linux内核死锁调试

使用内核的kernel hacking功能

打开以下配置

CONFIG_PROVE_LOCKING=y
CONFIG_LOCK_STAT=y
CONFIG_DEBUG_LOCKDEP=y

重新编译内核后，

proc目录下会有lockdep lockdep_stats lockdep_chains说明lockdep模块已经生效

测试code

#include 
#include 
#include 
#include 


static DEFINE_SPINLOCK(hack_spinA);
static DEFINE_SPINLOCK(hack_spinB);
#ifdef MEMORY_TEST
static char* buf;
static void create_slue_err(void){
    buf = kmalloc(32,GFP_KERNEL);
    if(buf)  {
        /*memset(buf,0x00,33);*/
    /*  kfree(buf);
        printk("%sn","free buf" );*/
        /*kfree(buf);*/
        memset(buf,0x00,33);
    }

    return;
}
#endif
void hack_spinBA(void) {
    printk("%sn","hack_spin:B=>An" );
    spin_lock(&hack_spinA);
    spin_lock(&hack_spinB);
}

void hack_spinAB(void) {
    printk("%sn","hack_spin:A=>Bn" );
    spin_lock(&hack_spinB);
}

static int __init my_test_init(void) {
    printk("init %sn", "my_test_init 1");
#ifdef MEMORY_TEST 
    create_slue_err();
#endif
    hack_spinBA();
    hack_spinAB();
    printk("init %sn", "my_test_init 2");
    return 0;
}

static void __exit my_test_exit(void) {
    printk("%sn","my_test_exit" );
    return ;
}
MODULE_LICENSE("GPL");
module_init(my_test_init);
module_exit(my_test_exit);

测试打印：

[  188.596504@1] init my_test_init 1
[  188.596548@1] hack_spin:B=>A
[  188.596548@1] 
[  188.598679@1] hack_spin:A=>B
[  188.598679@1] 
[  189.920389@1] BUG: spinlock lockup suspected on CPU#1, insmod/6999
[  189.920943@1]  lock: hack_spinB+0x0/0xfffffffffffffef4 [slub_test], .magic: dead4ead, .owner: insmod/6999, .owner_cpu: 1
[  189.931758@1] CPU: 1 PID: 6999 Comm: insmod Tainted: G           O 3.14.29 #6
[  189.938784@1] Call trace:
[  189.941406@1] [] dump_backtrace+0x0/0x144
[  189.946896@1] [] show_stack+0x1c/0x28
[  189.952120@1] [] dump_stack+0x74/0xb8
[  189.957247@1] [] spin_dump+0x78/0x98
[  189.962369@1] [] do_raw_spin_lock+0x170/0x1a8
[  189.968199@1] [] _raw_spin_lock+0x68/0x88
[  189.973764@1] [] hack_spinAB+0x30/0x3c [slub_test]
[  189.980015@1] [] $x+0x28/0x4c [slub_test]
[  189.985566@1] [] do_one_initcall+0xd4/0x13c
[  189.991257@1] [] load_module+0x16d8/0x1e08
[  189.996834@1] [] SyS_finit_module+0x80/0x90

[  211.641019@1] INFO: rcu_sched detected stalls on CPUs/tasks: { 1} (detected by 2, t=2104 jiffies, g=18446744073709551337, c=18446744073709551336, q=4)
[  211.648804@2] Task dump for CPU 1:
[  211.652167@2] insmod          R  running task        0  6999   5821 0x0000000a
[  211.659325@2] Call trace:
[  211.661920@2] [] __switch_to+0x74/0x8c

很明显可以看出死锁的发生路径与调用栈

hack_spinAB在申请spin_lock(&hack_spinB)时死锁了

总结

以上为个人经验，希望能给大家一个参考，也希望大家多多支持IT俱乐部。