0
登录后你可以
  • 下载海量资料
  • 学习在线课程
  • 观看技术视频
  • 写文章/发帖/加入社区
创作中心
发布
  • 发文章

  • 发资料

  • 发帖

  • 提问

  • 发视频

创作活动
CA91C142D-33IEV

CA91C142D-33IEV

  • 厂商:

    RENESAS(瑞萨)

  • 封装:

    BGA313

  • 描述:

    IC INTERFACE SPECIALIZED 313BGA

  • 数据手册
  • 价格&库存
CA91C142D-33IEV 数据手册
® Universe IID/IIBTM User Manual May 12, 2010 6024 Silver Creek Valley Road, San Jose, California 95138 Telephone: (800) 345-7015 • (408) 284-8200 • FAX: (408) 284-2775 Printed in U.S.A. ©2009 Integrated Device Technology, Inc. GENERAL DISCLAIMER Integrated Device Technology, Inc. reserves the right to make changes to its products or specifications at any time, without notice, in order to improve design or performance and to supply the best possible product. IDT does not assume any responsibility for use of any circuitry described other than the circuitry embodied in an IDT product. The Company makes no representations that circuitry described herein is free from patent infringement or other rights of third parties which may result from its use. No license is granted by implication or otherwise under any patent, patent rights or other rights, of Integrated Device Technology, Inc. CODE DISCLAIMER Code examples provided by IDT are for illustrative purposes only and should not be relied upon for developing applications. Any use of the code examples below is completely at your own risk. IDT MAKES NO REPRESENTATIONS OR WARRANTIES OF ANY KIND CONCERNING THE NONINFRINGEMENT, QUALITY, SAFETY OR SUITABILITY OF THE CODE, EITHER EXPRESS OR IMPLIED, INCLUDING WITHOUT LIMITATION ANY IMPLIED WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE, OR NON-INFRINGEMENT. FURTHER, IDT MAKES NO REPRESENTATIONS OR WARRANTIES AS TO THE TRUTH, ACCURACY OR COMPLETENESS OF ANY STATEMENTS, INFORMATION OR MATERIALS CONCERNING CODE EXAMPLES CONTAINED IN ANY IDT PUBLICATION OR PUBLIC DISCLOSURE OR THAT IS CONTAINED ON ANY IDT INTERNET SITE. IN NO EVENT WILL IDT BE LIABLE FOR ANY DIRECT, CONSEQUENTIAL, INCIDENTAL, INDIRECT, PUNITIVE OR SPECIAL DAMAGES, HOWEVER THEY MAY ARISE, AND EVEN IF IDT HAS BEEN PREVIOUSLY ADVISED ABOUT THE POSSIBILITY OF SUCH DAMAGES. The code examples also may be subject to United States export control laws and may be subject to the export or import laws of other countries and it is your responsibility to comply with any applicable laws or regulations. LIFE SUPPORT POLICY Integrated Device Technology's products are not authorized for use as critical components in life support devices or systems unless a specific written agreement pertaining to such intended use is executed between the manufacturer and an officer of IDT. 1. Life support devices or systems are devices or systems which (a) are intended for surgical implant into the body or (b) support or sustain life and whose failure to perform, when properly used in accordance with instructions for use provided in the labeling, can be reasonably expected to result in a significant injury to the user. 2. A critical component is any components of a life support device or system whose failure to perform can be reasonably expected to cause the failure of the life support device or system, or to affect its safety or effectiveness. IDT, the IDT logo, and Integrated Device Technology are trademarks or registered trademarks of Integrated Device Technology, Inc. 3 Contents 1. Functional Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17 1.1 1.2 2. Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.1.1 Universe II Features . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.1.2 Universe II Benefits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.1.3 Universe II Typical Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Main Interfaces. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2.1 VMEbus Interface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2.2 PCI Bus Interface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2.3 Interrupter and Interrupt Handler . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1.2.4 DMA Controller . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17 18 19 19 20 22 22 23 24 VMEbus Interface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25 2.1 2.2 2.3 2.4 2.5 2.6 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . VMEbus Requester . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2.1 Internal Arbitration for VMEbus Requests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2.2 Request Modes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2.3 VMEbus Release . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Universe II as VMEbus Master . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3.1 Addressing Capabilities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3.2 Data Transfer Capabilities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3.3 Cycle Terminations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Universe II as VMEbus Slave . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.4.1 Coupled Transfers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.4.2 Posted Writes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.4.3 Prefetched Block Reads . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.4.4 VMEbus Lock Commands (ADOH Cycles) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.4.5 VMEbus Read-Modify-Write Cycles (RMW Cycles) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.4.6 Register Accesses. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.4.7 Location Monitors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.4.8 Generating PCI Configuration Cycles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . VMEbus Configuration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.5.1 First Slot Detector . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.5.2 VMEbus Register Access at Power-up . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Automatic Slot Identification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.6.1 Auto Slot ID: VME64 Specified . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.6.2 Auto-ID: A Proprietary IDT Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.6.3 System Controller Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.6.4 IACK Daisy-Chain Driver Module . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.6.5 VMEbus Time-out . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.6.6 Bus Isolation Mode (BI-Mode) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Integrated Device Technology www.idt.com 25 25 25 26 27 28 28 30 31 32 32 33 34 36 37 37 38 38 41 41 41 42 42 43 44 45 45 46 Universe II User Manual May 12, 2010 4 3. Contents PCI Interface. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49 3.1 3.2 3.3 3.4 4. 49 49 49 50 50 52 52 53 53 54 55 55 56 57 57 57 58 60 61 64 65 Slave Image Programming . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67 4.1 4.2 4.3 4.4 5. Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . PCI Cycles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2.1 32-Bit Versus 64-Bit PCI . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2.2 PCI Bus Request and Parking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2.3 Address Phase . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2.4 Data Transfer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2.5 Termination Phase . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2.6 Parity Checking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Universe II as PCI Master . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3.1 Command Types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3.2 PCI Burst Transfers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3.3 Termination . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3.4 Parity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Universe II as PCI Target. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.4.1 Command Types . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.4.2 Data Transfer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.4.3 Coupled Transfers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.4.4 Posted Writes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.4.5 Special Cycle Generator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.4.6 Using the VOWN bit . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.4.7 Terminations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . VME Slave Image Programming. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2.1 VMEbus Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2.2 PCI Bus Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2.3 Control Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . PCI Bus Target Images . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.3.1 PCI Bus Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.3.2 VMEbus Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.3.3 Control Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Special PCI Target Image . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67 67 68 69 69 70 71 71 73 73 Registers Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75 5.1 5.2 5.3 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Register Access from the PCI Bus. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2.1 PCI Configuration Access . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2.2 Memory or I/O Access. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2.3 Locking the Register Block from the PCI bus. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Register Access from the VMEbus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.3.1 VMEbus Register Access Image (VRAI) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.3.2 CR/CSR Accesses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.3.3 RMW and ADOH Register Access Cycles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Universe II User Manual May 12, 2010 75 76 76 77 78 79 79 80 81 Integrated Device Technology www.idt.com Contents 5 5.4 5.5 Mailbox Registers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83 Semaphores . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83 6. DMA Controller . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85 6.1 6.2 6.3 6.4 6.5 6.6 6.7 6.8 7. Interrupt Generation and Handling. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109 7.1 7.2 7.3 8. Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85 DMA Registers. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85 6.2.1 Source and Destination Addresses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86 6.2.2 Non-incrementing DMA Mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 87 6.2.3 Transfer Size. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89 6.2.4 Transfer Data Width . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89 6.2.5 DMA Command Packet Pointer. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90 6.2.6 DMA Control and Status . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 90 Direct Mode Operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93 Linked-list Mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96 6.4.1 Linked-list Updating . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100 FIFO Operation and Bus Ownership . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101 6.5.1 PCI-to-VMEbus Transfers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101 6.5.2 VMEbus-to-PCI Transfers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103 DMA Interrupts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104 DMA Channel Interactions with Other Channels . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105 DMA Error Handling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105 6.8.1 DMA Software Response to Error . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105 6.8.2 DMA Hardware Response to Error . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106 6.8.3 Resuming DMA Transfers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Interrupt Generation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.2.1 PCI Interrupt Generation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.2.2 VMEbus Interrupt Generation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Interrupt Handling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.3.1 PCI Interrupt Handling. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.3.2 VMEbus Interrupt Handling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.3.3 Internal Interrupt Handling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.3.4 VME64 Auto-ID . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109 111 111 113 116 116 116 118 123 Error Handling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125 8.1 8.2 8.3 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Errors on Coupled Cycles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Errors on Decoupled Transactions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.3.1 Posted Writes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.3.2 Prefetched Reads . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.3.3 DMA Errors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.3.4 Parity Errors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Integrated Device Technology www.idt.com 125 125 125 126 127 127 127 Universe II User Manual May 12, 2010 6 9. Contents Resets, Clocks and Power-up Options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .129 9.1 9.2 9.3 9.4 9.5 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Resets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9.2.1 Universe II Reset Circuitry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9.2.2 Reset Implementation Cautions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Power-Up Options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9.3.1 Power-up Option Descriptions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9.3.2 Power-up Option Implementation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9.3.3 Hardware Initialization (Normal Operating Mode) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Test Modes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9.4.1 Auxiliary Test Modes. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9.4.2 JTAG support . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Clocks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129 129 132 133 135 136 139 140 141 141 141 142 10. Signals and Pinout . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .143 10.1 10.2 10.3 10.4 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . VMEbus Signals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . PCI Bus Signals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Pin-out . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10.4.1 313-pin Plastic BGA Package (PBGA) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143 144 147 151 151 11. Electrical Characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .153 11.1 11.2 11.3 11.4 DC Characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.1.1 Non-PCI Characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.1.2 PCI Characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.1.3 Pin List and DC Characteristics for all Signals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Operating Conditions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11.2.1 Absolute Maximum Ratings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Power Dissipation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Power Sequencing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153 153 154 155 160 161 161 162 12. Registers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .163 12.1 12.2 12.3 A. 163 164 171 171 Packaging Information . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .333 A.1 B. Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Register Map . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Register Description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12.3.1 PCI Configuration Space ID Register (PCI_ID) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 313 Pin PBGA Package . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 333 Performance. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .335 B.1 B.2 B.3 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . PCI Slave Channel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . B.2.1 Coupled Cycles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . B.2.2 Decoupled Cycles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . VME Slave Channel . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Universe II User Manual May 12, 2010 335 336 336 338 340 Integrated Device Technology www.idt.com Contents B.4 B.5 B.6 C. Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Physical Characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Thermal Characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Universe II Ambient Operating Calculations. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Thermal Vias . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 355 355 356 357 358 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 359 Little-endian Mode. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 359 Typical Applications. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 363 E.1 E.2 E.3 E.4 E.5 F. 340 342 347 347 347 348 350 350 350 352 Endian Mapping . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 359 D.1 D.2 E. B.3.1 Coupled Cycles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . B.3.2 Decoupled Cycles. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . DMA Channel and Relative FIFO Sizes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . B.4.1 VMEbus Ownership Modes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . B.4.2 VME Transfers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . B.4.3 PCI Transfers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Universe II Specific Register . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . B.5.1 Overview of the U2SPEC Register . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . B.5.2 Adjustable VME Timing Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Performance Summary. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Reliability Prediction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 355 C.1 C.2 C.3 C.4 C.5 D. 7 Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . VME Interface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . E.2.1 Transceivers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . E.2.2 Direction control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . E.2.3 Power-up Options. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . PCI Bus Interface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . E.3.1 Resets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . E.3.2 Local Interrupts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Manufacturing Test Pins. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Decoupling VDD and VSS on the Universe II . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 363 363 363 367 367 368 369 370 370 371 Ordering Information . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 373 F.1 Ordering Information . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 373 Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 8 Universe II User Manual May 12, 2010 Contents Integrated Device Technology www.idt.com 9 Figures Figure 1: Figure 2: Figure 3: Figure 4: Figure 5: Figure 6: Figure 7: Figure 8: Figure 9: Figure 10: Figure 11: Figure 12: Figure 13: Figure 14: Figure 15: Figure 16: Figure 17: Figure 18: Figure 19: Figure 20: Figure 21: Figure 22: Figure 23: Figure 24: Figure 25: Figure 26: Figure 27: Figure 28: Figure 29: Figure 30: Figure 31: Figure 32: Figure 33: Figure 34: Figure 35: Figure 36: Figure 37: Figure 38: Figure 39: Figure 40: Figure 41: Figure 42: Universe II Block Diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18 Universe II In Single Board Computer Application . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20 Universe II Data Flow Diagram . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21 VMEbus Slave Channel Dataflow. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32 Timing for Auto-ID Cycle. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44 PCI Bus Target Channel Dataflow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58 Register Fields for the Special Cycle Generator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62 Address Translation Mechanism for VMEbus to PCI Bus Transfers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69 Address Translation Mechanism for PCI Bus to VMEbus Transfers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72 Memory Mapping in the Special PCI Target Image . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74 Universe II Control and Status Register Space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76 PCI Bus Access to UCSR as Memory or I/O Space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77 UCSR Access from the VMEbus Register Access Image. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80 UCSR Access in VMEbus CR/CSR Space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 82 Direct Mode DMA transfers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94 Command Packet Structure and Linked List Operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97 DMA Linked List Operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98 Universe Interrupt Circuitry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110 STATUS/ID Provided by Universe II . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115 Sources of Internal Interrupts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119 Reset Circuitry. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133 Resistor-Capacitor Circuit Ensuring Power-Up Reset Duration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134 Power-up Options Timing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139 UCSR Access Mechanisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163 313 PBGA - Bottom View . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 333 313 PBGA - Top and Side View . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 334 Coupled Read Cycle - Universe II as VME Master . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 337 Several Coupled Read Cycles - Universe II as VME Master . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 337 Coupled Write Cycle - Universe II as VME Master . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 338 Several Non-Block Decoupled Writes - Universe II as VME Master. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 340 BLT Decoupled Write - Universe II as VME Master . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 340 Coupled Read Cycle - Universe II as VME Slave . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 341 Coupled Write Cycle - Universe II as VME Slave (bus parked at Universe II) . . . . . . . . . . . . . . . . . . . . . . . . . . 342 Non-Block Decoupled Write Cycle - Universe II as VME Slave . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 343 BLT Decoupled Write Cycle - Universe II as VME Slave . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 344 MBLT Decoupled Write Cycle - Universe II as VME Slave . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 344 BLT Pre-fetched Read Cycle - Universe II as VME Slave . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 346 PCI Read Transactions During DMA Operation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 349 Multiple PCI Read Transactions During DMA Operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 349 Universe II Connections to the VMEbus Through TTL Buffers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 364 Universe II Connections to the VMEbus Through TTL Buffers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 365 Power-up Configuration Using Passive Pull-ups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 368 Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 10 Figure 43: Figure 44: Figure 45: Figures Power-up Configuration Using Active Circuitry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 368 Analog Isolation Scheme . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 371 Noise Filter Scheme . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 371 Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 11 Tables Table 1: Table 2: Table 3: Table 4: Table 5: Table 6: Table 7: Table 8: Table 9: Table 10: Table 11: Table 12: Table 13: Table 14: Table 15: Table 16: Table 17: Table 18: Table 19: Table 20: Table 21: Table 22: Table 23: Table 24: Table 25: Table 26: Table 27: Table 28: Table 29: Table 30: Table 31: Table 32: Table 33: Table 34: Table 35: Table 36: Table 37: Table 38: Table 39: Table 40: Table 41: Table 42: VMEbus Address Modifier Codes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28 PCI Address Line Asserted as a Function of VA[15:11] . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39 Command Type Encoding for Transfer Type . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51 VMEbus Fields for VMEbus Slave Image. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67 PCI Bus Fields for VMEbus Slave Image . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68 Control Fields for VMEbus Slave Image . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68 PCI Bus Fields for the PCI Bus Target Image . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70 VMEbus Fields for the PCI Bus Target Image . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70 Control Fields for PCI Bus Target Image . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71 PCI Bus Fields for the Special PCI Target Image . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73 VMEbus Fields for the Special PCI Bus Target Image . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73 Control Fields for the Special PCI Bus Target Image . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73 Programming the VMEbus Register Access Image . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79 VON Settings for Non-Inc Mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88 DMA Interrupt Sources and Enable Bits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104 Source, Enabling, Mapping, and Status of PCI Interrupt Output. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112 Source, Enabling, Mapping, and Status of VMEbus Interrupt Outputs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114 Internal Interrupt Routing. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118 Hardware Reset Mechanisms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129 Software Reset Mechanism . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130 Functions Affected by Reset Initiators. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132 Power-Up Options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135 VRAI Base Address Power-up Options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137 Manufacturing Pin Requirements for Normal Operating Mode . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140 Test Mode Operation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 141 VMEbus Signals. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144 PCI Bus Signals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147 Non-PCI Electrical Characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153 AC/DC PCI Electrical Characteristics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 154 Pin List and DC Characteristics for Universe II Signals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155 Operating Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160 Absolute Maximum Ratings. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161 Power Dissipation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 161 Universe II Register Map . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164 Power-up Option Behavior of the VAS field in VRAI_CTL. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 304 PCI Slave Channel Performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 352 VME Slave Channel Performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 352 DMA Channel Performance. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 353 Ambient to Junction Thermal Impedance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 357 Maximum Universe II Junction Temperature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 357 Thermal Characteristics of Universe II . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 358 Mapping of 32-bit Little-Endian PCI Bus to 32-bit VMEbus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 360 Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 12 Table 43: Table 44: Table 45: Table 46: Table 47: Tables Mapping of 32-bit Little-Endian PCI Bus to 64-bit VMEbus . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . VMEbus Signal Drive Strength Requirements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . VMEbus Transceiver Requirements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Reset Signals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Standard Ordering Information . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Universe II User Manual May 12, 2010 361 366 366 369 373 Integrated Device Technology www.idt.com 13 About this Document This section discusses the following topics: • “Scope” on page 13 • “Document Conventions” on page 13 • “Revision History” on page 15 Scope The Universe IID/IIB User Manual discusses the features, capabilities, and configuration requirements for the Universe II. It is intended for hardware and software engineers who are designing system interconnect applications with the device. Document Conventions This document uses the following conventions. Signal Notation Signals are either active high or active low. Active low signals are defined as true (asserted) when they are at a logic low. Similarly, active high signals are defined as true at a logic high. Signals are considered asserted when active and negated when inactive, irrespective of voltage levels. For voltage levels, the use of 0 indicates a low voltage while a 1 indicates a high voltage. For voltage levels, the use of 0 indicates a low voltage while a 1 indicates a high voltage. For voltage levels, the use of 0 indicates a low voltage while a 1 indicates a high voltage. Each signal that assumes a logic low state when asserted is followed by an underscore sign, “_”. For example, SIGNAL_ is asserted low to indicate an active low signal. Signals that are not followed by an underscore are asserted when they assume the logic high state. For example, SIGNAL is asserted high to indicate an active high signal. The asterisk sign “*” is used in this manual to show that a signal is asserted low and that is used on the on the VMEbus backplane. For example, SIGNAL* is asserted to low to indicate an active low signal on the VMEbus backplane. Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 14 About this Document Object Size Notation • A byte is an 8-bit object. • A word is a 16-bit object. • A doubleword (Dword) is a 32-bit object. • A quadword is a 64-bit (8 byte) object. • A Kword is 1024 16-bit words. Numeric Notation • Hexadecimal numbers are denoted by the prefix 0x (for example, 0x04). • Binary numbers are denoted by the prefix 0b (for example, 0b010). • Registers that have multiple iterations are denoted by {x..y} in their names; where x is first register and address, and y is the last register and address. For example, REG{0..1} indicates there are two versions of the register at different addresses: REG0 and REG1. Symbols Ti This symbol indicates a basic design concept or information considered helpful. p This symbol indicates important configuration information or suggestions. This symbol indicates procedures or operating levels that may result in misuse or damage to the device. Document Status Information • Advance – Contains information that is subject to change, and is available once prototypes are released to customers. • Preliminary – Contains information about a product that is near production-ready, and is revised as required. • Formal – Contains information about a final, customer-ready product, and is available once the product is released to production. Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com About this Document 15 Revision History May 12, 2010, Formal This document fixed a number of minor typographical errors. No technical changes were made. October 2009, Formal This document was rebranded as IDT. No technical changes were made. June 2009, Formal There have been changes throughout the manual. August 2007, Formal There have been numerous edits throughout the manual. The formatting of the document has also been updated. November 2002, Formal This document information applies to both the Universe IIB and the Universe IID devices. The Universe IID is recommended for all new designs. For more information about the two devices, see the Universe IID/IIB Differences Summary. The following chapter was updated for the release of this manual: • “Reliability Prediction” on page 355 October 2002, Formal This document information applies to both the Universe IIB and the Universe IID devices. The Universe IID is recommended for all new designs. For more information about the two devices, see the Universe IID/IIB Differences Summary. There was an erratum found in the 361 DBGA package drawing. Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 16 Universe II User Manual May 12, 2010 About this Document Integrated Device Technology www.idt.com 17 1. Functional Overview This chapter outlines the functionality of the Universe II. This chapter discusses the following topics: 1.1 • “Overview” on page 17 • “VMEbus Interface” on page 22 • “PCI Bus Interface” on page 22 • “Interrupter and Interrupt Handler” on page 23 • “DMA Controller” on page 24 Overview The IDT Universe II is the industry's leading high performance PCI-to-VMEbus interconnect. Universe II is fully compliant with the VME64 bus standard, and tailored for the next-generation of advanced PCI processors and peripherals. With a zero-wait state implementation, multi-beat transactions, and support for bus-parking, Universe II provides high performance on the PCI bus. The Universe II eases development of multi-master, multi-processor architectures on VMEbus and PCI bus systems. The device is ideally suited for CPU boards functioning as both master and slave in the VMEbus system, and that require access to PCI systems. Bridging is accomplished through a decoupled architecture with independent FIFOs for inbound, outbound, and DMA traffic. With this architecture, throughput is maximized without sacrificing bandwidth on either bus. With the Universe II, you know that as your system becomes more complex, you have proven silicon that continues to provide everything you need in a PCI-to-VME bridge. Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 18 1. Functional Overview > Overview Figure 1: Universe II Block Diagram Four Location Monitors To Support VMEbus Broadcast Capability Location Monitor Fixed Priority, Round Robin, Single Level Modes VMEbus Arbiter 32-bit Address / 64-bit Data 33 MHz PCI Bus DMA Channel Bidirectional FIFO Direct/Linked List Mode VMEbus Interface PCI Interface 32-bit Address / 64-bit Data 33 MHz PCI Bus VMEbus Slave Channel Posted Writes, Prefetched Reads, Coupled Reads Register Channel Configuraion Registers, Mailbox Registers, Semaphores Interrupt Channel Interrupt Handler, Interrupter PCI Target Channel Posted Writes, Coupled Read JTAG 8091142_BK001_05 IEEE1149.1 Boundary Scan 1.1.1 Universe II Features The Universe II has the following features: • Industry-proven, high performance 64-bit VMEbus interconnect • Fully compliant, 32-bit or 64-bit, 33 MHz PCI bus interconnect • Integral FIFOs for write posting to maximize bandwidth utilization • Programmable DMA controller with Linked-List mode (Scatter/Gather) support • Flexible interrupt logic • Sustained transfer rates up to 60-70 Mbytes/s • Extensive suite of VMEbus address and data transfer modes • Automatic initialization for slave-only applications • Flexible register set, programmable from both the PCI bus and VMEbus ports • Full VMEbus system controller • Support for RMWs, ADOH, PCI LOCK_ cycles, and semaphores Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 1. Functional Overview > Overview • Commercial, industrial, and extended temperature variants • IEEE 1149.1 JTAG • Available packaging: 19 — 35mm x 35mm, 313-contact plastic BGA (PBGA) package 1.1.2 Universe II Benefits The Universe II offers the following benefits to designers: 1.1.3 • Industry proven device • Reliable customer support with experience in hundreds of customer designs Universe II Typical Applications The Universe II is targeted at today’s technology demands, such as the following: 1.1.3.1 • Single-board computers • Telecommunications equipment • Test equipment • Command and control systems • Factory automation equipment • Medical equipment • Military • Aerospace Typical Application Example: Single Board Computers The Universe II is widely used on VME-based Single Board Computers (SBC) that employ PCI as their local bus and VME as the backplane bus, as shown in the accompanying diagram. These SBC cards support a variety of applications including telecom, datacom, medical, industrial, and military equipment. The Universe II high performance architecture seamlessly bridges the PCI and VME busses, and is the VME industry's standard for single board computer interconnect device. Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 20 1. Functional Overview > Main Interfaces Figure 2: Universe II In Single Board Computer Application PMC Connection Memory Processor Processor Bus Processorto-PCI Bridge PCI Bus 32-bit/64-bit Data 33 MHz I/O Controller Universe II 64-bit Data VMEbus 1.2 8091142_TA001_02 Main Interfaces The Universe II has two main interfaces: the PCI Bus Interface and the VMEbus Interface. Each of the interfaces, VMEbus and PCI bus, there are three functionally distinct modules: master module, slave module, and interrupt module. These modules are connected to the different functional channels operating in the Universe II. The device had the following channels: • VMEbus Slave Channel • PCI Bus Target Channel • DMA Channel • Interrupt Channel • Register Channel Figure 3 shows the Universe II in terms of the different modules and channels. Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 1. Functional Overview > Main Interfaces 21 Figure 3: Universe II Data Flow Diagram DMA Channel PCI Bus Interface DMA bidirectional FIFO VMEbus Interface VMEbus Slave Channel PCI Master posted writes FIFO prefetch read FIFO coupled read VME Slave PCI Bus Slave Channel PCI BUS PCI Slave posted writes FIFO coupled read logic VME Master VMEbus Interrupt Channel PCI Interrupts Interrupt Handler Interrupter VME Interrupts 8091142_BK001_01 Register Channel Mailbox Registers Semaphores Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 22 1.2.1 1. Functional Overview > Main Interfaces VMEbus Interface The VME Interface is a VME64 Specification compliant interface. 1.2.1.1 Universe II as VMEbus Slave The Universe II VMEbus Slave Channel accepts all of the addressing and data transfer modes documented in the VME64 Specification - except A64 and those intended to augment 3U applications. Incoming write transactions from the VMEbus can be treated as either coupled or posted, depending upon the programming of the VMEbus slave image (see “VME Slave Image Programming” on page 67). With posted write transactions, data is written to a Posted Write Receive FIFO (RXFIFO), and the VMEbus master receives data acknowledgment from the Universe II. Write data is transferred to the PCI resource from the RXFIFO without the involvement of the initiating VMEbus master (see “Posted Writes” on page 33 for a full explanation of this operation). With a coupled cycle, the VMEbus master only receives data acknowledgment when the transaction is complete on the PCI bus. This means that the VMEbus is unavailable to other masters while the PCI bus transaction is executed. Read transactions may be either prefetched or coupled. A prefetched read is initiated when a VMEbus master requests a block read transaction (BLT or MBLT) and this mode is enabled. When the Universe II receives the block read request, it begins to fill its Read Data FIFO (RDFIFO) using burst transactions from the PCI resource. The initiating VMEbus master then acquires its block read data from the RDFIFO instead of directly from the PCI resources. As VMEbus slave, the Universe II does not assert RETRY* as a termination of the transaction. 1.2.1.2 Universe II as VMEbus Master The Universe II becomes VMEbus master when the VMEbus Master Interface is internally requested by the PCI Bus Target Channel, the DMA Channel, or the Interrupt Channel. The Interrupt Channel always has priority over the other two channels. Several mechanisms are available to configure the relative priority that the PCI Bus Target Channel and DMA Channel have over ownership of the VMEbus Master Interface. The Universe II’s VMEbus Master Interface generates all of the addressing and data transfer modes documented in the VME64 Specification - except A64 and those intended to augment 3U applications. The Universe II is also compatible with all VMEbus modules conforming to pre-VME64 specifications. As VMEbus master, the Universe II supports Read-Modify-Write (RMW), and Address-Only-with-Handshake (ADOH) but does not accept RETRY* as a termination from the VMEbus slave. The ADOH cycle is used to implement the VMEbus Lock command allowing a PCI master to lock VMEbus resources. 1.2.2 PCI Bus Interface The PCI Interface is a PCI 2.1 Specification compliant interface Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 1. Functional Overview > Main Interfaces 1.2.2.1 23 Universe II as PCI Target Read transactions from the PCI bus are always processed as coupled transactions. Write transactions can be either coupled or posted, depending upon the setting of the PCI bus target image (see “PCI Bus Target Images” on page 70). With a posted write transaction, write data is written to a Posted Write Transmit FIFO (TXFIFO) and the PCI bus master receives data acknowledgment from the Universe II with zero wait-states. Meanwhile, the Universe II obtains the VMEbus and writes the data to the VMEbus resource independent of the initiating PCI master (see “Posted Writes” on page 60 for a full description of this operation). The Universe II has a Special Cycle Generator that enables PCI masters to perform RMW and ADOH cycles. The Special Cycle Generator must be used in combination with a VMEbus ownership function to guarantee PCI masters exclusive access to VMEbus resources over several VMEbus transactions (see “Special Cycle Generator” on page 61 and “Using the VOWN bit” on page 64 for a full description of this functionality). 1.2.2.2 Universe II as PCI Master The Universe II becomes PCI master when the PCI Master Interface is internally requested by the VMEbus Slave Channel or the DMA Channel. There are mechanisms provided which allow the user to configure the relative priority of the VMEbus Slave Channel and the DMA Channel. 1.2.3 Interrupter and Interrupt Handler The Universe II has both interrupt generation and interrupt handling capability. 1.2.3.1 Interrupter The Universe II Interrupt Channel provides a flexible scheme to map interrupts to the PCI bus or VMEbus Interface. Interrupts are generated from hardware or software sources (see “Interrupt Generation” on page 111 and “Interrupt Handling” on page 116 for a full description of hardware and software sources). Interrupt sources can be mapped to any of the PCI bus or VMEbus interrupt output pins. Interrupt sources mapped to VMEbus interrupts are generated on the VMEbus interrupt output pins VIRQ_ [7:1]. When a software and hardware source are assigned to the same VIRQ_ pin, the software source always has higher priority. Interrupt sources mapped to PCI bus interrupts are generated on one of the INT_ [7:0] pins. To be fully PCI compliant, all interrupt sources must be routed to a single INT_ pin. For VMEbus interrupt outputs, the Universe II interrupter supplies an 8-bit STATUS/ID to a VMEbus interrupt handler during the IACK cycle. The interrupter also generates an internal interrupt in this situation if the SW_IACK bit, in the PCI Interrupt Status (LINT_STAT) register, is set to 1 (see “VMEbus Interrupt Generation” on page 113). Interrupts mapped to PCI bus outputs are serviced by the PCI interrupt controller. The CPU determines which interrupt sources are active by reading an interrupt status register in the Universe II. The source negates its interrupt when it has been serviced by the CPU (see “PCI Interrupt Generation” on page 111). Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 24 1.2.3.2 1. Functional Overview > Main Interfaces VMEbus Interrupt Handling A VMEbus interrupt triggers the Universe II to generate a normal VMEbus IACK cycle and generate the specified interrupt output. When the IACK cycle is complete, the Universe II releases the VMEbus and the interrupt vector is read by the PCI resource servicing the interrupt output. Software interrupts are ROAK, while hardware, and internal interrupts are RORA. 1.2.4 DMA Controller The Universe II has an internal DMA controller for high performance data transfer between the PCI and VMEbus. DMA operations between the source and destination bus are decoupled through the use of a single bidirectional FIFO (DMAFIFO). Parameters for the DMA transfer are software configurable in the Universe II registers (see “DMA Controller” on page 85). The principal mechanism for DMA transfers is the same for operations in either direction (PCI-to-VMEbus, or VMEbus-to-PCI), only the relative identity of the source and destination bus changes. In a DMA transfer, the Universe II gains control of the source bus and reads data into its DMAFIFO. Following specific rules of DMAFIFO operation (see “FIFO Operation and Bus Ownership” on page 101), it then acquires the destination bus and writes data from its DMAFIFO. The DMA controller can be programmed to perform multiple blocks of transfers using linked-list mode. The DMA works through the transfers in the linked-list following pointers at the end of each linked-list entry. Linked-list operation is initiated through a pointer in an internal Universe II register, but the linked-list itself resides in PCI bus memory. Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 25 2. VMEbus Interface This chapter explains the operation of the VMEbus Interface.This chapter discusses the following topics: 2.1 • “VMEbus Requester” on page 25 • “Universe II as VMEbus Master” on page 28 • “Universe II as VMEbus Slave” on page 32 • “VMEbus Configuration” on page 41 • “Automatic Slot Identification” on page 42 • “System Clock Driver” on page 44 Overview The VMEbus Interface incorporates all operations associated with the VMEbus. This includes master and slave functions, VMEbus configuration and system controller functions. 2.2 VMEbus Requester There are different channels in the Universe II which require the use of the VMEbus. They are referred to as VMEbus requesters and are described in the following sections. 2.2.1 Internal Arbitration for VMEbus Requests Different internal channels within the Universe II require use of the VMEbus: the Interrupt Channel, the PCI Target Channel, and the DMA Channel. These three channels do not directly request the VMEbus, instead they compete internally for ownership of the VMEbus Master Interface. 2.2.1.1 Interrupt Channel The Interrupt Channel always has the highest priority for access to the VMEbus Master Interface (see Figure 3 on page 21). The DMA and PCI Target Channel requests are handled in a fair manner. The channel awarded VMEbus mastership maintains ownership of the VMEbus until it is has completed the transaction. The definition of a complete transaction for each channel is in “VMEbus Release” on page 27. The Interrupt Channel requests the VMEbus master when it detects an enabled VMEbus interrupt line asserted and must run an interrupt acknowledge cycle to acquire the STATUS/ID. Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 26 2.2.1.2 2. VMEbus Interface > VMEbus Requester PCI Target Channel The PCI Target Channel requests the VMEbus Master Interface to service the following conditions: 2.2.1.3 • TXFIFO contains a complete transaction • A coupled cycle request DMA Channel The DMA Channel requests the VMEbus Master Interface in the following instances: • The DMAFIFO has 64 bytes available (if it is reading from the VMEbus) or 64 bytes in its FIFO (if it is writing to the VMEbus) • The DMA block is complete (see “DMA Controller” on page 85). In the case of the DMA Channel, the DMA Channel VMEbus-off-timer can be used to further qualify requests from this channel. The VMEbus-off-timer controls how long the DMA remains off the VMEbus before making another request (see “PCI-to-VMEbus Transfers” on page 101). The Universe II provides a software mechanism for VMEbus acquisition through the VMEbus ownership bit (VOWN in the “Master Control Register (MAST_CTL)” on page 271). When the VMEbus ownership bit is set, the Universe II acquires the VMEbus and sets an acknowledgment bit (VOWN_ACK in the MAST_CTL register) and optionally generates an interrupt to the PCI bus (see “VME Lock Cycles—Exclusive Access to VMEbus Resources” on page 63). The Universe II maintains VMEbus ownership until the ownership bit is cleared. During the VMEbus tenure initiated by setting the ownership bit, only the PCI Target Channel and Interrupt Channel can access the VMEbus Master Interface. 2.2.2 Request Modes The Universe II has configurable request modes of operation. 2.2.2.1 Request Levels The Universe II is software configurable to request on any one of the four VMEbus request levels: BR3*, BR2*, BR1*, and BR0*. The default setting is for level 3 VMEbus request. The request level is a global programming option set through the VMEbus Release Mode (VRL) field in the “Master Control Register (MAST_CTL)” on page 271. The programmed request level is used by the VMEbus Master Interface regardless of the channel (Interrupt Channel, DMA Channel, or PCI Target Channel) currently accessing the VMEbus Master Interface. 2.2.2.2 Fair and Demand Modes The Universe II requester can be programmed for either Fair or Demand mode. The request mode is a global programming option set through the VMEbus Request Mode (VRM) bits in the “Master Control Register (MAST_CTL)” on page 271. In Fair mode, the Universe II does not request the VMEbus until there are no other VMEbus requests pending at its programmed level. This mode ensures that every requester on an equal level has access to the bus. Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 2. VMEbus Interface > VMEbus Requester 27 In the default setting of Demand mode, the requester asserts its bus request regardless of the state of the BRn* line. By requesting the bus frequently, requesters far down the daisy chain may be prevented from ever obtaining bus ownership. This is referred to as starving those requesters. Note that in order to achieve fairness, all bus requesters in a VMEbus system must be set to fair mode. 2.2.3 VMEbus Release The Universe II VMEbus requester can be configured as either RWD (release when done) or ROR (release on request) using the VREL bit in the “Master Control Register (MAST_CTL)” on page 271. The default setting is for RWD. ROR means the Universe II releases BBSY* only if a bus request is pending from another VMEbus master and once the channel that is the current owner of the VMEbus Master Interface is done. Ownership of the bus can be assumed by another channel without re-arbitration on the bus if there are no pending requests on any level on the VMEbus. When set for RWD, the VMEbus Master Interface releases BBSY* when the channel accessing the VMEbus Master Interface is done (see below). Note that the MYBBSY status bit in the “Miscellaneous Status Register (MISC_STAT)” on page 275 is 0 when the Universe II asserts BBSY*. In RWD mode, the VMEbus is released when the channel (for example, the DMA Channel) is done, even if another channel has a request pending (for example, the PCI Target Channel). A re-arbitration of the VMEbus is required for any pending channel requests. Each channel has a set of rules that determine when it is ‘done’ with its VMEbus transaction. 2.2.3.1 Transaction Complete The interrupt is complete when a single interrupt acknowledge cycle is complete. The PCI Target Channel is complete under the following conditions: • when the TXFIFO is empty, the TXFE bit is clear (the TXFE bit is set by the Universe II in the MISC_STAT register, • when the maximum number of bytes per PCI Target Channel tenure has been reached (as programmed with the PWON field in the “Master Control Register (MAST_CTL)” on page 2711, • after each posted write, if the PWON is equal to 0b1111, as programmed in the MAST_CTL register • when the coupled cycle is complete and the Coupled Window Timer has expired, • if the Coupled Request Timer (page 58) expires before a coupled cycle is retried by a PCI master, or • when VMEbus ownership is acquired with the VOWN bit in the MAST_CTL register and then the VOWN bit is cleared (in other words, if the VMEbus is acquired through the use of the VOWN bit, the Universe II does not release BBSY* until the VOWN bit is cleared—see “VME Lock Cycles—Exclusive Access to VMEbus Resources” on page 63). The DMA Channel is complete under the following conditions: • DMAFIFO full during VMEbus to PCI bus transfers, • DMAFIFO empty during PCI bus to VMEbus transfers, 1. This setting is overridden if the VOWN mechanism is used. Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 28 2. VMEbus Interface > Universe II as VMEbus Master • if an error is encountered during the DMA operation, • the DMA VMEbus Tenure Byte Counter has expired, or • DMA block is complete. Refer to “FIFO Operation and Bus Ownership” on page 101 and “DMA Error Handling” on page 105 for more information. Universe II ownership of the VMEbus is not affected by the assertion of BCLR* because it does not monitor BCLR*. 2.3 Universe II as VMEbus Master The Universe II becomes VMEbus master under the following circumstances: 1. PCI master accesses a Universe II PCI target image (leading to VMEbus access) or the DMA Channel initiates a transaction, 2. Either the Universe II PCI Target Channel or the DMA Channel wins access to the VMEbus Master Interface through internal arbitration 3. Universe II Master Interface requests and obtains ownership of the VMEbus The Universe II also becomes VMEbus master if the VMEbus ownership bit is set (see “VME Lock Cycles—Exclusive Access to VMEbus Resources” on page 63) and in its role in VMEbus interrupt handling (see “VMEbus Interrupt Handling” on page 116). The following sections describe the function of the Universe II as a VMEbus master in terms of the different phases of a VMEbus transaction: addressing, data transfer, cycle termination, and bus release. 2.3.1 Addressing Capabilities Depending upon the programming of the PCI target image (see “PCI Bus Target Images” on page 70), the Universe II generates A16, A24, A32, and CR/CSR address phases on the VMEbus. The address mode and type (supervisor/non-privileged and program/data) are also programmed through the PCI target image. Address pipelining is provided, except during MBLT cycles. The VMEbus Specification does not permit pipelining during MBLT cycles. The address and Address Modifier (AM) codes that are generated by the Universe II are functions of the PCI address and PCI target image programming (see “PCI Bus Target Images” on page 70) or through DMA programming. Table 1 shows the AM codes used for the VMEbus. Table 1: VMEbus Address Modifier Codes Address Modifier Address Bits 0x3F 24 A24 supervisory block transfer (BLT) 0x3E 24 A24 supervisory program access 0x3D 24 A24 supervisory data access 0x3C 24 A24 supervisory 64-bit block transfer (MBLT) Universe II User Manual May 12, 2010 Description Integrated Device Technology www.idt.com 2. VMEbus Interface > Universe II as VMEbus Master 29 Table 1: VMEbus Address Modifier Codes Address Modifier Address Bits 0x3B 24 A24 non-privileged block transfer (BLT) 0x3A 24 A24 non-privileged program access 0x39 24 A24 non-privileged data access 0x38 24 A24 non-privileged 64-bit block transfer (MBLT) 0x37 40 A40BLT [MD32] 0x35 40 A40 lock command (LCK) 0x34 40 A40 access 0x32 24 A24 lock command 0x2F 24 CR/CSR 0x2D 16 A16 supervisory access 0x2C 16 A16 lock command 0x29 16 A16 non-privileged access 0x21 32 or 40 2eVME for 3U bus modules (address size in XAM code) 0x20 32 or 40 2eVME for 6U bus modules (address size in XAM code) 0x10 - 0x1F Undefined 0xF 32 A32 supervisory block transfer (BLT) 0xE 32 A32 supervisory program access 0xD 32 A32 supervisory data access 0xC 32 A32 supervisory 64-bit block transfer (MBLT) 0xB 32 A32 non-privileged block transfer (BLT) 0xA 32 A32 non-privileged program access 0x9 32 A32 non-privileged data access 0x8 32 A32 non-privileged 64-bit block transfer (MBLT) 0x5 32 A32 lock command 0x4 64 A64 lock command 0x3 64 A64 block transfer (BLT) 0x1 64 A64 single transfer access 0x0 64 A64 64-bit block transfer (MBLT) Integrated Device Technology www.idt.com Description User-defined Universe II User Manual May 12, 2010 30 2. VMEbus Interface > Universe II as VMEbus Master The Universe II generates Address-Only-with-Handshake (ADOH) cycles in support of lock commands for A16, A24, and A32 spaces. ADOH cycles can only be generated through the Special Cycle Generator (see “Special Cycle Generator” on page 61). There are two User Defined AM codes that can be programmed through the “User AM Codes Register (USER_AM)” on page 276. The USER_AM register can only be used to generate and accept AM codes 0x10 through 0x1F. The default USER_AM code is 0x10. These AM codes are designated as USERAM codes in the VMEbus Specification. After power-up, the two values in the USER_AM register default to the same VME64 User-defined AM code. If USER_AM code is used with the VMEbus Slave Interface, the cycles must use 32-bit addressing, and only single cycle accesses are used. BLTs and MBLTs with USER_AM codes will lead to unpredictable behavior. 2.3.2 Data Transfer Capabilities PCI and VMEbus protocols have different data transfer capabilities. The maximum data width for a VMEbus data transfer is programmed with the VMEbus Maximum Datawidth (VDW) field in the “PCI Target Image 0 Control (LSI0_CTL)” on page 181. For example, consider a 32-bit PCI transaction accessing a PCI target image with VDW set to 16 bits. A data beat with all byte lanes enabled will be broken into two 16-bit cycles on the VMEbus. If the PCI target image is also programmed with block transfers enabled, the 32-bit PCI data beat will result in a D16 block transfer on the VMEbus. Write data is unpacked to the VMEbus and read data is packed to the PCI bus data width. If the data width of the PCI data beat is the same as the maximum data width of the PCI target image, then the Universe II maps the data beat to an equivalent VMEbus cycle. For example, consider a 32-bit PCI transaction accessing a PCI target image with VDW set to 32 bits. A data beat with all byte lanes enabled is translated to a single 32-bit cycle on the VMEbus. As the general rule, if the PCI bus data width is less than the VMEbus data width then there is no packing or unpacking between the two buses. The only exception to this is during 32-bit PCI multi-data beat transactions to a PCI target image programmed with maximum VMEbus data width of 64 bits. In this case, packing/unpacking occurs to make maximum use of the full bandwidth on both buses. Only aligned VMEbus transactions are generated, so if the requested PCI data beat has unaligned, or non-, byte enables, then it is broken into multiple aligned VMEbus transactions no wider than the programmed VMEbus data width. For example, consider a three-byte PCI data beat (on a 32-bit PCI bus) accessing a PCI target image with VDW set to 16 bits. The three-byte PCI data beat is broken into three aligned VMEbus cycles: three single-byte cycle (the ordering of the cycles depends on the arrangement of the byte enables in the PCI data beat). If in the above example the PCI target image has a VDW set to 8-bit, then the three-byte PCI data beat is broken into three single-byte VMEbus cycles. Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 2. VMEbus Interface > Universe II as VMEbus Master 31 BLT/MBLT cycles are initiated on the VMEbus if the PCI target image has been programmed with this capacity (see “PCI Bus Target Images” on page 70). The length of the BLT/MBLT transactions on the VMEbus is determined by the initiating PCI transaction. For example, a single data beat PCI transaction queued in the TXFIFO results in a single data beat block transfer on the VMEbus. With the PWON field, the user can specify a transfer byte count that is queued from the TXFIFO before the VMEbus Master Interface relinquishes the VMEbus. The PWON field specifies the minimum tenure of the Universe II on the VMEbus. However, tenure is extended if the VOWN bit in the MAST_CTL register is set (see “Using the VOWN bit” on page 64). During DMA operations, the Universe II attempts block transfers to the maximum length permitted by the VMEbus specification (256 bytes for BLT, 2 Kbytes for MBLT) and is limited by the VON counter (see “DMA VMEbus Ownership” on page 91). The Universe II provides indivisible transactions with the VMEbus lock commands and the VMEbus ownership bit (see “VME Lock Cycles—Exclusive Access to VMEbus Resources” on page 63). 2.3.3 Cycle Terminations The Universe II accepts BERR* or DTACK* as cycle terminations from the VMEbus slave. It does not support RETRY*. The assertion of BERR* indicates that some type of system error occurred and the transaction did not complete properly. The assertion of BERR* during an IACK also causes the error to be logged. A VMEbus BERR* received by the Universe II during a coupled transaction is communicated to the PCI master as a Target-Abort. No information is logged if the Universe II receives BERR* in a coupled transaction. If an error occurs during a posted write to the VMEbus or during an IACK cycle, the Universe II uses the “VMEbus AM Code Error Log (V_AMERR)” on page 307 to log the AM code of the transaction (AMERR [5:0]), and the state of the IACK* signal (IACK bit, to indicate whether the error occurred during an IACK cycle). The current transaction in the FIFO is purged. The V_AMERR register also records if multiple errors have occurred (with the M_ERR bit), although the actual number of errors is not given. The error log is qualified by the value of the V_STAT bit. The address of the errored transaction is latched in the “VMEbus Address Error Log (VAERR)” on page 308. When the Universe II receives a VMEbus error during a posted write, it generates an interrupt on the VMEbus and/or PCI bus depending upon whether the VERR and LERR interrupts are enabled (see “Interrupt Handling” on page 116). DTACK* signals the successful completion of the transaction. Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 32 2.4 2. VMEbus Interface > Universe II as VMEbus Slave Universe II as VMEbus Slave This section describes the VMEbus Slave Channel and other aspects of the Universe II as VMEbus slave. The Universe II becomes VMEbus slave when one of its eight programmed slave images or register images are accessed by a VMEbus master. Depending upon the programming of the slave image, different possible transaction types can result (see “VME Slave Image Programming” on page 67). The Universe II cannot reflect a cycle on the VMEbus and access itself. For reads, the transaction can be coupled or prefetched. Write transactions can be coupled or posted. The type of read or write transaction allowed by the slave image depends on the programming of that particular VMEbus slave image (see Figure 4 and “VME Slave Image Programming” on page 67). To ensure sequential consistency, prefetched reads, coupled reads, and coupled write operations are only processed once all previously posted write operations have completed (the RXFIFO is empty). Figure 4: VMEbus Slave Channel Dataflow PREFETCHED READ DATA RDFIFO PCI BUS MASTER INTERFACE COUPLED READ DATA COUPLED WRITE DATA VMEbus SLAVE INTERFACE POSTED WRITE DATA RXFIFO Incoming cycles from the VMEbus can have data widths of 8-bit, 16-bit, 32-bit, and 64-bit. Although the PCI bus supports only two port sizes (32-bit and 64-bit), the byte lanes on the PCI bus can be individually enabled, which allows each type of VMEbus transaction to be directly mapped to the PCI data bus. In order for a VMEbus slave image to respond to an incoming cycle, the PCI Master Interface must be enabled by setting the bit BM in the “PCI Configuration Space Control and Status Register (PCI_CSR)” on page 172. If data is queued in the VMEbus Slave Channel FIFO and the BM bit is cleared, the FIFO empties but no additional transfers are received. 2.4.1 Coupled Transfers A coupled transfer means that no FIFO is involved in the transaction and handshakes are relayed directly through the Universe II. Coupled mode is the default setting for the VMEbus slave images. Coupled transfers only proceed once all posted write entries in the RXFIFO have completed (see “Posted Writes” on page 33). Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 2. VMEbus Interface > Universe II as VMEbus Slave 33 A coupled cycle with multiple data beats (such as block transfers) on the VMEbus side is always mapped to single data beat transactions on the PCI bus, where each data beat on the VMEbus is mapped to a single data beat transaction on the PCI bus regardless of data beat size. No packing or unpacking is performed. The only exception to this is when a D64 VMEbus transaction is mapped to D32 on the PCI bus. The data width of the PCI bus depends on the programming of the VMEbus slave image (32-bit or 64-bit, see “VME Slave Image Programming” on page 67). The Universe II enables the appropriate byte lanes on the PCI bus as required by the VMEbus transaction. For example, a VMEbus slave image programmed to generate 32-bit transactions on the PCI bus is accessed by a VMEbus D08 BLT read transaction (prefetching is not enabled in this slave image). The transaction is mapped to single data beat 32-bit transfers on the PCI bus with only one byte lane enabled. Target-Retry from a PCI target is not communicated to the VMEbus master. PCI transactions terminated with Target-Abort or Master-Abort are terminated on the VMEbus with BERR*. The Universe II sets the R_TA or R_MA bits in the “PCI Configuration Space Control and Status Register (PCI_CSR)” on page 172 when it receives a Target-Abort or Master-Abort. 