UG654
Spartan-6 FPGA
Integrated
Endpoint Block for
PCI Express
User Guide
UG654 (v3.0) April 19, 2010
Xilinx is providing this product documentation, hereinafter “Information,” to you “AS IS” with no warranty of any kind, express or implied.
Xilinx makes no representation that the Information, or any particular implementation thereof, is free from any claims of infringement. You
are responsible for obtaining any rights you may require for any implementation based on the Information. All specifications are subject to
change without notice.
XILINX EXPRESSLY DISCLAIMS ANY WARRANTY WHATSOEVER WITH RESPECT TO THE ADEQUACY OF THE INFORMATION OR
ANY IMPLEMENTATION BASED THEREON, INCLUDING BUT NOT LIMITED TO ANY WARRANTIES OR REPRESENTATIONS THAT
THIS IMPLEMENTATION IS FREE FROM CLAIMS OF INFRINGEMENT AND ANY IMPLIED WARRANTIES OF MERCHANTABILITY OR
FITNESS FOR A PARTICULAR PURPOSE.
Except as stated herein, none of the Information may be copied, reproduced, distributed, republished, downloaded, displayed, posted, or
transmitted in any form or by any means including, but not limited to, electronic, mechanical, photocopying, recording, or otherwise, without
the prior written consent of Xilinx.
© Copyright 2009-2010 Xilinx, Inc. XILINX, the Xilinx logo, Virtex, Spartan, ISE, and other designated brands included herein are
trademarks of Xilinx in the United States and other countries. PCI, PCI Express, PCIe, and PCI-X are trademarks of PCI-SIG. All other
trademarks are the property of their respective owners.
Spartan-6 FPGA Integrated Endpoint Block
www.xilinx.com
UG654 (v3.0) April 19, 2010
Revision History
The following table shows the revision history for this document.
Date
Version
Revision
06/24/09
1.0
Initial Xilinx release.
09/16/09
2.0
Updated core to v1.2 and ISE to v11.3. Added VHDL support.
12/02/09
2.5
Added Appendix F, Board Design Guidelines.
04/19/10
3.0
Removed note from System Interface, page 20 and Clocking, page 107 about 250 MHz
being deprecated from future versions. Updated the ISE® software version to 12.1
throughout the document.
Chapter 2: Changed “blocked TLPs” to “blocked Non-Posted TLPs” in the description of
signal trn_rnp_ok_n in Table 2-7. Removed the cfg_err_cpl_unexpect_n signal from
Table 2-9.
Chapter 3: Updated licensing information in accordance with changes to requirements
in ISE v12.1 software.
Chapter 4: Updated simulation tool versions and added ISE Simulator tool to
Example Design Elements. Changed “gate-level simulation” to “functional
simulation” in Setting up for Simulation. Changed “Verilog model” to “Verilog or
VHDL model” in Running the Simulation. Added Using the ISE Project Navigator
GUI Tool section. Added PIO project file to Table 4-1. In Table 4-7, added vhd files;
removed pci_exp_expect_tasks.v and pcie_2_0_v6.v; added
test_interface.vhd, pcie_2_0_v6_rp.v[hd],
pcie_upconfig_fix_3451_v6.v[hd], gtx_rx_valid_filter_v6.v[hd], and
gtx_drp_chanalign_fix_3752_v6.v[hd]; changed board.f entry; and added
simulation and waveform setup scripts. Revised the contents of the tests directory in
Table 4-9. Revised the files for the GTPA1 wrapper in Table 4-10.
Chapter 5: Added sentence about selecting the correct part/package combination to
Xilinx Reference Boards. Replaced Figure 5-9.
Chapter 6: Added bullet about completions in response to user-implemented
Configuration Space requests on page 65. Changed “should not assert trn_tsrc_rdy_n if
trn_tstr_n is asserted” to “should not deassert trn_tsrc_rdy_n during the middle of a
transfer if trn_tstr_n is asserted” in Source Throttling on the Transmit Datapath.
Updated Figure 6-24. Removed signal cfg_err_cpl_unexpected_n from Table 6-14. In
Table 6-15, added table note defining checkmarks and X’s and removed “Unexpected
Completion” column. Added sentence about other types of completions to Unexpected
Completions.
Chapter 7: Updated description of the GTPA1_DUAL tile in Core I/O Assignments.
Removed “mgt/” from the path names in the examples in Required Modifications and
Core I/O Assignments. Added “(0)” to the NET statement on page 114. Added
introductory paragraph before Table 7-1.
Chapter 8: Revised Figure 8-1. Added paragraph about FPGA power assumptions to
Board Power in Real-World Systems on page 121. Added Hot-Plug Systems subsection.
Chapter 9: Changed this chapter’s location in document. Added Master Data Parity
Error Bit Set Incorrectly and Non-Posted UpdateFC During PPM Transition sections.
Added PCI Configuration Space Header address to Master Data Parity Error Bit Set
Incorrectly. Updated Detailed Description section in Non-Posted UpdateFC During
PPM Transition.
Appendix B: Removed note about only Verilog source being available from page 143.
Added ISE simulator as fourth simulator to Simulating the Design. In Verilog Test
Selection, changed the test file name to tests.v and added an ISim example. Added
ISim to Table B-2.
UG654 (v3.0) April 19, 2010
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
Spartan-6 FPGA Integrated Endpoint Block
www.xilinx.com
UG654 (v3.0) April 19, 2010
Table of Contents
Revision History . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Preface: About This Guide
Guide Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
Additional Documentation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
Additional Resources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
Chapter 1: Introduction
About the Core . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
System Requirements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Recommended Design Experience . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Additional Core Resources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
13
13
14
14
Chapter 2: Core Overview
Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Protocol Layers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
PCI Configuration Space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Core Interfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Transaction Interface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Configuration Interface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Error Reporting Signals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
15
16
18
20
20
24
28
Chapter 3: Licensing the Core
Chapter 4: Getting Started Example Design
Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Generating the Core . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Simulating the Example Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Implementing the Example Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Directory Structure and File Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
33
36
38
39
44
Chapter 5: Generating and Customizing the Core
Customizing the Core through the CORE Generator Software . . . . . . . . . . . . . . . . 49
Chapter 6: Designing with the Core
Design with Configuration Space Registers and Configuration Interface . . . . . 87
Additional Packet Handling Requirements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
Reporting User Error Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
5
Flow Control Credit Information . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
Power Management . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
Generating Interrupt Requests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
Clocking and Reset of the Integrated Endpoint Block Core . . . . . . . . . . . . . . . . . . 107
Chapter 7: Core Constraints
Contents of the User Constraints File . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Required Modifications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Device Selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Core I/O Assignments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Core Physical Constraints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Core Timing Constraints . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Relocating the Integrated Endpoint Block . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Supported Core Pinouts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
111
112
112
113
113
113
114
114
Chapter 8: FPGA Configuration
Configuration Terminology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Configuration Access Time . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Board Power in Real-World Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Recommendations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
117
118
120
121
Chapter 9: Known Restrictions
Master Data Parity Error Bit Set Incorrectly . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
Non-Posted UpdateFC During PPM Transition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
Appendix A: Programmed Input/Output Example Design
System Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
PIO Hardware . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
PIO Operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
129
130
141
142
Appendix B: Root Port Model Test Bench
Architecture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Simulating the Design . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Test Selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Waveform Dumping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Output Logging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Parallel Test Programs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Test Description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Expanding the Root Port Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
6
www.xilinx.com
144
145
146
148
149
149
150
153
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Appendix C: Migration Considerations
Integrated PHY. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
System Clocking and Reset . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Interface Changes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Block RAM Settings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Signal Change Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
163
163
163
165
165
Appendix D: Debugging Designs
Finding Help on Xilinx.com . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Contacting Xilinx Technical Support . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Debug Tools . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Debug Ports . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Hardware Debug . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Simulation Debug . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
167
169
169
173
175
185
Appendix E: Managing Receive-Buffer Space for Inbound Completions
General Considerations and Concepts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
Methods of Managing Completion Space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 191
Appendix F: Board Design Guidelines
Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Example PCB Reference . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Board Stackup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Data Routing Guidelines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Reference Clock Considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
195
195
196
197
201
Appendix G: PCIE_A1 Port Descriptions
Clock and Reset Interface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Transaction Layer Interface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Block RAM Interface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
GTP Transceiver Interface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Configuration Management Interface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
Debug Interface Ports . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
205
206
209
209
212
222
Appendix H: PCIE_A1 Attribute Descriptions
Appendix I: PCIE_A1 Timing Parameter Descriptions
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
7
8
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Preface
About This Guide
This user guide describes the function and operation of the Spartan®-6 FPGA Integrated
Endpoint Block for PCI Express® core, including how to design, customize, and
implement the core.
Guide Contents
This manual contains these chapters and appendices:
•
Chapter 1, Introduction, describes the core and related information, including
recommended design experience, additional resources, technical support, and
submitting feedback to Xilinx.
•
Chapter 2, Core Overview, describes the main components of the integrated Endpoint
block core architecture.
•
Chapter 3, Licensing the Core, contains information about licensing the core.
•
Chapter 4, Getting Started Example Design, provides instructions for quickly
generating, simulating, and implementing the example design using the
demonstration test bench.
•
Chapter 5, Generating and Customizing the Core, describes how to use the graphical
user interface (GUI) to configure the integrated Endpoint block using the
CORE Generator™ software.
•
Chapter 6, Designing with the Core, provides instructions on how to design a device
using the integrated Endpoint block core.
•
Chapter 7, Core Constraints, discusses the required and optional constraints for the
integrated Endpoint block core.
•
Chapter 8, FPGA Configuration, discusses considerations for FPGA configuration
and PCI Express.
•
Chapter 9, Known Restrictions, describes any known restrictions for this core.
•
Appendix A, Programmed Input/Output Example Design, describes the
Programmed Input/Output (PIO) example design for use with the core.
•
Appendix B, Root Port Model Test Bench, describes the test bench environment,
which provides a test program interface for use with the PIO example design.
•
Appendix C, Migration Considerations, defines the differences in behaviors and
options between the integrated Endpoint block and Endpoint PIPE core.
•
Appendix D, Debugging Designs, provides information on resources available on the
Xilinx support website, available debug tools, and a step-by-step process for
debugging designs that use the Spartan-6 FPGA Integrated Endpoint Block for PCI
Express.
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
9
Preface: About This Guide
•
Appendix E, Managing Receive-Buffer Space for Inbound Completions, provides
example methods for handling finite receive buffer space for inbound completions
with regards to the PCI Express Endpoint requirement to advertise infinite
completion credits.
•
Appendix F, Board Design Guidelines, discusses topics related to implementing a
PCI Express design that uses the Spartan-6 FPGA on a printed circuit board.
•
Appendix G, PCIE_A1 Port Descriptions.
•
Appendix H, PCIE_A1 Attribute Descriptions.
•
Appendix I, PCIE_A1 Timing Parameter Descriptions.
Additional Documentation
These documents are also available for download at:
http://www.xilinx.com/products/spartan6.
•
Spartan-6 Family Overview
This overview outlines the features and product selection of the Spartan-6 family.
•
Spartan-6 FPGA Data Sheet: DC and Switching Characteristics
This data sheet contains the DC and Switching Characteristic specifications for the
Spartan-6 family.
•
Spartan-6 FPGA Packaging and Pinout Specifications
This specification includes the tables for device/package combinations and maximum
I/Os, pin definitions, pinout tables, pinout diagrams, mechanical drawings, and
thermal specifications.
•
Spartan-6 FPGA SelectIO Resources User Guide
This guide describes the SelectIO™ resources available in all Spartan-6 devices.
•
Spartan-6 FPGA Clocking Resources User Guide
This guide describes the clocking resources available in all Spartan-6 devices,
including the DCMs and the PLLs.
•
Spartan-6 FPGA Block RAM Resources User Guide
This guide describes the Spartan-6 device block RAM capabilities.
•
Spartan-6 FPGA Configurable Logic Blocks User Guide
This guide describes the capabilities of the configurable logic blocks (CLBs) available
in all Spartan-6 devices.
•
Spartan-6 FPGA Memory Controller User Guide
This guide describes the Spartan-6 FPGA memory controller block, a dedicated,
embedded multi-port memory controller that greatly simplifies interfacing Spartan-6
FPGAs to the most popular memory standards.
•
Spartan-6 FPGA GTP Transceivers User Guide
This guide describes the GTP transceivers available in Spartan-6 LXT FPGAs.
•
Spartan-6 FPGA DSP48A1 Slice User Guide
This guide describes the architecture of the DSP48A1 slice in Spartan-6 FPGAs and
provides configuration examples.
10
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Additional Resources
•
Spartan-6 FPGA PCB Designer’s Guide
This guide provides information on PCB design for Spartan-6 devices, with a focus on
strategies for making design decisions at the PCB and interface level.
Additional Resources
To find additional documentation, see the Xilinx website at:
http://www.xilinx.com/support/documentation/index.htm.
To search the Answer Database of silicon, software, and IP questions and answers, or to
create a technical support WebCase, see the Xilinx website at:
http://www.xilinx.com/support/mysupport.htm.
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
11
Preface: About This Guide
12
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Chapter 1
Introduction
This chapter introduces the Spartan®-6 FPGA Integrated Endpoint Block for PCI Express®
core and provides related information including system requirements, recommended
design experience, additional core resources, technical support, and submitting feedback
to Xilinx.
About the Core
The Spartan-6 FPGA Integrated Endpoint Block for PCI Express core is a reliable, highbandwidth, scalable serial interconnect building block for use with the Spartan-6 FPGA
family. The core instantiates the Spartan-6 FPGA Integrated Endpoint Block for PCI
Express found in the Spartan-6 family, and supports both Verilog-HDL and VHDL.
The Spartan-6 FPGA Integrated Endpoint Block for PCI Express core is a
CORE Generator™ IP core, included in the latest IP Update on the Xilinx IP Center. For
detailed information about the core, see the Spartan-6 FPGA Integrated Endpoint Block for
PCI Express product page. For information about licensing options, see Chapter 3,
Licensing the Core.
System Requirements
Windows
•
Windows XP Professional 32-bit/64-bit
•
Windows Vista Business 32-bit/64-bit
Linux
•
Red Hat Enterprise Linux WS v4.0 32-bit/64-bit
•
Red Hat Enterprise Desktop v5.0 32-bit/64-bit (with Workstation Option)
•
SUSE Linux Enterprise (SLE) v10.1 32-bit/64-bit
Software
•
ISE® v12.1 software or later
Check the release notes for the required Service Pack; ISE software Service Packs can be
downloaded from www.xilinx.com/support/download/index.htm.
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
13
Chapter 1: Introduction
Recommended Design Experience
Although the Spartan-6 FPGA Integrated Endpoint Block for PCI Express core is a fully
verified solution, the challenge associated with implementing a complete design varies
depending on the configuration and functionality of the application. For best results,
previous experience building high-performance, pipelined FPGA designs using Xilinx
implementation software and User Constraints Files (UCF) is recommended.
Additional Core Resources
For detailed information and updates about the integrated Endpoint block core, see these
documents:
•
LogiCORE™ IP Spartan-6 FPGA Integrated Endpoint Block for PCI Express Data Sheet
•
LogiCORE IP Spartan-6 FPGA Integrated Endpoint Block for PCI Express Release Notes
Additional information and resources related to the PCI Express technology are available
from the following websites:
14
•
PCI Express at PCI-SIG
•
PCI Express Developer’s Forum
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Chapter 2
Core Overview
This chapter describes the main components of the Spartan®-6 FPGA Integrated Endpoint
Block for PCI Express® core architecture.
Overview
Table 2-1 defines the Spartan-6 FPGA Integrated Endpoint Block for PCI Express solution.
Table 2-1:
Product Overview
Product Name
1-lane Integrated
Endpoint Block
User
PCI Express Base
FPGA
Lane Widths Link Speeds
Interface
Specification
Supported
Supported
Architecture
Width
Compliance
Spartan-6
32
x1
2.5 Gb/s
v1.1
The Spartan-6 FPGA Integrated Endpoint Block for PCI Express core internally instantiates
the Spartan-6 FPGA Integrated Endpoint Block for PCI Express. See Appendix E,
Managing Receive-Buffer Space for Inbound Completions, Appendix G, PCIE_A1 Port
Descriptions, and Appendix H, PCIE_A1 Attribute Descriptions, for information about the
software primitive, PCIE_A1, which represents the hardened-IP integrated Endpoint
block. The integrated Endpoint block follows the PCI Express Base Specification layering
model, which consists of the Physical, Data Link, and Transaction Layers.
Figure 2-1 illustrates the interfaces to the core, as defined below:
•
System (SYS) interface
•
PCI Express (PCI_EXP) interface
•
Configuration (CFG) interface
•
Transaction (TRN) interface
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
15
Chapter 2: Core Overview
X-Ref Target - Figure 2-1
LogiCORE IP Spartan-6 FPGA
Integrated Endpoint Block for PCI Express
User
Logic
Host
Interface
Transaction
(TRN)
Configuration
(CFG)
PCI Express
(PCI_EXP)
Spartan-6 FPGA
Integrated Endpoint
Block for
PCI Express
(PCIE_A1)
Transceiver
System
(SYS)
Figure 2-1:
PCI
Express
Fabric
Clock
and
Reset
Top-Level Functional Blocks and Interfaces
The core uses packets to exchange information between the various modules. Packets are
formed in the Transaction and Data Link Layers to carry information from the transmitting
component to the receiving component. Necessary information is added to the packet
being transmitted, which is required to handle the packet at those layers. At the receiving
end, each layer of the receiving element processes the incoming packet, strips the relevant
information and forwards the packet to the next layer.
As a result, the received packets are transformed from their Physical Layer representation
to their Data Link Layer representation and the Transaction Layer representation.
Protocol Layers
The functions of the protocol layers, as defined by the PCI Express Base Specification, include
generation and processing of Transaction Layer Packets (TLPs), flow control management,
initialization, power management, data protection, error checking and retry, physical link
interface initialization, maintenance and status tracking, serialization, deserialization and
other circuitry for interface operation. Each layer is defined in the remainder of this
section.
Transaction Layer
The Transaction Layer is the upper layer of the PCI Express architecture, and its primary
function is to accept, buffer, and disseminate Transaction Layer packets or TLPs. TLPs
communicate information through the use of memory, I/O, configuration, and message
transactions. To maximize the efficiency of communication between devices, the
Transaction Layer enforces PCI compliant Transaction ordering rules and manages TLP
buffer space via credit-based flow control.
Data Link Layer
The Data Link Layer acts as an intermediate stage between the Transaction Layer and the
Physical Layer. Its primary responsibility is to provide a reliable mechanism for the
exchange of TLPs between two components on a link.
16
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Protocol Layers
Services provided by the Data Link Layer include data exchange (TLPs), error detection
and recovery, initialization services and the generation and consumption of Data Link
Layer Packets (DLLPs). DLLPs are used to transfer information between Data Link Layers
of two directly connected components on the link. DLLPs convey information such as
Power Management, Flow Control, and TLP acknowledgments.
Physical Layer
The Physical Layer interfaces the Data Link Layer with signalling technology for link data
interchange, and is subdivided into the Logical sub-block and the Electrical sub-block.
•
The Logical sub-block is responsible for framing and deframing of TLPs and DLLPs. It
also implements the Link Training and Status State machine (LTSSM) which handles
link initialization, training, and maintenance. Scrambling, descrambling and 8B/10B
encoding and decoding of data is also performed in this sub-block.
•
The Electrical sub-block defines the input and output buffer characteristics that
interfaces the device to the PCIe® link.
The Physical Layer also supports Lane Polarity Inversion, as indicated in the PCI Express
Base Specification rev 1.1 requirement.
Configuration Management
The Configuration Management layer maintains the PCI Type0 Endpoint configuration
space and supports these features:
•
Implements PCI Configuration Space
•
Supports Configuration Space accesses
•
Power Management functions
•
Implements error reporting and status functionality
•
Implements packet processing functions
•
Receive
-
•
Configuration Reads and Writes
Transmit
-
Completions with or without data
-
TLM Error Messaging
-
User Error Messaging
-
Power Management Messaging/Handshake
•
Implements MSI and INTx interrupt emulation
•
Implements the Device Serial Number Capability in the PCIe Extended Capability
space
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
17
Chapter 2: Core Overview
PCI Configuration Space
The configuration space consists of three primary parts, illustrated in Table 2-4. These
include:
•
Legacy PCI v3.0 Type 0 Configuration Header
•
Legacy Extended Capability Items
•
•
PCIe Capability Item
•
Power Management Capability Item
•
Message Signaled Interrupt (MSI) Capability Item
PCIe Extended Capabilities
•
Device Serial Number Extended Capability Structure (optional)
The core implements three legacy extended capability items. The remaining legacy
extended capability space from address 0x6C to 0xFF is reserved. The core returns
0x00000000 when this address range is read.
The core also optionally implements one PCIe Extended Capability. The remaining PCIe
Extended Capability space is reserved. If the Device Serial Number Capability is
implemented, addresses from 0x10C to 0xFFF are reserved; otherwise addresses from
0x104 to 0xFFF are reserved. The core returns a Completion with Data of 0x00000000 if
there is a configuration read to addresses in the reserved space range; writes are ignored.
18
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
PCI Configuration Space
Table 2-2:
PCI Configuration Space Header
31
16
15
0
Device ID
Vendor ID
000h
Status
Command
004h
Class Code
BIST
Header
Lat Timer
Rev ID
008h
Cache Ln
00Ch
Base Address Register 0
010h
Base Address Register 1
014h
Base Address Register 2
018h
Base Address Register 3
01Ch
Base Address Register 4
020h
Base Address Register 5
024h
Cardbus CIS Pointer
028h
Subsystem ID
Subsystem Vendor ID
Expansion ROM Base Address
Reserved
030h
CapPtr
Reserved
Max Lat
Min Gnt
PM Capability
Data
BSE
MSI Control
038h
Intr Pin
Intr Line
03Ch
NxtCap
PM Cap
040h
PMCSR
NxtCap
MSI Cap
048h
Message Address (Upper)
050h
Message Data
NxtCap
PE Cap
PCI Express Device Capabilities
Device Status
Device Control
PCI Express Link Capabilities
Link Status
Link Control
Reserved Legacy Configuration
Space (Returns 0x00000000)
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
044h
04Ch
PE Capability
Next Cap
034h
Message Address (Lower)
Reserved
Optional
Returns 0 if not
implemented
02Ch
Capability
Version
PCI Express Extended
Capability - DSN
054h
058h
05Ch
060h
064h
068h
06Ch0FFh
100h
PCI Express Device Serial Number (1st)
104h
PCI Express Device Serial Number (2nd)
108h
Reserved Extended Configuration
Space (Returns Completion with 0x00000000)
10ChFFFh
www.xilinx.com
19
Chapter 2: Core Overview
Core Interfaces
The Integrated Endpoint Block for PCI Express core includes top-level signal interfaces
that have sub-groups for the receive direction, transmit direction, and signals common to
both directions.
System Interface
The System (SYS) interface consists of the system reset signal, sys_reset_n, the system clock
signal, sys_clk, and a hot reset indicator, received_hot_reset, as described in Table 2-3.
Table 2-3:
System Interface Signals
Function
Signal Name
Direction
System Reset
sys_reset_n
Input
Asynchronous, active Low signal.
System Clock
sys_clk
Input
Reference clock: 100 or 125 MHz.
received_hot_reset
Output
Hot Reset
Description
The core received a hot reset.
The system reset signal is an asynchronous active-Low input. The assertion of sys_reset_n
causes a hard reset of the entire core. The system input clock must be either 100 MHz or
125 MHz, as selected in the CORE Generator software GUI.
PCI Express Interface
The PCI Express (PCI_EXP) interface consists of differential transmit and receive pairs. A
PCI Express lane consists of a pair of transmit differential signals {pci_exp_txp,
pci_exp_txn} and a pair of receive differential signals {pci_exp_rxp, pci_exp_rxn}. The
1-lane core supports only Lane 0. Transmit and receive signals of the PCI_EXP interface are
defined in Table 2-4.
Table 2-4:
PCI Express Interface Signals for the 1-lane Endpoint Core
Lane
Number
Name
Direction
Description
0
pci_exp_txp0
Output
PCI Express Transmit Positive: Serial Differential
Output 0 (+)
0
pci_exp_txn0
Output
PCI Express Transmit Negative: Serial
Differential Output 0 (–)
0
pci_exp_rxp0
Input
PCI Express Receive Positive: Serial Differential
Input 0 (+)
0
pci_exp_rxn0
Input
PCI Express Receive Negative: Serial Differential
Input 0 (-)
Transaction Interface
The Transaction (TRN) interface provides a mechanism for the user design to generate and
consume TLPs. The signal descriptions for this interface are provided in Table 2-5,
Table 2-6, and Table 2-7.
20
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Transaction Interface
Common TRN Interface
Table 2-5 defines the common TRN interface signals.
Table 2-5:
Common Transaction Interface Signals
Name
trn_clk
Direction
Description
Output Transaction Clock: Transaction and Configuration interface
operations are referenced to and synchronous with the rising
edge of this clock. trn_clk is unavailable when the core
sys_reset_n is held asserted. trn_clk is guaranteed to be stable at
the nominal operating frequency only after trn_reset_n is
deasserted. The trn_clk clock output is a fixed frequency clock
output. trn_clk does not change frequencies in case of link
recovery.
• 1-lane Integrated Endpoint Block Frequency: 62.5 MHz
trn_reset_n
Output Transaction Reset: Active Low. User logic interacting with the
Transaction and Configuration interfaces must use trn_reset_n to
return to its quiescent state. trn_reset_n is deasserted
synchronously with respect to trn_clk, trn_reset_n is asserted
asynchronously with sys_reset_n assertion. The trn_reset_n
signal is asserted for core in-band reset events like Hot Reset or
Link Disable.
trn_lnk_up_n
Output Transaction Link Up: Active Low. Transaction link-up is asserted
when the core and the connected upstream link partner port are
ready and able to exchange data packets. Transaction link-up is
deasserted when the core and link partner are attempting to
establish communication, or when communication with the link
partner is lost due to errors on the transmission channel.
trn_lnk_up_n is also deasserted when the core is driven to Hot
Reset or Link Disable states by the link partner, and all TLPs
stored in the core are lost.
trn_fc_sel[2:0]
Input
Flow Control Informational Select: Selects the type of flow
control information presented on the trn_fc_* signals. Possible
values:
000 == receive buffer available space
001 == receive credits granted to the link partner
010 == receive credits consumed
100 == transmit user credits available
101 == transmit credit limit
110 == transmit credits consumed
trn_fc_ph[7:0]
Output Posted Header Flow Control Credits: The number of Posted
Header FC credits for the selected flow control type
trn_fc_pd[11:0]
Output Posted Data Flow Control Credits: The number of Posted Data
FC credits for the selected flow control type.
trn_fc_nph[7:0]
Output Non-Posted Header Flow Control Credits: The number of NonPosted Header FC credits for the selected flow control type.
trn_fc_npd[11:0]
Output Non-Posted Data Flow Control Credits: The number of NonPosted Data FC credits for the selected flow control type.
trn_fc_cplh[7:0]
Output Completion Header Flow Control Credits: The number of
Completion Header FC credits for the selected flow control type.
trn_fc_cpld[11:0]
Output Completion Data Flow Control Credits: The number of
Completion Data FC credits for the selected flow control type.
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
21
Chapter 2: Core Overview
Transmit TRN Interface
Table 2-6 defines the transmit (Tx) TRN interface signals.
Table 2-6:
Transaction Transmit Interface Signals
Name
Description
trn_tsof_n
Input
Transmit Start-of-Frame (SOF): Active Low. Signals the
start of a packet. Valid only along with assertion of
trn_tsrc_rdy_n.
trn_teof_n
Input
Transmit End-of-Frame (EOF): Active Low. Signals the
end of a packet. Valid only along with assertion of
trn_tsrc_rdy_n.
trn_td[31:0]
Input
Transmit Data: Packet data to be transmitted.
trn_tsrc_rdy_n
Input
Transmit Source Ready: Active Low. Indicates that the
user application is presenting valid data on trn_td[31:0].
trn_tdst_rdy_n
Output
Transmit Destination Ready: Active Low. Indicates that
the core is ready to accept data on trn_td[31:0]. The
simultaneous assertion of trn_tsrc_rdy_n and
trn_tdst_rdy_n marks the successful transfer of one data
beat on trn_td[31:0].
trn_tsrc_dsc_n
Input
Transmit Source Discontinue: Active Low. Can be
asserted any time starting on the first cycle after SOF to
EOF, inclusive.
trn_tbuf_av[5:0]
Output
Transmit Buffers Available: Indicates the number of
transmit buffers available in the core. Each free transmit
buffer can accommodate one TLP up to the supported
Maximum Payload Size. Maximum number of Transmit
buffers is determined by the Supported Maximum
Payload Size and block RAM configuration selected.
trn_terr_drop_n
Output
Transmit Error Drop: Active Low. Indicates that the core
discarded a packet because of a length violation or, when
streaming, data was not presented on consecutive clock
cycles. Length violations include packets longer than
supported.
Input
Transmit Streamed: Active Low. Indicates a packet is
presented on consecutive clock cycles and transmission
on the link can begin before the entire packet has been
written to the core. Commonly referred to as transmit cutthrough mode.
trn_tstr_n
trn_tcfg_req_n
22
Direction
Output
Transmit Configuration Request: Active Low. Asserted
when the core is ready to transmit a Configuration
Completion or other internally-generated TLP.
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Transaction Interface
Table 2-6:
Transaction Transmit Interface Signals (Cont’d)
Name
Direction
Description
trn_tcfg_gnt_n
Input
Transmit Configuration Grant: Active Low. Asserted by
the user application in response to trn_tcfg_req_n, to
allow the core to transmit an internally generated TLP.
Holding trn_tcfg_gnt_n deasserted after trn_tcfg_req_n
allows user-initiated TLPs to be given higher priority of
transmission over core generated TLPs. trn_tcfg_req_n is
asserted once for each internally generated packet. It
cannot be deasserted immediately following
trn_cfg_gnt_n if there are no transmit buffers available. If
the user does not wish to alter the prioritization of the
transmission of internally generated TLPs, this signal can
be continuously asserted.
trn_terrfwd_n
Input
Transmit Error Forward: Active Low. This input marks
the current packet in progress as error-poisoned. It can be
asserted any time between SOF and EOF, inclusive.
trn_terrfwd_n must not be asserted if trn_tstr_n is
asserted.
Receive TRN Interface
Table 2-7 defines the receive (RX) TRN interface signals.
Table 2-7:
Receive Transaction Interface Signals
Name
Direction
Description
trn_rsof_n
Output
Receive Start-of-Frame (SOF): Active Low.
Signals the start of a packet. Valid only if
trn_rsrc_rdy_n is also asserted.
trn_reof_n
Output
Receive End-of-Frame (EOF): Active Low.
Signals the end of a packet. Valid only if
trn_rsrc_rdy_n is also asserted.
trn_rd[31:0]
Output
Receive Data: Packet data being received. Valid
only if trn_rsrc_rdy_n is also asserted.
trn_rerrfwd_n
Output
Receive Error Forward: Active Low. Marks the
packet in progress as error poisoned. Asserted by
the core for the entire length of the packet.
trn_rsrc_rdy_n
Output
Receive Source Ready: Active Low. Indicates the
core is presenting valid data on trn_rd[31:0]
trn_rdst_rdy_n
Input
Receive Destination Ready: Active Low.
Indicates the user application is ready to accept
data on trn_rd[31:0]. The simultaneous assertion
of trn_rsrc_rdy_n and trn_rdst_rdy_n marks the
successful transfer of one data beat on
trn_td[31:0].
trn_rsrc_dsc_n
Output
Receive Source Discontinue: Active Low.
Indicates the core is aborting the current packet.
Asserted when the physical link is going into
reset.
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
23
Chapter 2: Core Overview
Table 2-7:
Receive Transaction Interface Signals (Cont’d)
Name
trn_rnp_ok_n
Direction
Description
Input
Receive Non-Posted OK: Active Low. The user
application asserts this signal when it is ready to
accept a Non-Posted Request TLP. trn_rnp_ok_n
must be deasserted when the user application
cannot process received Non-Posted TLPs, so
that these can be buffered within the core's
receive queue. In this case, Posted and
Completion TLPs received after the Non-Posted
TLPs will bypass the blocked Non-Posted TLPs.
When the user application approaches a state
where it is unable to service Non-Posted
Requests, it must deassert trn_rnp_ok_n one
clock cycle before the core presents EOF of the
next-to-last Non-Posted TLP the user application
can accept.
trn_rbar_hit_n[6:0]
Output
Receive BAR Hit: Active Low. Indicates BAR(s)
targeted by the current receive transaction.
Asserted throughout the packet, from trn_rsof_n
to trn_reof_n.
•
•
•
•
•
•
•
trn_rbar_hit_n[0] => BAR0
trn_rbar_hit_n[1] => BAR1
trn_rbar_hit_n[2] => BAR2
trn_rbar_hit_n[3] => BAR3
trn_rbar_hit_n[4] => BAR4
trn_rbar_hit_n[5] => BAR5
trn_rbar_hit_n[6] => Expansion ROM
Address.
If two BARs are configured into a single 64-bit
address, both corresponding trn_rbar_hit_n bits
are asserted.
Configuration Interface
The Configuration (CFG) interface enables the user design to inspect the state of the
Endpoint for PCIe configuration space. The user provides a 10-bit configuration address,
which selects one of the 1024 configuration space double word (DWORD) registers. The
endpoint returns the state of the selected register over the 32-bit data output port. Table 2-8
defines the Configuration interface signals. See Design with Configuration Space Registers
and Configuration Interface, page 87 for usage.
24
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Configuration Interface
Table 2-8:
Configuration Interface Signals
Name
Direction
Description
cfg_do[31:0]
Output
Configuration Data Out: A 32-bit data output port
used to obtain read data from the configuration
space inside the core.
cfg_rd_wr_done_n
Output
Configuration Read Write Done: Active-Low, readwrite done signal indicates a successful completion
of the user configuration register access operation.
• For a user configuration register read operation,
the signal validates the cfg_do[31:0] data-bus
value.
• Writes to the configuration space are not
supported.
cfg_dwaddr[9:0]
Input
Configuration DWORD Address: A 10-bit address
input port used to provide a configuration register
DWORD address during configuration register
accesses.
cfg_rd_en_n
Input
Configuration Read Enable: Active Low readenable for configuration register access.
Note: cfg_rd_en_n must be asserted for no more
than 1 trn_clk cycle for each access.
cfg_interrupt_n
cfg_interrupt_rdy_n
cfg_interrupt_assert_n
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Input
Configuration Interrupt: Active-Low interrupt
request signal. The user application can assert this
to cause the selected interrupt message type to be
transmitted by the core. The signal should be held
Low until cfg_interrupt_rdy_n is asserted.
Output
Configuration Interrupt Ready: Active-Low
interrupt grant signal. The simultaneous assertion
of cfg_interrupt_rdy_n and cfg_interrupt_n
indicates that the core has successfully transmitted
the requested interrupt message.
Input
www.xilinx.com
Configuration Legacy Interrupt Assert/Deassert
Select: Selects between Assert and Deassert
messages for Legacy interrupts when
cfg_interrupt_n is asserted. Not used for MSI
interrupts.
Value Message Type
0
Assert
1
Deassert
25
Chapter 2: Core Overview
Table 2-8:
Configuration Interface Signals (Cont’d)
Name
cfg_interrupt_di[7:0]
Direction
Description
Input
Configuration Interrupt Data In: For Message
Signaling Interrupts (MSI), the portion of the
Message Data that the endpoint must drive to
indicate MSI vector number, if Multi-Vector
Interrupts are enabled. The value indicated by
cfg_interrupt_mmenable[2:0] determines the
number of lower-order bits of Message Data that
the endpoint provides; the remaining upper bits of
cfg_interrupt_di[7:0] are not used.
For Single-Vector Interrupts, cfg_interrupt_di[7:0]
is not used. For Legacy interrupt messages
(Assert_INTx, Deassert_INTx), this list defines the
type of message to be sent:
Value
Legacy Interrupt
00h
INTA
01h
INTB
02h
INTC
03h
INTD
cfg_interrupt_do[7:0]
Output
Configuration Interrupt Data Out: The value of the
lowest eight bits of the Message Data field in the
endpoint's MSI capability structure. This value is
not used and is provided for informational
purposes and backwards compatibility.
cfg_interrupt_mmenable[2:0]
Output
Configuration Interrupt Multiple Message Enable:
This is the value of the Multiple Message Enable
field and defines the number of vectors the system
allows for multi-vector MSI. Values range from
000b to 101b. A value of 000b indicates that single
vector MSI is enabled, while other values indicate
the number of lower-order bits that can be used for
cfg_interrupt_di[7:0].
cfg_interrupt_mmenable[2:0] values:
•
•
•
•
•
•
cfg_interrupt_msienable
26
Output
www.xilinx.com
000b, 0 bits
001b, 1 bit
010b, 2 bits
011b, 3 bits
100b, 4 bits
101b, 5 bits
Configuration Interrupt MSI Enabled: Indicates
that the Message Signaling Interrupt (MSI)
messaging is enabled. If 0, then only Legacy (INTx)
interrupts can be sent. If 1, only MSI interrupts can
be sent.
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Configuration Interface
Table 2-8:
Configuration Interface Signals (Cont’d)
Name
Direction
Description
cfg_bus_number[7:0]
Output
Configuration Bus Number: Provides the assigned
bus number for the device. The user application
must use this information in the Bus Number field
of outgoing TLP requests. Default value after reset
is 00h. Refreshed whenever a Type 0 Configuration
Write packet is received.
cfg_device_number[4:0]
Output
Configuration Device Number: Provides the
assigned device number for the device. The user
application must use this information in the Device
Number field of outgoing TLP requests. Default
value after reset is 00000b. Refreshed whenever a
Type 0 Configuration Write packet is received.
cfg_function_number[2:0]
Output
Configuration Function Number: Provides the
function number for the device. The user
application must use this information in the
Function Number field of outgoing TLP request.
Function number is hardwired to 000b.
cfg_status[15:0]
Output
Configuration Status: Status register from the
Configuration Space Header.
cfg_command[15:0]
Output
Configuration Command: Command register from
the Configuration Space Header.
cfg_dstatus[15:0]
Output
Configuration Device Status: Device Status register
from the PCI Express Extended Capability
Structure.
cfg_dcommand[15:0]
Output
Configuration Device Command: Device Control
register from the PCI Express Extended Capability
Structure.
cfg_lstatus[15:0]
Output
Configuration Link Status: Link Status register
from the PCI Express Extended Capability
Structure.
cfg_lcommand[15:0]
Output
Configuration Link Command: Link Control
register from the PCI Express Extended Capability
Structure.
cfg_to_turnoff_n
Output
Configuration To Turnoff: Active Low. This output
notifies the user that a PME_TURN_Off message
has been received and the CMM will start polling
the cfg_turnoff_ok_n input coming in from the user.
After cfg_turnoff_ok_n is asserted, CMM sends a
PME_To_Ack message to the upstream device.
cfg_turnoff_ok_n
Input
Configuration Turnoff OK: Active Low. The user
application can assert this to notify the integrated
Endpoint block core that it is safe to turn the power
off.
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
27
Chapter 2: Core Overview
Table 2-8:
Configuration Interface Signals (Cont’d)
Name
cfg_pm_wake_n
Direction
Description
Input
Configuration Power Management Wake: A oneclock cycle active Low assertion signals the core to
generate and send a Power Management Wake
Event (PM_PME) Message TLP to the upstream link
partner.
Note: The user is required to assert this input only
under stable link conditions as reported on the
cfg_pcie_link_state[2:0] bus. Assertion of this signal
when the PCI Express Link is in transition results in
incorrect behavior on the PCI Express Link.
cfg_pcie_link_state_n[2:0]
Output
PCI Express Link State: This encoded bus reports
the PCIe Link State Information to the user.
•
•
•
•
cfg_trn_pending_n
Input
110b - PCI Express Link State is "L0"
101b - PCI Express Link State is "L0s"
011b - PCI Express Link State is "L1"
111b - PCI Express Link State is "in transition"
User Transaction Pending: If asserted, sets the
Transactions Pending bit in the Device Status
register.
Note: The user is required to assert this input if the
user application has not received a completion to an
upstream request. Active Low.
cfg_dsn[63:0]
cfg_ltssm_state[4:0]
Input
Configuration Device Serial Number: Serial
Number register fields of the Device Serial Number
extended capability. Not used if DSN capability is
disabled.
Output
LTSSM State: Indicates the current state of the Link
Training and Status State Machine. For state
encodings, see CFGLTSSMSTATE in Table G-10,
page 220.
Error Reporting Signals
Table 2-9 defines the user application error-reporting signals.
Table 2-9:
User Application Error-Reporting Signals
Port Name
28
Direction
Description
cfg_err_ecrc_n
Input
ECRC Error Report: Active Low. The user can
assert this signal to report an ECRC error (end-toend CRC).
cfg_err_ur_n
Input
Configuration Error Unsupported Request:
Active Low. The user can assert this signal to
report that an unsupported request was received.
This signal is ignored if cfg_err_cpl_rdy_n is
deasserted.
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Error Reporting Signals
Table 2-9:
User Application Error-Reporting Signals (Cont’d)
Port Name
cfg_err_cpl_timeout_n
Direction
Description
Input
Configuration Error Completion Timeout: Active
Low. The user can assert this signal to report a
completion timed out.
Note: The user should assert this signal only if the
device power state is D0. Asserting this signal in
non-D0 device power states might result in an
incorrect operation on the PCIe link. For
additional information, see the PCI Express Base
Specification, Rev.1.1, Section 5.3.1.2.
cfg_err_cpl_abort_n
Input
Configuration Error Completion Aborted: Active
Low. The user can assert this signal to report that
a completion was aborted. This signal is ignored
if cfg_err_cpl_rdy_n is deasserted.
cfg_err_posted_n
Input
Configuration Error Posted: Active Low. This
signal is used to further qualify any of the
cfg_err_* input signals. When this input is
asserted concurrently with one of the other
signals, it indicates that the transaction which
caused the error was a posted transaction.
cfg_err_cor_n
Input
Configuration Error Correctable Error: Active
Low. The user can assert this signal to report that
a correctable error was detected.
cfg_err_tlp_cpl_header[47:0]
Input
Configuration Error TLP Completion Header:
Accepts the header information from the user
when an error is signaled. This information is
required so that the core can issue a correct
completion, if required.
This information should be extracted from the
received error TLP and presented in the indicated
format:
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
[47:41]
Lower Address
[40:29]
Byte Count
[28:26]
TC
[25:24]
Attr
[23:8]
Requester ID
[7:0]
Tag
29
Chapter 2: Core Overview
Table 2-9:
User Application Error-Reporting Signals (Cont’d)
Port Name
Direction
Description
cfg_err_cpl_rdy_n
Output
Configuration Error Completion Ready: Active
Low. When asserted, this signal indicates that the
core can accept assertions on cfg_err_ur_n and
cfg_err_cpl_abort_n for Non-Posted transactions.
Assertions on cfg_err_ur_n and
cfg_err_cpl_abort_n are ignored when
cfg_err_cpl_rdy_n is deasserted.
cfg_err_locked_n
Input
Configuration Error Locked: Active Low. This
signal is used to further qualify any of the
cfg_err_* input signals. When this input is
asserted concurrently with one of the other
signals, it indicates that the transaction that
caused the error was a locked transaction.
This signal is intended to be used in Legacy
mode. If the user needs to signal an unsupported
request or an aborted completion for a locked
transaction, this signal can be used to return a
Completion Locked with UR or CA status.
Note: When not in Legacy mode, the core
automatically returns a Completion Locked, if
appropriate.
30
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Chapter 3
Licensing the Core
This version of the Spartan®-6 FPGA Integrated Endpoint Block for PCI Express® core
does not require a license key. Previous versions of the core released in ISE® v11.2 and
earlier required a license key. Refer to the corresponding version of this User Guide for
information on obtaining a license key. The Spartan-6 FPGA Integrated Endpoint Block for
PCI Express core is provided under the terms of the Xilinx End User Agreement.
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
31
Chapter 3: Licensing the Core
32
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Chapter 4
Getting Started Example Design
This chapter provides an overview of the Spartan®-6 FPGA Integrated Endpoint Block for
PCI Express® example design and instructions for generating the core. It also includes
information about simulating and implementing the example design using the provided
demonstration test bench.
Overview
The example simulation design consists of two discrete parts:
•
The Root Port Model, a test bench that generates, consumes, and checks PCI Express
bus traffic.
•
The Programmed Input/Output (PIO) example design, a completer application for
PCI Express. The PIO example design responds to Read and Write requests to its
memory space and can be synthesized for testing in hardware.
Simulation Design Overview
For the simulation design, transactions are sent from the Root Port Model to the integrated
Endpoint block core and processed by the PIO example design. Figure 4-1 illustrates the
simulation design provided with the integrated Endpoint block core. For more information
about the Root Port Model, see , Root Port Model Test Bench.
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
33
Chapter 4: Getting Started Example Design
X-Ref Target - Figure 4-1
Output
Logs
Root Port
Model TPI for
PCI Express
usrapp_com
Test
Program
usrapp_tx
usrapp_rx
dsport
PCI Express Fabric
Endpoint Core for
PCI Express
PIO
Design
Endpoint DUT for PCI Express
Figure 4-1: Simulation Example Design Block Diagram
34
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Overview
Implementation Design Overview
The implementation design consists of a simple PIO example that can accept read and
write transactions and respond to requests, as illustrated in Figure 4-2. Source code for the
example is provided with the core. For more information about the PIO example design,
see Appendix A, Programmed Input/Output Example Design.
X-Ref Target - Figure 4-2
Spartan-6 FPGA Integrated Endpoint Block for PCI Express
ep_mem0
PIO_TO_CTRL
ep_mem1
EP_TX
EP_RX
ep_mem2
ep_mem3
EP_MEM
PIO_EP
PIO
Figure 4-2:
Implementation Example Design Block Diagram
Example Design Elements
The PIO example design elements include:
•
Core wrapper
•
An example Verilog HDL or VHDL wrapper (instantiates the cores and example
design)
•
A customizable demonstration test bench to simulate the example design
The example design has been tested and verified with Xilinx ISE® v12.1 software and these
simulators:
•
Mentor Graphics ModelSim v6.5c
•
Cadence Incisive Enterprise Simulator (IES) 9.2
•
Synopsys VCS and VCS MX 2009.12
•
ISE Simulator (ISim)
Note: The VHDL example design supports only ModelSim.
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
35
Chapter 4: Getting Started Example Design
Generating the Core
To generate a core using the default values in the CORE Generator software GUI, follow
these steps:
1.
Start the CORE Generator tool.
For help starting and using the CORE Generator tool, see the Xilinx CORE Generator
Guide, available from the ISE Design Suite web page.
2.
Choose File > New Project.
X-Ref Target - Figure 4-3
Figure 4-3: New Project Dialog Box
3.
Enter a project name and location, then click OK. <project_dir> is used in this
example. The Project Options dialog box appears.
X-Ref Target - Figure 4-4
Figure 4-4: Project Options
36
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Generating the Core
4.
Set the project options:
From the Part tab, select these options:
•
Family: Spartan6
•
Device: xc6slx45t
•
Package: fgg484
•
Speed Grade: -2
Note: If an unsupported silicon device is selected, the core is dimmed (unavailable) in the list of
cores.
From the Generation tab, select these parameters, and then click OK.
5.
•
Design Entry. Select Verilog or VHDL.
•
Vendor. Select Synplicity or ISE (for XST).
Locate the core in the selection tree under Standard Bus Interfaces/PCI Express;
then double-click the core name to display the integrated Endpoint block main screen.
X-Ref Target - Figure 4-5
UG654_c4_05_041510
Figure 4-5:
Integrated Endpoint Block Main Screen
6.
In the Component Name field, enter a name for the core. <component_name> is used
in this example.
7.
Click Finish to generate the core using the default parameters. The core and its
supporting files, including the PIO example design and Root Port Model test bench,
are generated in the project directory.
For detailed information about the example design files and directories see Directory
Structure and File Contents, page 44. See the README file.
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
37
Chapter 4: Getting Started Example Design
Simulating the Example Design
The example design provides a quick way to simulate and observe the behavior of the core.
The simulation environment provided with the integrated Endpoint block core performs
simple memory access tests on the PIO example design. Transactions are generated by the
Root Port Model and responded to by the PIO example design.
•
PCI Express Transaction Layer Packets (TLPs) are generated by the test bench
transmit user application (pci_exp_usrapp_tx). As it transmits TLPs, it also
generates a log file, tx.dat.
•
PCI Express TLPs are received by the test bench receive user application
(pci_exp_usrapp_rx). As the user application receives the TLPs, it generates a log
file, rx.dat.
For more information about the test bench, see , Root Port Model Test Bench.
Setting up for Simulation
To run the functional simulation the Xilinx Simulation Libraries must be compiled for the
user system. See the Compiling Xilinx Simulation Libraries (COMPXLIB) in the Xilinx ISE
Synthesis and Verification Design Guide, and the Xilinx ISE Software Manuals and Help.
Documents can be downloaded from www.xilinx.com/support/software_manuals.htm.
Simulator Requirements
Spartan-6 device designs require a Verilog LRM-IEEE 1364-2005 encryption-compliant
simulator.
Note for Cadence IUS users: The work construct must be manually inserted into the
CDS.LIB file as shown below.
DEFINE WORK WORK
Running the Simulation
The simulation scripts provided with the example design support pre-implementation
(RTL) simulation. The existing test bench can be used to simulate with a
post-implementation version of the example design.
The pre-implementation simulation consists of these components:
•
Verilog or VHDL model of the test bench
•
Verilog or VHDL RTL example design
•
The Verilog or VHDL model of the Integrated Endpoint Block for PCI Express
1.
