"eth%d" - net dev name in 2.6?

"eth%d" - net dev name in 2.6?

[Jun Sun]

With swarm running on 2.6 I just saw the net dev names are
not set correctly.  See below.

eth%d: SiByte Ethernet at 0x10064000, address: 00-02-4C-FE-0C-B2                
eth%d: enabling TCP rcv checksum                                                

It appears alloc_netdev() assigns this initial name and nobody
later resets it to a more meaningful name.  

Any body has a clue here?  I don't think it is driver's job though ...

Thanks.

Jun

Re: "eth%d" - net dev name in 2.6?

[Martin Michlmayr]

* Jun Sun <jsun@mvista.com> [2004-03-09 18:33]:
> eth%d: SiByte Ethernet at 0x10064000, address: 00-02-4C-FE-0C-B2                
> eth%d: enabling TCP rcv checksum                                                
> 
> Any body has a clue here?  I don't think it is driver's job though ...

I have no idea, but I've seen a similar bug report at
http://bugs.debian.org/234817

-- 
Martin Michlmayr
tbm@cyrius.com

Re: "eth%d" - net dev name in 2.6?

[Ralf Baechle]

On Wed, Mar 10, 2004 at 02:53:46AM +0000, Martin Michlmayr wrote:

> I have no idea, but I've seen a similar bug report at
> http://bugs.debian.org/234817

This affects drivers/net/ne.c, so I'm adding netdev to the cc list.

Same problem, dereferencing net_device->name before register_netdev().
That's caused by a bit toooo quick and sloopy conversion from
init_etherdev to alloc_etherdev & register_netdev.

  Ralf

Re: "eth%d" - net dev name in 2.6?

[Kumba]

Jun Sun wrote:

> With swarm running on 2.6 I just saw the net dev names are
> not set correctly.  See below.
> 
> eth%d: SiByte Ethernet at 0x10064000, address: 00-02-4C-FE-0C-B2                
> eth%d: enabling TCP rcv checksum                                                
> 
> It appears alloc_netdev() assigns this initial name and nobody
> later resets it to a more meaningful name.  
> 
> Any body has a clue here?  I don't think it is driver's job though ...
> 
> Thanks.
> 
> Jun

I've seen this for ages on 2.4 and 2.6.  Seems to be some kind of typo 
or something in several archs (my Blade 100 shows this in dmesg as well).


--Kumba

-- 
"Such is oft the course of deeds that move the wheels of the world: 
small hands do them because they must, while the eyes of the great are 
elsewhere."  --Elrond

Re: "eth%d" - net dev name in 2.6?

[Ralf Baechle]

On Tue, Mar 09, 2004 at 11:14:37PM -0500, Kumba wrote:

> I've seen this for ages on 2.4 and 2.6.  Seems to be some kind of typo 
> or something in several archs (my Blade 100 shows this in dmesg as well).

What driver?

  Ralf

Re: "eth%d" - net dev name in 2.6?

[Kumba]

Ralf Baechle wrote:

> On Tue, Mar 09, 2004 at 11:14:37PM -0500, Kumba wrote:
> 
> 
>>I've seen this for ages on 2.4 and 2.6.  Seems to be some kind of typo 
>>or something in several archs (my Blade 100 shows this in dmesg as well).
> 
> 
> What driver?
> 
>   Ralf
> 

Sun GEM driver.  Looks like it's fixed in 2.6, but 2.4.22/23 on the 
system had this in dmesg:

sungem.c:v0.97 3/20/02 David S. Miller (davem@redhat.com)
eth%d: MII PHY ID: 437420 Enable Semiconductor
eth0: Sun GEM (PCI) 10/100/1000BaseT Ethernet 00:03:ba:04:be:a4

Seen it else where too, Maybe on a 2.4 driver in my x86 box, but that 
dmesg isn't available anymore to double check.


--Kumba

-- 
"Such is oft the course of deeds that move the wheels of the world: 
small hands do them because they must, while the eyes of the great are 
elsewhere."  --Elrond

Re: "eth%d" - net dev name in 2.6?

[Ralf Baechle]

On Tue, Mar 09, 2004 at 06:33:08PM -0800, Jun Sun wrote:

> With swarm running on 2.6 I just saw the net dev names are
> not set correctly.  See below.
> 
> eth%d: SiByte Ethernet at 0x10064000, address: 00-02-4C-FE-0C-B2                
> eth%d: enabling TCP rcv checksum                                                
> 
> It appears alloc_netdev() assigns this initial name and nobody
> later resets it to a more meaningful name.  
> 
> Any body has a clue here?  I don't think it is driver's job though ...

It's always the driver :-)

It's referencing net_device->name before register_netdev.

There's plenty of other small candy in that driver, for example in
sbmac_cleanup_module():

                dev = dev_sbmac[idx];
                if (!dev) {
                        struct sbmac_softc *sc = dev->priv;
                        unregister_netdev(dev);
                        sbmac_uninitctx(sc);
                        free_netdev(dev);
                }

Better make sure the pointer is NULL before we dereference it.  We don't
want to miss a crash, do we ;-)

  Ralf