supported (non SGI) PCI gigabit cards?

[dj26]

I went and ordered 4 of these and they called me to say the balance of their stock is damaged (broken caps etc) they are trying to get more of them but it looks like this source has dried up atleast temporarily.

DJ

One possible source is this site:

http://www.cc-solutions.com

They apparently have some Alteon ACEnics (Dell OEM) in stock which is what I have right now. I placed an online order earlier today - guess we'll see if something arrives in a few days or not.

[nekonoko]

Anyone feel like a little risk-taking ? It's also Broadcom 5700 ...

I just bought a cheap Broadcom based card on eBay to play with - I want to see if they are more forgiving as far as power draw in the Octane PCI shoeboxes. They're also a little easier to find.

[artherd]

neko- no ikidding on power draw! My Octane is questionable stability when loaded with 2 Atelons and a QLA2200F.

[nekonoko]

And to make matters worse - now my Octane isn't stable even with the Alteon removed - still crashing with PCI bus errors after a couple hours uptime. I guess it's time to look for a new card cage - too bad there aren't any for sale on eBay at the moment :/

[artherd]

Aaaaack! I hope we didn't fry our's!

I have removed one of the Atelon cards, right now running just one gigabit card and Qla2200F.

So far it's been stable, but only a week or so will tell for sure.

[hamei]

And to make matters worse - now my Octane isn't stable even with the Alteon removed - still crashing with PCI bus errors after a couple hours uptime. I guess it's time to look for a new card cage - too bad there aren't any for sale on eBay at the moment :/

Maybe you guys will have to run a supplementary wallwart to power your shoeboxes. Eeeeeuw !

[nekonoko]

I'm hoping it's just a bad compression connector on the shoebox - I have another shoebox lined up and will do more tests then. In the meantime the current shoebox works well enough to establish whether or not a Broadcom board (HP NC7770) can be made to work. Stay tuned!

[Intel-OUTSIDE]

if you damage a cage(in the electrical sense) i may buy it from you, it's probably the only way i'l get one.

[hamei]

if you damage a cage(in the electrical sense) i may buy it from you, it's probably the only way i'l get one.

Someone in here has a box with a bad connector - maybe one of the jbrothers ? jwhat ? Give him a pm .....

[schleusel]

In the meantime the current shoebox works well enough to establish whether or not a Broadcom board (HP NC7770) can be made to work. Stay tuned!

ah, cool.. I had planned to hunt for a tigon3 card to test it myself in the next weeks but this way its even better Good luck!

[nekonoko]

Okay I got the card in this morning and it's working fine. For the record I'm using a Compaq NC7770 PCI-X Gigabit Server Adapter.

The following command provides the PCI IDs we need to look at (14e4:1645, 14e4:1648 and 14e4:16a8 being the supported tigon3 card families):

Code: Select all

[Koyori:cpu/sysgen/IP30boot] neko 7% od -x if_tg.o | grep 14e4
 ...
 0223060 14e4 1645 10a9 8010 0001 0100 0000 0000
 0223100 14e4 1645 10a9 8011 0001 0100 0000 0000
 0223120 14e4 1645 10a9 8012 0100 0000 0000 0000
 0223140 14e4 1648 10a9 8013 0001 0000 0000 0000
 0223160 14e4 16a8 10a9 8014 0100 0000 0000 0000
 ...

Next, using the information from this page:

http://pciids.sourceforge.net/iii/?i=14e41645

The subsystems for my card are either 0e11:007c or 0e11:009a. I decided to start at the top and try 007c first which turned out to be the correct choice.

The new subsystem should match the SGI equivalent. Since the NC7770 is a copper device we need to replace the reference to the equivalent SGI copper PCI card (subsystem 10a9:8011).

It should now look like this:

Code: Select all

[Koyori:cpu/sysgen/IP30boot] neko 4% od -x if_tg.o | grep 14e4
...
0223060 14e4 1645 10a9 8010 0001 0100 0000 0000
0223100 14e4 1645 0e11 007c 0001 0100 0000 0000
0223120 14e4 1645 10a9 8012 0100 0000 0000 0000
0223140 14e4 1648 10a9 8013 0001 0000 0000 0000
0223160 14e4 16a8 10a9 8014 0100 0000 0000 0000

After completing the edit I ran a 'autoconfig -vf' to build the new kernel.

After a reboot:

Code: Select all

[Koyori:cpu/sysgen/IP30boot] neko 5# hinv
2 300 MHZ IP30 Processors
CPU: MIPS R12000 Processor Chip Revision: 2.4
FPU: MIPS R12010 Floating Point Chip Revision: 0.0
Main memory size: 2048 Mbytes
Xbow ASIC: Revision 1.4
Instruction cache size: 32 Kbytes
Data cache size: 32 Kbytes
Secondary unified instruction/data cache size: 1 Mbyte
Integral SCSI controller 0: Version QL1040B (rev. 2), single ended
  Disk drive: unit 1 on SCSI controller 0
  Disk drive: unit 2 on SCSI controller 0
Integral SCSI controller 1: Version QL1040B (rev. 2), single ended
Integral SCSI controller 2: Version QL1080, low voltage differential
Integral SCSI controller 3: Version QL1080, low voltage differential
IOC3/IOC4 serial port: tty1
IOC3/IOC4 serial port: tty2
IOC3 parallel port: plp1
Graphics board: V8
Integral Fast Ethernet: ef0, version 1, pci 2
Gigabit Ethernet: tg1, PCI slot 1
Iris Audio Processor: version RAD revision 12.0, number 1

tgcmd works though it's a bit confused by the subsystem change (can't grab the board vendor and media type which isn't important). Note that it still knows it's working with the 5701 chip which is what we want:

Code: Select all

[Koyori:cpu/sysgen/IP30boot] neko 6# tgcmd inventory
        Chip   Chip   Sybsys Subsys Board
I/F     Vendor Device Vendor Device Rev
===     ====== ====== ====== ====== ======
tg1     0x14e4 0x1645 0x0e11 0x007c 0x0000 (UNKNOWN VENDOR) 5701 (UNKNOWN MEDIA)

[Koyori:cpu/sysgen/IP30boot] neko 12# tgcmd configs tg1
accept_bcast                   1
autoneg                        ON
batch_receive                  OFF
coal_frames_rx                 0
coal_frames_tx                 0
coal_mode                      ADAPTIVE
coal_usecs_rx                  0
coal_usecs_tx                  0
data_rrbs                      4
fill_thresh                    10
flow_ctrl                      ON
hw_cksum_rx                    1
hw_cksum_tx                    1
intr_cpu                       -1
link_pref                      AUTO
mac_addr                       0x0
mac_mask                       0x0
mbuf_chain_len                 8
sendQ_size                     2048
l2ipfrag                       ON
l2tcpseg                       ON

[Koyori:cpu/sysgen/IP30boot] neko 13# tgcmd get_cpu tg1
tg1 interrupts are running on CPU 1.

[Koyori:cpu/sysgen/IP30boot] neko 14# tgcmd get_stats tg1

Driver Statistics:
                   1  Max Pkts seen on I/F Send Queue
                2090  Tx pkts with H/W generated checksum
                1275  Tx pkts with S/W generated checksum
                3365  Tx Ring producer index updated
                   2  Max Bfrs on Send Ring
                8298  Interrupts
                  11  Max Bfrs added to RxStd Ring
                   4  Nested interrupts
                2196  Rx pkts with H/W generated checksum
                2811  Rx pkts with S/W generated checksum
                8302  Status Block updates detected
                3365  Send lock obtained

NIC Statistics:
              380414  ifHCInOctets
               10465  <0x408>
                3640  ifHCInUcastPkts
                1364  ifHCInBroadcastPkts
                1092  etherStatsPkts64Octets
                3593  etherStatsPkts65to127Octets
                 309  etherStatsPkts128to255Octets
                   9  etherStatsPkts256to511Octets
                   1  etherStatsPkts512to1023Octets
              892255  ifHCOutOctets
                2635  ifHCOutUcastPkts
                 728  ifHCOutBroadcastPkts
                5004  COSIfHCInPkts[1]
                3363  COSIfHCOutPkts[1]
                3375  nicRingSetSendProdIndex
                8312  nicRingStatusUpdate
                8311  nicInterrupts
                   1  nicAvoidedInterrupts


Network up and running:

Code: Select all

Oct  7 12:49:38 5A:Koyori unix: NOTICE: tg1: Link up: 1000 Mbps, FULL duplex, flow_ctrl is ON

Code: Select all

[Koyori:cpu/sysgen/IP30boot] neko 7# ifconfig -v tg1
tg1: flags=8f15c43<UP,BROADCAST,RUNNING,FILTMULTI,MULTICAST,CKSUM,DRVRLOCK,LINK0,L2IPFRAG,L2TCPSEG,IPALIAS,HIGHBW,IPV6>
        inet 64.81.247.23 netmask 0xffffff00 broadcast 64.81.247.255
        recvspace 262144 sendspace 262144
        speed 1000.00 Mbit/s full-duplex

And that's all there is to it

As far as running one alongside a DM10 - well, it doesn't crash but the network doesn't seem to work despite having a healthy link. All commands work without error but I can't ping anything. Removing the DM10 clears it back up - those things are just beastly in an Octane. I think the only real solution would be to offload the DM10 to a shoehorn which would hopefully steal it's power from elsewhere.

[nekonoko]

I'm hoping it's just a bad compression connector on the shoebox - I have another shoebox lined up and will do more tests then. In the meantime the current shoebox works well enough to establish whether or not a Broadcom board (HP NC7770) can be made to work. Stay tuned!

Just a follow-up that the cage seems fine - I wonder if I just wasn't seating it forcefully enough. I slammed it in there pretty good this last time and had the machine up for about 10 hours without a fuss.

[artherd]

NICE work neko! Harder than the Atelon hack, I'll have to study what you wrote for a while )

Damn, you're amazing.

Is this working with 6.5.25?

[nekonoko]

Thanks Yes this was done using 6.5.25 on an Octane - the same procedure should also work on IP27.

[msunix]

Hi!

Has anyone here already tried this trick with a 3com 3c996? This card seems to be very similar to Neko's compaq-card and is available everywhere.

Michael