Compaq EV68A Hardware Reference Manual
  1. Manuals
  2. Brands
  3. Compaq Manuals
  4. Computer Hardware
  5. EV68A
  6. Hardware reference manual
Compaq EV68A Hardware Reference Manual

Compaq EV68A Hardware Reference Manual

Compaq microprocessor reference manual
Hide thumbs
Table of Contents

Advertisement

Quick Links

Download this manual
21264/EV68A Microprocessor
Hardware Reference Manual
Part Number: DS–0038B–TE
This manual is directly derived from the internal 21264/EV68A Specifications, Revi-
sion 1.1. You can access this hardware reference manual in PDF format from the
following site:
ftp://ftp.compaq.com/pub/products/alphaCPUdocs
Revision/Update Information:
Revision 1.1, March 2002
Compaq Computer Corporation
Shrewsbury, Massachusetts

Advertisement

Table of Contents
loading
Need help?

Need help?

Do you have a question about the EV68A and is the answer not in the manual?

Questions and answers

Related Manuals for Compaq EV68A

Summary of Contents for Compaq EV68A

  • Page 1 21264/EV68A Microprocessor Hardware Reference Manual Part Number: DS–0038B–TE This manual is directly derived from the internal 21264/EV68A Specifications, Revi- sion 1.1. You can access this hardware reference manual in PDF format from the following site: ftp://ftp.compaq.com/pub/products/alphaCPUdocs Revision/Update Information: Revision 1.1, March 2002...
  • Page 2 © Compaq Computer Corporation 2002. All rights reserved. Printed in the U.S.A. COMPAQ, the Compaq logo, the Digital logo, and VAX Registered in United States Patent and Trademark Office. Pentium is a registered trademark of Intel Corporation. Other product names mentioned herein may be trademarks and/or registered trademarks of their respective compa- nies.

  • Page 3: Table Of Contents

    21264/EV68A Microprocessor Features ........
  • Page 4 21264/EV68A Microprocessor Logic Symbol ........
  • Page 5 21264/EV68A-to-System Commands ........
  • Page 6 Hardware Structure of Explicitly Written IPRs ....... . 21264/EV68A Hardware Reference Manual 5–4...
  • Page 7 PLL Output Clocks ........... . . 21264/EV68A Hardware Reference Manual 6–9...
  • Page 8 Differential 21264/EV68A Clocks ........
  • Page 9 Opcodes Reserved for Compaq........
  • Page 10 D.40 Restriction 44: Not Applicable to the 21264/EV68A ....... D.41 Restriction 45: No HW_JMP or JMP Intructions in PALcode .
  • Page 11 21264/EV68A Microprocessor Logic Symbol ........
  • Page 12 Example for Initializing Bcache ..........7–5 21264/EV68A Reset State Machine State Diagram ....... 10–1 Type 1 Heat Sink .
  • Page 13 21264/EV68A Signal Descriptions by Function ........
  • Page 14 ProfileMe Mode PCTR_CTL Input Select Fields ........7–1 21264/EV68A Reset State Machine Major Operations ......7–2 Signal Pin Reset State .
  • Page 15 21264/EV68A Error Detection Mechanisms ........
  • Page 17 EV68A microprocessor (referred to as the 21264/EV68A). Content This manual contains the following chapters and appendixes: Chapter 1, Introduction, introduces the 21264/EV68A and provides an overview of the Alpha architecture. Chapter 2, Internal Architecture, describes the major hardware functions and the inter- nal chip architecture.
  • Page 18 PALcode. Appendix E, 21264/EV68A-to-Bcache Pin Interface, provides the pin interface between the 21264/EV68A and Bcache SSRAMs. The Glossary lists and defines terms associated with the 21264/EV68A. An Index is provided at the end of the document. Documentation Included by Reference The companion volume to this manual, the Alpha Architecture Reference Manual, Fourth Edition, can be accessed from the following website: ftp.compaq.com/...

  • Page 19: Terminology And Conventions

    Bits and fields are cleared when read. Unless otherwise specified, such bits cannot be written. Reserved Bits and fields are reserved by Compaq and should not be used; however, zeros can be written to reserved fields that cannot be masked. Read Only The value may be read by software.
  • Page 20 [:]. For example, [9:7,5,2:0] specifies bits 9,8,7,5,2,1, and 0. Similarly, single bits are frequently indicated with square brackets. For example, [27] specifies bit 27. See also Field Notation. Caution Cautions indicate potential damage to equipment or loss of data. 21264/EV68A Hardware Reference Manual...
  • Page 21 The bit range may, but need not necessarily, correspond to the bit Extent in the register. See the explanation above Table 5–1 for more information. Signal Names The following examples describe signal-name conventions used in this document. 21264/EV68A Hardware Reference Manual Words Bytes ½...
  • Page 22 Boldface, mixed-case type denotes signal names that are assigned internal and external to the 21264/EV68A (that is, the signal traverses a chip interface pin). When a signal has high and low assertion states, a lower- case italic x represents the assertion states.
  • Page 23 Do not care. A capital X represents any valid value. 21264/EV68A Hardware Reference Manual xxiii...

  • Page 25: Introduction

    This use of resources makes it easy to build implementations that issue multiple instructions every CPU cycle. The 21264/EV68A uses a set of subroutines, called privileged architecture library code (PALcode), that is specific to a particular Alpha operating system implementation and hardware platform.

  • Page 26: Addressing

    1.1.1 Addressing The basic addressable unit in the Alpha architecture is the 8-bit byte. The 21264/ EV68A supports a 48-bit or 43-bit virtual address (selectable under IPR control). Virtual addresses as seen by the program are translated into physical memory addresses by the memory-management mechanism.

  • Page 27: 21264/Ev68A Microprocessor Features

    The 21264/EV68A can issue four Alpha instructions in a single cycle, thereby minimiz- ing the average cycles per instruction (CPI). A number of low-latency and/or high- throughput features in the instruction issue unit and the onchip components of the mem- ory subsystem further reduce the average CPI.
  • Page 28 • A 2.0-V external interface. Refer to Chapter 9 for 21264/EV68A dc and ac electrical characteristics. Refer to the lpha Architecture Handbook, Version 4, Appendix E, for waivers and any other implementation-dependent information. Introduction 1–4...

  • Page 29: Internal Architecture

    This chapter provides both an overview of the 21264/EV68A microarchitecture and a sys- tem designer’s view of the 21264/EV68A implementation of the Alpha architecture. The combination of the 21264/EV68A microarchitecture and privileged architecture library code (PALcode) defines the chip’s implementation of the Alpha architecture. If a certain piece of hardware seems to be “architecturally incomplete,”...

  • Page 30: Instruction Fetch, Issue, And Retire Unit

    There can be up to 80 instructions, in 20 successive fetch slots, in flight between the register rename mappers and the end of the pipeline. The VPC logic contains a 20-entry table to store these fetched VPC addresses. Internal Architecture 2–2 21264/EV68A Hardware Reference Manual...

  • Page 31: Ev68A Block Diagram

    Figure 2–1 21264/EV68A Block Diagram Ibox Virtual Address Fetch Unit Next Address Queue Branch Predictor Integer Issue Queue (20 Entries) Ebox Address ALU 0 UNIT UNIT (L0) (U0) (U1) Integer Registers 0 Integer Registers 1 (80 Registers) (80 Registers) Mbox...

  • Page 32: Branch Predictor

    4K entry table of 2-bit saturating counters. The value of the saturating counter determines the predication, taken/not-taken, of the cur- rent branch. Internal Architecture 2–4 Choice Predictor Branch Address FM-05810.AI4 Local Index Predictor 1K x 3 Local Branch Prediction FM-05811.AI4 21264/EV68A Hardware Reference Manual...

  • Page 33: Global Predictor

    The ITB supports an 8-bit ASN and contains an ASM bit. The Icache is virtually addressed and contains the access-check information, so the ITB is accessed only for Istream references that miss in the Icache. Istream transactions to I/O address space are UNDEFINED. 21264/EV68A Hardware Reference Manual 21264/EV68A Microarchitecture Global Predictor 4K x 2 FM-05812.AI4...

  • Page 34: Integer Issue Queue

    2.1.1.6 Integer Issue Queue The 20-entry integer issue queue (IQ), associated with the integer execution units (Ebox), issues the following types of instructions at a maximum rate of four per cycle: Internal Architecture 2–6 21264/EV68A Hardware Reference Manual...

  • Page 35: Floating-Point Issue Queue

    The add and multiply arbiters pick one requester per cycle, while the store pipeline arbiter picks two requesters per cycle, one for each store pipeline. 21264/EV68A Hardware Reference Manual 21264/EV68A Microarchitecture Internal Architecture 2–7...

  • Page 36: Exception And Interrupt Logic

    “subclusters”, named upper (U) and lower (L). Fig- ure 2–6 shows the integer execution unit. In the figure, iop_wr is the cross-cluster bus for moving integer result values between clusters. Internal Architecture 2–8 21264/EV68A Hardware Reference Manual...

  • Page 37: Integer Execution Unit-Clusters 0 And 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2-7 Floating-Point Execution Units

    • One fully-pipelined unit (located in U0), with 3-cycle latency, that executes the fol- lowing instructions: – CTLZ, CTPOP, CTTZ – PERR, MINxxx, MAXxxx, UNPKxx, PKxx 21264/EV68A Hardware Reference Manual 21264/EV68A Microarchitecture iop_wr iop_wr Register iop_wr iop_wr Load/Store Data Load/Store Data FM-05643.AI4...

  • Page 38: Floating-Point Execution Unit

    The 72 Fbox register file entries contain storage for the values of the 31 Alpha floating- point registers (F31 is not stored) and 41 values written by instructions that have not been retired. Internal Architecture 2–10 LK98-0004A 21264/EV68A Hardware Reference Manual...

  • Page 39: External Cache And System Interface Unit

    The duplicate Dcache tag (DTAG) array holds a duplicate copy of the Dcache tags and is used by the Cbox when processing Dcache fills, Icache fills, and system port probes. 2.1.5 Onchip Caches The 21264/EV68A contains two onchip primary-level caches. 2.1.5.1 Instruction Cache The instruction cache (Icache) is a 64KB virtual-addressed, 2-way set-predict cache.

  • Page 40: Data Cache

    The Dcache contains two sets, each with 512 rows containing 64-byte blocks per row (that is, 32K bytes of data per set). The 21264/EV68A requires two additional bits of virtual address beyond the bits that specify an 8KB page, in order to specify a Dcache row index.

  • Page 41: Load Queue

    2.2 Pipeline Organization The 7-stage pipeline provides an optimized environment for executing Alpha instruc- tions. The pipeline stages (0 to 6) are shown in Figure 2–8 and described in the follow- ing paragraphs. 21264/EV68A Hardware Reference Manual Pipeline Organization Internal Architecture 2–13...
  • Page 42 ALU Shifter Multiplier Integer Register File Address Address 64KB Data Cache Floating-Point Add, Divide, Floating- and Square Root Point Register Floating-Point File Multiply 21264/EV68A Hardware Reference Manual System (64 Bits) Interface Cache Unit (128 Bits) Physical Address (44 Bits) FM-05575.AI4...

  • Page 43: Pipeline Organization

    IQ or FQ two cycles after they are issued. For example, if an instruction is issued in cycle n, it remains in the FQ or IQ in cycle n+1 but does not request service, and is deleted in cycle n+2. 21264/EV68A Hardware Reference Manual Internal Architecture 2–15...

  • Page 44: Pipeline Abort Delay (Gclk Cycles)

    Internal Architecture 2–16 Penalty (Cycles) Comments Integer or floating-point conditional branch misprediction. Memory format JSR or HW_RET. Load-load order or store-load order. — — — — 13+latency Add latency of instruction. See Section 2.3.3 for instruction latencies. 21264/EV68A Hardware Reference Manual...

  • Page 45: Instruction Group Definitions

    None 21264/EV68A Hardware Reference Manual Instruction Issue Rules Instruction Type All integer load instructions All floating-point load instructions All integer store instructions LDA, LDAH WH64, ECB, WMB RPCC RS, RC HW_MTPR, HW_MFPR...

  • Page 46: Ebox Slotting

    U E E E U L U L U E E L U L U L U E E U 21264/EV68A Hardware Reference Manual Slotting 3 2 1 0 L L L L L L L U L L U U...
  • Page 47 L E U U L L E E L L E L L L E U L L L E — 21264/EV68A Hardware Reference Manual Slotting Instruction Class 3 2 1 0 3 2 1 0 L L U U...

  • Page 48: Instruction Latencies

    Possible 1-cycle Ebox cross-cluster delay. imul Possible 1-cycle Ebox cross-cluster delay. imisc Possible 1-cycle Ebox cross-cluster delay. fcbr — Does not produce register value. Internal Architecture 2–20 Ebox IPRs = 1. Ibox and Mbox IPRs = 3. 21264/EV68A Hardware Reference Manual...

  • Page 49: Instruction Retire Rules

    Instruction Class Retire Stage Integer conditional branch Integer multiply 7/13 Integer operate Memory Floating-point add Floating-point multiply 21264/EV68A Hardware Reference Manual Instruction Retire Rules Comments — Latency is 13 cycles for the MUL/V instruction. — — — — Internal Architecture...

  • Page 50: Floating-Point Divide/Square Root Early Retire

    11 plus 9 from Table 2–4. Latency is 11 if hardware detects that no exception is possible (see Section 2.4.1). Branch instruction mispredict is reported in stage 7. JSR instruction mispredict is reported in stage 8. ) OR (EXP <= 2 ) OR (EXP <= 382 21264/EV68A Hardware Reference Manual...

  • Page 51: Load Instructions To R31 And F31

    See Table 6–3. 2.6.2 Prefetch with Modify Intent: LDS Instruction The 21264/EV68A processes an LDS instruction, with F31 as the destination, as a prefetch with modify intent transaction (ReadBlkMod command). If the transaction hits a dirty Dcache block, the instruction is dismissed. Otherwise, the addressed cache block is allocated into the Dcache for write access, with its dirty and modified bits set.

  • Page 52: Prefetch, Evict Next: Ldq And Hw_Ldq Instructions

    2.7 Special Cases of Alpha Instruction Execution This section describes the mechanisms that the 21264/EV68A uses to process irregular instructions in the Alpha instruction set, and cases in which the 21264/EV68A pro- cesses instructions in a non-intuitive way. 2.7.1 Load Hit Speculation The latency of integer load instructions that hit in the Dcache is three cycles.

  • Page 53: Pipeline Timing For Floating-Point Load Instructions

    Dcache, even if they are not dependent on the load data. However, if software misses are likely, the 21264/EV68A can still benefit from scheduling the instruction stream for Dcache miss latency. The 21264/EV68A includes a saturating counter that is incremented when load instructions hit and is decremented when load instructions miss.

  • Page 54: Floating-Point Store Instructions

    IQ. 2.7.3 CMOV Instruction For the 21264/EV68A, the Alpha CMOV instruction has three operands, and so pre- sents a special case. The required operation is to move either the value in register Rb or the value from the old physical destination register into the new destination register, based upon the value in Ra.

  • Page 55: Memory And I/O Address Space Instructions

    The Fbox add pipeline executes floating-point CMOV instructions as two distinct 4-cycle latency operations. 2.8 Memory and I/O Address Space Instructions This section provides an overview of the way the 21264/EV68A processes memory and I/O address space instructions. The 21264/EV68A supports, and internally recognizes, a 44-bit physical address space that is divided equally between memory address space and I/O address space.