2.4.2 Posted Writes A posted write involves the VMEbus master writing data into the Universe II’s RXFIFO, instead of directly to the PCI address. Write transactions from the VMEbus are processed as posted if the PWEN bit is set in the VMEbus slave image control register (see “VME Slave Image Programming” on page 67). If the bit is cleared (default setting) the transaction bypasses the FIFO and is performed as a coupled transfer. Incoming posted writes from the VMEbus are queued in the 64-entry deep RXFIFO. Each entry in the RXFIFO can contain 32 address bits, or 64 data bits. Each incoming VMEbus address phase, whether it is 16-bit, 24-bit, or 32-bit, constitutes a single entry in the RXFIFO and is followed by subsequent data entries. The address entry contains the translated PCI address space and command information mapping relevant to the particular VMEbus slave image that has been accessed (see “VME Slave Image Programming” on page 67). For this reason, any reprogramming of VMEbus slave image attributes are only reflected in RXFIFO entries queued after the reprogramming. Transactions queued before the re-programming are delivered to the PCI bus with the VMEbus slave image attributes that were in use before the reprogramming. The RXFIFO is the same structure as the RDFIFO. The different names are used for the FIFO’s two roles. In each FIFO, only one role, either the RXFIFO or the RDFIFO, can used at one time. 2.4.2.1 FIFO Entries Incoming non-block write transactions from the VMEbus require two entries in the RXFIFO: one address entry (with accompanying command information) and one data entry. The size of the data entry corresponds to the data width of the VMEbus transfer. Block transfers require at least two entries: one entry for address and command information, and one or more data entries. The VMEbus Slave Channel packs data received during block transfers to the full 64-bit width of the RXFIFO. For example, a ten data phase D16 BLT transfer (20 bytes in total) does not require ten data entries in the RXFIFO. Instead, eight of the ten data phases (16 bits per data phase for a total of 128 bits) are packed into two 64-bit data entries in the RXFIFO. The final two data phases (32 bits combined) are queued in the next RXFIFO entry. When the address entry is added to the three data entries, this VMEbus block write has been stored in a total of five RXFIFO entries. Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 34 2. VMEbus Interface > Universe II as VMEbus Slave Unlike the PCI Target Channel (see “Universe II as PCI Target”), the VMEbus Slave Channel does not retry the VMEbus if the RXFIFO does not have enough space to hold an incoming VMEbus write transaction. Instead, the DTACK* response from the VMEbus Slave Interface is delayed until space becomes available in the RXFIFO. Since single transfers require two entries in the RXFIFO, two entries must be available before the VMEbus Slave Interface asserts DTACK*. Similarly, the VMEbus Slave Channel requires two available RXFIFO entries before it can acknowledge the first data phase of a BLT or MBLT transfer (one entry for the address phase and one for the first data phase). If the RXFIFO has no available space for subsequent data phases in the block transfer, then the VMEbus Slave Interface delays assertion of DTACK* until a single entry is available for the next data phase in the block transfer. The PCI Master Interface uses transactions queued in the RXFIFO to generate transactions on the PCI bus. No address phase deletion is performed, so the length of a transaction on the PCI bus corresponds to the length of the queued VMEbus transaction. Non-block transfers are generated on the PCI bus as single data beat transactions. Block transfers are generated as one or more burst transactions, where the length of the burst transaction is programmed by the PABS field in the “Master Control Register (MAST_CTL)” on page 271. The Universe II always packs or unpacks data from the VMEbus transaction to the PCI bus data width programmed into the VMEbus slave image (with all PCI bus byte lanes enabled). The data width for a VMEbus transaction to the PCI bus is programmed in the LD64EN bit in the “VMEbus Slave Image 1 Control (VSI1_CTL)” on page 285. The LD64EN bit enables 64-bit PCI bus transactions For example, consider a VMEbus slave image programmed for posted writes and a D32 PCI bus that is accessed with a VMEbus D16 block write transaction. VMEbus D16 write transactions are mapped to D32 write transactions on the PCI bus with all byte lanes enabled. (However, a single D16 transaction from the VMEbus is mapped to the PCI bus as D32 with only two byte lanes enabled). During block transfers, the Universe II packs data to the full negotiated width of the PCI bus. This may imply that for block transfers that begin or end on addresses not aligned to the PCI bus width different byte lanes may be enabled during each data beat. 2.4.2.2 Errors If an error occurs during a posted write to the PCI bus, the Universe II uses the “PCI Command Error Log Register (L_CMDERR)” on page 210 to log the command information for the transaction (CMDERR [3:0]). The L_CMDERR register also records if multiple errors have occurred (with the M_ERR bit) although the actual number of errors is not given. The error log is qualified with the L_STAT bit. The address of the errored transaction is latched in the “PCI Address Error Log (LAERR)” on page 211. An interrupt is generated on the VMEbus and/or PCI bus depending upon whether the VERR and LERR interrupts are enabled (see “Error Handling” on page 125 and “Interrupt Generation and Handling” on page 109). 2.4.3 Prefetched Block Reads Prefetching of read data occurs for VMEbus block transfers (BLT, MBLT) in those slave images that have the Prefetch Enable (PREN) bit set (see “VME Slave Image Programming” on page 67). In the VMEbus Slave Channel, prefetching is not supported for non- BLT/MBLT transfers. Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 2. VMEbus Interface > Universe II as VMEbus Slave 35 Without prefetching, block read transactions from a VMEbus master are handled by the VMEbus Slave Channel as coupled reads. This means that each data phase of the block transfer is translated to a single data beat transaction on the PCI bus. In addition, only the amount of data requested during the relevant data phase is fetched from the PCI bus. For example, a D16 block read transaction with 32 data phases on the VMEbus maps to 32 PCI bus transactions, where each PCI bus transaction has only two byte lanes enabled. The VMEbus lies idle during the arbitration time required for each PCI bus transaction, resulting in a performance degradation. With prefetching enabled, the VMEbus Slave Channel uses a 64-entry deep RDFIFO to provide read data to the VMEbus with minimum latency. The RDFIFO is 64-bit, with additional bits for control information. If a VMEbus slave image is programmed for prefetching (see “VME Slave Image Programming” on page 67), then a block read access to that image causes the VMEbus Slave Channel to generate aligned burst read transactions on the PCI bus (the size of the burst read transactions is determined by the setting of the aligned burst size, PABS in the MAST_CTL register). These PCI burst read transaction are queued in the RDFIFO and the data is then delivered to the VMEbus. The first data phase provided to the VMEbus master is essentially a coupled read, but subsequent data phases in the VMEbus block read are delivered from the RDFIFO and are decoupled (see “Prefetched Reads” on page 127 for the impact on bus error handling). The RXFIFO is the same structure as the RDFIFO. The different names are used for the FIFO’s two roles. In each FIFO, only one role, either the RXFIFO or the RDFIFO, can used at one time. 2.4.3.1 FIFO Entries When there is a transactions from the Universe PCI Slave to the Universe VME Master, the data width of the transaction on the PCI bus (32-bit or 64-bit) depends on the setting of the LD64EN bit in the “VMEbus Slave Image 1 Control (VSI1_CTL)” on page 285 and the capabilities of the accessed PCI target. Internally, the prefetched read data is packed to 64-bit, regardless of the width of the PCI bus or the data width of the original VMEbus block read (no address information is stored with the data). Once one entry is queued in the RDFIFO, the VMEbus Slave Interface delivers the data to the VMEbus, unpacking the data as necessary to fit with the data width of the original VMEbus block read (D16, or D32). The VMEbus Slave Interface continuously delivers data from the RDFIFO to the VMEbus master performing the block read transaction. Because PCI bus data transfer rates exceed those of the VMEbus, it is unlikely that the RDFIFO will be unable to deliver data to the VMEbus master. For this reason, block read performance on the VMEbus is similar to that observed with block writes. However, if the RDFIFO be unable to deliver data to the VMEbus master (which can happen if there is considerable traffic on the PCI bus or the PCI bus target has a slow response) the VMEbus Slave Interface delays DTACK* assertion until an entry is queued and is available for the VMEbus block read. Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 36 2. VMEbus Interface > Universe II as VMEbus Slave On the PCI bus, prefetching continues as long as there is room for another transaction in the RDFIFO and the initiating VMEbus block read is still active. The space required in the RDFIFO for another PCI burst read transaction is determined by the setting of the PCI aligned burst size (PABS in the “Master Control Register (MAST_CTL)” on page 271). If PABS is set for 32 bytes, there must be four entries available in the RDFIFO; for aligned burst size set to 64 bytes, eight entries must be available, for aligned burst size set to 128 bytes, there must be 16 entries available. When there is insufficient room in the RDFIFO to hold another PCI burst read, the read transactions on the PCI bus are terminated and only resume if room becomes available for another aligned burst and the original VMEbus block read is still active. When the VMEbus block transfer terminates, any remaining data in the RDFIFO is removed. Reading on the PCI bus does not cross a 2048-byte boundary. The PCI Master Interface releases FRAME_ and the VMEbus Slave Channel relinquishes internal ownership of the PCI Master Interface when it reaches this boundary. The VMEbus Slave Channel re-requests internal ownership of the PCI Master Interface as soon as possible, in order to continue reading from the external PCI target. The PABS setting determines how much data must be available in the RDFIFO before the VMEbus Slave Channel continues reading. Regardless of the read request, the data width of prefetching on the PCI side is full width with all byte lanes enabled. If LD64EN is set in the VMEbus Slave image, the Universe II requests D64 on the PCI bus by asserting REQ64_ during the address phase. If the PCI target does not respond with ACK64_, subsequent data beats are D32. 2.4.3.2 Errors If an error occurs on the PCI bus, the Universe II does not translate the error condition into a BERR* on the VMEbus; the Universe II does not directly map the error. By doing nothing, the Universe II forces the external VMEbus error timer to expire. 2.4.4 VMEbus Lock Commands (ADOH Cycles) The Universe II supports VMEbus lock commands as described in the VME64 Specification. Under the specification, ADOH cycles are used to execute the lock command with a special AM code (see Table 1 on page 28). The purpose of the Lock command is to lock the resources on a card so a master on the card cannot modify the resource. Any resource locked on the VMEbus cannot be accessed by any other resource during the bus tenure of the VMEbus master. When the Universe II receives a VMEbus lock command, it asserts LOCK_ to the addressed resource on the PCI bus. The PCI Master Interface processes this as a 32-bit read transfer with all byte lanes enabled (no data). All subsequent slave VMEbus transactions are coupled while the Universe II owns PCI LOCK_. The Universe II holds the PCI bus lock until the VMEbus lock command is terminated (by negating BBSY*). The VMEbus Slave Channel has dedicated access to the PCI Master Interface during the locked transaction. Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 2. VMEbus Interface > Universe II as VMEbus Slave 37 The Universe II accepts ADOH cycles in any of the slave images when the Universe II PCI Master Interface is enabled (BM bit in PCI_CSR register) and the images are programmed to map transactions into PCI Memory Space. 2.4.4.1 Errors If an error occurs on the PCI bus, a bus error will occur on the VMEbus because they are coupled. In the event a bus error occurs on the VMEbus once a LOCK_ has been established, the VMEbus master which locked the VMEbus must terminate the LOCK_ by negating BBSY*. 2.4.4.2 DMA Access Once an external VMEbus masters locks the PCI bus, the Universe II DMA does not perform transfers on the PCI bus until the bus is unlocked. LOCK_ is negated on the PCI bus when AS* is negated on the VMEbus. LOCK_ is not negated when AS* is negated if LOCK_ was asserted by an ADOH/lock command 2.4.5 VMEbus Read-Modify-Write Cycles (RMW Cycles) A read-modify-write (RMW) cycle allows a VMEbus master to read from a VMEbus slave and then write to the same resource without relinquishing bus tenure between the two operations. Each of the Universe II slave images can be programmed to map RMW transactions to PCI locked transactions. If the LLRMW enable bit is set in the selected “VMEbus Slave Image 1 Control (VSI1_CTL)” on page 285), then every non-block slave read is mapped to a coupled PCI locked read. LOCK_ is held on the PCI bus until AS* is negated on the VMEbus. Every non-block slave read is assumed to be a RMW since there is no possible indication from the VMEbus master that the single cycle read is just a read or the beginning of a RMW. RMW cycles are not supported with unaligned or D24 cycles. If the LLRMW enable bit is not set and the Universe II receives a VMEbus RMW cycle, the read and write portions of the cycle are treated as independent transactions on the PCI bus (a read followed by a write). The write can be coupled or decoupled depending on the state of the PWEN bit in the accessed slave image. There can be an adverse performance impact for reads that are processed through a RMW-capable slave image This can be increased if LOCK_ is currently owned by another PCI master. When an external VMEbus Master begins a RMW cycle, at some point a read cycle appears on the PCI bus. During the time between when the read cycle occurs on the PCI bus and when the associated write cycle occurs on the PCI bus, no DMA transfers occurs on the PCI bus. 2.4.6 Register Accesses See “Registers” on page 163 for a full description of register mapping and register access. Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 38 2.4.7 2. VMEbus Interface > Universe II as VMEbus Slave Location Monitors Universe II has four location monitors to support a VMEbus broadcast capability. The location monitors’ image is a 4-Kbyte image in A16, A24 or A32 space on the VMEbus. If enabled, an access to a location monitor causes the PCI Master Interface to generate an interrupt. The “Location Monitor Control Register (LM_CTL)” on page 300 controls the Universe II’s location monitoring. The EN field of the LM_CTL register enables the capability. The PGM[1:0] field sets the Program/Data AM code. The SUPER[1:0] field of the LM_CTL register sets the Supervisor/User AM code to which the Universe II responds. The VAS[3:0] field of the LM_CTL register specifies the address space that is monitored. The BS[31:12] field of the “Location Monitor Base Address Register (LM_BS)” on page 302 specifies the lowest address in the 4 Kbyte range that is decoded as a location monitor access. While the Universe II is has four location monitors, they all share the same LM_CTL and LM_BS registers. In address spaces A24 and A16, the respective upper address bits are ignored. When an access to a location monitor is detected, an interrupt is generated on the PCI bus. VMEbus address bits [4:3] determine which Location Monitor is used, and hence which of four PCI interrupts to generate (see “Location Monitors” on page 122). The location monitors do not store write data. Read data from the location monitors is undefined. Location monitors do not support BLT or MBLT transfers. Each Universe II on the VMEbus must be programmed to monitor the same 4 Kbytes of addresses on the VMEbus. If the Universe II accesses its own (enabled) location monitor, the same Universe II generates DTACK* on the VMEbus and terminates its own cycle. This removes the necessity of the system integrator ensuring that there is another card enabled to generate DTACK*. The generation of DTACK* happens after the Universe II has decoded and responded to the cycle. If the location monitor is accessed by a different master, the Universe II does not respond with DTACK*. 2.4.8 Generating PCI Configuration Cycles PCI Configuration cycles can be generated by accessing a VMEbus slave image whose Local Address Space field (LAS) is set for Configuration Space. ADOH, BLT and MBLT cycles must not be attempted when the LAS field of an image is programmed for PCI Configuration Space. PCI Configuration cycles can only be generated when the VAS field in the appropriate VSIx_CTRL register is programmed for either A32, USER1, or USER2. Both Type 0 and Type 1 cycles are generated and handled through the same mechanism. Once a VMEbus cycle is received and mapped to a configuration cycle, the Universe II compares bits [23:16] of the incoming address with the value stored in the Bus Number field (BUS_NO[7:0]) in the “Master Control Register (MAST_CTL)” on page 271. If the bits are the same as the BUS_NO field, then a TYPE 0 access is generated. If they are not the same, a Type 1 configuration access is generated. The PCI bus-generated address then becomes an unsigned addition of the incoming VMEbus address and the VMEbus slave image translation offset. Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 2. VMEbus Interface > Universe II as VMEbus Slave 2.4.8.1 39 Generating Configuration Type 0 Cycles The Universe II asserts one of AD[31:11] on the PCI bus to select a device during a configuration Type 0 access. To perform a configuration Type 0 cycle on the PCI bus, the following steps must be completed: 1. Program the LAS field of VSIx_CTL for Configuration Space 2. Program the VSIx_BS, VSIx_BD registers to some suitable value 3. Program the VSIx_TO register to 0 4. Program the BUS_NO field of the MAST_CTL register to some value Perform a VMEbus access where: • VA[7:2] identifies the PCI Register Number and will be mapped directly to AD[7:2] • VA[10:8] identifies the PCI Function Number and will be mapped directly to AD[10:8] • VA[15:11] selects the device on the PCI bus and will be mapped to AD[31:12] according to Table 2 • VA[23:16] matches the BUS_NO in MAST_CTL register • Other address bits are not important—they are not mapped to the PCI bus Table 2: PCI Address Line Asserted as a Function of VA[15:11] VA[15:11]a PCI Address Line Assertedb 00000 11 00001 12 00010 13 00011 14 00100 15 00101 16 00110 17 00111 18 01000 19 01001 20 01010 21 01011 22 01100 23 01101 24 01110 25 Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 40 2. VMEbus Interface > Universe II as VMEbus Slave Table 2: PCI Address Line Asserted as a Function of VA[15:11] VA[15:11]a PCI Address Line Assertedb 01111 26 10000 27 10001 28 10010 29 10011 30 10100 31 a. The other values of VA[15:11] are not defined and must not be used. b. Only one of AD[31:11] is asserted; the other address lines in AD[31:11] are negated. ADOH, BLT and MBLT cycles must not be attempted when the LAS of an image is programmed to PCI Configuration space. 2.4.8.2 Generating Configuration Type 1 Cycles The following steps are used to generate a configuration Type 1 cycle on the VMEbus: 1. Program LAS field of VSIx_CTL to Configuration Space 2. Program the VSIx_BS, VSIx_BD registers to some suitable value 3. Program the VSIx_TO register to 0 4. Program the BUS_NO field of the MAST_CTL register to some value Perform a VMEbus access where: • VMEbus Address[7:2] identifies the PCI Register Number, • VMEbus Address[10:8] identifies the PCI Function Number, • VMEbus Address[15:11] identifies the PCI Device Number, • VMEbus Address[23:16] does not match the BUS_NO in MAST_CTL register, and • VMEbus Address[31:24] are mapped directly through to the PCI bus. Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 2. VMEbus Interface > VMEbus Configuration 2.5 41 VMEbus Configuration The Universe II provides the following functions to assist in the initial configuration of the VMEbus system: 2.5.1 • First Slot Detector • Register Access at Power-up • Auto Slot ID (two methods) First Slot Detector As specified by the VME64 Specification the First Slot Detector module on the Universe II samples BG3IN* immediately after reset to determine whether the Universe II’s host board resides in slot 1. The VME64 Specification requires that BG[3:0]* lines be driven high after reset. This means that if a card is preceded by another card in the VMEbus system, it always sample BG3IN* high after reset. BG3IN* can only be sampled low after reset by the first card in the system — there is no preceding card to drive BG3IN* high. If BG3IN* is sampled at logic low immediately after reset (due to the Universe II’s internal pull-down), then the Universe II’s host board is in slot 1 and the Universe II becomes SYSCON: otherwise, the SYSCON module is disabled. This mechanism may be overridden by software through clearing or setting the SYSCON bit in the “Miscellaneous Control Register (MISC_CTL)” on page 273. The Universe II monitors IACK*, instead of IACKIN*, when it is configured as SYSCON. This permits it to operate as SYSCON in a VMEbus chassis slot other than slot 1, provided there are only empty slots to its left. The slot with SYSCON in it becomes a virtual slot 1. 2.5.2 VMEbus Register Access at Power-up The Universe II provides a VMEbus slave image that allows access to all Universe II registers. The base address for the slave image is programmed through the “VMEbus Register Access Image Base Address Register (VRAI_BS)” on page 304. At power-up, the Universe II can program the VRAI_BS and “VMEbus Register Access Image Control Register (VRAI_CTL)” on page 303 registers with information specifying the Universe II Control/Status (UCSR) register slave image (see “Power-Up Options” on page 135). Register access at power-up is used in systems where the Universe II’s card has no CPU, or where register access for that card needs to be independent of the local CPU. Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 42 2.6 2. VMEbus Interface > Automatic Slot Identification Automatic Slot Identification The Universe II supports two types of Auto ID functionality. One type uses the Auto Slot ID technique as described in the VME64 Specification. The other type uses a proprietary method developed by DY4 Systems and implemented in the IDT SCV64 and the Universe II. Neither system identifies geographical addressing, only the relative position amongst the boards present in the system, for example, fourth board versus fourth slot. Both the VME64 Auto Slot ID and the DY4 method of automatic slot identification are activated through a power-up option. Refer “Power-Up Options” on page 135 for more information. Auto ID prevents the need for jumpers to uniquely identify cards in a system. The benefits of this feature are: 2.6.1 • Increases the speed of system level repairs in the field • Reduces the possibility of incorrect configurations • Reduces the number of unique spare cards that must be stocked Auto Slot ID: VME64 Specified The VME64 Auto Slot ID cycle, as described in the VME64 Specification, requires at power-up that the Auto ID slave takes the following actions: • Generate IRQ2* • Negate SYSFAIL* When the Auto Slot ID slave responds to the Monarch’s IACK cycle, the following actions are taken: 1. Enable accesses to its CR/CSR space 2. Provide a Status/ID to the Monarch indicating the interrupt is an Auto-ID request 3. Assert DTACK* 4. Release IRQ2* The Universe II participates in the VME64 Auto Slot ID cycle in either an automatic or semi-automatic mode. In its fully automatic mode, it holds SYSFAIL* asserted until SYSRST* is negated. When SYSRST* is negated, the Universe II asserts IRQ2* and releases SYSFAIL*. In its semi-automatic mode, the Universe II still holds SYSFAIL* asserted until SYSRST* is negated. However, when SYSRST* is negated, the local CPU performs diagnostics and local logic sets the AUTOID bit in the “Miscellaneous Control Register (MISC_CTL)” on page 273. This asserts IRQ2* and releases SYSFAIL*. After SYSFAIL* is released and the Universe II detects a level 2 IACK cycle, it responds with the STATUS/ID stored in its STATID register. The default value is 0xFE. Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 2. VMEbus Interface > Automatic Slot Identification 43 The Universe II can be programmed so that it does not release SYSFAIL* until the SYSFAIL bit in the “VMEbus CSR Bit Clear Register (VCSR_CLR)” on page 329 is cleared by local logic. SYSFAIL* is asserted if the SYSFAIL bit in the “VMEbus CSR Bit Set Register (VCSR_SET)” on page 330 is set at power-up. Since the system Monarch does not service the Auto-ID slave until after SYSFAIL* is negated, not clearing the SYSFAIL bit allows the Auto-ID process to be delayed until the CPU completes local diagnostics. Once local diagnostics are complete, the CPU clears the SYSFAIL bit and the Auto Slot ID cycle proceeds. The Monarch can perform CR/CSR reads and writes at A[23:19]= 0x00 in CR/CSR space and re-locate the Universe II’s CR/CSR base address. 2.6.1.1 Universe II and the Auto Slot ID Monarch At power-up an Auto Slot ID Monarch waits to run a IACK cycle until after SYSFAIL* goes high. After the IACK cycle is performed and it has received a Status/ID indicating an Auto Slot ID request, the monarch software does the following: 1. Masks IRQ2* — It will not service other interrupters at that interrupt level until current Auto Slot ID cycle is completed 2. Performs an access at 0x00 in CR/CSR space to get information about Auto Slot ID slave 3. Moves the CR/CSR base address to a new location 4. Unmasks IRQ2* (to allow it to service the next Auto-ID slave) The Universe II supports monarch activity through its capability to be a level 2 interrupt handler. All other activity must be handled through software residing on the board. 2.6.2 Auto-ID: A Proprietary IDT Method The Universe II uses a proprietary Auto-ID scheme when enabled through a power-up option (see “Auto-ID” on page 137). The IDT proprietary Auto-ID function identifies the relative position of each board in the system, without using jumpers or on-board information. The ID number generated by Auto-ID can then be used to determine the board’s base address. After any system reset (assertion of SYSRST*), the Auto-ID logic responds to the first level one IACK cycle on the VMEbus. After the level one IACK* signal has been asserted (either through IRQ1* or with a synthesized version), the Universe II in slot 1 counts five clocks from the start of the cycle and then asserts IACKOUT* to the second board in the system (see Figure 5). All other boards continue counting until they receive IACKIN*, then count four more clocks and assert IACKOUT* to the next board. Finally, the last board asserts IACKOUT* and the bus pauses until the data transfer time-out circuit ends the bus cycle by asserting BERR*. Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 44 2. VMEbus Interface > Automatic Slot Identification Figure 5: Timing for Auto-ID Cycle A B C D SYSCLK AS* DS0✴ IACK✴ IACKOUT✴ (CARD 1) ID = 5 IACKOUT✴ (CARD 2) ID = 9 IACKOUT✴ (CARD 3) ID COUNTER VALUE ID = 13 0 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 Because all boards are four clocks wide, the value in the clock counter is divided by four to identify the slot in which the board is installed; any remainder is discarded. Note that since the start of the IACK cycle is not synchronized to SYSCLK, a one count variation from the theoretical value of the board can occur. However, in all cases the ID value of a board is greater than that of a board in a lower slot number. The result is placed in the DY4AUTOID [7:0] field and the DY4DONE bit is set located in the “Miscellaneous Status Register (MISC_STAT)” on page 275. 2.6.3 System Controller Functions When located in Slot 1 of the VMEbus system (see “First Slot Detector” on page 41), the Universe II assumes the role of SYSCON and sets the SYSCON status bit in the MISC_CTL register. In accordance with the VME64 Specification, as SYSCON the Universe II has the following functions: 2.6.3.1 • A system clock driver • An arbitration module • An IACK Daisy Chain Driver (DCD) • A bus timer System Clock Driver The Universe II provides a 16 MHz SYSCLK signal derived from CLK64 when configured as SYSCON. 2.6.3.2 VMEbus Arbiter When the Universe II is SYSCON, the Arbitration Module is enabled. The Arbitration Module supports the following arbitration modes: • Fixed Priority Arbitration Mode (PRI) • Round Robin Arbitration Mode (RRS) (default setting) Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 2. VMEbus Interface > Automatic Slot Identification 45 These modes are selected with the VARB bit in the “Miscellaneous Status Register (MISC_STAT)” on page 275. 2.6.3.3 Fixed Priority Arbitration Mode (PRI) In this mode, the order of priority is VRBR_[3], VRBR_[2], VRBR_[1], and VRBR_[0] as defined by the VME64 Specification. The Arbitration Module issues a Bus Grant (VBGO [3:0]_) to the highest requesting level. If a Bus Request of higher priority than the current bus owner is asserted, the Arbitration Module asserts VBCLR_ until the owner releases the bus (VRBBSY_ is negated). 2.6.3.4 Round Robin Arbitration Mode (RRS) This mode arbitrates all levels in a round robin mode, by scanning from levels 3 to 0. Only one grant is issued per level and one owner is never forced from the bus in favor of another requester (VBCLR_ is never asserted). Since only one grant is issued per level on each round robin cycle, several scans are required to service a queue of requests at one level. 2.6.3.5 VMEbus Arbiter Time-out The Universe II’s VMEbus arbiter can be programmed to time-out if the requester does not assert BBSY* within a specified period. This allows BGOUT to be negated so that the arbiter can continue with other requesters. The timer is programmed using the VARBTO field in the MISC_CTL register, and can be set to 16 µs, 256 µs, or disabled. The default setting for the timer is 16 µs. The arbitration time-out timer has a granularity of 8 µs; setting the timer for 16 µs means the timer can timeout in as little as 8 µs. 2.6.4 IACK Daisy-Chain Driver Module The IACK Daisy-Chain Driver module is enabled when the Universe II becomes system controller. This module guarantees that IACKIN* stays high for at least 30 ns as specified in rule 40 of the VME64 specification. 2.6.5 VMEbus Time-out A programmable bus timer allows users to select a VMEbus time-out period. The time-out period is programmed through the VBTO field in the “Miscellaneous Control Register (MISC_CTL)” on page 273 and can be set to 16µs, 32µs, 64µs, 128 µs, 256 µs, 512 µs, 1024 µs, or disabled. The default setting for the timer is 64 µs. The VMEbus Timer module asserts VXBERR_ if a VMEbus transaction times out (indicated by one of the VMEbus data strobes remaining asserted beyond the time-out period). The VME Bus Time-out value has a 4us resolution. Therefore, the time-out may occur up to 4us earlier than the selected VME Bus Time-out period. Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 46 2.6.6 2. VMEbus Interface > Automatic Slot Identification Bus Isolation Mode (BI-Mode) ® BI-Mode is a mechanism for logically isolating the Universe II from the VMEbus. This mechanism is useful for the following purposes: • Implementing hot-standby systems — A system may have two identically configured boards, one in BI-Mode. If the board that is not in BI-Mode fails, it can be put in BI-Mode while the spare board is removed from BI-Mode. • System diagnostics for routine maintenance • Fault isolation in the event of a card failure — The faulty board can be isolated While in BI-Mode, the Universe II data channels cannot be used to communicate between VMEbus and PCI (Universe II mailboxes do provide a means of communication). The only traffic permitted is to Universe II registers either through configuration cycles, the PCI register image, the VMEbus register image, or CR/CSR space. No IACK cycles are generated or responded to. No DMA activity occurs. Any access to other PCI images result in a Target-Retry. Access to other VMEbus images are ignored. Entering BI-Mode has the following effects: • The VMEbus Master Interface becomes inactive — PCI Target Channel coupled accesses are retried. The PCI Target Channel Posted Writes FIFO continues to accept transactions but eventually fills and no further posted writes are accepted. The DMA FIFO eventually empties or fills and no further DMA activity takes place on the PCI bus. The Universe II VMEbus Master does not service interrupts while in BI-Mode. • The Universe II does not respond as a VMEbus slave — Except for accesses to the register image and CR/CSR image. • The Universe II does not respond to any interrupt it had outstanding. — All VMEbus outputs from the Universe II are tri-stated, so that the Universe II are not driving any VMEbus signals. The only exception to this is the IACK and BG daisy chains which must remain in operation as before. There are four ways to cause the Universe II to enter BI-Mode. The Universe II is put into BI-Mode for the following reasons: • If the BI-Mode power-up option is selected (See “Power-Up Options” on page 135 and Table 22 on page 135) • When SYSRST* or RST_ is asserted any time after the Universe II has been powered-up in BI-Mode • When VRIRQ_ [1] is asserted, provided that the ENGBI bit in the“Miscellaneous Control Register (MISC_CTL)” on page 273 has been set • When the BI bit in the MISC_CTL register is set Either of the following actions remove the Universe II from BI-Mode: • Power-up the Universe II with the BI-Mode option off (see “BI-Mode” on page 137), or Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 2. VMEbus Interface > Automatic Slot Identification • 47 Clear the BI bit in the MISC_CTL register. — This is effective only if the source of the BI-Mode is no longer active. If VRIRQ_ [1] is still being asserted while the ENGBI bit in the MISC_CTL register is set, then attempting to clear the BI bit in the MISC_CTL register does not work. Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 48 Universe II User Manual May 12, 2010 2. VMEbus Interface > Automatic Slot Identification Integrated Device Technology www.idt.com 49 3. PCI Interface Peripheral Component Interconnect (PCI) is a bus protocol that defines how devices communicate on a peripheral bus and with a host processor. If a device is referred to as PCI compliant it must be compliant with the PCI Local Bus Specification (Revision 2.1). The Universe II PCI bus supports frequencies up to 33 MHz, and 32-bit or 64-bit transfers. This chapter describes the Universe II’s PCI Interface. This chapter discusses the following topics: 3.1 • “PCI Cycles” on page 49 • “Universe II as PCI Master” on page 53 • “Universe II as PCI Target” on page 57 Overview The Universe II PCI Bus Interface is a directly connected to the PCI bus. For information concerning the different types of PCI accesses available, see “PCI Bus Target Images” on page 70. 3.2 PCI Cycles The PCI Bus Interface of the Universe II operates as a PCI compliant port with a 64-bit multiplexed address/data bus. The Universe II PCI Bus Interface is configured as little-endian using address invariant translation when mapping between the VMEbus and the PCI bus. Address invariant translation preserves the byte ordering of a data structure in a little-endian memory map and a big-endian memory map (see “Endian Mapping” on page 359 and the PCI 2.1 Specification). The Universe II has all the PCI signals described in the PCI 2.1 Specification) with the exception of SBO_ and SDONE. These pins are exception because the Universe II does not provide cache support. Universe II PCI cycles are synchronous, meaning that bus and control input signals are externally synchronized to the PCI clock (CLK). PCI cycles are divided into the following phases: 1. Request 2. Address 3. Data transfer 4. Cycle termination 3.2.1 32-Bit Versus 64-Bit PCI The Universe II is configured with a 32-bit or 64-bit PCI data bus at power-up (see “PCI Bus Width” on page 138). Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 50 3. PCI Interface > PCI Cycles Each of the Universe II’s VMEbus slave images can be programmed so that VMEbus transactions are mapped to a 64-bit data bus on the PCI Interface through the LD64EN bit, in the “DMA Transfer Control Register (DCTL)” on page 232. If the VMEbus slave image is programmed with a 64-bit PCI bus data width and Universe II is powered-up in a 64-bit PCI environment, the Universe II asserts REQ64_ during the address phase of the PCI transaction. REQ64_ is asserted if LD64EN is set in a 64-bit PCI system independent of whether the Universe II has a full 64-bit transfer. This can result in a performance degradation because of the extra clocks required to assert REQ64_ and to sample ACK64_. Also, there can be some performance degradation when accessing 32-bit targets with LD64EN set. Do not set the LD64EN bit unless there are 64-bit targets in the slave image window. If the VMEbus slave images are not programmed for a 64-bit wide PCI data bus, then the Universe operates transparently in a 32-bit PCI environment. Independent of the setting of the LD64EN bit, the Universe II never attempts a 64-bit cycle on the PCI bus if it is powered-up as a 32-bit device. 3.2.2 PCI Bus Request and Parking The Universe II supports bus parking. If the Universe II requires the PCI bus it asserts REQ_ only if its GNT_ is not currently asserted. When the PCI Master Module is ready to begin a transaction and its GNT_ is asserted, the transfer begins immediately. This eliminates a possible one clock cycle delay before beginning a transaction on the PCI bus which would exist if the Universe II did not implement bus parking. Refer to the PCI 2.1 Specification for more information 3.2.3 Address Phase PCI transactions are initiated by asserting FRAME_ and driving address and command information onto the bus. In the VMEbus Slave Channel, the Universe II calculates the address for the PCI transaction by adding a translation offset to the VMEbus address (see “Universe II as VMEbus Slave” on page 32). Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 3. PCI Interface > PCI Cycles 51 The command signals (on the C/BE_ lines) contain information about Memory space, cycle type and whether the transaction is read or write. Table 3 shows the PCI command type encoding implemented with the Universe II. Table 3: Command Type Encoding for Transfer Type 3.2.3.1 C/BE_ [3:0] for PCI, C/BE_ [7:4] for non-multiplexed Command Type Universe II Capability 0000 Interrupt Acknowledge N/A 0001 Special Cycle N/A 0010 I/O Read Target/Master 0011 I/O Write Target/Master 0100 Reserved N/A 0101 Reserved N/A 0110 Memory Read Target/Master 0111 Memory Write Target/Master 1000 Reserved N/A 1001 Reserved N/A 1010 Configuration Read Target/Master 1011 Configuration Write Target/Master 1100 Memory Read Multiple See “Memory Read Types” on page 51 1101 Dual Address Cycle N/A 1110 Memory Read Line See “Memory Read Types” on page 51 1111 Memory Write and Invalidate See “Memory Read Types” on page 51 Memory Read Types Memory Read Multiple and Memory Read Line transactions are aliased to Memory Read transactions when the Universe II is accessed as a PCI target with these commands. Likewise, Memory Write and Invalidate is aliased to Memory Write. As a PCI master, the Universe II can generate Memory Read Multiple but not Memory Read Line. Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 52 3. PCI Interface > PCI Cycles PCI targets must assert DEVSEL_ if they have decoded the access. During a Configuration cycle, the target is selected by its particular ID Select (IDSEL). If a target does not respond with DEVSEL_ within six clocks, a Master-abort is generated. The role of configuration cycles is described in the PCI 2.1 Specification. 3.2.4 Data Transfer Acknowledgment of a data phase occurs on the first rising clock edge after both IRDY_ and TRDY_ are asserted by the master and target, respectively. REQ64_ can be driven during the address phase to indicate that the master wishes to initiate a 64-bit transaction. The PCI target asserts ACK64_ if it is able to respond to the 64-bit transaction. Wait cycles are introduced by either the master or the target by de-asserting IRDY_ or TRDY_. For write cycles, data is valid on the first rising edge after IRDY_ is asserted. Data is acknowledged by the target on the first rising edge with TRDY_ asserted. For read cycles, data is transferred and acknowledged on first rising edge with both IRDY_ and TRDY_ asserted. A single data transfer cycle is repeated every time IRDY_ and TRDY_ are both asserted. The transaction only enters the termination phase when FRAME_ is de-asserted (master-initiated termination) or if STOP_ is asserted (target-initiated). When both FRAME_ and IRDY_ are de-asserted (final data phase is complete), the bus is defined as idle. 3.2.5 Termination Phase The PCI Bus Interface permits the following types of PCI terminations: 1. Master-abort: The PCI bus master negates FRAME_ when no target responds (DEVSEL_ not asserted) after six clock cycles. 2. Target-disconnect: A termination is requested by the target (STOP_ is asserted) because it is unable to respond within the latency requirements of the PCI specification or it requires a new address phase. • Target-disconnect with data: The transaction is terminated after data is transferred. The Universe II de-asserts REQ_ for at least two clock cycles if it receives STOP_ from the PCI target. • Target-disconnect without data: The transaction is terminated before data is transferred. The Universe II de-asserts REQ_ for at least two clock cycles if it receives STOP_ from the PCI target. 3. Target-retry: Termination is requested (STOP_ is asserted) by the target because it cannot currently process the transaction. Retry means that the transaction is terminated after the address phase without any data transfer. 4. Target-abort: Is a modified version of target-disconnect where the target requests a termination (asserts STOP_) of a transaction which it will never be able to respond to, or during which a fatal error occurred. Although there may be a fatal error for the initiating application, the transaction completes gracefully, ensuring normal PCI operation for other PCI resources. Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 3. PCI Interface > Universe II as PCI Master 3.2.6 53 Parity Checking The Universe II both monitors and generates parity information using the PAR signal. The Universe II monitors PAR when it accepts data as a master during a read or a target during a write. The Universe II drives PAR when it provides data as a target during a read or a master during a write. The Universe II also drives PAR during the address phase of a transaction when it is a master and monitors PAR during an address phase when it is the PCI target. In both address and data phases, the PAR signal provides even parity for C/BE_[3:0] and AD[31:0]. The Universe II continues with a transaction independent of any parity error reported during the transaction. The Universe II can also be programmed to report address parity errors. It does this by asserting the SERR_ signal and setting a status bit in its registers. No interrupt is generated, and regardless of whether assertion of SERR_ is enabled, the Universe II does not respond to the errored access. When the Universe II is powered-up in a 64-bit PCI environment, it uses PAR64 in the same way as PAR, except for AD[63:32] and C/BE[7:4]. Universe II reports parity errors during all transactions with the PERR_ signal. The Universe II drives PERR_ high within two clocks of receiving a parity error on incoming data, and holds PERR_ for at least one clock for each errored data phase. 3.3 Universe II as PCI Master The Universe II requests PCI bus mastership through its PCI Master Interface. The PCI Master Interface is available to either the VMEbus Slave Channel (access from a remote VMEbus master) or the DMA Channel. The VMEbus Slave Channel makes an internal request for the PCI Master Interface when the following conditions are met: • RXFIFO contains a complete transaction, • Sufficient data exists in the RXFIFO to generate a transaction of length defined by the programmable aligned burst size (PABS) • There is a coupled cycle request The DMA Channel makes an internal request for the PCI Master Interface when the following conditions are met: • DMAFIFO has room for 128 bytes to be read from PCI • DMAFIFO has queued 128 bytes to be written to PCI • DMA block is completely queued during a write to the PCI bus Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 54 3. PCI Interface > Universe II as PCI Master Arbitration between the two channels for the PCI Master Interface follows a round robin protocol. Each channel is given access to the PCI bus for a single transaction. Once that transaction completes, ownership of the PCI Master Interface is granted to the other channel if it requires the bus. The VMEbus Slave Channel and the DMA Channel each have a set of rules that determine when the transaction is complete and the channels no longer need the PCI Master Interface. The VMEbus Slave Channel no longer needs the PCI Master Interface under the following conditions: • An entire transaction (no greater in length than the programmed aligned burst size) is emptied from the RXFIFO • The coupled cycle is complete The DMA Channel no longer needs the PCI Master Interface when the following conditions are met: • The boundary programmed into the PCI aligned burst size is emptied from the DMAFIFO during writes to the PCI bus • The boundary programmed into the PCI aligned burst size is queued to the DMAFIFO during reads from the PCI bus Access from the VMEbus can be either coupled or decoupled. For a full description of the operation of these data paths, see “Universe II as VMEbus Slave” on page 32. 3.3.1 Command Types The PCI Master Interface can generate the following command types: • I/O Read • I/O Write • Memory Read • Memory Read Multiple • Memory Write • Configuration Read (Type 0 and 1) • Configuration Write (Type 0 and 1) The type of cycle the Universe II generates on the PCI bus depends on which VMEbus slave image is accessed and how it is programmed. For example, one slave image might be programmed as an I/O space, another as Memory space and another for Configuration space (see “VME Slave Image Programming” on page 67). When generating a memory transaction, the addressing is either 32-bit or 64-bit aligned, depending upon the PCI target. When generating an I/O transaction, the addressing is 32-bit aligned and all incoming transactions are coupled. Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 3. PCI Interface > Universe II as PCI Master 3.3.2 55 PCI Burst Transfers The Universe II generates aligned burst transfers of some maximum alignment, according to the programmed PCI aligned burst size (PABS field in the “Master Control Register (MAST_CTL)” on page 271). The PCI aligned burst size can be programmed at 32, 64 or 128 bytes. Burst transfers do not cross the programmed boundaries. For example, when programmed for 32-byte boundaries, a new burst begins at XXXX_XX20, XXXX_XX40, etc. If necessary, a new burst begins at an address with the programmed alignment. To optimize PCI bus usage, the Universe II always attempts to transfer data in aligned bursts at the full width of the PCI bus. The Universe II can perform a 64-bit data transfer over the AD [63:0] lines, if operated in a 64-bit PCI environment or against a 64-bit capable target or master. The LD64EN bit must be set if the access is being made through a VMEbus slave image; the LD64EN bit must be set if the access is being performed with the DMA. The Universe II generates burst cycles on the PCI bus if it is performing the following tasks: • Emptying the RXFIFO — The TXFE bit in the “Miscellaneous Status Register (MISC_STAT)” on page 275 is clear • Filling the RDFIFO receives a block read request from a VMEbus master to an appropriately programmed VMEbus slave image • Performing DMA transfers All other accesses are treated as single data beat transactions on the PCI bus. During PCI burst transactions, the Universe II dynamically enables byte lanes on the PCI bus by changing the BE_ signals during each data phase. 3.3.3 Termination The Universe II performs a Master-Abort if the target does not respond within six clock cycles. Coupled PCI transactions terminated with Target-Abort or Master-Abort are terminated on the VMEbus with BERR*. The R_TA or R_MA bits in the PCI_CS register are set when the Universe II receives a Target-Abort or generates a Master-Abort independent of whether the transaction was coupled, decoupled, prefetched, or initiated by the DMA. If the Universe II receives a retry from the PCI target, then it relinquishes the PCI bus and re-requests within three PCI clock cycles. No other transactions are processed by the PCI Master Interface until the retry condition is cleared. The Universe II can be programmed to perform a maximum number of retries using the MAXRTRY field in the “Master Control Register (MAST_CTL)” on page 271. When this number of retries has been reached, the Universe II responds in the same way as it does to a Target-Abort on the PCI bus. The Universe II can issue a BERR* signal on the VMEbus. All VMEbus slave coupled transactions and decoupled transactions encounter a delayed DTACK once the FIFO fills until the condition clears either due to success or a retry time-out. Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 56 3. PCI Interface > Universe II as PCI Master If the error occurs during a posted write to the PCI bus (see also “Error Handling” on page 125), the Universe II uses the “PCI Command Error Log Register (L_CMDERR)” on page 210 to log the command information for the transaction (CMDERR [3:0]) and the address of the errored transaction is latched in the “PCI Address Error Log (LAERR)” on page 211. The L_CMDERR register also records if multiple errors occur (with the M_ERR bit) although the number of errors is not given. The error log is qualified with the L_STAT bit. The rest of the transaction is purged from the RXFIFO if some portion of the write encounters an error. An interrupt is generated on the VMEbus and/or PCI bus depending upon whether the VERR and LERR interrupts are enabled (see “Interrupt Generation and Handling” on page 109). If an error occurs on the PCI bus, the Universe II does not translate the error condition into a BERR* on the VMEbus; the Universe II does not directly map the error. By taking no action, the Universe II forces the external VMEbus error timer to expire. 3.3.4 Parity The Universe II monitors PAR when it accepts data as a master during a read and drives PAR when it provides data as a master during a write. The Universe II also drives PAR during the address phase of a transaction when it is a master. In both address and data phases, the PAR signal provides even parity for C/BE_[3:0] and AD[31:0]. When the Universe II is powered-up in a 64-bit PCI environment, it uses PAR64 in the same way as PAR, except for AD[63:32] and C/BE[7:4]. The PERESP bit in the “PCI Configuration Space Control and Status Register (PCI_CSR)” on page 172 determines whether or not the Universe II responds to parity errors as PCI master. Data parity errors are reported through the assertion of PERR_ if the PERESP bit is set. Regardless of the setting of these two bits, the D_PE (Detected Parity Error) bit in the PCI_CS register is set if the Universe II encounters a parity error as a master. The DP_D (Data Parity Detected) bit in the same register is only set if parity checking is enabled through the PERESP bit and the Universe II detects a parity error while it is PCI master (i.e. it asserts PERR_ during a read transaction or receives PERR_ during a write). No interrupts are generated by the Universe II in response to parity errors reported during a transaction. Parity errors are reported by the Universe II through assertion of PERR_ and by setting the appropriate bits in the PCI_CS register. If PERR_ is asserted to the Universe II while it is PCI master, the only action it takes is to set the DP_D. The Universe II continues with a transaction independent of any parity errors reported during the transaction. As a master, the Universe II does not monitor SERR_. It is expected that a central resource on the PCI bus monitors SERR_ and take appropriate action. Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 3. PCI Interface > Universe II as PCI Target 3.4 57 Universe II as PCI Target The Universe II becomes PCI bus target when one of its nine programmed PCI target images, or one of its registers, is accessed by a PCI bus master. The Universe II cannot access its own images or registers and master the PCI bus. Refer to “Registers” on page 163 for more information on register accesses. When one of its PCI target images is accessed, the Universe II responds with DEVSEL_ within two clocks of FRAME_. This makes the Universe II a medium speed device, as reflected by the DEVSEL field in the PCI_CS register). 3.4.1 Command Types As PCI target, the Universe II responds to the following command types: • I/O Read • I/O Write • Memory Read • Memory Write • Configuration Read (Type 0) • Configuration Write (Type 0) • Memory Read Multiple (aliased to Memory Read) • Memory Line Read (aliased to Memory Read) • Memory Write and Invalidate (aliased to Memory Write) Type 0 Configuration accesses can only be made to the Universe II’s PCI configuration registers. The PCI target images do not accept Type 0 accesses. Address parity errors are reported if both PERESP and SERR_EN are set in the “PCI Configuration Space Control and Status Register (PCI_CSR)” on page 172. Address parity errors are reported by the Universe II by asserting the SERR_ signal and setting the S_SERR (Signalled SERR_) bit in the PCI_CS register. Assertion of SERR_ can be disabled by clearing the SERR_EN bit in the PCI_CS register. No interrupt is generated, and regardless of whether assertion of SERR_ is enabled or not, the Universe II does not respond to the access with DEVSEL_. Typically, the master of the transaction times out with a Master-Abort. If the Universe II is accessed with REQ64_ in Memory space as a 64-bit target, then it responds with ACK64_ if it is powered up as a 64-bit device. 3.4.2 Data Transfer Read transactions are always coupled, as opposed to VMEbus slave reads which can be pre-fetched (see “Universe II as VMEbus Slave” on page 32). Write transactions can be coupled or posted (see Figure 6 and “PCI Bus Target Images” on page 70). To ensure sequential consistency, coupled operations (reads or writes) are only processed once all previously posted write operations have completed (the TXFIFO is empty). Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 58 3. PCI Interface > Universe II as PCI Target Figure 6: PCI Bus Target Channel Dataflow POSTED WRITE DATA TXFIFO PCI BUS SLAVE INTERFACE COUPLED WRITE DATA VMEbus MASTER INTERFACE COUPLED READ DATA The PCI bus and the VMEbus can have different data width capabilities. The maximum VMEbus data width is programmed into the PCI target image through the VDW bit in the “PCI Target Image 0 Control (LSI0_CTL)” on page 181. For example, consider a 32-bit PCI transaction accessing a PCI target image with VDW set to 16 bits. A data beat with all byte lanes enabled will be broken into two 16-bit cycles on the VMEbus. If the PCI target image is also programmed with block transfers enabled, the 32-bit PCI data beat will result in a D16 block transfer on the VMEbus. Write data is unpacked to the VMEbus and read data is packed to the PCI bus data width. If the data width of the PCI data beat is the same as the maximum data width of the PCI target image, then the Universe II maps the data beat to an equivalent VMEbus cycle. For example, consider a 32-bit PCI transaction accessing a PCI target image with VDW set to 32 bits. A data beat with all byte lanes enabled is translated to a single 32-bit cycle on the VMEbus. If the PCI bus data width is less than the VMEbus data width then there is no packing or unpacking between the two buses. The only exception to this is during 32-bit PCI multi-data beat transactions to a PCI target image programmed with maximum VMEbus data width of 64 bits. In this case, packing/unpacking occurs to make maximum use of the full bandwidth on both buses. Only aligned VMEbus transactions are generated, so if the requested PCI data beat has unaligned or non-contiguous byte enables, then it is broken into multiple aligned VMEbus transactions no wider than the programmed VMEbus data width. For example, consider a three-byte PCI data beat (on a 32-bit PCI bus) accessing a PCI target image with VDW set to 16-bit. The three-byte PCI data beat will be broken into three aligned VMEbus cycles: three single-byte cycles. If in the above example the PCI target image has a VDW set to 8-bit, then the three-byte PCI data beat will be broken into three single-byte VMEbus cycles. 3.4.3 Coupled Transfers The PCI Target Channel supports coupled transfers. A coupled transfer through the PCI Target Channel is a transfer between PCI and VME where the Universe II maintains ownership of the VMEbus from the beginning to the end of the transfer on the PCI bus, and where the termination of the cycle on the VMEbus is relayed directly to the PCI master in the normal manner (Target-Abort or Target Completion), rather than through error-logging and interrupts. Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 3. PCI Interface > Universe II as PCI Target 59 By default, all PCI target images are set for coupled transfers. Coupled transfers typically cause the Universe II to go through three phases: the Coupled Data-Transfer Phase, and then the Coupled Wait Phase. When an external PCI Master attempts a data transfer through a slave image programmed for coupled cycles, and the Universe II currently owns the VMEbus, the PCI Target Channel moves directly to the Coupled Data-Transfer Phase 3.4.3.1 Coupled Data-Transfer Phase At the beginning of the Coupled Data-Transfer Phase, the Universe II latches the PCI command, byte enable, address and (in the case of a write) data. Regardless of the state of FRAME_, the Universe II retries1 the master, and then performs the transaction on the VMEbus. The Universe II continues to signal Target-Retry to the external PCI master until the transfer completes on the VMEbus. If the transfer completes normally on the VMEbus then, in the case of a read, the data is transmitted to the PCI bus master. If a data phase of a coupled transfer requires packing or unpacking on the VMEbus, acknowledgment of the transfer is not given to the PCI bus master until all data has been packed or unpacked on the VMEbus. Successful termination is signalled on the PCI bus—the data beat is acknowledged with a Target-Disconnect, forcing all multi-beat transfers into single beat. At this point, the Universe II enters the Coupled Wait Phase. If a bus error is signalled on the VMEbus or an error occurs during packing or unpacking, then the transaction is terminated on the PCI bus with Target-Abort. For more information refer to “Data Transfer” on page 57. 3.4.3.2 Coupled Wait Phase The Coupled Wait Phase is entered after the successful completion of a Coupled Data-Transfer phase. The Coupled Wait Phase allows consecutive coupled transactions to occur without releasing the VMEbus. If a new coupled transaction is attempted while the Universe II is in the Coupled Wait Phase, the Universe II moves directly to the Coupled Data-Transfer Phase. The Coupled Window Timer determines the maximum duration of the Coupled Wait Phase. When the Universe II enters the Coupled Wait Phase, the Coupled Window Timer starts. The period of this timer is specified in PCI clocks and is programmable through the CWT field of the “PCI Miscellaneous Register (LMISC)” on page 206. If this field is programmed to 0000, the Universe II does an early release of BBSY* during the coupled transfer on the VMEbus and will not enter the Coupled Wait Phase. In this case, VMEbus ownership is relinquished immediately by the PCI Target Channel after each coupled cycle. Once the timer associated with the Coupled Wait Phase expires, the Universe II releases the VMEbus if release mode is set for RWD, or the release mode is set for ROR and there is a pending (external) request on the VMEbus. 1. PCI latency requirements (as described in revision 2.1 of the PCI Specification) require that only 16 clock cycles can elapse between the first and second data beat of a transaction. Since the Universe II cannot guarantee that data acknowledgment will be received from the VMEbus in time to meet these PCI latency requirements, the Universe II performs a target-disconnect after the first data beat of every coupled write transaction. Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 60 3.4.4 3. PCI Interface > Universe II as PCI Target Posted Writes Posted writes are enabled for a PCI target image by setting the PWEN bit in the control register of the PCI target image (see “PCI Bus Target Images” on page 70) to 1 and setting the LAS bit to 0. Write transactions are relayed from the PCI bus to the VMEbus through a 64-entry deep TXFIFO. The TXFIFO allows each entry to contain 32 address bits (with extra bits provided for command information), or to a full 64-bit width. For each posted write transaction received from the PCI bus, the PCI Target Interface queues an address entry in the FIFO. This entry contains the translated address space and mapped VMEbus attributes information relevant to the particular PCI target image that has been accessed (see “PCI Bus Target Images” on page 70). For this reason, any reprogramming of PCI bus target image attributes will only be reflected in TXFIFO entries queued after the reprogramming. Transactions queued before the re-programming are delivered to the VMEbus with the PCI bus target image attributes that were in use before the reprogramming. Care must be taken before reprogramming target images. To ensure the FIFO is empty use one of the following options: 3.4.4.1 • Perform a coupled read. The coupled read does not complete until all posted-write data has been queued • Read the MISC_STAT register until the TXFE bit has a value of 0 FIFO Entries Once the address phase is queued in one TXFIFO entry, the PCI Target Interface may pack the subsequent data beats to a full 64-bit width before queuing the data into new entries in the TXFIFO. For 32-bit PCI transfers in the Universe II, the TXFIFO accepts a single burst of one address phase and 59 data phases when it is empty. For 64-bit PCI, the TXFIFO accepts a single burst of one address phase and 31 data phases when it is empty. To improve PCI bus utilization, the TXFIFO does not accept a new address phase if it does not have room for a burst of one address phase and 128 bytes of data. If the TXFIFO does not have enough space for an aligned burst, then the posted write transaction is terminated with a Target-Retry immediately after the address phase. When an external PCI Master posts writes to the PCI Target Channel of the Universe II, the Universe II issues a disconnect if the address crosses a 256-byte boundary. Before a transaction can be delivered to the VMEbus from the TXFIFO, the PCI Target Channel must obtain ownership of the VMEbus Master Interface. Ownership of the VMEbus Master Interface is granted to the different channels on a round robin basis (see “VMEbus Release” on page 27). Once the PCI Target Channel obtains the VMEbus through the VMEbus Master Interface, the manner in which the TXFIFO entries are delivered depends on the programming of the VMEbus attributes in the PCI target image (see “PCI Bus Target Images” on page 70). For example, if the VMEbus data width is programmed to 16-bit, and block transfers are disabled, then each data entry in the TXFIFO corresponds to four transactions on the VMEbus. If block transfers are enabled in the PCI target image, then each transaction queued in the TXFIFO, independent of its length, is delivered to the VMEbus as a block transfer. This means that if a single data beat transaction is queued in the TXFIFO, it appears on the VMEbus as a single data phase block transfer. Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 3. PCI Interface > Universe II as PCI Target 61 Any PCI master attempting coupled transactions is retried while the TXFIFO contains data. If posted writes are continually written to the PCI Target Channel by another master, and the FIFO does not empty, coupled transactions requested by the first PCI master in the PCI Target Channel does not proceed and are continually retried. This presents a potential starvation scenario. This functionality is intended to support earlier versions of PCI-to-PCI bridges. 