POSIX皮肤中的Xenomai clock_nanosleep跳到Linux内核 [英] Xenomai clock_nanosleep in POSIX skin jumps to Linux Kernel

问题描述

我正在Xenomai中测试POSIX皮肤.我正在尝试从Raspberry Pi上的一些GPIO读取和写入，并且在执行该程序时，上下文切换的数量越来越多(在/proc/xenomai/stat/中).

I'm testing POSIX skin in Xenomai. I'm trying to read and write from some GPIOs on a Raspberry Pi, and when I execute the program, there is an increasing number of context switching (in /proc/xenomai/stat/).

程序的主程序将GPIO映射到内存并启动pthread.造成麻烦的pthread是这样的:

The main of the program maps the GPIOs to memory and starts the pthreads. The pthread that makes trouble is this:

void *productive_thread(void *arg)
{
    struct timespec delay, sleep;
    unsigned long over;

    delay.tv_sec = 0;
    delay.tv_nsec = 10000; // 10 usec

    sleep.tv_sec = 0;
    sleep.tv_nsec = *(long *)arg;

    while(1)
    {
            // This are the read and write macros (gpio is the address of the GPIO mapping):
            // #define GPIO_SET *(gpio+7)
            // #define GPIO_CLR *(gpio+10)
            // #define GPIO_READ(g) (*(gpio + 13)&(1<<(g)))>>4
        while(GPIO_READ(4) != 1);
        GPIO_SET = 1 << 17;
        clock_nanosleep(CLOCK_REALTIME, 0, &delay, NULL);
        GPIO_CLR = 1 << 17;
        clock_nanosleep(CLOCK_REALTIME, 0, &sleep, NULL);
    }
    return NULL;
}

上下文切换的次数每个循环都会增加.我怀疑问题是clock_nanosleep，因为所有其他操作都是算术运算，但是clock_nanosleep是在 Xenomai文档.可以通过使用POSIX皮肤以某种方式进行改进吗?

The number of context switching increases by every loop. I suspect the problem is clock_nanosleep, because all other operations are arithmetic, but clock_nanosleep is defined in Xenomai documentation. Can it be improved somehow (using POSIX skin)?

推荐答案

当进程进入休眠状态时，它将执行自愿的上下文切换.很好，除非您实际上看到错过了最后期限.

When a process sleeps it performs a voluntary context switch. This is fine, unless you are actually seeing your deadlines missed.

这篇关于POSIX皮肤中的Xenomai clock_nanosleep跳到Linux内核的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！