[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Intel Triton 430FX chipset and bus mastering PCI cards

Hello PCI Experts.

I have observed a problem with a bus mastering PCI card
when used in any of several PCs based on the Intel 430FX
(Triton) chipset. This problem does NOT occur in PCs based
on the Neptune (older Pentium), 430HX (new Marl Pentium ATX
Motherboard) or 440FX (Pentium Pro ATX motherboard).

The 430FX's PCI arbiter appears (mostly) to respond to bus
requesters with a 2 cycle assertion of GNT#. Occasionally,
however, we observe longer (perhaps 8 cycle) assertions,
but these happen after the bus master's activity, in an
interval before it does another DMA, i.e. when it has NOT
asserted REQ#. It is as if the arbiter is clever, has spotted
the demand from the bus master and tries parking the bus
with it during a quiet period (in terms of Pentium activity) 
in case of further demands, but when nothing happens,
GNT# is recinded. This is a little strange, but in
itself not a problem. What kills the system is when a Pentium
to PCI cycle occurs just prior to the 'unusual' GNT# assertion
(the Pentium accesses the display card - I varied this card
without success). The consequence is that the arbiter
*appears* to crash in the sense that the bus master card cannot
get back on the bus the next time it tries, but no other PCI cycles
occur, and the Pentium cannot do  anything so the PC freezes
completely (it is running Windows 95).

My apologies in advance if this has been dealt with previously.
Any words of wisdom much appreciated.

Adam Barnes
Transtech Parallel Systems