5-18
PCI Interface
4.
PLB master reads are accepted if the PLB write post buffer and the PCI write post buffer are both
empty.
5.8.2.1 PCI Producer-Consumer Model
The PCI Producer-Consumer model is followed with one exception: PCI master reads do not flush PLB
writes and PCI master writes do not cause PLB prefetched read data to be discarded. Thus, if the “flag” is
stored in system memory (PLB side), but the “data” is store in a PCI target, the control software must man-
ually force coherency. This can be done by following two rules:
1.
To ensure data written by a PLB master has reached the intended PCI target, the PLB master
should execute a read from PCI, to any non-destructive address. This is only necessary if the write
is postable.
2.
To ensure data read by a PLB master is current (rather than old prefetched data), the PLB master
should execute a read from PCI to any other non-destructive address. This is only necessary if the
read is prefetchable.
5.8.3 PCI Frequency Options
The PCI bus may be run either synchronously or asynchronously to the processor bus. When asynchro-
nous mode is used, access latencies are increased, typically by 2 to 5 clocks than when in synchronous
mode. Various options are available for selecting PCI operating frequencies through the use of strapping
pins where pullup or pulldown resistors on certain CPC700 I/O signals are read during system reset. The
strapping pins and the corresponding frequency modes are indicated in Table 46.
Note that by placing a pulldown resistor on TSIZ[1] and a pullup on TSIZ[2], the PCI signal M66EN could
be attached to the PCI_66_EN input to allow a system to switch between 33MHz and 66MHz PCI fre-
quency. The PCI_66_EN signal is a dedicated strapping pin on the CPC700.
In synchronous mode the internal PCI logic is clocked from the system input clock. The PCI clock input
should be tied low. The PCI interface will operate at 1/2 of the frequency of the processor bus.
5.8.3.1 Effects on Performance of the Asynchronous Interface
Because of the inherent delays associated with crossing the synchronization interface, as well as other
design limitations, the performance (latency and bandwidth) through the PCI bridge is always less when
run asynchronously than when the busses are run synchronously. The following information summarizes
these differences. The numerical penalties associated with the asynchronous interface are stated as
Table 46.PCI Frequency Modes
CPC700 I/O
Clocking Mode
PCI Frequency Range
TSIZ(1)
TSIZ(2)
PCI_66_EN
Async
25 - 35 MHz
pulldown
don’t care
pulldown
Async
34 - 50 MHz
pulldown
pulldown
pullup
Async
49 - 67 MHz
pulldown
pullup
pullup
Sync (2:1)
25 - 35 MHz
pullup
don’t care
pulldown
Содержание CPC700
Страница 1: ...CPC700 Memory Controller and PCI Bridge User s Manual Version 1 1 Issue Date 3 22 00 Preliminary...
Страница 10: ...Table of Contents x Table of Contents...
Страница 16: ...Tables xvi List of Tables...
Страница 28: ...1 12 CPC700 User s Manual Preliminary...
Страница 72: ...3 36 Processor Interface...
Страница 132: ...4 60 Memory Controller...
Страница 184: ...5 52 PCI Interface...
Страница 194: ...6 10 Clock Power Management and Reset...
Страница 224: ...8 18 IIC...
Страница 244: ...10 10 Interrupt Controller...
Страница 246: ...I 11 2 JTAG...
Страница 250: ...12 4 Processor Local Bus PLB...
Страница 262: ...14 10 Register Summary...
Страница 267: ...CPC700 User s Manual Preliminary...