我要评论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 <linux/kernel.h>
#include <linux/module.h>
#include <linux/init.h>
#include <linux/slab.h>
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("%s\n","free buf" );*/
/*kfree(buf);*/
memset(buf,0x00,33);
}
return;
}
#endif
void hack_spinba(void) {
printk("%s\n","hack_spin:b=>a\n" );
spin_lock(&hack_spina);
spin_lock(&hack_spinb);
}
void hack_spinab(void) {
printk("%s\n","hack_spin:a=>b\n" );
spin_lock(&hack_spinb);
}
static int __init my_test_init(void) {
printk("init %s\n", "my_test_init 1");
#ifdef memory_test
create_slue_err();
#endif
hack_spinba();
hack_spinab();
printk("init %s\n", "my_test_init 2");
return 0;
}
static void __exit my_test_exit(void) {
printk("%s\n","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] [<ffffffc0010893f8>] dump_backtrace+0x0/0x144
[ 189.946896@1] [<ffffffc001089558>] show_stack+0x1c/0x28
[ 189.952120@1] [<ffffffc001b5b480>] dump_stack+0x74/0xb8
[ 189.957247@1] [<ffffffc0010fb6d4>] spin_dump+0x78/0x98
[ 189.962369@1] [<ffffffc0010fb904>] do_raw_spin_lock+0x170/0x1a8
[ 189.968199@1] [<ffffffc001b670a8>] _raw_spin_lock+0x68/0x88
[ 189.973764@1] [<ffffffbffc00507c>] hack_spinab+0x30/0x3c [slub_test]
[ 189.980015@1] [<ffffffbffc00d028>] $x+0x28/0x4c [slub_test]
[ 189.985566@1] [<ffffffc0010816d4>] do_one_initcall+0xd4/0x13c
[ 189.991257@1] [<ffffffc00112c2f8>] load_module+0x16d8/0x1e08
[ 189.996834@1] [<ffffffc00112cba0>] 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] [<ffffffc001085f10>] __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代码网。
发表评论