树莓派内核默认没有启用看门狗功能，当内核挂死时将进入“死机”状态或kgdb调试状态，并不会自动重启系统。本文为树莓派开启看门狗功能并通过内核线程周期性喂狗，当出现系统崩溃时会自动重启Linux系统。
环境说明：（1）单板：树莓派b
（2）Linux内核：Linux-4.1.15
（3）Bootloader：u-boot-2015.10
源码文件：linux-rpi-4.1.y/drivers/watchdog/bcm2835_wdt.c
1、看门狗驱动源码分析
树莓派的看门狗驱动程序为内核drivers/watchdog/bcm2835_wdt.c文件，该驱动程序实现了开关看门狗和喂狗的功能（不提供喂狗策略），它向内核看门狗子系统注册驱动设备，将喂狗策略移交应用程序，由应用程序打开/dev/watchdogX标准接口并完成周期性喂狗的操作。简单分析一下该驱动程序的源码：
static const struct of_device_id bcm2835_wdt_of_match[] = {
{ .compatible = "brcm,bcm2835-pm-wdt", },
{},
};
MODULE_DEVICE_TABLE(of, bcm2835_wdt_of_match);
static struct platform_driver bcm2835_wdt_driver = {
.probe = bcm2835_wdt_probe,
.remove = bcm2835_wdt_remove,
.shutdown = bcm2835_wdt_shutdown,
.driver = {
.name = "bcm2835-wdt",
.of_match_table = bcm2835_wdt_of_match,
},
};
module_platform_driver(bcm2835_wdt_driver);module_param(heartbeat, uint, 0);
MODULE_PARM_DESC(heartbeat, "Initial watchdog heartbeat in seconds");
module_param(nowayout, bool, 0);
MODULE_PARM_DESC(nowayout, "Watchdog cannot be stopped once started (default="
__MODULE_STRING(WATCHDOG_NOWAYOUT) ")");static int bcm2835_wdt_probe(struct platform_device *pdev)
{
struct device *dev = &pdev->dev;
struct device_node *np = dev->of_node;
struct bcm2835_wdt *wdt;
int err;
wdt = devm_kzalloc(dev, sizeof(struct bcm2835_wdt), GFP_KERNEL);
if (!wdt)
return -ENOMEM;
platform_set_drvdata(pdev, wdt);
spin_lock_init(&wdt->lock);
wdt->base = of_iomap(np, 0);
if (!wdt->base) {
dev_err(dev, "Failed to remap watchdog regs");
return -ENODEV;
}
watchdog_set_drvdata(&bcm2835_wdt_wdd, wdt);
watchdog_init_timeout(&bcm2835_wdt_wdd, heartbeat, dev);
watchdog_set_nowayout(&bcm2835_wdt_wdd, nowayout);
err = watchdog_register_device(&bcm2835_wdt_wdd);
if (err) {
dev_err(dev, "Failed to register watchdog device");
iounmap(wdt->base);
return err;
}
dev_info(dev, "Broadcom BCM2835 watchdog timer");
return 0;
}其中的watchdog_device结构实例bcm2835_wdt_wdd定义如下：
static struct watchdog_device bcm2835_wdt_wdd = {
.info = &bcm2835_wdt_info,
.ops = &bcm2835_wdt_ops,
.min_timeout = 1,
.max_timeout = WDOG_TICKS_TO_SECS(PM_WDOG_TIME_SET),
.timeout = WDOG_TICKS_TO_SECS(PM_WDOG_TIME_SET),
};static struct watchdog_ops bcm2835_wdt_ops = {
.owner = THIS_MODULE,
.start = bcm2835_wdt_start,
.stop = bcm2835_wdt_stop,
.set_timeout = bcm2835_wdt_set_timeout,
.get_timeleft = bcm2835_wdt_get_timeleft,
};bcm2835_wdt_probe函数在完成看门狗的注册之后，在文件系统的/dev目录下就会生成watchdog0设备文件，后面应用程序就可以通过操作它来实现看门狗的控制了。Linux内核的看门狗子系统实现在drivers/watchdog/目录下的watchdog_core.c和watchdog_dev.c文件中，其中watchdog_core.c实现了子系统的初始化以及提供了面向驱动的注册接口函数watchdog_register_device等；watchdog_dev.c实现了字符设备的初始化和注册，同时提供了设备文件控制接口watchdog_fops：
static const struct file_operations watchdog_fops = {
.owner = THIS_MODULE,
.write = watchdog_write,
.unlocked_ioctl = watchdog_ioctl,
.open = watchdog_open,
.release = watchdog_release,
};#define WDIOC_GETSUPPORT _IOR(WATCHDOG_IOCTL_BASE, 0, struct watchdog_info) //获取看门狗info信息
#define WDIOC_GETSTATUS _IOR(WATCHDOG_IOCTL_BASE, 1, int) //查询看门狗status状态信息
#define WDIOC_GETBOOTSTATUS _IOR(WATCHDOG_IOCTL_BASE, 2, int) //查询看门狗bootstatus信息
......
