Re: [PATCH] rcu: tree_stall: Correctly unlock root node in rcu_check_gp_start_stall

[Neeraj Upadhyay <neeraju@codeaurora.org>]

On 3/30/19 2:57 AM, Paul E. McKenney wrote:
> On Fri, Mar 29, 2019 at 07:52:15PM +0530, Neeraj Upadhyay wrote:
>>
>>
>> On 3/29/19 6:58 PM, Mukesh Ojha wrote:
>>>
>>> On 3/29/2019 4:57 PM, Neeraj Upadhyay wrote:
>>>> Only unlock the root node, if current node (rnp) is not
>>>> root node.
>>>>
>>>> Signed-off-by: Neeraj Upadhyay <neeraju@codeaurora.org>
>>>
>>>
>>>
>>>> ---
>>>>    kernel/rcu/tree_stall.h | 4 +++-
>>>>    1 file changed, 3 insertions(+), 1 deletion(-)
>>>>
>>>> diff --git a/kernel/rcu/tree_stall.h b/kernel/rcu/tree_stall.h
>>>> index f65a73a..0651833 100644
>>>> --- a/kernel/rcu/tree_stall.h
>>>> +++ b/kernel/rcu/tree_stall.h
>>>
>>>
>>> why this is showing as under tree_stall.h while it is under
>>> "kernel/rcu/tree.c"
>>
>> It's moved in https://git.kernel.org/pub/scm/linux/kernel/git/paulmck/linux-rcu.git/commit/?h=dev&id=10462d6f58fb6dbde7563e9343505d98d5bfba3d
>>
>> Please see linux-rcu dev tree for other changes, which moves code to
>> this file.
>>
>>
>> Thanks
>> Neeraj
>>
>>>
>>>> @@ -630,7 +630,9 @@ static void rcu_check_gp_start_stall(struct
>>>> rcu_node *rnp, struct rcu_data *rdp,
>>>>            time_before(j, rcu_state.gp_req_activity + gpssdelay) ||
>>>>            time_before(j, rcu_state.gp_activity + gpssdelay) ||
>>>>            atomic_xchg(&warned, 1)) {
>>>> -        raw_spin_unlock_rcu_node(rnp_root); /* irqs remain disabled. */
>>>> +        if (rnp_root != rnp)
>>>> +            /* irqs remain disabled. */
>>>> +            raw_spin_unlock_rcu_node(rnp_root);
>>>
>>> Looks good as it will balance the lock .if it is the root_node,
>>> which was not there earlier, and unlock was happening without any
>>> lock on root.
>>>
>>> Reviewed-by: Mukesh Ojha <mojha@codeaurora.org>
> 
> Applied, again thank you both!
> 
> In both cases, I updated the commit log, so please check to make sure
> that I didn't mess anything up.
> 
> 							Thanx, Paul
> 

Thanks Paul. One minor comment on 
https://git.kernel.org/pub/scm/linux/kernel/git/paulmck/linux-rcu.git/commit/?h=dev&id=ec6530e763046b6bb1f4c2c2aed49ebc68aae2a0 


"it clearly does not make sense to release
both rnp->lock and rnp->lock"

should be rnp->lock and rnp_root->lock



Thanks
Neeraj

>>> Cheers,
>>> -Mukesh
>>>
>>>>            raw_spin_unlock_irqrestore_rcu_node(rnp, flags);
>>>>            return;
>>>>        }
>>
>> -- 
>> QUALCOMM INDIA, on behalf of Qualcomm Innovation Center, Inc. is a
>> member of the Code Aurora Forum, hosted by The Linux Foundation
>>
> 

-- 
QUALCOMM INDIA, on behalf of Qualcomm Innovation Center, Inc. is a
member of the Code Aurora Forum, hosted by The Linux Foundation