30044-23

GXM NSC | Alldatasheet

Document overview

• Manufacturer or author: Provided By ALLDATASHEET.COM(FREE DATASHEET DOWNLOAD SITE)
• PDF pages: 244

Technical content

© 2000 National Semiconductor Corporation www.national.com Geode™ GXm Processor Integrated x86 Solution with MMX Support April 2000 Geode™ GXm Processor Integrated x86 Solution with MMX Support General Description The National Semiconductor® Geode™ GXm processor is an advanced 32-bit x86 compatible processor offering high performance, fully accelerated 2D graphics, a 64-bit synchronous DRAM controller and a PCI bus controller, all on a single chip that is compatible with Intel’s MMX technology. The GXm processor core is a proven design that offers competitive CPU performance. It has integer and floating point execution units that are based on sixth-generation technology. The integer core contains a single, six-stage execution pipeline and offers advanced features such as operand forwarding, branch target buffers, and extensive write buffering. A 16 KB write-back L1 cache is accessed in a unique fashion that eliminates pipeline stalls to fetch operands that hit in the cache. In addition to the advanced CPU features, the GXm pro- cessor integrates a host of functions which are typically implemented with external components. A full-function graphics accelerator provides pixel processing and ren- dering functions. A separate on-chip video buffer enables >30 fps MPEG1 video playback when used together with the CS5530 I/O companion chip. Graphics and system memory accesses are supported by a tightly-coupled synchronous DRAM (SDRAM) memory controller. This tightly coupled memory subsystem eliminates the need for an external L2 cache. The GXm processor includes Virtual System Architec- ture ® (VSA™ technology) enabling XpressGRAPHICS and XpressAUDIO subsystems as well as generic emula- tion capabilities. Software handler routines for the Xpress- GRAPHICS and XpressAUDIO subsystems can be included in the BIOS and provide compatible VGA and 16- bit industry standard audio emulation. XpressAUDIO tech- nology eliminates much of the hardware traditionally asso- ciated with audio functions. Geode™ GXm Processor Internal Block Diagram Write-Back Unit FPU Internal Bus Interface Unit Graphics Memory Display PCI SDRAM Port CS5530 PCI Bus Integer Cache Unit Integrated Functions MMU (CRT/LCD TFT) X-Bus Pipeline Controller Controller Controller C-Bus National Semiconductor and Virtual System Architecture are registered trademarks of National Semiconductor Corporation. Geode and VSA are trademarks of National Semiconductor Corporation. For a complete listing of National Semiconductor trademarks, please visit www.national.com/trademarks.

www.national.com 2 Revision 3.1 Geode™ GXm Processor

Features

/G01 Packaged in: — 352-T erminal Ball Grid Array (BGA) or — 320-Pin Staggered Pin Grid Array (SPGA) /G01 0.35-micron four layer metal CMOS process /G01 Split rail design (3.3V I/O and 2.9V core) 32-Bit x86 Processor /G01 Supports the MMX instruction set extension for the acceleration of multimedia applications /G01 Speeds offered up to 266 MHz /G01 16 KB unified L1 cache /G01 Integrated Floating Point Unit (FPU) /G01 Re-entrant System Management Mode (SMM) enhanced for VSA PCI Controller /G01 Fixed, rotating, hybrid, or ping-pong arbitration /G01 Supports up to three PCI bus masters /G01 Synchronous CPU and PCI bus clock frequency /G01 Supports concurrency between PCI master and L1 cache Power Management /G01 Designed to support CS5530 power management architecture /G01 CPU only Suspend or full 3V Suspend supported: — Clocks to CPU core stopped for CPU Suspend — All on-chip clocks stopped for 3V Suspend — Suspend refresh supported for 3V Suspend Virtual Systems Architecture Technology /G01 Architecture allows OS independent (software) virtual- ization of hardware functions /G01 Provides compatible high performance legacy VGA core functionality Note: GUI (Graphical User Interface) graphics accel- eration is pure hardware. /G01 Provides 16-bit XpressAUDIO subsystem 2D Graphics Accelerator /G01 Graphics pipeline performance significantly increased over previous generations by pipelining burst reads/writes /G01 Accelerates BitBLTs, line draw, text /G01 Supports all 256 raster operations /G01 Supports transparent BLTs /G01 Runs at core clock frequency /G01 Full VGA and VESA mode support /G01 Special "Driver level” instructions utilize internal scratchpad for enhanced performance Display Controller /G01 Video Generator (VG) improves memory efficiency for display refresh with SDRAM /G01 Supports a separate MPEG1 video buffer and data path to enable video acceleration in the CS5530 /G01 Internal palette RAM for use with the CS5530 /G01 Direct interface to CS5530 for CRT and TFT flat panel support which eliminates need for external RAMDAC /G01 Hardware frame buffer compressor/decompressor /G01 Hardware cursor /G01 Supports up to 1280x1024x8 bpp and 1024x768x16 bpp XpressRAM Subsystem /G01 Memory control/interface directly from CPU /G01 64-Bit wide memory bus /G01 Support for: — Two 168-pin unbuffered DIMMs — Up to 16 open banks simultaneously — Single or 16-byte reads (burst length of two)

Revision 3.1 3 www.national.com Table of Contents Geode™ GXm Processor

www.national.com 4 Revision 3.1 Table of Contents(Continued) Geode™ GXm Processor

Revision 3.1 5 www.national.com Table of Contents(Continued) Geode™ GXm Processor

www.national.com 6 Revision 3.1 Table of Contents(Continued) Geode™ GXm Processor

Revision 3.1 7 www.national.com Table of Contents(Continued) Geode™ GXm Processor

www.national.com 8 Revision 3.1 Geode™ GXm Processor

1.0 Architecture Overview

The National Semiconductor Geode GXm processor is an x86-compatible 32-bit microprocessor. The decoupled load/store unit (within the memory management unit) allows multiple instructions in a single clock cycle. Other features include single-cycle execution, single-cycle instruction decode, 16 KB write-back cache, and clock r a t e su pt o2 6 6M H z .T h e s ef e a t u r e sa r em a d ep o s s i b l e by the use of advanced-process technologies and super- pipelining. The GXm processor has low power consumption at all clock frequencies. Where additional power savings are required, designers can make use of Suspend mode, Stop Clock capability, and System Management Mode (SMM). The GXm processor is divided into major functional blocks (as shown in Figure 1-1):  Integer Unit  Floating Point Unit (FPU)  Write-Back Cache Unit  Memory Management Unit (MMU)  Internal Bus Interface Unit  Integrated Functions Instructions are executed in the integer unit and in the floating point unit. The cache unit stores the most recently used data and instructions and provides fast access to this information for the integer and floating point units.

1.1 INTEGER UNIT

The integer unit consists of:  Instruction Buffer  Instruction Fetch  Instruction Decoder and Execution The superpipelined integer unit fetches, decodes, and executes x86 instructions through the use of a six-stage integer pipeline. The instruction fetch pipeline stage generates, from the on-chip cache, a continuous high-speed instruction stream for use by the processor. Up to 128 bits of code are read during a single clock cycle. Branch prediction logic within the prefetch unit generates a predicted target address for unconditional or conditional branch instructions. When a branch instruction is detected, the instruction fetch stage starts loading instruc- tions at the predicted address within a single clock cycle. Up to 48 bytes of code are queued prior to the instruction decode stage. The instruction decode stage evaluates the code stream provided by the instruction fetch stage and determines the number of bytes in each instruction and the instruction type. Instructions are processed and decoded at a maxi- mum rate of one instruction per clock. The address calculation function is super-pipelined and contains two stages, AC1 and AC2. If the instruction refers to a memory operand, AC1 calculates a linear memory address for the instruction. The AC2 stage performs any required memory manage- ment functions, cache accesses, and register file accesses. If a floating point instruction is detected by AC2, the instruction is sent to the floating point unit for processing. The execution stage, under control of microcode, exe- cutes instructions using the operands provided by the address calculation stage. Write-back,the last stage of the integer unit, updates the register file within the integer unit or writes to the load/store unit within the memory management unit. Figure 1-1. Internal Block Diagram Write-Back Unit FPU Internal Bus Interface Unit Graphics Memory Display PCI SDRAM Port CS5530 PCI Bus Integer Cache Unit Integrated Functions MMU (CRT/LCD TFT) X-Bus Pipeline Controller Controller Controller C-Bus

Revision 3.1 9 www.national.com Architecture Overview (Continued) Geode™ GXm Processor

1.2 FLOATING POINT UNIT

The FPU (Floating Point Unit) interfaces to the integer unit and the cache unit through a 64-bit bus. The FPU is x87- instruction-set compatible and adheres to the IEEE-754 standard. Because almost all applications that contain FPU instructions also contain integer instructions, the GXm processor’s FPU achieves high performance by completing integer and FPU operations in parallel. FPU instructions are dispatched to the pipeline within the integer unit. The address calculation stage of the pipeline checks for memory management exceptions and accesses memory operands for use by the FPU. Once the instructions and operands have been provided to the FPU, the FPU completes instruction execution independently of the integer unit.

1.3 WRITE-BACK CACHE UNIT

The 16 KB write-back unified cache is a data/instruction cache and is configured as four-way set associative. The cache stores up to 16 KB of code and data in 1024 cache lines. The GXm processor provides the ability to allocate a por- tion of the L1 cache as a scratchpad, which is used to accelerate the Virtual Systems Architecture algorithms as well as for some graphics operations.

1.4 MEMORY MANAGEMENT UNIT

The memory management unit (MMU) translates the lin- ear address supplied by the integer unit into a physical address to be used by the cache unit and the internal bus interface unit. Memory management procedures are x86- compatible, adhering to standard paging mechanisms. The MMU also contains a load/store unit that is responsi- ble for scheduling cache and external memory accesses. The load/store unit incorporates two performance- enhancing features:  Load-store reorderingthat gives priority to memory reads required by the integer unit over writes to external memory.  Memory-read bypassingthat eliminates unnecessary memory reads by using valid data from the execution unit.

1.4.1 Internal Bus Interface Unit

The internal bus interface unit provides a bridge from the GXm processor to the integrated system functions (i.e., memory subsystem, display controller, graphics pipeline) and the PCI bus interface. When external memory access is required, the physical address is calculated by the memory management unit and then passed to the internal bus interface unit, which translates the cycle to an X-Bus cycle (the X-Bus is a National Semiconductor proprietary internal bus which provides a common interface for all of the system mod- ules). The X-Bus memory cycle now is arbitrated between other pending X-Bus memory requests to the SDRAM controller before completing. In addition, the internal bus interface unit provides config- uration control for up to 20 different regions within system memory with separate controls for read access, write access, cacheability, and PCI access.

1.5 INTEGRATED FUNCTIONS

The GXm processor integrates the following functions tra- ditionally implemented using external devices:  High-performance 2D graphics accelerator  Separate CRT and TFT data paths from the display controller  SDRAM memory controller  PCI bridge The processor has also been enhanced to support National Semiconductor’s proprietary Virtual System Architecture (VSA) implementation. The GXm processor implements a Unified Memory Archi- tecture (UMA). By using National Semiconductor’s Dis- play Compression Technology (DCT), the performance degradation inherent in traditional UMA systems is elimi- nated.

1.5.1 Graphics Accelerator

The graphics accelerator is a full-featured GUI (Graphical User Interface) accelerator. The graphics pipeline imple- ments a bitBLT engine for frame buffer bitBLTs and rect- angular fills. Additional instructions in the integer unit may be processed, as the bitBLT engine assists the CPU in the bitBLT operations that take place between system mem- ory and the frame buffer. This combination of hardware and software is used by the display driver to provide very fast transfers in both directions between system memory and the frame buffer. The bitBLT engine also draws ran- domly-oriented vectors, and scanlines for polygon fill. All of the pipeline operations described in the following list can be applied to any bitBLT operation.  Pattern Memory.Render with 8x8 dither, 8x8 mono- chrome, or 8x1 color pattern.  Color Expansion.Expand monochrome bitmaps to full-depth 8- or 16-bit colors.  Transparency.Suppresses drawing of background pixels for transparent text.  Raster Operations.Boolean operation combines source, destination, and pattern bitmaps.

www.national.com 10 Revision 3.1 Architecture Overview (Continued) Geode™ GXm Processor

1.5.2 Display Controller

The display port is a direct interface to the CS5530 which drives a TFT flat panel display, LCD panel, or a CRT dis- play. The display controller (video generator) retrieves image data from the frame buffer region of memory, performs a color-look-up if required, inserts the cursor overlay into the pixel stream, generates display timing, and formats the pixel data for output to a variety of display devices. The display controller contains Display Compression Technology (DCT) that allows the GXm processor to refresh the display from a compressed copy of the frame buffer. DCT typically decreases the screen-refresh band- width requirement by a factor of 15 to 20, further minimiz- ing bandwidth contention.

1.5.3 XpressRAM Memory Subsystem

The memory controller drives a 64-bit SDRAM port directly. The SDRAM memory array contains both the main system memory and the graphics frame buffer. Up to four module banks of SDRAM are supported. Each mod- ule bank will have two or four component banks depend- ing on the memory size and organization. The maximum configuration is four module banks with four component banks providing a total of 16 open banks. The maximum memory size is 1 GB. The memory controller handles multiple requests for memory data from the GXm processor, the graphics accelerator and the display controller. The memory con- troller contains extensive buffering logic that helps mini- mize contention for memory bandwidth between graphics and CPU requests. The memory controller cooperates with the internal bus controller to determine the cacheabil- ity of all memory references.

1.5.4 PCI Controller

The GXm processor incorporates a full-function PCI inter- face module that includes the PCI arbiter. All accesses to external I/O devices are sent over the PCI bus, although most memory accesses are serviced by the SDRAM con- troller. The Internal Bus Interface Unit contains address mapping logic that determines if memory accesses are targeted for the SDRAM or for the PCI bus.

Revision 3.1 11 www.national.com Architecture Overview (Continued) Geode™ GXm Processor

1.6 GEODE GXM/CS5530 SYSTEM DESIGNS

The GXm Integrated Subsystem with MMX support con- sists of two chips, the GXm Processor and the CS5530 I/O companion. The subsystem provides high perfor- mance using 32-bit x86 processing. The two chips inte- grate video, audio and memory interface functions normally performed by external hardware. As described in separate manuals, the CS5530 enables the full features of the GXm processor with MMX support. These features include full VGA and VESA video, 16-bit stereo sound, IDE interface, ISA interface, SMM power management, and AT compatibility logic. In addition, the newer CS5530 provides an Ultra DMA/33 interface, MPEG2 assist, and AC97 Version 2.0 compliant audio. Figure 1-2 shows a basic block system diagram (refer to Figure 2-4 on page 34 for detailed subsystem intercon- nection signals). It includes the National Semiconductor CS9210 Dual-Scan Flat Panel Display Controller for designs that need to interface to a DSTN panel (instead of TFT panel). Figure 1-2. Geode™ GXm/CS5530 System Block Diagram YUV Port (Video) RGB Port PCI Interface SDRAMMD[63:0] PCI Bus Geode™ CS5530 I/O Companion Graphics Data Video Data Analog RGB Digital RGB(to TFT or DSTN Panel) CRT TFT Panel USB (2 Ports) AC97 Codec Speakers CD ROM Audio Micro- phone GPIO Port (Graphics) Super ISA Bus SDRAM Serial Packet Clocks I/O BIOS IDE Devices14.31818 MHz Crystal IDE Control System Clocks DC-DC & Battery CS9210 DSTN Controller DSTN Panel Geode™ GXm Processor Geode™

www.national.com 12 Revision 3.1 Architecture Overview (Continued) Geode™ GXm Processor The CS9210 converts the digital RGB output of the CS5530 I/O companion chip to the digital output suitable for driving a dual-scan color STN (DSTN) flat panel LCD. It connects to the digital RGB output of a GXm processor or 55x0 and drives the graphics data onto a dual-scan flat panel LCD. It can drive all standard dual-scan color STN flat panels up to 1024x768 resolution. Figure 1-3 shows an example of a CS9210 interface in a typical GXm Inte- grated Subsystem. Figure 1-3. CS9210 Interface System Diagram DRAM Data Address Control13 Panel Control 6 24Panel Data DSTN Pixel Port Pixel Data LCD CS5530 CS9210 DSTN Controller I/O DRAM-B 256Kx16 Bit DRAM-A 256Kx16 Bit Address Control13 DRAM Data 16 Serial ConfigurationCompanion (Control & Data) Geode™ GXm Processor Geode™ Geode™

Revision 3.1 13 www.national.com Geode™ GXm Processor

2.0 Signal Definitions

This section describes the external interface of the Geode GXm processor. Figure 2-1 shows the signals organized by their functional interface groups (internal test and elec- trical pins are not shown).

2.1 PIN ASSIGNMENTS

The tables in this section use several common abbrevia- tions. Table 2-1 lists the mnemonics and their meanings. Figure 2-2 on page 14 shows the pin assignment for the

352 BGA with Tables 2-2 and 2-3 listing the pin assign-

ments sorted by pin number and alphabetically by signal name, respectively. Figure 2-3 on page 19 shows the pin assignment for the

320 SPGA with Tables 2-4 and 2-5 listing the pin assign-

ments sorted by pin number and alphabetically by signal name, respectively. In Section 2.2 “Signal Descriptions” starting on Page 24 a description of each signal is provided within its associated functional group. Following the signal descriptions, information regarding subsystem signal connections and split power planes and decoupling is provided. Figure 2-1. Functional Block Diagram Table 2-1. Pin Type Definitions Mnemonic Definition I Standard input pin. I/O Bidirectional pin. O T otem-pole output. OD Open-drain output structure that allows multiple devices to share the pin in a wired-OR configuration PU Pull-up resistor PD Pull-down resistor s/t/s Sustained tri-state, an active-low tri-state signal owned and driven by one and only one agent at a time. The agent that drives an s/t/s pin low must drive it high for at least one clock before letting it float. A new agent cannot start driving an s/t/s signal any sooner than one clock after the previous owner lets it float. A pull-up resistor is required to sustain the inactive state until another agent drives it, and must be provided by the central resource. VCC (PWR) Power pin. VSS (GND) Ground pin # The "#" symbol at the end of a signal name indicates that the active, or asserted state occurs when the signal is at a low voltage level. When "#" is not present after the signal name, the signal is asserted when at a high voltage level. SYSCLK CLKMODE[2:0] RESET INTR IRQ13 SMI# SUSP# SUSPA# SERIALP AD[31:0] C/BE[3:0]# PAR FRAME# IRDY# TRDY# STOP# LOCK# DEVSEL# PERR# SERR# REQ[2:0]# GNT[2:0]# MD[63:0] MA[12:0] BA[1:0] RASA#, RASB# CASA#, CASB# CS[3:0]# WEA#, WEB# DQM[7:0] CKEA, CKEB SDCLK[3:0] SDCLK_IN SDCLK_OUT PCLK VID_CLK DCLK CRT_HSYNC CRT_VSYNC FP_VSYNC FP_HSYNC ENA_DISP VID_RDY VID_VAL VID_DA TA[7:0] PIXEL[17:0] Memory Controller Interface Video Interface Signals PCI Interface Signals System Interface Signals Signals Geode™ GXm Processor

www.national.com 14 Revision 3.1 Signal Definitions (Continued) Geode™ GXm Processor Figure 2-2. 352 BGA Pin Assignment Diagram For order information refer to Section A.1 “Order Information” on page 236. 123456789 1 0 1 1 1 2 1 3 1 4 1 5 1 6 1 7 1 8 1 9 2 0 21 22 23 24 25 26 A B C D E F G H J K L M N P R T U V W Y AA AB AC AD AE AF Index Corner VSS VSS AD27 AD24 AD21 AD16 VCC2 FRAM#DEVS# VCC3 PERR# AD15 VSS AD11 CBE0# AD6 VCC2 AD4 AD2 VCC3 AD0 AD1 TEST2 MD2 VSS VSS VSS VSS AD28 AD25 AD22 AD18 VCC2 CBE2# TRDY# VCC3 LOCK# PAR AD14 AD12 AD9 AD7 VCC2 INTR AD3 VCC3 TEST1 TEST3 MD1 MD33 VSS VSS AD29 AD31 AD30 AD26 AD23 AD19 VCC2 AD17 IRDY# VCC3 STOP#SERR# CBE1# AD13 AD10 AD8 VCC2 AD5 SMI# VCC3 TEST0 IRQ13 MD32 MD34 MD3 MD35 GNT0# TDI REQ2# VSS CBE3# VSS VCC2 VSS VSS VCC3 VSS VSS VSS VSS VSS VSS VCC2 VSS VSS VCC3 VSS MD0 VSS MD4 MD36 TDN GNT2# SUSPA# REQ0# AD20 MD6 TDP MD5 MD37 TD0 GNT1# TEST VSS VSS MD38 MD7 MD39 VCC3 VCC3 VCC3 VCC3 VCC3 VCC3 VCC3 VCC3 TMS SUSP#REQ1# VSS VSS MD8 MD40 MD9 FPVSY TCLK RESET VSS VSS MD41 MD10 MD42 VCC2 VCC2 VCC2 VCC2 VCC2 VCC2 VCC2 VCC2 CKM1 FPHSYSERLP VSS VSS MD11 MD43 MD12 CKM2 VIDVAL CKM0 VSS VSS MD44 MD13 MD45 VSS PIX1 PIX0 VSS VSS MD14 MD46 MD15 VIDCLK PIX3 PIX2 VSS VSS MD47 CASA# SYSCLK PIX4 PIX5 PIX6 VSS VSS WEB# WEA# CASB# PIX7 PIX8 PIX9 VSS VSS DQM0 DQM4 DQM1 VCC3 VCC3 VCC3 VCC3 VCC3 VCC3 VCC3 VCC3 PIX10 PIX11 PIX12 VSS VSS DQM5 CS2# CS0# PIX13 CRTHS PIX14 VSS VSS RASA#RASB# MA0 VCC2 VCC2 VCC2 VCC2 VCC2 VCC2 VCC2 VCC2 PIX15 PIX16 CRTVS VSS VSS MA1 MA2 MA3 DCLK PIX17 VDA T6 VDA T7 MA4 MA5 MA6 MA7 PCLK FLT# VDA T4 VSS VOLDET VSS VCC2 VSS VSS VCC3 VSS VSS VSS VSS VSS VSS VCC2 VSS VSS VCC3 VSS DQM6 VSS MA8 MA9 MA10 VRDY VDA T5 VDA T3 VDA T0 EDISP MD63 VCC2 MD62 MD29 VCC3 MD59 MD26 MD56 MD55 MD22 CKEB VCC2 MD51 MD18 VCC3 MD48 DQM3 CS1# MA11 BA0 BA1 VSS VSS VDA T2 SCLK3 SCLK1RWCLK VCC2 SCKIN MD61 VCC3 MD28 MD58 MD25 MD24 MD54 MD21 VCC2 MD20 MD50 VCC3 MD17 DQM7 CS3# MA12 VSS VSS VSS VSS VDA T1 SCLK0 SCLK2 MD31 VCC2SCKOUTMD30 VCC3 MD60 MD27 MD57 VSS MD23 MD53 VCC2 MD52 MD19 VCC3 MD49 MD16 DQM2 CKEA VSS VSS 123456789 1 0 1 1 1 2 1 3 1 4 1 5 1 6 1 7 1 8 1 9 2 0 21 22 23 24 25 26 A B C D E F G H J K L M N P R T U V W Y AA AB AC AD AE AF Geode™ GXm

352 BGA - Top View

Note:Signal names have been abbreviated in this figure due to space constraints. = GND terminal = PWR terminal (VCC2 = VCC_CORE; VCC3 = VCC_IO) Processor

Revision 3.1 15 www.national.com Signal Definitions (Continued) Geode™ GXm Processor Table 2-2. 352 BGA Pin Assignments - Sorted by Pin Number Pin No. Signal Name A1 VSS A2 VSS A3 AD27 A4 AD24 A5 AD21 A6 AD16 A7 VCC2 A8 FRAME# A9 DEVSEL# A10 VCC3 A11 PERR# A12 AD15 A13 VSS A14 AD11 A15 C/BE0# A16 AD6 A17 VCC2 A18 AD4 A19 AD2 A20 VCC3 A21 AD0 A22 AD1 A23 TEST2 A24 MD2 A25 VSS A26 VSS B1 VSS B2 VSS B3 AD28 B4 AD25 B5 AD22 B6 AD18 B7 VCC2 B8 C/BE2# B9 TRDY# B10 VCC3 B11 LOCK# B12 PA R B13 AD14 B14 AD12 B15 AD9 B16 AD7 B17 VCC2 B18 INTR B19 AD3 B20 VCC3 B21 TEST1 B22 TEST3 B23 MD1 B24 MD33 B25 VSS B26 VSS C1 AD29 C2 AD31 C3 AD30 C4 AD26 C5 AD23 C6 AD19 C7 VCC2 C8 AD17 C9 IRDY# C10 VCC3 C11 STOP# C12 SERR# C13 C/BE1# C14 AD13 C15 AD10 C16 AD8 C17 VCC2 C18 AD5 C19 SMI# C20 VCC3 C21 TEST0 C22 IRQ13 C23 MD32 C24 MD34 C25 MD3 C26 MD35 D1 GNT0# D2 TDI D3 REQ2# D4 VSS D5 C/BE3# D6 VSS D7 VCC2 D8 VSS D9 VSS D10 VCC3 D11 VSS D12 VSS D13 VSS D14 VSS D15 VSS D16 VSS D17 VCC2 D18 VSS Pin No. Signal Name D19 VSS D20 VCC3 D21 VSS D22 MD0 D23 VSS D24 MD4 D25 MD36 D26 TDN E1 GNT2# E2 SUSP A# E3 REQ0# E4 AD20 E23 MD6 E24 TDP E25 MD5 E26 MD37 F1 TDO F2 GNT1# F3 TEST F4 VSS F23 VSS F24 MD38 F25 MD7 F26 MD39 G1 VCC3 G2 VCC3 G3 VCC3 G4 VCC3 G23 VCC3 G24 VCC3 G25 VCC3 G26 VCC3 H1 TMS H2 SUSP# H3 REQ1# H4 VSS H23 VSS H24 MD8 H25 MD40 H26 MD9 J1 FP_VSYNC J2 TCLK J3 RESET J4 VSS J23 VSS J24 MD41 J25 MD10 J26 MD42 Pin No. Signal Name K1 VCC2 K2 VCC2 K3 VCC2 K4 VCC2 K23 VCC2 K24 VCC2 K25 VCC2 K26 VCC2 L1 CLKMODE1 L2 FP_HSYNC L3 SERIALP L4 VSS L23 VSS L24 MD11 L25 MD43 L26 MD12 M1 CLKMODE2 M2 VID_VAL M3 CLKMODE0 M4 VSS M23 VSS M24 MD44 M25 MD13 M26 MD45 N1 VSS N2 PIXEL1 N3 PIXEL0 N4 VSS N23 VSS N24 MD14 N25 MD46 N26 MD15 P1 VID_CLK P2 PIXEL3 P3 PIXEL2 P4 VSS P23 VSS P24 MD47 P25 CASA# P26 SYSCLK R1 PIXEL4 R2 PIXEL5 R3 PIXEL6 R4 VSS R23 VSS R24 WEB# R25 WEA# R26 CASB# Pin No. Signal Name T1 PIXEL7 T2 PIXEL8 T3 PIXEL9 T4 VSS T23 VSS T24 DQM0 T25 DQM4 T26 DQM1 U1 VCC3 U2 VCC3 U3 VCC3 U4 VCC3 U23 VCC3 U24 VCC3 U25 VCC3 U26 VCC3 V1 PIXEL10 V2 PIXEL11 V3 PIXEL12 V4 VSS V23 VSS V24 DQM5 V25 CS2# V26 CS0# W1 PIXEL13 W2 CRT_HSYNC W3 PIXEL14 W4 VSS W23 VSS W24 RASA# W25 RASB# W26 MA0 Y1 VCC2 Y2 VCC2 Y3 VCC2 Y4 VCC2 Y23 VCC2 Y24 VCC2 Y25 VCC2 Y26 VCC2 AA1 PIXEL15 AA2 PIXEL16 AA3 CRT_VSYNC AA4 VSS AA23 VSS AA24 MA1 AA25 MA2 AA26 MA3 Pin No. Signal Name

www.national.com 16 Revision 3.1 Signal Definitions (Continued) Geode™ GXm Processor AB1 DCLK AB2 PIXEL17 AB3 VID_DATA6 AB4 VID_DATA7 AB23 MA4 AB24 MA5 AB25 MA6 AB26 MA7 AC1 PCLK AC2 FLT# AC3 VID_DATA4 AC4 VSS AC5 VOLDET AC6 VSS AC7 VCC2 AC8 VSS AC9 VSS AC10 VCC3 AC11 VSS AC12 VSS AC13 VSS AC14 VSS AC15 VSS Pin No. Signal Name AC16 VSS AC17 VCC2 AC18 VSS AC19 VSS AC20 VCC3 AC21 VSS AC22 DQM6 AC23 VSS AC24 MA8 AC25 MA9 AC26 MA10 AD1 VID_RDY AD2 VID_DATA5 AD3 VID_DATA3 AD4 VID_DATA0 AD5 ENA_DISP AD6 MD63 AD7 VCC2 AD8 MD62 AD9 MD29 AD10 VCC3 AD11 MD59 AD12 MD26 Pin No. Signal Name AD13 MD56 AD14 MD55 AD15 MD22 AD16 CKEB AD17 VCC2 AD18 MD51 AD19 MD18 AD20 VCC3 AD21 MD48 AD22 DQM3 AD23 CS1# AD24 MA11 AD25 BA0 AD26 BA1 AE1 VSS AE2 VSS AE3 VID_DATA2 AE4 SDCLK3 AE5 SDCLK1 AE6 RW_CLK AE7 VCC2 AE8 SDCLK_IN AE9 MD61 Pin No. Signal Name AE10 VCC3 AE11 MD28 AE12 MD58 AE13 MD25 AE14 MD24 AE15 MD54 AE16 MD21 AE17 VCC2 AE18 MD20 AE19 MD50 AE20 VCC3 AE21 MD17 AE22 DQM7 AE23 CS3# AE24 MA12 AE25 VSS AE26 VSS AF1 VSS AF2 VSS AF3 VID_DATA1 AF4 SDCLK0 AF5 SDCLK2 AF6 MD31 Pin No. Signal Name AF7 VCC2 AF8 SDCLK_OUT AF9 MD30 AF10 VCC3 AF11 MD60 AF12 MD27 AF13 MD57 AF14 VSS AF15 MD23 AF16 MD53 AF17 VCC2 AF18 MD52 AF19 MD19 AF20 VCC3 AF21 MD49 AF22 MD16 AF23 DQM2 AF24 CKEA AF25 VSS AF26 VSS Pin No. Signal Name Table 2-2. 352 BGA Pin Assignments - Sorted by Pin Number (Continued)

Revision 3.1 17 www.national.com Signal Definitions (Continued) Geode™ GXm Processor Table 2-3. 352 BGA Pin Assignments - Sorted Alphabetically by Signal Name Signal Name Type Pin No. AD0 I/O A21 AD1 I/O A22 AD2 I/O A19 AD3 I/O B19 AD4 I/O A18 AD5 I/O C18 AD6 I/O A16 AD7 I/O B16 AD8 I/O C16 AD9 I/O B15 AD10 I/O C15 AD11 I/O A14 AD12 I/O B14 AD13 I/O C14 AD14 I/O B13 AD15 I/O A12 AD16 I/O A6 AD17 I/O C8 AD18 I/O B6 AD19 I/O C6 AD20 I/O E4 AD21 I/O A5 AD22 I/O B5 AD23 I/O C5 AD24 I/O A4 AD25 I/O B4 AD26 I/O C4 AD27 I/O A3 AD28 I/O B3 AD29 I/O C1 AD30 I/O C3 AD31 I/O C2 BA0 O AD25 BA1 O AD26 CASA# O P25 CASB# O R26 C/BE0# I/O A15 C/BE1# I/O C13 C/BE2# I/O B8 C/BE3# I/O D5 CKEA O AF24 CKEB O AD16 CLKMODE0 I M3 CLKMODE1 I L1 CLKMODE2 I M1 CRT_HSYNC O W2 CRT_VSYNC O AA3 CS0# O V26 CS1# O AD23 CS2# O V25 CS3# O AE23 DCLK I AB1 DEVSEL# s/t/s A9 (PU) DQM0 O T24 DQM1 O T26 DQM2 O AF23 DQM3 O AD22 DQM4 O T25 DQM5 O V24 DQM6 O AC22 DQM7 O AE22 ENA_DISP O AD5 FLT# I AC2 FP_HSYNC O L2 FP_VSYNC O J1 FRAME# s/t/s A8 (PU) GNT0# O D1 GNT1# O F2 GNT2# O E1 INTR I B18 IRDY# s/t/s C9 (PU) IRQ13 O C22 LOCK# s/t/s B11 (PU) MA0 O W26 MA1 O AA24 MA2 O AA25 MA3 O AA26 MA4 O AB23 MA5 O AB24 MA6 O AB25 MA7 O AB26 MA8 O AC24 MA9 O AC25 MA10 O AC26 MA11 O AD24 MA12 O AE24 MD0 I/O D22 MD1 I/O B23 MD2 I/O A24 MD3 I/O C25 MD4 I/O D24 MD5 I/O E25 MD6 I/O E23 MD7 I/O F25 MD8 I/O H24 MD9 I/O H26 MD10 I/O J25 MD11 I/O L24 MD12 I/O L26 MD13 I/O M25 MD14 I/O N24 MD15 I/O N26 MD16 I/O AF22 MD17 I/O AE21 MD18 I/O AD19 MD19 I/O AF19 Signal Name Type Pin No. MD20 I/O AE18 MD21 I/O AE16 MD22 I/O AD15 MD23 I/O AF15 MD24 I/O AE14 MD25 I/O AE13 MD26 I/O AD12 MD27 I/O AF12 MD28 I/O AE11 MD29 I/O AD9 MD30 I/O AF9 MD31 I/O AF6 MD32 I/O C23 MD33 I/O B24 MD34 I/O C24 MD35 I/O C26 MD36 I/O D25 MD37 I/O E26 MD38 I/O F24 MD39 I/O F26 MD40 I/O H25 MD41 I/O J24 MD42 I/O J26 MD43 I/O L25 MD44 I/O M24 MD45 I/O M26 MD46 I/O N25 MD47 I/O P24 MD48 I/O AD21 MD49 I/O AF21 MD50 I/O AE19 MD51 I/O AD18 MD52 I/O AF18 MD53 I/O AF16 MD54 I/O AE15 MD55 I/O AD14 MD56 I/O AD13 MD57 I/O AF13 MD58 I/O AE12 MD59 I/O AD11 MD60 I/O AF11 MD61 I/O AE9 MD62 I/O AD8 MD63 I/O AD6 PAR I/O B12 PCLK O AC1 PERR# s/t/s A11 (PU) PIXEL0 O N3 PIXEL1 O N2 PIXEL2 O P3 PIXEL3 O P2 PIXEL4 O R1 PIXEL5 O R2 Signal Name Type Pin No. PIXEL6 O R3 PIXEL7 O T1 PIXEL8 O T2 PIXEL9 O T3 PIXEL10 O V1 PIXEL11 O V2 PIXEL12 O V3 PIXEL13 O W1 PIXEL14 O W3 PIXEL15 O AA1 PIXEL16 O AA2 PIXEL17 O AB2 RASA# O W24 RASB# O W25 REQ0# I E3 (PU) REQ1# I H3 (PU) REQ2# I D3 (PU) RESET I J3 RW_CLK O AE6 SDCLK_IN I AE8 SDCLK_OUT O AF8 SDCLK0 O AF4 SDCLK1 O AE5 SDCLK2 O AF5 SDCLK3 O AE4 SERIALP O L3 SERR# OD C12 (PU) SMI# I C19 STOP# s/t/s C11 (PU) SUSP# I H2 (PU) SUSPA# O E2 SYSCLK I P26 TCLK I J2 (PU) TDI I D2 (PU) TDN O D26 TDO O F1 TDP O E24 TEST I F3 (PD) TEST0 O C21 TEST1 O B21 TEST2 O A23 TEST3 O B22 TMS I H1 (PU) TRDY# s/t/s B9 (PU) VCC2 PWR A7 VCC2 PWR A17 VCC2 PWR B7 VCC2 PWR B17 VCC2 PWR C7 VCC2 PWR C17 VCC2 PWR D7 VCC2 PWR D17 VCC2 PWR K1 Signal Name Type Pin No.

www.national.com 18 Revision 3.1 Signal Definitions (Continued) Geode™ GXm Processor Note: PU/PD indicates pin is internally connected to a 20-kohm pull-up/- down resistor. VCC2 PWR K2 VCC2 PWR K3 VCC2 PWR K4 VCC2 PWR K23 VCC2 PWR K24 VCC2 PWR K25 VCC2 PWR K26 VCC2 PWR Y1 VCC2 PWR Y2 VCC2 PWR Y3 VCC2 PWR Y4 VCC2 PWR Y23 VCC2 PWR Y24 VCC2 PWR Y25 VCC2 PWR Y26 VCC2 PWR AC7 VCC2 PWR AC17 VCC2 PWR AD7 VCC2 PWR AD17 VCC2 PWR AE7 VCC2 PWR AE17 VCC2 PWR AF7 VCC2 PWR AF17 VCC3 PWR A10 VCC3 PWR A20 VCC3 PWR B10 VCC3 PWR B20 VCC3 PWR C10 VCC3 PWR C20 VCC3 PWR D10 VCC3 PWR D20 VCC3 PWR G1 VCC3 PWR G2 VCC3 PWR G3 VCC3 PWR G4 VCC3 PWR G23 VCC3 PWR G24 Signal Name Type Pin No. VCC3 PWR G25 VCC3 PWR G26 VCC3 PWR U1 VCC3 PWR U2 VCC3 PWR U3 VCC3 PWR U4 VCC3 PWR U23 VCC3 PWR U24 VCC3 PWR U25 VCC3 PWR U26 VCC3 PWR AC10 VCC3 PWR AC20 VCC3 PWR AD10 VCC3 PWR AD20 VCC3 PWR AE10 VCC3 PWR AE20 VCC3 PWR AF10 VCC3 PWR AF20 VID_CLK O P1 VID_DATA0 O AD4 VID_DATA1 O AF3 VID_DATA2 O AE3 VID_DATA3 O AD3 VID_DATA4 O AC3 VID_DATA5 O AD2 VID_DATA6 O AB3 VID_DATA7 O AB4 VID_RDY I AD1 VID_VAL O M2 VOLDET O AC5 VSS GND A1 VSS GND A2 VSS GND A13 VSS GND A25 VSS GND A26 VSS GND B1 VSS GND B2 Signal Name Type Pin No. VSS GND B25 VSS GND B26 VSS GND D4 VSS GND D6 VSS GND D8 VSS GND D9 VSS GND D11 VSS GND D12 VSS GND D13 VSS GND D14 VSS GND D15 VSS GND D16 VSS GND D18 VSS GND D19 VSS GND D21 VSS GND D23 VSS GND F4 VSS GND F23 VSS GND H4 VSS GND H23 VSS GND J4 VSS GND J23 VSS GND L4 VSS GND L23 VSS GND M4 VSS GND M23 VSS GND N1 VSS GND N4 VSS GND N23 VSS GND P4 VSS GND P23 VSS GND R4 VSS GND R23 VSS GND T4 VSS GND T23 VSS GND V4 VSS GND V23 Signal Name Type Pin No. VSS GND W4 VSS GND W23 VSS GND AA4 VSS GND AA23 VSS GND AC4 VSS GND AC6 VSS GND AC8 VSS GND AC9 VSS GND AC11 VSS GND AC12 VSS GND AC13 VSS GND AC14 VSS GND AC15 VSS GND AC16 VSS GND AC18 VSS GND AC19 VSS GND AC21 VSS GND AC23 VSS GND AE1 VSS GND AE2 VSS GND AE25 VSS GND AE26 VSS GND AF1 VSS GND AF2 VSS GND AF14 VSS GND AF25 VSS GND AF26 WEA# O R25 WEB# O R24 Signal Name Type Pin No. Table 2-3. 352 BGA Pin Assignments - Sorted Alphabetically by Signal Name (Continued)

Revision 3.1 19 www.national.com Signal Definitions (Continued) Geode™ GXm Processor Figure 2-3. 320 SPGA Pin Assignment Diagram For order information refer to Section A.1 “Order Information” on page 236. 123456789 1 0 1 1 1 2 1 3 1 4 1 5 1 6 1 7 1 8 1 9 2 0 21 22 23 24 25 26 A B C D E F G H J K L M N P Q R S T U V AA AB AC AD AE AF Index Corner 27 28 29 30 31 32 33 34 35 36 37 AG AH AJ AK AL AM W Y X Z AN A B C D E F G H J K L M N P Q R S T U V AA AB AC AD AE AF AG AH AJ AK AL AM W Y X Z AN 123456789 1 0 1 1 1 2 1 3 1 4 1 5 1 6 1 7 1 8 1 9 2 0 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 VCC3 AD25 VSS VCC2 AD16 VCC3 STOP# SERR# VSS AD11 AD8 VCC3 AD2 VCC2 VSS TST0 VCC3 VSS VSS AD27 CBE3# AD21 AD19 CBE2# TRDY# LOCK# CBE1# AD13 AD9 AD6 AD3 SMI# AD1 TST2 MD33 MD2 VCC3 AD31 AD26 AD23 VCC2 AD18 FRAME# VSS PAR VCC3 AD10 VSS AD4 AD0 VCC2 IRQ13 MD1 MD34 VCC3 AD30 AD29 AD24 AD22 AD20 AD17 IRDY# PERR# AD14 AD12 AD7 INTR TST1 TST3 MD0 MD32 MD3 MD35 REQ0# REQ2# AD28 VSS VCC2 VCC2 VSS DEVSEL# AD15 VSS CBE0# AD5 VSS VCC2 VCC2 VSS MD4 MD36 TDN GNT0# TDI MD5 TDP VSS CLKMODE2 VSS VSS MD37 VSS GNT2# SUSPA# TDO VSS TEST REQ1# GNT1# VCC2 VCC2 VCC2 RESET SUSP# VCC3 TMS VSS FPVSYN TCK SERIALP VSS NC CKMD1 FPHSYN CKMD0 VID_VAL PIX0 PIX1 PIX2 VSS VCC3 VSS PIX3 VID_CLK PIX6 PIX5 PIX4 NC PIX9 PIX8 VSS PIX7 NC PIX10 VCC3 PIX11 VSS PIX12 PIX13 VCC2 VCC2 VCC2 CRTHSYN DCLK PIX14 VSS VCC2 PIX15 PIX16 VSS PIX17 VSS CRTVSYN VDAT6 MD6 MD38 VCC2 VSS MD7 MD39 MD8 VCC2 VCC2 VCC2 MD40 MD9 VSS MD41 VCC3 MD10 MD42 MD11 VSS MD43 MD44 MD12 MD14 MD13 MD45 MD15 MD46 VSS VCC3 VSS SYSCLK MD47 WEA# WEB# CASA# DQM0 CASB# DQM1 VSS DQM4 CS2# DQM5 VSS CS0# VCC3 RASB# RASA# VCC2 VCC2 VCC2 VCC2 VSS MA1 MA2 MA0 MA4 MA3 VSS MA5 VSS MA8 MA6MA10 PCLK FLT# VDAT5 VSS VCC2 MD31 VSS MD60 MD57 VSS MD22 MD52 VSS VCC2 VCC2 VSS BA1 MA9 MA7 VRDY VSS VDAT0 SDCLK0 SDCLK2 SDCLKIN MD29 MD27 MD56 MD55 MD21 MD20 MD50 MD16 DQM3 CS3# VSS BA0 VCC2 VDA T4 VDAT2 SDCLK1 VCC2 RWCLK SDCLKOUT VSS MD58 VCC3 MD23 VSS MD19 MD49 VCC2 DQM6 CKEA MA11 VCC3 VDAT7 VDAT3 ENDIS SDCLK3 MD63 MD30 MD61 MD59 MD25 MD24 MD53 MD51 MD18 MD48 DQM7 DQM2 MA12 VOLDET VSS VCC2 VDAT1 VSS VCC2 MD62 VCC3 MD28 MD26 VSS MD54 CKEB VCC3 MD17 VCC2 VSS CS1# VCC3 VSS Note:Signal names have been abbreviated in this figure due to space constraints. = Denotes GND terminal = Denotes PWR terminal (VCC2 = VCC_CORE; VCC3 = VCC_IO)

320 SPGA - Top View

Geode™ GXm

www.national.com 20 Revision 3.1 Signal Definitions (Continued) Geode™ GXm Processor Table 2-4. 320 SPGA Pin Assignments - Sorted by Pin Number Pin No. Signal Name A3 VCC3 A5 AD25 A7 VSS A9 VCC2 A11 AD16 A13 VCC3 A15 STOP# A17 SERR# A19 VSS A21 AD11 A23 AD8 A25 VCC3 A27 AD2 A29 VCC2 A31 VSS A33 TEST0 A35 VCC3 A37 VSS B2 VSS B4 AD27 B6 C/BE3# B8 AD21 B10 AD19 B12 C/BE2# B14 TRDY# B16 LOCK# B18 C/BE1# B20 AD13 B22 AD9 B24 AD6 B26 AD3 B28 SMI# B30 AD1 B32 TEST2 B34 MD33 B36 MD2 C1 VCC3 C3 AD31 C5 AD26 C7 AD23 C9 VCC2 C11 AD18 C13 FRAME# C15 VSS C17 PA R C19 VCC3 C21 AD10 C23 VSS C25 AD4 C27 AD0 C29 VCC2 C31 IRQ13 C33 MD1 C35 MD34 C37 VCC3 D2 AD30 D4 AD29 D6 AD24 D8 AD22 D10 AD20 D12 AD17 D14 IRDY# D16 PERR# D18 AD14 D20 AD12 D22 AD7 D24 INTR D26 TEST1 D28 TEST3 D30 MD0 D32 MD32 D34 MD3 D36 MD35 E1 REQ0# E3 REQ2# E5 AD28 E7 VSS E9 VCC2 E11 VCC2 E13 VSS E15 DEVSEL# E17 AD15 E19 VSS E21 C/BE0# E23 AD5 E25 VSS E27 VCC2 E29 VCC2 E31 VSS E33 MD4 E35 MD36 E37 TDN F2 GNT0# F4 TDI F34 MD5 F36 TDP Pin No. Signal Name G1 VSS G3 CLKMODE2 G5 VSS G33 VSS G35 MD37 G37 VSS H2 GNT2# H4 SUSP A# H34 MD6 H36 MD38 J1 TDO J3 VSS J5 TEST J33 VCC2 J35 VSS J37 MD7 K2 REQ1# K4 GNT1# K34 MD39 K36 MD8 L1 VCC2 L3 VCC2 L5 VCC2 L33 VCC2 L35 VCC2 L37 VCC2 M2 RESET M4 SUSP# M34 MD40 M36 MD9 N1 VCC3 N3 TMS N5 VSS N33 VSS N35 MD41 N37 VCC3 P2 FP_VSYNC P4 TCLK P34 MD10 P36 MD42 Q1 SERIALP Q3 VSS Q5 NC Q33 MD11 Q35 VSS Q37 MD43 R2 CLKMODE1 R4 FP_HSYNC Pin No. Signal Name R34 MD44 R36 MD12 S1 CLKMODE0 S3 VID_VAL S5 PIXEL0 S33 MD14 S35 MD13 S37 MD45 T2 PIXEL1 T4 PIXEL2 T34 MD15 T36 MD46 U1 VSS U3 VCC3 U5 VSS U33 VSS U35 VCC3 U37 VSS V2 PIXEL3 V4 VID_CLK V34 SYSCLK V36 MD47 W1 PIXEL6 W3 PIXEL5 W5 PIXEL4 W33 WEA# W35 WEB# W37 CASA# X2 NC X4 PIXEL9 X34 DQM0 X36 CASB# Y1 PIXEL8 Y3 VSS Y5 PIXEL7 Y33 DQM1 Y35 VSS Y37 DQM4 Z2 NC Z4 PIXEL10 Z34 CS2# Z36 DQM5 AA1 VCC3 AA3 PIXEL11 AA5 VSS AA33 VSS AA35 CS0# AA37 VCC3 Pin No. Signal Name AB2 PIXEL12 AB4 PIXEL13 AB34 RASB# AB36 RASA# AC1 VCC2 AC3 VCC2 AC5 VCC2 AC33 VCC2 AC35 VCC2 AC37 VCC2 AD2 CRT_HSYNC AD4 DCLK AD34 MA2 AD36 MA0 AE1 PIXEL14 AE3 VSS AE5 VCC2 AE33 VCC2 AE35 VSS AE37 MA1 AF2 PIXEL15 AF4 PIXEL16 AF34 MA4 AF36 MA3 AG1 VSS AG3 PIXEL17 AG5 VSS AG33 VSS AG35 MA5 AG37 VSS AH2 CRT_VSYNC AH4 VID_DATA6 AH32 MA10 AH34 MA8 AH36 MA6 AJ1 PCLK AJ3 FTL# AJ5 VID_DATA5 AJ7 VSS AJ9 VCC2 AJ11 MD31 AJ13 VSS AJ15 MD60 AJ17 MD57 AJ19 VSS AJ21 MD22 AJ23 MD52 AJ25 VSS Pin No. Signal Name

Revision 3.1 21 www.national.com Signal Definitions (Continued) Geode™ GXm Processor AJ27 VCC2 AJ29 VCC2 AJ31 VSS AJ33 BA1 AJ35 MA9 AJ37 MA7 AK2 VID_RDY AK4 VSS AK6 VID_DATA0 AK8 SDCLK0 AK10 SDCLK2 AK12 SDCLK_IN AK14 MD29 AK16 MD27 AK18 MD56 AK20 MD55 AK22 MD21 Pin No. Signal Name AK24 MD20 AK26 MD50 AK28 MD16 AK30 DQM3 AK32 CS3# AK34 VSS AK36 BA0 AL1 VCC2 AL3 VID_DATA4 AL5 VID_DATA2 AL7 SDCLK1 AL9 VCC2 AL11 RW_CLK AL13 SDCLK_OUT AL15 VSS AL17 MD58 AL19 VCC3 Pin No. Signal Name AL21 MD23 AL23 VSS AL25 MD19 AL27 MD49 AL29 VCC2 AL31 DQM6 AL33 CKEA AL35 MA11 AL37 VCC3 AM2 VID_DATA7 AM4 VID_DATA3 AM6 ENA_DISP AM8 SDCLK3 AM10 MD63 AM12 MD30 AM14 MD61 AM16 MD59 Pin No. Signal Name AM18 MD25 AM20 MD24 AM22 MD53 AM24 MD51 AM26 MD18 AM28 MD48 AM30 DQM7 AM32 DQM2 AM34 MA12 AM36 VOLDET AN1 VSS AN3 VCC2 AN5 VID_DATA1 AN7 VSS AN9 VCC2 AN11 MD62 AN13 VCC3 Pin No. Signal Name AN15 MD28 AN17 MD26 AN19 VSS AN21 MD54 AN23 CKEB AN25 VCC3 AN27 MD17 AN29 VCC2 AN31 VSS AN33 CS1# AN35 VCC3 AN37 VSS Pin No. Signal Name Table 2-4. 320 SPGA Pin Assignments - Sorted by Pin Number (Continued)

www.national.com 22 Revision 3.1 Signal Definitions (Continued) Geode™ GXm Processor Table 2-5. 320 SPGA Pin Assignments - Sorted Alphabetically by Signal Name Signal Name Type Pin. No. AD0 I/O C27 AD1 I/O B30 AD2 I/O A27 AD3 I/O B26 AD4 I/O C25 AD5 I/O E23 AD6 I/O B24 AD7 I/O D22 AD8 I/O A23 AD9 I/O B22 AD10 I/O C21 AD11 I/O A21 AD12 I/O D20 AD13 I/O B20 AD14 I/O D18 AD15 I/O E17 AD16 I/O A11 AD17 I/O D12 AD18 I/O C11 AD19 I/O B10 AD20 I/O D10 AD21 I/O B8 AD22 I/O D8 AD23 I/O C7 AD24 I/O D6 AD25 I/O A5 AD26 I/O C5 AD27 I/O B4 AD28 I/O E5 AD29 I/O D4 AD30 I/O D2 AD31 I/O C3 BA0 O AK36 BA1 O AJ33 CASA# O W37 CASB# O X36 C/BE0# I/O E21 C/BE1# I/O B18 C/BE2# I/O B12 C/BE3# I/O B6 CKEA O AL33 CKEB O AN23 CLKMODE0 I S1 CLKMODE1 I R2 CLKMODE2 I G3 CRT_HSYNC O AD2 CRT_VSYNC O AH2 CS0# O AA35 CS1# O AN33 CS2# O Z34 CS3# O AK32 DCLK I AD4 DEVSEL# s/t/s E15 (PU) DQM0 O X34 DQM1 O Y33 DQM2 O AM32 DQM3 O AK30 DQM4 O Y37 DQM5 O Z36 DQM6 O AL31 DQM7 O AM30 ENA_DISP O AM6 FLT# I AJ3 FP_HSYNC O R4 FP_VSYNC O P2 FRAME# s/t/s C13 (PU) GNT0# O F2 GNT1# O K4 GNT2# O H2 INTR I D24 IRDY# s/t/s D14 (PU) IRQ13 O C31 LOCK# s/t/s B16 (PU) MA0 O AD36 MA1 O AE37 MA2 O AD34 MA3 O AF36 MA4 O AF34 MA5 O AG35 MA6 O AH36 MA7 O AJ37 MA8 O AH34 MA9 O AJ35 MA10 O AH32 MA11 O AL35 MA12 O AM34 MD0 I/O D30 MD1 I/O C33 MD2 I/O B36 MD3 I/O D34 MD4 I/O E33 MD5 I/O F34 MD6 I/O H34 MD7 I/O J37 MD8 I/O K36 MD9 I/O M36 MD10 I/O P34 MD11 I/O Q33 MD12 I/O R36 MD13 I/O S35 MD14 I/O S33 MD15 I/O T34 MD16 I/O AK28 MD17 I/O AN27 MD18 I/O AM26 MD19 I/O AL25 Signal Name Type Pin. No. MD20 I/O AK24 MD21 I/O AK22 MD22 I/O AJ21 MD23 I/O AL21 MD24 I/O AM20 MD25 I/O AM18 MD26 I/O AN17 MD27 I/O AK16 MD28 I/O AN15 MD29 I/O AK14 MD30 I/O AM12 MD31 I/O AJ11 MD32 I/O D32 MD33 I/O B34 MD34 I/O C35 MD35 I/O D36 MD36 I/O E35 MD37 I/O G35 MD38 I/O H36 MD39 I/O K34 MD40 I/O M34 MD41 I/O N35 MD42 I/O P36 MD43 I/O Q37 MD44 I/O R34 MD45 I/O S37 MD46 I/O T36 MD47 I/O V36 MD48 I/O AM28 MD49 I/O AL27 MD50 I/O AK26 MD51 I/O AM24 MD52 I/O AJ23 MD53 I/O AM22 MD54 I/O AN21 MD55 I/O AK20 MD56 I/O AK18 MD57 I/O AJ17 MD58 I/O AL17 MD59 I/O AM16 MD60 I/O AJ15 MD61 I/O AM14 MD62 I/O AN11 MD63 I/O AM10 NC Q5 NC X2 NC Z2 PAR I/O C17 PCLK O AJ1 PERR# s/t/s D16 (PU) PIXEL0 O S5 PIXEL1 O T2 PIXEL2 O T4 Signal Name Type Pin. No. PIXEL3 O V2 PIXEL4 O W5 PIXEL5 O W3 PIXEL6 O W1 PIXEL7 O Y5 PIXEL8 O Y1 PIXEL9 O X4 PIXEL10 O Z4 PIXEL11 O AA3 PIXEL12 O AB2 PIXEL13 O AB4 PIXEL14 O AE1 PIXEL15 O AF2 PIXEL16 O AF4 PIXEL17 O AG3 RASA# O AB36 RASB# O AB34 REQ0# I E1 (PU) REQ1# I K2 (PU) REQ2# I E3 (PU) RESET I M2 RW_CLK O AL11 SDCLK_IN I AK12 SDCLK_OUT O AL13 SDCLK0 O AK8 SDCLK1 O AL7 SDCLK2 O AK10 SDCLK3 O AM8 SERIALP O Q1 SERR# OD A17 (PU) SMI# I B28 STOP# s/t/s A15 (PU) SUSP# I M4 (PU) SUSPA# O H4 SYSCLK I V34 TCLK I P4 (PU) TDI I F4 (PU) TDN O E37 TDO O J1 TDP O F36 TEST I J5 (PD) TEST0 O A33 TEST1 O D26 TEST2 O B32 TEST3 O D28 TMS I N3 (PU) TRDY# s/t/s B14 (PU) VCC2 PWR A9 VCC2 PWR A29 VCC2 PWR C9 VCC2 PWR C29 VCC2 PWR E9 VCC2 PWR E11 Signal Name Type Pin. No.

Revision 3.1 23 www.national.com Signal Definitions (Continued) Geode™ GXm Processor Note: PU/PD indicates pin is internally connected to a 20-kohm pull-up/ down resistor VCC2 PWR E27 VCC2 PWR E29 VCC2 PWR J33 VCC2 PWR L1 VCC2 PWR L3 VCC2 PWR L5 VCC2 PWR L33 VCC2 PWR L35 VCC2 PWR L37 VCC2 PWR AC1 VCC2 PWR AC3 VCC2 PWR AC5 VCC2 PWR AC33 VCC2 PWR AC35 VCC2 PWR AC37 VCC2 PWR AE5 VCC2 PWR AE33 VCC2 PWR AJ9 VCC2 PWR AJ27 VCC2 PWR AJ29 VCC2 PWR AL1 VCC2 PWR AL9 VCC2 PWR AL29 VCC2 PWR AN3 VCC2 PWR AN9 VCC2 PWR AN29 VCC3 PWR A3 VCC3 PWR A13 VCC3 PWR A25 Signal Name Type Pin. No. VCC3 PWR A35 VCC3 PWR C1 VCC3 PWR C19 VCC3 PWR C37 VCC3 PWR N1 VCC3 PWR N37 VCC3 PWR U3 VCC3 PWR U35 VCC3 PWR AA1 VCC3 PWR AA37 VCC3 PWR AL19 VCC3 PWR AL37 VCC3 PWR AN13 VCC3 PWR AN25 VCC3 PWR AN35 VID_CLK O V4 VID_DATA0 O AK6 VID_DATA1 O AN5 VID_DATA2 O AL5 VID_DATA3 O AM4 VID_DATA4 O AL3 VID_DATA5 O AJ5 VID_DATA6 O AH4 VID_DATA7 O AM2 VID_RDY I AK2 VID_VAL O S3 VOLDET O AM36 VSS GND A7 VSS GND A19 Signal Name Type Pin. No. VSS GND A31 VSS GND A37 VSS GND B2 VSS GND C15 VSS GND C23 VSS GND E7 VSS GND E13 VSS GND E19 VSS GND E25 VSS GND E31 VSS GND G1 VSS GND G5 VSS GND G33 VSS GND G37 VSS GND J3 VSS GND J35 VSS GND N5 VSS GND N33 VSS GND Q3 VSS GND Q35 VSS GND U1 VSS GND U5 VSS GND U33 VSS GND U37 VSS GND Y3 VSS GND Y35 VSS GND AA5 VSS GND AA33 VSS GND AE3 Signal Name Type Pin. No. VSS GND AE35 VSS GND AG1 VSS GND AG5 VSS GND AG33 VSS GND AG37 VSS GND AJ7 VSS GND AJ13 VSS GND AJ19 VSS GND AJ25 VSS GND AJ31 VSS GND AK4 VSS GND AK34 VSS GND AL15 VSS GND AL23 VSS GND AN1 VSS GND AN7 VSS GND AN19 VSS GND AN31 VSS GND AN37 WEA# O W33 WEB# O W35 Signal Name Type Pin. No. Table 2-5. 320 SPGA Pin Assignments - Sorted Alphabetically by Signal Name (Continued)

www.national.com 24 Revision 3.1 Signal Definitions (Continued) Geode™ GXm Processor

2.2 SIGNAL DESCRIPTIONS

2.2.1 System Interface Signals

Pin No. SPGA Pin No. Type Description SYSCLK P26 V34 I System Clock System Clock runs synchronously with the PCI bus. The internal clock of the GXm processor is generated by an internal PLL which multiplies the SYSCLK input and can run up to eight times faster. The SYSCLK to core clock multiplier is configured using the CLKMOD[2:0] inputs. The SYSCLK input is a fixed frequency which can only be stopped or varied when the GXm processor is in a full 3V Sus- pend. (Section 6.4 “3-Volt Suspend Mode” on page 174 for details regarding this mode.) CLKMODE[2:0] M1, L1, G3, R2, I Clock Mode T h e s es i g n a l sa r eu s e dt os e tt h ec o r ec l o c km u l t i p l i e r .T h eP C I clock "SYSCLK" is multiplied by the value programmed by CLK- MODE[2:0] to generate the GXm processor’s core clock. CLKMODE2 is valid only for GXm processor revision 4.0 and up. The value read from DIR1 (Device ID Register 1, refer to page 51) affects the definition of the CLKMODE pins. If DIR1 = 30h-33h then CLKMODE[1:0]: 00 = SYSCLK multiplied by 4 (Test mode only) 01 = SYSCLK multiplied by 6 10 = SYSCLK multiplied by 7 11 = SYSCLK multiplied by 5 If DIR1 = 34h-4Fh then CLKMODE[1:0]: 00 = SYSCLK multiplied by 4 (Test mode only) 01 = SYSCLK multiplied by 6 10 = SYSCLK multiplied by 7 11 = SYSCLK multiplied by 8 If DIR1 > or = 50h then CLKMODE[2:0]: 000 = SYSCLK multiplied by 4 (T est mode only) 001 = SYSCLK multiplied by 10 010 = SYSCLK multiplied by 9 011 = SYSCLK multiplied by 5 100 = SYSCLK multiplied by 4 101 = SYSCLK multiplied by 6 110 = SYSCLK multiplied by 7 111 = SYSCLK multiplied by 8 RESET J3 M2 I Reset RESET aborts all operations in progress and places the GXm processor into a reset state. RESET forces the CPU and peripheral functions to begin executing at a known state. All data in the on-chip cache is invalidated. RESET is an asynchronous input but must meet specified setup and hold times to guarantee recognition at a particular clock edge. This input is typically generated during the Power-On- Reset sequence. Note: Warm Reset does not require an input on the GXm pro- cessor since the function is virtualized using SMM.

Revision 3.1 25 www.national.com Signal Definitions (Continued) Geode™ GXm Processor INTR B18 D24 I (Maskable) Interrupt Request INTR is a level-sensitive input that causes the GXm processor to Suspend execution of the current instruction stream and begin execution of an interrupt service routine. The INTR input can be masked through the Flags Register IF bit. (See Table 3-4 on page 43 for bit definitions.) IRQ13 C22 C31 O Interrupt Request Level 13 IRQ13 is asserted if an on-chip floating point error occurs. When a floating point error occurs, the GXm processor asserts the IRQ13 pin. The floating point interrupt handler then performs an OUT instruction to I/O address F0h or F1h. The GXm proces- sor accepts either of these cycles and clears the IRQ13 pin. Refer to Section 3.4.1 “I/O Address Space” on page 60 for fur- ther information on IN/OUT instructions. SMI# C19 B28 I System Management Interrupt SMI# is a level-sensitive interrupt. SMI# puts the GXm processor into System Management Mode (SMM). SUSP# H2 (PU) (PU) I Suspend Request This signal is used to request that the GXm processor enter Sus- pend mode. After recognition of an active SUSP# input, the pro- cessor completes execution of the current instruction, any pending decoded instructions and associated bus cycles. SUSP# is ignored following RESET# and is enabled by setting the SUSP bit in CCR2. (See Table 16 on page 44 for CCR2 bit definitions.) Since the GXm processor includes system logic functions as well as the CPU core, there are special modes designed to support the different power management states associated with APM, ACPI, and portable designs. The part can be configured to stop only the CPU core clocks, or all clocks. When all clocks are stopped, the external clock can also be stopped. (See Section 6.0 “Power Management” on page 174 for more details regarding power management states.) This pin is internally connected to a 20-kohm pull-up resistor. S U S P #i sp u l l e du pw h e nn o ta c t i v e . SUSPA# E2 H4 O Suspend Acknowledge Suspend Acknowledge indicates that the GXm processor has entered low-power Suspend mode as a result of SUSP# asser- tion or execution of a HALT instruction. SUSPA# is enabled by setting the SUSP bit in CCR2. (See Table 16 on page 44 for CCR2 bit definitions.) The SYSCLK input may be stopped after SUSPA# has been asserted to further reduce power consumption if the system is configured for 3V Suspend mode. (Section 6.4 “3-Volt Suspend Mode” on page 174 for details regarding this mode.) SERIALP L3 Q1 O Serial Packet Serial Packet is the single wire serial-transmission signal to the CS5530 chip. The clock used for this interface is the PCI clock (SYSCLK). This interface carries packets of miscellaneous infor- mation to the chipset to be used by the VSA software handlers.

2.2.1 System Interface Signals (Continued)

Pin No. SPGA Pin No. Type Description

www.national.com 26 Revision 3.1 Signal Definitions (Continued) Geode™ GXm Processor

2.2.2 PCI Interface Signals

Pin No. SPGA Pin No Type Description AD[31:0] Refer toT able 2-3 Refer toT able 2-5 I/O Multiplexed Address and Data Addresses and data are multiplexed on the same PCI pins. A bus transaction consists of an address phase in the cycle in which FRAME# is asserted followed by one or more data phases. Dur- ing the address phase, AD[31:0] contain a physical 32-bit address. For I/O, this is a byte address, for configuration and memory it is a DWORD address. During data phases, AD[7:0] contain the least significant byte (LSB) and AD[31:24] contain the most significant byte (MSB). Write data is stable and valid when IRDY# is asserted and read data is stable and valid when TRDY# is asserted. Data is transferred during those SYSCLKS where both IRDY# and TRDY# are asserted. C/BE[3:0]# D5, B8, C13, A15 B6, B12, B18, E21 I/O Multiplexed Command and Byte Enables Bus command and byte enables are multiplexed on the same PCI pins. During the address phase of a transaction when FRAME# is active, C/BE[3:0]# define the bus command. During the data phase C/BE[3:0]# are used as byte enables. The byte enables are valid for the entire data phase and determine which byte lanes carry meaningful data. C/BE0# applies to byte 0 (LSB) and C/BE3# applies to byte 3 (MSB). The command encoding and types are listed below. 0000 = Interrupt Acknowledge 0001 = Special Cycle 0010 = I/O Read 0011 = I/O Write 0100 = Reserved 0101 = Reserved 0110 = Memory Read 0111 = Memory Write 1000 = Reserved 1001 = Reserved 1010 = Configuration Read 1011 = Configuration Write 1100 = Memory Read Multiple 1101 = Dual Address Cycle (Reserved) 1110 = Memory Read Line 1111 = Memory Write and Invalidate PAR B12 C17 I/O Parity Parity generation is required by all PCI agents: the master drives PAR for address and write-data phases, the target drives PAR for read-data phases. Parity is even across AD[31:0] and C/BE[3:0]#. For address phases, PAR is stable and valid one SYSCLK after the address phase. It has the same timing as AD[31:0] but delayed by one SYSCLK. For data phases, PAR is stable and valid one SYSCLK after either IRDY# is asserted on a write transaction or after TRDY# is asserted on a read transaction. Once PAR is valid, it remains valid until one SYSCLK after the completion of the data phase. (Also see PERR#.)

Revision 3.1 27 www.national.com Signal Definitions (Continued) Geode™ GXm Processor FRAME# A8 (PU) C13 (PU) s/t/s Frame Cycle Frame is driven by the current master to indicate the beginning and duration of an access. FRAME# is asserted to indicate a bus transaction is beginning. While FRAME# is asserted, data transfers continue. When FRAME# is deasserted, the transaction is in the final data phase. This pin is internally connected to a 20-kohm pull-up resistor. IRDY# C9 (PU) D14 (PU) s/t/s Initiator Ready Initiator Ready is asserted to indicate that the bus master is able to complete the current data phase of the transaction. IRDY# is used in conjunction with TRDY#. A data phase is completed on any SYSCLK in which both IRDY# and TRDY# are sampled asserted. During a write, IRDY# indicates valid data is present on AD[31:0]. During a read, it indicates the master is prepared to accept data. Wait cycles are inserted until both IRDY# and TRDY# are asserted together. This pin is internally connected to a 20-kohm pull-up resistor. TRDY# B9 (PU) B14 (PU) s/t/s Target Ready TRDY# is asserted to indicate that the target agent is able to complete the current data phase of the transaction. TRDY# is used in conjunction with IRDY#. A data phase is complete on any SYSCLK in which both TRDY# and IRDY# are sampled asserted. During a read, TRDY# indicates that valid data is present on AD[31:0]. During a write, it indicates the target is pre- pared to accept data. Wait cycles are inserted until both IRDY# and TRDY# are asserted together. This pin is internally connected to a 20-kohm pull-up resistor. STOP# C11 (PU) A15 (PU) s/t/s Target Stop STOP# is asserted to indicate that the current target is request- ing the master to stop the current transaction. This signal is used with DEVSEL# to indicate retry, disconnect or target abort. If STOP# is sampled active while a master, FRAME# will be deas- serted and the cycle stopped within three SYSCLK cycles. As an input, STOP# can be asserted in the following cases. 1) If a PCI master tries to access memory that has been locked by another master. This condition is detected if FRAME# and LOCK# are asserted during an address phase. 2) STOP# will also be asserted if the PCI write buffers are full or if a previously buffered cycle has not completed. 3) Finally, STOP# can be asserted on read cycles that cross cache line boundaries. This is conditional based upon the programming of bit 1 in PCI Control Function 2 Register. (See T able 4-37 on page 156 for programming details.) This pin is internally connected to a 20-kohm pull-up resistor.

2.2.2 PCI Interface Signals (Continued)

Pin No. SPGA Pin No Type Description

www.national.com 28 Revision 3.1 Signal Definitions (Continued) Geode™ GXm Processor LOCK# B11 (PU) B16 (PU) s/t/s Lock Operation LOCK# indicates an atomic operation that may require multiple transactions to complete. When LOCK# is asserted, nonexclu- sive transactions may proceed to an address that is not currently locked (at least 16 bytes must be locked). A grant to start a trans- action on PCI does not guarantee control of LOCK#. Control of LOCK# is obtained under it own protocol in conjunction with GNT#. It is possible for different agents to use PCI while a single master retains ownership of LOCK#. The arbiter can implement a complete system lock. In this mode, if LOCK# is active, no other master can gain access to the system until the LOCK# is deasserted. This pin is internally connected to a 20-kohm pull-up resistor. DEVSEL# A9 (PU) E15 (PU) s/t/s Device Select DEVSEL# indicates that the driving device has decoded its address as the target of the current access. As an input, DEVSEL# indicates whether any device on the bus has been selected. DEVSEL# will also be driven by any agent that has the ability to accept cycles on a subtractive decode basis. As a mas- ter, if no DEVSEL# is detected within and up to the subtractive decode clock, a master abort cycle will result expect for special cycles which do not expect a DEVSEL# returned. This pin is internally connected to a 20-kohm pull-up resistor. PERR# A11 (PU) D16 (PU) s/t/s Parity Error PERR# is used for reporting of data parity errors during all PCI transactions except a Special Cycle. The PERR# line is driven two SYSCLKs after the data in which the error was detected. This is one SYSCLK after the PAR that is attached to the data. The minimum duration of PERR# is one SYSCLK for each data phase in which a data parity error is detected. PERR# must be driven high for one SYSCLK before being in TRI-STATE mode. A target asserts PERR# on write cycles if it has claimed the cycle with DEVSEL#. The master asserts PERR# on read cycles. This pin is internally connected to a 20-kohm pull-up resistor. SERR# C12 (PU) A17 (PU) OD System Error System Error may be asserted by any agent for reporting errors other than PCI parity. The intent is to have the PCI central agent assert NMI to the processor. When the Parity Enable bit is set in the Memory Controller Configuration register, SERR# will be asserted upon detecting a parity error on read operations from DRAM. REQ[2:0]# D3, H3, (PU) E3, K2, (PU) I Request Lines Request indicates to the arbiter that an agent desires use of the bus. Each master has its own REQ# line. REQ# priorities are based on the arbitration scheme chosen. Each of these pins are internally connected to a 20-kohm pull-up resistor. Pin No. SPGA Pin No Type Description

Revision 3.1 29 www.national.com Signal Definitions (Continued) Geode™ GXm Processor GNT[2:0]# E1, F2, H2, K4, O Grant Lines Grant indicates to the requesting master that it has been granted access to the bus. Each master has its own GNT# line. GNT# can be pulled away at any time a higher REQ# is received or if the master does not begin a cycle within a minimum period of time (16 SYSCLKs). Pin No. SPGA Pin No Type Description

2.2.3 Memory Controller Interface Signals

Pin No. SPGA Pin No. Type Description Note: The memory controller interface supports two types of memory configurations: SDRAM modules on the sys- tem board and JEDEC DIMM connectors. Refer to Section 4.3 “Memory Controller” on page 103 for detailed information regarding signal connections. MD[63:0] Refer toT able 2-3 Refer toT able 2-5 I/O Memory Data Bus The data bus lines driven to/from system memory. MA[12:0] Refer toT able 2-3 Refer toT able 2-5 O Memory Address Bus The multiplexed row/column address lines driven to the system memory. Supports 256 Mbit SDRAM. BA[1:0] AD26, AD25 AJ33, AK36 O Bank Address Bits These bits are used to select the component bank within the SDRAM. CS[3:0]# AE23, V25, AD23, V26 AK32, Z34, AN33, AA35 O Chip Selects The chip selects are used to select the module bank within the system memory. Each chip select corresponds to a specific mod- ule bank. If CS# is high, the bank(s) do not respond to RAS#, CAS#, WE# until the bank is selected again. RASA#, RASB# W24, W25 AB36, AB34 O Row Address Strobe RAS#, CAS#, WE# and CKE are encoded to support the differ- ent SDRAM commands. RASA# is used with CS[1:0]#. RASB# is used with CS[3:2]#. CASA#, CASB# P25, R26 W37, X36 O Column Address Strobe RAS#, CAS#, WE# and CKE are encoded to support the differ- ent SDRAM commands. CASA# is used with CS[1:0]#. CASB# is used with CS[3:2]#. WEA#, WEB# R25, R24 W33, W35 O Write Enable RAS#, CAS#, WE# and CKE are encoded to support the differ- ent SDRAM commands. WEA# is used with CS[1:0]#. WEB# is used with CS[3:2]#.

www.national.com 30 Revision 3.1 Signal Definitions (Continued) Geode™ GXm Processor DQM[7:0] Refer toT able 2-3 Refer toT able 2-5 O Data Mask Control Bits During memory read cycles, these outputs control whether the SDRAM output buffers are driven on the MD bus or not. All DQM signals are asserted during read cycles. During memory write cycles, these outputs control whether or not MD data will be written into the SDRAM. DQM[7:0] connect directly to the DQM7-0 pins of each connec- tor. CKEA, CKEB AF24, AD16 AL33, AN23 O Clock Enable For normal operation CKE is held high. CKE goes low during Suspend. SDCLK[3:0] AE4, AF5, AE5, AF4 AM8, AK10, AL7, AK8 O SDRAM Clocks The SDRAM samples all the control, address, and data using these clocks. SDCLK_IN AE8 AK12 I SDRAM Clock Input The GXm processor samples the memory read data on this clock. Works in conjunction with the SDCLK_OUT signal. SDCLK_OUT AF8 AL13 O SDRAM Clock Output This output is routed back to SDCLK_IN. The board designer should vary the length of the board trace to control skew between SDCLK_IN and SDCLK.

