2. PCI Interface
36
PowerSpan II User Manual
80A1010_MA001_09
Integrated Device Technology
www.idt.com
Writes and reads from one source are queued and arbitrated for the use of the master interface with
DMA in a round robin design. A DMA transaction can be given a lower priority by programming the
DMA Channel Off Counter (OFF) bit in the
“DMA x General Control and Status Register” on
. The OFF bit provides programmable control over the amount of source bus traffic generated
by the DMA channel. The channel interleaves source bus transfers with a period of idle processor bus
clocks where no source bus requests are generated. When source and destination interfaces are
different, 256 bytes of source bus traffic occurs before the idle period. If source and destination
interfaces are the same, 64 bytes of source bus traffic occur before the idle period. This helps prevent
PowerSpan II from interfering with processor bus instruction fetches.
All transactions (writes/reads/DMA) from two source interfaces arbitrate in a round robin scheme on a
per interface basis. Refer to
“Transactions Between the PB Interface and the PCI Interfaces” on
for more information.
2.1.4.4
PCI Transaction Ordering Rules
The
PCI 2.2 Specification
outlines transaction ordering rules for PCI transactions. PowerSpan II does
not comply with the following PCI transaction ordering rules:
•
PowerSpan II only completes the writes that are destined for the same bus as the initiated read
when it is processing a read request. It does not complete writes in both directions before
processing a read request. PowerSpan II does not prioritize writes over reads.
•
PowerSpan II does not allow posted memory writes to pass delayed read requests. This implies that
deadlock conditions may occur when the customer uses bridges that do not support delayed
transactions. Deadlock conditions are broken by the PowerSpan II maximum retry counter.