  • Page 56: Memory Address Space Store Instructions

    Dcache. This restriction assists in STx_C instruction and Dcache ECC processing. Internal Architecture 2–28 Load Byte/Word Load Longword No merge No merge No merge Merge up to 32 bytes No merge No merge 21264/EV68A Hardware Reference Manual Load Quadword No merge No merge Merge up to 64 bytes...

  • Page 57: I/O Address Space Store Instructions

    1024 cycles. After the IOWB merge register has closed its merge window, the Cbox sends I/O space store requests offchip in the order that they were received from the Mbox. 21264/EV68A Hardware Reference Manual Memory and I/O Address Space Instructions Store...

  • Page 58: Maf Memory Address Space Merging Rules

    Because all memory transactions are to 64-byte blocks, efficiency is improved by merg- ing several small data transactions into a single larger data transaction. Table 2–9 lists the rules the 21264/EV68A uses when merging memory transactions into 64-byte natu- rally aligned data block transactions. Rows represent the merged instruction in the MAF and columns represent the new issued transaction.

  • Page 59: Mbox Order Traps

    The 21264/EV68A maintains the default I/O instruction ordering as shown in Table 2– 11 (assume address X and address Y are different). Table 2–11 I/O Reference Ordering First Instruction in Pair Load I/O to address X Load I/O to address X...

  • Page 60: Other Mbox Replay Traps

    15 (saturating the counter), the Cbox will stall MAF and IOWB processing until at least one of the pending transactions has been committed. Probe pro- cessing is not interrupted by the state of this counter. Internal Architecture 2–32 21264/EV68A Hardware Reference Manual...

  • Page 61: Mb Instruction Processing

    Monitors Cbox CSR MB_CNT[3:0], a 4-bit counter of outstanding committed events. When the counter decrements from one to zero, the Cbox marks the youngest probe queue entry. 21264/EV68A Hardware Reference Manual I/O Write Buffer and the WMB Instruction Internal Architecture...

  • Page 62: Tb Fill Flow

    Table 2–13 TB Fill Flow Example Sequence 2 Write Datai Write PTEi The 21264/EV68A processes Dstream loads to the PTE by injecting, in hardware, some memory barrier processing between the PTE transaction and any subsequent load or store instruction. This is accomplished by the following mechanism: 1.

  • Page 63: Performance Measurement Support-Performance Counters

    SYSBUS_MB_ENABLE), the Cbox sig- nals the Ibox to clear IPR scoreboard bit [0]. The 21264/EV68A uses a similar mechanism to process Istream TB misses and fills to the PTE for the Istream. 1. The integer queue issues a HW_LD instruction with VPTE.

  • Page 64: Floating-Point Control Register

    UNFD is used in conjunction with UNDZ as follows: UNDZ [60] Underflow to zero. When UNDZ is set together with UNFD, underflow traps are disabled and the 21264/EV68A places a true zero in the destination register. See UNFD, above. Internal Architecture 2–36...

  • Page 65: Amask And Implver Instruction Values

    Reserved [47:0] — — Alpha architecture FPCR bit 47 (DNOD) is not implemented by the 21264/EV68A. 2.15 AMASK and IMPLVER Instruction Values The AMASK and IMPLVER instructions return the supported architecture extensions and processor type , respectively. 21264/EV68A Hardware Reference Manual...

  • Page 66: Amask

    Design Examples 2.15.1 AMASK The 21264/EV68A returns the AMASK instruction values provided in Table 2–15. The I_CTL register reports the 21264/EV68A pass level (see I_CTL[CHIP_ID], Section 5.2.15). Table 2–15 21264/EV68A AMASK Values 21264/EV68A Pass Level See I_CTL[CHIP_ID], Table 5–11 The AMASK bit definitions provided in Table 2–15 are defined in Table 2–16.

  • Page 67: Typical Uniprocessor Configuration

    Each interface controller must employ a duplicate tag store to maintain cache coherency. This system configuration could be used in a networked database server application. Figure 2–13 Typical Multiprocessor Configuration 21264 Cache 21264 Cache 21264/EV68A Hardware Reference Manual 21272 Core 21264 Logic Chipset Address Control Chips Address...

  • Page 69: Hardware Interface

    This chapter contains the 21264/EV68A microprocessor logic symbol and provides information about signal names, their function, and their location. This chapter also describes the mechanical specifications of the 21264/EV68A. It is organized as fol- lows: • The 21264/EV68A logic symbol •...

  • Page 70: 21264/Ev68A Microprocessor Logic Symbol

    21264/EV68A Microprocessor Logic Symbol Figure 3–1 21264/EV68A Microprocessor Logic Symbol 21264 System Interface SysAddIn_L[14:0] SysAddInClk_L SysAddOut_L[14:0] SysAddOutClk_L SysVref SysData_L[63:0] SysCheck_L[7:0] SysDataInClk_H[7:0] SysDataOutClk_L[7:0] SysDataInValid_L SysDataOutValid_L SysFillValid_L Clocks ClkIn_x FrameClk_x EV6Clk_x 2.5 V PLL_VDD Miscellaneous IRQ_H[5:0] ClkFwdRst_H SromData_H Tms_H Trst_L Tick_H Tdi_H...

  • Page 71: 21264/Ev68A Signal Names And Functions

    3.2 21264/EV68A Signal Names and Functions Table 3–1 defines the 21264/EV68A signal types referred to in this section. Table 3–1 Signal Pin Types Definitions Signal Type Inputs I_DC_REF I_DA I_DA_CLK Outputs O_OD O_OD_TP O_PP O_PP_CLK Bidirectional B_DA_OD B_DA_PP Other Spare...
  • Page 72 Bcache burst enable. Bcache tag bits. Tag dirty state bit. During cache write operations, the 21264/ EV68A will assert this signal if the Bcache data has been mod- ified. Bcache tag input clock. The 21264/EV68A uses this input clock to latch the tag information on Bcache read operations.
  • Page 73 Count Description A skew-controlled differential 50% duty cycle copy of the sys- tem clock. It is used by the 21264/EV68A as a reference, or framing, clock. These six interrupt signal lines may be asserted by the system. The response of the 21264/EV68A is determined by the system software.

  • Page 74: Ev68A Signal Descriptions By Function

    BiST. If the Icache BiST passes, the pin is deasserted at the end of the BiST operation; otherwise, it remains high. The 21264/EV68A generates a timeout reset signal if an instruction is not retired within one billion cycles. The 21264/EV68A signals the timeout reset event by output- ting a 256 GCLK cycle wide pulse on TestStat_H.
  • Page 75 Provides an external test point to measure phase alignment of the PLL. A skew-controlled differential 50% duty cycle copy of the system clock. It is used by the 21264/EV68A as a reference, or framing, clock. 2.5-V dedicated power supply for the 21264/EV68A PLL.

  • Page 76: Pin Assignments

    I_DA Pin Assignments The 21264/EV68A package has 587 pins aligned in a pin grid array (PGA) design. There are 380 functional signal pins, 1 dedicated 2.5-V pin for the PLL, 112 ground VSS pins, and 94 VDD pins. Table 3–4 lists the signal pins and their corresponding pin grid array (PGA) locations in alphabetical order for the signal type.
  • Page 77 AJ45 BcData_H_54 AP44 BcData_H_57 AU41 BcData_H_6 BcData_H_62 BB38 BcData_H_65 BcData_H_68 BcData_H_70 BcData_H_73 BcData_H_76 BcData_H_79 BcData_H_81 BcData_H_84 BcData_H_87 21264/EV68A Hardware Reference Manual PGA Location Signal Name BcData_H_107 BcData_H_11 BcData_H_112 AC43 BcData_H_115 AG45 BcData_H_118 AM42 BcData_H_120 AP40 BcData_H_123 BB44 BcData_H_126 BA37 BcData_H_14...
  • Page 78 BB16 NoConnect Reset_L BD16 Spare Spare Spare Spare BC21 SromOE_L BE17 SysAddIn_L_10 BB24 SysAddIn_L_13 BE23 SysAddIn_L_3 BE29 21264/EV68A Hardware Reference Manual PGA Location BcData_H_91 BcData_H_94 BcData_H_97 BcDataInClk_H_0 BcDataInClk_H_3 BcDataInClk_H_6 AH44 BcDataOutClk_H_0 J5 BcDataOutClk_H_3 AR43 BcDataOutClk_L_2 BcLoad_L BcTag_H_22 BcTag_H_25 BcTag_H_28 BcTag_H_31...
  • Page 79 SysData_L_52 AH38 SysData_L_55 AK38 SysData_L_58 AR39 SysData_L_60 AY38 SysData_L_63 AV34 SysData_L_9 SysDataInClk_H_2 SysDataInClk_H_5 SysDataInValid_L BD22 SysDataOutClk_L_2 AG7 21264/EV68A Hardware Reference Manual PGA Location Signal Name SysAddIn_L_6 BD28 SysAddIn_L_9 BC25 SysAddOut_L_1 BE39 SysAddOut_L_12 BC31 SysAddOut_L_2 BD36 SysAddOut_L_5 AY32 SysAddOut_L_8 BB32 SysCheck_L_0...

  • Page 80: Pin List Sorted By Pga Location

    Spare AJ43 Spare BcData_H_20 AK42 BcData_H_52 BcData_H_21 SysData_L_23 AM40 SysCheck_L_6 ClkIn_H AN41 BcData_H_55 BcData_H_23 AP44 BcData_H_54 21264/EV68A Hardware Reference Manual PGA Location SysDataOutClk_L_7 AW39 SysVref BA25 Tdo_H BB20 Trst_L AY20 PGA Location Signal Name BcTag_H_33 BcAdd_H_6 BcData_H_96 BcData_H_65 AA43 BcData_H_111...
  • Page 81 BA27 SysAddIn_L_5 BA33 SysAddOut_L_4 BA43 BcData_H_59 BB10 BcData_H_30 BcData_H_25 BB24 SysAddIn_L_10 BB32 SysAddOut_L_8 BcData_H_27 BcData_H_28 BC17 SromData_H 21264/EV68A Hardware Reference Manual PGA Location Signal Name Spare AR45 BcData_H_119 AT38 SysData_L_59 AT44 BcCheck_H_14 AU41 BcData_H_57 AV10 SysData_L_28 AV18 IRQ_H_3 AV28 SysAddOut_L_13...
  • Page 82 BcTag_H_38 BcDataWr_L BcData_H_97 BcData_H_101 BcTag_H_20 BcTag_H_39 BcData_H_5 SysDataInClk_H_4 BcCheck_H_12 SysData_L_0 BcTag_H_37 SysData_L_32 SysData_L_36 Spare SysData_L_1 BcTagShared_H BcData_H_32 21264/EV68A Hardware Reference Manual PGA Location Signal Name BcData_H_91 BC37 BcCheck_H_7 BcDataInClk_H_3 BD12 PllBypass_H NoConnect BD28 SysAddIn_L_6 BD36 SysAddOut_L_2 BD42 BcData_H_61 BE13 IRQ_H_1...
  • Page 83 BcData_H_10 BcData_H_106 BcCheck_H_4 SysData_L_8 SysData_L_41 BcData_H_74 SysData_L_42 BcDataInClk_H_1 BcData_H_12 BcData_H_108 BcData_H_77 BcData_H_109 Spare SysData_L_13 SysData_L_46 BcData_H_78 SysData_L_47 21264/EV68A Hardware Reference Manual PGA Location Signal Name BcData_H_38 SysData_L_5 BcTag_H_21 BcTagOE_L SysData_L_34 BcData_H_102 BcData_H_8 BcDataOutClk_H_0 BcDataOutClk_L_0 SysData_L_7 SysCheck_L_4 SysCheck_L_0 SysData_L_40 BcCheck_H_8 BcData_H_41...

  • Page 84: Ground And Power (Vss And Vdd) Pin List

    Pin Assignments Table 3–6 lists the 21264/EV68A ground and power (VSS and VDD, respectively) pin list. Table 3–6 Ground and Power (VSS and VDD) Pin List Signal PGA Location AA45 AG43 AR41 AW37 AW41 BA35 BA41 BC45 AB40 AH42 AK40...

  • Page 85: Mechanical Specifications

    Mechanical Specifications This section shows the 21264/EV68A mechanical package dimensions without a heat sink. For heat sink information and dimensions, refer to Chapter 10. Figure 3–2 shows the package physical dimensions without a heat sink. Figure 3–2 Package Dimensions 04 06 08 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 42 44 27.94 mm...

  • Page 86: 21264/Ev68A Packaging

    21264/EV68A Packaging 21264/EV68A Packaging Figure 3–3 shows the 21264/EV68A pinout from the top view with pins facing down. Figure 3–3 21264/EV68A Top View (Pin Down) 42 40 38 36 34 32 30 28 26 24 22 20 18 16 14 12 10 08 06 04 02 Hardware Interface 3–18...

  • Page 87: Ev68A Bottom View (Pin Up)

    Figure 3–4 shows the 21264/EV68A pinout from the bottom view with pins facing up. Figure 3–4 21264/EV68A Bottom View (Pin Up) 04 06 08 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 42 44...

  • Page 89: Cache And External Interfaces

    Bcache port • Interrupts Chapter 3 lists and defines all 21264/EV68A hardware interface signal pins. Chapter 9 describes the 21264/EV68A hardware interface electrical requirements. 4.1 Introduction to the External Interfaces A 21264/EV68A-based system can be divided into three major sections: •...
  • Page 90 GCLK. The period of BcDataOutClk_x[3:0] is a programmable mul- tiple of GCLK. – The Bcache turns the BcDataOutClk_x[3:0] clocks around and returns them to the 21264/EV68A as BcDataInClk_H[7:0]. Likewise, BcTagOutClk_x returns as BcTagInClk_H. – The Bcache interface supports a 64-byte block size.

  • Page 91: System Interface

    Figure 4–1 21264/EV68A System and Bcache Interfaces SysAddIn_L[14:0] SysAddInClk_L SysAddOut_L[14:0] SysAddOutClk_L SysVref SysData_L[63:0] SysCheck_L[7:0] SysDataInClk_H[7:0] SysDataOutClk_L[7:0] SysDataInValid_L SysDataOutValid_L SysFillValid_L BcAdd_H[23:4] [23:4] 21264 BcLoad_L BcData_H[127:0] BcCheck_H[15:0] BcDataInClk_H[7:0] BcDataOutClk_ x [3:0] BcDataOE_L BcDataWr_L BcTag_H[42:20] BcTagInClk_H BcTagOutClk_ x BcVref BcTagWr_L BcTagOE_L BcTagValid_H BcTagDir ty_H...

  • Page 92: Commands And Addresses

    Bcache. The 21264/EV68A supports a Bcache from 1MB to 16MB, with 64- byte blocks. A 128-bit data bus is used for transfers between the 21264/EV68A and the Bcache. The Bcache must be comprised of synchronous static RAMs (SSRAMs) and must contain either one, two, or three internal registers.