To run the simulation, go to this directory:
<project_dir>/<component_name>/simulation/functional
2.
38
Run the script that corresponds to the user simulation tool using one of these:
•
ModelSim: vsim -do simulate_mti.do
•
VCS: >./simulate_vcs.sh
•
IUS: > ./simulate_ncsim.sh
•
ISIM (UNIX): > ./simulate_isim.sh
•
ISIM (Windows): > simulate_isim.bat
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Implementing the Example Design
Implementing the Example Design
After generating the core, the netlists and the example design can be processed using the
Xilinx implementation tools. The generated output files include scripts to assist in running
the Xilinx software.
To implement the example design:
Open a command prompt or terminal window and type:
Windows
ms-dos> cd <project_dir>\<component_name>\implement
ms-dos> implement.bat
Linux
% cd <project_dir>/<component_name>/implement
% ./implement.sh
These commands execute a script that synthesizes, builds, maps, and place-and-routes the
example design, and then generates a post-par simulation model for use in timing
simulation. The resulting files are placed in the results directory and execute these
processes:
1.
Removes data files from the previous runs.
2.
Synthesizes the example design using XST or Synplify.
3.
ngdbuild. Builds a Xilinx design database for the example design.
-
Inputs:
Part-Package-Speed Grade selection:
For example, XC6SLX45T-FGG484-1
Example design UCF:
xilinx_pcie_1_lane_ep_<device>.ucf
4.
map: Maps design to the selected FPGA using the constraints provided.
5.
par: Places cells onto FPGA resources and routes connectivity.
6.
trce: Performs static timing analysis on design using constraints specified.
7.
netgen: Generates a logical Verilog HDL or VHDL representation of the design and
an SDF file for post-layout verification.
8.
bitgen: Generates a bitstream file for programming the FPGA.
These FPGA implementation related files are generated in the results directory:
•
routed.bit
FPGA configuration information.
•
routed.v[hd]
Verilog or VHDL functional Model.
•
routed.sdf
Timing model Standard Delay File.
•
mapped.mrp
Xilinx map report.
•
routed.par
Xilinx place and route report.
•
routed.twr
Xilinx timing analysis report.
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
39
Chapter 4: Getting Started Example Design
The script file starts from Verilog or VHDL source files and results in a bitstream file.
Users can also use the ISE Project Navigator GUI tool to implement designs. An example
ISE software project file is provided when the core is generated.
Using the ISE Project Navigator GUI Tool
To build a core and PIO example design with the ISE Project Navigator GUI tool:
1.
Start the ISE Project Navigator GUI tool.
For help starting and using the ISE Project Navigator tool, see the ISE Project Navigator
Guide, available from the ISE tool documentation web page
(http://www.xilinx.com/support/documentation/dt_ise.htm).
2.
Choose File → New Project.
3.
Enter a project name and location, then click Next > (see Figure 4-6).
X-Ref Target - Figure 4-6
UG654_c4_06_031110
Figure 4-6:
40
www.xilinx.com
Create New Project
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Implementing the Example Design
4.
Set the project options (see Figure 4-7):
Family: Spartan6
Device: Any LXT device
X-Ref Target - Figure 4-7
UG654_c4_07_031110
Figure 4-7:
Project Settings
5.
Click Next > and then Finish to create the project.
6.
Choose Project → New Source.
7.
Select IP (Core Generator & Architecture Wizard).
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
41
Chapter 4: Getting Started Example Design
8.
Enter a file name and ensure the “Add to project” checkbox is checked (see Figure 4-8).
X-Ref Target - Figure 4-8
UG654_c4_08_031110
Figure 4-8:
9.
Select Source Type
Select Spartan-6 Integrated Block for PCI Express. Click Next > and then Finish
(see Figure 4-9).
X-Ref Target - Figure 4-9
UG654_c4_09_031110
Figure 4-9:
42
www.xilinx.com
Select IP
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Implementing the Example Design
10. Configure the core as described in Chapter 5, Generating and Customizing the Core.
11. Choose File → Open Project.
12. Enter the ipcore_dir directory (see Figure 4-10).
X-Ref Target - Figure 4-10
UG654_c4_10_031110
Figure 4-10:
Directory ipcore_dir
13. Select the <component_name>_<lang>_example_project.xise file to load an
example project with the PIO example design along with the core built in step 1
through step 9 (see Figure 4-11).
X-Ref Target - Figure 4-11
UG654_c4_11_031110
Figure 4-11:
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
Load Example Project
43
Chapter 4: Getting Started Example Design
Directory Structure and File Contents
The integrated Endpoint block example design directories and their associated files are
defined in the sections that follow. Click a directory name to go to the desired directory and
its associated files.
Example Design
<project directory>
Top-level project directory; name is user-defined
topdirectory
<project directory>/<component name>
Core release notes readme file
<component name>/doc
Product documentation
<component name>/example_design
Verilog or VHDL design files
<component name>/implement
Implementation script files
implement/results
Results directory, created after implementation scripts are run, and
contains implement script results
<component name>/simulation
simulation/dsport
Root Port Bus Functional Model
simulation/functional
Functional simulation files
simulation/tests
Test command files
<component name>/source
Core source files
<project directory>
The project directory contains all the CORE Generator tool project files.
Table 4-1:
Project Directory
Name
Description
<project_dir>
<component_name>.xco
CORE Generator tool project-specific option file; can
be used as an input to the CORE Generator software.
<component_name>_flist.txt
List of files delivered with core.
<component_name>_<lang>_
example_project.xise
ISE software Project Navigator project file for the PIO
example design.
<component_name>.v[eo|ho]
Verilog or VHDL instantiation template.
Back to Top
44
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Directory Structure and File Contents
<project directory>/<component name>
The component name directory contains the release notes readme file provided with the
core, which can includes tool requirements, last-minute changes, updates, and issue
resolution.
Table 4-2:
Component Name Directory
Name
Description
<project_dir>/<component_name>
Readme file.
s6_pcie_readme.txt
Back to Top
<component name>/doc
The doc directory contains the PDF documentation provided with the core.
Table 4-3:
Doc Directory
Name
Description
<project_dir>/<component_name>/doc
s6_pcie_ug654.pdf
Spartan-6 FPGA Integrated Endpoint Block for PCI Express
User Guide.
s6_pcie_ds718.pdf
Spartan-6 FPGA Integrated Endpoint Block for PCI Express
Data Sheet.
Back to Top
<component name>/example_design
The example design directory contains the example design files provided with the core.
Table 4-4:
Example Design Directory
Name
Description
<project_dir>/<component_name>/example_design
xilinx_pcie_1_lane_ep_<device>.ucf
Example design UCF. Filename varies by
part, package, and speed grade.
xilinx_pcie_1_1_ep_s6.v[hd]
Top-level PIO example design files for
1-lane cores.
pcie_app_s6.v[hd]
PIO_EP_MEM.v[hd]
PIO.v[hd]
PIO_EP.v[hd]
PIO_EP_MEM_ACCESS.v[hd]
PIO example design files.
PIO_TO_CTRL.v[hd]
PIO_32.v[hd]
PIO_32_RX_ENGINE.v[hd]
PIO_32_TX_ENGINE.v[hd]
Back to Top
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
45
Chapter 4: Getting Started Example Design
<component name>/implement
The implement directory contains the core implementation script files.
Table 4-5:
Implement Directory
Name
Description
<project_dir>/<component_name>/implement
xst.scr
XST synthesis script.
implement.bat
implement.sh
DOS and Linux implementation scripts.
synplify.prj
Synplify synthesis script.
xst.prj
XST project file for the example design.
Back to Top
implement/results
The results directory is created by the implement script, after which the implement script
results are placed in the results directory.
Table 4-6:
Results Directory
Name
Description
<project_dir>/<component_name>/implement/results
Implement script result files.
Back to Top
46
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Directory Structure and File Contents
<component name>/simulation
simulation/dsport
The dsport directory contains the Root Port Bus Functional model files provided with the
core.
Table 4-7:
dsport Directory
Name
Description
<project_dir>/<component_name>/simulation/dsport
gtx_drp_chanalign_fix_3752_v6.v[hd]
gtx_rx_valid_filter_v6.v[hd]
gtx_tx_sync_rate_v6.v[hd]
gtx_wrapper_v6.v[hd]
pci_exp_usrapp_cfg.v[hd]
pci_exp_usrapp_com.v
pci_exp_usrapp_pl.v[hd]
pci_exp_usrapp_rx.v[hd]
pci_exp_usrapp_tx.v[hd]
pcie_2_0_rport_v6.v[hd]
pcie_2_0_v6_rp.v[hd]
pcie_bram_top_v6.v[hd]
pcie_bram_v6.v[hd]
pcie_brams_v6.v[hd]
pcie_clocking_v6.v[hd]
pcie_gtx_v6.v[hd]
pcie_pipe_lane_v6.v[hd]
pcie_pipe_misc_v6.v[hd]
pcie_pipe_v6.v[hd]
pcie_reset_delay_v6.v[hd]
pcie_upconfig_fix_3451_v6.v[hd]
test_interface.vhd
xilinx_pcie_2_0_rport_v6.v[hd]
Root port model files.
Back to Top
simulation/functional
The functional directory contains functional simulation scripts provided with the core.
Table 4-8:
Functional Directory
Name
Description
<project_dir>/<component_name>/simulation/functional
board_common.v
Contains test bench definitions.
board.f
List of files for RTL simulations.
board.v[hd]
Top-level simulation module.
isim_cmd.tcl
Simulation helper script for ISim.
simulate_isim.bat/simulate_isim.sh
Simulation scripts for ISIM DOS/UNIX.
simulate_mti.do
Simulation script for ModelSim.
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
47
Chapter 4: Getting Started Example Design
Table 4-8:
Functional Directory (Cont’d)
Name
Description
simulate_ncsim.sh
Simulation script for Cadence IUS.
simulate_vcs.sh
Simulation script for VCS.
sys_clk_gen_ds.v[hd]
System differential clock source.
sys_clk_gen.v[hd]
System clock source.
wave.{do, sv, tcl, wcfg}
Waveform setup scripts.
Back to Top
simulation/tests
The tests directory contains test definitions for the example test bench.
Table 4-9:
Tests Directory
Name
Description
<project_dir>/<component_name>/simulation/tests
Test definitions for example test bench.
tests.v[hd]
Back to Top
<component name>/source
This directory contains the source files for the core.
Table 4-10:
Source Directory
Name
Description
<project_dir>/<component_name>/source
<component name>.v[hd]
Verilog or VHDL top-level wrapper, which
instantiates the Endpoint block, block RAMs,
GTP transceiver, and clocking resources.
gtpa1_dual_wrapper_tile.v[hd]
gtpa1_dual_wrapper.v[hd]
Wrapper for the GTPA1, which configures the
transceiver and presents the interfaces required
for use with the integrated Endpoint block.
pcie_bram_top_s6.v[hd]
pcie_brams_s6.v[hd]
pcie_bram_s6.v[hd]
Configures and instantiates block RAMs for use
with the integrated Endpoint block.
Back to Top
48
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Chapter 5
Generating and Customizing the Core
The Spartan-6 FPGA Integrated Endpoint Block for PCI Express core is a fully configurable
and highly customizable solution. The integrated Endpoint block is customized using the
CORE Generator software GUI.
Note: The screen captures in this chapter are conceptual representatives of their subjects and
provide general information only. For the latest information, see the CORE Generator tool.
Customizing the Core through the CORE Generator Software
The CORE Generator software GUI for the Spartan-6 FPGA Integrated Endpoint Block for
PCI Express consists of nine screens:
•
Screen 1: Basic Parameter Settings
•
Screen 2: Base Address Registers
•
Screen 3: PCI Registers
•
Screens 4 and 5: Configuration Register Settings
•
Screen 6: Interrupt Capabilities
•
Screen 7: Power Management Registers
•
Screen 8: PCI Express Extended Capabilities
•
Screen 9: Advanced Settings
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
49
Chapter 5: Generating and Customizing the Core
Basic Parameter Settings
The initial customization screen shown in Figure 5-1 is used to define the basic parameters
for the core, including the component name, lane width and link speed.
X-Ref Target - Figure 5-1
UG654_c4_05_041510
Figure 5-1:
Screen 1: Integrated Endpoint Block for PCI Express Parameters
Component Name
Base name of the output files generated for the core. The name must begin with a letter and
can be composed of these characters: a to z, 0 to 9, and “_.”
PCIe Device / Port Type
•
50
Device Port Type: Indicates the PCI Express logical device type.
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Customizing the Core through the CORE Generator Software
Base Address Registers
The Base Address Register (BAR) screen shown in Figure 5-2 lets the user set the base
address register space. Each Bar (0 through 5) represents a 32-bit parameter.
X-Ref Target - Figure 5-2
Figure 5-2:
Screen 2: BAR Options
Base Address Register Overview
The Endpoint for PCIe supports up to six 32-bit Base Address Registers (BARs) or three
64-bit BARs, and the Expansion ROM BAR. BARs can be one of two sizes:
•
32-bit BARs: The address space can be as small as 128 bytes for Memory or 16 bytes
for I/O, or as large as 2 gigabytes. Used for Memory to I/O.
•
64-bit BARs: The address space can be as small as 128 bytes or as large as 8 exabytes.
Used for Memory only.
All BAR registers share these options:
•
Checkbox: Click the checkbox to enable the BAR; deselect the checkbox to disable the
BAR.
•
Type: BARs can either be I/O or Memory.
•
I/O: I/O BARs can only be 32-bit; the Prefetchable option does not apply to I/O
BARs.
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
51
Chapter 5: Generating and Customizing the Core
•
•
Memory: Memory BARs can be either 64-bit or 32-bit and can be prefetchable.
When a BAR is set as 64 bits, it uses the next BAR for the extended address space
and makes the next BAR inaccessible to the user
Size
•
Memory: When Memory and 64-bit are not selected, the size can range from
128 bytes to 2 gigabytes. When Memory and 64-bit are selected, the size can range
between 128 bytes and 8 exabytes.
•
I/O: When selected, the size can range from 16 bytes to 2 gigabytes.
•
Prefetchable: Identifies the ability of the memory space to be prefetched.
•
Value: The value assigned to the BAR based on the current selections.
For more information about managing the Base Address Register settings, see Managing
Base Address Register Settings.
Expansion ROM Base Address Register
If selected, the Expansion ROM is activated and can be a value from 2 KB to 4 GB.
Managing Base Address Register Settings
Memory, I/O, Type, and Prefetchable settings are handled by setting the appropriate GUI
settings for the desired base address register.
Memory or I/O settings indicate whether the address space is defined as memory or I/O.
The base address register only responds to commands that access the specified address
space. Generally, memory spaces less than 4Kbytes in size should be avoided. The
minimum I/O space allowed is 16 bytes; use of I/O space should be avoided in all new
designs.
Prefetchability is the ability of memory space to be prefetched. A memory space is
prefetchable if there are no side effects on reads (that is, data is not destroyed by reading, as
from a RAM). Byte write operations can be merged into a single doubleword write, when
applicable.
When configuring the core as an Endpoint for PCIe (non-Legacy), 64-bit addressing must
be supported for all BARs (except BAR5) that have the prefetchable bit set. 32-bit
addressing is permitted for all BARs that do not have the prefetchable bit set. The
prefetchable bit related requirement does not apply to a Legacy Endpoint. In either of the
above cases (Endpoint for PCI Express or Legacy Endpoint), the minimum memory
address range supported by a BAR is 128 bytes.
Disabling Unused Resources
For best results, disable unused base address registers to conserve system resources. A
base address register is disabled by deselecting unused BARs in the GUI.
52
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Customizing the Core through the CORE Generator Software
PCI Registers
The PCI Registers Screen shown in Figure 5-3 is used to customize the IP initial values,
class code and Cardbus CIS pointer information.
X-Ref Target - Figure 5-3
Figure 5-3:
PCI Registers: Screen 3
ID Initial Values
•
Vendor ID: Identifies the manufacturer of the device or application. Valid identifiers
are assigned by the PCI Special Interest Group to guarantee that each identifier is
unique. The default value, 10EEh, is the Vendor ID for Xilinx. Enter a vendor
identification number here. FFFFh is reserved.
•
Device ID: A unique identifier for the application; the default value is 0007h. This
field can be any value; change this value for the application.
•
Revision ID: Indicates the revision of the device or application; an extension of the
Device ID. The default value is 00h; enter values appropriate for the application.
•
Subsystem Vendor ID: Further qualifies the manufacturer of the device or
application. Enter a Subsystem Vendor ID here; the default value is 10EE. Typically,
this value is the same as Vendor ID. Setting the value to 0000h can cause compliance
testing issues.
•
Subsystem ID: Further qualifies the manufacturer of the device or application. This
value is typically the same as the Device ID; default value is 0007h. Setting the value
to 0000h can cause compliance testing issues.
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
53
Chapter 5: Generating and Customizing the Core
Class Code
The Class Code identifies the general function of a device, and is divided into three bytesize fields:
•
Base Class: Broadly identifies the type of function performed by the device.
•
Sub-Class: More specifically identifies the device function.
•
Interface: Defines a specific register-level programming interface, if any, allowing
device-independent software to interface with the device.
Class code encoding can be found at www.pcisig.com.
Cardbus CIS Pointer
Used in cardbus systems and points to the Card Information Structure for the cardbus
card. If this field is non-zero, an appropriate Card Information Structure must exist in the
correct location. The default value is 0000_0000h; value range is 0000_0000h through
FFFF_FFFFh.
Configuration Register Settings
The Configuration Registers screens shown in Figure 5-4 and Figure 5-5 show the options
for the Device Capabilities and Registers, the Block RAM Configuration Options, the Link
Capabilities Register, and the Link Status Register.
X-Ref Target - Figure 5-4
Figure 5-4:
54
Screen 4: Configuration Settings
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Customizing the Core through the CORE Generator Software
X-Ref Target - Figure 5-5
Figure 5-5:
Screen 5: Configuration Settings
Capabilities Register
•
Capability Version: Indicates PCI-SIG defined PCI Express capability structure
version number; this value cannot be changed.
•
Device Port Type: Indicates the PCI Express logical device type.
•
Capabilities Register: Displays the value of the Capabilities register presented by the
integrated Endpoint block, and is not editable.
Device Capabilities Register
•
Max Payload Size: Indicates the maximum payload size that the device/function can
support for TLPs.
•
Extended Tag Field: Indicates the maximum supported size of the Tag field as a
Requester. When selected, indicates 8-bit Tag field support. When deselected,
indicates 5-bit Tag field support.
•
Phantom Functions: Indicates the support for use of unclaimed function numbers to
extend the number of outstanding transactions allowed by logically combining
unclaimed function numbers (called Phantom Functions) with the Tag identifier. See
Section 2.2.6.2 of the PCI Express Base Specification version 1.1 for a description of
Tag Extensions. This field indicates the number of most significant bits of the function
number portion of Requester ID that are logically combined with the Tag identifier.
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
55
Chapter 5: Generating and Customizing the Core
•
Acceptable L0s Latency: Indicates the acceptable total latency that an Endpoint can
withstand due to the transition from L0s state to the L0 state.
•
Acceptable L1 Latency: Indicates the acceptable latency that an Endpoint can
withstand due to the transition from L1 state to the L0 state.
•
Device Capabilities Register: Displays the value of the Device Capabilities register
presented by the integrated Endpoint block and is not editable.
Block RAM Configuration Options
•
Performance Level: Selects the Performance Level settings, which determines the
Receiver and Transmitter Sizes. The table displayed specifies the Receiver and
Transmitter settings - number of TLPs buffered in the Transmitter, the Receiver Size,
the credits advertised by the core to the Link Partner and the number of block RAMs
required for the configuration, corresponding to the Max Payload Size selected, for
each of the Performance Level options.
•
Finite Completions: If selected, causes the device to advertise to the Link Partner the
actual amount of space available for completions in the receiver. For an Endpoint, this
is not compliant to the PCI Express Base Specification version 1.1, as endpoints are
required to advertise an infinite amount of completion space. Finite completions are
not supported in this release of the core.
Link Capabilities Register
This section is used to set the Link Capabilities register.
•
Maximum Link Speed: Indicates the maximum link speed of the given PCI Express
Link. This value is set to 2.5 Gb/s and is not editable.
•
Maximum Link Width: This value is set to 1 lane.
•
Enable ASPM L1 Support: Indicates the level of ASPM supported on the given PCI
Express Link. L0s is always supported by the integrated Endpoint block core; L1
support is optional and is enabled if this box is checked.
•
Link Capabilities Register: Displays the value of the Link Capabilities register
presented by the Endpoint and is not editable.
Link Status Register
•
56
Enable Slot Clock Configuration: Indicates that the Endpoint uses the platformprovided physical reference clock available on the connector. Must be cleared if the
Endpoint uses an independent reference clock.
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Customizing the Core through the CORE Generator Software
Interrupt Capabilities
The Interrupt Settings screen shown in Figure 5-6 sets the Legacy Interrupt Settings and
MSI Capabilities.
X-Ref Target - Figure 5-6
Figure 5-6:
Interrupt Capabilities: Screen 6
Legacy Interrupt Settings
•
Interrupt PIN: Indicates the mapping for Legacy Interrupt messages. A setting of
“None” indicates no Legacy Interrupts are used.
MSI Capabilities
•
Multiple Message Capable: Selects the number of MSI vectors to request from the
Root Complex.
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
57
Chapter 5: Generating and Customizing the Core
Power Management Registers
The Power Management Registers screen shown in Figure 5-7 includes settings for the
Power Management Registers, power consumption and power dissipation options.
X-Ref Target - Figure 5-7
Figure 5-7:
Power Management Registers: Screen 7
Power Management Registers
58
•
Device Specific Initialization: This bit indicates whether special initialization of this
function is required (beyond the standard PCI configuration header) before the
generic class device driver is able to use it. When selected, this option indicates that
the function requires a device specific initialization sequence following transition to
the D0 uninitialized state. See section 3.2.3 of the PCI Bus Power Management Interface
Specification Revision 1.2.
•
D1 Support: When selected, this option indicates that the function supports the D1
Power Management State. See section 3.2.3 of the PCI Bus Power Management Interface
Specification Revision 1.2.
•
D2 Support: When selected, this option indicates that the function supports the D2
Power Management State. See section 3.2.3 of the PCI Bus Power Management Interface
Specification Revision 1.2.
•
PME Support From: When this option is selected, indicates the power states in which
the function can assert PME#. See section 3.2.3 of the PCI Bus Power Management
Interface Specification Revision 1.2.
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Customizing the Core through the CORE Generator Software
Power Consumption
For information about power consumption, see section 3.2.6 of the PCI Bus Power
Management Interface Specification Revision 1.2
Power Dissipated
For information about power dissipation, see section 3.2.6 of the PCI Bus Power
Management Interface Specification Revision 1.2.
PCI Express Extended Capabilities
The PCIe Extended Capabilities screen shown in Figure 5-8 includes settings for Device
Serial Number Capability and optional user-defined Configuration capabilities.
X-Ref Target - Figure 5-8
Figure 5-8:
Screen 8: PCIe Extended Capabilities
Device Serial Number Capability
•
Device Serial Number Capability: An optional PCIe Extended Capability containing
a unique Device Serial Number. If enabled, the core presents the Device Serial
Number Capability using the value presented on the Device Serial Number input pin
of the port. If disabled, no Device Serial Number Extended Capability is presented.
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
59
Chapter 5: Generating and Customizing the Core
User Defined Configuration Capabilities
•
PCI Configuration Space Enable: Allows the user application to add/implement PCI
Legacy capability registers. This option should be selected if the user application
implements a legacy capability configuration space. This option enables the routing of
Configuration Requests to addresses outside the built-in PCI-Compatible
Configuration Space address range to the Transaction Interface.
•
PCI Express Extended Configuration Space Enable: Allows the user application to
add/implement PCI Express Extended capability registers. This option should be
selected if the user application implements such an extended capability configuration
space. This enables the routing of Configuration Requests to addresses outside the
built-in PCI Express Extended Configuration Space address range to the user
application.
Advanced Settings
The Advanced Settings screen shown in Figure 5-9 includes settings for Transaction Layer,
Physical Layer, Reference Clock Frequency and Xilinx Reference Boards options.
X-Ref Target - Figure 5-9
UG654_c5_09_030910
Figure 5-9:
60
Screen 9: Advanced Settings 1
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Customizing the Core through the CORE Generator Software
Transaction Layer Module
•
Trim TLP Digest ECRC: Causes the core to trim any TLP digest from an inbound
packet and clear the TLP Digest bit in the TLP header, before presenting it to the user.
•
Pipeline Registers for Transaction Block RAM Buffers: Selects the Pipeline registers
enabled for the Transaction Buffers. Pipeline registers can be enabled on either the
Write path or both the Read and Write paths of the Transaction Block RAM buffers.
Advanced Physical Layer
•
Force No Scrambling: Used for diagnostic purposes only and should never be
enabled in a working design. Setting this bit results in the data scramblers being
turned off so that the serial data stream can be analyzed.
Xilinx Reference Boards
Selecting this option enables the generation of Xilinx Reference Board specific design files.
Selecting the SP605 board configures the Reference Clock Frequency, Transceiver Location
and Transceiver Channel corresponding with the PCI Express edge connector on the
reference board. It also sets the corresponding pin locations in the UCF file. The user must
select the correct part/package combination when setting up the project to generate Xilinx
reference board specific design files.
Reference Clock Frequency
Selects the frequency of the reference clock provided on sys_clk. For important information
about clocking the Spartan-6 FPGA Integrated Endpoint Block for PCI Express, see
Clocking and Reset of the Integrated Endpoint Block Core, page 107.
Transceiver Selection
•
Transceiver Location: Selects the GTPA1_DUAL location for the PCI Express link.
•
Transceiver Channel: Selects the channel within the GTPA1_DUAL.
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
61
Chapter 5: Generating and Customizing the Core
62
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Chapter 6
Designing with the Core
This chapter provides design instructions for the Spartan®-6 FPGA Integrated Endpoint
Block for PCI Express® user interface and assumes knowledge of the PCI Express
Transaction Layer Packet (TLP) header fields. Header fields are defined in PCI Express Base
Specification v1.1, Chapter 2, Transaction Layer Specification.
This chapter includes the following design guidelines:
•
Transmitting Outbound Packets
•
Receiving Inbound Packets
•
Design with Configuration Space Registers and Configuration Interface
•
Additional Packet Handling Requirements
•
Power Management
•
Generating Interrupt Requests
•
Clocking and Reset of the Integrated Endpoint Block Core
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
63
Chapter 6: Designing with the Core
TLP Format on the Transaction Interface
Data is transmitted and received in Big-Endian order as required by the PCI Express Base
Specification. See Chapter 2 of the PCI Express Base Specification for detailed information
about TLP packet ordering. Figure 6-1 represents a typical 32-bit addressable Memory
Write Request TLP (as illustrated in Chapter 2 of the specification).
X-Ref Target - Figure 6-1
+0
+1
+2
+3
7 6 5 4 3 2 10 7 6 5 4 3 2 1 0 7 6 54 3 2 1 0 7 6 5 4 3 2 10
Byte 0 > R Fmt
x 0
Type
R TC
T E Attr R
Rsvd D
P
Requester ID
Byte 4 >
Tag
Address[31:2]
Byte 8 >
Byte 12 >
Data 0
Byte 16 >
Data 1
Byte 20 >
Data 2
Byte 24 >
TLP Digest
Figure 6-1:
Length
Last DW 1st DW
BE
BE
R
PCI Express Base Specification Byte Order
When using the 32-bit Transaction interface, packets are arranged on the 32-bit datapath in
the same order as shown in Figure 6-1. Byte 0 of the packet appears on trn_td[31:24]
(outbound) or trn_rd[31:24] (inbound) of the first DWORD, byte 1 on trn_td[23:16] or
trn_rd[23:16], and so forth. Byte 4 of the packet then appears on trn_td[31:24] or
trn_rd[31:24] of the second DWORD. The Header section of the packet consists of either
three or four DWORDs, determined by the TLP format and type as described in section 2.2
of the PCI Express Base Specification.
Packets sent to the core for transmission must follow the formatting rules for Transaction
Layer Packets (TLPs) as specified in Chapter 2 of the PCI Express Base Specification. The user
application is responsible for ensuring its packets’ validity, as the core does not check
packet validity or validate packets. The exact fields of a given TLP vary depending on the
type of packet being transmitted.
The core allows the user application to add an extra level of error checking by using the
optional TLP Digest field in the TLP header. The presence of a TLP Digest or ECRC is
indicated by the value of TD field in the TLP Header section. When TD=1, a correctly
computed CRC32 remainder DWORD is expected to be presented as the last DWORD of
the packet. The CRC32 remainder DWORD is not included in the length field of the TLP
header. The user application must calculate and present the TLP Digest as part of the
packet when transmitting packets. Upon receiving packets with a TLP Digest present, the
user application must check the validity of the CRC32 based on the contents of the packet.
The core does not check the TLP Digest for incoming packets.
The PCI Express Base Specification requires Advanced Error Reporting (AER) capability
when implementing ECRC. Although the integrated Endpoint block does not support
AER, users can still implement ECRC for custom solutions that do not require PCI Express
Base Specification Compliance.
64
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Transmitting Outbound Packets
Basic TLP Transmit Operation
The Endpoint for PCIe automatically transmits the following types of packets:
•
Completions to a remote device in response to Configuration Space requests.
•
Error-message responses to inbound requests malformed or unrecognized by the core.
Note: Certain unrecognized requests, for example, unexpected completions, can only be
detected by the user application, which is responsible for generating the appropriate response.
The user application is responsible for constructing these types of outbound packets:
•
Memory and I/O Requests to remote devices.
•
Completions in response to requests to the user application, for example, a Memory
Read Request.
•
Completions in response to user-implemented Configuration Space requests when
enabled. These requests include PCI Legacy capability registers beyond address BFh
and PCI Express extended capability registers beyond address 1FFh.
Note: For important information about accessing user-implemented Configuration Space while
in a low-power state, see Power Management, page 101.
The Spartan-6 FPGA Integrated Endpoint Block for PCI Express core notifies the user
application of pending internally generated TLPs that will arbitrate for the transmit
datapath by asserting trn_tcfg_req_n (0b). The user application can choose to give priority
to core-generated TLPs by driving trn_tcfg_gnt_n asserted (0b) permanently, without
regard to trn_tcfg_req_n. Doing so prevents user-application-generated TLPs from being
transmitted when a core-generated TLP is pending. Alternatively, the user application can
reserve priority for a user-application-generated TLP over core-generated TLPs, by
holding trn_tcfg_gnt_n deasserted (1b) until the user transaction is complete. Then, it is
asserted (0b) for one clock cycle. Users must not delay asserting trn_cfg_gnt_n indefinitely,
as this might cause a completion time-out in the Requester. See the PCI Express Base
Specification for more information on the Completion Timeout Mechanism.
Table 2-9, page 28 defines the transmit-direction Transaction interface signals. To transmit
a TLP, the user application must perform the following sequence of events on the transmit
Transaction interface:
1.
The user application logic asserts trn_tsrc_rdy_n, trn_tsof_n and presents the first TLP
DWORD on trn_td[31:0] when it is ready to transmit data.
2.
The user application asserts trn_tsrc_rdy_n and presents the remainder of the TLP
DWORDs on trn_td[31:0] for subsequent clock cycles (for which the core asserts
trn_tdst_rdy_n).
3.
The user application asserts trn_tsrc_rdy_n and trn_teof_n together with the last
DWORD of data.
4.
At the next clock cycle, the user application deasserts trn_tsrc_rdy_n to signal the end
of valid transfers on trn_td[31:0].
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
65
Chapter 6: Designing with the Core
Figure 6-2 illustrates a 3-DW TLP header without a data payload; an example is a 32-bit
addressable Memory Read request.
X-Ref Target - Figure 6-2
trn_clk
trn_td[31:00]
HDR 1
HDR 2
HDR 3
trn_tsof_n
trn_teof_n
trn_tsrc_rdy_n
trn_tdst_rdy_n
trn_tsrc_dsc_n
trn_terrfwd_n
trn_tbuf_av[5:0]
Figure 6-2:
66
TLP 3-DW Header without Payload
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Figure 6-3 illustrates a 4-DW TLP header without a data payload; an example is a 64-bit
addressable Memory Read request.
X-Ref Target - Figure 6-3
trn_clk
trn_td[31:00]
HDR 1
HDR 2
HDR 3
HDR 4
trn_tsof_n
trn_teof_n
trn_tsrc_rdy_n
trn_tdst_rdy_n
trn_tsrc_dsc_n
trn_terrfwd_n
trn_tbuf_av[5:0]
Figure 6-3:
TLP 4-DW Header without Payload
Figure 6-4 illustrates a 3-DW TLP header with a data payload; an example is a 32-bit
addressable Memory Write request.
X-Ref Target - Figure 6-4
trn_clk
trn_td[31:00]
HDR 1
HDR 2
HDR 3
DATA 1
DATA
DATA n
trn_tsof_n
trn_teof_n
trn_tsrc_rdy_n
trn_tdst_rdy_n
trn_tsrc_dsc_n
trn_terrfwd_n
trn_tbuf_av[5:0]
Figure 6-4:
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
TLP with 3-DW Header with Payload
www.xilinx.com
67
Chapter 6: Designing with the Core
Figure 6-5 illustrates a 4-DW TLP header with a data payload; an example is a 32-bit
addressable Memory Write request.
X-Ref Target - Figure 6-5
trn_clk
trn_td[31:00]
HDR 1
HDR 2
HDR 3
DATA
HDR 4
DATA n
trn_tsof_n
trn_teof_n
trn_tsrc_rdy_n
trn_tdst_rdy_n
trn_tsrc_dsc_n
trn_terrfwd_n
trn_tbuf_av[5:0]
Figure 6-5:
TLP with 4-DW Header with Payload
Presenting Back-to-Back Transactions on the Transmit Interface
The user application can present back-to-back TLPs on the transmit Transaction interface
to maximize bandwidth utilization. Figure 6-6 illustrates back-to-back TLPs presented on
the transmit interface. The user application asserts trn_tsof_n and presents a new TLP on
the next clock cycle after asserting trn_teof_n for the previous TLP.
X-Ref Target - Figure 6-6
trn_clk
trn_td[31:00]
HDR 1
HDR 2
HDR 3
DATA
HDR 1
HDR 2
HDR 3
trn_tsof_n
trn_teof_n
trn_trsc_rdy_n
trn_tdst_rdy_n
trn_tsrc_dsc_n
trn_terrfwd_n
trn_tbuf_av[5:0]
TLP1
Figure 6-6:
68
TLP2
Back-to-Back Transaction on Transmit Interface
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Source Throttling on the Transmit Datapath
The Transaction interface lets the user application throttle back if it has no data present on
trn_td[31:0]. When this condition occurs, the user application deasserts trn_tsrc_rdy_n,
which instructs the core Transaction interface to disregard data presented on trn_td[31:0].
Figure 6-7 illustrates the source throttling mechanism, where the user application does not
have data to present every clock cycle, and for this reason must deassert trn_tsrc_rdy_n
during these cycles. The user application should not deassert trn_tsrc_rdy_n during the
middle of a transfer if trn_tstr_n is asserted.
X-Ref Target - Figure 6-7
trn_clk
trn_td[31:0]
HDR 1
HDR 2
HDR 3
DATA 1
DATA 2
DATA 3
trn_tsof_n
trn_teof_n
trn_tsrc_rdy_n
trn_tstr_n
trn_tdst_rdy_n
trn_tsrc_dsc_n
trn_terrfwd_n
trn_tbuf_av[5:0]
Figure 6-7: Source Throttling on the Transmit Datapath
Destination Throttling of the Transmit Datapath
The core Transaction interface throttles the user application if there is no space left for a
new TLP in its transmit buffer pool. This can occur if the link partner is not processing
incoming packets at a rate equal to or greater than the rate at which the user application is
presenting TLPs. Figure 6-8 illustrates the deassertion of trn_tdst_rdy_n to throttle the user
application when the core’s internal transmit buffers are full.
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
69
Chapter 6: Designing with the Core
X-Ref Target - Figure 6-8
trn_clk
trn_td[31:00]
TLP1
dw1
TLP1
dw2
TLP1
dw3
TLP1
dw4
TLP1
dw25
TLP2
dw1
TLP2
dw2
TLP2
dw3
trn_tsof_n
trn_teof_n
trn_tsrc_rdy_n
trn_tdst_rdy_n
trn_tsrc_dsc_n
trn_terrfwd_n
trn_tbuf_av[5:0]
000000B
000001B
000000B
New Buffer Available
Figure 6-8:
000001B
000000B
New Buffer Available
Destination Throttling of the Endpoint Transmit Transaction Interface
If the core transmit Transaction interface accepts the start of a TLP by asserting
trn_tdst_rdy_n, it is guaranteed to accept the complete TLP with a size up to the value
contained in the Max_Payload_Size field of the PCI Express Device Capability Register
(offset 04H). To stay compliant to the PCI Express Base Specification, users should not violate
the Max_Payload_Size field of the PCI Express Device Control Register (offset 08H). The
core transmit Transaction interface deasserts trn_tdst_rdy_n only under these conditions:
•
After it has accepted the TLP completely and has no buffer space available for a new
TLP.
•
When the core is transmitting an internally generated TLP (configuration Completion
TLP, error Message TLP or error response as requested by the user application on the
cfg_err interface), after it has been granted use of the transmit datapath by the user
application, by assertion of trn_tcfg_gnt_n. The core subsequently asserts
trn_tdst_rdy_n after transmitting the internally generated TLP.
On deassertion of trn_tdst_rdy_n by the core, the user application needs to hold all control
and data signals until the core asserts trn_tdst_rdy_n. The core transmit Transaction
interface throttles the user application when the Power State field in Power Management
Control/Status Register (Offset 0x4) of the PCI Power Management Capability Structure
is changed to a non-D0 state. When this occurs, any ongoing TLP is accepted completely
and trn_tdst_rdy_n is subsequently deasserted, disallowing the user application from
initiating any new transactions—for the duration that the core is in the non-D0 power
state.
70
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Discontinuing Transmission of Transaction by Source
The core Transaction interface lets the user application terminate transmission of a TLP by
asserting trn_tsrc_dsc_n. Both trn_tsrc_rdy_n and trn_tdst_rdy_n must be asserted
together with trn_tsrc_dsc_n for the TLP to be discontinued. The signal trn_tsrc_dsc_n
must not be asserted together with trn_tsof_n. It can be asserted on any cycle after
trn_sof_n is deasserted up to and including the assertion of trn_teof_n. Asserting
trn_tsrc_dsc_n has no effect if no TLP transaction is in progress on the transmit interface.
Figure 6-9 illustrates the user application discontinuing a packet using trn_tsrc_dsc_n.
Asserting trn_teof_n together with trn_tsrc_dsc_n is optional.
X-Ref Target - Figure 6-9
trn_clk
trn_td[31:0]
TLP1
dw1
TLP1
dw2
TLP1
dw3
TLP1
dw4
TLP1
dw5
TLP
dw p-3
TLP
dw p-2
TLP
dw p-1
TLP
dw p
trn_tsof_n
trn_teof_n
trn_tsrc_rdy_n
trn_tdst_rdy_n
trn_tsrc_dsc_n
trn_terr_drop_n
Figure 6-9:
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Source Driven Transaction Discontinue on Transmit Interface
www.xilinx.com
71
Chapter 6: Designing with the Core
Discarding of Transaction by Destination
The core transmit Transaction interface discards a TLP for three reasons:
•
The PCI Express link goes down.
•
Presented TLP violates the Max_Payload_Size field of the Device Capability Register
(offset 04H) for PCI Express (it is left to the user to not violate the Max_Payload_Size
field of the Device Control Register (offset 08H)).
•
trn_tstr_n is asserted and data is not presented on consecutive clock cycles; that is,
trn_tsrc_rdy_n is deasserted in the middle of a TLP transfer.
When any of these occurs, the transmit Transaction interface continues to accept the
remainder of the presented TLP and asserts trn_terr_drop_n no later than the second clock
cycle following the EOF of the discarded TLP. Figure 6-10 illustrates the core signaling that
a packet was discarded using trn_terr_drop_n due to a length violation.
X-Ref Target - Figure 6-10
trn_clk
trn_td[31:00]
TLP
dw1
TLP
dw2
TLP
dw3
TLP
dw4
TLP
dw5
TLP
dw p-3
TLP
dw1
TLP
dw2
TLP
dw3
TLP
dw4
TLP
dw5
trn_tsof_n
trn_teof_n
trn_tsrc_rdy_n
trn_tdst_rdy_n
trn_tsrc_dsc_n
trn_terr_drop_n
trn_terrfwd_n
trn_tbuf_av[5:0]
Figure 6-10:
72
Destination Driven Transaction Discontinue on Transmit Interface
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Packet Data Poisoning on the Transmit Transaction Interface
The user application can use one of these mechanisms to mark the data payload of a
transmitted TLP as poisoned:
•
Set EP = 1 in the TLP header. This mechanism can be used if the payload is known to
be poisoned when the first DWORD of the header is presented to the core on the
Transaction interface.
•
Assert trn_terr_fwd_n for at least 1 valid data transfer cycle any time during the
packet transmission, as shown in Figure 6-11. This causes the core to set EP = 1 in the
TLP header when it transmits the packet onto the PCI Express fabric. This mechanism
can be used if the user application does not know whether a packet can be poisoned at
the start of packet transmission. Use of trn_terr_fwd_n is not supported for packets
when trn_tstr_n is asserted (streamed transmit packets).
X-Ref Target - Figure 6-11
trn_clk
trn_td[31:0]
TLP
dw1
TLP
dw2
TLP
dw3
TLP
dw4
TLP
dw5
TLP
dw p-3
TLP
dw p-2
TLP
dw p-1
TLP
dw p
trn_tsof_n
trn_teof_n
trn_tsrc_rdy_n
trn_tstr_n
trn_tdst_rdy_n
trn_tsrc_dsc_n
trn_terr_drop_n
trn_terrfwd_n
Figure 6-11:
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Packet Data Poisoning on the Transmit Transaction Interface
www.xilinx.com
73
Chapter 6: Designing with the Core
Streaming Mode for Transactions on the Transmit Interface
The Spartan-6 FPGA Integrated Endpoint Block for PCI Express core allows the user
application to enable Streaming (cut-through) mode for transmission of a TLP, when
possible, to reduce latency of operation. To enable this feature, the user application must
hold trn_tstr_n asserted for the entire duration of the transmitted TLP. In addition, the user
application must present valid frames on every clock cycle until the final cycle of the TLP.
In other words, the user application must not deassert trn_tsrc_rdy_n for the duration of
the presented TLP. Source throttling of the transaction while in streaming mode of
operation causes the transaction to be dropped (trn_terr_drop_n is asserted) and a
nullified TLP to be signaled on the PCI Express link. Figure 6-12 illustrates the streaming
mode of operation, where the first TLP is streamed and the second TLP is dropped due to
source throttling.
X-Ref Target - Figure 6-12
trn_clk
trn_td[31:0]
TLP1
dw1
TLP2
dw2
TLP1
dw25
TLP1
dw26
TLP2
dw1
TLP2
dw2
TLP
dw3
TLP2
dw4
TLP2
dw5
TLP2
dw6
trn_tsof_n
trn_teof_n
trn_tsrc_rdy_n
trn_tstr_n
trn_tdst_rdy_n
trn_terr_drop_n
Figure 6-12:
Streaming Mode on the Transmit Interface
Appending ECRC to Protect TLPs
If the user application needs to send a TLP Digest associated with a TLP, it must construct
the TLP header such that the TD bit is set and the user application must properly compute
and append the 1-DWORD TLP Digest after the last valid TLP payload section (if
applicable). TLPs originating within the core, for example Completions, Error Messages,
and Interrupts, do not have a TLP Digest appended.
Maximum Payload Size
TLP size is restricted by the capabilities of both link partners. After the link is trained, the
root complex sets the MAX_PAYLOAD_SIZE value in the Device Control register. This
value is equal to or less than the value advertised by the core's Device Capability register.
The advertised value in the Device Capability register of the integrated Endpoint block
core is either 128, 256, or 512 bytes, depending on the setting in the CORE Generator
software GUI. For more information about these registers, see section 7.8 of the PCI Express
Base Specification. The value of the core's Device Control register is provided to the user
application on the cfg_dcommand[15:0] output. See Design with Configuration Space
Registers and Configuration Interface, page 87 for information about this output.
74
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Transmit Buffers
The Endpoint for PCIe transmit Transaction interface provides trn_tbuf_av, an
instantaneous indication of the number of Max_Payload_Size buffers available for use in
the transmit buffer pool. Table 6-1 defines the number of transmit buffers available and
maximum supported payload size for a specific core.
Table 6-1:
Transmit Buffers Available
Capability Max
Performance Level(1)
Payload Size
Good (Minimize Block RAM Usage)
High (Maximize Performance)
(Bytes)
128
13
27
256
14
29
512
15
30
Notes:
1. Performance level is set through a CORE Generator software GUI selection.
Each buffer can hold one maximum sized TLP. A maximum sized TLP is a TLP with a
4-DWORD header plus a data payload equal to the MAX_PAYLOAD_SIZE of the core (as
defined in the Device Capability register) plus a TLP Digest. After the link is trained, the
root complex sets the MAX_PAYLOAD_SIZE value in the Device Control register. This
value is equal to or less than the value advertised by the core's Device Capability register.
For more information about these registers, see section 7.8 of the PCI Express Base
Specification. A TLP is held in the core's transmit buffer until the link partner
acknowledges receipt of the packet, at which time the buffer is released and a new TLP can
be loaded into it by the user application.
For example, if the Capability Max Payload Size selected for the Endpoint core is 256 bytes,
and the performance level selected is high, there are 29 total transmit buffers. Each of these
buffers can hold at a maximum one 64-bit Memory Write Request (4 DWORD header) plus
256 bytes of data (64 DWORDs) plus TLP Digest (1 DWORD) for a total of 69 DWORDs.
This example assumes the root complex set the MAX_PAYLOAD_SIZE register of the
Device Control register to 256 bytes, which is the maximum capability advertised by this
core. For this reason, at any given time, this core could have 29 of these 69 DWORD TLPs
awaiting transmittal. There is no sharing of buffers among multiple TLPs, so even if user is
sending smaller TLPs such as 32-bit Memory Read request with no TLP Digest totaling
3 DWORDs only per TLP, each transmit buffer still holds only one TLP at any time.
The internal transmit buffers are shared between the user application and the core’s
configuration management module (CMM). Due to this, the trn_tbuf_av bus can fluctuate
even if the user application is not transmitting packets. The CMM generates completion
TLPs in response to configuration reads or writes, interrupt TLPs at the request of the user
application, and message TLPs when needed.
The Transmit Buffers Available indication enables the user application to completely
utilize the PCI transaction ordering feature of the core transmitter. The transaction
ordering rules allow for Posted and Completion TLPs to bypass Non-Posted TLPs. See
section 2.4 of the PCI Express Base Specification for more information about ordering rules.
The core supports the transaction ordering rules and promotes Posted and Completion
packets ahead of blocked Non-Posted TLPs. Non-Posted TLPs can become blocked if the
link partner is in a state where it momentarily has no Non-Posted receive buffers available,
which it advertises through Flow Control updates. In this case, the core promotes
Completion and Posted TLPs ahead of these blocked Non-Posted TLPs. However, this can
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
75
Chapter 6: Designing with the Core
only occur if the Completion or Posted TLP has been loaded into the core by the user
application. By monitoring the trn_tbuf_av bus, the user application can ensure there is at
least one free buffer available for any Completion or Posted TLP. Promotion of Completion
and Posted TLPs only occurs when Non-Posted TLPs are blocked; otherwise packets are
sent on the link in the order they are received from the user application.
Receiving Inbound Packets
Basic TLP Receive Operation
Table 2-7, page 23 defines the receive Transaction interface signals. This sequence of events
must occur on the receive Transaction interface for the core to present a TLP to the user
application logic:
1.
When the user application is ready to receive data, it asserts trn_rdst_rdy_n.
2.
When the core is ready to transfer data, the core asserts trn_rsrc_rdy_n with trn_rsof_n
and presents the first complete TLP DWORD on trn_rd[31:0].
3.
The core then deasserts trn_rsof_n, asserts trn_rsrc_rdy_n, and presents TLP DWORDs
on trn_rd[31:0] for subsequent clock cycles, for which the user application logic asserts
trn_rdst_rdy_n.
4.
The core then asserts trn_reof_n and presents either the last DWORD on trn_td[31:0].
5.
If no further TLPs are available, at the next clock cycle, the core deasserts
trn_rsrc_rdy_n to signal the end of valid transfers on trn_rd[31:0].
Figure 6-13 illustrates a 3-DW TLP header without a data payload; an example is a 32-bit
addressable Memory Read request.
X-Ref Target - Figure 6-13
trn_clk
trn_rd[31:00]
HDR 1
HDR 2
HDR 3
trn_rsof_n
trn_reof_n
trn_rsrc_rdy_n
trn_rdst_rdy_n
trn_rsrc_dsc_n
trn_rerrfwd_n
trn_rnp_ok_n
trn_rbar_hit_n[6:0]
Figure 6-13:
76
TLP 3-DW Header without Payload
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Figure 6-14 illustrates a 4-DW TLP header without a data payload; an example is a 64-bit
addressable Memory Read request.
X-Ref Target - Figure 6-14
trn_clk
trn_rd[31:00]
HDR 1
HDR 2
HDR 3
HDR 4
trn_rsof_n
trn_reof_n
trn_rsrc_rdy_n
trn_rdst_rdy_n
trn_rsrc_dsc_n
trn_rerrfwd_n
trn_rnp_ok_n
trn_rbar_hit_n[6:0]
Figure 6-14:
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
TLP 4-DW Header without Payload
www.xilinx.com
77
Chapter 6: Designing with the Core
Figure 6-15 illustrates a 3-DW TLP header with a data payload; an example is a 32-bit
addressable Memory Write request.