#define WDIOC_SETOPTIONS _IOR(WATCHDOG_IOCTL_BASE, 4, int) //设置开关看门狗
#define WDIOC_KEEPALIVE _IOR(WATCHDOG_IOCTL_BASE, 5, int) //喂狗
#define WDIOC_SETTIMEOUT _IOWR(WATCHDOG_IOCTL_BASE, 6, int) //设置看门狗超时时间
#define WDIOC_GETTIMEOUT _IOR(WATCHDOG_IOCTL_BASE, 7, int) //查询看门狗超时时间
......
#define WDIOC_GETTIMELEFT _IOR(WATCHDOG_IOCTL_BASE, 10, int) //查询看门狗距离超时的剩余时间
2、看门狗驱动源码修改
由于看门狗驱动程序并没有实现喂狗策略，因此要启用看门狗可以通过编写应用程序，在应用层创建一个守护进程实现周期喂狗操作。但本文并不通过该方式实现，而是修改该驱动程序，在内核中创建一个内核线程来实现喂狗的动作。在驱动源码中添加以下3个函数：
static void bcm2835_wdt_set_prio(unsigned int policy, unsigned int prio)
{
struct sched_param param = { .sched_priority = prio };
sched_setscheduler(current, policy, ¶m);
}
static int bcm2835_wdt_kthread(void *data)
{
struct watchdog_device *wdog = (struct watchdog_device *)data;
/* 调整内核线程的调度策略和优先级 */
bcm2835_wdt_set_prio(SCHED_FIFO, MAX_RT_PRIO - 1);
while(1) {
if (kthread_should_stop()) {
bcm2835_wdt_stop(wdog);
break;
}
if (bcm2835_wdt_wdd.timeout != 0) {
(void)bcm2835_wdt_start(&bcm2835_wdt_wdd);
msleep((bcm2835_wdt_wdd.timeout >> 2)*1000);
dev_dbg(wdog->dev, "BCM2835 ping watchdog");
} else {
msleep(1000);
}
}
return 0;
}
static void bcm2835_wdt_start_kthread(struct watchdog_device *wdog)
{
/* 启动看门狗并创建内核线程执行喂狗操作 */
mutex_lock(&bcm2835_wdt_lock);
if (kwdt_task == NULL) {
kwdt_task = kthread_run(bcm2835_wdt_kthread, &bcm2835_wdt_wdd, "bcm2835_kwdt");
if (IS_ERR(kwdt_task)) {
dev_err(wdog->dev, "Failed to Create BCM2835 kernel watchdog thread");
kwdt_task = NULL;
}
dev_info(wdog->dev, "Create BCM2835 kernel watchdog thread OK!"
" TimeOut = %d sec", bcm2835_wdt_wdd.timeout);
}
mutex_unlock(&bcm2835_wdt_lock);
}在看门狗启动函数bcm2835_wdt_start中添加如下：
static int bcm2835_wdt_start(struct watchdog_device *wdog)
{
/* 喂狗操作 */
......
