Skip to contentSkip to conversation

: sunvnet netdev ops conversion...

PATCH In review David Miller 5 participants 7 messages Mar 20, 2009

Subsystems (2)
lore
Share & tools
lore

David Miller

author maintainer
#0 lore raw

PATCH 1/1 : sunvnet netdev ops conversion...

Committed to net-next-2.6...

sunvnet: Convert to net_device_ops.

---
 drivers/net/sunvnet.c |   18 +++++++++++-------
 1 files changed, 11 insertions(+), 7 deletions(-)
drivers/net/sunvnet.c +11 -7
index 611230f..a82fb2a 100644
@@ -1012,6 +1012,16 @@ err_out:
 static LIST_HEAD(vnet_list);
 static DEFINE_MUTEX(vnet_list_mutex);
 
+static const struct net_device_ops vnet_ops = {
+	.ndo_open		= vnet_open,
+	.ndo_stop		= vnet_close,
+	.ndo_set_multicast_list	= vnet_set_rx_mode,
+	.ndo_set_mac_address	= vnet_set_mac_addr,
+	.ndo_tx_timeout		= vnet_tx_timeout,
+	.ndo_change_mtu		= vnet_change_mtu,
+	.ndo_start_xmit		= vnet_start_xmit,
+};
+
 static struct vnet * __devinit vnet_new(const u64 *local_mac)
 {
 	struct net_device *dev;
@@ -1040,15 +1050,9 @@ static struct vnet * __devinit vnet_new(const u64 *local_mac)
 	INIT_LIST_HEAD(&vp->list);
 	vp->local_mac = *local_mac;
 
-	dev->open = vnet_open;
-	dev->stop = vnet_close;
-	dev->set_multicast_list = vnet_set_rx_mode;
-	dev->set_mac_address = vnet_set_mac_addr;
-	dev->tx_timeout = vnet_tx_timeout;
+	dev->netdev_ops = &vnet_ops;
 	dev->ethtool_ops = &vnet_ethtool_ops;
 	dev->watchdog_timeo = VNET_TX_TIMEOUT;
-	dev->change_mtu = vnet_change_mtu;
-	dev->hard_start_xmit = vnet_start_xmit;
 
 	err = register_netdev(dev);
 	if (err) {
Signature · 1 line
1.6.2.1.222.g570cc
#1 lore raw
Quoted 3 lines · David Miller

Committed to net-next-2.6...

sunvnet: Convert to net_device_ops.

I noticed that there are other drivers's net_device_ops need to be converted,is these convert nessary?

Quoted 44 lines
---
 drivers/net/sunvnet.c |   18 +++++++++++-------
 1 files changed, 11 insertions(+), 7 deletions(-)
drivers/net/sunvnet.c +11 -7
index 611230f..a82fb2a 100644
@@ -1012,6 +1012,16 @@ err_out:
 static LIST_HEAD(vnet_list);
 static DEFINE_MUTEX(vnet_list_mutex);
 
+static const struct net_device_ops vnet_ops = {
+	.ndo_open		= vnet_open,
+	.ndo_stop		= vnet_close,
+	.ndo_set_multicast_list	= vnet_set_rx_mode,
+	.ndo_set_mac_address	= vnet_set_mac_addr,
+	.ndo_tx_timeout		= vnet_tx_timeout,
+	.ndo_change_mtu		= vnet_change_mtu,
+	.ndo_start_xmit		= vnet_start_xmit,
+};
+
 static struct vnet * __devinit vnet_new(const u64 *local_mac)
 {
 	struct net_device *dev;
@@ -1040,15 +1050,9 @@ static struct vnet * __devinit vnet_new(const u64 *local_mac)
 	INIT_LIST_HEAD(&vp->list);
 	vp->local_mac = *local_mac;
 
-	dev->open = vnet_open;
-	dev->stop = vnet_close;
-	dev->set_multicast_list = vnet_set_rx_mode;
-	dev->set_mac_address = vnet_set_mac_addr;
-	dev->tx_timeout = vnet_tx_timeout;
+	dev->netdev_ops = &vnet_ops;
 	dev->ethtool_ops = &vnet_ethtool_ops;
 	dev->watchdog_timeo = VNET_TX_TIMEOUT;
-	dev->change_mtu = vnet_change_mtu;
-	dev->hard_start_xmit = vnet_start_xmit;
 
 	err = register_netdev(dev);
 	if (err) {
Signature · 2 lines
Regards
Yang Hongyang
author maintainer
#2 lore raw

From: Yang Hongyang <[email protected]>
Date: Fri, 20 Mar 2009 16:30:59 +0800

Quoted 7 lines
Quoted 3 lines · David Miller

Committed to net-next-2.6...

sunvnet: Convert to net_device_ops.

I noticed that there are other drivers's net_device_ops need to be converted,is these convert nessary?

Not necessary, but desirable. Once all drivers are converted we can remove the compatability code.

#3 lore raw
Quoted 6 lines · David Miller

From: Yang Hongyang <[email protected]>
Date: Fri, 20 Mar 2009 16:30:59 +0800

Quoted 3 lines
Quoted 2 lines · David Miller

Committed to net-next-2.6...

Quoted 7 lines
Quoted 4 lines
Quoted 1 line

sunvnet: Convert to net_device_ops.

I noticed that there are other drivers's net_device_ops need to be converted,is these convert nessary?

Not necessary, but desirable. Once all drivers are converted we can remove the compatability code.

Just thinking... Would it make sense, at some point before removing the compatibility layer, to add tests for partial conversions? I noticed the sungem bug just fixed by Stephen Hemminger
( http://permalink.gmane.org/gmane.linux.network/122756 ) which could have been caught by the patch below.

This will also ease the transition of "driver sets", like usbnet and it's children, where a number of drivers need to make the conversion together. Don't know if that's a big problem, but it did stop my first attempt to convert usbnet because I was uncertain whether I had caught all of the necessary mini driver conversions.

---

 net/core/dev.c |   28 ++++++++++++++++++++++++++++
 1 files changed, 28 insertions(+), 0 deletions(-)
net/core/dev.c +28 -0
index f112970..ad3e780 100644
@@ -4359,6 +4359,34 @@ int register_netdevice(struct net_device *dev)
 	 * This is temporary until all network devices are converted.
 	 */
 	if (dev->netdev_ops) {
+		/* drivers using netdev_ops should not set any of these */
+		if (dev->init ||
+		    dev->uninit ||
+		    dev->open ||
+		    dev->change_rx_flags ||
+		    dev->set_rx_mode ||
+		    dev->set_multicast_list ||
+		    dev->set_mac_address ||
+		    dev->validate_addr ||
+		    dev->do_ioctl ||
+		    dev->set_config ||
+		    dev->change_mtu ||
+		    dev->neigh_setup ||
+		    dev->tx_timeout ||
+		    dev->get_stats ||
+		    dev->vlan_rx_register ||
+		    dev->vlan_rx_add_vid ||
+		    dev->vlan_rx_kill_vid
+#ifdef CONFIG_NET_POLL_CONTROLLER
+		    || dev->poll_controller
+#endif
+			) {
+			char drivername[64];
+
+			pr_info("%s (%s): mixing netdev_ops and old API\n",
+				dev->name, netdev_drivername(dev, drivername, 
+							     64));
+		}
 		netdev_resync_ops(dev);
 	} else {
 		char drivername[64];
#4 lore raw
Quoted 63 lines · Fri, Mar 20, 2009 at 12:43:00PM CET, [email protected] wrote:
Quoted 6 lines · David Miller

From: Yang Hongyang <[email protected]>
Date: Fri, 20 Mar 2009 16:30:59 +0800

Quoted 3 lines
Quoted 2 lines · David Miller

Committed to net-next-2.6...

Quoted 7 lines
Quoted 4 lines
Quoted 1 line

sunvnet: Convert to net_device_ops.

I noticed that there are other drivers's net_device_ops need to be converted,is these convert nessary?

Not necessary, but desirable. Once all drivers are converted we can remove the compatability code.

Just thinking... Would it make sense, at some point before removing the compatibility layer, to add tests for partial conversions? I noticed the sungem bug just fixed by Stephen Hemminger
( http://permalink.gmane.org/gmane.linux.network/122756 ) which could have been caught by the patch below.

