[PATCH] netfront: Lockdep fixes

[PATCH] netfront: Lockdep fixes

[Jeremy Fitzhardinge]

netfront contains two locking problems found by lockdep:

1. rx_lock is a normal spinlock, and tx_lock is an irq spinlock.  This
   means that in normal use, tx_lock may be taken by an interrupt routine
   while rx_lock is held.  However, netif_disconnect_backend takes them
   in the order tx_lock->rx_lock, which could lead to a deadlock.  Reverse
   them.
2. rx_lock can also be used in softirq context, so it should be taken/released
   with spin_(un)lock_bh.

Signed-off-by: Jeremy Fitzhardinge <jeremy@xensource.com>
Cc: Chris Wright <chrisw@sous-sol.org>
Cc: Christian Limpach <Christian.Limpach@cl.cam.ac.uk>

diff -r a69029562c74 linux-2.6-xen-sparse/drivers/xen/netfront/netfront.c
--- a/linux-2.6-xen-sparse/drivers/xen/netfront/netfront.c	Thu Apr 12 14:13:33 2007 -0700
+++ b/linux-2.6-xen-sparse/drivers/xen/netfront/netfront.c	Thu Apr 12 14:19:42 2007 -0700
@@ -622,14 +622,14 @@ static int network_open(struct net_devic
 
 	memset(&np->stats, 0, sizeof(np->stats));
 
-	spin_lock(&np->rx_lock);
+	spin_lock_bh(&np->rx_lock);
 	if (netfront_carrier_ok(np)) {
 		network_alloc_rx_buffers(dev);
 		np->rx.sring->rsp_event = np->rx.rsp_cons + 1;
 		if (RING_HAS_UNCONSUMED_RESPONSES(&np->rx))
 			netif_rx_schedule(dev);
 	}
-	spin_unlock(&np->rx_lock);
+	spin_unlock_bh(&np->rx_lock);
 
 	network_maybe_wake_tx(dev);
 
@@ -1307,10 +1307,10 @@ static int netif_poll(struct net_device 
 	int pages_flipped = 0;
 	int err;
 
-	spin_lock(&np->rx_lock);
+	spin_lock_bh(&np->rx_lock);
 
 	if (unlikely(!netfront_carrier_ok(np))) {
-		spin_unlock(&np->rx_lock);
+		spin_unlock_bh(&np->rx_lock);
 		return 0;
 	}
 
@@ -1478,7 +1478,7 @@ err:
 		local_irq_restore(flags);
 	}
 
-	spin_unlock(&np->rx_lock);
+	spin_unlock_bh(&np->rx_lock);
 
 	return more_to_do;
 }
@@ -1520,7 +1520,7 @@ static void netif_release_rx_bufs(struct
 
 	skb_queue_head_init(&free_list);
 
-	spin_lock(&np->rx_lock);
+	spin_lock_bh(&np->rx_lock);
 
 	for (id = 0; id < NET_RX_RING_SIZE; id++) {
 		if ((ref = np->grant_rx_ref[id]) == GRANT_INVALID_REF) {
@@ -1588,7 +1588,7 @@ static void netif_release_rx_bufs(struct
 	while ((skb = __skb_dequeue(&free_list)) != NULL)
 		dev_kfree_skb(skb);
 
-	spin_unlock(&np->rx_lock);
+	spin_unlock_bh(&np->rx_lock);
 }
 
 static int network_close(struct net_device *dev)
@@ -1708,8 +1708,8 @@ static int network_connect(struct net_de
 	IPRINTK("device %s has %sing receive path.\n",
 		dev->name, np->copying_receiver ? "copy" : "flipp");
 
+	spin_lock_bh(&np->rx_lock);
 	spin_lock_irq(&np->tx_lock);
-	spin_lock(&np->rx_lock);
 
 	/*
 	 * Recovery procedure:
@@ -1761,7 +1761,7 @@ static int network_connect(struct net_de
 	network_tx_buf_gc(dev);
 	network_alloc_rx_buffers(dev);
 
