Intel NETWORK PROCESSOR IXP2800 - Manual (Page 205)

Hardware Reference Manual 205
Intel
®
IXP2800 Network Processor
DRAM
When a requestor gets a pull command on the CMD_BUS, the requestor sends the command to the
pull arbiter. This is enqueued into a requestor-dedicated FIFO. The pull request FIFOs are much
smaller than the push request FIFOs because pull requests can request up to 128 bytes of data. It is
eight entries deep and asserts full when it has six entries to account for in-flight requests.
The pull arbiter monitors the heads of each of the three FIFOs. A round robin arbitration scheme is
applied to all valid requests. When a request is granted, the request is completed regardless of how
many data transfers are required. Therefore, one request can take as many as 16 – 32 DRAM
cycles. The push data bus can only use 32 bits when delivering data to the Microengines, PCI, and
the Intel XScale
® core. For these data sources, it takes two cycles to pull every eight bytes
requested; otherwise, it takes only one cycle per eight bytes. On four byte cycles, data is delivered
as pulled.
Figure 73. DRAM Pull Arbiter Functional Blocks
A9733-02
D0_PUSH_REQ
DPXX_PUSH_ID
Round
Robin
A
R
B
I
T
E
R
ME_CLUSTER_0_DATA
D0_PULL_ID
D1_PULL_ID
D2_PULL_ID
D1_PUSH_REQ
D2_PUSH_REQ
ME_CLUSTER_1_DATA
XSCALE*_DATA
* Intel XScale ® Microarchitecture
PCI_PULL_DATA
MSF_PULL_DATA
DPXX_PULL_DATA[63:0]
DPXX_TAKE