内核崩溃中自旋锁并发的内核崩溃

Question

I want to clear a list which type is defined by Kernel. I have two main structs, num_wrapper and num. num_wapper has a list of num, and kernel crashs when I do the del_all_node() function.

I try to mark the list_del, and the kernel will not be crash. I don't understand why there will be crash problem since I have use spin_lock to protect this num_list.

Any tips will be appreciate.

The following is the simplified code.

#include <linux/init.h>
#include <linux/module.h>
#include <linux/list.h>
#include <linux/slab.h>

struct num_wrapper {
    struct list_head num_list;
    spinlock_t list_lock;
    u8 check_num;
};

struct num {
    struct list_head node;
    int number;
    struct num_wrapper* num_w_ptr;
};

s32 del_all_node(struct num_wrapper *number_wrap)
{
    struct num *tmp;    
    struct num *num_head; 
    spin_lock(&number_wrap->list_lock); 
    list_for_each_entry_safe(num_head, tmp, &number_wrap->num_list, node) {
        printk("num_head is %d\n", num_head->number); 
        list_del(&num_head->node);//this line seems to have problem 
    }
    spin_unlock(&number_wrap->list_lock); 
    return 0;
}

static int __init hello_init(void)
{
    /*Setup Scenario*/     
    struct num_wrapper *number_wrap = kzalloc(sizeof(struct num_wrapper)
            , GFP_KERNEL);  
    struct num *number = kzalloc(sizeof(struct num), GFP_KERNEL);  
    number->number = 10; 
    number_wrap->check_num = 20;
    INIT_LIST_HEAD(&number->node);   
    INIT_LIST_HEAD(&number_wrap->num_list); 
    list_add_tail(&number->node, &number_wrap->num_list); 

    del_all_node(number_wrap);
    return 0;
}

static void hello_exit(void)
{
    printk(KERN_ALERT "Good, haha\n");
}

module_init(hello_init);
module_exit(hello_exit);

update

After doing some debug, I seems to find root cause... my scenario is as follows :

I have a private data struct which is included in the net_device struct. And the following is the simplified scenario:

struct xx_if *xx_if; //this is private data in net_device
struct xx_if *tmp;
list_for_each_entry_safe(xx_if, tmp, xx_if_wrap->if_list, list) {
    free_netdev(xx_if->ndev);
    list_del(&xx_if->list);
} 

Since free_netdev will also free the private data xx_if, the code broken... My fix is change the sequence to these two statements, and it fix the crash problem.

Still strange thing is I have check whether xx_if is NULL, but still lead to crash if I don't interchange these two statements.

Answer 1

I don't clearly understand following code:

INIT_LIST_HEAD(&number->node);   
INIT_LIST_HEAD(&number_wrap->num_list); 
list_add_tail(&number->node, &number_wrap->num_list); 

You init two different structures, then you add one type of list to another type of list.

Is that the way you can do?

I think, that you need something like this:

  struct num{
        u8 check_num;
        struct list_head list;
    };

    struct num_wrapper{
        struct num* num_ptr;
        spinlock_t list_lock;
    };

    int init_num_wrapper(struct num_wrapper** prt){
        if(!ptr && *ptr){
            return -EINVAL;
        }

        *ptr = kzalloc(sizeof(struct num_wrapper), GFP_KERNEL);
        if(!*ptr){
            return -ENOMEM;
        }
        INIT_LIST_HEAD(& (*ptr)->num_ptr->list);
        ... init spinlock
        return 0;
    }

    int add_num(num_wrapper* prt_wrap, u8 check_num){
        ... checking pointers
        struct num num* = NULL;
        num = kmalloc(sizeof(struct num), GFP_KERNEL);
        if(! num){
            return -ENOMEM;
        }

        INIT_LIST_HEAD(&num->list);
        num->check_num = check_num;
        spin_lock(&prt_wrap->list_lock);
        list_add_tail(&num->list, &prt_wrap->num_ptr.list);
        spin_unlock(&prt_wrap->list_lock);
        return 0;
    }

    int remove_all_nodes(num_wrapper* prt_wrap){
        ... checking pointer
        struct num *tmp = NULL;    
        struct num *num_head = NULL; 
        spin_lock(&number_wrap->list_lock); 
        list_for_each_entry_safe(num_head, tmp, &prt_wrap->list, list)
        {
            printk("num_head is %d\n", num_head->number); 
            list_del(&num_head->node);//this line seems to have problem 
        }
        spin_unlock(&num_wrapper->list_lock); 
        return 0;
    }

Update

Then, you can use above functions for manipulating of num_wrapper. For example:

 //...
    struct num_wrapper* nums = NULL;
    init_num_wrapper(&nums); // after this call, you will have inited nums var, which can be used with others functions for manipulating with num_wrapper list.
     u8 num = 2;
     add_num(nums, num); // after this call new node with num will be added to num_wrapper 
    //...