linux设备驱动之控制台驱动

------------------------------------------ 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 我们在之前分析过input子系统和tty设备驱动架构.今天需要将两者结合起来.看看linux中的控制台是怎么样实现的. 二:控制台驱动的初始化 之前在分析tty驱动架构的时候曾分析到.主设备为4,次设备为0的设备节点,即/dev/tty0为当前的控制终端. 有tty_init()中,有以下代码段: static int __init tty_init(void) {          ……          ……          #ifdef CONFIG_VT          cdev_init(&vc0_cdev, &console_fops);          if (cdev_add(&vc0_cdev, MKDEV(TTY_MAJOR, 0), 1) ||              register_chrdev_region(MKDEV(TTY_MAJOR, 0), 1, "/dev/vc/0") < 0)                    panic("Couldn't register /dev/tty0 driver\n");          device_create(tty_class, NULL, MKDEV(TTY_MAJOR, 0), "tty0");            vty_init(); #endif          return 0; } CONFIG_VT:是指配置虚拟终端.即我们所说的控制台.在此可以看到TTY_MAJOR(4),0对应的设备节点操作集为console_fops. 继续跟进vty_init() int __init vty_init(void) {          vcs_init();            console_driver = alloc_tty_driver(MAX_NR_CONSOLES);          if (!console_driver)                    panic("Couldn't allocate console driver\n");          console_driver->owner = THIS_MODULE;          console_driver->name = "tty";          console_driver->name_base = 1;          console_driver->major = TTY_MAJOR;          console_driver->minor_start = 1;          console_driver->type = TTY_DRIVER_TYPE_CONSOLE;          console_driver->init_termios = tty_std_termios;          console_driver->flags = TTY_DRIVER_REAL_RAW | TTY_DRIVER_RESET_TERMIOS;          tty_set_operations(console_driver, &con_ops);          if (tty_register_driver(console_driver))                    panic("Couldn't register console driver\n");            kbd_init();          console_map_init(); #ifdef CONFIG_PROM_CONSOLE          prom_con_init(); #endif #ifdef CONFIG_MDA_CONSOLE          mda_console_init(); #endif          return 0; } 经过我们之前的tty驱动架构分析,这段代码看起来就比较简单了,它就是注册了一个tty驱动.这个驱动对应的操作集是位于con_ops里面的. 仔细看.在之后还会调用kbd_init().顾名思义,这个是一个有关键盘的初始化.控制终端跟键盘有什么关系呢?在之前分析tty的时候,曾提到过,.对于控制台而言,它的输入设备是键盘鼠标,它的输出设备是当前显示器.这两者是怎么关联起来的呢?不着急.请看下面的分析.   三:控制台的open操作 在前面分析了,对应console的操作集为con_ops.定义如下: static const struct file_operations console_fops = {          .llseek                = no_llseek,          .read                   = tty_read,          .write                  = redirected_tty_write,          .poll           = tty_poll,          .ioctl          = tty_ioctl,          .compat_ioctl    = tty_compat_ioctl,          .open                  = tty_open,          .release    = tty_release,          .fasync               = tty_fasync, }; 里面的函数指针值我们都不陌生了,在之前分析的tty驱动中已经分析过了. 结合前面的tty驱动分析.我们知道在open的时候,会调用ldisc的open和tty_driver.open. 对于ldisc默认是tty_ldiscs[0].我们来看下它的具体赋值. console_init(): void __init console_init(void) {          initcall_t *call;            /* Setup the default TTY line discipline. */          (void) tty_register_ldisc(N_TTY, &tty_ldisc_N_TTY);            /*           * set up the console device so that later boot sequences can           * inform about problems etc..           */          call = __con_initcall_start;          while (call < __con_initcall_end) {                    (*call)();                    call++;          } } 在这里,通过tty_register_ldisc.将tty_ldisc_N_TTY注册为了第N_TTY项.即第1项. tty_ldisc_N_TTY定义如下: struct tty_ldisc tty_ldisc_N_TTY = {          .magic           = TTY_LDISC_MAGIC,          .name            = "n_tty",          .open            = n_tty_open,          .close           = n_tty_close,          .flush_buffer    = n_tty_flush_buffer,          .chars_in_buffer = n_tty_chars_in_buffer,          .read            = read_chan,          .write           = write_chan,          .ioctl           = n_tty_ioctl,          .set_termios     = n_tty_set_termios,          .poll            = normal_poll,          .receive_buf     = n_tty_receive_buf,          .write_wakeup    = n_tty_write_wakeup } 对应的open操作为n_tty_open: static int n_tty_open(struct tty_struct *tty) {          if (!tty)                    return -EINVAL;            /* This one is ugly. Currently a malloc failure here can panic */          if (!tty->read_buf) {                    tty->read_buf = alloc_buf();                    if (!tty->read_buf)                             return -ENOMEM;          }          memset(tty->read_buf, 0, N_TTY_BUF_SIZE);          reset_._flags(tty);          tty->column = 0;          n_tty_set_termios(tty, NULL);          tty->minimum_to_wake = 1;          tty->closing = 0;          return 0; } 它为tty->read_buf分配内存.