fxp0: device timeout and sk0: watchdog timeout problems, onboard pcn problem

[Ganbold]

At 06:08 PM 1/19/2006, you wrote:
>hi
>maybe a stupid hint but did you checked the TP cable and pin order? I often
>saw such problems when a gigabit connected to an older /10 HUB or when using
>incorrect pin orders on the cables or even to long cables

I will ask person to check the cables.

>Appearently the sk down events are gone on my server with Pyun's patched
>drivers, did you checked if you compiled with them in the right place?

I put sk drivers in /usr/src/sys/pci and compiled and installed kernel again.

>My servers are now almost 24h up with the new sk driver and none of them is
>having the problem any more until now. Until this I had crontab run "ifconfig
>sk0 up" all 10 minutes what helped me so long.

ifconffig sk0 up might be the solution until I find the problem.

>On my SMP system when sk stopped it caused also watchdog timeout on all other
>NICs, on UP kernel not.

This system is SMP system, so I guess that is why fxp is timing out too.

>Do you use interface polling? Check with vmstat -i because you may have
>overlapping IRQs and you cards may not support it. Look into the MB manual
>and see which pci slots are shared and don't use them or try another bios
>setting. What MB you are using?

It is not using polling. I should ask the person to check the MB.
gw# vmstat -i
interrupt                          total       rate
irq1: atkbd0                         188          0
irq6: fdc0                            10          0
irq13: npx0                            1          0
irq14: ata0                            1          0
irq16: ahc0 ahc1                   56984         41
irq18: skc0 ohci0                 411469        300
irq19: fxp0                       104891         76
cpu0: timer                      2738746       1997
cpu1: timer                      2720830       1984
Total                            6033120       4400

>you may try ifconfig fxp link0 to load its microcode which could help but in
>my opinion and experience the fxp cards are bad when using more than one and
>I through them all out but interesting is that they work well on until
>releng_5, btw the sk  also works well on releng_5

Yes, I have at least several machines running 
FreeBSD-5.3 with 2 fxp cards and they seem to be OK.
Also I have another 2 FreeBSD 6.0 machine with 2 
fxp NIC and they are also fine.

I tried onboard pcn card:
pcn0 at pci0:9:0:  class=0x020000 card=0x20001014 
chip=0x20001022 rev=0x36 hdr=0x00
     vendor   = 'Advanced Micro Devices (AMD)'
     device   = 'Am79C970/1/2/3/5/6 PCnet LANCE PCI Ethernet Controller'
     class    = network
     subclass = ethernet

However system crashes after connecting the 
cable. After reboot it is the same, crashes.

Ganbold


>João