Re: [PATCH] rcu: fix possible bug in include/linux/pagemap.h

Re: [PATCH] rcu: fix possible bug in include/linux/pagemap.h

[Lai Jiangshan]

On 04/25/2011 09:48 PM, Paul E. McKenney wrote:
> On Mon, Apr 25, 2011 at 04:20:37PM +0800, Lai Jiangshan wrote:
>> I read the code and comments written by Nick Piggin(mainly from e286781d),
>> page_cache_get_speculative() is protected by rcu_read_lock(),
>> it may be preempted when preemptible RCU, so we must use get_page_unless_zero()
>> in this situation.
>>
>> In the days of e286781d, we only have CLASSIC_RCU and (old)PREEMPT_RCU,
>> so "defined(CONFIG_CLASSIC_RCU)" means non-preemptible RCU and the code is correct.
>>
>> In the days of b560d8ad, we only have TREE_RCU and TREE_PREEMPT_RCU,
>> so "defined(CONFIG_TREE_RCU)" means non-preemptible RCU and the code is correct.
>>
>> But in nowadays, we have TREE_RCU, TREE_PREEMPT_RCU, TINY_RCU and TINY_PREEMPT_RCU,
>> so the "defined(CONFIG_TREE_RCU)" for non-preemptible RCU is incorrect, and it may causes bugs.
>> we should use "!defined(CONFIG_PREEMPT_RCU)" for non-preemptible RCU code block.
>>
>> CC: Nick Piggin <npiggin@kernel.dk>
>> Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
>> Reported-by: WANG Cong <amwang@redhat.com>
>> Signed-off-by: Lai Jiangshan <laijs@cn.fujitsu.com>
> 
> Good catch!!!
> 
> Acked-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
> 


Sorry, I forgot to CC it to LKML. Added it.


Hi, Nick Piggin,

Could you review it?

Thanks,
Lai

Re: [PATCH] rcu: fix possible bug in include/linux/pagemap.h

[Américo Wang]

On Wed, Apr 27, 2011 at 12:03 PM, Lai Jiangshan <laijs@cn.fujitsu.com> wrote:
> On 04/25/2011 09:48 PM, Paul E. McKenney wrote:
>> On Mon, Apr 25, 2011 at 04:20:37PM +0800, Lai Jiangshan wrote:
>>> I read the code and comments written by Nick Piggin(mainly from e286781d),
>>> page_cache_get_speculative() is protected by rcu_read_lock(),
>>> it may be preempted when preemptible RCU, so we must use get_page_unless_zero()
>>> in this situation.
>>>
>>> In the days of e286781d, we only have CLASSIC_RCU and (old)PREEMPT_RCU,
>>> so "defined(CONFIG_CLASSIC_RCU)" means non-preemptible RCU and the code is correct.
>>>
>>> In the days of b560d8ad, we only have TREE_RCU and TREE_PREEMPT_RCU,
>>> so "defined(CONFIG_TREE_RCU)" means non-preemptible RCU and the code is correct.
>>>
>>> But in nowadays, we have TREE_RCU, TREE_PREEMPT_RCU, TINY_RCU and TINY_PREEMPT_RCU,
>>> so the "defined(CONFIG_TREE_RCU)" for non-preemptible RCU is incorrect, and it may causes bugs.
>>> we should use "!defined(CONFIG_PREEMPT_RCU)" for non-preemptible RCU code block.
>>>
>>> CC: Nick Piggin <npiggin@kernel.dk>
>>> Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
>>> Reported-by: WANG Cong <amwang@redhat.com>
>>> Signed-off-by: Lai Jiangshan <laijs@cn.fujitsu.com>
>>
>> Good catch!!!
>>
>> Acked-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
>>
>
>
> Sorry, I forgot to CC it to LKML. Added it.
>

Thanks for fixing this!

The description explains quite well, so I am okay with this patch,

Reviewed-by: WANG Cong <xiyou.wangcong@gmail.com>

BTW, you added my company email in Reported-by, but Cc'ed my gmail. ;-)