Re: [PATCH -mm 1/2] irq_work, Use llist in irq_work

[Huang Ying <ying.huang@intel.com>]

On 09/01/2011 03:58 PM, Peter Zijlstra wrote:
> On Thu, 2011-09-01 at 11:20 +0800, Huang Ying wrote:
>> On 09/01/2011 09:46 AM, Huang Ying wrote:
>>>>> -static void __irq_work_queue(struct irq_work *entry)
>>>>> +static void __irq_work_queue(struct irq_work *work)
>>>>>  {
>>>>> -	struct irq_work *next;
>>>>> +	struct irq_work_list *irq_work_list;
>>>>>  
>>>>> -	preempt_disable();
>>>>> +	irq_work_list = &get_cpu_var(irq_work_lists);
>>>>>  
>>>>> -	do {
>>>>> -		next = __this_cpu_read(irq_work_list);
>>>>> -		/* Can assign non-atomic because we keep the flags set. */
>>>>> -		entry->next = next_flags(next, IRQ_WORK_FLAGS);
>>>>> -	} while (this_cpu_cmpxchg(irq_work_list, next, entry) != next);
>>>>> +	llist_add(&work->llnode, &irq_work_list->llist);
>>>>>  
>>>>>  	/* The list was empty, raise self-interrupt to start processing. */
>>>>> -	if (!irq_work_next(entry))
>>>>> +	if (!test_and_set_bit(LIST_NONEMPTY_BIT, &irq_work_list->flags))
>>>>>  		arch_irq_work_raise();
>>>>
>>>> So why can't you simply test work->llnode->next?
>>>
>>> Yes.  That is better.  Even if there may be a small race window, it is
>>> not a big issue to raise one extra self interrupt seldom.
>>
>> Remember something about this.  I didn't test work->llnode->next here
>> because I didn't want expose the implementation details like that here.
>>  How about make llist_add() return whether list is empty before adding?
>>  Because it will be an inline function, that should be optimized out if
>> the caller do not need the information.
> 
> You could also use llist_empty() although that brings me to that
> ACCESS_ONCE thing in there, what's the point?

Something as follow with llist_empty() seems not work.

  empty = llist_empty(irq_work_list);
  llist_add(&work->llnode, irq_work_list);
  if (empty)
          arch_irq_work_raise();

Because irq_work IRQ handler or timer IRQ handler may be executed just
before "llist_add(&work->llnode, irq_work_list)", so that, although
"empty == false", arch_irq_work_raise() still should be executed.

Can you tell me how to that with llist_empty()?


For ACCESS_ONCE, Mathiew suggest me to add it,

Hi, Mathiew,

Can you explain why ACCESS_ONCE should be used here?

Best Regards,
Huang Ying