/* added by zhangyi 2016.5.7 */
bcm2835_wdt_start_kthread(wdog);
/* * */
return 0;
}static int bcm2835_wdt_remove(struct platform_device *pdev)
{
struct bcm2835_wdt *wdt = platform_get_drvdata(pdev);
/* added by zhangyi 2016.5.7 */
mutex_lock(&bcm2835_wdt_lock);
if (kwdt_task) {
kthread_stop(kwdt_task);
kwdt_task = NULL;
}
mutex_unlock(&bcm2835_wdt_lock);
/* * */
watchdog_unregister_device(&bcm2835_wdt_wdd);
iounmap(wdt->base);
return 0;
}
3、看门狗实验效果
（2）修改dts文件arch/arm/boot/dts/bcm2708_common.dtsi，将其中的watchdog部分修改如下：
watchdog: watchdog@7e100000 {
compatible = "brcm,bcm2835-pm-wdt";
reg = <0x7e100000 0x28>;
timeout-sec = <15>; //added by zhangyi 2016.5.7
//status = "disabled";
};
（3）启动看门狗
启动树莓派进入Linux系统后，可以发现在看门狗驱动模块已经顺利加载了：
root@apple:~# lsmod
Module Size Used by
...
bcm2835_wdt 4142 0
...
内核启动日志输出如下：
[ 12.868529] bcm2835-wdt 20100000.watchdog: Broadcom BCM2835 watchdog timer
下面启动看门狗内核线程，超时时间为15s
root@apple:~# echo 0 > /dev/watchdog0
root@apple:~# dmesg -c
[ 958.846139] watchdog watchdog0: Create BCM2835 kernel watchdog thread OK! TimeOut = 15 sec
（4）手动触发看门狗复位
接下来为了验证看门狗的功能，这里手动触发内核kernel panic，使得内核进程无法在调度，从而无法执行喂狗，最终系统重启。
root@apple:~# echo c > /proc/sysrq-trigger
[ 1299.165014] sysrq: SysRq : Trigger a crash
[ 1299.169352] Unable to handle kernel NULL pointer dereference at virtual address 00000000
[ 1299.177544] pgd = c5ee8000
[ 1299.180279] [00000000] *pgd=05e50831, *pte=00000000, *ppte=00000000
[ 1299.186665] Internal error: Oops: 817 [#1] ARM
Entering kdb (current=0xc5838da0, pid 507) Oops: (null)
due to oops @ 0xc0318624
CPU: 0 PID: 507 Comm: bash Tainted: G O 4.1.15 #5
Hardware name: BCM2708
task: c5838da0 ti: c5ed0000 task.ti: c5ed0000
PC is at sysrq_handle_crash+0x28/0x34
LR is at __handle_sysrq+0x9c/0x16c
pc : [<c0318624>] lr : [<c0318edc>] psr: 60000013
sp : c5ed1e60 ip : c5ed1e70 fp : c5ed1e6c
r10: 00000002 r9 : c5ed0000 r8 : 00000000
r7 : 00000005 r6 : 00000063 r5 : c0bc6bc4 r4 : c0c0cb48
r3 : 00000000 r2 : 00000001 r1 : c0c32590 r0 : 00000063
Flags: nZCv IRQs on FIQs on Mode SVC_32 ISA ARM Segment user
Control: 00c5387d Table: 05ee8008 DAC: 00000015
CPU: 0 PID: 507 Comm: bash Tainted: G O 4.1.15 #5
Hardware name: BCM2708
[<c0016660>] (unwind_backtrace) from [<c0013524>] (show_stack+0x20/0x24)
[<c0013524>] (show_stack) from [<c05263c4>] (dump_stack+0x20/0x28)
[<c05263c4>] (dump_stack) from [<c0010ae4>] (show_regs+0x1c/0x20)
[<c0010ae4>] (show_regs) from [<c00939f0>] (kdb_main_loop+0x33c/0x740)
[<c00939f0>] (kdb_main_loop) from [<c00963e8>] (kdb_stub+0x18c/0x3cc)
[<c00963e8>] (kdb_stub) from [<c008ccfc>] (kgdb_handle_exception+0x27c/0x7c0)
more>
U-Boot 2015.10 (Jan 02 2016 - 10:49:06 +0800)
DRAM: 128 MiB
RPI Model B rev2
MMC: bcm2835_sdhci: 0
reading uboot.env
In: serial
Out: lcd
Err: lcd
Net: Net Initialization Skipped
No ethernet found.
Hit any key to stop autoboot: 0
可以看到在触发了panic以后的15s后，系统自动重启，重新加载u-boot并引导Linux系统启动，至此树莓派的看门狗功能添加完成。
最后，本文中仅作为一种简单的实现方式，亦可根据需求实现看门狗在用户主进程中或硬件中断处理程序中执行喂狗策略（中断中可监测业务进程的执行状态并判断是否需要喂狗），实现方法和种类很多，原理一样。
6589
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