3.4.5 Special Cycle Generator The Special Cycle Generator in the PCI Target Channel of the Universe II can be used in conjunction with one of the PCI Target Images to generate Read-Modify-Write (RMW) and Address Only With Handshake (ADOH) cycles. The address programmed into the “Special Cycle PCI Bus Address Register (SCYC_ADDR)” on page 202, in the address space specified by the LAS field of the SCYC_CTL register (Memory or I/O), must appear on the PCI bus during the address phase of a transfer for the Special Cycle Generator to perform its function. Whenever this address on the PCI bus (bits [31:2]) is used to matches the address in the SCYC_ADDR register, the Universe II does not respond with ACK64_ (since the Special Cycle Generator only processes up to 32-bit cycles). The cycle that is produced on the VMEbus uses attributes programmed into the Image Control Register of the image that contains the address programmed in the SCYC_ADDR register. Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 62 3. PCI Interface > Universe II as PCI Target The Special Cycle Generator is configured through the register fields shown in Table 7. Figure 7: Register Fields for the Special Cycle Generator Field Register Bits Description 32-bit address ADDR in “Special Cycle PCI Bus Address Register (SCYC_ADDR)” on page 202 Specifies PCI bus target image address PCI Address Space LAS in “Special Cycle Control Register (SCYC_CTL)” on page 201 Specifies whether the address specified in the ADDR field lies in PCI memory or I/O space Special cycle SCYC[1:0] in “Special Cycle Control Register (SCYC_CTL)” on page 201 Disabled, RMW or ADOH 32-bit enable EN [31:0] in “Special Cycle Swap/Compare Enable Register (SCYC_EN)” on page 203 A bit mask to select the bits to be modified in the VMEbus read data during a RMW cycle 32-bit compare CMP [31:0] in “Special Cycle Compare Data Register (SCYC_CMP)” on page 204 Data which is compared to the VMEbus read data during a RMW cycle 32-bit swap SWP [31:0] in “Special Cycle Swap Data Register (SCYC_SWP)” on page 205 Data which is swapped with the VMEbus read data and written to the original address during a RMW cycle The following sections describe the specific properties for each of the transfer types: RMW and ADOH. 3.4.5.1 Read-Modify-Write When the SCYC field is set to RMW, any PCI bus read access to the specified PCI bus address (SCYC_ADDR register) results in a RMW cycle on the VMEbus (provided the constraints listed below are satisfied). RMW cycles on the VMEbus consist of a single read followed by a single write operation. The data from the read portion of the RMW on the VMEbus is returned as the read data on the PCI bus. RMW cycles make use of three 32-bit registers (see Table 7). The bit enable field is a bit mask which lets the user specify which bits in the read data are compared and modified in the RMW cycle. This bit enable setting is completely independent of the RMW cycle data width, which is determined by the data width of the initiating PCI transaction. During a RMW, the VMEbus read data is bitwise compared with the SCYC_CMP and SCYC_EN registers. The valid compared and enabled bits are then swapped using the SCYC_SWP register. Each enabled bit that compares equal is swapped with the corresponding bit in the 32-bit swap field. A false comparison results in the original bit being written back. Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 3. PCI Interface > Universe II as PCI Target 63 Once the RMW cycle completes, the VMEbus read data is returned to the waiting PCI bus master and the PCI cycle terminates. RMW Constraints Certain restrictions apply to the use of RMW cycles. If a write transaction is initiated to the VMEbus address when the special cycle field (SCYC in “Special Cycle Control Register (SCYC_CTL)” on page 201) is set for RMW, then a standard write occurs with the attributes programmed in the PCI target image (in other words, the special cycle generator is not used). The Universe II performs no packing and unpacking of data on the VMEbus during a RMW operation. The following constraints must also be met. 1. The Special Cycle Generator only generates a RMW if it is accessed with an 8-bit, aligned 16-bit, or aligned 32-bit read cycle. 2. The Special Cycle Generator only generates a RMW if the size of the request is less than or equal to the programmed VMEbus Maximum Data width. 3. The destination VMEbus address space must be one of A16, A24 or A32. In the event that the Special Cycle Generator is accessed with a read cycle that does not meet the RMW criteria, the Universe II generates a Target-Abort. The Universe II must be correctly programmed and accessed with correct byte-lane information. 3.4.5.2 VME Lock Cycles—Exclusive Access to VMEbus Resources The VME Lock cycle is used, in combination with the VOWN bit in the “Master Control Register (MAST_CTL)” on page 271, to lock resources on the VMEbus. The VME Lock cycle can be used by the Universe II to inform the resource that a locked cycle is intended. The VOWN bit in the MAST_CTL register can be set to ensure that when the Universe II acquires the VMEbus, it is the only master given access to the bus (until the VOWN bit is cleared). It may also be necessary for the PCI master to have locked the Universe II using the PCI LOCK_ signal. When the SCYC field is set to VME Lock, any write access to the specified VMEbus address will result in a VME Lock cycle on the VMEbus. A VME Lock cycle is coupled: the cycle does not complete on the PCI bus until it completes on the VMEbus. Reads to the specified address translate to VMEbus reads in the standard fashion. The data during writes is ignored. The AM code generated on the VMEbus is determined by the PCI target image definition for the specified VMEbus address (see Table 8 on page 70). However, after the VME Lock cycle is complete, there is no guarantee that the Universe II remains VMEbus master unless it has set the VOWN bit. If the Universe II loses VMEbus ownership, the VMEbus resouce does not remain locked. The following procedure is required to lock the VMEbus through an ADOH cycle: 1. If there is more than one master on the PCI bus, it may be necessary to use PCI LOCK_ to ensure that the PCI master driving the ADOH cycle has sole PCI access to the Universe II registers and the VMEbus 2. program the VOWN bit in the MAST_CTL register to a value of 1 (see “Using the VOWN bit” on page 64) Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 64 3. PCI Interface > Universe II as PCI Target 3. wait until the VOWN_ACK bit in the MAST_CTL register is a value of 1 4. generate an ADOH cycle with the Special Cycle Generator 5. perform transactions to be locked on the VMEbus 6. release the VMEbus by programming the VOWN bit in the MAST_CTL register to a value of 0 7. wait until the VOWN_ACK bit in the MAST_CTL register is a value of 0 In the event that BERR* is asserted on the VMEbus once the Universe II has locked and owns the VMEbus, it is the responsibility of the user to release ownership of the VMEbus by programming the VOWN bit in the MAST_CTL register to a value of 0. The following restrictions apply to the use of VME Lock cycles: 3.4.6 • All byte lane information is ignored for VME Lock cycles • The Universe II generates a VME Lock cycle on the VMEbus if the PCI Target Image, which includes the special cycle, has posted writes disabled • The Universe II Special Cycle Generator does not generate VME Lock cycles if the address space is not one of A16, A24 or A32 — it produces regular cycles instead Using the VOWN bit The Universe II provides a VMEbus ownership bit (VOWN bit in the “Master Control Register (MAST_CTL)” on page 271) to ensure that the Universe II has access to the locked VMEbus resource for an indeterminate period. The Universe II can be programmed to assert an interrupt on the PCI bus when it acquires the VMEbus and the VOWN bit is set in the “PCI Interrupt Enable Register (LINT_EN)” on page 242. While the VMEbus is held using the VOWN bit, the Universe II sets the VOWN_ACK bit in the MAST_CTL register. The act of changing the VOWN_ACK bit from 0 to 1 generates and interrupt. The VMEbus Master Interface maintains bus tenure while the ownership bit is set, and only releases the VMEbus when the ownership bit is cleared. 3.4.6.1 Reasons for Using the VOWN Bit If the VMEbus Master Interface is programmed for RWD, through the VREL bit in “Master Control Register (MAST_CTL)” on page 271, it may release the VMEbus when the PCI Target Channel has completed a transaction. If exclusive access to the VMEbus resource is required for multiple transactions, then the VMEbus ownership bit holds the bus until the exclusive access is no longer required. Alternatively, if the VMEbus Master Interface is programmed for ROR, the VMEbus ownership bit ensures VMEbus tenure even if other VMEbus requesters require the VMEbus. Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 3. PCI Interface > Universe II as PCI Target 3.4.7 65 Terminations The Universe II performs the following terminations as PCI target: 1. Target-Disconnect • When registers are accessed with FRAME_ asserted (no bursts allowed to registers) • After the first data beat of a coupled cycle with FRAME_ asserted A Target-disconnect with data only occurs if FRAME_ is asserted. • After the first data phase of a PCI Memory command (with FRAME_ asserted) if AD[1:0] is not equal to 00 (refer to Revision 2.1 of the PCI Specification) 2. Target-Retry • When a new posted write is attempted and the TXFIFO does not have room for a burst of one address phase and sixteen 64-bit data phases (for 64-bit PCI) • When a coupled transaction is attempted and the Universe II does not own the VMEbus • When a coupled transaction is attempted while the TXFIFO has entries to process • Register Channel is locked by the VME Slave Channel if a register access (including a RMW access) is in progress or the registers have been locked by an ADOH access. If the registers are locked by the VME Slave Channel, register accesses by external PCI masters are retried 3. Target-Abort • When the Universe II receives BERR* on the VMEbus during a coupled cycle (BERR* translated as Target-Abort on the PCI side and the S_TA bit is set in the “PCI Configuration Space Control and Status Register (PCI_CSR)” on page 172 Whether to terminate a transaction or for retry purposes, the Universe II keeps STOP_ asserted until FRAME_ is de-asserted, independent of the logic levels of IRDY_ and TRDY_. If STOP_ is asserted while TRDY_ is de-asserted, it means that the Universe II does not transfer any more data to the master. 3.4.7.1 Error During Posted Write If an error occurs during a posted write to the VMEbus, the Universe II uses the AMERR field in the “VMEbus AM Code Error Log (V_AMERR)” on page 307 to log the AM code of the transaction, and the state of the IACK* signal, through the IACK bit, to indicate whether the error occurred during an IACK cycle. The FIFO entries for the cycle are purged. The V_AMERR register also records whether multiple errors have occurred (with the M_ERR bit) although the number is not given. The error log is qualified with the V_STAT bit because logs are valid if the V_STAT bit is set. The address of the errored transaction is latched in the “VMEbus Address Error Log (VAERR)” on page 308. When the Universe II receives a VMEbus error during a posted write, it generates an interrupt on the VMEbus and/or PCI bus depending upon whether the VERR and VERR interrupts are enabled (see “Interrupt Generation and Handling” on page 109). Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 66 Universe II User Manual May 12, 2010 3. PCI Interface > Universe II as PCI Target Integrated Device Technology www.idt.com 67 4. Slave Image Programming This chapter describes the Slave Image Programming functionality of the Universe II. This chapter discusses the following topics: 4.1 • “VME Slave Image Programming” on page 67 • “PCI Bus Target Images” on page 70 • “Special PCI Target Image” on page 73 Overview The Universe II recognizes two types of accesses on its bus interfaces: accesses destined for the other bus, and accesses decoded for its own register space. Address decoding for the Universe II’s register space is described in “Registers” on page 163. This section describes the slave images used to map transactions between the PCI bus and VMEbus. 4.2 VME Slave Image Programming The Universe II accepts accesses from the VMEbus within specific programmed slave images. Each VMEbus slave image opens a window to the resources of the PCI bus and, through its specific attributes, allows the user to control the type of access to those resources. The tables below describe programming for the VMEbus slave images by dividing them into VMEbus, PCI bus and Control fields. Table 4: VMEbus Fields for VMEbus Slave Image Field Register Bits Description Base BS[31:12] or BS[31:16] in VSIx_BS Multiples of 4 or 64 Kbytes (base to bound: maximum of 4 GBytes) Bound BD[31:12] or BD[31:16] in VSIx_BD Address space VAS in VSIx_CTL A16, A24, A32, User 1, User 2 Mode SUPER in VSIx_CTL Supervisor and/or non-privileged Type PGM in VSIx_CTL Program and/or data Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 68 4. Slave Image Programming > VME Slave Image Programming Table 5: PCI Bus Fields for VMEbus Slave Image Field Register Bits Description Translation offset TO[31:12] or TO[31:16] in VSIx_TO Offsets VMEbus slave address to a selected PCI address Address space LAS in VSIx_CTL Memory, I/O, Configuration RMW LLRMW in VSIx_CTL RMW enable bit Table 6: Control Fields for VMEbus Slave Image Field Register Bits Description Image enable EN in VSIx_CTL Enable bit Posted write PWEN in VSIx_CTL Posted write enable bit Prefetched read PREN in VSIx_CTL Prefetched read enable bit Enable PCI D64 LD64EN in VSIx_CTL Enables 64-bit PCI bus transactions The Bus Master Enable (BM) bit of the PCI_CS register must be set in order for the image to accept posted writes from an external VMEbus master. If this bit is cleared while there is data in the VMEbus Slave Posted Write FIFO, the data is written to the PCI bus but no further data is accepted into this FIFO until the bit is set. IDT recommends that the attributes in a slave image not be changed while data is enqueued in the Posted Writes FIFO. To ensure data is queued from the FIFO, check the RXFE status bit in the “Miscellaneous Status Register (MISC_STAT)” on page 275 or perform a read from that image. If the programming for an image is changed after the transaction is queued in the FIFO, the transaction’s attributes are not changed. Only subsequent transactions are affected by the change in attributes. 4.2.1 VMEbus Fields Decoding for VMEbus accesses is based on the address, and address modifiers produced by the VMEbus master. Before responding to an external VMEbus master, the address must lie in the window defined by the base and bound addresses, and the Address Modifier must match one of those specified by the address space, mode, and type fields. The Universe II’s eight VMEbus slave images (images 0 to 7) are bounded by A32 space. The first and fourth of these images (VMEbus slave image 0 and 4) have a 4-Kbyte resolution while VMEbus slave images 1 to 3 and 5 to 7 have 64-Kbyte resolution (maximum image size of 4 Gbytes). The address space of a VMEbus slave image must not overlap with the address space for the Universe II’s control and status registers. Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 4. Slave Image Programming > VME Slave Image Programming 4.2.2 69 PCI Bus Fields The PCI bus fields specify how the VMEbus transaction is mapped to the appropriate PCI bus transaction. The translation offset field allows the user to translate the VMEbus address to a different address on the PCI bus. The translation of VMEbus transactions beyond 4 Gbytes results in wrap-around to the low portion of the address range. The LAS field controls generation of the PCI transaction command. The LLRMW bit allows indivisible mapping of incoming VMEbus RMW cycles to the PCI bus via the PCI LOCK_ mechanism (see “VMEbus Read-Modify-Write Cycles (RMW Cycles)” on page 37). When the LLRMW bit is set, single cycle reads are always be mapped to single data beat locked PCI transactions. Setting this bit has no effect on non-block writes: they can be coupled or decoupled. Only accesses to PCI Memory Space are decoupled, accesses to I/O or Configuration Space are always coupled. Figure 8: Address Translation Mechanism for VMEbus to PCI Bus Transfers A32 Image Offset [31..12] PCI [31..12] VME [31..12] 4.2.3 PCI [11..0] VME [11..0] Control Fields The control fields enable a VMEbus slave image (using the EN bit), as well as specify how reads and writes are processed. At power-up, all images are disabled and are configured for coupled reads and writes. If the PREN bit is set, the Universe II prefetches for incoming VMEbus block read cycles. It is the user's responsibility to ensure that prefetched reads are not destructive and that the entire image contains prefetchable resources. Prefetching is only possible in PCI Memory Space. If the PWEN bit is set, incoming write data from the VMEbus is loaded into the RXFIFO (see “Posted Writes” on page 33). Note that posted write transactions can only be mapped to Memory space on the PCI bus. Setting the LAS bit in the PCI fields to I/O or Configuration Space will force all incoming cycles to be coupled independent of this bit. Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 70 4. Slave Image Programming > PCI Bus Target Images If the LD64EN bit is set, the Universe II generates 64-bit transactions on the PCI bus by asserting REQ64_. The REQ64_ line is asserted during the address phase in a 64-bit PCI system, and is the means of determining whether the PCI target is a 64-bit port. If the target asserts ACK64_ with DEVSEL_, then the Universe II uses the 64-bit data bus. If the target does not assert ACK64_ with DEVSEL_, then the Universe II uses a 32-bit data bus. However, note that use of REQ64_ requires extra clocks internally. If no 64-bit targets are expected on the PCI bus then performance can be improved by disabling LD64EN on the VMEbus slave images. Universe II only performs 64-bit PCI transactions if the power-up option LCLSIZE bit in the “Miscellaneous Status Register (MISC_STAT)” on page 275 is set to 1. If the Universe II is set for a 32-bit PCI transaction (LCLSIZE bit set to 0) it does not perform 64-bit PCI transactions. 4.3 PCI Bus Target Images The Universe II accepts accesses from the PCI bus with programmed PCI target images. Each image opens a window to the resources of the VMEbus and allows the user to control the type of access to those resources. The Table 7, Table 8 and Table 9 describe programming for the eight standard PCI bus target images (numbered 0 to 7) by dividing them into VMEbus, PCI bus and Control fields. One special PCI target image is described in “Special PCI Target Image” on page 73. Table 7: PCI Bus Fields for the PCI Bus Target Image Field Register Bits Description Base BS[31:12] or BS[31:16] in LSIx_BS Multiples of 4 or 64 Kbytes (base to bound: maximum of 4 GBytes) Bound BD[31:12] or BD[31:16] in LSIx_BD Address space LAS in LSIx_CTL Memory or I/O Table 8: VMEbus Fields for the PCI Bus Target Image Field Register Bits Description Translation offset TO[31:12] or TO[31:16] in LSIx_TO Translates address supplied by PCI master to a specified VMEbus address Maximum data width VDW in LSIx_CTL 8, 16, 32, or 64 bits Address space VAS in LSIx_CTL A16, A24, A32, CR/CSR, User1, User2 Mode SUPER in LSIx_CTL Supervisor or non-privileged Type PGM in LSIx_CTL Program or data Cycle VCT in LSIx_CTL single or block Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 4. Slave Image Programming > PCI Bus Target Images 71 Table 9: Control Fields for PCI Bus Target Image Field Register Bits Description Image enable EN in LSIx_CTL Enable bit Posted write PWEN in LSIx_CTL enable bit IDT recommends that the attributes in a target image not be changed while data is enqueued in the Posted Writes FIFO. To ensure data is queued from the FIFO, check the TXFE status bit in the “Miscellaneous Status Register (MISC_STAT)” on page 275 or perform a read from that image. If the programming for an image is changed after the transaction is queued in the FIFO, the transaction’s attributes are not changed. Only subsequent transactions are affected by the change in attributes. 4.3.1 PCI Bus Fields All decoding for VMEbus accesses are based on the address and command information produced by a PCI bus master. The PCI Target Interface claims a cycle if there is an address match and if the command matches certain criteria. All of the Universe II’s eight PCI target images are A32-capable only. The first and fifth of them (PCI target images 0 and 4) have a 4 Kbyte resolution while PCI target images 1 to 3 and 5 to 7 have 64 Kbyte resolution. Typically, image 0 or image 4 would be used for an A16 image since they have the finest granularity. The address space of a VMEbus slave image must not overlap with the address space for the Universe II’s registers. 4.3.2 VMEbus Fields The VMEbus fields map PCI transactions to a VMEbus transaction, causing the Universe II to generate the appropriate VMEbus address, AM code, and cycle type. Some invalid combinations exist within the PCI target image definition fields. For example, A16 and CR/CSR spaces do not support block transfers, and A16 space does not support 64-bit transactions. Note that the Universe II does not attempt to detect or prevent these invalid programmed combinations, and that use of these combinations may cause illegal activity on the VMEbus. The 21-bit translation offset allows the user to translate the PCI address to a different address on the VMEbus. Figure 9 illustrates the translation process. Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 72 4. Slave Image Programming > PCI Bus Target Images Figure 9: Address Translation Mechanism for PCI Bus to VMEbus Transfers A32 Image Offset [31..12] PCI [31..12] VME [31..12] PCI [11..0] VME [11..0] Translations beyond the 4 Gbyte limit will wrap around to the low address range. The Universe II provides support for user defined AM codes. The “User AM Codes Register (USER_AM)” on page 276 contains AM codes identified as User1 and User2. The USER_AM register can only be used to generate and accept AM codes 0x10 through 0x1F. These AM codes are designated as USERAM codes in the VMEbus Specification. If the user selects one of these two, then the corresponding AM code from the global register is generated on the VMEbus. This approach results in standard single cycle transfers to A32 VMEbus address space independent of other settings in the VMEbus fields. The VCT bits in the “PCI Target Image 0 Control (LSI0_CTL)” on page 181 determine whether or not the VMEbus Master Interface will generate BLT transfers. The VCT bit will only be used if the VAS field is programmed for A24 or A32 space and the VDW bits are programmed for 8, 16, or 32 bits. If VAS bits of the control register are programmed to A24 or A32 and the VDW bits are programmed for 64-bit, the Universe II may perform MBLT transfers independent of the state of the VCT bit. 4.3.2.1 Transfers Transfers appear on the VMEbus as 16-bit transfers when the Universe II is programmed in the following manner: • PWEN= 1 • VDW=16-bit, 32-bit or 64-bit • VCT= 0 External PCI master begins a burst 32-bit write with A2=0 and BE_=0011, followed by a transfer with BE_=1100. These criteria optimize performance of 32-bit PCI systems which regularly perform 16-bit transfers. A series of 16-bit transfers is also performed if 64-bit posted write is received with BE_=11000011. Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 4. Slave Image Programming > Special PCI Target Image 4.3.3 73 Control Fields The control fields enable a PCI target image (the EN bit), as well as specify how writes are processed. If the PWEN bit is set, then the Universe II performs posted writes when that particular PCI target image is accessed. Posted write transactions are only decoded within PCI Memory space. Accesses from I/O spaces results in coupled cycles independent of the setting of the PWEN bit. 4.4 Special PCI Target Image The Universe II provides a special PCI target image located in Memory or I/O space. Its base address is aligned to 64-Mbyte boundaries and its size is fixed at 64 Mbytes (decoded using PCI address lines [31:26]). The “Special PCI Target Image (SLSI)” on page 208 is divided into four 16 Mbyte regions numbered 0 to 3 (see Figure 10 on page 74). These separate regions are selected with PCI address bits AD [25:24]. For example, if AD[25:24] = 01, then region 1 is decoded. Within each region, the upper 64Kbytes map to VMEbus A16 space, while the remaining portion of the 16 Mbytes maps to VMEbus A24 space. Note that no offsets are provided, so address information from the PCI transaction is mapped directly to the VMEbus. The general attributes of each region are programmed according to the tables below. Table 10: PCI Bus Fields for the Special PCI Target Image Field Register Bits Description Base BS[5:0] 64 Mbyte aligned base address for the image Address space LAS Places image in Memory or I/O Table 11: VMEbus Fields for the Special PCI Bus Target Image Field Register Bits Description Maximum data width VDW Separately sets each region for 16 or 32 bits Mode SUPER Separately sets each region as supervisor or non-privileged Type PGM Separately sets each region as program or data Table 12: Control Fields for the Special PCI Bus Target Image Field Register Bits Description Image enable EN Enable bit for the image Posted write PWEN Enable bit for posted writes for the image Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 74 4. Slave Image Programming > Special PCI Target Image The special PCI target image provides access to all of A16 and most of A24 space (all except the upper 64 Kbytes). By using the special PCI target image for A16 and A24 transactions, it is possible to free the eight standard PCI target images (see “PCI Bus Target Images” on page 70), which are typically programmed to access A32 space. Address space redundancy is provided in A16 space. The VMEbus specification requires only two A16 spaces, while the special PCI target image allows for four A16 address spaces. Figure 10: Memory Mapping in the Special PCI Target Image BASE+400 0000 64 Kbytes A16 BASE+3FF 0000 3 A24 16 Mbytes BASE+300 0000 BASE+2FF 0000 2 BASE+200 0000 A16 A24 A16 BASE+1FF 0000 1 BASE+100 0000 A24 A16 BASE+0FF 0000 0 A24 BASE+000 0000 Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 75 5. Registers Overview The Universe II Control and Status Registers (UCSR) occupy 4 Kbytes of internal memory. This chapter discusses the following topics: 5.1 • “Register Access from the PCI Bus” on page 76 • “Register Access from the VMEbus” on page 79 • “Mailbox Registers” on page 83 • “Semaphores” on page 83 Overview The Universe II Control and Status Registers (UCSR) occupy 4 Kbytes of internal memory. This 4 Kbytes is logically divided into the following three groups (see Figure 11): • PCI Configuration Space (PCICS) • Universe II Device Specific Registers (UDSR) • VMEbus Control and Status Registers (VCSR) The Universe II registers are little-endian. The access mechanisms for the UCSR are different depending upon whether the register space is accessed from the PCI bus or VMEbus. Register access from the PCI bus and VMEbus is discussed in the following sections. Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 76 5. Registers Overview > Register Access from the PCI Bus Figure 11: Universe II Control and Status Register Space VMEbus Configuration and Status Registers (VCSR) UNIVERSE DEVICE SPECIFIC REGISTERS (UDSR) 4 Kbytes UCSR Space PCI CONFIGURATION SPACE (PCICS) 5.2 Register Access from the PCI Bus There are different mechanisms to access the UCSR space from the PCI bus: Configuration space, PCI Memory or I/O space. 5.2.1 PCI Configuration Access When the UCSR space is accessed as Configuration space, it means that the access is externally decoded and the Universe II is notified through IDSEL (much like a standard chip select signal). Since the register location is encoded by a 6-bit register number (a value used to index a 32-bit area of Configuration space), only the lower 256 bytes of the UCSR can be accessed as Configuration space (this corresponds to the PCICS in the UCSR space, see Figure 12 on page 77). Only the PCI configuration registers are accessible through PCI Configuration cycles. Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 5. Registers Overview > Register Access from the PCI Bus 77 Figure 12: PCI Bus Access to UCSR as Memory or I/O Space VMEbus Configuration and Status Registers (VCSR) 4 Gbytes of Memory or I/O Space UNIVERSE DEVICE SPECIFIC REGISTERS (UDSR) Accessible through PCI Configuration Cycle 5.2.2 All 4 Kbytes Accessible as Memory or I/O Space PCI CONFIGURATION SPACE (PCICS) PCI_BS Memory or I/O Access Two 4-Kbyte ranges of addresses in PCI Memory space and/or PCI I/O space can be dedicated to the Universe II registers. There is one 4-Kbyte range in PCI Memory space and one 4-Kbyte range in PCI I/O space. The Universe II has the following two programmable registers: “PCI Configuration Base Address Register (PCI_BS0)” on page 178 and “PCI Configuration Base Address 1 Register (PCI_BS1)” on page 179. These register each specify the base address and address space for PCI access to the Universe II’s registers. The PCI_BSx registers can be programmed through PCI Configuration space or through a VMEbus access, to make the Universe II registers available anywhere in the 32-bit Memory space and in I/O space (as offsets of the BS[31:12] field in PCIBSx). Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 78 5. Registers Overview > Register Access from the PCI Bus The SPACE bit of the PCI_BSx registers specifies whether the address lies in Memory space or I/O space. The SPACE bit of these two registers are read-only. There is a power-up option that determines the value of the SPACE bit of the PCI_BSx registers. At power-up the SPACE bit of the PCI_BS1 register is the negation of the SPACE bit of the PCI_BS0 register. • When the VA[1] pin is sampled low at power-up, the PCI_BS0 register’s SPACE bit is set to 1, which signifies I/O space, and the PCI_BS1 register’s SPACE bit is set to 0, which signifies Memory space. • When VA[1] is sampled high at power-up, the PCI_BS0 register’s SPACE register’s bit is set to 0, which signifies Memory space, and the PCI_BS1 register’s SPACE bit is set to 1, which signifies I/O space. Universe II registers are not prefetchable and do not accept burst writes. 5.2.2.1 Conditions of Target-Retry Attempts to access UCSR space from the PCI bus can be retried by the Universe II under the following conditions: 5.2.3 • While UCSR space is being accessed by a VMEbus master, PCI masters are retried. • If a VMEbus master is performing a RMW access to the UCSRs then PCI attempts to access the USCR space results in a Target-Retry until AS* is negated. • If the Universe II registers are accessed through an ADOH cycle from the VMEbus, any PCI attempt to access the UCSRs is retried until BBSY* is negated. Locking the Register Block from the PCI bus The Universe II registers can be locked by a PCI master by using a PCI locked transaction. When an external PCI master locks the register block of the Universe II, an access to the register block from the VMEbus does not terminate with the assertion of DTACK* until the register block is unlocked. Hence a prolonged lock of the register block by a PCI resource may cause the VMEbus to timeout with a BERR*. Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 5. Registers Overview > Register Access from the VMEbus 5.3 79 Register Access from the VMEbus There are two mechanisms to access the UCSR space from the VMEbus. One method uses a VMEbus Register Access Image (VRAI) which can put the UCSR in an A16, A24 or A32 address space. The VRAI approach is useful in systems not implementing CR/CSR space as defined in the VME64 Specification. The other way to access the UCSR is as CR/CSR space, where each slot in the VMEbus system is assigned 512 Kbytes of CR/CSR space. 5.3.1 VMEbus Register Access Image (VRAI) The VMEbus Register Access Image occupies 4 Kbytes in A16, A24 or A32 space (depending upon the programming of the address space described in Table 13, and Figure 13). All registers are accessed as address offsets from the VRAI base address programmed in the “VMEbus Register Access Image Base Address Register (VRAI_BS)” on page 304. The image can be enabled or disabled using the EN bit in the“VMEbus Register Access Image Control Register (VRAI_CTL)” on page 303. The VMEbus register access image is defined by Table 13. Table 13: Programming the VMEbus Register Access Image Field Register Bits Description Address space VAS in the VRAI_CTL One of A16, A24, A32 Base address BS[31:12] in the VRAI_BSi Lowest address in the 4Kbyte slave image Slave image enable EN in the VRAI_CTL Enables VMEbus register access image Mode SUPER in the VRAI_CTL Supervisor and/or Non-Privileged Type PGM in the VRAI_CTL Program and/or Data Note that the VRAI base address can be configured as a power-up option (see “Resets, Clocks and Power-up Options” on page 129). Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 80 5. Registers Overview > Register Access from the VMEbus Figure 13: UCSR Access from the VMEbus Register Access Image VMEbus Configuration and Status Registers (VCSR) Total Memory in A16, A24 or A32 Address Space UNIVERSE DEVICE SPECIFIC REGISTERS (UDSR) 4 Kbytes of UCSR PCI CONFIGURATION SPACE (PCICS) VRAI_BS 5.3.2 CR/CSR Accesses The VME64 Specification assigns a total of 16 Mbytes of CR/CSR space for the VMEbus system. The CR/CSR image is enabled with the EN bit in the “VMEbus CSR Control Register (VCSR_CTL)” on page 305. This 16 Mbytes is broken up into 512 Kbytes per slot for a total of 32 slots. The first 512 Kbyte block is reserved for use by the Auto-ID mechanism. The UCSR space occupies the upper 4 Kbytes of the 512 Kbytes available for its slot position (see Figure 14). The base address of the CR/CSR space allocated to the Universe II’s slot is programmed in the “VMEbus CSR Base Address Register (VCSR_BS)” on page 331. Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 5. Registers Overview > Register Access from the VMEbus 81 For CSRs not supported in the Universe II and for CR accesses, the LAS field in the VCSR_CTL register specifies the PCI bus command that is generated when the cycle is mapped to the PCI bus. There is also a translation offset added to the 24-bit VMEbus address to produce a 32-bit PCI bus address which is programmed in the “VMEbus CSR Translation Offset (VCSR_TO)” on page 306. The registers in the UCSR space are located as address offsets from VCSR_BS. These offsets are different from those used in the VRAI mechanisms, where the first register in the UCSR has address offset of zero (see Table 34 on page 164). When accessing the UCSR in CR/CSR space, the first register has an address offset of 508 Kbytes (512 Kbytes minus 4 Kbytes). A simple approach for determining the register offset when accessing the UCSR in CR/CSR space is to add 508 Kbytes (0x7F000) to the address offsets given in Table 34 on page 164. 5.3.3 RMW and ADOH Register Access Cycles The Universe II supports RMW and ADOH accesses to its registers. A read-modify-write (RMW) cycle allows a VMEbus master to read from a VMEbus slave and then write to the same resource without relinquishing VMEbus tenure between the two operations. The Universe II accepts RMW cycles to any of its registers. This prevents an external PCI Master from accessing the registers of the Universe II until VMEbus AS* is asserted. This is useful if a single RMW access to the ADOH is required. If a sequence of accesses to the Universe registers must be performed without intervening PCI access to UCSR is required, then the VMEbus master should lock the Universe II through the use of ADOH. This prevents an external PCI Master from accessing the registers of the Universe II until VMEbus BBSY* is negated. It also prevents other VMEbus masters from accessing the Universe II registers. Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 82 5. Registers Overview > Register Access from the VMEbus Figure 14: UCSR Access in VMEbus CR/CSR Space VMEbus Configuration and Status Registers (VCSR) UNIVERSE DEVICE SPECIFIC REGISTERS (UDSR) 4 Kbytes of UCSR PCI CONFIGURATION SPACE (PCICS) 512 Kbytes of VMEbus CR/CSR Space (Portion of 16 Mbyte Total for Entire VMEbus System) Mapped to PCI VCSR_BS Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 5. Registers Overview > Mailbox Registers 5.4 83 Mailbox Registers The Universe II has four 32-bit mailbox registers which provide an additional communication path between the VMEbus and the PCI bus (see “Mailbox 0 Register (MBOX0)” on page 265 to “Mailbox 3 Register (MBOX3)” on page 268). The mailboxes support read and write accesses from either bus, and may be enabled to generate interrupts on either bus when they are written to. The mailboxes are accessible from the same address spaces and in the same manner as the other Universe II registers, as described above. Mailbox registers are useful for the communication of concise command, status, and parameter data. The specific uses of mailboxes depend on the application. For example, they can be used when a master on one bus needs to pass information (a message) on the other bus, without knowing where the information should be stored in the other bus’s address space. Or they can be used to store the address of a longer message written by the processor on one bus to the address space on the other bus, through the Universe II. They can also be used to initiate larger transfers through the FIFO, in a user-defined manner. Often users will enable and map mailbox interrupts, so that when the processor writes to a mailbox from one bus, the Universe II will interrupt the opposite bus. The interrupt service routine on the opposite bus would then cause a read from this same mailbox. Reading a mailbox cannot automatically trigger an interrupt. However, a similar effect can be achieved by reading the mailbox and then triggering an interrupt through hardware or software. Or one may use a “polling” approach, where one designates a bit in a mailbox register to indicate whether one has read from the mailbox. For details on how the mailbox interrupts are enabled and mapped, see “Interrupt Generation and Handling” on page 109 and “Mailbox Register Access Interrupts” on page 122. Applications will sometimes designate two mailboxes on one interface as being read/write from the PCI bus, and read-only from the VMEbus, and the two other mailboxes as read/write from the VMEbus and read-only from the PCI bus. This eliminates the need to implement locking. The Universe II provides semaphores which can be also be used to synchronize access to the mailboxes. Semaphores are described in the next section. 5.5 Semaphores The Universe II has two general-purpose semaphore registers each containing four semaphores ( “Semaphore 0 Register (SEMA0)” on page 269 and “Semaphore 1 Register (SEMA1)” on page 270). To gain ownership of a semaphore, a process writes a logic 1 to the semaphore bit and a unique pattern to the associated tag field. If a subsequent read of the tag field returns the same pattern, the process can consider itself the owner of the semaphore. A process writes a value of 0 to the semaphore to release it. When a semaphore bit is a value of 1, the associated tag field cannot be updated. Only when a semaphore is a value of 0 can the associated tag field be updated. These semaphores shares resources in the system. While the Universe II provides the semaphores, it is up to the user to determine access to which part of the system will be controlled by semaphores, and to design the system to enforce these rules. Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 84 5. Registers Overview > Semaphores An example of a use of the semaphore involves gating access to the Special Cycle Generator (“Special Cycle Generator” on page 61). It may be necessary to ensure that while one process uses the Special Cycle Generator on an address, no other process accesses this address. Before performing a Special Cycle, a process would be required to obtain the semaphore. This process would hold the semaphore until the Special Cycle completes. A separate process that intends to modify the same address would need to obtain the semaphore before proceeding (it need not verify the state of the SCYC[1:0] bit). This mechanism requires that processes know which addresses might be accessed through the Special Cycle Generator. Each of the four semaphores in a semaphore register are intended to be accessed with 8-bit transfers. Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 85 6. DMA Controller Direct memory access (DMA) allows a transaction to occur between two devices without involving the host processor (for example, a read transaction between a peripheral device and host processor memory). Because less time is required to complete transactions, applications that contain one or more DMA channels support faster read and write transfers than applications that support only host-assisted transactions. This chapter discusses the following topics: 6.1 • “DMA Registers” on page 85 • “Direct Mode Operation” on page 93 • “Linked-list Mode” on page 96 • “FIFO Operation and Bus Ownership” on page 101 • “DMA Interrupts” on page 104 • “DMA Channel Interactions with Other Channels” on page 105 • “DMA Error Handling” on page 105 Overview The Universe II has a DMA controller for high performance data transfer between the PCI bus and VMEbus. It is operated through a series of registers that control the source and destination for the data, length of the transfer and the transfer protocol to be used. There are two modes of operation for the DMA: Direct Mode, and Linked List Mode. In direct mode, the DMA registers are programmed directly by the external PCI master. In linked list mode, the registers are loaded from PCI memory by the Universe II, and the transfer described by these registers is executed. A block of DMA registers stored in PCI memory is called a command packet. A command packet can be linked to another command packet, so that when the DMA has completed the operations described by one command packet, it automatically moves on to the next command packed in the linked-list of command packets. 6.2 DMA Registers The DMA registers reside in a register block starting at offset 0x200. The registers describe the following information for a single DMA transfer: • Source: where to transfer data from • Destination: where to transfer data to • Size: how much data to transfer • Attributes: the transfer attributes to use on the PCI bus and VMEbus. Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 86 6. DMA Controller > DMA Registers A final register contains status and control information for the transfer. While the DMA is active, the registers are locked against any changes so that any writes to the registers will have no impact. In direct-mode operation, these registers are programmed directly. In linked-list operation, they are repeatedly loaded by the Universe II from command packets residing in PCI memory until the end of the linked-list is reached (see “Linked-list Mode” on page 96). 6.2.1 Source and Destination Addresses The source and destination addresses for the DMA reside in two registers: the “DMA PCI Bus Address Register (DLA)” on page 235), and the “DMA VMEbus Address Register (DVA)” on page 236. The determination of which is the source address, and which is the destination is made by the L2V bit in the “DMA Transfer Control Register (DCTL)” on page 232. When set, the DMA transfers data from the PCI to the VMEbus. The DLA becomes the PCI source register and DVA becomes the VMEbus destination register. When cleared, the DMA transfers data from the VMEbus to PCI bus and DLA becomes the PCI destination register; DVA becomes the VMEbus source register. The PCI address may be programmed to any byte address in PCI Memory space. It cannot transfer to or from PCI I/O or Configuration spaces. The VMEbus address can also be programmed to any byte address, and can access any VMEbus address space from A16 to A32 in supervisory or non-privileged space, and data or program space. The setting of address space, A16, A24 or A32, is programmed in the VAS field of the “DMA Transfer Control Register (DCTL)” on page 232. The sub-spaces are programmed in the PGM and SUPER fields of the same register. Although the PCI and VMEbus addresses may be programmed to any byte aligned address, they must be 8-byte aligned to each other (the low three bits of each must be identical). If not programmed with aligned source and destination addresses and an attempt to start the DMA is made, the DMA does not start. It sets the protocol error bit (P_ERR) in the “DMA General Control/Status Register (DGCS)” on page 238 and,if enabled to, generates an interrupt. When this occurs linked-list operations cease. In direct mode the user must reprogram the source and destination address registers (DMA, DLA) before each transfer. These registers are not updated in direct mode. In linked-list mode, these registers are updated by the DMA when the DMA is stopped, halted, or at the completion of processing a command packet. If read during DMA activity, they return the number of bytes remaining to transfer on the PCI side. All of the DMA registers are locked against any changes by the user while the DMA is active. When stopped due to an error situation, the DLA and DVA registers must not be used, but the DTBC is valid (see “DMA Error Handling” on page 105 for details). At the end of a successful linked-list transfer, the DVA and DLA registers point to the next address at the end of the transfer block, and the DTBC register is 0. Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 6. DMA Controller > DMA Registers 6.2.2 87 Non-incrementing DMA Mode The VMEbus Non-Incrementing Mode (Non-Inc Mode) enables the DMA Controller to perform transfers to or from a fixed VMEbus address. This means that the specified VMEbus address is not incremented during DMA reads or writes. This applies to both Direct and Linked List modes of DMA operation. For more information on these two types of DMA operation, refer to “Direct Mode Operation” on page 93 and “Linked-list Mode” on page 96. Unlike incrementing DMA operation, in Non-Inc Mode the DMA Controller can only perform 8-,16or 32-bit single cycle transfers on the VMEbus. This means that BLT and MBLT transfers cannot be performed when operating in Non-Inc Mode. 6.2.2.1 Using Non-Inc Mode The VMEbus Non-Inc Mode is enabled by writing a 1 to the NO_VINC bit in the “DMA Transfer Control Register (DCTL)” on page 232. In order to set-up and initiate DMA operation, the same steps which are described in the DMA Controller Section in the Universe II User Manual must be followed for Non-Inc Mode. The steps in setting-up and initiating DMA operation are as follows: 1. Program the tenure and interrupt requirements in the DGCS register (offset 0x220). 2. Program the source and destination addresses in the DLA and DVA registers. 3. Set the GO bit in the DGCS register. 6.2.2.2 Issues with Non-Inc Mode The VMEbus Address In Non-Inc Mode, the DVA register (offset 0x210) does not necessarily contain the fixed VMEbus address. This register must not be read during a DMA Non-Inc Mode transfer. Once a DMA transfer has been stopped — by setting the STOP bit of the DGCS register — the Non-Inc Mode transfer cannot be restarted by simply writing a 1 to the GO bit of the DGCS register. The DVA register must be reprogrammed with the required address before setting the DGCS GO bit. The VON Counter When the VON counter in the DGCS register reaches its programmed limit, the VMEbus Master Interface of the Universe II stops transferring data until the VOFF timer expires. If the device is operating in Non-Inc Mode, the VON counter has different limits than those indicated in the DMA Controller section. Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 88 6. DMA Controller > DMA Registers The different settings are detailed in Table 14. Table 14: VON Settings for Non-Inc Mode VON VMEbus Aligned DMA Transfer Count 001 128 bytes 010 256 bytes 011 512 bytes 100 1024 bytes 101 2048 bytes 110 4096 bytes 111 8192 bytes P_ERR Flag Behavior When the GO bit is set in Non-Inc Mode, the P_ERR flag of the “DMA General Control/Status Register (DGCS)” on page 238 is 1 when the following conditions are true: • VCT bit of the “DMA Transfer Control Register (DCTL)” on page 232 has a value of 1 • VDW field of the DCTL register has a value of 01 and bit 0 of the DVA register is a value of 0 • VDW field of the DCTL register has a value of 10 and bits 1 and 0 of the DVA register are non-zero • VDW field of the DCTL register has a value of 11. Single Cycle Transfers The Universe II performs 8-,16- or 32-bit single cycle transfers on the VMEbus. BLT and MBLT transfers cannot be performed when operating in Non-Inc Mode. 6.2.2.3 Non-Inc Mode Performance The transfer performance of DMA in Non-Inc Mode has been simulated at 14 MB/s for 32-bit writes and at 8 MB/s for 32-bit reads. This performance was determined using ideal slave responses; lower performance can be expected in actual systems. Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 6. DMA Controller > DMA Registers 6.2.3 89 Transfer Size The DMA can be programmed through the “DMA Transfer Byte Count Register (DTBC)” on page 234 to transfer any number of bytes from 1 byte to 16 MBytes. There are no alignment requirements to the source or destination addresses. If the width of the data turnovers (8- through 64-bit on VMEbus and 32- or 64-bit on PCI) do not align to the length of the transfer or the source/destination addresses, the DMA inserts transfers of smaller width on the appropriate bus. For example, if a 15-byte transfer is programmed to start at address 0x1000 on the VMEbus, and the width is set for D32, the DMA will perform three D32 transfers, followed by a D08 transfer. The Universe II does not generate unaligned transfers. On a 32-bit PCI bus, if the start address was 0x2000, the DMA would generate three data beats with all byte lanes enabled, and a fourth with three byte lanes enabled. The DTBC register is not updated while the DMA is active (DMA is indicated as active by the ACT bit in the DGCS register). At the end of a transfer it contains 0. However, if stopped by the user (through the STOP bit in the DGCS register) or the DMA encounters an error, the DTBC register contains the number of bytes remaining to transfer on the source side (see “DMA Error Handling” on page 105). Starting the DMA while DTBC is 0 results in one of two situations. If the CHAIN bit in the “DMA General Control/Status Register (DGCS)” on page 238 is not set, the DMA does not start; it performs no action. If the CHAIN bit is set, then the DMA loads the DMA registers with the contents of the command packet pointed to by the“DMA Command Packet Pointer (DCPP)” on page 237, and starts the transfers described by that packet. Note that the DCPP[31:5] field of the DCPP register implies that the command packets be 32-byte aligned (bits 4:0 of this register must be 0). 6.2.4 Transfer Data Width The VMEbus and PCI bus data widths are determined by three fields in the “DMA Transfer Control Register (DCTL)” on page 232. These fields affect the speed of the transfer. They should be set for the maximum allowable width that the destination device is capable of accepting. On the VMEbus, the DMA supports the following data widths: • D08(EO) • D8BLT • D16 • D16BLT • D32 • D64 • D32BLT • D64BLT (MBLT) The width of the transfer is set with the VDW field in the DCTL register. The VCT bit determines whether or not the Universe II VMEbus Master will generate BLT transfers. The value of this bit only has meaning if the address space is A24 or A32 and the data width is not 64 bits. If the data width is 64 bits the Universe II may perform MBLT transfers independent of the state of the VCT bit. Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 90 6. DMA Controller > DMA Registers The Universe II can perform data transfers smaller than that programmed in the VDW field in order to bring itself into alignment with the programmed width. For example if the width is set for D32 and the starting VMEbus address is 0x101, the DMA performs a D08 cycle. Only once it has achieved the alignment set in the VDW field does it start D32 transfers. At the end of the transfer, the DMA also performs more low-width transfers if the last address is not aligned to VDW. Similarly, if the VCT bit is set to enable block transfers, the DMA can perform non-block transfers to bring itself into alignment. On the PCI bus, the DMA provides the option of performing 32- or 64-bit PCI transactions through the LD64EN bit in the “DMA Transfer Control Register (DCTL)” on page 232 If the Universe II has powered-up on a 32-bit bus (see “Power-Up Options” on page 135), this bit has have no effect. If powered-up on a 64-bit bus, this bit can provide some performance improvements when accessing 32-bit targets on that bus. Following the PCI 2.1 Specification, before a 64-bit PCI initiator starts a 64-bit transaction, it engages in a protocol with the intended target to determine if it is 64-bit capable. This protocol typically consumes one clock period. To save bandwidth, the LD64EN bit can be cleared to bypass this protocol when it is known that the target is only 32-bit capable. 6.2.5 DMA Command Packet Pointer The “DMA Command Packet Pointer (DCPP)” on page 237 points to a 32-byte aligned address location in PCI Memory space that contains the next command packet to be loaded once the transfer currently programmed into the DMA registers has been successfully completed. When it has been completed (or the DTBC register is 0 when the GO bit is set) the DMA reads the 32-byte command packet from PCI memory and executes the transfer it describes. 6.2.6 DMA Control and Status “DMA General Control/Status Register (DGCS)” on page 238 contains a number of fields that control initiation and operation of the DMA as well as actions to be taken on completion. 6.2.6.1 DMA Initiation Once all the parameters associated with the transfer have been programmed (source/destination addresses, transfer length and data widths, and if desired, linked lists enabled), the DMA transfer is started by setting the GO bit in the DGCS register. This causes the DMA first to examine the DTBC register. If it is non-zero, it latches the values programmed into the DCTL, DTBC, DLA, and DVA registers and initiates the transfer programmed into those registers. If DTBC=0, it checks the CHAIN bit in the DGCS register and if that bit is cleared it assumes the transfer to have completed and stops. Otherwise, if the CHAIN bit is set, it loads into the DMA registers the command packet pointed to by the DCPP register and initiates the transfer describe there. If the GO bit is set, but the Universe II has not been enabled as a PCI master with the BM (bus master enable) bit in the “PCI Configuration Space Control and Status Register (PCI_CSR)” on page 172, or if the DVA and DLA contents are not 64-bit aligned to each other, the transfer does not start, a protocol error is indicated by the P_ERR bit in the “DMA General Control/Status Register (DGCS)” on page 238 and, if enabled, an interrupt is generated. Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 6. DMA Controller > DMA Registers 91 If the DMA has been terminated (stopped, halted, or error), all DMA registers contain values indicating where the DMA terminated. Once all status bits have been cleared, the DMA may be restarted from where it left off by simply setting the GO bit. The GO bit only has an effect if all status bits have been cleared. These bits include STOP, HALT, DONE, LERR, VERR, and P_ERR and are located in the “DMA General Control/Status Register (DGCS)” on page 238. These bits are all cleared by writing 1 to them, either before or while setting the GO bit. The GO bit always returns a 0 when read independent of the DMA’s current state. Clearing the bit has no impact at any time. The ACT bit in the DGCS register indicates whether the DMA is currently active. It is set by the DMA once the GO bit is set, and cleared when the DMA is idle. Generally, when the ACT bit is cleared, one of the other status bits in the DGCS register is set (DONE, STOP, HALT, LERR, VERR, or P_ERR), indicating why the DMA is no longer active. 6.2.6.2 DMA VMEbus Ownership Two fields in the DGCS register determine how the DMA shares the VMEbus with the other two potential masters in the Universe II (PCI Target Channel, and Interrupt Channel), and with other VMEbus masters on the bus. These fields are: VON and VOFF. VON VON affects how much data the DMA transfers before giving the opportunity to another master (either the Universe II or an external master) to assume ownership of the bus. The VON counter is used to temporarily stop the DMA from transferring data once a programmed number of bytes have been transferred (256 bytes, 512 bytes, 1K, 2K, 4K, 8K, or 16K). When performing MBLT transfers on the VMEbus, the DMA stops performing transfers within 2048 bytes after the programmed VON limit has been reached. When not performing MBLT transfers, the DMA will stop performing transfers within 256 bytes once the programmed limit has been reached. When programmed for Release-When-Done operation, the Universe II performs an early release of BBSY* when the VON counter reaches its programmed limit. VON may be disabled by setting the field to zero. When the VON bit is set to 0, the DMA continues transferring data as long as it is able. There are other conditions under which the DMA may relinquish bus ownership. See“FIFO Operation and Bus Ownership” on page 101 for details on the VMEbus request and release conditions for the DMA. VOFF VOFF affects how long the DMA waits before re-requesting the bus after the VON limit has been reached. By setting VOFF to 0, the DMA immediately re-requests the bus once the VON boundary has been reached. Since the DMA operates in a round-robin fashion with the PCI Target Channel, and in a priority fashion with the Interrupt Channel, if either of these channels require ownership of the VMEbus, they receive it at this time. VOFF is only invoked when VMEbus tenure is relinquished due to encountering the VON boundary. When the VMEbus is released due to other conditions (for example, the DMAFIFO has become full while reading from the VMEbus), it will be re-requested as soon as that condition is cleared. The VOFF timer can be programmed to various time intervals from 0µs to 1024µs. See “FIFO Operation and Bus Ownership” on page 101 for details on the VMEbus request and release conditions for the DMA. Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 92 6. DMA Controller > DMA Registers See “DMA Channel Interactions with Other Channels” on page 105 for information on other mechanisms which can delay the DMA Channel from acquiring the VMEbus or the PCI bus. 6.2.6.3 DMA Completion and Termination Normally, the DMA continues processing its transfers and command packets until either it completes all requests, or it encounters an error. There are also two methods for the user to interrupt this process and cause the DMA to terminate prematurely: STOP and HALT. STOP causes the DMA to terminate immediately, while HALT causes the DMA to terminate when it has completed processing the current command packet in a linked list. STOP When the STOP_REQ bit in the DGCS register is set, it tells the DMA to cease its operations on the source bus immediately. Remaining data in the FIFO continues to be written to the destination bus until the FIFO is empty. Once the FIFO is empty, the STOP bit in the same register is set and, if enabled, an interrupt generated. The DMA registers contain the values that the DMA stopped at: the DTBC register contains the number of bytes remaining in the transfer, the source and destination address registers contain the next address to be read/written, the DCPP register contains the next command packet in the linked-list, and the DCTL register contains the transfer attributes. If read transactions are occurring on the VMEbus, then setting a stop request can be affected by the VOFF timer. If the STOP_REQ bit is set while the DMA is lying idle waiting for VOFF to expire before re-starting reads, then the request remains pending until the VOFF timer has expired and the bus has been granted. HALT HALT provides a mechanism to interrupt the DMA at command packet boundaries during a linked-list mode transfer. In contrast, a STOP requests the DMA to be interrupted immediately, while halt takes effect only when the current command packet is complete. A halt is requested of the DMA by setting the HALT_REQ bit in the DGCS register. This causes the DMA to complete the transfers defined by the current contents of the DMA registers and, if the CHAIN bit is set, load in the next command packet. The DMA then terminates, the HALT bit in the DGCS register is set, and, if enabled, an interrupt generated. After a stop or halt, the DMA can be restarted from the point it left off by setting the GO bit; but before it can be re-started, the STOP and HALT bits must be cleared. Regardless of how the DMA stops—whether normal, bus error or user interrupted—the DMA indicates in the DGCS register why it stopped. The STOP and HALT bits are set in response to a stop or halt request. The DONE bit gets set when the DMA has successfully completed the DMA transfer, including all entries in the linked-list (if operating in that mode). There are also three bits that are set in response to error conditions: LERR in the case of Target-Abort encountered on the PCI bus; VERR in the case of a bus error encountered on the VMEbus; and P_ERR in the case that the DMA has not been properly programmed (the DMA was started with the BM bit in the PCI_CSR register not enabled, or the DLA and DVA registers were not 64-bit aligned, (see “Source and Destination Addresses” on page 86). Before the DMA can be restarted, each of these status bits must be cleared. Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 6. DMA Controller > Direct Mode Operation 93 When the DMA terminates, an interrupt may be generated to VMEbus or PCI bus. The user has control over which DMA termination conditions cause the interrupt through the INT_STOP, INT_HALT, INT_DONE, INT_LERR, INT_VERR, and INT_P_ERR bits in the DGCS register. 6.3 Direct Mode Operation When operated in direct mode, the Universe II DMA is set through manual register programming. Once the transfer described by the DVA, DLA, DTBC and DCTL registers has been completed, the DMA sits idle awaiting the next manual programming of the registers. Figure 15 describes the steps involved in operating the DMA in direct mode. Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 94 6. DMA Controller > Direct Mode Operation Figure 15: Direct Mode DMA transfers Step 1: Program DGCS with tenure and interrupt requirements Step 2: Program source/destination addresses, & transfer size/attributes Step 3: Ensure status bits are clear Step 4: Set GO bit Step 5: Await termination of DMA Normal Termination? No Handle error Yes Yes More transfers required? No Done Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 6. DMA Controller > Direct Mode Operation 95 In Step 1, the “DMA General Control/Status Register (DGCS)” on page 238is set up — The CHAIN bit is cleared, VON and VOFF are programmed with the appropriate values for controlling DMA VMEbus tenure, and the interrupt bits (INT_STOP, INT_HALT, INT_DONE, INT_LERR, INT_VERR, and INT_P_ERR) are programmed to enable generation of interrupts based on DMA termination events. DMA interrupt enable bits in the LINT_EN or VINT_EN bits should also be enabled as necessary (see “PCI Interrupt Generation” on page 111 and “VMEbus Interrupt Generation” on page 113 for details on generating interrupts). In Step 2, the actual transfer is programmed into the DMA — Source and destination start addresses into the DLA and DVA registers, transfer count into the DTBC register, and transfer width, direction and VMEbus address space into the DCTL register. These should be reprogrammed after each transfer. In Step 3, ensure that if any status bits (DONE, STOP, HALT, LERR, VERR, or P_ERR) remain set from a previous transfer they are cleared. — P_ERR must not be updated at the same time as Step 4, otherwise the P_ERR that may be generated by setting GO may be missed (see Step 4). These bits can be cleared as part of Step 1. In Step 4, with the transfer programmed, the GO bit in DGCS must be set. — If the DMA has been improperly programmed, either because the BM bit in the PCI_CSR has not been set to enable PCI bus mastership, or the source and destination start addresses are not aligned, then P_ERR will be asserted. Otherwise, the ACT bit will be set, and the DMA will then start transferring data, sharing ownership of the VMEbus with the PCI Target and Interrupt channels and the PCI bus with the VMEbus Slave Channel. In Step 5, the DMA waits for termination of the DMA transfers. — The DMA continues with the transfers until it: – Completes all transfers – Terminates early with the STOP_REQ bit – Encounters an error on the PCI bus or VMEbus Each of these conditions cause the ACT bit to clear, and a corresponding status bit to be set in the DGCS register. If enabled in Step 1, an interrupt is generated. Once the software has set the GO bit, the software can monitor for DMA completion by either waiting for generation of an interrupt, or by polling the status bits. It is recommended that a background timer also be initiated to time-out the transfer. This ensures the DMA has not been hung up by a busy VMEbus, or other such system issues. If an early termination is desired (perhaps because a higher priority operation is required) the STOP_REQ bit in the DGCS register can be set. This stops all DMA operations on the source bus immediately, and set the STOP bit in the same register when the last piece of queued data in the DMA FIFO has been written to the destination bus. Attempting to terminate the transfer with the HALT_REQ bit has no effect in direct mode operation since this bit only requests the DMA to stop between command packets in linked-list mode operation. Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 96 6. DMA Controller > Linked-list Mode When the software has detected completion, it must verify the status bits in the DGCS register to see the reason for completion. If one of the error bits have been set it proceeds into an error handling routine (see “DMA Error Handling” on page 105). If the STOP bit was set, the software must take whatever actions were programmed when it set the STOP_REQ bit. For example, if it was stopped for a higher priority transfer, it might record the DLA, DVA and DTBC registers, and then reprogram them with the higher priority transfer. When that has completed it can restore the DVA, DLA and DTBC registers to complete the remaining transfers. If the DONE bit was set, it indicates that the DMA completed its requested transfer successfully, and if more transfers are required, the software can proceed to Step 2 to start a new transfer. 6.4 Linked-list Mode Unlike direct mode, in which the DMA performs a single block of data at a time, linked-list mode allows the DMA to transfer a series of non-contiguous blocks of data without software intervention. Each entry in the linked-list is described by a command packet which parallels the DMA register layout. The data structure for each command packet is the same (see Figure 16 below), and contains all the necessary information to program the DMA address and control registers. It could be described in software as a record of eight 32-bit data elements. Four of the elements represent the four core registers required to define a DMA transfer: DCTL, DTBC, DVA, and DLA. A fifth element represents the DCPP register which points to the next command packet in the list. The least two significant bits of the DCPP element (the PROCESSED and NULL bits) provide status and control information for linked list processing. The PROCESSED bit indicates whether a command packet has been processed or not. When the DMA processes the command packet and has successfully completed all transfers described by this packet, it sets the PROCESSED bit to 1 before reading in the next command packet in the list. The PROCESSED bit must be initially set for 0. This bit, when set to 1, indicates that this command packet has been disposed of by the DMA and its memory can be de-allocated or reused for another transfer description. Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 6. DMA Controller > Linked-list Mode 97 Figure 16: Command Packet Structure and Linked List Operation First Command Packet in Linked-List DCTL Register Linked-List Start Address in Command Packet Pointer Register DTBC Register Register information copied to DMA Control and Address Registers DLA Register reserved DVA Register reserved DCPP Register r PN reserved DCPP points to next command packet in Linked-List Second Command Packet in Linked-List r = reserved P = processed bit (after command packet is processed the bit is set to 1) N = null bit DCTL Register DTBC Register DLA Register reserved DVA Register reserved DCPP Register r PN reserved N = 0 for another command packet Last Command Packet in Linked-List DCTL Register DTBC Register DLA Register reserved DVA Register reserved DCPP Register reserved r PN N = 1 for last command packet The NULL bit indicates the termination of the entire linked list. If the NULL bit is set to 0, the DMA processes the next command packet pointed to by the command packet pointer. If the NULL bit is set to 1 then the address in the command packet pointer is considered invalid and the DMA stops at the completion of the transfer described by the current command packet. Figure 17 outlines the steps in programming the DMA for linked-list operation. Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 98 6. DMA Controller > Linked-list Mode Figure 17: DMA Linked List Operation Step 1: Program DGCS with tenure and interrupt requirements Step 2 : Set up linked-list in PCI memory space Step 3 : Clear DTBC register, program DCPP Step 4 : Set GO bit Step 5 : Await termination of DMA Normal Termination? No handle error Yes Done In Step 1, the DGCS register is set up — The CHAIN bit is set, VON and VOFF are programmed with the appropriate values for controlling DMA VMEbus tenure, and the interrupt bits (INT_STOP, INT_HALT, INT_DONE, INT_LERR, INT_VERR, and INT_P_ERR) are programmed to enable generation of interrupts based on DMA termination events. DMA interrupt enable bits in the LINT_EN or VINT_EN bits should also be enabled as necessary (“PCI Interrupt Generation” on page 111 and “VMEbus Interrupt Generation” on page 113). In Step 2, the linked-list structure is programmed with the required transfers. — The actual structure may be set up at any time with command packet pointers pre-programmed and then only the remaining DMA transfer elements need be programmed later. One common way is to set up the command packets as a circular queue: each packet points to the next in the list, and the last points to the first. This allows continuous programming of the packets without having to set-up or tear down packets later. Once the structure for the linked-list is established, the individual packets are programmed with the appropriate source and destination addresses, transfer sizes and attributes. Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 6. DMA Controller > Linked-list Mode 99 In Step 3, Clear the DTBC register and program the DCPP register to point to the first command packet in the list. When using the DMA to perform linked-list transfers, it is important to ensure that the DTBC register contains a value of zero before setting the GO bit of the DGCS register. Otherwise, the DMA cannot read the first command packet but instead performs a direct mode transfer based on the contents of the DCTL, DTBC, DLA, DVA and DGCS registers. After this direct mode transfer is completed, the PROCESSED bit of the first command packet is programmed with a value of 1 even though the packet was not actually processed. The DMA continues as expected with the next command packet. In Step 4, to start the linked-list transfer, set the GO bit in the DGCS register. — The DMA first performs the transfers defined by the current contents of the DCTL, DTBC, DVA and DLA registers. Once that is complete it then starts the transfers defined by the linked-list pointed to in the DCPP register. In Step 5, await and deal with termination of the DMA. — Once the DMA channel is enabled, it processes the first command packet as specified by the DCPP register. The DMA transfer registers are programmed by information in the command packets and the DMA transfer steps along each command packet in sequence (see Figure 16). The DMA terminate when one of the following conditions are met: – Processes a command packet with the NULL bit set indicating the last packet of the list – Stops with the STOP_REQ bit in the DGCS register – Halts with the HALT_REQ bit in the DGCS register – Encounters an error on either the PCI bus or VMEbus Each of these conditions cause the ACT bit to clear, and a corresponding status bit to be set in the DGCS register. If enabled in step 1, an interrupt is generated. Once the software has set the GO bit, the software can monitor for DMA completion by either waiting for generation of an interrupt, by polling the status bits in the DGCS register, or by polling the PROCESSED bits of the command packets. It is recommended that a background timer also be initiated to time-out the transfer. This ensures that the DMA has not been hung up by a busy VMEbus, or other such system issues. Linked-list mode can be halted by setting the HALT_REQ bit in the “DMA Transfer Control Register (DCTL)” on page 232. When the HALT_REQ bit is set, the DMA terminates when all transfers defined by the current command packet is complete. It then loads the next command packet into its registers. The HALT bit in the DGCS register is asserted, and the ACT bit in the DGCS register is cleared. The PROCESSED bit in the linked-list is set to 1 approximately 1 s after the HALT bit is set: therefore after a DMA halt the user should wait at least 1 s before checking the status of the PROCESSED bit. The DMA can be restarted by clearing the HALT status bit and setting the GO bit during the same register write. If the DMA is restarted, the ACT bit is set by the Universe II and execution continues as if no HALT had occurred (that is, the Universe II processes the current command packet (see Figure 16)). Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 100 6. DMA Controller > Linked-list Mode In contrast to a halt, the DMA can also be immediately terminated through the STOP_REQ bit. This stops all DMA operations on the source bus immediately, and set the STOP bit in the same register when the last piece of queued data in the DMA FIFO has been written to the destination bus. Once stopped, the DVA, DLA and DTBC registers contain values indicating the next addresses to read/write and the number of bytes remaining in the transfer. Clearing the STOP bit and setting the GO bit causes the DMA to start-up again from where it left off, including continuing with subsequent command packets in the list. If the DMA is being stopped to insert a high priority DMA transfer, the remaining portion of the DMA transfer can be stored as a new command packet inserted at the top of the linked list. A new command packet with the attributes of the high priority transfer is then placed before that one in the list. Now the linked list is set up with the high priority packet first, followed be the remainder of the interrupted packet, followed in turn by the rest of the linked list. Finally, the DTBC register is cleared and the DCPP programmed with a pointer to the top of the list where the high priority command packet has been placed. When the GO bit is set (after clearing the STOP status bit in the DGCS register), the DMA performs the transfers in the order set in the linked list. For more details on updating the linked list see “Linked-list Updating” on page 100. DMA transfers continue until the DMA encounters a command packet with the NULL bit set to 1, indicating that the last packet has been reached. At this point, the DMA stops, the DONE bit is set, and the ACT flag is cleared. As it completes the transfers indicated by each command packet, the DMA sets the PROCESSED bit in that command packet before reading in the next command packet and processing its contents. 6.4.1 Linked-list Updating The Universe II provides a mechanism which enables the linked-list to be updated with additional linked list entries without halting or stopping the DMA. This takes place through the use of a semaphore in the device: the UPDATE bit in the “DMA Linked List Update Enable Register (D_LLUE)” on page 241. This bit is meant to ensure that the DMA does not read a command packet into the DMA registers while the command packet (outside the Universe II) is being updated. This semaphore does not prevent external masters from updating the DMA registers. Adding to a linked list begins by writing a 1 to the UPDATE bit. The DMA checks this bit before proceeding to the next command packet. If the UPDATE bit is 0, then the DMA locks the UPDATE bit against writes and proceeds to the next command packet. If the UPDATE bit is 1, then the DMA waits until the bit is cleared before proceeding to the next command packet. Setting the UPDATE bit is a means of stalling the DMA at command packet boundaries while local logic updates the linked-list. In order to ensure that the DMA is not currently reading a command packet during updates, the update logic must write a 1 to the UPDATE bit and read a value back. If a 0 is read back from the UPDATE bit, then the DMA is currently reading a command packet and has locked the UPDATE bit against writes. If a 1 is read back from the UPDATE bit, then the DMA is idle or processing a transaction and command packets can be updated. If the DMA attempts to proceed to the next command packet during the update, it encounters the set UPDATE bit and wait until the bit is cleared. Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 6. DMA Controller > FIFO Operation and Bus Ownership 101 If a set of linked command packets has already been created with empty packets at the end of new transfers, adding to the end of the current linked list is accomplished by the following steps: 1. Get UPDATE valid (write 1, read back 1) 2. Program attributes for new transfer in next available packet in list 3. Change null pointer (on previous tail of linked list) 4. Release update (clear the UPDATE bit) After updating the linked list, the DMA controller is in one of the following conditions: 1. It can be active and working its way through the linked list. In this case, no further steps are required. 2. The DMA can be idle (done) because it reached the final command packet. If a full set of linked command packets had already been created ahead of time, then the DCPP register points to the most recently programmed command packet, and the DTBC register would be zero. The DMA can be started on the new packet by simply clearing the DONE bit and setting the GO bit in the DGCS register. If a set of command packets have not been created ahead of time, the DCPP register can not be programmed to any valid packet, and needs programming to the newly programmed packet. 3. The DMA has encountered an error. In this circumstance, see “DMA Error Handling” on page 105 for how to handle DMA errors. Operation can be considerably simplified by ensuring that sufficient command packets have been created during system initialization, probably in a circular queue. In this fashion, when a new entry is added to the list, it is simply a matter of programming the next available entry in the list with the new transfer attributes and changing the previously last packet's NULL bit to zero. The DCPP register is guaranteed to point to a valid command packet, so upon updating the list, both cases 1 and 2 above can be covered by clearing the DONE bit and setting the GO bit. This has no effect for case 1 since the DMA is still active, and restarts the DMA for case 2. If an error has been encountered by the DMA (case 3), setting the GO bit and clearing the DONE bit is not be sufficient to restart the DMA—the error bits in the DGCS register also has to be cleared before operation can continue. 6.5 FIFO Operation and Bus Ownership The DMA uses a 256-byte, 64-bit FIFO (DMAFIFO). This supports high performance DMA transfers. In general, the DMA reads data from the source, and stores it as transactions in the FIFO. On the destination side, the DMA requests ownership of the master and once granted begins transfers. Transfers stop on the source side when the FIFO fills, and on the destination side when the FIFO empties. 6.5.1 PCI-to-VMEbus Transfers PCI-to-VMEbus transfers involve the Universe II reading from the PCI bus and writing to the VMEbus. Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 102 6. DMA Controller > FIFO Operation and Bus Ownership The PCI bus is requested for the current read once 128 bytes are available in the DMAFIFO. The DMA Channel fills the DMAFIFO using PCI read transactions with each transaction broken at address boundaries determined by the programmed PCI aligned burst size (PABS field in the “Master Control Register (MAST_CTL)” on page 271). This ensures that the DMA makes optimal use of the PCI bus by always generating bursts of 32, 64 or 128 bytes with zero wait states. The DMA packs read data into the DMAFIFO to the full 64-bit width of the FIFO, independent of the width of the PCI bus, or the data width of the ensuing VMEbus transaction. The PCI read transactions continue until either the DMA has completed the full programmed transfer, or there is insufficient room available in the DMAFIFO for a full transaction. The available space required for another burst read transaction is again 128 bytes. Since the VMEbus is typically much slower than the PCI bus, the DMAFIFO can fill frequently during PCI to VMEbus transfers, though the depth of the FIFO helps to minimize this situation. When the DMAFIFO fills, the PCI bus is free for other transactions (for example, between other devices on the bus or possibly for use by the Universe II’s VMEbus Slave Channel). The DMA only resumes read transactions on the PCI bus when the DMAFIFO has space for another aligned burst size transaction. The DMA can prefetch extra read data from the external PCI target. This means that the DMA must only be used with memory on the PCI bus which has no adverse side-effects when prefetched. The Universe II prefetches up to the aligned address boundary defined in the PABS field of the MASC_CTL register. On the VMEbus, the actual programmed number of bytes in the “DMA Transfer Byte Count Register (DTBC)” on page 234 are written. Prefetching can be avoided by programming the DMA for transfers that terminate at the PABS boundary. If further data is required beyond the boundary, but before the next boundary, the DTBC register may be programmed to eight byte transfers. The DMA fetches the full eight bytes, and nothing more. Programming the DTBC to less than eight bytes still results in eight bytes fetched from PCI. The DMA requests ownership of the Universe II's VMEbus Master Interface once 64 bytes of data have been queued in the DMAFIFO (see “VMEbus Requester” on page 25 on how the VMEbus Master Interface is shared between the DMA, the PCI Target Channel, and the Interrupt Channel). The Universe II maintains ownership of the Master Interface until one of the following conditions are met: • DMAFIFO is empty • DMA block is complete • DMA is stopped • a linked list is halted • DMA encounters an error • DMA VMEbus tenure limit (VON in the DGCS register) Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 6. DMA Controller > FIFO Operation and Bus Ownership 103 The DMA can be programmed to limit its VMEbus tenure to fixed block sizes using the VON field in the “DMA General Control/Status Register (DGCS)” on page 238. With VON enabled, the DMA relinquishes ownership of the Master Interface at defined address boundaries. See “DMA VMEbus Ownership” on page 91. To further control the DMA’s VMEbus ownership, the VOFF timer in the DGCS register can be used to program the DMA to remain off the VMEbus for a specified period when VMEbus tenure is relinquished. See “DMA VMEbus Ownership” on page 91. The DMA Channel unpacks the 64-bit data queued in the DMAFIFO to whatever the programmed transfer width is on the VMEbus (D16, D32, or D64). The VMEbus Master Interface delivers the data in the DMAFIFO according to the VMEbus cycle type programmed into the “DMA Transfer Control Register (DCTL)” on page 232. The DMA provides data to the VMEbus until one of the following conditions are met: • DMA FIFO empties • DMA VMEbus Tenure Byte Count expires. — Set in the VON in the “DMA General Control/Status Register (DGCS)” on page 238 If the DMAFIFO empties transfers on the VMEbus stop and, if the cycle being generated is a block transfer, then the block is terminated (AS* negated). The VMEbus ownership is relinquished by the DMA. The DMA does not re-request VMEbus ownership until another eight entries are queued in the DMAFIFO, or the DMA Channel has completed the current Transfer Block on the PCI bus (see “VMEbus Release” on page 27). PCI bus transactions are the full width of the PCI data bus with appropriate byte lanes enabled. The maximum VMEbus data width is programmable to 8, 16, 32, or 64-bit. Byte transfers can be only of type DO8 (EO). Because the PCI bus has a more flexible byte lane enabling scheme than the VMEbus, the Universe II can be required to generate a variety of VMEbus transaction types to handle the byte resolution of the starting and ending addresses (see “Data Transfer” on page 52). 6.5.2 VMEbus-to-PCI Transfers VMEbus-to-PCI transfers involve the Universe II reading from the VMEbus and writing to the PCI bus. With DMA transfers in this direction, the DMA Channel begins to queue data in the DMAFIFO as soon as there is room for 64 bytes in the DMAFIFO. When this watermark is reached, the DMA requests the VMEbus (through the VMEbus Master Interface) and begins reading data from the VMEbus. The Universe II maintains VMEbus ownership until one of the following conditions are met: • DMAFIFO is full • DMA block is complete • DMA is stopped • a linked list is halted • DMA encounters an error • VMEbus tenure limit is reached (VON in the DGCS register) Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 104 6. DMA Controller > DMA Interrupts The DMA can be programmed to limit its VMEbus tenure to fixed block sizes using the VON field in the “DMA General Control/Status Register (DGCS)” on page 238. With VON enabled, the DMA will relinquish ownership of the Master Interface at defined address boundaries (see “DMA VMEbus Ownership” on page 91). To further control the DMA’s VMEbus ownership, the VOFF timer in the DGCS register can be used to program the DMA to remain off the VMEbus for a specified period when VMEbus tenure is relinquished. See “DMA VMEbus Ownership” on page 91. Entries in the DMAFIFO are delivered to the PCI bus as PCI write transactions as soon as there are 128 bytes available in the DMAFIFO. If the PCI bus responds too slowly, the DMAFIFO runs the risk of filling before write transactions can begin at the PCI Master Interface. Once the DMAFIFO reaches a nearly full state (three entries remaining) the DMA requests that the VMEbus Master Interface complete its pending operations and stop. The pending read operations fill the DMAFIFO. Once the pending VMEbus reads are completed (or the VON timer expires), the DMA relinquishes VMEbus ownership and only re-requests the VMEbus Master Interface once 64 bytes again become available in the DMAFIFO. If the bus was released due to encountering a VON boundary, the bus is not re-requested until the VOFF timer expires. PCI bus transactions are the full width of the PCI data bus with appropriate byte lanes enabled. The maximum VMEbus data width is programmable to 8, 16, 32, or 64 bits. Byte transfers can be only of type DO8 (EO). Because the PCI bus has a more flexible byte lane enabling scheme than the VMEbus, the Universe II can be required to generate a variety of VMEbus transaction types to handle the byte resolution of the starting and ending addresses (see “Universe II as PCI Target” on page 57). 6.6 DMA Interrupts The Interrupt Channel in the Universe II handles a single interrupt sourced from the DMA Channel which it routes to either the VMEbus or PCI bus through the DMA bits in the LINT_EN and VINT_EN registers. There are six internal DMA sources of interrupts and these are all routed to this single interrupt. Each of these six sources can be individually enabled, and are listed in Table 15. Setting the enable bit enables the corresponding interrupt source. Table 15: DMA Interrupt Sources and Enable Bits Interrupt Source Enable Bit Stop Request INT_STOP Halt Request INT_HALT DMA Completion INT_DONE PCI Target-Abort or Master-Abort INT_LERR VMEbus Error INT_VERR Protocol Error INT_M_ERR Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 6. DMA Controller > DMA Channel Interactions with Other Channels 105 Once an enabled DMA interrupt has occurred the corresponding DMA bit in the “PCI Interrupt Status Register (LINT_STAT)” on page 244 and “VMEbus Interrupt Status Register (VINT_STAT)” on page 251 are set — regardless of whether the LINT_EN or VINT_EN enable bits have been set. Each one must be cleared independently. Clearing either the LINT_STAT or VINT_STAT registers does not clear the other. See “Interrupt Generation and Handling” on page 109. 6.7 DMA Channel Interactions with Other Channels This section describes the impact that the PCI Bus Target Channel and the VMEbus Slave Channel can have on the DMA Channel. The Universe II does not apply PCI 2.1 Specification transaction ordering requirements to the DMA Controller. Reads and writes through the DMA Controller can occur independently of the other channels. ADOH cycles and RMW cycles through the VMEbus Slave Channel do impact on the DMA Channel. Once an external VMEbus master locks the PCI bus, the DMA Controller does not perform transfers on the PCI bus until the Universe II is unlocked (see “VMEbus Lock Commands (ADOH Cycles)” on page 36). When an external VMEbus Master begins a RMW cycle, at some point a read cycle appears on the PCI bus. During the time between when the read cycle occurs on the PCI bus and when the associated write cycle occurs on the PCI bus, no DMA transfers occurs on the PCI bus (see “VMEbus Read-Modify-Write Cycles (RMW Cycles)” on page 37). If the PCI Target Channel locks the VMEbus using VOWN, no DMA transfers takes place on the VMEbus (see “Using the VOWN bit” on page 64). 6.8 DMA Error Handling This section describes how the Universe II responds to errors involving the DMA, and how the user can recover from them. The software source of a DMA error is a protocol, and the hardware source of a DMA error is a VMEbus error, or PCI bus Target-Abort or Master-Abort. 6.8.1 DMA Software Response to Error While the DMA is operating normally, the ACT bit in the “DMA General Control/Status Register (DGCS)” on page 238. Once the DMA has terminated, it clears this bit, and sets one of six status bits in the same register. The DONE bit will be set if the DMA completed all its programmed operations normally. If the DMA is interrupted, either the STOP or HALT bits are set. If an error has occurred, one of the remaining three bits, LERR, VERR, or P_ERR, is set. All six forms of DMA terminations can be optionally set to generate a DMA interrupt by setting the appropriate enable bit in the DGCS register (see “DMA Interrupts” on page 104). • LERR is set if the DMA encounters an error on the PCI bus (either a Master-Abort or Target-Abort) — Bits in the “PCI Configuration Space Control and Status Register (PCI_CSR)” on page 172 indicate which of these conditions caused the error. Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 106 6. DMA Controller > DMA Error Handling • VERR is set if the DMA encounters a bus error on the VMEbus. This is through a detected assertion of BERR* during a DMA cycle. • P_ERR is set if the GO bit in the DGCS register is set to start the DMA, and the DMA has been improperly programmed either because the BM bit in the PCI_CSR disables PCI bus mastership, or the source and destination start addresses are not aligned (see “Source and Destination Addresses” on page 86). Whether the error occurs on the destination or source bus, the DMA_CTL register contains the attributes relevant to the particular DMA transaction. The DTBC register provides the number of bytes remaining to transfer on the PCI side. The DTBC register contains valid values after an error. The DLA and DVA registers should not be used for error recovery. 6.8.2 DMA Hardware Response to Error When the error condition (VMEbus Error, Target-Abort, or Master-Abort) occurs on the source bus while the DMA is reading from the source bus, the DMA stops reading from the source bus. Any data previously queued within the DMAFIFO is written to the destination bus. Once the DMAFIFO empties, the error status bit is set and the DMA generates an interrupt (if enabled by INT_LERR or INT_VERR in the DGCS register—see “DMA Interrupts” on page 104). When the error condition (VMEbus Error, Target-Abort, or Master-Abort) occurs on the destination bus while the DMA is writing data to the destination bus, the DMA stops writing to the destination bus, and it also stops reading from the source bus. The error bit in the DGCS register is set and an interrupt asserted (if enabled). 6.8.2.1 Interrupt Generation During Bus Errors To generate an interrupt from a DMA error, there are two bits in the DGCS register, and one bit each in the VINT_EN and LINT_EN registers. In the DGCS register the INT_LERR bit enables the DMA to generate an interrupt to the Interrupt Channel after encountering an error on the PCI bus. The INT_VERR enables the DMA to generate an interrupt to the Interrupt Channel upon encountering an error on the VMEbus. Upon reaching the Interrupt Channel, all DMA interrupts can be routed to either the PCI bus or VMEbus by setting the appropriate bit in the enable registers. All DMA sources of interrupts (Done, Stopped, Halted, VMEbus Error, and PCI Error) constitute a single interrupt into the Interrupt Channel. 6.8.3 Resuming DMA Transfers When a DMA error occurs (on the source or destination bus), the status bits must be read in order to determine the source of the error. If it is possible to resume the transfer, the transfer should be resumed at the address that was in place up to 256 bytes from the current byte count. The original addresses (DLA and DVA) are required in order to resume the transfer at the appropriate location. However, the values in the DLA and the DVA registers should not be used to reprogram the DMA, because they are not valid once the DMA begins. In direct mode, it is the user’s responsibility to record the original state of the DVA and DLA registers for error recovery. In Linked-List mode, the user can refer to the current Command Packet stored on the PCI bus (whose location is specified by the DCPP register) for the location of the DVA and DLA information. Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 6. DMA Controller > DMA Error Handling 107 The DTBC register contains the number of bytes remaining to transfer on the source side. The Universe II does not store a count of bytes to transfer on the destination side. If the error occurred on the source side, then the location of the error is simply the latest source address plus the byte count. If the error occurred on the destination side, then one cannot infer specifically where the error occurred, because the byte count only refers to the number of data queued from the source, not what has been written to the destination. In this case, the error will have occurred up to 256 bytes before: the original address plus the byte count. Given this background, the following procedure can be implemented to recover from errors. 1. Read the value contained in the DTBC register. 2. Read the record of the DVA and DLA that is stored on the PCI bus or elsewhere (not the value stored in the Universe II registers of the same name). 3. If the difference between the value contained in the DTBC register and the original value is less than 512 bytes (the FIFO depth of the Universe II), reprogram all the DMA registers with their original values. 4. If the difference between the value contained in the DTBC register and the original value is greater than 512 bytes (the FIFO depth of the Universe II), add 512 bytes to the value contained in the DTBC register. 5. Add the difference between the original value in the DTBC and the new value in the DTBC register to the original value in the DLA register. 6. Add the difference between the original value in the DTBC and the new value in the DTBC register to the original value in the DVA register. 7. Clear the status flags. 8. Restart the DMA (see “DMA Initiation” on page 90). Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 108 Universe II User Manual May 12, 2010 6. DMA Controller > DMA Error Handling Integrated Device Technology www.idt.com 109 7. Interrupt Generation and Handling An interrupt is a signal informing a program that an event (for example, an error) has occurred. When a program receives an interrupt signal, it temporarily suspends normal processing and diverts the execution of instructions to a sub-routine handled by an interrupt controller. The controller communicates with the host processor and the device that initiated the interrupt to determine how to handle the interrupt. Interrupt signals can come from a variety of sources. Interrupt signals generated by devices (for example, a printer) indicate an event has occurred and are called hardware interrupts. Interrupt signals generated by programs are called software interrupts. This chapter discussed the interrupt generation and handling functionality of the Universe II. This chapter discusses the following topics: 7.1 • “Interrupt Generation” on page 111 • “Interrupt Handling” on page 116 Overview The Universe II has two types of interrupt capability: it is a generator of interrupts and an interrupt handler. The Interrupt Channel handles the prioritization and routing of interrupt sources to interrupt outputs on the PCI bus and VMEbus. The interrupt sources are: • PCI LINT_[7:0] • VMEbus IRQ*[7:1] • ACFAIL* and SYSFAIL* • Various internal events These sources can be routed to either the PCI LINT_ [7:0] lines or the VMEbus IRQ* [7:1] lines. Each interrupt source is individually maskable and can be mapped to various interrupt outputs. Most interrupt sources can be mapped to one particular destination bus. The PCI sources, LINT_[7:0], can only be mapped to the VMEbus interrupt outputs. The VMEbus sources, VIRQ[7:1], are not mapped to the PCI bus interrupts. However, the VIRQ[7:1] bits are status bits which indicate whether or not a STATUS/ID vector has been acquired. This indication can be used to generate an interrupt on the PCI bus. Some internal sources (for example, error conditions or DMA activity) can be mapped to either bus. Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 110 7. Interrupt Generation and Handling > Overview Figure 18: Universe Interrupt Circuitry VRACFAIL_ VRSYSFAIL_ Mapping and Enabling LINT [7:0] Internal Interrupt Handler VRIQ[7:1]_ Internal Sources Mapping VXIRQ[7:1]_ and Enabling Figure 18 illustrates the circuitry inside the Universe II Interrupt Channel. The PCI hardware interrupts are listed on the left, and the VMEbus interrupt inputs and outputs are on the right. Internal interrupts are also illustrated. The figure shows that the interrupt sources may be mapped and enabled. The Internal Interrupt Handler is a block within the Universe II that detects assertion of the VRIRQ_[7:1] pins and generates the VME IACK through the VME Master. Upon completion of the IACK cycle, the Internal Interrupt Handler notifies the Mapping Block which in turn asserts the local LINT_, if enabled. (Whereas the Internal Interrupt Handler implies a delay between assertion of an interrupt condition to the Universe II and the Universe’s mapping of the interrupt, all other interrupt sources get mapped immediately to their destination—assertion of LINT_ immediately causes an IRQ, assertion of ACFAIL immediately causes an LINT_, etc.) Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 7. Interrupt Generation and Handling > Interrupt Generation 7.2 111 Interrupt Generation The Universe II has the ability to generate interrupts on both the PCI bus and VMEbus. 7.2.1 PCI Interrupt Generation The Universe II expands on the basic PCI specification which permits “single function” devices to assert only a single interrupt line. Eight PCI interrupt outputs provide maximum flexibility, although if full PCI compliancy is required, all interrupt sources can be routed to a single PCI interrupt output. PCI interrupts may be generated from multiple sources: • VMEbus sources of PCI interrupts — IRQ*[7:1] — SYSFAIL* — ACFAIL* – internal sources of PCI interrupts — DMA — VMEbus bus error encountered — PCI Target-Abort or Master-Abort encountered — VMEbus ownership has been granted while the VOWN bit is set (see “VME Lock Cycles—Exclusive Access to VMEbus Resources” on page 63) — Software interrupt — Mailbox access — Location monitor access — VMEbus IACK cycle performed in response to a software interrupt Each source can be individually enabled in the “PCI Interrupt Enable Register (LINT_EN)” on page 242 and mapped to a single LINT_ signal through the “PCI Interrupt Map 0 Register (LINT_MAP0)” on page 246, “PCI Interrupt Map 1 Register (LINT_MAP1)” on page 247, and “PCI Interrupt Map 2 Register (LINT_MAP2)” on page 263. When an interrupt is received on any of the enabled sources, the Universe II asserts the appropriate LINT_ pin and sets a matching bit in the “PCI Interrupt Status Register (LINT_STAT)” on page 244. See Table 17 on page 114 for a list of the enable, mapping and status bits for PCI interrupt sources. Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 112 7. Interrupt Generation and Handling > Interrupt Generation Table 16: Source, Enabling, Mapping, and Status of PCI Interrupt Output Enable Bit in LINT_EN Interrupt Source Register Mapping Field in LINT_MAPx Register Status Bit in LINT_STAT Register ACFAIL* ACFAIL ACFAIL ACFAIL SYSFAIL* SYSFAIL SYSFAIL SYSFAIL PCI Software Interrupt SW_INT SW_INT SW_INT VMEbus Software IACK SW_IACK SW_IACK SW_IACK VMEbus Error occurred during a posted write VERR VERR VERR PCI Target-Abort or Master-Abort occurred during a posted write LERR LERR LERR DMA Event DMA DMA DMA VMEbus Interrupt Input VIRQ7-1 VIRQ7-1 VIRQ7-1 Location Monitor LM3-0 LM3-0 LM3-0 Mailbox Access MBOX3-0 MBOX3-0 MBOX3-0 VMEbus Ownership VOWN VOWN VOWN The “PCI Interrupt Status Register (LINT_STAT)” on page 244 shows the status of all sources of PCI interrupts, independent of whether that source has been enabled. This implies that an interrupt handling routine must mask out those bits in the register that do not correspond to enabled sources on the active LINT_ pin. Except for SYSFAIL* and ACFAIL*, all sources of PCI interrupts are edge-sensitive. Enabling of the ACFAIL* or SYSFAIL* sources (ACFAIL and SYSFAIL bits in the LINT_EN register) causes the status bit and mapped PCI interrupt pin to assert synchronously with the assertion of the ACFAIL* or SYSFAIL* source. The PCI interrupt is negated once the ACFAIL or SYSFAIL status bit is cleared. The status bit cannot be cleared if the source is still active. Therefore, if SYSFAIL* or ACFAIL* is still asserted while the interrupt is enabled the interrupt will continue to be asserted. Both of these sources are synchronized and filtered with multiple edges of the PCI clock at their inputs. Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 7. Interrupt Generation and Handling > Interrupt Generation 113 All other sources of PCI interrupts are edge-sensitive. The VMEbus source for PCI interrupts actually comes out of the VMEbus Interrupt Handler block and reflects acquisition of a VMEbus STATUS/ID. Therefore, even though VMEbus interrupts externally are level-sensitive as required by the VMEbus Specification, they are internally mapped to edge-sensitive interrupts (see “VMEbus Interrupt Handling” on page 116). The interrupt source status bit (in the LINT_STAT register) and the mapped LINT_ pin remain asserted with all interrupts. The status bit and the PCI interrupt output pin are only released when the interrupt is cleared by writing a 1 to the appropriate status bit. 7.2.2 VMEbus Interrupt Generation This section details the conditions under which the Universe II generates interrupts to the VMEbus. Interrupts may be generated on any combination of VMEbus interrupt lines (IRQ*[7:1]) from multiple sources: • PCI sources of VMEbus interrupts — LINT_[7:0] • Internal sources of VMEbus interrupts — DMA — VMEbus bus error encountered — PCI Target-Abort or Master-Abort encountered — Mailbox register access — Software interrupt Each of these sources may be individually enabled through the “VMEbus Interrupt Enable Register (VINT_EN)” on page 248 and mapped to a particular VMEbus Interrupt level using the “VME Interrupt Map 0 Register (VINT_MAP0)” on page 253, “VME Interrupt Map 1 Register (VINT_MAP1)” on page 254, and “VME Interrupt Map 2 Register (VINT_MAP2)” on page 264. Multiple sources may be mapped to any VMEbus level. Mapping interrupt sources to level 0 effectively disables the interrupt. Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 114 7. Interrupt Generation and Handling > Interrupt Generation Once an interrupt has been received from any of the sources, the Universe II sets the corresponding status bit in the“VMEbus Interrupt Status Register (VINT_STAT)” on page 251, and asserts the appropriate VMEbus interrupt output signal (if enabled). When a VMEbus interrupt handler receives the interrupt, it will perform an IACK cycle at that interrupt level. When the Universe II decodes that IACK cycle together with IACKIN* asserted, it provides the STATUS/ID previously stored in the “Interrupt STATUS/ID Out Register (STATID)” on page 255, unless it is configured as SYSCON in which case it does not monitor IACKIN*. See Table 17 for a list of the enable, mapping and status bits for VMEbus interrupt sources. Table 17: Source, Enabling, Mapping, and Status of VMEbus Interrupt Outputs Interrupt Source Enable Bit in the “VMEbus Interrupt Enable Register (VINT_EN)” on page 248 VINT_MAPx Status Bit in the “VMEbus Interrupt Status Register (VINT_STAT)” on page 251 Mapping Field in VMEbus Software Interrupt SW_INT7-1 N/Aa SW_INT7-1 VMEbus Error VERR VERR in the “VME Interrupt Map 1 Register (VINT_MAP1)” on page 254 VERR PCI Target-Abort or Master-Abort LERR LERR in the “VME Interrupt Map 1 Register (VINT_MAP1)” on page 254 LERR DMA Event DMA DMA in the “VME Interrupt Map 1 Register (VINT_MAP1)” on page 254 DMA Mailbox Register MBOX3-0 MBOX3-0 in the “VME Interrupt Map 2 Register (VINT_MAP2)” on page 264 MBOX3-0 PCI bus Interrupt Input LINT7-0 LINT7-0 in the “VME Interrupt Map 0 Register (VINT_MAP0)” on page 253 LINT7-0 VMEbus Software Interrupt SW_INT SW_INT in the “VME Interrupt Map 1 Register (VINT_MAP1)” on page 254 SW_INT (mappable) a. This set of software interrupts cannot be mapped. That is, setting the SW_INT1 bit triggers VXIRQ1, setting the SW_INT2 bit triggers VXIRQ2, etc. For all VMEbus interrupts, the Universe II interrupter supplies a pre-programmed 8-bit STATUS/ID; a common value for all interrupt levels. The upper seven bits are programmed in the STATID register. The lowest bit is cleared if the source of the interrupt was the software interrupt, and is set for all other interrupt sources. If a software interrupt source and another interrupt source are active and mapped to the same VMEbus interrupt level, the Universe II gives priority to the software source. Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 7. Interrupt Generation and Handling > Interrupt Generation 115 Figure 19: STATUS/ID Provided by Universe II STATUS/ID Programmed from VME_STATUS/ID Registers 0 if S/W Interrupt Source 1 if Internal or LINT Interrupt Source Once the Universe II has provided the STATUS/ID to an interrupt handler during a software initiated VMEbus interrupt, it generates an internal interrupt, SW_IACK. If enabled, this interrupt feeds back to the PCI bus (through one of the LINT_ pins) to signal a process that the interrupt started through software has been completed. All VMEbus interrupts generated by the Universe II are RORA, except for the software interrupts which are ROAK. This means that if the interrupt source was a software interrupt, then the VMEbus interrupt output is automatically negated when the Universe II receives the IACK cycle. However, for any other interrupt, the VMEbus interrupt output remains asserted until cleared by a register access. Writing 1 to the relevant bit in the VINT_STAT register clears that interrupt source. However, since PCI interrupts are level-sensitive, if an attempt is made to clear the VMEbus interrupt while the LINT_ pin is still asserted, the VMEbus interrupt remains asserted. For this reason, a VMEbus interrupt handler should clear the source of the PCI interrupt before clearing the VMEbus interrupt. Since software interrupts are ROAK, the respective bits in the VINT_STAT register are cleared automatically on completion of the IACK cycle, simultaneously with the negation of the IRQ. Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 116 7.3 7. Interrupt Generation and Handling > Interrupt Handling Interrupt Handling The Universe II can handle interrupts from both the PCI bus and the VMEbus. 7.3.1 PCI Interrupt Handling All eight PCI interrupt lines, LINT_[7:0], can act as interrupt inputs to the Universe II. They are level-sensitive and, if enabled in the “VMEbus Interrupt Enable Register (VINT_EN)” on page 248, immediately generate an interrupt to the VMEbus. It is expected that when a VMEbus interrupt handler receives the Universe II’s STATUS/ID from the Universe II, the interrupt handler clears the VMEbus interrupt by first clearing the source of the interrupt on the PCI bus, and then clearing the VMEbus interrupt (by writing a 1 to the appropriate bit in the “VMEbus Interrupt Status Register (VINT_STAT)” on page 251. Note that since PCI interrupts are level-sensitive, if an attempt is made to clear the VMEbus interrupt while the LINT_ pin is still asserted, the VMEbus interrupt remains asserted. This causes a second interrupt to be generated to the VMEbus. For this reason, a VMEbus interrupt handler should clear the source of the PCI interrupt before clearing the VMEbus interrupt. 7.3.2 VMEbus Interrupt Handling As a VMEbus interrupt handler, the Universe II can monitor any or all of the VMEbus interrupt levels. It can also monitor SYSFAIL* and ACFAIL*, although IACK cycles are not generated for these inputs. Each interrupt is enabled through the “PCI Interrupt Enable Register (LINT_EN)” on page 242. Once enabled, assertion of any of the VMEbus interrupt levels, IRQ[7:1]*, causes the internal interrupt handler circuitry to request ownership of the Universe II's VMEbus Master Interface on the level programmed in the “Master Control Register (MAST_CTL)” on page 271 (see “VMEbus Requester” on page 25). This interface is shared between several channels in the Universe II: the PCI Target Channel, the DMA Channel, and the Interrupt Channel. The Interrupt Channel has the highest priority over all other channels and, if an interrupt is pending, assumes ownership of the VMEbus Master Interface when the previous owner has relinquished ownership. The Universe II latches the first interrupt that appears on the VMEbus and begins to process it immediately. If an interrupt at a higher priority is asserted on the VMEbus before BBSY* is asserted the Universe II performs an interrupt acknowledge for the first interrupt it detected. Upon completion of that IACK cycle, the Universe II then performs IACK cycles for the higher of any remaining active interrupts. There may be some latency between reception of a VMEbus interrupt and generation of the IACK cycle. This arises because of the latency involved in the Interrupt Channel gaining control of the VMEbus Master Interface, and because of possible latency in gaining ownership of the VMEbus if the VMEbus Master Interface is programmed for release-when-done. In addition, the Universe II only generates an interrupt on the PCI bus once the IACK cycle has completed on the VMEbus. Because of these combined latencies (time to acquire VMEbus and time to run the IACK cycle), systems should be designed to accommodate a certain worst case latency from VMEbus interrupt generation to its translation to the PCI bus. Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 7. Interrupt Generation and Handling > Interrupt Handling 117 When the Universe II receives a STATUS/ID in response to an IACK cycle, it stores that value in one of seven registers. These registers, “VIRQ1 STATUS/ID Register (V1_STATID)” on page 256 through “VIRQ7 STATUS/ID Register (V7_STATID)” on page 262, store the STATUS/ID corresponding to each IACK level in the STATID field. Once an IACK cycle has been generated and the resulting STATUS/ID is latched, another IACK cycle is not run on that level until the level has been cleared by writing a 1 to the corresponding status bit in the “PCI Interrupt Status Register (LINT_STAT)” on page 244. If other interrupts (at different levels) are pending while the interrupt is waiting to be cleared, IACK cycles are run on those levels in order of priority and the STATUS/IDs stored in their respective registers. Once the IACK cycle is complete and the STATUS/ID stored, an interrupt is generated to the PCI bus on one of LINT_[7:0] depending on the mapping for that VMEbus level in the “PCI Interrupt Map 0 Register (LINT_MAP0)” on page 246. The interrupt is cleared and the VMEbus interrupt level is re-armed by clearing the correct bit in the “PCI Interrupt Status Register (LINT_STAT)” on page 244 register. 7.3.2.1 Bus Error During VMEbus IACK Cycle A bus error encountered on the VMEbus while the Universe II is performing an IACK cycle is handled by the Universe II in two ways. The first is through the error logs in the VMEbus Master Interface. These logs store address and command information whenever the Universe II encounters a bus error on the VMEbus (see “Error Handling” on page 125). If the error occurs during an IACK cycle, the IACK_ bit is set in the “VMEbus AM Code Error Log (V_AMERR)” on page 307. The VMEbus Master Interface also generates an internal interrupt to the Interrupt Channel indicating a VMEbus error occurred. This internal interrupt can be enabled and mapped to either the VMEbus or PCI bus. As well as generating an interrupt indicating an error during the IACK cycle, the Universe II also generates an interrupt as though the IACK cycle completed successfully. If an error occurs during the fetching of the STATUS/ID, the Universe II sets the ERR bit in the “Interrupt STATUS/ID Out Register (STATID)” on page 255, and generates an interrupt on the appropriate LINT_ pin (as mapped in the “PCI Interrupt Map 0 Register (LINT_MAP0)” on page 246). The PCI resource, upon receiving the PCI interrupt, is expected to read the“Interrupt STATUS/ID Out Register (STATID)” on page 255, and take appropriate actions if the ERR bit is set. Note that the STATUS/ID cannot be considered valid if the ERR bit is set in the STATUS/ID register. It is important to recognize that the IACK cycle error can generate two PCI interrupts: one through the VMEbus master bus error interrupt and another through the standard PCI interrupt translation. If an error occur during acquisition of a STATUS/ID, the “PCI Interrupt Status Register (LINT_STAT)” on page 244 shows that both VIRQx, and VERR are active. Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 118 7.3.3 7. Interrupt Generation and Handling > Interrupt Handling Internal Interrupt Handling The Universe II's internal interrupts are routed from several processes in the device. There is an interrupt from the VMEbus Master Interface to indicate a VMEbus error, another from the PCI Master Interface to indicate an error on that bus, another from the DMA to indicate various conditions in that channel, along with several others. Table 18 shows to which bus each interrupt source can be routed. Some sources can be mapped to both buses, but mapping interrupts to a single bus is recommended. Table 18: Internal Interrupt Routing May be Routed to: Interrupt Source VMEbus  PCI s/w interrupt VMEbus s/w interrupt PCI Bus   IACK cycle complete for s/w interrupt DMA event   Mailbox access    Location monitor PCI Target-Abort or Master-Abort   VMEbus bus error   VMEbus bus ownership granted Universe II User Manual May 12, 2010  Integrated Device Technology www.idt.com 7. Interrupt Generation and Handling > Interrupt Handling 119 Figure 20 shows the sources of interrupts, and the interfaces from which they originate. Figure 20: Sources of Internal Interrupts PCI Bus Interface VMEbus Interface VMEbus Slave Channel PCI Master posted writes FIFO VME Slave prefetch read FIFO coupled path DMA Channel DMA bidirectional FIFO PCI Bus Slave Channel posted writes FIFO PCI Target coupled read logic VME Master Interrupt Channel Interrupt Handler DMA PCI error PCI software interrupt Integrated Device Technology www.idt.com VME error VME ownership bit software IACK VME software interrupt Universe II User Manual May 12, 2010 120 7. Interrupt Generation and Handling > Interrupt Handling 7.3.3.1 VMEbus and PCI Software Interrupts It is possible to interrupt the VMEbus and the PCI bus through software. These interrupts may be triggered by writing a 1 to the respective enable bits. Interrupting the VMEbus Through Software The following methods trigger software interrupts on the VMEbus: 1. The first method for interrupting the VMEbus through software involves writing 1 to one of the SW_INT7-1 bits in the VINT_EN register while the mask bit is 0.1 This causes an interrupt to be generated on the corresponding IRQ7-1 line. For example, setting the SW_INT1 bit triggers VXIRQ1, setting the SW_INT2 bit triggers VXIRQ2, etc. 2. The second method for interrupting the VMEbus through software involves an extra step. Writing a 1 to the SW_INT bit in the“VMEbus Interrupt Enable Register (VINT_EN)” on page 248 when this bit is 0 triggers one interrupt on the VMEbus on the level programmed in the “VME Interrupt Map 1 Register (VINT_MAP1)” on page 254. This method requires that the user specify in the VINT_MAP1 register to which line the interrupt is to be generated. When the SW_INT interrupt (method 2) is active at the same level as one of SW_INT7-1 interrupts (method 1), the SW_INT interrupt (method 2) takes priority.While this interrupt source is active, the SW_INT status bit in the VINT_STAT register is set. This method is provided for compatibility with the original Universe device. With both methods, the mask bit (SW_INTx or SW_INT) in the “VMEbus Interrupt Enable Register (VINT_EN)” on page 248 must be 0 in order for writing 1 to the bit to have any effect. Regardless of the software interrupt method used, when an IACK cycle is serviced on the VMEbus, the Universe II can be programmed to generate an interrupt on the PCI bus by setting the SW_IACK enable bit in the “PCI Interrupt Enable Register (LINT_EN)” on page 242 (see “Software IACK Interrupt” on page 121). Interrupting the PCI bus Through Software On the PCI bus, there is only one method of directly triggering a software interrupt. This method is the same as the second method described in “Interrupting the VMEbus Through Software” on page 120. Causing a 0 to 1 transition in the SW_INT in the LINT_EN register generates an interrupt to the PCI bus. While this interrupt source is active, the SW_INT status bit in LINT_STAT is set. The SW_INT field in the “PCI Interrupt Map 1 Register (LINT_MAP1)” on page 247 determines which interrupt line is asserted on the PCI interface. 1. The term “enable” is more meaningful with respect to the other fields in this register, i.e., excluding the software interrupts. Writing to the software interrupt fields of this register does not enable an interrupt, it triggers an interrupt. Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 7. Interrupt Generation and Handling > Interrupt Handling 121 Termination of Software Interrupts Any software interrupt can be cleared by clearing the respective bit in the “VMEbus Interrupt Enable Register (VINT_EN)” on page 248 or “PCI Interrupt Enable Register (LINT_EN)” on page 242. However, this method is not recommend for VME bus software interrupts because it can result in a false interrupts on that bus. These false interrupts are caused because the Universe II does not respond to the interrupt handler’s IACK cycle, and the handler is left without a STATUS/ID for the interrupt. Since the software interrupt is edge-sensitive, the software interrupt bit in the VINT_EN or LINT_EN register should be cleared any time between the last interrupt finishing and the generation of another interrupt. It is recommended that the appropriate interrupt handler clear this bit once it has completed its operations. Alternatively, the process generating a software interrupt could clear this bit before re-asserting it. Software interrupts on the VMEbus have priority over other interrupts mapped internally to the same level on the VMEbus. When a VMEbus interrupt handler generates an IACK cycle on a level mapped to both a software interrupt and another interrupt, the Universe II always provides the STATUS/ID for the software interrupt (bit zero of the Status/ID is cleared). If there are no other active interrupts on that level, the interrupt is automatically cleared upon completion of the IACK cycle (since software interrupts are ROAK). While the software interrupt STATUS/ID has priority over other interrupt sources, the user can give other interrupt sources priority over the software interrupt. This is done by reading the “PCI Interrupt Status Register (LINT_STAT)” on page 244 when handling a Universe II interrupt. This register indicates all active interrupt sources. Using this information, the interrupt handler can then handle the interrupt sources in any system-defined order. 7.3.3.2 Software IACK Interrupt The Universe II generates an internal interrupt when it provides the software STATUS/ID to the VMEbus. This interrupt can only be routed to a PCI interrupt output. A PCI interrupt is generated upon completion of an IACK cycle that had been initiated by the Universe II’s software interrupt if the following occurs: • The SW_IACK bit is set in the “PCI Interrupt Status Register (LINT_STAT)” on page 244 • The SW_IACK field in the “PCI Interrupt Map 1 Register (LINT_MAP1)” on page 247 is mapped to a corresponding PCI interrupt line This interrupt could be used by a PCI process to indicate that the software interrupt generated to the VMEbus has been received by the device and acknowledged. Like other interrupt sources, this interrupt source can be independently enabled through the “PCI Interrupt Enable Register (LINT_EN)” on page 242 and mapped to a particular LINT_ pin using the LINT_MAP1 register. A status bit in the LINT_STAT register indicates when the interrupt source is active, and is used to clear the interrupt once it has been serviced. Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 122 7.3.3.3 7. Interrupt Generation and Handling > Interrupt Handling VMEbus Ownership Interrupt The VMEbus ownership interrupt is generated when the Universe II acquires the VMEbus in response to programming of the VOWN bit in the MAST_CTL register (“Master Control Register (MAST_CTL)” on page 271). This interrupt source can be used to indicate that ownership of the VMEbus is ensured during an exclusive access (see “VME Lock Cycles—Exclusive Access to VMEbus Resources” on page 63). The interrupt is cleared by writing a one to the matching bit in the “PCI Interrupt Status Register (LINT_STAT)” on page 244. 7.3.3.4 DMA Interrupt The DMA module provides the following possible interrupt sources: • The DMA is stopped (INT_STOP) • The DMA is halted (INT_HALT) • The DMA is done (INT_DONE) • PCI Target-Abort or Master-Abort (INT_LERR) • VMEbus errors (INT_VERR) • A PCI protocol error or if the Universe II is not enabled as PCI master (INT_P_ERR) All of these interrupt sources are ORed to a single DMA interrupt output line. When an interrupt comes from the DMA module, software must read the DMA status bits in the “PCI Interrupt Status Register (LINT_STAT)” on page 244 and the “VMEbus Interrupt Status Register (VINT_STAT)” on page 251 to discover the originating interrupt source. The DMA interrupt can be mapped to either the VMEbus or one of the PCI interrupt output lines (see “DMA Interrupts” on page 104). 7.3.3.5 Mailbox Register Access Interrupts The Universe II can be programmed to generate an interrupt on the PCI bus and/or the VMEbus when any one of its mailbox registers is written to (see “Mailbox Registers” on page 83). The user may enable or disable an interrupt response to the access of any mailbox register. Each register access may be individually mapped to a specific interrupt on the PCI bus through the “PCI Interrupt Map 2 Register (LINT_MAP2)” on page 263 and/or the VMEbus through the “VME Interrupt Map 2 Register (VINT_MAP2)” on page 264. The status of the PCI interrupt and the VMEbus are recorded in the “PCI Interrupt Status Register (LINT_STAT)” on page 244 and “VMEbus Interrupt Status Register (VINT_STAT)” on page 251. 7.3.3.6 Location Monitors The Universe II can be programmed to generate an interrupt on the PCI bus when one of its four location monitors is accessed (see “Location Monitors” on page 38). In order for an incoming VMEbus transaction to activate the location monitor of the Universe II the following criteria must be met: • Location monitor must be enabled Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 7. Interrupt Generation and Handling > Interrupt Handling 123 • Access must be within 4 kbytes of the location monitor base address (see “Location Monitor Base Address Register (LM_BS)” on page 302) • It must be in the specified address space When an access to a location monitor is detected, an interrupt may be generated on the PCI bus (if the location monitor is enabled). There are four location monitors: • VA[4:3] = 00 selects Location Monitor 1, • VA[4:3] = 01 selects Location Monitor 2, • VA[4:3] = 10 selects Location Monitor 3, and • VA[4:3] = 11 selects Location Monitor 4. An interrupt response to the access of any location monitor can be enabled or disabled with bits in the “PCI Interrupt Enable Register (LINT_EN)” on page 242. Access to each location monitor can be individually mapped to a specific interrupt on the PCI bus through the “PCI Interrupt Map 2 Register (LINT_MAP2)” on page 263—not to the VMEbus bus. The status of the PCI interrupt is logged in the LM bit of the “PCI Interrupt Status Register (LINT_STAT)” on page 244. 7.3.3.7 PCI and VMEbus Error Interrupts Interrupts from VMEbus errors, PCI Target-Aborts or Master-Aborts are generated only when bus errors arise during decoupled writes. The bus error interrupt (from either a PCI or VMEbus error) can be mapped to either a VMEbus or PCI interrupt output line. 7.3.4 VME64 Auto-ID The Universe II includes a power-up option for participation in the VME64 Auto-ID process. When this option is enabled, the Universe II generates a level 2 interrupt on the VMEbus before release of SYSFAIL*. When the level 2 IACK cycle is run by the system Monarch, the Universe II responds with the Auto-ID Status/ID, 0xFE, and enables access to a CR/CSR image at base address 0x00_0000. When the Monarch detects an Auto-ID STATUS/ID on level 2, it is expected to access the enabled CR/CSR space of the interrupter. From there it completes identification and configuration of the card. The Monarch functionality is typically implemented in software on one card in the VMEbus system. See “Automatic Slot Identification” on page 42. Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 124 Universe II User Manual May 12, 2010 7. Interrupt Generation and Handling > Interrupt Handling Integrated Device Technology www.idt.com 125 8. Error Handling Errors occur in a system as a result of parity, bus, or internal problems. In order to handle errors so that they have minimum effects on an application, devices have a logic module called an error handler. The error handler logs data about the error then communicates the information to another device (for example, a host processor) that is capable of resolving the error condition. This chapter discusses the following topics: 8.1 • “Errors on Coupled Cycles” on page 125 • “Errors on Decoupled Transactions” on page 125 Overview There are different conditions under which bus errors can occur with the Universe II: during coupled cycles or during decoupled cycles. In a coupled transaction, the completion status is returned to the transaction master, which can then take some action. However, in a decoupled transaction, the master is not involved in the data acknowledgment at the destination bus and higher level protocols are required. The error handling provided by the Universe II is described for both coupled and decoupled transactions. 8.2 Errors on Coupled Cycles During coupled cycles, the Universe II provides immediate indication of an errored cycle to the originating bus. VMEbus to PCI transactions terminated with Target-Abort or Master-Abort are terminated on the VMEbus with BERR*. The R_TA or R_MA bits in the “PCI Configuration Space Control and Status Register (PCI_CSR)” on page 172 are set when the Universe II receives a Target-Abort or Master-Abort. For PCI to VMEbus transactions, a VMEbus BERR* received by the Universe II is communicated to the PCI master as a Target-Abort and the S_TA bit is set in the PCI_CSR register. No information is logged in either direction, nor is an interrupt generated. 8.3 Errors on Decoupled Transactions During decoupled transactions, there is a possibility that an error in the transaction can occur. The following sections detail the decoupled transactions supported by Universe II and the types of error handling supported for these transactions. Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 126 8.3.1 8. Error Handling > Errors on Decoupled Transactions Posted Writes The Universe II provides the option of performing posted writes in both the PCI Target Channel and the VMEbus Slave Channel. Once data is written into the RXFIFO or TXFIFO by the initiating master (VMEbus or PCI bus respectively), the Universe II provides immediate acknowledgment of the cycle's termination. When the data in the FIFO is written to the destination slave or target by the Universe II, the Universe II can receive a bus error instead of a normal termination. The Universe II handles this situation by logging the errored transactions in one of two error logs and generating an interrupt. Each error log (one for VMEbus errors and one for PCI bus errors) is comprised of two registers: one for address and one for command or address space logging. 