82870P2P64H2 Datasheet, PDF (127/217 Pages) –

Functional Description
R
4.2.7
4.2.8
4.2.9
Arbitration Among Multiple Split Completions
The P64H2 will fairly arbitrate amongst all active split completions such that each completion will
get a fair shot at running on PCI. If there are multiple completions waiting to use PCI, the P64H2
will internally arbitrate based upon its MLT value, even if no other agents are requesting on the
bus. Therefore, the P64H2 will end one transaction when its MLT expires, reload, and start
another transaction.
If any particular transaction runs out of data and there are other active transactions to run, the
P64H2 will also switch to the next agent, even if the MLT has not expired for that transaction.
Finally, the prefetch algorithm will be altered such that several transactions can be active at one
time. See Section 3.2.48 for more information on prefetch algorithms.
Transaction Termination as a PCI-X Target
Retry
The P64H2 will retry a cycle when the Split Request queue is full. (i.e. we already have 4 current
and 4 pending Split Transactions). It will always have room to accept a split completion as it has a
dedicated buffer for split completions. It will also retry a cycle when the bus is locked. The P64H2
stores no state from the transaction on a retry.
Split Response
All cycles that cross the bridge receive a split response termination, if they are not retried.
Master-Abort
Any I/O transaction that would cross from PCI-X to either the hub interface or the peer bridge are
not decoded and result in a master abort to the PCI-X initiator.
Arbitration
The P64H2 always parks on the last agent to use PCI. This allows PCI devices operating as a
single stream to stay on the PCI bus for the duration of their transfer.
Intel^® 82870P2 P64H2 Datasheet
127