这个buffer空间大小为N_TTY_BUF_SIZE.read_buf实际上就是从按键的缓存区.然后调用reset_flags()来初始化tty中的一些字段: static void reset_buffer_flags(struct tty_struct *tty) {          unsigned long flags;            spin_lock_irqsave(&tty->read_lock, flags);          tty->read_head = tty->read_tail = tty->read_cnt = 0;          spin_unlock_irqrestore(&tty->read_lock, flags);          tty->canon_head = tty->canon_data = tty->erasing = 0;          memset(&tty->read_flags, 0, sizeof tty->read_flags);          n_tty_set_room(tty);          check_unthrottle(tty); } 这里比较简,不再详细分析.在这里要注意几个tty成员的含义: Tty->read_head, tty->read_tail , tty->read_cnt分别代表read_buf中数据的写入位置,读取位置和数据总数.read_buf是一个环形缓存区. n_tty_set_room()是设备read_buf中的可用缓存区 check_unthrottle():是用来判断是否需要打开”阀门”,允许输入数据流入   对于console tty_driver对应的open函数如下示: static int con_open(struct tty_struct *tty, struct file *filp) {          unsigned int currcons = tty->index;          int ret = 0;            acquire_console_sem();          if (tty->driver_data == NULL) {                    ret = vc_allocate(currcons);                    if (ret == 0) {                             struct vc_data *vc = vc_cons[currcons].d;                             tty->driver_data = vc;                             vc->vc_tty = tty;                               if (!tty->winsize.ws_row && !tty->winsize.ws_col) {                                      tty->winsize.ws_row = vc_cons[currcons].d->vc_rows;                                      tty->winsize.ws_col = vc_cons[currcons].d->vc_cols;                             }                             release_console_sem();                             vcs_make_sysfs(tty);                             return ret;                    }          }          release_console_sem();          return ret; } tty->index表示的是tty_driver所对示的设备节点序号.在这里也就是控制台的序列.用alt+fn就可以切换控制终端. 在这里,它主要为vc_cons[ ]数组中的对应项赋值.并将tty和vc建立关联.   四:控制台的read操作 从tty驱动架构中分析可得到,最终的read操作会转入到ldsic->read中进行. 相应tty_ldisc_N_TTY的read操作如下.这个函数代码较长,分段分析如下: static ssize_t read_chan(struct tty_struct *tty, struct file *file,                              unsigned char __user *buf, size_t nr) {          unsigned char __user *b = buf;          DECLARE_WAITQUEUE(wait, current);          int c;          int minimum, time;          ssize_t retval = 0;          ssize_t size;          long timeout;          unsigned long flags;   do_it_again:            if (!tty->read_buf) {                    printk(KERN_ERR "n_tty_read_chan: read_buf == NULL?!?\n");                    return -EIO;          }            c = job_control(tty, file);          if (c < 0)                    return c;            minimum = time = 0;          timeout = MAX_SCHEDULE_TIMEOUT;                   if (!tty->icanon) {                    time = (HZ / 10) * TIME_CHAR(tty);                    minimum = MIN_CHAR(tty);                      if (minimum) {                             if (time)                                      tty->minimum_to_wake = 1;                             else if (!waitqueue_active(&tty->read_wait) ||                                       (tty->minimum_to_wake > minimum))                                      tty->minimum_to_wake = minimum;                    } else {                             timeout = 0;                             if (time) {                                      timeout = time;                                      time = 0;                             }                             tty->minimum_to_wake = minimum = 1;                    }          } 首先,检查read操作的合法性,read_buf是否已经建立.