fxp going quiescent in current

[Robert Watson]

On Tue, 14 Nov 2006, Randy Bush wrote:

>> FreeBSD rip.psg.com 7.0-CURRENT FreeBSD 7.0-CURRENT #3: Sat Nov 11 19:18:23 GMT 2006     root at rip.psg.com:/usr/obj/usr/src/sys/RIP  i386
>>
>> and for the last four or five days, fxp0 goes dead.  it shows up
>> and active, but no packets move.
>>
>> down/up does not help.  only way out has been reboot.
>>
>> suggestions on how to debug?
>
> this is killing me.  noc woke me up twice in night to reboot the sucker. 
> any clues?

Do you have serial console access to the box?

The usual questions read:

(1) When it's "dead", do interrupts still fire for the interface when packets
     go near by?  See vmstat -i.

(2) Does the driver think the link is still negotiated?  What are the
     interface flags set to?  See ifconfig.

(3) When you run tcpdump on the interace, does it see packets from the outside
     world?

(4) If you ping out the interface, does tcpdump see packets?  Do you get
     ENOBUFS?  Have ping send at least 256 packets.  Do you get errors?  Send
     to the broadcast address so that arp doesn't need to be working to
     transmit.

(5) What does netstat -m show?

(6) Any unusual dmesg output?  In particular, any mention of fxp state changes
     or watchdogs firing?

(7) Does lowering the interface, waiting ten seconds, then raising it help?

Notice that these are all much easier if you have a serial console.  If not, 
you might want to do the above using cron and a temporary file followed by a 
reboot.  :-)

Robert N M Watson
Computer Laboratory
University of Cambridge