Re: [PATCH 4/9] firewire: don't use PREPARE_DELAYED_WORK

[Peter Hurley <peter@hurleysoftware.com>]

On 02/20/2014 09:13 PM, Tejun Heo wrote:
> On Thu, Feb 20, 2014 at 09:07:27PM -0500, Peter Hurley wrote:
>> On 02/20/2014 08:59 PM, Tejun Heo wrote:
>>> Hello,
>>>
>>> On Thu, Feb 20, 2014 at 08:44:46PM -0500, Peter Hurley wrote:
>>>>> +static void fw_device_workfn(struct work_struct *work)
>>>>> +{
>>>>> +	struct fw_device *device = container_of(to_delayed_work(work),
>>>>> +						struct fw_device, work);
>>>>
>>>> I think this needs an smp_rmb() here.
>>>
>>> The patch is equivalent transformation and the whole thing is
>>> guaranteed to have gone through pool->lock.  No explicit rmb
>>> necessary.
>>
>> The spin_unlock_irq(&pool->lock) only guarantees completion of
>> memory operations _before_ the unlock; memory operations which occur
>> _after_ the unlock may be speculated before the unlock.
>>
>> IOW, unlock is not a memory barrier for operations that occur after.
>
> It's not just unlock.  It's lock / unlock pair on the same lock from
> both sides.  Nothing can sip through that.

CPU 0                            | CPU 1
                                  |
  INIT_WORK(fw_device_workfn)     |
                                  |
  workfn = funcA                  |
  queue_work_on()                 |
  .                               | process_one_work()
  .                               |   ..
  .                               |   worker->current_func = work->func
  .                               |
  .                               |   speculative load of workfn = funcA
  .                               |   .
  workfn = funcB                  |   .
  queue_work_on()                 |   .
    local_irq_save()              |   .
    test_and_set_bit() == 1       |   .
                                  |   set_work_pool_and_clear_pending()
    work is not queued            |     smp_wmb
     funcB never runs             |     set_work_data()
                                  |       atomic_set()
                                  |   spin_unlock_irq()
                                  |
                                  |   worker->current_func(work)  @ fw_device_workfn
                                  |      workfn()  @ funcA


The speculative load of workfn on CPU 1 is valid because no rmb will occur
between the load and the execution of workfn() on CPU 1.

Thus funcB will never execute because, in this circumstance, a second
worker is not queued (because PENDING had not yet been cleared).

Regards,
Peter Hurley