driver: net: smsc911x: set NOCARRIER flag in driver initialization

As smsc driver supports carrier detection, it should unset NOCARRIER flag only after carrier state determination. By default that flag is off so driver should set it before starting auto-negotiation

---
 drivers/net/ethernet/smsc/smsc911x.c | 2 ++
 1 file changed, 2 insertions(+)

index a0fc151..69016a5 100644
@@ -2525,6 +2525,8 @@ static int smsc911x_drv_probe(struct platform_device *pdev)
 
 	netdev_info(dev, "MAC Address: %pM\n", dev->dev_addr);
 
+	netif_carrier_off(dev);
+
 	return 0;
 
 out_unregister_netdev_5:

1.8.3.2

From: Balakumaran Kannan <[email protected]>
Date: Sun, 25 May 2014 00:24:46 +0530

Hi David,

IMHO calling netif_carrier_off after registering device will not affect functionality. Because,

* Device starts carrier detection once driver's open function
calls phy_start.
* Kernel starts network layer initialization only after __dev_open
function sets IFF_UP bit of dev->flags.
And __dev_open sets this bit only after driver's open function returns.
* So we could place netif_carrier_off call anywhere before
phy_start in driver's open function gets called.

Also I have referred some other drivers.

* Intel i40e driver calls netif_carrier_off at the beginning of
driver's open (i40e_open) function.
* e1000 driver places netif_carrier_off at the end of driver probe
(e1000_probe) function.
* realtek r8169 driver calls netif_carrier_off at the end of
driver init (rtl_init_one) function.

All these drivers call netif_carrier_off function after registering the network device using register_netdev.

This explanation is based on my understanding. Please explain to me if I'm wrong or I miss something. In such case I'll modify the patch accordingly.

Thank you.

Regards,
K.Balakumaran

From: Balakumaran Kannan <[email protected]>
Date: Fri, 30 May 2014 18:35:43 +0530

I can quote more counter-examples than your list, starting with the Broadcom tg3 driver.

Thank you David for this information. I'll send updated patch as version-2.

Still I couldn't understand the reason behind this. I should be grateful if you would explain me little bit.

Regards,
K.Balakumaran

2014-06-03 7:15 GMT-07:00 Balakumaran Kannan <[email protected]>:

I believe one reason is that as soon as you call register_netdev(), the network device notifiers run as part of call_netdevice_notifiers() and it is valid for a notified callback to open your network device. This means that you can call the driver's ndo_open() before you returned from register_netdev(). If netif_carrier_off() is called after register_netdev(), there is a risk for ndo_open() not to get a consistent carrier state, and the actual call to netif_carrier_off() to mess things up.

Florian

From: Balakumaran Kannan <[email protected]>
Date: Tue, 3 Jun 2014 19:45:39 +0530

At the moment you call register_netdev() the device is visible, notifications are sent to userspace, and userland tools can try to bring the interface up and see the incorrect link state, before you do the netif_carrier_off().

Said another way, between the register_netdev() and netif_carrier_off() call, userspace can see the device in an inconsistent state.

Thank you very much Florian and David for great explanation.

Regards,
K.Balakumaran