Re: [PATCH] kthread: Don't depend on work queues

[ebiederm@xmission.com (Eric W. Biederman)]

Oleg Nesterov <oleg@tv-sign.ru> writes:

> On 04/10, Eric W. Biederman wrote:
>> 
>> +int kthreadd(void *unused)
>> +{
>> +	/* Setup a clean context for our children to inherit. */
>> +	kthreadd_setup();
>> +
>> +	current->flags |= PF_NOFREEZE;
>> +
>> +	for (;;) {
>> +		wait_event(kthread_create_work,
>> +			   !list_empty(&kthread_create_list));
>> +
>> +		spin_lock(&kthread_create_lock);
>> +		while (!list_empty(&kthread_create_list)) {
>
> Do we need to check the condition under lock? We can miss an event,
> but then it will be noticed by wait_event() above.

We need to be certain there is something on the list before we remove
it.  Otherwise we will start dereferencing bad pointers.

> IOW,
>
> 	for (;;) {
> 		wait_event(kthread_create_work,
> 			   !list_empty(&kthread_create_list));
>
> 		while (!list_empty(&kthread_create_list)) {
> 			struct kthread_create_info *create;
>
> 			spin_lock(&kthread_create_lock);
> 			create = list_entry(kthread_create_list.next,
> 					    struct kthread_create_info, list);
> 			list_del_init(&create->list);
> 			spin_unlock(&kthread_create_lock);
>
> 			create_kthread(create);
> 		}
> 	}

I guess since we are the only process to ever remove things from the
list that would be safe.

Eric