This will also ease the transition of "driver sets", like usbnet and it's children, where a number of drivers need to make the conversion together. Don't know if that's a big problem, but it did stop my first attempt to convert usbnet because I was uncertain whether I had caught all of the necessary mini driver conversions.

---

net/core/dev.c | 28 ++++++++++++++++++++++++++++

1 files changed, 28 insertions(+), 0 deletions(-)
net/core/dev.c +21 -0
index f112970..ad3e780 100644
@@ -4359,6 +4359,34 @@ int register_netdevice(struct net_device *dev)
 	 * This is temporary until all network devices are converted.
 	 */
 	if (dev->netdev_ops) {
+		/* drivers using netdev_ops should not set any of these */
+		if (dev->init ||
+		    dev->uninit ||
+		    dev->open ||
+		    dev->change_rx_flags ||
+		    dev->set_rx_mode ||
+		    dev->set_multicast_list ||
+		    dev->set_mac_address ||
+		    dev->validate_addr ||
+		    dev->do_ioctl ||
+		    dev->set_config ||
+		    dev->change_mtu ||
+		    dev->neigh_setup ||
+		    dev->tx_timeout ||
+		    dev->get_stats ||
+		    dev->vlan_rx_register ||
+		    dev->vlan_rx_add_vid ||
+		    dev->vlan_rx_kill_vid
+#ifdef CONFIG_NET_POLL_CONTROLLER
+		    || dev->poll_controller
+#endif

net/core/dev.c int register_netdevice(struct net_device *dev) · line 4392

Quoted 10 lines
net/core/dev.c +7 -0
+			) {
+			char drivername[64];
+
+			pr_info("%s (%s): mixing netdev_ops and old API\n",
+				dev->name, netdev_drivername(dev, drivername, 
+							     64));
+		}
 		netdev_resync_ops(dev);
 	} else {
 		char drivername[64];

cc'ing Dave.

This seems like a good idea. It would make it more visible. Besides, it's temporary so why not?

Jirka

Quoted 4 lines

--
To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html

#5 lore raw
Quoted 22 lines · On Fri, 20 Mar 2009 15:19:50 +0100 Jiri Pirko <[email protected]> wrote:
Quoted 21 lines · Fri, Mar 20, 2009 at 12:43:00PM CET, [email protected] wrote:
Quoted 6 lines · David Miller

From: Yang Hongyang <[email protected]>
Date: Fri, 20 Mar 2009 16:30:59 +0800

Quoted 3 lines
Quoted 2 lines · David Miller

Committed to net-next-2.6...

Quoted 7 lines
Quoted 4 lines
Quoted 1 line

sunvnet: Convert to net_device_ops.

I noticed that there are other drivers's net_device_ops need to be converted,is these convert nessary?

Not necessary, but desirable. Once all drivers are converted we can remove the compatability code.

Just thinking... Would it make sense, at some point before removing the compatibility layer, to add tests for partial conversions? I noticed the sungem bug just fixed by Stephen Hemminger
( http://permalink.gmane.org/gmane.linux.network/122756 ) which could have been caught by the patch below.

Quoted 8 lines
Quoted 8 lines

This will also ease the transition of "driver sets", like usbnet and it's children, where a number of drivers need to make the conversion together. Don't know if that's a big problem, but it did stop my first attempt to convert usbnet because I was uncertain whether I had caught all of the necessary mini driver conversions.

---

I intend to get all drivers in x86 allmodconfig converted, then change expose the COMPAT_NETDEV_OPS as a config option (which is how I test). Need to wait until after in tree drivers all converted because kernel developers expect random configurations to build.

author maintainer
#6 lore raw

From: Stephen Hemminger <[email protected]>
Date: Fri, 20 Mar 2009 09:16:18 -0700

Quoted 3 lines

I intend to get all drivers in x86 allmodconfig converted, then change expose the COMPAT_NETDEV_OPS as a config option (which is how I test).

If it were not for the randconfig maniacs I would say that I think you should expose it somehow now.

Turning it off is how I personally do conversions in fact.