轮询运行时间比超时时间长得多

Question

我有一个简单的 C 程序，我在我的 Ubuntu 22.04 VM 上编译了它，它通过 VirtualBox 在我的 MacBook 上运行：

#include <poll.h>
#include <stdio.h>
#include <unistd.h>

int main() {
    int fds[2];
    struct pollfd poller = {.events = POLLIN};

    if ( pipe(fds) != 0 ) {
        perror("pipe");
        return 1;
    }

    poller.fd = fds[0];
    poll(&poller, 1, 1000);

    close(fds[0]);
    close(fds[1]);

    return 0;
}

显然，

fds[0]

永远不会准备好读取，所以

poll

应该在 1000 毫秒后超时。

然而，当我跑步时

while true; do time ./a.out; done

我明白了

./a.out  0.01s user 0.00s system 0% cpu 9.924 total
./a.out  0.01s user 0.00s system 0% cpu 10.837 total
./a.out  0.01s user 0.01s system 0% cpu 4.070 total
./a.out  0.01s user 0.00s system 0% cpu 8.606 total
./a.out  0.00s user 0.00s system 0% cpu 1.131 total
./a.out  0.01s user 0.01s system 1% cpu 1.018 total
./a.out  0.00s user 0.01s system 0% cpu 2.682 total
./a.out  0.00s user 0.00s system 0% cpu 1.598 total
./a.out  0.01s user 0.00s system 1% cpu 1.018 total
./a.out  0.01s user 0.00s system 1% cpu 1.017 total
./a.out  0.01s user 0.00s system 1% cpu 1.055 total
./a.out  0.01s user 0.00s system 1% cpu 1.019 total
./a.out  0.01s user 0.01s system 1% cpu 1.045 total
./a.out  0.00s user 0.01s system 1% cpu 1.078 total
./a.out  0.00s user 0.01s system 0% cpu 8.336 total
./a.out  0.01s user 0.01s system 0% cpu 3.344 total
./a.out  0.01s user 0.00s system 1% cpu 1.019 total
./a.out  0.01s user 0.02s system 2% cpu 1.050 total
./a.out  0.00s user 0.01s system 1% cpu 1.022 total
./a.out  0.01s user 0.00s system 1% cpu 1.014 total
./a.out  0.00s user 0.00s system 0% cpu 2.352 total
./a.out  0.00s user 0.00s system 0% cpu 1.023 total
./a.out  0.01s user 0.00s system 0% cpu 1.323 total
...

在单独的终端中运行

htop

显示最小的 CPU/内存使用率（VM 有 4 个 2 GHz 处理器和 4 GB RAM）。

为什么

poll

会产生这种不一致的行为？