The Request Agent

The Request Agent Types

There are two types of Request Agents:

• Symmetric Request Agents— Most typically, these are the processors. With regard to FSB arbitration, the symmetric Request Agents have equal importance with respect to each other and use a rotational (symmetrical) priority scheme for FSB arbitration. Note that a custom-designed Request Agent other than a processor could be designed to operate as a symmetric agent. The symmetric agent FSB arbitration scheme supports up to but no more than four symmetric Request Agents in the rotation (eight if Hyper-Threading is enabled in four physical processors on the FSB).

• Priority Request Agents— The system designer may include one or more Request Agents that are not processors (and that don't emulate a symmetric FSB agent). If a Priority Agent is competing against the symmetric agents for bus ownership, it wins and they lose (with one exception that is highlighted in a later chapter).

The Agent ID

The Purpose of the Agent ID

When a Request Agent issues a transaction request, two of the items of information that it provides to the addressed Response Agent are:

• The Request Agent's unique Agent ID.

• A unique transaction ID assigned by the Request Agent.

This information is only used by the Response Agent if it chooses to memorize the transaction and issue a Deferred response to the Request Agent. The Response Agent then processes the read or write request off-line and, when the data has been read or written, it arbitrates for ownership of the FSB and issues a Deferred Reply transaction. Using the agent and transaction IDs delivered in the earlier transaction request as the Deferred Reply transaction's address, it addresses the Request Agent that initially issued the read or write request and completes the transaction. A detailed description of deferred transactions can be found in “Pentium® 4 FSB Transaction Deferral” on page 1277.

How the Agent ID Is Assigned

Each agent that is capable of initiating transaction requests must be assigned a unique agent ID at startup time. The agent typically accomplishes this by sampling one or more of its inputs on the trailing-edge of reset or when the power supply output voltages stabilize (i.e., on the rising-edge of the POWERGOOD signal). The manner in which the processors obtain their agent IDs was covered in “Assignment of IDs to the Processor” on page 860.