[1374] in linux-net channel archive
Re: eth0: Bus master arbitration failure, status 88f2.
daemon@ATHENA.MIT.EDU (Dave Platt)
Wed Nov 15 04:32:50 1995
Date: Tue, 14 Nov 95 17:54:57 PST
From: dplatt@rhett.3do.com (Dave Platt)
To: linux-net@vger.rutgers.edu, carlos@snfep1.if.usp.br
> I have a system with Triton chipset and the following cards: Adaptec
> 2940UW, Diamond Stealth Video VRAM and Adtron 650E plus (uses the AMD
> Lance PCI chip). They're all PCI cards; there are no ISA ones.
> I'm getting the msg. eth0: Bus master arbitration failure, status 88f2
> quite often. The ethernet card is on slot 0. If I move it to another
> PCI slot it simply doesn't work.
>
> Any clues? Is this a hardware defect or a problem intrinsic to this
> combination of cards?
I'm not certain. The 88F2 status indicates that a memory-access error
occurred. The PCnet-PCI '970 chip reports a busmaster error if it asks
for PCI-bus memory access with /REQ, and doesn't get a /GNT grant within
50 microseconds. This really shouldn't happen in a PCI machine unless you've
got something pathological taking place.
Any or all of the following might be taking place:
[1] Your motherboard's PCI controller may be defective (either in design, or
just a bit toasted) and it might not be "seeing" the /REQ requests.
The fact that your ethercard doesn't work in other slots is suspicious.
Are all of your PCI slots supposed to be busmaster-capable, or are some
of them slave-only?
Diagnosis method: swap out the motherboard for a different model and
see if things get better.
[2] The Adaptec controller might be hogging the bus in a pathological fashion
(although I'd be surprised if the PCI controller would permit it to do so).
Diagnosis method: swap in a different disk controller model (e.g. an
ISA type).
[3] The Diamond Stealth card might be doing something really strange.
I mention this because certain other PCI busmaster peripherals
(specifically the NCR 53c810 SCSI controllers) seem to have difficulty
working in PCI systems that have certain Diamond Stealth cards installed.
Symbios had to put a workaround in their latest SDMS BIOS for the '810.
Diagnosis method: swap in another video card (preferably _not_ a Diamond).
[4] You might have a bad ethercard. The '970 chip is known to require _very_
careful attention to power-supply and ground-plane layout and bypassing,
or it can run into trouble on either the PCI bus or the Ethernet due to
noise and ground bounce. At least one vendor (Boca, may dogs grub up
their bones) ignored AMD's design recommendations and built their card
in a way which makes it very unreliable. If you don't see at least
half-a-dozen surface-mount chip capacitors near the '970 chip, and at
least a couple of surface-mount tantalums on the board, then your board
may not meet the AMD recommendations and might be noise-prone.
Diagnosis: swap in a different Ethercard (e.g. an ISA-based Lance, which
uses the same driver) and see if matters improve.
Good luck. Chasing down this sort of problem is a rather nasty job.
