Memory and I/O Write TLP Processing - 4.4 English

UltraScale Devices Gen3 Integrated Block for PCI Express LogiCORE IP Product Guide (PG156)

Document ID
PG156
Release Date
2024-12-06
Version
4.4 English

When the Endpoint for PCIe receives a Memory or I/O Write TLP, the TLP destination address and transaction type are compared with the values in the core BARs. If the TLP passes this comparison check, the core passes the TLP to the Receive AXI4-Stream interface of the PIO design. The PIO design handles Memory writes and I/O TLP writes in different ways: the PIO design responds to I/O writes by generating a Completion Without Data (cpl), a requirement of the PCI Express specification.

Along with the start of packet, end of packet, and ready handshaking signals, the Completer Requester AXI4-Stream interface also asserts the appropriate (BAR ID[2:0]), Completer Request Descriptor[114:112] signal to indicate to the PIO design the specific destination BAR that matched the incoming TLP. On reception, the PIO design RX State Machine processes the incoming Write TLP and extracts the TLPs data and relevant address fields so that it can pass this along to the PIO design internal block RAM write request controller.

Based on the specific BAR ID[2:0] signals asserted, the RX state machine indicates to the internal write controller the appropriate 2 KB block RAM to use prior to asserting the write enable request. For example, if an I/O Write Request is received by the core targeting BAR0, the core passes the TLP to the PIO design and sets BAR ID[2:0] to 000b. The RX state machine extracts the lower address bits and the data field from the I/O Write TLP and instructs the internal Memory Write controller to begin a write to the block RAM.

In this example, the assertion of setting BAR ID[2:0] to 000b instructed the PIO memory write controller to access ep_mem0 (which by default represents 2 KB of I/O space). While the write is being carried out to the FPGA block RAM, the PIO design RX state machine deasserts m_axis_cq_tready, causing the Receive AXI4-Stream interface to stall receiving any further TLPs until the internal Memory Write controller completes the write to the block RAM. Deasserting m_axis_cq_tready in this way is not required for all designs using the core; the PIO design uses this method to simplify the control logic of the RX state machine.