  • Page 93: Translation Of Internal References To External Interface Reference

    WH64 Memory WH64 Memory WH64 I/O ECB Memory ECB I/O MB/WMB TBFill Flows 21264/EV68A Hardware Reference Manual Physical Address Considerations BcHit Status and Action Dcache hit, done. Bcache hit, done. Miss, generate RdBlk command. RdBytes, RdLWs, or RdQWs based on size.
  • Page 94 CSR BC_CLEAN_VICTIM is set and there is a Bcache index match valid but not dirty in the 21264/EV68A cache system. WrVictimBlk is issued for any Bcache match of the ECB address that is dirty in the 21264/EV68A cache sys- tem.

  • Page 95: System Duplicate Tag Stores

    The 21264/EV68A Cbox provides control signals and an interface for a second-level cache (Bcache). The 21264/EV68A supports a Bcache from 1MB to 16MB, with 64-byte blocks. A 128- bit bidirectional data bus is used for transfers between the 21264/EV68A and the Bcache.

  • Page 96: Victim Data Buffer

    • Filtering out all probe misses from the 21264/EV68A cache system If a probe misses in the 21264/EV68A cache system (Bcache miss and VAF miss), the 21264/EV68A stalls probe processing with the expectation that a SysDc fill will allo- cate this block. Because of this, in duplicate tag mode, the 21264/EV68A can never generate a probe miss response.

  • Page 97: Ev68A-Supported Cache Block States

    The 21264/EV68A requires the system to allow only one change to a block at a time. This means that if the 21264/EV68A gains the bus to read or write a block, no other node on the bus should be allowed to access that block until the data has been moved.

  • Page 98: Cache Block State Transitions

    State Name Description Clean/Shared This 21264/EV68A holds a read-only copy of the block, and at least one other agent in the system may hold a copy of the block. Upon eviction, the block is not written to memory. Dirty This 21264/EV68A holds a read-write copy of the block, and must write it to memory after it is evicted from the cache.

  • Page 99: Using Sysdc Commands

    Rdiox commands are noncached references to I/O address space. • Evict and STCChangeToDirty commands are generated by ECB and STx_C instructions, respectively. Table 4–5 shows the system responses to 21264/EV68A commands and 21264/EV68A reactions. Table 4–5 System Responses to 21264/EV68A Commands and Reactions 21264/EV68A...
  • Page 100 The original data in the Dcache is replaced with the filled data. The block is not writable, so the 21264/EV68A will retry the store instruc- tion and generate another ChxToDirty class command. To avoid a potential livelock situation, the STC_ENABLE CSR bit must be set.

  • Page 101: Dcache States And Duplicate Tags

    Both commands allocate buffers in the VAF (victim address file). This buffer is a coherent part of the 21264/EV68A cache system. Write data con- trol and deallocation of the VAF can be directly controlled by using the SysDc Write- Data and ReleaseBuffer commands.

  • Page 102: Lock Mechanism

    Dcache without interrupting load/store instruction processing in the processor core. 4.6 Lock Mechanism The 21264/EV68A does not contain a dedicated lock register, nor are system compo- nents required to do so. When a load-lock (LDx_L) instruction executes, data is accessed from the Dcache or Bcache.

  • Page 103: In-Order Processing Of Ldx_L/Stx_C Instructions

    Thus, a STx_C instruction is not issued until the older LDx_L is out of the IQ. 4.6.2 Internal Eviction of LDx_L Blocks The 21264/EV68A prevents the eviction of cache blocks in the Dcache due to either of the following references: •...

  • Page 104: Managing Speculative Store Issues With Multiprocessor Systems

    8-bit primary counter that tracks branch mispre- dicts and conditional branch retires, and a backup counter that places the 21264/ EV68A in conservative mode for a period of 16K cycles every 2 million cycles. The 8-bit counter is enabled by placing M_CTL[SMC] in periodic conservative mode.

  • Page 105: System Port Pins

    SysFillValid_L IRQ_H[5:0] 4.7.1 System Port Pins Table 3–1 defines the 21264/EV68A signal types referred to in this section. Table 4–6 lists the system port pin groups along with their type, number, and functional descrip- tion. Table 4–6 System Port Pins...

  • Page 106: Programming The System Interface Clocks

    System Port 4.7.2 Programming the System Interface Clocks The system forwarded clocks are free running and derived from the 21264/EV68A GCLK. The period of the system forwarded clocks is controlled by three Cbox CSRs, based on the bit-rate ratio (similar to the Bcache bit-rate ratio) except that all transfers are dual-data.

  • Page 107: 21264/Ev68A-To-System Commands

    This section describes the 21264/EV68A-to-system commands format and operation. The command, address, ID, and mask bits are transmitted in four consecutive cycles on SysAddOut_L[14:0]. The 21264/EV68A sends the command information in one of the two following modes as selected by the Cbox CSR bit.

  • Page 108: Page Hit Mode

    M1 and M2 are not asserted simultaneously. Reporting probe results as soon as possible is critical to high-speed operation, so when a result is known the 21264/EV68A uses the ear- liest opportunity to send an M signal to the system. M bit assertion can occur either in a valid command or a NZNOP.

  • Page 109: 21264/Ev68A-To-System Commands Descriptions

    Illegal combination Page hit mode + both SysAddOut_L[1:0] are unused Function The 21264/EV68A drives this command on idle cycles during reset. After the clock forward reset period, the first NZNOP is generated and this command is no longer generated. Returns probe status and ID number of the VDB entry holding the requested cache block.
  • Page 110 System Port Table 4–14 21264/EV68A-to-System Commands Descriptions (Continued) Command Command [4:0] ReadBlkMod 10001 ReadBlkI 10010 FetchBlk 10011 ReadBlkSpec 10100 ReadBlkModSpec 10101 ReadBlkSpecI 10110 FetchBlkSpec 10111 ReadBlkVic 11000 ReadBlkModVic 11001 ReadBlkVicI 11010 WrVictimBlk 00100 CleanVictimBlk 00101 Evict 00110 ReadBytes 01000 ReadLWs...

  • Page 111: Programming Inval_To_Dirty_Enable[1:0]

    Dcache with a status of dirty/shared, clean/ shared, or clean respectively. Table 4–15 Programming INVAL_TO_DIRTY_ENABLE[1:0] INVAL_TO_DIRTY_ENABLE[1:0] 21264/EV68A Hardware Reference Manual Cbox Action WH64 instructions are converted to RdModx commands at the interface. Beyond this point, no other agent sees the WH64 instruction. This mode is useful for microprocessors that do not want to support InvalToDirty transac- tions.

  • Page 112: Proberesponse Commands (Command[4:0] = 00001)

    LDx_L/STx_C sequence. 4.7.5 ProbeResponse Commands (Command[4:0] = 00001) The 21264/EV68A responds to system probes that did not miss with a 4-cycle transfer on SysAddOut_L[14:0]. As shown in Table 4–14, the Command[4:0] field for a Prob- eResponse command equals 00001.

  • Page 113: Sysack And 21264/Ev68A-To-System Commands Flow Control

    4.7.6 SysAck and 21264/EV68A-to-System Commands Flow Control Controlling the flow of 21264/EV68A-to-system commands is a joint task of the 21264/EV68A and the system. The flow is controlled using the A bit, which is asserted by the system, and the Cbox CSR SYSBUS_ACK_LIMIT[4:0] counter. The counter has the following properties: •...

  • Page 114: System-To-21264/Ev68A Commands

    System Port • There is no mechanism for the system to reject a 21264/EV68A-to-system com- mand. ProbeResponse, VDBFlushReq, NOP, NZNOP, and RdBlkxSpec (with a clear RV bit) commands do not require a response from the system. Systems must provide adequate resources for responses to all probes sent to the 21264/EV68A.

  • Page 115: System-To-21264/Ev68A Probe Commands Fields Descriptions

    Probe type and next tag state (see Tables 4–21 and 4–22). SysDc[4:0] Controls data movement in and out of the 21264/EV68A. See Table 4–24 for a list of data movement types. Clears the victim or I/O write buffer (IOWB) valid bit specified in ID[3:0].

  • Page 116: Data Transfer Commands (Two Cycles)

    21264/EV68A queue. The 21264/EV68A removes probes from the internal probe queue when the probe response is sent. The 21264/EV68A expects to hit in cache on a probe response, so it always fetches a cache block from the Bcache on system probes. This can become a performance prob- lem for systems that do not monitor the Bcache tags, so the 21264/EV68A provides Cbox CSR PRB_TAG_ONLY[0], which only accesses Bcache tags for system probes.

  • Page 117: Sysdc[4:0] Field Description

    010xx The A bit in the first cycle indicates that the command is acknowledged. When A = 1, the 21264/EV68A decrements its command outstanding counter, but the A bit is not neces- sarily related to the current SysDc command. Probe commands can combine a SysDc command along with MBDone. In that event, the probe is considered ahead of the SysDc command.

  • Page 118: Data Movement In And Out Of The 21264/Ev68A

    (the last cycle) to the delivery of the SysDc command to the processor. 4.7.8 Data Movement In and Out of the 21264/EV68A There are two modes of operation for data movement in and out of the 21264/EV68A: fast mode and fast mode disable. The data movement mode is selected using Cbox CSR FAST_MODE_DISABLE[0].

  • Page 119: Fast Data Mode

    Because there is a bandwidth difference between address (4 cycles) and data (8 cycles) transfers, the 21264/EV68A tries to fully use fast data mode by delaying the next SysAddOut write command until a fast data mode slot is available on the SysDataOut bus.

  • Page 120: Fast Transfer Timing Example

    2. The 21264/EV68A drivers stay off until the last piece of fill data is received, or a new SysDc write command overrides the current SysDc fill command. It is the responsibility of the external system to schedule SysDc fill or write commands so that there is no conflict on the SysData bus.

  • Page 121: Fast Data Disable Mode

    21264/EV68A is preceded by a SysDc command. The 21264/ EV68A drivers are only enabled for the duration of an 8-cycle transfer of data from the 21264/EV68A to the system. Systems must ensure that there is no overlap of enabled drivers and that there is adequate settle time on the SysData bus.

  • Page 122: Sysdatainvalid_L And Sysdataoutvalid_L

    BPHASE, before the start of a write transfer, and disabled on the succeeding GCLK BPHASE at the end of the write transfer. The write data is enveloped by the 21264/ EV68A drivers to guarantee that every data transfer has the same data valid win- dow.

  • Page 123: Sysfillvalid_L

    If SysFillValid_L is asserted at D1 time, the fill will continue uninterrupted. If it is not asserted, the 21264/EV68A cancels the fill, but expects all eight QWs of data to arrive at its system bus before continuing to the next fill. Also, the 21264/EV68A maintains the state of the MAF, expecting another valid fill to the same MAF entry.

  • Page 124: Data Wrapping

    SysAddOut_L[5:3] STByte/Word The order in which data is provided to the 21264/EV68A (for a memory or I/O fill) or moved from the 21264/EV68A (write victims or probe reads) can be determined by the system. The system chooses to reflect back the same low-order address bits and the cor- responding octaword found in the SysAddOut field or the system chooses any other starting point within the block.

  • Page 125: System Wrap And Deliver Data

    QW pointed to by the 21264/EV68A; however, some systems may find it more beneficial to begin the transfer elsewhere. The system must always indicate the starting point to the 21264/EV68A. The wrap order for subsequent QWs is interleaved.

  • Page 126: Nonexistent Memory Processing

    I/O space. However, unlike the earlier Alpha microprocessor implementa- tions, the 21264/EV68A can generate speculative references to memory space. To accommodate the speculative nature of the 21264/EV68A, the system must not gener- ate or lock error registers because of speculative references. The 21264/EV68A trans-...

  • Page 127: Ev68A Commands With Nxm Addresses And System Response

    21264/EV68A provides an all ones pattern as a signature for this bug. The NXM block is not cached in the Dcache or Bcache.

  • Page 128: Ordering Of System Port Transactions

    • Probe processing can stall inside the 21264/EV68A when the probe entry index matches PA[19:6] of a previous probe entry in the VAF. •...

  • Page 129: 21264/Ev68A Response To System Probe And In-Flight Command Interaction

    This case assumes that a SetDirty command has been sent to the system environment SharedToDirty because of a store instruction that hit in the 21264/EV68A caches and that another proces- sor has performed a load/store instruction to the same address. The 21264/EV68A pro- vides MAF hit information so that the system can correctly respond to the Set/Dirty command.

  • Page 130: System Probes And Sysdc Commands

    Bcache size can be 1MB, 2MB, 4MB, 8MB, or 16MB. The Bcache port has a 144-bit data bus that is used for data transfers between the 21264/EV68A and the Bcache. All Bcache control and address signal lines are clocked synchronously on Bcache clock cycle boundaries.

  • Page 131: Bcache Port Pins

    The Bcache supports the following multiples of the GCLK period: 1.5X (dual-data mode only), 2X, 2.5X, 3X, 3.5X, 4X, 5X, 6X, 7X, and 8X. However, the 21264/EV68A imposes a maximum Bcache clock period based on the SYSCLK ratio. Table 4–35 lists the range of maximum Bcache clock periods.

  • Page 132: Bcache Clocking

    Bcache tag parity bit BcTagInClk_H ⇒ input Int_Data_BcClk ⇒ output Bcache tag shared bit BcTagInClk_H ⇒ input Int_Data_BcClk ⇒ output Bcache tag valid bit BcTagInClk_H ⇒ input Input reference voltage for tag data Int_Index_BcClk Bcache data write enable 21264/EV68A Hardware Reference Manual...

  • Page 133: Setting The Period Of The Cache Clock

    Table 4–37 BC_CPU_CLK_DELAY[1:0] Values BC_CPU_CLK_DELAY[1:0] Value In the 21264/EV68A topology, the index pins are loaded by all the SSRAMs, while the clock and data pins see a limit load. This arrangement requires a relatively large amount of delay between the index pins and the Bcache clock pins to meet the setup constraints at the SSRAMs.

  • Page 134: Program Values To Set The Cache Clock Period (Single-Data)

    With the exception of the 2.5X-SD and 3.5X-SD cases, the clock waveform generated by the 21264/EV68A for the forwarded clocks has a 50-50 duty cycle. In the 2.5X-SD case, the 21264/EV68A produces an asymmetric clock that is high for two GCLK phases and low for three phases.

  • Page 135: Bcache Transactions

    The following three sections describe these Bcache transactions. 4.8.3.1 Bcache Data Read and Tag Read Transactions The 21264/EV68A always reads four pieces of data (64 bytes) from the Bcache during a data read transaction, and always interrogates the tag array on the first cycle. Once started, data read transactions are never cancelled.

  • Page 136: Bcache Data Write Transactions

    SSRAM setup/hold Bcache address latch requirements, a Bcache read command proceeds through the 21264/EV68A Cbox as follows: 1. When the 21264/EV68A clocks out the first address value on the Bcache index pins with the appropriate Int_Add_BcClk value, the Cbox loads the values of Cbox CSR...

  • Page 137: Bubbles On The Bcache Data Bus

    3. The difference between the data delivery (Int_Data_BcClk) and forwarded clocks out provides the setup for the data at the Bcache data flip-flop. 4. For Bcache writes, the 21264/EV68A drivers are enabled on the GCLK BPHASE preceding the start of a write transfer, and disabled on the succeeding GCLK BPHASE at the end of a write transfer.