2.2.3 Memory Controller Interface Signals (Continued)

Pin No. SPGA Pin No. Type Description

2.2.4 Video Interface Signals

PCLK AC1 AJ1 O Pixel Port Clock Pixel Port Clock represents the pixel dotclock or a 2x multiple of the dotclock for some 16-bit-per-pixel modes. It determines the data transfer rate from the GXm processor to the CS5530. VID_CLK P1 V4 O Video Clock Video Clock represents the video port clock to the CS5530. This pin is only used if the Video Port is enabled. DCLK AB1 AD4 I DOT Clock The DCLK input is driven from the CS5530 and represents the pixel dot clock. In some cases, such as when displaying 16 BPP data with an eight-bit-graphics pixel port, this clock will actually be a 2x multiple of the dotclock. CRT_HSYNC W2 AD2 O CRT Horizontal Sync CRT Horizontal Sync establishes the line rate and horizontal retrace interval for an attached CRT. The polarity is programma- ble and depends on the display mode.

Revision 3.1 31 www.national.com Signal Definitions (Continued) Geode™ GXm Processor CRT_VSYNC AA3 AH2 O CRT Vertical Sync CRT Vertical Sync establishes the screen refresh rate and verti- cal retrace interval for an attached CRT. The polarity is program- mable and depends on the display mode. FP_HSYNC L2 R4 O Flat Panel Horizontal Sync Flat Panel Horizontal Sync establishes the line rate and horizon- tal retrace interval for a TFT display. Polarity is programmable and depends on the display mode. This signal is an input to the CS5530. The CS5530 re-drives this signal to the flat panel. If no flat panel is used in the system, this signal does not need to be connected. FP_VSYNC J1 P2 O Flat Panel Vertical Sync Flat Panel Vertical Sync establishes the screen refresh rate and vertical retrace interval for a TFT display. Polarity is programma- ble and depends on the display mode. This signal is an input to the CS5530. The CS5530 re-drives this signal to the flat panel. If no flat panel is used in the system, this signal does not need to be connected. ENA_DISP AD5 AM6 O Display Enable Display Enable indicates the active display portion of a scan line to the CS5530. In a CS5530-based system, this signal is required to be con- nected even if there is no TFT panel in the system. VID_RDY AD1 AK2 I Video Ready This input signal indicates that the video FIFO in the CS5530 is ready to receive more data. VID_VAL M2 S3 O Video Valid VID_VAL qualifies valid video data to the CS5530. VID_DATA[7:0] Refer to Refer toT able 2-5 O Video Data Bus When the Video Port is enabled, this bus drives Video (Y -U-V) data synchronous to the VID_CLK output. PIXEL[17:0] Refer toT able 2-3 Refer toT able 2-5 O Graphics Pixel Data Bus This bus drives graphics pixel data synchronous to the PCLK output.

2.2.4 Video Interface Signals (Continued)

www.national.com 32 Revision 3.1 Signal Definitions (Continued) Geode™ GXm Processor

2.2.5 Power, Ground, and No Connect Signals

Pin No. SPGA Pin No. Type Description VOLDET AC5 AM36 O Voltage Detect In early schematic revisions this pin was identified as VOLDET . However, in the production version this pin is a "no connect" and should be left disconnected. VSS Refer toT able 2-3 (T otal of 71) Refer toT able 2-5 (T otal of 50) GND Ground Connection VCC2 Refer toT able 2-3 (T otal of 32) Refer toT able 2-5 (T otal of 32) PWR 2.9V (nominal) Core Power Connection VCC3 Refer toT able 2-3 (T otal of 32) Refer toT able 2-5 (T otal of 18) PWR 3.3V (nominal) I/O Power Connection NC -- Q5, X2, No Connection A line designated as NC should be left disconnected.

2.2.6 Internal Test and Measurement Signals

Pin No. SPGA Pin No. Type Description FLT# AC2 AJ3 I Float Float Outputs force the GXm processor to float all outputs in the high-impedance state and to enter a power-down state. RW_CLK AE6 AL11 O Raw Clock This output is the GXm processor clock. This debug signal can be used to verify clock operation. TEST[3:0] B22, A23, B21, C21 D28, B32, D26, A33 O SDRAM Test Outputs These outputs are used for internal debug only. TCLK J2 (PU) (PU) I Test Clock JTAG test clock. This pin is internally connected to a 20-kohm pull-up resistor. TDI D2 (PU) (PU) I Test Data Input JTAG serial test-data input. This pin is internally connected to a 20-kohm pull-up resistor. TDO F1 J1 O Test Data Output JTAG serial test-data output.

Revision 3.1 33 www.national.com Signal Definitions (Continued) Geode™ GXm Processor TMS H1 (PU) (PU) I Test Mode Select JTAG test-mode select. This pin is internally connected to a 20-kohm pull-up resistor. TEST F3 (PD) (PD) I Test Test-mode input. This pin is internally connected to a 20-kohm pull-down resistor. TDP E24 F36 O Thermal Diode Positive TDP is the positive terminal of the thermal diode on the die. The diode is used to do thermal characterization of the device in a system. This signal works in conjunction with TDN. TDN D26 E37 O Thermal Diode Negative TDN is the negative terminal of the thermal diode on the die. The diode is used to do thermal characterization of the device in a system. This signal works in conjunction with TDP . Pin No. SPGA Pin No. Type Description

www.national.com 34 Revision 3.1 Signal Definitions (Continued) Geode™ GXm Processor

2.3 SUBSYSTEM SIGNAL CONNECTIONS

As previously stated, the GXm Integrated Subsystem with MMX support consists of two chips: the GXm Processor and the CS5530 I/O companion. Figure 2-4 shows the signal connections between the processor and the I/O companion chip. Figure 2-4. Subsystem Signal Connections SYSCLK SERIALP IRQ13 SMI# PCLK CRT_HSYNC CRT_VSYNC PIXEL[17:0] FP_HSYNC FP_VSYNC ENA_DISP VID_VAL VID_CLK VID_DATA[7:0] VID_RDY INTR SUSP# SUSPA# AD[31:0] C/BE[3:0]# PAR FRAME# IRDY# TRDY# STOP# LOCK# DEVSEL# PERR# SERR# REQ0# GX_CLK PSERIAL IRQ13 SMI# PCLK HSYNC VSYNC PIXEL[23:0] FP_HSYNC FP_VSYNC FP_ENA_DISP VID_VAL VID_CLK VID_DATA[7:0] VID_RDY CPU_RST INTR SUSP# SUSPA# AD[31:0] C/BE[3:0]# PAR FRAME# IRDY# TRDY# STOP# LOCK# DEVSEL# PERR# SERR# REQ# GNT#GNT0# Geode™ CS5530 I/O Companion Exclusive Interconnect Signals (Do not connect to any other device) Nonexclusive Interconnect Signals (May also connect to other circuitry) Not needed if CRT only (no TFT) (Note) Note: Refer to Figure 2-5 for interconnection of these lines. RESET DCLK DCLK Geode™ GXm Processor

Revision 3.1 35 www.national.com Signal Definitions (Continued) Geode™ GXm Processor Figure 2-5. PIXEL Signal Connections PIXEL17 PIXEL16 PIXEL15 PIXEL14 PIXEL13 PIXEL12 PIXEL11 PIXEL10 PIXEL9 PIXEL8 PIXEL7 PIXEL6 PIXEL5 PIXEL4 PIXEL3 PIXEL2 PIXEL1 Geode™ CS5530 I/O Companion PIXEL0 PIXEL23 PIXEL22 PIXEL21 PIXEL20 PIXEL19 PIXEL18 PIXEL17 PIXEL16 PIXEL15 PIXEL14 PIXEL13 PIXEL12 PIXEL11 PIXEL10 PIXEL9 PIXEL8 PIXEL7 PIXEL6 PIXEL5 PIXEL4 PIXEL3 PIXEL2 PIXEL1 PIXEL0 Geode™ GXm Processor

www.national.com 36 Revision 3.1 Signal Definitions (Continued) Geode™ GXm Processor

2.4 POWER PLANES

Figure 2-6 shows layout recommendations for splitting the power plane between VCC2 (core: 2.9V) and VCC3 (I/O: 3.3V) volts in the BGA package. The illustration assumes there is one power plane, and no components on the back of the board Figure 2-7 shows layout recommendations for splitting the power plane between VCC2 (core: 2.9V) and VCC3 (I/O: 3.3V) volts for the GXm in the SPGA package. Figure 2-6. BGA Recommended Split Power Plane and Decoupling 1 26 A AF AF A 261 = High frequency capacitor = 220µF , low ESR capacitor = 3.3V connection = 2.9V connection Geode™ GXm 2.9V Plane (VCC2) 3.3V Plane (VCC3) 3.3V Plane (VCC3) 3.3V Plane (VCC3) 2.9V Plane (VCC2) Legend Processor

Revision 3.1 37 www.national.com Signal Definitions (Continued) Geode™ GXm Processor Figure 2-7. SPGA Recommended Split Power Plane and Decoupling 1 37 A AN A AN 1 37 2.9V Plane (VCC2) 3.3V Plane (VCC3) 3.3V Plane (VCC3) 3.3V Plane (VCC3) 3.3V Plane (VCC3) 2.9V Plane (VCC2) To 2.9V Regulator Note: Where signals cross plane splits, it is recommended to include AC decoupling between planes with 47pF capacitors. = High frequency capacitor = 220 µF , low ESR capacitor = 3.3V connection = 2.9V connection Legend Geode™ GXm Processor

www.national.com 38 Revision 3.1 Geode™ GXm Processor

3.0 Processor Programming

This section describes the internal operations of the Geode GXm processor from a programmer’s point of view. It includes a description of the traditional “core” process- ing and FPU operations. The integrated function registers are described at the end of this chapter. The primary register sets within the processor core include:  Application Register Set  System Register Set  Model Specific Register Set  Floating Point Unit Register Set. The initialization of the major registers within in core are s h o w ni nT a b l e3 - 1 . T h ei n t e g r a t e df u n c t i o ns e t sa r el o c a t e di nm a i nm e m o r y space and include:  Internal Bus Interface Unit Register Set  Graphics Pipeline Register Set  Display Controller Register Set  Memory Controller Register Set  Power Management Register Set

3.1 CORE PROCESSOR INITIALIZATION

The GXm processor is initialized when the RESET signal is asserted. The processor is placed in real mode and the registers listed in Table 3-1 are set to their initialized val- ues. RESET invalidates and disables the CPU cache, and turns off paging. When RESET is asserted, the CPU ter- minates all local bus activity and all internal execution. During the entire time that RESET is asserted, the inter- nal pipeline is flushed and no instruction execution or bus activity occurs. Approximately 150 to 250 external clock cycles after RESET is deasserted, the processor begins executing instructions at the top of physical memory (address loca- tion FFFFFFF0h). The actual time depends on the clock scaling in use. Also, an additional 2 20 clock cycles are needed when self-test is requested. Typically, an intersegment jump is placed at FFFFFFF0h. This instruction will force the processor to begin execution in the lowest 1 MB of address space. T h ef o l l o w i n gt a b l e ,T a b l e3 - 1 ,l i s t st h ec o r er e g i s t e r sa n d illustrates how they are initialized. Table 3-1. Initialized Core Register Controls Register Register Name Initialized Contents Comments EAX Accumulator xxxxxxxxh 00000000h indicates self-test passed. EBX Base xxxxxxxxh ECX Count xxxxxxxxh EDX Data xxxx 04 [DIR0]h DIR0 = Device ID EBP Base Pointer xxxxxxxxh ESI Source Index xxxxxxxxh EDI Destination Index xxxxxxxxh ESP Stack Pointer xxxxxxxxh EFLAGS Extended FLAGS 00000002h See Table 3-4 on page 43 for bit definitions. EIP Instruction Pointer 0000FFF0h ES Extra Segment 0000h Base address set to 00000000h. Limit set to FFFFh. CS Code Segment F000h Base address set to FFFF0000h. Limit set to FFFFh. SS Stack Segment 0000h Base address set to 00000000h. Limit set to FFFFh. DS Data Segment 0000h Base address set to 00000000h. Limit set to FFFFh. FS Extra Segment 0000h Base address set to 00000000h. Limit set to FFFFh. GS Extra Segment 0000h Base address set to 00000000h. Limit set to FFFFh. IDTR Interrupt Descriptor Table Register Base = 0, Limit = 3FFh GDTR Global Descriptor Table Register xxxx xxxxh LDTR Local Descriptor Table Register xxxxh TR Task Register xxxxh CR0 Machine Status Word 60000010h See Table 3-7 on page 49 for bit definitions. CR2 Control Register 2 xxxxxxxxh See Table 3-7 on page 49 for bit definitions. CR3 Control Register 3 xxxxxxxxh See Table 3-7 on page 49 for bit definitions. CR4 Control Register 4 00000000h See Table 3-7 on page 49 for bit definitions. CCR1 Configuration Control 1 00h See Table 3-11 on page 49 for bit definitions. CCR2 Configuration Control 2 00h See Table 3-11 on page 49 for bit definitions. CCR3 Configuration Control 3 00h See Table 3-11 on page 49 for bit definitions. CCR7 Configuration Control 7 00h See Table 3-11 on page 50 for bit definitions. SMAR0 SMM Address 0 00h See Table 3-11 on page 51 for bit definitions.

Revision 3.1 39 www.national.com Processor Programming (Continued) Geode™ GXm Processor

3.2 INSTRUCTION SET OVERVIEW

The GXm processor instruction set can be divided into nine types of operations:  Arithmetic  Bit Manipulation  Shift/Rotate  String Manipulation  Control Transfer D a t a T r a n s f e r  Floating Point  High-Level Language Support  Operating System Support GXm processor instructions operate on as few as zero operands and as many as three operands. An NOP instruction (no operation) is an example of a zero-operand instruction. Two-operand instructions allow the specifica- tion of an explicit source and destination pair as part of the instruction. These two-operand instructions can be divided into ten groups according to operand types:  Register to Register  Register to Memory  Memory to Register  Memory to Memory R e g i s t e r t o I / O  I/O to Register  Memory to I/O  I/O to Memory  Immediate Data to Register  Immediate Data to Memory A no p e r a n dc a nb eh e l di nt h ei n s t r u c t i o ni t s e l f( a si nt h e case of an immediate operand), in one of the processor’s registers or I/O ports, or in memory. An immediate oper- and is fetched as part of the opcode for the instruction. Operand lengths of 8, 16, 32 or 48 bits are supported as well as 64 or 80 bits associated with floating-point instruc- tions. Operand lengths of 8 or 32 bits are generally used when executing code written for 386- or 486-class (32-bit code) processors. Operand lengths of 8 or 16 bits are generally used when executing existing 8086 or 80286 code (16-bit code). The default length of an operand can be overridden by placing one or more instruction prefixes in front of the opcode. For example, the use of prefixes allows a 32-bit operand to be used with 16-bit code or a 16-bit operand to be used with 32-bit code. Section 9.1 “General Instruction Set Format” on page 202 contains the clock count table that lists each instruction in t h eC P Ui n s t r u c t i o ns e t .I n c l u d e di nt h et a b l ea r et h e associated opcodes, execution clock counts, and effects on the EFLAGS register.

3.2.1 Lock Prefix

The LOCK prefix may be placed before certain instruc- tions that read, modify, then write back to memory. The PCI will not be granted access in the middle of locked instructions. The LOCK prefix can be used with the follow- ing instructions only when the result is a write operation to memory.  Bit Test Instructions (BTS, BTR, BTC)  Exchange Instructions (XADD, XCHG, CMPXCHG)  One-Operand Arithmetic and Logical Instructions (DEC, INC, NEG, NOT)  Two-Operand Arithmetic and Logical Instructions (ADC, ADD, AND, OR, SBB, SUB, XOR). An invalid opcode exception is generated if the LOCK pre- fix is used with any other instruction or with one of the instructions above when no write operation to memory occurs (for example, when the destination is a register). SMAR1 SMM Address 1 00h See Table 3-11 on page 50 for bit definitions. SMAR2 SMM Address 2 / SMAR Size 00h See Table 3-11 on page 50 for bit definitions. DIR0 Device Identification 0 4xh Device ID and reads back initial CPU clock-speed set- ting. See Table 3-11 on page 51 for bit definitions. DIR1 Device Identification 1 xxh Stepping and Revision ID (RO). See Table 3-11 on page 51 for bit definitions. DR7 Debug Register 7 00000400h See Table 3-13 on page 53 for bit definitions. Note: x = Undefined value Table 3-1. Initialized Core Register Controls (Continued) Register Register Name Initialized Contents Comments

www.national.com 40 Revision 3.1 Processor Programming (Continued) Geode™ GXm Processor

3.3 REGISTER SETS

The accessible registers in the processor are grouped into three sets: 1) The Application Register Setcontains the registers frequently used by application programmers. Table 3- 2 shows the general purpose, segment, the instruc- tion pointer and the EFLAGS Registers. 2) The System Register Setcontains the registers typi- cally reserved for operating-systems programmers: control, system address, debug, configuration, and test registers. 3) The Model Specific Register (MSR) Setis used to monitor the performance of the processor or a specific component within the processor. The model specific register set has one 64-bit register called the Time Stamp Counter. Each of these register sets are discussed in detail in the subsections that follow. Additional registers to support integrated GXm processor subsystems are described in Section 4.1 “Integrated Functions Programming Interface” on page 92.

3.3.1 Application Register Set

The Application Register Set consists of the registers most often used by the applications programmer. These regis- ters are generally accessible, although some bits in the EFLAGS register are protected. The General Purpose Registercontents are frequently modified by instructions and typically contain arithmetic and logical instruction operands. In real mode,Segment Registers contain the base address for each segment. In protected mode, the seg- ment registers contain segment selectors. The segment selectors provide indexing for tables (located in memory) that contain the base address for each segment, as well as other memory addressing information. The Instruction Pointer Registerpoints to the next instruction that the processor will execute. This register is automatically incremented by the processor as execution progresses. The EFLAGS Register contains control bits used to reflect the status of previously executed instructions. This register also contains control bits that affect the operation of some instructions.

3.3.1.1 General Purpose Registers

The General Purpose Registers are divided into four data registers, two pointer registers, and two index registers as shown in Table 3-2. The Data Registersare used by the applications pro- grammer to manipulate data structures and to hold the results of logical and arithmetic operations. Different por- tions of general data registers can be addressed by using different names. An “E” prefix identifies the complete 32-bit register. An “X” suffix without the “E” prefix identifies the lower 16 bits of the register. The lower two bytes of a data register are addressed with an “H” suffix (identifies the upper byte) or an “L” suffix (iden- tifies the lower byte). These _L and _H portions of the data registers act as independent registers. For example, if the AH register is written to by an instruction, the AL reg- ister bits remain unchanged. The Pointer and Index Registersare listed below. SI or ESI Source Index DI or EDI Destination Index SP or ESP Stack Pointer BP or EBP Base Pointer These registers can be addressed as 16- or 32-bit registers, with the “E” prefix indicating 32 bits. The pointer and index registers can be used as general purpose registers; how- ever, some instructions use a fixed assignment of these registers. For example, repeated string operations always use ESI as the source pointer, EDI as the destination pointer, and ECX as a counter. The instructions that use fixed registers include multiply and divide, I/O access, string operations, stack operations, loop, variable shift and rotate, and translate instructions. The GXm processor implements a stack using the ESP register. This stack is accessed during the PUSH and POP instructions, procedure calls, procedure returns, interrupts, exceptions, and interrupt/exception returns. The GXm processor automatically adjusts the value of the ESP during operations that result from these instructions. The EBP register may be used to refer to data passed on the stack during procedure calls. Local data may also be placed on the stack and accessed with BP . This register provides a mechanism to access tack data in high-level languages.

Revision 3.1 41 www.national.com Processor Programming (Continued) Geode™ GXm Processor Table 3-2. Application Register Set 3 1 3 0 2 9 2 8 2 7 2 6 2 5 2 4 2 3 2 2 2 1 2 0 1 9 1 8 1 7 1 6 1 5 1 4 1 3 1 2 1 1 1 0 9876543210 General Purpose Registers AX AH AL EAX (Extended A Register) BX BH BL EBX (Extended B Register) CX CH CL ECX (Extended C Register) DX DH DL EDX (Extended D Register) SI (Source Index) ESI (Extended Source Index) DI (Destination Index) EDI (Extended Destination Index) BP (Base Pointer) EBP (Extended Base Pointer) SP (Stack Pointer) ESP (Extended Stack Pointer) Segment (Selector) Registers CS (Code Segment) SS (Stack Segment) DS (D Data Segment) ES (E Data Segment) FS (F Data Segment) GS (G Data Segment) Instruction Pointer and EFLAGS Registers EIP (Extended Instruction Pointer) ESP (Extended FLAGS Register)

www.national.com 42 Revision 3.1 Processor Programming (Continued) Geode™ GXm Processor

3.3.1.2 Segment Registers

The 16-bit segment registers, part of the main memory addressing mechanism, are described in Section 3.5 “Off- set, Segment, and Paging Mechanisms” on page 61. The six segment registers are: CS - Code Segment DS - Data Segment SS - Stack Segment ES - Extra Segment FS - Additional Data Segment GS - Additional Data Segment The segment registers are used to select segments in main memory. A segment acts as private memory for dif- ferent elements of a program such as code space, data space, and stack space. There are two segment mechanisms, one for real and vir- tual 8086 operating modes and one for protective mode. Initialization and transition to protective mode is described in Section 3.13.4 “Initialization and Transition to Protected Mode” on page 87. The segment mechanisms are described in Section 3.7 “Descriptors and Segment Mech- anisms” on page 62. The active segment register is selected according to the r u l e sl i s t e di nT a b l e3 - 3a n dt h et y p eo fi n s t r u c t i o nb e i n g currently processed. In general, the DS register selector is used for data references. Stack references use the SS register, and instruction fetches use the CS register. While some of these selections may be overridden, instruction fetches, stack operations, and the destination write opera- tion of string operations cannot be overridden. Special segment-override instruction prefixes allow the use of alternate segment registers. These segment registers include the ES, FS, and GS registers.

3.3.1.3 Instruction Pointer Register

The Instruction Pointer (EIP) Registercontains the off- set into the current code segment of the next instruction to be executed. The register is normally incremented by the length of the current instruction with each instruction exe- cution unless it is implicitly modified through an interrupt, exception, or an instruction that changes the sequential execution flow (for example JMP and CALL). Table 3-3 illustrates the code segment selection rules. Table 3-3. Segment Register Selection Rules Type of Memory Reference Implied (Default) Segment Segment-Override Prefix Code Fetch CS None Destination of PUSH, PUSHF , INT, CALL, PUSHA instructions SS None Source of POP , POPA, POPF , IRET , RET instructions SS None Destination of STOS, MOVS, REP STOS, REP MOVS instructions ES None Other data references with effective address using base registers of: EAX, EBX, ECX, EDX, ESI, EDI, EBP , ESP DS SS C S ,E S ,F S ,G S ,S S CS, DS, ES, FS, GS

Revision 3.1 43 www.national.com Processor Programming (Continued) Geode™ GXm Processor

3.3.1.4 EFLAGS Register

The EFLAGS Register contains status information and controls certain operations on the GXm processor. The lower 16 bits of this register are referred to as the EFLAGS register that is used when executing 8086 or 80286 code. Table 3-4 gives the bit formats for the EFLAGS Register Table 3-4. EFLAGS Register Bit Name Flag Type Description 31:22 RSVD -- Reserved:S e tt o0 . 21 ID System Identification Bit: The ability to set and clear this bit indicates that the CPUID instruction is sup- ported. The ID can be modified only if the CPUID bit in CCR4 (Index E8h[7]) is set. 20:19 RSVD -- Reserved:S e tt o0 .

18 AC System Alignment Check Enable: In conjunction with the AM flag (bit 18) in CR0, the AC flag deter-

mines whether or not misaligned accesses to memory cause a fault. If AC is set, alignment faults are enabled. 17 VM System Virtual 8086 Mode: If set while in protected mode, the processor switches to virtual 8086 opera- tion handling segment loads as the 8086 does, but generating exception 13 faults on privileged opcodes. The VM bit can be set by the IRET instruction (if current privilege level is 0) or by task switches at any privilege level. 16 RF Debug Resume Flag : Used in conjunction with debug register breakpoints. RF is checked at instruction boundaries before breakpoint exception processing. If set, any debug fault is ignored on the next instruction. 15 RSVD -- Reserved:S e tt o0 . 14 NT System Nested Task: While executing in protected mode, NT indicates that the execution of the current task is nested within another task. 13:12 IOPL System I/O Privilege Level: While executing in protected mode, IOPL indicates the maximum current privilege level (CPL) permitted to execute I/O instructions without generating an exception 13 fault or consulting the I/O permission bit map. IOPL also indicates the maximum CPL allowing alteration of the IF bit when new values are popped into the EFLAGS register. 11 OF Arithmetic Overflow Flag: Set if the operation resulted in a carry or borrow into the sign bit of the result but did not result in a carry or borrow out of the high-order bit. Also set if the operation resulted in a carry or borrow out of the high-order bit but did not result in a carry or borrow into the sign bit of the result. 10 DF Control Direction Flag: When cleared, DF causes string instructions to auto-increment (default) the appropriate index registers (ESI and/or EDI). Setting DF causes auto-decrement of the index registers to occur.

9 IF System Interrupt Enable Flag: When set, maskable interrupts (INTR input pin) are acknowledged and

serviced by the CPU. 8 TF Debug Trap Enable Flag: Once set, a single-step interrupt occurs after the next instruction completes execution. TF is cleared by the single-step interrupt. 7 SF Arithmetic Sign Flag: Set equal to high-order bit of result (0 indicates positive, 1 indicates negative). 6 ZF Arithmetic Zero Flag: Set if result is zero; cleared otherwise. 5R S V D - - Reserved:S e tt o0 . 4 AF Arithmetic Auxiliary Carry Flag: Set when a carry out of (addition) or borrow into (subtraction) bit position 3 of the result occurs; cleared otherwise. 3R S V D - - Reserved:S e tt o0 . 2 PF Arithmetic Parity Flag: Set when the low-order 8 bits of the result contain an even number of ones; other- wise PF is cleared. 1R S V D Reserved:S e tt o1 . 0 CF Arithmetic Carry Flag: Set when a carry out of (addition) or borrow into (subtraction) the most significant bit of the result occurs; cleared otherwise.

www.national.com 44 Revision 3.1 Processor Programming (Continued) Geode™ GXm Processor

3.3.2 System Register Set

The system register set, shown in Table 3-5, consists of registers not generally used by application programmers. These registers are typically employed by system level programmers who generate operating systems and mem- ory management programs. Associated with the system register set are certain tables and segments which are listed in Table 3-5. The Control Registerscontrol certain aspects of the GXm processor such as paging, coprocessor functions, and segment protection. The Descriptor Tableshold descriptors that manage memory segments and tables, interrupts and task switch- ing. The tables are defined by corresponding registers. The two Task State Segments Tables defined by TSS reg- ister are used to save and load the computer state when switching tasks. The Configuration Registersa r eu s e dt od e f i n et h e GXm CPU setup including cache management. The ID registers allow BIOS and other software to identify the specific CPU and stepping. System Management Mode (SMM) control information is stored in the SMM reg- isters. The Debug Registersprovide debugging facilities for the GXm processor and enable the use of data access break- points and code execution breakpoints. The Test Registersprovide a mechanism to test the con- tents of both the on-chip 16 KB cache and the T ranslation Lookaside Buffer (TLB). The TLB is used as a cache for the tables that are used in to translate linear addresses to physical addresses while paging is enabled. Table 3-5 lists the system register sets along with their size and function. Table 3-5. System Register Set Group Name Function Width (Bits) Control Registers CR0 System Control Register CR2 Page Fault Linear Address Register CR3 Page Directory Base Register CR4 Time Stamp Counter 32 Descriptor Tables GDT General Descriptor Table 32 IDT Interrupt Descriptor Table LDT Local Descriptor Table 16 Descriptor Table Registers GDTR GDT Register 32 IDTR IDT Register 32 LDTR LDT Register 16 Tas k St ate Segment and Registers TSS Task State Segment Tables TR TSS Register Setup 16 Configuration Registers CCRn Configuration Control Registers ID Registers DIRn Device Identification Registers SMM Registers SMARn SMM Address Region Registers SMHRn SMM Header Addresses 8 Performance Registers PCR0 Performance Control Register Debug Registers DR0 Linear Breakpoint Address 0 DR1 Linear Breakpoint Address 1 DR2 Linear Breakpoint Address 2 DR3 Linear Breakpoint Address 3 DR6 Breakpoint Status 32 DR7 Breakpoint Control 32 Tes t Registers TR3 Cache Test 32 TR4 Cache Test 32 TR5 Cache Test 32 TR6 TLB Test Control 32 TR7 TLB Test Status 32

Revision 3.1 45 www.national.com Processor Programming (Continued) Geode™ GXm Processor

3.3.2.1 Control Registers

A map of the Control Registers (CR0, CR2, CR3, and CR4) is shown in Table 3-6 and the bit definitions are given in Table 3-7. (These registers should not be confused with the CRRn registers.) The CR0 register contains system control bits which configure operating modes and indicate the general state of the CPU. The lower 16 bits of CR0 are referred to as the Machine Status Word (MSW). When operating in real mode, any program can read and write the control registers. In protected mode, however, only privilege level 0 (most-privileged) programs can read and write these registers. Table 3-6. Control Registers Map 3 1 3 0 2 9 2 8 2 7 2 6 2 5 2 4 2 3 2 2 2 1 2 0 1 9 1 8 1 7 1 6 1 5 1 4 1 3 1 2 1 1 1 0 9876543210 CR4 Register Control Register 4 (R/W) RSVD T S C RSVD CR3 Register Control Register 3 (R/W) PDBR (Page Directory Base Register) RSVD 0 0 RSVD CR2 Register Control Register 2 (R/W) PFLA (Page Fault Linear Address) CR1 Register Control Register 1 (R/W) RSVD CR0 Register Control Register 0 (R/W) P G C D N W RSVD A M R S V D W P RSVD N E R S V D T S E M M P P E Machine Status Word (MSW) Table 3-7. CR4-CR0 Bit Definitions Bit Name Description CR4 Register Control Register 4 (R/W) 31:3 RSVD Reserved: Set to 0 (always returns 0 when read). 2T S C Time Stamp Counter Instruction: If = 1 RDTSC instruction enabled for CPL = 0 only; reset state. If = 0 RDTSC instruction enabled for all CPL states. 1:0 RSVD Reserved: Set to 0 (always returns 0 when read). CR3 Register Control Register 3 (R/W) 31:12 PDBR Page Directory Base Register:Identifies page directory base address on a 4 KB page boundary. 11:0 RSVD Reserved: Set to 0. CR2 Register Control Register 2 (R/W) 31:0 PFLA Page Fault Linear Address:With paging enabled and after a page fault, PFLA contains the linear address of the address that caused the page fault. CR1 Register Control Register 1 (R/W) 31:0 RSVD Reserved CR0 Register Control Register 0 (R/W) 31 PG Paging Enable Bit: If PG = 1 and protected mode is enabled (PE = 1), paging is enabled. After changing the state of PG, software must execute an unconditional branch instruction (e.g., JMP , CALL) to have the change take effect.

www.national.com 46 Revision 3.1 Processor Programming (Continued) Geode™ GXm Processor 30 CD Cache Disable: If CD = 1, no further cache line fills occur. However, data already present in the cache continues to be used if the requested address hits in the cache. Writes continue to update the cache and cache invalida- tions due to inquiry cycles occur normally. The cache must also be invalidated with a WBINVD instruction to com- pletely disable any cache activity. 29 NW Not Write-Through: If NW = 1, the on-chip cache operates in write-back mode. In write-back mode, writes are issued to the external bus only for a cache miss, a line replacement of a modified line, execution of a locked instruction, or a line eviction as the result of a flush cycle. If NW = 0, the on-chip cache operates in write-through mode. In write-through mode, all writes (including cache hits) are issued to the external bus. This bit cannot be changed if LOCK_NW = 1 in CCR2. 18 AM Alignment Check Mask: If AM = 1, the AC bit in the EFLAGS register is unmasked and allowed to enable align- ment check faults. Setting AM = 0 prevents AC faults from occurring. 16 WP Write Protect: Protects read-only pages from supervisor write access. WP = 0 allows a read-only page to be written from privilege level 0-2. WP = 1 forces a fault on a write to a read-only page from any privilege level. 5N E Numerics Exception: NE = 1 to allow FPU exceptions to be handled by interrupt 16. NE = 0 if FPU exceptions are to be handled by external interrupts. 4R S V D Reserved: Do not attempt to modify, always 1. 3T S Task Switched: Set whenever a task switch operation is performed. Execution of a floating point instruction with TS = 1 causes a DNA fault. If MP = 1 and TS = 1, a WAIT instruction also causes a DNA fault. 2E M Emulate Processor Extension: If EM = 1, all floating point instructions cause a DNA fault 7. 1M P Monitor Processor Extension: If MP = 1 and TS = 1, a WAIT instruction causes Device Not Available (DNA) fault 7. The TS bit is set to 1 on task switches by the CPU. Floating point instructions are not affected by the state of the MP bit. The MP bit should be set to one during normal operations. 0P E Protected Mode Enable: Enables the segment based protection mechanism. If PE = 1, protected mode is enabled. If PE = 0, the CPU operates in real mode and addresses are formed as in an 8086-style CPU. Refer to Section 3.13 “Protection” on page 86. Table 3-7. CR4-CR0 Bit Definitions (Continued) Bit Name Description Table 3-8. Effects of Various Combinations of EM, TS, and MP Bits CR0[3:1] Instruction Type TS EM MP WAIT ESC 0 0 0 Execute Execute 0 0 1 Execute Execute 1 0 0 Execute Fault 7 1 0 1 Fault 7 Fault 7 0 1 0 Execute Fault 7 0 1 1 Execute Fault 7 1 1 0 Execute Fault 7 1 1 1 Fault 7 Fault 7

Revision 3.1 47 www.national.com Processor Programming (Continued) Geode™ GXm Processor

3.3.2.2 Configuration Registers

The configuration registers listed in T able 3-9 are CPU registers and are selected by register index numbers. The registers are accessed through I/O memory locations 22h and 23h. Registers are selected for access by writing an index number to I/O Port 22h using an OUT instruction prior to transferring data through I/O Port 23h. Each data transfer through I/O Port 23h must be preceded by a register index selection through I/O Port 22h; other- wise, subsequent I/O Port 23h operations are directed off- chip and produce external I/O cycles. If MAPEN, bit 4 of CCR3 (Index C3h[4]) = 0, external I/O cycles occur if the register index number is outside the range C0h-CFh, FEh, and FFh. The MAPEN bit should remain 0 during normal operation to allow system regis- ters located at I/O Port 22h to be accessed. Table 3-9. Configuration Register Summary Index Type Name Access Controlled By* Default Value Reference (Bit Formats) C1h R/W CCR1 — Configuration Control 1 SMI_LOCK 00h Table 3-11 on page 49 C2h R/W CCR2 — Configuration Control 2 -- 00h Table 3-11 on page 49 C3h R/W CCR3 — Configuration Control 3 SMI_LOCK 00h Table 3-11 on page 49 E8h R/W CCR4 — Configuration Control 4 MAPEN 85h Table 3-11 on page 50 EBh R/W CCR7 — Configuration Control 7 -- 00h Table 3-11 on page 50 20h R/W PCR — Performance Control MAPEN 07h Table 3-11 on page 50 B0h R/W SMHR0 — SMM Header Address 0 MAPEN xxh Table 3-11 on page 50 B1h R/W SMHR1 — SMM Header Address 1 MAPEN xxh Table 3-11 on page 50 B2h R/W SMHR2 — SMM Header Address 2 MAPEN xxh Table 3-11 on page 50 B3h R/W SMHR3 — SMM Header Address 3 MAPEN xxh Table 3-11 on page 50 B8h R/W GCR — Graphics Control Register MAPEN 00h Table 4-1 on page 92 B9h VGACTL — VGA Control Register -- 00h Table 5-5 on page 173 BAh- BDh VGAM0 — VGA Mask Register -- 00h Table 5-5 on page 173 CDh R/W SMAR0 — SMM Address 0 SMI_LOCK 00h Table 3-11 on page 51 CEh R/W SMAR1 — SMM Address 1 SMI_LOCK 00h Table 3-11 on page 51 CFh R/W SMAR2 — SMM Address 2 SMI_LOCK 00h Table 3-11 on page 51 FEh RO DIR0 — Device ID 0 -- 4xh Table 3-11 on page 51 FFh RO DIR1 — Device ID 1 -- xxh Table 3-11 on page 51 Note: *MAPEN = Index C3h[4] (CCR3) and SMI_LOCK = Index C3h[0] (CCR3).

www.national.com 48 Revision 3.1 Processor Programming (Continued) Geode™ GXm Processor Table 3-10. Configuration Register Map Register (Index) Bit 7 Bit 6 Bit 5 Bit 4 Bit 3 Bit 2 Bit 1 Bit 0 Control Registers CCR1 (C1h) RSVD SMAC USE_SMI RSVD CCR2 (C2h) USE_SUSP RSVD WT1 SUSP_HLT LOCK_NW RSVD CCR3 (C3h) LSS_34 LSS_23 LSS_12 MAPEN RSVD NMI_EN SMI_LOCK CCR4 (E8h) CPUID SMI_NEST RSVD DTE_EN MEM_BYP IORT2 IORT1 IORT0 CCR7 (EBh) RSVD NMI RSVD EMMX PCR (20h) LSSER RSVD Device ID Registers DIR0 (FEh) DID3 DID2 DID1 DID0 RSVD CLKMODE1 RSVD CLMODE0 DIR1 (FFh) SID3 SID2 SID1 SID0 RID3 RID2 RID1 RID0 SMM Base Header Address Registers SMAR0 (CDh) A31 A30 A29 A28 A27 A26 A25 A24 SMAR1 (CEh) A23 A22 A21 A20 A19 A18 A17 A16 SMAR2 (CFh) A15 A14 A13 A12 SIZE3 SIZE2 SIZE1 SIZE0 SMHR0 (B0h) A7 A6 A5 A4 A3 A2 A1 A0 SMHR1 (B1h) A15 A14 A13 A12 A11 A10 A9 A8 SMHR2 (B2h) A23 A22 A21 A20 A19 A18 A17 A16 SMHR3 (B3h) A31 A30 A29 A28 A27 A26 A26 A24 Graphics/VGA Related Registers GCR (B8h) RSVD Scratchpad Size Base Address Code VGACTL (B9h) RSVD Enable SMI for VGA memory B8000h to BFFFFh Enable SMI for VGA memory B0000h to B7FFFh Enable SMI for VGA memory A0000h to AFFFFh VGAM0 (BAh) VGA Mask Register Bits [7:0] VGAM1 (BBh) VGA Mask Register Bits [15:8] VGAM2 (BCh) VGA Mask Register Bits [23:16] VGAM3 (BDh) VGA Mask Register Bits [31:24]

Revision 3.1 49 www.national.com Processor Programming (Continued) Geode™ GXm Processor Table 3-11. Configuration Registers Bit Name Description Index C1h CCR1 — Configuration Control Register 1 (R/W) Default Value = 00h 7:3 RSVD Reserved:S e tt o0 . 2:1 SMAC System Management Memory Access: If = 00: SMM is disabled. If = 01: SMI# pin is active to enter SMM. SMINT instruction is inactive. If = 10: SMM is disabled. If = 11: SMINT instruction is active to enter SMM. SMI# pin is inactive. Note: SMI_LOCK (CCR3[0]) must = 0, or the CPU must be in SMI mode, to write this bit. 0R S V D Reserved:S e tt o0 . Note: Bits 1 and 2 are cleared to zero at reset. Index C2h CCR2 — Configuration Control Register 2 (R/W) Default Value = 00h 7U S E _ S U S P Enable Suspend Pins: If = 1: SUSP# input and SUSPA# output are enabled. If = 0: SUSP# input is ignored. 6R S V D Reserved: This is a test bit that must be set to 0. 5R S V D Reserved:S e tt o0 . 4W T 1 Write-Through Region 1: If = 1: Forces all writes to the address region between 640 KB to 1 MB that hit in the on-chip cache to be issued on the external bus.

3 SUSP_HLT Suspend on HALT :

If = 1: CPU enters suspend mode following execution of a HALT instruction. 2L O C K _ N W Lock NW Bit: If = 1: Prohibits changing the state of the NW bit (CR0[29]) (refer to Table 3-7 on page 45). Set to 1 after setting NW. 1:0 RSVD Reserved: Set to 0. Note: All bits are cleared to zero at reset. Index C3h CCR3 — Configuration Control Register 3 (R/W) Default Value = 00h

7 LSS_34 L o a d / S t o r eS e r i a l i z e3G Bt o4G B:

If = 1: Strong R/W ordering imposed in address range C0000000h to FFFFFFFFh:

6 LSS_23 L o a d / S t o r eS e r i a l i z e2G Bt o3G B:

If = 1: Strong R/W ordering imposed in address range 80000000h to BFFFFFFFh:

5 LSS_12 L o a d / S t o r eS e r i a l i z e1G Bt o2G B:

If = 1: Strong R/W ordering imposed in address range 40000000h to 7FFFFFFFh

4 MAPEN Map Enable:

If = 1: All configuration registers are accessible. All accesses to Port 22h are trapped. If = 0: Only configuration registers Index C1h through CFh, FEh, FFh (CCRn, SMAR, DIRn) are accessible. Other configuration registers (including PCR, SMHRn, GCR, VGACTL, VGAM0) are not accessible.

3 SUSP_SMM_EN Enable Suspend in SMM Mode:

0 = SUSP# ignored in SMM mode. 1 = SUSP# recognized in SMM mode. 2R S V D Reserved: Set to 0. 1N M I _ E N NMI Enable: If = 1: NMI is enabled during SMM. If = 0: NMI is not recognized during SMM. Note: SMI_LOCK (CCR3[0]) must = 0 or the CPU must be in SMI mode to write to this bit.

0 SMI_LOCK SMM Register Lock:

If = 1: SMM Address Region Register (SMAR[31:0]), SMAC (CCR1[2]), USE_SMI (CCR1[1]) cannot be modified unless in SMM routine. Once set, SMI_LOCK can only be cleared by asserting the RESET pin. Note: All bits are cleared to zero at reset.

www.national.com 50 Revision 3.1 Processor Programming (Continued) Geode™ GXm Processor Index E8h CCR4 — Configuration Control Register 4 (R/W) Default Value = 85h 7C P U I D Enable CPUID Instruction: If = 1: The ID bit in the EFLAGS register to be modified and execution of the CPUID instruction occurs as documented in Table 9-2 on page 202. If = 0: The ID bit can not be modified and execution of the CPUID instruction causes an invalid opcode exception. 6S M I _ N E S T SMI Nest: If = 1: SMI interrupts can occur during SMM mode. SMI handlers can optionally set SMI_NEST high to allow higher-priority SMI interrupts while handling the current event

5 FPU_FAST_EN FPU Fast Mode Enable:

If = 0: Disable FPU Fast Mode If = 1: Enable FPU Fast Mode. 4D T E _ E N Directory Table Entry Cache: If = 1: Enables directory table entry to be cached. Cleared to 0 at reset. 3M E M _ B Y P Memory Read Bypassing: If = 1: Enables memory read bypassing. Cleared to 0 at reset. 2:0 IORT(2:0) I/O Recovery Time: Specifies the minimum number of bus clocks between I/O accesses: 000 = No clock delay 100 = 16-clock delay 001 = 2-clock delay 101 = 32-clock delay (default value after reset) 010 = 4-clock delay 110 = 64-clock delay 011 = 8-clock delay 111 = 128-clock delay Cleared to 0 at reset. Note: MAPEN (CCR3[4]) must = 1 to read or write to this register. Index EBh CCR7 — Configuration Control Register 7 (R/W) Default Value = 00h 7:3 RSVD Reserved: Set to 0. 2N M I Generate NMI: 0 = Do nothing 1 = Generate NMI In order to generate multiple NMIs, this bit must be set to zero between each setting of 1. 1R S V D Reserved: Set to 0. 0E M M X Extended MMX Instructions Enable: If = 1: extended MMX instructions are enabled Index 20h PCR — Performance Control Register (R/W) Default Value = 07h

7 LSSER Load/Store Serialize Enable (Reorder Disable):LSSER should be set to ensure that memory-

mapped I/O devices operating outside of the address range 640K to 1M will operate correctly. For memory accesses above 1 GB, refer to CCR3[7:5] (LSS_34, LSS_23, LSS_12.) If = 1: All memory read and write operations will occur in execution order (load/store serializing enabled, reordering disabled). If = 0: Memory reads and write can be reordered for optimum performance (load/store serializing disabled, reordering enabled). Memory accesses in the address range 640K to 1M will always be issued in execution order. 6R S V D Reserved: Set to 0. 5R S V D Reserved: This is a test bit that must be set to 0. 4:0 RSVD Reserved: Set to 0. Note: MAPEN (CCR3[4]) must = 1 to read or write to this register. Index B0h, B1h, B2h, B3h SMHR — SMI Header Address Register (R/W) Default Value = xxh Table 3-11. Configuration Registers (Continued) Bit Name Description

Revision 3.1 51 www.national.com Processor Programming (Continued) Geode™ GXm Processor Index SMHR Bits SMM Header Address Bits [A31:0]:SMHR address bits [31:0] contain the physical base address for the SMM header space. For example, bits [31:24] correspond with Index B3h Refer to Section 3.11.4 “SMM Configuration Registers” on page 80 for more information. B3h B2h B1h B0h A[31:24] A[23:16] A[15:12] A[7:0] Note: MAPEN (CCR3[4]) must = 1 to read or write to this register. Index CDh, CEh, CFh SMAR — SMM Address Region/Size Register (R/W) Default Value = 00h Index SMAR Bits SMM Address Region Bits [A31:A12]:SMAR address bits [31:12] contain the base address for the SMM region. Bits [31:24] correspond with Index CDh Bits [23:16] correspond with Index CEh Bits [15:12] correspond with Index CFh[7:4] Index CFh allows simultaneous access to SMAR address regions bits SMAR[15:12] and size code bits SIZE[3:0]. During access, the upper 4-bits of Port 23h hold SMAR[15:12]. Refer to Section 3.11.4 “SMM Configuration Registers” on page 80 for more information. CDh CEh CFh[7:4] A[31:24] A[23:16] A[15:12] CFh[3:0] SIZE[3:0] SMM Region Size Bits [3:0]:SIZE bits contain the size code for the SMM region. During access the lower 4-bits of port 23 hold SIZE[3:0]. Index CFh allows simultaneous access to SMAR address regions bits SMAR[15:12] (see above) and size code bits. 0000 = SMM Disabled 0100 = 32 KB 1000 = 512 KB 1100 = 8 MB 0001 = 4 KB 0101 = 64 KB 1001 = 1 MB 1101 = 16 MB 0010 = 8 KB 0110 = 128 KB 1010 = 2 MB 1110 = 32 MB 0011 = 16 KB 0111 = 256 KB 1011 = 4 MB 1111 = 4 KB (same as 0001) Note: SMI_LOCK (CCR3[0]) must = 0, or the CPU must be in SMI mode, to write these registers/bits. Index FEh DIR0 — Device Identification Register 0 (RO) Default Value = 4xh 7:4 DID[3:0] Device ID (Read Only):Identifies device as GXm processor. 3:0 MULT[3:0] Core Multiplier (Read Only)— Identifies the core multiplier set by the CLKMODE[2:0] pins (see signal descriptions page 21) If DIR1 (Index FFh) is 30h-4Fh then MULT[3:0]: 0000 = SYSCLK multiplied by 4 (Test mode only) 0001 = SYSCLK multiplied by 6 0010 = SYSCLK multiplied by 4 (Test mode only) 0011 = SYSCLK multiplied by 6 0100 = SYSCLK multiplied by 7 0101 = SYSCLK multiplied by 8 0110 = SYSCLK multiplied by 7 0111 = SYSCLK multiplied by 5 1xxx = Reserved If DIR1 (Index FFh) is 50h or greater then MULT[3:0]: 0000 = SYSCLK multiplied by 4 (Test mode only) 0001 = SYSCLK multiplied by 10 0010 = SYSCLK multiplied by 4 (Test mode only) 0011 = SYSCLK multiplied by 6 0100 = SYSCLK multiplied by 9 0101 = SYSCLK multiplied by 5 0110 = SYSCLK multiplied by 7 0111 = SYSCLK multiplied by 8 1xxx = Reserved Index FFh DIR1 -- Device Identification Register 1 (RO) Default Value = xxh 7:0 DIR1 Device Identification Revision (Read Only)— DIR1 indicates device revision number. If DIR1 is 30h-33h = GXm processor revision 1.0 - 2.3 If DIR1 is 34h-4Fh = GXm processor revision 2.4 - 3.x If DIR1 is 50h or greater = GXm processor revision 5.0 - 5.4.. Table 3-11. Configuration Registers (Continued) Bit Name Description

www.national.com 52 Revision 3.1 Processor Programming (Continued) Geode™ GXm Processor

3.3.2.3 Debug Registers

Six debug registers (DR0-DR3, DR6 and DR7) support debugging on the GXm processor. Memory addresses loaded in the debug registers, referred to as “breakpoints,” generate a debug exception when a memory access of t h es p e c i f i e dt y p eo c c u r st ot h es p e c i f i e da d d r e s s .A breakpoint can be specified for a particular kind of mem- ory access such as a read or write operation. Code and data breakpoints can also be set allowing debug excep- tions to occur whenever a given data access (read or write operation) or code access (execute) occurs. The size of the debug target can be set to 1, 2, or 4 bytes. The debug registers are accessed through MOV instructions that can be executed only at privilege level 0 (real mode is always privilege level 0). The Debug Address Registers (DR0-DR3) each contain the linear address for one of four possible breakpoints. Each breakpoint is further specified by bits in the Debug Control Register (DR7). For each breakpoint address in DR0-DR3, there are corresponding fields L, R/W, and LEN in DR7 that specify the type of memory access asso- ciated with the breakpoint. The R/W field can be used to specify instruction execution as well as data access breakpoints. Instruction execution breakpoints are always taken before execution of the instruction that matches the breakpoint. The Debug Reg- isters are mapped in T able 3-12 Table 3-12. Debug Registers 3 1 3 0 2 9 2 8 2 7 2 6 2 5 2 4 2 3 2 2 2 1 2 0 1 9 1 8 1 7 1 6 1 5 1 4 1 3 1 2 1 1 1 0 9876543210 DR7 Register Debug Control Register 7 (R/W) LEN3 R/W3 LEN2 R/W2 LEN1 R/W1 LEN0 R/W0 0 0 G D 00100G L G L G L G DR6 Register Debug Status Register 6 (RO) 0000000000000000B T B S

0111111111 B 3 B 2 B 1 B 0

DR3 Register‘ Debug Address Register 3 (R/W)) Breakpoint 3 Linear Address DR2 Register Debug Address Register 2 (R/W) Breakpoint 2 Linear Address DR1 Register Debug Address Register 1 (R/W) Breakpoint 1 Linear Address DR0 Register Debug Address Register 0 (R/W) Breakpoint 0 Linear Address Note: All bits marked as 0 or 1 are reserved and should not be modified.

Revision 3.1 53 www.national.com Processor Programming (Continued) Geode™ GXm Processor The Debug Status Register (DR6) reflects conditions that were in effect at the time the debug exception occurred. The contents of the DR6 register are not automatically cleared by the processor after a debug exception occurs, and therefore should be cleared by software at the appro- priate time. Table 3-13 lists the field definitions for the DR6 and DR7 registers. Code execution breakpoints may also be generated by placing the breakpoint instruction (INT3) at the location where control is to be regained. The single-step feature m a yb ee n a b l e db ys e t t i n gt h eT Ff l a g( b i t8 )i nt h e EFLAGS register. This causes the processor to perform a debug exception after the execution of every instruction. Debug Registers 6 and 7 are shown in Table 3-13. Table 3-13. DR7 and DR6 Bit Definitions Field(s) Number of Bits Description DR7 Register Debug Control Register 7 (R/W) R/Wn 2 Applies to the DRn breakpoint address register: 00 = Break on instruction execution only 01 = Break on data write operations only 10 = Not used 11 = Break on data reads or write operations. LENn 2 Applies to the DRn breakpoint address register: 00 = One-byte length 01 = Two-byte length 10 = Not used 11 = Four-byte length. Gn 1 If = 1: breakpoint in DRn is globally enabled for all tasks and is not cleared by the processor as the result of a task switch. Ln 1 If = 1: breakpoint in DRn is locally enabled for the current task and is cleared by the processor as the result of a task switch. GD 1 Global disable of debug register access. GD bit is cleared whenever a debug exception occurs. DR6 Register Debug Status Register 6 (RO) Bn 1 Bn is set by the processor if the conditions described by DRn, R/Wn, and LENn occurred when the debug exception occurred, even if the breakpoint is not enabled via the Gn or Ln bits. BT 1 BT is set by the processor before entering the debug handler if a task switch has occurred to a task with the T bit in the TSS set. BS 1 BS is set by the processor if the debug exception was triggered by the single-step execution mode (TF flag, bit 8, in EFLAGS set). Note: n = 0, 1, 2, and 3

www.national.com 54 Revision 3.1 Processor Programming (Continued) Geode™ GXm Processor

3.3.2.4 Test Registers

The five test registers are used in testing the processor’s Translation Lookaside Buffer (TLB) and on-chip cache. TR6 and TR7 are used for TLB testing, and TR3-TR5 are used for cache testing. Table 3-14 is a register map for the Test Registers with their bit definitions given in T ables 3-15 and 3-17. TLB Test Registers The processor’s TLB is a 32-entry, four-way set associa- tive memory. Each TLB entry consists of a 24-bit tag and 20-bit data. The 24-bit tag represents the high-order 20 bits of the linear address, a valid bit, and three attribute bits. The 20-bit data portion represents the upper 20 bits of the physical address that corresponds to the linear address. The TLB Test Data Register (TR7) contains the upper 20 bits of the physical address (TLB data field), three LRU bits and a control bit. During TLB write operations, the physical address in TR7 is written into the TLB entry selected by the contents of TR6. During TLB lookup oper- ations, the TLB data selected by the contents of TR6 is loaded into TR7. T able 3-15 lists the bit definitions for TR7 and TR6. The TLB Test Control Register (TR6) contains a com- mand bit, the upper 20 bits of a linear address, a valid bit and the attribute bits used in the test operation. The con- tents of TR6 are used to create the 24-bit TLB tag during both write and read (TLB lookup) test operations. The command bit defines whether the test operation is a read or a write. Table 3-14. TLB Test Registers 3 1 3 0 2 9 2 8 2 7 2 6 2 5 2 4 2 3 2 2 2 1 2 0 1 9 1 8 1 7 1 6 1 5 1 4 1 3 1 2 1 1 1 0 9876543210 TR7 Register TLB Test Data Register (R/W) Physical Address 0 0 TLB LRU 0 0 PL REP 0 0 TR6 Register TLB Test Control Register (R/W) Linear Address V D D# U U# R R# 0000C

Revision 3.1 55 www.national.com Processor Programming (Continued) Geode™ GXm Processor Table 3-15. TR7-TR6 Bit Definitions Bit Name Description TR7 Register TLB Test Data Register (R/W) 31:12 Physical Address Physical Address: TLB lookup: Data field from the TLB. TLB write: Data field written into the TLB. 11:10 RSVD Reserved: Set to 0. 9:7 TLB LRU LRU Bits: TLB lookup: LRU bits associated with the TLB entry before the TLB lookup. TLB write: Ignored. 4P L PL Bit: TLB lookup: If PL = 1, read hit occurred. If PL = 0, read miss occurred. TLB write: If PL = 1, REP field is used to select the set. If PL = 0, the pseudo-LRU replacement algorithm is used to select the set. 3:2 REP Set Selection: TLB lookup: If PL = 1, this field indicates the set in which the tag was found. If PL = 0, undefined data. TLB write: If PL = 1, this field selects one of the four sets for replacement. If PL = 0, ignored. 1:0 RSVD Reserved: Set to 0. TR6 Register TLB Test Control Register (R/W) 31:12 Linear Address Linear Address: TLB lookup: The TLB is interrogated per this address. If one and only one match occurs in the TLB, the rest of the fields in TR6 and TR7 are updated per the matching TLB entry. TLB write: A TLB entry is allocated to this linear address.

11 V Valid Bit:

TLB write: If V = 1, the TLB entry contains valid data. If V = 0, target entry is invalidated. 10:9 8:7 6:5 D, D# U, U# R, R# Dirty Attribute Bit and its Complement (D, D#) User/Supervisor Attribute Bit and its Complement (U, U#) Read/Write Attribute Bit and its Complement (R, R#) Effect on TLB Lookup Effect on TLB Write 00 = Do not match Undefined 01 = Match if D, U, or R bit is a 0 Clear the bit 10 = Match if D, U, or R bit is a 1 Set the bit 11 = Match if D, U, or R bit is either a 1 or 0 Undefined 4:1 RSVD Reserved: Set to 0. 0C Command Bit : If C = 1: TLB lookup. If C = 0: TLB write.

www.national.com 56 Revision 3.1 Processor Programming (Continued) Geode™ GXm Processor Cache Test Registers The processor’s 16 KB on-chip cache is a four-way set associative memory that is configured as write-back cache. Each cache set contains 256 entries. Each entry consists of a 20-bit tag address, a 16-byte data field, a valid bit, and four dirty bits. The 20-bit tag represents the high-order 20 bits of the physical address. The 16-byte data represents the 16 bytes of data currently in memory at the physical address represented by the tag. The valid bit indicates whether the data bytes in the cache actually contain valid data. The four dirty bits indicate if the data bytes in the cache have been modified internally without updating external mem- ory (write-back configuration). Each dirty bit indicates the status for one double-word (4 bytes) within the 16-byte data field. F o re a c hl i n ei nt h ec a c h e ,t h e r ea r et h r e eL R Ub i t st h a t indicate which of the four sets was most recently accessed. A line is selected using bits [11:4] of the physi- cal address. Figure 3-1 illustrates the CPU cache archi- tecture. The CPU contains three test registers (TR5-TR3) that allow testing of its internal cache. Bit definitions for the cache test registers are shown in Table 3-17. Using a 16- byte cache fill buffer and a 16-byte cache flush buffer, cache reads and writes may be performed. Figure 3-1. CPU Cache Architecture D E C O D E 255 254 A11-A4 Line Address = Cache Entry (153 bits) Tag Address (20 bits) Data (128 bits) Valid Status (1 bit) Dirty Status (4 bits) Set 0 Set 1 Set 2 Set 3 LRU Table 3-16. Cache Test Registers 3 1 3 0 2 9 2 8 2 7 2 6 2 5 2 4 2 3 2 2 2 1 2 0 1 9 1 8 1 7 1 6 1 5 1 4 1 3 1 2 1 1 1 0 987654 3 2 1 0 TR5 Register (R/W) RSVD Line Selection Set/ DWORD Control Bits TR4 Register (R/W) Cache Tag Address 0 Valid LRU Bits Dirty Bits 0 0 0 TR3 Register (R/W) Cache Data

Revision 3.1 57 www.national.com Processor Programming (Continued) Geode™ GXm Processor Table 3-17. TR5-TR3 Bit Definitions Bit Name Description TR5 Register (R/W) 11:4 Line Selection Line Selection: Physical address bits 11-4 used to select one of 256 lines. 3:2 Set/DWORD Set/DWord Selection: Cache read: Selects which of the four sets in the cache is used as the source for data transferred to the cache flush buffer. Cache write: Selects which of the four sets in the cache is used as the destination for data transferred from the cache fill buffer. Flush buffer read: Selects which of the four Dword in the flush buffer is used during a TR3 read. Fill buffer write: Selects which of the four Dword in the fill buffer is written during a TR3 write. 1:0 Control Bits Control Bits: If = 00: flush read or fill buffer write. If = 01: cache write. If = 10: cache read. If = 11: cache flush. TR4 Register (R/W) 31:12 Upper Tag Address Upper Tag Address: Cache read: Upper 20 bits of tag address of the selected entry. Cache write: Data written into the upper 20 bits of the tag address of the selected entry.

10 Valid Bit Valid Bit:

Cache read: Valid bit for the selected entry. Cache write: Data written into the valid bit for the selected entry. 9:7 LRU Bits LRU Bits: Cache read: The LRU bits for the selected line. xx1 = Set 0 or Set 1 most recently accessed. xx0 = Set 2 or Set 3 most recently accessed. x1x = Most recent access to Set 0 or Set 1 was to Set 0. x0x = Most recent access to Set 0 or Set 1 was to Set 1. 1xx = Most recent access to Set 2 or Set 3 was to Set 2. 0xx = Most recent access to Set 2 or Set 3 was to Set 3. Cache write: Ignored. 6:3 Dirty Bits Dirty Bits: Cache read: The dirty bits for the selected entry (one bit per DWord). Cache write: Data written into the dirty bits for the selected entry. 2:0 RSVD Reserved: Set to 0. TR3 Register (R/W) 31:0 Cache Data Cache Data: Flush buffer read: Data accessed from the cache flush buffer. Fill buffer write: Data to be written into the cache fill buffer.

www.national.com 58 Revision 3.1 Processor Programming (Continued) Geode™ GXm Processor There are five types of test operations that can be exe- cuted:  Flush buffer read F i l l b u f f e r w r i t e C a c h e w r i t e  Cache read  Cache flush These operations are described in detail in Table 3-18. To fill a cache line with data, the fill buffer must be written four times. Once the fill buffer holds a complete cache line of data (16 bytes), a cache write operation transfers the data from the fill buffer to the cache. To read the contents of a cache line, a cache read opera- tion transfers the data in the selected cache line to the flush buffer. Once the flush buffer is loaded, the program- mer accesses the contents of the flush buffer by executing four flush buffer read operations. Table 3-18. Cache Test Operations Test Operation Code Sequence Action Taken Flush Buffer Read MOV TR5, 0h MOV dest,TR3 Set DWORD = 0, control = 00 = flush buffer read. Flush buffer (31:0) --> dest. MOV TR5, 4h MOV dest,TR3 Set DWORD = 1, control = 00 = flush buffer read. Flush buffer (63:32) --> dest. MOV TR5, 8h MOV dest,TR3 Set DWORD = 2, control = 00 = flush buffer read. Flush buffer (95:64) --> dest. MOV TR5, Ch MOV dest,TR3 Set DWORD = 3, control = 00 = flush buffer read. Flush buffer (127:96) --> dest. Fill Buffer Write MOV TR5, 0h MOV TR3, cache_data Set DWORD = 0, control = 00 = fill buffer write. Cache_data --> fill buffer (31:0). MOV TR5, 4h MOV TR3, cache_data Set DWORD = 1, control = 00 = fill buffer write. Cache_data --> fill buffer (63:32). MOV TR5, 8h MOV TR3, cache_data Set DWORD = 2, control = 00 = fill buffer write. Cache_data --> fill buffer (95:64). MOV TR5, Ch MOV TR3, cache_data Set DWORD = 3, control = 00 = fill buffer write. Cache_data --> fill buffer (127:96). Cache Write MOV TR4, cache_tag Cache_tag --> tag address, valid and dirty bits. MOV TR5, line+set+control=01 Fill buffer (127:0) --> cache line (127:0). Cache Read MOV TR5, line+set+control=10 MOV dest, TR4 Cache line (127:0) --> flush buffer (127:0). Cache line tag address, valid/LRU/dirty bits --> dest. Cache Flush MOV TR5, 3h Control = 11 = cache flush, all cache valid bits = 0.

Revision 3.1 59 www.national.com Processor Programming (Continued) Geode™ GXm Processor

3.3.3 Model Specific Register Set

The Model Specific Register (MSR) Set is used to monitor the performance of the processor or a specific component within the processor. A MSR can be read using the RDMSR instruction, opcode 0F32h. During a MSR read, the contents of the particular MSR, specified by the ECX Register, is loaded into the EDX:EAX Registers. A MSR can be written using the WRMSR instruction, opcode 0F30h. During a MSR write, the contents of EX:EAX are loaded into the MSR specified in the ECX Register. The RDMSR and WRMSR instructions are privileged instructions. The GXm processor contains one 64-bit Model Specific Register (MSR10) the Time Stamp Counter (TSC).

3.3.4 Time Stamp Counter

The processor contains a model specific register (MSR) called the Time Stamp Counter (TSC). The TSC, (MSR[10]), is a 64-bit counter that counts the internal CPU clock cycles since the last reset. The TSC uses a continuous CPU core clock and will continue to count clock cycles even when the processor is in suspend or shutdown mode. The TSC is read using a RDMSR instruction, opcode 0F 32h, with the ECX register set to 10h. During a TSC read, the contents of the TSC is loaded into the EDX:EAX regis- ters. The TSC is written to using a WRMSR instruction, opcode 0F 30h with the ECX register set to 10h. During a TSC write, the contents of EX:EAX are loaded into the TSC. The RDMSR and WRMSR instructions are privileged instructions. I na d d i t i o n ,t h eT S Cc a nb er e a du s i n gt h eR D T S C instruction, opcode 0F 31h. The RDTSC instruction loads the contents of the TSC into EDX:EAX. The use of the RDTSC instruction is restricted by the TSC flag (bit 2) in the CR4 register (refer to Tables 3-6 and 3-7 on pages 45 and 45 for CR4 register information). When the TSC bit = 0, the RDTSC instruction can be executed at any privilege level. When the TSC bit = 1, the RDTSC instruction can only be executed at privilege level 0.

www.national.com 60 Revision 3.1 Processor Programming (Continued) Geode™ GXm Processor

3.4 ADDRESS SPACES

The GXm processor can directly address either memory or I/O space. Figure 3-2 illustrates the range of addresses available for memory address space and I/O address space. For the CPU, the addresses for physical memory range between 00000000h and FFFFFFFFh (4 GBytes). The accessible I/O addresses space ranges between 00000000h and 0000FFFFh (64 KB). The CPU does not use coprocessor communication space in upper I/O space between 800000F8h and 800000FFh as do the 386-style CPUs. The I/O locations 22h and 23h are used for GXm processor configuration register access.

3.4.1 I/O Address Space

The CPU I/O address space is accessed using IN and OUT instructions to addresses referred to as “ports.”T h e accessible I/O address space is 64 KB and can be accessed as 8-bit, 16-bit or 32-bit ports. The GXm processor configuration registers reside within the I/O address space at port addresses 22h and 23h and are accessed using the standard IN and OUT instructions. The configuration registers are modified by writing the index of the configuration register to port 22h, and then transferring the data through port 23h. Accesses to the on-chip configuration registers do not generate external I/O cycles. However, each operation on port 23h must be preceded by a write to port 22h with a valid index value. Otherwise, subsequent port 23h operations will communi- cate through the I/O port to produce external I/O cycles with- out modifying the on-chip configuration registers. Write operations to port 22h outside of the CPU index range (C0h-CFh and FEh-FFh) result in external I/O cycles and do not affect the on-chip configuration registers. Reading port 22h generates external I/O cycles. I/O accesses to port address range 3B0h through 3DFh can be trapped to SMI by the CPU if this option is enabled in the BC_XMAP_1 register (see SMIB, SMIC, and SMID bits in Table 4-10 on page 101). Figure 3-2 illustrates the I/O address space.

3.4.2 Memory Address Space

The processor directly addresses up to 4 GB of physical memory even though the memory controller addresses only 1 GB of DRAM. Memory address space is accessed as bytes, words (16 bits) or DWORDs (32 bits). Words a n dD W O R D sa r es t o r e di nc o n s e c u t i v em e m o r yb y t e s with the low-order byte located in the lowest address. The physical address of a word or DWORD is the byte address of the low-order byte. The processor allows memory to be addressed using nine different addressing modes. These addressing modes are used to calculate an offset address, often referred to as an effective address. Depending on the operating mode of the CPU, the offset is then combined, using memory man- agement mechanisms, into a physical address that is applied to the physical memory devices. Memory management mechanisms consist of segmenta- tion and paging. Segmentation allows each program to use several independent, protected address spaces. Pag- ing translates a logical address into a physical address using translation lookup tables. Virtual memory is often implemented using paging. Either or both of these mecha- nisms can be used for management of the GXm proces- sor memory address space. Figure 3-2. Memory and I/O Address Spaces Physical Memory Space Accessible Programmed I/O Space FFFFFFFFh 0000FFFFh 00000000h FFFFFFFFh 00000000h Physical Memory 4G B Not Accessible 64 KB CPU General Configuration Register I/O Space 00000023h 00000022h

Revision 3.1 61 www.national.com Processor Programming (Continued) Geode™ GXm Processor

3.5 OFFSET, SEGMENT, AND PAGING MECHANISMS

The mapping of address space into a sequence of mem- ory locations (often cached) is performed by the offset, segment and paging mechanisms. In general, the offset, segment and paging mechanisms work in tandem as shown below: instruction offset/G01 offset mechanism/G01 offset address offset address/G01 /G02 segment mechanism/G01 linear address linear address/G01 paging mechanism/G01 physical page. As will be explained, the actual operations depend on sev- eral factors such as the current operating mode and if paging is enabled. Note: the paging mechanism uses part of the linear address as an offset on the physical page.

3.6 OFFSET MECHANISM

In all operating modes, the offset mechanism computes an offset (effective) address by adding together up to three values: a base, an index and a displacement. The base, if present, is the value in one of eight general regis- ters at the time of the execution of the instruction. The index, like the base, is a value that is contained in one of the general registers (except the ESP register) when the instruction is executed. The index differs from the base in that the index is first multiplied by a scale factor of 1, 2, 4 or 8 before the summation is made. The third component added to the memory address calculation is the displace- ment that is a value supplied as part of the instruction. Figure 3-3 illustrates the calculation of the offset address. Nine valid combinations of the base, index, scale factor and displacement can be used with the CPU instruction set. These combinations are listed in Table 3-19 on page 61. The base and index both refer to contents of a register as indicated by [Base] and [Index]. In real mode operation, the CPU only addresses the low- est 1 MB of memory and the offset contains 16-bits. In protective mode the offset contains 32 bits. Initialization and transition to protective mode is described in Section 3.13.4 “Initialization and T ransition to Protected Mode” on page 87. Figure 3-3. Offset Address Calculation Table 3-19. Memory Addressing Modes Index Base Displacement Scaling x1, x2, x4, x8 Offset Address (Effective Address) Addressing Mode Base Index Scale Factor (SF) Displacement (DP) Offset Address (OA) Calculation Direct x OA = DP Register Indirect x OA = [BASE] Based x x OA = [BASE] + DP Index x x OA = [INDEX] + DP Scaled Index x x x OA = ([INDEX] * SF) + DP Based Index x x OA = [BASE] + [INDEX] Based Scaled Index x x x OA = [BASE] + ([INDEX] * SF) Based Index with Displacement x x x OA = [BASE] + [INDEX] + DP Based Scaled Index with Displacement x x x x OA = [BASE] + ([INDEX] * SF) + DP

www.national.com 62 Revision 3.1 Processor Programming (Continued) Geode™ GXm Processor

3.7 DESCRIPTORS AND SEGMENT MECHANISMS

Memory is divided into contiguous regions called “seg- ments.” The segments allow the partitioning of individual elements of a program. Each segment provides a zero address-based private memory for such elements as code, data and stack space. The segment mechanisms select a segment in memory. Memory is divided into an arbitrary number of segments, each containing usually much less than the 2 32 byte (4 GByte) maximum. There are two segment mechanisms, one for Real and Virtual 8086 Operating Modes, and one for Protective Mode.

3.7.1 Real and Virtual 8086 Mode Segment

Real Mode Segment Mechanism In real mode operation, the CPU addresses only the low- est 1 MB of memory. In this mode a selector located in one of the segment registers is used to locate a segment. To calculate a physical memory address, the 16-bit seg- ment base address located in the selected segment regis- ter is multiplied by 16 and then a 16-bit offset address is added. The resulting 20-bit address is then extended with twelve zeros in the upper address bits to crate 32-bit phys- ical address. The value of the selector (the INDEX field) is multiplied by 16 to produce a base address (Figure 3-4.) The base address is summed with the instruction offset value to pro- duce a physical address. Virtual 8086 Mode Segment Mechanism In Virtual 8086 mode the operation is performed as in real mode except that a paging mechanism is added. When paging is enabled, the paging mechanism translates the linear address into a physical address using cached look- up tables (refer to Section 3.9 “Paging Mechanism” on page 72). Figure 3-4. Real Mode Address Calculation Offset Mechanism Selected Segment Register Offset Address 000h X1 6

32 Linear Address

(Physical Address) Base Address

12 High Order Address Bits

Revision 3.1 63 www.national.com Processor Programming (Continued) Geode™ GXm Processor

3.7.2 Segment Mechanism in Protective Mode

T h es e g m e n tm e c h a n i s mi np r o t e c t i v em o d ei sm o r ec o m - plex. Basically as in Real and Virtual 8086 modes the off- set address is added to the segment base address to produce a linear address (Figure 3-5). However, the cal- culation of the segment base address is based on the contents of descriptor tables. Again, if paging is enabled the linear address is further processed by the paging mechanism. A more detailed look at the segment mechanisms for real, virtual 8086 and protective modes is illustrated in Figure 3-6. In protective mode, the segment selector is cached. This is illustrated in Figure 3-7 on page 65.

