ld; } 这个函数的操作为到tty_ldiscs[ ]找到对应项.这个数组中的成员是调用tty_register_ldisc()将其设置进去的.   tty_ldisc_assign操作如下： static void tty_ldisc_assign(struct tty_struct *tty, struct tty_ldisc *ld) {          tty->ldisc = *ld;          tty->ldisc.refcount = 0; } 即将取出来的idisc作为tty->ldisc字段. 在这段代码中涉及到了tty_driver,tty_struct, struct tty_ldisc.这三者之间的关系用下图表示如下：   在这里,为tty_struct的ldisc是默认指定为tty_ldiscs[N_TTY].该ldisc对应的是控制终端的线路规范。可以在用空间用带TIOCSETD的ioctl调用进行更改. 将上述open用流程图的方式表示如下：   4.2:设备文件的write操作 设备文件的write操作对应tty_fops->write即tty_write().代码如下： static ssize_t tty_write(struct file *file, const char __user *buf,                                                         size_t count, loff_t *ppos) {          struct tty_struct *tty;          struct inode *inode = file->f_path.dentry->d_inode;          ssize_t ret;          struct tty_ldisc *ld;            tty = (struct tty_struct *)file->private_data;          if (tty_paranoia_check(tty, inode, "tty_write"))                    return -EIO;          if (!tty || !tty->driver->write ||                    (test_bit(TTY_IO_ERROR, &tty->flags)))                             return -EIO;            ld = tty_ldisc_ref_wait(tty);          if (!ld->write)                    ret = -EIO;          else                    ret = do_tty_write(ld->write, tty, file, buf, count);          tty_ldisc_deref(ld);          return ret; } 在open的过程中，将tty_struct存放在file的私有区。在write中,从file的私有区中就可以取到要操作的tty_struct. 如果tty_driver中没有write.如果tty有错误都会有效性判断失败返回。如果一切正常，递增ldsic的引用计数。将用do_tty_wirte()再行写操作。写完之后，再递减ldsic的引用计数. Do_tty_write代码分段分析如下： static inline ssize_t do_tty_write(          ssize_t (*write)(struct tty_struct *, struct file *, const unsigned char *, size_t),          struct tty_struct *tty,          struct file *file,          const char __user *buf,          size_t count) {          ssize_t ret, written = 0;          unsigned int chunk;            ret = tty_write_lock(tty, file->f_flags & O_NDELAY);          if (ret < 0)                    return ret;            /*           * We chunk up writes into a temporary buffer. This           * simplifies low-level drivers immensely, since they           * don't have locking issues and user mode accesses.           *           * But if TTY_NO_WRITE_SPLIT is set, we should use a           * big chunk-size..           *           * The default chunk-size is 2kB, because the NTTY           * layer has problems with bigger chunks. It will           * claim to be able to handle more characters than           * it actually does.           *           * FIXME: This can probably go away now except that 64K chunks           * are too likely to fail unless switched to vmalloc...           */          chunk = 2048;          if (test_bit(TTY_NO_WRITE_SPLIT, &tty->flags))                    chunk = 65536;          if (count < chunk)                    chunk = count;            /* write_buf/write_cnt is protected by the atomic_write_lock mutex */          if (tty->write_cnt < chunk) {                    unsigned char *buf;                      if (chunk < 1024)                             chunk = 1024;                      buf = kmalloc(chunk, GFP_KERNEL);                    if (!buf) {                             ret = -ENOMEM;                             goto out;                    }                    kfree(tty->write_buf);                    tty->write_cnt = chunk;                    tty->write_buf = buf;          } 默认一次写数据的大小为2K.如果设置了TTY_NO_WRITE_SPLIT.则将一次写的数据量扩大为65536. Tty->write_buf是写操作的临时缓存区。即将用户空的数据暂时存放到这里 Tty->write_cnt是临时缓存区的大小。 在这里，必须要根据一次写的数据量对这个临时缓存区做调整            /* Do the write .. */          for (;;) {                    size_t size = count;                    if (size > chunk)                             size = chunk;                    ret = -EFAULT;                    if (copy_from_user(tty->write_buf, buf, size))                             break;                    lock_kernel();                    ret = write(tty, file, tty->write_buf, size);                    unlock_kernel();                    if (ret <= 0)                             break;                    written += ret;                    buf += ret;                    count -= ret;                    if (!count)                             break;                    ret = -ERESTARTSYS;                    if (signal_pending(current))                             break;                    cond_resched();          }          if (written) {                    struct inode *inode = file->f_path.dentry->d_inode;                    inode->i_mtime = current_fs_time(inode->i_sb);                    ret = written;          } out:          tty_write_unlock(tty);          return ret; } 后面的操作就比较简单了。先将用户空间的数据copy到临时缓存区，然后再调用ldisc->write()完成这次写操作.最后再更新设备结点的时间戳. Write操作的流程图如下示：   在这里，我们只看到将数据写放到了ldisc->write().没有看到与tty_driver相关的部份。实际上在ldisc中对写入的数据做预处理过后，还是会调用tty_driver->write()将其写入硬件.   4.3:设备文件的read操作 static ssize_t tty_read(struct file *file, char __user *buf, size_t count,                             loff_t *ppos) {          int i;          struct tty_struct *tty;          struct inode *inode;          struct tty_ldisc *ld;            tty = (struct tty_struct *)file->private_data;          inode = file->f_path.dentry->d_inode;          if (tty_paranoia_check(tty, inode, "tty_read"))                    return -EIO;          if (!tty || (test_bit(TTY_IO_ERROR, &tty->flags)))                    return -EIO;            /* We want to wait for the line discipline to sort out in this             situation */          ld = tty_ldisc_ref_wait(tty);          lock_kernel();          if (ld->read)                    i = (ld->read)(tty, file, buf, count);          else                    i = -EIO;          tty_ldisc_deref(ld);          unlock_kernel();          if (i > 0)                    inode->i_atime = current_fs_time(inode->i_sb);          return i; } 这个read操作就更简单。直接调用ldsic->read()完成工作 流程图如下：   五：小结 在tty设备文件的操作中。Open操作会进行一系统初始化。然后调用ldsic->open tty_driver->open。在write和read调用中只tty_core只会用到ldisc->wirte/ldisc->read.除了上面分析的几个操作之外，还有一个ioctl操作，以及它封装的几个termios。这些ioctl类的操作会直接和tty_driver相关联. 在这一节里，只对tty的构造做一个分析，具体ldisc的操作我们之后以控制终端为例进行分析.