Re: PATCH 2.4.3.6: fix netdevice initialization

[Jeff Garzik]

Andrew Morton wrote:
> We should propagate dev_alloc_name's return value:
> And register_netdevice's, so

ok

> More significantly, the driver probe functions now become:
> 
> xxx_probe()
> {
>         dev = alloc_etherdev(sizeof(xxx_private));
>         ...
>         printk(KERN_INFO "%s: stuff\n", dev->name);
>         ...
>         ret = register_netdev(dev);
>         if (ret < 0)
>                 kfree(ret);
>         return ret;
> }
> 
> yes?

correct.

> And the printk() will say "eth%d: stuff", so we'll need to
> change the messages:
> 
> -       printk(KERN_INFO "%s: stuff\n", dev->name);
> +       printk(KERN_INFO "xxx: stuff\n");
> 
> Correct?

correct.  For PCI drivers, change to something like

  printk(KERN_INFO "tulip(%s): stuff\n", pci_dev->slot_name);

> My quibble with this is things like wait_for_completion(),
> which are called from both the probe() function and
> the mainline driver code.  These also print dev->name,
> and there's no obvious fix for that.

hrm.  I'm not sure it's necessary to pass driver_name to alloc_etherdev
(to set dev->name), just to be able to reference solely from the driver
during the probe phase.  Further there is a dev->name size limit you
will run into with "myverylongdrivernameskimbosh."

Just pass 'name' arg to wait_for_completion ;-)

-- 
Jeff Garzik       | May you have warm words on a cold evening,
Building 1024     | a full mooon on a dark night,
MandrakeSoft      | and a smooth road all the way to your door.