X-Ref Target - Figure 6-15
trn_clk
trn_rd[31:00]
HDR 1
HDR 2
HDR 3
DATA
trn_rsof_n
trn_reof_n
trn_rsrc_rdy_n
trn_rdst_rdy_n
trn_rsrc_dsc_n
trn_rerrfwd_n
trn_rnp_ok_n
trn_rbar_hit_n[6:0]
Figure 6-15:
78
TLP 3-DW Header with Payload
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Figure 6-16 illustrates a 4-DW TLP header with a data payload; an example is a 64-bit
addressable Memory Write request.
X-Ref Target - Figure 6-16
trn_clk
trn_rd[31:00]
HDR 1
HDR 2
HDR 3
HDR 4
DATA
trn_rsof_n
trn_reof_n
trn_rsrc_rdy_n
trn_rdst_rdy_n
trn_rsrc_dsc_n
trn_rerrfwd_n
trn_rnp_ok_n
trn_rbar_hit_n[6:0]
Figure 6-16:
TLP 4-DW Header with Payload
Throttling the Datapath on the Receive Transaction Interface
The user application can stall the transfer of data from the core at any time by deasserting
trn_rdst_rdy_n. If the user deasserts trn_rdst_rdy_n while no transfer is in progress and if
a TLP becomes available, the core asserts trn_rsrc_rdy_n and trn_rsof_n and presents the
first TLP DWORD on trn_rd[31:0]. The core remains in this state until the user asserts
trn_rdst_rdy_n to signal the acceptance of the data presented on trn_rd[31:0]. At that point,
the core presents subsequent TLP DWORDs as long as trn_rdst_rdy_n remains asserted. If
the user deasserts trn_rdst_rdy_n during the middle of a transfer, the core stalls the
transfer of data until the user asserts trn_rdst_rdy_n again. There is no limit to the number
of cycles the user can keep trn_rdst_rdy_n deasserted. The core pauses until the user is
again ready to receive TLPs.
Figure 6-17 illustrates the core asserting trn_rsrc_rdy_n and trn_rsof_n along with
presenting data on trn_rd[31:0]. The user application logic inserts wait states by
deasserting trn_rdst_rdy_n. The core does not present the next TLP DWORD until it
detects trn_rdst_rdy_n assertion. The user application logic can assert or deassert
trn_rdst_rdy_n as required to balance receipt of new TLP transfers with the rate of TLP
data processing inside the application logic.
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
79
Chapter 6: Designing with the Core
X-Ref Target - Figure 6-17
trn_clk
trn_rd[31:00]
TLP dw1
TLP dw2 TLP dw3
TLP dw4
TLP dw5 TLP dw6 TLP dw7
trn_rsof_n
trn_reof_n
trn_rsrc_rdy_n
trn_rdst_rdy_n
trn_rsrc_dsc_n
trn_rerrfwd_n
trn_rnp_ok_n
trn_rbar_hit_n[6:0]
Figure 6-17:
User Application Throttling Receive TLP
Receiving Back-to-Back Transactions on the Receive Transaction Interface
The user application logic must be designed to handle presentation of back-to-back TLPs
on the receive interface Transaction interface by the core. The core can assert trn_rsof_n for
a new TLP at the clock cycle after trn_reof_n assertion for the previous TLP. Figure 6-18
illustrates back-to-back TLPs presented on the receive interface.
X-Ref Target - Figure 6-18
trn_clk
trn_rd[31:00]
TLP1
dw1
TLP1
dw2
TLP1
dw3
TLP1
dw4
TLP1
dw5
TLP1
dw6
TLP2
dw1
TLP2
dw2
TLP2
dw3
TLP2
dw4
TLP2
dw5
TLP2
dw6
trn_rsof_n
trn_reof_n
trn_rsrc_rdy_n
trn_rdst_rdy_n
trn_rsrc_dsc_n
trn_rerrfwd_n
trn_rnp_ok_n
trn_rbar_hit_n[6:0]
Figure 6-18:
80
Receive Back-to-Back Transactions
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
If the user application cannot accept back-to-back packets, it can stall the transfer of the
TLP by deasserting trn_rdst_rdy_n as discussed in the previous section. Figure 6-19 shows
an example of using trn_rdst_rdy_n to pause the acceptance of the second TLP.
X-Ref Target - Figure 6-19
trn_clk
trn_rd[31:00]
TLP1
dw1
TLP1
dw2
TLP1
dw3
TLP1
dw4
TLP1
dw5
TLP1
dw6
TLP2
dw2
TLP2
dw1
TLP2
dw3
TLP2
dw4
trn_rsof_n
trn_reof_n
trn_rsrc_rdy_n
trn_rdst_rdy_n
trn_rsrc_dsc_n
Figure 6-19:
User Application Throttling of Back-to-Back TLPs
Packet Re-ordering on Receive Transaction Interface
Transaction processing in the core receiver is fully compliant with the PCI transaction
ordering rules. The transaction ordering rules allow for Posted and Completion TLPs to
bypass blocked Non-Posted TLPs.
The user application can deassert trn_rnp_ok_n if it is not ready to accept Non-Posted
Transactions from the core, (as shown in Figure 6-20) but can receive Posted and
Completion Transactions. The user application must deassert trn_rnp_ok_n at least one
clock cycle before trn_eof_n of the next-to-last Non-Posted packet the user can accept.
While trn_rnp_ok_n is deasserted, received Posted and Completion Transactions pass
Non-Posted Transactions. After the user application is ready to accept Non-Posted
Transactions, it must reassert trn_rnp_ok_n. Previously bypassed Non-Posted
Transactions are presented to the user application before other received TLPs.
X-Ref Target - Figure 6-20
trn_clk
trn_rd[31:0]
HDR 1
HDR 2
HDR 3
HDR 1
HDR 2
HDR 3
HDR 1
HDR 2
trn_rsof_n
trn_reof_n
trn_rsrc_rdy_n
trn_rdst_rdy_n
trn_rsrc_dsc_n
trn_rerrfwd_n
trn_rnp_ok_n
trn_rbar_hit_n[6:0]
Non-Posted TLP1
Figure 6-20:
Non-Posted TLP2
Posted/Cmpl TLP3
Packet Re-ordering on Receive Transaction Interface
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
81
Chapter 6: Designing with the Core
Packet re-ordering allows the user application to optimize the rate at which Non-Posted
TLPs are processed, while continuing to receive and process Posted and Completion TLPs
in a non-blocking fashion. The trn_rnp_ok_n signaling restrictions require that the user
application be able to receive and buffer at least three Non-Posted TLPs. The following
algorithm describes the process of managing the Non-Posted TLP buffers.
Consider that Non-Posted_Buffers_Available denotes the size of Non-Posted buffer space
available to user application. The size of the Non-Posted buffer space is three Non-Posted
TLPs. Non-Posted_Buffers_Available is decremented when a Non-Posted TLP is accepted
for processing from the core, and is incremented when Non-Posted TLP is drained for
processing by the user application.
For every clock cycle, do
{
if (Valid transaction Start-Of-Frame accepted by user application)
Extract TLP Format and Type from the 1st TLP DW
if (TLP type == Non Posted) {
if (Non-Posted_Buffers_Available <= 2)
// Accounts for the
current and possibly the next NP TLP
Deassert trn_rnp_ok_n on the following clock cycle.
else if (Other optional user policies to stall Non-Posted
transactions)
Deassert trn_rnp_ok_n on the following clock cycle.
else // (Non-Posted_Buffers_Available > 2)
Assert trn_rnp_ok_n on the following clock cycle.
Decrement Non-Posted_Buffers_Available in User Application
} else { // Posted and Completion TLPs
Process the received TLPs
}
}
}
82
www.xilinx.com
{
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Packet Data Poisoning and TLP Digest on Receive Transaction Interface
To simplify logic within the user application, the core performs automatic pre-processing
based on values of TLP Digest (TD) and Data Poisoning (EP) header bit fields on the
received TLP.
All received TLPs with the Data Poisoning bit in the header set (EP=1) are presented to the
user. The core asserts the trn_rerrfwd_n signal for the duration of each poisoned TLP, as
illustrated in Figure 6-21.
If the TLP Digest bit field in the TLP header is set (TD = 1), the TLP contains an End-to-End
CRC (ECRC). The core performs the following operations based on how the user
configured the core during core generation:
•
If the Trim TLP Digest option is on, the core removes and discards the ECRC field
from the received TLP and clears the TLP Digest bit in the TLP header.
•
If the Trim TLP Digest option is off, the core does not remove the ECRC field from the
received TLP and presents the entire TLP including TLP Digest to the user application
receiver interface.
See Chapter 5, Generating and Customizing the Core, for more information about how to
enable the Trim TLP Digest option during core generation.
X-Ref Target - Figure 6-21
trn_clk
trn_rd[31:00]
TLP1
dw1
TLP1
dw2
TLP1
dw3
TLP1
dw4
TLP1
dw18
TLP1
dw19
TLP1
dw20
TLP1
dw21
TLP1
dw22
trn_rsof_n
trn_reof_n
trn_rsrc_rdy_n
trn_rdst_rdy_n
trn_rsrc_dsc_n
trn_rerrfwd_n
trn_rnp_ok_n
trn_rbar_hit_n[6:0]
Figure 6-21:
Receive Transaction Data Poisoning
Packet Base Address Register Hit on Receive Transaction Interface
The core decodes incoming Memory and I/O TLP request addresses to determine which
Base Address Register (BAR) in the core's Type0 configuration space is being targeted, and
indicates the decoded base address on trn_rbar_hit_n[6:0]. For each received Memory or
I/O TLP, a minimum of one and a maximum of two (adjacent) bit(s) are set to 0. If the
received TLP targets a 32-bit Memory or I/O BAR, only one bit is asserted. If the received
TLP targets a 64-bit Memory BAR, two adjacent bits are asserted. If the core receives a TLP
that is not decoded by one of the BARs (that is, a misdirected TLP), then the core drops it
without presenting it to the user and an Unsupported Request message is automatically
generated. Even if the core is configured for a 64-bit BAR, the system might not always
allocate a 64-bit address, in which case only one trn_rbar_hit_n[6:0] signal is asserted.
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
83
Chapter 6: Designing with the Core
Table 6-2 illustrates mapping between trn_rbar_hit_n[6:0] and the BARs, and the
corresponding byte offsets in the core Type0 configuration header.
Table 6-2:
trn_rbar_hit_n to Base Address Register Mapping
trn_rbar_hit_n[x]
BAR
Byte Offset
0
0
10h
1
1
14h
2
2
18h
3
3
1Ch
4
4
20h
5
5
24h
6
Expansion ROM BAR
30h
For a Memory or I/O TLP Transaction on the receive interface, trn_rbar_hit_n[6:0] is valid
for the entire TLP, starting with the assertion of trn_rsof_n, as shown in Figure 6-22. When
receiving non-Memory and non-I/O transactions, trn_rbar_hit_n[6:0] is undefined.
X-Ref Target - Figure 6-22
trn_clk
trn_rd[31:00]
TLP1
dw1
TLP1
dw2
TLP1
dw3
TLP1
dw4
TLP1
dw5
TLP1
dw6
TLP2
dw1
TLP2
dw2
TLP2
dw3
TLP2
dw4
trn_rsof_n
trn_reof_n
trn_rsrc_rdy_n
trn_rdst_rdy_n
trn_rbar_hit_n[6:0]
1111101B
Figure 6-22:
1110011B
BAR Target Determination using trn_rbar_hit
The signal trn_rbar_hit_n[6:0] enables received Memory and I/O transactions to be
directed to the appropriate destination Memory and I/O apertures in the user application.
By utilizing trn_rbar_hit_n[6:0], application logic can inspect only the lower order Memory
and I/O address bits within the address aperture to simplify decoding logic.
84
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Packet Transfer Discontinue on Receive Transaction Interface
The Endpoint for PCIe asserts trn_rsrc_dsc_n if communication with the link partner is
lost, which results in the termination of an in-progress TLP. The loss of communication with
the link partner is signaled by deassertion of trn_lnk_up_n. When trn_lnk_up_n is
deasserted, it effectively acts as a Hot Reset to the entire core. For this reason, all TLPs
stored inside the core or being presented to the receive interface are irrecoverably lost.
Figure 6-23 illustrates packet transfer discontinue scenario.
X-Ref Target - Figure 6-23
trn_clk
trn_reset_n
trn_lnk_up_n
trn_rd[31:00]
TLP
dw1
TLP
dw2
TLP
dw3
TLP
dw1
TLP
dw p-1
TLP
dw p
trn_rsof_n
trn_reof_n
trn_rsrc_rdy_n
trn_rdst_rdy_n
trn_rsrc_dsc_n
trn_rerrfwd_n
trn_rnp_ok_n
trn_rbar_hit_n[6:0]
Figure 6-23:
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Receive Transaction Discontinue
www.xilinx.com
85
Chapter 6: Designing with the Core
Receiver Flow Control Credits Available
The Spartan-6 FPGA Integrated Endpoint Block for PCI Express provides the user
application information about the state of the receiver buffer pool queues. This
information represents the current space available for the Posted, Non-Posted, and
Completion queues.
One Header Credit is equal to either a 3 or 4 DWORD TLP Header and one Data Credit is
equal to 16 bytes of payload data. Table 6-3 provides values on credits available
immediately after trn_lnk_up_n assertion but before the reception of any TLP. If space
available for any of the above categories is exhausted, the corresponding credit available
signals indicate a value of zero. Credits available return to initial values after the receiver
has drained all TLPs.
Table 6-3:
Transaction Receiver Credits Available Initial Values
Credit Category
Non-Posted Header
Performance
128 byte
Level
Capability MPS
Good
Posted Data
Completion Header
Completion Data
512 byte
Capability MPS
8
High
Posted Header
256 byte
Capability MPS
Good
16
24
32
High
30
32
32
Good
41
96
211
High
89
211
467
Good
16
24
40
High
30
40
40
Good
41
96
211
High
89
211
467
The user application can use the trn_fc_ph[7:0], trn_fc_pd[11:0], trn_fc_nph[7:0],
trn_fc_npd[11:0], trn_fc_cplh[7:0], trn_fc_cpld[11:0], and trn_fc_sel[2:0] signals to
efficiently utilize and manage receiver buffer space available in the core and the core
application. For additional information, see Flow Control Credit Information, page 98.
Endpoint cores for PCI Express have a unique requirement where the user application
must use advanced methods to prevent buffer overflows while requesting Non-Posted
Read Requests from an upstream component. According to the specification, a PCI Express
Endpoint is required to advertise infinite storage credits for Completion Transactions in its
receivers. This means that endpoints must internally manage Memory Read Requests
transmitted upstream and not overflow the receiver when the corresponding Completions
are received. The user application transmit logic must use Completion credit information
presented to modulate the rate and size of Memory Read requests, to stay within the
instantaneous Completion space available in the core receiver. For additional information,
see Appendix E, Managing Receive-Buffer Space for Inbound Completions.
86
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Design with Configuration Space Registers and Configuration Interface
Design with Configuration Space Registers and Configuration
Interface
This section describes the use of the Configuration Interface for accessing the PCI Express
Configuration Space Type 0 registers that are part of the integrated Endpoint block core.
The Configuration Interface includes a read Configuration Port for accessing the registers.
In addition, some commonly used registers are mapped directly on the Configuration
Interface for convenience.
Registers Mapped Directly onto the Configuration Interface
The integrated Endpoint block core provides direct access to select command and status
registers in its Configuration Space. Values in these registers are modified by
Configuration Writes received from the Root Complex and cannot be modified by the user
application. Table 6-4 defines the command and status registers mapped to the
configuration port.
Table 6-4:
Command and Status Registers Mapped to the Configuration Port
Port Name
Direction
Description
cfg_bus_number[7:0]
Output
Bus Number: Default value after reset is
00h. Refreshed whenever a Type 0
Configuration Write packet is received.
cfg_device_number[4:0]
Output
Device Number: Default value after reset is
00000b. Refreshed whenever a Type 0
Configuration Write packet is received.
cfg_function_number[2:0]
Output
Function Number: Function number of the
core, hard wired to 000b.
cfg_status[15:0]
Output
Status Register: Status register from the
Configuration Space Header.
cfg_command[15:0]
Output
Command Register: Command register
from the Configuration Space Header.
cfg_dstatus[15:0]
Output
Device Status Register: Device status
register from the PCI Express Extended
Capability Structure.
cfg_dcommand[15:0]
Output
Device Command Register: Device control
register from the PCI Express Extended
Capability Structure.
cfg_lstatus[15:0]
Output
Link Status Register: Link status register
from the PCI Express Extended Capability
Structure.
cfg_lcommand[15:0]
Output
Link Command Register: Link control
register from the PCI Express Extended
Capability Structure.
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
87
Chapter 6: Designing with the Core
Device Control and Status Register Definitions
cfg_bus_number[7:0], cfg_device_number[4:0], cfg_function_number[2:0]
Together, these three values comprise the core ID, which the core captures from the
corresponding fields of inbound Type 0 Configuration Write accesses. The user application
is responsible for using this core ID as the Requestor ID on any requests it originates, and
using it as the Completer ID on any Completion response it sends. This core supports only
one function; for this reason, the function number is hardwired to 000b.
cfg_status[15:0]
This bus allows the user application to read the Status register in the PCI Configuration
Space Header. Table 6-5 defines these bits. See the PCI Express Base Specification for detailed
information.
Table 6-5:
Bit Mapping on Header Status Register
Bit
88
Name
cfg_status[15]
Detected Parity Error
cfg_status[14]
Signaled System Error
cfg_status[13]
Received Master Abort
cfg_status[12]
Received Target Abort
cfg_status[11]
Signaled Target Abort
cfg_status[10:9]
DEVSEL Timing (hardwired to 00b)
cfg_status[8]
Master Data Parity Error
cfg_status[7]
Fast Back-to-Back Transactions Capable (hardwired to 0)
cfg_status[6]
Reserved
cfg_status[5]
66 MHz Capable (hardwired to 0)
cfg_status[4]
Capabilities List Present (hardwired to 1)
cfg_status[3]
Interrupt Status
cfg_status[2:0]
Reserved
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Design with Configuration Space Registers and Configuration Interface
cfg_command[15:0]
This bus reflects the value stored in the Command register in the PCI Configuration Space
Header. Table 6-6 provides the definitions for each bit in this bus. See the PCI Express Base
Specification for detailed information.
Table 6-6:
Bit Mapping on Header Command Register
Bit
Name
cfg_command[15:11]
Reserved
cfg_command[10]
Interrupt Disable
cfg_command[9]
Fast Back-to-Back Transactions Enable (hardwired to 0)
cfg_command[8]
SERR Enable
cfg_command[7]
IDSEL Stepping/Wait Cycle Control (hardwired to 0)
cfg_command[6]
Parity Error Enable
cfg_command[5]
VGA Palette Snoop (hardwired to 0)
cfg_command[4]
Memory Write and Invalidate (hardwired to 0)
cfg_command[3]
Special Cycle Enable (hardwired to 0)
cfg_command[2]
Bus Master Enable
cfg_command[1]
Memory Address Space Decoder Enable
cfg_command[0]
I/O Address Space Decoder Enable
The user application must monitor the Bus Master Enable bit (cfg_command[2]) and
refrain from transmitting requests while this bit is not set. This requirement applies only to
requests; completions can be transmitted regardless of this bit.
cfg_dstatus[15:0]
This bus reflects the value stored in the Device Status register of the PCI Express Extended
Capabilities Structure. Table 6-7 defines each bit in the cfg_dstatus bus. See the PCI Express
Base Specification for detailed information.
Table 6-7:
Bit Mapping on PCI Express Device Status Register
Bit
Name
cfg_dstatus[15:6]
Reserved
cfg_dstatus[5]
Transaction Pending
cfg_dstatus[4]
AUX Power Detected
cfg_dstatus[3]
Unsupported Request Detected
cfg_dstatus[2]
Fatal Error Detected
cfg_dstatus[1]
Non-Fatal Error Detected
cfg_dstatus[0]
Correctable Error Detected
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
89
Chapter 6: Designing with the Core
cfg_dcommand[15:0]
This bus reflects the value stored in the Device Control register of the PCI Express
Extended Capabilities Structure. Table 6-8 defines each bit in the cfg_dcommand bus. See
the PCI Express Base Specification for detailed information.
Table 6-8:
Bit Mapping of PCI Express Device Control Register
Bit
Name
cfg_dcommand[15]
Reserved
cfg_dcommand[14:12]
Max_Read_Request_Size
cfg_dcommand[11]
Enable No Snoop
cfg_dcommand[10]
Auxiliary Power PM Enable
cfg_dcommand[9]
Phantom Functions Enable
cfg_dcommand[8]
Extended Tag Field Enable
cfg_dcommand[7:5]
Max_Payload_Size
cfg_dcommand[4]
Enable Relaxed Ordering
cfg_dcommand[3]
Unsupported Request Reporting Enable
cfg_dcommand[2]
Fatal Error Reporting Enable
cfg_dcommand[1]
Non-Fatal Error Reporting Enable
cfg_dcommand[0]
Correctable Error Reporting Enable
cfg_lstatus[15:0]
This bus reflects the value stored in the Link Status register in the PCI Express Extended
Capabilities Structure. Table 6-9 defines each bit in the cfg_lstatus bus. See the PCI Express
Base Specification for details.
Table 6-9:
Bit Mapping of PCI Express Link Status Register
Bit
90
Name
cfg_lstatus[15:13]
Reserved
cfg_lstatus[12]
Slot Clock Configuration
cfg_lstatus[11]
Reserved
cfg_lstatus[10]
Reserved
cfg_lstatus[9:4]
Negotiated Link Width
cfg_lstatus[3:0]
Link Speed
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Design with Configuration Space Registers and Configuration Interface
cfg_lcommand[15:0]
This bus reflects the value stored in the Link Control register of the PCI Express Extended
Capabilities Structure. Table 6-10 provides the definition of each bit in cfg_lcommand bus.
See the PCI Express Base Specification for more details.
Table 6-10:
Bit Mapping of PCI Express Link Control Register
Bit
Name
cfg_lcommand[15:8]
Reserved
cfg_lcommand [7]
Extended Synch
cfg_lcommand [6]
Common Clock Configuration
cfg_lcommand [5]
Retrain Link (Reserved for an endpoint device)
cfg_lcommand [4]
Link Disable
cfg_lcommand [3]
Read Completion Boundary
cfg_lcommand[2]
Reserved
cfg_lcommand [1:0]
Active State Link PM Control
Accessing Additional Registers through the Configuration Port
Configuration registers that are not directly mapped to the user interface can be accessed
by configuration-space address using the ports shown in Table 2-9, page 28.
The user application must supply the read address as a DWORD address, not a byte
address. To calculate the DWORD address for a register, divide the byte address by four. For
example:
•
The DWORD address of the Command/Status Register in the PCI Configuration Space
Header is 01h. (The byte address is 04h.)
•
The DWORD address for BAR0 is 04h. (The byte address is 10h.)
To read any register in the configuration space shown in Table 2-2, page 19, the user
application drives the register DWORD address onto cfg_dwaddr[9:0]. The core drives the
content of the addressed register onto cfg_do[31:0]. The value on cfg_do[31:0] is qualified
by signal assertion on cfg_rd_wr_done_n. Figure 6-24 illustrates an example with two
consecutive reads from the Configuration Space.
X-Ref Target - Figure 6-24
trn_clk
cfg_dwaddr [9:0]
A0
A1
cfg_rd_en_n
cfg_do [31:0]
D0
D1
cfg_rd_wr_done_n
Figure 6-24:
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Example Configuration Space Access
www.xilinx.com
91
Chapter 6: Designing with the Core
User Implemented Configuration Space
The Spartan-6 FPGA Integrated Endpoint Block for PCI Express enables users to optionally
implement registers in the PCI Configuration Space, the PCI Express Extended
Configuration Space, or both, in the user application. The user application is required to
return Config Completions for all address within this space. For more information about
enabling and customizing this feature, see Chapter 5, Generating and Customizing the
Core.
PCI Configuration Space
If the user chooses to implement registers within 0x6C to 0xFF in the PCI Configuration
Space, the start address of the address region they wish to implement can be defined
during the core generation process.
The user application is responsible for generating all Completions to Configuration Reads
and Writes from the user-defined start address to the end of PCI Configuration Space
(0xFF). Configuration Reads to unimplemented registers within this range should be
responded to with a Completion with 0x00000000 as the data, and configuration writes
should be responded to with a successful Completion.
For example, to implement address range 0xC0 to 0xCF, there are several address ranges
defined that should be treated differently depending on the access. Table 6-11 shows more
details on this example.
Table 6-11:
Example: User Implemented Space 0xC0 to 0xCF
Configuration Writes
Configuration Reads
0x00 to 0xBF
Core responds automatically
Core responds automatically
0xC0 to 0xCF
User application responds with
Successful Completion
User application responds with
register contents
0xD0 to 0xFF
User application responds with
Successful Completion
User application responds with
0x00000000
PCI Express Extended Configuration Space
The starting address of the region in the PCI Express Extended Configuration Space that is
optionally available for users to implement depends on the PCI Express Extended
Capabilities the user has enabled in the Spartan-6 FPGA Integrated Endpoint Block for
PCI Express, as shown in Table 6-12.
Table 6-12:
Min Start Addresses of the User Implemented Extended Capabilities
No Capabilities Selected
DSN
100h
10Ch
Starting byte address available
The Spartan-6 FPGA Integrated Endpoint Block for PCI Express allows the user to select
the start address of the user implemented PCI Express Extended Configuration Space. This
space must be implemented in the user application. The user application is required to
generate a CplD with 0x00000000 for Configuration Read and successful Cpl for
Configuration Write to addresses in this selected range not implemented in the user
application. The user can choose to implement a Configuration Space with a start address
other than that allowed by the integrated Endpoint block for PCI Express. In such a case,
the core returns a completion with 0x00000000 for configuration accesses to the region
that the user has chosen to not implement. Table 6-13 illustrates this scenario.
92
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Additional Packet Handling Requirements
Table 6-13:
Example: User Defined Start Address for Configuration Space
Configuration Space
Byte Address
DSN Capability
100h - 108h
Reserved Extended Configuration Space
10Ch - 164h
(Core Returns Successful Completion with 0x00000000)
User Implemented PCI Express Extended Configuration Space
168h - 47Ch
User Implemented Reserved PCI Express Extended Configuration Space
480h - FFFh
(User application Returns Successful Completion with 0x00000000)
Table 6-13 illustrates an example Configuration of the PCI Express Extended
Configuration Space, with these settings:
•
DSN Capability Enabled
•
User Implemented PCI Express Extended Configuration Space Enabled
•
User Implemented PCI Express Extended Configuration Space Start Address 168h
In this configuration, the DSN Capability occupies the registers at 100h-108h. The
remaining PCI Express Extended Configuration Space, starting at address 10Ch is
available to the user to implement. For this example, the user has chosen to implement
registers in the address region starting 168h.
In this scenario, the core returns successful Completions with 0x00000000 for
Configuration accesses to registers 10Ch-164h. Table 6-13 also illustrates a case where the
user only implements the registers from 168h to 47Ch. In this case, the user is responsible
for returning successful Completions with 0x00000000 for configuration accesses to
480h-FFFh.
Additional Packet Handling Requirements
The user application must manage the following mechanisms to ensure protocol
compliance, because the core does not manage them automatically.
Generation of Completions
The integrated Endpoint block core does not generate Completions for Memory Reads or
I/O requests made by a remote device. The user is expected to service these completions
according to the rules specified in the PCI Express Base Specification.
Tracking Non-Posted Requests and Inbound Completions
The Integrated Endpoint Block for PCIe does not track transmitted I/O requests or
Memory Reads that have yet to be serviced with inbound Completions. The user
application is required to keep track of such requests using the Tag ID or other information.
Keep in mind that one Memory Read request can be answered by several Completion
packets. The user application must accept all inbound Completions associated with the
original Memory Read until all requested data has been received.
The PCI Express Base Specification requires that an endpoint advertise infinite Completion
Flow Control credits as a receiver; the endpoint can only transmit Memory Reads and I/O
requests if it has enough space to receive subsequent Completions.
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
93
Chapter 6: Designing with the Core
The integrated Endpoint block core does not keep track of receive-buffer space for
Completion. Rather, it sets aside a fixed amount of buffer space for inbound Completions.
The user application must keep track of this buffer space to know if it can transmit requests
requiring a Completion response. See Appendix E, Managing Receive-Buffer Space for
Inbound Completions for more information.
Reporting User Error Conditions
The user application must report errors that occur during Completion handling using
dedicated error signals on the core interface, and must observe the Device Power State
before signaling an error to the core. If the user application detects an error (for example, a
Completion Timeout) while the device has been programmed to a non-D0 state, the user
application is responsible to signal the error after the device is programmed back to the D0
state.
After the user application signals an error, the core reports the error on the PCI Express
Link and also sets the appropriate status bit(s) in the Configuration Space. Because status
bits must be set in the appropriate Configuration Space register, the user application
cannot generate error reporting packets on the transmit interface. The type of errorreporting packets transmitted depends on whether or not the error resulted from a Posted
or Non-Posted Request. User-reported Posted errors cause Message packets to be sent to
the Root Complex if enabled to do so through the Device Control Error Reporting bits
and/or the Status SERR Enable bit. User-reported non-Posted errors cause Completion
packets with non-successful status to be sent to the Root Complex unless the error is
regarded as an Advisory Non-Fatal Error. For more information about Advisory Non-Fatal
Errors, see Chapter 6 of the PCI Express Base Specification. Errors on Non-Posted Requests
can result in either Messages to the Root Complex or Completion packets with nonSuccessful status sent to the original Requester.
Error Types
The user application triggers six types of errors using the signals defined in Table 2-9,
page 28.
•
End-to-end CRC ECRC Error
•
Unsupported Request Error
•
Completion Timeout Error
•
Unexpected Completion Error
•
Completer Abort Error
•
Correctable Error
Multiple errors can be detected in the same received packet; for example, the same packet
can be an Unsupported Request and have an ECRC error. If this happens, only one error
should be reported. Because all user-reported errors have the same severity, the user
application design can determine which error to report. The cfg_err_posted_n signal,
combined with the appropriate error reporting signal, indicates what type of errorreporting packets are transmitted. The user can signal only one error per clock cycle. See
Figure 6-25, Figure 6-26, and Figure 6-27, and Table 6-14 and Table 6-15.
94
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Reporting User Error Conditions
Table 6-14:
User-Indicated Error Signaling
Reported Error
cfg_err_posted_n
None
Don’t care
cfg_err_ur_n
Action
No Action Taken
0 or 1
0: If enabled, a Non-Fatal Error
Message is sent.
1: A Completion with a
“status=unsupported request” is sent.
0 or 1
cfg_err_cpl_abort_n
0: If enabled, a Non-Fatal Error
message is sent.
1: A Completion with a
“status=unsupported request” is sent.
Table 6-15:
cfg_err_cpl_timeout_n
Don’t care
If enabled, a Non-Fatal Error Message
is sent.
cfg_err_ecrc_n
Don’t care
If enabled, a Non-Fatal Error Message
is sent.
cfg_err_cor_n
Don’t care
If enabled, a Correctable Error
Message is sent.
Possible Error Conditions for TLPs Received by the User Application
Received TLP Type
Possible Error Condition
Error Qualifying Signal Status
Unsupported
Request
(cfg_err_ur_n)
Completion
Abort
(cfg_err_cpl_
abort_n)
Memory Write
✓
X
N/A
Memory Read
✓
✓
I/O
✓
Completion
X
Value to Drive on
(cfg_err_
posted_n)
Drive Data on
(cfg_err_tlp_cpl_
header[47:0])
✓
0
No
N/A
✓
1
Yes
✓
N/A
✓
1
Yes
X
N/A
✓
0
No
Correctable Error
ECRC Error
(cfg_err_cor_n) (cfg_err_ecrc_n)
Notes:
1. A checkmark indicates a possible error condition for a given TLP type. For example, users can signal Unsupported Request or ECRC Error
for a Memory Write TLP, if these errors are detected. An X indicates not a valid error condition for a given TLP type. For example, users
should never signal Completion Abort in response to a Memory Write TLP.
Whenever an error is detected in a Non-Posted Request, the user application deasserts
cfg_err_posted_n and provides header information on cfg_err_tlp_cpl_header[47:0]
during the same clock cycle the error is reported, as illustrated in Figure 6-25. The
additional header information is necessary to construct the required Completion with nonSuccessful status. Additional information about when to assert or deassert
cfg_err_posted_n is provided in the following sections.
If an error is detected on a Posted Request, the user application instead asserts
cfg_err_posted_n, but otherwise follows the same signaling protocol. This results in a
Non-Fatal Message to be sent, if enabled.
The core's ability to generate error messages can be disabled by the Root Complex issuing
a configuration write to the Endpoint core's Device Control register and the PCI Command
register setting the appropriate bits to 0. For more information about these registers, see
Chapter 7 of the PCI Express Base Specification. However, error-reporting status bits are
always set in the Configuration Space whether or not their Messages are disabled.
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
95
Chapter 6: Designing with the Core
If several non-Posted errors are signaled on cfg_err_ur_n or cfg_err_cpl_abort_n in a short
amount of time, it is possible for the core to be unable to buffer them all. If that occurs, then
cfg_err_cpl_rdy_n is deasserted, and the user must cease signaling those types of errors on
the same cycle. In addition, the user must not resume signaling those types of errors until
cfg_err_cpl_rdy_n is reasserted.
X-Ref Target - Figure 6-25
clk
cfg_err_ur_n
cfg_err_cpl_ready_n
cfg_err_locked_n
cfg_err_posted_n
cfg_dcommand[3]
tx_data[31:0]*
UR Completion Locked
*Internal signal not appearing on the User Interface
Figure 6-25: Signaling Unsupported Request for Non-Posted TLP
X-Ref Target - Figure 6-26
trn_clk
cfg_err_ur_n
cfg_err_posted_n
cfg_err_tlp_cpl_header[47:0]
header
cfg_dcommand[1]
tx_data[31:0]*
Non-Fatal Message
* Internal signals not appearing on User Interface
Figure 6-26: Signaling Unsupported Request for Posted TLP
96
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Reporting User Error Conditions
X-Ref Target - Figure 6-27
clk
cfg_err_ur_n
cfg_err_cpl_ready_n
cfg_err_locked_n
cfg_err_posted_n
cfg_dcommand[3]
tx_data[31:0]*
UR Completion Locked
*Internal signal not appearing on the User Interface
Figure 6-27:
Signaling Locked Unsupported Request for Locked Non-Posted TLP
Completion Timeouts
The integrated Endpoint block core does not implement Completion timers; for this
reason, the user application must track how long its pending Non-Posted Requests have
each been waiting for a Completion and trigger timeouts on them accordingly. The core
has no method of knowing when such a timeout has occurred, and for this reason does not
filter out inbound Completions for expired requests.
If a request times out, the user application must assert cfg_err_cpl_timeout_n, which
causes an error message to be sent to the Root Complex. If a Completion is later received
after a request times out, the user application must treat it as an Unexpected Completion.
Unexpected Completions
The integrated Endpoint block core automatically reports Unexpected Completions in
response to inbound Completions whose Requestor ID is different than the Endpoint ID
programmed in the Configuration Space. These completions are not passed to the user
application. The current version of the core regards an Unexpected Completion to be an
Advisory Non-Fatal Error (ANFE), and no message is sent. Other types of unexpected
completions are passed to the user application, and the user determines how to handle
these.
Completer Abort
If the user application is unable to transmit a normal Completion in response to a
Non-Posted Request it receives, it must signal cfg_err_cpl_abort_n. The cfg_err_posted_n
signal can also be set to 1 simultaneously to indicate Non-Posted and the appropriate
request information placed on cfg_err_tlp_cpl_header[47:0]. This sends a Completion with
non-Successful status to the original Requester, but does not send an Error Message. When
in Legacy mode if the cfg_err_locked_n signal is set to 0 (to indicate the transaction causing
the error was a locked transaction), a Completion Locked with Non-Successful status is
sent. If the cfg_err_posted_n signal is set to 0 (to indicate a Posted transaction), no
Completion is sent, but a Non-Fatal Error Message is sent (if enabled).
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
97
Chapter 6: Designing with the Core
Unsupported Request
If the user application receives an inbound Request it does not support or recognize, it
must assert cfg_err_ur_n to signal an Unsupported Request. The cfg_err_posted_n signal
must also be asserted or deasserted depending on whether the packet in question is a
Posted or Non-Posted Request. If the packet is Posted, a Non-Fatal Error Message is sent
out (if enabled); if the packet is Non-Posted, a Completion with a non-Successful status is
sent to the original Requester. When in Legacy mode if the cfg_err_locked_n signal is set to
0 (to indicate the transaction causing the error was a locked transaction), a Completion
Locked with Unsupported Request status is sent.
The Unsupported Request condition can occur for several reasons, including:
•
An inbound Memory Write packet violates the user application's programming
model, for example, if the user application has been allotted a 4 KB address space but
only uses 3 KB, and the inbound packet addresses the unused portion. (Note: If this
occurs on a Non-Posted Request, the user application should use cfg_err_cpl_abort_n
to flag the error.)
•
An inbound packet uses a packet Type not supported by the user application, for
example, an I/O request to a memory-only device.
ECRC Error
The integrated Endpoint block core does not check the ECRC field for validity. If the user
application chooses to check this field, and finds the CRC is in error, it can assert
cfg_err_ecrc_n, causing a Non-Fatal Error Message to be sent.
Flow Control Credit Information
Using the Flow Control Credit Signals
The integrated Endpoint block provides the user application with information about the
state of the Transaction Layer transmit and receive buffer credit pools. This information
represents the current space available, as well as the credit “limit” and “consumed”
information for the Posted, Non-Posted, and Completion pools.
Table 2-8, page 25 defines the Flow Control Credit signals. Credit status information is
presented on these signals:
•
trn_fc_ph[7:0]
•
trn_fc_pd[11:0]
•
trn_fc_nph[7:0]
•
trn_fc_npd[11:0]
•
trn_fc_cplh[7:0]
•
trn_fc_cpld[11:0]
Collectively, these signals are referred to as trn_fc_*.
The trn_fc_* signals provide information about each of the six credit pools defined in the
PCI Express Base Specification: Header and Data Credits for Each of Posted, Non-Posted,
and Completion.
98
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Flow Control Credit Information
Six different types of flow control information can be read by the user application. The
trn_fc_sel[2:0] input selects the type of flow control information represented by the
trn_fc_* outputs. The Flow Control Information Types are shown in Table 6-16.
Table 6-16:
Flow Control Information Types
trn_fc_sel[2:0]
Flow Control Information Type
000
Receive Credits Available Space
001
Receive Credits Limit
010
Receive Credits Consumed
011
Reserved
100
Transmit Credits Available Space
101
Transmit Credit Limit
110
Transmit Credits Consumed
111
Reserved
The trn_fc_sel[2:0] signals can be changed on every clock cycle to indicate a different Flow
Control Information Type. There is a two clock-cycle delay between the value of
trn_fc_sel[2:0] changing and the corresponding Flow Control Information Type being
presented on the trn_fc_* outputs. Figure 6-28 illustrates the timing of the Flow Control
Credits signals.
X-Ref Target - Figure 6-28
trn_clk
trn_fc_sel[2:0]
000b
trn_fc_*
001b
110b
RX Avail
Figure 6-28:
RX Limit
TX Consumed
Flow Control Credits
The output values of the trn_fc_* signals represent credit values as defined in the PCI
Express Base Specification. One Header Credit is equal to either a 3 or 4 DWORD TLP
Header and one Data Credit is equal to 16 bytes of payload data. Initial credit information
is available immediately after trn_lnk_up_n assertion, but before the reception of any TLP.
Table 6-17 defines the possible values presented on the trn_fc_* signals. Initial credit
information varies depending on the size of the receive buffers within the integrated
Endpoint block and the Link Partner.
Table 6-17:
trn_fc_* Value Definition
Header Credit Value
Data Credit Value
00 – 7F
000 – 7FF
FF-80
FFF-800
7F
7FF
Meaning
User credits
Negative credits available(1)
Infinite credits available(1)
Notes:
1. Only Transmit Credits Available Space indicate Negative or Infinite credits available.
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
99
Chapter 6: Designing with the Core
Receive Credit Flow Control Information
Receive Credit Flow Control Information can be obtained by setting trn_fc_sel[2:0] to
000b, 001b, or 010b. The Receive Credit Flow Control information indicates the current
status of the receive buffers within the integrated Endpoint block.
Receive Credits Available Space: trn_fc_sel[2:0] = 000b
Receive Credits Available space shows the credit space available in the integrated
Endpoint block's Transaction Layer local receive buffers for each credit pool. If space
available for any of the credit pools is exhausted, the corresponding trn_fc_* signal
indicates a value of zero. Receive Credits Available Space returns to its initial values after
the user application has drained all TLPs from the integrated Endpoint block.
In the case where infinite credits have been advertised to the Link Partner for a specific
Credit pool, such as Completion Credits for Endpoints, the user application should use this
value along with the methods described in Appendix E, Managing Receive-Buffer Space
for Inbound Completions to avoid completion buffer overflow.
Receive Credits Limit: trn_fc_sel[2:0] = 001b
Receive Credits Limit show the credits granted to the link partner. The trn_fc_* values are
initialized with the values advertised by the integrated Endpoint block during Flow
Control initialization and are updated as a cumulative count as TLPs are read out of the
Transaction Layer's receive buffers via the TRN interface. This value is referred to as
CREDITS_ALLOCATED within the PCI Express Base Specification.
In the case where infinite credits have been advertised for a specific credit pool, the Receive
Buffer Credits Limit for that pool will always indicate zero credits.
Receive Credits Consumed: trn_fc_sel[2:0] = 010b
Receive Buffer Credits Consumed show the credits consumed by the link partner (and
received by the integrated Endpoint block). The initial trn_fc_* values are always zero
and are updated as a cumulative count, as packets are received by the Transaction Layers
receive buffers. This value is referred to as CREDITS_RECEIVED in the PCI Express Base
Specification.
Transmit Credit Flow Control Information
Transmit Credit Flow Control Information can be obtained by setting trn_fc_sel[2:0] to
100b, 101b, or 110b. The Transmit Credit Flow Control information indicates the current
status of the receive buffers within the Link Partner.
Transmit Credits Available Space: trn_fc_sel[2:0] = 100b
Transmit Credits Available Space indicates the available credit space within the receive
buffers of the Link Partner for each credit pool. If space available for any of the credit pools
is exhausted, the corresponding trn_fc_* signal indicates a value of zero or negative.
Transmit Credits Available Space returns to its initial values after the integrated Endpoint
block has successfully sent all TLPs to the Link Partner.
If the value is negative, more header or data has been written into the integrated Endpoint
block's local transmit buffers than the Link Partner can currently consume. Because the
block does not allow posted packets to pass completions, a posted packet that is written is
not transmitted if there is a completion ahead of it waiting for credits (as indicated by a
zero or negative value). Similarly, a completion that is written is not transmitted if a posted
packet is ahead of it waiting for credits. The user application can monitor the Transmit
100
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Power Management
Credits Available Space to ensure that these temporary blocking conditions do not occur,
and that the bandwidth of the PCI Express Link is fully utilized by only writing packets to
the integrated Endpoint block that have sufficient space within the Link Partner's Receive
buffer. Non-Posted packets can always be bypassed within the integrated Endpoint block;
so, any Posted or Completion packet written will pass Non-Posted packets waiting for
credits.
The Link Partner can advertise infinite credits for one or more of the three traffic types.
Infinite credits are indicated to the user by setting the Header and Data credit outputs to
their maximum value as indicated in Table 6-17.
Transmit Credits Limit: trn_fc_sel[2:0] = 101b
Transmit Credits Limit shows the receive buffer limits of the Link Partner for each credit
pool. The trn_fc_* values are initialized with the values advertised by the Link Partner
during Flow Control initialization and are updated as a cumulative count as Flow Control
updates are received from the Link Partner. This value is referred to as CREDITS_LIMIT in
the PCI Express Base Specification.
In the case where infinite credits have been advertised for a specific Credit pool, the
Transmit Buffer Credits Limit always indicates zero credits for that pool.
Transmit Credits Consumed: trn_fc_sel[2:0] = 110b
Transmit Credits Consumed show the credits consumed of the Receive Buffer of the Link
Partner by the integrated Endpoint block. The initial value is always zero and is updated as
a cumulative count, as packets are transmitted to the Link Partner. This value is referred to
as CREDITS_CONSUMED in the PCI Express Base Specification.
Power Management
The Spartan-6 FPGA Integrated Endpoint Block for PCI Express core supports these power
management modes:
•
Active State Power Management (ASPM)
•
Programmed Power Management (PPM)
Implementing these power management functions as part of the PCI Express design
enables the PCI Express hierarchy to seamlessly exchange power-management messages
to save system power. All power management message identification functions are
implemented. The sections below describe the user logic definition to support the ASPM
and PPM modes of power management.
For additional information on ASPM and PPM implementation, see the PCI Express Base
Specification.
Active State Power Management
The Active State Power Management (ASPM) functionality is autonomous and
transparent from a user-logic function perspective. The core supports the conditions
required for ASPM.
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
101
Chapter 6: Designing with the Core
Programmed Power Management
To achieve considerable power savings on the PCI Express hierarchy tree, the core
supports these link states of Programmed Power Management (PPM):
•
L0: Active State (data exchange state)
•
L1: Higher Latency, lower power standby state
•
L3: Link Off State
All PPM messages are always initiated by an upstream link partner. Programming the core
to a non-D0 state, results in PPM messages being exchanged with the upstream linkpartner. The PCI Express Link transitions to a lower power state after completing a
successful exchange.
PPM L0 State
The L0 state represents normal operation and is transparent to the user logic. The core
reaches the L0 (active state) after a successful initialization and training of the PCI Express
Link(s) as per the protocol.
PPM L1 State
These steps outline the transition of the core to the PPM L1 state:
1.
The transition to a lower power PPM L1 state is always initiated by an upstream
device, by programming the PCI Express device power state to D3-hot (or to D1 or D2
if they are supported).
2.
The core then throttles/stalls the user logic from initiating any new transactions on the
user interface by deasserting trn_tdst_rdy_n. Any pending transactions on the user
interface are however accepted fully and can be completed later.
3.
The core exchanges appropriate power management messages with its link partner to
successfully transition the link to a lower power PPM L1 state. This action is
transparent to the user logic.
4.
All user transactions are stalled for the duration of time when the device power state is
non-D0.
5.
The device power state is communicated to the user logic through the user
configuration port interface. The user logic is responsible for performing a successful
read operation to identify the device power state.
6.
The user logic, after identifying the device power state as non-D0, can initiate a request
through cfg_pm_wake_n to the upstream link partner to configure the device back to
the D0 power state.
7.
The user logic must poll the PME_Status bit of the PMCSR (via the Configuration
Interface). If a PME message is not acknowledged by the host within 100 ms
(+50%/-5%) by the host clearing the PME_Status bit, the Endpoint is required to
retransmit. This functionality is not provided by the Spartan-6 FPGA Integrated
Endpoint Block for PCI Express. For more information, see section 5.3.3.3.1 of the PCI
Express Base Specification v1.1.
Note: If the upstream link partner has not configured the device to allow the generation of
PM_PME messages (PME_En bit of PMCSR = 0), the assertion of cfg_pm_wake_n is ignored
by the core.
102
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Generating Interrupt Requests
PPM L3 State
These steps outline the transition of the Integrated Endpoint Block for PCIe core to the
PPM L3 state:
1.
The core negotiates a transition to the L23 Ready Link State upon receiving a
PME_Turn_Off message from the upstream link partner.
2.
Upon receiving a PME_Turn_Off message, the Endpoint core initiates a handshake
with the user logic through cfg_to_turnoff_n (see Table 2-8, page 25) and expects a
cfg_turnoff_ok_n back from the user logic.
3.
A successful handshake results in a transmission of the Power Management Turn-off
Acknowledge (PME-turnoff_ack) Message by the Endpoint core to its upstream link
partner.
4.
The Endpoint core closes all its interfaces, disables the Physical/DataLink/Transaction layers and is ready for removal of power to the core.
Power-down negotiation follows these steps:
1.
Before power and clock are turned off, the Root Complex or the Hot-Plug controller in
a downstream switch issues a PME_Turn_Off broadcast message.
2.
When the Endpoint PIPE for PCIe core receives this TLP, it asserts cfg_to_turnoff_n to
the user application and starts polling the cfg_turnoff_ok_n input.
3.
When the user application detects the assertion of cfg_to_turnoff_n, it must complete
any packet in progress and stop generating any new packets. After the user
application is ready to be turned off, it asserts cfg_turnoff_ok_n to the core. After
assertion or of cfg_turnoff_ok_n, the user application has committed to being turned
off.
4.
The Endpoint core sends a PME_TO_Ack when it detects assertion of
cfg_turnoff_ok_n.