  • Page 138: Pin Descriptions

    In that case, amend the formula. For example, it is impossible to have rd_wr = 6 in the 1.5x dual-data rate mode case. 4.8.4 Pin Descriptions This section describes the characteristics of the Bcache interface pins. Cache and External Interfaces 4–50 21264/EV68A Hardware Reference Manual...

  • Page 139: Bcadd_H[23:4]

    1111 When the Cbox CSR BC_BANK_ENABLE[0] is not set, the unused BcAdd_H[23:4] pins are tied to zero. For example, when configured as a 4MB cache, the 21264/EV68A never changes BcAdd_H[23:22] from logic zero, and when BC_BANK_ENABLE[0] is asserted, the 21264/EV68A drives the complement of the MSB index on the next higher BcAdd_H pin.

  • Page 140: Control Pin Assertion For Ram_Type B

    BcTagOE_L read data by providing an extra cycle of output enable. Using these Cbox CSRs, late-write nonbursting and dual-data rate SSRAMs can be connected to the 21264/EV68A as described in Appendix E. Cache and External Interfaces 4–52 NOP WA0 WA1 WA2 WA3 NOP...

  • Page 141: Bcdatainclk_H And Bctaginclk_H

    Bcache banking is possible by decoding the index MSB (as determined by Cbox CSR BC_SIZE[3:0]) and asserting Cbox CSR BC_BANK_ENABLE[0]. To facilitate bank- ing, the 21264/EV68A provides the complement of the MSB bit in the next higher unused index bit. For example, when configured as an 8MB cache with banking enabled, the 21264/EV68A drives the inversion of PA[22] on BcAdd_H[23] for use as a chip enable in a banked configuration.

  • Page 142: Interrupts

    They may be asynchronous, are level sensitive, and can be indi- vidually masked by way of the EIE field of the CM_IER IPR. The system designer determines how these signals are used and selects their relative priority. Cache and External Interfaces 4–54 = 0x0 21264/EV68A Hardware Reference Manual...

  • Page 143: Internal Processor Registers

    This chapter describes 21264/EV68A internal processor registers (IPRs). They are sep- arated into the following circuit logic groups: Ebox, Ibox, Mbox, and Cbox. The gray areas in register figures indicate reserved fields. Bit ranges that are coupled with the field name specify those bits in that named field that are included in the IPR.
  • Page 144 1010 0100 DTB_ASN0 0010 0101 DTB_ASN1 1010 0101 MM_STAT 0010 0111 — M_CTL 0010 1000 DC_CTL 0010 1001 DC_STAT 0010 1010 21264/EV68A Hardware Reference Manual MT/MF Latency Issued from Ebox MFPR Access Pipe (Cycles) — — — — — —...

  • Page 145: Ebox Iprs

    CC register may be written and its associated counter enabled and dis- abled. Figure 5–2 shows the cycle counter control register. Figure 5–2 Cycle Counter Control Register CC_ENA COUNTER[31:4] 21264/EV68A Hardware Reference Manual Score- Index Board (Binary)

  • Page 146: Virtual Address Register - Va

    Counter Enable. When set, this bit allows the cycle counter to increment. CC[31:4] may be written by way of this field. Write transactions to CC_CTL result in CC[3:0] being cleared. — — 21264/EV68A Hardware Reference Manual LK99-0010A 3 2 1 LK99-0014A...

  • Page 147: Virtual Address Format Register - Va_Form

    Figure 5–5 Virtual Address Format Register (VA_48 = 0, VA_FORM_32 = 0) VPTB[63:33] VA[42:13] Figure 5–6 shows VA_FORM when VA_CTL(VA_48) equals 1 and VA_CTL(VA_FORM_32) equals 0. 21264/EV68A Hardware Reference Manual Type Description Virtual Page Table Base. See the VA_FORM register section for details.

  • Page 148: Ibox Iprs

    ITB_TAG and ITB_PTE registers are written into the ITB entry. The specific ITB entry that is written is determined by a round-robin algorithm; the algorithm writes to entry number 0 as the first entry after the 21264/EV68A is reset. Figure 5–8 shows the ITB tag array write register.

  • Page 149: Itb Invalidate All Process (Asm=0) Register - Itb_Iap

    Because the Icache is virtually indexed and tagged, it is normally not nec- Note: essary to flush the Icache when paging. Therefore, a write to ITB_IS will not flush the Icache. 21264/EV68A Hardware Reference Manual Ibox IPRs 12 11 10 9 8 7 6 5...

  • Page 150: Profileme Pc Register - Pmpc

    Figure 5–12 Exception Address Register PC[63:2] Internal Processor Registers 5–8 Type Description Address of the profiled instruction Read as zero Indicates that the PC field contains a physical-mode PALmode address 21264/EV68A Hardware Reference Manual 2 1 0 LK99-0018A 2 1 0 LK99-0018A...

  • Page 151: Instruction Virtual Address Format Register - Iva_Form

    IER field and bit[0] corresponds to the processor mode field. A HW_MFPR instruction to this register returns the values in both fields. Figure 5–16 shows the interrupt enable and current processor mode register. 21264/EV68A Hardware Reference Manual 38 37 Internal Processor Registers...

  • Page 152: Software Interrupt Request Register - Sirr

    Performance Counter Interrupt Enables Software Interrupt Enables AST Interrupt Enable When set, enables those AST interrupt requests that are also enabled by the value in ASTER. — — Current Mode Kernel Executive Supervisor User — — 21264/EV68A Hardware Reference Manual LK99-0022A...

  • Page 153: Interrupt Summary Register - Isum

    Usage of ISUM in performance monitoring is described in Section 6.10. Figure 5–18 shows the interrupt summary register. Figure 5–18 Interrupt Summary Register EI[5:0] PC[1:0] SI[15:1] ASTU ASTS ASTE ASTK 21264/EV68A Hardware Reference Manual 29 28 Type Description — — Software Interrupt Requests — — Internal Processor Registers...

  • Page 154: Hardware Interrupt Clear Register - Hw_Int_Clr

    ASTRR bits and whether the processor mode value held in the IER_CM register is greater than or equal to the value for the mode. — — 33 32 31 30 29 28 27 26 25 21264/EV68A Hardware Reference Manual LK99-0025A...

  • Page 155: Exception Summary Register - Exc_Sum

    Dstream exceptions: The REG field contains the register number of either the source specifier (for stores) or the destination specifier (for loads) of the instruction that triggered the trap. Figure 5–20 shows the exception summary register. 21264/EV68A Hardware Reference Manual Type Description —...

  • Page 156: Exception Summary Register

    This bit should be used by the IACV PALcode routine to deter- mine whether the offending I-stream virtual address is latched in the EXC_ADDR register or the VA register. If BAD_IVA is clear, EXC_ADDR contains the address; if BAD_IVA is set, VA con- tains the address. 21264/EV68A Hardware Reference Manual LK99-0026A...

  • Page 157: Pal Base Register - Pal_Base

    The Ibox control register (I_CTL) is a read-write register that controls various Ibox functions. Its contents are cleared by chip reset. Figure 5–22 shows the Ibox control register. 21264/EV68A Hardware Reference Manual Type Description Destination register of load or operate instruction that triggered the trap OR source register of store that triggered the trap.

  • Page 158: Ibox Control Register

    Virtual Page Table Base. See Section 5.1.5 for details. This is a read-only field that supplies the revision ID number for the 21264/EV68A part. 21264/EV68A pass 2.1 ID is 010010 RO,0 Indicates the status of BiST (clear = pass, set = fail), described in Section 11.5.1.
  • Page 159 [13] [12] BP_MODE[1:0] [11:10] SBE[1:0] [9:8] 21264/EV68A Hardware Reference Manual Type Description RW,0 The stWait table is used to reduce load/store order traps. When set, the stWait table is cleared after 64K cycles. When clear, the stWait table is cleared after 16K cycles. See Sec- tion 2.11.

  • Page 160: Ibox Status Register - I_Stat

    PALshadow Register Enable. Enables access to the PALshadow registers. If SDE[1] is set, R4-R7 and R20-R23 are used as PALshadow registers. SDE[0] does not affect 21264/EV68A operation. RW,0 Super Page Mode Enable. Identical to the SPE bits in the Mbox M_CTL SPE[2:0]. See Section 5.3.9.

  • Page 161: Ibox Status Register

    Name Extent Type Reserved [63:41] [40] [39] [38] 21264/EV68A Hardware Reference Manual 41 40 34 33 32 28 27 Description Reserved for COMPAQ. ProfileMe Mispredict Trap. If the I_STAT[TRP] bit is set, this bit indicates that the profiled instruc- tion caused a mispredict trap. JSR/JMP/RET/COR or HW_JSR/...
  • Page 162 When set, indicates that an error in the line predictor set I_STAT[PAR]. I_STAT[LAM] is set only when I_STAT[PAR] is set; is subsequently cleared when I_STAT[PAR] is cleared (when a 1 is written to I_STAT[PAR] ). Reserved for COMPAQ. 21264/EV68A Hardware Reference Manual...

  • Page 163: Icache Flush Register - Ic_Flush

    IPR Index Bit A HW_MFPR from this register returns the values in all of its component bit fields. Figure 5–24 shows the process context register. 21264/EV68A Hardware Reference Manual , this register is selected. Bits [4:0] of the Register Field...

  • Page 164: Process Context Register

    Kernel Mode AST enable register—used to individually enable each of the four AST interrupt requests. The bit order with this field is: User Mode Supervisor Mode Executive Mode Kernel Mode — — 21264/EV68A Hardware Reference Manual 2 1 0 LK99-0032A...

  • Page 165: Performance Counter Control Register - Pctr_Ctl

    Table 5–15 describes the performance counter control register fields. Table 5–15 Performance Counter Control Register Fields Description Name Extent Type Description SEXT(PCTR0_CTL[47]) [63:48] RO 21264/EV68A Hardware Reference Manual Type Description RW,1 Floating-point enable—if clear, floating-point instructions generate FEN exceptions. This bit is set by hardware on reset.
  • Page 166 Reads to this field return zero. Writes to this field are ignored. Selector 0. 0 = Aggregate counting mode 1 = ProfileMe mode See Table 5–16 for more information. Selector 1. Selects counter PCTR0 and PCTR1 modes. See Table 5–16 for more information. 21264/EV68A Hardware Reference Manual...

  • Page 167: Mbox Iprs

    Figure 5–26 shows the DTB tag array write registers 0 and 1. Figure 5–26 DTB Tag Array Write Registers 0 and 1 48 47 VA[47:13] 21264/EV68A Hardware Reference Manual Profiled instruction valid. When set, indicates a nontrapping profiled instruction retired valid. When clear, indicates that a nontrapping profiled instruction was killed after the cycle in which it was mapped.

  • Page 168: Dtb Pte Array Write Registers 0 And 1 - Dtb_Pte0, Dtb_Pte1

    Table 5–17 describes the DTB_ALTMODE register fields. Table 5–17 DTB Alternate Processor Mode Register Fields Description Name Extent Reserved [63:2] Internal Processor Registers 5–26 Type Description — — 21264/EV68A Hardware Reference Manual 9 8 7 LK99-0036A 2 1 0 LK99-0037A...

  • Page 169: Dstream Tb Invalidate All Process (Asm=0) Register - Dtb_Iap

    The DTB entry’s virtual page number matches DTB_IS[47:13] and its ASM bit is set. Figure 5–29 shows the Dstream translation buffer invalidate single registers. Figure 5–29 Dstream Translation Buffer Invalidate Single Registers 48 47 VA[47:13] 21264/EV68A Hardware Reference Manual Type Description Alt_Mode: ALT_MODE[1:0] Mode...

  • Page 170: Dstream Tb Address Space Number Registers 0 And 1 - Dtb_Asn0,1

    Opcode of the instruction that caused the error. HW_LD is displayed as 3 and HW_ST is displayed as 7. This bit is set when a fault-on-write error occurs during a write transaction and PTE[FOW] was set. 21264/EV68A Hardware Reference Manual LK99-0038A 4 3 2 LK99-0039A...

  • Page 171: Mbox Control Register - M_Ctl

    The Mbox control register (M_CTL) is a write-only register. Its contents are cleared by chip reset. Figure 5–32 shows the Mbox control register. Figure 5–32 Mbox Control Register SMC[1:0] SPE[2:0] 21264/EV68A Hardware Reference Manual Type Description This bit is set when a fault-on-read error occurs during a read transaction and PTE[FOR] was set.

  • Page 172: Dcache Control Register - Dc_Ctl

    (see Section 2.6.2) to behave like normal prefetches. Place 21264/EV68A in periodic conservative mode by using an 8-bit counter to add by 4 each time a branch mispredict happens and sub- tract by one each time a conditional branch retires. Enter conserva- tive mode if the MSB of the counter is set.

  • Page 173: Dcache Status Register – Dc_Stat

    The Dcache status register (DC_STAT) is a read-write register. If a Dcache tag parity error or data ECC error occurs, information about the error is latched in this register. Figure 5–34 shows the Dcache status register. 21264/EV68A Hardware Reference Manual Type Description —...

  • Page 174: Cbox Csrs And Iprs

    • The hardware allocates 367 bits for the WRITE_ONCE chain, of which the 21264/ EV68A uses 304 bits. During hardware reset (after BiST), 367 bits are always shifted into the WRITE_ONCE chain from the SROM, MSB first, so that the unused bits are shifted out the end of the WRITE_ONCE chain.

  • Page 175: Cbox Data Register - C_Data

    Many CSRs are duplicated for ease of hardware implementation. These CSRs are indicated in italics. They must be written with values that are identical to the values written to the original CSRs. 21264/EV68A Hardware Reference Manual Description — Cbox data register. A HW_MTPR instruction to this register causes six bits of data to be placed into a serial shift register.

  • Page 176: Cbox Write_Once Chain Order

    2.0X 0100 2.5X 1000 3.0X Enable duplicate tag mode in the 21264/EV68A. Enable probe-tag only mode in the 21264/EV68A. When asserted, disables fast data movement mode. Enables RdVictim mode on the pins. Duplicate CSR. Enable inhibition of incrementing acknowledge counter for RdVic commands.
  • Page 177 ENABLE_STC_COMMAND[0] BC_LATE_WRITE_NUM[0:2] BC_CPU_LATE_WRITE_NUM[0:1] BC_BURST_MODE_ENABLE[0] BC_PENTIUM_MODE[0] BC_CLK_RATIO[1] BC_FRM_CLK[0] BC_CLK_DELAY[0:1] BC_DDMR_ENABLE[0] BC_DDMF_ENABLE[0] BC_LATE_WRITE_UPPER[0] 21264/EV68A Hardware Reference Manual Description Write to read GCLK bubbles. Duplicate CSR. Duplicate CSR. Duplicate CSR. Duplicate CSR. Duplicate CSR. Duplicate CSR. Duplicate CSR. Duplicate CSR. Enable error checking during probe processing.
  • Page 178 Internal Processor Registers 5–36 Description Enables the update of the 21264/EV68A Bcache tag outputs based on the falling edge of the forwarded clock. Enables the update of the 21264/EV68A Bcache tag outputs based on the rising edge of the forwarded clock.
  • Page 179 SYS_DDMR_ENABLE SYS_DDMF_ENABLE BC_DDM_FALL_EN BC_DDM_RISE_EN BC_CLKFWD_ENABLE BC_RCV_MUX_CNT_PRESET[1:0] SYS_CLK_DELAY[0:1] SYS_DDMR_ENABLE SYS_DDMF_ENABLE SYS_DDM_FALL_EN SYS_DDM_RISE_EN SYS_CLKFWD_ENABLE SYS_RCV_MUX_CNT_PRESET[0:1] 21264/EV68A Hardware Reference Manual Description Duplicate CSR. Duplicate CSR. Duplicate CSR. Duplicate CSR. Duplicate CSR. Duplicate CSR. Duplicate CSR. Duplicate CSR. Duplicate CSR. Duplicate CSR. Duplicate CSR.

