Re: nfe problem on 8.0-BETA2

Sam Fourman Jr. wrote:
> On Mon, Jul 20, 2009 at 8:48 PM, Ryan
> Rogers<webmaster_at_doghouserepair.com> wrote:
>> I'm running 8.0-BETA2/amd64 on a system that has on-board ethernet.  It is
>> detected by FreeBSD as the following:
>>
>> nfe0: <NVIDIA nForce MCP55 Networking Adapter> port 0xac00-0xac07 mem
>> 0xcfffa000-0xcfffafff,0xcfff9000-0xcfff90ff,0xcfff8000-0xcfff800f irq 22 at
>> device 17.0 on pci0
>> miibus1: <MII bus> on nfe0
>> e1000phy0: <Marvell 88E1116 Gigabit PHY> PHY 1 on miibus1
>> e1000phy0:  10baseT, 10baseT-FDX, 100baseTX, 100baseTX-FDX, 1000baseT,
>> 1000baseT-FDX, auto
>>
>> The problem is, I can't get packets to move across the interface.  I know
>> that the interface works in Windows as well as Ubuntu, but it looks like
>> it's failing to get fully configured in FreeBSD.  The output of ifconfig
>> nfe0 is:
>>
>> nfe0: flags=8843<UP,BROADCAST,RUNNING,SIMPLEX,MULTICAST> metric 0 mtu 1500
>>        options=19b<RXCSUM,TXCSUM,VLAN_MTU,VLAN_HWTAGGING,VLAN_HWCSUM,TSO4>
>>        ether 00:04:4b:01:8c:0b
>>        inet 10.10.10.3 netmask 0xffffff00 broadcast 10.10.10.255
>>        media: Ethernet autoselect (none)
>>        status: active
>>
>> The part that I find odd is "media: Ethernet autoselect (none)".  If I
>> manually force it to "media 1000baseT mediaopt full-duplex", that line
>> becomes "media: 1000baseT full-duplex (10baseT/UTP half-duplex)". Still, the
>> interface is dead.
>>
>> I found PR kern/127910 which describes the same problem, except for
>> 7.0-RELEASE.  There's been no activity on that for 9 months though (aside
>> from my update today).  Can anyone offer me any insight on this?
>>
>> Thanks,
>> Ryan
> 
> 
> I can confirm this is a nfe problem, I first reported it to the
> mailing list a few weeks back
> under the subject FreeBSD 8 BETA1 DHCP trouble.
> http://groups.google.co.jp/group/muc.lists.freebsd.current/browse_thread/thread/ba7b35e561d3e868
> I have several different motherboards with nfe nics, My findings are as follows.
> 
> on a FreeBSD -CURRENT snapshot dated 6-1-2009 dhcp works as expected.
> somewhere after ~ 6-6-2009 something broke.
> 
> dhclient nfe0
> DHCPDISCOVER on nfe0 to 255.255.255.255 port 67 interval 4
> DHCPDISCOVER on nfe0 to 255.255.255.255 port 67 interval 9
> DHCPDISCOVER on nfe0 to 255.255.255.255 port 67 interval 9
> DHCPDISCOVER on nfe0 to 255.255.255.255 port 67 interval 12
> DHCPDISCOVER on nfe0 to 255.255.255.255 port 67 interval 18
> DHCPDISCOVER on nfe0 to 255.255.255.255 port 67 interval 9
> No DHCPOFFERS received.
> No working leases in persistent database - sleeping.
> 
> the odd thing is that on motherboards that have dual nfe nics nfe1
> always works,and nfe0 is always broke.
> on motherboards that have only a single nfe interface nfe0 is indeed broke.
> 
> on a dual nic motherboard, if you go into BIOS and disable one of the
> LAN devices. DHCP will fail if the interface name is nfe0
> even though it worked 5 min before when that same nic and MAC address
> had the nfe1 name.
> 
> This entire thing is odd :)
> 
> 
> Sam Fourman Jr.
> 
> 

I have 7.2-RELEASE/i386 running perfectly using the same motherboard 
that is causing me nfe problems in 8.0-BETA2/amd64.  I then decided to 
see if perhaps it was something with i386 vs amd64, so I downloaded a 
7.2-RELEASE/amd64 iso, and the nics worked perfectly there as well.

Seeing that nfe apparently worked on June 1st, I headed to the FTP to 
see which snapshots were available.  The earliest -CURRENT June snapshot 
was dated the 8th, so I grabbed that one.  The nics didn't work at all.

So, there's a small window where something changed which broke nfe. 
Looking at the SVN commit logs for if_nfe.c, nothing really fits in that 
window.  To me, this looks like something about general network device 
handling changed which nfe nics can't cope with.  I have no idea what 
that could be though, so I'm hoping that someone on this list would be 
able to point me in the right direction.

Ryan