X-Ref Target - Figure 6-29
trn_clk
rx_data [63 :0]*
PME_Turn_Off
cfg_to_turnoff_n
cfg_turnoff_ok_n
tx_data [31:0]*
PME_To_Ack
* Internal signals not appearing on User Interface
Figure 6-29:
Power Management Handshaking
Generating Interrupt Requests
The integrated Endpoint block supports sending interrupt requests as either legacy
interrupts or Message Signaled Interrupts (MSI). The mode is programmed using the MSI
Enable bit in the Message Control Register of the MSI Capability Structure. For more
information on the MSI capability structure, refer to section 6.8 of the PCI Local Base
Specification v3.0. The state of the MSI Enable bit is reflected by the
cfg_interrupt_msienable output:
•
cfg_interrupt_msienable = 0: Legacy Interrupt (INTx) mode
•
cfg_interrupt_msienable = 1: MSI mode
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
103
Chapter 6: Designing with the Core
If the MSI Enable bit is set to a 1, the core generates MSI requests by sending Memory Write
TLPs. If the MSI Enable bit is set to 0, the core generates legacy interrupt messages as long
as the Interrupt Disable bit in the PCI Command Register is set to 0:
•
cfg_command[10] = 0: interrupts enabled
•
cfg_command[10] = 1: interrupts disabled (request are blocked by the core)
The user application requests interrupt service in one of two ways, each of which are
described below. The user application must determine which method to use based on the
value of the cfg_interrupt_msienable output. When 0, the Legacy Interrupt method must
be used; when 1, the MSI method.
The MSI Enable bit in the MSI control register and the Interrupt Disable bit in the PCI
Command register are programmed by the Root Complex. The user application has no
direct control over these bits. Regardless of the interrupt type used, the user initiates
interrupt requests through the use of cfg_interrupt_n and cfg_interrupt_rdy_n as shown in
Table 6-18.
Table 6-18:
Interrupt Signalling
Port Name
cfg_interrupt_n
cfg_interrupt_rdy_n
Direction
Description
Input
Assert to request an interrupt. Leave asserted until the
interrupt is serviced.
Output
Asserted when the core accepts the signaled interrupt
request.
The user application requests interrupt service in one of two ways, each of which are
described below.
MSI Mode
•
As shown in Figure 6-30, the user application first asserts cfg_interrupt_n.
Additionally the user application supplies a value on cfg_interrupt_di[7:0] if
Multi-Vector MSI is enabled (see below).
•
The core asserts cfg_interrupt_rdy_n to signal that the interrupt has been accepted
and the core sends a MSI Memory Write TLP. On the following clock cycle, the user
application deasserts cfg_interrupt_n if no further interrupts are to be sent.
The MSI request is either a 32-bit addressable Memory Write TLP or a 64-bit addressable
Memory Write TLP. The address is taken from the Message Address and Message Upper
Address fields of the MSI Capability Structure, while the payload is taken from the
Message Data field. These values are programmed by system software through
configuration writes to the MSI Capability structure. When the core is configured for
Multi-Vector MSI, system software can permit Multi-Vector MSI messages by
programming a non-zero value to the Multiple Message Enable field.
The type of MSI TLP sent (32-bit addressable or 64-bit addressable) depends on the value
of the Upper Address field in the MSI capability structure. By default, MSI messages are
sent as 32-bit addressable Memory Write TLPs. MSI messages use 64-bit addressable
Memory Write TLPs only if the system software programs a non-zero value into the Upper
Address register.
When Multi-Vector MSI messages are enabled, the user application can override one or
more of the lower-order bits in the Message Data field of each transmitted MSI TLP to
differentiate between the various MSI messages sent upstream. The number of lower-order
bits in the Message Data field available to the user application is determined by the lesser
104
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Generating Interrupt Requests
of the value of the Multiple Message Capable field, as set in the CORE Generator software,
and the Multiple Message Enable field, as set by system software and available as the
cfg_interrupt_mmenable[2:0] core output. The core masks any bits in cfg_interrupt_di[7:0]
which are not configured by system software via Multiple Message Enable.
This pseudo-code shows the processing required:
// Value MSI_Vector_Num must be in range: 0 ≤ MSI_Vector_Num ≤
(2^cfg_interrupt_mmenable)-1
if (cfg_interrupt_msienable) {
// MSI Enabled
if (cfg_interrupt_mmenable > 0) { // Multi-Vector MSI Enabled
cfg_interrupt_di[7:0] = {Padding_0s, MSI_Vector_Num};
} else {
// Single-Vector MSI Enabled
cfg_interrupt_di[7:0] = Padding_0s;
}
} else {
// Legacy Interrupts Enabled
}
For example:
1.
If cfg_interrupt_mmenable[2:0] == 000b, i.e., 1 MSI Vector Enabled,
then cfg_interrupt_di[7:0] = 00h;
2.
if cfg_interrupt_mmenable[2:0] == 101b, i.e., 32 MSI Vectors Enabled,
then cfg_interrupt_di[7:0] = {{000b}, {MSI_Vector#}};
where MSI_Vector# is a 5-bit value and is allowed to be 00000b ≤ MSI_Vector# ≤ 11111b
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
105
Chapter 6: Designing with the Core
Legacy Interrupt Mode
•
As shown in Figure 6-30, the user application first asserts cfg_interrupt_n and
cfg_interrupt_assert_n to assert the interrupt. The user application should select a
specific interrupt (INTA, INTB, INTC, or INTD) using cfg_interrupt_di[7:0] as shown
in Table 6-19.
•
The core then asserts cfg_interrupt_rdy_n to indicate the interrupt has been accepted.
On the following clock cycle, the user application deasserts cfg_interrupt_n and, if the
Interrupt Disable bit in the PCI Command register is set to 0, the core sends an assert
interrupt message (Assert_INTA, Assert_INTB, and so forth).
•
After the user application has determined that the interrupt has been serviced, it
asserts cfg_interrupt_n while deasserting cfg_interrupt_assert_n to deassert the
interrupt. The appropriate interrupt must be indicated via cfg_interrupt_di[7:0].
•
The core then asserts cfg_interrupt_rdy_n to indicate the interrupt deassertion has
been accepted. On the following clock cycle, the user application deasserts
cfg_interrupt_n and the core sends a deassert interrupt message (Deassert_INTA,
Deassert_INTB, and so forth).
X-Ref Target - Figure 6-30
trn_clk
cfg_interrupt_msienable
cfg_interrupt_n
Legacy
Mode
cfg_interrupt_di
INTA
INTA
01h
07h
cfg_interrupt_assert_n
cfg_interrupt_rdy_n
cfg_interrupt_msienable
cfg_interrupt_n
MSI
Mode
cfg_interrupt_di
cfg_interrupt_rdy_n
Figure 6-30:
Table 6-19:
106
Requesting Interrupt Service: MSI and Legacy Mode
Legacy Interrupt Mapping
cfg_interrupt_di[7:0] value
Legacy Interrupt
00h
INTA
01h
INTB
02h
INTC
03h
INTD
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Clocking and Reset of the Integrated Endpoint Block Core
Clocking and Reset of the Integrated Endpoint Block Core
Reset
The Integrated Endpoint Block for PCI Express core uses sys_reset_n to reset the system,
an asynchronous, active-Low reset signal asserted during the PCI Express Fundamental
Reset. Asserting this signal causes a hard reset of the entire core, including the transceivers.
After the reset is released, the core attempts to link train and resume normal operation. In
a typical endpoint application, for example, an add-in card, a sideband reset signal is
normally present and should be connected to sys_reset_n. For endpoint applications that
do not have a sideband system reset signal, the initial hardware reset should be generated
locally. Three reset events can occur in PCI Express:
•
Cold Reset. A Fundamental Reset that occurs at the application of power. The signal
sys_reset_n is asserted to cause the cold reset of the core.
•
Warm Reset. A Fundamental Reset triggered by hardware without the removal and
re-application of power. The sys_reset_n signal is asserted to cause the warm reset to
the core.
•
Hot Reset: In-band propagation of a reset across the PCI Express Link through the
protocol. In this case, sys_reset_n is not used. In the case of Hot Reset, the
received_hot_reset signal is asserted to indicate the source of the reset.
The user application interface of the core has an output signal called trn_reset_n. This
signal is deasserted synchronously with respect to trn_clk. trn_reset_n is asserted as a
result of any of these conditions:
•
Fundamental Reset: Occurs (cold or warm) due to assertion of sys_reset_n.
•
PLL within the core: Loses lock, indicating an issue with the stability of the clock
input.
•
Loss of Transceiver PLL Lock: The Lane 0 transceiver loses lock, indicating an issue
with the PCI Express link.
The trn_reset_n signal deasserts synchronously with trn_clk after all of the above reasons
are resolved, allowing the core to attempt to train and resume normal operation.
Important Note: Systems designed to the PCI Express electro-mechanical specification
provide a sideband reset signal, which uses 3.3V signaling levels—see the FPGA device
data sheet to understand the requirements for interfacing to such signals.
Clocking
The integrated Endpoint block core input system clock signal is called sys_clk. The core
accepts a 100 or 125 MHz clock input. The clock frequency used must match the clock
frequency selection in the CORE Generator software GUI. For more information, see
Answer Record 18329.
In a typical PCI Express solution, the PCI Express reference clock is a spread spectrum
clock (SSC), provided at 100 MHz. In most commercial PCI Express systems, SSC cannot be
disabled. For more information regarding SSC and PCI Express, see section 4.3.1.1.1 of the
PCI Express Base Specification.
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
107
Chapter 6: Designing with the Core
Synchronous and Non-Synchronous Clocking
There are two ways to clock the PCI Express system:
•
Using synchronous clocking, where a shared clock source is used for all devices.
•
Using non-synchronous clocking, where each device has its own clock source.
Important Note: Xilinx recommends that designers use synchronous clocking when
using the core. All add-in card designs must use synchronous clocking due to the
characteristics of the provided reference clock. See the Spartan-6 FPGA GTP
Transceivers User Guide and device data sheet for additional information regarding
reference clock requirements.
For synchronous clocked systems, each link partner device shares the same clock source.
When using the 125 MHz reference clock option, an external PLL must be used to do a
multiply of 5/4 to convert the 100 MHz clock to 125 MHz, as illustrated in Figure 6-32 and
Figure 6-33. See Answer Record 18329 for more information about clocking requirements.
Further, even if the device is part of an embedded system, if the system uses commercial
PCI Express root complexes or switches along with typical mother board clocking
schemes, synchronous clocking should still be used as shown in Figure 6-32.
Figure 6-31, Figure 6-32, and Figure 6-33 illustrate high-level representations of the board
layouts. Designers must ensure that proper coupling, termination, and so forth are used
when laying out the board.
X-Ref Target - Figure 6-31
PCI Express Gen 1 Add-In Card
Spartan-6 FPGA
PCI Express
Endpoint
Device
100 MHz with SSC
PCI Express Clock
PCIe Link
PCIe Link
GTP
Transceivers
_
PCIe Link
+
PCIe Link
PCI Express Connector
UG654_c6_31_032510
Figure 6-31:
108
Spartan-6 FPGA PCI Express Gen 1 Using 100 MHz Reference Clock
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Clocking and Reset of the Integrated Endpoint Block Core
X-Ref Target - Figure 6-32
PCI Express
Switch or Root
Complex
Device
GTP
Transceiver
Embedded System Board
PCIe Link
PCIe Link
Spartan-6 FPGA
Endpoint
100 MHz
125 MHz
100 MHz
PCI Express
Clock Oscillator
External PLL
UG654_c6_32_032510
Figure 6-32:
Embedded System Using 125 MHz Reference Clock
X-Ref Target - Figure 6-33
PCI Express Add-In Card
+
External PLL
+
Spartan-6 FPGA
Endpoint
125 MHz
_
GTP
Transceivers
_
PCIe Link
PCIe Link
100 MHz with SSC
PCI Express Clock
_
PCIe Link
+
PCIe Link
PCI Express Connector
UG654_c6_33_032510
Figure 6-33:
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Open System Add-In Card Using 125 MHz Reference Clock
www.xilinx.com
109
Chapter 6: Designing with the Core
110
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Chapter 7
Core Constraints
The Spartan®-6 FPGA Integrated Endpoint Block for PCI Express® solution requires the
specification of timing and other physical implementation constraints to meet specified
performance requirements for PCI Express. These constraints are provided with the
Endpoint Solution in a User Constraints File (UCF). Pinouts and hierarchy names in the
generated UCF correspond to the provided example design.
To achieve consistent implementation results, a UCF containing these original, unmodified
constraints must be used when a design is run through the Xilinx tools. For additional
details on the definition and use of a UCF or specific constraints, see the Xilinx Libraries
Guide and/or Development System Reference Guide.
Constraints provided with the integrated Endpoint block solution have been tested in
hardware and provide consistent results. Constraints can be modified, but modifications
should only be made with a thorough understanding of the effect of each constraint.
Additionally, support is not provided for designs that deviate from the provided
constraints.
Contents of the User Constraints File
Although the UCF delivered with each core shares the same overall structure and sequence
of information, the content of each core’s UCF varies. The sections that follow define the
structure and sequence of information in a generic UCF file.
Part Selection Constraints: Device, Package, and Speed Grade
The first section of the UCF specifies the exact device for the implementation tools to
target, including the specific part, package, and speed grade. In some cases, device-specific
options can be included. The device in the UCF reflects the device chosen in the
CORE Generator software project.
User Timing Constraints
The User Timing constraints section is not populated; it is a placeholder for the designer to
provide timing constraints on user-implemented logic.
User Physical Constraints
The User Physical constraints section is not populated; it is a placeholder for the designer
to provide physical constraints on user-implemented logic.
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
111
Chapter 7: Core Constraints
Core Pinout and I/O Constraints
The Core Pinout and I/O constraints section contains constraints for I/Os belonging to the
core's System (SYS) and PCI Express (PCI_EXP) interfaces. It includes location constraints
for pins and I/O logic as well as I/O standard constraints.
Core Physical Constraints
Physical constraints are used to limit the core to a specific area of the device and to specify
locations for clock buffering and other logic instantiated by the core.
Core Timing Constraints
This Core Timing constraints section defines clock frequency requirements for the core and
specifies which nets the timing analysis tool should ignore.
Required Modifications
Some constraints provided in the UCF utilize hierarchical paths to elements within the
integrated Endpoint block core. These constraints assume an instance name of core for the
core. If a different instance name is used, replace core with the actual instance name in all
hierarchical constraints.
For example:
Using xilinx_pcie_ep as the instance name, the physical constraint
INST core/GT_i/tile0_gtpa1_dual_wrapper_i/gtpa1_dual_i
LOC = GTPA1_DUAL_X0Y0
becomes
INST xilinx_pcie_ep/GT_i/tile0_gtpa1_dual_wrapper_i/gtpa1_dual_i
LOC = GTPA1_DUAL_X0Y0
The provided UCF includes a line specifying attributes for the sys_reset_n pin, but it is up
to the user to un-comment that line and provide a pin location. In addition, the UCF
includes blank sections for constraining user-implemented logic. While the constraints
provided adequately constrain the integrated Endpoint block core itself, they cannot
adequately constrain user-implemented logic interfaced to the core. Additional constraints
must be implemented by the designer.
Device Selection
The device selection portion of the UCF informs the implementation tools which part,
package, and speed grade to target for the design. Because integrated Endpoint block cores
are designed for specific part and package combinations, this section should not be
modified by the designer.
The device selection section always contains a part selection line, but can also contain part
or package-specific options. An example part selection line:
CONFIG PART = xc6slx45t-fgg484-2;
112
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Core I/O Assignments
Core I/O Assignments
This section controls the placement and options for I/Os belonging to the core's System
(SYS) interface and PCI Express (PCI_EXP) interface. NET constraints in this section
control the pin location and I/O options for signals in the SYS group. Locations and
options vary depending on which derivative of the core is used and should not be changed
without fully understanding the system requirements.
For example:
NET sys_clk_p LOC = Y4;
NET sys_clk_n LOC = Y3;
See Clocking and Reset of the Integrated Endpoint Block Core, page 107 for detailed
information about reset and clock requirements.
Each GTPA1_DUAL tile contains two transceivers. Any GTPA1_DUAL tile along the top
edge of the device can be used with the integrated Endpoint block for PCIe. Either of the
two transceivers in the GTPA1_DUAL tile can be used. For GTPA1_DUAL pinout
information, see the Spartan-6 FPGA GTP Transceivers User Guide.
INST constraints are used to control placement of signals that belong to the PCI_EXP
group. These constraints control the location of the transceiver(s) used, which implicitly
controls pin locations for the transmit and receive differential pair. The provided
transceiver wrapper file consumes both transceivers in a tile even though only one is used.
For example:
INST core/GT_i/tile0_gtpa1_dual_wrapper_i/gtpa1_dual_i LOC =
GTPA1_DUAL_X0Y0;
Core Physical Constraints
Physical constraints can be included in the constraints file to control the location of
clocking and memory elements. Specific physical constraints are chosen to match each
supported device and package combination—it is very important to leave these constraints
unmodified except for changing the hierarchical name, as described above.
Core Timing Constraints
Timing constraints are provided for all integrated Endpoint block solutions, although they
differ based on core configuration. In all cases they are crucial and must not be modified,
except to specify the top-level hierarchical name. Timing constraints are divided into two
categories:
•
TIG constraints. Used on paths where specific delays are unimportant, to instruct the
timing analysis tools to refrain from issuing Unconstrained Path warnings.
•
Frequency constraints. Group clock nets into time groups and assign properties and
requirements to those groups.
TIG constraints example:
NET sys_reset_n TIG;
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
113
Chapter 7: Core Constraints
Clock constraints example:
First, the input reference clock period is specified, which can be either 100 or 125 MHz
(selected in the CORE Generator™ software GUI).
NET sys_clk_c PERIOD = 8ns;
Next, the internally generated clock net and period is specified, which can be 100 or
125 MHz. (Both clock constraints must be specified as having the same period.)
NET core/gt_refclk_out(0) TNM_NET = GT_REFCLK_OUT ;
TIMESPEC TS_GT_REFCLK_OUT = PERIOD GT_REFCLK_OUT 8ns HIGH 50 % ;
Relocating the Integrated Endpoint Block
While Xilinx does not provide technical support for designs whose system clock input,
GTP transceivers, or block RAM locations are different from the provided examples, it is
possible to relocate the core within the FPGA. The locations selected in the provided
examples are the recommended pinouts. These locations have been chosen based on the
proximity to the Endpoint Block, which enables meeting timing, and because they are
conducive to layout requirements for add-in card design. If the core is moved, the relative
location of all transceivers and clocking resources should be maintained to ensure timing
closure.
Supported Core Pinouts
Table 7-1 defines the supported core pinouts for the available LXT part and package
combinations. The CORE Generator software provides a UCF for the selected part and
package that matches the content of this table.
XC6SLX45T
XC6SLX45T
CSG484
XC6SLX45T
XC6SLX45T
XC6SLX45T
114
0
A8
B8
A4
B4
C5
D5
1
C9
D9
A6
B6
C7
D7
0
A10
B10
A12
B12
C11
D11
1
E10
F10
A14
B14
C13
D13
0
A10
B10
A6
B6
C7
D7
1
C11
D11
A8
B8
C9
D9
0
A12
B12
A14
B14
C13
D13
1
E14
F14
A16
B16
C15
D15
X0Y0
XC6SLX45T
XC6SLX45T
pci_exp_rxp
CSG324
XC6SLX25T
pci_exp_rxn
XC6SLX45T
pci_exp_txp
XC6SLX25T
pci_exp_txn
GTPA1_DUAL
sys_clk_p
Part
sys_clk_n
Package
Spartan-6 FPGA LXT Pinout
Channel
Table 7-1:
X1Y0
X0Y0
X1Y0
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Supported Core Pinouts
pci_exp_txn
pci_exp_txp
pci_exp_rxn
pci_exp_rxp
GTPA1_DUAL
sys_clk_p
Part
sys_clk_n
Package
Spartan-6 FPGA LXT Pinout (Cont’d)
Channel
Table 7-1:
0
A10
B10
A6
B6
C7
D7
1
C11
D11
A8
B8
C9
D9
0
A12
B12
A14
B14
C13
D13
1
E14
F14
A16
B16
C15
D15
0
B10
A10
A6
B6
C7
D7
1
D11
C11
A8
B8
C9
D9
0
B12
A12
A14
B14
C13
D13
1
F12
E12
A16
B16
C15
D15
0
B10
A10
A6
B6
C7
D7
1
D11
C11
A8
B8
C9
D9
0
B12
A12
A14
B14
C13
D13
1
F12
E12
A16
B16
C15
D15
XC6SLX75T
XC6SLX100T
XC6SLX150T
XC6SLX75T
X0Y1
XC6SLX100T
CSG484
XC6SLX150T
XC6SLX75T
XC6SLX100T
XC6SLX150T
XC6SLX75T
X1Y1
XC6SLX100T
XC6SLX150T
XC6SLX25T
XC6SLX45T
FGG484
XC6SLX25T
X0Y0
XC6SLX45T
XC6SLX45T
XC6SLX45T
X1Y0
XC6SLX75T
XC6SLX100T
XC6SLX150T
XC6SLX75T
X0Y1
XC6SLX100T
FGG484
XC6SLX150T
XC6SLX75T
XC6SLX100T
XC6SLX150T
XC6SLX75T
X1Y1
XC6SLX100T
XC6SLX150T
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
115
Chapter 7: Core Constraints
pci_exp_txn
pci_exp_txp
pci_exp_rxn
pci_exp_rxp
GTPA1_DUAL
sys_clk_p
Part
sys_clk_n
Package
Spartan-6 FPGA LXT Pinout (Cont’d)
Channel
Table 7-1:
0
A10
B10
A6
B6
C7
D7
1
C11
D11
A8
B8
C9
D9
0
C15
D15
A18
B18
C17
D17
1
A16
B16
A20
B20
C19
D19
0
A13
B13
A9
B9
C10
D10
1
C14
D14
A11
B11
C12
D12
0
C18
D18
A21
B21
C20
D20
1
A19
B19
A23
B23
C22
D22
XC6SLX75T
XC6SLX100T
XC6SLX150T
XC6SLX75T
X0Y1
XC6SLX100T
FGG676
XC6SLX150T
XC6SLX75T
XC6SLX100T
XC6SLX150T
XC6SLX75T
X1Y1
XC6SLX100T
XC6SLX150T
XC6SLX100T
XC6SLX150T
XC6SLX100T
FGG900
XC6SLX150T
XC6SLX100T
XC6SLX150T
XC6SLX100T
XC6SLX150T
116
X0Y1
X1Y1
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Chapter 8
FPGA Configuration
This chapter discusses how to configure the Spartan®-6 FPGA so that the device can link
up and be recognized by the system. This information is provided so the user can choose
the correct FPGA configuration method for the system and verify that it works as expected.
This chapter discusses how specific requirements of the PCI Express Base Specification and
PCI Express Card Electromechanical Specification apply to FPGA configuration. Where
appropriate, Xilinx recommends that the user read the actual specifications for detailed
information. This chapter is divided into four sections:
•
Configuration Terminology. Defines terms used in this chapter.
•
Configuration Access Time. Several specification items govern when an Endpoint
device needs to be ready to receive configuration accesses from the host (Root
Complex).
•
Board Power in Real-World Systems. Understanding real-world system constraints
related to board power and how they affect the specification requirements.
•
Recommendations. Describes methods for FPGA configuration and includes sample
problem analysis for FPGA configuration timing issues.
Configuration Terminology
In this chapter, these terms are used to differentiate between FPGA configuration and
configuration of the PCI Express device:
•
Configuration of the FPGA. FPGA configuration is used.
•
Configuration of the PCI Express device. After the link is active, configuration is used.
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
117
Chapter 8: FPGA Configuration
Configuration Access Time
In standard systems for PCI Express, when the system is powered up, configuration
software running on the processor starts scanning the PCI Express bus to discover the
machine topology.
The process of scanning the PCI Express hierarchy to determine its topology is referred to
as the enumeration process. The root complex accomplishes this by initiating configuration
transactions to devices as it traverses and determines the topology.
All PCI Express devices are expected to have established the link with their link partner
and be ready to accept configuration requests during the enumeration process. As a result,
there are requirements as to when a device needs to be ready to accept configuration
requests after power up; if the requirements are not met, the following occurs:
•
If a device is not ready and does not respond to configuration requests, the root
complex does not discover it and treats it as non-existent.
•
The operating system does not report the device's existence and the user's application
is not able to communicate with the device.
Choosing the appropriate FPGA configuration method is key to ensuring the device is able
to communicate with the system in time to achieve link up and respond to the
configuration accesses.
Configuration Access Specification Requirements
Two PCI Express specification items are relevant to configuration access:
1.
Section 6.6 of PCI Express Base Specification, rev 1.1 states “A system must guarantee
that all components intended to be software visible at boot time are ready to receive
Configuration Requests within 100 ms of the end of Fundamental Reset at the Root
Complex.” For detailed information about how this is accomplished, see the
specification; it is beyond the scope of this discussion.
Xilinx compliance to this specification is validated by the PCI Express-CV tests. The PCI
Special Interest Group (PCI-SIG) provides the PCI Express Configuration Test Software to
verify the device meets the requirement of being able to receive configuration accesses
within 100 ms of the end of the fundamental reset. The software, available to any member
of the PCI-SIG, generates several resets using the in-band reset mechanism and PERST#
toggling to validate robustness and compliance to the specification.
2.
Section 6.6 of PCI Express Base Specification, rev 1.1 defines three parameters necessary
“where power and PERST# are supplied.” The parameter TPVPERL applies to FPGA
configuration timing and is defined as:
TPVPERL - PERST# must remain active at least this long after power becomes valid.
118
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Configuration Access Time
The PCI Express Base Specification does not give a specific value for TPVPERL – only its
meaning is defined. The most common form factor used by designers with the integrated
Endpoint block core is an ATX-based form factor. The PCI Express Card Electromechanical
Specification focuses on requirements for ATX-based form factors. This applies to most
designs targeted to standard desktop or server type motherboards. Figure 8-1 shows the
relationship between Power Stable and PERST#. (This figure is based on Figure 2-10 from
section 2.1 of PCI Express Card Electromechanical Specification, rev 1.1.)
X-Ref Target - Figure 8-1
Power Stable
3.3 Vaux
3.3V/12V
PERST#
PCI Express Link
Inactive
Active
100 ms
T
PVPERL
UG654_c8_01_030910
Figure 8-1: Power Up
Section 2.6.2 of the PCI Express Card Electromechanical Specification defines TPVPREL as a
minimum of 100 ms, indicating that from the time power is stable the system reset is
asserted for at least 100 ms (as shown in Table 8-1).
Table 8-1:
TPVPERL Specification
Symbol
TPVPERL
Parameter
Min
Power stable to
PERST# inactive
100
Max
Units
ms
From Figure 8-1 and Table 8-1, it is possible to obtain a simple equation to define the FPGA
configuration time as follows:
FPGA Configuration Time ≤ TPWRVLD + TPVPERL
Equation 8-1
Given that TPVPERL is defined as 100 ms minimum, this becomes:
FPGA Configuration Time ≤ TPWRVLD + 100 ms
Equation 8-2
Note: Although TPWRVLD is included in Equation 8-2, it has yet to be defined in this discussion
because it depends on the type of system in use. The Board Power in Real-World Systems section
defines TPWRVLD for both ATX-based and non ATX-based systems.
FPGA configuration time is only relevant at cold boot; subsequent warm or hot resets do
not cause reconfiguration of the FPGA. If the design appears to be having problems due to
FPGA configuration, the user should issue a warm reset as a simple test, which resets the
system, including the PCI Express link, but keeps the board powered. If the problem does
not appear, the issue could be FPGA configuration time related.
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
119
Chapter 8: FPGA Configuration
Board Power in Real-World Systems
Several boards are used in PCI Express systems. The ATX Power Supply Design
specification, endorsed by Intel, is used as a guideline and for this reason followed in the
majority of mother boards and 100% of the time if it is an Intel-based motherboard. The
relationship between power rails and power valid signaling is described in the ATX 12V
Power Supply Design Guide. Figure 8-2, redrawn here and simplified to show the
information relevant to FPGA configuration, is based on the information and diagram
found in section 3.3 of the ATX 12V Power Supply Design Guide. For the entire diagram and
definition of all parameters, see the ATX 12V Power Supply Design Guide.
Figure 8-2 shows that power stable indication from Figure 8-1 for the PCI Express system is
indicated by the assertion of PWR_OK. PWR_OK is asserted High after some delay once
the power supply has reached 95% of nominal.
X-Ref Target - Figure 8-2
T1
VAC
PS_ON#
+12VDC
+5VDC
+3.3VDC
}
95%
O/P's
10%
T2
T3
PWR_OK
T4
T1 = Power On Time (T1 < 500 ms)
T2 = Risetime (0.1 ms <= T2 <= 20 ms)
T3 = PWR_OK Delay (100 ms < T3 < 500 ms)
T4 = PWR_OK risetime (T4 <= 10 ms)
Figure 8-2:
ATX Power Supply
Figure 8-2 shows that power is actually valid before PWR_OK is asserted High. This is
represented by T3 and is the PWR_OK delay. The ATX 12V Power Supply Design Guide
defines PWR_OK as 100 ms < T3 < 500 ms, indicating the following: From the point at
which the power level reaches 95% of nominal, there is a minimum of at least 100 ms but
no more than 500 ms of delay before PWR_OK is asserted. Remember, according to the PCI
Express Card Electromechanical Specification, the PERST# is guaranteed to be asserted a
minimum of 100 ms from when power is stable indicated in an ATX system by the
assertion of PWR_OK.
Again, the FPGA configuration time equation is:
FPGA Configuration Time ≤ TPWRVLD + 100 ms
120
www.xilinx.com
Equation 8-3
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Recommendations
TPWRVLD is defined as PWR_OK delay period, that is, TPWRVLD represents the amount of
time that power is valid in the system before PWR_OK is asserted. This time can be added
to the amount of time the FPGA has to configure. The minimum values of T2 and T4 are
negligible and considered zero for purposes of these calculations. For ATX-based
motherboards, which represent the majority of real-world motherboards in use, TPWRVLD
can be defined as:
100 ms ≤ TPWRVLD ≤ 500 ms
This provides the following requirement for FPGA configuration time in both ATX and
non-ATX-based motherboards:
•
FPGA Configuration Time ≤ 200 ms (for ATX based motherboard)
•
FPGA Configuration Time ≤ 100 ms (for non-ATX based motherboard)
The second equation for the non-ATX based motherboards assumes a TPWRVLD value of
0 ms because it is not defined in this context. Designers with non-ATX based motherboards
should evaluate their own power supply design to obtain a value for TPWRVLD.
This chapter assumes that the FPGA power (VCCINT) is stable before or at the same time as
PWR_OK is asserted. If this is not the case, additional time must be subtracted from the
available time for FPGA configuration. Xilinx recommends to avoid designing add-in
cards that have staggered voltage regulators with long delays.
Hot-Plug Systems
Hot-plug systems generally employ the use of a hot-plug power controller located on the
system motherboard. Many discrete hot-plug power controllers extend TPVPERL beyond
the minimum 100 ms. Add-in card designers should consult the hot-plug power controller
data sheet to determine the value of TPVPERL. If the hot-plug power controller is unknown,
a TPVPERL value of 100 ms should be assumed.
Recommendations
Xilinx recommends using a Quad-SPI Flash device in Master Serial/SPI mode with a
CCLK frequency of 33 MHz, which allows time for the FPGA configuration of the
Spartan-6 FPGA in ATX-based motherboards. Configuration options are shown as green
cells in Table 8-2 and Table 8-3 depending on the type of system in use. This section
discusses these recommendations and includes sample analysis of potential problems that
might arise during FPGA configuration.
FPGA Configuration Times for Spartan-6 Devices
During power up, the FPGA configuration sequence is performed in four steps:
1.
Wait for POR (Power on Reset) for all voltages (VccInt, Vccaux, and VccO) in the
FPGA to trip, referred to as POR Trip Time
2.
Wait for completion (deassertion) of INIT to allow the FPGA to initialize before
accepting a bitstream transfer.
Note: As a general rule, steps 1 and 2 require ≤ 50ms
3.
Wait for assertion of DONE, the actual time required for a bitstream to transfer, and
depends on:
•
Bitstream size
•
Clock frequency
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
121
Chapter 8: FPGA Configuration
•
Transfer mode used in the Flash Device
-
SPI = Serial Peripheral Interface
-
BPI = Byte Peripheral Interface
-
PFP = Platform Flash PROMs
For detailed information about the configuration process, see the Spartan-6 FPGA
Configuration User Guide.
Table 8-2 and Table 8-3 show the comparative data for all Spartan-6 FPGA LXT devices
with respect to a variety of flash devices and programming modes. The default clock rate
for configuring the device is always 2 MHz. Any reference to a different clock rate implies
a change in the settings of the device being used to program the FPGA. The configuration
clock (CCLK), when driven by the FPGA, has variation and is not exact. See UG380,
Spartan-6 FPGA Configuration Guide, for more information on CCLK tolerances.
Configuration Time Matrix: ATX Motherboards
Table 8-2 shows the configuration methods that allow the device to be configured before
the end of the fundamental reset in ATX-based systems. The table values represent the
bitstream transfer time only. The matrix is color-coded to show which configuration
methods allow the device to configure within 200 ms once the FPGA initialization time is
included. Choose a configuration method shaded in green when using ATX-based systems
to ensure that the device is recognized.
Table 8-2: Configuration Time Matrix (ATX Motherboards): Spartan-6 FPGA
Bitstream Transfer Time in Milliseconds
Bitstream (Bits)
SPIx4(1)
XCF32P(2)
(Slave-SMAPx8)
XC6SLX25T
6,411,440
36
25
XC6SLX45T
11,875,104
66
45
XC6SLX75T
19,624,608
110
75
XC6SLX100T
26,543,136
148
101
XC6SLX150T
33,761,568
188
128
Spartan-6 FPGA
GREEN: Bitstream Transfer Time + FPGA INIT Time (50 ms) <= 200 ms
YELLOW: Bitstream Transfer Time + FPGA INIT Time (50 ms) > 200 ms
RED: Bitstream Transfer Time > 200 ms
Notes:
1. CCLK assumptions: 45 MHz
2. CCLK assumptions: 33 MHz
122
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Recommendations
Configuration Time Matrix: Non-ATX-Based Motherboards
Table 8-3 shows the configuration methods that allow the device to be configured before
the end of the fundamental reset in non-ATX-based systems. This assumes TPWRVLD is
zero. The table values represent the bitstream transfer time only. The matrix is color-coded
to show which configuration methods allow the device to configure within 100 ms once the
FPGA initialization time is included. Choose a configuration method shaded in green
when using non-ATX-based systems to ensure that the device is recognized.
For some of the larger FPGAs, it might not be possible to configure within the 100 ms
window. In these cases, the user system needs to be evaluated to see if any margin is
available that can be assigned to TPWRVLD.
Table 8-3: Configuration Time Matrix (Generic Platforms: Non-ATX Motherboards):
Spartan-6 FPGA Bitstream Transfer Time in Milliseconds
Bitstream (Bits)
SPIx4(1)
XCF32P(2)
(Slave-SMAPx8)
XC6SLX25T
6,411,440
36
25
XC6SLX45T
11,875,104
66
45
XC6SLX75T
19,624,608
110
75
XC6SLX100T
26,543,136
148
101
XC6SLX150T
33,761,568
188
128
Spartan-6 FPGA
GREEN: Bitstream Transfer Time + FPGA INIT Time (50 ms) <= 100 ms
YELLOW: Bitstream Transfer Time + FPGA INIT Time (50 ms) > 100 ms
RED: Bitstream Transfer Time > 100 ms
Notes:
1. CCLK assumptions: 45 MHz
2. CCLK assumptions: 33 MHz
Sample Problem Analysis
This section presents data from an ASUS PL5 system to demonstrate the relationships
between Power Valid, FPGA Configuration, and PERST#. Figure 8-3 shows a case where
the endpoint failed to be recognized due to a FPGA configuration time issue. Figure 8-4
shows a successful FPGA configuration with the endpoint being recognized by the system.
Failed FPGA Recognition
Figure 8-3 illustrates a failed cold boot test using the default configuration time on an
LX50T FPGA. In this example, the host failed to recognize the Xilinx FPGA. Although a
second PERST# pulse assists in allowing more time for the FPGA to configure, the
slowness of the FPGA configuration clock (2 MHz) causes configuration to complete well
after this second deassertion. During this time, the system enumerated the bus and did not
recognize the FPGA.
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
123
Chapter 8: FPGA Configuration
X-Ref Target - Figure 8-3
Figure 8-3:
Default Configuration Time on LX50T Device (2 MHz Clock)
Successful FPGA Recognition
Figure 8-4 illustrates a successful cold boot test on the same system. In this test, the CCLK
was running at 50 MHz, allowing the FPGA to configure in time to be enumerated and
recognized. The figure shows that the FPGA began initialization approximately 250 ns
before PWR_OK. DONE going High shows that the FPGA was configured even before
PWR_OK was asserted.
X-Ref Target - Figure 8-4
Figure 8-4: Fast Configuration Time on LX50T Device (50 MHz Clock)
124
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Recommendations
Workarounds for Closed Systems
For failing FPGA configuration combinations, as represented by pink cells and yellow cells
in Table 8-2 and Table 8-3, designers might be able to work around the problem in closed
systems or systems where they can guarantee behavior. These options are not
recommended for products where the targeted end system is unknown.
1.
Check if the motherboard and BIOS generate multiple PERST# pulses at start-up. This
can be determined by capturing the signal on the board using an oscilloscope. (This is
similar to what is shown in Figure 8-3. If multiple PERST#s are generated, this
typically adds extra time for FPGA configuration.
Define TPERSTPERIOD as the total sum of the pulse width of PERST# and deassertion
period before the next PERST# pulse arrives. Because the FPGA is not power cycled or
reconfigured with additional PERST# assertions, the TPERSTPERIOD number can be
added to the FPGA configuration equation.
FPGA Configuration Time ≤ TPWRVLD + TPERSTPERIOD + 100 ms
2.
In closed systems, it might be possible to create scripts to force the system to perform
a warm reset after the FPGA is configured, after the initial power up sequence. This
resets the system along with the PCI Express sub-system allowing the device to be
recognized by the system.
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
125
Chapter 8: FPGA Configuration
126
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Chapter 9
Known Restrictions
This chapter describes restrictions or issues where the integrated Endpoint block deviates
from the PCI Express Base Specification, Rev.1.1, or in cases where the specification is
ambiguous. All issues listed in this chapter are considered low impact and are not a
concern for most applications. The Comments sections describe where the associated
problem might occur so that designers can decide quickly if further investigation is
needed.
Master Data Parity Error Bit Set Incorrectly
The Master Data Parity Error bit of the Status Register is erroneously set when the Error
Poisoned status bit is set in Completion with Data TLPs.
Area of Impact
Configuration Space
Detailed Description
Transmitting a Completion with Data TLP (CplD) with the Error Poisoned (EP) bit set to 1b
causes the Master Data Parity Error bit (bit 8) in the Status Register (PCI Configuration
Space Header address 0x006) to be set. This is not allowed for Endpoints.
Comments
There are no hardware-related side effects to setting the Master Data Parity Error bit.
System software effects are system dependent; however, it is unlikely that software will
react to this bit being set in an Endpoint.
Non-Posted UpdateFC During PPM Transition
A Non-Posted UpdateFC DLLP is not sent immediately after the link transitions from
power state L1 to L0 (due to PPM non-D0).
Area of Impact
Programmed Power Management (PPM)
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
127
Chapter 9: Known Restrictions
Detailed Description
When the link partner has used up all of the integrated Endpoint block’s Non-Posted
credits, and then places the integrated Endpoint block into a non-D0 PPM state, the
integrated Endpoint block must immediately send a Non-Posted UpdateFC DLLP
following an exit from state L1 to L0 because of a PPM change from the non-D0 state. The
integrated Endpoint block does not send the Non-Posted UpdateFC immediately upon
entry to D0. It waits for an internal timer to time-out, which leads to temporary reduced
performance.
Comments
The probability of this error occurring is extremely low. The link partner would have to
deplete all non-posted credits in the integrated Endpoint block, and then immediately put
the integrated Endpoint block into a non-D0 PPM state. It is unlikely that users would
want to change to a non-D0 PPM state while there are outstanding non-posted requests.
To avoid this temporary condition of reduced performance, users should ensure there are
no outstanding non-posted requests before moving to a non-D0 PPM state.
128
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Appendix A
Programmed Input/Output Example
Design
Programmed Input/Output (PIO) transactions are generally used by a PCI Express®
system host CPU to access Memory Mapped Input/Output (MMIO) and Configuration
Mapped Input/Output (CMIO) locations in the PCI Express fabric. Endpoints for
PCI Express accept Memory and I/O Write transactions and respond to Memory and I/O
Read transactions with Completion with Data transactions.
The PIO example design (PIO design) is included with the Endpoint for PCI Express
generated by the CORE Generator™ software, which allows users to easily bring up their
system board with a known established working design to verify the link and functionality
of the board.
Note: The PIO design Port Model is shared by the Spartan®-6 FPGA Integrated Endpoint Block for
PCI Express, Endpoint Block Plus for PCI Express, and Endpoint PIPE for PCI Express solutions.
This appendix represents all the solutions generically using the name Endpoint for PCI Express (or
Endpoint for PCIe®).
System Overview
The PIO design is a simple target-only application that interfaces with the Endpoint for
PCIe core’s Transaction (TRN) interface and is provided as a starting point for customers to
build their own designs. These features are included:
•
Four transaction-specific 2 KB target regions using the internal block RAMs,
providing a total target space of 8192 bytes
•
Supports single DWORD payload Read and Write PCI Express transactions to
32/64-bit address memory spaces and I/O space with support for completion TLPs
•
Utilizes the core’s trn_rbar_hit_n[6:0] signals to differentiate between TLP
destination Base Address Registers
•
Provides separate implementations optimized for 32-bit and 64-bit TRN interfaces
Figure A-1 illustrates the PCI Express system architecture components, consisting of a
Root Complex, a PCI Express switch device, and an Endpoint for PCIe. PIO operations
move data downstream from the Root Complex (CPU register) to the Endpoint, and/or
upstream from the Endpoint to the Root Complex (CPU register). In either case, the
PCI Express protocol request to move the data is initiated by the host CPU.
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
129
Appendix A: Programmed Input/Output Example Design
X-Ref Target - Figure A-1
PCIe
Root Complex
CPU
Main
Memory
Memory
Controller
Device
PCI_BUS_0
PCIe
Port
PCI_BUS_1
PCIe
Switch
PCI_BUS_X
PCIe
Endpoint
Figure A-1:
System Overview
Data is moved downstream when the CPU issues a store register to a MMIO address
command. The Root Complex typically generates a Memory Write TLP with the
appropriate MMIO location address, byte enables and the register contents. The
transaction terminates when the Endpoint receives the Memory Write TLP and updates the
corresponding local register.
Data is moved upstream when the CPU issues a load register from a MMIO address
command. The Root Complex typically generates a Memory Read TLP with the
appropriate MMIO location address and byte enables. The Endpoint generates a
Completion with Data TLP once it receives the Memory Read TLP. The Completion is
steered to the Root Complex and payload is loaded into the target register, completing the
transaction.
PIO Hardware
The PIO design implements a 8192 byte target space in FPGA block RAM, behind the
Endpoint for PCIe. This 32-bit target space is accessible through single DWORD I/O Read,
I/O Write, Memory Read 64, Memory Write 64, Memory Read 32, and Memory Write 32
TLPs.
The PIO design generates a completion with 1 DWORD of payload in response to a valid
Memory Read 32 TLP, Memory Read 64 TLP, or I/O Read TLP request presented to it by
the core. In addition, the PIO design returns a completion without data with successful
status for I/O Write TLP request.
The PIO design processes a Memory or I/O Write TLP with 1 DWORD payload by
updating the payload into the target address in the FPGA block RAM space.
130
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
PIO Hardware
Base Address Register Support
The PIO design supports four discrete target spaces, each consisting of a 2 KB block of
memory represented by a separate Base Address Register (BAR). Using the default
parameters, the CORE Generator software produces a core configured to work with the
PIO design defined in this section, consisting of:
•
One 64-bit addressable Memory Space BAR
•
One 32-bit Addressable Memory Space BAR
Users can change the default parameters used by the PIO design; however, in some cases
they might need to change the back-end user application depending on their system. See
Changing CORE Generator Software Default BAR Settings for information about changing
the default CORE Generator software parameters and the effect on the PIO design.
Each of the four 2 KB address spaces represented by the BARs corresponds to one of four
2 KB address regions in the PIO design. Each 2 KB region is implemented using a 2 KB
dual-port block RAM. As transactions are received by the core, the core decodes the
address and determines which of the four regions is being targeted. The core presents the
TLP to the PIO design and asserts the appropriate bits of trn_rbar_hit_n[6:0], as defined in
Table A-1.
Table A-1:
TLP Traffic Types
Block RAM
TLP Transaction Type
Default BAR
trn_rbar_hit_n[6:0]
Disabled
Disabled
ep_io_mem
I/O TLP transactions
ep_mem_32
32-bit address Memory
TLP transactions
2
111_1011b
ep_mem64
64-bit address Memory
TLP transactions
0-1
111_1100b
Exp. ROM
011_1111b
ep_mem_erom
32-bit address Memory
TLP transactions destined
for EROM
Changing CORE Generator Software Default BAR Settings
Users can change the CORE Generator software parameters and continue to use the PIO
design to create customized Verilog or VHDL source to match the selected BAR settings.
However, because the PIO design parameters are more limited than the core parameters,
consider these example design limitations when changing the default CORE Generator
software parameters:
•
The example design supports one I/O space BAR, one 32-bit Memory space (that
cannot be the Expansion ROM space), and one 64-bit Memory space. If these limits are
exceeded, only the first space of a given type is active—accesses to the other spaces do
not result in completions.
•
Each space is implemented with a 2 KB memory. If the corresponding BAR is
configured to a wider aperture, accesses beyond the 2 KB limit wrap around and
overlap the 2 KB memory space.
•
The PIO design supports one I/O space BAR, which by default is disabled, but can be
changed if desired.
Although there are limitations to the PIO design, Verilog or VHDL source code is provided
so the user can tailor the example design to their specific needs.
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
131
Appendix A: Programmed Input/Output Example Design
TLP Data Flow
This section defines the data flow of a TLP successfully processed by the PIO design. For
detailed information about the interface signals within the sub-blocks of the PIO design,
see Receive Path, page 136 and Transmit Path, page 138.
The PIO design successfully processes single DWORD payload Memory Read and Write
TLPs and I/O Read and Write TLPs. Memory Read or Memory Write TLPs of lengths
larger than one DWORD are not processed correctly by the PIO design; however, the core
does accept these TLPs and passes them along to the PIO design. If the PIO design receives
a TLP with a length of greater than 1 DWORD, the TLP is received completely from the
core and discarded. No corresponding completion is generated.
Memory and I/O Write TLP Processing
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 TRN 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 Receive
TRN interface also asserts the appropriate trn_rbar_hit_n[6:0] signal to indicate to the PIO
design the specific destination BAR that matched the incoming TLP. On reception, the PIO
design’s 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’s internal block
RAM write request controller.
Based on the specific trn_rbar_hit_n[6:0] signal 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 asserts trn_rbar_hit_n[0]. 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 trn_rbar_hit_n[0] instructed the PIO memory write
controller to access ep_io_mem (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 the trn_rdst_rdy_n signal, causing the Receive TRN interface to stall receiving
any further TLPs until the internal Memory Write controller completes the write to the
block RAM. Deasserting trn_rdst_rdy_n 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.
Memory and I/O Read TLP Processing
When the Endpoint for PCIe receives a Memory or I/O Read TLP, the TLP destination
address and transaction type are compared with the values programmed in the core BARs.
If the TLP passes this comparison check, the core passes the TLP to the Receive TRN
interface of the PIO design.
Along with the start of packet, end of packet, and ready handshaking signals, the Receive
TRN interface also asserts the appropriate trn_rbar_hit_n[6:0] signal to indicate to the PIO
design the specific destination BAR that matched the incoming TLP. On reception, the PIO
design’s state machine processes the incoming Read TLP and extracts the relevant TLP
132
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
PIO Hardware
information and passes it along to the PIO design's internal block RAM read request
controller.
Based on the specific trn_rbar_hit_n[6:0] signal asserted, the RX state machine indicates to
the internal read request controller the appropriate 2 KB block RAM to use before asserting
the read enable request. For example, if a Memory Read 32 Request TLP is received by the
core targeting the default MEM32 BAR2, the core passes the TLP to the PIO design and
asserts trn_rbar_hit_n[2]. The RX State machine extracts the lower address bits from the
Memory 32 Read TLP and instructs the internal Memory Read Request controller to start a
read operation.
In this example, the assertion of trn_rbar_hit_n[2] instructs the PIO memory read
controller to access the Mem32 space, which by default represents 2 KB of memory space.
A notable difference in handling of memory write and read TLPs is the requirement of the
receiving device to return a Completion with Data TLP in the case of memory or I/O read
request.
While the read is being processed, the PIO design RX state machine deasserts
trn_rdst_rdy_n, causing the Receive TRN interface to stall receiving any further TLPs until
the internal Memory Read controller completes the read access from the block RAM and
generates the completion. Deasserting trn_rst_rdy_n 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.
PIO File Structure
Table A-2 defines the PIO design file structure. Based on the specific core targeted, not all
files delivered by CORE Generator software are necessary, and some files might not be
delivered. The major difference is that some of the Endpoint for PCIe solutions use a 32-bit
user datapath, others use a 64-bit datapath, and the PIO design works with both. The
width of the datapath depends on the specific core being targeted.
