Trouble with a 3C597 NIC

[ajw99uk]

I bought serafina's 3C597 which arrived last week - however, my signature update may have been premature, as I've yet to achieve a working connection at 10Mb let alone 100. I've PM'd serafina (and sent messages via ebay) to ask how he had it set up, but until I hear back I would appreciate wider views. The card seems to be in very good physical condition, though obviously inspection by eye won't reveal bad joints or the like.

I have an I2 Extreme, so had to move the graphics up a notch to get at the bottom EISA slot. Ian's fast ethernet guide suggests Extreme and 3C597 should co-exist, although that was in an R10K machine, so presumably with an IMPACT-ready backplane. Is there a known problem with R4K-series I2 trying to combine Extreme and a 3C597 card? For completeness, I've also tried the top slot, with the same results as described below.

The other end of the cable is a Netgear FS608 switch (also tried a Linksys WAG354 modem-router). This purports to do automatic detection of speed.

After installing the card, installing phobos_fe and copying the "hacked" if_fe.o file to /usr/cpu/sysgen/IP22boot, I run "autoconfig -vf" and get a slightly bigger kernel. So far so good, apparently, using both 6.5 and 6.5.22m. Reboot, and the Phobos driver is shown early in the boot sequence: autosense 10Mbps and full duplex.

Amend /etc/config/netif.options to refer to "fe0" in place of "ec0" for the primary interface. Stop/start network or reboot - seems fine, but although I can ping the machine's own address and ifconfig returns sensible info, there is no traffic. "netstat -i" shows no incoming packets at all, and 100% errors on the outgoing packets. Half or full duplex seems to make no difference. Cables are good (for 10Mbps at least), as the connection is fine when I use the onboard NIC.

If I force 100Mbps ("fe_control -i fe0 -s 100" or using the GUI), the corresponding light on the switch goes out, and console starts reporting "lost beat". Is this a symptom of terminal failure of 100Mbps?

Is there any other diagnostic I can try? I'll have to borrow my parent's PC to try its EISA slot and Windows/Linux, but in the meantime any IRIX-based suggestions most welcome.

[robespierre]

Non-IMPACT backplane has 4 EISA slots so you should be able to fit it in the top or bottom slot. There is no problem I know of with R4K machines, I use one with a 3c597.

I can tell you how it's setup later today, but I suspect the netif defaults may not be quite right.

[ajw99uk]

Non-IMPACT backplane has 4 EISA slots so you should be able to fit it in the top or bottom slot. There is no problem I know of with R4K machines, I use one with a 3c597.

I can tell you how it's setup later today, but I suspect the netif defaults may not be quite right.

Thanks, that would be helpful. I think I followed the instructions for netif correctly, but if that's a possible quirk a working example to compare would be great.

Andrew

[robespierre]

OK, I installed this several years ago, so I don't remember all the details of how I set it up. But I can check config files to see what's there now.

In /etc/config/netif.options, I have
if1name=ec0
if1addr=$HOSTNAME
if2name=fe0
if2addr=anothername

you can also use gate-$HOSTNAME as the address of the second interface, or make fe0 the primary. you MUST have entries for both if1 and if2. and you MUST have the name of your host in /etc/sys_id, and entries for BOTH that name and anothername in /etc/hosts. They should be on different subnets.

I'm not quite sure how to check that the driver is really installed in your kernel (kernel size could increase for various other reasons). You should have if_fe.o in /var/sysgen/boot, and if_fe in /var/sysgen/master.d when you run autoconfig. I don't know if it matters, but mine are mode 0444 root:sys.

Good luck.

[ajw99uk]

Thanks, I'll have another go with those settings. Ian's instructions suggest amending if1name to makes the EISA card the primary interface, so I'll find out tonight if it is happier as secondary.

I put if_fe.o in /usr/cpu/sysgen/IP22boot (to which var/sysgen/boot is linked - I had tried /var/sysgen/boot first), and autoconfig did produce a bigger kernel -about 40KB bigger. The third line shown after booting the larger kernel shows that the Phobos driver (or rather the hacked one) is loading, and it does reach the stage where fe_control affects the card - as when selecting 100Mbps apparently causes the card to turn off!

I'll check the mode too, but as above it seems to get past that stage OK, so I hope the netif options are the key bit to fix.

[ajw99uk]

Still no joy, despite a variety of netif / ifconfig combinations. I can get as far as pinging the IP address attached to the 3C597, but not using it to reach any further.