然后再根据操作的类型来设置tty-> minimum_to_wake.这个成员的含义即为: 如果读进程在因数据不足而睡眠的情况下,数据到达并超过了minimum_to_wake.就将这个读进程唤醒.具体的唤醒过程我们在遇到的时候再进行分析.            /*           *      Internal serialization of reads.           */           //不允许阻塞          if (file->f_flags & O_NONBLOCK) {                    if (!mutex_trylock(&tty->atomic_read_lock))                             return -EAGAIN;          } else {                    if (mutex_lock_interruptible(&tty->atomic_read_lock))                             return -ERESTARTSYS;          }            add_wait_queue(&tty->read_wait, &wait); 在不允许睡眠的情况下,调用mutex_trylock()去获得锁.如果锁被占用,马上返回.否则用可中断的方式去获取锁,如果取锁错误,返回失败.如果取锁成功,将进程加至等待队列.在没有数据可读的情况下,直接睡眠.如果有数据可读,将其移出等待队列即可.            while (nr) {                    /* First test for status change. */                    if (tty->packet && tty->link->ctrl_status) {                             unsigned char cs;                             if (b != buf)                                      break;                             cs = tty->link->ctrl_status;                             tty->link->ctrl_status = 0;                             if (tty_put_user(tty, cs, b++)) {                                      retval = -EFAULT;                                      b--;                                      break;                             }                             nr--;                             break;                    } 接下来就是一个漫长的while循环,用来读取数据,一直到数据取满为止.如果tty->packet被置为1.即为信包模式,通常用在伪终端设备.如果tty->link->ctrl_status有数据.则说明如果链路状态发生改变,需要提交此信息.在这种情况下,将其直接copy到用户空间即可.                      /* This statement must be first before checking for input                       so that any interrupt will set the state back to                       TASK_RUNNING. */                    set_current_state(TASK_INTERRUPTIBLE);                      if (((minimum - (b - buf)) < tty->minimum_to_wake) &&                        ((minimum - (b - buf)) >= 1))                             tty->minimum_to_wake = (minimum - (b - buf));                      if (!input_available_p(tty, 0)) {                                        if (test_bit(TTY_OTHER_CLOSED, &tty->flags)) {                                      retval = -EIO;                                      break;                             }                             if (tty_hung_up_p(file))                                      break;                             if (!timeout)                                      break;                             if (file->f_flags & O_NONBLOCK) {                                      retval = -EAGAIN;                                      break;                             }                             if (signal_pending(current)) {                                      retval = -ERESTARTSYS;                                      break;                             }                             n_tty_set_room(tty);                             timeout = schedule_timeout(timeout);                             continue;                    }                    __set_current_state(TASK_RUNNING);                    先将进程设为TASK_INTERRUPTIBLE状态.再调用input_available_p()来判断可数据供读取.如果没有.则进程睡眠.如果有数据,则将进程状态设为TASK_RUNNING.在终端接收数据的处理过程中,有两种方式,一种是规范模式.一种是原始模式.在规范模式下,终端需要对数据里面的一些特殊字符做处理.在原始模式下.终端不会对接收到的数据做任何的处理.