Table A-2:
PIO Design File Structure
File
Description
PIO.[v|vhd]
Top-level design wrapper
PIO_EP.[v|vhd]
PIO application module
PIO_TO_CTRL.[v|vhd]
PIO turn-off controller module
PIO_32.v
32b interface macro define
PIO_64.v
64b macro define
PIO_32_RX_ENGINE.[v|vhd]
32b Receive engine
PIO_32_TX_ENGINE.[v|vhd]
32b Transmit engine
PIO_64_RX_ENGINE.[v|vhd]
64b Receive engine
PIO_64_TX_ENGINE.[v|vhd]
64b Transmit engine
PIO_EP_MEM_ACCESS.[v|vhd]
Endpoint memory access module
PIO_EP_MEM.[v|vhd]
Endpoint memory
Two configurations of the PIO Design are provided: PIO_32 and PIO_64, with 32 and 64-bit
TRN interfaces, respectively. The PIO configuration generated depends on the selected
endpoint type (that is, Spartan-6 FPGA integrated Endpoint block, PIPE, PCI Express, and
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
133
Appendix A: Programmed Input/Output Example Design
Block Plus) as well as the number of PCI Express lanes selected by the user. Table A-3
identifies the PIO configuration generated based on the user’s selection.
Table A-3:
PIO Configuration
Core
x1
x2
x4
x8
Endpoint for PIPE
PIO_32
NA
NA
NA
Endpoint for PCI Express (Soft-IP)
PIO_32
NA
PIO_64
PIO_64
Endpoint for PCI Express Block Plus
PIO_64
NA
PIO_64
PIO_64
Spartan-6 FPGA Integrated Endpoint Block
PIO_32
NA
NA
NA
Virtex®-6 FPGA Integrated Block
PIO_64
PIO_64
PIO_64
PIO_64
Figure A-2 shows the various components of the PIO design, which is separated into four
main parts: the TX Engine, RX Engine, Memory Access Controller, and Power
Management Turn-Off Controller.
X-Ref Target - Figure A-2
Spartan-6 FPGA Integrated Endpoint Block for PCI Express
ep_mem0
PIO_TO_CTRL
ep_mem1
EP_TX
EP_RX
ep_mem2
ep_mem3
EP_MEM
PIO_EP
PIO
Figure A-2:
134
PIO Design Components
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
PIO Hardware
PIO Application
Figure A-3 and Figure A-4 depict 64-bit and 32-bit PIO application top-level connectivity,
respectively. The datapath width, either 32 bits or 64 bits, depends on which Endpoint for
PCIe core is used The PIO_EP module contains the PIO FPGA block RAM memory
modules and the transmit and receive engines. The PIO_TO_CTRL module is the Endpoint
Turn-Off controller unit, which responds to power turn-off message from the host CPU
with an acknowledgement.
The PIO_EP module connects to the Endpoint Transaction (TRN) and Configuration (CFG)
interfaces.
X-Ref Target - Figure A-3
Figure A-3:
PIO 64-bit Application
Figure A-4:
PIO 32-bit Application
X-Ref Target - Figure A-4
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
135
Appendix A: Programmed Input/Output Example Design
Receive Path
Figure A-5 illustrates the PIO_32_RX_ENGINE and PIO_64_RX_ENGINE modules. The
datapath of the module must match the datapath of the core being used. These modules
connect with Endpoint for PCIe Transaction Receive (trn_r*) interface.
X-Ref Target - Figure A-5
PIO_32_Rx_Engine
PIO_64_Rx_Engine
trn_rdst_rdy_n
trn_rdst_rdy_n
req_compl_o
req_compl_o
req_td_o
req_td_o
req_ep_o
clk
wr_en_o
rst_n
requ_ep_o
clk
wr_en_o
rst_n
trn_rsof_n
req_tc_o[2:0]
trn_rsof_n
req_tc_o[2:0]
trn_reof_n
req_attr_o[1:0]
trn_reof_n
req_attr_o[1:0]
trn_rsrc_rdy_n
req_len_o[9:0]
trn_rsrc_rdy_n
req_len_o[9:0]
trn_rsrc_dsc_n
req_rid_o[15:0]
trn_rsrc_dsc_n
req_rid_o[15:0]
compl_done_i
req_tag_o[7:0]
compl_done_i
req_tag_o[7:0]
req_be_o[7:0]
wr_busy_i
req_addr_o[31:0]
trn_rd[31:0]
wr_addr_o[10:0]
tr_rbar_hit [6:0]
wr_be_o[7:0]
req_be_o[7:0]
wr_busy_i
req_addr_o[31:0]
trn_rd[63:0]
wr_addr_o[10:0]
trn_rrem_n_[7:0]
wr_be_o[7:0]
tr_rbar_hit [6:0]
wr_data_o[31:0]
wr_data_o[31:0]
EP_Rx
EP_Rx
Figure A-5:
Rx Engines
The PIO_32_RX_ENGINE and PIO_64_RX_ENGINE modules receive and parse incoming
read and write TLPs.
The RX engine parses 1 DWORD 32 and 64-bit addressable memory and I/O read requests.
The RX state machine extracts needed information from the TLP and passes it to the
memory controller, as defined in Table A-4.
Table A-4:
Rx Engine: Read Outputs
Port
136
Description
req_compl_o
Completion request (active High)
req_td_o
Request TLP Digest bit
req_ep_o
Request Error Poisoning bit
req_tc_o[2:0]
Request Traffic Class
req_attr_o[1:0]
Request Attributes
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
PIO Hardware
Table A-4:
Rx Engine: Read Outputs (Cont’d)
Port
Description
req_len_o[9:0]
Request Length
req_rid_o[15:0]
Request Requester Identifier
req_tag_o[7:0]
Request Tag
req_be_o[7:0]
Request Byte Enable
req_addr_o[10:0]
Request Address
The RX Engine parses 1 DWORD 32- and 64-bit addressable memory and I/O write
requests. The RX state machine extracts needed information from the TLP and passes it to
the memory controller, as defined in Table A-5.
Table A-5:
Rx Engine: Write Outputs
Port
Description
wr_en_o
Write received
wr_addr_o[10:0]
Write address
wr_be_o[7:0]
Write byte enable
wr_data_o[31:0]
Write data
The read datapath stops accepting new transactions from the core while the application is
processing the current TLP. This is accomplished by trn_rdst_rdy_n deassertion. For an
ongoing Memory or I/O Read transaction, the module waits for the compl_done_i input to
be asserted before it accepts the next TLP, while an ongoing Memory or I/O Write
transaction is deemed complete after wr_busy_i is deasserted.
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
137
Appendix A: Programmed Input/Output Example Design
Transmit Path
Figure A-6 shows the PIO_32_TX_ENGINE and PIO_64_TX_ENGINE modules. The
datapath of the module must match the datapath of the core being used. These modules
connect with the core Transaction Transmit (trn_r*) interface.
X-Ref Target - Figure A-6
PIO_32_Tx_Engine
PIO_64_Tx_Engine
clk
clk
rst_n
rst_n
trn_tdst_rdy_n
trn_tdst_rdy_n
trn_tdst_dsc_n
trn_tdst_dsc_n
requ_compl_i
trn_tsof_n
requ_compl_i
req_td_i
trn_teof_n
req_td_i
req_ep_i
trn_tsrc_rdy_n
req_ep_i
trn_tsof_n
cfg_bus_mstr_enable_i
trn_tsrc_dsc_n
cfg_bus_mstr_enable_i
trn_teof_n
req_tc_i[2:0]
compl_done_o
req_tc_i[2:0]
trn_tsrc_rdy_n
req_attr_i[1:0]
trn_td[31:0]
req_attr_i[1:0]
trn_tsrc_dsc_n
req_len_i[9:0]
rd_addr_o[10:0]
req_len_i[9:0]
compl_done_o
rd_be_o[3:0]
req_rid_i[15:0]
trn_td[63:0]
req_tag_i[7:0]
req_tag_i[7:0]
trn_trem_n[7:0]
req_be_i[7:0]
req_be_i[7:0]
rd_addr_o[10:0]
req_addr_i[31:0]
req_addr_i[31:0]
rd_data_i[31:0]
rd_data_i[31:0]
completer_id_i[15:0]
completer_id_i[15:0]
req_rid_i[15:0]
EP_Tx
rd_be_o[3:0]
EP_Tx
Figure A-6:
Tx Engines
The PIO_32_TX_ENGINE and PIO_64_TX_ENGINE modules generate completions for
received memory and I/O read TLPs. The PIO design does not generate outbound read or
write requests. However, users can add this functionality to further customize the design.
The PIO_32_TX_ENGINE and PIO_64_TX_ENGINE modules generate completions in
response to 1 DWORD 32 and 64-bit addressable memory and I/O read requests.
Information necessary to generate the completion is passed to the TX Engine, as defined in
Table A-6.
Table A-6:
Tx Engine Inputs
Port
138
Description
req_compl_i
Completion request (active High)
req_td_i
Request TLP Digest bit
req_ep_i
Request Error Poisoning bit
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
PIO Hardware
Table A-6:
Tx Engine Inputs (Cont’d)
Port
Description
req_tc_i[2:0]
Request Traffic Class
req_attr_i[1:0]
Request Attributes
req_len_i[9:0]
Request Length
req_rid_i[15:0]
Request Requester Identifier
req_tag_i[7:0]
Request Tag
req_be_i[7:0]
Request Byte Enable
req_addr_i[10:0]
Request Address
After the completion is sent, the TX engine asserts the compl_done_i output indicating to
the RX engine that it can assert trn_rdst_rdy_n and continue receiving TLPs.
Endpoint Memory
Figure A-7 displays the PIO_EP_MEM_ACCESS module. This module contains the
Endpoint memory space.
X-Ref Target - Figure A-7
PIO_EP_MEM_ACCESS
clk
rst_n
wr_en_i
rd_addr_i[10:0]
wr_busy_o
rd_be_i[3:0]
rd_data_o[31:0]
wr_addr_i[10:0]
wr_be_i[7:0]
wr_data_i[31:0]
EP_MEM
Figure A-7: EP Memory Access
The PIO_EP_MEM_ACCESS module processes data written to the memory from incoming
Memory and I/O Write TLPs and provides data read from the memory in response to
Memory and I/O Read TLPs.
The EP_MEM module processes 1 DWORD 32- and 64-bit addressable Memory and I/O
Write requests based on the information received from the RX Engine, as defined in
Table A-7. While the memory controller is processing the write, it asserts the wr_busy_o
output indicating it is busy.
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
139
Appendix A: Programmed Input/Output Example Design
Table A-7:
EP Memory: Write Inputs
Port
Description
wr_en_i
Write received
wr_addr_i[10:0]
Write address
wr_be_i[7:0]
Write byte enable
wr_data_i[31:0]
Write data
Both 32 and 64-bit Memory and I/O Read requests of one DWORD are processed based on
the inputs defined in Table A-8. After the read request is processed, the data is returned on
rd_data_o[31:0].
Table A-8:
EP Memory: Read Inputs
Port
140
Description
req_be_i[7:0]
Request Byte Enable
req_addr_i[31:0]
Request Address
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
PIO Operation
PIO Operation
PIO Read Transaction
Figure A-8 depicts a Back-To-Back Memory Read request to the PIO design. The receive
engine deasserts trn_rdst_rdy_n as soon as the first TLP is completely received. The
next Read transaction is accepted only after compl_done_o is asserted by the transmit
engine, indicating that Completion for the first request was successfully transmitted.
X-Ref Target - Figure A-8
clk
trn_rd[63:0]
0
trn_rrem_n[7:0]
00
HDR1+HDR2
HDR3+NULL
0F
HDR1+HDR2
HDR3+NULL
00
0F
00
trn_rsof_n
trn_reof_n
trn_rsrc_rdy_n
trn_rsrc_dsc_n
trn_rdst_rdy_n
trn_rbar_hit_n[6:0]
7F
7D
7F
7D
7F
compl_done_o
trn_td_[63:0]
00
trn_trem_n[7:0]
00
HDR1+HDR2
HDR3+DATA
00
trn_tsof_n
trn_teof_n
trn_tsrc_rdy_n
trn_tsrc_dsc_n
trn_tdst_dsc_n
trn_tdst_rdy_n
TLP1
Figure A-8:
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
TLP2
Back-to-Back Read Transactions
www.xilinx.com
141
Appendix A: Programmed Input/Output Example Design
PIO Write Transaction
Figure A-9 depicts a back-to-back Memory Write to the PIO design. The next Write
transaction is accepted only after wr_busy_o is deasserted by the memory access unit,
indicating that data associated with the first request was successfully written to the
memory aperture.
X-Ref Target - Figure A-9
Figure A-9:
Back-to-Back Write Transactions
Device Utilization
Table A-9 shows the PIO design FPGA resource utilization.
Table A-9:
PIO Design FPGA Resources
Resources
Utilization
LUTs
300
Flip-Flops
500
Block RAMs
4
Summary
The PIO design demonstrates the Endpoint for PCIe and its interface capabilities. In
addition, it enables rapid bring-up and basic validation of end user Endpoint add-in card
FPGA hardware on PCI Express platforms. Users can leverage standard operating system
utilities that enable generation of read and write transactions to the target space in the
reference design.
142
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Appendix B
Root Port Model Test Bench
The Endpoint for PCI Express® Root Port Model is a robust test bench environment that
provides a test program interface that can be used with the provided PIO design or with
the user’s design. The purpose of the Root Port Model is to provide a source mechanism
for generating downstream PCI Express TLP traffic to stimulate the customer design, and
a destination mechanism for receiving upstream PCI Express TLP traffic from the customer
design in a simulation environment.
Note: The Root Port Model is shared by the Spartan®-6 FPGA Integrated Endpoint Block, Endpoint
for PCI Express, Endpoint Block Plus for PCI Express, and Endpoint PIPE for PCI Express solutions.
This appendix represents all the solutions generically using the name Endpoint for PCI Express or
Endpoint (or Endpoint for PCIe®).
Source code for the Root Port Model is included to provide the model for a starting point
for the user test bench. All the significant work for initializing the core’s configuration
space, creating TLP transactions, generating TLP logs, and providing an interface for
creating and verifying tests are complete, allowing the user to dedicate efforts to verifying
the correct functionality of the design rather than spending time developing an Endpoint
core test bench infrastructure.
The Root Port Model consists of:
•
Test Programming Interface (TPI), which allows the user to stimulate the Endpoint
device for the PCI Express
•
Example tests that illustrate how to use the test program TPI
•
Verilog or VHDL source code for all Root Port Model components, which allow the
user to customize the test bench
Figure B-1 illustrates the illustrates the Root Port Model coupled with the PIO design.
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
143
Appendix B: Root Port Model Test Bench
X-Ref Target - Figure B-1
Output
Logs
Root Port
Model TPI for
PCI Express
usrapp_com
Test
Program
usrapp_tx
usrapp_rx
dsport
PCI Express Fabric
Endpoint Core for
PCI Express
PIO
Design
Endpoint DUT for PCI Express
Figure B-1:
Root Port Model and Top-Level Endpoint
Architecture
The Root Port Model consists of these blocks, illustrated in Figure B-1.
•
dsport (downstream port)
•
usrapp_tx
•
usrapp_rx
•
usrapp_com (Verilog only)
The usrapp_tx and usrapp_rx blocks interface with the dsport block for transmission and
reception of TLPs to and from the Endpoint Design Under Test (DUT). The Endpoint DUT
consists of the Endpoint for PCIe and the PIO design (displayed) or customer design.
144
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Simulating the Design
The usrapp_tx block sends TLPs to the dsport block for transmission across the PCI
Express Link to the Endpoint DUT. In turn, the Endpoint DUT device transmits TLPs
across the PCI Express Link to the dsport block, which are subsequently passed to the
usrapp_rx block. The dsport and core are responsible for the data link layer and physical
link layer processing when communicating across the PCI Express logic. Both the
usrapp_tx and usrapp_rx utilize the usrapp_com block for shared functions, for example,
TLP processing and log file outputting. Transaction sequences or test programs are
initiated by the usrapp_tx block to stimulate the endpoint device's fabric interface. TLP
responses from the endpoint device are received by the usrapp_rx block. Communication
between the usrapp_tx and usrapp_rx blocks allow the usrapp_tx block to verify correct
behavior and act accordingly when the usrapp_rx block has received TLPs from the
Endpoint device.
The Root Port Model has a 128-byte MPS capability in the receive direction and a 512-byte
MPS capability in the transmit direction.
Simulating the Design
Four simulation script files are provided with the model to facilitate simulation with
Synopsys VCS, Cadence IUS, Mentor Graphics ModelSim, and, Xilinx® ISE® Simulator
(ISim) simulators:
•
simulate_vcs.sh (Verilog only)
•
simulate_ncsim.sh
•
simulate_mti.do
•
simulate_isim.bat/simulate_isim.sh
The example simulation script files are located in this directory:
<project_dir>/<component_name>/simulation/functional
Instructions for simulating the PIO design using the Root Port Model are provided in the
LogiCORE IP Endpoint for PCI Express Getting Started Guide.
For IUS users, the work construct must be manually inserted into the CDS.LIB file:
DEFINE WORK WORK
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
145
Appendix B: Root Port Model Test Bench
Test Selection
Table B-1 describes the tests provided with the Root Port Model, followed by specific
sections for VHDL and Verilog test selection.
Table B-1:
Root Port Model Provided Tests
Test Name
Test in
VHDL/Verilog
Description
sample_smoke_test0
Verilog and
VHDL
Issues a PCI Type 0 Configuration Read TLP and waits for the
completion TLP; then compares the value returned with the
expected Device/Vendor ID value.
sample_smoke_test1
Verilog
Performs the same operation as sample_smoke_test0 but makes
use of expectation tasks. This test uses two separate test program
threads: one thread issues the PCI Type 0 Configuration Read TLP
and the second thread issues the Completion with Data TLP
expectation task. This test illustrates the form for a parallel test that
uses expectation tasks. This test form allows for confirming
reception of any TLPs from the customer’s design. Additionally,
this method can be used to confirm reception of TLPs when
ordering is unimportant.
pio_writeReadBack_test0
Verilog and
VHDL
Transmits a 1 DWORD Write TLP followed by a 1 DWORD Read
TLP to each of the example design’s active BARs, and then waits
for the Completion TLP and verifies that the write and read data
match. The test sends the appropriate TLP to each BAR based on
the BARs address type (for example, 32 bit or 64 bit) and space type
(for example, I/O or Memory).
pio_testByteEnables_test0
Verilog
Issues four sequential Write TLPs enabling a unique byte enable
for each Write TLP, and then issues a 1 DWORD Read TLP to
confirm that the data was correctly written to the example design.
The test sends the appropriate TLP to each BAR based on the BARs
address type (for example, 32 bit or 64 bit) and space type (for
example, I/O or Memory).
pio_memTestDataBus
Verilog
Determines if the PIO design’s FPGA block RAMs data bus
interface is correctly connected by performing a 32-bit walking
ones data test to the first available BAR in the example design.
pio_memTestAddrBus
Verilog
Determines whether the PIO design's FPGA block RAM's
address bus interface is correctly connected by performing a
walking ones address test. This test should only be called after
successful completion of pio_memTestDataBus.
pio_memTestDevice
Verilog
Checks the integrity of each bit of the PIO design’s FPGA block
RAM by performing an increment/decrement test. This test
should only be called after successful completion of
pio_memTestAddrBus.
pio_timeoutFailureExpected
Verilog
Sends a Memory 32 Write TLP followed by Memory 32 Read TLP
to an invalid address and waits for a Completion with data TLP.
This test anticipates that waiting for the completion TLP times out
and illustrates how the test programmer can gracefully handle this
event.
pio_tlp_test0
Verilog
Issues a sequence of Read and Write TLPs to the example design's
RX interface. Some of the TLPs, for example, burst writes, are not
supported by the PIO design.
(illustrative example only)
146
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Test Selection
VHDL Test Selection
Test selection is implemented in the VHDL Root Port Model by means of overriding the
test_selector generic within the tests entity. The test_selector generic is a string with
a one-to-one correspondence to each test within the tests entity.
The user can modify the generic mapping of the instantiation of the tests entity within the
pci_exp_usrapp_tx entity. The default generic mapping is to override the
test_selector with the test name pio_writeReadBack_test0. Currently, there are two
tests defined inside the tests entity, sample_smoke_test0 and
pio_writeReadBack_test0. Additional customer-defined tests should be added inside
tests.vhd. Currently, specific tests cannot be selected from the VHDL simulation scripts.
Verilog Test Selection
The Verilog test model used for the Root Port Model lets the user specify the name of the
test to be run as a command line parameter to the simulator. For example, the
simulate_ncsim.sh script file, used to start the IUS simulator explicitly specifies the
test sample_smoke_test0 to be run using this command line syntax:
ncsim work.boardx01 +TESTNAME=sample_smoke_test0
To change the test to be run, change the value provided to TESTNAME defined in the test
file tests.v. The same mechanism is used for VCS and ModelSim.
ISim uses the -testplusarg options to specify TESTNAME, for example:
demo_tb.exe -gui -view wave.wcfg -wdb wave_isim -tclbatch isim_cmd.tcl
-testplusarg TESTNAME=sample_smoke_test0
VHDL and Verilog Root Port Model Differences
The following subsections identify differences between the VHDL and Verilog Root Port
Model.
Verilog Expectation Tasks
The most significant difference between the Verilog and the VHDL test bench is that the
Verilog test bench has Expectation Tasks. Expectation tasks are API calls used in
conjunction with a bus mastering customer design. The test program issues a series of
expectation task calls; that is, the task calls expect a memory write TLP and a memory read
TLP. If the customer design does not respond with the expected TLPs, the test program
fails. This functionality was implemented using the fork-join construct in Verilog, which is
not available in VHDL and subsequently not implemented.
Verilog Command Line versus VHDL tests.vhd Module
The Verilog test bench allows test programs to be specified at the command line, while the
VHDL test bench specifies test programs within the tests.vhd module.
Generating Wave Files
•
The Verilog test bench uses recordvars and dumpfile commands within the code to
generate wave files.
•
The VHDL test bench leaves the generating wave file functionality up to the
simulator.
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
147
Appendix B: Root Port Model Test Bench
Speed Differences
The VHDL test bench is slower than the Verilog test bench, especially when testing the
x8 core. For initial design simulation and speed enhancement, the user might want to use
the x1 core, identify basic functionality issues, and then move to x4 or x8 simulation when
testing design performance.
Waveform Dumping
Table B-2 describes the available simulator waveform dump file formats, each of which is
provided in the simulator’s native file format. The same mechanism is used for VCS and
ModelSim.
Table B-2:
Simulator Dump File Format
Simulator
Dump File Format
Synopsys VCS
.vpd
ModelSim
.vcd
Cadence IUS
.trn
ISim
.wdb
VHDL Flow
Waveform dumping in the VHDL flow does not use the +dump_all mechanism described
in the Verilog flow section. Because the VHDL language itself does not provide a common
interface for dumping waveforms, each VHDL simulator has its own interface for
supporting waveform dumping. For both the supported ModelSim and IUS flows,
dumping is supported by invoking the VHDL simulator command line with a command
line option that specifies the respective waveform command file, wave.do (ModelSim),
wave.sv (IUS), and wave.wcfg (ISim). This command line can be found in the respective
simulation script files simulate_mti.do, simulate_ncsim.sh, and
simulate_isim.bat[.sh].
ModelSim
This command line initiates waveform dumping for the ModelSim flow using the VHDL
test bench:
>vsim +notimingchecks –do wave.do –L unisim –L work work.board
IUS
This command line initiates waveform dumping for the IUS flow using the VHDL test
bench:
>ncsim –gui work.board -input @”simvision –input wave.sv”
Verilog Flow
The Root Port Model provides a mechanism for outputting the simulation waveform to file
by specifying the +dump_all command line parameter to the simulator.
148
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Output Logging
For example, the script file simulate_ncsim.sh (used to start the IUS simulator) can
indicate to the Root Port Model that the waveform should be saved to a file using this
command line:
ncsim work.boardx01 +TESTNAME=sample_smoke_test0 +dump_all
Output Logging
When a test fails on the example or customer design, the test programmer debugs the
offending test case. Typically, the test programmer inspects the wave file for the simulation
and cross reference this to the messages displayed on the standard output. Because this
approach can be very time consuming, the Root Port Model offers an output logging
mechanism to assist the tester with debugging failing test cases to speed the process.
The Root Port Model creates three output files (tx.dat, rx.dat, and error.dat) during
each simulation run. Log files rx.dat and tx.dat each contain a detailed record of every
TLP that was received and transmitted, respectively, by the Root Port Model. With an
understanding of the expected TLP transmission during a specific test case, the test
programmer can more easily isolate the failure.
The log file error.dat is used in conjunction with the expectation tasks. Test programs
that utilize the expectation tasks generate a general error message to standard output.
Detailed information about the specific comparison failures that have occurred due to the
expectation error is located within error.dat.
Parallel Test Programs
There are two classes of tests are supported by the Root Port Model:
•
Sequential tests. Tests that exist within one process and behave similarly to
sequential programs. The test depicted in Test Program: pio_writeReadBack_test0,
page 152 is an example of a sequential test. Sequential tests are very useful when
verifying behavior that have events with a known order.
•
Parallel tests. Tests involving more than one process thread. The test
sample_smoke_test1 is an example of a parallel test with two process threads.
Parallel tests are very useful when verifying that a specific set of events have
occurred, however the order of these events are not known.
A typical parallel test uses the form of one command thread and one or more expectation
threads. These threads work together to verify a device's functionality. The role of the
command thread is to create the necessary TLP transactions that cause the device to receive
and generate TLPs. The role of the expectation threads is to verify the reception of an
expected TLP. The Root Port Model TPI has a complete set of expectation tasks to be used
in conjunction with parallel tests.
Because the example design is a target-only device, only Completion TLPs can be expected
by parallel test programs while using the PIO design. However, the full library of
expectation tasks can be used for expecting any TLP type when used in conjunction with
the customer's design (which can include bus-mastering functionality). Currently, the
VHDL version of the Root Port Model Test Bench does not support Parallel tests.
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
149
Appendix B: Root Port Model Test Bench
Test Description
The Root Port Model provides a Test Program Interface (TPI). The TPI provides the means
to create tests by simply invoking a series of Verilog tasks. All Root Port Model tests should
follow the same six steps:
1.
Perform conditional comparison of a unique test name
2.
Set up master timeout in case simulation hangs
3.
Wait for Reset and link-up
4.
Initialize the configuration space of the endpoint
5.
Transmit and receive TLPs between the Root Port Model and the Endpoint DUT
6.
Verify that the test succeeded
Test Program: pio_writeReadBack_test0, page 152 displays the listing of a simple test
program pio_writeReadBack_test0, written for use in conjunction with the PIO endpoint.
This test program is located in the file pio_tests.v. As the test name implies, this test
performs a one DWORD write operation to the PIO Design followed by a 1 DWORD read
operation from the PIO Design, after which it compares the values to confirm that they are
equal. The test is performed on the first location in each of the active Mem32 BARs of the
PIO Design. For the default configuration, this test performs the write and read back to
BAR2 and to the EROM space (BAR6) (Block Plus only). The following section outlines the
steps performed by the test program.
•
Line 1 of the sample program determines if the user has selected the test program
pio_writeReadBack_test1 when invoking the Verilog simulator.
•
Line 4 of the sample program invokes the TPI call TSK_SIMULATION_TIMEOUT
which sets the master timeout value to be long enough for the test to complete.
•
Line 5 of the sample program invokes the TPI call TSK_SYSTEM_INITIALIZATION.
This task causes the test program to wait for the system reset to deassert as well as the
Endpoint's trn_lnk_up_n signal to assert. This is an indication that the Endpoint is
ready to be configured by the test program via the Root Port Model.
•
Line 6 of the sample program uses the TPI call TSK_BAR_INIT. This task performs a
series of Type 0 Configuration Writes and Reads to the Endpoint core's PCI
Configuration Space, determines the memory and I/O requirements of the Endpoint,
and then programs the Endpoint's Base Address Registers so that it is ready to receive
TLPs from the Root Port Model.
•
Lines 7, 8, and 9 of the sample program work together to cycle through all the
Endpoint's BARs and determine whether they are enabled, and if so to determine
their type, for example, Mem32, Mem64, or I/O).
All PIO tests provided with the Root Port Model are written in a form that does not assume
that a specific BAR is enabled or is of a specific type (for example, Mem32, Mem64, I/O).
These tests perform on-the-fly BAR determination and execute TLP transactions
dependent on BAR types (that is, Memory32 TLPs to Memory32 Space, I/O TLPs to I/O
Space, and so forth). This means that if a user reconfigures the BARs of the Endpoint, the
PIO continues to work because it dynamically explores and configures the BARs. Users are
not required to follow the form used and can create tests that assume their own specific
BAR configuration.
150
•
Line 7 sets a counter to increment through all of the endpoint's BARs.
•
Line 8 determines whether the BAR is enabled by checking the global array
BAR_INIT_P_BAR_ENABLED[]. A non-zero value indicates that the corresponding
BAR is enabled. If the BAR is not enabled, then test program flow moves on to check
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Test Description
the next BAR. The previous call to TSK_BAR_INIT performed the necessary
configuration TLP communication to the endpoint device and filled in the
appropriate values into the BAR_INIT_P_BAR_ENABLED[] array.
•
Line 9 performs a case statement on the same global array
BAR_INIT_P_BAR_ENABLED[]. If the array element is enabled (that is, non-zero),
the element's value indicates the BAR type. A value of 1, 2, and 3 indicates I/O,
Memory 32, and Memory 64 spaces, respectively.
If the BAR type is either I/O or Memory 64, then the test does not perform any TLP
transactions. If the BAR type is Memory 32, program control continues to line 16 and
starts transmitting Memory 32 TLPs.
•
Lines 21-26 use the TPI call TSK_TX_MEMORY_WRITE_32 and transmits a Memory
32 Write TLP with the payload DWORD '01020304' to the PIO endpoint.
•
Lines 32-33 use the TPI calls TSK_TX_MEMORY_READ_32 followed by
TSK_WAIT_FOR_READ_DATA in order to transmit a Memory 32 Read TLP and then
wait for the next Memory 32 Completion with Data TLP. In case the Root Port Model
never receives the Completion with Data TLP, the TPI call
TSK_WAIT_FOR_READ_DATA would locally timeout and display an error message.
•
Line 34 compares the DWORD received from the Completion with Data TLP with the
DWORD that was transmitted to the PIO endpoint and displays the appropriate
success or failure message.
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
151
Appendix B: Root Port Model Test Bench
Test Program: pio_writeReadBack_test0
1.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
16.
17.
18.
19.
20.
21.
22.
23.
24.
25.
26.
27.
28.
29.
30.
31.
32.
33.
34.
35.
36.
37.
38.
39.
40.
41.
42.
43.
44.
45.
46.
47.
48.
49.
50.
51.
52.
53.
else if(testname == "pio_writeReadBack_test1")
begin
// This test performs a 32 bit write to a 32 bit Memory space and performs a read back
TSK_SIMULATION_TIMEOUT(10050);
TSK_SYSTEM_INITIALIZATION;
TSK_BAR_INIT;
for (ii = 0; ii <= 6; ii = ii + 1) begin
if (BAR_INIT_P_BAR_ENABLED[ii] > 2'b00) // bar is enabled
case(BAR_INIT_P_BAR_ENABLED[ii])
2'b01 : // IO SPACE
begin
$display("[%t] : NOTHING: to IO 32 Space BAR %x", $realtime, ii);
end
2'b10 : // MEM 32 SPACE
begin
$display("[%t] : Transmitting TLPs to Memory 32 Space BAR %x",
$realtime, ii);
//-----------------------------------------------------------------------// Event : Memory Write 32 bit TLP
//-----------------------------------------------------------------------DATA_STORE[0] = 8'h04;
DATA_STORE[1] = 8'h03;
DATA_STORE[2] = 8'h02;
DATA_STORE[3] = 8'h01;
P_READ_DATA = 32'hffff_ffff; // make sure P_READ_DATA has known initial value
TSK_TX_MEMORY_WRITE_32(DEFAULT_TAG, DEFAULT_TC, 10'd1, BAR_INIT_P_BAR[ii][31:0] , 4'hF,
4'hF, 1'b0);
TSK_TX_CLK_EAT(10);
DEFAULT_TAG = DEFAULT_TAG + 1;
//-----------------------------------------------------------------------// Event : Memory Read 32 bit TLP
//-----------------------------------------------------------------------TSK_TX_MEMORY_READ_32(DEFAULT_TAG, DEFAULT_TC, 10'd1, BAR_INIT_P_BAR[ii][31:0], 4'hF,
4'hF);
TSK_WAIT_FOR_READ_DATA;
if (P_READ_DATA != {DATA_STORE[3], DATA_STORE[2], DATA_STORE[1], DATA_STORE[0] })
begin
$display("[%t] : Test FAILED --- Data Error Mismatch, Write Data %x != Read Data %x",
$realtime,{DATA_STORE[3], DATA_STORE[2], DATA_STORE[1], DATA_STORE[0]}, P_READ_DATA);
end
else
begin
$display("[%t] : Test PASSED --- Write Data: %x successfully received", $realtime,
P_READ_DATA);
end
TSK_TX_CLK_EAT(10);
DEFAULT_TAG = DEFAULT_TAG + 1;
end
2'b11 : // MEM 64 SPACE
begin
$display("[%t] : NOTHING: to Memory 64 Space BAR %x", $realtime, ii);
end
default : $display("Error case in usrapp_tx\n");
endcase
end
$display("[%t] : Finished transmission of PCI-Express TLPs", $realtime);
$finish;
54.end
152
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Expanding the Root Port Model
Expanding the Root Port Model
The Root Port Model was created to work with the PIO design, and for this reason is
tailored to make specific checks and warnings based on the limitations of the PIO design.
These checks and warnings are enabled by default when the Root Port Model is generated
by the CORE Generator™ software. However, these limitations can easily be disabled so
that they do not affect the customer's design.
Because the PIO design was created to support at most one I/O BAR, one Mem64 BAR,
and two Mem32 BARs (one of which must be the EROM space), the Root Port Model by
default makes a check during device configuration that verifies that the core has been
configured to meet this requirement. A violation of this check causes a warning message to
be displayed as well as for the offending BAR to be gracefully disabled in the test bench.
This check can be disabled by setting the pio_check_design variable to zero in the
pci_exp_usrapp_tx.v file.
Root Port Model TPI Task List
The Root Port Model TPI tasks include the following, which are further defined in
Tables B-3 through B-7.
Table B-3:
•
Test Setup Tasks
•
TLP Tasks
•
BAR Initialization Tasks
•
Example PIO Design Tasks
•
Expectation Tasks
Test Setup Tasks
Name
TSK_SYSTEM_INITIALIZATION
Input(s)
Description
None
Waits for transaction interface reset and
link-up between the Root Port Model and
the Endpoint DUT.
This task must be invoked prior to the
Endpoint core initialization.
TSK_USR_DATA_SETUP_SEQ
None
TSK_TX_CLK_EAT
clock count
31:30
Waits clock_count transaction interface
clocks.
TSK_SIMULATION_TIMEOUT
timeout
31:0
Sets master simulation timeout value in
units of transaction interface clocks. This
task should be used to ensure that all DUT
tests complete.
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
Initializes global 4096 byte DATA_STORE
array entries to sequential values from
zero to 4095.
153
Appendix B: Root Port Model Test Bench
Table B-4:
TLP Tasks
Name
TSK_TX_TYPE0_CONFIGURATION_READ
TSK_TX_TYPE1_CONFIGURATION_READ
Input(s)
Description
tag_
7:0
reg_addr_
11:0
first_dw_be_
3:0
tag_
7:0
reg_addr_
11:0
first_dw_be_
3:0
Waits for transaction interface reset and linkup between the Root Port Model and the
Endpoint DUT.
This task must be invoked prior to Endpoint
core initialization.
Sends a Type 1 PCI Express Config Read TLP
from Root Port Model to reg_addr_ of
Endpoint DUT with tag_ and first_dw_be_
inputs.
CplD returned from the Endpoint DUT uses
the contents of global COMPLETE_ID_CFG as
the completion ID.
TSK_TX_TYPE0_CONFIGURATION_WRITE
TSK_TX_TYPE1_CONFIGURATION_WRITE
TSK_TX_MEMORY_READ_32
TSK_TX_MEMORY_READ_64
154
tag_
7:0
reg_addr_
11:0
reg_data_
31:0
first_dw_be_
3:0
tag_
7:0
reg_addr_
11:0
reg_data_
31:0
first_dw_be_
3:0
tag_
7:0
tc_
2:0
len_
9:0
addr_
31:0
last_dw_be_
3:0
first_dw_be_
3:0
tag_
7:0
tc_
2:0
len_
9:0
addr_
63:0
last_dw_be_
3:0
first_dw_be_
3:0
www.xilinx.com
Sends a Type 0 PCI Express Config Write TLP
from Root Port Model to reg_addr_ of
Endpoint DUT with tag_ and first_dw_be_
inputs.
Cpl returned from the Endpoint DUT uses the
contents of global COMPLETE_ID_CFG as the
completion ID.
Sends a Type 1 PCI Express Config Write TLP
from Root Port Model to reg_addr_ of
Endpoint DUT with tag_ and first_dw_be_
inputs.
Cpl returned from the Endpoint DUT uses the
contents of global COMPLETE_ID_CFG as the
completion ID.
Sends a PCI Express Memory Read TLP from
downstream port to 32 bit memory address
addr_ of Endpoint DUT.
CplD returned from the Endpoint DUT uses
the contents of global COMPLETE_ID_CFG as
the completion ID.
Sends a PCI Express Memory Read TLP from
Root Port Model to 64 bit memory address
addr_ of Endpoint DUT.
CplD returned from the Endpoint DUT uses
the contents of global COMPLETE_ID_CFG as
the completion ID.
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Expanding the Root Port Model
Table B-4:
TLP Tasks (Cont’d)
Name
TSK_TX_MEMORY_WRITE_32
TSK_TX_MEMORY_WRITE_64
TSK_TX_COMPLETION
TSK_TX_COMPLETION_DATA
TSK_TX_MESSAGE
Input(s)
Description
tag_
7:0
tc_
2:0
len_
9:0
addr_
31:0
last_dw_be_
3:0
first_dw_be_
3:0
ep_
–
tag_
7:0
tc_
2:0
len_
9:0
addr_
63:0
last_dw_be_
3:0
first_dw_be_
3:0
ep_
–
tag_
7:0
tc_
2:0
len_
9:0
comp_status_
2:0
tag_
7:0
tc_
2:0
len_
9:0
byte_count
11:0
lower_addr
6:0
comp_status
2:0
ep_
–
tag_
7:0
tc_
2:0
len_
9:0
data
63:0
message_rtg
2:0
Sends a PCI Express Memory Write TLP from
Root Port Model to 32 bit memory address
addr_ of Endpoint DUT.
CplD returned from the Endpoint DUT uses
the contents of global COMPLETE_ID_CFG as
the completion ID.
The global DATA_STORE byte array is used to
pass write data to task.
Sends a PCI Express Memory Write TLP from
Root Port Model to 64 bit memory address
addr_ of Endpoint DUT.
CplD returned from the Endpoint DUT uses
the contents of global COMPLETE_ID_CFG as
the completion ID.
The global DATA_STORE byte array is used to
pass write data to task.
Sends a PCI Express Completion TLP from
Root Port Model to the Endpoint DUT using
global COMPLETE_ID_CFG as the completion
ID.
Sends a PCI Express Completion with Data
TLP from Root Port Model to the Endpoint
DUT using global COMPLETE_ID_CFG as the
completion ID.
The global DATA_STORE byte array is used to
pass completion data to task.
Sends a PCI Express Message TLP from Root
Port Model to Endpoint DUT.
Completion returned from the Endpoint DUT
uses the contents of global
COMPLETE_ID_CFG as the completion ID.
message_code 7:0
TSK_TX_MESSAGE_DATA
tag_
7:0
tc_
2:0
len_
9:0
data
63:0
message_rtg
2:0
message_code 7:0
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
Sends a PCI Express Message with Data TLP
from Root Port Model to Endpoint DUT.
The global DATA_STORE byte array is used to
pass message data to task.
Completion returned from the Endpoint DUT
uses the contents of global
COMPLETE_ID_CFG as the completion ID.
155
Appendix B: Root Port Model Test Bench
Table B-4:
TLP Tasks (Cont’d)
Name
TSK_TX_IO_READ
TSK_TX_IO_WRITE
TSK_TX_BAR_READ
Input(s)
Description
tag_
7:0
addr_
31:0
first_dw_be_
3:0
tag_
7:0
addr_
31:0
first_dw_be_
3:0
data
31:0
bar_index
2:0
byte_offset
31:0
tag_
7:0
tc_
2:0
Sends a PCI Express I/O Read TLP from Root
Port Model to I/O address addr_[31:2] of the
Endpoint DUT.
CplD returned from the Endpoint DUT uses
the contents of global COMPLETE_ID_CFG as
the completion ID.
Sends a PCI Express I/O Write TLP from Root
Port Model to I/O address addr_[31:2] of the
Endpoint DUT.
CplD returned from the Endpoint DUT uses
the contents of global COMPLETE_ID_CFG as
the completion ID.
Sends a PCI Express 1 DWORD Memory 32,
Memory 64, or I/O Read TLP from the Root
Port Model to the target address
corresponding to offset byte_offset from BAR
bar_index of the Endpoint DUT. This task
sends the appropriate Read TLP based on how
BAR bar_index has been configured during
initialization. This task can only be called after
TSK_BAR_INIT has successfully completed.
CplD returned from the Endpoint DUT uses
the contents of global COMPLETE_ID_CFG as
the completion ID.
156
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Expanding the Root Port Model
Table B-4:
TLP Tasks (Cont’d)
Name
TSK_TX_BAR_WRITE
TSK_WAIT_FOR_READ_DATA
Input(s)
Description
bar_index
2:0
byte_offset
31:0
tag_
7:0
tc_
2:0
data_
31:0
None
Sends a PCI Express 1 DWORD Memory 32,
Memory 64, or I/O Write TLP from the Root
Port to the target address corresponding to
offset byte_offset from BAR bar_index of the
Endpoint DUT.
This task sends the appropriate Write TLP
based on how BAR bar_index has been
configured during initialization. This task can
only be called after TSK_BAR_INIT has
successfully completed.
Waits for the next completion with data TLP
that was sent by the Endpoint DUT. On
successful completion, the first DWORD of
data from the CplD is stored in the global
P_READ_DATA. This task should be called
immediately following any of the read tasks in
the TPI that request Completion with Data
TLPs to avoid any race conditions.
By default this task will locally time out and
terminate the simulation after 1000 transaction
interface clocks. The global cpld_to_finish can
be set to zero so that local time out returns
execution to the calling test and does not result
in simulation timeout. For this case test
programs should check the global cpld_to,
which when set to one indicates that this task
has timed out and that the contents of
P_READ_DATA are invalid.
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
157
Appendix B: Root Port Model Test Bench
Table B-5:
BAR Initialization Tasks
Name
TSK_BAR_INIT
Input(s)
None
Description
Performs a standard sequence of Base Address
Register initialization tasks to the Endpoint
device using the PCI Express fabric. Performs a
scan of the Endpoint's PCI BAR range
requirements, performs the necessary memory
and I/O space mapping calculations, and finally
programs the Endpoint so that it is ready to be
accessed.
On completion, the user test program can begin
memory and I/O transactions to the device. This
function displays to standard output a memory
and I/O table that details how the Endpoint has
been initialized. This task also initializes global
variables within the Root Port Model that are
available for test program usage. This task should
only be called after
TSK_SYSTEM_INITIALIZATION.
TSK_BAR_SCAN
None
Performs a sequence of PCI Type 0 Configuration
Writes and Configuration Reads using the PCI
Express fabric to determine the memory and I/O
requirements for the Endpoint.
The task stores this information in the global array
BAR_INIT_P_BAR_RANGE[]. This task should
only be called after
TSK_SYSTEM_INITIALIZATION.
TSK_BUILD_PCIE_MAP
None
Performs memory and I/O mapping algorithm
and allocates Memory 32, Memory 64, and I/O
space based on the Endpoint requirements.
This task has been customized to work in
conjunction with the limitations of the PIO design
and should only be called after completion of
TSK_BAR_SCAN.
TSK_DISPLAY_PCIE_MAP
158
None
Displays the memory mapping information of the
Endpoint core’s PCI Base Address Registers. For
each BAR, the BAR value, the BAR range, and
BAR type is given. This task should only be called
after completion of TSK_BUILD_PCIE_MAP.
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Expanding the Root Port Model
Table B-6:
Example PIO Design Tasks
Name
TSK_TX_READBACK_CONFIG
Input(s)
Description
None
Performs a sequence of PCI Type 0 Configuration
Reads to the Endpoint device's Base Address
Registers, PCI Command Register, and PCIe
Device Control Register using the PCI Express
fabric.
This task should only be called after
TSK_SYSTEM_INITIALIZATION.
TSK_MEM_TEST_DATA_BUS
bar_index
2:0
Tests whether the PIO design FPGA block RAM
data bus interface is correctly connected by
performing a 32-bit walking ones data test to the
I/O or memory address pointed to by the input
bar_index.
For an exhaustive test, this task should be called
four times, once for each block RAM used in the
PIO design.
TSK_MEM_TEST_ADDR_BUS
bar_index
2:0
nBytes
31:0
Tests whether the PIO design FPGA block RAM
address bus interface is accurately connected by
performing a walking ones address test starting at
the I/O or memory address pointed to by the
input bar_index.
For an exhaustive test, this task should be called
four times, once for each block RAM used in the
PIO design. Additionally, the nBytes input should
specify the entire size of the individual block
RAM.
TSK_MEM_TEST_DEVICE
bar_index
2:0
nBytes
31:0
Tests the integrity of each bit of the PIO design
FPGA block RAM by performing an
increment/decrement test on all bits starting at
the block RAM pointed to by the input bar_index
with the range specified by input nBytes.
For an exhaustive test, this task should be called
four times, once for each block RAM used in the
PIO design. Additionally, the nBytes input should
specify the entire size of the individual block
RAM.
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
159
Appendix B: Root Port Model Test Bench
Table B-7:
Expectation Tasks
Name
TSK_EXPECT_CPLD
Input(s)
2:0
traffic_class
td
-
ep
-
attr
1:0
length
9:0
completer_id
15:0
completer_status
2:0
160
expect status
Waits for a Completion with Data
TLP that matches traffic_class, td,
ep, attr, length, and payload.
Returns a 1 on successful
completion; 0 otherwise.
byte_count
11:0
requester_id
15:0
tag
7:0
address_low
6:0
traffic_class
2:0
Expect
td
-
status
ep
-
attr
1:0
completer_id
15:0
completer_status
2:0
Waits for a Completion without
Data TLP that matches
traffic_class, td, ep, attr, and
length.
Returns a 1 on successful
completion; 0 otherwise.
-
bcm
TSK_EXPECT_MEMRD
Description
-
bcm
TSK_EXPECT_CPL
Output
byte_count
11:0
requester_id
15:0
tag
7:0
address_low
6:0
traffic_class
2:0
td
-
ep
-
attr
1:0
length
9:0
requester_id
15:0
tag
7:0
last_dw_be
3:0
first_dw_be
3:0
address
29:0
www.xilinx.com
Expect status
Waits for a 32-bit Address
Memory Read TLP with matching
header fields.
Returns a 1 on successful
completion; 0 otherwise. This task
can only be used in conjunction
with Bus Master designs.
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Expanding the Root Port Model
Table B-7:
Expectation Tasks (Cont’d)
Name
TSK_EXPECT_MEMRD64
TSK_EXPECT_MEMWR
TSK_EXPECT_MEMWR64
TSK_EXPECT_IOWR
Input(s)
Output
Description
Expect status
Waits for a 64-bit Address
Memory Read TLP with matching
header fields. Returns a 1 on
successful completion; 0
otherwise.
traffic_class
2:0
td
-
ep
-
attr
1:0
length
9:0
requester_id
15:0
tag
7:0
last_dw_be
3:0
first_dw_be
3:0
address
61:0
traffic_class
2:0
Expect
td
-
status
ep
-
attr
1:0
length
9:0
requester_id
15:0
tag
7:0
last_dw_be
3:0
first_dw_be
3:0
address
29:0
traffic_class
2:0
Expect
td
-
status
ep
-
attr
1:0
length
9:0
requester_id
15:0
tag
7:0
last_dw_be
3:0
first_dw_be
3:0
address
61:0
td
-
Expect
ep
-
status
requester_id
15:0
tag
7:0
first_dw_be
3:0
address
31:0
data
31:0
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
This task can only be used in
conjunction with Bus Master
designs.
Waits for a 32 bit Address Memory
Write TLP with matching header
fields. Returns a 1 on successful
completion; 0 otherwise.
This task can only be used in
conjunction with Bus Master
designs.
Waits for a 64 bit Address Memory
Write TLP with matching header
fields. Returns a 1 on successful
completion; 0 otherwise.
This task can only be used in
conjunction with Bus Master
designs.
Waits for an I/O Write TLP with
matching header fields. Returns a
1 on successful completion; 0
otherwise.
This task can only be used in
conjunction with Bus Master
designs.
161
Appendix B: Root Port Model Test Bench
162
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Appendix C
Migration Considerations
Migrating a design from a LogiCORE™ Endpoint PIPE for PCI Express to a
Spartan®-6 FPGA Integrated Endpoint Block for PCI Express is a simple process. This
appendix describes the changes in the interfaces and signals that are necessary when
migrating from the Spartan-3 FPGA Endpoint PIPE core to a Spartan-6 FPGA integrated
Endpoint block core.
Integrated PHY
The first and most notable change between the Spartan-3 FPGA Endpoint PIPE core and
the Spartan-6 FPGA integrated Endpoint block core is that the external SerDes PHY device
that previously resided outside of the device has been integrated into the Spartan-6
architecture. This means that all of the PXPIPE signals (33 ports) are replaced with the
serial interface (4 ports).
System Clocking and Reset
For the Spartan-6 FPGA integrated Endpoint block, the sys_clk signal was added.
In the Spartan-3 FPGA design, the system clock was provided by the external PHY on the
port, RXCLK. For more information about sys_clk and how to set it up, see Clocking in
Chapter 6.
