Strange ethN numbering problem.

Strange ethN numbering problem.

[John Clark]

On a system which has one Intel Ethernet 1 Gb interface, and 4 'Marvel', 
interfaces,
during kernel initialization the interfaces indicate they have 'normal' 
ethernet ethN names,
ie, eth0, eth1, eth2, eth3, eth4, eth5 are reported from the 'dmesg' output.

However, when the system comes up and attempt to do an ifconfig, the 
'ethN' numbers
have changed to a some what intermengled seriese starting with eth6... 
eth10.

I have never seen this sort of problem before, and so I have no clue 
what is causing
the later changes in the numbering scheme.

Does anyone have any idea where to look.

Thanks
John Clark

Re: Strange ethN numbering problem.

[Bernd Eckenfels]

In article <45A2E19F.4070307@metricsystems.com> you wrote:
> However, when the system comes up and attempt to do an ifconfig, the 
> 'ethN' numbers
> have changed to a some what intermengled seriese starting with eth6... 
> eth10.

maybe a system startup script is renaming them (in order to give them well
known numbers)? 

What kind of distribution is that? is this a new problem? Have a look in
/etc/mactab.

Gruss
Bernd

Re: Strange ethN numbering problem.

[John Clark]

Bernd Eckenfels schrieb:
> In article <45A2E19F.4070307@metricsystems.com> you wrote:
>   
>> However, when the system comes up and attempt to do an ifconfig, the 
>> 'ethN' numbers
>> have changed to a some what intermengled seriese starting with eth6... 
>> eth10.
>>     
>
> maybe a system startup script is renaming them (in order to give them well
> known numbers)? 
>
> What kind of distribution is that? is this a new problem? Have a look in
> /etc/mactab.

This is not a 'new' distribtution. In fact, the disk was used for a 
previous hardware box, of the same
manufacturer and allegedly the same cpu mother board.

The kernel is 2.6.19.1 the at-that-moment current linux kernel.

What should I look for in terms of interface renaming. What is also sort 
of strange is that they all
have the same 'mac' address vendor unique id... even though two 
interfaces are for an Intel
ethernet chip, and the othe 4 are from the Marvel chip.

Thanks
John Clark

Re: Strange ethN numbering problem.

[Andrey Borzenkov]

John Clark wrote:

> Bernd Eckenfels schrieb:
>> In article <45A2E19F.4070307@metricsystems.com> you wrote:
>>   
>>> However, when the system comes up and attempt to do an ifconfig, the
>>> 'ethN' numbers
>>> have changed to a some what intermengled seriese starting with eth6...
>>> eth10.
>>>     
>>
>> maybe a system startup script is renaming them (in order to give them
>> well known numbers)?
>>
>> What kind of distribution is that? is this a new problem? Have a look in
>> /etc/mactab.
> 
> This is not a 'new' distribtution. In fact, the disk was used for a
> previous hardware box, of the same
> manufacturer and allegedly the same cpu mother board.
> 

Then quite likely it remembered lower numbers for "old" interfaces and
starts renaming with next available.

> The kernel is 2.6.19.1 the at-that-moment current linux kernel.
> 
> What should I look for in terms of interface renaming.

I guess in udev rules; look also if you have /etc/iftab. The best you can do
is asking in lists/groups dedicated to your distribution.

-andrey

> What is also sort 
> of strange is that they all
> have the same 'mac' address vendor unique id... even though two
> interfaces are for an Intel
> ethernet chip, and the othe 4 are from the Marvel chip.
> 
> Thanks
> John Clark

Re: Strange ethN numbering problem.

[John Clark]

Andrey Borzenkov schrieb:
> John Clark wrote:
>
>   
>
> Then quite likely it remembered lower numbers for "old" interfaces and
> starts renaming with next available.
>
>   
>> The kernel is 2.6.19.1 the at-that-moment current linux kernel.
>>
>> What should I look for in terms of interface renaming.
>>     
>
> I guess in udev rules; look also if you have /etc/iftab. The best you can do
> is asking in lists/groups dedicated to your distribution.
>
> -andrey

Thanks.

It was 'udev rules' that were messing things up, left over from using 
the disk on a different
piece of hardware. To date I've been making only 'embedded'  systems 
using busybox, and other
similarly limited root environments and never really dealt with a  
pretty much full up distribution
outside of my host development systems. Hence, never really had gotten 
in to 'udev'. I've been
using devfs mostly till recently... However, in anticipation of large 
capacity flash systems, I've moved
to making my embedded systems almost as full up as most host systems.

Is there some startup command line option for the linux kernel to force 
the 'udev' management
program to basically 'ignore and refresh' device names?


John Clark.