在这里input_available_p()在判断是否有数据可读也分两种情况进行,对于规范模式,看是否有已经转换好的数据,对于原始模式,判断接收的信息总数                      /* Deal with packet mode. */                    //packet模式`忽略                    if (tty->packet && b == buf) {                             if (tty_put_user(tty, TIOCPKT_DATA, b++)) {                                      retval = -EFAULT;                                      b--;                                      break;                             }                             nr--;                    }                      if (tty->icanon) {                             /* N.B. avoid overrun if nr == 0 */                             while (nr && tty->read_cnt) {                                      int eol;                                      eol = test_and_clear_bit(tty->read_tail,                                                         tty->read_flags);                                      c = tty->read_buf[tty->read_tail];                                      spin_lock_irqsave(&tty->read_lock, flags);                                      tty->read_tail = ((tty->read_tail+1) &                                                           (N_TTY_BUF_SIZE-1));                                      tty->read_cnt--;                                      if (eol) {                                                /* this test should be redundant:                                                 * we shouldn't be reading data if                                                 * canon_data is 0                                                 */                                                if (--tty->canon_data < 0)                                                         tty->canon_data = 0;                                      }                                      spin_unlock_irqrestore(&tty->read_lock, flags);                                        //如果没有到结束字符,将字符copy到数据空间                                      //__DISABLED_CHAR是不需要copy到用户空间的                                      if (!eol || (c != __DISABLED_CHAR)) {                                                if (tty_put_user(tty, c, b++)) {                                                         retval = -EFAULT;                                                         b--;                                                         break;                                                }                                                nr--;                                      }                                      if (eol) {                                                //如果遇到行结束符.就可以退出了                                                tty_audit_push(tty);                                                break;                                      }                             }                             if (retval)                                      break;                    } else {                             //非加工模式,直接copy                             int uncopied;                             //环形缓存,copy两次                             uncopied = copy_from_read_buf(tty, &b, &nr);                             uncopied += copy_from_read_buf(tty, &b, &nr);                             if (uncopied) {                                      retval = -EFAULT;                                      break;                             }                    } 对于规范模式,要读满一行才会返回用户空间.例如我们在shell上输入指令的时候,要按下enter键指令才会进行处理.在tty->read_flags数组中定义了一些满行的标志,如果read_buf中对应的数据在tty->read_flags中被置位.就会认为这次读入已经到结尾了.在这里还要注意的是,不要将__DISABLED_CHAR即’/0’拷贝到用户空间. 对于原始模式,只需要将read_buf中的数据读入到用户空间就可以返回了.在这里需要注意read_buf是一个环形缓存,需要copy两次.例如tail在head之前的情况.                      /* If there is enough space in the read buffer now, let the                     * low-level driver know. We use n_tty_chars_in_buffer() to                     * check the buffer, as it now knows about canonical mode.                     * Otherwise, if the driver is throttled and the line is                     * longer than TTY_THRESHOLD_UNTHROTTLE in canonical mode,                     * we won't get any more characters.                     */                    if (n_tty_chars_in_buffer(tty) <= TTY_THRESHOLD_UNTHROTTLE) {                             n_tty_set_room(tty);                             check_unthrottle(tty);                    } OK.到这里,read_buf中或多或少已经有数据被取出了.如果当前的数据量少于TTY_THRESHOLD_UNTHROTTLE.就可以调用check_unthrottle()将其它的写进程唤醒了                      if (b - buf >= minimum)                             break;                    if (time)                             timeout = time;          }            mutex_unlock(&tty->atomic_read_lock);          remove_wait_queue(&tty->read_wait, &wait);            if (!waitqueue_active(&tty->read_wait))                    tty->minimum_to_wake = minimum;            __set_current_state(TASK_RUNNING);   已经读完了数据,是该到清理的时候了.将进程移出等待队列,并当进程状态设为TASK_RUNNING            size = b - buf;          if (size) {                    retval = size;                    if (nr)                             clear_bit(TTY_PUSH, &tty->flags);          } else if (test_and_clear_bit(TTY_PUSH, &tty->flags))                     goto do_it_again;            //更新剩余空间数          n_tty_set_room(tty);            return retval; } TTY_PUSH:是由底层驱动程序在读到一个EOF字符并将其放入缓存区造成的,表示用户要尽快将缓存区数据取走. 如果本次操作没有读取任何数据,且被设置了TTY_PUSH,则跳转到do_it_again,继续执行.如果本次操作读取了数据,可以等到下一次read的时候再来取. 最后,更新read_buf的剩余空间数.   五:控制终端数据的来源 从这个函数里面我们可以看到,数据是从read_buf中取出来的,但是谁将数据放入到read_buf中的呢?为了探究出它的根源.我们还得要从vty_init()说起. 在之前分析过. vty_init()会调用一个表面字义看起来与键盘相关的一个子函数: kbd_init().跟踪这个函数: int __init kbd_init(void) {          int i;          int error;           for (i = 0; i < MAX_NR_CONSOLES; i++) {                    kbd_table[i].ledflagstate = KBD_DEFLEDS;                    kbd_table[i].default_ledflagstate = KBD_DEFLEDS;                    kbd_table[i].ledmode = LED_SHOW_FLAGS;                    kbd_table[i].lockstate = KBD_DEFLOCK;                    kbd_table[i].slockstate = 0;                    kbd_table[i].modeflags = KBD_DEFMODE;                    kbd_table[i].kbdmode = default_utf8 ? VC_UNICODE : VC_XLATE;          }            error = input_register_handler(&kbd_handler);          if (error)                    return error;            tasklet_enable(&keyboard_tasklet);          tasklet_schedule(&keyboard_tasklet);            return 0; } 暂时用不到的部份我们先不与分析。 在这里注册了一个input handler。结合前面我们分析的input子系统，在handler里会处理input device上报的事件。跟进这个handler看一下： kbd_handler定义如下： static struct input_handler kbd_handler = {          .event                 = kbd_event,          .connect   = kbd_connect,          .disconnect       = kbd_disconnect,          .start                   = kbd_start,          .name                = "kbd",          .id_table   = kbd_ids, }; Id_table是用来匹配input device的。跟进去看一下，看哪些device的事件，才会交给它处理： static const struct input_device_id kbd_ids[] = {          {                 .flags = INPUT_DEVICE_ID_MATCH_EVBIT,                 .evbit = { BIT_MASK(EV_KEY) },         },            {                 .flags = INPUT_DEVICE_ID_MATCH_EVBIT,                 .evbit = { BIT_MASK(EV_SND) },         },            { },    /* Terminating entry */ };   从这个id_table中看来，只要是能支持EV_KEY或者是EV_SND的设备都会被这个hnadler匹配到。相应的。也就能够处理input device上报的事件了. 根据之前的input子系统分析，在input device和handler 进行匹配的时候会调用handler->connect.即kbd_connect().代码如下： static int kbd_connect(struct input_handler *handler, struct input_dev *dev,                             const struct input_device_id *id) {          struct input_handle *handle;          int error;          int i;            for (i = KEY_RESERVED; i < BTN_MISC; i++)                    if (test_bit(i, dev->keybit))                             break;            if (i == BTN_MISC && !test_bit(EV_SND, dev->evbit))                    return -ENODEV;            handle = kzalloc(sizeof(struct input_handle), GFP_KERNEL);          if (!handle)                    return -ENOMEM;            handle->dev = dev;          handle->handler = handler;          handle->name = "kbd";            error = input_register_handle(handle);          if (error)                    goto err_free_handle;            error = input_open_device(handle);          if (error)                    goto err_unregister_handle;            return 0;    err_unregister_handle:          input_unregister_handle(handle);  err_free_handle:          kfree(handle);          return error; } 在这段代码里，它申请分初始化了一个hande结构，并将其注册。