3.7.2.1 Segment Selectors

The segment registers are used to store segment selec- tors. In protective mode, the segment selectors are divided in to three fields: the RPL, TI and INDEX fields as s h o w ni nF i g u r e3 - 6 . The segments are assigned permission levels to prevent application program errors from disrupting operating pro- grams. The Requested Privilege Level (RPL) determines the Effective Privilege Level of an instruction. RPL = 0 indi- cates the most privileged level, and RPL = 3 indicates the least privileged level. Refer to Section 3.13 “Protection” on page 86. Descriptor tables hold descriptors that allow management of segments and tables in address space while in protec- tive mode. The Table Indicator Bit (TI) in the selector selects either the General Descriptor Table (GDT) or one Local Descriptor Tables (LDT) tables. If TI = 0, GDT is selected; if TI =1, LDT is selected. The 13-bit INDEX field in the segment selector is used to index a GDT or LDT table. Figure 3-5. Protected Mode Address Calculation Offset Mechanism Selector Mechanism Offset Address 32Optional Physical Segment Base Address AddressPaging Mechanism Linear Address Memory

www.national.com 64 Revision 3.1 Processor Programming (Continued) Geode™ GXm Processor Figure 3-6. Selector Mechanisms 15 3 2 1 0 INDEX TI INSTRUCTION OFFSET Segment Selector Segment Descriptor Base GDT or LDT Descriptor Table Main Memory Segmentp p = Paging Mechanism RPL + Linear Address Address Physical Address 15 0 INDEX INSTRUCTION OFFSET Logical Address Base Main Memory Segmentp+ Linear Address Address Physical Address x1 6 p= Paging Mechanism for Virtual 8086 Mode only Real and Virtual 8086 Modes Address Logical Segment Selector Protective Mode Logical Address

Revision 3.1 65 www.national.com Processor Programming (Continued) Geode™ GXm Processor Figure 3-7. Selector Mechanism Caching INDEX TI RPL Selector Load Instruction 15 0 Selector In Segment Register Segment Descriptor Segment Descriptor Global Descriptor Table Local Descriptor Table TI = 0 TI = 1 Cached Segment Segment Segment Segment Register Selected By Decoded Instruction Caching Cached and Descriptor Selector Used If Available Base Address

www.national.com 66 Revision 3.1 Processor Programming (Continued) Geode™ GXm Processor

3.7.3 GDTR and LDTR Registers

The GDT , and LDT descriptor tables are defined by the Global Descriptor T able Register (GDTR) and the Local Descriptor Table Register (LDTR) respectively. Some texts refer to these registers as GDT, and LDT descriptors. The following instructions are used in conjunction with the GDT and LDT registers:  LGDT - Load memory to GDTR  LLDT - Load memory to LDTR  SGDT - Store GDTR to memory  SLDT - Store LDTR to memory The GDTR is set up in REAL mode using the LGDT instruction. This is possible as the LGDT instruction is one of two instructions that directly load a linear address (instead of a segment relative address) in protective mode. (The other instruction is the Load Interrupt Descrip- tor Table [LIDT]). As shown in T able 3-20, the GDT registers contain a BASE ADDRESS field and a LIMIT field that define the GDT tables. (The IDTR is described in Section 3.7.3.2 “T ask, Gate and Interrupt Descriptors” on page 67.) Also shown in T able 3-20, the LDTR is only two bytes wide as it contains only a SELECTOR field. The contents of the SELECTOR field point to a descriptor in the GDT .

3.7.3.1 Segment Descriptors

There are several types of descriptors. A segment descriptor defines the base address, limit and attributes of a memory segment. The GDT or LDT can hold several types of descriptors. In particular, the segment descriptors are stored in either of two registers, the GDT, or the LDT as shown in Table 3- 20). Either of these tables can store as many as 8,192 13) eight-byte selectors taking as much as 64 KB of memory. The first descriptor in the GDT (location 0) is not used by the CPU and is referred to as the “null descriptor.” Types of Segment Descriptors The type of memory segments are defined by correspond- ing types of segment descriptors:  Code Segment Descriptors  Data Segment Descriptors  Stack Segment Descriptors  LDT Segment Descriptors Table 3-20. GDTR, LDTR and IDTR Registers 4 7 1 6 1 5 1 4 1 3 1 2 1 1 1 0 9876543210 GDT Register Global Descriptor Table Register BASE LIMIT IDT Register Interrupt Descriptor Table Register BASE LIMIT LDT Register Local Descriptor Table Register SELECTOR

Revision 3.1 67 www.national.com Processor Programming (Continued) Geode™ GXm Processor

3.7.3.2 Task, Gate and Interrupt Descriptors

Besides segment descriptors there are descriptors used in task switching, switching between tasks with different priority and those used to control interrupt functions:  Task State Segment Table Descriptors  Gate Table Descriptors  Interrupt Descriptors. All descriptors have some things in common. They are all eight bytes in length and have three fields (BASE, LIMIT and TYPE). The BASE field defines the starting location for the table or segment. The LIMIT field defines the size and the TYPE field depends on the type of descriptor. One of the main functions of the TYPE field is to define the access rights to the associated segment or table. Interrupt Descriptor Table The Interrupt Descriptor Table is an array of 256 8-byte (4- byte for real mode) interrupt descriptors, each of which is used to point to an interrupt service routine. Every inter- rupt that may occur in the system must have an associ- ated entry in the IDT. The contents of the IDTR are completely visible to the programmer through the use of the SIDT instruction. The IDT descriptor table is defined by the Interrupt Descriptor Table Register (IDTR). Some texts refer to this register as an IDT descriptor. T h ef o l l o w i n gi n s t r u c t i o n sa r eu s e di nc o n j u n c t i o nw i t ht h e IDTR registers:  LIDT - Load memory to IDTR  S I D T-S t o r eI D T Rt om e m o r y The IDTR is set up in REAL mode using the LIDT instruc- tion. This is possible as the LIDT instructions is only one of two instructions that directly load a linear address (instead of a segment relative address) in protective mode. As previously shown in Table 3-20, the IDTR contains a BASE ADDRESS field and a LIMIT field that define the IDT tables.

3.7.4 Descriptor Bit Structure

The bit structure for application and system descriptors is shown in T able 3-21. The explanation of the TYPE field is shown in T able 3-23 on page 68. Table 3-21. Application and System Segment Descriptors 3 1 3 1 2 9 2 8 2 7 2 6 2 5 2 4 2 3 2 2 2 1 2 0 1 9 1 8 1 7 1 6 1 5 1 4 1 3 1 2 1 1 1 0 9876543210 Memory Offset +4 BASE[31:24] G D 0 A V L LIMIT[19:16] P DPL S TYPE BASE[23:16] Memory Offset +0 BASE[15:0] LIMIT[15:0] Table 3-22. Application and System Segment Descriptors Bit Definitions Bit Memory Offset Name Description 31:24 +4 BASE Segment Base Address: Three fields which collectively define the base location for the segment in 4 GB physical address space.7:0 +4 31:16 +0 19:16 +4 LIMIT Segment Limit: Two fields that define the size of the segment based on the Segment Limit Granularity Bit. If G = 1: Limit value interpreted in units of 4 KB. If G = 0: Limit value is interpreted in bytes. 15:0 +0 23 +4 G Segment Limit Granularity Bit:Defines LIMIT multiplier. If G = 1: Limit value interpreted in units of 4 KB. Segment size ranges from 1 byte to 1 MB. If G = 0: Limit value is interpreted in bytes. Segment size ranges from 4 KB to 4 GB. 22 +4 D Default Length for Operands and Effective Addresses: If D = 1: Code segment = 32-bit length for operands and effective addresses If D = 0: Code segment = 16-bit length for operands and effective addresses If D = 1: Data segment = Pushes, calls and pop instructions use 32-bit ESP register If D = 0: Data segment = Stack operations use 16-bit SP register 20 +4 AVL Segment Available: This field is available for use by system software.

www.national.com 68 Revision 3.1 Processor Programming (Continued) Geode™ GXm Processor 15 +4 P Segment Present: If = 1: Segment is memory segment allocated. If = 0: The BASE and LIMIT fields become available for use by the system. Also, If = 0, a segment- not-present exception generated when selector for the descriptor is loaded into a segment register allowing virtual memory management. 14:13 +4 DPL Descriptor Privilege Level: If = 00: Highest privilege level If = 11: Low privilege level 12 +4 S Descriptor Type: If = 1: Code or data segment If = 0: System segment 11:8 +4 TYPE Segment Type: Refer to Table 3-23 on page 68 for TYPE bit definitions. Bit 11 = Executable Bit 10 = Conforming if bit 12 = 1 Bit 10 = Expand Down if bit 12 = 0 Bit 9 = Readable, if Bit 12 = 1 Bit 9 = Writable, if Bit 12 = 0 Bit 8 = Accessed Table 3-22. Application and System Segment Descriptors Bit Definitions (Continued) Bit Memory Offset Name Description Table 3-23. Application and System Segment Descriptors TYPE Bit Definitions TYPE Bits [11:8] System Segment and Gate Types Bit 12 = 0 Application Segment Types Bit 12 = 1 Num SEWA TYPE (Data Segments) 0 0000 Reserved Data Read-Only 1 0001 Available 16-Bit TSS Data Read-Only, accessed 2 0010 LDT Data Read/Write 3 0011 Busy 16-Bit TSS Data Read/Write accessed 4 0100 16-Bit Call Gate Data Read-Only, expand down 5 0101 Task Gate Data Read-Only, expand down, accessed 6 0110 16-Bit Interrupt Gate Data Read/Write, expand down 7 0111 16-Bit Trap Gate Data Read/Write, expand down, accessed Num SCRA TYPE (Code Segments) 8 1000 Reserved Code Execute-Only 9 1001 Available 32-Bit TSS Code Execute-Only, accessed A 1010 Reserved Code Execute/Read B 1011 Busy 32-Bit TSS Code Execute/Read, accessed C 1100 32-Bit Call Gate Code Execute/Read, conforming D 1101 Reserved Code Execute/Read, conforming, accessed E 1110 32-Bit Interrupt Gate Code Execute/Read-Only, conforming F 1111 32-Bit Trap Gate Code Execute/Read-Only, conforming accessed S = Code Segment (not Data Segment) E = Expand Down W = Write Enable A = Accessed C = Conforming Code Segment R = Read Enable

Revision 3.1 69 www.national.com Processor Programming (Continued) Geode™ GXm Processor

3.7.5 Gate Descriptors

Four kinds of gate descriptors are used to provide protec- tion during control transfers: call gates, trap gates, inter- rupt gates and task gates. (For more information on protection refer to Section 3.13 “Protection” on page 86.) Call Gate Descriptor (CGD).Call gates are used to define legal entry points to a procedure with a higher priv- ilege level. The call gates are used by CALL and JUMP instructions in much the same manner as code segment descriptors. When the CPU decodes an instruction and sees it refers to a call gate descriptor in the GDT table or a LDT table, the call gate is used to point to another descriptor in the table that defines the destination code segment. The following privilege levels are tested during the transfer through the call gate:  CPL = Current Privilege Level  RPL = Segment Selector Field  DPL = Descriptor Privilege Level in the call gate descriptor.  DPL = Descriptor Privilege Level in the destination code segment. The maximum value of the CPL and RPL must be equal or less than the gate DPL. For a JMP instruction the desti- nation DPL equals the CPL. For a CALL instruction the destination DPL is less or equals the CPL. Conforming Code Segments. Transfer to a procedure with a higher privilege level can also be accomplished by bypassing the use of call gates, if the requested proce- dure is to be executed in a conforming code segment. Conforming code segments have the C bit set in the TYPE field in their descriptor. The bit structure and definitions for gate descriptors are s h o w ni nT a b l e s3 - 2 4a n dT a b l e3 - 2 5o np a g e6 9 . Table 3-24. Gate Descriptors 3 1 3 0 2 9 2 8 2 7 2 6 2 5 2 4 2 3 2 2 2 1 2 0 1 9 1 8 1 7 1 6 1 5 1 4 1 3 1 2 1 1 1 0 9876543210 Memory Offset +4 OFFSET[31:16] P DPL 0 TYPE 0 0 0 PARAMETERS Memory Offset +0 SELECTOR[15:0] OFFSET[15:0] Table 3-25. Gate Descriptors Bit Definitions Bit Memory Offset Name Description 31:16 +4 OFFSET Offset:Offset used during a call gate to calculate the branch target. 15:0 +0 31:16 +0 SELECTOR Segment Selector 15 +4 P Segment Present 14:13 +4 DPL Descriptor Privilege Level 11:8 +4 TYPE Segment Type: 0100 = 16-bit call gate 1100 = 32-bit call gate 0101 = Task gate 1110 = 32-bit interrupt gate 0110 = 16-bit interrupt gate 1111 = 32-bit trap gate 0111 = 16-bit trap gate 4:0 +4 PARAMETERS Parameters: Number of parameters to copy from the caller’s stack to the called proce- dure’s stack.

www.national.com 70 Revision 3.1 Processor Programming (Continued) Geode™ GXm Processor

3.8 MULTITASKING AND TASK STATE SEGMENTS

The CPU enables rapid task switching using JMP and CALL instructions that refer to Task State Segments ( T S S ) .D u r i n gas w i t c h ,t h ec o m p l e t et a s ks t a t eo ft h ec u r - rent task is stored in its TSS, and the task state of the requested task is loaded from its TSS. The TSSs are defined through special segment descriptors and gates. The Task Register (TR)holds 16-bit descriptors that con- tain the base address and segment limit for each task state segment. The TR is loaded and stored via the LTR and STR instructions, respectively. The TR can only be accessed only during protected mode and can be loaded when the privilege level is 0 (most privileged). When the TR is loaded, the TR selector field indexes a TSS descrip- tor that must reside in the Global Descriptor Table (GDT). Only the 16-bit selector of a TSS descriptor in the TR is accessible. The BASE, TSS LIMT and ACCESS RIGHT fields are program invisible. During task switching, the processor saves the current CPU state in the TSS before starting a new task. The TSS can be either a 386/486-type 32-bit TSS (see Table 3-26) or a 286-type 16-bit TSS (see Table 3-27 on page 71). Task Gate Descriptors. A task gate descriptor provides controlled access to the descriptor for a task switch. The DPL of the task gate is used to control access. The selec- tor’s RPL and the CPL of the procedure must be a higher level (numerically less) than the DPL of the descriptor. The RPL in the task gate is not used. T h eI / OM a pB a s eA d d r e s sf i e l di nt h e3 2 - b i tT S Sp o i n t s to an I/O permission bit map that often follows the TSS at location +68h. Table 3-26. 32-Bit Task State Segment (TSS) Table 31 16 15 0 I / O M a p B a s e A d d r e s s 000000000000000T + 6 4 h

0000000000000000 S e l e c t o r f o r T a s k ’ s L D T + 6 0 h

0000000000000000 G S + 5 C h

0000000000000000 F S + 5 8 h

0000000000000000 D S + 5 4 h

0000000000000000 S S + 5 0 h

0000000000000000 C S + 4 C h

0000000000000000 E S + 4 8 h

0000000000000000 S S f o r C P L = 2 + 1 8 h

ESP for CPL = 2 +14h

0000000000000000 S S f o r C P L = 1 + 1 0 h

ESP for CPL = 1 +Ch

0000000000000000 S S f o r C P L = 0 + 8 h

ESP for CPL = 0 +4h

0000000000000000 B a c k L i n k ( O l d T S S S e l e c t o r ) + 0 h

Note: 0=R e s e r v e d

Revision 3.1 71 www.national.com Processor Programming (Continued) Geode™ GXm Processor Table 3-27. 16-Bit Task State Segment (TSS) Table 15 0 Selector for Task’s LDT +2Ah DS +28h SS +26h CS +24h ES +22h DI +20h SI +1Eh BP +1Ch SP +1Ah BX +18h DX +16h CX +14h AX +12h EFLAGS +10h IP +Eh SS for Privilege Level 0 +Ch SP for Privilege Level 1 +Ah SS for Privilege Level 1 +8h SP for Privilege Level 1 +6h SS for Privilege Level 0 +4h SP for Privilege Level 0 +2h Back Link (Old TSS Selector) +0h

www.national.com 72 Revision 3.1 Processor Programming (Continued) Geode™ GXm Processor

3.9 PAGING MECHANISM

The paging mechanism translates a linear address to its corresponding physical address. If the required page is not currently present in RAM, an exception is generated. When the operating system services the exception, the required page can be loaded into memory and the instruc- tion restarted. Pages are either 4 KB or 1 MB in size. The CPU defaults to 4 KB pages that are aligned to 4 KB boundaries. A page is addressed by using two levels of tables as illus- trated in Figure 3-8. Bits[31:22] of the 32-bit linear address, the Directory Table Index (DTI) are used to locate an entry in the page directory table. The page directory table acts as a 32-bit master index to up to 1 K individual second-level page tables. The selected entry in the page directory table, referred to as the directory table entry (DTE), identifies the starting address of the second- level page table. The page directory table itself is a page a n di s ,t h e r e f o r e ,a l i g n e dt oa4K Bb o u n d a r y .T h ep h y s i - cal address of the current page directory table is stored in the CR3 control register, also referred to as the Page Directory Base Register (PDBR). Figure 3-8. Paging Mechanism Directory Table Index (DTI) Page Table Index (PTI) Page Frame Offset (PFO) 31 22 21 12 11 0 Linear Address DTE Cache 2-Entry Fully Associative Main TLB 32-Entry 4-Way Set Associative DTE 4K B PTE 4K B Physical Page 4G B -4 KB External Memory Directory Table Page Table Memory CR3 Control Register

Revision 3.1 73 www.national.com Processor Programming (Continued) Geode™ GXm Processor Bits [21:12] of the 32-bit linear address, referred to as the Page Table Index (PTI), locate a 32-bit entry in the sec- ond-level page table. This Page Table Entry (PTE) con- tains the base address of the desired page frame. The second-level page table addresses up to 1K individual page frames. A second-level page table is 4 KB in size and is itself a page. Bits [11:0] of the 32-bit linear address, the Page Frame Offset (PFO), locate the desired physical data within the page frame. Since the page directory table can point to 1 K page tables, and each page table can point to 1 K page frames, a total of 1 M page frames can be implemented. Since each page frame contains 4 KB, up to 4 GB of virtual memory can be addressed by the CPU with a single page directory table. Along with the base address of the page table or the page frame, each directory table entry or page table entry con- tains attribute bits and a present bit as illustrated in T able 3-28. If the present bit (P) is set in the DTE, the page table is present and the appropriate page table entry is read. If P = 1 in the corresponding PTE (indicating that the page is in memory), the accessed and dirty bits are updated, if necessary, and the operand is fetched. Both accessed bits are set (DTE and PTE), if necessary, to indicate that the table and the page have been used to translate a linear address. The dirty bit (D) is set before the first write is made to a page. The present bits must be set to validate the remaining bits in the DTE and PTE. If either of the present bits are not set, a page fault is generated when the DTE or PTE is accessed. If P = 0, the remaining DTE/PTE bits are avail- able for use by the operating system. For example, the operating system can use these bits to record where on the hard disk the pages are located. A page fault is also generated if the memory reference violates the page pro- tection attributes. Translation Look-Aside Buffer The translation look-aside buffer (TLB) is a cache for the paging mechanism and replaces the two-level page table lookup procedure for TLB hits. The TLB is a four-way set associative 32-entry page table cache that automatically keeps the most commonly used page table entries in the p r o c e s s o r .T h e3 2 - e n t r yT L B ,c o u p l e dw i t ha4Kp a g e size, results in coverage of 128 KB of memory addresses. The TLB must be flushed when entries in the page tables are changed. The TLB is flushed whenever the CR3 regis- ter is loaded. An individual entry in the TLB can be flushed using the INVLPG instruction. DTE Cache The DTE cache caches the two most recent DTEs so that future TLB misses only require a single page table read to calculate the physical address. The DTE cache is dis- abled following reset and can be enabled by setting the DTE_EN bit in CCR4[4] (Index E8h). Table 3-28. Directory Table Entry (DTE) and Page Table Entry (PTE) Bit Name Description 31:12 BASE ADDRESS Base Address:Specifies the base address of the page or page table. 11:9 AVAILABLE Available:Undefined and Available to the Programmer 8:7 RSVD Reserved: Unavailable to programmer 6D Dirty Bit: PTE format — If = 1: Indicates that a write access has occurred to the page. DTE format — Reserved. 5A Accessed Flag:If set, indicates that a read access or write access has occurred to the page. 4:3 RSVD Reserved: Set to 0. 2U / S User/Supervisor Attribute: If = 1: Page is accessible by User at privilege level 3. If = 0: Page is accessible by Supervisor only when CPL≤ 2. 1W / R Write/Read Attribute: If = 1: Page is writable. If = 0: Page is read only. 0P Present Flag: If = 1: The page is present in RAM and the remaining DTE/PTE bits are validated If = 0: The page is not present in RAM and the remaining DTE/PTE bits are available for use by the pro- grammer.

www.national.com 74 Revision 3.1 Processor Programming (Continued) Geode™ GXm Processor

3.10 INTERRUPTS AND EXCEPTIONS

The processing of either an interrupt or an exception changes the normal sequential flow of a program by trans- ferring program control to a selected service routine. Except for SMM interrupts, the location of the selected service routine is determined by one of the interrupt vec- tors stored in the interrupt descriptor table. True interrupts are hardware interrupts and are generated by signal sources external to the processor. All exceptions (including so-called software interrupts) are produced inter- nally by the processor.

3.10.1 Interrupts

External events can interrupt normal program execution by using one of the three interrupt pins on the GXm pro- cessor:  Non-maskable Interrupt (NMI pin)  Maskable Interrupt (INTR pin)  SMM Interrupt (SMI# pin) For most interrupts, program transfer to the interrupt rou- tine occurs after the current instruction has been com- pleted. When the execution returns to the original program, it begins immediately following the interrupted instruction. The NMI interruptcannot be masked by software and always uses interrupt vector 2 to locate its service routine. Since the interrupt vector is fixed and is supplied inter- nally, no interrupt acknowledge bus cycles are performed. This interrupt is normally reserved for unusual situations such as parity errors and has priority over INTR interrupts. Once NMI processing has started, no additional NMIs are processed until an IRET instruction is executed, typically at the end of the NMI service routine. If NMI is re-asserted before execution of the IRET instruction, one and only one NMI rising edge is stored and then processed after execu- tion of the next IRET . During the NMI service routine, maskable interrupts may be enabled. If an unmasked INTR occurs during the NMI service routine, the INTR is serviced and execution returns to the NMI service routine following the next IRET. If a HALT instruction is executed within the NMI service routine, the CPU restarts execution only in response to RESET , an unmasked INTR or a System Management Mode (SMM) interrupt. NMI does not restart CPU execu- tion under this condition. The INTR interruptis unmasked when the Interrupt Enable Flag (IF , bit 9) in the EFLAGS register is set to 1. Except for string operations, INTR interrupts are acknowl- edged between instructions. Long string operations have interrupt windows between memory moves that allow INTR interrupts to be acknowledged. When an INTR interrupt occurs, the processor performs an interrupt-acknowledge bus cycle. During this cycle, the CPU reads an 8-bit vector that is supplied by an external interrupt controller. This vector selects which of the 256 possible interrupt handlers will be executed in response to the interrupt. The SMM interrupthas higher priority than either INTR or NMI. After SMI# is asserted, program execution is passed to an SMI service routine that runs in SMM address space reserved for this purpose. The remainder of this section does not apply to the SMM interrupts. SMM interrupts are described in greater detail later in this section.

3.10.2 Exceptions

Exceptions are generated by an interrupt instruction or a program error. Exceptions are classified as traps, faults or aborts depending on the mechanism used to report them and the restartability of the instruction which first caused the exception. A Trap exceptionis reported immediately following the instruction that generated the trap exception. Trap excep- tions are generated by execution of a software interrupt instruction (INTO, INT3, INTn, BOUND), by a single-step operation or by a data breakpoint. Software interrupts can be used to simulate hardware interrupts. For example, an INTn instruction causes the processor to execute the interrupt service routine pointed to by the nth vector in the interrupt table. Execution of the interrupt service routine occurs regardless of the state of the IF flag (bit 9) in the EFLAGS register. The one byte INT3, or breakpoint interrupt (vector 3), is a particular case of the INTn instruction. By inserting this one byte instruction in a program, the user can set break- points in the code that can be used during debug.

Revision 3.1 75 www.national.com Processor Programming (Continued) Geode™ GXm Processor Single-step operation is enabled by setting the TF bit (bit 8) in the EFLAGS register. When TF is set, the CPU gen- erates a debug exception (vector 1) after the execution of every instruction. Data breakpoints also generate a debug exception and are specified by loading the debug regis- ters (DR0-DR7) with the appropriate values. A Fault exceptionis reported before completion of the instruction that generated the exception. By reporting the fault before instruction completion, the processor is left in a state that allows the instruction to be restarted and the effects of the faulting instruction to be nullified. Fault exceptions include divide-by-zero errors, invalid opcodes, page faults and coprocessor errors. Debug exceptions (vector 1) are also handled as faults (except for data breakpoints and single-step operations). After execution of the fault service routine, the instruction pointer points to the instruction that caused the fault. An Abort exceptioni sat y p eo ff a u l te x c e p t i o nt h a ti s severe enough that the CPU cannot restart the program at the faulting instruction. The double fault (vector 8) is the only abort exception that occurs on the processor.

3.10.3 Interrupt Vectors

When the processor services an interrupt or exception, the current program’s instruction pointer and flags are pushed onto the stack to allow resumption of execution of the interrupted program. In protected mode, the processor also saves an error code for some exceptions. Program control is then transferred to the interrupt handler (also called the interrupt service routine). Upon execution of an IRET at the end of the service routine, program execution resumes at the instruction pointer address saved on the stack when the interrupt was serviced.

3.10.3.1 Interrupt Vector Assignments

Each interrupt (except SMI#) and exception is assigned one of 256 interrupt vector numbers as shown in Table 3- 29. The first 32 interrupt vector assignments are defined or reserved. INT instructions acting as software interrupts may use any of interrupt vectors, 0 through 255. The non-maskable hardware interrupt (NMI) is assigned vector 2. Illegal opcodes including faulty FPU instructions will cause an illegal opcode exception, interrupt vector 6. NMI interrupts are enabled by setting bit 2 of the CCR7 r e g i s t e r( I n d e xE B h [ 2 ]=1 ,s e eT a b l e3 - 1 1o np a g e4 9f o r register format). In response to a maskable hardware interrupt (INTR), the processor issues interrupt acknowledge bus cycles used to read the vector number from external hardware. These vec- tors should be in the range 32 to 255 as vectors 0 to 31 are predefined. In PCs, vectors 8 through 15 are used.

3.10.3.2 Interrupt Descriptor Table

The interrupt vector number is used by the processor to locate an entry in the interrupt descriptor table (IDT). In real mode, each IDT entry consists of a four-byte far pointer to the beginning of the corresponding interrupt service routine. In protected mode, each IDT entry is an 8-byte descriptor. The Interrupt Descriptor Table Register (IDTR) specifies the beginning address and limit of the IDT. Following reset, the IDTR contains a base address of 0h with a limit of 3FFh. The IDT can be located anywhere in physical memory as determined by the IDTR. The IDT may contain different types of descriptors: interrupt gates, trap gates and task gates. Interrupt gates are used primarily to enter a hard- ware interrupt handler. Trap gates are generally used to enter an exception handler or software interrupt handler. If an interrupt gate is used, the Interrupt Enable Flag (IF) in the EFLAGS register is cleared before the interrupt han- dler is entered. Task gates are used to make the transition to a new task. Table 3-29. Interrupt Vector Assignments Interrupt Vector Function Exception Type

0 Divide error Fault

1 Debug exception Trap/Fault*

2 NMI interrupt

3 Breakpoint Trap

4 Interrupt on overflow Trap

5 BOUND range exceeded Fault

6 Invalid opcode Fault

7 Device not available Fault

8 Double fault Abort

9 Reserved

10 Invalid TSS Fault

11 Segment not present Fault

12 Stack fault Fault

13 General protection fault Trap/Fault

14 Page fault Fault

15 Reserved

16 FPU error Fault

17 Alignment check exception Fault

18:31 Reserved 32:55 Maskable hardware interrupts Trap 0:255 Programmed interrupt Trap Note: *Data breakpoints and single steps are traps. All other debug exceptions are faults.

www.national.com 76 Revision 3.1 Processor Programming (Continued) Geode™ GXm Processor

3.10.4 Interrupt and Exception Priorities

As the CPU executes instructions, it follows a consistent policy for prioritizing exceptions and hardware interrupts. The priorities for competing interrupts and exceptions are listed in Table 3-30. SMM interrupts always take prece- dence. Debug traps for the previous instruction and next instructions are handled as the next priority. When NMI and maskable INTR interrupts are both detected at the same instruction boundary, the GXm processor services the NMI interrupt first. The CPU checks for exceptions in parallel with instruction decoding and execution. Several exceptions can result from a single instruction. However, only one exception is generated upon each attempt to execute the instruction. Each exception service routine should make the appropri- ate corrections to the instruction and then restart the instruction. In this way, exceptions can be serviced until the instruction executes properly. The CPU supports instruction restart after all faults, except when an instruction causes a task switch to a task whose task state segment (TSS) is partially not present. A TSS can be partially not present if the TSS is not page aligned and one of the pages where the TSS resides is not currently in memory. Table 3-30. Interrupt and Exception Priorities Priority Description Notes 0 Warm Reset. Caused by the assertion of WM_RST. 1 SMM hardware interrupt. SMM interrupts are caused by SMI# asserted and always have highest priority. 2 Debug traps and faults from previous instruction. Includes single-step trap and data breakpoints specified in the debug registers. 3 Debug traps for next instruction. Includes instruction execution breakpoints specified in the debug registers. 4 Non-maskable hardware interrupt. Caused by NMI asserted. 5 Maskable hardware interrupt. Caused by INTR asserted and IF = 1. 6 Faults resulting from fetching the next instruction. Includes segment not present, general protection fault and page fault. 7 Faults resulting from instruction decoding. Includes illegal opcode, instruction too long, or privilege violation. 8 WAIT instruction and TS = 1 and MP = 1. Device not available exception generated. 9 ESC instruction and EM = 1 or TS = 1. Device not available exception generated. 10 Floating point error exception. Caused by unmasked floating point exception with NE = 1.

11 Segmentation faults (for each memory reference

required by the instruction) that prevent transferring the entire memory operand. Includes segment not present, stack fault, and general protection fault.

12 Page Faults that prevent transferring the entire

memory operand. 13 Alignment check fault.

Revision 3.1 77 www.national.com Processor Programming (Continued) Geode™ GXm Processor

3.10.5 Exceptions in Real Mode

Many of the exceptions described in Table 3-29 on page 75 are not applicable in real mode. Exceptions 10, 11, and 14 do not occur in real mode. Other exceptions have slightly different meanings in real mode as listed in T able 3-31.

3.10.6 Error Codes

When operating in protected mode, the following exceptions generate a 16-bit error code:  Double Fault  Alignment Check I n v a l i d T S S  Segment Not Present S t a c k F a u l t  General Protection Fault  Page Fault The error code format and bit definitions are shown in T able 3-32. Bits [15:3] (selector index) are not meaningful if the error code was generated as the result of a page fault. The error code is always zero for double faults and alignment check exceptions. Table 3-31. Exception Changes in Real Mode Vector Number Protected Mode Function Real Mode Function 8 Double fault. Interrupt table limit overrun. 10 Invalid TSS. Does not occur.

11 Segment not

present. Does not occur. 12 Stack fault. SS segment limit overrun.

13 General protection

fault. C S ,D S ,E S ,F S ,G Ss e g - ment limit overrun. In pro- tected mode, an error is pushed. In real mode, no error is pushed. 14 Page fault. Does not occur. Table 3-32. Error Codes 1 5 1 4 1 3 1 2 1 1 1 0 9876543 2 1 0 Selector Index S2 S1 S0 Table 3-33. Error Code Bit Definitions Fault Type Selector Index (Bits 15:3) S2 (Bit 2) S1 (Bit 1) S0 (Bit 0) Page Fault Reserved. Fault caused by: 0 = Not present page 1 = Page-level protection violation Fault occurred during: 0 = Read access 1=W r i t ea c c e s s Fault occurred during 0 = Supervisor access 1 = User access. IDT Fault Index of faulty IDT selector. Reserved 1 If = 1, exception occurred while trying to invoke exception or hardware interrupt handler. Segment Fault Index of faulty selector. TI bit of faulty selector 0 If =1, exception occurred while trying to invoke exception or hardware interrupt handler.

www.national.com 78 Revision 3.1 Processor Programming (Continued) Geode™ GXm Processor

3.11 SYSTEM MANAGEMENT MODE

System Management Mode (SMM) is usually employed for system power management or software-transparent emulation of I/O peripherals. SMM mode is entered through a hardware signal “System Management Inter- rupt” (SMI# pin) that has a higher priority than any other interrupt, including NMI. An SMM interrupt can also be triggered from software using an SMINT instruction. Fol- lowing an SMM interrupt, portions of the CPU state are automatically saved, SMM mode is entered, and program execution begins at the base of SMM address space (Fig- ure 3-9). The GXm processor extends System Management Mode (SMM) to support the virtualization of many devices, including VGA video. The SMM mechanism can be trig- gered not only by I/O activity, but by access to selected memory regions. For example, SMM interrupts are gener- ated when VGA addresses are accessed. As will be described, other SMM enhancements have reduced SMM overhead and improved virtualization-software perfor- mance Figure 3-9. System Management Memory Address Space FFFFFFFFh 00000000h Non-SMM SMM Potential SMM Address Space Physical Memory Space FFFFFFFFh 00000000h 4K Bt o3 2M B Physical Memory 4G B Defined SMM Address Space

Revision 3.1 79 www.national.com Processor Programming (Continued) Geode™ GXm Processor

3.11.1 SMM Enhancements

Eight SMM instructions have been added to the x86 instruction set that permit initiating SMM through software a n ds a v i n ga n dr e s t o r i n gt h et o t a lC P Us t a t ew h e ni n SMM. The SMM header now:  Stores 32-bits memory addresses.  Stores 32-bit memory data.  Differentiates memory and I/O accesses.  Indicates if an SMM interrupt was generated by access to a VGA region. The SMM service code is now cacheable. An SMAR reg- ister specifies the SMM region code base and limit. An SMHR register specifies the physical address for the SMM header. The SMI_NEST bit enables the nesting of SMM interrupts.

3.11.2 SMM Operation

SMM execution flow is summarized in Figure 3-10. Enter- ing SMM requires the assertion of the SMI# pin for at least two SYSCLK periods or execution of the SMINT instruction. For the SMI# signal or SMINT instruction to be recog- nized, configuration register bits must be set as shown in Table 3-34. (The configuration registers are discussed in detail in Section 3.3.2.2 “Configuration Registers” on page 47.) After triggering an SMM through the SMI# pin or a SMINT instruction, selected CPU state information is automati- cally saved in the SMM memory space header located at the top of SMM memory space. After saving the header, the CPU enters real mode and begins executing the SMM service routine starting at the SMM memory region base address. The SMM service routine is user definable and may con- tain system or power management software. If the power management software forces the CPU to power down or if the SMM service routine modifies more registers than are automatically saved, the complete CPU state information should be saved. Figure 3-10. SMM Execution Flow Table 3-34. SMI# and SMINT Recognition Requirements Register Bits SMI# SMINT USE_SMI, CCR1[1] (Index C1h) 1 1 SMAC, CCR1[2] (Index C1h) 0 1 SIZE[3:0], SMAR3[3:0] (Index CFh) >0 >0 SMI# Sampled Active or SMINT Instruction Executed CPU State Stored in SMM Address Space Header Program Flow Transfers to SMM Address Space CPU Enters Real Mode Execution Begins at SMM Address Space Base Address RSM Instruction Restores CPU State Using Header Information Normal Execution Resumes

www.national.com 80 Revision 3.1 Processor Programming (Continued) Geode™ GXm Processor

3.11.3 The SMI# Pin

External chipsets can generate an SMI based on numer- ous asynchronous events, including power management timers, I/O address trapping, external devices, audio FIFO events, and others. Since SMI# is edge sensitive, the chipset must generate an edge for each of the events above, requiring arbitration and storage of multiple SMM events. These functions are provided by the CS5530 I/O companion device. The processor generates an SMI when the external pin changes from high-to-low or when an RSM occurs if SMI# has not remained low since the initiation of the previous SMI.

3.11.4 SMM Configuration Registers

The SMAR register specifies the base location of SMM code region and its size limit. This SMAR register is identi- cal to many of the National Semconductor processors. A new configuration control register called SMHR has been added to specify the 32-bit physical address of the SMM header. The SMHR address must be 32-bit aligned as the bottom two bits are ignored by the microcode. Hardware will detect write operations to SMHR, and sig- nal the microcode to recompute the header address. Access to these registers is enabled by MAPEN (Index C3h[4]). The SMAR register writes to the SMM header when the SMAR register is changed. For this reason, changes to the SMAR register should be completed prior to setting up the SMM header. The configuration registers bit formats a r ed e t a i l e di nT a b l e3 - 1 1o np a g e4 9 .

3.11.5 SMM Memory Space Header

Tables 3-35 and show the SMM header. A memory address field has been added to the end (offset –40h) of the header for the GXm processor. Memory data will be stored overlapping the I/O data, since these events can- not occur simultaneously. The I/O address is valid for both IN and OUT instructions, and I/O data is valid only for OUT . The memory address is valid for read and write operations, and memory data is valid only for write opera- tions. With every SMI interrupt or SMINT instruction, selected CPU state information is automatically saved in the SMM memory space header located at the top of SMM address space. The header contains CPU state information that is modified when servicing an SMM interrupt. Included in this information are two pointers. The current IP points to the instruction executing when the SMI was detected, but it is valid only for an internal I/O SMI. The Next IP points to the instruction that will be executed after exiting SMM. The contents of Debug Register 7 (DR7), the Extended FLAGS Register (EFLAGS), and Control Register 0 (CR0) are also saved. If SMM has been entered due to an I/O trap for a REP INSx or REP OUTSx instruction, the Current IP and Next IP fields con- tain the same addresses. In addition, the I and P fields con- tain valid information. If entry into SMM is the result of an I/O trap, it is useful for the programmer to know the port address, data size and data value associated with that I/O operation. This informa- tion is also saved in the header and is valid only if SMI# is asserted during an I/O bus cycle. The I/O trap information is not restored within the CPU when executing a RSM instruction. Table 3-35. SMM Memory Space Header Mem. O f f s e t 3 1 3 0 2 9 2 8 2 7 2 6 2 5 2 4 2 3 2 2 2 1 2 0 1 9 1 8 1 7 1 6 1 5 1 4 1 3 1 2 1 1 1 0 9876543210 -4h DR7 –8h EFLAGS –Ch CR0 –10h Current IP –14h Next IP –18h RSVD CS Selector –1Ch CS Descriptor [63:32] –20h CS Descriptor [31:0] –24h RSVD RSVD N V X M H S P I C –28h I/O Data Size I/O Address [15:0] –2Ch I/O or Memory Data [31:0] (Note) –30h Restored ESI or EDI –32h Memory Address [31:0] Note: Check the M bit at offset 24h to determine if the data is memory or I/O.

Revision 3.1 81 www.national.com Processor Programming (Continued) Geode™ GXm Processor Table 3-36. SMM Memory Space Header Description Name Description Size DR7 Debug Register 7:The contents of Debug Register 7. 4 Bytes EFLAGS Extended FLAGS Register:The contents of Extended FLAGS Register. 4 Bytes CR0 Control Register 0:The contents of Control Register 0. 4 Bytes Current IP Current Instruction Pointer:The address of the instruction executed prior to servicing SMM interrupt. 4B y t e s Next IP Next Instruction Pointer:The address of the next instruction that will be executed after exiting SMM. 4B y t e s CS Selector Code Segment Selector:Code segment register selector for the current code segment. 2 Bytes CS Descriptor Code Segment Descriptor:Encoded descriptor bits for the current code segment. 8 Bytes N Nested SMI Status:Flag that determines whether an SMI occurred during SMM (i.e., nested) 1 Bit V SoftVGA SMI Status:SMI was generated by an access to VGA region. 1 Bit X External SMI Status: If = 1: SMI generated by external SMI# pin If = 0: SMI internally generated by Internal Bus Interface Unit. 1B i t M Memory or I/O Access:0 = I/O access; 1 = Memory access. 1 Bit H Halt Status:Indicates that the processor was in a halt or shutdown prior to servicing the SMM interrupt. 1B i t S Software SMM Entry Indicator: If = 1: Current SMM is the result of an SMINT instruction. If = 0: Current SMM is not the result of an SMINT instruction. 1B i t P REP INSx/OUTSx Indicator: If = 1: Current instruction has a REP prefix. If = 0: Current instruction does not have a REP prefix. 1B i t I IN, INSx, OUT, or OUTSx Indicator: If = 1: Current instruction performed is an I/O WRITE. If = 0: Current instruction performed is an I/O READ. 1B i t C CS Writable:Code Segment Writable If = 1: CS is writable If = 0: CS is not writable 1B i t I/O Data Size Indicates size of data for the trapped I/O cycle: 01h = byte 03h = word 0Fh = DWORD 2B y t e s I/O Address Processor port used for the trapped I/O cycle. 2B y t e s I/O or Memory Data Data associated with the trapped I/O or memory cycleS. 4B y t e s Restored ESI or EDI Restored ESI or EDI Value:Used when it is necessary to repeat a REP OUTSx or REP INSx instruction when one of the I/O cycles caused an SMI# trap. 4B y t e s Memory Address Physical address of the operation that caused the SMI. 4B y t e s Note: INSx = INS, INSB, INSW or INSD instruction. OUTSx = OUTS, OUTSB, OUTSW and OUTSD instruction.

www.national.com 82 Revision 3.1 Processor Programming (Continued) Geode™ GXm Processor

3.11.6 SMM Instructions

The GXm processor core automatically saves the minimal amount of CPU state information when entering an SMM cycle that allows fast SMM service-routine entry and exit. After entering the SMM service routine, the MOV, SVDC, SVLDT and SVTS instructions can be used to save the complete CPU state information. If the SMM service rou- tine modifies more state information than is automatically saved or if it forces the CPU to power down, the complete CPU state information must be saved. Since the CPU is a static device, its internal state is retained when the input clock is stopped. Therefore, an entire CPU-state save is not necessary before stopping the input clock. The SMM instructions, listed in T able 3-37, can be exe- cuted only if all the conditions listed below are met. 1) USE_SMI = 1. 2) SMAR SIZE > 0. 3) Current Privilege level = 0. 4) SMAC bit is high or the CPU is in an SMI service routine. If any one of the conditions above is not met and an attempt is made to execute an SVDC, RSDC, SVLDT, RSLDT, SVTS, RSTS, or RSM instruction, an invalid opcode exception is generated. The SMM instructions can be executed outside of defined SMM space provided the conditions above are met. The SMINT instruction can be used by software to enter SMM. The SMINT instruction can only be used outside an SMM routine if all the conditions listed below are true. 1) USE_SMI = 1 2) SMAR size > 0 3) Current Privilege Level = 0 4) SMAC = 1 If SMI# is asserted to the CPU during a software SMI, the hardware SMI# is serviced after the software SMI has been exited by execution of the RSM instruction. All the SMM instructions (except RSM and SMINT) save or restore 80 bits of data, allowing the saved values to include the hidden portion of the register contents. Table 3-37. SMM Instruction Set Instruction Opcode Format Description SVDC 0F 78h [mod sreg3 r/m] SVDC mem80, sreg3 Save Segment Register and Descriptor Saves reg (DS, ES, FS, GS, or SS) to mem80. RSDC 0F 79h [mod sreg3 r/m] RSDC sreg3, mem80 Restore Segment Register and Descriptor Restores reg (DS, ES, FS, GS, or SS) from mem80. Use RSM to restore CS. Note: Processing “RSDC CS, Mem80” will produce an excep- tion. SVLDT 0F 7Ah [mod 000 r/m] SVLDT mem80 Save LDTR and Descriptor Saves Local Descriptor Table (LDTR) to mem80. RSLDT 0F 7Bh [mod 000 r/m] RSLDT mem80 Restore LDTR and Descriptor Restores Local Descriptor Table (LDTR) from mem80. SVTS 0F 7Ch [mod 000 r/m] SVTS mem80 Save TSR and Descriptor Saves Task State Register (TSR) to mem80. RSTS 0F 7Dh [mod 000 r/m] RSTS mem80 Restore TSR and Descriptor Restores Task State Register (TSR) from mem80. SMINT 0F 38h SMINT Software SMM Entry CPU enters SMM. CPU state information is saved in SMM memory space header and execution begins at SMM base address. RSM 0F AAh RSM Resume Normal Mode Exits SMM. The CPU state is restored using the SMM memory space header and execution resumes at interrupted point. Note: smem80 = 80-bit memory location.

Revision 3.1 83 www.national.com Processor Programming (Continued) Geode™ GXm Processor

3.11.7 SMM Memory Space

SMM memory space is defined by specifying the base address and size of the SMM memory space in the SMAR register. The base address must be a multiple of the SMM memory space size. For example, a 32 KB SMM memory space must be located at a 32 KB address boundary. The memory space size can range from 4 KB to 32 MB. Execu- tion of the interrupt begins at the base of the SMM memory space. SMM memory space accesses are always cacheable, which allows SMM routines to run faster.

3.11.8 SMI Generation

Virtualization software depends on processor-specific hardware to generate SMI interrupts for each memory or I/O access to the device being implemented. The GXm processor implements SMI generation for VGA accesses. Memory write operations in regions A0000h to AFFFFh, B0000h to B7FFFh, and B8000h to BFFFFh generate an SMI. Memory reads are not trapped by the GXm processor. The GXm processor traps I/O addresses for VGA in the following regions: 3B0h to 3BFh, 3C0h to 3CFh, and 3D0h to 3DFh. Memory-write trapping is performed during instruction decode in the processor core. I/O read and write trapping is implemented in the Internal Bus Interface Unit of the GXm processor. The SMI-generation hardware requires two additional configuration registers to control and mask SMI interrupts in the VGA memory space: VGACTL and VGAM. The VGACTL register has a control bit for each address range shown above. The VGAM register has 32 bits that can selectively disable 2 KB regions within the VGA memory. The VGAM applies only to the A0000h-to-AFFFFh region. If this region is not enabled in VGA_CTL, then the con- tents of VGAM is ignored. The purpose of VGAM is to pre- vent SMI from occurring when non-displayed VGA memory is accessed. This is an enhancement which improves performance for double-buffered applications. T h ef o r m a to fe a c hr e g i s t e ri ss h o w ni nC h a p t e r4o ft h i s document.

3.11.9 SMI Service Routine Execution

Upon entry into SMM, after the SMM header has been saved, the CR0, EFLAGS, and DR7 registers are set to their reset values. The Code Segment (CS) register is loaded with the base, as defined by the SMAR register, and a limit of 4 GBytes. The SMI service routine then begins execution at the SMM base address in real mode. The programmer must save the value of any registers that may be changed by the SMI service routine. For data accesses immediately after entering the SMI service rou- tine, the programmer must use CS as a segment override. I/O port access is possible during the routine but care must be taken to save registers modified by the I/O instructions. Before using a segment register, the register and the register’s descriptor cache contents should be saved using the SVDC instruction. Hardware interrupts, INTRs and NMIs, may be serviced d u r i n ga nS M Is e r v i c er o u t i n e .I fi n t e r r u p t sa r et ob es e r - viced while executing in the SMM memory space, the SMM memory space must be within the address range of 0 to 1 MB to guarantee proper return to the SMI service routine after handling the interrupt. INTRs are automatically disabled when entering SMM since the IF flag (EFLAGS register, bit 9) is set to its reset value. Once in SMM, the INTR can be enabled by setting the IF flag. An NMI event in SMM can be enabled by set- ting NMI_EN high in the CCR3 register (Index C3h[1]). If NMI is not enabled while in SMM, the CPU latches one NMI event and services the interrupt after NMI has been enabled or after exiting SMM through the RSM instruction. The processor is always in real mode in SMM, but it may exit to either real or protected mode depending on its state when SMM was initiated. The IDT (Interrupt Descrip- tor T able) indicates which state it will exit to. Within the SMI service routine, protected mode may be entered and exited as required, and real or protected mode device drivers may be called. To exit the SMI service routine, a Resume (RSM) instruc- tion, rather than an IRET, is executed. The RSM instruc- tion causes the GXm processor core to restore the CPU state using the SMM header information and resume exe- cution at the interrupted point. If the full CPU state was saved by the programmer, the stored values should be reloaded before executing the RSM instruction using the MOV, RSDC, RSLDT and RSTS instructions.

3.11.9.1 SMI Nesting

The SMI mechanism supports nesting of SMI interrupts through the SMI handler, the SMI_NEST bit in CCR4[6] (Index E8h), and the Nested SMI Status bit (bit N in the SMM header, see T able on page 80). Nesting is an impor- tant capability in allowing high-priority events, such as audio virtualization, to interrupt lower-priority SMI code for VGA virtualization or power management. SMI_NEST controls whether SMI interrupts can occur during SMM. SMI handlers can optionally set SMI_NEST high to allow higher-priority SMI interrupts while handling the current event. The SMI handler is responsible for managing the SMI header data for nested SMI interrupts. The SMI header must be saved before SMI_NEST is set high, and SMI_NEST must be cleared and its header information restored before an RSM instruction is executed. The Nested SMI Status bit has been added to the SMM header to show whether the current SMI is nested. The processor sets Nested SMI Status high if the processor was in SMM when the SMI was taken. The processor uses Nested SMI Status on exit to determine whether the processor should stay in SMM. When SMI nesting is disabled, the processor holds off external SMI interrupts until the currently executing SMM code exits. When SMI nesting is enabled, the processor can proceed with the SMI. The SMI handler will guarantee

www.national.com 84 Revision 3.1 Processor Programming (Continued) Geode™ GXm Processor that no internal SMIs are generated in SMM, so the pro- cessor ignores such events. If the internal and external SMI signals are received simultaneously, then the internal SMI is given priority to avoid losing the event. The state diagram of the SMI_NEST and Nested SMI Sta- t u sb i t sa r es h o w ni nF i g u r e3 - 1 1w i t he a c hs t a t e explained next. A. When the processor is outside of SMM, Nested SMI Status is always clear and SMI_NEST is set high. B. The first-level SMI interrupt is received by the processor. The microcode clears SMI_NEST, sets Nested SMI Status high and saves the previous value of Nested SMI Status (0) in the SMI header. C. The first-level SMI handler saves the header and sets SMI_NEST high to re-enable SMI interrupts from SMM. D. A second-level (nested) SMI interrupt is received by the processor. This SMI is taken even though the processor is in SMM because the SMI_NEST bit is set high. The microcode clears SMI_NEST, sets Nested SMI Status high and saves the previous value of Nested SMI Status (1) in the SMI header. E. The second-level SMI handler saves the header and sets SMI_NEST to re-enable SMI interrupts within SMM. Another level of nesting could occur during this period. F. The second-level SMI handler clears SMI_NEST to disable SMI interrupts, then restores its SMI header. G. The second-level SMI handler executes an RSM. The microcode sets SMI_NEST, and restores the Nested SMI Status (1) based on the SMI header. H. The first-level SMI handler clears SMI_NEST to disable SMI interrupts, then restores its SMI header. I. The first-level SMI handler executes an RSM. The microcode sets SMI_NEST high and restores the Nested SMI Status (0) based on the SMI header. When the processor is outside of SMM, Nested SMI Sta- tus is always clear and SMI_NEST is set high. Figure 3-11. SMI Nesting State Machine SMI_NEST Nested SMI Status AB CD E F G H I

Revision 3.1 85 www.national.com Processor Programming (Continued) Geode™ GXm Processor

3.11.9.2 CPU States Related to SMM and Suspend

The state diagram shown in Figure 3-12 illustrates the var- ious CPU states associated with SMM and Suspend mode. While in the SMI service routine, the GXm proces- sor core can enter Suspend mode either by (1) executing a halt (HLT) instruction or (2) by asserting the SUSP# input. During SMM operations and while in SUSP#-initiated Suspend mode, an occurrence of either NMI or INTR is latched. (In order for INTR to be latched, the IF flag, EFLAGS register bit 9, must be set.) The INTR or NMI is serviced after exiting Suspend mode. If Suspend mode is entered through a HLT instruction from the operating system or application software, the reception of an SMI# interrupt causes the CPU to exit Suspend mode and enter SMM. If Suspend mode is entered through the hardware (SUSP# = 0) while the operating system or application software is active, the CPU latches one occurrence of INTR, NMI, and SMI#. Figure 3-12. SMM and Suspend Mode State Diagram Suspend Mode (SUSPA# = 0) Suspend Mode (SUSPA# = 0) Suspend Mode (SUSPA# = 0) NMI or INTR HLT* IRET* RSM* SMI# = 0 SMINT* SUSP# = 1 SUSP# = 0 Interrupt Service Routine Interrupt Service Routine OS/Application Software SMI Service Routine (SMI# = 0) NMI or INTR RESET SMI# = 0 (INTR, NMI and SMI# latched) Non-SMM Operations SMM Operations Interrupt Service Routine Suspend Mode (SUSPA# = 0) (INTR and NMI latched) NMI or INTR IRET* SUSP# = 0 SUSP# = 1 IRET* HLT* NMI or INTR *Instructions

www.national.com 86 Revision 3.1 Processor Programming (Continued) Geode™ GXm Processor

3.12 SHUTDOWN AND HALT

The Halt Instruction (HLT) stops program execution and generates a special Halt bus cycle. The GXm processor core then drives out a special Stop Grant bus cycle and enters a low-power Suspend mode if the SUSP_HLT bit in CCR2 (Index C2h[3]) is set. SMI#, NMI, INTR with inter- rupts enabled (IF bit in EFLAGS = 1), or RESET forces the CPU out of the halt state. If the halt state is inter- rupted, the saved code segment and instruction pointer specify the instruction following the HLT. Shutdown occurs when a severe error is detected that prevents further processing. The most common severe error is the triple fault, a fault event while handling a dou- b l ef a u l t .S e t t i n gt h eI D To rt h eG D Tl i m i tt oz e r ow i l l cause a triple fault. An NMI input or a reset can bring the processor out of shutdown. An NMI will work if the IDT limit is large enough, at least 000Fh, to contain the NMI interrupt vec- tor and if the stack has enough room. The stack must be large enough to contain the vector and flag information (the stack pointer must be greater than 0005h).

3.13 PROTECTION

Segment protection and page protection are safeguards built into the GXm processor’s protected-mode architec- ture that deny unauthorized or incorrect access to selected memory addresses. These safeguards allow multitasking programs to be isolated from each other and from the operating system. This section concentrates on segment protection. Selectors and descriptors are the key elements in the seg- ment protection mechanism. The segment base address, size, and privilege level are established by a segment descriptor. Privilege levels control the use of privileged instructions, I/O instructions and access to segments and segment descriptors. Selectors are used to locate seg- ment descriptors. Segment accesses are divided into two basic types, those involving code segments (e.g., control transfers) and those involving data accesses. The ability of a task to access a segment depends on the:  segment type  instruction requesting access  type of descriptor used to define the segment  associated privilege levels (described next) Data stored in a segment can be accessed only by code executing at the same or a more privileged level. A code segment or procedure can only be called by a task exe- cuting at the same or a less privileged level.

3.13.1 Privilege Levels

The values for privilege levels range between 0 and 3. Level 0 is the highest privilege level (most privileged), and level 3 is the lowest privilege level (least privileged). The privilege level in real mode is zero. The Descriptor Privilege Level(DPL) is the privilege level defined for a segment in the segment descriptor. The DPL field specifies the minimum privilege level needed to access the memory segment pointed to by the descriptor. The Current Privilege Level(CPL) is defined as the cur- rent task’s privilege level. The CPL of an executing task is stored in the hidden portion of the code segment register and essentially is the DPL for the current code segment. The Requested Privilege Level(RPL) specifies a selec- tor’s privilege level. RPL is used to distinguish between the privilege level of a routine actually accessing memory (the CPL), and the privilege level of the original requester (the RPL) of the memory access. If the level requested by RPL is less than the CPL, the RPL level is accepted and the Effective Privilege Level (EPL) is changed to the RPL value. If the level requested by RPL is greater than CPL, the CPL overrides the requested RPL and EPL becomes the CPL value. T h el e s s e ro ft h eR P La n dC P Li sc a l l e dt h eE f f e c t i v eP r i v i - lege Level (EPL). Therefore, if RPL = 0 in a segment selec- tor, the EPL is always determined by the CPL. If RPL = 3, t h eE P Li sa l w a y s3r e g a r d l e s so ft h eC P L . For a memory access to succeed, the EPL must be at least as privileged as the Descriptor Privilege Level (EPL ≤ DPL). If the EPL is less privileged than the DPL (EPL > DPL), a general protection fault is generated. For exam- ple, if a segment has a DPL = 2, an instruction accessing the segment only succeeds if executed with an EPL ≤2.

3.13.2 I/O Privilege Levels

The I/O Privilege Level (IOPL) allows the operating sys- tem executing at CPL = 0 to define the least privileged level at which IOPL-sensitive instructions can uncondition- ally be used. The IOPL-sensitive instructions include CLI, IN, OUT, INS, OUTS, REP INS, REP OUTS, and STI. Modification of the IF bit in the EFLAGS register is also sensitive to the I/O privilege level. T h eI O P Li ss t o r e di nt h eE F L A G Sr e g i s t e r( b i t s[ 3 1 : 1 2 ] ) . An I/O permission bit map is available as defined by the 32-bit Task State Segment (TSS). Since each task can have its TSS, access to individual I/O ports can be granted through separate I/O permission bit maps. If CPL ≤ IOPL, IOPL-sensitive operations can be per- formed. If CPL > IOPL, a general protection fault is gener- ated if the current task is associated with a 16-bit TSS. If the current task is associated with a 32-bit TSS and CPL > IOPL, the CPU consults the I/O permission bitmap in the TSS to determine on a port-by-port basis whether or not I/O instructions (IN, OUT, INS, OUTS, REP INS, REP OUTS) are permitted. The remaining IOPL-sensitive operations generate a general protection fault.

Revision 3.1 87 www.national.com Processor Programming (Continued) Geode™ GXm Processor

3.13.3 Privilege Level Transfers

A task’s CPL can be changed only through intersegment control transfers using gates or task switches to a code segment with a different privilege level. Control transfers result from exception and interrupt servicing and from execution of the CALL, JMP , INT, IRET and RET instruc- tions. There are five types of control transfers that are summa- rized in Table 3-38. Control transfers can be made only when the operation causing the control transfer references the correct descriptor type. Any violation of these descriptor usage rules causes a general protection fault. Any control transfer that changes the CPL within a task results in a change of stack. The initial values for the stack segment (SS) and stack pointer (ESP) for privilege levels 0, 1, and 2 are stored in the TSS. During a JMP or CALL control transfer, the SS and ESP are loaded with the new stack pointer and the previous stack pointer is saved on the new stack. When returning to the original privilege level, the RET or IRET instruction restores the SS and ESP of the less-privileged stack.

3.13.3.1 Gates

Gate descriptors described in Section 3.7.5 “Gate Descriptors” on page 69, provide protection for privilege transfers among executable segments. Gates are used to transition to routines of the same or a more privileged level. Call gates, interrupt gates and trap gates are used for privilege transfers within a task. Task gates are used to transfer between tasks. Gates conform to the standard rules of privilege. In other words, gates can be accessed by a task if the effective privilege level (EPL) is the same or more privileged than the gate descriptor’s privilege level (DPL).

3.13.4 Initialization and Transition to Protected Mode

The GXm processor core switches to real mode immedi- ately after RESET . While operating in real mode, the sys- tem tables and registers should be initialized. The GDTR and IDTR must point to a valid GDT and IDT, respectively. The size of the IDT should be at least 256 bytes, and the GDT must contain descriptors that describe the initial code and data segments. T h ep r o c e s s o rc a nb ep l a c e di np r o t e c t e dm o d eb ys e t t i n g the PE bit (CR0 register bit 0). After enabling protected mode, the CS register should be loaded and the instruc- tion decode queue should be flushed by executing an intersegment JMP . Finally, all data segment registers should be initialized with appropriate selector values. Table 3-38. Descriptor Types Used for Control Transfer Type of Control Transfer Operation Types Descriptor Referenced Descriptor Table Intersegment within the same privilege level. JMP , CALL, RET, IRET* Code Segment GDT or LDT Intersegment to the same or a more privileged level. Interrupt within task (could change CPL level). CALL Gate Call GDT or LDT Interrupt Instruction, Exception, External Interrupt Trap or Interrupt Gate IDT Intersegment to a less privileged level (changes task CPL). RET, IRET* Code Segment GDT or LDT Task Switch via TSS CALL, JMP Task State Segment GDT Task Switch via Task Gate CALL, JMP Task Gate GDT or LDT IRET**, Interrupt Instruction, Exception, External Interrupt Task Gate IDT Note: *NT = 0 (Nested Task bit in EFLAGS, bit 14) **NT =1 (Nested Task bit in EFLAGS, bit 14)

www.national.com 88 Revision 3.1 Processor Programming (Continued) Geode™ GXm Processor

3.14 VIRTUAL 8086 MODE

Both real mode and virtual 8086 (V86) modes are sup- ported by the GXm processor, allowing execution of 8086 application programs and 8086 operating systems. V86 mode allows the execution of 8086-type applications, yet still permits use of the paging and protection mechanisms. V86 tasks run at privilege level 3. Before entry, all seg- ment limits must be set to FFFFh (64K) as in real mode.

3.14.1 Memory Addressing

While in V86 mode, segment registers are used in an identical fashion to real mode. The contents of the Seg- ment register are multiplied by 16 and added to the offset to form the Segment Base Linear Address. The GXm pro- cessor permits the operating system to select which pro- grams use the V86 address mechanism and which programs use protected mode addressing for each task. The GXm processor also permits the use of paging when operating in V86 mode. Using paging, the 1 MB address space of the V86 task can be mapped to any region in the 4 GB linear address space. The paging hardware allows multiple V86 tasks to run concurrently, and provides protection and operating sys- tem isolation. The paging hardware must be enabled to run multiple V86 tasks or to relocate the address space of a V86 task to physical address space other than 0.

3.14.2 Protection

All V86 tasks operate with the least amount of privilege (level 3) and are subject to all CPU protected mode protec- t i o nc h e c k s .A sar e s u l t ,a n ya t t e m p tt oe x e c u t eap r i v i - leged instruction within a V86 task results in a general protection fault. In V86 mode, a slightly different set of instructions are sensitive to the I/O privilege level (IOPL) than in protected mode. These instructions are: CLI, INT n, IRET, POPF , PUSHF , and STI. The INT3, INTO and BOUND variations of the INT instruction are not IOPL sensitive.

3.14.3 Interrupt Handling

To fully support the emulation of an 8086-type machine, interrupts in V86 mode are handled as follows. When an interrupt or exception is serviced in V86 mode, program execution transfers to the interrupt service routine at privi- lege level 0 (i.e., transition from V86 to protected mode occurs). The VM bit in the EFLAGS register (bit 17) is cleared. The protected mode interrupt service routine then determines if the interrupt came from a protected mode or V86 application by examining the VM bit in the EFLAGS image stored on the stack. The interrupt service routine may then choose to allow the 8086 operating sys- tem to handle the interrupt or may emulate the function of the interrupt handler. Following completion of the interrupt service routine, an IRET instruction restores the EFLAGS register (restores VM = 1) and segment selectors and control returns to the interrupted V86 task.

3.14.4 Entering and Leaving Virtual 8086 Mode

V86 mode is entered from protected mode by either exe- cuting an IRET instruction at CPL = 0 or by task switching. If an IRET is used, the stack must contain an EFLAGS image with VM = 1. If a task switch is used, the TSS must contain an EFLAGS image containing a 1 in the VM bit position. The POPF instruction cannot be used to enter V86 mode since the state of the VM bit is not affected. V86 mode can only be exited as the result of an interrupt or exception. The transition out must use a 32-bit trap or interrupt gate that must point to a non-conforming privi- lege level 0 segment (DPL = 0), or a 32-bit TSS. These restrictions are required to permit the trap handler to IRET back to the V86 program.

Revision 3.1 89 www.national.com Processor Programming (Continued) Geode™ GXm Processor

3.15 FLOATING POINT UNIT OPERATIONS

The FPU is x87-instruction-set compatible and adheres to the IEEE-754 standard. Because most applications that contain FPU instructions intermix with integer instructions, the GXm processor’s FPU achieves high performance by completing integer and FPU operations in parallel.

3.15.1 FPU (Floating Point Unit) Register Set

In addition to the registers described to this point, the FPU within the CPU provides the user eight data registers accessed in a stack-like manner, a control register, and a status register. The CPU also provides a data register tag word that improves context switching and stack perfor- mance by maintaining empty/non-empty status for each of the eight data registers. In addition, registers contain pointers to (a) the memory location containing the current instruction word and (b) the memory location containing the operand associated with the current instruction word (if any).

3.15.2 FPU Tag Word Register

The CPU maintains a tag word register that is divided into eight tag word fields. These fields assume one of four val- ues depending on the contents of their associated data registers: Valid (00), Zero (01), Special (10), and Empty (11). Note: Denormal, Infinity, QNaN, SNaN and unsup- ported formats are tagged as “Special”. Tag values are maintained transparently by the CPU and are only avail- able to the programmer indirectly through the FSTENV and FSAVE instructions. The tag word with tag fields for each associated physical register, tag(n), is shown in Table 3-39 on page 90.

3.15.3 FPU Status Register

The FPU communicates status information and operation results to the CPU through the status register. The fields in the FPU status register are detailed in T able 3-39 on page 90. These fields include information related to exception status, operation execution status, register sta- tus, operand class, and comparison results. This register is continuously accessible to the CPU regardless of the state of the Control or Execution Units.

3.15.4 FPU Mode Control Register

The FPU Mode Control Register (MCR) shown in T able 3- 39 on page 90 is used by the GXm processor to specify the operating mode of the FPU. The MCR register fields include information related to the rounding mode selected, the amount of precision to be used in the calculations, and the exception conditions which should be reported to the GXm processor using traps. The user controls precision, rounding, and exception reporting by setting or clearing appropriate bits in the MCR.

www.national.com 90 Revision 3.1 Processor Programming (Continued) Geode™ GXm Processor Table 3-39. FPU Registers Bit Name Description FPU Tag Word Register (R/W) (Note) 15:14 TAG7 TAG7: 00 = Valid; 01 = Zero; 10 = Special; 11 = Empty. 13:12 TAG6 TAG6: 00 = Valid; 01 = Zero; 10 = Special; 11 = Empty. 11:10 TAG5 TAG5: 00 = Valid; 01 = Zero; 10 = Special; 11 = Empty. 9:8 T AG4 TAG4: 00 = Valid; 01 = Zero; 10 = Special; 11 = Empty. 7:6 T AG3 TAG3: 00 = Valid; 01 = Zero; 10 = Special; 11 = Empty. 5:4 T AG2 TAG2: 00 = Valid; 01 = Zero; 10 = Special; 11 = Empty. 3:2 T AG1 TAG1: 00 = Valid; 01 = Zero; 10 = Special; 11 = Empty. 1:0 T AG0 TAG0: 00 = Valid; 01 = Zero; 10 = Special; 11 = Empty. FPU Status Register (R/W) (Note)

15 B Copy of ES bit(bit 7 this register)

14 C3 Condition code bit 3

13:11 S Top-of-Stack:Register number that points to the current TOS. 10:8 C[2:0] Condition code bits [2:0] 7E S Error indicator:Set to 1 if unmasked exception detected. 6S F Stack Full:FPU Status Register: or invalid register operation bit.

5 P Precision error exception bit

4U Underflow error exception bit 3O Overflow error exception bit 2Z Divide-by-zero exception bit 1D Denormalized-operand error exception bit 0I Invalid operation exception bit FPU Mode Control Register (R/W) (Note) 15:12 RSVD Reserved: Set to 0. 11:10 RC Rounding Control Bits: 00 = Round to nearest or even 01 = Round towards minus infinity 10 = Round towards plus infinity 11 = Truncate 9:8 PC Precision Control Bits: 00 = 24-bit mantissa 01 = Reserved 10 = 53-bit mantissa 11 = 64-bit mantissa 7:6 RSVD Reserved: Set to 0. 5P Precision error exception bit 4U FPU Mode Control Register

3 O Overflow error exception bit

2 Z Divide-by-zero exception bit

1 D Denormalized-operand error exception bit

0 I Invalid-operation exception bit

Note: R/W only through the environment at store and restore commands.

Revision 3.1 91 www.national.com Geode™ GXm Processor

4.0 Integrated Functions

The Geode GXm processor integrates a memory control- ler, graphics pipeline and display controller in a Unified Memory Architecture (UMA). UMA simplifies system designs and significantly reduces overall system costs associated with high chip count, small footprint notebook designs. Performance degradation in traditional UMA sys- tems is reduced through the use of National Semiconduc- tor’s Display Compression Technology (DCT). Figure 4-1 shows the major functional blocks of the GXm processor and how the internal bus interface unit operates as the interface between the processor’s core units and the integrated functions. This section details how the integrated functions and inter- nal bus interface unit operate and their respective regis- ters. Figure 4-1. Internal Block Diagram Write-Back Unit FPU Internal Bus Interface Unit Graphics Memory Display PCI SDRAM Port CS5530 PCI Bus Integer Cache Unit Integrated Functions MMU (CRT/LCD TFT) X-Bus Pipeline Controller Controller Controller C-Bus

www.national.com 92 Revision 3.1 Integrated Functions (Continued) Geode™ GXm Processor

4.1 INTEGRATED FUNCTIONS PROGRAMMING INTERFACE

The GXm processor performs mapping for the dedicated cache, graphics pipeline, display controller, memory con- troller, and graphics memory, including the frame buffer. It maps these to high memory addresses or GXm processor memory space. The base address for these is controlled by the Graphics Configuration Register (GCR, Index B8h), which specifies address bits [31:30] in physical memory. Figure 4-2 on page 93 shows the address map for the GXm processor. When accessing the GXm processor memory space, address bits [29:24] must be zero. This allows the GXm processor a linear address space with a total of 16 MB. Address bit 23 divides this space into 8 MB for control (bit 23 = 0) and 8 MB for graphics memory (bit 23 = 1). In control space, bits [22:16] are not decoded, so the programmer should set them to zero. Address bit 15 divides the remaining 64 KB address space into scratch- pad RAM and PCI access (bit 15 = 0) and control regis- ters (bit 15 = 1). Device drivers must be responsible for performing physi- cal-to-virtual memory-address translation, including allo- cation of selectors that point to the GXm processor. The processor may be accessed in protected mode by creat- ing a selector with the physical address shown in T able 4- 1, and a limit of 16 MB. A selector with a 64 KB limit is large enough to access all of the GXm processor’s regis- ters and scratchpad RAM.

4.1.1 Graphics Control Register

The GXm processor incorporates graphics functions that require registers to implement and control them. Most of these registers are memory mapped and physically located in the logical units they control. The mapping of these units is controlled by this configuration register. The Graphics Control Register (GCR, Index B8h) is I/O- mapped because it must be accessed before memory mapping can be enabled. Refer to Section 3.3.2.2 “Con- figuration Registers” on page 47 for information on how to access this register. Table 4-1. GCR Register Bit Name Description Index B8h GCR Register (R/W) Default Value = 00h 7:4 RSVD Reserved: Set to 0. 3:2 SP Scratchpad Size:Specifies the size of the scratchpad cache. 00 = 0 KB 01 = 2 KB 10 = 3 KB 11 = 4 KB 1:0 GX GXm Base Address: Specifies the physical address for the base (GX_BASE) of the scratchpad RAM, the graphics memory (frame buffer, compression buffer, etc.) and the other memory mapped registers. 00 = Scratchpad RAM, Graphics Subsystem, and memory-mapped configuration registers are disabled. 01 = Scratchpad RAM and control registers start at GX_BASE = 40000000h. 10 = Scratchpad RAM and control registers start at GX_BASE = 80000000h. 11 = Scratchpad RAM and control registers start at GX_BASE = C0000000h.

Revision 3.1 93 www.national.com Integrated Functions (Continued) Geode™ GXm Processor Figure 4-2. Geode™ GXm Processor Memory Space Conventional Memory UMBs and Expansion ROMs Video BIOS System BIOS Extended Memory PCI Access Scratchpad VGA/MDA Frame Buffers (Soft VGA and/or PCA/ISA) Internal Bus IF Unit Registers Graphics Pipeline Registers S M MS y s t e mC o d e (Frame Buffer, etc.) PCI Access ROM Access (256 KB) A0000h (640 KB) C0000h E0000h 100000h (1 MB) E8000h GX_BASE+8000h GX_BASE+9000h GX_BASE+400000h GX_BASE+800000h GX_BASE+1000000h FFFC0000h FFFFFFFFh (4 GB) Extended Memory Graphics Memory (Frame Buffer, etc.) A0000h (640 KB) C0000h E0000h 100000h (1 MB) E8000h Shadowed Video BIOS Shadowed System BIOS S M MS y s t e mC o d e Physical Address Map DRAM Map MAX *GBADD or Top of DRAM *Top of DRAM PCI Access PCI Access Display Controller Registers Memory Controller Registers Graphics Memory * See BC_DRAM_TOP Table 4-10 on page 101 or MC_GBASE_ADD on page 111. (See Table 4-29 on page 136) (See Table 4-24 on page 124) (See Table 4-9 on page 101) GX_BASE (See Table 4-1 on page 92) FFFF FFFFh MAX Conventional Memory UMBs and Expansion ROMs GX_BASE+8500h GX_BASE+8400h (See Table 4-15 on page 108) GX_BASE+8300h GX_BASE+8100h (See Table 4-5 on page 97) GX_BASE+1000h Power Management Registers (See Table 6-1 on page 179)

www.national.com 94 Revision 3.1 Integrated Functions (Continued) Geode™ GXm Processor

4.1.2 Control Registers

The control registers for the GXm processor use 32 KB of the memory map, starting at GX_BASE+8000h (see Fig- ure 4-2 on page 93). This area is divided into internal bus interface unit, graphics pipeline, display controller, mem- ory controller, and power management sections:  The internal bus interface unit maps 100h locations starting at GX_BASE+8000h.  The graphics pipeline maps 200h locations starting at GX_BASE+8100h.  The display controller maps 100h locations starting at GX_BASE+8300h.  The memory controller maps 100h locations starting at GX_BASE+8400h  GX_BASE+8500h-8FFFh is dedicated to power management registers for the serial packet transmis- sion control, the user-defined power management address space, Suspend Refresh, and SMI status for Suspend/Resume. The register descriptions are contained in the individual subsections of this chapter. Accesses to undefined regis- ters in the GXm processor control register space will not cause a hardware error.

