R-Tile Avalon® Streaming Intel® FPGA IP for PCI Express* User Guide

ID 683501
Date 6/26/2023
Public

A newer version of this document is available. Customers should click here to go to the newest version.

Document Table of Contents

2.2.3.3.1. Deferrable Memory Write (DMWr)

Deferrable Memory Write (DMWr) transactions are a new type of TLP supported by the PCI Specifications. This new feature allows the completer to return an acknowledgement to the requester of the DMWr transaction and provides the completer a mechanism to temporarily refuse the request. For additional details, refer to the Deferrable Memory Write (DMWr) ECN.

In the R-Tile Avalon Streaming Intel FPGA IP for PCIe, the requirements for supporting this Deferrable Memory Write feature are:
  1. During the credit initialization phase between the IP and the Application logic, the Application logic needs to advertise infinite credits for Non-Posted Data (NPD) transactions to the IP. Note that the Application logic can still throttle the Non-Posted traffic with the credits for the Non-Posted Headers (NPH). Refer to Credit Initialization for more details on the procedure required to advertise infinite credits.
  2. With infinite credits being advertised between the R-Tile Avalon Streaming IP and the Application logic, the minimum size of the Application logic Rx buffer for Non-Posted Data (NPD) transactions needs to be equal to the amount of Non-Posted Header (NPH) advertised credits multiplied by 128 bytes.
Note:
DMWr support is only available in the following OPNs:
  • AGIx027R29AxxxxR2
  • AGIx027R29AxxxxR3
  • AGIx027R29BxxxxR3
  • AGIx023R18AxxxxR0
  • AGIx041R29DxxxxR0
  • AGIx041R29DxxxxR1
For more details on OPN decoding, refer to the Available Options section of the Intel Agilex® 7 FPGAs and SoCs Device Overview.