Open。这些都是我们之前分析过的东东。在注册handle的时候。又会调用到hande->start.函数如下： static void kbd_start(struct input_handle *handle) {          unsigned char leds = ledstate;            tasklet_disable(&keyboard_tasklet);          if (leds != 0xff) {                    input_inject_event(handle, EV_LED, LED_SCROLLL, !!(leds & 0x01));                    input_inject_event(handle, EV_LED, LED_NUML,    !!(leds & 0x02));                    input_inject_event(handle, EV_LED, LED_CAPSL,   !!(leds & 0x04));                    input_inject_event(handle, EV_SYN, SYN_REPORT, 0);          }          tasklet_enable(&keyboard_tasklet); } 这里就是对键盘上的LED进行操作。启用了tasklent。这些都不是我们所关心的重点。 来看下它的事件处理过程： static void kbd_event(struct input_handle *handle, unsigned int event_type,                          unsigned int event_code, int value) {          if (event_type == EV_MSC && event_code == MSC_RAW && HW_RAW(handle->dev))                    kbd_rawcode(value);          if (event_type == EV_KEY)                    kbd_keycode(event_code, value, HW_RAW(handle->dev));          tasklet_schedule(&keyboard_tasklet);          do_poke_blanked_console = 1;          schedule_console_callback(); } 不管对应键盘的那一种模式。后面的数据流程都会转入到input_queue()进等处理。 实际上。控制终端由vc_cons[ ]数组表示。数组中的每一个项都表示一个控制终端。由全局变量fg_console来指示当前所用的cosole/另外。对于键盘等输出设备也对应一个数组。即kbd_table[ ].用来表示当前终端的控制信息. 其余的都不是我们想关心的。来跟踪一下这个函数的实现： static void put_queue(struct vc_data *vc, int ch) {          struct tty_struct *tty = vc->vc_tty;            if (tty) {                    tty_insert_flip_char(tty, ch, 0);                    con_schedule_flip(tty);          } } 这里的参数vc就是指的在vc_cons[ ]中的当前项。回忆在console open的时候。初始化了这一项。并建立了VC和tty的关联。就这样。在vc中可以寻着关联关系找到tty了. Tty_insert_filp_char( )将数据ch存入tty的一个缓存中，具体代码如下示： static inline int tty_insert_flip_char(struct tty_struct *tty,                                                unsigned char ch, char flag) {          struct tty_buffer *tb = tty->buf.tail;          if (tb && tb->used < tb->size) {                    tb->flag_buf_ptr[tb->used] = flag;                    tb->char_buf_ptr[tb->used++] = ch;                    return 1;          }          return tty_insert_flip_string_flags(tty, &ch, &flag, 1); } 在这里，将数存先存进了tty->buf中。后面的tty_insert_flip_string_flags是在当前buf不够的情况下，扩张buf使用的。代码比较简单，请自行分析。   将数据暂存之后，会调用con_schedule_flip(tty)去唤醒一个软中断的工作队列.代码如下： static inline void con_schedule_flip(struct tty_struct *t) {          unsigned long flags;          spin_lock_irqsave(&t->buf.lock, flags);          if (t->buf.tail != NULL)                    t->buf.tail->commit = t->buf.tail->used;          spin_unlock_irqrestore(&t->buf.lock, flags);          schedule_delayed_work(&t->buf.work, 0); } 对应的工作队列为t->buf.work.这个工作队列是怎么定义的呢？这就要回到我们之前分析的tty驱动的tty_struct的初始化. 代码片段如下所示： static void initialize_tty_struct(struct tty_struct *tty) {          。。。。。。          。。。。。。。          INIT_DELAYED_WORK(&tty->buf.work, flush_to_ldisc);          。。。。。。 } 这就是这个工作队列的定义了. 在这里，特别提醒一下。在上面的put_queue()处理是处于一个中断环境。回想一想整个事件的流程。是键盘中断àinput device上报事件àhandler处理这个事件àput_queue() 在中断中，将工作队列唤醒。将比较繁重的工作交由这个工作队列处理。虽然工作队列也是工作在中断状态。但它是开中断执行的.这也就是软中断存在的目的.   跟进flush_to_ldisc(): static void flush_to_ldisc(struct work_struct *work) {          struct tty_struct *tty =                    container_of(work, struct tty_struct, buf.work.work);          unsigned long          flags;          struct tty_ldisc *disc;          struct tty_buffer *tbuf, *head;          char *char_buf;          unsigned char *flag_buf;            disc = tty_ldisc_ref(tty);          if (disc == NULL)       /*  !TTY_LDISC */                    return; 工作队列所调用的参数是它本身所表示的work_queue.