4.1.3 Graphics Memory

The GXm processor’s graphics memory is mapped into 8 MB starting at GX_BASE+800000h. This area includes the frame buffer memory and storage for internal display controller state. The frame buffer is a linear map whose size depends on the current resolution setup in the mem- ory controller. Frame buffer scan lines are not contiguous in many resolutions, so software that renders to the frame buffer must use a skip count to advance between scan lines. The display controller uses the graphics memory that lies between scan lines for internal state. For this rea- son,accessing graphics memory between the end of a scan line and the start of another can cause display prob- lems. The skip count for all supported resolutions is s h o w ni nT a b l e4 - 2 . Graphics memory is allocated from system DRAM by the system BIOS. The graphics memory size is programmed by setting the graphics memory base address in the mem- ory controller. Display drivers communicate with system BIOS about resolution changes, to ensure that the correct amount of graphics memory is allocated.When a graphics resolution change requires an increased amount of graph- ics memory, the system must be rebooted!T h er e a s o nf o r this restriction is that no mechanism exists to recover sys- tem DRAM from the operating system without rebooting. Table 4-2. Display Resolution Skip Counts Screen Resolution Pixel Depth Skip Count 640x480 8 bits 1024 640x480 16 bits 2048 800x600 8 bits 1024 800x600 16 bits 2048 1024x768 8 bits 1024 1024x768 16 bits 2048

Revision 3.1 95 www.national.com Integrated Functions (Continued) Geode™ GXm Processor

4.1.4 L1 Cache Controller

The GXm processor contains an on-board 16 KB unified data/instruction L1 cache. It operates in write-back mode. Since the memory controller is also on-board, the L1 cache requires no external logic to maintain coherency. All DMA cycles automatically snoop the L1 cache. For improved graphics performance, part of the L1 cache operates as a scratchpad RAM to be used by the graphics pipeline as a BLT Buffer. The CD bit (Cache Disable, bit 30) in CR0 globally con- trols the operating mode of the L1 cache. LCD and LWT, Local Cache Disable and Local Write-through bits in the Translation Lookaside Buffer, control the mode on a page- by-page basis. Additionally, memory configuration control can specify certain memory regions as non-cacheable. If the cache is disabled, no further cache line fills occur. However, data already present in the cache continues to be used. For the cache to be completely disabled, the cache must be invalidated with a WBINVD instruction after the cache has been disabled. Write-back caching improves performance by relieving congestion on slower external buses. With four dirty bits, the cache marks dirty locations on a double-word basis. This further reduces the number of double-word bus write operations needed during a replacement or flush opera- tion. The GXm processor will cache SMM regions. This speeds up system management overhead to allow for hardware emulation such as VGA. The cache of the GXm processor provides the ability to r e d e f i n e2K B ,3K B ,o r4K Bo ft h eL 1c a c h et ob e scratchpad memory. The scratchpad area is memory mapped to the upper memory region defined by the GCR register (Index B8h). The valid bits for the scratchpad R A Mw i l la l w a y sb et r u ea n dt h es c r a t c h p a dR A Ml o c a - tions will never be flushed to memory. The scratchpad RAM serves as a general purpose high speed RAM and as a BLT buffer for the graphics pipeline. Incrementing BLT buffer address registers have been added to enable the graphics pipeline to access this memory as a BLT buffer. A 16-byte line buffer dedicated to the graphics pipeline accesses has been added to minimize graphics interference with normal CPU operation. Table 4-3 summarizes the registers contained in the L1 cache. These registers do not have default values and must be initialized before use. Table 4-4 on page 96 gives the register/bit formats. Table 4-3. L1 Cache BitBLT Register Summary Mnemonic Name Function L1_BB0_BASE L1 Cache BitBLT 0 Base Address Contains the address offset to the first byte of BLT Buffer 0 in the scratch- pad memory. L1_BB0_POINTER L1 Cache BitBLT 0 Pointer Contains the address offset to the current line of BLT Buffer 0 in the scratchpad memory. L1_BB1_BASE L1 Cache BitBLT 1 Base Address Contains the offset to the first byte of BLT Buffer 1 in the scratchpad mem- ory. L1_BB1_POINTER L1 Cache BitBLT 1 Pointer Contains the address offset to the current line of BLT Buffer 1 in the scratchpad memory. Note: For information on accessing these registers, refer to Section 4.1.6 “CPU_READ/CPU_WRITE Instructions” on page 99.

www.national.com 96 Revision 3.1 Integrated Functions (Continued) Geode™ GXm Processor Table 4-4. L1 Cache BitBLT Registers Bit Name Description L1_BB0_BASE Register (R/W) Default Value = None 15:12 RSVD Reserved: Set to 0. 11:4 INDEX BitBLT 0 Base Index:T h ei n d e xt ot h es t a r t i n gl i n eo fB L TB u f f e r0 . 3:0 BYTE BitBLT 0 Starting Byte:Determines which byte of the starting line is the beginning of BLT Buffer 0. L1_BB0_POINTER Register (R/W) Default Value = None 15:12 RSVD Reserved: Set to 0. 11:4 INDEX BitBLT 0 Pointer Index:The index to the current line of BLT Buffer 0. 3:0 RSVD Reserved: Set to 0. L1_BB1_Base Register (R/W) Default Value = None 15:12 RSVD Reserved: Set to 0. 11:4 INDEX BitBLT 1 Base Index:T h ei n d e xt ot h es t a r t i n gl i n eo fB L TB u f f e r1 . 3:0 BYTE BitBLT 1 Starting Byte:Determines which byte of the starting line is the beginning of BLT Buffer 1. L1_BB1_POINTER Register (R/W) Default Value = None 15:12 RSVD Reserved: Set to 0. 11:4 INDEX BitBLT 1 Pointer Index:The index to the current line of BLT Buffer 1. 3:0 RSVD Reserved: Set to 0.

Revision 3.1 97 www.national.com Integrated Functions (Continued) Geode™ GXm Processor

4.1.4.1 Scratchpad Memory

The scratchpad RAM is a dedicated high-speed memory cache that contains BLT buffers, SMM header, and a scratchpad area for display drivers. It provides both L1 cache performance and a dedicated resource that cannot be thrown out by other system activity. The configuration of the scratchpad is based on graphics resolution and is described in Table 4-5. The scratchpad memory is part of the on-chip L1 cache memory. The memory size is controlled by bits in the GCR register (Index B8h). The scratchpad memory can be dis- abled, or sized to 2 KB, 3 KB, or 4 KB. The remaining L1 c a c h es i z ei s1 6K Bm i n u st h es c r a t c h p a ds i z e ,a n da l lo f the scratchpad area is subtracted from a single way. The scratchpad memory is used by display drivers and virtualization software. Because this resource must be tightly controlled to avoid conflicts, application software and third-party drivers should avoid accesses to the scratchpad area. The display driver creates and manages two BLT buffers from within the scratchpad area. These BLT buffers are used to transfer source data from system memory into the frame buffer, or for destination data from system memory or the frame buffer. The graphics pipeline accesses the BLT buffers for many common operations, including Bit- BLT transfers, output primitives, and raster text. Display drivers also use a small portion of the scratchpad as an extended register file, since scratchpad read and write accesses are very fast compared to normal memory oper- ations. The virtualization software uses the scratchpad area to store critical SMM information, including the SMI header and SMM system state. No SMM code currently resides in the scratchpad area, although this is an option for future products. When the BLT buffer pointer is used (refer to Table 4-8 on page 99) addresses outside the scratchpad range will wrap around back into the scratchpad RAM. T able 4-5 shows the allocation of scratchpad memory for the 2 KB and 3 KB configurations of the scratchpad. The 2 KB con- figuration uses GX_BASE+0800h to GX_BASE+1000h. The 3 KB configuration uses GX_BASE+0400h to GX_BASE+1000h. These configurations are fixed by the system BIOS during boot and cannot be changed without rebooting the system. Table 4-5. Scratchpad Organization

2 KB Configuration 3 KB Configuration

DescriptionOffset Size Offset Size GX_BASE + 0EE0h 288 bytes GX_BASE + 0EE0h 288 bytes SMM scratchpad GX_BASE + 0E60h 128 bytes GX_BASE + 0E60h 128 bytes Driver scratchpad GX_BASE + 0B30h 816 bytes GX_BASE + 0930h 1328 bytes BLT Buffer 0 GX_BASE + 0800h 816 bytes GX_BASE + 0400h 1328 bytes BLT Buffer 1

www.national.com 98 Revision 3.1 Integrated Functions (Continued) Geode™ GXm Processor

4.1.5 Display Driver Instructions

The GXm processor has four instructions to access pro- cessor core registers. Table 4-6 shows these instructions. Adding CPU instructions does not create a compatibility problem for applications that may depend on receiving illegal opcode traps. The solution is to make these instruc- tions generate an illegal opcode trap unless a compatibil- ity bit is explicitly set. The GXm processor uses the scratchpad size field (bits [3:2] in GCR, Index B8h) to enable or disable all of the graphics instructions. If the s c r a t c h p a ds i z eb i t sa r ez e r o ,m e a n i n gt h a tn o n eo ft h e cache is defined as scratchpad, then hardware will assume that the graphics controller is not being used and the graphics instructions will be disabled. Any other scratchpad size will enable all of the new instructions. Note that the base address of the memory map in the GCR register can still be set up to allow access to the memory controller registers. Table 4-6. Display Driver Instructions Syntax Opcode Description BB0_RESET 0F3A Reset the BLT Buffer 0 pointer to the base. BB1_RESET 0F3B Reset the BLT Buffer 1 pointer to the base. CPU_WRITE 0F3C Write data to CPU internal register. CPU_READ 0F3D Read data from CPU internal register.

Revision 3.1 99 www.national.com Integrated Functions (Continued) Geode™ GXm Processor

4.1.6 CPU_READ/CPU_WRITE Instructions

The GXm processor has several internal registers that control the BLT buffer and power management circuitry in the dedicated cache subsystem. T o avoid adding addi- tional instructions to read and write these registers, the GXm processor has a general mechanism to access inter- nal CPU registers with reasonable performance. The GXm processor has two special instructions to read and write CPU registers: CPU_READ and CPU_WRITE. Both instructions fetch a 32-bit register address fromEBX as s h o w ni nT a b l e4 - 7a n dT a b l e4 - 8.CPU_WRITE uses EAX for the source data, and CPU_READ usesEAX as the destination. Both instructions always transfer 32 bits of data. These instructions work by initiating a special I/O transac- tion where the high address bit is set. This provides a very large address space for internal CPU registers. The BLT buffer base registers define the starting physical addresses of the BLT buffers located within the dedicated L1 cache. The dedicated cache can be configured for up to 4 KB, so 12 address bits are required for each base address. Table 4-7. CPU-Access Instructions Syntax Opcode Registers Length CPU_WRITE 0F3Ch EBX = 32-bit address, EAX = Source 2 bytes CPU_READ 0F3Dh EBX = 32-bit address, EAX = Destination 2 bytes Table 4-8. Address Map for CPU-Access Registers Register EBX Address Description L1_BB0_BASE FFFFFF0Ch BLT Buffer 0 base address (see Table 4-4 on page 96). L1_BB1_BASE FFFFFF1Ch BLT Buffer 1 base address (see Table 4-4 on page 96). L1_BB0_POINTER FFFFFF2Ch BLT Buffer 0 pointer address (see Table 4-4 on page 96). L1_BB1_POINTER FFFFFF3Ch BLT Buffer 1 pointer address (see Table 4-4 on page 96). PM_BASE FFFFFF6Ch Power management base address (see Table 6-3 on page 181). PM_MASK FFFFFF7Ch Power management address mask (see T able 6-3 on page 181).

www.national.com 100 Revision 3.1 Integrated Functions (Continued) Geode™ GXm Processor

4.2 INTERNAL BUS INTERFACE UNIT

The Geode GXm processor’s internal bus interface unit provides control and interface functions to the internal C- Bus (processor core, FPU, graphics pipeline, and L1 cache) and X-Bus (PCI controller, display controller, mem- ory controller, and graphics accelerator) paths, provides control for several sections of memory, and plays an important part in the virtual VGA function. The internal bus interface unit performs, without loss of compatibility, the functions that previously required the external pins IGNNE# and A20M#. The internal bus interface unit provides configuration con- t r o lf o ru pt o2 0d i f f e r e n tr e g i o n sw i t h i ns y s t e mm e m o r y .I t provides 19 configurable memory regions in the address space between 640 KB and 1 MB, with separate control for read access, write access, cacheability, and PCI access. The memory configuration control includes a top-of-mem- ory register and hardware support for VGA emulation plus, the capability to program 20 regions of the memory map for different ROM configurations, and to locate mem- ory-mapped I/O.

4.2.1 FPU Error Support

The FERR# (floating point error) and IGNNE# (ignore numeric error) pins of the 486 microprocessor have been replaced with an IRQ13 (interrupt request 13) pin. In DOS systems, FPU errors are reported by the external vector 13. This mode of operation is specified by clearing the NE bit (bit 5) in the CR0 register. If the NE bit is active, the IRQ13 output of the GXm processor is always driven inac- tive. If the NE bit is cleared, the GXm processor drives I R Q 1 3a c t i v ew h e nt h eE Sb i t( b i t7 )i nt h eF P US t a t u s Register is set high. Software must respond to this inter- rupt with an OUT instruction of an 8-bit operand to F0h or F1h. When the OUT cycle occurs, the IRQ13 pin is driven inactive and the FPU starts ignoring numeric errors. When the ES bit is cleared, the FPU resumes monitoring numeric errors.

4.2.2 A20M Support

The GXm processor provides an A20M bit in the BC_XMAP_1 Register (GX_BASE+ 8004h[21]) to replace the A20M# pin on the 486 microprocessor. When the A20M bit is set high, all non-SMI accesses will have address bit 20 forced to zero. External hardware must do an SMI trap on I/O locations that toggle the A20M# pin. The SMI software can then change the A20M bit as desired. This maintains compatibility with software that depends on wrapping the address at bit 20.

4.2.3 SMI Generation

The internal bus interface unit can generate SMI inter- rupts whenever an I/O cycle in the VGA address range is 3B0h-3BFh and 3C0h-3CFh. An I/O cycle to 3D0h-3DFh can be trapped. In case an external VGA card is present, the Internal Bus Interface Unit default values will not gen- erate an interrupt on VGA accesses. (Refer to Section 5.2.3.1 “SMI Generation” on page 168 for instructions on how to configure the registers to generate the SMI inter- rupt.) 4.2.4 640 KB to 1 MB Region There are 19 configurable memory regions located b e t w e e n6 4 0K Ba n d1M B .T h r e eo ft h er e g i o n sa r e A0000h-AFFFFh, B0000h-B7FFFh, and B8000h- BFFFFh. The area between C0000h and FFFFFh is divided into 16 KB segments to form the remaining 16 regions. Each of these regions has four control bits to allow any combination of read-access, write-access, cache, and PCI-access capabilities (Table 4-11 on page 102). In addition, each of the three regions defined in the A0000h-BFFFh area of memory has a VGA control bit that can cause the graphics pipeline to handle accesses to that section of memory (see T able 5-3 on page 170).

Revision 3.1 101 www.national.com Integrated Functions (Continued) Geode™ GXm Processor

4.2.5 Internal Bus Interface Unit Registers

The Internal Bus Interface Unit maps 100h locations start- ing at GX_BASE+8000h. Refer to Section 4.1.2 “Control Registers” on page 94 for instructions on accessing these registers. Table 4-9 summarizes the four 32-bit registers contained in the Internal Bus Interface Unit and T able 4-10 gives the register/bit formats. Table 4-9. Internal Bus Interface Unit Register Summary GX_BASE+ Memory Offset Type Name/Function Default Value 8000h-8003h R/W BC_DRAM_TOP Top of DRAM: Contains the highest available address of system memory not including the memory that is set aside for graphics memory, which corresponds to 1 GByte of memory. The largest possible value for the register is 3FFFFFFFh. 3FFFFFFFh 8004h-8007h R/W BC_XMAP_1 Memory X-Bus Map Register 1 (A and B Region Control: Contains the region control of the A and B regions and the SMI controls required for VGA emulation. PCI access to internal registers and the A20M function are also controlled by this register. 00000000h 8008h-800Bh R/W BC_XMAP_2 Memory X-Bus Map Register 2 (C and D Region Control): Contains region con- trol fields for eight regions in the address range C0h through DCh. 00000000h 800Ch-800Fh R/W BC_XMAP_3 Memory X-Bus Map Register 3 (E and F Region Control): Contains the region control fields for memory regions in the address range E0h through FCh. 00000000h Table 4-10. Internal Bus Interface Unit Registers Bit Name Description GX_BASE+8000h-8003h BC_DRAM_TOP Register (R/W) Default Value = 3FFFFFFFh 31:30 RSVD Reserved: Set to 0. 29:17 TOP OF DRAM Top of DRAM: Maximum value is FFFh. 16:0 1FFFF Granularity:Must be set to 1FFFFh (128 KB). GX_BASE+8004h-8007h BC_XMAP_1 Register (R/W) Default Value = 00000000h 31:29 RSVD Reserved: Set to 0.

28 GEB8 Graphics Enable for B8 Region:Allow memory R/W operations for address range B8000h-BFFFFh be

directed to the graphics pipeline: 0 = Disable; 1 = Enable. (Used for VGA emulation.) 27:24 B8 B8 Region:Region control field for address range B8000h-BFFFFh. Note: Refer to Table 4-11 for decode. 23 RSVD Reserved: Set to 0.

22 PRAE PCI Register Access Enable:Allow PCI Slave to access internal registers on the X-Bus:

0 = Disable; 1 = Enable.

21 A20M Address Bit 20 Mask:Address bit 20 is always forced to a zero except for SMI accesses:

0 = Disable; 1 = Enable.

20 GEB0 Graphics Enable for B0 Region:Allow memory R/W operations for address range B0000h-B7FFFh be

directed to the graphics pipeline: 0 = Disable; 1 = Enable. (Used for VGA emulation.) 19:16 B0 B0 Region:Region control field for address range B0000h-B7FFFh. Note: Refer to Table 4-11 for decode. 15 SMID SMID: All I/O accesses for address range 3D0h-3DFh generate an SMI: 0 = Disable; 1 = Enable. (Used for VGA virtualization.) 14 SMIC SMIC: All I/O accesses for address range 3C0h-3CFh generate an SMI: 0 = Disable; 1 = Enable. (Used for VGA virtualization.)

13 SMIB SMIB: All I/O accesses for address range 3B0h-3BFh generate an SMI: 0 = Disable; 1 = Enable

(Used for VGA virtualization.)

www.national.com 102 Revision 3.1 Integrated Functions (Continued) Geode™ GXm Processor 12:8 RSVD Reserved: Set to 0. 7X P D X-Bus Pipeline Disable:When cleared, the address for the next cycle can be driven on the internal X- Bus before the completion of the data phase of the current cycle. 6G N W S X - B u sG r a p h i c sP i p eN oW a i tS t a t e :Data driven on X-Bus from graphics pipeline: 0 = 1 full clock before X_DSX is asserted 1 = On the same clock in which X_RDY is asserted 5X N W S X-Bus No Wait State:Data driven on X-Bus from Internal Bus Interface Unit: 0 = 1 full clock before X_DSX is asserted 1 = On the same clock in which X_RDY is asserted 4G E A Graphics Enable for A Region:Memory R/W operations for address range A0000h-AFFFFh are directed to the graphics pipeline: 0 = Disable; 1 = Enable. (Used for VGA emulation.) 3:0 A0 A0 Region:Region control field for address range A0000h-AFFFFh. Note: Refer to Table 4-11 for decode. GX_BASE+8008h-800Bh BC_XMAP_2 Register (R/W) Default Value = 00000000h 31:28 DC DC Region: Region control field for address range DC000h to DFFFFh. 27:24 D8 D8 Region:Region control field for address range D8000h to DBFFFh. 23:20 D4 D4 Region:Region control field for address range D4000h to D7FFFh. 19:16 D0 D0 Region:Region control field for address range D0000h to D3FFFh. 15:12 CC CC Region: Region control field for address range CC000h to CFFFFh. 11:8 C8 C8 Region:Region control field for address range C8000h to CBFFF . 7:4 C4 C4 Region:Region control field for address range C4000h to C7FFFh. 3:0 C0 C0 Region:Region control field for address range C0000h to C3FFFh. Note: Refer to Table 4-11 for decode. GX_BASE+800Ch-800Fh BC_XMAP_3 Register (R/W) Default Value = 00000000h 31:28 FC FC Region:Region control field for address range FC000h to FFFFFh. 27:24 F8 F8 Region:Region control field for address range F8000h to FBFFFh. 23:20 F4 F4 Region:Region control field for address range F4000h to F7FFFh. 19:16 F0 F0 Region:Region control field for address range F0000h to F3FFFh. 15:12 EC EC Region: Region control field for address range EC000h to EFFFFh. 11:8 E8 E8 Region:Region control field for address range E8000h to EBFFFh. 7:4 E4 E4 Region:Region control field for address range E4000h to E7FFFh. 3:0 E0 E0 Region:Region control field for address range E0000h to E3FFFh. Note: Refer to Table 4-11 for decode. Table 4-10. Internal Bus Interface Unit Registers (Continued) Bit Name Description Table 4-11. Region-Control-Field Bit Definitions Bit Position Function 3 PCI Accessible:The PCI slave can access this memory if this bit is set high and if the appropriate Read or Write Enable bit is also set high. 2 Cache Enable:Caching this region of memory is inhibited if this bit is cleared. 1 Write Enable:Write operations to this region of memory are allowed if this bit is set high. If this bit is cleared, then write operations in this region are directed to the PCI master. 0 Read Enable:Read operations to this region of memory are allowed if this bit is set high. If this bit is cleared then read operations in this region are directed to the PCI master. Note: If Cache Enable = 1 and Write Enable = 1, the Write Enable determination occurs after the data has passed the cache. Since the cache does write update, write data will change the cache if the address is cached. If a read then occurs to that address. the data will come from the written data that is in the cache even though the address is not writable. If this must be avoided then do not make the region cacheable.

Revision 3.1 103 www.national.com Integrated Functions (Continued) Geode™ GXm Processor

4.3 MEMORY CONTROLLER

The memory controller operates with the Processor Inter- face (X-Bus), Display Controller Interface, Graphics Pipe- line Interface, and the SDRAM Interface. The GXm processor supports LVTTL (low voltage TTL) technology. LVTTL technology allows the SDRAM inter- face of the memory controller to run at frequencies up to 100 MHz. The SDRAM clock is a function of the core clock. The SDRAM bus can be run at speeds that range between 66 MHz and 100 MHz. The core clock can be divided down from two to five in half clock increments to generate the SDRAM clock. SDRAM frequencies between 79 MHz and

100 MHz are only supported for certain types of closed

systems and strict design rules must be adhered to. For further details, please contact your local National Semi- conductor technical support representative. A basic block diagram of the memory controller is shown in Figure 4-3. Figure 4-3. Memory Controller Block Diagram Address Processor/PCI Display Controller Graphics Pipeline Processor/PCI Address Processor I/F Display Controller I/F Graphics Pipeline I/F Arbiter SDRAM RASA#,RASB# CKEA#, CKEB# WEA#/WEB# Configuration MA[12:0] BA[1:0] Display Controller Address Graphics Pipeline Address Processor/PCI Data Display Controller Data Graphics Pipeline Data Processor/PCI Display Controller MD[63:0] Read Buffer (16 Bytes) Sequence Controller Timing Controller Registers Control/MUX Write Buffer (16 Bytes) Write Buffer (16 Bytes) Graphics Controller Write Buffer (16 Bytes) Control Control Control DQM[7:0] CASA#,CASB# CS[3:0]# REF Clock Divider 2, 2.5, 3, 3.5, 4 SDCLK[3:0]Core Clock (ph2)

www.national.com 104 Revision 3.1 Integrated Functions (Continued) Geode™ GXm Processor

4.3.1 Memory Array Configuration

The memory controller supports up to two 64-bit, 168-pin unbuffered SDRAM modules (DIMM). Each DIMM receives a unique set of RAS, CAS, WE, and CKE lines. Each DIMM can have one or two 64-bit DIMM banks. Each DIMM bank is selected by a unique chip select (CS). There are four chip select signals to choose between a total of four DIMM banks. Each DIMM bank also receives au n i q u eS D C L K .E a c hD I M Mb a n kc a nh a v et w oo rf o u r component banks. Component bank selection is done through the bank address (BA) lines. For example, 16 Mb SDRAMS have two component banks and 64 Mb SDRAMs have two or four component banks. For single DIMM bank modules, the memory con- troller can support two DIMMS with a maximum of eight component banks. For dual DIMM bank modules, the memory controller can support two DIMMs with a maxi- mum of 16 component banks. Up to 16 banks can be open at the same time. Figure 4-4. Memory Array Configuration MA[12:0] BA[1:0] MD[63:0] DQM[7:0] RASA# CASA# WEA# CS1# CS0# CKEA SDCLK0 SDCLK1 RASB# CASB# WEB# CS3# CS2# CKEB SDCLK2 SDCLK3 A[12:0] BA[1:0] MD[63:0] DQM[7:0] RAS# CAS# WE# S0#, S2# CKE0 CK0, CK2 A[12:0] BA[1:0] MD[63:0] DQM[7:0] RAS# CAS# WE# S1#, S3# CKE1 CK1, CK3 Bank 0 Bank 1 A[12:0] BA[1:0] MD[63:0] DQM[7:0] RAS# CAS# WE# S0#, S2# CKE0 CK0, CK2 A[12:0] BA[1:0] MD[63:0] DQM[7:0] RAS# CAS# WE# S1#, S3# CKE1 CK1, CK3 Bank 0 Bank 1 DIMM 1 DIMM 0 Geode™ GXm Processor

Revision 3.1 105 www.national.com Integrated Functions (Continued) Geode™ GXm Processor

4.3.2 Memory Organizations

The memory controller supports JEDEC standard synchronous DRAMs in 16 Mb and 64 Mb configurations. Supported configurations are shown in Table 4-12. Table 4-12. Synchronous DRAM Configurations Depth Organization Row Address Column Address Bank Address Total # of Address bits 1 1 Mx16 A10-A0 A7-A0 BA0 20 2 2 Mx8 A10-A0 A8-A0 BA0 21

2 Mx32 A10-A0 A7-A0 BA1-BA0 21

2 Mx32 A10-A0 A8-A0 BA0 21

2 Mx32 A11-A0 A6-A0 BA1-BA0 21

2 Mx32 A12-A0 A6-A0 BA0 21

4 4 Mx4 A10-A0 A9-A0 BA0 22

4 Mx16 A11-A0 A7-A0 BA1-BA0 22

4 Mx16 A12-A0 A7-A0 BA0 22

4 Mx16 A10-A0 A9-A0 BA0 22

8 8 Mx8 A11-A0 A8-A0 BA1-BA0 23

8 Mx8 A12-A0 A8-A0 BA0 23

8 Mx32 A11-A0 A8-A0 BA1-BA0 23

8 Mx32 A12-A0 A7-A0 BA1-BA0 23

16 16 Mx4 A11-A0 A9-A0 BA1-BA0 24

16 Mx4 A12-A0 A9-A0 BA0 24

16 Mx16 A12-A0 A8-A0 BA1-BA0 24

16 Mx16 A11-A0 A9-A0 BA1-BA0 24

32 32 Mx8 A12-A0 A9-A0 BA1-BA0 25 64 64 Mx4 A12-A0 A9-A0,A11 BA1-BA0 26

www.national.com 106 Revision 3.1 Integrated Functions (Continued) Geode™ GXm Processor

4.3.3 SDRAM Commands

This subsection discusses the SDRAM commands sup- ported by the memory controller. Table 4-13 summarizes these commands followed by detailed operational infor- mation regarding each command. MRS — The Mode Register command defines the specific mode of operation of the SDRAM. This definition includes the selection of burst length, burst type, and CAS latency. CAS latency is the delay, in clock cycles, between the reg- istration of a read command and the availability of the first piece of output data. The burst length is programmed by address bits MA[2:0], the burst type by address bit MA3 and the CAS latency by address bits MA[6:4]. The memory controller only supports a burst length of two and burst type of interleave. The field value on MA[12:0] and BA[1:0] during the MRS c y c l ea r ea ss h o w ni nT a b l e4 - 1 4 . PRE — The precharge command is used to deactivate the open row in a particular bank or the open row in both component banks. Address pin MA10 determines whether one or both banks are to be precharged. In the case where only one component bank is to be precharged, BA[1:0] selects which bank. Once a bank has been pre- charged, it is in the Idle state and must be activated prior to any read or write commands. Table 4-13. Basic Command Truth Table Name Command CS RAS CAS WE MRS Mode Register Set L L L L PRE Bank Precharge L L H L ACT Bank activate/row- address entry LLH H WRT Column address entry/Write operation LHLL READ Column address entry/Read operation LHLH DESL Control input inhibit/ No operation HXXX REFR * CBR Refresh or Auto Refresh LLLH Note: *This command is CBR (CAS-before-RAS) refresh when CKE is high and self refresh when CKE is low. Table 4-14. Address Line Programming during MRS Cycles BA[1:0] MA[12:7] MA[6:4] MA3 MA2 MA1 MA0 00 000000 CAS Latency: 000 = Reserved 010 = 2 CLK 100 = 4 CLK 110 = 6 CLK 001 = 1 CLK 011 = 3 CLK 101 = 5 CLK 111 = 7 CLK 1001

Revision 3.1 107 www.national.com Integrated Functions (Continued) Geode™ GXm Processor ACT — The activate command is used to open a row in a particular bank for a subsequent access. The value on the BA lines selects the bank, and the address on the MA lines selects the row. This row remains open for accesses until a precharge command is issued to that bank. A pre- charge command must be issued before opening a differ- ent row in the same bank. READ —T h er e a dc o m m a n di su s e dt oi n i t i a t eab u r s t read access to an active row. The value on the BA lines select the component bank, and the address provided by the MA lines select the starting column location. The memory controller does not perform auto precharge dur- ing read operations. Valid data-out from the starting col- umn address is available following the CAS latency after the read command. The DQM signals are asserted low during read operations. WRT — The write command is used to initiate a burst w r i t ea c c e s st oa na c t i v er o w .T h ev a l u eo nt h eB Al i e n s select the component bank, and the address provided by the MA lines select the starting column location. The memory controller does not perform auto precharge dur- ing write operations. This leaves the page open for subse- quent accesses. Data appearing on the MD lines is written to the DQM logic level appearing coincident with the data. If the DQM signal is registered low, the corre- sponding data will be written to memory. If the DQM is driven high, the corresponding data will be ignored, and a write will not be executed to that location. REF — Auto refresh is used during normal operation and is analogous to the CAS-before-RAS (CBR) refresh in conventional DRAMs.During auto refresh the address bits are "don’t care". The memory controller precharges all banks prior to an auto refresh cycle. Auto refresh cycles are issued approximately 15 µs apart. The self refresh command is used to retain data in the SDRAMs even when the rest of the system is powered down. The self refresh command is similar to an auto refresh command except CKE is disabled (low). The memory controller issues a self refresh command during 3V Suspend mode when all the internal clocks are stopped.

4.3.3.1 SDRAM Initialization Sequence

After the clocks have started and stabilized, the memory controller SDRAM initialization sequence begins: 1) Precharge all component banks 2) Perform eight refresh cycles 3) Perform an MRS cycle 4) Perform eight refresh cycles This sequence is compatible with the majority of SDRAMs available from the various vendors.

www.national.com 108 Revision 3.1 Integrated Functions (Continued) Geode™ GXm Processor

4.3.4 Memory Controller Register Description

The Memory Controller maps 100h locations starting at GX_BASE+8400h. Refer to Section 4.1.2 “Control Regis- ters” on page 94 for instructions on accessing these regis- ters. Table 4-15 summarizes the 32-bit registers contained in the memory controller. Table 4-16 gives detailed regis- ter/bit formats. Table 4-15. Memory Controller Register Summary GX_BASE+ Memory Offset Type Name/Function Default Value 8400h-8403h R/W MC_MEM_CNTRL1 Memory Controller Control Register 1: Memory controller configuration informa- tion e.g., refresh interval, SDCLK ratio, etc. 248C0040h 8404h-8407h R/W MC_MEM_CNTRL2 Memory Controller Control Register 2: Memory controller configuration informa- tion to control SDCLK. 00000801h 8408h-840Bh R/W MC_BANK_CFG Memory Controller Bank Configuration: Contains the configuration information for the each of the two DIMMs in the memory array. BIOS programs this register dur- ing boot by running an autosizing routine on the memory. 41104110h 840Ch-840Fh R/W MC_SYNC_TIM1 Memory Controller Synchronous Timing Register 1: SDRAM memory timing information - This register controls the memory timing of all four banks of DRAM. BIOS programs this register based on the processor frequency and the SDCLK divide ratio. 2A733225h 8414h-8417h R/W MC_GBASE_ADD Memory Controller Graphics Base Address Register: This register sets the graphics memory base address, which is programmable on 512 KB boundaries. The display controller and the graphics pipeline generate a 20-bit DWORD offset that is added to the graphics memory base address to form the physical memory address. Typically, the graphics memory region is located at the top of physical memory. 00000000h 8418h-841Bh R/W MC_DR_ADD Memory Controller Dirty RAM Address Register: This register is used to set the Dirty RAM address index for processor diagnostic access. This register should be initialized before accessing the MC_DR_ACC register 00000000h 841Ch-841Fh R/W MC_DR_ACC Memory Controller Dirty RAM Access Register: This register is used to access the Dirty RAM. A read/write to this register will access the Dirty RAM at the address specified in the MC_DR_ADD register. 0000000xh Table 4-16. Memory Controller Registers Bit Name Description GX_BASE+ 8400h-8403h MC_MEM_CNTRL1 (R/W) Default Value = 248C0040h 31:29 MDHDCTL MD High Drive Control:Controls the high drive and slew rate of the memory data bus (MD[63:0]): 000 = T ristate 001 = Smallest drive strength 010 -110 = Represents gradual drive strength increase 111 = Highest drive strength 28:26 MABAHDCTL MA/BA High Drive Control:Controls the high drive and slew rate of the memory address bus includ- ing the memory bank address bus (MA[12:0] and BA[1:0]): 000 = T ristate 001 = Smallest drive strength 010 -110 = Represents gradual drive strength increase 111 = Highest drive strength

Revision 3.1 109 www.national.com Integrated Functions (Continued) Geode™ GXm Processor 25:23 MEMHDCTL Control High Drive/Slew Control:Controls the high drive and slew rate of the memory control sig- nals (CASA#, CASB#, RASA#, RASB#, CKEA, CKEB, WEA#, WEA#, DQM[7:0], and CS[3:0]#): 000 = T ristate 001 = Smallest drive strength 010 -110 = Represents gradual drive strength increase 111 = Highest drive strength 22 RSVD Reserved: Set to 0. 21 RSVD Reserved: Must be set to 0. Wait state on the X-Bus x_data during read cycles - for debug only. 20:18 SDCLKRATE SDRAM Clock Ratio:Selects SDRAM clock ratio: 000 = Reserved 100 = ÷ 3.5 001 = ÷ 2 101 = ÷ 4 011 = ÷ 3 (Default) 111 = ÷ 5 Ratio does not take effect until the SDCLKSTRT bit (bit 17 of this register) transitions from 0 to 1. 17 SDCLKSTRT Start SDCLK:Start operating SDCLK using the new ratio and shift value (selected in bits [20:18] of this register): 0 = Clear; 1 = Enable. This bit should be cleared every time before a one is written to it in order to start SDCLK or to change the shift value. 16:8 RFSHRATE Refresh Interval:This field determines the number of processor core clocks multiplied by 64 between refresh cycles to the DRAM. By default, the refresh interval is 00h. This implies that refresh is turned off by default. 7:6 RFSHSTAG Refresh Staggering:This field determines number of clocks between REF commands to different banks during refresh cycles: 00 = 0 SDRAM clocks 10 = 2 SDRAM clocks 01 = 1 SDRAM clocks (Default) 11 = 4 SDRAM clocks Staggering is used to help reduce power spikes during refresh. When only DIMM0 is installed and it has only one DIMM bank, then this field must be set to 00. 5 2CLKADDR Two Clock Address Setup:Assert memory address for one extra clock before CS# is asserted: 0 = Disable; 1 = Enable. This can be used to compensate for address setup at high frequencies. 4R F S H T S T Test Refresh:This bit, when set high, generates a refresh request. This bit is only used for testing purposes. 3 XBUSARB X-Bus Round Robin:When enabled, processor requests are arbitrated at the same priority level than graphics pipeline requests and non-critical display controller requests. When disabled, processor requests are arbitrated at a higher priority level. High priority display controller requests always have the highest arbitration priority: 0 = Enable; 1 = Disable. 2V G A W R P VGA Wrap Enable: Allow memory wrapping into the VGA memory address space from A0000h to BFFFFh: 0 = Disable; 1 = Enable. 1R S V D Reserved: Set to 0. 0 SDRAMPRG Program SDRAM: When this bit is set the memory controller will program the SDRAM MRS register using LTMODE in MC_SYNC_TIM1. This bit should be cleared every time before a one is written to it in order to program the SDRAM. GX_BASE+8404h-8407h MC_MEM_CNTRL2 (R/W) Default Value = 00000801h 31:18 RSVD Reserved: Set to 0. 17:16 SDCLKRISE SDCLK Rising Delay:Controls the delay between the core clock and the rising edge of SDCLK dur- ing all modes. (Set by BIOS.) 15:14 SDCLKFALL SDCLK Falling Delay:Controls the delay between the core clock and the falling edge of SDCLK dur- ing 2.5 and 3.5 clock modes. (Set by BIOS.) 13:11 SDCLKHDCTL SDCLK High Drive/Slew Control:Controls the high drive and slew rate of SDCLK[3:0] and SDCLK_OUT . 000 = Highest drive strength. (No braking applied in the pads) 001 = Smallest drive strength 010 -110 = Represent gradual drive strength increase 111 = Highest drive strength 10 SDCLKOMSK Mask SDCLK_OUT: 0 = Not masked; 1 = Mask. Table 4-16. Memory Controller Registers (Continued) Bit Name Description

www.national.com 110 Revision 3.1 Integrated Functions (Continued) Geode™ GXm Processor 9 SDCLK3MSK Mask SDCLK3: 0 = Not masked; 1 = Mask.

8 SDCLK2MSK Mask SDCLK2: 0 = Not masked; 1 = Mask

7 SDCLK1MSK Mask SDCLK1: 0 = Not masked; 1 = Mask.

6 SDCLK0MSK Mask SDCLK0: 0 = Not masked; 1 = Mask

5:3 SHFTSDCLK Shift SDCLK:This function allows shifting SDCLK to meet SDRAM setup and hold time requirements. The shift function will not take effect until the SDCLKSTRT bit (bit 17 of MC_MEM_CNTRL1) transi- tions from 0 to 1: 000 = No shift 100 = Shift 2 core clocks 001 = Shift 0.5 core clock 101 = Shift 2.5 core clocks 010 = Shift 1 core clock 110 = Shift 3 core clocks 011 = Shift 1.5 core clock 111 = Reserved Note: Refer to Figure 4-10 for an example of SDCLK shifting. 2R S V D Reserved: Set to 0. 1R D Read Data Phase:Selects if read data is latched one or two core clock after the rising edge of S D C L K :0=1c o r ec l o c k ;1=2c o r ec l o c k s . 0 FSTRDMSK Fast Read Mask:Do not allow core reads to bypass the request FIFO: 0 = Disable; 1 = Enable. GX_BASE+8408h-840Bh MC_BANK_CFG (R/W) Default Value = 41104110h 31 RSVD Reserved: Set to 0.

30 DIMM1_

MOD_BNK DIMM1 Module Banks: Selects the number of module banks per DIMM for DIMM1: 0 = 1 Module bank 1 = 2 Module banks 29 RSVD Reserved: Set to 0.

28 DIMM1_

COMP_BNK DIMM1 Component Banks: Selects the number of component banks per module bank for DIMM1: 0 = 2 Component banks 1 = 4 Component banks 27 RSVD Reserved: Set to 0. 26:24 DIMM1_SZ DIMM1 Size: Selects the size of DIMM1: 000 = 4 MB 010 = 16 MB 100 = 64 MB 110 = 256 MB 001 = 8 MB 011 = 32 MB 101 = 128 MB 111 = 512 MB 23 RSVD Reserved: Set to 0. 22:20 DIMM1_PG_SZ DIMM1 Page Size:Selects the page size of DIMM1: 000 = 1 KB 010 = 4 KB 1xx = 16 KB 001 = 2 KB 011 = 8 KB 111 = DIMM1 not installed When DIMM1 is not installed, program all other DIMM1 fields to 0. 19:15 RSVD Reserved: Set to 0.

14 DIMM0_

MOD_BNK DIMM0 Module Banks: Selects number of module banks per DIMM for DIMM0: 0 = 1 Module bank 1 = 2 Module banks 13 RSVD Reserved: Set to 0.

12 DIMM0_

COMP_BNK DIMM0 Component Banks: Selects the number of component banks per module bank for DIMM0: 0 = 2 Component banks 1 = 4 Component banks 11 RSVD Reserved: Set to 0. 10:8 DIMM0_SZ DIMM0 Size:Selects the size of DIMM1: 000 = 4 MB 010 = 16 MB 100 = 64 MB 110 = 256 MB 001 = 8 MB 011 = 32 MB 101 = 128 MB 111 = 512 MB 7R S V D Reserved: Set to 0. 6:4 DIMM0_PG_SZ DIMM0 Page Size:Selects the page size of DIMM0: 000 = 1 KB 010 = 4 KB 1xx = 16 KB 001 = 2 KB 011 = 8 KB 111 = DIMM0 not installed When DIMM0 is not installed, program all other DIMM0 fields to 0. 3:0 RSVD Reserved: Set to 0. Table 4-16. Memory Controller Registers (Continued) Bit Name Description

Revision 3.1 111 www.national.com Integrated Functions (Continued) Geode™ GXm Processor GX_BASE+840Ch-840Fh MC_SYNC_TIM1 (R/W) Default Value = 2A733225h 31 RSVD Reserved: Set to 0. 30:28 LTMODE CAS Latency (LTMODE): CAS latency is the delay, in clock cycles, between the registration of a read command and the availability of the first piece of output data (BIOS interrogates EEPROM across the I2C interface to determine this value): 000 = Reserved 010 = 2 CLK 100 = 4 CLK 110 = 6 CLK 001 = 1 CLK 011 = 3 CLK 101 = 5 CLK 111 = 7 CLK This field will not take effect until SDRAMPRG (bit 0 of MC_MEM_CNTRL1) transitions from 0 to 1. ERRATA: CAS Latency of 1 CLK is not currently supported. 27:24 RC REF to REF/ACT Command Period (tRC):Minimum number of SDRAM clock between REF and REF/ACT commands: 0000 = Reserved 0100 = 5 CLK 1000 = 9 CLK 1100 = 13 CLK 0001 = 2 CLK 0101 = 6 CLK 1001 = 10 CLK 1101 = 14 CLK 0010 = 3 CLK 0110 = 7 CLK 1010 = 11 CLK 1110 = 15 CLK 0011 = 4 CLK 0111 = 8 CLK 1011 = 12 CLK 1111 = 16 CLK 23:20 RAS ACT to PRE Command Period (tRAS):Minimum number of SDRAM clocks between ACT and PRE commands: 0000 = Reserved 0100 = 5 CLK 1000 = 9 CLK 1100 = 13 CLK 0001 = 2 CLK 0101 = 6 CLK 1001 = 10 CLK 1101 = 14 CLK 0010 = 3 CLK 0110 = 7 CLK 1010 = 11 CLK 1110 = 15 CLK 0011 = 4 CLK 0111 = 8 CLK 1011 = 12 CLK 1111 = 16 CLK 19 RSVD Reserved: Set to 0. 18:16 RP PRE to ACT Command Period (tRP):Minimum number of SDRAM clocks between PRE and ACT commands: 000 = Reserved 010 = 2 CLK 100 = 4 CLK 110 = 6 CLK 001 = 1 CLK 011 = 3 CLK 101 = 5 CLK 111 = 7 CLK 15 RSVD Reserved: Set to 0. 14:12 RCD Delay Time ACT to READ/WRT Command (tRCD): Minimum number of SDRAM clock between ACT and READ/WRT commands: 000 = Reserved 010 = 2 CLK 100 = 4 CLK 110 = 6 CLK 001 = 1 CLK 011 = 3 CLK 101 = 5 CLK 111 = 7 CLK 11 RSVD Reserved: Set to 0. 10:8 RRD ACT(0) to ACT(1) Command Period (tRRD):Minimum number of SDRAM clocks between ACT and ACT command to two different component banks within the same module bank. The memory control- ler does not perform back-to-back Activate commands to two different component banks without a READ or WIRTE command between them. Hence, this field should be set to 001. 7R S V D Reserved: Set to 0. 6:4 DPL Data-in to PRE command period (tDPL):Minimum number of SDRAM clocks from the time the last write datum is sampled till the bank is precharged: 000 = Reserved 010 = 2 CLK 100 = 4 CLK 110 = 6 CLK 001 = 1 CLK 011 = 3 CLK 101 = 5 CLK 111 = 7 CLK 3:0 RSVD Reserved: Set to 0 or leave unchanged. GX_BASE+8414h-8417h MC_GBASE_ADD (R/W) Default Value = 00000000h 31:18 RSVD Reserved: Set to 0.

17 TE Test Enable TEST[3:0]:

0 = TEST[3:0] are driven low 1 = TEST[3:0] pins are used to output test information

16 TECTL Test Enable Shared Control Pins:

0 = RASB#, CASB#, CKEB, WEB# are driven low 1 = RASB#, CASB#, CKEB, WEB# are used to output test information 15:12 SEL Select:This field is used for debug purposes only. 11 RSVD Reserved: Set to 0. Table 4-16. Memory Controller Registers (Continued) Bit Name Description

www.national.com 112 Revision 3.1 Integrated Functions (Continued) Geode™ GXm Processor

4.3.5 Address Translation

The memory controller supports two address translations depending on the method used to interleave pages.

4.3.5.1 High Order Interleaving

High Order Interleaving (HOI) uses the most significant address bits to select which bank the page is located in. This has the effect of allowing any mixture of DIMM types. However, it spreads the pages over wide address ranges. For example, two 8 MB DIMMs contain a total of four com- ponent pages. Two pages are together in one DIMM sepa- rated from the other two pages by 8 MB.

4.3.5.2 Low Order Interleaving

Low Order Interleaving (LOI) uses the least significant bits after the page bits to select which bank the page is located in. This requires that memory is a power of 2, that the number of banks is a power of 2, and that the page sizes are the same. In other words, the DIMMs have to be of the same type. However, LOI does give a good benefit by providing a moving page throughout memory. Using the same example as above, two banks would be on one D I M Ma n dt h en e x tt w ob a n k sw o u l db eo nt h es e c o n d DIMM, but they would be linear in address space. For an eight bank system that has 1 KB address (8 KB data) pages, there would be an effective moving page of 64 KB of data.

4.3.5.3 Physical Address to DRAM Address

Auto LOI is in effect whenever the two DIMMs have the same number of DIMM banks, component banks, module sizes and page sizes. T a b l e s4 - 1 7a n dT a b l e4 - 1 8o np a g e1 1 3g i v eA u t oL O I address conversion examples when two DIMMs of the same size are used in a system. Table 4-17 shows a one DIMM bank conversion example, while Table 4-18 shows a two DIMM bank example. Table 4-19 and T able 4-20 on page 114 give Non-Auto LOI address conversion examples when either one or two DIMMs of different sizes are used in a system. Table 4-19 shows a one DIMM bank address conversion example, while Table 4-20 shows a two DIMM bank example. The addresses are computed on a per DIMM basis. Since the DRAM interface is 64 bits wide, the lower three bits of the physical address get mapped onto the DQM[7:0] lines. Thus, the address conversion tables (Tables 4-17 through 4-20) show the physical address starting from A3. 10:0 GBADD Graphics Base Address:This field indicates the graphics memory base address, which is program- mable on 512 KB boundaries. This field corresponds to address bits [29:19]. Note that BC_DRAM_TOP must be set to a value lower than the Graphics Base Address. GX_BASE+8418h-841Bh MC_DR_ADD (R/W) Default Value = 00000000h 31:10 RSVD Reserved: Set to 0. 9:0 DRADD Dirty RAM Address:This field is the address index that is used to access the Dirty RAM with the MC_DR_ACC register. This field does not auto increment. GX_BASE+841Ch-841Fh MC_DR_ACC (R/W) Default Value = 0000000xh 31:2 RSVD Reserved: Set to 0. 1D Dirty Bit:This bit is read/write accessible. 0V Valid Bit:This bit is read/write accessible. Table 4-16. Memory Controller Registers (Continued) Bit Name Description

Revision 3.1 113 www.national.com Integrated Functions (Continued) Geode™ GXm Processor Table 4-17. Auto LOI -- 2 DIMMs, Same Size, 1 DIMM Bank

1 K Page Size 2 K Page Size 4 K Page Size 1 K Page Size 2 K Page Size 4 Page Size

Row Col Row Col Row Col Row Col Row Col Row Col Address 2 Component Banks 4 Component Banks MA12 A24 -- A25 -- A26 A25 -- A26 -- A27 MA11 A23 -- A24 -- A25 A24 -- A25 -- A26 MA10 A22 -- A23 -- A24 A23 -- A24 -- A25 MA9 A21 -- A22 -- A23 A22 A9 A23 -- A24 MA8 A20 -- A21 -- A22 A11 A21 A8 A22 -- A23 A11 MA7 A19 -- A20 A10 A21 A10 A20 A7 A21 A10 A22 A10 MA6 A18 A9 A19 A9 A20 A9 A19 A6 A20 A9 A21 A9 MA5 A17 A8 A18 A8 A19 A8 A18 A5 A19 A8 A20 A8 MA4 A16 A7 A17 A7 A18 A7 A17 A4 A18 A7 A19 A7 MA3 A15 A6 A16 A6 A17 A6 A16 A3 A17 A6 A18 A6 MA2 A14 A5 A15 A5 A16 A5 A15 A8 A16 A5 A17 A5 MA1 A13 A4 A14 A4 A15 A4 A14 A7 A15 A4 A16 A4 MA0 A12 A3 A13 A3 A14 A3 A13 A6 A14 A3 A15 A3 CS0#/CS1# A11 A12 A13 A12 A13 A14 BA0/BA1 A10 A11 A12 A11/A10 A12/A11 A13/A12 T a b l e4 - 1 8 . A u t oL O I- -2D I M M s ,S a m eS i z e ,2D I M MB a n k s

1 K Page Size 2 K Page Size 4 K Page Size 1 K Page Size 2 Page Size 4 K Page Size

Row Col Row Col Row Col Row Col Row Col Row Col Address 2 Component Banks 4 Component Banks MA12 A25 -- A26 -- A27 A26 -- A27 -- A28 -- MA11 A24 -- A25 -- A26 A25 -- A26 -- A27 -- MA10 A23 -- A24 -- A25 A24 -- A25 -- A26 -- MA9 A22 -- A23 -- A24 A23 -- A24 -- A25 -- MA8 A21 -- A22 -- A23 A11 A22 -- A23 -- A24 A11 MA7 A20 -- A21 A10 A22 A10 A21 -- A22 A10 A23 A10 MA6 A19 A9 A20 A9 A21 A9 A20 A9 A21 A9 A22 A9 MA5 A18 A8 A19 A8 A20 A8 A19 A8 A20 A8 A21 A8 MA4 A17 A7 A18 A7 A19 A7 A18 A7 A19 A7 A20 A7 MA3 A16 A6 A17 A6 A18 A6 A17 A6 A18 A6 A19 A6 MA2 A15 A5 A16 A5 A17 A5 A16 A5 A17 A5 A18 A5 MA1 A14 A4 A15 A4 A16 A4 A15 A4 A16 A4 A17 A4 MA0 A13 A3 A14 A3 A15 A3 A14 A3 A15 A3 A16 A3 CS0#/CS1# A12 A13 A14 A13 A14 A15 CS2#/CS3# A11 A12 A13 A12 A13 A14 BA0/BA1 A10 A11 A12 A11/A10 A12/A11 A13/A12

www.national.com 114 Revision 3.1 Integrated Functions (Continued) Geode™ GXm Processor Table 4-19. Non-Auto LOI -- 1 or 2 DIMMs, Different Sizes, 1 DIMM Bank

1 K Page Size 2 K Page Size 4 K Page Size 1 K Page Size 2 K Page Size 4 K Page Size

Row Col Row Col Row Col Row Col Row Col Row Col Address 2 Component Banks 4 Component Banks MA12 A23 -- A24 -- A25 -- A24 -- A25 -- A26 MA11 A22 -- A23 -- A24 -- A23 -- A24 -- A25 MA10 A21 -- A22 -- A23 -- A22 -- A23 -- A24 MA9 A20 -- A21 -- A22 -- A21 -- A22 -- A23 MA8 A19 -- A20 -- A21 A11 A20 -- A21 -- A22 A11 MA7 A18 -- A19 A10 A20 A10 A19 -- A20 A10 A21 A10 MA6 A17 A9 A18 A9 A19 A9 A18 A9 A19 A9 A20 A9 MA5 A16 A8 A17 A8 A18 A8 A17 A8 A18 A8 A19 A8 MA4 A15 A7 A16 A7 A17 A7 A16 A7 A17 A7 A18 A7 MA3 A14 A6 A15 A6 A16 A6 A15 A6 A16 A6 A17 A6 MA2 A13 A5 A14 A5 A15 A5 A14 A5 A15 A5 A16 A5 MA1 A12 A4 A13 A4 A14 A4 A13 A4 A14 A4 A15 A4 MA0 A11 A3 A12 A3 A13 A3 A12 A3 A13 A3 A14 A3 BA0/BA1 A10 A11 A12 A11/A10 A12/A11 A13/A12 Table 4-20. Non-Auto LOI -- 1 or 2 DIMMs, Different Sizes, 2 DIMM Banks Row Col Row Col Row Col Row Col Row Col Row Col Address 2 Component Banks 4 Component Banks MA8 A20 -- A21 -- A22 A11 A21 -- A22 -- A23 A11 MA7 A19 -- A20 A10 A21 A10 A20 -- A21 A10 A22 A10 MA6 A18 A9 A19 A9 A20 A9 A19 A9 A20 A9 A21 A9 MA5 A17 A8 A18 A8 A19 A8 A18 A8 A19 A8 A20 A8 MA4 A16 A7 A17 A7 A18 A7 A17 A7 A18 A7 A19 A7 MA3 A15 A6 A16 A6 A17 A6 A16 A6 A17 A6 A18 A6 MA2 A14 A5 A15 A5 A16 A5 A15 A5 A16 A5 A17 A5 MA1 A13 A4 A14 A4 A15 A4 A14 A4 A15 A4 A16 A4 MA0 A12 A3 A13 A3 A14 A3 A13 A3 A14 A3 A15 A3 CS0#/CS1# A11 A12 A13 A12 A13 A14 CS2#/CS3# -- -- BA0/BA1 A10 A11 A12 A11/A10 A12/A11 A13/A12

Revision 3.1 115 www.national.com Integrated Functions (Continued) Geode™ GXm Processor

4.3.6 Memory Cycles

Figures 4-5 through Figure 4-8 on page 117 illustrate vari- ous memory cycles that the memory controller supports. The following subsections describe some of the sup- ported cycles. SDRAM Read Cycle F i g u r e4 - 5s h o w saS D R A Mr e a dc y c l e .T h ef i g u r e assumes that a previous ACT command has presented the row address for the read operation. Note that the burst length for the READ command is always two. Figure 4-5. Basic Read Cycle with a CAS Latency of Two SDCLK CS# RAS# CAS# WE# DQM MD MA COL n nn + 1

www.national.com 118 Revision 3.1 Integrated Functions (Continued) Geode™ GXm Processor

4.3.7 SDRAM Interface Clocking

The GXm processor drives the SDCLK to the SDRAMs; one for each DIMM bank. All the control, data, and address signals driven by the memory controller are sam- pled by the SDRAM at the rising edge of SDCLK. SDCLK- OUT is a reference signal used to generate SDCLKIN. Read data is sampled by the memory controller at the ris- ing edge of SDCLKIN. The delay for SDCLKIN must be designed so that it lags the SDCLKs at the DRAM by approximately 2ns. The delay should also include the SDCLK transmission line delay. The SDCLK traces on the board need to be laid out so there is no skew between each of the four sinks. These guidelines allow the memory interface to be closer to the DRAM specifications. They improve performance by run- ning the SDCLK up to frequencies of 100 MHz and a CAS latency of two. Figure 4-9. SDCLKIN Clocking DIMM DIMM SDCLK[3:0] Delay SDCLKOUT SDCLKIN SDCLK0 SDCLK1 SDCLK2 SDCLK3 Geode™ GXm Processor

www.national.com 120 Revision 3.1 Integrated Functions (Continued) Geode™ GXm Processor

4.4 GRAPHICS PIPELINE

The graphics pipeline of the GXm processor includes a BitBLT/vector engine which has been optimized for Microsoft Windows. The hardware supports pattern gen- eration, source expansion, pattern/source transparency, and 256 ternary raster operations. The block diagram of the graphics pipeline is shown in Figure 4-11.

4.4.1 BitBLT/Vector Engine

BLTs are initiated by writing to the GP_BLT_MODE regis- ter, which specifies the type of source data (none, frame buffer, or BLT buffer), the type of the destination data (none, frame buffer, or BLT buffer), and a source expan- sion flag. Figure 4-11. Graphics Pipeline Block Diagram Pattern Hardware Raster Operation Output Aligner BE PAT SRC DST BE Internal Bus Interface Unit Graphics Scratchpad RAM BitBLT Buffers and Memory X-Bus C-Bus Pipeline BE = Byte Enable PAT = Pattern Data SRC = Source Data DST = Destination Data Output Aligner Source Expansion Control Logic DRAM Interface Register Access Controller Key:

Revision 3.1 121 www.national.com Integrated Functions (Continued) Geode™ GXm Processor The BLT buffers in the dedicated cache temporarily store s o u r c ea n dd e s t i n a t i o nd a t a ,t y p i c a l l yo nas c a nl i n eb a s i s . The hardware automatically loads frame-buffer data (source or destination) into the BLT buffers for each scan line. The software is responsible for making sure that this does not overflow the memory allocated for the BLT buff- ers. When the source data is a bitmap, the data is loaded directly into the BLT buffer before starting the BLT. Vectors are initiated by writing to the GP_VECTOR_MODE register (GX_BASE+8204h), which specifies the direction of the vector and a “read destina- tion data” flag. If the flag is set, the hardware will read destination data along the vector and store it temporarily in BLT Buffer 0.

4.4.2 Master/Slave Registers

When starting a BitBLT or vector operation, the graphics pipeline registers are latched from the master registers to the slave registers. A second BitBLT or vector operation can then be loaded into the master registers while the first operation is rendered. If a second BLT is pending in the master registers, any write operations to the graphics pipeline registers will corrupt the values of the pending BLT . Software must prevent this from happening by check- ing the “BLT Pending” bit in the GP_BLT_ST ATUS register (GX_BASE+820Ch[2]. Most of the graphics pipeline registers are latched directly from the master registers to the slave registers when starting a new BitBLT or vector operation. Some registers, however, use the updated slave values if the master regis- ters have not been written, which allows software to ren- der successive primitives without loading some of the registers as outlined in Table 4-21.

4.4.3 Pattern Generation

The graphics pipeline contains hardware support for 8x8 monochrome patterns (expanded to two colors), 8x8 dither patterns (expanded to four colors), and 8x1 color patterns. The pattern hardware, however, does not main- tain a pattern origin, so the pattern data must be justified before it is loaded into the GXm processor’s registers. For solid primitives, the pattern hardware is disabled and the pattern color is always sourced from the GP_PAT_COLOR_0 register (GX_BASE+8110h). Table 4-21. Graphics Pipeline Registers Master Function GP_DST_XCOOR Next X position along vector. Master register if written, otherwise: Unchanged slave if BLT , source mode = bitmap. Slave + width if BLT, source mode = text glyph GP_DST_YCOOR Next Y position along vector. Master register if written, otherwise: Slave +/- height if BLT, source mode = bitmap. Unchanged slave if BLT , source mode = text glyph. GP_INIT_ERROR Master register if written, otherwise: Initial error for the next pixel along the vector. GP_SRC_YCOOR Master register if written, otherwise: Slave +/- height if BLT, source mode = bitmap.

www.national.com 122 Revision 3.1 Integrated Functions (Continued) Geode™ GXm Processor

4.4.3.1 Monochrome Patterns

Monochrome patterns are selected by setting the pattern mode to 01b in the GP_RASTER_MODE register (GX_BASE+ 8200h). Those pixels corresponding to a clear bit (0) in the pattern are rendered using the color specified in the GP_PAT_COLOR_0 register, and those pixels corresponding to a set bit (1) in the pattern are ren- dered using the color specified in the GP_PAT_COLOR_1 register (GX_BASE+8112h). If the pattern transparency bit is set high in the GP_RASTER_MODE register, those pixels corresponding to a clear bit in the pattern data are not drawn. Monochrome patterns use bits [63:0] of the pattern data. Bits [7:0] correspond to the first row of the pattern, and bit 7 corresponds to the leftmost pixel on the screen. This is illustrated in Figure 4-12. Figure 4-12. Example of Monochrome Patterns

4.4.3.2 Dither Patterns

Dither patterns are selected by setting the pattern mode to 10 b in the GP_RASTER_MODE register (Table 4-25 on page 125). Two bits of pattern data are used for each pixel, allowing color expansion to four colors. The colors are specified in the GP_PAT_COLOR_0 through GP_PAT_COLOR_3 registers (Table 4-25 on page 125). Dither patterns use all 128 bits of pattern data. Bits [15:0] correspond to the first row of the pattern (the lower byte contains the LSB of the pattern color and the upper byte contains the MSB of the pattern color). This is illustrated in Figure 4-13. Figure 4-13. Example of Dither Patterns GP_PAT_DATA_0 = 0x80412214 GP_PAT_DATA_1 = 0x08142241 00AA 4411 00AA 1155 00AA 4411 00AA 1155 GP_PAT_DATA_0 = 0x441100AA GP_PAT_DATA_1 = 0x115500AA GP_PAT_DATA_2 = 0x441100AA GP_PAT_DATA_3 = 0x115500AA

Revision 3.1 123 www.national.com Integrated Functions (Continued) Geode™ GXm Processor

4.4.3.3 Color Patterns

Color patterns are selected by setting the pattern mode to 11 b in the GP_RASTER_MODE register. Bits [63:0] are used to hold a row of pattern data for an 8-BPP pattern, with bits [7:0] corresponding to the leftmost pixel of the row. Likewise, bits [127:0] are used for a 16-BPP color pattern, with bits [15:0] corresponding to the leftmost pixel of the row. To support an 8x8 color pattern, software must load the pattern data for each row.

4.4.4 Source Expansion

The graphics pipeline contains hardware support for color expansion of source data (primarily used for text). Those pixels corresponding to a clear bit (0) in the source data are rendered using the color specified in the GP_SRC_COLOR_0 register (GX_BASE+810Ch), and those pixels corresponding to a set bit (1) in the source data are rendered using the color specified in the GP_SRC_COLOR_1 register (GX_BASE+810Eh). If the source transparency bit is set in the GP_RASTER_MODE register, those pixels corresponding to a clear bit (0) in the source data are not drawn.

4.4.5 Raster Operations

The GP_RASTER_MODE register specifies how the pat- tern data, source data (color-expanded if necessary), and destination data are combined to produce the output from the graphics pipeline. The definition of the ROP valu matches that of the Microsoft API. This allows Windows display drivers to load the raster operation directly into hardware. T able 4-22 illustrates this definition. Some com- mon raster operations are described in Table 4-23. Table 4-22. GP_RASTER_MODE Bit Patterns Pattern (bit) Source (bit) Destination (bit) Output (bit) 00 0R O P [ 0 ] 00 1R O P [ 1 ] 01 0R O P [ 2 ] 01 1R O P [ 3 ] 10 0R O P [ 4 ] 10 1R O P [ 5 ] 11 0R O P [ 6 ] 11 1R O P [ 7 ] Table 4-23. Common Raster Operations ROP Description F0h Output = Pattern CCh Output = source 5Ah Output = Pattern XOR destination 66h Output = Source XOR destination 55h Output = ~Destination

www.national.com 124 Revision 3.1 Integrated Functions (Continued) Geode™ GXm Processor

4.4.6 Graphics Pipeline Register Descriptions

The graphics pipeline maps 200h locations starting at GX_BASE+8100h. Refer to Section 4.1.2 “Control Regis- ters” on page 94 for instructions on accessing these regis- ters. Table 4-24 summarizes the graphics pipeline registers and T able 4-25 on page 125 gives detailed register/bit for- mats. Table 4-24. Graphics Pipeline Configuration Register Summary GX_BASE+ Memory Offset Type Name / Function Default Value 8100h-8103h R/W GP_DST/START_Y/XCOOR Destination/Starting Y and X Coordinates Register — In BLT mode this register specifies the destination Y and X positions for a BLT operation. In Vector mode it specifies the starting Y and X positions in a vector. 00000000h 8104-8107h R/W GP_WIDTH/HEIGHT and GP_VECTOR_LENGTH/INIT_ERROR Width/Height or Vector Length/Initial Error Register — In BLT mode this register specifies the BLT width and height in pixels. In Vector mode it specifies the vector initial error and pixel length. 00000000h 8108h-810Bh R/W GP_SRC_X/YCOOR and GP_AXIAL/DIAG_ERROR Source X/Y Coordinate Axial/Diagonal Error Register — In BLT mode this register specifies the BLT X and Y source. In Vector mode it specifies the axial and diago- nal error for rendering a vector. 00000000h 810Ch-810Fh R/W GP_SRC_COLOR_0 and GP_SCR_COLOR_1 Source Color Register — Determines the colors used when expanding mono- chrome source data in either the 8-BPP mode or the 16-BPP mode. 00000000h 8110h-8113h R/W GP_PAT_COLOR_0 and GP_PAT_COLOR_1 Graphics Pipeline Pattern Color 0 and 1 Registers — These two registers deter- mine the colors used when expanding pattern data. 00000000h 8114h-8117h R/W GP_PAT_COLOR_2 and GP_PAT_COLOR_3 Graphics Pipeline Pattern Color 2 and 3 Registers — These two registers deter- mine the colors used when expanding pattern data. 00000000h 8120h-8123h R/W GP_PAT_DATA 0 through 3 Graphics Pipeline Pattern Data Registers 0 through 3 — Together these registers contain 128 bits of pattern data. GP_PAT_DATA_0 corresponds to bits [31:0] of the pattern data. GP_PAT_DATA_1 corresponds to bits [63:32] of the pattern data. GP_PAT_DATA_2 corresponds to bits [95:64] of the pattern data. GP_PAT_DATA_3 corresponds to bits [127:96] of the pattern data. 00000000h 8124h-8127h R/W 00000000h 8128h-812Bh R/W 00000000h 812Ch-812Fh R/W 00000000h 8140h-8143h (Note) R/W GP_VGA_WRITE Graphics Pipeline VGA Write Patch Control Register — Controls the VGA mem- ory write path in the graphics pipeline. xxxxxxxxh 8144h-8147h (Note) R/W GP_VGA_READ Graphics Pipeline VGA Read Patch Control Register — Controls the VGA mem- ory read path in the graphics pipeline. 00000000h 8200h-8203h R/W GP_RASTER_MODE Graphics Pipeline Raster Mode Register — This register controls the manipula- tion of the pixel data through the graphics pipeline. Refer to Section 4.4.5 “Raster Operations” on page 123. 00000000h 8204h-8207h R/W GP_VECTOR_MODE Graphics Pipeline Vector Mode Register — Writing to this register initiates the rendering of a vector. 00000000h 8208h-820Bh R/W GP_BLT_MODE Graphics Pipeline BLT Mode Register — Writing to this initiates a BLT operation. 00000000h Note: The registers at GX_BASE+8140, 8144h, 8210h, and 8217h are located in the area designated for the graphics pipeline but are used for VGA emulation purposes. Refer to Table 5-5 on page 173 for these register’s bit formats.

