Motorola MVME5100 用户手册

下载
页码 330
2-26
Computer Group Literature Center Web Site
Hawk PCI Host Bridge & Multi-Processor Interrupt Controller
2
PCI FIFO
A 64-bit by 16 entry FIFO (4 cache lines total) is used to hold data between 
the PCI Slave and the PPC Master to ensure that optimum data throughput 
is maintained. The same FIFO is used for both read and write transactions. 
A 52-bit by 4 entry FIFO is used to hold command information being 
passed between the PCI Slave and the PPC Master. If write posting is 
enabled, then the maximum number of transactions that may be posted is 
limited by the abilities of either the data FIFO or the command FIFO. For 
example, one burst transaction, 16 double words long, would make the 
data FIFO the limiting factor for write posting. Four single beat 
transactions would make the command FIFO be the limiting factor. If 
either limit is exceeded then any pending PCI transactions are delayed 
(TRDY_ is not asserted) until the PPC Master has completed a portion of 
the previously posted transactions and created some room within the 
command and/or data FIFOs.
PCI Master
The PCI Master, in conjunction with the capabilities of the PPC Slave, 
attempts to move data in either single beat or four-beat (burst) transactions. 
The PCI Master supports 32-bit and 64-bit transactions in the following 
manner:
All PPC60x single beat transactions, regardless of the byte count, 
are subdivided into one or two 32-bit transfers, depending on the 
alignment and the size of the transaction. This includes single beat 
8-byte transactions.
All PPC60x burst transactions are transferred in 64-bit mode if the 
PCI bus has 64-bit mode enabled. If at any time during the 
transaction the PCI target indicates it can not support 64-bit mode, 
the PCI Master continues to transfer the remaining data within that 
transaction in 32-bit mode.
The PCI Master can support Critical Word First (CWF) burst transfers. 
The PCI Master divides this transaction into two parts. The first part starts 
on the address presented with the CWF transfer request and continues up 
to the end of the current cache line. The second transfer starts at the 
beginning of the associated cache line and works its way up to (but not 
including) the word addressed by the CWF request.