8.3.1.1 Error Logs If the error occurs during a posted write to the VMEbus, the Universe II uses the “VMEbus AM Code Error Log (V_AMERR)” on page 307 to log the AM code of the transaction in the AMERR field. The state of the IACK* signal is logged in the IACK bit, to indicate whether the error occurred during an IACK cycle. The address of the errored transaction is latched in the V_AERR register (Table 12.3.108). An interrupt is generated on the VMEbus and/or PCI bus depending upon whether the VERR interrupts are enabled (see “Interrupt Generation and Handling” on page 109). The remaining entries of the transaction are removed from the FIFO. If the error occurs during a posted write to the PCI bus, the Universe II uses the “PCI Command Error Log Register (L_CMDERR)” on page 210 to log the command information for the transaction (CMDERR [3:0]). The address of the errored transaction is latched in the “PCI Address Error Log (LAERR)” on page 211. An interrupt is generated on the VMEbus and/or PCI bus depending upon whether the VERR and LERR interrupts are enabled (see “Interrupt Generation and Handling” on page 109). Under either of the conditions (VMEbus-to-PCI, or PCI-to-VMEbus), the address that is stored in the log represents the most recent address the Universe II generated before the bus error was encountered. For single cycle transactions, the address represents the address for the actual errored transaction. However, for multi-data beat transactions (block transfers on the VMEbus or burst transactions on the PCI bus) the log only indicates that an error occurred somewhere after the latched address. For a VMEbus block transfer, the logged address will represent the start of the block transfer. In the PCI Target Channel, the Universe II generates block transfers that do not cross 256-byte boundaries, the error will have occurred from the logged address up to the next 256-byte boundary. In the VMEbus Slave Channel, the error will have occurred anywhere from the logged address up to the next burst aligned address. In the case of PCI-initiated transactions, all data from the errored address up to the end of the initiating transaction is flushed from the TXFIFO. Since the Universe II breaks PCI transactions at 256-byte boundaries (or earlier if the TXFIFO is full), the data is not flushed past this point. If the PCI master is generating bursts that do not cross the 256-byte boundary, then (again) only data up to the end of that transaction is flushed. Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 8. Error Handling > Errors on Decoupled Transactions 127 In a posted write from the VMEbus, all data subsequent to the error in the transaction is flushed from the RXFIFO. However, the length of a VMEbus transaction differs from the length of the errored PCI bus transaction. For non-block transfers, the length always corresponds to one so only the errored data beat is flushed. However, if an error occurs on the PCI bus during a transaction initiated by a VMEbus block transfer, all data subsequent to the errored data beat in the block transfer is flushed from the RXFIFO. In the case of BLTs, this implies that potentially all data up to the next 256-byte boundary may be flushed. For MBLTs, all data up to the next 2-KByte boundary may be flushed. Once an error is captured in a log, that set of registers is frozen against further errors until the error is acknowledged. The log is acknowledged and made available to latch another error by clearing the corresponding status bit in the VINT_STAT or LINT_STAT registers. if a second error occur before the CPU has the opportunity to acknowledge the first error, another bit in the logs is set to indicate this situation (M_ERR bit). 8.3.2 Prefetched Reads In response to a block read from the VMEbus, the Universe II initiates prefetching on the PCI bus (if the VMEbus slave image is programmed with this option, see “VME Slave Image Programming” on page 67). The transaction generated on the PCI bus is an aligned memory read transaction with multiple data beats extending to the aligned burst boundary (as programmed by PABS in the “Master Control Register (MAST_CTL)” on page 271). Once an acknowledgment is given for the first data beat, an acknowledgment is sent to the VMEbus initiator by the assertion of DTACK*. Therefore, the first data beat of a prefetched read is coupled while all subsequent reads in the transaction are decoupled. If an error occurs on the PCI bus, the Universe II does not translate the error condition into a BERR* on the VMEbus. Indeed, the Universe II does not directly map the error. By doing nothing, the Universe II forces the external VMEbus error timer to expire. 8.3.3 DMA Errors The Universe II’s response to a bus error during a transfer controlled by the DMA Channel is described in “DMA Error Handling” on page 105. 8.3.4 Parity Errors The Universe II both monitors and generates parity information using the PAR signal. The Universe II monitors PAR when it accepts data as a master during a read or as a target during a write. The Universe II drives PAR when it provides data as a target during a read or a master during a write. The Universe II also drives PAR during the address phase of a transaction when it is a master and monitors PAR during an address phase when it is the PCI target. In both address and data phases, the PAR signal provides even parity for C/BE_[3:0] and AD[31:0]. If the Universe II is powered up in a 64-bit PCI environment, then PAR64 provides even parity for C/BE_[7:4] and AD[63:32]. Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 128 8. Error Handling > Errors on Decoupled Transactions The PERESP and SERR_EN bits in the “PCI Configuration Space Control and Status Register (PCI_CSR)” on page 172 determine whether or not the Universe II responds to parity errors. Data parity errors are reported through the assertion of PERR_ if the PERESP bit is set. Address parity errors, reported through the SERR_ signal, are reported if both PERESP and SERR_EN are set. Regardless of the setting of these two bits, the D_PE (Detected Parity Error) bit in the PCI_CS register is set if the Universe II encounters a parity error as a master or as a target. The DP_D (Data Parity Detected) bit in the same register is only set if parity checking is enabled through the PERESP bit and the Universe II detects a parity error while it is PCI master (that is, it asserts PERR_ during a read transaction or receives PERR_ during a write). No interrupts are generated by the Universe II either as a master or as a target in response to parity errors reported during a transaction. Parity errors are reported by the Universe II through assertion of PERR_ and by setting the appropriate bits in the PCI_CSR register. If PERR_ is asserted to the Universe II while it is PCI master, the only action it takes is to set the DP_D. Regardless of whether the Universe II is the master or target of the transaction, and regardless which agent asserted PERR_, the Universe II does not take any action other than to set bits in the PCI_CSR register. The Universe II continues with a transaction independent of any parity errors reported during the transaction. Similarly, address parity errors are reported by the Universe II (if the SERR_EN bit and the PERESP bit are set) by asserting the SERR_ signal for one clock cycle and setting the S_SERR (Signalled SERR_) bit in the PCI_CSR register. Assertion of SERR_ can be disabled by clearing the SERR_EN bit in the PCI_CSR register. No interrupt is generated, and regardless of whether assertion of SERR_ is enabled or not, the Universe II does not respond to the access with DEVSEL_. Typically the master of the transaction times-out with a Master-Abort. As a master, the Universe II does not monitor SERR_. It is expected that a central resource on the PCI bus monitors SERR_ and takes appropriate action. Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 129 9. Resets, Clocks and Power-up Options This chapter highlights utility functions in the Universe II. This chapter discusses the following topics: 9.1 • “Resets” on page 129 • “Power-Up Options” on page 135 • “Test Modes” on page 141 • “Clocks” on page 142 Overview The Universe II has many programmable reset options and power-up options that impact the functionality of the device. 9.2 Resets The Universe II provides a number of pins and registers for reset support. Pin support is summarized in Table 19. Table 19: Hardware Reset Mechanisms Interface and Direction Pin Name Long Name Effectsa VMEbus Input VRSYSRST_ VMEbus Reset Input Asserts LRST_ on the local bus, resets the Universe II, and re-configures power-up options. VMEbus Output VXSYSRST VMEbus System Reset Universe II output for SYSRST* (resets the VMEbus) PCI Input PWRRST_ Power-up Reset Resets the Universe II and re-configures power-up options. RST_ PCI Reset Input Resets the Universe II from the PCI bus. VME_ RESET_ VMEbus Reset Initiator Causes Universe II to assert VXSYSRST PCI Output LRST_ PCI Bus Reset Output Resets PCI resources JTAG Input TRST_ JTAG Test Reset Provides asynchronous initialization of the TAP controller in the Universe II. a. A more detailed account of the effects of reset signals is provided in “Reset Implementation Cautions” on page 133 Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 130 9. Resets, Clocks and Power-up Options > Resets The Universe II is only reset through hardware. Software can make the Universe II assert its reset outputs. In order to reset the Universe II through software, the Universe II reset outputs must be connected to the Universe II reset inputs. For example, the SW_LRST bit in the MISC_CTL register, which asserts the LRST_ output, does not reset the Universe II itself unless LRST_ is looped back to RST_. As described in “Reset Implementation Cautions” on page 133, there are potential loopback configurations resulting in permanent reset. Table 20: Software Reset Mechanism Register Name Type “Master Control Register (MAST_CTL)” on page 271 SW_LRST W Function Software PCI Reset 0=No effect 1=Initiate LRST_ A read always returns 0. SW_SYSRST W Software VMEbus SYSRESET 0=No effect 1=Initiate SYSRST* A read always returns 0 “VMEbus CSR Bit Set Register (VCSR_SET)” on page 330 RESET R/W Board Reset Reads: 0=LRST_ not asserted 1=LRST_ asserted Writes: 0=no effect 1=assert LRST_ SYSFAIL R/W VMEbus SYSFAIL Reads: 0=VXSYSFAIL not asserted 1=VXSYSFAIL asserted Writes: 0=no effect 1=assert VXSYSFAIL Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 9. Resets, Clocks and Power-up Options > Resets 131 Table 20: Software Reset Mechanism Register Name Type “VMEbus CSR Bit Clear Register (VCSR_CLR)” on page 329 RESET R/W Function Board Reset Reads: 0=LRST_ not asserted 1=LRST_ asserted Writes: 0=no effect 1=negate LRST_ SYSFAIL R/W VMEbus SYSFAIL Reads: 0=VXSYSFAIL not asserted 1=VXSYSFAIL asserted Writes: 0=no effect 1=negate VXSYSFAIL Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 132 9.2.1 9. Resets, Clocks and Power-up Options > Resets Universe II Reset Circuitry Table 21 on page 132 and Figure 21 on page 133 shows how to reset various aspects of the Universe II. For example, it shows that in order to reset the clock services (SYSCLK, CLK64 enables, and PLL divider), PWRRST_ must be asserted. PWRRST_ resets all aspects of Universe II listed in column 1 of Table 21. Table 21 also indicates the reset effects that are extended in time. For example, VXSYSRST_ remains asserted for 256 ms after all initiators are removed—this satisfies VMEbus Specification (minimum of 200 ms SYSRST*). The external 64 MHz clock controls this assertion time. LRST_ is asserted for 5 ms or more from all sources except VRSYSRST_. Table 21: Functions Affected by Reset Initiators Effect of Reseta Clock Services Reset Source PWRRST_ SYSCLK CLK64 enables PLL Divider VMEbus Services PWRRST_, or VRSYSRST_ VMEbus Arbiter VMEbus Timer VCSR Registers General Services PWRRST_, RST_ or Most registers VRSYSRST_ Power-Up and Reset State Machine Power-up the device PWRRST_, or VRSYSRST_ Reset Registers PWRRST_, or VMEbus Reset Output VXSYSRST_ (asserted for more than 200 ms) VME_RESET_, or SW_SYSRST bit in MISC_CTL register PWRRST_, or VRSYSRST_, or PCI Bus Reset Output LRST_ (asserted for at least 5 ms) SW_LRST bit in MISC_CTL register, or RESET bit in VCSR_SET registerb a. On PWRRST_, options are loaded from pins. On SYSRST and RST_, options are loaded from values that were latched at the previous PWRRST_. b. LRST_ can be cleared by writing 1 to the RESET bit in the CSR_CLR register. Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 9. Resets, Clocks and Power-up Options > Resets 133 Figure 21: Reset Circuitry PWRRST_ VME Services (VME Arbiter, VMEbus timer, VCSR registers) VRSYSRST_ Clock Services (SYSCLK, CLK64 enables, PLL divider) General Services (Most Registers) RST_ Power-Up and Reset (Power-up, reset registers, assert VOE_) VOE_ VME_RESET_ hold for > 200ms VXSYSRST_ MISC_CTL Register SW_SYSRST SW_LRST VCSR_CLR and VCSR_SET Registers RESET hold for >= 5ms LRST_ Notes: 1. On PWRRST_, options are loaded from pins. On SYSRST and RST_, options are loaded from values that were latched at the previous PWRRST_. 2. Refer to “Registers” on page 163 to find the effects of various reset events. 9.2.2 Reset Implementation Cautions To prevent the Universe II from resetting the PCI bus, the LRST_ output can be left unconnected. Otherwise, LRST_ must be grouped with other PCI reset generators to assert the RST_ signal so that the following conditions are met: RST_ = LRST_ & reset_source1 & reset_source2 &... If the Universe II is the only initiator of PCI reset, LRST_ can be directly connected to RST_. Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 134 9. Resets, Clocks and Power-up Options > Resets Assertion of VME_RESET_ causes the Universe II to assert VXSYSRST_. Since VME_RESET_ causes assertion of SYSRST*, and since SYSRST* causes assertion of LRST_, tying both VME_RESET_ and LRST_ to RST_ will put the Universe II into permanent reset. If VME_RESET_ is driven by PCI reset logic, ensure that the logic is designed to break this feedback path. The PWRRST_ input keeps the Universe II in reset until the power supply has reached a stable level (see Table 21). It must be held asserted for over 100 milliseconds after power is stable. Typically this can be achieved through a resistor/capacitor combination (see Figure 22)or under voltage sensing circuits. Figure 22: Resistor-Capacitor Circuit Ensuring Power-Up Reset Duration 47 K PWRRST_ 10μF The Universe II supports the VMEbus CSR Bit Clear and Bit Set registers. The “VMEbus CSR Bit Set Register (VCSR_SET)” on page 330 register allows the user to assert LRST_ or SYSFAIL by writing to the RESET or SYSFAIL bits. LRST_ or SYSFAIL remains asserted until the corresponding bit is cleared in the“VMEbus CSR Bit Clear Register (VCSR_CLR)” on page 329. The FAIL bit in each of these registers is a status bit and is set by the software to indicate board failure. Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 9. Resets, Clocks and Power-up Options > Power-Up Options 9.3 135 Power-Up Options The Universe II can be automatically configured at power-up to operate in different functional modes. These power-up options allow the Universe II to be set in a particular mode independent of any local intelligence. Table 22: Power-Up Optionsa Option Register Field Default Pins VMEbus Register Access Slave Image VRAI_CTL EN Disabled VA[31] VAS A16 VA[30:29] VRAI_BS BS 0x00 VA[28:21] VCSR_CTL LAS Memory VA[20] VCSR_TO TO 0x00 VA[19:15] MISC_STAT DY4AUTO Disabled VD[30] MISC_CTL V64AUTO Disabled VD[29] VINT_EN SW_INT 0 VINT_STAT SW_INT 0 VINT_MAP1 SW_INT 000 BI-Mode MISC_CTL BI Disabled VD[28] Auto-Syscon Detect MISC_CTL SYSCON Enabled VBGIN[3]* SYSFAIL* Assertion VCSR_SET SYSFAIL Asserted VD[27] VCSR_CLR SYSFAIL LSI0_CTL EN Disabled VA[13] LAS Memory VA[12] VAS A16 VA[11:10] LSI0_BS BS 0x0 VA[9:6] LSI0_BD BD 0x0 VA[5:2] PCI Register Access PCI_BS0, PCI_BS1 SPACE See “Registers” on page 163 VA[1] PCI Bus Sizeb MISC_STAT LCLSIZE 32-bit REQ64_ PCI CSR Master Enable PCI_CSR BM Disabled VA[14] VMEbus CR/CSR Slave Image Auto-ID PCI Target Image a. All power-up options are latched only at the rising-edge of PWRRST_. They are loaded when PWRRST_, SYSRST* and RST_ are negated. Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 136 9. Resets, Clocks and Power-up Options > Power-Up Options b. The PCI Bus Size is loaded on any RST_ event (PCI 2.1 Specification). The majority of the Universe II power-up options are loaded from the VMEbus address and data lines after any PWRRST_ (see Table 22 on page 135). The PCI bus width (a power-up option required by the PCI 2.1 Specification) is loaded on any RST_ event from the REQ64_ pin. The second special power-up option is VMEbus SYSCON enabling, required by the VMEbus specification. The SYSCON option is loaded during a SYSRST* event from the BG3IN* signal. All power-up options are latched from the state of a particular pin or group of pins on the rising edge of PWRST_. Each of these pins, except REQ64_, has a weak internal pull-down to put the Universe II into a default configuration. (REQ64_ has an internal pull-up). If a non-default configuration is required, a pull-up of approximately 10k  is required on the signal. See “PCI Bus Width” on page 138 and the VMEbus Specification. The Universe II may be restored to the state it was in immediately following the previous power-up without re-asserting PWRRST_. After SYSRST* or RST_ (with PWRRST_ negated), the values that were originally latched at the rising edge of PWRRST_ are reloaded into the Universe II (except for PCI bus width and VMEbus SYSCON enabling, which are loaded from their pins). Table 22 lists the power-up options of the Universe II, the pins which determine the options, and the register settings that are set by this option. Each option is described in more detail in “Power-up Option Descriptions” on page 136. 9.3.1 Power-up Option Descriptions This section describes each of the groups of power-up options that were listed in Table 22. 9.3.1.1 VMEbus Register Access Image The Universe II has several VMEbus slave images, each of which can provide a different mapping of VMEbus cycles to PCI cycles. All VMEbus slave images are configurable through a set of VMEbus slave image registers: VSIx_CTL, VSIx_BS, VSIx_BD, and VSIx_TO. No VMEbus to PCI transaction is possible until these registers are programmed. The VMEbus Register Access Image (VRAI) power-up option permits access from the VMEbus to the Universe II internal registers at power-up. The power-up option allows programming of the VMEbus register access image address space and the upper five bits of its base address; all other bits are 0 (see Table 23). Once access is provided to the registers, then all other Universe II features (such as further VMEbus slave images) can be configured from the VMEbus. Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 9. Resets, Clocks and Power-up Options > Power-Up Options 137 Table 23 shows how the upper bits in the VRAI base address are programmed for A16, A24, and A32 VMEbus register access images. Table 23: VRAI Base Address Power-up Options VRAI_CTL: VAS BS [31:24] BS [23:16] BS [15:12] A16 0 0 Power-up Option VA [28:25] A24 0 Power-up Option 0 VA [28:21] A32 Power-up Option 0 0 VA [28:21] 9.3.1.2 VMEbus CR/CSR Slave Image CR/CSR space is an address space introduced in the VME64 Specification. The CR/CSR space on any VMEbus device is 512 Kbytes in size; the upper region of the 512 Kbytes dedicated to register space, and the lower region is dedicated to configuration ROM. The Universe II maps its internal registers to the upper region of the CR/CSR space, and passes all other accesses through to the PCI bus (see “Registers” on page 163). The VMEbus CR/CSR Slave Image power-up option maps CR/CSR accesses to the PCI bus. CR/CSR space can be mapped to memory or I/O space with a 5-bit offset. This allows mapping to any 128 Mbyte page on the PCI bus. As part of this implementation, ensure that the PCI Master Interface is enabled through the MAST_EN bit power-up option or configured through a register access before accessing configuration ROM. 9.3.1.3 Auto-ID There are two Auto-ID mechanisms provided by the Universe II. One is the VME64 Specification version which relies upon use of the CR/CSR space for configuration of the VMEbus system, and a IDT proprietary system which uses the IACK daisy chain for identifying cards in a system. Either of these mechanisms can be enabled at power-up (see “Automatic Slot Identification” on page 42). Because VME64 Auto-ID relies upon SYSFAIL to operate correctly, this power-up option overrides the SYSFAIL power-up option described in “SYSFAIL* Assertion” on page 138. 9.3.1.4 BI-Mode BI-Mode (Bus Isolation Mode) is a mechanism for logically isolating the Universe II from the VMEbus for diagnostic, maintenance and failure recovery purposes. BI-Mode can be enabled as a power-up option (see “Bus Isolation Mode (BI-Mode)” on page 46). When the Universe II has been powered-up in BI-Mode, then any subsequent SYSRST* or RST_ restores the Universe II to BI-Mode, Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 138 9.3.1.5 9. Resets, Clocks and Power-up Options > Power-Up Options Auto-Syscon Detect The VMEbus SYSCON enabling, required by the VMEbus Specification, is a special power-up option in that it does not return to its after-power-up state following RST_ or SYSRST_. The SYSCON option is loaded during a SYSRST* event from the VBG3IN* signal. 9.3.1.6 SYSFAIL* Assertion This power-up option causes the Universe II to assert SYSFAIL* immediately upon entry into reset. The SYSFAIL* pin is released through a register access. Note that this power-up option is over-ridden if VME64 Auto-ID has been enabled. This option is used when extensive on-board diagnostics need to be performed before release of SYSFAIL*. After completion of diagnostics, SYSFAIL* can be released through software or through initiation of the VME64 Auto-ID sequence (if enabled, see “Auto Slot ID: VME64 Specified” on page 42). 9.3.1.7 PCI Target Image The PCI Target Image power-up option provides for default enabling of a PCI target image (automatically mapping PCI cycles to the VMEbus). The default target image can be mapped with base and bounds at 256MB resolution in Memory or I/O space, and map PCI transactions to different VMEbus address spaces. Beyond the settings provided for in this power-up option, the target image possesses its other default conditions: the translation offset is 0, posted writes are disabled, and only 32-bit (maximum) non-block VMEbus cycles in the non-privileged data space are generated. This option is typically used to access permits the use of Boot ROM on another card in the VMEbus system. 9.3.1.8 PCI Register Access A power-up option determines if the registers are mapped into Memory or I/O space. 9.3.1.9 PCI Bus Width The PCI Interface can be used as a 32-bit bus or 64-bit bus. The PCI bus width is determined during a PCI reset (see the PCI 2.1 Specification). The Universe II is configured as 32-bit PCI if REQ64_ is high on RST_; it is configured as 64-bit if REQ64_ is low. The Universe II has an internal pull-up on REQ64_, so the Universe II defaults to 32-bit PCI. On a 32-bit PCI bus, the Universe II drives all its 64-bit extension bi-direct signals at all times; these signals include: C/BE[7:4]_, AD[63:32], REQ64_, PAR64 and ACK64_ to unknown values. If used as a 32-bit interface, the 64-bit pins, AD[63:32], C/BE[7:4], PAR64 and ACK64_ can be left un-terminated. Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 9. Resets, Clocks and Power-up Options > Power-Up Options 9.3.1.10 139 PCI CSR Image Space There is a power-up option (using the VA[1] pin) that determines the value of the SPACE bit of the PCI_BSx registers. At power-up the SPACE bit of the PCI_BS1 register is the negation of the SPACE bit of the PCI_BS0 register. • When the VA pin is sampled low at power-up, the PCI_BS0 register’s SPACE bit is set to 1, which signifies I/O space, and the PCI_BS1 register’s SPACE bit is set to 0, which signifies Memory space. • When VA is sampled high at power-up, the PCI_BS0 register’s SPACE register’s bit is set to 0, which signifies Memory space, and the PCI_BS1 register’s SPACE bit is set to 1, which signifies I/O space. Once set, this mapping is constant until the next power-up sequence. See “Memory or I/O Access” on page 77. 9.3.2 Power-up Option Implementation In order to implement power-up requirements for the Universe II weak pull-up resistors are required. 9.3.2.1 Pull-up Requirements The pull-ups for the general power-up options (if other than default values are required) must be placed on the VA[31:1] and VD[31:27] lines. During reset, the Universe II negates VOE_, putting these signals into a high-impedance state. While VOE_ is negated the pull-ups (or internal pull-downs) bring the option pins (on A[31:1] and D[31:27]) to their appropriate state. The internal pull-downs are very weak. The leakage current on many transceivers can be sufficient to override these pull-downs. To ensure proper operation designers must ensure power-up option pins go to the correct state. Within two CLK64 periods after PWRRST_ is negated, the Universe II latches the levels on the option pins, and then negates VOE_ one clock later. This enables the VMEbus transceivers inwards. Figure 23: Power-up Options Timing minimum 100ms CLK64 PWRRST_ VOE_ VA, VD Integrated Device Technology www.idt.com power-up options Universe II User Manual May 12, 2010 140 9. Resets, Clocks and Power-up Options > Power-Up Options The power-up options are subsequently loaded into their respective registers several PCI clock periods after PWRRST_, SYSRST* and RST_ have all been negated. Because of the power-up configuration, the VMEbus buffers are not enabled until several CLK64 periods after release of SYSRST* (approximately 45 ns). Allowing for worst case backplane skew of 25 ns, the Universe II is not prepared to receive a slave access until 70 ns after release of SYSRST*. 9.3.3 Hardware Initialization (Normal Operating Mode) The Universe II has I/O capabilities that are specific to manufacturing test functions. These pins are not required in a non-manufacturing test setting. Table 24 shows how these pins must be terminated. Table 24: Manufacturing Pin Requirements for Normal Operating Mode Pin Name Pin Value tmode[2] VSS (or pulled-down if board tests are performed, see “Auxiliary Test Modes” on page 141) tmode[1] tmode[0] pll_testsel VSS enid VSS pll_testout No connect VCOCTL VSS Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 9. Resets, Clocks and Power-up Options > Test Modes 9.4 141 Test Modes The Universe II provides two types of test modes: auxiliary modes (NAND tree simulation and High Impedance) and JTAG (IEEE 1149.1). 9.4.1 Auxiliary Test Modes Two auxiliary test modes are supported: NAND tree and high impedance. The Universe II has three test mode input pins (TMODE[2:0]). For normal operations these inputs should be tied to ground (or pulled to ground through resistors). Table 25 below indicates the 3 operating modes of the Universe II. At reset the TMODE[2:0] inputs are latched by the Universe II to determine the mode of operation. The Universe II remains in this mode until the TMODE[2:0] inputs have changed and a reset event has occurred. PLL_TESTSEL must be high for any test mode. Table 25: Test Mode Operation Operation Mode TMODE[2:0] PLL_TESTSEL Normal Mode 000 0 Accelerate 001 0 PLL Test 010 1 Scan Mode 011 1 NAND Tree Simulation 100 1 RAM Test 101 1 High Impedance 110 0/1 Reserved 111 1 For NAND Tree Simulation, the values of the TMODE pins are latched during the active part of PWRRST_. These pins can change state during the NAND Tree tests. The timers are always accelerated in this mode. All outputs are tristated in this mode, except for the VXSYSFAIL output pin. For High Impedance mode, the values of the TMODE pins are also latched during the active part of PWRRST_. All outputs are tristated in this mode, except for the VXSYSFAIL output pin. 9.4.2 JTAG support The Universe II includes dedicated user-accessible test logic that is fully compatible with the IEEE 1149.1 Standard Test Access Port (TAP) and Boundary Scan Architecture. This standard was developed by the Test Technology Technical Committee of IEEE Computer Society and the Joint Test Action Group (JTAG). The Universe II’s JTAG support includes: • five-pin JTAG interface (TCK, TDI, TDO, TMS, and TRST_) • JTAG TAP controller • three-bit instruction register Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 142 9. Resets, Clocks and Power-up Options > Clocks • boundary scan register • bypass register • an IDCODE register The following required public instructions are supported: BYPASS (3'b111), SAMPLE(3'b100), and EXTEST(3'b000). The optional public instruction IDCODE(3'b011) selects the IDCODE register which returns 32'b01e201d. The following external pins are not part of the boundary scan register: LCLK, PLL_TESTOUT, PLL_TESTSEL, TMODE[3:0], and VCOCTL. 9.5 Clocks CLK64 is a 64 MHz clock that is required by the Universe II in order to synchronize internal Universe II state machines and to produce the VMEbus system clock (VSYSCLK) when the Universe II is system controller (SYSCON). This clock is specified to have a minimum 50-50 duty cycle with a maximum rise time of 5 ns. Using a different clock frequency is not recommended. It will alter various internal timers and change VME timing. Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 143 10. Signals and Pinout This chapter discusses the following topics: 10.1 • “VMEbus Signals” on page 144 • “PCI Bus Signals” on page 147 • “Pin-out” on page 151 Overview The following detailed description of the Universe II signals is organized according to these functional groups: • VMEbus Signals • PCI Signals Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 144 10. Signals and Pinout > VMEbus Signals 10.2 VMEbus Signals Table 26: VMEbus Signals CLK64 Input Reference Clock – this 64MHz clock is used to generate fixed timing parameters. It requires a 50-50 duty cycle (±20%) with a 5ns maximum rise time. CLK64 is required to synchronize the internal state machines of the VME side of the Universe II. VA [31:1] Bidirectional VMEbus Address Lines 31 to 01 – during MBLT transfers, VA 31-01 serve as data bits D63-D33. VA03-01 are used to indicate interrupt level on the VMEbus. VA_DIR Output VMEbus Address Transceiver Direction Control – the Universe II controls the direction of the address (VA31-01, VLWORD_) transceivers as required for master, slave and bus isolation modes. When the Universe II is driving lines on the VMEbus, this signal is driven high; when the VMEbus is driving the Universe II, this signal is driven low. VAM [5:0] Bidirectional VMEbus Address Modifier Codes – these codes indicate the address space being accessed (A16, A24, A32), the privilege level (user, supervisor), the cycle type (standard, BLT, MBLT) and the data type (program, data). VAM_DIR Output VMEbus AM Code Direction Control – controls the direction of the AM code transceivers as required for master, slave and bus isolation modes. When the Universe II is driving lines on the VMEbus, this signal is driven high; when the VMEbus is driving the Universe II, this signal is driven low. VAS_ Bidirectional VMEbus Address Strobe – the falling edge of VAS_ indicates a valid address on the bus. By continuing to assert VAS_, ownership of the bus is maintained during a RMW cycle. VAS_DIR Output VMEbus Address Strobe Direction Control – controls the direction of the address strobe transceiver as required for master, slave and bus isolation modes. When the Universe II is driving lines on the VMEbus, this signal is driven high; when the VMEbus is driving the Universe II, this signal is driven low. VBCLR_ Output VMEbus Bus Clear – requests that the current owner release the bus. Asserted by the Universe II when configured as SYSCON and the arbiter detects a higher level pending request. VBGI[3:0]_ Input VMEbus Bus Grant Inputs – The VME arbiter awards use of the data transfer bus by driving these bus grant lines low. The signal propagates down the bus grant daisy chain and is either accepted by a requester if it requesting at the appropriate level, or passed on as a VBGO [3:0]_ to the next board in the bus grant daisy chain. Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 10. Signals and Pinout > VMEbus Signals 145 Table 26: VMEbus Signals (Continued) VBGO[3:0]_ Output VMEbus Bus Grant Outputs – Only one output is asserted at any time, according to the level at which the VMEbus is being granted. VD[31:0]_ Bidirectional VMEbus Data Lines – 31 through 0 VD_DIR Output VMEbus Data Transceiver Direction Control – the Universe II controls the direction of the data (VD [31:0]) transceivers as required for master, slave and bus isolation modes. When the Universe II is driving lines on the VMEbus, this signal is driven high; when the VMEbus is driving the Universe II, this signal is driven low. VDS[1:0]_ Bidirectional VMEbus Data Strobes – the level of these signals are used to indicate active byte lanes During write cycles, the falling edge indicates valid data on the bus. During read cycles, assertion indicates a request to a slave to provide data. VDS_DIR Output VMEbus Data Strobe Direction Control – controls the direction of the data strobe transceivers as required for master, slave and bus isolation modes. When the Universe II is driving lines on the VMEbus, this signal is driven high; when the VMEbus is driving the Universe II, this signal is driven low. VDTACK_ Bidirectional VMEbus Data Transfer Acknowledge – VDTACK_ driven low indicates that the addressed slave has responded to the transfer. The Universe II always rescinds DTACK*. It is tristated once the initiating master negates AS*. VIACK_ Bidirectional VMEbus Interrupt Acknowledge – Indicates that the cycle just beginning is an interrupt acknowledge cycle. VIACKI_ Input VMEbus Interrupt Acknowledge In – Input for IACK daisy chain driver. If interrupt acknowledge is at same level as interrupt currently generated by the Universe II, then the cycle is accepted. If interrupt acknowledge is not at same level as current interrupt or Universe II is not generating an interrupt, then the Universe II propagates VIACKO_. VIACKO_ Output VMEbus Interrupt Acknowledge Out– Generated by the Universe II if it receives VIACKI_ and is not currently generating an interrupt at the level being acknowledged. VLWORD_ Bidirectional VMEbus Longword Data Transfer Size Indicator – This signal is used in conjunction with the two data strobes VDS [1:0]_ and VA 01 to indicate the number of bytes (1 – 4) in the current transfer. During MBLT transfers VLWORD_ serves as data bit D32. Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 146 10. Signals and Pinout > VMEbus Signals Table 26: VMEbus Signals (Continued) VOE_ Output VMEbus Transceiver Output Enable – Used to control transceivers to isolate the Universe II from the VMEbus during a reset or BI-mode. On power-up, VOE_ is high (to disable the buffers). VOE_ is negated during some VMEbus Slave Channel read operations. VRACFAIL_ Input VMEbus ACFAIL Input signal – Warns the VMEbus system of imminent power failure. This gives the modules in the system time to shut down in an orderly fashion before power-down. ACFAIL is mapped to a PCI interrupt. VRBBSY_ Input VMEbus Receive Bus Busy – Allows the Universe II to monitor whether the VMEbus is owned by another VMEbus master VRBERR_ Input VMEbus Receive Bus Error – A low level signal indicates that the addressed slave has not responded, or is signalling an error. VRBR[3:0]_ Input VMEbus Receive Bus Request Lines – If the Universe II is the Syscon, the arbiter logic monitors these signals and generates the appropriate Bus Grant signals. Also monitored by requester in ROR mode. VRIRQ[7:1]_ Input VMEbus Receive Interrupts 7 through 1 – These interrupts can be mapped to any of the Universe II’s PCI interrupt outputs. VRIRQ7-1_ are individually maskable, but cannot be read. VRSYSFAIL_ Input VMEbus Receive SYSFAIL – Asserted by a VMEbus system to indicate some system failure. VRSYSFAIL_ is mapped to a PCI interrupt. VRSYSRST_ Input VMEbus Receive System Reset – Causes assertion of LRST_ on the local bus and resets the Universe II. VSLAVE_DIR Output VMEbus Slave Direction Control – Transceiver control that allows the Universe II to drive DTACK* on the VMEbus. When the Universe II is driving lines on the VMEbus, this signal is driven high; when the VMEbus is driving the Universe II, this signal is driven low. VSYSCLK Bidirectional VMEbus System Clock – Generated by the Universe II when it is the Syscon and monitored during DY4 Auto ID sequence Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 10. Signals and Pinout > PCI Bus Signals 147 Table 26: VMEbus Signals (Continued) VSCON_DIR Output Syscon Direction Control – Transceiver control that allows the Universe II to drive VBCLR_ and SYSCLK. When the Universe II is driving lines on the VMEbus, this signal is driven high; when the VMEbus is driving the Universe II, this signal is driven low. VWRITE_ Bidirectional VMEbus Write signal – Indicates the direction of data transfer. VXBBSY Output VMEbus Transmit Bus Busy Signal – Generated by the Universe II when it is VMEbus master VXBERR Output VMEbus Transmit Bus Error Signal – Generated by the Universe II when PCI target generates Target-Abort on coupled PCI access from VMEbus. VXBR [3:0] Output VMEbus Transmit Bus Request – The Universe II requests the VMEbus when it needs to become VMEbus master. VXIRQ [7:1] Output VMEbus Transmit Interrupts – The VMEbus interrupt outputs are individually maskable. VXSYSFAIL Output VMEbus System Failure – Asserted by the Universe II during reset and plays a role in VME64 Auto ID. VXSYSRST Output VMEbus System Reset – The Universe II output for SYSRST*. 10.3 PCI Bus Signals Table 27: PCI Bus Signals ACK64_ Bidirectional Acknowledge 64-bit Transfer – It indicates slave can perform a 64-bit transfer when driven by the PCI slave (target). AD [31:0] Bidirectional PCI Address/Data Bus – Address and data are multiplexed over these pins providing a 32-bit address and 32-bit data bus. Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 148 10. Signals and Pinout > PCI Bus Signals Table 27: PCI Bus Signals AD [63:32] Bidirectional PCI Address/Data Bus – Address and data are multiplexed over these pins providing 64-bit address and data capability. C/BE_ [7:0] Bidirectional PCI Bus Command and Byte Enable Lines – Command and byte enable information is multiplexed over all eight C/BE lines. C/BE [7:4]_ are only used in a 64-bit PCI bus DEVSEL_ Bidirectional PCI Device Select – This signal is driven by the Universe II when it is accessed as PCI slave. ENID Input Enable IDD Tests – Required for ASIC manufacturing test, tie to ground for normal operation. FRAME_ Bidirectional Cycle Frame – This signal is driven by the Universe II when it is PCI initiator, and is monitored by the Universe II when it is PCI target GNT_ Input PCI Grant – indicates to the Universe II that it has been granted ownership of the PCI bus. IDSEL Input PCI Initialization Device Select – This signal is used as a chip select during configuration read and write transactions LINT[7:0]_ Bidirectional (Open Drain) PCI Interrupt Inputs – These PCI interrupt inputs can be mapped to any PCI bus or VMEbus interrupt output. IRDY_ Bidirectional Initiator Ready – Is used by the Universe II as PCI master to indicate that is ready to complete a current data phase. LCLK Input PCI Clock – Provides timing for all transactions on the PCI bus. PCI signals are sampled on the rising edge of CLK, and all timing parameters are defined relative to this signal. The PCI clock frequency of the Universe II must be between 25 and 33MHz. Lower frequencies result in invalid VME timing. LOCK_ Bidirectional Lock – Used by the Universe II to indicate an exclusive operation with a PCI device. While the Universe II drives LOCK_, other PCI masters are excluded from accessing that particular PCI device. When the Universe II samples LOCK_, it can be excluded from a particular PCI device. LRST_ Universe II User Manual May 12, 2010 Output Integrated Device Technology www.idt.com 10. Signals and Pinout > PCI Bus Signals 149 Table 27: PCI Bus Signals PCI Reset Output – Used to reset PCI resources. PAR Bidirectional Parity – Parity is even across AD [31:0] and C/BE [3:0] (the number of 1s summed across these lines and PAR equal an even number). PAR64 Bidirectional Parity Upper DWORD – Parity is even across AD [63:32] and C/BE [7:4] (the number of 1s summed across these lines and PAR equal an even number). PERR_ Bidirectional Parity Error – Reports parity errors during all transactions. The Universe II drives PERR_ high within two clocks of receiving a parity error on incoming data, and holds PERR_ for at least one clock for each errored data phase. PLL_TESTOUT Output Manufacturing Test Output—No connect PLL_TESTSEL Input Manufacturing Test Select—Tie to ground for normal operation PWRRST_ Input Power-up Reset – All Universe II circuitry is reset by this input. REQ_ Output Bus Request – Used by the Universe II to indicate that it requires the use of the PCI bus. REQ64_ Bidirectional 64-Bit Bus Request— Used to request a 64-bit PCI transaction. If the target does not respond with ACK64_, 32-bit operation is assumed. RST_ Input PCI Reset Input— Resets the Universe II from the PCI bus. SERR_ Output System Error – Reports address parity errors or any other system error. STOP_ Bidirectional Stop – Used by the Universe II as PCI slave when it wishes to signal the PCI master to stop the current transaction. As PCI master, the Universe II terminates the transaction if it receives STOP_ from the PCI slave. Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 150 10. Signals and Pinout > PCI Bus Signals Table 27: PCI Bus Signals TCK Input JTAG Test Clock Input – Used to clock the Universe II’s TAP controller. Tie to any logic level if JTAG is not used in the system. TDI Input JTAG Test Data Input – Used to serially shift test data and test instructions into the Universe II. Tie to any logic level if JTAG is not used in the system. TDO Output JTAG Test Data Output – Used to serially shift test data and test instructions out of the Universe II TMODE [2:0] Input Test Mode Enable – Used for chip testing, tie to ground for normal operation. TMS Input JTAG Test Mode Select – Controls the state of the Test Access Port (TAP) controller in the Universe II. Tie to any logic level if JTAG is not used in the system. TRDY_ Bidirectional Target Ready – Used by the Universe II as PCI slave to indicate that it is ready to complete the current data phase. During a read with Universe II as PCI master, the slave asserts TRDY_ to indicate to the Universe II that valid data is present on the data bus. TRST_ Input JTAG Test Reset – Provides asynchronous initialization of the TAP controller in the Universe II. Tie to ground if JTAG is not used in the system. VCOCTL Input Manufacturing testing – Tie to ground for normal operation VME_RESET_ Input VMEbus Reset Input — Generates a VME bus system reset. Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 10. Signals and Pinout > Pin-out 151 10.4 Pin-out 10.4.1 313-pin Plastic BGA Package (PBGA) A 1 vd[22] 2 3 vrberr_ 9 vam[5] 10 11 vam[3] 15 vas_ 18 19 va[12] 22 23 vxsysrst clk64 va[25] vrirq_[2] vxbr[3] Integrated Device Technology www.idt.com vxbr[0] vbgo_[2] int_[0] VDD vxirq[4] vracfail_ vbgo_[0] vxirq[5] vbgo_[1] ad[3] vrirq_[6] vbgi_[3] vbgi_[2] VDD ad[5] ad[34] ad[36] ad[40] tmode[2] VDD lock_ ad[44] VDD ad[6] 21 22 viacki_ VDD VSS 19 20 enid ad[42] ad[7] 17 18 ad[46] ad[10] ad[8] VDD ad[37] req64_ ad[11] tmode[1] 15 16 gnt_ ad[12] ad[43] VSS ad[47] stop_ 13 14 cbe[4] VDD rst_ ad[41] vme_res et_ VSS vxirq[7] ad[45] ad[39] cbe[5] idsel ad[13] VDD ad[38] ad[4] vxbr[2] VSS 11 12 VSS irdy_ 9 10 ad[16] cbe[7] VSS ad[14] frame_ ad[35] VDD ad[1] vbgo_[3] vrirq_[5] VDD vrsysfail_ 7 8 ad[19] ad[17] cbe[0] ad[20] ad[50] tmode[0] VDD VDD ad[2] vbgi_[0] vxirq[6] vxirq[2] VSS ad[52] 5 6 trdy_ VSS VDD ad[49] cbe[1] ad[15] vrirq_[7] ad[33] viacko_ req_ vxbr[1] ad[0] ad[53] cbe[3] par64 devsel_ 3 4 VSS ad[54] VSS ad[48] cbe[2] serr_ ack64_ VSS VSS VSS VDD vrirq_[4] vrirq_[1] va[31] vrirq_[3] vxirq[3] cbe[6] VSS vbgi_[1] ad[51] vcoctl AVDD 1 2 ad[22] ad[55] ad[21] ad[18] VSS VSS VSS ad[32] vxirq[1] va[27] va[29] va[30] va[26] VDD va[22] VSS VSS VDD VSS VDD AVSS VDD int_[1] AE VDD perr_ ad[23] VDD VDD VSS VSS VSS VDD VDD va[28] VDD VDD va[24] va[16] va[21] vlword_ va[7] ad[31] VSS ad[28] AD PLL_ testsel lclk vrbr_[0] AC PLL_ testout ad[56] int_[3] ad[29] ad[62] VSS VSS va[4] va[6] va[23] vrsysrst_ va[20] 24 25 va[17] va[19] va[1] va[13] vxsysfail vas_DIR VDD par VSS VSS VSS vdtack_ va[14] va[18] va[15] va[3] va[10] VSS tck vsysclk int_[6] int_[2] vds_DIR va_DIR VDD va[9] va[11] 20 21 va[5] va[8] vam[4] vds_[0] vrbr_[3] vd[1] VDD VDD trst_ vslave_D IR VDD vd_DIR vxberr vd[0] VDD vam[2] vam[0] tdo va[2] 16 17 vam[1] voe_ VDD vd[29] vd[31] tms tdi 14 viack_ vwrite_ VDD vds_[1] 12 13 vd[28] vam_DIR AB ad[58] ad[57] ad[26] AA ad[24] ad[25] VSS Y ad[27] VDD ad[60] W ad[59] pwrrst_ ad[30] V vrbr_[1] vrbr_[2] ad[63] U ad[61] VSS vrbbsy_ T lrst_ vbclr_ VDD R vxbbsy int_[4] vd[7] P VSS int_[5] vd[8] N vscon_ DIR vd[4] vd[10] M int_[7] vd[3] vd[17] L vd[2] vd[11] vd[16] K VDD vd[12] VDD J vd[6] vd[15] vd[25] H vd[5] vd[20] vd[24] G vd[13] VDD vd[27] F vd[9] vd[21] vd[30] E vd[14] vd[26] 8 D vd[19] vd[23] 6 7 C vd[18] 4 5 B 23 24 ad[9] 25 Universe II User Manual May 12, 2010 152 Universe II User Manual May 12, 2010 10. Signals and Pinout > Pin-out Integrated Device Technology www.idt.com 153 11. Electrical Characteristics This chapter discusses the following topics: • “DC Characteristics” on page 153 • “Operating Conditions” on page 160 • “Power Dissipation” on page 161 11.1 DC Characteristics 11.1.1 Non-PCI Characteristics Table 28 specifies the required DC characteristics of all non-PCI signals pins. Table 28: Non-PCI Electrical Characteristics Tested at 0°C to 70°C Symbols Parameters Test conditions Min Max VIH_TTL Voltage Input high VOUT = 0.1V or VDD – 0.1V; [IOUT] = 20 µA 2.2 V VDD + 0.3V VIH_CMOS Voltage Input high VOUT = 0.1V or VDD – 0.1V; [IOUT] = 20 µA 0.7 Vdd VDD + 0.3V VIL_TTL Voltage Input low VOUT = 0.1V or VDD – 0.1V; [IOUT] = 20 µA –0.3 V 0.8V VIL_CMOS Voltage Input low VOUT = 0.1V or VDD – 0.1V; [IOUT] = 20 µA –0.3 V 0.3VDD VT+_TTL Voltage Input high (Schmitt trigger) VOUT = 0.1V or VDD – 0.1V; [IOUT] = 20 µA - 2.4 V VT+_CMOS Voltage Input high (Schmitt trigger) VOUT = 0.1V or VDD – 0.1V; [IOUT] = 20 µA - 0.7VDD VT -_ttl Voltage Input low (Schmitt trigger) VOUT = 0.1V or VDD – 0.1V; [IOUT] = 20 µA 0.8V - VT -_CMOS Voltage Input low (Schmitt trigger) VOUT = 0.1V or VDD – 0.1V; [IOUT] = 20 µA 0.25VDD - IIN Input leakage current With no pull-up or pull-down resistance (Vin = Vss or VDD ) -5.0A 5.0A Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 154 11. Electrical Characteristics > DC Characteristics Table 28: Non-PCI Electrical Characteristics Tested at 0°C to 70°C Symbols Parameters Test conditions Min Max IIH Input leakage current high Inputs with pull-down resistance (Vin = Vdd) 10A 180A IIL Input leakage current low Inputs with pull-up resistance -180A -10A Tristate output leakage Vout = Vdd or Vss -10.0A 10.0A IOZ 11.1.2 (Vin = Vss) PCI Characteristics Table 29 specifies the required AC and DC characteristics of all PCI Universe II signal pins. Table 29: AC/DC PCI Electrical Characteristics Symbols Parameters VIL Condition Min Max Units Voltage Input low -0.5 0.8 V VIH Voltage Input high 2.0 Vdd + 0.5 V IIN Input leakage current Vin = 2.7V or 0.5V -10 10 A IIL Input leakage current low Vin = 0.5V -70 -10 A (Pin with pull-up) VOL Voltage output low Iout = 3mA, 6mA - 0.4 V VOH Voltage output high Iout = -2mA 2.4 - V IOH (AC)a Switching current high 0 < Vout21.4 -44 - mA 1.4 < Vout < 2.4 -44 + (Vout -1.4) / 0.024 - mA 3.1 < Vout < Vdd - Equation A mA (Test point) Vout = 3.1V - -142 mA Switching current high 0 < Vout31.4 95 - mA 0.6Vdd < Vout < 0.1Vdd Vout / 0.023 - mA 0.71 < Vout < 0 - Equation B mA Vout = 0.71V - 206 mA IOL (AC)b (Test point) Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 11. Electrical Characteristics > DC Characteristics 155 Table 29: AC/DC PCI Electrical Characteristics Symbols Parameters Condition Min Max Units ICL Low clamp current -5 Vin2-1 -25 + (Vin +10) / 0.015 - mA SLEWR Output rise slew rate 0.4V to 2.4V load 1 5 V/ns SLEWR Output fall slew rate 2.4V to 0.4V load 1 5 V/ns a. Equation A: Ioh = 11.9 * (Vout - 5.25) * (Vout + 2.45) for Vdd > Vout > 3.1V b. Equation B: Iol = 78.5 * Vout * (4.4 - Vout) for 0V < Vout < 0.71V 11.1.3 Pin List and DC Characteristics for all Signals Table 30 specifies the required DC characteristics of all Universe II signal pins Table 30: Pin List and DC Characteristics for Universe II Signals Pin Name Pin Number Type Input Type Output Type IOL (mA) IOH (mA) ack64# W11 I/O TTL 3S 6 –2 PCI Acknowledge 64 Bit Transfer AD [63:0] Listed in: “313 Pin PBGA Package” on page 333 I/O TTL 3S 6 –2 PCI Address/Data Pins C/BE# [0] Y14 I/O TTL 3S 6 –2 PCI Command and Byte Enables C/BE# [1] V14 C/BE# [2] T14 C/BE# [3] W13 C/BE# [4] AE15 C/BE# [5] AD14 C/BE# [6] T12 C/BE# [7] AD12 clk64 C23 I TTL – – – VME Clock 64 MHz—60-40 duty, 5 ns rise time devsel# AC7 I/O – 3S 6 –2 PCI Device Select enid AE21 I CMOS – – – Enable IDD Tests Integrated Device Technology www.idt.com Signal Description Universe II User Manual May 12, 2010 156 11. Electrical Characteristics > DC Characteristics Table 30: Pin List and DC Characteristics for Universe II Signals (Continued) Pin Name Pin Number Type Input Type Output Type IOL (mA) IOH (mA) frame# W17 I/O TTL 3S 6 –2 PCI Cycle Frame gnt# AE17 I TTL – – – PCI Grant idsel AB16 I TTL – – – PCI Initialization Device Select lint# [0] K20 I/O TTL OD 12 –12 lint# [1] AA5 I/O TTL OD 4 –4 lint# [2] L9 lint# [3] V6 lint# [4] M4 lint# [5] L3 lint# [6] M8 lint# [7] L1 irdy# AC15 I/O TTL 3S 6 –2 PCI Initiator Ready lclk AA3 I TTL – – – PCI Clock Signal lock# AA23 I/O TTL 3S 6 –2 PCI Lock lrst# R1 O – 3S 6 –2 PCI Reset Output par P8 I/O TTL 3S 6 –2 PCI parity par64 AE5 I/O TTL 3S 6 –2 PCI Parity Upper DWORD perr# AB4 I/O TTL 3S 6 –2 PCI Parity Error pll_testout AB2 For factory testing pll_testsel AC1 For factory testing pwrrst# T4 I TTL/Schm – – – Power–up Reset req# K22 O – 3S 6 –2 PCI Request req64# AD18 I/O TTL 3S 6 –2 PCI Request 64 Bit Transfer rst# AA19 I TTL – – – PCI Reset serr# AA7 O TTL OD 12 –12 PCI System Error stop# AB18 I/O TTL 3S 6 –2 PCI Stop Universe II User Manual May 12, 2010 Signal Description PCI Interrupt Integrated Device Technology www.idt.com 11. Electrical Characteristics > DC Characteristics 157 Table 30: Pin List and DC Characteristics for Universe II Signals (Continued) Pin Name Pin Number Type Input Type Output Type IOL (mA) IOH (mA) tck H12 I TTL – – – JTAG Test Clock Input tdi A13 I TTL – – – JTAG Test Data Input Signal Description (PU) tdo C13 O – 3S – – JTAG Test Data OUTput tmode [0] AA13 I TTL – – – Test Mode Enable tmode [1] AA21 tmode [2] W23 tms C11 TTL – – – JTAG Test Mode Select TTL 3S 6 –2 PCI Target Ready TTL – – – JTAG Test Reset 3S 3 –3 VMEbus Address Pins I (PU) trdy# AD8 I/O trst# E13 I (PU) Listed in: “313 Pin PBGA Package” on page 333 I/O vam [0] E11 I/O TTL 3S 3 –3 VMEbus Address Modifier Signals vam [1] D10 vam [2] G9 vam [3] B10 vam [4] H10 vam [5] A9 vam_dir B8 O – 3S 6 –6 VMEbus AM Signal Direction Control vas# B14 I/O TTL/Schm 3S 3 –3 VMEbus Address Strobe VA [31:1] TTL (PD) (PU) vas_dir K12 O – 3S 6 –6 VMEbus AS Direction Control va_dir G13 O – 3S 12 –12 VMEbus Address Direction Control vbclr# N3 O – 3S 3 –3 VMEbus BCLR* Signal Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 158 11. Electrical Characteristics > DC Characteristics Table 30: Pin List and DC Characteristics for Universe II Signals (Continued) Pin Name Pin Number Type Input Type Output Type IOL (mA) IOH (mA) vbgi# [0] N21 I TT – – – vbgi# [1] M16 vbgi# [2] N25 vbgi# [3] N23 vbgo# [0] M20 vbgo# [1] L25 vbgo# [2] M18 vbgo# [3] M24 vcoctl Signal Description VMEbus Bus Grant In TT (PD) O – 3S 12 –12 VMEbus Bus Grant Out AE3 I – – – – Factory testing VD [31:0] Listed in: “313 Pin PBGA Package” on page 333 I/O TTL 3S 3 –3 VMEbus Data Pins vd_dir F10 O – 3S 12 –12 VMEbus Data Direction Control vds# [0] F12 I/O TTL 3S 3 –3 VMEbus Data Strobes vds# [1] A11 vds_dir J11 O – 3S 6 –6 VMEbus Data Strobe Direction Control vdtack# G15 I/O TTL/Schm 3S 3 –3 VMEbus DTACK* Signal (PU) (PU) viack# E7 I/O TTL 3S 3 –3 VMEbus IACK* Signal viacki# AE23 I TTL – – – VMEbus IACKIN* Signal viacko# L21 O – 3S 12 –12 VMEbus IACKOUT* Signal vlword# K14 I/O TTL 3S 3 –3 VMEbus LWORD* Signal (PD) VME_RESET# V22 I TTL voe# B12 O – 3S 24 –24 vracfail# P18 I TTL/Schm – – – Universe II User Manual May 12, 2010 VMEbus Reset Input VMEbus Transceiver Output Enable VMEbus ACFAIL* Signal Integrated Device Technology www.idt.com 11. Electrical Characteristics > DC Characteristics 159 Table 30: Pin List and DC Characteristics for Universe II Signals (Continued) Pin Name Pin Number Type Input Type Output Type IOL (mA) IOH (mA) vrbbsy# M6 I TTL/Schm – – – VMEbus Received BBSY* Signal vrberr# A7 I TTL/Schm – – – VMEbus Receive Bus Error vrbr# [0] W5 I TTL/Schm – – – VMEbus Receive Bus Request vrbr# [1] U1 vrbr# [2] R3 vrbr# [3] L7 vrirq# [1] H22 I TTL/Schm – – – VMEbus Receive Interrupts vrirq# [2] H20 vrirq# [3] E25 vrirq# [4] J21 vrirq# [5] V16 vrirq# [6] P20 vrirq# [7] R17 vrsysfail# AC13 I TTL – – – VMEbus Receive SYSFAIL Signal vrsysrst# C21 I TTL/Schm – – – VMEbus Receive SYSRESET* Signal vscon_dir M2 O – 3S 6 –6 SYSCON signals direction control vslave_dir C15 O – 3S 6 –6 DTACK/BERR direction control vsysclk N7 I/O TTL 3S 3 –3 VMEbus SYSCLK Signal vwrite# D8 I/O TTL 3S 3 –3 VMEbus Write vxbbsy P2 O – 3S 3 –3 VMEbus Transmit BBSY* Signal vxberr D12 O – 3S 3 –3 VMEbus Transmit Bus Error (BERR*) Integrated Device Technology www.idt.com Signal Description Universe II User Manual May 12, 2010 160 11. Electrical Characteristics > Operating Conditions Table 30: Pin List and DC Characteristics for Universe II Signals (Continued) Pin Name Pin Number Type Input Type Output Type IOL (mA) IOH (mA) vxbr [0] G25 O – 3S 3 –3 VMEbus Transmit Bus Request vxbr [1] H24 vxbr [2] P24 vxbr [3] G23 vxirq [1] J19 O – 3S 3 –3 VMEbus Transmit Interrupts vxirq [2] K24 vxirq [3] K18 vxirq [4] J25 vxirq [5] L23 vxirq [6] M22 vxirq [7] R25 vxsysfail M10 O – 3S 3 –3 VMEbus Transmit SYSFAIL Signal vxsysrst A23 O – 3S 3 –3 VMEbus Transmit SYSRESET* Signal 11.2 Signal Description Operating Conditions The following table specifies recommended operating condition for the Universe II. Table 31: Operating Conditions Frequency Operation (Mhz) Symbols Parameters Min Max Vdd DC Supply Voltage (5V ± 10%) 4.5V 5.5V Ta (Commercial) Ambient Temperature 0ºC +70ºC 25 - 33 Ta (Industrial) Ambient Temperature -40ºC +85ºC 25 - 33 Ta (Extended) Ambient Temperature -55ºC +125ºC 25 Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 11. Electrical Characteristics > Power Dissipation 11.2.1 161 Absolute Maximum Ratings Table 32: Absolute Maximum Ratings Parameter Range DC Supply Voltage (VSS to VDD) -0.3 to 7.0 V DC Input Voltage (VIN) -0.5 to vdd + 0.5V DC Current Drain per Pin, Any Single Input or Output ± 50ma DC Current Drain per Pin, Any Paralleled Outputs ± 100ma DC Current Drain VDD and VSS Pins ± 75ma Storage Temperature, (TSTG) -40 ºC to 125 ºC Stresses beyond those listed above may cause permanent damage to the devices. These are stress ratings only, and functional operation of the devices at these or any other conditions beyond those indicated in the operational sections of this document is not implied. Exposure to maximum rating conditions for extended periods may affect device reliability. 11.3 Power Dissipation Table 33: Power Dissipation Parameter Rating IDLE Power Dissipation 1.50 W Typical Power Dissipation (32-bit PCI) 2.00 W Power Dissipation (64-bit PCI) 2.20 W Maximum Power Dissipation (32-bit PCI) 2.70W Power Dissipation (64-bit PCI) 3.20W Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 162 11.4 11. Electrical Characteristics > Power Sequencing Power Sequencing When designing with the Universe II device, care must be taken when powering the device to ensure proper operation. During power-up, no signals must be applied to any Universe II signal pins prior to stable power being applied to the device. In a mixed 3.3V and 5V design, IDT recommends that 5V power be stable prior to other devices coming out reset. If other devices come out of reset before the 5V power is stable, make certain that no signals are driven to the Universe II signal pins - including possible signals from the VME backplane. Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 163 12. Registers This appendix discusses the following topics: 12.1 • “Overview” on page 163 • “Register Map” on page 164 • “Register Description” on page 171 Overview The Universe II Control and Status Registers facilitate host system configuration and allow the user to control Universe II operational characteristics. The registers are divided into three groups: • PCI Configuration Space • VMEbus Configuration and Status Registers • Universe II Device Specific Status Registers Universe II registers have little-endian byte-ordering. Figure 24 below summarizes the supported register access mechanisms. Figure 24: UCSR Access Mechanisms VMEbus Configuration and Status Registers (VCSR) UNIVERSE DEVICE SPECIFIC REGISTERS (UDSR) 4 Kbytes UCSR Space PCI CONFIGURATION SPACE (PCICS) Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 164 12. Registers > Register Map Bits listed as reserved must be programmed with a value of 0. Reserved bits always read a value of zero. 12.2 Register Map Table 34 lists the Universe II registers by address offset. Table 34: Universe II Register Map Offset Register Name 0x000 “PCI Configuration Space ID Register (PCI_ID)” on page 171 0x004 “PCI Configuration Space Control and Status Register (PCI_CSR)” on page 172 0x008 “PCI Configuration Class Register (PCI_CLASS)” on page 176 PCI_CLASS 0x00C “PCI Configuration Miscellaneous 0 Register (PCI_MISC0)” on page 177 PCI_MISC0 0x010 “PCI Configuration Base Address Register (PCI_BS0)” on page 178 PCI_BS0 0x014 “PCI Configuration Base Address 1 Register (PCI_BS1)” on page 179 PCI_BS1 0x018-0x024 PCI_ID PCI Unimplemented 0x028 PCI Reserved 0x02C PCI Reserved 0x030 PCI Unimplemented 0x034 PCI Reserved 0x038 PCI Reserved 0x03C “PCI Configuration Miscellaneous 1 Register (PCI_MISC1)” on page 180 0x040-0x0FF PCI_CSR PCI_MISC1 PCI Unimplemented 0x100 “PCI Target Image 0 Control (LSI0_CTL)” on page 181 LSI0_CTL 0x104 “PCI Target Image 0 Base Address Register (LSI0_BS)” on page 183 LSI0_BS 0x108 “PCI Target Image 0 Bound Address Register (LSI0_BD)” on page 184 LSI0_BD 0x10C “PCI Target Image 0 Translation Offset (LSI0_TO)” on page 185 LSI0_TO 0x110 Reserved 0x114 “PCI Target Image 1 Control (LSI1_CTL)” on page 186 LSI1_CTL 0x118 “PCI Target Image 1 Base Address Register (LSI1_BS)” on page 188 LSI1_BS 0x11C “PCI Target Image 1 Bound Address Register (LSI1_BD)” on page 189 LSI1_BD Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 12. Registers > Register Map 165 Table 34: Universe II Register Map (Continued) Offset Register Name 0x120 “PCI Target Image 1 Translation Offset (LSI1_TO)” on page 190 0x124 Reserved 0x128 “PCI Target Image 2 Control (LSI2_CTL)” on page 191 LSI2_CTL 0x12C “PCI Target Image 2 Base Address Register (LSI2_BS)” on page 193 LSI2_BS 0x130 “PCI Target Image 2 Bound Address Register (LSI2_BD)” on page 194 LSI2_BD 0x134 “PCI Target Image 2 Translation Offset (LSI2_TO)” on page 195 LSI2_TO 0x138 Reserved 0x13C “PCI Target Image 3 Control (LSI3_CTL)” on page 196 LSI3_CTL 0x140 “PCI Target Image 3 Base Address Register (LSI3_BS)” on page 198 LSI3_BS 0x144 “PCI Target Image 3 Bound Address Register (LSI3_BD)” on page 199 LSI3_BD 0x148 “PCI Target Image 3 Translation Offset (LSI3_TO)” on page 200 LSI3_TO 0x14C-0x16C LSI1_TO Reserved 0x170 “Special Cycle Control Register (SCYC_CTL)” on page 201 0x174 “Special Cycle PCI Bus Address Register (SCYC_ADDR)” on page 202 0x178 “Special Cycle Swap/Compare Enable Register (SCYC_EN)” on page 203 0x17C “Special Cycle Compare Data Register (SCYC_CMP)” on page 204 SCYC_CMP 0x180 “Special Cycle Swap Data Register (SCYC_SWP)” on page 205 SCYC_SWP 0x184 “PCI Miscellaneous Register (LMISC)” on page 206 0x188 “Special PCI Target Image (SLSI)” on page 208 0x18C “PCI Command Error Log Register (L_CMDERR)” on page 210 0x190 “PCI Address Error Log (LAERR)” on page 211 0x194-0x19C SCYC_CTL SCYC_ADDR SCYC_EN LMISC SLSI L_CMDERR LAERR Reserved 0x1A0 “PCI Target Image 4 Control Register (LSI4_CTL)” on page 212 LSI4_CTL 0x1A4 “PCI Target Image 4 Base Address Register (LSI4_BS)” on page 214 LSI4_BS 0x1A8 “PCI Target Image 4 Bound Address Register (LSI4_BD)” on page 215 LSI4_BD 0x1AC “PCI Target Image 4 Translation Offset (LSI4_TO)” on page 216 LSI4_TO 0x1B0 Reserved Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 166 12. Registers > Register Map Table 34: Universe II Register Map (Continued) Offset Register Name 0x1B4 “PCI Target Image 5 Control Register (LSI5_CTL)” on page 217 LSI5_CTL 0x1B8 “PCI Target Image 5 Base Address Register (LSI5_BS)” on page 219 LSI5_BS 0x1BC “PCI Target Image 5 Bound Address Register (LSI5_BD)” on page 220 LSI5_BD 0x1C0 “PCI Target Image 5 Translation Offset (LSI5_TO)” on page 221 LSI5_TO 0x1C4 Reserved 0x1C8 “PCI Target Image 6 Control Register (LSI6_CTL)” on page 222 LSI6_CTL 0x1CC “PCI Target Image 6 Base Address Register (LSI6_BS)” on page 224 LSI6_BS 0x1D0 “PCI Target Image 6 Bound Address Register (LSI6_BD)” on page 225 LSI6_BD 0x1D4 “PCI Target Image 6 Translation Offset (LSI6_TO)” on page 226 LSI6_TO 0x1D8 Reserved 0x1DC “PCI Target Image 7 Control Register (LSI7_CTL)” on page 227 LSI7_CTL 0x1E0 “PCI Target Image 7 Base Address Register (LSI7_BS)” on page 229 LSI7_BS 0x1E4 “PCI Target Image 7 Bound Address Register (LSI7_BD)” on page 230 LSI7_BD 0x1E8 “PCI Target Image 7 Translation Offset (LSI7_TO)” on page 231 LSI7_TO 0x1EC-0x1FC Reserved 0x200 “DMA Transfer Control Register (DCTL)” on page 232 DCTL 0x204 “DMA Transfer Byte Count Register (DTBC)” on page 234 DTBC 0x208 “DMA PCI Bus Address Register (DLA)” on page 235 0x20C Reserved 0x210 “DMA VMEbus Address Register (DVA)” on page 236 0x214 Reserved 0x218 “DMA Command Packet Pointer (DCPP)” on page 237 0x21C Reserved 0x220 “DMA General Control/Status Register (DGCS)” on page 238 0x224 “DMA Linked List Update Enable Register (D_LLUE)” on page 241 0x2280x-2FC 0x300 Universe II User Manual May 12, 2010 DLA DVA DCPP DGCS D_LLUE Reserved “PCI Interrupt Enable Register (LINT_EN)” on page 242 LINT_EN Integrated Device Technology www.idt.com 12. Registers > Register Map 167 Table 34: Universe II Register Map (Continued) Offset Register Name 0x304 “PCI Interrupt Status Register (LINT_STAT)” on page 244 LINT_STAT 0x308 “PCI Interrupt Map 0 Register (LINT_MAP0)” on page 246 LINT_MAP0 0x30C “PCI Interrupt Map 1 Register (LINT_MAP1)” on page 247 LINT_MAP1 0x310 “VMEbus Interrupt Enable Register (VINT_EN)” on page 248 0x314 “VMEbus Interrupt Status Register (VINT_STAT)” on page 251 VINT_STAT 0x318 “VME Interrupt Map 0 Register (VINT_MAP0)” on page 253 VINT_MAP0 0x31C “VME Interrupt Map 1 Register (VINT_MAP1)” on page 254 VINT_MAP1 0x320 “Interrupt STATUS/ID Out Register (STATID)” on page 255 STATID 0x324 “VIRQ1 STATUS/ID Register (V1_STATID)” on page 256 V1_STATID 0x328 “VIRQ2 STATUS/ID Register (V2_STATID)” on page 257 V2_STATID 0x32C “VIRQ3 STATUS/ID Register (V3_STATID)” on page 258 V3_STATID 0x330 “VIRQ4 STATUS/ID Register (V4_STATID)” on page 259 V4_STATID 0x334 “VIRQ5 STATUS/ID Register (V5_STATID)” on page 260 V5_STATID 0x338 “VIRQ6 STATUS/ID Register (V6_STATID)” on page 261 V6_STATID 0x33C “VIRQ7 STATUS/ID Register (V7_STATID)” on page 262 V7_STATID 0x340 “PCI Interrupt Map 2 Register (LINT_MAP2)” on page 263 LINT_MAP2 0x344 “VME Interrupt Map 2 Register (VINT_MAP2)” on page 264 VINT_MAP2 0x348 “Mailbox 0 Register (MBOX0)” on page 265 MBOX0 0x34C “Mailbox 1 Register (MBOX1)” on page 266 MBOX1 0x350 “Mailbox 2 Register (MBOX2)” on page 267 MBOX2 0x354 “Mailbox 3 Register (MBOX3)” on page 268 MBOX3 0x358 “Semaphore 0 Register (SEMA0)” on page 269 SEMA0 0x35C “Semaphore 1 Register (SEMA1)” on page 270 SEMA1 0x360-0x3FC VINT_EN Reserved 0x400 “Master Control Register (MAST_CTL)” on page 271 MAST_CTL 0x404 “Miscellaneous Control Register (MISC_CTL)” on page 273 MISC_CTL 0x408 “Miscellaneous Status Register (MISC_STAT)” on page 275 MISC_STAT Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 168 12. Registers > Register Map Table 34: Universe II Register Map (Continued) Offset 0x40C 0x410-0x4F8 0x4FC 0x500-0xEFC Register “User AM Codes Register (USER_AM)” on page 276 Name USER_AM Reserved “Universe II Specific Register (U2SPEC)” on page 277 U2SPEC Reserved 0xF00 “VMEbus Slave Image 0 Control Register (VSI0_CTL)” on page 279 VSI0_CTL 0xF04 “VMEbus Slave Image 0 Base Address Register (VSI0_BS)” on page 282 VSI0_BS 0xF08 “VMEbus Slave Image 0 Bound Address Register (VSI0_BD)” on page 283 VSI0_BD 0xF0C “VMEbus Slave Image 0 Translation Offset (VSI0_TO)” on page 284 VSI0_TO 0xF10 Reserved 0xF14 “VMEbus Slave Image 1 Control (VSI1_CTL)” on page 285 VSI1_CTL 0xF18 “VMEbus Slave Image 1 Base Address Register (VSI1_BS)” on page 287 VSI1_BS 0xF1C “VMEbus Slave Image 1 Bound Address Register (VSI1_BD)” on page 288 VSI1_BD 0xF20 “VMEbus Slave Image 1 Translation Offset (VSI1_TO)” on page 289 VSI1_TO 0xF24 Reserved 0xF28 “VMEbus Slave Image 2 Control (VSI2_CTL)” on page 290 VSI2_CTL 0xF2C “VMEbus Slave Image 2 Base Address Register (VSI2_BS)” on page 292 VSI2_BS 0xF30 “VMEbus Slave Image 2 Bound Address Register (VSI2_BD)” on page 293 VSI2_BD 0xF34 “VMEbus Slave Image 2 Translation Offset (VSI2_TO)” on page 294 VSI2_TO 0xF38 Reserved 0xF3C “VMEbus Slave Image 3 Control (VSI3_CTL)” on page 295 VSI3_CTL 0xF40 “VMEbus Slave Image 3 Base Address Register (VSI3_BS)” on page 297 VSI3_BS 0xF44 “VMEbus Slave Image 3 Bound Address Register (VSI3_BD)” on page 298 VSI3_BD 0xF48 “VMEbus Slave Image 3 Translation Offset (VSI3_TO)” on page 299 VSI3_TO Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 12. Registers > Register Map 169 Table 34: Universe II Register Map (Continued) Offset 0xF4C-0xF60 Register Name Reserved 0xF64 “Location Monitor Control Register (LM_CTL)” on page 300 LM_CTL 0xF68 “Location Monitor Base Address Register (LM_BS)” on page 302 LM_BS 0xF6C Reserved 0xF70 “VMEbus Register Access Image Control Register (VRAI_CTL)” on page 303 VRAI_CTL 0xF74 “VMEbus Register Access Image Base Address Register (VRAI_BS)” on page 304 VRAI_BS 0xF78-0xF7C Reserved 0xF80 “VMEbus CSR Control Register (VCSR_CTL)” on page 305 VCSR_CTL 0xF84 “VMEbus CSR Translation Offset (VCSR_TO)” on page 306 VCSR_TO 0xF88 “VMEbus AM Code Error Log (V_AMERR)” on page 307 V_AMERR 0xF8C “VMEbus Address Error Log (VAERR)” on page 308 0xF90 “VMEbus Slave Image 4 Control (VSI4_CTL)” on page 309r VSI4_CTL 0xF94 “VMEbus Slave Image 4 Base Address Register (VSI4_BS)” on page 311 VSI4_BS 0xF98 “VMEbus Slave Image 4 Bound Address Register (VSI4_BD)” on page 312 VSI4_BD 0xF9C “VMEbus Slave Image 4 Translation Offset (VSI4_TO)” on page 313 VSI4_TO 0xFA0 Reserved 0xFA4 “VMEbus Slave Image 5 Control (VSI5_CTL)” on page 314 VSI5_CTL 0xFA8 “VMEbus Slave Image 5 Base Address Register (VSI5_BS)” on page 316 VSI5_BS 0xFAC “VMEbus Slave Image 5 Bound Address Register (VSI5_BD)” on page 317 VSI5_BD 0xFB0 “VMEbus Slave Image 5 Translation Offset (VSI5_TO)” on page 318 VSI5_TO 0xFB4 Reserved 0xFB8 “VMEbus Slave Image 6 Control (VSI6_CTL)” on page 319 VSI6_CTL 0xFBC “VMEbus Slave Image 6 Base Address Register (VSI6_BS)” on page 321 VSI6_BS 0xFC0 “VMEbus Slave Image 6 Bound Address Register (VSI6_BD)” on page 322 VSI6_BD Integrated Device Technology www.idt.com VAERR Universe II User Manual May 12, 2010 170 12. Registers > Register Map Table 34: Universe II Register Map (Continued) Offset Register Name 0xFC4 “VMEbus Slave Image 6 Translation Offset (VSI6_TO)” on page 323 0xFC8 Reserved 0xFCC “VMEbus Slave Image 7 Control (VSI7_CTL)” on page 324 VSI7_CTL 0xFD0 “VMEbus Slave Image 7 Base Address Register (VSI7_BS)” on page 326 VSI7_BS 0xFD4 “VMEbus Slave Image 7 Bound Address Register (VSI7_BD)” on page 327 VSI7_BD 0xFD8 “VMEbus Slave Image 7 Translation Offset (VSI7_TO)” on page 328 VSI7_TO 0xFDC-0xFEC VSI6_TO Reserved 0xFF0 VME CR/CSR Reserved 0xFF4 “VMEbus CSR Bit Clear Register (VCSR_CLR)” on page 329 VCSR_CLR 0xFF8 “VMEbus CSR Bit Set Register (VCSR_SET)” on page 330 VCSR_SET 0xFFC “VMEbus CSR Base Address Register (VCSR_BS)” on page 331 Universe II User Manual May 12, 2010 VCSR_BS Integrated Device Technology www.idt.com 12. Registers > Register Description 12.3 171 Register Description The following tables describe the Universe II registers. 