而它本身又是封装在tty_strcut里面的。调用container_of()宏就可以获取到封装它的tty_struct.然后增加tty->ldisc的引用计数            spin_lock_irqsave(&tty->buf.lock, flags);          /* So we know a flush is running */          set_bit(TTY_FLUSHING, &tty->flags);          head = tty->buf.head;          if (head != NULL) {                    tty->buf.head = NULL;                    for (;;) {                             int count = head->commit - head->read;                             if (!count) {                                      if (head->next == NULL)                                                break;                                      tbuf = head;                                      head = head->next;                                      tty_buffer_free(tty, tbuf);                                      continue;                             }                             /* Ldisc or user is trying to flush the buffers                                we are feeding to the ldisc, stop feeding the                                line discipline as we want to empty the queue */                             if (test_bit(TTY_FLUSHPENDING, &tty->flags))                                      break;                             if (!tty->receive_room) {                                      schedule_delayed_work(&tty->buf.work, 1);                                      break;                             }                             if (count > tty->receive_room)                                      count = tty->receive_room;                             char_buf = head->char_buf_ptr + head->read;                             flag_buf = head->flag_buf_ptr + head->read;                             head->read += count;                             spin_unlock_irqrestore(&tty->buf.lock, flags);                             disc->receive_buf(tty, char_buf, flag_buf, count);                             spin_lock_irqsave(&tty->buf.lock, flags);                    }                    /* Restore the queue head */                    tty->buf.head = head;          } 对于tty->buf中的每个缓存区，如果缓存区中没有数据，则将其释放，这个释放是有优化的。如果数据少于512就将其放到tty->buf->free中。下次要放分存放空间的时候可以直接到这里面取。如果设置了TTY_ FLUSHPENDING就会跳出循环。 如果tty的接收缓存区不够，则跳出循环，定时器到达过后再来调用这个工作队列. 最后调用tty->receive_buf()来处理这个数据了.            /* We may have a deferred request to flush the input buffer,             if so pull the chain under the lock and empty the queue */          if (test_bit(TTY_FLUSHPENDING, &tty->flags)) {                    __tty_buffer_flush(tty);                    clear_bit(TTY_FLUSHPENDING, &tty->flags);                    wake_up(&tty->read_wait);          }          clear_bit(TTY_FLUSHING, &tty->flags);          spin_unlock_irqrestore(&tty->buf.lock, flags);            tty_ldisc_deref(disc); } 数据最终会通过tty-> receive_buf()将数据放入read_buf. 在这段代码中，有几个很有意思的处理。在进入工作队列的时候，首先会置TTY_FLUSHING标志.如果有进程在读read_buf的时候，如果此标志被置位，就会设置TTY_FLUSHPENDING标志，并进行睡眠。在数据处理完成之后，判断是否有TTY_FLUSHPENDING标志。如果有，则将读进程唤醒.并清除TTY_FLUSHPENDING和TTY_FLUSHING 想一想。为什么会这么处理呢？为什么这里需要两个缓存区，一个buf.一个read_buf。为什么要这样麻烦呢？ 首先，对于缓存区的数目问题：我们在后面会看到。对接收数据还有一系列的预处理过程，这些过程是比较费时的。不宜在中断中进行费时的操作。所以需要选用软中断机制。这就需要将数据先放置一个buf.再由软中断进行预处理之后，再将它放入到read_buf.这就是两个缓存区的原因. 另外：在存数据到read_buf的时候。会有进程从read_buf中读数据。这样就会造成一个竞争。注意到在软中断情况下是不可睡眠的。我们只能选用自旋锁一类的机制。而这种机制是禁止中断和抢占的。这又违背了软中断机制的初衷。怎么办呢？这就是这样标志的作用了。在设计中，我们必须首先得要保证软中断处理机制的快速完成。所以一进入软中断，就置了一个标志。如果有进程来读数据了，也就是说竞争条件发生了，先将读进程置睡眠。不管怎样，先让软中断处理完之后再说。软中断的工作over这后，再唤醒读进程。 我们之前讲的一系统加锁机制是在两者同样平等的情况。而原子置位与判断置位一般是为了保证一方的工作先完成。   好了，到这一步，我们终于看到跟踪read_buf中数据来源问题的一丝曙光了。数据经过tty->receive_buf之后，这个过程就清晰明朗了。 