-	spin_unlock(&np->rx_lock);
+	spin_unlock_bh(&np->rx_lock);
 	spin_unlock_irq(&np->tx_lock);
 
 	return 0;
@@ -1818,7 +1818,7 @@ static ssize_t store_rxbuf_min(struct cl
 	if (target > RX_MAX_TARGET)
 		target = RX_MAX_TARGET;
 
-	spin_lock(&np->rx_lock);
+	spin_lock_bh(&np->rx_lock);
 	if (target > np->rx_max_target)
 		np->rx_max_target = target;
 	np->rx_min_target = target;
@@ -1827,7 +1827,7 @@ static ssize_t store_rxbuf_min(struct cl
 
 	network_alloc_rx_buffers(netdev);
 
-	spin_unlock(&np->rx_lock);
+	spin_unlock_bh(&np->rx_lock);
 	return len;
 }
 
@@ -1861,7 +1861,7 @@ static ssize_t store_rxbuf_max(struct cl
 	if (target > RX_MAX_TARGET)
 		target = RX_MAX_TARGET;
 
-	spin_lock(&np->rx_lock);
+	spin_lock_bh(&np->rx_lock);
 	if (target < np->rx_min_target)
 		np->rx_min_target = target;
 	np->rx_max_target = target;
@@ -1870,7 +1870,7 @@ static ssize_t store_rxbuf_max(struct cl
 
 	network_alloc_rx_buffers(netdev);
 
-	spin_unlock(&np->rx_lock);
+	spin_unlock_bh(&np->rx_lock);
 	return len;
 }
 
@@ -2033,10 +2033,10 @@ static void netif_disconnect_backend(str
 static void netif_disconnect_backend(struct netfront_info *info)
 {
 	/* Stop old i/f to prevent errors whilst we rebuild the state. */
+	spin_lock_bh(&info->rx_lock);
 	spin_lock_irq(&info->tx_lock);
-	spin_lock(&info->rx_lock);
 	netfront_carrier_off(info);
-	spin_unlock(&info->rx_lock);
+	spin_unlock_bh(&info->rx_lock);
 	spin_unlock_irq(&info->tx_lock);
 
 	if (info->irq)

Re: [PATCH] netfront: Lockdep fixes

[Keir Fraser]

On 12/4/07 23:58, "Jeremy Fitzhardinge" <jeremy@xensource.com> wrote:

> netfront contains two locking problems found by lockdep:

I want to port lockdep down into Xen at some point. It's sweet!

 -- Keir

Re: [PATCH] netfront: Lockdep fixes

[Herbert Xu]

Hi Jeremy:

Jeremy Fitzhardinge <jeremy@xensource.com> wrote:
> 
> @@ -1307,10 +1307,10 @@ static int netif_poll(struct net_device 
>        int pages_flipped = 0;
>        int err;
> 
> -       spin_lock(&np->rx_lock);
> +       spin_lock_bh(&np->rx_lock);

The ->poll method is guaranteed to be called with BH disabled so
this isn't necessary.

> @@ -1588,7 +1588,7 @@ static void netif_release_rx_bufs(struct
>        while ((skb = __skb_dequeue(&free_list)) != NULL)
>                dev_kfree_skb(skb);
> 
> -       spin_unlock(&np->rx_lock);
> +       spin_unlock_bh(&np->rx_lock);
> }

Just a minor nit.  This is normally called with BH disabled,
except from uninit so you could put a local_bh_disable there
instead.

> static int network_close(struct net_device *dev)
> @@ -1708,8 +1708,8 @@ static int network_connect(struct net_de
>        IPRINTK("device %s has %sing receive path.\n",
>                dev->name, np->copying_receiver ? "copy" : "flipp");
> 
> +       spin_lock_bh(&np->rx_lock);
>        spin_lock_irq(&np->tx_lock);
> -       spin_lock(&np->rx_lock);
> 
>        /*
>         * Recovery procedure:
> @@ -1761,7 +1761,7 @@ static int network_connect(struct net_de
>        network_tx_buf_gc(dev);
>        network_alloc_rx_buffers(dev);
> 
> -       spin_unlock(&np->rx_lock);
> +       spin_unlock_bh(&np->rx_lock);
>        spin_unlock_irq(&np->tx_lock);

You can't enable BH with IRQs disabled.  Besides, for the sake of
symmetry these two should be reversed.