Revision 3.1 125 www.national.com Integrated Functions (Continued) Geode™ GXm Processor 820Ch-820Fh R/W GP_BLT_STATUS Graphics Pipeline BLT Status Register — Contains configuration and status infor- mation for the BLT engine. The status bits are contained in the lower byte of the register. 00000000h 8210h-8213h (Note) R/W GP_VGA_BASE Graphics Pipeline VGA Memory Base Address Register — Specifies the offset of the VGA memory, starting from the base of graphics memory. xxxxxxxxh 8214h-8217h (Note) R/W GP_VGA_LATCH Graphics Pipeline VGA Display Latch Register — Provides a memory mapped way to read or write the VGA display latch. xxxxxxxxh Table 4-24. Graphics Pipeline Configuration Register Summary (Continued) GX_BASE+ Memory Offset Type Name / Function Default Value Note: The registers at GX_BASE+8140, 8144h, 8210h, and 8217h are located in the area designated for the graphics pipeline but are used for VGA emulation purposes. Refer to Table 5-5 on page 173 for these register’s bit formats. Table 4-25. Graphics Pipeline Configuration Registers Bit Name Description GX_BASE+8100h-8103h GP_DST/START_X/YCOOR Register (R/W) Default Value = 00000000h 31:16 DESTINATION/STARTING Y POSITION (SIGNED): BLT Mode — Specifies the destination Y position for a BLT operation. Vector Mode — Specifies the starting Y position in a vector. 15:0 DESTINATION/STARTING X POSITION (SIGNED): BLT Mode — Specifies the destination X position for a BLT operation. Vector Mode — Specifies the starting X position in a vector. GX_BASE+8104h-8107h GP_WIDTH/HEIGHT and Default Value = 00000000h GP_VECTOR_LENGTH/INIT_ERROR Register (R/W) 31:16 PIXEL_WIDTH or VECTOR_LENGTH (UNSIGNED): BLT Mode — Specifies the width, in pixels, of a BLT operation. No pixels are rendered for a width of zero. Vector Mode — Bits [31:30] are reserved in this mode allowing this 14-bit field to specify the length, in pixels, of a vector. No pixels are rendered for a length of zero. This field is limited to 14 bits due to a lack of precision in the registers used to hold the error terms. 15:0 PIXEL_HEIGHT or VECTOR_INITIAL_ERROR (UNSIGNED): BLT Mode — Specifies the height, in pixels, of a BLT operation. No pixels are rendered for a height of zero. Vector Mode — Specifies the initial error for renderng a vector. GX_BASE+8108h-810Bh GP_SCR_X/YCOOR and GP_AXIAL/DIAG_ERROR Register (R/W) Default Value = 00000000h 31:16 SRC_X_POS or VECTOR_AXIAL_ERROR (SIGNED): BLT Mode — Specifies the source X position for a BLT operation. Vector Mode — Specifies the axial error for rendering a vector. 15:0 SRC_Y_POS or VECTOR_DIAG_ERROR (SIGNED): Source Y Position (Signed) — Specifies the source Y position for a BLT operation. Vector Mode — Specifies the diagonal error for rendering a vector. GX_BASE+810Ch-810Fh GP_SRC_COLOR Register (R/W) Default Value = 00000000h 8-BPP Mode 31:24 GP_SRC_COLOR_0: 8-BPP Color Index — The color index must be duplicated in the upper byte of GP_SRC_COLOR_0 when rendering 8-BPP data. 23:16 15:8 GP_SRC_COLOR_1: 8-BPP Color Index — The color index must be duplicated in the upper byte of GP_SRC_COLOR_1 when rendering 8-BPP data. 7:0 16-BPP Mode 31:16 GP_SRC_COLOR_0: 16-BPP Color (RGB) 15:0 GP_SRC_COLOR_1: 16-BPP Color (RGB)

www.national.com 126 Revision 3.1 Integrated Functions (Continued) Geode™ GXm Processor Note: The Graphics Pipeline Source Color Register specifies the colors used when expanding monochrome source data in either the 8-BPP mode or the 16-BPP mode. Those pixels corresponding to clear bits (0) in the source data are rendered using GP_SRC_COLOR_0 and those pixels corresponding to set bits (1) in the source data are rendered using GP_SRC_COLOR_1. GX_BASE+8110h-8113h GP_PAT_COLOR_A Register (R/W) Default Value = 00000000h 8-BPP Mode 31:24 GP_PAT_COLOR_0 : 8-BPP Color Index — The color index must be duplicated in the upper byte of GP_PAT_COLOR_0 when rendering 8-BPP data. 23:16 15:8 GP_PAT_COLOR_1: 8-BPP Color Index — The color index must be duplicated in the upper byte of GP_PAT_COLOR_1 when rendering 8-BPP data. 7:0 16-BPP Mode 31:16 GP_PAT_COLOR_0: 16-BPP Color (RGB) 15:0 GP_PAT_COLOR_1: 16-BPP Color (RGB) Note: The Graphics Pipeline Pattern Color A and B Registers specify the colors used when expanding pattern data. GX_BASE+8114h-8117h GP_PAT_COLOR_B Register (R/W) Default Value = 00000000h 8-BPP Mode 31:24 GP_PAT_COLOR_2: 8-BPP Color Index — The color index must be duplicated in the upper byte of GP_PAT_COLOR_2 when rendering 8-BPP data. 23:16 15:8 GP_PAT_COLOR_3: 8-BPP Color Index — The color index must be duplicated in the upper byte of GP_PAT_COLOR_3 when rendering 8-BPP data. 7:0 16-BPP Mode 31:16 GP_PAT_COLOR_2: 16-BPP Color (RGB) 15:0 GP_PAT_COLOR_3: 16-BPP Color (RGB) Note: The Graphics Pipeline Pattern Color A and B Registers specify the colors used when expanding pattern data. GX_BASE+8120h-8123h GP_PAT_DATA_0 Register (R/W) Default Value = 00000000h 31:0 GP Pattern Data Register 0:The Graphics Pipeline Pattern Data Registers 0 through 3 together contain 128 bits of pat- tern data. The GP_PAT_DATA_0 register corresponds to bits [31:0] of the pattern data. GX_BASE+8124h-8127h GP_PAT_DATA_1 Register (R/W) Default Value = 00000000h 31:0 GP Pattern Data Register 1:The Graphics Pipeline Pattern Data Registers 0 through 3 together contain 128 bits of pat- tern data. The GP_PAT_DATA_1 register corresponds to bits [63:32] of the pattern data. GX_BASE+8128h-812Bh GP_PAT_DATA_2 Register (R/W) Default Value = 00000000h 31:0 GP Pattern Data Register 2:The Graphics Pipeline Pattern Data Registers 0 through 3 together contain 128 bits of pat- tern data. The GP_PAT_DATA_2 register corresponds to bits [95:64] of the pattern data. GX_BASE+812Ch-812Fh GP_PAT_DATA_3 Register (R/W) Default Value = 00000000h 31:0 GP Pattern Data Register 3:The Graphics Pipeline Pattern Data Registers 0 through 3 together contain 128 bits of pat- tern data. The GP_PAT_DATA_3 register corresponds to bits [127:96] of the pattern data. GX_BASE+8140h-8143h GP_VGA_WRITE Register (R/W) Default Value = xxxxxxxxh Note that the registers at GX_BASE+82140h and 8144h are located in the area designated for the graphics pipeline but are used for VGA emulation purposes. Refer to Table 5-5 on page 173 for these register’s bit formats. GX_BASE+8144h-8147h GP_VGA_READ Register (R/W) Default Value = 00000000h Note that the registers at GX_BASE+82140h and 8144h are located in the area designated for the graphics pipeline but are used for VGA emulation purposes. Refer to Table 5-5 on page 173 for these register’s bit formats. Table 4-25. Graphics Pipeline Configuration Registers (Continued) Bit Name Description

Revision 3.1 127 www.national.com Integrated Functions (Continued) Geode™ GXm Processor GX_BASE+8200h-8203h GP_RASTER_MODE Register (R/W) Default Value = 00000000h 31:13 RSVD Reserved: Set to 0. 12 TB Transparent BLIT:When set, this bit enables transparent BLIT. The source color data will be compared to a color key and if it matches, that pixel will not be drawn. The color key value is stored in the BLIT buffer as destination data. The raster operation must be set to C6h, and the pattern registers must be all F’s for this mode to work properly. 11 ST Source Transparency:Enables transparency for monochrome source data. Those pixels corresponding to clear bits in the source data are not drawn. 10 PT Pattern Transparency:Enables transparency for monochrome pattern data. Those pixels corresponding to clear bits in the pattern data are not drawn. 9:8 PM Pattern Mode:Specifies the format of the pattern data. 00 = Indicates a solid pattern. The pattern data is always sourced from the GP_PAT_COLOR_0 register. 01 = Indicates a monochrome pattern. The pattern data is sourced from the GP_PAT_COLOR_0 and GP_PAT_COLOR_1 registers. 10 = Indicates a dither pattern. All four pattern color registers are used. 11 =Indicates a color pattern. The pattern data is sourced directly from the pattern data registers. 7:0 ROP Raster Operation:Specifies the raster operation for pattern, source, and destination data. GX_BASE+8204h-8207h GP_VECTOR_MODE Register (R/W) Default Value = 00000000h 31:4 RSVD Reserved: Set to 0. 3D E S T Read Destination Data:Indicates that frame-buffer destination data is required. 2D M I N Minor Direction:Indicates a positive minor axis step. 1D M A J Major Direction:Indicates a positive major axis step. 0Y M A J Major Direction:Indicates a Y Major vector. GX_BASE+8208h-820Bh GP_BLT_MODE Register (R/W) Default Value = 00000000h 31:9 RSVD Reserved: Set to 0. 8Y Reverse Y Direction:Indicates a negative increment for the Y position. This bit is used to control the direc- tion of screen to screen BLTs to prevent data corruption in overlapping windows. 7:6 SM Source Mode: Specifies the format of the source data. 00 = Source is a color bitmap. 01 = Source is a monochrome bitmap (use source color expansion). 10 = Unused. 11 = Source is a text glyph (use source color expansion). This differs from a monochrome bitmap in that the X position is adjusted by the width of the BLT and the Y position remains the same. 5R S V D Reserved: Set to 0. 4:2 RD Destination Data:Specifies the destination data location. 000 = No destination data is required. The destination data into the raster operation unit is all ones. 010 = Read destination data from BLT Buffer 0. 011 = Read destination data from BLT Buffer 1. 100 = Read destination data from the frame buffer (store temporarily in BLT Buffer 0). 101 = Read destination data from the frame buffer (store temporarily in BLT Buffer 1). 1:0 RS Source Data:Specifies the source data location. 00 = No source data is required. The source data into the raster operation unit is all ones. 01 = Read source data from the frame buffer (temporarily stored in BLT Buffer 0). 10 = Read source data from BLT Buffer 0. 11 = Read source data from BLT Buffer 1. Table 4-25. Graphics Pipeline Configuration Registers (Continued) Bit Name Description

www.national.com 128 Revision 3.1 Integrated Functions (Continued) Geode™ GXm Processor GX_BASE+820Ch-820Fh GP_BLT_STATUS Register (R/W) Default Value = 00000000h 31:10 RSVD Reserved: Set to 0. 9W Screen Width:Selects a frame-buffer width of 2048 bytes (default is 1024 bytes). 8M 16-BPP Mode: Selects a pixel data format of 16 BPP (default is 8 BPP). 7:3 RSVD Reserved: Set to 0. 2B P ( R O ) BLT Pending (Read Only):Indicates that a BLT operation is pending in the master registers. The “BLT Pending” bit must be clear before loading any of the graphics pipeline registers. Loading registers when this bit is set high will destroy the values for the pending BLT. 1P B ( R O ) Pipeline Busy (Read Only):Indicates that the graphics pipeline is processing data. The “Pipeline Busy” bit differs from the “BLT Busy” bit in that the former only indicates that the graphics pipe- line is processing data. The “BLT Busy” bit also indicates that the memory controller has not yet processed all of the requests for the current operation. The “Pipeline Busy” bit must be clear before loading a BLT buffer if the previous BLT operation used the same BLT buffer. 0B B ( R O ) BLT Busy (Read Only):Indicates that a BLT / vector operation is in progress. The “BLT Busy” bit must be clear before accessing the frame buffer directly. GX_BASE+8210h-8213h GP_VGA_BASE (R/W) Default Value = xxxxxxxxh Note that the registers at GX_BASE+8210h and 8214h are located in the area designated for the graphics pipeline but are used for VGA emulation purposes. Refer to Table 5-5 on page 173 for these register’s bit formats. GX_BASE+8214h-8217h GP_VGA_LATCH Register (R/W) Default Value = xxxxxxxxh Note that the registers at GX_BASE+8210h and 8214h are located in the area designated for the graphics pipeline but are used for VGA emulation purposes. Refer to Table 5-5 on page 173 for these register’s bit formats. Table 4-25. Graphics Pipeline Configuration Registers (Continued) Bit Name Description

Revision 3.1 129 www.national.com Integrated Functions (Continued) Geode™ GXm Processor

4.5 DISPLAY CONTROLLER

The GXm processor incorporates a display controller that retrieves display data from the memory controller and for- mats it for output on a variety of display devices. The GXm processor can directly connect to an active matrix TFT LCD flat panel or to an external RAMDAC for CRT display or both. The display controller includes a display FIFO, compression/decompression (CODEC) hardware, hard- ware cursor, a 256-entry-by-18-bit palette RAM (plus three extension colors), display timing generator, dither and frame-rate-modulation circuitry for TFT panels, and flexible output formatting logic. A diagram of the display controller subsystem is shown in Figure 4-14. Figure 4-14. Display Controller Block Diagram Memory Data Compressed Codec Cursor Palette Extensions Palette Dither Output RAMDAC Panel Control Registers TimingMemoryMemory Address Output Control Pseudo/True Color Mux Line Buffer (64x32 bit) Display FIFO (64x64 bit) Latch RAM (264x18 Palette Data and FRM Format Addr. Logic Generatorand Control Logic Address Generator bit)

www.national.com 130 Revision 3.1 Integrated Functions (Continued) Geode™ GXm Processor

4.5.1 Display FIFO

The display controller contains a large (64x64 bit) FIFO for queuing up display data from the memory controller as it is required for output to the screen. The memory control- ler must arbitrate between the display controller requests and other requests for memory access from the micropro- cessor core, L1 cache controller, and the graphics pipe- line. Since display data is required in real time, this data is the highest priority in the system. Without efficient memory management, system performance would suffer dramati- cally due to the constant display-refresh requests from the display controller. The large size of the display FIFO is desirable so that the FIFO may primarily be loaded during times when there is no other request pending to the DRAM controller and so that the memory controller can stay in page mode for a long period of time when servicing the display FIFO. When a priority request from the cache or graphics pipeline occurs, if the display FIFO has enough data queued up, the DRAM controller can imme- diately service the request without concern that the dis- play FIFO will underflow. If the display FIFO is below a programmable threshold, a high-priority request will be sent to the DRAM controller, which will take precedence over any other requests that are pending. The display FIFO is 64 bits wide to accommodate high- speed burst read operations from the DRAM controller at maximum memory bandwidth. In addition to the normal pixel data stream, the display FIFO also queues up cursor patterns.

4.5.2 Compression Technology

To reduce the system memory contention caused by the display refresh, the display controller contains compres- sion and decompression logic for compressing the frame buffer image in real time as it is sent to the display. It com- bines this compressed display buffer into the extra off- screen memory within the graphics memory aperture. Coherency of the compressed display buffer is maintained by use of dirty and valid bits for each line. The dirty and valid RAM is contained on-chip for maximum efficiency. Whenever a line has been validly compressed, it will be retrieved from the compressed display buffer for all future accesses until the line becomes dirty again. Dirty lines will be retrieved from the normal uncompressed frame buffer. The compression logic has the ability to insert a program- mable number of "static" frames, during which time dirty bits are ignored and the valid bits are read to determine whether a line should be retrieved from the frame buffer or compressed display buffer. The less frequently the dirty bits are sampled, the more frequently lines will be retrieved from the compressed display buffer. This allows a programmable screen image update rate (as opposed to refresh rate). Generally, an update rate of 30 frames per second is adequate for displaying most types of data, including real- time video. However, if a flat panel display is used that has a slow response time, such as 100 ms, the image need not be updated faster than ten frames per second, since the panel could not display changes beyond that rate. The compression algorithm used in the GXm processor commonly achieves compression ratios between 10:1 and 20:1, depending on the nature of the display data. This high level of compression provides higher system perfor- mance by reducing typical latency for normal system memory access, higher graphics performance by increas- ing available drawing bandwidth to the DRAM array, and much lower power consumption by significantly reducing the number of off-chip DRAM accesses required for refreshing the display. These advantages become even more pronounced as display resolution, color depth, and refresh rate are increased and as the size of the installed DRAM increases. As uncompressed lines are fed to the display, they will be compressed and stored in an on-chip compressed line buffer (64x32 bits). Lines will not be written back to the compressed display buffer in the DRAM unless a valid compression has resulted, so there is no penalty for pathological frame buffer images where the compression algorithm breaks down.

4.5.3 Motion Video Acceleration Support

The display controller of the GXm processor supports the CS5530 and future I/O companion chips’ hardware motion video acceleration by reading the off-screen video buffer and serializing the video data onto the RAMDAC port. The display controller supplies video data to the I/O compan- ion chips in either interleaved YUV4:2:2 format or RGB5:6:5 format. The I/O companion chips can then scale and filter the data, apply color space conversion to YUV data, and mix the video data with graphics data also supplied by the display controller.

Revision 3.1 131 www.national.com Integrated Functions (Continued) Geode™ GXm Processor

4.5.4 Hardware Cursor

The display controller contains hardware cursor logic to allow overlay of the cursor image onto the pixel data stream. Overhead for updating this image on the screen is kept to a minimum by requiring that only the X and Y posi- tion be changed. This eliminates "submarining" effects commonly associated with software cursors. The cursor, 32x32 pixels with two bits per pixel, is loaded into off- screen memory within the graphics memory aperture. The two-bit code selects color 0, color 1, transparent, or back- ground-color inversion for each pixel in the cursor (see Table 4-31 on page 144. The two cursor colors will be stored as extensions to the normal 256-entry palette at locations 100h and 101h. These palette extensions will be used when driving a flat panel or a RAMDAC operating in 16 BPP (bits per pixel) mode. For 8 BPP operation using an external RAMDAC, the DC_CURSOR_COLOR regis- ter (GX_BASE+8360h) should be programmed to set the indices for the cursor colors. To avoid corruption of the cursor colors by an application program that modifies the external palette, care should be taken to program the cur- sor color indices to one of the static color indices. Since Microsoft Windows typically uses only black and white cursor colors and these are static colors, this kind of prob- lem should rarely occur.

4.5.5 Display Timing Generator

The display controller features a fully programmable tim- ing generator for generating all timing control signals for the display. The timing control signals include horizontal and vertical sync and blank signals in addition to timing for active and overscan regions of the display. The timing generator is similar in function to the CRTC of the original VGA, although programming is more straightforward. Pro- gramming of the timing registers will generally happen via a BIOS INT10 call during a mode set. When programming the timing registers directly, extreme care should be taken to ensure that all timing is compatible with the display device. The timing generator supports overscan to maintain full backward compatibility with the VGA. This feature is sup- ported primarily for CRT display devices since flat panel displays have fixed resolutions and do not provide for overscan. However, the GXm processor supports a mech- anism to center the display when a display mode is selected having a lower resolution than the panel resolu- tion. The border region is effectively stretched to fill the remainder of the screen. The border color is at palette extension 104h. For 8 BPP operation with an external RAMDAC, the DC_BORDER_COLOR register (GX_BASE+8368h) should also be programmed.

4.5.6 Dither and Frame-Rate Modulation

The display controller supports 2x2 dither and two-level frame-rate modulation (FRM) to increase the apparent number of colors displayed on 9-bit or 12-bit TFT panels. Dither and FRM are individually programmable. With dith- ering and FRM enabled, 185,193 colors are possible on a 9-bit TFT panel, and 226,981 colors are possible on a 12- bit TFT panel.

4.5.7 Display Modes

The GXm processor supports 640x480, 800x600, and 1024x768 display resolutions at both 8 and 16 bits per pixel. In addition, 1280x1024 resolution is supported at 8 bits per pixel only. Two 16-bit display formats are sup- ported: RGB 5-6-5 and RGB 5-5-5. All CRT modes use VESA-compatible timing. T able 4-26 on page 132 gives the supported CRT display modes.

www.national.com 132 Revision 3.1 Integrated Functions (Continued) Geode™ GXm Processor Table 4-26. TFT Panel Display Modes Resolution Simultaneous Colors Refresh Rate (Hz) DOTCLK Rate (MHz) PCLK (MHz) Panel Type Maximum Displayed Colors (Note 1) 640x480 (Note 2)

8 BPP

60 25.175 25.175 9-bit 57 3 = 185,193 12-bit 61 3 = 226,981 18-bit 4 3 = 262,144

16 BPP

64 K colors

60 25.175 25.175 9-bit 29x57x29 = 47,937 12-bit 31x61x31 = 58,621 18-bit 32x64x32 = 65,535 800x600 (Note 2) 60 40.0 40.0 9-bit 57 3 = 185,193 12-bit 61 3 = 226,981 18-bit 64 3 = 262,144 60 40.0 40.0 9-bit 29x57x29 = 47,937 12-bit 31x61x31 = 58,621 18-bit 32x64x32 = 65,535 1024x768 8 BPP 256 colors out of a palette of 256 60 65 32.5 9-bit/18-I/F 57 3 = 185,193 60 65 32.5 9-bit/18-I/F 29x57x29 = 47,937 Notes: 1) 9-bit and 12-bit panels use FRM and dither to increase displayed colors. (See Section 4.5.6 “Dither and Frame-Rate Modulation” on page 131.) 2) All 640x480 and 800x600 modes can be run in simultaneous display with CRT.

Revision 3.1 133 www.national.com Integrated Functions (Continued) Geode™ GXm Processor Table 4-27. TFT Panel Data Bus Formats Panel Data Bus Bit 18-Bit TFT 12-Bit TFT 9-Bit TFT 640x480 1024x768

17 R5 R5 R5 R5 Even

16 R4 R4 R4 R4

15 R3 R3 R3 R3

14 R2 R2 R5 Odd

13 R1 R4

12 R0 R3

11 G5 G5 G5 G5 Even

10 G4 G4 G4 G4

www.national.com 134 Revision 3.1 Integrated Functions (Continued) Geode™ GXm Processor Table 4-28. CRT Display Modes Resolution Simultaneous Colors Refresh Rate (Hz) DOTCLK Rate (MHz) PCLK (MHz) Graphics Port Width (Bits) 640x480 8 BPP 256 colors out of a palette of 256 60 25.175 25.175 8 72 31.5 31.5 8 75 31.5 31.5 8 60 25.175 50.35 8 25.175 16 72 31.5 63.0 8 31.5 16 75 31.5 63.0 8 31.5 16 800x600 8 BPP 256 colors out of a palette of 256 60 40.0 40.0 8 72 50.0 50.0 8 75 49.5 49.5 8 60 40.0 80.0 8 40.0 16 72 50.0 100 8 50.0 16 75 49.5 99 8 49.5 16 1024x768 8 BPP 256 colors out of a palette of 256 60 65.0 65.0 8 70 75.0 75.0 8 75 78.5 78.5 8 60 65.0 65.0 16 70 75.0 75.0 16 75 78.5 78.5 16 1280x1024 8 BPP 256 colors out of a palette of 256 60 108.0 108.0 8 54.0 16 75 135.0 67.5 16

Revision 3.1 135 www.national.com Integrated Functions (Continued) Geode™ GXm Processor

4.5.8 Graphics Memory Map

The GXm processor supports a maximum of 4 MB of graphics memory and will map it to an address space (see Figure 4-2 on page 93) higher than the maximum amount of installed RAM. The graphics memory aperture physi- cally resides at the top of the installed system RAM. The start address and size of the graphics memory aperture are programmable on 128 KB boundaries. Typically, the system BIOS sets the size and start address of the graph- ics memory aperture during the boot process based on the amount of installed RAM, user defined CMOS set- tings, and display resolution. The graphics pipeline and display controller address the graphics memory with a 20- bit offset (address bits [21:2]) and four byte enables into the graphics memory aperture. The graphics memory stores several buffers that are used to generate the dis- play: the frame buffer, compressed display buffer, VGA memory, and cursor pattern(s). Any remaining off-screen memory within the graphics aperture may be used by the display driver as desired or not at all.

4.5.8.1 DC Memory Organization Registers

The display controller contains a number of registers that allow full programmability of the graphics memory organi- zation. This includes starting offsets for each of the buffer regions described above, line delta parameters for the frame buffer and compression buffer, as well as com- pressed line-buffer size information. The starting offsets for the various buffers are programmable for a high degree of flexibility in memory organization.

4.5.8.2 Frame Buffer and Compression Buffer

The GXm processor supports primary display modes 640x480, 800x600, and 1024x768 at both 8 BPP and 16 BPP , and 1280x1024 at 8 BPP . Pixels will be packed into DWORDs as shown in Figure 4-15. In order to simplify address calculations by the rendering hardware, the frame buffer is organized in an XY fashion where the offset is simply a concatenation of the X and Y pixel addresses. All 8 BPP display modes with the excep- tion of 1280x1024 resolution will use a 1024-byte line delta between the starting offsets of adjacent lines. All 16 BPP display modes and 1280x1024x8 BPP display modes will use a 2048-byte line delta between the starting offsets of adjacent lines. If there is room, the space b e t w e e nt h ee n do fal i n ea n dt h es t a r to ft h en e x tl i n ew i l l be filled with the compressed display data for that line, thus allowing efficient memory utilization. For 1024x768 display modes, the frame-buffer line size is the same as the line delta, so no room is left for the compressed dis- play data between lines. In this case, the compressed dis- play buffer begins at the end of the frame buffer region and is linearly mapped. Figure 4-15. Pixel Arrangement Within a DWORD B i tP o s i t i o n 3 13 02 92 82 72 62 52 42 32 22 12 01 91 81 71 61 51 41 31 21 11 0 9 8 7 6 5 4 3 2 1 0 Address 3h 2h 1h 0h Pixel Org - 8 BPP (3,0) (2,0) (1,0) (0,0) Pixel Org - 16 BPP (1,0) (0,0) (1023,0) (1023, 1023) (0, 0) (0, 1023) DWORD 0

www.national.com 136 Revision 3.1 Integrated Functions (Continued) Geode™ GXm Processor

4.5.8.3 VGA Display Support

The graphics pipeline contains full hardware support for t h eV G Af r o n te n d .T h eV G Ad a t ai ss t o r e di na2 5 6K B buffer located in graphics memory. The main task for Soft- VGA is converting the data in the VGA buffer to an 8 BPP frame buffer that can be displayed by the GXm proces- sor’s hardware. For some modes, the display controller can display the VGA data directly and the data conversion is not neces- sary. This includes standard VGA mode 13h and the vari- ations of that mode used in several games; the display controller can also directly display VGA planar graphics modes D, E, F , 10, 11, and 12. Likewise, the hardware can directly display all of the higher-resolution VESA modes. Since the frame buffer data is written directly to memory instead of travelling across an external bus, the GXm pro- cessor outperforms typical VGA cards for these modes. The display controller, however, does not directly support text modes. SoftVGA must then convert the characters and attributes in the VGA buffer to an 8 BPP frame buffer the hardware uses for display refresh. See Section 4 “Vir- tual Subsystem Architecture” for SoftVGA details.

4.5.8.4 Cursor Pattern Memory Organization

The cursor overlay patterns are loaded to independent memory locations, usually mapped above the frame buffer and compressed display buffer (off-screen). The cursor buffer must start on a 16-byte aligned boundary. It is lin- early mapped, and is always 256 bytes in size. If there is enough room (256 bytes) after the compression-buffer line but before the next frame-buffer line starts, the cursor pat- tern may be loaded into this area to make efficient use of the graphics memory. Each pattern is a 32x32-pixel array of 2-bit codes. The codes are a combination of AND mask and XOR mask for a particular pixel. Each line of an overlay pattern is stored as two DWORDs, with each DWORD containing the AND masks for 16 pixels in the upper word and the XOR masks for 16 pixels in the lower word. DWORDs are arranged with the leftmost pixel block being least significant and the rightmost pixel block being most significant. Pixels within words are arranged with the leftmost pixels being most significant and the rightmost pixels being least significant. Multiple cursor patterns may be loaded into the off-screen memory. An application may simply change the cursor start offset to select a new cursor pattern. The new cursor pattern will be used at the start of the next frame scan.

4.5.9 Display Controller Registers

The Display Controller maps 100h locations starting at GX_BASE+8300h. Refer to Section 4.1.2 “Control Regis- ters” on page 94 for instructions on accessing these regis- ters. The Display Controller Registers are divided into six cate- gories:  Configuration and Status Registers  Memory Organization Registers  Timing Registers  Cursor and Line Compare Registers  Color Registers  Palette and RAM Diagnostic Registers Table 4-29 summarizes these registers and locations and the following subsections give detailed register/bit for- mats. Table 4-29. Display Controller Register Summary GX_BASE+ Memory Offset Type Name/Function Default Value Configuration and Status Registers 8300h-8303h R/W DC_UNLOCK Display Controller Unlock — This register is provided to lock the most critical memory- mapped display controller registers to prevent unwanted modification (write operations). Read operations are always allowed. 00000000h 8304h-8307h R/W DC_GENERAL_CFG Display Controller General Configuration — General control bits for the display controller. 00000000h 8308h-830Bh R/W DC_TIMING_CFG Display Controller Timing Configuration — Status and control bits for various display timing functions. xx000000h 830Ch-830Fh R/W DC_OUTPUT_CFG Display Controller Output Configuration — Status and control bits for pixel output formatting functions. xx000000h Memory Organization Registers 8310h-8313h R/W DC_FB_ST_OFFSET Display Controller Frame Buffer Start Address — Specifies offset at which the frame buffer starts. xxxxxxxxh

Revision 3.1 137 www.national.com Integrated Functions (Continued) Geode™ GXm Processor 8314h-8317h R/W DC_CB_ST_OFFSET Display Controller Compression Buffer Start Address — Specifies offset at which the com- pressed display buffer starts. xxxxxxxxh 8318h-831Bh R/W DC_CURS_ST_OFFSET Display Controller Cursor Buffer Start Address — Specifies offset at which the cursor mem- ory buffer starts. xxxxxxxxh 831Ch-831Fh -- Reserved 00000000h 8320h-8323h R/W DC_VID_ST_OFFSET Display Controller Video Start Address — Specifies offset at which the video buffer starts. xxxxxxxxh 8324h-8327h R/W DC_LINE_DELTA Display Controller Line Delta — Stores line delta for the graphics display buffers. xxxxxxxxh 8328h-832Bh R/W DC_BUF_SIZE Display Controller Buffer Size — Specifies the number of bytes to transfer for a line of frame buffer data and the size of the compressed line buffer. xxxxxxxxh 832Ch-832Fh -- Reserved 00000000h Timing Registers 8330h-8333h R/W DC_H_TIMING_1 Display Controller Horizontal and Total Timing — Horizontal active and total timing information. xxxxxxxxh 8334h-8337h R/W DC_H_TIMING_2 Display Controller CRT Horizontal Blanking Timing — CRT horizontal blank timing information. xxxxxxxxh 8338h-833Bh R/W DC_H_TIMING_3 Display Controller CRT Sync Timing — CRT horizontal sync timing information. xxxxxxxxh 833Ch-833Fh R/W DC_FP_H_TIMING Display Controller Flat Panel Horizontal Sync Timing: Horizontal sync timing information for an attached flat panel display. xxxxxxxxh 8340h-8343h R/W DC_V_TIMING_1 Display Controller Vertical and Total Timing — Vertical active and total timing information. The parameters pertain to both CRT and flat panel display. xxxxxxxxh 8344h-8247h R/W DC_V_TIMING_2 Display Controller CRT Vertical Blank Timing — Vertical blank timing information. xxxxxxxxh 8348h-834Bh R/W DC_V_TIMING_3 Display Controller CRT Vertical Sync Timing — CRT vertical sync timing information. xxxxxxxxh 834Ch-834Fh R/W DC_FP_V_TIMING Display Controller Flat Panel Vertical Sync Timing — Flat panel vertical sync timing information. xxxxxxxxh Cursor and Line Compare Registers 8350h-8353h R/W DC_CURSOR_X Display Controller Cursor X Position — X position information of the hardware cursor. xxxxxxxxh 8354h-8357h RO DC_V_LINE_CNT Display Controller Vertical Line Count — This read only register provides the current scan- line for the display. It is used by software to time update of the frame buffer to avoid tearing artifacts. xxxxxxxxh 8358h-835Bh R/W DC_CURSOR_Y Display Controller Cursor Y Position — Y position information of the hardware cursor. xxxxxxxxh 835Ch-835Fh R/W DC_SS_LINE_CMP Display Controller Split-Screen Line Compare — Contains the line count at which the lower screen begins in a VGA split-screen mode. xxxxxxxxh Table 4-29. Display Controller Register Summary (Continued) GX_BASE+ Memory Offset Type Name/Function Default Value

www.national.com 138 Revision 3.1 Integrated Functions (Continued) Geode™ GXm Processor Color Registers 8360h-8363h R/W DC_CURSOR_COLOR Display Controller Cursor Color — Contains the 8-bit indices for the cursor colors. xxxxxxxxh 8364h-8367h -- Reserved 00000000h 8368h-836Bh R/W DC_BORDER_COLOR Display Controller Border Color — Contains the 8-bit index for the border or overscan color. xxxxxxxxh 836Ch-836Fh -- Reserved 00000000h Palette and RAM Diagnostic Registers 8370h-8373h R/W DC_PAL_ADDRESS Display Controller Palette Address — This register should be written with the address (index) location to be used for the next access to the DC_PAL_DATA register. xxxxxxxxh 8374h-8377h R/W DC_PAL_DATA Display Controller Palette Data — Contains the data for a palette access cycle. xxxxxxxxh 8378h-837Bh R/W DC_DFIFO_DIAG Display Controller Display FIFO Diagnostic — This register is provided to enable testability of the Display FIFO RAM. xxxxxxxxh 837Ch-837Fh R/W DC_CFIFO_DIAG Display Controller Compression FIFO Diagnostic — This register is provided to enable test- ability of the Compressed Line Buffer (FIFO) RAM. xxxxxxxxh Table 4-29. Display Controller Register Summary (Continued) GX_BASE+ Memory Offset Type Name/Function Default Value

Revision 3.1 139 www.national.com Integrated Functions (Continued) Geode™ GXm Processor

4.5.9.1 Configuration and Status Registers

The Configuration and Status Registers group consists of four 32-bit registers located at GX_BASE+8300h-830Ch. These registers are described below and T able 4-30 gives their bit formats.  Display Controller Unlock (DC_UNLOCK) - This register is provided to lock the most critical memory-mapped display controller registers to prevent unwanted modification (write operations). Read operations are always allowed.  Display Controller General Configuration (DC_GENERAL_CFG) - General control bits for the display controller.  Display Controller Timing Configuration (DC_TIMING_CFG) - Status and control bits for various display timing functions.  Display Controller Output Configuration (DC_OUTPUT_CFG) - Status and control bits for pixel output formatting functions. Table 4-30. Display Controller Configuration and Status Registers Bit Name Description GX_BASE+8300h-8303h DC_UNLOCK Register (R/W) Default Value = 00000000h 31:16 RSVD Reserved: Set to 0. 15:0 UNLOCK_ CODE Unlock Code:This register must be written with the value 4758h in order to write to the protected regis- ters. The following registers are protected by the locking mechanism. DC_GENERAL_CFG DC_CB_ST_OFFSET, DC_BUF_SIZE, DC_V_TIMING_2 DC_TIMING_CFG, DC_CURS_ST_OFFSET , DC_OUTPUT_CFG, DC_H_TIMNG_2, DC_FP_H_TIMING DC_FB_ST_OFFSET, DC_LINE_DELTA, DC_FP_V_TIMING GX_BASE+8304h-8307h DC_GENERAL_CFG (R/W) Default Value = 00000000h

31 DDCK Divide Dot Clock:Divide internal DOTCLK by two relative to PCLK (pertains only to 16 BPP display

modes utilizing an eight-bit RAMDAC): 0 = Disable; 1 = Enable. 30 DPCK Divide Pixel Clock:Divide PCLK by two relative to internal DOTCLK (pertains only to display modes that pack two pixels together such as 1280x1024 on an external CRT only): 0 = Disable; 1 = Enable. 29 VRDY Video Ready Protocol:0 = Low speed video port, use with V2.3 and older. 1 = High speed video port, use with V2.4 and newer. 28 VIDE Video Enable:Motion video port: 0 = Disable; 1 = Enable. 27 SSLC Split-screen Line Compare:VGA line compare function: 0 = Disable; 1 = Enable. When enabled, the internal line counter will be compared to the value programmed in the DC_SS _LINE_CMP register. If it matches, the frame buffer address will be reset to zero. This enables a split screen function. 26 CH4S Chain 4 Skip:Allow display controller to read every 4th DWORD from the frame buffer for compatibility with the VGA: 0 = Disable; 1 = Enable.

25 DIAG FIFO Diagnostic Mode:This bit allows testability of the on-chip Display FIFO and Compressed Line

Buffer via the diagnostic access registers. A low-to-high transition will reset the Display FIFO’s R/W point- ers and the Compressed Line Buffer’s read pointer. 0 = Normal operation; 1 = Enable. 24 LDBL Line Double:Allow line doubling for emulated VGA modes: 0 = Disable; 1 = Enable. If enabled, this will cause each odd line to be replicated from the previous line as the data is sent to the dis- play. Timing parameters should be programmed as if no pixel doubling is used, however, the frame buffer should be loaded with half the normal number of lines. 23 CKWR Clock Write:This bit will be output directly to an external clock chip or SYNDAC. The bit should be pulsed high and low by the software to strobe data into the chip. Note that this bit can be used in conjunction with the DACRS[2:0] pins. 22:20 DAC_RS[2:0] RAMDAC Register Selects:This 3-bit field sets the register select inputs to the external RAMDAC for the next cycle. It is used to allow access to the extended register set of the RAMDAC. Alternatively, these bits may be used in selecting the frequency for an external clock chip or SYNDAC. If more than eight frequency selections are required, the RAMDAC extended register programming sequence must be used or the addi- tional select bit must be provided by some other means.

www.national.com 140 Revision 3.1 Integrated Functions (Continued) Geode™ GXm Processor

19 RTPM Real-Time Performance Monitoring:Allows real-time monitoring of a variety of internal GXm

processor signals by multiplexing the signals onto the CLKWR and DACRS[2:0] pins: 0 = Disable (Normal operation); 1 = Enable. The CLKWR pin should not be fed to a clock chip or SYNDAC when this mode of operation is used, a dif- ferent programming scheme should be used for the clock chip using the DACRS[2:0] signals and RAM- DACRD# and RAMDACWR# signals. The selection of output signals is made using bits [27:16] of the DC_BUF_SIZE register. The lower 12 bits of this field will select one of eight outputs for each pin. 18 FDTY Frame Dirty Mode:Allow entire frame to be flagged as dirty whenever a pixel write occurs to the frame buffer (this is provided for modes that use a linearly mapped frame buffer for which the line delta is not equal to 1024 or 2048 bytes): 0 = Disable; 1 = Enable. When disabled, dirty bits are set according to the Y address of the pixel write. 17 RSVD Reserved: Set to 0. 16 CMPI Compressor Insert Mode:Insert one static frame between update frames: 0 = Disable; 1 = Enable. An update frame is referred to as a frame in which dirty lines will be allowed to be updated. Conversely, a static frame is referred to as a frame in which dirty lines will not be updated (although the image may not be static, since lines that are not compressed successfully must be retrieved from the uncompressed frame buffer). 15:12 DFIFO HI-PRI END LVL Display FIFO High Priority End Level:This field specifies the depth of the display FIFO (in 64-bit entries x 4) at which a high-priority request previously issued to the memory controller will end. The value is dependent upon display mode. This register should always be non-zero and should be larger than the start level. 11:8 DFIFO HI-PRI ST ART LVL Display FIFO High Priority Start Level:This field specifies the depth of the display FIFO (in 64-bit entries x 4) at which a high-priority request will be sent to the memory controller to fill up the FIFO. The value is dependent upon display mode. This register should always be nonzero and should be less than the high-priority end level. 7:6 DCLK_ MUL DCLK Multiplier:This 2-bit field specifies the clock multiplier for the input DCLK pin. After the input clock is optionally multiplied, the internal DOTCLK, PCLK, and FPCLK may be divided as necessary. 00 = Forced Low 01 = 1 x DCLK 10 = 2 x DCLK 11 = 4 x DCLK 5D E C E Decompression Enable:Allow operation of internal decompression hardware: 0 = Disable; 1 = Enable. 4C M P E Compression Enable:Allow operation of internal compression hardware: 0 = Disable; 1 = Enable 3P P C Pixel Panning Compatibility:This bit has the same function as that found in the VGA. Allow pixel alignment to change when crossing a split-screen boundary - it will force the pixel alignment to be 16-byte aligned: 0 = Disable; 1 = Enable. If disabled, the previous alignment will be preserved when crossing a split-screen boundary. 2D V C K Divide Video Clock:Selects frequency of VID_CLK pin: 0 = VID_CLK pin frequency is equal to one-half (½) the frequency of the core clock. 1 = VID_CLK pin frequency is equal to one-fourth (¼) the frequency of the core clock. Note: Bit 28 (VIDE) must be set to 1 for this bit to be valid. 1 CURE Cursor Enable:Allow operation of internal hardware cursor: 0 = Disable; 1 = Enable. 0D F L E Display FIFO Load Enable:Allow the display FIFO to be loaded from memory: 0 = Disable; 1 = Enable. If disabled, no write or read operations will occur to the display FIFO. If enabled, a flat panel should be powered down prior to setting this bit low. Similarly, if active, a CRT should be blanked prior to setting this bit low. Table 4-30. Display Controller Configuration and Status Registers (Continued) Bit Name Description

Revision 3.1 141 www.national.com Integrated Functions (Continued) Geode™ GXm Processor GX_BASE+8308h-830Bh DC_TIMING_CFG Register (R/W) Default Value = xxx00000h

31 VINT

(RO) Vertical Interrupt (Read Only):Is a vertical interrupt pending? 0 = No; 1 = Y es. This bit is provided to maintain backward compatibility with the VGA. It corresponds to VGA port 3C2h bit

30 VNA

(RO) Vertical Not Active (Read Only):Is the active part of a vertical scan is in progress (i.e. retrace, blanking, or border)? 0 = Y es; 1 = No. This bit is provided to maintain backward compatibility with the VGA. It corresponds to VGA port 3BA/3DA bit 3.

29 DNA

(RO) Display Not Active (Read Only):Is the active part of a line is being displayed (i.e. retrace, blanking, or border)? 0 = Yes; 1 = No. This bit is provided to maintain backward compatibility with the VGA. It corresponds to VGA port 3BA/3DA bit 0.

28 SENS

(RO) Monitor Sense (Read Only):This bit returns the result of the voltage comparator test of the RGB lines from the external RAMDAC. The value will be a low level if one or more of the comparators exceed the 340 mV level indicating an unloaded line. This bit can be tested repeatedly to determine the loading on the red, green, and blue lines by loading the palette with various values. The BIOS can then determine whether a color, monochrome, or no monitor is attached. If no RAMDAC is attached, the BIOS should assume that a color panel is attached and operate in color mode. For VGA emulation, read operations to port 3C2 bit 4 are redirected here.

27 DDCI

(RO) DDC Input (Read Only):This bit returns the value from the DDCIN pin that should reflect the value from pin 12 of the VGA connector. It is used to provide support for the VESA Display Data Channel standard level DDC1. 26:20 RSVD Reserved: Set to 0. 19:17 PWR_SEQ DELAY Power Sequence Delay:This 3-bit field sets the delay between edges for the power sequencing control logic. The actual delay is this value multiplied by one frame period (typically 16ms). Note that a value of zero will result in a delay of only one DOTCLK period.

16 BKRT Blink Rate:

0 = Cursor blinks on every 16 frames for a duration of 8 frames (approximately 4 times per second) and VGA text characters will blink on every 32 frames for a duration of 16 frames (approximately 2 times per second). 1 = Cursor blinks on every 32 frames for a duration of 16 frames (approximately 2 times per second) and VGA text characters blink on every 64 frames for a duration of 32 frames (approximately 1 time per sec- ond).

15 PXDB Pixel Double:Allow pixel doubling to stretch the displayed image in the horizontal dimension:

0 = Disable; 1 = Enable. If bit 15 is enabled, timing parameters should be programmed as if no pixel doubling is used, however, the frame buffer should be loaded with half the normal pixels per line. Also, the FB_LINE_SIZE parameter in DC_BUF_SIZE should be set for the number of bytes to be transferred for the line rather than the number displayed.

14 INTL Interlace Scan:Allow interlaced scan mode:

0 = Disable (non-interlaced scanning is supported) 1 = Enable (If a flat panel is attached, it should be powered down before setting this bit.) 13 PLNR VGA Planar Mode: This bit must be set high for all VGA planar display modes. 12 FCEN Flat Panel Center:Allows the border and active portions of a scan line to be qualified as “active” to a flat panel display via the ENADISP signal. This allows the use of a large border region for centering the flat panel display. 0 = Disable; 1 = Enable. When disabled, only the normal active portion of the scan line will be qualified as active.

11 FVSP Flat Panel Vertical Sync Polarity:

0 = Causes TFT vertical sync signal to be normally low, generating a high pulse during sync interval. 1 = Causes TFT vertical sync signal to be normally high, generating a low pulse during sync interval.

10 FHSP Flat Panel Horizontal Sync Polarity:

0 = Causes TFT horizontal sync signal to be normally low, generating a high pulse during sync interval. 1 = Causes TFT horizontal sync signal to be normally high, generating a low pulse during sync interval. Table 4-30. Display Controller Configuration and Status Registers (Continued) Bit Name Description

www.national.com 142 Revision 3.1 Integrated Functions (Continued) Geode™ GXm Processor

9 CVSP CRT Vertical Sync Polarity:

0 = Causes CRT VSYNC signal to be normally low, generating a high pulse during the retrace interval. 1 = Cause CRT VSYNC signal to be normally high, generating a low pulse during the retrace interval.

8 CHSP CRT Horizontal Sync Polarity:

0 = Causes CRT HSYNC signal to be normally low, generating a high pulse during the retrace interval. 1 = Causes CRT HSYNC signal to be normally high, generating a low pulse during the retrace interval. 7B L N K Blink Enable:Blink circuitry: 0 = Disable; 1 = Enable. If enabled, the hardware cursor will blink as well as any pixels. This is provided to maintain compatibility with VGA text modes. The blink rate is determined by the bit 16 (BKRT). 6V I E N Vertical Interrupt Enable:Generate a vertical interrupt on the occurrence of the next vertical sync pulse: 0 = Disable, vertical interrupt is cleared; 1 = Enable. This bit is provided to maintain backward compatibility with the VGA. 5T G E N Timing Generator Enable:Allow timing generator to generate the timing control signals for the display. 0 = Disable, the Timing Registers may be reprogrammed, and all circuitry operating on the DOTCLK will be reset. 1 = Enable, no write operations are permitted to the Timing Registers. 4 DDCK DDC Clock: This bit is used to provide the serial clock for reading the DDC data pin. This bit is multiplexed onto the CRTVSYNC pin, but in order for it to have an effect, the VSYE bit must be set low to disable the normal vertical sync. Software should then pulse this bit high and low to clock data into the GXm proces- sor. This feature is provided to allow support for the VESA Display Data Channel standard level DDC1. 3B L K E Blank Enable:Allow generation of the composite blank signal to the display device: 0 = Disable; 1 = Enable. When disabled, the BLANK# output will be a static low level. This allows VESA DPMS compliance. 2V S Y E Horizontal Sync Enable:Allow generation of the horizontal sync signal to a CRT display device: 0 = Disable; 1 = Enable. When disabled, the HSYNC output will be a static low level. This allows VESA DPMS compliance. Note that this bit only applies to the CRT; the flat panel HSYNC is controlled by the automatic power sequencing logic.

1 HSYE Vertical Sync Enable:Allow generation of the vertical sync signal to a CRT display device:

0 = Disable; 1 = Enable. When disabled, the VSYNC output will be a static low level. This allows VESA DPMS compliance. Note that this bit only applies to the CRT; the flat panel VSYNC is controlled by the automatic power sequencing logic. 0F P P E Flat Panel Power Enable:On a low-to-high transition this bit will enable the flat panel power-up sequence to begin. This will first turn on VDD to the panel, then start the clocks, syncs, and pixel bus, then turn on the LCD bias voltage, and finally the backlight. On a high-to-low transition, this bit will disable the outputs in the reverse order. GX_BASE+830Ch-830Fh DC_OUTPUT_CFG Register (R/W) Default Value = xxx00000h 31:16 RSVD Reserved: Set to 0. 15 DIAG Compressed Line Buffer Diagnostic Mode:This bit will allow testability of the Compressed Line Buffer via the diagnostic access registers. A low-to-high transition will reset the Compressed Line Buffer write pointer. 0 = Disable (Normal operation); 1 = Enable.

14 CFRW Compressed Line Buffer Read/Write Select:Enables the read/write address to the Compressed Line

Buffer for use in diagnostic testing of the RAM. 0 = Write address enabled 1 = Read address enabled

13 PDEH Panel Data Enable High:

0 = The PANEL[17:9] data bus to be driven to a logic low level to effectively blank an attached flat panel display or disable the upper pixel data bus for 16-bit pixel port RAMDACs. 1 = If no flat panel is attached, the PANEL[17:9] data bus will be driven with active pixel data. If a flat panel is attached, setting this bit high will have no effect −the upper panel bus will be driven based upon the power sequencing logic. Table 4-30. Display Controller Configuration and Status Registers (Continued) Bit Name Description

Revision 3.1 143 www.national.com Integrated Functions (Continued) Geode™ GXm Processor

12 PDEL Panel Data Enable Low:

0 = This bit will cause the PANEL[8:0] data bus to be driven to a logic low level to effectively blank an attached flat panel display or disable the lower panel data bus if it is not required. 1= If no flat panel is attached, the PANEL[8:0] data bus will be driven with active pixel data. If a flat panel is attached, setting this bit high will have no effect− the lower panel bus will be driven based upon the power sequencing logic.

11 PRMP Palette Re-map:

0 = The modified codes are sent to the RAMDAC and the external palette should uses the modified map- ping. 1 = Bits [8:1] of the palette output register are routed to the RAMDAC data bus. The GXm processor inter- nal palette RAM may be loaded with 8-bit VGA indices to translate the modified codes stored in display memory so that the RAMDAC data bus will contain the expected indices. The modified codes are used to achieve character blinking in VGA text modes. This mode should be set high set high only for desktop sys- tems with no flat panel attached. It should only be necessary when 8514/A or VESA standard feature con- nector support is required.

10 CKSL Clock Select:Selects output used to clock PANEL[17:0], FPHSYNC, FPVSYNC, and ENADISP output

pins. 1=P C L K 0 = FPCLK (based upon the power sequencing logic) This bit should be high when using a 16-bit RAMDAC. 9F R M S Frame Rate Modulation Select: 0 = Enables FRM circuitry to change the pattern displayed every frame. 1 = Enables FRM circuitry to change the pattern displayed every two frames (to allow for slower response time liquid crystal materials). 83 / 4 A D D 3- or 4-bit Add: 0 = Enables dither and FRM circuitry to operate on the 3 most significant bits of each color component for 9-bit TFT panels. 1 = Enables the dither and FRM circuitry to operate on the 4 most significant bits of each color component for 12-bit TFT panels. 7R S V D Reserved: Must be set to 0. 6R S V D Reserved: Must be set to 0. 5R S V D Reserved: Must be set to 0. 4D I T E Dither Enable:Allow a 2x2 spatial dither on the 3-bit or 4-bit color value. Note that dither will not be sup- ported for 12-bit TFT panels when FRM is enabled. 0 = Disable; 1 = Enable. 3F R M E Frame-Rate Modulation Enable:Allow FRM to be performed on the 3-bit or 4-bit color value using the next most significant bit after the least significant bit sent to the panel. 0 = Disable (no FRM performed); 1 = Enable. 2P C K E PCLK Enable: 0 = PCLK is disabled and a low logic level is driven off-chip. Also, the RAMDAC data bus is driven low. 1 = Enable PCLK to be driven off-chip. This clock operates the RAMDAC interface. 1 16FMT 16 BPP Format:Selects RGB display mode: 0 = RGB 5-6-5 mode 1 = RGB 5-5-5 display mode This bit is only significant if 8 BPP is low, indicating 16 BPP mode.

08 B P P 8B P P/1 6B P PS e l e c t :

0 = 16-bit per pixel display mode is selected. (Bit 1 of OUTPUT_CONFIG will indicate the format of the 16 bit data.) 1 = 8-bit-per-pixel display mode is selected. This is the also the mode used in VGA emulation. Table 4-30. Display Controller Configuration and Status Registers (Continued) Bit Name Description

www.national.com 144 Revision 3.1 Integrated Functions (Continued) Geode™ GXm Processor

4.5.10 Memory Organization Registers

The GXm processor utilizes a graphics memory aperture that is up to 4 MB in size. The base address of the graph- ics memory aperture is stored in the DRAM controller. The graphics memory is made up of the normal uncom- pressed frame buffer, compressed display buffer, and cur- sor buffer. Each buffer begins at a programmable offset within the graphics memory aperture. T h ev a r i o u sm e m o r yb u f f e r sa r ea r r a n g e ds oa st oe f f i - ciently pack the data within the graphics memory aper- ture. This requires flexibility in the way that the buffers are arranged when different display modes are in use. The cursor buffer is a linear block so addressing is straightfor- ward. The frame buffer and compressed display buffer are arranged based upon scan lines. Each scan line has a maximum number of valid or active DWORDs and a delta, that when added to the previous line offset, points to the next line. In this way, the buffers may be stored as linear blocks or as logical blocks as may be desired. The Memory Organization Registers group consists of six 32-bit registers located at GX_BASE+8310h-8328h. These registers are described below and T able 4-31 gives their bit formats.  Display Controller Frame Buffer Start Address (DC_FB_ST_OFFSET) - Specifies the offset at which the frame buffer starts.  Display Controller Compression Buffer Start Address (DC_CB_ST_OFFSET) - Specifies the offset at which the compressed display buffer starts.  Display Controller Cursor Buffer Start Address (DC_CURS_ST_OFFSET) - Specifies the offset at which the cursor memory buffer starts.  Display Controller Video Start Address (DC_VID_ST_OFFSET) - Specifies the offset at which the video buffer starts.  Display Controller Line Delta (DC_LINE_DELTA) - Stores the line delta for the graphics display buffers.  Display Controller Buffer Size (DC_BUF_SIZE) - Specifies the number of bytes to transfer for a line of frame buffer data and the size of the compressed line buffer. (The compressed line buffer will be invali- dated if it exceeds the CB_LINE_SIZE, bits [15:9].) Table 4-31. Display Controller Memory Organization Registers Bit Name Description GX_BASE+8310h-8313h DC_FB_ST_OFFSET Register (R/W) Default Value = xxxxxxxxh 31:22 RSVD Reserved: Set to 0. 21:0 FB_START _OFFSET Frame Buffer Start Offset:This value represents the byte offset of the starting location of the displayed frame buffer. This value may be changed to achieve panning across a virtual desktop or to allow multiple buffering. When this register is programmed to a nonzero value, the compression logic should be disabled. The memory address defined by bits [21:4] will take effect at the start of the next frame scan. The pixel offset defined by bits [3:0] will take effect immediately (in general, it should only change during vertical blank- ing). GX_BASE+8314h-8317h DC_CB_ST_OFFSET Register (R/W) Default Value = xxxxxxxxh 31:22 RSVD Reserved: Set to 0. 21:0 CB_START _OFFSET Compressed Display Buffer Start Offset:This value represents the byte offset of the starting location of the compressed display buffer. Bits [3:0] should always be programmed to zero so that the start offset is aligned to a 16-byte boundary. This value should change only when a new display mode is set due to a change in size of the frame buffer.

Revision 3.1 145 www.national.com Integrated Functions (Continued) Geode™ GXm Processor GX_BASE+8318h-831Bh DC_CUR_ST_OFFSET Register (R/W) Default Value = xxxxxxxxh 31:22 RSVD Reserved: Set to 0. 21:0 CUR_START _OFFSET Cursor Start Offset:This value represents the byte offset of the starting location of the cursor display pattern. Bits [1:0] should always be programmed to zero so that the start offset is DWORD aligned. The cursor data will be stored as a linear block of data. The active cursor will always be 32x32x2 bits in size. Multiple cursor patterns may be loaded into off-screen memory. The start offset is loaded at the start of a frame. Each cursor pattern will be exactly 256 bytes in size. Note that if there is a Y offset for the cursor pattern, the cursor start offset should be set to point to the first displayed line of the cursor pattern. The cursor code for a given pixel is determined by an AND mask and an XOR mask. Each line of a cursor will be stored as two DWORDs, with each DWORD containing the AND masks for 16 pixels in the upper word and the XOR masks for 16 pixels in the lower word. DWORDs will be arranged with the leftmost block of 16 pixels being least significant and the rightmost block being most significant. Pixels within words will be arranged with the leftmost pixels being most significant and the rightmost pixels being least significant. The 2-bit cursor codes are as follows. AND XOR Displayed 0 0 Cursor Color 0 0 1 Cursor Color 1 1 0 Transparent − Background Pixel 11I n v e r t e d − Bit-wise Inversion of Background Pixel GX_BASE+831Ch-831Fh Reserved Default Value = 00000000h GX_BASE+8320h-8323h DC_VID_ST_OFFSET Register (R/W) Default Value = xxxxxxxxh 31:21 RSVD Reserved: Set to 0. 20:0 VID_ST ART _OFFSET Video Buffer Start Offset Value:This is the value for the Video Buffer Start Offset. It represents the starting location for Video Buffer. Bits [3:0] should always be programmed as zero so that the start offset is aligned to a 16 byte boundary. GX_BASE+8324h-8327h DC_LINE_DELTA Register (R/W) Default Value = xxxxxxxxh 31:22 RSVD Reserved: Set to 0. 21:12 CB_LINE_ DELTA Compressed Display Buffer Line Delta:This value represents number of DWORDs that, when added to the starting offset of the previous line, will point to the start of the next compressed line in memory. It is used to always maintain a pointer to the starting offset for the compressed display buffer line being loaded into the display FIFO. 11:10 RSVD Reserved: Set to 0. 9:0 FB_LINE_ DELTA Frame Buffer Line Delta:This value represents number of DWORDs that, when added to the starting offset of the previous line, will point to the start of the next frame buffer line in memory. It is used to always maintain a pointer to the starting offset for the frame buffer line being loaded into the display FIFO. GX_BASE+8328h-832Bh DC_BUF_SIZE Register (R/W) Default Value = xxxxxxxxh 31:30 RSVD Reserved: Set to 0. 29:16 VID_BUF_ SIZE Video Buffer Size:These bits set the video buffer size, in 64-byte segments. The maximum size is 1 MB. 15:9 CB_LINE_ SIZE Compressed Display Buffer Line Size:This value represents the number of DWORDs for a valid com- pressed line plus 1. It is used to detect an overflow of the compressed data FIFO. It should never be larger than 41h or 65Dh since the maximum size of the compressed data FIFO is 64 DWORDs. 8:0 FB_LINE_ SIZE F r a m eB u f f e rL i n eS i z e :This value specifies the number of QWORDS (8-byte segments) to transfer for each display line from the frame buffer. If panning is enabled, this value can generally be programmed to the displayed number of QWORDS + 2 so that enough data is transferred to handle any possible alignment. Extra pixel data in the FIFO at the end of a line will automatically be discarded. GX_BASE+832Ch-832Fh Reserved Default Value = 00000000h Table 4-31. Display Controller Memory Organization Registers (Continued) Bit Name Description

www.national.com 146 Revision 3.1 Integrated Functions (Continued) Geode™ GXm Processor

4.5.11 Timing Registers

The GXm processor timing registers control the genera- tion of sync, blanking, and active display regions. They provide complete flexibility in interfacing to both CRT and flat panel displays. These registers will generally be pro- grammed by the BIOS from an INT 10h call or by the extended mode driver from a display timing file. Note that the horizontal timing parameters are specified in character clocks, which actually means pixels divided by 8, since all characters are bit mapped. For interlaced display the ver- tical counter will be incremented twice during each display line, so vertical timing parameters should be programmed with reference to the total frame rather than a single field. The Timing Registers group consists of six 32-bit registers located at GX_BASE+8330h-834Ch. These registers are described below and Table 4-32 gives their bit formats.  Display Controller Horizontal and Total Timing (DC_H_TIMING_1) - Contains horizontal active and total timing informa- tion.  Display Controller CRT Horizontal Blanking Timing (DC_H_TIMING_2 Register) - Contains CRT horizontal blank timing information.  Display Controller CRT Sync Timing (DC_H_TIMING_3) - Contains CRT horizontal sync timing information. N o t e ,h o w e v e r ,t h a tt h i sr e g i s t e rs h o u l da l s ob e programmed appropriately for flat panel only display since the horizontal sync transition determines when to advance the vertical counter.  Display Controller Flat Panel Horizontal Sync Timing (DC_FP_H_TIMING) - Contains horizontal sync timing information for an attached flat panel display.  Display Controller Vertical and Total Timing (DC_V_TIMING_1) - Contains vertical active and total timing information. The parameters pertain to both CRT and flat panel display.  Display Controller CRT Vertical Blank Timing (DC_V_TIMING_2) - Contains vertical blank timing information.  Display Controller CRT Vertical Sync Timing (DC_V_TIMING_3) - Contains CRT vertical sync timing information.  Display Controller Flat Panel Vertical Sync Timing (DC_FP_V_TIMING) - Contains flat panel vertical sync timing information. Table 4-32. Display Controller Timing Registers Bit Name Description GX_BASE+8330h-8333h DC_H_TIMING_1 Register (R/W) Default Value = xxxxxxxxh 31:27 RSVD Reserved: Set to 0. 26:19 H_TOTAL Horizontal Total:This field represents the total number of character clocks for a given scan line minus 1. Note that the value is necessarily greater than the H_ACTIVE field because it includes bor- der pixels and blanked pixels. For flat panels, this value will never change. The field [26:16] may be programmed with the pixel count minus 1, although bits [18:16] are ignored. The horizontal total is programmable on 8-pixel boundaries only. 18:16 RSVD Reserved: These bits are readable and writable but have no effect. 15:11 RSVD Reserved: Set to 0. 10:3 H_ACTIVE Horizontal Active:This field represents the total number of character clocks for the displayed por- tion of a scan line minus 1. The field [10:0] may be programmed with the pixel count minus 1, although bits [2:0] are ignored. The active count is programmable on 8-pixel boundaries only. Note that for flat panels, if this value is less than the panel active horizontal resolution (H_PANEL), the parameters H_BLANK_START, H_BLANK_END, H_SYNC_START, and H_SYNC_END should be reduced by the value of H_ADJUST (or the value of H_PANEL - H_ACTIVE / 2)to achieve horizontal centering. 2:0 RSVD Reserved: These bits are readable and writable but have no effect. Note: Note also that for simultaneous CRT and flat panel display the H_ACTIVE and H_TOTAL parameters pertain to both. GX_BASE+8334h-8337h DC_H_TIMING_2 Register (R/W) Default Value = xxxxxxxxh 31:27 RSVD Reserved: Set to 0. 26:19 H_BLK_END Horizontal Blank End:This field represents the character clock count at which the horizontal blank- ing signal becomes inactive minus 1. The field [26:16] may be programmed with the pixel count minus 1, although bits [18:16] are ignored. The blank end position is programmable on 8-pixel boundaries only. 18:16 RSVD Reserved: These bits are readable and writable but have no effect. 15:11 RSVD Reserved: Set to 0.

Revision 3.1 147 www.national.com Integrated Functions (Continued) Geode™ GXm Processor 10:3 H_BLK_START Horizontal Blank Start:This field represents the character clock count at which the horizontal blanking signal becomes active minus 1. The field [10:0] may be programmed with the pixel count minus 1, although bits [2:0] are ignored. The blank start position is programmable on 8-pixel bound- aries only. 2:0 RSVD Reserved: These bits are readable and writable but have no effect. Note: A minimum of four character clocks is required for the horizontal blanking portion of a line in order for the timing generator to function correctly. GX_BASE+8338h-833Bh DC_H_TIMING_3 Register (R/W) Default Value = xxxxxxxxh 31:27 RSVD Reserved: Set to 0. 26:19 H_SYNC_END Horizontal Sync End:This field represents the character clock count at which the CRT horizontal sync signal becomes inactive minus 1. The field [26:16] may be programmed with the pixel count minus 1, although bits [18:16] are ignored. The sync end position is programmable on 8-pixel bound- aries only. 18:16 RSVD Reserved: These bits are readable and writable but have no effect. 15:11 RSVD Reserved: Set to 0. 10:3 H_SYNC_START Horizontal Sync Start:This field represents the character clock count at which the CRT horizontal sync signal becomes active minus 1. The field [10:0] may be programmed with the pixel count minus 1, although bits [2:0] are ignored. The sync start position is programmable on 8-pixel boundaries only. 2:0 RSVD Reserved: These bits are readable and writable but have no effect. Note: This register should also be programmed appropriately for flat panel only display since the horizontal sync transition deter- mines when to advance the vertical counter. GX_BASE+833Ch-833Fh C_FP_H_TIMING Register (R/W) Default Value = xxxxxxxxh 31:27 RSVD Reserved: Set to 0. 26:16 FP_H_SYNC _END Flat Panel Horizontal Sync End:This field represents the pixel count at which the flat panel hori- zontal sync signal becomes inactive minus 1. 15:11 RSVD Reserved: Set to 0. 10:0 FP_H_SYNC _START Flat Panel Horizontal Sync Start:This field represents the pixel count at which the flat panel hori- zontal sync signal becomes active minus 1. Note: All values are specified in pixels rather than character clocks to allow precise control over sync position. Note, however, that for flat panels which combine two pixels per panel clock, these values should be odd numbers (even pixel boundary) to guarantee that the sync signal will meet proper setup and hold times. GX_BASE+8340h-8343h DC_V_TIMING_1 Register (R/W) Default Value = xxxxxxxxh 31:27 RSVD Reserved: Set to 0. 26:16 V_TOTAL Vertical Total:This field represents the total number of lines for a given frame scan minus 1. Note that the value is necessarily greater than the V_ACTIVE field because it includes border lines and blanked lines. If the display is interlaced, the total number of lines must be odd, so this value should be an even number. 15:11 RSVD Reserved: Set to 0. 10:0 V_ACTIVE Vertical Active:This field represents the total number of lines for the displayed portion of a frame scan minus 1. Note that for flat panels, if this value is less than the panel active vertical resolution (V_PANEL), the parameters V_BLANK_START, V_BLANK_END, V_SYNC_START, and V_SYNC_END should be reduced by the following value (V_ADJUST) to achieve vertical centering: V_ADJUST = (V_P ANEL - V_ACTIVE) / 2 If the display is interlaced, the number of active lines should be even, so this value should be an odd number. Note: All values are specified in lines. GX_BASE+8344h-8347h DC_V_TIMING_2 Register (R/W) Default Value = xxxxxxxxh 31:27 RSVD Reserved: Set to 0. 26:16 V_BLANK_END Vertical Blank End:This field represents the line at which the vertical blanking signal becomes inactive minus 1. If the display is interlaced, no border is supported, so this value should be identical to V_TOTAL. 15:11 RSVD Reserved: Set to 0. Table 4-32. Display Controller Timing Registers (Continued) Bit Name Description

www.national.com 148 Revision 3.1 Integrated Functions (Continued) Geode™ GXm Processor 10:0 V_BLANK_ ST ART Vertical Blank Start:This field represents the line at which the vertical blanking signal becomes active minus 1. If the display is interlaced, this value should be programmed to V_ACTIVE plus 1. Note: All values are specified in lines. For interlaced display, no border is supported, so blank timing is implied by the total/active tim- ing. GX_BASE+8348h-834Bh DC_V_TIMING_3 Register (R/W) Default Value = xxxxxxxxh 31:27 RSVD Reserved: Set to 0. 26:16 V_SYNC_END Vertical Sync End:This field represents the line at which the CRT vertical sync signal becomes inactive minus 1. 15:11 RSVD Reserved: Set to 0. 10:0 V_SYNC_ST ART Vertical Sync Start:This field represents the line at which the CRT vertical sync signal becomes active minus 1. For interlaced display, note that the vertical counter is incremented twice during each line and since there are an odd number of lines, the vertical sync pulse will trigger in the middle of a line for one field and at the end of a line for the subsequent field. Note: All values are specified in lines. GX_BASE+834Ch-834Fh DC_FP_V_TIMING Register (R/W) Default Value = xxxxxxxxh 31:27 RSVD Reserved: Set to 0. 26:16 FP_V_SYNC _END Flat Panel Vertical Sync End:This field represents the line at which the flat panel vertical sync sig- nal becomes inactive minus 2. Note that the internal flat panel vertical sync is latched by the flat panel horizontal sync prior to being output to the panel. 15:11 RSVD Reserved: Set to 0. 10:0 FP_VSYNC _START Flat Panel Vertical Sync Start:This field represents the line at which the internal flat panel vertical sync signal becomes active minus 2. Note that the internal flat panel vertical sync is latched by the flat panel horizontal sync prior to being output to the panel. Note: All values are specified in lines. Table 4-32. Display Controller Timing Registers (Continued) Bit Name Description

Revision 3.1 149 www.national.com Integrated Functions (Continued) Geode™ GXm Processor

4.5.12 Cursor Position Registers

The Cursor Position Registers contain pixel coordinate information for the cursor. These values are not latched by the timing generator until the start of the frame to avoid tearing artifacts when moving the cursor. The Cursor Position group consists of four 32-bit registers located at GX_BASE+8350h-835Ch. These registers are described below and Table 4-33 gives their bit formats.  Display Controller Cursor X Position (DC_CURSOR_X) - Contains the X position information of the hardware cursor.  Display Controller Vertical Line Count (DC_V_LINE_CNT) - This register is read only. It provides the current scanline for the display. It is used by software to time update of the frame buffer to avoid tearing artifacts.  Display Controller Cursor Y Position (DC_CURSOR_Y) - Contains the Y position information of the hardware cursor.  Display Controller Split-Screen Line Compare (DC_SS_LINE_CMP) - Contains the line count at which the lower screen begins in a VGA split-screen mode. Table 4-33. Display Controller Cursor Position Registers Bit Name Description GX_BASE+8350h-8353h DC_CURSOR_X Register (R/W) Default Value = xxxxxxxxh 31:16 RSVD Reserved: Set to 0. 15:11 X_OFFSET X Offset:This field represents the X pixel offset within the 32x32 cursor pattern at which the displayed portion of the cursor is to begin. Normally, this value is set to zero to display the entire cursor pattern, but for cursors for which the "hot spot" is not at the left edge of the pattern, it may be necessary to display the rightmost pixels of the cursor only as the cursor moves close to the left edge of the display. 10:0 CURSOR_X Cursor X:This field represents the X coordinate of the pixel at which the upper left corner of the cursor is to be displayed. This value is referenced to the screen origin (0,0) which is the pixel in the upper left corner of the screen. GX_BASE+8354h-8357h DC_V_LINE_CNT Register (RO) Default Value = xxxxxxxxh 31:11 RSVD Reserved (Read Only) 10:0 V_LINE_CNT (RO) Vertical Line Count (Read Only):This value is the current scanline of the display. Note: The value in this register is driven directly off of the DOTCLK, and consequently it is not synchronized with the CPU clock. Soft- ware should read this register twice and compare the result to ensure that the value is not transitioning. GX_BASE+8358h-835Bh DC_CURSOR_Y Register (R/W) Default Value = xxxxxxxxh 31:16 RSVD Reserved: Set to 0. 15:11 Y_OFFSET Y Offset:This field represents the Y line offset within the 32x32 cursor pattern at which the displayed portion of the cursor is to begin. Normally, this value is set to zero to display the entire cursor pattern, but for cursors for which the "hot spot" is not at the top edge of the pattern, it may be necessary to display the bottommost lines of the cursor only as the cursor moves close to the top edge of the display. Note that if this value is nonzero, the CUR_START_OFFSET must be set to point to the first cursor line to be displayed. 10 RSVD Reserved: Set to 0. 9:0 CURSOR_Y Cursor Y:This field represents the Y coordinate of the line at which the upper left corner of the cursor is to be displayed. This value is referenced to the screen origin (0,0) which is the pixel in the upper left cor- ner of the screen. This field is alternately used as the line-compare value for a newly-programmed frame buffer start offset. This is necessary for VGA programs that change the start offset in the middle of a frame. In order to use this function, the hardware cursor function should be disabled. GX_BASE+835Ch-835Fh DC_SS_LINE_CMP Register (R/W) Default Value = xxxxxxxxh 31:11 RSVD Reserved: Set to 0. 10:0 SS_LINE_CM P Split-Screen Line Compare:This is the line count at which the lower screen begins in a VGA split- screen mode. Note: When the internal line counter hits this value, the frame buffer address is reset to 0. This function is enabled with the SSLC bit in the DC_GENERAL_CFG register.

www.national.com 150 Revision 3.1 Integrated Functions (Continued) Geode™ GXm Processor

4.5.13 Color Registers

T h e s er e g i s t e r sa r eu s e di n8B P Pd i s p l a ym o d ew i t ha n external RAMDAC for passing cursor and border color indices to the palette in the RAMDAC. For the flat panel color translation, the cursor and border color data is loaded into palette extensions as described in the Palette Access Registers section. The Color Registers group consists of two 32-bit registers located at GX_BASE+8360h-8368h. These registers are described below and Table 4-34 gives their bit formats.  Display Controller Cursor Color (DC_CURSOR_COLOR) - Contains the 8-bit indices for the cursor colors.  Display Controller Border Color (DC_BORDER_COLOR) Contains the 8-bit index for the border or overscan color. Table 4-34. Display Controller Color Registers Bit Name Description GX_BASE+8360h-8363h DC_CURSOR_COLOR Register (R/W) Default Value = xxxxxxxxh 31:16 RSVD Reserved: Set to 0. 15:8 CURS_CLR_1 Cursor Color 1:This is the 8-bit index to the external palette for the cursor color 1. It should point to a reserved or static color. 7:0 CURS_CLR_0 Cursor Color 0:This is the 8-bit index to the external palette for the cursor color 0. It should point to a reserved or static color. GX_BASE+8364h-8367h Reserved Default Value = 00000000h GX_BASE+8368h-836Bh DC_BORDER_COLOR Register (RO) Default Value = xxxxxxxxh 31:8 RSVD Reserved: Set to 0. 7:0 BORDER_CLR Border Color:This is the 8-bit index to the external palette for the border color. It should point to a reserved or static color. GX_BASE+836Ch-836Fh Reserved Default Value = 00000000h

Revision 3.1 151 www.national.com Integrated Functions (Continued) Geode™ GXm Processor

4.5.14 Palette Access Registers

These registers are used for accessing the internal palette RAM and extensions. In addition to the standard 256 entries for 8 BPP color translation, the GXm processor palette has extensions for cursor colors and overscan (border) color. The Palette Access Register group consists of four 32-bit registers located at GX_BASE+8370h-837Ch. These reg- isters are described below and Table 4-35 gives their bit formats.  Display Controller Palette Address (DC_P AL_ADDRESS) - This register should be written with the address (index) location to be used for the next access to the DC_PAL_DATA register.  Display Controller Palette Data (DC_PAL_DATA) - Contains the data for a palette access cycle.  Display Controller Display FIFO Diagnostic (DC_DFIFO_DIAG) - This register is provided to enable testability of the Display FIFO RAM.  Display Controller Compression FIFO Diagnostic (DC_CFIFO_DIAG) - This register is provided to enable testability of the Compressed Line Buffer (FIFO) RAM. Table 4-35. Display Controller Palette and RAM Diagnostic Registers Bit Name Description GX_BASE+8370h-8373h DC_PAL_ADDRESS Register (R/W) Default Value = xxxxxxxxh 31:9 RSVD Reserved: Set to 0. 8:0 PALETTE_ADDR Palette Address:This 9-bit field specifies the address to be used for the next access to the DC_PAL_DATA register. Each access to the data register will automatically increment the palette address register. If non-sequential access is made to the palette, the address register must be loaded between each non-sequential data block. The address ranges are as follows. Address Color 0h - FFh Standard Palette Colors 100h Cursor Color 0 101h Cursor Color 1 102h Reserved 103h Reserved 104h Overscan Color 105h - 1FFh Not Valid Note that in general, 18-bit values will be loaded for all color extensions. However, if a 16 BPP mode is active, only the appropriate most significant bits will be used (5-5-5 or 5-6-5). If an 8 BPP display mode is active and an external RAMDAC is used, the cursor index will be obtained from the DC_CURSOR_COLOR register. The border index will be obtained from the DC_BORDER_COLOR register. GX_BASE+8374h-8377h DC_PAL_DATA Register (R/W) Default Value = xxxxxxxxh 31:18 RSVD Reserved: Set to 0. 17:0 PALETTE_DATA Palette Data:This 18-bit field contains the read or write data for a palette access. Note: When a read or write to the palette RAM occurs, the previous output value will be held for one additional DOTCLK period. This effect should go unnoticed and will provide for sparkle-free update. Prior to a read or write to this register, the DC_PAL_ADDRESS register should be loaded with the appropriate address. The address automatically increments after each access to this register, so for sequential access, the address register need only be loaded once GX_BASE+8378h-837Bh DC_DFIFO_DIAG Register (R/W) Default Value = xxxxxxxxh

www.national.com 152 Revision 3.1 Integrated Functions (Continued) Geode™ GXm Processor 31:0 DISPLAY FIFO DIAGNOSTIC DATA Display FIFO Diagnostic Read or Write Data:Before this register is accessed, the DIAG bit in DC_GENERAL_CFG register should be set high and the DFLE bit should be set low. Since, each FIFO entry is 64 bits, an even number of write operations should be performed. Each pair of write operations will cause the FIFO write pointer to increment automatically. After all write operations have been performed, a single read of don't care data should be performed to load data into the out- put latch. Each subsequent read will contain the appropriate data which was previously written. Each pair of read operations will cause the FIFO read pointer to increment automatically. A pause of at least four core clocks should be allowed between subsequent read operations to allow adequate time for the shift to take place. GX_BASE+837Ch-837Fh DC_CFIFO_DIAG Register (R/W) Default Value = xxxxxxxxh 31:0 COMPRESSED FIFO DIAGNOS- TIC DATA Compressed Data FIFO Diagnostic Read or Write Data:Before this register is accessed, the DIAG bit in DC_GENERAL_CFG register should be set high and the DFLE bit should be set low. Also, the DIAG bit in DC_OUTPUT_CFG should be set high and the CFRW bit in DC_OUTPUT_CFG should be set low. After each write, the FIFO write pointer will automatically increment. After all write operations have been performed, the CFRW bit of DC_OUTPUT_CFG should be set high to enable read addresses to the FIFO and a single read of don't care data should be performed to load data into the output latch. Each subsequent read will contain the appropriate data which was previously written. After each read, the FIFO read pointer will automatically incre- ment. Table 4-35. Display Controller Palette and RAM Diagnostic Registers (Continued) Bit Name Description

Revision 3.1 153 www.national.com Integrated Functions (Continued) Geode™ GXm Processor

4.5.15 CS5530 Display Controller Interface

As previously stated in Section 1.6 “Geode GXM/CS5530 System Designs” on page 11, the GXm processor inter- faces with either the CS5530 I/O companion chip. This section will discuss the specifics on signal connections between the two devices with regards to the display con- troller. W h e nt h eG X mp r o c e s s o ri su s e di nas y s t e mw i t ht h e CS5530 I/O companion chip, the need for an external RAMDAC is eliminated. The CS5530 contains the DACs, a video accelerator engine, and the TFT interface. A GXm processor and CS5530-based system supports both portable and desktop configurations. Figure 4-16 shows the signal connections for both types of systems. Figure 4-16. Display Controller Signal Connections DCLK PCLK FP_HSYNC FP_VSYNC ENA_DISP VID_RDY VID_CLK VID_DATA[7:0] PIXEL[17:12] PIXEL[11:6] HSYNC VSYNC R[5:0] G[5:0] B[5:0] CLK VDD 12VBKL Pin 13 Pin 14 Pin 3 Pin 2 Pin 1 Geode™ GXm Processor Power Control TFT ENAB VGA Pin 15 Pin 12 Flat Geode™ CS5530 I/O Companion DCLK PCLK FP_HSYNC FP_VSYNC FP_DISP_ENA VID_RDY VID_CLK VID_DATA[7:0] PIXEL[23:18]* PIXEL[15:10]* PIXEL[5:0] VID_VAL CRT_HSYNC CRT_VSYNC PIXEL[7:2]* VID_VAL HSYNC VSYNC FP_ENA_VDD FP_ENA_BKL FP_DISP_ENA_OUT FP_HSYNC FP_VSYNC FP_CLK FP_DATA[17:12] FP_DATA[11:16] FP_DATA[5:0] Logic HSYNC_OUT VSYNC_OUT IOUTR IOUTG IOUTB DDC_SCL DDC_SDA Panel Port Portable Configuration

www.national.com 154 Revision 3.1 Integrated Functions (Continued) Geode™ GXm Processor

4.5.15.1 CS5530 Video Port Data Transfer

VID_VAL indicates that the GXm processor has placed valid data on VID_DATA[7:0]. VID_RDY indicates that the CS5530 is ready to accept the next byte of video data. VID_DATA[7:0] is advanced when both VID_VAL and VID_RDY are asserted. VID_RDY is driven one clock early to the GXm processor while VID_VAL is driven coin- cident with VID_DATA[7:0]. A sample interface functional timing diagram is shown in Figure 4-17. Figure 4-17. Video Port Data Transfer (CS5530) VID_CLK VID_VAL 8C L K s 8+3C L K s VID_RDY 3C L K s 4C L K sVID_DATA 8C L K s 1 2 CLK CLKs CLK CLKs CLKs 4C L K s Note: VID_CLK = CORE_CLK/2 [7:0]

Revision 3.1 155 www.national.com Integrated Functions (Continued) Geode™ GXm Processor

4.6 PCI CONTROLLER

The GXm processor includes an integrated PCI controller with the following features.