对于tty_ldisc_N_TTY. receive_buf接口如下所示： static void n_tty_receive_buf(struct tty_struct *tty, const unsigned char *cp,                                   char *fp, int count) {          const unsigned char *p;          char *f, flags = TTY_NORMAL;          int     i;          char buf[64];          unsigned long cpuflags;            if (!tty->read_buf)                    return;            if (tty->real_raw) {                    spin_lock_irqsave(&tty->read_lock, cpuflags);                    i = min(N_TTY_BUF_SIZE - tty->read_cnt,                             N_TTY_BUF_SIZE - tty->read_head);                    i = min(count, i);                    memcpy(tty->read_buf + tty->read_head, cp, i);                    tty->read_head = (tty->read_head + i) & (N_TTY_BUF_SIZE-1);                    tty->read_cnt += i;                    cp += i;                    count -= i;                      i = min(N_TTY_BUF_SIZE - tty->read_cnt,                             N_TTY_BUF_SIZE - tty->read_head);                    i = min(count, i);                    memcpy(tty->read_buf + tty->read_head, cp, i);                    tty->read_head = (tty->read_head + i) & (N_TTY_BUF_SIZE-1);                    tty->read_cnt += i;                    spin_unlock_irqrestore(&tty->read_lock, cpuflags);          } else {                    for (i = count, p = cp, f = fp; i; i--, p++) {                             if (f)                                      flags = *f++;                             switch (flags) {                             case TTY_NORMAL:                                      n_tty_receive_char(tty, *p);                                      break;                             case TTY_BREAK:                                      n_tty_receive_break(tty);                                      break;                             case TTY_PARITY:                             case TTY_FRAME:                                      n_tty_receive_parity_error(tty, *p);                                      break;                             case TTY_OVERRUN:                                      n_tty_receive_overrun(tty);                                      break;                             default:                                      printk(KERN_ERR "%s: unknown flag %d\n",                                             tty_name(tty, buf), flags);                                      break;                             }                    }                    if (tty->driver->flush_chars)                             tty->driver->flush_chars(tty);          } 对于原始模式。直接将数据copy到read_buf中。对于加工模式，将数据预处理之后，再加入到read_buf中。这个预处理过程比较繁杂，这里先忽略.          n_tty_set_room(tty);            if (!tty->icanon && (tty->read_cnt >= tty->minimum_to_wake)) {                    kill_fasync(&tty->fasync, SIGIO, POLL_IN);                    if (waitqueue_active(&tty->read_wait))                             wake_up_interruptible(&tty->read_wait);          }            /*           * Check the remaining room for the input canonicalization           * mode.  We don't want to throttle the driver if we're in           * canonical mode and don't have a newline yet!           */          if (tty->receive_room < TTY_THRESHOLD_THROTTLE) {                    /* check TTY_THROTTLED first so it indicates our state */                    if (!test_and_set_bit(TTY_THROTTLED, &tty->flags) &&                        tty->driver->throttle)                             tty->driver->throttle(tty);          } } 重新计数read_buf的剩余空间量。如果可读数据大于tty->minimum_to_wake.就将它的读进程唤醒。 如果当前read_buf剩余空间不足TTY_THRESHOLD_THROTTLE.就调用tty->driver->throttle(tty)将数程流入进程先阻塞.   六：控制终端的write操作 在输入shell指令的时候，屏幕上会出现我们键入的字符。在输入密码的时候，屏幕上一般不会显示我们当前按入了什么键。就就是终端的两种模式，回显和非回显（ECHO）。当设置为回显模式的时候，会将键入的值在屏幕上面显示出来。这个显示的过程就是通过tty driver->write来实现的。 屏幕上的显示操作跟显示驱动有很重要的联系。一般就是调用显卡驱动的显示接口来实现。在切换终端的时候。设置显示区域。由于这部份跟显卡驱动关联较深，而功能又比较单一。在这里不做详细分析。   七:总结 在这一节里，将之前分析过的input子系统，tty驱动架构联系在了一起。我们渐渐体会到，Linux中大量的使用分层架构。层与层之前的联系很紧密而维护也很简单。深入体会其中的架构思想。对于我们平时做开发是很有裨益的.