Interface Changes
Streaming Signal Added
The trn_tstr_n signal was added to the integrated Endpoint block to allow packets to be
streamed in the transmit direction. For more information on trn_tstr_n, see Table 2-6,
page 22.
TRN Transmit Destination Discontinue Removed
The trn_tdst_dsc_n signal has been replaced with trn_tx_terr_drop_n. The signal serves
the same purpose to signal when a packet has been dropped. However, the signal now is
asserted 1 to 2 clock cycles after end of the packet that was dropped. The User Application
is not required to do anything in response to trn_terr_drop_n; it is intended for diagnosing
problems when bringing up new designs. This makes timing significantly easier to meet.
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
163
Appendix C: Migration Considerations
TRN Buffer Available Size Change
The Spartan-3 FPGA Endpoint PIPE signal trn_tbuf_av[4:0] is one bit wider. The signal is
now trn_tbuf_av[5:0]. This change reflects the increased number of transmit buffers
supported by the Spartan-6 FPGA core.
CMM Arbitration
The Spartan-6 FPGA integrated Endpoint block design has two signals that allow for the
user to control the arbitration between the CMM and the TRN interfaces for transmitted
packets. These signals are trn_tcfg_req_n and trn_tcfg_gnt_n.
To maintain the same behavior as the Spartan-3 FPGA Endpoint PIPE, assign the signal
trn_tcfg_gnt_n asserted (1'b0).
TRN Credit Buses Additional Functionality
The TRN credit buses have different names in the Spartan-6 FPGA integrated Endpoint
block design. The letter “r” has been removed from the signal names. Table C-1 shows the
old and new names.
Table C-1:
Credit Bus Name Change from Spartan-3 to Spartan-6 Devices
Spartan-3 FPGAs
Spartan-6 FPGAs
trn_rfc_nph
trn_fc_nph
trn_rfc_npd
trn_fc_npd
trn_rfc_ph
trn_fc_ph
trn_rfc_pd
trn_fc_pd
trn_rfc_cplh
trn_fc_cplh
trn_rfc_cpld
trn_fc_cpld
There is also a signal named trn_fc_sel[2:0] that controls what values are placed on the
trn_* bus.
To maintain the same behavior as the Spartan-3 FPGA Endpoint PIPE, set the
trn_fc_sel[2:0] signal to 3'b000. For more information, see Flow Control Credit
Information in Chapter 6.
Configuration Error Completion Ready
A signal named cfg_err_cpl_rdy_n has been added for the Spartan-6 FPGA integrated
Endpoint block. For more information, see Table 2-9, page 28.
Configuration Error Locked
A signal named cfg_err_locked_n has been added for the Spartan-6 FPGA integrated
Endpoint block. For more information, see Table 2-9, page 28.
164
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Block RAM Settings
Removed Configuration Signals
These signals were removed because they were either unused or not needed:
•
cfg_di
•
cfg_wr_en_n
•
cfg_byte_en_n
•
cfg_err_cpl_unexpected_n
Hot Reset
A signal named received_hot_reset has been added for the Spartan-6 FPGA integrated
Endpoint block. For more information, see Table 2-3, page 20.
Block RAM Settings
The block RAM settings can now be customized. See the CORE Generator™ software GUI
for supported settings.
Signal Change Summary
These signals have been added for the Spartan-6 FPGA integrated Endpoint block:
•
sys_clk
•
trn_tstr_n
•
trn_tcfg_req_n
•
trn_tcfg_gnt_n
•
cfg_err_cpl_rdy_n
•
cfg_err_locked_n
•
received_hot_reset
•
trn_fc_sel[2:0]
These signals have been removed from the Spartan-6 FPGA integrated Endpoint block:
•
cfg_di
•
cfg_wr_en_n
•
cfg_byte_en_n
•
cfg_err_cpl_unexpected_n
Table C-2 shows the signal name changes.
Table C-2:
Signal Name Change from Spartan-3 to Spartan-6 Devices
Spartan-3 FPGAs
Spartan-6 FPGAs
trn_rfc_nph
trn_fc_nph
trn_rfc_npd
trn_fc_npd
trn_rfc_ph
trn_fc_ph
trn_rfc_pd
trn_fc_pd
trn_rfc_cplh
trn_fc_cplh
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
165
Appendix C: Migration Considerations
Table C-2:
Signal Name Change from Spartan-3 to Spartan-6 Devices (Cont’d)
Spartan-3 FPGAs
166
Spartan-6 FPGAs
trn_rfc_cpld
trn_fc_cpld
trn_tbuf_av[4:0]
trn_tbuf_av[5:0]
trn_tdst_dsn_n
trn_tx_terr_drop_n
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Appendix D
Debugging Designs
This appendix provides information on using resources available on the Xilinx Support
website, available debug tools, and a step-by-step process for debugging designs that use
the Spartan®-6 Integrated Endpoint Block for PCI Express®. This appendix uses flow
diagrams to guide the user through the debug process.
The following information is found in this appendix:
•
Finding Help on Xilinx.com
•
Contacting Xilinx Technical Support
•
Debug Tools
•
Hardware Debug
•
Simulation Debug
Finding Help on Xilinx.com
To help in the design and debug process when using the Integrated Endpoint Block for PCI
Express, the Xilinx Support webpage (www.xilinx.com/support) contains key resources
such as Product documentation, Release Notes, Answer Records, and links to opening a
Technical Support case.
Documentation
The Data Sheet and User Guide are the main documents associated with the
Spartan-6 FPGA integrated Endpoint block, as shown in Table D-1.
Table D-1: Spartan-6 FPGA Integrated Endpoint Block for PCI Express
Documentation
Designation
Description
DS
Data Sheet: provides a high-level description of the integrated Endpoint
block and key features. It includes information on which ISE software
version is supported by the current LogiCORE™ IP version used to
instantiate the integrated Endpoint block.
UG
User Guide: provides information on generating an integrated Endpoint
block design, detailed descriptions of the interface and how to use the
product. The User Guide contains waveforms to show interactions with the
block and other important information needed to design with the product.
These Integrated Endpoint Block for PCI Express documents along with documentation
related to all products that aid in the design process can be found on the Xilinx Support
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
167
Appendix D: Debugging Designs
webpage. Documentation is sorted by product family at the main support page or by
solution at the Documentation Center.
To see the available documentation by device family:
•
Navigate to www.xilinx.com/support.
•
Select Spartan-6 from the Device List drop-down menu.
•
This will sort all available Spartan-6 FPGA documentation by Hardware
Documentation, Configuration Solutions Documentation, Related Software
Documentation, Tools, IP, and Data Files.
To see the available documentation by solution:
•
Navigate to www.xilinx.com/support.
•
Select the Documentation tab located at the top of the webpage.
•
This is the Documentation Center where Xilinx documentation is sorted by Devices,
Boards, IP, Design Tools, Doc Type, and Topic.
Release Notes and Known Issues
Known issues for all cores, including the Spartan-6 FPGA Integrated Endpoint Block for
PCI Express, are described in the IP Release Notes Guide.
Answer Records
Answer Records include information on commonly encountered problems, helpful
information on how to resolve these problems, and any known issues with a product.
Answer Records are created and maintained daily ensuring users have access to the most
up-to-date information on Xilinx products. Answer Records can be found by searching the
Answers Database.
To use the Answers Database Search:
168
•
Navigate to www.xilinx.com/support. The Answers Database Search is located at the
top of this webpage.
•
Enter keywords in the provided search field and select Search.
•
Examples of searchable keywords are product names, error messages, or a generic
summary of the issue encountered.
•
To see all answer records directly related to the Spartan-6 FPGA Integrated
Endpoint Block for PCI Express, search for the phrase Spartan-6 Integrated
Endpoint Block for PCI Express.
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Contacting Xilinx Technical Support
Contacting Xilinx Technical Support
Xilinx provides premier technical support for customers encountering issues that require
additional assistance.
To contact Technical Support:
•
Navigate to www.xilinx.com/support.
•
Open a WebCase by selecting the WebCase link located under Support Quick Links.
When opening a WebCase, include:
•
Target FPGA including package and speed grade
•
All applicable software versions of the ISE tool, synthesis (if not XST), and simulator
•
The xco file created during generation of the LogiCORE IP wrapper
•
This file is located in the directory targeted for the CORE Generator™ software
project
Additional files might be required based on the specific issue. See the relevant sections in
this debug guide for further information on specific files to include with the WebCase.
Debug Tools
There are many tools available to debug PCI Express design issues. It is important to know
which tools would be useful for debugging for the various situations encountered. This
chapter references these tools:
Example Design
Xilinx Endpoint for PCI Express products come with a synthesizable back-end application
called the PIO design that has been tested and is proven to be interoperable in available
systems. The design appropriately handles all incoming 1 DWORD read and write
transactions. It returns completions for non-posted transactions and updates the target
memory space for writes. For more information, see Appendix A, Programmed
Input/Output Example Design.
ChipScope Pro Tool
The ChipScope™ Pro tool inserts logic analyzer, bus analyzer, and virtual I/O software
cores directly into the user design. The ChipScope Pro tool allows the user to set trigger
conditions to capture application and integrated Endpoint block port signals in hardware.
Captured signals can then be analyzed through the ChipScope Pro Logic Analyzer tool.
For detailed information on the ChipScope Pro tool, visit www.xilinx.com/chipscope.
Link Analyzers
Third-party link analyzers show link traffic in a graphical or text format. Lecroy, Agilent,
and Vmetro are companies that make common analyzers available today. These tools
greatly assist in debugging link issues and allow users to capture data which Xilinx
support representatives can view to assist in interpreting link behavior.
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
169
Appendix D: Debugging Designs
Third-Party Software Tools
This section describes third-party software tools that can be useful in debugging.
LSPCI (Linux)
LSPCI is available on Linux platforms and allows users to view the PCI Express device
configuration space. LSPCI is usually found in the /sbin directory. LSPCI displays a list of
devices on the PCI buses in the system. See the LSPCI manual for all command options.
Some useful commands for debugging include:
•
lspci -x -d [<vendor>]:[<device>]
This displays the first 64 bytes of configuration space in hexadecimal form for the
device with vendor and device ID specified (omit the -d option to display information
for all devices). The default Vendor/Device ID for Xilinx cores is 10EE:6012. Here is a
sample of a read of the configuration space of a Xilinx device:
> lspci -x -d 10EE:6012
81:00.0 Memory controller: Xilinx
00: ee 10 12 60 07 00 10 00 00 00
10: 00 00 80 fa 00 00 00 00 00 00
20: 00 00 00 00 00 00 00 00 00 00
30: 00 00 00 00 40 00 00 00 00 00
Corporation: Unknown device 6012
80 05 10 00 00 00
00 00 00 00 00 00
00 00 ee 10 6f 50
00 00 05 01 00 00
Included in this section of the configuration space are the Device ID, Vendor ID, Class
Code, Status and Command registers, and Base Address Registers.
•
lspci -xxxx -d [<vendor>]:[<device>]
This displays the extended configuration space of the device. It might be useful to read
the extended configuration space on the root and look for the Advanced Error
Reporting (AER) registers. These registers provide more information on why the
device has flagged an error (for example, it might show that a correctable error was
issued because of a replay timer time-out).
•
lspci -k
Shows kernel drivers handling each device and kernel modules capable of handling it
(works with kernel 2.6 or later).
PCItree (Windows)
PCItree can be downloaded at www.pcitree.de and allows the user to view the PCI Express
device configuration space and perform 1 DWORD memory writes and reads to the
aperture.
170
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Debug Tools
The configuration space is displayed by default in the lower right corner when the device
is selected, as shown in Figure D-1.
X-Ref Target - Figure D-1
Figure D-1:
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
PCItree with Read of Configuration Space
www.xilinx.com
171
Appendix D: Debugging Designs
HWDIRECT (Windows)
HWDIRECT can be purchased at www.eprotek.com and allows the user to view the PCI
Express device configuration space as well as the extended configuration space (including
the AER registers on the root).
X-Ref Target - Figure D-2
Figure D-2:
HWDIRECT with Read of Configuration Space
PCI-SIG Software Suites
PCI-SIG® software suites such as PCIE-CV can be used to test compliance with the
specification. This software can be downloaded at www.pcisig.com.
172
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Debug Ports
Debug Ports
The Spartan-6 FPGA Integrated Endpoint Block for PCI Express has debug ports described
in Table D-3 providing insight to why the different error conditions occur. The receiver
might detect different problems that result in either a Fatal, Non-fatal, or correctable error.
Also, the receiver might detect an unsupported request. Four of the debug signals shown
in Table D-2 mirror the lower four bits of the PCI Express device status register. When one
of these conditions occurs, another signal is asserted for one clock cycle to show the reason
causing the error.
Table D-2:
Device Status Register Debug Ports
Name
Device Status Bit
dbg_reg_detected_correctable
Bit 0 - correctable
dbg_reg_detected_non_fatal
Bit 1 - Non-Fatal
dbg_reg_detected_fatal
Bit 2 - Fatal
dbg_reg_detected_unsupported
Bit 3 - Unsupported Request
Table D-3 defines the debug port signals.
Table D-3:
Spartan-6 FPGA Integrated Block for PCI Express Debug Ports
Port
Direction Clock Domain
Description
dbg_bad_dllp_status
Output
USERCLK
This signal pulses High for one USERCLK cycle when a
DLLP CRC error is detected.
dbg_bad_tlp_lcrc
Output
USERCLK
This signal pulses High for one USERCLK cycle when a
TLP with an LCRC error is detected.
dbg_bad_tlp_seq_num
Output
USERCLK
This signal pulses High for one USERCLK cycle when a
TLP with an invalid sequence number is detected.
dbg_bad_tlp_status
Output
USERCLK
This signal pulses High for one USERCLK cycle when a
bad TLP is detected, for reasons other than a bad LCRC
or a bad sequence number.
dbg_dl_protocol_status
Output
USERCLK
This signal pulses High for one USERCLK cycle if an outof-range ACK or NAK is received.
dbg_fc_protocol_err_status
Output
USERCLK
This signal pulses High for one USERCLK cycle if there
is a protocol error with the received flow control updates.
dbg_mlfrmd_length
Output
USERCLK
This signal pulses High for one USERCLK cycle to
indicate that a received TLP had a length that did not
match what was in the TLP header.
dbg_mlfrmd_mps
Output
USERCLK
This signal pulses High for one USERCLK cycle to
indicate that a received TLP had a length in violation of
the negotiated MPS.
dbg_mlfrmd_tcvc
Output
USERCLK
This signal pulses High for one USERCLK cycle to
indicate that a received TLP had an invalid TC or VC
value.
dbg_mlfrmd_tlp_status
Output
USERCLK
This signal pulses High for one USERCLK cycle when a
malformed TLP is received. See the other
DBGMLFRMD* signals for further clarification.
Note: There is skew between DBGMLFRMD* and
DBGMLFRMDTLPSTATUS.
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
173
Appendix D: Debugging Designs
Table D-3:
Spartan-6 FPGA Integrated Block for PCI Express Debug Ports (Cont’d)
Port
Direction Clock Domain
Description
dbg_mlfrmd_unrec_type
Output
USERCLK
This signal pulses High for one USERCLK cycle to
indicate that a received TLP had an
invalid/unrecognized type field value.
dbg_poistlpstatus
Output
USERCLK
This signal pulses High for one USERCLK cycle to
indicate that a TLP was received with the EP (poisoned)
status bit set.
dbg_rcvr_overflow_status
Output
USERCLK
This signal pulses High for one USERCLK cycle if a
received TLP violates the advertised credit.
dbg_reg_detected_correctable
Output
USERCLK
This signal is a mirror of the internal signal used to
indicate a correctable error is detected. The error is
cleared upon a read by the Root Complex (RC).
dbg_reg_detected_fatal
Output
USERCLK
This signal is a mirror of the internal signal used to
indicate that a fatal error has been detected. The error is
cleared upon a read by the RC.
dbg_reg_detected_non_fatal
Output
USERCLK
This signal is a mirror of the internal signal used to
indicate that a non-fatal error has been detected. The
error is cleared upon a read by the RC.
dbg_reg_detected_unsupported
Output
USERCLK
This signal is a mirror of the internal signal used to
indicate that an unsupported request has been detected.
The error is cleared upon a read by the RC.
dbg_rply_rollover_status
Output
USERCLK
This signal pulses High for one USERCLK cycle when
the rollover counter expires.
dbg_rply_timeout_status
Output
USERCLK
This signal pulses High for one USERCLK cycle when
the replay time-out counter expires.
dbg_ur_no_bar_hit
Output
USERCLK
This signal pulses High for one USERCLK cycle to
indicate that a received read or write request did not
match any configured BAR.
dbg_ur_pois_cfg_wr
Output
USERCLK
This signal pulses High for one USERCLK cycle to
indicate that a CfgWr TLP with the Error/Poisoned bit
(EP) = 1 was received.
dbg_ur_status
Output
USERCLK
This signal pulses High for one USERCLK cycle when an
unsupported request is received. See the DBGUR*
signals for further clarification.
Note: There is skew between DBGUR* and
DBGURSTATUS.
dbg_ur_unsup_msg
Output
USERCLK
This signal pulses High for one USERCLK cycle to
indicate that an Msg or MsgD TLP with an unsupported
type was received.
Using the Debug Ports
The debug ports are outputs on the integrated Endpoint block and users can access them
by opening the wrapper source file in the source directory. This file is named
<corename>.v[hd] where <corename> represents core name entered in the
CORE Generator tool. Signals are defined for each of these ports as either wires in the
Verilog version or signals in the VHDL version.
174
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Hardware Debug
The debug ports can be used in both simulation and in hardware to debug problems. In
simulation, these signals can easily be added to the waveform viewer without any changes
to the code because they are already defined in the wrapper file. In hardware, users might
want to use the ChipScope Pro tool to monitor these signals or probe them to external ports
or add additional logic such as counters to enable more in depth analysis. Users might
need to bring these signals to upper layers in the design and this can be done by modifying
the port description and instantiations of the files.
Figure D-3 shows a common problem faced by many users. This illustrates the behavior of
the debug ports when a non-fatal error condition occurs. The scenario is a memory write is
sent from the downstream port to the Spartan-6 FPGA Endpoint. This memory write does
not correctly target any of the available BARs in the design. This results in a non-fatal error
condition. However, there are other reasons that cause a non-fatal error, so monitoring the
cfg_dev_status_nonfatal_err_detected output of the core might not be sufficient to debug
the problem. By using the debug ports, the user can see that the non-fatal error was caused
by a BAR miss due to dbg_ur_no_bar_hit asserted for one cycle.
X-Ref Target - Figure D-3
Figure D-3:
Debug Wave Screenshot
In the ChipScope tool, the user needs to decide how best to trigger the ChipScope tool to
capture these problems. There are various ways to do this, but one suggestion would be to
use the signals in Table D-3 as triggers. At least one of these signals is asserted. When the
ChipScope tool triggers, the rest of the signals can be analyzed to find out exactly what
caused the error condition.
Hardware Debug
Hardware issues can range from device recognition issues to problems seen after hours of
testing. This section provides debug flow diagrams for some of the most common issues
experienced by users. Endpoints that are shaded gray indicate that more information can
be found in sections after Figure D-4.
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
175
Appendix D: Debugging Designs
X-Ref Target - Figure D-4
Design Fails in Hardware
No
Using probes, an LED, ChipScope
or some other method, determine if
trn_lnk_up_n is asserted. When
trn_lnk_up_n is low, it indicates
the core has achieved link up
meaning the LTSSM is in L0 state
and the data link layer is in the
DL_Active state.
Yes
Is trn_lnk_up_n asserted?
(trn_lnk_up_n = 0)
See “Link is Training Debug” section.
No
To eliminate FPGA configuration
as a root cause, perform a soft
restart of the system. Performing a
soft reset on the system will keep
power applied and forces
re-enumeration of the device.
Yes
Does a soft reset fix the problem?
(trn_lnk_up=0)
See "FPGA Configuration Time
Debug" section.
No
One reason trn_reset_n stays
asserted other than the system
reset being asserted, is due to a
faulty clock. This may keep the
PLL from locking which holds
trn_reset_n asserted.
No
Is trn_reset_n deasserted?
(trn_reset_n = 1)
Multi-lane links are susceptible to
crosstalk and noise when all lanes
are switching during training.
A quick test for this is forcing one
lane operation. This can be done
by using an interposer or adapter
to isolate the upper lanes or use
a tape such as Scotch tape and
tape off the upper lanes on the
connector. If its a embedded board,
remove the AC capacitors if
possible to isolate the lanes.
See "Clock Debug" section.
Yes
Yes
Is it a multi-lane link?
Force x1 Operation
Does trn_lnk_up_n = 0 when using
as x1 only?
No
Yes
Do you have a link analyzer?
Yes
No
The ChipScope tool may be used to try and
determine the point of failure.
Figure D-4:
176
There are potentially problems
with the board layout causing
interference when all lanes are
switching. See board debug
suggestions.
Use the link analyzer to monitor the training
sequence and to determine the point of failure.
Have the analyzer trigger on the first TS1 that it
recognizes and then compare the output to the
LTSSM state machine sequences outlined in
Chapter 4 of the PCI Express Base Specification.
Design Fails in Hardware Debug Flow Diagram
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Hardware Debug
FPGA Configuration Time Debug
Device initialization and configuration issues can be caused by not having the FPGA
configured fast enough to enter link training and be recognized by the system. Section 6.6
of PCI Express Base Specification, v1.1 states two rules that might be impacted by FPGA
Configuration Time:
•
A component must enter the LTSSM Detect state within 20 ms of the end of the
Fundamental reset.
•
A system must guarantee that all components intended to be software visible at boot
time are ready to receive Configuration Requests within 100 ms of the end of
Conventional Reset at the Root Complex.
These statements basically mean there is a finite time in which the FPGA must be
configured by, and not meeting these requirements could cause problems with link
training and device recognition.
Configuration can be accomplished using an onboard PROM or dynamically using JTAG.
When using JTAG to configure the device, configuration typically occurs after the Chipset
has enumerated each peripheral. After configuring the FPGA, a soft reset is required to
restart enumeration and configuration of the device. A soft reset on a Windows based PC
is performed by going to Start -> Shut Down and then selecting Restart.
To eliminate FPGA configuration as a root cause, perform a soft restart of the system.
Performing a soft reset on the system keeps power applied and forces re-enumeration of
the device. If the device links up and is recognized after a soft reset is performed, then
FPGA configuration is most likely the problem. Most typical systems use ATX power
supplies which provides some margin on this 100 ms window as the power supply is
normally valid before the 100 ms window starts. For more information on FPGA
configuration, see Chapter 8, FPGA Configuration.
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
177
Appendix D: Debugging Designs
Link is Training Debug
X-Ref Target - Figure D-5
Link is Training
(trn_lnk_up_n = 0)
PCITREE and lspci will scan the
the system and display devices
recognized during startup. These
tools show the PCI configuration
space and its settings within
the device.
Is the device recognized by the system?
Can it be seen by PCITREE (Windows) or
lspci (Linux)?
Yes
See “Data Transfer Failing Debug”
section.
No
To eliminate FGPA configuration
as a root cause, perform a soft
restart of the system. Performing a
soft reset on the system will keep
power applied and forces
re-enumeration of the device.
If this fixes the problem, then it is
likely the FPGA is not configured in
time for the host to access the card.
The PIO design is known to work.
Often, the PIO design will work when
a user design will not. This usually
indicates some parameter or resource
conflict due to settings used for the
user design configuration.
It is recommended to mirror the PIO
CORE Generator GUI settings into
the user design. Even though the
design may not function, it should
still be recognized by the system.
Does a soft reset fix the problem?
(trn_lnk_up=0)
Yes
See "FPGA Configuration Time
Debug" section.
No
Does using the PIO example
design fix the problem?
Does mirroring the PIO
CORE Generator GUI settings for
the user design fix the problem?
Yes
No
Yes
Do you have a link analyzer?
No
Check for configuration settings
conflict. See the "Debugging
PCI Configuration Space Parameters"
section.
No
Yes
With no link analyzer, it is possible to use
ChipScope to gather the same information.
If the PIO design works, but mirroring the
configuration parameters does not fix the
problem, then attention should be focused on
the user application design. See the "Application
Requirements" section.
It is likely, the problem is due to the device
not responding properly to some type of access. A
link analyzer allows you to view the link traffic
and determine if something is incorrect. See
the "Using a Link Analyzer to Debug
Device Recognition Issues” section.
Figure D-5:
178
Link Trained Debug Flow Diagram
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Hardware Debug
FPGA Configuration Time Debug
Device initialization and configuration issues can be caused by not having the FPGA
configured fast enough to enter link training and be recognized by the system. Section 6.6
of PCI Express Base Specification, v2.0 states two rules that might be impacted by FPGA
Configuration Time:
•
A component must enter the LTSSM Detect state within 20 ms of the end of the
Fundamental reset.
•
A system must guarantee that all components intended to be software visible at boot
time are ready to receive Configuration Requests within 100 ms of the end of
Conventional Reset at the Root Complex.
These statements basically mean there is a finite time in which the FPGA must be
configured by, and not meeting these requirements could cause problems with link
training and device recognition.
Configuration can be accomplished using an onboard PROM or dynamically using JTAG.
When using JTAG to configure the device, configuration typically occurs after the Chipset
has enumerated each peripheral. After configuring the FPGA, a soft reset is required to
restart enumeration and configuration of the device. A soft reset on a Windows based PC
is performed by going to Start -> Shut Down and then selecting Restart.
To eliminate FPGA configuration as a root cause, perform a soft restart of the system.
Performing a soft reset on the system keeps power applied and forces re-enumeration of
the device. If the device links up and is recognized after a soft reset is performed, then
FPGA configuration is most likely the problem. Most typical systems use ATX power
supplies which provides some margin on this 100 ms window as the power supply is
normally valid before the 100 ms window starts. For more information on FPGA
configuration, see Chapter 8, FPGA Configuration.
Debugging PCI Configuration Space Parameters
Often, a user application fails to be recognized by the system, but the Xilinx PIO Example
design works. In these cases, the user application is often using a PCI configuration space
setting that is interfering with the system systems ability to recognize and allocate
resources to the card.
Xilinx PCI Express solutions handle all configuration transactions internally and generate
the correct responses to incoming configuration requests. Chipsets have limits as to the
amount of system resources they can allocate, and the core must be configured to adhere to
these limitations.
The resources requested by the endpoint are identified by the BAR settings within the
Endpoint configuration space. Verify that the resources requested in each BAR can be
allocated by the chipset. I/O BARs are especially limited so configuring a large I/O BAR
typically prevents the chipset from configuring the device. Generate a core that
implements a small amount of Memory (approximately 2 KB) to identify if this is the root
cause.
The Class Code setting selected in the CORE Generator software GUI can also affect
configuration. The Class Code informs the Chipset as to what type of device the Endpoint
is. Chipsets might expect a certain type of device to be plugged into the PCI Express slot
and configuration might fail if it reads an unexpected Class Code. The BIOS could be
configurable to workaround this issue.
Use the PIO design with default settings to rule out any device allocation issues. The PIO
design default settings have proven to work in all systems encountered when debugging
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
179
Appendix D: Debugging Designs
problems. If the default settings allow the device to be recognized, then change the PIO
design settings to match the intended user application by changing the PIO configuration
the CORE Generator software GUI. Trial and error might be required to pinpoint the
problem if a link analyzer is not available.
Using a link analyzer, it is possible to monitor the link traffic and possibly determine when
during the enumeration and configuration process problems occur.
Application Requirements
During enumeration, it is possible for the chipset to issue TLP traffic that is passed from the
core to the backend application. A common oversight when designing custom backend
applications is to not have logic which handles every type incoming request. As a result, no
response is created and problems arise. The PIO design has the necessary backend
functions to respond correctly to any incoming request. It is the responsibility of the
application to generate the correct response. These packet types are presented to the
application:
•
Requests targeting the Expansion ROM (if enabled)
•
Message TLPs
•
Memory or I/O requests targeting a BAR
•
All completion packets
The PIO design, can be used to rule out any of these types of concerns, as the PIO design
responds to all incoming transactions to the user application in some way to ensure the
host receives the proper response allowing the system to progress. If the PIO design works,
but the custom application does not, this means that some transaction is not being handled
properly.
The ChipScope analyzer should be implemented on the wrapper TRN Receive interface to
identify if requests targeting the backend application are drained and completed
successfully. The TRN interface signals that should be probed in the ChipScope analyzer
are defined in Table D-4, page 182.
Using a Link Analyzer to Debug Device Recognition Issues
In cases where the link is up (trn_lnk_up_n = 0), but the device is not recognized by the
system, a link analyzer can help solve the problem. It is likely the FPGA is not responding
properly to some type of access. Use the link view to analyzer the traffic and see if anything
looks out of place.
To focus on the problem, it might be necessary to try different triggers. Here are some
trigger examples:
180
•
Trigger on the first INIT_FC1 and/or UPDATE_FC in either direction. This allows the
analyzer to begin capture after link up.
•
The first TLP normally transmitted to an endpoint is the Set Slot Power Limit
Message. This usually occurs before Configuration traffic begins. This might be a
good trigger point.
•
Trigger on Configuration TLPs.
•
Trigger on Memory Read or Memory Write TLPs.
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Hardware Debug
Data Transfer Failing Debug
X-Ref Target - Figure D-6
Link is Up (trn_lnk_up_n = 0)
Device is recognized by system.
Data Transfers failing.
The most often cause of a system
freeze or hang is due to a
completion timeout occurring
on the host. This happens, when
the host issues a non-posted
transaction (usually a memory
read) to the Endpoint and the
Endpoint's user application does
not properly respond.
Is the system freezing or hanging?
Yes
Ensure that completions are returned
for all incoming Non-Posted traffic.
No
If trn_lnk_up_n is toggling, it usually
means the physical link is marginal.
In these cases, the link may be
established but may then fail once
traffic begins to flow. Use ChipScope
Pro or probe trn_lnk_up_n to a logic
analyzer and determine if it is toggling.
Is trn_lnk_up_n toggling?
Yes
Link could be marginal and packets
are failing to pass LCRC check.
No
Errors are reported to the user
interface on the output cfg_dstatus[3:0].
This is a copy of the device status
register. Using ChipScope monitor
this bus for errors.
Fatal Error? Blue screen?
Other errors?
Yes
Errors flagged by the core are due
to problems on the receive data path.
Use a link analyzer if possible to
check incoming packets. See the
"Identifying Errors" section.
No
Receive
Is the problem with receiving
or transmitting TLPs?
Do incoming packets appear
on TRN receive interface?
Transmit
Do outgoing packets arrive
at destination?
No
No
If read or write transactions do not
appear on the trn interface, it means
that most likely the incoming packet
did not hit a BAR. Verify incoming
TLP addresses against BAR
allocation.
If completion packets fail to reach their
destination, ensure the packet
contained the correct requester ID as
captured from the original
Non-Posted TLP.
A memory write that misses a BAR
results in a Non-Fatal error message.
A non-posted transaction that misses a
BAR results in a Completion with
UR status.
Figure D-6:
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
If other packets fail, ensure the address
targeted is valid.
Data Transfer Debug Flow Diagram
www.xilinx.com
181
Appendix D: Debugging Designs
Identifying Errors
Hardware symptoms of system lock up issues are indicated when the system hangs or a
blue screen appears (PC systems). The PCI Express Base Specification v2.0 requires that error
detection be implemented at the receiver. A system lock up or hang is commonly the result
of a Fatal Error and is reported in bit 2 of the receivers Device Status register. Using the
ChipScope tool, monitor the core’s device status register to see if a fatal error is being
reported.
A fatal error reported at the Root complex implies an issue on the transmit side of the EP.
The Root Complex Device Status register can often times be seen using PCITree (Windows)
or LSPCI (Linux). If a fatal error is detected, refer to the Transmit section. A Root Complex
can often implement Advanced Error Reporting (AER) which further distinguishes the
type of error reported. AER provides valuable information as to why a certain error was
flagged and is provided as an extended capability within a devices configuration space.
Section 7.10 of the PCI Express Base Specification v2.0 provides more information on AER
registers.
Transmit
Fatal Error Detected on Root or Link Partner
Check to make sure the TLP is correctly formed and that the payload (if one is attached)
matches what is stated in the header length field. The Endpoints device status register does
not report errors created by traffic on the transmit channel.
The signals shown in Table D-4 should be monitored on the Transmit interface to verify all
traffic being initiated is correct.
Table D-4:
TRN Transmit Interface Signals
Name
Direction
Description
Output
Transaction Link Up: Active Low. Transaction link-up is
asserted when the core and the connected upstream link
partner port are ready and able to exchange data packets.
Transaction link-up is deasserted when the core and link
partner are attempting to establish communication, and
when communication with the link partner is lost due to
errors on the transmission channel. When the core is driven
to Hot Reset and Link Disable states by the link partner,
trn_lnk_up_n is deasserted and all TLP’s stored in the
endpoint core are lost.
trn_tsof_n
Input
Transmit Start-of-Frame (SOF): Active Low. Signals the start
of a packet. Valid only along with assertion of
trn_tsrc_rdy_n.
trn_teof_n
Input
Transmit End-of-Frame (EOF): Active Low. Signals the end
of a packet. Valid only along with assertion of
trn_tsrc_rdy_n.
trn_td[63:0]
Input
Transmit Data: Packet data to be transmitted.
trn_trem_n
Input
Transmit Data Remainder: Valid only if trn_teof_n,
trn_tsrc_rdy_n, and trn_tdst_rdy_n are all asserted. Legal
values are:
trn_lnk_up_n
• 0 = packet data on all of trn_td[63:0]
• 1 = packet data only on trn_td[63:32]
182
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Hardware Debug
Table D-4:
TRN Transmit Interface Signals (Cont’d)
Name
Direction
Description
trn_tsrc_rdy_n
Input
Transmit Source Ready: Active Low. Indicates that the User
Application is presenting valid data on trn_td[63:0].
trn_tdst_rdy_n
Output
Transmit Destination Ready: Active Low. Indicates that the
core is ready to accept data on trn_td[63:0]. The
simultaneous assertion of trn_tsrc_rdy_n and
trn_tdst_rdy_n marks the successful transfer of one data
beat on trn_td[63:0].
Fatal Error Not Detected
Ensure that the address provided in the TLP header is valid. The kernel mode driver
attached to the device is responsible for obtaining the system resources allocated to the
device. In a Bus Mastering design, the driver is also responsible for providing the
application with a valid address range. System hangs or blue screens might occur if a TLP
contains an address which does not target the designated address range for that device.
Receive
Xilinx solutions for PCI Express provide the Device Status register to the application on
CFG_DSTATUS[3:0]. Debug ports are available to help users determine the exact cause of
errors on the Endpoint's receiver. See Debug Ports, page 173 for information on these ports.
Table D-5:
Description of CFG_DSTATUS[3:0]
CFG_DSTATUS[3:0]
Description
CFG_DSTATUS[0]
Correctable Error Detected
CFG_DSTATUS[1]
Non-Fatal Error Detected
CFG_DSTATUS[2]
Fatal Error Detected
CFG_DSTATUS[3]
UR Detected
System lock up conditions due to issues on the receive channel of the PCI Express core are
often result of an error message being sent upstream to the root. Error messages are only
sent when error reporting is enabled in the Device Control register.
A fatal condition is reported if any of these occur:
•
Training Error
•
DLL Protocol Error
•
Flow Control Protocol Error
•
Malformed TLP
•
Receiver Overflow
The first four bullets are not common in hardware because both Xilinx PCI Express
solutions and connected components have been thoroughly tested in simulation and
hardware. However, a receiver overflow is a possibility. Users must ensure they follow
requirements discussed in Receiver Flow Control Credits Available in Chapter 6 when
issuing memory reads.
Debug ports are available to help users determine the exact cause of errors on the
endpoint's receiver. See Debug Ports, page 173 for information on these ports.
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
183
Appendix D: Debugging Designs
Non-Fatal Errors
This section lists conditions that are reported as Non-Fatal errors. See the PCI Express Base
Specification for more details.
If the error is being reported by the root, the Advanced Error Reporting (AER) registers can
be read to determine the condition that led to the error. Use a tool such as HWDIRECT,
discussed in Third-Party Software Tools, page 170, to read the root's AER registers.
Chapter 7 of the PCI Express Base Specification defines the AER registers. If the error is
signaled by the endpoint, debug ports are available to help determine the specific cause of
the error.
Correctable Non-Fatal errors are:
•
Receiver Error
•
Bad TLP
•
Bad DLLP
•
Replay Timeout
•
Replay NUM Rollover
The first three errors listed above are detected by the receiver and are not common in
hardware systems. The replay error conditions are signaled by the transmitter. If an ACK is
not received for a packet within the allowed time, it is replayed by the transmitter.
Throughput can be reduced if many packets are being replayed, and the source can usually
be determined by examining the link analyzer or ChipScope tool captures.
Uncorrectable Non-Fatal errors are:
•
Poisoned TLP
•
Received ECRC Check Failed
•
Unsupported Request (UR)
•
Completion Timeout
•
Completer Abort
•
Unexpected Completion
•
ACS Violation
An unsupported request usually indicates that the address in the TLP did not fall within
the address space allocated to the BAR. This often points to a problem with the address
translation performed by the driver. Ensure also that the BAR has been assigned correctly
by the root at start-up. LSPCI or PCItree discussed in Third-Party Software Tools, page 170
can be used to read the BAR values for each device.
A completion timeout indicates that no completion was returned for a transmitted TLP
and is reported by the requester. This can cause the system to hang (could include a blue
screen on Windows) and is usually caused when one of the devices locks up and stops
responding to incoming TLPs. If the root is reporting the completion timeout, the
ChipScope analyzer can be used to investigate why the User Application did not respond
to a TLP (for example, the User Application is busy, there are no transmit buffers available,
or trn_tdst_rdy_n is deasserted). If the endpoint is reporting the Completion timeout,
a link analyzer would show the traffic patterns during the time of failure and would be
useful in determining the root cause.
184
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Simulation Debug
Next Steps
If the debug suggestions listed above do not resolve the issue, open a support case to have
the appropriate Xilinx expert assist with the issue.
To create a technical support case in Webcase, see the Xilinx website at:
www.xilinx.com/support/clearexpress/websupport.htm
Items to include when opening a case:
•
Detailed description of the issue and results of the steps listed above.
•
Attach ChipScope analyzer VCD captures taken in the steps above.
To discuss possible solutions, use the Xilinx User Community:
forums.xilinx.com/xlnx/
Simulation Debug
This section provides simulation debug flow diagrams for some of the most common
issues experienced by users. Endpoints that are shaded gray indicate that more
information can be found in sections below the figure.
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
185
Appendix D: Debugging Designs
ModelSim Debug
X-Ref Target - Figure D-7
ModelSim
Simulation Debug
SecureIP models are used to
simulate the integrated block
for PCI Express and the MGTs.
To use these models, a Verilog
LRM-IEEE 1364-2005 encryptioncompliant simulator is required.
Are you using ModelSim version 6.4a
or later?
No
Update ModelSim to
version 6.4a or later.
Yes
A Verilog license is required to
simulate with the SecureIP models.
If the user design uses VHDL, a
mixed-mode simulation license is
required.
If using VHDL, do you have a
mixed-mode simulation license?
No
Obtain a mixed-mode
simulation license.
Yes
The PIO Example design should
allow the user to quickly determine
if the simulator is set up correctly.
The default test will achieve link up
(trn_lnk_up_n = 0) and issue a
Configuration Read to the core's
Device and VendorID.
Does simulating the PIO Example
Design give the expected output?
Yes
See "PIO Simulator Expected
Output" section.
Yes
Need to compile and map the
proper libraries. See "Compiling
Simulation Libraries Section."
No
If the libraries are not compiled and
mapped correctly, it will cause errors
such as:
# ** Error: (vopt-19) Failed to access
library 'secureip' at "secureip".
# No such file or directory.
(errno = ENOENT)
# ** Error: ../../example_design/
xilinx_pcie_2_0_ep_v6.v(820):
Library secureip not found.
To model the Integrated block for
PCI Express and the MGTs, the
SecureIP models are used. These
models must be referenced during
the vsim call. Also, it is necessary to
reference the unisims library and
possibly xilinxcorelib depending
on the design.
Do you get errors referring to
failing to access library?
No
Do you get errors indicating
"PCIE_2_0" or other elements like
"BUFG" not defined?
Yes
Add the "-L" switch with the appropriate
library reference to the vsim command
line. For example: -L secureip or
-L unisims_ver.
No
Are you able to receive packets
on the TRN RX interface and transmit
packets on the TRN TX interface?
Yes
If problem is more design specific, open
a case with Xilinx Technical Support
and include a wlf file dump of the simulation.
For the best results, dump the entire design
hierarchy.
No
One of the most common mistakes in
simulation of an Endpoint is forgetting
to set the Memory, IO, and Bus Master
Enable bits to a 1 in the PCI Command
register in the configuration space.
In the DSPORT test bench application,
issue a Configuration Write to the PCI
Command register at DWORD address
offset 0x04 and set bits [2:0] to 111b.
Figure D-7:
186
ModelSim Debug FLow Diagram
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Simulation Debug
PIO Simulator Expected Output
The PIO design simulation should give the output as follows:
# Loading work.board(fast)
# Loading unisims_ver.IBUFDS_GTXE1(fast)
# Loading work.pcie_clocking_v6(fast)
# Loading unisims_ver.PCIE_2_0(fast)
# Loading work.pcie_gtx_v6(fast)
# Loading unisims_ver.GTXE1(fast)
# Loading unisims_ver.RAMB36(fast)
# Loading unisims_ver.RAMB16_S36_S36(fast)
# Loading unisims_ver.PCIE_2_0(fast__1)
# Loading work.glbl(fast)
# [
0] board.EP.core.pcie_2_0_i.pcie_bram_i ROWS_TX 1 COLS_TX 2
# [
0] board.EP.core.pcie_2_0_i.pcie_bram_i ROWS_RX 1 COLS_RX 2
# [
0] board.EP.core.pcie_2_0_i.pcie_bram_i.pcie_brams_tx NUM_BRAMS 2
DOB_REG 1 WIDTH 36 RAM_WRITE_LATENCY 0 RAM_RADDR_LATENCY 0 RAM_RDATA_LATENCY 2
# [
0] board.EP.core.pcie_2_0_i.pcie_bram_i.pcie_brams_rx NUM_BRAMS 2
DOB_REG 1 WIDTH 36 RAM_WRITE_LATENCY 0 RAM_RADDR_LATENCY 0 RAM_RDATA_LATENCY 2
# [
0] board.RP.rport.pcie_2_0_i.pcie_bram_i ROWS_TX 1 COLS_TX 2
# [
0] board.RP.rport.pcie_2_0_i.pcie_bram_i ROWS_RX 1 COLS_RX 2
# [
0] board.RP.rport.pcie_2_0_i.pcie_bram_i.pcie_brams_tx NUM_BRAMS 2
DOB_REG 1 WIDTH 36 RAM_WRITE_LATENCY 0 RAM_RADDR_LATENCY 0 RAM_RDATA_LATENCY 2
# [
0] board.RP.rport.pcie_2_0_i.pcie_bram_i.pcie_brams_rx NUM_BRAMS 2
DOB_REG 1 WIDTH 36 RAM_WRITE_LATENCY 0 RAM_RADDR_LATENCY 0 RAM_RDATA_LATENCY 2
# Running test {sample_smoke_test0}......
# [
0] : System Reset Asserted...
# [
4995000] : System Reset De-asserted...
# [
64069100] : Transaction Reset Is De-asserted...
# [
73661100] : Transaction Link Is Up...
# [
73661100] : Expected Device/Vendor ID = 000710ee
# [
73661100] : Reading from PCI/PCI-Express Configuration Register 0x00
# [
73673000] : TSK_PARSE_FRAME on Transmit
# [
74941000] : TSK_PARSE_FRAME on Receive
# [
75273000] : TEST PASSED --- Device/Vendor ID 000710ee successfully received
# ** Note: $finish
: ../tests/sample_tests1.v(29)
#
Time: 75273 ns Iteration: 3 Instance: /board/RP/tx_usrapp
Compiling Simulation Libraries
Use the compxlib command to compile simulation libraries. This tool is delivered as part of
the Xilinx software. For more information see the ISE Software Manuals and specifically
the Development System Reference Guide under the section titled compxlib.
Assuming the Xilinx and ModelSim environments are set up correctly, this is an example of
compiling the SecureIP and UniSims libraries for Verilog into the current directory
compxlib -s mti_se -arch spartan6 -l verilog -lib secureip -lib unisims
-dir ./
There are many other options available for compxlib described in the Development System
Reference Guide.
Compxlib produces a modelsim.ini file containing the library mappings. In ModelSim,
to see the current library mappings type vmap at the prompt. The mappings can be
updated in the INI file or to map a library at the ModelSim prompt type:
vmap [<logical_name>] [<path>]
For example:
Vmap unisims_ver C:\my_unisim_lib
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
187
Appendix D: Debugging Designs
Next Step
If the debug suggestions listed above do not resolve the issue, a support case should be
opened to have the appropriate Xilinx expert assist with the issue.
To create a technical support case in Webcase, see the Xilinx website at:
www.xilinx.com/support/clearexpress/websupport.htm
Items to include when opening a case:
•
Detailed description of the issue and results of the steps listed above.
•
Attach a VCD or WLF dump of the simulation.
To discuss possible solutions, use the Xilinx User Community:
forums.xilinx.com/xlnx/
188
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Appendix E
Managing Receive-Buffer Space for
Inbound Completions
The PCI Express Base Specification requires all Endpoints to advertise infinite Flow
Control credits for received Completions to their link partners. This means that an
Endpoint must only transmit Non-Posted Requests for which it has space to accept
Completion responses. This appendix describes how a User Application can manage the
receive-buffer space in the PCI Express Endpoint core to fulfill this requirement.
General Considerations and Concepts
Completion Space
Table E-1 defines the completion space reserved in the receive buffer by the core. The
values differ for different versions of the core, and also differ based on whether the
designer chooses to have TLP Digests (ECRC) removed from the incoming packet stream.
Values are credits, expressed in decimal.
Table E-1:
Receiver-Buffer Completion Space
Performance Level : Good
Capability Max Payload
Size (bytes)
Performance Level : High
Cpl. Hdr.
(Total_CplH)
Cpl. Data
(Total_CplD)
Cpl. Hdr.
(Total_CplH)
Cpl. Data
(Total_CplD)
128
8
64
16
128
256
16
128
32
256
512
32
256
32
256
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
189
Appendix E: Managing Receive-Buffer Space for Inbound Completions
Maximum Request Size
A Memory Read cannot request more than the value stated in Max_Request_Size, which is
given by Configuration bits cfg_dcommand[14:12] as defined in Table E-2. If the User
Application chooses not to read the Max_Request_Size value, it must use the default value
of 128 bytes.
Table E-2:
Max Request Size Settings
Max_Request_Size
cfg_dcommand[14:12]
Bytes
DW
QW
Credits
000b
128
32
16
8
001b
256
64
32
16
010b
512
128
64
32
011b
1024
256
128
64
100b
2048
512
256
128
101b
4096
1024
512
256
Reserved
110b–111b
Read Completion Boundary
A Memory Read can be answered with multiple Completions, which when put together
return all requested data. To make room for packet-header overhead, the User Application
must allocate enough space for the maximum number of Completions that might be
returned.
To make this process easier, the Base Specification quantizes the length of all Completion
packets such that each must start and end on a naturally aligned Read Completion
Boundary (RCB), unless it services the starting or ending address of the original request.
The value of RCB is determined by Configuration bit cfg_lcommand[3] as defined in
Table E-3. If the User Application chooses not to read the RCB value, it must use the default
value of 64 bytes.
Table E-3:
Read Completion Boundary Settings
Read Completion Boundary
cfg_lcommand[3]
Bytes
DW
QW
Credits
0
64
16
8
4
1
128
32
16
8
When calculating the number of Completion credits a Non-Posted Request requires, the
user must determine how many RCB-bounded blocks the Completion response might
require; this is the same as the number of Completion Header credits required.
190
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Methods of Managing Completion Space
Methods of Managing Completion Space
A User Application can choose one of four methods to manage receive-buffer Completion
space, as listed in Table E-4. For convenience, this discussion refers to these methods as
LIMIT_FC, PACKET_FC, RCB_FC, and DATA_FC. Each has advantages and
disadvantages that the designer needs to consider when developing the user application.
Table E-4:
Managing Receive Completion Space Methods
Method
LIMIT_FC
PACKET_FC
Description
Advantage
Disadvantage
Limit the total number of
outstanding NP Requests
Simplest method to
implement in user
logic
Much Completion
capacity goes
unused
Track the number of
outstanding CplH and CplD
credits; allocate and
deallocate on a per-packet
basis
Relatively simple
user logic; finer
allocation
granularity means
less wasted capacity
than LIMIT_FC
As with LIMIT_FC,
credits for an NP are
still tied up until the
Request is
completely satisfied
RCB_FC
Track the number of
Ties up credits for
outstanding CplH and CplD less time than
credits; allocate and
PACKET_FC
deallocate on a per-RCB basis
More complex user
logic than LIMIT_FC
or PACKET_FC
DATA_FC
Track the number of
Lowest amount of
outstanding CplH and CplD wasted capacity
credits; allocate and
deallocate on a per-RCB basis
Most complex user
logic
The LIMIT_FC Method
The LIMIT_FC method is the simplest to implement. The User Application assesses the
maximum number of outstanding Non-Posted Requests allowed at one time, MAX_NP. To
calculate this value, perform these steps:
1.
Determine the number of CplH credits required by a Max_Request_Size packet:
Max_Header_Count = ceiling(Max_Request_Size / RCB)
2.
Determine the greatest number of maximum-sized Completions supported by the
CplD credit pool:
Max_Packet_Count_CplD = floor(CplD / Max_Request_Size)
3.
