死锁问题----打印此刻系统中所有的task

一、task_struct的遍历

进程，线程和内核线程的基本属性都是由struct task_struct结构体定义的。
设备上运行的所有task都是通过双链表连成一串，其中第一个task就是init_task，最终又指向会init_task。因此可以通过首个task（init_task）通过双向链表（tasks）遍历所有task，最终回到init_task方式遍历所有的task。      

首进程init_task在内核启动的时候静态赋值创建。(/init/init_task.c)

struct task_struct init_task
= {
       .state                	= 0,
       .stack                	= init_stack,
       .usage                	= REFCOUNT_INIT(2),
       .flags                	= PF_KTHREAD,
       .prio                	= MAX_PRIO - 20,
       .static_prio        		= MAX_PRIO - 20,
       .normal_prio        		= MAX_PRIO - 20,
       .policy                	= SCHED_NORMAL,
       .cpus_ptr        		= &init_task.cpus_mask,
       .cpus_mask        		= CPU_MASK_ALL,
       .nr_cpus_allowed			= NR_CPUS,
       .mm                		= NULL,
       .active_mm        		= &init_mm,
       .tasks					= LIST_HEAD_INIT(init_task.tasks),
       ......
       .signal                	= &init_signals,
       .sighand        			= &init_sighand,
       .nsproxy        			= &init_nsproxy,
       .pending        			= {
               .list 	= LIST_HEAD_INIT(init_task.pending.list),
               .signal 	= {{0}}
       },
       .blocked        			= {{0}},
       .alloc_lock        		= __SPIN_LOCK_UNLOCKED(init_task.alloc_lock),
       .journal_info        	= NULL,
       INIT_CPU_TIMERS(init_task)
       .pi_lock        			= __RAW_SPIN_LOCK_UNLOCKED(init_task.pi_lock),
       .timer_slack_ns 			= 50000, /* 50 usec default slack */
       .thread_pid        		= &init_struct_pid,
       .thread_group        	= LIST_HEAD_INIT(init_task.thread_group),
       .thread_node        		= LIST_HEAD_INIT(init_signals.thread_head),
       ......
};
EXPORT_SYMBOL(init_task);

可以在编译生成的linux镜像中查找init_task符号表

nm vmlinux | grep init_task

然而一个进程中往往有很多线程，线程也与进程类似，通过双向链表串联，最终挂在进程task_struct下的thread_group
遍历系统中所有的进程和进程中所有的线程。

其中打印堆栈的方法：https://blog.csdn.net/sydyh43/article/details/119707079

二、遍历的实现

#define BKTRACE_DEPTH	30

#define next_task(p)	list_entry_rcu((p)->tasks.next, struct task_struct, tasks)
#define next_thread(p)	list_entry_rcu((p)->thread_group.next, struct task_struct, thread_group)

static void print_tasks_stack(void)
{
	struct task_struct *p = NULL, *tmp = NULL, *pt = NULL;
	unsigned long backtrace[BKTRACE_DEPTH];
	struct stack_trace trace;
			
	printk("dump stack....\n");
	rcu_read_lock();

	for (p = &init_task; (tmp = next_task(p)) != &init_task; ) {
		pt = p;
		do {
			if (pt->state == TASK_RUNNING) {
				printk("task name:%s\n", pt->comm);
				memset(backtrace, 0x00, sizeof(unsigned long) * BKTRACE_DEPTH);
				memset(&trace, 0x00, sizeof(struct stack_trace));
				trace.max_entries = BKTRACE_DEPTH;
				trace.entries = backtrace;
				
				save_stack_trace_tsk(pt, &trace);
				print_stack_trace(&trace, 0);			
			}
			
			if (pt->state & TASK_INTERRUPTIBLE) {
				printk("task name:%s\n", pt->comm);
				memset(backtrace, 0x00, sizeof(unsigned long) * BKTRACE_DEPTH);
				memset(&trace, 0x00, sizeof(struct stack_trace));
				trace.max_entries = BKTRACE_DEPTH;
				trace.entries = backtrace;
				
				save_stack_trace_tsk(pt, &trace);
				print_stack_trace(&trace, 0);
			}
			
			if (pt->state & TASK_UNINTERRUPTIBLE) {
				printk("task name:%s\n", pt->comm);
				memset(backtrace, 0x00, sizeof(unsigned long) * BKTRACE_DEPTH);
				memset(&trace, 0x00, sizeof(struct stack_trace));
				trace.max_entries = BKTRACE_DEPTH;
				trace.entries = backtrace;
				
				save_stack_trace_tsk(pt, &trace);
				print_stack_trace(&trace, 0);
			}
		}while((pt = next_thread(pt)) != p);
		
		p = tmp;
	}

	rcu_read_unlock();
}

遍历结果

发现个问题，创建线程时，printk("task name:%s\n", pt->comm);打印出来的线程名就是进程的名字，最终不好定位问题。

因此，在创建线程的时候最好指定线程名。创建线程名的方法。

#include <sys/prctl.h>

static void *thread_fun0(void *arg)
{
	prctl(PR_SET_NAME, "thread_fun0");	
	while (1) {
		fun_b();
		
		sleep(5);
	}
	
	return NULL;
}

 三、应用

1、模拟死锁的代码

static void fun_a(void)
{
	pthread_mutex_lock(&mtx0);
	sleep(2);
	printf("1%s\n", __func__);
	pthread_mutex_lock(&mtx1);
	printf("2%s\n", __func__);
	
	return;
}

static void fun_b(void)
{
	pthread_mutex_lock(&mtx1);
	sleep(1);
	printf("1%s\n", __func__);	
	pthread_mutex_lock(&mtx0);
	printf("1%s\n", __func__);	

	return;
}

static void *thread_fun0(void *arg)
{
	prctl(PR_SET_NAME, "thread_fun0");	
	while (1) {
		fun_a();
		
		sleep(5);
	}
	
	return NULL;
}

static void *thread_fun1(void *arg)
{
	prctl(PR_SET_NAME, "thread_fun1");	
	while (1) {
		
		fun_b();
		
		sleep(7);
	}
	
	return NULL;
}

2、打印堆栈

3、因此当设备出现狗咬死复位的前一刻，需要把堆栈都打印出来，方便后续的问题定位