最高效的进(线)程间通信机制--eventfd
我们常用的进程（线程）间通信机制有管道，信号，消息队列，信号量，共享内存，socket等等，其中主要作为进程（线程）间通知/等待的有管道pipe和socketpair。线程还有特别的condition。
今天来看一个liunx较新的系统调用，它是从LINUX 2.6.27版本开始增加的，主要用于进程或者线程间的通信（如通知/等待机制的实现）。
首先来看一下函数原型：
#include
int eventfd(unsigned int initval, int flags);
下面是它man手册中的描述，我照着翻译了一遍（我英语四级434，你们要是怀疑下文的话，可以直接去man eventfd ^^）：
eventfd()创建了一个"eventfd object"，能在用户态用做事件wait/notify机制，通过内核取唤醒用户态的事件。这个对象保存了一个内核维护的uint64_t类型的整型counter。这个counter初始值被参数initval指定，一般初值设置为0。
它的标记可以有以下属性：
EFD_CLOECEX，EFD_NONBLOCK，EFD_SEMAPHORE。
在linux直到版本2.6.26，这个flags参数是没用的，必须指定为0。
它返回了一个引用eventfd object的描述符。这个描述符可以支持以下操作：
read：如果计数值counter的值不为0，读取成功，获得到该值。如果counter的值为0，非阻塞模式，会直接返回失败，并把errno的值指纹EINVAL。如果为阻塞模式，一直会阻塞到counter为非0位置。
write：会增加8字节的整数在计数器counter上，如果counter的值达到0xfffffffffffffffe时，就会阻塞。直到counter的值被read。阻塞和非阻塞情况同上面read一样。
close：这个操作不用说了。
重点是支持这个：
poll(2), select(2) (and similar)
The returned file descriptor supports poll(2) (and analogously epoll(7)) and select(2), as follows:
* The file descriptor is readable (the select(2) readfds argument; the poll(2) POLLIN flag) if the counter has a
value greater than 0.
* The file descriptor is writable (the select(2) writefds argument; the poll(2) POLLOUT flag) if it is possible to
write a value of at least "1" without blocking.
* If an overflow of the counter value was detected, then select(2) indicates the file descriptor as being both
readable and writable, and poll(2) returns a POLLERR event. As noted above, write(2) can never overflow the
counter. However an overflow can occur if 2^64 eventfd "signal posts" were performed by the KAIO subsystem (the‐
oretically possible, but practically unlikely). If an overflow has occurred, then read(2) will return that maxi‐
mum uint64_t value (i.e., 0xffffffffffffffff).
The eventfd file descriptor also supports the other file-descriptor multiplexing APIs: pselect(2) and ppoll(2).
它的内核代码实现是这样子的：
int eventfd_signal(struct eventfd_ctx *ctx, int n)
{
unsigned long flags;
if (n < 0)
return -EINVAL;
spin_lock_irqsave(&ctx->wqh.lock, flags);
if (ULLONG_MAX - ctx->count < n)
n = (int) (ULLONG_MAX - ctx->count);
ctx->count += n;
if (waitqueue_active(&ctx->wqh))
wake_up_locked_poll(&ctx->wqh, POLLIN);
spin_unlock_irqrestore(&ctx->wqh.lock, flags);
return n;
}本质就是做了一次唤醒，不用read，也不用write，与eventfd_write的区别是不用阻塞。
说了这么多，我们来看一个例子，理解理解其中的含义：
#include
#include
#include
#include
#include /* Definition of uint64_t */
#define handle_error(msg) \
do { perror(msg); exit(EXIT_FAILURE); } while (0)
int
main(int argc, char *argv[])
{
int efd, j;
uint64_t u;
ssize_t s;
if (argc < 2) {
fprintf(stderr, "Usage: %s ...\n", argv[0]);
exit(EXIT_FAILURE);
}
efd = eventfd(0, 0);
if (efd == -1)
handle_error("eventfd");
switch (fork()) {
case 0:
for (j = 1; j < argc; j++) {
printf("Child writing %s to efd\n", argv[j]);
u = strtoull(argv[j], NULL, 0);
/* strtoull() allows various bases */
s = write(efd, &u, sizeof(uint64_t));
if (s != sizeof(uint64_t))
handle_error("write");
}
printf("Child completed write loop\n");
exit(EXIT_SUCCESS);
default:
sleep(2);
printf("Parent about to read\n");
s = read(efd, &u, sizeof(uint64_t));
if (s != sizeof(uint64_t))
handle_error("read");
printf("Parent read %llu (0x%llx) from efd\n",
(unsigned long long) u, (unsigned long long) u);
exit(EXIT_SUCCESS);
case -1:
handle_error("fork");
}
}
$ ./a.out 1 2 4 7 14
Child writing 1 to efd
Child writing 2 to efd
Child writing 4 to efd
Child writing 7 to efd
Child writing 14 to efd
Child completed write loop
Parent about to read
Parent read 28 (0x1c) from efd
注意：这里用了sleep(2)保证子进程循环写入完毕，得到的值就是综合28。如果不用sleep(2)来保证时序，当子进程写入一个值，父进程会立马从eventfd读出该值。