Determine the greatest number of maximum-sized Completions supported by the
CplH credit pool:
Max_Packet_Count_CplH = floor(CplH / Max_Header_Count)
4.
Use the smaller of the two quantities from steps 2 and 3 to obtain the maximum number
of outstanding Non-Posted requests:
MAX_NP = min(Max_Packet_Count_CplH, Max_Packet_Count_CplD)
With knowledge of MAX_NP, the User Application can load a register NP_PENDING with
zero at reset and make sure it always stays with the range 0 to MAX_NP. When a NonPosted Request is transmitted, NP_PENDING decrements by one. When all Completions
for an outstanding NP Request are received, NP_PENDING increments by one.
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
191
Appendix E: Managing Receive-Buffer Space for Inbound Completions
Although this method is the simplest to implement, it potentially wastes the most receiver
space because an entire Max_Request_Size block of Completion credit is allocated for each
Non-Posted Request, regardless of actual request size. The amount of waste becomes
greater when the User Application issues a larger proportion of short Memory Reads (on
the order of a single DWORD), I/O Reads and I/O Writes.
The PACKET_FC Method
The PACKET_FC method allocates blocks of credit in finer granularities than LIMIT_FC,
using the receive Completion space more efficiently with a small increase in user logic.
Start with two registers, CPLH_PENDING and CPLD_PENDING, (loaded with zero at
reset), and then perform these steps:
1.
When the User Application needs to send an NP request determine the potential
number of CplH and CplD credits, it might require:
NP_CplH = ceiling[((Start_Address mod RCB) + Request_Size) / RCB]
NP_CplD = ceiling[((Start_Address mod 16 bytes) + Request_Size) /16 bytes]
(except I/O Write, which returns zero data)
The modulo and ceiling functions ensure that any fractional RCB or credit blocks are
rounded up. For example, if a Memory Read requests 8 bytes of data from address
7Ch, the returned data can potentially be returned over two Completion packets (7Ch7Fh, followed by 80h-83h). This would require two RCB blocks and two data credits.
2.
Check the following:
CPLH_PENDING + NP_CplH < Total_CplH (from Table E-1)
CPLD_PENDING + NP_CplD < Total_CplD (from Table E-1)
3.
If both inequalities are true, transmit the Non-Posted Request, increase
CPLH_PENDING by NP_CplH and CPLD_PENDING by NP_CplD. For each NP
Request transmitted, keep NP_CplH and NP_CplD for later use.
4.
When all Completion data is returned for an NP Request, decrement
CPLH_PENDING and CPLD_PENDING accordingly.
This method is less wasteful than LIMIT_FC but still ties up all of an NP Request’s
Completion space until the entire request is satisfied. RCB_FC and DATA_FC provide finer
de-allocation granularity at the expense of more logic.
The RCB_FC Method
The RCB_FC method allocates and de-allocates blocks of credit in RCB granularity. Credit
is freed on a per-RCB basis.
As with PACKET_FC, start with two registers, CPLH_PENDING and CPLD_PENDING
(loaded with zero at reset).
1.
Calculate the number of data credits per RCB:
CplD_PER_RCB = RCB / 16 bytes
2.
When the User Application needs to send an NP request, determine the potential
number of CplH credits it might require. Use this to allocate CplD credits with RCB
granularity:
NP_CplH = ceiling[((Start_Address mod RCB) + Request_Size) / RCB]
NP_CplD = NP_CplH × CplD_PER_RCB
192
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Methods of Managing Completion Space
3.
Check the following:
CPLH_PENDING + NP_CplH < Total_CplH
CPLD_PENDING + NP_CplD < Total_CplD
4.
If both inequalities are true, transmit the Non-Posted Request, increase
CPLH_PENDING by NP_CplH and CPLD_PENDING by NP_CplD.
5.
At the start of each incoming Completion, or when that Completion begins at or
crosses an RCB without ending at that RCB, decrement CPLH_PENDING by 1 and
CPLD_PENDING by CplD_PER_RCB. Any Completion can cross more than one RCB.
The number of RCB crossings can be calculated by:
RCB_CROSSED = ceiling[((Lower_Address mod RCB) + Length) / RCB]
Lower_Address and Length are fields that can be parsed from the Completion header.
Alternatively, a designer can load a register CUR_ADDR with Lower_Address at the
start of each incoming Completion, increment per DW or QW as appropriate, then
count an RCB whenever CUR_ADDR rolls over.
This method is less wasteful than PACKET_FC but still gives us an RCB granularity. If a
User Application transmits I/O requests, the User Application could adopt a policy of only
allocating one CplD credit for each I/O Read and zero CplD credits for each I/O Write. The
User Application would have to match each incoming Completion’s Tag with the Type
(Memory Write, I/O Read, I/O Write) of the original NP Request.
The DATA_FC Method
The DATA_FC method provides the finest allocation granularity at the expense of logic.
As with PACKET_FC and RCB_FC, start with two registers, CPLH_PENDING and
CPLD_PENDING (loaded with zero at reset).
1.
When the User Application needs to send an NP request, determine the potential
number of CplH and CplD credits it might require:
NP_CplH = ceiling[((Start_Address mod RCB) + Request_Size) / RCB]
NP_CplD = ceiling[((Start_Address mod 16 bytes) + Request_Size) / 16 bytes]
(except I/O Write, which returns zero data)
2.
Check the following:
CPLH_PENDING + NP_CplH < Total_CplH
CPLD_PENDING + NP_CplD < Total_CplD
3.
If both inequalities are true, transmit the Non-Posted Request, increase
CPLH_PENDING by NP_CplH and CPLD_PENDING by NP_CplD.
4.
At the start of each incoming Completion, or when that Completion begins at or
crosses an RCB without ending at that RCB, decrement CPLH_PENDING by 1. The
number of RCB crossings can be calculated by:
RCB_CROSSED = ceiling[((Lower_Address mod RCB) + Length) / RCB]
Lower_Address and Length are fields that can be parsed from the Completion header.
Alternatively, a designer can load a register CUR_ADDR with Lower_Address at the
start of each incoming Completion, increment per DW or QW as appropriate, then
count an RCB whenever CUR_ADDR rolls over.
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
193
Appendix E: Managing Receive-Buffer Space for Inbound Completions
5.
At the start of each incoming Completion, or when that Completion begins at or
crosses at a naturally aligned credit boundary, decrement CPLD_PENDING by 1. The
number of credit-boundary crossings is given by:
DATA_CROSSED = ceiling[((Lower_Address mod 16 B) + Length) / 16 B]
Alternatively, a designer can load a register CUR_ADDR with Lower_Address at the
start of each incoming Completion, increment per DW or QW as appropriate, then
count an RCB whenever CUR_ADDR rolls over each 16-byte address boundary.
This method is the least wasteful but requires the greatest amount of user logic. If even
finer granularity is desired, the user can scale the Total_CplD value by 2 or 4 to get the
number of Completion QWORDs or DWORDs, respectively, and adjust the data
calculations accordingly.
194
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Appendix F
Board Design Guidelines
Overview
This appendix discusses topics related to implementing a PCI Express® design that uses
the Spartan®-6 FPGA on a printed circuit board (PCB). Optimal performance requires an
understanding of the functionality of the device pins and needs to address issues such as
device interfacing, protocol specifications, and signal integrity.
Recommendations made in this chapter are guidelines and do not guarantee a working
design.
The information presented here discusses PCB considerations specific to the PCI Express
specifications. This chapter should be used in conjuction with these documents for a
comprehensive understanding of PCB design with Xilinx FPGAs.
•
UG386, Spartan-6 FPGA GTP Transceivers User Guide - Specifically, see the “Board
Design Guidelines” chapter.
•
UG393, Spartan-6 FPGA PCB Design Guide.
The PCI-SIG maintains multiple specifications that can apply depending on the form factor
of the design. This document only considers the subset of these specifications focused on
chip-to-chip and add-in card implementations. Table F-1 shows the specifications that
correlate to the applicable form factors.
Table F-1:
PCI-SIG Specifications and Form Factor
Specification Name
Form-factor
PCI Express Base Specification Revision 1.1
Chip-to-chip on a single PCB
PCI Express Card Electromechanical
Specification (CEM) Revision 1.1
ATX: desktop/server consisting of System
card and Add-in card
Example PCB Reference
Xilinx delivers the SP605 board with an x1 PCI Express add-in card connection. This
chapter uses this board as an example for certain recommendations.
For documentation such as schematics, gerbers, and a bill-of-material for the SP605 board,
see the Spartan-6 FPGA SP605 Evaluation Kit product page:
www.xilinx.com /sp605
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
195
Appendix F: Board Design Guidelines
Board Stackup
Board stackup design is dependent on many variables, including design, manufacturing,
and cost constraints. See the information onboard stackup design in UG393 and UG386.
Generally speaking, signal layers for high-speed signals such as PCI Express data signals
should be sandwiched between ground planes. It is also preferable to use the layers closest
to the top or bottom of the device so that via stubs are minimized.
SP605 Example
Figure F-1 shows the stackup that the SP605 Add-in Card reference board employs. All
internal signal layers are sandwiched between (uninterrupted) ground and power planes.
X-Ref Target - Figure F-1
TOP SIDE
LAYER 1 TOP
PRE-PREG
LAYER 2 PWR1
CORE
LAYER 3 SIG1
PRE-PREG
LAYER 4 GND1
CORE
LAYER 5 SIG2
PRE-PREG
LAYER 6 GND2
CORE
LAYER 7 PWR2
PRE-PREG
LAYER 8 PWR3
CORE
LAYER 9 GND3
PRE-PREG
LAYER 10 SIG3
CORE
LAYER 11 GND4
PRE-PREG
LAYER 12 SIG4
CORE
LAYER 13 GND5
PRE-PREG
LAYER 14 BOT
Spartan-6 FPGA
and
PCI Express Edge Connector Side B (RX)
located on top
PCI Express Edge Connector Side A
located on bottom
BOTTOM SIDE
Figure F-1:
SP605 Board Stackup
Transmit (TX) data lines initiate from the FPGA on the top layer, immediately drop to SIG3
(Layer 10) for routing across the PCB, and then terminate at the PCI Express edge
connector side A on the bottom layer.
196
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Data Routing Guidelines
Receive (RX) data lines initiate from the FPGA on the top layer, immediately drop to SIG3
(Layer 10) for routing across the PCB, and then terminate at the PCI Express edge
connector side B on the top layer.
X-Ref Target - Figure F-2
Edge
Connector
Side B
FPGA
Signal Layer 10 Receive
Signal Layer 10 Transmit
Edge
Connector
Side A
TX capacitors
Figure F-2:
Transmit and Receive Data Lines
Power Supply Design
UG393 discusses general Power Distribution System (PDS) design for the FPGA, including
the required decoupling capacitors for the VCCINT, VCCO, and VCCAUX supplies.
It is also imperative to ensure a clean power supply on MGTAVCC and MGTAVTT power
supplies. Consult UG386 for more details on GTP transceiver power supply layout and
other requirements for filtering and design.
Data Routing Guidelines
Breakout from FPGA BGA
UG386 discusses how to break out the high-speed GTP transceiver signals from the BGA
and provides examples of such. Design constraints might require microstrips for the BGA
exit path or from via to the PCI Express edge connector launch or SMT pads. In such cases,
the microstrip trace must be kept as short as possible.
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
197
Appendix F: Board Design Guidelines
An example Receive and Transmit breakout pattern from the SP605 board are shown in
Figure F-3. Transmit lines are shown in green, and receive lines are shown in red.
X-Ref Target - Figure F-3
Via
Breakout Trace
BGA Pin
Microstrip Trace
Figure F-3:
Receive Breakout Pattern
Microstrip vs. Stripline
Striplines are to be used whenever possible, as are the uppermost and lowermost stripline
layers to minimize via stubs. When the stackup is being planned, these layers should be
placed as close to the top and bottom layers whenever possible.
Plane Reference and Splits
Ground planes should be used as reference planes for signals, as opposed to noisier power
planes. Each reference plane should be contiguous for the length of the trace, because
routing over plane-splits creates an impedance discontinuity. In this case, the impedance of
the trace changes because its coupling to the reference plane is changed abruptly at the
plane split.
Bends
Follow the recommendations in UG393 regarding microstrip and stripline bends. Tight
bends (such as 90 degrees) should be avoided; only mitered, 45-degree or less, bends are
recommended.
198
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Data Routing Guidelines
Propagation Delay
PCI Express generally does not specify a maximum propagation delay for data signals,
with the exception of add-in cards. Add-in card designs should meet the propagation
delay specification in the CEM specification for data traces. The delay from the edge finger
to the GTP transceiver must not exceed 750 ps.
Intrapair Skew
Intrapair skew refers to the skew between a P and N leg of a differential pair. Skew can
introduce common-mode effects which lead to increased EMI, crosstalk and other DC
effects. It is important to match the skew for differential pairs as close as possible.
Xilinx recommends intrapair trace length-matching to within 5 mils to minimize these
effects.
Symmetrical Routing
Always use symmetrical routing. This prevents common-mode effects, such as EMI, from
being introduced into the system.
Figure F-4 illustrates two examples of non-symmetrical routing, which should be avoided.
X-Ref Target - Figure F-4
Figure F-4:
Non-Symmetrical Routing Examples
Vias
Users should follow the recommendations in UG393 for differential vias. Specifically,
wherever high-speed signals must transition signal layers, a Ground-Signal-SignalGround (GSSG) type via should be used if possible. This provides a low inductance return
current path.
All vias for a differential pair should employ symmetrical routing rules.
Trace Impedance
Differential data-line trace impedance was not specified in the Rev 1.0, 1.0a, or 1.1 (1.x) of
the PCI Express Base and PCI Express CEM Specifications. The transmitters and receivers
were specified to have 100Ω nominal differential impedance; therefore, most 1.x designs
opt for a default 100Ω differential trace impedance for all PCI Express differential
connections.
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
199
Appendix F: Board Design Guidelines
Xilinx recommends using simulation techniques to determine the optimum trace
impedance. Simulation using HSPICE or Hyperlinx can help determine the optimum trace
impedance to reduce signal loss.
PCB dielectric material, board stack up, microstrip, and strip line traces affect signal
impedence. It is important that all of these factors are taken into consideration together.
If a simulator is not available, Xilinx recommends these basic guidelines for differential
data-line trace impedance targets:
•
100Ω ±10% for 2.5 Gb/s only links
Trace Separation
Generally, simulation or post-layout analysis tools should be used to determine the
optimum spacing required to reduce crosstalk from nearby aggressor signals. In the
absence of these tools, Xilinx suggests that spacing between differential pairs and other
non-PCI Express signals should be at least three times the dielectric height above the
reference planes to minimize crosstalk. Exceptions to this are allowed in the break-out area
of the FPGA; however, these sections should be kept as short as possible.
Lane Polarity Inversion
The PCI Express Base Specification (1.x) requires that all PCI Express receivers support
polarity inversion. This gives the PCB designer flexibility to avoid having to cross P and N
lines within a given differential pair.
GTP receivers support lane polarity inversion on a per transceiver basis.
AC Coupling
System and Add-in Cards
AC coupling capacitors should be placed on the TX pairs. Place the capacitors either near
the edge connector or the FPGA—not in the middle of the interconnect.
Chip-to-Chip
AC coupling capacitors can be placed anywhere on the interconnect, except in the very
middle.
General Guidelines
Capacitors for coupled traces should always be located at the same relative place as its
partner, that is, symmetrical routing guidelines apply for differential pairs.
Use 0.1 µF ceramic chip capacitors in the smallest package possible.
Data Signal Termination
No external resistor terminators are required with the exception of a precision 50Ω resistor
connected to the RCAL circuitry for the GTP transceiver column. Make sure the trace
length and geometry to both legs of the resistor are equal. See UG386 for more information.
200
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Reference Clock Considerations
Additional Considerations for Add-In Card Designs
1.
Board thickness for add-in cards should not to exceed 0.062 inches.
2.
Care must be taken when connecting the RX and TX data lines to the edge connector.
The edge connector pin names for the TX and RX data lines as defined in the CEM
specification are named from the view of the system board. That is, the RX (PERxx)
lines are connected to the receiver on the system board and the transmitter on the
add-in card. Similarly the TX (PETxx) lines are connected to the transmitter on the
system board and the receiver on the add-in card. That means the add-in card should
route the edge connector PERxx pins to the transmitter and the PETxx pins to the
receiver on an Endpoint configured FPGA. Figure F-5 illustrates how to connect the
data lines for an add-in card design.
X-Ref Target - Figure F-5
Edge
Connector
System
Motherboard
Add-in Card
TX
PETp1
PETn1
RX
RX
PERp1
PERn1
TX
TX
PETp2
PETn2
RX
PERp2
PERn2
.
.
.
RX
.
.
.
.
.
.
Figure F-5:
TX
.
.
.
.
.
.
Add-In Card Design Connections
Reference Clock Considerations
Jitter
Reference clock jitter has the potential to close both the TX and RX eyes, depending of the
frequency content of the phase jitter. Therefore, it is very important to maintain as clean a
reference clock as possible.
Reduce crosstalk on the REFCLK signal by isolating the clock signal from nearby highspeed traces. Maintain a separation of at least 25 mils from the nearest aggressor signals.
Ensure a clean power supply on MGTAVCC power supply. See UG386 for more details on
GTP transceiver power supply layout and design.
In some cases where the designer has no control over the clock source, it might be desirable
to add a jitter attenuator device.
If an external PLL or jitter attenuator device is used, ensure that it meets the specifications
for PLL bandwidth as defined in the PCI Express Base Specification. The PLL bandwidth
specification is different for 1.x and 2.0 versions of the specification.
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
201
Appendix F: Board Design Guidelines
Trace Impedance
The reference clock should use a 100Ω differential trace impedance.
Termination
The REFCLK signal should be routed to the dedicated reference clock input pins on the
GTP transceiver, and the user design should instantiate an IBUFDS primitive in the user
design. An internal 100Ω differential termination biased to 2/3 MGTAVCC is automatically
included on these input pins when the IBUFDS is used, and no external termination is
required or needed for Spartan-6 devices. This is true for both HSCL and LVDS clocks.
See UG386 for more information on GTP transceiver reference clock termination.
AC Coupling
The REFCLK signal should be AC coupled at the input to the FPGA. Xilinx recommends
0.1 µF ceramic-chip capacitors for this purpose. See UG386 for more information
Fanout
If the reference clock needs to be routed to more than one location, then a dedicated clock
fanout device should be used. Make sure to follow the specifications for the fanout device.
For instance, 100Ω termination might be required on the input to the fanout device.
Figure F-6 shows an example of a clock fanout device used to route the reference clock to
multiple locations. The Spartan-6 FPGA requires no external resistive termination (just AC
coupling capacitors). The fanout device is shown with a single resistor terminator at its
clock input pins.
X-Ref Target - Figure F-6
PCIe Edge Connector
Spartan-6
FPGA
REFCLK+
REFCLK-
Clock
Fanout
Chip
Figure F-6:
Other
Location
Fanout Block Diagram
Sideband PCI Express Signals
PERST#
The PERST# signal must be routed to the FPGA for add-in cards. This 3.3V signal should
be routed to an 3.3 V I/O bank (that is, VCCO connected to 3.3V). If a non-3.3V I/O bank is
used, an external circuit is necessary to interface with the Spartan-6 FPGA inputs. This
external circuit could consist of a level translator such as the ST Micro ST2378E, a resistor
network, or other transistor-based circuit. There is no termination required for this signal,
although the integrated Endpoint Block core implements a pull-up on the input from
within the example UCF file.
202
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Reference Clock Considerations
PRSNT#
The PRSNT# pins should be connected as recommended in the CEM specification. Also see
the SP605 board for an example
Summary Checklist
Table F-2 provides a checklist which summarizes the items discussed in this chapter.
Table F-2:
Board Design Checklist
Item
Board Stackup
Follow guidelines in UG393 and UG386.
Power Supply Design
Follow guidelines in UG393 and UG386.
High-Speed Data Signal Routing
Use stripline routing when possible.
Avoid routing over reference plane splits or voids.
Bends < 45 degrees.
Add-in cards must not exceed 750 ps propagation delay.
Length match intrapair skew to within 3 ps.
Use Ground-Signal-Signal-Ground (GSSG) type vias when possible.
Limit the number of vias.
100Ω differential trace impedance for 2.5 Gb/s data signals.
20 mil trace separation between differential pairs (exception in breakout area).
AC coupling 0.1 µF ceramic chip capacitors on all TX lines.
50Ω precision resistor connected to the RCAL circuit for GTP transceivers (see UG386).
Add-in cards must not exceed 0.062 inches in thickness.
Reference Clock (REFCLK)
100Ω differential trace impedance.
Maintain separation of at least 25 mils from nearby aggressor signals.
Ensure clean power supply on MGTAVCC.
No external termination required at input to FPGA (however, user must instantiate
IBUFDS primitive).
AC coupling 0.1 µF ceramic chip capacitors.
Sideband Signals for Add-In Cards
PERST# routes directly to 3.3V I/O bank. Use external circuitry if routing to non-3.3V
I/O bank.
PRSNT# connects as recommended in CEM specification.
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
203
Appendix F: Board Design Guidelines
204
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Appendix G
PCIE_A1 Port Descriptions
This appendix describes the physical interfaces visible on the Spartan®-6 FPGA integrated
Endpoint block’s software primitive, PCIE_A1.
This appendix contains these sections:
•
Clock and Reset Interface
•
Transaction Layer Interface
•
Block RAM Interface
•
GTP Transceiver Interface
•
Configuration Management Interface
•
Debug Interface Ports
Clock and Reset Interface
Table G-1 defines the ports in the Clock and Reset interface.
Table G-1:
Clock and Reset Interface Port Descriptions
Port
Direction
Clock Domain
CLOCKLOCKED
Input
USERCLK
LOCKED signal from the PLL.
MGTCLK
Input
MGTCLK
PIPE interface clock.
Output
MGTCLK
Received hot reset. When asserted, this output indicates
when an in-band hot reset has been received.
SYSRESETN
Input
NONE
USERCLK
Input
USERCLK
User interface clock.
USERRSTN
Output
USERCLK
User interface reset (active Low). This output should be
used to reset the user design logic (it is asserted when the
integrated Endpoint block is reset).
RECEIVEDHOTRST
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Description
Asynchronous system reset (active Low). When this input
is asserted, the integrated Endpoint block is held in reset
until PLL LOCK; thus it can be used to reset the integrated
Endpoint block.
www.xilinx.com
205
Appendix G: PCIE_A1 Port Descriptions
Transaction Layer Interface
Packets are presented to and received from the integrated Endpoint block’s Transaction
Layer through the Transaction Layer interface. Table G-2 defines the ports in the
Transaction Layer interface.
Table G-2:
Transaction Layer Interface Port Descriptions
Port
Direction
Clock Domain
Description
TRNFCCPLD[11:0]
Output
USERCLK
Completion Data Flow Control Credits. This output contains the
number of Completion Data FC credits for the selected flow
control type.
TRNFCCPLH[7:0]
Output
USERCLK
Completion Header Flow Control Credits. This output contains
the number of Completion Header FC credits for the selected
flow control type.
TRNFCNPD[11:0]
Output
USERCLK
Non-Posted Data Flow Control Credits. This output contains the
number of Non-Posted Data FC credits for the selected flow
control type.
TRNFCNPH[7:0]
Output
USERCLK
Non-Posted Header Flow Control Credits. This output contains
the number of Non-Posted Header FC credits for the selected
flow control type.
TRNFCPD[11:0]
Output
USERCLK
Posted Data Flow Control Credits. This output contains the
number of Posted Data FC credits for the selected flow control
type.
TRNFCPH[7:0]
Output
USERCLK
Posted Header Flow Control Credits. This output contains the
number of Posted Header FC credits for the selected flow control
type.
TRNFCSEL[2:0]
Input
USERCLK
Flow Control Informational Select. This input selects the type of
flow control information presented on the TRNFC* signals. Valid
values are:
000b: Receive buffer available space
001b: Receive credits granted to the link partner
010b: Receive credits consumed
100b: Transmit user credits available
101b: Transmit credit limit
110b: Transmit credits consumed
TRNLNKUPN
206
Output
USERCLK
Transaction Link Up (active Low). This output is asserted when
the core and the connected upstream link partner port are ready
and able to exchange data packets. It is deasserted when the core
and link partner are attempting to establish communication, and
when communication with the link partner is lost due to errors
on the transmission channel. When the core is driven to the Hot
Reset and Link Disable states by the link partner, TRNLNKUPN
is deasserted and all TLPs stored in the core are lost.
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Transaction Layer Interface
Table G-2:
Transaction Layer Interface Port Descriptions (Cont’d)
Port
Direction
Clock Domain
Description
TRNRBARHITN[6:0]
Output
USERCLK
Receive BAR Hit (active Low). This output indicates the BAR(s)
targeted by the current receive transaction:
TRNRBARHITN[0]: BAR0
TRNRBARHITN[1]: BAR1
TRNRBARHITN[2]: BAR2
TRNRBARHITN[3]: BAR3
TRNRBARHITN[4]: BAR4
TRNRBARHITN[5]: BAR5
TRNRBARHITN[6]: Expansion ROM Address
If two BARs are configured into a single 64-bit address, both
corresponding TRNRBARHITN bits are asserted.
TRNRD[31:0]
Output
USERCLK
Receive Data. This bus contains the packet data being received.
Input
USERCLK
Receive Destination Ready (active Low). This input is asserted to
indicate that the user application is ready to accept data on
TRNRD. Simultaneous assertion of TRNRSRCRDYN and
TRNRDSTRDYN marks the successful transfer of data on
TRNRD.
TRNREOFN
Output
USERCLK
Receive End-of-Frame (active Low). When asserted, this output
indicates the end of a packet.
TRNRERRFWDN
Output
USERCLK
Receive Error Forward (active Low). This output marks the
current packet in progress as error-poisoned. It is asserted by the
integrated Endpoint block for the entire length of the packet.
Input
USERCLK
Receive Non-Posted OK (active Low). The user application
asserts this input whenever it is ready to accept a Non-Posted
Request packet. This allows Posted and Completion packets to
bypass Non-Posted packets in the inbound queue if necessitated
by the user application. When the user application approaches a
state where it is unable to service Non-Posted Requests, it must
deassert TRNRNPOKN one clock cycle before the integrated
Endpoint block presents TRNREOFN of the last Non-Posted TLP
the user application can accept.
TRNRSOFN
Output
USERCLK
Receive Start-of-Frame (active Low). When asserted, this output
indicates the start of a packet.
TRNRSRCDSCN
Output
USERCLK
Receive Source Discontinue (active Low). When asserted, this
output indicates that the integrated Endpoint block is aborting
the current packet transfer. It is asserted when the physical link
is going into reset.
TRNRSRCRDYN
Output
USERCLK
Receive Source Ready (active Low). When asserted, this output
indicates that the integrated Endpoint block is presenting valid
data on TRNRD.
TRNTBUFAV[5:0]
Output
USERCLK
Transmit Buffers Available. This output provides the number of
transmit buffers available in the integrated Endpoint block. The
maximum number is 32. Each transmit buffer can accommodate
one TLP up to the supported Maximum Payload Size.
TRNRDSTRDYN
TRNRNPOKN
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
207
Appendix G: PCIE_A1 Port Descriptions
Table G-2:
Transaction Layer Interface Port Descriptions (Cont’d)
Port
Direction
Clock Domain
Description
TRNTCFGGNTN
Input
USERCLK
Transmit Configuration Grant (active Low). The user application
asserts this input in response to TRNTCFGREQN, to allow the
integrated Endpoint block to transmit an internally generated
TLP. If the user does not need to postpone internally generated
TLPs, this signal can be continuously asserted.
TRNTCFGREQN
Output
USERCLK
Transmit Configuration Request (active Low). This output is
asserted when the integrated Endpoint block is ready to transmit
a Configuration Completion or other internally generated TLP.
Input
USERCLK
Transmit Data. This bus contains the packet data to be
transmitted.
Output
USERCLK
Transmit Destination Ready (active Low). When asserted, this
output indicates that the integrated Endpoint block is ready to
accept data on TRNTD. Simultaneous assertion of
TRNTSRCRDYN and TRNTDSTRDYN marks a successful
transfer of data on TRNTD.
Input
USERCLK
Transmit End-of-Frame (active Low). This input signals the end
of a packet.
TRNTERRDROPN
Output
USERCLK
Transmit Error Drop (active Low). When asserted, this output
indicates that the integrated Endpoint block discarded a packet
because of a length violation or, when streaming, data was not
presented on consecutive clock cycles. Length violations only
include packets longer than the supported maximum payload
size and do not include packets whose payload does not match
the payload advertised in the TLP header length field.
TRNTERRFWDN
Input
USERCLK
Transmit Error Forward (active Low). This input marks the
current packet in progress as error-poisoned. If TRNTSTRN is
deasserted, TRNTERRFWDN can be asserted any time between
start of frame (SOF) and end of frame (EOF), inclusive. If
TRNTSTRN is asserted, TRNTERRFWDN can only be asserted at
SOF.
TRNTSOFN
Input
USERCLK
Transmit Start-of-Frame (active Low). When asserted, this input
indicates the start of a packet.
TRNTSRCRDYN
Input
USERCLK
Transmit Source Ready (active Low). When asserted, this input
indicates that the user application is presenting valid data on
TRNTD.
TRNTSTRN
Input
USERCLK
Transmit Streamed (active Low). When asserted, this input
indicates a packet is presented on consecutive clock cycles and
transmission on the link can begin before the entire packet has
been written to the integrated Endpoint block.
TRNTD[31:0]
TRNTDSTRDYN
TRNTEOFN
208
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Block RAM Interface
Block RAM Interface
The Transmit (TX) and Receive (RX) buffers are implemented with block RAM. Table G-3
defines the TX buffer and RX buffer ports for the Block RAM interface.
Table G-3:
Block RAM Interface Port Descriptions
Port
Direction
Clock Domain
Description
MIMRXRADDR[11:0]
Output
USERCLK
RX buffer read address
MIMRXRDATA[34:0]
Input
USERCLK
RX buffer read data
MIMRXREN
Output
USERCLK
RX buffer read enable
MIMRXWADDR[11:0]
Output
USERCLK
RX buffer write address
MIMRXWDATA[34:0]
Output
USERCLK
RX buffer write data
MIMRXWEN
Output
USERCLK
RX buffer write enable
MIMTXRADDR[11:0]
Output
USERCLK
TX buffer read address
MIMTXRDATA[35:0]
Input
USERCLK
TX buffer read data
MIMTXREN
Output
USERCLK
TX buffer read enable
MIMTXWADDR[11:0]
Output
USERCLK
TX buffer write address
MIMTXWDATA[35:0]
Output
USERCLK
TX buffer write data
MIMTXWEN
Output
USERCLK
TX buffer write enable
GTP Transceiver Interface
Table G-4 defines the PIPE per Lane ports within the GTP Transceiver interface. There are
two copies of the PIPE per Lane ports, one for each port (n = A or B). Depending on which
GTP transceiver is used, the LogiCORE IP core for PCI Express selects the correct port to
use for the design.
Table G-4:
PIPE per Lane Port Descriptions for the GTP Transceiver Interface
Port
Direction Clock Domain
Description
PIPEGTRESETDONEn
Input
MGTCLK
When asserted, this input indicates that the GTP
transceiver has finished reset and is ready for use.
PIPEPHYSTATUSn
Input
MGTCLK
PIPEPHYSTATUSn is asserted for a single cycle to
indicate completion of GTP transceiver functions such
as Power Management state transitions and receiver
detection on lane n.
PIPERXCHARISKn[1:0]
Input
MGTCLK
This output defines the control bit(s) for received data:
0b: Data byte
1b: Control byte
The lower bit corresponds to the lower byte of
PIPERXDATAn[7:0] while the upper bit describes of
PIPERXDATAn[15:8].
PIPERXDATAn[15:0]
Input
MGTCLK
This input contains the received data.
PIPERXENTERELECIDLEn
Input
MGTCLK
This input indicates an electrical idle on the Receiver.
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
209
Appendix G: PCIE_A1 Port Descriptions
Table G-4:
PIPE per Lane Port Descriptions for the GTP Transceiver Interface (Cont’d)
Port
Direction Clock Domain
Description
PIPERXPOLARITYn
Output
MGTCLK
When High, this output instructs the GTP transceiver to
invert polarity (on the RX differential pair).
PIPERXRESETn
Output
MGTCLK
When asserted, this output resets the receive portion of
the GTP transceiver.
Input
MGTCLK
This input encodes the receiver status and error codes
for the received data stream and receiver detection on
lane n:
PIPERXSTATUSn[2:0]
000b: Data received OK
001b: Reserved
010b: Reserved
011b: Receiver Detected
100b: 8B/10B decode error
101b: Elastic Buffer overflow
110b: Elastic Buffer underflow
111b: Receive disparity error
PIPETXCHARDISPMODEn[1:0]
Output
MGTCLK
PIPETXCHARDISPMODE and
PIPETXCHARDISPVAL allow the 8B/10B disparity of
outgoing data to be controlled when 8B/10B encoding
is enabled.
PIPETXCHARDISPMODE[1] corresponds to
TXDATA[15:8] and PIPETXCHARDISPMODE[0]
corresponds to PIPETXDATA[7:0].
For PCI Express operation, PIPETXCHARDISPMODE
maps to the PIPE signal TXCOMPLIANCE given that
PIPETXCHARDISPVAL is Low. When
PIPETXCHARDISPMODE is High and
PIPETXCHARDISPVAL is Low, the running disparity
is set to negative. This functionality is used when
transmitting the compliance pattern.
PIPETXCHARDISPVALn[1:0]
Output
MGTCLK
PIPETXCHARDISPMODE and
PIPETXCHARDISPVAL allow the 8B/10B disparity of
outgoing data to be controlled when 8B/10B encoding
is enabled.
TXCHARDISPVAL[1] corresponds to TXDATA[15:8]
and TXCHARDISPVAL[0] corresponds to
TXDATA[7:0].
For PCI Express operation, PIPETXCHARDISPVAL
should always be Low.
PIPETXCHARISKn[1:0]
Output
MGTCLK
This output determines the control bit(s) for received
data:
0b: Data byte
1b: Control byte
The lower bit corresponds to the lower byte of
PIPETXDATAn[7:0] while the upper bit describes
PIPETXDATAn[15:8].
210
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
GTP Transceiver Interface
Table G-4:
PIPE per Lane Port Descriptions for the GTP Transceiver Interface (Cont’d)
Port
Direction Clock Domain
Description
PIPETXDATAn[15:0]
Output
MGTCLK
This output contains the transmit data.
PIPETXELECIDLEn
Output
MGTCLK
This output forces the transmit output to electrical idle
in all power states.
PIPETXPOWERDOWNn[1:0]
Output
MGTCLK
This output is the Power Management signal for the
transmitter for lane n:
00b: P0 (Normal operation)
01b: P0s (Low recovery time power-saving state)
10b: P1 (Longer recovery time power state)
11b: Reserved
PIPETXRCVRDETn
Output
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
MGTCLK
When asserted, this output enables the GTP transceiver
to begin either a receiver detection operation or
loopback.
www.xilinx.com
211
Appendix G: PCIE_A1 Port Descriptions
Configuration Management Interface
The Configuration Management Interface contains these signal groupings:
•
Management Interface Ports
•
Error Reporting Ports
•
Interrupt Generation and Status Ports
•
Power Management Ports
•
Configuration Specific Register Ports
•
Miscellaneous Configuration Management Ports
Management Interface Ports
Table G-5 defines the Management Interface ports within the Configuration Management
interface. These ports are used when reading and writing the Configuration Space
Registers.
Table G-5:
Management Interface Port Descriptions
Port
Direction
Clock Domain
Description
Output
USERCLK
Management Data Out. This 32-bit data output obtains
read data from the configuration space inside the
integrated Endpoint block.
CFGDWADDR[9:0]
Input
USERCLK
Management DWORD Address. This 10-bit address
input provides a configuration register DWORD
address during configuration register accesses.
CFGRDENN
Input
USERCLK
Management Read Enable (active Low). This input is
the read-enable for configuration register accesses.
Output
USERCLK
Management Read or Write Done (active Low). The
read-write done signal indicates successful completion
of the user configuration register access operation. For a
user configuration register read operation, this signal
validates the value of the CFGDO[31:0] data bus. The
integrated Endpoint block does not support write
operations.
CFGDO[31:0]
CFGRDWRDONEN
Error Reporting Ports
Table G-6 defines the Error Reporting ports within the Configuration Management
interface.
Table G-6:
Error Reporting Port Descriptions
Port
Direction Clock Domain
Description
CFGERRCORN
Input
USERCLK
Configuration Error Correctable Error (active Low).
The user asserts this signal to report a Correctable
Error.
CFGERRCPLABORTN
Input
USERCLK
Configuration Error Completion Aborted (active
Low). The user asserts this signal to report a
completion was aborted. This signal is ignored if
CFGERRCPLRDYN is deasserted.
212
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Configuration Management Interface
Table G-6:
Error Reporting Port Descriptions (Cont’d)
Port
CFGERRCPLRDYN
Direction Clock Domain
Description
Output
USERCLK
Configuration Error Completion Ready (active Low).
When asserted, this signal indicates that the core can
accept assertions on CFGERRURN and
CFGERRCPLABORTN for Non-Posted Transactions.
Assertions on CFGERRURN and
CFGERRCPLABORTN are ignored when
CFGERRCPLRDYN is deasserted.
CFGERRCPLTIMEOUTN
Input
USERCLK
Configuration Error Completion Time-out (active
Low). The user asserts this signal to report a
completion timed out.
CFGERRECRCN
Input
USERCLK
ECRC Error Report (active Low). The user asserts this
signal to report an end-to-end CRC (ECRC) error.
CFGERRLOCKEDN
Input
USERCLK
Configuration Error Locked (active Low). This input is
used to further qualify the CFGERRURN or
CFGERRCPLABORTN input signal. When this input
is asserted concurrently with one of those two signals,
it indicates that the transaction that caused the error
was an MRdLk transaction and not an MRd. The
integrated Endpoint block generates a CplLk instead
of a Cpl if the appropriate response is to send a
Completion.
CFGERRPOSTEDN
Input
USERCLK
Configuration Error Posted (active Low). This input is
used to further qualify any of the CFGERR* input
signals. When this input is asserted concurrently with
one of the other signals, it indicates that the
transaction that caused the error was a posted
transaction.
CFGERRTLPCPLHEADER[47:0]
Input
USERCLK
Configuration Error TLP Completion Header. This
48-bit input accepts the header information from the
user when an error is signaled. This information is
required so that the integrated Endpoint block can
issue a correct completion, if required.
This information should be extracted from the
received error TLP and presented in the listed format:
[47:41] Lower Address
[40:29] Byte Count
[28:26] TC
[25:24] Attr
[23:8] Requester ID
[7:0] Tag
CFGERRURN
Input
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
USERCLK
Configuration Error Unsupported Request (active
Low). The user asserts this signal to report that an
Unsupported Request (UR) was received. This signal
is ignored if CFGERRCPLRDYN is deasserted.
www.xilinx.com
213
Appendix G: PCIE_A1 Port Descriptions
Interrupt Generation and Status Ports
Table G-7 defines the Interrupt Generation and Status ports within the Configuration
Management interface.
Table G-7:
Interrupt Generation and Status Port Descriptions
Port
CFGINTERRUPTASSERTN
Direction
Clock Domain
Input
USERCLK
Description
Configuration Legacy Interrupt
Assert/Deassert Select. This input selects
between Assert and Deassert messages for
Legacy interrupts when CFGINTERRUPTN is
asserted. It is not used for MSI interrupts.
Value Message Type:
0b: Assert
1b: Deassert
CFGINTERRUPTDI[7:0]
Input
USERCLK
Configuration Interrupt Data In. For Message
Signaling Interrupts (MSI), this input provides
the portion of the Message Data that the
Endpoint must drive to indicate MSI vector
number, if Multi-Vector Interrupts are enabled.
The value indicated by
CFGINTERRUPTMMENABLE[2:0] determines
the number of lower-order bits of Message Data
that the Endpoint provides; the remaining upper
bits of CFGINTERRUPTDI[7:0] are not used.
For Single-Vector Interrupts,
CFGINTERRUPTDI[7:0] is not used.
For Legacy Interrupt Messages (ASSERTINTX,
DEASSERTINTX), this input indicates which
message type is sent, where Value Legacy
Interrupt is:
00h: INTA
01h: INTB
02h: INTC
03h: INTD
CFGINTERRUPTDO[7:0]
Output
USERCLK
Configuration Interrupt Data Out. This output is
the value of the lowest eight bits of the Message
Data field in the Endpoint’s MSI capability
structure. This value is used in conjunction with
CFGINTERRUPTMMENABLE[2:0] to drive
CFGINTERRUPTDI[7:0].
CFGINTERRUPTMMENABLE[2:0]
Output
USERCLK
Configuration Interrupt Multiple Message
Enabled. This output has the value of the
Multiple Message Enable field, where values
range from 000b to 101b. A value of 000b
indicates that single vector MSI is enabled.
Other values indicate the number of bits that can
be used for multi-vector MSI.
214
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Configuration Management Interface
Table G-7:
Interrupt Generation and Status Port Descriptions (Cont’d)
Port
CFGINTERRUPTMSIENABLE
Direction
Clock Domain
Output
USERCLK
Description
Configuration Interrupt MSI Enabled.
0: Only Legacy (INTx) interrupts can be sent
1: The Message Signaling Interrupt (MSI)
messaging is enabled
CFGINTERRUPTN
CFGINTERRUPTRDYN
Input
USERCLK
Configuration Interrupt Request (active Low).
When asserted, this input causes the selected
interrupt message type to be transmitted by the
integrated Endpoint block. The signal should be
asserted until CFGINTERRUPTRDYN is
asserted.
Output
USERCLK
Configuration Interrupt Ready (active Low).
This output is the interrupt grant signal. The
simultaneous assertion of
CFGINTERRUPTRDYN and
CFGINTERRUPTN indicates that the integrated
Endpoint block has successfully transmitted the
requested interrupt message.
Power Management Ports
Table G-8 defines the Power Management ports within the Configuration Management
interface.
Table G-8:
Power Management Port Descriptions
Port
Direction
Clock Domain
Description
Input
USERCLK
Send PMPME Message (active Low). A one-clock
cycle assertion of this input signals the integrated
Endpoint block to send a Power Management
Wake Event (PMPME) Message TLP to the
upstream link partner.
CFGTOTURNOFFN
Output
USERCLK
Configuration To Turnoff: This output signal
notifies the user that a PME_TURN_Off message
has been received, and the Configuration and
Capabilities Module (CCM) starts polling the
CFGTURNOFFOKN input coming in from the
user. When CFGTURNOFFOKN is asserted, the
CMM sends a PME_To_Ack message to the
upstream device.
CFGTURNOFFOKN
Input
USERCLK
Configuration Turnoff OK (active Low). This input
is the power turn-off ready signal. The user
application can assert this input to notify the
Endpoint that it is safe for power to be turned off.
CFGPMWAKEN
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
215
Appendix G: PCIE_A1 Port Descriptions
Configuration Specific Register Ports
Table G-9 defines the Configuration Specific Register ports within the Configuration
Management interface. These ports directly mirror the contents of commonly used
registers located within the PCI Express Configuration Space.
Table G-9:
Configuration Specific Register Port Descriptions
Port
CFGCOMMANDBUSMASTERENABLE
Direction Clock Domain
Output
USERCLK
Description
Configuration Command, Bus Master
Enable, Command[2]. The integrated
Endpoint block takes no action based
on this setting; the user logic must.
When this output is asserted, the user
logic is allowed to issue Memory or I/O
Requests (including MSI interrupts);
otherwise, the user logic must not issue
those requests.
CFGCOMMANDINTERRUPTDISABLE
Output
USERCLK
Configuration Command, Interrupt
Disable, Command[10]. When this
output is asserted, the integrated
Endpoint block is prevented from
asserting INTx interrupts.
CFGCOMMANDIOENABLE
Output
USERCLK
Configuration Command, I/O Space
Enable, Command[0].
0: The integrated Endpoint block
filters these accesses and responds
with a UR.
1: Allows the device to receive I/O
Space accesses.
CFGCOMMANDMEMENABLE
Output
USERCLK
Configuration Command, Memory
Space Enable, Command[1].
0: The integrated Endpoint block
filters these accesses and responds
with a UR.
1: Allows the device to receive
Memory Space accesses.
CFGCOMMANDSERREN
Output
USERCLK
Configuration Command, SERR Enable
(active Low), Command[8].
When this output is asserted, reporting
of Non-fatal and Fatal errors is enabled.
If enabled, errors are reported either
through this bit or through the PCI
Express specific bits in the Device
Control Register.
CFGDEVCONTROLAUXPOWEREN
Output
USERCLK
Not used.
CFGDEVCONTROLCORRERRREPORTINGEN
Output
USERCLK
Configuration Device Control,
Correctable Error Reporting Enable,
DEVICECTRL[0]. This bit, in
conjunction with other bits, controls
sending ERRCOR messages.
216
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Configuration Management Interface
Table G-9:
Configuration Specific Register Port Descriptions (Cont’d)
Port
Direction Clock Domain
Description
CFGDEVCONTROLENABLERO
Output
USERCLK
Configuration Device Control, Enable
Relaxed Ordering, DEVICECTRL[4].
When this output is asserted, the user
logic is permitted to set the Relaxed
Ordering bit in the Attributes field of
transactions it initiates that do not
require strong write ordering.
CFGDEVCONTROLEXTTAGEN
Output
USERCLK
Configuration Device Control, Tag
Field Enable, DEVICECTRL[8]. When
this output is asserted, the user logic
can use an 8-bit Tag field as a Requester.
When this output is deasserted, the user
logic is restricted to a 5-bit Tag field. The
integrated Endpoint block does not
enforce the number of Tag bits used,
either in outgoing request TLPs or
incoming Completions.
CFGDEVCONTROLFATALERRREPORTINGEN
Output
USERCLK
Configuration Device Control, Fatal
Error Reporting Enable,
DEVICECTRL[2]. This bit, in
conjunction with other bits, controls
sending ERRFATAL messages.
CFGDEVCONTROLMAXPAYLOAD[2:0]
Output
USERCLK
Configuration Device Control,
MAXPAYLOADSIZE,
DEVICECTRL[7:5]. This field sets the
maximum TLP payload size. As a
Receiver, the user logic must handle
TLPs as large as the set value. As a
Transmitter, the user logic must not
generate TLPs exceeding the set value.
000b: 128-byte maximum payload
size
001b: 256-byte maximum payload
size
010b: 512-byte maximum payload
size
CFGDEVCONTROLMAXREADREQ[2:0]
Output
USERCLK
Configuration Device Control,
MAXREADREQUESTSIZE,
DEVICECTRL[14:12]. This field sets the
maximum Read Request size for the
user logic as a Requester. The user logic
must not generate Read Requests with
size exceeding the set value.
000b: 128-byte maximum Read
Request size
001b: 256-byte maximum Read
Request size
010b: 512-byte maximum Read
Request size
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
217
Appendix G: PCIE_A1 Port Descriptions
Table G-9:
Configuration Specific Register Port Descriptions (Cont’d)
Port
Direction Clock Domain
Description
CFGDEVCONTROLNONFATALREPORTINGEN
Output
USERCLK
Configuration Device Control, NonFatal Error Reporting Enable,
DEVICECTRL[1]. This bit, in
conjunction with other bits, controls
sending ERRNONFATAL messages.
CFGDEVCONTROLNOSNOOPEN
Output
USERCLK
Configuration Device Control, Enable
No Snoop, DEVICECTRL[11]. When
this output is asserted, the user logic is
permitted to set the No Snoop bit in
TLPs it initiates that do not require
hardware-enforced cache coherency.
CFGDEVCONTROLPHANTOMEN
Output
USERCLK
Configuration Device Control,
Phantom Functions Enable,
DEVICECTRL[9]. When this output is
asserted, the user logic can use
unclaimed Functions as Phantom
Functions to extend the number of
outstanding transaction identifiers. If
this output is deasserted, the user logic
is not allowed to use Phantom
Functions.
CFGDEVCONTROLURERRREPORTINGEN
Output
USERCLK
Configuration Device Control, UR
Reporting Enable, DEVICECTRL[3].
This bit, in conjunction with other bits,
controls the signaling of URs by
sending Error messages.
CFGDEVSTATUSCORRERRDETECTED
Output
USERCLK
Configuration Device Status,
Correctable Error Detected,
DEVICESTATUS[0]. This output
indicates the status of correctable errors
detected. Errors are logged in this
register regardless of whether error
reporting is enabled or not in the Device
Control Register.
CFGDEVSTATUSFATALERRDETECTED
Output
USERCLK
Configuration Device Status, Fatal
Error Detected, DEVICESTATUS[2].
This output indicates the status of Fatal
errors detected. Errors are logged in this
register regardless of whether error
reporting is enabled or not in the Device
Control Register.
CFGDEVSTATUSNONFATALERRDETECTED
Output
USERCLK
Configuration Device Status, Non-Fatal
Error Detected, DEVICESTATUS[1].
This output indicates the status of Nonfatal errors detected. Errors are logged
in this register regardless of whether
error reporting is enabled or not in the
Device Control Register.
218
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Configuration Management Interface
Table G-9:
Configuration Specific Register Port Descriptions (Cont’d)
Port
Direction Clock Domain
Description
CFGDEVSTATUSURDETECTED
Output
USERCLK
Configuration Device Status,
Unsupported Request Detected,
DEVICESTATUS[3]. This output
indicates that the integrated Endpoint
block received a UR. Errors are logged
in this register regardless of whether
error reporting is enabled or not in the
Device Control Register.