On Friday evening I started again from scratch. Both interfaces are wired to a Netgear FS068 v3 10/100 switch (which seems to have no problem with 100Mbps on other machines and 10Mbps for ec0 or the RiscPC).

1. return I2 to original hardware state as regards GIO/EISA: Extreme in bottom slot
2. move Extreme up a slot, solid orange LED - graphics error
3. take out and reseat midplane
4. put Extreme back in upper slot, boots fine into 6.5.22
5. add 3C597 to bottom EISA slot, boot, install FE_control software and Phobos driver, which generates a new kernel
6. copy if_fe.o to /var/sysgen/boot and generate a new kernel again - the result is about 40KB bigger than the plain
7. amend netif.options to show fe0 as the primary interface, as per the sgidepot instructions
8. reboot: Phobos driver is definitely loaded; it reports half-duplex, 10Mbps (autosense)
9. ping IP address of the primary interface = fine; ping anything else = all packets lost; netstat shows as many outbounds errors as packets and no inbound packets (but, the light on the switch flickers a few times, which suggests something is happening).

Since my opening post in this thread, I've heard back from serafina (hope you had a good vacation!) that the card was a straight pull from his I2. I've handled it carefully, used an ESD bag, wrist-strap, etc and cannot see any damage to the card.

Any further thoughts welcome - and a UK offer to try the card in another I2 even more so!

[jan-jaap]

* I've used the 3c597 (a long time ago). Extreme gfx in the bottom 3 slots, 3c597 in the top slot.

* To use the 3c597 with the Phobos driver, the driver has to be hacked with a hex editor, because it is locked to something which identifies it as a Phobos card rather than a vanilla 3com card. I may still have a modified driver.

[SAQ]

I have one in Stylosa, but haven't played around with it in a while.

One thing to try might be wiring a crossover cable to another machine and running tcpdump/etherape/something of that ilk on the second box and seeing what you get.

Can't ping anything else sounds like it might be routing. Check to make sure the default route goes out over fe0 instead of ec0.

Per SGI

If the destination is directly reachable via an interface requiring no
intermediary system to act as a gateway, the -interface modifier should
be specified to indicate a special type of gateway value; the gateway
value given is the address of this host on the common network (not he
interface name), indicating the interface to be used for transmission.
When a route has been added with the -interface modifier 'netstat -rn'
ouput will show link# in the gateway column and the flags of the
interface by default.

[ajw99uk]

* I've used the 3c597 (a long time ago). Extreme gfx in the bottom 3 slots, 3c597 in the top slot.

Tried both, though only the lower since removing/reseating the midplane. Same results using top or bottom EISA slot.

* To use the 3c597 with the Phobos driver, the driver has to be hacked with a hex editor, because it is locked to something which identifies it as a Phobos card rather than a vanilla 3com card. I may still have a modified driver.

I have the already-hacked driver from mapesdhs's sgidepot - the fact that fe_control speed settings do affect the card suggest it is being loaded and operating (to some extent at least). But an second version would be welcome, just in case there's some glitch in downloading/extracting the first.

Thanks for your interest.

[ajw99uk]

I have one in Stylosa, but haven't played around with it in a while.

One thing to try might be wiring a crossover cable to another machine and running tcpdump/etherape/something of that ilk on the second box and seeing what you get.

Can't ping anything else sounds like it might be routing. Check to make sure the default route goes out over fe0 instead of ec0.

Thanks for the suggestions - I hope one of my many RJ45 cables is crossover ....

[ajw99uk]

Just had another go - booted first with ec0 as the primary interface in netif.options, pinged it fine from another machine, and could ping to that machine also. The two machines were connected via the switch, as none of my ethernet cables is a crossover.
I saved outputs of "ifconfig", "netstat -i" and "netstat -rn", edited netif.options to make fe0 the primary interface and powered down/up (KVM doesn't like a simple reboot, sadly). Then captured the same info to file, to compare and post here for comments.