  • Page 180: Cbox Write_Many Chain Description

    CSR to program the system forwarded clock shift register load val- ues. CSR to program the system forwarded clock b-phase enables. CSR to program the ratio between frame clock and system for- warded clock. Fifth SYSDC_DELAY bit. 21264/EV68A Hardware Reference Manual...

  • Page 181: Write_Many Chain Write Transaction Example

    Figure 5–37 shows an example of PALcode used to write to the WRITE_MANY chain. Figure 5–37 WRITE_MANY Chain Write Transaction Example ; Initialize the Bcache configuration in the Cbox BC_VALID_MODE = 1 BC_ENABLE = 1 21264/EV68A Hardware Reference Manual Description Control Bcache block parity calculation Enable initialize mode Bcache size...
  • Page 182 ; shift in 6 bits ; decrement R0 ; done if R0 is zero ; align next 6 bits ; continue shifting ; wait until last shift ; predicts fall thru ; predict infinite loop ; nop ; nop 21264/EV68A Hardware Reference Manual...

  • Page 183: Cbox Read Register (Ipr) Description

    Table 5–26 Cbox Read IPR Fields Description Name Description C_SYNDROME_1[7:0] Syndrome for upper QW in OW of victim that was scrubbed. C_SYNDROME_0[7:0] Syndrome for lower QW in OW of victim that was scrubbed. 21264/EV68A Hardware Reference Manual Cbox CSRs and IPRs Internal Processor Registers 5–41...
  • Page 184 Table 5–26 Cbox Read IPR Fields Description (Continued) Name Description C_STAT[4:0] As follows: Bits 0 0 0 0 0 Either no error, or error on a speculative load, or a Bcache victim read 0 0 0 0 1 BC_PERR (Bcache tag parity error) 0 0 0 1 0 DC_PERR (duplicate tag parity error) 0 0 0 1 1 DSTREAM_MEM_ERR 0 0 1 0 0 DSTREAM_BC_ERR...

  • Page 185: Privileged Architecture Library Code

    Privileged Architecture Library Code This chapter describes the 21264/EV68A privileged architecture library code (PAL- code). The chapter is organized as follows: • PALcode description • PALmode environment • Required PALcode function codes • Opcodes reserved for PALcode • Internal processor register access mechanisms •...

  • Page 186: Palmode Environment

    There are a few extra instructions that are only available in PALmode, and will cause a dispatch to the OPCDEC PALcode entry point if attempted while not in PALmode. The Alpha architecture allows some flexibility in what these special PALmode instructions do. In the 21264/EV68A, the special PALmode-only instructions perform the following func- tions: •...

  • Page 187: Required Palcode Function Codes

    When executing in PALmode, there are certain restrictions for using the privileged instructions because PALmode gives the programmer complete access to many of the internal details of the 21264/EV68A. Refer to Section 6.4 for information on these spe- cial PALmode instructions.

  • Page 188: Hw_St Instruction

    Access checks for FOR, FOW, read and write protection. Access checks use DTB_ ALT_MODE IPR. Access length is longword. Access length is quadword. Holds a 12-bit signed byte displacement. 26 25 21 20 16 15 21264/EV68A Hardware Reference Manual DISP FM-05654.AI4 DISP FM-05654.AI4...

  • Page 189: Hw_Ret Instruction

    JSR_COROUTINE, the stack can be managed by setting the HINT bits accordingly. See Section D.25 for more information about the HW_RET instruction. Figure 6–3 shows the HW_RET instruction format. 21264/EV68A Hardware Reference Manual Opcodes Reserved for PALcode Description The opcode value.

  • Page 190: Hw_Mfpr And Hw_Mtpr Instructions

    HW_COROUTINE — The prediction is popped off the stack and used as the target. The PC is pushed onto the stack. If set, the fetcher is stalled until the HW_RET instruction is retired or aborted. The 21264/EV68A will: • Force a mispredict •...

  • Page 191: Internal Processor Register Access Mechanisms

    For example, a load instruction that generates an access violation is an implicit writer of the VA, MM_STAT, and EXC_ADDR IPRs. In the 21264/ EV68A, only instructions that generate an exception will act as implicit IPR writ- ers.

  • Page 192: Ipr Scoreboard Bits

    Software was required to schedule HW_MTPR and HW_MFPR instructions for each machine’s pipeline organization in order to ensure correct behavior. This software scheduling task is more difficult in the 21264/EV68A because the Ibox performs dynamic scheduling. Hence, eight extra scoreboard bits are used within the IQ to help maintain correct IPR access order.

  • Page 193: Hardware Structure Of Implicitly Written Iprs

    PALcode routine reordered. invoked by the exception associated with the writer, then ordering is guaran- teed. 21264/EV68A Hardware Reference Manual Internal Processor Register Access Mechanisms First Instruction Explicit Reader Explicit Writer A variety of mechanisms are be reordered. used to ensure order: scoreboard bits to stall issue of reader;...

  • Page 194: Correct Ordering Of Explicit Writers Followed By Implicit Readers

    See Appendix D for a listing of cases where this method is recommended. Privileged Architecture Library Code 6–10 First Instruction Reader reads second Scoreboard bits stall second register. Writer cannot writer in map stage until first write second register writer is retired. until it is retired. 21264/EV68A Hardware Reference Manual...

  • Page 195: Correct Ordering Of Explicit Readers Followed By Implicit Writers

    HW_MFPR and the LD. 6.6 PALshadow Registers The 21264/EV68A contains eight extra virtual integer registers, called shadow regis- ters, which are available to PALcode for use as scratch space and storage for com- monly used values. These registers are made available under the control of the SDE[1] field of the I_CTL IPR.

  • Page 196: Status Flags

    6.7.1 Status Flags The FPCR status bits in the 21264/EV68A are set with PALcode assistance. Floating- point exceptions, for which the associated FPCR status bit is clear or for which the associated trap is enabled, result in a hardware trap to the ARITH PALcode routine. The EXC_SUM register contains information to allow this routine to update the FPCR appropriately, and to decide whether to report the exception to the operating system.

  • Page 197: Palcode Exception Entry Points

    DTBM_SINGLE Fault DFAULT Fault OPCDEC Fault IACV Fault 21264/EV68A Hardware Reference Manual to 7F inclusive and 3F inclusive, and IER_CM[CM] is not equal to the kernel Offset Description Dstream TB miss on virtual page table entry fetch. Use three- level flow.

  • Page 198: Translation Buffer (Tb) Fill Flows

    Interrupt 6.9 Translation Buffer (TB) Fill Flows This section shows the expected PALcode flows for DTB miss and ITB miss. Familiar- ity with 21264/EV68A IPRs is assumed. 6.9.1 DTB Fill Figure 6–5 shows single-miss DTB instructions flow. Figure 6–5 Single-Miss DTB Instructions Flow Example...
  • Page 199 (J) page table and this processor’s virtual memory area such that if this processor sees the write to the PTE from (I) it will see the new data. Processor I Wr Data Wr PTE 21264/EV68A Hardware Reference Manual Translation Buffer (TB) Fill Flows ; branch for mb ; return ; return ;...

  • Page 200: Itb Fill

    This behavior is functionally correct because DTB loads that tag-match an existing DTB entry are ignored by the 21264/EV68A and the second DTB miss execution will load exactly the same entry as the first. 6.9.2 ITB Fill Figure 6–6 shows the ITB miss instructions flow.

  • Page 201: Performance Counter Support

    6.10 Performance Counter Support The 21264/EV68A provides hardware support for two methods of obtaining program performance feedback information. The two methods do not require program modifica- tion. Instead, performance monitoring utilities make calls to the PALcode to set up the counters and contain interrupt handlers that call PALcode to retrieve the collected data.

  • Page 202: Aggregate Mode Programming Guidelines

    Mnemonic Relevant Fields PMPC All fields IER_CM PCEN[1:0] ISUM PC[1:0] I_CTL SPCE, PCT0_EN, PCT1_EN I_STAT OVR, ICM, TRAP-TYPE, LSO, TRP, MIS PCTX PPCE PCTR_CTL All fields 21264/EV68A Hardware Reference Manual Described in Section 5.2.6 5.2.9 5.2.11 5.2.15 5.2.16 5.2.21 5.2.22...

  • Page 203: Operation

    The handler may also choose to write the counters to control the frequency of inter- rupts. Table 6–10 Aggregate Mode Returned IPR Contents PCTR_CTL 21264/EV68A Hardware Reference Manual Performance Counter Support Field Contents PCTR0[19:0]...

  • Page 204: Aggregate Counting Mode Description

    I_CTL[PCT0_EN] and either I_CTL[SPCE] or PCTX[PPCE]. On overflow, an inter- rupt is triggered as ISUM[PC0] if enabled via IER_CM[PCEN0]. The 21264/EV68A can retire up to 11 instructions per cycle, which exceeds PCTR0's maximum increment of 8 per cycle. However, no retires go uncounted because the 21264/EV68A cannot sustain 11 retires per cycle, and the 21264/EV68A corrects PCTR0 in subsequent cycles.

  • Page 205: Operation

    SL0 & SL1. 4. End window The last cycle of the window depends on whether the instruction traps, retires, aborts, and/or is squashed by the fetcher. 21264/EV68A Hardware Reference Manual Performance Counter Support New Instructions CMOV1 Ra, oldRc CMOV2 newRc1, Rb...

  • Page 206: Profileme Mode Returned Ipr Contents

    Branch direction if instruction is a conditional branch. Instruction stalled for at least one cycle between fetch and map stages of pipeline. Instruction killed during or before cycle in which it was mapped. Counter 0 value. Counter 1 value. 21264/EV68A Hardware Reference Manual...

  • Page 207: Profileme Counting Mode Description

    Counts cycles that a profiled instruction delayed the retire pointer advance during the ProfileMe window. The 21264/EV68A tracks instructions in the pipeline by allocating them "inums" near the front of the pipeline. All inums are retired in the order in which they were allocated at the end of the pipeline.

  • Page 208: Counter Modes For Profileme Mode

    Table 6–14 ProfileMe Mode PCTR_CTL Input Select Fields SL0[4] SL1[3:2] PCTR0 Privileged Architecture Library Code 6–24 PCTR1 Retired instructions Cycle counting Cycle counting Inum retire delay Retired instructions Bcache miss or long latency probes Cycle counting Mbox replay traps 21264/EV68A Hardware Reference Manual...

  • Page 209: Initialization And Configuration

    GCLK frequency during sleep mode. 7.1 Power-Up Reset Flow and the Reset_L and DCOK_H Pins The 21264/EV68A reset sequence is triggered using the two input signals Reset_L and DCOK_H in a sequence that is described in Section 7.1.1. After Reset_L is deasserted,...

  • Page 210: Power-Up Reset Flow And The Reset_L And Dcok_H Pins

    Setup (A0) and hold (A1) for IRQ’s to be latched by DCOK (2 ns for each). Enough time for Reset_L to propagate through 5 stages of RESET synchronizer (clocked by the inter- nal framing clock, which is driven by EV6Clk_x). Worst case for the 21264/EV68A would be 5x8x15 = 600 GCLK cycles.

  • Page 211: Power Sequencing And Reset State For Signal Pins

    Tristated BcDataWr_L Tristated BcLoad_L Tristated BcTag_H[42:20] Tristated BcTagDirty_H Tristated System Interface IRQ_H[5:0] NA (input) SysAddIn_L[14:0] NA (input) SysAddInClk_L NA (input) 21264/EV68A Hardware Reference Manual RAMP1 RAMP2 WAIT_ClkFwdRst0 no min Signal BcTagInClk_H BcTagOE_L BcTagOutClk_x BcTagParity_H BcTagShared_H BcTagValid_H BcTagWr_L BcVref SysDataInClk_H[7:0] SysDataInValid_L...

  • Page 212: Clock Forwarding And System Clock Ratio Configuration

    Prior to DCOK_H being asserted, the logic internal to the 21264/EV68A is being reset and the internal clock network is running (either clocked by the PLL VCO, which is at a nominal speed, or by ClkIn_H, if the PLL is bypassed).

  • Page 213: Pin Signal Names And Initialization State

    IRQ_H pins. The IRQ_H pins are sampled on the rising edge of DCOK_H, during which time the 21264/EV68A is in reset and is not generating any system activity. During normal operation, the IRQ_H pins supply interrupt requests to the 21264/EV68A.

  • Page 214: Pll Ramp Up

    PLL ramp up sequence. Ramping up of the PLL is required to guarantee that the dynamic change in frequency will not cause the supply on the 21264/ EV68A to fall due to the supply loop inductance. Clock control circuitry steps GCLK from power-up/reset clocking to 1/16 and finally normal operating frequency.

  • Page 215: Clock Forward Reset And System Interface Initialization

    As BiST completes, the TestStat_H pin is held low for 16 GCLK cycles. Then, if BiST succeeds, the pin remains low. Otherwise, it is asserted. After successfully completing BiST, the 21264/EV68A then performs the SROM load sequence (described in Chapter 11). After the SROM load sequence is finished, the 21264/EV68A deasserts SromOE_L.

  • Page 216: Fault Reset Flow

    RUN state. ClkFwdRst_H is asserted by the system, which causes the state machine to transition to the WAIT_FAULT_RESET state. The 21264/EV68A internally resets a minimum amount of internal state. Note the effects of that reset on the IPRs in Table 7–5 Table 7–5 Effect on IPRs After Fault Reset...

  • Page 217: Energy Star Certification And Sleep Mode Flow

    Clks 7.3 Energy Star Certification and Sleep Mode Flow The 21264/EV68A is Energy Star compliant. Energy Star is a program administered by the Environmental Protection Agency to reduce energy consumption. For compliance, a computer must automatically enter a low power sleep mode using 30 watts or less after a specified period of inactivity.

  • Page 218: Effect On Iprs After Transition Through Sleep Mode

    WRITE_MANY Cleared (That is, the WRITE_MANY chain is initialized and the Bcache is Note that Interrupt enables are maintained during sleep mode, enabling the 21264/ EV68A to wake up. The 21264/EV68A waits for either an unmasked clock interrupt or an unmasked device interrupt from the system.

  • Page 219: Warm Reset Flow

    The warm reset sequence of operation is triggered by the assertion of the Reset_L sig- nal line. The reset state machine is initially in RUN state. The 21264/EV68A then, by default, ramps down the PLL (similar to the sleep flow sequence) and the reset state machine ends up in the WAIT_RESET state.

  • Page 220: Array Initialization

    After BiST and SROM loading have completed, SromOE_L deasserts and the 21264/ EV68A waits for ClkFwdRst_H to deassert before starting the external clocks and, like the other flows, waits for 264 cycles before starting instructions. 7.5 Array Initialization The following arrays are initialized by BiST: •...

