Re: More Ethernet card problems.
[Dale, IMHO debian-user is the appropriate list]
Hi,
On Wed, 2 Jun 1999, Dale Scheetz wrote:
[hassle with peecee hardware]
Maybe it's an interrupt conflict. I've actually seen interrupt problems
with 3c509 cards a couple of times. Interrupt problems are often spurious
and give you a hard time figuring out where the problem lies. They
are one of the main reasons why PC hardware sucks, PnP is in fact an
attempt to alleviate the pain (turns out PnP is often a pain itself, oh
well.)
Had the PC been specified in the early eighties to use edge-triggered
interrupts instead of level-triggered interrupts, you'd have no more
interrupt conflicts or shortages.
> I suspected that the card had, somehow, gotten back into PnP mode, so I
> ran the 3c5x9setup program again to make sure. All output from this
> program was the same as before, with no errors.
Are you sure, you've got a true PnP 3c509 card? The ones I've come
accross had a "jumperless mode" which is some 3com-proprietary variant of
PnP and not really the same.
It is sometimes possible that buggy firmware doesn't really disable the
PnP or jumperless mode and still listens to PnP commands. It is unlikely
however. It is really my guess that you have an interrupt conflict. The
situation would then be that some other hardware is trying to use the same
interrupt as the 3c509. This confuses the hardware, the driver or even
both.
Do you run isapnp during boot? Did you set up any PnP cards in
/etc/isapnp.conf? If so, then it might be the case that the PnP hardware,
before getting configured by isapnp, uses some of the resources the 3c509
is trying to use as well. You can check this by taking out the
potentially conflicting hardware or moving the 3c509 to a system without
potentially conflicting other hardware.
> On the possibility that it could be the Windows '95 machine that was at
> fault (the hub reports all cards active once the Debian box appears to be
> configured properly), I booted up the Sparc, which is also on the hub, and
> was able to ping the '95 box, but not the Debian one.
If the interrupts are messed up, the kernel never knows when it should let
the 3c509 driver talk to the hardware, eg. to respond to ICMP (ping)
packets.
> Even though the card tells the hub it is active, and seems to be
> functional from the point of view of the driver, ifconfig, route, and the
> kernel, I can't ping out with it, and the rest of the net can't ping into
> it.
Ifconfig and route and all such tools merely read and write to kernel
state tables, there is no actual communication with the card or the
network. Ping, traceroute, telnet etc. all do actually generate traffic
and it is very well possible that some packets actually get out, you're
just not seeing them back. Actually, the card could be seeing them, it
just doesn't know how to tell the kernel. You can check this condition by
verifying that the light on the back of the card is on, it is a sort of
carrier detect flag. If it's on, the card works physically on the network
side.
> I have my son getting me another Ethernet card, so I can swap out and
> verify that the card _is_ bad, but this is not the most desirable
> solution. (specially when he tells me that the card I got in the kit, with
> two cards and a hub for a bit over $100, will cost over $100 all by
> itself!)
Tsss... Get a cheap ne2000 clone. Should be no more than $10 nowadays.
At 10% of the cost, you get 10% of the hassle compared to 3Com ISA cards.
Or even better, get a PCI ne2000 clone, should cost no more than the ISA
one and you no longer have even potential interrupt problems, since the PC
designers learned after more than 10 years and designed the PCI bus to use
edge-triggered interrupts, so the different slots can share interrupt
lines.
Cheers,
Joost
Reply to: