Re: [RFC] llist: Fix code comments about llist_del_first locking

[Huang\, Ying]

Joel Fernandes <joelaf@xxxxxxxxxx> writes:

> On Thu, Dec 8, 2016 at 4:42 PM, Joel Fernandes <joelaf@xxxxxxxxxx> wrote:
>> On Thu, Dec 8, 2016 at 4:35 PM, Huang, Ying <ying.huang@xxxxxxxxx> wrote:
>>> Joel Fernandes <joelaf@xxxxxxxxxx> writes:
>>>
>>>> Usage llist_del_first needs lock protection, however the table in the
>>>> comments of llist.h show a '-'. Correct this, and also add better
>>>> comments on top.
>>>>
>>>> Cc: Huang Ying <ying.huang@xxxxxxxxx>
>>>> Cc: Ingo Molnar <mingo@xxxxxxxxxx>
>>>> Cc: Will Deacon <will.deacon@xxxxxxx>
>>>> Cc: Paul McKenney <paulmck@xxxxxxxxxxxxxxxxxx>
>>>> Signed-off-by: Joel Fernandes <joelaf@xxxxxxxxxx>
>>>> ---
>>>>  include/linux/llist.h | 19 ++++++++++---------
>>>>  1 file changed, 10 insertions(+), 9 deletions(-)
>>>>
>>>> diff --git a/include/linux/llist.h b/include/linux/llist.h
>>>> index fd4ca0b..15e4949 100644
>>>> --- a/include/linux/llist.h
>>>> +++ b/include/linux/llist.h
>>>> @@ -3,14 +3,15 @@
>>>>  /*
>>>>   * Lock-less NULL terminated single linked list
>>>>   *
>>>> - * If there are multiple producers and multiple consumers, llist_add
>>>> - * can be used in producers and llist_del_all can be used in
>>>> - * consumers.  They can work simultaneously without lock.  But
>>>> - * llist_del_first can not be used here.  Because llist_del_first
>>>> - * depends on list->first->next does not changed if list->first is not
>>>> - * changed during its operation, but llist_del_first, llist_add,
>>>> - * llist_add (or llist_del_all, llist_add, llist_add) sequence in
>>>> - * another consumer may violate that.
>>>> + * If there are multiple producers and multiple consumers, llist_add can be
>>>> + * used in producers and llist_del_all can be used in consumers.  They can work
>>>> + * simultaneously without lock.  But llist_del_first will need to use a lock
>>>> + * with any other operation (ABA problem).  This is because llist_del_first
>>>> + * depends on list->first->next not changing but there's no way to be sure
>>>> + * about that and the cmpxchg in llist_del_first may succeed if list->first is
>>>> + * the same after concurrent operations. For example, a llist_del_first,
>>>> + * llist_add, llist_add (or llist_del_all, llist_add, llist_add) sequence in
>>>> + * another consumer may cause violations.
>>>>   *
>>>>   * If there are multiple producers and one consumer, llist_add can be
>>>>   * used in producers and llist_del_all or llist_del_first can be used
>>>> @@ -19,7 +20,7 @@
>>>>   * This can be summarized as follow:
>>>>   *
>>>>   *           |   add    | del_first |  del_all
>>>> - * add       |    -     |     -     |     -
>>>> + * add       |    -     |     L     |     -
>>>
>>> If there are only one consumer which only calls llist_del_first(), lock
>>> is unnecessary.  So '-' is shown here originally.  But if there are
>>> multiple consumers which call llist_del_first() or llist_del_all(), lock
>>> is needed.
>>
>> I think this needs to be made more clear in the table. The table
>> doesn't clear say whether it describes the preceding paragraph
>> (multiple producers and one consumer), or if it describes the multiple
>> producers and one consumer case. So either we should have 2 tables, or
>
> Sorry, I meant "or if it describes the multiple producer and multiple
> consumer case".

I tried to describe both cases in the original table.

  *           |   add    | del_first |  del_all
  * add       |    -     |     -     |     -
  * del_first |          |     L     |     L
  * del_all   |          |           |     -

The 'L' for "del_first * del_first" means multiple consumers uses
llist_del_first() need lock.  And the 'L' for 'del_first * del_all'
means multiple consumers uses llist_del_first() and llist_del_all() need
lock.

Best Regards,
Huang, Ying