[PATCH] Do not register ne2k-pci device until initialized

[PATCH] Do not register ne2k-pci device until initialized

[Lubomir Rintel]

Doing it in reverse order causes uevent to be sent before
we have a MAC address, which confuses udev.

Signed-off-by: Lubomir Rintel <lkundrak@v3.sk>
---
 drivers/net/ne2k-pci.c |    7 +++----
 1 files changed, 3 insertions(+), 4 deletions(-)

diff --git a/drivers/net/ne2k-pci.c b/drivers/net/ne2k-pci.c
index eb66f65..7d83896 100644
--- a/drivers/net/ne2k-pci.c
+++ b/drivers/net/ne2k-pci.c
@@ -374,18 +374,17 @@ static int __devinit ne2k_pci_init_one (struct pci_dev *pdev,
 	dev->ethtool_ops = &ne2k_pci_ethtool_ops;
 	NS8390_init(dev, 0);
 
+	memcpy(dev->dev_addr, SA_prom, 6);
+	memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
+
 	i = register_netdev(dev);
 	if (i)
 		goto err_out_free_netdev;
 
-	for(i = 0; i < 6; i++)
-		dev->dev_addr[i] = SA_prom[i];
 	printk("%s: %s found at %#lx, IRQ %d, %pM.\n",
 	       dev->name, pci_clone_list[chip_idx].name, ioaddr, dev->irq,
 	       dev->dev_addr);
 
-	memcpy(dev->perm_addr, dev->dev_addr, dev->addr_len);
-
 	return 0;
 
 err_out_free_netdev:

-- 
"Excuse all the blood" -- Dead

Re: [PATCH] Do not register ne2k-pci device until initialized

[Jeff Garzik]

Lubomir Rintel wrote:
> Doing it in reverse order causes uevent to be sent before
> we have a MAC address, which confuses udev.
> 
> Signed-off-by: Lubomir Rintel <lkundrak@v3.sk>

Acked-by: Jeff Garzik <jgarzik@redhat.com>

Good spotting, that fixes a clear startup race

Re: [PATCH] Do not register ne2k-pci device until initialized

[Andreas Mohr]

Hi,

> Good spotting, that fixes a clear startup race

I saw this and thought that this must be one of those non-singular bugs
again... ;)

(I checked for dev_addr manipulations _after_ register_netdev)

And, well...:

smsc911x.c looks like it has the same issue (in case of a non-specified
MAC address - invalid EEPROM content).

Dito ucc_geth.c

Dito arm/ixp4xx_eth.c

Dito wireless/netwave_cs.c


Given that several drivers seem to violate this rule: is there a way
to add a check against this problem?

Andreas Mohr

Re: [PATCH] Do not register ne2k-pci device until initialized

[David Miller]

From: Andreas Mohr <andi@lisas.de>
Date: Sat, 2 May 2009 19:58:11 +0200

> And, well...:
> 
> smsc911x.c looks like it has the same issue (in case of a non-specified
> MAC address - invalid EEPROM content).
> 
> Dito ucc_geth.c
> 
> Dito arm/ixp4xx_eth.c
> 
> Dito wireless/netwave_cs.c
> 
> Given that several drivers seem to violate this rule: is there a way
> to add a check against this problem?

I'm pretty sure we have a valid ethernet address check
somewhere already... hmmm..

Re: [PATCH] Do not register ne2k-pci device until initialized

[David Miller]

From: Jeff Garzik <jeff@garzik.org>
Date: Sat, 02 May 2009 13:05:34 -0400

> Lubomir Rintel wrote:
>> Doing it in reverse order causes uevent to be sent before
>> we have a MAC address, which confuses udev.
>> Signed-off-by: Lubomir Rintel <lkundrak@v3.sk>
> 
> Acked-by: Jeff Garzik <jgarzik@redhat.com>
> 
> Good spotting, that fixes a clear startup race

Applied, thanks.

Re: [PATCH] Do not register ne2k-pci device until initialized

[Krzysztof Halasa]

Andreas Mohr <andi@lisas.de> writes:

> Dito arm/ixp4xx_eth.c

Right, will have a patch soon (mainline only, IMHO it's not worth it
to bother "stable" people because of no practical impact of this bug,
especially on this ARM IXP4xx arch).

Thanks.
-- 
Krzysztof Halasa