Subject: Re: Kernel WARNING: at net/core/dev.c:1330
	__netif_schedule+0x2c/0x98()
From: Peter Zijlstra <peterz@infradead.org>
To: Jarek Poplawski <jarkao2@gmail.com>
Cc: David Miller <davem@davemloft.net>, Larry.Finger@lwfinger.net,
       kaber@trash.net, torvalds@linux-foundation.org,
       akpm@linux-foundation.org, netdev@vger.kernel.org,
       linux-kernel@vger.kernel.org, linux-wireless@vger.kernel.org,
       mingo@redhat.com
In-Reply-To: <20080723085452.GB4561@ff.dom.local>
References: <20080722.160409.216536011.davem@davemloft.net>
	 <20080723062036.GA4561@ff.dom.local>
	 <20080723.005921.113868915.davem@davemloft.net>
	 <20080723085452.GB4561@ff.dom.local>
Content-Type: text/plain
Date: Wed, 23 Jul 2008 11:03:06 +0200
Message-Id: <1216803786.7257.146.camel@twins>
Mime-Version: 1.0
Content-Transfer-Encoding: 7bit
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 2149
Lines: 64

On Wed, 2008-07-23 at 08:54 +0000, Jarek Poplawski wrote:
> On Wed, Jul 23, 2008 at 12:59:21AM -0700, David Miller wrote:
> > From: Jarek Poplawski <jarkao2@gmail.com>
> > Date: Wed, 23 Jul 2008 06:20:36 +0000
> > 
> > > PS: if there is nothing new in lockdep the classical method would
> > > be to change this static array:
> > > 
> > > static struct lock_class_key
> > > netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)];
> > > 
> > > to
> > > 
> > > static struct lock_class_key
> > > netdev_xmit_lock_key[ARRAY_SIZE(netdev_lock_type)][MAX_NUM_TX_QUEUES];
> > > 
> > > and set lockdep classes per queue as well. (If we are sure we don't
> > > need lockdep subclasses anywhere this could be optimized by using
> > > one lock_class_key per 8 queues and spin_lock_nested()).
> > 
> > Unfortunately MAX_NUM_TX_QUEUES is USHORT_MAX, so this isn't really
> > a feasible approach.
> 
> Is it used by real devices already? Maybe for the beginning we could
> start with something less?
> 
> > spin_lock_nested() isn't all that viable either, as the subclass
> > limit is something like 8.
> 
> This method would need to do some additional counting: depending of
> a queue number each 8 subsequent queues share (are set to) the same
> class and their number mod 8 gives the subqueue number for
> spin_lock_nested().
> 
> I'll try to find if there is something new around this in lockdep.
> (lockdep people added to CC.)

There isn't.

Is there a static data structure that the driver needs to instantiate to
'create' a queue? Something like:

/* this imaginary e1000 hardware has 16 hardware queues */
static struct net_tx_queue e1000e_tx_queues[16]; 

In that case you can stick the key in there and do:

int e1000e_init_tx_queue(struct net_tx_queue *txq)
{
	...

	spin_lock_init(&txq->tx_lock);
	lockdep_set_class(&txq->tx_lock, &txq->tx_lock_key);

	...
}

( This is what the scheduler runqueues also do )

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/