  • Page 221: Example For Initializing Bcache

    WrVictimBlk command generation to the system interface are squashed. Using the INVAL_TO_DIRTY_ENABLE and INIT_MODE registers, initialization code loaded from the SROM can generate and delete blocks inside the 21264/EV68A without system interaction. This behavior is very useful for initialization and startup processing, when the system interfaces are not fully functional.

  • Page 222: External Interface Initialization

    ;bc_wrt_sts_a ;bc_bank_enable_a ;This loop generates legal ECC data, and ;invalidate tags which are written to the ;Bcache for all but the final 64KB of address. ;and cleans up the Dcache also. Reset State Comments — — 21264/EV68A Hardware Reference Manual...
  • Page 223 DTB_ALTMODE DTB alternate processor mode DTB_IAP DTB invalidate all process ASM = 0 DTB_IA DTB invalidate all process 21264/EV68A Hardware Reference Manual Internal Processor Register Power-Up Reset State Reset State Comments — Must be written to in PALcode. — —...

  • Page 224: Ieee 1149.1 Test Port Reset

    Cbox shift control 7.9 IEEE 1149.1 Test Port Reset Signal Trst_L must be asserted when powering up the 21264/EV68A. Trst_L must not be deasserted prior to assertion of DCOK_H. Trst_L can remain asserted during nor- mal operation of the 21264/EV68A.

  • Page 225: Ev68A Reset State Machine State Diagram

    Figure 7–5 21264/EV68A Reset State Machine State Diagram PLL Ramp Up Reset_L deasserted WAIT_ DCOK_H NOMINAL asserted [16,32] WAIT_ SETTLE [16,32] Reset_L asserted Enabled COLD Interrupt Reset_L asserted WAIT_ WAIT_ RESET INTERRUPT Counter Counter finished & finished & Sleep Mode...
  • Page 226 PAL_BASE + 0x780. DOWN1 21264/EV68A was in a state in which GCLK was at its highest speed and Reset_L was asserted. Internal chip functions are reset and the internal duration counter is set to 8205 cycles.

  • Page 227: Phase-Lock Loop (Pll) Functional Description

    7.11 Phase-Lock Loop (PLL) Functional Description The PLL multiplies the clock frequency of a differential input reference clock and aligns the phase of its output to that differential input clock. Thus, the 21264/EV68A can communicate synchronously on clock boundaries with clock periods that are defined by the system.

  • Page 228: Power-Up/Reset Clocking

    — — Divider 16 is out of range for the 21264/EV68A and reserved for future use. Valid reference clock (ClkIn_x) frequencies for the 21264/EV68A are specified in the range from 80 to 200. Divider values that are out of that range are displayed as a dash “—”.

  • Page 229: Error Detection And Error Handling

    Error Detection and Error Handling This chapter gives an overview of the 21264/EV68A error detection and error handling mechanisms, and is organized as follows: • Data error correction code • Icache data or tag parity error • Dcache tag parity error •...

  • Page 230: Data Error Correction Code

    Data Error Correction Code 8.1 Data Error Correction Code The 21264/EV68A supports a quadword error correction code (ECC) for the system data bus. ECC is generated by the 21264/EV68A for all memory write transactions (WrVictimBlk) emitted from the 21264/EV68A and for all probe data. ECC is also checked on every memory read transaction for single-bit correction and double-bit error detection.

  • Page 231: Dcache Data Single-Bit Correctable Ecc Error

    Cbox scrubs the block in the Dcache by evicting the block into the victim buffer (thereby scrubbing it) and writing it back into the Dcache as follows: – C_STAT[DSTREAM_DC_ERR] is set. 21264/EV68A Hardware Reference Manual Dcache Data Single-Bit Correctable ECC Error Error Detection and Error Handling 8–3...

  • Page 232: Store Instruction (Quadword Or Smaller)

    The Dcache duplicate tag has the correct version of the Dcache coherence state for the 21264/EV68A, allowing it to be used for correct tag/status data when the Dcache tags generate a parity error. These tags are parity protected also; however, the Dcache dupli- cate tag cell is designed to be much more tolerant of soft errors.

  • Page 233: Bcache Tag Parity Error

    If an error is detected, the following actions are taken: • The hardware flushes the Icache. • C_STAT[ISTREAM_BC_ERR] is set. 21264/EV68A Hardware Reference Manual Bcache Tag Parity Error Error Detection and Error Handling 8–5...

  • Page 234: Dcache Fill From Bcache

    The Ibox will invoke a replay trap at an instruction that is older than (or equal to) any instruction that consumes the load instruction’s data. The 21264/ EV68A then stalls the replayed Istream in the map stage of the pipeline, until the error is corrected.

  • Page 235: Bcache Victim Read

    C_ADDR contains bits [42:6] of the system memory fill address of the block that contains the error. • C_SYNDROME_0[7:0] and C_SYNDROME_1[7:0] contain the syndrome of quadword 0 and 1, respectively, of the octaword subblock that contains the error. 21264/EV68A Hardware Reference Manual Error Detection and Error Handling 8–7...

  • Page 236: Dcache Fill From Memory

    The PALcode cannot scrub the data because C_STAT is zero, and C_ADDR does not have the address of the block with the error. Error Detection and Error Handling 8–8 21264/EV68A Hardware Reference Manual...

  • Page 237: Bcache Data Single-Bit Correctable Ecc Error On A Probe

    The PALcode on the probed processor may choose to scrub the error, though it will probably be scrubbed by the requesting processor. 8.12 Double-Bit Fill Errors Double-bit errors for fills are detected, but not corrected, in the 21264/EV68A. The fol- lowing events may cause a double-bit fill error: •...

  • Page 238: Error Case Summary

    Log as MCHK Uncorrectable Log as MCHK Icache flushed Scrub error as described Section D.36 Log as CRD Corrected and Scrub error as described Section D.36 scrubbed in Dcache Log as CRD None Log as CRD None None 21264/EV68A Hardware Reference Manual...
  • Page 239 For a single-bit error on a non-target quadword, the error is not corrected in hardware, but is corrected by PALcode during the scrub operation. The contents of C_ADDR may not be accurate when there is heavy cache fill traffic. 21264/EV68A Hardware Reference Manual Error Case Summary Hardware...

  • Page 241: Electrical Data

    This chapter describes the electrical characteristics of the 21264/EV68A and its inter- face pins. The chapter contains both ac and dc electrical characteristics and power sup- ply considerations, and is organized as follows: • Electrical characteristics • DC characteristics •...

  • Page 242: Dc Characteristics

    The test load must be mal operation, these inputs are coupled with a 680-pF capacitor. 3. Functional operation of the 21264/EV68A with less than all VDD and VSS pins connected is not implied.

  • Page 243: Input Differential Amplifier Clock Receiver (I_Da_Clk)

    Current out of a 21264/EV68A pin is represented by a – symbol while a + Note: symbol indicates current flowing into a 21264/EV68A pin. Table 9–3 VDD (I_DC_POWER) Parameter Symbol Description Processor core supply voltage Power (sleep) Processor power required (sleep)

  • Page 244: Bidirectional, Differential Amplifier Receiver, Open-Drain Output Driver (B_Da_Od)

    VREF + 200 mV — = 6 mA — = –6 mA VDD – 400 mV VDD — Freq =10 MHz — 21264/EV68A Hardware Reference Manual Maximum 400 mV 150 A 5.7 pF Note 6 Maximum VREF –200 mv 400 mV 150 A 5.7 pF...

  • Page 245: Power Supply Sequencing And Avoiding Potential Failure Mechanisms

    Before the power-on sequencing can occur, systems should ensure that DCOK_H is deasserted and Reset_L is asserted. Then, systems ramp power to the 21264/EV68A PLL_VDD @ 2.5 V and the 21264/EV68A power planes (VDD @ 1.65 V, not to exceed 1.70 V under any circumstances), with PLL_VDD leading VDD. Systems should supply differential clocks to the 21264/EV68A on ClkIn_H and ClkIn_L.

  • Page 246: Ac Characteristics

    To avoid failure mechanism number two, systems must sequence and control external signal flow in such a way as to avoid zero differential into the 21264/EV68A input receivers (I_DA, I_DA_CLK, B_DA_OD, B_DA_PP, and B_DA_PP). Finally, to avoid failure mechanism number three, systems must sequence input and bidirectional pins (I_DA, I_DA_CLK, B_DA_OD, B_DA_PP, and I_DC_REF) such that the 21264/ EV68A does not see a voltage above its VDD.

  • Page 247: Ac Specifications

    SysDataInClk_H[7:0] I_DA SysDataOutClk_L[7:0] O_OD BcAdd_H[23:4] O_PP BcDataOE_L O_PP BcLoad_L O_PP BcDataWr_L O_PP BcData_H[127:0] B_DA_PP 21264/EV68A Hardware Reference Manual Reference Signal SysAddInClk_L 400 ps 400 ps SysAddInClk_L 400 ps 400 ps SysAddInClk_L 400 ps 400 ps SysAddInClk_L 400 ps 400 ps...
  • Page 248 2.0 ns 2.0 ns Tck_H Tck_H 2.0 ns 2.0 ns Tck_H IEEE 1149.1 Port Freq. = 5.0 MHz Max. EV6Clk_x 21264/EV68A Hardware Reference Manual TSkew Duty Cycle TSlew 45-55% ± 350 ps ± 350 ps 1.0 V/ns 1.0 V/ns 38-63% 1.0 V/ns...
  • Page 249 The TSU and TDH of FrameClk_x are referenced to the deasserting edge of ClkIn_x. This signal is a feedback to the internal PLL and may be monitored for overall 21264/EV68A jitter. It can also be used as a feedback signal to an external PLL when in PLL bypass mode. Proper termina- tion of EV6Clk_x is imperative.

  • Page 251: Operating Temperature At Heat Sink Center (Tc)

    • Thermal design considerations 10.1 Operating Temperature The 21264/EV68A is specified to operate when the temperature at the center of the heat sink (T is as shown in Table 10–1. Temperature T the heat sink, between the two package studs. The GRAFOIL pad is the interface mate- rial between the package and the heat sink.
  • Page 252 3 ( C/W) Heat sink type 3 has a 80 mm Table 10–3 Maximum T for 21264/EV68A @ 750 MHz and @ 1.7 V with Various Airflows Airflow (linear ft/min) Maximum T with heat sink type 1 ( C)

  • Page 253: Heat Sink Specifications

    Figure 10–1 shows the heat sink type 1, along with its approximate dimensions. Figure 10–1 Type 1 Heat Sink 1.55 in 2.05 in 0.10 in 0.30 in 21264/EV68A Hardware Reference Manual 4.10 in 0.05 in Heat Sink Specifications 1.00 in 4.10 in 1.50 in...

  • Page 254: Type 2 Heat Sink

    Figure 10–2 shows the heat sink type 2, along with its approximate dimensions. Figure 10–2 Type 2 Heat Sink 1.375 in 1.875 in 1.015 in 0.055 in 0.25 in Thermal Management 10–4 3.75 in 1.875 in 0.075 in 0.04 in 21264/EV68A Hardware Reference Manual 1.00 in 3.75 in 2.35 in LK99-0211a...

  • Page 255: Type 3 Heat Sink

    1.600 in 1.600 in 1.100 in 0.060 in 0.080 in 0.040 in 2.00 in 0.30 in 3.200 in 21264/EV68A Hardware Reference Manual Heat Sink Specifications 80 mm 15 mm 3.20 in 1.00 in 0.200 in 0.575 in 0.20 in 0.100 in 0.05 in...

  • Page 256: Thermal Design Considerations

    10.3 Thermal Design Considerations Follow these guidelines for printed circuit board (PCB) component placement: • Orient the 21264/EV68A on the PCB with the heat sink fins aligned with the air- flow direction. • Avoid preheating ambient air. Place the 21264/EV68A on the PCB so that inlet air is not preheated by any other PCB components.

  • Page 257: Test Pins

    11.1 Test Pins The 21264/EV68A test access ports include the IEEE 1149.1 test access port, a dual- purpose SROM/Serial diagnostic terminal port, and a test status output pin. Table 11–1 lists the test access port pins.

  • Page 258: Srom/Serial Diagnostic Terminal Port

    The SromOE_L pin supplies the output enable as well as the reset to the serial ROM. (Refer to the serial ROM specifications for details.) The 21264/EV68A asserts this signal low for the duration of the Icache load from the serial ROM.

  • Page 259: Ieee 1149.1 Port

    Tdi_H, Tms_H, and Trst_L pins, as required by the present standard. The scope of 1149.1 compliant features on the 21264/EV68A is limited to the board level assembly verification test. The systems that do not intend to drive this port must terminate the port pins as follows: pull-ups on Tdi_H and Tms_H, pull-downs on Tck_H and Trst_L.

  • Page 260: Teststat_H Pin

    A system designer may sample the TestStat_H pin on the first rising edge of the SromClk_H pin to determine BiST results. After the power-up dur- ing the normal chip operation, whenever the 21264/EV68A does not retire an instruction for 2K CPU cycles, the pin is asserted high for 3 CPU cycles.

  • Page 261: Power-Up Self-Test And Initialization

    11.5.2 SROM Initialization Power-up initialization on the 21264/EV68A is different from previous generation Alpha systems in two aspects. First, in the 21264/EV68A systems, the presence of serial ROMs is mandatory as initialization of several Cbox configuration registers depends on them. Second, it is possible to skip or partially fill Icache from serial ROMs.

  • Page 262: Serial Instruction Cache Load Operation

    Data(0,n) plus MBZ(m,0) must equal 367 bits. (If Cbox Config Data(0,n) is (0,366), MBZ would be zero.) For the 21264/EV68A, Cbox Config Data is 304 bits; the value for n is 303. Therefore, the value MBZ field for Pass 3 is: MBZ(m,0) = 367 minus 304 = 63 = (62,0) Tables 11–4 and 5–24 describe the details of the Icache and Cbox bit fields, respec-...

  • Page 263: Notes On Ieee 1149.1 Operation And Compliance

    Icache load. 11.6 Notes on IEEE 1149.1 Operation and Compliance 1. IEEE 1149.1 port pins on the 21264/EV68A are not pulled up or pulled down on the chip. The necessary pull-up or pull-down function must be implemented on the board.

  • Page 265: A.1 Alpha Instruction Summary

    This appendix provides a summary of the Alpha instruction set and describes the 21264/EV68A IEEE floating-point conformance. It is organized as follows: • Alpha instruction summary • Reserved opcodes • IEEE floating-point instructions • VAX floating-point instructions • Independent floating-point instructions •...