4.6.1 X-Bus PCI Slave

 16-byte PCI write buffer  16-byte PCI read buffer from X-bus  Supports cache line bursting  Write/Inv line support  Pacing of data for read or write operations with X-bus  No active byte enable transfers supported

4.6.2 X-Bus PCI Master

 1 6b y t eX - b u st oP C Iw r i t eb u f f e r  Configuration read/write Support  Int Acknowledge support  Lock conversion  Support fast back-to-back cycles as slave

4.6.3 PCI Arbiter

 Fixed, rotating, hybrid, or ping-pong arbitration (programmable)  Support four masters, three on PCI  Internal REQ for CPU  Master retry mask counter  Master dead timer  Resource or total system lock support

4.6.4 Generating Configuration Cycles

Configuration space is a physical address space unique to PCI. Configuration Mechanism #1 must be used by soft- ware to generate configuration cycles. Two DWORD I/O locations are used in this mechanism. The first DWORD location (CF8h) references a read/write register that is named CONFIG_ADDRESS. The second DWORD address (CFCh) references a register named CONFIG_DATA. The general method for accessing con- figuration space is to write a value into CONFIG_ADDRESS that specifies the PCI bus, device on that bus, and configuration register in that device being accessed. A read or write to CONFIG_DATA will then cause the bridge to translate that CONFIG_ADDRESS value to the requested configuration cycle on the PCI bus.

4.6.5 Generating Special Cycles

A special cycle is a broadcast message to the PCI bus. Two hardcoded special cycle messages are defined in the command encode: HALT and SHUTDOWN. Software can also generate special cycles by using special cycle gener- ation for configuration mechanism #1 as described in the initiate a special cycle from software, the host must write a value to CONFIG_ADDRESS encoded as shown in Table 4-36. The next value written to CONFIG_DATA is the encoded special cycle. Type 0 or Type 1 conversion will be based on the Bus Bridge number matching the GXm processor’s bus number of 00h. Table 4-36. Special-Cycle Code to CONFIG_ADDRESS 3 1 3 0 2 4 2 3 2 2 2 1 2 0 1 9 1 8 1 7 1 6 1 5 1 4 1 3 1 2 1 1 1 09 8 765432 1 0 1 0000000 B u sN o .=B r i d g e 11111 1 1 1 000000T T CONFIG ENABLE RSVD BUS NUMBER DEVICE NUMBER FUNCTION NUMBER REGISTER NUMBER TRANS LATION TYPE Note: See Table 4-37 on page 156, bits [1:0] for translation type.

www.national.com 156 Revision 3.1 Integrated Functions (Continued) Geode™ GXm Processor

4.6.6 PCI Configuration Space Control Registers

There are two registers in this category: CONFIG_ADDRESS and CONFIG_DATA. The CONFIG_ADDRESS register contains the address information for the next configuration space access to CONFIG_DATA. Only DWORD accesses are permitted to this register all others will be forwarded as normal I/O cycles to the PCI bus. The CONFIG_DATA register contains the data that is sent or received during a PCI configuration space access. Table 4-37 gives the bit formats for these two registers. Table 4-37. PCI Configuration Registers Bit Name Description I/O Offset 0CF8h-0CFBh CONFIG_ADDRESS Register (R/W) Default Value = 00000000h

31 CFG_EN Config Enable:Determines when accesses should be translated to configuration cycles on the PCI

bus, or treated as a normal I/O operation. This register will be updated only on full DWORD I/O oper- ations to the CONFIG_ADDRESS. Any other accesses are treated as normal I/O cycles in order to allow I/O devices to use BYTE or WORD registers at the same address an remain unaffected. Once bit 31 is set high, subsequent accesses to CONFIG_DATA are then translated to configuration cycles. 1 = Generate configuration cycles 0 = Normal I/O cycles 30:24 RSVD Reserved: Set to 0. 23:16 BUS Bus: Specifies a PCI bus number in the hierarchy of 1 to 256 buses. 15:11 DEVICE Device:Selects a device on a specified bus. A device value of 00h will select the GXm processor if the bus number is also 00h. DEVICE values of 01h to 15h will be mapped to AD[31:11], so only 21 of the 32 possible devices are supported. A DEVICE value of 00001b will map to AD[11] while a device of 10101b will map to AD[31]. 10:8 FUNCTION Function:Selects a function in a multi-function device. 7:2 REGISTER Register:Chooses a configuration DWORD space register in the selected device. 1:0 TT Translation Type Bits:These bits indicate if the configuration access is local or one that requires translation through other bridges to another PCI bus. When an access occurs to the CONFIG_DATA address and the specified bus number matches the GXm processor’s bus number (00h), then a Type 0 translation takes place. For a Type 0 translation, the CONFIG_ADDRESS register values are translated to AD lines on the PCI bus. Note that bits 10:2 are passed unchanged. The DEVICE value is mapped to one of 21 AD lines. The translation type bits are set to 00 to indicate a transaction on the local PCI bus. When an access occurs to the CONFIG_DATA address and the specified bus number is not 00h (Type 1), the GXm processor passes this cycle to the PCI bus by copying the contents of the CONFIG_ADDRESS register onto the AD lines during the address phase of the cycle while driving the translation type bits AD[1:0] to 01. I/O Offset 0CFCh-0CFFh CONFIG_DATA (R/W) Default Value = 00000000h 31:0 CONFIG_DATA Configuration Data Register:Contains the data that is sent or received during a PCI configuration space access. The register accessed is determined by the value in the CONFIG_ADDRESS regis- ter. The CONFIG_DATA register supports BYTE, WORD, or DWORD accesses. To access this reg- ister, bit 31 of the CONFIG_ADDRESS register must be set to 0 and a full DWORD I/O access must be done. Configuration cycles are performed when bit 31 of the CONFIG_ADDRESS register is set to 1

Revision 3.1 157 www.national.com Integrated Functions (Continued) Geode™ GXm Processor

4.6.7 PCI Configuration Space Registers

To access the internal PCI configuration registers of the GXm processor, the Configuration Address Register (CONFIG_ADDRESS) must be written as a DWORD using the format shown in Table 4-38. Any other size will b ei n t e r p r e t e da sa nI / Ow r i t et oP o r t0 C F 8 h .A l s o ,w h e n entering the Configuration Index, only the six most signifi- cant bits of the offset are used, and the two least signifi- cant bits must be 00b. Table 4-39 summarizes the registers located within the Configuration Space. The tables that follow, give detailed register/bit formats. Table 4-38. Format for Accessing the Internal PCI Configuration Registers 3 1 3 0 2 9 2 8 2 7 2 6 2 5 2 4 2 3 2 2 2 1 2 0 1 9 1 8 1 7 1 6 1 5 1 4 1 3 1 2 1 1 1 0 9876543210

1 R E S E R V E D 0000000000000000 C o n f i g u r a t i o n I n d e x 00

Table 4-39. PCI Configuration Space Register Summary Index Type Name Default Value 00h-01h RO Vendor Identification 1078h 02h-03h RO Device Identification 0001h 04h-05h R/W PCI Command 0007h 06h-07h R/W Device Status 0280h 08h RO Revision Identification 00h 09h-0Bh RO Class Code 060000h 0Ch RO Cache Line Size 00h 0Dh R/W Latency Timer 0Dh 0Eh-3Fh -- Reserved 00h 40h R/W PCI Control Function 1 00h 41h R/W PCI Control Function 2 96h 42h -- Reserved 00h 43h R/W PCI Arbitration Control 1 80h 44h R/W PCI Arbitration Control 2 00h 45h-FFh -- Reserved 00h

www.national.com 158 Revision 3.1 Integrated Functions (Continued) Geode™ GXm Processor Table 4-40. PCI Configuration Registers Bit Name Description Index 00h-01h Vendor Identification Register (RO) Default Value = 1078h 31:0 VID (RO) Vendor Identification Register (Read Only):The combination of this value and the device ID uniquely identifies any PCI device. The Vendor ID is the ID given to national Semiconductor Corporation by the PCI SIG. Index 02h-03h Device Identification Register (RO) Default Value = 0001h 31:0 DIR (RO) Device Identification Register (Read Only):This value along with the vendor ID uniquely identifies any PCI device. Index 04h-05h PCI Command Register (R/W) Default Value = 0007h 15:10 RSVD Reserved: Set to 0. 9F B E Fast Back-to-Back Enable:As a master, the GXm processor does not support this function. This bit returns 0. 8S E R R SERR# Enable: This is used as an output enable gate for the SERR# driver. 7W A T Wait Cycle Control:GXm processor does not do address/ data stepping. This bit is always set to 0. 6P E Parity Error Response: 0 = GXm processor ignores parity errors on the PCI bus. 1 = GXm processor checks for parity errors. 5V P S VGA Palette Snoop:GXm processor does not support this function. This bit is always set to 0. 4M S Memory Write and Invalidate Enable:As a master, the GXm processor does not support this function. This bit is always set to 0. 3S P C Special Cycles:GXm processor does not respond to special cycles on the PCI bus. This bit is always set to 0. 2B M Bus Master: 0 = GXm processor does not perform master cycles on the PCI. 1 = GXm processor can act as a bus master on the PCI. 1M S Memory Space: GXm processor will always respond to memory cycles on the PCI. This bit is always set to 1. 0I O S I/O Space:GXm processor will not respond to I/O accesses from the PCI. This bit is always set to 1. Index 06h-07h PCI Device Status Register (RO, R/W Clear) Default Value = 0280h 15 DPE Detected Parity Error:When a parity error is detected, this bit is set to 1. This bit can be cleared to 0 by writing a 1 to it. 14 SSE Signaled System Error:This bit is set whenever SERR# is driven active. 13 RMA Received Master Abort:This bit is set whenever a master abort cycle occurs. A master abort will occur whenever a PCI cycle is not claimed except for special cycles. This bit can be cleared to 0 by writing a 1 to it. 12 RTA Received Target Abort:This bit is set whenever a target abort is received while the GXm processor is master of the cycle. This bit can be cleared to 0 by writing a 1 to it. 11 STA Signaled Target Abort:This bit is set whenever the GXm processor signals a target abort. A target abort is signaled when an address parity occurs for an address that hits in the GXm processor’s address space. This bit can be cleared to 0 by writing a 1 to it. 10:9 DT Devise Timing: 00 = Fast 01 = Medium 10 = Slow 11 = Reserved The GXm processor performs medium DEVSEL# active for addresses that hit into the GXm processor address space. These two bits are always set to 01.

Revision 3.1 159 www.national.com Integrated Functions (Continued) Geode™ GXm Processor 8D P D Data Parity Detected:This bit is set when three conditions are met. 1) GXm processor asserted PERR# or observed PERR# asserted; 2) GXm processor is the master for the cycle in which the PERR# occurred; and 3) PE (bit 6 of Command Register) is enabled. This bit can be cleared to 0 by writing a 1 to it. 7F B S Fast Back-to-Back Capable:As a target, the processor is capable of accepting Fast Back-to-Back transactions. This bit is always set to 1. 6:0 RSVD Reserved: Set to 0. Index 08h Revision Identification Register (RO) Default Value = 00h 7:0 RID (RO) Revision ID (Read Only):This register contains the revision number of the GXm design. Index 09h-0Bh Class Code Register (RO) Default Value = 060000h 23:16 CLASS Class Code:The class code register is used to identify the generic function of the device. The GXm processor is classified as a host bridge device (06). 15:0 RSVD (RO) Reserved (Read Only) Index 0Ch Cache Line Size Register (RO) Default Value = 00h 7:0 CACHELINE Cache Line Size (Read Only):The cache line size register specifies the system cacheline size in units of 32-bit words. This function is not supported in the GXm processor. Index 0Dh Latency Timer Register (R/W) Default Value = 00h 7:5 RSVD Reserved: Set to 0. 4:0 LAT_TIMER Latency Timer:The latency timer as used in this implementation will prevent a system lockup resulting from a slave the does not responded to the master. If the register value is set to 00h, the timer is dis- abled. Otherwise, Timer represents the 5 MSBs of an 8-bit counter. The counter will reset on each valid data transfer. If the counter expires before the next TRDY# is received active, then the slave is consid- ered to be incapable of responding, and the master will stop the transaction with a master abort and flag an SERR# active. This would also keep the master from being retried forever by a slave device that con- tinues to issue retries. In these cases, the master will also stop the cycle with a master abort. Index 0Eh-3Fh Reserved Default Value = 00h Index 40h PCI Control Function 1 Register (R/W) Default Value = 00h 7R S V D Reserved: Set to 0. 6S W Single Write Mode:PCI slave supports: 0 = Multiple PCI write cycles 1 = Single cycle write transfers on the PCI bus. The slave will perform a target disconnect with the first data transferred. 5S R Single Read Mode:PCI slave supports: 0 = Multiple PCI read cycles. 1 = Single cycle read transfers on the PCI bus. The slave will perform a target disconnect with the first data transferred. 4R X B N E Force Retry when X-Bus Buffers are Not Empty: 0 = PCI slave accepts the PCI cycle with data in the PCI master write buffers. The data in the PCI master write buffers will not be affected or corrupted. The PCI master holds request active indicating the need to access the PCI bus. 1 = PCI slave retries cycles if the PCI master X-bus write buffers contain buffered data. 3S W B E PCI Slave Write Buffer Enable:PCI slave write buffers: 0 = Disable; 1 = Enable. 2C L R E PCI Cache Line Read Enable:Read operations from the PCI into the GXm processor: 0 = Single cycle unless a read multiple or memory read line command is used. 1 = Cause a cache line read to occur. 1X B E X-Bus Burst Enable:PCI slave acting as a master performs burst cycles on the X-bus on write-back invalidate cycles from the PCI. 0 = Disable; 1 = Enable. (This bit does not control read bursting; bit 2 does.) 0R S V D Reserved: Should return a value of 0. Table 4-40. PCI Configuration Registers (Continued) Bit Name Description

www.national.com 160 Revision 3.1 Integrated Functions (Continued) Geode™ GXm Processor Index 41h PCI Control Function 2 Register (R/W) Default Value = 96h 7R S V D Reserved: Set to 0. 6R W _ C L K RAW Clock: A debug signal used to view internal clock operation. 0 = Disable; 1 = Enable. 5P F S PERR# forces SERR#: PCI master drives an active SERR# anytime it also drives or receives an active PERR#: 0 = Disable; 1 = Enable. 4X W B X-Bus to PCI Write Buffer:Enable GXm processor PCI master’s X-Bus write buffers (non-locked mem- ory cycles are buffered, I/O cycles and lock cycles are not buffered): 0 = Disable; 1 = Enable. 3:2 SDB Slave Disconnect Boundary:PCI slave issues a disconnect with data when it crosses line boundary: 00 = 128 bytes 01 = 256 bytes 10 = 512 bytes 11 = 1024 bytes Works in conjunction with bit 1. 1S D B E Slave Disconnect Boundary Enable: 0 = PCI slave disconnects on boundaries set by bits [3:2]. 1 = PCI disconnects on cache line boundary which is 16 bytes. 0X W S X-Bus Wait State Enable:The PCI slave acting as a master on the X-bus will insert wait states on write cycles for data setup time. 0 = Disable; 1 = Enable. Index 43h PCI Arbitration Control 1 Register (R/W) Default Value = 80h 7B G Bus Grant: 0 = Grants bus regardless of X-bus buffers. 1 = Grants bus only if X-bus buffers are empty. 6R S V D Reserved: Set to 1. 5R M E 2 REQ2# Retry Mask Enable:Arbiter allows the REQ2# to be masked based on the master retry mask in bits [2:1]: 0 = Disable; 1 = Enable. 4R M E 1 REQ1# Retry Mask Enable:Arbiter allows the REQ1# to be masked based on the master retry mask in bits [2:1]: 0 = Disable; 1 = Enable. 3R M E 0 REQ0# Retry Mask Enable:Arbiter allows the REQ0# to be masked based on the master retry mask in bits [2:1]: 0 = Disable; 1 = Enable. 2:1 MRM Master Retry Mask:When a target issues a retry to a master, the arbiter can mask the request from the retried master in order to allow other lower order masters to gain access to the PCI bus: 00 = No retry mask 01 = Mask for 16 PCI clocks 10 = Mask for 32 PCI clocks 11 = Mask for 64 PCI clocks 0H X R Hold X-bus on Retries:Arbiter holds the X-Bus X_HOLD for two additional clocks to see if the retried master will request the bus again: 0 = Disable; 1 = Enable (This may prevent retry thrashing in some cases.) Table 4-40. PCI Configuration Registers (Continued) Bit Name Description

Revision 3.1 161 www.national.com Integrated Functions (Continued) Geode™ GXm Processor Index 44h PCI Arbitration Control 2 Register (R/W) Default Value = 00h 7P P Ping-Pong: 0 = Arbiter grants the processor bus per the setting of bits [2:0]. 1 = Arbiter grants the processor bus ownership of the PCI bus every other arbitration cycle. 6:4 FAC Fixed Arbitration Controls:These bits control the priority under fixed arbitration. The priority table is as follows (priority listed highest to lowest): 000 = REQ0#, REQ1#, REQ2# 001 = REQ1#, REQ0#, REQ2# 010 = REQ0#, REQ2#,REQ1# 011 = Reserved 100 = REQ1#, REQ2#, REQ0# 101 = Reserved 110 = REQ2#, REQ1#, REQ0# 111 = REQ2#, REQ0#, REQ1# Note: The rotation arbitration bits [2:0] must be set to 000 for full fixed arbitration. If rotation bits are not set to 000, then hybrid arbitration will occur. If Ping-Pong is enabled (bit 7 = 1), the processor will have priority every other arbitration. In this mode, the arbiter grants the PCI bus to a master and ignores all other requests. When the master finishes, the processor will be guaranteed access. At this point PCI requests will again be recognized. This will switch arbitration from CPU-to-PCI to CPU-to-PCI, etc. 3R S V D Reserved: Set to 0. 45h-FFh -- Reserved Table 4-40. PCI Configuration Registers (Continued) Bit Name Description

www.national.com 162 Revision 3.1 Integrated Functions (Continued) Geode™ GXm Processor

4.6.8 PCI Cycles

The following sections and diagrams provide the func- tional relationships for PCI cycles.

4.6.8.1 PCI Read Transaction

A PCI read transaction consists of an address phase and one or more data phases. Data phases may consist of wait cycles and a data transfer. Figure 4-18 illustrates a PCI read transaction. In this example, there are three data phases. The address phase begins on clock 2 when FRAME# is asserted. During the address phase, AD[31:0] contains a valid address and C/BE[3:0]# contains a valid bus com- mand. The first data phase begins on clock 3. During the data phase, AD[31:0] contains data and C/BE[3:0]# indi- cate which byte lanes of AD[31:0] carry valid data. The first data phase completes with zero delay cycles. How- ever, the second phase is delayed one cycle because the target was not ready so it deasserted TRDY# on clock 5. The last data phase is delayed one cycle because the master deasserted IRDY# on clock 7. For additional information refer to Chapter 3.3.1, Read Transaction, of the PCI Local Bus Specification, Revision 2.1. Figure 4-18. Basic Read Operation CLK FRAME# AD C/BE# DATA-1 DATA-2 DATA-3ADDR BUS CMD BE#s IRDY# TRDY# DEVSEL# BUS TRANSACTION ADDR PHASE DATA PHASE DATA PHASE DATA PHASE DATA TRANSFER WAIT WAIT DATA TRANSFER WAIT DATA TRANSFER

Revision 3.1 163 www.national.com Integrated Functions (Continued) Geode™ GXm Processor

4.6.8.2 PCI Write Transaction

A PCI write transaction is similar to a PCI read transac- tion, consisting of an address phase and one or more data phases. Since the master provides both address and data, no turnaround cycle is required following the address phase. The data phases work the same for both read and write transactions. Figure 4-19 illustrates a write transaction. T h ea d d r e s sp h a s eb e g i n so nc l o c k2w h e nF R A M E #i s asserted. The first and second data phases complete without delays. During data phase 3, the target inserts three wait cycles by deasserting TRDY#. For additional information refer to Chapter 3.3.2, Write Transaction, of the PCI Local Bus Specification, Revision 2.1. Figure 4-19. Basic Write Operation CLK FRAME# AD C/BE# DATA-2 DATA-3ADDR IRDY# TRDY# DEVSEL# BUS TRANSACTION ADDR PHASE DATA PHASE DATA PHASE DATA PHASE DATA TRANSFER WAIT WAIT WAIT DATA TRANSFER DATA-1 BE#’s-2 BE#’s-3BUS CMD BE#’s-1 DATA TRANSFER

www.national.com 164 Revision 3.1 Integrated Functions (Continued) Geode™ GXm Processor

4.6.8.3 PCI Arbitration

An agent requests the bus by asserting its REQ#. Based on the arbitration scheme set in the PCI Arbitration Con- trol 2 Register (Index 44h), the GX PCI arbiter will grant the request by asserting GNT#. Figure 4-20 illustrates basic arbitration. REQ#-a is asserted at clock 1. The PCI GXm processor arbiter grants access to Agent A by asserting GNT#-a on clock 2. Agent A must begin a transaction by asserting FRAME# within 16 clocks, or the GX PCI arbiter will remove GNT#. Also, it is possible for Agent A to lose bus ownership sooner if another agent with higher priority requests the bus. However, in this example, Agent A starts the transaction on clock 3 by asserting FRAME# and completes its transaction. Since Agent A requests another transaction, REQ#-a remains asserted. When FRAME# is asserted on clock 3, the GXm processor’s PCI arbiter determines Agent B should go next, asserts GNT#-b and deasserts GNT#-a on clock 4. Agent B requires only a single transaction. It completes the trans- action, then deasserts FRAME# and REQ#-b on clock 6. The GXm processor’s PCI arbiter can then grant access to agent A, and does so on clock 7. Note that all buffers must flush before a grant is given to a new agent. For additional information refer to Chapter 3.4.1, Arbitra- tion Signaling Protocol, of the PCI Local Bus Specifica- tion, Revision 2.1.

4.6.8.4 PCI Halt Command

Halt is a broadcast message from the processor indicating it has executed a halt instruction. The PCI Special Cycle command is used to broadcast the message to all agents on the bus segment. During the address phase of the Halt Special cycle, C/BE[3:0]# = 0001 and AD[31:0] are driven to random values. During the data phase, C/BE[3:0]# = 1100 indicating bytes 1 and 0 are valid and AD[15:0] = 0001h. For additional information, refer to Chapter 3.7.2, Special Cycle, and Appendix A, Special Cycle Messages, of the PCI Local Bus Specification, Revision 2.1. Figure 4-20. Basic Arbitration CLK REQ#-a REQ#-b GNT#-a GNT#-b FRAME# AD DATAADDR DATAADDR access-a access-b

Revision 3.1 165 www.national.com Geode™ GXm Processor

5.0 Virtual Subsystem Architecture

This section describes the Virtual Subsystem Architec- ture® (VSA™) as implemented with the Geode GXm pro- cessor(s) and VSA enhanced I/O companion device(s). VSA provides a framework to enable software implemen- tation of traditionally hardware-only components. VSA software executes in System Management Mode (SMM), enabling it to execute transparently to the operating sys- tem, drivers and applications. The VSA design is based upon a simple model for replac- ing hardware components with software. Hardware to be virtualized is merely replaced with simple access detec- tion circuitry which asserts the processor’s SMI# (System Management Interrupt) pin when hardware accesses are detected. The current execution stream is immediately preempted, and the processor enters SMM. The SMM system software then saves the processor state, initializes the VSA execution environment, decodes the SMI source and dispatches handler routines which have registered requests to service the decoded SMI source. Once all handler routines have completed, the processor state is restored and normal execution resumes. In this manner, hardware accesses are transparently replaced with the execution of SMM handler software. Historically, SMM software was used primarily for the sin- gle purpose of facilitating active power management for notebook designs. That software’s only function was to manage the power up and down of devices to save power. With high performance processors now available, it is fea- sible to implement, primarily in SMM software, PC capa- bilities traditionally provided by hardware. In contrast to power management code, this virtualization software gen- erally has strict performance requirements to prevent application performance from being significantly impacted. Several functions can be virtualized in a GXm processor based design using the VSA environment. The VSA enhanced chipsets provide programmable resources to trap both memory and I/O accesses. However, specific hardware is included to support the virtualization of VGA core compatibility and audio functionality in the system. The hardware support for VGA emulation resides com- pletely inside the GXm processor. Legacy VGA accesses do not generate off-chip bus cycles. However, the VSA support hardware for XpressAUDIO resides in the CS5530 I/O companion device and is described in the CS5530 specification.

5.1 VIRTUAL VGA

The GXm processor reduces the burden of PC-legacy hardware by using a balanced mix of hardware and soft- ware to provide the same functionality. The graphics pipe- line contains full hardware support for the VGA “front- end”, the logic that controls read and write operations to the VGA frame buffer (located in graphics memory). For some modes, the hardware can also provide direct display of the data in the VGA buffer. Virtual VGA traps frame buffer accesses only when necessary, but it must trap all VGA I/O accesses to maintain the VGA state and properly program the graphics pipeline and display controller. VGA functionality with the GXm processor includes the standard VGA modes (VGA, EGA, CGA, and MDA) as well as the higher-resolution VESA modes. The CGA and MDA modes (modes 0 through 7) require that Virtual VGA convert the data in the VGA buffer to a separate 8-BPP frame buffer that the hardware can use for display refresh. The remaining modes, VGA, EGA, and VESA, can be dis- played directly by the hardware, with no data conversion required. For these modes, Virtual VGA outperforms typi- cal VGA cards because the frame buffer data does not travel across an external bus. Display drivers for popular GUI (graphical user interface) based operating systems are provided by National Semi- conductor which enable a full featured 2D hardware accel- erator to be used instead of the emulated VGA core.

5.1.1 Traditional VGA Hardware

A VGA card consists of display memory and control regis- ters. The VGA display memory shows up in system mem- ory between addresses A0000h and BFFFFh. It is possible to map this memory to three different ranges within this 128 KB block. T h ef i r s tr a n g ei s - A0000h to B0000h for EGA and VGA modes, the second range is - B0000h to B7FFFh for MDA modes, and the third range is - B8000h to BFFFFh for CGA modes. The VGA control registers are mapped to the I/O address range from 3B0h to 3DFh. The VGA registers are a c c e s s e dw i t ha ni n d e x i n gs c h e m et h a tp r o v i d e sm o r e registers than would normally fit into this range. Some registers are mapped at two locations, one for mono- chrome, and another for color. The VGA hardware can be accessed by calling BIOS rou- tines or by directly writing to VGA memory and control registers. DOS always calls BIOS to set up the display mode and render characters. Many other applications access the VGA memory and control registers directly. The VGA card can be set up to a virtually unlimited num- ber of modes. However, many applications use one of the predefined modes specified by the BIOS routine which sets up the display mode. The predefined modes are translated into specific VGA control register setups by the BIOS. The standard modes supported by VGA cards are shown in Table 5-1.

www.national.com 166 Revision 3.1 Virtual Subsystem Architecture (Continued) Geode™ GXm Processor A VGA is made up of several functional units. T h eframe bufferis 256 KB of memory that provides data for the video display. It is organized as 64 K 32-bit DWORDs. T h esequencer decomposes word and DWORD CPU accesses into byte operations for the graphics controller. It also controls a number of miscellaneous functions, including reset and some clocking controls. T h egraphics controllerprovides most of the interface between CPU data and the frame buffer. It allows the programmer to read and write frame buffer data in different formats. Plus provides ROP (raster operation) and masking functions. T h eCRT controllerprovides video timing signals and address generation for video refresh. It also provides a text cursor. T h eattribute controllercontains the video refresh datapath, including text rasterization and palette lookup. T h egeneral registersprovide status information for the programmer as well as control over VGA-host address mapping and clock selection. This is all handled in hardware by the graphics pipeline. It is important to understand that a VGA is constructed of numerous independent functions. Most of the register fields correspond to controls that were originally built out of discrete logic or were part of a dedicated controller such as the 6845. The notion of a VGA “mode” is a higher- level convention to denote a particular set of values for the registers. Many popular programs do not use standard modes, preferring instead to produce their own VGA set- ups that are optimal for their purposes.

5.1.1.1 VGA Memory Organization

T h eV G Am e m o r yi so r g a n i z e da s6 4K3 2 - b i tD W O R D s . This organization is usually presented as four 64 KB “planes”. A plane consists of one byte out of every DWORD. Thus, plane 0 refers to the least significant byte from every one of the 64 K DWORDs. The addressing granularity of this memory is a DWORD, not a byte; that is, consecutive addresses refer to consecutive DWORDs. The only provision for byte-granularity addressing is the four-byte enable signals used for writes. In C parlance, single_plane_byte = (dword_fb[address] >> (plane * 8)) & 0xFF; When dealing with VGA, it is important to recognize the distinction between host addresses, frame buffer addresses, and the refresh address pipe. A VGA control- ler contains lots of hardware to translate between these address spaces in different ways, and understanding these translations is critical to understanding the entire device. In standard four-plane graphics modes, a frame- buffer DWORD provides eight 4-bit pixels. The left-most pixel comes from bit 7 of each plane, with plane 3 provid- ing the most significant bit. pixel[i].bit[j] = dword_fb[address].bit[j*8 + (7-i)]

5.1.1.2 VGA Front End

The VGA front end consists of address and data transla- tions between the CPU and the frame buffer. This func- tionality is contained within the graphics controller and sequencer components. Most of the front end functionality is implemented in the VGA read and write hardware of the GXm processor. An important axiom of the VGA is that the front end and back end are controlled independently. There are no register fields that control the behavior of both pieces. Terms like “VGA odd/even mode” are there- fore somewhat misleading; there are two different controls for odd/even functionality in the front end, and two sepa- rate controls in the refresh path to cause “sensible” refresh behavior for frame buffer contents written in odd/even mode. Normally, all these fields would be set up together, but they don’t have to be. This sort of orthogonal behavior gives rise to the enormous number of possible VGA “modes”. The CPU end of the read and write pipes is one byte wide. Word and DWORD accesses from the Table 5-1. Standard VGA Modes Category Mode Text or Graphics Resolution Format Type Software 0,1 Text 40x25 Characters CGA 2,3 Text 80x25 Characters CGA 4,5 Graphics 320x200 2 BPP CGA

6 Graphics 640x200 1 BPP CGA

7 Text 80x25 Characters MDA

Hardware 0Dh Graphics 320x200 4 BPP EGA 0Eh Graphics 640x200 4 BPP EGA 0Fh Graphics 640x350 1 BPP EGA 10h Graphics 640x350 4 BPP EGA 11h Graphics 640x480 1 BPP VGA 12h Graphics 640x480 4 BPP VGA 13h Graphics 320x200 8 BPP VGA

Revision 3.1 167 www.national.com Virtual Subsystem Architecture (Continued) Geode™ GXm Processor C P Ut oV G Am e m o r ya r eb r o k e nd o w ni n t om u l t i p l eb y t e accesses by the sequencer. For example, a word write to A0000h (in a VGA graphics mode) is processed as if it were two-byte write operations to A0000h and A0001h.

5.1.1.3 Address Mapping

When a VGA card sees an address on the host bus, bits [31:15] determine whether the transaction is for the VGA. Depending on the mode, addresses 000AXXXX, 000B{0XXX}XXX, or 000B{1XXX}XXXX can decode into VGA space. If the access is for the VGA, bits [15:0] pro- vide the DWORD address into the frame buffer (however, see odd/even and Chain 4 modes, below). Thus, each byte address on the host bus addresses a DWORD in VGA memory. On a write transaction, the byte enables are normally driven from the sequencer’s MapMask register. The VGA has two other write address mappings that modify this behavior. In odd/even (Chain 2) write mode, bit 0 of the a d d r e s si su s e dt oe n a b l eb y t e s0a n d2( i fz e r o )o rb y t e s 1 and 3 (if one). In addition, the address presented to the frame buffer has bit 0 replaced with the PageBit field of the Miscellaneous Output register. Chain 4 write mode is similar; only one of the four byte enables is asserted, based on bits [1:0] of the address, and bits [1:0] of the frame buffer address are set to zero. In each of these modes, the MapMask enables are logically ANDed into the enables that result from the address.

5.2 GXM VIRTUAL VGA

The GXm processor provides VGA compatibility through a mixture of hardware and software. The processor core contains SMI generation hardware for VGA memory write operations. The bus controller contains SMI generation hardware for VGA I/O read and write operations. The graphics pipeline contains hardware to detect and pro- cess reads and writes to VGA memory. VGA memory is partitioned from system memory.

5.2.1 Datapath Elements

The graphics controller contains several elements that convert between host data and frame buffer data. The rotator simply rotates the byte written from the host b y0t o7b i t st ot h er i g h t ,b a s e do nt h eR o t a t e C o u n tf i e l d of the DataRotate register. It has no effect in the read path. The display latch is a 32-bit register that is loaded on every read access to the frame buffer. All 32 bits of the frame buffer DWORDs are loaded into the latch. The write-mode unitconverts a byte from the host into a 32-bit value. A VGA has four write modes: W r i t e M o d e 0 : - B i tno fb y t ebc o m e sf r o mo n eo ft w op l a c e s , depending on bit b of the EnableSetReset register. If that bit is zero, it comes from bit n of the host data. If that bit is one, it comes from bit b of the SetReset register. This mode allows the programmer to set some planes from the host data and the others from SetReset.  Write Mode 1: - All 32 bits come directly out of the display latch; the host data is ignored. This mode is used for screen- to-screen copies.  Write Mode 2: - Bit n of byte b comes from bit b of the host data; that is, the four LSBs of the host data are each replicated through a byte of the result. In conjunction with the BitMask register, this mode allows the programmer to directly write a 4-bit color to one or more pixels.  Write Mode 3: - B i tno fb y t ebc o m e sf r o mb i tbo ft h eS e t R e s e t register. The host data is ANDed with the BitMask register to provide the bit mask for the write (see below). The read mode unit converts a 32-bit value from the frame buffer into a byte. A VGA has two read modes:  Read Mode 0: - One of the four bytes from the frame buffer is returned, based on the value of the ReadMapSelect register. In Chain 4 mode, bits [1:0] of the read address select a plane. In odd/even read mode, bit 0 of the read address replaces bit 0 of ReadMapSe- lect.  Read Mode 1: - Bit n of the result is set to 1 if bit n in every byte b matches bit b of the ColorCompare register; other- wise it is set to 0. There is a ColorDon’tCare register that can exclude planes from this comparison. In four-plane graphics modes, this provides a conver- sion from 4 BPP to 1 BPP . The ALU is a simple two-operand ROP unit that operates on writes. Its operating modes are COPY , AND, OR, and XOR. The 32-bit inputs are: 1) the output of the write-mode unit and 2) the display latch (not necessarily the value at the frame buffer address of the write). An application that wishes to performs ROPs on the source and destination must first byte read the address (to load the latch) and then immediately write a byte to the same address. The ALU has no effect in Write Mode 1. T h eb i tm a s ku n i td o e sn o tp r o v i d eat r u eb i tm a s k . Instead, it selects between the ALU output and the display latch. The mask is an 8-bit value, and bit n of the mask makes the selection for bit n of all four bytes of the result (a zero selects the latch). No bit masking occurs in Write Mode 1. The VGA hardware of the GXm processor does not imple- ment Write Mode 1 directly, but it can be indirectly imple- mented by setting the BitMask to zero and the ALU mode to COPY .

www.national.com 168 Revision 3.1 Virtual Subsystem Architecture (Continued) Geode™ GXm Processor

5.2.2 Video Refresh

VGA refresh is controlled by two units: the CRT controller (CRTC) and the attribute controller (ATTR). The CRTC provides refresh addresses and video control; the ATTR provides the refresh datapath, including pixel formatting and internal palette lookup. The VGA back end contains two basic clocks: the dot clock (or pixel clock) and the character clock. The Clock- Select field of the Miscellaneous Output register selects a “master clock” of either 25 MHz or 28 MHz. This master clock, optionally divided by two, drives the dot clock. The character clock is simply the dot clock divided by eight or nine. The VGA supports four basic pixel formats. Using text for- mat, the VGA interprets frame buffer values as ASCII characters, foreground/background attributes, and font data. The other three formats are all “graphics modes”, known as APA (All Points Addressable) modes. These for- mats could be called CGA-compatible (odd/even four bits/pixel), EGA-compatible (4-plane four bits/pixel), and VGA-compatible (pixel-per-byte eight bits/pixel). The for- mat is chosen by the ShiftRegister field of the Graphics Controller Mode register. The refresh address pipe is an integral part of the CRTC, and has many configuration options. Refresh can begin at any frame buffer address. The display width and the frame buffer pitch (scan-line delta) are set separately. Multiple scan lines can be refreshed from the same frame buffer addresses. The LineCompare register causes the refresh address to be reset to zero at a particular scan line, pro- viding support for vertical split-screen. W i t h i nt h ec o n t e x to fas i n g l es c a nl i n e ,t h er e f r e s h address increments by one on every character clock. Before being presented to the frame buffer, refresh addresses can be shifted by 0, 1, or 2 bits to the left. These options are often mis-named Byte, Word, and Dou- bleword modes. Using this shifter, the refresh unit can be programmed to skip one out of two or three out of four DWORDs of refresh data. As an example of the utility of this function, consider Chain 4 mode, described earlier. Pixels written in Chain 4 mode occupy one out of every four DWORDs in the frame buffer. If the refresh path is put into “Doubleword” mode, the refresh will come only from those DWORDs writable in Chain 4. This is how VGA mode 13h works. In text mode, the ATTR has a lot of work to do. At each character clock, it pulls a DWORD of data out of the frame b u f f e r .I nt h a tD W O R D ,p l a n e0c o n t a i n st h eA S C I Ic h a r - acter code, and plane 1 contains an attribute byte. The ATTR uses plane 0 to generate a font lookup address and read another DWORD. In plane 2, this DWORD contains a bit-per-pixel representation of one scan line in the appro- priate character glyph. The ATTR transforms these bits into eight pixels, obtaining foreground and background colors from the attribute byte. The CRTC must refresh from the same memory addresses for all scan lines that make up a character row; within that row, the ATTR must f e t c hs u c c e s s i v es c a nl i n e sf r o mt h eg l y p ht a b l es oa st o draw proper characters. Graphics modes are somewhat simpler. In CGA-compatible mode, a DWORD provides eight pixels. The first four pixels come from planes 0 and 2; each 4-bit pixel gets bits [3:2] from plane 2, and bits [1:0] from plane 0. The remaining four pixels come from planes 1 and 3. The EGA-compatible mode also gets eight pixels from a DWORD, but each pixel gets one bit from each plane, with plane 3 providing bit 3. Finally, VGA-compatible mode gets four pixels from each DWORD; plane 0 provides the first pixel, plane 1 the next, and so on. The 8 BPP mode uses an option to provide every pixel for two dot clocks, thus allowing the refresh pipe to keep up (it only increments on character clocks) and meaning that the 320-pixel-wide mode 13h really has 640 visible pixels per line. The VGA color model is unusual. The ATTR contains a 16-entry color palette with 6 bits per entry. Except for 8 BPP modes, all VGA configu- rations drive four bits of pixel data into the palette, which produces a 6-bit result. Based on various control regis- ters, this value is then combined with other register con- tents to produce an 8-bit index into the DAC. There is a ColorPlaneEnable register to mask bits out of the pixel data before it goes to the palette; this is used to emulate four-color CGA modes by ignoring the top two bits of each pixel. In 8 BPP modes, the palette is bypassed and the pixel data goes directly to the DAC

5.2.3 GXm VGA Hardware

The GXm processor core contains hardware to detect VGA accesses and generate SMI interrupts. The graphics pipeline contains hardware to detect and process reads and writes to VGA memory. The VGA memory on the GXm processor is partitioned from system memory. The GXm processor has the following hardware components to assist the VGA emulation software.  SMI Generation  VGA Range Detection  VGA Sequencer V G A W r i t e / R e a d P a t h  VGA Address Generator  VGA Memory

5.2.3.1 SMI Generation

VGA emulation software is notified of VGA memory accesses by an SMI generated in dedicated circuitry in the processor core that detects and traps memory accesses. The SMI generation hardware for VGA memory addresses is in the second stage of instruction decoding on the processor core. This is the earliest stage of instruc- tion decode where virtual addresses have been translated to physical addresses. Trapping after the execution stage is impractical, because memory write buffering will allow subsequent instructions to execute. The VGA emulation code requires the SMI to be gener- ated immediately when a VGA access occurs. The SMI generation hardware can optionally exclude areas of VGA memory, based on a 32-bit register which has a control bit for each 2 KB region of the VGA memory window. The

Revision 3.1 169 www.national.com Virtual Subsystem Architecture (Continued) Geode™ GXm Processor control bit determines whether or not an SMI interrupt is generated for the corresponding region. The purpose of this hardware is to allow the VGA emulation software to disable SMI interrupts in VGA memory regions that are not currently displayed. For direct display modes (8 BPP or 16 BPP) in the display controller, Virtual VGA can operate without SMI genera- tion. The SMI generation circuit on the GXm processor has configuration registers to control and mask SMI interrupts in the VGA memory space.

5.2.3.2 VGA Memory Addresses

SMI generation can be configured to trap VGA memory accesses in one of the following ranges: A0000h to AFFFFh (EGA,VGA), B0000h to B7FFFh (MDA), or B8000h to BFFFFh (CGA). Range selection is accomplished through programmable bits in the VGACTL register (Index B9h). Fine control can be exercised within the range selected to allow off-screen accesses to occur without generating SMIs. SMI generation can also separately control the following I/O ranges: 3B0h to 3BFh, 3C0h to 3CFh, and 3D0h to 3DFh. The BC_XMAP_1 register (GX_BASE+8004h) in the Internal Bus Interface Unit has an enable/disable bit for each of the address ranges above.

5.2.3.3 VGA Configuration Registers

Table 5-2 summarizes the VGA Configuration Registers. Detailed register/bit formats are given in T able 5-3.

5.2.3.4 VGA Control Register

The VGA control register (VGACTL) provides control for SMI generation through an enable bit for memory address ranges A0000h to BFFFFh. Each bit controls whether or not SMI is generated for accesses to the corresponding address range. The default value of this register is zero so that VGA accesses will not be trapped on systems with an external VGA card.

5.2.3.5 VGA Mask Registers

The VGA Mask register (VGAM) has 32 bits that can selectively mask 2 KB regions within the VGA memory region A0000h to AFFFFh. If none of the three regions is enabled in VGACTL, then the contents of VGAM are i g n o r e d .V G A Mc a nb eu s e dt op r e v e n tt h eo c c u r r e n c eo f SMI when non-displayed VGA memory is accessed. This is an enhancement that improves performance for double- buffered applications only. Table 5-2. VGA Configuration Registers Summary Index Name Description Default B9h VGACTL VGA Control Register 00h (SMI generation disabled) BAh-BDh VGAM VGA Mask Register Don’t Care

www.national.com 170 Revision 3.1 Virtual Subsystem Architecture (Continued) Geode™ GXm Processor Table 5-3. VGA Configuration Registers Bit Description Index B9h VGACTL Register (R/W) Default Value = 00h 7:3 Reserved: Set to 0.

2 SMI generation for VGA memory range B8000h to BFFFFh: 0 = Disable; 1 = Enable

1 SMI generation for VGA memory range B0000h to B7FFFh: 0 = Disable; 1 = Enable.

0 SMI generation for VGA memory range A0000h to AFFFFh: 0 = Disable; 1 = Enable

Index BAh-BDh VGAM Register (R/W) Default Value = xxxxxxxxh 31 SMI generation for address range AF800h to AFFFFh: 0 = Disable; 1 = Enable. 30 SMI generation for address range AF000h to AF7FFh: 0 = Disable; 1 = Enable. 29 SMI generation for address range AE800h to AEFFFh: 0 = Disable; 1 = Enable. 28 SMI generation for address range AE000h to AE7FFh: 0 = Disable; 1 = Enable. 27 SMI generation for address range AD800h to ADFFFh: 0 = Disable; 1 = Enable. 26 SMI generation for address range AD000h to AD7FFh: 0 = Disable; 1 = Enable. 25 SMI generation for address range AC800h to ACFFFh: 0 = Disable; 1 = Enable. 24 SMI generation for address range AC000h to AC7FFh: 0 = Disable; 1 = Enable. 23 SMI generation for address range AB800h to ABFFFh: 0 = Disable; 1 = Enable. 22 SMI generation for address range AB000h to AB7FFh: 0 = Disable; 1 = Enable. 21 SMI generation for address range AA800h to AAFFFh: 0 = Disable; 1 = Enable. 20 SMI generation for address range AA000h to AA7FFh: 0 = Disable; 1 = Enable. 19 SMI generation for address range A9800h to A9FFFh: 0 = Disable; 1 = Enable. 18 SMI generation for address range A9000h to A97FFh: 0 = Disable; 1 = Enable. 17 SMI generation for address range A8800h to A8FFFh: 0 = Disable; 1 = Enable. 16 SMI generation for address range A8000h to A87FFh: 0 = Disable; 1 = Enable. 15 SMI generation for address range A7800h to A7FFFh: 0 = Disable; 1 = Enable. 14 SMI generation for address range A7000h to A77FFh: 0 = Disable; 1 = Enable. 13 SMI generation for address range A6800h to A6FFFh: 0 = Disable; 1 = Enable. 12 SMI generation for address range A6000h to A67FFh: 0 = Disable; 1 = Enable. 11 SMI generation for address range A5800h to A5FFFh: 0 = Disable; 1 = Enable. 10 SMI generation for address range A5000h to A57FFh: 0 = Disable; 1 = Enable. 9 SMI generation for address range A4800h to A4FFFh: 0 = Disable; 1 = Enable. 8 SMI generation for address range A4000h to A47FFh: 0 = Disable; 1 = Enable. 7 SMI generation for address range A3800h to A3FFFh: 0 = Disable; 1 = Enable. 6 SMI generation for address range A3000h to A37FFh: 0 = Disable; 1 = Enable. 5 SMI generation for address range A2800h to A2FFFh: 0 = Disable; 1 = Enable. 4 SMI generation for address range A2000h to A27FFh: 0 = Disable; 1 = Enable. 3 SMI generation for address range A1800h to A1FFFh: 0 = Disable; 1 = Enable. 2 SMI generation for address range A1000h to A17FFh: 0 = Disable; 1 = Enable. 1 SMI generation for address range A0800h to A0FFFh: 0 = Disable; 1 = Enable. 0 SMI generation for address range A0000h to A07FFh: 0 = Disable; 1 = Enable.

Revision 3.1 171 www.national.com Virtual Subsystem Architecture (Continued) Geode™ GXm Processor

5.2.3.6 VGA Range Detection

The VGA range detection circuit is similar to the SMI gen- eration hardware, however, it resides in the bus controller address mapping unit. The purpose of this hardware is to notify the graphics pipeline when accesses to the VGA memory range A0000h to BFFFFh are detected. The graphics pipeline has VGA read and write path hardware to process VGA memory accesses. The VGA range detection can be configured to trap VGA memory accesses in one or more of the following ranges: A0000h to AFFFFh (EGA,VGA), B0000h to B7FFFh (MDA), or B8000h to BFFFFh (CGA).

5.2.3.7 VGA Sequencer

The VGA sequencer is located at the front end of the graphics pipeline. The purpose of the VGA sequencer is to divide up multiple-byte read and write operations into a sequence of single-byte read and write operations. 16-bit or 32-bit X-bus write operations to VGA memory are divided into 8-bit write operations and sent to the VGA write path. 16-bit or 32-bit X-bus read operations from VGA memory are accumulated from 8-bit read operations over the VGA read path. The sequencer generates the lower two bits of the address.

5.2.3.8 VGA Write/Read Path

The VGA write path implements standard VGA write oper- ations into VGA memory. No SMI is generated for write path operations when the VGA access is not displayed. When the VGA access is displayed, an SMI is generated so that the SMI emulation can update the frame buffer. The VGA write path converts 8-bit write operations from the sequencer into 32-bit VGA memory write operations. The operations performed by the VGA write path include data rotation, raster operation (ALU), bit masking, plane select, plane enable, and write modes. The VGA read path implements standard VGA read oper- ations from VGA memory. No SMI is needed for read-path operations. The VGA read path converts 32-bit read oper- ations from VGA memory to 8-bit data back to the sequencer. The basic operations performed by the VGA read path include color compare, plane-read select, and read modes.

5.2.3.9 VGA Address Generator

The VGA address generator translates VGA memory addresses up to address where the VGA memory resides on the GXm processor. The VGA address generator requires the address from the VGA access (A0000h to BFFFFh), the base of the VGA memory on the GXm pro- cessor, and various control bits. The control bits are nec- essary because addressing is complicated by odd/even and Chain 4 addressing modes.

5.2.3.10 VGA Memory

The VGA memory requires 256 KB of memory organized as 64 KB by 32 bits. The VGA memory is implemented as part of system memory. The GXm processor partitions system memory into two areas, normal system memory and graphics memory. System memory is mapped to the normal physical address of the DRAM, starting at zero and ending at memory size. Graphics memory is mapped into high physical memory, contiguous to the registers and dedicated cache of the GXm processor. The graphics memory includes the frame buffer, compression buffer, cursor memory, and VGA memory. The VGA memory is mapped on a 256 KB boundary to simplify the address generation.

5.2.4 VGA Video BIOS

The video BIOS supports the VESA BIOS Extensions (VBE) Version 1.2 and 2.0, as well as all standard VGA BIOS calls. It interacts with Virtual VGA through the use of several extended VGA registers. These are virtual regis- ters contained in the VSA code for Virtual VGA. (These registers are defined in a separate document.)

www.national.com 172 Revision 3.1 Virtual Subsystem Architecture (Continued) Geode™ GXm Processor

5.2.5 Virtual VGA Register Descriptions

This section describes the registers contained in the graphics pipeline used for VGA emulation. The graphics pipeline maps 200h locations starting at GX_BASE+8100h. Refer to Section 4.1.2 “Control Regis- ters” on page 94 for instructions on accessing these regis- ters. The registers are summarized in Table 5-4, followed by detailed bit formats in T able 5-5 on page 173. Table 5-4. Virtual VGA Register Summary GX_BASE+ Memory Offset Type Function Default Value 8210h-8213h R/W GP_VGA_BASE VGA Graphics Pipeline VGA Memory Base Address Register — Specifies the offset of the VGA memory, starting from the base of graphics memory. xxxxxxxxh 8214h-8217h R/W GP_VGA_LATCH Graphics Pipeline VGA Display Latch Register — Provides a memory mapped way to read or write the VGA display latch. xxxxxxxxh 8140h-8143h R/W GP_VGA_WRITE Graphics Pipeline VGA Write Patch Control Register — Controls the VGA mem- ory write path in the graphics pipeline. xxxxxxxxh 8144h-8147h R/W GP_VGA_READ Graphics Pipeline VGA Read Patch Control Register — Controls the VGA mem- ory read path in the graphics pipeline. 00000000h

Revision 3.1 173 www.national.com Virtual Subsystem Architecture (Continued) Geode™ GXm Processor Table 5-5. Virtual VGA Registers Bit Name Description GX_BASE+8210h-8213h GP_VGA_BASE (R/W) Default Value = xxxxxxxxh 31:14 RSVD Reserved: Set to 0. 13:8 VGA_BASE (RO) Base Address (Read Only):The VGA base address is added to the graphics memory base to specify where VGA memory starts. The VGA base address provides longword address bits 19:14 when mapping VGA accesses into graphics memory. This allows the VGA base address to start on any 64 KB boundary within the 4 MB of graphics memory. 7:6 RSVD Reserved: Set to 0. 5:0 VGA_BASE (WO) Base Address (Write Only):The VGA base address is added to the graphics memory base to specify where VGA memory starts. The VGA base address provides longword address bits 19:14 when mapping VGA accesses into graphics memory. This allows the VGA base address to start on any 64 KB boundary within the 4 MB of graphics memory. GX_BASE+8214h-8217h GP_VGA_LATCH Register (R/W) Default Value = xxxxxxxxh 31:0 LATCH Display Latch:Specifies the value in the VGA display latch. VGA read operations cause VGA frame-buffer data to be latched in the display latch. VGA write operations can use the display latch as a source of data for VGA frame-buffer write operations. GX_BASE+8140h-8143h GP_VGA_WRITE Register (R/W) Default Value = xxxxxxxxh 31:28 RSVD Reserved: Set to 0. 27:24 MAP_MASK Map Mask: Enables planes 3 through 0 for writing. Combined with chain control to determine the final enables. 23:21 RSVD Reserved: Set to 0. 20 W3 Write Mode 3:Selects write mode 3 by using the bit mask with the rotated data. 19 W2 Write Mode 2:Selects write mode 2 by controlling set/reset. 18:16 RC Rotate Count:Controls the eight bit rotator. 15:12 SRE Set/Reset Enable:Enables the set/reset value for each plane. 11:8 SR Set/Reset:Selects 1 or 0 for each plane if enabled. 7:0 BIT_MASK Bit Mask:Selects data from the data latches (last read data). GX_BASE+8144h-8147h GP_VGA_READ Register (R/W) Default Value = 00000000h 31:18 RSVD Reserved: Set to 0. 17:16 RMS Read Map Select:Selects which plane to read in read mode 0 (Chain 2 and Chain 4 inactive). 15 F15 Force Address Bit 15:Forces address bit 15 to 0. 14 PC4 Packed Chain 4:— Provides 64 KB of packed pixel addressing when used with Chain 4 mode. This bit causes the VGA addresses to be shifted right by 2 bits. 13 C4 Chain 4 Mode:Selects Chain 4 mode for both read operations and write operations. 12 PB Page Bit:Becomes LSB of address if COE is set high. 11 COE Chain Odd/Even:Selects PB rather than A0 for least-significant VGA address bit. 10 W2 W r i t eC h a i n2M o d e :Selects Chain 2 mode for write operations. 9R 2 Read Chain 2 Mode:Selects Chain 2 mode for read operations. 8R M Read Mode: Selects between read mode 0 (normal) and read mode 1 (color compare). 7:4 CCM Color Compare Mask: Selects planes to include in the color comparison (read mode 1). 3:0 CC Color Compare:Specifies value of each plane for color comparison (read mode 1).

www.national.com 174 Revision 3.1 Geode™ GXm Processor

6.0 Power Management

The power management resources provided by a com- bined Geode GXm processor and CS5530-based system have been designed to support a full-featured notebook implementation. The extent to which these resources are employed depends on the application and the discretion of the system designer. The three greatest power consumers in a notebook sys- tem are the display, the hard drive and the CPU. Manag- ing power for the first two is relatively straightforward and is discussed in the Geode CS5530 I/O companion speci- fication. Managing CPU power can be more difficult since detecting inactive (Idle) states by monitoring external activity is imperfect as well as inefficient. The GXm processor and CS5530 I/O companion chip contain the most advanced power management features for reducing the power consumption of the processor in the system while delivering the highest performance in any mobile processor. The GXm processor supports the following CPU power management features:  APM Support  CPU Suspend Command Registers (CS5530)  Suspend Modulation 3 V o l t S u s p e n d  GXm Integrated Processor Serial Bus

6.1 APM SUPPORT

Many notebook computers rely solely on the APM (Advanced Power Management) driver for DOS, Windows 3.1 and Windows 95 operating systems to manage power to the CPU. APM provides several services that enhance the system power management by determining when the CPU is idle. For the CPU, APM is theoretically the best approach but there are some drawbacks. 1. APM is an OS-specific driver which may not be avail- able for some operating systems. 2. Application support is inconsistent. Some applica- tions in foreground may prevent idle calls. The components for APM support are:  Software CPU Suspend control via the CS5530 CPU Suspend Command Register (ACh).  Software SMI entry via the Software SMI Register (D0h). This allows the APM BIOS to be part of the SMI handler.

6.2 CPU SUSPEND COMMAND REGISTERS

Power management system software can invoke the SUSP#/SUSPA# protocol with the “CPU Suspend Com- mand” and the “Suspend Notebook Command” registers in the CS5530. If the SUSP#/SUSPA# protocol is invoked, all pending SMIs are serviced and SMI# is deasserted. Then SUSP# is asserted by the CS5530 and, subse- quently, SUSPA# is returned by the GXm processor. When a condition that ends the “Suspend” state exists, SMI# is re-asserted. At this point, if the PLL in the GXm processor has not been stopped, then SUSP# is deas- serted. SUSP# is never deasserted until SUSPA# has been sampled active (low). Note: The SMI# pin is a unidirectional line from the CS5530 to the GXm processor. It is active low. When SMI is initiated from a normal mode, the SMI# pin is asserted low and is held low until the SMI source is cleared. At that time, SMI# is de- asserted.

6.3 SUSPEND MODULATION

The hardware provided to support the GXm processor’s power management works by assuming that the GXm processor is Idle and reducing power until activity is detected. Most power management schemes in the indus- try run the system at full speed until a period of inactivity is detected. National Semiconductor’s more aggressive approach yields lower power consumption. When activity is detected, the GXm processor is instantly converted to full speed for a programmed duration. This is called Sus- pend Modulation. Suspend Modulation acts as backup for cases where APM doesn’t correctly detect an Idle condition in the sys- tem. As long as it is enabled, it will only become active in the background. The “Suspend Modulation Enable Regis- ter” in the CS5530 enables the Suspend Modulation fea- ture. The “Suspend Modulation ON Count Register” in the CS5530 is an 8-bit counter that represents the number of 32 µs intervals that the SUSP# pin will be asserted to the GXm processor. This counter, together with the “Suspend Modulation OFF Count Register” and the IRQ/Video Speedup Registers, performs the Suspend Modulation function for GXm processor’s power management. The ratio of the on count to the off count sets up an effective (emulated) clock frequency, allowing the power manager in the system to reduce the GXm processor’s power con- sumption. 6.4 3-VOLT SUSPEND MODE The GXm processor and CS5530 support stopping the processor and system clocks using the 3-Volt Suspend Mode. If configured (refer to the CS5530 specification), the CS5530 asserts the SUSP_3V pin after the SUSP#/SUSPA# handshake. SUSP_3V is intended to be connected to the output enable of a clock synthesizer or buffer chip so that the clocks to the GXm processor (SYSCLK), the CS5530 (PCI_CLK), and other system devices are stopped. The SUSP_3V pin is asserted on any write to the CS5530’s “CPU Suspend Command Reg- ister” or “Suspend Notebook Command Register” with bit 0 of the “Clock Stop Control Register” set. The GXm processor has two low-power Suspend modes. The mode implemented is determined by bit 0 in the PM Clock Stop Control Register. One mode (bit 0 clear) turns off the internal clocks to everything except the internal dis- play and memory controllers, thereby keeping the display active. The second mode, which is lower power, turns off all internal clocks generated from SYSCLK. This mode is

Revision 3.1 175 www.national.com Power Management (Continued) Geode™ GXm Processor selected by setting bit 0 in the PM Clock Stop Control Register. If you are using DRAMs without self refresh, you must supply a 32 kHz clock to the CLK32KHZ bit to keep t h er e f r e s hc i r c u i t r ya c t i v ew h e nu s i n gt h el o w e r - p o w e r Suspend mode. While also in 3-Volt Suspend Mode, the CS5530 contin- ues to decrement all of its device timers, and it responds to external SMI interrupts using the 32 kHz clock input (CLK32KHz) pin. Any SMI event, timer or pin, causes the CS5530 to deassert the SUSP_3V pin, starting the sys- tem clocks. The CS5530 holds SUSP# active for a pre- programmed period that varies from 0 to 16 ms, which allows the clocks to settle. After this period expires, the CS5530 deasserts SUSP#. SMI# is held active for the entire period, so that the GXm processor status registers are updated. The SUSP_3V pin can be active either high or low. The pin is an input during POR, and is sampled to determine its inactive state. This allows a designer to match the active state of SUSP_3V to the inactive state for a clock driver output enable with a pull-up or pull-down resistor.

6.5 SUSPEND MODE AND BUS CYCLES

The following subsections describe the bus cycles when the Suspend mode is implemented.

6.5.1 Initiating Suspend with SUSP#

The GXm processor has two low-power Suspend modes. The mode is selected by bit 0 in the PM Clock Stop Con- trol Register. One mode (bit 0 cleared) turns off the inter- nal clocks to everything but the internal Display and Memory Controllers, keeping the display active. A lower- power mode turns off all internal clocks generated from SYSCLK. This mode is selected by setting bit 0 in the PM Clock Stop Control Register. If the bit is set and DRAMS without self-refresh are used, a 32 KHz clock must be supplied to the CLK32KHZ input to keep the refresh circuit active. The GXm processor enters the Suspend mode in response to SUSP# input assertion only when certain conditions are met. First, the USE_SUSP bit must be set in CCR2 (Index C2h[7]). In addition, execution of the cur- rent instructions and any pending decoded instructions and associated bus cycles must be completed. SUSP# is sampled on the rising edge of SYSCLK, and must meet specified setup and hold times to be recognized at a par- ticular SYSCLK edge. When all conditions are met, the SUSPA# output is asserted. The time from assertion of SUSP# to the activa- tion of SUSPA# depends on which instructions were decoded prior to assertion of SUSP#. Normally, once SUSP# has been sampled inactive the SUSPA# output will be deactivated within two clocks. However, the deacti- vation of SUSPA# may be delayed until the end of an active refresh cycle. If the CPU is already in a Suspend mode initiated by SUSP#, one occurrence of NMI, INTR and SMI# is stored for execution after Suspend mode is exited. The CPU also allows PCI accesses during a SUSP#-initiated Suspend mode (see Figure 6-1). If the CPU is in the middle of a PCI access when SUSP# is asserted, the assertion of SUSPA# will be delayed until the PCI access is com- pleted. Figure 6-1. SUSP#-Initiated Suspend Mode SYSCLK SUSP# SUSPA#

www.national.com 176 Revision 3.1 Power Management (Continued) Geode™ GXm Processor

6.5.2 Initiating Suspend with HALT

The CPU also enters Suspend mode as a result of execut- ing a HALT instruction if the SUSP_HALT bit in CCR2 (Index C2h[3]) is set. Suspend mode is then exited upon recognition of an NMI, an unmasked INTR, or an SMI#. Normally SUSPA# is deactivated within six SYSCLKS from the detection of an active interrupt. However, the deactivation of SUSPA# may be delayed until the end of an active refresh cycle. The CPU also allows PCI accesses during a HALT -initi- ated Suspend mode. If the CPU is in the middle of a PCI access when the Halt instruction is executed, the asser- tion of SUSPA# will be delayed until the PCI access is completed. Figure 6-2. HALT-Initiated Suspend Mode SYSCLK FRAME# C/BE[3:0]# AD[15:0] IRDY# INTR, NMI, SUSPA# O XI I XX SMI# HALT

Revision 3.1 177 www.national.com Power Management (Continued) Geode™ GXm Processor

6.5.3 Responding to a PCI Access During Suspend

The GXm processor can temporarily exit Suspend mode to handle PCI accesses. If an unmasked REQx# is asserted, the GXm processor will deassert SUSPA# and exit the Suspend mode to respond to the PCI access. A PCI access is completed when FRAME# is inactive and TRDY# or STOP# are active. If SUSP# is asserted when the PCI access is completed, the GXm processor will assert SUSPA# and return to a SUSP#-initiated Suspend mode. If it was a HALT -initiated Suspend mode and no active interrupts have been recognized, the CPU will a s s e r tS U S P A #a n dr e t u r nt oaH A L T - i n i t i a t e dS u s p e n d mode. Figure 6-3. PCI Access During Suspend Mode SYSCLK REQx# TRDY# SUSP# SUSPA# FRAME#

www.national.com 178 Revision 3.1 Power Management (Continued) Geode™ GXm Processor

6.5.4 Stopping the Input Clock

Because the GXm processor is a static device, the input clock (SYSCLK) can be stopped and restarted without any loss of internal CPU data. If DRAMS are used that do not have self-refresh, bit 0 of the PM Clock Stop Control Register must be set to a one and the CLK32KHZ input must be continuously applied to keep the refresh circuitry running. The SYSCLK input can be stopped at either a logic high or logic low state. The required sequence for stopping SYSCLK is to initiate CPU Suspend mode, wait for the assertion of SUSPA# by the processor, and then stop the input clock. The CPU remains suspended until SYSCLK is restarted and the Suspend mode is exited as described earlier. While SYSCLK is stopped, the processor can no longer sample and respond to any input stimulus including REQx#, NMI, SMI#, INTR, and RESET inputs. Figure 6-4 illustrates the recommended sequence for stopping the SYSCLK using SUSP# to initiate Suspend mode. SYSCLK may be started prior to or following nega- tion of the SUSP# input. The figure includes the SUSP_3V pin from the CS5530 which is used to stop the external clocks. Figure 6-4. Stopping SYSCLK During Suspend Mode SYSCLK SUSP# SUSP_3V SMI Event, Timer or Pin SUSPA# (CS5530)

Revision 3.1 179 www.national.com Power Management (Continued) Geode™ GXm Processor

6.6 GXM PROCESSOR SERIAL BUS

The power management logic of the GXm processor pro- vides the CS5530 with information regarding the GXm processor productivity. If the GXm processor is deter- mined to be relatively inactive, the GXm processor power consumption can be greatly reduced by entering the Sus- pend Modulation mode. Although the majority of the system power management logic is implemented in the CS5530, a small amount of logic is required within the GXm processor to provide information from the graphics controller that is not exter- nally visible otherwise. The GXm processor implements a simple serial communications mechanism to transmit the CPU status to the CS5530. The GXm processor accumu- lates CPU events in a 8-bit register, “PM Serial Packet Register” (GX_BASE+850Ch, see Table 6-1), which is serially transmitted out of the GXm processor every 1 to 10 µs. The transmission frequency is set with the “PM Serial Packet Control Register” (GX_BASE+8504h, see Table 6-1).

6.6.1 Serial Packet Transmission

The GXm processor transmits the contents of the “PM Serial Packet Register” on the SERIALP output pin to the PSERIAL input pin of the CS5530. The GXm processor holds SERIALP low until the transmission interval counter (GX_BASE+8504h[4:3]) has elapsed. Once the counter has elapsed, PSERIAL is held high for two SYSCLKs to indicate the start of packet transmission. The contents of the packet register are then shifted out starting from bit 7 down to bit 0. PSERIAL is held high for one SYSCLK to indicate the end of packet transmission and then remains low until the next transmission interval. After the packet transmission has completed, the packet contents are cleared.

6.7 POWER MANAGEMENT REGISTERS

The GXm processor contains the power management registers for the serial packet transmission control, the user-defined power management address space, Sus- pend Refresh, and SMI status for Suspend/Resume. These registers are memory mapped (GX_BASE+8500h- 8FFFh) in the address space of the GXm processor and are described in the following sections. Refer to Section 4.1.2 “Control Registers” on page 94 for instructions on accessing these registers. Note, however, the PM_BASE and PM_MASK registers a r ea c c e s s e dw i t ht h eC P U _ R E A Da n dC P U _ W R I T E instructions. Refer to Section 4.1.6 “CPU_READ/CPU_WRITE Instructions” on page 99 for more information regarding these instructions. Table 6-1 summarizes the above mentioned registers. T a b l e s6 - 2a n d6 - 3s t a r t i n go np a g e1 8 0g i v et h e s er e g i s - ter’s bit formats. Table 6-1. Power Management Register Summary GX_BASE+ Memory Offset Type Name/Function Default Value Control and Status Registers 8500h-8503h R/W PM_STAT_SMI PM SMI Status Register — Contains System Management Mode (SMM) status i n f o r m a t i o nu s e db yS o f t V G A . xxxxxx00h 8504h-8507h R/W PM_CNTRL_TEN PM Serial Packet Control Register — Sets the serial packet transmission frequency and enables specific CPU events to be recorded in the serial packet. xxxxxx00h 8508h-850Bh R/W PM_CNTRL_CSTP PM Clock Stop Control Register — Enables the 3-V Suspend Mode for the GXm processor. xxxxxx00h 850Ch-850Fh R/W PM_SER_PACK PM Serial Packet Register — Transmits the contents of the serial packet. xxxxxx00h Programmable Address Region Registers FFFF FF6Ch R/W PM_BASE PM Base Register — Contains the base address for the programmable memory range decode. This register, in combination with the PM_MASK register, is used to generate a memory range decode which sets bit 1 in the serial transmission packet. 00000000h FFFF FF7Ch R/W PM_MASK PM Mask Register — The address mask for the PM_BASE register 00000000h

www.national.com 180 Revision 3.1 Power Management (Continued) Geode™ GXm Processor Table 6-2. Power Management Control and Status Registers Bit Name Description GX_BASE+8500h-8503h PM_STAT_SMI Register (R/W) Default Value = xxxxxx00h 31:8 RSVD Reserved — These bits are not used. Do not write to these bits. 7:3 RSVD Reserved — Set to 0. 2S M I _ M E M SMI VGA Emulation Memory — This bit is set high if a SMI was generated for VGA emulation in response to a VGA memory access. An SMI can be generated on a memory access to one of three regions in the A0000h-to-BFFFFh range as specified in the BC_XMAP_1 register. 1S M I _ I O SMI VGA Emulation I/O— This bit is set high if a SMI was generated for VGA emulation in response to an I/O access. An SMI can be generated on a I/O access to one of three regions in the 3B0h-to-3DFh range as specified in the BC_XMAP_1 register. 0S M I _ P I N SMI Pin— When set high, this bit indicates that the SMI# input pin has been asserted to the GXm processor. Note: These bits are “sticky” bits and can only be cleared with a write of ‘1’ to the respective bit. GX_BASE+8504h-8507h PM_CNTRL_TEN Register (R/W) Default Value = xxxxxx00h 31:8 RSVD Reserved — These bits are not used. Do not write to these bits. 7:6 RSVD Reserved — Set to 0. 5X _ T E S T ( W O ) Transmission Test (Write Only)— Setting this bit causes the GXm processor to immediately trans- mit the current contents of the serial packet. This bit is write only and is used primarily for test. This bit returns 0 on a read. 4:3 X_FREQ Transmission Frequency— This field indicates the time between serial packet transmissions. Serial packet transmissions occur at the selected interval only if at least one of the packet bits is set high: 00 = Disable transmitter; 01 = 1 ms; 10 = 5 ms; 11 = 10 ms. 2C P U _ R D CPU Activity Read Enable— Setting this bit high enables reporting of CPU Level-1 cache read misses that are not a result of an instruction fetch. This bit is a don’t-care if the CMEN bit is not set high 1C P U _ E N CPU Activity Master Enable— Setting this bit high enables reporting of CPU Level-1 cache misses in bit 6 of the serial transmission packet. When enabled, the CPU Level-1 cache miss activity is reported on any read (assuming the CREN is set high) or write access excluding misses that resulted from an instruction fetch. 0V I D _ E N Video Event Enable— Setting this bit high enables video decode events to be reported in bit 0 of the serial transmission packet. CPU or graphics-pipeline accesses to the graphics memory and dis- play-controller-register accesses are also reported. GX_BASE+8508h-850Bh PM_CNTRL_CSTP Register (R/W) Default Value = xxxxxx00h 31:8 RSVD Reserved — These bits are not used. Do not write to these bits. 7:1 RSVD Reserved — Set to 0. 0C L K _ S T P Clock Stop— This bit configures the GXm processor for Suspend Refresh Mode or 3-Volt Suspend Mode: 0 = Suspend Refresh Mode. The clocks to the memory and display controller are active. 1 = 3-Volt Suspend Mode. All internal clocks are stopped. Note: When this register is set high and the Suspend input pin (SUSP#) is asserted, the GXm processor stops all it’s internal clocks, and asserts the Suspend Acknowledge output pin (SUSPA#). Once SUSPA# is asserted the GXm processor’s SYSCLK input can be stopped. If this register is cleared, the internal memory-controller and display-controller clocks are not stopped on the SUSP#/SUSPA# sequence, and the SYSCLK input can not be stopped. GX_BASE+850Ch-850Fh PM_SER_PACK Register (R/W) Default Value = xxxxxx00h 31:8 RSVD Reserved — These bits are not used. Do not write to these bits. 7V I D _ I R Q Video IRQ— This bit indicates the occurrence of a video vertical sync pulse. This bit is set at the same timer that the VINT (Vertical Interrupt) bit is set in the DC_TIMING_CFG register. The VINT bit has a corresponding enable bit (VIEN) in the DC_TIM_CFG register. 6C P U _ A C T CPU Activity— This bit indicates the occurrence of a level 1 cache miss that was not a result of an instruction fetch. This bit has a corresponding enable bit in the PM_CNTL_TEN register. 5:2 RSVD Reserved — Set to 0. 1U S R _ D E F Programmable Address Decode — This bit indicates the occurrence of a programmable memory address decode. This bit is set based on the values of the PM_BASE register and the PM_MASK register. The PM_BASE register can be initialized to any address in the full 128 MB address range.

Revision 3.1 181 www.national.com Power Management (Continued) Geode™ GXm Processor 0V I D _ D E C Video Decode — This bit indicates that the CPU has accessed either the Display Controller regis- ters or the graphics memory region. This bit has a corresponding enable bit in the PM_CNTRL_TEN. Note: The GXm processor transmits the contents of the serial packet only when a bit in the packet register is set and the interval counter has elapsed. The CS5530 decodes the serial packet after each transmission. Once a bit in the packet is set, it will remain set until the completion of the next packet transmission. Successive events of the same type that occur between packet transmissions are ignored. Multiple unique events between packet transmissions will accumulate in this register. Table 6-2. Power Management Control and Status Registers (Continued) Bit Name Description Table 6-3. Power Management Programmable Address Region Registers Bit Name Description Index FFFFFF6Ch PM_BASE Register (R/W) Default Value = 0000000h 31:28 RSVD Reserved — Set to 0. 27:2 BASE_ADDR Base Address — This is the word-aligned base address for the programmable memory range com- pare. The actual address range is determined with this field and the PM_MASK register value. 1:0 RSVD Reserved — Set to 0. Index FFFFFF7Ch PM_MASK Register (R/W) Default Value = 0000000h 31:28 RSVD Reserved — Set to 0. 27:2 ADR_MASK Address Mask — This field is the address mask for the BASE_ADDR field in the PM_BASE regis- ter. If a bit in the ADR_MASK field is cleared the corresponding bit in the BASE_ADDR field must match the processor address. If a bit in the mask field is set high, the corresponding bit in the BASE_ADDR field always compares. If the processor cycle type matches the values of the WE and RE bits, and all bits in the BADD field match the processor address based on the ADR_MASK field, bit 1 will be set high in the serial transmission packet. 1W E Write Enable— Compare memory write cycles with BASE_ADDR and ADR_MASK: 0 = Disable; 1 = Enable. 0R E Read Enable — Compare memory read cycles with BASE_ADDR and ADR_MASK: 0 = Disable; 1 = Enable

www.national.com 182 Revision 3.1 Geode™ GXm Processor

7.0 Electrical Specifications

This section provides information on electrical connec- tions, absolute maximum ratings, required operating con- ditions, DC characteristics, and AC characteristics for the Geode GXm processor. All voltage values in the Electrical Specifications are with respect to V SS unless otherwise noted. For detailed information on the PCI bus electrical specification refer to Chapter 4 of the PCI Bus Specifica- tion, Revision 2.1.