CFGLINKCONTROLASPMCONTROL[1:0]
Output
USERCLK
Configuration Link Control, ASPM
Control, LINKCTRL[1:0]. This 2-bit
output indicates the level of ASPM
supported, where:
00b: Disabled
01b: L0s Entry Enabled
10b: Not used
11b: Not used
CFGLINKCONTROLCOMMONCLOCK
Output
USERCLK
Configuration Link Control, Common
Clock Configuration, LINKCTRL[6].
When this output is asserted, this
component and the component at the
opposite end of this Link are operating
with a distributed common reference
clock. When this output is deasserted,
the components are operating with an
asynchronous reference clock.
CFGLINKCONTROLEXTENDEDSYNC
Output
USERCLK
Configuration Link Control, Extended
Synch, LINKCTRL[7]. When this
output is asserted, the transmission of
additional Ordered Sets is forced when
exiting the L0s state and when in the
Recovery state.
CFGLINKCONTROLRCB
Output
USERCLK
Configuration Link Control, RCB,
LINKCTRL[3]. This output indicates
the Read Completion Boundary value,
where:
0: 64B
1: 128B
CFGTRNPENDINGN
Input
USERCLK
User Transaction Pending (active Low).
When asserted, this input sets the
Transactions Pending bit in the Device
Status Register (DEVICESTATUS[5]).
Note: The user is required to assert this
input if the User Application has not
received a completion to a request.
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
219
Appendix G: PCIE_A1 Port Descriptions
Miscellaneous Configuration Management Ports
Table G-10 defines the Miscellaneous ports within the Configuration Management
interface.
Table G-10:
Miscellaneous Configuration Management Port Descriptions
Port
Direction Clock Domain
Description
CFGBUSNUMBER[7:0]
Output
USERCLK
Configuration Bus Number. This 8-bit output provides the
assigned bus number for the device. The user application
must use this information in the Bus Number field of
outgoing TLP requests. The default value after reset is
00h. This output is refreshed whenever a Type 0
Configuration Write packet is received.
CFGDEVICENUMBER[4:0]
Output
USERCLK
Configuration Device Number: This 5-bit output provides
the assigned device number for the device. The user
application must use this information in the Device
Number field of outgoing TLP requests. The default value
after reset is 00000b. This output is refreshed whenever a
Type 0 Configuration Write packet is received.
CFGDEVID[15:0]
Input
USERCLK
Device ID value. This 16-bit input must be stable when
SYSRESETN is deasserted.
CFGDSN[63:0]
Input
USERCLK
Configuration Device Serial Number. This 64-bit input
indicates the value that should be transferred to the Device
Serial Number Capability.
CFGFUNCTIONNUMBER[2:0]
Output
USERCLK
Configuration Function Number. This 3-bit output
provides the function number for the device. The user
application must use this information in the Function
Number field of outgoing TLP requests. The function
number is hardwired to 000b.
220
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Configuration Management Interface
Table G-10:
Miscellaneous Configuration Management Port Descriptions (Cont’d)
Port
CFGLTSSMSTATE[4:0]
Direction Clock Domain
Output
MGTCLK
Description
This 5-bit output is a mirror of the LTSSM state machine
bits:
00000b: Detect.Quiet
00001b: Detect.Active
00010b: Polling.Active
00011b: Polling.Config
00100b: Polling Compliance
00101b: Configuration.Linkwidth.Start
00110b: Configuration.Linkwidth.Start
00111b: Configuration.Linkwidth.Accept
01000b: Configuration.Linkwidth.Accept
01001b: Configuration.Lanenum.Wait
01010b: Configuration.Lanenum.Accept
01011b: Configuration.Complete
01100b: Configuration.Idle
01101b: L0
01110b: L1.Entry
01111b: L1.Entry
10000b: L1.Entry
10001b: L1.Idle
10010b: L1.Exit-to-recovery
10011b: Recovery.RcvrLock
10100b: Recovery.RcvrCfg
10101b: Recovery.Idle
10110b: Hot Reset
10111b: Disabled
11000b: Disabled
11001b: Disabled
11010b: Disabled
11011b: Detect.Quiet
CFGPCIELINKSTATEN[2:0]
Output
USERCLK
PCI Express Link State. This encoded bus reports the PCI
Express Link State Information to the user:
110b: L0 state
101b: L0s state
011b: L1 state
111b: Under transition
CFGREVID[7:0]
Input
USERCLK
Revision ID Value. This input must be stable when
SYSRESETN is deasserted.
CFGSUBSYSID[15:0]
Input
USERCLK
Subsystem ID Value. This input must be stable when
SYSRESETN is deasserted.
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
221
Appendix G: PCIE_A1 Port Descriptions
Table G-10:
Miscellaneous Configuration Management Port Descriptions (Cont’d)
Port
Direction Clock Domain
Description
CFGSUBSYSVENID[15:0]
Input
USERCLK
Subsystem Vendor ID Reset Value. This input must be
stable when SYSRESETN is deasserted.
CFGVENID[15:0]
Input
USERCLK
Vendor ID Value. This input must be stable when
SYSRESETN is deasserted.
Debug Interface Ports
Table G-11 describes the Debug Interface ports.
Table G-11:
Debug Interface Port Descriptions
Port
Direction
Clock Domain
Description
DBGBADDLLPSTATUS
Output
USERCLK
This signal pulses High for one USERCLK cycle
when a DLLP CRC error is detected.
DBGBADTLPLCRC
Output
USERCLK
This signal pulses High for one USERCLK cycle
when a TLP with an LCRC error is detected.
DBGBADTLPSEQNUM
Output
USERCLK
This signal pulses High for one USERCLK cycle
when a TLP with an invalid sequence number
is detected.
DBGBADTLPSTATUS
Output
USERCLK
This signal pulses High for one USERCLK cycle
when a bad TLP is detected, for reasons other
than a bad LCRC or a bad sequence number.
DBGDLPROTOCOLSTATUS
Output
USERCLK
This signal pulses High for one USERCLK cycle
if an out-of-range ACK or NAK is received.
DBGFCPROTOCOLERRSTATUS
Output
USERCLK
This signal pulses High for one USERCLK cycle
if there is a protocol error with the received
flow control updates.
DBGMLFRMDLENGTH
Output
USERCLK
This signal pulses High for one USERCLK cycle
to indicate that a received TLP had a length that
did not match what was in the TLP header.
DBGMLFRMDMPS
Output
USERCLK
This signal pulses High for one USERCLK cycle
to indicate that a received TLP had a length in
violation of the negotiated MPS.
DBGMLFRMDTCVC
Output
USERCLK
This signal pulses High for one USERCLK cycle
to indicate that a received TLP had an invalid
TC or VC value.
DBGMLFRMDTLPSTATUS
Output
USERCLK
This signal pulses High for one USERCLK cycle
when a malformed TLP is received. See the
other DBGMLFRMD* signals for further
clarification.
Note: There is skew between DBGMLFRMD*
and DBGMLFRMDTLPSTATUS.
DBGMLFRMDUNRECTYPE
222
Output
USERCLK
www.xilinx.com
This signal pulses High for one USERCLK cycle
to indicate that a received TLP had an
invalid/unrecognized type field value.
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Debug Interface Ports
Table G-11:
Debug Interface Port Descriptions (Cont’d)
Port
Direction
Clock Domain
Description
DBGPOISTLPSTATUS
Output
USERCLK
This signal pulses High for one USERCLK cycle
if a poisoned TLP is received.
DBGRCVROVERFLOWSTATUS
Output
USERCLK
This signal pulses High for one USERCLK cycle
if a received TLP violates the advertised credit.
DBGREGDETECTEDCORRECTABLE
Output
USERCLK
This signal is a mirror of the internal signal
used to indicate a correctable error is detected.
The error is cleared upon a read by the Root
Complex (RC).
DBGREGDETECTEDFATAL
Output
USERCLK
This signal is a mirror of the internal signal
used to indicate that a fatal error has been
detected. The error is cleared upon a read by
the RC.
DBGREGDETECTEDNONFATAL
Output
USERCLK
This signal is a mirror of the internal signal
used to indicate that a non-fatal error has been
detected. The error is cleared upon a read by
the RC.
DBGREGDETECTEDUNSUPPORTED
Output
USERCLK
This signal is a mirror of the internal signal
used to indicate that an unsupported request
has been detected. The error is cleared upon a
read by the RC.
DBGRPLYROLLOVERSTATUS
Output
USERCLK
This signal pulses High for one USERCLK cycle
when the rollover counter expires.
DBGRPLYTIMEOUTSTATUS
Output
USERCLK
This signal pulses High for one USERCLK cycle
when the replay time-out counter expires.
DBGURNOBARHIT
Output
USERCLK
This signal pulses High for one USERCLK cycle
to indicate that a received read or write request
did not match any configured BAR.
DBGURPOISCFGWR
Output
USERCLK
This signal pulses High for one USERCLK cycle
to indicate that a CfgWr TLP with the
Error/Poisoned bit (EP) = 1 was received.
DBGURSTATUS
Output
USERCLK
This signal pulses High for one USERCLK cycle
when an unsupported request is received. See
the DBGUR* signals for further clarification.
Note: There is skew between DBGUR* and
DBGURSTATUS.
DBGURUNSUPMSG
Output
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
USERCLK
www.xilinx.com
This signal pulses High for one USERCLK cycle
to indicate that an Msg or MsgD TLP with an
unsupported type was received.
223
Appendix G: PCIE_A1 Port Descriptions
224
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Appendix H
PCIE_A1 Attribute Descriptions
Table H-1 defines the attributes on the PCIE_A1 library primitive for the Spartan®-6 FPGA
Integrated Endpoint Block for PCI Express® designs. All attributes are set in the
LogiCORE™ IP; they are documented in this chapter for reference. Users should not
change the attribute settings as set in the CORE Generator™ software GUI for proper
operation of the design.
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
225
Appendix H: PCIE_A1 Attribute Descriptions
Table H-1:
PCIE_A1 Attributes
Attribute Name
BAR0
Type
32-bit
Hex
Description
This attribute specifies the mask/settings for Base
Address Register (BAR) 0. If BAR is not to be
implemented, this attribute is set to 32'h00000000.
Bits are defined as follows:
• Memory Space BAR:
0: Mem Space Indicator (set to 0)
[2:1]: Type field (10 for 64-bit, 00 for 32-bit)
3: Prefetchable (0 or 1)
[31:4]: Mask for writable bits of BAR. For a 32-bit
BAR, the uppermost 31:n bits are set to 1, where
2n = memory aperture size in bytes. For a 64-bit BAR,
the uppermost 63:n bits of {BAR1, BAR0} are set to 1.
• I/O Space BAR:
0: I/O Space Indicator (set to 1)
1: Reserved (set to 0)
[31:2]: Mask for writable bits of BAR. The uppermost
31:n bits are set to 1, where 2n = I/O aperture size in
bytes
BAR1
32-bit
Hex
This attribute specifies the mask/settings for BAR1 if
BAR0 is a 32-bit BAR, or the upper bits of {BAR1, BAR0}
if BAR0 is a 64-bit BAR. If BAR is not to be implemented,
this attribute is set to 32'h00000000. See the BAR0
description if this attribute functions as the upper bits of
a 64-bit BAR. Bits are defined as follows:
• Memory Space BAR (not the upper bits of BAR0):
0: Mem Space Indicator (set to 0)
[2:1]: Type field (10 for 64-bit, 00 for 32-bit)
3: Prefetchable (0 or 1)
[31:4]: Mask for writable bits of BAR. For a 32-bit
BAR, the uppermost 31:n bits are set to 1, where
2n = memory aperture size in bytes. For a 64-bit BAR,
the uppermost 63:n bits of {BAR2, BAR1} are set to 1.
• I/O Space BAR:
0: I/O Space Indicator (set to 1)
1: Reserved (set to 0)
[31:2]: Mask for writable bits of BAR. The uppermost
31:n bits are set to 1, where 2n = I/O aperture size in
bytes
226
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Table H-1:
PCIE_A1 Attributes (Cont’d)
Attribute Name
BAR2
Type
Description
32-bit
Hex
For an Endpoint, this attribute specifies the
mask/settings for BAR2 if BAR1 is a 32-bit BAR, or the
upper bits of {BAR2, BAR1} if BAR1 is the lower part of
a 64-bit BAR. If BAR is not to be implemented, this
attribute is set to 32'h00000000. See the BAR1
description if this attribute functions as the upper bits of
a 64-bit BAR.
For an Endpoint, bits are defined as follows:
• Memory Space BAR (not upper bits of BAR1):
0: Mem Space Indicator (set to 0)
[2:1]: Type field (10 for 64-bit, 00 for 32-bit)
3: Prefetchable (0 or 1)
[31:4]: Mask for writable bits of BAR. For a 32-bit
BAR, the uppermost 31:n bits are set to 1, where
2n = memory aperture size in bytes. For a 64-bit BAR,
the uppermost 63:n bits of {BAR3, BAR2} are set to 1.
• I/O Space BAR:
0: I/O Space Indicator (set to 1)
1: Reserved (set to 0)
[31:2]: Mask for writable bits of BAR. The uppermost
31:n bits are set to 1, where 2n = I/O aperture size in
bytes
BAR3
32-bit
Hex
For an Endpoint, this attribute specifies the
mask/settings for BAR3 if BAR2 is a 32-bit BAR, or the
upper bits of {BAR3, BAR2} if BAR2 is the lower part of
a 64-bit BAR. If BAR is not to be implemented, this
attribute is set to 32'h00000000. See the BAR2
description if this functions as the upper bits of a 64-bit
BAR.
For an Endpoint, bits are defined as follows:
• Memory Space BAR (not upper bits of BAR2):
0: Mem Space Indicator (set to 0)
[2:1]: Type field (10 for 64-bit, 00 for 32-bit)
3: Prefetchable (0 or 1)
[31:4]: Mask for writable bits of BAR. For a 32-bit
BAR, the uppermost 31:n bits are set to 1, where
2n = memory aperture size in bytes. For a 64-bit BAR,
the uppermost 63:n bits of {BAR4, BAR3} are set to 1.
• I/O Space BAR:
0: I/O Space Indicator (set to 1)
1: Reserved (set to 0)
[31:2]: Mask for writable bits of BAR. The uppermost
31:n bits are set to 1, where 2n = I/O aperture size in
bytes
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
227
Appendix H: PCIE_A1 Attribute Descriptions
Table H-1:
PCIE_A1 Attributes (Cont’d)
Attribute Name
BAR4
Type
Description
32-bit
Hex
For an Endpoint, this attribute specifies mask/settings
for Base Address Register (BAR) 4 if BAR3 is a 32-bit
BAR, or the upper bits of {BAR4, BAR3}, if BAR3 is the
lower part of a 64-bit BAR. If BAR is not to be
implemented, this attribute is set to 32'h00000000.
See the BAR3 description if this functions as the upper
bits of a 64-bit BAR.
For an Endpoint, bits are defined as follows:
• Memory Space BAR (not upper bits of BAR3):
0: Mem Space Indicator (set to 0)
[2:1]: Type field (10 for 64-bit, 00 for 32-bit)
3: Prefetchable (0 or 1)
[31:4]: Mask for writable bits of BAR. For a 32-bit
BAR, the uppermost 31:n bits are set to 1, where
2n = memory aperture size in bytes. For a 64-bit BAR,
the uppermost 63:n bits of {BAR5, BAR4} to 1.
• I/O Space BAR:
0: I/O Space Indicator (set to 1)
1: Reserved (set to 0)
[31:2]: Mask for writable bits of BAR. The uppermost
31:n bits are set to 1, where 2n = I/O aperture size in
bytes
BAR5
32-bit
Hex
For an Endpoint, this attribute specifies mask/settings
for BAR5 if BAR4 is a 32-bit BAR or the upper bits of
{BAR5, BAR4} if BAR4 is the lower part of a 64-bit BAR.
If BAR is not to be implemented, this attribute is set to
32'h00000000. See the BAR4 description if this
functions as the upper bits of a 64-bit BAR.
For an Endpoint, bits are defined as follows:
• Memory Space BAR (not upper bits of BAR4):
0: Mem Space Indicator (set to 0)
[2:1]: Type field (00 for 32-bit; BAR5 cannot be the
lower part of a 64-bit BAR)
3: Prefetchable (0 or 1)
[31:4]: Mask for writable bits of BAR. The uppermost
31:n bits are set to 1, where 2n = memory aperture
size in bytes
• I/O Space BAR:
0: I/O Space Indicator (set to 1)
1: Reserved (set to 0)
[31:2]: Mask for writable bits of BAR. The uppermost
31:n bits are set to 1, where 2n = I/O aperture size in
bytes
CARDBUS_CIS_POINTER
228
32-bit
Hex
Pointer to the Cardbus data structure. This value is
transferred to the Cardbus CIS Pointer Register. It is set
to 0 if the Cardbus pointer is not implemented.
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Table H-1:
PCIE_A1 Attributes (Cont’d)
Attribute Name
Type
Description
CLASS_CODE
24-bit
Hex
Code identifying basic function, subclass, and
applicable programming interface. This value is
transferred to the Class Code Register.
DEV_CAP_ENDPOINT_L0S_LATENCY
3-bit
Binary
Endpoint L0s Acceptable Latency. This attribute records
the latency that the Endpoint can withstand on
transitions from the L0s state to the L0 state. Valid
settings are:
000b: Less than 64 ns
001b: 64 ns to 128 ns
010b: 128 ns to 256 ns
011b: 256 ns to 512 ns
100b: 512 ns to 1 µs
101b: 1 µs to 2 µs
110b: 2 µs to 4 µs
111b: More than 4 µs
DEV_CAP_ENDPOINT_L1_LATENCY
3-bit
Binary
Endpoint L1 Acceptable Latency. Records the latency
that the endpoint can withstand on transitions from the
L1 state to the L0 state (if the L1 state is supported).
Valid settings are:
000b: Less than 1 µs
001b: 1 µs to 2 µs
010b: 2 µs to 4 µs
011b: 4 µs to 8 µs
100b: 8 µs to 16 µs
101b: 16 µs to 32 µs
110b: 32 µs to 64 µs
111b: More than 64 µs
DEV_CAP_EXT_TAG_SUPPORTED
Boolean
Extended Tags support.
FALSE: 5-bit tag
TRUE: 8-bit tag
DEV_CAP_MAX_PAYLOAD_SUPPORTED
3-bit
Binary
This attribute specifies the maximum payload
supported. Valid (supported) settings are:
000b: 128 bytes
001b: 256 bytes
010b: 512 bytes
This value is transferred to the Device Capabilities
Register.
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
229
Appendix H: PCIE_A1 Attribute Descriptions
Table H-1:
PCIE_A1 Attributes (Cont’d)
Attribute Name
Type
Description
DEV_CAP_PHANTOM_FUNCTIONS_SUPPOR
T
2-bit
Binary
Phantom Function Support. This attribute indicates the
number of functions re-allocated as Tag bits. Valid
settings are:
00b: 0
01b: 1
10b: 2
11b: 3
DEV_CAP_ROLE_BASED_ERROR
Boolean
When this attribute is set to TRUE, compliant error
reporting is supported.
DISABLE_BAR_FILTERING
Boolean
When this attribute is set to TRUE, BAR filtering is
disabled. This setting does not change the behavior of
the BAR hit outputs.
DISABLE_ID_CHECK
Boolean
When this attribute is set to TRUE, checking for
Requester ID of received completions is disabled.
DISABLE_SCRAMBLING
Boolean
When this attribute is TRUE, Scrambling of transmit
data is turned off.
ENABLE_RX_TD_ECRC_TRIM
Boolean
When this attribute is set to TRUE, received TLPs have
their td bit set to 0 and the ECRC is removed.
EXPANSION_ROM
22-bit
Hex
This attribute specifies the mask/settings for the
Expansion ROM BAR. If the BAR is not to be
implemented, this attribute is set to 22'h00000000.
Bits are defined as follows:
0: Expansion ROM implemented (set to 1 to
implement ROM)
[21:1]: Mask for writable bits of BAR. The uppermost
21:n bits are set to 1, where 2n = ROM aperture size in
bytes
FAST_TRAIN
Boolean
When this attribute is set to TRUE, the timers in the
LTSSM state machine are shortened to reduce
simulation time. Specifically, the transition out of
Polling.Active requires sending 16 TS1s and receiving 8
TS1s. The LTSSM timer values of 1 ms, 2 ms, 12 ms,
24 ms, and 48 ms are reduced to 3.9 µs, 7.81 µs, 46.8 µs,
93.75 µs, and 187.5 µs, respectively (reduced by a factor
of 256). This attribute must be set to FALSE for silicon
designs.
GTP_SEL
Boolean
This attribute indicates which port interface is used:
FALSE: Transceiver A port interface
TRUE: Transceiver B port interface
230
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Table H-1:
PCIE_A1 Attributes (Cont’d)
Attribute Name
LINK_CAP_ASPM_SUPPORT
Type
2-bit
Binary
Description
Active State PM Support. This attribute indicates the
level of active state power management supported by
the selected PCI Express Link:
00b: Reserved
01b: L0s entry supported
10b: Reserved
11b: Reserved
LINK_CAP_L0S_EXIT_LATENCY
3-bit
Binary
This attribute sets the exit latency from the L0s state to
be applied (at 2.5 Gb/s) where a common clock is used.
This value is transferred to the Link Capabilities
Register.
Valid settings are:
000b: Less than 64 ns
001b: 64 ns to less than 128 ns
010b: 128 ns to less than 256 ns
011b: 256 ns to less than 512 ns
100b: 512 ns to less than 1 µs
101b: 1 µs to less than 2 µs
110b: 2 µs to 4 µs
111b: More than 4 µs
LINK_CAP_L1_EXIT_LATENCY
3-bit
Binary
This attribute sets the exit latency from the L1 state to be
applied (at 2.5 Gb/s) where a common clock is used.
This value is transferred to the Link Capabilities
Register.
Valid settings are:
000b: Less than 1 µs
001b: 1 µs to less than 2 µs
010b: 2 µs to less than 4 µs
011b: 4 µs to less than 8 µs
100b: 8 µs to less than 16 µs
101b: 16 µs to less than 32 µs
110b: 32 µs to 64 µs
111b: More than 64 µs
LINK_STATUS_SLOT_CLOCK_CONFIG
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Boolean
Slot Clock Configuration. This attribute indicates where
the component uses the same physical reference clock
that the platform provides on the connector. For a port
that connects to the slot, this attribute indicates that it
uses a clock with a common source to that used by the
slot. For an adaptor inserted in the slot, this attribute
indicates that it uses the same clock source as the slot,
not a locally derived clock source. This value is
transferred to the Link Status Register, bit 12.
www.xilinx.com
231
Appendix H: PCIE_A1 Attribute Descriptions
Table H-1:
PCIE_A1 Attributes (Cont’d)
Attribute Name
LL_ACK_TIMEOUT
LL_ACK_TIMEOUT_EN
Type
Description
15-bit
Hex
This attribute sets an ACK time-out counter override
value. The value is in increments of USERCLK periods.
It should be set to 0 unless the user wishes to override
the default (internal) setting.
Boolean
When set to TRUE, the value specified by
LL_ACK_TIMEOUT is added to the internal value,
increasing the ACK Timeout delay.
When set to FALSE, the value provided on
LL_ACK_TIMEOUT is subtracted from the internal
value, decreasing the ACK Timeout delay
LL_REPLAY_TIMEOUT
LL_REPLAY_TIMEOUT_EN
15-bit
Hex
This attribute sets a replay timer override value. The
value is in increments of USERCLK periods. It should be
set to 0 unless the user wishes to override the default
(internal) setting.
Boolean
When set to TRUE, the value specified by
LL_REPLAY_TIMEOUT is added to the internal value,
increasing the Replay Timeout delay.
When set to FALSE, the value provided on
LL_REPLAY_TIMEOUT is subtracted from the internal
value, decreasing the Replay Timeout delay
MSI_CAP_MULTIMSG_EXTENSION
Boolean
Multiple Message Capable Extension. When set to
TRUE, this attribute allows 256 unique messages to be
sent by the user regardless of the setting of
MSI_CAP_MULTIMSGCAP).
Note: Enabling this feature (TRUE) violates the PCI
Express Base Specification and should only be used in
closed systems.
MSI_CAP_MULTIMSGCAP
3-bit
Binary
Multiple Message Capable. Each MSI function can
request up to 32 unique messages. System software can
read this field to determine the number of messages
requested. The number of messages requested are
encoded as follows:
000b: 1 vector
001b: 2 vectors
010b: 4 vectors
011b: 8 vectors
100b: 16 vectors
101b: 32 vectors
110b - 111b: Reserved
PCIE_CAP_CAPABILITY_VERSION
232
4-bit
Hex
This attribute indicates the version number of the
PCI-SIG defined PCI Express capability structure. It
must be set to 0001b.
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Table H-1:
PCIE_A1 Attributes (Cont’d)
Attribute Name
PCIE_CAP_DEVICE_PORT_TYPE
Type
4-bit
Hex
Description
This attribute identifies the type of device/port. Valid
settings are (all other values are unsupported):
0000b: PCI Express Endpoint device
0001b: Legacy PCI Express Endpoint device
This value is transferred to the PCI Express Capabilities
Register.
PCIE_CAP_INT_MSG_NUM
PCIE_CAP_SLOT_IMPLEMENTED
PCIE_GENERIC
5-bit
Hex
Interrupt Message Number. This value is transferred to
the PCI Express Cap Register [13:9]. It is not used
internally by the integrated Endpoint block.
Boolean
This attribute must be set to FALSE.
12-bit
Hex
The 12 bits are assigned as follows:
11: This bit must be 0.
10:
• 0: Electrical idle is not received until an Electrical
Idle Ordered Set (EIOS) is received, if no EIOS core
enters the LTSSM RECOVERY state
• 1: An electrical idle can occur without an EIOS
(the EIOS is assumed). This is the default and
recommended setting.
[9:7]: These bits drive the Interrupt Pin Register in the
PCI Configuration Space. A value of 0 indicates no
Legacy interrupts are implemented. Values of 1, 2, 3,
and 4 indicate INTA, INTB, INTC, and INTD,
respectively. Other values are not permitted.
6:
• 0: The DSN Extended Capability is not
implemented
• 1: The DSN Extended Capability is implemented
5:
• 0: 8B/10B Not_in_table is not inferred
• 1: 8B/10B Not_in_table from the GTP transceiver
is inferred from RXSTATUS. This is the default
and recommended setting.
4:
• 0: A read to an unimplemented config space
returns completion with data of zero. This is the
default and recommended setting.
• 1: A read to an unimplemented config space
returns a UR (legacy behavior of PIPE)
[3:0]: These bits drive nFTS[7:4]. The lower bits of
nFTS are set to Fh. The default value is 0xF.
PLM_AUTO_CONFIG
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Boolean
This attribute must be set to FALSE.
www.xilinx.com
233
Appendix H: PCIE_A1 Attribute Descriptions
Table H-1:
PCIE_A1 Attributes (Cont’d)
Attribute Name
PM_CAP_AUXCURRENT
Type
Description
3-bit
Binary
AUX Current. Requested auxiliary current allocation.
This value is transferred to the PM Capabilities Register,
bits [24:22]. The integrated Endpoint block does not
support AUX power, so this field should be set to 000b.
PM_CAP_D1SUPPORT
Boolean
D1 Support. This value is transferred to the PM
Capabilities Register, bit 25.
PM_CAP_D2SUPPORT
Boolean
D2 Support. This value is transferred to the PM
Capabilities Register, bit 26.
PM_CAP_DSI
Boolean
Device Specific Initialization (DSI). This value is
transferred to the PM Capabilities Register, bit 21.
PM_CAP_PME_CLOCK
Boolean
When this attribute is set to TRUE, a PCI™ clock is
required for PME generation. This attribute must be set
to FALSE per the specification. The value is transferred
to the PM Capabilities Register, bit 19.
PM_CAP_PMESUPPORT
5-bit
Hex
PME Support. These five bits indicate support for
D3cold, D3hot, D2, D1, and D0, respectively. This value
is transferred to the PM Capabilities Register, bits
[31:27].
PM_CAP_VERSION
3-bit
Binary
The version of Power Management specification
followed. This value is transferred to the PM
Capabilities Register, bits [18:16].
This attribute must be set to 3.
PM_DATA_SCALE0
2-bit
Hex
Power Management Data Scale Register 0. This attribute
specifies the scale applied to PM_DATA0. The power
consumption of the device is determined by multiplying
the contents of the Base Power Data Register field with
the value corresponding to the encoding returned by
this field. Defined encodings are:
00b: 1.0x
01b: 0.1x
10b: 0.01x
11b: 0.001x
PM_DATA_SCALE1
2-bit
Hex
Power Management Data Scale Register 1. This attribute
specifies the scale applied to PM_DATA1. The power
consumption of the device is determined by multiplying
the contents of the Base Power Data Register field with
the value corresponding to the encoding returned by
this field. Defined encodings are:
00b: 1.0x
01b: 0.1x
10b: 0.01x
11b: 0.001x
234
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Table H-1:
PCIE_A1 Attributes (Cont’d)
Attribute Name
PM_DATA_SCALE2
Type
Description
2-bit
Hex
Power Management Data Scale Register 2. This attribute
specifies the scale applied to PM_DATA2. The power
consumption of the device is determined by multiplying
the contents of the Base Power Data Register field with
the value corresponding to the encoding returned by
this field. Defined encodings are:
00b: 1.0x
01b: 0.1x
10b: 0.01x
11b: 0.001x
PM_DATA_SCALE3
2-bit
Hex
Power Management Data Scale Register 3. This attribute
specifies the scale applied to PM_DATA3. The power
consumption of the device is determined by multiplying
the contents of the Base Power Data Register field with
the value corresponding to the encoding returned by
this field. Defined encodings are:
00b: 1.0x
01b: 0.1x
10b: 0.01x
11b: 0.001x
PM_DATA_SCALE4
2-bit
Hex
Power Management Data Scale Register 4. This attribute
specifies the scale applied to PM_DATA4. The power
consumption of the device is determined by multiplying
the contents of the Base Power Data Register field with
the value corresponding to the encoding returned by
this field. Defined encodings are:
00b: 1.0x
01b: 0.1x
10b: 0.01x
11b: 0.001x
PM_DATA_SCALE5
2-bit
Hex
Power Management Data Scale Register 5. This attribute
specifies the scale applied to PM_DATA5. The power
consumption of the device is determined by multiplying
the contents of the Base Power Data Register field with
the value corresponding to the encoding returned by
this field. Defined encodings are:
00b: 1.0x
01b: 0.1x
10b: 0.01x
11b: 0.001x
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
235
Appendix H: PCIE_A1 Attribute Descriptions
Table H-1:
PCIE_A1 Attributes (Cont’d)
Attribute Name
PM_DATA_SCALE6
Type
Description
2-bit
Hex
Power Management Data Scale Register 6. This attribute
specifies the scale applied to PM_DATA6. The power
consumption of the device is determined by multiplying
the contents of the Base Power Data Register field with
the value corresponding to the encoding returned by
this field. Defined encodings are:
00b: 1.0x
01b: 0.1x
10b: 0.01x
11b: 0.001x
PM_DATA_SCALE7
2-bit
Hex
Power Management Data Scale Register 7. This attribute
specifies the scale applied to PM_DATA7. The power
consumption of the device is determined by multiplying
the contents of the Base Power Data Register field with
the value corresponding to the encoding returned by
this field. Defined encodings are:
00b: 1.0x
01b: 0.1x
10b: 0.01x
11b: 0.001x
PM_DATA0
8-bit
Hex
Power Management Data Register 0 (D0 Power
Consumed). This value appears in the Data field of the
PM Status Register if the host has written the value
0000b to the Data Select field of the PM Control
Register.
PM_DATA1
8-bit
Hex
Power Management Data Register 1 (D1 Power
Consumed). This value appears in the Data field of the
PM Status Register if the host has written the value
0001b to the Data Select field of the PM Control
Register.
PM_DATA2
8-bit
Hex
Power Management Data Register 2 (D2 Power
Consumed). This value appears in the Data field of the
PM Status Register if the host has written the value
0010b the Data Select field of the PM Control Register.
PM_DATA3
8-bit
Hex
Power Management Data Register 3 (D3 Power
Consumed). This value appears in the Data field of the
PM Status Register if the host has written the value
0011b to the Data Select field of the PM Control
Register.
PM_DATA4
8-bit
Hex
Power Management Data Register 4 (D0 Power
Dissipated). This value appears in the Data field of the
PM Status Register if the host has written the value
0100b to the Data Select field of the PM Control
Register.
236
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Table H-1:
PCIE_A1 Attributes (Cont’d)
Attribute Name
Type
Description
PM_DATA5
8-bit
Hex
Power Management Data Register 5 (D1 Power
Dissipated). This value appears in the Data field of the
PM Status Register if the host has written the value
0101b to the Data Select field of the PM Control
Register.
PM_DATA6
8-bit
Hex
Power Management Data Register 6 (D2 Power
Dissipated). This value appears in the Data field of the
PM Status Register if the host has written the value
0110b to the Data Select field of the PM Control
Register.
PM_DATA7
8-bit
Hex
Power Management Data Register 7 (D3 Power
Dissipated). This value appears in the Data field of the
PM Status Register if the host has written the value
0111b to the Data Select field of the PM Control
Register.
Boolean
Attention Button Present. When this attribute is TRUE,
an Attention Button is implemented on the chassis for
this slot. This value is transferred to the Slot Capabilities
Register.
SLOT_CAP_ATT_BUTTON_PRESENT
This attribute must be set to FALSE for Endpoints.
SLOT_CAP_ATT_INDICATOR_PRESENT
Boolean
Attention Indicator Present. When this attribute is
TRUE, an Attention Indicator is implemented on the
chassis for this slot. This value is transferred to the Slot
Capabilities Register.
This attribute must be set to FALSE for Endpoints.
SLOT_CAP_POWER_INDICATOR_PRESENT
Boolean
Power Indicator Present. When this attribute is TRUE, a
Power Indicator is implemented on the chassis for this
slot. This value is transferred to the Slot Capabilities
Register.
This attribute must be set to FALSE for Endpoints.
TL_RX_RAM_RADDR_LATENCY
Boolean
This attribute specifies the read address latency for RX
RAMs in terms of USER_CLK cycles.
FALSE: No fabric pipeline register is on the read
address and enable block RAM inputs
TRUE: A fabric pipeline register is on the read
address and enable block RAM inputs
TL_RX_RAM_RDATA_LATENCY
2-bit
Binary
This attribute specifies the read data latency for RX
RAMs in terms of USER_CLK cycles.
01b: The block RAM output register is disabled
10b: The block RAM output register is enabled
11b: The block RAM output register is enabled and a
fabric pipeline register is added to the block RAM
data output
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
237
Appendix H: PCIE_A1 Attribute Descriptions
Table H-1:
PCIE_A1 Attributes (Cont’d)
Attribute Name
TL_RX_RAM_WRITE_LATENCY
Type
Description
Boolean
This attribute specifies the write latency for RX RAMs in
terms of cycles of USER_CLK.
FALSE: No fabric pipeline register is on the write
address and enable block RAM inputs
TRUE: A fabric pipeline register is on the write
address and enable block RAM inputs
TL_TFC_DISABLE
Boolean
When this attribute is set to TRUE, checking of flow
control values and transmit packets in the order they
were presented on the TRN TX interface is disabled.
TL_TX_CHECKS_DISABLE
Boolean
When this attribute is set to TRUE, all TLM checks of
incoming data are disabled.
TL_TX_RAM_RADDR_LATENCY
Boolean
This attribute specifies the read address latency for TX
RAMs in terms of USER_CLK cycles.
FALSE: No fabric pipeline register on the read
address and enable block RAM inputs
TRUE: A fabric pipeline register is on the read
address and enable block RAM inputs
TL_TX_RAM_RDATA_LATENCY
2-bit
Binary
This attribute specifies the read data latency for TX
RAMs in terms of USER_CLK cycles.
01b: The block RAM output register is disabled
01b: The block RAM output register is enabled
11b: The block RAM output register is enabled and a
fabric pipeline register is added to the block RAM
data output
USR_CFG
Boolean
When this attribute is set to TRUE, the user application
is permitted to add or implement PCI Legacy capability
registers beyond address BFh. This option should be
selected when the user application implements such a
legacy capability configuration space, starting at C0h.
USR_EXT_CFG
Boolean
When this attribute is set to TRUE, the user application
is permitted to add or implement PCI Express extended
capability registers beyond address 1FFh. This box
should be checked when the user application
implements such an extended capability configuration
space starting at 200h.
VC0_CPL_INFINITE
Boolean
When this attribute is set to TRUE, the block advertises
infinite completions.
Note: For Endpoints, this attribute must be set to TRUE
for compliance.
VC0_RX_RAM_LIMIT
238
12-bit
Hex
This attribute must be set to RX buffer bytes/4.
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Table H-1:
PCIE_A1 Attributes (Cont’d)
Attribute Name
VC0_TOTAL_CREDITS_CD
Type
Description
11-bit
Hex
Number of credits that should be advertised for
Completion data received on Virtual Channel 0. The
bytes advertised must be less than or equal to the block
RAM bytes available.
The equation to calculate bytes advertised is:
(ph * (rx_td_ecrc_trim ? 16 : 20)) + (pd * 16) + (nph * 20)
+ (ch * 16) + (cd * 16)
The equation to calculate block RAM bytes available is:
(vc0_rx_ram_limit + 1) * 4
See Table H-2, page 240 for valid settings.
VC0_TOTAL_CREDITS_CH
7-bit
Hex
Number of credits that should be advertised for
Completion headers received on Virtual Channel 0. The
sum of the Posted, Non-Posted, and Completion header
credits must be < 80.
See Table H-2, page 240 for valid settings.
VC0_TOTAL_CREDITS_NPH
7-bit
Hex
Number of credits that should be advertised for NonPosted headers received on Virtual Channel 0. The
number of Non-Posted data credits advertised by the
block is equal to the number of Non-Posted header
credits. The sum of the Posted, Non-Posted, and
Completion header credits must be < 80.
This attribute must be set to 8.
VC0_TOTAL_CREDITS_PD
11-bit
Hex
Number of credits that should be advertised for Posted
data received on Virtual Channel 0. The bytes advertised
must be less than or equal to the block RAM bytes
available.
The equation to calculate bytes advertised is:
(ph * (rx_td_ecrc_trim ? 16 : 20)) + (pd * 16) + (nph * 20)
+ (ch * 16) + (cd * 16)
The equation to calculate block RAM bytes available is:
(vc0_rx_ram_limit + 1) * 4
See Table H-2, page 240 for valid settings.
VC0_TOTAL_CREDITS_PH
7-bit
Hex
Number of credits that should be advertised for Posted
headers received on Virtual Channel 0. The sum of the
Posted, Non-Posted, and Completion header credits
must be < 80.
VC0_TX_LASTPACKET
5-bit
Hex
Index of the last packet buffer used by TX TLM (that is,
the number of buffers – 1). This value is calculated from
the maximum payload size supported and the number
of block RAMs configured for transmit.
The equation is:
((TX buffer bytes) / (MPS_in_bytes + 20) - 1)
See Table H-2 for valid settings.
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
239
Appendix H: PCIE_A1 Attribute Descriptions
Table H-2:
Valid Data Credit Combinations
Parameter Name
240
Valid Combinations
DEV_CAP_MAX_PAYLOAD_SUPPORTED
0
0
1
1
2
2
VC0_TOTAL_CREDITS_PD
36
92
92
204
204
716
VC0_TOTAL_CREDITS_CD
64
128
128
256
256
256
VC0_TOTAL_CREDITS_CH
8
16
16
32
32
32
VC0_TX_LAST_PACKET
12
26
13
28
14
29
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Appendix I
PCIE_A1 Timing Parameter
Descriptions
This appendix lists the timing parameter names and descriptions related to the
Spartan®-6 FPGA Integrated Endpoint Block for PCI Express® designs. This information
is useful for debugging timing issues. Values for these timing parameters can be obtained
by running the Speedprint tool. Usage of Speedprint is documented in the Development
System Reference Guide.
The timing parameters on the integrated Endpoint block consist of either Setup/Hold or
Clock-to-Out parameters. Table I-1 lists the timing parameter names, descriptions, signal
grouping, and related clock domain for a given parameter. In the table, parameter
Tpcicck_XXX is a setup time (before the clock edge), and parameter Tpcickc_XXX is a hold
time (after clock edge).
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
241
Appendix I: PCIE_A1 Timing Parameter Descriptions
Table I-1:
PCIE_A1 Timing Parameters
Name
Clock Domain
Signal Grouping
Sequential Setup and Hold Times for Integrated Endpoint Block Inputs
Tpcicck_CFG /
Tpcickc_CFG
USERCLK
CFGDEVID[15:0]
CFGDSN[63:0]
CFGDWADDR[9:0]
CFGERRCORN
CFGERRCPLTIMEOUTN
CFGERRECRCN
CFGERRLOCKEDN
CFGERRPOSTEDN
CFGERRTLPCPLHEADER[47:0]
CFGERRURN
CFGINTERRUPTASSERTN
CFGINTERRUPTDI[7:0]
CFGINTERRUPTN
CFGRDENN
CFGREVID[7:0]
CFGSUBSYSID[15:0]
CFGSUBSYSVENID[15:0]
CFGTRNPENDINGN
CFGVENID[15:0]
TRNTCFGGNTN
Tpcicck_ERR /
Tpcickc_ERR
242
USERCLK
CFGERRCPLABORTN
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Table I-1:
PCIE_A1 Timing Parameters (Cont’d)
Name
Tpcicck_MGT /
Tpcickc_MGT
Clock Domain
MGTCLK
Signal Grouping
PIPEGTRESETDONEA
PIPEGTRESETDONEB
PIPEPHYSTATUSA
PIPEPHYSTATUSB
PIPERXCHARISKA[1:0]
PIPERXCHARISKB[1:0]
PIPERXDATAA[15:0]
PIPERXDATAB[15:0]
PIPERXENTERELECIDLEA
PIPERXENTERELECIDLEB
PIPERXSTATUSA[2:0]
PIPERXSTATUSB[2:0]
Tpcicck_PWR /
Tpcickc_PWR
USERCLK
Tpcicck_SCAN /
Tpcickc_SCAN
USERCLK
CFGPMWAKEN
CFGTURNOFFOKN
SCANEN
SCANIN[4:0]
SCANRESETMASK
Tpcidck_LOCKED /
Tpcickd_LOCKED
USERCLK
CLOCKLOCKED
Tpcidck_RESET /
Tpcickd_RESET
USERCLK
SYSRESETN
Tpcidck_RXRAM /
Tpcickd_RXRAM
USERCLK
MIMRXRDATA[34:0]
Tpcidck_TRNFC /
Tpcickd_TRNFC
USERCLK
TRNFCSEL[2:0]
Tpcidck_TRNRD /
Tpcickd_TRNRD
USERCLK
TRNRDSTRDYN
Tpcidck_TRNRN /
Tpcickd_TRNRN
USERCLK
TRNRNPOKN
Tpcidck_TRNTD /
Tpcickd_TRNTD
USERCLK
TRNTD[31:0]
Tpcidck_TRNTE /
Tpcickd_TRNTE
USERCLK
TRNTEOFN
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
TRNTERRFWDN
www.xilinx.com
243
Appendix I: PCIE_A1 Timing Parameter Descriptions
Table I-1:
PCIE_A1 Timing Parameters (Cont’d)
Name
Tpcidck_TRNTS /
Tpcickd_TRNTS
Clock Domain
USERCLK
Signal Grouping
TRNTSOFN
TRNTSRCDSCN
TRNTSRCRDYN
TRNTSTRN
Tpcidck_TXRAM /
Tpcickd_TXRAM
USERCLK
MIMTXRDATA[35:0]
Sequential Clock to Output Times for Integrated Endpoint Block Outputs
Tpcicko_CFG
USERCLK
CFGCOMMANDINTERRUPTDISABLE
CFGDEVCONTROLMAXPAYLOAD[2:0]
CFGDEVCONTROLMAXREADREQ[2:0]
Tpcicko_CFGBUS
USERCLK
CFGBUSNUMBER[7:0]
Tpcicko_CFGCOMMAND
USERCLK
CFGCOMMANDSERREN
Tpcicko_CFGDEV
USERCLK
CFGDEVCONTROLCORRERRREPORTINGEN
CFGDEVCONTROLEXTTAGEN
CFGDEVCONTROLFATALERRREPORTINGEN
CFGDEVCONTROLNONFATALREPORTINGEN
CFGDEVCONTROLNOSNOOPEN
CFGDEVCONTROLPHANTOMEN
CFGDEVCONTROLURERRREPORTINGEN
CFGDEVICENUMBER[4:0]
CFGDEVSTATUSCORRERRDETECTED
CFGDEVSTATUSFATALERRDETECTED
CFGDEVSTATUSNONFATALERRDETECTED
CFGDEVSTATUSURDETECTED
Tpcicko_CFGDO
USERCLK
CFGDO[31:0]
Tpcicko_CFGDONE
USERCLK
CFGRDWRDONEN
Tpcicko_CFGERR
USERCLK
CFGERRCPLRDYN
Tpcicko_CFGFCN
USERCLK
CFGFUNCTIONNUMBER[2:0]
Tpcicko_CFGINT
USERCLK
CFGINTERRUPTDO[7:0]
CFGINTERRUPTRDYN
244
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Table I-1:
PCIE_A1 Timing Parameters (Cont’d)
Name
Tpcicko_CFGLINK
Clock Domain
USERCLK
Signal Grouping
CFGLINKCONTOLRCB
CFGLINKCONTROLASPMCONTROL[1:0]
CFGLINKCONTROLCOMMONCLOCK
CFGLINKCONTROLEXTENDEDSYNC
Tpcicko_CFGOFF
USERCLK
CFGTOTURNOFFN
Tpcicko_CFGSTATE
MGTCLK
CFGLTSSMSTATE[4:0]
USERCLK
CFGPCIELINKSTATEN[2:0]
USERCLK
DBGBADTLPLCRC
Tpcicko_DBG
DBGBADTLPSEQNUM
DBGMLFRMDLENGTH
DBGMLFRMDMPS
DBGMLFRMDTCVC
DBGMLFRMDUNRECTYPE
DBGREGDETECTEDCORRECTABLE
DBGREGDETECTEDFATAL
DBGREGDETECTEDNONFATAL
DBGREGDETECTEDUNSUPPORTED
DBGURNOBARHIT
DBGURPOISCFGWR
DBGURUNSUPMSG
Tpcicko_ENA
USERCLK
CFGCOMMANDBUSMASTERENABLE
CFGCOMMANDIOENABLE
CFGCOMMANDMEMENABLE
CFGDEVCONTROLENABLERO
CFGINTERRUPTMMENABLE[2:0]
Tpcicko_MSG
USERCLK
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
CFGINTERRUPTMSIENABLE
www.xilinx.com
245
Appendix I: PCIE_A1 Timing Parameter Descriptions
Table I-1:
PCIE_A1 Timing Parameters (Cont’d)
Name
Tpcicko_PIPE
Clock Domain
MGTCLK
Signal Grouping
PIPEGTTXELECIDLEA
PIPEGTTXELECIDLEB
PIPERXPOLARITYA
PIPERXPOLARITYB
PIPERXRESETA
PIPERXRESETB
PIPETXCHARDISPMODEA[1:0]
PIPETXCHARDISPMODEB[1:0]
PIPETXCHARDISPVALA[1:0]
PIPETXCHARDISPVALB[1:0]
PIPETXCHARISKA[1:0]
PIPETXCHARISKB[1:0]
PIPETXDATAA[15:0]
PIPETXDATAB[15:0]
PIPETXRCVRDETA
PIPETXRCVRDETB
Tpcicko_PWR
MGTCLK
PIPEGTPOWERDOWNA[1:0]
PIPEGTPOWERDOWNB[1:0]
Tpcicko_RXRAM
USERCLK
CFGDEVCONTROLAUXPOWEREN
USERCLK
MIMRXRADDR[11:0]
MIMRXREN
MIMRXWADDR[11:0]
MIMRXWDATA[34:0]
MIMRXWEN
Tpcicko_SCANOUT
246
USERCLK
SCANOUT[4:0]
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Table I-1:
PCIE_A1 Timing Parameters (Cont’d)
Name
Tpcicko_STATUS
Clock Domain
USERCLK
Signal Grouping
DBGBADDLLPSTATUS
DBGBADTLPSTATUS
DBGDLPROTOCOLSTATUS
DBGFCPROTOCOLERRSTATUS
DBGMLFRMDTLPSTATUS
DBGPOISTLPSTATUS
DBGRCVROVERFLOWSTATUS
DBGRPLYROLLOVERSTATUS
DBGRPLYTIMEOUTSTATUS
DBGURSTATUS
Tpcicko_TRN
USERCLK
TRNFCCPLD[11:0]
TRNFCCPLH[7:0]
TRNFCNPD[11:0]
TRNFCNPH[7:0]
TRNFCPD[11:0]
TRNFCPH[7:0]
TRNLNKUPN
TRNRBARHITN[6:0]
TRNRD[31:0]
TRNREOFN
TRNRERRFWDN
TRNRSOFN
TRNRSRCDSCN
TRNRSRCRDYN
TRNTBUFAV[5:0]
TRNTCFGREQN
TRNTDSTRDYN
TRNTERRDROPN
Tpcicko_TXRAM
USERCLK
MIMTXRADDR[11:0]
MIMTXREN
MIMTXWADDR[11:0]
MIMTXWDATA[35:0]
MIMTXWEN
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
www.xilinx.com
247
Appendix I: PCIE_A1 Timing Parameter Descriptions
248
www.xilinx.com
Spartan-6 FPGA Integrated Endpoint Block
UG654 (v3.0) April 19, 2010
Was this manual useful for you? yes no
Thank you for your participation!

* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project

Download PDF

advertisement