Linux sentinel写法

原创已于 2024-09-23 16:42:03 修改 · 615 阅读

0 ·

CC 4.0 BY-SA版权

文章标签：

#linux #sentinel

于 2024-08-29 19:05:01 首次发布

Linux 同时被 2 个专栏收录

19 篇文章

订阅专栏

Linux_内核开发

15 篇文章

订阅专栏

在linux驱动里我们经常能看到类似下面的写法：

static const struct of_device_id asensm6_of_match[] = {
	{ .compatible = DRIVER_COMPATIBLE },
	{ /* sentinel */ },
};

static const struct of_device_id rockchip_pinctrl_dt_match[] = {
	{ .compatible = "rockchip,px30-pinctrl",
		.data = &px30_pin_ctrl },
	{ .compatible = "rockchip,rv1108-pinctrl",
		.data = &rv1108_pin_ctrl },
	{ .compatible = "rockchip,rv1126-pinctrl",
		.data = &rv1126_pin_ctrl },
	{ .compatible = "rockchip,rk1808-pinctrl",
		.data = &rk1808_pin_ctrl },
	{ .compatible = "rockchip,rk2928-pinctrl",
		.data = &rk2928_pin_ctrl },
	{ .compatible = "rockchip,rk3036-pinctrl",
		.data = &rk3036_pin_ctrl },
	{ .compatible = "rockchip,rk3066a-pinctrl",
		.data = &rk3066a_pin_ctrl },
	{ .compatible = "rockchip,rk3066b-pinctrl",
		.data = &rk3066b_pin_ctrl },
	{ .compatible = "rockchip,rk3128-pinctrl",
		.data = (void *)&rk3128_pin_ctrl },
	{ .compatible = "rockchip,rk3188-pinctrl",
		.data = &rk3188_pin_ctrl },
	{ .compatible = "rockchip,rk3228-pinctrl",
		.data = &rk3228_pin_ctrl },
	{ .compatible = "rockchip,rk3288-pinctrl",
		.data = &rk3288_pin_ctrl },
	{ .compatible = "rockchip,rk3308-pinctrl",
		.data = &rk3308_pin_ctrl },
	{ .compatible = "rockchip,rk3328-pinctrl",
		.data = &rk3328_pin_ctrl },
	{ .compatible = "rockchip,rk3368-pinctrl",
		.data = &rk3368_pin_ctrl },
	{ .compatible = "rockchip,rk3399-pinctrl",
		.data = &rk3399_pin_ctrl },
	{},
};

可以看到数组声明初始化时的尾部会有一个空白的{}, 对于GCC编译器来说它会被初始化为0，也就是说{}等价于{.a = 0, .b = 0}, linux的用意在于防止数组越界以及不需要去动态计算大小。
例如我们现在想遍历一个数组，那么终止条件就是判断当前是否为结束符，如果是结束符则遍历结束。
在声明中使用哨兵主要目的简化链表与数组的算法步骤，使得判断终止变得比较单一了，无论什么条件只要遇到了哨兵那就意味着这个数组的遍历该结束了，因为判断比较单一这样能够提升CPU分支预测的命中率，因为如果判断语句较为复杂会导致CPU分支预测失败率提高，当分支预测失败时CPU为了保证程序能够正常运行下去会重新回滚去执行这就间接导致了这个代码被放入热路径的概率会非常小。
此外在遍历数组时不需要去关心它的大小了，也不需要去计算，只需要简单的判断是否指向了哨兵元素就可以了，因为它判断的单一性，会让CPU很容易预测并将代码放入热路径（CPU缓存）里加快寻址与执行过程让循环代码变得更紧凑。