7.1 PART NUMBERS

The following part numbers designate the various speeds available. For all speeds, the V CC2 voltage is 2.9V nominal and the VCC3 voltage is 3.3V nominal.

7.2 ELECTRICAL CONNECTIONS

7.2.1 Power/Ground Connections and Decoupling

Testing and operating the GXm processor requires the use of standard high frequency techniques to reduce par- asitic effects. These effects can be minimized by filtering the DC power leads with low-inductance decoupling capacitors, using low-impedance wiring, and by utilizing all of the V CC2 ,VCC3 , and VSS pins.

7.2.2 Power Sequencing

With two voltages connected to the GXm processor, it is important that the voltages come up in the correct order. VCC2 should come up at or before VCC3 . There are no additional timing requirements related to this sequence.

7.2.3 NC-Designated Pins

Pins designated NC (No Connection) should be left dis- connected. Connecting an NC pin to a pull-up/-down resistor, or an active signal could cause unexpected results and possible circuit malfunctions.

7.2.4 Pull-Up and Pull-Down Resistors

Table 7-2 lists the input pins that are internally connected to a 20-kohm pull-up/-down resistor. When unused, these inputs do not require connection to an external pull-up/- down resistor.

7.2.5 Unused Input Pins

All inputs not used by the system designer and not listed in T able 7-2 should be kept at either ground or VCC3 .T o prevent possible spurious operation, connect active-high inputs to ground through a 20-kohm (±10%) pull-down resistor and active-low inputs to VCC3 through a 20-kohm (±10%) pull-up resistor. Table 7-1. Part Numbers Core Frequency (MHz) Temperature (Degree C) Part Marking 266 70 GXm-266P 2.9V 70C 85 GXm-266P 2.9V 85C 70 GXm-266B 2.9V 70C 85 GXm-266B 2.9V 85C 233 70 GXm-233P 2.9V 70C 85 GXm-233P 2.9V 85C 70 GXm-233B 2.9V 70C 85 GXm-233B 2.9V 85C 200 70 GXm-200P 2.9V 70C 85 GXm-200P 2.9V 85C 70 GXm-200B 2.9V 70C 85 GXm-200B 2.9V 85C 180 70 GXm-180P 2.9V 70C 85 GXm-180P 2.9V 85C 70 GXm-180B 2.9V 70C 85 GXm-180B 2.9V 85C Note: B=B G AP a c k a g e P=S P G AP a c k a g e T a b l e7 - 2 . P i n sw i t h2 0 - k o h m Internal Resistor Signal Name BGA Ball No. SPGA Pin No. PU/PD SUSP#* H2 M4 Pull-up FRAME# A8 C13 Pull-up IRDY# C9 D14 Pull-up TRDY# B9 B14 Pull-up STOP# C11 A15 Pull-up LOCK# B11 B16 Pull-up DEVSEL# A9 E15 Pull-up PERR# A11 D16 Pull-up SERR# C12 A17 Pull-up REQ[2:0]# D3, H3, E3, K2, Pull-up TCLK J2 P4 Pull-up TMS H1 N3 Pull-up TDI D2 F4 Pull-up TEST F3 J5 Pull-down Note: *SUSP# is pulled up when not active.

Revision 3.1 183 www.national.com Electrical Specifications (Continued) Geode™ GXm Processor

7.3 ABSOLUTE MAXIMUM RATINGS

Table 7-3 lists absolute maximum ratings for the GXm pro- cessor. Stresses beyond the listed ratings may cause per- manent damage to the device. Exposure to conditions beyond these limits may (1) reduce device reliability and (2) result in premature failure even when there is no immediately apparent sign of failure. Prolonged exposure to conditions at or near the absolute maximum ratings may also result in reduced useful life and reliability. These are stress rat- ings only and do not imply that operation under any condi- tions other than those listed under T able 7-4 on page 184 is possible. Table 7-3. Absolute Maximum Ratings Parameter Min Max Units Notes Operating Case Temperature –65 110 °C Power Applied Storage Temperature –65 150 °C No Bias Supply Voltage 3.2 V Voltage On Any Pin –0.5 6.0 V Input Clamp Current, I IK –0.5 10 mA Power Applied Output Clamp Current, IOK 25 mA Power Applied

www.national.com 184 Revision 3.1 Electrical Specifications (Continued) Geode™ GXm Processor

7.4 OPERATING CONDITIONS

Table 7-4 lists the operating conditions for the GXm processor. Table 7-4. Operating Conditions Symbol Parameter Min Max Units Notes TC Operating Case Temperature 0 70 °C For Desktop Applications TC Operating Case Temperature 0 85 °C For Notebook Applications VCC2 Supply Voltage (2.9V nominal) 2.75 3.05 V VCC3 Supply Voltage (3.3V nominal) 3.14 3.46 V VIH High-Level Input Voltage: All input and I/O pins except SDRAM Interface and SYSCLK 2.0 5.5 V Note 1 SDRAM Interface 2.0 V CC3 +0.5 V Note 2 SYSCLK 2.7 5.5 V Note 1 VIL Low-Level Input Voltage: All except PCI bus and SYSCLK –0.5 0.8 V PCI bus –0.5 0.3*V CC3 V SYSCLK –0.5 0.4 V IOH High-Level Output Current –2 mA V O =V OH (Min) IOL Low-Level Output Current 5 mA V O =V OL (Max) Notes: 1) This parameter indicates that these pins are tolerant to the PCI 5 Volt Signaling Environment DC specification. 2) SDRAM Interface Pins: BA[1:0], CAS[A:B]#, CKE[A:B], CS[3:0]#, DQM[7:0], MA[12:0], MD[63:0], RASA#, RASB#, SDCLK_IN, SDCLK_OUT, SDCLK[3:0], TEST[3:0], WE[A:B]#

Revision 3.1 185 www.national.com Electrical Specifications (Continued) Geode™ GXm Processor

7.5 DC CHARACTERISTICS

DC characteristics were measured under the operating conditions listed in Table 7-4 on page 184. Table 7-5. DC Characteristics Symbol Parameter Min Max Units Notes VOL Output Low Voltage 0.4 V I OL =5m A VOH Output High Voltage 2.4 V I OH =– 2m A II Input Leakage Current for all input pins except those with internal PU/PDs ±10 µA0 < V IN <V CC3 , See Table 7-2 IIH Input Leakage Current for all pins with internal PDs. 200 µAV IH =2 . 4V , See Table 7-2 IIL Input Leakage Current for all pins with internal PUs. –400 µAV IL=0 . 3 5V , See Table 7-2 ICC Active ICC : Core ICC 2a tfCLK = 180 MHz I/O ICC 3a tfCLK = 180 MHz 2.25 0.400 AN o t e 1 Core ICC 2a tfCLK = 200 MHz I/O ICC 3a tfCLK = 200 MHz 2.55 0.405 Core ICC 2a tfCLK = 233 MHz I/O ICC 3a tfCLK = 233M Hz 2.85 0.410 Core ICC 2a tfCLK = 266 MHz I/O ICC 3a tfCLK = 266 MHz 3.10 0.415 ICCSM Suspend Mode ICC : Core ICC 2a tfCLK = 180 MHz I/O ICC 3a tfCLK = 180 MHz mA Notes 1 and 4 Core ICC 2a tfCLK = 200 MHz I/O ICC 3a tfCLK = 200 MHz Core ICC 2a tfCLK = 233 MHz I/O ICC 3a tfCLK = 233 MHz Core ICC 2a tfCLK = 266 MHz I/O ICC 3a tfCLK = 266 MHz ICCSS Standby ICC (Suspend and CLK Stopped): Core ICC 2a tfCLK =0M H z I/O ICC 3a tfCLK =0M H z mA Notes 1 and 3 C IN Input Capacitance 16 pF f = 1 MHz, Note 2 C OUT Output or I/O Capacitance 16 pF f = 1 MHz, Note 2 C CLK CLK Capacitance 12 pF f = 1 MHz, Note 2 Notes: 1. fCLK ratings refer to internal clock frequency. 2. Not 100% tested. 3. All inputs are at 0.2 V or VCC3 – 0.2 (CMOS levels). All inputs are held static and all outputs are unloaded (static IOUT = 0 mA). 4. All inputs are at 0.2 V or VCC3 – 0.2 (CMOS levels). All inputs except clock are held static and all outputs are unloaded (static IOUT = 0 mA).

www.national.com 186 Revision 3.1 Electrical Specifications (Continued) Geode™ GXm Processor

7.6 AC CHARACTERISTICS

The following tables list the AC characteristics including output delays, input setup requirements, input hold requirements and output float delays. The rising-clock- edge reference level V REF, and other reference levels are shown in Table 7-6. Input or output signals must cross these levels during testing. Input setup and hold times are specified minimums that define the smallest acceptable sampling window for which a synchronous input signal must be stable for correct oper- ation. All AC tests are at V CC2 = 2.75V to 3.05V (2.9V nominal), TC =0 oCt o7 0oCo r8 5o,C L = 50 pF unless otherwise specified. Figure 7-1. Drive Level and Measurement Points for Switching Characteristics Table 7-6. Drive Level and Measurement Points for Switching Characteristics Symbol Voltage (V) VREF 1.5 VIHD 2.4 VILD 0.4 CLK OUTPUTS INPUTS VIHD VILD VREF Valid Input Valid Outputn+1Valid Outputn VREF VREF VILD VIHD Min Max Legend: A = Maximum Output Delay Specification B = Minimum Output Delay Specification C = Minimum Input Setup Specification D = Minimum Input Hold Specification TX B A CD

Revision 3.1 187 www.national.com Electrical Specifications (Continued) Geode™ GXm Processor Figure 7-1 SYSCLK Timing and Measurement Points Table 7-7. Clock Signals Symbol ParameterT

180 MHz (6x)

(Note)

200 MHz (6x)

(Note)

233 MHz (7x)

(Note)

266 MHz (8x)

(Note) UnitsMin Max Min Max Min Max Min Max t1 SYSCLK Period 33.3 30.0 30.0 30.0 ns t2 SYSCLK Period Stability ±250 ±250 ±250 ±250 ps t3 SYSCLK High Time 10 10 10 10 ns t4 SYSCLK Low Time 10 10 10 10 ns t7 DCLK Period 7.3 7.3 7.3 7.3 ns t8 DCLK Rise/Fall Time 3.0 3.0 3.0 3.0 ns t9 SDCLK_OUT, SDCLK[3:0] Period 14.5 19.5 13 17 11 16 10 13 ns t10 SDCLK_OUT, SDCLK[3:0] High Time 7.5 6.5 5.5 5 ns t11 SDCLK_OUT, SDCLK[3:0] Low Time 7.5 6.5 5.5 5 ns t12 SDCLK_OUT, SDCLK[3:0] Fall Time t13 SDCLK_OUT, SDCLK[3:0] Rise Time Note: SDCLK timings (t9-t13) assume an SDCLK that is a "divide by 3" from the internal core clock. Hence: 180 MHz (6x) = 60.0 MHz SDCLK 200 MHz (6x) = 66.7 MHz SDCLK 233 MHz (7x) = 77.7 MHz SDCLK 266 MHz (8x) = 88.7 MHz SDCLK SYSCLK 1.5V VIH (Min) VIL (Max) t6 t5t4

Revision 3.1 191 www.national.com Electrical Specifications (Continued) Geode™ GXm Processor Table 7-11. Video Interface Signals Figure 7-8. Graphics Port Timing Symbol Parameter Min Max Unit t1 PCLK Period 7.4 40 ns t2 PCLK High Time 3 ns t3 PCLK Low Time 3 ns t4 PIXEL[17:0], CRT_HSYNC, CRT_VSYNC, FP_HSYNC, FP_VSYNC, ENA_DISP Valid Delay from PCLK Rising Edge 25 n s t5 VID_CLK Period 8.5 ns t6 VID_RDY Setup to VID_CLK Rising Edge 5 ns t7 VID_RDY Hold to VID_CLK Rising Edge 2 ns t8 VID_VAL, VID_DATA[7:0] Valid Delay from VID_CLK Rising Edge 2 5 ns t9 DCLK Period 7.4 ns t10 DCLK Rise/Fall Time 3 ns tcyc DCLK Duty Cycle 40 60 % t2 t3 PCLK PIXEL[17:0], CRT_HSYNC, CRT_VSYNC, FP_HSYNC, FP_VSYNC, ENA_DISP

Revision 3.1 193 www.national.com Electrical Specifications (Continued) Geode™ GXm Processor Table 7-12. JTAG AC Specification Figure 7-11. TCK Timing and Measurement Points Symbol Parameter Min Max Unit TCK Frequency (MHz) 25 MHz t1 TCK Period 40 ns t2 TCK High Time 10 ns t3 TCK Low Time 10 ns t4 TCK Rise Time 4 ns t5 TCK Fall Time 4 ns t6 TDO Valid Delay 3 25 ns t7 Non-test Outputs Valid Delay 3 25 ns t8 TDO Float Delay 30 ns t9 Non-test Outputs Float Delay 36 ns t10 TDI, TMS Setup Time 8 ns t11 Non-test Inputs Setup Time 8 ns t12 TDI, TMS Hold Time 7 ns t13 Non-test Inputs Hold Time 7 ns TCK 1.5 V VIH(Min) VIL(Max) t4 t5

www.national.com 194 Revision 3.1 Electrical Specifications (Continued) Geode™ GXm Processor Figure 7-12. JTAG Test Timings TCK TDI, TMS 1.5 V t10 t12 t7 t9 t11 t13 TDO Output Signals Input Signals

Revision 3.1 195 www.national.com Geode™ GXm Processor

8.0 Package Specifications

The thermal characteristics and mechanical dimensions for the Geode GXm processor are provided on the follow- ing pages.

8.1 THERMAL CHARACTERISTICS

Table 8-1 shows the junction-to-case thermal resistance of the SPGA and BGA package and can be used to calcu- late the junction (die) temperature under any given cir- cumstance. Note that there is no specification for maximum junction temperature given since the operation of both SPGA and BGA devices are guaranteed to a case temperature range of 0°C to 85°C. As long as the case temperature of the device is maintained within this range, the junction tem- perature of the die will also be maintained within its allow- able operating range. However, the die (junction) temperature under a given operating condition can be cal- culated by using the following equation: T J =T C +( P* θJC) where: TJ = Junction temperature (°C) TC = Case temperature at top center of package (°C) P = Power dissipation (W) θJC = Junction-to-case thermal resistance (°C/W) As stated previously, a maximum junction temperature is not specified since a maximum case temperature is. Therefore, the following equation can be used to calculate the maximum thermal resistance required of the thermal solution for a given maximum ambient temperature: where: θ CS = Max case-to-heatsink thermal resistance (°C/W) allowed for thermal solution. θSA = Max heatsink-to-ambient thermal resistance (°C/W) allowed for thermal solution. TA = Max ambient temperature (°C) TC = Max case temperature at top center of package (°C) P = Max power dissipation (W) If thermal grease is used between the case and heatsink, θCS will reduce to about 0.01 °C/W. Therefore, the above equation can be simplified to: where: θCA = θCS = Max case-to-ambient thermal resistance (°C/W) allowed for thermal solution. The calculatedθCA value (examples shown in Table 8-2) represents the maximum allowed thermal resistance of the selected cooling solution which is required to maintain the 85°C case temperature for the application in which the device is used. These examples are given for reference only. The actual value used for Maximum Power (P) and ambient tempera- ture (T A) is determined by the system designer based on system configuration, extremes of the operating environ- ment, and whether active thermal management (via Sus- pend Modulation) of the processor is employed. Table 8-1. Junction-to-Case Thermal Resistance for SPGA and BGA Packages Package θJC SPGA 1.7 °C/W BGA 1.1 °C/W θC S θSA+ TC TA– θCA TC TA–

www.national.com 196 Revision 3.1 Package Specifications (Continued) Geode™ GXm Processor

8.1.1 Heatsink Considerations

As described previously, T able 8-2 shows the maximum allowed thermal resistance of a heatsink for particular operating environments. The calculated values, defined as θCA , represent the required ability of a particular heat- sink to transfer heat generated by the processor from its case into the air, thereby maintaining the case tempera- ture at or below 85°C. BecauseθCA is a measure of ther- mal resistivity, it is inversely proportional to the heatsink’s ability to dissipate heat or it’s thermalconductivity. Note: A" p e r f e c t "h e a t s i n kw o u l db ea b l et om a i n t a i na case temperature equal to that of the ambient air inside the system chassis. Looking at Table 8-2, it can be seen that as ambient tem- perature (TA) increases,θCA decreases, and that as power consumption of the processor (P) increases,θCA decreases. Thus, the ability of the heatsink to dissipate thermal energy must increase as the processor power increases and as the temperature inside the enclosure increases. WhileθCA is a useful parameter to calculate, heatsinks are not typically specified in terms of a singleθCA .T h i si s because the thermal resistivity of a heatsink is not con- stant across power or temperature. In fact, heatsinks become slightly less efficient as the amount of heat they are trying to dissipate increases. For this reason, heatsinks are typically specified by graphs that plot heat dissipation (in watts) vs. mounting surface (case) temperature rise above ambient (in °C). This method is necessary because ambient and case temperatures fluctuate constantly dur- ing normal operation of the system. The system designer must be careful to choose the proper heatsink by match- ing the requiredθ CA with the thermal dissipation curve of the device under the entire range of operating conditions in order to make sure that a case temperature of 85°C is never surpassed. Table 8-2. Case-to-Ambient Thermal Resistance Examples @ 85°C Core Voltage (VCC2 ) Core Frequency Maximum Power θCA for Different Ambient Temperatures (°C/W) 20°C 25°C 30°C 35°C 40°C 2.9V (Nominal)

Revision 3.1 197 www.national.com Package Specifications (Continued) Geode™ GXm Processor To choose the proper heatsink, the system designer must make sure that the calculatedθCA falls above the curve (shaded area). The curve itself defines the minimum tem- perature rise above ambient that the heatsink can main- tain. See Figure 8-1 as an example of a particular heatsink under consideration. Figure 8-1. Heatsink Example Example 1 Assume P (max) = 5W and TA (max) = 40°C. Therefore: In this case, the heatsink under consideration is more than adequate since at 5W worst case, it can maintain a 40°C case temperature rise above ambient (θCA = 9) when a maximum of 45°C (θCA = 8) is required. Example 2 Assume P (max) = 10W and TA (max) = 40°C. Therefore: In this case, the heatsink under consideration is NOT ade- quate to maintain the 45°C case temperature rise above ambient for a 9W processor. For more information on thermal design considerations or heatsink properties, refer to the Product Selection Guide of any leading vendor of Thermal Engineering solutions. 24 68 10 θCA = 45/9 = 5 Heat Dissipated - Watts θCA = 45/5 = 9 Mounting Surface Temperature Rise Above Ambient – °C θCA TC TA– θCA 85 40–() θCA 9= θCA TC TA– θCA 85 40–() θCA 5=

www.national.com 198 Revision 3.1 Package Specifications (Continued) Geode™ GXm Processor

8.2 MECHANICAL PACKAGE OUTLINES

Dimensions for the BGA package are shown in Figure 8-2. Figure 8-3 shows the SPGA dimensions. Table 8-3 gives the legend for the symbols used in both package outlines. Figure 8-2. 352-Terminal BGA Mechanical Package Outline Sym Millimeters Inches M i nM a xM i nM a x A 1.45 2.23 0.057 0.088 A1 0.50 0.70 0.020 0.028 A2 0.43 0.83 0.017 0.033 aaa 0.20 0.008 B 0.60 0.90 0.024 0.035 D 34.80 35.20 1.370 1.386 D1 31.55 31.95 1.242 1.258 D2 32.80 35.20 1.291 1.386 E1 1.12 1.42 0.044 0.056 F 0.35 0.014 S1 1.42 1.82 0.056 0.072 F A01 Index Chamfer 1 . 5m mo nas i d e

45 Degree Angle

.889 REF. S1D1 D D 1.5 1.5 B A Seating Plane aaa Z Z

Revision 3.1 199 www.national.com Package Specifications (Continued) Geode™ GXm Processor Figure 8-3. 320-Pin SPGA Mechanical Package Outline F D A01 index mark .030" blank circle inside .060" filled circle to form donut

45 CHAMFER

2.29 1.52REF. (INDEX CORNER) 1.65 REF. D D o Sym Millimeters Inches Min Max Min Max A 2.51 3.07 0.099 0.121 B 0.43 0.51 0.017 0.020 D 49.28 49.91 1.940 1.965 D1 45.47 45.97 1.790 1.810 E1 2.41 2.67 0.095 0.105 E2 1.14 1.40 0.045 0.055 F -- 0.127 Diag -- 0.005 Diag L 2.97 3.38 0.117 0.133 S1 1.65 2.16 0.065 0.085 SEATING PLANE L E2 E1 B A

www.national.com 200 Revision 3.1 Package Specifications (Continued) Geode™ GXm Processor Table 8-3. Mechanical Package Outline Legend Symbol Meaning A Distance from seating plane datum to highest point of body A1 Solder ball height A2 Laminate thickness (excluding heat spreader) aaa Coplanarity B Pin or solder ball diameter D Largest overall package outline dimension D1 Length from outer pin center to outer pin center D2 Heat spreader outline dimension E1 BGA: Solder ball pitch SPGA: Linear spacing between true pin position centerlines E2 Diagonal spacing between true pin position centerlines F Flatness L Distance from seating plane to tip of pin S1 Length from outer pin/ball center to edge of laminate

Revision 3.1 201 www.national.com Geode™ GXm Processor

9.0 Instruction Set

This section summarizes the Geode GXm processor instruction set and provides detailed information on the instruction encodings. The instruction set is broken into four categories:  Processor Core Instruction Set - listed in T able 9-27 on page 213  FPU Instruction Set - listed in T able 9-29 on page 225  MMX Instruction Set - listed in T able 9-31 on page 230  National Semiconductor Extended MMX Instruction Set - listed in Table 9-33 on page 235 These tables provide information on the instruction encod- ing, and the instruction clock counts for each instruction. The clock count values for these tables are based on the following assumptions 1. All clock counts refer to the microprocessor core internal clock frequency. For example, clock doubled GXm processor cores will reference a clock frequency that is twice the bus frequency. 2. The instruction has been prefetched, decoded and is ready for execution. 3. Bus cycles do not require wait states. 4. There are no local bus HOLD requests delaying processor access to the bus. 5. No exceptions are detected during instruction execu- tion. 6. If an effective address is calculated, it does not use two general register components. One register, scaling and displacement can be used within the clock count shown. However, if the effective address calculation uses two general register components, add one clock to the clock count shown. 7. All clock counts assume aligned 32-bit memory/IO operands. 8. If instructions access a 32-bit operand on odd addresses, add one clock for read or write and add two clocks for read and write. 9. For non-cached memory accesses, add two clocks (clock doubled GXm processor cores) or four clocks (clock tripled GXm processor cores), assuming zero wait state memory accesses. 10. Locked cycles are not cacheable. Therefore, using the LOCK prefix with an instruction adds additional clocks as specified in item 9 above.

www.national.com 202 Revision 3.1 Instruction Set (Continued) Geode™ GXm Processor

9.1 GENERAL INSTRUCTION SET FORMAT

Depending on the instruction, the GXm processor core instructions follow the general instruction format shown in Table 9-1. These instructions vary in length and can start at any byte address. An instruction consists of one or more bytes that can include prefix bytes, at least one opcode byte, a mod r/m byte, an s-i-b byte, address displacement, and imme- diate data. An instruction can be as short as one byte and as long as 15 bytes. If there are more than 15 bytes in the instruction, a general protection fault (error code 0) is gen- erated. The fields in the general instruction format at the byte level are summarized in Table 9-2 and detailed in the fol- lowing subsections. Table 9-1. General Instruction Set Format Prefix (optional) Opcode Register and Address Mode Specifier Address Displacement Immediate Data mod r/m Byte s-i-b Byte mod reg r/m ss index base 0 or More Bytes 1 or 2 Bytes 7:6 5:3 2:0 7:6 5:3 2:0 0, 8, 16, or 32 Bits 0, 8, 16, or 32 Bits Table 9-2. Instruction Fields Field Name Description Prefix (optional) Prefix Field(s): One or more optional fields that are used to specify segment register override, address and operand size, repeat elements in string instruction, LOCK# assertion. Opcode Opcode Field: Identifies instruction operation. mod Address Mode Specifier: Used with r/m field to select addressing mode. reg General Register Specifier: Uses reg, sreg3 or sreg2 encoding depending on opcode field. r/m Address Mode Specifier: Used with mod field to select addressing mode. ss Scale factor: Determines scaled-index address mode. index Index: Determines general register to be used as index register. base Base: Determines general register to be used as base register. Address Displacement Displacement: Determines address displacement. Immediate Data Immediate Data: Immediate-data operand used by instruction.

Revision 3.1 203 www.national.com Instruction Set (Continued) Geode™ GXm Processor

9.1.1 Prefix (Optional)

Prefix bytes can be placed in front of any instruction to modify the operation of that instruction. When more than one prefix is used, the order is not important. There are five types of prefixes that can be used: 1. Segment Override explicitly specifies which segment register the instruction will use for effective address calculation. 2. Address Size switches between 16-bit and 32-bit addressing by selecting the non-default address size. 3. Operand Size switches between 16-bit and 32-bit operand size by selecting the non-default operand size. 4. Repeat is used with a string instruction to cause the instruction to be repeated for each element of the string. 5. Lock is used to assert the hardware LOCK# signal during execution of the instruction. Table 9-3 lists the encoding for different types of prefix bytes.

9.1.2 Opcode

The opcode field specifies the operation to be performed by the instruction. The opcode field is either one or two bytes in length and may be further defined by additional bits in the mod r/m byte. Some operations have more than one opcode, each specifying a different form of the opera- tion. Certain opcodes name instruction groups. For exam- ple, opcode 80h names a group of operations that have an immediate operand and a register or memory operand. The reg field may appear in the second opcode byte or in the mod r/m byte. The opcode may contain w, d, s and eee opcode fields as s h o w ni nt h eT a b l e9 - 2 7o np a g e2 1 3 9.1.2.1 w Field (Operand Size) When used, the 1-bit w field selects the operand size dur- ing 16-bit and 32-bit data operations. See Table 9-4. Table 9-3. Instruction Prefix Summary Prefix Encoding Description ES: 26h Override segment default, use ES for memory operand. CS: 2Eh Override segment default, use CS for memory operand. SS: 36h Override segment default, use SS for memory operand. DS: 3Eh Override segment default, use DS for memory operand. FS: 64h Override segment default, use FS for memory operand. GS: 65h Override segment default, use GS for memory operand. Operand Size 66h Make operand size attribute the inverse of the default. Address Size 67h Make address size attribute the inverse of the default. LOCK F0h Assert LOCK# hardware signal. REPNE F2h Repeat the following string instruction. REP/REPE F3h Repeat the following string instruction. Table 9-4. w Field Encoding w Field Operand Size 16-Bit Data Operations 32-Bit Data Operations 08 b i t s 8 b i t s 1 1 6b i t s 3 2b i t s

www.national.com 204 Revision 3.1 Instruction Set (Continued) Geode™ GXm Processor 9.1.2.2 d Field (Operand Direction) When used, the d field (bit 1) determines which operand is taken as the source operand and which operand is taken as the destination. See T able 9-5. 9.1.2.3 s Field (Immediate Data Field Size) When used, the s field (bit 1) determines the size of the immediate data field. If the s bit is set, the immediate field of the opcode is 8 bits wide and is sign-extended to match the operand size of the opcode. See Table 9-6. 9.1.2.4 eee Field (MOV-Instruction Register Selection) The eee field (bits [5:3]) is used to select the control, debug and test registers in the MOV instructions. The type of register and base registers selected by the eee field are listed in Table 9-7. The values shown in Table 9-7 are the only valid encodings for the eee bits. Table 9-5. d Field Encoding d Field Direction of Operation Source Operand Destination Operand

0 Register-to-Register

1 Register-to-Register

Table 9-6. s Field Encoding s Field Immediate Field Size 8-Bit Operand Size 16-Bit Operand Size 32-Bit Operand Size 0 (or not present) 8b i t s 1 6b i t s 3 2b i t s 18 b i t s 8 b i t s (sign-extended) 8b i t s (sign-extended) Table 9-7. eee Field Encoding eee Field Register Type Base Register

000 Control Register CR0

010 Control Register CR2

011 Control Register CR3

100 Control Register CR4

000 Debug Register DR0

001 Debug Register DR1

010 Debug Register DR2

011 Debug Register DR3

110 Debug Register DR6

111 Debug Register DR7

011 Test Register TR3

100 Test Register TR4

101 Test Register TR5

110 Test Register TR6

111 Test Register TR7

Revision 3.1 205 www.national.com Instruction Set (Continued) Geode™ GXm Processor 9.1.3 mod and r/m Byte (Memory Addressing) T h em o da n dr / mf i e l d sw i t h i nt h em o dr / mb y t e ,s e l e c tt h e type of memory addressing to be used. Some instructions use a fixed addressing mode (e.g., PUSH or POP) and therefore, these fields are not present. Table 9-8 lists the addressing method when 16-bit addressing is used and a mod r/m byte is present. Some mod r/m field encodings are dependent on the w field and are shown in Table 9-8. Table 9-8. General Registers Selected by mod r/m Fields and w Field mod r/m 16-Bit Operation 32-Bit Operation w = 0w = 1w = 0w = 1 11 000 AL AX AL EAX 11 001 CL CX CL ECX 11 010 DL DX DL EDX 11 011 BL BX BL EBX 11 100 AH SP AH ESP 11 101 CH BP CH EBP 11 110 DH SI DH ESI 11 111 BH DI BH EDI Table 9-9. mod r/m Field Encoding mod Field r/m Field 16-Bit Address Mode with mod r/m Byte 32-Bit Address Mode with mod r/m Byte and No s-i-b Byte Present 00 000 DS:[BX+SI] DS:[EAX] 00 001 DS:[BX+DI] DS:[ECX] 00 010 SS:[BP+SI] DS:[EDX] 00 011 SS:[BP+DI] DS:[EBX] 00 100 DS:[SI] s-i-b is present (See Table 9-15) 00 101 DS:[DI] DS:[d32] 00 110 DS:[d16] DS:[ESI] 00 111 DS:[BX] DS:[EDI] 01 000 DS:[BX+SI+d8] DS:[EAX+d8] 01 001 DS:[BX+DI+d8] DS:[ECX+d8] 01 010 SS:[BP+SI+d8] DS:[EDX+d8] 01 011 SS:[BP+DI+d8] DS:[EBX+d8] 01 100 DS:[SI+d8] s-i-b is present (See Table 9-15) 01 101 DS:[DI+d8] SS:[EBP+d8] 01 110 SS:[BP+d8] DS:[ESI+d8] 01 111 DS:[BX+d8] DS:[EDI+d8] 10 000 DS:[BX+SI+d16] DS:[EAX+d32] 10 001 DS:[BX+DI+d16] DS:[ECX+d32] 10 010 SS:[BP+SI+d16] DS:[EDX+d32] 10 011 SS:[BP+DI+d16] DS:[EBX+d32] 10 100 DS:[SI+d16] s-i-b is present (See Table 9-15) 10 101 DS:[DI+d16] SS:[EBP+d32] 10 110 SS:[BP+d16] DS:[ESI+d32] 10 111 DS:[BX+d16] DS:[EDI+d32] 11 xxx See Table 9-8. See Table 9-8 Note: Note: d8 refers to 8-bit displacement, and d16 refers to 16-bit displacement.

www.national.com 206 Revision 3.1 Instruction Set (Continued) Geode™ GXm Processor 9.1.4 reg Field The reg field (Table 9-10) determines which general regis- ters are to be used. The selected register is dependent on whether a 16- or 32-bit operation is current and on the sta- tus of the w bit. 9.1.4.1 sreg2 Field (ES, CS, SS, DS Register Selection) The sreg2 field (T able 9-11) is a 2-bit field that allows one of the four 286-type segment registers to be specified. 9.1.4.2 sreg3 Field (FS and GS Segment Register Selection) The sreg3 field (Table 9-12) is 3-bit field that is similar to the sreg2 field, but allows use of the FS and GS segment registers. Table 9-10. General Registers Selected by reg Field reg 16-Bit Operation 32-Bit Operation w = 0w = 1w = 0w = 1

000 AL AX AL EAX

001 CL CX CL ECX

010 DL DX DL EDX

011 BL BX BL EBX

100 AH SP AH ESP

101 CH BP CH EBP

110 DH SI DH ESI

111 BH DI BH EDI

Table 9-11. sreg2 Field Encoding sreg2 Field Segment Register Selected 00 ES 01 CS 10 SS 11 DS Table 9-12. sreg3 Field Encoding sreg3 Field Segment Register Selected 000 ES 001 CS 010 SS 011 DS 100 FS 101 GS

110 Undefined

111 Undefined

Revision 3.1 207 www.national.com Instruction Set (Continued) Geode™ GXm Processor 9.1.5 s-i-b Byte (Scale, Indexing, Base) The s-i-b fields provide scale factor, indexing and a base field for address selection. The ss, index and base fields are described next. 9.1.5.1 ss Field (Scale Selection) The ss field (T able 9-13) specifies the scale factor used in the offset mechanism for address calculation. The scale factor multiplies the index value to provide one of the com- ponents used to calculate the offset address. 9.1.5.2 index Field (Index Selection) The index field (T able 9-14) specifies the index register used by the offset mechanism for offset address calcula- tion. When no index register is used (index field = 100), the ss value must be 00 or the effective address is unde- fined.

9.1.5.3 Base Field (s-i-b Present)

In Table 9-9 on page 205, the note “s-i-b present” for cer- tain entries forces the use of the mod and base field as listed in Table 9-15. The first two digits in the first column of Table 9-15 identify the mod bits in the mod r/m byte. The last three digits in the first column of this table identi- f i e st h eb a s ef i e l d si nt h es - i - bb y t e . Table 9-13. ss Field Encoding ss Field Scale Factor 00 x1 01 x2 01 x4 11 x8 Table 9-14. index Field Encoding Index Field Index Register

000 EAX

001 ECX

010 EDX

011 EBX

101 EBP

110 ESI

111 EDI

Table 9-15. mod base Field Encoding mod Field within mode/rm Byte (bits 7:6) base Field within s-i-b Byte (bits 2:0) 32-Bit Address Mode with mod r/m and s-i-b Bytes Present 00 000 DS:[EAX+(scaled index)] 00 001 DS:[ECX+(scaled index)] 00 010 DS:[EDX+(scaled index)] 00 011 DS:[EBX+(scaled index)] 00 100 SS:[ESP+(scaled index)] 00 101 DS:[d32+(scaled index)] 00 110 DS:[ESI+(scaled index)] 00 111 DS:[EDI+(scaled index)] 01 000 DS:[EAX+(scaled index)+d8] 01 001 DS:[ECX+(scaled index)+d8] 01 010 DS:[EDX+(scaled index)+d8] 01 011 DS:[EBX+(scaled index)+d8] 01 100 SS:[ESP+(scaled index)+d8] 01 101 SS:[EBP+(scaled index)+d8] 01 110 DS:[ESI+(scaled index)+d8] 01 111 DS:[EDI+(scaled index)+d8] 10 000 DS:[EAX+(scaled index)+d32] 10 001 DS:[ECX+(scaled index)+d32] 10 010 DS:[EDX+(scaled index)+d32] 10 011 DS:[EBX+(scaled index)+d32] 10 100 SS:[ESP+(scaled index)+d32] 10 101 SS:[EBP+(scaled index)+d32] 10 110 DS:[ESI+(scaled index)+d32] 10 111 DS:[EDI+(scaled index)+d32]

www.national.com 208 Revision 3.1 Instruction Set (Continued) Geode™ GXm Processor

9.2 CPUID INSTRUCTION

The CPUID instruction (opcode 0FA2) allows the software to make processor inquiries as to the vendor, family, model, stepping, features and also provides cache infor- mation. The GXm with MMX supports both the standard and National Semiconductor extended CPUID levels. The presence of the CPUID instruction is indicated by the ability to change the value of the ID Flag, bit 21 in the EFLAGS register. The CPUID level allows the CPUID instruction to return different information in the EAX, EBX, ECX, and EDX reg- isters. The level is determined by the initialized value of the EAX register before the instruction is executed. A summary of the CPUID levels is shown in Table 9-16.

9.2.1 Standard CPUID Levels

The standard CPUID levels are part of the standard x86 instruction set.

9.2.1.1 CPUID Instruction with EAX = 00000000h

Standard function 0h (EAX = 0) of the CPUID instruction returns the maximum standard CPUID levels as well as the processor vendor string. After the instruction is executed, the EAX register contains the maximum standard CPUID levels supported. The maximum standard CPUID level is the highest acceptable value for the EAX register input. This does not include the extended CPUID levels. The EBX through EDX registers contain the vendor string of the processor as shown in Table 9-17.

9.2.1.2 CPUID Instruction with EAX = 00000001h

Standard function 01h (EAX = 1) of the CPUID instruction returns the processor type, family, model, and stepping information of the current processor in the EAX register (see Table 9-18). The EBX and ECX registers are reserved. The standard feature flags supported are returned in the EDX register as shown in Table 9-19 on page 209. Each flag refers to a specific feature and indicates if that feature is present on the processor. Some of these features have protection control in CR4. Before using any of these fea- tures on the processor, the software should check the cor- responding feature flag. Attempting to execute an unavailable feature can cause exceptions and unexpected behavior. For example, software must check bit 4 before attempting to use the Time Stamp Counter instruction. Table 9-16. CPUID Levels Summary CPUID Type Initialized EAX Register Returned Data in EAX, EBX, ECX, EDX Registers Standard 00000000h Maximum standard levels, CPU vendor string Standard 00000001h Model, family, type and features Standard 00000002h TLB and cache information Extended 80000000h Maximum extended levels Extended 80000001h Extended model, family, type and Extended 80000002h CPU marketing name string Extended 80000003h Extended 80000004h Extended 80000005h TLB and L1 cache description Table 9-17. CPUID Data Returned when EAX = 0 Register (Note) Returned Contents Description EAX 2 Maximum Standard Level EBX 69 72 7943 (iryC) Vendor ID String 1 EDX 73 6E 4978 (snlx) Vendor ID String 2 ECX 64 61 6574 (daet) Vendor ID String 3 Note: The register column is intentionally out of order. Table 9-18. EAX, EBX, ECX CPUID Data Returned when EAX = 1 Register Returned Contents Description EAX[3:0] xx Stepping ID EAX[7:4] 4 Model EAX[11:8] 5 Family EAX[15:12] 0 Type EAX[31:16] - Reserved EBX - Reserved ECX - Reserved

Revision 3.1 209 www.national.com Instruction Set (Continued) Geode™ GXm Processor

9.2.1.3 CPUID Instruction with EAX = 00000002h

Standard function 02h (EAX = 02h) of the CPUID instruc- tion returns information that is specific to the National Semiconductor family of processors. Information about the TLB is returned in EAX as shown in Table 9-20. Infor- mation about the L1 cache is returned in EDX. Table 9-19. EDX CPUID Data Returned when EAX = 1 EDX Returned Contents* Feature Flag CR4 Bit EDX[0] 1 FPU On-Chip - EDX[1] 0 Virtual Mode Extension - EDX[2] 0 Debug Extensions - EDX[3] 0 Page Size Extensions - EDX[4] 1 Time Stamp Counter 2 EDX[5] 1 RDMSR / WRMSR Instructions EDX[6] 0 Physical Address Extensions EDX[7] 0 Machine Check Exception - EDX[8] 1 CMPXCHG8B Instruction - EDX[9] 0 On-Chip APIC Hardware - EDX[10] 0 Reserved - EDX[11] 0 SYSENTER / SYSEXIT Instructions EDX[12] 0 Memory Type Range Registers EDX[13] 0 Page Global Enable - EDX[14] 0 Machine Check Architecture EDX[15] 1 Conditional Move Instructions EDX[16] 0 Page Attribute Table - EDX[22:17] 0 Reserved - EDX[23] 1 MMX Instructions - EDX[24] 0 Fast FPU Save and Restore EDX[31:25] 0 Reserved - Note: *0 = Not supported Table 9-20. Standard CPUID with EAX = 00000002h Register Returned Contents Description E A X x xx x7 0x x h T L Bi s3 2E n t r y ,4 - w a ys e ta s s o - ciative, and has 4 KByte Pages EAX xx xx xx 01h The CPUID instruction needs to be executed only once with an input value of 02h to retrieve complete information about the cache and TLB EBX Reserved ECX Reserved EDX xx xx xx 80h L1 cache is 16 KBytes, 4-way set associated, and has 16 bytes per line.

www.national.com 210 Revision 3.1 Instruction Set (Continued) Geode™ GXm Processor

9.2.2 Extended CPUID Levels

Testing for extended CPUID instruction support can be accomplished by executing a CPUID instruction with the EAX register initialized to 80000000h. If a value greater than or equal to 80000000h is returned to the EAX regis- ter by the CPUID instruction, the processor supports extended CPUID levels.

9.2.2.1 CPUID Instruction with

EAX = 8000000h Extended function 80000000h (EAX = 80000000h) of the CPUID instruction returns the maximum extended CPUID levels supported by the current processor in EAX (Table 9- 21). The EBX, ECX, and EDX registers are currently reserved.

9.2.2.2 CPUID Instruction with

EAX = 8000 0001h Extended function 80000001h (EAX = 80000001h) of the CPUID instruction returns the processor type, family, model, and stepping information of the current processor in EAX. The EBX and ECX registers are reserved. The extended feature flags supported are returned in the EDX register as shown in Table 9-23. Each flag refers to a specific feature and indicates if that feature is present on the processor. Some of these features have protection control in CR4. Before using any of these features on the processor, the software should check the corresponding feature flag. Table 9-21. Maximum Extended CPUID Level Register Returned Contents Description EAX 80000005h Maximum Extended CPUID Level (six levels) EBX - Reserved ECX - Reserved EDX - Reserved Table 9-22. EAX, EBX, ECX CPUID Data Returned when EAX = 80000001h Register Returned Contents Description EAX[3:0] xx Stepping ID EAX[7:4] 4 Model EAX[11:8] 5 Family EAX[15:12] 0 Processor Type EAX[31:16] - Reserved EBX - Reserved ECX - Reserved Table 9-23. EDX CPUID Data Returned when EAX = 80000001h EDX Returned Contents* Feature Flag CR4 Bit EDX[0] 1 FPU On-Chip - EDX[1] 0 Virtual Mode Extension - EDX[2] 0 Debugging Extension - EDX[3] 0 Page Size Extension (4 MB) EDX[4] 1 Time Stamp Counter 2 EDX[5] 1 National Semiconductor Model-Specific Registers (via RDMSR / WRMSR Instructions) EDX[6] 0 Reserved - EDX[7] 0 Machine Check Exception - EDX[8] 1 CMPXCHG8B Instruction - EDX[9] 0 Reserved - EDX[10] 0 Reserved - EDX[11] 0 SYSCALL / SYSRET Instruction EDX[12] 0 Reserved - EDX[13] 0 Page Global Enable - EDX[14] 0 Reserved - EDX[15] 1 Integer Conditional Move Instruction EDX[16] 0 FPU Conditional Move Instruction EDX[22:17] 0 Reserved - EDX[23] 1 MMX - EDX[24] 1 6x86MX Multimedia Extensions Note: 0 = Not supported

Revision 3.1 211 www.national.com Instruction Set (Continued) Geode™ GXm Processor

9.2.2.3 CPUID Instruction with

EAX = 80000002h, 80000003h, 80000004h Extended functions 80000002h through 80000004h (EAX = 80000002h, EAX = 80000003h, EAX = 80000004h) of the CPUID instruction return an ASCII string containing the name of the current processor. These functions elimi- nate the need to look up the processor name in a lookup table. Software can simply call these functions to obtain the name of the processor. The string may be 48 ASCII characters long, and is returned in little endian format. If the name is shorter than 48 characters long, the remain- ing bytes will be filled with ASCII NUL character (00h).

9.2.2.4 CPUID Instruction with

EAX = 80000005h Extended function 80000005h (EAX = 80000005h) of the CPUID instruction returns information about the TLB and L 1c a c h et ob el o o k e du pi nal o o k u pt a b l e .R e f e rt oT a b l e 9-25. Table 9-24. Official CPU Name 80000002h 80000003h 80000004h EAX CPU Name EAX CPU Name EAX CPU Name EBX CPU Name EBX CPU Name EBX CPU Name ECX CPU Name ECX CPU Name ECX CPU Name EDX CPU Name EDX CPU Name EDX CPU Name Table 9-25. Standard CPUID with EAX = 80000005h Register Returned Contents Description EAX -- Reserved EBX xx xx 70 xxh TLB is 32 Entry, 4-way set associative, and has 4 KByte Pages EBX xx xx xx 01h The CPUID instruction needs to be executed only once with an input value of 02h to retrieve complete information about the cache and TLB ECX xx xx xx 80h L1 cache is 16 KBytes, 4-way set associated, and has 16 bytes per line. EDX -- Reserved

www.national.com 212 Revision 3.1 Instruction Set (Continued) Geode™ GXm Processor

9.3 PROCESSOR CORE INSTRUCTION SET

The instruction set for the GXm processor core is summa- rized in Table 9-27 on page 213. The table uses several symbols and abbreviations that are described next and listed in Table 9-26. Opcodes Opcodes are given as hex values except when they appear within brackets as binary values. Clock Counts The clock counts listed in the instruction set summary table are grouped by operating mode (Real and Pro- tected) and whether there is a register/cache hit or a cache miss. In some cases, more than one clock count is s h o w ni nac o l u m nf o rag i v e ni n s t r u c t i o n ,o rav a r i a b l ei s used in the clock count. Flags There are nine flags that are affected by the execution of instructions. The flag names have been abbreviated and vari- ous conventions used to indicate what effect the instruc- tion has on the particular flag. Table 9-26. Processor Core Instruction Set Table Legend Symbol or Abbreviation Description Opcode # Immediate 8-bit data ## Immediate 16-bit data ### Full immediate 32-bit data (8, 16, 32 bits) + 8-bit signed displacement +++ Full signed displacement (16, 32 bits) Clock Count / Register operand/memory operand. n Number of times operation is repeated. L Level of the stack frame. | Conditional jump taken | Conditional jump not taken. (e.g. “4|1” = 4 clocks if jump taken, 1 clock if jump not taken) \\C P L ≤ IOPL \\ CPL > IOPL (where CPL = Current Privilege Level, IOPL = I/O Privilege Level) Flags OF Overflow Flag DF Direction Flag IF Interrupt Enable Flag TF Trap Flag SF Sign Flag ZF Zero Flag AF Auxiliary Flag PF Parity Flag CF Carry Flag x Flag is modified by the instruction. - Flag is not changed by the instruction. 0 Flag is reset to “0”. 1 Flag is set to “1”. u Flag is undefined following execution the instruction.

Revision 3.1 213 www.national.com Instruction Set (Continued) Geode™ GXm Processor Table 9-27. Processor Core Instruction Set Summary Instruction Opcode Flags Real Mode Prot’d Mode Real Mode Prot’d Mode ODI T SZ APC FFFFFFFFF Clock Count (Reg/Cache Hit) Notes AAA A S C I IA d j u s tA La f t e rA d d 3 7 u -- -uuxux 3 3 AAD ASCII Adjust AX before Divide D 5 0 A u- - - xxuxu 7 7 AAM ASCII Adjust AX after Multiply D 4 0 A u- - - xxuxu 1 9 1 9 AAS ASCII Adjust AL after Subtract 3 F u- - - uuxux 3 3 ADC Add with Carry R e g i s t e r t o R e g i s t e r 1 [ 0 0 d w ] [ 1 1 r e g r / m ] x- - - xxxxx 1 1 b h Register to Memory 1 [000w] [mod reg r/m] 1 1 Memory to Register 1 [001w] [mod reg r/m] 1 1 Immediate to Register/Memory 8 [00sw] [mod 010 r/m]### 1 1 Immediate to Accumulator 1 [010w] ### 1 1 ADD Integer Add R e g i s t e r t o R e g i s t e r 0 [ 0 0 d w ] [ 1 1 r e g r / m ] x- - - xxxxx 1 1 b h Register to Memory 0 [000w] [mod reg r/m] 1 1 Memory to Register 0 [001w] [mod reg r/m] 1 1 Immediate to Register/Memory 8 [00sw] [mod 000 r/m]### 1 1 Immediate to Accumulator 0 [010w] ### 1 1 AND Boolean AND Register to Register 2 [00dw] [11 reg r/m] 0 - - - x x u x 0 1 1 b h Register to Memory 2 [000w] [mod reg r/m] 1 1 Memory to Register 2 [001w] [mod reg r/m] 1 1 Immediate to Register/Memory 8 [00sw] [mod 100 r/m]### 1 1 Immediate to Accumulator 2 [010w] ### 1 1 ARPL Adjust Requested Privilege Level F r o m R e g i s t e r / M e m o r y 6 3 [ m o d r e g r / m ] -----x --- 9 a h BB0_Reset Set BLT Buffer 0 Pointer to the Base0F 3A 2 2 BB1_Reset Set BLT Buffer 1 Pointer to the Base0F 3B 2 2 BOUND Check Array Boundaries If In Range 77 BSF Scan Bit Forward R e g i s t e r , R e g i s t e r / M e m o r y 0 F B C [ m o d r e g r / m ] -----x --- 4 / 9 + n 4 / 9 + n b h BSR Scan Bit Reverse R e g i s t e r , R e g i s t e r / M e m o r y 0 F B D [ m o d r e g r / m ] -----x --- 4 / 1 1 + n4 / 1 1 + n b h BT Test Bit Register/Memory, Register 0F A3 [mod reg r/m] 1/7 1/7 BTC Test Bit and Complement Register/Memory, Register 0F BB [mod reg r/m] 2/8 2/8 BTR Test Bit and Reset Register/Memory, Register 0F B3 [mod reg r/m 2/8 2/8 BTS Test Bit and Set Register (short form) 0F AB [mod reg r/m] 2/8 2/8

www.national.com 214 Revision 3.1 Instruction Set (Continued) Geode™ GXm Processor CALL Subroutine Call Register/Memory Indirect Within Segment FF [mod 010 r/m] 3/4 3/4 Direct Intersegment - C a l lG a t et oS a m eP r i v i l e g e -Call Gate to Different Privilege No Par’s -Call Gate to Different Privilege m Par’s -16-bit Task to 16-bit TSS -16-bit Task to 32-bit TSS -16-bit Task to V86 Task -32-bit Task to 16-bit TSS -32-bit Task to 32-bit TSS -32-bit Task to V86 Task 9A [unsigned full offset, selector] 91 4 51+2m 183 189 123 186 192 126 Indirect Intersegment - C a l lG a t et oS a m eP r i v i l e g e -Call Gate to Different Privilege No Par’s -Call Gate to Different Privilege m Par’s -16-bit Task to 16-bit TSS -16-bit Task to 32-bit TSS -16-bit Task to V86 Task -32-bit Task to 16-bit TSS -32-bit Task to 32-bit TSS -32-bit Task to V86 Task FF [mod 011 r/m] 11 15 52+2m 184 190 124 187 193 127 CMOVA/CMOVNBE Move if Above/Not Below or Equal CMOVBE/CMOVNA Move if Below or Equal/Not Above CMOVAE/CMOVNB/CMOVNC Move if Above or Equal/Not Below/Not Carry CMOVB/CMOVC/CMOVNAE Move if Below/Carry/Not Above or Equal CMOVE/CMOVZ Move if Equal/Zero CMOVNE/CMOVNZ Move if Not Equal/Not Zero CMOVG/CMOVNLE Move if Greater/Not Less or Equal CMOVLE/CMOVNG Move if Less or Equal/Not Greater CMOVL/CMOVNGE Move if Less/Not Greater or Equal CMOVGE/CMOVNL Move if Greater or Equal/Not Less CMOVO Move if Overflow CMOVNO Move if No Overflow CMOVP/CMOVPE Move if Parity/Parity Even CMOVNP/CMOVPO Move if Not Parity/Parity Odd Table 9-27. Processor Core Instruction Set Summary (Continued) Instruction Opcode Flags Real Mode Prot’d Mode Real Mode Prot’d Mode ODI T SZ APC FFFFFFFFF Clock Count (Reg/Cache Hit) Notes

Revision 3.1 215 www.national.com Instruction Set (Continued) Geode™ GXm Processor CMOVS Move if Sign CMOVNS Move if Not Sign CMP Compare Integers R e g i s t e r t o R e g i s t e r 3 [ 1 0 d w ] [ 1 1 r e g r / m ] x- - - xxxxx 1 1 b h Register to Memory 3 [101w] [mod reg r/m] 1 1 Memory to Register 3 [100w] [mod reg r/m] 1 1 Immediate to Register/Memory 8 [00sw] [mod 111 r/m] ### 1 1 Immediate to Accumulator 3 [110w] ### 1 1 CMPS Compare String A [ 0 1 1 w ] x- - - xxxxx 6 6 b h CMPXCHG Compare and Exchange R e g i s t e r 1 , R e g i s t e r 2 0 F B [ 0 0 0 w ] [ 1 1 r e g 2 r e g 1 ] x- - - xxxxx 6 6 Memory, Register 0F B [000w] [mod reg r/m] 6 6 CPU_READ Read Special CPU Register 0F 3C 1 1 CPU_WRITE Write Special CPU Register 0F 3D 1 1 DAA Decimal Adjust AL after Add 2 7 ---- xxxxx 2 2 DAS Decimal Adjust AL after Subtract 2 F ---- xxxxx 2 2 DEC Decrement by 1 R e g i s t e r / M e m o r y F [ 1 1 1 w ] [ m o d 0 0 1 r / m ] x- - - xxxx- 1 1 b h Register (short form) 4 [1 reg] 1 1 DIV Unsigned Divide Accumulator by Register/Memory Divisor: Byte Word Doubleword F [ 0 1 1 w ] [ m o d 1 1 0 r / m ] ----x x u u - b,e e,h ENTER Enter New Stack Frame Level = 1 17 17 Level (L) > 1 17+2*L 17+2*L IDIVInteger (Signed) Divide Accumulator by Register/Memory Divisor: Byte Word Doubleword F [ 0 1 1 w ] [ m o d 1 1 1 r / m ] ----x x u u - b,e e,h IMUL Integer (Signed) Multiply Accumulator by Register/Memory Multiplier: Byte Word Doubleword F [011w] [mod 101 r/m] x - - - x x u u x bh Register with Register/Memory Multiplier: Word Doubleword 0F AF [mod reg r/m] Register/Memory with Immediate to Register2 Multiplier: Word Doubleword 6 [10s1] [mod reg r/m] ### IN Input from I/O Port Variable Port E [110w] 8 8/22 Table 9-27. Processor Core Instruction Set Summary (Continued) Instruction Opcode Flags Real Mode Prot’d Mode Real Mode Prot’d Mode ODI T SZ APC FFFFFFFFF Clock Count (Reg/Cache Hit) Notes

www.national.com 216 Revision 3.1 Instruction Set (Continued) Geode™ GXm Processor INC Increment by 1 R e g i s t e r / M e m o r y F [ 1 1 1 w ] [ m o d 0 0 0 r / m ] x- - - xxxx- 1 1 b h Register (short form) 4 [0 reg] 1 1 INT Software Interrupt I N T i C D # --x 0 ----- 1 9 b , e g , j , k , r Protected Mode: -Interrupt or Trap to Same Privilege -Interrupt or Trap to Different Privilege -16-bit Task to 16-bit TSS by Task Gate -16-bit Task to 32-bit TSS by Task Gate -16-bit Task to V86 by Task Gate -16-bit Task to 16-bit TSS by Task Gate -32-bit Task to 32-bit TSS by Task Gate -32-bit Task to V86 by Task Gate -V86 to 16-bit TSS by Task Gate -V86 to 32-bit TSS by Task Gate -V86 to Privilege 0 by Trap Gate/Int Gate 184 190 124 187 193 127 187 193 INT 3 CC INT INT INTO If OF==0 If OF==1 (INT 4) CE INT INT IRET Interrupt Return R e a l M o d e C F xxxxxxxxx 1 3 g , h , j , k , r Protected Mode: - W i t h i nT a s kt oS a m eP r i v i l e g e -Within Task to Different Privilege -16-bit Task to 16-bit Task -16-bit Task to 32-bit TSS -16-bit Task to V86 Task -32-bit Task to 16-bit TSS -32-bit Task to 32-bit TSS -32-bit Task to V86 Task 169 175 109 172 178 112 JB/JNAE/JC Jump on Below/Not Above or Equal/Carry Full Displacement 0F 82 +++ 1 1 JBE/JNA Jump on Below or Equal/Not Above Full Displacement 0F 86 +++ 1 1 JE/JZ Jump on Equal/Zero Full Displacement 0F 84 +++ 1 1 JL/JNGE Jump on Less/Not Greater or Equal Full Displacement 0F 8C +++ 1 1 JLE/JNG Jump on Less or Equal/Not Greater Full Displacement 0F 8E +++ 1 1 Table 9-27. Processor Core Instruction Set Summary (Continued) Instruction Opcode Flags Real Mode Prot’d Mode Real Mode Prot’d Mode ODI T SZ APC FFFFFFFFF Clock Count (Reg/Cache Hit) Notes

Revision 3.1 217 www.national.com Instruction Set (Continued) Geode™ GXm Processor JMP Unconditional Jump Full Displacement E9 +++ 1 1 Register/Memory Indirect Within Segment FF [mod 100 r/m] 1/3 1/3 Direct Intersegment -Call Gate Same Privilege Level -16-bit Task to 16-bit TSS -16-bit Task to 32-bit TSS -16-bit Task to V86 Task -32-bit Task to 16-bit TSS -32-bit Task to 32-bit TSS -32-bit Task to V86 Task EA [unsigned full offset, selector] 81 2 186 192 126 189 195 129 Indirect Intersegment -Call Gate Same Privilege Level -16-bit Task to 16-bit TSS -16-bit Task to 32-bit TSS -16-bit Task to V86 Task -32-bit Task to 16-bit TSS -32-bit Task to 32-bit TSS -32-bit Task to V86 Task FF [mod 101 r/m] 10 13 187 193 127 190 196 130 JNB/JAE/JNC Jump on Not Below/Above or Equal/Not Carry Full Displacement 0F 83 +++ 1 1 JNBE/JA Jump on Not Below or Equal/Above Full Displacement 0F 87 +++ 1 1 JNE/JNZ Jump on Not Equal/Not Zero Full Displacement 0F 85 +++ 1 1 JNL/JGE Jump on Not Less/Greater or Equal Full Displacement 0F 8D +++ 1 1 JNLE/JG Jump on Not Less or Equal/Greater Full Displacement 0F 8F +++ 1 1 JNO Jump on Not Overflow Full Displacement 0F 81 +++ 1 1 JNP/JPO Jump on Not Parity/Parity Odd Full Displacement 0F 8B +++ 1 1 JNS Jump on Not Sign Full Displacement 0F 89 +++ 1 1 JO Jump on Overflow Full Displacement 0F 80 +++ 1 1 JP/JPE Jump on Parity/Parity Even Full Displacement 0F 8A +++ 1 1 JS Jump on Sign Full Displacement 0F 88 +++ 1 1 LAR Load Access Rights F r o m R e g i s t e r / M e m o r y 0 F 0 2 [ m o d r e g r / m ] -----x --- 9 a g , h , j , p Table 9-27. Processor Core Instruction Set Summary (Continued) Instruction Opcode Flags Real Mode Prot’d Mode Real Mode Prot’d Mode ODI T SZ APC FFFFFFFFF Clock Count (Reg/Cache Hit) Notes

www.national.com 218 Revision 3.1 Instruction Set (Continued) Geode™ GXm Processor LEA Load Effective Address With Index Register 11 LLDT Load LDT Register LMSW Load Machine Status Word LSL Load Segment Limit F r o m R e g i s t e r / M e m o r y 0 F 0 3 [ m o d r e g r / m ] -----x --- 9 a g , h , j , p LTR Load Task Register MOV Move Data Register to Memory 8 [100w] [mod reg r/m] 1 1 Register/Memory to Register 8 [101w] [mod reg r/m] 1 1 Immediate to Register/Memory C [011w] [mod 000 r/m] ### 1 1 Immediate to Register (short form) B [w reg] ### 1 1 Memory to Accumulator (short form) A [000w] +++ 1 1 Accumulator to Memory (short form) A [001w] +++ 1 1 Register/Memory to Segment Register 8E [mod sreg3 r/m] 1 6 Segment Register to Register/Memory 8C [mod sreg3 r/m] 1 1 MOV Move to/from Control/Debug/Test Regs CR0/CR2/CR3/CR4 to Register 0F 20 [11 eee reg] 6 6 Register to DR0-DR3 0F 23 [11 eee reg] 10 10 DR0-DR3 to Register 0F 21 [11 eee reg] 9 9 Register to DR6-DR7 0F 23 [11 eee reg] 10 10 DR6-DR7 to Register 0F 21 [11 eee reg] 9 9 Register to TR3-5 0F 26 [11 eee reg] 16 16 TR3-5 to Register 0F 24 [11 eee reg] 8 8 Register to TR6-TR7 0F 26 [11 eee reg] 11 11 TR6-TR7 to Register 0F 24 [11 eee reg] 3 3 MOVSX Move with Sign Extension MOVZX Move with Zero Extension MUL Unsigned Multiply Accumulator with Register/Memory Multiplier: Byte Word Doubleword F [011w] [mod 100 r/m] x - - - x x u u x bh Table 9-27. Processor Core Instruction Set Summary (Continued) Instruction Opcode Flags Real Mode Prot’d Mode Real Mode Prot’d Mode ODI T SZ APC FFFFFFFFF Clock Count (Reg/Cache Hit) Notes

Revision 3.1 219 www.national.com Instruction Set (Continued) Geode™ GXm Processor NEG Negate Integer F [ 0 1 1 w ] [ m o d 0 1 1 r / m ] x- - - xxxxx 1 1 b h OIO Official Invalid Opcode 0 F F F --x 0 ----- 1 8 - 1 2 5 OR Boolean OR Register to Register 0 [10dw] [11 reg r/m] 0 - - - x x u x 0 1 1 b h Register to Memory 0 [100w] [mod reg r/m] 1 1 Memory to Register 0 [101w] [mod reg r/m] 1 1 Immediate to Register/Memory 8 [00sw] [mod 001 r/m] ### 1 1 Immediate to Accumulator 0 [110w] ### 1 1 OUT Output to Port Variable Port E [111w] 14 14/28 POP Pop Value off Stack Register (short form) 5 [1 reg] 1 1 Segment Register (ES, SS, DS) [000 sreg2 111] 1 6 Segment Register (FS, GS) 0F [10 sreg3 001] 1 6 POPF Pop Stack into FLAGS 9 D xxxxxxxxx 8 8 b h , n PREFIX BYTES Address Size Prefix 67 Operand Size Prefix 66 Segment Override Prefix -CS -DS -ES -FS -GS -SS PUSH Push Value onto Stack Register (short form) 5 [0 reg] 1 1 Segment Register (ES, CS, SS, DS) [000 sreg2 110] 1 1 Segment Register (FS, GS) 0F [10 sreg3 000] 1 1 Immediate 6 [10s0] ### 1 1 RCL Rotate Through Carry Left RCR Rotate Through Carry Right 32+4n bh , m Table 9-27. Processor Core Instruction Set Summary (Continued) Instruction Opcode Flags Real Mode Prot’d Mode Real Mode Prot’d Mode ODI T SZ APC FFFFFFFFF Clock Count (Reg/Cache Hit) Notes

www.national.com 220 Revision 3.1 Instruction Set (Continued) Geode™ GXm Processor 39+4n bh , m REPE CMPS Compare String F i n d n o n - m a t c h F 3 A [ 0 1 1 w ] x- - - xxxxx 1 1 + 4 n 1 1 + 4 n b h REPE SCAS Scan String F i n d n o n - A L / A X / E A X F 3 A [ 1 1 1 w ] x- - - xxxxx 9 + 3 n 9 + 3 n b h REPNE CMPS Compare String F i n d m a t c h F 2 A [ 0 1 1 w ] x- - - xxxxx 1 1 + 4 n 1 1 + 4 n b h REPNE SCAS Scan String F i n d A L / A X / E A X F 2 A [ 1 1 1 w ] x- - - xxxxx 9 + 3 n 9 + 3 n b h RET Return from Subroutine Within Segment Adding Immediate to SP C2 ## 3 3 Intersegment CB 10 13 Intersegment Adding Immediate to SP CA ## 10 13 Protected Mode: Different Privilege Level -Intersegment -Intersegment Adding Immediate to SP ROL Rotate Left ROR Rotate Right RSM Resume from SMM Mode 0 F A A xxxxxxxxx 5 7 5 7 s s SAHF Store AH in FLAGS 9 E ---- xxxxx 1 1 SAL Shift Left Arithmetic Register/Memory by 1 D[000w] [mod 100 r/m] x - - - x x u x x 1 1 b h Register/Memory by CL D[001w] [mod 100 r/m] u - - - x x u x x 2 2 Register/Memory by Immediate C[000w] [mod 100 r/m] # u - - - x x u x x 1 1 SAR Shift Right Arithmetic Register/Memory by 1 D[000w] [mod 111 r/m] x - - - x x u x x 2 2 b h Register/Memory by CL D[001w] [mod 111 r/m] u - - - x x u x x 2 2 Register/Memory by Immediate C[000w] [mod 111 r/m] # u - - - x x u x x 2 2 SBB Integer Subtract with Borrow R e g i s t e r t o R e g i s t e r 1 [ 1 0 d w ] [ 1 1 r e g r / m ] x- - - xxxxx 1 1 b h Register to Memory 1[100w] [mod reg r/m] 1 1 Memory to Register 1[101w] [mod reg r/m] 1 1 Immediate to Register/Memory 8[00sw] [mod 011 r/m] ### 1 1 Immediate to Accumulator (short form) 1[110w] ### 1 1 SCAS Scan String A [ 1 1 1 w ] x- - - xxxxx 2 2 b h SETB/SETNAE/SETC Set Byte on Below/Not Above or Equal/Carry SETBE/SETNA Set Byte on Below or Equal/Not Above SETE/SETZ Set Byte on Equal/Zero Table 9-27. Processor Core Instruction Set Summary (Continued) Instruction Opcode Flags Real Mode Prot’d Mode Real Mode Prot’d Mode ODI T SZ APC FFFFFFFFF Clock Count (Reg/Cache Hit) Notes

Revision 3.1 221 www.national.com Instruction Set (Continued) Geode™ GXm Processor SETL/SETNGE Set Byte on Less/Not Greater or Equal SETLE/SETNG Set Byte on Less or Equal/Not Greater SETNB/SETAE/SETNC Set Byte on Not Below/Above or Equal/Not Carry SETNBE/SETA Set Byte on Not Below or Equal/Above SETNE/SETNZ S e tB y t eo nN o tE q u a l / N o tZ e r o SETNL/SETGE Set Byte on Not Less/Greater or Equal SETNLE/SETG Set Byte on Not Less or Equal/Greater SETNO Set Byte on Not Overflow SETNP/SETPO S e tB y t eo nN o tP a r i t y / P a r i t yO d d SETNS Set Byte on Not Sign SETO Set Byte on Overflow SETP/SETPE Set Byte on Parity/Parity Even SETS Set Byte on Sign SGDT Store GDT Register SIDT Store IDT Register SLDT Store LDT Register STR Store Task Register SHL Shift Left Logical Register/Memory by 1 D [000w] [mod 100 r/m] x - - - x x u x x 1 1 b h Register/Memory by CL D [001w] [mod 100 r/m] u - - - x x u x x 2 2 Register/Memory by Immediate C [000w] [mod 100 r/m] # u - - - x x u x x 1 1 SHLD S h i f tL e f tD o u b l e Register/Memory by Immediate 0F A4 [mod reg r/m] # u - - - x x u x x 3 3 b h Register/Memory by CL 0F A5 [mod reg r/m] 6 6 SHR Shift Right Logical Register/Memory by 1 D [000w] [mod 101 r/m] x - - - x x u x x 2 2 b h Register/Memory by CL D [001w] [mod 101 r/m] u - - - x x u x x 2 2 Register/Memory by Immediate C [000w] [mod 101 r/m] # u - - - x x u x x 2 2 SHRD Shift Right Double Register/Memory by Immediate 0F AC [mod reg r/m] # u - - - x x u x x 3 3 b h Register/Memory by CL 0F AD [mod reg r/m] 6 6 Table 9-27. Processor Core Instruction Set Summary (Continued) Instruction Opcode Flags Real Mode Prot’d Mode Real Mode Prot’d Mode ODI T SZ APC FFFFFFFFF Clock Count (Reg/Cache Hit) Notes

www.national.com 222 Revision 3.1 Instruction Set (Continued) Geode™ GXm Processor SUB Integer Subtract R e g i s t e r t o R e g i s t e r 2 [ 1 0 d w ] [ 1 1 r e g r / m ] x -- - xxxxx 1 1 b h Register to Memory 2 [100w] [mod reg r/m] 1 1 Memory to Register 2 [101w] [mod reg r/m] 1 1 Immediate to Register/Memory 8 [00sw] [mod 101 r/m] ### 1 1 Immediate to Accumulator (short form) 2 [110w] ### 1 1 TEST Test Bits Register/Memory and Register 8 [010w] [mod reg r/m] 0 - - - x x u x 0 1 1 b h Immediate Data and Register/Memory F [011w] [mod 000 r/m] ### 1 1 Immediate Data and Accumulator A [100w] ### 1 1 VERR Verify Read Access T o R e g i s t e r / M e m o r y 0 F 0 0 [ m o d 1 0 0 r / m ] -----x --- 8 a g , h , j , p VERW Verify Write Access T o R e g i s t e r / M e m o r y 0 F 0 0 [ m o d 1 0 1 r / m ] -----x --- 8 a g , h , j , p XADD Exchange and Add R e g i s t e r 1 , R e g i s t e r 2 0 F C [ 0 0 0 w ] [ 1 1 r e g 2 r e g 1 ] x- - - xxxxx 2 2 Memory, Register 0F C[000w] [mod reg r/m] 2 2 XCHG Exchange Register with Accumulator 9[0 reg] 2 2 XOR Boolean Exclusive OR Register to Register 3 [00dw] [11 reg r/m] 0 - - - x x u x 0 1 1 b h Register to Memory 3 [000w] [mod reg r/m] 1 1 Memory to Register 3 [001w] [mod reg r/m] 1 1 Immediate to Register/Memory 8 [00sw] [mod 110 r/m] ### 1 1 Immediate to Accumulator (short form) 3 [010w] ### 1 1 Table 9-27. Processor Core Instruction Set Summary (Continued) Instruction Opcode Flags Real Mode Prot’d Mode Real Mode Prot’d Mode ODI T SZ APC FFFFFFFFF Clock Count (Reg/Cache Hit) Notes

Revision 3.1 223 www.national.com Instruction Set (Continued) Geode™ GXm Processor Instruction Notes for Instruction Set Summary Notes a through c apply to Real Address Mode only: a. This is a Protected Mode instruction. Attempted execution in Real Mode will result in exception 6 (invalid opcode). b. Exception 13 fault (general protection) will occur in Real Mode if an operand reference is made that partially or fully extends beyond the maximum CS, DS, ES, FS, or GS seg- ment limit (FFFFH). Exception 12 fault (stack segment limit vio- lation or not present) will occur in Real Mode if an operand reference is made that partially or fully extends beyond the maximum SS limit. c. This instruction may be executed in Real Mode. In Real Mode, its purpose is primarily to initialize the CPU for Pro- tected Mode. d. - Notes e through g apply to Real Address Mode and Protected Virtual Address Mode: e. An exception may occur, depending on the value of the oper- and. f. LOCK# is automatically asserted, regardless of the presence or absence of the LOCK prefix. g. LOCK# is asserted during descriptor table accesses. Notes h through r apply to Protected Virtual Address Mode only: h. Exception 13 fault will occur if the memory operand in CS, DS, ES, FS, or GS cannot be used due to either a segment limit violation or an access rights violation. If a stack limit is violated, an exception 12 occurs. i. For segment load operations, the CPL, RPL, and DPL must agree with the privilege rules to avoid an exception 13 fault. The segment’s descriptor must indicate “present” or exception 11 (CS, DS, ES, FS, GS not present). If the SS register is loaded and a stack segment not present is detected, an exception 12 occurs. j. All segment descriptor accesses in the GDT or LDT made by this instruction will automatically assert LOCK# to maintain descriptor integrity in multiprocessor systems. k. JMP, CALL, INT, RET, and IRET instructions referring to another code segment will cause an exception 13, if an appli- cable privilege rule is violated. l. An exception 13 fault occurs if CPL is greater than 0 (0 is the most privileged level). m. An exception 13 fault occurs if CPL is greater than IOPL. n. The IF bit of the flag register is not updated if CPL is greater than IOPL. The IOPL and VM fields of the flag register are updated only if CPL = 0. o. The PE bit of the MSW (CR0) cannot be reset by this instruc- tion. Use MOV into CRO if desiring to reset the PE bit. p. Any violation of privilege rules as apply to the selector oper- and does not cause a Protection exception, rather, the zero flag is cleared. q. If the coprocessor’s memory operand violates a segment limit or segment access rights, an exception 13 fault will occur before the ESC instruction is executed. An exception 12 fault will occur if the stack limit is violated by the operand’s starting address. r. The destination of a JMP , CALL, INT, RET, or IRET must be in the defined limit of a code segment or an exception 13 fault will occur. Note s applies to National Semiconductor-specific SMM in- structions: s. All memory accesses to SMM space are non-cacheable. An invalid opcode exception 6 occurs unless SMI is enabled and SMAR size > 0, and CPL = 0 and [SMAC is set or if in an SMI handler]. Note t applies to cache invalidation instruction with the cache operating in write-back mode: t. The total clock count is the clock count shown plus the num- ber of clocks required to write all “modified” cache lines to external memory.

www.national.com 224 Revision 3.1 Instruction Set (Continued) Geode™ GXm Processor