12.3.1 PCI Configuration Space ID Register (PCI_ID) Register name: PCI_VID Bits 7 Register offset: 0x00 6 5 4 3 31:24 DID 23:16 DID 15:08 VID 07:00 VID Bits Name 31:16 DID[15:0] Description Device ID 2 1 0 Type Reset by Reset value R all 0x10E3 R all 0x10E3 IDT allocated device identifier 15:00 VID[15:0] Vendor ID PCI SIG allocated vendor identifier Note: IDT acquired Tundra Semiconductor. Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 172 12.3.2 12. Registers > Register Description PCI Configuration Space Control and Status Register (PCI_CSR) Register offset: 0x004 Register name: PCI_CSR Bits 7 6 5 4 3 31:24 D_PE S_SERR R_MA R_TA S_TA 23:16 TFBBC 1 DEVSEL PCI Reserved WAIT Bits Name 31 D_PE 0 DP_D PCI Reserved 15:08 07:00 2 PERESP VGAPS MWI_EN SC Description Detected Parity Error This bit is always set by the Universe II when the PCI master interface detects a data parity error or the PCI target interface detects address or data parity errors. BM MFBBC SERR_EN MS IOS Type Reset by Reset value R/Write 1 to Clear All 0x0 R/Write 1 to Clear All 0x0 R/Write 1 to Clear All 0x0 R/Write 1 to Clear All 0x0 R/Write 1 to Clear All 0x0 0=No parity error 1=Parity error 30 S_SERR Signalled SERR_ The Universe II PCI target interface sets this bit when it asserts SERR_ to signal an address parity error. SERR_EN must be set before SERR_ can be asserted. 0=SERR_ not asserted 1=SERR_ asserted. 29 R_MA Received Master-Abort The Universe II PCI master interface sets this bit when a transaction it initiated had to be terminated with a Master-Abort. 0=Master did not generate Master-Abort 1=Master generated Master-Abort 28 R_TA Received Target-Abort The Universe II PCI master interface sets this bit when a transaction it initiated was terminated with a Target-Abort. 0=Master did not detect Target-Abort 1=Master detected Target-Abort. 27 S_TA Signalled Target-Abort 0=Target did not terminate transaction with Target-Abort 1=Target terminated transaction with Target-Abort. Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 12. Registers > Register Description Bits Name 26:25 DEVSEL 173 Type Reset by Reset value R All 0x1 R/Write 1 to Clear All 0x0 R All 0x0 Reserved R All 0x0 Master Fast Back-to-Back Enable R All 0x0 R/W All 0x0 R All 0x0 R/W All 0x0 Description Device Select Timing The Universe II is a medium speed device 24 DP_D Master Data Parity Error The Universe II PCI master interface sets this bit if the Parity Error Response bit is set, if it is the master of transaction in which it asserts PERR_, or the addressed target asserts PERR_. 0=Master did not detect/generate data parity error, 1=Master detected/generated data parity error. 23 TFBBC Target Fast Back to Back Capable Universe II cannot accept Back to Back cycles from a different agent. 22:10 Reserved 09 MFBBC The Universe II master never generates fast back to back transactions. 0=No fast back-to-back transactions 08 SERR_EN SERR_ Enable Setting this and PERESP allows the Universe II PCI target interface to report address parity errors with SERR_. 0=Disable SERR_ driver 1=Enable SERR_ driver Note: The Universe II only refuses PCI addresses with parity errors when both the PERESP and SERR_EN bits are programmed to a value of 1. 07 WAIT Wait Cycle Control 0=No address/data stepping 06 PERESP Parity Error Response Controls the Universe II response to data and address parity errors. When enabled, it allows the assertion of PERR_ to report data parity errors. When this bit and SERR_EN are asserted, the Universe II can report address parity errors on SERR_. Universe II parity generation is unaffected by this bit. 0=Disable 1=Enable Note: The Universe II only refuses PCI addresses with parity errors when both the PERESP and SERR_EN bits are programmed to a value of 1. Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 174 12. Registers > Register Description Bits Name 05 VGAPS Description VGA Palette Snoop Type Reset by Reset value R All 0x0 R All 0x0 R All 0x0 R/W PWR VME This bit is 1 after reset if the VMEbus Address [14] equals 1 during power-on reset. The Universe II treats palette accesses like all other accesses. 0=Disable 04 MWI_EN Memory Write and Invalidate Enable The Universe II PCI master interface never generates a Memory Write and Invalidate command. 0=Disable 03 SC Special Cycles 0=Disable The Universe II PCI target interface never responds to special cycles. 02 BM Master Enable 0=Disable 1=Enable For a VMEbus slave image to respond to an incoming cycle, this bit must be set. If this bit is cleared while there is data in the VMEbus Slave Posted Write FIFO, the data will be written to the PCI bus but no further data will be accepted into this FIFO until the bit is set. 01 MS Target Memory Enable 0=Disable 1=Enable Universe II User Manual May 12, 2010 This bit is reloaded by “all”. R/W PWR VME This bit is 1 after reset if the VMEbus Address [13:12] equals 10 during power-on reset. This bit is reloaded by “all” Integrated Device Technology www.idt.com 12. Registers > Register Description Bits Name 00 IOS Description Target IO Enable 0=Disable 1=Enable Integrated Device Technology www.idt.com 175 Type Reset by R/W PWR VME Reset value This bit is 1 after reset if the VMEbus Address [13:12] equals 10 during power-on reset. This bit is reloaded by “all” Universe II User Manual May 12, 2010 176 12.3.3 12. Registers > Register Description PCI Configuration Class Register (PCI_CLASS) Register name: PCI_CLASS Bits 7 Register offset: 0x08 6 5 4 3 31:24 BASE 23:16 SUB 15:08 PROG 07:00 RID Bits Name 31:24 BASE [7:0] Description Base Class Code 2 1 0 Type Reset by Reset value R All 0x06 R All 0x80 R All 0x00 R All 0x02 The Universe II is defined as a PCI bridge device 23:16 SUB [7:0] Sub Class Code The Universe II sub-class is “other bridge device” 15:08 PROG [7:0] Programming Interface The Universe II does not have a standardized register-level programming interface 07:00 RID [7:0] Universe II User Manual May 12, 2010 Revision ID Integrated Device Technology www.idt.com 12. Registers > Register Description 12.3.4 177 PCI Configuration Miscellaneous 0 Register (PCI_MISC0) The Universe II is not a multi-function device. Register name: PCI_MISC0 Register offset: 0x0C Bits 7 6 31:24 BISTC SBIST 23:16 MFUNCT 5 4 3 2 1 Reserved 0 CCODE LAYOUT 15:08 LTIMER Reserved 07:00 Reserved Bits Name 31 BISTC 30 SBIST Type Reset by Reset value The Universe II is not BIST Capable R All 0x0 Start BIST R All 0x0 Reserved R All 0x0 Completion Code R All 0x0 R All 0x0 R All 0x0 R/W All 0x0 R All 0x0 Description The Universe II is not BIST capable 29:28 Reserved 27:24 CCODE The Universe II is not BIST capable 23 MFUNCT Multifunction Device The Universe II is not a multi-function device. 0=No 1=Yes 22:16 LAYOUT Configuration Space Layout 15:11 LTIMER Latency Timer The latency timer has a resolution of eight clocks. When the Universe II latency timer is programmed for eight clock periods and FRAME_, IRDY_ and TRDY_ are asserted while GNT_ is not asserted, FRAME_ is negated on the next clock edge. 10:00 Reserved Integrated Device Technology www.idt.com Reserved Universe II User Manual May 12, 2010 178 12.3.5 12. Registers > Register Description PCI Configuration Base Address Register (PCI_BS0) This register specifies the 4 Kbyte aligned base address of the 4 Kbyte Universe II register space on PCI. Register name: PCI_BS0 Bits 7 Register offset: 0x10 6 5 4 3 31:24 BS 23:16 BS 15:08 2 1 BS 0 Reserved 07:00 Reserved SPACE Type Reset by Reset value R/W All 0x00 Reserved R All 0x00 PCI Bus Address Space R All Undefined Bits Name Description 31:12 BS[31:12] Base Address 11:02 Reserved 00 SPACE 0=Memory 1=I/O A power-up option determines if the registers are mapped into Memory or I/O space in relation to this base address (see “Power-Up Options” on page 135). If mapped into Memory space, the user is free to locate the registers anywhere in the 32-bit address space. If PCI_BS0 is mapped to Memory space, PCI_BS1 is mapped to I/O space; if PCI_BS0 is mapped to I/O space, then PCI_BS1 is mapped to Memory space. • When the VA[1] pin is sampled low at power-up, the PCI_BS0 register’s SPACE bit is set to 1, which signifies I/O space, and the PCI_BS1 register’s SPACE bit is set to 0, which signifies memory space. • When VA[1] is sampled high at power-up, the PCI_BS0 register’s SPACE register’s bit is set to 0, which signifies Memory space, and the PCI_BS1 register’s SPACE bit is set to 1, which signifies I/O space. A write must occur to this register before the Universe II Device Specific Registers can be accessed. This write can be performed with a PCI configuration transaction or a VMEbus register access. Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 12. Registers > Register Description 12.3.6 179 PCI Configuration Base Address 1 Register (PCI_BS1) This register specifies the 4 KByte aligned base address of the 4 KByte Universe II register space in PCI Register name: PCI_BS1 Bits 7 Register offset: 0x14 6 5 4 3 31:24 BS 23:16 BS 15:08 2 1 BS 0 Reserved 07:00 Reserved SPACE Type Reset by Reset value R/W All 0x00 Reserved R All 0x00 PCI Bus Address Space R All Undefined Bits Name Description 31:12 BS[31:12] Base Address 11:02 Reserved 00 SPACE 0=Memory 1=I/O A power-up option determines the value of the SPACE bit. This determines whether the registers are mapped into Memory or I/O space in relation to this base address (see “Power-Up Options” on page 135). If mapped into Memory space, the user is free to locate the Universe registers anywhere in the 32-bit address space. If PCI_BS0 is mapped to Memory space, PCI_BS1 is mapped to I/O space; if PCI_BS0 is mapped to I/O space, then PCI_BS1 is mapped to Memory space. • When the VA[1] pin is sampled low at power-up, the PCI_BS0 register’s SPACE bit is set to “1”, which signifies I/O space, and the PCI_BS1 register’s SPACE bit is set to “0”, which signifies memory space. • When VA[1] is sampled high at power-up, the PCI_BS0 register’s SPACE register’s bit is set to “0”, which signifies Memory space, and the PCI_BS1 register’s SPACE bit is set to “1”, which signifies I/O space. A write must occur to this register before the Universe II Device Specific Registers can be accessed. This write can be performed with a PCI configuration transaction or a VMEbus register access. The SPACE bit in this register is an inversion of the SPACE field in PCI_BS0. Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 180 12.3.7 12. Registers > Register Description PCI Configuration Miscellaneous 1 Register (PCI_MISC1) Register name: PCI_MISC1 Bits 7 Register offset: 0x3C 6 5 4 3 31:24 MAX_LAT [7:0] 23:16 MIN_GNT [7:0] 15:08 INT_PIN [7:0] 07:00 INT_LINE [7:0] Bits Name 31:24 MAX_LAT [7:0] 23:16 MIN_GNT [7:0] 2 1 0 Type Reset by Reset value Maximum Latency: This device has no special latency requirements R All 0x00 Minimum Grant R All 0b11 R All 0b01 R/W All 0x00 Description 250 ns increments The MIN_GNT parameter assumes the Universe II master is transferring an aligned burst size of 64 bytes to a 32-bit target with no wait states. This would require roughly 20 clocks (at a clock frequency of 33 MHz, this is about 600 ns). MIN_GNT is set to three, or 750 ns. 15:08 INT_PIN [7:0] Interrupt Pin Universe II pin INT_ [0] has a PCI compliant I/O buffer 07:00 INT_LINE [7:0] Interrupt Line Used by some PCI systems to record interrupt routing information Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 12. Registers > Register Description 12.3.8 181 PCI Target Image 0 Control (LSI0_CTL) In the PCI Target Image Control register, setting the VCT bit will only have effect if the VAS bits are programmed for A24 or A32 space and the VDW bits are programmed for 8-bit, 16-bit, or 32-bit. If VAS bits are programmed to A24 or A32 and the VDW bits are programmed for 64-bit, the Universe II may perform MBLT transfers independent of the state of the VCT bit. The setting of the PWEN bit is ignored if the LAS bit is programmed for PCI Bus I/O Space, forcing all transactions through this image to be coupled. Register name: LSI0_CTL Register offset: 0x100 Bits 7 6 31:24 EN PWEN 23:16 15:08 5 4 2 1 Reserved PGM Reserved 07:00 SUPER VAS Reserved VCT Reserved Bits Name 31 EN 0 Reserved VDW Reserved 3 Description Image Enable LAS Type Reset by Reset value R/W All Undefined R/W All 0x00 R All 0x00 R/W All 0b10 R All 0x00 0=Disable 1=Enable 30 PWEN Posted Write Enable 0=Disable 1=Enable 29:24 Reserved 23:22 VDW Reserved VMEbus Maximum Data width 00=8-bit data width 01=16 bit data width 10=32-bit data width 11=64-bit data width 21:19 Reserved Integrated Device Technology www.idt.com Reserved Universe II User Manual May 12, 2010 182 12. Registers > Register Description Bits Name 18:16 VAS Description VMEbus Address Space Type Reset by Reset value R/W All Undefined R All 0x00 R/W All 0x00 R All 0x00 R/W All 0x00 R All 0x00 R/W All 0x00 R All 0x00 R/W All Undefined 000=A16 001=A24 010=A32 011= Reserved 100=Reserved 101=CR/CSR 110=User1 111=User2 15 Reserved 14 PGM Reserved Program/Data AM Code 0=Data 1=Program 13 Reserved 12 SUPER Reserved Supervisor/User AM Code 0=Non-Privileged 1=Supervisor 11:09 Reserved 08 VCT Reserved VMEbus Cycle Type 0=No BLTs on VMEbus 1=Single BLTs on VMEbus 07:01 Reserved 00 LAS Reserved PCI Bus Memory Space 0=PCI Bus Memory Space 1=PCI Bus I/O Space Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 12. Registers > Register Description 12.3.9 183 PCI Target Image 0 Base Address Register (LSI0_BS) The base address specifies the lowest address in the address range that will be decoded. The base address for PCI Target Image 0 and PCI Target Image 4 have a 4 Kbyte resolution. PCI Target Images 1, 2, 3, 5, 6, and 7 have a 64 Kbyte resolution. Register name: LSI0_BS Bits 7 Register offset: 0x104 6 5 4 3 31:24 BS 23:16 BS 15:08 2 1 BS 0 Reserved 07:00 Reserved Type Reset by Reset value Base Address R/W All Undefined BS[27:12] Base Address R/W All 0x00 Reserved Reserved R All 0x00 Bits Name 31:28 BS[31:28] 27:12 11:00 Integrated Device Technology www.idt.com Description Universe II User Manual May 12, 2010 184 12.3.10 12. Registers > Register Description PCI Target Image 0 Bound Address Register (LSI0_BD) The addresses decoded in a slave image are those which are greater than or equal to the base address and less than the bound register. If the bound address is 0, then the addresses decoded are those greater than or equal to the base address. The bound address for PCI Target Image 0 and PCI Target Image 4 have a 4Kbyte resolution. PCI Target Images 1, 2, 3, 5, 6, and 7 have a 64Kbyte resolution. Register name: LSI0_BD Bits 7 Register offset: 0x108 6 5 4 3 31:24 BD 23:16 BD 15:08 2 1 BD 0 Reserved 07:00 Reserved Type Reset by Reset value Bound Address R/W All Undefined BD[27:12] Bound Address R/W All 0x00 Reserved Reserved R All 0x00 Bits Name 31:28 BD[31:28] 27:12 11:00 Universe II User Manual May 12, 2010 Description Integrated Device Technology www.idt.com 12. Registers > Register Description 12.3.11 185 PCI Target Image 0 Translation Offset (LSI0_TO) The translation offset for PCI Target Image 0 and PCI Target Image 4 have a 4Kbyte resolution. PCI Target Images 1, 2, 3, 5, 6, and 7 have a 64Kbyte resolution. Address bits [31:12] generated on the VMEbus in response to an image decode are a two’s complement addition of address bits [31:12] on the PCI Bus and bits [31:12] of the image’s translation offset. Register name: LSI0_TO Bits 7 Register offset: 0x10C 6 5 4 3 31:24 TO 23:16 TO 15:08 2 1 TO 0 Reserved 07:00 Reserved Bits Name 31:12 TO[31:12] Translation Offset 11:00 Reserved Reserved Integrated Device Technology www.idt.com Description Type Reset by Reset value R/W All 0x00 R All 0x00 Universe II User Manual May 12, 2010 186 12.3.12 12. Registers > Register Description PCI Target Image 1 Control (LSI1_CTL) In the PCI Target Image Control register, setting the VCT bit will only have effect if the VAS bits are programmed for A24 or A32 space and the VDW bits are programmed for 8-bit, 16-bit, or 32-bit. If VAS bits are programmed to A24 or A32 and the VDW bits are programmed for 64-bit, the Universe II may perform MBLT transfers independent of the state of the VCT bit. The setting of the PWEN bit is ignored if the LAS bit is programmed for PCI Bus I/O Space, forcing all transactions through this image to be coupled. Register name: LSI1_CTL Register offset: 0x114 Bits 7 6 31:24 EN PWEN 23:16 15:08 5 4 2 1 Reserved PGM Reserved 07:00 SUPER VAS Reserved VCT Reserved Bits Name 31 EN 0 Reserved VDW Reserved 3 Description Image Enable LAS Type Reset by Reset value R/W All Undefined R/W All 0x00 R All 0x00 R/W All 0b10 R All 0x00 0=Disable 1=Enable 30 PWEN Posted Write Enable 0=Disable 1=Enable 29:24 Reserved 23:22 VDW Reserved VMEbus Maximum Data width 00=8-bit data width 01=16 bit data width 10=32-bit data width 11=64-bit data width 21:19 Reserved Universe II User Manual May 12, 2010 Reserved Integrated Device Technology www.idt.com 12. Registers > Register Description Bits Name 18:16 VAS 187 Description VMEbus Address Space Type Reset by Reset value R/W All Undefined R All 0x00 R/W All 0x00 R All 0x00 R/W All 0x00 R All 0x00 R/W All 0x00 R All 0x00 R/W All Undefined 000=A16 001=A24 010=A32 011= Reserved 100=Reserved 101=CR/CSR 110=User1 111=User2 15 Reserved 14 PGM Reserved Program/Data AM Code 0=Data 1=Program 13 Reserved 12 SUPER Reserved Supervisor/User AM Code 0=Non-Privileged 1=Supervisor 11:09 Reserved 08 VCT Reserved VMEbus Cycle Type 0=No BLTs on VMEbus 1=Single BLTs on VMEbus 07:01 Reserved 00 LAS Reserved PCI Bus Memory Space 0=PCI Bus Memory Space 1=PCI Bus I/O Space Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 188 12.3.13 12. Registers > Register Description PCI Target Image 1 Base Address Register (LSI1_BS) The base address specifies the lowest address in the address range that will be decoded. Register name: LSI1_BS Bits 7 Register offset: 0x118 6 5 4 3 31:24 BS 23:16 BS 15:08 Reserved 07:00 Reserved Bits Name 31:16 BS[31:16] Base Address 15:00 Reserved Reserved Universe II User Manual May 12, 2010 Description 2 1 0 Type Reset by Reset value R/W All 0x00 R All 0x00 Integrated Device Technology www.idt.com 12. Registers > Register Description 12.3.14 189 PCI Target Image 1 Bound Address Register (LSI1_BD) The addresses decoded in a slave image are those which are greater than or equal to the base address and less than the bound register. If the bound address is 0, then the addresses decoded are those greater than or equal to the base address. The bound address for PCI Target Image 0 and PCI Target Image 4 have a 4Kbyte resolution. PCI Target Images 1, 2, 3, 5, 6, and 7 have a 64Kbyte resolution. Register name: LSI1_BD Bits 7 Register offset: 0x11C 6 5 4 3 31:24 BD 23:16 BD 15:08 Reserved 07:00 Reserved Bits Name 31:16 BD[31:16] Bound Address 15:00 Reserved Reserved Integrated Device Technology www.idt.com Description 2 1 0 Type Reset by Reset value R/W All 0x00 R All 0x00 Universe II User Manual May 12, 2010 190 12.3.15 12. Registers > Register Description PCI Target Image 1 Translation Offset (LSI1_TO) Address bits [31:16] generated on the VMEbus in response to an image decode are a two’s complement addition of address bits [31:16] on the PCI Bus and bits [31:16] of the image’s translation offset. Register name: LSI1_TO Bits 7 Register offset: 0x120 6 5 4 3 31:24 TO 23:16 TO 15:08 Reserved 07:00 Reserved Bits Name 31:16 TO[31:16] Translation Offset 15:00 Reserved Reserved Universe II User Manual May 12, 2010 Description 2 1 0 Type Reset by Reset value R/W All 0x00 R All 0x00 Integrated Device Technology www.idt.com 12. Registers > Register Description 12.3.16 191 PCI Target Image 2 Control (LSI2_CTL) In the PCI Target Image Control register, setting the VCT bit will only have effect if the VAS bits are programmed for A24 or A32 space and the VDW bits are programmed for 8-bit, 16-bit, or 32-bit. If VAS bits are programmed to A24 or A32 and the VDW bits are programmed for 64-bit, the Universe II may perform MBLT transfers independent of the state of the VCT bit. The setting of the PWEN bit is ignored if the LAS bit is programmed for PCI Bus I/O Space, forcing all transactions through this image to be coupled. Register name: LSI2_CTL Register offset: 0x13C Bits 7 6 31:24 EN PWEN 23:16 15:08 5 4 2 1 Reserved PGM Reserved 07:00 SUPER VAS Reserved VCT Reserved Bits Name 31 EN 0 Reserved VDW Reserved 3 Description Image Enable LAS Type Reset by Reset value R/W All Undefined R/W All 0x00 R All 0x00 R/W All 0b10 R All 0x00 0=Disable 1=Enable 30 PWEN Posted Write Enable 0=Disable 1=Enable 29:24 Reserved 23:22 VDW Reserved VMEbus Maximum Data width 00=8-bit data width 01=16 bit data width 10=32-bit data width 11=64-bit data width 21:19 Reserved Integrated Device Technology www.idt.com Reserved Universe II User Manual May 12, 2010 192 12. Registers > Register Description Bits Name 18:16 VAS Description VMEbus Address Space Type Reset by Reset value R/W All Undefined R All 0x00 R/W All 0x00 R All 0x00 R/W All 0x00 R All 0x00 R/W All 0x00 R All 0x00 R/W All Undefined 000=A16 001=A24 010=A32 011= Reserved 100=Reserved 101=CR/CSR 110=User1 111=User2 15 Reserved 14 PGM Reserved Program/Data AM Code 0=Data 1=Program 13 Reserved 12 SUPER Reserved Supervisor/User AM Code 0=Non-Privileged 1=Supervisor 11:09 Reserved 08 VCT Reserved VMEbus Cycle Type 0=No BLTs on VMEbus 1=Single BLTs on VMEbus 07:01 Reserved 00 LAS Reserved PCI Bus Memory Space 0=PCI Bus Memory Space 1=PCI Bus I/O Space Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 12. Registers > Register Description 12.3.17 193 PCI Target Image 2 Base Address Register (LSI2_BS) The base address specifies the lowest address in the address range that will be decoded. Register name: LSI12_BS Bits 7 Register offset: 0x12C 6 5 4 3 31:24 BS 23:16 BS 15:08 Reserved 07:00 Reserved Bits Name 31:16 BS[31:16] Base Address 15:00 Reserved Reserved Integrated Device Technology www.idt.com Description 2 1 0 Type Reset by Reset value R/W All 0x00 R All 0x00 Universe II User Manual May 12, 2010 194 12.3.18 12. Registers > Register Description PCI Target Image 2 Bound Address Register (LSI2_BD) The addresses decoded in a slave image are those which are greater than or equal to the base address and less than the bound register. If the bound address is 0, then the addresses decoded are those greater than or equal to the base address. Register name: LSI2_BD Bits 7 Register offset: 0x130 6 5 4 3 31:24 BD 23:16 BD 15:08 Reserved 07:00 Reserved Bits Name 31:16 BD[31:16] Bound Address 15:00 Reserved Reserved Universe II User Manual May 12, 2010 Description 2 1 0 Type Reset by Reset value R/W All 0x00 R All 0x00 Integrated Device Technology www.idt.com 12. Registers > Register Description 12.3.19 195 PCI Target Image 2 Translation Offset (LSI2_TO) Address bits [31:16] generated on the VMEbus in response to an image decode are a two’s complement addition of address bits [31:16] on the PCI Bus and bits [31:16] of the image’s translation offset. Register name: LSI2_TO Bits 7 Register offset: 0x134 6 5 4 3 31:24 TO 23:16 TO 15:08 Reserved 07:00 Reserved Bits Name 31:16 TO[31:16] Translation Offset 15:00 Reserved Reserved Integrated Device Technology www.idt.com Description 2 1 0 Type Reset by Reset value R/W All 0x00 R All 0x00 Universe II User Manual May 12, 2010 196 12.3.20 12. Registers > Register Description PCI Target Image 3 Control (LSI3_CTL) In the PCI Target Image Control register, setting the VCT bit will only have effect if the VAS bits are programmed for A24 or A32 space and the VDW bits are programmed for 8-bit, 16-bit, or 32-bit. If VAS bits are programmed to A24 or A32 and the VDW bits are programmed for 64-bit, the Universe II may perform MBLT transfers independent of the state of the VCT bit. The setting of the PWEN bit is ignored if the LAS bit is programmed for PCI Bus I/O Space, forcing all transactions through this image to be coupled. Register name: LSI3_CTL Register offset: 0x13C Bits 7 6 31:24 EN PWEN 23:16 15:08 5 4 2 1 Reserved PGM Reserved 07:00 SUPER VAS Reserved VCT Reserved Bits Name 31 EN 0 Reserved VDW Reserved 3 Description Image Enable LAS Type Reset by Reset value R/W All Undefined R/W All 0x00 R All 0x00 R/W All 10 R All 0x00 0=Disable 1=Enable 30 PWEN Posted Write Enable 0=Disable 1=Enable 29:24 Reserved 23:22 VDW Reserved VMEbus Maximum Data width 00=8-bit data width 01=16 bit data width 10=32-bit data width 11=64-bit data width 21:19 Reserved Universe II User Manual May 12, 2010 Reserved Integrated Device Technology www.idt.com 12. Registers > Register Description Bits Name 18:16 VAS 197 Description VMEbus Address Space Type Reset by Reset value R/W All Undefined R All 0x00 R/W All 0x00 R All 0x00 R/W All 0x00 R All 0x00 R/W All 0x00 R All 0x00 R/W All Undefined 000=A16 001=A24 010=A32 011= Reserved 100=Reserved 101=CR/CSR 110=User1 111=User2 15 Reserved 14 PGM Reserved Program/Data AM Code 0=Data 1=Program 13 Reserved 12 SUPER Reserved Supervisor/User AM Code 0=Non-Privileged 1=Supervisor 11:09 Reserved 08 VCT Reserved VMEbus Cycle Type 0=No BLTs on VMEbus 1=Single BLTs on VMEbus 07:01 Reserved 00 LAS Reserved PCI Bus Memory Space 0=PCI Bus Memory Space 1=PCI Bus I/O Space Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 198 12.3.21 12. Registers > Register Description PCI Target Image 3 Base Address Register (LSI3_BS) The base address specifies the lowest address in the address range that will be decoded. Register name: LSI3_BS Bits 7 Register offset: 0x140 6 5 4 3 31:24 BS 23:16 BS 15:08 Reserved 07:00 Reserved Bits Name 31:16 BS[31:16] Base Address 15:00 Reserved Reserved Universe II User Manual May 12, 2010 Description 2 1 0 Type Reset by Reset value R/W All 0x00 R All 0x00 Integrated Device Technology www.idt.com 12. Registers > Register Description 12.3.22 199 PCI Target Image 3 Bound Address Register (LSI3_BD) The addresses decoded in a slave image are those which are greater than or equal to the base address and less than the bound register. If the bound address is 0, then the addresses decoded are those greater than or equal to the base address. Register name: LSI3_BD Bits 7 Register offset: 0x144 6 5 4 3 31:24 BD 23:16 BD 15:08 Reserved 07:00 Reserved Bits Name 31:16 BD[31:16] Bound Address 15:00 Reserved Reserved Integrated Device Technology www.idt.com Description 2 1 0 Type Reset by Reset value R/W All 0x00 R All 0x00 Universe II User Manual May 12, 2010 200 12.3.23 12. Registers > Register Description PCI Target Image 3 Translation Offset (LSI3_TO) Address bits [31:16] generated on the VMEbus in response to an image decode are a two’s complement addition of address bits [31:16] on the PCI Bus and bits [31:16] of the image’s translation offset. Register name: LSI3_TO Bits 7 Register offset: 0x148 6 5 4 3 31:24 TO 23:16 TO 15:08 Reserved 07:00 Reserved Bits Name 31:16 TO[31:16] Translation Offset 15:00 Reserved Reserved Universe II User Manual May 12, 2010 Description 2 1 0 Type Reset by Reset value R/W All 0x00 R All 0x00 Integrated Device Technology www.idt.com 12. Registers > Register Description 12.3.24 201 Special Cycle Control Register (SCYC_CTL) The special cycle generator generates an ADOH or RMW cycle for the 32-bit PCI Bus address which matches the programmed address in SCYC_ADDR, in the address space specified in the LAS field of the SCYC_CTL register. A Read-Modify-Write command is initiated by a read to the specified address. Address-Only cycles are initiated by either read or write cycles. Register Name: SCYC_CTL Bits 7 Register offset: 0x170 6 5 4 3 31:24 Reserved 23:16 Reserved 15:08 Reserved 07:00 Reserved Bits Name 31:03 Reserved 02 LAS Description Reserved PCI Bus Address Space 2 1 LAS 0 SCYC Type Reset by Reset value R All 0x00 R/W All 0x00 R/W All 0x00 0=PCI Bus Memory Space 1=PCI Bus I/O Space For a RMW cycle only 01:00 SCYC Special Cycle 00=Disable 01=RMW 10=ADOH 11=Reserved Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 202 12.3.25 12. Registers > Register Description Special Cycle PCI Bus Address Register (SCYC_ADDR) This register designates the special cycle address. This address must appear on the PCI Bus during the address phase of a transfer for the Special Cycle Generator to perform its function. Whenever the addresses match, the Universe II does not respond with ACK64_. Register name: SCYC_ADDR Bits 7 6 Register offset: 0x174 5 4 3 31:24 ADDR 23:16 ADDR 15:08 ADDR 07:00 2 1 ADDR Bits Name 31:02 ADDR 01:00 Reserved Universe II User Manual May 12, 2010 0 Reserved Type Reset by Reset value Address R/W All 0x00 Reserved R/W All 0x00 Description Integrated Device Technology www.idt.com 12. Registers > Register Description 12.3.26 203 Special Cycle Swap/Compare Enable Register (SCYC_EN) The bits enabled in this register determine the bits that are involved in the compare and swap operations for VME RMW cycles. Register name: SCYC_EN Bits 7 Register offset: 0x178 6 5 4 3 31:24 EN 23:16 EN 15:08 EN 07:00 EN Bits Name 31:00 EN Description Bit Enable 2 1 0 Type Reset by Reset value R/W All 0x00 0=Disable 1=Enable Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 204 12.3.27 12. Registers > Register Description Special Cycle Compare Data Register (SCYC_CMP) The data returned from the read portion of a VMEbus RMW is compared with the contents of this register. SCYC_EN is used to control which bits are compared. Register name: SCYC_CMP Bits 7 Register offset: 0x17C 6 5 4 3 31:24 CMP 23:16 CMP 15:08 CMP 07:00 CMP 2 1 0 Bits Name Description Type Reset by Reset value 31:00 CMP The data returned from the VMEbus is compared with the contents of this register. R/W All 0x00 Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 12. Registers > Register Description 12.3.28 205 Special Cycle Swap Data Register (SCYC_SWP) If enabled bits matched with the value in the compare register, then the contents of the swap data register is written back to VME. SCYC_EN is used to control which bits are written back to VME. Register name: SCYC_SWP Bits 7 Register offset: 0x180 6 5 4 3 31:24 SWP 23:16 SWP 15:08 SWP 07:00 SWP Bits Name 31:00 SWP Integrated Device Technology www.idt.com Description Swap data 2 1 0 Type Reset by Reset value R/W All 0x00 Universe II User Manual May 12, 2010 206 12. Registers > Register Description 12.3.29 PCI Miscellaneous Register (LMISC) This register can only be set at configuration or after disabling all PCI Target Images. Register name: LMISC Bits 7 Register offset: 0x184 6 31:24 5 4 3 CRT[3:0] Reserved 15:08 Reserved 07:00 Reserved CRT[3:0] 1 Reserved 23:16 Name 2 Description CRT 0 CWT Type Reset By Reset State R/W all 000 R/W all 000 This field is provided for backward compatibility with the Universe I. It has no effect on the operation of the Universe II. CWT [2:0] Coupled Window Timer 000=Disable - release after first coupled transaction 001=16 PCI Clocks 010=32 PCI Clocks 011=64 PCI Clocks 100=128 PCI Clocks 101=256 PCI Clocks 110=512 PCI Clocks others= Reserved Bits Name Description Type Reset by Reset value 31:28 CRT This field is provided for backward compatibility with the Universe I. It has no effect on the operation of the Universe II. R/W All 0x00 27 Reserved R All 0x00 Universe II User Manual May 12, 2010 Reserved Integrated Device Technology www.idt.com 12. Registers > Register Description Bits Name 26:24 CWT 207 Description Coupled Window Timer Type Reset by Reset value R/W All 0x00 R All 0x00 000=Disable - release after first coupled transaction 001=16 PCI Clocks 010=32 PCI Clocks 011=64 PCI Clocks 100=128 PCI Clocks 101=256 PCI Clocks 110=512 PCI Clocks others= Reserved 23:00 Reserved Reserved The Universe II uses CWT to determine how long to hold ownership of the VMEbus after processing a coupled transaction. The timer is restarted each time the Universe II processes a coupled transaction. If this timer expires, the PCI Slave Channel releases the VMEbus. Device behavior is unpredictable if CWT is changed during coupled cycle activity. Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 208 12. Registers > Register Description 12.3.30 Special PCI Target Image (SLSI) This register fully specifies an A32 capable special PCI Target Image. The base is programmable to a 64 Mbyte alignment, and the size is fixed at 64 Mbytes. Incoming address lines [31:26] (in Memory or I/O) must match this field for the Universe II to decode the access. This special PCI Target Image has lower priority than any other PCI Target Image. The 64 Mbytes of the SLSI is partitioned into four 16 Mbyte regions, numbered 0 to 3 (0 is at the lowest address). PCI address bits [25:24] are used to select regions. The top 64 Kbyte of each region is mapped to VMEbus A16 space, and the rest of each 16 Mbyte region is mapped to A24 space. The user can use the PGM, SUPER, and VDW fields to specify the AM code and the maximum port size for each region. The PGM field is ignored for the portion of each region mapped to A16 space. No block transfer AM codes are generated. Register name: SLSI Register offset: 0x188 Bits 7 6 31:24 EN PWN 5 4 3 2 1 0 Reserved LAS Reserved 23:16 VDW Reserved 15:08 PGM SUPER 07:00 BS Name EN Description Image Enable Type Reset by Reset value R/W All 0 R/W All 0 R/W All 0 0=Disable 1=Enable PWEN Posted Write Enable 0=Disable 1=Enable VDW VMEbus Maximum Datawidth Each of the four bits specifies a data width for the corresponding 16 MByte region. Low order bits correspond to the lower address regions. 0=16-bit 1=32-bit Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 12. Registers > Register Description Name PGM 209 Description Program/Data AM Code Type Reset by Reset value R/W All 0 R/W All 0 R/W All 0 R/W All 0 Each of the four bits specifies Program/Data AM code for the corresponding 16 MByte region. Low order bits correspond to the lower address regions. 0=Data 1=Program SUPER Supervisor/User AM Code Each of the four bits specifies Supervisor/User AM code for the corresponding 16 MByte region. Low order bits correspond to the lower address regions. 0=Non-Privileged 1=Supervisor BS Base Address Specifies a 64 MByte aligned base address for this 64 MByte image. LAS PCI Bus Address Space 0=PCI Bus Memory Space 1=PCI Bus I/O Space Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 210 12. Registers > Register Description 12.3.31 PCI Command Error Log Register (L_CMDERR) The Universe II PCI Master Interface is responsible for logging errors under the following conditions: • Posted write transaction results in a target abort, • Posted write transaction results in a master abort, or • Maximum retry counter expires during retry of posted write transaction. This register logs the command information. Register name: L_CMDERR Bits 7 Register offset: 0x18C 6 31:24 23:16 5 4 CMDERR L_STAT 3 M_ERR Reserved 07:00 Reserved CMDERR M_ERR 1 0 Reserved Reserved 15:08 Name 2 Description Type Reset by Reset value PCI Command Error Log R All 0111 Multiple Error Occurred R All 0 R/W All 0 0=Single error 1=At least one error has occurred since the logs were frozen. L_STAT PCI Error Log Status Reads: 0=logs invalid 1=logs are valid and error logging halted Writes: 0=no effect 1=clears L_STAT and enables error logging Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 12. Registers > Register Description 12.3.32 211 PCI Address Error Log (LAERR) The starting address of an errored PCI transaction is logged in this register under the following conditions: • a posted write transaction results in a target abort, • a posted write transaction results in a master abort, or • a maximum retry counter expires during retry of posted write transaction. Contents are qualified by bit L_STAT of the L_CMDERR register. Register name: LAERR Bits 7 Register offset: 0x190 6 5 4 3 31:24 LAERR 23:16 LAERR 15:08 LAERR 07:00 1 LAERR Name LAERR 2 Description PCI Address Error Log Integrated Device Technology www.idt.com 0 Reserved Type Reset by Reset value R All 0 Universe II User Manual May 12, 2010 212 12. Registers > Register Description 12.3.33 PCI Target Image 4 Control Register (LSI4_CTL) In the PCI Target Image Control register, setting the VCT bit will only have effect if the VAS bits are programmed for A24 or A32 space and the VDW bits are programmed for 8-bit, 16-bit, or 32-bit. If VAS bits are programmed to A24 or A32 and the VDW bits are programmed for 64-bit, the Universe II may perform MBLT transfers independent of the state of the VCT bit. The setting of the PWEN bit is ignored if the LAS bit is programmed for PCI Bus I/O Space, forcing all transactions through this image to be coupled. Register name: LSI4_CTL Register offset: 0x1A0 Bits 7 6 31:24 EN PWEN 23:16 15:08 5 2 Reserved PGM Reserved 07:00 1 0 SUPER VAS Reserved VCT Reserved Name EN 3 Reserved VDW Reserved 4 Description Image Enable LAS Type Reset by Reset value R/W All 0 R/W All 0 R/W All 10 R/W All 0 0=Disable 1=Enable PWEN Posted Write Enable 0=Disable 1=Enable VDW VMEbus Maximum Datawidth 00=8-bit data width 01=16 bit data width 10=32-bit data width 11=64-bit data width VAS VMEbus Address Space 000=A16 001=A24 010=A32 011= Reserved 100=Reserved 101=CR/CSR 110=User1 111=User2 Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 12. Registers > Register Description Name PGM 213 Description Program/Data AM Code Type Reset by Reset value R/W All 0 R/W All 0 R/W All 0 R/W All 0 0=Data 1=Program SUPER Supervisor/User AM Code 0=Non-Privileged 1=Supervisor VCT VMEbus Cycle Type 0=no BLTs on VMEbus 1=BLTs on VMEbus LAS PCI Bus Memory Space 0=PCI Bus Memory Space 1=PCI Bus I/O Space Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 214 12. Registers > Register Description 12.3.34 PCI Target Image 4 Base Address Register (LSI4_BS) The base address specifies the lowest address in the address range that will be decoded. Register name: LSI4_BS Bits 7 Register offset: 0x1A4 6 5 4 3 31:24 BS 23:16 BS 15:08 BS 1 0 Reserved 07:00 Reserved Name BS 2 Description Base Address Universe II User Manual May 12, 2010 Type Reset by Reset value R/W All 0 Integrated Device Technology www.idt.com 12. Registers > Register Description 12.3.35 215 PCI Target Image 4 Bound Address Register (LSI4_BD) The addresses decoded in a slave image are those which are greater than or equal to the base address and less than the bound register. If the bound address is 0, then the addresses decoded are those greater than or equal to the base address. The bound address for PCI Target Image 0 and PCI Target Image 4 have a 4Kbyte resolution. PCI Target Images 1, 2, 3, 5, 6, and 7 have a 64Kbyte resolution. Register name: LSI4_BD Bits 7 Register offset: 0x1A8 6 5 4 3 31:24 BD 23:16 BD 15:08 BD 1 0 Reserved 07:00 Reserved Name BD 2 Description Bound Address Integrated Device Technology www.idt.com Type Reset by Reset value R/W All 0 Universe II User Manual May 12, 2010 216 12. Registers > Register Description 12.3.36 PCI Target Image 4 Translation Offset (LSI4_TO) Address bits [31:12] generated on the VMEbus in response to an image decode are a two’s complement addition of address bits [31:12] on the PCI Bus and bits [31:12] of the image’s translation offset. Register name: LSI4_TO Bits 7 Register offset: 0x1AC 6 5 4 3 31:24 TO 23:16 TO 15:08 TO 1 0 Reserved 07:00 Reserved Name TO 2 Description Translation offset Universe II User Manual May 12, 2010 Type Reset by Reset value R/W All 0 Integrated Device Technology www.idt.com 12. Registers > Register Description 12.3.37 217 PCI Target Image 5 Control Register (LSI5_CTL) In the PCI Target Image Control register, setting the VCT bit will only have effect if the VAS bits are programmed for A24 or A32 space and the VDW bits are programmed for 8-bit, 16-bit, or 32-bit. If VAS bits are programmed to A24 or A32 and the VDW bits are programmed for 64-bit, the Universe II may perform MBLT transfers independent of the state of the VCT bit. The setting of the PWEN bit is ignored if the LAS bit is programmed for PCI Bus I/O Space, forcing all transactions through this image to be coupled. Register name: LSI5_CTL Register offset: 0x1B4 Bits 7 6 31:24 EN PWEN 23:16 5 Reserved 3 2 Reserved PGM 1 0 Reserved VDW 15:08 4 Reserved 07:00 SUPER VAS Reserved VCT Reserved Name EN Description Image Enable LAS Type Reset by Reset value R/W All 0 R/W All 0 R/W All 10 R/W All 0 0=Disabl 1=Enable PWEN Posted Write Enable 0=Disable 1=Enable VDW VMEbus Maximum Datawidth 00=8-bit data width 01=16 bit data width 10=32-bit data width 11=64-bit data width VAS VMEbus Address Space 000=A16 001=A24 010=A32 011= Reserved 100=Reserved 101=CR/CSR 110=User1 111=User2 Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 218 12. Registers > Register Description Name Description Type Reset by Reset value PGM Program/Data AM Code R/W All 0 R/W All 0 R/W All 0 R/W All 0 0=Data 1=Program SUPER Supervisor/User AM Code 0=Non-Privileged 1=Supervisor VCT VMEbus Cycle Type 0=no BLTs on VMEbus 1=BLTs on VMEbus LAS PCI Bus Memory Space 0=PCI Bus Memory Space 1=PCI Bus I/O Space Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 12. Registers > Register Description 12.3.38 219 PCI Target Image 5 Base Address Register (LSI5_BS) The base address specifies the lowest address in the address range that will be decoded. Register name: LSI5_BS Bits Function 31:24 BS 23:16 BS 15:08 Reserved 07:00 Reserved Name BS Register offset: 0x1B8 Description Base Address Integrated Device Technology www.idt.com Type Reset by Reset value R/W All 0 Universe II User Manual May 12, 2010 220 12. Registers > Register Description 12.3.39 PCI Target Image 5 Bound Address Register (LSI5_BD) The addresses decoded in a slave image are those which are greater than or equal to the base address and less than the bound register. If the bound address is 0, then the addresses decoded are those greater than or equal to the base address. Register name: LSI5_BD Register offset: 0x1BC Bits Function 31:24 BD 23:16 BD 15:08 Reserved 07:00 Reserved Name BD Description Bound Address Type Reset by Reset value R/W All 0 The bound address for PCI Target Image 0 and PCI Target Image 4 have a 4Kbyte resolution. PCI Target Images 1, 2, 3, 5, 6, and 7 have a 64Kbyte resolution. Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 12. Registers > Register Description 12.3.40 221 PCI Target Image 5 Translation Offset (LSI5_TO) Address bits [31:16] generated on the VMEbus in response to an image decode are a two’s complement addition of address bits [31:16] on the PCI Bus and bits [31:16] of the image’s translation offset. Register name: LSI5_TO Register offset: 0x1C0 Bits Function 31:24 TO 23:16 TO 15:08 Reserved 07:00 Reserved Name TO Description Translation offset Integrated Device Technology www.idt.com Type Reset by Reset value R/W All 0 Universe II User Manual May 12, 2010 222 12. Registers > Register Description 12.3.41 PCI Target Image 6 Control Register (LSI6_CTL) In the PCI Target Image Control register, setting the VCT bit will only have effect if the VAS bits are programmed for A24 or A32 space and the VDW bits are programmed for 8-bit, 16-bit, or 32-bit. If VAS bits are programmed to A24 or A32 and the VDW bits are programmed for 64-bit, the Universe II may perform MBLT transfers independent of the state of the VCT bit. The setting of the PWEN bit is ignored if the LAS bit is programmed for PCI Bus I/O Space, forcing all transactions through this image to be coupled. Register name: LSI6_CTL Register offset: 0x1C8 Bits 7 6 31:24 EN PWEN 23:16 15:08 5 2 1 Reserved PGM Reserved 07:00 0 SUPER VAS Reserved VCT Reserved Name EN 3 Reserved VDW Reserved 4 Description Image Enable LAS Type Reset by Reset value R/W All 0 R/W All 0 R/W All 10 R/W All 0 0=Disable 1=Enable PWEN Posted Write Enable 0=Disable 1=Enable VDW VMEbus Maximum Datawidth 00=8-bit data width 01=16 bit data width 10=32-bit data width 11=64-bit data width VAS VMEbus Address Space 000=A16 001=A24 010=A32 011= Reserved 100=Reserved 101=CR/CSR 110=User1 111=User2 Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 12. Registers > Register Description Name PGM 223 Description Program/Data AM Code Type Reset by Reset value R/W All 0 R/W All 0 R/W All 0 R/W All 0 0=Data 1=Program SUPER Supervisor/User AM Code 0=Non-Privileged 1=Supervisor VCT VMEbus Cycle Type 0=no BLTs on VMEbus 1=BLTs on VMEbus LAS PCI Bus Memory Space 0=PCI Bus Memory Space 1=PCI Bus I/O Space Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 224 12. Registers > Register Description 12.3.42 PCI Target Image 6 Base Address Register (LSI6_BS) The base address specifies the lowest address in the address range that will be decoded. Register name: LSI6_BS Register offset: 0x1CC Bits Function 31:24 BS 23:16 BS 15:08 Reserved 07:00 Reserved Name BS Description Base Address Universe II User Manual May 12, 2010 Type Reset by Reset value R/W All 0 Integrated Device Technology www.idt.com 12. Registers > Register Description 12.3.43 225 PCI Target Image 6 Bound Address Register (LSI6_BD) The addresses decoded in a slave image are those which are greater than or equal to the base address and less than the bound register. If the bound address is 0, then the addresses decoded are those greater than or equal to the base address. Register name: LSI6_BD Register offset: 0x1D0 Bits Function 31:24 BD 23:16 BD 15:08 Reserved 07:00 Reserved Name BD Description Bound Address Type Reset by Reset value R/W All 0 The bound address for PCI Target Image 0 and PCI Target Image 4 have a 4Kbyte resolution. PCI Target Images 1, 2, 3, 5, 6, and 7 have a 64Kbyte resolution. Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 226 12. Registers > Register Description 12.3.44 PCI Target Image 6 Translation Offset (LSI6_TO) Address bits [31:16] generated on the VMEbus in response to an image decode are a two’s complement addition of address bits [31:16] on the PCI Bus and bits [31:16] of the image’s translation offset. Register name: LSI6_TO Register offset: 0x1D4 Bits Function 31:24 TO 23:16 TO 15;08 Reserved 07:00 Reserved Name TO Description Translation offset Universe II User Manual May 12, 2010 Type Reset by Reset value R/W All 0 Integrated Device Technology www.idt.com 12. Registers > Register Description 12.3.45 227 PCI Target Image 7 Control Register (LSI7_CTL) In the PCI Target Image Control register, setting the VCT bit will only have effect if the VAS bits are programmed for A24 or A32 space and the VDW bits are programmed for 8-bit, 16-bit, or 32-bit. If VAS bits are programmed to A24 or A32 and the VDW bits are programmed for 64-bit, the Universe II may perform MBLT transfers independent of the state of the VCT bit. The setting of the PWEN bit is ignored if the LAS bit is programmed for PCI Bus I/O Space, forcing all transactions through this image to be coupled. Register name: LSI7_CTL Register offset: 0x1DC Bits 7 6 31:24 EN PWEN 23;16 15:08 5 2 Reserved PGM Reserved 07:00 1 0 SUPER VAS Reserved VCT Reserved Name EN 3 Reserved VDW Reserved 4 Description Image Enable LAS Type Reset by Reset value R/W All 0 R/W All 0 R/W All 10 R/W All 0 R/W All 0 R/W All 0 0=Disable 1=Enable PWEN Posted Write Enable 0=Disable 1=Enable VDW VMEbus Maximum Datawidth 00=8-bit data width 01=16 bit data width 10=32-bit data width 11=64-bit data width VAS VMEbus Address Space 000=A16, 001=A24, 010=A32, 011= Reserved, 100=Reserved, 101=CR/CSR, 110=User1, 111=User2 PGM Program/Data AM Code 0=Data 1=Program SUPER Supervisor/User AM Code 0=Non-Privileged 1=Supervisor Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 228 12. Registers > Register Description Name VCT Description VMEbus Cycle Type Type Reset by Reset value R/W All 0 R/W All 0 0=no BLTs on VMEbus 1=BLTs on VMEbus LAS PCI Bus Memory Space 0=PCI Bus Memory Space 1=PCI Bus I/O Space Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 12. Registers > Register Description 12.3.46 229 PCI Target Image 7 Base Address Register (LSI7_BS) The base address specifies the lowest address in the address range that is decoded. Register name: LSI7_BS Register offset: 0x1E0 Bits Function 31:24 BS 23:16 BS 15:08 Reserved 07:00 Reserved Name BS Description Base Address Integrated Device Technology www.idt.com Type Reset by Reset value R/W All 0 Universe II User Manual May 12, 2010 230 12. Registers > Register Description 12.3.47 PCI Target Image 7 Bound Address Register (LSI7_BD) The addresses decoded in a slave image are those which are greater than or equal to the base address and less than the bound register. If the bound address is 0, then the addresses decoded are those greater than or equal to the base address. Register name: LSI7_BD Register offset: 0x1E4 Bits Function 31:24 BD 23:16 BD 15:08 Reserved 07:00 Reserved Name BD Description Bound Address Type Reset by Reset value R/W All 0 The bound address for PCI Target Image 0 and PCI Target Image 4 have a 4Kbyte resolution. PCI Target Images 1, 2, 3, 5, 6, and 7 have a 64Kbyte resolution. Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 12. Registers > Register Description 12.3.48 231 PCI Target Image 7 Translation Offset (LSI7_TO) Address bits [31:16] generated on the VMEbus in response to an image decode are a two’s complement addition of address bits [31:16] on the PCI Bus and bits [31:16] of the image’s translation offset. Register name: LSI7_TO Register offset: 0x1E8 Bits Function 31:24 TO 23:16 TO 15:08 Reserved 07:00 Reserved Name TO Description Translation offset Integrated Device Technology www.idt.com Type Reset by Reset value R/W All 0 Universe II User Manual May 12, 2010 232 12. Registers > Register Description 12.3.49 DMA Transfer Control Register (DCTL) This register is programmed from either bus or is programmed by the DMAC when it loads the command packet. The DMA only accesses PCI Bus Memory space. The VCT bit determines whether or not the Universe II VME Master will generate BLT transfers. The value of this bit only has meaning if the address space is A24 or A32 and the data width is not 64 bits. If the data width is 64 bits the Universe II may perform MBLT transfers independent of the state of the VCT bit. Register name: DCTL Bits 7 31:24 L2V Register offset: 0x200 6 5 4 3 2 1 0 Reserved 23:16 VDW 15:08 PGM Reserved SUPER VAS Reserved NO_ VCT VINC 07:00 LD64EN Name L2V Reserved Description Direction Type Reset by Reset value R/W All 0 R/W All 0 R/W All 0 R/W All 0 0=Transfer from VMEbus to PCI Bus 1=Transfer from PCI Bus to VMEbus VDW VMEbus Maximum Datawidth 00=8-bit data width 01=16 bit data width 10=32-bit data width 11=64-bit data width VAS VMEbus Address Space 000=A16 001=A24 010=A32 011= Reserved 100=Reserved 101=Reserved 110=User1 111=User2 PGM Program/Data AM Code 00=Data 01=Program others= Reserved Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 12. Registers > Register Description Name SUPER 233 Description Supervisor/User AM Code Type Reset by Reset value R/W All 0 R/W All 0 R/W All 0 R/W All 1 00=Non-Privileged 01=Supervisor others= Reserved NO_VINC VMEbus Non-Incrementing Mode (Non-Inc Mode) 0=disabled 1=enabled VCT VMEbus Cycle Type 0=no BLTs on VMEbus 1=BLTs on VMEbus LD64EN Enable 64-bit PCI Bus Transactions 0=Disable 1=Enable Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 234 12. Registers > Register Description 12.3.50 DMA Transfer Byte Count Register (DTBC) This register specifies the number of bytes to be moved by the DMA before the start of the DMA transfer, or the number of remaining bytes in the transfer while the DMA is active. This register is programmed from either bus or is programmed by the DMA Controller when it loads a command packet from a linked-list. Register name: DTBC Register offset: 0x204 Bits Function 31:24 Reserved 23:16 DTBC 15:08 DTBC 07:00 DTBC Name DTBC Description DMA Transfer Byte Count Type Reset by Reset value R/W All 0 In direct mode the user must re-program the DTBC register before each transfer. When using the DMA to perform linked-list transfers, it is essential that the DTBC register contains a value of zero before setting the GO bit of the DGCS register or undefined behaviors may occur. Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 12. Registers > Register Description 12.3.51 235 DMA PCI Bus Address Register (DLA) This register is programmed from either bus or by the DMA Controller when it loads a command packet. In direct mode the user must reprogram the DLA register before each transfer. In linked-list mode, this register is only updated when the DMA is stopped, halted, or at the completion of processing a command packet. Register name: DLA Register offset: 0x208 Bits Function 31:24 LA 23:16 LA 15:08 LA 07:00 LA Name Description Type Reset by Reset value LA[31:3] PCI Bus Address R/W All 0 LA[2:0] PCI Bus Address R/W All 0 Address bits [2:0] must be programmed the same as those in the DVA. After a Bus Error, a Target-Abort, or a Master-Abort, the value in the DLA register must not be used to reprogram the DMA because it has no usable information. Some offset from its original value must be used. Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 236 12. Registers > Register Description 12.3.52 DMA VMEbus Address Register (DVA) This register is programmed from either bus or is programmed by the DMA Controller when it loads a command packet. In direct mode the user must reprogram the DVA register before each transfer. In linked-list operation, this register is only updated when the DMA is stopped, halted, or at the completion of processing a command packet. Register name: DVA Register offset: 0x210 Bits Function 31:24 VA 23:16 VA 15:08 VA 07:00 VA Name VA Description VMEbus Address Type Reset by Reset value R/W All 0 Address bits [2:0] must be programmed the same as those in the DLA. After a bus error, a Target-Abort, or a Master-Abort, the value in the DVA register must not be used to reprogram the DMA because it has no usable information. Some offset from its original value must be used. Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 12. Registers > Register Description 12.3.53 237 DMA Command Packet Pointer (DCPP) This register contains the pointer into the current command packet. Initially it is programmed to the starting packet of the linked-list, and is updated with the address to a new command packet at the completion of a packet. The packets must be aligned to a 32-byte address. Register name: DCPP Bits 7 Register offset: 0x218 6 5 4 3 31:24 DCPP 23:16 DCPP 15:08 DCPP 07:00 DCPP Name DCPP [31:5] 1 0 Reserved Description DMA Command Packet Pointer Integrated Device Technology www.idt.com 2 Type Reset by Reset value R/W All 0 Universe II User Manual May 12, 2010 238 12. Registers > Register Description 12.3.54 DMA General Control/Status Register (DGCS) Register name: DGCS Register offset: 0x220 Bits 7 6 5 4 3 2 1 0 31:24 GO STOP_ HALT_ 0 CHAIN 0 0 0 REQ REQ VERR P_ERR 23:16 Reserved 15:08 ACT 07:00 0 VON STOP HALT 0 INT_ INT_ 0 STOP HALT Name GO VOFF DONE LERR INT_ INT_ INT_ INT_ DONE LERR VERR P_ERR Description DMA Go Bit 0=No effect Type Reset by Reset value W/Read 0 always All 0 W/Read 0 always All 0 W/Read 0 always All 0 R/W All 0 R/W All 0 1=Enable DMA Transfers STOP_ REQ DMA Stop Request 0=No effect 1=Stop DMA transfer when all buffered data has been written HALT_ REQ DMA Halt Request 0=No effect 1=Halt the DMA transfer at the completion of the current command packet CHAIN DMA Chaining 0=DMA Direct Mode 1=DMA Linked List mode VON [2:0] VMEbus “On” counter 000=Until done 001=256 bytes 010=512 bytes 011=1024 bytes 100=2048 bytes 101=4096 bytes 110=8192 bytes 111=16384 bytes others= Reserved Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 12. Registers > Register Description Name VOFF [3:0] 239 Description VMEbus “Off” Counter Type Reset by Reset value R/W All 0 R All 0 R/Write 1 to Clear All 0 R/Write 1 to Clear All 0 R/Write 1 to Clear All 0 R/Write 1 to Clear All 0 R/Write 1 to Clear All 0 R/Write 1 to Clear All 0 0000=0µs 0001=16µs 0010=32µs 0011=64µs 0100=128µs 0101=256µs 0110=512µs 0111=1024µs 1000=2µs 1001=4µs 1010=8µs others= Reserved The DMA will not re-request the VME Master until this timer expires. ACT DMA Active Status Bit 0=Not Active 1=Active STOP DMA Stopped Status Bit 0=Not Stopped 1=Stopped HALT DMA Halted Status Bit 0=Not Halted 1=Halted DONE DMA Done Status Bit 0=Not Complete 1=Complete LERR DMA PCI Bus Error Status Bit 0=No Error 1=Error VERR DMA VMEbus Error Status Bit 0=No Error 1=Error P_ERR DMA Programming Protocol Error Status Bit Asserted if PCI master interface disabled or lower three bits of PCI and VME addresses differ 0=No Error 1=Error Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 240 12. Registers > Register Description Name INT_STOP Description Interrupt when Stopped Type Reset by Reset value R/W All 0 R/W All 0 R/W All 0 R/W All 0 R/W All 0 R/W All 0 0=Disable 1=Enable INT_HALT Interrupt when Halted 0=Disable 1=Enable INT_ DONE Interrupt when Done 0=Disable 1=Enable INT_LERR Interrupt on LERR 0=Disable 1=Enable INT_VERR Interrupt on VERR 0=Disable 1=Enable INT_P_ ERR Interrupt on Master Enable Error 0=Disable 1=Enable STOP, HALT, DONE, LERR, VERR, and P_ERR must be cleared before the GO bit is enabled. Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 12. Registers > Register Description 12.3.55 241 DMA Linked List Update Enable Register (D_LLUE) The PCI Resource must read back a logic 1 in the UPDATE field before proceeding to modify the linked list. After the Linked List has been modified the PCI Resource must clear the UPDATE field by writing a logic 0. The Universe II does not prevent an external master, from the PCI bus or the VMEbus, from writing to the other DMA registers (see “Linked-list Updating” on page 100). Register name: D_LLUE Bits 7 31:24 UPDATE Register offset: 0x224 6 5 4 3 Reserved 15:08 Reserved 07:00 Reserved UPDATE 1 0 Reserved 23:16 Name 2 Description DMA Linked List Update Enable Type Reset by Reset value R/W All 0 0=PCI Resource not Updating Linked List 1=PCI Resource Updating Linked List Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 242 12. Registers > Register Description 12.3.56 PCI Interrupt Enable Register (LINT_EN) Bits VIRQ7-VIRQ1 enable the Universe II to respond as a VME Interrupt Handler to interrupts on the VIRQ[x] lines. When a VIRQx interrupt is enabled, and the corresponding VIRQ[x] pin is asserted, the Universe II requests the VMEbus and performs a VME IACK cycle for that interrupt level. When the interrupt acknowledge cycle completes, the STATUS/ID is stored in the corresponding VINT_ID register, the VIRQx bit of the LINT_STAT register is set, and a PCI interrupt is generated. The Universe II does not acquire further interrupt STATUS/ID vectors at the same interrupt level until the VIRQx bit in the LINT_STAT register is cleared. The other bits enable the respective internal or external sources to interrupt the PCI side. Register name: LINT_EN Bits 7 Register offset: 0x300 6 5 4 3 2 1 0 LM0 MBOX3 MBOX2 MBOX1 MBOX0 SW_ Reserved VERR LERR DMA VIRQ3 VIRQ2 VIRQ1 VOWN 31:24 Reserved 23:16 LM3 LM2 LM1 15:08 ACFAIL SYSFAIL SW_INT IACK 07:00 VIRQ7 VIRQ6 Name LM3 VIRQ5 Description Location Monitor 3 VIRQ4 Type Reset by Reset value R/W All 0 R/W All 0 R/W All 0 R/W All 0 R/W All 0 R/W All 0 0=LM3 Interrupt Disabled 1=LM3 Interrupt Enabled LM2 Location Monitor 2 0=LM2 Interrupt Disabled 1=LM2 Interrupt Enabled LM1 Location Monitor 1 0=LM1 Interrupt Disabled 1=LM1 Interrupt Enabled LM0 Location Monitor 0 0=LM0 Interrupt Disabled 1=LM0 Interrupt Enabled MBOX3 Mailbox 3 0=MBOX3 Interrupt Disabled 1=MBOX3 Interrupt Enabled MBOX2 Mailbox 2 0=MBOX2 Interrupt Disabled 1=MBOX2 Interrupt Enabled Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 12. Registers > Register Description Name MBOX1 243 Description Mailbox 1 Type Reset by Reset value R/W All 0 R/W All 0 R/W All 0 R/W All 0 R/W All 0 R/W All 0 R/W All 0 R/W All 0 R/W All 0 R/W All 0 R/W All 0 0=MBOX1 Interrupt Disabled 1=MBOX1 Interrupt Enabled MBOX0 Mailbox 0 0=MBOX0 Interrupt Disabled 1=MBOX0 Interrupt Enabled ACFAIL ACFAIL Interrupt 0=ACFAIL Interrupt Disabled 1=ACFAIL Interrupt Enabled SYSFAIL SYSFAIL Interrupt 0=SYSFAIL Interrupt Disabled 1=SYSFAIL Interrupt Enabled SW_INT Local Software Interrupt 0=PCI Software Interrupt Disabled 1=PCI Software Interrupt Enabled A zero-to-one transition will cause the PCI software interrupt to be asserted. Subsequent zeroing of this bit will cause the interrupt to be masked, but will not clear the PCI Software Interrupt Status bit. SW_IACK “VME Software IACK” 0 =“VME Software IACK” Interrupt Disabled 1 =“VME Software IACK” Interrupt Enabled VERR PCI VERR Interrupt 0 =PCI VERR Interrupt Disabled 1=PCI VERR Interrupt Enabled LERR PCI LERR Interrupt 0 =PCI LERR Interrupt Disabled 1 =PCI LERR Interrupt Enabled DMA PCI DMA Interrupt 0=PCI DMA Interrupt Disabled 1=PCI DMA Interrupt Enabled VIRQ7-VIRQ1 VIRQx Interrupt 0=VIRQx Interrupt Disabled 1 =VIRQx Interrupt Enabled VOWN VOWN Interrupt 0=VOWN Interrupt Disabled 1=VOWN Interrupt Enabled Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 244 12. Registers > Register Description 12.3.57 PCI Interrupt Status Register (LINT_STAT) Status bits indicated as “R/Write 1 to Clear” are edge sensitive: the status is latched when the interrupt event occurs. These status bits can be cleared independently of the state of the interrupt source by writing a “1” to the status register. Clearing the status bit does not imply the source of the interrupt is cleared. However, ACFAIL and SYSFAIL are level-sensitive. Clearing ACFAIL or SYSFAIL while their respective pins are still asserted will have no effect. Register name: LINT_STAT Bits 7 Register offset: 0x304 6 5 4 3 2 1 0 LM0 MBOX3 MBOX2 MBOX1 MBOX0 SW_ Reserved VERR LERR DMA VIRQ3 VIRQ2 VIRQ1 VOWN 31:24 Reserved 23:16 LM3 LM2 LM1 15:08 ACFAIL SYSFAIL SW_INT IACK 07:00 VIRQ7 Name LM3 VIRQ6 VIRQ5 Description Location Monitor 3 Status/Clear 0=no Location Monitor 3 Interrupt, VIRQ4 Type Reset by Reset value R/Write 1 to Clear All 0 R/Write 1 to Clear All 0 R/Write 1 to Clear All 0 R/Write 1 to Clear All 0 R/Write 1 to Clear All 0 R/Write 1 to Clear All 0 R/Write 1 to Clear All 0 1=Location Monitor 3 Interrupt active LM2 Location Monitor 2 Status/Clear 0=no Location Monitor 2 Interrupt, 1=Location Monitor 2 Interrupt active LM1 Location Monitor 1 Status/Clear 0=no Location Monitor 1 Interrupt, 1=Location Monitor 1 Interrupt active LM0 Location Monitor 0 Status/Clear 0=no Location Monitor 0 Interrupt, 1=Location Monitor 0 Interrupt active MBOX3 Mailbox 3 Status/Clear 0=no Mailbox 3 Interrupt, 1=Mailbox 3 Interrupt active MBOX2 Mailbox 2 Status/Clear 0=no Mailbox 2 Interrupt, 1=Mailbox 2 Interrupt active MBOX1 Mailbox 1 Status/Clear 0=no Mailbox 1 Interrupt, 1=Mailbox 1 Interrupt active Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 12. Registers > Register Description Name MBOX0 245 Description Mailbox 0 Status/Clear 0=no Mailbox 0 Interrupt, Type Reset by Reset value R/Write 1 to Clear All 0 R/Write 1 to Clear All 0 R/Write 1 to Clear All 0 R/Write 1 to Clear All 0 R/Write 1 to Clear All 0 R/Write 1 to Clear All 0 R/Write 1 to Clear All 0 R/Write 1 to Clear All 0 R/Write 1 to Clear All 0 R/Write 1 to Clear All 0 1=Mailbox 0 Interrupt active ACFAIL ACFAIL Interrupt Status/Clear 0=no ACFAIL Interrupt, 1=ACFAIL Interrupt active SYSFAIL SYSFAIL Interrupt Status/Clear 0=no SYSFAIL Interrupt, 1=SYSFAIL Interrupt active SW_INT Local Software Interrupt Status/Clear 0=no PCI Software Interrupt, 1=PCI Software Interrupt active SW_IACK “VME Software IACK” Status/Clear 0=no “VME Software IACK” Interrupt, 1=“VME Software IACK” Interrupt active VERR Local VERR Interrupt Status/Clear 0=Local VERR Interrupt masked, 1=Local VERR Interrupt active LERR Local LERR Interrupt Status/Clear 0=Local LERR Interrupt masked, 1=Local LERR Interrupt active DMA Local DMA Interrupt Status/Clear 0=Local DMA Interrupt masked, 1=Local DMA Interrupt active VIRQ7-VIRQ1 VIRQx Interrupt Status/Clear 0=VIRQx Interrupt masked, 1=VIRQx Interrupt active VOWN VOWN Interrupt Status/Clear 0=no VOWN Interrupt masked, 1=VOWN Interrupt active Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 246 12. Registers > Register Description 12.3.58 PCI Interrupt Map 0 Register (LINT_MAP0) This register maps various interrupt sources to one of the eight PCI interrupt pins. For example, a value of 000 maps the corresponding interrupt source to LINT_ [0]. Register name: LINT_MAP0 Register offset: 0x308 Bits 7 31:24 Reserved VIRQ7 Reserved VIRQ6 23:16 Reserved VIRQ5 Reserved VIRQ4 15:08 Reserved VIRQ3 Reserved VIRQ2 07:00 Reserved VIRQ1 Reserved VOWN Name VIRQ7-VIRQ1 VOWN 6 5 4 Description 3 2 1 0 Type Reset by Reset value PCI interrupt destination (LINT[7:0]) for VIRQx R/W All 0 VMEbus ownership bit interrupt map to PCI interrupt R/W All 0 Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 12. Registers > Register Description 12.3.59 247 PCI Interrupt Map 1 Register (LINT_MAP1) This register maps various interrupt sources to one of the eight PCI interrupt pins. For example, a value of 000 maps the corresponding interrupt source to LINT_ [0]. Register name: LINT_MAP1 Register offset: 0x30C Bits 7 6 5 31:24 Reserved ACFAIL Reserved SYSFAIL 23:16 Reserved SW_INT Reserved SW_IACK 15:08 07:00 4 3 2 Reserved Reserved Name LERR 0 VERR Reserved DMA Type Reset by Reset value ACFAIL interrupt destination R/W All 0 SYSFAIL SYSFAIL interrupt destination R/W All 0 SW_INT PCI software interrupt destination R/W All 0 VMEbus Software IACK interrupt destination R/W All 0 VERR VMEbus Error interrupt destination R/W All 0 LERR PCI Bus Error interrupt destination R/W All 0 DMA DMA interrupt destination R/W All 0 ACFAIL SW_IACK Description 1 Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 248 12. Registers > Register Description 12.3.60 VMEbus Interrupt Enable Register (VINT_EN) This register enables the various sources of VMEbus interrupts. SW_INT can be enabled with the VME64AUTO power-up option. Register name: VINT_EN Register offset: 0x310 Bits 7 6 5 4 3 2 1 0 31:24 SW_ INT7 SW_ INT6 SW_ INT5 SW_ INT4 SW_ INT3 SW_ INT2 SW_ INT1 Reserved MBOX3 MBOX2 MBOX1 MBOX0 SW_INT Reserved VERR LERR DMA LINT4 LINT3 LINT2 LINT1 LINT0 23:16 Reserved 15:08 07:00 Reserved LINT7 Name SW_INT7 LINT6 LINT5 Description Type Reset by Reset value VME Software 7 Interrupt Mask 0=VME Software 7 Interrupt masked, R/W All 0 R/W All 0 R/W All 0 R/W All 0 1=VME Software 7 Interrupt enabled A zero-to-one transition will cause a VME level 7 interrupt to be generated. Subsequent zeroing of this bit will cause the interrupt to be masked, but will not clear the VME Software 7 Interrupt Status bit. SW_INT6 VME Software 6 Interrupt Mask 0=VME Software 6 Interrupt masked, 1=VME Software 6 Interrupt enabled A zero-to-one transition will cause a VME level 6 interrupt to be generated. Subsequent zeroing of this bit will cause the interrupt to be masked, but will not clear the VME Software 6 Interrupt Status bit. SW_INT5 VME Software 5 Interrupt Mask 0=VME Software 5 Interrupt masked, 1=VME Software 5 Interrupt enabled A zero-to-one transition will cause a VME level 5 interrupt to be generated. Subsequent zeroing of this bit will cause the interrupt to be masked, but will not clear the VME Software 5 Interrupt Status bit. SW_INT4 VME Software 4 Interrupt Mask 0=VME Software 4 Interrupt masked, 1=VME Software 4 Interrupt enabled A zero-to-one transition will cause a VME level 4 interrupt to be generated. Subsequent zeroing of this bit will cause the interrupt to be masked, but will not clear the VME Software 4 Interrupt Status bit. Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 12. Registers > Register Description Name SW_INT3 249 Description Type Reset by Reset value VME Software 3 Interrupt Mask 0=VME Software 3 Interrupt masked, R/W All 0 R/W All 0 R/W All 0 R/W All 0 R/W All 0 R/W All 0 R/W All 0 R/W All Power-up Option 1=VME Software 3 Interrupt enabled A zero-to-one transition will cause a VME level 3 interrupt to be generated. Subsequent zeroing of this bit will cause the interrupt to be masked, but will not clear the VME Software 3 Interrupt Status bit. SW_INT2 VME Software 2 Interrupt Mask 0=VME Software 2 Interrupt masked, 1=VME Software 2 Interrupt enabled A zero-to-one transition will cause a VME level 2 interrupt to be generated. Subsequent zeroing of this bit will cause the interrupt to be masked, but will not clear the VME Software 2 Interrupt Status bit. SW_INT1 VME Software 1 Interrupt Mask 0=VME Software 1 Interrupt masked, 1=VME Software 1 Interrupt enabled A zero-to-one transition will cause a VME level 1 interrupt to be generated. Subsequent zeroing of this bit will cause the interrupt to be masked, but will not clear the VME Software 1 Interrupt Status bit. MBOX3 Mailbox 3 Mask 0=MBOX3 Interrupt masked, 1=MBOX3 Interrupt enabled MBOX2 Mailbox 2 Mask 0=MBOX2 Interrupt masked, 1=MBOX2 Interrupt enabled MBOX1 Mailbox 1 Mask 0=MBOX1 Interrupt masked, 1=MBOX1 Interrupt enabled MBOX0 Mailbox 0 Mask 0=MBOX0 Interrupt masked, 1=MBOX0 Interrupt enabled SW_INT “VME Software Interrupt” Mask 0 = VME Software Interrupt masked 1 =VME Software Interrupt enabled A zero-to-one transition causes the VME software interrupt to be asserted. Subsequent zeroing of this bit causes the interrupt to be masked and the VMEbus interrupt negated, but does not clear the VME software interrupt status bit. Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 250 12. Registers > Register Description Name VERR Description VERR Interrupt Mask Type Reset by Reset value R/W All 0 R/W All 0 R/W All 0 R/W All 0 0 =PCI VERR Interrupt masked 1=PCI VERR Interrupt enabled LERR LERR Interrupt Mask 0 =PCI LERR Interrupt masked 1 =PCI LERR Interrupt enabled DMA DMA Interrupt Mask 0=PCI DMA Interrupt masked 1=PCI DMA Interrupt enabled LINT7-LINT0 PCI Interrupt Mask 0=LINTx Interrupt masked 1 =LINTx Interrupt enabled Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 12. Registers > Register Description 12.3.61 251 VMEbus Interrupt Status Register (VINT_STAT) This register maps PCI Bus interrupt sources to one of the seven VMEbus interrupt pins. A value of 001 maps the corresponding interrupt source to VIRQ*[1], a value of 002 maps to VIRQ*[2], etc. A value of 000 effectively masks the interrupt since there is no corresponding VIRQ*[0]. Register name: VINT_STAT Register offset: 0x314 Bits 7 6 5 4 3 2 1 0 31:24 SW_ SW_ SW_ SW_ SW_ SW_ SW_ Reserved INT7 INT6 INT5 INT4 INT3 INT2 INT1 MBOX3 MBOX2 MBOX1 MBOX0 SW_INT Reserved VERR LERR DMA LINT4 LINT3 LINT2 LINT1 LINT0 23:16 Reserved 15:08 07:00 Reserved LINT7 Name SW_INT7 LINT6 LINT5 Description VME Software 7 Interrupt Status/Clear 0=no VME Software 7 Interrupt, Type Reset by Reset value R/Write 1 to clear All 0 R/Write 1 to clear All 0 R/Write 1 to clear All 0 R/Write 1 to clear All 0 R/Write 1 to clear All 0 R/Write 1 to clear All 0 R/Write 1 to clear All 0 R/Write 1 to clear All 0 1=VME Software 7 Interrupt active SW_INT6 VME Software 6 Interrupt Status/Clear 0=no VME Software 6 Interrupt, 1=VME Software 6 Interrupt active SW_INT5 VME Software 5 Interrupt Status/Clear 0=no VME Software 5 Interrupt, 1=VME Software 5 Interrupt active SW_INT4 VME Software 4 Interrupt Status/Clear 0=no VME Software 4 Interrupt, 1=VME Software 4 Interrupt active SW_INT3 VME Software 3 Interrupt Status/Clear 0=no VME Software 3 Interrupt, 1=VME Software 3 Interrupt active SW_INT2 VME Software 2 Interrupt Status/Clear 0=no VME Software 2 Interrupt, 1=VME Software 2 Interrupt active SW_INT1 VME Software 1 Interrupt Status/Clear 0=no VME Software 1 Interrupt, 1=VME Software 1 Interrupt active MBOX3 Mailbox 3 Status/Clear 0=no Mailbox 3 Interrupt, 1=Mailbox 3 Interrupt active Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 252 12. Registers > Register Description Name MBOX2 Description Mailbox 2 Status/Clear 0=no Mailbox 2 Interrupt, Type Reset by Reset value R/Write 1 to clear All 0 R/Write 1 to clear All 0 R/Write 1 to clear All 0 R/Write 1 to Clear All Power-up Option R/Write 1 to Clear All 0 R/Write 1 to Clear All 0 R/Write 1 to Clear All 0 R/Write 1 to Clear All 0 1=Mailbox 2 Interrupt active MBOX1 Mailbox 1 Status/Clear 0=no Mailbox 1 Interrupt, 1=Mailbox 1 Interrupt active MBOX0 Mailbox 0 Status/Clear 0=no Mailbox 0 Interrupt, 1=Mailbox 0 Interrupt active SW_INT VME Software Interrupt Status/Clear 0=VME Software Interrupt inactive, 1=VME Software Interrupt active VERR VERR Interrupt Status/Clear 0=VME VERR Interrupt masked, 1=VME VERR Interrupt enabled LERR LERR Interrupt Status/Clear 0=VME LERR Interrupt masked, 1=VME LERR Interrupt enabled DMA DMA Interrupt Status/Clear 0=VME DMA Interrupt masked, 1=VME DMA Interrupt enabled LINT7-LINT0 LINTx Interrupt Status/Clear 0=LINTx Interrupt masked, 1=LINTx Interrupt enabled Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 12. Registers > Register Description 12.3.62 253 VME Interrupt Map 0 Register (VINT_MAP0) Register name: VINT_MAP0 Register offset: 0x318 Bits 7 31:24 Reserved LINT7 Reserved LINT6 23:16 Reserved LINT5 Reserved LINT4 15:08 Reserved LINT3 Reserved LINT2 07:00 Reserved LINT1 Reserved LINT0 Name LINT7-LINT0 6 5 4 Description VMEbus destination of PCI Bus interrupt source Integrated Device Technology www.idt.com 3 2 1 0 Type Reset by Reset value R/W all 0 Universe II User Manual May 12, 2010 254 12. Registers > Register Description 12.3.63 VME Interrupt Map 1 Register (VINT_MAP1) This register maps VMEbus interrupt sources to one of the seven VMEbus interrupt pins. A value of 001 maps the corresponding interrupt source to VIRQ*[1], a value of 010 maps to VIRQ*[2], etc. A value of 000 effectively masks the interrupt since there is no corresponding VIRQ*[0]. SW_INT is set to 010 with the VME64AUTO power-up option. Register name: VINT_MAP1 Bits 7 Register offset: 0x31C 6 5 31:24 4 3 2 1 Reserved 23:16 Reserved SW_INT 15:08 Reserved VERR 07:00 0 Reserved Name LERR DMA Type Reset by Reset value VMEbus Software interrupt destination R/W all Power-up Option VERR VMEbus Error interrupt destination R/W all 0 LERR PCI Bus Error interrupt destination R/W all 0 DMA DMA interrupt destination R/W all 0 SW_INT Description Reserved Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 12. Registers > Register Description 12.3.64 255 Interrupt STATUS/ID Out Register (STATID) When the Universe II responds to an interrupt acknowledge cycle on VMEbus it returns an 8-bit STATUS/ID. STATID [7:1] can be written by software to uniquely identify the VMEbus module within the system. STATID [0] is a value of 0 if the Universe II is generating a software interrupt (SW_IACK) at the same level as the interrupt acknowledge cycle, otherwise it is a value of 1. Register name: STATID Bits 7 Register offset: 0x320 6 5 4 3 31:24 STATID [7:0] 23:16 Reserved 15:08 Reserved 07:00 Reserved Name STATID [7:1] STATID [0] Description Bits [7:1] of the STATUS/ID byte are returned when the Universe II responds to a VMEbus IACK cycle. 0 = the Universe II is generating a SW_IACK at the same level as the interrupt acknowledge cycle. 2 1 0 Type Reset by Reset value R/W all 1111111 R all undefined 1 = the Universe II is not generating a SW_IACK at the same level as the interrupt acknowledge cycle. The reset state is designed to support the VME64 Auto ID STATUS/ID value. Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 256 12. Registers > Register Description 12.3.65 VIRQ1 STATUS/ID Register (V1_STATID) The Vx_STATID registers are read-only registers that hold the 8-bit VMEbus STATUS/ID that is acquired when the Universe II performs a IACK cycle for a given interrupt level. Register name: V1_STATID Bits 7 Register offset: 0x324 6 5 4 3 31:24 Reserved 23:16 Reserved 15:08 1 Reserved 07:00 0 ERR STATID [7:0] Name ERR 2 Description Error Status Bit Type Reset by Reset value R all 0 R all 0 0=STATUS/ID was acquired without bus error 1=bus error occurred during acquisition of the STATUS/ID STATID [7:0] STATUS/ID acquired during IACK cycle for level 1 VMEbus interrupt The Universe II is enabled as the interrupt handler for a given interrupt level via the VIRQx bits of the LINT_EN register. Once a vector for a given level is acquired, the Universe II does not perform a subsequent interrupt acknowledge cycle at that level until the corresponding VIRQx bit in the LINT_STAT register is cleared. The acquisition of a level x STATUS/ID by the Universe II updates the STATUS/ID field of the corresponding Vx_STATID register and generation of a PCI interrupt. A VMEbus error during the acquisition of the STATUS/ID vector sets the ERR bit, which means the STATUS/ID field may not contain a valid vector. Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 12. Registers > Register Description 12.3.66 257 VIRQ2 STATUS/ID Register (V2_STATID) The Vx_STATID registers are read-only registers that hold the 8-bit VMEbus STATUS/ID that is acquired when the Universe II performs a IACK cycle for a given interrupt level. Register name: V2_STATID Bits 7 Register offset: 0x328 6 5 4 3 31:24 Reserved 23:16 Reserved 15:08 1 Reserved 07:00 0 ERR STATID [7:0] Name ERR 2 Description Error Status Bit Type Reset by Reset value R all 0 R all 0 0=STATUS/ID was acquired without bus error 1=bus error occurred during acquisition of the STATUS/ID STATID [7:0] STATUS/ID acquired during IACK cycle for level 1 VMEbus interrupt The Universe II is enabled as the interrupt handler for a given interrupt level via the VIRQx bits of the LINT_EN register. Once a vector for a given level is acquired, the Universe II does not perform a subsequent interrupt acknowledge cycle at that level until the corresponding VIRQx bit in the LINT_STAT register is cleared. The acquisition of a level x STATUS/ID by the Universe II updates the STATUS/ID field of the corresponding Vx_STATID register and generation of a PCI interrupt. A VMEbus error during the acquisition of the STATUS/ID vector sets the ERR bit, which means the STATUS/ID field may not contain a valid vector. Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 258 12. Registers > Register Description 12.3.67 VIRQ3 STATUS/ID Register (V3_STATID) The Vx_STATID registers are read-only registers that hold the 8-bit VMEbus STATUS/ID that is acquired when the Universe II performs a IACK cycle for a given interrupt level. Register name: V3_STATID Bits 7 Register offset: 0x32C 6 5 4 3 31:24 Reserved 23:16 Reserved 15:08 1 Reserved 07:00 0 ERR STATID [7:0] Name ERR 2 Description Error Status Bit Type Reset by Reset value R all 0 R all 0 0=STATUS/ID was acquired without bus error 1=bus error occurred during acquisition of the STATUS/ID STATID [7:0] STATUS/ID acquired during IACK cycle for level 3VMEbus interrupt The Universe II is enabled as the interrupt handler for a given interrupt level via the VIRQx bits of the LINT_EN register. Once a vector for a given level is acquired, the Universe II does not perform a subsequent interrupt acknowledge cycle at that level until the corresponding VIRQx bit in the LINT_STAT register is cleared. The acquisition of a level x STATUS/ID by the Universe II updates the STATUS/ID field of the corresponding Vx_STATID register and generation of a PCI interrupt. A VMEbus error during the acquisition of the STATUS/ID vector sets the ERR bit, which means the STATUS/ID field may not contain a valid vector. Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 12. Registers > Register Description 12.3.68 259 VIRQ4 STATUS/ID Register (V4_STATID) The Vx_STATID registers are read-only registers that hold the 8-bit VMEbus STATUS/ID that is acquired when the Universe II performs a IACK cycle for a given interrupt level. Register name: V4_STATID Bits 7 Register offset: 0x330 6 5 4 3 31:24 Reserved 23:16 Reserved 15:08 1 Reserved 07:00 0 ERR STATID [7:0] Name ERR 2 Description Error Status Bit Type Reset by Reset value R all 0 R all 0 0=STATUS/ID was acquired without bus error 1=bus error occurred during acquisition of the STATUS/ID STATID [7:0] STATUS/ID acquired during IACK cycle for level 4 VMEbus interrupt The Universe II is enabled as the interrupt handler for a given interrupt level via the VIRQx bits of the LINT_EN register. Once a vector for a given level is acquired, the Universe II does not perform a subsequent interrupt acknowledge cycle at that level until the corresponding VIRQx bit in the LINT_STAT register is cleared. The acquisition of a level x STATUS/ID by the Universe II updates the STATUS/ID field of the corresponding Vx_STATID register and generation of a PCI interrupt. A VMEbus error during the acquisition of the STATUS/ID vector sets the ERR bit, which means the STATUS/ID field may not contain a valid vector. Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 260 12. Registers > Register Description 12.3.69 VIRQ5 STATUS/ID Register (V5_STATID) The Vx_STATID registers are read-only registers that hold the 8-bit VMEbus STATUS/ID that is acquired when the Universe II performs a IACK cycle for a given interrupt level. Register name: V5_STATID Bits 7 Register offset: 0x334 6 5 4 3 31:24 Reserved 23:16 Reserved 15:08 1 Reserved 07:00 0 ERR STATID [7:0] Name ERR 2 Description Error Status Bit Type Reset by Reset value R all 0 R all 0 0=STATUS/ID was acquired without bus error 1=bus error occurred during acquisition of the STATUS/ID STATID [7:0] STATUS/ID acquired during IACK cycle for level 5 VMEbus interrupt The Universe II is enabled as the interrupt handler for a given interrupt level via the VIRQx bits of the LINT_EN register. Once a vector for a given level is acquired, the Universe II does not perform a subsequent interrupt acknowledge cycle at that level until the corresponding VIRQx bit in the LINT_STAT register is cleared. The acquisition of a level x STATUS/ID by the Universe II updates the STATUS/ID field of the corresponding Vx_STATID register and generation of a PCI interrupt. A VMEbus error during the acquisition of the STATUS/ID vector sets the ERR bit, which means the STATUS/ID field may not contain a valid vector. Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 12. Registers > Register Description 12.3.70 261 VIRQ6 STATUS/ID Register (V6_STATID) The Vx_STATID registers are read-only registers that hold the 8-bit VMEbus STATUS/ID that is acquired when the Universe II performs a IACK cycle for a given interrupt level. Register name: V6_STATID Bits 7 Register offset: 0x338 6 5 4 3 31:24 Reserved 23:16 Reserved 15:08 1 Reserved 07:00 0 ERR STATID [7:0] Name ERR 2 Description Error Status Bit Type Reset by Reset value R all 0 R all 0 0=STATUS/ID was acquired without bus error 1=bus error occurred during acquisition of the STATUS/ID STATID [7:0] STATUS/ID acquired during IACK cycle for level 6 VMEbus interrupt The Universe II is enabled as the interrupt handler for a given interrupt level via the VIRQx bits of the LINT_EN register. Once a vector for a given level is acquired, the Universe II does not perform a subsequent interrupt acknowledge cycle at that level until the corresponding VIRQx bit in the LINT_STAT register is cleared. The acquisition of a level x STATUS/ID by the Universe II updates the STATUS/ID field of the corresponding Vx_STATID register and generation of a PCI interrupt. A VMEbus error during the acquisition of the STATUS/ID vector sets the ERR bit, which means the STATUS/ID field may not contain a valid vector. Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 262 12. Registers > Register Description 12.3.71 VIRQ7 STATUS/ID Register (V7_STATID) The Vx_STATID registers are read-only registers that hold the 8-bit VMEbus STATUS/ID that is acquired when the Universe II performs a IACK cycle for a given interrupt level. Register name: V7_STATID Bits 7 Register offset: 0x33C 6 5 4 3 31:24 Reserved 23:16 Reserved 15:08 1 Reserved 07:00 0 ERR STATID [7:0] Name ERR 2 Description Error Status Bit Type Reset by Reset value R all 0 R all 0 0=STATUS/ID was acquired without bus error 1=bus error occurred during acquisition of the STATUS/ID STATID [7:0] STATUS/ID acquired during IACK cycle for level 7 VMEbus interrupt The Universe II is enabled as the interrupt handler for a given interrupt level via the VIRQx bits of the LINT_EN register. Once a vector for a given level is acquired, the Universe II does not perform a subsequent interrupt acknowledge cycle at that level until the corresponding VIRQx bit in the LINT_STAT register is cleared. The acquisition of a level x STATUS/ID by the Universe II updates the STATUS/ID field of the corresponding Vx_STATID register and generation of a PCI interrupt. A VMEbus error during the acquisition of the STATUS/ID vector sets the ERR bit, which means the STATUS/ID field may not contain a valid vector. Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 12. Registers > Register Description 12.3.72 263 PCI Interrupt Map 2 Register (LINT_MAP2) This register maps interrupt sources to one of the eight PCI interrupt pins. For example, a value of 000 maps the corresponding interrupt source to LINT_ [0]. Register name: LINT_MAP2 Register offset: 0x340 Bits 7 31:24 Reserved LM3 Reserved LM2 23:16 Reserved LM1 Reserved LM0 15:08 Reserved MBOX3 Reserved MBOX2 07:00 Reserved MBOX1 Reserved MBOX0 Name 6 5 Description 4 3 2 1 0 Type Reset by Reset value LM3 [2:0] Location Monitor 3 Interrupt destination R/W all 0 LM2 [2:0] Location Monitor 2 Interrupt destination R/W all 0 LM1 [2:0] Location Monitor 1 Interrupt destination R/W all 0 LM0 [2:0] Location Monitor 0 Interrupt destination R/W all 0 MBOX3 [2:0] Mailbox 3 Interrupt destination R/W all 0 MBOX2 [2:0] Mailbox 2 Interrupt destination R/W all 0 MBOX1 [2:0] Mailbox 1 Interrupt destination R/W all 0 MBOX0 [2:0] Mailbox 0 Interrupt destination R/W all 0 Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 264 12. Registers > Register Description 12.3.73 VME Interrupt Map 2 Register (VINT_MAP2) This register maps interrupt sources to one of the seven VMEbus interrupt pins. A value of 001 maps the corresponding interrupt source to VIRQ*[1], a value of 002 maps to VIRQ*[2], etc. A value of 000 effectively masks the interrupt since there is no corresponding VIRQ*[0]. Register name: VINT_MAP2 Bits 7 Register offset: 0x344 6 5 4 3 31:24 Reserved 23:16 Reserved 2 1 15:08 Reserved MBOX3 Reserved MBOX2 07:00 Reserved MBOX1 Reserved MBOX0 Name Description 0 Type Reset by Reset value MBOX3 [2:0] Mailbox 3 Interrupt destination R/W all 0 MBOX2 [2:0] Mailbox 2 Interrupt destination R/W all 0 MBOX1 [2:0] Mailbox 1 Interrupt destination R/W all 0 MBOX0 [2:0] Mailbox 0 Interrupt destination R/W all 0 Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 12. Registers > Register Description 12.3.74 265 Mailbox 0 Register (MBOX0) This register is a general purpose mailbox register. Register name: MBOX0 Register offset: 0x348 Bits Function 31:24 MBOX0 23:16 MBOX0 15:08 MBOX0 07:00 MBOX0 Name MBOX0 [31:0] Description Mailbox Integrated Device Technology www.idt.com Type Reset by Reset value R/W all 0 Universe II User Manual May 12, 2010 266 12. Registers > Register Description 12.3.75 Mailbox 1 Register (MBOX1) This register is a general purpose mailbox register. Register name: MBOX1 Register offset: 0x34C Bits Function 31:24 MBOX1 23:16 MBOX1 15:08 MBOX1 07:00 MBOX1 Name MBOX1 [31:0] Description Mailbox Universe II User Manual May 12, 2010 Type Reset by Reset value R/W all 0 Integrated Device Technology www.idt.com 12. Registers > Register Description 12.3.76 267 Mailbox 2 Register (MBOX2) This register is a general purpose mailbox register. Register name: MBOX2 Register offset: 0x350 Bits Function 31:24 MBOX2 23:16 MBOX2 15:08 MBOX2 07:00 MBOX2 Name MBOX2 [31:0] Description Mailbox Integrated Device Technology www.idt.com Type Reset by Reset value R/W all 0 Universe II User Manual May 12, 2010 268 12. Registers > Register Description 12.3.77 Mailbox 3 Register (MBOX3) This register is a general purpose mailbox register. Register name: MBOX3 Register offset: 0x354 Bits Function 31:24 MBOX3 23:16 MBOX3 15:08 MBOX3 07:00 MBOX3 Name MBOX3 [31:0] Description Mailbox Universe II User Manual May 12, 2010 Type Reset by Reset value R/W all 0 Integrated Device Technology www.idt.com 12. Registers > Register Description 12.3.78 269 Semaphore 0 Register (SEMA0) This register can only be accessed through byte-wide access. Register name: SEMA0 Register offset: 0x358 Bits 7 31:24 SEM3 TAG3 23:16 SEM2 TAG2 15:08 SEM1 TAG1 07:00 SEM0 TAG0 Name SEM3 TAG3 [6:0] SEM2 TAG2 [6:0] SEM1 TAG1 [6:0] SEM0 TAG0 [6:0] 6 5 Description 4 3 2 1 0 Type Reset by Reset value Semaphore 3 R/W all 0 Tag 3 R/W all 0 Semaphore 2 R/W all 0 Tag2 R/W all 0 Semaphore 1 R/W all 0 Tag 1 R/W all 0 Semaphore 0 R/W all 0 Tag 0 R/W all 0 If a semaphore bit is a value of 0, the associated tag field can be written to. If a semaphore bit is a value of 1, the associated tag field cannot be written to (see “Semaphores” on page 83). Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 270 12. Registers > Register Description 12.3.79 Semaphore 1 Register (SEMA1) This register can only be accessed through byte-wide access. Register name: SEMA1 Register offset: 0x35C Bits 7 31:24 SEM7 TAG7 23:16 SEM6 TAG6 15:08 SEM5 TAG5 07:00 SEM4 TAG4 Name SEM3 TAG3 [6:0] SEM2 TAG2 [6:0] SEM1 TAG1 [6:0] SEM0 TAG0 [6:0] 6 5 Description 4 3 2 1 0 Type Reset by Reset value Semaphore 7 R/W all 0 Tag 7 R/W all 0 Semaphore 6 R/W all 0 Tag 6 R/W all 0 Semaphore 5 R/W all 0 Tag 5 R/W all 0 Semaphore 4 R/W all 0 Tag 4 R/W all 0 If a semaphore bit is a value of 0, the associated tag field can be written to. If a semaphore bit is a value of 1, the associated tag field cannot be written to (see “Semaphores” on page 83). Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 12. Registers > Register Description 12.3.80 271 Master Control Register (MAST_CTL) Writing a 1 to the VOWN bit in the MAST_CTL register has the effect of asserting BBSY* until a 0 is written to the VOWN bit. It does not affect the transactions in the PCI Target Channel. The Universe II will not do an early release of BBSY* if the VMEbus was owned during a transaction by means of VOWN, regardless of the value of PWON. It is important to wait until VOWN_ACK is a value of 0 before writing a value of 1 to the VOWN bit. In the event that BERR* is asserted on the VMEbus once the Universe II owns the VMEbus, the user must release ownership by programming the VOWN bit to a value of 0, if the VMEbus was gained by setting the VOWN bit. VMEbus masters must not write a value of 1 to the VOWN bit since this will lock up the VMEbus. Once the value programmed in the PWON field is reached during de-queuing of posted writes, the Universe II will do an early release of BBSY*. If the PWON field is programmed to a value of 1111, the Universe II will do an early release of BBSY* at the completion of each transaction. Note that the VOWN setting described above overrides the POWN setting. BUS_NO is used by the VMEbus Slave Channel when mapping VME transactions into PCI Configuration space. If the bus number of the VMEbus address (bits [23:16]) is equal to the BUS_NO field, then the Universe II generates a Type 0 configuration cycle, otherwise Type 1 is generated. Register name: MAST_CTL Bits 7 6 31:24 23:16 Register offset: 0x400 5 4 3 2 MAXRTRY VRL 1 0 PWON VRM VREL VOWN VOWN_ Reserved ACK 15:08 Reserved PABS 07:00 BUS_NO Name MAXRTRY [3:0] Reserved Description Maximum Number of Retries Type Reset by Reset value R/W all 1000 R/W all 0000 0000=Retry Forever, Multiples of 64 (0001 through 1111). Maximum Number of retries before the PCI master interface signals error condition PWON [3:0] Posted Write Transfer Count 0000=128 bytes, 0001=256 bytes, 0010=512 bytes, 0011=1024 bytes, 0100=2048 bytes, 0101=4096 bytes, 0110 - 1110 = Reserved, 1111=Early release of BBSY*. Transfer count at which the PCI Slave Channel Posted Writes FIFO gives up the VME Master Interface. Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 272 12. Registers > Register Description Name VRL [1:0] Description VMEbus Request Level Type Reset by Reset value R/W all 11 R/W all 0 R/W all 0 R/W all 0 R all 0 R/W all 00 R/W all 0000 0000 00=Level 0,01=Level 1,10=Level 2, 11=Level 3 VRM VMEbus Request Mode 0=Demand,1=Fair VREL VMEbus Release Mode 0=Release When Done (RWD), 1=Release on Request (ROR) VOWN VME Ownership Bit 0=Release VMEbus, 1=Acquire and Hold VMEbus VOWN_ACK VME Ownership Bit Acknowledge 0=VMEbus not owned, 1=VMEbus acquired and held due to assertion of VOWN PABS [1:0] PCI Aligned Burst Size 00=32 byte 01=64 byte 10=128 byte 11=256 byte Controls the PCI address boundary at which the Universe II breaks up a PCI transaction in the VME Slave channel (see “VME Slave Image Programming” on page 67) and the DMA Channel (see “FIFO Operation and Bus Ownership” on page 101). This field also determines when the PCI Master Module as part of the VME Slave Channel will request the PCI bus (i.e., when 32, 64,128, or 256 bytes are available). It does not have this effect on the DMA Channel, which has a fixed watermark of 128 bytes (see “FIFO Operation and Bus Ownership” on page 101). BUS_NO [7:0] PCI Bus Number Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 12. Registers > Register Description 12.3.81 273 Miscellaneous Control Register (MISC_CTL) VMEbus masters must not write to SW_SYSRST, and PCI masters must not write to SW_LRST. The bits VBTO, VARB and VARBTO support SYSCON functionality. Universe II participation in the VME64 Auto ID mechanism is controlled by the VME64AUTO bit. When this bit is detected high, the Universe II uses the SW_IACK mechanism to generate VXIRQ2 on the VMEbus, then releases VXSYSFAIL. Access to the CR/CSR image is enabled when the level 2 interrupt acknowledge cycle completes. This sequence can be initiated with a power-up option or by software writing a 1 to this bit. Register name: MISC_CTL Bits Register offset: 0x404 7 6 31:24 23:16 15:08 5 VBTO SW_ LRST SW_ SYSRST Reserved SYSRE-SE T Reserved BI 3 2 Reserved VARB ENGBI RE- 1 0 VARBTO SYSCON SCIND V64AUTO Reserved 07:00 Reserved Name VBTO 4 Description VME Bus Time-out Type Reset by Reset value R/W all 0011 R/W all 0 0000=Disable 0001=16 µsec 0010=32 µsec 0011=64 µsec 0100=128 µsec 0101=256 µsec 0110=512 µsec 0111=1024 µsec Others= Reserved Note: The VBTO value has a 4us resolution. The time-out may occur up to 4us earlier than the selected VBTO period. VARB VMEbus Arbitration Mode 0=Round Robin 1=Priority Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 274 12. Registers > Register Description Name VARBTO Description Type Reset by Reset value R/W all 01 R/W all 0 R/W all 0 R/W all Power-up Option R/W all 0 RESCIND is unused in the Universe II. R/W all 1 SYSCON R/W all Power-up Option R/W all Power-up Option R/W all 0 VMEbus Arbitration Time-out 00=Disable Timer 01=16 µs (minimum 8µs) 10=256 µs others= Reserved SW_LRST Software PCI Reset 0=No effect 1=Initiate LRST_ A read always returns 0. SW_ SYSRST Software VMEbus SYSRESET 0=No effect 1=Initiate SYSRST* A read always returns 0. BI BI-Mode 0=Universe II is not in BI-Mode, 1=Universe II is in BI-Mode Write to this bit to change the Universe II BI-Mode status. This bit is also affected by the global BI-Mode initiator VRIRQ1*, if this feature is enabled. ENGBI Enable Global BI-Mode Initiator 0=Assertion of VIRQ1 ignored 1=Assertion of VIRQ1 puts device in BI-Mode RESCIND SYSCON 0=Universe II is not VMEbus System Controller, 1=Universe II is VMEbus System Controller V64AUTO VME64 Auto ID Write: 0=No effect 1=Initiate sequence This bit initiates Universe II VME64 Auto ID Slave participation. SYSRESET System Reset 0=No effect 1=Initiate SYSRST* Universe II asserts SYSRESET without resetting itself. Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 12. Registers > Register Description 12.3.82 275 Miscellaneous Status Register (MISC_STAT) Register name: MISC_STAT Register offset: 0x408 Bits 7 6 31:24 Reserved LCLSIZE 23:16 Reserved 5 4 3 Reserved MYBBSY DY4AUTO Reserved DY4_DON E 15:08 DY4AUTOID 07:00 Reserved Name LCLSIZE 2 Description PCI Bus Size 1 0 Reserved TXFE RXFE Reserved Type Reset by Reset value R all Power-up Option R all Power-up Option R all 1 R all 0 R all 1 R all 1 R Power-up reset and VMEbus SYS- 0 At the rising edge of RST_, the Universe II samples REQ64_ to determine the PCI Bus size. This bit reflects the result. 0=32-bit 1=64-bit DY4AUTO DY4 Auto ID Enable 0=Disable 1=Enable MYBBSY Universe II BBSY 0=Asserted 1=Negated DY4DONE DY4 Auto ID Done 0=Not done 1=Done TXFE PCI Target Channel Posted Writes FIFO 0=data in the FIFO 1=no data in the FIFO RXFE VME Slave Channel Posted Writes FIFO 0=data in the FIFO 1=no data in the FIFO DY4 DY4 Auto ID AUTOID RESET* Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 276 12. Registers > Register Description 12.3.83 User AM Codes Register (USER_AM) The USER_AM register can only be used to generate and accept AM codes 0x10 through 0x1F. These AM codes are designated as USERAM codes in the VMEbus Specification. Register name: USER_AM Register offset: 0x40C Bits 7 6 5 4 3 31:24 0 1 USER1AM Reserved 23:16 0 1 USER2AM Reserved 15:08 Reserved 07:00 Reserved Name Description 2 1 0 Type Reset by Reset value USER1AM [3:0] User AM Code 1 R/W all 0000 USER2AM [3:0] User AM Code 2 R/W all 0000 Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 12. Registers > Register Description 12.3.84 277 Universe II Specific Register (U2SPEC) Register name: U2SPEC Bits Register offset: 0x4FC 7 6 5 4 3 31:24 Universe Reserved 23:16 Universe Reserved 15:08 Universe Reserved DS0/DS1 07:00 AS Reserved Universe Reserved Name DS0/DS1 DTKFLTR Description Data Strobe Filtering 2 1 MASt11 POSt28 0 READt27 Reserved PREt28 Type Reset by Reset value R/W all 0 R/W all 0 R/W all 0 R/W all 0 R/W all 00 R/W all 0 This filter eliminates the noise on DS1* and DS0*. 0=Disable 1=Enable AS Address Strobe Filtering This filter eliminates the noise on AS*. 0=Disable 1=Enable DTKFLTR VME DTACK* Inactive Filter 0=Slower but better filter 1=Faster but poorer filter MASt11 VME Master Parameter t11 Control (DS* high time during BLT’s and MBLT’s) 0=Default 1=Faster READt27 VME Master Parameter t27 Control (Delay of DS* negation after read) 00=Default 01=Faster 10=No Delay POSt28 VME Slave Parameter t28 Control (Time of DS* to DTACK* for posted-write) 0=Default 1=Faster Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 278 12. Registers > Register Description Name Description Type Reset by Reset value PREt28 VME Slave Parameter t28 Control (Time of DS* to DTACK* for prefetch read) R/W all 0 0=Default 1=Faster Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 12. Registers > Register Description 12.3.85 279 VMEbus Slave Image 0 Control Register (VSI0_CTL) The state of PWEN and PREN are ignored if LAS is not programmed memory space. Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 280 12. Registers > Register Description Register name: VSI0_CTL Register offset: 0xF00 Bits 7 6 5 31:24 EN PWEN PREN 23:16 PGM 4 SUPER 1 0 Reserved VAS Reserved LD64EN LLRMW Name EN 2 Reserved 15:08 07:00 3 Reserved Description Image Enable LAS Type Reset by Reset value R/W PWR VME 0 R/W PWR VME 0 R/W PWR VME 0 R/W PWR VME 11 R/W PWR VME 11 R/W PWR VME 0 0=Disable 1=Enable PWEN Posted Write Enable 0=Disable 1=Enable PREN Prefetch Read Enable 0=Disable 1=Enable PGM Program/Data AM Code 00=Reserved 01=Data 10=Program 11=Both SUPER Supervisor/User AM Code 00=Reserved 01=Non-Privileged 10=Supervisor 11=Both VAS VMEbus Address Space 000=A16 001=A24 010=A32 011= Reserved 100=Reserved 101=Reserved 110=User1 111=User2 Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 12. Registers > Register Description Name LD64EN 281 Description Enable 64-bit PCI Bus Transactions Type Reset by Reset value R/W PWR VME 0 R/W PWR VME 0 R/W PWR VME 0 0=Disable 1=Enable LLRMW Enable PCI Bus Lock of VMEbus RMW 0=Disable 1=Enable LAS PCI Bus Address Space 00=PCI Bus Memory Space 01=PCI Bus I/O Space 10=PCI Bus Configuration Space 11=Reserved Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 282 12. Registers > Register Description 12.3.86 VMEbus Slave Image 0 Base Address Register (VSI0_BS) The base address specifies the lowest address in the address range that is decoded. This image has 4 Kbyte resolution. Register name: VSI0_BS Register offset: 0xF04 Bits Function 31:24 BS 23:16 BS 15:08 BS 07:00 Reserved Name BS[31:12] Reserved Description Base Address Universe II User Manual May 12, 2010 Type Reset by Reset value R/W PWR VME 0 Integrated Device Technology www.idt.com 12. Registers > Register Description 12.3.87 283 VMEbus Slave Image 0 Bound Address Register (VSI0_BD) The addresses decoded in a slave image are those which are greater than or equal to the base address and less than the bound register. This image has 4 Kbyte resolution. Register name: VSI0_BD Register offset: 0xF08 Bits Function 31:24 BD 23:16 BD 15:08 BD 07:00 Reserved Name BD[31:12] Reserved Description Bound Address Integrated Device Technology www.idt.com Type Reset by Reset value R/W PWR VME 0 Universe II User Manual May 12, 2010 284 12. Registers > Register Description 12.3.88 VMEbus Slave Image 0 Translation Offset (VSI0_TO) This image has 4 Kbyte resolution. Register name: VSI0_TO Register offset: 0xF0C Bits Function 31:24 TO 23:16 TO 15:08 TO 07:00 Reserved Name TO[31:12] Reserved Description Translation Offset Universe II User Manual May 12, 2010 Type Reset by Reset value R/W PWR VME 0 Integrated Device Technology www.idt.com 12. Registers > Register Description 12.3.89 285 VMEbus Slave Image 1 Control (VSI1_CTL) This register provides the general, VMEbus and PCI controls for this slave image. Note that only transactions destined for PCI Memory space are decoupled (the posted write RXFIFO generates on Memory space transactions on the PCI Bus). In order for a VMEbus slave image to respond to an incoming cycle, the BM bit in the PCI_CSR register must be enabled. The state of PWEN and PREN are ignored if LAS is not programmed memory space. Register name: VSI1_CTL Register offset: 0xF14 Bits 7 6 5 31:24 EN PWEN PREN 23:16 PGM 4 SUPER 1 0 Reserved VAS Reserved LD64EN LLRMW Name EN 2 Reserved 15:08 07:00 3 Reserved Description Image Enable LAS Type Reset by Reset value R/W PWR VME 0 R/W PWR VME 0 R/W PWR VME 0 R/W PWR VME 11 R/W PWR VME 11 0=Disable 1=Enable PWEN Posted Write Enable 0=Disable 1=Enable PREN Prefetch Read Enable 0=Disable 1=Enable PGM Program/Data AM Code 00=Reserved 01=Data 10=Program 11=Both SUPER Supervisor/User AM Code 00=Reserved 01=Non-Privileged 10=Supervisor 11=Both Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 286 12. Registers > Register Description Name VAS Description VMEbus Address Space Type Reset by Reset value R/W PWR VME 0 R/W PWR VME 1 R/W PWR VME 1 R/W PWR VME 0 000=Reserved 001=A24 010=A32 011= Reserved 100=Reserved 101=Reserved 110=User1 111=User2 LD64EN Enable 64-bit PCI Bus Transactions 0=Disable 1=Enable LLRMW Enable PCI Bus Lock of VMEbus RMW 0=Disable 1=Enable LAS PCI Bus Address Space 00=PCI Bus Memory Space 01=PCI Bus I/O Space 10=PCI Bus Configuration Space 11=Reserved Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 12. Registers > Register Description 12.3.90 287 VMEbus Slave Image 1 Base Address Register (VSI1_BS) The base address specifies the lowest address in the address range that will be decoded. Register name: VSI1_BS Register offset: 0xF18 Bits Function 31:24 BS 23:16 BS 15:08 Reserved 07:00 Reserved Name BS[31:16] Description Base Address Integrated Device Technology www.idt.com Type Reset by Reset value R/W PWR VME 0 Universe II User Manual May 12, 2010 288 12. Registers > Register Description 12.3.91 VMEbus Slave Image 1 Bound Address Register (VSI1_BD) The addresses decoded in a slave image are those which are greater than or equal to the base address and less than the bound register. Register name: VSI1_BD Register offset: 0xF1C Bits Function 31:24 BD 23:16 BD 15:08 Reserved 07:00 Reserved Name BD[31:16] Description Bound Address Universe II User Manual May 12, 2010 Type Reset by Reset value R/W PWR VME 0 Integrated Device Technology www.idt.com 12. Registers > Register Description 12.3.92 289 VMEbus Slave Image 1 Translation Offset (VSI1_TO) Register name: VSI1_TO Register offset: 0xF20 Bits Function 31:24 TO 23:16 TO 15:08 Reserved 07:00 Reserved Name TO[31:16] Description Translation Offset Integrated Device Technology www.idt.com Type Reset by Reset value R/W PWR VME 0 Universe II User Manual May 12, 2010 290 12. Registers > Register Description 12.3.93 VMEbus Slave Image 2 Control (VSI2_CTL) This register provides the general, VMEbus and PCI controls for this slave image. Note that only transactions destined for PCI Memory space are decoupled (the posted write RXFIFO generates on Memory space transactions on the PCI Bus). In order for a VMEbus slave image to respond to an incoming cycle, the BM bit in the PCI_CSR register must be enabled. The state of PWEN and PREN are ignored if LAS is not programmed memory space. Register name: VSI2_CTL Register offset: 0xF28 Bits 7 6 5 31:24 EN PWEN PREN 23:16 PGM 4 SUPER 1 0 Reserved VAS Reserved LD64EN LLRMW Name EN 2 Reserved 15:08 07:00 3 Reserved Description Image Enable LAS Type Reset by Reset value R/W PWR VME 0 R/W PWR VME 0 R/W PWR VME 0 R/W PWR VME 11 R/W PWR VME 11 0=Disable 1=Enable PWEN Posted Write Enable 0=Disable 1=Enable PREN Prefetch Read Enable 0=Disable 1=Enable PGM Program/Data AM Code 00=Reserved 01=Data 10=Program 11=Both SUPER Supervisor/User AM Code 00=Reserved 01=Non-Privileged 10=Supervisor 11=Both Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 12. Registers > Register Description Name VAS 291 Description VMEbus Address Space Type Reset by Reset value R/W PWR VME 0 R/W PWR VME 0 R/W PWR VME 0 R/W PWR VME 0 000=Reserved 001=A24 010=A32 011= Reserved 100=Reserved 101=Reserved 110=User1 111=User2 LD64EN Enable 64-bit PCI Bus Transactions 0=Disable 1=Enable LLRMW Enable PCI Bus Lock of VMEbus RMW 0=Disable 1=Enable LAS PCI Bus Address Space 00=PCI Bus Memory Space 01=PCI Bus I/O Space 10=PCI Bus Configuration Space 11=Reserved Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 292 12. Registers > Register Description 12.3.94 VMEbus Slave Image 2 Base Address Register (VSI2_BS) Register name: VSI2_BS Register offset: 0xF2C Bits Function 31:24 BS 23:16 BS 15:08 Reserved 07:00 Reserved Name BS[31:16] Description Base Address Universe II User Manual May 12, 2010 Type Reset by Reset value R/W PWR VME 0 Integrated Device Technology www.idt.com 12. Registers > Register Description 12.3.95 293 VMEbus Slave Image 2 Bound Address Register (VSI2_BD) Register name: VSI2_BD Register offset: 0xF30 Bits Function 31:24 BD 23:16 BD 15:08 Reserved 07:00 Reserved Name BD[31:16] Description Bound Address Integrated Device Technology www.idt.com Type Reset by Reset value R/W PWR VME 0 Universe II User Manual May 12, 2010 294 12. Registers > Register Description 12.3.96 VMEbus Slave Image 2 Translation Offset (VSI2_TO) Register name: VSI2_TO Register offset: 0xF34 Bits Function 31:24 TO 23:16 TO 15:08 Reserved 07:00 Reserved Name TO[31:16] Description Translation Offset Universe II User Manual May 12, 2010 Type Reset by Reset value R/W PWR VME 0 Integrated Device Technology www.idt.com 12. Registers > Register Description 12.3.97 295 VMEbus Slave Image 3 Control (VSI3_CTL) This register provides the general, VMEbus and PCI controls for this slave image. Note that only transactions destined for PCI Memory space are decoupled (the posted write RXFIFO generates on Memory space transactions on the PCI Bus). In order for a VMEbus slave image to respond to an incoming cycle, the BM bit in the PCI_CSR register must be enabled. The state of PWEN and PREN are ignored if LAS is not programmed memory space. Register name: VSI3_CTL Register offset: 0xF3C Bits 7 6 5 31:24 EN PWEN PREN 23:16 PGM 4 SUPER 1 0 Reserved VAS Reserved LD64EN LLRMW Name EN 2 Reserved 15:08 07:00 3 Reserved Description Image Enable LAS Type Reset by Reset value R/W PWR VME 0 R/W PWR VME 0 R/W PWR VME 0 R/W PWR VME 11 R/W PWR VME 11 0=Disable 1=Enable PWEN Posted Write Enable 0=Disable 1=Enable PREN Prefetch Read Enable 0=Disable 1=Enable PGM Program/Data AM Code 00=Reserved 01=Data 10=Program 11=Both SUPER Supervisor/User AM Code 00=Reserved 01=Non-Privileged 10=Supervisor 11=Both Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 296 12. Registers > Register Description Name VAS Description VMEbus Address Space Type Reset by Reset value R/W PWR VME 0 R/W PWR VME 0 R/W PWR VME 0 R/W PWR VME 0 000=Reserved 001=A24 010=A32 011= Reserved 100=Reserved, 101=Reserved 110=User1 111=User2 LD64EN Enable 64-bit PCI Bus Transactions 0=Disable 1=Enable LLRMW Enable PCI Bus Lock of VMEbus RMW 0=Disable 1=Enable LAS PCI Bus Address Space 00=PCI Bus Memory Space 01=PCI Bus I/O Space 10=PCI Bus Configuration Space 11=Reserved Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 12. Registers > Register Description 12.3.98 297 VMEbus Slave Image 3 Base Address Register (VSI3_BS) Register name: VSI3_BS Register offset: 0xF40 Bits Function 31:24 BS 23:16 BS 15:08 Reserved 07:00 Reserved Name BS[31:16] Description Base Address Integrated Device Technology www.idt.com Type Reset by Reset value R/W PWR VME 0 Universe II User Manual May 12, 2010 298 12. Registers > Register Description 12.3.99 VMEbus Slave Image 3 Bound Address Register (VSI3_BD) Register name: VSI3_BD Register offset: 0xF44 Bits Function 31:24 BD 23:16 BD 15:08 Reserved 07:00 Reserved Name BD[31:16] Description Bound Address Universe II User Manual May 12, 2010 Type Reset by Reset value R/W PWR VME 0 Integrated Device Technology www.idt.com 12. Registers > Register Description 12.3.100 299 VMEbus Slave Image 3 Translation Offset (VSI3_TO) Register name: VSI3_TO Register offset: 0xF48 Bits Function 31:24 TO 23:16 TO 15:08 Reserved 07:00 Reserved Name TO[31:16] Description Translation Offset Integrated Device Technology www.idt.com Type Reset by Reset value R/W PWR VME 0 Universe II User Manual May 12, 2010 300 12. Registers > Register Description 12.3.101 Location Monitor Control Register (LM_CTL) This register specifies the VMEbus controls for the location monitor image. This image has a 4 Kbyte resolution and a 4 Kbyte size. The image responds to a VME read or write within the 4 Kbyte space and matching one of the address modifier codes specified. BLTs and MBLTs are not supported. VMEbus address bits [4:3] are used to set the status bit in LINT_STAT for one of the four location monitor interrupts. If the Universe II VMEbus master is the owner of the VMEbus, the Universe II VMEbus slave will generate DTACK* to terminate the transaction. The Location Monitor does not store write data and read data is undefined. Register name: LM_CTL Bits 7 31:24 EN 23:16 Register offset: 0xF64 6 5 4 3 1 0 Reserved PGM SUPER Reserved 15:08 Reserved 07:00 Reserved Name EN 2 Description Image Enable VAS Type Reset by Reset value R/W PWR VME 0 R/W PWR VME 11 R/W PWR VME 11 0=Disable 1=Enable PGM Program/Data AM Code 00=Reserved 01=Data 10=Program 11=Both SUPER Supervisor/User AM Code 00=Reserved 01=Non-Privileged 10=Supervisor 11=Both Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 12. Registers > Register Description Name VAS 301 Description VMEbus Address Space Type Reset by Reset value R/W PWR VME 0 000=A16 001=A24 010=A32 011=Reserved 100=Reserved 101=Reserved 110=User1 111=User2 others= Reserved Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 302 12. Registers > Register Description 12.3.102 Location Monitor Base Address Register (LM_BS) The base address specifies the lowest address in the 4 Kbyte range that will be decoded as a location monitor access. Register name: LM_BS Bits 7 Register offset: 0xF68 6 5 4 3 31:24 BS 23:16 BS 15:08 Reserved 07:00 Reserved Name BS [31:12] Description Base Address Universe II User Manual May 12, 2010 2 1 0 Type Reset by Reset value R/W PWR VME 0 Integrated Device Technology www.idt.com 12. Registers > Register Description 12.3.103 303 VMEbus Register Access Image Control Register (VRAI_CTL) The VME Register Access Image allows access to the Universe II registers with standard VMEbus cycles. Only single cycle and lock AM codes are accepted. When a register is accessed with a RMW, it is locked for the duration of the transaction. Register name: VRAI_CTL Bits 7 31:24 EN 23:16 Register offset: 0xF70 6 5 4 3 1 0 Reserved PGM SUPER Reserved 15:08 Reserved 07:00 Reserved Name EN 2 Description Image Enable VAS Type Reset by Reset value R/W PWR VME Power-up Option R/W PWR VME 11 R/W PWR VME 11 R/W PWR VME Power-up Option 0=Disable 1=Enable PGM Program/Data AM Code 00=Reserved 01=Data 10=Program 11=Both SUPER Supervisor/User AM Code 00=Reserved 01=Non-Privileged 10=Supervisor 11=Both VAS VMEbus Address Space 00=A16 01=A24 10=A32 all others are reserved Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 304 12. Registers > Register Description 12.3.104 VMEbus Register Access Image Base Address Register (VRAI_BS) The base address specifies the lowest address in the 4 Kbyte VMEbus Register Access Image. Register name: VRAI_BS Register offset: 0xF74 Bits Function 31:24 BS 23:16 BS 15:08 BS Reserved 07:00 Reserved Name Description Type Reset by Reset value BS[31:12] The base address specifies the lowest address in the 4 Kbyte VMEbus Register Access Image. R/W PWR VME Power-up Option The reset state is a function of the Power-up Option behavior of the VAS field in VRAI_CTL. Table 35 shows the behavior of the VAS field. Table 35: Power-up Option Behavior of the VAS field in VRAI_CTL VRAI_CTL: VAS BS [31:24] BS [23:16] BS [15:12] A16 0 0 Power-up Option VA [28:25] A24 0 Power-up Option VA [28:21] 0 A32 Power-up Option VA [28:21] 0 0 Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 12. Registers > Register Description 12.3.105 305 VMEbus CSR Control Register (VCSR_CTL) Register name: VCSR_CTL Bits 7 31:24 EN Register offset: 0xF80 6 5 4 3 1 Reserved 15:08 Reserved 07:00 Reserved Name 0 Reserved 23:16 EN 2 Description Image Enable LAS Type Reset by Reset value R/W PWR,VME 0 R/W PWR,VME 0 The EN bit is set to a value of 1 whenever a VME64 monarch acquires the Status/ID vector for the level 2 interrupt during VME64 Auto ID. 0=Disable 1=Enable LAS PCI Bus Address Space 00= PCI Bus Address Space 01= PCI Bus I/O Space Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 306 12. Registers > Register Description 12.3.106 VMEbus CSR Translation Offset (VCSR_TO) For CSR's not supported in the Universe II and for CR accesses, the translation offset is added to the 24-bit VMEbus address to produce a 32-bit PCI Bus address. Register name: VCSR_TO Bits 7 Register offset: 0xF84 6 5 31:24 4 3 2 1 0 TO 23:16 TO Reserved 15:08 Reserved 07:00 Reserved Name Description Type Reset by Reset value TO [31:24] Translation Offset R/W PWR VME 0 TO [23:19] Translation Offset R/W PWR VME Power-up Option Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 12. Registers > Register Description 12.3.107 307 VMEbus AM Code Error Log (V_AMERR) The Universe II VMEbus Master Interface is responsible for logging the parameters of a posted write transaction that results in a bus error. This register holds the address modifier code and the state of the IACK* signal. The register contents are qualified by the V_STAT bit. Register name: V_AMERR Bits 7 Register offset: 0xF88 6 5 31:24 23:16 4 3 AMERR V_STAT Reserved 07:00 Reserved Name IACK M_ERR 1 0 IACK M_ERR Reserved 15:08 AMERR [5:0] 2 Description Type Reset by Reset value VMEbus AM Code Error Log R PWR, VME 0 VMEbus IACK Signal R PWR, VME 0 Multiple Error Occurred R PWR, VME 0 R/W PWR, VME 0 0=Single error 1=At least one error has occurred since the logs were frozen V_STAT VME Error Log Status Reads: 0=logs invalid 1=logs are valid and error logging halted Writes: 0=no effect 1=clears V_STAT and enables error logging Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 308 12. Registers > Register Description 12.3.108 VMEbus Address Error Log (VAERR) The Universe II VMEbus Master Interface is responsible for logging the parameters of a posted write transaction that results in a bus error. This register holds the address. The register contents are qualified by the V_STAT bit of the V_AMERR register. Register name: VAERR Bits 7 Register offset: 0xF8C 6 5 4 3 31:24 VAERR 23:16 VAERR 15:08 VAERR 07:00 VAERR Name VAERR [31:1] Description VMEbus address error log Universe II User Manual May 12, 2010 2 1 0 Type Reset by Reset value R PWR, VME 0 Integrated Device Technology www.idt.com 12. Registers > Register Description 12.3.109 309 VMEbus Slave Image 4 Control (VSI4_CTL) This register provides the general, VMEbus and PCI controls for this slave image. Note that only transactions destined for PCI Memory space are decoupled (the posted write RXFIFO generates on Memory space transactions on the PCI Bus). This image has 4 Kbyte resolution. In order for a VMEbus slave image to respond to an incoming cycle, the BM bit in the PCI_CSR register must be enabled. The state of PWEN and PREN are ignored if LAS is not programmed memory space. Register name: VSI4_CTL Register offset: 0xF90 Bits 7 6 5 31:24 EN PWEN PREN 23:16 PGM 4 SUPER 1 0 Reserved VAS Reserved LD64EN LLRMW Name EN 2 Reserved 15:08 07:00 3 Reserved Description Image Enable LAS Type Reset by Reset value R/W PWR, VME 0 R/W PWR, VME 0 R/W PWR, VME 0 R/W PWR, VME 11 R/W PWR, VME 11 0=Disable 1=Enable PWEN Posted Write Enable 0=Disable 1=Enable PREN Prefetch Read Enable 0=Disable 1=Enable PGM Program/Data AM Code 00=Reserved 01=Data 10=Program 11=Both SUPER Supervisor/User AM Code 00=Reserved 01=Non-Privileged 10=Supervisor 11=Both Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 310 12. Registers > Register Description Name VAS Description VMEbus Address Space Type Reset by Reset value R/W PWR, VME 0 R/W PWR, VME 0 R/W PWR, VME 0 R/W PWR, VME 0 000=A16 001=A24 010=A32 011= Reserved 100=Reserved 101=Reserved 110=User1 111=User2 LD64EN Enable 64-bit PCI Bus Transactions 0=Disable 1=Enable LLRMW Enable PCI Bus Lock of VMEbus RMW 0=Disable 1=Enable LAS PCI Bus Address Space 00=PCI Bus Memory Space 01=PCI Bus I/O Space 10=PCI Bus Configuration Space 11=Reserved Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 12. Registers > Register Description 12.3.110 311 VMEbus Slave Image 4 Base Address Register (VSI4_BS) The base address specifies the lowest address in the address range that is decoded. This image has a 4 Kbyte resolution. Register name: VSI4_BS Register offset: 0xF94 Bits Function 31:24 BS 23:16 BS 15:08 BS 07:00 Reserved Reserved Name Description Type Reset by Reset value BS[31:12] Base Address R/W PWR VME 0 Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 312 12. Registers > Register Description 12.3.111 VMEbus Slave Image 4 Bound Address Register (VSI4_BD) The addresses decoded in a slave image are those which are greater than or equal to the base address and less than the bound register. If the bound register is 0, then the addresses decoded are those greater than or equal to the base address. This image has 4 Kbyte resolution. Register name: VSI4_BD Register offset: 0xF98 Bits Function 31:24 BD 23:16 BD 15:08 BD 07:00 Reserved Name BD[31:12] Reserved Description Bound Address Universe II User Manual May 12, 2010 Type Reset by Reset value R/W PWR VME 0 Integrated Device Technology www.idt.com 12. Registers > Register Description 12.3.112 313 VMEbus Slave Image 4 Translation Offset (VSI4_TO) The translation offset is added to the source address that is decoded and this new address becomes the destination address. If a negative offset is desired, the offset must be expressed as a two’s complement. This image has 4 Kbyte resolution. Register name: VSI4_TO Register offset: 0xF9C Bits Function 31:24 TO 23:16 TO 15:08 TO 07:00 Reserved Name TO[31:12] Reserved Description Translation Offset Integrated Device Technology www.idt.com Type Reset by Reset value R/W PWR VME 0 Universe II User Manual May 12, 2010 314 12. Registers > Register Description 12.3.113 VMEbus Slave Image 5 Control (VSI5_CTL) This register provides the general, VMEbus and PCI controls for this slave image. Note that only transactions destined for PCI Memory space are decoupled (the posted write RXFIFO generates on Memory space transactions on the PCI Bus). In order for a VMEbus slave image to respond to an incoming cycle, the BM bit in the PCI_CSR register must be enabled. The state of PWEN and PREN are ignored if LAS is not programmed memory space. Register name: VSI5_CTL Register offset: 0xFA4 Bits 7 6 5 31:24 EN PWEN PREN 23:16 PGM 4 SUPER 1 0 Reserved VAS Reserved LD64EN LLRMW Name EN 2 Reserved 15:08 07:00 3 Reserved Description Image Enable LAS Type Reset by Reset value R/W PWR VME 0 R/W PWR VME 0 R/W PWR VME 0 R/W PWR VME 11 R/W PWR VME 11 0=Disable 1=Enable PWEN Posted Write Enable 0=Disable 1=Enable PREN Prefetch Read Enable 0=Disable 1=Enable PGM Program/Data AM Code 00=Reserved 01=Data 10=Program 11=Both SUPER Supervisor/User AM Code 00=Reserved 01=Non-Privileged 10=Supervisor 11=Both Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 12. Registers > Register Description Name VAS 315 Description VMEbus Address Space Type Reset by Reset value R/W PWR VME 0 R/W PWR VME 0 R/W PWR VME 0 R/W PWR VME 0 000=Reserved 001=A24 010=A32 011= Reserved 100=Reserved 101=Reserved 110=User1 111=User2 LD64EN Enable 64-bit PCI Bus Transactions 0=Disable 1=Enable LLRMW Enable PCI Bus Lock of VMEbus RMW 0=Disable 1=Enable LAS PCI Bus Address Space 00=PCI Bus Memory Space 01=PCI Bus I/O Space 10=PCI Bus Configuration Space 11=Reserved Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 316 12. Registers > Register Description 12.3.114 VMEbus Slave Image 5 Base Address Register (VSI5_BS) The base address specifies the lowest address in the address range that is decoded. Register name: VSI5_BS Register offset: 0xFA8 Bits Function 31:24 BS 23:16 BS 15:08 Reserved 07:00 Reserved Name BS[31:16] Description Base Address Universe II User Manual May 12, 2010 Type Reset by Reset value R/W PWR VME 0 Integrated Device Technology www.idt.com 12. Registers > Register Description 12.3.115 317 VMEbus Slave Image 5 Bound Address Register (VSI5_BD) The addresses decoded in a slave image are those which are greater than or equal to the base address and less than the bound register. If the bound register is 0, then the addresses decoded are those greater than or equal to the base address. Register name: VSI5_BD Register offset: 0xFAC Bits Function 31:24 BD 23:16 BD 15:08 Reserved 07:00 Reserved Name BD[31:16] Description Bound Address Integrated Device Technology www.idt.com Type Reset by Reset value R/W PWR VME 0 Universe II User Manual May 12, 2010 318 12. Registers > Register Description 12.3.116 VMEbus Slave Image 5 Translation Offset (VSI5_TO) The translation offset is added to the source address that is decoded and this new address becomes the destination address. If a negative offset is desired, the offset must be expressed as a two’s complement. Register name: VSI5_TO Register offset: 0xFB0 Bits Function 31:24 TO 23:16 TO 15:08 Reserved 07:00 Reserved Name TO[31:16] Description Translation Offset Universe II User Manual May 12, 2010 Type Reset by Reset value R/W PWR VME 0 Integrated Device Technology www.idt.com 12. Registers > Register Description 12.3.117 319 VMEbus Slave Image 6 Control (VSI6_CTL) This register provides the general, VMEbus and PCI controls for this slave image. Note that only transactions destined for PCI Memory space are decoupled (the posted write RXFIFO generates on Memory space transactions on the PCI Bus). In order for a VMEbus slave image to respond to an incoming cycle, the BM bit in the PCI_CSR register must be enabled. The state of PWEN and PREN are ignored if LAS is not programmed memory space. Register name: VSI6_CTL Register offset: 0xFB8 Bits 7 6 5 31:24 EN PWEN PREN 23:16 PGM 4 SUPER 1 0 Reserved VAS Reserved LD64EN LLRMW Name EN 2 Reserved 15:08 07:00 3 Reserved Description Image Enable LAS Type Reset by Reset value R/W PWR VME 0 R/W PWR VME 0 R/W PWR VME 0 R/W PWR VME 11 R/W PWR VME 11 0=Disable 1=Enable PWEN Posted Write Enable 0=Disable 1=Enable PREN Prefetch Read Enable 0=Disable 1=Enable PGM Program/Data AM Code 00=Reserved 01=Data 10=Program 11=Both SUPER Supervisor/User AM Code 00=Reserved 01=Non-Privileged 10=Supervisor 11=Both Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 320 12. Registers > Register Description Name VAS Description VMEbus Address Space Type Reset by Reset value R/W PWR VME 0 R/W PWR VME 0 R/W PWR VME 0 R/W PWR VME 0 000=Reserved 001=A24 010=A32 011= Reserved 100=Reserved 101=Reserved 110=User1 111=User2 LD64EN Enable 64-bit PCI Bus Transactions 0=Disable 1=Enable LLRMW Enable PCI Bus Lock of VMEbus RMW 0=Disable 1=Enable LAS PCI Bus Address Space 00=PCI Bus Memory Space 01=PCI Bus I/O Space 10=PCI Bus Configuration Space 11=Reserved Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 12. Registers > Register Description 12.3.118 321 VMEbus Slave Image 6 Base Address Register (VSI6_BS) The base address specifies the lowest address in the address range that is decoded. Register name: VSI6_BS Register offset: 0xFBC Bits Function 31:24 BS 23:16 BS 15:08 Reserved 07:00 Reserved Name BS[31:16] Description Base Address Integrated Device Technology www.idt.com Type Reset by Reset value R/W PWR VME 0 Universe II User Manual May 12, 2010 322 12. Registers > Register Description 12.3.119 VMEbus Slave Image 6 Bound Address Register (VSI6_BD) The addresses decoded in a slave image are those which are greater than or equal to the base address and less than the bound register. If the bound register is 0, then the addresses decoded are those greater than or equal to the base address. Register name: VSI6_BD Register offset: 0xFC0 Bits Function 31:24 BD 23:16 BD 15:08 Reserved 07:00 Reserved Name BD[31:16] Description Bound Address Universe II User Manual May 12, 2010 Type Reset by Reset value R/W PWR VME 0 Integrated Device Technology www.idt.com 12. Registers > Register Description 12.3.120 323 VMEbus Slave Image 6 Translation Offset (VSI6_TO) The translation offset is added to the source address that is decoded and this new address becomes the destination address. If a negative offset is desired, the offset must be expressed as a two’s complement. Register name: VSI6_TO Register offset: 0xFC4 Bits Function 31:24 TO 23:16 TO 15:08 Reserved 07:00 Reserved Name TO[31:16] Description Translation Offset Integrated Device Technology www.idt.com Type Reset by Reset value R/W PWR VME 0 Universe II User Manual May 12, 2010 324 12. Registers > Register Description 12.3.121 VMEbus Slave Image 7 Control (VSI7_CTL) This register provides the general, VMEbus and PCI controls for this slave image. Note that only transactions destined for PCI Memory space are decoupled (the posted write RXFIFO generates on Memory space transactions on the PCI Bus). In order for a VMEbus slave image to respond to an incoming cycle, the BM bit in the PCI_CSR register must be enabled. The state of PWEN and PREN are ignored if LAS is not programmed memory space. Register name: VSI7_CTL Register offset: 0xFCC Bits 7 6 5 31:24 EN PWEN PREN 23:16 PGM 4 SUPER 1 0 Reserved VAS Reserved LD64EN LLRMW Name EN 2 Reserved 15:08 07:00 3 Reserved Description Image Enable LAS Type Reset by Reset value R/W PWR VME 0 R/W PWR VME 0 R/W PWR VME 0 R/W PWR VME 11 R/W PWR VME 11 0=Disable 1=Enable PWEN Posted Write Enable 0=Disable 1=Enable PREN Prefetch Read Enable 0=Disable 1=Enable PGM Program/Data AM Code 00=Reserved 01=Data 10=Program 11=Both SUPER Supervisor/User AM Code 00=Reserved 01=Non-Privileged 10=Supervisor 11=Both Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 12. Registers > Register Description Name VAS 325 Description VMEbus Address Space Type Reset by Reset value R/W PWR VME 0 R/W PWR VME 0 R/W PWR VME 0 R/W PWR VME 0 000=Reserved 001=A24 010=A32 011= Reserved 100=Reserved 101=Reserved 110=User1 111=User2 LD64EN Enable 64-bit PCI Bus Transactions 0=Disable 1=Enable LLRMW Enable PCI Bus Lock of VMEbus RMW 0=Disable 1=Enable LAS PCI Bus Address Space 00=PCI Bus Memory Space 01=PCI Bus I/O Space 10=PCI Bus Configuration Space 11=Reserved Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 326 12. Registers > Register Description 12.3.122 VMEbus Slave Image 7 Base Address Register (VSI7_BS) The base address specifies the lowest address in the address range that is decoded. Register name: VSI7_BS Register offset: 0xFD0 Bits Function 31:24 BS 23:16 BS 15:08 Reserved 07:00 Reserved Name BS[31:16] Description Base Address Universe II User Manual May 12, 2010 Type Reset by Reset value R/W PWR VME 0 Integrated Device Technology www.idt.com 12. Registers > Register Description 12.3.123 327 VMEbus Slave Image 7 Bound Address Register (VSI7_BD) The addresses decoded in a slave image are those which are greater than or equal to the base address and less than the bound register. If the bound register is 0, then the addresses decoded are those greater than or equal to the base address. Register name: VSI7_BD Register offset: 0xFD4 Bits Function 31:24 BD 23:16 BD 15:08 Reserved 07:00 Reserved Name BD[31:16] Description Bound Address Integrated Device Technology www.idt.com Type Reset by Reset value R/W PWR VME 0 Universe II User Manual May 12, 2010 328 12. Registers > Register Description 12.3.124 VMEbus Slave Image 7 Translation Offset (VSI7_TO) Register name: VSI7_TO Register offset: 0xFD8 Bits Function 31:24 TO 23:16 TO 15:08 Reserved 07:00 Reserved Name TO[31:16] Description Translation Offset Universe II User Manual May 12, 2010 Type Reset by Reset value R/W PWR VME 0 Integrated Device Technology www.idt.com 12. Registers > Register Description 12.3.125 329 VMEbus CSR Bit Clear Register (VCSR_CLR) This register implements the Bit Clear Register as defined in the VME64 Specification. The RESET bit must be written to only from the VMEbus. Register name: VCSR_CLR Register offset: 0xFF4 Bits 7 6 5 31:24 RESET SYSFAIL FAIL 4 3 Reserved 15:08 Reserved 07:00 Reserved RESET Description Board Reset 1 0 Reserved 23:16 Name 2 Type Reset by Reset value R/W PWR VME 0 R/W all Power-up Option R PWR VME 0 Reads: 0=LRST_ not asserted 1=LRST_ asserted Writes: 0=no effect 1=negate LRST_ SYSFAIL VMEbus SYSFAIL Reads: 0=VXSYSFAIL not asserted 1=VXSYSFAIL asserted Writes: 0=no effect 1=negate VXSYSFAIL FAIL Board Fail 0=Board has not failed Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 330 12. Registers > Register Description 12.3.126 VMEbus CSR Bit Set Register (VCSR_SET) This register implements the Bit Set Register as defined in the VME64 Specification. The RESET bit must be written to only from the VMEbus. Writing 1 to the RESET bit asserts LRST_. The PCI reset remains asserted until a 1 is written to the RESET bit of the VCSR_CLR register. Register name: VCSR_SET Register offset: 0xFF8 Bits 7 6 5 31:24 RESET SYSFAIL FAIL 4 3 Reserved 15:08 Reserved 07:00 Reserved RESET Description Board Reset 1 0 Reserved 23:16 Name 2 Type Reset by Reset value R/W PWR VME 0 R/W all Power-up Option R PWR VME 0 Reads: 0=LRST_ not asserted 1=LRST_ asserted Writes: 0=no effect 1=assert LRST_ SYSFAIL VMEbus SYSFAIL Reads: 0=VXSYSFAIL not asserted 1=VXSYSFAIL asserted Writes: 0=no effect 1=assert VXSYSFAIL FAIL Board Fail 0=Board has not failed Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 12. Registers > Register Description 12.3.127 331 VMEbus CSR Base Address Register (VCSR_BS) The base address specifies one of 31 available CR/CSR windows as defined in the VME64 Specification. Each window consumes 512 Kbytes of CR/CSR space. Register name: VCSR_BS Bits 7 31:24 Register offset: 0xFFC 6 5 4 3 BS Reserved 15:08 Reserved 07:00 Reserved Name 1 0 Reserved 23:16 BS [23:19] 2 Description Base Address Type Reset by Reset value R/W PWR VME 0 VCSR_BS register is accessed with an 8-bit transfer. Bits [31:27] of the register are compared with address lines [23:19]. Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 332 Universe II User Manual May 12, 2010 12. Registers > Register Description Integrated Device Technology www.idt.com 333 A. Packaging Information This appendix discusses the following topics: • A.1 “313 Pin PBGA Package” on page 333 313 Pin PBGA Package Figure 25: 313 PBGA - Bottom View Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 334 A. Packaging Information > 313 Pin PBGA Package Figure 26: 313 PBGA - Top and Side View Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 335 B. Performance This chapter discusses the following topics: B.1 • “PCI Slave Channel” on page 336 • “VME Slave Channel” on page 340 • “Decoupled Cycles” on page 342 • “DMA Channel and Relative FIFO Sizes” on page 347 • “Universe II Specific Register” on page 350 • “Performance Summary” on page 352 Overview As a VMEbus bridge, the Universe II's most important function is data transfer. This function is performed by its three channels: the PCI Slave Channel, the VME Slave Channel, and the DMA Channel. Since each channel operates independently of the others and because each has its own unique characteristics, the following analysis reviews the data transfer performance for each channel: • “PCI Slave Channel” on page 336 • “VME Slave Channel” on page 340 • “DMA Channel and Relative FIFO Sizes” on page 347 • “Performance Summary” on page 352 Where relevant, descriptions of factors affecting performance and how they might be controlled in different environments are discussed. The decoupled nature of the Universe II can cause some confusion in discussing performance parameters. This is because, in a fully decoupled bus bridge each of the two opposing buses operates at its peak performance independently of the other. The Universe II, however, because of the finite size of its FIFOs does not represent a 100% decoupled bridge. As the FIFOs fill or empty (depending on the direction of data movement) the two buses tend to migrate to matched performance where the higher performing bus is forced to slow down to match the other bus. This limits the sustained performance of the device. Some factors such as the PCI Aligned Burst Size and VME request/release modes can limit the effect of FIFO size and enhance performance. Another aspect in considering the performance of a device is bandwidth consumption. The greater bandwidth consumed to transfer a given amount of data, the less is available for other bus masters. Decoupling significantly improves the Universe II's bandwidth consumption, and on the PCI bus allows it to use the minimum permitted by the PCI specification. To simplify the analysis and allow comparison with other devices, Universe II performance has been calculated using the following assumptions: Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 336 B. Performance > PCI Slave Channel • As a PCI master: — one clock bus grant latency — zero wait state PCI target • As a VME master: — ideal VME slave response (DS* to DTACK* = 30ns) Assumed as part of any calculation on VME performance is the inclusion of VME transceivers with propagation delay of 4 ns. This appendix presents sustained performance values. In contrast, the original Universe User Manual (9000000.MD303.01) provided peak performance numbers. This explains why some of the performance numbers in this document appear to be lower than for the original Universe. B.2 PCI Slave Channel This channel supports both coupled and decoupled transactions. Each type of transaction, and the performance of each, are discussed in the following sections. B.2.1 Coupled Cycles The Universe II has a Coupled Window Timer (CWT in the LMISC register) which permits the coupled channel to maintain ownership of the VMEbus for an extended period beyond the completion of a cycle. This permits subsequent coupled accesses to the VMEbus to occur back-to-back without requirement for re-arbitration. B.2.1.1 Request of VMEbus The CWT should be set for the expected latency between sequential coupled accesses attempted by the CPU. In calculating the latency expected here, the designer needs to account for latency across their host PCI bridge as well as latency encountered in re-arbitration for the PCI bus between each coupled access. Care must be taken not to set the CWT greater than necessary as the Universe II blocks all decoupled write transactions with target-retry, while the coupled channel owns the VMEbus. It is only when the CWT has expired that the PCI bus is permitted to enqueue transactions in the TXFIFO. When a coupled access to the VMEbus is attempted, the Universe II generates a target-retry to the PCI initiator if the coupled path does not currently own the VMEbus. This occurs if the Universe II is not currently VMEbus master, or if the DMA is currently VMEbus master or if entries exist in the TXFIFO. If the Universe II does not have ownership of the VMEbus when a coupled access is attempted, the Universe II generates a target-retry with a single wait state (See Figure 27). The request for the VMEbus occurs shortly after the cycle is retried. Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com B. Performance > PCI Slave Channel B.2.1.2 337 Read Cycles Once the coupled channel owns the VMEbus, the Universe II propagates the cycle out to the VMEbus. Figure 27 shows such a coupled read cycle against an ideal VME slave. There are 10 wait states inserted by the Universe II on the PCI bus before it responds with TRDY_. Further wait states are inserted for each extra 30ns in slave response. Performing 32-bit PCI reads from VME gives a sustained performance of approximately 8.5 MB/s. Figure 28 shows several of these accesses occurring consecutively. Figure 27: Coupled Read Cycle - Universe II as VME Master VMEbus A[31:1] AS* D[31:0] WRITE* DS0* DS1* DTACK* PCI CLK FRAME# AD[31:0] C/BE[3:0] IRDY# TRDY# STOP# DEVSEL# Figure 28: Several Coupled Read Cycles - Universe II as VME Master VMEbus A[31:1] AS* D[31:0] WRITE* DS0* DS1* DTACK* PCI CLK FRAME# AD[31:0] C/BE[3:0] IRDY# TRDY# STOP# DEVSEL# Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 338 B. Performance > PCI Slave Channel B.2.1.3 Write Cycles The performance of coupled write cycles is similar to that of coupled read cycles except that an extra wait state is inserted. Figure 29 shows a coupled write cycle against an ideal VME slave. Ten wait states are inserted on the PCI bus by the Universe II before it responds with TRDY_. A slower VME slave response translates directly to more wait states on the PCI bus. The sustained performance, when generating write cycles from a 32-bit PCI bus against an ideal VME slave is approximately 9.3 MB/s. Figure 29: Coupled Write Cycle - Universe II as VME Master VMEbus A[31:1] AS* D[31:0] WRITE* DS0* DS1* DTACK* PCI CLK FRAME# AD[31:0] C/BE[3:0] IRDY# TRDY# STOP# DEVSEL# B.2.2 Decoupled Cycles Only write transactions can be decoupled in the PCI Target Channel. B.2.2.1 Effect of the PWON Counter The Posted Write On Counter (PWON in the MAST_CTL register) controls the maximum tenure that the PCI Slave Channel will have on the VMEbus. Once this channel has gained ownership of the VMEbus for use by the TXFIFO, it only relinquishes it if the FIFO becomes empty or if the number of bytes programmed in the counter expires. In most situations, the FIFO empties before the counter expires. However, if a great deal of data is being transferred by a PCI initiator to the VMEbus, then this counter ensures that only a fixed amount of VME bandwidth is consumed. Limiting the size of the PWON counter imposes greater arbitration overhead on data being transferred out from the FIFO. This is true even when programmed for ROR mode since an internal arbitration cycle will still occur. The value for the PWON counter must be weighed from the system perspective with the impact of imposing greater latency on other channels (the DMA and Interrupt Channels) and other VME masters in gaining ownership of the VMEbus. On a Universe II equipped card which is only performing system control functions, the counter would be set to minimum. On a card which is responsible for transferring considerable amounts of performance-critical data the counter will be set much higher at the expense of system latency. Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com B. Performance > PCI Slave Channel B.2.2.2 339 PCI Target Response As the PCI target during decoupled write operations to the VMEbus, the Universe II responds in one of two manners: 1. It immediately issues a target retry because the FIFO does not have sufficient room for a burst of one address phase and 128 bytes of data. (There are no programmable watermarks in the PCI Target Channel. The PCI Aligned Burst Size (PABS) does not affect the PCI Target Channel.) 2. It responds as a zero-wait state target receiving up to 256 bytes in a transaction. When the FIFO is full or a 256-byte boundary has been reached, the Universe II issues a Target-Disconnect. In either case, the Universe II will consume the minimum possible PCI bandwidth, never inserting wait states. B.2.2.3 VME Master Performance As a VME master, the Universe II waits until a full transaction has been enqueued in the Tx-FIFO before requesting the VMEbus and generating a VME cycle. If the VMEbus is already owned by the decoupled path (see “Effect of the PWON Counter” on page -338), the Universe II still waits until a full transaction is enqueued in the FIFO before processing it. If configured to generate non-block transfers, the Universe II can generate back-to-back VME transfers with cycle times of approximately 180ns (AS* to AS*) against an ideal VME slave (30-45 ns). A greater cycle time is required between the termination of one full enqueued transaction and the start of the next. This inter-transaction time is approximately 210ns. As such, the longer the PCI transaction, the greater the sustained performance on the VMEbus. With 64-byte PCI transactions, the sustained rate is 43 MB/s. With 32-byte transactions, this drops to 23 MB/s. Each of these numbers is calculated with no initial arbitration or re-arbitration for the bus. Figure 30 shows the Universe II queueing a transaction with multiple non-block VME transfers. Block transfers significantly increase performance. The inter-transaction period remains at approximately 210 ns for BLTs and MBLTs, but the data beat cycle time (DS* to DS*) drops to about 120ns against the same ideal slave. Again the length of the burst size affects the sustained performance because of the inter-transaction time. For BLTs operating with a burst size of 64 bytes, the sustained performance is 37 MB/s, dropping to 33 MB/s for a burst size of 32 bytes. MBLTs operating with 64-byte bursts perform at a sustained rate of 66 MB/s, dropping to 50 MB/s for 32 bytes. Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 340 B. Performance > VME Slave Channel Figure 30: Several Non-Block Decoupled Writes - Universe II as VME Master VMEbus A[31:1] AS* D[31:0] WRITE* DS0* DS1* DTACK* Figure 31: BLT Decoupled Write - Universe II as VME Master VMEbus A[31:1] AS* D[31:0] WRITE* DS0* DS1* DTACK* B.3 VME Slave Channel This channel supports both coupled and decoupled transactions. Each type of transaction, and the performance of each, are discussed in the following sections. B.3.1 Coupled Cycles The Universe II VME Slave Channel handles both block and non-block coupled accesses in similar manners. Each data beat is translated to a single PCI transaction. Once the transaction has been acknowledged on the PCI bus, the Universe II asserts DTACK* to terminate the VME data beat. B.3.1.1 Block vs. non-Block Transfers A non-block transfer and the first beat of a BLT transfer have identical timing. In each, the Universe II decodes the access and then provides a response to the data beat. Subsequent data beats in the BLT transfer are shorter than the first due to the fact that no address decoding need be performed in these beats. MBLT transfers behave somewhat differently. The first beat of an MBLT transfer is address only, and so the response is relatively fast. Subsequent data beats require acknowledgment from the PCI bus. With a 32-bit PCI bus, the MBLT data beat (64 bits of data) requires a two data beat PCI transaction. Because of this extra data beat required on the PCI bus, the slave response of the Universe II during coupled MBLT cycles is at least one PCI clock greater (depending upon the response from the PCI target) than that during BLT cycles. Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com B. Performance > VME Slave Channel B.3.1.2 341 Read Cycles During coupled cycles, the Universe II does not acknowledge a VME transaction until it has been acknowledged on the PCI bus. Because of this the VME slave response during coupled reads is directly linked to the response time for the PCI target. Each clock of latency in the PCI target response translates directly to an extra clock of latency in the Universe II's VME coupled slave response. The address of an incoming VME transaction is decoded and translated to an equivalent PCI transaction. Typically, four PCI clock periods elapse between the initial assertion of AS* on the VMEbus and the assertion of REQ_ on the PCI bus. During the data only portion of subsequent beats in block transfers, the time from DS* assertion to REQ_ is about 4 clocks. If the PCI bus is parked at the Universe II, no REQ_ is asserted and FRAME_ is asserted 4 clocks after AS*. From assertion of REQ_, the Universe II does not insert any extra wait states in its operations as an initiator on the PCI bus. Upon receiving GNT_ asserted, the Universe II asserts FRAME_ in the next clock and after the required turn-around phase, asserts IRDY_ to begin data transfer. Once TRDY_ is sampled asserted, the Universe II responds back to the VMEbus by asserting DTACK*. If the initiating VME transaction is 64-bit and the PCI bus or PCI bus target are 32 bit, then two data transfers are required on PCI before the Universe II can respond with DTACK*. No wait states are inserted by the Universe II between these two data beats on PCI. The assertion of DTACK* from the assertion of TRDY_ has a latency of 1 clock. Figure 32 shows a typical non-block coupled read cycle. When accessing a PCI target with a zero wait state response, the Universe II VME response becomes approximately 10 PCI clock periods (about 301ns in a 33MHz system) during single cycles, and the first beat of a BLT. During pure data beats in both BLT and MBLTs, the slave response becomes 8 clocks. Figure 32: Coupled Read Cycle - Universe II as VME Slave VMEbus A[31:1] AS* D[31:0] WRITE* DS0* DS1* DTACK* PCI CLK FRAME# AD[31:0] C/BE[3:0] IRDY# TRDY# STOP# DEVSEL# Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 342 B. Performance > VME Slave Channel B.3.1.3 Write Cycles Coupled writes in the VME Slave Channel operate in a similar fashion to the coupled reads. The VME slave response is directly linked to the response of the PCI target. In generating the request to the PCI bus, coupled write cycles require one further clock over reads. Hence, during single cycles, or the first beat of a BLT, the time from AS* to REQ_ asserted is 3-4 PCI clocks, while DS* to REQ_ is 3 clocks for the data beat portion of a block transfer. If the PCI bus is parked at the Universe II, REQ_ is not asserted and the transaction begins immediately with assertion of FRAME_. As with reads, the response from the PCI target's assertion of TRDY_ to DTACK* assertion by the Universe II adds one clock to the transfer. Figure 33 shows a typical non-block coupled write cycle. Because write cycles on the PCI bus require one less clock than reads, due to the absence of the turn-around phase between address and data phases, the overall slave response during coupled writes works out to the same as coupled reads against an identical target. In accessing a zero-wait state PCI target, the Universe II's coupled write slave response then is approximately 10 PCI clocks. During subsequent data beats of a block transfer (either BLT or MBLT), the slave response (DS* to DTACK*) is 8 clocks. Figure 33: Coupled Write Cycle - Universe II as VME Slave (bus parked at Universe II) VMEbus A[31:1] AS* D[31:0] WRITE* DS0* DS1* DTACK* PCI CLK FRAME# AD[31:0] C/BE[3:0] IRDY# TRDY# STOP# DEVSEL# B.3.2 Decoupled Cycles B.3.2.1 Write Cycles Effect of the PCI Aligned Burst Size The PCI Aligned Burst Size (PABS in the MAST_CTL register) affects the maximum burst size that the Universe II generates onto the PCI bus; either 32, 64, or 128 bytes. Note that the VME Slave Channel only generates PCI bursts in response to incoming block transfers. Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com B. Performance > VME Slave Channel 343 The greater burst size means less arbitration and addressing overhead. However, incumbent in this is the greater average latency for other devices in the PCI system. Hence, in the VME Slave Channel, the burst size is a trade-off between performance and latency. VME Slave Response As a VME slave, the Universe II accepts data into its RXFIFO with minimum delay provided there is room in the FIFO for a further data beat. Assertion of DTACK* is delayed if there is insufficient room in the FIFO for the next data beat. During non-block transfers, the Universe II must both decode the address and enqueue the data before asserting DTACK* to acknowledge the transfer. Because of this, the slave response during non-block transfers is considerably slower than block transfers. This slave response time is 127ns. During BLT transfers, the slave response in the first data beat being both address decode and data transfer is the same as a non-block transfer, i.e., 127ns. Subsequent data beats, however, are much faster. Response time for these is 50 to 56ns. During MBLT transfers, the first phase is address only and the slave response is 127ns. Subsequent phases are data only and so the slave response is the same as with BLTs i.e., 50 to 56ns. Note that the slave response is independent of the data size. D16 non-block transfers have a slave response identical to D32. BLT data beats have slave responses identical to MBLT data beats. Figure 34: Non-Block Decoupled Write Cycle - Universe II as VME Slave VMEbus A[31:1] AM[5:0] LWORD* AS* D[31:0] WRITE* DS0* DS1* DTACK* PCI CLK FRAME# AD[31:0] C/BE[3:0] IRDY# TRDY# STOP# DEVSEL# Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 344 B. Performance > VME Slave Channel Figure 35: BLT Decoupled Write Cycle - Universe II as VME Slave VMEbus A[31:1] AM[5:0] AS* D[31:0] WRITE* DS0* DS1* DTACK* PCI CLK FRAME# AD[31:0] C/BE[3:0] IRDY# TRDY# STOP# DEVSEL# Figure 36: MBLT Decoupled Write Cycle - Universe II as VME Slave VMEbus A[31:1] AM[5:0] AS* D[31:0] WRITE* DS0* DS1* DTACK* PCI CLK REQ# GNT# FRAME# AD[31:0] C/BE[3:0] IRDY# TRDY# STOP# DEVSEL# PCI Master Performance The Universe II supports bus parking. If the Universe II requires the PCI bus it will assert REQ_ only if its GNT_ is not currently asserted. When the PCI Master Module is ready to begin a transaction and its GNT_ is asserted, the transfer begins immediately. This eliminates a possible one clock cycle delay before beginning a transaction on the PCI bus which would exist if the Universe II did not implement bus parking. Bus parking is described in Section 3.4.3 of the PCI Specification (Rev. 2.1). Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com B. Performance > VME Slave Channel 345 On the PCI bus, the Universe II deliquesce data from the RXFIFO once a complete VME transaction has been enqueued or once sufficient data has been enqueued to form a PCI transaction of length defined by the PABS field. Since the Universe II does not perform any address phase deletion, non-block transfers are decreed from the RXFIFO as single data beat transactions. Only block transfers result in multi-data beat PCI transactions; typically 8, 16 or 32 data beats. In either case, the Universe II does not insert any wait states as a PCI master. The clock, after the bus has been granted to the Universe II, drives out FRAME_ to generate the address phase. The data phases begin immediately on the next clock. If there is more than one data phase, each phase will immediately follow the acknowledgment of the previous phase. In each case, because of the lack of any wait states as a PCI master, the Universe II is consuming the minimum possible bandwidth on the PCI bus, and data will be written to the PCI bus at an average sustained rate equal to the rate at which the VME master is capable of writing it. The sustained performance on the PCI bus performing single data beat write transactions to a 32-bit PCI bus is 15 MB/s; double this for a 64-bit bus. When performing 32-byte transactions the sustained performance increases to 106 MB/s; 120 MB/s with 64-byte transactions. Again, these can be doubled for a 64-bit PCI bus. Bear in mind that the PCI bus can only dequeue data as fast as it is being enqueued on the VMEbus. Hence, as the RXFIFO empties, the sustained performance on the PCI will drop down to match the lower performance on the VME side. However, even with the decreased sustained performance, the consumed bandwidth will remain constant (no extra wait states are inserted while the Universe II is master of the PCI bus.) These numbers assume the PCI bus is granted to the Universe II immediately and that the writes are to a zero-wait state PCI target capable of accepting the full burst length. Figure 27 through Figure 36 show the Universe II responding to non-block, BLT and MBLT write transactions to a 32-bit PCI bus. Even better performance is obtained with PCI bus parking. B.3.2.2 Prefetched Read Cycles To minimize its slave response, the Universe II generates prefetched reads to the PCI bus in response to BLT and MBLT reads coming in from the VMEbus. This option must first be enabled on a per image basis. When enabled, the Universe II will respond to a block read by performing burst reads on the PCI bus of length defined by the PCI Aligned Burst Size (PABS in the MAST_CTL register). These burst reads continue while the block transfer is still active on the VMEbus (AS* not negated) and there is room in the RDFIFO. If there is insufficient room in the RDFIFO to continue (a common occurrence since the Universe II is capable of fetching data from the PCI bus at a much faster rate than a VME master is capable of receiving it), then pre-fetching stops and only continues once enough room exists in the RDFIFO for another full burst size. The first data beat of a block transfer must wait for the first data beat to be retrieved from the PCI bus—this is essentially a coupled transfer. See the section on coupled transfers for details on coupled performance. However, once the pre-fetching begins, data is provided by the Universe II in subsequent data beats with a slave response of 57ns. This continues while there is data in the RDFIFO. If the RDFIFO empties because data is being fetched from the PCI bus too slowly, wait states are inserted on the VMEbus awaiting the enqueueing of more data. Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 346 B. Performance > VME Slave Channel On the PCI bus, the Universe II fetches data at 89 MB/s with PABS set to 32-byte transactions; 106 MB/s when set to 64-byte transactions. Even better performance is obtained if PABS is set for 128-byte transactions. Once the RDFIFO fills, pre-fetching slows to match the rate at which it is being read by the external VMEbus master. Bandwidth consumption, however, remains constant, only the idle time between transactions increases. Figure 37: BLT Pre-fetched Read Cycle - Universe II as VME Slave Cursor1 = 16,600.88 ns Cursor2 = 18,345.94 ns 16,600.88 17,000 17,500 18,000 18,345.94 ns Group: A Group: VME Va[31:0] = 'h zzzzzzzz vam[5:0] = 'h 3F 3F Vas = z Vd[31:0] = 'h zzzzzzzz vwrite = 1 Vds0 = z Vds1 = z Vdtack = z FFF41000 0B 0* xxxxxxxx 5D1* 0B 0B * * * * 5* * * * * * 3F Group: PCI pclk = 1 reqnn[0] = 1 gntnn[0] = 1 framenn = 1 ad_low[31:0] = 'h A5A5A5A5 A5A5A5A5 cxbenn_low[3:0] = 'h 5 5 irdynn = 1 trdynn = 1 stopnn = 1 devselnn = 1 Universe II User Manual May 12, 2010 0 0004107C C Integrated Device Technology www.idt.com B. Performance > DMA Channel and Relative FIFO Sizes B.4 347 DMA Channel and Relative FIFO Sizes Two fixed “watermarks” in the DMA Channel control the Universe’s II requisition of the PCI bus and VMEbus. The DMAFIFO PCI Watermark is 128 bytes. This means that during reads from the PCI bus, the Universe II will wait for 128 bytes to be free in the DMAFIFO before requesting the PCI bus. For PCI writes, the Universe II waits for 128 bytes of data to be in the FIFO before requesting the PCI bus. The DMAFIFO VMEbus watermark is 64 bytes. This means that during reads from the VMEbus, the Universe II will wait for 64 bytes to be free in the DMAFIFO before requesting the Vmebus. For VMEbus writes, the Universe II waits for 64 bytes of data to be in the FIFO before requesting the VMEbus. These watermarks have been tailored for the relative speeds of each bus, and provide near optimal use of the DMA channel. B.4.1 VMEbus Ownership Modes The DMA has two counters that control its access to the VMEbus: the VON (VMEbus On) counter and the VOFF (VMEbus Off) timer. The VON counter controls the number of bytes that are transferred by the DMA during any VMEbus tenure, while the VOFF timer controls the period before the next request after a VON time-out. While the bus is more optimally shared between various masters in the system, and average latency drops as the value programmed for the VON counter drops, the sustained performance of the DMA also drops. The DMA is typically limited by its performance on the VMEbus. As this drops off with greater re-arbitration cycles, the average VMEbus throughput will drop. Even if the Universe II is programmed for ROR mode, and no other channels or masters are requesting the bus, there will be a period of time during which the DMA will pause its transfers on the bus, due to the VON counter expiring. An important point to consider when programming these timers is the more often the DMA relinquishes its ownership of the bus, the more frequently the PCI Slave Channel will have access to the VMEbus. If DMA tenure is too long, the TXFIFO may fill up causing any further accesses to the bus to be retried. In the same fashion, all coupled accesses will be retried while the DMA has tenure on the bus. This can significantly affect transfer latency and should be considered when calculating the overall system latency. B.4.2 VME Transfers On the VMEbus, the Universe II can perform D08 through D64 transactions in either block or non-block mode. The time to perform a single beat, however, is independent of the bus width being used. Hence, a D08 transaction will transfer data at 25% the rate of a D32, which in turn is half that for D64. There is a significant difference between the performance for block vs. non-block operations. Because of the extra addressing required for each data transfer in non-block operations, the DMA performance is about half that compared to operating in block mode. Moreover, considering that most VME slaves respond less quickly in non-block mode, the overall performance may drop to one-quarter of that achievable in block mode. Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 348 B. Performance > DMA Channel and Relative FIFO Sizes When programmed for Release-When-Done operation, the Universe II will perform an early release of BBSY* when the VON counter reaches its programmed limit. This gives other masters a chance to use the VMEbus (and possibly access the VME Slave Channel), but may decrease performance of the DMA Channel; this factor may also play in favor of the DMA Channel, by pausing the PCI Target Channel’s use of the VMEbus. B.4.2.1 Read Transfers When performing non-block reads on the VMEbus, the Universe II cycle time (AS* to next AS*) is approximately 209ns, which translates to about 20 MB/s when performing D32 transfers. For block transfers the cycle time (DS* to next DS*) falls to about 156ns, or 25 MB/s for D32 transfers. For multiplexed block transfers (MBLTs) the cycle time remains the same, but because the data width doubles, the transfer rate increases to about 50MB/s. B.4.2.2 Write Transfers Non-block writes to the VMEbus occur at 180ns cycle time (AS* to next AS*), or 23MB/s during D32 transfers. Block writes, however, are significantly faster with a 116ns cycle time (DS* to next DS*), or 36 MB/s. Multiplexed block transfers have slightly longer cycle times at about 112ns (DS* to next DS*), or 62 MB/s with D64 MBLTs. B.4.3 PCI Transfers As a master on the PCI bus, the Universe II DMA follows the same general set of rules as the VME Slave channel does: it never inserts any wait states into the transfer (i.e., it never negates IRDY_ until the transaction is complete) and will whenever possible, generate full aligned bursts as set in the PABS field of the MAST_CTL register. Between transactions on the PCI bus, the Universe II DMA typically sits idle for 6 clocks. Hence, minimizing the number of idle periods and re-arbitration times by setting PABS to its maximum value of 128 bytes may increase the performance of the DMA on this bus. Higher PABS values imply that the Universe II will hold on to both the PCI bus and the VMEbus for longer periods of time. The reason that PABS also may impact on VMEbus tenure is that (in the case of PCI writes), the DMA FIFO is less likely to fill, and (in the case of PCI reads) the DMA is less likely to go empty. However, given the relative speeds of the buses, and the relative watermarks, the effect of PABS on VMEbus utilization is not as significant as its effects on the PCI bus. While higher values of PABS increase DMA throughput, they may increase system latency. That is, there will be a longer latency for other PCI transactions, including possible transactions coming through the VME Slave Channel (since the DMA channel will own the PCI bus for longer periods of time). Also, accesses between other PCI peripherals will, on average, have a longer wait before being allowed to perform their transactions. PCI latency must be traded off against possible DMA performance. Although both read and write transactions occur on the PCI bus with zero wait states, there is a period of six PCI clocks during which the Universe II remains idle before re-requesting the bus for the next transaction. PCI bus parking may be used to eliminate the need for re-arbitration. Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com B. Performance > DMA Channel and Relative FIFO Sizes 349 With PABS set for 32-byte transactions on a 32-bit PCI bus, this translates to a peak transfer rate of 97 MB/s for reads (including pre-fetching), 98 MB/s for writes, doubling to 194 and 196 for a 64-bit PCI bus. With PABS set for 64-byte transactions, the peak transfer rate increases to 118 MB/s for reads, 125 MB/s for writes on a 32-bit PCI bus—236 MB/s and 250 MB/s respectively for 64-bit PCI buses. The numbers for writes to PCI assume that data are read from VME using BLTs. Figure 38: PCI Read Transactions During DMA Operation PCI CLK REQ# GNT# FRAME# AD[31:0] C/BE[3:0] IRDY# TRDY# STOP# DEVSEL# Figure 39: Multiple PCI Read Transactions During DMA Operation PCI CLK REQ# GNT# FRAME# AD[31:0] C/BE[3:0] IRDY# TRDY# STOP# DEVSEL# Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 350 B.5 B. Performance > Universe II Specific Register Universe II Specific Register The Universe II Specific Register, U2SPEC, offset 0x4FC, can be used to improve the performance of the Universe II by reducing the latency of key VMEbus timing elements. This register is present in versions of the Universe device which have a Revision ID of 01or 02 — defined in the PCI_CLASS register, offset 008. B.5.1 Overview of the U2SPEC Register Although the VMEbus is asynchronous, there are a number of maximum and minimum timing parameters which must be followed. These requirements are detailed in the VME64 Specification. In order to qualify as compliant the master, slave and location monitor devices must guarantee they meet these timing parameters independent of their surroundings. They must assume zero latency between themselves and the VMEbus. This, in practice, is never the case. Buffers, transceivers and the backplane itself, all introduce latencies that combine to produce additional system delay. The consequence of such delay is the degradation of overall performance. The Universe II’s U2SPEC register enables users to compensate for the latencies which are inherent to their VMEbus system designs. Through the use of this register, users can reduce the inherent delay associated with five key VMEbus timing parameters. Use of the U2SPEC register may result in violation of the VME64 Specification. B.5.2 Adjustable VME Timing Parameters B.5.2.1 VME DTACK* Inactive Filter (DTKFLTR) In order to overcome the DTACK* noise typical of most VME systems, the Universe II quadruple samples this signal with the 64 MHz clock. The extra sampling is a precaution that results in decreased performance. Users who believe their systems to have little noise on their DTACK* lines can elect to filter this signal less, and therefore increase their Universe II response time. B.5.2.2 VME Master Parameter t11 Control (MASt11) According to the VME64 Specification, a VMEbus master must not drive DS0* low until both it and DS1* have been simultaneously high for a minimum of 40 ns. The MASt11 parameter in the U2SPEC register, however, allows DS0* to be driven low in less than 40 ns. B.5.2.3 VME Master Parameter t27 Control (READt27) During read cycles, the VMEbus master must guarantee the data lines are valid within 25 ns after DTACK* is asserted. The master must not latch the data and terminate the cycle for a minimum of 25 ns after the falling edge of DTACK*. The READt27 parameter in the U2SPEC register supports faster cycle termination with one of two settings. One setting allows data to be latched and the cycle terminated with an associated delay that is less than 25 ns. The second setting results in no delay in latching and termination. Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com B. Performance > Universe II Specific Register B.5.2.4 351 VME Slave Parameter t28 Control (POSt28) According to the VME64 Specification, VMEbus slaves must wait at least 30 ns after the assertion of DS* before driving DTACK* low. When the Universe II is acting as a VME slave, the POSt28 parameter in the U2SPEC register enables DTACK* to be asserted in less than 30 ns when executing posted writes. B.5.2.5 VME Slave Parameter t28 Control (PREt28) VMEbus slaves must wait at least 30ns after the assertion of DS* before driving DTACK* low. When the Universe II is acting as a VME slave in the transaction, PREt28 parameter in the U2SPEC register enables DTACK* to be asserted in less than 30 ns when executing pre-fetched reads. Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 352 B.6 B. Performance > Performance Summary Performance Summary Table 36: PCI Slave Channel Performance Cycle Type Performance Coupled Read — PCI target response 8 PCI clocks Coupled Write — PCI target response 9 PCI clocks Decoupled Write: Non-block D32 — VME cycle time 180 ns — sustained perf (32-byte PABS) 23 Mbytes/s — sustained perf (64-byte PABS) 43 Mbytes/s D32 BLT — VME cycle time — sustained perf (32-byte PABS) 119 ns — sustained perf (64-byte PABS) 32 Mbytes/s 35 Mbytes/s D64 MBLT — VME cycle time — sustained performance (32-byte PABS) 119 ns — sustained perf (64-byte PABS) 53 Mbytes/s 59 Mbytes/s Table 37: VME Slave Channel Performance Performance Cycle Type VME Slave Response (ns) Coupled Read — non-block 301 — D32 BLT 293 — D64 BLT 322 Coupled Write — non-block 278 — D32 BLT 264 — D64 BLT 292 Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com B. Performance > Performance Summary 353 Table 37: VME Slave Channel Performance Performance Cycle Type VME Slave Response (ns) Pre-fetched Read — VME slave response (1st data beat) 293 — VME slave response (other data beats) 57 Decoupled Write — non-block slave response 127 — block slave response (1st data beat) 127 — block slave response (other data beats) 50 Table 38: DMA Channel Performance Cycle Type Performance Mbytes/s PCI Reads — 32-byte PABS 97 (194)a — 64byte PABS 118 (236) PCI Writes — 32-byte PABS 98 (196) — 64byte PABS 125 (250) VME Reads — non-block D32 18 — D32 BLT 22 — D64 MBLT 45 VME Writes — non-block D32 22 — D32 BLT 32 — D64 MBLT 65 a. 64-bit PCI performance in brackets. Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 354 Universe II User Manual May 12, 2010 B. Performance > Performance Summary Integrated Device Technology www.idt.com 355 C. Reliability Prediction This appendix discusses the following topics: C.1 • “Physical Characteristics” on page 355 • “Thermal Characteristics” on page 356 • “Universe II Ambient Operating Calculations” on page 357 • “Thermal Vias” on page 358 Overview This section is designed to help the user to estimate the inherent reliability of the Universe II. The information serves as a guide only; meaningful results will be obtained only through careful consideration of the device, its operating environment, and its application. C.2 Physical Characteristics • CMOS gate array • 120,000 two-input NAND gate equivalence • 0.5 µm feature size • 309 mils x 309 mils scribed die size Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 356 C.3 C. Reliability Prediction > Thermal Characteristics Thermal Characteristics • Idle power consumption: 1.50 Watts • Typical power consumption* (32-bit PCI): 2.00 Watts • Maximum power consumption (32-bit PCI): 2.70 Watts • Typical power consumption (64-bit PCI): 2.20 Watts • Maximum power consumption (64-bit PCI): 3.20 Watts Maximum power consumption is worst case consumption when the Universe II is performing DMA reads from the VME bus with alternating worst case data patterns ($FFFF_FFFF, $0000_0000 on consecutive cycles), and 100pF loading on the PCI bus In the majority of system applications, the Universe II consumes typical values or less. Typical power consumption numbers are based on the Universe II remaining idle 30%-50% of the time, which is significantly less than what is considered likely in most systems. For this reason, it is recommended that typical power consumption numbers be used for power estimation and ambient temperature calculations, as described below. The HTOL FIT rate is 67 FITs. The HTOL test showed 67 FITs based on the calculation of 60% Confidence Level (C.L.) and Activation Energy Ea=0.7eV for 0.5um process at stress condition 1.1 x Vcc at 125°C ambient temperature This FIT rate is approximately equivalent to 90% C.L. 167 FITs. Calculations were based on a 100 piece sample size for three lots. The test conditions were at 125°C/Bias, 5.5 V at 1 MHz per MIL-STD-883D.M1015.8. Ti p Universe II User Manual May 12, 2010 FIT is the basic reliability rate expressed as failures per billion (1e-9) device hours. Mean Time Between Failures (MTBF) is the reciprocal of FIT. MTBF is the predicted number of device hours before a failure will occur. Integrated Device Technology www.idt.com C. Reliability Prediction > Universe II Ambient Operating Calculations C.4 357 Universe II Ambient Operating Calculations The maximum ambient temperature of the Universe II can be calculated as follows: Ta  Tj - ja * P Where, Ta = Ambient temperature (°C) Tj = Maximum Universe II Junction Temperature (°C) ja = Ambient to Junction Thermal Impedance (°C / Watt) P = Universe II power consumption (Watts) The ambient to junction thermal impedance (ja) is dependent on the air flow in linear feet per minute over the Universe II. The values for ja over different values of air flow are shown in Table 39. Table 39: Ambient to Junction Thermal Impedance Air Flow (m/s) 0 1 2 313 PBGA 15.10 13.10 11.70 For example, the maximum ambient temperature of the 313 PBGA, 32-bit PCI environment with 100 LFPM blowing past the Universe II is: Ta  Tj - ja * P Ta  125 - 13.1* 2.0 Ta  98.8 °C Therefore the maximum rated ambient temperature for the Universe II in this environment is 98.8°C. Further improvements can be made by adding heat sinks to the PBGA package. Tj values of Universe II are calculated as follows (Tj = ja * P + Ta) Table 40: Maximum Universe II Junction Temperature Extended (125 °C Ambient)a Industrial (85 °C Ambient) Commercial: (70 °C Ambient) Tj=13.1*2.7+125=160.37°C Tj=13.1*2.7+85=120.37°C Tj=13.1*2.7+70=105.37°C a. IDT recommends that the maximum junction temperature of the Universe II does not exceed 150 °C. This temperature limit can be achieved by using heat dissipation techniques, such as heat sinks and forced airflows. Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 358 C. Reliability Prediction > Thermal Vias Table 41 shows the simulated Psi jt and Theta jc thermal characteristics of the Universe II package. Table 41: Thermal Characteristics of Universe II Interface Result Psi jt 0.25 C/watt Theta jc (junction to case) 4.7 C/watt These values were obtained under the following PCB and environmental conditions: • PCB condtionsMaximum junction — PCB standard: JEDEC JESD51-9 — PCB layers: 4 — PCB dimensions: 101.6 mm x 114.3 mm — PCB thickness: 1.6 mm • Environmental conditions — Maximum junction temperature: 125C — Ambient temperature: 70C — Power dissipation: 3 W C.5 Thermal Vias The 313-pin plastic BGA package contains thermal vias which directly pipe heat from the die to the solder balls on the underside of the package. The solder balls use the capabilities of the power and ground planes of the printed circuit board to draw heat out of the package. Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 359 D. Endian Mapping Universe II has Little-endian mapping. Little-endian refers to a method of formatting data where address 0 (or the smallest address referencing the data) points to the least significant byte of the data. Data in a system must be consistent; that is, the system must be entirely big-endian or little-endian. This chapter discusses the following topics: • D.1 “Little-endian Mode” on page 359 Overview The Universe II always performs Address Invariant translation between the PCI and VMEbus ports. Address Invariant mapping preserves the byte ordering of a data structure in a little-endian memory map and a big-endian memory map. D.2 Little-endian Mode Table 42 shows the byte lane swapping and address translation between a 32-bit little-endian PCI bus and the VMEbus for the address invariant translation scheme. Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 360 D. Endian Mapping > Little-endian Mode Table 42: Mapping of 32-bit Little-Endian PCI Bus to 32-bit VMEbus PCI Bus Byte Enables Address VMEbus 3 2 1 0 1 0 1 1 1 0 0 0 D0-D7 1 1 0 1 0 1 D8-D15 1 0 1 1 1 0 0 1 1 1 1 1 1 0 0 0 1 0 1 0 0 0 0 0 0 0 0 1 0 0 0 Universe II User Manual May 12, 2010 1 1 0 1 0 0 1 0 0 0 DS1 DS0 A1 LW D8-D15 0 1 0 1 D0-D7 1 0 0 1 D16-D23 D8-D15 0 1 1 1 1 D24-D31 D0-D7 1 0 1 1 0 D0-D7 D8-D15 0 0 0 1 D8-D15 D0-D7 D8-D15 D16-D23 0 0 1 0 D16-D23 D8-D15 D16-D23 D8-D15 0 0 1 1 D24-D31 D0-D7 D0-D7 D24-D31 0 1 0 0 D8-D15 D16-D23 D16-D23 D8-D15 D8-D15 D16-D23 1 0 0 0 D16-D23 D8-D15 D24-D31 D0-D7 D0-D7 D24-D31 0 0 0 0 D8-D15 D16-D23 D16-D23 D8-D15 D24-D31 D0-D7 1 0 0 1 0 Byte Lane Mapping Integrated Device Technology www.idt.com D. Endian Mapping > Little-endian Mode 361 The unpacking of multiplexed 64-bit data from the VMEbus into two 32-bit quantities on a little-endian PCI bus is outlined in Table 43 below. Table 43: Mapping of 32-bit Little-Endian PCI Bus to 64-bit VMEbus Byte Enables 3 2 1 Address 0 2 1 0 PCI to VME Byte Lane Mapping First Transfer (D32-D63) 0 0 0 0 0 0 0 D0-D7 A24-A31 (D56-D63) D8-D15 A16-A23 (D48-D55) D16-D23 A8-A15 (D40-D47) D24-D31 LWORD, A1-A7 (D32-D39) D0-D7 D24-D31 D8-D15 D16-D23 D16-D23 D8-D15 D24-D31 D0-D7 Second Transfer (D0-D31) 0 0 0 0 Integrated Device Technology www.idt.com 1 0 0 Universe II User Manual May 12, 2010 362 Universe II User Manual May 12, 2010 D. Endian Mapping > Little-endian Mode Integrated Device Technology www.idt.com 363 E. Typical Applications This chapter discusses the following topics: E.1 • “VME Interface” on page 363 • “PCI Bus Interface” on page 368 • “Manufacturing Test Pins” on page 370 • “Decoupling VDD and VSS on the Universe II” on page 371 Overview Being a bridge between standard interfaces, the Universe II requires minimal external logic to interface to either the VMEbus or to the PCI bus. In most applications, only transceivers to buffer the Universe II from the VMEbus, plus some reset logic are all that is required. The following information should be used only as a guide in designing the Universe II into a PCI/VME application. Each application will have its own set-up requirements. E.2 VME Interface E.2.1 Transceivers The Universe II has been designed such that it requires full buffering from VMEbus signals. Necessary drive current to the VMEbus is provided by the transceivers while at the same time isolating the Universe II from potentially noisy VMEbus backplanes. In particular, complete isolation of the Universe II from the VMEbus backplane allows use of ETL transceivers which provide high noise immunity as well as use in live insertion environments. The VME community has recently standardized “VME64 Extensions” (ANSI VITA 1.1) which among other new VME features, facilitates live insertion environments. If neither live insertion nor noise immunity are a concern, those buffers that provide input only (U15 and U17 in Figure 40) may be omitted. The daisy chain input signals, BGIN[3:0] and IACKIN, have Schmitt trigger inputs, which should rectify any minor noise on these signals. If considerable noise is expected, the designer may wish to put external filters on these signals. Bear in mind that any filtering done on these signals will detrimentally affect the propagation of bus grants down the daisy chain. Only extremely noisy systems or poorly designed backplanes should require these filters. Figure 40 shows one example of how to connect the Universe II to the VMEbus. The transceivers in this example were chosen to meet the following criteria: • Provide sufficient drive strength as required by the VME specification • Meet Universe II skew requirements, and • Minimize part counts. Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 364 E. Typical Applications > VME Interface U15 and U17 in Figure 40 are optional devices and provide better noise immunity for the system. Figure 40: Universe II Connections to the VMEbus Through TTL Buffers Universe VMEbus A[31:1], LWORD* VOE_ VA_DIR U1 U2 G DIR G DIR A0 A1 B0 B1 B2 A2 A3 A4 A0 A1 A5 A6 B6 B7 A7 B0 B1 B2 B3 B4 A2 A3 A4 A5 A6 B3 B4 B5 U3 U4 G DIR B5 B6 B7 A7 A0 A1 A2 A3 A4 A5 A6 A7 G DIR B0 B1 B2 B3 B4 B5 B6 B7 B0 A0 A1 A2 A3 B1 B2 B3 B4 B5 B6 A4 A5 A6 A7 B7 VA[31:1], LWORD* D[31:0] VD_DIR U5 U6 G DIR U7 G DIR U8 G DIR G DIR A0 B0 A0 B0 A0 B0 A1 B1 B2 A1 A2 B1 B2 A1 A2 B1 B2 B3 B4 A3 A4 B3 B4 A3 A4 B3 B4 A5 A6 B5 B6 A5 A6 B5 B6 A5 A6 B5 B6 A7 B7 A7 B7 A7 B7 A2 A3 A4 A0 A1 B0 B1 A2 A3 B2 B3 A4 A5 B4 B5 A6 A7 B6 B7 VD[31:0] G DIR VAM[5:0] VWRITE_ VIACK_ U10:A VAS_ U11: U11:A VAS_DIR VDS0_ VDS1_ U11:B A0 B0 A1 A2 B1 A3 A4 B3 A5 B5 B2 B4 B6 B7 U10:B U10:C WRITE* IACK* AS* DS[1:0]* U11:C VDS_DIR VDTACK_ AM[5:0] U9 VAM_DIR U10:D DTACK* U11:D VSLAVE_DIR SYSCLK_ U12:A SYSCLK U13:A VBCLR_ U12:B BCLR* VSCON_DIR Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com E. Typical Applications > VME Interface 365 Figure 41: Universe II Connections to the VMEbus Through TTL Buffers Universe VMEbus Vcc U14:A VXBBSY BBSY* U15:A VRBBSY_ U15:B VRACFAIL_ ACFAIL* U14:C VXSYSFAIL SYSFAIL* U15:C VRSYSFAIL_ U14:D VXSYSRST SYSRST* U15:D VRSYSRST_ U14:E VXBERR BERR* U15:E VRBERR_ U14:F-H, U16:A-D VXIRQ[7:1] VRIRQ[7:1]_ 7 IRQ[7:1]* 7 U15:F-H, U17:A-D U16:E-H VXBR[3:0] 4 VRBR[3:0]_ 4 BR[3:0]* U17:E-H VBGIN[3:0]_ 4 VBGOUT[3:0]_ 4 BG[3:0]IN* BG[3:0]OUT* VIACKIN_ IACKIN* VIACKOUT_ IACKOUT* U1-U9 U10, U12 U11, U13 U15, U17 ’245, SN74VMEH22501 (TI) ’126 ’125 ’241 (optional) U14 and U16 were originally 642 buffers, but the 642 buffer is now obsolete. IDT recommends any other open drain buffer as a replacement, such as a F06. Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 366 E. Typical Applications > VME Interface The Universe II, with the addition of external transceivers, is designed to meet the timing requirements of the VME specification. Refer to the VME64 specification (ANSI VITA 1.0) for details on the VME timing. In order to meet the requirements outlined in this specification, the external transceivers must meet certain characteristics as outlined in Table 45. Table 44: VMEbus Signal Drive Strength Requirements VME bus Signal Required Drive Strength A[31:1], D[31:0], AM[5:0], IACK*, LWORD*, WRITE*, DTACK* IOL  48mA IOH  3mA IOL  64mA AS*, DS[1:0]*, IOH  3mA IOL  64mA SYSCLK* IOH  3mA BR[3:0]*, BSY*, IRQ[7:0]*, BERR*, SYSFAIL*, SYSRESET* IOL 48mA Table 45: VMEbus Transceiver Requirements Parameter From To (Input) (Output) Timing Min Max VA, VD, VAM, VIACK, VLWORD, VWRITE, VAS, VDSx, VDTACKa skew (pkg to pkg) A B 8 ns skew (pkg to pkg) B A 4 ns tProp DIR A 1 ns 5ns tProp DIR B 2 ns 10ns Cin A 25 pf a. There are no limits on propagation delay or skew on the remaining buffered VME signals: VSYSCLK, VBCLR, VXBBSY, VRBBSY, VRACFAIL, VXSYSFAIL, VRSYSFAIL, VXSYSRST, VRSYSRST, VXBERR, VRBERR, VXIRQ, VRIRQ, VXBR, VRBR. F Series transceivers meet the requirements specified in Table 44 and Table 45. A faster family such as ABT, may also be used. Care should be taken in the choice of transceivers to avoid ground bounces and also to minimize crosstalk incurred during switching. To limit the effects of crosstalk, the amount of routing under these transceivers must be kept to a minimum. Daisy chain signals can be especially susceptible to crosstalk. Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com E. Typical Applications > VME Interface 367 Should the designer wish to put any further circuitry between the Universe II and the VMEbus, that circuitry must meet the same timing requirements as the transceivers in order for the combined circuit to remain compliant with the VME64 specification. E.2.1.1 Pull-down resistors The Universe II has internal pull-down resistors which are used for its default power-up option state. (Note that REQ64_ has an internal pull-up.) These internal pull-down resistors, ranging from 25k-500k, are designed to sink between 10µA-200µA. F-series buffers, however, can source up to 650 µA of current (worst case). This sourced current has the ability to override the internal power up resistors on the Universe II. This may cause the Universe II to incorrectly sample a logic “1” on the pins. To counteract this potential problem, assuming a worst case scenario of a 650 µA current, IDT recommends connecting a 1K resistor to ground, in parallel, with the internal pull-down resistor. IDT recommends that any pins controlling the power-up options which are critical to the application at power-up be connected to ground with a pull-down resistor as described above. If these options are not critical and if it is possible to reprogram these options after reset, additional resistors need not be added. E.2.2 Direction control When the Universe II is driving VMEbus lines, it drives the direction control signals high (i.e., VA_DIR, VAM_DIR, VAS_DIR, VD_DIR, VDS_DIR, VSLAVE_DIR, and VSCON_DIR). When the VMEbus is driving the Universe II, these signals are driven low. The control signals in the Universe II do not all have the same functionality. Since the Universe II implements early bus release, VAS_DIR must be a separate control signal. Contention between the Universe II and the VME buffers is handled since the Universe II tristates its outputs one 64MHz clock period before the buffer direction control is faced inwards. E.2.3 Power-up Options Power-up options for the automatic configuration of slave images and other Universe II features are provided through the state of the VME address and data pins, VA[31:1] and VD[31:27]. All of these signals are provided with internal pull-downs to bias these signals to their default conditions. Should values other than the defaults be required here, either pull-ups or active circuitry may be applied to these signals to provide alternate configurations. Power-up options are described in “Resets, Clocks and Power-up Options” on page 129. Since the power-up configurations lie on pins that may be driven by the Universe II or by the VME transceivers, care must be taken to ensure that there is no conflict. During any reset event, the Universe II does not drive the VA or VD signals. As well, during any VMEbus reset (SYSRST*) and for several CLK64 periods after, the Universe II negates VOE_ to tri-state the transceivers. During the period that these signals are tri-stated, the power-up options are loaded with their values latched on the rising edge of PWRRST_. Configuration of power-up options is most easily accomplished through passive 10k pull-up resistors on the appropriate VA and VD pins. The configurations may be made user-configurable through jumpers or switches as shown in Figure 42 Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 368 E. Typical Applications > PCI Bus Interface Figure 42: Power-up Configuration Using Passive Pull-ups Universe II VDD Power-up Pins Alternatively, an active circuit may be designed which drives the VA and VD pins with pre-set (or pre-programmed) values. This sort of circuit would be of value when power-up configurations such as the register access slave image are stored in an external programmable register. To implement this circuit, the VOE_ output from the Universe II must be monitored. When the Universe II negates this signal, the appropriate VA and VD signals may be driven and upon re-assertion the drive must be removed. To avoid conflict with the transceivers, logic must be designed such that the enabling of the transceivers does not occur until some point after the configuration options have been removed from the VD and VA signals. Figure 43 shows one such implementation. The delay for enabling of the VMEbus transceivers could be implemented though clocked latches. Figure 43: Power-up Configuration Using Active Circuitry Universe II OE_ Tranceivers Delay OE_ Power-up Pins E.2.3.1 VMEbus VOE_ External Register Auto-Syscon and PCI Bus Width Power-up Options The VME64 specification provides for automatic enabling of the system controller in a VME system through monitoring of the BGIN3* signal. If at the end of SYSRST* this pin is low, then the system controller is enabled; otherwise it is disabled. The Universe II provides an internal pull-down resistor for this function. If it is in slot one, this pin will be sampled low. If not in slot one, then it will be driven high by the previous board in the system and system controller functions will be disabled. No external logic is required to implement this feature. E.3 PCI Bus Interface The Universe II provides a fully standard PCI bus interface compliant for both 32-bit and 64-bit designs. No external transceivers or glue logic is required in interfacing the Universe II to any other PCI compliant devices. All signals may be routed directly to those devices. Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com E. Typical Applications > PCI Bus Interface 369 The Universe II’s PCI interface can be used as a 32-bit bus or 64-bit bus. If used as a 32-bit interface, the 64-bit pins, AD[32:63] and ACK64_ are left unterminated. On a 32-bit PCI bus, the Universe II drives all its 64-bit extension bi-direct signals (C/BE[7:4]_, AD[63:32], REQ64_, PAR64 and ACK64_) at all times to unknown values. Independent of the setting of the LD64EN bit, the Universe II will never attempt a 64-bit cycle on the PCI bus if it is powered up as 32-bit. REQ64_ must be pulled-down (with a 4.7kresistor) at reset for 64-bit PCI (see “PCI Bus Width” on page 138). There is an internal pull-up on this pin which causes the Universe II to default to 32-bit PCI. This power-up option provides the necessary information to the Universe II so that these unused pins may be left unterminated. E.3.1 Resets The Universe II provides several reset input and outputs which are asserted under various conditions. These can be grouped into three types as shown in Table 46. Table 46: Reset Signals Group Signal Name Direction VMEbus VXSYSRST output VRSYSRST_ input LRST_ output RST_ input VME_RESET_ input PWRRST_ input PCI bus Power-up E.3.1.1 VMEbus Resets The VMEbus resets are connected to the VMEbus as indicated in Figure 40 on page 364 through external buffers. E.3.1.2 PCI bus Resets Use of the PCI bus resets will be application dependent. The RST_ input to the Universe II should typically be tied in some fashion to the PCI bus reset signal of the same name. This will ensure that all Universe II PCI related functions are reset together with the PCI bus. The LRST_ pin is a totem-pole output which is asserted due to any of the following initiators: • PWRRST_, • VRSYSRST_, • local software reset (in the MISC_CTL register), or • VME CSR reset (in the VCSR_SET register). Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 370 E. Typical Applications > Manufacturing Test Pins The designer may wish to disallow the Universe II from resetting the PCI bus in which case this output may be left unconnected. Otherwise LRST_ should be grouped with other PCI reset generators to assert the RST_ signal such that: RST_ = LRST_ and reset_source1 and reset_source2 and... If the Universe II is the only initiator of PCI reset, LRST_ may be directly connected to RST_. Assertion of VME_RESET causes the Universe II to assert VXSYSRST. This signal must not by tied to the PCI RST_ signal unless the Universe II LRST_ output will not generate a PCI bus reset. Connecting both LRST_ and VME_RESET_ to RST_ will cause a feedback loop on the reset circuitry forcing the entire system into a endless reset. To reset the VMEbus through this signal it is recommended that it be asserted for several clock cycles, until the Universe II asserts RST_, and then released. This ensures a break is made in the feedback path. E.3.1.3 Power-Up Reset The PWRRST_ input is used to provide reset to the Universe II until the power supply has reached a stable level. It should be held asserted for 100 milliseconds after power is stable. Typically this can be achieved through a resistor/capacitor combination although more accurate solutions using under voltage sensing circuits (e.g. MC34064) are often implemented. The power-up options are latched on the rising edge of PWRRST_. E.3.1.4 JTAG Reset The JTAG reset, TRST_, should be tied into the master system JTAG controller. It resets the Universe II internal JTAG controller. If JTAG is not being used, this pin should be tied to ground. E.3.2 Local Interrupts The Universe II provides eight local bus interrupts, only one of which has drive strength that is fully PCI compliant. If any of the other seven interrupts are to be used as interrupt outputs to the local bus (all eight may be defined as either input or output), an analysis must be done on the design to determine whether the 4 mA of drive that the Universe II provides on these lines is sufficient for the design. If more drive is required, the lines may simply be buffered. All Universe II interrupts are initially defined as inputs. To prevent excess power dissipation, any interrupts defined as inputs should always be driven to either high or low. Pull-ups should be used for this purpose rather than direct drive since a mis-programming of the interrupt registers may cause the local interrupts to be configured as outputs and potentially damage the device. E.4 Manufacturing Test Pins The Universe II has several signals used for manufacturing test purposes. They are listed in Table 24 on page 140, along with the source to which they should be tied. Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com E. Typical Applications > Decoupling VDD and VSS on the Universe II E.5 371 Decoupling VDD and VSS on the Universe II This section is intended to be a guide for decoupling the power and ground pins on the Universe II. A separate analog power and ground plane is not required to provide power to the analog portion of the Universe II. However, to ensure a jitter free PLL operation, the analog AVDD and AVSS pins must be noise free. The following are recommended solutions for noise free PLL operation. The design could implement one of these solutions, but not both. The Analog Isolation Scheme consists of the following: • a 0.1µF capacitor between the AVDD and AVSS pins, and • corresponding inductors between the pins and the board power and ground planes (See Figure 44). These inductors are not necessary, but they are recommended. Figure 44: Analog Isolation Scheme Board VDD 1.5 - 220 µH AVDD 0.1 µF AVSS 1.5 - 220 µH Board VSS The Noise Filter Scheme filters out the noise using two capacitors to filter high and low frequencies (See Figure 45). Figure 45: Noise Filter Scheme Board VDD 10-38 † AVDD 0.01 µF (High Freq. Bias) 22 µF (Low Freq. Bias) AVSS Board VSS For both schemes, it is recommended that the components involved be tied as close as possible to the associated analog pins. Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 372 E. Typical Applications > Decoupling VDD and VSS on the Universe II In addition to the decoupling schemes shown above, it is recommended that 0.1µF bypass capacitors should be tied between every three pairs of VDD pins and the board ground plane. These bypass capacitors should also be tied as close as possible to the package. Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com 373 F. Ordering Information This appendix discusses Universe II’s ordering information. F.1 Ordering Information IDT products are designated by a product code. When ordering, refer to products by their full code. Table 47 details the available part numbers. Table 47: Standard Ordering Information Part Number PCI Frequency Voltage Temperature Package CA91C142D-33CE 33 MHz 5V Commercial (0° to 70°C) PBGA CA91C142D-33CEV 33 MHz 5V Commercial (0° to 70°C) CA91C142D-33IE 33 MHz 5V Industrial (-40° to 85°C) CA91C142D-33IEV 33 MHz 5V Industrial (-40° to 85°C) CA91C142D-25EE Integrated Device Technology www.idt.com 25 MHz 5V Extended (-55° to 125°C) PBGA (RoHS/Green) PBGA PBGA (RoHS/Green) PBGA Universe II User Manual May 12, 2010 374 F. Ordering Information > Ordering Information The IDT “Tsi” part numbering system is explained as follows. Prototype version status Operating environment Package type RoHS/Green compliance Operating frequency IDT product identifier Product number Tsi NNN(N) - SS(S) E P G (Z#) • ( ) – Indicates optional characters. • Tsi – IDT “Tsi” product identifier. • NNNN – Product number (may be three or four digits). • SS(S) – Maximum operating frequency or data transfer rate of the fastest interface. For operating frequency numbers, M and G represent MHz and GHz. For transfer rate numbers, M and G represent Mbps and Gbps. • E – Operating environment in which the product is guaranteed. This code may be one of the following characters: — C - Commercial temperature range (0 to +70°C) — I - Industrial temperature range (-40 to +85°C) — E - Extended temperature range (-55 to +125°C) • P – The Package type of the product: — B - Ceramic ball grid array (CBGA) — E, L, J, and K - Plastic ball grid array (PBGA) — G - Ceramic pin grid array (CPGA) — M - Small outline integrated circuit (SOIC) — Q - Plastic quad flatpack (QFP) Universe II User Manual May 12, 2010 Integrated Device Technology www.idt.com F. Ordering Information > Ordering Information • 375 G – IDT “Tsi” products fit into three RoHS-compliance categories: — Y - RoHS Compliant (6of6) – These products contain none of the six restricted substances above the limits set in the EU Directive 2002/95/EC. — Y - RoHS Compliant (Flip Chip) – These products contain only one of the six restricted substances: Lead (Pb). These flip-chip products are RoHS compliant through the Lead exemption for Flip Chip technology, Commission Decision 2005/747/EC, which allows Lead in solders to complete a viable electrical connection between semiconductor die and carrier within integrated circuit Flip Chip packages. — V - RoHS Compliant/Green - These products follow the above definitions for RoHS Compliance and meet JIG (Joint Industry Guide) Level B requirements for Brominated Flame Retardants (other than PBBs and PBDEs). • Z# – Prototype version status (optional). If a product is released as a prototype then a “Z” is added to the end of the part number. Further revisions to the prototype prior to production release would add a sequential numeric digit. For example, the first prototype version of device would have a “Z,” a second version would have “Z1,” and so on. The prototype version code is dropped once the product reaches production status. Integrated Device Technology www.idt.com Universe II User Manual May 12, 2010 376 Universe II User Manual May 12, 2010 F. Ordering Information > Ordering Information Integrated Device Technology www.idt.com CORPORATE HEADQUARTERS 6024 Silver Creek Valley Road San Jose, CA 95138 for SALES: 800-345-7015 or 408-284-8200 fax: 408-284-2775 www.idt.com for Tech Support: email: VMEhelp@idt.com phone: 408-360-1538 Document: May 12, 2010 DISCLAIMER Integrated Device Technology, Inc. (IDT) and its subsidiaries reserve the right to modify the products and/or specifications described herein at any time and at IDT’s sole discretion. All information in this document, including descriptions of product features and performance, is subject to change without notice. Performance specifications and the operating parameters of the described products are determined in the independent state and are not guaranteed to perform the same way when installed in customer products. The information contained herein is provided without representation or warranty of any kind, whether express or implied, including, but not limited to, the suitability of IDT’s products for any particular purpose, an implied warranty of merchantability, or non-infringement of the intellectual property rights of others. This document is presented only as a guide and does not convey any license under intellectual property rights of IDT or any third parties. IDT’s products are not intended for use in life support systems or similar devices where the failure or malfunction of an IDT product can be reasonably expected to significantly affect the health or safety of users. Anyone using an IDT product in such a manner does so at their own risk, absent an express, written agreement by IDT. Integrated Device Technology, IDT and the IDT logo are registered trademarks of IDT. Other trademarks and service marks used herein, including protected names, logos and designs, are the property of IDT or their respective third party owners. May 2010  2010 Integrated Device Technology, Inc *Notice: The information in this document is subject to change without notice
CA91C142D-33IEV 价格&库存

很抱歉,暂时无法提供与“CA91C142D-33IEV”相匹配的价格&库存,您可以联系我们找货

免费人工找货