Using nested locking for spin_lock_bh

Using nested locking for spin_lock_bh

[Sven Eckelmann]

[-- Attachment #1: Type: text/plain, Size: 810 bytes --]

Hi,

I started to run some experimental batman-adv code inside some virtual 
machines with different debugging functions enabled. One of them was lockdep. 
There we noticed a false positive for a lock class that is actually used in 
two different data structures. Therefore, lockdep noticed that there could be 
a deadlock for this class in a function that locks the two different data 
structures.

To resolve this problem, I wanted to define two different subclasses and 
replace the spin_lock_bh with its nested version. The only problem now is the 
absence of spin_lock_bh_nested. I could only find spin_lock_bh and 
spin_lock_irqsave_nested.

Is there another way how this should be done or is there a general problem why 
there is no nested support for this incarnation of spin_lock?

Kind regards,
	Sven

[-- Attachment #2: This is a digitally signed message part. --]
[-- Type: application/pgp-signature, Size: 836 bytes --]

Re: Using nested locking for spin_lock_bh

[Yong Zhang]

On Wed, Mar 28, 2012 at 06:36:39PM +0200, Sven Eckelmann wrote:
> Hi,
> 
> I started to run some experimental batman-adv code inside some virtual 
> machines with different debugging functions enabled. One of them was lockdep. 
> There we noticed a false positive for a lock class that is actually used in 
> two different data structures.

Hmm, I think you can initialize the lock class in the two differnent data
structures with differnt lock_class_key.

Thanks,
Yong

> Therefore, lockdep noticed that there could be 
> a deadlock for this class in a function that locks the two different data 
> structures.
> 
> To resolve this problem, I wanted to define two different subclasses and 
> replace the spin_lock_bh with its nested version. The only problem now is the 
> absence of spin_lock_bh_nested. I could only find spin_lock_bh and 
> spin_lock_irqsave_nested.
> 
> Is there another way how this should be done or is there a general problem why 
> there is no nested support for this incarnation of spin_lock?
> 
> Kind regards,
> 	Sven



-- 
Only stand for myself

Re: Using nested locking for spin_lock_bh

[Sven Eckelmann]

[-- Attachment #1: Type: text/plain, Size: 664 bytes --]

On Thursday, March 29, 2012 11:11:37 AM Yong Zhang wrote:
> On Wed, Mar 28, 2012 at 06:36:39PM +0200, Sven Eckelmann wrote:
> > Hi,
> > 
> > I started to run some experimental batman-adv code inside some virtual
> > machines with different debugging functions enabled. One of them was
> > lockdep. There we noticed a false positive for a lock class that is
> > actually used in two different data structures.
> 
> Hmm, I think you can initialize the lock class in the two differnent data
> structures with differnt lock_class_key.

Thanks. I think I understand the problem now a little bit better and could 
resolve it using lockdep_set_class.

Kind regards,
	Sven

[-- Attachment #2: This is a digitally signed message part. --]
[-- Type: application/pgp-signature, Size: 836 bytes --]

Re: Using nested locking for spin_lock_bh

[Peter Zijlstra]

On Wed, 2012-03-28 at 18:36 +0200, Sven Eckelmann wrote:
> 
> Is there another way how this should be done or is there a general problem why 
> there is no nested support for this incarnation of spin_lock? 

No reason other than that it wasn't needed before now, but I see you've
resolved things with a lockdep_set_class() which works too.