sparse: possible false report of context imbalance

sparse: possible false report of context imbalance

[Larry Finger]

Hi,

Sparse reports the following:

   CHECK   drivers/staging/rtl8188eu/core/rtw_mlme.c
drivers/staging/rtl8188eu/core/rtw_mlme.c:1003:9: warning: context imbalance in 
'rtw_free_assoc_resources' - different lock contexts for basic block

The code in question is as follows:

         if (lock_scanned_queue)
                 spin_lock_bh(&(pmlmepriv->scanned_queue.lock));

         pwlan = rtw_find_network(&pmlmepriv->scanned_queue, 
tgt_network->network.MacAddress);

         if (lock_scanned_queue)
                 spin_unlock_bh(&(pmlmepriv->scanned_queue.lock));

As this fragment uses the identical test to unlock that is used to lock, and the 
test variable is not touched, I think this is a false indication. I am using 
version 0.4.4 of sparse.

Thanks,

Larry

Re: sparse: possible false report of context imbalance

[Josh Triplett]

On Wed, Oct 16, 2013 at 10:23:56PM -0500, Larry Finger wrote:
> Sparse reports the following:
> 
>   CHECK   drivers/staging/rtl8188eu/core/rtw_mlme.c
> drivers/staging/rtl8188eu/core/rtw_mlme.c:1003:9: warning: context
> imbalance in 'rtw_free_assoc_resources' - different lock contexts
> for basic block
> 
> The code in question is as follows:
> 
>         if (lock_scanned_queue)
>                 spin_lock_bh(&(pmlmepriv->scanned_queue.lock));
> 
>         pwlan = rtw_find_network(&pmlmepriv->scanned_queue,
> tgt_network->network.MacAddress);
> 
>         if (lock_scanned_queue)
>                 spin_unlock_bh(&(pmlmepriv->scanned_queue.lock));
> 
> As this fragment uses the identical test to unlock that is used to
> lock, and the test variable is not touched, I think this is a false
> indication. I am using version 0.4.4 of sparse.

Sparse can't track conditional contexts like this; sparse intentionally
complains here that you're running the same basic block (the
rtw_find_network call) with and without a lock held.

The following workaround works when this is legitimate, though it isn't
ideal:

if (condition) {
	lock
	do_thing
	unlock
} else {
	do_thing
}

Ideally, Sparse should be able to track conditional contexts, but that
would require some form of abstract evaluation, or as a simplistic hack,
looking for identical side-effect-free conditional expressions.

- Josh Triplett

Re: sparse: possible false report of context imbalance

[Larry Finger]

On 10/17/2013 11:25 AM, Josh Triplett wrote:
> On Wed, Oct 16, 2013 at 10:23:56PM -0500, Larry Finger wrote:
>> Sparse reports the following:
>>
>>    CHECK   drivers/staging/rtl8188eu/core/rtw_mlme.c
>> drivers/staging/rtl8188eu/core/rtw_mlme.c:1003:9: warning: context
>> imbalance in 'rtw_free_assoc_resources' - different lock contexts
>> for basic block
>>
>> The code in question is as follows:
>>
>>          if (lock_scanned_queue)
>>                  spin_lock_bh(&(pmlmepriv->scanned_queue.lock));
>>
>>          pwlan = rtw_find_network(&pmlmepriv->scanned_queue,
>> tgt_network->network.MacAddress);
>>
>>          if (lock_scanned_queue)
>>                  spin_unlock_bh(&(pmlmepriv->scanned_queue.lock));
>>
>> As this fragment uses the identical test to unlock that is used to
>> lock, and the test variable is not touched, I think this is a false
>> indication. I am using version 0.4.4 of sparse.
>
> Sparse can't track conditional contexts like this; sparse intentionally
> complains here that you're running the same basic block (the
> rtw_find_network call) with and without a lock held.
>
> The following workaround works when this is legitimate, though it isn't
> ideal:
>
> if (condition) {
> 	lock
> 	do_thing
> 	unlock
> } else {
> 	do_thing
> }
>
> Ideally, Sparse should be able to track conditional contexts, but that
> would require some form of abstract evaluation, or as a simplistic hack,
> looking for identical side-effect-free conditional expressions.

Thanks for the answer, which is about what I expected. Sparse is a great tool, 
particularly for endian-related issues.

Larry