Re: [PATCH net-next v4 1/6] bonding: simplify and use RCU protection for 3ad xmit path

[Ding Tianhong <dthxman@gmail.com>]

于 2013/9/7 23:03, Veaceslav Falico 写道:
> On Sat, Sep 07, 2013 at 04:45:05PM +0200, Nikolay Aleksandrov wrote:
>>
>> On 09/07/2013 04:20 PM, Veaceslav Falico wrote:
>>> On Fri, Sep 06, 2013 at 03:28:07PM +0800, Ding Tianhong wrote:
> ...snip...
>>> diff --git a/include/linux/rculist.h b/include/linux/rculist.h
>>> index f4b1001..37b49d1 100644
>>> --- a/include/linux/rculist.h
>>> +++ b/include/linux/rculist.h
>>> @@ -23,6 +23,7 @@
>>> * way, we must not access it directly
>>> */
>>> #define list_next_rcu(list) (*((struct list_head __rcu
>>> **)(&(list)->next)))
>>> +#define list_prev_rcu(list) (*((struct list_head __rcu
>>> **)(&(list)->prev)))
>>>
>>> /*
>>> * Insert a new entry between two known consecutive entries.
>>> @@ -271,6 +272,12 @@ static inline void list_splice_init_rcu(struct
>>> list_head *list,
>>> likely(__ptr != __next) ? container_of(__next, type, member) : NULL; \
>>> })
>>>
>>> +#define list_last_or_null_rcu(ptr, type, member) \
>>> + ({struct list_head *__ptr = (ptr); \
>>> + struct list_head __rcu *__last = list_prev_rcu(__ptr); \
>>> + likely(__ptr != __last) ? container_of(__prev, type, member) : 
>>> NULL; \
>>> + })
>>> +
>> Hi,
>> Actually I don't think you can dereference ->prev and use the standard
>> list_del_rcu because it guarantees only the ->next ptr will be valid and
>> ->prev is set to LIST_POISON2.
>> IMO, you'll need something like this: 
>> https://lkml.org/lkml/2012/7/25/193
>> with the bidir_del and all that.
>
> Yeah, right, my bad - we can rely only on the ->next pointer, indeed,
> missed that part. RCU is hard :).
>
> So it'll be a lot harder to implement bond_last_slave_rcu() in a
> 'straightforward' approach.
>
> I'd rather go in the opposite direction here - i.e. drop the 'reverse'
> traversal completely, and all the use cases for bond_last_slave_rcu(). 
> I've
> got some patches already - http://patchwork.ozlabs.org/patch/272076/ 
> doing
> that, and hopefully will remove the whole 'backword' traversal completely
> in the future.
>

Although RCU is truely difficult to understand, but it is very 
interesting and beautiful,
I will follow your footsteps and finish it.

>>
>> But in any case I complete agree with Veaceslav here. Read all the
>> documentation carefully :-)
>>
>> Cheers,
>> Nik
>>

good lession, read it complete and agreed, it is a hard work to do.
I need more time to review the Veaceslav's patchset and improve my
thought, thanks Nik and Veaceslav.

Best regards
Ding

>>> /**
>>> * list_for_each_entry_rcu - iterate over rcu list of given type
>>> * @pos: the type * to use as a loop cursor.
>>> ------- END OF PATCH ------
>>>
>>> Anyway, it's up to you.
>>>
>>> Hope that helps.
>>
> -- 
> To unsubscribe from this list: send the line "unsubscribe netdev" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at http://vger.kernel.org/majordomo-info.html
>