contiki-事件调度

　　事件驱动机制广泛应用于嵌入式系统，类似于中断机制，当有事件到来时（比如按键、数据到达），系统响应并处理该事件。相对于轮询机制，事件机制优势很明星，低功耗（系统处于休眠状态，当有事件到达时才被唤醒）和MCU利用率高。

　　Contiki将事件机制融入Protothreads机制，每个事件绑定一个进程（广播事件例外），进程间的消息也是通过事件来传递的。用无符号字符型来标示事件。事件结构体event_data定义如下：

struct event_data {
  process_event_t ev;
  process_data_t data;
  struct process *p;
};
typedef unsigned char process_event_t;
typedef void *        process_data_t;

　　用无符号字符型标识一个事件，Contiki定义了10个事件（0x80~0x8A），其它的供用户使用。

#define PROCESS_EVENT_NONE            0x80
#define PROCESS_EVENT_INIT            0x81
#define PROCESS_EVENT_POLL            0x82
#define PROCESS_EVENT_EXIT            0x83
#define PROCESS_EVENT_SERVICE_REMOVED 0x84
#define PROCESS_EVENT_CONTINUE        0x85
#define PROCESS_EVENT_MSG             0x86
#define PROCESS_EVENT_EXITED          0x87
#define PROCESS_EVENT_TIMER           0x88
#define PROCESS_EVENT_COM             0x89
#define PROCESS_EVENT_MAX             0x8a

　　每个事件绑定一个进程，如果p为NULL，表示该事件绑定所有进程（即广播事件PROCESS_BROADCAST）。除此之外，事件可以携带数据data，可以用着电进行进程间的通信（向另一进程传递带数据的事件）。

static process_num_events_t nevents, fevent;
static struct event_data events[PROCESS_CONF_NUMEVENTS];

#define PROCESS_CONF_NUMEVENTS 32

　　Contiki用一个全局的静态数组存放事件，这意味着事件数目在系统运行之前就要指定（用户可以通过PROCESS_CONF_NUMEVENTS自选配置大小），通过数组下标可以快速访问事件。系统还定义另两个全局静态变量nevents和fevent，分别用于记录未处理事件总数及下一个待处理的位置。事件逻辑组成环形队列，存储在数组里。如下图：

　　可见，对于Contiki系统而言，事件并没有优先级之分，而是先到先服务的策略，全局变量fevent记录了下一次待处理事件的下标。

事件产生

　　Contiki有两种方式产生事件，即同步和异步。同步事件通过process_post_synch函数产生，事件触发后直接处理（调用call_process函数）。而异步事件产生是由process_post产生，并没有及时处理，而是放入事件队列等待处理，process_post流程图如下：

int process_post(struct process *p, process_event_t ev, process_data_t data)
{
  static process_num_events_t snum;

  if(PROCESS_CURRENT() == NULL) {
    PRINTF("process_post: NULL process posts event %d to process '%s', nevents %d\n",
       ev,PROCESS_NAME_STRING(p), nevents);
  } 
  else {
    PRINTF("process_post: Process '%s' posts event %d to process '%s', nevents %d\n",
       PROCESS_NAME_STRING(PROCESS_CURRENT()), ev,
       p == PROCESS_BROADCAST? "<broadcast>": PROCESS_NAME_STRING(p), nevents);
  }
  /*nevents未处理事件总数，事件队列满，返回PROCESS_ERR_FULL*/
  if(nevents == PROCESS_CONF_NUMEVENTS) {
#if DEBUG
    if(p == PROCESS_BROADCAST) {
      printf("soft panic: event queue is full when broadcast event %d was posted from %s\n", ev, PROCESS_NAME_STRING(process_current));
    } else {
      printf("soft panic: event queue is full when event %d was posted to %s frpm %s\n", ev, PROCESS_NAME_STRING(p), PROCESS_NAME_STRING(process_current));
    }
#endif /* DEBUG */
    return PROCESS_ERR_FULL;
  }
  //事件队列未满，继续
  snum = (process_num_events_t)(fevent + nevents) % PROCESS_CONF_NUMEVENTS;//取得循环队列中下一个空闲位置
  events[snum].ev = ev;//将事件加入队列
  events[snum].data = data;
  events[snum].p = p;
  ++nevents;

#if PROCESS_CONF_STATS
  if(nevents > process_maxevents) {
    process_maxevents = nevents;
  }
#endif /* PROCESS_CONF_STATS */
  
  return PROCESS_ERR_OK;
}

　　process_post首先判断事件队列是否已满，若满返回错误，否则取得下一个空闲位置（因为环形队列，需要做余操作），而后，设置该事件并将未处理事件总数加1。

事件调度

　　事件没有优先级，采用先到先服务策略，每一次系统轮询（process_run函数）只处理一个事件，do_event函数用于处理事件，其流程图如下：

/*
 * Process the next event in the event queue and deliver it to
 * listening processes.
 */
/*---------------------------------------------------------------------------*/
static void
do_event(void)
{
  static process_event_t ev;
  static process_data_t data;
  static struct process *receiver;
  static struct process *p;
  
  /*
   * If there are any events in the queue, take the first one and walk
   * through the list of processes to see if the event should be
   * delivered to any of them. If so, we call the event handler
   * function for the process. We only process one event at a time and
   * call the poll handlers inbetween.
   */

  if(nevents > 0) {
    
    /* There are events that we should deliver. */
      /*取出待处理事件*/
    ev = events[fevent].ev;
    
    data = events[fevent].data;
    receiver = events[fevent].p;

    /* Since we have seen the new event, we move pointer upwards
       and decrese the number of events. */
      /*更新fevent和nevents*/
    fevent = (fevent + 1) % PROCESS_CONF_NUMEVENTS;
    --nevents;

    /* If this is a broadcast event, we deliver it to all events, in
       order of their priority. */
      /*是否广播事件*/
    if(receiver == PROCESS_BROADCAST) {
      for(p = process_list; p != NULL; p = p->next) {

        /* If we have been requested to poll a process, we do this in
           between processing the broadcast event. */
         /*有高优先级进程，运行所有高优先级进程*/
        if(poll_requested) {
          do_poll();
        }
        call_process(p, ev, data);//处理事件
      }
    } 
    else {
          /* This is not a broadcast event, so we deliver it to the
         specified process.不是广播事件，提供给特定进程 */
          /* If the event was an INIT event, we should also update the
         state of the process. 如果是初始化事件，设置进程状态为RUNNING*/
        if(ev == PROCESS_EVENT_INIT) {
        receiver->state = PROCESS_STATE_RUNNING;
      }

      /* Make sure that the process actually is running. */
      call_process(receiver, ev, data);//处理事件
    }
  }
}

　　do_event首先取出该事件（即，将事件的值复制到一个新变量），更新总的未处理事件总数及下一个待处理事件的数组下标（环形队列，需要取余操作）。接着判断事件是否为广播事件PROCESS_BROADCAST，若是，考虑到处理广播事件可能需要更多的时间，为保证系统实时性，先运行高优先级的进程，而后再去处理事件（调用call_process函数）。如果事件是初始化事件PROCESS_EVENT_INIT（创建进程的时候会触发此事件），需要将进程状态设为PROCESS_STATE_RUNNING。

事件处理

　　实际的事件处理是在进程的函数体thread，正如上文所述，call_process会调用thread函数，执行该进程。关键代码如下：

ret = p->thread(&p->pt, ev, data);

参考Jelline大神博客： http://jelline.blog.chinaunix.net