[tulip] LNE100TX... not solved

[Donald Becker]

On Fri, 15 Dec 2000, Hank Barta wrote:

>     My clue was that the same card in another system seemed to work
>     fine. (I probably haven't thrashed it enough to really say it
>     works flawlessly.) I then want back to the troublesome system
>     and tried moving the card to another PCI slot and that did the
>     trick. Note that I it still required the 'test' version of the
>     tulip driver.
...
>     FWIW, my motherboard is an ABIT KT-7. Several of the PCI slots
>     are 'shared' with onboard peripherals such as the VIA chip set
>     (USB, IDE, Video.) From reading the ABIT news group, I get the

That's the problem.  We have a bunch of ABit BP-6 boards.  (They are a cheap
way to test software on a large-CPU-count cluster.)  The BP-6 has two slots
that are both nominally "bus masters", but they actually share bus master
pins.  Thus they nominally comply with the PCI bus requirements when tested
individually, but you can't use two bus masters cards in those slots.

>     I thought that problems like this were not 'supposed' to happen.
>     Are there any good resources on identifying and resolving them
>     beyond shooting in the dark with trial and error?

A goal of the PCI specs was to avoid problems such as this.  But it seems
that no matter how the specs are written, there is always some way around
doing the right thing:

char *getcwd(void) { return "."; }

Donald Becker				becker@scyld.com
Scyld Computing Corporation		http://www.scyld.com
410 Severn Ave. Suite 210		Second Generation Beowulf Clusters
Annapolis MD 21403			410-990-9993