Functional Description
R
Intel
®
82845 MCH for SDR Datasheet 113
Table 15. AGP Commands Supported by the Intel
®
MCH When Acting as an AGP Target
MCH Host Bridge AGP
Command
C/BE[3:0]#
Encoding
Cycle Destination Response as PCIx Target
Read 0000 System memory Low-priority read
0000 Hub interface Complete with random data
Hi-Priority Read 0001 System memory High-priority read
0000 The Hub interface Complete with random data
Reserved 0010 N/A No response
Reserved 0011 N/A No response
Write 0100 System memory Low-priority write
0100 Hub interface Cycle goes to DRAM with byte
enables inactive
Hi-Priority Write 0101 System memory High-priority write
0101 Hub interface Cycle goes to DRAM with byte
enables inactive; does not go to
the hub interface
Reserved 0110 N/A No response
Reserved 0111 N/A No response
Long Read 1000 System memory Low-priority read
Hub interface Complete locally with random data;
does not go to the hub interface
Hi-Priority Long
Read
1001 System memory High-priority read
Hub interface Complete with random data
Flush 1010 MCH Complete with QW of random data
Reserved 1011 N/A No response
Fence 1100 MCH No response; Flag inserted in
MCH request queue
Reserved 1101 N/A No response
Reserved 1110 N/A No response
Reserved 1111 N/A No response
NOTES:
1. N/A refers to a function that is not applicable
As a target of an AGP cycle, the MCH supports all the transactions targeting system memory
(summarized in Table 15). The MCH supports both normal and high-priority read and write
requests. The MCH does not support AGP cycles to the hub interface. PIPE# and SBA cycles do
not require coherency management and all AGP initiator accesses to system memory, using AGP
PIPE# or SBA protocol, are treated as non-snoopable cycles. These accesses are directed to the
AGP aperture in system memory that is programmed as either uncacheable (UC) memory or write
combining (WC) in the processor’s MTRRs.