  • Page 266: A–2 Architecture Instructions

    Branch if zero Branch if > zero 11.08 Bit clear 11.20 Logical sum Branch if low bit clear Branch if low bit set Branch if zero Branch if < zero Branch if zero Unconditional branch 21264/EV68A Hardware Reference Manual...
  • Page 267 CMPTLT CMPTUN CMPULE CMPULT CPYS CPYSE CPYSN CTLZ CTPOP CTTZ CVTDG CVTGD CVTGF 21264/EV68A Hardware Reference Manual Alpha Instruction Summary Opcode Description Branch to subroutine Trap to PALcode 11.24 CMOVE if zero 11.46 CMOVE if zero 11.66 CMOVE if > zero 11.16...
  • Page 268 Extract word low Floating branch if zero Floating branch if zero Floating branch if > zero Floating branch if zero Floating branch if < zero Floating branch if zero 17.02A FCMOVE if zero 17.02D FCMOVE if zero 21264/EV68A Hardware Reference Manual...
  • Page 269 INSQH INSQL INSWH INSWL ITOFF ITOFS ITOFT JSR_COROUTINE Mbr LDAH LDBU LDL_L LDQ_L LDQ_U 21264/EV68A Hardware Reference Manual Alpha Instruction Summary Opcode Description 17.02F FCMOVE if > zero 17.02E FCMOVE if zero 17.02C FCMOVE if < zero 17.02B FCMOVE if zero 18.8000 Prefetch data...
  • Page 270 13.40 Multiply longword with integer overflow enable 13.20 Multiply quadword 13.60 Multiply quadword with integer overflow enable 16.082 Multiply S_floating 16.0A2 Multiply T_floating 11.28 Logical sum with complement 1C.31 Pixel error 1C.37 Pack longwords to bytes 21264/EV68A Hardware Reference Manual...
  • Page 271 S8SUBQ SEXTB SEXTW SQRTF SQRTG SQRTS SQRTT STL_C STQ_C STQ_U 21264/EV68A Hardware Reference Manual Alpha Instruction Summary Opcode Description 1C.36 Pack words to bytes 18.E000 Read and clear 1A.2 Return from subroutine 18.C000 Read process cycle counter 18.F000 Read and set 10.02...

  • Page 272: A.2 Reserved Opcodes

    ZAPNOT A.2 Reserved Opcodes This section describes the opcodes that are reserved in the Alpha architecture. They can be reserved for Compaq or for PALcode. A.2.1 Opcodes Reserved for Compaq Table A–3 lists opcodes reserved for Compaq. Table A–3 Opcodes Reserved for Compaq...

  • Page 273: Opcodes Reserved For Palcode

    A.2.2 Opcodes Reserved for PALcode Table A–4 lists the 21264/EV68A-specific instructions. See Chapter 2 for more information. Table A–4 Opcodes Reserved for PALcode 21264/EV68A Architecture Mnemonic Opcode Mnemonic HW_LD PAL1B HW_ST PAL1F HW_REI PAL1E HW_MFPR PAL19 HW_MTPR PAL1D A.3 IEEE Floating-Point Instructions Table A–5 lists the hexadecimal value of the 11-bit function code field for the IEEE...
  • Page 274 CVTQT CVTTS DIVS DIVT MULS MULT SQRTS SQRTT SUBS SUBT Mnemonic None CVTST Mnemonic None CVTTQ Mnemonic CVTTQ Alpha Instruction Set A–10 /SUC /SUM /SUD /SUI /SVD /SVID 21264/EV68A Hardware Reference Manual /SUIC /SUIM /SUID /SVC /SVI /SVIC /SVM /SVIM...

  • Page 275: Vax Floating-Point Instructions

    Table A–7 lists the hexadecimal value of the 11-bit function code field for the floating- point instructions that are not directly tied to IEEE or VAX floating point. The opcode for the following instructions is 17 21264/EV68A Hardware Reference Manual VAX Floating-Point Instructions /SUC...

  • Page 276: A.6 Opcode Summary

    INTS*, column contains the symbol JSR*, representing all jump MISC* (mem) (mem) (mem) \ PAL \ (mem) (mem) JSR* LDL_L (mem) (mem) (mem) 21264/EV68A Hardware Reference Manual BLBC (br) (br) FBEQ (br) (br) FBLT (br) (br)

  • Page 277: Required Palcode Function Codes

    6-bit opcode and ffff is the hexadecimal 26-bit function code. Table A–10 Required PALcode Function Codes Mnemonic DRAINA HALT 21264/EV68A Hardware Reference Manual Required PALcode Function Codes \ PAL \ LDQ_L (mem) (mem) FPTI*...

  • Page 278: Ieee Floating-Point Conformance

    Alpha architecture. In addition, the 21264/ EV68A delivers a canonical Quiet NaN when an operation is required to produce a NaN value and none of its inputs are NaNs. Encodings for Signaling NaN and Quiet NaN are defined by the Alpha Architecture Reference Manual, Fourth Edi- tion.

  • Page 279: Exceptional Input And Output Conditions

    The 21264/EV68A does not produce a denormal result for the underflow exception. Instead, a true zero (+0) is written to the destination register. In the 21264/EV68A, the FPCR underflow to zero (UNDZ) bit must be set if the underflow disable (UNFD) bit is set.
  • Page 280 IEEE Floating-Point Conformance Table A–11 Exceptional Input and Output Conditions (Continued) Alpha Instructions Exponent overflow Exponent underflow Inexact result Alpha Instruction Set A–16 21264/EV68A Hardware Supplied Result Exception ±Inf or ±MAX Overflow Underflow Result Inexact 21264/EV68A Hardware Reference Manual...
  • Page 281 Inf operand QNaN operand SNaN operand CMPTLT CMPTLE INPUT Inf operand QNaN operand SNaN operand CVTfi INPUT Inf operand QNaN operand 21264/EV68A Hardware Reference Manual IEEE Floating-Point Conformance 21264/EV68A Hardware Supplied Result Exception ±Inf (none) QNaN (none) QNaN Invalid Op...
  • Page 282 CVTff OUTPUT (same as ADDx) FBEQ FBNE FBLT FBLE FBGT FBGE LDS LDT STS STT CPYS CPYSN FCMOVx See Section 2.14 for information about the floating-point control register (FPCR). 21264/EV68A Hardware Supplied Result Exception Invalid Op Result Inexact Truncated result...

  • Page 283: B.1.1 Bsdl Description Of The Alpha 21264/Ev68A Boundary-Scan Register

    B.1 Boundary-Scan Register The Boundary-Scan Register (BSR) on the 21264/EV68A is 367 bits long. It is accessed by the three public (SAMPLE, EXTEST, CLAMP) instructions. The register operation for the public instructions is compliant with the IEEE 1149.1 standard.

  • Page 284: Boundary-Scan Register

    F32 , F34 , H34 , G35 , F40 , G39 , K38 , J41 , "& M40 , N39 , P40 , T38 , V40 , W41 , W39 , Y40 , "& 21264/EV68A Hardware Reference Manual " &...
  • Page 285 "ClkIn_H "ClkIn_L "FrameClk_H "FrameClk_L "PllBypass_H 21264/EV68A Hardware Reference Manual Boundary-Scan Register AB38, AC39, AD38, AF40, AH38, AJ39, AL41, AK38, "& AN39, AP38, AR39, AT38, AY38, AV36, AW35, AV34),"& : (L7 , AA5 , AK8 , BA13, L39 , AA41, AM40, AY34),"&...
  • Page 286 , 147 , 58 , 169 , 148 , 16 , 104 , 82 , 170 , 61 , 105 ), 21264/EV68A Hardware Reference Manual , AN3 , C7 , C19 , "& , "& , AE7 , BC13, AW21, "&...
  • Page 287 "Spare_4 "Spare_5 "PLL_VDD "Spare_0 "MiscVref "Spare_2 "DCOK_H 21264/EV68A Hardware Reference Manual Boundary-Scan Register 124 , : (26 , 71 , 46 , 45 179 , 186 , 195 , 202 , 234 , 242 , 250 , 267 , "&...
  • Page 288 OUTPUT2, OUTPUT2, OUTPUT2, 21264/EV68A Hardware Reference Manual , 146 , 60 , 278 , "& , 53 , "& , 535 , 18 , 541 , 264 , "& , 144 , 103 , 152 , "&...
  • Page 289 " 304 ( BC_3, BcDataInClk_H(2), " 303 ( BC_2, SysDataOutClk_L(2), OUTPUT2, " 302 ( BC_3, *, " 301 ( BC_3, SysDataInClk_H(2), " 300 ( BC_2, BcData_H(19), " 299 ( BC_2, BcData_H(83), 21264/EV68A Hardware Reference Manual Boundary-Scan Register INPUT, INPUT, INPUT, INPUT,...
  • Page 290 236, WEAK1 ), "& -- CONTROL, INPUT, CONTROL, INPUT, BIDIR, x, 239, 21264/EV68A Hardware Reference Manual ), "& -- ), "& -- ), "& -- ), "& -- ), "& -- ), "& -- ), "& -- ), "& -- ), "&...
  • Page 291 " 174 ( BC_2, BcAdd_H(16), " 173 ( BC_2, BcAdd_H(17), " 172 ( BC_2, BcAdd_H(18), " 171 ( BC_2, BcAdd_H(19), " 170 ( BC_2, BcAdd_H(20), " 169 ( BC_2, BcAdd_H(21), 21264/EV68A Hardware Reference Manual Boundary-Scan Register BIDIR, x, 239, BIDIR, x, 236, WEAK1 ), "&...
  • Page 292 116, BIDIR, x, 119, WEAK1 ), "& -- BIDIR, x, 116, BIDIR, x, 116, 21264/EV68A Hardware Reference Manual ), "& -- ), "& -- ), "& -- ), "& -- ), "& -- ), "& -- ), "& -- ), "&...
  • Page 293 ( BC_2, SysData_L(62), " 42 ( BC_2, BcData_H(126), " 41 ( BC_2, BcData_H(62), " 40 ( BC_2, SysData_L(63), " 39 ( BC_2, BcData_H(127), 21264/EV68A Hardware Reference Manual Boundary-Scan Register BIDIR, x, 119, WEAK1 ), "& -- BIDIR, x, 116, BIDIR, x, 116,...
  • Page 294 INPUT, INPUT, INPUT, INPUT, INPUT, INPUT, INPUT, INPUT, INPUT, 21264/EV68A Hardware Reference Manual ), "& -- ), "& -- ), "& -- ), "& -- ), "& -- ), "& -- ), "& -- ), "& -- ), "& -- ), "&...

  • Page 295: Serial Icache Load Predecode Values

    Serial Icache Load Predecode Values See the Alpha Motherboards Software Developer’s Kit (SDK) for information. 21264/EV68A Hardware Reference Manual Serial Icache Load Predecode Values C–1...

  • Page 297: Palcode Restrictions And Guidelines

    21264/EV68A Hardware Reference Manual flushed /* initialize Int. Reg. 0*/ /* initialize Int. Reg. 1*/ /* initialize F.P. Reg. 0*/ /* initialize F.P. Reg. 1*/ /* initialize Int. Reg. 2*/ /* initialize Int. Reg. 3*/ PALcode Restrictions and Guidelines D–1...
  • Page 298 /* initialize Int. Reg. 23*/ /* initialize F.P. Reg. 22*/ /* initialize F.P. Reg. 23*/ /* initialize Int. Reg. 24*/ /* initialize Int. Reg. 25*/ /* initialize F.P. Reg. 24*/ /* initialize F.P. Reg. 25*/ /* initialize Int. Reg. 26*/ 21264/EV68A Hardware Reference Manual...
  • Page 299 DTB_ALT_MODE (SCRBRD=6)*/ ** MAP_SHADOW_REGISTERS ** The shadow registers are mapped. This code may be done by the SROM 21264/EV68A Hardware Reference Manual /* initialize Int. Reg. 27*/ /* initialize F.P. Reg. 26*/ /* initialize F.P. Reg. 27*/ /* initialize Int.
  • Page 300 /* continue executing in next block*/ /* fetch in next block*/ /* initialize Shadow Reg. 2*/ /* initialize Shadow Reg. 3*/ /* continue executing in next block*/ /* fetch in next block*/ /* initialize Shadow Reg. 4*/ 21264/EV68A Hardware Reference Manual...
  • Page 301 .align 6 bccshf:mtpr r1,EV6__DATA subq r0,1,r0 r0,bccend r1,6,r1 21264/EV68A Hardware Reference Manual /* initialize Shadow Reg. 5*/ /* continue executing in next block*/ /* fetch in next block*/ /* initialize Shadow Reg. 6*/ /* initialize Shadow Reg. 7*/ /* continue executing in next block*/ /* go back to 1st block and start executing*/ /* data<35:32>...
  • Page 302 /* under the above mtpr to SCRBRD=4*/ /* value = 0x000000000000001F*/ /* clear bits in DC_STAT (SCRBRD=6)*/ /* nop*/ /* and 1st clear PCTR_CTL (SCRBRD=4)*/ /* set up value for demon write*/ /* set up value for demon write*/ /* nop*/ 21264/EV68A Hardware Reference Manual...
  • Page 303 * already been initialized. This technique * can sometimes be used to debug snippets of * this code. 21264/EV68A Hardware Reference Manual /* and 2nd clear PCTR_CTL (SCRBRD=4)*/ /* set up value for demon write*/ /* set up value for demon write*/...

  • Page 304: Restriction 2 : No Multiple Writers To Iprs In Same Scoreboard Group

    LD, ST, or any memory operation, including all types of JMP instructions and HW_RET instructions that do not have the STALL bit set. PALcode Restrictions and Guidelines D–8 /* r0 <- current location */ /* r1 <- cc_ctl enable bit */ /* Enable/clear the cycle counter. */ 21264/EV68A Hardware Reference Manual...

  • Page 305: Guideline

    D.5 Restriction 7 : Replay Trap, Interrupt Code Sequence, and STF/ ITOF On an Mbox replay trap, the 21264/EV68A Ibox guarantees that the refetched load or store instruction that caused the trap is issued before any newer load or store instruc- tions.

  • Page 306: Restriction 9 : Palmode Istream Address Ranges

    Istream ACV limits, and the IVA_FORM format selection. The VA_CTL[VA_48] bit determines the VA_FORM format selection and the Dstream ACV limits. IPR mode bits I_CTL[VA_FORM_32] and VA_CTL[VA_FORM_32] should be consistent when executing in native mode. PALcode Restrictions and Guidelines D–10 21264/EV68A Hardware Reference Manual...

  • Page 307: D.11 Restriction 14 : Hw_Ret

    (a non-PALmode trap) until the register has been unlocked. After being unlocked, a subsequent new path exception condition will be allowed to reload the register and trap to PALcode. The 21264/EV68A may complete execution of the first PALcode flow, encountering the second exception condition before the delimit- ing instruction is retired, hence the need for the locking mechanism to ensure visibility of the initial register value.

  • Page 308: Guideline 16 : Jsr-Bad Va

    The following example contains a code sequence that creates the dependency chain. :Assume Ra holds value to write to ASN0/ASN1 HW_MFPR R0, VA, SCBD<7,6,5,4> XOR R0, R0, R0 BIS R0, R9, R9 PALcode Restrictions and Guidelines D–12 21264/EV68A Hardware Reference Manual...

  • Page 309: Restriction 22: Hw_Ret/Stall After Hw_Mtpr Is0/Is1

    A HW_ST/P/CONDITIONAL will not clear the lock flag such that a successive store- conditional (either STx_C or HW_ST/C) might succeed even in the absence of a load- locked instruction. In the 21264/EV68A, a store-conditional is forced to fail if there is an intervening memory operation between the store-conditional and its address-match- ing LDxL.

  • Page 310: Restriction 25: Hw_Mtpr Itb_Ia After Reset

    (up to four instructions) of all PALcode flows except CALL_PAL flows. Conditional branches should be avoided in this window. PALcode Restrictions and Guidelines D–14 21264/EV68A Hardware Reference Manual...

