| View previous topic :: View next topic |
| Author |
Message |
Mnemia
Guru
Joined: 17 May 2002
Posts: 476
|
 Posted: Wed Jul 23, 2003 9:09 pm Post subject: Nasty problem with Adaptec quad Ethernet controller |
 |
|
I'm having some problems with the Adaptec ANA-64044LV four port Ethernet card (64 bit). I'm using vanilla kernel 2.4.21.
This card uses the starfire driver and that loads up okay. All four ports are detected by the driver and I can use them to send and receive data. However, when the driver initializes each of the chips on the card, I get a message like this:
| Code: |
eth6: Dropping NETIF_F_SG since no checksum feature.
eth6: Adaptec Starfire 6915 at 0xc8a67000, 00:00:d1:f0:29:c1, IRQ 11.
eth6: MII PHY found at address 1, status 0x7809 advertising 0x01e1.
eth6: scatter-gather and hardware TCP cksumming disabled.
|
I KNOW this card has hardware checksumming and that it is supported under Linux based on what I've read on Donald Becker's site and in the course of some Googling. So why is it "disabled"?
I tried turning on the verbose debugging mode for the driver but all it did was print out some register dumps.
Also, I'm having another apparent problem with these cards as well. If I have all four ports plugged into the same network with different IP addresses, ARP doesn't work properly from other hosts on the network. All four ports respond to any request for the MAC address of any of the IPs assigned to any of the ports!!! This is despite the fact that all the ports have different MAC addresses and different IPs. They actually all send a simultaneous ARP response trying to claim an IP address that only one of them has. The practical effect of this is that since the card always transmits from each port in a sequential order, the first port always gets all four IPs when I look in the ARP tables of other hosts on the network. I figured out that this was happening by examining ARP tables and watching the output from tcpdump. My question here is - is this normal behavior? It doesn't seem normal to me. Will this have negative performance consequences? |
|
| Back to top |
|
 |
xplosiv
n00b
Joined: 26 Jul 2003
Posts: 29
|
| Posted: Sat Jul 26, 2003 3:36 am Post subject: |
|
|
What a coinsidence, I am trying this same card using the gentoo sources. I can't even get the card to work right, I see the ports listed (eth2-eth5), and the Rx Bytes and TxBytes counter does increase, I have link lights, the boot up log shows it established a link, but I can't do anything IP wise i.e. icmp/tcp/udp.
My onboard nics work just fine, and I have no error messages at all. The only thing I can think of is that this card is a 64 bit card, and I inserted it into a 32bit pci slot (according to adaptec, it's backwards compatible). Many stores list that it is Linux compatible, Adaptec's site doesnt mention linux at all.
Did you have to do anything special besides selecting the starfire driver in make menuconfig? I am really desperate here  |
|
| Back to top |
|
|
Mnemia
Guru
Joined: 17 May 2002
Posts: 476
|
| Posted: Sat Jul 26, 2003 8:26 am Post subject: |
|
|
Actually, I have the card in a 32 bit PCI slot as well. I was wondering if that might be related to my troubles as well.
Short story - no, I didn't have to do anything too special to get the cards to work.
Long story - my computer actually wouldn't even boot with the cards in at first. It would lock up during the POST before the bootloader even loaded. I fixed that problem with a BIOS update from Intel, and then the cards worked to a degree. They now work fine for me except for the two problems I mentioned above. I have starfire compiled into the kernel, not as a module, of that means anything. I'm using them with static IPs (I've never tried them with DHCP). Other than the checksum problem they seem to work fine when only one interface is on the same network segment, as this avoids the ARP problem I mentioned.
Here's some links I found when trying to solve my problem that might be helpful to you:
Adaptec's Linux Page
http://www.adaptec.com/worldwide/common/editorial.html?sess=no&prodkey=linux_home
Their page isn't all that helpful but I saw a mention in there somewhere that they will try to provide assistance if you have problems with Linux. So if all else fails you might try emailing their technical support.
Donald Becker's Page:
http://www.scyld.com/diag_index.html
This is the guy who wrote the starfire and most other Ethernet drivers in the Linux kernel. His page has several diagnostic utilities you can compile including a starfire specific one.
Your problem sounds like it might be related to messed up ARP stuff since the link appears to be up. Have you tried checking out the ARP tables on various other hosts on the network after pinging the card you're having trouble with? Try that and try using tcpdump to view the traffic from ARP.
Good luck - I'm beginning to think these boards just have major trouble with a lot of motherboards or something. |
|
| Back to top |
|
|
xplosiv
n00b
Joined: 26 Jul 2003
Posts: 29
|
| Posted: Sat Jul 26, 2003 1:58 pm Post subject: |
|
|
thanks for the links, I installed those diagnostic tools, and everything seems to be ok, no errors. No luck with the arp thing, I tried downloading the latest starfire driver by replacing the starfire.c file in drivers/net and recompile the kernel, but that didn't work.
I wrote the author of those drivers, but am sure he is too busy writing back, but am pretty desperate right now lol |
|
| Back to top |
|
|
Mnemia
Guru
Joined: 17 May 2002
Posts: 476
|
| Posted: Mon Jul 28, 2003 6:33 pm Post subject: |
|
|
| Let me know if you hear anything from him. If so I'd be interested to know what he said. Not to mention his opinion of these cards. |
|
| Back to top |
|
|
xplosiv
n00b
Joined: 26 Jul 2003
Posts: 29
|
| Posted: Mon Jul 28, 2003 7:05 pm Post subject: |
|
|
| I plugged this card into a cheap p3 clone , and it worked perfectly, no errors, no packet loss, nothing. Incredible. I flashed the bios on my server board, but it didn't help much. It looks like the problem is with the IRQ's, he is helping me trying to track down where the issue is. Very nice guy! |
|
| Back to top |
|
|
BradN
Advocate
Joined: 19 Apr 2002
Posts: 2391
Location: Wisconsin (USA)
|
| Posted: Tue Jul 29, 2003 4:18 am Post subject: |
|
|
| Try moving cards to different PCI slots - this solved some interrupt problems for me once. |
|
| Back to top |
|
|
xplosiv
n00b
Joined: 26 Jul 2003
Posts: 29
|
| Posted: Tue Jul 29, 2003 12:12 pm Post subject: |
|
|
| it's a 1u server, so only 1 pci slot unfortunately. I noticed that /proc/interrupts shows 0 for the adaptec ethernet interfaces, even if some of the packets get through. The on board ethernet interfaces don't have this problem. |
|
| Back to top |
|
|
|
Kernel & Hardware