9.4 FPU INSTRUCTION SET

The processor core is functionally divided into the FPU, and the integer unit. The FPU processes floating point instructions only and does so in parallel with the integer unit. For example, when the integer unit detects a floating point instruction without memory operands, after two clock cycles the instruction passes to the FPU for execution. The integer unit continues to execute instructions while the FPU executes the floating point instruction. If another FPU instruction is encountered, the second FPU instruc- tion is placed in the FPU queue. Up to four FPU instruc- tions can be queued. In the event of an FPU exception, while other FPU instructions are queued, the state of the CPU is saved to ensure recovery. The instruction set for the FPU is summarized in Table 9- 29 on page 225. The table uses abbreviations that are described in Table 9-28. Table 9-28. FPU Instruction Set Table Legend Abbr. Description n Stack register number TOS Top of stack register pointed to by SSS in the status register. ST(1) FPU register next to TOS ST(n) A specific FPU register, relative to TOS M.WI 16-bit integer operand from memory M.SI 32-bit integer operand from memory M.LI 64-bit integer operand from memory M.SR 32-bit real operand from memory M.DR 64-bit real operand from memory M.XR 80-bit real operand from memory M.BCD 18-digit BCD integer operand from memory CC FPU condition code Env Regs Status, Mode Control and Tag Registers, Instruction Pointer and Operand Pointer

Revision 3.1 225 www.national.com Instruction Set (Continued) Geode™ GXm Processor Table 9-29. FPU Instruction Set Summary FPU Instruction Opcode Operation Clock Count Notes F2XM1 Function Evaluation 2x-1 D9 F0 TOS <--- 2 TOS -1 92 - 108 2 FABS Floating Absolute Value D9 E1 TOS <--- | TOS | 2 2 FADD Floating Point Add T o po fS t a c k D C[ 1 1 0 00n ] S T ( n )< - - -S T ( n )+T O S 4-9 80-bit Register D8 [1100 0 n] TOS <--- TOS + ST(n) 4 - 9 64-bit Real DC [mod 000 r/m] TOS <--- TOS + M.DR 4 - 9 32-bit Real D8 [mod 000 r/m] TOS <--- TOS + M.SR 4 - 9 FADDP Floating Point Add, Pop DE [1100 0 n] ST(n) <--- ST(n) + TOS; then pop TOS FIADD Floating Point Integer Add 32-bit integer DA [mod 000 r/m] TOS <--- TOS + M.SI 8 - 14 16-bit integer DE [mod 000 r/m] TOS <--- TOS + M.WI 8 - 14 FCHS Floating Change Sign D9 E0 TOS <--- - TOS 2 FCLEX Clear Exceptions (9B) DB E2 Wait then Clear Exceptions 5 FNCLEX Clear Exceptions DB E2 Clear Exceptions 3 FCMOVB Floating Point Conditional Move if Below DA [1100 0 n] If (CF=1) ST(0) <--- ST(n) 4 FCMOVE Floating Point Conditional Move if Equal DA [1100 1 n] If (ZF=1) ST(0) <--- ST(n) 4 FCMOVBE Floating Point Conditional Move if Below or Equal DA [1101 0 n] If (CF=1 or ZF=1) ST(0) <--- ST(n) 4 FCMOVU Floating Point Conditional Move if Unordered DA [1101 1 n] If (PF=1) ST(0) <--- ST(n) 4 FCMOVNB Floating Point Conditional Move if Not Below DB [1100 0 n] If (CF=0) ST(0) <--- ST(n) 4 FCMOVNE Floating Point Conditional Move if Not Equal DB [1100 1 n] If (ZF=0) ST(0) <--- ST(n) 4 FCMOVNBE Floating Point Conditional Move if Not Below or Equal DB [1101 0 n] If (CF=0 and ZF=0) ST(0) <--- ST(n) 4 FCMOVNU Floating Point Conditional Move if Not Unordered DB [1101 1 n] If (DF=0) ST(0) <--- ST(n) 4 FCOM Floating Point Compare 80-bit Register D8 [1101 0 n] CC set by TOS - ST(n) 4 64-bit Real DC [mod 010 r/m] CC set by TOS - M.DR 4 32-bit Real D8 [mod 010 r/m] CC set by TOS - M.SR 4 FCOMP Floating Point Compare, Pop 80-bit Register D8 [1101 1 n] CC set by TOS - ST(n); then pop TOS 4 64-bit Real DC [mod 011 r/m] CC set by TOS - M.DR; then pop TOS 4 32-bit Real D8 [mod 011 r/m] CC set by TOS - M.SR; then pop TOS 4 FCOMPP Floating Point Compare, Pop Two Stack Elements DE D9 CC set by TOS - ST(1); then pop TOS and ST(1) FCOMI Floating Point Compare Real and Set EFLAGS 80-bit Register DB [1111 0 n] EFLAG set by TOS - ST(n) 4 FCOMIP Floating Point Compare Real and Set EFLAGS, Pop 80-bit Register DF [1111 0 n] EFLAG set by TOS - ST(n); then pop TOS 4 FUCOMI Floating Point Unordered Compare Real and Set EFLAGS 80-bit Integer DB [1110 1 n] EFLAG set by TOS - ST(n) 9 - 10 FUCOMIP Floating Point Unordered Compare Real and Set EFLAGS, Pop 80-bit Integer DF [1110 1 n] EFLAG set by TOS - ST(n); then pop TOS 9 - 10 FICOM Floating Point Integer Compare 32-bit integer DA [mod 010 r/m] CC set by TOS - M.WI 9 - 10 16-bit integer DE [mod 010 r/m] CC set by TOS - M.SI 9 - 10 FICOMP Floating Point Integer Compare, Pop 32-bit integer DA [mod 011 r/m] CC set by TOS - M.WI; then pop TOS 9 - 10 16-bit integer DE [mod 011 r/m CC set by TOS - M.SI; then pop TOS 9 - 10

www.national.com 226 Revision 3.1 Instruction Set (Continued) Geode™ GXm Processor FCOS Function Evaluation: Cos(x) D9 FF TOS <--- COS(TOS) 92 - 141 1 FDECSTP Decrement Stack pointer D9 F6 Decrement top of stack pointer 4 FDIV Floating Point Divide Top of Stack DC [1111 1 n] ST(n) <--- ST(n) / TOS 24 - 34 80-bit Register D8 [1111 0 n] TOS <--- TOS / ST(n) 24 - 34 64-bit Real DC [mod 110 r/m] TOS <--- TOS / M.DR 24 - 34 32-bit Real D8 [mod 110 r/m] TOS <--- TOS / M.SR 24 - 34 FDIVP Floating Point Divide, Pop DE [1111 1 n] ST(n) <--- ST(n) / TOS; then pop TOS 24 - 34 FDIVR Floating Point Divide Reversed Top of Stack DC [1111 0 n] TOS <--- ST(n) / TOS 24 - 34 80-bit Register D8 [1111 1 n] ST(n) <--- TOS / ST(n) 24 - 34 64-bit Real DC [mod 111 r/m] TOS <--- M.DR / TOS 24 - 34 32-bit Real D8 [mod 111 r/m] TOS <--- M.SR / TOS 24 - 34 FDIVRP Floating Point Divide Reversed, PopDE [1111 0 n] ST(n) <--- TOS / ST(n); then pop TOS 24 - 34 FIDIVFloating Point Integer Divide 32-bit Integer DA [mod 110 r/m] TOS <--- TOS / M.SI 34 - 38 16-bit Integer DE [mod 110 r/m] TOS <--- TOS / M.WI 34 - 38 FIDIVR Floating Point Integer Divide Reversed 32-bit Integer DA [mod 111 r/m] TOS <--- M.SI / TOS 34 - 38 16-bit Integer DE [mod 111 r/m] TOS <--- M.WI / TOS 34 - 38 FFREE Free Floating Point Register DD [1100 0 n] TAG(n) <--- Empty 4 FINCSTP Increment Stack Pointer D9 F7 Increment top-of-stack pointer 2 FINITInitialize FPU (9B)DB E3 Wait, then initialize 8 FNINIT Initialize FPU DB E3 Initialize 6 FLD Load Data to FPU Register Top of Stack D9 [1100 0 n] Push ST(n) onto stack 2 64-bit Real DD [mod 000 r/m] Push M.DR onto stack 2 32-bit Real D9 [mod 000 r/m] Push M.SR onto stack 2 FBLD Load Packed BCD Data to FPU Register DF [mod 100 r/m] Push M.BCD onto stack 41 - 45 FILD Load Integer Data to FPU Register 64-bit Integer DF [mod 101 r/m] Push M.LI onto stack 4 - 8 32-bit Integer DB [mod 000 r/m] Push M.SI onto stack 4 - 6 16-bit Integer DF [mod 000 r/m] Push M.WI onto stack 3 - 6 FLD1 Load Floating Const.= 1.0 D9 E8 Push 1.0 onto stack 4 FLDCW Load FPU Mode Control Register D9 [mod 101 r/m] Ctl Word <--- Memory 4 FLDENV Load FPU Environment D9 [mod 100 r/m] Env Regs <--- Memory 30 FLDL2E Load Floating Const.= Log 2(e) D9 EA Push Log 2(e) onto stack 4 FLDL2T Load Floating Const.= Log2(10) D9 E9 Push Log 2(10) onto stack 4 FLDLG2 Load Floating Const.= Log10(2) D9 EC Push Log 10(2) onto stack 4 FLDLN2 Load Floating Const.= Ln(2) D9 ED Push Log e(2) onto stack 4 FLDPI Load Floating Const.=π D9 EB Push π onto stack 4 FLDZ Load Floating Const.= 0.0 D9 EE Push 0.0 onto stack 4 FMUL Floating Point Multiply T o po fS t a c k D C[ 1 1 0 01n ] S T ( n )< - - -S T ( n ) × TOS 4 - 9 80-bit Register D8 [1100 1 n] TOS <--- TOS × ST(n) 4 - 9 64-bit Real DC [mod 001 r/m] TOS <--- TOS ×M.DR 4 - 8 32-bit Real D8 [mod 001 r/m] TOS <--- TOS × M.SR 4 - 6 FMULP Floating Point Multiply & Pop DE [1100 1 n] ST(n) <--- ST(n) × TOS; then pop TOS 4 - 9 FIMUL Floating Point Integer Multiply 32-bit Integer DA [mod 001 r/m] TOS <--- TOS × M.SI 9 - 11 16-bit Integer DE [mod 001 r/m] TOS <--- TOS × M.WI 8 - 10 Table 9-29. FPU Instruction Set Summary (Continued) FPU Instruction Opcode Operation Clock Count Notes

Revision 3.1 227 www.national.com Instruction Set (Continued) Geode™ GXm Processor FNOP No Operation D9 D0 No Operation 2 FPATAN Function Eval: Tan-1(y/x) D9 F3 ST(1) <--- ATAN[ST(1) / TOS]; then pop TOS 97 - 161 3 FPREM Floating Point Remainder D9 F8 TOS <--- Rem[TOS / ST(1)] 82 - 91 FPREM1 Floating Point Remainder IEEE D9 F5 TOS <--- Rem[TOS / ST(1)] 82 - 91 FPTAN Function Eval: Tan(x) D9 F2 TOS <--- TAN(TOS); then push 1.0 onto stack 117 - 129 1 FRNDINT Round to Integer D9 FC TOS <--- Round(TOS) 10 - 20 FRSTOR Load FPU Environment and Register DD [mod 100 r/m] Restore state 56 - 72 FSAVE Save FPU Environment and Register (9B)DD [mod 110 r/m] Wait, then save state 57 - 67 FNSAVE Save FPU Environment and Register DD [mod 110 r/m] Save state 55 - 65 FSCALE Floating Multiply by 2n D9 FD TOS <--- TOS × 2(ST(1)) 7-1 4 FSIN Function Evaluation: Sin(x) D9 FE TOS <--- SIN(TOS) 76 - 140 1 FSINCOS Function Eval.: Sin(x)& Cos(x) D9 FB temp <--- TOS; TOS <--- SIN(temp); then push COS(temp) onto stack 145 - 161 1 FSQRT Floating Point Square Root D9 FA TOS <--- Square Root of TOS 59 - 60 FST Store FPU Register T o po fS t a c k D D[ 1 1 0 10n ] S T ( n )< - - -T O S 2 64-bit Real DD [mod 010 r/m] M.DR <--- TOS 2 32-bit Real D9 [mod 010 r/m] M.SR <--- TOS 2 FSTP Store FPU Register, Pop Top of Stack DB [1101 1 n] ST(n) <--- TOS; then pop TOS 2 80-bit Real DB [mod 111 r/m] M.XR <--- TOS; then pop TOS 2 64-bit Real DD [mod 011 r/m] M.DR <--- TOS; then pop TOS 2 32-bit Real D9 [mod 011 r/m] M.SR <--- TOS; then pop TOS 2 FBSTP Store BCD Data, Pop DF [mod 110 r/m] M.BCD <--- TOS; then pop TOS 57 - 63 FIST Store Integer FPU Register 32-bit Integer DB [mod 010 r/m] M.SI <--- TOS 8 - 13 16-bit Integer DF [mod 010 r/m] M.WI <--- TOS 7 - 10 FISTP Store Integer FPU Register, Pop 64-bit Integer DF [mod 111 r/m] M.LI <--- TOS; then pop TOS 10 - 13 32-bit Integer DB [mod 011 r/m] M.SI <--- TOS; then pop TOS 8 - 13 16-bit Integer DF [mod 011 r/m] M.WI <--- TOS; then pop TOS 7 - 10 FSTCW Store FPU Mode Control Register (9B)D9 [mod 111 r/m] Wait Memory <--- Control Mode Register 5 FNSTCW Store FPU Mode Control Register D9 [mod 111 r/m] Memory <--- Control Mode Register 3 FSTENV Store FPU Environment (9B)D9 [mod 110 r/m] Wait Memory <--- Env. Registers 14 - 24 FNSTENV Store FPU Environment D9 [mod 110 r/m] Memory <--- Env. Registers 12 - 22 FSTSW Store FPU Status Register (9B)DD [mod 111 r/m] Wait Memory <--- Status Register 6 FNSTSW Store FPU Status Register DD [mod 111 r/m] Memory <--- Status Register 4 FSTSW AX Store FPU Status Register to AX (9B)DF E0 Wait AX <--- Status Register 4 FNSTSW AX Store FPU Status Register to AXDF E0 AX <--- Status Register 2 FSUB Floating Point Subtract T o po fS t a c k D C[ 1 1 1 01n ] S T ( n )< - - -S T ( n )-T O S 4-9 80-bit Register D8 [1110 0 n] TOS <--- TOS - ST(n 4 - 9 64-bit Real DC [mod 100 r/m] TOS <--- TOS - M.DR 4 - 9 32-bit Real D8 [mod 100 r/m] TOS <--- TOS - M.SR 4 - 9 FSUBP Floating Point Subtract, Pop DE [1110 1 n] ST(n) <--- ST(n) - TOS; then pop TOS 4 - 9 FSUBR Floating Point Subtract Reverse T o po fS t a c k D C[ 1 1 1 00n ] T O S< - - -S T ( n )-T O S 4-9 80-bit Register D8 [1110 1 n] ST(n) <--- TOS - ST(n) 4 - 9 64-bit Real DC [mod 101 r/m] TOS <--- M.DR - TOS 4 - 9 32-bit Real D8 [mod 101 r/m] TOS <--- M.SR - TOS 4 - 9 FSUBRP Floating Point Subtract Reverse, PopD E[ 1 1 1 00n ] S T ( n )< - - -T O S-S T ( n ) ;t h e np o pT O S 4-9 Table 9-29. FPU Instruction Set Summary (Continued) FPU Instruction Opcode Operation Clock Count Notes

www.national.com 228 Revision 3.1 Instruction Set (Continued) Geode™ GXm Processor FPU Instruction Summary Notes All references to TOS and ST(n) refer to stack layout prior to exe- cution. Values popped off the stack are discarded. A pop from the stack increments the top of stack pointer. A push to the stack decrements the top of stack pointer. Notes: 1. For FCOS, FSIN, FSINCOS and FPTAN, time shown is for absolute value of TOS < 3p/4. Add 90 clock counts for argu- ment reduction if outside this range. For FCOS, clock count is 141 if TOS <π/4 and clock count is 92 ifπ/4 < TOS >π/2. For FSIN, clock count is 81 to 82 if absolute value of TOS < π/4. 2. For F2XM1, clock count is 92 if absolute value of TOS < 0.5. 3. For FPATAN, clock count is 97 if ST(1)/TOS <π/32. 4. For FYL2XP1, clock count is 170 if TOS is out of range and regular FYL2X is called. 5. The following opcodes are reserved: D9D7, D9E2, D9E7, DDFC, DED8, DEDA, DEDC, DEDD, DEDE, DFFC. If a reserved opcode is executed, and unpredictable results may occur (exceptions are not generated). FISUB Floating Point Integer Subtract 32-bit Integer DA [mod 100 r/m] TOS <--- TOS - M.SI 14 - 29 16-bit Integer DE [mod 100 r/m] TOS <--- TOS - M.WI 14 - 27 FISUBR Floating Point Integer Subtract Reverse 32-bit Integer Reversed DA [mod 101 r/m] TOS <--- M.SI - TOS 14 - 29 16-bit Integer Reversed DE [mod 101 r/m] TOS <--- M.WI - TOS 14 - 27 FTST Test Top of Stack D9 E4 CC set by TOS - 0.0 4 FUCOM Unordered Compare DD [1110 0 n] CC set by TOS - ST(n) 4 FUCOMP Unordered Compare, Pop DD [1110 1 n] CC set by TOS - ST(n); then pop TOS 4 FUCOMPP Unordered Compare, Pop two elements DA E9 CC set by TOS - ST(I); then pop TOS and ST(1) FWAIT Wait 9B Wait for FPU not busy 2 FXAM Report Class of Operand D9 E5 CC <--- Class of TOS 4 FXCH Exchange Register with TOS D9 [1100 1 n] TOS <--> ST(n) Exchange 3 FXTRACT Extract Exponent D9 F4 temp <--- TOS; TOS <--- exponent (temp); then push significant (temp) onto stack 11 - 16 FLY2X Function Eval. y× Log2(x) D9 F1 ST(1) <--- ST(1) ×Log 2(TOS); then pop TOS 145 - 154 FLY2XP1 Function Eval. y× Log2(x+1) D9 F9 ST(1) <--- ST(1) ×Log2(1+TOS); then pop TOS 131 - 133 4 Table 9-29. FPU Instruction Set Summary (Continued) FPU Instruction Opcode Operation Clock Count Notes

Revision 3.1 229 www.national.com Instruction Set (Continued) Geode™ GXm Processor

9.5 MMX INSTRUCTION SET

The CPU is functionally divided into the FPU unit, and the integer unit. The FPU has been extended to process both MMX instructions and floating point instructions in parallel with the integer unit. Revision 3.1 For example, when the integer unit detects a MMX instruction, the instruction passes to the FPU unit for exe- cution. The integer unit continues to execute instructions while the FPU unit executes the MMX instruction. If another MMX instruction is encountered, the second MMX instruction is placed in the MMX queue. Up to four MMX instructions can be queued. T h eM M Xi n s t r u c t i o ns e ti ss u m m a r i z e di nT a b l e9 - 3 1o n page 230. The abbreviations used in the table are listed in Table 9-30. Table 9-30. MMX Instruction Set Table Legend Abbreviation Description <---- Result written [11 mm reg] Binary or binary groups of digits mm One of eight 64-bit MMX registers reg A general purpose register <--sat-- If required, the resultant data is saturated to remain in the associated data range <--move-- Source data is moved to result location [byte] Eight 8-bit bytes are processed in parallel [word] Four 16-bit word are processed in parallel [dword] Two 32-bit double words are processed in parallel [qword] One 64-bit quad word is processed [sign xxx] The byte, word, double word or quad word most significant bit is a sign bit mm1, mm2 MMX Register 1, MMX Register 2 mod r/m Mod and r/m byte encoding (page 6-6 of this manual) pack Source data is truncated or saturated to next smaller data size, then concatenated. packdw Pack two double words from source and two double words from destination into four words in destination register. packwb Pack four words from source and four words from destination into eight bytes in destination register.

www.national.com 230 Revision 3.1 Instruction Set (Continued) Geode™ GXm Processor Table 9-31. MMX Instruction Set Summary MMX Instructions Opcode Operation and Clock Count (Latency/Throughput) EMMS Empty MMX State 0F77 Tag Word <--- FFFFh (empties the floating point tag word) 1/1 MOVD Move Doubleword Register to MMX Register 0F6E [11 mm reg] MMX reg [qword] <--move, zero extend-- reg [dword] 1/1 MMX Register to Register 0F7E [11 mm reg] reg [qword] <--move-- MMX reg [low dword] 5/1 Memory to MMX Register 0F6E [mod mm r/m] MMX regr[qword] <--move, zero extend-- memory[dword] 1/1 MMX Register to Memory 0F7E [mod mm r/m] Memory [dword] <--move-- MMX reg [low dword] 1/1 MOVQ Move Quardword MMX Register 2 to MMX Register 1 0F6F [11 mm1 mm2] MMX reg 1 [qword] <--move-- MMX reg 2 [qword] 1/1 MMX Register 1 to MMX Register 2 0F7F [11 mm1 mm2] MMX reg 2 [qword] <--move-- MMX reg 1 [qword] 1/1 Memory to MMX Register 0F6F [mod mm r/m] MMX reg [qword] <--move-- memory[qword] 1/1 MMX Register to Memory 0F7F [mod mm r/m] Memory [qword] <--move-- MMX reg [qword] 1/1 PACKSSDW Pack Dword with Signed Saturation MMX Register 2 to MMX Register 1 0F6B [11 mm1 mm2] MMX reg 1 [qword] <--packdw, signed sat-- MMX reg 2, MMX reg 1 1/1 Memory to MMX Register 0F6B [mod mm r/m] MMX reg [qword] <--packdw, signed sat-- memory, MMX reg 1/1 PACKSSWB Pack Word with Signed Saturation MMX Register 2 to MMX Register 1 0F63 [11 mm1 mm2] MMX reg 1 [qword] <--packwb, signed sat-- MMX reg 2, MMX reg 1 1/1 Memory to MMX Register 0F63 [mod mm r/m] MMX reg [qword] <--packwb, signed sat-- memory, MMX reg 1/1 PACKUSWB Pack Word with Unsigned Saturation MMX Register 2 to MMX Register 1 0F67 [11 mm1 mm2] MMX reg 1 [qword] <--packwb, unsigned sat-- MMX reg 2, MMX reg 1 1/1 Memory to MMX Register 0F67 [mod mm r/m] MMX reg [qword] <--packwb, unsigned sat-- memory, MMX reg 1/1 PADDB Packed Add Byte with Wrap-Around MMX Register 2 to MMX Register 1 0FFC [11 mm1 mm2] MMX reg 1 [byte] <---- MMX reg 1 [byte] + MMX reg 2 [byte] 1/1 Memory to MMX Register 0FFC [mod mm r/m] MMX reg[byte] <---- memory [byte] + MMX reg [byte] 1/1 PADDD Packed Add Dword with Wrap-Around MMX Register 2 to MMX Register 1 0FFE [11 mm1 mm2] MMX reg 1 [sign dword] <---- MMX reg 1 [sign dword] + MMX reg 2 [sign dword] 1/1 Memory to MMX Register 0FFE [mod mm r/m] MMX reg [sign dword] <---- memory [sign dword] + MMX reg [sign dword] 1/1 PADDSB Packed Add Signed Byte with Saturation MMX Register 2 to MMX Register 1 0FEC [11 mm1 mm2] MMX reg 1 [sign byte] <--sat-- MMX reg 1 [sign byte] + MMX reg 2 [sign byte] 1/1 Memory to Register 0FEC [mod mm r/m] MMX reg [sign byte] <--sat-- memory [sign byte] + MMX reg [sign byte] 1/1 PADDSW Packed Add Signed Word with Saturation MMX Register 2 to MMX Register 1 0FED [11 mm1 mm2] MMX reg 1 [sign word] <--sat-- MMX reg 1 [sign word] + MMX reg 2 [sign word] 1/1 Memory to Register 0FED [mod mm r/m] MMX reg [sign word] <--sat-- memory [sign word] + MMX reg [sign word] 1/1 PADDUSB Add Unsigned Byte with Saturation MMX Register 2 to MMX Register 1 0FDC [11 mm1 mm2] MMX reg 1 [byte] <--sat-- MMX reg 1 [byte] + MMX reg 2 [byte] 1/1 Memory to Register 0FDC [mod mm r/m] MMX reg [byte] <--sat-- memory [byte] + MMX reg [byte] 1/1 PADDUSW Add Unsigned Word with Saturation MMX Register 2 to MMX Register 1 0FDD [11 mm1 mm2] MMX reg 1 [word] <--sat-- MMX reg 1 [word] + MMX reg 2 [word] 1/1 Memory to Register 0FDD [mod mm r/m] MMX reg [word] <--sat-- memory [word] + MMX reg [word] 1/1 PADDW Packed Add Word with Wrap-Around MMX Register 2 to MMX Register 1 0FFD [11 mm1 mm2] MMX reg 1 [word] <---- MMX reg 1 [word] + MMX reg 2 [word] 1/1 Memory to MMX Register 0FFD [mod mm r/m] MMX reg [word] <---- memory [word] + MMX reg [word] 1/1 PAND Bitwise Logical AND MMX Register 2 to MMX Register 1 0FDB [11 mm1 mm2] MMX reg 1 [qword] <--logic AND-- MMX reg 1 [qword], MMX reg 2 [qword] 1/1 Memory to MMX Register 0FDB [mod mm r/m] MMX reg [qword] <--logic AND-- memory [qword], MMX reg [qword] PANDN Bitwise Logical AND NOT MMX Register 2 to MMX Register 1 0FDF [11 mm1 mm2] MMX reg 1 [qword] <--logic AND -- NOT MMX reg 1 [qword], MMX reg 2 [qword] 1/1 Memory to MMX Register 0FDF [mod mm r/m] MMX reg [qword] <--logic AND-- NOT MMX reg [qword], Memory [qword] 1/1 PCMPEQB Packed Byte Compare for Equality MMX Register 2 with MMX Register 1 0F74 [11 mm1 mm2] MMX reg 1 [byte] <--FFh-- if MMX reg 1 [byte] = MMX reg 2 [byte] MMX reg 1 [byte]<--00h-- if MMX reg 1 [byte] NOT = MMX reg 2 [byte] Memory with MMX Register 0F74 [mod mm r/m] MMX reg [byte] <--FFh-- if memory[byte] = MMX reg [byte] MMX reg [byte] <--00h-- if memory[byte] NOT = MMX reg [byte]

Revision 3.1 231 www.national.com Instruction Set (Continued) Geode™ GXm Processor PCMPEQD Packed Dword Compare for Equality MMX Register 2 with MMX Register 1 0F76 [11 mm1 mm2] MMX reg 1 [dword] <--FFFF FFFFh-- if MMX reg 1 [dword] = MMX reg 2 [dword] MMX reg 1 [dword]<--0000 0000h--if MMX reg 1[dword] NOT = MMX reg 2 [dword] Memory with MMX Register 0F76 [mod mm r/m] MMX reg [dword] <--FFFF FFFFh-- if memory[dword] = MMX reg [dword] MMX reg [dword] <--0000 0000h-- if memory[dword] NOT = MMX reg [dword] PCMPEQW Packed Word Compare for Equality MMX Register 2 with MMX Register 1 0F75 [11 mm1 mm2] MMX reg 1 [word] <--FFFFh-- if MMX reg 1 [word] = MMX reg 2 [word] MMX reg 1 [word]<--0000h-- if MMX reg 1 [word] NOT = MMX reg 2 [word] Memory with MMX Register 0F75 [mod mm r/m] MMX reg [word] <--FFFFh-- if memory[word] = MMX reg [word] MMX reg [word] <--0000h-- if memory[word] NOT = MMX reg [word] PCMPGTB Pack Compare Greater Than Byte MMX Register 2 to MMX Register 1 0F64 [11 mm1 mm2] MMX reg 1 [byte] <--FFh-- if MMX reg 1 [byte] > MMX reg 2 [byte] MMX reg 1 [byte]<--00h-- if MMX reg 1 [byte] NOT > MMX reg 2 [byte] Memory with MMX Register 0F64 [mod mm r/m] MMX reg [byte] <--FFh-- if memory[byte] > MMX reg [byte] MMX reg [byte] <--00h-- if memory[byte] NOT > MMX reg [byte] PCMPGTD Pack Compare Greater Than Dword MMX Register 2 to MMX Register 1 0F66 [11 mm1 mm2] MMX reg 1 [dword] <--FFFF FFFFh-- if MMX reg 1 [dword] > MMX reg 2 [dword] MMX reg 1 [dword]<--0000 0000h--if MMX reg 1 [dword]NOT > MMX reg 2 [dword] Memory with MMX Register 0F66 [mod mm r/m] MMX reg [dword] <--FFFF FFFFh-- if memory[dword] > MMX reg [dword] MMX reg [dword] <--0000 0000h-- if memory[dword] NOT > MMX reg [dword] PCMPGTW Pack Compare Greater Than Word MMX Register 2 to MMX Register 1 0F65 [11 mm1 mm2] MMX reg 1 [word] <--FFFFh-- if MMX reg 1 [word] > MMX reg 2 [word] MMX reg 1 [word]<--0000h-- if MMX reg 1 [word] NOT > MMX reg 2 [word] Memory with MMX Register 0F65 [mod mm r/m] MMX reg [word] <--FFFFh-- if memory[word] > MMX reg [word] MMX reg [word] <--0000h-- if memory[word] NOT > MMX reg [word] PMADDWD Packed Multiply and Add MMX Register 2 to MMX Register 1 0FF5 [11 mm1 mm2] MMX reg 1 [dword] <--add-- [dword]<---- MMX reg 1 [sign word]*MMX reg 2[sign word] Memory to MMX Register 0FF5 [mod mm r/m] MMX reg 1 [dword] <--add-- [dword] <---- memory [sign word] * Memory [sign word] PMULHW Packed Multiply High MMX Register 2 to MMX Register 1 0FE5 [11 mm1 mm2] MMX reg 1 [word] <--upper bits-- MMX reg 1 [sign word] * MMX reg 2 [sign word] Memory to MMX Register 0FE5 [mod mm r/m] MMX reg 1 [word] <--upper bits-- memory [sign word] * Memory [sign word] 2/1 PMULLW Packed Multiply Low MMX Register 2 to MMX Register 1 0FD5 [11 mm1 mm2] MMX reg 1 [word] <--lower bits-- MMX reg 1 [sign word] * MMX reg 2 [sign word] 2/1 Memory to MMX Register 0FD5 [mod mm r/m] MMX reg 1 [word] <--lower bits-- memory [sign word] * Memory [sign word] 2/1 POR Bitwise OR MMX Register 2 to MMX Register 1 0FEB [11 mm1 mm2] MMX reg 1 [qword] <--logic OR-- MMX reg 1 [qword], MMX reg 2 [qword] 1/1 Memory to MMX Register 0FEB [mod mm r/m] MMX reg [qword] <--logic OR-- MMX reg [qword], memory[qword] 1/1 PSLLD Packed Shift Left Logical Dword MMX Register 1 by MMX Register 2 0FF2 [11 mm1 mm2] MMX reg 1 [dword] <--shift left, shifting in zeroes by MMX reg 2 [dword]-- 1/1 MMX Register by Memory 0FF2 [mod mm r/m] MMX reg [dword] <--shift left, shifting in zeroes by memory[dword]-- 1/1 MMX Register by Immediate 0F72 [11 110 mm] # MMX reg [dword] <--shift left, shifting in zeroes by [im byte]-- 1/1 PSLLQ Packed Shift Left Logical Qword MMX Register 1 by MMX Register 2 0FF3 [11 mm1 mm2] MMX reg 1 [qword] <--shift left, shifting in zeroes by MMX reg 2 [qword]-- 1/1 MMX Register by Memory 0FF3 [mod mm r/m] MMX reg [qword] <--shift left, shifting in zeroes by [qword]-- 1/1 MMX Register by Immediate 0F73 [11 110 mm] # MMX reg [qword] <--shift left, shifting in zeroes by [im byte]-- 1/1 PSLLW Packed Shift Left Logical Word MMX Register 1 by MMX Register 2 0FF1 [11 mm1 mm2] MMX reg 1 [word] <--shift left, shifting in zeroes by MMX reg 2 [word]-- 1/1 MMX Register by Memory 0FF1 [mod mm r/m] MMX reg [word] <--shift left, shifting in zeroes by memory[word]-- 1/1 MMX Register by Immediate 0F71 [11 110mm] # MMX reg [word] <--shift left, shifting in zeroes by [im byte]-- 1/1 Table 9-31. MMX Instruction Set Summary (Continued) MMX Instructions Opcode Operation and Clock Count (Latency/Throughput)

www.national.com 232 Revision 3.1 Instruction Set (Continued) Geode™ GXm Processor PSRAD Packed Shift Right Arithmetic Dword MMX Register 1 by MMX Register 2 0FE2 [11 mm1 mm2] MMX reg 1 [dword] <--arith shift right, shifting in zeroes by MMX reg 2 [dword--] 1/1 MMX Register by Memory 0FE2 [mod mm r/m] MMX reg [dword] <--arith shift right, shifting in zeroes by memory[dword]-- 1/1 MMX Register by Immediate 0F72 [11 100 mm] # MMX reg [dword] <--arith shift right, shifting in zeroes by [im byte]-- 1/1 PSRAW Packed Shift Right Arithmetic Word MMX Register 1 by MMX Register 2 0FE1 [11 mm1 mm2] MMX reg 1 [word] <--arith shift right, shifting in zeroes by MMX reg 2 [word]-- 1/1 MMX Register by Memory 0FE1 [mod mm r/m] MMX reg [word] <--arith shift right, shifting in zeroes by memory[word--] 1/1 MMX Register by Immediate 0F71 [11 100 mm] # MMX reg [word] <--arith shift right, shifting in zeroes by [im byte]-- 1/1 PSRLD Packed Shift Right Logical Dword MMX Register 1 by MMX Register 2 0FD2 [11 mm1 mm2] MMX reg 1 [dword] <--shift right, shifting in zeroes by MMX reg 2 [dword]-- 1/1 MMX Register by Memory 0FD2 [mod mm r/m] MMX reg [dword] <--shift right, shifting in zeroes by memory[dword]-- 1/1 MMX Register by Immediate 0F72 [11 010 mm] # MMX reg [dword] <--shift right, shifting in zeroes by [im byte]-- 1/1 PSRLQ Packed Shift Right Logical Qword MMX Register 1 by MMX Register 2 0FD3 [11 mm1 mm2] MMX reg 1 [qword] <--shift right, shifting in zeroes by MMX reg 2 [qword] 1/1 MMX Register by Memory 0FD3 [mod mm r/m] MMX reg [qword] <--shift right, shifting in zeroes by memory[qword] 1/1 MMX Register by Immediate 0F73 [11 010 mm] # MMX reg [qword] <--shift right, shifting in zeroes by [im byte] 1/1 PSRLW Packed Shift Right Logical Word MMX Register 1 by MMX Register 2 0FD1 [11 mm1 mm2] MMX reg 1 [word] <--shift right, shifting in zeroes by MMX reg 2 [word] 1/1 MMX Register by Memory 0FD1 [mod mm r/m] MMX reg [word] <--shift right, shifting in zeroes by memory[word] 1/1 MMX Register by Immediate 0F71 [11 010 mm] # MMX reg [word] <--shift right, shifting in zeroes by imm[word] 1/1 PSUBB Subtract Byte With Wrap-Around MMX Register 2 to MMX Register 1 0FF8 [11 mm1 mm2] MMX reg 1 [byte] <---- MMX reg 1 [byte] subtract MMX reg 2 [byte] 1/1 Memory to MMX Register 0FF8 [mod mm r/m] MMX reg [byte] <---- MMX reg [byte] subtract memory [byte] 1/1 PSUBD Subtract Dword With Wrap-Around MMX Register 2 to MMX Register 1 0FFA [11 mm1 mm2] MMX reg 1 [dword] <---- MMX reg 1 [dword] subtract MMX reg 2 [dword] 1/1 Memory to MMX Register 0FFA [mod mm r/m] MMX reg [dword] <---- MMX reg [dword] subtract memory [dword] 1/1 PSUBSB Subtract Byte Signed With Saturation MMX Register 2 to MMX Register 1 0FE8 [11 mm1 mm2] MMX reg 1 [sign byte] <--sat-- MMX reg 1 [sign byte] subtract MMX reg 2 [sign byte] Memory to MMX Register 0FE8 [mod mm r/m] MMX reg [sign byte] <--sat-- MMX reg [sign byte] subtract memory [sign byte] 1/1 PSUBSW Subtract Word Signed With Saturation MMX Register 2 to MMX Register 1 0FE9 [11 mm1 mm2] MMX reg 1 [sign word] <--sat-- MMX reg 1 [sign word] subtract MMX reg 2 [sign word] Memory to MMX Register 0FE9 [mod mm r/m] MMX reg [sign word] <--sat-- MMX reg [sign word] subtract memory [sign word] 1/1 PSUBUSB Subtract Byte Unsigned With Saturation MMX Register 2 to MMX Register 1 0FD8 [11 mm1 mm2] MMX reg 1 [byte] <--sat-- MMX reg 1 [byte] subtract MMX reg 2 [byte] 1/1 Memory to MMX Register 0FD8 [11 mm reg] MMX reg [byte] <--sat-- MMX reg [byte] subtract memory [byte] 1/1 PSUBUSW Subtract Word Unsigned With Saturation MMX Register 2 to MMX Register 1 0FD9 [11 mm1 mm2] MMX reg 1 [word] <--sat-- MMX reg 1 [word] subtract MMX reg 2 [word] 1/1 Memory to MMX Register 0FD9 [11 mm reg] MMX reg [word] <--sat-- MMX reg [word] subtract memory [word] 1/1 PSUBW Subtract Word With Wrap-Around MMX Register 2 to MMX Register 1 0FF9 [11 mm1 mm2] MMX reg 1 [word] <---- MMX reg 1 [word] subtract MMX reg 2 [word] 1/1 Memory to MMX Register 0FF9 [mod mm r/m] MMX reg [word] <---- MMX reg [word] subtract memory [word] 1/1 PUNPCKHBW Unpack High Packed Byte, Data to Packed Words MMX Register 2 to MMX Register 1 0F68 [11 mm1 mm2] MMX reg 1 [byte] <--interleave-- MMX reg 1 [up byte], MMX reg 2 [up byte] 1/1 Memory to MMX Register 0F68 [11 mm reg] MMX reg [byte] <--interleave-- memory [up byte], MMX reg [up byte] 1/1 PUNPCKHDQ Unpack High Packed Dword, Data to Qword MMX Register 2 to MMX Register 1 0F6A [11 mm1 mm2] MMX reg 1 [dword] <--interleave-- MMX reg 1 [up dword], MMX reg 2 [up dword] Memory to MMX Register 0F6A [11 mm reg] MMX reg [dword] <--interleave-- memory [up dword], MMX reg [up dword] 1/1 PUNPCKHWD Unpack High Packed Word, Data to Packed Dwords MMX Register 2 to MMX Register 1 0F69 [11 mm1 mm2] MMX reg 1 [word] <--interleave-- MMX reg 1 [up word], MMX reg 2 [up word] 1/1 Memory to MMX Register 0F69 [11 mm reg] MMX reg [word] <--interleave-- memory [up word], MMX reg [up word] 1/1 Table 9-31. MMX Instruction Set Summary (Continued) MMX Instructions Opcode Operation and Clock Count (Latency/Throughput)

Revision 3.1 233 www.national.com Instruction Set (Continued) Geode™ GXm Processor PUNPCKLBW Unpack Low Packed Byte, Data to Packed Words MMX Register 2 to MMX Register 1 0F60 [11 mm1 mm2] MMX reg 1 [word] <--interleave-- MMX reg 1 [low byte], MMX reg 2 [low byte] 1/1 Memory to MMX Register 0F60 [11 mm reg] MMX reg [word] <--interleave-- memory [low byte], MMX reg [low byte] 1/1 PUNPCKLDQ Unpack Low Packed Dword, Data to Qword MMX Register 2 to MMX Register 1 0F62 [11 mm1 mm2] MMX reg 1 [word] <--interleave-- MMX reg 1 [low dword], MMX reg 2 [low dword] Memory to MMX Register 0F62 [11 mm reg] MMX reg [word] <--interleave-- memory [low dword], MMX reg [low dword] 1/1 PUNPCKLWD Unpack Low Packed Word, Data to Packed Dwords MMX Register 2 to MMX Register 1 0F61 [11 mm1 mm2] MMX reg 1 [word] <--interleave-- MMX reg 1 [low word], MMX reg 2 [low word] 1/1 Memory to MMX Register 0F61 [11 mm reg] MMX reg [word] <--interleave-- memory [low word], MMX reg [low word] 1/1 PXOR Bitwise XOR MMX Register 2 to MMX Register 1 0FEF [11 mm1 mm2] MMX reg 1 [qword] <--logic exclusive OR-- MMX reg 1 [qword], MMX reg 2 [qword] Memory to MMX Register 0FEF [11 mm reg] MMX reg [qword] <--logic exclusive OR-- memory[qword], MMX reg [qword] 1/1 Table 9-31. MMX Instruction Set Summary (Continued) MMX Instructions Opcode Operation and Clock Count (Latency/Throughput)

www.national.com 234 Revision 3.1 Instruction Set (Continued) Geode™ GXm Processor

9.6 NATIONAL SEMICONDUCTOR EXTENDED MMX INSTRUCTION SET

National Semiconductor has added instructions to its implementation of the MMX Architecture in order to facili- tate writing of multimedia applications. In general, these instructions allow more efficient implementation of multi- media algorithms, or more precision in computation than can be achieved using the basic set of MMX instructions. All of the added instructions follow the SIMD (single instruction, multiple data) format. Many of the instructions add flexibility to the MMX architecture by allowing both source operands of an instruction to be preserved, while the result goes to a separate register that is derived from the input. Table 9-33 on page 235 summarizes the Extended MMX Instructions. The abbreviations used in the table are listed in Table 9-32. Configuration control register CCR7(0) at location EBh must be set to allow the execution of the Extended MMX instructions. Table 9-32. Extend MMX Instruction Set Table Legend Abbreviation Description <---- Result written [11 mm reg] Binary or binary groups of digits mm One of eight 64-bit MMX registers reg A general purpose register <--sat-- If required, the resultant data is saturated to remain in the associated data range <--move-- Source data is moved to result location [byte] Eight 8-bit bytes are processed in parallel [word] Four 16-bit WORD are processed in paral- lel [dword] Two 32-bit DWORDs are processed in par- allel [qword] One 64-bit QWORD is processed [sign xxx] The BYTE, WORD, DWORD or QWORD most significant bit is a sign bit mm1, mm2 MMX Register 1, MMX Register 2 mod r/m Mod and r/m byte encoding (page 6-6 of this manual) pack Source data is truncated or saturated to next smaller data size, then concatenated. packdw Pack two DWORDs from source and two DWORDs from destination into four WORDs in destination register. packwb Pack four WORDs from source and four WORDs from destination into eight BYTEs in destination register.

Revision 3.1 235 www.national.com Instruction Set (Continued) Geode™ GXm Processor Table 9-33. Extended MMX Instruction Set Summary MMX Instructions Opcode Operation and Clock Count PADDSIW Packed Add Signed Word with Saturation Using Implied Destination MMX Register plus MMX Register to Implied Register 0F51 [11 mm1 mm2] Sum signed packed word from MMX register/memory ---> signed packed word in MMX register, saturate, and write result ---> implied register Memory plus MMX Register to Implied Register 0F51 [mod mm r/m] 1 PAVEB Packed Average Byte MMX Register 2 with MMX Register 1 0F50 [11 mm1 mm2] Average packed byte from the MMX register/memory with packed byte in the MMX register. Result is placed in the MMX register. Memory with MMX Register 0F50 [mod mm r/m] 1 PDISTIB Packed Distance and Accumulate with Implied Register Memory, MMX Register to Implied Register 0F54 [mod mm r/m] Find absolute value of difference between packed byte in memory and packed byte in the MMX register. Using unsigned saturation, accumulate with value in implied destination regis- ter. PMACHRIW Packed Multiply and Accumulate with Rounding Memory to MMX Register 0F5E[mod mm r/m] Multiply the packed word in the MMX register by the packed word in memory. Sum the 32-bit results pairwise. Accumulate the result with the packed signed word in the implied destina- tion register. PMAGW Packed Magnitude MMX Register 2 to MMX Register 1 0F52 [11 mm1 mm2] Set the destination equal ---> the packed word with the largest magnitude, between the packed word in the MMX regis- ter/memory and the MMX register. Memory to MMX Register 0F52 [mod mm r/m] 2 PMULHRIW Packed Multiply High with Rounding, Implied Destination MMX Register 2 to MMX Register1 0F5D [11 mm1 mm2] Packed multiply high with rounding and store bits 30 - 15 in implied register. Memory to MMX Register 0F5D [mod mm r/m] 2 PMULHRW Packed Multiply High with Rounding MMX Register 2 to MMX Register 1 0F59 [11 mm1 mm2] Multiply the signed packed word in the MMX register/memory with the signed packed word in the MMX register. Round with 1/2 bit 15, and store bits 30 - 15 of result in the MMX register. Memory to MMX Register 0F59 [mod mm r/m] 2 PMVGEZB Packed Conditional Move If Greater Than or Equal to Zero Memory to MMX Register 0F5C [mod mm r/m] Conditionally move packed byte from memory ---> packed byte in the MMX register if packed byte in implied MMX register is greater than or equal ---> zero. PMVLZB Packed Conditional Move If Less Than Zero Memory to MMX Register 0F5B [mod mm r/m] Conditionally move packed byte from memory ---> packed byte in the MMX register if packed byte in implied MMX register is less than zero. PMVNZB Packed Conditional Move If Not Zero Memory to MMX Register 0F5A [mod mm r/m] Conditionally move packed byte from memory ---> packed byte in the MMX register if packed byte in implied MMX register is not zero. PMVZB Packed Conditional Move If Zero Memory to MMX Register 0F58 [mod mm r/m] Conditionally move packed byte from memory ---> packed byte in the MMX register if packed byte in implied the MMX register is zero. PSUBSIW Packed Subtracted with Saturation Using Implied Destination MMX Register 2 to MMX Register 1 0F55 [11 mm1 mm2] Subtract signed packed word in the MMX register/memory from signed packed word in the MMX register, saturate, and write result ---> implied register. Memory to MMX Register 0F55 [mod mm r/m] 1

www.national.com 236 Revision 3.1 Geode™ GXm Processor Appendix A Support Documentation A.1 ORDER INFORMATION A.2 DATA BOOK REVISION HISTORY This document is a report of the revision/creation process of the data book for the GXm processor. Any revisions (i.e., additions, deletions, parameter corrections, etc.) are recorded in the tables below. Order Number Part Marking Core Frequency (MHz) Temperature (Degree C) Package 30070-53 GXm-266P 2.9V 70C 266 70 SPGA 30071-53 GXm-266P 2.9V 85C 266 85 SPGA 30170-53 GXm-266B 2.9V 70C 266 70 BGA 30171-53 GXm-266B 2.9V 85C 266 85 BGA 30050-33 GXm-233P 2.9V 70C 233 70 SPGA 30054-33 GXm-233P 2.9V 85C 233 85 SPGA 30150-33 GXm-233B 2.9V 70C 233 70 BGA 30151-33 GXm-233B 2.9V 85C 233 85 BGA 30040-23 GXm-200P 2.9V 70C 200 70 SPGA 30044-23 GXm-200P 2.9V 85C 200 85 SPGA 30140-23 GXm-200B 2.9V 70C 200 70 BGA 30141-23 GXm-200B 2.9V 85C 200 85 BGA 30030-23 GXm-180P 2.9V 70C 180 70 SPGA 30034-23 GXm-180P 2.9V 85C 180 85 SPGA 30130-23 GXm-180B 2.9V 70C 180 70 BGA 30131-23 GXm-180B 2.9V 85C 180 85 BGA Table A-1. Revision History Revision # (PDF Date) Revisions / Comments 0.0 (2/5/98) Creation phase 0.1 (2/25/98) Creation phase continues - added functional description. 0.2 (3/24/98) Creation phase continues - added 233 MHz parameters. 0.3 (4/22/98) Creation phase continues - added 266 MHz numbers. 1.0 (8/13/98) All sections complete - added 300 MHz numbers, added Index. 2.0 (10/29/98) Major change is new values for 352 BGA Mechanical. 3.0 (9/21/99) Converted to National Semiconductor format and updated for addendum revision 3.0. 3.1 (4/6/00) Formatting changes and one change from engineering. See Table A-2 for details. Table A-2. Edits to Current Revision Section Revision

3.0 Processor

 Combined bits 1 and 2 of Configuration Control Register 1 in Table 3-11 on page 52. 7.0 Electricals  All references to Recommended Operating Conditions became Operating Conditions.  T able 7-3 on page 183 - Changed supply voltage from 3.6V to 3.2V .

Revision 3.1 237 www.national.com Index Geode™ GXm Processor A Absolute Maximum Ratings 183 AC Characteristics 186 Accessing 157 Address Spaces 60 Directory Table Entry (DTE) 73 DTE Cache 73 I/O Address Space 60 Memory Address Space 60 Memory Addressing Modes 61 Offset Mechanism 61 Page Frame Offset (PFO) 73 Page Table Entry (PTE) 73 Paging Mechanism 72 Translation Look-Aside Buffer 74 Address Translation 108 Application Register Set 40 B BGA Ball Assignments by Ball Number 15 BGA Ball Assignments by Pin Name 17 BGA Ball Assignments Diagram 14 C Cache BB0_BASE 95 BB0_POINTER 95 BB1_BASE 95 BB1_POINTER 95 GCR register (Index B8h) 95 L1 cache 95 scratchpad memory 95 Write-back caching 95 Cache Controller 95 Cache Disable, bit 30 95 Cache Test Operations 58 call gate 69 Current Privilege Level 69 Descriptor Privilege Level 69 Descriptor Privilege Level in Destination 69 Descriptors Bit Definitions 69 Segment Selector Field 69 CCR1 System Management Memory Access 49 CCR1 Configuration Control Register 1 Index C1h 49 CCR2 Enable Suspend Pins 49 Lock NW Bit 49 Suspend on HALT 49 Write-Through Region 1 49 CCR2 Configuration Control Register 2 Index C2h 49 CCR3 Load/Store Serialize 1 GByte to 2 GBytes 49 Load/Store Serialize 2 GBytes to 3 GBytes 49 Load/Store Serialize 3 GBytes to 4 GBytes 49 Map Enable 49 NMI Enable 49 SMM Register Lock 49 CCR3 Configuration Control Register 3 Index C3h 49 CCR4 Directory Table Entry Cache 50 Enable CPUID Instruction 50 I/O Recovery Time 50 Memory Read Bypassing 50 SMI Nest 50 CCR4 Configuration Control Register 4 Index E8h 50 CCR7 Cyrix Extended MMX Instructions Enable 50 NMI Enabl 50 CCR7 Configuration Control Register 7 Index EBh 50 Clock Mode 24 Configuration Register Map 48 Control Registers 48 Device ID Registers 48 Graphics/VGA Related Registers 48 SMM Base Header Address Registers 48 Configuration Register Summary 47 Conforming Code Segments 69 Control Transfer 87 CPU_READ 99 CPU_READ/WRITE EAX instructions 99 EBX instructions 99 CPU_WRITE 99 CPUID Instruction 208 EAX = 0000 0000h 208 EAX = 0000 0001h 208 EAX = 0000 0002h 209 EAX = 8000 0000h 210 EAX = 8000 0001h 210 EAX = 8000 0002h 211 EAX = 8000 0003h 211 EAX = 8000 0004h 211 EAX = 8000 0005h 211 CPUID Levels 208 CPUID Levels, Extended 210 CR0 Register 45 Alignment Check Mask 46 Cache Disable 46 Emulate Processor Extension 46 Monitor Processor Extension 46 Not Write-Through 46 Numerics Exception 46 Paging Enable Bit 45 Protected Mode Enable 46 Task Switched 46 Write Protect 46 CR2 Register 45 Page Fault Linear Address 45 CR3 Register 45 Page Directory Base Register 45 CR4 Register 45 Time Stamp Counter Instruction 45 D DC Characteristics 185 Descriptor Bit Structure 67 Descriptor Types 87 Descriptors Gate 67 gate 69 Interrupt 67 Task 67 Device Select 28 DEVSEL 28 DIMM 112 DIR0 Device ID 51 DIR0 Device Identification Register 0 Index FEh 51 DIR1 Device Identification Revision 51 DIR1 Device Identification Register 1 Index FFh 51

www.national.com 238 Revision 3.1 Index (Continued) Geode™ GXm Processor Directory Table Entry 73 Display Controller 129 –154 Buffer Organization 135 CODEC hardware 129 Compression Logic 130 Compression Technology 130 CRT Display Modes 134 Cursor Pattern Memory 136 DC Memory Organization 135 DC_CURSOR_COLOR Register (BX_BASE+8360h) 131 Display FIFO 130 Display Modes 131 Display Timing 131 Dither/Frame-Rate Modulation (FRM) 131 Graphics Memory Map 135 Hardware Cursor 131 Memory Management 130 Pixel Arrangement Within a DWORD 135 RAMDAC 129 TFT LCD flat panel 129 TFT Panel Data Bus Formats 133 TFT Panel Display Modes 132 VESA-compatible 131 VGA Display Support 136 Display Controller Block Diagram 129 Display Controller Registers 136 Configuration and Status Registers 139 DC_BORDER_COLOR (8368h-836Bh) 138 DC_BUF_SIZE (8328h-832Bh) 137 DC_CB_ST_OFFSET (8314h-8317h) 137 DC_CFIFO_DIAG (837Ch-837Fh) 138 DC_CURS_ST_OFFSET (8318h-831Bh) 137 DC_CURSOR_COLOR (83680h-8363h) 138 DC_CURSOR_X (8350h-8353h) 137 DC_CURSOR_Y (8358h-835Bh) 137 DC_DFIFO_DIAG (8378h-837Bh) 138 DC_FB_ST_OFFSET (8310h-8313h) 136 DC_FP_H_TIMING (833Ch-833Fh) 137 DC_FP_V_TIMING (834Ch-834Fh) 137 DC_GENERAL_CFG (8304h-8307h) 136 DC_H_TIMING_1 (8330h-8333h) 137 DC_H_TIMING_2 (8334h-8337h) 137 DC_H_TIMING_3 (8338h-833Bh) 137 DC_LINE_DELTA (8324h-8327h) 137 DC_OUTPUT_CFG (830Ch-830Fh) 136 DC_PAL_ADDRESS (8370h-8373h) 138 DC_PAL_DATA (8374h-8377h) 138 DC_SS_LINE_CMP (835Ch-835Fh) 137 DC_TIMING_CFG (8308h-830Bh) 136 DC_UNLOCK (8300h-8303h) 136 DC_V_LINE_CNT (8354h-8357h) 137 DC_V_TIMING_1 (8340h-8343h) 137 DC_V_TIMING_2 (8344h-8247h) 137 DC_V_TIMING_3 (8348h-834Bh) 137 DC_VID_ST_OFFSET (8320h-8323h) 137 Memory Organization Registers 144 Display Driver BB0_RESET 98 BB1_RESET 98 CPU_READ 98 CPU_WRITE 98 Scratchpad 98 Display Driver Instructions 98 DR6 Register 53 Bn 53 BS 53 BT 53 DR7 and DR6 Bit Definitions 53 DR7 Register 53 GD 53 Gn 53 LENn 53 Ln 53 R/Wn 53 DRAM Address Conversion 112 E EBP register 40 EFLAGS Register 43 Alignment Check Enable (AM) 43 Auxiliary Carry Flag 43 Carry Flag 43 CPUID instruction 43 Direction Flag (DF) 43 I/O Privilege Level (IOPL) 43 Identification Bit 43 Interrupt Enable 43 Nested Task (NT) 43 Resume Flag (RF) 43 Sign Flag 43 Trap Enable Flag 43 Virtual 8086 Mode (VM) 43 EFLAGS register, bit 9 83 EGA 165 Electrical Connections 182 NC-Designated Pins 182 Power/Ground Connections 182 Pull-Up/Pull-Down Resisters 182 Unused Input Pins 182 Electrical Specifications 182 Absolute Maximum Ratings 183 AC Characteristics 186 Clock Signals 187 DC Characteristics Table 185 DCLK Timing 192 Graphics Port Timing 191 JTAG AC Specification 193 JTAG Test Timings 194 Output Valid Timing 190 Part Numbers 182 PCI Interface Signals 189 SDRAM Interface Signals 190 Setup and Hold Timings 190 SYSCLK Timing 187 System Signals 188 TCK Timing and Measurement Points 193 Video Interface Signals 191 Video Port Timing 192 Exceptions 74 Abort 75 Fault 75 Trap 74 Extended MMX Instruction Set 234 Extended MMX™ Instruction Set Configuration Control Rregister 234 Legend 234 F Fields - index 207 Fields - mod and r/m 205 Fields - sreg3 206 Fields - ss 207 floating point error 25

Revision 3.1 239 www.national.com Index (Continued) Geode™ GXm Processor FPU Mode Control Register 89 Register Set 89 Status Register 89 Tag Word Register 89 FPU Instruction Set 224 Summary Notes 228 FPU Mode Control Register 90 Denormalized-operand error exception bit 90 Divide-by-zero exception bit 90 Invalid-operation exception bit 90 Overflow error exception bit 90 Precision Control Bits 90 Precision error exception bit 90 Rounding Control Bits 90 FPU Operations 89 FPU Registers 90 FPU Status Register 90 Condition code bit 3 90 Condition code bits 90 Copy of ES bit 90 Denormalized-operand error exception bit 90 Divide-by-zero exception bit 90 Error indicator 90 Invalid operation exception bit 90 Overflow error exception bit 90 Precision error exception bit 90 Stack Full 90 Top-of-Stack 90 Underflow error exception bit 90 FPU Tag Word Register (TAG7:0] 90 frame buffer 94 G Gates 87 General Purpose Registers 40 Global Descriptor Table Register (GDTR) 66 Grant Lines 29 Graphics Memory (GX_BASE+800000h) 94 Graphics Pipeline 120 –128 BitBLT/vector engine 120 Color Patterns 123 Diagonal Error Register (108h-810Bh) 124 Dither Patterns 122 Error Register (8104-8107h) 124 GP_BLT_MODE 120 GP_BLT_MODE (8208h-820Bh) 124 GP_BLT_STATUS (820Ch-820Fh) 125 GP_DST/START_Y/XCOOR (8100h-8103h) 124 GP_DST_XCOOR 121 GP_DST_YCOOR 121 GP_INIT_ERROR 121 GP_PAT_COLOR_0 register 122 GP_PAT_COLOR_1 (GX_BASE+8112h) 122 GP_PAT_COLOR_A (8110h) 124 GP_PAT_COLOR_B (8114h) 124 GP_PAT_DATA (8120h-812Fh) 124 GP_RASTER_MODE (8200h-8203h) 124 GP_RASTER_MODE (GX_BASE+ 8200h) 122 GP_RASTER_MODE Bit Patterns 123 GP_SRC_COLOR (810Ch-810Fh) 124 GP_SRC_COLOR_0 (GX_BASE+810Ch) 123 GP_SRC_YCOOR 121 GP_VECTOR_MODE (8204h-8207h) 124 GP_VGA_BASE (8210h-8213h) 125 GP_VGA_LATCH (8214h-8217h) 125 GP_VGA_READ (8200h-8203h) 124 GP_VGA_WRITE (8140h-8143h) 124 Master/Slave Registers 121 Monochrome Patterns 122 Pattern Generation 121 graphics pipeline 165 ,166 Graphics Pipeline Block Diagram 120 H HALT 25 High Order Interleaving 112 I I/O Address Space 60 Initialization 38 Initialization, CPU 38 Initiator Ready IRDY 27 TRDY 27 Instruction Fields 202 Instruction Set 39 eee Field Encoding 204 Index Field 207 Memory Addressing 205 mod base Field Encoding table 207 mod r/m Field Encoding 205 Opcode 203 prefix bytes 203 reg Field 206 s-i-b Byte 207 s-i-b present 207 sreg2 field 206 sreg3 206 ss Field 207 w Field Operand Size 203 instruction set 201 Instruction Set Format Table 202 Instruction Set Formats 201 Instruction Set Overview 39 Instructions Bit Test Instructions 39 Exchange Instructions 39 One-operand Arithmetic and Logical 39 Two-operand Arithmetic and Logical 39 Instuction Prefix Summary 203 Integrated Functions 91 Integrated Functions Programming Interface 92 Interleaving 112 Internal Bus Interface 100 –102 Internal Bus Interface Unit 640KB to 1MB 100 C-Bus 100 FPU Error Support 100 Graphics 100 IRQ13 100 L1 cache 100 Processor Core 100 Region Control Field Bit Definitions 102 Registers (GX_BASE+8000h) 101 SMI Interrupts 100 VGA Access 100 X-Bus 100 Internal Bus Interface Unit Diagram 91 Internal Bus Interface Unit Registers

www.national.com 240 Revision 3.1 Index (Continued) Geode™ GXm Processor BC_DRAM_TOP (8000h-8003h) 101 BC_XMAP_1 (8004h-8007h) 101 BC_XMAP_2 (8008h-800Bh) 101 BC_XMAP_3 (800Ch-800Fh) 101 Internal Test Signals Ifloat 32 Raw Clock 32 SDRAM Test Outputs 32 Test 33 Test Clock 32 Test Data Input 32 Test Data Output 32 Thermal Diode Negative (TDN) 33 Thermal Diode Positive (TDP) 33 Interrupt Interrupt and Exception Priorities 76 Interrupt Descriptor Table 75 Interrupt Request Level 13 25 Interrupts 74 INTR 74 NMI 74 Real Mode Error Codes 77 Real Mode, Exceptions 77 SMM 74 Vector A 75 INTR 43 ,74,76,83,85,86 invalid opcode 39 IRET instruction 43 L Legacy VGA 165 Local Descriptor Table Register (LDTR) 66 LOCK 28 Lock Prefix 39 Low Order Interleaving 112 M MediaGX™ Virtual VGA 167 Memory Address Space 60 Memory Addressing Paging Mechanism 72 Memory Addressing Modes 61 Memory Controller 103 –119 Auto LOI 112

1 DIMM Bank 113

2 DIMM Banks 113

DRAM Address Conversion 112 DRAM Configuration 105 Graphics Pipeline 103 Memory Array Configuration 104 Memory Cycles 115 Memory Organization 105 Non-Auto LOI

1 DIMM Bank 114

2 DIMM Banks 114

SDRAM Initialization Sequence 107 SDRAM Interface 103 SDRAM Interface Clocking 118 CAS latency 118 SDRAM Read Cycle 115 SDRAM Refresh Cycle 117 SDRAM Write Cycle 116 SHFTSDCLK 119 X-Bus 103 Memory Controller Interface Signals Bank Address Bits 29 Chip Selects 29 Clock Enable 30 Column Address Strobe 29 Data Mask Control Bits 30 Memory Address Bus 29 Row Address Strobe 29 SDRAM Clocks 30 Write Enable 29 Memory Controller Register 108 MC_BANK_CFG (8408h-840Bh) 108 MC_DR_ACC (841Ch-841Fh) 108 MC_DR_ADD (8418h-841Bh) 108 MC_GBASE_ADD (8414h-8417h) 108 MC_MEM_CNTRL1 (8400h-8403h) 108 MC_MEM_CNTRL2 (8404h-8407h) 108 MC_SYNC_TIM1 (840Ch-840Fh) 108 Memory Data Bus 29 MMX Instruction Set 229 Multiplexed Address PCI pins 26 Multiplexed Command Configuration Read 26 Configuration Write 26 Dual Address Cycle 26 Memory Read Line 26 Memory Read Multiple 26 Memory Write and Invalidate 26 Special Cycle 26 Multiplexed Command and Byte Enables Interrupt Acknowledge 26 Multitasking 70 N NMI 49 ,74,75,76,78,83,85,86 notebook computers 174 O Overflow Flag 43 P Package Outlines 198 Package Specifications 195 Page Table Entry 73 palette lookup 166 PCI Arbitration 164 PCI Configuration Registers Access Format 157 Bus 156 Cache Line Size (0Ch) 157 Class Code (09h-0Bh) 157 CONFIG ENABLE 156 CONFIG_DATA 0CFCh-0CFFh 156 Device 156 Device Identification (02h-03h) 157 Device Status (06h-07h) 157 Latency Timer (0Dh) 157 PCI Arbitration Control 1 (43h) 157 PCI Arbitration Control 2 (44h) 157

Revision 3.1 241 www.national.com Index (Continued) Geode™ GXm Processor PCI Command (04h-05h) 157 PCI Control Function 1 (40h) 157 PCI Control Function 2 (41h) 157 Register 156 Revision Identification (08h) 157 Translation Type Bits 1 0 156 Vendor Identification (00h-01h) 157 PCI Configuration Registers 0CF8h-0CFBh 156 PCI Controller 155 CONFIG_ADDRESS 155 Configuration Cycles 155 PCI Arbiter 155 Space Control Registers 156 Special Cycles 155 X-Bus PCI Master 155 X-Bus PCI Slave 155 PCI Cycles 162 PCI Halt Command 164 PCI Interface Signals Frame 27 Initiator Ready 27 Lock Operation 28 Multiplexed Address and Data 26 Multiplexed Command and Byte Enables 26 Parity 26 Parity Error 28 Request Lines 28 Target Ready 27 Target Stop 27 PCI Local Bus Specification 162 PCI Read Transactions 162 PCI Special Cycle Command 164 PCI Write Transactions 163 PCR Performance Control Register Index 20h 50 PERR 28 Pixel Arrangement Within a DWORD 135 Pointer and Index Registers ECX Counter 40 EDI Destination Pointer 40 ESI Source Pointer 40 ESP Register 40 PUSH and POP Instructions 40 Power and Ground Connections and Decoupling 182 Power Management 174 3-Volt Suspend Mode 174 Advanced Power Management (APM) 174 CPU Suspend Command Registers 174 Initiating Suspend with HALT 176 Initiating Suspend with SUSP 175 Processor Serial Bus 179 Responding to a PCI Access During Suspend Mode 177 Serial Packet Transmission 179 Stopping the Input Clock 178 Suspend Mode and Bus Cycles 175 Suspend Modulation 174 Power Management Registers 179 PM_BASE (FFFF FF6Ch) 179 PM_CNTRL_CSTP (8508h-850Bh) 179 PM_CNTRL_TEN (8504h-8507h) 179 PM_MASK (FFFF FF7Ch) 179 PM_SER_PACK (850Ch-850Fh) 179 PM_STAT_SMI (8500h-8503h) 179 Power Planes 36 –37 Power, Ground, No Connect Ground (VSS) 32 Power, Ground, No Connect Signals Ground (VSS) 32 No Connect (NC) 32 Power Connect (VCC2) 32 Power Connect (VCC3) 32 Voltage Detect(VOLDET) 32 Privilege Level Transfers 87 Privilege Levels (CPL, DPL and RPL) 86 Privilege Levels (I/O) 86 Processor Core Instruction Set 212 Clock Counts 212 Flags 212 Legend 212 Opcodes 212 Processor Initialization 38 Programming Interface 38 Protected Mode, Initialization and Transition 87 Protection 86 Current Privilege Level (CPL) 86 Descriptor Privilege Level (DPL) 86 Requested Privilege Level (RPL) 86 Protection - V86 Mode 88 R Register Controls 38 Register Sets 40 Application Flags Register 40 General Purpose Register 40 Instruction Pointer Register 40 Segment Registers 40 Flags Register 43 General Purpose 40 Data Registers 40 Pointer and Index Registers 40 Instruction Pointer 42 Selection Rules 42 Model Specific Register 40 System Register Set 40 ,44 Registers Application Register 40 Model Specific Register 59 REQ 28 RESET 38 ROP (raster operation) 166 Row Address Strobe CAS 29 CKE 29 RAS 29 RASA 29 RASB 29 WE 29 S Scratchpad 2KB configurations 97 3KB configurations 97 SMM information 97 Scratchpad RAM 97 SDRAM Clocks SDCLK_IN 30 SDCLK_OUT 30 Segment Register Selection Rules 42 Segment Registers 42 Serial Packet CX5520 25

www.national.com 242 Revision 3.1 Index (Continued) Geode™ GXm Processor VSA 25 Shutdown and Halt 86 Signal Definitions 13 –23 Signal Descriptions 24 Cyrix Internal Test and Measurement Signals 32 Memory Controller Interface Signals 29 –30 PCI Interface Signals 26 –29 Power, Ground and No Connect Signals 32 System Interface Signals 24 –25 Video Interface Signals 30 –31 Signals - INTR 74 Signals - NMI 74 Signals - SMM 74 SIZE SMM Region Size Bits 51 Skip Counts 94 SMAR SMM Address Region Bits 51 SMAR SMM Address Region Register Indices CDh, CEH, CFh SMHR SMM Header Address 51 SMHR SMI Header Address Indices B0h, B1h, B2h, B3h 50 SMI Configuration Registers 80 Generation 83 SMI# 74 ,75,78 pin 80 SMI# pin 174 SMM 78 CPU States 85 Instructions 82 Memory Space 83 Memory Space Header 80 Operation 79 SMI Enhancements 79 SMI Events 80 SMI Nested States 84 SMI Nesting 83 SMI Service Routine Execution 83 SMI# Pin 80 Suspend Mode 85 Suspend Mode CPU States 85 SMM Memory Space Header Description 81 SPGA Pin Assignments by Pin Number 20 SPGA Pin Assignments by Signal Name 22 SPGA Pin Assignments Diagram 19 STOP 27 Subsystem Signal Connections 34 –35 Suspend 59 ,85,86 Suspend Mode 25 System Error 28 NMI 28 System Interface Signals Interrupt Request 25 Reset 24 Serial Packet 25 Suspend Acknowledge 25 Suspend Request 25 System Clock 24 System Management Interrupt 25 System Management Interrupt (SMI#) 165 System Register Set 44 System Register Sets Cache Test Registers 56 Configuration Registers 47 Debug Registers 47 Gate Descriptors 69 Task Register 69 ,70 System Registers 44 –59 Configuration Registers 47 Control Registers 45 Debug Registers 52 Model Specific Register (MSR) 59 Segment Descriptor Table Registers 66 Test Registers 54 T Task Gate Descriptors 70 Task Register (TR) 70 Task State Segments 70 Thermal Characteristics 195 TR3 Register 57 Cache Data 57 TR4 Register 57 Dirty Bits 57 LRU Bits 57 Upper Tag Address 57 Valid Bit 57 TR5 Register 57 Control Bits 57 Line Selection 57 TR6 Register 55 Command Bit 55 Dirty Attribute Bit 55 Linear Address 55 Valid Bit 55 TR7 Register 55 LRU Bits 55 Physical Address 55 PL Bit 55 Set Selection 55 Translation Lookaside Buffer 95 V V86 Mode Entering and Leaving 88 Interrupt Handling 88 Memory Addressing 88 VESA 165 VGA Address Mapping 167 MapMask register 167 Miscellaneous Output register 167 VGA Configuration Registers 169 VGA Control Register (B9h) 169 VGA Mask Register (BAh-BDh) 169 VGA Front End 166 VGA function attribute controller 166 CRT controller 166 frame buffer 166 general registers 166 graphics controller 166 sequencer 166 VGA Hardware 165 ,168 SMI Generation 168 VGA Address Generator 168 VGA Memory 168 VGA Range Detection 168 VGA Sequencer 168 VGA Write/Read Path 168 VGA Memory 171

Revision 3.1 243 www.national.com Index (Continued) Geode™ GXm Processor frame buffer address 166 host address 166 refresh address 166 VGA Memory Addresses 169 VGA Memory Organization 166 VGA Range Detection 171 VGA Sequencer 171 VGA Video BIOS 171 VGA Video Refresh 168 All Points Addressable mode (APA) 168 attribute controller (ATTR) 168 CGA mode 168 Chain 4 mode 168 ClockSelect field 168 ColorPlaneEnable register 168 CRT controller (CRTC) 168 LineCompare register 168 Miscellaneous Output register 168 ShiftRegister field 168 VGA Write/Read Path 171 Video Data Bus VID_CLK 31 Video Interface Signals CRT Horizontal Sync 30 CRT Vertical Sync 31 Display Enable 31 Dotclock 30 Flat Panel Horizontal Sync 31 Flat Panel Vertical Sync 31 Graphics Pixel Data Bus 31 Pixel Port Clock 30 Video Clock 30 Video Data Bus 31 Video Ready 31 Video Valid 31 video refresh 166 Virtual 8086 Mode (V86) 88 Virtual Subsystem Architecture (VSA) 165 Virtual VGA 165 ColorCompare register 167 ColorDon’tCare register 167 Datapath Elements 167 read mode unit 167 write-mode unit 167 DataRotate register 167 ReadMapSelect register 167 SetReset register 167 SMI Generation 168 Virtual VGA Register Descriptions 172 VIrtual VGA Registers GP_VGA_WRITE (8140h-8143h) 172 Virtual VGA Registers GP_VGA_BASE VGA (8210h-8213h) 172 GP_VGA_LATCH (8214h-8217h) 172 GP_VGA_READ (8144h-8147h) 172 X XpressAUDIO 165

NATIONAL’S PRODUCTS ARE NOT AUTHORIZED FOR USE AS CRITICAL COMPONENTS IN LIFE SUPPORT DEVICES OR SYSTEMS WITHOUT THE EXPRESS WRITTEN APPROVAL OF THE PRESIDENT AND GENERAL COUNSEL OF NATIONAL SEMICONDUCTOR CORPORATION. As used herein: 1. Life support devices or systems are devices or systems which, (a) are intended for surgical implant into the body, or (b) support or sustain life, and whose failure to perform when properly used in accordance with instructions for use provided in the labeling, can be reasonably expected to result in a significant injury to the user. 2. A critical component is any component of a life support device or system whose failure to perform can be reasonably expected to cause the failure of the life support device or system, or to affect its safety or effectiveness. National Semiconductor Corporation Americas Tel: 1-800-272-9959 Fax: 1-800-737-7018 Email: support@nsc.com National Semiconductor Europe Fax: +49 (0) 180-530 85 86 Email: europe.support@nsc.com Deutsch Tel: +49 (0) 69 9508 6208 English Tel: +44 (0) 870 24 0 2171 F r a n c a i sT e l : + 3 3( 0 )14 19 18 7 9 0 National Semiconductor Asia Pacific Customer Response Group Tel: 65-2544466 Fax: 65-2504466 Email: ap.support@nsc.com National Semiconductor Japan Ltd. T el: 81-3-5639-7560 Fax: 81-3-5639-7507 www.national.com Geode™ GXm Processor Integrated x86 Solution with MMX Support National does not assume any responsibility for use of any circuitry described, no circuit patent licenses are implied and National reserves the right at any time without notice to change said circuitry and specifications.