Outputs from "ifconfig -av":
With ec0 as primary interface:

Code: Select all

ec0: flags=8400c43<UP,BROADCAST,RUNNING,FILTMULTI,MULTICAST,IPALIAS,IPV6>
   inet 10.236.2.97 netmask 0xffffff00 broadcast 10.236.2.255
   speed 10.00 Mbit/s half-duplex
fe0: flags=822<BROADCAST,NOTRAILERS,MULTICAST>
lo0: flags=8001849<UP,LOOPBACK,RUNNING,MULTICAST,CKSUM,IPV6>
   inet 127.0.0.1 netmask 0xff000000

With fe0 as primary interface:

Code: Select all

fe0: flags=c63<UP,BROADCAST,NOTRAILERS,RUNNING,FILTMULTI,MULTICAST>
   inet 10.236.2.97 netmask 0xffffff00 broadcast 10.236.2.255
ec0: flags=8400802<BROADCAST,MULTICAST,IPALIAS,IPV6>
   speed 10.00 Mbit/s half-duplex
lo0: flags=8001849<UP,LOOPBACK,RUNNING,MULTICAST,CKSUM,IPV6>
   inet 127.0.0.1 netmask 0xff000000

Outputs from "netstat -i" - note that the fe0 interface can receive packets without error, but seems unable to send outgoing packets. Pinging the I2 from another machine causes the "Ipkts" number to increase.
With ec0 as active interface:

Code: Select all

Name   Mtu   Network         Address            Ipkts Ierrs    Opkts Oerrs  Coll
ec0    1500  10.236.2        bluebox               60     0       20     0     0
fe0*   1500  none            none                   0     0        0     0     0
lo0    8304  loopback        localhost           1355     0     1355     0     0

With fe0 as active interface:

Code: Select all

Name   Mtu   Network         Address            Ipkts Ierrs    Opkts Oerrs  Coll
fe0    1500  10.236.2        bluebox               98     0       20    20     0
ec0*   1500  none            none                   0     0        0     0     0
lo0    8304  loopback        localhost           1373     0     1373     0     0

Finally, output of "netstat -rn" using ec0 as primary interface, then with fe0

Code: Select all

Routing tables
Internet:
Destination      Gateway            Netmask    Flags    Refs     Use  Interface
default          10.236.2.1                    UGS         1        0  ec0
10.236.2         link#1             0xffffff00 UC          0        0  ec0
10.236.2.97      127.0.0.1                     UGHS        0        0  lo0
127.0.0.1        127.0.0.1                     UH          7        2  lo0
224              link#1             0xf0000000 UCS         0        0  ec0
255.255.255.255  10.236.2.255                  UGHS        0        0  ec0

This is the equivalent when fe0 is the active interface:

Code: Select all

Routing tables
Internet:
Destination      Gateway            Netmask    Flags    Refs     Use  Interface
default          10.236.2.1                    UGS         1        0  fe0
10.236.2         link#2             0xffffff00 UC          0        0  fe0
10.236.2.97      127.0.0.1                     UGHS        0        0  lo0
127.0.0.1        127.0.0.1                     UH          6        2  lo0
224              link#2             0xf0000000 UCS         0        0  fe0
255.255.255.255  10.236.2.255                  UGHS        0        0  fe0

The only difference I can see is the "link#" entries (and the localhost flags number). Should link#2 here be link#1, and if so how do I change it?

Any other thoughts?

Thanks,
Andrew

[ajw99uk]

One last thought before I give up on this.

If I can get incoming packets through the EISA card, would it be possible to:
1. set up the onboard interface to send outgoing only
2. use the EISA for incoming only
3. link the two so as to appear to the network as a single address (or does this not matter? 2 IP addresses on the same subnet?)

Even if possible, would this actually yield better performance that the onboard i/f by itself? (bearing in mind that the EISA card seems unable to provide 100mb/s - "fe_control -s 100" kills the link).

[ajw99uk]

Just to close off this thread, I have resolved the issue by buying a G160!

I should also note that having seen this thread and PMs recording my struggles with the 3C597, serafina (from whom I bought it) offered to send me a second 3C597 to try. However, I now have a spare but doubtful 3C597 and (I think) serafina has sold the I2 from which it came!

[foetz]

a quick addition here although not entirely on topic:

if you should have i/o blocking on 6.2 with nfs use nfs v2 for mounting on the indigo. sending was fine but receiving unusable otherwise.

[hamei]

Since we're doing thread necrophilia ...

If I force 100Mbps ("fe_control -i fe0 -s 100" or using the GUI), the corresponding light on the switch goes out, and console starts reporting "lost beat". Is this a symptom of terminal failure of 100Mbps?

Cisco is adamant that either both ends of a link should auto-negotiate or both ends should be manually set. Apparently sometimes half-half works but it is not something you can depend upon. Have run across this in real life as well.