Re: [PATCH net] cxgb4: fix refcount init for TC-MQPRIO offload

[Rahul Lakkireddy <rahul.lakkireddy@chelsio.com>]

On Tuesday, December 12/17/19, 2019 at 14:09:30 -0800, David Miller wrote:
> From: Rahul Lakkireddy <rahul.lakkireddy@chelsio.com>
> Date: Mon, 16 Dec 2019 19:54:02 +0530
> 
> > @@ -205,7 +205,11 @@ static int cxgb4_mqprio_alloc_hw_resources(struct net_device *dev)
> >  			cxgb4_enable_rx(adap, &eorxq->rspq);
> >  	}
> >  
> > -	refcount_inc(&adap->tc_mqprio->refcnt);
> > +	if (!refcount_read(&adap->tc_mqprio->refcnt))
> > +		refcount_set(&adap->tc_mqprio->refcnt, 1);
> > +	else
> > +		refcount_inc(&adap->tc_mqprio->refcnt);
> > +
> 
> This is not correct.
> 
> You're bypassing the whole point of the refcount_t type which is to make sure
> that code that initializes a new object explicitly calls refcount_set() and
> that once the refcount goes to zero the object is freed or the refcount is
> initialized again using refcount_set() or similar.
> 
> I'm not applying this, it's just papering around the real problem.

Ok, I'll move refcount_set() closer to where the hardware queue
arrays, that it's tracking, have been allocated. Will send a v2.

Thanks,
Rahul