Re: [PATCH -mm 4/4] irq_work, Use llist in irq_work

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

On 09/06/2011 06:57 PM, Mathieu Desnoyers wrote:
> * Huang Ying (ying.huang@intel.com) wrote:
>> Use llist in irq_work instead of the lock-less linked list
>> implementation in irq_work to avoid the code duplication.
>>
>> Signed-off-by: Huang Ying <ying.huang@intel.com>
>> Cc: Peter Zijlstra <peterz@infradead.org>
>> ---
>>  include/linux/irq_work.h |   15 ++++---
>>  kernel/irq_work.c        |   92 ++++++++++++++++-------------------------------
>>  2 files changed, 41 insertions(+), 66 deletions(-)
>>
> [...]
>> -static void __irq_work_queue(struct irq_work *entry)
>> +static void __irq_work_queue(struct irq_work *work)
>>  {
>> -	struct irq_work *next;
>> -
>>  	preempt_disable();
>>  
>> -	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, &__get_cpu_var(irq_work_list));
>>  	/* The list was empty, raise self-interrupt to start processing. */
>> -	if (!irq_work_next(entry))
>> +	if (!work->llnode.next)
> 
> 
> Hrm. What happens if this function gets delayed between llist_add and
> "if (!work->llnode.next)" ? It seems like the threads performing
> llist_del_all would be within its right to free the memory pointed to by
> work in the meantime.

Yes.  This is an issue.  This can be fixed in several way

1) use another flag to indicate whether list is empty
2) make llist_add return whether list is empty before adding
3) request irq_work users to free the memory with call_rcu() or after
synchronize_rcu().

Personally I prefer 1), which will not expose llist implementation
details too.

Best Regards,
Huang Ying