 
> @@ -2033,10 +2033,10 @@ static void netif_disconnect_backend(str
> static void netif_disconnect_backend(struct netfront_info *info)
> {
>        /* Stop old i/f to prevent errors whilst we rebuild the state. */
> +       spin_lock_bh(&info->rx_lock);
>        spin_lock_irq(&info->tx_lock);
> -       spin_lock(&info->rx_lock);
>        netfront_carrier_off(info);
> -       spin_unlock(&info->rx_lock);
> +       spin_unlock_bh(&info->rx_lock);
>        spin_unlock_irq(&info->tx_lock);

Ditto.

Cheers,
-- 
Visit Openswan at http://www.openswan.org/
Email: Herbert Xu ~{PmV>HI~} <herbert@gondor.apana.org.au>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt

Re: [PATCH] netfront: Lockdep fixes

[Keir Fraser]

On 13/4/07 13:10, "Herbert Xu" <herbert@gondor.apana.org.au> wrote:

> The ->poll method is guaranteed to be called with BH disabled so
> this isn't necessary.

Thanks.

>> @@ -1588,7 +1588,7 @@ static void netif_release_rx_bufs(struct
> 
> Just a minor nit.  This is normally called with BH disabled,
> except from uninit so you could put a local_bh_disable there
> instead.

That function's *only* called from uninit()!

> You can't enable BH with IRQs disabled.  Besides, for the sake of
> symmetry these two should be reversed.

Ack, I should have spotted this one, and the ditto.

 -- Keir

Re: [PATCH] netfront: Lockdep fixes

[Jeremy Fitzhardinge]

Herbert Xu wrote:
> The ->poll method is guaranteed to be called with BH disabled so
> this isn't necessary.
>   

OK.  As you can probably tell, I took a fairly blunt search-and-replace
approach once I'd worked out what lockdep was complaining about.

>> @@ -1588,7 +1588,7 @@ static void netif_release_rx_bufs(struct
>>        while ((skb = __skb_dequeue(&free_list)) != NULL)
>>                dev_kfree_skb(skb);
>>
>> -       spin_unlock(&np->rx_lock);
>> +       spin_unlock_bh(&np->rx_lock);
>> }
>>     
>
> Just a minor nit.  This is normally called with BH disabled,
> except from uninit so you could put a local_bh_disable there
> instead.
>   

OK.

>> static int network_close(struct net_device *dev)
>> @@ -1708,8 +1708,8 @@ static int network_connect(struct net_de
>>        IPRINTK("device %s has %sing receive path.\n",
>>                dev->name, np->copying_receiver ? "copy" : "flipp");
>>
>> +       spin_lock_bh(&np->rx_lock);
>>        spin_lock_irq(&np->tx_lock);
>> -       spin_lock(&np->rx_lock);
>>
>>        /*
>>         * Recovery procedure:
>> @@ -1761,7 +1761,7 @@ static int network_connect(struct net_de
>>        network_tx_buf_gc(dev);
>>        network_alloc_rx_buffers(dev);
>>
>> -       spin_unlock(&np->rx_lock);
>> +       spin_unlock_bh(&np->rx_lock);
>>        spin_unlock_irq(&np->tx_lock);
>>     
>
> You can't enable BH with IRQs disabled.  Besides, for the sake of
> symmetry these two should be reversed.
>   

Will do.

Thanks for looking over it.

    J

Re: [PATCH] netfront: Lockdep fixes

[Herbert Xu]

On Fri, Apr 13, 2007 at 03:22:41PM +0100, Keir Fraser wrote:
> 
> >> @@ -1588,7 +1588,7 @@ static void netif_release_rx_bufs(struct
> > 
> > Just a minor nit.  This is normally called with BH disabled,
> > except from uninit so you could put a local_bh_disable there
> > instead.
> 
> That function's *only* called from uninit()!

Heh, don't know how I saw that one getting called elsewhere :)

Cheers,
-- 
Visit Openswan at http://www.openswan.org/
Email: Herbert Xu ~{PmV>HI~} <herbert@gondor.apana.org.au>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt