Athlon Gigabyte GA-7IX mainboard is unstable with 3Com ethernet-cards, partially solved

[Donald Becker]

On 28 Dec 1999, Moritz Franosch wrote:

> We use an AMD Athlon K7 processor with a Gigabyte GA-7IX mainboard.

Some Athalon systems have know reliability problems.

> During transferring about 100 MB over the network by NFS the system
> locked totally (you could not ping it, the keyboard LEDs did not react,
> you could not switch virtual consoles).
> 
> This was the case for both 3Com 3c900 Boomerang and 3Com 3c905C
..
> An SMC Ultra ISA ethernet card worked stable.

The ISA cards use CPU-driven data movement.
All reasonable PCI network adapters use bus mastering.
(The ne2k-pci design is the only one that doesn't, and it does not count as
"reasonable".)

> We thought the reason for the instability could be the bus-master DMA
> feature of the network cards. So we patched the 3c59x driver (the
> patch is appended at the bottom).
> This patch disabled the bus-master DMA. The 3c900 Boomerang works
> stable with it, the 3c905C Tornado does not work at all with the
> patch.

3Com has not supported programmed-I/O since the conversion to the '900
series from the '590 series.  They must have finally removed the circuitry
in the 905C "Tornado" redesign.

Using programmed-I/O is *much* slower than bus mastering.

> The 3c90x driver provided by 3Com
> (http://support.3com.com/infodeli/tools/nic/linuxdownload.htm) shows
> the same symptoms.
> All 3Com-cards work stable on other mainboards.

The great thing about having the 3Com driver available is that if neither
driver work, it's a hardware problem.  The drivers are completely unrelated.
The 3Com people have extensive experience with their hardware, and weird
machines.  I have extensive experience with Linux and typical user problems.

It's very clear that you have a unique hardware problem.

> I have already tried to use the debug-option of the driver, but either 
> I got (almost) no output or far too much.

If the machine just hangs the debug output won't tell you very much.  You
would have to log to the console screen, and even then you might not learn
anything.

> ampa kernel: 3c59x.c:v0.99H 11/17/98 Donald Becker http://cesdis.gsfc.nasa.gov/linux/drivers/vo
> ampa kernel: eth0: 3Com 3c905C Tornado at 0xe400,  00:50:04:ef:bc:9e, IRQ 11

Grrr, the real v0.99H driver does not detect the Tornado.  The v0.99L is the
correct driver.  This, however, is unrelated to your problem.

Donald Becker
Scyld Computing Corporation, and
USRA-CESDIS,   becker@cesdis.gsfc.nasa.gov