  • Page 311: D.26 Restriction 30 : Hw_Mtpr And Hw_Mfpr To The Cbox Csr

    Istream fill requests and stalls instruction fetch until after the desired MTPR/MFPR action is completed. This code disables Istream prefetching by way of a HW_MTPR to I_CTL[SBE], IC_FLUSH, and HW_RET_STALL sequence. 21264/EV68A Hardware Reference Manual PALcode Restrictions and Guidelines D–15...

  • Page 312: Restriction 30 : Hw_Mtpr And Hw_Mfpr To The Cbox Csr

    ; (6,0L) read cbox data ; nop ; go to block 5 ; touch block 5 ; block 5 ; clean to <5:0> ; accumulate shift data ; go to block 6 ; touch block 6 21264/EV68A Hardware Reference Manual...

  • Page 313: D.27 Restriction 31 : I_Ctl[Va_48] Update

    In normal operation, with counters enabled, a counter overflow will produce an overflow pulse, clear the counter, and produce a performance counter interrupt. Interrupts can only be blocked for one cycle. 21264/EV68A Hardware Reference Manual Restriction 31 : I_CTL[VA_48] Update ; block 6 ;...

  • Page 314: Restriction 33 : Hw_Ld Physical/Lock Use

    The new update value must not be within one cycle of overflow (within 16 for SL0, within 4 for SL1) as required by Section D.28. PALcode Restrictions and Guidelines D–18 21264/EV68A Hardware Reference Manual...

  • Page 315: Guideline 39: Writing Multiple Dtb Entries In The Same Pal Flow

    ; Current state: hw_ldq/p r4, MCHK_CRD__C_ADDR(r5) r31, r31, r31 r31, r31, r31 r31, r31, r31 21264/EV68A Hardware Reference Manual ; (2&6) write tag0 ; (1&5) write tag 1 ; (0&4) write pte0 ; (3&7) write pte1 ; force new fetch block ;...
  • Page 316 ; these 2 lines... ; ...stop pre-fetching ; clean the cbox error chain ; (7,1L) flush dtb ; wait for retire ; get a 1 ; (4,0L) clear crd ; W1C bits ; (6,0L) ; stall till they retire 21264/EV68A Hardware Reference Manual...

  • Page 317: Restriction 41: Mtpr Itb_Tag, Mtpr Itb_Pte Must Be In The Same Fetch Block

    There cannot be any mispredictable/trappable instructions together with an HW_MTPR in the current fetch block. D.40 Restriction 44: Not Applicable to the 21264/EV68A D.41 Restriction 45: No HW_JMP or JMP Intructions in PALcode Do not include HW_JMP or JMP instructions in PALcode; use HW_RET instead.

  • Page 318: Restriction 46: Avoiding Livelocks In Speculative Load Crd Handlers

    Because CBOX_ERR[C_ADDR] is not guaranteed, the CRD_HANDLER might not evict the sberr. PALcode Restrictions and Guidelines D–22 ; make sure hw_ret goes ; Hold up loads ; Hold up loads ; Hold up loads ; Ignore restriction 43 ; Return 21264/EV68A Hardware Reference Manual...
  • Page 319 <.+4> r31, next_reread s7, sys__cbox 21264/EV68A Hardware Reference Manual ; Loop dec value ; Start offset ; Block size (64K -> size of dcache) ; 2X bcache size ; Skip next instruction if ADDR ;...

  • Page 320: Restriction 48: Mb Bracketing Of Dcache Writes To Force Bad Data Ecc And Force Bad Tag Parity

    Writes to DC_CTL[F_BAD_DECC] and DC_CTL[DCDAT_ERR_EN] must be brack- eted by MB instructions to quiesce the memory system. The Istream must also be qui- esced before and during the sequence, as described in Section D.26. PALcode Restrictions and Guidelines D–24 21264/EV68A Hardware Reference Manual...

  • Page 321: E.1 Forwarding Clock Pin Groupings

    This appendix provides the pin interface between the21264/EV68A and Bcache SSRAMs. E.1 Forwarding Clock Pin Groupings Table E–1 lists the correspondance between the clock signals for the 21264/EV68A and Bcache (late-write non-bursting and dual-data rate) SSRAMs. Table E–1 Bcache Forwarding Clock Pin Groupings...

  • Page 322: Bcache Forwarding Clock Pin Groupings

    BcTagValid_H E.2 Late-Write Non-Bursting SSRAMs Table E–2 provides the data pin connections between late-write non-bursting SSRAMs and the 21264/EV68A or the system board. Table E–3 provides the same information for the tag pins. Data Pin Usage Table E–2 Late-Write Non-Bursting SSRAMs Data Pin Usage...

  • Page 323: E.3 Dual-Data Rate Ssrams

    Unconnected E.3 Dual-Data Rate SSRAMs Table E–4 provides the data pin connections between dual-data rate SSRAMs and the 21264/EV68A or the system board. Table E–5 provides the same information for the tag pins. Data Pin Usage Table E–4 Dual-Data Rate SSRAM Data Pin Usage...

  • Page 324: Dual-Data Rate Ssram Tag Pin Usage

    Dual-Data Rate SSRAMs Table E–4 Dual-Data Rate SSRAM Data Pin Usage (Continued) 21264/EV68A Signal Name or Board Connection From board, pulled up to VDD From board, pulled up to VDD Unconnected or pulled down to VSS BcDataOE_L From board, pulled down to VSS Tag Pin Usage Unused Bcache tag pins should be pulled to ground through a 200-ohm resister.
  • Page 325 Table E–5 Dual-Data Rate SSRAM Tag Pin Usage (Continued) 21264/EV68A Signal Name or Board Connection Unconnected From board, pulled down to VSS From board, pulled up to VDD 21264/EV68A Hardware Reference Manual Dual-Data Rate SSRAMs Dual-Data Rate SSRAM Tag Pin Name...

  • Page 327: Address Translation

    This glossary provides definitions for specific terms and acronyms associated with the Alpha 21264/EV68A microprocessor and chips in general. abort The unit stops the operation it is performing, without saving status, to perform some other operation. address space number (ASN) An optionally implemented register used to reduce the need for invalidation of cached address translations for process-specific addresses when a context switch occurs.
  • Page 328 All data and commands are associated with a clock and the receiver’s latch on both the rise and fall of the clock. Bit times are a multiple of the 21264/EV68A clocks. Systems must produce a bit time identical to 21264/EV68A’s bit time. The bit time is one-half the period of the forwarding clock.
  • Page 329 Also called cache consis- tency. cache fill An operation that loads an entire cache block by using multiple read cycles from main memory. cache flush An operation that marks all cache blocks as invalid. 21264/EV68A Hardware Reference Manual...

  • Page 330: Cache Memory

    The 21264/EV68A microprocessor contains two onchip internal caches. See also write- through cache and write-back cache.

  • Page 331: Conditional Branch Instructions

    Dual-data rate. A dual-data rate SSRAM can provide data on both the rising and falling edges of the clock signal. denormal An IEEE floating-point bit pattern that represents a number whose magnitude lies between zero and the smallest finite number. Dual inline package. 21264/EV68A Hardware Reference Manual...
  • Page 332 The error may be correctable (soft error) or uncorrectable (hard error). Emitter-coupled logic. EEPROM Electrically erasable programmable read-only memory. A memory device that can be byte-erased, written to, and read from. Contrast with FEPROM. Glossary –6 21264/EV68A Hardware Reference Manual...
  • Page 333 The framing clock defines the start of a transmission either from the system to the 21264/EV68A or from the 21264/EV68A to the system. The framing clock is a power- of-2 multiple of the 21264/EV68A GCLK frequency, and is usually the system clock.
  • Page 334 Instruction cache. A cache reserved for storage of instructions. One of the three areas of primary cache (located on the 21264/EV68A) used to store instructions. The Icache contains 8KB of memory space. It is a direct-mapped cache. Icache blocks, or lines, contain 32 bytes of instruction stream data with associated tag as well as a 6-bit ASM field and an 8-bit branch history field per block.

  • Page 335: Interface Reset

    (LW) Four contiguous bytes starting on an arbitrary byte boundary. The bits are numbered from right to left, 0 through 31. Load queue. Least significant bit. 21264/EV68A Hardware Reference Manual...
  • Page 336 A board on which logic devices (such as transistors, resistors, and memory chips) are mounted and connected to perform a specific system function. module-level cache See second-level cache. Metal-oxide semiconductor. MOSFET Metal-oxide semiconductor field-effect transistor. Glossary –10 21264/EV68A Hardware Reference Manual...
  • Page 337 Sixteen contiguous bytes starting on an arbitrary byte boundary. The bits are numbered from right to left, 0 through 127. OpenVMS Alpha operating system The version of the open VMS operating system for Alpha platforms. operand The data or register upon which an operation is performed. 21264/EV68A Hardware Reference Manual...
  • Page 338 A CPU design technique whereby multiple instructions are simultaneously overlapped in execution. Programmable logic array. PLCC Plastic leadless chip carrier or plastic-leaded chip carrier. Programmable logic device. Phase-locked loop. PMOS P-type metal-oxide semiconductor. Probe queue. Glossary –12 21264/EV68A Hardware Reference Manual...
  • Page 339 Random-access memory. Row address select. Read-after-write. READ_BLOCK A transaction where the 21264/EV68A requests that an external logic unit fetch read data. read data wrapping System feature that reduces apparent memory latency by allowing read data cycles to differ the usual low-to-high sequence. Requires cooperation between the 21264/EV68A and external hardware.
  • Page 340 Read-only memory. Register-transfer logic. Serial access memory. Should be one. Should be zero. scheduling The process of ordering instruction execution to obtain optimum performance. Glossary –14 21264/EV68A Hardware Reference Manual...
  • Page 341 An area of memory set aside for temporary data storage or for procedure and interrupt service linkages. A stack uses the last-in/first-out concept. As items are added to (pushed on) the stack, the stack pointer decrements. As items are retrieved from (popped off) the stack, the stack pointer increments. 21264/EV68A Hardware Reference Manual...
  • Page 342 An operation that may halt the processor or cause it to lose information. Only privileged software (that is, software running in kernel mode) can trigger an UNDEFINED opera- tion. (This meaning only applies when the word is written in all upper case.) Glossary –16 21264/EV68A Hardware Reference Manual...
  • Page 343 A cache that is addressed with virtual addresses. The tag of the cache is a virtual address. This process allows direct addressing of the cache without having to go through the translation buffer making cache hit times faster. VLSI Very-large-scale integration. Virtual program counter. VRAM Video random-access memory. 21264/EV68A Hardware Reference Manual...

  • Page 344: Write-Back Cache

    WRITE_BLOCK A transaction where the 21264/EV68A requests that an external logic unit process write data. write data wrapping System feature that reduces apparent memory latency by allowing write data cycles to differ the usual low-to-high sequence.
  • Page 345 BC_CLK_DELAY Cbox CSR 4–45 defined 5–35 BC_CLK_LD_VECTOR Cbox CSR defined 5–38 BC_CLKFWD_ENABLE Cbox CSR defined 5–36 21264/EV68A Hardware Reference Manual BC_CLOCK_OUT Cbox CSR BC_CPU_CLK_DELAY Cbox CSR defined BC_CPU_LATE_WRITE_NUM Cbox CSR defined BC_DDM_FALL_EN Cbox CSR defined BC_DDM_RISE_EN Cbox CSR defined BC_DDMF_ENABLE Cbox CSR...
  • Page 346 C_STAT Cbox read register field C_STS Cbox read register field C_SYNDROME_0 Cbox read register field C_SYNDROME_1 Cbox read register field Cache block states response to 21264/EV68A commands transitions Cache coherency CALL_PAL entry points 4–43 Caution convention 21264/EV68A Hardware Reference Manual 3–4...
  • Page 347 CMOV instruction, special cases of COLD reset machine state 7–17 Commands 21264/EV68A to system 4–19 system to 21264/EV68A 4–26 when to NXM 4–38 21264/EV68A Hardware Reference Manual Conventions abbreviations address aligned 4–13 bit notation D–15 caution data units do not care...
  • Page 348 VA_CTL virtual address format register VA_FORM virtual address register ECB instruction, external interface reference 21264/EV68A Hardware Reference Manual 7–16 5–28 5–28 7–16 5–27 7–15 7–15 5–27 7–16 5–27...
  • Page 349 2–10 executed in pipeline 2–16 FEN fault 6–13 FetchBlk, 21264/EV68A command system probes, with 4–41 21264/EV68A Hardware Reference Manual FetchBlkSpec, 21264/EV68A command 8–2 4–39 8–3 Field notation convention Floating-point arithmetic trap, pipeline abort delay with Floating-point control register 5–39 PALcode emulation of 8–2...
  • Page 350 Instruction slot, pipelined Instruction-stream translation buffer Int_Add_BcClk internal forwarded clock 4–48 Int_Data_BcClk internal forwarded clock 4–49 5–10 INT_FWD_CLK clock queue Integer arithmetic trap, pipeline abort delay with 21264/EV68A Hardware Reference Manual 7–15 5–21 8–2 8–10 8–5 8–7 5–21 7–12 7–12 11–7...
  • Page 351 7–15 ITB_MISS fault 6–14 ITB_PTE array write register 5–6 at power-on reset state 7–14 21264/EV68A Hardware Reference Manual ITB_TAG array write register at power-on reset state IVA_FORM instruction virtual address format register at power-on reset state JITTER_CMD Cbox CSR, defined...
  • Page 352 Open-drain output driver. See O_OD pin type Operating temperature 4–5 Packaging 3–18 Paired instruction fetch order PAL_BASE register after fault reset after warm reset at power-on reset state through sleep mode 21264/EV68A Hardware Reference Manual 7–16 6–12 6–14 3–3 4–38 4–21 4–21 3–3 9–2 9–4 3–3...
  • Page 353 PLL_IDD, values for 9–3 PLL_VDD signal pin 3–5 PLL_VDD, values for 9–3 PllBypass_H signal pin 3–5 PMPC ProfileMe register 5–8 21264/EV68A Hardware Reference Manual Ports IEEE 1149.1 serial terminal SROM load Power maximum sleep defined Power supply sequencing Power-on flow signals and constraints...
  • Page 354 SromClk_H signal pin SromData_H signal pin SromOE_L signal pin SSRAMs dual-data rate pin assignments late-write non-bursting pin assignments STC_ENABLE Cbox CSR STCChangeToDirty, 21264/EV68A command 4–13 Storage temperature 21264/EV68A Hardware Reference Manual xxii 11–2 4–23 5–39 4–24 4–22 4–41 3–3 D–19 8–2...
  • Page 355 SysDataInValid_L signal pin rules for SysDataOutClk_L signal pin 4–18 SysDataOutValid_L signal pin rules for 5–34 SysDc commands system probes, with SysDc field, system to 21264/EV68A commands 5–36 4–29 4–18 SYSDC_DELAY Cbox CSR defined 5–34 SysFillValid_L signal pin rules for System clock ratio configuration...
  • Page 356 WRITE_ONCE chain description Write-after-read. See WAR 4–21 Write-after-write. See WAW WrLWs, 21264/EV68A command WrQWs, 21264/EV68A command WrVictimBlk, 21264/EV68A command 4–39 system probes, with X convention 21264/EV68A Hardware Reference Manual 7–18 7–18 7–19 7–17 7–18 7–17 6–14 2–6 7–11 2–6 2–33...

This manual is also suitable for:

21264

Table of Contents